linux-stable/drivers/iommu/tegra-smmu.c

1183 lines
28 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-only
/*
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
* Copyright (C) 2011-2014 NVIDIA CORPORATION. All rights reserved.
*/
#include <linux/bitops.h>
#include <linux/debugfs.h>
#include <linux/err.h>
#include <linux/iommu.h>
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#include <linux/kernel.h>
#include <linux/of.h>
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#include <linux/of_device.h>
#include <linux/platform_device.h>
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/dma-mapping.h>
#include <soc/tegra/ahb.h>
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#include <soc/tegra/mc.h>
struct tegra_smmu_group {
struct list_head list;
struct tegra_smmu *smmu;
const struct tegra_smmu_group_soc *soc;
struct iommu_group *group;
unsigned int swgroup;
};
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
struct tegra_smmu {
void __iomem *regs;
struct device *dev;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
struct tegra_mc *mc;
const struct tegra_smmu_soc *soc;
struct list_head groups;
unsigned long pfn_mask;
unsigned long tlb_mask;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
unsigned long *asids;
struct mutex lock;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
struct list_head list;
struct dentry *debugfs;
struct iommu_device iommu; /* IOMMU Core code handle */
};
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
struct tegra_smmu_as {
struct iommu_domain domain;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
struct tegra_smmu *smmu;
unsigned int use_count;
spinlock_t lock;
u32 *count;
struct page **pts;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
struct page *pd;
dma_addr_t pd_dma;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
unsigned id;
u32 attr;
};
static struct tegra_smmu_as *to_smmu_as(struct iommu_domain *dom)
{
return container_of(dom, struct tegra_smmu_as, domain);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static inline void smmu_writel(struct tegra_smmu *smmu, u32 value,
unsigned long offset)
{
writel(value, smmu->regs + offset);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static inline u32 smmu_readl(struct tegra_smmu *smmu, unsigned long offset)
{
return readl(smmu->regs + offset);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#define SMMU_CONFIG 0x010
#define SMMU_CONFIG_ENABLE (1 << 0)
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#define SMMU_TLB_CONFIG 0x14
#define SMMU_TLB_CONFIG_HIT_UNDER_MISS (1 << 29)
#define SMMU_TLB_CONFIG_ROUND_ROBIN_ARBITRATION (1 << 28)
#define SMMU_TLB_CONFIG_ACTIVE_LINES(smmu) \
((smmu)->soc->num_tlb_lines & (smmu)->tlb_mask)
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#define SMMU_PTC_CONFIG 0x18
#define SMMU_PTC_CONFIG_ENABLE (1 << 29)
#define SMMU_PTC_CONFIG_REQ_LIMIT(x) (((x) & 0x0f) << 24)
#define SMMU_PTC_CONFIG_INDEX_MAP(x) ((x) & 0x3f)
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#define SMMU_PTB_ASID 0x01c
#define SMMU_PTB_ASID_VALUE(x) ((x) & 0x7f)
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#define SMMU_PTB_DATA 0x020
#define SMMU_PTB_DATA_VALUE(dma, attr) ((dma) >> 12 | (attr))
#define SMMU_MK_PDE(dma, attr) ((dma) >> SMMU_PTE_SHIFT | (attr))
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#define SMMU_TLB_FLUSH 0x030
#define SMMU_TLB_FLUSH_VA_MATCH_ALL (0 << 0)
#define SMMU_TLB_FLUSH_VA_MATCH_SECTION (2 << 0)
#define SMMU_TLB_FLUSH_VA_MATCH_GROUP (3 << 0)
#define SMMU_TLB_FLUSH_VA_SECTION(addr) ((((addr) & 0xffc00000) >> 12) | \
SMMU_TLB_FLUSH_VA_MATCH_SECTION)
#define SMMU_TLB_FLUSH_VA_GROUP(addr) ((((addr) & 0xffffc000) >> 12) | \
SMMU_TLB_FLUSH_VA_MATCH_GROUP)
#define SMMU_TLB_FLUSH_ASID_MATCH (1 << 31)
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#define SMMU_PTC_FLUSH 0x034
#define SMMU_PTC_FLUSH_TYPE_ALL (0 << 0)
#define SMMU_PTC_FLUSH_TYPE_ADR (1 << 0)
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#define SMMU_PTC_FLUSH_HI 0x9b8
#define SMMU_PTC_FLUSH_HI_MASK 0x3
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
/* per-SWGROUP SMMU_*_ASID register */
#define SMMU_ASID_ENABLE (1 << 31)
#define SMMU_ASID_MASK 0x7f
#define SMMU_ASID_VALUE(x) ((x) & SMMU_ASID_MASK)
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
/* page table definitions */
#define SMMU_NUM_PDE 1024
#define SMMU_NUM_PTE 1024
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#define SMMU_SIZE_PD (SMMU_NUM_PDE * 4)
#define SMMU_SIZE_PT (SMMU_NUM_PTE * 4)
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#define SMMU_PDE_SHIFT 22
#define SMMU_PTE_SHIFT 12
#define SMMU_PAGE_MASK (~(SMMU_SIZE_PT-1))
#define SMMU_OFFSET_IN_PAGE(x) ((unsigned long)(x) & ~SMMU_PAGE_MASK)
#define SMMU_PFN_PHYS(x) ((phys_addr_t)(x) << SMMU_PTE_SHIFT)
#define SMMU_PHYS_PFN(x) ((unsigned long)((x) >> SMMU_PTE_SHIFT))
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#define SMMU_PD_READABLE (1 << 31)
#define SMMU_PD_WRITABLE (1 << 30)
#define SMMU_PD_NONSECURE (1 << 29)
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#define SMMU_PDE_READABLE (1 << 31)
#define SMMU_PDE_WRITABLE (1 << 30)
#define SMMU_PDE_NONSECURE (1 << 29)
#define SMMU_PDE_NEXT (1 << 28)
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#define SMMU_PTE_READABLE (1 << 31)
#define SMMU_PTE_WRITABLE (1 << 30)
#define SMMU_PTE_NONSECURE (1 << 29)
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#define SMMU_PDE_ATTR (SMMU_PDE_READABLE | SMMU_PDE_WRITABLE | \
SMMU_PDE_NONSECURE)
static unsigned int iova_pd_index(unsigned long iova)
{
return (iova >> SMMU_PDE_SHIFT) & (SMMU_NUM_PDE - 1);
}
static unsigned int iova_pt_index(unsigned long iova)
{
return (iova >> SMMU_PTE_SHIFT) & (SMMU_NUM_PTE - 1);
}
static bool smmu_dma_addr_valid(struct tegra_smmu *smmu, dma_addr_t addr)
{
addr >>= 12;
return (addr & smmu->pfn_mask) == addr;
}
static dma_addr_t smmu_pde_to_dma(struct tegra_smmu *smmu, u32 pde)
{
return (dma_addr_t)(pde & smmu->pfn_mask) << 12;
}
static void smmu_flush_ptc_all(struct tegra_smmu *smmu)
{
smmu_writel(smmu, SMMU_PTC_FLUSH_TYPE_ALL, SMMU_PTC_FLUSH);
}
static inline void smmu_flush_ptc(struct tegra_smmu *smmu, dma_addr_t dma,
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
unsigned long offset)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
u32 value;
offset &= ~(smmu->mc->soc->atom_size - 1);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
if (smmu->mc->soc->num_address_bits > 32) {
#ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
value = (dma >> 32) & SMMU_PTC_FLUSH_HI_MASK;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#else
value = 0;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
#endif
smmu_writel(smmu, value, SMMU_PTC_FLUSH_HI);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
value = (dma + offset) | SMMU_PTC_FLUSH_TYPE_ADR;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
smmu_writel(smmu, value, SMMU_PTC_FLUSH);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static inline void smmu_flush_tlb(struct tegra_smmu *smmu)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
smmu_writel(smmu, SMMU_TLB_FLUSH_VA_MATCH_ALL, SMMU_TLB_FLUSH);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static inline void smmu_flush_tlb_asid(struct tegra_smmu *smmu,
unsigned long asid)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
u32 value;
if (smmu->soc->num_asids == 4)
value = (asid & 0x3) << 29;
else
value = (asid & 0x7f) << 24;
value |= SMMU_TLB_FLUSH_ASID_MATCH | SMMU_TLB_FLUSH_VA_MATCH_ALL;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
smmu_writel(smmu, value, SMMU_TLB_FLUSH);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static inline void smmu_flush_tlb_section(struct tegra_smmu *smmu,
unsigned long asid,
unsigned long iova)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
u32 value;
if (smmu->soc->num_asids == 4)
value = (asid & 0x3) << 29;
else
value = (asid & 0x7f) << 24;
value |= SMMU_TLB_FLUSH_ASID_MATCH | SMMU_TLB_FLUSH_VA_SECTION(iova);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
smmu_writel(smmu, value, SMMU_TLB_FLUSH);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static inline void smmu_flush_tlb_group(struct tegra_smmu *smmu,
unsigned long asid,
unsigned long iova)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
u32 value;
if (smmu->soc->num_asids == 4)
value = (asid & 0x3) << 29;
else
value = (asid & 0x7f) << 24;
value |= SMMU_TLB_FLUSH_ASID_MATCH | SMMU_TLB_FLUSH_VA_GROUP(iova);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
smmu_writel(smmu, value, SMMU_TLB_FLUSH);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static inline void smmu_flush(struct tegra_smmu *smmu)
{
smmu_readl(smmu, SMMU_PTB_ASID);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static int tegra_smmu_alloc_asid(struct tegra_smmu *smmu, unsigned int *idp)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
unsigned long id;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
mutex_lock(&smmu->lock);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
id = find_first_zero_bit(smmu->asids, smmu->soc->num_asids);
if (id >= smmu->soc->num_asids) {
mutex_unlock(&smmu->lock);
return -ENOSPC;
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
set_bit(id, smmu->asids);
*idp = id;
mutex_unlock(&smmu->lock);
return 0;
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static void tegra_smmu_free_asid(struct tegra_smmu *smmu, unsigned int id)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
mutex_lock(&smmu->lock);
clear_bit(id, smmu->asids);
mutex_unlock(&smmu->lock);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static bool tegra_smmu_capable(enum iommu_cap cap)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
return false;
}
static struct iommu_domain *tegra_smmu_domain_alloc(unsigned type)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
struct tegra_smmu_as *as;
if (type != IOMMU_DOMAIN_UNMANAGED)
return NULL;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
as = kzalloc(sizeof(*as), GFP_KERNEL);
if (!as)
return NULL;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
as->attr = SMMU_PD_READABLE | SMMU_PD_WRITABLE | SMMU_PD_NONSECURE;
as->pd = alloc_page(GFP_KERNEL | __GFP_DMA | __GFP_ZERO);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
if (!as->pd) {
kfree(as);
return NULL;
}
as->count = kcalloc(SMMU_NUM_PDE, sizeof(u32), GFP_KERNEL);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
if (!as->count) {
__free_page(as->pd);
kfree(as);
return NULL;
}
as->pts = kcalloc(SMMU_NUM_PDE, sizeof(*as->pts), GFP_KERNEL);
if (!as->pts) {
kfree(as->count);
__free_page(as->pd);
kfree(as);
return NULL;
}
spin_lock_init(&as->lock);
/* setup aperture */
as->domain.geometry.aperture_start = 0;
as->domain.geometry.aperture_end = 0xffffffff;
as->domain.geometry.force_aperture = true;
return &as->domain;
}
static void tegra_smmu_domain_free(struct iommu_domain *domain)
{
struct tegra_smmu_as *as = to_smmu_as(domain);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
/* TODO: free page directory and page tables */
WARN_ON_ONCE(as->use_count);
kfree(as->count);
kfree(as->pts);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
kfree(as);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static const struct tegra_smmu_swgroup *
tegra_smmu_find_swgroup(struct tegra_smmu *smmu, unsigned int swgroup)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
const struct tegra_smmu_swgroup *group = NULL;
unsigned int i;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
for (i = 0; i < smmu->soc->num_swgroups; i++) {
if (smmu->soc->swgroups[i].swgroup == swgroup) {
group = &smmu->soc->swgroups[i];
break;
}
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
return group;
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static void tegra_smmu_enable(struct tegra_smmu *smmu, unsigned int swgroup,
unsigned int asid)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
const struct tegra_smmu_swgroup *group;
unsigned int i;
u32 value;
group = tegra_smmu_find_swgroup(smmu, swgroup);
if (group) {
value = smmu_readl(smmu, group->reg);
value &= ~SMMU_ASID_MASK;
value |= SMMU_ASID_VALUE(asid);
value |= SMMU_ASID_ENABLE;
smmu_writel(smmu, value, group->reg);
} else {
pr_warn("%s group from swgroup %u not found\n", __func__,
swgroup);
/* No point moving ahead if group was not found */
return;
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
for (i = 0; i < smmu->soc->num_clients; i++) {
const struct tegra_mc_client *client = &smmu->soc->clients[i];
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
if (client->swgroup != swgroup)
continue;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
value = smmu_readl(smmu, client->smmu.reg);
value |= BIT(client->smmu.bit);
smmu_writel(smmu, value, client->smmu.reg);
}
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static void tegra_smmu_disable(struct tegra_smmu *smmu, unsigned int swgroup,
unsigned int asid)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
const struct tegra_smmu_swgroup *group;
unsigned int i;
u32 value;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
group = tegra_smmu_find_swgroup(smmu, swgroup);
if (group) {
value = smmu_readl(smmu, group->reg);
value &= ~SMMU_ASID_MASK;
value |= SMMU_ASID_VALUE(asid);
value &= ~SMMU_ASID_ENABLE;
smmu_writel(smmu, value, group->reg);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
for (i = 0; i < smmu->soc->num_clients; i++) {
const struct tegra_mc_client *client = &smmu->soc->clients[i];
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
if (client->swgroup != swgroup)
continue;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
value = smmu_readl(smmu, client->smmu.reg);
value &= ~BIT(client->smmu.bit);
smmu_writel(smmu, value, client->smmu.reg);
}
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static int tegra_smmu_as_prepare(struct tegra_smmu *smmu,
struct tegra_smmu_as *as)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
u32 value;
int err;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
if (as->use_count > 0) {
as->use_count++;
return 0;
}
as->pd_dma = dma_map_page(smmu->dev, as->pd, 0, SMMU_SIZE_PD,
DMA_TO_DEVICE);
if (dma_mapping_error(smmu->dev, as->pd_dma))
return -ENOMEM;
/* We can't handle 64-bit DMA addresses */
if (!smmu_dma_addr_valid(smmu, as->pd_dma)) {
err = -ENOMEM;
goto err_unmap;
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
err = tegra_smmu_alloc_asid(smmu, &as->id);
if (err < 0)
goto err_unmap;
smmu_flush_ptc(smmu, as->pd_dma, 0);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
smmu_flush_tlb_asid(smmu, as->id);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
smmu_writel(smmu, as->id & 0x7f, SMMU_PTB_ASID);
value = SMMU_PTB_DATA_VALUE(as->pd_dma, as->attr);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
smmu_writel(smmu, value, SMMU_PTB_DATA);
smmu_flush(smmu);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
as->smmu = smmu;
as->use_count++;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
return 0;
err_unmap:
dma_unmap_page(smmu->dev, as->pd_dma, SMMU_SIZE_PD, DMA_TO_DEVICE);
return err;
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static void tegra_smmu_as_unprepare(struct tegra_smmu *smmu,
struct tegra_smmu_as *as)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
if (--as->use_count > 0)
return;
tegra_smmu_free_asid(smmu, as->id);
dma_unmap_page(smmu->dev, as->pd_dma, SMMU_SIZE_PD, DMA_TO_DEVICE);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
as->smmu = NULL;
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static int tegra_smmu_attach_dev(struct iommu_domain *domain,
struct device *dev)
{
struct tegra_smmu *smmu = dev_iommu_priv_get(dev);
struct tegra_smmu_as *as = to_smmu_as(domain);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
struct device_node *np = dev->of_node;
struct of_phandle_args args;
unsigned int index = 0;
int err = 0;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
while (!of_parse_phandle_with_args(np, "iommus", "#iommu-cells", index,
&args)) {
unsigned int swgroup = args.args[0];
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
if (args.np != smmu->dev->of_node) {
of_node_put(args.np);
continue;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
of_node_put(args.np);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
err = tegra_smmu_as_prepare(smmu, as);
if (err < 0)
return err;
tegra_smmu_enable(smmu, swgroup, as->id);
index++;
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
if (index == 0)
return -ENODEV;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
return 0;
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static void tegra_smmu_detach_dev(struct iommu_domain *domain, struct device *dev)
{
struct tegra_smmu_as *as = to_smmu_as(domain);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
struct device_node *np = dev->of_node;
struct tegra_smmu *smmu = as->smmu;
struct of_phandle_args args;
unsigned int index = 0;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
while (!of_parse_phandle_with_args(np, "iommus", "#iommu-cells", index,
&args)) {
unsigned int swgroup = args.args[0];
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
if (args.np != smmu->dev->of_node) {
of_node_put(args.np);
continue;
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
of_node_put(args.np);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
tegra_smmu_disable(smmu, swgroup, as->id);
tegra_smmu_as_unprepare(smmu, as);
index++;
}
}
static void tegra_smmu_set_pde(struct tegra_smmu_as *as, unsigned long iova,
u32 value)
{
unsigned int pd_index = iova_pd_index(iova);
struct tegra_smmu *smmu = as->smmu;
u32 *pd = page_address(as->pd);
unsigned long offset = pd_index * sizeof(*pd);
/* Set the page directory entry first */
pd[pd_index] = value;
/* The flush the page directory entry from caches */
dma_sync_single_range_for_device(smmu->dev, as->pd_dma, offset,
sizeof(*pd), DMA_TO_DEVICE);
/* And flush the iommu */
smmu_flush_ptc(smmu, as->pd_dma, offset);
smmu_flush_tlb_section(smmu, as->id, iova);
smmu_flush(smmu);
}
static u32 *tegra_smmu_pte_offset(struct page *pt_page, unsigned long iova)
{
u32 *pt = page_address(pt_page);
return pt + iova_pt_index(iova);
}
static u32 *tegra_smmu_pte_lookup(struct tegra_smmu_as *as, unsigned long iova,
dma_addr_t *dmap)
{
unsigned int pd_index = iova_pd_index(iova);
struct tegra_smmu *smmu = as->smmu;
struct page *pt_page;
u32 *pd;
pt_page = as->pts[pd_index];
if (!pt_page)
return NULL;
pd = page_address(as->pd);
*dmap = smmu_pde_to_dma(smmu, pd[pd_index]);
return tegra_smmu_pte_offset(pt_page, iova);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static u32 *as_get_pte(struct tegra_smmu_as *as, dma_addr_t iova,
dma_addr_t *dmap, struct page *page)
{
unsigned int pde = iova_pd_index(iova);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
struct tegra_smmu *smmu = as->smmu;
if (!as->pts[pde]) {
dma_addr_t dma;
dma = dma_map_page(smmu->dev, page, 0, SMMU_SIZE_PT,
DMA_TO_DEVICE);
if (dma_mapping_error(smmu->dev, dma)) {
__free_page(page);
return NULL;
}
if (!smmu_dma_addr_valid(smmu, dma)) {
dma_unmap_page(smmu->dev, dma, SMMU_SIZE_PT,
DMA_TO_DEVICE);
__free_page(page);
return NULL;
}
as->pts[pde] = page;
tegra_smmu_set_pde(as, iova, SMMU_MK_PDE(dma, SMMU_PDE_ATTR |
SMMU_PDE_NEXT));
*dmap = dma;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
} else {
u32 *pd = page_address(as->pd);
*dmap = smmu_pde_to_dma(smmu, pd[pde]);
}
return tegra_smmu_pte_offset(as->pts[pde], iova);
}
static void tegra_smmu_pte_get_use(struct tegra_smmu_as *as, unsigned long iova)
{
unsigned int pd_index = iova_pd_index(iova);
as->count[pd_index]++;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
}
static void tegra_smmu_pte_put_use(struct tegra_smmu_as *as, unsigned long iova)
{
unsigned int pde = iova_pd_index(iova);
struct page *page = as->pts[pde];
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
/*
* When no entries in this page table are used anymore, return the
* memory page to the system.
*/
if (--as->count[pde] == 0) {
struct tegra_smmu *smmu = as->smmu;
u32 *pd = page_address(as->pd);
dma_addr_t pte_dma = smmu_pde_to_dma(smmu, pd[pde]);
tegra_smmu_set_pde(as, iova, 0);
dma_unmap_page(smmu->dev, pte_dma, SMMU_SIZE_PT, DMA_TO_DEVICE);
__free_page(page);
as->pts[pde] = NULL;
}
}
static void tegra_smmu_set_pte(struct tegra_smmu_as *as, unsigned long iova,
u32 *pte, dma_addr_t pte_dma, u32 val)
{
struct tegra_smmu *smmu = as->smmu;
unsigned long offset = SMMU_OFFSET_IN_PAGE(pte);
*pte = val;
dma_sync_single_range_for_device(smmu->dev, pte_dma, offset,
4, DMA_TO_DEVICE);
smmu_flush_ptc(smmu, pte_dma, offset);
smmu_flush_tlb_group(smmu, as->id, iova);
smmu_flush(smmu);
}
static struct page *as_get_pde_page(struct tegra_smmu_as *as,
unsigned long iova, gfp_t gfp,
unsigned long *flags)
{
unsigned int pde = iova_pd_index(iova);
struct page *page = as->pts[pde];
/* at first check whether allocation needs to be done at all */
if (page)
return page;
/*
* In order to prevent exhaustion of the atomic memory pool, we
* allocate page in a sleeping context if GFP flags permit. Hence
* spinlock needs to be unlocked and re-locked after allocation.
*/
if (!(gfp & __GFP_ATOMIC))
spin_unlock_irqrestore(&as->lock, *flags);
page = alloc_page(gfp | __GFP_DMA | __GFP_ZERO);
if (!(gfp & __GFP_ATOMIC))
spin_lock_irqsave(&as->lock, *flags);
/*
* In a case of blocking allocation, a concurrent mapping may win
* the PDE allocation. In this case the allocated page isn't needed
* if allocation succeeded and the allocation failure isn't fatal.
*/
if (as->pts[pde]) {
if (page)
__free_page(page);
page = as->pts[pde];
}
return page;
}
static int
__tegra_smmu_map(struct iommu_domain *domain, unsigned long iova,
phys_addr_t paddr, size_t size, int prot, gfp_t gfp,
unsigned long *flags)
{
struct tegra_smmu_as *as = to_smmu_as(domain);
dma_addr_t pte_dma;
struct page *page;
u32 pte_attrs;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
u32 *pte;
page = as_get_pde_page(as, iova, gfp, flags);
if (!page)
return -ENOMEM;
pte = as_get_pte(as, iova, &pte_dma, page);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
if (!pte)
return -ENOMEM;
/* If we aren't overwriting a pre-existing entry, increment use */
if (*pte == 0)
tegra_smmu_pte_get_use(as, iova);
pte_attrs = SMMU_PTE_NONSECURE;
if (prot & IOMMU_READ)
pte_attrs |= SMMU_PTE_READABLE;
if (prot & IOMMU_WRITE)
pte_attrs |= SMMU_PTE_WRITABLE;
tegra_smmu_set_pte(as, iova, pte, pte_dma,
SMMU_PHYS_PFN(paddr) | pte_attrs);
return 0;
}
static size_t
__tegra_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
size_t size, struct iommu_iotlb_gather *gather)
{
struct tegra_smmu_as *as = to_smmu_as(domain);
dma_addr_t pte_dma;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
u32 *pte;
pte = tegra_smmu_pte_lookup(as, iova, &pte_dma);
if (!pte || !*pte)
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
return 0;
tegra_smmu_set_pte(as, iova, pte, pte_dma, 0);
tegra_smmu_pte_put_use(as, iova);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
return size;
}
static int tegra_smmu_map(struct iommu_domain *domain, unsigned long iova,
phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
{
struct tegra_smmu_as *as = to_smmu_as(domain);
unsigned long flags;
int ret;
spin_lock_irqsave(&as->lock, flags);
ret = __tegra_smmu_map(domain, iova, paddr, size, prot, gfp, &flags);
spin_unlock_irqrestore(&as->lock, flags);
return ret;
}
static size_t tegra_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
size_t size, struct iommu_iotlb_gather *gather)
{
struct tegra_smmu_as *as = to_smmu_as(domain);
unsigned long flags;
spin_lock_irqsave(&as->lock, flags);
size = __tegra_smmu_unmap(domain, iova, size, gather);
spin_unlock_irqrestore(&as->lock, flags);
return size;
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static phys_addr_t tegra_smmu_iova_to_phys(struct iommu_domain *domain,
dma_addr_t iova)
{
struct tegra_smmu_as *as = to_smmu_as(domain);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
unsigned long pfn;
dma_addr_t pte_dma;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
u32 *pte;
pte = tegra_smmu_pte_lookup(as, iova, &pte_dma);
if (!pte || !*pte)
return 0;
pfn = *pte & as->smmu->pfn_mask;
return SMMU_PFN_PHYS(pfn) + SMMU_OFFSET_IN_PAGE(iova);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static struct tegra_smmu *tegra_smmu_find(struct device_node *np)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
struct platform_device *pdev;
struct tegra_mc *mc;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
pdev = of_find_device_by_node(np);
if (!pdev)
return NULL;
mc = platform_get_drvdata(pdev);
if (!mc)
return NULL;
return mc->smmu;
}
static int tegra_smmu_configure(struct tegra_smmu *smmu, struct device *dev,
struct of_phandle_args *args)
{
const struct iommu_ops *ops = smmu->iommu.ops;
int err;
err = iommu_fwspec_init(dev, &dev->of_node->fwnode, ops);
if (err < 0) {
dev_err(dev, "failed to initialize fwspec: %d\n", err);
return err;
}
err = ops->of_xlate(dev, args);
if (err < 0) {
dev_err(dev, "failed to parse SW group ID: %d\n", err);
iommu_fwspec_free(dev);
return err;
}
return 0;
}
static struct iommu_device *tegra_smmu_probe_device(struct device *dev)
{
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
struct device_node *np = dev->of_node;
struct tegra_smmu *smmu = NULL;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
struct of_phandle_args args;
unsigned int index = 0;
int err;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
while (of_parse_phandle_with_args(np, "iommus", "#iommu-cells", index,
&args) == 0) {
smmu = tegra_smmu_find(args.np);
if (smmu) {
err = tegra_smmu_configure(smmu, dev, &args);
of_node_put(args.np);
if (err < 0)
return ERR_PTR(err);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
/*
* Only a single IOMMU master interface is currently
* supported by the Linux kernel, so abort after the
* first match.
*/
dev_iommu_priv_set(dev, smmu);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
break;
}
of_node_put(args.np);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
index++;
}
if (!smmu)
return ERR_PTR(-ENODEV);
return &smmu->iommu;
}
static void tegra_smmu_release_device(struct device *dev)
{
dev_iommu_priv_set(dev, NULL);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
}
static const struct tegra_smmu_group_soc *
tegra_smmu_find_group(struct tegra_smmu *smmu, unsigned int swgroup)
{
unsigned int i, j;
for (i = 0; i < smmu->soc->num_groups; i++)
for (j = 0; j < smmu->soc->groups[i].num_swgroups; j++)
if (smmu->soc->groups[i].swgroups[j] == swgroup)
return &smmu->soc->groups[i];
return NULL;
}
static void tegra_smmu_group_release(void *iommu_data)
{
struct tegra_smmu_group *group = iommu_data;
struct tegra_smmu *smmu = group->smmu;
mutex_lock(&smmu->lock);
list_del(&group->list);
mutex_unlock(&smmu->lock);
}
static struct iommu_group *tegra_smmu_group_get(struct tegra_smmu *smmu,
unsigned int swgroup)
{
const struct tegra_smmu_group_soc *soc;
struct tegra_smmu_group *group;
struct iommu_group *grp;
/* Find group_soc associating with swgroup */
soc = tegra_smmu_find_group(smmu, swgroup);
mutex_lock(&smmu->lock);
/* Find existing iommu_group associating with swgroup or group_soc */
list_for_each_entry(group, &smmu->groups, list)
if ((group->swgroup == swgroup) || (soc && group->soc == soc)) {
grp = iommu_group_ref_get(group->group);
mutex_unlock(&smmu->lock);
return grp;
}
group = devm_kzalloc(smmu->dev, sizeof(*group), GFP_KERNEL);
if (!group) {
mutex_unlock(&smmu->lock);
return NULL;
}
INIT_LIST_HEAD(&group->list);
group->swgroup = swgroup;
group->smmu = smmu;
group->soc = soc;
group->group = iommu_group_alloc();
if (IS_ERR(group->group)) {
devm_kfree(smmu->dev, group);
mutex_unlock(&smmu->lock);
return NULL;
}
iommu_group_set_iommudata(group->group, group, tegra_smmu_group_release);
if (soc)
iommu_group_set_name(group->group, soc->name);
list_add_tail(&group->list, &smmu->groups);
mutex_unlock(&smmu->lock);
return group->group;
}
static struct iommu_group *tegra_smmu_device_group(struct device *dev)
{
struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
struct tegra_smmu *smmu = dev_iommu_priv_get(dev);
struct iommu_group *group;
group = tegra_smmu_group_get(smmu, fwspec->ids[0]);
if (!group)
group = generic_device_group(dev);
return group;
}
static int tegra_smmu_of_xlate(struct device *dev,
struct of_phandle_args *args)
{
u32 id = args->args[0];
return iommu_fwspec_add_ids(dev, &id, 1);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static const struct iommu_ops tegra_smmu_ops = {
.capable = tegra_smmu_capable,
.domain_alloc = tegra_smmu_domain_alloc,
.domain_free = tegra_smmu_domain_free,
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
.attach_dev = tegra_smmu_attach_dev,
.detach_dev = tegra_smmu_detach_dev,
.probe_device = tegra_smmu_probe_device,
.release_device = tegra_smmu_release_device,
.device_group = tegra_smmu_device_group,
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
.map = tegra_smmu_map,
.unmap = tegra_smmu_unmap,
.iova_to_phys = tegra_smmu_iova_to_phys,
.of_xlate = tegra_smmu_of_xlate,
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
.pgsize_bitmap = SZ_4K,
};
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
static void tegra_smmu_ahb_enable(void)
{
static const struct of_device_id ahb_match[] = {
{ .compatible = "nvidia,tegra30-ahb", },
{ }
};
struct device_node *ahb;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
ahb = of_find_matching_node(NULL, ahb_match);
if (ahb) {
tegra_ahb_enable_smmu(ahb);
of_node_put(ahb);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
}
static int tegra_smmu_swgroups_show(struct seq_file *s, void *data)
{
struct tegra_smmu *smmu = s->private;
unsigned int i;
u32 value;
seq_printf(s, "swgroup enabled ASID\n");
seq_printf(s, "------------------------\n");
for (i = 0; i < smmu->soc->num_swgroups; i++) {
const struct tegra_smmu_swgroup *group = &smmu->soc->swgroups[i];
const char *status;
unsigned int asid;
value = smmu_readl(smmu, group->reg);
if (value & SMMU_ASID_ENABLE)
status = "yes";
else
status = "no";
asid = value & SMMU_ASID_MASK;
seq_printf(s, "%-9s %-7s %#04x\n", group->name, status,
asid);
}
return 0;
}
DEFINE_SHOW_ATTRIBUTE(tegra_smmu_swgroups);
static int tegra_smmu_clients_show(struct seq_file *s, void *data)
{
struct tegra_smmu *smmu = s->private;
unsigned int i;
u32 value;
seq_printf(s, "client enabled\n");
seq_printf(s, "--------------------\n");
for (i = 0; i < smmu->soc->num_clients; i++) {
const struct tegra_mc_client *client = &smmu->soc->clients[i];
const char *status;
value = smmu_readl(smmu, client->smmu.reg);
if (value & BIT(client->smmu.bit))
status = "yes";
else
status = "no";
seq_printf(s, "%-12s %s\n", client->name, status);
}
return 0;
}
DEFINE_SHOW_ATTRIBUTE(tegra_smmu_clients);
static void tegra_smmu_debugfs_init(struct tegra_smmu *smmu)
{
smmu->debugfs = debugfs_create_dir("smmu", NULL);
if (!smmu->debugfs)
return;
debugfs_create_file("swgroups", S_IRUGO, smmu->debugfs, smmu,
&tegra_smmu_swgroups_fops);
debugfs_create_file("clients", S_IRUGO, smmu->debugfs, smmu,
&tegra_smmu_clients_fops);
}
static void tegra_smmu_debugfs_exit(struct tegra_smmu *smmu)
{
debugfs_remove_recursive(smmu->debugfs);
}
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
struct tegra_smmu *tegra_smmu_probe(struct device *dev,
const struct tegra_smmu_soc *soc,
struct tegra_mc *mc)
{
struct tegra_smmu *smmu;
size_t size;
u32 value;
int err;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
if (!smmu)
return ERR_PTR(-ENOMEM);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
/*
* This is a bit of a hack. Ideally we'd want to simply return this
* value. However the IOMMU registration process will attempt to add
* all devices to the IOMMU when bus_set_iommu() is called. In order
* not to rely on global variables to track the IOMMU instance, we
* set it here so that it can be looked up from the .probe_device()
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
* callback via the IOMMU device's .drvdata field.
*/
mc->smmu = smmu;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
size = BITS_TO_LONGS(soc->num_asids) * sizeof(long);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
smmu->asids = devm_kzalloc(dev, size, GFP_KERNEL);
if (!smmu->asids)
return ERR_PTR(-ENOMEM);
INIT_LIST_HEAD(&smmu->groups);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
mutex_init(&smmu->lock);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
smmu->regs = mc->regs;
smmu->soc = soc;
smmu->dev = dev;
smmu->mc = mc;
smmu->pfn_mask =
BIT_MASK(mc->soc->num_address_bits - SMMU_PTE_SHIFT) - 1;
dev_dbg(dev, "address bits: %u, PFN mask: %#lx\n",
mc->soc->num_address_bits, smmu->pfn_mask);
smmu->tlb_mask = (1 << fls(smmu->soc->num_tlb_lines)) - 1;
dev_dbg(dev, "TLB lines: %u, mask: %#lx\n", smmu->soc->num_tlb_lines,
smmu->tlb_mask);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
value = SMMU_PTC_CONFIG_ENABLE | SMMU_PTC_CONFIG_INDEX_MAP(0x3f);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
if (soc->supports_request_limit)
value |= SMMU_PTC_CONFIG_REQ_LIMIT(8);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
smmu_writel(smmu, value, SMMU_PTC_CONFIG);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
value = SMMU_TLB_CONFIG_HIT_UNDER_MISS |
SMMU_TLB_CONFIG_ACTIVE_LINES(smmu);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
if (soc->supports_round_robin_arbitration)
value |= SMMU_TLB_CONFIG_ROUND_ROBIN_ARBITRATION;
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
smmu_writel(smmu, value, SMMU_TLB_CONFIG);
smmu_flush_ptc_all(smmu);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
smmu_flush_tlb(smmu);
smmu_writel(smmu, SMMU_CONFIG_ENABLE, SMMU_CONFIG);
smmu_flush(smmu);
tegra_smmu_ahb_enable();
err = iommu_device_sysfs_add(&smmu->iommu, dev, NULL, dev_name(dev));
if (err)
return ERR_PTR(err);
iommu_device_set_ops(&smmu->iommu, &tegra_smmu_ops);
iommu_device_set_fwnode(&smmu->iommu, dev->fwnode);
err = iommu_device_register(&smmu->iommu);
if (err) {
iommu_device_sysfs_remove(&smmu->iommu);
return ERR_PTR(err);
}
err = bus_set_iommu(&platform_bus_type, &tegra_smmu_ops);
if (err < 0) {
iommu_device_unregister(&smmu->iommu);
iommu_device_sysfs_remove(&smmu->iommu);
return ERR_PTR(err);
}
if (IS_ENABLED(CONFIG_DEBUG_FS))
tegra_smmu_debugfs_init(smmu);
memory: Add NVIDIA Tegra memory controller support The memory controller on NVIDIA Tegra exposes various knobs that can be used to tune the behaviour of the clients attached to it. Currently this driver sets up the latency allowance registers to the HW defaults. Eventually an API should be exported by this driver (via a custom API or a generic subsystem) to allow clients to register latency requirements. This driver also registers an IOMMU (SMMU) that's implemented by the memory controller. It is supported on Tegra30, Tegra114 and Tegra124 currently. Tegra20 has a GART instead. The Tegra SMMU operates on memory clients and SWGROUPs. A memory client is a unidirectional, special-purpose DMA master. A SWGROUP represents a set of memory clients that form a logical functional unit corresponding to a single device. Typically a device has two clients: one client for read transactions and one client for write transactions, but there are also devices that have only read clients, but many of them (such as the display controllers). Because there is no 1:1 relationship between memory clients and devices the driver keeps a table of memory clients and the SWGROUPs that they belong to per SoC. Note that this is an exception and due to the fact that the SMMU is tightly integrated with the rest of the Tegra SoC. The use of these tables is discouraged in drivers for generic IOMMU devices such as the ARM SMMU because the same IOMMU could be used in any number of SoCs and keeping such tables for each SoC would not scale. Acked-by: Joerg Roedel <jroedel@suse.de> Signed-off-by: Thierry Reding <treding@nvidia.com>
2014-04-16 07:24:44 +00:00
return smmu;
}
void tegra_smmu_remove(struct tegra_smmu *smmu)
{
iommu_device_unregister(&smmu->iommu);
iommu_device_sysfs_remove(&smmu->iommu);
if (IS_ENABLED(CONFIG_DEBUG_FS))
tegra_smmu_debugfs_exit(smmu);
}