linux-stable/drivers/pci/controller/pci-aardvark.c

2014 lines
59 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
/*
* Driver for the Aardvark PCIe controller, used on Marvell Armada
* 3700.
*
* Copyright (C) 2016 Marvell
*
* Author: Hezi Shahmoon <hezi.shahmoon@marvell.com>
*/
#include <linux/bitfield.h>
#include <linux/delay.h>
#include <linux/gpio/consumer.h>
#include <linux/interrupt.h>
#include <linux/irq.h>
#include <linux/irqdomain.h>
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/pci.h>
PCI: Unify ECAM constants in native PCI Express drivers Add ECAM-related constants to provide a set of standard constants defining memory address shift values to the byte-level address that can be used to access the PCI Express Configuration Space, and then move native PCI Express controller drivers to use the newly introduced definitions retiring driver-specific ones. Refactor pci_ecam_map_bus() function to use newly added constants so that limits to the bus, device function and offset (now limited to 4K as per the specification) are in place to prevent the defective or malicious caller from supplying incorrect configuration offset and thus targeting the wrong device when accessing extended configuration space. This refactor also allows for the ".bus_shift" initialisers to be dropped when the user is not using a custom value as a default value will be used as per the PCI Express Specification. Thanks to Qian Cai <qcai@redhat.com>, Michael Walle <michael@walle.cc>, and Vladimir Oltean <olteanv@gmail.com> for reporting a pci_ecam_create() issue with .bus_shift and to Vladimir for proposing the fix. [bhelgaas: incorporate Vladimir's fix, update commit log] Suggested-by: Bjorn Helgaas <bhelgaas@google.com> Link: https://lore.kernel.org/r/20201129230743.3006978-2-kw@linux.com Tested-by: Michael Walle <michael@walle.cc> Signed-off-by: Krzysztof Wilczyński <kw@linux.com> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com> Reviewed-by: Jon Derrick <jonathan.derrick@intel.com> Reviewed-by: Bjorn Helgaas <bhelgaas@google.com>
2020-11-29 23:07:39 +00:00
#include <linux/pci-ecam.h>
#include <linux/init.h>
#include <linux/phy/phy.h>
#include <linux/platform_device.h>
#include <linux/msi.h>
#include <linux/of_address.h>
#include <linux/of_gpio.h>
#include <linux/of_pci.h>
#include "../pci.h"
#include "../pci-bridge-emul.h"
/* PCIe core registers */
#define PCIE_CORE_DEV_ID_REG 0x0
#define PCIE_CORE_CMD_STATUS_REG 0x4
#define PCIE_CORE_DEV_REV_REG 0x8
#define PCIE_CORE_SSDEV_ID_REG 0x2c
#define PCIE_CORE_PCIEXP_CAP 0xc0
#define PCIE_CORE_PCIERR_CAP 0x100
#define PCIE_CORE_ERR_CAPCTL_REG 0x118
#define PCIE_CORE_ERR_CAPCTL_ECRC_CHK_TX BIT(5)
#define PCIE_CORE_ERR_CAPCTL_ECRC_CHK_TX_EN BIT(6)
#define PCIE_CORE_ERR_CAPCTL_ECRC_CHCK BIT(7)
#define PCIE_CORE_ERR_CAPCTL_ECRC_CHCK_RCV BIT(8)
/* PIO registers base address and register offsets */
#define PIO_BASE_ADDR 0x4000
#define PIO_CTRL (PIO_BASE_ADDR + 0x0)
#define PIO_CTRL_TYPE_MASK GENMASK(3, 0)
#define PIO_CTRL_ADDR_WIN_DISABLE BIT(24)
#define PIO_STAT (PIO_BASE_ADDR + 0x4)
#define PIO_COMPLETION_STATUS_SHIFT 7
#define PIO_COMPLETION_STATUS_MASK GENMASK(9, 7)
#define PIO_COMPLETION_STATUS_OK 0
#define PIO_COMPLETION_STATUS_UR 1
#define PIO_COMPLETION_STATUS_CRS 2
#define PIO_COMPLETION_STATUS_CA 4
#define PIO_NON_POSTED_REQ BIT(10)
#define PIO_ERR_STATUS BIT(11)
#define PIO_ADDR_LS (PIO_BASE_ADDR + 0x8)
#define PIO_ADDR_MS (PIO_BASE_ADDR + 0xc)
#define PIO_WR_DATA (PIO_BASE_ADDR + 0x10)
#define PIO_WR_DATA_STRB (PIO_BASE_ADDR + 0x14)
#define PIO_RD_DATA (PIO_BASE_ADDR + 0x18)
#define PIO_START (PIO_BASE_ADDR + 0x1c)
#define PIO_ISR (PIO_BASE_ADDR + 0x20)
#define PIO_ISRM (PIO_BASE_ADDR + 0x24)
/* Aardvark Control registers */
#define CONTROL_BASE_ADDR 0x4800
#define PCIE_CORE_CTRL0_REG (CONTROL_BASE_ADDR + 0x0)
#define PCIE_GEN_SEL_MSK 0x3
#define PCIE_GEN_SEL_SHIFT 0x0
#define SPEED_GEN_1 0
#define SPEED_GEN_2 1
#define SPEED_GEN_3 2
#define IS_RC_MSK 1
#define IS_RC_SHIFT 2
#define LANE_CNT_MSK 0x18
#define LANE_CNT_SHIFT 0x3
#define LANE_COUNT_1 (0 << LANE_CNT_SHIFT)
#define LANE_COUNT_2 (1 << LANE_CNT_SHIFT)
#define LANE_COUNT_4 (2 << LANE_CNT_SHIFT)
#define LANE_COUNT_8 (3 << LANE_CNT_SHIFT)
#define LINK_TRAINING_EN BIT(6)
#define LEGACY_INTA BIT(28)
#define LEGACY_INTB BIT(29)
#define LEGACY_INTC BIT(30)
#define LEGACY_INTD BIT(31)
#define PCIE_CORE_CTRL1_REG (CONTROL_BASE_ADDR + 0x4)
#define HOT_RESET_GEN BIT(0)
#define PCIE_CORE_CTRL2_REG (CONTROL_BASE_ADDR + 0x8)
#define PCIE_CORE_CTRL2_RESERVED 0x7
#define PCIE_CORE_CTRL2_TD_ENABLE BIT(4)
#define PCIE_CORE_CTRL2_STRICT_ORDER_ENABLE BIT(5)
#define PCIE_CORE_CTRL2_OB_WIN_ENABLE BIT(6)
#define PCIE_CORE_CTRL2_MSI_ENABLE BIT(10)
#define PCIE_CORE_REF_CLK_REG (CONTROL_BASE_ADDR + 0x14)
#define PCIE_CORE_REF_CLK_TX_ENABLE BIT(1)
#define PCIE_CORE_REF_CLK_RX_ENABLE BIT(2)
#define PCIE_MSG_LOG_REG (CONTROL_BASE_ADDR + 0x30)
#define PCIE_ISR0_REG (CONTROL_BASE_ADDR + 0x40)
#define PCIE_MSG_PM_PME_MASK BIT(7)
#define PCIE_ISR0_MASK_REG (CONTROL_BASE_ADDR + 0x44)
#define PCIE_ISR0_MSI_INT_PENDING BIT(24)
#define PCIE_ISR0_CORR_ERR BIT(11)
#define PCIE_ISR0_NFAT_ERR BIT(12)
#define PCIE_ISR0_FAT_ERR BIT(13)
#define PCIE_ISR0_ERR_MASK GENMASK(13, 11)
#define PCIE_ISR0_INTX_ASSERT(val) BIT(16 + (val))
#define PCIE_ISR0_INTX_DEASSERT(val) BIT(20 + (val))
#define PCIE_ISR0_ALL_MASK GENMASK(31, 0)
#define PCIE_ISR1_REG (CONTROL_BASE_ADDR + 0x48)
#define PCIE_ISR1_MASK_REG (CONTROL_BASE_ADDR + 0x4C)
#define PCIE_ISR1_POWER_STATE_CHANGE BIT(4)
#define PCIE_ISR1_FLUSH BIT(5)
#define PCIE_ISR1_INTX_ASSERT(val) BIT(8 + (val))
#define PCIE_ISR1_ALL_MASK GENMASK(31, 0)
#define PCIE_MSI_ADDR_LOW_REG (CONTROL_BASE_ADDR + 0x50)
#define PCIE_MSI_ADDR_HIGH_REG (CONTROL_BASE_ADDR + 0x54)
#define PCIE_MSI_STATUS_REG (CONTROL_BASE_ADDR + 0x58)
#define PCIE_MSI_MASK_REG (CONTROL_BASE_ADDR + 0x5C)
#define PCIE_MSI_ALL_MASK GENMASK(31, 0)
#define PCIE_MSI_PAYLOAD_REG (CONTROL_BASE_ADDR + 0x9C)
#define PCIE_MSI_DATA_MASK GENMASK(15, 0)
PCI: aardvark: Configure PCIe resources from 'ranges' DT property In commit 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration") was removed aardvark PCIe outbound window configuration and commit description said that was recommended solution by HW designers. But that commit completely removed support for configuring PCIe IO resources without removing PCIe IO 'ranges' from DTS files. After that commit PCIe IO space started to be treated as PCIe MEM space and accessing it just caused kernel crash. Moreover implementation of PCIe outbound windows prior that commit was incorrect. It completely ignored offset between CPU address and PCIe bus address and expected that in DTS is CPU address always same as PCIe bus address without doing any checks. Also it completely ignored size of every PCIe resource specified in 'ranges' DTS property and expected that every PCIe resource has size 128 MB (also for PCIe IO range). Again without any check. Apparently none of PCIe resource has in DTS specified size of 128 MB. So it was completely broken and thanks to how aardvark mask works, configuration was completely ignored. This patch reverts back support for PCIe outbound window configuration but implementation is a new without issues mentioned above. PCIe outbound window is required when DTS specify in 'ranges' property non-zero offset between CPU and PCIe address space. To address recommendation by HW designers as specified in commit description of 6df6ba974a55, set default outbound parameters as PCIe MEM access without translation and therefore for this PCIe 'ranges' it is not needed to configure PCIe outbound window. For PCIe IO space is needed to configure aardvark PCIe outbound window. This patch fixes kernel crash when trying to access PCIe IO space. Link: https://lore.kernel.org/r/20210624215546.4015-2-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Cc: stable@vger.kernel.org # 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration")
2021-06-24 21:55:45 +00:00
/* PCIe window configuration */
#define OB_WIN_BASE_ADDR 0x4c00
#define OB_WIN_BLOCK_SIZE 0x20
#define OB_WIN_COUNT 8
#define OB_WIN_REG_ADDR(win, offset) (OB_WIN_BASE_ADDR + \
OB_WIN_BLOCK_SIZE * (win) + \
(offset))
#define OB_WIN_MATCH_LS(win) OB_WIN_REG_ADDR(win, 0x00)
#define OB_WIN_ENABLE BIT(0)
#define OB_WIN_MATCH_MS(win) OB_WIN_REG_ADDR(win, 0x04)
#define OB_WIN_REMAP_LS(win) OB_WIN_REG_ADDR(win, 0x08)
#define OB_WIN_REMAP_MS(win) OB_WIN_REG_ADDR(win, 0x0c)
#define OB_WIN_MASK_LS(win) OB_WIN_REG_ADDR(win, 0x10)
#define OB_WIN_MASK_MS(win) OB_WIN_REG_ADDR(win, 0x14)
#define OB_WIN_ACTIONS(win) OB_WIN_REG_ADDR(win, 0x18)
#define OB_WIN_DEFAULT_ACTIONS (OB_WIN_ACTIONS(OB_WIN_COUNT-1) + 0x4)
#define OB_WIN_FUNC_NUM_MASK GENMASK(31, 24)
#define OB_WIN_FUNC_NUM_SHIFT 24
#define OB_WIN_FUNC_NUM_ENABLE BIT(23)
#define OB_WIN_BUS_NUM_BITS_MASK GENMASK(22, 20)
#define OB_WIN_BUS_NUM_BITS_SHIFT 20
#define OB_WIN_MSG_CODE_ENABLE BIT(22)
#define OB_WIN_MSG_CODE_MASK GENMASK(21, 14)
#define OB_WIN_MSG_CODE_SHIFT 14
#define OB_WIN_MSG_PAYLOAD_LEN BIT(12)
#define OB_WIN_ATTR_ENABLE BIT(11)
#define OB_WIN_ATTR_TC_MASK GENMASK(10, 8)
#define OB_WIN_ATTR_TC_SHIFT 8
#define OB_WIN_ATTR_RELAXED BIT(7)
#define OB_WIN_ATTR_NOSNOOP BIT(6)
#define OB_WIN_ATTR_POISON BIT(5)
#define OB_WIN_ATTR_IDO BIT(4)
#define OB_WIN_TYPE_MASK GENMASK(3, 0)
#define OB_WIN_TYPE_SHIFT 0
#define OB_WIN_TYPE_MEM 0x0
#define OB_WIN_TYPE_IO 0x4
#define OB_WIN_TYPE_CONFIG_TYPE0 0x8
#define OB_WIN_TYPE_CONFIG_TYPE1 0x9
#define OB_WIN_TYPE_MSG 0xc
/* LMI registers base address and register offsets */
#define LMI_BASE_ADDR 0x6000
#define CFG_REG (LMI_BASE_ADDR + 0x0)
#define LTSSM_SHIFT 24
#define LTSSM_MASK 0x3f
#define RC_BAR_CONFIG 0x300
/* LTSSM values in CFG_REG */
enum {
LTSSM_DETECT_QUIET = 0x0,
LTSSM_DETECT_ACTIVE = 0x1,
LTSSM_POLLING_ACTIVE = 0x2,
LTSSM_POLLING_COMPLIANCE = 0x3,
LTSSM_POLLING_CONFIGURATION = 0x4,
LTSSM_CONFIG_LINKWIDTH_START = 0x5,
LTSSM_CONFIG_LINKWIDTH_ACCEPT = 0x6,
LTSSM_CONFIG_LANENUM_ACCEPT = 0x7,
LTSSM_CONFIG_LANENUM_WAIT = 0x8,
LTSSM_CONFIG_COMPLETE = 0x9,
LTSSM_CONFIG_IDLE = 0xa,
LTSSM_RECOVERY_RCVR_LOCK = 0xb,
LTSSM_RECOVERY_SPEED = 0xc,
LTSSM_RECOVERY_RCVR_CFG = 0xd,
LTSSM_RECOVERY_IDLE = 0xe,
LTSSM_L0 = 0x10,
LTSSM_RX_L0S_ENTRY = 0x11,
LTSSM_RX_L0S_IDLE = 0x12,
LTSSM_RX_L0S_FTS = 0x13,
LTSSM_TX_L0S_ENTRY = 0x14,
LTSSM_TX_L0S_IDLE = 0x15,
LTSSM_TX_L0S_FTS = 0x16,
LTSSM_L1_ENTRY = 0x17,
LTSSM_L1_IDLE = 0x18,
LTSSM_L2_IDLE = 0x19,
LTSSM_L2_TRANSMIT_WAKE = 0x1a,
LTSSM_DISABLED = 0x20,
LTSSM_LOOPBACK_ENTRY_MASTER = 0x21,
LTSSM_LOOPBACK_ACTIVE_MASTER = 0x22,
LTSSM_LOOPBACK_EXIT_MASTER = 0x23,
LTSSM_LOOPBACK_ENTRY_SLAVE = 0x24,
LTSSM_LOOPBACK_ACTIVE_SLAVE = 0x25,
LTSSM_LOOPBACK_EXIT_SLAVE = 0x26,
LTSSM_HOT_RESET = 0x27,
LTSSM_RECOVERY_EQUALIZATION_PHASE0 = 0x28,
LTSSM_RECOVERY_EQUALIZATION_PHASE1 = 0x29,
LTSSM_RECOVERY_EQUALIZATION_PHASE2 = 0x2a,
LTSSM_RECOVERY_EQUALIZATION_PHASE3 = 0x2b,
};
#define VENDOR_ID_REG (LMI_BASE_ADDR + 0x44)
/* PCIe core controller registers */
#define CTRL_CORE_BASE_ADDR 0x18000
#define CTRL_CONFIG_REG (CTRL_CORE_BASE_ADDR + 0x0)
#define CTRL_MODE_SHIFT 0x0
#define CTRL_MODE_MASK 0x1
#define PCIE_CORE_MODE_DIRECT 0x0
#define PCIE_CORE_MODE_COMMAND 0x1
/* PCIe Central Interrupts Registers */
#define CENTRAL_INT_BASE_ADDR 0x1b000
#define HOST_CTRL_INT_STATUS_REG (CENTRAL_INT_BASE_ADDR + 0x0)
#define HOST_CTRL_INT_MASK_REG (CENTRAL_INT_BASE_ADDR + 0x4)
#define PCIE_IRQ_CMDQ_INT BIT(0)
#define PCIE_IRQ_MSI_STATUS_INT BIT(1)
#define PCIE_IRQ_CMD_SENT_DONE BIT(3)
#define PCIE_IRQ_DMA_INT BIT(4)
#define PCIE_IRQ_IB_DXFERDONE BIT(5)
#define PCIE_IRQ_OB_DXFERDONE BIT(6)
#define PCIE_IRQ_OB_RXFERDONE BIT(7)
#define PCIE_IRQ_COMPQ_INT BIT(12)
#define PCIE_IRQ_DIR_RD_DDR_DET BIT(13)
#define PCIE_IRQ_DIR_WR_DDR_DET BIT(14)
#define PCIE_IRQ_CORE_INT BIT(16)
#define PCIE_IRQ_CORE_INT_PIO BIT(17)
#define PCIE_IRQ_DPMU_INT BIT(18)
#define PCIE_IRQ_PCIE_MIS_INT BIT(19)
#define PCIE_IRQ_MSI_INT1_DET BIT(20)
#define PCIE_IRQ_MSI_INT2_DET BIT(21)
#define PCIE_IRQ_RC_DBELL_DET BIT(22)
#define PCIE_IRQ_EP_STATUS BIT(23)
#define PCIE_IRQ_ALL_MASK GENMASK(31, 0)
#define PCIE_IRQ_ENABLE_INTS_MASK PCIE_IRQ_CORE_INT
/* Transaction types */
#define PCIE_CONFIG_RD_TYPE0 0x8
#define PCIE_CONFIG_RD_TYPE1 0x9
#define PCIE_CONFIG_WR_TYPE0 0xa
#define PCIE_CONFIG_WR_TYPE1 0xb
PCI: aardvark: Increase polling delay to 1.5s while waiting for PIO response Measurements in different conditions showed that aardvark hardware PIO response can take up to 1.44s. Increase wait timeout from 1ms to 1.5s to ensure that we do not miss responses from hardware. After 1.44s hardware returns errors (e.g. Completer abort). The previous two patches fixed checking for PIO status, so now we can use it to also catch errors which are reported by hardware after 1.44s. After applying this patch, kernel can detect and print PIO errors to dmesg: [ 6.879999] advk-pcie d0070000.pcie: Non-posted PIO Response Status: CA, 0xe00 @ 0x100004 [ 6.896436] advk-pcie d0070000.pcie: Posted PIO Response Status: COMP_ERR, 0x804 @ 0x100004 [ 6.913049] advk-pcie d0070000.pcie: Posted PIO Response Status: COMP_ERR, 0x804 @ 0x100010 [ 6.929663] advk-pcie d0070000.pcie: Non-posted PIO Response Status: CA, 0xe00 @ 0x100010 [ 6.953558] advk-pcie d0070000.pcie: Posted PIO Response Status: COMP_ERR, 0x804 @ 0x100014 [ 6.970170] advk-pcie d0070000.pcie: Non-posted PIO Response Status: CA, 0xe00 @ 0x100014 [ 6.994328] advk-pcie d0070000.pcie: Posted PIO Response Status: COMP_ERR, 0x804 @ 0x100004 Without this patch kernel prints only a generic error to dmesg: [ 5.246847] advk-pcie d0070000.pcie: config read/write timed out Link: https://lore.kernel.org/r/20210722144041.12661-3-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Reviewed-by: Marek Behún <kabel@kernel.org> Cc: stable@vger.kernel.org # 7fbcb5da811b ("PCI: aardvark: Don't rely on jiffies while holding spinlock")
2021-07-22 14:40:39 +00:00
#define PIO_RETRY_CNT 750000 /* 1.5 s */
#define PIO_RETRY_DELAY 2 /* 2 us*/
#define LINK_WAIT_MAX_RETRIES 10
#define LINK_WAIT_USLEEP_MIN 90000
#define LINK_WAIT_USLEEP_MAX 100000
#define RETRAIN_WAIT_MAX_RETRIES 10
#define RETRAIN_WAIT_USLEEP_US 2000
#define MSI_IRQ_NUM 32
#define CFG_RD_CRS_VAL 0xffff0001
struct advk_pcie {
struct platform_device *pdev;
void __iomem *base;
PCI: aardvark: Configure PCIe resources from 'ranges' DT property In commit 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration") was removed aardvark PCIe outbound window configuration and commit description said that was recommended solution by HW designers. But that commit completely removed support for configuring PCIe IO resources without removing PCIe IO 'ranges' from DTS files. After that commit PCIe IO space started to be treated as PCIe MEM space and accessing it just caused kernel crash. Moreover implementation of PCIe outbound windows prior that commit was incorrect. It completely ignored offset between CPU address and PCIe bus address and expected that in DTS is CPU address always same as PCIe bus address without doing any checks. Also it completely ignored size of every PCIe resource specified in 'ranges' DTS property and expected that every PCIe resource has size 128 MB (also for PCIe IO range). Again without any check. Apparently none of PCIe resource has in DTS specified size of 128 MB. So it was completely broken and thanks to how aardvark mask works, configuration was completely ignored. This patch reverts back support for PCIe outbound window configuration but implementation is a new without issues mentioned above. PCIe outbound window is required when DTS specify in 'ranges' property non-zero offset between CPU and PCIe address space. To address recommendation by HW designers as specified in commit description of 6df6ba974a55, set default outbound parameters as PCIe MEM access without translation and therefore for this PCIe 'ranges' it is not needed to configure PCIe outbound window. For PCIe IO space is needed to configure aardvark PCIe outbound window. This patch fixes kernel crash when trying to access PCIe IO space. Link: https://lore.kernel.org/r/20210624215546.4015-2-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Cc: stable@vger.kernel.org # 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration")
2021-06-24 21:55:45 +00:00
struct {
phys_addr_t match;
phys_addr_t remap;
phys_addr_t mask;
u32 actions;
} wins[OB_WIN_COUNT];
u8 wins_count;
struct irq_domain *rp_irq_domain;
struct irq_domain *irq_domain;
struct irq_chip irq_chip;
raw_spinlock_t irq_lock;
struct irq_domain *msi_domain;
struct irq_domain *msi_inner_domain;
raw_spinlock_t msi_irq_lock;
DECLARE_BITMAP(msi_used, MSI_IRQ_NUM);
struct mutex msi_used_lock;
int link_gen;
struct pci_bridge_emul bridge;
struct gpio_desc *reset_gpio;
struct phy *phy;
};
static inline void advk_writel(struct advk_pcie *pcie, u32 val, u64 reg)
{
writel(val, pcie->base + reg);
}
static inline u32 advk_readl(struct advk_pcie *pcie, u64 reg)
{
return readl(pcie->base + reg);
}
static u8 advk_pcie_ltssm_state(struct advk_pcie *pcie)
{
u32 val;
u8 ltssm_state;
val = advk_readl(pcie, CFG_REG);
ltssm_state = (val >> LTSSM_SHIFT) & LTSSM_MASK;
return ltssm_state;
}
static inline bool advk_pcie_link_up(struct advk_pcie *pcie)
{
/* check if LTSSM is in normal operation - some L* state */
u8 ltssm_state = advk_pcie_ltssm_state(pcie);
return ltssm_state >= LTSSM_L0 && ltssm_state < LTSSM_DISABLED;
}
static inline bool advk_pcie_link_active(struct advk_pcie *pcie)
{
/*
* According to PCIe Base specification 3.0, Table 4-14: Link
* Status Mapped to the LTSSM, and 4.2.6.3.6 Configuration.Idle
* is Link Up mapped to LTSSM Configuration.Idle, Recovery, L0,
* L0s, L1 and L2 states. And according to 3.2.1. Data Link
* Control and Management State Machine Rules is DL Up status
* reported in DL Active state.
*/
u8 ltssm_state = advk_pcie_ltssm_state(pcie);
return ltssm_state >= LTSSM_CONFIG_IDLE && ltssm_state < LTSSM_DISABLED;
}
static inline bool advk_pcie_link_training(struct advk_pcie *pcie)
{
/*
* According to PCIe Base specification 3.0, Table 4-14: Link
* Status Mapped to the LTSSM is Link Training mapped to LTSSM
* Configuration and Recovery states.
*/
u8 ltssm_state = advk_pcie_ltssm_state(pcie);
return ((ltssm_state >= LTSSM_CONFIG_LINKWIDTH_START &&
ltssm_state < LTSSM_L0) ||
(ltssm_state >= LTSSM_RECOVERY_EQUALIZATION_PHASE0 &&
ltssm_state <= LTSSM_RECOVERY_EQUALIZATION_PHASE3));
}
static int advk_pcie_wait_for_link(struct advk_pcie *pcie)
{
int retries;
/* check if the link is up or not */
for (retries = 0; retries < LINK_WAIT_MAX_RETRIES; retries++) {
if (advk_pcie_link_up(pcie))
return 0;
usleep_range(LINK_WAIT_USLEEP_MIN, LINK_WAIT_USLEEP_MAX);
}
return -ETIMEDOUT;
}
static void advk_pcie_wait_for_retrain(struct advk_pcie *pcie)
{
size_t retries;
for (retries = 0; retries < RETRAIN_WAIT_MAX_RETRIES; ++retries) {
if (advk_pcie_link_training(pcie))
break;
udelay(RETRAIN_WAIT_USLEEP_US);
}
}
static void advk_pcie_issue_perst(struct advk_pcie *pcie)
{
if (!pcie->reset_gpio)
return;
/* 10ms delay is needed for some cards */
dev_info(&pcie->pdev->dev, "issuing PERST via reset GPIO for 10ms\n");
gpiod_set_value_cansleep(pcie->reset_gpio, 1);
usleep_range(10000, 11000);
gpiod_set_value_cansleep(pcie->reset_gpio, 0);
}
PCI: aardvark: Fix link training Fix multiple link training issues in aardvark driver. The main reason of these issues was misunderstanding of what certain registers do, since their names and comments were misleading: before commit 96be36dbffac ("PCI: aardvark: Replace custom macros by standard linux/pci_regs.h macros"), the pci-aardvark.c driver used custom macros for accessing standard PCIe Root Bridge registers, and misleading comments did not help to understand what the code was really doing. After doing more tests and experiments I've come to the conclusion that the SPEED_GEN register in aardvark sets the PCIe revision / generation compliance and forces maximal link speed. Both GEN3 and GEN2 values set the read-only PCI_EXP_FLAGS_VERS bits (PCIe capabilities version of Root Bridge) to value 2, while GEN1 value sets PCI_EXP_FLAGS_VERS to 1, which matches with PCI Express specifications revisions 3, 2 and 1 respectively. Changing SPEED_GEN also sets the read-only bits PCI_EXP_LNKCAP_SLS and PCI_EXP_LNKCAP2_SLS to corresponding speed. (Note that PCI Express rev 1 specification does not define PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers and when SPEED_GEN is set to GEN1 (which also sets PCI_EXP_FLAGS_VERS set to 1), lspci cannot access PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers.) Changing PCIe link speed can be done via PCI_EXP_LNKCTL2_TLS bits of PCI_EXP_LNKCTL2 register. Armada 3700 Functional Specifications says that the default value of PCI_EXP_LNKCTL2_TLS is based on SPEED_GEN value, but tests showed that the default value is always 8.0 GT/s, independently of speed set by SPEED_GEN. So after setting SPEED_GEN, we must also set value in PCI_EXP_LNKCTL2 register via PCI_EXP_LNKCTL2_TLS bits. Triggering PCI_EXP_LNKCTL_RL bit immediately after setting LINK_TRAINING_EN bit actually doesn't do anything. Tests have shown that a delay is needed after enabling LINK_TRAINING_EN bit. As triggering PCI_EXP_LNKCTL_RL currently does nothing, remove it. Commit 43fc679ced18 ("PCI: aardvark: Improve link training") introduced code which sets SPEED_GEN register based on negotiated link speed from PCI_EXP_LNKSTA_CLS bits of PCI_EXP_LNKSTA register. This code was added to fix detection of Compex WLE900VX (Atheros QCA9880) WiFi GEN1 PCIe cards, as otherwise these cards were "invisible" on PCIe bus (probably because they crashed). But apparently more people reported the same issues with these cards also with other PCIe controllers [1] and I was able to reproduce this issue also with other "noname" WiFi cards based on Atheros QCA9890 chip (with the same PCI vendor/device ids as Atheros QCA9880). So this is not an issue in aardvark but rather an issue in Atheros QCA98xx chips. Also, this issue only exists if the kernel is compiled with PCIe ASPM support, and a generic workaround for this is to change PCIe Bridge to 2.5 GT/s link speed via PCI_EXP_LNKCTL2_TLS_2_5GT bits in PCI_EXP_LNKCTL2 register [2], before triggering PCI_EXP_LNKCTL_RL bit. This workaround also works when SPEED_GEN is set to value GEN2 (5 GT/s). So remove this hack completely in the aardvark driver and always set SPEED_GEN to value from 'max-link-speed' DT property. Fix for Atheros QCA98xx chips is handled separately by patch [2]. These two things (code for triggering PCI_EXP_LNKCTL_RL bit and changing SPEED_GEN value) also explain why commit 6964494582f5 ("PCI: aardvark: Train link immediately after enabling training") somehow fixed detection of those problematic Compex cards with Atheros chips: if triggering link retraining (via PCI_EXP_LNKCTL_RL bit) was done immediately after enabling link training (via LINK_TRAINING_EN), it did nothing. If there was a specific delay, aardvark HW already initialized PCIe link and therefore triggering link retraining caused the above issue. Compex cards triggered link down event and disappeared from the PCIe bus. Commit f4c7d053d7f7 ("PCI: aardvark: Wait for endpoint to be ready before training link") added 100ms sleep before calling 'Start link training' command and explained that it is a requirement of PCI Express specification. But the code after this 100ms sleep was not doing 'Start link training', rather it triggered PCI_EXP_LNKCTL_RL bit via PCIe Root Bridge to put link into Recovery state. The required delay after fundamental reset is already done in function advk_pcie_wait_for_link() which also checks whether PCIe link is up. So after removing the code which triggers PCI_EXP_LNKCTL_RL bit on PCIe Root Bridge, there is no need to wait 100ms again. Remove the extra msleep() call and update comment about the delay required by the PCI Express specification. According to Marvell Armada 3700 Functional Specifications, Link training should be enabled via aardvark register LINK_TRAINING_EN after selecting PCIe generation and x1 lane. There is no need to disable it prior resetting card via PERST# signal. This disabling code was introduced in commit 5169a9851daa ("PCI: aardvark: Issue PERST via GPIO") as a workaround for some Atheros cards. It turns out that this also is Atheros specific issue and affects any PCIe controller, not only aardvark. Moreover this Atheros issue was triggered by juggling with PCI_EXP_LNKCTL_RL, LINK_TRAINING_EN and SPEED_GEN bits interleaved with sleeps. Now, after removing triggering PCI_EXP_LNKCTL_RL, there is no need to explicitly disable LINK_TRAINING_EN bit. So remove this code too. The problematic Compex cards described in previous git commits are correctly detected in advk_pcie_train_link() function even after applying all these changes. Note that with this patch, and also prior this patch, some NVMe disks which support PCIe GEN3 with 8 GT/s speed are negotiated only at the lowest link speed 2.5 GT/s, independently of SPEED_GEN value. After manually triggering PCI_EXP_LNKCTL_RL bit (e.g. from userspace via setpci), these NVMe disks change link speed to 5 GT/s when SPEED_GEN was configured to GEN2. This issue first needs to be properly investigated. I will send a fix in the future. On the other hand, some other GEN2 PCIe cards with 5 GT/s speed are autonomously by HW autonegotiated at full 5 GT/s speed without need of any software interaction. Armada 3700 Functional Specifications describes the following steps for link training: set SPEED_GEN to GEN2, enable LINK_TRAINING_EN, poll until link training is complete, trigger PCI_EXP_LNKCTL_RL, poll until signal rate is 5 GT/s, poll until link training is complete, enable ASPM L0s. The requirement for triggering PCI_EXP_LNKCTL_RL can be explained by the need to achieve 5 GT/s speed (as changing link speed is done by throw to recovery state entered by PCI_EXP_LNKCTL_RL) or maybe as a part of enabling ASPM L0s (but in this case ASPM L0s should have been enabled prior PCI_EXP_LNKCTL_RL). It is unknown why the original pci-aardvark.c driver was triggering PCI_EXP_LNKCTL_RL bit before waiting for the link to be up. This does not align with neither PCIe base specifications nor with Armada 3700 Functional Specification. (Note that in older versions of aardvark, this bit was called incorrectly PCIE_CORE_LINK_TRAINING, so this may be the reason.) It is also unknown why Armada 3700 Functional Specification says that it is needed to trigger PCI_EXP_LNKCTL_RL for GEN2 mode, as according to PCIe base specification 5 GT/s speed negotiation is supposed to be entirely autonomous, even if initial speed is 2.5 GT/s. [1] - https://lore.kernel.org/linux-pci/87h7l8axqp.fsf@toke.dk/ [2] - https://lore.kernel.org/linux-pci/20210326124326.21163-1-pali@kernel.org/ Link: https://lore.kernel.org/r/20211005180952.6812-12-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Reviewed-by: Marek Behún <kabel@kernel.org>
2021-10-05 18:09:50 +00:00
static void advk_pcie_train_link(struct advk_pcie *pcie)
{
PCI: aardvark: Fix link training Fix multiple link training issues in aardvark driver. The main reason of these issues was misunderstanding of what certain registers do, since their names and comments were misleading: before commit 96be36dbffac ("PCI: aardvark: Replace custom macros by standard linux/pci_regs.h macros"), the pci-aardvark.c driver used custom macros for accessing standard PCIe Root Bridge registers, and misleading comments did not help to understand what the code was really doing. After doing more tests and experiments I've come to the conclusion that the SPEED_GEN register in aardvark sets the PCIe revision / generation compliance and forces maximal link speed. Both GEN3 and GEN2 values set the read-only PCI_EXP_FLAGS_VERS bits (PCIe capabilities version of Root Bridge) to value 2, while GEN1 value sets PCI_EXP_FLAGS_VERS to 1, which matches with PCI Express specifications revisions 3, 2 and 1 respectively. Changing SPEED_GEN also sets the read-only bits PCI_EXP_LNKCAP_SLS and PCI_EXP_LNKCAP2_SLS to corresponding speed. (Note that PCI Express rev 1 specification does not define PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers and when SPEED_GEN is set to GEN1 (which also sets PCI_EXP_FLAGS_VERS set to 1), lspci cannot access PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers.) Changing PCIe link speed can be done via PCI_EXP_LNKCTL2_TLS bits of PCI_EXP_LNKCTL2 register. Armada 3700 Functional Specifications says that the default value of PCI_EXP_LNKCTL2_TLS is based on SPEED_GEN value, but tests showed that the default value is always 8.0 GT/s, independently of speed set by SPEED_GEN. So after setting SPEED_GEN, we must also set value in PCI_EXP_LNKCTL2 register via PCI_EXP_LNKCTL2_TLS bits. Triggering PCI_EXP_LNKCTL_RL bit immediately after setting LINK_TRAINING_EN bit actually doesn't do anything. Tests have shown that a delay is needed after enabling LINK_TRAINING_EN bit. As triggering PCI_EXP_LNKCTL_RL currently does nothing, remove it. Commit 43fc679ced18 ("PCI: aardvark: Improve link training") introduced code which sets SPEED_GEN register based on negotiated link speed from PCI_EXP_LNKSTA_CLS bits of PCI_EXP_LNKSTA register. This code was added to fix detection of Compex WLE900VX (Atheros QCA9880) WiFi GEN1 PCIe cards, as otherwise these cards were "invisible" on PCIe bus (probably because they crashed). But apparently more people reported the same issues with these cards also with other PCIe controllers [1] and I was able to reproduce this issue also with other "noname" WiFi cards based on Atheros QCA9890 chip (with the same PCI vendor/device ids as Atheros QCA9880). So this is not an issue in aardvark but rather an issue in Atheros QCA98xx chips. Also, this issue only exists if the kernel is compiled with PCIe ASPM support, and a generic workaround for this is to change PCIe Bridge to 2.5 GT/s link speed via PCI_EXP_LNKCTL2_TLS_2_5GT bits in PCI_EXP_LNKCTL2 register [2], before triggering PCI_EXP_LNKCTL_RL bit. This workaround also works when SPEED_GEN is set to value GEN2 (5 GT/s). So remove this hack completely in the aardvark driver and always set SPEED_GEN to value from 'max-link-speed' DT property. Fix for Atheros QCA98xx chips is handled separately by patch [2]. These two things (code for triggering PCI_EXP_LNKCTL_RL bit and changing SPEED_GEN value) also explain why commit 6964494582f5 ("PCI: aardvark: Train link immediately after enabling training") somehow fixed detection of those problematic Compex cards with Atheros chips: if triggering link retraining (via PCI_EXP_LNKCTL_RL bit) was done immediately after enabling link training (via LINK_TRAINING_EN), it did nothing. If there was a specific delay, aardvark HW already initialized PCIe link and therefore triggering link retraining caused the above issue. Compex cards triggered link down event and disappeared from the PCIe bus. Commit f4c7d053d7f7 ("PCI: aardvark: Wait for endpoint to be ready before training link") added 100ms sleep before calling 'Start link training' command and explained that it is a requirement of PCI Express specification. But the code after this 100ms sleep was not doing 'Start link training', rather it triggered PCI_EXP_LNKCTL_RL bit via PCIe Root Bridge to put link into Recovery state. The required delay after fundamental reset is already done in function advk_pcie_wait_for_link() which also checks whether PCIe link is up. So after removing the code which triggers PCI_EXP_LNKCTL_RL bit on PCIe Root Bridge, there is no need to wait 100ms again. Remove the extra msleep() call and update comment about the delay required by the PCI Express specification. According to Marvell Armada 3700 Functional Specifications, Link training should be enabled via aardvark register LINK_TRAINING_EN after selecting PCIe generation and x1 lane. There is no need to disable it prior resetting card via PERST# signal. This disabling code was introduced in commit 5169a9851daa ("PCI: aardvark: Issue PERST via GPIO") as a workaround for some Atheros cards. It turns out that this also is Atheros specific issue and affects any PCIe controller, not only aardvark. Moreover this Atheros issue was triggered by juggling with PCI_EXP_LNKCTL_RL, LINK_TRAINING_EN and SPEED_GEN bits interleaved with sleeps. Now, after removing triggering PCI_EXP_LNKCTL_RL, there is no need to explicitly disable LINK_TRAINING_EN bit. So remove this code too. The problematic Compex cards described in previous git commits are correctly detected in advk_pcie_train_link() function even after applying all these changes. Note that with this patch, and also prior this patch, some NVMe disks which support PCIe GEN3 with 8 GT/s speed are negotiated only at the lowest link speed 2.5 GT/s, independently of SPEED_GEN value. After manually triggering PCI_EXP_LNKCTL_RL bit (e.g. from userspace via setpci), these NVMe disks change link speed to 5 GT/s when SPEED_GEN was configured to GEN2. This issue first needs to be properly investigated. I will send a fix in the future. On the other hand, some other GEN2 PCIe cards with 5 GT/s speed are autonomously by HW autonegotiated at full 5 GT/s speed without need of any software interaction. Armada 3700 Functional Specifications describes the following steps for link training: set SPEED_GEN to GEN2, enable LINK_TRAINING_EN, poll until link training is complete, trigger PCI_EXP_LNKCTL_RL, poll until signal rate is 5 GT/s, poll until link training is complete, enable ASPM L0s. The requirement for triggering PCI_EXP_LNKCTL_RL can be explained by the need to achieve 5 GT/s speed (as changing link speed is done by throw to recovery state entered by PCI_EXP_LNKCTL_RL) or maybe as a part of enabling ASPM L0s (but in this case ASPM L0s should have been enabled prior PCI_EXP_LNKCTL_RL). It is unknown why the original pci-aardvark.c driver was triggering PCI_EXP_LNKCTL_RL bit before waiting for the link to be up. This does not align with neither PCIe base specifications nor with Armada 3700 Functional Specification. (Note that in older versions of aardvark, this bit was called incorrectly PCIE_CORE_LINK_TRAINING, so this may be the reason.) It is also unknown why Armada 3700 Functional Specification says that it is needed to trigger PCI_EXP_LNKCTL_RL for GEN2 mode, as according to PCIe base specification 5 GT/s speed negotiation is supposed to be entirely autonomous, even if initial speed is 2.5 GT/s. [1] - https://lore.kernel.org/linux-pci/87h7l8axqp.fsf@toke.dk/ [2] - https://lore.kernel.org/linux-pci/20210326124326.21163-1-pali@kernel.org/ Link: https://lore.kernel.org/r/20211005180952.6812-12-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Reviewed-by: Marek Behún <kabel@kernel.org>
2021-10-05 18:09:50 +00:00
struct device *dev = &pcie->pdev->dev;
u32 reg;
PCI: aardvark: Fix link training Fix multiple link training issues in aardvark driver. The main reason of these issues was misunderstanding of what certain registers do, since their names and comments were misleading: before commit 96be36dbffac ("PCI: aardvark: Replace custom macros by standard linux/pci_regs.h macros"), the pci-aardvark.c driver used custom macros for accessing standard PCIe Root Bridge registers, and misleading comments did not help to understand what the code was really doing. After doing more tests and experiments I've come to the conclusion that the SPEED_GEN register in aardvark sets the PCIe revision / generation compliance and forces maximal link speed. Both GEN3 and GEN2 values set the read-only PCI_EXP_FLAGS_VERS bits (PCIe capabilities version of Root Bridge) to value 2, while GEN1 value sets PCI_EXP_FLAGS_VERS to 1, which matches with PCI Express specifications revisions 3, 2 and 1 respectively. Changing SPEED_GEN also sets the read-only bits PCI_EXP_LNKCAP_SLS and PCI_EXP_LNKCAP2_SLS to corresponding speed. (Note that PCI Express rev 1 specification does not define PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers and when SPEED_GEN is set to GEN1 (which also sets PCI_EXP_FLAGS_VERS set to 1), lspci cannot access PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers.) Changing PCIe link speed can be done via PCI_EXP_LNKCTL2_TLS bits of PCI_EXP_LNKCTL2 register. Armada 3700 Functional Specifications says that the default value of PCI_EXP_LNKCTL2_TLS is based on SPEED_GEN value, but tests showed that the default value is always 8.0 GT/s, independently of speed set by SPEED_GEN. So after setting SPEED_GEN, we must also set value in PCI_EXP_LNKCTL2 register via PCI_EXP_LNKCTL2_TLS bits. Triggering PCI_EXP_LNKCTL_RL bit immediately after setting LINK_TRAINING_EN bit actually doesn't do anything. Tests have shown that a delay is needed after enabling LINK_TRAINING_EN bit. As triggering PCI_EXP_LNKCTL_RL currently does nothing, remove it. Commit 43fc679ced18 ("PCI: aardvark: Improve link training") introduced code which sets SPEED_GEN register based on negotiated link speed from PCI_EXP_LNKSTA_CLS bits of PCI_EXP_LNKSTA register. This code was added to fix detection of Compex WLE900VX (Atheros QCA9880) WiFi GEN1 PCIe cards, as otherwise these cards were "invisible" on PCIe bus (probably because they crashed). But apparently more people reported the same issues with these cards also with other PCIe controllers [1] and I was able to reproduce this issue also with other "noname" WiFi cards based on Atheros QCA9890 chip (with the same PCI vendor/device ids as Atheros QCA9880). So this is not an issue in aardvark but rather an issue in Atheros QCA98xx chips. Also, this issue only exists if the kernel is compiled with PCIe ASPM support, and a generic workaround for this is to change PCIe Bridge to 2.5 GT/s link speed via PCI_EXP_LNKCTL2_TLS_2_5GT bits in PCI_EXP_LNKCTL2 register [2], before triggering PCI_EXP_LNKCTL_RL bit. This workaround also works when SPEED_GEN is set to value GEN2 (5 GT/s). So remove this hack completely in the aardvark driver and always set SPEED_GEN to value from 'max-link-speed' DT property. Fix for Atheros QCA98xx chips is handled separately by patch [2]. These two things (code for triggering PCI_EXP_LNKCTL_RL bit and changing SPEED_GEN value) also explain why commit 6964494582f5 ("PCI: aardvark: Train link immediately after enabling training") somehow fixed detection of those problematic Compex cards with Atheros chips: if triggering link retraining (via PCI_EXP_LNKCTL_RL bit) was done immediately after enabling link training (via LINK_TRAINING_EN), it did nothing. If there was a specific delay, aardvark HW already initialized PCIe link and therefore triggering link retraining caused the above issue. Compex cards triggered link down event and disappeared from the PCIe bus. Commit f4c7d053d7f7 ("PCI: aardvark: Wait for endpoint to be ready before training link") added 100ms sleep before calling 'Start link training' command and explained that it is a requirement of PCI Express specification. But the code after this 100ms sleep was not doing 'Start link training', rather it triggered PCI_EXP_LNKCTL_RL bit via PCIe Root Bridge to put link into Recovery state. The required delay after fundamental reset is already done in function advk_pcie_wait_for_link() which also checks whether PCIe link is up. So after removing the code which triggers PCI_EXP_LNKCTL_RL bit on PCIe Root Bridge, there is no need to wait 100ms again. Remove the extra msleep() call and update comment about the delay required by the PCI Express specification. According to Marvell Armada 3700 Functional Specifications, Link training should be enabled via aardvark register LINK_TRAINING_EN after selecting PCIe generation and x1 lane. There is no need to disable it prior resetting card via PERST# signal. This disabling code was introduced in commit 5169a9851daa ("PCI: aardvark: Issue PERST via GPIO") as a workaround for some Atheros cards. It turns out that this also is Atheros specific issue and affects any PCIe controller, not only aardvark. Moreover this Atheros issue was triggered by juggling with PCI_EXP_LNKCTL_RL, LINK_TRAINING_EN and SPEED_GEN bits interleaved with sleeps. Now, after removing triggering PCI_EXP_LNKCTL_RL, there is no need to explicitly disable LINK_TRAINING_EN bit. So remove this code too. The problematic Compex cards described in previous git commits are correctly detected in advk_pcie_train_link() function even after applying all these changes. Note that with this patch, and also prior this patch, some NVMe disks which support PCIe GEN3 with 8 GT/s speed are negotiated only at the lowest link speed 2.5 GT/s, independently of SPEED_GEN value. After manually triggering PCI_EXP_LNKCTL_RL bit (e.g. from userspace via setpci), these NVMe disks change link speed to 5 GT/s when SPEED_GEN was configured to GEN2. This issue first needs to be properly investigated. I will send a fix in the future. On the other hand, some other GEN2 PCIe cards with 5 GT/s speed are autonomously by HW autonegotiated at full 5 GT/s speed without need of any software interaction. Armada 3700 Functional Specifications describes the following steps for link training: set SPEED_GEN to GEN2, enable LINK_TRAINING_EN, poll until link training is complete, trigger PCI_EXP_LNKCTL_RL, poll until signal rate is 5 GT/s, poll until link training is complete, enable ASPM L0s. The requirement for triggering PCI_EXP_LNKCTL_RL can be explained by the need to achieve 5 GT/s speed (as changing link speed is done by throw to recovery state entered by PCI_EXP_LNKCTL_RL) or maybe as a part of enabling ASPM L0s (but in this case ASPM L0s should have been enabled prior PCI_EXP_LNKCTL_RL). It is unknown why the original pci-aardvark.c driver was triggering PCI_EXP_LNKCTL_RL bit before waiting for the link to be up. This does not align with neither PCIe base specifications nor with Armada 3700 Functional Specification. (Note that in older versions of aardvark, this bit was called incorrectly PCIE_CORE_LINK_TRAINING, so this may be the reason.) It is also unknown why Armada 3700 Functional Specification says that it is needed to trigger PCI_EXP_LNKCTL_RL for GEN2 mode, as according to PCIe base specification 5 GT/s speed negotiation is supposed to be entirely autonomous, even if initial speed is 2.5 GT/s. [1] - https://lore.kernel.org/linux-pci/87h7l8axqp.fsf@toke.dk/ [2] - https://lore.kernel.org/linux-pci/20210326124326.21163-1-pali@kernel.org/ Link: https://lore.kernel.org/r/20211005180952.6812-12-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Reviewed-by: Marek Behún <kabel@kernel.org>
2021-10-05 18:09:50 +00:00
int ret;
PCI: aardvark: Fix link training Fix multiple link training issues in aardvark driver. The main reason of these issues was misunderstanding of what certain registers do, since their names and comments were misleading: before commit 96be36dbffac ("PCI: aardvark: Replace custom macros by standard linux/pci_regs.h macros"), the pci-aardvark.c driver used custom macros for accessing standard PCIe Root Bridge registers, and misleading comments did not help to understand what the code was really doing. After doing more tests and experiments I've come to the conclusion that the SPEED_GEN register in aardvark sets the PCIe revision / generation compliance and forces maximal link speed. Both GEN3 and GEN2 values set the read-only PCI_EXP_FLAGS_VERS bits (PCIe capabilities version of Root Bridge) to value 2, while GEN1 value sets PCI_EXP_FLAGS_VERS to 1, which matches with PCI Express specifications revisions 3, 2 and 1 respectively. Changing SPEED_GEN also sets the read-only bits PCI_EXP_LNKCAP_SLS and PCI_EXP_LNKCAP2_SLS to corresponding speed. (Note that PCI Express rev 1 specification does not define PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers and when SPEED_GEN is set to GEN1 (which also sets PCI_EXP_FLAGS_VERS set to 1), lspci cannot access PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers.) Changing PCIe link speed can be done via PCI_EXP_LNKCTL2_TLS bits of PCI_EXP_LNKCTL2 register. Armada 3700 Functional Specifications says that the default value of PCI_EXP_LNKCTL2_TLS is based on SPEED_GEN value, but tests showed that the default value is always 8.0 GT/s, independently of speed set by SPEED_GEN. So after setting SPEED_GEN, we must also set value in PCI_EXP_LNKCTL2 register via PCI_EXP_LNKCTL2_TLS bits. Triggering PCI_EXP_LNKCTL_RL bit immediately after setting LINK_TRAINING_EN bit actually doesn't do anything. Tests have shown that a delay is needed after enabling LINK_TRAINING_EN bit. As triggering PCI_EXP_LNKCTL_RL currently does nothing, remove it. Commit 43fc679ced18 ("PCI: aardvark: Improve link training") introduced code which sets SPEED_GEN register based on negotiated link speed from PCI_EXP_LNKSTA_CLS bits of PCI_EXP_LNKSTA register. This code was added to fix detection of Compex WLE900VX (Atheros QCA9880) WiFi GEN1 PCIe cards, as otherwise these cards were "invisible" on PCIe bus (probably because they crashed). But apparently more people reported the same issues with these cards also with other PCIe controllers [1] and I was able to reproduce this issue also with other "noname" WiFi cards based on Atheros QCA9890 chip (with the same PCI vendor/device ids as Atheros QCA9880). So this is not an issue in aardvark but rather an issue in Atheros QCA98xx chips. Also, this issue only exists if the kernel is compiled with PCIe ASPM support, and a generic workaround for this is to change PCIe Bridge to 2.5 GT/s link speed via PCI_EXP_LNKCTL2_TLS_2_5GT bits in PCI_EXP_LNKCTL2 register [2], before triggering PCI_EXP_LNKCTL_RL bit. This workaround also works when SPEED_GEN is set to value GEN2 (5 GT/s). So remove this hack completely in the aardvark driver and always set SPEED_GEN to value from 'max-link-speed' DT property. Fix for Atheros QCA98xx chips is handled separately by patch [2]. These two things (code for triggering PCI_EXP_LNKCTL_RL bit and changing SPEED_GEN value) also explain why commit 6964494582f5 ("PCI: aardvark: Train link immediately after enabling training") somehow fixed detection of those problematic Compex cards with Atheros chips: if triggering link retraining (via PCI_EXP_LNKCTL_RL bit) was done immediately after enabling link training (via LINK_TRAINING_EN), it did nothing. If there was a specific delay, aardvark HW already initialized PCIe link and therefore triggering link retraining caused the above issue. Compex cards triggered link down event and disappeared from the PCIe bus. Commit f4c7d053d7f7 ("PCI: aardvark: Wait for endpoint to be ready before training link") added 100ms sleep before calling 'Start link training' command and explained that it is a requirement of PCI Express specification. But the code after this 100ms sleep was not doing 'Start link training', rather it triggered PCI_EXP_LNKCTL_RL bit via PCIe Root Bridge to put link into Recovery state. The required delay after fundamental reset is already done in function advk_pcie_wait_for_link() which also checks whether PCIe link is up. So after removing the code which triggers PCI_EXP_LNKCTL_RL bit on PCIe Root Bridge, there is no need to wait 100ms again. Remove the extra msleep() call and update comment about the delay required by the PCI Express specification. According to Marvell Armada 3700 Functional Specifications, Link training should be enabled via aardvark register LINK_TRAINING_EN after selecting PCIe generation and x1 lane. There is no need to disable it prior resetting card via PERST# signal. This disabling code was introduced in commit 5169a9851daa ("PCI: aardvark: Issue PERST via GPIO") as a workaround for some Atheros cards. It turns out that this also is Atheros specific issue and affects any PCIe controller, not only aardvark. Moreover this Atheros issue was triggered by juggling with PCI_EXP_LNKCTL_RL, LINK_TRAINING_EN and SPEED_GEN bits interleaved with sleeps. Now, after removing triggering PCI_EXP_LNKCTL_RL, there is no need to explicitly disable LINK_TRAINING_EN bit. So remove this code too. The problematic Compex cards described in previous git commits are correctly detected in advk_pcie_train_link() function even after applying all these changes. Note that with this patch, and also prior this patch, some NVMe disks which support PCIe GEN3 with 8 GT/s speed are negotiated only at the lowest link speed 2.5 GT/s, independently of SPEED_GEN value. After manually triggering PCI_EXP_LNKCTL_RL bit (e.g. from userspace via setpci), these NVMe disks change link speed to 5 GT/s when SPEED_GEN was configured to GEN2. This issue first needs to be properly investigated. I will send a fix in the future. On the other hand, some other GEN2 PCIe cards with 5 GT/s speed are autonomously by HW autonegotiated at full 5 GT/s speed without need of any software interaction. Armada 3700 Functional Specifications describes the following steps for link training: set SPEED_GEN to GEN2, enable LINK_TRAINING_EN, poll until link training is complete, trigger PCI_EXP_LNKCTL_RL, poll until signal rate is 5 GT/s, poll until link training is complete, enable ASPM L0s. The requirement for triggering PCI_EXP_LNKCTL_RL can be explained by the need to achieve 5 GT/s speed (as changing link speed is done by throw to recovery state entered by PCI_EXP_LNKCTL_RL) or maybe as a part of enabling ASPM L0s (but in this case ASPM L0s should have been enabled prior PCI_EXP_LNKCTL_RL). It is unknown why the original pci-aardvark.c driver was triggering PCI_EXP_LNKCTL_RL bit before waiting for the link to be up. This does not align with neither PCIe base specifications nor with Armada 3700 Functional Specification. (Note that in older versions of aardvark, this bit was called incorrectly PCIE_CORE_LINK_TRAINING, so this may be the reason.) It is also unknown why Armada 3700 Functional Specification says that it is needed to trigger PCI_EXP_LNKCTL_RL for GEN2 mode, as according to PCIe base specification 5 GT/s speed negotiation is supposed to be entirely autonomous, even if initial speed is 2.5 GT/s. [1] - https://lore.kernel.org/linux-pci/87h7l8axqp.fsf@toke.dk/ [2] - https://lore.kernel.org/linux-pci/20210326124326.21163-1-pali@kernel.org/ Link: https://lore.kernel.org/r/20211005180952.6812-12-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Reviewed-by: Marek Behún <kabel@kernel.org>
2021-10-05 18:09:50 +00:00
/*
* Setup PCIe rev / gen compliance based on device tree property
* 'max-link-speed' which also forces maximal link speed.
*/
reg = advk_readl(pcie, PCIE_CORE_CTRL0_REG);
reg &= ~PCIE_GEN_SEL_MSK;
PCI: aardvark: Fix link training Fix multiple link training issues in aardvark driver. The main reason of these issues was misunderstanding of what certain registers do, since their names and comments were misleading: before commit 96be36dbffac ("PCI: aardvark: Replace custom macros by standard linux/pci_regs.h macros"), the pci-aardvark.c driver used custom macros for accessing standard PCIe Root Bridge registers, and misleading comments did not help to understand what the code was really doing. After doing more tests and experiments I've come to the conclusion that the SPEED_GEN register in aardvark sets the PCIe revision / generation compliance and forces maximal link speed. Both GEN3 and GEN2 values set the read-only PCI_EXP_FLAGS_VERS bits (PCIe capabilities version of Root Bridge) to value 2, while GEN1 value sets PCI_EXP_FLAGS_VERS to 1, which matches with PCI Express specifications revisions 3, 2 and 1 respectively. Changing SPEED_GEN also sets the read-only bits PCI_EXP_LNKCAP_SLS and PCI_EXP_LNKCAP2_SLS to corresponding speed. (Note that PCI Express rev 1 specification does not define PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers and when SPEED_GEN is set to GEN1 (which also sets PCI_EXP_FLAGS_VERS set to 1), lspci cannot access PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers.) Changing PCIe link speed can be done via PCI_EXP_LNKCTL2_TLS bits of PCI_EXP_LNKCTL2 register. Armada 3700 Functional Specifications says that the default value of PCI_EXP_LNKCTL2_TLS is based on SPEED_GEN value, but tests showed that the default value is always 8.0 GT/s, independently of speed set by SPEED_GEN. So after setting SPEED_GEN, we must also set value in PCI_EXP_LNKCTL2 register via PCI_EXP_LNKCTL2_TLS bits. Triggering PCI_EXP_LNKCTL_RL bit immediately after setting LINK_TRAINING_EN bit actually doesn't do anything. Tests have shown that a delay is needed after enabling LINK_TRAINING_EN bit. As triggering PCI_EXP_LNKCTL_RL currently does nothing, remove it. Commit 43fc679ced18 ("PCI: aardvark: Improve link training") introduced code which sets SPEED_GEN register based on negotiated link speed from PCI_EXP_LNKSTA_CLS bits of PCI_EXP_LNKSTA register. This code was added to fix detection of Compex WLE900VX (Atheros QCA9880) WiFi GEN1 PCIe cards, as otherwise these cards were "invisible" on PCIe bus (probably because they crashed). But apparently more people reported the same issues with these cards also with other PCIe controllers [1] and I was able to reproduce this issue also with other "noname" WiFi cards based on Atheros QCA9890 chip (with the same PCI vendor/device ids as Atheros QCA9880). So this is not an issue in aardvark but rather an issue in Atheros QCA98xx chips. Also, this issue only exists if the kernel is compiled with PCIe ASPM support, and a generic workaround for this is to change PCIe Bridge to 2.5 GT/s link speed via PCI_EXP_LNKCTL2_TLS_2_5GT bits in PCI_EXP_LNKCTL2 register [2], before triggering PCI_EXP_LNKCTL_RL bit. This workaround also works when SPEED_GEN is set to value GEN2 (5 GT/s). So remove this hack completely in the aardvark driver and always set SPEED_GEN to value from 'max-link-speed' DT property. Fix for Atheros QCA98xx chips is handled separately by patch [2]. These two things (code for triggering PCI_EXP_LNKCTL_RL bit and changing SPEED_GEN value) also explain why commit 6964494582f5 ("PCI: aardvark: Train link immediately after enabling training") somehow fixed detection of those problematic Compex cards with Atheros chips: if triggering link retraining (via PCI_EXP_LNKCTL_RL bit) was done immediately after enabling link training (via LINK_TRAINING_EN), it did nothing. If there was a specific delay, aardvark HW already initialized PCIe link and therefore triggering link retraining caused the above issue. Compex cards triggered link down event and disappeared from the PCIe bus. Commit f4c7d053d7f7 ("PCI: aardvark: Wait for endpoint to be ready before training link") added 100ms sleep before calling 'Start link training' command and explained that it is a requirement of PCI Express specification. But the code after this 100ms sleep was not doing 'Start link training', rather it triggered PCI_EXP_LNKCTL_RL bit via PCIe Root Bridge to put link into Recovery state. The required delay after fundamental reset is already done in function advk_pcie_wait_for_link() which also checks whether PCIe link is up. So after removing the code which triggers PCI_EXP_LNKCTL_RL bit on PCIe Root Bridge, there is no need to wait 100ms again. Remove the extra msleep() call and update comment about the delay required by the PCI Express specification. According to Marvell Armada 3700 Functional Specifications, Link training should be enabled via aardvark register LINK_TRAINING_EN after selecting PCIe generation and x1 lane. There is no need to disable it prior resetting card via PERST# signal. This disabling code was introduced in commit 5169a9851daa ("PCI: aardvark: Issue PERST via GPIO") as a workaround for some Atheros cards. It turns out that this also is Atheros specific issue and affects any PCIe controller, not only aardvark. Moreover this Atheros issue was triggered by juggling with PCI_EXP_LNKCTL_RL, LINK_TRAINING_EN and SPEED_GEN bits interleaved with sleeps. Now, after removing triggering PCI_EXP_LNKCTL_RL, there is no need to explicitly disable LINK_TRAINING_EN bit. So remove this code too. The problematic Compex cards described in previous git commits are correctly detected in advk_pcie_train_link() function even after applying all these changes. Note that with this patch, and also prior this patch, some NVMe disks which support PCIe GEN3 with 8 GT/s speed are negotiated only at the lowest link speed 2.5 GT/s, independently of SPEED_GEN value. After manually triggering PCI_EXP_LNKCTL_RL bit (e.g. from userspace via setpci), these NVMe disks change link speed to 5 GT/s when SPEED_GEN was configured to GEN2. This issue first needs to be properly investigated. I will send a fix in the future. On the other hand, some other GEN2 PCIe cards with 5 GT/s speed are autonomously by HW autonegotiated at full 5 GT/s speed without need of any software interaction. Armada 3700 Functional Specifications describes the following steps for link training: set SPEED_GEN to GEN2, enable LINK_TRAINING_EN, poll until link training is complete, trigger PCI_EXP_LNKCTL_RL, poll until signal rate is 5 GT/s, poll until link training is complete, enable ASPM L0s. The requirement for triggering PCI_EXP_LNKCTL_RL can be explained by the need to achieve 5 GT/s speed (as changing link speed is done by throw to recovery state entered by PCI_EXP_LNKCTL_RL) or maybe as a part of enabling ASPM L0s (but in this case ASPM L0s should have been enabled prior PCI_EXP_LNKCTL_RL). It is unknown why the original pci-aardvark.c driver was triggering PCI_EXP_LNKCTL_RL bit before waiting for the link to be up. This does not align with neither PCIe base specifications nor with Armada 3700 Functional Specification. (Note that in older versions of aardvark, this bit was called incorrectly PCIE_CORE_LINK_TRAINING, so this may be the reason.) It is also unknown why Armada 3700 Functional Specification says that it is needed to trigger PCI_EXP_LNKCTL_RL for GEN2 mode, as according to PCIe base specification 5 GT/s speed negotiation is supposed to be entirely autonomous, even if initial speed is 2.5 GT/s. [1] - https://lore.kernel.org/linux-pci/87h7l8axqp.fsf@toke.dk/ [2] - https://lore.kernel.org/linux-pci/20210326124326.21163-1-pali@kernel.org/ Link: https://lore.kernel.org/r/20211005180952.6812-12-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Reviewed-by: Marek Behún <kabel@kernel.org>
2021-10-05 18:09:50 +00:00
if (pcie->link_gen == 3)
reg |= SPEED_GEN_3;
PCI: aardvark: Fix link training Fix multiple link training issues in aardvark driver. The main reason of these issues was misunderstanding of what certain registers do, since their names and comments were misleading: before commit 96be36dbffac ("PCI: aardvark: Replace custom macros by standard linux/pci_regs.h macros"), the pci-aardvark.c driver used custom macros for accessing standard PCIe Root Bridge registers, and misleading comments did not help to understand what the code was really doing. After doing more tests and experiments I've come to the conclusion that the SPEED_GEN register in aardvark sets the PCIe revision / generation compliance and forces maximal link speed. Both GEN3 and GEN2 values set the read-only PCI_EXP_FLAGS_VERS bits (PCIe capabilities version of Root Bridge) to value 2, while GEN1 value sets PCI_EXP_FLAGS_VERS to 1, which matches with PCI Express specifications revisions 3, 2 and 1 respectively. Changing SPEED_GEN also sets the read-only bits PCI_EXP_LNKCAP_SLS and PCI_EXP_LNKCAP2_SLS to corresponding speed. (Note that PCI Express rev 1 specification does not define PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers and when SPEED_GEN is set to GEN1 (which also sets PCI_EXP_FLAGS_VERS set to 1), lspci cannot access PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers.) Changing PCIe link speed can be done via PCI_EXP_LNKCTL2_TLS bits of PCI_EXP_LNKCTL2 register. Armada 3700 Functional Specifications says that the default value of PCI_EXP_LNKCTL2_TLS is based on SPEED_GEN value, but tests showed that the default value is always 8.0 GT/s, independently of speed set by SPEED_GEN. So after setting SPEED_GEN, we must also set value in PCI_EXP_LNKCTL2 register via PCI_EXP_LNKCTL2_TLS bits. Triggering PCI_EXP_LNKCTL_RL bit immediately after setting LINK_TRAINING_EN bit actually doesn't do anything. Tests have shown that a delay is needed after enabling LINK_TRAINING_EN bit. As triggering PCI_EXP_LNKCTL_RL currently does nothing, remove it. Commit 43fc679ced18 ("PCI: aardvark: Improve link training") introduced code which sets SPEED_GEN register based on negotiated link speed from PCI_EXP_LNKSTA_CLS bits of PCI_EXP_LNKSTA register. This code was added to fix detection of Compex WLE900VX (Atheros QCA9880) WiFi GEN1 PCIe cards, as otherwise these cards were "invisible" on PCIe bus (probably because they crashed). But apparently more people reported the same issues with these cards also with other PCIe controllers [1] and I was able to reproduce this issue also with other "noname" WiFi cards based on Atheros QCA9890 chip (with the same PCI vendor/device ids as Atheros QCA9880). So this is not an issue in aardvark but rather an issue in Atheros QCA98xx chips. Also, this issue only exists if the kernel is compiled with PCIe ASPM support, and a generic workaround for this is to change PCIe Bridge to 2.5 GT/s link speed via PCI_EXP_LNKCTL2_TLS_2_5GT bits in PCI_EXP_LNKCTL2 register [2], before triggering PCI_EXP_LNKCTL_RL bit. This workaround also works when SPEED_GEN is set to value GEN2 (5 GT/s). So remove this hack completely in the aardvark driver and always set SPEED_GEN to value from 'max-link-speed' DT property. Fix for Atheros QCA98xx chips is handled separately by patch [2]. These two things (code for triggering PCI_EXP_LNKCTL_RL bit and changing SPEED_GEN value) also explain why commit 6964494582f5 ("PCI: aardvark: Train link immediately after enabling training") somehow fixed detection of those problematic Compex cards with Atheros chips: if triggering link retraining (via PCI_EXP_LNKCTL_RL bit) was done immediately after enabling link training (via LINK_TRAINING_EN), it did nothing. If there was a specific delay, aardvark HW already initialized PCIe link and therefore triggering link retraining caused the above issue. Compex cards triggered link down event and disappeared from the PCIe bus. Commit f4c7d053d7f7 ("PCI: aardvark: Wait for endpoint to be ready before training link") added 100ms sleep before calling 'Start link training' command and explained that it is a requirement of PCI Express specification. But the code after this 100ms sleep was not doing 'Start link training', rather it triggered PCI_EXP_LNKCTL_RL bit via PCIe Root Bridge to put link into Recovery state. The required delay after fundamental reset is already done in function advk_pcie_wait_for_link() which also checks whether PCIe link is up. So after removing the code which triggers PCI_EXP_LNKCTL_RL bit on PCIe Root Bridge, there is no need to wait 100ms again. Remove the extra msleep() call and update comment about the delay required by the PCI Express specification. According to Marvell Armada 3700 Functional Specifications, Link training should be enabled via aardvark register LINK_TRAINING_EN after selecting PCIe generation and x1 lane. There is no need to disable it prior resetting card via PERST# signal. This disabling code was introduced in commit 5169a9851daa ("PCI: aardvark: Issue PERST via GPIO") as a workaround for some Atheros cards. It turns out that this also is Atheros specific issue and affects any PCIe controller, not only aardvark. Moreover this Atheros issue was triggered by juggling with PCI_EXP_LNKCTL_RL, LINK_TRAINING_EN and SPEED_GEN bits interleaved with sleeps. Now, after removing triggering PCI_EXP_LNKCTL_RL, there is no need to explicitly disable LINK_TRAINING_EN bit. So remove this code too. The problematic Compex cards described in previous git commits are correctly detected in advk_pcie_train_link() function even after applying all these changes. Note that with this patch, and also prior this patch, some NVMe disks which support PCIe GEN3 with 8 GT/s speed are negotiated only at the lowest link speed 2.5 GT/s, independently of SPEED_GEN value. After manually triggering PCI_EXP_LNKCTL_RL bit (e.g. from userspace via setpci), these NVMe disks change link speed to 5 GT/s when SPEED_GEN was configured to GEN2. This issue first needs to be properly investigated. I will send a fix in the future. On the other hand, some other GEN2 PCIe cards with 5 GT/s speed are autonomously by HW autonegotiated at full 5 GT/s speed without need of any software interaction. Armada 3700 Functional Specifications describes the following steps for link training: set SPEED_GEN to GEN2, enable LINK_TRAINING_EN, poll until link training is complete, trigger PCI_EXP_LNKCTL_RL, poll until signal rate is 5 GT/s, poll until link training is complete, enable ASPM L0s. The requirement for triggering PCI_EXP_LNKCTL_RL can be explained by the need to achieve 5 GT/s speed (as changing link speed is done by throw to recovery state entered by PCI_EXP_LNKCTL_RL) or maybe as a part of enabling ASPM L0s (but in this case ASPM L0s should have been enabled prior PCI_EXP_LNKCTL_RL). It is unknown why the original pci-aardvark.c driver was triggering PCI_EXP_LNKCTL_RL bit before waiting for the link to be up. This does not align with neither PCIe base specifications nor with Armada 3700 Functional Specification. (Note that in older versions of aardvark, this bit was called incorrectly PCIE_CORE_LINK_TRAINING, so this may be the reason.) It is also unknown why Armada 3700 Functional Specification says that it is needed to trigger PCI_EXP_LNKCTL_RL for GEN2 mode, as according to PCIe base specification 5 GT/s speed negotiation is supposed to be entirely autonomous, even if initial speed is 2.5 GT/s. [1] - https://lore.kernel.org/linux-pci/87h7l8axqp.fsf@toke.dk/ [2] - https://lore.kernel.org/linux-pci/20210326124326.21163-1-pali@kernel.org/ Link: https://lore.kernel.org/r/20211005180952.6812-12-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Reviewed-by: Marek Behún <kabel@kernel.org>
2021-10-05 18:09:50 +00:00
else if (pcie->link_gen == 2)
reg |= SPEED_GEN_2;
else
reg |= SPEED_GEN_1;
advk_writel(pcie, reg, PCIE_CORE_CTRL0_REG);
/*
PCI: aardvark: Fix link training Fix multiple link training issues in aardvark driver. The main reason of these issues was misunderstanding of what certain registers do, since their names and comments were misleading: before commit 96be36dbffac ("PCI: aardvark: Replace custom macros by standard linux/pci_regs.h macros"), the pci-aardvark.c driver used custom macros for accessing standard PCIe Root Bridge registers, and misleading comments did not help to understand what the code was really doing. After doing more tests and experiments I've come to the conclusion that the SPEED_GEN register in aardvark sets the PCIe revision / generation compliance and forces maximal link speed. Both GEN3 and GEN2 values set the read-only PCI_EXP_FLAGS_VERS bits (PCIe capabilities version of Root Bridge) to value 2, while GEN1 value sets PCI_EXP_FLAGS_VERS to 1, which matches with PCI Express specifications revisions 3, 2 and 1 respectively. Changing SPEED_GEN also sets the read-only bits PCI_EXP_LNKCAP_SLS and PCI_EXP_LNKCAP2_SLS to corresponding speed. (Note that PCI Express rev 1 specification does not define PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers and when SPEED_GEN is set to GEN1 (which also sets PCI_EXP_FLAGS_VERS set to 1), lspci cannot access PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers.) Changing PCIe link speed can be done via PCI_EXP_LNKCTL2_TLS bits of PCI_EXP_LNKCTL2 register. Armada 3700 Functional Specifications says that the default value of PCI_EXP_LNKCTL2_TLS is based on SPEED_GEN value, but tests showed that the default value is always 8.0 GT/s, independently of speed set by SPEED_GEN. So after setting SPEED_GEN, we must also set value in PCI_EXP_LNKCTL2 register via PCI_EXP_LNKCTL2_TLS bits. Triggering PCI_EXP_LNKCTL_RL bit immediately after setting LINK_TRAINING_EN bit actually doesn't do anything. Tests have shown that a delay is needed after enabling LINK_TRAINING_EN bit. As triggering PCI_EXP_LNKCTL_RL currently does nothing, remove it. Commit 43fc679ced18 ("PCI: aardvark: Improve link training") introduced code which sets SPEED_GEN register based on negotiated link speed from PCI_EXP_LNKSTA_CLS bits of PCI_EXP_LNKSTA register. This code was added to fix detection of Compex WLE900VX (Atheros QCA9880) WiFi GEN1 PCIe cards, as otherwise these cards were "invisible" on PCIe bus (probably because they crashed). But apparently more people reported the same issues with these cards also with other PCIe controllers [1] and I was able to reproduce this issue also with other "noname" WiFi cards based on Atheros QCA9890 chip (with the same PCI vendor/device ids as Atheros QCA9880). So this is not an issue in aardvark but rather an issue in Atheros QCA98xx chips. Also, this issue only exists if the kernel is compiled with PCIe ASPM support, and a generic workaround for this is to change PCIe Bridge to 2.5 GT/s link speed via PCI_EXP_LNKCTL2_TLS_2_5GT bits in PCI_EXP_LNKCTL2 register [2], before triggering PCI_EXP_LNKCTL_RL bit. This workaround also works when SPEED_GEN is set to value GEN2 (5 GT/s). So remove this hack completely in the aardvark driver and always set SPEED_GEN to value from 'max-link-speed' DT property. Fix for Atheros QCA98xx chips is handled separately by patch [2]. These two things (code for triggering PCI_EXP_LNKCTL_RL bit and changing SPEED_GEN value) also explain why commit 6964494582f5 ("PCI: aardvark: Train link immediately after enabling training") somehow fixed detection of those problematic Compex cards with Atheros chips: if triggering link retraining (via PCI_EXP_LNKCTL_RL bit) was done immediately after enabling link training (via LINK_TRAINING_EN), it did nothing. If there was a specific delay, aardvark HW already initialized PCIe link and therefore triggering link retraining caused the above issue. Compex cards triggered link down event and disappeared from the PCIe bus. Commit f4c7d053d7f7 ("PCI: aardvark: Wait for endpoint to be ready before training link") added 100ms sleep before calling 'Start link training' command and explained that it is a requirement of PCI Express specification. But the code after this 100ms sleep was not doing 'Start link training', rather it triggered PCI_EXP_LNKCTL_RL bit via PCIe Root Bridge to put link into Recovery state. The required delay after fundamental reset is already done in function advk_pcie_wait_for_link() which also checks whether PCIe link is up. So after removing the code which triggers PCI_EXP_LNKCTL_RL bit on PCIe Root Bridge, there is no need to wait 100ms again. Remove the extra msleep() call and update comment about the delay required by the PCI Express specification. According to Marvell Armada 3700 Functional Specifications, Link training should be enabled via aardvark register LINK_TRAINING_EN after selecting PCIe generation and x1 lane. There is no need to disable it prior resetting card via PERST# signal. This disabling code was introduced in commit 5169a9851daa ("PCI: aardvark: Issue PERST via GPIO") as a workaround for some Atheros cards. It turns out that this also is Atheros specific issue and affects any PCIe controller, not only aardvark. Moreover this Atheros issue was triggered by juggling with PCI_EXP_LNKCTL_RL, LINK_TRAINING_EN and SPEED_GEN bits interleaved with sleeps. Now, after removing triggering PCI_EXP_LNKCTL_RL, there is no need to explicitly disable LINK_TRAINING_EN bit. So remove this code too. The problematic Compex cards described in previous git commits are correctly detected in advk_pcie_train_link() function even after applying all these changes. Note that with this patch, and also prior this patch, some NVMe disks which support PCIe GEN3 with 8 GT/s speed are negotiated only at the lowest link speed 2.5 GT/s, independently of SPEED_GEN value. After manually triggering PCI_EXP_LNKCTL_RL bit (e.g. from userspace via setpci), these NVMe disks change link speed to 5 GT/s when SPEED_GEN was configured to GEN2. This issue first needs to be properly investigated. I will send a fix in the future. On the other hand, some other GEN2 PCIe cards with 5 GT/s speed are autonomously by HW autonegotiated at full 5 GT/s speed without need of any software interaction. Armada 3700 Functional Specifications describes the following steps for link training: set SPEED_GEN to GEN2, enable LINK_TRAINING_EN, poll until link training is complete, trigger PCI_EXP_LNKCTL_RL, poll until signal rate is 5 GT/s, poll until link training is complete, enable ASPM L0s. The requirement for triggering PCI_EXP_LNKCTL_RL can be explained by the need to achieve 5 GT/s speed (as changing link speed is done by throw to recovery state entered by PCI_EXP_LNKCTL_RL) or maybe as a part of enabling ASPM L0s (but in this case ASPM L0s should have been enabled prior PCI_EXP_LNKCTL_RL). It is unknown why the original pci-aardvark.c driver was triggering PCI_EXP_LNKCTL_RL bit before waiting for the link to be up. This does not align with neither PCIe base specifications nor with Armada 3700 Functional Specification. (Note that in older versions of aardvark, this bit was called incorrectly PCIE_CORE_LINK_TRAINING, so this may be the reason.) It is also unknown why Armada 3700 Functional Specification says that it is needed to trigger PCI_EXP_LNKCTL_RL for GEN2 mode, as according to PCIe base specification 5 GT/s speed negotiation is supposed to be entirely autonomous, even if initial speed is 2.5 GT/s. [1] - https://lore.kernel.org/linux-pci/87h7l8axqp.fsf@toke.dk/ [2] - https://lore.kernel.org/linux-pci/20210326124326.21163-1-pali@kernel.org/ Link: https://lore.kernel.org/r/20211005180952.6812-12-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Reviewed-by: Marek Behún <kabel@kernel.org>
2021-10-05 18:09:50 +00:00
* Set maximal link speed value also into PCIe Link Control 2 register.
* Armada 3700 Functional Specification says that default value is based
* on SPEED_GEN but tests showed that default value is always 8.0 GT/s.
*/
PCI: aardvark: Fix link training Fix multiple link training issues in aardvark driver. The main reason of these issues was misunderstanding of what certain registers do, since their names and comments were misleading: before commit 96be36dbffac ("PCI: aardvark: Replace custom macros by standard linux/pci_regs.h macros"), the pci-aardvark.c driver used custom macros for accessing standard PCIe Root Bridge registers, and misleading comments did not help to understand what the code was really doing. After doing more tests and experiments I've come to the conclusion that the SPEED_GEN register in aardvark sets the PCIe revision / generation compliance and forces maximal link speed. Both GEN3 and GEN2 values set the read-only PCI_EXP_FLAGS_VERS bits (PCIe capabilities version of Root Bridge) to value 2, while GEN1 value sets PCI_EXP_FLAGS_VERS to 1, which matches with PCI Express specifications revisions 3, 2 and 1 respectively. Changing SPEED_GEN also sets the read-only bits PCI_EXP_LNKCAP_SLS and PCI_EXP_LNKCAP2_SLS to corresponding speed. (Note that PCI Express rev 1 specification does not define PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers and when SPEED_GEN is set to GEN1 (which also sets PCI_EXP_FLAGS_VERS set to 1), lspci cannot access PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers.) Changing PCIe link speed can be done via PCI_EXP_LNKCTL2_TLS bits of PCI_EXP_LNKCTL2 register. Armada 3700 Functional Specifications says that the default value of PCI_EXP_LNKCTL2_TLS is based on SPEED_GEN value, but tests showed that the default value is always 8.0 GT/s, independently of speed set by SPEED_GEN. So after setting SPEED_GEN, we must also set value in PCI_EXP_LNKCTL2 register via PCI_EXP_LNKCTL2_TLS bits. Triggering PCI_EXP_LNKCTL_RL bit immediately after setting LINK_TRAINING_EN bit actually doesn't do anything. Tests have shown that a delay is needed after enabling LINK_TRAINING_EN bit. As triggering PCI_EXP_LNKCTL_RL currently does nothing, remove it. Commit 43fc679ced18 ("PCI: aardvark: Improve link training") introduced code which sets SPEED_GEN register based on negotiated link speed from PCI_EXP_LNKSTA_CLS bits of PCI_EXP_LNKSTA register. This code was added to fix detection of Compex WLE900VX (Atheros QCA9880) WiFi GEN1 PCIe cards, as otherwise these cards were "invisible" on PCIe bus (probably because they crashed). But apparently more people reported the same issues with these cards also with other PCIe controllers [1] and I was able to reproduce this issue also with other "noname" WiFi cards based on Atheros QCA9890 chip (with the same PCI vendor/device ids as Atheros QCA9880). So this is not an issue in aardvark but rather an issue in Atheros QCA98xx chips. Also, this issue only exists if the kernel is compiled with PCIe ASPM support, and a generic workaround for this is to change PCIe Bridge to 2.5 GT/s link speed via PCI_EXP_LNKCTL2_TLS_2_5GT bits in PCI_EXP_LNKCTL2 register [2], before triggering PCI_EXP_LNKCTL_RL bit. This workaround also works when SPEED_GEN is set to value GEN2 (5 GT/s). So remove this hack completely in the aardvark driver and always set SPEED_GEN to value from 'max-link-speed' DT property. Fix for Atheros QCA98xx chips is handled separately by patch [2]. These two things (code for triggering PCI_EXP_LNKCTL_RL bit and changing SPEED_GEN value) also explain why commit 6964494582f5 ("PCI: aardvark: Train link immediately after enabling training") somehow fixed detection of those problematic Compex cards with Atheros chips: if triggering link retraining (via PCI_EXP_LNKCTL_RL bit) was done immediately after enabling link training (via LINK_TRAINING_EN), it did nothing. If there was a specific delay, aardvark HW already initialized PCIe link and therefore triggering link retraining caused the above issue. Compex cards triggered link down event and disappeared from the PCIe bus. Commit f4c7d053d7f7 ("PCI: aardvark: Wait for endpoint to be ready before training link") added 100ms sleep before calling 'Start link training' command and explained that it is a requirement of PCI Express specification. But the code after this 100ms sleep was not doing 'Start link training', rather it triggered PCI_EXP_LNKCTL_RL bit via PCIe Root Bridge to put link into Recovery state. The required delay after fundamental reset is already done in function advk_pcie_wait_for_link() which also checks whether PCIe link is up. So after removing the code which triggers PCI_EXP_LNKCTL_RL bit on PCIe Root Bridge, there is no need to wait 100ms again. Remove the extra msleep() call and update comment about the delay required by the PCI Express specification. According to Marvell Armada 3700 Functional Specifications, Link training should be enabled via aardvark register LINK_TRAINING_EN after selecting PCIe generation and x1 lane. There is no need to disable it prior resetting card via PERST# signal. This disabling code was introduced in commit 5169a9851daa ("PCI: aardvark: Issue PERST via GPIO") as a workaround for some Atheros cards. It turns out that this also is Atheros specific issue and affects any PCIe controller, not only aardvark. Moreover this Atheros issue was triggered by juggling with PCI_EXP_LNKCTL_RL, LINK_TRAINING_EN and SPEED_GEN bits interleaved with sleeps. Now, after removing triggering PCI_EXP_LNKCTL_RL, there is no need to explicitly disable LINK_TRAINING_EN bit. So remove this code too. The problematic Compex cards described in previous git commits are correctly detected in advk_pcie_train_link() function even after applying all these changes. Note that with this patch, and also prior this patch, some NVMe disks which support PCIe GEN3 with 8 GT/s speed are negotiated only at the lowest link speed 2.5 GT/s, independently of SPEED_GEN value. After manually triggering PCI_EXP_LNKCTL_RL bit (e.g. from userspace via setpci), these NVMe disks change link speed to 5 GT/s when SPEED_GEN was configured to GEN2. This issue first needs to be properly investigated. I will send a fix in the future. On the other hand, some other GEN2 PCIe cards with 5 GT/s speed are autonomously by HW autonegotiated at full 5 GT/s speed without need of any software interaction. Armada 3700 Functional Specifications describes the following steps for link training: set SPEED_GEN to GEN2, enable LINK_TRAINING_EN, poll until link training is complete, trigger PCI_EXP_LNKCTL_RL, poll until signal rate is 5 GT/s, poll until link training is complete, enable ASPM L0s. The requirement for triggering PCI_EXP_LNKCTL_RL can be explained by the need to achieve 5 GT/s speed (as changing link speed is done by throw to recovery state entered by PCI_EXP_LNKCTL_RL) or maybe as a part of enabling ASPM L0s (but in this case ASPM L0s should have been enabled prior PCI_EXP_LNKCTL_RL). It is unknown why the original pci-aardvark.c driver was triggering PCI_EXP_LNKCTL_RL bit before waiting for the link to be up. This does not align with neither PCIe base specifications nor with Armada 3700 Functional Specification. (Note that in older versions of aardvark, this bit was called incorrectly PCIE_CORE_LINK_TRAINING, so this may be the reason.) It is also unknown why Armada 3700 Functional Specification says that it is needed to trigger PCI_EXP_LNKCTL_RL for GEN2 mode, as according to PCIe base specification 5 GT/s speed negotiation is supposed to be entirely autonomous, even if initial speed is 2.5 GT/s. [1] - https://lore.kernel.org/linux-pci/87h7l8axqp.fsf@toke.dk/ [2] - https://lore.kernel.org/linux-pci/20210326124326.21163-1-pali@kernel.org/ Link: https://lore.kernel.org/r/20211005180952.6812-12-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Reviewed-by: Marek Behún <kabel@kernel.org>
2021-10-05 18:09:50 +00:00
reg = advk_readl(pcie, PCIE_CORE_PCIEXP_CAP + PCI_EXP_LNKCTL2);
reg &= ~PCI_EXP_LNKCTL2_TLS;
if (pcie->link_gen == 3)
reg |= PCI_EXP_LNKCTL2_TLS_8_0GT;
else if (pcie->link_gen == 2)
reg |= PCI_EXP_LNKCTL2_TLS_5_0GT;
else
reg |= PCI_EXP_LNKCTL2_TLS_2_5GT;
advk_writel(pcie, reg, PCIE_CORE_PCIEXP_CAP + PCI_EXP_LNKCTL2);
/* Enable link training after selecting PCIe generation */
reg = advk_readl(pcie, PCIE_CORE_CTRL0_REG);
reg |= LINK_TRAINING_EN;
advk_writel(pcie, reg, PCIE_CORE_CTRL0_REG);
/*
* Reset PCIe card via PERST# signal. Some cards are not detected
* during link training when they are in some non-initial state.
*/
advk_pcie_issue_perst(pcie);
/*
* PERST# signal could have been asserted by pinctrl subsystem before
* probe() callback has been called or issued explicitly by reset gpio
* function advk_pcie_issue_perst(), making the endpoint going into
PCI: aardvark: Fix link training Fix multiple link training issues in aardvark driver. The main reason of these issues was misunderstanding of what certain registers do, since their names and comments were misleading: before commit 96be36dbffac ("PCI: aardvark: Replace custom macros by standard linux/pci_regs.h macros"), the pci-aardvark.c driver used custom macros for accessing standard PCIe Root Bridge registers, and misleading comments did not help to understand what the code was really doing. After doing more tests and experiments I've come to the conclusion that the SPEED_GEN register in aardvark sets the PCIe revision / generation compliance and forces maximal link speed. Both GEN3 and GEN2 values set the read-only PCI_EXP_FLAGS_VERS bits (PCIe capabilities version of Root Bridge) to value 2, while GEN1 value sets PCI_EXP_FLAGS_VERS to 1, which matches with PCI Express specifications revisions 3, 2 and 1 respectively. Changing SPEED_GEN also sets the read-only bits PCI_EXP_LNKCAP_SLS and PCI_EXP_LNKCAP2_SLS to corresponding speed. (Note that PCI Express rev 1 specification does not define PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers and when SPEED_GEN is set to GEN1 (which also sets PCI_EXP_FLAGS_VERS set to 1), lspci cannot access PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers.) Changing PCIe link speed can be done via PCI_EXP_LNKCTL2_TLS bits of PCI_EXP_LNKCTL2 register. Armada 3700 Functional Specifications says that the default value of PCI_EXP_LNKCTL2_TLS is based on SPEED_GEN value, but tests showed that the default value is always 8.0 GT/s, independently of speed set by SPEED_GEN. So after setting SPEED_GEN, we must also set value in PCI_EXP_LNKCTL2 register via PCI_EXP_LNKCTL2_TLS bits. Triggering PCI_EXP_LNKCTL_RL bit immediately after setting LINK_TRAINING_EN bit actually doesn't do anything. Tests have shown that a delay is needed after enabling LINK_TRAINING_EN bit. As triggering PCI_EXP_LNKCTL_RL currently does nothing, remove it. Commit 43fc679ced18 ("PCI: aardvark: Improve link training") introduced code which sets SPEED_GEN register based on negotiated link speed from PCI_EXP_LNKSTA_CLS bits of PCI_EXP_LNKSTA register. This code was added to fix detection of Compex WLE900VX (Atheros QCA9880) WiFi GEN1 PCIe cards, as otherwise these cards were "invisible" on PCIe bus (probably because they crashed). But apparently more people reported the same issues with these cards also with other PCIe controllers [1] and I was able to reproduce this issue also with other "noname" WiFi cards based on Atheros QCA9890 chip (with the same PCI vendor/device ids as Atheros QCA9880). So this is not an issue in aardvark but rather an issue in Atheros QCA98xx chips. Also, this issue only exists if the kernel is compiled with PCIe ASPM support, and a generic workaround for this is to change PCIe Bridge to 2.5 GT/s link speed via PCI_EXP_LNKCTL2_TLS_2_5GT bits in PCI_EXP_LNKCTL2 register [2], before triggering PCI_EXP_LNKCTL_RL bit. This workaround also works when SPEED_GEN is set to value GEN2 (5 GT/s). So remove this hack completely in the aardvark driver and always set SPEED_GEN to value from 'max-link-speed' DT property. Fix for Atheros QCA98xx chips is handled separately by patch [2]. These two things (code for triggering PCI_EXP_LNKCTL_RL bit and changing SPEED_GEN value) also explain why commit 6964494582f5 ("PCI: aardvark: Train link immediately after enabling training") somehow fixed detection of those problematic Compex cards with Atheros chips: if triggering link retraining (via PCI_EXP_LNKCTL_RL bit) was done immediately after enabling link training (via LINK_TRAINING_EN), it did nothing. If there was a specific delay, aardvark HW already initialized PCIe link and therefore triggering link retraining caused the above issue. Compex cards triggered link down event and disappeared from the PCIe bus. Commit f4c7d053d7f7 ("PCI: aardvark: Wait for endpoint to be ready before training link") added 100ms sleep before calling 'Start link training' command and explained that it is a requirement of PCI Express specification. But the code after this 100ms sleep was not doing 'Start link training', rather it triggered PCI_EXP_LNKCTL_RL bit via PCIe Root Bridge to put link into Recovery state. The required delay after fundamental reset is already done in function advk_pcie_wait_for_link() which also checks whether PCIe link is up. So after removing the code which triggers PCI_EXP_LNKCTL_RL bit on PCIe Root Bridge, there is no need to wait 100ms again. Remove the extra msleep() call and update comment about the delay required by the PCI Express specification. According to Marvell Armada 3700 Functional Specifications, Link training should be enabled via aardvark register LINK_TRAINING_EN after selecting PCIe generation and x1 lane. There is no need to disable it prior resetting card via PERST# signal. This disabling code was introduced in commit 5169a9851daa ("PCI: aardvark: Issue PERST via GPIO") as a workaround for some Atheros cards. It turns out that this also is Atheros specific issue and affects any PCIe controller, not only aardvark. Moreover this Atheros issue was triggered by juggling with PCI_EXP_LNKCTL_RL, LINK_TRAINING_EN and SPEED_GEN bits interleaved with sleeps. Now, after removing triggering PCI_EXP_LNKCTL_RL, there is no need to explicitly disable LINK_TRAINING_EN bit. So remove this code too. The problematic Compex cards described in previous git commits are correctly detected in advk_pcie_train_link() function even after applying all these changes. Note that with this patch, and also prior this patch, some NVMe disks which support PCIe GEN3 with 8 GT/s speed are negotiated only at the lowest link speed 2.5 GT/s, independently of SPEED_GEN value. After manually triggering PCI_EXP_LNKCTL_RL bit (e.g. from userspace via setpci), these NVMe disks change link speed to 5 GT/s when SPEED_GEN was configured to GEN2. This issue first needs to be properly investigated. I will send a fix in the future. On the other hand, some other GEN2 PCIe cards with 5 GT/s speed are autonomously by HW autonegotiated at full 5 GT/s speed without need of any software interaction. Armada 3700 Functional Specifications describes the following steps for link training: set SPEED_GEN to GEN2, enable LINK_TRAINING_EN, poll until link training is complete, trigger PCI_EXP_LNKCTL_RL, poll until signal rate is 5 GT/s, poll until link training is complete, enable ASPM L0s. The requirement for triggering PCI_EXP_LNKCTL_RL can be explained by the need to achieve 5 GT/s speed (as changing link speed is done by throw to recovery state entered by PCI_EXP_LNKCTL_RL) or maybe as a part of enabling ASPM L0s (but in this case ASPM L0s should have been enabled prior PCI_EXP_LNKCTL_RL). It is unknown why the original pci-aardvark.c driver was triggering PCI_EXP_LNKCTL_RL bit before waiting for the link to be up. This does not align with neither PCIe base specifications nor with Armada 3700 Functional Specification. (Note that in older versions of aardvark, this bit was called incorrectly PCIE_CORE_LINK_TRAINING, so this may be the reason.) It is also unknown why Armada 3700 Functional Specification says that it is needed to trigger PCI_EXP_LNKCTL_RL for GEN2 mode, as according to PCIe base specification 5 GT/s speed negotiation is supposed to be entirely autonomous, even if initial speed is 2.5 GT/s. [1] - https://lore.kernel.org/linux-pci/87h7l8axqp.fsf@toke.dk/ [2] - https://lore.kernel.org/linux-pci/20210326124326.21163-1-pali@kernel.org/ Link: https://lore.kernel.org/r/20211005180952.6812-12-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Reviewed-by: Marek Behún <kabel@kernel.org>
2021-10-05 18:09:50 +00:00
* fundamental reset. As required by PCI Express spec (PCI Express
* Base Specification, REV. 4.0 PCI Express, February 19 2014, 6.6.1
* Conventional Reset) a delay for at least 100ms after such a reset
* before sending a Configuration Request to the device is needed.
* So wait until PCIe link is up. Function advk_pcie_wait_for_link()
* waits for link at least 900ms.
*/
PCI: aardvark: Fix link training Fix multiple link training issues in aardvark driver. The main reason of these issues was misunderstanding of what certain registers do, since their names and comments were misleading: before commit 96be36dbffac ("PCI: aardvark: Replace custom macros by standard linux/pci_regs.h macros"), the pci-aardvark.c driver used custom macros for accessing standard PCIe Root Bridge registers, and misleading comments did not help to understand what the code was really doing. After doing more tests and experiments I've come to the conclusion that the SPEED_GEN register in aardvark sets the PCIe revision / generation compliance and forces maximal link speed. Both GEN3 and GEN2 values set the read-only PCI_EXP_FLAGS_VERS bits (PCIe capabilities version of Root Bridge) to value 2, while GEN1 value sets PCI_EXP_FLAGS_VERS to 1, which matches with PCI Express specifications revisions 3, 2 and 1 respectively. Changing SPEED_GEN also sets the read-only bits PCI_EXP_LNKCAP_SLS and PCI_EXP_LNKCAP2_SLS to corresponding speed. (Note that PCI Express rev 1 specification does not define PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers and when SPEED_GEN is set to GEN1 (which also sets PCI_EXP_FLAGS_VERS set to 1), lspci cannot access PCI_EXP_LNKCAP2 and PCI_EXP_LNKCTL2 registers.) Changing PCIe link speed can be done via PCI_EXP_LNKCTL2_TLS bits of PCI_EXP_LNKCTL2 register. Armada 3700 Functional Specifications says that the default value of PCI_EXP_LNKCTL2_TLS is based on SPEED_GEN value, but tests showed that the default value is always 8.0 GT/s, independently of speed set by SPEED_GEN. So after setting SPEED_GEN, we must also set value in PCI_EXP_LNKCTL2 register via PCI_EXP_LNKCTL2_TLS bits. Triggering PCI_EXP_LNKCTL_RL bit immediately after setting LINK_TRAINING_EN bit actually doesn't do anything. Tests have shown that a delay is needed after enabling LINK_TRAINING_EN bit. As triggering PCI_EXP_LNKCTL_RL currently does nothing, remove it. Commit 43fc679ced18 ("PCI: aardvark: Improve link training") introduced code which sets SPEED_GEN register based on negotiated link speed from PCI_EXP_LNKSTA_CLS bits of PCI_EXP_LNKSTA register. This code was added to fix detection of Compex WLE900VX (Atheros QCA9880) WiFi GEN1 PCIe cards, as otherwise these cards were "invisible" on PCIe bus (probably because they crashed). But apparently more people reported the same issues with these cards also with other PCIe controllers [1] and I was able to reproduce this issue also with other "noname" WiFi cards based on Atheros QCA9890 chip (with the same PCI vendor/device ids as Atheros QCA9880). So this is not an issue in aardvark but rather an issue in Atheros QCA98xx chips. Also, this issue only exists if the kernel is compiled with PCIe ASPM support, and a generic workaround for this is to change PCIe Bridge to 2.5 GT/s link speed via PCI_EXP_LNKCTL2_TLS_2_5GT bits in PCI_EXP_LNKCTL2 register [2], before triggering PCI_EXP_LNKCTL_RL bit. This workaround also works when SPEED_GEN is set to value GEN2 (5 GT/s). So remove this hack completely in the aardvark driver and always set SPEED_GEN to value from 'max-link-speed' DT property. Fix for Atheros QCA98xx chips is handled separately by patch [2]. These two things (code for triggering PCI_EXP_LNKCTL_RL bit and changing SPEED_GEN value) also explain why commit 6964494582f5 ("PCI: aardvark: Train link immediately after enabling training") somehow fixed detection of those problematic Compex cards with Atheros chips: if triggering link retraining (via PCI_EXP_LNKCTL_RL bit) was done immediately after enabling link training (via LINK_TRAINING_EN), it did nothing. If there was a specific delay, aardvark HW already initialized PCIe link and therefore triggering link retraining caused the above issue. Compex cards triggered link down event and disappeared from the PCIe bus. Commit f4c7d053d7f7 ("PCI: aardvark: Wait for endpoint to be ready before training link") added 100ms sleep before calling 'Start link training' command and explained that it is a requirement of PCI Express specification. But the code after this 100ms sleep was not doing 'Start link training', rather it triggered PCI_EXP_LNKCTL_RL bit via PCIe Root Bridge to put link into Recovery state. The required delay after fundamental reset is already done in function advk_pcie_wait_for_link() which also checks whether PCIe link is up. So after removing the code which triggers PCI_EXP_LNKCTL_RL bit on PCIe Root Bridge, there is no need to wait 100ms again. Remove the extra msleep() call and update comment about the delay required by the PCI Express specification. According to Marvell Armada 3700 Functional Specifications, Link training should be enabled via aardvark register LINK_TRAINING_EN after selecting PCIe generation and x1 lane. There is no need to disable it prior resetting card via PERST# signal. This disabling code was introduced in commit 5169a9851daa ("PCI: aardvark: Issue PERST via GPIO") as a workaround for some Atheros cards. It turns out that this also is Atheros specific issue and affects any PCIe controller, not only aardvark. Moreover this Atheros issue was triggered by juggling with PCI_EXP_LNKCTL_RL, LINK_TRAINING_EN and SPEED_GEN bits interleaved with sleeps. Now, after removing triggering PCI_EXP_LNKCTL_RL, there is no need to explicitly disable LINK_TRAINING_EN bit. So remove this code too. The problematic Compex cards described in previous git commits are correctly detected in advk_pcie_train_link() function even after applying all these changes. Note that with this patch, and also prior this patch, some NVMe disks which support PCIe GEN3 with 8 GT/s speed are negotiated only at the lowest link speed 2.5 GT/s, independently of SPEED_GEN value. After manually triggering PCI_EXP_LNKCTL_RL bit (e.g. from userspace via setpci), these NVMe disks change link speed to 5 GT/s when SPEED_GEN was configured to GEN2. This issue first needs to be properly investigated. I will send a fix in the future. On the other hand, some other GEN2 PCIe cards with 5 GT/s speed are autonomously by HW autonegotiated at full 5 GT/s speed without need of any software interaction. Armada 3700 Functional Specifications describes the following steps for link training: set SPEED_GEN to GEN2, enable LINK_TRAINING_EN, poll until link training is complete, trigger PCI_EXP_LNKCTL_RL, poll until signal rate is 5 GT/s, poll until link training is complete, enable ASPM L0s. The requirement for triggering PCI_EXP_LNKCTL_RL can be explained by the need to achieve 5 GT/s speed (as changing link speed is done by throw to recovery state entered by PCI_EXP_LNKCTL_RL) or maybe as a part of enabling ASPM L0s (but in this case ASPM L0s should have been enabled prior PCI_EXP_LNKCTL_RL). It is unknown why the original pci-aardvark.c driver was triggering PCI_EXP_LNKCTL_RL bit before waiting for the link to be up. This does not align with neither PCIe base specifications nor with Armada 3700 Functional Specification. (Note that in older versions of aardvark, this bit was called incorrectly PCIE_CORE_LINK_TRAINING, so this may be the reason.) It is also unknown why Armada 3700 Functional Specification says that it is needed to trigger PCI_EXP_LNKCTL_RL for GEN2 mode, as according to PCIe base specification 5 GT/s speed negotiation is supposed to be entirely autonomous, even if initial speed is 2.5 GT/s. [1] - https://lore.kernel.org/linux-pci/87h7l8axqp.fsf@toke.dk/ [2] - https://lore.kernel.org/linux-pci/20210326124326.21163-1-pali@kernel.org/ Link: https://lore.kernel.org/r/20211005180952.6812-12-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Reviewed-by: Marek Behún <kabel@kernel.org>
2021-10-05 18:09:50 +00:00
ret = advk_pcie_wait_for_link(pcie);
if (ret < 0)
dev_err(dev, "link never came up\n");
else
dev_info(dev, "link up\n");
}
PCI: aardvark: Configure PCIe resources from 'ranges' DT property In commit 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration") was removed aardvark PCIe outbound window configuration and commit description said that was recommended solution by HW designers. But that commit completely removed support for configuring PCIe IO resources without removing PCIe IO 'ranges' from DTS files. After that commit PCIe IO space started to be treated as PCIe MEM space and accessing it just caused kernel crash. Moreover implementation of PCIe outbound windows prior that commit was incorrect. It completely ignored offset between CPU address and PCIe bus address and expected that in DTS is CPU address always same as PCIe bus address without doing any checks. Also it completely ignored size of every PCIe resource specified in 'ranges' DTS property and expected that every PCIe resource has size 128 MB (also for PCIe IO range). Again without any check. Apparently none of PCIe resource has in DTS specified size of 128 MB. So it was completely broken and thanks to how aardvark mask works, configuration was completely ignored. This patch reverts back support for PCIe outbound window configuration but implementation is a new without issues mentioned above. PCIe outbound window is required when DTS specify in 'ranges' property non-zero offset between CPU and PCIe address space. To address recommendation by HW designers as specified in commit description of 6df6ba974a55, set default outbound parameters as PCIe MEM access without translation and therefore for this PCIe 'ranges' it is not needed to configure PCIe outbound window. For PCIe IO space is needed to configure aardvark PCIe outbound window. This patch fixes kernel crash when trying to access PCIe IO space. Link: https://lore.kernel.org/r/20210624215546.4015-2-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Cc: stable@vger.kernel.org # 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration")
2021-06-24 21:55:45 +00:00
/*
* Set PCIe address window register which could be used for memory
* mapping.
*/
static void advk_pcie_set_ob_win(struct advk_pcie *pcie, u8 win_num,
phys_addr_t match, phys_addr_t remap,
phys_addr_t mask, u32 actions)
{
advk_writel(pcie, OB_WIN_ENABLE |
lower_32_bits(match), OB_WIN_MATCH_LS(win_num));
advk_writel(pcie, upper_32_bits(match), OB_WIN_MATCH_MS(win_num));
advk_writel(pcie, lower_32_bits(remap), OB_WIN_REMAP_LS(win_num));
advk_writel(pcie, upper_32_bits(remap), OB_WIN_REMAP_MS(win_num));
advk_writel(pcie, lower_32_bits(mask), OB_WIN_MASK_LS(win_num));
advk_writel(pcie, upper_32_bits(mask), OB_WIN_MASK_MS(win_num));
advk_writel(pcie, actions, OB_WIN_ACTIONS(win_num));
}
static void advk_pcie_disable_ob_win(struct advk_pcie *pcie, u8 win_num)
{
advk_writel(pcie, 0, OB_WIN_MATCH_LS(win_num));
advk_writel(pcie, 0, OB_WIN_MATCH_MS(win_num));
advk_writel(pcie, 0, OB_WIN_REMAP_LS(win_num));
advk_writel(pcie, 0, OB_WIN_REMAP_MS(win_num));
advk_writel(pcie, 0, OB_WIN_MASK_LS(win_num));
advk_writel(pcie, 0, OB_WIN_MASK_MS(win_num));
advk_writel(pcie, 0, OB_WIN_ACTIONS(win_num));
}
static void advk_pcie_setup_hw(struct advk_pcie *pcie)
{
phys_addr_t msi_addr;
u32 reg;
PCI: aardvark: Configure PCIe resources from 'ranges' DT property In commit 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration") was removed aardvark PCIe outbound window configuration and commit description said that was recommended solution by HW designers. But that commit completely removed support for configuring PCIe IO resources without removing PCIe IO 'ranges' from DTS files. After that commit PCIe IO space started to be treated as PCIe MEM space and accessing it just caused kernel crash. Moreover implementation of PCIe outbound windows prior that commit was incorrect. It completely ignored offset between CPU address and PCIe bus address and expected that in DTS is CPU address always same as PCIe bus address without doing any checks. Also it completely ignored size of every PCIe resource specified in 'ranges' DTS property and expected that every PCIe resource has size 128 MB (also for PCIe IO range). Again without any check. Apparently none of PCIe resource has in DTS specified size of 128 MB. So it was completely broken and thanks to how aardvark mask works, configuration was completely ignored. This patch reverts back support for PCIe outbound window configuration but implementation is a new without issues mentioned above. PCIe outbound window is required when DTS specify in 'ranges' property non-zero offset between CPU and PCIe address space. To address recommendation by HW designers as specified in commit description of 6df6ba974a55, set default outbound parameters as PCIe MEM access without translation and therefore for this PCIe 'ranges' it is not needed to configure PCIe outbound window. For PCIe IO space is needed to configure aardvark PCIe outbound window. This patch fixes kernel crash when trying to access PCIe IO space. Link: https://lore.kernel.org/r/20210624215546.4015-2-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Cc: stable@vger.kernel.org # 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration")
2021-06-24 21:55:45 +00:00
int i;
/*
* Configure PCIe Reference clock. Direction is from the PCIe
* controller to the endpoint card, so enable transmitting of
* Reference clock differential signal off-chip and disable
* receiving off-chip differential signal.
*/
reg = advk_readl(pcie, PCIE_CORE_REF_CLK_REG);
reg |= PCIE_CORE_REF_CLK_TX_ENABLE;
reg &= ~PCIE_CORE_REF_CLK_RX_ENABLE;
advk_writel(pcie, reg, PCIE_CORE_REF_CLK_REG);
/* Set to Direct mode */
reg = advk_readl(pcie, CTRL_CONFIG_REG);
reg &= ~(CTRL_MODE_MASK << CTRL_MODE_SHIFT);
reg |= ((PCIE_CORE_MODE_DIRECT & CTRL_MODE_MASK) << CTRL_MODE_SHIFT);
advk_writel(pcie, reg, CTRL_CONFIG_REG);
/* Set PCI global control register to RC mode */
reg = advk_readl(pcie, PCIE_CORE_CTRL0_REG);
reg |= (IS_RC_MSK << IS_RC_SHIFT);
advk_writel(pcie, reg, PCIE_CORE_CTRL0_REG);
/*
* Replace incorrect PCI vendor id value 0x1b4b by correct value 0x11ab.
* VENDOR_ID_REG contains vendor id in low 16 bits and subsystem vendor
* id in high 16 bits. Updating this register changes readback value of
* read-only vendor id bits in PCIE_CORE_DEV_ID_REG register. Workaround
* for erratum 4.1: "The value of device and vendor ID is incorrect".
*/
reg = (PCI_VENDOR_ID_MARVELL << 16) | PCI_VENDOR_ID_MARVELL;
advk_writel(pcie, reg, VENDOR_ID_REG);
/*
* Change Class Code of PCI Bridge device to PCI Bridge (0x600400),
* because the default value is Mass storage controller (0x010400).
*
* Note that this Aardvark PCI Bridge does not have compliant Type 1
* Configuration Space and it even cannot be accessed via Aardvark's
* PCI config space access method. Something like config space is
* available in internal Aardvark registers starting at offset 0x0
* and is reported as Type 0. In range 0x10 - 0x34 it has totally
* different registers.
*
* Therefore driver uses emulation of PCI Bridge which emulates
* access to configuration space via internal Aardvark registers or
* emulated configuration buffer.
*/
reg = advk_readl(pcie, PCIE_CORE_DEV_REV_REG);
reg &= ~0xffffff00;
reg |= PCI_CLASS_BRIDGE_PCI_NORMAL << 8;
advk_writel(pcie, reg, PCIE_CORE_DEV_REV_REG);
/* Disable Root Bridge I/O space, memory space and bus mastering */
reg = advk_readl(pcie, PCIE_CORE_CMD_STATUS_REG);
reg &= ~(PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER);
advk_writel(pcie, reg, PCIE_CORE_CMD_STATUS_REG);
/* Set Advanced Error Capabilities and Control PF0 register */
reg = PCIE_CORE_ERR_CAPCTL_ECRC_CHK_TX |
PCIE_CORE_ERR_CAPCTL_ECRC_CHK_TX_EN |
PCIE_CORE_ERR_CAPCTL_ECRC_CHCK |
PCIE_CORE_ERR_CAPCTL_ECRC_CHCK_RCV;
advk_writel(pcie, reg, PCIE_CORE_ERR_CAPCTL_REG);
/* Set PCIe Device Control register */
reg = advk_readl(pcie, PCIE_CORE_PCIEXP_CAP + PCI_EXP_DEVCTL);
reg &= ~PCI_EXP_DEVCTL_RELAX_EN;
reg &= ~PCI_EXP_DEVCTL_NOSNOOP_EN;
reg &= ~PCI_EXP_DEVCTL_PAYLOAD;
reg &= ~PCI_EXP_DEVCTL_READRQ;
reg |= PCI_EXP_DEVCTL_PAYLOAD_512B;
reg |= PCI_EXP_DEVCTL_READRQ_512B;
advk_writel(pcie, reg, PCIE_CORE_PCIEXP_CAP + PCI_EXP_DEVCTL);
/* Program PCIe Control 2 to disable strict ordering */
reg = PCIE_CORE_CTRL2_RESERVED |
PCIE_CORE_CTRL2_TD_ENABLE;
advk_writel(pcie, reg, PCIE_CORE_CTRL2_REG);
/* Set lane X1 */
reg = advk_readl(pcie, PCIE_CORE_CTRL0_REG);
reg &= ~LANE_CNT_MSK;
reg |= LANE_COUNT_1;
advk_writel(pcie, reg, PCIE_CORE_CTRL0_REG);
/* Set MSI address */
msi_addr = virt_to_phys(pcie);
advk_writel(pcie, lower_32_bits(msi_addr), PCIE_MSI_ADDR_LOW_REG);
advk_writel(pcie, upper_32_bits(msi_addr), PCIE_MSI_ADDR_HIGH_REG);
/* Enable MSI */
reg = advk_readl(pcie, PCIE_CORE_CTRL2_REG);
reg |= PCIE_CORE_CTRL2_MSI_ENABLE;
advk_writel(pcie, reg, PCIE_CORE_CTRL2_REG);
/* Clear all interrupts */
advk_writel(pcie, PCIE_MSI_ALL_MASK, PCIE_MSI_STATUS_REG);
advk_writel(pcie, PCIE_ISR0_ALL_MASK, PCIE_ISR0_REG);
advk_writel(pcie, PCIE_ISR1_ALL_MASK, PCIE_ISR1_REG);
advk_writel(pcie, PCIE_IRQ_ALL_MASK, HOST_CTRL_INT_STATUS_REG);
/* Disable All ISR0/1 and MSI Sources */
advk_writel(pcie, PCIE_ISR0_ALL_MASK, PCIE_ISR0_MASK_REG);
advk_writel(pcie, PCIE_ISR1_ALL_MASK, PCIE_ISR1_MASK_REG);
advk_writel(pcie, PCIE_MSI_ALL_MASK, PCIE_MSI_MASK_REG);
/* Unmask summary MSI interrupt */
reg = advk_readl(pcie, PCIE_ISR0_MASK_REG);
reg &= ~PCIE_ISR0_MSI_INT_PENDING;
advk_writel(pcie, reg, PCIE_ISR0_MASK_REG);
PCI: aardvark: Fix support for PME requester on emulated bridge Enable aardvark PME interrupt unconditionally by unmasking it and read PME requester ID to emulated bridge config space immediately after receiving interrupt. PME requester ID is stored in the PCIE_MSG_LOG_REG register, which contains the last inbound message. So when new inbound message is received by HW (including non-PM), the content in PCIE_MSG_LOG_REG register is replaced by a new value. PCIe specification mandates that subsequent PMEs are kept pending until the PME Status Register bit is cleared by software by writing a 1b. Support for masking/unmasking PME interrupt on emulated bridge via PCI_EXP_RTCTL_PMEIE bit is now implemented only in emulated bridge config space, to ensure that we do not miss any aardvark PME interrupt. Reading of PCI_EXP_RTCAP and PCI_EXP_RTSTA registers is simplified as final value is now always stored into emulated bridge config space by the interrupt handler, so there is no need to implement support for these registers in read_pcie callback. Clearing of W1C bit PCI_EXP_RTSTA_PME is now also simplified as it is done by pci-bridge-emul.c code for emulated bridge config space. So there is no need to implement support for clearing this bit in write_pcie callback. Link: https://lore.kernel.org/r/20220110015018.26359-18-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
2022-01-10 01:50:12 +00:00
/* Unmask PME interrupt for processing of PME requester */
reg = advk_readl(pcie, PCIE_ISR0_MASK_REG);
reg &= ~PCIE_MSG_PM_PME_MASK;
advk_writel(pcie, reg, PCIE_ISR0_MASK_REG);
/* Enable summary interrupt for GIC SPI source */
reg = PCIE_IRQ_ALL_MASK & (~PCIE_IRQ_ENABLE_INTS_MASK);
advk_writel(pcie, reg, HOST_CTRL_INT_MASK_REG);
PCI: aardvark: Configure PCIe resources from 'ranges' DT property In commit 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration") was removed aardvark PCIe outbound window configuration and commit description said that was recommended solution by HW designers. But that commit completely removed support for configuring PCIe IO resources without removing PCIe IO 'ranges' from DTS files. After that commit PCIe IO space started to be treated as PCIe MEM space and accessing it just caused kernel crash. Moreover implementation of PCIe outbound windows prior that commit was incorrect. It completely ignored offset between CPU address and PCIe bus address and expected that in DTS is CPU address always same as PCIe bus address without doing any checks. Also it completely ignored size of every PCIe resource specified in 'ranges' DTS property and expected that every PCIe resource has size 128 MB (also for PCIe IO range). Again without any check. Apparently none of PCIe resource has in DTS specified size of 128 MB. So it was completely broken and thanks to how aardvark mask works, configuration was completely ignored. This patch reverts back support for PCIe outbound window configuration but implementation is a new without issues mentioned above. PCIe outbound window is required when DTS specify in 'ranges' property non-zero offset between CPU and PCIe address space. To address recommendation by HW designers as specified in commit description of 6df6ba974a55, set default outbound parameters as PCIe MEM access without translation and therefore for this PCIe 'ranges' it is not needed to configure PCIe outbound window. For PCIe IO space is needed to configure aardvark PCIe outbound window. This patch fixes kernel crash when trying to access PCIe IO space. Link: https://lore.kernel.org/r/20210624215546.4015-2-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Cc: stable@vger.kernel.org # 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration")
2021-06-24 21:55:45 +00:00
/*
* Enable AXI address window location generation:
* When it is enabled, the default outbound window
* configurations (Default User Field: 0xD0074CFC)
* are used to transparent address translation for
* the outbound transactions. Thus, PCIe address
* windows are not required for transparent memory
* access when default outbound window configuration
* is set for memory access.
*/
reg = advk_readl(pcie, PCIE_CORE_CTRL2_REG);
reg |= PCIE_CORE_CTRL2_OB_WIN_ENABLE;
advk_writel(pcie, reg, PCIE_CORE_CTRL2_REG);
PCI: aardvark: Configure PCIe resources from 'ranges' DT property In commit 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration") was removed aardvark PCIe outbound window configuration and commit description said that was recommended solution by HW designers. But that commit completely removed support for configuring PCIe IO resources without removing PCIe IO 'ranges' from DTS files. After that commit PCIe IO space started to be treated as PCIe MEM space and accessing it just caused kernel crash. Moreover implementation of PCIe outbound windows prior that commit was incorrect. It completely ignored offset between CPU address and PCIe bus address and expected that in DTS is CPU address always same as PCIe bus address without doing any checks. Also it completely ignored size of every PCIe resource specified in 'ranges' DTS property and expected that every PCIe resource has size 128 MB (also for PCIe IO range). Again without any check. Apparently none of PCIe resource has in DTS specified size of 128 MB. So it was completely broken and thanks to how aardvark mask works, configuration was completely ignored. This patch reverts back support for PCIe outbound window configuration but implementation is a new without issues mentioned above. PCIe outbound window is required when DTS specify in 'ranges' property non-zero offset between CPU and PCIe address space. To address recommendation by HW designers as specified in commit description of 6df6ba974a55, set default outbound parameters as PCIe MEM access without translation and therefore for this PCIe 'ranges' it is not needed to configure PCIe outbound window. For PCIe IO space is needed to configure aardvark PCIe outbound window. This patch fixes kernel crash when trying to access PCIe IO space. Link: https://lore.kernel.org/r/20210624215546.4015-2-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Cc: stable@vger.kernel.org # 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration")
2021-06-24 21:55:45 +00:00
/*
* Set memory access in Default User Field so it
* is not required to configure PCIe address for
* transparent memory access.
*/
advk_writel(pcie, OB_WIN_TYPE_MEM, OB_WIN_DEFAULT_ACTIONS);
/*
* Bypass the address window mapping for PIO:
* Since PIO access already contains all required
* info over AXI interface by PIO registers, the
* address window is not required.
*/
reg = advk_readl(pcie, PIO_CTRL);
reg |= PIO_CTRL_ADDR_WIN_DISABLE;
advk_writel(pcie, reg, PIO_CTRL);
PCI: aardvark: Configure PCIe resources from 'ranges' DT property In commit 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration") was removed aardvark PCIe outbound window configuration and commit description said that was recommended solution by HW designers. But that commit completely removed support for configuring PCIe IO resources without removing PCIe IO 'ranges' from DTS files. After that commit PCIe IO space started to be treated as PCIe MEM space and accessing it just caused kernel crash. Moreover implementation of PCIe outbound windows prior that commit was incorrect. It completely ignored offset between CPU address and PCIe bus address and expected that in DTS is CPU address always same as PCIe bus address without doing any checks. Also it completely ignored size of every PCIe resource specified in 'ranges' DTS property and expected that every PCIe resource has size 128 MB (also for PCIe IO range). Again without any check. Apparently none of PCIe resource has in DTS specified size of 128 MB. So it was completely broken and thanks to how aardvark mask works, configuration was completely ignored. This patch reverts back support for PCIe outbound window configuration but implementation is a new without issues mentioned above. PCIe outbound window is required when DTS specify in 'ranges' property non-zero offset between CPU and PCIe address space. To address recommendation by HW designers as specified in commit description of 6df6ba974a55, set default outbound parameters as PCIe MEM access without translation and therefore for this PCIe 'ranges' it is not needed to configure PCIe outbound window. For PCIe IO space is needed to configure aardvark PCIe outbound window. This patch fixes kernel crash when trying to access PCIe IO space. Link: https://lore.kernel.org/r/20210624215546.4015-2-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Cc: stable@vger.kernel.org # 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration")
2021-06-24 21:55:45 +00:00
/*
* Configure PCIe address windows for non-memory or
* non-transparent access as by default PCIe uses
* transparent memory access.
*/
for (i = 0; i < pcie->wins_count; i++)
advk_pcie_set_ob_win(pcie, i,
pcie->wins[i].match, pcie->wins[i].remap,
pcie->wins[i].mask, pcie->wins[i].actions);
/* Disable remaining PCIe outbound windows */
for (i = pcie->wins_count; i < OB_WIN_COUNT; i++)
advk_pcie_disable_ob_win(pcie, i);
advk_pcie_train_link(pcie);
}
static int advk_pcie_check_pio_status(struct advk_pcie *pcie, bool allow_crs, u32 *val)
{
struct device *dev = &pcie->pdev->dev;
u32 reg;
unsigned int status;
char *strcomp_status, *str_posted;
int ret;
reg = advk_readl(pcie, PIO_STAT);
status = (reg & PIO_COMPLETION_STATUS_MASK) >>
PIO_COMPLETION_STATUS_SHIFT;
/*
* According to HW spec, the PIO status check sequence as below:
* 1) even if COMPLETION_STATUS(bit9:7) indicates successful,
* it still needs to check Error Status(bit11), only when this bit
* indicates no error happen, the operation is successful.
* 2) value Unsupported Request(1) of COMPLETION_STATUS(bit9:7) only
* means a PIO write error, and for PIO read it is successful with
* a read value of 0xFFFFFFFF.
* 3) value Completion Retry Status(CRS) of COMPLETION_STATUS(bit9:7)
* only means a PIO write error, and for PIO read it is successful
* with a read value of 0xFFFF0001.
* 4) value Completer Abort (CA) of COMPLETION_STATUS(bit9:7) means
* error for both PIO read and PIO write operation.
* 5) other errors are indicated as 'unknown'.
*/
switch (status) {
case PIO_COMPLETION_STATUS_OK:
if (reg & PIO_ERR_STATUS) {
strcomp_status = "COMP_ERR";
ret = -EFAULT;
break;
}
/* Get the read result */
if (val)
*val = advk_readl(pcie, PIO_RD_DATA);
/* No error */
strcomp_status = NULL;
ret = 0;
break;
case PIO_COMPLETION_STATUS_UR:
strcomp_status = "UR";
ret = -EOPNOTSUPP;
break;
case PIO_COMPLETION_STATUS_CRS:
if (allow_crs && val) {
/* PCIe r4.0, sec 2.3.2, says:
* If CRS Software Visibility is enabled:
* For a Configuration Read Request that includes both
* bytes of the Vendor ID field of a device Function's
* Configuration Space Header, the Root Complex must
* complete the Request to the host by returning a
* read-data value of 0001h for the Vendor ID field and
* all '1's for any additional bytes included in the
* request.
*
* So CRS in this case is not an error status.
*/
*val = CFG_RD_CRS_VAL;
strcomp_status = NULL;
ret = 0;
break;
}
/* PCIe r4.0, sec 2.3.2, says:
* If CRS Software Visibility is not enabled, the Root Complex
* must re-issue the Configuration Request as a new Request.
* If CRS Software Visibility is enabled: For a Configuration
* Write Request or for any other Configuration Read Request,
* the Root Complex must re-issue the Configuration Request as
* a new Request.
* A Root Complex implementation may choose to limit the number
* of Configuration Request/CRS Completion Status loops before
* determining that something is wrong with the target of the
* Request and taking appropriate action, e.g., complete the
* Request to the host as a failed transaction.
*
* So return -EAGAIN and caller (pci-aardvark.c driver) will
* re-issue request again up to the PIO_RETRY_CNT retries.
*/
strcomp_status = "CRS";
ret = -EAGAIN;
break;
case PIO_COMPLETION_STATUS_CA:
strcomp_status = "CA";
ret = -ECANCELED;
break;
default:
strcomp_status = "Unknown";
ret = -EINVAL;
break;
}
if (!strcomp_status)
return ret;
if (reg & PIO_NON_POSTED_REQ)
str_posted = "Non-posted";
else
str_posted = "Posted";
dev_dbg(dev, "%s PIO Response Status: %s, %#x @ %#x\n",
str_posted, strcomp_status, reg, advk_readl(pcie, PIO_ADDR_LS));
return ret;
}
static int advk_pcie_wait_pio(struct advk_pcie *pcie)
{
struct device *dev = &pcie->pdev->dev;
int i;
for (i = 1; i <= PIO_RETRY_CNT; i++) {
u32 start, isr;
start = advk_readl(pcie, PIO_START);
isr = advk_readl(pcie, PIO_ISR);
if (!start && isr)
return i;
udelay(PIO_RETRY_DELAY);
}
PCI: aardvark: Fix kernel panic during PIO transfer Trying to start a new PIO transfer by writing value 0 in PIO_START register when previous transfer has not yet completed (which is indicated by value 1 in PIO_START) causes an External Abort on CPU, which results in kernel panic: SError Interrupt on CPU0, code 0xbf000002 -- SError Kernel panic - not syncing: Asynchronous SError Interrupt To prevent kernel panic, it is required to reject a new PIO transfer when previous one has not finished yet. If previous PIO transfer is not finished yet, the kernel may issue a new PIO request only if the previous PIO transfer timed out. In the past the root cause of this issue was incorrectly identified (as it often happens during link retraining or after link down event) and special hack was implemented in Trusted Firmware to catch all SError events in EL3, to ignore errors with code 0xbf000002 and not forwarding any other errors to kernel and instead throw panic from EL3 Trusted Firmware handler. Links to discussion and patches about this issue: https://git.trustedfirmware.org/TF-A/trusted-firmware-a.git/commit/?id=3c7dcdac5c50 https://lore.kernel.org/linux-pci/20190316161243.29517-1-repk@triplefau.lt/ https://lore.kernel.org/linux-pci/971be151d24312cc533989a64bd454b4@www.loen.fr/ https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/1541 But the real cause was the fact that during link retraining or after link down event the PIO transfer may take longer time, up to the 1.44s until it times out. This increased probability that a new PIO transfer would be issued by kernel while previous one has not finished yet. After applying this change into the kernel, it is possible to revert the mentioned TF-A hack and SError events do not have to be caught in TF-A EL3. Link: https://lore.kernel.org/r/20210608203655.31228-1-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com> Reviewed-by: Marek Behún <kabel@kernel.org> Cc: stable@vger.kernel.org # 7fbcb5da811b ("PCI: aardvark: Don't rely on jiffies while holding spinlock")
2021-06-08 20:36:55 +00:00
dev_err(dev, "PIO read/write transfer time out\n");
return -ETIMEDOUT;
}
static pci_bridge_emul_read_status_t
advk_pci_bridge_emul_base_conf_read(struct pci_bridge_emul *bridge,
int reg, u32 *value)
{
struct advk_pcie *pcie = bridge->data;
switch (reg) {
case PCI_COMMAND:
*value = advk_readl(pcie, PCIE_CORE_CMD_STATUS_REG);
return PCI_BRIDGE_EMUL_HANDLED;
case PCI_INTERRUPT_LINE: {
/*
* From the whole 32bit register we support reading from HW only
* two bits: PCI_BRIDGE_CTL_BUS_RESET and PCI_BRIDGE_CTL_SERR.
* Other bits are retrieved only from emulated config buffer.
*/
__le32 *cfgspace = (__le32 *)&bridge->conf;
u32 val = le32_to_cpu(cfgspace[PCI_INTERRUPT_LINE / 4]);
if (advk_readl(pcie, PCIE_ISR0_MASK_REG) & PCIE_ISR0_ERR_MASK)
val &= ~(PCI_BRIDGE_CTL_SERR << 16);
else
val |= PCI_BRIDGE_CTL_SERR << 16;
if (advk_readl(pcie, PCIE_CORE_CTRL1_REG) & HOT_RESET_GEN)
val |= PCI_BRIDGE_CTL_BUS_RESET << 16;
else
val &= ~(PCI_BRIDGE_CTL_BUS_RESET << 16);
*value = val;
return PCI_BRIDGE_EMUL_HANDLED;
}
default:
return PCI_BRIDGE_EMUL_NOT_HANDLED;
}
}
static void
advk_pci_bridge_emul_base_conf_write(struct pci_bridge_emul *bridge,
int reg, u32 old, u32 new, u32 mask)
{
struct advk_pcie *pcie = bridge->data;
switch (reg) {
case PCI_COMMAND:
advk_writel(pcie, new, PCIE_CORE_CMD_STATUS_REG);
break;
case PCI_INTERRUPT_LINE:
/*
* According to Figure 6-3: Pseudo Logic Diagram for Error
* Message Controls in PCIe base specification, SERR# Enable bit
* in Bridge Control register enable receiving of ERR_* messages
*/
if (mask & (PCI_BRIDGE_CTL_SERR << 16)) {
u32 val = advk_readl(pcie, PCIE_ISR0_MASK_REG);
if (new & (PCI_BRIDGE_CTL_SERR << 16))
val &= ~PCIE_ISR0_ERR_MASK;
else
val |= PCIE_ISR0_ERR_MASK;
advk_writel(pcie, val, PCIE_ISR0_MASK_REG);
}
if (mask & (PCI_BRIDGE_CTL_BUS_RESET << 16)) {
u32 val = advk_readl(pcie, PCIE_CORE_CTRL1_REG);
if (new & (PCI_BRIDGE_CTL_BUS_RESET << 16))
val |= HOT_RESET_GEN;
else
val &= ~HOT_RESET_GEN;
advk_writel(pcie, val, PCIE_CORE_CTRL1_REG);
}
break;
default:
break;
}
}
static pci_bridge_emul_read_status_t
advk_pci_bridge_emul_pcie_conf_read(struct pci_bridge_emul *bridge,
int reg, u32 *value)
{
struct advk_pcie *pcie = bridge->data;
switch (reg) {
PCI: aardvark: Fix support for PME requester on emulated bridge Enable aardvark PME interrupt unconditionally by unmasking it and read PME requester ID to emulated bridge config space immediately after receiving interrupt. PME requester ID is stored in the PCIE_MSG_LOG_REG register, which contains the last inbound message. So when new inbound message is received by HW (including non-PM), the content in PCIE_MSG_LOG_REG register is replaced by a new value. PCIe specification mandates that subsequent PMEs are kept pending until the PME Status Register bit is cleared by software by writing a 1b. Support for masking/unmasking PME interrupt on emulated bridge via PCI_EXP_RTCTL_PMEIE bit is now implemented only in emulated bridge config space, to ensure that we do not miss any aardvark PME interrupt. Reading of PCI_EXP_RTCAP and PCI_EXP_RTSTA registers is simplified as final value is now always stored into emulated bridge config space by the interrupt handler, so there is no need to implement support for these registers in read_pcie callback. Clearing of W1C bit PCI_EXP_RTSTA_PME is now also simplified as it is done by pci-bridge-emul.c code for emulated bridge config space. So there is no need to implement support for clearing this bit in write_pcie callback. Link: https://lore.kernel.org/r/20220110015018.26359-18-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
2022-01-10 01:50:12 +00:00
/*
* PCI_EXP_SLTCAP, PCI_EXP_SLTCTL, PCI_EXP_RTCTL and PCI_EXP_RTSTA are
* also supported, but do not need to be handled here, because their
* values are stored in emulated config space buffer, and we read them
* from there when needed.
PCI: aardvark: Fix support for PME requester on emulated bridge Enable aardvark PME interrupt unconditionally by unmasking it and read PME requester ID to emulated bridge config space immediately after receiving interrupt. PME requester ID is stored in the PCIE_MSG_LOG_REG register, which contains the last inbound message. So when new inbound message is received by HW (including non-PM), the content in PCIE_MSG_LOG_REG register is replaced by a new value. PCIe specification mandates that subsequent PMEs are kept pending until the PME Status Register bit is cleared by software by writing a 1b. Support for masking/unmasking PME interrupt on emulated bridge via PCI_EXP_RTCTL_PMEIE bit is now implemented only in emulated bridge config space, to ensure that we do not miss any aardvark PME interrupt. Reading of PCI_EXP_RTCAP and PCI_EXP_RTSTA registers is simplified as final value is now always stored into emulated bridge config space by the interrupt handler, so there is no need to implement support for these registers in read_pcie callback. Clearing of W1C bit PCI_EXP_RTSTA_PME is now also simplified as it is done by pci-bridge-emul.c code for emulated bridge config space. So there is no need to implement support for clearing this bit in write_pcie callback. Link: https://lore.kernel.org/r/20220110015018.26359-18-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
2022-01-10 01:50:12 +00:00
*/
case PCI_EXP_LNKCAP: {
u32 val = advk_readl(pcie, PCIE_CORE_PCIEXP_CAP + reg);
/*
* PCI_EXP_LNKCAP_DLLLARC bit is hardwired in aardvark HW to 0.
* But support for PCI_EXP_LNKSTA_DLLLA is emulated via ltssm
* state so explicitly enable PCI_EXP_LNKCAP_DLLLARC flag.
*/
val |= PCI_EXP_LNKCAP_DLLLARC;
*value = val;
return PCI_BRIDGE_EMUL_HANDLED;
}
case PCI_EXP_LNKCTL: {
/* u32 contains both PCI_EXP_LNKCTL and PCI_EXP_LNKSTA */
u32 val = advk_readl(pcie, PCIE_CORE_PCIEXP_CAP + reg) &
~(PCI_EXP_LNKSTA_LT << 16);
if (advk_pcie_link_training(pcie))
val |= (PCI_EXP_LNKSTA_LT << 16);
if (advk_pcie_link_active(pcie))
val |= (PCI_EXP_LNKSTA_DLLLA << 16);
*value = val;
return PCI_BRIDGE_EMUL_HANDLED;
}
case PCI_EXP_DEVCAP:
case PCI_EXP_DEVCTL:
case PCI_EXP_DEVCAP2:
case PCI_EXP_DEVCTL2:
case PCI_EXP_LNKCAP2:
case PCI_EXP_LNKCTL2:
*value = advk_readl(pcie, PCIE_CORE_PCIEXP_CAP + reg);
return PCI_BRIDGE_EMUL_HANDLED;
default:
return PCI_BRIDGE_EMUL_NOT_HANDLED;
}
}
static void
advk_pci_bridge_emul_pcie_conf_write(struct pci_bridge_emul *bridge,
int reg, u32 old, u32 new, u32 mask)
{
struct advk_pcie *pcie = bridge->data;
switch (reg) {
case PCI_EXP_LNKCTL:
advk_writel(pcie, new, PCIE_CORE_PCIEXP_CAP + reg);
if (new & PCI_EXP_LNKCTL_RL)
advk_pcie_wait_for_retrain(pcie);
break;
case PCI_EXP_RTCTL: {
PCI: aardvark: Fix support for PME requester on emulated bridge Enable aardvark PME interrupt unconditionally by unmasking it and read PME requester ID to emulated bridge config space immediately after receiving interrupt. PME requester ID is stored in the PCIE_MSG_LOG_REG register, which contains the last inbound message. So when new inbound message is received by HW (including non-PM), the content in PCIE_MSG_LOG_REG register is replaced by a new value. PCIe specification mandates that subsequent PMEs are kept pending until the PME Status Register bit is cleared by software by writing a 1b. Support for masking/unmasking PME interrupt on emulated bridge via PCI_EXP_RTCTL_PMEIE bit is now implemented only in emulated bridge config space, to ensure that we do not miss any aardvark PME interrupt. Reading of PCI_EXP_RTCAP and PCI_EXP_RTSTA registers is simplified as final value is now always stored into emulated bridge config space by the interrupt handler, so there is no need to implement support for these registers in read_pcie callback. Clearing of W1C bit PCI_EXP_RTSTA_PME is now also simplified as it is done by pci-bridge-emul.c code for emulated bridge config space. So there is no need to implement support for clearing this bit in write_pcie callback. Link: https://lore.kernel.org/r/20220110015018.26359-18-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
2022-01-10 01:50:12 +00:00
u16 rootctl = le16_to_cpu(bridge->pcie_conf.rootctl);
/* Only emulation of PMEIE and CRSSVE bits is provided */
rootctl &= PCI_EXP_RTCTL_PMEIE | PCI_EXP_RTCTL_CRSSVE;
bridge->pcie_conf.rootctl = cpu_to_le16(rootctl);
break;
}
PCI: aardvark: Fix support for PME requester on emulated bridge Enable aardvark PME interrupt unconditionally by unmasking it and read PME requester ID to emulated bridge config space immediately after receiving interrupt. PME requester ID is stored in the PCIE_MSG_LOG_REG register, which contains the last inbound message. So when new inbound message is received by HW (including non-PM), the content in PCIE_MSG_LOG_REG register is replaced by a new value. PCIe specification mandates that subsequent PMEs are kept pending until the PME Status Register bit is cleared by software by writing a 1b. Support for masking/unmasking PME interrupt on emulated bridge via PCI_EXP_RTCTL_PMEIE bit is now implemented only in emulated bridge config space, to ensure that we do not miss any aardvark PME interrupt. Reading of PCI_EXP_RTCAP and PCI_EXP_RTSTA registers is simplified as final value is now always stored into emulated bridge config space by the interrupt handler, so there is no need to implement support for these registers in read_pcie callback. Clearing of W1C bit PCI_EXP_RTSTA_PME is now also simplified as it is done by pci-bridge-emul.c code for emulated bridge config space. So there is no need to implement support for clearing this bit in write_pcie callback. Link: https://lore.kernel.org/r/20220110015018.26359-18-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
2022-01-10 01:50:12 +00:00
/*
* PCI_EXP_RTSTA is also supported, but does not need to be handled
* here, because its value is stored in emulated config space buffer,
* and we write it there when needed.
*/
case PCI_EXP_DEVCTL:
case PCI_EXP_DEVCTL2:
case PCI_EXP_LNKCTL2:
advk_writel(pcie, new, PCIE_CORE_PCIEXP_CAP + reg);
break;
default:
break;
}
}
static pci_bridge_emul_read_status_t
advk_pci_bridge_emul_ext_conf_read(struct pci_bridge_emul *bridge,
int reg, u32 *value)
{
struct advk_pcie *pcie = bridge->data;
switch (reg) {
case 0:
*value = advk_readl(pcie, PCIE_CORE_PCIERR_CAP + reg);
/*
* PCI_EXT_CAP_NEXT bits are set to offset 0x150, but Armada
* 3700 Functional Specification does not document registers
* at those addresses.
*
* Thus we clear PCI_EXT_CAP_NEXT bits to make Advanced Error
* Reporting Capability header the last Extended Capability.
* If we obtain documentation for those registers in the
* future, this can be changed.
*/
*value &= 0x000fffff;
return PCI_BRIDGE_EMUL_HANDLED;
case PCI_ERR_UNCOR_STATUS:
case PCI_ERR_UNCOR_MASK:
case PCI_ERR_UNCOR_SEVER:
case PCI_ERR_COR_STATUS:
case PCI_ERR_COR_MASK:
case PCI_ERR_CAP:
case PCI_ERR_HEADER_LOG + 0:
case PCI_ERR_HEADER_LOG + 4:
case PCI_ERR_HEADER_LOG + 8:
case PCI_ERR_HEADER_LOG + 12:
case PCI_ERR_ROOT_COMMAND:
case PCI_ERR_ROOT_STATUS:
case PCI_ERR_ROOT_ERR_SRC:
*value = advk_readl(pcie, PCIE_CORE_PCIERR_CAP + reg);
return PCI_BRIDGE_EMUL_HANDLED;
default:
return PCI_BRIDGE_EMUL_NOT_HANDLED;
}
}
static void
advk_pci_bridge_emul_ext_conf_write(struct pci_bridge_emul *bridge,
int reg, u32 old, u32 new, u32 mask)
{
struct advk_pcie *pcie = bridge->data;
switch (reg) {
/* These are W1C registers, so clear other bits */
case PCI_ERR_UNCOR_STATUS:
case PCI_ERR_COR_STATUS:
case PCI_ERR_ROOT_STATUS:
new &= mask;
fallthrough;
case PCI_ERR_UNCOR_MASK:
case PCI_ERR_UNCOR_SEVER:
case PCI_ERR_COR_MASK:
case PCI_ERR_CAP:
case PCI_ERR_HEADER_LOG + 0:
case PCI_ERR_HEADER_LOG + 4:
case PCI_ERR_HEADER_LOG + 8:
case PCI_ERR_HEADER_LOG + 12:
case PCI_ERR_ROOT_COMMAND:
case PCI_ERR_ROOT_ERR_SRC:
advk_writel(pcie, new, PCIE_CORE_PCIERR_CAP + reg);
break;
default:
break;
}
}
static const struct pci_bridge_emul_ops advk_pci_bridge_emul_ops = {
.read_base = advk_pci_bridge_emul_base_conf_read,
.write_base = advk_pci_bridge_emul_base_conf_write,
.read_pcie = advk_pci_bridge_emul_pcie_conf_read,
.write_pcie = advk_pci_bridge_emul_pcie_conf_write,
.read_ext = advk_pci_bridge_emul_ext_conf_read,
.write_ext = advk_pci_bridge_emul_ext_conf_write,
};
/*
* Initialize the configuration space of the PCI-to-PCI bridge
* associated with the given PCIe interface.
*/
static int advk_sw_pci_bridge_init(struct advk_pcie *pcie)
{
struct pci_bridge_emul *bridge = &pcie->bridge;
bridge->conf.vendor =
cpu_to_le16(advk_readl(pcie, PCIE_CORE_DEV_ID_REG) & 0xffff);
bridge->conf.device =
cpu_to_le16(advk_readl(pcie, PCIE_CORE_DEV_ID_REG) >> 16);
bridge->conf.class_revision =
cpu_to_le32(advk_readl(pcie, PCIE_CORE_DEV_REV_REG) & 0xff);
/* Support 32 bits I/O addressing */
bridge->conf.iobase = PCI_IO_RANGE_TYPE_32;
bridge->conf.iolimit = PCI_IO_RANGE_TYPE_32;
/* Support 64 bits memory pref */
bridge->conf.pref_mem_base = cpu_to_le16(PCI_PREF_RANGE_TYPE_64);
bridge->conf.pref_mem_limit = cpu_to_le16(PCI_PREF_RANGE_TYPE_64);
/* Support interrupt A for MSI feature */
bridge->conf.intpin = PCI_INTERRUPT_INTA;
/*
* Aardvark HW provides PCIe Capability structure in version 2 and
* indicate slot support, which is emulated.
*/
bridge->pcie_conf.cap = cpu_to_le16(2 | PCI_EXP_FLAGS_SLOT);
/*
* Set Presence Detect State bit permanently since there is no support
* for unplugging the card nor detecting whether it is plugged. (If a
* platform exists in the future that supports it, via a GPIO for
* example, it should be implemented via this bit.)
*
* Set physical slot number to 1 since there is only one port and zero
* value is reserved for ports within the same silicon as Root Port
* which is not our case.
*/
bridge->pcie_conf.slotcap = cpu_to_le32(FIELD_PREP(PCI_EXP_SLTCAP_PSN,
1));
bridge->pcie_conf.slotsta = cpu_to_le16(PCI_EXP_SLTSTA_PDS);
/* Indicates supports for Completion Retry Status */
bridge->pcie_conf.rootcap = cpu_to_le16(PCI_EXP_RTCAP_CRSVIS);
bridge->subsystem_vendor_id = advk_readl(pcie, PCIE_CORE_SSDEV_ID_REG) & 0xffff;
bridge->subsystem_id = advk_readl(pcie, PCIE_CORE_SSDEV_ID_REG) >> 16;
bridge->has_pcie = true;
bridge->pcie_start = PCIE_CORE_PCIEXP_CAP;
bridge->data = pcie;
bridge->ops = &advk_pci_bridge_emul_ops;
return pci_bridge_emul_init(bridge, 0);
}
static bool advk_pcie_valid_device(struct advk_pcie *pcie, struct pci_bus *bus,
int devfn)
{
if (pci_is_root_bus(bus) && PCI_SLOT(devfn) != 0)
return false;
/*
* If the link goes down after we check for link-up, we have a problem:
* if a PIO request is executed while link-down, the whole controller
* gets stuck in a non-functional state, and even after link comes up
* again, PIO requests won't work anymore, and a reset of the whole PCIe
* controller is needed. Therefore we need to prevent sending PIO
* requests while the link is down.
*/
Merge branch 'pci/host-probe-refactor' - Use pci_host_bridge.windows list directly instead of splicing in a temporary list for cadence, mvebu, host-common (Rob Herring) - Use pci_host_probe() instead of open-coding all the pieces for altera, brcmstb, iproc, mobiveil, rcar, rockchip, tegra, v3, versatile, xgene, xilinx, xilinx-nwl (Rob Herring) - Convert to devm_platform_ioremap_resource_byname() instead of open-coding platform_get_resource_byname() and devm_ioremap_resource() for altera, cadence, mediatek, rockchip, tegra, xgene (Dejin Zheng) - Convert to devm_platform_ioremap_resource() instead of open-coding platform_get_resource() and devm_ioremap_resource() for aardvark, brcmstb, exynos, ftpci100, versatile (Dejin Zheng) - Remove redundant error messages from devm_pci_remap_cfg_resource() callers (Dejin Zheng) - Drop useless PCI_ENABLE_PROC_DOMAINS from versatile driver (Rob Herring) - Default host bridge parent device to the platform device (Rob Herring) - Drop unnecessary zeroing of host bridge fields (Rob Herring) - Use pci_is_root_bus() instead of tracking root bus number separately in aardvark, designware (imx6, keystone, designware-host), mobiveil, xilinx-nwl, xilinx, rockchip, rcar (Rob Herring) - Set host bridge bus number in pci_scan_root_bus_bridge() instead of each driver for aardvark, designware-host, host-common, mediatek, rcar, tegra, v3-semi (Rob Herring) - Use bridge resources instead of parsing DT 'ranges' again for cadence (Rob Herring) - Remove private bus number and range from cadence (Rob Herring) - Use devm_pci_alloc_host_bridge() to simplify rcar (Rob Herring) - Use struct pci_host_bridge.windows list directly rather than a temporary (Rob Herring) - Reduce OF "missing non-prefetchable window" from error to warning message (Rob Herring) - Convert rcar-gen2 from old Arm-specific pci_common_init_dev() to new arch-independent interfaces (Rob Herring) - Move DT resource setup into devm_pci_alloc_host_bridge() (Rob Herring) - Set bridge map_irq and swizzle_irq to default functions; drivers that don't support legacy IRQs (iproc) need to undo this (Rob Herring) * pci/host-probe-refactor: PCI: Set bridge map_irq and swizzle_irq to default functions PCI: Move DT resource setup into devm_pci_alloc_host_bridge() PCI: rcar-gen2: Convert to use modern host bridge probe functions PCI: of: Reduce missing non-prefetchable memory region to a warning PCI: rcar: Use struct pci_host_bridge.windows list directly PCI: rcar: Use devm_pci_alloc_host_bridge() PCI: cadence: Remove private bus number and range storage PCI: cadence: Use bridge resources for outbound window setup PCI: Move setting pci_host_bridge.busnr out of host drivers PCI: rcar: Use pci_is_root_bus() to check if bus is root bus PCI: rockchip: Use pci_is_root_bus() to check if bus is root bus PCI: xilinx: Use pci_is_root_bus() to check if bus is root bus PCI: xilinx-nwl: Use pci_is_root_bus() to check if bus is root bus PCI: mobiveil: Use pci_is_root_bus() to check if bus is root bus PCI: designware: Use pci_is_root_bus() to check if bus is root bus PCI: aardvark: Use pci_is_root_bus() to check if bus is root bus PCI: Drop unnecessary zeroing of bridge fields PCI: Set default bridge parent device PCI: versatile: Drop flag PCI_ENABLE_PROC_DOMAINS PCI: controller: Remove duplicate error message PCI: controller: Convert to devm_platform_ioremap_resource() PCI: controller: Convert to devm_platform_ioremap_resource_byname() PCI: xilinx: Use pci_host_probe() to register host PCI: xilinx-nwl: Use pci_host_probe() to register host PCI: rockchip: Use pci_host_probe() to register host PCI: rcar: Use pci_host_probe() to register host PCI: iproc: Use pci_host_probe() to register host PCI: altera: Use pci_host_probe() to register host PCI: xgene: Use pci_host_probe() to register host PCI: versatile: Use pci_host_probe() to register host PCI: v3: Use pci_host_probe() to register host PCI: tegra: Use pci_host_probe() to register host PCI: mobiveil: Use pci_host_probe() to register host PCI: brcmstb: Use pci_host_probe() to register host PCI: host-common: Use struct pci_host_bridge.windows list directly PCI: mvebu: Use struct pci_host_bridge.windows list directly PCI: cadence: Use struct pci_host_bridge.windows list directly # Conflicts: # drivers/pci/controller/cadence/pcie-cadence-host.c
2020-08-05 23:24:21 +00:00
if (!pci_is_root_bus(bus) && !advk_pcie_link_up(pcie))
return false;
return true;
}
PCI: aardvark: Fix kernel panic during PIO transfer Trying to start a new PIO transfer by writing value 0 in PIO_START register when previous transfer has not yet completed (which is indicated by value 1 in PIO_START) causes an External Abort on CPU, which results in kernel panic: SError Interrupt on CPU0, code 0xbf000002 -- SError Kernel panic - not syncing: Asynchronous SError Interrupt To prevent kernel panic, it is required to reject a new PIO transfer when previous one has not finished yet. If previous PIO transfer is not finished yet, the kernel may issue a new PIO request only if the previous PIO transfer timed out. In the past the root cause of this issue was incorrectly identified (as it often happens during link retraining or after link down event) and special hack was implemented in Trusted Firmware to catch all SError events in EL3, to ignore errors with code 0xbf000002 and not forwarding any other errors to kernel and instead throw panic from EL3 Trusted Firmware handler. Links to discussion and patches about this issue: https://git.trustedfirmware.org/TF-A/trusted-firmware-a.git/commit/?id=3c7dcdac5c50 https://lore.kernel.org/linux-pci/20190316161243.29517-1-repk@triplefau.lt/ https://lore.kernel.org/linux-pci/971be151d24312cc533989a64bd454b4@www.loen.fr/ https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/1541 But the real cause was the fact that during link retraining or after link down event the PIO transfer may take longer time, up to the 1.44s until it times out. This increased probability that a new PIO transfer would be issued by kernel while previous one has not finished yet. After applying this change into the kernel, it is possible to revert the mentioned TF-A hack and SError events do not have to be caught in TF-A EL3. Link: https://lore.kernel.org/r/20210608203655.31228-1-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com> Reviewed-by: Marek Behún <kabel@kernel.org> Cc: stable@vger.kernel.org # 7fbcb5da811b ("PCI: aardvark: Don't rely on jiffies while holding spinlock")
2021-06-08 20:36:55 +00:00
static bool advk_pcie_pio_is_running(struct advk_pcie *pcie)
{
struct device *dev = &pcie->pdev->dev;
/*
* Trying to start a new PIO transfer when previous has not completed
* cause External Abort on CPU which results in kernel panic:
*
* SError Interrupt on CPU0, code 0xbf000002 -- SError
* Kernel panic - not syncing: Asynchronous SError Interrupt
*
* Functions advk_pcie_rd_conf() and advk_pcie_wr_conf() are protected
* by raw_spin_lock_irqsave() at pci_lock_config() level to prevent
* concurrent calls at the same time. But because PIO transfer may take
* about 1.5s when link is down or card is disconnected, it means that
* advk_pcie_wait_pio() does not always have to wait for completion.
*
* Some versions of ARM Trusted Firmware handles this External Abort at
* EL3 level and mask it to prevent kernel panic. Relevant TF-A commit:
* https://git.trustedfirmware.org/TF-A/trusted-firmware-a.git/commit/?id=3c7dcdac5c50
*/
if (advk_readl(pcie, PIO_START)) {
dev_err(dev, "Previous PIO read/write transfer is still running\n");
return true;
}
return false;
}
static int advk_pcie_rd_conf(struct pci_bus *bus, u32 devfn,
int where, int size, u32 *val)
{
struct advk_pcie *pcie = bus->sysdata;
int retry_count;
bool allow_crs;
u32 reg;
int ret;
if (!advk_pcie_valid_device(pcie, bus, devfn))
return PCIBIOS_DEVICE_NOT_FOUND;
if (pci_is_root_bus(bus))
return pci_bridge_emul_conf_read(&pcie->bridge, where,
size, val);
/*
* Completion Retry Status is possible to return only when reading all
* 4 bytes from PCI_VENDOR_ID and PCI_DEVICE_ID registers at once and
* CRSSVE flag on Root Bridge is enabled.
*/
allow_crs = (where == PCI_VENDOR_ID) && (size == 4) &&
(le16_to_cpu(pcie->bridge.pcie_conf.rootctl) &
PCI_EXP_RTCTL_CRSSVE);
if (advk_pcie_pio_is_running(pcie))
goto try_crs;
/* Program the control register */
reg = advk_readl(pcie, PIO_CTRL);
reg &= ~PIO_CTRL_TYPE_MASK;
if (pci_is_root_bus(bus->parent))
reg |= PCIE_CONFIG_RD_TYPE0;
else
reg |= PCIE_CONFIG_RD_TYPE1;
advk_writel(pcie, reg, PIO_CTRL);
/* Program the address registers */
PCI: Unify ECAM constants in native PCI Express drivers Add ECAM-related constants to provide a set of standard constants defining memory address shift values to the byte-level address that can be used to access the PCI Express Configuration Space, and then move native PCI Express controller drivers to use the newly introduced definitions retiring driver-specific ones. Refactor pci_ecam_map_bus() function to use newly added constants so that limits to the bus, device function and offset (now limited to 4K as per the specification) are in place to prevent the defective or malicious caller from supplying incorrect configuration offset and thus targeting the wrong device when accessing extended configuration space. This refactor also allows for the ".bus_shift" initialisers to be dropped when the user is not using a custom value as a default value will be used as per the PCI Express Specification. Thanks to Qian Cai <qcai@redhat.com>, Michael Walle <michael@walle.cc>, and Vladimir Oltean <olteanv@gmail.com> for reporting a pci_ecam_create() issue with .bus_shift and to Vladimir for proposing the fix. [bhelgaas: incorporate Vladimir's fix, update commit log] Suggested-by: Bjorn Helgaas <bhelgaas@google.com> Link: https://lore.kernel.org/r/20201129230743.3006978-2-kw@linux.com Tested-by: Michael Walle <michael@walle.cc> Signed-off-by: Krzysztof Wilczyński <kw@linux.com> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com> Reviewed-by: Jon Derrick <jonathan.derrick@intel.com> Reviewed-by: Bjorn Helgaas <bhelgaas@google.com>
2020-11-29 23:07:39 +00:00
reg = ALIGN_DOWN(PCIE_ECAM_OFFSET(bus->number, devfn, where), 4);
advk_writel(pcie, reg, PIO_ADDR_LS);
advk_writel(pcie, 0, PIO_ADDR_MS);
/* Program the data strobe */
advk_writel(pcie, 0xf, PIO_WR_DATA_STRB);
retry_count = 0;
do {
/* Clear PIO DONE ISR and start the transfer */
advk_writel(pcie, 1, PIO_ISR);
advk_writel(pcie, 1, PIO_START);
ret = advk_pcie_wait_pio(pcie);
if (ret < 0)
goto try_crs;
retry_count += ret;
/* Check PIO status and get the read result */
ret = advk_pcie_check_pio_status(pcie, allow_crs, val);
} while (ret == -EAGAIN && retry_count < PIO_RETRY_CNT);
if (ret < 0)
goto fail;
if (size == 1)
*val = (*val >> (8 * (where & 3))) & 0xff;
else if (size == 2)
*val = (*val >> (8 * (where & 3))) & 0xffff;
return PCIBIOS_SUCCESSFUL;
try_crs:
/*
* If it is possible, return Completion Retry Status so that caller
* tries to issue the request again instead of failing.
*/
if (allow_crs) {
*val = CFG_RD_CRS_VAL;
return PCIBIOS_SUCCESSFUL;
}
fail:
*val = 0xffffffff;
return PCIBIOS_SET_FAILED;
}
static int advk_pcie_wr_conf(struct pci_bus *bus, u32 devfn,
int where, int size, u32 val)
{
struct advk_pcie *pcie = bus->sysdata;
u32 reg;
u32 data_strobe = 0x0;
int retry_count;
int offset;
int ret;
if (!advk_pcie_valid_device(pcie, bus, devfn))
return PCIBIOS_DEVICE_NOT_FOUND;
if (pci_is_root_bus(bus))
return pci_bridge_emul_conf_write(&pcie->bridge, where,
size, val);
if (where % size)
return PCIBIOS_SET_FAILED;
PCI: aardvark: Fix kernel panic during PIO transfer Trying to start a new PIO transfer by writing value 0 in PIO_START register when previous transfer has not yet completed (which is indicated by value 1 in PIO_START) causes an External Abort on CPU, which results in kernel panic: SError Interrupt on CPU0, code 0xbf000002 -- SError Kernel panic - not syncing: Asynchronous SError Interrupt To prevent kernel panic, it is required to reject a new PIO transfer when previous one has not finished yet. If previous PIO transfer is not finished yet, the kernel may issue a new PIO request only if the previous PIO transfer timed out. In the past the root cause of this issue was incorrectly identified (as it often happens during link retraining or after link down event) and special hack was implemented in Trusted Firmware to catch all SError events in EL3, to ignore errors with code 0xbf000002 and not forwarding any other errors to kernel and instead throw panic from EL3 Trusted Firmware handler. Links to discussion and patches about this issue: https://git.trustedfirmware.org/TF-A/trusted-firmware-a.git/commit/?id=3c7dcdac5c50 https://lore.kernel.org/linux-pci/20190316161243.29517-1-repk@triplefau.lt/ https://lore.kernel.org/linux-pci/971be151d24312cc533989a64bd454b4@www.loen.fr/ https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/1541 But the real cause was the fact that during link retraining or after link down event the PIO transfer may take longer time, up to the 1.44s until it times out. This increased probability that a new PIO transfer would be issued by kernel while previous one has not finished yet. After applying this change into the kernel, it is possible to revert the mentioned TF-A hack and SError events do not have to be caught in TF-A EL3. Link: https://lore.kernel.org/r/20210608203655.31228-1-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com> Reviewed-by: Marek Behún <kabel@kernel.org> Cc: stable@vger.kernel.org # 7fbcb5da811b ("PCI: aardvark: Don't rely on jiffies while holding spinlock")
2021-06-08 20:36:55 +00:00
if (advk_pcie_pio_is_running(pcie))
return PCIBIOS_SET_FAILED;
/* Program the control register */
reg = advk_readl(pcie, PIO_CTRL);
reg &= ~PIO_CTRL_TYPE_MASK;
if (pci_is_root_bus(bus->parent))
reg |= PCIE_CONFIG_WR_TYPE0;
else
reg |= PCIE_CONFIG_WR_TYPE1;
advk_writel(pcie, reg, PIO_CTRL);
/* Program the address registers */
PCI: Unify ECAM constants in native PCI Express drivers Add ECAM-related constants to provide a set of standard constants defining memory address shift values to the byte-level address that can be used to access the PCI Express Configuration Space, and then move native PCI Express controller drivers to use the newly introduced definitions retiring driver-specific ones. Refactor pci_ecam_map_bus() function to use newly added constants so that limits to the bus, device function and offset (now limited to 4K as per the specification) are in place to prevent the defective or malicious caller from supplying incorrect configuration offset and thus targeting the wrong device when accessing extended configuration space. This refactor also allows for the ".bus_shift" initialisers to be dropped when the user is not using a custom value as a default value will be used as per the PCI Express Specification. Thanks to Qian Cai <qcai@redhat.com>, Michael Walle <michael@walle.cc>, and Vladimir Oltean <olteanv@gmail.com> for reporting a pci_ecam_create() issue with .bus_shift and to Vladimir for proposing the fix. [bhelgaas: incorporate Vladimir's fix, update commit log] Suggested-by: Bjorn Helgaas <bhelgaas@google.com> Link: https://lore.kernel.org/r/20201129230743.3006978-2-kw@linux.com Tested-by: Michael Walle <michael@walle.cc> Signed-off-by: Krzysztof Wilczyński <kw@linux.com> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com> Reviewed-by: Jon Derrick <jonathan.derrick@intel.com> Reviewed-by: Bjorn Helgaas <bhelgaas@google.com>
2020-11-29 23:07:39 +00:00
reg = ALIGN_DOWN(PCIE_ECAM_OFFSET(bus->number, devfn, where), 4);
advk_writel(pcie, reg, PIO_ADDR_LS);
advk_writel(pcie, 0, PIO_ADDR_MS);
/* Calculate the write strobe */
offset = where & 0x3;
reg = val << (8 * offset);
data_strobe = GENMASK(size - 1, 0) << offset;
/* Program the data register */
advk_writel(pcie, reg, PIO_WR_DATA);
/* Program the data strobe */
advk_writel(pcie, data_strobe, PIO_WR_DATA_STRB);
retry_count = 0;
do {
/* Clear PIO DONE ISR and start the transfer */
advk_writel(pcie, 1, PIO_ISR);
advk_writel(pcie, 1, PIO_START);
ret = advk_pcie_wait_pio(pcie);
if (ret < 0)
return PCIBIOS_SET_FAILED;
retry_count += ret;
ret = advk_pcie_check_pio_status(pcie, false, NULL);
} while (ret == -EAGAIN && retry_count < PIO_RETRY_CNT);
return ret < 0 ? PCIBIOS_SET_FAILED : PCIBIOS_SUCCESSFUL;
}
static struct pci_ops advk_pcie_ops = {
.read = advk_pcie_rd_conf,
.write = advk_pcie_wr_conf,
};
static void advk_msi_irq_compose_msi_msg(struct irq_data *data,
struct msi_msg *msg)
{
struct advk_pcie *pcie = irq_data_get_irq_chip_data(data);
phys_addr_t msi_addr = virt_to_phys(pcie);
msg->address_lo = lower_32_bits(msi_addr);
msg->address_hi = upper_32_bits(msi_addr);
msg->data = data->hwirq;
}
static int advk_msi_set_affinity(struct irq_data *irq_data,
const struct cpumask *mask, bool force)
{
return -EINVAL;
}
static void advk_msi_irq_mask(struct irq_data *d)
{
struct advk_pcie *pcie = d->domain->host_data;
irq_hw_number_t hwirq = irqd_to_hwirq(d);
unsigned long flags;
u32 mask;
raw_spin_lock_irqsave(&pcie->msi_irq_lock, flags);
mask = advk_readl(pcie, PCIE_MSI_MASK_REG);
mask |= BIT(hwirq);
advk_writel(pcie, mask, PCIE_MSI_MASK_REG);
raw_spin_unlock_irqrestore(&pcie->msi_irq_lock, flags);
}
static void advk_msi_irq_unmask(struct irq_data *d)
{
struct advk_pcie *pcie = d->domain->host_data;
irq_hw_number_t hwirq = irqd_to_hwirq(d);
unsigned long flags;
u32 mask;
raw_spin_lock_irqsave(&pcie->msi_irq_lock, flags);
mask = advk_readl(pcie, PCIE_MSI_MASK_REG);
mask &= ~BIT(hwirq);
advk_writel(pcie, mask, PCIE_MSI_MASK_REG);
raw_spin_unlock_irqrestore(&pcie->msi_irq_lock, flags);
}
static void advk_msi_top_irq_mask(struct irq_data *d)
{
pci_msi_mask_irq(d);
irq_chip_mask_parent(d);
}
static void advk_msi_top_irq_unmask(struct irq_data *d)
{
pci_msi_unmask_irq(d);
irq_chip_unmask_parent(d);
}
static struct irq_chip advk_msi_bottom_irq_chip = {
.name = "MSI",
.irq_compose_msi_msg = advk_msi_irq_compose_msi_msg,
.irq_set_affinity = advk_msi_set_affinity,
.irq_mask = advk_msi_irq_mask,
.irq_unmask = advk_msi_irq_unmask,
};
static int advk_msi_irq_domain_alloc(struct irq_domain *domain,
unsigned int virq,
unsigned int nr_irqs, void *args)
{
struct advk_pcie *pcie = domain->host_data;
int hwirq, i;
mutex_lock(&pcie->msi_used_lock);
hwirq = bitmap_find_free_region(pcie->msi_used, MSI_IRQ_NUM,
order_base_2(nr_irqs));
mutex_unlock(&pcie->msi_used_lock);
if (hwirq < 0)
return -ENOSPC;
for (i = 0; i < nr_irqs; i++)
irq_domain_set_info(domain, virq + i, hwirq + i,
&advk_msi_bottom_irq_chip,
domain->host_data, handle_simple_irq,
NULL, NULL);
return 0;
}
static void advk_msi_irq_domain_free(struct irq_domain *domain,
unsigned int virq, unsigned int nr_irqs)
{
struct irq_data *d = irq_domain_get_irq_data(domain, virq);
struct advk_pcie *pcie = domain->host_data;
mutex_lock(&pcie->msi_used_lock);
bitmap_release_region(pcie->msi_used, d->hwirq, order_base_2(nr_irqs));
mutex_unlock(&pcie->msi_used_lock);
}
static const struct irq_domain_ops advk_msi_domain_ops = {
.alloc = advk_msi_irq_domain_alloc,
.free = advk_msi_irq_domain_free,
};
static void advk_pcie_irq_mask(struct irq_data *d)
{
struct advk_pcie *pcie = d->domain->host_data;
irq_hw_number_t hwirq = irqd_to_hwirq(d);
unsigned long flags;
u32 mask;
raw_spin_lock_irqsave(&pcie->irq_lock, flags);
mask = advk_readl(pcie, PCIE_ISR1_MASK_REG);
mask |= PCIE_ISR1_INTX_ASSERT(hwirq);
advk_writel(pcie, mask, PCIE_ISR1_MASK_REG);
raw_spin_unlock_irqrestore(&pcie->irq_lock, flags);
}
static void advk_pcie_irq_unmask(struct irq_data *d)
{
struct advk_pcie *pcie = d->domain->host_data;
irq_hw_number_t hwirq = irqd_to_hwirq(d);
unsigned long flags;
u32 mask;
raw_spin_lock_irqsave(&pcie->irq_lock, flags);
mask = advk_readl(pcie, PCIE_ISR1_MASK_REG);
mask &= ~PCIE_ISR1_INTX_ASSERT(hwirq);
advk_writel(pcie, mask, PCIE_ISR1_MASK_REG);
raw_spin_unlock_irqrestore(&pcie->irq_lock, flags);
}
static int advk_pcie_irq_map(struct irq_domain *h,
unsigned int virq, irq_hw_number_t hwirq)
{
struct advk_pcie *pcie = h->host_data;
irq_set_status_flags(virq, IRQ_LEVEL);
irq_set_chip_and_handler(virq, &pcie->irq_chip,
handle_level_irq);
irq_set_chip_data(virq, pcie);
return 0;
}
static const struct irq_domain_ops advk_pcie_irq_domain_ops = {
.map = advk_pcie_irq_map,
.xlate = irq_domain_xlate_onecell,
};
static struct irq_chip advk_msi_irq_chip = {
.name = "advk-MSI",
.irq_mask = advk_msi_top_irq_mask,
.irq_unmask = advk_msi_top_irq_unmask,
};
static struct msi_domain_info advk_msi_domain_info = {
.flags = MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
MSI_FLAG_MULTI_PCI_MSI | MSI_FLAG_PCI_MSIX,
.chip = &advk_msi_irq_chip,
};
static int advk_pcie_init_msi_irq_domain(struct advk_pcie *pcie)
{
struct device *dev = &pcie->pdev->dev;
raw_spin_lock_init(&pcie->msi_irq_lock);
mutex_init(&pcie->msi_used_lock);
pcie->msi_inner_domain =
irq_domain_add_linear(NULL, MSI_IRQ_NUM,
&advk_msi_domain_ops, pcie);
if (!pcie->msi_inner_domain)
return -ENOMEM;
pcie->msi_domain =
pci_msi_create_irq_domain(dev_fwnode(dev),
&advk_msi_domain_info,
pcie->msi_inner_domain);
if (!pcie->msi_domain) {
irq_domain_remove(pcie->msi_inner_domain);
return -ENOMEM;
}
return 0;
}
static void advk_pcie_remove_msi_irq_domain(struct advk_pcie *pcie)
{
irq_domain_remove(pcie->msi_domain);
irq_domain_remove(pcie->msi_inner_domain);
}
static int advk_pcie_init_irq_domain(struct advk_pcie *pcie)
{
struct device *dev = &pcie->pdev->dev;
struct device_node *node = dev->of_node;
struct device_node *pcie_intc_node;
struct irq_chip *irq_chip;
int ret = 0;
raw_spin_lock_init(&pcie->irq_lock);
pcie_intc_node = of_get_next_child(node, NULL);
if (!pcie_intc_node) {
dev_err(dev, "No PCIe Intc node found\n");
return -ENODEV;
}
irq_chip = &pcie->irq_chip;
irq_chip->name = devm_kasprintf(dev, GFP_KERNEL, "%s-irq",
dev_name(dev));
if (!irq_chip->name) {
ret = -ENOMEM;
goto out_put_node;
}
irq_chip->irq_mask = advk_pcie_irq_mask;
irq_chip->irq_unmask = advk_pcie_irq_unmask;
pcie->irq_domain =
irq_domain_add_linear(pcie_intc_node, PCI_NUM_INTX,
&advk_pcie_irq_domain_ops, pcie);
if (!pcie->irq_domain) {
dev_err(dev, "Failed to get a INTx IRQ domain\n");
ret = -ENOMEM;
goto out_put_node;
}
out_put_node:
of_node_put(pcie_intc_node);
return ret;
}
static void advk_pcie_remove_irq_domain(struct advk_pcie *pcie)
{
irq_domain_remove(pcie->irq_domain);
}
static struct irq_chip advk_rp_irq_chip = {
.name = "advk-RP",
};
static int advk_pcie_rp_irq_map(struct irq_domain *h,
unsigned int virq, irq_hw_number_t hwirq)
{
struct advk_pcie *pcie = h->host_data;
irq_set_chip_and_handler(virq, &advk_rp_irq_chip, handle_simple_irq);
irq_set_chip_data(virq, pcie);
return 0;
}
static const struct irq_domain_ops advk_pcie_rp_irq_domain_ops = {
.map = advk_pcie_rp_irq_map,
.xlate = irq_domain_xlate_onecell,
};
static int advk_pcie_init_rp_irq_domain(struct advk_pcie *pcie)
{
pcie->rp_irq_domain = irq_domain_add_linear(NULL, 1,
&advk_pcie_rp_irq_domain_ops,
pcie);
if (!pcie->rp_irq_domain) {
dev_err(&pcie->pdev->dev, "Failed to add Root Port IRQ domain\n");
return -ENOMEM;
}
return 0;
}
static void advk_pcie_remove_rp_irq_domain(struct advk_pcie *pcie)
{
irq_domain_remove(pcie->rp_irq_domain);
}
PCI: aardvark: Fix support for PME requester on emulated bridge Enable aardvark PME interrupt unconditionally by unmasking it and read PME requester ID to emulated bridge config space immediately after receiving interrupt. PME requester ID is stored in the PCIE_MSG_LOG_REG register, which contains the last inbound message. So when new inbound message is received by HW (including non-PM), the content in PCIE_MSG_LOG_REG register is replaced by a new value. PCIe specification mandates that subsequent PMEs are kept pending until the PME Status Register bit is cleared by software by writing a 1b. Support for masking/unmasking PME interrupt on emulated bridge via PCI_EXP_RTCTL_PMEIE bit is now implemented only in emulated bridge config space, to ensure that we do not miss any aardvark PME interrupt. Reading of PCI_EXP_RTCAP and PCI_EXP_RTSTA registers is simplified as final value is now always stored into emulated bridge config space by the interrupt handler, so there is no need to implement support for these registers in read_pcie callback. Clearing of W1C bit PCI_EXP_RTSTA_PME is now also simplified as it is done by pci-bridge-emul.c code for emulated bridge config space. So there is no need to implement support for clearing this bit in write_pcie callback. Link: https://lore.kernel.org/r/20220110015018.26359-18-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
2022-01-10 01:50:12 +00:00
static void advk_pcie_handle_pme(struct advk_pcie *pcie)
{
u32 requester = advk_readl(pcie, PCIE_MSG_LOG_REG) >> 16;
advk_writel(pcie, PCIE_MSG_PM_PME_MASK, PCIE_ISR0_REG);
/*
* PCIE_MSG_LOG_REG contains the last inbound message, so store
* the requester ID only when PME was not asserted yet.
* Also do not trigger PME interrupt when PME is still asserted.
*/
if (!(le32_to_cpu(pcie->bridge.pcie_conf.rootsta) & PCI_EXP_RTSTA_PME)) {
pcie->bridge.pcie_conf.rootsta = cpu_to_le32(requester | PCI_EXP_RTSTA_PME);
/*
* Trigger PME interrupt only if PMEIE bit in Root Control is set.
* Aardvark HW returns zero for PCI_EXP_FLAGS_IRQ, so use PCIe interrupt 0.
*/
if (!(le16_to_cpu(pcie->bridge.pcie_conf.rootctl) & PCI_EXP_RTCTL_PMEIE))
return;
if (generic_handle_domain_irq(pcie->rp_irq_domain, 0) == -EINVAL)
PCI: aardvark: Fix support for PME requester on emulated bridge Enable aardvark PME interrupt unconditionally by unmasking it and read PME requester ID to emulated bridge config space immediately after receiving interrupt. PME requester ID is stored in the PCIE_MSG_LOG_REG register, which contains the last inbound message. So when new inbound message is received by HW (including non-PM), the content in PCIE_MSG_LOG_REG register is replaced by a new value. PCIe specification mandates that subsequent PMEs are kept pending until the PME Status Register bit is cleared by software by writing a 1b. Support for masking/unmasking PME interrupt on emulated bridge via PCI_EXP_RTCTL_PMEIE bit is now implemented only in emulated bridge config space, to ensure that we do not miss any aardvark PME interrupt. Reading of PCI_EXP_RTCAP and PCI_EXP_RTSTA registers is simplified as final value is now always stored into emulated bridge config space by the interrupt handler, so there is no need to implement support for these registers in read_pcie callback. Clearing of W1C bit PCI_EXP_RTSTA_PME is now also simplified as it is done by pci-bridge-emul.c code for emulated bridge config space. So there is no need to implement support for clearing this bit in write_pcie callback. Link: https://lore.kernel.org/r/20220110015018.26359-18-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
2022-01-10 01:50:12 +00:00
dev_err_ratelimited(&pcie->pdev->dev, "unhandled PME IRQ\n");
}
}
static void advk_pcie_handle_msi(struct advk_pcie *pcie)
{
u32 msi_val, msi_mask, msi_status, msi_idx;
msi_mask = advk_readl(pcie, PCIE_MSI_MASK_REG);
msi_val = advk_readl(pcie, PCIE_MSI_STATUS_REG);
msi_status = msi_val & ((~msi_mask) & PCIE_MSI_ALL_MASK);
for (msi_idx = 0; msi_idx < MSI_IRQ_NUM; msi_idx++) {
if (!(BIT(msi_idx) & msi_status))
continue;
advk_writel(pcie, BIT(msi_idx), PCIE_MSI_STATUS_REG);
if (generic_handle_domain_irq(pcie->msi_inner_domain, msi_idx) == -EINVAL)
dev_err_ratelimited(&pcie->pdev->dev, "unexpected MSI 0x%02x\n", msi_idx);
}
advk_writel(pcie, PCIE_ISR0_MSI_INT_PENDING,
PCIE_ISR0_REG);
}
static void advk_pcie_handle_int(struct advk_pcie *pcie)
{
u32 isr0_val, isr0_mask, isr0_status;
u32 isr1_val, isr1_mask, isr1_status;
int i;
isr0_val = advk_readl(pcie, PCIE_ISR0_REG);
isr0_mask = advk_readl(pcie, PCIE_ISR0_MASK_REG);
isr0_status = isr0_val & ((~isr0_mask) & PCIE_ISR0_ALL_MASK);
isr1_val = advk_readl(pcie, PCIE_ISR1_REG);
isr1_mask = advk_readl(pcie, PCIE_ISR1_MASK_REG);
isr1_status = isr1_val & ((~isr1_mask) & PCIE_ISR1_ALL_MASK);
PCI: aardvark: Fix support for PME requester on emulated bridge Enable aardvark PME interrupt unconditionally by unmasking it and read PME requester ID to emulated bridge config space immediately after receiving interrupt. PME requester ID is stored in the PCIE_MSG_LOG_REG register, which contains the last inbound message. So when new inbound message is received by HW (including non-PM), the content in PCIE_MSG_LOG_REG register is replaced by a new value. PCIe specification mandates that subsequent PMEs are kept pending until the PME Status Register bit is cleared by software by writing a 1b. Support for masking/unmasking PME interrupt on emulated bridge via PCI_EXP_RTCTL_PMEIE bit is now implemented only in emulated bridge config space, to ensure that we do not miss any aardvark PME interrupt. Reading of PCI_EXP_RTCAP and PCI_EXP_RTSTA registers is simplified as final value is now always stored into emulated bridge config space by the interrupt handler, so there is no need to implement support for these registers in read_pcie callback. Clearing of W1C bit PCI_EXP_RTSTA_PME is now also simplified as it is done by pci-bridge-emul.c code for emulated bridge config space. So there is no need to implement support for clearing this bit in write_pcie callback. Link: https://lore.kernel.org/r/20220110015018.26359-18-kabel@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Marek Behún <kabel@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
2022-01-10 01:50:12 +00:00
/* Process PME interrupt as the first one to do not miss PME requester id */
if (isr0_status & PCIE_MSG_PM_PME_MASK)
advk_pcie_handle_pme(pcie);
/* Process ERR interrupt */
if (isr0_status & PCIE_ISR0_ERR_MASK) {
advk_writel(pcie, PCIE_ISR0_ERR_MASK, PCIE_ISR0_REG);
/*
* Aardvark HW returns zero for PCI_ERR_ROOT_AER_IRQ, so use
* PCIe interrupt 0
*/
if (generic_handle_domain_irq(pcie->rp_irq_domain, 0) == -EINVAL)
dev_err_ratelimited(&pcie->pdev->dev, "unhandled ERR IRQ\n");
}
/* Process MSI interrupts */
if (isr0_status & PCIE_ISR0_MSI_INT_PENDING)
advk_pcie_handle_msi(pcie);
/* Process legacy interrupts */
for (i = 0; i < PCI_NUM_INTX; i++) {
if (!(isr1_status & PCIE_ISR1_INTX_ASSERT(i)))
continue;
advk_writel(pcie, PCIE_ISR1_INTX_ASSERT(i),
PCIE_ISR1_REG);
if (generic_handle_domain_irq(pcie->irq_domain, i) == -EINVAL)
dev_err_ratelimited(&pcie->pdev->dev, "unexpected INT%c IRQ\n",
(char)i + 'A');
}
}
static irqreturn_t advk_pcie_irq_handler(int irq, void *arg)
{
struct advk_pcie *pcie = arg;
u32 status;
status = advk_readl(pcie, HOST_CTRL_INT_STATUS_REG);
if (!(status & PCIE_IRQ_CORE_INT))
return IRQ_NONE;
advk_pcie_handle_int(pcie);
/* Clear interrupt */
advk_writel(pcie, PCIE_IRQ_CORE_INT, HOST_CTRL_INT_STATUS_REG);
return IRQ_HANDLED;
}
static int advk_pcie_map_irq(const struct pci_dev *dev, u8 slot, u8 pin)
{
struct advk_pcie *pcie = dev->bus->sysdata;
/*
* Emulated root bridge has its own emulated irq chip and irq domain.
* Argument pin is the INTx pin (1=INTA, 2=INTB, 3=INTC, 4=INTD) and
* hwirq for irq_create_mapping() is indexed from zero.
*/
if (pci_is_root_bus(dev->bus))
return irq_create_mapping(pcie->rp_irq_domain, pin - 1);
else
return of_irq_parse_and_map_pci(dev, slot, pin);
}
static void advk_pcie_disable_phy(struct advk_pcie *pcie)
{
phy_power_off(pcie->phy);
phy_exit(pcie->phy);
}
static int advk_pcie_enable_phy(struct advk_pcie *pcie)
{
int ret;
if (!pcie->phy)
return 0;
ret = phy_init(pcie->phy);
if (ret)
return ret;
ret = phy_set_mode(pcie->phy, PHY_MODE_PCIE);
if (ret) {
phy_exit(pcie->phy);
return ret;
}
ret = phy_power_on(pcie->phy);
if (ret) {
phy_exit(pcie->phy);
return ret;
}
return 0;
}
static int advk_pcie_setup_phy(struct advk_pcie *pcie)
{
struct device *dev = &pcie->pdev->dev;
struct device_node *node = dev->of_node;
int ret = 0;
pcie->phy = devm_of_phy_get(dev, node, NULL);
if (IS_ERR(pcie->phy) && (PTR_ERR(pcie->phy) == -EPROBE_DEFER))
return PTR_ERR(pcie->phy);
/* Old bindings miss the PHY handle */
if (IS_ERR(pcie->phy)) {
dev_warn(dev, "PHY unavailable (%ld)\n", PTR_ERR(pcie->phy));
pcie->phy = NULL;
return 0;
}
ret = advk_pcie_enable_phy(pcie);
if (ret)
dev_err(dev, "Failed to initialize PHY (%d)\n", ret);
return ret;
}
static int advk_pcie_probe(struct platform_device *pdev)
{
struct device *dev = &pdev->dev;
struct advk_pcie *pcie;
struct pci_host_bridge *bridge;
PCI: aardvark: Configure PCIe resources from 'ranges' DT property In commit 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration") was removed aardvark PCIe outbound window configuration and commit description said that was recommended solution by HW designers. But that commit completely removed support for configuring PCIe IO resources without removing PCIe IO 'ranges' from DTS files. After that commit PCIe IO space started to be treated as PCIe MEM space and accessing it just caused kernel crash. Moreover implementation of PCIe outbound windows prior that commit was incorrect. It completely ignored offset between CPU address and PCIe bus address and expected that in DTS is CPU address always same as PCIe bus address without doing any checks. Also it completely ignored size of every PCIe resource specified in 'ranges' DTS property and expected that every PCIe resource has size 128 MB (also for PCIe IO range). Again without any check. Apparently none of PCIe resource has in DTS specified size of 128 MB. So it was completely broken and thanks to how aardvark mask works, configuration was completely ignored. This patch reverts back support for PCIe outbound window configuration but implementation is a new without issues mentioned above. PCIe outbound window is required when DTS specify in 'ranges' property non-zero offset between CPU and PCIe address space. To address recommendation by HW designers as specified in commit description of 6df6ba974a55, set default outbound parameters as PCIe MEM access without translation and therefore for this PCIe 'ranges' it is not needed to configure PCIe outbound window. For PCIe IO space is needed to configure aardvark PCIe outbound window. This patch fixes kernel crash when trying to access PCIe IO space. Link: https://lore.kernel.org/r/20210624215546.4015-2-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Cc: stable@vger.kernel.org # 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration")
2021-06-24 21:55:45 +00:00
struct resource_entry *entry;
int ret, irq;
bridge = devm_pci_alloc_host_bridge(dev, sizeof(struct advk_pcie));
if (!bridge)
return -ENOMEM;
pcie = pci_host_bridge_priv(bridge);
pcie->pdev = pdev;
platform_set_drvdata(pdev, pcie);
PCI: aardvark: Configure PCIe resources from 'ranges' DT property In commit 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration") was removed aardvark PCIe outbound window configuration and commit description said that was recommended solution by HW designers. But that commit completely removed support for configuring PCIe IO resources without removing PCIe IO 'ranges' from DTS files. After that commit PCIe IO space started to be treated as PCIe MEM space and accessing it just caused kernel crash. Moreover implementation of PCIe outbound windows prior that commit was incorrect. It completely ignored offset between CPU address and PCIe bus address and expected that in DTS is CPU address always same as PCIe bus address without doing any checks. Also it completely ignored size of every PCIe resource specified in 'ranges' DTS property and expected that every PCIe resource has size 128 MB (also for PCIe IO range). Again without any check. Apparently none of PCIe resource has in DTS specified size of 128 MB. So it was completely broken and thanks to how aardvark mask works, configuration was completely ignored. This patch reverts back support for PCIe outbound window configuration but implementation is a new without issues mentioned above. PCIe outbound window is required when DTS specify in 'ranges' property non-zero offset between CPU and PCIe address space. To address recommendation by HW designers as specified in commit description of 6df6ba974a55, set default outbound parameters as PCIe MEM access without translation and therefore for this PCIe 'ranges' it is not needed to configure PCIe outbound window. For PCIe IO space is needed to configure aardvark PCIe outbound window. This patch fixes kernel crash when trying to access PCIe IO space. Link: https://lore.kernel.org/r/20210624215546.4015-2-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Cc: stable@vger.kernel.org # 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration")
2021-06-24 21:55:45 +00:00
resource_list_for_each_entry(entry, &bridge->windows) {
resource_size_t start = entry->res->start;
resource_size_t size = resource_size(entry->res);
unsigned long type = resource_type(entry->res);
u64 win_size;
/*
* Aardvark hardware allows to configure also PCIe window
* for config type 0 and type 1 mapping, but driver uses
* only PIO for issuing configuration transfers which does
* not use PCIe window configuration.
*/
if (type != IORESOURCE_MEM && type != IORESOURCE_IO)
PCI: aardvark: Configure PCIe resources from 'ranges' DT property In commit 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration") was removed aardvark PCIe outbound window configuration and commit description said that was recommended solution by HW designers. But that commit completely removed support for configuring PCIe IO resources without removing PCIe IO 'ranges' from DTS files. After that commit PCIe IO space started to be treated as PCIe MEM space and accessing it just caused kernel crash. Moreover implementation of PCIe outbound windows prior that commit was incorrect. It completely ignored offset between CPU address and PCIe bus address and expected that in DTS is CPU address always same as PCIe bus address without doing any checks. Also it completely ignored size of every PCIe resource specified in 'ranges' DTS property and expected that every PCIe resource has size 128 MB (also for PCIe IO range). Again without any check. Apparently none of PCIe resource has in DTS specified size of 128 MB. So it was completely broken and thanks to how aardvark mask works, configuration was completely ignored. This patch reverts back support for PCIe outbound window configuration but implementation is a new without issues mentioned above. PCIe outbound window is required when DTS specify in 'ranges' property non-zero offset between CPU and PCIe address space. To address recommendation by HW designers as specified in commit description of 6df6ba974a55, set default outbound parameters as PCIe MEM access without translation and therefore for this PCIe 'ranges' it is not needed to configure PCIe outbound window. For PCIe IO space is needed to configure aardvark PCIe outbound window. This patch fixes kernel crash when trying to access PCIe IO space. Link: https://lore.kernel.org/r/20210624215546.4015-2-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Cc: stable@vger.kernel.org # 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration")
2021-06-24 21:55:45 +00:00
continue;
/*
* Skip transparent memory resources. Default outbound access
* configuration is set to transparent memory access so it
* does not need window configuration.
*/
if (type == IORESOURCE_MEM && entry->offset == 0)
PCI: aardvark: Configure PCIe resources from 'ranges' DT property In commit 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration") was removed aardvark PCIe outbound window configuration and commit description said that was recommended solution by HW designers. But that commit completely removed support for configuring PCIe IO resources without removing PCIe IO 'ranges' from DTS files. After that commit PCIe IO space started to be treated as PCIe MEM space and accessing it just caused kernel crash. Moreover implementation of PCIe outbound windows prior that commit was incorrect. It completely ignored offset between CPU address and PCIe bus address and expected that in DTS is CPU address always same as PCIe bus address without doing any checks. Also it completely ignored size of every PCIe resource specified in 'ranges' DTS property and expected that every PCIe resource has size 128 MB (also for PCIe IO range). Again without any check. Apparently none of PCIe resource has in DTS specified size of 128 MB. So it was completely broken and thanks to how aardvark mask works, configuration was completely ignored. This patch reverts back support for PCIe outbound window configuration but implementation is a new without issues mentioned above. PCIe outbound window is required when DTS specify in 'ranges' property non-zero offset between CPU and PCIe address space. To address recommendation by HW designers as specified in commit description of 6df6ba974a55, set default outbound parameters as PCIe MEM access without translation and therefore for this PCIe 'ranges' it is not needed to configure PCIe outbound window. For PCIe IO space is needed to configure aardvark PCIe outbound window. This patch fixes kernel crash when trying to access PCIe IO space. Link: https://lore.kernel.org/r/20210624215546.4015-2-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Cc: stable@vger.kernel.org # 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration")
2021-06-24 21:55:45 +00:00
continue;
/*
* The n-th PCIe window is configured by tuple (match, remap, mask)
* and an access to address A uses this window if A matches the
* match with given mask.
* So every PCIe window size must be a power of two and every start
* address must be aligned to window size. Minimal size is 64 KiB
* because lower 16 bits of mask must be zero. Remapped address
* may have set only bits from the mask.
*/
while (pcie->wins_count < OB_WIN_COUNT && size > 0) {
/* Calculate the largest aligned window size */
win_size = (1ULL << (fls64(size)-1)) |
(start ? (1ULL << __ffs64(start)) : 0);
win_size = 1ULL << __ffs64(win_size);
if (win_size < 0x10000)
break;
dev_dbg(dev,
"Configuring PCIe window %d: [0x%llx-0x%llx] as %lu\n",
pcie->wins_count, (unsigned long long)start,
(unsigned long long)start + win_size, type);
if (type == IORESOURCE_IO) {
pcie->wins[pcie->wins_count].actions = OB_WIN_TYPE_IO;
pcie->wins[pcie->wins_count].match = pci_pio_to_address(start);
} else {
pcie->wins[pcie->wins_count].actions = OB_WIN_TYPE_MEM;
pcie->wins[pcie->wins_count].match = start;
}
pcie->wins[pcie->wins_count].remap = start - entry->offset;
pcie->wins[pcie->wins_count].mask = ~(win_size - 1);
if (pcie->wins[pcie->wins_count].remap & (win_size - 1))
break;
start += win_size;
size -= win_size;
pcie->wins_count++;
}
if (size > 0) {
dev_err(&pcie->pdev->dev,
"Invalid PCIe region [0x%llx-0x%llx]\n",
(unsigned long long)entry->res->start,
(unsigned long long)entry->res->end + 1);
return -EINVAL;
}
}
pcie->base = devm_platform_ioremap_resource(pdev, 0);
if (IS_ERR(pcie->base))
return PTR_ERR(pcie->base);
irq = platform_get_irq(pdev, 0);
if (irq < 0)
return irq;
ret = devm_request_irq(dev, irq, advk_pcie_irq_handler,
IRQF_SHARED | IRQF_NO_THREAD, "advk-pcie",
pcie);
if (ret) {
dev_err(dev, "Failed to register interrupt\n");
return ret;
}
pcie->reset_gpio = devm_gpiod_get_optional(dev, "reset", GPIOD_OUT_LOW);
ret = PTR_ERR_OR_ZERO(pcie->reset_gpio);
if (ret) {
if (ret != -EPROBE_DEFER)
dev_err(dev, "Failed to get reset-gpio: %i\n", ret);
return ret;
}
ret = gpiod_set_consumer_name(pcie->reset_gpio, "pcie1-reset");
if (ret) {
dev_err(dev, "Failed to set reset gpio name: %d\n", ret);
return ret;
}
ret = of_pci_get_max_link_speed(dev->of_node);
if (ret <= 0 || ret > 3)
pcie->link_gen = 3;
else
pcie->link_gen = ret;
ret = advk_pcie_setup_phy(pcie);
if (ret)
return ret;
advk_pcie_setup_hw(pcie);
ret = advk_sw_pci_bridge_init(pcie);
if (ret) {
dev_err(dev, "Failed to register emulated root PCI bridge\n");
return ret;
}
ret = advk_pcie_init_irq_domain(pcie);
if (ret) {
dev_err(dev, "Failed to initialize irq\n");
return ret;
}
ret = advk_pcie_init_msi_irq_domain(pcie);
if (ret) {
dev_err(dev, "Failed to initialize irq\n");
advk_pcie_remove_irq_domain(pcie);
return ret;
}
ret = advk_pcie_init_rp_irq_domain(pcie);
if (ret) {
dev_err(dev, "Failed to initialize irq\n");
advk_pcie_remove_msi_irq_domain(pcie);
advk_pcie_remove_irq_domain(pcie);
return ret;
}
bridge->sysdata = pcie;
bridge->ops = &advk_pcie_ops;
bridge->map_irq = advk_pcie_map_irq;
ret = pci_host_probe(bridge);
if (ret < 0) {
advk_pcie_remove_rp_irq_domain(pcie);
advk_pcie_remove_msi_irq_domain(pcie);
advk_pcie_remove_irq_domain(pcie);
return ret;
}
return 0;
}
static int advk_pcie_remove(struct platform_device *pdev)
{
struct advk_pcie *pcie = platform_get_drvdata(pdev);
struct pci_host_bridge *bridge = pci_host_bridge_from_priv(pcie);
u32 val;
PCI: aardvark: Configure PCIe resources from 'ranges' DT property In commit 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration") was removed aardvark PCIe outbound window configuration and commit description said that was recommended solution by HW designers. But that commit completely removed support for configuring PCIe IO resources without removing PCIe IO 'ranges' from DTS files. After that commit PCIe IO space started to be treated as PCIe MEM space and accessing it just caused kernel crash. Moreover implementation of PCIe outbound windows prior that commit was incorrect. It completely ignored offset between CPU address and PCIe bus address and expected that in DTS is CPU address always same as PCIe bus address without doing any checks. Also it completely ignored size of every PCIe resource specified in 'ranges' DTS property and expected that every PCIe resource has size 128 MB (also for PCIe IO range). Again without any check. Apparently none of PCIe resource has in DTS specified size of 128 MB. So it was completely broken and thanks to how aardvark mask works, configuration was completely ignored. This patch reverts back support for PCIe outbound window configuration but implementation is a new without issues mentioned above. PCIe outbound window is required when DTS specify in 'ranges' property non-zero offset between CPU and PCIe address space. To address recommendation by HW designers as specified in commit description of 6df6ba974a55, set default outbound parameters as PCIe MEM access without translation and therefore for this PCIe 'ranges' it is not needed to configure PCIe outbound window. For PCIe IO space is needed to configure aardvark PCIe outbound window. This patch fixes kernel crash when trying to access PCIe IO space. Link: https://lore.kernel.org/r/20210624215546.4015-2-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Cc: stable@vger.kernel.org # 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration")
2021-06-24 21:55:45 +00:00
int i;
/* Remove PCI bus with all devices */
pci_lock_rescan_remove();
pci_stop_root_bus(bridge->bus);
pci_remove_root_bus(bridge->bus);
pci_unlock_rescan_remove();
/* Disable Root Bridge I/O space, memory space and bus mastering */
val = advk_readl(pcie, PCIE_CORE_CMD_STATUS_REG);
val &= ~(PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER);
advk_writel(pcie, val, PCIE_CORE_CMD_STATUS_REG);
/* Disable MSI */
val = advk_readl(pcie, PCIE_CORE_CTRL2_REG);
val &= ~PCIE_CORE_CTRL2_MSI_ENABLE;
advk_writel(pcie, val, PCIE_CORE_CTRL2_REG);
/* Clear MSI address */
advk_writel(pcie, 0, PCIE_MSI_ADDR_LOW_REG);
advk_writel(pcie, 0, PCIE_MSI_ADDR_HIGH_REG);
/* Mask all interrupts */
advk_writel(pcie, PCIE_MSI_ALL_MASK, PCIE_MSI_MASK_REG);
advk_writel(pcie, PCIE_ISR0_ALL_MASK, PCIE_ISR0_MASK_REG);
advk_writel(pcie, PCIE_ISR1_ALL_MASK, PCIE_ISR1_MASK_REG);
advk_writel(pcie, PCIE_IRQ_ALL_MASK, HOST_CTRL_INT_MASK_REG);
/* Clear all interrupts */
advk_writel(pcie, PCIE_MSI_ALL_MASK, PCIE_MSI_STATUS_REG);
advk_writel(pcie, PCIE_ISR0_ALL_MASK, PCIE_ISR0_REG);
advk_writel(pcie, PCIE_ISR1_ALL_MASK, PCIE_ISR1_REG);
advk_writel(pcie, PCIE_IRQ_ALL_MASK, HOST_CTRL_INT_STATUS_REG);
/* Remove IRQ domains */
advk_pcie_remove_rp_irq_domain(pcie);
advk_pcie_remove_msi_irq_domain(pcie);
advk_pcie_remove_irq_domain(pcie);
/* Free config space for emulated root bridge */
pci_bridge_emul_cleanup(&pcie->bridge);
/* Assert PERST# signal which prepares PCIe card for power down */
if (pcie->reset_gpio)
gpiod_set_value_cansleep(pcie->reset_gpio, 1);
/* Disable link training */
val = advk_readl(pcie, PCIE_CORE_CTRL0_REG);
val &= ~LINK_TRAINING_EN;
advk_writel(pcie, val, PCIE_CORE_CTRL0_REG);
PCI: aardvark: Configure PCIe resources from 'ranges' DT property In commit 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration") was removed aardvark PCIe outbound window configuration and commit description said that was recommended solution by HW designers. But that commit completely removed support for configuring PCIe IO resources without removing PCIe IO 'ranges' from DTS files. After that commit PCIe IO space started to be treated as PCIe MEM space and accessing it just caused kernel crash. Moreover implementation of PCIe outbound windows prior that commit was incorrect. It completely ignored offset between CPU address and PCIe bus address and expected that in DTS is CPU address always same as PCIe bus address without doing any checks. Also it completely ignored size of every PCIe resource specified in 'ranges' DTS property and expected that every PCIe resource has size 128 MB (also for PCIe IO range). Again without any check. Apparently none of PCIe resource has in DTS specified size of 128 MB. So it was completely broken and thanks to how aardvark mask works, configuration was completely ignored. This patch reverts back support for PCIe outbound window configuration but implementation is a new without issues mentioned above. PCIe outbound window is required when DTS specify in 'ranges' property non-zero offset between CPU and PCIe address space. To address recommendation by HW designers as specified in commit description of 6df6ba974a55, set default outbound parameters as PCIe MEM access without translation and therefore for this PCIe 'ranges' it is not needed to configure PCIe outbound window. For PCIe IO space is needed to configure aardvark PCIe outbound window. This patch fixes kernel crash when trying to access PCIe IO space. Link: https://lore.kernel.org/r/20210624215546.4015-2-pali@kernel.org Signed-off-by: Pali Rohár <pali@kernel.org> Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> Cc: stable@vger.kernel.org # 6df6ba974a55 ("PCI: aardvark: Remove PCIe outbound window configuration")
2021-06-24 21:55:45 +00:00
/* Disable outbound address windows mapping */
for (i = 0; i < OB_WIN_COUNT; i++)
advk_pcie_disable_ob_win(pcie, i);
/* Disable phy */
advk_pcie_disable_phy(pcie);
return 0;
}
static const struct of_device_id advk_pcie_of_match_table[] = {
{ .compatible = "marvell,armada-3700-pcie", },
{},
};
MODULE_DEVICE_TABLE(of, advk_pcie_of_match_table);
static struct platform_driver advk_pcie_driver = {
.driver = {
.name = "advk-pcie",
.of_match_table = advk_pcie_of_match_table,
},
.probe = advk_pcie_probe,
.remove = advk_pcie_remove,
};
module_platform_driver(advk_pcie_driver);
MODULE_DESCRIPTION("Aardvark PCIe controller");
MODULE_LICENSE("GPL v2");