From 4fe204977096e900cb91a3298b05c794ac24f540 Mon Sep 17 00:00:00 2001 From: Niklas Schnelle Date: Wed, 7 Jul 2021 10:42:43 +0200 Subject: [PATCH 1/7] s390/pci: refresh function handle in iomap The function handle of a PCI function is updated when disabling or enabling it as well as when the function's availability changes or it enters the error state. Until now this only occurred either while there is no struct pci_dev associated with the function yet or the function became unavailable. This meant that leaving a stale function handle in the iomap either didn't happen because there was no iomap yet or it lead to errors on PCI access but so would the correct disabled function handle. In the future a CLP Set PCI Function Disable/Enable cycle during PCI device recovery may be done while the device is bound to a driver. In this case we must update the iomap associated with the now-stale function handle to ensure that the resulting zPCI instruction references an accurate function handle. Since the function handle is accessed by the PCI accessor helpers without locking use READ_ONCE()/WRITE_ONCE() to mark this access and prevent compiler optimizations that would move the load/store. With that infrastructure in place let's also properly update the function handle in the existing cases. This makes sure that in the future debugging of a zPCI function access through the handle will show an up to date handle reducing the chance of confusion. Also it makes sure we have one single place where a zPCI function handle is updated after initialization. Reviewed-by: Pierre Morel Reviewed-by: Matthew Rosato Signed-off-by: Niklas Schnelle Signed-off-by: Vasily Gorbik --- arch/s390/include/asm/pci.h | 1 + arch/s390/pci/pci.c | 36 ++++++++++++++++++++++++++++++++---- arch/s390/pci/pci_event.c | 6 +++--- arch/s390/pci/pci_insn.c | 4 ++-- 4 files changed, 38 insertions(+), 9 deletions(-) diff --git a/arch/s390/include/asm/pci.h b/arch/s390/include/asm/pci.h index 6b3c366af78e..35adc0cd0e6a 100644 --- a/arch/s390/include/asm/pci.h +++ b/arch/s390/include/asm/pci.h @@ -213,6 +213,7 @@ bool zpci_is_device_configured(struct zpci_dev *zdev); int zpci_register_ioat(struct zpci_dev *, u8, u64, u64, u64); int zpci_unregister_ioat(struct zpci_dev *, u8); void zpci_remove_reserved_devices(void); +void zpci_update_fh(struct zpci_dev *zdev, u32 fh); /* CLP */ int clp_setup_writeback_mio(void); diff --git a/arch/s390/pci/pci.c b/arch/s390/pci/pci.c index 872d772b73d2..fe01ca3b33fa 100644 --- a/arch/s390/pci/pci.c +++ b/arch/s390/pci/pci.c @@ -481,6 +481,34 @@ static void zpci_free_iomap(struct zpci_dev *zdev, int entry) spin_unlock(&zpci_iomap_lock); } +static void zpci_do_update_iomap_fh(struct zpci_dev *zdev, u32 fh) +{ + int bar, idx; + + spin_lock(&zpci_iomap_lock); + for (bar = 0; bar < PCI_STD_NUM_BARS; bar++) { + if (!zdev->bars[bar].size) + continue; + idx = zdev->bars[bar].map_idx; + if (!zpci_iomap_start[idx].count) + continue; + WRITE_ONCE(zpci_iomap_start[idx].fh, zdev->fh); + } + spin_unlock(&zpci_iomap_lock); +} + +void zpci_update_fh(struct zpci_dev *zdev, u32 fh) +{ + if (!fh || zdev->fh == fh) + return; + + zdev->fh = fh; + if (zpci_use_mio(zdev)) + return; + if (zdev->has_resources && zdev_enabled(zdev)) + zpci_do_update_iomap_fh(zdev, fh); +} + static struct resource *__alloc_res(struct zpci_dev *zdev, unsigned long start, unsigned long size, unsigned long flags) { @@ -668,7 +696,7 @@ int zpci_enable_device(struct zpci_dev *zdev) if (clp_enable_fh(zdev, &fh, ZPCI_NR_DMA_SPACES)) rc = -EIO; else - zdev->fh = fh; + zpci_update_fh(zdev, fh); return rc; } @@ -679,14 +707,14 @@ int zpci_disable_device(struct zpci_dev *zdev) cc = clp_disable_fh(zdev, &fh); if (!cc) { - zdev->fh = fh; + zpci_update_fh(zdev, fh); } else if (cc == CLP_RC_SETPCIFN_ALRDY) { pr_info("Disabling PCI function %08x had no effect as it was already disabled\n", zdev->fid); /* Function is already disabled - update handle */ rc = clp_refresh_fh(zdev->fid, &fh); if (!rc) { - zdev->fh = fh; + zpci_update_fh(zdev, fh); rc = -EINVAL; } } else { @@ -776,7 +804,7 @@ int zpci_scan_configured_device(struct zpci_dev *zdev, u32 fh) { int rc; - zdev->fh = fh; + zpci_update_fh(zdev, fh); /* the PCI function will be scanned once function 0 appears */ if (!zdev->zbus->bus) return 0; diff --git a/arch/s390/pci/pci_event.c b/arch/s390/pci/pci_event.c index 6a5bfa9dc1f2..8df8b3210c5b 100644 --- a/arch/s390/pci/pci_event.c +++ b/arch/s390/pci/pci_event.c @@ -78,7 +78,7 @@ void zpci_event_error(void *data) static void zpci_event_hard_deconfigured(struct zpci_dev *zdev, u32 fh) { - zdev->fh = fh; + zpci_update_fh(zdev, fh); /* Give the driver a hint that the function is * already unusable. */ @@ -121,7 +121,7 @@ static void __zpci_event_availability(struct zpci_ccdf_avail *ccdf) if (!zdev) zpci_create_device(ccdf->fid, ccdf->fh, ZPCI_FN_STATE_STANDBY); else - zdev->fh = ccdf->fh; + zpci_update_fh(zdev, ccdf->fh); break; case 0x0303: /* Deconfiguration requested */ if (zdev) { @@ -130,7 +130,7 @@ static void __zpci_event_availability(struct zpci_ccdf_avail *ccdf) */ if (zdev->state != ZPCI_FN_STATE_CONFIGURED) break; - zdev->fh = ccdf->fh; + zpci_update_fh(zdev, ccdf->fh); zpci_deconfigure_device(zdev); } break; diff --git a/arch/s390/pci/pci_insn.c b/arch/s390/pci/pci_insn.c index 2e43996159f0..28d863aaafea 100644 --- a/arch/s390/pci/pci_insn.c +++ b/arch/s390/pci/pci_insn.c @@ -163,7 +163,7 @@ static inline int zpci_load_fh(u64 *data, const volatile void __iomem *addr, unsigned long len) { struct zpci_iomap_entry *entry = &zpci_iomap_start[ZPCI_IDX(addr)]; - u64 req = ZPCI_CREATE_REQ(entry->fh, entry->bar, len); + u64 req = ZPCI_CREATE_REQ(READ_ONCE(entry->fh), entry->bar, len); return __zpci_load(data, req, ZPCI_OFFSET(addr)); } @@ -244,7 +244,7 @@ static inline int zpci_store_fh(const volatile void __iomem *addr, u64 data, unsigned long len) { struct zpci_iomap_entry *entry = &zpci_iomap_start[ZPCI_IDX(addr)]; - u64 req = ZPCI_CREATE_REQ(entry->fh, entry->bar, len); + u64 req = ZPCI_CREATE_REQ(READ_ONCE(entry->fh), entry->bar, len); return __zpci_store(data, req, ZPCI_OFFSET(addr)); } From da995d538d3a17610d89fea0f5813cf7921b3c2c Mon Sep 17 00:00:00 2001 From: Niklas Schnelle Date: Thu, 1 Jul 2021 15:49:11 +0200 Subject: [PATCH 2/7] s390/pci: implement reset_slot for hotplug slot This is done by adding a zpci_hot_reset_device() call which does a low level reset of the PCI function without changing its higher level function state. This way it can be used while the zPCI function is bound to a driver and with DMA tables being controlled either through the IOMMU or DMA APIs which is prohibited when using zpci_disable_device() as that drop existing DMA translations. As this reset, unlike a normal FLR, also calls zpci_clear_irq() we need to implement arch_restore_msi_irqs() and make sure we re-enable IRQs for the PCI function if they were previously disabled. Reviewed-by: Pierre Morel Reviewed-by: Matthew Rosato Signed-off-by: Niklas Schnelle Signed-off-by: Vasily Gorbik --- arch/s390/include/asm/pci.h | 1 + arch/s390/pci/pci.c | 59 ++++++++++++++++++++++++++++++ arch/s390/pci/pci_irq.c | 9 +++++ drivers/pci/hotplug/s390_pci_hpc.c | 24 ++++++++++++ 4 files changed, 93 insertions(+) diff --git a/arch/s390/include/asm/pci.h b/arch/s390/include/asm/pci.h index 35adc0cd0e6a..a47068a73969 100644 --- a/arch/s390/include/asm/pci.h +++ b/arch/s390/include/asm/pci.h @@ -210,6 +210,7 @@ int zpci_deconfigure_device(struct zpci_dev *zdev); void zpci_device_reserved(struct zpci_dev *zdev); bool zpci_is_device_configured(struct zpci_dev *zdev); +int zpci_hot_reset_device(struct zpci_dev *zdev); int zpci_register_ioat(struct zpci_dev *, u8, u64, u64, u64); int zpci_unregister_ioat(struct zpci_dev *, u8); void zpci_remove_reserved_devices(void); diff --git a/arch/s390/pci/pci.c b/arch/s390/pci/pci.c index fe01ca3b33fa..e95b08e5cc5b 100644 --- a/arch/s390/pci/pci.c +++ b/arch/s390/pci/pci.c @@ -723,6 +723,65 @@ int zpci_disable_device(struct zpci_dev *zdev) return rc; } +/** + * zpci_hot_reset_device - perform a reset of the given zPCI function + * @zdev: the slot which should be reset + * + * Performs a low level reset of the zPCI function. The reset is low level in + * the sense that the zPCI function can be reset without detaching it from the + * common PCI subsystem. The reset may be performed while under control of + * either DMA or IOMMU APIs in which case the existing DMA/IOMMU translation + * table is reinstated at the end of the reset. + * + * After the reset the functions internal state is reset to an initial state + * equivalent to its state during boot when first probing a driver. + * Consequently after reset the PCI function requires re-initialization via the + * common PCI code including re-enabling IRQs via pci_alloc_irq_vectors() + * and enabling the function via e.g.pci_enablde_device_flags().The caller + * must guard against concurrent reset attempts. + * + * In most cases this function should not be called directly but through + * pci_reset_function() or pci_reset_bus() which handle the save/restore and + * locking. + * + * Return: 0 on success and an error value otherwise + */ +int zpci_hot_reset_device(struct zpci_dev *zdev) +{ + int rc; + + zpci_dbg(3, "rst fid:%x, fh:%x\n", zdev->fid, zdev->fh); + if (zdev_enabled(zdev)) { + /* Disables device access, DMAs and IRQs (reset state) */ + rc = zpci_disable_device(zdev); + /* + * Due to a z/VM vs LPAR inconsistency in the error state the + * FH may indicate an enabled device but disable says the + * device is already disabled don't treat it as an error here. + */ + if (rc == -EINVAL) + rc = 0; + if (rc) + return rc; + } + + rc = zpci_enable_device(zdev); + if (rc) + return rc; + + if (zdev->dma_table) + rc = zpci_register_ioat(zdev, 0, zdev->start_dma, zdev->end_dma, + (u64)zdev->dma_table); + else + rc = zpci_dma_init_device(zdev); + if (rc) { + zpci_disable_device(zdev); + return rc; + } + + return 0; +} + /** * zpci_create_device() - Create a new zpci_dev and add it to the zbus * @fid: Function ID of the device to be created diff --git a/arch/s390/pci/pci_irq.c b/arch/s390/pci/pci_irq.c index 3823e159bf74..954bb7a83124 100644 --- a/arch/s390/pci/pci_irq.c +++ b/arch/s390/pci/pci_irq.c @@ -387,6 +387,15 @@ void arch_teardown_msi_irqs(struct pci_dev *pdev) airq_iv_free(zpci_ibv[0], zdev->msi_first_bit, zdev->msi_nr_irqs); } +void arch_restore_msi_irqs(struct pci_dev *pdev) +{ + struct zpci_dev *zdev = to_zpci(pdev); + + if (!zdev->irqs_registered) + zpci_set_irq(zdev); + default_restore_msi_irqs(pdev); +} + static struct airq_struct zpci_airq = { .handler = zpci_floating_irq_handler, .isc = PCI_ISC, diff --git a/drivers/pci/hotplug/s390_pci_hpc.c b/drivers/pci/hotplug/s390_pci_hpc.c index dcefdb42ac46..a89b7de72dcf 100644 --- a/drivers/pci/hotplug/s390_pci_hpc.c +++ b/drivers/pci/hotplug/s390_pci_hpc.c @@ -57,6 +57,29 @@ static int disable_slot(struct hotplug_slot *hotplug_slot) return zpci_deconfigure_device(zdev); } +static int reset_slot(struct hotplug_slot *hotplug_slot, bool probe) +{ + struct zpci_dev *zdev = container_of(hotplug_slot, struct zpci_dev, + hotplug_slot); + + if (zdev->state != ZPCI_FN_STATE_CONFIGURED) + return -EIO; + /* + * We can't take the zdev->lock as reset_slot may be called during + * probing and/or device removal which already happens under the + * zdev->lock. Instead the user should use the higher level + * pci_reset_function() or pci_bus_reset() which hold the PCI device + * lock preventing concurrent removal. If not using these functions + * holding the PCI device lock is required. + */ + + /* As long as the function is configured we can reset */ + if (probe) + return 0; + + return zpci_hot_reset_device(zdev); +} + static int get_power_status(struct hotplug_slot *hotplug_slot, u8 *value) { struct zpci_dev *zdev = container_of(hotplug_slot, struct zpci_dev, @@ -76,6 +99,7 @@ static int get_adapter_status(struct hotplug_slot *hotplug_slot, u8 *value) static const struct hotplug_slot_ops s390_hotplug_slot_ops = { .enable_slot = enable_slot, .disable_slot = disable_slot, + .reset_slot = reset_slot, .get_power_status = get_power_status, .get_adapter_status = get_adapter_status, }; From dfd5bb23ad75bdabde89ac3166705a450bf16acb Mon Sep 17 00:00:00 2001 From: Niklas Schnelle Date: Wed, 5 May 2021 14:00:06 +0200 Subject: [PATCH 3/7] PCI: Export pci_dev_lock() Commit e3a9b1212b9d ("PCI: Export pci_dev_trylock() and pci_dev_unlock()") already exported pci_dev_trylock()/pci_dev_unlock() however in some circumstances such as during error recovery it makes sense to block waiting to get full access to the device so also export pci_dev_lock(). Link: https://lore.kernel.org/all/20210928181014.GA713179@bhelgaas/ Acked-by: Pierre Morel Acked-by: Bjorn Helgaas Signed-off-by: Niklas Schnelle Signed-off-by: Vasily Gorbik --- drivers/pci/pci.c | 3 ++- include/linux/pci.h | 1 + 2 files changed, 3 insertions(+), 1 deletion(-) diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c index da75c422ba85..193bfcda9c26 100644 --- a/drivers/pci/pci.c +++ b/drivers/pci/pci.c @@ -5074,12 +5074,13 @@ static int pci_reset_bus_function(struct pci_dev *dev, bool probe) return pci_parent_bus_reset(dev, probe); } -static void pci_dev_lock(struct pci_dev *dev) +void pci_dev_lock(struct pci_dev *dev) { pci_cfg_access_lock(dev); /* block PM suspend, driver probe, etc. */ device_lock(&dev->dev); } +EXPORT_SYMBOL_GPL(pci_dev_lock); /* Return 1 on successful lock, 0 on contention */ int pci_dev_trylock(struct pci_dev *dev) diff --git a/include/linux/pci.h b/include/linux/pci.h index b4dbcc86b3f1..d307b071b65e 100644 --- a/include/linux/pci.h +++ b/include/linux/pci.h @@ -1664,6 +1664,7 @@ void pci_cfg_access_lock(struct pci_dev *dev); bool pci_cfg_access_trylock(struct pci_dev *dev); void pci_cfg_access_unlock(struct pci_dev *dev); +void pci_dev_lock(struct pci_dev *dev); int pci_dev_trylock(struct pci_dev *dev); void pci_dev_unlock(struct pci_dev *dev); From 4cdf2f4e24ff0d345fc36ef6d6aec059333a261e Mon Sep 17 00:00:00 2001 From: Niklas Schnelle Date: Wed, 7 Jul 2021 11:00:01 +0200 Subject: [PATCH 4/7] s390/pci: implement minimal PCI error recovery When the platform detects an error on a PCI function or a service action has been performed it is put in the error state and an error event notification is provided to the OS. Currently we treat all error event notifications the same and simply set pdev->error_state = pci_channel_io_perm_failure requiring user intervention such as use of the recover attribute to get the device usable again. Despite requiring a manual step this also has the disadvantage that the device is completely torn down and recreated resulting in higher level devices such as a block or network device being recreated. In case of a block device this also means that it may need to be removed and added to a software raid even if that could otherwise survive with a temporary degradation. This is of course not ideal more so since an error notification with PEC 0x3A indicates that the platform already performed error recovery successfully or that the error state was caused by a service action that is now finished. At least in this case we can assume that the error state can be reset and the function made usable again. So as not to have the disadvantage of a full tear down and recreation we need to coordinate this recovery with the driver. Thankfully there is already a well defined recovery flow for this described in Documentation/PCI/pci-error-recovery.rst. The implementation of this is somewhat straight forward and simplified by the fact that our recovery flow is defined per PCI function. As a reset we use the newly introduced zpci_hot_reset_device() which also takes the PCI function out of the error state. Reviewed-by: Pierre Morel Acked-by: Matthew Rosato Signed-off-by: Niklas Schnelle Signed-off-by: Vasily Gorbik --- arch/s390/include/asm/pci.h | 4 +- arch/s390/pci/pci.c | 53 +++++++++ arch/s390/pci/pci_event.c | 224 +++++++++++++++++++++++++++++++++++- 3 files changed, 277 insertions(+), 4 deletions(-) diff --git a/arch/s390/include/asm/pci.h b/arch/s390/include/asm/pci.h index a47068a73969..90824be5ce9a 100644 --- a/arch/s390/include/asm/pci.h +++ b/arch/s390/include/asm/pci.h @@ -296,8 +296,10 @@ void zpci_debug_exit(void); void zpci_debug_init_device(struct zpci_dev *, const char *); void zpci_debug_exit_device(struct zpci_dev *); -/* Error reporting */ +/* Error handling */ int zpci_report_error(struct pci_dev *, struct zpci_report_error_header *); +int zpci_clear_error_state(struct zpci_dev *zdev); +int zpci_reset_load_store_blocked(struct zpci_dev *zdev); #ifdef CONFIG_NUMA diff --git a/arch/s390/pci/pci.c b/arch/s390/pci/pci.c index e95b08e5cc5b..2f9b78fa82a5 100644 --- a/arch/s390/pci/pci.c +++ b/arch/s390/pci/pci.c @@ -990,6 +990,59 @@ int zpci_report_error(struct pci_dev *pdev, } EXPORT_SYMBOL(zpci_report_error); +/** + * zpci_clear_error_state() - Clears the zPCI error state of the device + * @zdev: The zdev for which the zPCI error state should be reset + * + * Clear the zPCI error state of the device. If clearing the zPCI error state + * fails the device is left in the error state. In this case it may make sense + * to call zpci_io_perm_failure() on the associated pdev if it exists. + * + * Returns: 0 on success, -EIO otherwise + */ +int zpci_clear_error_state(struct zpci_dev *zdev) +{ + u64 req = ZPCI_CREATE_REQ(zdev->fh, 0, ZPCI_MOD_FC_RESET_ERROR); + struct zpci_fib fib = {0}; + u8 status; + int cc; + + cc = zpci_mod_fc(req, &fib, &status); + if (cc) { + zpci_dbg(3, "ces fid:%x, cc:%d, status:%x\n", zdev->fid, cc, status); + return -EIO; + } + + return 0; +} + +/** + * zpci_reset_load_store_blocked() - Re-enables L/S from error state + * @zdev: The zdev for which to unblock load/store access + * + * Re-enables load/store access for a PCI function in the error state while + * keeping DMA blocked. In this state drivers can poke MMIO space to determine + * if error recovery is possible while catching any rogue DMA access from the + * device. + * + * Returns: 0 on success, -EIO otherwise + */ +int zpci_reset_load_store_blocked(struct zpci_dev *zdev) +{ + u64 req = ZPCI_CREATE_REQ(zdev->fh, 0, ZPCI_MOD_FC_RESET_BLOCK); + struct zpci_fib fib = {0}; + u8 status; + int cc; + + cc = zpci_mod_fc(req, &fib, &status); + if (cc) { + zpci_dbg(3, "rls fid:%x, cc:%d, status:%x\n", zdev->fid, cc, status); + return -EIO; + } + + return 0; +} + static int zpci_mem_init(void) { BUILD_BUG_ON(!is_power_of_2(__alignof__(struct zpci_fmb)) || diff --git a/arch/s390/pci/pci_event.c b/arch/s390/pci/pci_event.c index 8df8b3210c5b..2e3e5b278925 100644 --- a/arch/s390/pci/pci_event.c +++ b/arch/s390/pci/pci_event.c @@ -47,18 +47,223 @@ struct zpci_ccdf_avail { u16 pec; /* PCI event code */ } __packed; +static inline bool ers_result_indicates_abort(pci_ers_result_t ers_res) +{ + switch (ers_res) { + case PCI_ERS_RESULT_CAN_RECOVER: + case PCI_ERS_RESULT_RECOVERED: + case PCI_ERS_RESULT_NEED_RESET: + return false; + default: + return true; + } +} + +static bool is_passed_through(struct zpci_dev *zdev) +{ + return zdev->s390_domain; +} + +static bool is_driver_supported(struct pci_driver *driver) +{ + if (!driver || !driver->err_handler) + return false; + if (!driver->err_handler->error_detected) + return false; + if (!driver->err_handler->slot_reset) + return false; + if (!driver->err_handler->resume) + return false; + return true; +} + +static pci_ers_result_t zpci_event_notify_error_detected(struct pci_dev *pdev, + struct pci_driver *driver) +{ + pci_ers_result_t ers_res = PCI_ERS_RESULT_DISCONNECT; + + ers_res = driver->err_handler->error_detected(pdev, pdev->error_state); + if (ers_result_indicates_abort(ers_res)) + pr_info("%s: Automatic recovery failed after initial reporting\n", pci_name(pdev)); + else if (ers_res == PCI_ERS_RESULT_NEED_RESET) + pr_debug("%s: Driver needs reset to recover\n", pci_name(pdev)); + + return ers_res; +} + +static pci_ers_result_t zpci_event_do_error_state_clear(struct pci_dev *pdev, + struct pci_driver *driver) +{ + pci_ers_result_t ers_res = PCI_ERS_RESULT_DISCONNECT; + struct zpci_dev *zdev = to_zpci(pdev); + int rc; + + pr_info("%s: Unblocking device access for examination\n", pci_name(pdev)); + rc = zpci_reset_load_store_blocked(zdev); + if (rc) { + pr_err("%s: Unblocking device access failed\n", pci_name(pdev)); + /* Let's try a full reset instead */ + return PCI_ERS_RESULT_NEED_RESET; + } + + if (driver->err_handler->mmio_enabled) { + ers_res = driver->err_handler->mmio_enabled(pdev); + if (ers_result_indicates_abort(ers_res)) { + pr_info("%s: Automatic recovery failed after MMIO re-enable\n", + pci_name(pdev)); + return ers_res; + } else if (ers_res == PCI_ERS_RESULT_NEED_RESET) { + pr_debug("%s: Driver needs reset to recover\n", pci_name(pdev)); + return ers_res; + } + } + + pr_debug("%s: Unblocking DMA\n", pci_name(pdev)); + rc = zpci_clear_error_state(zdev); + if (!rc) { + pdev->error_state = pci_channel_io_normal; + } else { + pr_err("%s: Unblocking DMA failed\n", pci_name(pdev)); + /* Let's try a full reset instead */ + return PCI_ERS_RESULT_NEED_RESET; + } + + return ers_res; +} + +static pci_ers_result_t zpci_event_do_reset(struct pci_dev *pdev, + struct pci_driver *driver) +{ + pci_ers_result_t ers_res = PCI_ERS_RESULT_DISCONNECT; + + pr_info("%s: Initiating reset\n", pci_name(pdev)); + if (zpci_hot_reset_device(to_zpci(pdev))) { + pr_err("%s: The reset request failed\n", pci_name(pdev)); + return ers_res; + } + pdev->error_state = pci_channel_io_normal; + ers_res = driver->err_handler->slot_reset(pdev); + if (ers_result_indicates_abort(ers_res)) { + pr_info("%s: Automatic recovery failed after slot reset\n", pci_name(pdev)); + return ers_res; + } + + return ers_res; +} + +/* zpci_event_attempt_error_recovery - Try to recover the given PCI function + * @pdev: PCI function to recover currently in the error state + * + * We follow the scheme outlined in Documentation/PCI/pci-error-recovery.rst. + * With the simplification that recovery always happens per function + * and the platform determines which functions are affected for + * multi-function devices. + */ +static pci_ers_result_t zpci_event_attempt_error_recovery(struct pci_dev *pdev) +{ + pci_ers_result_t ers_res = PCI_ERS_RESULT_DISCONNECT; + struct pci_driver *driver; + + /* + * Ensure that the PCI function is not removed concurrently, no driver + * is unbound or probed and that userspace can't access its + * configuration space while we perform recovery. + */ + pci_dev_lock(pdev); + if (pdev->error_state == pci_channel_io_perm_failure) { + ers_res = PCI_ERS_RESULT_DISCONNECT; + goto out_unlock; + } + pdev->error_state = pci_channel_io_frozen; + + if (is_passed_through(to_zpci(pdev))) { + pr_info("%s: Cannot be recovered in the host because it is a pass-through device\n", + pci_name(pdev)); + goto out_unlock; + } + + driver = to_pci_driver(pdev->dev.driver); + if (!is_driver_supported(driver)) { + if (!driver) + pr_info("%s: Cannot be recovered because no driver is bound to the device\n", + pci_name(pdev)); + else + pr_info("%s: The %s driver bound to the device does not support error recovery\n", + pci_name(pdev), + driver->name); + goto out_unlock; + } + + ers_res = zpci_event_notify_error_detected(pdev, driver); + if (ers_result_indicates_abort(ers_res)) + goto out_unlock; + + if (ers_res == PCI_ERS_RESULT_CAN_RECOVER) { + ers_res = zpci_event_do_error_state_clear(pdev, driver); + if (ers_result_indicates_abort(ers_res)) + goto out_unlock; + } + + if (ers_res == PCI_ERS_RESULT_NEED_RESET) + ers_res = zpci_event_do_reset(pdev, driver); + + if (ers_res != PCI_ERS_RESULT_RECOVERED) { + pr_err("%s: Automatic recovery failed; operator intervention is required\n", + pci_name(pdev)); + goto out_unlock; + } + + pr_info("%s: The device is ready to resume operations\n", pci_name(pdev)); + if (driver->err_handler->resume) + driver->err_handler->resume(pdev); +out_unlock: + pci_dev_unlock(pdev); + + return ers_res; +} + +/* zpci_event_io_failure - Report PCI channel failure state to driver + * @pdev: PCI function for which to report + * @es: PCI channel failure state to report + */ +static void zpci_event_io_failure(struct pci_dev *pdev, pci_channel_state_t es) +{ + struct pci_driver *driver; + + pci_dev_lock(pdev); + pdev->error_state = es; + /** + * While vfio-pci's error_detected callback notifies user-space QEMU + * reacts to this by freezing the guest. In an s390 environment PCI + * errors are rarely fatal so this is overkill. Instead in the future + * we will inject the error event and let the guest recover the device + * itself. + */ + if (is_passed_through(to_zpci(pdev))) + goto out; + driver = to_pci_driver(pdev->dev.driver); + if (driver && driver->err_handler && driver->err_handler->error_detected) + driver->err_handler->error_detected(pdev, pdev->error_state); +out: + pci_dev_unlock(pdev); +} + static void __zpci_event_error(struct zpci_ccdf_err *ccdf) { struct zpci_dev *zdev = get_zdev_by_fid(ccdf->fid); struct pci_dev *pdev = NULL; + pci_ers_result_t ers_res; zpci_dbg(3, "err fid:%x, fh:%x, pec:%x\n", ccdf->fid, ccdf->fh, ccdf->pec); zpci_err("error CCDF:\n"); zpci_err_hex(ccdf, sizeof(*ccdf)); - if (zdev) - pdev = pci_get_slot(zdev->zbus->bus, zdev->devfn); + if (zdev) { + zpci_update_fh(zdev, ccdf->fh); + if (zdev->zbus->bus) + pdev = pci_get_slot(zdev->zbus->bus, zdev->devfn); + } pr_err("%s: Event 0x%x reports an error for PCI function 0x%x\n", pdev ? pci_name(pdev) : "n/a", ccdf->pec, ccdf->fid); @@ -66,7 +271,20 @@ static void __zpci_event_error(struct zpci_ccdf_err *ccdf) if (!pdev) return; - pdev->error_state = pci_channel_io_perm_failure; + switch (ccdf->pec) { + case 0x003a: /* Service Action or Error Recovery Successful */ + ers_res = zpci_event_attempt_error_recovery(pdev); + if (ers_res != PCI_ERS_RESULT_RECOVERED) + zpci_event_io_failure(pdev, pci_channel_io_perm_failure); + break; + default: + /* + * Mark as frozen not permanently failed because the device + * could be subsequently recovered by the platform. + */ + zpci_event_io_failure(pdev, pci_channel_io_frozen); + break; + } pci_dev_put(pdev); } From 213fca9e23b59581c573d558aa477556f00b8198 Mon Sep 17 00:00:00 2001 From: Sven Schnelle Date: Tue, 2 Nov 2021 10:55:30 +0100 Subject: [PATCH 5/7] s390/tape: fix timer initialization in tape_std_assign() commit 9c6c273aa424 ("timer: Remove init_timer_on_stack() in favor of timer_setup_on_stack()") changed the timer setup from init_timer_on_stack(() to timer_setup(), but missed to change the mod_timer() call. And while at it, use msecs_to_jiffies() instead of the open coded timeout calculation. Cc: stable@vger.kernel.org Fixes: 9c6c273aa424 ("timer: Remove init_timer_on_stack() in favor of timer_setup_on_stack()") Signed-off-by: Sven Schnelle Reviewed-by: Vasily Gorbik Signed-off-by: Vasily Gorbik --- drivers/s390/char/tape_std.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/drivers/s390/char/tape_std.c b/drivers/s390/char/tape_std.c index 1f5fab617b67..f7e75d9fedf6 100644 --- a/drivers/s390/char/tape_std.c +++ b/drivers/s390/char/tape_std.c @@ -53,7 +53,6 @@ int tape_std_assign(struct tape_device *device) { int rc; - struct timer_list timeout; struct tape_request *request; request = tape_alloc_request(2, 11); @@ -70,7 +69,7 @@ tape_std_assign(struct tape_device *device) * So we set up a timeout for this call. */ timer_setup(&request->timer, tape_std_assign_timeout, 0); - mod_timer(&timeout, jiffies + 2 * HZ); + mod_timer(&request->timer, jiffies + msecs_to_jiffies(2000)); rc = tape_do_io_interruptible(device, request); From 9d48c7afedf91a02d03295837ec76b2fb5e7d3fe Mon Sep 17 00:00:00 2001 From: Thomas Richter Date: Wed, 3 Nov 2021 13:13:04 +0100 Subject: [PATCH 6/7] s390/cpumf: cpum_cf PMU displays invalid value after hotplug remove When a CPU is hotplugged while the perf stat -e cycles command is running, a wrong (very large) value is displayed immediately after the CPU removal: Check the values, shouldn't be too high as in time counts unit events 1.001101919 29261846 cycles 2.002454499 17523405 cycles 3.003659292 24361161 cycles 4.004816983 18446744073638406144 cycles 5.005671647 cycles ... The CPU hotplug off took place after 3 seconds. The issue is the read of the event count value after 4 seconds when the CPU is not available and the read of the counter returns an error. This is treated as a counter value of zero. This results in a very large value (0 - previous_value). Fix this by detecting the hotplugged off CPU and report 0 instead of a very large number. Cc: stable@vger.kernel.org Fixes: a029a4eab39e ("s390/cpumf: Allow concurrent access for CPU Measurement Counter Facility") Reported-by: Sumanth Korikkar Signed-off-by: Thomas Richter Reviewed-by: Sumanth Korikkar Signed-off-by: Vasily Gorbik --- arch/s390/kernel/perf_cpum_cf.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/arch/s390/kernel/perf_cpum_cf.c b/arch/s390/kernel/perf_cpum_cf.c index 6f431fa9e4d7..ee8707abdb6a 100644 --- a/arch/s390/kernel/perf_cpum_cf.c +++ b/arch/s390/kernel/perf_cpum_cf.c @@ -687,8 +687,10 @@ static void cpumf_pmu_stop(struct perf_event *event, int flags) false); if (cfdiag_diffctr(cpuhw, event->hw.config_base)) cfdiag_push_sample(event, cpuhw); - } else + } else if (cpuhw->flags & PMU_F_RESERVED) { + /* Only update when PMU not hotplugged off */ hw_perf_event_update(event); + } hwc->state |= PERF_HES_UPTODATE; } } From a4751f157c194431fae9e9c493f456df8272b871 Mon Sep 17 00:00:00 2001 From: Vineeth Vijayan Date: Fri, 5 Nov 2021 16:44:51 +0100 Subject: [PATCH 7/7] s390/cio: check the subchannel validity for dev_busid Check the validity of subchanel before reading other fields in the schib. Fixes: d3683c055212 ("s390/cio: add dev_busid sysfs entry for each subchannel") CC: Reported-by: Cornelia Huck Signed-off-by: Vineeth Vijayan Reviewed-by: Cornelia Huck Link: https://lore.kernel.org/r/20211105154451.847288-1-vneethv@linux.ibm.com Signed-off-by: Vasily Gorbik --- drivers/s390/cio/css.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/s390/cio/css.c b/drivers/s390/cio/css.c index 2bc55ccf3f23..ce9e7517430f 100644 --- a/drivers/s390/cio/css.c +++ b/drivers/s390/cio/css.c @@ -437,8 +437,8 @@ static ssize_t dev_busid_show(struct device *dev, struct subchannel *sch = to_subchannel(dev); struct pmcw *pmcw = &sch->schib.pmcw; - if ((pmcw->st == SUBCHANNEL_TYPE_IO || - pmcw->st == SUBCHANNEL_TYPE_MSG) && pmcw->dnv) + if ((pmcw->st == SUBCHANNEL_TYPE_IO && pmcw->dnv) || + (pmcw->st == SUBCHANNEL_TYPE_MSG && pmcw->w)) return sysfs_emit(buf, "0.%x.%04x\n", sch->schid.ssid, pmcw->dev); else