From fb51ccbf217c1c994607b6519c7d85250928553d Mon Sep 17 00:00:00 2001 From: Jan Kiszka Date: Fri, 4 Nov 2011 09:45:59 +0100 Subject: PCI: Rework config space blocking services pci_block_user_cfg_access was designed for the use case that a single context, the IPR driver, temporarily delays user space accesses to the config space via sysfs. This assumption became invalid by the time pci_dev_reset was added as locking instance. Today, if you run two loops in parallel that reset the same device via sysfs, you end up with a kernel BUG as pci_block_user_cfg_access detect the broken assumption. This reworks the pci_block_user_cfg_access to a sleeping service pci_cfg_access_lock and an atomic-compatible variant called pci_cfg_access_trylock. The former not only blocks user space access as before but also waits if access was already locked. The latter service just returns false in this case, allowing the caller to resolve the conflict instead of raising a BUG. Adaptions of the ipr driver were originally written by Brian King. Acked-by: Brian King Acked-by: Michael S. Tsirkin Signed-off-by: Jan Kiszka Signed-off-by: Jesse Barnes --- drivers/pci/access.c | 74 ++++++++++++++++++++++++++++++++++------------------ 1 file changed, 48 insertions(+), 26 deletions(-) (limited to 'drivers/pci/access.c') diff --git a/drivers/pci/access.c b/drivers/pci/access.c index fdaa42aac7c..0c4c71712df 100644 --- a/drivers/pci/access.c +++ b/drivers/pci/access.c @@ -127,20 +127,20 @@ EXPORT_SYMBOL(pci_write_vpd); * We have a bit per device to indicate it's blocked and a global wait queue * for callers to sleep on until devices are unblocked. */ -static DECLARE_WAIT_QUEUE_HEAD(pci_ucfg_wait); +static DECLARE_WAIT_QUEUE_HEAD(pci_cfg_wait); -static noinline void pci_wait_ucfg(struct pci_dev *dev) +static noinline void pci_wait_cfg(struct pci_dev *dev) { DECLARE_WAITQUEUE(wait, current); - __add_wait_queue(&pci_ucfg_wait, &wait); + __add_wait_queue(&pci_cfg_wait, &wait); do { set_current_state(TASK_UNINTERRUPTIBLE); raw_spin_unlock_irq(&pci_lock); schedule(); raw_spin_lock_irq(&pci_lock); - } while (dev->block_ucfg_access); - __remove_wait_queue(&pci_ucfg_wait, &wait); + } while (dev->block_cfg_access); + __remove_wait_queue(&pci_cfg_wait, &wait); } /* Returns 0 on success, negative values indicate error. */ @@ -153,7 +153,8 @@ int pci_user_read_config_##size \ if (PCI_##size##_BAD) \ return -EINVAL; \ raw_spin_lock_irq(&pci_lock); \ - if (unlikely(dev->block_ucfg_access)) pci_wait_ucfg(dev); \ + if (unlikely(dev->block_cfg_access)) \ + pci_wait_cfg(dev); \ ret = dev->bus->ops->read(dev->bus, dev->devfn, \ pos, sizeof(type), &data); \ raw_spin_unlock_irq(&pci_lock); \ @@ -172,7 +173,8 @@ int pci_user_write_config_##size \ if (PCI_##size##_BAD) \ return -EINVAL; \ raw_spin_lock_irq(&pci_lock); \ - if (unlikely(dev->block_ucfg_access)) pci_wait_ucfg(dev); \ + if (unlikely(dev->block_cfg_access)) \ + pci_wait_cfg(dev); \ ret = dev->bus->ops->write(dev->bus, dev->devfn, \ pos, sizeof(type), val); \ raw_spin_unlock_irq(&pci_lock); \ @@ -401,36 +403,56 @@ int pci_vpd_truncate(struct pci_dev *dev, size_t size) EXPORT_SYMBOL(pci_vpd_truncate); /** - * pci_block_user_cfg_access - Block userspace PCI config reads/writes + * pci_cfg_access_lock - Lock PCI config reads/writes * @dev: pci device struct * - * When user access is blocked, any reads or writes to config space will - * sleep until access is unblocked again. We don't allow nesting of - * block/unblock calls. + * When access is locked, any userspace reads or writes to config + * space and concurrent lock requests will sleep until access is + * allowed via pci_cfg_access_unlocked again. */ -void pci_block_user_cfg_access(struct pci_dev *dev) +void pci_cfg_access_lock(struct pci_dev *dev) +{ + might_sleep(); + + raw_spin_lock_irq(&pci_lock); + if (dev->block_cfg_access) + pci_wait_cfg(dev); + dev->block_cfg_access = 1; + raw_spin_unlock_irq(&pci_lock); +} +EXPORT_SYMBOL_GPL(pci_cfg_access_lock); + +/** + * pci_cfg_access_trylock - try to lock PCI config reads/writes + * @dev: pci device struct + * + * Same as pci_cfg_access_lock, but will return 0 if access is + * already locked, 1 otherwise. This function can be used from + * atomic contexts. + */ +bool pci_cfg_access_trylock(struct pci_dev *dev) { unsigned long flags; - int was_blocked; + bool locked = true; raw_spin_lock_irqsave(&pci_lock, flags); - was_blocked = dev->block_ucfg_access; - dev->block_ucfg_access = 1; + if (dev->block_cfg_access) + locked = false; + else + dev->block_cfg_access = 1; raw_spin_unlock_irqrestore(&pci_lock, flags); - /* If we BUG() inside the pci_lock, we're guaranteed to hose - * the machine */ - BUG_ON(was_blocked); + return locked; } -EXPORT_SYMBOL_GPL(pci_block_user_cfg_access); +EXPORT_SYMBOL_GPL(pci_cfg_access_trylock); /** - * pci_unblock_user_cfg_access - Unblock userspace PCI config reads/writes + * pci_cfg_access_unlock - Unlock PCI config reads/writes * @dev: pci device struct * - * This function allows userspace PCI config accesses to resume. + * This function allows PCI config accesses to resume. */ -void pci_unblock_user_cfg_access(struct pci_dev *dev) +void pci_cfg_access_unlock(struct pci_dev *dev) { unsigned long flags; @@ -438,10 +460,10 @@ void pci_unblock_user_cfg_access(struct pci_dev *dev) /* This indicates a problem in the caller, but we don't need * to kill them, unlike a double-block above. */ - WARN_ON(!dev->block_ucfg_access); + WARN_ON(!dev->block_cfg_access); - dev->block_ucfg_access = 0; - wake_up_all(&pci_ucfg_wait); + dev->block_cfg_access = 0; + wake_up_all(&pci_cfg_wait); raw_spin_unlock_irqrestore(&pci_lock, flags); } -EXPORT_SYMBOL_GPL(pci_unblock_user_cfg_access); +EXPORT_SYMBOL_GPL(pci_cfg_access_unlock); -- cgit v1.2.3-70-g09d2 From a2e27787f893621c5a6b865acf6b7766f8671328 Mon Sep 17 00:00:00 2001 From: Jan Kiszka Date: Fri, 4 Nov 2011 09:46:00 +0100 Subject: PCI: Introduce INTx check & mask API These new PCI services allow to probe for 2.3-compliant INTx masking support and then use the feature from PCI interrupt handlers. The services are properly synchronized with concurrent config space access via sysfs or on device reset. This enables generic PCI device drivers like uio_pci_generic or KVM's device assignment to implement the necessary kernel-side IRQ handling without any knowledge about device-specific interrupt status and control registers. Acked-by: Michael S. Tsirkin Signed-off-by: Jan Kiszka Signed-off-by: Jesse Barnes --- drivers/pci/access.c | 2 +- drivers/pci/pci.c | 110 +++++++++++++++++++++++++++++++++++++++++++++++++++ drivers/pci/pci.h | 2 + include/linux/pci.h | 3 ++ 4 files changed, 116 insertions(+), 1 deletion(-) (limited to 'drivers/pci/access.c') diff --git a/drivers/pci/access.c b/drivers/pci/access.c index 0c4c71712df..2a581642c23 100644 --- a/drivers/pci/access.c +++ b/drivers/pci/access.c @@ -13,7 +13,7 @@ * configuration space. */ -static DEFINE_RAW_SPINLOCK(pci_lock); +DEFINE_RAW_SPINLOCK(pci_lock); /* * Wrappers for all PCI configuration access functions. They just check diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c index c3cca7cdc6e..924193ef4fe 100644 --- a/drivers/pci/pci.c +++ b/drivers/pci/pci.c @@ -2767,6 +2767,116 @@ pci_intx(struct pci_dev *pdev, int enable) } } +/** + * pci_intx_mask_supported - probe for INTx masking support + * @pdev: the PCI device to operate on + * + * Check if the device dev support INTx masking via the config space + * command word. + */ +bool pci_intx_mask_supported(struct pci_dev *dev) +{ + bool mask_supported = false; + u16 orig, new; + + pci_cfg_access_lock(dev); + + pci_read_config_word(dev, PCI_COMMAND, &orig); + pci_write_config_word(dev, PCI_COMMAND, + orig ^ PCI_COMMAND_INTX_DISABLE); + pci_read_config_word(dev, PCI_COMMAND, &new); + + /* + * There's no way to protect against hardware bugs or detect them + * reliably, but as long as we know what the value should be, let's + * go ahead and check it. + */ + if ((new ^ orig) & ~PCI_COMMAND_INTX_DISABLE) { + dev_err(&dev->dev, "Command register changed from " + "0x%x to 0x%x: driver or hardware bug?\n", orig, new); + } else if ((new ^ orig) & PCI_COMMAND_INTX_DISABLE) { + mask_supported = true; + pci_write_config_word(dev, PCI_COMMAND, orig); + } + + pci_cfg_access_unlock(dev); + return mask_supported; +} +EXPORT_SYMBOL_GPL(pci_intx_mask_supported); + +static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask) +{ + struct pci_bus *bus = dev->bus; + bool mask_updated = true; + u32 cmd_status_dword; + u16 origcmd, newcmd; + unsigned long flags; + bool irq_pending; + + /* + * We do a single dword read to retrieve both command and status. + * Document assumptions that make this possible. + */ + BUILD_BUG_ON(PCI_COMMAND % 4); + BUILD_BUG_ON(PCI_COMMAND + 2 != PCI_STATUS); + + raw_spin_lock_irqsave(&pci_lock, flags); + + bus->ops->read(bus, dev->devfn, PCI_COMMAND, 4, &cmd_status_dword); + + irq_pending = (cmd_status_dword >> 16) & PCI_STATUS_INTERRUPT; + + /* + * Check interrupt status register to see whether our device + * triggered the interrupt (when masking) or the next IRQ is + * already pending (when unmasking). + */ + if (mask != irq_pending) { + mask_updated = false; + goto done; + } + + origcmd = cmd_status_dword; + newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE; + if (mask) + newcmd |= PCI_COMMAND_INTX_DISABLE; + if (newcmd != origcmd) + bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd); + +done: + raw_spin_unlock_irqrestore(&pci_lock, flags); + + return mask_updated; +} + +/** + * pci_check_and_mask_intx - mask INTx on pending interrupt + * @pdev: the PCI device to operate on + * + * Check if the device dev has its INTx line asserted, mask it and + * return true in that case. False is returned if not interrupt was + * pending. + */ +bool pci_check_and_mask_intx(struct pci_dev *dev) +{ + return pci_check_and_set_intx_mask(dev, true); +} +EXPORT_SYMBOL_GPL(pci_check_and_mask_intx); + +/** + * pci_check_and_mask_intx - unmask INTx of no interrupt is pending + * @pdev: the PCI device to operate on + * + * Check if the device dev has its INTx line asserted, unmask it if not + * and return true. False is returned and the mask remains active if + * there was still an interrupt pending. + */ +bool pci_check_and_unmask_intx(struct pci_dev *dev) +{ + return pci_check_and_set_intx_mask(dev, false); +} +EXPORT_SYMBOL_GPL(pci_check_and_unmask_intx); + /** * pci_msi_off - disables any msi or msix capabilities * @dev: the PCI device to operate on diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h index b74084e9ca1..3b6e4ed306b 100644 --- a/drivers/pci/pci.h +++ b/drivers/pci/pci.h @@ -136,6 +136,8 @@ static inline void pci_remove_legacy_files(struct pci_bus *bus) { return; } /* Lock for read/write access to pci device and bus lists */ extern struct rw_semaphore pci_bus_sem; +extern raw_spinlock_t pci_lock; + extern unsigned int pci_pm_d3_delay; #ifdef CONFIG_PCI_MSI diff --git a/include/linux/pci.h b/include/linux/pci.h index 72401596b2a..4286b853956 100644 --- a/include/linux/pci.h +++ b/include/linux/pci.h @@ -804,6 +804,9 @@ int __must_check pci_set_mwi(struct pci_dev *dev); int pci_try_set_mwi(struct pci_dev *dev); void pci_clear_mwi(struct pci_dev *dev); void pci_intx(struct pci_dev *dev, int enable); +bool pci_intx_mask_supported(struct pci_dev *dev); +bool pci_check_and_mask_intx(struct pci_dev *dev); +bool pci_check_and_unmask_intx(struct pci_dev *dev); void pci_msi_off(struct pci_dev *dev); int pci_set_dma_max_seg_size(struct pci_dev *dev, unsigned int size); int pci_set_dma_seg_boundary(struct pci_dev *dev, unsigned long mask); -- cgit v1.2.3-70-g09d2