aboutsummaryrefslogtreecommitdiff
path: root/arch/powerpc/kernel/eeh_driver.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/powerpc/kernel/eeh_driver.c')
-rw-r--r--arch/powerpc/kernel/eeh_driver.c167
1 files changed, 141 insertions, 26 deletions
diff --git a/arch/powerpc/kernel/eeh_driver.c b/arch/powerpc/kernel/eeh_driver.c
index 7bb30dca4e1..420da61d4ce 100644
--- a/arch/powerpc/kernel/eeh_driver.c
+++ b/arch/powerpc/kernel/eeh_driver.c
@@ -143,17 +143,43 @@ static void eeh_disable_irq(struct pci_dev *dev)
static void eeh_enable_irq(struct pci_dev *dev)
{
struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
- struct irq_desc *desc;
if ((edev->mode) & EEH_DEV_IRQ_DISABLED) {
edev->mode &= ~EEH_DEV_IRQ_DISABLED;
-
- desc = irq_to_desc(dev->irq);
- if (desc && desc->depth > 0)
+ /*
+ * FIXME !!!!!
+ *
+ * This is just ass backwards. This maze has
+ * unbalanced irq_enable/disable calls. So instead of
+ * finding the root cause it works around the warning
+ * in the irq_enable code by conditionally calling
+ * into it.
+ *
+ * That's just wrong.The warning in the core code is
+ * there to tell people to fix their assymetries in
+ * their own code, not by abusing the core information
+ * to avoid it.
+ *
+ * I so wish that the assymetry would be the other way
+ * round and a few more irq_disable calls render that
+ * shit unusable forever.
+ *
+ * tglx
+ */
+ if (irqd_irq_disabled(irq_get_irq_data(dev->irq)))
enable_irq(dev->irq);
}
}
+static bool eeh_dev_removed(struct eeh_dev *edev)
+{
+ /* EEH device removed ? */
+ if (!edev || (edev->mode & EEH_DEV_REMOVED))
+ return true;
+
+ return false;
+}
+
/**
* eeh_report_error - Report pci error to each device driver
* @data: eeh device
@@ -170,10 +196,8 @@ static void *eeh_report_error(void *data, void *userdata)
enum pci_ers_result rc, *res = userdata;
struct pci_driver *driver;
- /* We might not have the associated PCI device,
- * then we should continue for next one.
- */
- if (!dev) return NULL;
+ if (!dev || eeh_dev_removed(edev))
+ return NULL;
dev->error_state = pci_channel_io_frozen;
driver = eeh_pcid_get(dev);
@@ -213,6 +237,9 @@ static void *eeh_report_mmio_enabled(void *data, void *userdata)
enum pci_ers_result rc, *res = userdata;
struct pci_driver *driver;
+ if (!dev || eeh_dev_removed(edev))
+ return NULL;
+
driver = eeh_pcid_get(dev);
if (!driver) return NULL;
@@ -250,7 +277,8 @@ static void *eeh_report_reset(void *data, void *userdata)
enum pci_ers_result rc, *res = userdata;
struct pci_driver *driver;
- if (!dev) return NULL;
+ if (!dev || eeh_dev_removed(edev))
+ return NULL;
dev->error_state = pci_channel_io_normal;
driver = eeh_pcid_get(dev);
@@ -290,7 +318,8 @@ static void *eeh_report_resume(void *data, void *userdata)
struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
struct pci_driver *driver;
- if (!dev) return NULL;
+ if (!dev || eeh_dev_removed(edev))
+ return NULL;
dev->error_state = pci_channel_io_normal;
driver = eeh_pcid_get(dev);
@@ -326,7 +355,8 @@ static void *eeh_report_failure(void *data, void *userdata)
struct pci_dev *dev = eeh_dev_to_pci_dev(edev);
struct pci_driver *driver;
- if (!dev) return NULL;
+ if (!dev || eeh_dev_removed(edev))
+ return NULL;
dev->error_state = pci_channel_io_perm_failure;
driver = eeh_pcid_get(dev);
@@ -362,10 +392,24 @@ static void *eeh_rmv_device(void *data, void *userdata)
*/
if (!dev || (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE))
return NULL;
- driver = eeh_pcid_get(dev);
- if (driver && driver->err_handler)
+
+ /*
+ * We rely on count-based pcibios_release_device() to
+ * detach permanently offlined PEs. Unfortunately, that's
+ * not reliable enough. We might have the permanently
+ * offlined PEs attached, but we needn't take care of
+ * them and their child devices.
+ */
+ if (eeh_dev_removed(edev))
return NULL;
+ driver = eeh_pcid_get(dev);
+ if (driver) {
+ eeh_pcid_put(dev);
+ if (driver->err_handler)
+ return NULL;
+ }
+
/* Remove it from PCI subsystem */
pr_debug("EEH: Removing %s without EEH sensitive driver\n",
pci_name(dev));
@@ -396,6 +440,48 @@ static void *eeh_pe_detach_dev(void *data, void *userdata)
return NULL;
}
+/*
+ * Explicitly clear PE's frozen state for PowerNV where
+ * we have frozen PE until BAR restore is completed. It's
+ * harmless to clear it for pSeries. To be consistent with
+ * PE reset (for 3 times), we try to clear the frozen state
+ * for 3 times as well.
+ */
+static void *__eeh_clear_pe_frozen_state(void *data, void *flag)
+{
+ struct eeh_pe *pe = (struct eeh_pe *)data;
+ int i, rc;
+
+ for (i = 0; i < 3; i++) {
+ rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
+ if (rc)
+ continue;
+ rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA);
+ if (!rc)
+ break;
+ }
+
+ /* The PE has been isolated, clear it */
+ if (rc) {
+ pr_warn("%s: Can't clear frozen PHB#%x-PE#%x (%d)\n",
+ __func__, pe->phb->global_number, pe->addr, rc);
+ return (void *)pe;
+ }
+
+ return NULL;
+}
+
+static int eeh_clear_pe_frozen_state(struct eeh_pe *pe)
+{
+ void *rc;
+
+ rc = eeh_pe_traverse(pe, __eeh_clear_pe_frozen_state, NULL);
+ if (!rc)
+ eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
+
+ return rc ? -EIO : 0;
+}
+
/**
* eeh_reset_device - Perform actual reset of a pci slot
* @pe: EEH PE
@@ -430,19 +516,33 @@ static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus)
eeh_pe_dev_traverse(pe, eeh_rmv_device, &removed);
}
- /* Reset the pci controller. (Asserts RST#; resets config space).
+ /*
+ * Reset the pci controller. (Asserts RST#; resets config space).
* Reconfigure bridges and devices. Don't try to bring the system
* up if the reset failed for some reason.
+ *
+ * During the reset, it's very dangerous to have uncontrolled PCI
+ * config accesses. So we prefer to block them. However, controlled
+ * PCI config accesses initiated from EEH itself are allowed.
*/
+ eeh_pe_state_mark(pe, EEH_PE_RESET);
rc = eeh_reset_pe(pe);
- if (rc)
+ if (rc) {
+ eeh_pe_state_clear(pe, EEH_PE_RESET);
return rc;
+ }
pci_lock_rescan_remove();
/* Restore PE */
eeh_ops->configure_bridge(pe);
eeh_pe_restore_bars(pe);
+ eeh_pe_state_clear(pe, EEH_PE_RESET);
+
+ /* Clear frozen state */
+ rc = eeh_clear_pe_frozen_state(pe);
+ if (rc)
+ return rc;
/* Give the system 5 seconds to finish running the user-space
* hotplug shutdown scripts, e.g. ifdown for ethernet. Yes,
@@ -552,7 +652,6 @@ static void eeh_handle_normal_event(struct eeh_pe *pe)
result = PCI_ERS_RESULT_NEED_RESET;
} else {
pr_info("EEH: Notify device drivers to resume I/O\n");
- result = PCI_ERS_RESULT_NONE;
eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result);
}
}
@@ -564,10 +663,17 @@ static void eeh_handle_normal_event(struct eeh_pe *pe)
if (rc < 0)
goto hard_fail;
- if (rc)
+ if (rc) {
result = PCI_ERS_RESULT_NEED_RESET;
- else
+ } else {
+ /*
+ * We didn't do PE reset for the case. The PE
+ * is still in frozen state. Clear it before
+ * resuming the PE.
+ */
+ eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
result = PCI_ERS_RESULT_RECOVERED;
+ }
}
/* If any device has a hard failure, then shut off everything. */
@@ -629,8 +735,17 @@ perm_error:
/* Notify all devices that they're about to go down. */
eeh_pe_dev_traverse(pe, eeh_report_failure, NULL);
- /* Shut down the device drivers for good. */
+ /* Mark the PE to be removed permanently */
+ pe->freeze_count = EEH_MAX_ALLOWED_FREEZES + 1;
+
+ /*
+ * Shut down the device drivers for good. We mark
+ * all removed devices correctly to avoid access
+ * the their PCI config any more.
+ */
if (frozen_bus) {
+ eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED);
+
pci_lock_rescan_remove();
pcibios_remove_pci_devices(frozen_bus);
pci_unlock_rescan_remove();
@@ -655,14 +770,13 @@ static void eeh_handle_special_event(void)
eeh_serialize_lock(&flags);
/* Purge all events */
- eeh_remove_event(NULL);
+ eeh_remove_event(NULL, true);
list_for_each_entry(hose, &hose_list, list_node) {
phb_pe = eeh_phb_pe_get(hose);
if (!phb_pe) continue;
- eeh_pe_state_mark(phb_pe,
- EEH_PE_ISOLATED | EEH_PE_PHB_DEAD);
+ eeh_pe_state_mark(phb_pe, EEH_PE_ISOLATED);
}
eeh_serialize_unlock(flags);
@@ -675,11 +789,10 @@ static void eeh_handle_special_event(void)
eeh_serialize_lock(&flags);
/* Purge all events of the PHB */
- eeh_remove_event(pe);
+ eeh_remove_event(pe, true);
if (rc == EEH_NEXT_ERR_DEAD_PHB)
- eeh_pe_state_mark(pe,
- EEH_PE_ISOLATED | EEH_PE_PHB_DEAD);
+ eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
else
eeh_pe_state_mark(pe,
EEH_PE_ISOLATED | EEH_PE_RECOVERING);
@@ -703,12 +816,14 @@ static void eeh_handle_special_event(void)
if (rc == EEH_NEXT_ERR_FROZEN_PE ||
rc == EEH_NEXT_ERR_FENCED_PHB) {
eeh_handle_normal_event(pe);
+ eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
} else {
pci_lock_rescan_remove();
list_for_each_entry(hose, &hose_list, list_node) {
phb_pe = eeh_phb_pe_get(hose);
if (!phb_pe ||
- !(phb_pe->state & EEH_PE_PHB_DEAD))
+ !(phb_pe->state & EEH_PE_ISOLATED) ||
+ (phb_pe->state & EEH_PE_RECOVERING))
continue;
/* Notify all devices to be down */
generated by cgit v1.2.3-70-g09d2 (git 2.46.0) at 2025-11-09 21:57:09 +0000