]> pilppa.com Git - linux-2.6-omap-h63xx.git/commitdiff
[PATCH] powerpc: PCI Error Recovery: PPC64 core recovery routines
authorLinas Vepstas <linas@linas.org>
Fri, 4 Nov 2005 00:52:49 +0000 (18:52 -0600)
committerPaul Mackerras <paulus@samba.org>
Tue, 10 Jan 2006 04:28:32 +0000 (15:28 +1100)
Various PCI bus errors can be signaled by newer PCI controllers.  The
core error recovery routines are architecture dependent.  This patch adds
a recovery infrastructure for the  PPC64 pSeries systems.

Signed-off-by: Linas Vepstas <linas@austin.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
(cherry picked from e8ca11b460c4c9c7fa6b529be221529ebd770e38 commit)

arch/powerpc/platforms/pseries/Makefile
arch/powerpc/platforms/pseries/eeh.c
arch/powerpc/platforms/pseries/eeh_driver.c [new file with mode: 0644]
arch/powerpc/platforms/pseries/eeh_event.c
include/asm-powerpc/eeh.h
include/asm-powerpc/eeh_event.h
include/asm-powerpc/ppc-pci.h

index 6accdd1555058861a08ab43fadf6caa0582ad415..0b7d5cb6417712bd4b1534cce09116486b13001f 100644 (file)
@@ -4,7 +4,7 @@ obj-$(CONFIG_SMP)       += smp.o
 obj-$(CONFIG_IBMVIO)   += vio.o
 obj-$(CONFIG_XICS)     += xics.o
 obj-$(CONFIG_SCANLOG)  += scanlog.o
-obj-$(CONFIG_EEH)      += eeh.o eeh_event.o
+obj-$(CONFIG_EEH)      += eeh.o eeh_driver.o eeh_event.o
 
 obj-$(CONFIG_HVC_CONSOLE)      += hvconsole.o
 obj-$(CONFIG_HVCS)             += hvcserver.o
index 7fbfd16d72b703b91b2e1e91a5f60935018e0af2..d6560c45637bb06afaee30d56cbef3c52fe1d151 100644 (file)
@@ -485,6 +485,11 @@ static void __eeh_mark_slot (struct device_node *dn, int mode_flag)
                if (PCI_DN(dn)) {
                        PCI_DN(dn)->eeh_mode |= mode_flag;
 
+                       /* Mark the pci device driver too */
+                       struct pci_dev *dev = PCI_DN(dn)->pcidev;
+                       if (dev && dev->driver)
+                               dev->error_state = pci_channel_io_frozen;
+
                        if (dn->child)
                                __eeh_mark_slot (dn->child, mode_flag);
                }
@@ -544,6 +549,7 @@ int eeh_dn_check_failure(struct device_node *dn, struct pci_dev *dev)
        int rets[3];
        unsigned long flags;
        struct pci_dn *pdn;
+       enum pci_channel_state state;
        int rc = 0;
 
        __get_cpu_var(total_mmio_ffs)++;
@@ -648,8 +654,13 @@ int eeh_dn_check_failure(struct device_node *dn, struct pci_dev *dev)
        eeh_mark_slot (dn, EEH_MODE_ISOLATED);
        spin_unlock_irqrestore(&confirm_error_lock, flags);
 
-       eeh_send_failure_event (dn, dev, rets[0], rets[2]);
-       
+       state = pci_channel_io_normal;
+       if ((rets[0] == 2) || (rets[0] == 4))
+               state = pci_channel_io_frozen;
+       if (rets[0] == 5)
+               state = pci_channel_io_perm_failure;
+       eeh_send_failure_event (dn, dev, state, rets[2]);
+
        /* Most EEH events are due to device driver bugs.  Having
         * a stack trace will help the device-driver authors figure
         * out what happened.  So print that out. */
@@ -953,8 +964,10 @@ static void *early_enable_eeh(struct device_node *dn, void *data)
         * But there are a few cases like display devices that make sense.
         */
        enable = 1;     /* i.e. we will do checking */
+#if 0
        if ((*class_code >> 16) == PCI_BASE_CLASS_DISPLAY)
                enable = 0;
+#endif
 
        if (!enable)
                pdn->eeh_mode |= EEH_MODE_NOCHECK;
diff --git a/arch/powerpc/platforms/pseries/eeh_driver.c b/arch/powerpc/platforms/pseries/eeh_driver.c
new file mode 100644 (file)
index 0000000..f6cff06
--- /dev/null
@@ -0,0 +1,366 @@
+/*
+ * PCI Error Recovery Driver for RPA-compliant PPC64 platform.
+ * Copyright (C) 2004, 2005 Linas Vepstas <linas@linas.org>
+ *
+ * All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or (at
+ * your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ * NON INFRINGEMENT.  See the GNU General Public License for more
+ * details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ * Send feedback to <linas@us.ibm.com>
+ *
+ */
+#include <linux/delay.h>
+#include <linux/irq.h>
+#include <linux/interrupt.h>
+#include <linux/notifier.h>
+#include <linux/pci.h>
+#include <asm/eeh.h>
+#include <asm/eeh_event.h>
+#include <asm/ppc-pci.h>
+#include <asm/pci-bridge.h>
+#include <asm/prom.h>
+#include <asm/rtas.h>
+
+
+static inline const char * pcid_name (struct pci_dev *pdev)
+{
+       if (pdev->dev.driver)
+               return pdev->dev.driver->name;
+       return "";
+}
+
+/**
+ * Return the "partitionable endpoint" (pe) under which this device lies
+ */
+static struct device_node * find_device_pe(struct device_node *dn)
+{
+       while ((dn->parent) && PCI_DN(dn->parent) &&
+             (PCI_DN(dn->parent)->eeh_mode & EEH_MODE_SUPPORTED)) {
+               dn = dn->parent;
+       }
+       return dn;
+}
+
+
+#ifdef DEBUG
+static void print_device_node_tree (struct pci_dn *pdn, int dent)
+{
+       int i;
+       if (!pdn) return;
+       for (i=0;i<dent; i++)
+               printk(" ");
+       printk("dn=%s mode=%x \tcfg_addr=%x pe_addr=%x \tfull=%s\n",
+               pdn->node->name, pdn->eeh_mode, pdn->eeh_config_addr,
+               pdn->eeh_pe_config_addr, pdn->node->full_name);
+       dent += 3;
+       struct device_node *pc = pdn->node->child;
+       while (pc) {
+               print_device_node_tree(PCI_DN(pc), dent);
+               pc = pc->sibling;
+       }
+}
+#endif
+
+/** 
+ * irq_in_use - return true if this irq is being used 
+ */
+static int irq_in_use(unsigned int irq)
+{
+       int rc = 0;
+       unsigned long flags;
+   struct irq_desc *desc = irq_desc + irq;
+
+       spin_lock_irqsave(&desc->lock, flags);
+       if (desc->action)
+               rc = 1;
+       spin_unlock_irqrestore(&desc->lock, flags);
+       return rc;
+}
+
+/* ------------------------------------------------------- */
+/** eeh_report_error - report an EEH error to each device,
+ *  collect up and merge the device responses.
+ */
+
+static void eeh_report_error(struct pci_dev *dev, void *userdata)
+{
+       enum pcierr_result rc, *res = userdata;
+       struct pci_driver *driver = dev->driver;
+
+       dev->error_state = pci_channel_io_frozen;
+
+       if (!driver)
+               return;
+
+       if (irq_in_use (dev->irq)) {
+               struct device_node *dn = pci_device_to_OF_node(dev);
+               PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED;
+               disable_irq_nosync(dev->irq);
+       }
+       if (!driver->err_handler)
+               return;
+       if (!driver->err_handler->error_detected)
+               return;
+
+       rc = driver->err_handler->error_detected (dev, pci_channel_io_frozen);
+       if (*res == PCIERR_RESULT_NONE) *res = rc;
+       if (*res == PCIERR_RESULT_NEED_RESET) return;
+       if (*res == PCIERR_RESULT_DISCONNECT &&
+            rc == PCIERR_RESULT_NEED_RESET) *res = rc;
+}
+
+/** eeh_report_reset -- tell this device that the pci slot
+ *  has been reset.
+ */
+
+static void eeh_report_reset(struct pci_dev *dev, void *userdata)
+{
+       struct pci_driver *driver = dev->driver;
+       struct device_node *dn = pci_device_to_OF_node(dev);
+
+       if (!driver)
+               return;
+
+       if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) {
+               PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED;
+               enable_irq(dev->irq);
+       }
+       if (!driver->err_handler)
+               return;
+       if (!driver->err_handler->slot_reset)
+               return;
+
+       driver->err_handler->slot_reset(dev);
+}
+
+static void eeh_report_resume(struct pci_dev *dev, void *userdata)
+{
+       struct pci_driver *driver = dev->driver;
+
+       dev->error_state = pci_channel_io_normal;
+
+       if (!driver)
+               return;
+       if (!driver->err_handler)
+               return;
+       if (!driver->err_handler->resume)
+               return;
+
+       driver->err_handler->resume(dev);
+}
+
+static void eeh_report_failure(struct pci_dev *dev, void *userdata)
+{
+       struct pci_driver *driver = dev->driver;
+
+       dev->error_state = pci_channel_io_perm_failure;
+
+       if (!driver)
+               return;
+
+       if (irq_in_use (dev->irq)) {
+               struct device_node *dn = pci_device_to_OF_node(dev);
+               PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED;
+               disable_irq_nosync(dev->irq);
+       }
+       if (!driver->err_handler)
+               return;
+       if (!driver->err_handler->error_detected)
+               return;
+       driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
+}
+
+/* ------------------------------------------------------- */
+/**
+ * handle_eeh_events -- reset a PCI device after hard lockup.
+ *
+ * pSeries systems will isolate a PCI slot if the PCI-Host
+ * bridge detects address or data parity errors, DMA's
+ * occuring to wild addresses (which usually happen due to
+ * bugs in device drivers or in PCI adapter firmware).
+ * Slot isolations also occur if #SERR, #PERR or other misc
+ * PCI-related errors are detected.
+ *
+ * Recovery process consists of unplugging the device driver
+ * (which generated hotplug events to userspace), then issuing
+ * a PCI #RST to the device, then reconfiguring the PCI config
+ * space for all bridges & devices under this slot, and then
+ * finally restarting the device drivers (which cause a second
+ * set of hotplug events to go out to userspace).
+ */
+
+/**
+ * eeh_reset_device() -- perform actual reset of a pci slot
+ * Args: bus: pointer to the pci bus structure corresponding
+ *            to the isolated slot. A non-null value will
+ *            cause all devices under the bus to be removed
+ *            and then re-added.
+ *     pe_dn: pointer to a "Partionable Endpoint" device node.
+ *            This is the top-level structure on which pci
+ *            bus resets can be performed.
+ */
+
+static void eeh_reset_device (struct pci_dn *pe_dn, struct pci_bus *bus)
+{
+       if (bus)
+               pcibios_remove_pci_devices(bus);
+
+       /* Reset the pci controller. (Asserts RST#; resets config space).
+        * Reconfigure bridges and devices */
+       rtas_set_slot_reset(pe_dn);
+
+       /* Walk over all functions on this device */
+       rtas_configure_bridge(pe_dn);
+       eeh_restore_bars(pe_dn);
+
+       /* Give the system 5 seconds to finish running the user-space
+        * hotplug shutdown scripts, e.g. ifdown for ethernet.  Yes, 
+        * this is a hack, but if we don't do this, and try to bring 
+        * the device up before the scripts have taken it down, 
+        * potentially weird things happen.
+        */
+       if (bus) {
+               ssleep (5);
+               pcibios_add_pci_devices(bus);
+       }
+}
+
+/* The longest amount of time to wait for a pci device
+ * to come back on line, in seconds.
+ */
+#define MAX_WAIT_FOR_RECOVERY 15
+
+void handle_eeh_events (struct eeh_event *event)
+{
+       struct device_node *frozen_dn;
+       struct pci_dn *frozen_pdn;
+       struct pci_bus *frozen_bus;
+       int perm_failure = 0;
+
+       frozen_dn = find_device_pe(event->dn);
+       frozen_bus = pcibios_find_pci_bus(frozen_dn);
+
+       if (!frozen_dn) {
+               printk(KERN_ERR "EEH: Error: Cannot find partition endpoint for %s\n",
+                       pci_name(event->dev));
+               return;
+       }
+
+       /* There are two different styles for coming up with the PE.
+        * In the old style, it was the highest EEH-capable device
+        * which was always an EADS pci bridge.  In the new style,
+        * there might not be any EADS bridges, and even when there are,
+        * the firmware marks them as "EEH incapable". So another
+        * two-step is needed to find the pci bus.. */
+       if (!frozen_bus)
+               frozen_bus = pcibios_find_pci_bus (frozen_dn->parent);
+
+       if (!frozen_bus) {
+               printk(KERN_ERR "EEH: Cannot find PCI bus for %s\n",
+                       frozen_dn->full_name);
+               return;
+       }
+
+#if 0
+       /* We may get "permanent failure" messages on empty slots.
+        * These are false alarms. Empty slots have no child dn. */
+       if ((event->state == pci_channel_io_perm_failure) && (frozen_device == NULL))
+               return;
+#endif
+
+       frozen_pdn = PCI_DN(frozen_dn);
+       frozen_pdn->eeh_freeze_count++;
+       
+       if (frozen_pdn->eeh_freeze_count > EEH_MAX_ALLOWED_FREEZES)
+               perm_failure = 1;
+
+       /* If the reset state is a '5' and the time to reset is 0 (infinity)
+        * or is more then 15 seconds, then mark this as a permanent failure.
+        */
+       if ((event->state == pci_channel_io_perm_failure) &&
+           ((event->time_unavail <= 0) ||
+            (event->time_unavail > MAX_WAIT_FOR_RECOVERY*1000)))
+       {
+               perm_failure = 1;
+       }
+
+       /* Log the error with the rtas logger. */
+       if (perm_failure) {
+               /*
+                * About 90% of all real-life EEH failures in the field
+                * are due to poorly seated PCI cards. Only 10% or so are
+                * due to actual, failed cards.
+                */
+               printk(KERN_ERR
+                  "EEH: PCI device %s - %s has failed %d times \n"
+                  "and has been permanently disabled.  Please try reseating\n"
+                  "this device or replacing it.\n",
+                       pci_name (frozen_pdn->pcidev), 
+                       pcid_name(frozen_pdn->pcidev), 
+                       frozen_pdn->eeh_freeze_count);
+
+               eeh_slot_error_detail(frozen_pdn, 2 /* Permanent Error */);
+
+               /* Notify all devices that they're about to go down. */
+               pci_walk_bus(frozen_bus, eeh_report_failure, 0);
+
+               /* Shut down the device drivers for good. */
+               pcibios_remove_pci_devices(frozen_bus);
+               return;
+       }
+
+       eeh_slot_error_detail(frozen_pdn, 1 /* Temporary Error */);
+       printk(KERN_WARNING
+          "EEH: This PCI device has failed %d times since last reboot: %s - %s\n",
+               frozen_pdn->eeh_freeze_count,
+               pci_name (frozen_pdn->pcidev), 
+               pcid_name(frozen_pdn->pcidev));
+
+       /* Walk the various device drivers attached to this slot through
+        * a reset sequence, giving each an opportunity to do what it needs
+        * to accomplish the reset.  Each child gets a report of the
+        * status ... if any child can't handle the reset, then the entire
+        * slot is dlpar removed and added.
+        */
+       enum pcierr_result result = PCIERR_RESULT_NONE;
+       pci_walk_bus(frozen_bus, eeh_report_error, &result);
+
+       /* If all device drivers were EEH-unaware, then shut
+        * down all of the device drivers, and hope they
+        * go down willingly, without panicing the system.
+        */
+       if (result == PCIERR_RESULT_NONE) {
+               eeh_reset_device(frozen_pdn, frozen_bus);
+       }
+
+       /* If any device called out for a reset, then reset the slot */
+       if (result == PCIERR_RESULT_NEED_RESET) {
+               eeh_reset_device(frozen_pdn, NULL);
+               pci_walk_bus(frozen_bus, eeh_report_reset, 0);
+       }
+
+       /* If all devices reported they can proceed, the re-enable PIO */
+       if (result == PCIERR_RESULT_CAN_RECOVER) {
+               /* XXX Not supported; we brute-force reset the device */
+               eeh_reset_device(frozen_pdn, NULL);
+               pci_walk_bus(frozen_bus, eeh_report_reset, 0);
+       }
+
+       /* Tell all device drivers that they can resume operations */
+       pci_walk_bus(frozen_bus, eeh_report_resume, 0);
+}
+
+/* ---------- end of file ---------- */
index 92497333c2b65af53a4f007e2b1a09c5954f0f4a..9a9961f27480d5736981c5982716ee17fc7c307e 100644 (file)
@@ -21,6 +21,7 @@
 #include <linux/list.h>
 #include <linux/pci.h>
 #include <asm/eeh_event.h>
+#include <asm/ppc-pci.h>
 
 /** Overview:
  *  EEH error states may be detected within exception handlers;
@@ -36,31 +37,6 @@ LIST_HEAD(eeh_eventlist);
 static void eeh_thread_launcher(void *);
 DECLARE_WORK(eeh_event_wq, eeh_thread_launcher, NULL);
 
-/**
- * eeh_panic - call panic() for an eeh event that cannot be handled.
- * The philosophy of this routine is that it is better to panic and
- * halt the OS than it is to risk possible data corruption by
- * oblivious device drivers that don't know better.
- *
- * @dev pci device that had an eeh event
- * @reset_state current reset state of the device slot
- */
-static void eeh_panic(struct pci_dev *dev, int reset_state)
-{
-       /*
-        * Since the panic_on_oops sysctl is used to halt the system
-        * in light of potential corruption, we can use it here.
-        */
-       if (panic_on_oops) {
-               panic("EEH: MMIO failure (%d) on device:%s\n", reset_state,
-                     pci_name(dev));
-       }
-       else {
-               printk(KERN_INFO "EEH: Ignored MMIO failure (%d) on device:%s\n",
-                      reset_state, pci_name(dev));
-       }
-}
-
 /**
  * eeh_event_handler - dispatch EEH events.  The detection of a frozen
  * slot can occur inside an interrupt, where it can be hard to do
@@ -82,10 +58,16 @@ static int eeh_event_handler(void * dummy)
 
                spin_lock_irqsave(&eeh_eventlist_lock, flags);
                event = NULL;
+
+               /* Unqueue the event, get ready to process. */
                if (!list_empty(&eeh_eventlist)) {
                        event = list_entry(eeh_eventlist.next, struct eeh_event, list);
                        list_del(&event->list);
                }
+               
+               if (event)
+                       eeh_mark_slot(event->dn, EEH_MODE_RECOVERING);
+
                spin_unlock_irqrestore(&eeh_eventlist_lock, flags);
                if (event == NULL)
                        break;
@@ -93,8 +75,11 @@ static int eeh_event_handler(void * dummy)
                printk(KERN_INFO "EEH: Detected PCI bus error on device %s\n",
                       pci_name(event->dev));
 
-               eeh_panic (event->dev, event->state);
+               handle_eeh_events(event);
+
+               eeh_clear_slot(event->dn, EEH_MODE_RECOVERING);
 
+               pci_dev_put(event->dev);
                kfree(event);
        }
 
@@ -122,7 +107,7 @@ static void eeh_thread_launcher(void *dummy)
  */
 int eeh_send_failure_event (struct device_node *dn,
                             struct pci_dev *dev,
-                            int state,
+                            enum pci_channel_state state,
                             int time_unavail)
 {
        unsigned long flags;
index 4395b7bc1ed462b839d21881e7ef3f1382988bcf..b263fb2fa6e4376312677746cc00a10911f676df 100644 (file)
@@ -34,9 +34,11 @@ struct device_node;
 extern int eeh_subsystem_enabled;
 
 /* Values for eeh_mode bits in device_node */
-#define EEH_MODE_SUPPORTED     (1<<0)
-#define EEH_MODE_NOCHECK       (1<<1)
-#define EEH_MODE_ISOLATED      (1<<2)
+#define EEH_MODE_SUPPORTED     (1<<0)
+#define EEH_MODE_NOCHECK       (1<<1)
+#define EEH_MODE_ISOLATED      (1<<2)
+#define EEH_MODE_RECOVERING    (1<<3)
+#define EEH_MODE_IRQ_DISABLED  (1<<4)
 
 /* Max number of EEH freezes allowed before we consider the device
  * to be permanently disabled. */
index 5e11a00b6fa01722031e647e610f03d00faab3b9..93d55a2bebfdc9e4acc660679d9a4391d7905f73 100644 (file)
@@ -30,7 +30,7 @@ struct eeh_event {
        struct list_head     list;
        struct device_node      *dn;   /* struct device node */
        struct pci_dev       *dev;  /* affected device */
-       int                  state;
+       enum pci_channel_state state; /* PCI bus state for the affected device */
        int time_unavail;    /* milliseconds until device might be available */
 };
 
@@ -47,8 +47,11 @@ struct eeh_event {
  */
 int eeh_send_failure_event (struct device_node *dn,
                             struct pci_dev *dev,
-                            int reset_state,
+                            enum pci_channel_state state,
                             int time_unavail);
 
+/* Main recovery function */
+void handle_eeh_events (struct eeh_event *);
+
 #endif /* __KERNEL__ */
 #endif /* ASM_PPC64_EEH_EVENT_H */
index bdef312900a1d23c2b98708cf34aa42733492f53..caf67a3bcb78d454fa37243f5fca2538dd7c19f2 100644 (file)
@@ -52,6 +52,15 @@ extern unsigned long pci_probe_only;
 
 /* ---- EEH internal-use-only related routines ---- */
 #ifdef CONFIG_EEH
+/**
+ * eeh_slot_error_detail -- record and EEH error condition to the log
+ * @severity: 1 if temporary, 2 if permanent failure.
+ *
+ * Obtains the the EEH error details from the RTAS subsystem,
+ * and then logs these details with the RTAS error log system.
+ */
+void eeh_slot_error_detail (struct pci_dn *pdn, int severity);
+
 /**
  * rtas_set_slot_reset -- unfreeze a frozen slot
  *