powerpc/eeh: Replace pci_dn with eeh_dev for EEH aux components
authorGavin Shan <shangw@linux.vnet.ibm.com>
Mon, 27 Feb 2012 20:04:08 +0000 (20:04 +0000)
committerBenjamin Herrenschmidt <benh@kernel.crashing.org>
Fri, 9 Mar 2012 00:39:46 +0000 (11:39 +1100)
The original EEH implementation is heavily depending on struct pci_dn.
We have to put EEH related information to pci_dn. Actually, we could
split struct pci_dn so that the EEH sensitive information to form an
individual struct, then EEH looks more independent.

The patch replaces pci_dn with eeh_dev for EEH aux components like
event and driver. Also, the eeh_event struct has been adjusted for
a little bit since eeh_dev has linked the associated FDT (Flat Device
Tree) node and PCI device. It's not necessary for eeh_event struct to
trace FDT node and PCI device. We can just simply to trace eeh_dev in
eeh_event.

The patch also renames function pcid_name() to eeh_pcid_name(), which
should be missed in the previous patch where the EEH aux components
have been cleaned up.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
arch/powerpc/include/asm/eeh_event.h
arch/powerpc/platforms/pseries/eeh.c
arch/powerpc/platforms/pseries/eeh_driver.c
arch/powerpc/platforms/pseries/eeh_event.c

index 25ebf6a..c68b012 100644 (file)
  */
 struct eeh_event {
        struct list_head        list;   /* to form event queue  */
-       struct device_node      *dn;    /* struct device node   */
-       struct pci_dev          *dev;   /* affected device      */
+       struct eeh_dev          *edev;  /* EEH device           */
 };
 
-int eeh_send_failure_event(struct device_node *dn, struct pci_dev *dev);
-struct pci_dn *handle_eeh_events(struct eeh_event *);
+int eeh_send_failure_event(struct eeh_dev *edev);
+struct eeh_dev *handle_eeh_events(struct eeh_event *);
 
 #endif /* __KERNEL__ */
 #endif /* ASM_POWERPC_EEH_EVENT_H */
index aec10f6..9b1fd0c 100644 (file)
@@ -475,7 +475,7 @@ int eeh_dn_check_failure(struct device_node *dn, struct pci_dev *dev)
        eeh_mark_slot(dn, EEH_MODE_ISOLATED);
        raw_spin_unlock_irqrestore(&confirm_error_lock, flags);
 
-       eeh_send_failure_event(edev->dn, edev->pdev);
+       eeh_send_failure_event(edev);
 
        /* Most EEH events are due to device driver bugs.  Having
         * a stack trace will help the device-driver authors figure
index 3f25fab..baf92cd 100644 (file)
@@ -40,7 +40,7 @@
  * This routine is used to retrieve the name of PCI device driver
  * if that's valid.
  */
-static inline const char *pcid_name(struct pci_dev *pdev)
+static inline const char *eeh_pcid_name(struct pci_dev *pdev)
 {
        if (pdev && pdev->dev.driver)
                return pdev->dev.driver->name;
@@ -81,7 +81,7 @@ static void print_device_node_tree(struct pci_dn *pdn, int dent)
  */
 static void eeh_disable_irq(struct pci_dev *dev)
 {
-       struct device_node *dn = pci_device_to_OF_node(dev);
+       struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
 
        /* Don't disable MSI and MSI-X interrupts. They are
         * effectively disabled by the DMA Stopped state
@@ -93,7 +93,7 @@ static void eeh_disable_irq(struct pci_dev *dev)
        if (!irq_has_action(dev->irq))
                return;
 
-       PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED;
+       edev->mode |= EEH_MODE_IRQ_DISABLED;
        disable_irq_nosync(dev->irq);
 }
 
@@ -106,10 +106,10 @@ static void eeh_disable_irq(struct pci_dev *dev)
  */
 static void eeh_enable_irq(struct pci_dev *dev)
 {
-       struct device_node *dn = pci_device_to_OF_node(dev);
+       struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
 
-       if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) {
-               PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED;
+       if ((edev->mode) & EEH_MODE_IRQ_DISABLED) {
+               edev->mode &= ~EEH_MODE_IRQ_DISABLED;
                enable_irq(dev->irq);
        }
 }
@@ -270,20 +270,20 @@ static int eeh_report_failure(struct pci_dev *dev, void *userdata)
 
 /**
  * eeh_reset_device - Perform actual reset of a pci slot
- * @pe_dn: PE associated device node
+ * @edev: PE associated EEH device
  * @bus: PCI bus corresponding to the isolcated slot
  *
  * This routine must be called to do reset on the indicated PE.
  * During the reset, udev might be invoked because those affected
  * PCI devices will be removed and then added.
  */
-static int eeh_reset_device(struct pci_dn *pe_dn, struct pci_bus *bus)
+static int eeh_reset_device(struct eeh_dev *edev, struct pci_bus *bus)
 {
        struct device_node *dn;
        int cnt, rc;
 
        /* pcibios will clear the counter; save the value */
-       cnt = pe_dn->eeh_freeze_count;
+       cnt = edev->freeze_count;
 
        if (bus)
                pcibios_remove_pci_devices(bus);
@@ -292,21 +292,22 @@ static int eeh_reset_device(struct pci_dn *pe_dn, struct pci_bus *bus)
         * Reconfigure bridges and devices. Don't try to bring the system
         * up if the reset failed for some reason.
         */
-       rc = eeh_reset_pe(pe_dn);
+       rc = eeh_reset_pe(edev);
        if (rc)
                return rc;
 
        /* Walk over all functions on this device. */
-       dn = pe_dn->node;
-       if (!pcibios_find_pci_bus(dn) && PCI_DN(dn->parent))
+       dn = eeh_dev_to_of_node(edev);
+       if (!pcibios_find_pci_bus(dn) && of_node_to_eeh_dev(dn->parent))
                dn = dn->parent->child;
 
        while (dn) {
-               struct pci_dn *ppe = PCI_DN(dn);
+               struct eeh_dev *pedev = of_node_to_eeh_dev(dn);
+
                /* On Power4, always true because eeh_pe_config_addr=0 */
-               if (pe_dn->eeh_pe_config_addr == ppe->eeh_pe_config_addr) {
+               if (edev->pe_config_addr == pedev->pe_config_addr) {
                        eeh_ops->configure_bridge(dn);
-                       eeh_restore_bars(ppe);
+                       eeh_restore_bars(pedev);
                }
                dn = dn->sibling;
        }
@@ -321,7 +322,7 @@ static int eeh_reset_device(struct pci_dn *pe_dn, struct pci_bus *bus)
                ssleep(5);
                pcibios_add_pci_devices(bus);
        }
-       pe_dn->eeh_freeze_count = cnt;
+       edev->freeze_count = cnt;
 
        return 0;
 }
@@ -348,23 +349,22 @@ static int eeh_reset_device(struct pci_dn *pe_dn, struct pci_bus *bus)
  * drivers (which cause a second set of hotplug events to go out to
  * userspace).
  */
-struct pci_dn *handle_eeh_events(struct eeh_event *event)
+struct eeh_dev *handle_eeh_events(struct eeh_event *event)
 {
        struct device_node *frozen_dn;
-       struct pci_dn *frozen_pdn;
+       struct eeh_dev *frozen_edev;
        struct pci_bus *frozen_bus;
        int rc = 0;
        enum pci_ers_result result = PCI_ERS_RESULT_NONE;
        const char *location, *pci_str, *drv_str, *bus_pci_str, *bus_drv_str;
 
-       frozen_dn = eeh_find_device_pe(event->dn);
+       frozen_dn = eeh_find_device_pe(eeh_dev_to_of_node(event->edev));
        if (!frozen_dn) {
-
-               location = of_get_property(event->dn, "ibm,loc-code", NULL);
+               location = of_get_property(eeh_dev_to_of_node(event->edev), "ibm,loc-code", NULL);
                location = location ? location : "unknown";
                printk(KERN_ERR "EEH: Error: Cannot find partition endpoint "
                                "for location=%s pci addr=%s\n",
-                       location, eeh_pci_name(event->dev));
+                       location, eeh_pci_name(eeh_dev_to_pci_dev(event->edev)));
                return NULL;
        }
 
@@ -389,22 +389,21 @@ struct pci_dn *handle_eeh_events(struct eeh_event *event)
                return NULL;
        }
 
-       frozen_pdn = PCI_DN(frozen_dn);
-       frozen_pdn->eeh_freeze_count++;
+       frozen_edev = of_node_to_eeh_dev(frozen_dn);
+       frozen_edev->freeze_count++;
+       pci_str = eeh_pci_name(eeh_dev_to_pci_dev(event->edev));
+       drv_str = eeh_pcid_name(eeh_dev_to_pci_dev(event->edev));
 
-       pci_str = eeh_pci_name(event->dev);
-       drv_str = pcid_name(event->dev);
-       
-       if (frozen_pdn->eeh_freeze_count > EEH_MAX_ALLOWED_FREEZES)
+       if (frozen_edev->freeze_count > EEH_MAX_ALLOWED_FREEZES)
                goto excess_failures;
 
        printk(KERN_WARNING
           "EEH: This PCI device has failed %d times in the last hour:\n",
-               frozen_pdn->eeh_freeze_count);
+               frozen_edev->freeze_count);
 
-       if (frozen_pdn->pcidev) {
-               bus_pci_str = pci_name(frozen_pdn->pcidev);
-               bus_drv_str = pcid_name(frozen_pdn->pcidev);
+       if (frozen_edev->pdev) {
+               bus_pci_str = pci_name(frozen_edev->pdev);
+               bus_drv_str = eeh_pcid_name(frozen_edev->pdev);
                printk(KERN_WARNING
                        "EEH: Bus location=%s driver=%s pci addr=%s\n",
                        location, bus_drv_str, bus_pci_str);
@@ -425,7 +424,7 @@ struct pci_dn *handle_eeh_events(struct eeh_event *event)
        /* Get the current PCI slot state. This can take a long time,
         * sometimes over 3 seconds for certain systems.
         */
-       rc = eeh_ops->wait_state(frozen_pdn->node, MAX_WAIT_FOR_RECOVERY*1000);
+       rc = eeh_ops->wait_state(eeh_dev_to_of_node(frozen_edev), MAX_WAIT_FOR_RECOVERY*1000);
        if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) {
                printk(KERN_WARNING "EEH: Permanent failure\n");
                goto hard_fail;
@@ -435,14 +434,14 @@ struct pci_dn *handle_eeh_events(struct eeh_event *event)
         * don't post the error log until after all dev drivers
         * have been informed.
         */
-       eeh_slot_error_detail(frozen_pdn, EEH_LOG_TEMP);
+       eeh_slot_error_detail(frozen_edev, EEH_LOG_TEMP);
 
        /* 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 == PCI_ERS_RESULT_NONE) {
-               rc = eeh_reset_device(frozen_pdn, frozen_bus);
+               rc = eeh_reset_device(frozen_edev, frozen_bus);
                if (rc) {
                        printk(KERN_WARNING "EEH: Unable to reset, rc=%d\n", rc);
                        goto hard_fail;
@@ -451,7 +450,7 @@ struct pci_dn *handle_eeh_events(struct eeh_event *event)
 
        /* If all devices reported they can proceed, then re-enable MMIO */
        if (result == PCI_ERS_RESULT_CAN_RECOVER) {
-               rc = eeh_pci_enable(frozen_pdn, EEH_OPT_THAW_MMIO);
+               rc = eeh_pci_enable(frozen_edev, EEH_OPT_THAW_MMIO);
 
                if (rc < 0)
                        goto hard_fail;
@@ -465,7 +464,7 @@ struct pci_dn *handle_eeh_events(struct eeh_event *event)
 
        /* If all devices reported they can proceed, then re-enable DMA */
        if (result == PCI_ERS_RESULT_CAN_RECOVER) {
-               rc = eeh_pci_enable(frozen_pdn, EEH_OPT_THAW_DMA);
+               rc = eeh_pci_enable(frozen_edev, EEH_OPT_THAW_DMA);
 
                if (rc < 0)
                        goto hard_fail;
@@ -483,7 +482,7 @@ struct pci_dn *handle_eeh_events(struct eeh_event *event)
 
        /* If any device called out for a reset, then reset the slot */
        if (result == PCI_ERS_RESULT_NEED_RESET) {
-               rc = eeh_reset_device(frozen_pdn, NULL);
+               rc = eeh_reset_device(frozen_edev, NULL);
                if (rc) {
                        printk(KERN_WARNING "EEH: Cannot reset, rc=%d\n", rc);
                        goto hard_fail;
@@ -502,7 +501,7 @@ struct pci_dn *handle_eeh_events(struct eeh_event *event)
        /* Tell all device drivers that they can resume operations */
        pci_walk_bus(frozen_bus, eeh_report_resume, NULL);
 
-       return frozen_pdn;
+       return frozen_edev;
        
 excess_failures:
        /*
@@ -515,7 +514,7 @@ excess_failures:
                "has failed %d times in the last hour "
                "and has been permanently disabled.\n"
                "Please try reseating this device or replacing it.\n",
-               location, drv_str, pci_str, frozen_pdn->eeh_freeze_count);
+               location, drv_str, pci_str, frozen_edev->freeze_count);
        goto perm_error;
 
 hard_fail:
@@ -526,7 +525,7 @@ hard_fail:
                location, drv_str, pci_str);
 
 perm_error:
-       eeh_slot_error_detail(frozen_pdn, EEH_LOG_PERM);
+       eeh_slot_error_detail(frozen_edev, EEH_LOG_PERM);
 
        /* Notify all devices that they're about to go down. */
        pci_walk_bus(frozen_bus, eeh_report_failure, NULL);
index e98347c..4a47525 100644 (file)
@@ -56,8 +56,8 @@ DEFINE_MUTEX(eeh_event_mutex);
 static int eeh_event_handler(void * dummy)
 {
        unsigned long flags;
-       struct eeh_event        *event;
-       struct pci_dn *pdn;
+       struct eeh_event *event;
+       struct eeh_dev *edev;
 
        daemonize("eehd");
        set_current_state(TASK_INTERRUPTIBLE);
@@ -77,23 +77,26 @@ static int eeh_event_handler(void * dummy)
 
        /* Serialize processing of EEH events */
        mutex_lock(&eeh_event_mutex);
-       eeh_mark_slot(event->dn, EEH_MODE_RECOVERING);
+       edev = event->edev;
+       eeh_mark_slot(eeh_dev_to_of_node(edev), EEH_MODE_RECOVERING);
 
        printk(KERN_INFO "EEH: Detected PCI bus error on device %s\n",
-              eeh_pci_name(event->dev));
+              eeh_pci_name(edev->pdev));
+
+       edev = handle_eeh_events(event);
 
-       pdn = handle_eeh_events(event);
+       eeh_clear_slot(eeh_dev_to_of_node(edev), EEH_MODE_RECOVERING);
+       pci_dev_put(edev->pdev);
 
-       eeh_clear_slot(event->dn, EEH_MODE_RECOVERING);
-       pci_dev_put(event->dev);
        kfree(event);
        mutex_unlock(&eeh_event_mutex);
 
        /* If there are no new errors after an hour, clear the counter. */
-       if (pdn && pdn->eeh_freeze_count>0) {
+       if (edev && edev->freeze_count>0) {
                msleep_interruptible(3600*1000);
-               if (pdn->eeh_freeze_count>0)
-                       pdn->eeh_freeze_count--;
+               if (edev->freeze_count>0)
+                       edev->freeze_count--;
+
        }
 
        return 0;
@@ -114,17 +117,17 @@ static void eeh_thread_launcher(struct work_struct *dummy)
 
 /**
  * eeh_send_failure_event - Generate a PCI error event
- * @dev: pci device
+ * @edev: EEH device
  *
  * This routine can be called within an interrupt context;
  * the actual event will be delivered in a normal context
  * (from a workqueue).
  */
-int eeh_send_failure_event(struct device_node *dn,
-                            struct pci_dev *dev)
+int eeh_send_failure_event(struct eeh_dev *edev)
 {
        unsigned long flags;
        struct eeh_event *event;
+       struct device_node *dn = eeh_dev_to_of_node(edev);
        const char *location;
 
        if (!mem_init_done) {
@@ -140,11 +143,10 @@ int eeh_send_failure_event(struct device_node *dn,
                return 1;
        }
 
-       if (dev)
-               pci_dev_get(dev);
+       if (edev->pdev)
+               pci_dev_get(edev->pdev);
 
-       event->dn = dn;
-       event->dev = dev;
+       event->edev = edev;
 
        /* We may or may not be called in an interrupt context */
        spin_lock_irqsave(&eeh_eventlist_lock, flags);