bus/pci: support sigbus handler
authorJeff Guo <jia.guo@intel.com>
Mon, 15 Oct 2018 11:27:24 +0000 (19:27 +0800)
committerThomas Monjalon <thomas@monjalon.net>
Mon, 15 Oct 2018 20:17:26 +0000 (22:17 +0200)
This patch implements the ops for the PCI bus sigbus handler. It finds the
PCI device that is being hot-unplugged and calls the relevant ops of the
hot-unplug handler to handle the hot-unplug failure of the device.

Signed-off-by: Jeff Guo <jia.guo@intel.com>
Acked-by: Shaopeng He <shaopeng.he@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
drivers/bus/pci/pci_common.c

index be7cc1f..66b6330 100644 (file)
@@ -403,6 +403,36 @@ pci_find_device(const struct rte_device *start, rte_dev_cmp_t cmp,
        return NULL;
 }
 
+/*
+ * find the device which encounter the failure, by iterate over all device on
+ * PCI bus to check if the memory failure address is located in the range
+ * of the BARs of the device.
+ */
+static struct rte_pci_device *
+pci_find_device_by_addr(const void *failure_addr)
+{
+       struct rte_pci_device *pdev = NULL;
+       uint64_t check_point, start, end, len;
+       int i;
+
+       check_point = (uint64_t)(uintptr_t)failure_addr;
+
+       FOREACH_DEVICE_ON_PCIBUS(pdev) {
+               for (i = 0; i != RTE_DIM(pdev->mem_resource); i++) {
+                       start = (uint64_t)(uintptr_t)pdev->mem_resource[i].addr;
+                       len = pdev->mem_resource[i].len;
+                       end = start + len;
+                       if (check_point >= start && check_point < end) {
+                               RTE_LOG(DEBUG, EAL, "Failure address %16.16"
+                                       PRIx64" belongs to device %s!\n",
+                                       check_point, pdev->device.name);
+                               return pdev;
+                       }
+               }
+       }
+       return NULL;
+}
+
 static int
 pci_hot_unplug_handler(struct rte_device *dev)
 {
@@ -430,6 +460,29 @@ pci_hot_unplug_handler(struct rte_device *dev)
        return ret;
 }
 
+static int
+pci_sigbus_handler(const void *failure_addr)
+{
+       struct rte_pci_device *pdev = NULL;
+       int ret = 0;
+
+       pdev = pci_find_device_by_addr(failure_addr);
+       if (!pdev) {
+               /* It is a generic sigbus error, no bus would handle it. */
+               ret = 1;
+       } else {
+               /* The sigbus error is caused of hot-unplug. */
+               ret = pci_hot_unplug_handler(&pdev->device);
+               if (ret) {
+                       RTE_LOG(ERR, EAL,
+                               "Failed to handle hot-unplug for device %s",
+                               pdev->name);
+                       ret = -1;
+               }
+       }
+       return ret;
+}
+
 static int
 pci_plug(struct rte_device *dev)
 {
@@ -462,6 +515,7 @@ struct rte_pci_bus rte_pci_bus = {
                .get_iommu_class = rte_pci_get_iommu_class,
                .dev_iterate = rte_pci_dev_iterate,
                .hot_unplug_handler = pci_hot_unplug_handler,
+               .sigbus_handler = pci_sigbus_handler,
        },
        .device_list = TAILQ_HEAD_INITIALIZER(rte_pci_bus.device_list),
        .driver_list = TAILQ_HEAD_INITIALIZER(rte_pci_bus.driver_list),