powerpc/eeh: Permanently disable the removed device
authorGanesh Goudar <ganeshgr@linux.ibm.com>
Mon, 22 Apr 2024 07:57:37 +0000 (13:27 +0530)
committerMichael Ellerman <mpe@ellerman.id.au>
Mon, 29 Apr 2024 13:51:15 +0000 (23:51 +1000)
When a device is hot removed on powernv, the hotplug driver clears
the device's state. However, on pseries, if a device is removed by
phyp after reaching the error threshold, the kernel remains unaware,
leading to the device not being torn down. This prevents necessary
remediation actions like failover.

Permanently disable the device if the presence check fails.

Also, in eeh_dev_check_failure in we may consider the error as false
positive if the device is hotpluged out as the get_state call returns
EEH_STATE_NOT_SUPPORT and we may end up not clearing the device state,
so log the event if the state is not moved to permanent failure state.

Signed-off-by: Ganesh Goudar <ganeshgr@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://msgid.link/20240422075737.1405551-1-ganeshgr@linux.ibm.com
arch/powerpc/kernel/eeh.c
arch/powerpc/kernel/eeh_driver.c

index ab316e155ea9f5f16b05cb9c6bab16c65057d317..6670063a7a6c7d467ee43bea4ffd2df6427f7a36 100644 (file)
@@ -506,9 +506,18 @@ int eeh_dev_check_failure(struct eeh_dev *edev)
         * We will punt with the following conditions: Failure to get
         * PE's state, EEH not support and Permanently unavailable
         * state, PE is in good state.
+        *
+        * On the pSeries, after reaching the threshold, get_state might
+        * return EEH_STATE_NOT_SUPPORT. However, it's possible that the
+        * device state remains uncleared if the device is not marked
+        * pci_channel_io_perm_failure. Therefore, consider logging the
+        * event to let device removal happen.
+        *
         */
        if ((ret < 0) ||
-           (ret == EEH_STATE_NOT_SUPPORT) || eeh_state_active(ret)) {
+           (ret == EEH_STATE_NOT_SUPPORT &&
+            dev->error_state == pci_channel_io_perm_failure) ||
+           eeh_state_active(ret)) {
                eeh_stats.false_positives++;
                pe->false_positives++;
                rc = 0;
index 48773d2d9be3cb0a02080d49c7944c8fcb17438e..7efe04c68f0fe3fb1c3c13d97d58e79e47cf103b 100644 (file)
@@ -865,9 +865,18 @@ void eeh_handle_normal_event(struct eeh_pe *pe)
                                devices++;
 
        if (!devices) {
-               pr_debug("EEH: Frozen PHB#%x-PE#%x is empty!\n",
+               pr_warn("EEH: Frozen PHB#%x-PE#%x is empty!\n",
                        pe->phb->global_number, pe->addr);
-               goto out; /* nothing to recover */
+               /*
+                * The device is removed, tear down its state, on powernv
+                * hotplug driver would take care of it but not on pseries,
+                * permanently disable the card as it is hot removed.
+                *
+                * In the case of powernv, note that the removal of device
+                * is covered by pci rescan lock, so no problem even if hotplug
+                * driver attempts to remove the device.
+                */
+               goto recover_failed;
        }
 
        /* Log the event */