]> err.no Git - linux-2.6/commitdiff
[PATCH] ppc64: escape hatch for spinning interrupt deadlocks
authorLinas Vepstas <linas@linas.org>
Fri, 4 Nov 2005 00:49:31 +0000 (18:49 -0600)
committerPaul Mackerras <paulus@samba.org>
Thu, 10 Nov 2005 00:33:33 +0000 (11:33 +1100)
08-eeh-spin-counter.patch

One an EEH event is triggers, all further I/O to a device is blocked (until
reset).  Bad device drivers may end up spinning in their interrupt handlers,
trying to read an interrupt status register that will never change state.
This patch moves that spin counter to a per-device structure, and adds
some diagnostic prints to help locate the bad driver.

Signed-off-by: Linas Vepstas <linas@linas.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
arch/ppc64/kernel/eeh.c

index e7522f6da69d2bd03dc87318f3fa93328385921e..0c52c2de92e02b43d6dbee63c98878c77bb76bfa 100644 (file)
@@ -78,14 +78,12 @@ DECLARE_WORK(eeh_event_wq, eeh_event_handler, NULL);
 
 static struct notifier_block *eeh_notifier_chain;
 
-/*
- * If a device driver keeps reading an MMIO register in an interrupt
+/* If a device driver keeps reading an MMIO register in an interrupt
  * handler after a slot isolation event has occurred, we assume it
  * is broken and panic.  This sets the threshold for how many read
  * attempts we allow before panicking.
  */
-#define EEH_MAX_FAILS  1000
-static atomic_t eeh_fail_count;
+#define EEH_MAX_FAILS  100000
 
 /* RTAS tokens */
 static int ibm_set_eeh_option;
@@ -521,7 +519,6 @@ static void eeh_event_handler(void *dummy)
                       "%s\n", event->reset_state,
                       pci_name(event->dev));
 
-               atomic_set(&eeh_fail_count, 0);
                notifier_call_chain (&eeh_notifier_chain,
                                     EEH_NOTIFY_FREEZE, event);
 
@@ -657,12 +654,18 @@ int eeh_dn_check_failure(struct device_node *dn, struct pci_dev *dev)
        spin_lock_irqsave(&confirm_error_lock, flags);
        rc = 1;
        if (pdn->eeh_mode & EEH_MODE_ISOLATED) {
-               atomic_inc(&eeh_fail_count);
-               if (atomic_read(&eeh_fail_count) >= EEH_MAX_FAILS) {
+               pdn->eeh_check_count ++;
+               if (pdn->eeh_check_count >= EEH_MAX_FAILS) {
+                       printk (KERN_ERR "EEH: Device driver ignored %d bad reads, panicing\n",
+                               pdn->eeh_check_count);
+                       dump_stack();
+                       
                        /* re-read the slot reset state */
                        if (read_slot_reset_state(pdn, rets) != 0)
                                rets[0] = -1;   /* reset state unknown */
-                       eeh_panic(dev, rets[0]);
+
+                       /* If we are here, then we hit an infinite loop. Stop. */
+                       panic("EEH: MMIO halt (%d) on device:%s\n", rets[0], pci_name(dev));
                }
                goto dn_unlock;
        }
@@ -808,6 +811,8 @@ static void *early_enable_eeh(struct device_node *dn, void *data)
        struct pci_dn *pdn = PCI_DN(dn);
 
        pdn->eeh_mode = 0;
+       pdn->eeh_check_count = 0;
+       pdn->eeh_freeze_count = 0;
 
        if (status && strcmp(status, "ok") != 0)
                return NULL;    /* ignore devices with bad status */