efi_pstore: Avoid deadlock in non-blocking paths
Seiji Aguchi [Fri, 11 Jan 2013 18:10:05 +0000 (18:10 +0000)]
[Issue]

There is a scenario which efi_pstore may hang up:

 - cpuA grabs efivars->lock
 - cpuB panics and calls smp_send_stop
 - smp_send_stop sends IRQ to cpuA
 - after 1 second, cpuB gives up on cpuA and sends an NMI instead
 - cpuA is now in an NMI handler while still holding efivars->lock
 - cpuB is deadlocked

This case may happen if a firmware has a bug and
cpuA is stuck talking with it.

[Solution]

This patch changes a spin_lock to a spin_trylock in non-blocking paths.
and if the spin_lock has already taken by another cpu,
it returns without accessing to a firmware to avoid the deadlock.

Signed-off-by: Seiji Aguchi <seiji.aguchi@hds.com>
Acked-by: Don Zickus <dzickus@redhat.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

drivers/firmware/efivars.c

index 7b1c374..ef5070d 100644 (file)
@@ -1209,7 +1209,16 @@ static int efi_pstore_write(enum pstore_type_id type,
        u64 storage_space, remaining_space, max_variable_size;
        efi_status_t status = EFI_NOT_FOUND;
 
-       spin_lock(&efivars->lock);
+       if (pstore_cannot_block_path(reason)) {
+               /*
+                * If the lock is taken by another cpu in non-blocking path,
+                * this driver returns without entering firmware to avoid
+                * hanging up.
+                */
+               if (!spin_trylock(&efivars->lock))
+                       return -EBUSY;
+       } else
+               spin_lock(&efivars->lock);
 
        /*
         * Check if there is a space enough to log.