drm/i915/gem: Retry faulthandlers on ENOSPC

As we no longer use the shmemfs allocation directly, we do not expect to
receive -ENOSPC from a backing store allocation. The potential sources
for -ENOSPC are then our own internal eviction code, so the choice is
either to kill the potential application with SIGBUS or to retry the
faulthandler.

In this patch we retry the fault handler, but since this is a should
never happen condition, it is arguable that we gather up copious debug
and kill the application. At worst, we cause an interruptible busy-wait,
stalling the application -- all causes should be transient and the
system should eventually recover. A small stall is hopefully a better
outcome than random oomkiller.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200515200031.12034-1-chris@chris-wilson.co.uk

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
index 70f5f82..9d306dc 100644 (file)
--- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
@@ -216,12 +216,12 @@ static vm_fault_t i915_error_to_vmf_fault(int err)
        case -ENXIO: /* unable to access backing store (on device) */
                return VM_FAULT_SIGBUS;
 
-       case -ENOSPC: /* shmemfs allocation failure */
        case -ENOMEM: /* our allocation failure */
                return VM_FAULT_OOM;
 
        case 0:
        case -EAGAIN:
+       case -ENOSPC: /* transient failure to evict? */
        case -ERESTARTSYS:
        case -EINTR:
        case -EBUSY: