drm/i915: fix reset handling in the throttle ioctl
authorDaniel Vetter <daniel.vetter@ffwll.ch>
Wed, 14 Nov 2012 16:14:06 +0000 (17:14 +0100)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Sun, 20 Jan 2013 12:11:15 +0000 (13:11 +0100)
While auditing the code I've noticed one place (the throttle ioctl)
which does not yet wait for the reset handler to complete and doesn't
properly decode the wedge state into -EAGAIN/-EIO. Fix this up by
calling the right helpers. This might explain the oddball "my
compositor just died in a successfull gpu reset" reports. Or maybe not, since
current mesa doesn't use this ioctl to throttle command submission.

The throttle ioctl doesn't take the struct_mutex, so to avoid busy-looping
with -EAGAIN while a reset is in process, check for errors first and wait
for the handler to complete if a reset is pending by calling
i915_gem_wait_for_error.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/i915_gem.c

index 04b2f92..c96a501 100644 (file)
@@ -3390,8 +3390,13 @@ i915_gem_ring_throttle(struct drm_device *dev, struct drm_file *file)
        u32 seqno = 0;
        int ret;
 
-       if (atomic_read(&dev_priv->gpu_error.wedged))
-               return -EIO;
+       ret = i915_gem_wait_for_error(&dev_priv->gpu_error);
+       if (ret)
+               return ret;
+
+       ret = i915_gem_check_wedge(&dev_priv->gpu_error, false);
+       if (ret)
+               return ret;
 
        spin_lock(&file_priv->mm.lock);
        list_for_each_entry(request, &file_priv->mm.request_list, client_list) {