drm/i915: Mark contents as dirty on a write fault
authorChris Wilson <chris@chris-wilson.co.uk>
Fri, 20 Sep 2019 12:18:21 +0000 (13:18 +0100)
committerChris Wilson <chris@chris-wilson.co.uk>
Tue, 24 Sep 2019 08:58:56 +0000 (09:58 +0100)
Since dropping the set-to-gtt-domain in commit a679f58d0510 ("drm/i915:
Flush pages on acquisition"), we no longer mark the contents as dirty on
a write fault. This has the issue of us then not marking the pages as
dirty on releasing the buffer, which means the contents are not written
out to the swap device (should we ever pick that buffer as a victim).
Notably, this is visible in the dumb buffer interface used for cursors.
Having updated the cursor contents via mmap, and swapped away, if the
shrinker should evict the old cursor, upon next reuse, the cursor would
be invisible.

E.g. echo 80 > /proc/sys/kernel/sysrq ; echo f > /proc/sysrq-trigger

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111541
Fixes: a679f58d0510 ("drm/i915: Flush pages on acquisition")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.william.auld@gmail.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: <stable@vger.kernel.org> # v5.2+
Reviewed-by: Matthew Auld <matthew.william.auld@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190920121821.7223-1-chris@chris-wilson.co.uk
drivers/gpu/drm/i915/gem/i915_gem_mman.c

index 1748e63156a24634ca7a9f2d79d10ddfa5f6bb49..860b751c51f19ccbd6e55cb58f2dfc0b2350b2a2 100644 (file)
@@ -319,7 +319,11 @@ vm_fault_t i915_gem_fault(struct vm_fault *vmf)
                intel_wakeref_auto(&i915->ggtt.userfault_wakeref,
                                   msecs_to_jiffies_timeout(CONFIG_DRM_I915_USERFAULT_AUTOSUSPEND));
 
-       i915_vma_set_ggtt_write(vma);
+       if (write) {
+               GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj));
+               i915_vma_set_ggtt_write(vma);
+               obj->mm.dirty = true;
+       }
 
 err_fence:
        i915_vma_unpin_fence(vma);