drm/i915: Mark vm_free_page() as a potential sleeper agent

vm_free_page() may call down into set_pages_array_wb() (which itself
sleeps, on x86 at least) but only if on !llc and the caches overflow.
Since this is unlikely, we only rarely trigger the error in practice,
and so to make CI detection of this sleeping bug possible we want to
mark the common vm_free_page() as a potential sleep.

Suggested-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
References: https://bugs.freedesktop.org/show_bug.cgi?id=103638
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Matthew Auld <matthew.william.auld@gmail.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20171109213450.13875-1-chris@chris-wilson.co.uk

diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 2847a6b..1e40eeb 100644 (file)
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -454,6 +454,14 @@ static void vm_free_pages_release(struct i915_address_space *vm,
 
 static void vm_free_page(struct i915_address_space *vm, struct page *page)
 {
+       /*
+        * On !llc, we need to change the pages back to WB. We only do so
+        * in bulk, so we rarely need to change the page attributes here,
+        * but doing so requires a stop_machine() from deep inside arch/x86/mm.
+        * To make detection of the possible sleep more likely, use an
+        * unconditional might_sleep() for everybody.
+        */
+       might_sleep();
        if (!pagevec_add(&vm->free_pages, page))
                vm_free_pages_release(vm, false);
 }