4.14-stable review patch. If anyone has any objections, please let me know.
------------------
From: Jeremy Cline jcline@redhat.com
commit 889ad63d41eea20184b0483e7e585e5b20fb6cfe upstream.
"qxl_bo_unref" may sleep, but calling "qxl_release_map" causes "preempt_disable()" to be called and "preempt_enable()" isn't called until "qxl_release_unmap" is used. Move the call to "qxl_bo_unref" out from in between the two to avoid sleeping from an atomic context.
This issue can be demonstrated on a kernel with CONFIG_LOCKDEP=y by creating a VM using QXL, using a desktop environment using Xorg, then moving the cursor on or off a window.
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1571128 Fixes: 9428088c90b6 ("drm/qxl: reapply cursor after resetting primary") Cc: stable@vger.kernel.org Signed-off-by: Jeremy Cline jcline@redhat.com Link: http://patchwork.freedesktop.org/patch/msgid/20180601200532.13619-1-jcline@r... Signed-off-by: Gerd Hoffmann kraxel@redhat.com Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org
--- drivers/gpu/drm/qxl/qxl_display.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
--- a/drivers/gpu/drm/qxl/qxl_display.c +++ b/drivers/gpu/drm/qxl/qxl_display.c @@ -630,7 +630,7 @@ static void qxl_cursor_atomic_update(str struct qxl_cursor_cmd *cmd; struct qxl_cursor *cursor; struct drm_gem_object *obj; - struct qxl_bo *cursor_bo = NULL, *user_bo = NULL; + struct qxl_bo *cursor_bo = NULL, *user_bo = NULL, *old_cursor_bo = NULL; int ret; void *user_ptr; int size = 64*64*4; @@ -684,7 +684,7 @@ static void qxl_cursor_atomic_update(str cursor_bo, 0); cmd->type = QXL_CURSOR_SET;
- qxl_bo_unref(&qcrtc->cursor_bo); + old_cursor_bo = qcrtc->cursor_bo; qcrtc->cursor_bo = cursor_bo; cursor_bo = NULL; } else { @@ -704,6 +704,9 @@ static void qxl_cursor_atomic_update(str qxl_push_cursor_ring_release(qdev, release, QXL_CMD_CURSOR, false); qxl_release_fence_buffer_objects(release);
+ if (old_cursor_bo) + qxl_bo_unref(&old_cursor_bo); + qxl_bo_unref(&cursor_bo);
return;