drm/atomic: Fix double free in drm_atomic_state_default_clear
authorMaarten Lankhorst <maarten.lankhorst@linux.intel.com>
Tue, 31 Jan 2017 09:25:25 +0000 (10:25 +0100)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Tue, 31 Jan 2017 12:41:46 +0000 (13:41 +0100)
drm_atomic_helper_page_flip and drm_atomic_ioctl set their own events
in crtc_state->event. But when it's set the event is freed in 2 places.

Solve this by only freeing the event in the atomic ioctl when it
allocated its own event.

This has been broken twice. The first time when the code was introduced,
but only in the corner case when an event is allocated, but more crtc's
were included by atomic check and then failing. This can mostly
happen when you do an atomic modeset in i915 and the display clock is
changed, which forces all crtc's to be included to the state.

This has been broken worse by adding in-fences support, which caused
the double free to be done unconditionally.

[IGT] kms_rotation_crc: starting subtest primary-rotation-180
=============================================================================
BUG kmalloc-128 (Tainted: G     U         ): Object already free
-----------------------------------------------------------------------------

Disabling lock debugging due to kernel taint
INFO: Allocated in drm_atomic_helper_setup_commit+0x285/0x2f0 [drm_kms_helper] age=0 cpu=3 pid=1529
 ___slab_alloc+0x308/0x3b0
 __slab_alloc+0xd/0x20
 kmem_cache_alloc_trace+0x92/0x1c0
 drm_atomic_helper_setup_commit+0x285/0x2f0 [drm_kms_helper]
 intel_atomic_commit+0x35/0x4f0 [i915]
 drm_atomic_commit+0x46/0x50 [drm]
 drm_mode_atomic_ioctl+0x7d4/0xab0 [drm]
 drm_ioctl+0x2b3/0x490 [drm]
 do_vfs_ioctl+0x69c/0x700
 SyS_ioctl+0x4e/0x80
 entry_SYSCALL_64_fastpath+0x13/0x94
INFO: Freed in drm_event_cancel_free+0xa3/0xb0 [drm] age=0 cpu=3 pid=1529
 __slab_free+0x48/0x2e0
 kfree+0x159/0x1a0
 drm_event_cancel_free+0xa3/0xb0 [drm]
 drm_mode_atomic_ioctl+0x86d/0xab0 [drm]
 drm_ioctl+0x2b3/0x490 [drm]
 do_vfs_ioctl+0x69c/0x700
 SyS_ioctl+0x4e/0x80
 entry_SYSCALL_64_fastpath+0x13/0x94
INFO: Slab 0xffffde1f0997b080 objects=17 used=2 fp=0xffff92fb65ec2578 flags=0x200000000008101
INFO: Object 0xffff92fb65ec2578 @offset=1400 fp=0xffff92fb65ec2ae8

Redzone ffff92fb65ec2570: bb bb bb bb bb bb bb bb                          ........
Object ffff92fb65ec2578: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b  kkkkkkkkkkkkkkkk
Object ffff92fb65ec2588: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b  kkkkkkkkkkkkkkkk
Object ffff92fb65ec2598: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b  kkkkkkkkkkkkkkkk
Object ffff92fb65ec25a8: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b  kkkkkkkkkkkkkkkk
Object ffff92fb65ec25b8: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b  kkkkkkkkkkkkkkkk
Object ffff92fb65ec25c8: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b  kkkkkkkkkkkkkkkk
Object ffff92fb65ec25d8: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b  kkkkkkkkkkkkkkkk
Object ffff92fb65ec25e8: 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b a5  kkkkkkkkkkkkkkk.
Redzone ffff92fb65ec25f8: bb bb bb bb bb bb bb bb                          ........
Padding ffff92fb65ec2738: 5a 5a 5a 5a 5a 5a 5a 5a                          ZZZZZZZZ
CPU: 3 PID: 180 Comm: kworker/3:2 Tainted: G    BU          4.10.0-rc6-patser+ #5039
Hardware name:                  /NUC5PPYB, BIOS PYBSWCEL.86A.0031.2015.0601.1712 06/01/2015
Workqueue: events intel_atomic_helper_free_state [i915]
Call Trace:
 dump_stack+0x4d/0x6d
 print_trailer+0x20c/0x220
 free_debug_processing+0x1c6/0x330
 ? drm_atomic_state_default_clear+0xf7/0x1c0 [drm]
 __slab_free+0x48/0x2e0
 ? drm_atomic_state_default_clear+0xf7/0x1c0 [drm]
 kfree+0x159/0x1a0
 drm_atomic_state_default_clear+0xf7/0x1c0 [drm]
 ? drm_atomic_state_clear+0x30/0x30 [drm]
 intel_atomic_state_clear+0xd/0x20 [i915]
 drm_atomic_state_clear+0x1a/0x30 [drm]
 __drm_atomic_state_free+0x13/0x60 [drm]
 intel_atomic_helper_free_state+0x5d/0x70 [i915]
 process_one_work+0x260/0x4a0
 worker_thread+0x2d1/0x4f0
 kthread+0x127/0x130
 ? process_one_work+0x4a0/0x4a0
 ? kthread_stop+0x120/0x120
 ret_from_fork+0x29/0x40
FIX kmalloc-128: Object at 0xffff92fb65ec2578 not freed

Fixes: 3b24f7d67581 ("drm/atomic: Add struct drm_crtc_commit to track async updates")
Fixes: 9626014258a5 ("drm/fence: add in-fences support")
Cc: <stable@vger.kernel.org> # v4.8+
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Gustavo Padovan <gustavo.padovan@collabora.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1485854725-27640-1-git-send-email-maarten.lankhorst@linux.intel.com
drivers/gpu/drm/drm_atomic.c

index 50f5cf7b69d1dc55fd427efa61e7e8e5eeff80d0..fdfb1ec17e660efa0b1f2c1f7273fa8d6fd8567a 100644 (file)
@@ -2032,13 +2032,16 @@ static void complete_crtc_signaling(struct drm_device *dev,
        }
 
        for_each_crtc_in_state(state, crtc, crtc_state, i) {
+               struct drm_pending_vblank_event *event = crtc_state->event;
                /*
-                * TEST_ONLY and PAGE_FLIP_EVENT are mutually
-                * exclusive, if they weren't, this code should be
-                * called on success for TEST_ONLY too.
+                * Free the allocated event. drm_atomic_helper_setup_commit
+                * can allocate an event too, so only free it if it's ours
+                * to prevent a double free in drm_atomic_state_clear.
                 */
-               if (crtc_state->event)
-                       drm_event_cancel_free(dev, &crtc_state->event->base);
+               if (event && (event->base.fence || event->base.file_priv)) {
+                       drm_event_cancel_free(dev, &event->base);
+                       crtc_state->event = NULL;
+               }
        }
 
        if (!fence_state)