OSDN Git Service

drm/scheduler: fix rare NULL ptr race
authorYintian Tao <yttao@amd.com>
Mon, 23 Mar 2020 11:19:37 +0000 (19:19 +0800)
committerAlex Deucher <alexander.deucher@amd.com>
Thu, 26 Mar 2020 14:22:36 +0000 (10:22 -0400)
There is one one corner case at dma_fence_signal_locked
which will raise the NULL pointer problem just like below.
->dma_fence_signal
    ->dma_fence_signal_locked
->test_and_set_bit
here trigger dma_fence_release happen due to the zero of fence refcount.

->dma_fence_put
    ->dma_fence_release
->drm_sched_fence_release_scheduled
    ->call_rcu
here make the union fled “cb_list” at finished fence
to NULL because struct rcu_head contains two pointer
which is same as struct list_head cb_list

Therefore, to hold the reference of finished fence at drm_sched_process_job
to prevent the null pointer during finished fence dma_fence_signal

[  732.912867] BUG: kernel NULL pointer dereference, address: 0000000000000008
[  732.914815] #PF: supervisor write access in kernel mode
[  732.915731] #PF: error_code(0x0002) - not-present page
[  732.916621] PGD 0 P4D 0
[  732.917072] Oops: 0002 [#1] SMP PTI
[  732.917682] CPU: 7 PID: 0 Comm: swapper/7 Tainted: G           OE     5.4.0-rc7 #1
[  732.918980] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.8.2-0-g33fbe13 by qemu-project.org 04/01/2014
[  732.920906] RIP: 0010:dma_fence_signal_locked+0x3e/0x100
[  732.938569] Call Trace:
[  732.939003]  <IRQ>
[  732.939364]  dma_fence_signal+0x29/0x50
[  732.940036]  drm_sched_fence_finished+0x12/0x20 [gpu_sched]
[  732.940996]  drm_sched_process_job+0x34/0xa0 [gpu_sched]
[  732.941910]  dma_fence_signal_locked+0x85/0x100
[  732.942692]  dma_fence_signal+0x29/0x50
[  732.943457]  amdgpu_fence_process+0x99/0x120 [amdgpu]
[  732.944393]  sdma_v4_0_process_trap_irq+0x81/0xa0 [amdgpu]

v2: hold the finished fence at drm_sched_process_job instead of
    amdgpu_fence_process
v3: resume the blank line

Signed-off-by: Yintian Tao <yttao@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/scheduler/sched_main.c

index 71ce621..60c4c6a 100644 (file)
@@ -661,7 +661,9 @@ static void drm_sched_process_job(struct dma_fence *f, struct dma_fence_cb *cb)
 
        trace_drm_sched_process_job(s_fence);
 
+       dma_fence_get(&s_fence->finished);
        drm_sched_fence_finished(s_fence);
+       dma_fence_put(&s_fence->finished);
        wake_up_interruptible(&sched->wake_up_worker);
 }