drm/sched: Fix a use-after-free when tracing the scheduler.
Subsystems (3)
PATCH 1/1 drm/sched: Fix a use-after-free when tracing the scheduler.
With DEBUG_SLAB (poisoning on free) enabled, I could quickly produce an oops when tracing V3D.
Signed-off-by Eric Anholt <[email protected]>
---
I think this patch is correct (though maybe a bigger refactor could avoid the extra get/put?), but I've still got this with "vblank_mode=0 perf record -a -e v3d:.\* -e gpu_scheduler:.\* glxgears". Any ideas?
[ 139.842191] Unable to handle kernel NULL pointer dereference at virtual address 00000020 [ 139.850413] pgd = eab7bb57 [ 139.854424] [00000020] *pgd=80000040004003, *pmd=00000000 [ 139.860523] Internal error: Oops: 206 [#1] SMP ARM [ 139.865340] Modules linked in: [ 139.868404] CPU: 0 PID: 1161 Comm: v3d_render Not tainted 4.20.0-rc4+ #552 [ 139.875287] Hardware name: Broadcom STB (Flattened Device Tree) [ 139.881228] PC is at perf_trace_drm_sched_job_wait_dep+0xa8/0xf4 [ 139.887243] LR is at 0xe9790274 [ 139.890388] pc : [<c06e715c>] lr : [<e9790274>] psr: a0050013 [ 139.896662] sp : ed21dec0 ip : ed21dec0 fp : ed21df04 [ 139.901893] r10: ed267478 r9 : 00000000 r8 : ff7bde04 [ 139.907123] r7 : 00000000 r6 : 00000063 r5 : 00000000 r4 : c1208448 [ 139.913659] r3 : c1265690 r2 : ff7bf660 r1 : 00000034 r0 : ff7bf660 [ 139.920196] Flags: NzCv IRQs on FIQs on Mode SVC_32 ISA ARM Segment user [ 139.927339] Control: 30c5383d Table: 68fa3b40 DAC: fffffffd [ 139.933095] Process v3d_render (pid: 1161, stack limit = 0xb3c84b1b) [ 139.939457] Stack: (0xed21dec0 to 0xed21e000) [ 139.943821] dec0: 20050013 ffffffff eb9700cc 00000000 ec0e8e80 00000000 eb9700cc e9790274 [ 139.952009] dee0: 00000000 e2f59345 eb970078 eba8f680 c12ae00c c1208478 00000000 e8c2b048 [ 139.960197] df00: eb9700cc c06e92e4 c06e8f04 00000000 80050013 ed267478 eb970078 00000000 [ 139.968385] df20: ed267578 c0e45ae0 e9093080 c06e831c ed267630 c06e8120 c06e77d4 c1208448 [ 139.976573] df40: ee2e8acc 00000001 00000000 ee2e8640 c0272ab4 ed21df54 ed21df54 e2f59345 [ 139.984762] df60: ed21c000 ed1b4800 ed2d7840 00000000 ed21c000 ed267478 c06e8084 ee935cb0 [ 139.992950] df80: ed1b4838 c0249b44 ed21c000 ed2d7840 c02499e4 00000000 00000000 00000000 [ 140.001138] dfa0: 00000000 00000000 00000000 c02010ac 00000000 00000000 00000000 00000000 [ 140.009326] dfc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 [ 140.017514] dfe0: 00000000 00000000 00000000 00000000 00000013 00000000 00000000 00000000 [ 140.025707] [<c06e715c>] (perf_trace_drm_sched_job_wait_dep) from [<c06e92e4>] (drm_sched_entity_pop_job+0x394/0x438) [ 140.036332] [<c06e92e4>] (drm_sched_entity_pop_job) from [<c06e8120>] (drm_sched_main+0x9c/0x298) [ 140.045221] [<c06e8120>] (drm_sched_main) from [<c0249b44>] (kthread+0x160/0x168) [ 140.052716] [<c0249b44>] (kthread) from [<c02010ac>] (ret_from_fork+0x14/0x28)
drivers/gpu/drm/scheduler/sched_entity.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/gpu/drm/scheduler/sched_entity.c +3 -1
@@ -440,13 +440,15 @@ struct drm_sched_job *drm_sched_entity_pop_job(struct drm_sched_entity *entity)while ((entity->dependency = sched->ops->dependency (sched_job, entity))) { - +dma_fence_get (entity->dependency);if (drm_sched_entity_add_dependency_cb (entity)) {trace_drm_sched_job_wait_dep (sched_job, entity->dependency); +dma_fence_put (entity->dependency);return NULL ; } +dma_fence_put (entity->dependency); }/* skip jobs from entity that marked guilty */
Signature · 1 line
2.20.0.rc1
Am 03.12.18 um 21:14 schrieb Eric Anholt:
Quoted 2 lines
With
DEBUG_SLAB(poisoning on free) enabled, I could quickly produce an oops when tracing V3D.
Good catch, but the solution is a clear NAK.
drm_sched_entity_add_dependency_cb() can result in setting entity->dependency to NULL. That in turn can lead to a memory leak because we call the _put with a NULL fence.
Instead we should rather call trace_drm_sched_job_wait_dep() before even calling drm_sched_entity_add_dependency_cb(). This is also cleaner because we want to trace which dependencies the driver gave to the scheduler and not which we actually needed to add a callback to.
Regards,
Christian.
Quoted 65 lines
Signed-off-by Eric Anholt <[email protected]>---
I think this patch is correct (though maybe a bigger refactor could avoid the extra get/put?), but I've still got this with "vblank_mode=0 perf record -a -e v3d:.\* -e gpu_scheduler:.\* glxgears". Any ideas?
[ 139.842191] Unable to handle kernel NULL pointer dereference at virtual address 00000020 [ 139.850413] pgd = eab7bb57 [ 139.854424] [00000020] *pgd=80000040004003, *pmd=00000000 [ 139.860523] Internal error: Oops: 206 [#1] SMP ARM [ 139.865340] Modules linked in: [ 139.868404] CPU: 0 PID: 1161 Comm: v3d_render Not tainted 4.20.0-rc4+ #552 [ 139.875287] Hardware name: Broadcom STB (Flattened Device Tree) [ 139.881228] PC is at perf_trace_drm_sched_job_wait_dep+0xa8/0xf4 [ 139.887243] LR is at 0xe9790274 [ 139.890388] pc : [<c06e715c>] lr : [<e9790274>] psr: a0050013 [ 139.896662] sp : ed21dec0 ip : ed21dec0 fp : ed21df04 [ 139.901893] r10: ed267478 r9 : 00000000 r8 : ff7bde04 [ 139.907123] r7 : 00000000 r6 : 00000063 r5 : 00000000 r4 : c1208448 [ 139.913659] r3 : c1265690 r2 : ff7bf660 r1 : 00000034 r0 : ff7bf660 [ 139.920196] Flags: NzCv IRQs on FIQs on Mode SVC_32 ISA ARM Segment user [ 139.927339] Control: 30c5383d Table: 68fa3b40 DAC: fffffffd [ 139.933095] Process v3d_render (pid: 1161, stack limit = 0xb3c84b1b) [ 139.939457] Stack: (0xed21dec0 to 0xed21e000) [ 139.943821] dec0: 20050013 ffffffff eb9700cc 00000000 ec0e8e80 00000000 eb9700cc e9790274 [ 139.952009] dee0: 00000000 e2f59345 eb970078 eba8f680 c12ae00c c1208478 00000000 e8c2b048 [ 139.960197] df00: eb9700cc c06e92e4 c06e8f04 00000000 80050013 ed267478 eb970078 00000000 [ 139.968385] df20: ed267578 c0e45ae0 e9093080 c06e831c ed267630 c06e8120 c06e77d4 c1208448 [ 139.976573] df40: ee2e8acc 00000001 00000000 ee2e8640 c0272ab4 ed21df54 ed21df54 e2f59345 [ 139.984762] df60: ed21c000 ed1b4800 ed2d7840 00000000 ed21c000 ed267478 c06e8084 ee935cb0 [ 139.992950] df80: ed1b4838 c0249b44 ed21c000 ed2d7840 c02499e4 00000000 00000000 00000000 [ 140.001138] dfa0: 00000000 00000000 00000000 c02010ac 00000000 00000000 00000000 00000000 [ 140.009326] dfc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 [ 140.017514] dfe0: 00000000 00000000 00000000 00000000 00000013 00000000 00000000 00000000 [ 140.025707] [<c06e715c>] (perf_trace_drm_sched_job_wait_dep) from [<c06e92e4>] (drm_sched_entity_pop_job+0x394/0x438) [ 140.036332] [<c06e92e4>] (drm_sched_entity_pop_job) from [<c06e8120>] (drm_sched_main+0x9c/0x298) [ 140.045221] [<c06e8120>] (drm_sched_main) from [<c0249b44>] (kthread+0x160/0x168) [ 140.052716] [<c0249b44>] (kthread) from [<c02010ac>] (ret_from_fork+0x14/0x28)drivers/gpu/drm/scheduler/sched_entity.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)drivers/gpu/drm/scheduler/sched_entity.c +3
-1@@ -440,13 +440,15 @@ struct drm_sched_job *drm_sched_entity_pop_job(struct drm_sched_entity *entity)while ((entity->dependency = sched->ops->dependency (sched_job, entity))) { - +dma_fence_get (entity->dependency);if (drm_sched_entity_add_dependency_cb (entity)) {trace_drm_sched_job_wait_dep (sched_job, entity->dependency); +dma_fence_put (entity->dependency);return NULL ; } +dma_fence_put (entity->dependency); }/* skip jobs from entity that marked guilty */