[perf] perf_fuzzer triggers NULL pointer derefernce in i915 driver

[perf] perf_fuzzer triggers NULL pointer derefernce in i915 driver

[Vince Weaver]

Hello

with current git the perf_fuzzer was able to trigger this NULL pointer
de-reference in the i915 driver.

Vince

[40119.349172] BUG: kernel NULL pointer dereference, address: 0000000000000000
[40119.356626] #PF: supervisor read access in kernel mode
[40119.362121] #PF: error_code(0x0000) - not-present page
[40119.367606] PGD 0 P4D 0 
[40119.370320] Oops: 0000 [#1] SMP PTI
[40119.374035] CPU: 0 PID: 654 Comm: Xorg Tainted: G        W         5.5.0-rc1+ #138
[40119.382149] Hardware name: LENOVO 10AM000AUS/SHARKBAY, BIOS FBKT72AUS 01/26/2014
[40119.390053] RIP: 0010:get_timeline_name+0x13/0x20 [i915]
[40119.395747] Code: 0f 1f 80 00 00 00 00 0f 1f 44 00 00 48 c7 c0 12 56 4e a0 c3 0f 1f 00 0f 1f 44 00 00 48 8b 87 d0 00 00 00 48 c7 c2 1c 56 4e a0 <48> 8b 00 48 85 c0 48 0f 44 c2 c3 66 90 0f 1f 44 00 00 55 48 8b 87
[40119.415853] RSP: 0018:ffffc90000903938 EFLAGS: 00010206
[40119.421479] RAX: 0000000000000000 RBX: ffff88811093af00 RCX: 0000000000000000
[40119.429148] RDX: ffffffffa04e561c RSI: ffff88811093af00 RDI: ffff88811093af00
[40119.436782] RBP: ffffc90000903990 R08: 0000000000000000 R09: ffff88811093af00
[40119.444420] R10: ffff88811371d208 R11: 0000000000000005 R12: 000000000000000a
[40119.452063] R13: 0000000000000881 R14: ffffffff822f1840 R15: 00000000000a0018
[40119.459671] FS:  00007f61e3f0ff00(0000) GS:ffff88811ac00000(0000) knlGS:0000000000000000
[40119.468347] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[40119.474482] CR2: 0000000000000000 CR3: 0000000111f76003 CR4: 00000000001606f0
[40119.482134] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000009000000
[40119.489841] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000600
[40119.497477] Call Trace:
[40119.500099]  perf_trace_dma_fence+0x80/0x230
[40119.504687]  dma_fence_init+0x9a/0xb0
[40119.508657]  dma_fence_work_init+0x40/0xa0 [i915]
[40119.513726]  i915_vma_work+0x34/0x50 [i915]
[40119.518256]  i915_vma_pin+0x2a4/0x6d0 [i915]
[40119.522884]  eb_lookup_vmas+0x205/0xb80 [i915]
[40119.527667]  i915_gem_do_execbuffer+0x676/0x19c0 [i915]
[40119.533290]  ? __alloc_pages_nodemask+0x16f/0x300
[40119.538331]  ? xas_store+0x56/0x5e0
[40119.542051]  ? mem_cgroup_charge_statistics+0x4c/0xd0
[40119.547436]  i915_gem_execbuffer2_ioctl+0x1df/0x3d0 [i915]
[40119.553348]  ? i915_gem_execbuffer_ioctl+0x2e0/0x2e0 [i915]
[40119.559359]  drm_ioctl_kernel+0xaa/0xf0 [drm]
[40119.564034]  drm_ioctl+0x1f7/0x390 [drm]
[40119.568271]  ? i915_gem_execbuffer_ioctl+0x2e0/0x2e0 [i915]
[40119.574280]  ? _raw_spin_unlock+0xa/0x10
[40119.578475]  ? __handle_mm_fault+0x8ca/0x15f0
[40119.583163]  do_vfs_ioctl+0x45f/0x6e0
[40119.587075]  ksys_ioctl+0x5e/0x90
[40119.590653]  __x64_sys_ioctl+0x16/0x20
[40119.594665]  do_syscall_64+0x52/0x1a0
[40119.598592]  entry_SYSCALL_64_after_hwframe+0x44/0xa9

Re: [perf] perf_fuzzer triggers NULL pointer derefernce in i915 driver

[Jani Nikula]

On Thu, 12 Dec 2019, Vince Weaver <vincent.weaver@maine.edu> wrote:
> with current git the perf_fuzzer was able to trigger this NULL pointer
> de-reference in the i915 driver.

Please file a bug.

https://gitlab.freedesktop.org/drm/intel/wikis/How-to-file-i915-bugs

BR,
Jani.


-- 
Jani Nikula, Intel Open Source Graphics Center

Re: [perf] perf_fuzzer triggers NULL pointer derefernce in i915 driver

[Dave Hansen]

On 12/12/19 11:09 PM, Jani Nikula wrote:
> On Thu, 12 Dec 2019, Vince Weaver <vincent.weaver@maine.edu> wrote:
>> with current git the perf_fuzzer was able to trigger this NULL pointer
>> de-reference in the i915 driver.
> Please file a bug.
> 
> https://gitlab.freedesktop.org/drm/intel/wikis/How-to-file-i915-bugs

I'm seeing the same thing.  It's annoyingly and immediately reproducible
for me:

	https://gitlab.freedesktop.org/drm/intel/issues/826

Let me know if you want anything fancier done like a bisect.  Looking
back through my kernel logs, it appears to also have happened with
5.4.0-rc4.

Re: [perf] perf_fuzzer triggers NULL pointer derefernce in i915 driver

[Vince Weaver]

On Fri, 13 Dec 2019, Dave Hansen wrote:

> On 12/12/19 11:09 PM, Jani Nikula wrote:
> > On Thu, 12 Dec 2019, Vince Weaver <vincent.weaver@maine.edu> wrote:
> >> with current git the perf_fuzzer was able to trigger this NULL pointer
> >> de-reference in the i915 driver.
> > Please file a bug.
> > 
> > https://gitlab.freedesktop.org/drm/intel/wikis/How-to-file-i915-bugs
> 
> I'm seeing the same thing.  It's annoyingly and immediately reproducible
> for me:
> 
> 	https://gitlab.freedesktop.org/drm/intel/issues/826
> 
> Let me know if you want anything fancier done like a bisect.  Looking
> back through my kernel logs, it appears to also have happened with
> 5.4.0-rc4.

This patch was sent out in response to my report (but not as a direct 
reply).


>From chris@chris-wilson.co.uk Thu Dec 12 10:42:36 2019
>Date: Thu, 12 Dec 2019 15:42:24 +0000
>From: Chris Wilson <chris@chris-wilson.co.uk>
>To: intel-gfx@lists.freedesktop.org
>Cc: Chris Wilson <chris@chris-wilson.co.uk>, Vince Weaver <vincent.weaver@maine.edu>, Matthew Auld <matthew.auld@intel.com>
>Subject: [PATCH] drm/i915: Set fence_work.ops before dma_fence_init

Since dma_fence_init may call ops (because of a meaningless
trace_dma_fence), we need to set the worker ops prior to that call.

Reported-by: Vince Weaver <vincent.weaver@maine.edu>
Fixes: 8e458fe2ee05 ("drm/i915: Generalise the clflush dma-worker")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
---
 drivers/gpu/drm/i915/i915_sw_fence_work.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_sw_fence_work.c b/drivers/gpu/drm/i915/i915_sw_fence_work.c
index 07552cd544f2..8538ee7a521d 100644
--- a/drivers/gpu/drm/i915/i915_sw_fence_work.c
+++ b/drivers/gpu/drm/i915/i915_sw_fence_work.c
@@ -78,12 +78,11 @@ static const struct dma_fence_ops fence_ops = {
 void dma_fence_work_init(struct dma_fence_work *f,
 			 const struct dma_fence_work_ops *ops)
 {
+	f->ops = ops;
 	spin_lock_init(&f->lock);
 	dma_fence_init(&f->dma, &fence_ops, &f->lock, 0, 0);
 	i915_sw_fence_init(&f->chain, fence_notify);
 	INIT_WORK(&f->work, fence_work);
-
-	f->ops = ops;
 }
 
 int dma_fence_work_chain(struct dma_fence_work *f, struct dma_fence *signal)
-- 
2.24.0