lkml.org 
[lkml]   [2021]   [Jun]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
SubjectRe: nouveau broken on Riva TNT2 in 5.13.0-rc4: NULL pointer dereference in nouveau_bo_sync_for_device
From
Date


Am 10.06.21 um 19:59 schrieb Christian König:
> Am 10.06.21 um 19:50 schrieb Ondrej Zary:
>> [SNIP]
>>> I can't see how this is called from the nouveau code, only
>>> possibility I
>>> see is that it is maybe called through the AGP code somehow.
>> Yes, you're right:
>> [   13.192663] Call Trace:
>> [   13.192678]  dump_stack+0x54/0x68
>> [   13.192690]  ttm_tt_init+0x11/0x8a [ttm]
>> [   13.192699]  ttm_agp_tt_create+0x39/0x51 [ttm]
>> [   13.192840]  nouveau_ttm_tt_create+0x17/0x22 [nouveau]
>> [   13.192856]  ttm_tt_create+0x78/0x8c [ttm]
>> [   13.192864]  ttm_bo_handle_move_mem+0x7d/0xca [ttm]
>> [   13.192873]  ttm_bo_validate+0x92/0xc8 [ttm]
>> [   13.192883]  ttm_bo_init_reserved+0x216/0x243 [ttm]
>> [   13.192892]  ttm_bo_init+0x45/0x65 [ttm]
>> [   13.193018]  ? nouveau_bo_del_io_reserve_lru+0x48/0x48 [nouveau]
>> [   13.193150]  nouveau_bo_init+0x8c/0x94 [nouveau]
>> [   13.193273]  ? nouveau_bo_del_io_reserve_lru+0x48/0x48 [nouveau]
>> [   13.193407]  nouveau_bo_new+0x44/0x57 [nouveau]
>> [   13.193537]  nouveau_channel_prep+0xa3/0x269 [nouveau]
>> [   13.193665]  nouveau_channel_new+0x3c/0x5f7 [nouveau]
>> [   13.193679]  ? slab_free_freelist_hook+0x3b/0xa7
>> [   13.193686]  ? kfree+0x9e/0x11a
>> [   13.193781]  ? nvif_object_sclass_put+0xd/0x16 [nouveau]
>> [   13.193908]  nouveau_drm_device_init+0x2e2/0x646 [nouveau]
>> [   13.193924]  ? pci_enable_device_flags+0x1e/0xac
>> [   13.194052]  nouveau_drm_probe+0xeb/0x188 [nouveau]
>> [   13.194182]  ? nouveau_drm_device_init+0x646/0x646 [nouveau]
>> [   13.194195]  pci_device_probe+0x89/0xe9
>> [   13.194205]  really_probe+0x127/0x2a7
>> [   13.194212]  driver_probe_device+0x5b/0x87
>> [   13.194219]  device_driver_attach+0x2e/0x41
>> [   13.194226]  __driver_attach+0x7c/0x83
>> [   13.194232]  bus_for_each_dev+0x4c/0x66
>> [   13.194238]  driver_attach+0x14/0x16
>> [   13.194244]  ? device_driver_attach+0x41/0x41
>> [   13.194251]  bus_add_driver+0xc5/0x16c
>> [   13.194258]  driver_register+0x87/0xb9
>> [   13.194265]  __pci_register_driver+0x38/0x3b
>> [   13.194271]  ? 0xf0c0d000
>> [   13.194362]  nouveau_drm_init+0x14c/0x1000 [nouveau]
>>
>> How is ttm_dma_tt->dma_address allocated?
>
> Mhm, I need to double check how AGP is supposed to work.
>
> Since barely anybody is using it these days it is something which
> breaks from time to time.

I have no idea how that ever worked in the first place since AGP isn't
supposed to sync between CPU/GPU. Everything is coherent for that case.

Anyway here is a patch which adds a check to those functions if the
dma_address array is allocated in the first place. Please test it.

Thanks,
Christian.

>
> Thanks for the backtrace,
> Christian.
>
>>   I cannot find any assignment
>> executed (in the working code):
>>
>> $ git grep dma_address\ = drivers/gpu/
>> drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c:
>> sg->sgl->dma_address = addr;
>> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c: dma_address =
>> &dma->dma_address[offset >> PAGE_SHIFT];
>> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c: dma_address =
>> (mm_node->start << PAGE_SHIFT) + offset;
>> drivers/gpu/drm/i915/gvt/scheduler.c:   sg->dma_address = addr;
>> drivers/gpu/drm/i915/i915_gpu_error.c:  sg->dma_address = it;
>> drivers/gpu/drm/ttm/ttm_tt.c:   ttm->dma_address = (void *)
>> (ttm->ttm.pages + ttm->ttm.num_pages);
>> drivers/gpu/drm/ttm/ttm_tt.c:   ttm->dma_address =
>> kvmalloc_array(ttm->ttm.num_pages,
>> drivers/gpu/drm/ttm/ttm_tt.c:   ttm_dma->dma_address = NULL;
>> drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c: viter->dma_address =
>> &__vmw_piter_phys_addr;
>> drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c: viter->dma_address =
>> &__vmw_piter_dma_addr;
>> drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c: viter->dma_address =
>> &__vmw_piter_sg_addr;
>>
>> The 2 cases in ttm_tt.c are in ttm_dma_tt_alloc_page_directory() and
>> ttm_sg_tt_alloc_page_directory().
>> Confirmed by adding printk()s that they're NOT called.
>>
>>
>

From 5370102729c6ecb280712c40b92ff7b9f58c6e1e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig@amd.com>
Date: Fri, 11 Jun 2021 14:34:50 +0200
Subject: [PATCH] drm/nouveau: check dma_address array for CPU/GPU sync
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

AGP for example doesn't have a dma_address array.

Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/nouveau/nouveau_bo.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index 085023624fb0..1a52590f5303 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -551,7 +551,7 @@ nouveau_bo_sync_for_device(struct nouveau_bo *nvbo)
struct ttm_tt *ttm_dma = (struct ttm_tt *)nvbo->bo.ttm;
int i, j;

- if (!ttm_dma)
+ if (!ttm_dma || !ttm_dma->dma_address)
return;
if (!ttm_dma->pages) {
NV_DEBUG(drm, "ttm_dma 0x%p: pages NULL\n", ttm_dma);
@@ -587,7 +587,7 @@ nouveau_bo_sync_for_cpu(struct nouveau_bo *nvbo)
struct ttm_tt *ttm_dma = (struct ttm_tt *)nvbo->bo.ttm;
int i, j;

- if (!ttm_dma)
+ if (!ttm_dma || !ttm_dma->dma_address)
return;
if (!ttm_dma->pages) {
NV_DEBUG(drm, "ttm_dma 0x%p: pages NULL\n", ttm_dma);
--
2.25.1
\
 
 \ /
  Last update: 2021-06-11 14:38    [W:0.197 / U:0.164 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site