[PATCH] drm/radeon: Use kvmalloc for CS chunks

[Christian König]

Am 02.03.21 um 07:42 schrieb Chen Li:
> The number of chunks/chunks_array may be passed in
> by userspace and can be large.

I'm wondering if we shouldn't rather restrict the number of chunks.

> It has been observed to cause kcalloc failures from trinity fuzzy test:
>
> ```
>   WARNING: CPU: 0 PID: 5487 at mm/page_alloc.c:4385
>   __alloc_pages_nodemask+0x2d8/0x14d0
>
> ......
>
> Trace:
> __warn.part.4+0x11c/0x174
> __alloc_pages_nodemask+0x2d8/0x14d0
> warn_slowpath_null+0x84/0xb0
> __alloc_pages_nodemask+0x2d8/0x14d0
> __alloc_pages_nodemask+0x2d8/0x14d0
> alloc_pages_current+0xf0/0x1b0
> free_buffer_head+0x88/0xf0
> jbd2_journal_try_to_free_buffers+0x1e0/0x2a0
> ext4_releasepage+0x84/0x140
> release_pages+0x414/0x4c0
> release_pages+0x42c/0x4c0
> __find_get_block+0x1a4/0x5b0
> alloc_pages_current+0xcc/0x1b0
> kmalloc_order+0x30/0xb0
> __kmalloc+0x300/0x390
> kmalloc_order_trace+0x48/0x110
> __kmalloc+0x300/0x390
> radeon_cs_parser_init.part.1+0x74/0x670 [radeon]
> crypto_shash_update+0x5c/0x1c0
> radeon_cs_parser_init.part.1+0x74/0x670 [radeon]
> __wake_up_common_lock+0xb8/0x210
> radeon_cs_ioctl+0xc8/0xb80 [radeon]
> radeon_cs_ioctl+0x50/0xb80 [radeon]
> drm_ioctl_kernel+0xf4/0x160
> radeon_cs_ioctl+0x0/0xb80 [radeon]
> drm_ioctl_kernel+0xa0/0x160
> drm_ioctl+0x2dc/0x4f0
> radeon_drm_ioctl+0x80/0xf0 [radeon]
> new_sync_write+0x120/0x1c0
> timerqueue_add+0x88/0x140
> do_vfs_ioctl+0xe4/0x990
> ksys_ioctl+0xdc/0x110
> ksys_ioctl+0x78/0x110
> sys_ioctl+0x2c/0x50
> entSys+0xa0/0xc0

Please drop the backtrace, it doesn't add any value to the commit log.

> ```
>
> Obviously, the required order in this case is larger than MAX_ORDER.
> So, just use kvmalloc instead.
>
> Signed-off-by: Chen Li <chenli at uniontech.com>

Reviewed-by: Christian König <christian.koenig at amd.com>

The same patch should probably applied to amdgpu as well if we don't 
already use kvmalloc there as well.

Regards,
Christian.

> ---
>   drivers/gpu/drm/radeon/radeon_cs.c | 8 ++++----
>   1 file changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/radeon/radeon_cs.c b/drivers/gpu/drm/radeon/radeon_cs.c
> index 35e937d39b51..fb736ef9f9aa 100644
> --- a/drivers/gpu/drm/radeon/radeon_cs.c
> +++ b/drivers/gpu/drm/radeon/radeon_cs.c
> @@ -288,7 +288,7 @@ int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data)
>   	p->chunk_relocs = NULL;
>   	p->chunk_flags = NULL;
>   	p->chunk_const_ib = NULL;
> -	p->chunks_array = kcalloc(cs->num_chunks, sizeof(uint64_t), GFP_KERNEL);
> +	p->chunks_array = kvmalloc_array(cs->num_chunks, sizeof(uint64_t), GFP_KERNEL);
>   	if (p->chunks_array == NULL) {
>   		return -ENOMEM;
>   	}
> @@ -299,7 +299,7 @@ int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data)
>   	}
>   	p->cs_flags = 0;
>   	p->nchunks = cs->num_chunks;
> -	p->chunks = kcalloc(p->nchunks, sizeof(struct radeon_cs_chunk), GFP_KERNEL);
> +	p->chunks = kvmalloc_array(p->nchunks, sizeof(struct radeon_cs_chunk), GFP_KERNEL);
>   	if (p->chunks == NULL) {
>   		return -ENOMEM;
>   	}
> @@ -452,8 +452,8 @@ static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error, bo
>   	kvfree(parser->vm_bos);
>   	for (i = 0; i < parser->nchunks; i++)
>   		kvfree(parser->chunks[i].kdata);
> -	kfree(parser->chunks);
> -	kfree(parser->chunks_array);
> +	kvfree(parser->chunks);
> +	kvfree(parser->chunks_array);
>   	radeon_ib_free(parser->rdev, &parser->ib);
>   	radeon_ib_free(parser->rdev, &parser->const_ib);
>   }
> --
> 2.30.0
>
>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx