Re: [PATCH v3 1/3] accel/amdxdna: refuse an I/O memory mapping of an imported BO

[Lizhi Hou <lizhi.hou@amd.com>]

On 8/26/26 01:37, Christian König wrote:
> On 8/24/26 18:08, Lizhi Hou wrote:
>> On 8/24/26 07:06, Christian König wrote:
>>> On 8/17/26 19:53, Lizhi Hou wrote:
>>>> On 8/13/26 09:46, Taimuraz Kaitmazov wrote:
>>>>> amdxdna_gem_obj_vmap() takes whatever dma_buf_vmap() returns and only
>>>>> rejects a NULL vaddr. iosys_map is discriminated by is_iomem, so an
>>>>> exporter answering with an I/O mapping leaves a void __iomem pointer in
>>>>> abo->mem.kva, which amdxdna_cmd_set_error() memsets and memcpys through.
>>>>>
>>>>> amdxdna_drm_va_tbl takes a dmabuf_fd, so such a BO can be any exporter's
>>>>> buffer. amdgpu cannot reach this: its .pin forces GTT for a non peer to
>>>>> peer attachment like ours. An exporter on drm_gem_prime_dmabuf_ops has
>>>>> no .pin, and drm_gem_ttm_vmap() answers iomem for a VRAM resident
>>>>> object, so an NPU paired with nouveau or radeon does.
>>>>>
>>>>> Refuse the mapping. vmw_gem_vmap() does the same; unlike that one this
>>>>> path is reachable from an unprivileged ioctl, so it does not warn.
>>>>>
>>>>> Signed-off-by: Taimuraz Kaitmazov <taimuraz@kaitmazov.com>
>>>>> ---
>>>>>     drivers/accel/amdxdna/amdxdna_gem.c | 10 ++++++++--
>>>>>     1 file changed, 8 insertions(+), 2 deletions(-)
>>>>>
>>>>> diff --git a/drivers/accel/amdxdna/amdxdna_gem.c b/drivers/accel/amdxdna/amdxdna_gem.c
>>>>> index 1f190b319bb..b66ec9e4828 100644
>>>>> --- a/drivers/accel/amdxdna/amdxdna_gem.c
>>>>> +++ b/drivers/accel/amdxdna/amdxdna_gem.c
>>>>> @@ -683,10 +683,16 @@ static int amdxdna_gem_obj_vmap(struct drm_gem_object *obj, struct iosys_map *ma
>>>>>           dma_resv_assert_held(obj->resv);
>>>>>     -    if (is_import_bo(abo))
>>>>> +    if (is_import_bo(abo)) {
>>>>>             ret = dma_buf_vmap(abo->dma_buf, map);
>>> Mhm, why does amdxdna a vmap in the first place? For some workaround?
>> This is used for flushing the imported BO before. Based on our discussion before, the driver should not flush imported BO, so this becomes a invalid case.
>>
>> On the other hand, vmap on a io_mem should not happen. So I suggested to move the check to amdxdna_gem_vmap() for an extra check.
> Oh my, you should not use a vmap to flush CPU caches even if you own that GEM buffer! That is as buggy as it can be and most likely doesn't work 100% reliable.

We know that it would be problem on non-x86 architectures. And currently 
the device is X86 only. Is flushing through vmap buggy on x86 as well?

To make the flush running on all architectures,  I can remove the vmap 
completely here and only use drm_clflush_pages. Does this sound right?

>
> What is the complete command submission flow here?

So the submission flow is roughly:  alloc bo -> mmap BO -> write BO 
content (and flush CPU cache by user space runtime) -> submit/wait io 
command.


Thanks,

Lizhi

>
> Regards,
> Christian.
>
>>
>> Thanks,
>>
>> Lizhi
>>
>>> Usually DMA-buf only provides that framebuffer emulation scanout inside the kernel.
>>>
>>> On the other hand as far as I can see that here should work correctly.
>>>
>>> Regards,
>>> Christian.
>>>
>>>>> -    else
>>>>> +        /* Callers use mem.kva as an ordinary kernel address. */
>>>>> +        if (!ret && map->is_iomem) {
>>>>> +            dma_buf_vunmap(abo->dma_buf, map);
>>>>> +            return -EOPNOTSUPP;
>>>>> +        }
>>>> Thanks for the fix. The 'is_iomem' check should be moved to amdxdna_gem_vmap() to cover all the cases.
>>>>
>>>> Lizhi
>>>>
>>>>> +    } else {
>>>>>             ret = drm_gem_shmem_object_vmap(obj, map);
>>>>> +    }
>>>>>         if (ret)
>>>>>             return ret;
>>>>>         if (!map->vaddr)