Hi

Am 01.12.20 um 10:13 schrieb Christian König:
> Am 01.12.20 um 09:32 schrieb Thomas Zimmermann:
>> Hi
>>
>> Am 30.11.20 um 16:33 schrieb Christian König:
>>> Am 30.11.20 um 16:30 schrieb Daniel Vetter:
>>>> On Mon, Nov 30, 2020 at 01:04:26PM +0100, Thomas Zimmermann wrote:
>>>>> Mapping a GEM object's buffer into kernel address space prevents the
>>>>> buffer from being evicted from VRAM, which in turn may result in
>>>>> out-of-memory errors. It's therefore required to only vmap GEM BOs for
>>>>> short periods of time; unless the GEM implementation provides 
>>>>> additional
>>>>> guarantees.
>>>>>
>>>>> Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
>>>>> ---
>>>>>   drivers/gpu/drm/drm_prime.c |  6 ++++++
>>>>>   include/drm/drm_gem.h       | 16 ++++++++++++++++
>>>>>   2 files changed, 22 insertions(+)
>>>>>
>>>>> diff --git a/drivers/gpu/drm/drm_prime.c b/drivers/gpu/drm/drm_prime.c
>>>>> index 7db55fce35d8..9c9ece9833e0 100644
>>>>> --- a/drivers/gpu/drm/drm_prime.c
>>>>> +++ b/drivers/gpu/drm/drm_prime.c
>>>>> @@ -669,6 +669,12 @@ EXPORT_SYMBOL(drm_gem_unmap_dma_buf);
>>>>>    * callback. Calls into &drm_gem_object_funcs.vmap for device 
>>>>> specific handling.
>>>>>    * The kernel virtual address is returned in map.
>>>>>    *
>>>>> + * To prevent the GEM object from being relocated, callers must 
>>>>> hold the GEM
>>>>> + * object's reservation lock from when calling this function until 
>>>>> releasing the
>>>>> + * mapping. Holding onto a mapping and the associated reservation 
>>>>> lock for an
>>>>> + * unbound time may result in out-of-memory errors. Calls to 
>>>>> drm_gem_dmabuf_vmap()
>>>>> + * should therefore be accompanied by a call to 
>>>>> drm_gem_dmabuf_vunmap().
>>>>> + *
>>>>>    * Returns 0 on success or a negative errno code otherwise.
>>>> This is a dma-buf hook, which means just documenting the rules you'd 
>>>> like
>>>> to have here isn't enough. We need to roll this out at the dma-buf 
>>>> level,
>>>> and enforce it.
>>>>
>>>> Enforce it = assert_lock_held
>>>>
>>>> Roll out = review everyone. Because this goes through dma-buf it'll 
>>>> come
>>>> back through shmem helpers (and other helpers and other subsystems) 
>>>> back
>>>> to any driver using vmap for gpu buffers. This includes the media
>>>> subsystem, and the media subsystem definitely doesn't cope with just
>>>> temporarily mapping buffers. So there we need to pin them, which I 
>>>> think
>>>> means we'll need 2 version of dma_buf_vmap - one that's temporary and
>>>> requires we hold dma_resv lock, the other requires that the buffer is
>>>> pinned.
>>>
>>> OR start to proper use the dma_buf_pin/dma_buf_unpin functions which 
>>> I added to cover this use case as well.
>>
>> While I generally agree, here are some thoughts:
>>
>> I found all generic pin functions useless, because they don't allow 
>> for specifying where to pin. With fbdev emulation, this means that 
>> console buffers might never make it to VRAM for scanout. If anything, 
>> the policy should be that pin always pins in HW-accessible memory.
> 
> Yes, completely agree. The major missing part here are some kind of 
> usage flags what we want to do with the buffer.

Not sure, but wasn't that not wanted by someone? I don't really remember.

Given Daniel's reply about pin, it really feels like we have 
contradicting policies for this interface.

> 
>>
>> Pin has quite a bit of overhead (more locking, buffer movement), so it 
>> should be the second choice after regular vmap. To make both work 
>> together, pin probably relies on holding the reservation lock internally.
> 
> There is a dma_resv_assert_held() at the beginning of those two functions.
> 
>>
>> Therefore I think we still would want some additional helpers, such as:
>>
>>   pin_unlocked(), which acquires the resv lock, calls regular pin and 
>> then drops the resv lock. Same for unpin_unlocked()
>>
>>   vmap_pinned(), which enforces that the buffer has been pinned and 
>> then calls regalar vmap. Same for vunmap_pinned()
> 
> I would rather open code that in each driver, hiding the locking logic 
> in some midlayer is usually not a good idea.

These helpers are less about hiding logic and more about making drivers 
do the right thing. The idea behind pin_unlocked() is that it drops the 
reservation lock immediately before returning. I suspect that some 
driver might not do that with an open-coded version. And vmap_pinned() 
does check for the BO to be pinned. Regular vmap would assert for the 
reservation lock instead.

Best regards
Thomas

> 
> Regards,
> Christian.
> 
>>
>> A typical pattern with these functions would look like this.
>>
>>     drm_gem_object bo;
>>     dma_buf_map map;
>>
>>     init() {
>>         pin_unlocked(bo);
>>         vmap_pinned(bo, map);
>>     }
>>
>>     worker() {
>>         begin_cpu_access()
>>         // access bo via map
>>         end_cpu_access()
>>     }
>>
>>     fini() {
>>         vunmap_pinned(bo, map);
>>         unpin_unlocked(bo);
>>     }
>>
>>     init()
>>     while (...) {
>>         worker()
>>     }
>>     fini()
>>
>> Is that reasonable for media drivers?
>>
>> Best regards
>> Thomas
>>
>>
>>>
>>> Cheers,
>>> Christian.
>>>
>>>>
>>>> That's what I meant with that this approach here is very sprawling :-/
>>>> -Daniel
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer