Re: [PATCH 8/8] drm/amdgpu: add drm buddy support to amdgpu

["Christian König" <christian.koenig@amd.com>]

Am 04.11.21 um 09:49 schrieb Matthew Auld:
> On 04/11/2021 07:34, Christian König wrote:
>>
>>
>> Am 03.11.21 um 20:25 schrieb Matthew Auld:
>>> On 25/10/2021 14:00, Arunpravin wrote:
>>>> - Remove drm_mm references and replace with drm buddy functionalities
>>>> - Add res cursor support for drm buddy
>>>>
>>>> Signed-off-by: Arunpravin <Arunpravin.PaneerSelvam@amd.com>
>>>
>>> <snip>
>>>
>>>> +        spin_lock(&mgr->lock);
>>>> +        r = drm_buddy_alloc(mm, (uint64_t)place->fpfn << PAGE_SHIFT,
>>>> +                    (uint64_t)lpfn << PAGE_SHIFT,
>>>> +                    (uint64_t)n_pages << PAGE_SHIFT,
>>>> +                     min_page_size, &node->blocks,
>>>> +                     node->flags);
>>>
>>>
>>> Is spinlock + GFP_KERNEL allowed?
>>
>> Nope it isn't, but does that function really calls kmalloc()?
>
> It calls kmem_cache_zalloc(..., GFP_KERNEL)

Oh that's bad. In this case we either need a mutex here or some other 
approach to avoid the allocation.

Christian.

>
>>
>> Christian.
>>
>>>
>>>> + spin_unlock(&mgr->lock);
>>>> +
>>>> +        if (unlikely(r))
>>>> +            goto error_free_blocks;
>>>> +
>>>>           pages_left -= pages;
>>>>           ++i;
>>>>             if (pages > pages_left)
>>>>               pages = pages_left;
>>>>       }
>>>> -    spin_unlock(&mgr->lock);
>>>> +
>>>> +    /* Free unused pages for contiguous allocation */
>>>> +    if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
>>>> +        uint64_t actual_size = (uint64_t)node->base.num_pages << 
>>>> PAGE_SHIFT;
>>>> +
>>>> +        r = drm_buddy_free_unused_pages(mm,
>>>> +                        actual_size,
>>>> +                        &node->blocks);
>>>
>>> Needs some locking.
>>