[PATCH v2 2/7] drm/amdgpu: Implement mmap as GEM object function

[Christian König]

Am 15.04.21 um 14:11 schrieb Thomas Zimmermann:
> Hi
>
> Am 15.04.21 um 13:38 schrieb Christian König:
>> Am 15.04.21 um 12:17 schrieb Thomas Zimmermann:
>>> Moving the driver-specific mmap code into a GEM object function allows
>>> for using DRM helpers for various mmap callbacks.
>>>
>>> This change resolves several inconsistencies between regular mmap and
>>> prime-based mmap. The vm_ops field in vma is now set for all mmap'ed
>>> areas. Previously it way only set for regular mmap calls, prime-based
>>> mmap used TTM's default vm_ops. The function amdgpu_verify_access() is
>>> no longer being called and therefore removed by this patch.
>>>
>>> As a side effect, amdgpu_ttm_vm_ops and amdgpu_ttm_fault() are now
>>> implemented in amdgpu's GEM code.
>>>
>>> v2:
>>>     * rename amdgpu_ttm_vm_ops and amdgpu_ttm_fault() to
>>>       amdgpu_gem_vm_ops and amdgpu_gem_fault() (Christian)
>>>     * the check for kfd_bo has meanwhile been removed
>>>
>>> Signed-off-by: Thomas Zimmermann <tzimmermann at suse.de>
>>
>> [SNIP]
>>> +static int amdgpu_gem_prime_mmap(struct drm_gem_object *obj, struct 
>>> vm_area_struct *vma)
>>> +{
>>
>> Mhm, just double checking this function is now a core GEM function 
>> and not prime specific?
>>
>> If yes maybe drop the _prime part.
>
> No problem, but other functions in the GEM callbacks are also named 
> _prime_. Probably needs a clean-up.

Yeah, some still needs to be renamed. But there shouldn't be any left in 
amdgpu_gem.c

>
>>
>>> +    struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
>>> +    struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
>>> +    unsigned long asize = amdgpu_bo_size(bo);
>>> +
>>> +    if (!vma->vm_file)
>>> +        return -ENODEV;
>>> +
>>> +    if (!adev)
>>> +        return -ENODEV;
>>> +
>>> +    /* Check for valid size. */
>>> +    if (asize < vma->vm_end - vma->vm_start)
>>> +        return -EINVAL;
>>
>> Shouldn't we have that check in the common code?
>
> It's at [1]. I didn't really bother about tidying up the checks 
> themselves. We can drop all these except for the usermm branch below. 
> Same for the radeon patch.

In this case I would just drop that from both amdgpu and radeon.

Thanks,
Christian.

>
> Best regards
> Thomas
>
> [1] 
> https://elixir.bootlin.com/linux/v5.12-rc7/source/drivers/gpu/drm/drm_gem.c#L1059
>
>>
>> Apart from that looks good to me.
>>
>> Christian.
>>
>>> +
>>> +    if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm) ||
>>> +        (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)) {
>>> +        return -EPERM;
>>> +    }
>>> +
>>> +    return drm_gem_ttm_mmap(obj, vma);
>>> +}
>>> +
>>>   static const struct drm_gem_object_funcs amdgpu_gem_object_funcs = {
>>>       .free = amdgpu_gem_object_free,
>>>       .open = amdgpu_gem_object_open,
>>> @@ -212,6 +266,8 @@ static const struct drm_gem_object_funcs 
>>> amdgpu_gem_object_funcs = {
>>>       .export = amdgpu_gem_prime_export,
>>>       .vmap = drm_gem_ttm_vmap,
>>>       .vunmap = drm_gem_ttm_vunmap,
>>> +    .mmap = amdgpu_gem_prime_mmap,
>>> +    .vm_ops = &amdgpu_gem_vm_ops,
>>>   };
>>>   /*
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>>> index 1485f33c3cc7..d4083c19402b 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>>> @@ -152,25 +152,6 @@ static void amdgpu_evict_flags(struct 
>>> ttm_buffer_object *bo,
>>>       *placement = abo->placement;
>>>   }
>>> -/**
>>> - * amdgpu_verify_access - Verify access for a mmap call
>>> - *
>>> - * @bo:    The buffer object to map
>>> - * @filp: The file pointer from the process performing the mmap
>>> - *
>>> - * This is called by ttm_bo_mmap() to verify whether a process
>>> - * has the right to mmap a BO to their process space.
>>> - */
>>> -static int amdgpu_verify_access(struct ttm_buffer_object *bo, 
>>> struct file *filp)
>>> -{
>>> -    struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);
>>> -
>>> -    if (amdgpu_ttm_tt_get_usermm(bo->ttm))
>>> -        return -EPERM;
>>> -    return drm_vma_node_verify_access(&abo->tbo.base.vma_node,
>>> -                      filp->private_data);
>>> -}
>>> -
>>>   /**
>>>    * amdgpu_ttm_map_buffer - Map memory into the GART windows
>>>    * @bo: buffer object to map
>>> @@ -1522,7 +1503,6 @@ static struct ttm_device_funcs 
>>> amdgpu_bo_driver = {
>>>       .eviction_valuable = amdgpu_ttm_bo_eviction_valuable,
>>>       .evict_flags = &amdgpu_evict_flags,
>>>       .move = &amdgpu_bo_move,
>>> -    .verify_access = &amdgpu_verify_access,
>>>       .delete_mem_notify = &amdgpu_bo_delete_mem_notify,
>>>       .release_notify = &amdgpu_bo_release_notify,
>>>       .io_mem_reserve = &amdgpu_ttm_io_mem_reserve,
>>> @@ -1897,50 +1877,6 @@ void 
>>> amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool 
>>> enable)
>>>       adev->mman.buffer_funcs_enabled = enable;
>>>   }
>>> -static vm_fault_t amdgpu_ttm_fault(struct vm_fault *vmf)
>>> -{
>>> -    struct ttm_buffer_object *bo = vmf->vma->vm_private_data;
>>> -    vm_fault_t ret;
>>> -
>>> -    ret = ttm_bo_vm_reserve(bo, vmf);
>>> -    if (ret)
>>> -        return ret;
>>> -
>>> -    ret = amdgpu_bo_fault_reserve_notify(bo);
>>> -    if (ret)
>>> -        goto unlock;
>>> -
>>> -    ret = ttm_bo_vm_fault_reserved(vmf, vmf->vma->vm_page_prot,
>>> -                       TTM_BO_VM_NUM_PREFAULT, 1);
>>> -    if (ret == VM_FAULT_RETRY && !(vmf->flags & 
>>> FAULT_FLAG_RETRY_NOWAIT))
>>> -        return ret;
>>> -
>>> -unlock:
>>> -    dma_resv_unlock(bo->base.resv);
>>> -    return ret;
>>> -}
>>> -
>>> -static const struct vm_operations_struct amdgpu_ttm_vm_ops = {
>>> -    .fault = amdgpu_ttm_fault,
>>> -    .open = ttm_bo_vm_open,
>>> -    .close = ttm_bo_vm_close,
>>> -    .access = ttm_bo_vm_access
>>> -};
>>> -
>>> -int amdgpu_mmap(struct file *filp, struct vm_area_struct *vma)
>>> -{
>>> -    struct drm_file *file_priv = filp->private_data;
>>> -    struct amdgpu_device *adev = drm_to_adev(file_priv->minor->dev);
>>> -    int r;
>>> -
>>> -    r = ttm_bo_mmap(filp, vma, &adev->mman.bdev);
>>> -    if (unlikely(r != 0))
>>> -        return r;
>>> -
>>> -    vma->vm_ops = &amdgpu_ttm_vm_ops;
>>> -    return 0;
>>> -}
>>> -
>>>   int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset,
>>>                  uint64_t dst_offset, uint32_t byte_count,
>>>                  struct dma_resv *resv,
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>>> index dec0db8b0b13..6e51faad7371 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
>>> @@ -146,7 +146,6 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo,
>>>               struct dma_resv *resv,
>>>               struct dma_fence **fence);
>>> -int amdgpu_mmap(struct file *filp, struct vm_area_struct *vma);
>>>   int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo);
>>>   int amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo);
>>>   uint64_t amdgpu_ttm_domain_start(struct amdgpu_device *adev, 
>>> uint32_t type);
>>
>