[PATCH 3/8] drm/amdgpu: Implement mmap as GEM object function
Christian König
christian.koenig at amd.com
Wed Apr 7 11:25:55 UTC 2021
Hi,
Am 06.04.21 um 17:27 schrieb Felix Kuehling:
> Am 2021-04-06 um 6:38 a.m. schrieb Thomas Zimmermann:
>> Hi
>>
>> Am 06.04.21 um 11:35 schrieb Christian König:
>>> Am 06.04.21 um 11:08 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 check for kfd_bo has been taken
>>>> from amdgpu_verify_access(), which is not called any longer and has
>>>> been removed.
>>>>
>>>> As a side effect, amdgpu_ttm_vm_ops and amdgpu_ttm_fault() are now
>>>> implemented in amdgpu's GEM code.
>>>>
>>>> Signed-off-by: Thomas Zimmermann <tzimmermann at suse.de>
>>>> ---
>>>> drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c | 46 -------------
>>>> drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h | 2 -
>>>> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 4 +-
>>>> drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 64 +++++++++++++++++++
>>>> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 71
>>>> ---------------------
>>>> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 1 -
>>>> 6 files changed, 66 insertions(+), 122 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c
>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c
>>>> index e0c4f7c7f1b9..19c5ab08d9ec 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c
>>>> @@ -42,52 +42,6 @@
>>>> #include <linux/pci-p2pdma.h>
>>>> #include <linux/pm_runtime.h>
>>>> -/**
>>>> - * amdgpu_gem_prime_mmap - &drm_driver.gem_prime_mmap implementation
>>>> - * @obj: GEM BO
>>>> - * @vma: Virtual memory area
>>>> - *
>>>> - * Sets up a userspace mapping of the BO's memory in the given
>>>> - * virtual memory area.
>>>> - *
>>>> - * Returns:
>>>> - * 0 on success or a negative error code on failure.
>>>> - */
>>>> -int amdgpu_gem_prime_mmap(struct drm_gem_object *obj,
>>>> - struct vm_area_struct *vma)
>>>> -{
>>>> - struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
>>>> - struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
>>>> - unsigned asize = amdgpu_bo_size(bo);
>>>> - int ret;
>>>> -
>>>> - if (!vma->vm_file)
>>>> - return -ENODEV;
>>>> -
>>>> - if (adev == NULL)
>>>> - return -ENODEV;
>>>> -
>>>> - /* Check for valid size. */
>>>> - if (asize < vma->vm_end - vma->vm_start)
>>>> - return -EINVAL;
>>>> -
>>>> - if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm) ||
>>>> - (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)) {
>>>> - return -EPERM;
>>>> - }
>>>> - vma->vm_pgoff += amdgpu_bo_mmap_offset(bo) >> PAGE_SHIFT;
>>>> -
>>>> - /* prime mmap does not need to check access, so allow here */
>>>> - ret = drm_vma_node_allow(&obj->vma_node,
>>>> vma->vm_file->private_data);
>>>> - if (ret)
>>>> - return ret;
>>>> -
>>>> - ret = ttm_bo_mmap(vma->vm_file, vma, &adev->mman.bdev);
>>>> - drm_vma_node_revoke(&obj->vma_node, vma->vm_file->private_data);
>>>> -
>>>> - return ret;
>>>> -}
>>>> -
>>>> static int
>>>> __dma_resv_make_exclusive(struct dma_resv *obj)
>>>> {
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h
>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h
>>>> index 39b5b9616fd8..3e93b9b407a9 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.h
>>>> @@ -31,8 +31,6 @@ struct drm_gem_object
>>>> *amdgpu_gem_prime_import(struct drm_device *dev,
>>>> struct dma_buf *dma_buf);
>>>> bool amdgpu_dmabuf_is_xgmi_accessible(struct amdgpu_device *adev,
>>>> struct amdgpu_bo *bo);
>>>> -int amdgpu_gem_prime_mmap(struct drm_gem_object *obj,
>>>> - struct vm_area_struct *vma);
>>>> extern const struct dma_buf_ops amdgpu_dmabuf_ops;
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>>>> index 76f48f79c70b..e96d2758f4bb 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>>>> @@ -1656,7 +1656,7 @@ static const struct file_operations
>>>> amdgpu_driver_kms_fops = {
>>>> .flush = amdgpu_flush,
>>>> .release = drm_release,
>>>> .unlocked_ioctl = amdgpu_drm_ioctl,
>>>> - .mmap = amdgpu_mmap,
>>>> + .mmap = drm_gem_mmap,
>>>> .poll = drm_poll,
>>>> .read = drm_read,
>>>> #ifdef CONFIG_COMPAT
>>>> @@ -1719,7 +1719,7 @@ static const struct drm_driver
>>>> amdgpu_kms_driver = {
>>>> .prime_handle_to_fd = drm_gem_prime_handle_to_fd,
>>>> .prime_fd_to_handle = drm_gem_prime_fd_to_handle,
>>>> .gem_prime_import = amdgpu_gem_prime_import,
>>>> - .gem_prime_mmap = amdgpu_gem_prime_mmap,
>>>> + .gem_prime_mmap = drm_gem_prime_mmap,
>>>> .name = DRIVER_NAME,
>>>> .desc = DRIVER_DESC,
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>>>> index fb7171e5507c..fe93faad05f2 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
>>>> @@ -41,6 +41,36 @@
>>>> static const struct drm_gem_object_funcs amdgpu_gem_object_funcs;
>>>> +static vm_fault_t amdgpu_ttm_fault(struct vm_fault *vmf)
>>> Please name that function amdgpu_gem_fault or amdgpu_gem_object_fault
>>>
>>>> +{
>>>> + 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
>>>> +};
>>>> +
>>>> static void amdgpu_gem_object_free(struct drm_gem_object *gobj)
>>>> {
>>>> struct amdgpu_bo *robj = gem_to_amdgpu_bo(gobj);
>>>> @@ -201,6 +231,38 @@ static void amdgpu_gem_object_close(struct
>>>> drm_gem_object *obj,
>>>> ttm_eu_backoff_reservation(&ticket, &list);
>>>> }
>>>> +static int amdgpu_gem_prime_mmap(struct drm_gem_object *obj, struct
>>>> vm_area_struct *vma)
>>>> +{
>>>> + 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;
>>>> +
>>>> + /*
>>>> + * Don't verify access for KFD BOs. They don't have a GEM
>>>> + * object associated with them.
>>>> + */
>>>> + if (bo->kfd_bo)
>>>> + goto out;
>>> Who does the access verification now?
>> This is somewhat confusing.
>>
>> I took this check as-is, including the comment, from amdgpu's
>> verify_access function. The verify_access function was called by
>> ttm_bo_mmap. It returned 0 and ttm_bo_mmap did the mapping.
> This is probably a left-over from when we mapped BOs using /dev/kfd. We
> changed this to use /dev/dri/renderD* a long time ago to fix CPU mapping
> invalidations on memory evictions. I think we can let GEM do the access
> check.
Ok, good to know.
Thomas can you remove the extra handling in a separate prerequisite patch?
If anybody then bisects to this patch we at least know what to do to get
it working again.
Regards,
Christian.
>
> Regards,
> Felix
>
>
>>> Christian.
>>>
>>>> +
>>>> + if (amdgpu_ttm_tt_get_usermm(bo->tbo.ttm) ||
>>>> + (bo->flags & AMDGPU_GEM_CREATE_NO_CPU_ACCESS)) {
>>>> + return -EPERM;
>>>> + }
>>>> +
>>>> +out:
>>>> + 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,
>>>> @@ -208,6 +270,8 @@ 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_ttm_vm_ops,
>>>> };
>>>> /*
>>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>>>> index 1c6131489a85..d9de91a517c6 100644
>>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
>>>> @@ -152,32 +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);
>>>> -
>>>> - /*
>>>> - * Don't verify access for KFD BOs. They don't have a GEM
>>>> - * object associated with them.
>>>> - */
>>>> - if (abo->kfd_bo)
>>>> - return 0;
>>>> -
>>>> - if (amdgpu_ttm_tt_get_usermm(bo->ttm))
>>>> - return -EPERM;
>>>> - return drm_vma_node_verify_access(&abo->tbo.base.vma_node,
>>>> - filp->private_data);
>>>> -}
>> Here's the orignal verification code. It gives a free pass to KFD.
>>
>>>> -
>>>> /**
>>>> * amdgpu_ttm_map_buffer - Map memory into the GART windows
>>>> * @bo: buffer object to map
>>>> @@ -1531,7 +1505,6 @@ 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,
>>>> @@ -1906,50 +1879,6 @@ 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;
>>>> -}
>> And this was the mmap callback in struct file_operations. It calls
>> ttm_bo_mmap(), which skips verification for KFD BOs. To the best of my
>> knowledge, there was no additional verification for these KFD BOs.
>>
>> The original code in amdgpu_gem_prime_mmap() did seom verification,
>> but didn't handle KFD specially. I guess, PRIME needs GEM and KFD BOs
>> wouldn't quailify.
>>
>> In the end I went with the semantics I found in amdgpu_mmap() and
>> handled KFD specially. Let me know if this requires to be changed.
>>
>> Best regards
>> Thomas
>>
>>>> -
>>>> 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);
>>> _______________________________________________
>>> dri-devel mailing list
>>> dri-devel at lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/dri-devel
More information about the amd-gfx
mailing list