Re: [PATCH v4 09/15] drm/shmem-helper: Correct doc-comment of drm_gem_shmem_get_sg_table()

Re: [PATCH v4 09/15] drm/shmem-helper: Correct doc-comment of drm_gem_shmem_get_sg_table()

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 1851 bytes --]

Hi

Am 18.04.22 um 00:37 schrieb Dmitry Osipenko:
> drm_gem_shmem_get_sg_table() never returns NULL on error, but a ERR_PTR.
> Correct the doc comment which says that it returns NULL on error.
> 
> Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>


> ---
>   drivers/gpu/drm/drm_gem_shmem_helper.c | 5 +++--
>   1 file changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_gem_shmem_helper.c b/drivers/gpu/drm/drm_gem_shmem_helper.c
> index 8ad0e02991ca..30ee46348a99 100644
> --- a/drivers/gpu/drm/drm_gem_shmem_helper.c
> +++ b/drivers/gpu/drm/drm_gem_shmem_helper.c
> @@ -662,7 +662,7 @@ EXPORT_SYMBOL(drm_gem_shmem_print_info);
>    * drm_gem_shmem_get_pages_sgt() instead.
>    *
>    * Returns:
> - * A pointer to the scatter/gather table of pinned pages or NULL on failure.
> + * A pointer to the scatter/gather table of pinned pages or errno on failure.

', or an ERR_PTR()-encoded errno code on failure'

>    */
>   struct sg_table *drm_gem_shmem_get_sg_table(struct drm_gem_shmem_object *shmem)
>   {
> @@ -688,7 +688,8 @@ EXPORT_SYMBOL_GPL(drm_gem_shmem_get_sg_table);
>    * drm_gem_shmem_get_sg_table() should not be directly called by drivers.
>    *
>    * Returns:
> - * A pointer to the scatter/gather table of pinned pages or errno on failure.
> + * A pointer to the scatter/gather table of pinned pages ERR_PTR()-encoded

', or an' before ERR_PTR

With the improved grammar:

Acked-by: Thomas Zimmermann <tzimmermann@suse.de>


> + * error code on failure.
>    */
>   struct sg_table *drm_gem_shmem_get_pages_sgt(struct drm_gem_shmem_object *shmem)
>   {

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

[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]

[-- Attachment #2: Type: text/plain, Size: 183 bytes --]

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 10/15] drm/shmem-helper: Take reservation lock instead of drm_gem_shmem locks

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 11246 bytes --]

Hi

Am 18.04.22 um 00:37 schrieb Dmitry Osipenko:
> Replace drm_gem_shmem locks with the reservation lock to make GEM
> lockings more consistent.
> 
> Previously drm_gem_shmem_vmap() and drm_gem_shmem_get_pages() were
> protected by separate locks, now it's the same lock, but it doesn't
> make any difference for the current GEM SHMEM users. Only Panfrost
> and Lima drivers use vmap() and they do it in the slow code paths,
> hence there was no practical justification for the usage of separate
> lock in the vmap().
> 
> Suggested-by: Daniel Vetter <daniel@ffwll.ch>
> Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
> ---
>   drivers/gpu/drm/drm_gem_shmem_helper.c  | 38 ++++++++++++-------------
>   drivers/gpu/drm/lima/lima_gem.c         |  8 +++---
>   drivers/gpu/drm/panfrost/panfrost_mmu.c | 15 ++++++----
>   include/drm/drm_gem_shmem_helper.h      | 10 -------
>   4 files changed, 31 insertions(+), 40 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_gem_shmem_helper.c b/drivers/gpu/drm/drm_gem_shmem_helper.c
> index 30ee46348a99..3ecef571eff3 100644
> --- a/drivers/gpu/drm/drm_gem_shmem_helper.c
> +++ b/drivers/gpu/drm/drm_gem_shmem_helper.c
> @@ -86,8 +86,6 @@ __drm_gem_shmem_create(struct drm_device *dev, size_t size, bool private)
>   	if (ret)
>   		goto err_release;
>   
> -	mutex_init(&shmem->pages_lock);
> -	mutex_init(&shmem->vmap_lock);
>   	INIT_LIST_HEAD(&shmem->madv_list);
>   
>   	if (!private) {
> @@ -157,8 +155,6 @@ void drm_gem_shmem_free(struct drm_gem_shmem_object *shmem)
>   	WARN_ON(shmem->pages_use_count);
>   
>   	drm_gem_object_release(obj);
> -	mutex_destroy(&shmem->pages_lock);
> -	mutex_destroy(&shmem->vmap_lock);
>   	kfree(shmem);
>   }
>   EXPORT_SYMBOL_GPL(drm_gem_shmem_free);
> @@ -209,11 +205,11 @@ int drm_gem_shmem_get_pages(struct drm_gem_shmem_object *shmem)
>   
>   	WARN_ON(shmem->base.import_attach);
>   
> -	ret = mutex_lock_interruptible(&shmem->pages_lock);
> +	ret = dma_resv_lock_interruptible(shmem->base.resv, NULL);
>   	if (ret)
>   		return ret;
>   	ret = drm_gem_shmem_get_pages_locked(shmem);
> -	mutex_unlock(&shmem->pages_lock);
> +	dma_resv_unlock(shmem->base.resv);
>   
>   	return ret;
>   }
> @@ -248,9 +244,9 @@ static void drm_gem_shmem_put_pages_locked(struct drm_gem_shmem_object *shmem)
>    */
>   void drm_gem_shmem_put_pages(struct drm_gem_shmem_object *shmem)
>   {
> -	mutex_lock(&shmem->pages_lock);
> +	dma_resv_lock(shmem->base.resv, NULL);
>   	drm_gem_shmem_put_pages_locked(shmem);
> -	mutex_unlock(&shmem->pages_lock);
> +	dma_resv_unlock(shmem->base.resv);
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_put_pages);
>   
> @@ -310,7 +306,7 @@ static int drm_gem_shmem_vmap_locked(struct drm_gem_shmem_object *shmem,
>   	} else {
>   		pgprot_t prot = PAGE_KERNEL;
>   
> -		ret = drm_gem_shmem_get_pages(shmem);
> +		ret = drm_gem_shmem_get_pages_locked(shmem);
>   		if (ret)
>   			goto err_zero_use;
>   
> @@ -360,11 +356,11 @@ int drm_gem_shmem_vmap(struct drm_gem_shmem_object *shmem,
>   {
>   	int ret;
>   
> -	ret = mutex_lock_interruptible(&shmem->vmap_lock);
> +	ret = dma_resv_lock_interruptible(shmem->base.resv, NULL);
>   	if (ret)
>   		return ret;
>   	ret = drm_gem_shmem_vmap_locked(shmem, map);

Within drm_gem_shmem_vmap_locked(), there's a call to dma_buf_vmap() for 
imported pages. If the exporter side also holds/acquires the same 
reservation lock as our object, the whole thing can deadlock. We cannot 
move dma_buf_vmap() out of the CS, because we still need to increment 
the reference counter. I honestly don't know how to easily fix this 
problem. There's a TODO item about replacing these locks at [1]. As 
Daniel suggested this patch, we should talk to him about the issue.

Best regards
Thomas

[1] 
https://www.kernel.org/doc/html/latest/gpu/todo.html#move-buffer-object-locking-to-dma-resv-lock



> -	mutex_unlock(&shmem->vmap_lock);
> +	dma_resv_unlock(shmem->base.resv);
>   
>   	return ret;
>   }
> @@ -385,7 +381,7 @@ static void drm_gem_shmem_vunmap_locked(struct drm_gem_shmem_object *shmem,
>   		dma_buf_vunmap(obj->import_attach->dmabuf, map);
>   	} else {
>   		vunmap(shmem->vaddr);
> -		drm_gem_shmem_put_pages(shmem);
> +		drm_gem_shmem_put_pages_locked(shmem);
>   	}
>   
>   	shmem->vaddr = NULL;
> @@ -406,9 +402,11 @@ static void drm_gem_shmem_vunmap_locked(struct drm_gem_shmem_object *shmem,
>   void drm_gem_shmem_vunmap(struct drm_gem_shmem_object *shmem,
>   			  struct iosys_map *map)
>   {
> -	mutex_lock(&shmem->vmap_lock);
> +	dma_resv_lock(shmem->base.resv, NULL);
>   	drm_gem_shmem_vunmap_locked(shmem, map);
> -	mutex_unlock(&shmem->vmap_lock);
> +	dma_resv_unlock(shmem->base.resv);
> +
> +	drm_gem_shmem_update_purgeable_status(shmem);
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_vunmap);
>   
> @@ -442,14 +440,14 @@ drm_gem_shmem_create_with_handle(struct drm_file *file_priv,
>    */
>   int drm_gem_shmem_madvise(struct drm_gem_shmem_object *shmem, int madv)
>   {
> -	mutex_lock(&shmem->pages_lock);
> +	dma_resv_lock(shmem->base.resv, NULL);
>   
>   	if (shmem->madv >= 0)
>   		shmem->madv = madv;
>   
>   	madv = shmem->madv;
>   
> -	mutex_unlock(&shmem->pages_lock);
> +	dma_resv_unlock(shmem->base.resv);
>   
>   	return (madv >= 0);
>   }
> @@ -487,10 +485,10 @@ EXPORT_SYMBOL(drm_gem_shmem_purge_locked);
>   
>   bool drm_gem_shmem_purge(struct drm_gem_shmem_object *shmem)
>   {
> -	if (!mutex_trylock(&shmem->pages_lock))
> +	if (!dma_resv_trylock(shmem->base.resv))
>   		return false;
>   	drm_gem_shmem_purge_locked(shmem);
> -	mutex_unlock(&shmem->pages_lock);
> +	dma_resv_unlock(shmem->base.resv);
>   
>   	return true;
>   }
> @@ -549,7 +547,7 @@ static vm_fault_t drm_gem_shmem_fault(struct vm_fault *vmf)
>   	/* We don't use vmf->pgoff since that has the fake offset */
>   	page_offset = (vmf->address - vma->vm_start) >> PAGE_SHIFT;
>   
> -	mutex_lock(&shmem->pages_lock);
> +	dma_resv_lock(shmem->base.resv, NULL);
>   
>   	if (page_offset >= num_pages ||
>   	    WARN_ON_ONCE(!shmem->pages) ||
> @@ -561,7 +559,7 @@ static vm_fault_t drm_gem_shmem_fault(struct vm_fault *vmf)
>   		ret = vmf_insert_pfn(vma, vmf->address, page_to_pfn(page));
>   	}
>   
> -	mutex_unlock(&shmem->pages_lock);
> +	dma_resv_unlock(shmem->base.resv);
>   
>   	return ret;
>   }
> diff --git a/drivers/gpu/drm/lima/lima_gem.c b/drivers/gpu/drm/lima/lima_gem.c
> index 0f1ca0b0db49..5008f0c2428f 100644
> --- a/drivers/gpu/drm/lima/lima_gem.c
> +++ b/drivers/gpu/drm/lima/lima_gem.c
> @@ -34,7 +34,7 @@ int lima_heap_alloc(struct lima_bo *bo, struct lima_vm *vm)
>   
>   	new_size = min(new_size, bo->base.base.size);
>   
> -	mutex_lock(&bo->base.pages_lock);
> +	dma_resv_lock(bo->base.base.resv, NULL);
>   
>   	if (bo->base.pages) {
>   		pages = bo->base.pages;
> @@ -42,7 +42,7 @@ int lima_heap_alloc(struct lima_bo *bo, struct lima_vm *vm)
>   		pages = kvmalloc_array(bo->base.base.size >> PAGE_SHIFT,
>   				       sizeof(*pages), GFP_KERNEL | __GFP_ZERO);
>   		if (!pages) {
> -			mutex_unlock(&bo->base.pages_lock);
> +			dma_resv_unlock(bo->base.base.resv);
>   			return -ENOMEM;
>   		}
>   
> @@ -56,13 +56,13 @@ int lima_heap_alloc(struct lima_bo *bo, struct lima_vm *vm)
>   		struct page *page = shmem_read_mapping_page(mapping, i);
>   
>   		if (IS_ERR(page)) {
> -			mutex_unlock(&bo->base.pages_lock);
> +			dma_resv_unlock(bo->base.base.resv);
>   			return PTR_ERR(page);
>   		}
>   		pages[i] = page;
>   	}
>   
> -	mutex_unlock(&bo->base.pages_lock);
> +	dma_resv_unlock(bo->base.base.resv);
>   
>   	ret = sg_alloc_table_from_pages(&sgt, pages, i, 0,
>   					new_size, GFP_KERNEL);
> diff --git a/drivers/gpu/drm/panfrost/panfrost_mmu.c b/drivers/gpu/drm/panfrost/panfrost_mmu.c
> index d3f82b26a631..404b8f67e2df 100644
> --- a/drivers/gpu/drm/panfrost/panfrost_mmu.c
> +++ b/drivers/gpu/drm/panfrost/panfrost_mmu.c
> @@ -424,6 +424,7 @@ static int panfrost_mmu_map_fault_addr(struct panfrost_device *pfdev, int as,
>   	struct panfrost_gem_mapping *bomapping;
>   	struct panfrost_gem_object *bo;
>   	struct address_space *mapping;
> +	struct drm_gem_object *obj;
>   	pgoff_t page_offset;
>   	struct sg_table *sgt;
>   	struct page **pages;
> @@ -446,13 +447,15 @@ static int panfrost_mmu_map_fault_addr(struct panfrost_device *pfdev, int as,
>   	page_offset = addr >> PAGE_SHIFT;
>   	page_offset -= bomapping->mmnode.start;
>   
> -	mutex_lock(&bo->base.pages_lock);
> +	obj = &bo->base.base;
> +
> +	dma_resv_lock(obj->resv, NULL);
>   
>   	if (!bo->base.pages) {
>   		bo->sgts = kvmalloc_array(bo->base.base.size / SZ_2M,
>   				     sizeof(struct sg_table), GFP_KERNEL | __GFP_ZERO);
>   		if (!bo->sgts) {
> -			mutex_unlock(&bo->base.pages_lock);
> +			dma_resv_unlock(obj->resv);
>   			ret = -ENOMEM;
>   			goto err_bo;
>   		}
> @@ -462,7 +465,7 @@ static int panfrost_mmu_map_fault_addr(struct panfrost_device *pfdev, int as,
>   		if (!pages) {
>   			kvfree(bo->sgts);
>   			bo->sgts = NULL;
> -			mutex_unlock(&bo->base.pages_lock);
> +			dma_resv_unlock(obj->resv);
>   			ret = -ENOMEM;
>   			goto err_bo;
>   		}
> @@ -472,7 +475,7 @@ static int panfrost_mmu_map_fault_addr(struct panfrost_device *pfdev, int as,
>   		pages = bo->base.pages;
>   		if (pages[page_offset]) {
>   			/* Pages are already mapped, bail out. */
> -			mutex_unlock(&bo->base.pages_lock);
> +			dma_resv_unlock(obj->resv);
>   			goto out;
>   		}
>   	}
> @@ -483,13 +486,13 @@ static int panfrost_mmu_map_fault_addr(struct panfrost_device *pfdev, int as,
>   	for (i = page_offset; i < page_offset + NUM_FAULT_PAGES; i++) {
>   		pages[i] = shmem_read_mapping_page(mapping, i);
>   		if (IS_ERR(pages[i])) {
> -			mutex_unlock(&bo->base.pages_lock);
> +			dma_resv_unlock(obj->resv);
>   			ret = PTR_ERR(pages[i]);
>   			goto err_pages;
>   		}
>   	}
>   
> -	mutex_unlock(&bo->base.pages_lock);
> +	dma_resv_unlock(obj->resv);
>   
>   	sgt = &bo->sgts[page_offset / (SZ_2M / PAGE_SIZE)];
>   	ret = sg_alloc_table_from_pages(sgt, pages + page_offset,
> diff --git a/include/drm/drm_gem_shmem_helper.h b/include/drm/drm_gem_shmem_helper.h
> index d0a57853c188..70889533962a 100644
> --- a/include/drm/drm_gem_shmem_helper.h
> +++ b/include/drm/drm_gem_shmem_helper.h
> @@ -26,11 +26,6 @@ struct drm_gem_shmem_object {
>   	 */
>   	struct drm_gem_object base;
>   
> -	/**
> -	 * @pages_lock: Protects the page table and use count
> -	 */
> -	struct mutex pages_lock;
> -
>   	/**
>   	 * @pages: Page table
>   	 */
> @@ -79,11 +74,6 @@ struct drm_gem_shmem_object {
>   	 */
>   	struct sg_table *sgt;
>   
> -	/**
> -	 * @vmap_lock: Protects the vmap address and use count
> -	 */
> -	struct mutex vmap_lock;
> -
>   	/**
>   	 * @vaddr: Kernel virtual address of the backing memory
>   	 */

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

[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]

[-- Attachment #2: Type: text/plain, Size: 183 bytes --]

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 10/15] drm/shmem-helper: Take reservation lock instead of drm_gem_shmem locks

[Daniel Vetter]

On Mon, Apr 18, 2022 at 10:18:54PM +0300, Dmitry Osipenko wrote:
> Hello,
> 
> On 4/18/22 21:38, Thomas Zimmermann wrote:
> > Hi
> > 
> > Am 18.04.22 um 00:37 schrieb Dmitry Osipenko:
> >> Replace drm_gem_shmem locks with the reservation lock to make GEM
> >> lockings more consistent.
> >>
> >> Previously drm_gem_shmem_vmap() and drm_gem_shmem_get_pages() were
> >> protected by separate locks, now it's the same lock, but it doesn't
> >> make any difference for the current GEM SHMEM users. Only Panfrost
> >> and Lima drivers use vmap() and they do it in the slow code paths,
> >> hence there was no practical justification for the usage of separate
> >> lock in the vmap().
> >>
> >> Suggested-by: Daniel Vetter <daniel@ffwll.ch>
> >> Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
> >> ---
> ...
> >>   @@ -310,7 +306,7 @@ static int drm_gem_shmem_vmap_locked(struct
> >> drm_gem_shmem_object *shmem,
> >>       } else {
> >>           pgprot_t prot = PAGE_KERNEL;
> >>   -        ret = drm_gem_shmem_get_pages(shmem);
> >> +        ret = drm_gem_shmem_get_pages_locked(shmem);
> >>           if (ret)
> >>               goto err_zero_use;
> >>   @@ -360,11 +356,11 @@ int drm_gem_shmem_vmap(struct
> >> drm_gem_shmem_object *shmem,
> >>   {
> >>       int ret;
> >>   -    ret = mutex_lock_interruptible(&shmem->vmap_lock);
> >> +    ret = dma_resv_lock_interruptible(shmem->base.resv, NULL);
> >>       if (ret)
> >>           return ret;
> >>       ret = drm_gem_shmem_vmap_locked(shmem, map);
> > 
> > Within drm_gem_shmem_vmap_locked(), there's a call to dma_buf_vmap() for
> > imported pages. If the exporter side also holds/acquires the same
> > reservation lock as our object, the whole thing can deadlock. We cannot
> > move dma_buf_vmap() out of the CS, because we still need to increment
> > the reference counter. I honestly don't know how to easily fix this
> > problem. There's a TODO item about replacing these locks at [1]. As
> > Daniel suggested this patch, we should talk to him about the issue.
> > 
> > Best regards
> > Thomas
> > 
> > [1]
> > https://www.kernel.org/doc/html/latest/gpu/todo.html#move-buffer-object-locking-to-dma-resv-lock
> 
> Indeed, good catch! Perhaps we could simply use a separate lock for the
> vmapping of the *imported* GEMs? The vmap_use_count is used only by
> vmap/vunmap, so it doesn't matter which lock is used by these functions
> in the case of imported GEMs since we only need to protect the
> vmap_use_count.

Apologies for the late reply, I'm flooded.

I discussed this with Daniel Stone last week in a chat, roughly what we
need to do is:

1. Pick a function from shmem helpers.

2. Go through all drivers that call this, and make sure that we acquire
dma_resv_lock in the top level driver entry point for this.

3. Once all driver code paths are converted, add a dma_resv_assert_held()
call to that function to make sure you have it all correctly.

4. Repeate 1-3 until all shmem helper functions are converted over.

5. Ditch the 3 different shmem helper locks.

The trouble is that I forgot that vmap is a thing, so that needs more
work. I think there's two approaches here:
- Do the vmap at import time. This is the trick we used to untangle the
  dma_resv_lock issues around dma_buf_attachment_map()
- Change the dma_buf_vmap rules that callers must hold the dma_resv_lock.
- Maybe also do what you suggest and keep a separate lock for this, but
  the fundamental issue is that this doesn't really work - if you share
  buffers both ways with two drivers using shmem helpers, then the
  ordering of this vmap_count_mutex vs dma_resv_lock is inconsistent and
  you can get some nice deadlocks. So not a great approach (and also the
  reason why we really need to get everyone to move towards dma_resv_lock
  as _the_ buffer object lock, since otherwise we'll never get a
  consistent lock nesting hierarchy).

The trouble here is that trying to be clever and doing the conversion just
in shmem helpers wont work, because there's a lot of cases where the
drivers are all kinds of inconsistent with their locking.

Adding Daniel S, also maybe for questions it'd be fastest to chat on irc?
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 10/15] drm/shmem-helper: Take reservation lock instead of drm_gem_shmem locks

[Daniel Vetter]

On Thu, Apr 28, 2022 at 09:31:00PM +0300, Dmitry Osipenko wrote:
> Hello Daniel,
> 
> 27.04.2022 17:50, Daniel Vetter пишет:
> > On Mon, Apr 18, 2022 at 10:18:54PM +0300, Dmitry Osipenko wrote:
> >> Hello,
> >>
> >> On 4/18/22 21:38, Thomas Zimmermann wrote:
> >>> Hi
> >>>
> >>> Am 18.04.22 um 00:37 schrieb Dmitry Osipenko:
> >>>> Replace drm_gem_shmem locks with the reservation lock to make GEM
> >>>> lockings more consistent.
> >>>>
> >>>> Previously drm_gem_shmem_vmap() and drm_gem_shmem_get_pages() were
> >>>> protected by separate locks, now it's the same lock, but it doesn't
> >>>> make any difference for the current GEM SHMEM users. Only Panfrost
> >>>> and Lima drivers use vmap() and they do it in the slow code paths,
> >>>> hence there was no practical justification for the usage of separate
> >>>> lock in the vmap().
> >>>>
> >>>> Suggested-by: Daniel Vetter <daniel@ffwll.ch>
> >>>> Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
> >>>> ---
> >> ...
> >>>>   @@ -310,7 +306,7 @@ static int drm_gem_shmem_vmap_locked(struct
> >>>> drm_gem_shmem_object *shmem,
> >>>>       } else {
> >>>>           pgprot_t prot = PAGE_KERNEL;
> >>>>   -        ret = drm_gem_shmem_get_pages(shmem);
> >>>> +        ret = drm_gem_shmem_get_pages_locked(shmem);
> >>>>           if (ret)
> >>>>               goto err_zero_use;
> >>>>   @@ -360,11 +356,11 @@ int drm_gem_shmem_vmap(struct
> >>>> drm_gem_shmem_object *shmem,
> >>>>   {
> >>>>       int ret;
> >>>>   -    ret = mutex_lock_interruptible(&shmem->vmap_lock);
> >>>> +    ret = dma_resv_lock_interruptible(shmem->base.resv, NULL);
> >>>>       if (ret)
> >>>>           return ret;
> >>>>       ret = drm_gem_shmem_vmap_locked(shmem, map);
> >>>
> >>> Within drm_gem_shmem_vmap_locked(), there's a call to dma_buf_vmap() for
> >>> imported pages. If the exporter side also holds/acquires the same
> >>> reservation lock as our object, the whole thing can deadlock. We cannot
> >>> move dma_buf_vmap() out of the CS, because we still need to increment
> >>> the reference counter. I honestly don't know how to easily fix this
> >>> problem. There's a TODO item about replacing these locks at [1]. As
> >>> Daniel suggested this patch, we should talk to him about the issue.
> >>>
> >>> Best regards
> >>> Thomas
> >>>
> >>> [1]
> >>> https://www.kernel.org/doc/html/latest/gpu/todo.html#move-buffer-object-locking-to-dma-resv-lock
> >>
> >> Indeed, good catch! Perhaps we could simply use a separate lock for the
> >> vmapping of the *imported* GEMs? The vmap_use_count is used only by
> >> vmap/vunmap, so it doesn't matter which lock is used by these functions
> >> in the case of imported GEMs since we only need to protect the
> >> vmap_use_count.
> > 
> > Apologies for the late reply, I'm flooded.
> > 
> > I discussed this with Daniel Stone last week in a chat, roughly what we
> > need to do is:
> > 
> > 1. Pick a function from shmem helpers.
> > 
> > 2. Go through all drivers that call this, and make sure that we acquire
> > dma_resv_lock in the top level driver entry point for this.
> > 
> > 3. Once all driver code paths are converted, add a dma_resv_assert_held()
> > call to that function to make sure you have it all correctly.
> > 4. Repeate 1-3 until all shmem helper functions are converted over.
> Somehow I didn't notice the existence of dma_resv_assert_held(), thank
> you for the suggestion :)
> 
> > 
> > 5. Ditch the 3 different shmem helper locks.
> > 
> > The trouble is that I forgot that vmap is a thing, so that needs more
> > work. I think there's two approaches here:
> > - Do the vmap at import time. This is the trick we used to untangle the
> >   dma_resv_lock issues around dma_buf_attachment_map()
> 
> > - Change the dma_buf_vmap rules that callers must hold the dma_resv_lock.
> 
> I'll consider this option for v6, thank you.
> 
> I see now that you actually want to define the new rules for the
> dma-bufs in general and not only in the context of the DRM code, this
> now makes much more sense to me.

Yeah dma-buf is a cross driver interface, so we should try to be
consistent here. We didn't do this in the past, where the only reason you
didn't get lockdep splats was because you normally didn't run all possible
combinations of drivers and importer/exporter relationships in one system.
But that means it becomes very tricky to reason about how dma-buf really
works.

> > - Maybe also do what you suggest and keep a separate lock for this, but
> >   the fundamental issue is that this doesn't really work - if you share
> >   buffers both ways with two drivers using shmem helpers, then the
> >   ordering of this vmap_count_mutex vs dma_resv_lock is inconsistent and
> >   you can get some nice deadlocks. So not a great approach (and also the
> >   reason why we really need to get everyone to move towards dma_resv_lock
> >   as _the_ buffer object lock, since otherwise we'll never get a
> >   consistent lock nesting hierarchy).
> 
> The separate locks should work okay because it will be always the
> exporter that takes the dma_resv_lock. But I agree that it's less ideal
> than defining the new rules for dma-bufs since sometime you will take
> the resv lock and sometime not, potentially hiding bugs related to lockings.

That's the issue, some importers need to take the dma_resv_lock for
dma_buf_vmap too (e.g. to first nail the buffer in place when it's a
dynamic memory manager). In practice it'll work as well as what we have
currently, which is similarly inconsistent, except with per-driver locks
instead of shared locks from shmem helpers or dma-buf, so less obvious
that things are inconsistent.

So yeah if it's too messy maybe the approach is to have a separate lock
for vmap for now, land things, and then fix up dma_buf_vmap in a follow up
series.
-Daniel

> > The trouble here is that trying to be clever and doing the conversion just
> > in shmem helpers wont work, because there's a lot of cases where the
> > drivers are all kinds of inconsistent with their locking.
> > 
> > Adding Daniel S, also maybe for questions it'd be fastest to chat on irc?
> 
> My nickname is digetx on the #dri-devel channel, feel free to ping me if
> needed. Right now yours suggestions are clear to me, hence no extra
> questions.
> 
> Thank you for the review.

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 10/15] drm/shmem-helper: Take reservation lock instead of drm_gem_shmem locks

[Daniel Vetter]

On Wed, May 04, 2022 at 06:56:09PM +0300, Dmitry Osipenko wrote:
> On 5/4/22 11:21, Daniel Vetter wrote:
> ...
> >>> - Maybe also do what you suggest and keep a separate lock for this, but
> >>>   the fundamental issue is that this doesn't really work - if you share
> >>>   buffers both ways with two drivers using shmem helpers, then the
> >>>   ordering of this vmap_count_mutex vs dma_resv_lock is inconsistent and
> >>>   you can get some nice deadlocks. So not a great approach (and also the
> >>>   reason why we really need to get everyone to move towards dma_resv_lock
> >>>   as _the_ buffer object lock, since otherwise we'll never get a
> >>>   consistent lock nesting hierarchy).
> >>
> >> The separate locks should work okay because it will be always the
> >> exporter that takes the dma_resv_lock. But I agree that it's less ideal
> >> than defining the new rules for dma-bufs since sometime you will take
> >> the resv lock and sometime not, potentially hiding bugs related to lockings.
> > 
> > That's the issue, some importers need to take the dma_resv_lock for
> > dma_buf_vmap too (e.g. to first nail the buffer in place when it's a
> > dynamic memory manager). In practice it'll work as well as what we have
> > currently, which is similarly inconsistent, except with per-driver locks
> > instead of shared locks from shmem helpers or dma-buf, so less obvious
> > that things are inconsistent.
> > 
> > So yeah if it's too messy maybe the approach is to have a separate lock
> > for vmap for now, land things, and then fix up dma_buf_vmap in a follow up
> > series.
> 
> The amdgpu driver was the fist who introduced the concept of movable
> memory for dma-bufs. Now we want to support it for DRM SHMEM too. For
> both amdgpu ttm and shmem drivers we will want to hold the reservation
> lock when we're touching moveable buffers. The current way of denoting
> that dma-buf is movable is to implement the pin/unpin callbacks of the
> dma-buf ops, should be doable for shmem.

Hm that sounds like a bridge too far? I don't think we want to start
adding moveable dma-bufs for shmem, thus far at least no one asked for
that. Goal here is just to streamline the locking a bit and align across
all the different ways of doing buffers in drm.

Or do you mean something else and I'm just completely lost?

> A day ago I found that mapping of imported dma-bufs is broken at least
> for the Tegra DRM driver (and likely for others too) because driver
> doesn't assume that anyone will try to mmap imported buffer and just
> doesn't handle this case at all, so we're getting a hard lockup on
> touching mapped memory because we're mapping something else than the
> dma-buf.

Huh that sounds bad, how does this happen? Pretty much all pieces of
dma-buf (cpu vmap, userspace mmap, heck even dma_buf_attach) are optional
or at least can fail for various reasons. So exporters not providing mmap
support is fine, but importers then dying is not.

> My plan is to move the dma-buf management code to the level of DRM core
> and make it aware of the reservation locks for the dynamic dma-bufs.
> This way we will get the proper locking for dma-bufs and fix mapping of
> imported dma-bufs for Tegra and other drivers.

So maybe we're completely talking past each another, or coffee is not
working here on my end, but I've no idea what you mean.

We do have some helpers for taking care of the dma_resv_lock dance, and
Christian König has an rfc patch set to maybe unify this further. But that
should be fairly orthogonal to reworking shmem (it might help a bit with
reworking shmem though).
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 10/15] drm/shmem-helper: Take reservation lock instead of drm_gem_shmem locks

[Daniel Vetter]

On Fri, May 06, 2022 at 01:49:12AM +0300, Dmitry Osipenko wrote:
> On 5/5/22 11:12, Daniel Vetter wrote:
> > On Wed, May 04, 2022 at 06:56:09PM +0300, Dmitry Osipenko wrote:
> >> On 5/4/22 11:21, Daniel Vetter wrote:
> >> ...
> >>>>> - Maybe also do what you suggest and keep a separate lock for this, but
> >>>>>   the fundamental issue is that this doesn't really work - if you share
> >>>>>   buffers both ways with two drivers using shmem helpers, then the
> >>>>>   ordering of this vmap_count_mutex vs dma_resv_lock is inconsistent and
> >>>>>   you can get some nice deadlocks. So not a great approach (and also the
> >>>>>   reason why we really need to get everyone to move towards dma_resv_lock
> >>>>>   as _the_ buffer object lock, since otherwise we'll never get a
> >>>>>   consistent lock nesting hierarchy).
> >>>>
> >>>> The separate locks should work okay because it will be always the
> >>>> exporter that takes the dma_resv_lock. But I agree that it's less ideal
> >>>> than defining the new rules for dma-bufs since sometime you will take
> >>>> the resv lock and sometime not, potentially hiding bugs related to lockings.
> >>>
> >>> That's the issue, some importers need to take the dma_resv_lock for
> >>> dma_buf_vmap too (e.g. to first nail the buffer in place when it's a
> >>> dynamic memory manager). In practice it'll work as well as what we have
> >>> currently, which is similarly inconsistent, except with per-driver locks
> >>> instead of shared locks from shmem helpers or dma-buf, so less obvious
> >>> that things are inconsistent.
> >>>
> >>> So yeah if it's too messy maybe the approach is to have a separate lock
> >>> for vmap for now, land things, and then fix up dma_buf_vmap in a follow up
> >>> series.
> >>
> >> The amdgpu driver was the fist who introduced the concept of movable
> >> memory for dma-bufs. Now we want to support it for DRM SHMEM too. For
> >> both amdgpu ttm and shmem drivers we will want to hold the reservation
> >> lock when we're touching moveable buffers. The current way of denoting
> >> that dma-buf is movable is to implement the pin/unpin callbacks of the
> >> dma-buf ops, should be doable for shmem.
> > 
> > Hm that sounds like a bridge too far? I don't think we want to start
> > adding moveable dma-bufs for shmem, thus far at least no one asked for
> > that. Goal here is just to streamline the locking a bit and align across
> > all the different ways of doing buffers in drm.
> > 
> > Or do you mean something else and I'm just completely lost?
> 
> I'm talking about aligning DRM locks with the dma-buf locks. The problem
> is that the convention of dma-bufs isn't specified yet. In particular
> there is no convention for the mapping operations.
> 
> If we want to switch vmapping of shmem to use reservation lock, then
> somebody will have to hold this lock for dma_buf_vmap() and the locking
> convention needs to be specified firmly.

Ah yes that makes sense.

> In case of dynamic buffers, we will also need to specify whether
> dma_buf_vmap() should imply the implicit pinning by exporter or the
> buffer must be pinned explicitly by importer before dma_buf_vmap() is
> invoked.
> 
> Perhaps I indeed shouldn't care about this for this patchset. The
> complete locking model of dma-bufs must be specified first.

Hm I thought vmap is meant to pin itself, and not rely on any other
pinning done already. And from a quick look through the long call chain
for amd (which is currently the only driver supporting dynamic dma-buf)
that seems to be the case.

But yeah the locking isn't specificied yet, and that makes it a bit a mess
:-(

> >> A day ago I found that mapping of imported dma-bufs is broken at least
> >> for the Tegra DRM driver (and likely for others too) because driver
> >> doesn't assume that anyone will try to mmap imported buffer and just
> >> doesn't handle this case at all, so we're getting a hard lockup on
> >> touching mapped memory because we're mapping something else than the
> >> dma-buf.
> > 
> > Huh that sounds bad, how does this happen? Pretty much all pieces of
> > dma-buf (cpu vmap, userspace mmap, heck even dma_buf_attach) are optional
> > or at least can fail for various reasons. So exporters not providing mmap
> > support is fine, but importers then dying is not.
> 
> Those drivers that die don't have userspace that uses dma-bufs
> extensively. I noticed it only because was looking at this code too much
> for the last days.
> 
> Drivers that don't die either map imported BOs properly or don't allow
> mapping at all.

Ah yeah driver bugs as explanation makes sense :-/

> >> My plan is to move the dma-buf management code to the level of DRM core
> >> and make it aware of the reservation locks for the dynamic dma-bufs.
> >> This way we will get the proper locking for dma-bufs and fix mapping of
> >> imported dma-bufs for Tegra and other drivers.
> > 
> > So maybe we're completely talking past each another, or coffee is not
> > working here on my end, but I've no idea what you mean.
> > 
> > We do have some helpers for taking care of the dma_resv_lock dance, and
> > Christian König has an rfc patch set to maybe unify this further. But that
> > should be fairly orthogonal to reworking shmem (it might help a bit with
> > reworking shmem though).
> 
> The reservation lock itself doesn't help much shmem, IMO. It should help
> only in the context of dynamic dma-bufs and today we don't have a need
> in the dynamic shmem dma-bufs.
> 
> You were talking about making DRM locks consistent with dma-buf locks,
> so I thought that yours main point of making use of reservation locks
> for shmem is to prepare to the new locking scheme.
> 
> I wanted to try to specify the dma-buf locking convention for mapping
> operations because it's missing right now and it should affect how DRM
> should take the reservation locks, but this is not easy to do as I see now.
> 
> Could you please point at the Christian's RFC patch? He posted too many
> patches, can't find it :) I'm curious to take a look.

https://lore.kernel.org/dri-devel/20220504074739.2231-1-christian.koenig@amd.com/

Wrt this patch series here I'm wondering whether we could do an interim
solution that side-steps the dma_buf_vmap mess.

- in shmem helpers pin any vmapped buffer (it's how dma-buf works too),
  and that pinning would be done under dma_resv_lock (like with other
  drivers using dma_resv_lock for bo protection)

- switch over everything else except vmap code to dma_resv_lock, but leave
  vmap locking as-is

- shrinker then only needs to trylock dma_resv_trylock in the shrinker,
  which can check for pinned buffer and that's good enough to exclude
  vmap'ed buffer. And it avoids mixing the vmap locking into the new
  shrinker code and driver interfaces.

This still leaves the vmap locking mess as-is, but I think that's a mess
that's orthogonal to shrinker work.

Thoughts?
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 10/15] drm/shmem-helper: Take reservation lock instead of drm_gem_shmem locks

[Daniel Vetter]

On Tue, May 10, 2022 at 04:39:53PM +0300, Dmitry Osipenko wrote:
> On 5/9/22 16:42, Daniel Vetter wrote:
> > On Fri, May 06, 2022 at 01:49:12AM +0300, Dmitry Osipenko wrote:
> >> On 5/5/22 11:12, Daniel Vetter wrote:
> >>> On Wed, May 04, 2022 at 06:56:09PM +0300, Dmitry Osipenko wrote:
> >>>> On 5/4/22 11:21, Daniel Vetter wrote:
> >>>> ...
> >>>>>>> - Maybe also do what you suggest and keep a separate lock for this, but
> >>>>>>>   the fundamental issue is that this doesn't really work - if you share
> >>>>>>>   buffers both ways with two drivers using shmem helpers, then the
> >>>>>>>   ordering of this vmap_count_mutex vs dma_resv_lock is inconsistent and
> >>>>>>>   you can get some nice deadlocks. So not a great approach (and also the
> >>>>>>>   reason why we really need to get everyone to move towards dma_resv_lock
> >>>>>>>   as _the_ buffer object lock, since otherwise we'll never get a
> >>>>>>>   consistent lock nesting hierarchy).
> >>>>>>
> >>>>>> The separate locks should work okay because it will be always the
> >>>>>> exporter that takes the dma_resv_lock. But I agree that it's less ideal
> >>>>>> than defining the new rules for dma-bufs since sometime you will take
> >>>>>> the resv lock and sometime not, potentially hiding bugs related to lockings.
> >>>>>
> >>>>> That's the issue, some importers need to take the dma_resv_lock for
> >>>>> dma_buf_vmap too (e.g. to first nail the buffer in place when it's a
> >>>>> dynamic memory manager). In practice it'll work as well as what we have
> >>>>> currently, which is similarly inconsistent, except with per-driver locks
> >>>>> instead of shared locks from shmem helpers or dma-buf, so less obvious
> >>>>> that things are inconsistent.
> >>>>>
> >>>>> So yeah if it's too messy maybe the approach is to have a separate lock
> >>>>> for vmap for now, land things, and then fix up dma_buf_vmap in a follow up
> >>>>> series.
> >>>>
> >>>> The amdgpu driver was the fist who introduced the concept of movable
> >>>> memory for dma-bufs. Now we want to support it for DRM SHMEM too. For
> >>>> both amdgpu ttm and shmem drivers we will want to hold the reservation
> >>>> lock when we're touching moveable buffers. The current way of denoting
> >>>> that dma-buf is movable is to implement the pin/unpin callbacks of the
> >>>> dma-buf ops, should be doable for shmem.
> >>>
> >>> Hm that sounds like a bridge too far? I don't think we want to start
> >>> adding moveable dma-bufs for shmem, thus far at least no one asked for
> >>> that. Goal here is just to streamline the locking a bit and align across
> >>> all the different ways of doing buffers in drm.
> >>>
> >>> Or do you mean something else and I'm just completely lost?
> >>
> >> I'm talking about aligning DRM locks with the dma-buf locks. The problem
> >> is that the convention of dma-bufs isn't specified yet. In particular
> >> there is no convention for the mapping operations.
> >>
> >> If we want to switch vmapping of shmem to use reservation lock, then
> >> somebody will have to hold this lock for dma_buf_vmap() and the locking
> >> convention needs to be specified firmly.
> > 
> > Ah yes that makes sense.
> > 
> >> In case of dynamic buffers, we will also need to specify whether
> >> dma_buf_vmap() should imply the implicit pinning by exporter or the
> >> buffer must be pinned explicitly by importer before dma_buf_vmap() is
> >> invoked.
> >>
> >> Perhaps I indeed shouldn't care about this for this patchset. The
> >> complete locking model of dma-bufs must be specified first.
> > 
> > Hm I thought vmap is meant to pin itself, and not rely on any other
> > pinning done already. And from a quick look through the long call chain
> > for amd (which is currently the only driver supporting dynamic dma-buf)
> > that seems to be the case.
> 
> The vmapping behaviour is implementation-defined until it's documented
> explicitly, IMO.
> 
> > But yeah the locking isn't specificied yet, and that makes it a bit a mess
> > :-(
> > 
> >>>> A day ago I found that mapping of imported dma-bufs is broken at least
> >>>> for the Tegra DRM driver (and likely for others too) because driver
> >>>> doesn't assume that anyone will try to mmap imported buffer and just
> >>>> doesn't handle this case at all, so we're getting a hard lockup on
> >>>> touching mapped memory because we're mapping something else than the
> >>>> dma-buf.
> >>>
> >>> Huh that sounds bad, how does this happen? Pretty much all pieces of
> >>> dma-buf (cpu vmap, userspace mmap, heck even dma_buf_attach) are optional
> >>> or at least can fail for various reasons. So exporters not providing mmap
> >>> support is fine, but importers then dying is not.
> >>
> >> Those drivers that die don't have userspace that uses dma-bufs
> >> extensively. I noticed it only because was looking at this code too much
> >> for the last days.
> >>
> >> Drivers that don't die either map imported BOs properly or don't allow
> >> mapping at all.
> > 
> > Ah yeah driver bugs as explanation makes sense :-/
> > 
> >>>> My plan is to move the dma-buf management code to the level of DRM core
> >>>> and make it aware of the reservation locks for the dynamic dma-bufs.
> >>>> This way we will get the proper locking for dma-bufs and fix mapping of
> >>>> imported dma-bufs for Tegra and other drivers.
> >>>
> >>> So maybe we're completely talking past each another, or coffee is not
> >>> working here on my end, but I've no idea what you mean.
> >>>
> >>> We do have some helpers for taking care of the dma_resv_lock dance, and
> >>> Christian König has an rfc patch set to maybe unify this further. But that
> >>> should be fairly orthogonal to reworking shmem (it might help a bit with
> >>> reworking shmem though).
> >>
> >> The reservation lock itself doesn't help much shmem, IMO. It should help
> >> only in the context of dynamic dma-bufs and today we don't have a need
> >> in the dynamic shmem dma-bufs.
> >>
> >> You were talking about making DRM locks consistent with dma-buf locks,
> >> so I thought that yours main point of making use of reservation locks
> >> for shmem is to prepare to the new locking scheme.
> >>
> >> I wanted to try to specify the dma-buf locking convention for mapping
> >> operations because it's missing right now and it should affect how DRM
> >> should take the reservation locks, but this is not easy to do as I see now.
> >>
> >> Could you please point at the Christian's RFC patch? He posted too many
> >> patches, can't find it :) I'm curious to take a look.
> > 
> > https://lore.kernel.org/dri-devel/20220504074739.2231-1-christian.koenig@amd.com/
> > 
> > Wrt this patch series here I'm wondering whether we could do an interim
> > solution that side-steps the dma_buf_vmap mess.
> > 
> > - in shmem helpers pin any vmapped buffer (it's how dma-buf works too),
> >   and that pinning would be done under dma_resv_lock (like with other
> >   drivers using dma_resv_lock for bo protection)
> > 
> > - switch over everything else except vmap code to dma_resv_lock, but leave
> >   vmap locking as-is
> > 
> > - shrinker then only needs to trylock dma_resv_trylock in the shrinker,
> >   which can check for pinned buffer and that's good enough to exclude
> >   vmap'ed buffer. And it avoids mixing the vmap locking into the new
> >   shrinker code and driver interfaces.
> > 
> > This still leaves the vmap locking mess as-is, but I think that's a mess
> > that's orthogonal to shrinker work.
> > 
> > Thoughts?
> 
> Since vmapping implies implicit pinning, we can't use a separate lock in
> drm_gem_shmem_vmap() because we need to protect the
> drm_gem_shmem_get_pages(), which is invoked by drm_gem_shmem_vmap() to
> pin the pages and requires the dma_resv_lock to be locked.
> 
> Hence the problem is:
> 
> 1. If dma-buf importer holds the dma_resv_lock and invokes
> dma_buf_vmap() -> drm_gem_shmem_vmap(), then drm_gem_shmem_vmap() shall
> not take the dma_resv_lock.
> 
> 2. Since dma-buf locking convention isn't specified, we can't assume
> that dma-buf importer holds the dma_resv_lock around dma_buf_vmap().
> 
> The possible solutions are:
> 
> 1. Specify the dma_resv_lock convention for dma-bufs and make all
> drivers to follow it.
> 
> 2. Make only DRM drivers to hold dma_resv_lock around dma_buf_vmap().
> Other non-DRM drivers will get the lockdep warning.
> 
> 3. Make drm_gem_shmem_vmap() to take the dma_resv_lock and get deadlock
> if dma-buf importer holds the lock.
> 
> ...

Yeah this is all very annoying.

> There are actually very few drivers in kernel that use dma_buf_vmap()
> [1], so perhaps it's not really a big deal to first try to define the
> locking and pinning convention for the dma-bufs? At least for
> dma_buf_vmap()? Let me try to do this.
> 
> [1] https://elixir.bootlin.com/linux/v5.18-rc6/C/ident/dma_buf_vmap

Yeah looking through the code there's largely two classes of drivers that
need vmap:

- display drivers that need to do cpu upload (usb, spi, i2c displays).
  Those generally set up the vmap at import time or when creating the
  drm_framebuffer object (e.g. see
  drm_gem_cma_prime_import_sg_table_vmap()), because that's really the
  only place where you can safely do that without running into locking
  inversion issues sooner or later

- lots of other drivers (and shmem helpers) seem to do dma_buf_vmap just
  because they can, but only actually ever use vmap on native objects,
  never on imported objects. Or at least I think so.

So maybe another approach here:

1. In general drivers which need a vmap need to set that up at dma_buf
import time - the same way we pin the buffers at import time for
non-dynamic importers because that's the only place where across all
drivers it's ok to just take dma_resv_lock.

2. We remove the "just because we can" dma_buf_vmap support from
helpers/drivers - the paths all already can cope with NULL since
dma_buf_vmap can fail. vmap will only work on native objects, not imported
ones.

3. If there is any driver using shmem helpers that absolutely needs vmap
to also work on imported it needs a special import function (like cma
helpers) which sets up the vmap at import time.

So since this is all very tricky ... what did I miss this time around?

> I envision that the extra dma_resv_locks for dma-bufs potentially may
> create unnecessary bottlenecks for some drivers if locking isn't really
> necessary by a specific driver, so drivers will need to keep this in
> mind. On the other hand, I don't think that any of the today's drivers
> will notice the additional resv locks in practice.

Nah I don't think the extra locking will ever create a bottleneck,
especially not for vmap. Generally vmap is a fallback or at least cpu
operation, so at that point you're already going very slow.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 10/15] drm/shmem-helper: Take reservation lock instead of drm_gem_shmem locks

[Christian König]

Am 11.05.22 um 15:00 schrieb Daniel Vetter:
> On Tue, May 10, 2022 at 04:39:53PM +0300, Dmitry Osipenko wrote:
>> [SNIP]
>> Since vmapping implies implicit pinning, we can't use a separate lock in
>> drm_gem_shmem_vmap() because we need to protect the
>> drm_gem_shmem_get_pages(), which is invoked by drm_gem_shmem_vmap() to
>> pin the pages and requires the dma_resv_lock to be locked.
>>
>> Hence the problem is:
>>
>> 1. If dma-buf importer holds the dma_resv_lock and invokes
>> dma_buf_vmap() -> drm_gem_shmem_vmap(), then drm_gem_shmem_vmap() shall
>> not take the dma_resv_lock.
>>
>> 2. Since dma-buf locking convention isn't specified, we can't assume
>> that dma-buf importer holds the dma_resv_lock around dma_buf_vmap().
>>
>> The possible solutions are:
>>
>> 1. Specify the dma_resv_lock convention for dma-bufs and make all
>> drivers to follow it.
>>
>> 2. Make only DRM drivers to hold dma_resv_lock around dma_buf_vmap().
>> Other non-DRM drivers will get the lockdep warning.
>>
>> 3. Make drm_gem_shmem_vmap() to take the dma_resv_lock and get deadlock
>> if dma-buf importer holds the lock.
>>
>> ...
> Yeah this is all very annoying.

Ah, yes that topic again :)

I think we could relatively easily fix that by just defining and 
enforcing that the dma_resv_lock must have be taken by the caller when 
dma_buf_vmap() is called.

A two step approach should work:
1. Move the call to dma_resv_lock() into the dma_buf_vmap() function and 
remove all lock taking from the vmap callback implementations.
2. Move the call to dma_resv_lock() into the callers of dma_buf_vmap() 
and enforce that the function is called with the lock held.

It shouldn't be that hard to clean up. The last time I looked into it my 
main problem was that we didn't had any easy unit test for it.

Regards,
Christian.

>
>> There are actually very few drivers in kernel that use dma_buf_vmap()
>> [1], so perhaps it's not really a big deal to first try to define the
>> locking and pinning convention for the dma-bufs? At least for
>> dma_buf_vmap()? Let me try to do this.
>>
>> [1] https://elixir.bootlin.com/linux/v5.18-rc6/C/ident/dma_buf_vmap
> Yeah looking through the code there's largely two classes of drivers that
> need vmap:
>
> - display drivers that need to do cpu upload (usb, spi, i2c displays).
>    Those generally set up the vmap at import time or when creating the
>    drm_framebuffer object (e.g. see
>    drm_gem_cma_prime_import_sg_table_vmap()), because that's really the
>    only place where you can safely do that without running into locking
>    inversion issues sooner or later
>
> - lots of other drivers (and shmem helpers) seem to do dma_buf_vmap just
>    because they can, but only actually ever use vmap on native objects,
>    never on imported objects. Or at least I think so.
>
> So maybe another approach here:
>
> 1. In general drivers which need a vmap need to set that up at dma_buf
> import time - the same way we pin the buffers at import time for
> non-dynamic importers because that's the only place where across all
> drivers it's ok to just take dma_resv_lock.
>
> 2. We remove the "just because we can" dma_buf_vmap support from
> helpers/drivers - the paths all already can cope with NULL since
> dma_buf_vmap can fail. vmap will only work on native objects, not imported
> ones.
>
> 3. If there is any driver using shmem helpers that absolutely needs vmap
> to also work on imported it needs a special import function (like cma
> helpers) which sets up the vmap at import time.
>
> So since this is all very tricky ... what did I miss this time around?
>
>> I envision that the extra dma_resv_locks for dma-bufs potentially may
>> create unnecessary bottlenecks for some drivers if locking isn't really
>> necessary by a specific driver, so drivers will need to keep this in
>> mind. On the other hand, I don't think that any of the today's drivers
>> will notice the additional resv locks in practice.
> Nah I don't think the extra locking will ever create a bottleneck,
> especially not for vmap. Generally vmap is a fallback or at least cpu
> operation, so at that point you're already going very slow.
> -Daniel

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 10/15] drm/shmem-helper: Take reservation lock instead of drm_gem_shmem locks

[Daniel Vetter]

On Wed, May 11, 2022 at 04:24:28PM +0200, Christian König wrote:
> Am 11.05.22 um 15:00 schrieb Daniel Vetter:
> > On Tue, May 10, 2022 at 04:39:53PM +0300, Dmitry Osipenko wrote:
> > > [SNIP]
> > > Since vmapping implies implicit pinning, we can't use a separate lock in
> > > drm_gem_shmem_vmap() because we need to protect the
> > > drm_gem_shmem_get_pages(), which is invoked by drm_gem_shmem_vmap() to
> > > pin the pages and requires the dma_resv_lock to be locked.
> > > 
> > > Hence the problem is:
> > > 
> > > 1. If dma-buf importer holds the dma_resv_lock and invokes
> > > dma_buf_vmap() -> drm_gem_shmem_vmap(), then drm_gem_shmem_vmap() shall
> > > not take the dma_resv_lock.
> > > 
> > > 2. Since dma-buf locking convention isn't specified, we can't assume
> > > that dma-buf importer holds the dma_resv_lock around dma_buf_vmap().
> > > 
> > > The possible solutions are:
> > > 
> > > 1. Specify the dma_resv_lock convention for dma-bufs and make all
> > > drivers to follow it.
> > > 
> > > 2. Make only DRM drivers to hold dma_resv_lock around dma_buf_vmap().
> > > Other non-DRM drivers will get the lockdep warning.
> > > 
> > > 3. Make drm_gem_shmem_vmap() to take the dma_resv_lock and get deadlock
> > > if dma-buf importer holds the lock.
> > > 
> > > ...
> > Yeah this is all very annoying.
> 
> Ah, yes that topic again :)
> 
> I think we could relatively easily fix that by just defining and enforcing
> that the dma_resv_lock must have be taken by the caller when dma_buf_vmap()
> is called.
> 
> A two step approach should work:
> 1. Move the call to dma_resv_lock() into the dma_buf_vmap() function and
> remove all lock taking from the vmap callback implementations.
> 2. Move the call to dma_resv_lock() into the callers of dma_buf_vmap() and
> enforce that the function is called with the lock held.
> 
> It shouldn't be that hard to clean up. The last time I looked into it my
> main problem was that we didn't had any easy unit test for it.

Yeah I think it's doable or at least a lot less work than the map/unmap
side, which really was unfixable without just pinning at import time to
avoid the locking fun. But vmap is used a lot less, and mostly by display
drivers (where locking is a lot easier against dma_resv_lock), so it might
be possible to pull off.
-Daniel

> 
> Regards,
> Christian.
> 
> > 
> > > There are actually very few drivers in kernel that use dma_buf_vmap()
> > > [1], so perhaps it's not really a big deal to first try to define the
> > > locking and pinning convention for the dma-bufs? At least for
> > > dma_buf_vmap()? Let me try to do this.
> > > 
> > > [1] https://elixir.bootlin.com/linux/v5.18-rc6/C/ident/dma_buf_vmap
> > Yeah looking through the code there's largely two classes of drivers that
> > need vmap:
> > 
> > - display drivers that need to do cpu upload (usb, spi, i2c displays).
> >    Those generally set up the vmap at import time or when creating the
> >    drm_framebuffer object (e.g. see
> >    drm_gem_cma_prime_import_sg_table_vmap()), because that's really the
> >    only place where you can safely do that without running into locking
> >    inversion issues sooner or later
> > 
> > - lots of other drivers (and shmem helpers) seem to do dma_buf_vmap just
> >    because they can, but only actually ever use vmap on native objects,
> >    never on imported objects. Or at least I think so.
> > 
> > So maybe another approach here:
> > 
> > 1. In general drivers which need a vmap need to set that up at dma_buf
> > import time - the same way we pin the buffers at import time for
> > non-dynamic importers because that's the only place where across all
> > drivers it's ok to just take dma_resv_lock.
> > 
> > 2. We remove the "just because we can" dma_buf_vmap support from
> > helpers/drivers - the paths all already can cope with NULL since
> > dma_buf_vmap can fail. vmap will only work on native objects, not imported
> > ones.
> > 
> > 3. If there is any driver using shmem helpers that absolutely needs vmap
> > to also work on imported it needs a special import function (like cma
> > helpers) which sets up the vmap at import time.
> > 
> > So since this is all very tricky ... what did I miss this time around?
> > 
> > > I envision that the extra dma_resv_locks for dma-bufs potentially may
> > > create unnecessary bottlenecks for some drivers if locking isn't really
> > > necessary by a specific driver, so drivers will need to keep this in
> > > mind. On the other hand, I don't think that any of the today's drivers
> > > will notice the additional resv locks in practice.
> > Nah I don't think the extra locking will ever create a bottleneck,
> > especially not for vmap. Generally vmap is a fallback or at least cpu
> > operation, so at that point you're already going very slow.
> > -Daniel
> 

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 10/15] drm/shmem-helper: Take reservation lock instead of drm_gem_shmem locks

[Daniel Vetter]

On Wed, May 11, 2022 at 06:14:00PM +0300, Dmitry Osipenko wrote:
> On 5/11/22 17:24, Christian König wrote:
> > Am 11.05.22 um 15:00 schrieb Daniel Vetter:
> >> On Tue, May 10, 2022 at 04:39:53PM +0300, Dmitry Osipenko wrote:
> >>> [SNIP]
> >>> Since vmapping implies implicit pinning, we can't use a separate lock in
> >>> drm_gem_shmem_vmap() because we need to protect the
> >>> drm_gem_shmem_get_pages(), which is invoked by drm_gem_shmem_vmap() to
> >>> pin the pages and requires the dma_resv_lock to be locked.
> >>>
> >>> Hence the problem is:
> >>>
> >>> 1. If dma-buf importer holds the dma_resv_lock and invokes
> >>> dma_buf_vmap() -> drm_gem_shmem_vmap(), then drm_gem_shmem_vmap() shall
> >>> not take the dma_resv_lock.
> >>>
> >>> 2. Since dma-buf locking convention isn't specified, we can't assume
> >>> that dma-buf importer holds the dma_resv_lock around dma_buf_vmap().
> >>>
> >>> The possible solutions are:
> >>>
> >>> 1. Specify the dma_resv_lock convention for dma-bufs and make all
> >>> drivers to follow it.
> >>>
> >>> 2. Make only DRM drivers to hold dma_resv_lock around dma_buf_vmap().
> >>> Other non-DRM drivers will get the lockdep warning.
> >>>
> >>> 3. Make drm_gem_shmem_vmap() to take the dma_resv_lock and get deadlock
> >>> if dma-buf importer holds the lock.
> >>>
> >>> ...
> >> Yeah this is all very annoying.
> > 
> > Ah, yes that topic again :)
> > 
> > I think we could relatively easily fix that by just defining and
> > enforcing that the dma_resv_lock must have be taken by the caller when
> > dma_buf_vmap() is called.
> > 
> > A two step approach should work:
> > 1. Move the call to dma_resv_lock() into the dma_buf_vmap() function and
> > remove all lock taking from the vmap callback implementations.
> > 2. Move the call to dma_resv_lock() into the callers of dma_buf_vmap()
> > and enforce that the function is called with the lock held.
> 
> I've doubts about the need to move out the dma_resv_lock() into the
> callers of dma_buf_vmap()..
> 
> I looked through all the dma_buf_vmap() users and neither of them
> interacts with dma_resv_lock() at all, i.e. nobody takes the lock
> in/outside of dma_buf_vmap(). Hence it's easy and more practical to make
> dma_buf_mmap/vmap() to take the dma_resv_lock by themselves.

i915_gem_dmabuf_vmap -> i915_gem_object_pin_map_unlocked ->
  i915_gem_object_lock -> dma_resv_lock

And all the ttm drivers should work similarly. So there's definitely
drivers which grab dma_resv_lock from their vmap callback.

> It's unclear to me which driver may ever want to do the mapping under
> the dma_resv_lock. But if we will ever have such a driver that will need
> to map imported buffer under dma_resv_lock, then we could always add the
> dma_buf_vmap_locked() variant of the function. In this case the locking
> rule will sound like this:
> 
> "All dma-buf importers are responsible for holding the dma-reservation
> lock around the dmabuf->ops->mmap/vmap() calls."
> 
> > It shouldn't be that hard to clean up. The last time I looked into it my
> > main problem was that we didn't had any easy unit test for it.
> 
> Do we have any tests for dma-bufs at all? It's unclear to me what you
> are going to test in regards to the reservation locks, could you please
> clarify?

Unfortunately not really :-/ Only way really is to grab a driver which
needs vmap (those are mostly display drivers) on an imported buffer, and
see what happens.

2nd best is liberally sprinkling lockdep annotations all over the place
and throwing it at intel ci (not sure amd ci is accessible to the public)
and then hoping that's good enough. Stuff like might_lock and
dma_resv_assert_held.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 10/15] drm/shmem-helper: Take reservation lock instead of drm_gem_shmem locks

[Daniel Vetter]

On Wed, May 11, 2022 at 06:40:32PM +0300, Dmitry Osipenko wrote:
> On 5/11/22 18:29, Daniel Vetter wrote:
> > On Wed, May 11, 2022 at 06:14:00PM +0300, Dmitry Osipenko wrote:
> >> On 5/11/22 17:24, Christian König wrote:
> >>> Am 11.05.22 um 15:00 schrieb Daniel Vetter:
> >>>> On Tue, May 10, 2022 at 04:39:53PM +0300, Dmitry Osipenko wrote:
> >>>>> [SNIP]
> >>>>> Since vmapping implies implicit pinning, we can't use a separate lock in
> >>>>> drm_gem_shmem_vmap() because we need to protect the
> >>>>> drm_gem_shmem_get_pages(), which is invoked by drm_gem_shmem_vmap() to
> >>>>> pin the pages and requires the dma_resv_lock to be locked.
> >>>>>
> >>>>> Hence the problem is:
> >>>>>
> >>>>> 1. If dma-buf importer holds the dma_resv_lock and invokes
> >>>>> dma_buf_vmap() -> drm_gem_shmem_vmap(), then drm_gem_shmem_vmap() shall
> >>>>> not take the dma_resv_lock.
> >>>>>
> >>>>> 2. Since dma-buf locking convention isn't specified, we can't assume
> >>>>> that dma-buf importer holds the dma_resv_lock around dma_buf_vmap().
> >>>>>
> >>>>> The possible solutions are:
> >>>>>
> >>>>> 1. Specify the dma_resv_lock convention for dma-bufs and make all
> >>>>> drivers to follow it.
> >>>>>
> >>>>> 2. Make only DRM drivers to hold dma_resv_lock around dma_buf_vmap().
> >>>>> Other non-DRM drivers will get the lockdep warning.
> >>>>>
> >>>>> 3. Make drm_gem_shmem_vmap() to take the dma_resv_lock and get deadlock
> >>>>> if dma-buf importer holds the lock.
> >>>>>
> >>>>> ...
> >>>> Yeah this is all very annoying.
> >>> Ah, yes that topic again :)
> >>>
> >>> I think we could relatively easily fix that by just defining and
> >>> enforcing that the dma_resv_lock must have be taken by the caller when
> >>> dma_buf_vmap() is called.
> >>>
> >>> A two step approach should work:
> >>> 1. Move the call to dma_resv_lock() into the dma_buf_vmap() function and
> >>> remove all lock taking from the vmap callback implementations.
> >>> 2. Move the call to dma_resv_lock() into the callers of dma_buf_vmap()
> >>> and enforce that the function is called with the lock held.
> >> I've doubts about the need to move out the dma_resv_lock() into the
> >> callers of dma_buf_vmap()..
> >>
> >> I looked through all the dma_buf_vmap() users and neither of them
> >> interacts with dma_resv_lock() at all, i.e. nobody takes the lock
> >> in/outside of dma_buf_vmap(). Hence it's easy and more practical to make
> >> dma_buf_mmap/vmap() to take the dma_resv_lock by themselves.
> > i915_gem_dmabuf_vmap -> i915_gem_object_pin_map_unlocked ->
> >   i915_gem_object_lock -> dma_resv_lock
> > 
> > And all the ttm drivers should work similarly. So there's definitely
> > drivers which grab dma_resv_lock from their vmap callback.
> 
> Grr.. I'll take another look.
> 
> >> It's unclear to me which driver may ever want to do the mapping under
> >> the dma_resv_lock. But if we will ever have such a driver that will need
> >> to map imported buffer under dma_resv_lock, then we could always add the
> >> dma_buf_vmap_locked() variant of the function. In this case the locking
> >> rule will sound like this:
> >>
> >> "All dma-buf importers are responsible for holding the dma-reservation
> >> lock around the dmabuf->ops->mmap/vmap() calls."
> 
> Are you okay with this rule?

Yeah I think long-term it's where we want to be, just trying to find
clever ways to get there.

And I think Christian agrees with that?

> >>> It shouldn't be that hard to clean up. The last time I looked into it my
> >>> main problem was that we didn't had any easy unit test for it.
> >> Do we have any tests for dma-bufs at all? It's unclear to me what you
> >> are going to test in regards to the reservation locks, could you please
> >> clarify?
> > Unfortunately not really :-/ Only way really is to grab a driver which
> > needs vmap (those are mostly display drivers) on an imported buffer, and
> > see what happens.
> > 
> > 2nd best is liberally sprinkling lockdep annotations all over the place
> > and throwing it at intel ci (not sure amd ci is accessible to the public)
> > and then hoping that's good enough. Stuff like might_lock and
> > dma_resv_assert_held.
> 
> Alright

So throwing it at intel-gfx-ci can't hurt I think, but that only covers
i915 so doesn't really help with the bigger issue of catching all the
drivers.

Cheers, Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 10/15] drm/shmem-helper: Take reservation lock instead of drm_gem_shmem locks

[Christian König]

Am 11.05.22 um 21:05 schrieb Daniel Vetter:
> [SNIP]
>>>> It's unclear to me which driver may ever want to do the mapping under
>>>> the dma_resv_lock. But if we will ever have such a driver that will need
>>>> to map imported buffer under dma_resv_lock, then we could always add the
>>>> dma_buf_vmap_locked() variant of the function. In this case the locking
>>>> rule will sound like this:
>>>>
>>>> "All dma-buf importers are responsible for holding the dma-reservation
>>>> lock around the dmabuf->ops->mmap/vmap() calls."
>> Are you okay with this rule?
> Yeah I think long-term it's where we want to be, just trying to find
> clever ways to get there.
>
> And I think Christian agrees with that?

Yes, completely.

A design where most DMA-buf functions are supposed to be called with the 
reservation lock held is exactly what I have in mind for the long term.

>>>>> It shouldn't be that hard to clean up. The last time I looked into it my
>>>>> main problem was that we didn't had any easy unit test for it.
>>>> Do we have any tests for dma-bufs at all? It's unclear to me what you
>>>> are going to test in regards to the reservation locks, could you please
>>>> clarify?
>>> Unfortunately not really :-/ Only way really is to grab a driver which
>>> needs vmap (those are mostly display drivers) on an imported buffer, and
>>> see what happens.
>>>
>>> 2nd best is liberally sprinkling lockdep annotations all over the place
>>> and throwing it at intel ci (not sure amd ci is accessible to the public)
>>> and then hoping that's good enough. Stuff like might_lock and
>>> dma_resv_assert_held.
>> Alright
> So throwing it at intel-gfx-ci can't hurt I think, but that only covers
> i915 so doesn't really help with the bigger issue of catching all the
> drivers.

BTW: We have now somebody working on converting the existing 
libdrm_amdgpu unit tests over to igt.

Regards,
Christian.

>
> Cheers, Daniel

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 10/15] drm/shmem-helper: Take reservation lock instead of drm_gem_shmem locks

[Daniel Vetter]

On Thu, May 12, 2022 at 09:29:35AM +0200, Christian König wrote:
> Am 11.05.22 um 21:05 schrieb Daniel Vetter:
> > [SNIP]
> > > > > It's unclear to me which driver may ever want to do the mapping under
> > > > > the dma_resv_lock. But if we will ever have such a driver that will need
> > > > > to map imported buffer under dma_resv_lock, then we could always add the
> > > > > dma_buf_vmap_locked() variant of the function. In this case the locking
> > > > > rule will sound like this:
> > > > > 
> > > > > "All dma-buf importers are responsible for holding the dma-reservation
> > > > > lock around the dmabuf->ops->mmap/vmap() calls."
> > > Are you okay with this rule?
> > Yeah I think long-term it's where we want to be, just trying to find
> > clever ways to get there.
> > 
> > And I think Christian agrees with that?
> 
> Yes, completely.
> 
> A design where most DMA-buf functions are supposed to be called with the
> reservation lock held is exactly what I have in mind for the long term.
> 
> > > > > > It shouldn't be that hard to clean up. The last time I looked into it my
> > > > > > main problem was that we didn't had any easy unit test for it.
> > > > > Do we have any tests for dma-bufs at all? It's unclear to me what you
> > > > > are going to test in regards to the reservation locks, could you please
> > > > > clarify?
> > > > Unfortunately not really :-/ Only way really is to grab a driver which
> > > > needs vmap (those are mostly display drivers) on an imported buffer, and
> > > > see what happens.
> > > > 
> > > > 2nd best is liberally sprinkling lockdep annotations all over the place
> > > > and throwing it at intel ci (not sure amd ci is accessible to the public)
> > > > and then hoping that's good enough. Stuff like might_lock and
> > > > dma_resv_assert_held.
> > > Alright
> > So throwing it at intel-gfx-ci can't hurt I think, but that only covers
> > i915 so doesn't really help with the bigger issue of catching all the
> > drivers.
> 
> BTW: We have now somebody working on converting the existing libdrm_amdgpu
> unit tests over to igt.

This sounds awesome.

/me throws a happy dance

Cheers, Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 11/15] drm/shmem-helper: Add generic memory shrinker

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 38600 bytes --]

Hi

Am 18.04.22 um 00:37 schrieb Dmitry Osipenko:
> Introduce a common DRM SHMEM shrinker. It allows to reduce code
> duplication among DRM drivers that implement theirs own shrinkers.
> This is initial version of the shrinker that covers basic needs of
> GPU drivers, both purging and eviction of shmem objects are supported.
> 
> This patch is based on a couple ideas borrowed from Rob's Clark MSM
> shrinker and Thomas' Zimmermann variant of SHMEM shrinker.
> 
> In order to start using DRM SHMEM shrinker drivers should:
> 
> 1. Implement new purge(), evict() + swap_in() GEM callbacks.
> 2. Register shrinker using drm_gem_shmem_shrinker_register(drm_device).
> 3. Use drm_gem_shmem_set_purgeable_and_evictable(shmem) and alike API
>     functions to activate shrinking of GEMs.
> 
> Signed-off-by: Daniel Almeida <daniel.almeida@collabora.com>
> Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
> ---
>   drivers/gpu/drm/drm_gem_shmem_helper.c | 765 ++++++++++++++++++++++++-
>   include/drm/drm_device.h               |   4 +
>   include/drm/drm_gem.h                  |  35 ++
>   include/drm/drm_gem_shmem_helper.h     | 105 +++-
>   4 files changed, 877 insertions(+), 32 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_gem_shmem_helper.c b/drivers/gpu/drm/drm_gem_shmem_helper.c
> index 3ecef571eff3..3838fb8d6f3a 100644
> --- a/drivers/gpu/drm/drm_gem_shmem_helper.c
> +++ b/drivers/gpu/drm/drm_gem_shmem_helper.c
> @@ -88,6 +88,13 @@ __drm_gem_shmem_create(struct drm_device *dev, size_t size, bool private)
>   
>   	INIT_LIST_HEAD(&shmem->madv_list);
>   
> +	/*
> +	 * Eviction and purging are disabled by default, shmem user must enable
> +	 * them explicitly using drm_gem_shmem_set_evictable/purgeable().
> +	 */
> +	shmem->eviction_disable_count = 1;
> +	shmem->purging_disable_count = 1;
> +
>   	if (!private) {
>   		/*
>   		 * Our buffers are kept pinned, so allocating them
> @@ -126,6 +133,107 @@ struct drm_gem_shmem_object *drm_gem_shmem_create(struct drm_device *dev, size_t
>   }
>   EXPORT_SYMBOL_GPL(drm_gem_shmem_create);
>   
> +static void
> +drm_gem_shmem_add_pages_to_shrinker(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> +	size_t page_count = obj->size >> PAGE_SHIFT;
> +
> +	if (!shmem->pages_shrinkable) {
> +		WARN_ON(gem_shrinker->shrinkable_count + page_count < page_count);
> +		gem_shrinker->shrinkable_count += page_count;
> +		shmem->pages_shrinkable = true;
> +	}
> +}
> +
> +static void
> +drm_gem_shmem_remove_pages_from_shrinker(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> +	size_t page_count = obj->size >> PAGE_SHIFT;
> +
> +	if (shmem->pages_shrinkable) {
> +		WARN_ON(gem_shrinker->shrinkable_count < page_count);
> +		gem_shrinker->shrinkable_count -= page_count;
> +		shmem->pages_shrinkable = false;
> +	}
> +}
> +
> +static void
> +drm_gem_shmem_set_pages_state_locked(struct drm_gem_shmem_object *shmem,
> +				     enum drm_gem_shmem_pages_state new_state)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> +
> +	lockdep_assert_held(&gem_shrinker->lock);
> +	lockdep_assert_held(&obj->resv->lock.base);
> +
> +	if (new_state >= DRM_GEM_SHMEM_PAGES_STATE_PINNED) {
> +		if (drm_gem_shmem_is_evictable(shmem))
> +			new_state = DRM_GEM_SHMEM_PAGES_STATE_EVICTABLE;
> +
> +		if (drm_gem_shmem_is_purgeable(shmem))
> +			new_state = DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE;
> +
> +		if (!shmem->pages)
> +			new_state = DRM_GEM_SHMEM_PAGES_STATE_UNPINNED;
> +
> +		if (shmem->evicted)
> +			new_state = DRM_GEM_SHMEM_PAGES_STATE_EVICTED;
> +	}
> +
> +	if (shmem->pages_state == new_state)
> +		return;
> +
> +	switch (new_state) {
> +	case DRM_GEM_SHMEM_PAGES_STATE_UNPINNED:
> +	case DRM_GEM_SHMEM_PAGES_STATE_PURGED:
> +		drm_gem_shmem_remove_pages_from_shrinker(shmem);
> +		list_del_init(&shmem->madv_list);
> +		break;
> +
> +	case DRM_GEM_SHMEM_PAGES_STATE_PINNED:
> +		drm_gem_shmem_remove_pages_from_shrinker(shmem);
> +		list_move_tail(&shmem->madv_list, &gem_shrinker->lru_active);
> +		break;
> +
> +	case DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE:
> +		drm_gem_shmem_add_pages_to_shrinker(shmem);
> +		list_move_tail(&shmem->madv_list, &gem_shrinker->lru_purgeable);
> +		break;
> +
> +	case DRM_GEM_SHMEM_PAGES_STATE_EVICTABLE:
> +		drm_gem_shmem_add_pages_to_shrinker(shmem);
> +		list_move_tail(&shmem->madv_list, &gem_shrinker->lru_evictable);
> +		break;
> +
> +	case DRM_GEM_SHMEM_PAGES_STATE_EVICTED:
> +		drm_gem_shmem_remove_pages_from_shrinker(shmem);
> +		list_move_tail(&shmem->madv_list, &gem_shrinker->lru_evicted);
> +		break;
> +	}
> +
> +	shmem->pages_state = new_state;
> +}
> +
> +static void
> +drm_gem_shmem_set_pages_state(struct drm_gem_shmem_object *shmem,
> +			      enum drm_gem_shmem_pages_state new_state)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> +
> +	if (!gem_shrinker)
> +		return;
> +
> +	mutex_lock(&gem_shrinker->lock);
> +	drm_gem_shmem_set_pages_state_locked(shmem, new_state);
> +	mutex_unlock(&gem_shrinker->lock);
> +}
> +
>   /**
>    * drm_gem_shmem_free - Free resources associated with a shmem GEM object
>    * @shmem: shmem GEM object to free
> @@ -137,6 +245,9 @@ void drm_gem_shmem_free(struct drm_gem_shmem_object *shmem)
>   {
>   	struct drm_gem_object *obj = &shmem->base;
>   
> +	/* take out shmem GEM object from the memory shrinker */
> +	drm_gem_shmem_madvise(shmem, -1);
> +
>   	WARN_ON(shmem->vmap_use_count);
>   
>   	if (obj->import_attach) {
> @@ -148,7 +259,7 @@ void drm_gem_shmem_free(struct drm_gem_shmem_object *shmem)
>   			sg_free_table(shmem->sgt);
>   			kfree(shmem->sgt);
>   		}
> -		if (shmem->pages)
> +		if (shmem->pages_use_count)
>   			drm_gem_shmem_put_pages(shmem);
>   	}
>   
> @@ -159,18 +270,226 @@ void drm_gem_shmem_free(struct drm_gem_shmem_object *shmem)
>   }
>   EXPORT_SYMBOL_GPL(drm_gem_shmem_free);
>   
> -static int drm_gem_shmem_get_pages_locked(struct drm_gem_shmem_object *shmem)
> +static void drm_gem_shmem_update_pages_state_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> +	enum drm_gem_shmem_pages_state new_state;
> +
> +	if (!gem_shrinker || obj->import_attach)
> +		return;
> +
> +	mutex_lock(&gem_shrinker->lock);
> +
> +	if (!shmem->madv)
> +		new_state = DRM_GEM_SHMEM_PAGES_STATE_PINNED;
> +	else if (shmem->madv > 0)
> +		new_state = DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE;
> +	else
> +		new_state = DRM_GEM_SHMEM_PAGES_STATE_PURGED;
> +
> +	drm_gem_shmem_set_pages_state_locked(shmem, new_state);
> +
> +	mutex_unlock(&gem_shrinker->lock);
> +}
> +
> +static void drm_gem_shmem_update_pages_state(struct drm_gem_shmem_object *shmem)
> +{
> +	dma_resv_lock(shmem->base.resv, NULL);
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +	dma_resv_unlock(shmem->base.resv);
> +}
> +
> +static int
> +drm_gem_shmem_set_evictable_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	int ret = 0;
> +
> +	WARN_ON_ONCE(!shmem->eviction_disable_count--);
> +
> +	if (shmem->madv < 0)
> +		ret = -ENOMEM;
> +
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +
> +	return ret;
> +}
> +
> +static int
> +drm_gem_shmem_set_unevictable_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +	int err;
> +
> +	if (shmem->madv < 0)
> +		return -ENOMEM;
> +
> +	if (shmem->evicted) {
> +		err = obj->funcs->swap_in(obj);
> +		if (err)
> +			return err;
> +	}
> +
> +	shmem->eviction_disable_count++;
> +
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +
> +	return 0;
> +}
> +
> +static int
> +drm_gem_shmem_set_purgeable_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	int ret = 0;
> +
> +	WARN_ON_ONCE(!shmem->purging_disable_count--);
> +
> +	if (shmem->madv < 0)
> +		ret = -ENOMEM;
> +
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +
> +	return ret;
> +}
> +
> +/**
> + * drm_gem_shmem_set_purgeable() - Make GEM purgeable by memory shrinker
> + * @shmem: shmem GEM object
> + *
> + * Tell memory shrinker that this GEM can be purged. Initially purging is
> + * disabled for all GEMs. Each set_pureable() call must have corresponding
> + * set_unpureable() call. If GEM was purged, then -ENOMEM is returned.
> + *
> + * Returns:
> + * 0 on success or a negative error code on failure.
> + */
> +int drm_gem_shmem_set_purgeable(struct drm_gem_shmem_object *shmem)
> +{
> +	int ret;
> +
> +	dma_resv_lock(shmem->base.resv, NULL);
> +	ret = drm_gem_shmem_set_purgeable_locked(shmem);
> +	dma_resv_unlock(shmem->base.resv);
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_set_purgeable);
> +
> +static int
> +drm_gem_shmem_set_unpurgeable_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	if (shmem->madv < 0)
> +		return -ENOMEM;
> +
> +	shmem->purging_disable_count++;
> +
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +
> +	return 0;
> +}
> +
> +static int
> +drm_gem_shmem_set_purgeable_and_evictable_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	int ret;
> +
> +	ret = drm_gem_shmem_set_evictable_locked(shmem);
> +	if (!ret) {
> +		ret = drm_gem_shmem_set_purgeable_locked(shmem);
> +		if (ret)
> +			drm_gem_shmem_set_unevictable_locked(shmem);
> +	}
> +
> +	return ret;
> +}
> +
> +static int
> +drm_gem_shmem_set_unpurgeable_and_unevictable_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	int ret;
> +
> +	ret = drm_gem_shmem_set_unpurgeable_locked(shmem);
> +	if (!ret) {
> +		ret = drm_gem_shmem_set_unevictable_locked(shmem);
> +		if (ret)
> +			drm_gem_shmem_set_purgeable_locked(shmem);
> +	}
> +
> +	return ret;
> +}
> +
> +/**
> + * drm_gem_shmem_set_purgeable_and_evictable() - Make GEM unpurgeable and
> + * 						 unevictable by memory shrinker
> + * @shmem: shmem GEM object
> + *
> + * Tell memory shrinker that this GEM can't be purged and evicted. Each
> + * set_purgeable_and_evictable() call must have corresponding
> + * unpurgeable_and_unevictable() call. If GEM was purged, then -ENOMEM
> + * is returned.
> + *
> + * Returns:
> + * 0 on success or a negative error code on failure.
> + */
> +int drm_gem_shmem_set_purgeable_and_evictable(struct drm_gem_shmem_object *shmem)
> +{
> +	int ret;
> +
> +	dma_resv_lock(shmem->base.resv, NULL);
> +	ret = drm_gem_shmem_set_purgeable_and_evictable_locked(shmem);
> +	dma_resv_unlock(shmem->base.resv);
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_set_purgeable_and_evictable);
> +
> +/**
> + * drm_gem_shmem_set_unpurgeable_and_unevictable() - Make GEM purgeable and
> + * 						     evictable by memory shrinker
> + * @shmem: shmem GEM object
> + *
> + * Tell memory shrinker that this GEM can be purged and evicted. Each
> + * unpurgeable_and_unevictable() call must have corresponding
> + * set_purgeable_and_evictable() call. If GEM was purged, then -ENOMEM
> + * is returned.
> + *
> + * Returns:
> + * 0 on success or a negative error code on failure.
> + */
> +int drm_gem_shmem_set_unpurgeable_and_unevictable(struct drm_gem_shmem_object *shmem)
> +{
> +	int ret;
> +
> +	ret = dma_resv_lock_interruptible(shmem->base.resv, NULL);
> +	if (ret)
> +		return ret;
> +
> +	ret = drm_gem_shmem_set_unpurgeable_and_unevictable_locked(shmem);
> +	dma_resv_unlock(shmem->base.resv);
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_set_unpurgeable_and_unevictable);
> +
> +static int
> +drm_gem_shmem_acquire_pages_locked(struct drm_gem_shmem_object *shmem)
>   {
>   	struct drm_gem_object *obj = &shmem->base;
>   	struct page **pages;
>   
> -	if (shmem->pages_use_count++ > 0)
> +	if (shmem->madv < 0) {
> +		WARN_ON(shmem->pages);
> +		return -ENOMEM;
> +	}
> +
> +	if (shmem->pages) {
> +		WARN_ON(!shmem->evicted);
>   		return 0;
> +	}
>   
>   	pages = drm_gem_get_pages(obj);
>   	if (IS_ERR(pages)) {
>   		DRM_DEBUG_KMS("Failed to get pages (%ld)\n", PTR_ERR(pages));
> -		shmem->pages_use_count = 0;
>   		return PTR_ERR(pages);
>   	}
>   
> @@ -189,6 +508,25 @@ static int drm_gem_shmem_get_pages_locked(struct drm_gem_shmem_object *shmem)
>   	return 0;
>   }
>   
> +static int drm_gem_shmem_get_pages_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	int err;
> +
> +	if (shmem->madv < 0)
> +		return -ENOMEM;
> +
> +	if (shmem->pages_use_count++ > 0)
> +		return 0;
> +
> +	err = drm_gem_shmem_acquire_pages_locked(shmem);
> +	if (err) {
> +		shmem->pages_use_count = 0;
> +		return err;
> +	}
> +
> +	return 0;
> +}
> +
>   /*
>    * drm_gem_shmem_get_pages - Allocate backing pages for a shmem GEM object
>    * @shmem: shmem GEM object
> @@ -209,21 +547,38 @@ int drm_gem_shmem_get_pages(struct drm_gem_shmem_object *shmem)
>   	if (ret)
>   		return ret;
>   	ret = drm_gem_shmem_get_pages_locked(shmem);
> +
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +
>   	dma_resv_unlock(shmem->base.resv);
>   
>   	return ret;
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_get_pages);
>   
> -static void drm_gem_shmem_put_pages_locked(struct drm_gem_shmem_object *shmem)
> +static void drm_gem_shmem_get_pages_no_fail(struct drm_gem_shmem_object *shmem)
>   {
> -	struct drm_gem_object *obj = &shmem->base;
> +	WARN_ON(shmem->base.import_attach);
>   
> -	if (WARN_ON_ONCE(!shmem->pages_use_count))
> -		return;
> +	dma_resv_lock(shmem->base.resv, NULL);
>   
> -	if (--shmem->pages_use_count > 0)
> +	if (drm_gem_shmem_get_pages_locked(shmem))
> +		shmem->pages_use_count++;
> +
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +
> +	dma_resv_unlock(shmem->base.resv);
> +}
> +
> +static void
> +drm_gem_shmem_release_pages_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +
> +	if (!shmem->pages) {
> +		WARN_ON(!shmem->evicted && shmem->madv >= 0);
>   		return;
> +	}
>   
>   #ifdef CONFIG_X86
>   	if (shmem->map_wc)
> @@ -236,6 +591,21 @@ static void drm_gem_shmem_put_pages_locked(struct drm_gem_shmem_object *shmem)
>   	shmem->pages = NULL;
>   }
>   
> +static void drm_gem_shmem_put_pages_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +
> +	lockdep_assert_held(&obj->resv->lock.base);
> +
> +	if (WARN_ON(!shmem->pages_use_count))
> +		return;
> +
> +	if (--shmem->pages_use_count > 0)
> +		return;
> +
> +	drm_gem_shmem_release_pages_locked(shmem);
> +}
> +
>   /*
>    * drm_gem_shmem_put_pages - Decrease use count on the backing pages for a shmem GEM object
>    * @shmem: shmem GEM object
> @@ -246,6 +616,7 @@ void drm_gem_shmem_put_pages(struct drm_gem_shmem_object *shmem)
>   {
>   	dma_resv_lock(shmem->base.resv, NULL);
>   	drm_gem_shmem_put_pages_locked(shmem);
> +	drm_gem_shmem_update_pages_state_locked(shmem);
>   	dma_resv_unlock(shmem->base.resv);
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_put_pages);
> @@ -262,9 +633,21 @@ EXPORT_SYMBOL(drm_gem_shmem_put_pages);
>    */
>   int drm_gem_shmem_pin(struct drm_gem_shmem_object *shmem)
>   {
> +	int err;
> +
>   	WARN_ON(shmem->base.import_attach);
>   
> -	return drm_gem_shmem_get_pages(shmem);
> +	err = drm_gem_shmem_set_unpurgeable_and_unevictable(shmem);
> +	if (err)
> +		return err;
> +
> +	err = drm_gem_shmem_get_pages(shmem);
> +	if (err) {
> +		drm_gem_shmem_set_purgeable_and_evictable(shmem);
> +		return err;
> +	}
> +
> +	return 0;
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_pin);
>   
> @@ -280,6 +663,7 @@ void drm_gem_shmem_unpin(struct drm_gem_shmem_object *shmem)
>   	WARN_ON(shmem->base.import_attach);
>   
>   	drm_gem_shmem_put_pages(shmem);
> +	drm_gem_shmem_set_purgeable_and_evictable(shmem);
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_unpin);
>   
> @@ -359,7 +743,18 @@ int drm_gem_shmem_vmap(struct drm_gem_shmem_object *shmem,
>   	ret = dma_resv_lock_interruptible(shmem->base.resv, NULL);
>   	if (ret)
>   		return ret;
> +
> +	ret = drm_gem_shmem_set_unpurgeable_and_unevictable_locked(shmem);
> +	if (ret)
> +		goto unlock;
> +
>   	ret = drm_gem_shmem_vmap_locked(shmem, map);
> +	if (ret)
> +		drm_gem_shmem_set_purgeable_and_evictable_locked(shmem);
> +	else
> +		drm_gem_shmem_update_pages_state_locked(shmem);
> +
> +unlock:
>   	dma_resv_unlock(shmem->base.resv);
>   
>   	return ret;
> @@ -404,9 +799,9 @@ void drm_gem_shmem_vunmap(struct drm_gem_shmem_object *shmem,
>   {
>   	dma_resv_lock(shmem->base.resv, NULL);
>   	drm_gem_shmem_vunmap_locked(shmem, map);
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +	drm_gem_shmem_set_purgeable_and_evictable_locked(shmem);
>   	dma_resv_unlock(shmem->base.resv);
> -
> -	drm_gem_shmem_update_purgeable_status(shmem);
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_vunmap);
>   
> @@ -447,29 +842,140 @@ int drm_gem_shmem_madvise(struct drm_gem_shmem_object *shmem, int madv)
>   
>   	madv = shmem->madv;
>   
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +
>   	dma_resv_unlock(shmem->base.resv);
>   
>   	return (madv >= 0);
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_madvise);
>   
> -void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem)
> +/**
> + * drm_gem_shmem_swap_in_pages_locked() - Moves shmem pages back to memory
> + * @shmem: shmem GEM object
> + *
> + * This function moves pages back to memory if they were previously evicted
> + * by the memory shrinker.
> + *
> + * Returns:
> + * 0 on success or a negative error code on failure.
> + */
> +int drm_gem_shmem_swap_in_pages_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +	struct sg_table *sgt;
> +	int ret;
> +
> +	lockdep_assert_held(&obj->resv->lock.base);
> +
> +	if (shmem->evicted) {
> +		ret = drm_gem_shmem_acquire_pages_locked(shmem);
> +		if (ret)
> +			return ret;
> +
> +		sgt = drm_gem_shmem_get_sg_table(shmem);
> +		if (IS_ERR(sgt))
> +			return PTR_ERR(sgt);
> +
> +		ret = dma_map_sgtable(obj->dev->dev, sgt,
> +				      DMA_BIDIRECTIONAL, 0);
> +		if (ret) {
> +			sg_free_table(sgt);
> +			kfree(sgt);
> +			return ret;
> +		}
> +
> +		shmem->sgt = sgt;
> +		shmem->evicted = false;
> +		shmem->pages_state = DRM_GEM_SHMEM_PAGES_STATE_PINNED;
> +
> +		drm_gem_shmem_update_pages_state_locked(shmem);
> +	}
> +
> +	return shmem->pages ? 0 : -ENOMEM;
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_swap_in_pages_locked);
> +
> +/**
> + * drm_gem_shmem_swap_in_locked() - Moves shmem GEM back to memory
> + * @shmem: shmem GEM object
> + *
> + * This function moves shmem GEM back to memory if it was previously evicted
> + * by the memory shrinker. The GEM is ready to use on success.
> + *
> + * Returns:
> + * 0 on success or a negative error code on failure.
> + */
> +int drm_gem_shmem_swap_in_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +
> +	lockdep_assert_held(&obj->resv->lock.base);
> +
> +	if (shmem->evicted)
> +		return obj->funcs->swap_in(obj);
> +
> +	return 0;
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_swap_in_locked);
> +
> +static void drm_gem_shmem_unpin_pages_locked(struct drm_gem_shmem_object *shmem)
>   {
>   	struct drm_gem_object *obj = &shmem->base;
>   	struct drm_device *dev = obj->dev;
>   
> -	WARN_ON(!drm_gem_shmem_is_purgeable(shmem));
> +	if (shmem->evicted)
> +		return;
>   
>   	dma_unmap_sgtable(dev->dev, shmem->sgt, DMA_BIDIRECTIONAL, 0);
> +	drm_gem_shmem_release_pages_locked(shmem);
> +	drm_vma_node_unmap(&obj->vma_node, dev->anon_inode->i_mapping);
> +
>   	sg_free_table(shmem->sgt);
>   	kfree(shmem->sgt);
>   	shmem->sgt = NULL;
> +}
>   
> -	drm_gem_shmem_put_pages_locked(shmem);
> +/**
> + * drm_gem_shmem_evict_locked - Evict shmem pages
> + * @shmem: shmem GEM object
> + *
> + * This function unpins shmem pages, allowing them to be swapped out from
> + * memory.
> + */
> +void drm_gem_shmem_evict_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
>   
> -	shmem->madv = -1;
> +	lockdep_assert_held(&obj->resv->lock.base);
>   
> -	drm_vma_node_unmap(&obj->vma_node, dev->anon_inode->i_mapping);
> +	WARN_ON(!drm_gem_shmem_is_evictable(shmem));
> +	WARN_ON(shmem->madv < 0);
> +	WARN_ON(shmem->evicted);
> +
> +	drm_gem_shmem_unpin_pages_locked(shmem);
> +
> +	shmem->evicted = true;
> +	drm_gem_shmem_set_pages_state(shmem, DRM_GEM_SHMEM_PAGES_STATE_EVICTED);
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_evict_locked);
> +
> +/**
> + * drm_gem_shmem_purge_locked - Purge shmem pages
> + * @shmem: shmem GEM object
> + *
> + * This function permanently releases shmem pages.
> + */
> +void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +
> +	lockdep_assert_held(&obj->resv->lock.base);
> +
> +	WARN_ON(!drm_gem_shmem_is_purgeable(shmem));
> +	WARN_ON(shmem->madv < 0);
> +
> +	drm_gem_shmem_unpin_pages_locked(shmem);
>   	drm_gem_free_mmap_offset(obj);
>   
>   	/* Our goal here is to return as much of the memory as
> @@ -480,6 +986,9 @@ void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem)
>   	shmem_truncate_range(file_inode(obj->filp), 0, (loff_t)-1);
>   
>   	invalidate_mapping_pages(file_inode(obj->filp)->i_mapping, 0, (loff_t)-1);
> +
> +	shmem->madv = -1;
> +	drm_gem_shmem_set_pages_state(shmem, DRM_GEM_SHMEM_PAGES_STATE_PURGED);
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_purge_locked);
>   
> @@ -543,22 +1052,31 @@ static vm_fault_t drm_gem_shmem_fault(struct vm_fault *vmf)
>   	vm_fault_t ret;
>   	struct page *page;
>   	pgoff_t page_offset;
> +	bool pages_inactive;
> +	int err;
>   
>   	/* We don't use vmf->pgoff since that has the fake offset */
>   	page_offset = (vmf->address - vma->vm_start) >> PAGE_SHIFT;
>   
>   	dma_resv_lock(shmem->base.resv, NULL);
>   
> -	if (page_offset >= num_pages ||
> -	    WARN_ON_ONCE(!shmem->pages) ||
> -	    shmem->madv < 0) {
> +	pages_inactive = shmem->pages_state < DRM_GEM_SHMEM_PAGES_STATE_PINNED;
> +	WARN_ON_ONCE(!shmem->pages ^ pages_inactive);
> +
> +	if (page_offset >= num_pages || (!shmem->pages && !shmem->evicted)) {
>   		ret = VM_FAULT_SIGBUS;
>   	} else {
> +		err = drm_gem_shmem_swap_in_locked(shmem);
> +		if (err) {
> +			ret = VM_FAULT_OOM;
> +			goto unlock;
> +		}
> +
>   		page = shmem->pages[page_offset];
>   
>   		ret = vmf_insert_pfn(vma, vmf->address, page_to_pfn(page));
>   	}
> -
> +unlock:
>   	dma_resv_unlock(shmem->base.resv);
>   
>   	return ret;
> @@ -568,13 +1086,8 @@ static void drm_gem_shmem_vm_open(struct vm_area_struct *vma)
>   {
>   	struct drm_gem_object *obj = vma->vm_private_data;
>   	struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
> -	int ret;
> -
> -	WARN_ON(shmem->base.import_attach);
> -
> -	ret = drm_gem_shmem_get_pages(shmem);
> -	WARN_ON_ONCE(ret != 0);
>   
> +	drm_gem_shmem_get_pages_no_fail(shmem);
>   	drm_gem_vm_open(vma);
>   }
>   
> @@ -716,6 +1229,8 @@ struct sg_table *drm_gem_shmem_get_pages_sgt(struct drm_gem_shmem_object *shmem)
>   
>   	shmem->sgt = sgt;
>   
> +	drm_gem_shmem_update_pages_state(shmem);
> +
>   	return sgt;
>   
>   err_free_sgt:
> @@ -762,6 +1277,202 @@ drm_gem_shmem_prime_import_sg_table(struct drm_device *dev,
>   }
>   EXPORT_SYMBOL_GPL(drm_gem_shmem_prime_import_sg_table);
>   
> +static struct drm_gem_shmem_shrinker *
> +to_drm_shrinker(struct shrinker *shrinker)
> +{
> +	return container_of(shrinker, struct drm_gem_shmem_shrinker, base);
> +}
> +
> +static unsigned long
> +drm_gem_shmem_shrinker_count_objects(struct shrinker *shrinker,
> +				     struct shrink_control *sc)
> +{
> +	struct drm_gem_shmem_shrinker *gem_shrinker = to_drm_shrinker(shrinker);
> +	u64 count = READ_ONCE(gem_shrinker->shrinkable_count);
> +
> +	if (count >= SHRINK_EMPTY)
> +		return SHRINK_EMPTY - 1;
> +
> +	return count ?: SHRINK_EMPTY;
> +}
> +
> +static unsigned long
> +drm_gem_shmem_shrinker_run_objects_scan(struct shrinker *shrinker,
> +					unsigned long nr_to_scan,
> +					bool *lock_contention,
> +					bool evict)
> +{
> +	struct drm_gem_shmem_shrinker *gem_shrinker = to_drm_shrinker(shrinker);
> +	struct drm_gem_shmem_object *shmem;
> +	struct list_head still_in_list;
> +	struct drm_gem_object *obj;
> +	unsigned long freed = 0;
> +	struct list_head *lru;
> +	size_t page_count;
> +
> +	INIT_LIST_HEAD(&still_in_list);
> +
> +	mutex_lock(&gem_shrinker->lock);
> +
> +	if (evict)
> +		lru = &gem_shrinker->lru_evictable;
> +	else
> +		lru = &gem_shrinker->lru_purgeable;
> +
> +	while (freed < nr_to_scan) {
> +		shmem = list_first_entry_or_null(lru, typeof(*shmem), madv_list);
> +		if (!shmem)
> +			break;
> +
> +		obj = &shmem->base;
> +		page_count = obj->size >> PAGE_SHIFT;
> +		list_move_tail(&shmem->madv_list, &still_in_list);
> +
> +		if (evict && get_nr_swap_pages() < page_count)
> +			continue;
> +
> +		/*
> +		 * If it's in the process of being freed, gem_object->free()
> +		 * may be blocked on lock waiting to remove it.  So just
> +		 * skip it.
> +		 */
> +		if (!kref_get_unless_zero(&obj->refcount))
> +			continue;
> +
> +		mutex_unlock(&gem_shrinker->lock);
> +
> +		/* prevent racing with job-submission code paths */
> +		if (!dma_resv_trylock(obj->resv)) {
> +			*lock_contention |= true;
> +			goto shrinker_lock;
> +		}
> +
> +		/* prevent racing with the dma-buf exporting */
> +		if (!mutex_trylock(&gem_shrinker->dev->object_name_lock)) {
> +			*lock_contention |= true;
> +			goto resv_unlock;
> +		}
> +
> +		/* check whether h/w uses this object */
> +		if (!dma_resv_test_signaled(obj->resv, DMA_RESV_USAGE_WRITE))
> +			goto object_name_unlock;
> +
> +		/* GEM may've become unpurgeable while shrinker was unlocked */
> +		if (evict) {
> +			if (!drm_gem_shmem_is_evictable(shmem))
> +				goto object_name_unlock;
> +		} else {
> +			if (!drm_gem_shmem_is_purgeable(shmem))
> +				goto object_name_unlock;
> +		}
> +
> +		if (evict)
> +			freed += obj->funcs->evict(obj);
> +		else
> +			freed += obj->funcs->purge(obj);
> +object_name_unlock:
> +		mutex_unlock(&gem_shrinker->dev->object_name_lock);
> +resv_unlock:
> +		dma_resv_unlock(obj->resv);
> +shrinker_lock:
> +		drm_gem_object_put(&shmem->base);
> +		mutex_lock(&gem_shrinker->lock);
> +	}
> +
> +	list_splice_tail(&still_in_list, lru);
> +
> +	mutex_unlock(&gem_shrinker->lock);
> +
> +	return freed;
> +}
> +
> +static unsigned long
> +drm_gem_shmem_shrinker_scan_objects(struct shrinker *shrinker,
> +				    struct shrink_control *sc)
> +{
> +	unsigned long nr_to_scan = sc->nr_to_scan;
> +	bool lock_contention = false;
> +	unsigned long freed;
> +
> +	/* purge as many objects as we can */
> +	freed = drm_gem_shmem_shrinker_run_objects_scan(shrinker, nr_to_scan,
> +							&lock_contention, false);
> +	nr_to_scan -= freed;
> +
> +	/* evict as many objects as we can */
> +	if (freed < nr_to_scan)
> +		freed += drm_gem_shmem_shrinker_run_objects_scan(shrinker,
> +								 nr_to_scan,
> +								 &lock_contention,
> +								 true);
> +
> +	return (!freed && !lock_contention) ? SHRINK_STOP : freed;
> +}
> +
> +/**
> + * drm_gem_shmem_shrinker_register() - Register shmem shrinker
> + * @dev: DRM device
> + *
> + * Returns:
> + * 0 on success or a negative error code on failure.
> + */
> +int drm_gem_shmem_shrinker_register(struct drm_device *dev)
> +{
> +	struct drm_gem_shmem_shrinker *gem_shrinker;
> +	int err;
> +
> +	if (WARN_ON(dev->shmem_shrinker))
> +		return -EBUSY;
> +
> +	gem_shrinker = kzalloc(sizeof(*gem_shrinker), GFP_KERNEL);
> +	if (!gem_shrinker)
> +		return -ENOMEM;
> +
> +	gem_shrinker->base.count_objects = drm_gem_shmem_shrinker_count_objects;
> +	gem_shrinker->base.scan_objects = drm_gem_shmem_shrinker_scan_objects;
> +	gem_shrinker->base.seeks = DEFAULT_SEEKS;
> +	gem_shrinker->dev = dev;
> +
> +	INIT_LIST_HEAD(&gem_shrinker->lru_purgeable);
> +	INIT_LIST_HEAD(&gem_shrinker->lru_evictable);
> +	INIT_LIST_HEAD(&gem_shrinker->lru_evicted);
> +	INIT_LIST_HEAD(&gem_shrinker->lru_active);
> +	mutex_init(&gem_shrinker->lock);
> +
> +	dev->shmem_shrinker = gem_shrinker;
> +
> +	err = register_shrinker(&gem_shrinker->base);
> +	if (err) {
> +		dev->shmem_shrinker = NULL;
> +		kfree(gem_shrinker);
> +		return err;
> +	}
> +
> +	return 0;
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_shrinker_register);
> +
> +/**
> + * drm_gem_shmem_shrinker_unregister() - Unregister shmem shrinker
> + * @dev: DRM device
> + */
> +void drm_gem_shmem_shrinker_unregister(struct drm_device *dev)
> +{
> +	struct drm_gem_shmem_shrinker *gem_shrinker = dev->shmem_shrinker;
> +
> +	if (gem_shrinker) {
> +		unregister_shrinker(&gem_shrinker->base);
> +		WARN_ON(!list_empty(&gem_shrinker->lru_purgeable));
> +		WARN_ON(!list_empty(&gem_shrinker->lru_evictable));
> +		WARN_ON(!list_empty(&gem_shrinker->lru_evicted));
> +		WARN_ON(!list_empty(&gem_shrinker->lru_active));
> +		mutex_destroy(&gem_shrinker->lock);
> +		dev->shmem_shrinker = NULL;
> +		kfree(gem_shrinker);
> +	}
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_shrinker_unregister);
> +
>   MODULE_DESCRIPTION("DRM SHMEM memory-management helpers");
>   MODULE_IMPORT_NS(DMA_BUF);
>   MODULE_LICENSE("GPL v2");
> diff --git a/include/drm/drm_device.h b/include/drm/drm_device.h
> index 9923c7a6885e..929546cad894 100644
> --- a/include/drm/drm_device.h
> +++ b/include/drm/drm_device.h
> @@ -16,6 +16,7 @@ struct drm_vblank_crtc;
>   struct drm_vma_offset_manager;
>   struct drm_vram_mm;
>   struct drm_fb_helper;
> +struct drm_gem_shmem_shrinker;
>   
>   struct inode;
>   
> @@ -277,6 +278,9 @@ struct drm_device {
>   	/** @vram_mm: VRAM MM memory manager */
>   	struct drm_vram_mm *vram_mm;
>   
> +	/** @shmem_shrinker: SHMEM GEM memory shrinker */
> +	struct drm_gem_shmem_shrinker *shmem_shrinker;
> +
>   	/**
>   	 * @switch_power_state:
>   	 *
> diff --git a/include/drm/drm_gem.h b/include/drm/drm_gem.h
> index 9d7c61a122dc..390d1ce08ed3 100644
> --- a/include/drm/drm_gem.h
> +++ b/include/drm/drm_gem.h
> @@ -172,6 +172,41 @@ struct drm_gem_object_funcs {
>   	 * This is optional but necessary for mmap support.
>   	 */
>   	const struct vm_operations_struct *vm_ops;
> +
> +	/**
> +	 * @purge:
> +	 *
> +	 * Releases the GEM object's allocated backing storage to the system.
> +	 *
> +	 * Returns the number of pages that have been freed by purging the GEM object.
> +	 *
> +	 * This callback is used by the GEM shrinker.
> +	 */
> +	unsigned long (*purge)(struct drm_gem_object *obj);
> +
> +	/**
> +	 * @evict:
> +	 *
> +	 * Unpins the GEM object's allocated backing storage, allowing shmem pages
> +	 * to be swapped out.

What's the difference to the existing unpin() callback?

> +	 *
> +	 * Returns the number of pages that have been unpinned.
> +	 *
> +	 * This callback is used by the GEM shrinker.
> +	 */
> +	unsigned long (*evict)(struct drm_gem_object *obj);
> +
> +	/**
> +	 * @swap_in:
> +	 *
> +	 * Pins GEM object's allocated backing storage if it was previously evicted,
> +	 * moving swapped out pages back to memory.
> +	 *
> +	 * Returns 0 on success, or -errno on error.
> +	 *
> +	 * This callback is used by the GEM shrinker.
> +	 */
> +	int (*swap_in)(struct drm_gem_object *obj);

Why do you need swap_in()? This can be done on-demand as part of a pin 
or vmap operation.

>   };
>   
>   /**
> diff --git a/include/drm/drm_gem_shmem_helper.h b/include/drm/drm_gem_shmem_helper.h
> index 70889533962a..a65557b446e6 100644
> --- a/include/drm/drm_gem_shmem_helper.h
> +++ b/include/drm/drm_gem_shmem_helper.h
> @@ -6,6 +6,7 @@
>   #include <linux/fs.h>
>   #include <linux/mm.h>
>   #include <linux/mutex.h>
> +#include <linux/shrinker.h>
>   
>   #include <drm/drm_file.h>
>   #include <drm/drm_gem.h>
> @@ -15,8 +16,18 @@
>   struct dma_buf_attachment;
>   struct drm_mode_create_dumb;
>   struct drm_printer;
> +struct drm_device;
>   struct sg_table;
>   
> +enum drm_gem_shmem_pages_state {
> +	DRM_GEM_SHMEM_PAGES_STATE_PURGED = -2,
> +	DRM_GEM_SHMEM_PAGES_STATE_EVICTED = -1,
> +	DRM_GEM_SHMEM_PAGES_STATE_UNPINNED = 0,
> +	DRM_GEM_SHMEM_PAGES_STATE_PINNED = 1,
> +	DRM_GEM_SHMEM_PAGES_STATE_EVICTABLE = 2,
> +	DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE = 3,
> +};

These states can be detected by looking at the vmap and pin refcounts. 
No need to store them explicitly. In your patch, they also come with a 
big zoo of trivial helpers. None of that seems necessary AFAICT.

What's the difference between purge and evict BTW?

> +
>   /**
>    * struct drm_gem_shmem_object - GEM object backed by shmem
>    */
> @@ -43,8 +54,8 @@ struct drm_gem_shmem_object {
>   	 * @madv: State for madvise
>   	 *
>   	 * 0 is active/inuse.
> +	 * 1 is not-needed/can-be-purged
>   	 * A negative value is the object is purged.
> -	 * Positive values are driver specific and not used by the helpers.
>   	 */
>   	int madv;
>   
> @@ -91,6 +102,40 @@ struct drm_gem_shmem_object {
>   	 * @map_wc: map object write-combined (instead of using shmem defaults).
>   	 */
>   	bool map_wc;
> +
> +	/**
> +	 * @eviction_disable_count:
> +	 *
> +	 * The shmem pages are disallowed to be evicted by the memory shrinker
> +	 * while count is non-zero. Used internally by memory shrinker.
> +	 */
> +	unsigned int eviction_disable_count;
> +
> +	/**
> +	 * @purging_disable_count:
> +	 *
> +	 * The shmem pages are disallowed to be purged by the memory shrinker
> +	 * while count is non-zero. Used internally by memory shrinker.
> +	 */
> +	unsigned int purging_disable_count;
> +
> +	/**
> +	 * @pages_state: Current state of shmem pages. Used internally by
> +	 * memory shrinker.
> +	 */
> +	enum drm_gem_shmem_pages_state pages_state;
> +
> +	/**
> +	 * @evicted: True if shmem pages were evicted by the memory shrinker.
> +	 * Used internally by memory shrinker.
> +	 */
> +	bool evicted;
> +
> +	/**
> +	 * @pages_shrinkable: True if shmem pages can be evicted or purged
> +	 * by the memory shrinker. Used internally by memory shrinker.
> +	 */
> +	bool pages_shrinkable;

As commented before, this state can be foundby looking at existing 
fields. No need to store it separately.

Best regards
Thomas

>   };
>   
>   #define to_drm_gem_shmem_obj(obj) \
> @@ -111,15 +156,33 @@ int drm_gem_shmem_mmap(struct drm_gem_shmem_object *shmem, struct vm_area_struct
>   
>   int drm_gem_shmem_madvise(struct drm_gem_shmem_object *shmem, int madv);
>   
> +int drm_gem_shmem_set_purgeable(struct drm_gem_shmem_object *shmem);
> +int drm_gem_shmem_set_purgeable_and_evictable(struct drm_gem_shmem_object *shmem);
> +int drm_gem_shmem_set_unpurgeable_and_unevictable(struct drm_gem_shmem_object *shmem);
> +
> +static inline bool drm_gem_shmem_is_evictable(struct drm_gem_shmem_object *shmem)
> +{
> +	return (shmem->madv >= 0) && !shmem->eviction_disable_count &&
> +		shmem->base.funcs->evict && shmem->base.funcs->swap_in &&
> +		!shmem->vmap_use_count && !shmem->base.dma_buf &&
> +		!shmem->base.import_attach && shmem->sgt;
> +}
> +
>   static inline bool drm_gem_shmem_is_purgeable(struct drm_gem_shmem_object *shmem)
>   {
> -	return (shmem->madv > 0) &&
> -		!shmem->vmap_use_count && shmem->sgt &&
> -		!shmem->base.dma_buf && !shmem->base.import_attach;
> +	return (shmem->madv > 0) && !shmem->purging_disable_count &&
> +		!shmem->vmap_use_count && shmem->base.funcs->purge &&
> +		!shmem->base.dma_buf && !shmem->base.import_attach &&
> +		shmem->sgt;
>   }
>   
> -void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem);
> +int drm_gem_shmem_swap_in_pages_locked(struct drm_gem_shmem_object *shmem);
> +int drm_gem_shmem_swap_in_locked(struct drm_gem_shmem_object *shmem);
> +
> +void drm_gem_shmem_evict_locked(struct drm_gem_shmem_object *shmem);
> +
>   bool drm_gem_shmem_purge(struct drm_gem_shmem_object *shmem);
> +void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem);
>   
>   struct sg_table *drm_gem_shmem_get_sg_table(struct drm_gem_shmem_object *shmem);
>   struct sg_table *drm_gem_shmem_get_pages_sgt(struct drm_gem_shmem_object *shmem);
> @@ -262,6 +325,38 @@ static inline int drm_gem_shmem_object_mmap(struct drm_gem_object *obj, struct v
>   	return drm_gem_shmem_mmap(shmem, vma);
>   }
>   
> +/**
> + * struct drm_gem_shmem_shrinker - Generic memory shrinker for shmem GEMs
> + */
> +struct drm_gem_shmem_shrinker {
> +	/** @base: Shrinker for purging shmem GEM objects */
> +	struct shrinker base;
> +
> +	/** @lock: Protects @lru_* */
> +	struct mutex lock;
> +
> +	/** @lru_purgeable: List of shmem GEM objects available for purging */
> +	struct list_head lru_purgeable;
> +
> +	/** @lru_active: List of active shmem GEM objects */
> +	struct list_head lru_active;
> +
> +	/** @lru_evictable: List of shmem GEM objects that can be evicted */
> +	struct list_head lru_evictable;
> +
> +	/** @lru_evicted: List of evicted shmem GEM objects */
> +	struct list_head lru_evicted;
> +
> +	/** @dev: DRM device that uses this shrinker */
> +	struct drm_device *dev;
> +
> +	/** @shrinkable_count: Count of shmem GEM pages to be purged and evicted */
> +	u64 shrinkable_count;
> +};
> +
> +int drm_gem_shmem_shrinker_register(struct drm_device *dev);
> +void drm_gem_shmem_shrinker_unregister(struct drm_device *dev);
> +
>   /*
>    * Driver ops
>    */

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

[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]

[-- Attachment #2: Type: text/plain, Size: 183 bytes --]

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 11/15] drm/shmem-helper: Add generic memory shrinker

[Daniel Vetter]

On Tue, Apr 19, 2022 at 11:40:41PM +0300, Dmitry Osipenko wrote:
> On 4/19/22 10:22, Thomas Zimmermann wrote:
> > Hi
> > 
> > Am 18.04.22 um 00:37 schrieb Dmitry Osipenko:
> >> Introduce a common DRM SHMEM shrinker. It allows to reduce code
> >> duplication among DRM drivers that implement theirs own shrinkers.
> >> This is initial version of the shrinker that covers basic needs of
> >> GPU drivers, both purging and eviction of shmem objects are supported.
> >>
> >> This patch is based on a couple ideas borrowed from Rob's Clark MSM
> >> shrinker and Thomas' Zimmermann variant of SHMEM shrinker.
> >>
> >> In order to start using DRM SHMEM shrinker drivers should:
> >>
> >> 1. Implement new purge(), evict() + swap_in() GEM callbacks.
> >> 2. Register shrinker using drm_gem_shmem_shrinker_register(drm_device).
> >> 3. Use drm_gem_shmem_set_purgeable_and_evictable(shmem) and alike API
> >>     functions to activate shrinking of GEMs.
> >>
> >> Signed-off-by: Daniel Almeida <daniel.almeida@collabora.com>
> >> Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
> >> ---
> >>   drivers/gpu/drm/drm_gem_shmem_helper.c | 765 ++++++++++++++++++++++++-
> >>   include/drm/drm_device.h               |   4 +
> >>   include/drm/drm_gem.h                  |  35 ++
> >>   include/drm/drm_gem_shmem_helper.h     | 105 +++-
> >>   4 files changed, 877 insertions(+), 32 deletions(-)
> ...
> >> @@ -172,6 +172,41 @@ struct drm_gem_object_funcs {
> >>        * This is optional but necessary for mmap support.
> >>        */
> >>       const struct vm_operations_struct *vm_ops;
> >> +
> >> +    /**
> >> +     * @purge:
> >> +     *
> >> +     * Releases the GEM object's allocated backing storage to the
> >> system.
> >> +     *
> >> +     * Returns the number of pages that have been freed by purging
> >> the GEM object.
> >> +     *
> >> +     * This callback is used by the GEM shrinker.
> >> +     */
> >> +    unsigned long (*purge)(struct drm_gem_object *obj);

Hm I feel like drivers shouldn't need to know the difference here?

Like shmem helpers can track what's purgeable, and for eviction/purging
the driver callback should do the same?

The only difference is when we try to re-reserve the backing storage. When
the object has been evicted that should suceed, but when the object is
purged that will fail.

That's the difference between evict and purge for drivers?

> >> +
> >> +    /**
> >> +     * @evict:
> >> +     *
> >> +     * Unpins the GEM object's allocated backing storage, allowing
> >> shmem pages
> >> +     * to be swapped out.
> > 
> > What's the difference to the existing unpin() callback?
> 
> Drivers need to do more than just unpinning pages when GEMs are evicted.
> Unpinning is only a part of the eviction process. I'll improve the
> doc-comment in v5.
> 
> For example, for VirtIO-GPU driver we need to to detach host from the
> guest's memory before pages are evicted [1].
> 
> [1]
> https://gitlab.collabora.com/dmitry.osipenko/linux-kernel-rd/-/blob/932eb03198bce3a21353b09ab71e95f1c19b84c2/drivers/gpu/drm/virtio/virtgpu_object.c#L145
> 
> In case of Panfrost driver, we will need to remove mappings before pages
> are evicted.

It might be good to align this with ttm, otoh that all works quite a bit
differently for ttm since ttm supports buffer moves and a lot more fancy
stuff.

I'm bringing this up since I have this fancy idea that eventually we could
glue shmem helpers into ttm in some cases for managing buffers when they
sit in system memory (as opposed to vram).

> >> +     *
> >> +     * Returns the number of pages that have been unpinned.
> >> +     *
> >> +     * This callback is used by the GEM shrinker.
> >> +     */
> >> +    unsigned long (*evict)(struct drm_gem_object *obj);
> >> +
> >> +    /**
> >> +     * @swap_in:
> >> +     *
> >> +     * Pins GEM object's allocated backing storage if it was
> >> previously evicted,
> >> +     * moving swapped out pages back to memory.
> >> +     *
> >> +     * Returns 0 on success, or -errno on error.
> >> +     *
> >> +     * This callback is used by the GEM shrinker.
> >> +     */
> >> +    int (*swap_in)(struct drm_gem_object *obj);
> > 
> > Why do you need swap_in()? This can be done on-demand as part of a pin
> > or vmap operation.
> 
> Similarly to the unpinning, the pining of pages is only a part of what
> needs to be done for GPU drivers. Besides of returning pages back to
> memory, we also need to make them accessible to GPU and this is a
> driver-specific process. This why we need the additional callbacks.

This is a bit much midlayer. The way this works in ttm is you reserve all
the objects you need (which makes sure they're physically available
again), and then the driver goes through and makes sure the page tables
are all set up again.

Once you get towards gpu vm that's really the only approach, since your
swap_in has no idea for which vm it needs to restore pagetables (and
restoring it for all is a bit meh).

If drivers want to optimize this they can adjust/set any tracking
information from their evict callback as needed.

> 
> >>   };
> >>     /**
> >> diff --git a/include/drm/drm_gem_shmem_helper.h
> >> b/include/drm/drm_gem_shmem_helper.h
> >> index 70889533962a..a65557b446e6 100644
> >> --- a/include/drm/drm_gem_shmem_helper.h
> >> +++ b/include/drm/drm_gem_shmem_helper.h
> >> @@ -6,6 +6,7 @@
> >>   #include <linux/fs.h>
> >>   #include <linux/mm.h>
> >>   #include <linux/mutex.h>
> >> +#include <linux/shrinker.h>
> >>     #include <drm/drm_file.h>
> >>   #include <drm/drm_gem.h>
> >> @@ -15,8 +16,18 @@
> >>   struct dma_buf_attachment;
> >>   struct drm_mode_create_dumb;
> >>   struct drm_printer;
> >> +struct drm_device;
> >>   struct sg_table;
> >>   +enum drm_gem_shmem_pages_state {
> >> +    DRM_GEM_SHMEM_PAGES_STATE_PURGED = -2,
> >> +    DRM_GEM_SHMEM_PAGES_STATE_EVICTED = -1,
> >> +    DRM_GEM_SHMEM_PAGES_STATE_UNPINNED = 0,
> >> +    DRM_GEM_SHMEM_PAGES_STATE_PINNED = 1,
> >> +    DRM_GEM_SHMEM_PAGES_STATE_EVICTABLE = 2,
> >> +    DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE = 3,
> >> +};
> > 
> > These states can be detected by looking at the vmap and pin refcounts.
> > No need to store them explicitly.
> 
> I'll try to revisit this, but I was finding that it's much more
> difficult to follow and debug code without the explicit states.

purgeable/purged needs some state, but pinned shouldn't be duplicated, so
I concur here a bit.

> > In your patch, they also come with a
> > big zoo of trivial helpers. None of that seems necessary AFAICT.
> 
> There are couple functions which could be squashed, although this may
> hurt readability of the code a tad. I'll try to take another look at
> this for v5.
> 
> > What's the difference between purge and evict BTW?
> 
> The evicted pages are moved out from memory to a SWAP partition or file.
> 
> The purged pages are destroyed permanently.
> 
> >> +
> >>   /**
> >>    * struct drm_gem_shmem_object - GEM object backed by shmem
> >>    */
> >> @@ -43,8 +54,8 @@ struct drm_gem_shmem_object {
> >>        * @madv: State for madvise
> >>        *
> >>        * 0 is active/inuse.
> >> +     * 1 is not-needed/can-be-purged
> >>        * A negative value is the object is purged.
> >> -     * Positive values are driver specific and not used by the helpers.
> >>        */
> >>       int madv;
> >>   @@ -91,6 +102,40 @@ struct drm_gem_shmem_object {
> >>        * @map_wc: map object write-combined (instead of using shmem
> >> defaults).
> >>        */
> >>       bool map_wc;
> >> +
> >> +    /**
> >> +     * @eviction_disable_count:
> >> +     *
> >> +     * The shmem pages are disallowed to be evicted by the memory
> >> shrinker
> >> +     * while count is non-zero. Used internally by memory shrinker.
> >> +     */
> >> +    unsigned int eviction_disable_count;
> >> +
> >> +    /**
> >> +     * @purging_disable_count:
> >> +     *
> >> +     * The shmem pages are disallowed to be purged by the memory
> >> shrinker
> >> +     * while count is non-zero. Used internally by memory shrinker.
> >> +     */
> >> +    unsigned int purging_disable_count;

What are these disable counts for?

The way purgeable works in other drivers is that userspace sets purgeable
or not, and it's up to userspace to not make a mess of this.

There's also some interactions, and I guess a bunch of drivers get this
wrong in funny ways. Not sure how to best clean this up.

- Once you have a shrinker/dynamic memory management you should _not_ pin
  pages, except when it's truly permanent like for scanout. Instead
  drivers should attach dma_fence to the dma_resv to denote in-flight
  access.

- A pinned buffer object is not allowed to be put into purgeable state,
  and a bo in purgeable state should not be allowed to be pinned.

- Drivers need to hold dma_resv_lock for long enough in their command
  submission, i.e. from the point where the reserve the buffers and make
  sure that mappings exists, to the point where the request is submitted
  to hw or drm/sched and fences are installed.

But I think a lot of current shmem users just pin as part of execbuf, so
this won't work quite so well right out of the box.

Anyway with that design I don't think there should ever be a need to
disable shrinking.

> >> +
> >> +    /**
> >> +     * @pages_state: Current state of shmem pages. Used internally by
> >> +     * memory shrinker.
> >> +     */
> >> +    enum drm_gem_shmem_pages_state pages_state;
> >> +
> >> +    /**
> >> +     * @evicted: True if shmem pages were evicted by the memory
> >> shrinker.
> >> +     * Used internally by memory shrinker.
> >> +     */
> >> +    bool evicted;
> >> +
> >> +    /**
> >> +     * @pages_shrinkable: True if shmem pages can be evicted or purged
> >> +     * by the memory shrinker. Used internally by memory shrinker.
> >> +     */
> >> +    bool pages_shrinkable;
> > 
> > As commented before, this state can be foundby looking at existing
> > fields. No need to store it separately.
> 
> When we're transitioning from "evictable" to a "purgeable" state, we
> must not add pages twice to the "shrinkable_count" variable. Hence this
> is not a state, but a variable which prevents the double accounting of
> the pages. Please see drm_gem_shmem_add_pages_to_shrinker() in this patch.
> 
> Perhaps something like "pages_accounted_by_shrinker" could be a better
> name for the variable. I'll revisit this for v5.

Hm not sure we need to account this? Usually the shrinker just counts when
it's asked to do so, not practively maintain that count. Once you start
shrinking burning cpu time is generally not too terrible.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 11/15] drm/shmem-helper: Add generic memory shrinker

[Daniel Vetter]

On Thu, Apr 28, 2022 at 09:20:15PM +0300, Dmitry Osipenko wrote:
> 27.04.2022 18:03, Daniel Vetter wrote:
> >> ...
> >>>> @@ -172,6 +172,41 @@ struct drm_gem_object_funcs {
> >>>>        * This is optional but necessary for mmap support.
> >>>>        */
> >>>>       const struct vm_operations_struct *vm_ops;
> >>>> +
> >>>> +    /**
> >>>> +     * @purge:
> >>>> +     *
> >>>> +     * Releases the GEM object's allocated backing storage to the
> >>>> system.
> >>>> +     *
> >>>> +     * Returns the number of pages that have been freed by purging
> >>>> the GEM object.
> >>>> +     *
> >>>> +     * This callback is used by the GEM shrinker.
> >>>> +     */
> >>>> +    unsigned long (*purge)(struct drm_gem_object *obj);
> > 
> > Hm I feel like drivers shouldn't need to know the difference here?
> > 
> > Like shmem helpers can track what's purgeable, and for eviction/purging
> > the driver callback should do the same?
> > 
> > The only difference is when we try to re-reserve the backing storage. When
> > the object has been evicted that should suceed, but when the object is
> > purged that will fail.
> > 
> > That's the difference between evict and purge for drivers?
> 
> When buffer is purged, we can permanently release the backing storage
> and the reserved IOV space, re-using the freed space by new BOs.
> 
> When buffer is evicted, the BO's IOV should be kept reserved and the
> re-reservation of the backing storage should succeed.

Yeah but what's the difference for driver callbacks? In both cases the
driver callback needs to tear down gpu mappings and pagetables. The only
difference happens after that in the shmem helper: For purge we ditch the
shmem object, for evict we keep it. Drivers shouldn't need to care about
that difference, hence why the two callbacks?

> 
> >>>> +
> >>>> +    /**
> >>>> +     * @evict:
> >>>> +     *
> >>>> +     * Unpins the GEM object's allocated backing storage, allowing
> >>>> shmem pages
> >>>> +     * to be swapped out.
> >>>
> >>> What's the difference to the existing unpin() callback?
> >>
> >> Drivers need to do more than just unpinning pages when GEMs are evicted.
> >> Unpinning is only a part of the eviction process. I'll improve the
> >> doc-comment in v5.
> >>
> >> For example, for VirtIO-GPU driver we need to to detach host from the
> >> guest's memory before pages are evicted [1].
> >>
> >> [1]
> >> https://gitlab.collabora.com/dmitry.osipenko/linux-kernel-rd/-/blob/932eb03198bce3a21353b09ab71e95f1c19b84c2/drivers/gpu/drm/virtio/virtgpu_object.c#L145
> >>
> >> In case of Panfrost driver, we will need to remove mappings before pages
> >> are evicted.
> > 
> > It might be good to align this with ttm, otoh that all works quite a bit
> > differently for ttm since ttm supports buffer moves and a lot more fancy
> > stuff.
> > 
> > I'm bringing this up since I have this fancy idea that eventually we could
> > glue shmem helpers into ttm in some cases for managing buffers when they
> > sit in system memory (as opposed to vram).
> 
> I'll take a look at ttm for v6.
> 
> >>>> +     *
> >>>> +     * Returns the number of pages that have been unpinned.
> >>>> +     *
> >>>> +     * This callback is used by the GEM shrinker.
> >>>> +     */
> >>>> +    unsigned long (*evict)(struct drm_gem_object *obj);
> >>>> +
> >>>> +    /**
> >>>> +     * @swap_in:
> >>>> +     *
> >>>> +     * Pins GEM object's allocated backing storage if it was
> >>>> previously evicted,
> >>>> +     * moving swapped out pages back to memory.
> >>>> +     *
> >>>> +     * Returns 0 on success, or -errno on error.
> >>>> +     *
> >>>> +     * This callback is used by the GEM shrinker.
> >>>> +     */
> >>>> +    int (*swap_in)(struct drm_gem_object *obj);
> >>>
> >>> Why do you need swap_in()? This can be done on-demand as part of a pin
> >>> or vmap operation.
> >>
> >> Similarly to the unpinning, the pining of pages is only a part of what
> >> needs to be done for GPU drivers. Besides of returning pages back to
> >> memory, we also need to make them accessible to GPU and this is a
> >> driver-specific process. This why we need the additional callbacks.
> > 
> > This is a bit much midlayer. The way this works in ttm is you reserve all
> > the objects you need (which makes sure they're physically available
> > again), and then the driver goes through and makes sure the page tables
> > are all set up again.
> > 
> > Once you get towards gpu vm that's really the only approach, since your
> > swap_in has no idea for which vm it needs to restore pagetables (and
> > restoring it for all is a bit meh).
> > 
> > If drivers want to optimize this they can adjust/set any tracking
> > information from their evict callback as needed.
> 
> In practice, majority of BOs have only one mapping. Only shared BOs
> usually have extra mappings and shared BOs aren't evictable.

That seems like a fairly arbitrary limitations, and e.g. i915 doesn't have
this limitation, and also has a shrinker. I don't think it should be built
into the design.

> When memory pages are gone, then all the GPU mappings also should be
> gone. Perhaps it's indeed won't be a bad idea to move out the restoring
> of h/w VMs from the swap_in() and make drivers to handle the restoring
> by themselves, so swap_in() will be only about restoring the pages. I'll
> try to improve it in v6.

Sounds good.
-Daniel

> 
> >>>>   };
> >>>>     /**
> >>>> diff --git a/include/drm/drm_gem_shmem_helper.h
> >>>> b/include/drm/drm_gem_shmem_helper.h
> >>>> index 70889533962a..a65557b446e6 100644
> >>>> --- a/include/drm/drm_gem_shmem_helper.h
> >>>> +++ b/include/drm/drm_gem_shmem_helper.h
> >>>> @@ -6,6 +6,7 @@
> >>>>   #include <linux/fs.h>
> >>>>   #include <linux/mm.h>
> >>>>   #include <linux/mutex.h>
> >>>> +#include <linux/shrinker.h>
> >>>>     #include <drm/drm_file.h>
> >>>>   #include <drm/drm_gem.h>
> >>>> @@ -15,8 +16,18 @@
> >>>>   struct dma_buf_attachment;
> >>>>   struct drm_mode_create_dumb;
> >>>>   struct drm_printer;
> >>>> +struct drm_device;
> >>>>   struct sg_table;
> >>>>   +enum drm_gem_shmem_pages_state {
> >>>> +    DRM_GEM_SHMEM_PAGES_STATE_PURGED = -2,
> >>>> +    DRM_GEM_SHMEM_PAGES_STATE_EVICTED = -1,
> >>>> +    DRM_GEM_SHMEM_PAGES_STATE_UNPINNED = 0,
> >>>> +    DRM_GEM_SHMEM_PAGES_STATE_PINNED = 1,
> >>>> +    DRM_GEM_SHMEM_PAGES_STATE_EVICTABLE = 2,
> >>>> +    DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE = 3,
> >>>> +};
> >>>
> >>> These states can be detected by looking at the vmap and pin refcounts.
> >>> No need to store them explicitly.
> >>
> >> I'll try to revisit this, but I was finding that it's much more
> >> difficult to follow and debug code without the explicit states.
> > 
> > purgeable/purged needs some state, but pinned shouldn't be duplicated, so
> > I concur here a bit.
> > 
> >>> In your patch, they also come with a
> >>> big zoo of trivial helpers. None of that seems necessary AFAICT.
> >>
> >> There are couple functions which could be squashed, although this may
> >> hurt readability of the code a tad. I'll try to take another look at
> >> this for v5.
> >>
> >>> What's the difference between purge and evict BTW?
> >>
> >> The evicted pages are moved out from memory to a SWAP partition or file.
> >>
> >> The purged pages are destroyed permanently.
> >>
> >>>> +
> >>>>   /**
> >>>>    * struct drm_gem_shmem_object - GEM object backed by shmem
> >>>>    */
> >>>> @@ -43,8 +54,8 @@ struct drm_gem_shmem_object {
> >>>>        * @madv: State for madvise
> >>>>        *
> >>>>        * 0 is active/inuse.
> >>>> +     * 1 is not-needed/can-be-purged
> >>>>        * A negative value is the object is purged.
> >>>> -     * Positive values are driver specific and not used by the helpers.
> >>>>        */
> >>>>       int madv;
> >>>>   @@ -91,6 +102,40 @@ struct drm_gem_shmem_object {
> >>>>        * @map_wc: map object write-combined (instead of using shmem
> >>>> defaults).
> >>>>        */
> >>>>       bool map_wc;
> >>>> +
> >>>> +    /**
> >>>> +     * @eviction_disable_count:
> >>>> +     *
> >>>> +     * The shmem pages are disallowed to be evicted by the memory
> >>>> shrinker
> >>>> +     * while count is non-zero. Used internally by memory shrinker.
> >>>> +     */
> >>>> +    unsigned int eviction_disable_count;
> >>>> +
> >>>> +    /**
> >>>> +     * @purging_disable_count:
> >>>> +     *
> >>>> +     * The shmem pages are disallowed to be purged by the memory
> >>>> shrinker
> >>>> +     * while count is non-zero. Used internally by memory shrinker.
> >>>> +     */
> >>>> +    unsigned int purging_disable_count;
> > 
> > What are these disable counts for?
> 
> Some of BO types should stay pinned permanently, this applies to both
> VirtIO and Panfrost drivers that make use of the generic shrinker in
> this patchset. Hence I made objects unpurgeable and unevictable by default.
> 
> Initially the idea of these counts was to allow drivers to explicitly
> disable purging and eviction, and do it multiple times. If driver
> disables eviction in two different places in the code, then we need to
> track the eviction-disable count.
> 
> In the v5 of this patchset drivers don't need to explicitly disable
> shrinking anymore, they only need to enable it. The counts are also used
> internally by DRM SHMEM core to track the vmappings and pinnings, but
> perhaps pages_use_count could be used for that instead. I'll revisit it
> for v6.
> 
> > The way purgeable works in other drivers is that userspace sets purgeable
> > or not, and it's up to userspace to not make a mess of this.
> > 
> > There's also some interactions, and I guess a bunch of drivers get this
> > wrong in funny ways. Not sure how to best clean this up.
> > 
> > - Once you have a shrinker/dynamic memory management you should _not_ pin
> >   pages, except when it's truly permanent like for scanout. Instead
> >   drivers should attach dma_fence to the dma_resv to denote in-flight
> >   access.
> 
> By default pages are pinned when drm_gem_shmem_get_pages_sgt() is
> invoked by drivers during of BO creation time.
> 
> We could declare that pages_use_count=1 means the pages are allowed to
> be evicted and purged if shrinker is enabled. Then the further
> drm_gem_shmem_pin/vmap() calls will bump the pages_use_count,
> disallowing the eviction and purging, like you're suggesting, and we
> won't need the explicit counts.
> 
> > - A pinned buffer object is not allowed to be put into purgeable state,
> >   and a bo in purgeable state should not be allowed to be pinned.
> > 
> > - Drivers need to hold dma_resv_lock for long enough in their command
> >   submission, i.e. from the point where the reserve the buffers and make
> >   sure that mappings exists, to the point where the request is submitted
> >   to hw or drm/sched and fences are installed.
> > 
> > But I think a lot of current shmem users just pin as part of execbuf, so
> > this won't work quite so well right out of the box.
> 
> The current shmem users assume that BO is pinned permanently once it has
> been created.
> 
> > Anyway with that design I don't think there should ever be a need to
> > disable shrinking.
> 
> To me what you described mostly matches to what I did in the v5.
> 
> >>>> +
> >>>> +    /**
> >>>> +     * @pages_state: Current state of shmem pages. Used internally by
> >>>> +     * memory shrinker.
> >>>> +     */
> >>>> +    enum drm_gem_shmem_pages_state pages_state;
> >>>> +
> >>>> +    /**
> >>>> +     * @evicted: True if shmem pages were evicted by the memory
> >>>> shrinker.
> >>>> +     * Used internally by memory shrinker.
> >>>> +     */
> >>>> +    bool evicted;
> >>>> +
> >>>> +    /**
> >>>> +     * @pages_shrinkable: True if shmem pages can be evicted or purged
> >>>> +     * by the memory shrinker. Used internally by memory shrinker.
> >>>> +     */
> >>>> +    bool pages_shrinkable;
> >>>
> >>> As commented before, this state can be foundby looking at existing
> >>> fields. No need to store it separately.
> >>
> >> When we're transitioning from "evictable" to a "purgeable" state, we
> >> must not add pages twice to the "shrinkable_count" variable. Hence this
> >> is not a state, but a variable which prevents the double accounting of
> >> the pages. Please see drm_gem_shmem_add_pages_to_shrinker() in this patch.
> >>
> >> Perhaps something like "pages_accounted_by_shrinker" could be a better
> >> name for the variable. I'll revisit this for v5.
> > 
> > Hm not sure we need to account this? Usually the shrinker just counts when
> > it's asked to do so, not practively maintain that count. Once you start
> > shrinking burning cpu time is generally not too terrible.
> 
> We could count pages on demand by walking up the "evictable" list, but
> then the shrinker's lock needs to be taken by the
> drm_gem_shmem_shrinker_count_objects() to protect the list.
> 
> Previously Rob Clark said that the profiling of freedreno's shrinker
> showed that it's worthwhile to reduce the locks as much as possible,
> including the case of counting shrinkable objects.

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 11/15] drm/shmem-helper: Add generic memory shrinker

[Rob Clark]

On Thu, Apr 28, 2022 at 11:20 AM Dmitry Osipenko
<dmitry.osipenko@collabora.com> wrote:
>
> 27.04.2022 18:03, Daniel Vetter wrote:
> >> ...
> >>>> @@ -172,6 +172,41 @@ struct drm_gem_object_funcs {
> >>>>        * This is optional but necessary for mmap support.
> >>>>        */
> >>>>       const struct vm_operations_struct *vm_ops;
> >>>> +
> >>>> +    /**
> >>>> +     * @purge:
> >>>> +     *
> >>>> +     * Releases the GEM object's allocated backing storage to the
> >>>> system.
> >>>> +     *
> >>>> +     * Returns the number of pages that have been freed by purging
> >>>> the GEM object.
> >>>> +     *
> >>>> +     * This callback is used by the GEM shrinker.
> >>>> +     */
> >>>> +    unsigned long (*purge)(struct drm_gem_object *obj);
> >
> > Hm I feel like drivers shouldn't need to know the difference here?
> >
> > Like shmem helpers can track what's purgeable, and for eviction/purging
> > the driver callback should do the same?
> >
> > The only difference is when we try to re-reserve the backing storage. When
> > the object has been evicted that should suceed, but when the object is
> > purged that will fail.
> >
> > That's the difference between evict and purge for drivers?
>
> When buffer is purged, we can permanently release the backing storage
> and the reserved IOV space, re-using the freed space by new BOs.
>
> When buffer is evicted, the BO's IOV should be kept reserved and the
> re-reservation of the backing storage should succeed.
>
> >>>> +
> >>>> +    /**
> >>>> +     * @evict:
> >>>> +     *
> >>>> +     * Unpins the GEM object's allocated backing storage, allowing
> >>>> shmem pages
> >>>> +     * to be swapped out.
> >>>
> >>> What's the difference to the existing unpin() callback?
> >>
> >> Drivers need to do more than just unpinning pages when GEMs are evicted.
> >> Unpinning is only a part of the eviction process. I'll improve the
> >> doc-comment in v5.
> >>
> >> For example, for VirtIO-GPU driver we need to to detach host from the
> >> guest's memory before pages are evicted [1].
> >>
> >> [1]
> >> https://gitlab.collabora.com/dmitry.osipenko/linux-kernel-rd/-/blob/932eb03198bce3a21353b09ab71e95f1c19b84c2/drivers/gpu/drm/virtio/virtgpu_object.c#L145
> >>
> >> In case of Panfrost driver, we will need to remove mappings before pages
> >> are evicted.
> >
> > It might be good to align this with ttm, otoh that all works quite a bit
> > differently for ttm since ttm supports buffer moves and a lot more fancy
> > stuff.
> >
> > I'm bringing this up since I have this fancy idea that eventually we could
> > glue shmem helpers into ttm in some cases for managing buffers when they
> > sit in system memory (as opposed to vram).
>
> I'll take a look at ttm for v6.
>
> >>>> +     *
> >>>> +     * Returns the number of pages that have been unpinned.
> >>>> +     *
> >>>> +     * This callback is used by the GEM shrinker.
> >>>> +     */
> >>>> +    unsigned long (*evict)(struct drm_gem_object *obj);
> >>>> +
> >>>> +    /**
> >>>> +     * @swap_in:
> >>>> +     *
> >>>> +     * Pins GEM object's allocated backing storage if it was
> >>>> previously evicted,
> >>>> +     * moving swapped out pages back to memory.
> >>>> +     *
> >>>> +     * Returns 0 on success, or -errno on error.
> >>>> +     *
> >>>> +     * This callback is used by the GEM shrinker.
> >>>> +     */
> >>>> +    int (*swap_in)(struct drm_gem_object *obj);
> >>>
> >>> Why do you need swap_in()? This can be done on-demand as part of a pin
> >>> or vmap operation.
> >>
> >> Similarly to the unpinning, the pining of pages is only a part of what
> >> needs to be done for GPU drivers. Besides of returning pages back to
> >> memory, we also need to make them accessible to GPU and this is a
> >> driver-specific process. This why we need the additional callbacks.
> >
> > This is a bit much midlayer. The way this works in ttm is you reserve all
> > the objects you need (which makes sure they're physically available
> > again), and then the driver goes through and makes sure the page tables
> > are all set up again.
> >
> > Once you get towards gpu vm that's really the only approach, since your
> > swap_in has no idea for which vm it needs to restore pagetables (and
> > restoring it for all is a bit meh).
> >
> > If drivers want to optimize this they can adjust/set any tracking
> > information from their evict callback as needed.
>
> In practice, majority of BOs have only one mapping. Only shared BOs
> usually have extra mappings and shared BOs aren't evictable.
>
> When memory pages are gone, then all the GPU mappings also should be
> gone. Perhaps it's indeed won't be a bad idea to move out the restoring
> of h/w VMs from the swap_in() and make drivers to handle the restoring
> by themselves, so swap_in() will be only about restoring the pages. I'll
> try to improve it in v6.
>
> >>>>   };
> >>>>     /**
> >>>> diff --git a/include/drm/drm_gem_shmem_helper.h
> >>>> b/include/drm/drm_gem_shmem_helper.h
> >>>> index 70889533962a..a65557b446e6 100644
> >>>> --- a/include/drm/drm_gem_shmem_helper.h
> >>>> +++ b/include/drm/drm_gem_shmem_helper.h
> >>>> @@ -6,6 +6,7 @@
> >>>>   #include <linux/fs.h>
> >>>>   #include <linux/mm.h>
> >>>>   #include <linux/mutex.h>
> >>>> +#include <linux/shrinker.h>
> >>>>     #include <drm/drm_file.h>
> >>>>   #include <drm/drm_gem.h>
> >>>> @@ -15,8 +16,18 @@
> >>>>   struct dma_buf_attachment;
> >>>>   struct drm_mode_create_dumb;
> >>>>   struct drm_printer;
> >>>> +struct drm_device;
> >>>>   struct sg_table;
> >>>>   +enum drm_gem_shmem_pages_state {
> >>>> +    DRM_GEM_SHMEM_PAGES_STATE_PURGED = -2,
> >>>> +    DRM_GEM_SHMEM_PAGES_STATE_EVICTED = -1,
> >>>> +    DRM_GEM_SHMEM_PAGES_STATE_UNPINNED = 0,
> >>>> +    DRM_GEM_SHMEM_PAGES_STATE_PINNED = 1,
> >>>> +    DRM_GEM_SHMEM_PAGES_STATE_EVICTABLE = 2,
> >>>> +    DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE = 3,
> >>>> +};
> >>>
> >>> These states can be detected by looking at the vmap and pin refcounts.
> >>> No need to store them explicitly.
> >>
> >> I'll try to revisit this, but I was finding that it's much more
> >> difficult to follow and debug code without the explicit states.
> >
> > purgeable/purged needs some state, but pinned shouldn't be duplicated, so
> > I concur here a bit.
> >
> >>> In your patch, they also come with a
> >>> big zoo of trivial helpers. None of that seems necessary AFAICT.
> >>
> >> There are couple functions which could be squashed, although this may
> >> hurt readability of the code a tad. I'll try to take another look at
> >> this for v5.
> >>
> >>> What's the difference between purge and evict BTW?
> >>
> >> The evicted pages are moved out from memory to a SWAP partition or file.
> >>
> >> The purged pages are destroyed permanently.
> >>
> >>>> +
> >>>>   /**
> >>>>    * struct drm_gem_shmem_object - GEM object backed by shmem
> >>>>    */
> >>>> @@ -43,8 +54,8 @@ struct drm_gem_shmem_object {
> >>>>        * @madv: State for madvise
> >>>>        *
> >>>>        * 0 is active/inuse.
> >>>> +     * 1 is not-needed/can-be-purged
> >>>>        * A negative value is the object is purged.
> >>>> -     * Positive values are driver specific and not used by the helpers.
> >>>>        */
> >>>>       int madv;
> >>>>   @@ -91,6 +102,40 @@ struct drm_gem_shmem_object {
> >>>>        * @map_wc: map object write-combined (instead of using shmem
> >>>> defaults).
> >>>>        */
> >>>>       bool map_wc;
> >>>> +
> >>>> +    /**
> >>>> +     * @eviction_disable_count:
> >>>> +     *
> >>>> +     * The shmem pages are disallowed to be evicted by the memory
> >>>> shrinker
> >>>> +     * while count is non-zero. Used internally by memory shrinker.
> >>>> +     */
> >>>> +    unsigned int eviction_disable_count;
> >>>> +
> >>>> +    /**
> >>>> +     * @purging_disable_count:
> >>>> +     *
> >>>> +     * The shmem pages are disallowed to be purged by the memory
> >>>> shrinker
> >>>> +     * while count is non-zero. Used internally by memory shrinker.
> >>>> +     */
> >>>> +    unsigned int purging_disable_count;
> >
> > What are these disable counts for?
>
> Some of BO types should stay pinned permanently, this applies to both
> VirtIO and Panfrost drivers that make use of the generic shrinker in
> this patchset. Hence I made objects unpurgeable and unevictable by default.
>
> Initially the idea of these counts was to allow drivers to explicitly
> disable purging and eviction, and do it multiple times. If driver
> disables eviction in two different places in the code, then we need to
> track the eviction-disable count.
>
> In the v5 of this patchset drivers don't need to explicitly disable
> shrinking anymore, they only need to enable it. The counts are also used
> internally by DRM SHMEM core to track the vmappings and pinnings, but
> perhaps pages_use_count could be used for that instead. I'll revisit it
> for v6.
>
> > The way purgeable works in other drivers is that userspace sets purgeable
> > or not, and it's up to userspace to not make a mess of this.
> >
> > There's also some interactions, and I guess a bunch of drivers get this
> > wrong in funny ways. Not sure how to best clean this up.
> >
> > - Once you have a shrinker/dynamic memory management you should _not_ pin
> >   pages, except when it's truly permanent like for scanout. Instead
> >   drivers should attach dma_fence to the dma_resv to denote in-flight
> >   access.
>
> By default pages are pinned when drm_gem_shmem_get_pages_sgt() is
> invoked by drivers during of BO creation time.
>
> We could declare that pages_use_count=1 means the pages are allowed to
> be evicted and purged if shrinker is enabled. Then the further
> drm_gem_shmem_pin/vmap() calls will bump the pages_use_count,
> disallowing the eviction and purging, like you're suggesting, and we
> won't need the explicit counts.
>
> > - A pinned buffer object is not allowed to be put into purgeable state,
> >   and a bo in purgeable state should not be allowed to be pinned.
> >
> > - Drivers need to hold dma_resv_lock for long enough in their command
> >   submission, i.e. from the point where the reserve the buffers and make
> >   sure that mappings exists, to the point where the request is submitted
> >   to hw or drm/sched and fences are installed.
> >
> > But I think a lot of current shmem users just pin as part of execbuf, so
> > this won't work quite so well right out of the box.
>
> The current shmem users assume that BO is pinned permanently once it has
> been created.
>
> > Anyway with that design I don't think there should ever be a need to
> > disable shrinking.
>
> To me what you described mostly matches to what I did in the v5.
>
> >>>> +
> >>>> +    /**
> >>>> +     * @pages_state: Current state of shmem pages. Used internally by
> >>>> +     * memory shrinker.
> >>>> +     */
> >>>> +    enum drm_gem_shmem_pages_state pages_state;
> >>>> +
> >>>> +    /**
> >>>> +     * @evicted: True if shmem pages were evicted by the memory
> >>>> shrinker.
> >>>> +     * Used internally by memory shrinker.
> >>>> +     */
> >>>> +    bool evicted;
> >>>> +
> >>>> +    /**
> >>>> +     * @pages_shrinkable: True if shmem pages can be evicted or purged
> >>>> +     * by the memory shrinker. Used internally by memory shrinker.
> >>>> +     */
> >>>> +    bool pages_shrinkable;
> >>>
> >>> As commented before, this state can be foundby looking at existing
> >>> fields. No need to store it separately.
> >>
> >> When we're transitioning from "evictable" to a "purgeable" state, we
> >> must not add pages twice to the "shrinkable_count" variable. Hence this
> >> is not a state, but a variable which prevents the double accounting of
> >> the pages. Please see drm_gem_shmem_add_pages_to_shrinker() in this patch.
> >>
> >> Perhaps something like "pages_accounted_by_shrinker" could be a better
> >> name for the variable. I'll revisit this for v5.
> >
> > Hm not sure we need to account this? Usually the shrinker just counts when
> > it's asked to do so, not practively maintain that count. Once you start
> > shrinking burning cpu time is generally not too terrible.
>
> We could count pages on demand by walking up the "evictable" list, but
> then the shrinker's lock needs to be taken by the
> drm_gem_shmem_shrinker_count_objects() to protect the list.
>
> Previously Rob Clark said that the profiling of freedreno's shrinker
> showed that it's worthwhile to reduce the locks as much as possible,
> including the case of counting shrinkable objects.

Sorry I missed this earlier, but danvet is giving some bad advice here ;-)

You *really* need count_objects() to be lockless and fast, ie. no list
iteration.  It doesn't have to return the "perfect" value, so it is ok
if it is racy / not-atomic / etc.  Otherwise you will have bad system
performance issues when you start hitting do_shrink_slab() on many
threads at once.

BR,
-R
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 11/15] drm/shmem-helper: Add generic memory shrinker

[Thomas Zimmermann]

[-- Attachment #1.1.1: Type: text/plain, Size: 38598 bytes --]

Hi

Am 18.04.22 um 00:37 schrieb Dmitry Osipenko:
> Introduce a common DRM SHMEM shrinker. It allows to reduce code
> duplication among DRM drivers that implement theirs own shrinkers.
> This is initial version of the shrinker that covers basic needs of
> GPU drivers, both purging and eviction of shmem objects are supported.
> 
> This patch is based on a couple ideas borrowed from Rob's Clark MSM
> shrinker and Thomas' Zimmermann variant of SHMEM shrinker.
> 
> In order to start using DRM SHMEM shrinker drivers should:
> 
> 1. Implement new purge(), evict() + swap_in() GEM callbacks.
> 2. Register shrinker using drm_gem_shmem_shrinker_register(drm_device).
> 3. Use drm_gem_shmem_set_purgeable_and_evictable(shmem) and alike API
>     functions to activate shrinking of GEMs.

Honestly speaking, after reading the patch and the discussion here I 
really don't like where all tis is going. The interfaces and 
implementation are overengineered.  Descisions about evicting and 
purging should be done by the memory manager. For the most part, it's 
none of the driver's business.

I'd like to ask you to reduce the scope of the patchset and build the 
shrinker only for virtio-gpu. I know that I first suggested to build 
upon shmem helpers, but it seems that it's easier to do that in a later 
patchset.

Best regards
Thomas

> 
> Signed-off-by: Daniel Almeida <daniel.almeida@collabora.com>
> Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
> ---
>   drivers/gpu/drm/drm_gem_shmem_helper.c | 765 ++++++++++++++++++++++++-
>   include/drm/drm_device.h               |   4 +
>   include/drm/drm_gem.h                  |  35 ++
>   include/drm/drm_gem_shmem_helper.h     | 105 +++-
>   4 files changed, 877 insertions(+), 32 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_gem_shmem_helper.c b/drivers/gpu/drm/drm_gem_shmem_helper.c
> index 3ecef571eff3..3838fb8d6f3a 100644
> --- a/drivers/gpu/drm/drm_gem_shmem_helper.c
> +++ b/drivers/gpu/drm/drm_gem_shmem_helper.c
> @@ -88,6 +88,13 @@ __drm_gem_shmem_create(struct drm_device *dev, size_t size, bool private)
>   
>   	INIT_LIST_HEAD(&shmem->madv_list);
>   
> +	/*
> +	 * Eviction and purging are disabled by default, shmem user must enable
> +	 * them explicitly using drm_gem_shmem_set_evictable/purgeable().
> +	 */
> +	shmem->eviction_disable_count = 1;
> +	shmem->purging_disable_count = 1;
> +
>   	if (!private) {
>   		/*
>   		 * Our buffers are kept pinned, so allocating them
> @@ -126,6 +133,107 @@ struct drm_gem_shmem_object *drm_gem_shmem_create(struct drm_device *dev, size_t
>   }
>   EXPORT_SYMBOL_GPL(drm_gem_shmem_create);
>   
> +static void
> +drm_gem_shmem_add_pages_to_shrinker(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> +	size_t page_count = obj->size >> PAGE_SHIFT;
> +
> +	if (!shmem->pages_shrinkable) {
> +		WARN_ON(gem_shrinker->shrinkable_count + page_count < page_count);
> +		gem_shrinker->shrinkable_count += page_count;
> +		shmem->pages_shrinkable = true;
> +	}
> +}
> +
> +static void
> +drm_gem_shmem_remove_pages_from_shrinker(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> +	size_t page_count = obj->size >> PAGE_SHIFT;
> +
> +	if (shmem->pages_shrinkable) {
> +		WARN_ON(gem_shrinker->shrinkable_count < page_count);
> +		gem_shrinker->shrinkable_count -= page_count;
> +		shmem->pages_shrinkable = false;
> +	}
> +}
> +
> +static void
> +drm_gem_shmem_set_pages_state_locked(struct drm_gem_shmem_object *shmem,
> +				     enum drm_gem_shmem_pages_state new_state)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> +
> +	lockdep_assert_held(&gem_shrinker->lock);
> +	lockdep_assert_held(&obj->resv->lock.base);
> +
> +	if (new_state >= DRM_GEM_SHMEM_PAGES_STATE_PINNED) {
> +		if (drm_gem_shmem_is_evictable(shmem))
> +			new_state = DRM_GEM_SHMEM_PAGES_STATE_EVICTABLE;
> +
> +		if (drm_gem_shmem_is_purgeable(shmem))
> +			new_state = DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE;
> +
> +		if (!shmem->pages)
> +			new_state = DRM_GEM_SHMEM_PAGES_STATE_UNPINNED;
> +
> +		if (shmem->evicted)
> +			new_state = DRM_GEM_SHMEM_PAGES_STATE_EVICTED;
> +	}
> +
> +	if (shmem->pages_state == new_state)
> +		return;
> +
> +	switch (new_state) {
> +	case DRM_GEM_SHMEM_PAGES_STATE_UNPINNED:
> +	case DRM_GEM_SHMEM_PAGES_STATE_PURGED:
> +		drm_gem_shmem_remove_pages_from_shrinker(shmem);
> +		list_del_init(&shmem->madv_list);
> +		break;
> +
> +	case DRM_GEM_SHMEM_PAGES_STATE_PINNED:
> +		drm_gem_shmem_remove_pages_from_shrinker(shmem);
> +		list_move_tail(&shmem->madv_list, &gem_shrinker->lru_active);
> +		break;
> +
> +	case DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE:
> +		drm_gem_shmem_add_pages_to_shrinker(shmem);
> +		list_move_tail(&shmem->madv_list, &gem_shrinker->lru_purgeable);
> +		break;
> +
> +	case DRM_GEM_SHMEM_PAGES_STATE_EVICTABLE:
> +		drm_gem_shmem_add_pages_to_shrinker(shmem);
> +		list_move_tail(&shmem->madv_list, &gem_shrinker->lru_evictable);
> +		break;
> +
> +	case DRM_GEM_SHMEM_PAGES_STATE_EVICTED:
> +		drm_gem_shmem_remove_pages_from_shrinker(shmem);
> +		list_move_tail(&shmem->madv_list, &gem_shrinker->lru_evicted);
> +		break;
> +	}
> +
> +	shmem->pages_state = new_state;
> +}
> +
> +static void
> +drm_gem_shmem_set_pages_state(struct drm_gem_shmem_object *shmem,
> +			      enum drm_gem_shmem_pages_state new_state)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> +
> +	if (!gem_shrinker)
> +		return;
> +
> +	mutex_lock(&gem_shrinker->lock);
> +	drm_gem_shmem_set_pages_state_locked(shmem, new_state);
> +	mutex_unlock(&gem_shrinker->lock);
> +}
> +
>   /**
>    * drm_gem_shmem_free - Free resources associated with a shmem GEM object
>    * @shmem: shmem GEM object to free
> @@ -137,6 +245,9 @@ void drm_gem_shmem_free(struct drm_gem_shmem_object *shmem)
>   {
>   	struct drm_gem_object *obj = &shmem->base;
>   
> +	/* take out shmem GEM object from the memory shrinker */
> +	drm_gem_shmem_madvise(shmem, -1);
> +
>   	WARN_ON(shmem->vmap_use_count);
>   
>   	if (obj->import_attach) {
> @@ -148,7 +259,7 @@ void drm_gem_shmem_free(struct drm_gem_shmem_object *shmem)
>   			sg_free_table(shmem->sgt);
>   			kfree(shmem->sgt);
>   		}
> -		if (shmem->pages)
> +		if (shmem->pages_use_count)
>   			drm_gem_shmem_put_pages(shmem);
>   	}
>   
> @@ -159,18 +270,226 @@ void drm_gem_shmem_free(struct drm_gem_shmem_object *shmem)
>   }
>   EXPORT_SYMBOL_GPL(drm_gem_shmem_free);
>   
> -static int drm_gem_shmem_get_pages_locked(struct drm_gem_shmem_object *shmem)
> +static void drm_gem_shmem_update_pages_state_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> +	enum drm_gem_shmem_pages_state new_state;
> +
> +	if (!gem_shrinker || obj->import_attach)
> +		return;
> +
> +	mutex_lock(&gem_shrinker->lock);
> +
> +	if (!shmem->madv)
> +		new_state = DRM_GEM_SHMEM_PAGES_STATE_PINNED;
> +	else if (shmem->madv > 0)
> +		new_state = DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE;
> +	else
> +		new_state = DRM_GEM_SHMEM_PAGES_STATE_PURGED;
> +
> +	drm_gem_shmem_set_pages_state_locked(shmem, new_state);
> +
> +	mutex_unlock(&gem_shrinker->lock);
> +}
> +
> +static void drm_gem_shmem_update_pages_state(struct drm_gem_shmem_object *shmem)
> +{
> +	dma_resv_lock(shmem->base.resv, NULL);
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +	dma_resv_unlock(shmem->base.resv);
> +}
> +
> +static int
> +drm_gem_shmem_set_evictable_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	int ret = 0;
> +
> +	WARN_ON_ONCE(!shmem->eviction_disable_count--);
> +
> +	if (shmem->madv < 0)
> +		ret = -ENOMEM;
> +
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +
> +	return ret;
> +}
> +
> +static int
> +drm_gem_shmem_set_unevictable_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +	int err;
> +
> +	if (shmem->madv < 0)
> +		return -ENOMEM;
> +
> +	if (shmem->evicted) {
> +		err = obj->funcs->swap_in(obj);
> +		if (err)
> +			return err;
> +	}
> +
> +	shmem->eviction_disable_count++;
> +
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +
> +	return 0;
> +}
> +
> +static int
> +drm_gem_shmem_set_purgeable_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	int ret = 0;
> +
> +	WARN_ON_ONCE(!shmem->purging_disable_count--);
> +
> +	if (shmem->madv < 0)
> +		ret = -ENOMEM;
> +
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +
> +	return ret;
> +}
> +
> +/**
> + * drm_gem_shmem_set_purgeable() - Make GEM purgeable by memory shrinker
> + * @shmem: shmem GEM object
> + *
> + * Tell memory shrinker that this GEM can be purged. Initially purging is
> + * disabled for all GEMs. Each set_pureable() call must have corresponding
> + * set_unpureable() call. If GEM was purged, then -ENOMEM is returned.
> + *
> + * Returns:
> + * 0 on success or a negative error code on failure.
> + */
> +int drm_gem_shmem_set_purgeable(struct drm_gem_shmem_object *shmem)
> +{
> +	int ret;
> +
> +	dma_resv_lock(shmem->base.resv, NULL);
> +	ret = drm_gem_shmem_set_purgeable_locked(shmem);
> +	dma_resv_unlock(shmem->base.resv);
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_set_purgeable);
> +
> +static int
> +drm_gem_shmem_set_unpurgeable_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	if (shmem->madv < 0)
> +		return -ENOMEM;
> +
> +	shmem->purging_disable_count++;
> +
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +
> +	return 0;
> +}
> +
> +static int
> +drm_gem_shmem_set_purgeable_and_evictable_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	int ret;
> +
> +	ret = drm_gem_shmem_set_evictable_locked(shmem);
> +	if (!ret) {
> +		ret = drm_gem_shmem_set_purgeable_locked(shmem);
> +		if (ret)
> +			drm_gem_shmem_set_unevictable_locked(shmem);
> +	}
> +
> +	return ret;
> +}
> +
> +static int
> +drm_gem_shmem_set_unpurgeable_and_unevictable_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	int ret;
> +
> +	ret = drm_gem_shmem_set_unpurgeable_locked(shmem);
> +	if (!ret) {
> +		ret = drm_gem_shmem_set_unevictable_locked(shmem);
> +		if (ret)
> +			drm_gem_shmem_set_purgeable_locked(shmem);
> +	}
> +
> +	return ret;
> +}
> +
> +/**
> + * drm_gem_shmem_set_purgeable_and_evictable() - Make GEM unpurgeable and
> + * 						 unevictable by memory shrinker
> + * @shmem: shmem GEM object
> + *
> + * Tell memory shrinker that this GEM can't be purged and evicted. Each
> + * set_purgeable_and_evictable() call must have corresponding
> + * unpurgeable_and_unevictable() call. If GEM was purged, then -ENOMEM
> + * is returned.
> + *
> + * Returns:
> + * 0 on success or a negative error code on failure.
> + */
> +int drm_gem_shmem_set_purgeable_and_evictable(struct drm_gem_shmem_object *shmem)
> +{
> +	int ret;
> +
> +	dma_resv_lock(shmem->base.resv, NULL);
> +	ret = drm_gem_shmem_set_purgeable_and_evictable_locked(shmem);
> +	dma_resv_unlock(shmem->base.resv);
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_set_purgeable_and_evictable);
> +
> +/**
> + * drm_gem_shmem_set_unpurgeable_and_unevictable() - Make GEM purgeable and
> + * 						     evictable by memory shrinker
> + * @shmem: shmem GEM object
> + *
> + * Tell memory shrinker that this GEM can be purged and evicted. Each
> + * unpurgeable_and_unevictable() call must have corresponding
> + * set_purgeable_and_evictable() call. If GEM was purged, then -ENOMEM
> + * is returned.
> + *
> + * Returns:
> + * 0 on success or a negative error code on failure.
> + */
> +int drm_gem_shmem_set_unpurgeable_and_unevictable(struct drm_gem_shmem_object *shmem)
> +{
> +	int ret;
> +
> +	ret = dma_resv_lock_interruptible(shmem->base.resv, NULL);
> +	if (ret)
> +		return ret;
> +
> +	ret = drm_gem_shmem_set_unpurgeable_and_unevictable_locked(shmem);
> +	dma_resv_unlock(shmem->base.resv);
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_set_unpurgeable_and_unevictable);
> +
> +static int
> +drm_gem_shmem_acquire_pages_locked(struct drm_gem_shmem_object *shmem)
>   {
>   	struct drm_gem_object *obj = &shmem->base;
>   	struct page **pages;
>   
> -	if (shmem->pages_use_count++ > 0)
> +	if (shmem->madv < 0) {
> +		WARN_ON(shmem->pages);
> +		return -ENOMEM;
> +	}
> +
> +	if (shmem->pages) {
> +		WARN_ON(!shmem->evicted);
>   		return 0;
> +	}
>   
>   	pages = drm_gem_get_pages(obj);
>   	if (IS_ERR(pages)) {
>   		DRM_DEBUG_KMS("Failed to get pages (%ld)\n", PTR_ERR(pages));
> -		shmem->pages_use_count = 0;
>   		return PTR_ERR(pages);
>   	}
>   
> @@ -189,6 +508,25 @@ static int drm_gem_shmem_get_pages_locked(struct drm_gem_shmem_object *shmem)
>   	return 0;
>   }
>   
> +static int drm_gem_shmem_get_pages_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	int err;
> +
> +	if (shmem->madv < 0)
> +		return -ENOMEM;
> +
> +	if (shmem->pages_use_count++ > 0)
> +		return 0;
> +
> +	err = drm_gem_shmem_acquire_pages_locked(shmem);
> +	if (err) {
> +		shmem->pages_use_count = 0;
> +		return err;
> +	}
> +
> +	return 0;
> +}
> +
>   /*
>    * drm_gem_shmem_get_pages - Allocate backing pages for a shmem GEM object
>    * @shmem: shmem GEM object
> @@ -209,21 +547,38 @@ int drm_gem_shmem_get_pages(struct drm_gem_shmem_object *shmem)
>   	if (ret)
>   		return ret;
>   	ret = drm_gem_shmem_get_pages_locked(shmem);
> +
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +
>   	dma_resv_unlock(shmem->base.resv);
>   
>   	return ret;
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_get_pages);
>   
> -static void drm_gem_shmem_put_pages_locked(struct drm_gem_shmem_object *shmem)
> +static void drm_gem_shmem_get_pages_no_fail(struct drm_gem_shmem_object *shmem)
>   {
> -	struct drm_gem_object *obj = &shmem->base;
> +	WARN_ON(shmem->base.import_attach);
>   
> -	if (WARN_ON_ONCE(!shmem->pages_use_count))
> -		return;
> +	dma_resv_lock(shmem->base.resv, NULL);
>   
> -	if (--shmem->pages_use_count > 0)
> +	if (drm_gem_shmem_get_pages_locked(shmem))
> +		shmem->pages_use_count++;
> +
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +
> +	dma_resv_unlock(shmem->base.resv);
> +}
> +
> +static void
> +drm_gem_shmem_release_pages_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +
> +	if (!shmem->pages) {
> +		WARN_ON(!shmem->evicted && shmem->madv >= 0);
>   		return;
> +	}
>   
>   #ifdef CONFIG_X86
>   	if (shmem->map_wc)
> @@ -236,6 +591,21 @@ static void drm_gem_shmem_put_pages_locked(struct drm_gem_shmem_object *shmem)
>   	shmem->pages = NULL;
>   }
>   
> +static void drm_gem_shmem_put_pages_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +
> +	lockdep_assert_held(&obj->resv->lock.base);
> +
> +	if (WARN_ON(!shmem->pages_use_count))
> +		return;
> +
> +	if (--shmem->pages_use_count > 0)
> +		return;
> +
> +	drm_gem_shmem_release_pages_locked(shmem);
> +}
> +
>   /*
>    * drm_gem_shmem_put_pages - Decrease use count on the backing pages for a shmem GEM object
>    * @shmem: shmem GEM object
> @@ -246,6 +616,7 @@ void drm_gem_shmem_put_pages(struct drm_gem_shmem_object *shmem)
>   {
>   	dma_resv_lock(shmem->base.resv, NULL);
>   	drm_gem_shmem_put_pages_locked(shmem);
> +	drm_gem_shmem_update_pages_state_locked(shmem);
>   	dma_resv_unlock(shmem->base.resv);
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_put_pages);
> @@ -262,9 +633,21 @@ EXPORT_SYMBOL(drm_gem_shmem_put_pages);
>    */
>   int drm_gem_shmem_pin(struct drm_gem_shmem_object *shmem)
>   {
> +	int err;
> +
>   	WARN_ON(shmem->base.import_attach);
>   
> -	return drm_gem_shmem_get_pages(shmem);
> +	err = drm_gem_shmem_set_unpurgeable_and_unevictable(shmem);
> +	if (err)
> +		return err;
> +
> +	err = drm_gem_shmem_get_pages(shmem);
> +	if (err) {
> +		drm_gem_shmem_set_purgeable_and_evictable(shmem);
> +		return err;
> +	}
> +
> +	return 0;
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_pin);
>   
> @@ -280,6 +663,7 @@ void drm_gem_shmem_unpin(struct drm_gem_shmem_object *shmem)
>   	WARN_ON(shmem->base.import_attach);
>   
>   	drm_gem_shmem_put_pages(shmem);
> +	drm_gem_shmem_set_purgeable_and_evictable(shmem);
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_unpin);
>   
> @@ -359,7 +743,18 @@ int drm_gem_shmem_vmap(struct drm_gem_shmem_object *shmem,
>   	ret = dma_resv_lock_interruptible(shmem->base.resv, NULL);
>   	if (ret)
>   		return ret;
> +
> +	ret = drm_gem_shmem_set_unpurgeable_and_unevictable_locked(shmem);
> +	if (ret)
> +		goto unlock;
> +
>   	ret = drm_gem_shmem_vmap_locked(shmem, map);
> +	if (ret)
> +		drm_gem_shmem_set_purgeable_and_evictable_locked(shmem);
> +	else
> +		drm_gem_shmem_update_pages_state_locked(shmem);
> +
> +unlock:
>   	dma_resv_unlock(shmem->base.resv);
>   
>   	return ret;
> @@ -404,9 +799,9 @@ void drm_gem_shmem_vunmap(struct drm_gem_shmem_object *shmem,
>   {
>   	dma_resv_lock(shmem->base.resv, NULL);
>   	drm_gem_shmem_vunmap_locked(shmem, map);
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +	drm_gem_shmem_set_purgeable_and_evictable_locked(shmem);
>   	dma_resv_unlock(shmem->base.resv);
> -
> -	drm_gem_shmem_update_purgeable_status(shmem);
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_vunmap);
>   
> @@ -447,29 +842,140 @@ int drm_gem_shmem_madvise(struct drm_gem_shmem_object *shmem, int madv)
>   
>   	madv = shmem->madv;
>   
> +	drm_gem_shmem_update_pages_state_locked(shmem);
> +
>   	dma_resv_unlock(shmem->base.resv);
>   
>   	return (madv >= 0);
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_madvise);
>   
> -void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem)
> +/**
> + * drm_gem_shmem_swap_in_pages_locked() - Moves shmem pages back to memory
> + * @shmem: shmem GEM object
> + *
> + * This function moves pages back to memory if they were previously evicted
> + * by the memory shrinker.
> + *
> + * Returns:
> + * 0 on success or a negative error code on failure.
> + */
> +int drm_gem_shmem_swap_in_pages_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +	struct sg_table *sgt;
> +	int ret;
> +
> +	lockdep_assert_held(&obj->resv->lock.base);
> +
> +	if (shmem->evicted) {
> +		ret = drm_gem_shmem_acquire_pages_locked(shmem);
> +		if (ret)
> +			return ret;
> +
> +		sgt = drm_gem_shmem_get_sg_table(shmem);
> +		if (IS_ERR(sgt))
> +			return PTR_ERR(sgt);
> +
> +		ret = dma_map_sgtable(obj->dev->dev, sgt,
> +				      DMA_BIDIRECTIONAL, 0);
> +		if (ret) {
> +			sg_free_table(sgt);
> +			kfree(sgt);
> +			return ret;
> +		}
> +
> +		shmem->sgt = sgt;
> +		shmem->evicted = false;
> +		shmem->pages_state = DRM_GEM_SHMEM_PAGES_STATE_PINNED;
> +
> +		drm_gem_shmem_update_pages_state_locked(shmem);
> +	}
> +
> +	return shmem->pages ? 0 : -ENOMEM;
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_swap_in_pages_locked);
> +
> +/**
> + * drm_gem_shmem_swap_in_locked() - Moves shmem GEM back to memory
> + * @shmem: shmem GEM object
> + *
> + * This function moves shmem GEM back to memory if it was previously evicted
> + * by the memory shrinker. The GEM is ready to use on success.
> + *
> + * Returns:
> + * 0 on success or a negative error code on failure.
> + */
> +int drm_gem_shmem_swap_in_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +
> +	lockdep_assert_held(&obj->resv->lock.base);
> +
> +	if (shmem->evicted)
> +		return obj->funcs->swap_in(obj);
> +
> +	return 0;
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_swap_in_locked);
> +
> +static void drm_gem_shmem_unpin_pages_locked(struct drm_gem_shmem_object *shmem)
>   {
>   	struct drm_gem_object *obj = &shmem->base;
>   	struct drm_device *dev = obj->dev;
>   
> -	WARN_ON(!drm_gem_shmem_is_purgeable(shmem));
> +	if (shmem->evicted)
> +		return;
>   
>   	dma_unmap_sgtable(dev->dev, shmem->sgt, DMA_BIDIRECTIONAL, 0);
> +	drm_gem_shmem_release_pages_locked(shmem);
> +	drm_vma_node_unmap(&obj->vma_node, dev->anon_inode->i_mapping);
> +
>   	sg_free_table(shmem->sgt);
>   	kfree(shmem->sgt);
>   	shmem->sgt = NULL;
> +}
>   
> -	drm_gem_shmem_put_pages_locked(shmem);
> +/**
> + * drm_gem_shmem_evict_locked - Evict shmem pages
> + * @shmem: shmem GEM object
> + *
> + * This function unpins shmem pages, allowing them to be swapped out from
> + * memory.
> + */
> +void drm_gem_shmem_evict_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
>   
> -	shmem->madv = -1;
> +	lockdep_assert_held(&obj->resv->lock.base);
>   
> -	drm_vma_node_unmap(&obj->vma_node, dev->anon_inode->i_mapping);
> +	WARN_ON(!drm_gem_shmem_is_evictable(shmem));
> +	WARN_ON(shmem->madv < 0);
> +	WARN_ON(shmem->evicted);
> +
> +	drm_gem_shmem_unpin_pages_locked(shmem);
> +
> +	shmem->evicted = true;
> +	drm_gem_shmem_set_pages_state(shmem, DRM_GEM_SHMEM_PAGES_STATE_EVICTED);
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_evict_locked);
> +
> +/**
> + * drm_gem_shmem_purge_locked - Purge shmem pages
> + * @shmem: shmem GEM object
> + *
> + * This function permanently releases shmem pages.
> + */
> +void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem)
> +{
> +	struct drm_gem_object *obj = &shmem->base;
> +
> +	lockdep_assert_held(&obj->resv->lock.base);
> +
> +	WARN_ON(!drm_gem_shmem_is_purgeable(shmem));
> +	WARN_ON(shmem->madv < 0);
> +
> +	drm_gem_shmem_unpin_pages_locked(shmem);
>   	drm_gem_free_mmap_offset(obj);
>   
>   	/* Our goal here is to return as much of the memory as
> @@ -480,6 +986,9 @@ void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem)
>   	shmem_truncate_range(file_inode(obj->filp), 0, (loff_t)-1);
>   
>   	invalidate_mapping_pages(file_inode(obj->filp)->i_mapping, 0, (loff_t)-1);
> +
> +	shmem->madv = -1;
> +	drm_gem_shmem_set_pages_state(shmem, DRM_GEM_SHMEM_PAGES_STATE_PURGED);
>   }
>   EXPORT_SYMBOL(drm_gem_shmem_purge_locked);
>   
> @@ -543,22 +1052,31 @@ static vm_fault_t drm_gem_shmem_fault(struct vm_fault *vmf)
>   	vm_fault_t ret;
>   	struct page *page;
>   	pgoff_t page_offset;
> +	bool pages_inactive;
> +	int err;
>   
>   	/* We don't use vmf->pgoff since that has the fake offset */
>   	page_offset = (vmf->address - vma->vm_start) >> PAGE_SHIFT;
>   
>   	dma_resv_lock(shmem->base.resv, NULL);
>   
> -	if (page_offset >= num_pages ||
> -	    WARN_ON_ONCE(!shmem->pages) ||
> -	    shmem->madv < 0) {
> +	pages_inactive = shmem->pages_state < DRM_GEM_SHMEM_PAGES_STATE_PINNED;
> +	WARN_ON_ONCE(!shmem->pages ^ pages_inactive);
> +
> +	if (page_offset >= num_pages || (!shmem->pages && !shmem->evicted)) {
>   		ret = VM_FAULT_SIGBUS;
>   	} else {
> +		err = drm_gem_shmem_swap_in_locked(shmem);
> +		if (err) {
> +			ret = VM_FAULT_OOM;
> +			goto unlock;
> +		}
> +
>   		page = shmem->pages[page_offset];
>   
>   		ret = vmf_insert_pfn(vma, vmf->address, page_to_pfn(page));
>   	}
> -
> +unlock:
>   	dma_resv_unlock(shmem->base.resv);
>   
>   	return ret;
> @@ -568,13 +1086,8 @@ static void drm_gem_shmem_vm_open(struct vm_area_struct *vma)
>   {
>   	struct drm_gem_object *obj = vma->vm_private_data;
>   	struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
> -	int ret;
> -
> -	WARN_ON(shmem->base.import_attach);
> -
> -	ret = drm_gem_shmem_get_pages(shmem);
> -	WARN_ON_ONCE(ret != 0);
>   
> +	drm_gem_shmem_get_pages_no_fail(shmem);
>   	drm_gem_vm_open(vma);
>   }
>   
> @@ -716,6 +1229,8 @@ struct sg_table *drm_gem_shmem_get_pages_sgt(struct drm_gem_shmem_object *shmem)
>   
>   	shmem->sgt = sgt;
>   
> +	drm_gem_shmem_update_pages_state(shmem);
> +
>   	return sgt;
>   
>   err_free_sgt:
> @@ -762,6 +1277,202 @@ drm_gem_shmem_prime_import_sg_table(struct drm_device *dev,
>   }
>   EXPORT_SYMBOL_GPL(drm_gem_shmem_prime_import_sg_table);
>   
> +static struct drm_gem_shmem_shrinker *
> +to_drm_shrinker(struct shrinker *shrinker)
> +{
> +	return container_of(shrinker, struct drm_gem_shmem_shrinker, base);
> +}
> +
> +static unsigned long
> +drm_gem_shmem_shrinker_count_objects(struct shrinker *shrinker,
> +				     struct shrink_control *sc)
> +{
> +	struct drm_gem_shmem_shrinker *gem_shrinker = to_drm_shrinker(shrinker);
> +	u64 count = READ_ONCE(gem_shrinker->shrinkable_count);
> +
> +	if (count >= SHRINK_EMPTY)
> +		return SHRINK_EMPTY - 1;
> +
> +	return count ?: SHRINK_EMPTY;
> +}
> +
> +static unsigned long
> +drm_gem_shmem_shrinker_run_objects_scan(struct shrinker *shrinker,
> +					unsigned long nr_to_scan,
> +					bool *lock_contention,
> +					bool evict)
> +{
> +	struct drm_gem_shmem_shrinker *gem_shrinker = to_drm_shrinker(shrinker);
> +	struct drm_gem_shmem_object *shmem;
> +	struct list_head still_in_list;
> +	struct drm_gem_object *obj;
> +	unsigned long freed = 0;
> +	struct list_head *lru;
> +	size_t page_count;
> +
> +	INIT_LIST_HEAD(&still_in_list);
> +
> +	mutex_lock(&gem_shrinker->lock);
> +
> +	if (evict)
> +		lru = &gem_shrinker->lru_evictable;
> +	else
> +		lru = &gem_shrinker->lru_purgeable;
> +
> +	while (freed < nr_to_scan) {
> +		shmem = list_first_entry_or_null(lru, typeof(*shmem), madv_list);
> +		if (!shmem)
> +			break;
> +
> +		obj = &shmem->base;
> +		page_count = obj->size >> PAGE_SHIFT;
> +		list_move_tail(&shmem->madv_list, &still_in_list);
> +
> +		if (evict && get_nr_swap_pages() < page_count)
> +			continue;
> +
> +		/*
> +		 * If it's in the process of being freed, gem_object->free()
> +		 * may be blocked on lock waiting to remove it.  So just
> +		 * skip it.
> +		 */
> +		if (!kref_get_unless_zero(&obj->refcount))
> +			continue;
> +
> +		mutex_unlock(&gem_shrinker->lock);
> +
> +		/* prevent racing with job-submission code paths */
> +		if (!dma_resv_trylock(obj->resv)) {
> +			*lock_contention |= true;
> +			goto shrinker_lock;
> +		}
> +
> +		/* prevent racing with the dma-buf exporting */
> +		if (!mutex_trylock(&gem_shrinker->dev->object_name_lock)) {
> +			*lock_contention |= true;
> +			goto resv_unlock;
> +		}
> +
> +		/* check whether h/w uses this object */
> +		if (!dma_resv_test_signaled(obj->resv, DMA_RESV_USAGE_WRITE))
> +			goto object_name_unlock;
> +
> +		/* GEM may've become unpurgeable while shrinker was unlocked */
> +		if (evict) {
> +			if (!drm_gem_shmem_is_evictable(shmem))
> +				goto object_name_unlock;
> +		} else {
> +			if (!drm_gem_shmem_is_purgeable(shmem))
> +				goto object_name_unlock;
> +		}
> +
> +		if (evict)
> +			freed += obj->funcs->evict(obj);
> +		else
> +			freed += obj->funcs->purge(obj);
> +object_name_unlock:
> +		mutex_unlock(&gem_shrinker->dev->object_name_lock);
> +resv_unlock:
> +		dma_resv_unlock(obj->resv);
> +shrinker_lock:
> +		drm_gem_object_put(&shmem->base);
> +		mutex_lock(&gem_shrinker->lock);
> +	}
> +
> +	list_splice_tail(&still_in_list, lru);
> +
> +	mutex_unlock(&gem_shrinker->lock);
> +
> +	return freed;
> +}
> +
> +static unsigned long
> +drm_gem_shmem_shrinker_scan_objects(struct shrinker *shrinker,
> +				    struct shrink_control *sc)
> +{
> +	unsigned long nr_to_scan = sc->nr_to_scan;
> +	bool lock_contention = false;
> +	unsigned long freed;
> +
> +	/* purge as many objects as we can */
> +	freed = drm_gem_shmem_shrinker_run_objects_scan(shrinker, nr_to_scan,
> +							&lock_contention, false);
> +	nr_to_scan -= freed;
> +
> +	/* evict as many objects as we can */
> +	if (freed < nr_to_scan)
> +		freed += drm_gem_shmem_shrinker_run_objects_scan(shrinker,
> +								 nr_to_scan,
> +								 &lock_contention,
> +								 true);
> +
> +	return (!freed && !lock_contention) ? SHRINK_STOP : freed;
> +}
> +
> +/**
> + * drm_gem_shmem_shrinker_register() - Register shmem shrinker
> + * @dev: DRM device
> + *
> + * Returns:
> + * 0 on success or a negative error code on failure.
> + */
> +int drm_gem_shmem_shrinker_register(struct drm_device *dev)
> +{
> +	struct drm_gem_shmem_shrinker *gem_shrinker;
> +	int err;
> +
> +	if (WARN_ON(dev->shmem_shrinker))
> +		return -EBUSY;
> +
> +	gem_shrinker = kzalloc(sizeof(*gem_shrinker), GFP_KERNEL);
> +	if (!gem_shrinker)
> +		return -ENOMEM;
> +
> +	gem_shrinker->base.count_objects = drm_gem_shmem_shrinker_count_objects;
> +	gem_shrinker->base.scan_objects = drm_gem_shmem_shrinker_scan_objects;
> +	gem_shrinker->base.seeks = DEFAULT_SEEKS;
> +	gem_shrinker->dev = dev;
> +
> +	INIT_LIST_HEAD(&gem_shrinker->lru_purgeable);
> +	INIT_LIST_HEAD(&gem_shrinker->lru_evictable);
> +	INIT_LIST_HEAD(&gem_shrinker->lru_evicted);
> +	INIT_LIST_HEAD(&gem_shrinker->lru_active);
> +	mutex_init(&gem_shrinker->lock);
> +
> +	dev->shmem_shrinker = gem_shrinker;
> +
> +	err = register_shrinker(&gem_shrinker->base);
> +	if (err) {
> +		dev->shmem_shrinker = NULL;
> +		kfree(gem_shrinker);
> +		return err;
> +	}
> +
> +	return 0;
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_shrinker_register);
> +
> +/**
> + * drm_gem_shmem_shrinker_unregister() - Unregister shmem shrinker
> + * @dev: DRM device
> + */
> +void drm_gem_shmem_shrinker_unregister(struct drm_device *dev)
> +{
> +	struct drm_gem_shmem_shrinker *gem_shrinker = dev->shmem_shrinker;
> +
> +	if (gem_shrinker) {
> +		unregister_shrinker(&gem_shrinker->base);
> +		WARN_ON(!list_empty(&gem_shrinker->lru_purgeable));
> +		WARN_ON(!list_empty(&gem_shrinker->lru_evictable));
> +		WARN_ON(!list_empty(&gem_shrinker->lru_evicted));
> +		WARN_ON(!list_empty(&gem_shrinker->lru_active));
> +		mutex_destroy(&gem_shrinker->lock);
> +		dev->shmem_shrinker = NULL;
> +		kfree(gem_shrinker);
> +	}
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_shrinker_unregister);
> +
>   MODULE_DESCRIPTION("DRM SHMEM memory-management helpers");
>   MODULE_IMPORT_NS(DMA_BUF);
>   MODULE_LICENSE("GPL v2");
> diff --git a/include/drm/drm_device.h b/include/drm/drm_device.h
> index 9923c7a6885e..929546cad894 100644
> --- a/include/drm/drm_device.h
> +++ b/include/drm/drm_device.h
> @@ -16,6 +16,7 @@ struct drm_vblank_crtc;
>   struct drm_vma_offset_manager;
>   struct drm_vram_mm;
>   struct drm_fb_helper;
> +struct drm_gem_shmem_shrinker;
>   
>   struct inode;
>   
> @@ -277,6 +278,9 @@ struct drm_device {
>   	/** @vram_mm: VRAM MM memory manager */
>   	struct drm_vram_mm *vram_mm;
>   
> +	/** @shmem_shrinker: SHMEM GEM memory shrinker */
> +	struct drm_gem_shmem_shrinker *shmem_shrinker;
> +
>   	/**
>   	 * @switch_power_state:
>   	 *
> diff --git a/include/drm/drm_gem.h b/include/drm/drm_gem.h
> index 9d7c61a122dc..390d1ce08ed3 100644
> --- a/include/drm/drm_gem.h
> +++ b/include/drm/drm_gem.h
> @@ -172,6 +172,41 @@ struct drm_gem_object_funcs {
>   	 * This is optional but necessary for mmap support.
>   	 */
>   	const struct vm_operations_struct *vm_ops;
> +
> +	/**
> +	 * @purge:
> +	 *
> +	 * Releases the GEM object's allocated backing storage to the system.
> +	 *
> +	 * Returns the number of pages that have been freed by purging the GEM object.
> +	 *
> +	 * This callback is used by the GEM shrinker.
> +	 */
> +	unsigned long (*purge)(struct drm_gem_object *obj);
> +
> +	/**
> +	 * @evict:
> +	 *
> +	 * Unpins the GEM object's allocated backing storage, allowing shmem pages
> +	 * to be swapped out.
> +	 *
> +	 * Returns the number of pages that have been unpinned.
> +	 *
> +	 * This callback is used by the GEM shrinker.
> +	 */
> +	unsigned long (*evict)(struct drm_gem_object *obj);
> +
> +	/**
> +	 * @swap_in:
> +	 *
> +	 * Pins GEM object's allocated backing storage if it was previously evicted,
> +	 * moving swapped out pages back to memory.
> +	 *
> +	 * Returns 0 on success, or -errno on error.
> +	 *
> +	 * This callback is used by the GEM shrinker.
> +	 */
> +	int (*swap_in)(struct drm_gem_object *obj);
>   };
>   
>   /**
> diff --git a/include/drm/drm_gem_shmem_helper.h b/include/drm/drm_gem_shmem_helper.h
> index 70889533962a..a65557b446e6 100644
> --- a/include/drm/drm_gem_shmem_helper.h
> +++ b/include/drm/drm_gem_shmem_helper.h
> @@ -6,6 +6,7 @@
>   #include <linux/fs.h>
>   #include <linux/mm.h>
>   #include <linux/mutex.h>
> +#include <linux/shrinker.h>
>   
>   #include <drm/drm_file.h>
>   #include <drm/drm_gem.h>
> @@ -15,8 +16,18 @@
>   struct dma_buf_attachment;
>   struct drm_mode_create_dumb;
>   struct drm_printer;
> +struct drm_device;
>   struct sg_table;
>   
> +enum drm_gem_shmem_pages_state {
> +	DRM_GEM_SHMEM_PAGES_STATE_PURGED = -2,
> +	DRM_GEM_SHMEM_PAGES_STATE_EVICTED = -1,
> +	DRM_GEM_SHMEM_PAGES_STATE_UNPINNED = 0,
> +	DRM_GEM_SHMEM_PAGES_STATE_PINNED = 1,
> +	DRM_GEM_SHMEM_PAGES_STATE_EVICTABLE = 2,
> +	DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE = 3,
> +};
> +
>   /**
>    * struct drm_gem_shmem_object - GEM object backed by shmem
>    */
> @@ -43,8 +54,8 @@ struct drm_gem_shmem_object {
>   	 * @madv: State for madvise
>   	 *
>   	 * 0 is active/inuse.
> +	 * 1 is not-needed/can-be-purged
>   	 * A negative value is the object is purged.
> -	 * Positive values are driver specific and not used by the helpers.
>   	 */
>   	int madv;
>   
> @@ -91,6 +102,40 @@ struct drm_gem_shmem_object {
>   	 * @map_wc: map object write-combined (instead of using shmem defaults).
>   	 */
>   	bool map_wc;
> +
> +	/**
> +	 * @eviction_disable_count:
> +	 *
> +	 * The shmem pages are disallowed to be evicted by the memory shrinker
> +	 * while count is non-zero. Used internally by memory shrinker.
> +	 */
> +	unsigned int eviction_disable_count;
> +
> +	/**
> +	 * @purging_disable_count:
> +	 *
> +	 * The shmem pages are disallowed to be purged by the memory shrinker
> +	 * while count is non-zero. Used internally by memory shrinker.
> +	 */
> +	unsigned int purging_disable_count;
> +
> +	/**
> +	 * @pages_state: Current state of shmem pages. Used internally by
> +	 * memory shrinker.
> +	 */
> +	enum drm_gem_shmem_pages_state pages_state;
> +
> +	/**
> +	 * @evicted: True if shmem pages were evicted by the memory shrinker.
> +	 * Used internally by memory shrinker.
> +	 */
> +	bool evicted;
> +
> +	/**
> +	 * @pages_shrinkable: True if shmem pages can be evicted or purged
> +	 * by the memory shrinker. Used internally by memory shrinker.
> +	 */
> +	bool pages_shrinkable;
>   };
>   
>   #define to_drm_gem_shmem_obj(obj) \
> @@ -111,15 +156,33 @@ int drm_gem_shmem_mmap(struct drm_gem_shmem_object *shmem, struct vm_area_struct
>   
>   int drm_gem_shmem_madvise(struct drm_gem_shmem_object *shmem, int madv);
>   
> +int drm_gem_shmem_set_purgeable(struct drm_gem_shmem_object *shmem);
> +int drm_gem_shmem_set_purgeable_and_evictable(struct drm_gem_shmem_object *shmem);
> +int drm_gem_shmem_set_unpurgeable_and_unevictable(struct drm_gem_shmem_object *shmem);
> +
> +static inline bool drm_gem_shmem_is_evictable(struct drm_gem_shmem_object *shmem)
> +{
> +	return (shmem->madv >= 0) && !shmem->eviction_disable_count &&
> +		shmem->base.funcs->evict && shmem->base.funcs->swap_in &&
> +		!shmem->vmap_use_count && !shmem->base.dma_buf &&
> +		!shmem->base.import_attach && shmem->sgt;
> +}
> +
>   static inline bool drm_gem_shmem_is_purgeable(struct drm_gem_shmem_object *shmem)
>   {
> -	return (shmem->madv > 0) &&
> -		!shmem->vmap_use_count && shmem->sgt &&
> -		!shmem->base.dma_buf && !shmem->base.import_attach;
> +	return (shmem->madv > 0) && !shmem->purging_disable_count &&
> +		!shmem->vmap_use_count && shmem->base.funcs->purge &&
> +		!shmem->base.dma_buf && !shmem->base.import_attach &&
> +		shmem->sgt;
>   }
>   
> -void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem);
> +int drm_gem_shmem_swap_in_pages_locked(struct drm_gem_shmem_object *shmem);
> +int drm_gem_shmem_swap_in_locked(struct drm_gem_shmem_object *shmem);
> +
> +void drm_gem_shmem_evict_locked(struct drm_gem_shmem_object *shmem);
> +
>   bool drm_gem_shmem_purge(struct drm_gem_shmem_object *shmem);
> +void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem);
>   
>   struct sg_table *drm_gem_shmem_get_sg_table(struct drm_gem_shmem_object *shmem);
>   struct sg_table *drm_gem_shmem_get_pages_sgt(struct drm_gem_shmem_object *shmem);
> @@ -262,6 +325,38 @@ static inline int drm_gem_shmem_object_mmap(struct drm_gem_object *obj, struct v
>   	return drm_gem_shmem_mmap(shmem, vma);
>   }
>   
> +/**
> + * struct drm_gem_shmem_shrinker - Generic memory shrinker for shmem GEMs
> + */
> +struct drm_gem_shmem_shrinker {
> +	/** @base: Shrinker for purging shmem GEM objects */
> +	struct shrinker base;
> +
> +	/** @lock: Protects @lru_* */
> +	struct mutex lock;
> +
> +	/** @lru_purgeable: List of shmem GEM objects available for purging */
> +	struct list_head lru_purgeable;
> +
> +	/** @lru_active: List of active shmem GEM objects */
> +	struct list_head lru_active;
> +
> +	/** @lru_evictable: List of shmem GEM objects that can be evicted */
> +	struct list_head lru_evictable;
> +
> +	/** @lru_evicted: List of evicted shmem GEM objects */
> +	struct list_head lru_evicted;
> +
> +	/** @dev: DRM device that uses this shrinker */
> +	struct drm_device *dev;
> +
> +	/** @shrinkable_count: Count of shmem GEM pages to be purged and evicted */
> +	u64 shrinkable_count;
> +};
> +
> +int drm_gem_shmem_shrinker_register(struct drm_device *dev);
> +void drm_gem_shmem_shrinker_unregister(struct drm_device *dev);
> +
>   /*
>    * Driver ops
>    */

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

[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]

[-- Attachment #2: Type: text/plain, Size: 183 bytes --]

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 11/15] drm/shmem-helper: Add generic memory shrinker

[Daniel Vetter]

On Thu, May 05, 2022 at 10:34:02AM +0200, Thomas Zimmermann wrote:
> Hi
> 
> Am 18.04.22 um 00:37 schrieb Dmitry Osipenko:
> > Introduce a common DRM SHMEM shrinker. It allows to reduce code
> > duplication among DRM drivers that implement theirs own shrinkers.
> > This is initial version of the shrinker that covers basic needs of
> > GPU drivers, both purging and eviction of shmem objects are supported.
> > 
> > This patch is based on a couple ideas borrowed from Rob's Clark MSM
> > shrinker and Thomas' Zimmermann variant of SHMEM shrinker.
> > 
> > In order to start using DRM SHMEM shrinker drivers should:
> > 
> > 1. Implement new purge(), evict() + swap_in() GEM callbacks.
> > 2. Register shrinker using drm_gem_shmem_shrinker_register(drm_device).
> > 3. Use drm_gem_shmem_set_purgeable_and_evictable(shmem) and alike API
> >     functions to activate shrinking of GEMs.
> 
> Honestly speaking, after reading the patch and the discussion here I really
> don't like where all tis is going. The interfaces and implementation are
> overengineered.  Descisions about evicting and purging should be done by the
> memory manager. For the most part, it's none of the driver's business.
> 
> I'd like to ask you to reduce the scope of the patchset and build the
> shrinker only for virtio-gpu. I know that I first suggested to build upon
> shmem helpers, but it seems that it's easier to do that in a later patchset.

We have a few shrinkers already all over, so extracting that does make
sense I think. I do agree that there's probably a few more steps than
necessary involved right now in all this for the helper<->driver
interface.
-Daniel

> 
> Best regards
> Thomas
> 
> > 
> > Signed-off-by: Daniel Almeida <daniel.almeida@collabora.com>
> > Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
> > ---
> >   drivers/gpu/drm/drm_gem_shmem_helper.c | 765 ++++++++++++++++++++++++-
> >   include/drm/drm_device.h               |   4 +
> >   include/drm/drm_gem.h                  |  35 ++
> >   include/drm/drm_gem_shmem_helper.h     | 105 +++-
> >   4 files changed, 877 insertions(+), 32 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/drm_gem_shmem_helper.c b/drivers/gpu/drm/drm_gem_shmem_helper.c
> > index 3ecef571eff3..3838fb8d6f3a 100644
> > --- a/drivers/gpu/drm/drm_gem_shmem_helper.c
> > +++ b/drivers/gpu/drm/drm_gem_shmem_helper.c
> > @@ -88,6 +88,13 @@ __drm_gem_shmem_create(struct drm_device *dev, size_t size, bool private)
> >   	INIT_LIST_HEAD(&shmem->madv_list);
> > +	/*
> > +	 * Eviction and purging are disabled by default, shmem user must enable
> > +	 * them explicitly using drm_gem_shmem_set_evictable/purgeable().
> > +	 */
> > +	shmem->eviction_disable_count = 1;
> > +	shmem->purging_disable_count = 1;
> > +
> >   	if (!private) {
> >   		/*
> >   		 * Our buffers are kept pinned, so allocating them
> > @@ -126,6 +133,107 @@ struct drm_gem_shmem_object *drm_gem_shmem_create(struct drm_device *dev, size_t
> >   }
> >   EXPORT_SYMBOL_GPL(drm_gem_shmem_create);
> > +static void
> > +drm_gem_shmem_add_pages_to_shrinker(struct drm_gem_shmem_object *shmem)
> > +{
> > +	struct drm_gem_object *obj = &shmem->base;
> > +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> > +	size_t page_count = obj->size >> PAGE_SHIFT;
> > +
> > +	if (!shmem->pages_shrinkable) {
> > +		WARN_ON(gem_shrinker->shrinkable_count + page_count < page_count);
> > +		gem_shrinker->shrinkable_count += page_count;
> > +		shmem->pages_shrinkable = true;
> > +	}
> > +}
> > +
> > +static void
> > +drm_gem_shmem_remove_pages_from_shrinker(struct drm_gem_shmem_object *shmem)
> > +{
> > +	struct drm_gem_object *obj = &shmem->base;
> > +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> > +	size_t page_count = obj->size >> PAGE_SHIFT;
> > +
> > +	if (shmem->pages_shrinkable) {
> > +		WARN_ON(gem_shrinker->shrinkable_count < page_count);
> > +		gem_shrinker->shrinkable_count -= page_count;
> > +		shmem->pages_shrinkable = false;
> > +	}
> > +}
> > +
> > +static void
> > +drm_gem_shmem_set_pages_state_locked(struct drm_gem_shmem_object *shmem,
> > +				     enum drm_gem_shmem_pages_state new_state)
> > +{
> > +	struct drm_gem_object *obj = &shmem->base;
> > +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> > +
> > +	lockdep_assert_held(&gem_shrinker->lock);
> > +	lockdep_assert_held(&obj->resv->lock.base);
> > +
> > +	if (new_state >= DRM_GEM_SHMEM_PAGES_STATE_PINNED) {
> > +		if (drm_gem_shmem_is_evictable(shmem))
> > +			new_state = DRM_GEM_SHMEM_PAGES_STATE_EVICTABLE;
> > +
> > +		if (drm_gem_shmem_is_purgeable(shmem))
> > +			new_state = DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE;
> > +
> > +		if (!shmem->pages)
> > +			new_state = DRM_GEM_SHMEM_PAGES_STATE_UNPINNED;
> > +
> > +		if (shmem->evicted)
> > +			new_state = DRM_GEM_SHMEM_PAGES_STATE_EVICTED;
> > +	}
> > +
> > +	if (shmem->pages_state == new_state)
> > +		return;
> > +
> > +	switch (new_state) {
> > +	case DRM_GEM_SHMEM_PAGES_STATE_UNPINNED:
> > +	case DRM_GEM_SHMEM_PAGES_STATE_PURGED:
> > +		drm_gem_shmem_remove_pages_from_shrinker(shmem);
> > +		list_del_init(&shmem->madv_list);
> > +		break;
> > +
> > +	case DRM_GEM_SHMEM_PAGES_STATE_PINNED:
> > +		drm_gem_shmem_remove_pages_from_shrinker(shmem);
> > +		list_move_tail(&shmem->madv_list, &gem_shrinker->lru_active);
> > +		break;
> > +
> > +	case DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE:
> > +		drm_gem_shmem_add_pages_to_shrinker(shmem);
> > +		list_move_tail(&shmem->madv_list, &gem_shrinker->lru_purgeable);
> > +		break;
> > +
> > +	case DRM_GEM_SHMEM_PAGES_STATE_EVICTABLE:
> > +		drm_gem_shmem_add_pages_to_shrinker(shmem);
> > +		list_move_tail(&shmem->madv_list, &gem_shrinker->lru_evictable);
> > +		break;
> > +
> > +	case DRM_GEM_SHMEM_PAGES_STATE_EVICTED:
> > +		drm_gem_shmem_remove_pages_from_shrinker(shmem);
> > +		list_move_tail(&shmem->madv_list, &gem_shrinker->lru_evicted);
> > +		break;
> > +	}
> > +
> > +	shmem->pages_state = new_state;
> > +}
> > +
> > +static void
> > +drm_gem_shmem_set_pages_state(struct drm_gem_shmem_object *shmem,
> > +			      enum drm_gem_shmem_pages_state new_state)
> > +{
> > +	struct drm_gem_object *obj = &shmem->base;
> > +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> > +
> > +	if (!gem_shrinker)
> > +		return;
> > +
> > +	mutex_lock(&gem_shrinker->lock);
> > +	drm_gem_shmem_set_pages_state_locked(shmem, new_state);
> > +	mutex_unlock(&gem_shrinker->lock);
> > +}
> > +
> >   /**
> >    * drm_gem_shmem_free - Free resources associated with a shmem GEM object
> >    * @shmem: shmem GEM object to free
> > @@ -137,6 +245,9 @@ void drm_gem_shmem_free(struct drm_gem_shmem_object *shmem)
> >   {
> >   	struct drm_gem_object *obj = &shmem->base;
> > +	/* take out shmem GEM object from the memory shrinker */
> > +	drm_gem_shmem_madvise(shmem, -1);
> > +
> >   	WARN_ON(shmem->vmap_use_count);
> >   	if (obj->import_attach) {
> > @@ -148,7 +259,7 @@ void drm_gem_shmem_free(struct drm_gem_shmem_object *shmem)
> >   			sg_free_table(shmem->sgt);
> >   			kfree(shmem->sgt);
> >   		}
> > -		if (shmem->pages)
> > +		if (shmem->pages_use_count)
> >   			drm_gem_shmem_put_pages(shmem);
> >   	}
> > @@ -159,18 +270,226 @@ void drm_gem_shmem_free(struct drm_gem_shmem_object *shmem)
> >   }
> >   EXPORT_SYMBOL_GPL(drm_gem_shmem_free);
> > -static int drm_gem_shmem_get_pages_locked(struct drm_gem_shmem_object *shmem)
> > +static void drm_gem_shmem_update_pages_state_locked(struct drm_gem_shmem_object *shmem)
> > +{
> > +	struct drm_gem_object *obj = &shmem->base;
> > +	struct drm_gem_shmem_shrinker *gem_shrinker = obj->dev->shmem_shrinker;
> > +	enum drm_gem_shmem_pages_state new_state;
> > +
> > +	if (!gem_shrinker || obj->import_attach)
> > +		return;
> > +
> > +	mutex_lock(&gem_shrinker->lock);
> > +
> > +	if (!shmem->madv)
> > +		new_state = DRM_GEM_SHMEM_PAGES_STATE_PINNED;
> > +	else if (shmem->madv > 0)
> > +		new_state = DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE;
> > +	else
> > +		new_state = DRM_GEM_SHMEM_PAGES_STATE_PURGED;
> > +
> > +	drm_gem_shmem_set_pages_state_locked(shmem, new_state);
> > +
> > +	mutex_unlock(&gem_shrinker->lock);
> > +}
> > +
> > +static void drm_gem_shmem_update_pages_state(struct drm_gem_shmem_object *shmem)
> > +{
> > +	dma_resv_lock(shmem->base.resv, NULL);
> > +	drm_gem_shmem_update_pages_state_locked(shmem);
> > +	dma_resv_unlock(shmem->base.resv);
> > +}
> > +
> > +static int
> > +drm_gem_shmem_set_evictable_locked(struct drm_gem_shmem_object *shmem)
> > +{
> > +	int ret = 0;
> > +
> > +	WARN_ON_ONCE(!shmem->eviction_disable_count--);
> > +
> > +	if (shmem->madv < 0)
> > +		ret = -ENOMEM;
> > +
> > +	drm_gem_shmem_update_pages_state_locked(shmem);
> > +
> > +	return ret;
> > +}
> > +
> > +static int
> > +drm_gem_shmem_set_unevictable_locked(struct drm_gem_shmem_object *shmem)
> > +{
> > +	struct drm_gem_object *obj = &shmem->base;
> > +	int err;
> > +
> > +	if (shmem->madv < 0)
> > +		return -ENOMEM;
> > +
> > +	if (shmem->evicted) {
> > +		err = obj->funcs->swap_in(obj);
> > +		if (err)
> > +			return err;
> > +	}
> > +
> > +	shmem->eviction_disable_count++;
> > +
> > +	drm_gem_shmem_update_pages_state_locked(shmem);
> > +
> > +	return 0;
> > +}
> > +
> > +static int
> > +drm_gem_shmem_set_purgeable_locked(struct drm_gem_shmem_object *shmem)
> > +{
> > +	int ret = 0;
> > +
> > +	WARN_ON_ONCE(!shmem->purging_disable_count--);
> > +
> > +	if (shmem->madv < 0)
> > +		ret = -ENOMEM;
> > +
> > +	drm_gem_shmem_update_pages_state_locked(shmem);
> > +
> > +	return ret;
> > +}
> > +
> > +/**
> > + * drm_gem_shmem_set_purgeable() - Make GEM purgeable by memory shrinker
> > + * @shmem: shmem GEM object
> > + *
> > + * Tell memory shrinker that this GEM can be purged. Initially purging is
> > + * disabled for all GEMs. Each set_pureable() call must have corresponding
> > + * set_unpureable() call. If GEM was purged, then -ENOMEM is returned.
> > + *
> > + * Returns:
> > + * 0 on success or a negative error code on failure.
> > + */
> > +int drm_gem_shmem_set_purgeable(struct drm_gem_shmem_object *shmem)
> > +{
> > +	int ret;
> > +
> > +	dma_resv_lock(shmem->base.resv, NULL);
> > +	ret = drm_gem_shmem_set_purgeable_locked(shmem);
> > +	dma_resv_unlock(shmem->base.resv);
> > +
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL_GPL(drm_gem_shmem_set_purgeable);
> > +
> > +static int
> > +drm_gem_shmem_set_unpurgeable_locked(struct drm_gem_shmem_object *shmem)
> > +{
> > +	if (shmem->madv < 0)
> > +		return -ENOMEM;
> > +
> > +	shmem->purging_disable_count++;
> > +
> > +	drm_gem_shmem_update_pages_state_locked(shmem);
> > +
> > +	return 0;
> > +}
> > +
> > +static int
> > +drm_gem_shmem_set_purgeable_and_evictable_locked(struct drm_gem_shmem_object *shmem)
> > +{
> > +	int ret;
> > +
> > +	ret = drm_gem_shmem_set_evictable_locked(shmem);
> > +	if (!ret) {
> > +		ret = drm_gem_shmem_set_purgeable_locked(shmem);
> > +		if (ret)
> > +			drm_gem_shmem_set_unevictable_locked(shmem);
> > +	}
> > +
> > +	return ret;
> > +}
> > +
> > +static int
> > +drm_gem_shmem_set_unpurgeable_and_unevictable_locked(struct drm_gem_shmem_object *shmem)
> > +{
> > +	int ret;
> > +
> > +	ret = drm_gem_shmem_set_unpurgeable_locked(shmem);
> > +	if (!ret) {
> > +		ret = drm_gem_shmem_set_unevictable_locked(shmem);
> > +		if (ret)
> > +			drm_gem_shmem_set_purgeable_locked(shmem);
> > +	}
> > +
> > +	return ret;
> > +}
> > +
> > +/**
> > + * drm_gem_shmem_set_purgeable_and_evictable() - Make GEM unpurgeable and
> > + * 						 unevictable by memory shrinker
> > + * @shmem: shmem GEM object
> > + *
> > + * Tell memory shrinker that this GEM can't be purged and evicted. Each
> > + * set_purgeable_and_evictable() call must have corresponding
> > + * unpurgeable_and_unevictable() call. If GEM was purged, then -ENOMEM
> > + * is returned.
> > + *
> > + * Returns:
> > + * 0 on success or a negative error code on failure.
> > + */
> > +int drm_gem_shmem_set_purgeable_and_evictable(struct drm_gem_shmem_object *shmem)
> > +{
> > +	int ret;
> > +
> > +	dma_resv_lock(shmem->base.resv, NULL);
> > +	ret = drm_gem_shmem_set_purgeable_and_evictable_locked(shmem);
> > +	dma_resv_unlock(shmem->base.resv);
> > +
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL_GPL(drm_gem_shmem_set_purgeable_and_evictable);
> > +
> > +/**
> > + * drm_gem_shmem_set_unpurgeable_and_unevictable() - Make GEM purgeable and
> > + * 						     evictable by memory shrinker
> > + * @shmem: shmem GEM object
> > + *
> > + * Tell memory shrinker that this GEM can be purged and evicted. Each
> > + * unpurgeable_and_unevictable() call must have corresponding
> > + * set_purgeable_and_evictable() call. If GEM was purged, then -ENOMEM
> > + * is returned.
> > + *
> > + * Returns:
> > + * 0 on success or a negative error code on failure.
> > + */
> > +int drm_gem_shmem_set_unpurgeable_and_unevictable(struct drm_gem_shmem_object *shmem)
> > +{
> > +	int ret;
> > +
> > +	ret = dma_resv_lock_interruptible(shmem->base.resv, NULL);
> > +	if (ret)
> > +		return ret;
> > +
> > +	ret = drm_gem_shmem_set_unpurgeable_and_unevictable_locked(shmem);
> > +	dma_resv_unlock(shmem->base.resv);
> > +
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL_GPL(drm_gem_shmem_set_unpurgeable_and_unevictable);
> > +
> > +static int
> > +drm_gem_shmem_acquire_pages_locked(struct drm_gem_shmem_object *shmem)
> >   {
> >   	struct drm_gem_object *obj = &shmem->base;
> >   	struct page **pages;
> > -	if (shmem->pages_use_count++ > 0)
> > +	if (shmem->madv < 0) {
> > +		WARN_ON(shmem->pages);
> > +		return -ENOMEM;
> > +	}
> > +
> > +	if (shmem->pages) {
> > +		WARN_ON(!shmem->evicted);
> >   		return 0;
> > +	}
> >   	pages = drm_gem_get_pages(obj);
> >   	if (IS_ERR(pages)) {
> >   		DRM_DEBUG_KMS("Failed to get pages (%ld)\n", PTR_ERR(pages));
> > -		shmem->pages_use_count = 0;
> >   		return PTR_ERR(pages);
> >   	}
> > @@ -189,6 +508,25 @@ static int drm_gem_shmem_get_pages_locked(struct drm_gem_shmem_object *shmem)
> >   	return 0;
> >   }
> > +static int drm_gem_shmem_get_pages_locked(struct drm_gem_shmem_object *shmem)
> > +{
> > +	int err;
> > +
> > +	if (shmem->madv < 0)
> > +		return -ENOMEM;
> > +
> > +	if (shmem->pages_use_count++ > 0)
> > +		return 0;
> > +
> > +	err = drm_gem_shmem_acquire_pages_locked(shmem);
> > +	if (err) {
> > +		shmem->pages_use_count = 0;
> > +		return err;
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> >   /*
> >    * drm_gem_shmem_get_pages - Allocate backing pages for a shmem GEM object
> >    * @shmem: shmem GEM object
> > @@ -209,21 +547,38 @@ int drm_gem_shmem_get_pages(struct drm_gem_shmem_object *shmem)
> >   	if (ret)
> >   		return ret;
> >   	ret = drm_gem_shmem_get_pages_locked(shmem);
> > +
> > +	drm_gem_shmem_update_pages_state_locked(shmem);
> > +
> >   	dma_resv_unlock(shmem->base.resv);
> >   	return ret;
> >   }
> >   EXPORT_SYMBOL(drm_gem_shmem_get_pages);
> > -static void drm_gem_shmem_put_pages_locked(struct drm_gem_shmem_object *shmem)
> > +static void drm_gem_shmem_get_pages_no_fail(struct drm_gem_shmem_object *shmem)
> >   {
> > -	struct drm_gem_object *obj = &shmem->base;
> > +	WARN_ON(shmem->base.import_attach);
> > -	if (WARN_ON_ONCE(!shmem->pages_use_count))
> > -		return;
> > +	dma_resv_lock(shmem->base.resv, NULL);
> > -	if (--shmem->pages_use_count > 0)
> > +	if (drm_gem_shmem_get_pages_locked(shmem))
> > +		shmem->pages_use_count++;
> > +
> > +	drm_gem_shmem_update_pages_state_locked(shmem);
> > +
> > +	dma_resv_unlock(shmem->base.resv);
> > +}
> > +
> > +static void
> > +drm_gem_shmem_release_pages_locked(struct drm_gem_shmem_object *shmem)
> > +{
> > +	struct drm_gem_object *obj = &shmem->base;
> > +
> > +	if (!shmem->pages) {
> > +		WARN_ON(!shmem->evicted && shmem->madv >= 0);
> >   		return;
> > +	}
> >   #ifdef CONFIG_X86
> >   	if (shmem->map_wc)
> > @@ -236,6 +591,21 @@ static void drm_gem_shmem_put_pages_locked(struct drm_gem_shmem_object *shmem)
> >   	shmem->pages = NULL;
> >   }
> > +static void drm_gem_shmem_put_pages_locked(struct drm_gem_shmem_object *shmem)
> > +{
> > +	struct drm_gem_object *obj = &shmem->base;
> > +
> > +	lockdep_assert_held(&obj->resv->lock.base);
> > +
> > +	if (WARN_ON(!shmem->pages_use_count))
> > +		return;
> > +
> > +	if (--shmem->pages_use_count > 0)
> > +		return;
> > +
> > +	drm_gem_shmem_release_pages_locked(shmem);
> > +}
> > +
> >   /*
> >    * drm_gem_shmem_put_pages - Decrease use count on the backing pages for a shmem GEM object
> >    * @shmem: shmem GEM object
> > @@ -246,6 +616,7 @@ void drm_gem_shmem_put_pages(struct drm_gem_shmem_object *shmem)
> >   {
> >   	dma_resv_lock(shmem->base.resv, NULL);
> >   	drm_gem_shmem_put_pages_locked(shmem);
> > +	drm_gem_shmem_update_pages_state_locked(shmem);
> >   	dma_resv_unlock(shmem->base.resv);
> >   }
> >   EXPORT_SYMBOL(drm_gem_shmem_put_pages);
> > @@ -262,9 +633,21 @@ EXPORT_SYMBOL(drm_gem_shmem_put_pages);
> >    */
> >   int drm_gem_shmem_pin(struct drm_gem_shmem_object *shmem)
> >   {
> > +	int err;
> > +
> >   	WARN_ON(shmem->base.import_attach);
> > -	return drm_gem_shmem_get_pages(shmem);
> > +	err = drm_gem_shmem_set_unpurgeable_and_unevictable(shmem);
> > +	if (err)
> > +		return err;
> > +
> > +	err = drm_gem_shmem_get_pages(shmem);
> > +	if (err) {
> > +		drm_gem_shmem_set_purgeable_and_evictable(shmem);
> > +		return err;
> > +	}
> > +
> > +	return 0;
> >   }
> >   EXPORT_SYMBOL(drm_gem_shmem_pin);
> > @@ -280,6 +663,7 @@ void drm_gem_shmem_unpin(struct drm_gem_shmem_object *shmem)
> >   	WARN_ON(shmem->base.import_attach);
> >   	drm_gem_shmem_put_pages(shmem);
> > +	drm_gem_shmem_set_purgeable_and_evictable(shmem);
> >   }
> >   EXPORT_SYMBOL(drm_gem_shmem_unpin);
> > @@ -359,7 +743,18 @@ int drm_gem_shmem_vmap(struct drm_gem_shmem_object *shmem,
> >   	ret = dma_resv_lock_interruptible(shmem->base.resv, NULL);
> >   	if (ret)
> >   		return ret;
> > +
> > +	ret = drm_gem_shmem_set_unpurgeable_and_unevictable_locked(shmem);
> > +	if (ret)
> > +		goto unlock;
> > +
> >   	ret = drm_gem_shmem_vmap_locked(shmem, map);
> > +	if (ret)
> > +		drm_gem_shmem_set_purgeable_and_evictable_locked(shmem);
> > +	else
> > +		drm_gem_shmem_update_pages_state_locked(shmem);
> > +
> > +unlock:
> >   	dma_resv_unlock(shmem->base.resv);
> >   	return ret;
> > @@ -404,9 +799,9 @@ void drm_gem_shmem_vunmap(struct drm_gem_shmem_object *shmem,
> >   {
> >   	dma_resv_lock(shmem->base.resv, NULL);
> >   	drm_gem_shmem_vunmap_locked(shmem, map);
> > +	drm_gem_shmem_update_pages_state_locked(shmem);
> > +	drm_gem_shmem_set_purgeable_and_evictable_locked(shmem);
> >   	dma_resv_unlock(shmem->base.resv);
> > -
> > -	drm_gem_shmem_update_purgeable_status(shmem);
> >   }
> >   EXPORT_SYMBOL(drm_gem_shmem_vunmap);
> > @@ -447,29 +842,140 @@ int drm_gem_shmem_madvise(struct drm_gem_shmem_object *shmem, int madv)
> >   	madv = shmem->madv;
> > +	drm_gem_shmem_update_pages_state_locked(shmem);
> > +
> >   	dma_resv_unlock(shmem->base.resv);
> >   	return (madv >= 0);
> >   }
> >   EXPORT_SYMBOL(drm_gem_shmem_madvise);
> > -void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem)
> > +/**
> > + * drm_gem_shmem_swap_in_pages_locked() - Moves shmem pages back to memory
> > + * @shmem: shmem GEM object
> > + *
> > + * This function moves pages back to memory if they were previously evicted
> > + * by the memory shrinker.
> > + *
> > + * Returns:
> > + * 0 on success or a negative error code on failure.
> > + */
> > +int drm_gem_shmem_swap_in_pages_locked(struct drm_gem_shmem_object *shmem)
> > +{
> > +	struct drm_gem_object *obj = &shmem->base;
> > +	struct sg_table *sgt;
> > +	int ret;
> > +
> > +	lockdep_assert_held(&obj->resv->lock.base);
> > +
> > +	if (shmem->evicted) {
> > +		ret = drm_gem_shmem_acquire_pages_locked(shmem);
> > +		if (ret)
> > +			return ret;
> > +
> > +		sgt = drm_gem_shmem_get_sg_table(shmem);
> > +		if (IS_ERR(sgt))
> > +			return PTR_ERR(sgt);
> > +
> > +		ret = dma_map_sgtable(obj->dev->dev, sgt,
> > +				      DMA_BIDIRECTIONAL, 0);
> > +		if (ret) {
> > +			sg_free_table(sgt);
> > +			kfree(sgt);
> > +			return ret;
> > +		}
> > +
> > +		shmem->sgt = sgt;
> > +		shmem->evicted = false;
> > +		shmem->pages_state = DRM_GEM_SHMEM_PAGES_STATE_PINNED;
> > +
> > +		drm_gem_shmem_update_pages_state_locked(shmem);
> > +	}
> > +
> > +	return shmem->pages ? 0 : -ENOMEM;
> > +}
> > +EXPORT_SYMBOL_GPL(drm_gem_shmem_swap_in_pages_locked);
> > +
> > +/**
> > + * drm_gem_shmem_swap_in_locked() - Moves shmem GEM back to memory
> > + * @shmem: shmem GEM object
> > + *
> > + * This function moves shmem GEM back to memory if it was previously evicted
> > + * by the memory shrinker. The GEM is ready to use on success.
> > + *
> > + * Returns:
> > + * 0 on success or a negative error code on failure.
> > + */
> > +int drm_gem_shmem_swap_in_locked(struct drm_gem_shmem_object *shmem)
> > +{
> > +	struct drm_gem_object *obj = &shmem->base;
> > +
> > +	lockdep_assert_held(&obj->resv->lock.base);
> > +
> > +	if (shmem->evicted)
> > +		return obj->funcs->swap_in(obj);
> > +
> > +	return 0;
> > +}
> > +EXPORT_SYMBOL_GPL(drm_gem_shmem_swap_in_locked);
> > +
> > +static void drm_gem_shmem_unpin_pages_locked(struct drm_gem_shmem_object *shmem)
> >   {
> >   	struct drm_gem_object *obj = &shmem->base;
> >   	struct drm_device *dev = obj->dev;
> > -	WARN_ON(!drm_gem_shmem_is_purgeable(shmem));
> > +	if (shmem->evicted)
> > +		return;
> >   	dma_unmap_sgtable(dev->dev, shmem->sgt, DMA_BIDIRECTIONAL, 0);
> > +	drm_gem_shmem_release_pages_locked(shmem);
> > +	drm_vma_node_unmap(&obj->vma_node, dev->anon_inode->i_mapping);
> > +
> >   	sg_free_table(shmem->sgt);
> >   	kfree(shmem->sgt);
> >   	shmem->sgt = NULL;
> > +}
> > -	drm_gem_shmem_put_pages_locked(shmem);
> > +/**
> > + * drm_gem_shmem_evict_locked - Evict shmem pages
> > + * @shmem: shmem GEM object
> > + *
> > + * This function unpins shmem pages, allowing them to be swapped out from
> > + * memory.
> > + */
> > +void drm_gem_shmem_evict_locked(struct drm_gem_shmem_object *shmem)
> > +{
> > +	struct drm_gem_object *obj = &shmem->base;
> > -	shmem->madv = -1;
> > +	lockdep_assert_held(&obj->resv->lock.base);
> > -	drm_vma_node_unmap(&obj->vma_node, dev->anon_inode->i_mapping);
> > +	WARN_ON(!drm_gem_shmem_is_evictable(shmem));
> > +	WARN_ON(shmem->madv < 0);
> > +	WARN_ON(shmem->evicted);
> > +
> > +	drm_gem_shmem_unpin_pages_locked(shmem);
> > +
> > +	shmem->evicted = true;
> > +	drm_gem_shmem_set_pages_state(shmem, DRM_GEM_SHMEM_PAGES_STATE_EVICTED);
> > +}
> > +EXPORT_SYMBOL_GPL(drm_gem_shmem_evict_locked);
> > +
> > +/**
> > + * drm_gem_shmem_purge_locked - Purge shmem pages
> > + * @shmem: shmem GEM object
> > + *
> > + * This function permanently releases shmem pages.
> > + */
> > +void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem)
> > +{
> > +	struct drm_gem_object *obj = &shmem->base;
> > +
> > +	lockdep_assert_held(&obj->resv->lock.base);
> > +
> > +	WARN_ON(!drm_gem_shmem_is_purgeable(shmem));
> > +	WARN_ON(shmem->madv < 0);
> > +
> > +	drm_gem_shmem_unpin_pages_locked(shmem);
> >   	drm_gem_free_mmap_offset(obj);
> >   	/* Our goal here is to return as much of the memory as
> > @@ -480,6 +986,9 @@ void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem)
> >   	shmem_truncate_range(file_inode(obj->filp), 0, (loff_t)-1);
> >   	invalidate_mapping_pages(file_inode(obj->filp)->i_mapping, 0, (loff_t)-1);
> > +
> > +	shmem->madv = -1;
> > +	drm_gem_shmem_set_pages_state(shmem, DRM_GEM_SHMEM_PAGES_STATE_PURGED);
> >   }
> >   EXPORT_SYMBOL(drm_gem_shmem_purge_locked);
> > @@ -543,22 +1052,31 @@ static vm_fault_t drm_gem_shmem_fault(struct vm_fault *vmf)
> >   	vm_fault_t ret;
> >   	struct page *page;
> >   	pgoff_t page_offset;
> > +	bool pages_inactive;
> > +	int err;
> >   	/* We don't use vmf->pgoff since that has the fake offset */
> >   	page_offset = (vmf->address - vma->vm_start) >> PAGE_SHIFT;
> >   	dma_resv_lock(shmem->base.resv, NULL);
> > -	if (page_offset >= num_pages ||
> > -	    WARN_ON_ONCE(!shmem->pages) ||
> > -	    shmem->madv < 0) {
> > +	pages_inactive = shmem->pages_state < DRM_GEM_SHMEM_PAGES_STATE_PINNED;
> > +	WARN_ON_ONCE(!shmem->pages ^ pages_inactive);
> > +
> > +	if (page_offset >= num_pages || (!shmem->pages && !shmem->evicted)) {
> >   		ret = VM_FAULT_SIGBUS;
> >   	} else {
> > +		err = drm_gem_shmem_swap_in_locked(shmem);
> > +		if (err) {
> > +			ret = VM_FAULT_OOM;
> > +			goto unlock;
> > +		}
> > +
> >   		page = shmem->pages[page_offset];
> >   		ret = vmf_insert_pfn(vma, vmf->address, page_to_pfn(page));
> >   	}
> > -
> > +unlock:
> >   	dma_resv_unlock(shmem->base.resv);
> >   	return ret;
> > @@ -568,13 +1086,8 @@ static void drm_gem_shmem_vm_open(struct vm_area_struct *vma)
> >   {
> >   	struct drm_gem_object *obj = vma->vm_private_data;
> >   	struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
> > -	int ret;
> > -
> > -	WARN_ON(shmem->base.import_attach);
> > -
> > -	ret = drm_gem_shmem_get_pages(shmem);
> > -	WARN_ON_ONCE(ret != 0);
> > +	drm_gem_shmem_get_pages_no_fail(shmem);
> >   	drm_gem_vm_open(vma);
> >   }
> > @@ -716,6 +1229,8 @@ struct sg_table *drm_gem_shmem_get_pages_sgt(struct drm_gem_shmem_object *shmem)
> >   	shmem->sgt = sgt;
> > +	drm_gem_shmem_update_pages_state(shmem);
> > +
> >   	return sgt;
> >   err_free_sgt:
> > @@ -762,6 +1277,202 @@ drm_gem_shmem_prime_import_sg_table(struct drm_device *dev,
> >   }
> >   EXPORT_SYMBOL_GPL(drm_gem_shmem_prime_import_sg_table);
> > +static struct drm_gem_shmem_shrinker *
> > +to_drm_shrinker(struct shrinker *shrinker)
> > +{
> > +	return container_of(shrinker, struct drm_gem_shmem_shrinker, base);
> > +}
> > +
> > +static unsigned long
> > +drm_gem_shmem_shrinker_count_objects(struct shrinker *shrinker,
> > +				     struct shrink_control *sc)
> > +{
> > +	struct drm_gem_shmem_shrinker *gem_shrinker = to_drm_shrinker(shrinker);
> > +	u64 count = READ_ONCE(gem_shrinker->shrinkable_count);
> > +
> > +	if (count >= SHRINK_EMPTY)
> > +		return SHRINK_EMPTY - 1;
> > +
> > +	return count ?: SHRINK_EMPTY;
> > +}
> > +
> > +static unsigned long
> > +drm_gem_shmem_shrinker_run_objects_scan(struct shrinker *shrinker,
> > +					unsigned long nr_to_scan,
> > +					bool *lock_contention,
> > +					bool evict)
> > +{
> > +	struct drm_gem_shmem_shrinker *gem_shrinker = to_drm_shrinker(shrinker);
> > +	struct drm_gem_shmem_object *shmem;
> > +	struct list_head still_in_list;
> > +	struct drm_gem_object *obj;
> > +	unsigned long freed = 0;
> > +	struct list_head *lru;
> > +	size_t page_count;
> > +
> > +	INIT_LIST_HEAD(&still_in_list);
> > +
> > +	mutex_lock(&gem_shrinker->lock);
> > +
> > +	if (evict)
> > +		lru = &gem_shrinker->lru_evictable;
> > +	else
> > +		lru = &gem_shrinker->lru_purgeable;
> > +
> > +	while (freed < nr_to_scan) {
> > +		shmem = list_first_entry_or_null(lru, typeof(*shmem), madv_list);
> > +		if (!shmem)
> > +			break;
> > +
> > +		obj = &shmem->base;
> > +		page_count = obj->size >> PAGE_SHIFT;
> > +		list_move_tail(&shmem->madv_list, &still_in_list);
> > +
> > +		if (evict && get_nr_swap_pages() < page_count)
> > +			continue;
> > +
> > +		/*
> > +		 * If it's in the process of being freed, gem_object->free()
> > +		 * may be blocked on lock waiting to remove it.  So just
> > +		 * skip it.
> > +		 */
> > +		if (!kref_get_unless_zero(&obj->refcount))
> > +			continue;
> > +
> > +		mutex_unlock(&gem_shrinker->lock);
> > +
> > +		/* prevent racing with job-submission code paths */
> > +		if (!dma_resv_trylock(obj->resv)) {
> > +			*lock_contention |= true;
> > +			goto shrinker_lock;
> > +		}
> > +
> > +		/* prevent racing with the dma-buf exporting */
> > +		if (!mutex_trylock(&gem_shrinker->dev->object_name_lock)) {
> > +			*lock_contention |= true;
> > +			goto resv_unlock;
> > +		}
> > +
> > +		/* check whether h/w uses this object */
> > +		if (!dma_resv_test_signaled(obj->resv, DMA_RESV_USAGE_WRITE))
> > +			goto object_name_unlock;
> > +
> > +		/* GEM may've become unpurgeable while shrinker was unlocked */
> > +		if (evict) {
> > +			if (!drm_gem_shmem_is_evictable(shmem))
> > +				goto object_name_unlock;
> > +		} else {
> > +			if (!drm_gem_shmem_is_purgeable(shmem))
> > +				goto object_name_unlock;
> > +		}
> > +
> > +		if (evict)
> > +			freed += obj->funcs->evict(obj);
> > +		else
> > +			freed += obj->funcs->purge(obj);
> > +object_name_unlock:
> > +		mutex_unlock(&gem_shrinker->dev->object_name_lock);
> > +resv_unlock:
> > +		dma_resv_unlock(obj->resv);
> > +shrinker_lock:
> > +		drm_gem_object_put(&shmem->base);
> > +		mutex_lock(&gem_shrinker->lock);
> > +	}
> > +
> > +	list_splice_tail(&still_in_list, lru);
> > +
> > +	mutex_unlock(&gem_shrinker->lock);
> > +
> > +	return freed;
> > +}
> > +
> > +static unsigned long
> > +drm_gem_shmem_shrinker_scan_objects(struct shrinker *shrinker,
> > +				    struct shrink_control *sc)
> > +{
> > +	unsigned long nr_to_scan = sc->nr_to_scan;
> > +	bool lock_contention = false;
> > +	unsigned long freed;
> > +
> > +	/* purge as many objects as we can */
> > +	freed = drm_gem_shmem_shrinker_run_objects_scan(shrinker, nr_to_scan,
> > +							&lock_contention, false);
> > +	nr_to_scan -= freed;
> > +
> > +	/* evict as many objects as we can */
> > +	if (freed < nr_to_scan)
> > +		freed += drm_gem_shmem_shrinker_run_objects_scan(shrinker,
> > +								 nr_to_scan,
> > +								 &lock_contention,
> > +								 true);
> > +
> > +	return (!freed && !lock_contention) ? SHRINK_STOP : freed;
> > +}
> > +
> > +/**
> > + * drm_gem_shmem_shrinker_register() - Register shmem shrinker
> > + * @dev: DRM device
> > + *
> > + * Returns:
> > + * 0 on success or a negative error code on failure.
> > + */
> > +int drm_gem_shmem_shrinker_register(struct drm_device *dev)
> > +{
> > +	struct drm_gem_shmem_shrinker *gem_shrinker;
> > +	int err;
> > +
> > +	if (WARN_ON(dev->shmem_shrinker))
> > +		return -EBUSY;
> > +
> > +	gem_shrinker = kzalloc(sizeof(*gem_shrinker), GFP_KERNEL);
> > +	if (!gem_shrinker)
> > +		return -ENOMEM;
> > +
> > +	gem_shrinker->base.count_objects = drm_gem_shmem_shrinker_count_objects;
> > +	gem_shrinker->base.scan_objects = drm_gem_shmem_shrinker_scan_objects;
> > +	gem_shrinker->base.seeks = DEFAULT_SEEKS;
> > +	gem_shrinker->dev = dev;
> > +
> > +	INIT_LIST_HEAD(&gem_shrinker->lru_purgeable);
> > +	INIT_LIST_HEAD(&gem_shrinker->lru_evictable);
> > +	INIT_LIST_HEAD(&gem_shrinker->lru_evicted);
> > +	INIT_LIST_HEAD(&gem_shrinker->lru_active);
> > +	mutex_init(&gem_shrinker->lock);
> > +
> > +	dev->shmem_shrinker = gem_shrinker;
> > +
> > +	err = register_shrinker(&gem_shrinker->base);
> > +	if (err) {
> > +		dev->shmem_shrinker = NULL;
> > +		kfree(gem_shrinker);
> > +		return err;
> > +	}
> > +
> > +	return 0;
> > +}
> > +EXPORT_SYMBOL_GPL(drm_gem_shmem_shrinker_register);
> > +
> > +/**
> > + * drm_gem_shmem_shrinker_unregister() - Unregister shmem shrinker
> > + * @dev: DRM device
> > + */
> > +void drm_gem_shmem_shrinker_unregister(struct drm_device *dev)
> > +{
> > +	struct drm_gem_shmem_shrinker *gem_shrinker = dev->shmem_shrinker;
> > +
> > +	if (gem_shrinker) {
> > +		unregister_shrinker(&gem_shrinker->base);
> > +		WARN_ON(!list_empty(&gem_shrinker->lru_purgeable));
> > +		WARN_ON(!list_empty(&gem_shrinker->lru_evictable));
> > +		WARN_ON(!list_empty(&gem_shrinker->lru_evicted));
> > +		WARN_ON(!list_empty(&gem_shrinker->lru_active));
> > +		mutex_destroy(&gem_shrinker->lock);
> > +		dev->shmem_shrinker = NULL;
> > +		kfree(gem_shrinker);
> > +	}
> > +}
> > +EXPORT_SYMBOL_GPL(drm_gem_shmem_shrinker_unregister);
> > +
> >   MODULE_DESCRIPTION("DRM SHMEM memory-management helpers");
> >   MODULE_IMPORT_NS(DMA_BUF);
> >   MODULE_LICENSE("GPL v2");
> > diff --git a/include/drm/drm_device.h b/include/drm/drm_device.h
> > index 9923c7a6885e..929546cad894 100644
> > --- a/include/drm/drm_device.h
> > +++ b/include/drm/drm_device.h
> > @@ -16,6 +16,7 @@ struct drm_vblank_crtc;
> >   struct drm_vma_offset_manager;
> >   struct drm_vram_mm;
> >   struct drm_fb_helper;
> > +struct drm_gem_shmem_shrinker;
> >   struct inode;
> > @@ -277,6 +278,9 @@ struct drm_device {
> >   	/** @vram_mm: VRAM MM memory manager */
> >   	struct drm_vram_mm *vram_mm;
> > +	/** @shmem_shrinker: SHMEM GEM memory shrinker */
> > +	struct drm_gem_shmem_shrinker *shmem_shrinker;
> > +
> >   	/**
> >   	 * @switch_power_state:
> >   	 *
> > diff --git a/include/drm/drm_gem.h b/include/drm/drm_gem.h
> > index 9d7c61a122dc..390d1ce08ed3 100644
> > --- a/include/drm/drm_gem.h
> > +++ b/include/drm/drm_gem.h
> > @@ -172,6 +172,41 @@ struct drm_gem_object_funcs {
> >   	 * This is optional but necessary for mmap support.
> >   	 */
> >   	const struct vm_operations_struct *vm_ops;
> > +
> > +	/**
> > +	 * @purge:
> > +	 *
> > +	 * Releases the GEM object's allocated backing storage to the system.
> > +	 *
> > +	 * Returns the number of pages that have been freed by purging the GEM object.
> > +	 *
> > +	 * This callback is used by the GEM shrinker.
> > +	 */
> > +	unsigned long (*purge)(struct drm_gem_object *obj);
> > +
> > +	/**
> > +	 * @evict:
> > +	 *
> > +	 * Unpins the GEM object's allocated backing storage, allowing shmem pages
> > +	 * to be swapped out.
> > +	 *
> > +	 * Returns the number of pages that have been unpinned.
> > +	 *
> > +	 * This callback is used by the GEM shrinker.
> > +	 */
> > +	unsigned long (*evict)(struct drm_gem_object *obj);
> > +
> > +	/**
> > +	 * @swap_in:
> > +	 *
> > +	 * Pins GEM object's allocated backing storage if it was previously evicted,
> > +	 * moving swapped out pages back to memory.
> > +	 *
> > +	 * Returns 0 on success, or -errno on error.
> > +	 *
> > +	 * This callback is used by the GEM shrinker.
> > +	 */
> > +	int (*swap_in)(struct drm_gem_object *obj);
> >   };
> >   /**
> > diff --git a/include/drm/drm_gem_shmem_helper.h b/include/drm/drm_gem_shmem_helper.h
> > index 70889533962a..a65557b446e6 100644
> > --- a/include/drm/drm_gem_shmem_helper.h
> > +++ b/include/drm/drm_gem_shmem_helper.h
> > @@ -6,6 +6,7 @@
> >   #include <linux/fs.h>
> >   #include <linux/mm.h>
> >   #include <linux/mutex.h>
> > +#include <linux/shrinker.h>
> >   #include <drm/drm_file.h>
> >   #include <drm/drm_gem.h>
> > @@ -15,8 +16,18 @@
> >   struct dma_buf_attachment;
> >   struct drm_mode_create_dumb;
> >   struct drm_printer;
> > +struct drm_device;
> >   struct sg_table;
> > +enum drm_gem_shmem_pages_state {
> > +	DRM_GEM_SHMEM_PAGES_STATE_PURGED = -2,
> > +	DRM_GEM_SHMEM_PAGES_STATE_EVICTED = -1,
> > +	DRM_GEM_SHMEM_PAGES_STATE_UNPINNED = 0,
> > +	DRM_GEM_SHMEM_PAGES_STATE_PINNED = 1,
> > +	DRM_GEM_SHMEM_PAGES_STATE_EVICTABLE = 2,
> > +	DRM_GEM_SHMEM_PAGES_STATE_PURGEABLE = 3,
> > +};
> > +
> >   /**
> >    * struct drm_gem_shmem_object - GEM object backed by shmem
> >    */
> > @@ -43,8 +54,8 @@ struct drm_gem_shmem_object {
> >   	 * @madv: State for madvise
> >   	 *
> >   	 * 0 is active/inuse.
> > +	 * 1 is not-needed/can-be-purged
> >   	 * A negative value is the object is purged.
> > -	 * Positive values are driver specific and not used by the helpers.
> >   	 */
> >   	int madv;
> > @@ -91,6 +102,40 @@ struct drm_gem_shmem_object {
> >   	 * @map_wc: map object write-combined (instead of using shmem defaults).
> >   	 */
> >   	bool map_wc;
> > +
> > +	/**
> > +	 * @eviction_disable_count:
> > +	 *
> > +	 * The shmem pages are disallowed to be evicted by the memory shrinker
> > +	 * while count is non-zero. Used internally by memory shrinker.
> > +	 */
> > +	unsigned int eviction_disable_count;
> > +
> > +	/**
> > +	 * @purging_disable_count:
> > +	 *
> > +	 * The shmem pages are disallowed to be purged by the memory shrinker
> > +	 * while count is non-zero. Used internally by memory shrinker.
> > +	 */
> > +	unsigned int purging_disable_count;
> > +
> > +	/**
> > +	 * @pages_state: Current state of shmem pages. Used internally by
> > +	 * memory shrinker.
> > +	 */
> > +	enum drm_gem_shmem_pages_state pages_state;
> > +
> > +	/**
> > +	 * @evicted: True if shmem pages were evicted by the memory shrinker.
> > +	 * Used internally by memory shrinker.
> > +	 */
> > +	bool evicted;
> > +
> > +	/**
> > +	 * @pages_shrinkable: True if shmem pages can be evicted or purged
> > +	 * by the memory shrinker. Used internally by memory shrinker.
> > +	 */
> > +	bool pages_shrinkable;
> >   };
> >   #define to_drm_gem_shmem_obj(obj) \
> > @@ -111,15 +156,33 @@ int drm_gem_shmem_mmap(struct drm_gem_shmem_object *shmem, struct vm_area_struct
> >   int drm_gem_shmem_madvise(struct drm_gem_shmem_object *shmem, int madv);
> > +int drm_gem_shmem_set_purgeable(struct drm_gem_shmem_object *shmem);
> > +int drm_gem_shmem_set_purgeable_and_evictable(struct drm_gem_shmem_object *shmem);
> > +int drm_gem_shmem_set_unpurgeable_and_unevictable(struct drm_gem_shmem_object *shmem);
> > +
> > +static inline bool drm_gem_shmem_is_evictable(struct drm_gem_shmem_object *shmem)
> > +{
> > +	return (shmem->madv >= 0) && !shmem->eviction_disable_count &&
> > +		shmem->base.funcs->evict && shmem->base.funcs->swap_in &&
> > +		!shmem->vmap_use_count && !shmem->base.dma_buf &&
> > +		!shmem->base.import_attach && shmem->sgt;
> > +}
> > +
> >   static inline bool drm_gem_shmem_is_purgeable(struct drm_gem_shmem_object *shmem)
> >   {
> > -	return (shmem->madv > 0) &&
> > -		!shmem->vmap_use_count && shmem->sgt &&
> > -		!shmem->base.dma_buf && !shmem->base.import_attach;
> > +	return (shmem->madv > 0) && !shmem->purging_disable_count &&
> > +		!shmem->vmap_use_count && shmem->base.funcs->purge &&
> > +		!shmem->base.dma_buf && !shmem->base.import_attach &&
> > +		shmem->sgt;
> >   }
> > -void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem);
> > +int drm_gem_shmem_swap_in_pages_locked(struct drm_gem_shmem_object *shmem);
> > +int drm_gem_shmem_swap_in_locked(struct drm_gem_shmem_object *shmem);
> > +
> > +void drm_gem_shmem_evict_locked(struct drm_gem_shmem_object *shmem);
> > +
> >   bool drm_gem_shmem_purge(struct drm_gem_shmem_object *shmem);
> > +void drm_gem_shmem_purge_locked(struct drm_gem_shmem_object *shmem);
> >   struct sg_table *drm_gem_shmem_get_sg_table(struct drm_gem_shmem_object *shmem);
> >   struct sg_table *drm_gem_shmem_get_pages_sgt(struct drm_gem_shmem_object *shmem);
> > @@ -262,6 +325,38 @@ static inline int drm_gem_shmem_object_mmap(struct drm_gem_object *obj, struct v
> >   	return drm_gem_shmem_mmap(shmem, vma);
> >   }
> > +/**
> > + * struct drm_gem_shmem_shrinker - Generic memory shrinker for shmem GEMs
> > + */
> > +struct drm_gem_shmem_shrinker {
> > +	/** @base: Shrinker for purging shmem GEM objects */
> > +	struct shrinker base;
> > +
> > +	/** @lock: Protects @lru_* */
> > +	struct mutex lock;
> > +
> > +	/** @lru_purgeable: List of shmem GEM objects available for purging */
> > +	struct list_head lru_purgeable;
> > +
> > +	/** @lru_active: List of active shmem GEM objects */
> > +	struct list_head lru_active;
> > +
> > +	/** @lru_evictable: List of shmem GEM objects that can be evicted */
> > +	struct list_head lru_evictable;
> > +
> > +	/** @lru_evicted: List of evicted shmem GEM objects */
> > +	struct list_head lru_evicted;
> > +
> > +	/** @dev: DRM device that uses this shrinker */
> > +	struct drm_device *dev;
> > +
> > +	/** @shrinkable_count: Count of shmem GEM pages to be purged and evicted */
> > +	u64 shrinkable_count;
> > +};
> > +
> > +int drm_gem_shmem_shrinker_register(struct drm_device *dev);
> > +void drm_gem_shmem_shrinker_unregister(struct drm_device *dev);
> > +
> >   /*
> >    * Driver ops
> >    */
> 
> -- 
> Thomas Zimmermann
> Graphics Driver Developer
> SUSE Software Solutions Germany GmbH
> Maxfeldstr. 5, 90409 Nürnberg, Germany
> (HRB 36809, AG Nürnberg)
> Geschäftsführer: Ivo Totev




-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 11/15] drm/shmem-helper: Add generic memory shrinker

[Daniel Vetter]

On Fri, May 06, 2022 at 03:10:43AM +0300, Dmitry Osipenko wrote:
> On 5/5/22 11:34, Thomas Zimmermann wrote:
> > Hi
> > 
> > Am 18.04.22 um 00:37 schrieb Dmitry Osipenko:
> >> Introduce a common DRM SHMEM shrinker. It allows to reduce code
> >> duplication among DRM drivers that implement theirs own shrinkers.
> >> This is initial version of the shrinker that covers basic needs of
> >> GPU drivers, both purging and eviction of shmem objects are supported.
> >>
> >> This patch is based on a couple ideas borrowed from Rob's Clark MSM
> >> shrinker and Thomas' Zimmermann variant of SHMEM shrinker.
> >>
> >> In order to start using DRM SHMEM shrinker drivers should:
> >>
> >> 1. Implement new purge(), evict() + swap_in() GEM callbacks.
> >> 2. Register shrinker using drm_gem_shmem_shrinker_register(drm_device).
> >> 3. Use drm_gem_shmem_set_purgeable_and_evictable(shmem) and alike API
> >>     functions to activate shrinking of GEMs.
> > 
> > Honestly speaking, after reading the patch and the discussion here I
> > really don't like where all tis is going. The interfaces and
> > implementation are overengineered.  Descisions about evicting and
> > purging should be done by the memory manager. For the most part, it's
> > none of the driver's business.
> 
> Daniel mostly suggesting to make interface more flexible for future
> drivers, so we won't need to re-do it later on. My version of the
> interface is based on what drivers need today.
> 
> Why do you think it's a problem to turn shmem helper into the simple
> generic memory manager? I don't see how it's better to have drivers
> duplicating the exactly same efforts and making different mistakes.
> 
> The shmem shrinker implementation is mostly based on the freedreno's
> shrinker and it's very easy to enable generic shrinker for VirtIO and
> Panfrost drivers. I think in the future freedreno and other drivers
> could switch to use drm shmem instead of open coding the memory management.

Yeah I think we have enough shrinkers all over drm to actually design
something solid here.

There's also the i915 shrinker and some kinda shrinker in ttm too. So we
are definitely past the "have 3 examples to make sure you design something
solid" rule of thumb.

I also have a bit an idea that we could try to glue the shmem shrinker
into ttm, at least at a very high level that's something that would make
some sense.
 
> > I'd like to ask you to reduce the scope of the patchset and build the
> > shrinker only for virtio-gpu. I know that I first suggested to build
> > upon shmem helpers, but it seems that it's easier to do that in a later
> > patchset.
> 
> The first version of the VirtIO shrinker didn't support memory eviction.
> Memory eviction support requires page fault handler to be aware of the
> evicted pages, what should we do about it? The page fault handling is a
> part of memory management, hence to me drm-shmem is already kinda a MM.

Hm I still don't get that part, why does that also not go through the
shmem helpers? I'm still confused why drivers need to know the difference
between evition and purging. Or maybe I'm confused again.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 11/15] drm/shmem-helper: Add generic memory shrinker

[Daniel Vetter]

On Tue, May 10, 2022 at 04:47:52PM +0300, Dmitry Osipenko wrote:
> On 5/9/22 16:49, Daniel Vetter wrote:
> > On Fri, May 06, 2022 at 03:10:43AM +0300, Dmitry Osipenko wrote:
> >> On 5/5/22 11:34, Thomas Zimmermann wrote:
> >>> Hi
> >>>
> >>> Am 18.04.22 um 00:37 schrieb Dmitry Osipenko:
> >>>> Introduce a common DRM SHMEM shrinker. It allows to reduce code
> >>>> duplication among DRM drivers that implement theirs own shrinkers.
> >>>> This is initial version of the shrinker that covers basic needs of
> >>>> GPU drivers, both purging and eviction of shmem objects are supported.
> >>>>
> >>>> This patch is based on a couple ideas borrowed from Rob's Clark MSM
> >>>> shrinker and Thomas' Zimmermann variant of SHMEM shrinker.
> >>>>
> >>>> In order to start using DRM SHMEM shrinker drivers should:
> >>>>
> >>>> 1. Implement new purge(), evict() + swap_in() GEM callbacks.
> >>>> 2. Register shrinker using drm_gem_shmem_shrinker_register(drm_device).
> >>>> 3. Use drm_gem_shmem_set_purgeable_and_evictable(shmem) and alike API
> >>>>     functions to activate shrinking of GEMs.
> >>>
> >>> Honestly speaking, after reading the patch and the discussion here I
> >>> really don't like where all tis is going. The interfaces and
> >>> implementation are overengineered.  Descisions about evicting and
> >>> purging should be done by the memory manager. For the most part, it's
> >>> none of the driver's business.
> >>
> >> Daniel mostly suggesting to make interface more flexible for future
> >> drivers, so we won't need to re-do it later on. My version of the
> >> interface is based on what drivers need today.
> >>
> >> Why do you think it's a problem to turn shmem helper into the simple
> >> generic memory manager? I don't see how it's better to have drivers
> >> duplicating the exactly same efforts and making different mistakes.
> >>
> >> The shmem shrinker implementation is mostly based on the freedreno's
> >> shrinker and it's very easy to enable generic shrinker for VirtIO and
> >> Panfrost drivers. I think in the future freedreno and other drivers
> >> could switch to use drm shmem instead of open coding the memory management.
> > 
> > Yeah I think we have enough shrinkers all over drm to actually design
> > something solid here.
> > 
> > There's also the i915 shrinker and some kinda shrinker in ttm too. So we
> > are definitely past the "have 3 examples to make sure you design something
> > solid" rule of thumb.
> > 
> > I also have a bit an idea that we could try to glue the shmem shrinker
> > into ttm, at least at a very high level that's something that would make
> > some sense.
> 
> Before gluing the shmem shrinker into ttm, the drivers should be
> switched to ttm? Or do you mean something else by the gluing?

No, drivers which don't need ttm shouldn't be forced to use it.

> Perhaps it should be possible to have a common drm-shrinker helper that
> will do the basic-common things like tracking the eviction size and
> check whether BO is exported or locked, but we shouldn't consider doing
> this for now. For the starter more reasonable should be to create a
> common shrinker base for drivers that use drm-shmem, IMO.

Yeah that might be the more practical approach. But really this was just
an aside, absolutely no need to worry about this for now. I just wanted to
point out that there really is a lot of use for this.

> >>> I'd like to ask you to reduce the scope of the patchset and build the
> >>> shrinker only for virtio-gpu. I know that I first suggested to build
> >>> upon shmem helpers, but it seems that it's easier to do that in a later
> >>> patchset.
> >>
> >> The first version of the VirtIO shrinker didn't support memory eviction.
> >> Memory eviction support requires page fault handler to be aware of the
> >> evicted pages, what should we do about it? The page fault handling is a
> >> part of memory management, hence to me drm-shmem is already kinda a MM.
> > 
> > Hm I still don't get that part, why does that also not go through the
> > shmem helpers?
> 
> The drm_gem_shmem_vm_ops includes the page faults handling, it's a
> helper by itself that is used by DRM drivers.
> 
> I could try to move all the shrinker logic to the VirtIO and re-invent
> virtio_gem_shmem_vm_ops, but what is the point of doing this for each
> driver if we could have it once and for all in the common drm-shmem code?
> 
> Maybe I should try to factor out all the shrinker logic from drm-shmem
> into a new drm-shmem-shrinker that could be shared by drivers? Will you
> be okay with this option?

I think we're talking past each another a bit. I'm only bringing up the
purge vs eviction topic we discussed in the other subthread again.

> > I'm still confused why drivers need to know the difference
> > between evition and purging. Or maybe I'm confused again.
> 
> Example:
> 
> If userspace uses IOV addresses, then these addresses must be kept
> reserved while buffer is evicted.
> 
> If BO is purged, then we don't need to retain the IOV space allocated
> for the purged BO.

Yeah but is that actually needed by anyone? If userspace fails to allocate
another bo because of lack of gpu address space then it's very easy to
handle that:

1. Make a rule that "out of gpu address space" gives you a special errno
code like ENOSPC

2. If userspace gets that it walks the list of all buffers it marked as
purgeable and nukes them (whether they have been evicted or not). Then it
retries the bo allocation.

Alternatively you can do step 2 also directly from the bo alloc ioctl in
step 1. Either way you clean up va space, and actually a lot more (you
potentially nuke all buffers marked as purgeable, not just the ones that
have been purged already) and only when va cleanup is actually needed

Trying to solve this problem at eviction time otoh means:
- we have this difference between eviction and purging
- it's still not complete, you still need to glue step 2 above into your
  driver somehow, and once step 2 above is glued in doing additional
  cleanup in the purge function is just duplicated logic

So at least in my opinion this isn't the justification we need. And we
should definitely not just add that complication "in case, for the
future", if we don't have a real need right now. Adding it later on is
easy, removing it later on because it just gets in the way and confuses is
much harder.

> The drm-shmem only handles shmem pages, not the mappings of these pages.

Yeah that's why you need an evict callback into the driver. That part is
clear.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 11/15] drm/shmem-helper: Add generic memory shrinker

[Daniel Vetter]

On Wed, May 11, 2022 at 07:06:18PM +0300, Dmitry Osipenko wrote:
> On 5/11/22 16:09, Daniel Vetter wrote:
> >>>>> I'd like to ask you to reduce the scope of the patchset and build the
> >>>>> shrinker only for virtio-gpu. I know that I first suggested to build
> >>>>> upon shmem helpers, but it seems that it's easier to do that in a later
> >>>>> patchset.
> >>>> The first version of the VirtIO shrinker didn't support memory eviction.
> >>>> Memory eviction support requires page fault handler to be aware of the
> >>>> evicted pages, what should we do about it? The page fault handling is a
> >>>> part of memory management, hence to me drm-shmem is already kinda a MM.
> >>> Hm I still don't get that part, why does that also not go through the
> >>> shmem helpers?
> >> The drm_gem_shmem_vm_ops includes the page faults handling, it's a
> >> helper by itself that is used by DRM drivers.
> >>
> >> I could try to move all the shrinker logic to the VirtIO and re-invent
> >> virtio_gem_shmem_vm_ops, but what is the point of doing this for each
> >> driver if we could have it once and for all in the common drm-shmem code?
> >>
> >> Maybe I should try to factor out all the shrinker logic from drm-shmem
> >> into a new drm-shmem-shrinker that could be shared by drivers? Will you
> >> be okay with this option?
> > I think we're talking past each another a bit. I'm only bringing up the
> > purge vs eviction topic we discussed in the other subthread again.
> 
> Thomas asked to move the whole shrinker code to the VirtIO driver and
> I's saying that this is not a great idea to me, or am I misunderstanding
> the Thomas' suggestion? Thomas?

I think it was just me creating a confusion here.

fwiw I do also think that shrinker in shmem helpers makes sense, just in
case that was also lost in confusion.

> >>> I'm still confused why drivers need to know the difference
> >>> between evition and purging. Or maybe I'm confused again.
> >> Example:
> >>
> >> If userspace uses IOV addresses, then these addresses must be kept
> >> reserved while buffer is evicted.
> >>
> >> If BO is purged, then we don't need to retain the IOV space allocated
> >> for the purged BO.
> > Yeah but is that actually needed by anyone? If userspace fails to allocate
> > another bo because of lack of gpu address space then it's very easy to
> > handle that:
> > 
> > 1. Make a rule that "out of gpu address space" gives you a special errno
> > code like ENOSPC
> > 
> > 2. If userspace gets that it walks the list of all buffers it marked as
> > purgeable and nukes them (whether they have been evicted or not). Then it
> > retries the bo allocation.
> > 
> > Alternatively you can do step 2 also directly from the bo alloc ioctl in
> > step 1. Either way you clean up va space, and actually a lot more (you
> > potentially nuke all buffers marked as purgeable, not just the ones that
> > have been purged already) and only when va cleanup is actually needed
> > 
> > Trying to solve this problem at eviction time otoh means:
> > - we have this difference between eviction and purging
> > - it's still not complete, you still need to glue step 2 above into your
> >   driver somehow, and once step 2 above is glued in doing additional
> >   cleanup in the purge function is just duplicated logic
> > 
> > So at least in my opinion this isn't the justification we need. And we
> > should definitely not just add that complication "in case, for the
> > future", if we don't have a real need right now. Adding it later on is
> > easy, removing it later on because it just gets in the way and confuses is
> > much harder.
> 
> The IOVA space is only one example.
> 
> In case of the VirtIO driver, we may have two memory allocation for a
> BO. One is the shmem allcation in guest and the other is in host's vram.
> If we will only release the guest's memory on purge, then the vram will
> remain allocated until BO is destroyed, which unnecessarily sub-optimal.

Hm but why don't you just nuke the memory on the host side too when you
evict? Allowing the guest memory to be swapped out while keeping the host
memory allocation alive also doesn't make a lot of sense for me. Both can
be recreated (I guess at least?) on swap-in.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 11/15] drm/shmem-helper: Add generic memory shrinker

[Daniel Vetter]

On Thu, 12 May 2022 at 13:36, Dmitry Osipenko
<dmitry.osipenko@collabora.com> wrote:
>
> On 5/11/22 22:09, Daniel Vetter wrote:
> > On Wed, May 11, 2022 at 07:06:18PM +0300, Dmitry Osipenko wrote:
> >> On 5/11/22 16:09, Daniel Vetter wrote:
> >>>>>>> I'd like to ask you to reduce the scope of the patchset and build the
> >>>>>>> shrinker only for virtio-gpu. I know that I first suggested to build
> >>>>>>> upon shmem helpers, but it seems that it's easier to do that in a later
> >>>>>>> patchset.
> >>>>>> The first version of the VirtIO shrinker didn't support memory eviction.
> >>>>>> Memory eviction support requires page fault handler to be aware of the
> >>>>>> evicted pages, what should we do about it? The page fault handling is a
> >>>>>> part of memory management, hence to me drm-shmem is already kinda a MM.
> >>>>> Hm I still don't get that part, why does that also not go through the
> >>>>> shmem helpers?
> >>>> The drm_gem_shmem_vm_ops includes the page faults handling, it's a
> >>>> helper by itself that is used by DRM drivers.
> >>>>
> >>>> I could try to move all the shrinker logic to the VirtIO and re-invent
> >>>> virtio_gem_shmem_vm_ops, but what is the point of doing this for each
> >>>> driver if we could have it once and for all in the common drm-shmem code?
> >>>>
> >>>> Maybe I should try to factor out all the shrinker logic from drm-shmem
> >>>> into a new drm-shmem-shrinker that could be shared by drivers? Will you
> >>>> be okay with this option?
> >>> I think we're talking past each another a bit. I'm only bringing up the
> >>> purge vs eviction topic we discussed in the other subthread again.
> >>
> >> Thomas asked to move the whole shrinker code to the VirtIO driver and
> >> I's saying that this is not a great idea to me, or am I misunderstanding
> >> the Thomas' suggestion? Thomas?
> >
> > I think it was just me creating a confusion here.
> >
> > fwiw I do also think that shrinker in shmem helpers makes sense, just in
> > case that was also lost in confusion.
>
> Okay, good that we're on the same page now.
>
> >>>>> I'm still confused why drivers need to know the difference
> >>>>> between evition and purging. Or maybe I'm confused again.
> >>>> Example:
> >>>>
> >>>> If userspace uses IOV addresses, then these addresses must be kept
> >>>> reserved while buffer is evicted.
> >>>>
> >>>> If BO is purged, then we don't need to retain the IOV space allocated
> >>>> for the purged BO.
> >>> Yeah but is that actually needed by anyone? If userspace fails to allocate
> >>> another bo because of lack of gpu address space then it's very easy to
> >>> handle that:
> >>>
> >>> 1. Make a rule that "out of gpu address space" gives you a special errno
> >>> code like ENOSPC
> >>>
> >>> 2. If userspace gets that it walks the list of all buffers it marked as
> >>> purgeable and nukes them (whether they have been evicted or not). Then it
> >>> retries the bo allocation.
> >>>
> >>> Alternatively you can do step 2 also directly from the bo alloc ioctl in
> >>> step 1. Either way you clean up va space, and actually a lot more (you
> >>> potentially nuke all buffers marked as purgeable, not just the ones that
> >>> have been purged already) and only when va cleanup is actually needed
> >>>
> >>> Trying to solve this problem at eviction time otoh means:
> >>> - we have this difference between eviction and purging
> >>> - it's still not complete, you still need to glue step 2 above into your
> >>>   driver somehow, and once step 2 above is glued in doing additional
> >>>   cleanup in the purge function is just duplicated logic
> >>>
> >>> So at least in my opinion this isn't the justification we need. And we
> >>> should definitely not just add that complication "in case, for the
> >>> future", if we don't have a real need right now. Adding it later on is
> >>> easy, removing it later on because it just gets in the way and confuses is
> >>> much harder.
> >>
> >> The IOVA space is only one example.
> >>
> >> In case of the VirtIO driver, we may have two memory allocation for a
> >> BO. One is the shmem allcation in guest and the other is in host's vram.
> >> If we will only release the guest's memory on purge, then the vram will
> >> remain allocated until BO is destroyed, which unnecessarily sub-optimal.
> >
> > Hm but why don't you just nuke the memory on the host side too when you
> > evict? Allowing the guest memory to be swapped out while keeping the host
> > memory allocation alive also doesn't make a lot of sense for me. Both can
> > be recreated (I guess at least?) on swap-in.
>
> Shouldn't be very doable or at least worth the efforts. It's userspace
> that manages data uploading, kernel only provides transport for the
> virtio-gpu commands.
>
> Drivers are free to use the same function for both purge() and evict()
> callbacks if they want. Getting rid of the purge() callback creates more
> problems than solves, IMO.

Hm this still sounds pretty funny and defeats the point of
purgeable/evictable buffers a bit I think. But also I guess we'd
pushed this bikeshed to the max, so I think if you make ->purge
optional and just call ->evict if that's not present, and document it
all in the kerneldoc, then I think that's good.

I just don't think that encouraging drivers to distinguish between
evict/purge is a good idea for almost all of them.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH v4 11/15] drm/shmem-helper: Add generic memory shrinker

[Daniel Vetter]

On Thu, May 12, 2022 at 10:04:53PM +0300, Dmitry Osipenko wrote:
> On 5/12/22 20:04, Daniel Vetter wrote:
> > On Thu, 12 May 2022 at 13:36, Dmitry Osipenko
> > <dmitry.osipenko@collabora.com> wrote:
> >>
> >> On 5/11/22 22:09, Daniel Vetter wrote:
> >>> On Wed, May 11, 2022 at 07:06:18PM +0300, Dmitry Osipenko wrote:
> >>>> On 5/11/22 16:09, Daniel Vetter wrote:
> >>>>>>>>> I'd like to ask you to reduce the scope of the patchset and build the
> >>>>>>>>> shrinker only for virtio-gpu. I know that I first suggested to build
> >>>>>>>>> upon shmem helpers, but it seems that it's easier to do that in a later
> >>>>>>>>> patchset.
> >>>>>>>> The first version of the VirtIO shrinker didn't support memory eviction.
> >>>>>>>> Memory eviction support requires page fault handler to be aware of the
> >>>>>>>> evicted pages, what should we do about it? The page fault handling is a
> >>>>>>>> part of memory management, hence to me drm-shmem is already kinda a MM.
> >>>>>>> Hm I still don't get that part, why does that also not go through the
> >>>>>>> shmem helpers?
> >>>>>> The drm_gem_shmem_vm_ops includes the page faults handling, it's a
> >>>>>> helper by itself that is used by DRM drivers.
> >>>>>>
> >>>>>> I could try to move all the shrinker logic to the VirtIO and re-invent
> >>>>>> virtio_gem_shmem_vm_ops, but what is the point of doing this for each
> >>>>>> driver if we could have it once and for all in the common drm-shmem code?
> >>>>>>
> >>>>>> Maybe I should try to factor out all the shrinker logic from drm-shmem
> >>>>>> into a new drm-shmem-shrinker that could be shared by drivers? Will you
> >>>>>> be okay with this option?
> >>>>> I think we're talking past each another a bit. I'm only bringing up the
> >>>>> purge vs eviction topic we discussed in the other subthread again.
> >>>>
> >>>> Thomas asked to move the whole shrinker code to the VirtIO driver and
> >>>> I's saying that this is not a great idea to me, or am I misunderstanding
> >>>> the Thomas' suggestion? Thomas?
> >>>
> >>> I think it was just me creating a confusion here.
> >>>
> >>> fwiw I do also think that shrinker in shmem helpers makes sense, just in
> >>> case that was also lost in confusion.
> >>
> >> Okay, good that we're on the same page now.
> >>
> >>>>>>> I'm still confused why drivers need to know the difference
> >>>>>>> between evition and purging. Or maybe I'm confused again.
> >>>>>> Example:
> >>>>>>
> >>>>>> If userspace uses IOV addresses, then these addresses must be kept
> >>>>>> reserved while buffer is evicted.
> >>>>>>
> >>>>>> If BO is purged, then we don't need to retain the IOV space allocated
> >>>>>> for the purged BO.
> >>>>> Yeah but is that actually needed by anyone? If userspace fails to allocate
> >>>>> another bo because of lack of gpu address space then it's very easy to
> >>>>> handle that:
> >>>>>
> >>>>> 1. Make a rule that "out of gpu address space" gives you a special errno
> >>>>> code like ENOSPC
> >>>>>
> >>>>> 2. If userspace gets that it walks the list of all buffers it marked as
> >>>>> purgeable and nukes them (whether they have been evicted or not). Then it
> >>>>> retries the bo allocation.
> >>>>>
> >>>>> Alternatively you can do step 2 also directly from the bo alloc ioctl in
> >>>>> step 1. Either way you clean up va space, and actually a lot more (you
> >>>>> potentially nuke all buffers marked as purgeable, not just the ones that
> >>>>> have been purged already) and only when va cleanup is actually needed
> >>>>>
> >>>>> Trying to solve this problem at eviction time otoh means:
> >>>>> - we have this difference between eviction and purging
> >>>>> - it's still not complete, you still need to glue step 2 above into your
> >>>>>   driver somehow, and once step 2 above is glued in doing additional
> >>>>>   cleanup in the purge function is just duplicated logic
> >>>>>
> >>>>> So at least in my opinion this isn't the justification we need. And we
> >>>>> should definitely not just add that complication "in case, for the
> >>>>> future", if we don't have a real need right now. Adding it later on is
> >>>>> easy, removing it later on because it just gets in the way and confuses is
> >>>>> much harder.
> >>>>
> >>>> The IOVA space is only one example.
> >>>>
> >>>> In case of the VirtIO driver, we may have two memory allocation for a
> >>>> BO. One is the shmem allcation in guest and the other is in host's vram.
> >>>> If we will only release the guest's memory on purge, then the vram will
> >>>> remain allocated until BO is destroyed, which unnecessarily sub-optimal.
> >>>
> >>> Hm but why don't you just nuke the memory on the host side too when you
> >>> evict? Allowing the guest memory to be swapped out while keeping the host
> >>> memory allocation alive also doesn't make a lot of sense for me. Both can
> >>> be recreated (I guess at least?) on swap-in.
> >>
> >> Shouldn't be very doable or at least worth the efforts. It's userspace
> >> that manages data uploading, kernel only provides transport for the
> >> virtio-gpu commands.
> >>
> >> Drivers are free to use the same function for both purge() and evict()
> >> callbacks if they want. Getting rid of the purge() callback creates more
> >> problems than solves, IMO.
> > 
> > Hm this still sounds pretty funny and defeats the point of
> > purgeable/evictable buffers a bit I think. But also I guess we'd
> > pushed this bikeshed to the max, so I think if you make ->purge
> > optional and just call ->evict if that's not present, and document it
> > all in the kerneldoc, then I think that's good.
> 
> This is a good enough compromise to me.
> 
> > I just don't think that encouraging drivers to distinguish between
> > evict/purge is a good idea for almost all of them.
> 
> Intel's shrinker checks the "madvise" status of BOs and then decides
> what to do based on it. Perhaps we could move the decision-making about
> purging to drivers and then it will be single evict() callback, but will
> drivers really ever need to be responsible for this decision-making or
> this will be an unnecessary boilerplate code in the drivers? I'll think
> more about this.

tbh I wouldn't worry about details, you've convinced me that some
differentiation between evict and purged makes sense. And yeah maybe
drivers should have a helper to check that instead of explicit argument,
but that's a bikeshed color choice which should be fairly easy to adjust
later on still.

> Thank you all for taking time to look at this patchset. I'm preparing
> the new version.

Cheers, Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization