Re: [RFC PATCH v2 2/6] drm/shmem: Introduce the notion of sparse objects

[Boris Brezillon <boris.brezillon@collabora.com>]

On Wed, 26 Mar 2025 02:14:22 +0000
Adrián Larumbe <adrian.larumbe@collabora.com> wrote:

> Sparse DRM objects will store their backing pages in an xarray, to avoid
> the overhead of preallocating a huge struct page pointer array when only
> a very small range of indices might be assigned.
> 
> For now, only the definition of a sparse object as a union alternative
> to a 'dense' object is provided, with functions that exploit it being
> part of later commits.
> 
> Signed-off-by: Adrián Larumbe <adrian.larumbe@collabora.com>
> ---
>  drivers/gpu/drm/drm_gem_shmem_helper.c | 68 +++++++++++++++++++++++++-
>  include/drm/drm_gem_shmem_helper.h     | 23 ++++++++-
>  2 files changed, 88 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_gem_shmem_helper.c b/drivers/gpu/drm/drm_gem_shmem_helper.c
> index d99dee67353a..5f75eb1230f6 100644
> --- a/drivers/gpu/drm/drm_gem_shmem_helper.c
> +++ b/drivers/gpu/drm/drm_gem_shmem_helper.c
> @@ -128,6 +128,31 @@ struct drm_gem_shmem_object *drm_gem_shmem_create(struct drm_device *dev, size_t
>  }
>  EXPORT_SYMBOL_GPL(drm_gem_shmem_create);
>  
> +/**
> + * drm_gem_shmem_create_sparse - Allocate a sparse object with the given size
> + * @dev: DRM device
> + * @size: Size of the sparse object to allocate
> + *
> + * This function creates a sparse shmem GEM object.
> + *
> + * Returns:
> + * A struct drm_gem_shmem_object * on success or an ERR_PTR()-encoded negative
> + * error code on failure.
> + */
> +struct drm_gem_shmem_object *drm_gem_shmem_create_sparse(struct drm_device *dev, size_t size)
> +{
> +	struct drm_gem_shmem_object *shmem =
> +		__drm_gem_shmem_create(dev, size, false, NULL);
> +
> +	if (!IS_ERR(shmem)) {
> +		shmem->sparse = true;
> +		xa_init_flags(&shmem->xapages, XA_FLAGS_ALLOC);
> +	}
> +
> +	return shmem;
> +}
> +EXPORT_SYMBOL_GPL(drm_gem_shmem_create_sparse);
> +
>  /**
>   * drm_gem_shmem_create_with_mnt - Allocate an object with the given size in a
>   * given mountpoint
> @@ -173,8 +198,8 @@ void drm_gem_shmem_free(struct drm_gem_shmem_object *shmem)
>  			sg_free_table(shmem->sgt);
>  			kfree(shmem->sgt);
>  		}
> -		if (shmem->pages)
> -			drm_gem_shmem_put_pages(shmem);
> +
> +		drm_gem_shmem_put_pages(shmem);
>  
>  		drm_WARN_ON(obj->dev, shmem->pages_use_count);
>  
> @@ -196,6 +221,12 @@ static int drm_gem_shmem_get_pages(struct drm_gem_shmem_object *shmem)
>  	if (shmem->pages_use_count++ > 0)
>  		return 0;
>  
> +	/* We only allow increasing the user count in the case of
> +	 * sparse shmem objects with some backed pages for now
> +	 */
> +	if (shmem->sparse && xa_empty(&shmem->xapages))
> +		return -EINVAL;
> +
>  	pages = drm_gem_get_pages(obj);
>  	if (IS_ERR(pages)) {
>  		drm_dbg_kms(obj->dev, "Failed to get pages (%ld)\n",
> @@ -231,6 +262,14 @@ void drm_gem_shmem_put_pages(struct drm_gem_shmem_object *shmem)
>  
>  	dma_resv_assert_held(shmem->base.resv);
>  
> +	if (!shmem->sparse) {
> +		if (!shmem->pages)
> +			return;
> +	} else {
> +		/* Not implemented yet */
> +		return;
> +	}
> +
>  	if (drm_WARN_ON_ONCE(obj->dev, !shmem->pages_use_count))
>  		return;
>  
> @@ -404,8 +443,15 @@ void drm_gem_shmem_vunmap(struct drm_gem_shmem_object *shmem,
>  {
>  	struct drm_gem_object *obj = &shmem->base;
>  
> +	if (shmem->sparse) {
> +		drm_err(obj->dev, "UM unmapping of sparse shmem objects not implemented\n");
> +		return;
> +	}
> +
>  	if (drm_gem_is_imported(obj)) {
>  		dma_buf_vunmap(obj->dma_buf, map);
> +	} else if (obj->import_attach) {
> +		dma_buf_vunmap(obj->import_attach->dmabuf, map);
>  	} else {
>  		dma_resv_assert_held(shmem->base.resv);
>  
> @@ -541,6 +587,12 @@ static vm_fault_t drm_gem_shmem_fault(struct vm_fault *vmf)
>  	struct page *page;
>  	pgoff_t page_offset;
>  
> +	/* TODO: Implement UM mapping of sparse shmem objects */
> +	if (drm_WARN_ON(obj->dev, shmem->sparse)) {
> +		drm_err(obj->dev, "UM mapping of sparse shmem objects not implemented\n");
> +		return VM_FAULT_SIGBUS;
> +	}
> +
>  	/* We don't use vmf->pgoff since that has the fake offset */
>  	page_offset = (vmf->address - vma->vm_start) >> PAGE_SHIFT;
>  
> @@ -566,8 +618,14 @@ 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);
>  
> +	/* TODO: Implement UM mapping of sparse shmem objects */
> +	if (drm_WARN_ON(obj->dev, shmem->sparse))
> +		return;
> +
>  	drm_WARN_ON(obj->dev, drm_gem_is_imported(obj));
>  
> +	drm_WARN_ON(obj->dev, obj->import_attach);
> +
>  	dma_resv_lock(shmem->base.resv, NULL);
>  
>  	/*
> @@ -690,6 +748,9 @@ struct sg_table *drm_gem_shmem_get_sg_table(struct drm_gem_shmem_object *shmem)
>  {
>  	struct drm_gem_object *obj = &shmem->base;
>  
> +	if (drm_WARN_ON(obj->dev, shmem->sparse))
> +		return ERR_PTR(-EINVAL);
> +
>  	drm_WARN_ON(obj->dev, drm_gem_is_imported(obj));
>  
>  	return drm_prime_pages_to_sg(obj->dev, shmem->pages, obj->size >> PAGE_SHIFT);
> @@ -702,6 +763,9 @@ static struct sg_table *drm_gem_shmem_get_pages_sgt_locked(struct drm_gem_shmem_
>  	int ret;
>  	struct sg_table *sgt;
>  
> +	if (drm_WARN_ON(obj->dev, shmem->sparse))
> +		return ERR_PTR(-EINVAL);
> +
>  	if (shmem->sgt)
>  		return shmem->sgt;
>  
> diff --git a/include/drm/drm_gem_shmem_helper.h b/include/drm/drm_gem_shmem_helper.h
> index cef5a6b5a4d6..00e47512b30f 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/xarray.h>
>  
>  #include <drm/drm_file.h>
>  #include <drm/drm_gem.h>
> @@ -29,7 +30,10 @@ struct drm_gem_shmem_object {
>  	/**
>  	 * @pages: Page table
>  	 */
> -	struct page **pages;
> +	union {
> +		struct page **pages;
> +		struct xarray xapages;
> +	};

I've played with this code, and I must admit I'm not a huge fan of this
union. It's just super easy to forget an

   if (is_xarray) do_this else do_that

in one of the path, and end up accessing the wrong type without even
noticing (or noticing late).

I'd rather go for an drm_gem_shmem_sparse_backing object, and have an
optional pointer to this sparse_backing object in drm_gem_shmem_object.
I actually tried this approach here [1], and it seems to work.

>  
>  	/**
>  	 * @pages_use_count:
> @@ -91,12 +95,18 @@ struct drm_gem_shmem_object {
>  	 * @map_wc: map object write-combined (instead of using shmem defaults).
>  	 */
>  	bool map_wc : 1;
> +
> +	/**
> +	 * @sparse: the object is only partially backed by pages
> +	 */
> +	bool sparse : 1;
>  };
>  
>  #define to_drm_gem_shmem_obj(obj) \
>  	container_of(obj, struct drm_gem_shmem_object, base)
>  
>  struct drm_gem_shmem_object *drm_gem_shmem_create(struct drm_device *dev, size_t size);
> +struct drm_gem_shmem_object *drm_gem_shmem_create_sparse(struct drm_device *dev, size_t size);
>  struct drm_gem_shmem_object *drm_gem_shmem_create_with_mnt(struct drm_device *dev,
>  							   size_t size,
>  							   struct vfsmount *gemfs);
> @@ -210,6 +220,10 @@ static inline struct sg_table *drm_gem_shmem_object_get_sg_table(struct drm_gem_
>  {
>  	struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
>  
> +	/* Use the specific sparse shmem get_sg_table function instead */
> +	if (WARN_ON(shmem->sparse))
> +		return ERR_PTR(-EINVAL);
> +
>  	return drm_gem_shmem_get_sg_table(shmem);
>  }
>  
> @@ -229,6 +243,10 @@ static inline int drm_gem_shmem_object_vmap(struct drm_gem_object *obj,
>  {
>  	struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
>  
> +	/* TODO: Implement kernel mapping of sparse shmem objects */
> +	if (WARN_ON(shmem->sparse))
> +		return -EACCES;
> +

Okay, this is where things start to get messy. Sparse objects support a
subset of the features supported by regular shmem objects, which
not only makes the code error-prone, but makes it hard for people to
guess what a vmap/get_pages/sgt/... implementation should do. I think
we should define it right now, and IMO, the simpler is to just assume
that sparse objects, when operated as regular objects, provide the same
functionality. That is:

- vmap will populate all pages, pin them, and vmap them
- get_pages will populate all pages
- get_sgt will populate all pages and instantiate an sgt covering the
  whole GEM
- ...

Of course, this means that sparse objects are pointless when operated as
regular objects, but it makes them safe to use, and if we want to
discourage people to call vmap/get_pages/... on a sparse object, we can
always add a drm_warn() in those places.

>  	return drm_gem_shmem_vmap(shmem, map);
>  }
>  
> @@ -263,6 +281,9 @@ static inline int drm_gem_shmem_object_mmap(struct drm_gem_object *obj, struct v
>  {
>  	struct drm_gem_shmem_object *shmem = to_drm_gem_shmem_obj(obj);
>  
> +	if (shmem->sparse)
> +		return -EACCES;
> +
>  	return drm_gem_shmem_mmap(shmem, vma);
>  }
>  

[1]https://gitlab.freedesktop.org/panfrost/linux/-/merge_requests/16/diffs?commit_id=ec08f6c7a728bc6bfc8031ab5fd67ffe92453726