On Sun, 27 Aug 2023 20:54:38 +0300
Dmitry Osipenko <dmitry.osipenko@xxxxxxxxxxxxx> wrote:
> Add separate pages_pin_count for tracking of whether drm-shmem pages are
> moveable or not. With the addition of memory shrinker support to drm-shmem,
> the pages_use_count will no longer determine whether pages are hard-pinned
> in memory, but whether pages exit and are soft-pinned (and could be swapped
> out). The pages_pin_count > 1 will hard-pin pages in memory.
>
> Suggested-by: Boris Brezillon <boris.brezillon@xxxxxxxxxxxxx>
> Signed-off-by: Dmitry Osipenko <dmitry.osipenko@xxxxxxxxxxxxx>
> ---
> drivers/gpu/drm/drm_gem_shmem_helper.c | 22 +++++++++++++++++-----
> include/drm/drm_gem_shmem_helper.h | 10 ++++++++++
> 2 files changed, 27 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_gem_shmem_helper.c b/drivers/gpu/drm/drm_gem_shmem_helper.c
> index d545d3d227d7..1a7e5c332fd8 100644
> --- a/drivers/gpu/drm/drm_gem_shmem_helper.c
> +++ b/drivers/gpu/drm/drm_gem_shmem_helper.c
> @@ -234,14 +234,22 @@ static int drm_gem_shmem_pin_locked(struct drm_gem_shmem_object *shmem)
>
> dma_resv_assert_held(shmem->base.resv);
>
> + if (kref_get_unless_zero(&shmem->pages_pin_count))
> + return 0;
> +
> ret = drm_gem_shmem_get_pages_locked(shmem);
> + if (!ret)
> + kref_init(&shmem->pages_pin_count);
>
> return ret;
> }
>
> -static void drm_gem_shmem_unpin_locked(struct drm_gem_shmem_object *shmem)
> +static void drm_gem_shmem_kref_unpin_pages(struct kref *kref)
> {
> - dma_resv_assert_held(shmem->base.resv);
> + struct drm_gem_shmem_object *shmem;
> +
> + shmem = container_of(kref, struct drm_gem_shmem_object,
> + pages_pin_count);
>
> drm_gem_shmem_put_pages_locked(shmem);
> }
> @@ -263,6 +271,9 @@ int drm_gem_shmem_pin(struct drm_gem_shmem_object *shmem)
>
> drm_WARN_ON(obj->dev, obj->import_attach);
>
> + if (kref_get_unless_zero(&shmem->pages_pin_count))
> + return 0;
> +
> ret = dma_resv_lock_interruptible(shmem->base.resv, NULL);
> if (ret)
> return ret;
> @@ -286,9 +297,10 @@ void drm_gem_shmem_unpin(struct drm_gem_shmem_object *shmem)
>
> drm_WARN_ON(obj->dev, obj->import_attach);
>
> - dma_resv_lock(shmem->base.resv, NULL);
> - drm_gem_shmem_unpin_locked(shmem);
> - dma_resv_unlock(shmem->base.resv);
> + if (kref_put_dma_resv(&shmem->pages_pin_count,
> + drm_gem_shmem_kref_unpin_pages,
> + obj->resv, NULL))
> + dma_resv_unlock(obj->resv);
> }
> EXPORT_SYMBOL_GPL(drm_gem_shmem_unpin);
>
> diff --git a/include/drm/drm_gem_shmem_helper.h b/include/drm/drm_gem_shmem_helper.h
> index ec2d8b24e3cf..afb7cd671e2a 100644
> --- a/include/drm/drm_gem_shmem_helper.h
> +++ b/include/drm/drm_gem_shmem_helper.h
> @@ -39,6 +39,16 @@ struct drm_gem_shmem_object {
> */
> unsigned int pages_use_count;
>
> + /**
> + * @pages_pin_count:
> + *
> + * Reference count on the pinned pages table.
> + * The pages allowed to be evicted and purged by memory
> + * shrinker only when the count is zero, otherwise pages
> + * are hard-pinned in memory.
> + */
> + struct kref pages_pin_count;
I know it's tempting to use kref for the pages use/pin count, but I'm
wondering if we wouldn't be better using a refcount_t, which provides
overflow/underflow protection while still letting us control how we
want to handle the locking for 0 <-> 1 transitions. By doing that, we
avoid introducing core locking changes that might be more
controversial/longer to get accepted. Besides, I suspect the resulting
code (the one using a refcount_t) won't be more verbose/complicated (no
release functions needed if you don't use kref_put(), which makes
things closer to what we have right now).
> +
> /**
> * @madv: State for madvise
> *
