On Thu, 24 Aug 2023 02:34:48 +0100
Adrián Larumbe <adrian.larumbe@xxxxxxxxxxxxx> wrote:
> BO's RSS is updated every time new pages are allocated and mapped for the
> object, either in its entirety at creation time for non-heap buffers, or
> else on demand for heap buffers at GPU page fault's IRQ handler.
>
> Same calculations had to be done for imported PRIME objects, since backing
> storage for it might have already been allocated by the exporting driver.
>
> Signed-off-by: Adrián Larumbe <adrian.larumbe@xxxxxxxxxxxxx>
> ---
> drivers/gpu/drm/panfrost/panfrost_gem.c | 22 ++++++++++++++++++++++
> drivers/gpu/drm/panfrost/panfrost_gem.h | 5 +++++
> drivers/gpu/drm/panfrost/panfrost_mmu.c | 16 +++++++++++-----
> 3 files changed, 38 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpu/drm/panfrost/panfrost_gem.c b/drivers/gpu/drm/panfrost/panfrost_gem.c
> index aea16b0e4dda..c6bd1f16a6d4 100644
> --- a/drivers/gpu/drm/panfrost/panfrost_gem.c
> +++ b/drivers/gpu/drm/panfrost/panfrost_gem.c
> @@ -206,6 +206,17 @@ static enum drm_gem_object_status panfrost_gem_status(struct drm_gem_object *obj
>
> return res;
> }
> +
> +size_t panfrost_gem_rss(struct drm_gem_object *obj)
> +{
> + struct panfrost_gem_object *bo = to_panfrost_bo(obj);
> +
> + if (!bo->base.pages)
> + return 0;
> +
> + return bo->rss_size;
> +}
> +
> static const struct drm_gem_object_funcs panfrost_gem_funcs = {
> .free = panfrost_gem_free_object,
> .open = panfrost_gem_open,
> @@ -218,6 +229,7 @@ static const struct drm_gem_object_funcs panfrost_gem_funcs = {
> .vunmap = drm_gem_shmem_object_vunmap,
> .mmap = drm_gem_shmem_object_mmap,
> .status = panfrost_gem_status,
> + .rss = panfrost_gem_rss,
> .vm_ops = &drm_gem_shmem_vm_ops,
> };
>
> @@ -274,13 +286,23 @@ panfrost_gem_prime_import_sg_table(struct drm_device *dev,
> {
> struct drm_gem_object *obj;
> struct panfrost_gem_object *bo;
> + struct scatterlist *sgl;
> + unsigned int count;
> + size_t total = 0;
>
> obj = drm_gem_shmem_prime_import_sg_table(dev, attach, sgt);
> if (IS_ERR(obj))
> return ERR_CAST(obj);
>
> + for_each_sgtable_dma_sg(sgt, sgl, count) {
> + size_t len = sg_dma_len(sgl);
> +
> + total += len;
> + }
Why not simply have bo->rss_size = obj->size here? Not sure I see a
reason to not trust dma_buf?
> +
> bo = to_panfrost_bo(obj);
> bo->noexec = true;
> + bo->rss_size = total;
>
> return obj;
> }
> diff --git a/drivers/gpu/drm/panfrost/panfrost_gem.h b/drivers/gpu/drm/panfrost/panfrost_gem.h
> index e06f7ceb8f73..e2a7c46403c7 100644
> --- a/drivers/gpu/drm/panfrost/panfrost_gem.h
> +++ b/drivers/gpu/drm/panfrost/panfrost_gem.h
> @@ -36,6 +36,11 @@ struct panfrost_gem_object {
> */
> atomic_t gpu_usecount;
>
> + /*
> + * Object chunk size currently mapped onto physical memory
> + */
> + size_t rss_size;
> +
> bool noexec :1;
> bool is_heap :1;
> bool is_purgable :1;
> diff --git a/drivers/gpu/drm/panfrost/panfrost_mmu.c b/drivers/gpu/drm/panfrost/panfrost_mmu.c
> index c0123d09f699..e03a5a9da06f 100644
> --- a/drivers/gpu/drm/panfrost/panfrost_mmu.c
> +++ b/drivers/gpu/drm/panfrost/panfrost_mmu.c
> @@ -285,17 +285,19 @@ static void panfrost_mmu_flush_range(struct panfrost_device *pfdev,
> pm_runtime_put_autosuspend(pfdev->dev);
> }
>
> -static int mmu_map_sg(struct panfrost_device *pfdev, struct panfrost_mmu *mmu,
> +static size_t mmu_map_sg(struct panfrost_device *pfdev, struct panfrost_mmu *mmu,
> u64 iova, int prot, struct sg_table *sgt)
> {
> unsigned int count;
> struct scatterlist *sgl;
> struct io_pgtable_ops *ops = mmu->pgtbl_ops;
> u64 start_iova = iova;
> + size_t total = 0;
>
> for_each_sgtable_dma_sg(sgt, sgl, count) {
> unsigned long paddr = sg_dma_address(sgl);
> size_t len = sg_dma_len(sgl);
> + total += len;
>
> dev_dbg(pfdev->dev, "map: as=%d, iova=%llx, paddr=%lx, len=%zx", mmu->as, iova, paddr, len);
>
> @@ -315,7 +317,7 @@ static int mmu_map_sg(struct panfrost_device *pfdev, struct panfrost_mmu *mmu,
>
> panfrost_mmu_flush_range(pfdev, mmu, start_iova, iova - start_iova);
>
> - return 0;
> + return total;
> }
>
> int panfrost_mmu_map(struct panfrost_gem_mapping *mapping)
> @@ -326,6 +328,7 @@ int panfrost_mmu_map(struct panfrost_gem_mapping *mapping)
> struct panfrost_device *pfdev = to_panfrost_device(obj->dev);
> struct sg_table *sgt;
> int prot = IOMMU_READ | IOMMU_WRITE;
> + size_t mapped_size;
>
> if (WARN_ON(mapping->active))
> return 0;
> @@ -337,9 +340,10 @@ int panfrost_mmu_map(struct panfrost_gem_mapping *mapping)
> if (WARN_ON(IS_ERR(sgt)))
> return PTR_ERR(sgt);
>
> - mmu_map_sg(pfdev, mapping->mmu, mapping->mmnode.start << PAGE_SHIFT,
> + mapped_size = mmu_map_sg(pfdev, mapping->mmu, mapping->mmnode.start << PAGE_SHIFT,
> prot, sgt);
> mapping->active = true;
> + bo->rss_size += mapped_size;
Actually, the GEM might be resident even before panfrost_mmu_map() is
called: as soon as drm_gem_shmem_get_pages[_locked]() is called, it's
resident (might get evicted after that point though). That means any
mmap coming from userspace will make the buffer resident too. I know
we're automatically mapping GEMs to the GPU VM in panfrost_gem_open(),
so it makes no difference, but I think I'd prefer if we keep ->rss_size
for heap BOs only (we probably want to rename it heap_rss_size) and
then have
if (bo->is_heap)
return bo->heap_rss_size;
else if (bo->base.pages)
return bo->base.base.size;
else
return 0;
in panfrost_gem_rss().
>
> return 0;
> }
> @@ -447,6 +451,7 @@ static int panfrost_mmu_map_fault_addr(struct panfrost_device *pfdev, int as,
> pgoff_t page_offset;
> struct sg_table *sgt;
> struct page **pages;
> + size_t mapped_size;
>
> bomapping = addr_to_mapping(pfdev, as, addr);
> if (!bomapping)
> @@ -518,10 +523,11 @@ static int panfrost_mmu_map_fault_addr(struct panfrost_device *pfdev, int as,
> if (ret)
> goto err_map;
>
> - mmu_map_sg(pfdev, bomapping->mmu, addr,
> - IOMMU_WRITE | IOMMU_READ | IOMMU_NOEXEC, sgt);
> + mapped_size = mmu_map_sg(pfdev, bomapping->mmu, addr,
> + IOMMU_WRITE | IOMMU_READ | IOMMU_NOEXEC, sgt);
>
> bomapping->active = true;
> + bo->rss_size += mapped_size;
>
> dev_dbg(pfdev->dev, "mapped page fault @ AS%d %llx", as, addr);
>
