Commit c5b3cc41 authored by Sunil Khatri's avatar Sunil Khatri Committed by Alex Deucher
Browse files

drm/amdgpu: use hmm_pfns instead of array of pages 



we dont need to allocate local array of pages to hold
the pages returned by the hmm, instead we could use
the hmm_range structure itself to get to hmm_pfn
and get the required pages directly.

This avoids call to alloc/free quite a lot.

Signed-off-by: default avatarSunil Khatri <sunil.khatri@amd.com>
Suggested-by: default avatarChristian König <christian.koenig@amd.com>
Reviewed-by: default avatarChristian König <christian.koenig@amd.com>
Acked-by: default avatarFelix Kuehling <felix.kuehling@amd.com>
Signed-off-by: default avatarAlex Deucher <alexander.deucher@amd.com>
parent ae4d627e

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c

+7 −3
Original line number Diff line number Diff line
@@ -1089,7 +1089,7 @@ static int init_user_pages(struct kgd_mem *mem, uint64_t user_addr, 
		return 0;

	}



	ret = amdgpu_ttm_tt_get_user_pages(bo, bo->tbo.ttm->pages, &range);

	ret = amdgpu_ttm_tt_get_user_pages(bo, &range);

	if (ret) {

		if (ret == -EAGAIN)

			pr_debug("Failed to get user pages, try again\n");
@@ -1103,6 +1103,9 @@ static int init_user_pages(struct kgd_mem *mem, uint64_t user_addr, 
		pr_err("%s: Failed to reserve BO\n", __func__);

		goto release_out;

	}



	amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm, range);



	amdgpu_bo_placement_from_domain(bo, mem->domain);

	ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);

	if (ret)
@@ -2565,8 +2568,7 @@ static int update_invalid_user_pages(struct amdkfd_process_info *process_info, 
		}



		/* Get updated user pages */

		ret = amdgpu_ttm_tt_get_user_pages(bo, bo->tbo.ttm->pages,

						   &mem->range);

		ret = amdgpu_ttm_tt_get_user_pages(bo, &mem->range);

		if (ret) {

			pr_debug("Failed %d to get user pages\n", ret);
@@ -2595,6 +2597,8 @@ static int update_invalid_user_pages(struct amdkfd_process_info *process_info, 
			ret = 0;

		}



		amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm, mem->range);



		mutex_lock(&process_info->notifier_lock);



		/* Mark the BO as valid unless it was invalidated

drivers/gpu/drm/amd/amdgpu/amdgpu_bo_list.h

+0 −1
Original line number Diff line number Diff line
@@ -38,7 +38,6 @@ struct amdgpu_bo_list_entry { 
	struct amdgpu_bo		*bo;

	struct amdgpu_bo_va		*bo_va;

	uint32_t			priority;

	struct page			**user_pages;

	struct hmm_range		*range;

	bool				user_invalidated;

};

drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c

+6 −24
Original line number Diff line number Diff line
@@ -29,6 +29,7 @@ 
#include <linux/pagemap.h>

#include <linux/sync_file.h>

#include <linux/dma-buf.h>

#include <linux/hmm.h>



#include <drm/amdgpu_drm.h>

#include <drm/drm_syncobj.h>
@@ -885,24 +886,12 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p, 
		struct amdgpu_bo *bo = e->bo;

		int i;



		e->user_pages = kvcalloc(bo->tbo.ttm->num_pages,

					 sizeof(struct page *),

					 GFP_KERNEL);

		if (!e->user_pages) {

			drm_err(adev_to_drm(p->adev), "kvmalloc_array failure\n");

			r = -ENOMEM;

			goto out_free_user_pages;

		}



		r = amdgpu_ttm_tt_get_user_pages(bo, e->user_pages, &e->range);

		if (r) {

			kvfree(e->user_pages);

			e->user_pages = NULL;

		r = amdgpu_ttm_tt_get_user_pages(bo, &e->range);

		if (r)

			goto out_free_user_pages;

		}



		for (i = 0; i < bo->tbo.ttm->num_pages; i++) {

			if (bo->tbo.ttm->pages[i] != e->user_pages[i]) {

			if (bo->tbo.ttm->pages[i] != hmm_pfn_to_page(e->range->hmm_pfns[i])) {

				userpage_invalidated = true;

				break;

			}
@@ -946,7 +935,7 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p, 
		}



		if (amdgpu_ttm_tt_is_userptr(e->bo->tbo.ttm) &&

		    e->user_invalidated && e->user_pages) {

		    e->user_invalidated) {

			amdgpu_bo_placement_from_domain(e->bo,

							AMDGPU_GEM_DOMAIN_CPU);

			r = ttm_bo_validate(&e->bo->tbo, &e->bo->placement,
@@ -955,11 +944,8 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p, 
				goto out_free_user_pages;



			amdgpu_ttm_tt_set_user_pages(e->bo->tbo.ttm,

						     e->user_pages);

						     e->range);

		}



		kvfree(e->user_pages);

		e->user_pages = NULL;

	}



	amdgpu_cs_get_threshold_for_moves(p->adev, &p->bytes_moved_threshold,
@@ -1001,11 +987,7 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p, 
	amdgpu_bo_list_for_each_userptr_entry(e, p->bo_list) {

		struct amdgpu_bo *bo = e->bo;



		if (!e->user_pages)

			continue;

		amdgpu_ttm_tt_get_user_pages_done(bo->tbo.ttm, e->range);

		kvfree(e->user_pages);

		e->user_pages = NULL;

		e->range = NULL;

	}

	mutex_unlock(&p->bo_list->bo_list_mutex);

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c

+3 −2
Original line number Diff line number Diff line
@@ -572,8 +572,7 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data, 
		goto release_object;



	if (args->flags & AMDGPU_GEM_USERPTR_VALIDATE) {

		r = amdgpu_ttm_tt_get_user_pages(bo, bo->tbo.ttm->pages,

						 &range);

		r = amdgpu_ttm_tt_get_user_pages(bo, &range);

		if (r)

			goto release_object;
@@ -581,6 +580,8 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data, 
		if (r)

			goto user_pages_done;



		amdgpu_ttm_tt_set_user_pages(bo->tbo.ttm, range);



		amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);

		r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);

		amdgpu_bo_unreserve(bo);

drivers/gpu/drm/amd/amdgpu/amdgpu_hmm.c

+1 −10
Original line number Diff line number Diff line
@@ -167,13 +167,12 @@ void amdgpu_hmm_unregister(struct amdgpu_bo *bo) 


int amdgpu_hmm_range_get_pages(struct mmu_interval_notifier *notifier,

			       uint64_t start, uint64_t npages, bool readonly,

			       void *owner, struct page **pages,

			       void *owner,

			       struct hmm_range **phmm_range)

{

	struct hmm_range *hmm_range;

	unsigned long end;

	unsigned long timeout;

	unsigned long i;

	unsigned long *pfns;

	int r = 0;
@@ -222,14 +221,6 @@ int amdgpu_hmm_range_get_pages(struct mmu_interval_notifier *notifier, 
	hmm_range->start = start;

	hmm_range->hmm_pfns = pfns;



	/*

	 * Due to default_flags, all pages are HMM_PFN_VALID or

	 * hmm_range_fault() fails. FIXME: The pages cannot be touched outside

	 * the notifier_lock, and mmu_interval_read_retry() must be done first.

	 */

	for (i = 0; pages && i < npages; i++)

		pages[i] = hmm_pfn_to_page(pfns[i]);



	*phmm_range = hmm_range;



	return 0;