drm/i915/gt: Switch to object allocations for page directories

The GEM object is grossly overweight for the practicality of tracking
large numbers of individual pages, yet it is currently our only
abstraction for tracking DMA allocations. Since those allocations need
to be reserved upfront before an operation, and that we need to break
away from simple system memory, we need to ditch using plain struct page
wrappers.

In the process, we drop the WC mapping as we ended up clflushing
everything anyway due to various issues across a wider range of
platforms. Though in a future step, we need to drop the kmap_atomic
approach which suggests we need to pre-map all the pages and keep them
mapped.

v2: Verify our large scratch page is suitably DMA aligned; and manually
clear the scratch since we are allocating plain struct pages full of
prior content.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200729164219.5737-2-chris@chris-wilson.co.uk
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

1 files changed, 52 insertions, 37 deletions

diff --git a/drivers/gpu/drm/i915/gt/gen8_ppgtt.c b/drivers/gpu/drm/i915/gt/gen8_ppgtt.c
index 08cb65bfb313..a40cb7ea8c66 100644
--- a/drivers/gpu/drm/i915/gt/gen8_ppgtt.c
+++ b/drivers/gpu/drm/i915/gt/gen8_ppgtt.c
@@ -199,7 +199,7 @@ static u64 __gen8_ppgtt_clear(struct i915_address_space * const vm,
 			      struct i915_page_directory * const pd,
 			      u64 start, const u64 end, int lvl)
 {
-	const struct i915_page_scratch * const scratch = &vm->scratch[lvl];
+	const struct drm_i915_gem_object * const scratch = vm->scratch[lvl];
 	unsigned int idx, len;
 
 	GEM_BUG_ON(end > vm->total >> GEN8_PTE_SHIFT);
@@ -239,7 +239,7 @@ static u64 __gen8_ppgtt_clear(struct i915_address_space * const vm,
 
 			vaddr = kmap_atomic_px(pt);
 			memset64(vaddr + gen8_pd_index(start, 0),
-				 vm->scratch[0].encode,
+				 vm->scratch[0]->encode,
 				 count);
 			kunmap_atomic(vaddr);
 
@@ -296,12 +296,13 @@ static void __gen8_ppgtt_alloc(struct i915_address_space * const vm,
 			    __func__, vm, lvl + 1, idx);
 
 			pt = stash->pt[!!lvl];
-			GEM_BUG_ON(!pt);
+			__i915_gem_object_pin_pages(pt->base);
+			i915_gem_object_make_unshrinkable(pt->base);
 
 			if (lvl ||
 			    gen8_pt_count(*start, end) < I915_PDES ||
 			    intel_vgpu_active(vm->i915))
-				fill_px(pt, vm->scratch[lvl].encode);
+				fill_px(pt, vm->scratch[lvl]->encode);
 
 			spin_lock(&pd->lock);
 			if (likely(!pd->entry[idx])) {
@@ -356,16 +357,6 @@ static void gen8_ppgtt_alloc(struct i915_address_space *vm,
 			   &start, start + length, vm->top);
 }
 
-static __always_inline void
-write_pte(gen8_pte_t *pte, const gen8_pte_t val)
-{
-	/* Magic delays? Or can we refine these to flush all in one pass? */
-	*pte = val;
-	wmb(); /* cpu to cache */
-	clflush(pte); /* cache to memory */
-	wmb(); /* visible to all */
-}
-
 static __always_inline u64
 gen8_ppgtt_insert_pte(struct i915_ppgtt *ppgtt,
 		      struct i915_page_directory *pdp,
@@ -382,8 +373,7 @@ gen8_ppgtt_insert_pte(struct i915_ppgtt *ppgtt,
 	vaddr = kmap_atomic_px(i915_pt_entry(pd, gen8_pd_index(idx, 1)));
 	do {
 		GEM_BUG_ON(iter->sg->length < I915_GTT_PAGE_SIZE);
-		write_pte(&vaddr[gen8_pd_index(idx, 0)],
-			  pte_encode | iter->dma);
+		vaddr[gen8_pd_index(idx, 0)] = pte_encode | iter->dma;
 
 		iter->dma += I915_GTT_PAGE_SIZE;
 		if (iter->dma >= iter->max) {
@@ -406,10 +396,12 @@ gen8_ppgtt_insert_pte(struct i915_ppgtt *ppgtt,
 				pd = pdp->entry[gen8_pd_index(idx, 2)];
 			}
 
+			clflush_cache_range(vaddr, PAGE_SIZE);
 			kunmap_atomic(vaddr);
 			vaddr = kmap_atomic_px(i915_pt_entry(pd, gen8_pd_index(idx, 1)));
 		}
 	} while (1);
+	clflush_cache_range(vaddr, PAGE_SIZE);
 	kunmap_atomic(vaddr);
 
 	return idx;
@@ -465,7 +457,7 @@ static void gen8_ppgtt_insert_huge(struct i915_vma *vma,
 
 		do {
 			GEM_BUG_ON(iter->sg->length < page_size);
-			write_pte(&vaddr[index++], encode | iter->dma);
+			vaddr[index++] = encode | iter->dma;
 
 			start += page_size;
 			iter->dma += page_size;
@@ -490,6 +482,7 @@ static void gen8_ppgtt_insert_huge(struct i915_vma *vma,
 			}
 		} while (rem >= page_size && index < I915_PDES);
 
+		clflush_cache_range(vaddr, PAGE_SIZE);
 		kunmap_atomic(vaddr);
 
 		/*
@@ -521,7 +514,7 @@ static void gen8_ppgtt_insert_huge(struct i915_vma *vma,
 			if (I915_SELFTEST_ONLY(vma->vm->scrub_64K)) {
 				u16 i;
 
-				encode = vma->vm->scratch[0].encode;
+				encode = vma->vm->scratch[0]->encode;
 				vaddr = kmap_atomic_px(i915_pt_entry(pd, maybe_64K));
 
 				for (i = 1; i < index; i += 16)
@@ -575,27 +568,37 @@ static int gen8_init_scratch(struct i915_address_space *vm)
 		GEM_BUG_ON(!clone->has_read_only);
 
 		vm->scratch_order = clone->scratch_order;
-		memcpy(vm->scratch, clone->scratch, sizeof(vm->scratch));
-		px_dma(&vm->scratch[0]) = 0; /* no xfer of ownership */
+		for (i = 0; i <= vm->top; i++)
+			vm->scratch[i] = i915_gem_object_get(clone->scratch[i]);
+
 		return 0;
 	}
 
-	ret = setup_scratch_page(vm, __GFP_HIGHMEM);
+	ret = setup_scratch_page(vm);
 	if (ret)
 		return ret;
 
-	vm->scratch[0].encode =
-		gen8_pte_encode(px_dma(&vm->scratch[0]),
+	vm->scratch[0]->encode =
+		gen8_pte_encode(px_dma(vm->scratch[0]),
 				I915_CACHE_LLC, vm->has_read_only);
 
 	for (i = 1; i <= vm->top; i++) {
-		if (unlikely(setup_page_dma(vm, px_base(&vm->scratch[i]))))
+		struct drm_i915_gem_object *obj;
+
+		obj = vm->alloc_pt_dma(vm, I915_GTT_PAGE_SIZE_4K);
+		if (IS_ERR(obj))
 			goto free_scratch;
 
-		fill_px(&vm->scratch[i], vm->scratch[i - 1].encode);
-		vm->scratch[i].encode =
-			gen8_pde_encode(px_dma(&vm->scratch[i]),
-					I915_CACHE_LLC);
+		ret = pin_pt_dma(vm, obj);
+		if (ret) {
+			i915_gem_object_put(obj);
+			goto free_scratch;
+		}
+
+		fill_px(obj, vm->scratch[i - 1]->encode);
+		obj->encode = gen8_pde_encode(px_dma(obj), I915_CACHE_LLC);
+
+		vm->scratch[i] = obj;
 	}
 
 	return 0;
@@ -616,12 +619,20 @@ static int gen8_preallocate_top_level_pdp(struct i915_ppgtt *ppgtt)
 
 	for (idx = 0; idx < GEN8_3LVL_PDPES; idx++) {
 		struct i915_page_directory *pde;
+		int err;
 
 		pde = alloc_pd(vm);
 		if (IS_ERR(pde))
 			return PTR_ERR(pde);
 
-		fill_px(pde, vm->scratch[1].encode);
+		err = pin_pt_dma(vm, pde->pt.base);
+		if (err) {
+			i915_gem_object_put(pde->pt.base);
+			kfree(pde);
+			return err;
+		}
+
+		fill_px(pde, vm->scratch[1]->encode);
 		set_pd_entry(pd, idx, pde);
 		atomic_inc(px_used(pde)); /* keep pinned */
 	}
@@ -635,6 +646,7 @@ gen8_alloc_top_pd(struct i915_address_space *vm)
 {
 	const unsigned int count = gen8_pd_top_count(vm);
 	struct i915_page_directory *pd;
+	int err;
 
 	GEM_BUG_ON(count > ARRAY_SIZE(pd->entry));
 
@@ -642,12 +654,20 @@ gen8_alloc_top_pd(struct i915_address_space *vm)
 	if (unlikely(!pd))
 		return ERR_PTR(-ENOMEM);
 
-	if (unlikely(setup_page_dma(vm, px_base(pd)))) {
+	pd->pt.base = vm->alloc_pt_dma(vm, I915_GTT_PAGE_SIZE_4K);
+	if (IS_ERR(pd->pt.base)) {
 		kfree(pd);
 		return ERR_PTR(-ENOMEM);
 	}
 
-	fill_page_dma(px_base(pd), vm->scratch[vm->top].encode, count);
+	err = pin_pt_dma(vm, pd->pt.base);
+	if (err) {
+		i915_gem_object_put(pd->pt.base);
+		kfree(pd);
+		return ERR_PTR(err);
+	}
+
+	fill_page_dma(px_base(pd), vm->scratch[vm->top]->encode, count);
 	atomic_inc(px_used(pd)); /* mark as pinned */
 	return pd;
 }
@@ -682,12 +702,7 @@ struct i915_ppgtt *gen8_ppgtt_create(struct intel_gt *gt)
 	 */
 	ppgtt->vm.has_read_only = !IS_GEN_RANGE(gt->i915, 11, 12);
 
-	/*
-	 * There are only few exceptions for gen >=6. chv and bxt.
-	 * And we are not sure about the latter so play safe for now.
-	 */
-	if (IS_CHERRYVIEW(gt->i915) || IS_BROXTON(gt->i915))
-		ppgtt->vm.pt_kmap_wc = true;
+	ppgtt->vm.alloc_pt_dma = alloc_pt_dma;
 
 	err = gen8_init_scratch(&ppgtt->vm);
 	if (err)