Skip to contentSkip to conversation

GEN8 PPGTT cleanups + 4GB support

PATCH New Ben Widawsky 1 participant 12 messages Dec 24, 2013 – Dec 26, 2013

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover GEN8 PPGTT cleanups + 4GB support

The following patches rework a lot of the PPGTT cleanup and initialization code. In the short term, this should allow us to use all 4GB address space as well as get the code ready for adding full PPGTT support to BDW.

Aside from making the code clearer, the more discrete functions make finer grained page table allocation much easier to manage. Previously, we were allocating one gigantic contiguous chunk for the GEN8 page tables. This required a 4MB contiguous allocation. On modern Linux, such a large allocation is impossible, and so we fell back to a 2GB address space requiring only 2MB for page tables.

The GEN8 PPGTT code was in poor shape to begin with because the effort was on getting functionality quickly to new platforms, and this was achieved by largely copy and pasting the GEN6 code and altering where needed. I initially tried to do the finer grained allocation as a single patch, and was really unsatisfied with the results.

One nice outcome (I feel) is everything gets referenced as a PDPE/PDE/PTE, or index thereof. It becomes easy to convert these things to the graphics virtual address, or vice-versa. The win will be when we start to map address ranges more dynamically, specifically, allocate the page tables, PDEs, and map the PDEs.

The last patch is totally optional. I've or its equivalent quite a few times in various series, but it never seems to stick.

After we get this merged, I'll submit BDW full PPGTT. I also have a blog post about this stuff in the works.

Ben Widawsky (7):
drm/i915/bdw: Split up PPGTT cleanup drm/i915/bdw: Reorganize PPGTT init drm/i915/bdw: Split ppgtt initialization up drm/i915: Make clear/insert vfuncs args absolute drm/i915/bdw: Reorganize PT allocations Revert "drm/i915/bdw: Limit GTT to 2GB" drm/i915: Update i915_gem_gtt.c copyright

 drivers/gpu/drm/i915/i915_drv.h     |  13 +-
 drivers/gpu/drm/i915/i915_gem_gtt.c | 463 +++++++++++++++++++++++++-----------
 2 files changed, 332 insertions(+), 144 deletions(-)
Signature · 1 line
1.8.5.2
#1 lore raw

PATCH 1/7 drm/i915/bdw: Split up PPGTT cleanup

This will make the code more readable, and extensible which is needed for upcoming feature work. Eventually, we'll do the same for init.

---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 58 +++++++++++++++++++++++--------------
 1 file changed, 37 insertions(+), 21 deletions(-)
drivers/gpu/drm/i915/i915_gem_gtt.c +37 -21
index 998f9a0..d0ee9a9 100644
@@ -318,36 +318,52 @@ static void gen8_ppgtt_insert_entries(struct i915_address_space *vm,
 	kunmap_atomic(pt_vaddr);
 }
 
-static void gen8_ppgtt_cleanup(struct i915_address_space *vm)
+static void gen8_ppgtt_free(struct i915_hw_ppgtt *ppgtt)
+{
+	int i;
+
+	for (i = 0; i < ppgtt->num_pd_pages ; i++)
+		kfree(ppgtt->gen8_pt_dma_addr[i]);
+
+	__free_pages(ppgtt->gen8_pt_pages, get_order(ppgtt->num_pt_pages << PAGE_SHIFT));
+	__free_pages(ppgtt->pd_pages, get_order(ppgtt->num_pd_pages << PAGE_SHIFT));
+}
+
+static void gen8_ppgtt_unmap_pages(struct i915_hw_ppgtt *ppgtt)
 {
-	struct i915_hw_ppgtt *ppgtt =
-		container_of(vm, struct i915_hw_ppgtt, base);
 	int i, j;
 
-	list_del(&vm->global_link);
-	drm_mm_takedown(&vm->mm);
+	for (i = 0; i < ppgtt->num_pd_pages; i++) {
+		/* TODO: future may allow sparse mappings update this is not */
+		if (!ppgtt->pd_dma_addr[i])
+			continue;
 
-	for (i = 0; i < ppgtt->num_pd_pages ; i++) {
-		if (ppgtt->pd_dma_addr[i]) {
-			pci_unmap_page(ppgtt->base.dev->pdev,
-				       ppgtt->pd_dma_addr[i],
-				       PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
+		pci_unmap_page(ppgtt->base.dev->pdev,
+			       ppgtt->pd_dma_addr[i],
+			       PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
 
-			for (j = 0; j < GEN8_PDES_PER_PAGE; j++) {
-				dma_addr_t addr = ppgtt->gen8_pt_dma_addr[i][j];
-				if (addr)
-					pci_unmap_page(ppgtt->base.dev->pdev,
-						       addr,
-						       PAGE_SIZE,
-						       PCI_DMA_BIDIRECTIONAL);
+		for (j = 0; j < GEN8_PDES_PER_PAGE; j++) {
+			dma_addr_t addr = ppgtt->gen8_pt_dma_addr[i][j];
+			if (addr)
+				pci_unmap_page(ppgtt->base.dev->pdev,
+				       addr,
+				       PAGE_SIZE,
+				       PCI_DMA_BIDIRECTIONAL);
 
-			}
 		}
-		kfree(ppgtt->gen8_pt_dma_addr[i]);
 	}
+}
 
-	__free_pages(ppgtt->gen8_pt_pages, get_order(ppgtt->num_pt_pages << PAGE_SHIFT));
-	__free_pages(ppgtt->pd_pages, get_order(ppgtt->num_pd_pages << PAGE_SHIFT));
+static void gen8_ppgtt_cleanup(struct i915_address_space *vm)
+{
+	struct i915_hw_ppgtt *ppgtt =
+		container_of(vm, struct i915_hw_ppgtt, base);
+
+	list_del(&vm->global_link);
+	drm_mm_takedown(&vm->mm);
+
+	gen8_ppgtt_unmap_pages(ppgtt);
+	gen8_ppgtt_free(ppgtt);
 }
 
 /**
Signature · 1 line
1.8.5.2
#2 lore raw

PATCH 2/7 drm/i915/bdw: Reorganize PPGTT init

In reply to Ben Widawsky

Create 3 clear stages in PPGTT init. This will help with upcoming changes be more readable. The 3 stages are, allocation, dma mapping, and writing the P[DT]Es

One nice benefit to the patches is that it makes 2 very clear error points, allocation, and mapping, and avoids having to do any handling after writing PTEs (something which was likely buggy before). This simplified error handling I suspect will be helpful when we move to deferred/dynamic page table allocation and mapping.

The patches also attempts to break up some of the steps into more logical reviewable chunks, particularly when we free.

---
 drivers/gpu/drm/i915/i915_drv.h     |   2 +-
 drivers/gpu/drm/i915/i915_gem_gtt.c | 125 +++++++++++++++++++++---------------
 2 files changed, 74 insertions(+), 53 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +1 -1
index cc8afff..dc21963 100644
@@ -671,7 +671,7 @@ struct i915_hw_ppgtt {
 	};
 	union {
 		dma_addr_t *pt_dma_addr;
-		dma_addr_t *gen8_pt_dma_addr[4];
+		dma_addr_t **gen8_pt_dma_addr;
 	};
 
 	int (*enable)(struct i915_hw_ppgtt *ppgtt);
drivers/gpu/drm/i915/i915_gem_gtt.c +73 -52

196 lines not on this page · raw message

Signature · 1 line
1.8.5.2
#3 lore raw

PATCH v2 2/7 drm/i915/bdw: Reorganize PPGTT init

Create 3 clear stages in PPGTT init. This will help with upcoming changes be more readable. The 3 stages are, allocation, dma mapping, and writing the P[DT]Es

One nice benefit to the patches is that it makes 2 very clear error points, allocation, and mapping, and avoids having to do any handling after writing PTEs (something which was likely buggy before). This simplified error handling I suspect will be helpful when we move to deferred/dynamic page table allocation and mapping.

The patches also attempts to break up some of the steps into more logical reviewable chunks, particularly when we free.

v2: Don't call cleanup on the error path since that takes down the drm_mm and list entry, which aren't setup at this point.

---
 drivers/gpu/drm/i915/i915_drv.h     |   2 +-
 drivers/gpu/drm/i915/i915_gem_gtt.c | 124 +++++++++++++++++++++---------------
 2 files changed, 73 insertions(+), 53 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +1 -1
index cc8afff..dc21963 100644
@@ -671,7 +671,7 @@ struct i915_hw_ppgtt {
 	};
 	union {
 		dma_addr_t *pt_dma_addr;
-		dma_addr_t *gen8_pt_dma_addr[4];
+		dma_addr_t **gen8_pt_dma_addr;
 	};
 
 	int (*enable)(struct i915_hw_ppgtt *ppgtt);
drivers/gpu/drm/i915/i915_gem_gtt.c +72 -52

195 lines not on this page · raw message

Signature · 1 line
1.8.5.2
#4 lore raw

PATCH 3/7 drm/i915/bdw: Split ppgtt initialization up

In reply to Ben Widawsky

Like cleanup in an earlier patch, the code becomes much more readable, and easier to extend if we extract out helper functions for the various stages of init.

Note that with this patch it becomes really simple, and tempting to begin using the 'goto out' idiom with explicit free/fini semantics. I've avoiding that to keep the cleanup function (which is required anyway) as robust and widespread as possible.

---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 183 +++++++++++++++++++++++++-----------
 1 file changed, 126 insertions(+), 57 deletions(-)
drivers/gpu/drm/i915/i915_gem_gtt.c +126 -57
index b0b621d..25687cf 100644
@@ -364,93 +364,162 @@ static void gen8_ppgtt_cleanup(struct i915_address_space *vm)
 	gen8_ppgtt_free(ppgtt);
 }
 
-/**
- * GEN8 legacy ppgtt programming is accomplished through a max 4 PDP registers
- * with a net effect resembling a 2-level page table in normal x86 terms. Each
- * PDP represents 1GB of memory 4 * 512 * 512 * 4096 = 4GB legacy 32b address
- * space.
- *
- * FIXME: split allocation into smaller pieces. For now we only ever do this
- * once, but with full PPGTT, the multiple contiguous allocations will be bad.
- * TODO: Do something with the size parameter
- */
-static int gen8_ppgtt_init(struct i915_hw_ppgtt *ppgtt, uint64_t size)
+static int gen8_ppgtt_allocate_page_tables(struct i915_hw_ppgtt *ppgtt,
+					   const int max_pdp)
 {
 	struct page *pt_pages;
-	const int max_pdp = DIV_ROUND_UP(size, 1 << 30);
 	const int num_pt_pages = GEN8_PDES_PER_PAGE * max_pdp;
-	int i, j, ret;
-
-	if (size % (1<<30))
-		DRM_INFO("Pages will be wasted unless GTT size (%llu) is divisible by 1GB\n", size);
-
-	/* 1. Do all our allocations for page directories and page tables */
-	ppgtt->pd_pages = alloc_pages(GFP_KERNEL, get_order(max_pdp << PAGE_SHIFT));
-	if (!ppgtt->pd_pages)
-		return -ENOMEM;
 
 	pt_pages = alloc_pages(GFP_KERNEL, get_order(num_pt_pages << PAGE_SHIFT));
-	if (!pt_pages) {
-		__free_pages(ppgtt->pd_pages, get_order(max_pdp << PAGE_SHIFT));
+	if (!pt_pages)
 		return -ENOMEM;
-	}
 
 	ppgtt->gen8_pt_pages = pt_pages;
-	ppgtt->num_pd_pages = 1 << get_order(max_pdp << PAGE_SHIFT);
 	ppgtt->num_pt_pages = 1 << get_order(num_pt_pages << PAGE_SHIFT);
-	ppgtt->num_pd_entries = max_pdp * GEN8_PDES_PER_PAGE;
-	BUG_ON(ppgtt->num_pd_pages > GEN8_LEGACY_PDPS);
 
-	/* NB:From here on, ppgtt->base.cleanup() should function properly. */
+	return 0;
+}
+
+static int gen8_ppgtt_allocate_dma(struct i915_hw_ppgtt *ppgtt)
+{
+	int i;
 
-	ppgtt->gen8_pt_dma_addr = kcalloc(max_pdp,
+	ppgtt->gen8_pt_dma_addr = kcalloc(ppgtt->num_pd_entries,
 					  sizeof(*ppgtt->gen8_pt_dma_addr),
 					  GFP_KERNEL);
-	if (!ppgtt->gen8_pt_dma_addr) {
-		ret = -ENOMEM;
-		goto bail;
-	}
+	if (!ppgtt->gen8_pt_dma_addr)
+		return -ENOMEM;
 
-	for (i = 0; i < max_pdp; i++) {
+	for (i = 0; i < ppgtt->num_pd_entries; i++) {
 		ppgtt->gen8_pt_dma_addr[i] = kcalloc(GEN8_PDES_PER_PAGE,
 						     sizeof(dma_addr_t),
 						     GFP_KERNEL);
 		if (!ppgtt->gen8_pt_dma_addr[i]) {
-			ret = -ENOMEM;
-			goto bail;
+			kfree(ppgtt->gen8_pt_dma_addr);
+			while(i--)
+				kfree(ppgtt->gen8_pt_dma_addr[i]);
+
+			return -ENOMEM;
 		}
 	}
 
+	return 0;
+}
+
+static int gen8_ppgtt_allocate_page_directories(struct i915_hw_ppgtt *ppgtt,
+						const int max_pdp)
+{
+	ppgtt->pd_pages = alloc_pages(GFP_KERNEL, get_order(max_pdp << PAGE_SHIFT));
+	if (!ppgtt->pd_pages)
+		return -ENOMEM;
+
+	ppgtt->num_pd_pages = 1 << get_order(max_pdp << PAGE_SHIFT);
+
+	return 0;
+}
+
+static int gen8_ppgtt_alloc(struct i915_hw_ppgtt *ppgtt,
+			    const int max_pdp)
+{
+	int ret;
+
+	ret = gen8_ppgtt_allocate_page_directories(ppgtt, max_pdp);
+	if (ret)
+		return ret;
+
+	ret = gen8_ppgtt_allocate_page_tables(ppgtt, max_pdp);
+	if (ret) {
+		__free_pages(ppgtt->pd_pages, get_order(max_pdp << PAGE_SHIFT));
+		return ret;
+	}
+
+	ppgtt->num_pd_entries = max_pdp * GEN8_PDES_PER_PAGE;
+	BUG_ON(ppgtt->num_pd_pages > GEN8_LEGACY_PDPS);
+
+	ret = gen8_ppgtt_allocate_dma(ppgtt);
+	if (ret)
+		gen8_ppgtt_free(ppgtt);
+
+	return ret;
+}
+
+static int gen8_ppgtt_setup_page_directories(struct i915_hw_ppgtt *ppgtt,
+					     const int pd)
+{
+	dma_addr_t pd_addr;
+	int ret;
+
+	pd_addr = pci_map_page(ppgtt->base.dev->pdev,
+			       &ppgtt->pd_pages[pd], 0,
+			       PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
+
+	ret = pci_dma_mapping_error(ppgtt->base.dev->pdev, pd_addr);
+	if (ret)
+		return ret;
+
+	ppgtt->pd_dma_addr[pd] = pd_addr;
+
+	return 0;
+}
+
+static int gen8_ppgtt_setup_page_tables(struct i915_hw_ppgtt *ppgtt,
+					const int pd,
+					const int pt)
+{
+	dma_addr_t pt_addr;
+	struct page *p;
+	int ret;
+
+	p = &ppgtt->gen8_pt_pages[pd * GEN8_PDES_PER_PAGE + pt];
+	pt_addr = pci_map_page(ppgtt->base.dev->pdev,
+			       p, 0, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
+	ret = pci_dma_mapping_error(ppgtt->base.dev->pdev, pt_addr);
+	if (ret)
+		return ret;
+
+	ppgtt->gen8_pt_dma_addr[pd][pt] = pt_addr;
+
+	return 0;
+}
+
+/**
+ * GEN8 legacy ppgtt programming is accomplished through a max 4 PDP registers
+ * with a net effect resembling a 2-level page table in normal x86 terms. Each
+ * PDP represents 1GB of memory 4 * 512 * 512 * 4096 = 4GB legacy 32b address
+ * space.
+ *
+ * FIXME: split allocation into smaller pieces. For now we only ever do this
+ * once, but with full PPGTT, the multiple contiguous allocations will be bad.
+ * TODO: Do something with the size parameter
+ */
+static int gen8_ppgtt_init(struct i915_hw_ppgtt *ppgtt, uint64_t size)
+{
+	const int max_pdp = DIV_ROUND_UP(size, 1 << 30);
+	const int min_pt_pages = GEN8_PDES_PER_PAGE * max_pdp;
+	int i, j, ret;
+
+	if (size % (1<<30))
+		DRM_INFO("Pages will be wasted unless GTT size (%llu) is divisible by 1GB\n", size);
+
+	/* 1. Do all our allocations for page directories and page tables. */
+	ret = gen8_ppgtt_alloc(ppgtt, max_pdp);
+	if (ret)
+		return ret;
+
 	/*
-	 * 2. Create all the DMA mappings for the page directories and page
-	 * tables
+	 * 2. Create DMA mappings for the page directories and page tables.
+	 * NB:From here on, ppgtt->base.cleanup() should function properly.
 	 */
 	for (i = 0; i < max_pdp; i++) {
-		dma_addr_t pd_addr, pt_addr;
-
-		/* Get the page table mappings per page directory */
 		for (j = 0; j < GEN8_PDES_PER_PAGE; j++) {
-			struct page *p = &pt_pages[i * GEN8_PDES_PER_PAGE + j];
-
-			pt_addr = pci_map_page(ppgtt->base.dev->pdev,
-					       p, 0, PAGE_SIZE,
-					       PCI_DMA_BIDIRECTIONAL);
-			ret = pci_dma_mapping_error(ppgtt->base.dev->pdev, pt_addr);
+			ret = gen8_ppgtt_setup_page_tables(ppgtt, i, j);
 			if (ret)
 				goto bail;
-
-			ppgtt->gen8_pt_dma_addr[i][j] = pt_addr;
 		}
 
-		/* And the page directory mappings */
-		pd_addr = pci_map_page(ppgtt->base.dev->pdev,
-				       &ppgtt->pd_pages[i], 0,
-				       PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
-		ret = pci_dma_mapping_error(ppgtt->base.dev->pdev, pd_addr);
+		ret = gen8_ppgtt_setup_page_directories(ppgtt, i);
 		if (ret)
 			goto bail;
-
-		ppgtt->pd_dma_addr[i] = pd_addr;
 	}
 
 	/*
@@ -488,7 +557,7 @@ static int gen8_ppgtt_init(struct i915_hw_ppgtt *ppgtt, uint64_t size)
 			 ppgtt->num_pd_pages, ppgtt->num_pd_pages - max_pdp);
 	DRM_DEBUG_DRIVER("Allocated %d pages for page tables (%lld wasted)\n",
 			 ppgtt->num_pt_pages,
-			 (ppgtt->num_pt_pages - num_pt_pages) +
+			 (ppgtt->num_pt_pages - min_pt_pages) +
 			 size % (1<<30));
 	return 0;
 
Signature · 1 line
1.8.5.2
#5 lore raw

PATCH v2 3/7 drm/i915/bdw: Split ppgtt initialization up

Like cleanup in an earlier patch, the code becomes much more readable, and easier to extend if we extract out helper functions for the various stages of init.

Note that with this patch it becomes really simple, and tempting to begin using the 'goto out' idiom with explicit free/fini semantics. I've kept the error path as similar as possible to the cleanup() function to make sure cleanup is as robust as possible

v2: Remove comment "NB:From here on, ppgtt->base.cleanup() should function properly"
Update commit message to reflect above

---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 182 +++++++++++++++++++++++++-----------
 1 file changed, 126 insertions(+), 56 deletions(-)
drivers/gpu/drm/i915/i915_gem_gtt.c +126 -56
index 502b1dc..a19cc73 100644
@@ -364,91 +364,161 @@ static void gen8_ppgtt_cleanup(struct i915_address_space *vm)
 	gen8_ppgtt_free(ppgtt);
 }
 
-/**
- * GEN8 legacy ppgtt programming is accomplished through a max 4 PDP registers
- * with a net effect resembling a 2-level page table in normal x86 terms. Each
- * PDP represents 1GB of memory 4 * 512 * 512 * 4096 = 4GB legacy 32b address
- * space.
- *
- * FIXME: split allocation into smaller pieces. For now we only ever do this
- * once, but with full PPGTT, the multiple contiguous allocations will be bad.
- * TODO: Do something with the size parameter
- */
-static int gen8_ppgtt_init(struct i915_hw_ppgtt *ppgtt, uint64_t size)
+static int gen8_ppgtt_allocate_page_tables(struct i915_hw_ppgtt *ppgtt,
+					   const int max_pdp)
 {
 	struct page *pt_pages;
-	const int max_pdp = DIV_ROUND_UP(size, 1 << 30);
 	const int num_pt_pages = GEN8_PDES_PER_PAGE * max_pdp;
-	int i, j, ret;
-
-	if (size % (1<<30))
-		DRM_INFO("Pages will be wasted unless GTT size (%llu) is divisible by 1GB\n", size);
-
-	/* 1. Do all our allocations for page directories and page tables */
-	ppgtt->pd_pages = alloc_pages(GFP_KERNEL, get_order(max_pdp << PAGE_SHIFT));
-	if (!ppgtt->pd_pages)
-		return -ENOMEM;
 
 	pt_pages = alloc_pages(GFP_KERNEL, get_order(num_pt_pages << PAGE_SHIFT));
-	if (!pt_pages) {
-		__free_pages(ppgtt->pd_pages, get_order(max_pdp << PAGE_SHIFT));
+	if (!pt_pages)
 		return -ENOMEM;
-	}
 
 	ppgtt->gen8_pt_pages = pt_pages;
-	ppgtt->num_pd_pages = 1 << get_order(max_pdp << PAGE_SHIFT);
 	ppgtt->num_pt_pages = 1 << get_order(num_pt_pages << PAGE_SHIFT);
-	ppgtt->num_pd_entries = max_pdp * GEN8_PDES_PER_PAGE;
-	BUG_ON(ppgtt->num_pd_pages > GEN8_LEGACY_PDPS);
 
-	ppgtt->gen8_pt_dma_addr = kcalloc(max_pdp,
+	return 0;
+}
+
+static int gen8_ppgtt_allocate_dma(struct i915_hw_ppgtt *ppgtt)
+{
+	int i;
+
+	ppgtt->gen8_pt_dma_addr = kcalloc(ppgtt->num_pd_entries,
 					  sizeof(*ppgtt->gen8_pt_dma_addr),
 					  GFP_KERNEL);
-	if (!ppgtt->gen8_pt_dma_addr) {
-		ret = -ENOMEM;
-		goto bail;
-	}
+	if (!ppgtt->gen8_pt_dma_addr)
+		return -ENOMEM;
 
-	for (i = 0; i < max_pdp; i++) {
+	for (i = 0; i < ppgtt->num_pd_entries; i++) {
 		ppgtt->gen8_pt_dma_addr[i] = kcalloc(GEN8_PDES_PER_PAGE,
 						     sizeof(dma_addr_t),
 						     GFP_KERNEL);
 		if (!ppgtt->gen8_pt_dma_addr[i]) {
-			ret = -ENOMEM;
-			goto bail;
+			kfree(ppgtt->gen8_pt_dma_addr);
+			while(i--)
+				kfree(ppgtt->gen8_pt_dma_addr[i]);
+
+			return -ENOMEM;
 		}
 	}
 
+	return 0;
+}
+
+static int gen8_ppgtt_allocate_page_directories(struct i915_hw_ppgtt *ppgtt,
+						const int max_pdp)
+{
+	ppgtt->pd_pages = alloc_pages(GFP_KERNEL, get_order(max_pdp << PAGE_SHIFT));
+	if (!ppgtt->pd_pages)
+		return -ENOMEM;
+
+	ppgtt->num_pd_pages = 1 << get_order(max_pdp << PAGE_SHIFT);
+
+	return 0;
+}
+
+static int gen8_ppgtt_alloc(struct i915_hw_ppgtt *ppgtt,
+			    const int max_pdp)
+{
+	int ret;
+
+	ret = gen8_ppgtt_allocate_page_directories(ppgtt, max_pdp);
+	if (ret)
+		return ret;
+
+	ret = gen8_ppgtt_allocate_page_tables(ppgtt, max_pdp);
+	if (ret) {
+		__free_pages(ppgtt->pd_pages, get_order(max_pdp << PAGE_SHIFT));
+		return ret;
+	}
+
+	ppgtt->num_pd_entries = max_pdp * GEN8_PDES_PER_PAGE;
+	BUG_ON(ppgtt->num_pd_pages > GEN8_LEGACY_PDPS);
+
+	ret = gen8_ppgtt_allocate_dma(ppgtt);
+	if (ret)
+		gen8_ppgtt_free(ppgtt);
+
+	return ret;
+}
+
+static int gen8_ppgtt_setup_page_directories(struct i915_hw_ppgtt *ppgtt,
+					     const int pd)
+{
+	dma_addr_t pd_addr;
+	int ret;
+
+	pd_addr = pci_map_page(ppgtt->base.dev->pdev,
+			       &ppgtt->pd_pages[pd], 0,
+			       PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
+
+	ret = pci_dma_mapping_error(ppgtt->base.dev->pdev, pd_addr);
+	if (ret)
+		return ret;
+
+	ppgtt->pd_dma_addr[pd] = pd_addr;
+
+	return 0;
+}
+
+static int gen8_ppgtt_setup_page_tables(struct i915_hw_ppgtt *ppgtt,
+					const int pd,
+					const int pt)
+{
+	dma_addr_t pt_addr;
+	struct page *p;
+	int ret;
+
+	p = &ppgtt->gen8_pt_pages[pd * GEN8_PDES_PER_PAGE + pt];
+	pt_addr = pci_map_page(ppgtt->base.dev->pdev,
+			       p, 0, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
+	ret = pci_dma_mapping_error(ppgtt->base.dev->pdev, pt_addr);
+	if (ret)
+		return ret;
+
+	ppgtt->gen8_pt_dma_addr[pd][pt] = pt_addr;
+
+	return 0;
+}
+
+/**
+ * GEN8 legacy ppgtt programming is accomplished through a max 4 PDP registers
+ * with a net effect resembling a 2-level page table in normal x86 terms. Each
+ * PDP represents 1GB of memory 4 * 512 * 512 * 4096 = 4GB legacy 32b address
+ * space.
+ *
+ * FIXME: split allocation into smaller pieces. For now we only ever do this
+ * once, but with full PPGTT, the multiple contiguous allocations will be bad.
+ * TODO: Do something with the size parameter
+ */
+static int gen8_ppgtt_init(struct i915_hw_ppgtt *ppgtt, uint64_t size)
+{
+	const int max_pdp = DIV_ROUND_UP(size, 1 << 30);
+	const int min_pt_pages = GEN8_PDES_PER_PAGE * max_pdp;
+	int i, j, ret;
+
+	if (size % (1<<30))
+		DRM_INFO("Pages will be wasted unless GTT size (%llu) is divisible by 1GB\n", size);
+
+	/* 1. Do all our allocations for page directories and page tables. */
+	ret = gen8_ppgtt_alloc(ppgtt, max_pdp);
+	if (ret)
+		return ret;
+
 	/*
-	 * 2. Create all the DMA mappings for the page directories and page
-	 * tables
+	 * 2. Create DMA mappings for the page directories and page tables.
 	 */
 	for (i = 0; i < max_pdp; i++) {
-		dma_addr_t pd_addr, pt_addr;
-
-		/* Get the page table mappings per page directory */
 		for (j = 0; j < GEN8_PDES_PER_PAGE; j++) {
-			struct page *p = &pt_pages[i * GEN8_PDES_PER_PAGE + j];
-
-			pt_addr = pci_map_page(ppgtt->base.dev->pdev,
-					       p, 0, PAGE_SIZE,
-					       PCI_DMA_BIDIRECTIONAL);
-			ret = pci_dma_mapping_error(ppgtt->base.dev->pdev, pt_addr);
+			ret = gen8_ppgtt_setup_page_tables(ppgtt, i, j);
 			if (ret)
 				goto bail;
-
-			ppgtt->gen8_pt_dma_addr[i][j] = pt_addr;
 		}
 
-		/* And the page directory mappings */
-		pd_addr = pci_map_page(ppgtt->base.dev->pdev,
-				       &ppgtt->pd_pages[i], 0,
-				       PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
-		ret = pci_dma_mapping_error(ppgtt->base.dev->pdev, pd_addr);
+		ret = gen8_ppgtt_setup_page_directories(ppgtt, i);
 		if (ret)
 			goto bail;
-
-		ppgtt->pd_dma_addr[i] = pd_addr;
 	}
 
 	/*
@@ -486,7 +556,7 @@ static int gen8_ppgtt_init(struct i915_hw_ppgtt *ppgtt, uint64_t size)
 			 ppgtt->num_pd_pages, ppgtt->num_pd_pages - max_pdp);
 	DRM_DEBUG_DRIVER("Allocated %d pages for page tables (%lld wasted)\n",
 			 ppgtt->num_pt_pages,
-			 (ppgtt->num_pt_pages - num_pt_pages) +
+			 (ppgtt->num_pt_pages - min_pt_pages) +
 			 size % (1<<30));
 	return 0;
 
Signature · 1 line
1.8.5.2
#6 lore raw

PATCH 4/7 drm/i915: Make clear/insert vfuncs args absolute

In reply to Ben Widawsky

This patch converts insert_entries and clear_range, both functions which are specific to the VM. These functions tend to encapsulate the gen specific PTE writes. Passing absolute addresses to the insert_entries, and clear_range will help make the logic clearer within the functions as to what's going on. Currently, all callers simply do the appropriate page shift, which IMO, ends up looking weird with an upcoming change for the gen8 page table allocations.

Up until now, the PPGTT was a funky 2 level page table. GEN8 changes this to look more like a 3 level page table, and to that extent we need a significant amount more memory simply for the page tables. To address this, the allocations will be split up in finer amounts.

---
 drivers/gpu/drm/i915/i915_drv.h     |  6 +--
 drivers/gpu/drm/i915/i915_gem_gtt.c | 80 +++++++++++++++++++++----------------
 2 files changed, 49 insertions(+), 37 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +3 -3
index dc21963..43c2ee7 100644
@@ -614,12 +614,12 @@ struct i915_address_space {
 				     enum i915_cache_level level,
 				     bool valid); /* Create a valid PTE */
 	void (*clear_range)(struct i915_address_space *vm,
-			    unsigned int first_entry,
-			    unsigned int num_entries,
+			    uint64_t start,
+			    size_t length,
 			    bool use_scratch);
 	void (*insert_entries)(struct i915_address_space *vm,
 			       struct sg_table *st,
-			       unsigned int first_entry,
+			       uint64_t start,
 			       enum i915_cache_level cache_level);
 	void (*cleanup)(struct i915_address_space *vm);
 };
drivers/gpu/drm/i915/i915_gem_gtt.c +46 -34

230 lines not on this page · raw message

Signature · 1 line
1.8.5.2
#7 lore raw

PATCH 5/7 drm/i915/bdw: Reorganize PT allocations

In reply to Ben Widawsky

The previous allocation mechanism would get 2 contiguous allocations, one for the page directories, and one for the page tables. As each page table is 1 page, and there are 512 of these per page directory, this goes to 1MB. An unfriendly request at best. Worse still, our HW now supports 4 page directories, and a 2MB allocation is not allowed.

In order to fix this, this patch attempts to split up each page table allocation into a single, discrete allocation. There is nothing really fancy about the patch itself, it just has to manage an extra pointer indirection, and have a fancier bit of logic to free up the pages.

To accommodate some of the added complexity, two new helpers are introduced to allocate, and free the page table pages.

NOTE: I really wanted to split the way we do allocations, and the way in which we identify the page table/page directory being used. I found splitting this functionality up to be too unwieldy. I apologize in advance to the reviewer. I'd recommend looking at the result, rather than the diff.

---
 drivers/gpu/drm/i915/i915_drv.h     |   5 +-
 drivers/gpu/drm/i915/i915_gem_gtt.c | 135 +++++++++++++++++++++++++++---------
 2 files changed, 107 insertions(+), 33 deletions(-)
drivers/gpu/drm/i915/i915_drv.h +3 -2
index 43c2ee7..6b23b13 100644
@@ -653,6 +653,7 @@ struct i915_gtt {
 };
 #define gtt_total_entries(gtt) ((gtt).base.total >> PAGE_SHIFT)
 
+#define GEN8_LEGACY_PDPS 4
 struct i915_hw_ppgtt {
 	struct i915_address_space base;
 	struct kref ref;
@@ -660,14 +661,14 @@ struct i915_hw_ppgtt {
 	unsigned num_pd_entries;
 	union {
 		struct page **pt_pages;
-		struct page *gen8_pt_pages;
+		struct page **gen8_pt_pages[GEN8_LEGACY_PDPS];
 	};
 	struct page *pd_pages;
 	int num_pd_pages;
 	int num_pt_pages;
 	union {
 		uint32_t pd_offset;
-		dma_addr_t pd_dma_addr[4];
+		dma_addr_t pd_dma_addr[GEN8_LEGACY_PDPS];
 	};
 	union {
 		dma_addr_t *pt_dma_addr;
drivers/gpu/drm/i915/i915_gem_gtt.c +104 -31

209 lines not on this page · raw message

Signature · 1 line
1.8.5.2
#8 lore raw

PATCH 6/7 Revert "drm/i915/bdw: Limit GTT to 2GB"

In reply to Ben Widawsky

This reverts commit 3a2ffb65eec6dbda2fd8151894f51c18b42c8d41.

Now that the code is fixed to use smaller allocations, it should be safe to let the full GGTT be used on DW.

The testcase for this is anything which uses more than half of the GTT, thus eclipsing the old limit.

---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 5 -----
 1 file changed, 5 deletions(-)
drivers/gpu/drm/i915/i915_gem_gtt.c +0 -5
index 8946b1c..37955db 100644
@@ -1731,11 +1731,6 @@ static inline unsigned int gen8_get_total_gtt_size(u16 bdw_gmch_ctl)
 	bdw_gmch_ctl &= BDW_GMCH_GGMS_MASK;
 	if (bdw_gmch_ctl)
 		bdw_gmch_ctl = 1 << bdw_gmch_ctl;
-	if (bdw_gmch_ctl > 4) {
-		WARN_ON(!i915_preliminary_hw_support);
-		return 4<<20;
-	}
-
 	return bdw_gmch_ctl << 20;
 }
 
Signature · 1 line
1.8.5.2
#9 lore raw

PATCH 7/7 drm/i915: Update i915_gem_gtt.c copyright

In reply to Ben Widawsky

I keep meaning to do this... by now almost the entire file has been written by an Intel employee (including Daniel post-2010).

---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 1 +
 1 file changed, 1 insertion(+)
drivers/gpu/drm/i915/i915_gem_gtt.c +1 -0
index 37955db..c21641f 100644
@@ -1,5 +1,6 @@
 /*
  * Copyright © 2010 Daniel Vetter
+ * Copyright © 2011-2013 Intel Corporation
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
Signature · 1 line
1.8.5.2

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#10 lore raw

PATCH 8/9 drm/i915: Split GEN6 PPGTT cleanup

In reply to Ben Widawsky

This cleanup is similar to the GEN8 cleanup (though less necessary). Having everything split will make cleaning the initialization path error paths easier to understand.

---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 26 +++++++++++++++++++-------
 1 file changed, 19 insertions(+), 7 deletions(-)
drivers/gpu/drm/i915/i915_gem_gtt.c +19 -7
index 5aa6680..fc81daf 100644
@@ -1004,22 +1004,21 @@ static void gen6_ppgtt_insert_entries(struct i915_address_space *vm,
 	kunmap_atomic(pt_vaddr);
 }
 
-static void gen6_ppgtt_cleanup(struct i915_address_space *vm)
+static void gen6_ppgtt_unmap_pages(struct i915_hw_ppgtt *ppgtt)
 {
-	struct i915_hw_ppgtt *ppgtt =
-		container_of(vm, struct i915_hw_ppgtt, base);
 	int i;
 
-	list_del(&vm->global_link);
-	drm_mm_takedown(&ppgtt->base.mm);
-	drm_mm_remove_node(&ppgtt->node);
-
 	if (ppgtt->pt_dma_addr) {
 		for (i = 0; i < ppgtt->num_pd_entries; i++)
 			pci_unmap_page(ppgtt->base.dev->pdev,
 				       ppgtt->pt_dma_addr[i],
 				       4096, PCI_DMA_BIDIRECTIONAL);
 	}
+}
+
+static void gen6_ppgtt_free(struct i915_hw_ppgtt *ppgtt)
+{
+	int i;
 
 	kfree(ppgtt->pt_dma_addr);
 	for (i = 0; i < ppgtt->num_pd_entries; i++)
@@ -1028,6 +1027,19 @@ static void gen6_ppgtt_cleanup(struct i915_address_space *vm)
 	kfree(ppgtt);
 }
 
+static void gen6_ppgtt_cleanup(struct i915_address_space *vm)
+{
+	struct i915_hw_ppgtt *ppgtt =
+		container_of(vm, struct i915_hw_ppgtt, base);
+
+	list_del(&vm->global_link);
+	drm_mm_takedown(&ppgtt->base.mm);
+	drm_mm_remove_node(&ppgtt->node);
+
+	gen6_ppgtt_unmap_pages(ppgtt);
+	gen6_ppgtt_free(ppgtt);
+}
+
 static int gen6_ppgtt_init(struct i915_hw_ppgtt *ppgtt)
 {
 #define GEN6_PD_ALIGN (PAGE_SIZE * 16)
Signature · 1 line
1.8.5.2
#11 lore raw

PATCH 9/9 drm/i915: Split GEN6 PPGTT initialization up

Simply to match the GEN8 style of PPGTT initialization, split up the allocations and mappings. Unlike GEN8, we skip a separate dma_addr_t allocation function, as it is much simpler pre-gen8.

With this code it would be easy to make a more general PPGTT initialization function with per GEN alloc/map/etc. or use a common helper, similar to the ringbuffer code. I don't see a benefit to doing this just yet, but who knows...

---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 141 +++++++++++++++++++++++-------------
 1 file changed, 91 insertions(+), 50 deletions(-)
drivers/gpu/drm/i915/i915_gem_gtt.c +91 -50
index fc81daf..fd53789 100644
@@ -1040,14 +1040,14 @@ static void gen6_ppgtt_cleanup(struct i915_address_space *vm)
 	gen6_ppgtt_free(ppgtt);
 }
 
-static int gen6_ppgtt_init(struct i915_hw_ppgtt *ppgtt)
+static int gen6_ppgtt_allocate_page_directories(struct i915_hw_ppgtt *ppgtt)
 {
 #define GEN6_PD_ALIGN (PAGE_SIZE * 16)
 #define GEN6_PD_SIZE (GEN6_PPGTT_PD_ENTRIES * PAGE_SIZE)
 	struct drm_device *dev = ppgtt->base.dev;
 	struct drm_i915_private *dev_priv = dev->dev_private;
 	bool retried = false;
-	int i, ret;
+	int ret;
 
 	/* PPGTT PDEs reside in the GGTT and consists of 512 entries. The
 	 * allocator works in address space sizes, so it's multiplied by page
@@ -1074,42 +1074,60 @@ alloc:
 	if (ppgtt->node.start < dev_priv->gtt.mappable_end)
 		DRM_DEBUG("Forced to use aperture for PDEs\n");
 
-	ppgtt->base.pte_encode = dev_priv->gtt.base.pte_encode;
 	ppgtt->num_pd_entries = GEN6_PPGTT_PD_ENTRIES;
-	if (IS_GEN6(dev)) {
-		ppgtt->enable = gen6_ppgtt_enable;
-		ppgtt->switch_mm = gen6_mm_switch;
-	} else if (IS_HASWELL(dev)) {
-		ppgtt->enable = gen7_ppgtt_enable;
-		ppgtt->switch_mm = hsw_mm_switch;
-	} else if (IS_GEN7(dev)) {
-		ppgtt->enable = gen7_ppgtt_enable;
-		ppgtt->switch_mm = gen7_mm_switch;
-	} else
-		BUG();
-	ppgtt->base.clear_range = gen6_ppgtt_clear_range;
-	ppgtt->base.insert_entries = gen6_ppgtt_insert_entries;
-	ppgtt->base.cleanup = gen6_ppgtt_cleanup;
-	ppgtt->base.scratch = dev_priv->gtt.base.scratch;
-	ppgtt->base.start = 0;
-	ppgtt->base.total = GEN6_PPGTT_PD_ENTRIES * I915_PPGTT_PT_ENTRIES * PAGE_SIZE;
+	return ret;
+}
+
+static int gen6_ppgtt_allocate_page_tables(struct i915_hw_ppgtt *ppgtt)
+{
+	int i;
+
 	ppgtt->pt_pages = kcalloc(ppgtt->num_pd_entries, sizeof(struct page *),
 				  GFP_KERNEL);
-	if (!ppgtt->pt_pages) {
-		drm_mm_remove_node(&ppgtt->node);
+
+	if (!ppgtt->pt_pages)
 		return -ENOMEM;
-	}
 
 	for (i = 0; i < ppgtt->num_pd_entries; i++) {
 		ppgtt->pt_pages[i] = alloc_page(GFP_KERNEL);
-		if (!ppgtt->pt_pages[i])
-			goto err_pt_alloc;
+		if (!ppgtt->pt_pages[i]) {
+			gen6_ppgtt_free(ppgtt);
+			return -ENOMEM;
+		}
+	}
+
+	return 0;
+}
+
+static int gen6_ppgtt_alloc(struct i915_hw_ppgtt *ppgtt)
+{
+	int ret;
+
+	ret = gen6_ppgtt_allocate_page_directories(ppgtt);
+	if (ret)
+		return ret;
+
+	ret = gen6_ppgtt_allocate_page_tables(ppgtt);
+	if (ret) {
+		drm_mm_remove_node(&ppgtt->node);
+		return ret;
 	}
 
 	ppgtt->pt_dma_addr = kcalloc(ppgtt->num_pd_entries, sizeof(dma_addr_t),
 				     GFP_KERNEL);
-	if (!ppgtt->pt_dma_addr)
-		goto err_pt_alloc;
+	if (!ppgtt->pt_dma_addr) {
+		drm_mm_remove_node(&ppgtt->node);
+		gen6_ppgtt_free(ppgtt);
+		return -ENOMEM;
+	}
+
+	return 0;
+}
+
+static int gen6_ppgtt_setup_page_tables(struct i915_hw_ppgtt *ppgtt)
+{
+	struct drm_device *dev = ppgtt->base.dev;
+	int i;
 
 	for (i = 0; i < ppgtt->num_pd_entries; i++) {
 		dma_addr_t pt_addr;
@@ -1118,40 +1136,63 @@ alloc:
 				       PCI_DMA_BIDIRECTIONAL);
 
 		if (pci_dma_mapping_error(dev->pdev, pt_addr)) {
-			ret = -EIO;
-			goto err_pd_pin;
-
+			gen6_ppgtt_unmap_pages(ppgtt);
+			return -EIO;
 		}
+
 		ppgtt->pt_dma_addr[i] = pt_addr;
 	}
 
-	ppgtt->base.clear_range(&ppgtt->base, 0, ppgtt->base.total, true);
+	return 0;
+}
+
+static int gen6_ppgtt_init(struct i915_hw_ppgtt *ppgtt)
+{
+	struct drm_device *dev = ppgtt->base.dev;
+	struct drm_i915_private *dev_priv = dev->dev_private;
+	int ret;
+
+	ppgtt->base.pte_encode = dev_priv->gtt.base.pte_encode;
+	if (IS_GEN6(dev)) {
+		ppgtt->enable = gen6_ppgtt_enable;
+		ppgtt->switch_mm = gen6_mm_switch;
+	} else if (IS_HASWELL(dev)) {
+		ppgtt->enable = gen7_ppgtt_enable;
+		ppgtt->switch_mm = hsw_mm_switch;
+	} else if (IS_GEN7(dev)) {
+		ppgtt->enable = gen7_ppgtt_enable;
+		ppgtt->switch_mm = gen7_mm_switch;
+	} else
+		BUG();
+
+	ret = gen6_ppgtt_alloc(ppgtt);
+	if (ret)
+		return ret;
+
+	ret = gen6_ppgtt_setup_page_tables(ppgtt);
+	if (ret) {
+		gen6_ppgtt_free(ppgtt);
+		return ret;
+	}
+
+	ppgtt->base.clear_range = gen6_ppgtt_clear_range;
+	ppgtt->base.insert_entries = gen6_ppgtt_insert_entries;
+	ppgtt->base.cleanup = gen6_ppgtt_cleanup;
+	ppgtt->base.scratch = dev_priv->gtt.base.scratch;
+	ppgtt->base.start = 0;
+	ppgtt->base.total = GEN6_PPGTT_PD_ENTRIES * I915_PPGTT_PT_ENTRIES * PAGE_SIZE;
 	ppgtt->debug_dump = gen6_dump_ppgtt;
 
-	DRM_DEBUG_DRIVER("Allocated pde space (%ldM) at GTT entry: %lx\n",
-			 ppgtt->node.size >> 20,
-			 ppgtt->node.start / PAGE_SIZE);
 	ppgtt->pd_offset =
 		ppgtt->node.start / PAGE_SIZE * sizeof(gen6_gtt_pte_t);
 
-	return 0;
+	ppgtt->base.clear_range(&ppgtt->base, 0, ppgtt->base.total, true);
 
-err_pd_pin:
-	if (ppgtt->pt_dma_addr) {
-		for (i--; i >= 0; i--)
-			pci_unmap_page(dev->pdev, ppgtt->pt_dma_addr[i],
-				       4096, PCI_DMA_BIDIRECTIONAL);
-	}
-err_pt_alloc:
-	kfree(ppgtt->pt_dma_addr);
-	for (i = 0; i < ppgtt->num_pd_entries; i++) {
-		if (ppgtt->pt_pages[i])
-			__free_page(ppgtt->pt_pages[i]);
-	}
-	kfree(ppgtt->pt_pages);
-	drm_mm_remove_node(&ppgtt->node);
+	DRM_DEBUG_DRIVER("Allocated pde space (%ldM) at GTT entry: %lx\n",
+			 ppgtt->node.size >> 20,
+			 ppgtt->node.start / PAGE_SIZE);
 
-	return ret;
+	return 0;
 }
 
 int i915_gem_init_ppgtt(struct drm_device *dev, struct i915_hw_ppgtt *ppgtt)
Signature · 1 line
1.8.5.2