[PATCH v8 0/2] drm/i915: Hugepage manager and test for MTL

[PATCH v8 0/2] drm/i915: Hugepage manager and test for MTL

[Andrzej Hajda]

This patchset patches sent by Jonathan and Andi, with 
addressed CI failures:
1. Fixed checking alignment of 64K pages on both Pre-Gen12 and Gen12.
2. Fixed start alignment of 2M pages.

Regards
Andrzej

Jonathan Cavitt (2):
  drm/i915: Migrate platform-dependent mock hugepage selftests to live
  drm/i915: Use correct huge page manager for MTL

.../gpu/drm/i915/gem/selftests/huge_pages.c   | 88 +++++++++++++++----
 drivers/gpu/drm/i915/gt/gen8_ppgtt.c          |  3 +-
 2 files changed, 71 insertions(+), 20 deletions(-)

Cc: intel-gfx@lists.freedesktop.org
Cc: linux-kernel@vger.kernel.org
Cc: Jonathan Cavitt <jonathan.cavitt@intel.com>
Cc: Andrzej Hajda <andrzej.hajda@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
-- 
2.39.2

---
Changes in v8:
- described in patches
- Link to v7: https://lore.kernel.org/r/20230425-hugepage-migrate-v7-0-fa6605a986c9@intel.com

---
Jonathan Cavitt (2):
      drm/i915: Migrate platform-dependent mock hugepage selftests to live
      drm/i915: Use correct huge page manager for MTL

 drivers/gpu/drm/i915/gem/selftests/huge_pages.c | 106 ++++++++++++++++++------
 drivers/gpu/drm/i915/gt/gen8_ppgtt.c            |   3 +-
 2 files changed, 83 insertions(+), 26 deletions(-)
---
base-commit: 4d0066a1c0763d50b6fb017e27d12b081ce21b57
change-id: 20230425-hugepage-migrate-1869aaf31a6d

Best regards,
-- 
Andrzej Hajda <andrzej.hajda@intel.com>

[PATCH v8 1/2] drm/i915: Migrate platform-dependent mock hugepage selftests to live

[Andrzej Hajda]

From: Jonathan Cavitt <jonathan.cavitt@intel.com>

Convert the igt_mock_ppgtt_huge_fill and igt_mock_ppgtt_64K mock selftests
into live selftests as their requirements have recently become
platform-dependent. Additionally, apply necessary platform dependency
checks to these tests.

v8:
- handle properly 64K and 2M pages
v9:
- do not expect 64K pages if 2M are present
- fix hex printing
- obey commit message line limit

Signed-off-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Co-developed-by: Andrzej Hajda <andrzej.hajda@intel.com>
Signed-off-by: Andrzej Hajda <andrzej.hajda@intel.com>
---
 drivers/gpu/drm/i915/gem/selftests/huge_pages.c | 106 ++++++++++++++++++------
 1 file changed, 81 insertions(+), 25 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
index defece0bcb811f..cb5863f37f9d37 100644
--- a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
+++ b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
@@ -695,8 +695,7 @@ static int igt_mock_ppgtt_misaligned_dma(void *arg)
 	return err;
 }
 
-static void close_object_list(struct list_head *objects,
-			      struct i915_ppgtt *ppgtt)
+static void close_object_list(struct list_head *objects)
 {
 	struct drm_i915_gem_object *obj, *on;
 
@@ -710,17 +709,36 @@ static void close_object_list(struct list_head *objects,
 	}
 }
 
-static int igt_mock_ppgtt_huge_fill(void *arg)
+static int igt_ppgtt_huge_fill(void *arg)
 {
-	struct i915_ppgtt *ppgtt = arg;
-	struct drm_i915_private *i915 = ppgtt->vm.i915;
-	unsigned long max_pages = ppgtt->vm.total >> PAGE_SHIFT;
+	struct drm_i915_private *i915 = arg;
+	unsigned int supported = RUNTIME_INFO(i915)->page_sizes;
+	bool has_pte64 = GRAPHICS_VER_FULL(i915) >= IP_VER(12, 50);
+	struct i915_address_space *vm;
+	struct i915_gem_context *ctx;
+	unsigned long max_pages;
 	unsigned long page_num;
+	struct file *file;
 	bool single = false;
 	LIST_HEAD(objects);
 	IGT_TIMEOUT(end_time);
 	int err = -ENODEV;
 
+	if (supported == I915_GTT_PAGE_SIZE_4K)
+		return 0;
+
+	file = mock_file(i915);
+	if (IS_ERR(file))
+		return PTR_ERR(file);
+
+	ctx = hugepage_ctx(i915, file);
+	if (IS_ERR(ctx)) {
+		err = PTR_ERR(ctx);
+		goto out;
+	}
+	vm = i915_gem_context_get_eb_vm(ctx);
+	max_pages = vm->total >> PAGE_SHIFT;
+
 	for_each_prime_number_from(page_num, 1, max_pages) {
 		struct drm_i915_gem_object *obj;
 		u64 size = page_num << PAGE_SHIFT;
@@ -750,13 +768,14 @@ static int igt_mock_ppgtt_huge_fill(void *arg)
 
 		list_add(&obj->st_link, &objects);
 
-		vma = i915_vma_instance(obj, &ppgtt->vm, NULL);
+		vma = i915_vma_instance(obj, vm, NULL);
 		if (IS_ERR(vma)) {
 			err = PTR_ERR(vma);
 			break;
 		}
 
-		err = i915_vma_pin(vma, 0, 0, PIN_USER);
+		/* vma start must be aligned to BIT(21) to allow 2M PTEs */
+		err = i915_vma_pin(vma, 0, BIT(21), PIN_USER);
 		if (err)
 			break;
 
@@ -784,12 +803,13 @@ static int igt_mock_ppgtt_huge_fill(void *arg)
 		GEM_BUG_ON(!expected_gtt);
 		GEM_BUG_ON(size);
 
-		if (expected_gtt & I915_GTT_PAGE_SIZE_4K)
+		if (!has_pte64 && (obj->base.size < I915_GTT_PAGE_SIZE_2M ||
+				   expected_gtt & I915_GTT_PAGE_SIZE_2M))
 			expected_gtt &= ~I915_GTT_PAGE_SIZE_64K;
 
 		i915_vma_unpin(vma);
 
-		if (vma->page_sizes.sg & I915_GTT_PAGE_SIZE_64K) {
+		if (!has_pte64 && vma->page_sizes.sg & I915_GTT_PAGE_SIZE_64K) {
 			if (!IS_ALIGNED(vma->node.start,
 					I915_GTT_PAGE_SIZE_2M)) {
 				pr_err("node.start(%llx) not aligned to 2M\n",
@@ -808,7 +828,7 @@ static int igt_mock_ppgtt_huge_fill(void *arg)
 		}
 
 		if (vma->resource->page_sizes_gtt != expected_gtt) {
-			pr_err("gtt=%u, expected=%u, size=%zd, single=%s\n",
+			pr_err("gtt=%#x, expected=%#x, size=0x%zx, single=%s\n",
 			       vma->resource->page_sizes_gtt, expected_gtt,
 			       obj->base.size, str_yes_no(!!single));
 			err = -EINVAL;
@@ -823,19 +843,25 @@ static int igt_mock_ppgtt_huge_fill(void *arg)
 		single = !single;
 	}
 
-	close_object_list(&objects, ppgtt);
+	close_object_list(&objects);
 
 	if (err == -ENOMEM || err == -ENOSPC)
 		err = 0;
 
+	i915_vm_put(vm);
+out:
+	fput(file);
 	return err;
 }
 
-static int igt_mock_ppgtt_64K(void *arg)
+static int igt_ppgtt_64K(void *arg)
 {
-	struct i915_ppgtt *ppgtt = arg;
-	struct drm_i915_private *i915 = ppgtt->vm.i915;
+	struct drm_i915_private *i915 = arg;
+	bool has_pte64 = GRAPHICS_VER_FULL(i915) >= IP_VER(12, 50);
 	struct drm_i915_gem_object *obj;
+	struct i915_address_space *vm;
+	struct i915_gem_context *ctx;
+	struct file *file;
 	const struct object_info {
 		unsigned int size;
 		unsigned int gtt;
@@ -907,16 +933,41 @@ static int igt_mock_ppgtt_64K(void *arg)
 	if (!HAS_PAGE_SIZES(i915, I915_GTT_PAGE_SIZE_64K))
 		return 0;
 
+	file = mock_file(i915);
+	if (IS_ERR(file))
+		return PTR_ERR(file);
+
+	ctx = hugepage_ctx(i915, file);
+	if (IS_ERR(ctx)) {
+		err = PTR_ERR(ctx);
+		goto out;
+	}
+	vm = i915_gem_context_get_eb_vm(ctx);
+
 	for (i = 0; i < ARRAY_SIZE(objects); ++i) {
 		unsigned int size = objects[i].size;
 		unsigned int expected_gtt = objects[i].gtt;
 		unsigned int offset = objects[i].offset;
 		unsigned int flags = PIN_USER;
 
+		/*
+		 * For modern GTT models, the requirements for marking a page-table
+		 * as 64K have been relaxed.  Account for this.
+		 */
+		if (has_pte64) {
+			expected_gtt = 0;
+			if (size >= SZ_64K)
+				expected_gtt |= I915_GTT_PAGE_SIZE_64K;
+			if (size & (SZ_64K - 1))
+				expected_gtt |= I915_GTT_PAGE_SIZE_4K;
+		}
+
 		for (single = 0; single <= 1; single++) {
 			obj = fake_huge_pages_object(i915, size, !!single);
-			if (IS_ERR(obj))
-				return PTR_ERR(obj);
+			if (IS_ERR(obj)) {
+				err = PTR_ERR(obj);
+				goto out_vm;
+			}
 
 			err = i915_gem_object_pin_pages_unlocked(obj);
 			if (err)
@@ -928,7 +979,7 @@ static int igt_mock_ppgtt_64K(void *arg)
 			 */
 			obj->mm.page_sizes.sg &= ~I915_GTT_PAGE_SIZE_2M;
 
-			vma = i915_vma_instance(obj, &ppgtt->vm, NULL);
+			vma = i915_vma_instance(obj, vm, NULL);
 			if (IS_ERR(vma)) {
 				err = PTR_ERR(vma);
 				goto out_object_unpin;
@@ -945,7 +996,8 @@ static int igt_mock_ppgtt_64K(void *arg)
 			if (err)
 				goto out_vma_unpin;
 
-			if (!offset && vma->page_sizes.sg & I915_GTT_PAGE_SIZE_64K) {
+			if (!has_pte64 && !offset &&
+			    vma->page_sizes.sg & I915_GTT_PAGE_SIZE_64K) {
 				if (!IS_ALIGNED(vma->node.start,
 						I915_GTT_PAGE_SIZE_2M)) {
 					pr_err("node.start(%llx) not aligned to 2M\n",
@@ -964,9 +1016,10 @@ static int igt_mock_ppgtt_64K(void *arg)
 			}
 
 			if (vma->resource->page_sizes_gtt != expected_gtt) {
-				pr_err("gtt=%u, expected=%u, i=%d, single=%s\n",
+				pr_err("gtt=%#x, expected=%#x, i=%d, single=%s offset=%#x size=%#x\n",
 				       vma->resource->page_sizes_gtt,
-				       expected_gtt, i, str_yes_no(!!single));
+				       expected_gtt, i, str_yes_no(!!single),
+				       offset, size);
 				err = -EINVAL;
 				goto out_vma_unpin;
 			}
@@ -982,7 +1035,7 @@ static int igt_mock_ppgtt_64K(void *arg)
 		}
 	}
 
-	return 0;
+	goto out_vm;
 
 out_vma_unpin:
 	i915_vma_unpin(vma);
@@ -992,7 +1045,10 @@ static int igt_mock_ppgtt_64K(void *arg)
 	i915_gem_object_unlock(obj);
 out_object_put:
 	i915_gem_object_put(obj);
-
+out_vm:
+	i915_vm_put(vm);
+out:
+	fput(file);
 	return err;
 }
 
@@ -1910,8 +1966,6 @@ int i915_gem_huge_page_mock_selftests(void)
 		SUBTEST(igt_mock_exhaust_device_supported_pages),
 		SUBTEST(igt_mock_memory_region_huge_pages),
 		SUBTEST(igt_mock_ppgtt_misaligned_dma),
-		SUBTEST(igt_mock_ppgtt_huge_fill),
-		SUBTEST(igt_mock_ppgtt_64K),
 	};
 	struct drm_i915_private *dev_priv;
 	struct i915_ppgtt *ppgtt;
@@ -1962,6 +2016,8 @@ int i915_gem_huge_page_live_selftests(struct drm_i915_private *i915)
 		SUBTEST(igt_ppgtt_sanity_check),
 		SUBTEST(igt_ppgtt_compact),
 		SUBTEST(igt_ppgtt_mixed),
+		SUBTEST(igt_ppgtt_huge_fill),
+		SUBTEST(igt_ppgtt_64K),
 	};
 
 	if (!HAS_PPGTT(i915)) {

-- 
2.34.1

[PATCH v8 2/2] drm/i915: Use correct huge page manager for MTL

[Andrzej Hajda]

From: Jonathan Cavitt <jonathan.cavitt@intel.com>

MTL currently uses gen8_ppgtt_insert_huge when managing huge pages.
This is because MTL reports as not supporting 64K pages, or more
accurately, the system that reports whether a platform has 64K pages
reports false for MTL.  This is only half correct, as the 64K page support
reporting system only cares about 64K page support for LMEM, which MTL
doesn't have.

MTL should be using xehpsdv_ppgtt_insert_huge.  However, simply changing
over to using that manager doesn't resolve the issue because MTL is
expecting the virtual address space for the page table to be flushed after
initialization, so we must also add a flush statement there.

Signed-off-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Andrzej Hajda <andrzej.hajda@intel.com>
---
 drivers/gpu/drm/i915/gt/gen8_ppgtt.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/gt/gen8_ppgtt.c b/drivers/gpu/drm/i915/gt/gen8_ppgtt.c
index 4daaa6f5566888..9c571185395f49 100644
--- a/drivers/gpu/drm/i915/gt/gen8_ppgtt.c
+++ b/drivers/gpu/drm/i915/gt/gen8_ppgtt.c
@@ -570,6 +570,7 @@ xehpsdv_ppgtt_insert_huge(struct i915_address_space *vm,
 			}
 		} while (rem >= page_size && index < max);
 
+		drm_clflush_virt_range(vaddr, PAGE_SIZE);
 		vma_res->page_sizes_gtt |= page_size;
 	} while (iter->sg && sg_dma_len(iter->sg));
 }
@@ -707,7 +708,7 @@ static void gen8_ppgtt_insert(struct i915_address_space *vm,
 	struct sgt_dma iter = sgt_dma(vma_res);
 
 	if (vma_res->bi.page_sizes.sg > I915_GTT_PAGE_SIZE) {
-		if (HAS_64K_PAGES(vm->i915))
+		if (GRAPHICS_VER_FULL(vm->i915) >= IP_VER(12, 50))
 			xehpsdv_ppgtt_insert_huge(vm, vma_res, &iter, cache_level, flags);
 		else
 			gen8_ppgtt_insert_huge(vm, vma_res, &iter, cache_level, flags);

-- 
2.34.1

Re: [Intel-gfx] [PATCH v8 1/2] drm/i915: Migrate platform-dependent mock hugepage selftests to live

[Andi Shyti]

Hi Andrzej,

On Wed, Apr 26, 2023 at 11:28:48PM +0200, Andrzej Hajda wrote:
> From: Jonathan Cavitt <jonathan.cavitt@intel.com>
> 
> Convert the igt_mock_ppgtt_huge_fill and igt_mock_ppgtt_64K mock selftests
> into live selftests as their requirements have recently become
> platform-dependent. Additionally, apply necessary platform dependency
> checks to these tests.
> 
> v8:
> - handle properly 64K and 2M pages
> v9:
> - do not expect 64K pages if 2M are present
> - fix hex printing
> - obey commit message line limit
> 
> Signed-off-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
> Co-developed-by: Andrzej Hajda <andrzej.hajda@intel.com>
> Signed-off-by: Andrzej Hajda <andrzej.hajda@intel.com>

Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com> 

Andi

Re: [Intel-gfx] [PATCH v8 2/2] drm/i915: Use correct huge page manager for MTL

[Andi Shyti]

Hi Andrzej,

> MTL currently uses gen8_ppgtt_insert_huge when managing huge pages.
> This is because MTL reports as not supporting 64K pages, or more
> accurately, the system that reports whether a platform has 64K pages
> reports false for MTL.  This is only half correct, as the 64K page support
> reporting system only cares about 64K page support for LMEM, which MTL
> doesn't have.
> 
> MTL should be using xehpsdv_ppgtt_insert_huge.  However, simply changing
> over to using that manager doesn't resolve the issue because MTL is
> expecting the virtual address space for the page table to be flushed after
> initialization, so we must also add a flush statement there.
> 
> Signed-off-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
> Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> Signed-off-by: Andrzej Hajda <andrzej.hajda@intel.com>

Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com> 

Andi

Re: [Intel-gfx] [PATCH v8 0/2] drm/i915: Hugepage manager and test for MTL

[Andrzej Hajda]

On 26.04.2023 23:28, Andrzej Hajda wrote:
> This patchset patches sent by Jonathan and Andi, with
> addressed CI failures:
> 1. Fixed checking alignment of 64K pages on both Pre-Gen12 and Gen12.
> 2. Fixed start alignment of 2M pages.
> 
> Regards
> Andrzej
> 
> Jonathan Cavitt (2):
>    drm/i915: Migrate platform-dependent mock hugepage selftests to live
>    drm/i915: Use correct huge page manager for MTL
> 
> .../gpu/drm/i915/gem/selftests/huge_pages.c   | 88 +++++++++++++++----
>   drivers/gpu/drm/i915/gt/gen8_ppgtt.c          |  3 +-
>   2 files changed, 71 insertions(+), 20 deletions(-)
> 
> Cc: intel-gfx@lists.freedesktop.org
> Cc: linux-kernel@vger.kernel.org
> Cc: Jonathan Cavitt <jonathan.cavitt@intel.com>
> Cc: Andrzej Hajda <andrzej.hajda@intel.com>
> Cc: Matthew Auld <matthew.auld@intel.com>

Thx for comments, pushed.

Regards
Andrzej