[PATCH 1/3] drm/i915: Stop marking the unaccessible scratch page as UC

[PATCH 1/3] drm/i915: Stop marking the unaccessible scratch page as UC

[Chris Wilson]

Since by design, if not entirely by practice, nothing is allowed to
access the scratch page we use to background fill the VM, then we do not
need to ensure that it is coherent between the CPU and GPU.
set_pages_uc() does a stop_machine() after changing the PAT, and that
significantly impacts upon context creation throughput.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 3a82c97d5d53..70ef145ecbe5 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -422,16 +422,12 @@ static struct i915_page_scratch *alloc_scratch_page(struct drm_device *dev)
 		return ERR_PTR(ret);
 	}
 
-	set_pages_uc(px_page(sp), 1);
-
 	return sp;
 }
 
 static void free_scratch_page(struct drm_device *dev,
 			      struct i915_page_scratch *sp)
 {
-	set_pages_wb(px_page(sp), 1);
-
 	cleanup_px(dev, sp);
 	kfree(sp);
 }
-- 
2.9.3

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH 2/3] drm/i915: Embed the scratch page struct into each VM

[Chris Wilson]

As the scratch page is no longer shared between all VM, and each has
their own, forgo the small allocation and simply embed the scratch page
struct into the i915_address_space.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 76 +++++++++++++++----------------------
 drivers/gpu/drm/i915/i915_gem_gtt.h |  6 +--
 2 files changed, 32 insertions(+), 50 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 70ef145ecbe5..2e46ff79905f 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -407,29 +407,16 @@ static void fill_page_dma_32(struct drm_device *dev, struct i915_page_dma *p,
 	fill_page_dma(dev, p, v);
 }
 
-static struct i915_page_scratch *alloc_scratch_page(struct drm_device *dev)
+static int
+setup_scratch_page(struct drm_device *dev, struct i915_page_dma *scratch)
 {
-	struct i915_page_scratch *sp;
-	int ret;
-
-	sp = kzalloc(sizeof(*sp), GFP_KERNEL);
-	if (sp == NULL)
-		return ERR_PTR(-ENOMEM);
-
-	ret = __setup_page_dma(dev, px_base(sp), GFP_DMA32 | __GFP_ZERO);
-	if (ret) {
-		kfree(sp);
-		return ERR_PTR(ret);
-	}
-
-	return sp;
+	return __setup_page_dma(dev, scratch, GFP_DMA32 | __GFP_ZERO);
 }
 
-static void free_scratch_page(struct drm_device *dev,
-			      struct i915_page_scratch *sp)
+static void cleanup_scratch_page(struct drm_device *dev,
+				 struct i915_page_dma *scratch)
 {
-	cleanup_px(dev, sp);
-	kfree(sp);
+	cleanup_page_dma(dev, scratch);
 }
 
 static struct i915_page_table *alloc_pt(struct drm_device *dev)
@@ -475,7 +462,7 @@ static void gen8_initialize_pt(struct i915_address_space *vm,
 {
 	gen8_pte_t scratch_pte;
 
-	scratch_pte = gen8_pte_encode(px_dma(vm->scratch_page),
+	scratch_pte = gen8_pte_encode(vm->scratch_page.daddr,
 				      I915_CACHE_LLC, true);
 
 	fill_px(vm->dev, pt, scratch_pte);
@@ -486,9 +473,9 @@ static void gen6_initialize_pt(struct i915_address_space *vm,
 {
 	gen6_pte_t scratch_pte;
 
-	WARN_ON(px_dma(vm->scratch_page) == 0);
+	WARN_ON(vm->scratch_page.daddr == 0);
 
-	scratch_pte = vm->pte_encode(px_dma(vm->scratch_page),
+	scratch_pte = vm->pte_encode(vm->scratch_page.daddr,
 				     I915_CACHE_LLC, true, 0);
 
 	fill32_px(vm->dev, pt, scratch_pte);
@@ -772,7 +759,7 @@ static void gen8_ppgtt_clear_range(struct i915_address_space *vm,
 				   bool use_scratch)
 {
 	struct i915_hw_ppgtt *ppgtt = i915_vm_to_ppgtt(vm);
-	gen8_pte_t scratch_pte = gen8_pte_encode(px_dma(vm->scratch_page),
+	gen8_pte_t scratch_pte = gen8_pte_encode(vm->scratch_page.daddr,
 						 I915_CACHE_LLC, use_scratch);
 
 	if (!USES_FULL_48BIT_PPGTT(vm->dev)) {
@@ -878,9 +865,9 @@ static int gen8_init_scratch(struct i915_address_space *vm)
 	struct drm_device *dev = vm->dev;
 	int ret;
 
-	vm->scratch_page = alloc_scratch_page(dev);
-	if (IS_ERR(vm->scratch_page))
-		return PTR_ERR(vm->scratch_page);
+	ret = setup_scratch_page(dev, &vm->scratch_page);
+	if (ret)
+		return ret;
 
 	vm->scratch_pt = alloc_pt(dev);
 	if (IS_ERR(vm->scratch_pt)) {
@@ -914,7 +901,7 @@ free_pd:
 free_pt:
 	free_pt(dev, vm->scratch_pt);
 free_scratch_page:
-	free_scratch_page(dev, vm->scratch_page);
+	cleanup_scratch_page(dev, &vm->scratch_page);
 
 	return ret;
 }
@@ -958,7 +945,7 @@ static void gen8_free_scratch(struct i915_address_space *vm)
 		free_pdp(dev, vm->scratch_pdp);
 	free_pd(dev, vm->scratch_pd);
 	free_pt(dev, vm->scratch_pt);
-	free_scratch_page(dev, vm->scratch_page);
+	cleanup_scratch_page(dev, &vm->scratch_page);
 }
 
 static void gen8_ppgtt_cleanup_3lvl(struct drm_device *dev,
@@ -1455,7 +1442,7 @@ static void gen8_dump_ppgtt(struct i915_hw_ppgtt *ppgtt, struct seq_file *m)
 	struct i915_address_space *vm = &ppgtt->base;
 	uint64_t start = ppgtt->base.start;
 	uint64_t length = ppgtt->base.total;
-	gen8_pte_t scratch_pte = gen8_pte_encode(px_dma(vm->scratch_page),
+	gen8_pte_t scratch_pte = gen8_pte_encode(vm->scratch_page.daddr,
 						 I915_CACHE_LLC, true);
 
 	if (!USES_FULL_48BIT_PPGTT(vm->dev)) {
@@ -1572,7 +1559,7 @@ static void gen6_dump_ppgtt(struct i915_hw_ppgtt *ppgtt, struct seq_file *m)
 	uint32_t  pte, pde;
 	uint32_t start = ppgtt->base.start, length = ppgtt->base.total;
 
-	scratch_pte = vm->pte_encode(px_dma(vm->scratch_page),
+	scratch_pte = vm->pte_encode(vm->scratch_page.daddr,
 				     I915_CACHE_LLC, true, 0);
 
 	gen6_for_each_pde(unused, &ppgtt->pd, start, length, pde) {
@@ -1797,7 +1784,7 @@ static void gen6_ppgtt_clear_range(struct i915_address_space *vm,
 	unsigned first_pte = first_entry % GEN6_PTES;
 	unsigned last_pte, i;
 
-	scratch_pte = vm->pte_encode(px_dma(vm->scratch_page),
+	scratch_pte = vm->pte_encode(vm->scratch_page.daddr,
 				     I915_CACHE_LLC, true, 0);
 
 	while (num_entries) {
@@ -1943,14 +1930,15 @@ unwind_out:
 static int gen6_init_scratch(struct i915_address_space *vm)
 {
 	struct drm_device *dev = vm->dev;
+	int ret;
 
-	vm->scratch_page = alloc_scratch_page(dev);
-	if (IS_ERR(vm->scratch_page))
-		return PTR_ERR(vm->scratch_page);
+	ret = setup_scratch_page(dev, &vm->scratch_page);
+	if (ret)
+		return ret;
 
 	vm->scratch_pt = alloc_pt(dev);
 	if (IS_ERR(vm->scratch_pt)) {
-		free_scratch_page(dev, vm->scratch_page);
+		cleanup_scratch_page(dev, &vm->scratch_page);
 		return PTR_ERR(vm->scratch_pt);
 	}
 
@@ -1964,7 +1952,7 @@ static void gen6_free_scratch(struct i915_address_space *vm)
 	struct drm_device *dev = vm->dev;
 
 	free_pt(dev, vm->scratch_pt);
-	free_scratch_page(dev, vm->scratch_page);
+	cleanup_scratch_page(dev, &vm->scratch_page);
 }
 
 static void gen6_ppgtt_cleanup(struct i915_address_space *vm)
@@ -2505,7 +2493,7 @@ static void gen8_ggtt_clear_range(struct i915_address_space *vm,
 		 first_entry, num_entries, max_entries))
 		num_entries = max_entries;
 
-	scratch_pte = gen8_pte_encode(px_dma(vm->scratch_page),
+	scratch_pte = gen8_pte_encode(vm->scratch_page.daddr,
 				      I915_CACHE_LLC,
 				      use_scratch);
 	for (i = 0; i < num_entries; i++)
@@ -2537,7 +2525,7 @@ static void gen6_ggtt_clear_range(struct i915_address_space *vm,
 		 first_entry, num_entries, max_entries))
 		num_entries = max_entries;
 
-	scratch_pte = vm->pte_encode(px_dma(vm->scratch_page),
+	scratch_pte = vm->pte_encode(vm->scratch_page.daddr,
 				     I915_CACHE_LLC, use_scratch, 0);
 
 	for (i = 0; i < num_entries; i++)
@@ -2907,8 +2895,8 @@ static size_t gen9_get_stolen_size(u16 gen9_gmch_ctl)
 static int ggtt_probe_common(struct i915_ggtt *ggtt, u64 size)
 {
 	struct pci_dev *pdev = ggtt->base.dev->pdev;
-	struct i915_page_scratch *scratch_page;
 	phys_addr_t phys_addr;
+	int ret;
 
 	/* For Modern GENs the PTEs and register space are split in the BAR */
 	phys_addr = pci_resource_start(pdev, 0) + pci_resource_len(pdev, 0) / 2;
@@ -2929,16 +2917,14 @@ static int ggtt_probe_common(struct i915_ggtt *ggtt, u64 size)
 		return -ENOMEM;
 	}
 
-	scratch_page = alloc_scratch_page(ggtt->base.dev);
-	if (IS_ERR(scratch_page)) {
+	ret = setup_scratch_page(ggtt->base.dev, &ggtt->base.scratch_page);
+	if (ret) {
 		DRM_ERROR("Scratch setup failed\n");
 		/* iounmap will also get called at remove, but meh */
 		iounmap(ggtt->gsm);
-		return PTR_ERR(scratch_page);
+		return ret;
 	}
 
-	ggtt->base.scratch_page = scratch_page;
-
 	return 0;
 }
 
@@ -3020,7 +3006,7 @@ static void gen6_gmch_remove(struct i915_address_space *vm)
 	struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
 
 	iounmap(ggtt->gsm);
-	free_scratch_page(vm->dev, vm->scratch_page);
+	cleanup_scratch_page(vm->dev, &vm->scratch_page);
 }
 
 static int gen8_gmch_probe(struct i915_ggtt *ggtt)
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index 026cf671ac01..a90999fe2e57 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -312,10 +312,6 @@ struct i915_page_dma {
 #define px_page(px) (px_base(px)->page)
 #define px_dma(px) (px_base(px)->daddr)
 
-struct i915_page_scratch {
-	struct i915_page_dma base;
-};
-
 struct i915_page_table {
 	struct i915_page_dma base;
 
@@ -361,7 +357,7 @@ struct i915_address_space {
 
 	bool closed;
 
-	struct i915_page_scratch *scratch_page;
+	struct i915_page_dma scratch_page;
 	struct i915_page_table *scratch_pt;
 	struct i915_page_directory *scratch_pd;
 	struct i915_page_directory_pointer *scratch_pdp; /* GEN8+ & 48b PPGTT */
-- 
2.9.3

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH 3/3] drm/i915: Allow DMA pagetables to use highmem

[Chris Wilson]

As we never need to directly access the pages we allocate for scratch and
the pagetables, and always remap them into the GTT through the dma
remapper, we do not need to limit the allocations to lowmem i.e. we can
pass in the __GFP_HIGHMEM flag to the page allocation.

For backwards compatibility, e.g. certain old GPUs not liking highmem
for certain functions that may be accidentally mapped to the scratch
page by userspace, keep the GMCH probe as only allocating form DMA32.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 18 ++++++++++++------
 1 file changed, 12 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 2e46ff79905f..435c3fc73e18 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -32,6 +32,8 @@
 #include "i915_trace.h"
 #include "intel_drv.h"
 
+#define I915_GFP_DMA (GFP_KERNEL | __GFP_HIGHMEM)
+
 /**
  * DOC: Global GTT views
  *
@@ -345,7 +347,7 @@ static int __setup_page_dma(struct drm_device *dev,
 
 static int setup_page_dma(struct drm_device *dev, struct i915_page_dma *p)
 {
-	return __setup_page_dma(dev, p, GFP_KERNEL);
+	return __setup_page_dma(dev, p, I915_GFP_DMA);
 }
 
 static void cleanup_page_dma(struct drm_device *dev, struct i915_page_dma *p)
@@ -408,9 +410,11 @@ static void fill_page_dma_32(struct drm_device *dev, struct i915_page_dma *p,
 }
 
 static int
-setup_scratch_page(struct drm_device *dev, struct i915_page_dma *scratch)
+setup_scratch_page(struct drm_device *dev,
+		   struct i915_page_dma *scratch,
+		   gfp_t gfp)
 {
-	return __setup_page_dma(dev, scratch, GFP_DMA32 | __GFP_ZERO);
+	return __setup_page_dma(dev, scratch, gfp | __GFP_ZERO);
 }
 
 static void cleanup_scratch_page(struct drm_device *dev,
@@ -865,7 +869,7 @@ static int gen8_init_scratch(struct i915_address_space *vm)
 	struct drm_device *dev = vm->dev;
 	int ret;
 
-	ret = setup_scratch_page(dev, &vm->scratch_page);
+	ret = setup_scratch_page(dev, &vm->scratch_page, I915_GFP_DMA);
 	if (ret)
 		return ret;
 
@@ -1932,7 +1936,7 @@ static int gen6_init_scratch(struct i915_address_space *vm)
 	struct drm_device *dev = vm->dev;
 	int ret;
 
-	ret = setup_scratch_page(dev, &vm->scratch_page);
+	ret = setup_scratch_page(dev, &vm->scratch_page, I915_GFP_DMA);
 	if (ret)
 		return ret;
 
@@ -2917,7 +2921,9 @@ static int ggtt_probe_common(struct i915_ggtt *ggtt, u64 size)
 		return -ENOMEM;
 	}
 
-	ret = setup_scratch_page(ggtt->base.dev, &ggtt->base.scratch_page);
+	ret = setup_scratch_page(ggtt->base.dev,
+				 &ggtt->base.scratch_page,
+				 GFP_DMA32);
 	if (ret) {
 		DRM_ERROR("Scratch setup failed\n");
 		/* iounmap will also get called at remove, but meh */
-- 
2.9.3

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

✗ Ro.CI.BAT: failure for series starting with [1/3] drm/i915: Stop marking the unaccessible scratch page as UC

[Patchwork]

== Series Details ==

Series: series starting with [1/3] drm/i915: Stop marking the unaccessible scratch page as UC
URL   : https://patchwork.freedesktop.org/series/11398/
State : failure

== Summary ==

Series 11398v1 Series without cover letter
http://patchwork.freedesktop.org/api/1.0/series/11398/revisions/1/mbox

Test kms_cursor_legacy:
        Subgroup basic-flip-vs-cursor-legacy:
                pass       -> FAIL       (ro-byt-n2820)
                pass       -> FAIL       (ro-skl3-i5-6260u)
                pass       -> DMESG-FAIL (fi-skl-i7-6700k)

fi-hsw-i7-4770k  total:244  pass:222  dwarn:0   dfail:0   fail:0   skip:22 
fi-kbl-qkkr      total:244  pass:185  dwarn:29  dfail:0   fail:3   skip:27 
fi-skl-i7-6700k  total:244  pass:208  dwarn:4   dfail:2   fail:2   skip:28 
fi-snb-i7-2600   total:244  pass:202  dwarn:0   dfail:0   fail:0   skip:42 
ro-bdw-i5-5250u  total:240  pass:219  dwarn:4   dfail:0   fail:0   skip:17 
ro-bdw-i7-5557U  total:240  pass:217  dwarn:5   dfail:0   fail:0   skip:18 
ro-bdw-i7-5600u  total:240  pass:206  dwarn:0   dfail:0   fail:2   skip:32 
ro-bsw-n3050     total:240  pass:194  dwarn:0   dfail:0   fail:4   skip:42 
ro-byt-n2820     total:240  pass:196  dwarn:0   dfail:0   fail:4   skip:40 
ro-hsw-i3-4010u  total:240  pass:213  dwarn:0   dfail:0   fail:1   skip:26 
ro-hsw-i7-4770r  total:240  pass:185  dwarn:0   dfail:0   fail:0   skip:55 
ro-ilk1-i5-650   total:235  pass:174  dwarn:0   dfail:0   fail:2   skip:59 
ro-ivb-i7-3770   total:240  pass:204  dwarn:0   dfail:0   fail:1   skip:35 
ro-ivb2-i7-3770  total:240  pass:208  dwarn:0   dfail:0   fail:1   skip:31 
ro-skl3-i5-6260u total:240  pass:222  dwarn:0   dfail:0   fail:4   skip:14 

Results at /archive/results/CI_IGT_test/RO_Patchwork_1959/

6e092ef drm-intel-nightly: 2016y-08m-22d-07h-37m-35s UTC integration manifest
09de690 drm/i915: Allow DMA pagetables to use highmem
9f37ec8 drm/i915: Embed the scratch page struct into each VM
89baee41 drm/i915: Stop marking the unaccessible scratch page as UC

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH 1/3] drm/i915: Stop marking the unaccessible scratch page as UC

[Joonas Lahtinen]

On ma, 2016-08-22 at 08:44 +0100, Chris Wilson wrote:
> Since by design, if not entirely by practice, nothing is allowed to
> access the scratch page we use to background fill the VM, then we do not
> need to ensure that it is coherent between the CPU and GPU.
> set_pages_uc() does a stop_machine() after changing the PAT, and that
> significantly impacts upon context creation throughput.
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH 2/3] drm/i915: Embed the scratch page struct into each VM

[Joonas Lahtinen]

On ma, 2016-08-22 at 08:44 +0100, Chris Wilson wrote:
> As the scratch page is no longer shared between all VM, and each has
> their own, forgo the small allocation and simply embed the scratch page
> struct into the i915_address_space.
> 
> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

A fairly mechanical change.

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

For dropping the use of px_dma & friend macros, maybe an A-b from Mika.

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH 2/3] drm/i915: Embed the scratch page struct into each VM

[Mika Kuoppala]

Joonas Lahtinen <joonas.lahtinen@linux.intel.com> writes:

> On ma, 2016-08-22 at 08:44 +0100, Chris Wilson wrote:
>> As the scratch page is no longer shared between all VM, and each has
>> their own, forgo the small allocation and simply embed the scratch page
>> struct into the i915_address_space.
>> 
>> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
>
> A fairly mechanical change.
>
> Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
>
> For dropping the use of px_dma & friend macros, maybe an A-b from
> Mika.

Sure thing.

Acked-by: Mika Kuoppala <mika.kuoppala@intel.com>

>
> Regards, Joonas
> -- 
> Joonas Lahtinen
> Open Source Technology Center
> Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [PATCH 3/3] drm/i915: Allow DMA pagetables to use highmem

[Joonas Lahtinen]

On ma, 2016-08-22 at 08:44 +0100, Chris Wilson wrote:
> As we never need to directly access the pages we allocate for scratch and
> the pagetables, and always remap them into the GTT through the dma
> remapper, we do not need to limit the allocations to lowmem i.e. we can
> pass in the __GFP_HIGHMEM flag to the page allocation.
> 
> For backwards compatibility, e.g. certain old GPUs not liking highmem
> for certain functions that may be accidentally mapped to the scratch
> page by userspace, keep the GMCH probe as only allocating form DMA32.

s/form/from/

With that,

Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Regards, Joonas
-- 
Joonas Lahtinen
Open Source Technology Center
Intel Corporation
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx