Skip to contentSkip to conversation

Add Pooled EU support to BXT

PATCH Applied Arun Siluvery 4 participants 13 messages Jul 17, 2015 – Jul 21, 2015

Subsystems (3)

v1 of 3 Applied Latest is v3 →

lore
Share & tools
lore
#0 lore raw

cover Add Pooled EU support to BXT

This patches adds support to enable Pooled EU feature in BXT.

This series has a dependency with Patch1 from Mika hence sending it as part of the series (already reviewed).

Patch2 - adds a framework to extend the golden context batch through which we can add Gen based commands to enable specific features, in this case it is used to enabled Pooled EU (Patch3)

This is on the idea suggested by Chris Wilson to send two batches instead of the previous approach of patching the binary data which is unnecessarily complicated (http://www.spinics.net/lists/intel-gfx/msg71498.html).

Patch4 - option for the userspace to query it's availability.

Arun Siluvery (3):
drm/i915: Add provision to extend Golden context batch drm/i915:bxt: Enable Pooled EU support drm/i915/bxt: Add get_param to query Pooled EU availability

Mika Kuoppala (1):
drm/i915: Do kunmap if renderstate parsing fails

 drivers/gpu/drm/i915/i915_dma.c              |  3 ++
 drivers/gpu/drm/i915/i915_drv.c              |  1 +
 drivers/gpu/drm/i915/i915_drv.h              |  5 ++-
 drivers/gpu/drm/i915/i915_gem_render_state.c | 52 ++++++++++++++++++++++++++--
 drivers/gpu/drm/i915/i915_gem_render_state.h |  2 ++
 drivers/gpu/drm/i915/i915_reg.h              |  2 ++
 drivers/gpu/drm/i915/intel_lrc.c             |  6 ++++
 include/uapi/drm/i915_drm.h                  |  1 +
 8 files changed, 69 insertions(+), 3 deletions(-)
Signature · 6 lines
1.9.1

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

PATCH 1/4 drm/i915: Do kunmap if renderstate parsing fails

From: Mika Kuoppala <[email protected]>

Kunmap the renderstate page on error path.

---
 drivers/gpu/drm/i915/i915_gem_render_state.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/gpu/drm/i915/i915_gem_render_state.c +8 -2
index a0201fc..b6492fe 100644
@@ -96,8 +96,10 @@ static int render_state_setup(struct render_state *so)
 			s = lower_32_bits(r);
 			if (so->gen >= 8) {
 				if (i + 1 >= rodata->batch_items ||
-				    rodata->batch[i + 1] != 0)
-					return -EINVAL;
+				    rodata->batch[i + 1] != 0) {
+					ret = -EINVAL;
+					goto err_out;
+				}
 
 				d[i++] = s;
 				s = upper_32_bits(r);
@@ -120,6 +122,10 @@ static int render_state_setup(struct render_state *so)
 	}
 
 	return 0;
+
+err_out:
+	kunmap(page);
+	return ret;
 }
 
 void i915_gem_render_state_fini(struct render_state *so)
Signature · 6 lines
1.9.1

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

Re: [PATCH v1 1/4] drm/i915: Do kunmap if renderstate parsing fails

Quoted 6 lines · Arun Siluvery

From: Mika Kuoppala <[email protected]>

Kunmap the renderstate page on error path.

Queued for -next, thanks for the patch.
-Daniel

Quoted 39 lines
---
 drivers/gpu/drm/i915/i915_gem_render_state.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
drivers/gpu/drm/i915/i915_gem_render_state.c +8 -2
index a0201fc..b6492fe 100644
@@ -96,8 +96,10 @@ static int render_state_setup(struct render_state *so)
 			s = lower_32_bits(r);
 			if (so->gen >= 8) {
 				if (i + 1 >= rodata->batch_items ||
-				    rodata->batch[i + 1] != 0)
-					return -EINVAL;
+				    rodata->batch[i + 1] != 0) {
+					ret = -EINVAL;
+					goto err_out;
+				}
 
 				d[i++] = s;
 				s = upper_32_bits(r);
@@ -120,6 +122,10 @@ static int render_state_setup(struct render_state *so)
 	}
 
 	return 0;
+
+err_out:
+	kunmap(page);
+	return ret;
 }
 
 void i915_gem_render_state_fini(struct render_state *so)
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#3 lore raw

PATCH 2/4 drm/i915: Add provision to extend Golden context batch

In reply to Arun Siluvery

The Golden batch carries 3D state at the beginning so that HW starts with a known state. It is carried as a binary blob which is auto-generated from source. The idea was it would be easier to maintain and keep the complexity out of the kernel which makes sense as we don't really touch it. However if you really need to update it then you need to update generator source and keep the binary blob in sync with it.

There is a need to patch this in bxt to send one additional command to enable a feature. A solution was to patch the binary data with some additional data structures (included as part of auto-generator source) but it was unnecessarily complicated.

Chris suggested the idea of having a secondary batch and execute two batch buffers. It has clear advantages as we needn't touch the base golden batch, can customize secondary/auxiliary batch depending on Gen and can be carried in the driver with no dependencies.

This patch adds support for this auxiliary batch which is inserted at the end of golden batch and is completely independent from it. Thanks to Mika for the preliminary review.

Cc Mika Kuoppala <[email protected]>
Cc Chris Wilson <[email protected]>
Cc Armin Reese <[email protected]>
---
 drivers/gpu/drm/i915/i915_gem_render_state.c | 27 +++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_gem_render_state.h |  2 ++
 drivers/gpu/drm/i915/intel_lrc.c             |  6 ++++++
 3 files changed, 35 insertions(+)
drivers/gpu/drm/i915/i915_gem_render_state.c +27 -0
index b6492fe..b86e382 100644
@@ -73,6 +73,15 @@ free_gem:
 	return ret;
 }
 
+#define OUT_BATCH(batch, i, val)				\
+	do {							\
+		if (WARN_ON((i) >= PAGE_SIZE / sizeof(u32))) {	\
+			ret = -ENOSPC;				\
+			goto err_out;				\
+		}						\
+		(batch)[(i)++] = (val);				\
+	} while(0)
+
 static int render_state_setup(struct render_state *so)
 {
 	const struct intel_renderstate_rodata *rodata = so->rodata;
@@ -110,6 +119,15 @@ static int render_state_setup(struct render_state *so)
 
 		d[i++] = s;
 	}
+
+	while (i % CACHELINE_DWORDS)
+		OUT_BATCH(d, i, MI_NOOP);
+
+	so->aux_batch_offset = i * sizeof(u32);
+
+	OUT_BATCH(d, i, MI_BATCH_BUFFER_END);
+	so->aux_batch_size = (i * sizeof(u32)) - so->aux_batch_offset;
+
 	kunmap(page);
 
 	ret = i915_gem_object_set_to_gtt_domain(so->obj, false);
@@ -128,6 +146,8 @@ err_out:
 	return ret;
 }
 
+#undef OUT_BATCH
+
 void i915_gem_render_state_fini(struct render_state *so)
 {
 	i915_gem_object_ggtt_unpin(so->obj);
@@ -176,6 +196,13 @@ int i915_gem_render_state_init(struct drm_i915_gem_request *req)
 	if (ret)
 		goto out;
 
+	ret = req->ring->dispatch_execbuffer(req,
+					     (so.ggtt_offset + so.aux_batch_offset),
+					     so.aux_batch_size,
+					     I915_DISPATCH_SECURE);
+	if (ret)
+		goto out;
+
 	i915_vma_move_to_active(i915_gem_obj_to_ggtt(so.obj), req);
 
 out:
drivers/gpu/drm/i915/i915_gem_render_state.h +2 -0
index 7aa7372..79de101 100644
@@ -37,6 +37,8 @@ struct render_state {
 	struct drm_i915_gem_object *obj;
 	u64 ggtt_offset;
 	int gen;
+	u32 aux_batch_size;
+	u64 aux_batch_offset;
 };
 
 int i915_gem_render_state_init(struct drm_i915_gem_request *req);
drivers/gpu/drm/i915/intel_lrc.c +6 -0
index adb386d..5e4771e 100644
@@ -1661,6 +1661,12 @@ static int intel_lr_context_render_state_init(struct drm_i915_gem_request *req)
 	if (ret)
 		goto out;
 
+	ret = req->ring->emit_bb_start(req,
+				       (so.ggtt_offset + so.aux_batch_offset),
+				       I915_DISPATCH_SECURE);
+	if (ret)
+		goto out;
+
 	i915_vma_move_to_active(i915_gem_obj_to_ggtt(so.obj), req);
 
 out:
Signature · 6 lines
1.9.1

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

Re: [PATCH v1 2/4] drm/i915: Add provision to extend Golden context batch

Quoted 36 lines · Arun Siluvery

The Golden batch carries 3D state at the beginning so that HW starts with a known state. It is carried as a binary blob which is auto-generated from source. The idea was it would be easier to maintain and keep the complexity out of the kernel which makes sense as we don't really touch it. However if you really need to update it then you need to update generator source and keep the binary blob in sync with it.

There is a need to patch this in bxt to send one additional command to enable a feature. A solution was to patch the binary data with some additional data structures (included as part of auto-generator source) but it was unnecessarily complicated.

Chris suggested the idea of having a secondary batch and execute two batch buffers. It has clear advantages as we needn't touch the base golden batch, can customize secondary/auxiliary batch depending on Gen and can be carried in the driver with no dependencies.

This patch adds support for this auxiliary batch which is inserted at the end of golden batch and is completely independent from it. Thanks to Mika for the preliminary review.

Cc Mika Kuoppala <[email protected]>
Cc Chris Wilson <[email protected]>
Cc Armin Reese <[email protected]>
---
 drivers/gpu/drm/i915/i915_gem_render_state.c | 27 +++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_gem_render_state.h |  2 ++
 drivers/gpu/drm/i915/intel_lrc.c             |  6 ++++++
 3 files changed, 35 insertions(+)
drivers/gpu/drm/i915/i915_gem_render_state.c +0 -0
index b6492fe..b86e382 100644

drivers/gpu/drm/i915/i915_gem_render_state.c free_gem: · line 78

Quoted 6 lines
drivers/gpu/drm/i915/i915_gem_render_state.c +3 -0
@@ -73,6 +73,15 @@ free_gem:
 	return ret;
 }
 
+#define OUT_BATCH(batch, i, val)				\
+	do {							\
+		if (WARN_ON((i) >= PAGE_SIZE / sizeof(u32))) {	\

We have to be slightly more careful here, as we don't have the full page available since we put render state into the high arena of the golden bb. Something like WARN_ON(i > PAGE/sizeof(u32) || (batch)[i]) should suffice.

drivers/gpu/drm/i915/i915_gem_render_state.c static int render_state_setup(struct render_state *so) · line 129

Quoted 12 lines
drivers/gpu/drm/i915/i915_gem_render_state.c +8 -0
@@ -110,6 +119,15 @@ static int render_state_setup(struct render_state *so)
 
 		d[i++] = s;
 	}
+
+	while (i % CACHELINE_DWORDS)
+		OUT_BATCH(d, i, MI_NOOP);
+
+	so->aux_batch_offset = i * sizeof(u32);
+
+	OUT_BATCH(d, i, MI_BATCH_BUFFER_END);
+	so->aux_batch_size = (i * sizeof(u32)) - so->aux_batch_offset;

Strictly, and if we are passing the batch length we are being strictly conformant, then the aux_batch_size must be a multiple of 8.

Quoted 14 lines
drivers/gpu/drm/i915/i915_gem_render_state.c +3 -0
+
 	kunmap(page);
 
 	ret = i915_gem_object_set_to_gtt_domain(so->obj, false);
@@ -128,6 +146,8 @@ err_out:
 	return ret;
 }
 
+#undef OUT_BATCH
+
 void i915_gem_render_state_fini(struct render_state *so)
 {
 	i915_gem_object_ggtt_unpin(so->obj);

drivers/gpu/drm/i915/i915_gem_render_state.c int i915_gem_render_state_init(struct drm_i915_gem_request *req) · line 198

Quoted 3 lines
drivers/gpu/drm/i915/i915_gem_render_state.c +0 -0
@@ -176,6 +196,13 @@ int i915_gem_render_state_init(struct drm_i915_gem_request *req)
 	if (ret)
 		goto out;
 

Then we need only execute this BB if so.aux_batch_size > 8

drivers/gpu/drm/i915/i915_gem_render_state.c int i915_gem_render_state_init(struct drm_i915_gem_request *req) · line 206

Quoted 8 lines
drivers/gpu/drm/i915/i915_gem_render_state.c +7 -0
+	ret = req->ring->dispatch_execbuffer(req,
+					     (so.ggtt_offset + so.aux_batch_offset),
+					     so.aux_batch_size,
+					     I915_DISPATCH_SECURE);
+	if (ret)
+		goto out;
+
 	i915_vma_move_to_active(i915_gem_obj_to_ggtt(so.obj), req);

-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#5 lore raw

Re: [PATCH v1 2/4] drm/i915: Add provision to extend Golden context batch

Quoted 40 lines · Chris Wilson
Quoted 39 lines · Arun Siluvery

The Golden batch carries 3D state at the beginning so that HW starts with a known state. It is carried as a binary blob which is auto-generated from source. The idea was it would be easier to maintain and keep the complexity out of the kernel which makes sense as we don't really touch it. However if you really need to update it then you need to update generator source and keep the binary blob in sync with it.

There is a need to patch this in bxt to send one additional command to enable a feature. A solution was to patch the binary data with some additional data structures (included as part of auto-generator source) but it was unnecessarily complicated.

Chris suggested the idea of having a secondary batch and execute two batch buffers. It has clear advantages as we needn't touch the base golden batch, can customize secondary/auxiliary batch depending on Gen and can be carried in the driver with no dependencies.

This patch adds support for this auxiliary batch which is inserted at the end of golden batch and is completely independent from it. Thanks to Mika for the preliminary review.

Cc Mika Kuoppala <[email protected]>
Cc Chris Wilson <[email protected]>
Cc Armin Reese <[email protected]>
---
 drivers/gpu/drm/i915/i915_gem_render_state.c | 27 +++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_gem_render_state.h |  2 ++
 drivers/gpu/drm/i915/intel_lrc.c             |  6 ++++++
 3 files changed, 35 insertions(+)
drivers/gpu/drm/i915/i915_gem_render_state.c +0 -0
index b6492fe..b86e382 100644
@@ -73,6 +73,15 @@ free_gem:
 	return ret;
 }
 
Quoted 9 lines

drivers/gpu/drm/i915/i915_gem_render_state.c free_gem: · line 78

Quoted 3 lines
drivers/gpu/drm/i915/i915_gem_render_state.c +3 -0
+#define OUT_BATCH(batch, i, val)				\
+	do {							\
+		if (WARN_ON((i) >= PAGE_SIZE / sizeof(u32))) {	\

We have to be slightly more careful here, as we don't have the full page available since we put render state into the high arena of the golden bb. Something like WARN_ON(i > PAGE/sizeof(u32) || (batch)[i]) should suffice.

Null state gen makes the final batch with two passes. First it builds command and state separately. And when size of both are know, it compacts by relocating the state right after the commands (+some alignment).

So we should have the rest of the page usable for auxillary commands here as we have already copied the state part also.

-Mika

Quoted 35 lines

drivers/gpu/drm/i915/i915_gem_render_state.c static int render_state_setup(struct render_state *so) · line 129

Quoted 12 lines
drivers/gpu/drm/i915/i915_gem_render_state.c +8 -0
@@ -110,6 +119,15 @@ static int render_state_setup(struct render_state *so)
 
 		d[i++] = s;
 	}
+
+	while (i % CACHELINE_DWORDS)
+		OUT_BATCH(d, i, MI_NOOP);
+
+	so->aux_batch_offset = i * sizeof(u32);
+
+	OUT_BATCH(d, i, MI_BATCH_BUFFER_END);
+	so->aux_batch_size = (i * sizeof(u32)) - so->aux_batch_offset;

Strictly, and if we are passing the batch length we are being strictly conformant, then the aux_batch_size must be a multiple of 8.

Quoted 14 lines
drivers/gpu/drm/i915/i915_gem_render_state.c +3 -0
+
 	kunmap(page);
 
 	ret = i915_gem_object_set_to_gtt_domain(so->obj, false);
@@ -128,6 +146,8 @@ err_out:
 	return ret;
 }
 
+#undef OUT_BATCH
+
 void i915_gem_render_state_fini(struct render_state *so)
 {
 	i915_gem_object_ggtt_unpin(so->obj);

drivers/gpu/drm/i915/i915_gem_render_state.c int i915_gem_render_state_init(struct drm_i915_gem_request *req) · line 198

Quoted 3 lines
drivers/gpu/drm/i915/i915_gem_render_state.c +0 -0
@@ -176,6 +196,13 @@ int i915_gem_render_state_init(struct drm_i915_gem_request *req)
 	if (ret)
 		goto out;
 

Then we need only execute this BB if so.aux_batch_size > 8

Quoted 16 lines

drivers/gpu/drm/i915/i915_gem_render_state.c int i915_gem_render_state_init(struct drm_i915_gem_request *req) · line 206

Quoted 8 lines
drivers/gpu/drm/i915/i915_gem_render_state.c +7 -0
+	ret = req->ring->dispatch_execbuffer(req,
+					     (so.ggtt_offset + so.aux_batch_offset),
+					     so.aux_batch_size,
+					     I915_DISPATCH_SECURE);
+	if (ret)
+		goto out;
+
 	i915_vma_move_to_active(i915_gem_obj_to_ggtt(so.obj), req);

-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

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

#6 lore raw

Re: [PATCH v1 2/4] drm/i915: Add provision to extend Golden context batch

Quoted 52 lines · Mika Kuoppala
Quoted 40 lines · Chris Wilson
Quoted 39 lines · Arun Siluvery

The Golden batch carries 3D state at the beginning so that HW starts with a known state. It is carried as a binary blob which is auto-generated from source. The idea was it would be easier to maintain and keep the complexity out of the kernel which makes sense as we don't really touch it. However if you really need to update it then you need to update generator source and keep the binary blob in sync with it.

There is a need to patch this in bxt to send one additional command to enable a feature. A solution was to patch the binary data with some additional data structures (included as part of auto-generator source) but it was unnecessarily complicated.

Chris suggested the idea of having a secondary batch and execute two batch buffers. It has clear advantages as we needn't touch the base golden batch, can customize secondary/auxiliary batch depending on Gen and can be carried in the driver with no dependencies.

This patch adds support for this auxiliary batch which is inserted at the end of golden batch and is completely independent from it. Thanks to Mika for the preliminary review.

Cc Mika Kuoppala <[email protected]>
Cc Chris Wilson <[email protected]>
Cc Armin Reese <[email protected]>
---
 drivers/gpu/drm/i915/i915_gem_render_state.c | 27 +++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_gem_render_state.h |  2 ++
 drivers/gpu/drm/i915/intel_lrc.c             |  6 ++++++
 3 files changed, 35 insertions(+)
drivers/gpu/drm/i915/i915_gem_render_state.c +0 -0
index b6492fe..b86e382 100644
@@ -73,6 +73,15 @@ free_gem:
 	return ret;
 }
 
Quoted 9 lines

drivers/gpu/drm/i915/i915_gem_render_state.c free_gem: · line 78

Quoted 3 lines
drivers/gpu/drm/i915/i915_gem_render_state.c +3 -0
+#define OUT_BATCH(batch, i, val)				\
+	do {							\
+		if (WARN_ON((i) >= PAGE_SIZE / sizeof(u32))) {	\

We have to be slightly more careful here, as we don't have the full page available since we put render state into the high arena of the golden bb. Something like WARN_ON(i > PAGE/sizeof(u32) || (batch)[i]) should suffice.

Quoted 8 lines

Null state gen makes the final batch with two passes. First it builds command and state separately. And when size of both are know, it compacts by relocating the state right after the commands (+some alignment).

So we should have the rest of the page usable for auxillary commands here as we have already copied the state part also.

Ta. Maybe add some words of enlightenment here for future me as well? Also we will need to document that the kernel then relies on the packing to add extra commands after the batch to the null state generator. -Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#7 lore raw

PATCH 3/4 drm/i915:bxt: Enable Pooled EU support

In reply to Arun Siluvery

This mode allows to assign EUs to pools.
The command to enable this mode is sent in auxiliary golden context batch as this is only issued once with each context initialization. Thanks to Mika for the preliminary review.

Cc Mika Kuoppala <[email protected]>
Cc Chris Wilson <[email protected]>
Cc Armin Reese <[email protected]>
---
 drivers/gpu/drm/i915/i915_gem_render_state.c | 15 +++++++++++++++
 drivers/gpu/drm/i915/i915_reg.h              |  2 ++
 2 files changed, 17 insertions(+)
drivers/gpu/drm/i915/i915_gem_render_state.c +15 -0
index b86e382..a41a1b6 100644
@@ -84,6 +84,7 @@ free_gem:
 
 static int render_state_setup(struct render_state *so)
 {
+	struct drm_device *dev = so->obj->base.dev;
 	const struct intel_renderstate_rodata *rodata = so->rodata;
 	unsigned int i = 0, reloc_index = 0;
 	struct page *page;
@@ -125,6 +126,20 @@ static int render_state_setup(struct render_state *so)
 
 	so->aux_batch_offset = i * sizeof(u32);
 
+	if (IS_BROXTON(dev)) {
+		u32 pool_config = 0;
+		struct drm_i915_private *dev_priv = to_i915(dev);
+
+		OUT_BATCH(d, i, GEN9_MEDIA_POOL_STATE);
+		OUT_BATCH(d, i, GEN9_MEDIA_POOL_ENABLE);
+		if (dev_priv->info.subslice_total == 3)
+			pool_config = 0x00777000;
+		OUT_BATCH(d, i, pool_config);
+		OUT_BATCH(d, i, 0);
+		OUT_BATCH(d, i, 0);
+		OUT_BATCH(d, i, 0);
+	}
+
 	OUT_BATCH(d, i, MI_BATCH_BUFFER_END);
 	so->aux_batch_size = (i * sizeof(u32)) - so->aux_batch_offset;
 
drivers/gpu/drm/i915/i915_reg.h +2 -0
index 9a2ffad..e052499 100644
@@ -379,6 +379,8 @@
  */
 #define GFX_INSTR(opcode, flags) ((0x3 << 29) | ((opcode) << 24) | (flags))
 
+#define GEN9_MEDIA_POOL_STATE     ((0x3 << 29) | (0x2 << 27) | (0x5 << 16) | 4)
+#define   GEN9_MEDIA_POOL_ENABLE  (1 << 31)
 #define GFX_OP_RASTER_RULES    ((0x3<<29)|(0x7<<24))
 #define GFX_OP_SCISSOR         ((0x3<<29)|(0x1c<<24)|(0x10<<19))
 #define   SC_UPDATE_SCISSOR       (0x1<<1)
Signature · 6 lines
1.9.1

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

Re: [PATCH v1 3/4] drm/i915:bxt: Enable Pooled EU support

Quoted 4 lines · Arun Siluvery

This mode allows to assign EUs to pools.
The command to enable this mode is sent in auxiliary golden context batch as this is only issued once with each context initialization. Thanks to Mika for the preliminary review.

A quick explanation for why this has to be in the kernel would be nice. Privileged instruction?

Not fond of the split between this and patch 4. Patch 4 intoduces one feature flag that looks different to the one we use here to enable support.
-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#9 lore raw

Re: [PATCH v1 3/4] drm/i915:bxt: Enable Pooled EU support

Quoted 9 lines · Chris Wilson
Quoted 4 lines · Arun Siluvery

This mode allows to assign EUs to pools.
The command to enable this mode is sent in auxiliary golden context batch as this is only issued once with each context initialization. Thanks to Mika for the preliminary review.

A quick explanation for why this has to be in the kernel would be nice. Privileged instruction?

The pooled mode is global. Once set, it has to stay same across all contexts until subsequent fw reset.

-Mika

Quoted 11 lines

Not fond of the split between this and patch 4. Patch 4 intoduces one feature flag that looks different to the one we use here to enable support.
-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

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

#10 lore raw

Re: [PATCH v1 3/4] drm/i915:bxt: Enable Pooled EU support

In reply to Chris Wilson
Quoted 8 lines · Chris Wilson
Quoted 4 lines · Arun Siluvery

This mode allows to assign EUs to pools.
The command to enable this mode is sent in auxiliary golden context batch as this is only issued once with each context initialization. Thanks to Mika for the preliminary review.

A quick explanation for why this has to be in the kernel would be nice. Privileged instruction?

This purpose of auxiliary batch is explained in patch2, but I can add some explanation about this one also.

Quoted 4 lines

Not fond of the split between this and patch 4. Patch 4 intoduces one feature flag that looks different to the one we use here to enable support.

I will patch4 as separate as it deals with libdrm changes but use the feature flag in this one.

regards
Arun

Quoted 2 lines

-Chris

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

#11 lore raw

Re: [PATCH v1 3/4] drm/i915:bxt: Enable Pooled EU support

Quoted 7 lines · Arun
Quoted 6 lines · Chris Wilson
Quoted 4 lines · Arun Siluvery

This mode allows to assign EUs to pools.
The command to enable this mode is sent in auxiliary golden context batch as this is only issued once with each context initialization. Thanks to Mika for the preliminary review.

Quoted 5 lines
Quoted 2 lines

A quick explanation for why this has to be in the kernel would be nice. Privileged instruction?

This purpose of auxiliary batch is explained in patch2, but I can add some explanation about this one also.

Here, I am looking for an explanation of why these commands in particular are desired. Mika's short explanation that must be the same for all contexts on the system is sufficient.
-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#12 lore raw

PATCH 4/4 drm/i915/bxt: Add get_param to query Pooled EU availability

In reply to Arun Siluvery

User space clients need to know when the pooled EU feature is present and enabled on the hardware so that they can adapt work submissions. Create a new device info flag for this purpose, and create a new GETPARAM entry to allow user space to query its setting.

Set has_pooled_eu to true in the Broxton static device info - Broxton supports the feature in hardware and the driver will enable it by default.

---
 drivers/gpu/drm/i915/i915_dma.c | 3 +++
 drivers/gpu/drm/i915/i915_drv.c | 1 +
 drivers/gpu/drm/i915/i915_drv.h | 5 ++++-
 include/uapi/drm/i915_drm.h     | 1 +
 4 files changed, 9 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/i915_dma.c +3 -0
index 5e63076..6c31beb 100644
@@ -170,6 +170,9 @@ static int i915_getparam(struct drm_device *dev, void *data,
 	case I915_PARAM_HAS_RESOURCE_STREAMER:
 		value = HAS_RESOURCE_STREAMER(dev);
 		break;
+	case I915_PARAM_HAS_POOLED_EU:
+		value = HAS_POOLED_EU(dev);
+		break;
 	default:
 		DRM_DEBUG("Unknown parameter %d\n", param->param);
 		return -EINVAL;
drivers/gpu/drm/i915/i915_drv.c +1 -0
index e44dc0d..213f74d 100644
@@ -389,6 +389,7 @@ static const struct intel_device_info intel_broxton_info = {
 	.num_pipes = 3,
 	.has_ddi = 1,
 	.has_fbc = 1,
+	.has_pooled_eu = 1,
 	GEN_DEFAULT_PIPEOFFSETS,
 	IVB_CURSOR_OFFSETS,
 };
drivers/gpu/drm/i915/i915_drv.h +4 -1
index 768d1db..32850a8 100644
@@ -775,7 +775,8 @@ struct intel_csr {
 	func(supports_tv) sep \
 	func(has_llc) sep \
 	func(has_ddi) sep \
-	func(has_fpga_dbg)
+	func(has_fpga_dbg) sep \
+	func(has_pooled_eu)
 
 #define DEFINE_FLAG(name) u8 name:1
 #define SEP_SEMICOLON ;
@@ -2549,6 +2550,8 @@ struct drm_i915_cmd_table {
 #define HAS_RESOURCE_STREAMER(dev) (IS_HASWELL(dev) || \
 				    INTEL_INFO(dev)->gen >= 8)
 
+#define HAS_POOLED_EU(dev) (INTEL_INFO(dev)->has_pooled_eu)
+
 #define INTEL_PCH_DEVICE_ID_MASK		0xff00
 #define INTEL_PCH_IBX_DEVICE_ID_TYPE		0x3b00
 #define INTEL_PCH_CPT_DEVICE_ID_TYPE		0x1c00
include/uapi/drm/i915_drm.h +1 -0
index e7c29f1..9649577 100644
@@ -356,6 +356,7 @@ typedef struct drm_i915_irq_wait {
 #define I915_PARAM_EU_TOTAL		 34
 #define I915_PARAM_HAS_GPU_RESET	 35
 #define I915_PARAM_HAS_RESOURCE_STREAMER 36
+#define I915_PARAM_HAS_POOLED_EU         37
 
 typedef struct drm_i915_getparam {
 	int param;
Signature · 6 lines
1.9.1

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