Skip to contentSkip to conversation

drm/i915: resize the GuC WOPCM for rc6

PATCH In review Peter Antoine 6 participants 15 messages Apr 26, 2016 – May 16, 2016

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

PATCH v3 1/1 drm/i915: resize the GuC WOPCM for rc6

This patch resizes the GuC WOPCM to so that the GuC and the RC6 memory spaces do not overlap.

Issue: https://jira01.devtools.intel.com/browse/VIZ-6638

---
 drivers/gpu/drm/i915/i915_guc_reg.h     | 5 +++--
 drivers/gpu/drm/i915/intel_guc_loader.c | 6 +++++-
 2 files changed, 8 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/i915_guc_reg.h +3 -2
index 80786d9..6e01238 100644
@@ -68,10 +68,11 @@
 #define GUC_MAX_IDLE_COUNT		_MMIO(0xC3E4)
 
 #define GUC_WOPCM_SIZE			_MMIO(0xc050)
-#define   GUC_WOPCM_SIZE_VALUE  	  (0x80 << 12)	/* 512KB */
+#define   GUC_WOPCM_SIZE_VALUE		(0x80 << 12)	/* 512KB */
+#define   BXT_GUC_WOPCM_SIZE_VALUE	(0x70 << 12)	/* 448KB */
 
 /* GuC addresses below GUC_WOPCM_TOP don't map through the GTT */
-#define	GUC_WOPCM_TOP			(GUC_WOPCM_SIZE_VALUE)
+#define	GUC_WOPCM_TOP			(0x80 << 12)	/* 512KB */
 
 #define GEN8_GT_PM_CONFIG		_MMIO(0x138140)
 #define GEN9LP_GT_PM_CONFIG		_MMIO(0x138140)
drivers/gpu/drm/i915/intel_guc_loader.c +5 -1
index fc3ff68..38fb321 100644
@@ -312,7 +312,11 @@ static int guc_ucode_xfer(struct drm_i915_private *dev_priv)
 	intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
 
 	/* init WOPCM */
-	I915_WRITE(GUC_WOPCM_SIZE, GUC_WOPCM_SIZE_VALUE);
+	if (IS_BROXTON(dev))
+		I915_WRITE(GUC_WOPCM_SIZE, BXT_GUC_WOPCM_SIZE_VALUE);
+	else
+		I915_WRITE(GUC_WOPCM_SIZE, GUC_WOPCM_SIZE_VALUE);
+
 	I915_WRITE(DMA_GUC_WOPCM_OFFSET, GUC_WOPCM_OFFSET_VALUE);
 
 	/* Enable MIA caching. GuC clock gating is disabled. */
Signature · 6 lines
1.9.1

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

✗ Fi.CI.BAT: failure for drm/i915: resize the GuC WOPCM for rc6

== Series Details ==

== Series Details ==

Series: drm/i915: resize the GuC WOPCM for rc6
URL : https://patchwork.freedesktop.org/series/6313/
State : failure

== Summary ==

Series 6313v1 drm/i915: resize the GuC WOPCM for rc6
http://patchwork.freedesktop.org/api/1.0/series/6313/revisions/1/mbox/

Test kms_flip:

        Subgroup basic-flip-vs-wf_vblank:
                pass       -> FAIL       (hsw-gt2)
Test kms_force_connector_basic:
        Subgroup force-edid:
                pass       -> SKIP       (ivb-t430s)
Test kms_pipe_crc_basic:
        Subgroup read-crc-pipe-b-frame-sequence:
                skip       -> PASS       (bdw-nuci7)
bdw-nuci7        total:200  pass:188  dwarn:0   dfail:0   fail:0   skip:12 
bdw-ultra        total:200  pass:175  dwarn:0   dfail:0   fail:0   skip:25 
bsw-nuc-2        total:199  pass:158  dwarn:0   dfail:0   fail:0   skip:41 
byt-nuc          total:199  pass:158  dwarn:0   dfail:0   fail:0   skip:41 
hsw-brixbox      total:200  pass:174  dwarn:0   dfail:0   fail:0   skip:26 
hsw-gt2          total:200  pass:178  dwarn:0   dfail:0   fail:1   skip:21 
ilk-hp8440p      total:200  pass:139  dwarn:0   dfail:0   fail:0   skip:61 
ivb-t430s        total:200  pass:168  dwarn:0   dfail:0   fail:0   skip:32 
skl-i7k-2        total:200  pass:173  dwarn:0   dfail:0   fail:0   skip:27 
skl-nuci5        total:200  pass:189  dwarn:0   dfail:0   fail:0   skip:11 
snb-dellxps      total:200  pass:158  dwarn:0   dfail:0   fail:0   skip:42 
snb-x220t        total:200  pass:158  dwarn:0   dfail:0   fail:1   skip:41 

Results at /archive/results/CI_IGT_test/Patchwork_2068/

f814551aa7232ed36d71244dd148b48660b53a78 drm-intel-nightly: 2016y-04m-25d-11h-36m-27s UTC integration manifest c8a3953 drm/i915: resize the GuC WOPCM for rc6

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

#2 lore raw

Re: ✗ Fi.CI.BAT: failure for drm/i915: resize the GuC WOPCM for rc6

Quoted 6 lines · Patchwork

== Series Details ==

Series: drm/i915: resize the GuC WOPCM for rc6
URL : https://patchwork.freedesktop.org/series/6313/
State : failure

Quoted 8 lines

== Summary ==

Series 6313v1 drm/i915: resize the GuC WOPCM for rc6
http://patchwork.freedesktop.org/api/1.0/series/6313/revisions/1/mbox/

Test kms_flip:

       Subgroup basic-flip-vs-wf_vblank:
               pass       -> FAIL       (hsw-gt2)

Patch cannot effect HSW as platform does not have GuC and GuC code is all guarded by platform. False Failure.

Quoted 3 lines

Test kms_force_connector_basic:

       Subgroup force-edid:
               pass       -> SKIP       (ivb-t430s)

Nothing to do with display.

Quoted 3 lines

Test kms_pipe_crc_basic:

       Subgroup read-crc-pipe-b-frame-sequence:
               skip       -> PASS       (bdw-nuci7)

Same.

Quoted 14 lines
bdw-nuci7        total:200  pass:188  dwarn:0   dfail:0   fail:0   skip:12
bdw-ultra        total:200  pass:175  dwarn:0   dfail:0   fail:0   skip:25
bsw-nuc-2        total:199  pass:158  dwarn:0   dfail:0   fail:0   skip:41
byt-nuc          total:199  pass:158  dwarn:0   dfail:0   fail:0   skip:41
hsw-brixbox      total:200  pass:174  dwarn:0   dfail:0   fail:0   skip:26
hsw-gt2          total:200  pass:178  dwarn:0   dfail:0   fail:1   skip:21
ilk-hp8440p      total:200  pass:139  dwarn:0   dfail:0   fail:0   skip:61
ivb-t430s        total:200  pass:168  dwarn:0   dfail:0   fail:0   skip:32
skl-i7k-2        total:200  pass:173  dwarn:0   dfail:0   fail:0   skip:27
skl-nuci5        total:200  pass:189  dwarn:0   dfail:0   fail:0   skip:11
snb-dellxps      total:200  pass:158  dwarn:0   dfail:0   fail:0   skip:42
snb-x220t        total:200  pass:158  dwarn:0   dfail:0   fail:1   skip:41
Quoted 6 lines

Results at /archive/results/CI_IGT_test/Patchwork_2068/

f814551aa7232ed36d71244dd148b48660b53a78 drm-intel-nightly: 2016y-04m-25d-11h-36m-27s UTC integration manifest c8a3953 drm/i915: resize the GuC WOPCM for rc6

--
Peter Antoine (Android Graphics Driver Software Engineer)
---------------------------------------------------------------------
Intel Corporation (UK) Limited
Registered No. 1134945 (England)
Registered Office: Pipers Way, Swindon SN3 1RJ VAT No: 860 2173 47
_______________________________________________
Intel-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#3 lore raw

Re: ✗ Fi.CI.BAT: failure for drm/i915: resize the GuC WOPCM for rc6

Quoted 22 lines · Peter Antoine
Quoted 6 lines · Patchwork

== Series Details ==

Series: drm/i915: resize the GuC WOPCM for rc6
URL : https://patchwork.freedesktop.org/series/6313/
State : failure

Quoted 8 lines

== Summary ==

Series 6313v1 drm/i915: resize the GuC WOPCM for rc6
http://patchwork.freedesktop.org/api/1.0/series/6313/revisions/1/mbox/

Test kms_flip:

      Subgroup basic-flip-vs-wf_vblank:
              pass       -> FAIL       (hsw-gt2)

Patch cannot effect HSW as platform does not have GuC and GuC code is all guarded by platform. False Failure.

Quoted 3 lines

Test kms_force_connector_basic:

      Subgroup force-edid:
              pass       -> SKIP       (ivb-t430s)
Quoted 6 lines

Nothing to do with display.

Quoted 3 lines

Test kms_pipe_crc_basic:

      Subgroup read-crc-pipe-b-frame-sequence:
              skip       -> PASS       (bdw-nuci7)

Same.

The idea is to digg through bugzilla to make sure we have bugs for all of these, not just shrug them off.

Yes this is work, and yes it's meant to motivate people to fix up all the fail we still have.
-Daniel

Quoted 33 lines
Quoted 15 lines

bdw-nuci7        total:200  pass:188  dwarn:0   dfail:0   fail:0   skip:12
bdw-ultra        total:200  pass:175  dwarn:0   dfail:0   fail:0   skip:25
bsw-nuc-2        total:199  pass:158  dwarn:0   dfail:0   fail:0   skip:41
byt-nuc          total:199  pass:158  dwarn:0   dfail:0   fail:0   skip:41
hsw-brixbox      total:200  pass:174  dwarn:0   dfail:0   fail:0   skip:26
hsw-gt2          total:200  pass:178  dwarn:0   dfail:0   fail:1   skip:21
ilk-hp8440p      total:200  pass:139  dwarn:0   dfail:0   fail:0   skip:61
ivb-t430s        total:200  pass:168  dwarn:0   dfail:0   fail:0   skip:32
skl-i7k-2        total:200  pass:173  dwarn:0   dfail:0   fail:0   skip:27
skl-nuci5        total:200  pass:189  dwarn:0   dfail:0   fail:0   skip:11
snb-dellxps      total:200  pass:158  dwarn:0   dfail:0   fail:0   skip:42
snb-x220t        total:200  pass:158  dwarn:0   dfail:0   fail:1   skip:41
Quoted 6 lines

Results at /archive/results/CI_IGT_test/Patchwork_2068/

f814551aa7232ed36d71244dd148b48660b53a78 drm-intel-nightly: 2016y-04m-25d-11h-36m-27s UTC integration manifest c8a3953 drm/i915: resize the GuC WOPCM for rc6

--
Peter Antoine (Android Graphics Driver Software Engineer)
---------------------------------------------------------------------
Intel Corporation (UK) Limited
Registered No. 1134945 (England)
Registered Office: Pipers Way, Swindon SN3 1RJ VAT No: 860 2173 47
_______________________________________________
Intel-gfx mailing list [email protected]
https://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]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#4 lore raw

Re: ✗ Fi.CI.BAT: failure for drm/i915: resize the GuC WOPCM for rc6

-----Original Message-----
From: Daniel Vetter [mailto:[email protected]] On Behalf Of Daniel Vetter
Sent: Tuesday, April 26, 2016 3:20 PM
To: Antoine, Peter <[email protected]>
Cc: [email protected]
Subject: Re: [Intel-gfx] ✗ Fi.CI.BAT: failure for drm/i915: resize the GuC WOPCM for rc6

Quoted 14 lines · Peter Antoine
Quoted 12 lines · Patchwork

== Series Details ==

Series: drm/i915: resize the GuC WOPCM for rc6
URL : https://patchwork.freedesktop.org/series/6313/
State : failure

== Summary ==

Series 6313v1 drm/i915: resize the GuC WOPCM for rc6
http://patchwork.freedesktop.org/api/1.0/series/6313/revisions/1/mbox
/

Quoted 5 lines
Quoted 3 lines

Test kms_flip:

      Subgroup basic-flip-vs-wf_vblank:
              pass       -> FAIL       (hsw-gt2)

Patch cannot effect HSW as platform does not have GuC and GuC code is all guarded by platform. False Failure.

Already reported.
https://bugs.freedesktop.org/show_bug.cgi?id=94294

Quoted 5 lines
Quoted 4 lines

Test kms_force_connector_basic:

      Subgroup force-edid:
              pass       -> SKIP       (ivb-t430s)

Nothing to do with display.

Already reported:
https://bugs.freedesktop.org/show_bug.cgi?id=93123

Quoted 5 lines
Quoted 4 lines

Test kms_pipe_crc_basic:

      Subgroup read-crc-pipe-b-frame-sequence:
              skip       -> PASS       (bdw-nuci7)

Same.

The BAT should not fail on a pass.

The idea is to digg through bugzilla to make sure we have bugs for all of these, not just shrug them off.

Yes this is work, and yes it's meant to motivate people to fix up all the fail we still have. -Daniel

Quoted 23 lines
Quoted 15 lines

bdw-nuci7        total:200  pass:188  dwarn:0   dfail:0   fail:0   skip:12
bdw-ultra        total:200  pass:175  dwarn:0   dfail:0   fail:0   skip:25
bsw-nuc-2        total:199  pass:158  dwarn:0   dfail:0   fail:0   skip:41
byt-nuc          total:199  pass:158  dwarn:0   dfail:0   fail:0   skip:41
hsw-brixbox      total:200  pass:174  dwarn:0   dfail:0   fail:0   skip:26
hsw-gt2          total:200  pass:178  dwarn:0   dfail:0   fail:1   skip:21
ilk-hp8440p      total:200  pass:139  dwarn:0   dfail:0   fail:0   skip:61
ivb-t430s        total:200  pass:168  dwarn:0   dfail:0   fail:0   skip:32
skl-i7k-2        total:200  pass:173  dwarn:0   dfail:0   fail:0   skip:27
skl-nuci5        total:200  pass:189  dwarn:0   dfail:0   fail:0   skip:11
snb-dellxps      total:200  pass:158  dwarn:0   dfail:0   fail:0   skip:42
snb-x220t        total:200  pass:158  dwarn:0   dfail:0   fail:1   skip:41
Quoted 7 lines

Results at /archive/results/CI_IGT_test/Patchwork_2068/

f814551aa7232ed36d71244dd148b48660b53a78 drm-intel-nightly:
2016y-04m-25d-11h-36m-27s UTC integration manifest c8a3953 drm/i915: resize the GuC WOPCM for rc6

Quoted 11 lines

--
Peter Antoine (Android Graphics Driver Software Engineer)
---------------------------------------------------------------------
Intel Corporation (UK) Limited
Registered No. 1134945 (England)
Registered Office: Pipers Way, Swindon SN3 1RJ VAT No: 860 2173 47
_______________________________________________
Intel-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#5 lore raw
In reply to Peter Antoine
Quoted 35 lines · Peter Antoine

This patch resizes the GuC WOPCM to so that the GuC and the RC6 memory spaces do not overlap.

Issue: https://jira01.devtools.intel.com/browse/VIZ-6638

---
  drivers/gpu/drm/i915/i915_guc_reg.h     | 5 +++--
  drivers/gpu/drm/i915/intel_guc_loader.c | 6 +++++-
  2 files changed, 8 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/i915_guc_reg.h +3 -2
index 80786d9..6e01238 100644
@@ -68,10 +68,11 @@
  #define GUC_MAX_IDLE_COUNT		_MMIO(0xC3E4)
 
  #define GUC_WOPCM_SIZE			_MMIO(0xc050)
-#define   GUC_WOPCM_SIZE_VALUE  	  (0x80 << 12)	/* 512KB */
+#define   GUC_WOPCM_SIZE_VALUE		(0x80 << 12)	/* 512KB */
+#define   BXT_GUC_WOPCM_SIZE_VALUE	(0x70 << 12)	/* 448KB */
 
  /* GuC addresses below GUC_WOPCM_TOP don't map through the GTT */
-#define	GUC_WOPCM_TOP			(GUC_WOPCM_SIZE_VALUE)
+#define	GUC_WOPCM_TOP			(0x80 << 12)	/* 512KB */
 
  #define GEN8_GT_PM_CONFIG		_MMIO(0x138140)
  #define GEN9LP_GT_PM_CONFIG		_MMIO(0x138140)
drivers/gpu/drm/i915/intel_guc_loader.c +0 -0
index fc3ff68..38fb321 100644
@@ -312,7 +312,11 @@ static int guc_ucode_xfer(struct drm_i915_private *dev_priv)
  	intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
 

drivers/gpu/drm/i915/intel_guc_loader.c static int guc_ucode_xfer(struct drm_i915_private *dev_priv) · line 322

Quoted 10 lines
drivers/gpu/drm/i915/intel_guc_loader.c +5 -1
  	/* init WOPCM */
-	I915_WRITE(GUC_WOPCM_SIZE, GUC_WOPCM_SIZE_VALUE);
+	if (IS_BROXTON(dev))
+		I915_WRITE(GUC_WOPCM_SIZE, BXT_GUC_WOPCM_SIZE_VALUE);
+	else
+		I915_WRITE(GUC_WOPCM_SIZE, GUC_WOPCM_SIZE_VALUE);
+
  	I915_WRITE(DMA_GUC_WOPCM_OFFSET, GUC_WOPCM_OFFSET_VALUE);
 
  	/* Enable MIA caching. GuC clock gating is disabled. */

So, this gives the right result, but doesn't really show or explain why we have different values, or how the values are arrived at; they're just more magic numbers. Also, in the loader there's a check on the firmware size that uses different values. So I'd rather prefer the unified approach in the attached version ...

.Dave.

Quoted 1 line

From fa0f5e514eea4342933e237e693ea547b9b10f81 Mon Sep 17 00:00:00 2001

From: Peter Antoine <[email protected]>
Date: Tue, 26 Apr 2016 10:11:41 +0100
Subject: [PATCH] drm/i915/bxt: reserve space for RC6 in the the GuC WOPCM
Organization: Intel Corporation (UK) Ltd. - Co. Reg. #1134945 - Pipers Way, Swindon SN3 1RJ

This patch resizes the GuC WOPCM (specifically on BXT) so that the GuC and RC6 memory spaces do not overlap.

Issue: https://jira01.devtools.intel.com/browse/VIZ-6638

---
 drivers/gpu/drm/i915/i915_guc_reg.h     |  6 +++---
 drivers/gpu/drm/i915/intel_guc_loader.c | 17 +++++++++++++----
 2 files changed, 16 insertions(+), 7 deletions(-)
drivers/gpu/drm/i915/i915_guc_reg.h +3 -3
index 80786d9..cf5a65b 100644
@@ -67,11 +67,11 @@
 #define   GUC_WOPCM_OFFSET_VALUE	  0x80000	/* 512KB */
 #define GUC_MAX_IDLE_COUNT		_MMIO(0xC3E4)
 
+/* Defines WOPCM space available to GuC firmware */
 #define GUC_WOPCM_SIZE			_MMIO(0xc050)
-#define   GUC_WOPCM_SIZE_VALUE  	  (0x80 << 12)	/* 512KB */
-
 /* GuC addresses below GUC_WOPCM_TOP don't map through the GTT */
-#define	GUC_WOPCM_TOP			(GUC_WOPCM_SIZE_VALUE)
+#define   GUC_WOPCM_TOP			  (0x80 << 12)	/* 512KB */
+#define   BXT_GUC_WOPCM_RC6_RESERVED	  (0x10 << 12)	/* 64KB  */
 
 #define GEN8_GT_PM_CONFIG		_MMIO(0x138140)
 #define GEN9LP_GT_PM_CONFIG		_MMIO(0x138140)
drivers/gpu/drm/i915/intel_guc_loader.c +13 -4
index 876e5da..80fc1e0 100644
@@ -281,6 +281,17 @@ static int guc_ucode_xfer_dma(struct drm_i915_private *dev_priv)
 	return ret;
 }
 
+static u32 guc_wopcm_size(struct drm_i915_private *dev_priv)
+{
+	u32 wopcm_size = GUC_WOPCM_TOP;
+
+	/* On BXT, the top of WOPCM is reserved for RC6 context */
+	if (IS_BROXTON(dev_priv))
+		wopcm_size -= BXT_GUC_WOPCM_RC6_RESERVED;
+
+	return wopcm_size;
+}
+
 /*
  * Load the GuC firmware blob into the MinuteIA.
  */
@@ -308,7 +319,7 @@ static int guc_ucode_xfer(struct drm_i915_private *dev_priv)
 	intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
 
 	/* init WOPCM */
-	I915_WRITE(GUC_WOPCM_SIZE, GUC_WOPCM_SIZE_VALUE);
+	I915_WRITE(GUC_WOPCM_SIZE, guc_wopcm_size(dev_priv));
 	I915_WRITE(DMA_GUC_WOPCM_OFFSET, GUC_WOPCM_OFFSET_VALUE);
 
 	/* Enable MIA caching. GuC clock gating is disabled. */
@@ -552,9 +563,7 @@ static void guc_fw_fetch(struct drm_device *dev, struct intel_guc_fw *guc_fw)
 
 	/* Header and uCode will be loaded to WOPCM. Size of the two. */
 	size = guc_fw->header_size + guc_fw->ucode_size;
-
-	/* Top 32k of WOPCM is reserved (8K stack + 24k RC6 context). */
-	if (size > GUC_WOPCM_SIZE_VALUE - 0x8000) {
+	if (size > guc_wopcm_size(dev->dev_private)) {
 		DRM_ERROR("Firmware is too large to fit in WOPCM\n");
 		goto fail;
 	}
Signature · 1 line
1.9.1

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

#6 lore raw

The attached version still does not explain that the WOPCM_TOP is to tell the GuC not to use that space. The extra information does not aid anybody as the information is used internally within the GuC.

But, I have not actual objection to the patch.

Peter.

-----Original Message-----
From: Gordon, David S
Sent: Thursday, May 5, 2016 2:41 PM
To: Antoine, Peter <[email protected]>; [email protected]
Cc: Vivi, Rodrigo <[email protected]>
Subject: Re: [PATCH v3] drm/i915: resize the GuC WOPCM for rc6

Quoted 37 lines · Peter Antoine

This patch resizes the GuC WOPCM to so that the GuC and the RC6 memory spaces do not overlap.

Issue: https://jira01.devtools.intel.com/browse/VIZ-6638

---
  drivers/gpu/drm/i915/i915_guc_reg.h     | 5 +++--
  drivers/gpu/drm/i915/intel_guc_loader.c | 6 +++++-
  2 files changed, 8 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/i915_guc_reg.h +3 -2
index 80786d9..6e01238 100644
@@ -68,10 +68,11 @@
  #define GUC_MAX_IDLE_COUNT		_MMIO(0xC3E4)
 
  #define GUC_WOPCM_SIZE			_MMIO(0xc050)
-#define   GUC_WOPCM_SIZE_VALUE  	  (0x80 << 12)	/* 512KB */
+#define   GUC_WOPCM_SIZE_VALUE		(0x80 << 12)	/* 512KB */
+#define   BXT_GUC_WOPCM_SIZE_VALUE	(0x70 << 12)	/* 448KB */
 
  /* GuC addresses below GUC_WOPCM_TOP don't map through the GTT */
-#define	GUC_WOPCM_TOP			(GUC_WOPCM_SIZE_VALUE)
+#define	GUC_WOPCM_TOP			(0x80 << 12)	/* 512KB */
 
  #define GEN8_GT_PM_CONFIG		_MMIO(0x138140)
  #define GEN9LP_GT_PM_CONFIG		_MMIO(0x138140)
drivers/gpu/drm/i915/intel_guc_loader.c +0 -0
index fc3ff68..38fb321 100644
@@ -312,7 +312,11 @@ static int guc_ucode_xfer(struct drm_i915_private *dev_priv)
  	intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
 

drivers/gpu/drm/i915/intel_guc_loader.c static int guc_ucode_xfer(struct drm_i915_private *dev_priv) · line 322

Quoted 10 lines
drivers/gpu/drm/i915/intel_guc_loader.c +5 -1
  	/* init WOPCM */
-	I915_WRITE(GUC_WOPCM_SIZE, GUC_WOPCM_SIZE_VALUE);
+	if (IS_BROXTON(dev))
+		I915_WRITE(GUC_WOPCM_SIZE, BXT_GUC_WOPCM_SIZE_VALUE);
+	else
+		I915_WRITE(GUC_WOPCM_SIZE, GUC_WOPCM_SIZE_VALUE);
+
  	I915_WRITE(DMA_GUC_WOPCM_OFFSET, GUC_WOPCM_OFFSET_VALUE);
 
  	/* Enable MIA caching. GuC clock gating is disabled. */

So, this gives the right result, but doesn't really show or explain why we have different values, or how the values are arrived at; they're just more magic numbers. Also, in the loader there's a check on the firmware size that uses different values. So I'd rather prefer the unified approach in the attached version ...

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

#7 lore raw
Quoted 1 line · Peter

The attached version still does not explain that the WOPCM_TOP is to tell the GuC not to use that space.

That's NOT what WOPCM_TOP means. The GuC is allowed to use the space up to the value stored in the GUC_WOPCM_SIZE register (as the comment above the #define says). Architecturally, this is allowed to be any value greater than (16K+sizeof internal SRAM (64, 128, or 256K)) and less than or equal to GUC_WOPCM_TOP (which is a platform-independent constant), so we normally choose the maximm allowed. Howver on BXT, we need to leave some space at the top for the RC6 image, hence the logic (and comments!) in guc_wopcm_size().

Quoted 1 line

The extra information does not aid anybody as the information is used internally within the GuC.

It may help the next person who has to figure out what's gone wrong on some future chip that needs more than 64K for RC6!

.Dave.

Quoted 5 lines

But, I have not actual objection to the patch.

Peter.

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

#8 lore raw
Quoted 8 lines · Dave Gordon

On 05/05/2016 15:02, Antoine, Peter wrote:

The attached version still does not explain that the WOPCM_TOP is to tell the GuC not to use that space.

That's NOT what WOPCM_TOP means. The GuC is allowed to use the space up to the value stored in the GUC_WOPCM_SIZE register (as the comment above the #define says). Architecturally, this is allowed to be any value greater than (16K+sizeof internal SRAM (64, 128, or 256K)) and less than or equal to GUC_WOPCM_TOP (which is a platform-independent constant), so we normally choose the maximm allowed. Howver on BXT, we need to leave some space at the top for the RC6 image, hence the logic (and comments!) in guc_wopcm_size().

Yes, the firmware can use upto GUC_WOPCM_TOP and to leave the rest alone.

Quoted 4 lines

The extra information does not aid anybody as the information is used internally within the GuC.

It may help the next person who has to figure out what's gone wrong on some future chip that needs more than 64K for RC6!

You hid a if statement in a function (making the code harder to read and more prone to error). Where maybe a slightly clearer comment was required.

And this patch has been held up two weeks just for a better comment.

Peter.

Quoted 2 lines

.Dave.

And what if the next reserved space is not for RC6?

Quoted 9 lines

But, I have not actual objection to the patch.

Peter.

--
Peter Antoine (Android Graphics Driver Software Engineer)
---------------------------------------------------------------------
Intel Corporation (UK) Limited
Registered No. 1134945 (England)
Registered Office: Pipers Way, Swindon SN3 1RJ VAT No: 860 2173 47
_______________________________________________
Intel-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Tested-by Reviewed-by
#9 lore raw
Quoted 28 lines · Peter Antoine
Quoted 6 lines · Dave Gordon

On 05/05/2016 15:02, Antoine, Peter wrote:

The attached version still does not explain that the WOPCM_TOP is to tell the GuC not to use that space.

Quoted 9 lines

That's NOT what WOPCM_TOP means. The GuC is allowed to use the space up to the value stored in the GUC_WOPCM_SIZE register (as the comment above the #define says). Architecturally, this is allowed to be any value greater than
(16K+sizeof internal SRAM (64, 128, or 256K)) and less than or equal to GUC_WOPCM_TOP (which is a platform-independent constant), so we normally choose the maximm allowed. Howver on BXT, we need to leave some space at the top for the
RC6 image, hence the logic (and comments!) in guc_wopcm_size().

Yes, the firmware can use upto GUC_WOPCM_TOP and to leave the rest alone.

Quoted 6 lines

The extra information does not aid anybody as the information is used internally within the GuC.

It may help the next person who has to figure out what's gone wrong on some future chip that needs more than 64K for RC6!

You hid a if statement in a function (making the code harder to read and more prone to error). Where maybe a slightly clearer comment was required.

Quoted 17 lines

And this patch has been held up two weeks just for a better comment.

Peter.

Quoted 2 lines

.Dave.

And what if the next reserved space is not for RC6?

Quoted 9 lines

But, I have not actual objection to the patch.

Peter.

Quoted 12 lines

--
Peter Antoine (Android Graphics Driver Software Engineer)
---------------------------------------------------------------------
Intel Corporation (UK) Limited
Registered No. 1134945 (England)
Registered Office: Pipers Way, Swindon SN3 1RJ VAT No: 860 2173 47
_______________________________________________
Intel-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

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

#10 lore raw
In reply to Dave Gordon
Quoted 12 lines · Dave Gordon
Quoted 1 line · Peter

The attached version still does not explain that the WOPCM_TOP is to tell the GuC not to use that space.

That's NOT what WOPCM_TOP means. The GuC is allowed to use the space up to the value stored in the GUC_WOPCM_SIZE register (as the comment above the #define says). Architecturally, this is allowed to be any value greater than (16K+sizeof internal SRAM (64, 128, or 256K)) and less than or equal to GUC_WOPCM_TOP (which is a platform-independent constant), so we normally choose the maximm allowed. Howver on BXT, we need to leave some space at the top for the RC6 image, hence the logic (and comments!) in guc_wopcm_size().

Quoted 11 lines
Quoted 1 line

The extra information does not aid anybody as the information is used internally within the GuC.

It may help the next person who has to figure out what's gone wrong on some future chip that needs more than 64K for RC6!

.Dave.

Quoted 5 lines

But, I have not actual objection to the patch.

Peter.

Unfortunately Dave's patch locked my test system on bootup, so I've t-b & r-b'd Peter's.
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#11 lore raw
Quoted 21 lines · Nick Hoath
Quoted 20 lines · Dave Gordon
Quoted 2 lines · Peter

The attached version still does not explain that the WOPCM_TOP is to tell the GuC not to use that space.

That's NOT what WOPCM_TOP means. The GuC is allowed to use the space up to the value stored in the GUC_WOPCM_SIZE register (as the comment above the #define says). Architecturally, this is allowed to be any value greater than (16K+sizeof internal SRAM (64, 128, or 256K)) and less than or equal to GUC_WOPCM_TOP (which is a platform-independent constant), so we normally choose the maximm allowed. Howver on BXT, we need to leave some space at the top for the RC6 image, hence the logic (and comments!) in guc_wopcm_size().

Quoted 2 lines

The extra information does not aid anybody as the information is used internally within the GuC.

It may help the next person who has to figure out what's gone wrong on some future chip that needs more than 64K for RC6!

.Dave.

Quoted 1 line
Quoted 7 lines
Quoted 5 lines
Quoted 4 lines

But, I have not actual objection to the patch.

Peter.

Unfortunately Dave's patch locked my test system on bootup, so I've t-b & r-b'd Peter's.

They're equivalent, unless your firmware happens to be between 458752 and 491520 bytes in size (in which case you have a problem anyway).

To check, I've run both versions, with debug printing the value chosen (on SKL) and the value that would have been chosen on BXT, and they're identical (and both work). So I think your build had some other problem unrelated to the specific patch.

I've no problem with using Peter's patch for now, but it's not just a matter of the comments; there's also the other use(s) of GUC_WOP_(TOP,SIZE_VALUE), with ad-hoc additions or subtractions. So it still needs fixing properly.

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

Tested-by Reviewed-by
#12 lore raw
Quoted 38 lines · David S
Quoted 21 lines · Nick Hoath
Quoted 20 lines · Dave Gordon
Quoted 2 lines · Peter

The attached version still does not explain that the WOPCM_TOP is to tell the GuC not to use that space.

That's NOT what WOPCM_TOP means. The GuC is allowed to use the space up to the value stored in the GUC_WOPCM_SIZE register (as the comment above the #define says). Architecturally, this is allowed to be any value greater than (16K+sizeof internal SRAM (64, 128, or 256K)) and less than or equal to GUC_WOPCM_TOP (which is a platform-independent constant), so we normally choose the maximm allowed. Howver on BXT, we need to leave some space at the top for the RC6 image, hence the logic (and comments!) in guc_wopcm_size().

Quoted 2 lines

The extra information does not aid anybody as the information is used internally within the GuC.

It may help the next person who has to figure out what's gone wrong on some future chip that needs more than 64K for RC6!

.Dave.

Quoted 1 line
Quoted 7 lines
Quoted 5 lines
Quoted 4 lines

But, I have not actual objection to the patch.

Peter.

Unfortunately Dave's patch locked my test system on bootup, so I've t-b & r-b'd Peter's.

They're equivalent, unless your firmware happens to be between 458752 and 491520 bytes in size (in which case you have a problem anyway).

To check, I've run both versions, with debug printing the value chosen (on SKL) and the value that would have been chosen on BXT, and they're identical (and both work). So I think your build had some other problem unrelated to the specific patch.

Quoted 7 lines

I've no problem with using Peter's patch for now, but it's not just a matter of the comments; there's also the other use(s) of GUC_WOP_(TOP,SIZE_VALUE), with ad-hoc additions or subtractions. So it still needs fixing properly.

.Dave.

After a rebuild & a retest, Dave's patch works fine. Therefore for "drm/i915/bxt: reserve space for RC6 in the the GuC WOPCM":

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

#13 lore raw
In reply to Peter Antoine
Quoted 2 lines · On Tue, 26 Apr 2016 10:11:41 +0100 Peter Antoine <[email protected]>…

This patch resizes the GuC WOPCM to so that the GuC and the RC6 memory spaces do not overlap.

Hi Peter,

With this patch applied to our IOTG kernel tree, I see a regression with the RC6 residency values. The pm_rc6_residency fails the accuracy test.

The counter is still updating, but instead of seeing appx. 3000ms in RC6 I see appx. 100ms during the 3 second sleep.

Simply reverting this one change makes the RC6 residency counts go back to normal. Seems like this patch should fix a problem like this, not cause it.

This is on a Leaf Hill BXT CRB platform.

Any pointers or thoughts on how to debug this would be appreciated.

Thanks,
Bob

Quoted 43 lines

Issue: https://jira01.devtools.intel.com/browse/VIZ-6638

---
 drivers/gpu/drm/i915/i915_guc_reg.h     | 5 +++--
 drivers/gpu/drm/i915/intel_guc_loader.c | 6 +++++-
 2 files changed, 8 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/i915_guc_reg.h +3 -2
index 80786d9..6e01238 100644
@@ -68,10 +68,11 @@
 #define GUC_MAX_IDLE_COUNT		_MMIO(0xC3E4)
 
 #define GUC_WOPCM_SIZE			_MMIO(0xc050)
-#define   GUC_WOPCM_SIZE_VALUE  	  (0x80 << 12)	/* 512KB */
+#define   GUC_WOPCM_SIZE_VALUE		(0x80 << 12)	/* 512KB */
+#define   BXT_GUC_WOPCM_SIZE_VALUE	(0x70 << 12)	/* 448KB */
 
 /* GuC addresses below GUC_WOPCM_TOP don't map through the GTT */
-#define	GUC_WOPCM_TOP			(GUC_WOPCM_SIZE_VALUE)
+#define	GUC_WOPCM_TOP			(0x80 << 12)	/* 512KB */
 
 #define GEN8_GT_PM_CONFIG		_MMIO(0x138140)
 #define GEN9LP_GT_PM_CONFIG		_MMIO(0x138140)
drivers/gpu/drm/i915/intel_guc_loader.c +5 -1
index fc3ff68..38fb321 100644
@@ -312,7 +312,11 @@ static int guc_ucode_xfer(struct drm_i915_private *dev_priv)
 	intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
 
 	/* init WOPCM */
-	I915_WRITE(GUC_WOPCM_SIZE, GUC_WOPCM_SIZE_VALUE);
+	if (IS_BROXTON(dev))
+		I915_WRITE(GUC_WOPCM_SIZE, BXT_GUC_WOPCM_SIZE_VALUE);
+	else
+		I915_WRITE(GUC_WOPCM_SIZE, GUC_WOPCM_SIZE_VALUE);
+
 	I915_WRITE(DMA_GUC_WOPCM_OFFSET, GUC_WOPCM_OFFSET_VALUE);
 
 	/* Enable MIA caching. GuC clock gating is disabled. */
Signature · 11 lines
--
Bob Paauwe                  
[email protected]
IOTG / PED Software Organization
Intel Corp.  Folsom, CA
(916) 356-6193    

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

Hi Bob,

It was tested on a BXT-T RVP and LeafHill A and worked fine. The patch has been in the Android tree for at least a couple of months (probably more) and has seemed to be stable. But, not sure Android was run on the BXT CRB.

It could be the GuC firmware crashing, there are some fixes in 8.7. Has the size of the SRAM changed on that board? Is something else reserving space in that area?

The size that was reserved was passed onto us from the GuC firmware team via the RC6 team. It might be worth talking to the power team in Bangalore. [email protected] is probably your best bet for RC6 issues.

I hope that helps.

Peter.

-----Original Message-----
From: Paauwe, Bob J
Sent: Tuesday, May 10, 2016 4:59 PM
To: Antoine, Peter <[email protected]>
Cc: [email protected]; Vivi, Rodrigo <[email protected]>
Subject: Re: [Intel-gfx] [PATCH v3] drm/i915: resize the GuC WOPCM for rc6

Quoted 2 lines · On Tue, 26 Apr 2016 10:11:41 +0100 Peter Antoine <[email protected]>…

This patch resizes the GuC WOPCM to so that the GuC and the RC6 memory spaces do not overlap.

Hi Peter,

With this patch applied to our IOTG kernel tree, I see a regression with the RC6 residency values. The pm_rc6_residency fails the accuracy test.

The counter is still updating, but instead of seeing appx. 3000ms in RC6 I see appx. 100ms during the 3 second sleep.

Simply reverting this one change makes the RC6 residency counts go back to normal. Seems like this patch should fix a problem like this, not cause it.

This is on a Leaf Hill BXT CRB platform.

Any pointers or thoughts on how to debug this would be appreciated.

Thanks,
Bob

Quoted 35 lines

Issue: https://jira01.devtools.intel.com/browse/VIZ-6638

---
 drivers/gpu/drm/i915/i915_guc_reg.h     | 5 +++--
 drivers/gpu/drm/i915/intel_guc_loader.c | 6 +++++-
 2 files changed, 8 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/i915_guc_reg.h +3 -2
index 80786d9..6e01238 100644
@@ -68,10 +68,11 @@
 #define GUC_MAX_IDLE_COUNT		_MMIO(0xC3E4)
 
 #define GUC_WOPCM_SIZE			_MMIO(0xc050)
-#define   GUC_WOPCM_SIZE_VALUE  	  (0x80 << 12)	/* 512KB */
+#define   GUC_WOPCM_SIZE_VALUE		(0x80 << 12)	/* 512KB */
+#define   BXT_GUC_WOPCM_SIZE_VALUE	(0x70 << 12)	/* 448KB */
 
 /* GuC addresses below GUC_WOPCM_TOP don't map through the GTT */
-#define	GUC_WOPCM_TOP			(GUC_WOPCM_SIZE_VALUE)
+#define	GUC_WOPCM_TOP			(0x80 << 12)	/* 512KB */
 
 #define GEN8_GT_PM_CONFIG		_MMIO(0x138140)
 #define GEN9LP_GT_PM_CONFIG		_MMIO(0x138140)
drivers/gpu/drm/i915/intel_guc_loader.c +0 -0
index fc3ff68..38fb321 100644
@@ -312,7 +312,11 @@ static int guc_ucode_xfer(struct drm_i915_private *dev_priv)
 	intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
 

drivers/gpu/drm/i915/intel_guc_loader.c static int guc_ucode_xfer(struct drm_i915_private *dev_priv) · line 322

Quoted 10 lines
drivers/gpu/drm/i915/intel_guc_loader.c +5 -1
 	/* init WOPCM */
-	I915_WRITE(GUC_WOPCM_SIZE, GUC_WOPCM_SIZE_VALUE);
+	if (IS_BROXTON(dev))
+		I915_WRITE(GUC_WOPCM_SIZE, BXT_GUC_WOPCM_SIZE_VALUE);
+	else
+		I915_WRITE(GUC_WOPCM_SIZE, GUC_WOPCM_SIZE_VALUE);
+
 	I915_WRITE(DMA_GUC_WOPCM_OFFSET, GUC_WOPCM_OFFSET_VALUE);
 
 	/* Enable MIA caching. GuC clock gating is disabled. */

--
--
Bob Paauwe
[email protected]
IOTG / PED Software Organization
Intel Corp. Folsom, CA
(916) 356-6193

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