Skip to contentSkip to conversation

drm/i915/cnl: WaForceEnableNonCoherent

PATCH In review Rodrigo Vivi 5 participants 7 messages Aug 23, 2017 – Aug 28, 2017

Subsystems (2)
lore
Share & tools
lore

Rodrigo Vivi

author maintainer
#0 lore raw

PATCH 1/1 drm/i915/cnl: WaForceEnableNonCoherent

Must Force Non-Coherent whenever executing a 3D context. This is a workaround for a possible hang in the unlikely event
a TLB invalidation occurs during a PSD flush.
Set masked bit 4 in 0x7300 during boot.

Cc Mika Kuoppala <[email protected]>
Cc Oscar Mateo <[email protected]>
---
 drivers/gpu/drm/i915/intel_engine_cs.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_engine_cs.c +4 -1
index a6ac9d0a4156..7dfc78b038c4 100644
@@ -1071,8 +1071,11 @@ static int cnl_init_workarounds(struct intel_engine_cs *engine)
 	int ret;
 
 	/* WaForceContextSaveRestoreNonCoherent:cnl */
+	/* WaForceEnableNonCoherent:cnl */
 	WA_SET_BIT_MASKED(CNL_HDC_CHICKEN0,
-			  HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT);
+			  HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT |
+			  HDC_FORCE_NON_COHERENT);
+
 
 	/* WaDisableReplayBufferBankArbitrationOptimization:cnl */
 	WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2,
Signature · 6 lines
2.13.2

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

✓ Fi.CI.BAT: success for drm/i915/cnl: WaForceEnableNonCoherent

== Series Details ==

== Series Details ==

Series: drm/i915/cnl: WaForceEnableNonCoherent
URL : https://patchwork.freedesktop.org/series/29243/
State : success

== Summary ==

Series 29243v1 drm/i915/cnl: WaForceEnableNonCoherent
https://patchwork.freedesktop.org/api/1.0/series/29243/revisions/1/mbox/

Test kms_pipe_crc_basic:

        Subgroup suspend-read-crc-pipe-c:
                fail       -> PASS       (fi-skl-6700k) fdo#100367

fdo#100367 https://bugs.freedesktop.org/show_bug.cgi?id=100367

fi-bdw-5557u     total:279  pass:268  dwarn:0   dfail:0   fail:0   skip:11  time:449s
fi-bdw-gvtdvm    total:279  pass:265  dwarn:0   dfail:0   fail:0   skip:14  time:438s
fi-blb-e6850     total:279  pass:224  dwarn:1   dfail:0   fail:0   skip:54  time:356s
fi-bsw-n3050     total:279  pass:243  dwarn:0   dfail:0   fail:0   skip:36  time:558s
fi-bwr-2160      total:279  pass:184  dwarn:0   dfail:0   fail:0   skip:95  time:253s
fi-bxt-j4205     total:279  pass:260  dwarn:0   dfail:0   fail:0   skip:19  time:523s
fi-byt-j1900     total:279  pass:254  dwarn:1   dfail:0   fail:0   skip:24  time:528s
fi-byt-n2820     total:279  pass:251  dwarn:0   dfail:0   fail:0   skip:28  time:514s
fi-elk-e7500     total:279  pass:230  dwarn:0   dfail:0   fail:0   skip:49  time:439s
fi-glk-2a        total:279  pass:260  dwarn:0   dfail:0   fail:0   skip:19  time:606s
fi-hsw-4770      total:279  pass:263  dwarn:0   dfail:0   fail:0   skip:16  time:451s
fi-hsw-4770r     total:279  pass:263  dwarn:0   dfail:0   fail:0   skip:16  time:419s
fi-ilk-650       total:279  pass:229  dwarn:0   dfail:0   fail:0   skip:50  time:427s
fi-ivb-3520m     total:279  pass:261  dwarn:0   dfail:0   fail:0   skip:18  time:508s
fi-ivb-3770      total:279  pass:261  dwarn:0   dfail:0   fail:0   skip:18  time:473s
fi-kbl-7500u     total:279  pass:261  dwarn:0   dfail:0   fail:0   skip:18  time:478s
fi-kbl-7560u     total:279  pass:269  dwarn:0   dfail:0   fail:0   skip:10  time:603s
fi-kbl-r         total:279  pass:261  dwarn:0   dfail:0   fail:0   skip:18  time:601s
fi-pnv-d510      total:279  pass:223  dwarn:1   dfail:0   fail:0   skip:55  time:524s
fi-skl-6260u     total:279  pass:269  dwarn:0   dfail:0   fail:0   skip:10  time:468s
fi-skl-6700k     total:279  pass:261  dwarn:0   dfail:0   fail:0   skip:18  time:478s
fi-skl-6770hq    total:279  pass:269  dwarn:0   dfail:0   fail:0   skip:10  time:486s
fi-skl-gvtdvm    total:279  pass:266  dwarn:0   dfail:0   fail:0   skip:13  time:445s
fi-skl-x1585l    total:279  pass:268  dwarn:0   dfail:0   fail:0   skip:11  time:484s
fi-snb-2520m     total:279  pass:251  dwarn:0   dfail:0   fail:0   skip:28  time:546s
fi-snb-2600      total:279  pass:250  dwarn:0   dfail:0   fail:0   skip:29  time:408s

3adc9e3cacefe82619fc56ec78a0ca3a67a4b00c drm-tip: 2017y-08m-23d-21h-35m-35s UTC integration manifest bcab325ea91a drm/i915/cnl: WaForceEnableNonCoherent

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_5480/
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#2 lore raw
In reply to Rodrigo Vivi
Quoted 4 lines · Rodrigo Vivi

Must Force Non-Coherent whenever executing a 3D context. This is a workaround for a possible hang in the unlikely event
a TLB invalidation occurs during a PSD flush.
Set masked bit 4 in 0x7300 during boot.

This bug should not be present in HW anymore. A different reason to keep doing this is performance, though.

Quoted 14 lines
Cc Mika Kuoppala <[email protected]>
Cc Oscar Mateo <[email protected]>
---
  drivers/gpu/drm/i915/intel_engine_cs.c | 5 ++++-
  1 file changed, 4 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_engine_cs.c +0 -0
index a6ac9d0a4156..7dfc78b038c4 100644
@@ -1071,8 +1071,11 @@ static int cnl_init_workarounds(struct intel_engine_cs *engine)
  	int ret;
 

drivers/gpu/drm/i915/intel_engine_cs.c static int cnl_init_workarounds(struct intel_engine_cs *engine) · line 1081

Quoted 10 lines
drivers/gpu/drm/i915/intel_engine_cs.c +4 -1
  	/* WaForceContextSaveRestoreNonCoherent:cnl */
+	/* WaForceEnableNonCoherent:cnl */
  	WA_SET_BIT_MASKED(CNL_HDC_CHICKEN0,
-			  HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT);
+			  HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT |
+			  HDC_FORCE_NON_COHERENT);
+
 
  	/* WaDisableReplayBufferBankArbitrationOptimization:cnl */
  	WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2,

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

#3 lore raw
Quoted 10 lines · Oscar Mateo
Quoted 4 lines · Rodrigo Vivi

Must Force Non-Coherent whenever executing a 3D context. This is a workaround for a possible hang in the unlikely event
a TLB invalidation occurs during a PSD flush.
Set masked bit 4 in 0x7300 during boot.

This bug should not be present in HW anymore. A different reason to keep doing this is performance, though.

Doesn't userspace already has a choice between coherent and non-coherent access. Why would we want to cheat, and force non-coherent when it clearly wants to use coherent one?
(well, because that's what we've been always doing is one reason, the other one may be that "clearly" can sometimes turn out to be "by accident")

-Michał

Quoted 20 lines
Quoted 20 lines
Cc Mika Kuoppala <[email protected]>
Cc Oscar Mateo <[email protected]>
---
  drivers/gpu/drm/i915/intel_engine_cs.c | 5 ++++-
  1 file changed, 4 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_engine_cs.c +3 -1
index a6ac9d0a4156..7dfc78b038c4 100644
@@ -1071,8 +1071,11 @@ static int cnl_init_workarounds(struct intel_engine_cs *engine)
  	int ret;
  	/* WaForceContextSaveRestoreNonCoherent:cnl */
+	/* WaForceEnableNonCoherent:cnl */
  	WA_SET_BIT_MASKED(CNL_HDC_CHICKEN0,
-			  HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT);
+			  HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT |
+			  HDC_FORCE_NON_COHERENT);
Quoted 8 lines

drivers/gpu/drm/i915/intel_engine_cs.c static int cnl_init_workarounds(struct intel_engine_cs *engine) · line 1081

Quoted 3 lines
drivers/gpu/drm/i915/intel_engine_cs.c +1 -0
+
  	/* WaDisableReplayBufferBankArbitrationOptimization:cnl */
  	WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2,

_______________________________________________
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

#4 lore raw
Quoted 9 lines · Michał Winiarski
Quoted 8 lines · Oscar Mateo
Quoted 4 lines · Rodrigo Vivi

Must Force Non-Coherent whenever executing a 3D context. This is a workaround for a possible hang in the unlikely event
a TLB invalidation occurs during a PSD flush.
Set masked bit 4 in 0x7300 during boot.

Quoted 9 lines
Quoted 2 lines

This bug should not be present in HW anymore. A different reason to keep doing this is performance, though.

Doesn't userspace already has a choice between coherent and non-coherent access. Why would we want to cheat, and force non-coherent when it clearly wants to use coherent one?
(well, because that's what we've been always doing is one reason, the other one may be that "clearly" can sometimes turn out to be "by accident")

Caveat for below is that I haven't investigated the cnl particulars. But we did end up having system hangs on skl in the past by mismatching this, see commit e238659ddd889a5d3fbdfa1a2ab120f90404bf41

-Mika

Quoted 30 lines

-Michał

Quoted 20 lines
Quoted 20 lines
Cc Mika Kuoppala <[email protected]>
Cc Oscar Mateo <[email protected]>
---
  drivers/gpu/drm/i915/intel_engine_cs.c | 5 ++++-
  1 file changed, 4 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_engine_cs.c +3 -1
index a6ac9d0a4156..7dfc78b038c4 100644
@@ -1071,8 +1071,11 @@ static int cnl_init_workarounds(struct intel_engine_cs *engine)
  	int ret;
  	/* WaForceContextSaveRestoreNonCoherent:cnl */
+	/* WaForceEnableNonCoherent:cnl */
  	WA_SET_BIT_MASKED(CNL_HDC_CHICKEN0,
-			  HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT);
+			  HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT |
+			  HDC_FORCE_NON_COHERENT);
Quoted 8 lines

drivers/gpu/drm/i915/intel_engine_cs.c static int cnl_init_workarounds(struct intel_engine_cs *engine) · line 1081

Quoted 3 lines
drivers/gpu/drm/i915/intel_engine_cs.c +1 -0
+
  	/* WaDisableReplayBufferBankArbitrationOptimization:cnl */
  	WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2,

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

Quoted 4 lines

_______________________________________________
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

#5 lore raw
In reply to Michał Winiarski
Quoted 7 lines · Michał Winiarski
Quoted 6 lines · Oscar Mateo
Quoted 4 lines · Rodrigo Vivi

Must Force Non-Coherent whenever executing a 3D context. This is a workaround for a possible hang in the unlikely event
a TLB invalidation occurs during a PSD flush.
Set masked bit 4 in 0x7300 during boot.

Quoted 5 lines
Quoted 2 lines

This bug should not be present in HW anymore. A different reason to keep doing this is performance, though.

Doesn't userspace already has a choice between coherent and non-coherent access. Why would we want to cheat, and force non-coherent when it clearly wants to use coherent one?

But does it? I cannot find anything in i915 to allow userspace to flip this chicken bit (the register is not whitelisted and we don't have any execbuf flag/hint). Or am I missing something?

Quoted 31 lines

(well, because that's what we've been always doing is one reason, the other one may be that "clearly" can sometimes turn out to be "by accident")

-Michał

Quoted 26 lines
Quoted 12 lines
Cc Mika Kuoppala <[email protected]>
Cc Oscar Mateo <[email protected]>
---
   drivers/gpu/drm/i915/intel_engine_cs.c | 5 ++++-
   1 file changed, 4 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_engine_cs.c +0 -0
index a6ac9d0a4156..7dfc78b038c4 100644

drivers/gpu/drm/i915/intel_engine_cs.c static int cnl_init_workarounds(struct intel_engine_cs *engine) · line 1079

Quoted 10 lines
drivers/gpu/drm/i915/intel_engine_cs.c +4 -1
@@ -1071,8 +1071,11 @@ static int cnl_init_workarounds(struct intel_engine_cs *engine)
   	int ret;
   	/* WaForceContextSaveRestoreNonCoherent:cnl */
+	/* WaForceEnableNonCoherent:cnl */
   	WA_SET_BIT_MASKED(CNL_HDC_CHICKEN0,
-			  HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT);
+			  HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT |
+			  HDC_FORCE_NON_COHERENT);
+
   	/* WaDisableReplayBufferBankArbitrationOptimization:cnl */
   	WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2,

_______________________________________________
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

#6 lore raw
Quoted 12 lines · Oscar Mateo
Quoted 9 lines · Michał Winiarski
Quoted 8 lines · Oscar Mateo
Quoted 4 lines · Rodrigo Vivi

Must Force Non-Coherent whenever executing a 3D context. This is a workaround for a possible hang in the unlikely event
a TLB invalidation occurs during a PSD flush.
Set masked bit 4 in 0x7300 during boot.

This bug should not be present in HW anymore. A different reason to keep doing this is performance, though.

Quoted 7 lines
Quoted 3 lines

Doesn't userspace already has a choice between coherent and non-coherent access. Why would we want to cheat, and force non-coherent when it clearly wants to use coherent one?

But does it? I cannot find anything in i915 to allow userspace to flip this chicken bit (the register is not whitelisted and we don't have any execbuf flag/hint). Or am I missing something?

It's a hardware interface, not i915 interface.
See: IHD-OS-SKL-Vol7-05.16
Shared Functions -> Overview of memory accesses (Cache behavior) (page 144)
https://01.org/sites/default/files/documentation/intel-gfx-prm-osrc-skl-vol07-3d_media_gpgpu.pdf

The workaround is removing the ability to choose coherency model - even if userspace is opting in for coherent path, the HW will use non-coherent one.

-Michał

Quoted 24 lines
Quoted 24 lines

(well, because that's what we've been always doing is one reason, the other one may be that "clearly" can sometimes turn out to be "by accident")

-Michał

Quoted 18 lines
Quoted 18 lines
Cc Mika Kuoppala <[email protected]>
Cc Oscar Mateo <[email protected]>
---
   drivers/gpu/drm/i915/intel_engine_cs.c | 5 ++++-
   1 file changed, 4 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_engine_cs.c +2 -1
index a6ac9d0a4156..7dfc78b038c4 100644
@@ -1071,8 +1071,11 @@ static int cnl_init_workarounds(struct intel_engine_cs *engine)
   	int ret;
   	/* WaForceContextSaveRestoreNonCoherent:cnl */
+	/* WaForceEnableNonCoherent:cnl */
   	WA_SET_BIT_MASKED(CNL_HDC_CHICKEN0,
-			  HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT);
+			  HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT |
Quoted 9 lines
Quoted 8 lines
Quoted 8 lines

drivers/gpu/drm/i915/intel_engine_cs.c static int cnl_init_workarounds(struct intel_engine_cs *engine) · line 1081

Quoted 4 lines
drivers/gpu/drm/i915/intel_engine_cs.c +2 -0
+			  HDC_FORCE_NON_COHERENT);
+
   	/* WaDisableReplayBufferBankArbitrationOptimization:cnl */
   	WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2,

_______________________________________________
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