Skip to contentSkip to conversation

drm/i915/icl: Fix context slice count configuration

PATCH In review Tvrtko Ursulin 4 participants 14 messages Aug 22, 2018 – Aug 29, 2018

Subsystems (2)
lore
Share & tools
lore

Tvrtko Ursulin

author maintainer
#0 lore raw

PATCH 1/1 drm/i915/icl: Fix context slice count configuration

From: Tvrtko Ursulin <[email protected]>

Bitfield width for configuring the active slice count has grown in Gen11 so we need to program the GEN8_R_PWR_CLK_STATE accordingly.

Current code was always requesting eight times the number of slices (due writting to a bitfield starting three bits higher than it should). These requests were luckily a) capped by the hardware to the available number of slices, and b) we haven't yet exported the code to ask for reduced slice configurations.

Due both of the above there was no impact from this incorrect programming but we should still fix it.

Bspec: 12247

---
 drivers/gpu/drm/i915/i915_reg.h  |  2 ++
 drivers/gpu/drm/i915/intel_lrc.c | 12 ++++++++----
 2 files changed, 10 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_reg.h +2 -0
index 59d06d0055bb..640f7b774a26 100644
@@ -344,6 +344,8 @@ static inline bool i915_mmio_reg_valid(i915_reg_t reg)
 #define   GEN8_RPCS_S_CNT_ENABLE	(1 << 18)
 #define   GEN8_RPCS_S_CNT_SHIFT		15
 #define   GEN8_RPCS_S_CNT_MASK		(0x7 << GEN8_RPCS_S_CNT_SHIFT)
+#define   GEN11_RPCS_S_CNT_SHIFT	12
+#define   GEN11_RPCS_S_CNT_MASK		(0x3f << GEN11_RPCS_S_CNT_SHIFT)
 #define   GEN8_RPCS_SS_CNT_ENABLE	(1 << 11)
 #define   GEN8_RPCS_SS_CNT_SHIFT	8
 #define   GEN8_RPCS_SS_CNT_MASK		(0x7 << GEN8_RPCS_SS_CNT_SHIFT)
drivers/gpu/drm/i915/intel_lrc.c +8 -4
index 36050f085071..43b8b0675ba0 100644
@@ -2501,10 +2501,14 @@ make_rpcs(struct drm_i915_private *dev_priv)
 	 * enablement.
 	*/
 	if (INTEL_INFO(dev_priv)->sseu.has_slice_pg) {
-		rpcs |= GEN8_RPCS_S_CNT_ENABLE;
-		rpcs |= hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask) <<
-			GEN8_RPCS_S_CNT_SHIFT;
-		rpcs |= GEN8_RPCS_ENABLE;
+		rpcs = hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask);
+
+		if (INTEL_GEN(dev_priv) >= 11)
+			rpcs <<= GEN11_RPCS_S_CNT_SHIFT;
+		else
+			rpcs <<= GEN8_RPCS_S_CNT_SHIFT;
+
+		rpcs |= GEN8_RPCS_ENABLE | GEN8_RPCS_S_CNT_ENABLE;
 	}
 
 	if (INTEL_INFO(dev_priv)->sseu.has_subslice_pg) {
Signature · 6 lines
2.17.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw
Quoted 48 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

Bitfield width for configuring the active slice count has grown in Gen11 so we need to program the GEN8_R_PWR_CLK_STATE accordingly.

Current code was always requesting eight times the number of slices (due writting to a bitfield starting three bits higher than it should). These requests were luckily a) capped by the hardware to the available number of slices, and b) we haven't yet exported the code to ask for reduced slice configurations.

Due both of the above there was no impact from this incorrect programming but we should still fix it.

Bspec: 12247

---
  drivers/gpu/drm/i915/i915_reg.h  |  2 ++
  drivers/gpu/drm/i915/intel_lrc.c | 12 ++++++++----
  2 files changed, 10 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_reg.h +2 -0
index 59d06d0055bb..640f7b774a26 100644
@@ -344,6 +344,8 @@ static inline bool i915_mmio_reg_valid(i915_reg_t reg)
  #define   GEN8_RPCS_S_CNT_ENABLE	(1 << 18)
  #define   GEN8_RPCS_S_CNT_SHIFT		15
  #define   GEN8_RPCS_S_CNT_MASK		(0x7 << GEN8_RPCS_S_CNT_SHIFT)
+#define   GEN11_RPCS_S_CNT_SHIFT	12
+#define   GEN11_RPCS_S_CNT_MASK		(0x3f << GEN11_RPCS_S_CNT_SHIFT)
  #define   GEN8_RPCS_SS_CNT_ENABLE	(1 << 11)
  #define   GEN8_RPCS_SS_CNT_SHIFT	8
  #define   GEN8_RPCS_SS_CNT_MASK		(0x7 << GEN8_RPCS_SS_CNT_SHIFT)
drivers/gpu/drm/i915/intel_lrc.c +0 -2
index 36050f085071..43b8b0675ba0 100644
@@ -2501,10 +2501,14 @@ make_rpcs(struct drm_i915_private *dev_priv)
  	 * enablement.
  	*/
  	if (INTEL_INFO(dev_priv)->sseu.has_slice_pg) {
-		rpcs |= GEN8_RPCS_S_CNT_ENABLE;
-		rpcs |= hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask) <<

drivers/gpu/drm/i915/intel_lrc.c make_rpcs(struct drm_i915_private *dev_priv) · line 2511

Quoted 10 lines
drivers/gpu/drm/i915/intel_lrc.c +8 -2
-			GEN8_RPCS_S_CNT_SHIFT;
-		rpcs |= GEN8_RPCS_ENABLE;
+		rpcs = hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask);
+
+		if (INTEL_GEN(dev_priv) >= 11)
+			rpcs <<= GEN11_RPCS_S_CNT_SHIFT;
+		else
+			rpcs <<= GEN8_RPCS_S_CNT_SHIFT;
+
+		rpcs |= GEN8_RPCS_ENABLE | GEN8_RPCS_S_CNT_ENABLE;

I don't know if you saw that wording in the documentation :

 "

Note: In ICL, software programs this register as if GT consists of 2 slices with 4 subslices in each slice. Hardware maps this to the 1 slice/8-subslice physical layout.

"

My understanding is that it would make this function a bit more complicated ;)

It also implies that on ICL you cannot select 3 subslices, which is unfortunately what Tony was trying to do.

Maybe some opens need to be raised as to what's possible on ICL.

drivers/gpu/drm/i915/intel_lrc.c make_rpcs(struct drm_i915_private *dev_priv) · line 2514

Quoted 3 lines
drivers/gpu/drm/i915/intel_lrc.c +0 -0
  	}
 
  	if (INTEL_INFO(dev_priv)->sseu.has_subslice_pg) {

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

#2 lore raw
Quoted 68 lines · Lionel Landwerlin
Quoted 49 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

Bitfield width for configuring the active slice count has grown in Gen11 so we need to program the GEN8_R_PWR_CLK_STATE accordingly.

Current code was always requesting eight times the number of slices (due writting to a bitfield starting three bits higher than it should). These requests were luckily a) capped by the hardware to the available number of slices, and b) we haven't yet exported the code to ask for reduced slice configurations.

Due both of the above there was no impact from this incorrect programming but we should still fix it.

Bspec: 12247

---
  drivers/gpu/drm/i915/i915_reg.h  |  2 ++ drivers/gpu/drm/i915/intel_lrc.c | 12 ++++++++----

  2 files changed, 10 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_reg.h +2 -0
index 59d06d0055bb..640f7b774a26 100644
@@ -344,6 +344,8 @@ static inline bool i915_mmio_reg_valid(i915_reg_t 
 reg)
   #define   GEN8_RPCS_S_CNT_ENABLE    (1 << 18)
   #define   GEN8_RPCS_S_CNT_SHIFT        15
   #define   GEN8_RPCS_S_CNT_MASK        (0x7 << GEN8_RPCS_S_CNT_SHIFT)
+#define   GEN11_RPCS_S_CNT_SHIFT    12
+#define   GEN11_RPCS_S_CNT_MASK        (0x3f << GEN11_RPCS_S_CNT_SHIFT)
   #define   GEN8_RPCS_SS_CNT_ENABLE    (1 << 11)
   #define   GEN8_RPCS_SS_CNT_SHIFT    8

  #define   GEN8_RPCS_SS_CNT_MASK        (0x7 << GEN8_RPCS_SS_CNT_SHIFT)

drivers/gpu/drm/i915/intel_lrc.c +0 -2
index 36050f085071..43b8b0675ba0 100644
@@ -2501,10 +2501,14 @@ make_rpcs(struct drm_i915_private *dev_priv)
        * enablement.
       */
       if (INTEL_INFO(dev_priv)->sseu.has_slice_pg) {
-        rpcs |= GEN8_RPCS_S_CNT_ENABLE;
-        rpcs |= hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask) <<

drivers/gpu/drm/i915/intel_lrc.c make_rpcs(struct drm_i915_private *dev_priv) · line 2511

Quoted 10 lines
drivers/gpu/drm/i915/intel_lrc.c +8 -2
-            GEN8_RPCS_S_CNT_SHIFT;
-        rpcs |= GEN8_RPCS_ENABLE;
+        rpcs = hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask);
+
+        if (INTEL_GEN(dev_priv) >= 11)
+            rpcs <<= GEN11_RPCS_S_CNT_SHIFT;
+        else
+            rpcs <<= GEN8_RPCS_S_CNT_SHIFT;
+
+        rpcs |= GEN8_RPCS_ENABLE | GEN8_RPCS_S_CNT_ENABLE;

I don't know if you saw that wording in the documentation :

 "

Note: In ICL, software programs this register as if GT consists of 2 slices with 4 subslices in each slice. Hardware maps this to the 1 slice/8-subslice physical layout.

Quoted 8 lines

"

My understanding is that it would make this function a bit more complicated ;)

It also implies that on ICL you cannot select 3 subslices, which is unfortunately what Tony was trying to do.

Maybe some opens need to be raised as to what's possible on ICL.

I interpreted the note in my head as "In ICL, _if_ _the_ software programs.." so did not see a problem. Thought that would be just some hidden/under the covers remapping hw would do. But I see now that was wrong, and you are most likely right. I'll try to do some digging to understand this better.

But for the second part of it, I don't see why 1x3 configuration would be illegal. If software must assume hw is 2x4, even if in reality it is 1x8, then 1x3 would still be legal, no?

I thought the cause of the hang on ICL was that when Tony was asking for 1x3, the code was actually programming a request for 8x3 - which is illegal (as in slice count must be 1 to enable subslice pg) and so would fail to actually turn of the unwanted subslices.

But then I also though on ICL we deal with masks and not counts when programming the hardware. Since apparently it is counts both for slices and subslices, I am mightily confused as to how media-driver would even theoretically be able to turn off a _specific_ (sub)slice?!

Regards,

Tvrtko

Quoted 9 lines

drivers/gpu/drm/i915/intel_lrc.c make_rpcs(struct drm_i915_private *dev_priv) · line 2514

Quoted 3 lines
drivers/gpu/drm/i915/intel_lrc.c +0 -0
       }
       if (INTEL_INFO(dev_priv)->sseu.has_subslice_pg) {

_______________________________________________
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

#3 lore raw
Quoted 76 lines · Tvrtko Ursulin
Quoted 74 lines · Lionel Landwerlin
Quoted 50 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

Bitfield width for configuring the active slice count has grown in Gen11 so we need to program the GEN8_R_PWR_CLK_STATE accordingly.

Current code was always requesting eight times the number of slices (due writting to a bitfield starting three bits higher than it should). These requests were luckily a) capped by the hardware to the available number of slices, and b) we haven't yet exported the code to ask for reduced slice configurations.

Due both of the above there was no impact from this incorrect programming but we should still fix it.

Bspec: 12247

---
  drivers/gpu/drm/i915/i915_reg.h  |  2 ++ drivers/gpu/drm/i915/intel_lrc.c | 12 ++++++++----

  2 files changed, 10 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_reg.h +2 -0
index 59d06d0055bb..640f7b774a26 100644
@@ -344,6 +344,8 @@ static inline bool 
 i915_mmio_reg_valid(i915_reg_t reg)
   #define   GEN8_RPCS_S_CNT_ENABLE    (1 << 18)
   #define   GEN8_RPCS_S_CNT_SHIFT        15
   #define   GEN8_RPCS_S_CNT_MASK        (0x7 << GEN8_RPCS_S_CNT_SHIFT)
+#define   GEN11_RPCS_S_CNT_SHIFT    12
+#define   GEN11_RPCS_S_CNT_MASK        (0x3f << 
 GEN11_RPCS_S_CNT_SHIFT)
   #define   GEN8_RPCS_SS_CNT_ENABLE    (1 << 11)

  #define   GEN8_RPCS_SS_CNT_SHIFT    8
  #define   GEN8_RPCS_SS_CNT_MASK        (0x7 << GEN8_RPCS_SS_CNT_SHIFT)

drivers/gpu/drm/i915/intel_lrc.c +0 -2
index 36050f085071..43b8b0675ba0 100644
@@ -2501,10 +2501,14 @@ make_rpcs(struct drm_i915_private *dev_priv)
        * enablement.
       */
       if (INTEL_INFO(dev_priv)->sseu.has_slice_pg) {
-        rpcs |= GEN8_RPCS_S_CNT_ENABLE;
-        rpcs |= hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask) <<

drivers/gpu/drm/i915/intel_lrc.c make_rpcs(struct drm_i915_private *dev_priv) · line 2511

Quoted 10 lines
drivers/gpu/drm/i915/intel_lrc.c +8 -2
-            GEN8_RPCS_S_CNT_SHIFT;
-        rpcs |= GEN8_RPCS_ENABLE;
+        rpcs = hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask);
+
+        if (INTEL_GEN(dev_priv) >= 11)
+            rpcs <<= GEN11_RPCS_S_CNT_SHIFT;
+        else
+            rpcs <<= GEN8_RPCS_S_CNT_SHIFT;
+
+        rpcs |= GEN8_RPCS_ENABLE | GEN8_RPCS_S_CNT_ENABLE;

I don't know if you saw that wording in the documentation :

  "

Note: In ICL, software programs this register as if GT consists of 2 slices with 4 subslices in each slice. Hardware maps this to the 1 slice/8-subslice physical layout.

"

My understanding is that it would make this function a bit more complicated ;)

Quoted 7 lines
Quoted 3 lines

It also implies that on ICL you cannot select 3 subslices, which is unfortunately what Tony was trying to do.

Maybe some opens need to be raised as to what's possible on ICL.

I interpreted the note in my head as "In ICL, _if_ _the_ software programs.." so did not see a problem. Thought that would be just some hidden/under the covers remapping hw would do. But I see now that was wrong, and you are most likely right. I'll try to do some digging to understand this better.

But for the second part of it, I don't see why 1x3 configuration would be illegal. If software must assume hw is 2x4, even if in reality it is 1x8, then 1x3 would still be legal, no?

We still seem to put a subslice number in the register for ICL (values being 0b001, 0b010, 0b011 & 0b100).

My understanding is that the hardware will just multiply that value by 2 to map to the 1x8 underlying topology.

So if that's the case, you can't really do odd numbers... ¯\_(ツ)_/¯

Quoted 4 lines

I thought the cause of the hang on ICL was that when Tony was asking for 1x3, the code was actually programming a request for 8x3 - which is illegal (as in slice count must be 1 to enable subslice pg) and so would fail to actually turn of the unwanted subslices.

But then I also though on ICL we deal with masks and not counts when programming the hardware. Since apparently it is counts both for slices and subslices, I am mightily confused as to how media-driver would even theoretically be able to turn off a _specific_ (sub)slice?!

The fact that the feature needed isn't implemented at by the thread dispatcher is really strange to me too.

It sounds like we're forced to use a bigger hammer than what we really need.

As to how that maps to the right subslices is also unknown to me.

The only explanation I have is that subslices with no VME samplers get turn off first in the list of subslices to turn off.

Cheers,

-

Lionel

Quoted 15 lines

Regards,

Tvrtko

Quoted 9 lines

drivers/gpu/drm/i915/intel_lrc.c make_rpcs(struct drm_i915_private *dev_priv) · line 2514

Quoted 3 lines
drivers/gpu/drm/i915/intel_lrc.c +0 -0
       }
       if (INTEL_INFO(dev_priv)->sseu.has_subslice_pg) {

_______________________________________________
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
In reply to Lionel Landwerlin
Quoted 68 lines · Lionel Landwerlin
Quoted 49 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

Bitfield width for configuring the active slice count has grown in Gen11 so we need to program the GEN8_R_PWR_CLK_STATE accordingly.

Current code was always requesting eight times the number of slices (due writting to a bitfield starting three bits higher than it should). These requests were luckily a) capped by the hardware to the available number of slices, and b) we haven't yet exported the code to ask for reduced slice configurations.

Due both of the above there was no impact from this incorrect programming but we should still fix it.

Bspec: 12247

---
  drivers/gpu/drm/i915/i915_reg.h  |  2 ++ drivers/gpu/drm/i915/intel_lrc.c | 12 ++++++++----

  2 files changed, 10 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_reg.h +2 -0
index 59d06d0055bb..640f7b774a26 100644
@@ -344,6 +344,8 @@ static inline bool i915_mmio_reg_valid(i915_reg_t 
 reg)
   #define   GEN8_RPCS_S_CNT_ENABLE    (1 << 18)
   #define   GEN8_RPCS_S_CNT_SHIFT        15
   #define   GEN8_RPCS_S_CNT_MASK        (0x7 << GEN8_RPCS_S_CNT_SHIFT)
+#define   GEN11_RPCS_S_CNT_SHIFT    12
+#define   GEN11_RPCS_S_CNT_MASK        (0x3f << GEN11_RPCS_S_CNT_SHIFT)
   #define   GEN8_RPCS_SS_CNT_ENABLE    (1 << 11)
   #define   GEN8_RPCS_SS_CNT_SHIFT    8

  #define   GEN8_RPCS_SS_CNT_MASK        (0x7 << GEN8_RPCS_SS_CNT_SHIFT)

drivers/gpu/drm/i915/intel_lrc.c +0 -2
index 36050f085071..43b8b0675ba0 100644
@@ -2501,10 +2501,14 @@ make_rpcs(struct drm_i915_private *dev_priv)
        * enablement.
       */
       if (INTEL_INFO(dev_priv)->sseu.has_slice_pg) {
-        rpcs |= GEN8_RPCS_S_CNT_ENABLE;
-        rpcs |= hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask) <<

drivers/gpu/drm/i915/intel_lrc.c make_rpcs(struct drm_i915_private *dev_priv) · line 2511

Quoted 10 lines
drivers/gpu/drm/i915/intel_lrc.c +8 -2
-            GEN8_RPCS_S_CNT_SHIFT;
-        rpcs |= GEN8_RPCS_ENABLE;
+        rpcs = hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask);
+
+        if (INTEL_GEN(dev_priv) >= 11)
+            rpcs <<= GEN11_RPCS_S_CNT_SHIFT;
+        else
+            rpcs <<= GEN8_RPCS_S_CNT_SHIFT;
+
+        rpcs |= GEN8_RPCS_ENABLE | GEN8_RPCS_S_CNT_ENABLE;

I don't know if you saw that wording in the documentation :

 "

Note: In ICL, software programs this register as if GT consists of 2 slices with 4 subslices in each slice. Hardware maps this to the 1 slice/8-subslice physical layout.

Quoted 8 lines

"

My understanding is that it would make this function a bit more complicated ;)

It also implies that on ICL you cannot select 3 subslices, which is unfortunately what Tony was trying to do.

Maybe some opens need to be raised as to what's possible on ICL.

Happy to r-b this one since we clarified it is fine?

Regards,

Tvrtko

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

#5 lore raw
Quoted 78 lines · Tvrtko Ursulin
Quoted 76 lines · Lionel Landwerlin
Quoted 50 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

Bitfield width for configuring the active slice count has grown in Gen11 so we need to program the GEN8_R_PWR_CLK_STATE accordingly.

Current code was always requesting eight times the number of slices (due writting to a bitfield starting three bits higher than it should). These requests were luckily a) capped by the hardware to the available number of slices, and b) we haven't yet exported the code to ask for reduced slice configurations.

Due both of the above there was no impact from this incorrect programming but we should still fix it.

Bspec: 12247

---
  drivers/gpu/drm/i915/i915_reg.h  |  2 ++ drivers/gpu/drm/i915/intel_lrc.c | 12 ++++++++----

  2 files changed, 10 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_reg.h +2 -0
index 59d06d0055bb..640f7b774a26 100644
@@ -344,6 +344,8 @@ static inline bool 
 i915_mmio_reg_valid(i915_reg_t reg)
   #define   GEN8_RPCS_S_CNT_ENABLE    (1 << 18)
   #define   GEN8_RPCS_S_CNT_SHIFT        15
   #define   GEN8_RPCS_S_CNT_MASK        (0x7 << GEN8_RPCS_S_CNT_SHIFT)
+#define   GEN11_RPCS_S_CNT_SHIFT    12
+#define   GEN11_RPCS_S_CNT_MASK        (0x3f << 
 GEN11_RPCS_S_CNT_SHIFT)
   #define   GEN8_RPCS_SS_CNT_ENABLE    (1 << 11)

  #define   GEN8_RPCS_SS_CNT_SHIFT    8
  #define   GEN8_RPCS_SS_CNT_MASK        (0x7 << GEN8_RPCS_SS_CNT_SHIFT)

drivers/gpu/drm/i915/intel_lrc.c +0 -2
index 36050f085071..43b8b0675ba0 100644
@@ -2501,10 +2501,14 @@ make_rpcs(struct drm_i915_private *dev_priv)
        * enablement.
       */
       if (INTEL_INFO(dev_priv)->sseu.has_slice_pg) {
-        rpcs |= GEN8_RPCS_S_CNT_ENABLE;
-        rpcs |= hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask) <<

drivers/gpu/drm/i915/intel_lrc.c make_rpcs(struct drm_i915_private *dev_priv) · line 2511

Quoted 10 lines
drivers/gpu/drm/i915/intel_lrc.c +8 -2
-            GEN8_RPCS_S_CNT_SHIFT;
-        rpcs |= GEN8_RPCS_ENABLE;
+        rpcs = hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask);
+
+        if (INTEL_GEN(dev_priv) >= 11)
+            rpcs <<= GEN11_RPCS_S_CNT_SHIFT;
+        else
+            rpcs <<= GEN8_RPCS_S_CNT_SHIFT;
+
+        rpcs |= GEN8_RPCS_ENABLE | GEN8_RPCS_S_CNT_ENABLE;

I don't know if you saw that wording in the documentation :

  "

Note: In ICL, software programs this register as if GT consists of 2 slices with 4 subslices in each slice. Hardware maps this to the 1 slice/8-subslice physical layout.

"

My understanding is that it would make this function a bit more complicated ;)

It also implies that on ICL you cannot select 3 subslices, which is unfortunately what Tony was trying to do.

Quoted 9 lines
Quoted 1 line

Maybe some opens need to be raised as to what's possible on ICL.

Happy to r-b this one since we clarified it is fine?

Regards,

Tvrtko

I think there is still an issue here with regard to the subslice programming.
You can only program values in [1, 4].
But because we expose up to 8 subslices, you need to take that mask and alter the slice mask based on its value.

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

#6 lore raw
Quoted 83 lines · Lionel Landwerlin
Quoted 82 lines · Tvrtko Ursulin
Quoted 69 lines · Lionel Landwerlin
Quoted 50 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

Bitfield width for configuring the active slice count has grown in Gen11 so we need to program the GEN8_R_PWR_CLK_STATE accordingly.

Current code was always requesting eight times the number of slices (due writting to a bitfield starting three bits higher than it should). These requests were luckily a) capped by the hardware to the available number of slices, and b) we haven't yet exported the code to ask for reduced slice configurations.

Due both of the above there was no impact from this incorrect programming but we should still fix it.

Bspec: 12247

---
  drivers/gpu/drm/i915/i915_reg.h  |  2 ++ drivers/gpu/drm/i915/intel_lrc.c | 12 ++++++++----

  2 files changed, 10 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_reg.h +2 -0
index 59d06d0055bb..640f7b774a26 100644
@@ -344,6 +344,8 @@ static inline bool 
 i915_mmio_reg_valid(i915_reg_t reg)
   #define   GEN8_RPCS_S_CNT_ENABLE    (1 << 18)
   #define   GEN8_RPCS_S_CNT_SHIFT        15
   #define   GEN8_RPCS_S_CNT_MASK        (0x7 << GEN8_RPCS_S_CNT_SHIFT)
+#define   GEN11_RPCS_S_CNT_SHIFT    12
+#define   GEN11_RPCS_S_CNT_MASK        (0x3f << 
 GEN11_RPCS_S_CNT_SHIFT)
   #define   GEN8_RPCS_SS_CNT_ENABLE    (1 << 11)

  #define   GEN8_RPCS_SS_CNT_SHIFT    8
  #define   GEN8_RPCS_SS_CNT_MASK        (0x7 << GEN8_RPCS_SS_CNT_SHIFT)

drivers/gpu/drm/i915/intel_lrc.c +0 -2
index 36050f085071..43b8b0675ba0 100644
@@ -2501,10 +2501,14 @@ make_rpcs(struct drm_i915_private *dev_priv)
        * enablement.
       */
       if (INTEL_INFO(dev_priv)->sseu.has_slice_pg) {
-        rpcs |= GEN8_RPCS_S_CNT_ENABLE;
-        rpcs |= hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask) <<

drivers/gpu/drm/i915/intel_lrc.c make_rpcs(struct drm_i915_private *dev_priv) · line 2511

Quoted 10 lines
drivers/gpu/drm/i915/intel_lrc.c +8 -2
-            GEN8_RPCS_S_CNT_SHIFT;
-        rpcs |= GEN8_RPCS_ENABLE;
+        rpcs = hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask);
+
+        if (INTEL_GEN(dev_priv) >= 11)
+            rpcs <<= GEN11_RPCS_S_CNT_SHIFT;
+        else
+            rpcs <<= GEN8_RPCS_S_CNT_SHIFT;
+
+        rpcs |= GEN8_RPCS_ENABLE | GEN8_RPCS_S_CNT_ENABLE;

I don't know if you saw that wording in the documentation :

  "

Note: In ICL, software programs this register as if GT consists of 2 slices with 4 subslices in each slice. Hardware maps this to the 1 slice/8-subslice physical layout.

Quoted 8 lines

"

My understanding is that it would make this function a bit more complicated ;)

It also implies that on ICL you cannot select 3 subslices, which is unfortunately what Tony was trying to do.

Maybe some opens need to be raised as to what's possible on ICL.

Happy to r-b this one since we clarified it is fine?

Quoted 8 lines
Quoted 5 lines

Regards,

Tvrtko

I think there is still an issue here with regard to the subslice programming.
You can only program values in [1, 4].
But because we expose up to 8 subslices, you need to take that mask and alter the slice mask based on its value.

Hmm true, thanks. I think we need to somehow express this restriction on the uAPI level rather than silently mask it. Shall we reject attempts to configure subslice mask if hweight(mask) > max_subslices / 2 for ICL-LP?

Regards,

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

#7 lore raw
Quoted 87 lines · Tvrtko Ursulin
Quoted 85 lines · Lionel Landwerlin
Quoted 84 lines · Tvrtko Ursulin
Quoted 69 lines · Lionel Landwerlin
Quoted 50 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

Bitfield width for configuring the active slice count has grown in Gen11 so we need to program the GEN8_R_PWR_CLK_STATE accordingly.

Current code was always requesting eight times the number of slices (due writting to a bitfield starting three bits higher than it should). These requests were luckily a) capped by the hardware to the available number of slices, and b) we haven't yet exported the code to ask for reduced slice configurations.

Due both of the above there was no impact from this incorrect programming but we should still fix it.

Bspec: 12247

---
  drivers/gpu/drm/i915/i915_reg.h  |  2 ++ drivers/gpu/drm/i915/intel_lrc.c | 12 ++++++++----

  2 files changed, 10 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_reg.h +2 -0
index 59d06d0055bb..640f7b774a26 100644
@@ -344,6 +344,8 @@ static inline bool 
 i915_mmio_reg_valid(i915_reg_t reg)
   #define   GEN8_RPCS_S_CNT_ENABLE    (1 << 18)
   #define   GEN8_RPCS_S_CNT_SHIFT        15
   #define   GEN8_RPCS_S_CNT_MASK        (0x7 << GEN8_RPCS_S_CNT_SHIFT)
+#define   GEN11_RPCS_S_CNT_SHIFT    12
+#define   GEN11_RPCS_S_CNT_MASK        (0x3f << 
 GEN11_RPCS_S_CNT_SHIFT)
   #define   GEN8_RPCS_SS_CNT_ENABLE    (1 << 11)

  #define   GEN8_RPCS_SS_CNT_SHIFT    8
  #define   GEN8_RPCS_SS_CNT_MASK        (0x7 << GEN8_RPCS_SS_CNT_SHIFT)

drivers/gpu/drm/i915/intel_lrc.c +0 -2
index 36050f085071..43b8b0675ba0 100644
@@ -2501,10 +2501,14 @@ make_rpcs(struct drm_i915_private *dev_priv)
        * enablement.
       */
       if (INTEL_INFO(dev_priv)->sseu.has_slice_pg) {
-        rpcs |= GEN8_RPCS_S_CNT_ENABLE;
-        rpcs |= hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask) <<

drivers/gpu/drm/i915/intel_lrc.c make_rpcs(struct drm_i915_private *dev_priv) · line 2511

Quoted 10 lines
drivers/gpu/drm/i915/intel_lrc.c +8 -2
-            GEN8_RPCS_S_CNT_SHIFT;
-        rpcs |= GEN8_RPCS_ENABLE;
+        rpcs = hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask);
+
+        if (INTEL_GEN(dev_priv) >= 11)
+            rpcs <<= GEN11_RPCS_S_CNT_SHIFT;
+        else
+            rpcs <<= GEN8_RPCS_S_CNT_SHIFT;
+
+        rpcs |= GEN8_RPCS_ENABLE | GEN8_RPCS_S_CNT_ENABLE;

I don't know if you saw that wording in the documentation :

  "

Note: In ICL, software programs this register as if GT consists of 2 slices with 4 subslices in each slice. Hardware maps this to the 1 slice/8-subslice physical layout.

Quoted 8 lines

"

My understanding is that it would make this function a bit more complicated ;)

It also implies that on ICL you cannot select 3 subslices, which is unfortunately what Tony was trying to do.

Maybe some opens need to be raised as to what's possible on ICL.

Happy to r-b this one since we clarified it is fine?

Regards,

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

Tvrtko

I think there is still an issue here with regard to the subslice programming.
You can only program values in [1, 4].
But because we expose up to 8 subslices, you need to take that mask and alter the slice mask based on its value.

Hmm true, thanks. I think we need to somehow express this restriction on the uAPI level rather than silently mask it. Shall we reject attempts to configure subslice mask if hweight(mask) > max_subslices / 2 for ICL-LP?

No wait, I was in the dynamic sseu world, but this also applies to the current code base..

When we program SScount in drm-tip with the hweight of 8 we overflow the bitfield already and splat over SScountEn. Luckily with no ill effect since we set it anyway.

So the fix needs to be in make_rpcs today..

if (GEN11_LP) {
	if (hweight(slice_mask) == 1)
		subslice_count /= 2;
	else
		subslice_en = 0;
}

I think..

Regards,

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

#8 lore raw
Quoted 98 lines · Tvrtko Ursulin
Quoted 87 lines · Tvrtko Ursulin
Quoted 85 lines · Lionel Landwerlin
Quoted 84 lines · Tvrtko Ursulin
Quoted 69 lines · Lionel Landwerlin
Quoted 50 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

Bitfield width for configuring the active slice count has grown in Gen11 so we need to program the GEN8_R_PWR_CLK_STATE accordingly.

Current code was always requesting eight times the number of slices (due writting to a bitfield starting three bits higher than it should). These requests were luckily a) capped by the hardware to the available number of slices, and b) we haven't yet exported the code to ask for reduced slice configurations.

Due both of the above there was no impact from this incorrect programming but we should still fix it.

Bspec: 12247

---
  drivers/gpu/drm/i915/i915_reg.h  |  2 ++ drivers/gpu/drm/i915/intel_lrc.c | 12 ++++++++----

  2 files changed, 10 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_reg.h +2 -0
index 59d06d0055bb..640f7b774a26 100644
@@ -344,6 +344,8 @@ static inline bool 
 i915_mmio_reg_valid(i915_reg_t reg)
   #define   GEN8_RPCS_S_CNT_ENABLE    (1 << 18)
   #define   GEN8_RPCS_S_CNT_SHIFT        15
   #define   GEN8_RPCS_S_CNT_MASK        (0x7 << GEN8_RPCS_S_CNT_SHIFT)
+#define   GEN11_RPCS_S_CNT_SHIFT    12
+#define   GEN11_RPCS_S_CNT_MASK        (0x3f << 
 GEN11_RPCS_S_CNT_SHIFT)
   #define   GEN8_RPCS_SS_CNT_ENABLE    (1 << 11)

  #define   GEN8_RPCS_SS_CNT_SHIFT    8
  #define   GEN8_RPCS_SS_CNT_MASK        (0x7 << GEN8_RPCS_SS_CNT_SHIFT)

drivers/gpu/drm/i915/intel_lrc.c +0 -2
index 36050f085071..43b8b0675ba0 100644
@@ -2501,10 +2501,14 @@ make_rpcs(struct drm_i915_private *dev_priv)
        * enablement.
       */
       if (INTEL_INFO(dev_priv)->sseu.has_slice_pg) {
-        rpcs |= GEN8_RPCS_S_CNT_ENABLE;
-        rpcs |= hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask) <<

drivers/gpu/drm/i915/intel_lrc.c make_rpcs(struct drm_i915_private *dev_priv) · line 2511

Quoted 10 lines
drivers/gpu/drm/i915/intel_lrc.c +8 -2
-            GEN8_RPCS_S_CNT_SHIFT;
-        rpcs |= GEN8_RPCS_ENABLE;
+        rpcs = hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask);
+
+        if (INTEL_GEN(dev_priv) >= 11)
+            rpcs <<= GEN11_RPCS_S_CNT_SHIFT;
+        else
+            rpcs <<= GEN8_RPCS_S_CNT_SHIFT;
+
+        rpcs |= GEN8_RPCS_ENABLE | GEN8_RPCS_S_CNT_ENABLE;

I don't know if you saw that wording in the documentation :

  "

Note: In ICL, software programs this register as if GT consists of 2 slices with 4 subslices in each slice. Hardware maps this to the 1 slice/8-subslice physical layout.

Quoted 8 lines

"

My understanding is that it would make this function a bit more complicated ;)

It also implies that on ICL you cannot select 3 subslices, which is unfortunately what Tony was trying to do.

Maybe some opens need to be raised as to what's possible on ICL.

Happy to r-b this one since we clarified it is fine?

Regards,

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

Tvrtko

I think there is still an issue here with regard to the subslice programming.
You can only program values in [1, 4].
But because we expose up to 8 subslices, you need to take that mask and alter the slice mask based on its value.

Hmm true, thanks. I think we need to somehow express this restriction on the uAPI level rather than silently mask it. Shall we reject attempts to configure subslice mask if hweight(mask) > max_subslices / 2 for ICL-LP?

Quoted 8 lines

No wait, I was in the dynamic sseu world, but this also applies to the current code base..

When we program SScount in drm-tip with the hweight of 8 we overflow the bitfield already and splat over SScountEn. Luckily with no ill effect since we set it anyway.

So the fix needs to be in make_rpcs today..

if (GEN11_LP) {
    if (hweight(slice_mask) == 1)

&& hweight(subslice_mask) < max_subslices

So to leave the default full config alone.

Quoted 7 lines

        subslice_count /= 2;
    else
        subslice_en = 0;
}

I think..

Quoted 7 lines

Regards,

Tvrtko
_______________________________________________
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

#9 lore raw
Quoted 105 lines · Tvrtko Ursulin
Quoted 102 lines · Tvrtko Ursulin
Quoted 87 lines · Tvrtko Ursulin
Quoted 85 lines · Lionel Landwerlin
Quoted 84 lines · Tvrtko Ursulin
Quoted 69 lines · Lionel Landwerlin
Quoted 50 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

Bitfield width for configuring the active slice count has grown in Gen11 so we need to program the GEN8_R_PWR_CLK_STATE accordingly.

Current code was always requesting eight times the number of slices (due writting to a bitfield starting three bits higher than it should). These requests were luckily a) capped by the hardware to the available number of slices, and b) we haven't yet exported the code to ask for reduced slice configurations.

Due both of the above there was no impact from this incorrect programming but we should still fix it.

Bspec: 12247

---
  drivers/gpu/drm/i915/i915_reg.h  |  2 ++ drivers/gpu/drm/i915/intel_lrc.c | 12 ++++++++----

  2 files changed, 10 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_reg.h +2 -0
index 59d06d0055bb..640f7b774a26 100644
@@ -344,6 +344,8 @@ static inline bool 
 i915_mmio_reg_valid(i915_reg_t reg)
   #define   GEN8_RPCS_S_CNT_ENABLE    (1 << 18)
   #define   GEN8_RPCS_S_CNT_SHIFT        15
   #define   GEN8_RPCS_S_CNT_MASK        (0x7 << GEN8_RPCS_S_CNT_SHIFT)
+#define   GEN11_RPCS_S_CNT_SHIFT    12
+#define   GEN11_RPCS_S_CNT_MASK        (0x3f << 
 GEN11_RPCS_S_CNT_SHIFT)
   #define   GEN8_RPCS_SS_CNT_ENABLE    (1 << 11)

  #define   GEN8_RPCS_SS_CNT_SHIFT    8
  #define   GEN8_RPCS_SS_CNT_MASK        (0x7 << GEN8_RPCS_SS_CNT_SHIFT)

drivers/gpu/drm/i915/intel_lrc.c +0 -2
index 36050f085071..43b8b0675ba0 100644
@@ -2501,10 +2501,14 @@ make_rpcs(struct drm_i915_private *dev_priv)
        * enablement.
       */
       if (INTEL_INFO(dev_priv)->sseu.has_slice_pg) {
-        rpcs |= GEN8_RPCS_S_CNT_ENABLE;
-        rpcs |= hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask) <<

drivers/gpu/drm/i915/intel_lrc.c make_rpcs(struct drm_i915_private *dev_priv) · line 2511

Quoted 10 lines
drivers/gpu/drm/i915/intel_lrc.c +8 -2
-            GEN8_RPCS_S_CNT_SHIFT;
-        rpcs |= GEN8_RPCS_ENABLE;
+        rpcs = hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask);
+
+        if (INTEL_GEN(dev_priv) >= 11)
+            rpcs <<= GEN11_RPCS_S_CNT_SHIFT;
+        else
+            rpcs <<= GEN8_RPCS_S_CNT_SHIFT;
+
+        rpcs |= GEN8_RPCS_ENABLE | GEN8_RPCS_S_CNT_ENABLE;

I don't know if you saw that wording in the documentation :

  "

Note: In ICL, software programs this register as if GT consists of 2 slices with 4 subslices in each slice. Hardware maps this to the 1 slice/8-subslice physical layout.

Quoted 8 lines

"

My understanding is that it would make this function a bit more complicated ;)

It also implies that on ICL you cannot select 3 subslices, which is unfortunately what Tony was trying to do.

Maybe some opens need to be raised as to what's possible on ICL.

Happy to r-b this one since we clarified it is fine?

Regards,

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

Tvrtko

I think there is still an issue here with regard to the subslice programming.
You can only program values in [1, 4].
But because we expose up to 8 subslices, you need to take that mask and alter the slice mask based on its value.

Hmm true, thanks. I think we need to somehow express this restriction on the uAPI level rather than silently mask it. Shall we reject attempts to configure subslice mask if hweight(mask) > max_subslices / 2 for ICL-LP?

No wait, I was in the dynamic sseu world, but this also applies to the current code base..

When we program SScount in drm-tip with the hweight of 8 we overflow the bitfield already and splat over SScountEn. Luckily with no ill effect since we set it anyway.

Quoted 8 lines
Quoted 4 lines

So the fix needs to be in make_rpcs today..

if (GEN11_LP) {
     if (hweight(slice_mask) == 1)

&& hweight(subslice_mask) < max_subslices

So to leave the default full config alone.

Which means that drm-tip configures the 1x6x8 part as SScount=0b110, which is another undocumented value according to BSpec. Furthermore the allowed configuration table suggests only three subslices can be enabled when SScountEn is set, so like the situation on BXT/GLK, now I need to figure out if we are inadvertently only enabling half of the subslices on ICL-LP.

Tvrtko

Quoted 14 lines
Quoted 7 lines

         subslice_count /= 2;
     else
         subslice_en = 0;
}

I think..

Quoted 7 lines

Regards,

Tvrtko
_______________________________________________
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

#10 lore raw
Quoted 111 lines · Tvrtko Ursulin
Quoted 109 lines · Tvrtko Ursulin
Quoted 106 lines · Tvrtko Ursulin
Quoted 89 lines · Tvrtko Ursulin
Quoted 87 lines · Lionel Landwerlin
Quoted 86 lines · Tvrtko Ursulin
Quoted 71 lines · Lionel Landwerlin
Quoted 55 lines · Tvrtko Ursulin

From: Tvrtko Ursulin <[email protected]>

Bitfield width for configuring the active slice count has grown in Gen11 so we need to program the GEN8_R_PWR_CLK_STATE accordingly.

Current code was always requesting eight times the number of slices (due writting to a bitfield starting three bits higher than it should). These requests were luckily a) capped by the hardware to the available number of slices, and b) we haven't yet exported the code to ask for reduced slice configurations.

Due both of the above there was no impact from this incorrect programming but we should still fix it.

Bspec: 12247

---
  drivers/gpu/drm/i915/i915_reg.h  |  2 ++ drivers/gpu/drm/i915/intel_lrc.c | 12 ++++++++----

  2 files changed, 10 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_reg.h +2 -0
index 59d06d0055bb..640f7b774a26 100644
@@ -344,6 +344,8 @@ static inline bool 
 i915_mmio_reg_valid(i915_reg_t reg)
   #define   GEN8_RPCS_S_CNT_ENABLE    (1 << 18)
   #define   GEN8_RPCS_S_CNT_SHIFT        15
   #define   GEN8_RPCS_S_CNT_MASK        (0x7 << GEN8_RPCS_S_CNT_SHIFT)
+#define   GEN11_RPCS_S_CNT_SHIFT    12
+#define   GEN11_RPCS_S_CNT_MASK        (0x3f << 
 GEN11_RPCS_S_CNT_SHIFT)
   #define   GEN8_RPCS_SS_CNT_ENABLE    (1 << 11)

  #define   GEN8_RPCS_SS_CNT_SHIFT    8
  #define   GEN8_RPCS_SS_CNT_MASK        (0x7 << GEN8_RPCS_SS_CNT_SHIFT)

drivers/gpu/drm/i915/intel_lrc.c +0 -2
index 36050f085071..43b8b0675ba0 100644
@@ -2501,10 +2501,14 @@ make_rpcs(struct drm_i915_private 
 *dev_priv)
        * enablement.
       */
       if (INTEL_INFO(dev_priv)->sseu.has_slice_pg) {
-        rpcs |= GEN8_RPCS_S_CNT_ENABLE;
-        rpcs |= 

hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask) <<
-            GEN8_RPCS_S_CNT_SHIFT;
-        rpcs |= GEN8_RPCS_ENABLE;
+        rpcs = hweight8(INTEL_INFO(dev_priv)->sseu.slice_mask);

drivers/gpu/drm/i915/intel_lrc.c

Quoted 7 lines
drivers/gpu/drm/i915/intel_lrc.c +7 -0
+
+        if (INTEL_GEN(dev_priv) >= 11)
+            rpcs <<= GEN11_RPCS_S_CNT_SHIFT;
+        else
+            rpcs <<= GEN8_RPCS_S_CNT_SHIFT;
+
+        rpcs |= GEN8_RPCS_ENABLE | GEN8_RPCS_S_CNT_ENABLE;

I don't know if you saw that wording in the documentation :

  "

Note: In ICL, software programs this register as if GT consists of 2 slices with 4 subslices in each slice. Hardware maps this to the 1 slice/8-subslice physical layout.

Quoted 8 lines

"

My understanding is that it would make this function a bit more complicated ;)

It also implies that on ICL you cannot select 3 subslices, which is unfortunately what Tony was trying to do.

Maybe some opens need to be raised as to what's possible on ICL.

Happy to r-b this one since we clarified it is fine?

Regards,

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

Tvrtko

I think there is still an issue here with regard to the subslice programming.
You can only program values in [1, 4].
But because we expose up to 8 subslices, you need to take that mask and alter the slice mask based on its value.

Hmm true, thanks. I think we need to somehow express this restriction on the uAPI level rather than silently mask it. Shall we reject attempts to configure subslice mask if hweight(mask) > max_subslices / 2 for ICL-LP?

No wait, I was in the dynamic sseu world, but this also applies to the current code base..

When we program SScount in drm-tip with the hweight of 8 we overflow the bitfield already and splat over SScountEn. Luckily with no ill effect since we set it anyway.

So the fix needs to be in make_rpcs today..

Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

if (GEN11_LP) {
     if (hweight(slice_mask) == 1)

&& hweight(subslice_mask) < max_subslices

So to leave the default full config alone.

Which means that drm-tip configures the 1x6x8 part as SScount=0b110, which is another undocumented value according to BSpec. Furthermore the allowed configuration table suggests only three subslices can be enabled when SScountEn is set, so like the situation on BXT/GLK, now I need to figure out if we are inadvertently only enabling half of the subslices on ICL-LP.

Or maybe make the upper 4 bits of the subslice map to "big subslices" and lower bottom 4 to "small subslices".

Then have the restrictions checked in ioctl() context_set_param.

-

Lionel

Quoted 22 lines

Tvrtko

Quoted 18 lines
Quoted 7 lines

         subslice_count /= 2;
     else
         subslice_en = 0;
}

I think..

Quoted 7 lines

Regards,

Tvrtko
_______________________________________________
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

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

#11 lore raw

✗ Fi.CI.CHECKPATCH: warning for drm/i915/icl: Fix context slice count configuration

In reply to Tvrtko Ursulin
== Series Details ==

== Series Details ==

Series: drm/i915/icl: Fix context slice count configuration
URL : https://patchwork.freedesktop.org/series/48570/
State : warning

== Summary ==

$ dim checkpatch origin/drm-tip
5cf33a6a09d9 drm/i915/icl: Fix context slice count configuration
-:10: WARNING:TYPO_SPELLING: 'writting' may be misspelled - perhaps 'writing'?
#10: 
writting to a bitfield starting three bits higher than it should). These

total: 0 errors, 1 warnings, 0 checks, 26 lines checked

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

#12 lore raw

✓ Fi.CI.BAT: success for drm/i915/icl: Fix context slice count configuration

In reply to Tvrtko Ursulin
== Series Details ==

== Series Details ==

Series: drm/i915/icl: Fix context slice count configuration
URL : https://patchwork.freedesktop.org/series/48570/
State : success

== Summary ==

= CI Bug Log - changes from CI_DRM_4696 -> Patchwork_9992 =

== Summary - SUCCESS ==

No regressions found.

External URL: https://patchwork.freedesktop.org/api/1.0/series/48570/revisions/1/mbox/

== Known issues ==

Here are the changes found in Patchwork_9992 that come from known issues:

=== IGT changes ===

    ==== Issues hit ====

    igt@kms_pipe_crc_basic@hang-read-crc-pipe-b:
      {fi-byt-clapper}:   PASS -> FAIL (fdo#107362, fdo#103191)

    
    ==== Possible fixes ====

    igt@drv_selftest@live_hangcheck:
      {fi-skl-iommu}:     DMESG-FAIL (fdo#106560, fdo#107174) -> PASS
      fi-kbl-7567u:       DMESG-FAIL (fdo#106560, fdo#106947) -> PASS

    igt@kms_pipe_crc_basic@suspend-read-crc-pipe-a:
      {fi-byt-clapper}:   FAIL (fdo#107362, fdo#103191) -> PASS

    igt@kms_pipe_crc_basic@suspend-read-crc-pipe-b:
      {fi-cfl-8109u}:     INCOMPLETE (fdo#106070) -> PASS

    
    ==== Warnings ====

    {igt@pm_rpm@module-reload}:
      fi-cnl-psr:         FAIL -> WARN (fdo#107602)

{name}: This element is suppressed. This means it is ignored when computing
the status of the difference (SUCCESS, WARNING, or FAILURE).

fdo#103191 https://bugs.freedesktop.org/show_bug.cgi?id=103191
fdo#106070 https://bugs.freedesktop.org/show_bug.cgi?id=106070
fdo#106560 https://bugs.freedesktop.org/show_bug.cgi?id=106560
fdo#106947 https://bugs.freedesktop.org/show_bug.cgi?id=106947
fdo#107174 https://bugs.freedesktop.org/show_bug.cgi?id=107174
fdo#107362 https://bugs.freedesktop.org/show_bug.cgi?id=107362
fdo#107602 https://bugs.freedesktop.org/show_bug.cgi?id=107602

== Participating hosts (54 -> 46) ==

Missing (8): fi-ilk-m540 fi-hsw-4200u fi-byt-j1900 fi-byt-squawks fi-bsw-cyan fi-ctg-p8600 fi-glk-j4005 fi-gdg-551

== Build changes ==

* Linux: CI_DRM_4696 -> Patchwork_9992

CI_DRM_4696: ced152c46fc90f7c1ac8963850d64c9f1ce652d6 @ git://anongit.freedesktop.org/gfx-ci/linux
IGT_4608: 94ebd21177feedf03e8f6dd1e73dca1a6ec7a0ac @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools
Patchwork_9992: 5cf33a6a09d933f2cc8def1e8f43aa1812615aab @ git://anongit.freedesktop.org/gfx-ci/linux

== Linux commits ==

5cf33a6a09d9 drm/i915/icl: Fix context slice count configuration

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_9992/issues.html
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#13 lore raw

✓ Fi.CI.IGT: success for drm/i915/icl: Fix context slice count configuration

In reply to Tvrtko Ursulin
== Series Details ==

== Series Details ==

Series: drm/i915/icl: Fix context slice count configuration
URL : https://patchwork.freedesktop.org/series/48570/
State : success

== Summary ==

= CI Bug Log - changes from CI_DRM_4696_full -> Patchwork_9992_full =

== Summary - WARNING ==

Minor unknown changes coming with Patchwork_9992_full need to be verified manually.

If you think the reported changes have nothing to do with the changes introduced in Patchwork_9992_full, please notify your bug team to allow them to document this new failure mode, which will reduce false positives in CI.

== Possible new issues ==

Here are the unknown changes that may have been introduced in Patchwork_9992_full:

=== IGT changes ===

    ==== Warnings ====

    igt@kms_cursor_legacy@cursorb-vs-flipa-toggle:
      shard-hsw:          SKIP -> PASS

== Known issues ==

Here are the changes found in Patchwork_9992_full that come from known issues:

=== IGT changes ===

    ==== Issues hit ====

    igt@gem_ctx_isolation@vcs1-s3:
      shard-kbl:          PASS -> INCOMPLETE (fdo#103665)

    igt@kms_flip@modeset-vs-vblank-race:
      shard-apl:          PASS -> FAIL (fdo#103060)

    
    ==== Possible fixes ====

    igt@drv_suspend@shrink:
      shard-glk:          FAIL (fdo#106886) -> PASS

    igt@kms_flip@2x-flip-vs-expired-vblank-interruptible:
      shard-glk:          FAIL (fdo#105363) -> PASS +1

    igt@kms_flip@flip-vs-expired-vblank-interruptible:
      shard-glk:          FAIL (fdo#102887, fdo#105363) -> PASS

    igt@kms_setmode@basic:
      shard-apl:          FAIL (fdo#99912) -> PASS
      shard-glk:          FAIL (fdo#99912) -> PASS

    igt@perf@polling:
      shard-hsw:          FAIL (fdo#102252) -> PASS

    igt@pm_rpm@system-suspend:
      shard-kbl:          INCOMPLETE (fdo#107556, fdo#103665) -> PASS

fdo#102252 https://bugs.freedesktop.org/show_bug.cgi?id=102252
fdo#102887 https://bugs.freedesktop.org/show_bug.cgi?id=102887
fdo#103060 https://bugs.freedesktop.org/show_bug.cgi?id=103060
fdo#103665 https://bugs.freedesktop.org/show_bug.cgi?id=103665
fdo#105363 https://bugs.freedesktop.org/show_bug.cgi?id=105363
fdo#106886 https://bugs.freedesktop.org/show_bug.cgi?id=106886
fdo#107556 https://bugs.freedesktop.org/show_bug.cgi?id=107556
fdo#99912 https://bugs.freedesktop.org/show_bug.cgi?id=99912

== Participating hosts (5 -> 5) ==

No changes in participating hosts

== Build changes ==

* Linux: CI_DRM_4696 -> Patchwork_9992

CI_DRM_4696: ced152c46fc90f7c1ac8963850d64c9f1ce652d6 @ git://anongit.freedesktop.org/gfx-ci/linux
IGT_4608: 94ebd21177feedf03e8f6dd1e73dca1a6ec7a0ac @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools
Patchwork_9992: 5cf33a6a09d933f2cc8def1e8f43aa1812615aab @ git://anongit.freedesktop.org/gfx-ci/linux
piglit_4509: fdc5a4ca11124ab8413c7988896eec4c97336694 @ git://anongit.freedesktop.org/piglit

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_9992/shards.html
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx