Steer multicast register workaround verification
Subsystems (2)
cover Steer multicast register workaround verification
We're seeing some CI errors indicating that a workaround did not apply properly on EHL/JSL. The workaround in question is updating a multicast register, the failures are only seen on specific CI machines, and the failures only seem to happen on resets and such rather than on initial driver load. It seems likely that the culprit here is failure to steer the multicast register readback on a SKU that has slice0 / subslice0 fused off.
This series makes a couple changes:
* Workaround verification will explicitly steer MCR registers by calling read_subslice_reg rather than a regular read.
* New multicast ranges are added for gen11 and gen12. Sadly this information is still missing from the bspec (just like the updated forcewake tables). The hardware guys have given us a spreadsheet with both the forcewake and the multicast information while they work on getting the spec properly updated, so that's where the new ranges come from.
In addition to MCR and forcewake, there's supposed to be some more bspec updates coming soon that deal with steering (i.e., different MCR ranges should actually be using different registers to steer rather than just the 0xFDC register we're familiar with); I don't have the full details on that yet, so those updates will have to wait until we actually have an updated spec.
References: https://gitlab.freedesktop.org/drm/intel/issues/1222
Matt Roper (4):
drm/i915: Setup multicast register steering for all gen >= 10 drm/i915: Steer multicast register readback in wa_verify drm/i915: Don't skip verification of MCR engine workarounds drm/i915: Add MCR ranges for gen11 and gen12
drivers/gpu/drm/i915/gt/intel_engine.h | 3 + drivers/gpu/drm/i915/gt/intel_engine_cs.c | 17 +- drivers/gpu/drm/i915/gt/intel_workarounds.c | 146 ++++++++++++------ .../gpu/drm/i915/gt/intel_workarounds_types.h | 2 + 4 files changed, 110 insertions(+), 58 deletions(-)
Signature · 6 lines
2.24.1 _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 1/4 drm/i915: Setup multicast register steering for all gen >= 10
Steering of multicast registers for workarounds is needed on all platforms gen10 and above. Move the wa_init_mcr() call to the higher-level gt_init_workarounds() rather than re-calling it for each individual platform.
--- drivers/gpu/drm/i915/gt/intel_workarounds.c | 9 +++------ 1 file changed, 3 insertions(+), 6 deletions(-)
drivers/gpu/drm/i915/gt/intel_workarounds.c +3 -6
@@ -870,8 +870,6 @@ wa_init_mcr(struct drm_i915_private *i915, struct i915_wa_list *wal)static void cnl_gt_workarounds_init (struct drm_i915_private *i915,struct i915_wa_list *wal) { -wa_init_mcr (i915, wal); -/* WaDisableI2mCycleOnWRPort:cnl (pre-prod) */ if (IS_CNL_REVID (i915,CNL_REVID_B0 ,CNL_REVID_B0 )) wa_write_or(wal,
@@ -887,8 +885,6 @@ cnl_gt_workarounds_init(struct drm_i915_private *i915, struct i915_wa_list *wal)static void icl_gt_workarounds_init (struct drm_i915_private *i915,struct i915_wa_list *wal) { -wa_init_mcr (i915, wal); -/* WaInPlaceDecompressionHang:icl */ wa_write_or(wal, GEN9_GAMT_ECO_REG_RW_IA,
@@ -943,8 +939,6 @@ icl_gt_workarounds_init(struct drm_i915_private *i915, struct i915_wa_list *wal)static void tgl_gt_workarounds_init (struct drm_i915_private *i915,struct i915_wa_list *wal) { -wa_init_mcr (i915, wal); -/* Wa_1409420604:tgl */ if (IS_TGL_REVID (i915,TGL_REVID_A0 ,TGL_REVID_A0 )) wa_write_or(wal,
@@ -961,6 +955,9 @@ tgl_gt_workarounds_init(struct drm_i915_private *i915, struct i915_wa_list *wal)static void gt_init_workarounds (struct drm_i915_private *i915,struct i915_wa_list *wal) { +if (INTEL_GEN (i915) >=10 ) +wa_init_mcr (i915, wal); +if (IS_GEN (i915,12 ))tgl_gt_workarounds_init (i915, wal);else if (IS_GEN (i915,11 ))
Signature · 6 lines
2.24.1 _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 2/4 drm/i915: Steer multicast register readback in wa_verify
In reply to Matt RoperReads of multicast registers give the value for slice/subslice 0 by default unless we manually steer them to a specific slice/subslice. If slice/subslice 0 are fused off in hardware, we'll always read back a value of 0 rather than the value we wrote into the multicast register. Although wa_init_mcr() sets up some initial steering to a valid slice/subslice at startup, we appear to later lose that steering setting in some cases. Let's handle this more explicitly and have wa_verify use intel_read_subslice_reg() to directly steer the readback of workaround registers in an MCR range to a valid slice/subslice.
--- drivers/gpu/drm/i915/gt/intel_engine.h | 3 + drivers/gpu/drm/i915/gt/intel_engine_cs.c | 17 ++-- drivers/gpu/drm/i915/gt/intel_workarounds.c | 98 ++++++++++++------- .../gpu/drm/i915/gt/intel_workarounds_types.h | 2 + 4 files changed, 73 insertions(+), 47 deletions(-)
drivers/gpu/drm/i915/gt/intel_engine.h +3 -0
@@ -214,6 +214,9 @@ int intel_ring_submission_setup(struct intel_engine_cs *engine);int intel_engine_stop_cs (struct intel_engine_cs *engine);void intel_engine_cancel_stop_cs (struct intel_engine_cs *engine); +u32 intel_read_subslice_reg (struct intel_uncore *uncore, +int slice,int subslice,i915_reg_t reg); +void intel_engine_set_hwsp_writemask (struct intel_engine_cs *engine,u32 mask);u64 intel_engine_get_active_head (const struct intel_engine_cs *engine);
drivers/gpu/drm/i915/gt/intel_engine_cs.c +8 -9
@@ -951,12 +951,11 @@ const char *i915_cache_level_str(struct drm_i915_private *i915, int type) } } -static u32 -read_subslice_reg (const struct intel_engine_cs *engine, -int slice,int subslice,i915_reg_t reg) +u32 +intel_read_subslice_reg (struct intel_uncore *uncore, +int slice,int subslice,i915_reg_t reg) { -struct drm_i915_private *i915 = engine->i915; -struct intel_uncore *uncore = engine->uncore; +struct drm_i915_private *i915 = uncore->i915;u32 mcr_mask, mcr_ss, mcr, old_mcr, val;enum forcewake_domains fw_domains;
@@ -1027,11 +1026,11 @@ void intel_engine_get_instdone(const struct intel_engine_cs *engine, }for_each_instdone_slice_subslice (i915, sseu, slice, subslice) { instdone->sampler[slice][subslice] = -read_subslice_reg (engine, slice, subslice, -GEN7_SAMPLER_INSTDONE ); +intel_read_subslice_reg (uncore, slice, subslice, +GEN7_SAMPLER_INSTDONE ); instdone->row[slice][subslice] = -read_subslice_reg (engine, slice, subslice, -GEN7_ROW_INSTDONE ); +intel_read_subslice_reg (uncore, slice, subslice, +GEN7_ROW_INSTDONE ); }break ;case 7 :
drivers/gpu/drm/i915/gt/intel_workarounds.c +60 -38
147 lines not on this page · raw message
drivers/gpu/drm/i915/gt/intel_workarounds_types.h +2 -0
@@ -24,6 +24,8 @@ struct i915_wa_list {struct i915_wa *list;unsigned int count;unsigned int wa_count; +u32 mcrslice; + u32 mcrss; };#endif /* __INTEL_WORKAROUNDS_TYPES_H__ */
Signature · 6 lines
2.24.1 _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 3/4 drm/i915: Don't skip verification of MCR engine workarounds
In reply to Matt RoperNow that we manually steer multicast register reads during workaround verification, it should be safe to verify these ones too.
--- drivers/gpu/drm/i915/gt/intel_workarounds.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)
drivers/gpu/drm/i915/gt/intel_workarounds.c +1 -5
@@ -1812,13 +1812,9 @@ static int engine_wa_list_verify(struct intel_context *ce, } err =0 ; -for (i =0 , wa = wal->list; i < wal->count; i++, wa++) { -if (mcr_range (rq->i915,i915_mmio_reg_offset (wa->reg))) -continue ; - +for (i =0 , wa = wal->list; i < wal->count; i++, wa++)if (!wa_verify (rq->i915, wa, results[i], wal->name, from)) err = -ENXIO ; - }i915_gem_object_unpin_map (vma->obj);
Signature · 6 lines
2.24.1 _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 4/4 drm/i915: Add MCR ranges for gen11 and gen12
In reply to Matt RoperThe multicast register ranges are slightly different for gen11 and gen12 than the table we have for gen8. This information never got updated in the bspec, so this patch is based on a spreadsheet provided by the hardware team while they work on getting the official documentation updated.
--- drivers/gpu/drm/i915/gt/intel_workarounds.c | 45 ++++++++++++++++++--- 1 file changed, 39 insertions(+), 6 deletions(-)
drivers/gpu/drm/i915/gt/intel_workarounds.c +39 -6
@@ -1023,10 +1023,12 @@ wa_verify(struct drm_i915_private *i915, const struct i915_wa *wa, u32 cur,return true; } -static const struct { +struct mcr_range {u32 start;u32 end; -} mcr_ranges_gen8[] = { +}; + +static const struct mcr_range mcr_ranges_gen8[] = { { .start =0x5500 , .end =0x55ff }, { .start =0x7000 , .end =0x7fff }, { .start =0x9400 , .end =0x97ff },
@@ -1035,11 +1037,42 @@ static const struct { {}, }; +static const struct mcr_range mcr_ranges_gen11[] = { + { .start =0x5500 , .end =0x55ff }, + { .start =0x7000 , .end =0x7fff }, + { .start =0x8140 , .end =0x815f }, + { .start =0x8c00 , .end =0x8cff }, + { .start =0x94d0 , .end =0x955f }, + { .start =0xb000 , .end =0xb3ff }, + { .start =0xdf00 , .end =0xe8ff }, + { .start =0x24400 , .end =0x24fff }, + {}, +}; + +static const struct mcr_range mcr_ranges_gen12[] = { + { .start =0xb00 , .end =0xbff }, + { .start =0x1000 , .end =0x1fff }, + { .start =0x8150 , .end =0x815f }, + { .start =0x8700 , .end =0x87ff }, + { .start =0x9520 , .end =0x955f }, + { .start =0xb100 , .end =0xb3ff }, + { .start =0xde80 , .end =0xe8ff }, + { .start =0x24a00 , .end =0x24a7f }, + {}, +}; +static bool mcr_range (struct drm_i915_private *i915,u32 offset) { +const struct mcr_range *range_list;int i; -if (INTEL_GEN (i915) <8 ) +if (INTEL_GEN (i915) >=12 ) + range_list = mcr_ranges_gen12; +else if (INTEL_GEN (i915) >=11 ) + range_list = mcr_ranges_gen11; +else if (INTEL_GEN (i915) >=8 ) + range_list = mcr_ranges_gen8; +else return false;/*
@@ -1047,9 +1080,9 @@ static bool mcr_range(struct drm_i915_private *i915, u32 offset)* which only controls CPU initiated MMIO. Routing does not * work for CS access so we cannot verify them on this path. */ -for (i =0 ; mcr_ranges_gen8[i].start; i++) -if (offset >= mcr_ranges_gen8[i].start && - offset <= mcr_ranges_gen8[i].end) +for (i =0 ; range_list[i].start; i++) +if (offset >= range_list[i].start && + offset <= range_list[i].end)return true;return false;
Signature · 6 lines
2.24.1 _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Patchwork
CI[Intel-gfx] ✗ Fi.CI.CHECKPATCH: warning for Steer multicast register workaround verification
In reply to Matt Roper== Series Details ==
== Series Details ==
Series: Steer multicast register workaround verification
URL : https://patchwork.freedesktop.org/series/76792/
State : warning
== Summary ==
$ dim checkpatch origin/drm-tip c89d34e99992 drm/i915: Setup multicast register steering for all gen >= 10 4759117a6337 drm/i915: Steer multicast register readback in wa_verify -:77: CHECK:MULTIPLE_ASSIGNMENTS: multiple assignments should be avoided #77: FILE: drivers/gpu/drm/i915/gt/intel_workarounds.c:59: + wal->mcrslice = wal->mcrss = 0;
total: 0 errors, 0 warnings, 1 checks, 195 lines checked 1d656b938337 drm/i915: Don't skip verification of MCR engine workarounds d348e98bdcf9 drm/i915: Add MCR ranges for gen11 and gen12
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Patchwork
CI[Intel-gfx] ✗ Fi.CI.BAT: failure for Steer multicast register workaround verification
In reply to Matt Roper== Series Details ==
== Series Details ==
Series: Steer multicast register workaround verification
URL : https://patchwork.freedesktop.org/series/76792/
State : failure
== Summary ==
CI Bug Log - changes from CI_DRM_8403 -> Patchwork_17534
====================================================
Summary
-------
**FAILURE**
Serious unknown changes coming with Patchwork_17534 absolutely need to be verified manually.
If you think the reported changes have nothing to do with the changes introduced in Patchwork_17534, please notify your bug team to allow them to document this new failure mode, which will reduce false positives in CI.
External URL: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_17534/index.html
Possible new issues
-------------------
Here are the unknown changes that may have been introduced in Patchwork_17534:
### IGT changes ###
#### Possible regressions ####
* igt@i915_selftest@live@workarounds:
- fi-bsw-n3050: [PASS][1] -> [DMESG-FAIL][2]
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_8403/fi-bsw-n3050/igt@i915_selftest@[email protected]
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_17534/fi-bsw-n3050/igt@i915_selftest@[email protected]
- fi-bsw-kefka: [PASS][3] -> [DMESG-FAIL][4]
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_8403/fi-bsw-kefka/igt@i915_selftest@[email protected]
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_17534/fi-bsw-kefka/igt@i915_selftest@[email protected]
- fi-bsw-nick: [PASS][5] -> [DMESG-FAIL][6]
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_8403/fi-bsw-nick/igt@i915_selftest@[email protected]
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_17534/fi-bsw-nick/igt@i915_selftest@[email protected]
- fi-bdw-5557u: [PASS][7] -> [DMESG-FAIL][8]
[7]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_8403/fi-bdw-5557u/igt@i915_selftest@[email protected]
[8]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_17534/fi-bdw-5557u/igt@i915_selftest@[email protected]
Known issues
------------
Here are the changes found in Patchwork_17534 that come from known issues:
### IGT changes ###
#### Possible fixes ####
* igt@i915_selftest@live@gt_engines:
- fi-bwr-2160: [INCOMPLETE][9] ([i915#489]) -> [PASS][10]
[9]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_8403/fi-bwr-2160/igt@i915_selftest@live@gt_engines.html
[10]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_17534/fi-bwr-2160/igt@i915_selftest@live@gt_engines.html
[i915#489]: https://gitlab.freedesktop.org/drm/intel/issues/489
Participating hosts (51 -> 17)
------------------------------
ERROR: It appears as if the changes made in Patchwork_17534 prevented too many machines from booting.
Missing (34): fi-kbl-soraka fi-icl-u2 fi-apl-guc fi-icl-y fi-skl-lmem fi-icl-guc fi-icl-dsi fi-skl-6600u fi-cml-u2 fi-bxt-dsi fi-tgl-u fi-cml-s fi-glk-dsi fi-kbl-7500u fi-ctg-p8600 fi-kbl-7560u fi-skl-6700k2 fi-kbl-r fi-ilk-m540 fi-ehl-1 fi-tgl-dsi fi-skl-guc fi-cfl-8700k fi-hsw-4200u fi-byt-squawks fi-bsw-cyan fi-cfl-guc fi-kbl-guc fi-whl-u fi-kbl-x1275 fi-cfl-8109u fi-kbl-8809g fi-byt-clapper fi-bdw-samus
Build changes
-------------
* CI: CI-20190529 -> None
* Linux: CI_DRM_8403 -> Patchwork_17534
CI-20190529: 20190529
CI_DRM_8403: 09978e99929f6e5acfe1e959f6499a134f210887 @ git://anongit.freedesktop.org/gfx-ci/linux
IGT_5619: 94de923ca8d4cc8f532b8062d87aaad9da6ef956 @ git://anongit.freedesktop.org/xorg/app/intel-gpu-tools
Patchwork_17534: d348e98bdcf92355b0c475f858e733c87ff2c69a @ git://anongit.freedesktop.org/gfx-ci/linux
== Linux commits ==
d348e98bdcf9 drm/i915: Add MCR ranges for gen11 and gen12 1d656b938337 drm/i915: Don't skip verification of MCR engine workarounds 4759117a6337 drm/i915: Steer multicast register readback in wa_verify c89d34e99992 drm/i915: Setup multicast register steering for all gen >= 10
== Logs ==
For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_17534/index.html
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Hi,
Quoted 8 lines · Matt Roper
We're seeing some CI errors indicating that a workaround did not apply properly on EHL/JSL. The workaround in question is updating a multicast register, the failures are only seen on specific CI machines, and the failures only seem to happen on resets and such rather than on initial driver load. It seems likely that the culprit here is failure to steer the multicast register readback on a SKU that has slice0 / subslice0 fused off.
Quoted 9 lines
This series makes a couple changes:
* Workaround verification will explicitly steer MCR registers by callingread_subslice_regrather than a regular read.
* New multicast ranges are added for gen11 and gen12. Sadly this information is still missing from the bspec (just like the updated forcewake tables). The hardware guys have given us a spreadsheet with both the forcewake and the multicast information while they work on getting the spec properly updated, so that's where the new ranges come from.
I think there are multiple things here. To begin with, newly discovered ranges are of course a savior.
But I am not sure about the approach of using intel_read_subslice_reg in wa_verify. It is one suspicion that 0xfdc is lost on reset, but we do reprogram it afterwards don't we? And since it is the first register in the list it is supposed to be in place before the rest of verification runs, no?
A year or two I tried figuring this for Icelake and failed, but AFAIR (maybe my experiments can be found somewhere on trybot patchwork), I even tried both applying the affected ones via unicast (for each ss, or l3 where applicable) and also verifying a single register in all enabled ss. AFAIR there were still some issues there. Granted my memory could be leaky.. But I think this multiple write/verify could still be useful.
(Now that I think about it, I think that the problem area back when I experiementing with it was more suspend/resume.. hm..)
My main concern is that with current code we effectively have, after reset:
intel_gt_apply_workarounds:
program 0xfdc
program the rest of waverify_wa
do reads using configured 0xfdc
So MCR should be correct. This series seems to be doing:
intel_gt_apply_workarounds:
program 0xfdc
* store ss used for MCR configuration
program the rest of waverify_wa
Do reads but reconfigure 0xfdc before every register in range, but to the same value as in initial configuration.
Is this correct? Is the thinking then simply writing the same value to 0xfdc multiple times fixes things?
Regards,
Tvrtko
P.S. Update, found the experiments, listing some of them:
https://patchwork.freedesktop.org/series/64183/
https://patchwork.freedesktop.org/series/64013/
It reminded me that there were some unexplained issues with regards of where I used ffs or fls for finding the valid common MCR setting between L3 and SSEU. I think we use a different one than Windows but ours works better for our verification, empirically at least. Usual disclaimer about my leaky memory applies here.
Quoted 15 lines
In addition to MCR and forcewake, there's supposed to be some more bspec updates coming soon that deal with steering (i.e., different MCR ranges should actually be using different registers to steer rather than just the
0xFDCregister we're familiar with); I don't have the full details on that yet, so those updates will have to wait until we actually have an updated spec.References: https://gitlab.freedesktop.org/drm/intel/issues/1222
Matt Roper (4):
drm/i915: Setup multicast register steering for all gen >= 10 drm/i915: Steer multicast register readback inwa_verifydrm/i915: Don't skip verification of MCR engine workarounds drm/i915: Add MCR ranges for gen11 and gen12
Quoted 6 lines
drivers/gpu/drm/i915/gt/intel_engine.h | 3 + drivers/gpu/drm/i915/gt/intel_engine_cs.c | 17 +- drivers/gpu/drm/i915/gt/intel_workarounds.c | 146 ++++++++++++------ .../gpu/drm/i915/gt/intel_workarounds_types.h | 2 + 4 files changed, 110 insertions(+), 58 deletions(-)
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Quoted 58 lines · Tvrtko Ursulin
Hi,
Quoted 8 lines · Matt Roper
We're seeing some CI errors indicating that a workaround did not apply properly on EHL/JSL. The workaround in question is updating a multicast register, the failures are only seen on specific CI machines, and the failures only seem to happen on resets and such rather than on initial driver load. It seems likely that the culprit here is failure to steer the multicast register readback on a SKU that has slice0 / subslice0 fused off.
Quoted 9 lines
This series makes a couple changes:
* Workaround verification will explicitly steer MCR registers by callingread_subslice_regrather than a regular read.
* New multicast ranges are added for gen11 and gen12. Sadly this information is still missing from the bspec (just like the updated forcewake tables). The hardware guys have given us a spreadsheet with both the forcewake and the multicast information while they work on getting the spec properly updated, so that's where the new ranges come from.I think there are multiple things here. To begin with, newly discovered ranges are of course a savior.
But I am not sure about the approach of using
intel_read_subslice_reginwa_verify. It is one suspicion that0xfdcis lost on reset, but we do reprogram it afterwards don't we? And since it is the first register in the list it is supposed to be in place before the rest of verification runs, no?A year or two I tried figuring this for Icelake and failed, but AFAIR (maybe my experiments can be found somewhere on trybot patchwork), I even tried both applying the affected ones via unicast (for each ss, or l3 where applicable) and also verifying a single register in all enabled ss. AFAIR there were still some issues there. Granted my memory could be leaky.. But I think this multiple write/verify could still be useful.
(Now that I think about it, I think that the problem area back when I experiementing with it was more suspend/resume.. hm..)
My main concern is that with current code we effectively have, after reset:
intel_gt_apply_workarounds:
program0xfdc
program the rest of waverify_wa
do reads using configured0xfdcSo MCR should be correct. This series seems to be doing:
intel_gt_apply_workarounds:
program0xfdc
* store ss used for MCR configuration
program the rest of waverify_wa
Do reads but reconfigure0xfdcbefore every register in range, but to the same value as in initial configuration.
Quoted 2 lines
Is this correct? Is the thinking then simply writing the same value to
0xfdcmultiple times fixes things?
We stick the MCR steering at the beginning of the GT workaround list, but the workaround that CI is complaining about is one of the RCS engine workarounds, which is on a separate WA list. At startup I think the RCS workarounds are applied shortly after the GT workarounds, so the steering is still in place, but on things like engine resets and such that might not be the case.
A simpler approach might be to just stick the steering settings at the front of the RCS workaround list, just as we do for the GT list. That was actually what I intended to try initially, but I started thinking that we might want to handle registers invidually once we get the gen12 details about different steering registers for different ranges. But I guess we should probably hold off on that until we actually have all the details documented and just keep things simple for now.
I'll send a v2 of this series later today that just sticks the steering settings at the front of the RCS engine's WA list instead of trying to individually steer each register. That's probably a simpler and cleaner approach for now.
Matt
Quoted 37 lines
Regards,
Tvrtko
P.S. Update, found the experiments, listing some of them:https://patchwork.freedesktop.org/series/64183/
https://patchwork.freedesktop.org/series/64013/It reminded me that there were some unexplained issues with regards of where I used ffs or fls for finding the valid common MCR setting between L3 and SSEU. I think we use a different one than Windows but ours works better for our verification, empirically at least. Usual disclaimer about my leaky memory applies here.
Quoted 21 lines
In addition to MCR and forcewake, there's supposed to be some more bspec updates coming soon that deal with steering (i.e., different MCR ranges should actually be using different registers to steer rather than just the
0xFDCregister we're familiar with); I don't have the full details on that yet, so those updates will have to wait until we actually have an updated spec.References: https://gitlab.freedesktop.org/drm/intel/issues/1222
Matt Roper (4):
drm/i915: Setup multicast register steering for all gen >= 10 drm/i915: Steer multicast register readback inwa_verifydrm/i915: Don't skip verification of MCR engine workarounds drm/i915: Add MCR ranges for gen11 and gen12drivers/gpu/drm/i915/gt/intel_engine.h | 3 + drivers/gpu/drm/i915/gt/intel_engine_cs.c | 17 +- drivers/gpu/drm/i915/gt/intel_workarounds.c | 146 ++++++++++++------ .../gpu/drm/i915/gt/intel_workarounds_types.h | 2 + 4 files changed, 110 insertions(+), 58 deletions(-)
Signature · 9 lines
Matt Roper Graphics Software Engineer VTT-OSGC Platform Enablement Intel Corporation (916) 356-2795 _______________________________________________ Intel-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/intel-gfx