2 * SPDX-License-Identifier: MIT
4 * Copyright © 2014-2018 Intel Corporation
8 #include "intel_context.h"
9 #include "intel_engine_pm.h"
11 #include "intel_ring.h"
12 #include "intel_workarounds.h"
15 * DOC: Hardware workarounds
17 * This file is intended as a central place to implement most [1]_ of the
18 * required workarounds for hardware to work as originally intended. They fall
19 * in five basic categories depending on how/when they are applied:
21 * - Workarounds that touch registers that are saved/restored to/from the HW
22 * context image. The list is emitted (via Load Register Immediate commands)
23 * everytime a new context is created.
24 * - GT workarounds. The list of these WAs is applied whenever these registers
25 * revert to default values (on GPU reset, suspend/resume [2]_, etc..).
26 * - Display workarounds. The list is applied during display clock-gating
28 * - Workarounds that whitelist a privileged register, so that UMDs can manage
29 * them directly. This is just a special case of a MMMIO workaround (as we
30 * write the list of these to/be-whitelisted registers to some special HW
32 * - Workaround batchbuffers, that get executed automatically by the hardware
33 * on every HW context restore.
35 * .. [1] Please notice that there are other WAs that, due to their nature,
36 * cannot be applied from a central place. Those are peppered around the rest
37 * of the code, as needed.
39 * .. [2] Technically, some registers are powercontext saved & restored, so they
40 * survive a suspend/resume. In practice, writing them again is not too
41 * costly and simplifies things. We can revisit this in the future.
46 * Keep things in this file ordered by WA type, as per the above (context, GT,
47 * display, register whitelist, batchbuffer). Then, inside each type, keep the
50 * - Infrastructure functions and macros
51 * - WAs per platform in standard gen/chrono order
52 * - Public functions to init or apply the given workaround type.
56 * KBL revision ID ordering is bizarre; higher revision ID's map to lower
57 * steppings in some cases. So rather than test against the revision ID
58 * directly, let's map that into our own range of increasing ID's that we
59 * can test against in a regular manner.
62 const struct i915_rev_steppings kbl_revids
[] = {
63 [0] = { .gt_stepping
= KBL_REVID_A0
, .disp_stepping
= KBL_REVID_A0
},
64 [1] = { .gt_stepping
= KBL_REVID_B0
, .disp_stepping
= KBL_REVID_B0
},
65 [2] = { .gt_stepping
= KBL_REVID_C0
, .disp_stepping
= KBL_REVID_B0
},
66 [3] = { .gt_stepping
= KBL_REVID_D0
, .disp_stepping
= KBL_REVID_B0
},
67 [4] = { .gt_stepping
= KBL_REVID_F0
, .disp_stepping
= KBL_REVID_C0
},
68 [5] = { .gt_stepping
= KBL_REVID_C0
, .disp_stepping
= KBL_REVID_B1
},
69 [6] = { .gt_stepping
= KBL_REVID_D1
, .disp_stepping
= KBL_REVID_B1
},
70 [7] = { .gt_stepping
= KBL_REVID_G0
, .disp_stepping
= KBL_REVID_C0
},
73 const struct i915_rev_steppings tgl_uy_revids
[] = {
74 [0] = { .gt_stepping
= TGL_REVID_A0
, .disp_stepping
= TGL_REVID_A0
},
75 [1] = { .gt_stepping
= TGL_REVID_B0
, .disp_stepping
= TGL_REVID_C0
},
76 [2] = { .gt_stepping
= TGL_REVID_B1
, .disp_stepping
= TGL_REVID_C0
},
77 [3] = { .gt_stepping
= TGL_REVID_C0
, .disp_stepping
= TGL_REVID_D0
},
80 /* Same GT stepping between tgl_uy_revids and tgl_revids don't mean the same HW */
81 const struct i915_rev_steppings tgl_revids
[] = {
82 [0] = { .gt_stepping
= TGL_REVID_A0
, .disp_stepping
= TGL_REVID_B0
},
83 [1] = { .gt_stepping
= TGL_REVID_B0
, .disp_stepping
= TGL_REVID_D0
},
86 static void wa_init_start(struct i915_wa_list
*wal
, const char *name
, const char *engine_name
)
89 wal
->engine_name
= engine_name
;
92 #define WA_LIST_CHUNK (1 << 4)
94 static void wa_init_finish(struct i915_wa_list
*wal
)
96 /* Trim unused entries. */
97 if (!IS_ALIGNED(wal
->count
, WA_LIST_CHUNK
)) {
98 struct i915_wa
*list
= kmemdup(wal
->list
,
99 wal
->count
* sizeof(*list
),
111 DRM_DEBUG_DRIVER("Initialized %u %s workarounds on %s\n",
112 wal
->wa_count
, wal
->name
, wal
->engine_name
);
115 static void _wa_add(struct i915_wa_list
*wal
, const struct i915_wa
*wa
)
117 unsigned int addr
= i915_mmio_reg_offset(wa
->reg
);
118 unsigned int start
= 0, end
= wal
->count
;
119 const unsigned int grow
= WA_LIST_CHUNK
;
122 GEM_BUG_ON(!is_power_of_2(grow
));
124 if (IS_ALIGNED(wal
->count
, grow
)) { /* Either uninitialized or full. */
125 struct i915_wa
*list
;
127 list
= kmalloc_array(ALIGN(wal
->count
+ 1, grow
), sizeof(*wa
),
130 DRM_ERROR("No space for workaround init!\n");
135 memcpy(list
, wal
->list
, sizeof(*wa
) * wal
->count
);
142 while (start
< end
) {
143 unsigned int mid
= start
+ (end
- start
) / 2;
145 if (i915_mmio_reg_offset(wal
->list
[mid
].reg
) < addr
) {
147 } else if (i915_mmio_reg_offset(wal
->list
[mid
].reg
) > addr
) {
150 wa_
= &wal
->list
[mid
];
152 if ((wa
->clr
| wa_
->clr
) && !(wa
->clr
& ~wa_
->clr
)) {
153 DRM_ERROR("Discarding overwritten w/a for reg %04x (clear: %08x, set: %08x)\n",
154 i915_mmio_reg_offset(wa_
->reg
),
157 wa_
->set
&= ~wa
->clr
;
163 wa_
->read
|= wa
->read
;
169 wa_
= &wal
->list
[wal
->count
++];
172 while (wa_
-- > wal
->list
) {
173 GEM_BUG_ON(i915_mmio_reg_offset(wa_
[0].reg
) ==
174 i915_mmio_reg_offset(wa_
[1].reg
));
175 if (i915_mmio_reg_offset(wa_
[1].reg
) >
176 i915_mmio_reg_offset(wa_
[0].reg
))
179 swap(wa_
[1], wa_
[0]);
183 static void wa_add(struct i915_wa_list
*wal
, i915_reg_t reg
,
184 u32 clear
, u32 set
, u32 read_mask
)
186 struct i915_wa wa
= {
197 wa_write_masked_or(struct i915_wa_list
*wal
, i915_reg_t reg
, u32 clear
, u32 set
)
199 wa_add(wal
, reg
, clear
, set
, clear
);
203 wa_write(struct i915_wa_list
*wal
, i915_reg_t reg
, u32 set
)
205 wa_write_masked_or(wal
, reg
, ~0, set
);
209 wa_write_or(struct i915_wa_list
*wal
, i915_reg_t reg
, u32 set
)
211 wa_write_masked_or(wal
, reg
, set
, set
);
215 wa_write_clr(struct i915_wa_list
*wal
, i915_reg_t reg
, u32 clr
)
217 wa_write_masked_or(wal
, reg
, clr
, 0);
221 wa_masked_en(struct i915_wa_list
*wal
, i915_reg_t reg
, u32 val
)
223 wa_add(wal
, reg
, 0, _MASKED_BIT_ENABLE(val
), val
);
227 wa_masked_dis(struct i915_wa_list
*wal
, i915_reg_t reg
, u32 val
)
229 wa_add(wal
, reg
, 0, _MASKED_BIT_DISABLE(val
), val
);
232 #define WA_SET_BIT_MASKED(addr, mask) \
233 wa_masked_en(wal, (addr), (mask))
235 #define WA_CLR_BIT_MASKED(addr, mask) \
236 wa_masked_dis(wal, (addr), (mask))
238 #define WA_SET_FIELD_MASKED(addr, mask, value) \
239 wa_write_masked_or(wal, (addr), 0, _MASKED_FIELD((mask), (value)))
241 static void gen6_ctx_workarounds_init(struct intel_engine_cs
*engine
,
242 struct i915_wa_list
*wal
)
244 WA_SET_BIT_MASKED(INSTPM
, INSTPM_FORCE_ORDERING
);
247 static void gen7_ctx_workarounds_init(struct intel_engine_cs
*engine
,
248 struct i915_wa_list
*wal
)
250 WA_SET_BIT_MASKED(INSTPM
, INSTPM_FORCE_ORDERING
);
253 static void gen8_ctx_workarounds_init(struct intel_engine_cs
*engine
,
254 struct i915_wa_list
*wal
)
256 WA_SET_BIT_MASKED(INSTPM
, INSTPM_FORCE_ORDERING
);
258 /* WaDisableAsyncFlipPerfMode:bdw,chv */
259 WA_SET_BIT_MASKED(MI_MODE
, ASYNC_FLIP_PERF_DISABLE
);
261 /* WaDisablePartialInstShootdown:bdw,chv */
262 WA_SET_BIT_MASKED(GEN8_ROW_CHICKEN
,
263 PARTIAL_INSTRUCTION_SHOOTDOWN_DISABLE
);
265 /* Use Force Non-Coherent whenever executing a 3D context. This is a
266 * workaround for for a possible hang in the unlikely event a TLB
267 * invalidation occurs during a PSD flush.
269 /* WaForceEnableNonCoherent:bdw,chv */
270 /* WaHdcDisableFetchWhenMasked:bdw,chv */
271 WA_SET_BIT_MASKED(HDC_CHICKEN0
,
272 HDC_DONOT_FETCH_MEM_WHEN_MASKED
|
273 HDC_FORCE_NON_COHERENT
);
275 /* From the Haswell PRM, Command Reference: Registers, CACHE_MODE_0:
276 * "The Hierarchical Z RAW Stall Optimization allows non-overlapping
277 * polygons in the same 8x4 pixel/sample area to be processed without
278 * stalling waiting for the earlier ones to write to Hierarchical Z
281 * This optimization is off by default for BDW and CHV; turn it on.
283 WA_CLR_BIT_MASKED(CACHE_MODE_0_GEN7
, HIZ_RAW_STALL_OPT_DISABLE
);
285 /* Wa4x4STCOptimizationDisable:bdw,chv */
286 WA_SET_BIT_MASKED(CACHE_MODE_1
, GEN8_4x4_STC_OPTIMIZATION_DISABLE
);
289 * BSpec recommends 8x4 when MSAA is used,
290 * however in practice 16x4 seems fastest.
292 * Note that PS/WM thread counts depend on the WIZ hashing
293 * disable bit, which we don't touch here, but it's good
294 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
296 WA_SET_FIELD_MASKED(GEN7_GT_MODE
,
297 GEN6_WIZ_HASHING_MASK
,
298 GEN6_WIZ_HASHING_16x4
);
301 static void bdw_ctx_workarounds_init(struct intel_engine_cs
*engine
,
302 struct i915_wa_list
*wal
)
304 struct drm_i915_private
*i915
= engine
->i915
;
306 gen8_ctx_workarounds_init(engine
, wal
);
308 /* WaDisableThreadStallDopClockGating:bdw (pre-production) */
309 WA_SET_BIT_MASKED(GEN8_ROW_CHICKEN
, STALL_DOP_GATING_DISABLE
);
311 /* WaDisableDopClockGating:bdw
313 * Also see the related UCGTCL1 write in bdw_init_clock_gating()
314 * to disable EUTC clock gating.
316 WA_SET_BIT_MASKED(GEN7_ROW_CHICKEN2
,
317 DOP_CLOCK_GATING_DISABLE
);
319 WA_SET_BIT_MASKED(HALF_SLICE_CHICKEN3
,
320 GEN8_SAMPLER_POWER_BYPASS_DIS
);
322 WA_SET_BIT_MASKED(HDC_CHICKEN0
,
323 /* WaForceContextSaveRestoreNonCoherent:bdw */
324 HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT
|
325 /* WaDisableFenceDestinationToSLM:bdw (pre-prod) */
326 (IS_BDW_GT3(i915
) ? HDC_FENCE_DEST_SLM_DISABLE
: 0));
329 static void chv_ctx_workarounds_init(struct intel_engine_cs
*engine
,
330 struct i915_wa_list
*wal
)
332 gen8_ctx_workarounds_init(engine
, wal
);
334 /* WaDisableThreadStallDopClockGating:chv */
335 WA_SET_BIT_MASKED(GEN8_ROW_CHICKEN
, STALL_DOP_GATING_DISABLE
);
337 /* Improve HiZ throughput on CHV. */
338 WA_SET_BIT_MASKED(HIZ_CHICKEN
, CHV_HZ_8X8_MODE_IN_1X
);
341 static void gen9_ctx_workarounds_init(struct intel_engine_cs
*engine
,
342 struct i915_wa_list
*wal
)
344 struct drm_i915_private
*i915
= engine
->i915
;
347 /* WaCompressedResourceSamplerPbeMediaNewHashMode:skl,kbl
349 * Must match Display Engine. See
350 * WaCompressedResourceDisplayNewHashMode.
352 WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2
,
353 GEN9_PBE_COMPRESSED_HASH_SELECTION
);
354 WA_SET_BIT_MASKED(GEN9_HALF_SLICE_CHICKEN7
,
355 GEN9_SAMPLER_HASH_COMPRESSED_READ_ADDR
);
358 /* WaClearFlowControlGpgpuContextSave:skl,bxt,kbl,glk,cfl */
359 /* WaDisablePartialInstShootdown:skl,bxt,kbl,glk,cfl */
360 WA_SET_BIT_MASKED(GEN8_ROW_CHICKEN
,
361 FLOW_CONTROL_ENABLE
|
362 PARTIAL_INSTRUCTION_SHOOTDOWN_DISABLE
);
364 /* WaEnableYV12BugFixInHalfSliceChicken7:skl,bxt,kbl,glk,cfl */
365 /* WaEnableSamplerGPGPUPreemptionSupport:skl,bxt,kbl,cfl */
366 WA_SET_BIT_MASKED(GEN9_HALF_SLICE_CHICKEN7
,
367 GEN9_ENABLE_YV12_BUGFIX
|
368 GEN9_ENABLE_GPGPU_PREEMPTION
);
370 /* Wa4x4STCOptimizationDisable:skl,bxt,kbl,glk,cfl */
371 /* WaDisablePartialResolveInVc:skl,bxt,kbl,cfl */
372 WA_SET_BIT_MASKED(CACHE_MODE_1
,
373 GEN8_4x4_STC_OPTIMIZATION_DISABLE
|
374 GEN9_PARTIAL_RESOLVE_IN_VC_DISABLE
);
376 /* WaCcsTlbPrefetchDisable:skl,bxt,kbl,glk,cfl */
377 WA_CLR_BIT_MASKED(GEN9_HALF_SLICE_CHICKEN5
,
378 GEN9_CCS_TLB_PREFETCH_ENABLE
);
380 /* WaForceContextSaveRestoreNonCoherent:skl,bxt,kbl,cfl */
381 WA_SET_BIT_MASKED(HDC_CHICKEN0
,
382 HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT
|
383 HDC_FORCE_CSR_NON_COHERENT_OVR_DISABLE
);
385 /* WaForceEnableNonCoherent and WaDisableHDCInvalidation are
386 * both tied to WaForceContextSaveRestoreNonCoherent
387 * in some hsds for skl. We keep the tie for all gen9. The
388 * documentation is a bit hazy and so we want to get common behaviour,
389 * even though there is no clear evidence we would need both on kbl/bxt.
390 * This area has been source of system hangs so we play it safe
391 * and mimic the skl regardless of what bspec says.
393 * Use Force Non-Coherent whenever executing a 3D context. This
394 * is a workaround for a possible hang in the unlikely event
395 * a TLB invalidation occurs during a PSD flush.
398 /* WaForceEnableNonCoherent:skl,bxt,kbl,cfl */
399 WA_SET_BIT_MASKED(HDC_CHICKEN0
,
400 HDC_FORCE_NON_COHERENT
);
402 /* WaDisableSamplerPowerBypassForSOPingPong:skl,bxt,kbl,cfl */
403 if (IS_SKYLAKE(i915
) ||
405 IS_COFFEELAKE(i915
) ||
407 WA_SET_BIT_MASKED(HALF_SLICE_CHICKEN3
,
408 GEN8_SAMPLER_POWER_BYPASS_DIS
);
410 /* WaDisableSTUnitPowerOptimization:skl,bxt,kbl,glk,cfl */
411 WA_SET_BIT_MASKED(HALF_SLICE_CHICKEN2
, GEN8_ST_PO_DISABLE
);
414 * Supporting preemption with fine-granularity requires changes in the
415 * batch buffer programming. Since we can't break old userspace, we
416 * need to set our default preemption level to safe value. Userspace is
417 * still able to use more fine-grained preemption levels, since in
418 * WaEnablePreemptionGranularityControlByUMD we're whitelisting the
419 * per-ctx register. As such, WaDisable{3D,GPGPU}MidCmdPreemption are
420 * not real HW workarounds, but merely a way to start using preemption
421 * while maintaining old contract with userspace.
424 /* WaDisable3DMidCmdPreemption:skl,bxt,glk,cfl,[cnl] */
425 WA_CLR_BIT_MASKED(GEN8_CS_CHICKEN1
, GEN9_PREEMPT_3D_OBJECT_LEVEL
);
427 /* WaDisableGPGPUMidCmdPreemption:skl,bxt,blk,cfl,[cnl] */
428 WA_SET_FIELD_MASKED(GEN8_CS_CHICKEN1
,
429 GEN9_PREEMPT_GPGPU_LEVEL_MASK
,
430 GEN9_PREEMPT_GPGPU_COMMAND_LEVEL
);
432 /* WaClearHIZ_WM_CHICKEN3:bxt,glk */
433 if (IS_GEN9_LP(i915
))
434 WA_SET_BIT_MASKED(GEN9_WM_CHICKEN3
, GEN9_FACTOR_IN_CLR_VAL_HIZ
);
437 static void skl_tune_iz_hashing(struct intel_engine_cs
*engine
,
438 struct i915_wa_list
*wal
)
440 struct intel_gt
*gt
= engine
->gt
;
441 u8 vals
[3] = { 0, 0, 0 };
444 for (i
= 0; i
< 3; i
++) {
448 * Only consider slices where one, and only one, subslice has 7
451 if (!is_power_of_2(gt
->info
.sseu
.subslice_7eu
[i
]))
455 * subslice_7eu[i] != 0 (because of the check above) and
456 * ss_max == 4 (maximum number of subslices possible per slice)
460 ss
= ffs(gt
->info
.sseu
.subslice_7eu
[i
]) - 1;
464 if (vals
[0] == 0 && vals
[1] == 0 && vals
[2] == 0)
467 /* Tune IZ hashing. See intel_device_info_runtime_init() */
468 WA_SET_FIELD_MASKED(GEN7_GT_MODE
,
469 GEN9_IZ_HASHING_MASK(2) |
470 GEN9_IZ_HASHING_MASK(1) |
471 GEN9_IZ_HASHING_MASK(0),
472 GEN9_IZ_HASHING(2, vals
[2]) |
473 GEN9_IZ_HASHING(1, vals
[1]) |
474 GEN9_IZ_HASHING(0, vals
[0]));
477 static void skl_ctx_workarounds_init(struct intel_engine_cs
*engine
,
478 struct i915_wa_list
*wal
)
480 gen9_ctx_workarounds_init(engine
, wal
);
481 skl_tune_iz_hashing(engine
, wal
);
484 static void bxt_ctx_workarounds_init(struct intel_engine_cs
*engine
,
485 struct i915_wa_list
*wal
)
487 gen9_ctx_workarounds_init(engine
, wal
);
489 /* WaDisableThreadStallDopClockGating:bxt */
490 WA_SET_BIT_MASKED(GEN8_ROW_CHICKEN
,
491 STALL_DOP_GATING_DISABLE
);
493 /* WaToEnableHwFixForPushConstHWBug:bxt */
494 WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2
,
495 GEN8_SBE_DISABLE_REPLAY_BUF_OPTIMIZATION
);
498 static void kbl_ctx_workarounds_init(struct intel_engine_cs
*engine
,
499 struct i915_wa_list
*wal
)
501 struct drm_i915_private
*i915
= engine
->i915
;
503 gen9_ctx_workarounds_init(engine
, wal
);
505 /* WaToEnableHwFixForPushConstHWBug:kbl */
506 if (IS_KBL_GT_REVID(i915
, KBL_REVID_C0
, REVID_FOREVER
))
507 WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2
,
508 GEN8_SBE_DISABLE_REPLAY_BUF_OPTIMIZATION
);
510 /* WaDisableSbeCacheDispatchPortSharing:kbl */
511 WA_SET_BIT_MASKED(GEN7_HALF_SLICE_CHICKEN1
,
512 GEN7_SBE_SS_CACHE_DISPATCH_PORT_SHARING_DISABLE
);
515 static void glk_ctx_workarounds_init(struct intel_engine_cs
*engine
,
516 struct i915_wa_list
*wal
)
518 gen9_ctx_workarounds_init(engine
, wal
);
520 /* WaToEnableHwFixForPushConstHWBug:glk */
521 WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2
,
522 GEN8_SBE_DISABLE_REPLAY_BUF_OPTIMIZATION
);
525 static void cfl_ctx_workarounds_init(struct intel_engine_cs
*engine
,
526 struct i915_wa_list
*wal
)
528 gen9_ctx_workarounds_init(engine
, wal
);
530 /* WaToEnableHwFixForPushConstHWBug:cfl */
531 WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2
,
532 GEN8_SBE_DISABLE_REPLAY_BUF_OPTIMIZATION
);
534 /* WaDisableSbeCacheDispatchPortSharing:cfl */
535 WA_SET_BIT_MASKED(GEN7_HALF_SLICE_CHICKEN1
,
536 GEN7_SBE_SS_CACHE_DISPATCH_PORT_SHARING_DISABLE
);
539 static void cnl_ctx_workarounds_init(struct intel_engine_cs
*engine
,
540 struct i915_wa_list
*wal
)
542 /* WaForceContextSaveRestoreNonCoherent:cnl */
543 WA_SET_BIT_MASKED(CNL_HDC_CHICKEN0
,
544 HDC_FORCE_CONTEXT_SAVE_RESTORE_NON_COHERENT
);
546 /* WaDisableReplayBufferBankArbitrationOptimization:cnl */
547 WA_SET_BIT_MASKED(COMMON_SLICE_CHICKEN2
,
548 GEN8_SBE_DISABLE_REPLAY_BUF_OPTIMIZATION
);
550 /* WaPushConstantDereferenceHoldDisable:cnl */
551 WA_SET_BIT_MASKED(GEN7_ROW_CHICKEN2
, PUSH_CONSTANT_DEREF_DISABLE
);
553 /* FtrEnableFastAnisoL1BankingFix:cnl */
554 WA_SET_BIT_MASKED(HALF_SLICE_CHICKEN3
, CNL_FAST_ANISO_L1_BANKING_FIX
);
556 /* WaDisable3DMidCmdPreemption:cnl */
557 WA_CLR_BIT_MASKED(GEN8_CS_CHICKEN1
, GEN9_PREEMPT_3D_OBJECT_LEVEL
);
559 /* WaDisableGPGPUMidCmdPreemption:cnl */
560 WA_SET_FIELD_MASKED(GEN8_CS_CHICKEN1
,
561 GEN9_PREEMPT_GPGPU_LEVEL_MASK
,
562 GEN9_PREEMPT_GPGPU_COMMAND_LEVEL
);
564 /* WaDisableEarlyEOT:cnl */
565 WA_SET_BIT_MASKED(GEN8_ROW_CHICKEN
, DISABLE_EARLY_EOT
);
568 static void icl_ctx_workarounds_init(struct intel_engine_cs
*engine
,
569 struct i915_wa_list
*wal
)
571 struct drm_i915_private
*i915
= engine
->i915
;
573 /* WaDisableBankHangMode:icl */
576 intel_uncore_read(engine
->uncore
, GEN8_L3CNTLREG
) |
579 /* Wa_1604370585:icl (pre-prod)
580 * Formerly known as WaPushConstantDereferenceHoldDisable
582 if (IS_ICL_REVID(i915
, ICL_REVID_A0
, ICL_REVID_B0
))
583 WA_SET_BIT_MASKED(GEN7_ROW_CHICKEN2
,
584 PUSH_CONSTANT_DEREF_DISABLE
);
586 /* WaForceEnableNonCoherent:icl
587 * This is not the same workaround as in early Gen9 platforms, where
588 * lacking this could cause system hangs, but coherency performance
589 * overhead is high and only a few compute workloads really need it
590 * (the register is whitelisted in hardware now, so UMDs can opt in
591 * for coherency if they have a good reason).
593 WA_SET_BIT_MASKED(ICL_HDC_MODE
, HDC_FORCE_NON_COHERENT
);
595 /* Wa_2006611047:icl (pre-prod)
596 * Formerly known as WaDisableImprovedTdlClkGating
598 if (IS_ICL_REVID(i915
, ICL_REVID_A0
, ICL_REVID_A0
))
599 WA_SET_BIT_MASKED(GEN7_ROW_CHICKEN2
,
600 GEN11_TDL_CLOCK_GATING_FIX_DISABLE
);
602 /* Wa_2006665173:icl (pre-prod) */
603 if (IS_ICL_REVID(i915
, ICL_REVID_A0
, ICL_REVID_A0
))
604 WA_SET_BIT_MASKED(GEN11_COMMON_SLICE_CHICKEN3
,
605 GEN11_BLEND_EMB_FIX_DISABLE_IN_RCC
);
607 /* WaEnableFloatBlendOptimization:icl */
608 wa_write_masked_or(wal
,
610 0, /* write-only, so skip validation */
611 _MASKED_BIT_ENABLE(FLOAT_BLEND_OPTIMIZATION_ENABLE
));
613 /* WaDisableGPGPUMidThreadPreemption:icl */
614 WA_SET_FIELD_MASKED(GEN8_CS_CHICKEN1
,
615 GEN9_PREEMPT_GPGPU_LEVEL_MASK
,
616 GEN9_PREEMPT_GPGPU_THREAD_GROUP_LEVEL
);
618 /* allow headerless messages for preemptible GPGPU context */
619 WA_SET_BIT_MASKED(GEN10_SAMPLER_MODE
,
620 GEN11_SAMPLER_ENABLE_HEADLESS_MSG
);
622 /* Wa_1604278689:icl,ehl */
623 wa_write(wal
, IVB_FBC_RT_BASE
, 0xFFFFFFFF & ~ILK_FBC_RT_VALID
);
624 wa_write_masked_or(wal
, IVB_FBC_RT_BASE_UPPER
,
625 0, /* write-only register; skip validation */
628 /* Wa_1406306137:icl,ehl */
629 wa_masked_en(wal
, GEN9_ROW_CHICKEN4
, GEN11_DIS_PICK_2ND_EU
);
632 static void gen12_ctx_workarounds_init(struct intel_engine_cs
*engine
,
633 struct i915_wa_list
*wal
)
646 WA_SET_BIT_MASKED(GEN11_COMMON_SLICE_CHICKEN3
,
647 GEN12_DISABLE_CPS_AWARE_COLOR_PIPE
);
649 /* WaDisableGPGPUMidThreadPreemption:gen12 */
650 WA_SET_FIELD_MASKED(GEN8_CS_CHICKEN1
,
651 GEN9_PREEMPT_GPGPU_LEVEL_MASK
,
652 GEN9_PREEMPT_GPGPU_THREAD_GROUP_LEVEL
);
655 static void tgl_ctx_workarounds_init(struct intel_engine_cs
*engine
,
656 struct i915_wa_list
*wal
)
658 gen12_ctx_workarounds_init(engine
, wal
);
661 * Wa_1604555607:tgl,rkl
663 * Note that the implementation of this workaround is further modified
664 * according to the FF_MODE2 guidance given by Wa_1608008084:gen12.
665 * FF_MODE2 register will return the wrong value when read. The default
666 * value for this register is zero for all fields and there are no bit
667 * masks. So instead of doing a RMW we should just write the GS Timer
668 * and TDS timer values for Wa_1604555607 and Wa_16011163337.
672 FF_MODE2_GS_TIMER_MASK
| FF_MODE2_TDS_TIMER_MASK
,
673 FF_MODE2_GS_TIMER_224
| FF_MODE2_TDS_TIMER_128
,
677 static void dg1_ctx_workarounds_init(struct intel_engine_cs
*engine
,
678 struct i915_wa_list
*wal
)
680 gen12_ctx_workarounds_init(engine
, wal
);
683 WA_CLR_BIT_MASKED(GEN11_COMMON_SLICE_CHICKEN3
,
684 DG1_FLOAT_POINT_BLEND_OPT_STRICT_MODE_EN
);
687 WA_SET_BIT_MASKED(HIZ_CHICKEN
,
688 DG1_HZ_READ_SUPPRESSION_OPTIMIZATION_DISABLE
);
692 __intel_engine_init_ctx_wa(struct intel_engine_cs
*engine
,
693 struct i915_wa_list
*wal
,
696 struct drm_i915_private
*i915
= engine
->i915
;
698 if (engine
->class != RENDER_CLASS
)
701 wa_init_start(wal
, name
, engine
->name
);
704 dg1_ctx_workarounds_init(engine
, wal
);
705 else if (IS_ROCKETLAKE(i915
) || IS_TIGERLAKE(i915
))
706 tgl_ctx_workarounds_init(engine
, wal
);
707 else if (IS_GEN(i915
, 12))
708 gen12_ctx_workarounds_init(engine
, wal
);
709 else if (IS_GEN(i915
, 11))
710 icl_ctx_workarounds_init(engine
, wal
);
711 else if (IS_CANNONLAKE(i915
))
712 cnl_ctx_workarounds_init(engine
, wal
);
713 else if (IS_COFFEELAKE(i915
) || IS_COMETLAKE(i915
))
714 cfl_ctx_workarounds_init(engine
, wal
);
715 else if (IS_GEMINILAKE(i915
))
716 glk_ctx_workarounds_init(engine
, wal
);
717 else if (IS_KABYLAKE(i915
))
718 kbl_ctx_workarounds_init(engine
, wal
);
719 else if (IS_BROXTON(i915
))
720 bxt_ctx_workarounds_init(engine
, wal
);
721 else if (IS_SKYLAKE(i915
))
722 skl_ctx_workarounds_init(engine
, wal
);
723 else if (IS_CHERRYVIEW(i915
))
724 chv_ctx_workarounds_init(engine
, wal
);
725 else if (IS_BROADWELL(i915
))
726 bdw_ctx_workarounds_init(engine
, wal
);
727 else if (IS_GEN(i915
, 7))
728 gen7_ctx_workarounds_init(engine
, wal
);
729 else if (IS_GEN(i915
, 6))
730 gen6_ctx_workarounds_init(engine
, wal
);
731 else if (INTEL_GEN(i915
) < 8)
734 MISSING_CASE(INTEL_GEN(i915
));
739 void intel_engine_init_ctx_wa(struct intel_engine_cs
*engine
)
741 __intel_engine_init_ctx_wa(engine
, &engine
->ctx_wa_list
, "context");
744 int intel_engine_emit_ctx_wa(struct i915_request
*rq
)
746 struct i915_wa_list
*wal
= &rq
->engine
->ctx_wa_list
;
755 ret
= rq
->engine
->emit_flush(rq
, EMIT_BARRIER
);
759 cs
= intel_ring_begin(rq
, (wal
->count
* 2 + 2));
763 *cs
++ = MI_LOAD_REGISTER_IMM(wal
->count
);
764 for (i
= 0, wa
= wal
->list
; i
< wal
->count
; i
++, wa
++) {
765 *cs
++ = i915_mmio_reg_offset(wa
->reg
);
770 intel_ring_advance(rq
, cs
);
772 ret
= rq
->engine
->emit_flush(rq
, EMIT_BARRIER
);
780 gen4_gt_workarounds_init(struct drm_i915_private
*i915
,
781 struct i915_wa_list
*wal
)
783 /* WaDisable_RenderCache_OperationalFlush:gen4,ilk */
784 wa_masked_dis(wal
, CACHE_MODE_0
, RC_OP_FLUSH_ENABLE
);
788 g4x_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
790 gen4_gt_workarounds_init(i915
, wal
);
792 /* WaDisableRenderCachePipelinedFlush:g4x,ilk */
793 wa_masked_en(wal
, CACHE_MODE_0
, CM0_PIPELINED_RENDER_FLUSH_DISABLE
);
797 ilk_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
799 g4x_gt_workarounds_init(i915
, wal
);
801 wa_masked_en(wal
, _3D_CHICKEN2
, _3D_CHICKEN2_WM_READ_PIPELINED
);
805 snb_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
807 /* WaDisableHiZPlanesWhenMSAAEnabled:snb */
810 _3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB
);
812 /* WaDisable_RenderCache_OperationalFlush:snb */
813 wa_masked_dis(wal
, CACHE_MODE_0
, RC_OP_FLUSH_ENABLE
);
816 * BSpec recommends 8x4 when MSAA is used,
817 * however in practice 16x4 seems fastest.
819 * Note that PS/WM thread counts depend on the WIZ hashing
820 * disable bit, which we don't touch here, but it's good
821 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
825 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK
, GEN6_WIZ_HASHING_16x4
),
826 GEN6_WIZ_HASHING_16x4
);
828 wa_masked_dis(wal
, CACHE_MODE_0
, CM0_STC_EVICT_DISABLE_LRA_SNB
);
832 /* WaStripsFansDisableFastClipPerformanceFix:snb */
833 _3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL
|
836 * "This bit must be set if 3DSTATE_CLIP clip mode is set
837 * to normal and 3DSTATE_SF number of SF output attributes
840 _3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH
);
844 ivb_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
846 /* WaDisableEarlyCull:ivb */
847 wa_masked_en(wal
, _3D_CHICKEN3
, _3D_CHICKEN_SF_DISABLE_OBJEND_CULL
);
849 /* WaDisablePSDDualDispatchEnable:ivb */
850 if (IS_IVB_GT1(i915
))
852 GEN7_HALF_SLICE_CHICKEN1
,
853 GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE
);
855 /* WaDisable_RenderCache_OperationalFlush:ivb */
856 wa_masked_dis(wal
, CACHE_MODE_0_GEN7
, RC_OP_FLUSH_ENABLE
);
858 /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */
860 GEN7_COMMON_SLICE_CHICKEN1
,
861 GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC
);
863 /* WaApplyL3ControlAndL3ChickenMode:ivb */
864 wa_write(wal
, GEN7_L3CNTLREG1
, GEN7_WA_FOR_GEN7_L3_CONTROL
);
865 wa_write(wal
, GEN7_L3_CHICKEN_MODE_REGISTER
, GEN7_WA_L3_CHICKEN_MODE
);
867 /* WaForceL3Serialization:ivb */
868 wa_write_clr(wal
, GEN7_L3SQCREG4
, L3SQ_URB_READ_CAM_MATCH_DISABLE
);
871 * WaVSThreadDispatchOverride:ivb,vlv
873 * This actually overrides the dispatch
874 * mode for all thread types.
876 wa_write_masked_or(wal
, GEN7_FF_THREAD_MODE
,
878 GEN7_FF_TS_SCHED_HW
|
879 GEN7_FF_VS_SCHED_HW
|
880 GEN7_FF_DS_SCHED_HW
);
882 if (0) { /* causes HiZ corruption on ivb:gt1 */
883 /* enable HiZ Raw Stall Optimization */
884 wa_masked_dis(wal
, CACHE_MODE_0_GEN7
, HIZ_RAW_STALL_OPT_DISABLE
);
887 /* WaDisable4x2SubspanOptimization:ivb */
888 wa_masked_en(wal
, CACHE_MODE_1
, PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
);
891 * BSpec recommends 8x4 when MSAA is used,
892 * however in practice 16x4 seems fastest.
894 * Note that PS/WM thread counts depend on the WIZ hashing
895 * disable bit, which we don't touch here, but it's good
896 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
898 wa_add(wal
, GEN7_GT_MODE
, 0,
899 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK
, GEN6_WIZ_HASHING_16x4
),
900 GEN6_WIZ_HASHING_16x4
);
904 vlv_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
906 /* WaDisableEarlyCull:vlv */
907 wa_masked_en(wal
, _3D_CHICKEN3
, _3D_CHICKEN_SF_DISABLE_OBJEND_CULL
);
909 /* WaPsdDispatchEnable:vlv */
910 /* WaDisablePSDDualDispatchEnable:vlv */
912 GEN7_HALF_SLICE_CHICKEN1
,
913 GEN7_MAX_PS_THREAD_DEP
|
914 GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE
);
916 /* WaDisable_RenderCache_OperationalFlush:vlv */
917 wa_masked_dis(wal
, CACHE_MODE_0_GEN7
, RC_OP_FLUSH_ENABLE
);
919 /* WaForceL3Serialization:vlv */
920 wa_write_clr(wal
, GEN7_L3SQCREG4
, L3SQ_URB_READ_CAM_MATCH_DISABLE
);
923 * WaVSThreadDispatchOverride:ivb,vlv
925 * This actually overrides the dispatch
926 * mode for all thread types.
928 wa_write_masked_or(wal
,
931 GEN7_FF_TS_SCHED_HW
|
932 GEN7_FF_VS_SCHED_HW
|
933 GEN7_FF_DS_SCHED_HW
);
936 * BSpec says this must be set, even though
937 * WaDisable4x2SubspanOptimization isn't listed for VLV.
939 wa_masked_en(wal
, CACHE_MODE_1
, PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
);
942 * BSpec recommends 8x4 when MSAA is used,
943 * however in practice 16x4 seems fastest.
945 * Note that PS/WM thread counts depend on the WIZ hashing
946 * disable bit, which we don't touch here, but it's good
947 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
949 wa_add(wal
, GEN7_GT_MODE
, 0,
950 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK
, GEN6_WIZ_HASHING_16x4
),
951 GEN6_WIZ_HASHING_16x4
);
954 * WaIncreaseL3CreditsForVLVB0:vlv
955 * This is the hardware default actually.
957 wa_write(wal
, GEN7_L3SQCREG1
, VLV_B0_WA_L3SQCREG1_VALUE
);
961 hsw_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
963 /* L3 caching of data atomics doesn't work -- disable it. */
964 wa_write(wal
, HSW_SCRATCH1
, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE
);
968 _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE
),
969 0 /* XXX does this reg exist? */);
971 /* WaVSRefCountFullforceMissDisable:hsw */
972 wa_write_clr(wal
, GEN7_FF_THREAD_MODE
, GEN7_FF_VS_REF_CNT_FFME
);
976 /* WaDisable_RenderCache_OperationalFlush:hsw */
978 /* enable HiZ Raw Stall Optimization */
979 HIZ_RAW_STALL_OPT_DISABLE
);
981 /* WaDisable4x2SubspanOptimization:hsw */
982 wa_masked_en(wal
, CACHE_MODE_1
, PIXEL_SUBSPAN_COLLECT_OPT_DISABLE
);
985 * BSpec recommends 8x4 when MSAA is used,
986 * however in practice 16x4 seems fastest.
988 * Note that PS/WM thread counts depend on the WIZ hashing
989 * disable bit, which we don't touch here, but it's good
990 * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
992 wa_add(wal
, GEN7_GT_MODE
, 0,
993 _MASKED_FIELD(GEN6_WIZ_HASHING_MASK
, GEN6_WIZ_HASHING_16x4
),
994 GEN6_WIZ_HASHING_16x4
);
996 /* WaSampleCChickenBitEnable:hsw */
997 wa_masked_en(wal
, HALF_SLICE_CHICKEN3
, HSW_SAMPLE_C_PERFORMANCE
);
1001 gen9_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
1003 /* WaDisableKillLogic:bxt,skl,kbl */
1004 if (!IS_COFFEELAKE(i915
) && !IS_COMETLAKE(i915
))
1009 if (HAS_LLC(i915
)) {
1010 /* WaCompressedResourceSamplerPbeMediaNewHashMode:skl,kbl
1012 * Must match Display Engine. See
1013 * WaCompressedResourceDisplayNewHashMode.
1017 MMCD_PCLA
| MMCD_HOTSPOT_EN
);
1020 /* WaDisableHDCInvalidation:skl,bxt,kbl,cfl */
1023 BDW_DISABLE_HDC_INVALIDATION
);
1027 skl_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
1029 gen9_gt_workarounds_init(i915
, wal
);
1031 /* WaDisableGafsUnitClkGating:skl */
1034 GEN8_EU_GAUNIT_CLOCK_GATE_DISABLE
);
1036 /* WaInPlaceDecompressionHang:skl */
1037 if (IS_SKL_REVID(i915
, SKL_REVID_H0
, REVID_FOREVER
))
1039 GEN9_GAMT_ECO_REG_RW_IA
,
1040 GAMT_ECO_ENABLE_IN_PLACE_DECOMPRESS
);
1044 bxt_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
1046 gen9_gt_workarounds_init(i915
, wal
);
1048 /* WaInPlaceDecompressionHang:bxt */
1050 GEN9_GAMT_ECO_REG_RW_IA
,
1051 GAMT_ECO_ENABLE_IN_PLACE_DECOMPRESS
);
1055 kbl_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
1057 gen9_gt_workarounds_init(i915
, wal
);
1059 /* WaDisableDynamicCreditSharing:kbl */
1060 if (IS_KBL_GT_REVID(i915
, 0, KBL_REVID_B0
))
1063 GAMT_CHKN_DISABLE_DYNAMIC_CREDIT_SHARING
);
1065 /* WaDisableGafsUnitClkGating:kbl */
1068 GEN8_EU_GAUNIT_CLOCK_GATE_DISABLE
);
1070 /* WaInPlaceDecompressionHang:kbl */
1072 GEN9_GAMT_ECO_REG_RW_IA
,
1073 GAMT_ECO_ENABLE_IN_PLACE_DECOMPRESS
);
1077 glk_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
1079 gen9_gt_workarounds_init(i915
, wal
);
1083 cfl_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
1085 gen9_gt_workarounds_init(i915
, wal
);
1087 /* WaDisableGafsUnitClkGating:cfl */
1090 GEN8_EU_GAUNIT_CLOCK_GATE_DISABLE
);
1092 /* WaInPlaceDecompressionHang:cfl */
1094 GEN9_GAMT_ECO_REG_RW_IA
,
1095 GAMT_ECO_ENABLE_IN_PLACE_DECOMPRESS
);
1099 wa_init_mcr(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
1101 const struct sseu_dev_info
*sseu
= &i915
->gt
.info
.sseu
;
1102 unsigned int slice
, subslice
;
1103 u32 l3_en
, mcr
, mcr_mask
;
1105 GEM_BUG_ON(INTEL_GEN(i915
) < 10);
1108 * WaProgramMgsrForL3BankSpecificMmioReads: cnl,icl
1109 * L3Banks could be fused off in single slice scenario. If that is
1110 * the case, we might need to program MCR select to a valid L3Bank
1111 * by default, to make sure we correctly read certain registers
1112 * later on (in the range 0xB100 - 0xB3FF).
1114 * WaProgramMgsrForCorrectSliceSpecificMmioReads:cnl,icl
1115 * Before any MMIO read into slice/subslice specific registers, MCR
1116 * packet control register needs to be programmed to point to any
1117 * enabled s/ss pair. Otherwise, incorrect values will be returned.
1118 * This means each subsequent MMIO read will be forwarded to an
1119 * specific s/ss combination, but this is OK since these registers
1120 * are consistent across s/ss in almost all cases. In the rare
1121 * occasions, such as INSTDONE, where this value is dependent
1122 * on s/ss combo, the read should be done with read_subslice_reg.
1124 * Since GEN8_MCR_SELECTOR contains dual-purpose bits which select both
1125 * to which subslice, or to which L3 bank, the respective mmio reads
1126 * will go, we have to find a common index which works for both
1129 * Case where we cannot find a common index fortunately should not
1130 * happen in production hardware, so we only emit a warning instead of
1131 * implementing something more complex that requires checking the range
1132 * of every MMIO read.
1135 if (INTEL_GEN(i915
) >= 10 && is_power_of_2(sseu
->slice_mask
)) {
1137 intel_uncore_read(&i915
->uncore
, GEN10_MIRROR_FUSE3
) &
1140 drm_dbg(&i915
->drm
, "L3 fuse = %x\n", l3_fuse
);
1141 l3_en
= ~(l3_fuse
<< GEN10_L3BANK_PAIR_COUNT
| l3_fuse
);
1146 slice
= fls(sseu
->slice_mask
) - 1;
1147 subslice
= fls(l3_en
& intel_sseu_get_subslices(sseu
, slice
));
1149 drm_warn(&i915
->drm
,
1150 "No common index found between subslice mask %x and L3 bank mask %x!\n",
1151 intel_sseu_get_subslices(sseu
, slice
), l3_en
);
1152 subslice
= fls(l3_en
);
1153 drm_WARN_ON(&i915
->drm
, !subslice
);
1157 if (INTEL_GEN(i915
) >= 11) {
1158 mcr
= GEN11_MCR_SLICE(slice
) | GEN11_MCR_SUBSLICE(subslice
);
1159 mcr_mask
= GEN11_MCR_SLICE_MASK
| GEN11_MCR_SUBSLICE_MASK
;
1161 mcr
= GEN8_MCR_SLICE(slice
) | GEN8_MCR_SUBSLICE(subslice
);
1162 mcr_mask
= GEN8_MCR_SLICE_MASK
| GEN8_MCR_SUBSLICE_MASK
;
1165 drm_dbg(&i915
->drm
, "MCR slice/subslice = %x\n", mcr
);
1167 wa_write_masked_or(wal
, GEN8_MCR_SELECTOR
, mcr_mask
, mcr
);
1171 cnl_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
1173 wa_init_mcr(i915
, wal
);
1175 /* WaInPlaceDecompressionHang:cnl */
1177 GEN9_GAMT_ECO_REG_RW_IA
,
1178 GAMT_ECO_ENABLE_IN_PLACE_DECOMPRESS
);
1182 icl_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
1184 wa_init_mcr(i915
, wal
);
1186 /* WaInPlaceDecompressionHang:icl */
1188 GEN9_GAMT_ECO_REG_RW_IA
,
1189 GAMT_ECO_ENABLE_IN_PLACE_DECOMPRESS
);
1191 /* WaModifyGamTlbPartitioning:icl */
1192 wa_write_masked_or(wal
,
1193 GEN11_GACB_PERF_CTRL
,
1194 GEN11_HASH_CTRL_MASK
,
1195 GEN11_HASH_CTRL_BIT0
| GEN11_HASH_CTRL_BIT4
);
1197 /* Wa_1405766107:icl
1198 * Formerly known as WaCL2SFHalfMaxAlloc
1202 GEN11_LSN_UNSLCVC_GAFS_HALF_SF_MAXALLOC
|
1203 GEN11_LSN_UNSLCVC_GAFS_HALF_CL2_MAXALLOC
);
1206 * Formerly known as WaDisCtxReload
1209 GEN8_GAMW_ECO_DEV_RW_IA
,
1210 GAMW_ECO_DEV_CTX_RELOAD_DISABLE
);
1212 /* Wa_1405779004:icl (pre-prod) */
1213 if (IS_ICL_REVID(i915
, ICL_REVID_A0
, ICL_REVID_A0
))
1215 SLICE_UNIT_LEVEL_CLKGATE
,
1216 MSCUNIT_CLKGATE_DIS
);
1218 /* Wa_1406838659:icl (pre-prod) */
1219 if (IS_ICL_REVID(i915
, ICL_REVID_A0
, ICL_REVID_B0
))
1221 INF_UNIT_LEVEL_CLKGATE
,
1224 /* Wa_1406463099:icl
1225 * Formerly known as WaGamTlbPendError
1229 GAMT_CHKN_DISABLE_L3_COH_PIPE
);
1231 /* Wa_1607087056:icl,ehl,jsl */
1232 if (IS_ICELAKE(i915
) ||
1233 IS_JSL_EHL_REVID(i915
, EHL_REVID_A0
, EHL_REVID_A0
)) {
1235 SLICE_UNIT_LEVEL_CLKGATE
,
1236 L3_CLKGATE_DIS
| L3_CR2X_CLKGATE_DIS
);
1241 gen12_gt_workarounds_init(struct drm_i915_private
*i915
,
1242 struct i915_wa_list
*wal
)
1244 wa_init_mcr(i915
, wal
);
1248 tgl_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
1250 gen12_gt_workarounds_init(i915
, wal
);
1252 /* Wa_1409420604:tgl */
1253 if (IS_TGL_UY_GT_REVID(i915
, TGL_REVID_A0
, TGL_REVID_A0
))
1255 SUBSLICE_UNIT_LEVEL_CLKGATE2
,
1256 CPSSUNIT_CLKGATE_DIS
);
1258 /* Wa_1607087056:tgl also know as BUG:1409180338 */
1259 if (IS_TGL_UY_GT_REVID(i915
, TGL_REVID_A0
, TGL_REVID_A0
))
1261 SLICE_UNIT_LEVEL_CLKGATE
,
1262 L3_CLKGATE_DIS
| L3_CR2X_CLKGATE_DIS
);
1266 dg1_gt_workarounds_init(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
1268 gen12_gt_workarounds_init(i915
, wal
);
1270 /* Wa_1607087056:dg1 */
1271 if (IS_DG1_REVID(i915
, DG1_REVID_A0
, DG1_REVID_A0
))
1273 SLICE_UNIT_LEVEL_CLKGATE
,
1274 L3_CLKGATE_DIS
| L3_CR2X_CLKGATE_DIS
);
1276 /* Wa_1409420604:dg1 */
1279 SUBSLICE_UNIT_LEVEL_CLKGATE2
,
1280 CPSSUNIT_CLKGATE_DIS
);
1282 /* Wa_1408615072:dg1 */
1283 /* Empirical testing shows this register is unaffected by engine reset. */
1285 wa_write_or(wal
, UNSLICE_UNIT_LEVEL_CLKGATE2
,
1286 VSUNIT_CLKGATE_DIS_TGL
);
1290 gt_init_workarounds(struct drm_i915_private
*i915
, struct i915_wa_list
*wal
)
1293 dg1_gt_workarounds_init(i915
, wal
);
1294 else if (IS_TIGERLAKE(i915
))
1295 tgl_gt_workarounds_init(i915
, wal
);
1296 else if (IS_GEN(i915
, 12))
1297 gen12_gt_workarounds_init(i915
, wal
);
1298 else if (IS_GEN(i915
, 11))
1299 icl_gt_workarounds_init(i915
, wal
);
1300 else if (IS_CANNONLAKE(i915
))
1301 cnl_gt_workarounds_init(i915
, wal
);
1302 else if (IS_COFFEELAKE(i915
) || IS_COMETLAKE(i915
))
1303 cfl_gt_workarounds_init(i915
, wal
);
1304 else if (IS_GEMINILAKE(i915
))
1305 glk_gt_workarounds_init(i915
, wal
);
1306 else if (IS_KABYLAKE(i915
))
1307 kbl_gt_workarounds_init(i915
, wal
);
1308 else if (IS_BROXTON(i915
))
1309 bxt_gt_workarounds_init(i915
, wal
);
1310 else if (IS_SKYLAKE(i915
))
1311 skl_gt_workarounds_init(i915
, wal
);
1312 else if (IS_HASWELL(i915
))
1313 hsw_gt_workarounds_init(i915
, wal
);
1314 else if (IS_VALLEYVIEW(i915
))
1315 vlv_gt_workarounds_init(i915
, wal
);
1316 else if (IS_IVYBRIDGE(i915
))
1317 ivb_gt_workarounds_init(i915
, wal
);
1318 else if (IS_GEN(i915
, 6))
1319 snb_gt_workarounds_init(i915
, wal
);
1320 else if (IS_GEN(i915
, 5))
1321 ilk_gt_workarounds_init(i915
, wal
);
1322 else if (IS_G4X(i915
))
1323 g4x_gt_workarounds_init(i915
, wal
);
1324 else if (IS_GEN(i915
, 4))
1325 gen4_gt_workarounds_init(i915
, wal
);
1326 else if (INTEL_GEN(i915
) <= 8)
1329 MISSING_CASE(INTEL_GEN(i915
));
1332 void intel_gt_init_workarounds(struct drm_i915_private
*i915
)
1334 struct i915_wa_list
*wal
= &i915
->gt_wa_list
;
1336 wa_init_start(wal
, "GT", "global");
1337 gt_init_workarounds(i915
, wal
);
1338 wa_init_finish(wal
);
1341 static enum forcewake_domains
1342 wal_get_fw_for_rmw(struct intel_uncore
*uncore
, const struct i915_wa_list
*wal
)
1344 enum forcewake_domains fw
= 0;
1348 for (i
= 0, wa
= wal
->list
; i
< wal
->count
; i
++, wa
++)
1349 fw
|= intel_uncore_forcewake_for_reg(uncore
,
1358 wa_verify(const struct i915_wa
*wa
, u32 cur
, const char *name
, const char *from
)
1360 if ((cur
^ wa
->set
) & wa
->read
) {
1361 DRM_ERROR("%s workaround lost on %s! (%x=%x/%x, expected %x)\n",
1362 name
, from
, i915_mmio_reg_offset(wa
->reg
),
1363 cur
, cur
& wa
->read
, wa
->set
);
1372 wa_list_apply(struct intel_uncore
*uncore
, const struct i915_wa_list
*wal
)
1374 enum forcewake_domains fw
;
1375 unsigned long flags
;
1382 fw
= wal_get_fw_for_rmw(uncore
, wal
);
1384 spin_lock_irqsave(&uncore
->lock
, flags
);
1385 intel_uncore_forcewake_get__locked(uncore
, fw
);
1387 for (i
= 0, wa
= wal
->list
; i
< wal
->count
; i
++, wa
++) {
1389 intel_uncore_rmw_fw(uncore
, wa
->reg
, wa
->clr
, wa
->set
);
1391 intel_uncore_write_fw(uncore
, wa
->reg
, wa
->set
);
1392 if (IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM
))
1394 intel_uncore_read_fw(uncore
, wa
->reg
),
1395 wal
->name
, "application");
1398 intel_uncore_forcewake_put__locked(uncore
, fw
);
1399 spin_unlock_irqrestore(&uncore
->lock
, flags
);
1402 void intel_gt_apply_workarounds(struct intel_gt
*gt
)
1404 wa_list_apply(gt
->uncore
, >
->i915
->gt_wa_list
);
1407 static bool wa_list_verify(struct intel_uncore
*uncore
,
1408 const struct i915_wa_list
*wal
,
1415 for (i
= 0, wa
= wal
->list
; i
< wal
->count
; i
++, wa
++)
1417 intel_uncore_read(uncore
, wa
->reg
),
1423 bool intel_gt_verify_workarounds(struct intel_gt
*gt
, const char *from
)
1425 return wa_list_verify(gt
->uncore
, >
->i915
->gt_wa_list
, from
);
1428 static inline bool is_nonpriv_flags_valid(u32 flags
)
1430 /* Check only valid flag bits are set */
1431 if (flags
& ~RING_FORCE_TO_NONPRIV_MASK_VALID
)
1434 /* NB: Only 3 out of 4 enum values are valid for access field */
1435 if ((flags
& RING_FORCE_TO_NONPRIV_ACCESS_MASK
) ==
1436 RING_FORCE_TO_NONPRIV_ACCESS_INVALID
)
1443 whitelist_reg_ext(struct i915_wa_list
*wal
, i915_reg_t reg
, u32 flags
)
1445 struct i915_wa wa
= {
1449 if (GEM_DEBUG_WARN_ON(wal
->count
>= RING_MAX_NONPRIV_SLOTS
))
1452 if (GEM_DEBUG_WARN_ON(!is_nonpriv_flags_valid(flags
)))
1455 wa
.reg
.reg
|= flags
;
1460 whitelist_reg(struct i915_wa_list
*wal
, i915_reg_t reg
)
1462 whitelist_reg_ext(wal
, reg
, RING_FORCE_TO_NONPRIV_ACCESS_RW
);
1465 static void gen9_whitelist_build(struct i915_wa_list
*w
)
1467 /* WaVFEStateAfterPipeControlwithMediaStateClear:skl,bxt,glk,cfl */
1468 whitelist_reg(w
, GEN9_CTX_PREEMPT_REG
);
1470 /* WaEnablePreemptionGranularityControlByUMD:skl,bxt,kbl,cfl,[cnl] */
1471 whitelist_reg(w
, GEN8_CS_CHICKEN1
);
1473 /* WaAllowUMDToModifyHDCChicken1:skl,bxt,kbl,glk,cfl */
1474 whitelist_reg(w
, GEN8_HDC_CHICKEN1
);
1476 /* WaSendPushConstantsFromMMIO:skl,bxt */
1477 whitelist_reg(w
, COMMON_SLICE_CHICKEN2
);
1480 static void skl_whitelist_build(struct intel_engine_cs
*engine
)
1482 struct i915_wa_list
*w
= &engine
->whitelist
;
1484 if (engine
->class != RENDER_CLASS
)
1487 gen9_whitelist_build(w
);
1489 /* WaDisableLSQCROPERFforOCL:skl */
1490 whitelist_reg(w
, GEN8_L3SQCREG4
);
1493 static void bxt_whitelist_build(struct intel_engine_cs
*engine
)
1495 if (engine
->class != RENDER_CLASS
)
1498 gen9_whitelist_build(&engine
->whitelist
);
1501 static void kbl_whitelist_build(struct intel_engine_cs
*engine
)
1503 struct i915_wa_list
*w
= &engine
->whitelist
;
1505 if (engine
->class != RENDER_CLASS
)
1508 gen9_whitelist_build(w
);
1510 /* WaDisableLSQCROPERFforOCL:kbl */
1511 whitelist_reg(w
, GEN8_L3SQCREG4
);
1514 static void glk_whitelist_build(struct intel_engine_cs
*engine
)
1516 struct i915_wa_list
*w
= &engine
->whitelist
;
1518 if (engine
->class != RENDER_CLASS
)
1521 gen9_whitelist_build(w
);
1523 /* WA #0862: Userspace has to set "Barrier Mode" to avoid hangs. */
1524 whitelist_reg(w
, GEN9_SLICE_COMMON_ECO_CHICKEN1
);
1527 static void cfl_whitelist_build(struct intel_engine_cs
*engine
)
1529 struct i915_wa_list
*w
= &engine
->whitelist
;
1531 if (engine
->class != RENDER_CLASS
)
1534 gen9_whitelist_build(w
);
1537 * WaAllowPMDepthAndInvocationCountAccessFromUMD:cfl,whl,cml,aml
1539 * This covers 4 register which are next to one another :
1540 * - PS_INVOCATION_COUNT
1541 * - PS_INVOCATION_COUNT_UDW
1543 * - PS_DEPTH_COUNT_UDW
1545 whitelist_reg_ext(w
, PS_INVOCATION_COUNT
,
1546 RING_FORCE_TO_NONPRIV_ACCESS_RD
|
1547 RING_FORCE_TO_NONPRIV_RANGE_4
);
1550 static void cml_whitelist_build(struct intel_engine_cs
*engine
)
1552 struct i915_wa_list
*w
= &engine
->whitelist
;
1554 if (engine
->class != RENDER_CLASS
)
1555 whitelist_reg_ext(w
,
1556 RING_CTX_TIMESTAMP(engine
->mmio_base
),
1557 RING_FORCE_TO_NONPRIV_ACCESS_RD
);
1559 cfl_whitelist_build(engine
);
1562 static void cnl_whitelist_build(struct intel_engine_cs
*engine
)
1564 struct i915_wa_list
*w
= &engine
->whitelist
;
1566 if (engine
->class != RENDER_CLASS
)
1569 /* WaEnablePreemptionGranularityControlByUMD:cnl */
1570 whitelist_reg(w
, GEN8_CS_CHICKEN1
);
1573 static void icl_whitelist_build(struct intel_engine_cs
*engine
)
1575 struct i915_wa_list
*w
= &engine
->whitelist
;
1577 switch (engine
->class) {
1579 /* WaAllowUMDToModifyHalfSliceChicken7:icl */
1580 whitelist_reg(w
, GEN9_HALF_SLICE_CHICKEN7
);
1582 /* WaAllowUMDToModifySamplerMode:icl */
1583 whitelist_reg(w
, GEN10_SAMPLER_MODE
);
1585 /* WaEnableStateCacheRedirectToCS:icl */
1586 whitelist_reg(w
, GEN9_SLICE_COMMON_ECO_CHICKEN1
);
1589 * WaAllowPMDepthAndInvocationCountAccessFromUMD:icl
1591 * This covers 4 register which are next to one another :
1592 * - PS_INVOCATION_COUNT
1593 * - PS_INVOCATION_COUNT_UDW
1595 * - PS_DEPTH_COUNT_UDW
1597 whitelist_reg_ext(w
, PS_INVOCATION_COUNT
,
1598 RING_FORCE_TO_NONPRIV_ACCESS_RD
|
1599 RING_FORCE_TO_NONPRIV_RANGE_4
);
1602 case VIDEO_DECODE_CLASS
:
1603 /* hucStatusRegOffset */
1604 whitelist_reg_ext(w
, _MMIO(0x2000 + engine
->mmio_base
),
1605 RING_FORCE_TO_NONPRIV_ACCESS_RD
);
1606 /* hucUKernelHdrInfoRegOffset */
1607 whitelist_reg_ext(w
, _MMIO(0x2014 + engine
->mmio_base
),
1608 RING_FORCE_TO_NONPRIV_ACCESS_RD
);
1609 /* hucStatus2RegOffset */
1610 whitelist_reg_ext(w
, _MMIO(0x23B0 + engine
->mmio_base
),
1611 RING_FORCE_TO_NONPRIV_ACCESS_RD
);
1612 whitelist_reg_ext(w
,
1613 RING_CTX_TIMESTAMP(engine
->mmio_base
),
1614 RING_FORCE_TO_NONPRIV_ACCESS_RD
);
1618 whitelist_reg_ext(w
,
1619 RING_CTX_TIMESTAMP(engine
->mmio_base
),
1620 RING_FORCE_TO_NONPRIV_ACCESS_RD
);
1625 static void tgl_whitelist_build(struct intel_engine_cs
*engine
)
1627 struct i915_wa_list
*w
= &engine
->whitelist
;
1629 switch (engine
->class) {
1632 * WaAllowPMDepthAndInvocationCountAccessFromUMD:tgl
1635 * This covers 4 registers which are next to one another :
1636 * - PS_INVOCATION_COUNT
1637 * - PS_INVOCATION_COUNT_UDW
1639 * - PS_DEPTH_COUNT_UDW
1641 whitelist_reg_ext(w
, PS_INVOCATION_COUNT
,
1642 RING_FORCE_TO_NONPRIV_ACCESS_RD
|
1643 RING_FORCE_TO_NONPRIV_RANGE_4
);
1645 /* Wa_1808121037:tgl */
1646 whitelist_reg(w
, GEN7_COMMON_SLICE_CHICKEN1
);
1648 /* Wa_1806527549:tgl */
1649 whitelist_reg(w
, HIZ_CHICKEN
);
1652 whitelist_reg_ext(w
,
1653 RING_CTX_TIMESTAMP(engine
->mmio_base
),
1654 RING_FORCE_TO_NONPRIV_ACCESS_RD
);
1659 static void dg1_whitelist_build(struct intel_engine_cs
*engine
)
1661 struct i915_wa_list
*w
= &engine
->whitelist
;
1663 tgl_whitelist_build(engine
);
1665 /* GEN:BUG:1409280441:dg1 */
1666 if (IS_DG1_REVID(engine
->i915
, DG1_REVID_A0
, DG1_REVID_A0
) &&
1667 (engine
->class == RENDER_CLASS
||
1668 engine
->class == COPY_ENGINE_CLASS
))
1669 whitelist_reg_ext(w
, RING_ID(engine
->mmio_base
),
1670 RING_FORCE_TO_NONPRIV_ACCESS_RD
);
1673 void intel_engine_init_whitelist(struct intel_engine_cs
*engine
)
1675 struct drm_i915_private
*i915
= engine
->i915
;
1676 struct i915_wa_list
*w
= &engine
->whitelist
;
1678 wa_init_start(w
, "whitelist", engine
->name
);
1681 dg1_whitelist_build(engine
);
1682 else if (IS_GEN(i915
, 12))
1683 tgl_whitelist_build(engine
);
1684 else if (IS_GEN(i915
, 11))
1685 icl_whitelist_build(engine
);
1686 else if (IS_CANNONLAKE(i915
))
1687 cnl_whitelist_build(engine
);
1688 else if (IS_COMETLAKE(i915
))
1689 cml_whitelist_build(engine
);
1690 else if (IS_COFFEELAKE(i915
))
1691 cfl_whitelist_build(engine
);
1692 else if (IS_GEMINILAKE(i915
))
1693 glk_whitelist_build(engine
);
1694 else if (IS_KABYLAKE(i915
))
1695 kbl_whitelist_build(engine
);
1696 else if (IS_BROXTON(i915
))
1697 bxt_whitelist_build(engine
);
1698 else if (IS_SKYLAKE(i915
))
1699 skl_whitelist_build(engine
);
1700 else if (INTEL_GEN(i915
) <= 8)
1703 MISSING_CASE(INTEL_GEN(i915
));
1708 void intel_engine_apply_whitelist(struct intel_engine_cs
*engine
)
1710 const struct i915_wa_list
*wal
= &engine
->whitelist
;
1711 struct intel_uncore
*uncore
= engine
->uncore
;
1712 const u32 base
= engine
->mmio_base
;
1719 for (i
= 0, wa
= wal
->list
; i
< wal
->count
; i
++, wa
++)
1720 intel_uncore_write(uncore
,
1721 RING_FORCE_TO_NONPRIV(base
, i
),
1722 i915_mmio_reg_offset(wa
->reg
));
1724 /* And clear the rest just in case of garbage */
1725 for (; i
< RING_MAX_NONPRIV_SLOTS
; i
++)
1726 intel_uncore_write(uncore
,
1727 RING_FORCE_TO_NONPRIV(base
, i
),
1728 i915_mmio_reg_offset(RING_NOPID(base
)));
1732 rcs_engine_wa_init(struct intel_engine_cs
*engine
, struct i915_wa_list
*wal
)
1734 struct drm_i915_private
*i915
= engine
->i915
;
1736 if (IS_DG1_REVID(i915
, DG1_REVID_A0
, DG1_REVID_A0
) ||
1737 IS_TGL_UY_GT_REVID(i915
, TGL_REVID_A0
, TGL_REVID_A0
)) {
1739 * Wa_1607138336:tgl[a0],dg1[a0]
1740 * Wa_1607063988:tgl[a0],dg1[a0]
1743 GEN9_CTX_PREEMPT_REG
,
1744 GEN12_DISABLE_POSH_BUSY_FF_DOP_CG
);
1747 if (IS_TGL_UY_GT_REVID(i915
, TGL_REVID_A0
, TGL_REVID_A0
)) {
1750 * (see also Wa_1606682166:icl)
1754 GEN7_DISABLE_SAMPLER_PREFETCH
);
1756 /* Wa_1408615072:tgl */
1757 wa_write_or(wal
, UNSLICE_UNIT_LEVEL_CLKGATE2
,
1758 VSUNIT_CLKGATE_DIS_TGL
);
1761 if (IS_DG1(i915
) || IS_ROCKETLAKE(i915
) || IS_TIGERLAKE(i915
)) {
1762 /* Wa_1606931601:tgl,rkl,dg1 */
1763 wa_masked_en(wal
, GEN7_ROW_CHICKEN2
, GEN12_DISABLE_EARLY_READ
);
1766 * Wa_1407928979:tgl A*
1767 * Wa_18011464164:tgl[B0+],dg1[B0+]
1768 * Wa_22010931296:tgl[B0+],dg1[B0+]
1769 * Wa_14010919138:rkl, dg1
1771 wa_write_or(wal
, GEN7_FF_THREAD_MODE
,
1772 GEN12_FF_TESSELATION_DOP_GATE_DISABLE
);
1775 if (IS_DG1_REVID(i915
, DG1_REVID_A0
, DG1_REVID_A0
) ||
1776 IS_ROCKETLAKE(i915
) || IS_TIGERLAKE(i915
)) {
1777 /* Wa_1409804808:tgl,rkl,dg1[a0] */
1778 wa_masked_en(wal
, GEN7_ROW_CHICKEN2
,
1779 GEN12_PUSH_CONST_DEREF_HOLD_DIS
);
1783 * Wa_14010229206:tgl,rkl,dg1[a0]
1785 wa_masked_en(wal
, GEN9_ROW_CHICKEN4
, GEN12_DISABLE_TDL_PUSH
);
1790 * Wa_1607297627:tgl,rkl,dg1[a0]
1792 * On TGL and RKL there are multiple entries for this WA in the
1793 * BSpec; some indicate this is an A0-only WA, others indicate
1794 * it applies to all steppings so we trust the "all steppings."
1795 * For DG1 this only applies to A0.
1798 GEN6_RC_SLEEP_PSMI_CONTROL
,
1799 GEN12_WAIT_FOR_EVENT_POWER_DOWN_DISABLE
|
1800 GEN8_RC_SEMA_IDLE_MSG_DISABLE
);
1804 * Wa_22010271021:tgl,rkl
1807 GEN9_CS_DEBUG_MODE1
,
1808 FF_DOP_CLOCK_GATE_DISABLE
);
1811 if (IS_GEN(i915
, 12)) {
1812 /* Wa_1406941453:gen12 */
1818 if (IS_GEN(i915
, 11)) {
1819 /* This is not an Wa. Enable for better image quality */
1822 _3D_CHICKEN3_AA_LINE_QUALITY_FIX_ENABLE
);
1824 /* WaPipelineFlushCoherentLines:icl */
1827 GEN8_LQSC_FLUSH_COHERENT_LINES
);
1831 * Formerly known as WaGAPZPriorityScheme
1835 GEN11_ARBITRATION_PRIO_ORDER_MASK
);
1839 * Formerly known as WaL3BankAddressHashing
1841 wa_write_masked_or(wal
,
1843 GEN11_HASH_CTRL_EXCL_MASK
,
1844 GEN11_HASH_CTRL_EXCL_BIT0
);
1845 wa_write_masked_or(wal
,
1847 GEN11_BANK_HASH_ADDR_EXCL_MASK
,
1848 GEN11_BANK_HASH_ADDR_EXCL_BIT0
);
1852 * Formerly known as WaDisableCleanEvicts
1856 GEN11_LQSC_CLEAN_EVICT_DISABLE
);
1858 /* WaForwardProgressSoftReset:icl */
1860 GEN10_SCRATCH_LNCF2
,
1861 PMFLUSHDONE_LNICRSDROP
|
1862 PMFLUSH_GAPL3UNBLOCK
|
1863 PMFLUSHDONE_LNEBLK
);
1865 /* Wa_1406609255:icl (pre-prod) */
1866 if (IS_ICL_REVID(i915
, ICL_REVID_A0
, ICL_REVID_B0
))
1869 GEN7_DISABLE_DEMAND_PREFETCH
);
1871 /* Wa_1606682166:icl */
1874 GEN7_DISABLE_SAMPLER_PREFETCH
);
1876 /* Wa_1409178092:icl */
1877 wa_write_masked_or(wal
,
1879 GEN11_COHERENT_PARTIAL_WRITE_MERGE_ENABLE
,
1882 /* WaEnable32PlaneMode:icl */
1883 wa_masked_en(wal
, GEN9_CSFE_CHICKEN1_RCS
,
1884 GEN11_ENABLE_32_PLANE_MODE
);
1887 * Wa_1408615072:icl,ehl (vsunit)
1888 * Wa_1407596294:icl,ehl (hsunit)
1890 wa_write_or(wal
, UNSLICE_UNIT_LEVEL_CLKGATE
,
1891 VSUNIT_CLKGATE_DIS
| HSUNIT_CLKGATE_DIS
);
1893 /* Wa_1407352427:icl,ehl */
1894 wa_write_or(wal
, UNSLICE_UNIT_LEVEL_CLKGATE2
,
1895 PSDUNIT_CLKGATE_DIS
);
1897 /* Wa_1406680159:icl,ehl */
1899 SUBSLICE_UNIT_LEVEL_CLKGATE
,
1900 GWUNIT_CLKGATE_DIS
);
1903 * Wa_1408767742:icl[a2..forever],ehl[all]
1904 * Wa_1605460711:icl[a0..c0]
1907 GEN7_FF_THREAD_MODE
,
1908 GEN12_FF_TESSELATION_DOP_GATE_DISABLE
);
1910 /* Wa_22010271021:ehl */
1911 if (IS_JSL_EHL(i915
))
1913 GEN9_CS_DEBUG_MODE1
,
1914 FF_DOP_CLOCK_GATE_DISABLE
);
1917 if (IS_GEN_RANGE(i915
, 9, 12)) {
1918 /* FtrPerCtxtPreemptionGranularityControl:skl,bxt,kbl,cfl,cnl,icl,tgl */
1920 GEN7_FF_SLICE_CS_CHICKEN1
,
1921 GEN9_FFSC_PERCTX_PREEMPT_CTRL
);
1924 if (IS_SKYLAKE(i915
) ||
1925 IS_KABYLAKE(i915
) ||
1926 IS_COFFEELAKE(i915
) ||
1927 IS_COMETLAKE(i915
)) {
1928 /* WaEnableGapsTsvCreditFix:skl,kbl,cfl */
1931 GEN9_GAPS_TSV_CREDIT_DISABLE
);
1934 if (IS_BROXTON(i915
)) {
1935 /* WaDisablePooledEuLoadBalancingFix:bxt */
1937 FF_SLICE_CS_CHICKEN2
,
1938 GEN9_POOLED_EU_LOAD_BALANCING_FIX_DISABLE
);
1941 if (IS_GEN(i915
, 9)) {
1942 /* WaContextSwitchWithConcurrentTLBInvalidate:skl,bxt,kbl,glk,cfl */
1944 GEN9_CSFE_CHICKEN1_RCS
,
1945 GEN9_PREEMPT_GPGPU_SYNC_SWITCH_DISABLE
);
1947 /* WaEnableLbsSlaRetryTimerDecrement:skl,bxt,kbl,glk,cfl */
1950 GEN9_LBS_SLA_RETRY_TIMER_DECREMENT_ENABLE
);
1952 /* WaProgramL3SqcReg1DefaultForPerf:bxt,glk */
1953 if (IS_GEN9_LP(i915
))
1954 wa_write_masked_or(wal
,
1956 L3_PRIO_CREDITS_MASK
,
1957 L3_GENERAL_PRIO_CREDITS(62) |
1958 L3_HIGH_PRIO_CREDITS(2));
1960 /* WaOCLCoherentLineFlush:skl,bxt,kbl,cfl */
1963 GEN8_LQSC_FLUSH_COHERENT_LINES
);
1966 if (IS_GEN(i915
, 7))
1967 /* WaBCSVCSTlbInvalidationMode:ivb,vlv,hsw */
1970 GFX_TLB_INVALIDATE_EXPLICIT
| GFX_REPLAY_MODE
);
1972 if (IS_GEN_RANGE(i915
, 6, 7))
1974 * We need to disable the AsyncFlip performance optimisations in
1975 * order to use MI_WAIT_FOR_EVENT within the CS. It should
1976 * already be programmed to '1' on all products.
1978 * WaDisableAsyncFlipPerfMode:snb,ivb,hsw,vlv
1982 ASYNC_FLIP_PERF_DISABLE
);
1984 if (IS_GEN(i915
, 6)) {
1986 * Required for the hardware to program scanline values for
1988 * WaEnableFlushTlbInvalidationMode:snb
1992 GFX_TLB_INVALIDATE_EXPLICIT
);
1995 * From the Sandybridge PRM, volume 1 part 3, page 24:
1996 * "If this bit is set, STCunit will have LRA as replacement
1997 * policy. [...] This bit must be reset. LRA replacement
1998 * policy is not supported."
2002 CM0_STC_EVICT_DISABLE_LRA_SNB
);
2005 if (IS_GEN_RANGE(i915
, 4, 6))
2006 /* WaTimedSingleVertexDispatch:cl,bw,ctg,elk,ilk,snb */
2007 wa_add(wal
, MI_MODE
,
2008 0, _MASKED_BIT_ENABLE(VS_TIMER_DISPATCH
),
2009 /* XXX bit doesn't stick on Broadwater */
2010 IS_I965G(i915
) ? 0 : VS_TIMER_DISPATCH
);
2012 if (IS_GEN(i915
, 4))
2014 * Disable CONSTANT_BUFFER before it is loaded from the context
2015 * image. For as it is loaded, it is executed and the stored
2016 * address may no longer be valid, leading to a GPU hang.
2018 * This imposes the requirement that userspace reload their
2019 * CONSTANT_BUFFER on every batch, fortunately a requirement
2020 * they are already accustomed to from before contexts were
2023 wa_add(wal
, ECOSKPD
,
2024 0, _MASKED_BIT_ENABLE(ECO_CONSTANT_BUFFER_SR_DISABLE
),
2025 0 /* XXX bit doesn't stick on Broadwater */);
2029 xcs_engine_wa_init(struct intel_engine_cs
*engine
, struct i915_wa_list
*wal
)
2031 struct drm_i915_private
*i915
= engine
->i915
;
2033 /* WaKBLVECSSemaphoreWaitPoll:kbl */
2034 if (IS_KBL_GT_REVID(i915
, KBL_REVID_A0
, KBL_REVID_E0
)) {
2036 RING_SEMA_WAIT_POLL(engine
->mmio_base
),
2042 engine_init_workarounds(struct intel_engine_cs
*engine
, struct i915_wa_list
*wal
)
2044 if (I915_SELFTEST_ONLY(INTEL_GEN(engine
->i915
) < 4))
2047 if (engine
->class == RENDER_CLASS
)
2048 rcs_engine_wa_init(engine
, wal
);
2050 xcs_engine_wa_init(engine
, wal
);
2053 void intel_engine_init_workarounds(struct intel_engine_cs
*engine
)
2055 struct i915_wa_list
*wal
= &engine
->wa_list
;
2057 if (INTEL_GEN(engine
->i915
) < 4)
2060 wa_init_start(wal
, "engine", engine
->name
);
2061 engine_init_workarounds(engine
, wal
);
2062 wa_init_finish(wal
);
2065 void intel_engine_apply_workarounds(struct intel_engine_cs
*engine
)
2067 wa_list_apply(engine
->uncore
, &engine
->wa_list
);
2070 static struct i915_vma
*
2071 create_scratch(struct i915_address_space
*vm
, int count
)
2073 struct drm_i915_gem_object
*obj
;
2074 struct i915_vma
*vma
;
2078 size
= round_up(count
* sizeof(u32
), PAGE_SIZE
);
2079 obj
= i915_gem_object_create_internal(vm
->i915
, size
);
2081 return ERR_CAST(obj
);
2083 i915_gem_object_set_cache_coherency(obj
, I915_CACHE_LLC
);
2085 vma
= i915_vma_instance(obj
, vm
, NULL
);
2091 err
= i915_vma_pin(vma
, 0, 0,
2092 i915_vma_is_ggtt(vma
) ? PIN_GLOBAL
: PIN_USER
);
2099 i915_gem_object_put(obj
);
2100 return ERR_PTR(err
);
2108 static const struct mcr_range mcr_ranges_gen8
[] = {
2109 { .start
= 0x5500, .end
= 0x55ff },
2110 { .start
= 0x7000, .end
= 0x7fff },
2111 { .start
= 0x9400, .end
= 0x97ff },
2112 { .start
= 0xb000, .end
= 0xb3ff },
2113 { .start
= 0xe000, .end
= 0xe7ff },
2117 static const struct mcr_range mcr_ranges_gen12
[] = {
2118 { .start
= 0x8150, .end
= 0x815f },
2119 { .start
= 0x9520, .end
= 0x955f },
2120 { .start
= 0xb100, .end
= 0xb3ff },
2121 { .start
= 0xde80, .end
= 0xe8ff },
2122 { .start
= 0x24a00, .end
= 0x24a7f },
2126 static bool mcr_range(struct drm_i915_private
*i915
, u32 offset
)
2128 const struct mcr_range
*mcr_ranges
;
2131 if (INTEL_GEN(i915
) >= 12)
2132 mcr_ranges
= mcr_ranges_gen12
;
2133 else if (INTEL_GEN(i915
) >= 8)
2134 mcr_ranges
= mcr_ranges_gen8
;
2139 * Registers in these ranges are affected by the MCR selector
2140 * which only controls CPU initiated MMIO. Routing does not
2141 * work for CS access so we cannot verify them on this path.
2143 for (i
= 0; mcr_ranges
[i
].start
; i
++)
2144 if (offset
>= mcr_ranges
[i
].start
&&
2145 offset
<= mcr_ranges
[i
].end
)
2152 wa_list_srm(struct i915_request
*rq
,
2153 const struct i915_wa_list
*wal
,
2154 struct i915_vma
*vma
)
2156 struct drm_i915_private
*i915
= rq
->engine
->i915
;
2157 unsigned int i
, count
= 0;
2158 const struct i915_wa
*wa
;
2161 srm
= MI_STORE_REGISTER_MEM
| MI_SRM_LRM_GLOBAL_GTT
;
2162 if (INTEL_GEN(i915
) >= 8)
2165 for (i
= 0, wa
= wal
->list
; i
< wal
->count
; i
++, wa
++) {
2166 if (!mcr_range(i915
, i915_mmio_reg_offset(wa
->reg
)))
2170 cs
= intel_ring_begin(rq
, 4 * count
);
2174 for (i
= 0, wa
= wal
->list
; i
< wal
->count
; i
++, wa
++) {
2175 u32 offset
= i915_mmio_reg_offset(wa
->reg
);
2177 if (mcr_range(i915
, offset
))
2182 *cs
++ = i915_ggtt_offset(vma
) + sizeof(u32
) * i
;
2185 intel_ring_advance(rq
, cs
);
2190 static int engine_wa_list_verify(struct intel_context
*ce
,
2191 const struct i915_wa_list
* const wal
,
2194 const struct i915_wa
*wa
;
2195 struct i915_request
*rq
;
2196 struct i915_vma
*vma
;
2197 struct i915_gem_ww_ctx ww
;
2205 vma
= create_scratch(&ce
->engine
->gt
->ggtt
->vm
, wal
->count
);
2207 return PTR_ERR(vma
);
2209 intel_engine_pm_get(ce
->engine
);
2210 i915_gem_ww_ctx_init(&ww
, false);
2212 err
= i915_gem_object_lock(vma
->obj
, &ww
);
2214 err
= intel_context_pin_ww(ce
, &ww
);
2218 rq
= i915_request_create(ce
);
2224 err
= i915_request_await_object(rq
, vma
->obj
, true);
2226 err
= i915_vma_move_to_active(vma
, rq
, EXEC_OBJECT_WRITE
);
2228 err
= wa_list_srm(rq
, wal
, vma
);
2230 i915_request_get(rq
);
2232 i915_request_set_error_once(rq
, err
);
2233 i915_request_add(rq
);
2238 if (i915_request_wait(rq
, 0, HZ
/ 5) < 0) {
2243 results
= i915_gem_object_pin_map(vma
->obj
, I915_MAP_WB
);
2244 if (IS_ERR(results
)) {
2245 err
= PTR_ERR(results
);
2250 for (i
= 0, wa
= wal
->list
; i
< wal
->count
; i
++, wa
++) {
2251 if (mcr_range(rq
->engine
->i915
, i915_mmio_reg_offset(wa
->reg
)))
2254 if (!wa_verify(wa
, results
[i
], wal
->name
, from
))
2258 i915_gem_object_unpin_map(vma
->obj
);
2261 i915_request_put(rq
);
2263 intel_context_unpin(ce
);
2265 if (err
== -EDEADLK
) {
2266 err
= i915_gem_ww_ctx_backoff(&ww
);
2270 i915_gem_ww_ctx_fini(&ww
);
2271 intel_engine_pm_put(ce
->engine
);
2272 i915_vma_unpin(vma
);
2277 int intel_engine_verify_workarounds(struct intel_engine_cs
*engine
,
2280 return engine_wa_list_verify(engine
->kernel_context
,
2285 #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
2286 #include "selftest_workarounds.c"