2 * Copyright © 2014-2017 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
25 #include "intel_guc.h"
26 #include "intel_guc_ads.h"
27 #include "intel_guc_submission.h"
30 static void guc_init_ggtt_pin_bias(struct intel_guc
*guc
);
32 static void gen8_guc_raise_irq(struct intel_guc
*guc
)
34 struct drm_i915_private
*dev_priv
= guc_to_i915(guc
);
36 I915_WRITE(GUC_SEND_INTERRUPT
, GUC_SEND_TRIGGER
);
39 static inline i915_reg_t
guc_send_reg(struct intel_guc
*guc
, u32 i
)
41 GEM_BUG_ON(!guc
->send_regs
.base
);
42 GEM_BUG_ON(!guc
->send_regs
.count
);
43 GEM_BUG_ON(i
>= guc
->send_regs
.count
);
45 return _MMIO(guc
->send_regs
.base
+ 4 * i
);
48 void intel_guc_init_send_regs(struct intel_guc
*guc
)
50 struct drm_i915_private
*dev_priv
= guc_to_i915(guc
);
51 enum forcewake_domains fw_domains
= 0;
54 guc
->send_regs
.base
= i915_mmio_reg_offset(SOFT_SCRATCH(0));
55 guc
->send_regs
.count
= SOFT_SCRATCH_COUNT
- 1;
57 for (i
= 0; i
< guc
->send_regs
.count
; i
++) {
58 fw_domains
|= intel_uncore_forcewake_for_reg(dev_priv
,
60 FW_REG_READ
| FW_REG_WRITE
);
62 guc
->send_regs
.fw_domains
= fw_domains
;
65 void intel_guc_init_early(struct intel_guc
*guc
)
67 intel_guc_fw_init_early(guc
);
68 intel_guc_ct_init_early(&guc
->ct
);
69 intel_guc_log_init_early(&guc
->log
);
71 mutex_init(&guc
->send_mutex
);
72 spin_lock_init(&guc
->irq_lock
);
73 guc
->send
= intel_guc_send_nop
;
74 guc
->handler
= intel_guc_to_host_event_handler_nop
;
75 guc
->notify
= gen8_guc_raise_irq
;
78 static int guc_init_wq(struct intel_guc
*guc
)
80 struct drm_i915_private
*dev_priv
= guc_to_i915(guc
);
83 * GuC log buffer flush work item has to do register access to
84 * send the ack to GuC and this work item, if not synced before
85 * suspend, can potentially get executed after the GFX device is
87 * By marking the WQ as freezable, we don't have to bother about
88 * flushing of this work item from the suspend hooks, the pending
89 * work item if any will be either executed before the suspend
90 * or scheduled later on resume. This way the handling of work
91 * item can be kept same between system suspend & rpm suspend.
93 guc
->log
.relay
.flush_wq
=
94 alloc_ordered_workqueue("i915-guc_log",
95 WQ_HIGHPRI
| WQ_FREEZABLE
);
96 if (!guc
->log
.relay
.flush_wq
) {
97 DRM_ERROR("Couldn't allocate workqueue for GuC log\n");
102 * Even though both sending GuC action, and adding a new workitem to
103 * GuC workqueue are serialized (each with its own locking), since
104 * we're using mutliple engines, it's possible that we're going to
105 * issue a preempt request with two (or more - each for different
106 * engine) workitems in GuC queue. In this situation, GuC may submit
107 * all of them, which will make us very confused.
108 * Our preemption contexts may even already be complete - before we
109 * even had the chance to sent the preempt action to GuC!. Rather
110 * than introducing yet another lock, we can just use ordered workqueue
111 * to make sure we're always sending a single preemption request with a
114 if (HAS_LOGICAL_RING_PREEMPTION(dev_priv
) &&
115 USES_GUC_SUBMISSION(dev_priv
)) {
116 guc
->preempt_wq
= alloc_ordered_workqueue("i915-guc_preempt",
118 if (!guc
->preempt_wq
) {
119 destroy_workqueue(guc
->log
.relay
.flush_wq
);
120 DRM_ERROR("Couldn't allocate workqueue for GuC "
129 static void guc_fini_wq(struct intel_guc
*guc
)
131 struct drm_i915_private
*dev_priv
= guc_to_i915(guc
);
133 if (HAS_LOGICAL_RING_PREEMPTION(dev_priv
) &&
134 USES_GUC_SUBMISSION(dev_priv
))
135 destroy_workqueue(guc
->preempt_wq
);
137 destroy_workqueue(guc
->log
.relay
.flush_wq
);
140 int intel_guc_init_misc(struct intel_guc
*guc
)
142 struct drm_i915_private
*i915
= guc_to_i915(guc
);
145 guc_init_ggtt_pin_bias(guc
);
147 ret
= guc_init_wq(guc
);
151 intel_uc_fw_fetch(i915
, &guc
->fw
);
156 void intel_guc_fini_misc(struct intel_guc
*guc
)
158 intel_uc_fw_fini(&guc
->fw
);
162 static int guc_shared_data_create(struct intel_guc
*guc
)
164 struct i915_vma
*vma
;
167 vma
= intel_guc_allocate_vma(guc
, PAGE_SIZE
);
171 vaddr
= i915_gem_object_pin_map(vma
->obj
, I915_MAP_WB
);
173 i915_vma_unpin_and_release(&vma
);
174 return PTR_ERR(vaddr
);
177 guc
->shared_data
= vma
;
178 guc
->shared_data_vaddr
= vaddr
;
183 static void guc_shared_data_destroy(struct intel_guc
*guc
)
185 i915_gem_object_unpin_map(guc
->shared_data
->obj
);
186 i915_vma_unpin_and_release(&guc
->shared_data
);
189 int intel_guc_init(struct intel_guc
*guc
)
191 struct drm_i915_private
*dev_priv
= guc_to_i915(guc
);
194 ret
= guc_shared_data_create(guc
);
197 GEM_BUG_ON(!guc
->shared_data
);
199 ret
= intel_guc_log_create(&guc
->log
);
203 ret
= intel_guc_ads_create(guc
);
206 GEM_BUG_ON(!guc
->ads_vma
);
208 /* We need to notify the guc whenever we change the GGTT */
209 i915_ggtt_enable_guc(dev_priv
);
214 intel_guc_log_destroy(&guc
->log
);
216 guc_shared_data_destroy(guc
);
218 intel_uc_fw_fini(&guc
->fw
);
222 void intel_guc_fini(struct intel_guc
*guc
)
224 struct drm_i915_private
*dev_priv
= guc_to_i915(guc
);
226 i915_ggtt_disable_guc(dev_priv
);
227 intel_guc_ads_destroy(guc
);
228 intel_guc_log_destroy(&guc
->log
);
229 guc_shared_data_destroy(guc
);
230 intel_uc_fw_fini(&guc
->fw
);
233 static u32
guc_ctl_debug_flags(struct intel_guc
*guc
)
235 u32 level
= intel_guc_log_get_level(&guc
->log
);
239 ads
= intel_guc_ggtt_offset(guc
, guc
->ads_vma
) >> PAGE_SHIFT
;
240 flags
= ads
<< GUC_ADS_ADDR_SHIFT
| GUC_ADS_ENABLED
;
242 if (!GUC_LOG_LEVEL_IS_ENABLED(level
))
243 flags
|= GUC_LOG_DEFAULT_DISABLED
;
245 if (!GUC_LOG_LEVEL_IS_VERBOSE(level
))
246 flags
|= GUC_LOG_DISABLED
;
248 flags
|= GUC_LOG_LEVEL_TO_VERBOSITY(level
) <<
249 GUC_LOG_VERBOSITY_SHIFT
;
254 static u32
guc_ctl_feature_flags(struct intel_guc
*guc
)
258 flags
|= GUC_CTL_VCS2_ENABLED
;
260 if (USES_GUC_SUBMISSION(guc_to_i915(guc
)))
261 flags
|= GUC_CTL_KERNEL_SUBMISSIONS
;
263 flags
|= GUC_CTL_DISABLE_SCHEDULER
;
268 static u32
guc_ctl_ctxinfo_flags(struct intel_guc
*guc
)
272 if (USES_GUC_SUBMISSION(guc_to_i915(guc
))) {
275 base
= intel_guc_ggtt_offset(guc
, guc
->stage_desc_pool
);
276 ctxnum
= GUC_MAX_STAGE_DESCRIPTORS
/ 16;
279 flags
|= (base
<< GUC_CTL_BASE_ADDR_SHIFT
) |
280 (ctxnum
<< GUC_CTL_CTXNUM_IN16_SHIFT
);
285 static u32
guc_ctl_log_params_flags(struct intel_guc
*guc
)
287 u32 offset
= intel_guc_ggtt_offset(guc
, guc
->log
.vma
) >> PAGE_SHIFT
;
290 #if (((CRASH_BUFFER_SIZE) % SZ_1M) == 0)
292 #define FLAG GUC_LOG_ALLOC_IN_MEGABYTE
298 BUILD_BUG_ON(!CRASH_BUFFER_SIZE
);
299 BUILD_BUG_ON(!IS_ALIGNED(CRASH_BUFFER_SIZE
, UNIT
));
300 BUILD_BUG_ON(!DPC_BUFFER_SIZE
);
301 BUILD_BUG_ON(!IS_ALIGNED(DPC_BUFFER_SIZE
, UNIT
));
302 BUILD_BUG_ON(!ISR_BUFFER_SIZE
);
303 BUILD_BUG_ON(!IS_ALIGNED(ISR_BUFFER_SIZE
, UNIT
));
305 BUILD_BUG_ON((CRASH_BUFFER_SIZE
/ UNIT
- 1) >
306 (GUC_LOG_CRASH_MASK
>> GUC_LOG_CRASH_SHIFT
));
307 BUILD_BUG_ON((DPC_BUFFER_SIZE
/ UNIT
- 1) >
308 (GUC_LOG_DPC_MASK
>> GUC_LOG_DPC_SHIFT
));
309 BUILD_BUG_ON((ISR_BUFFER_SIZE
/ UNIT
- 1) >
310 (GUC_LOG_ISR_MASK
>> GUC_LOG_ISR_SHIFT
));
312 flags
= GUC_LOG_VALID
|
313 GUC_LOG_NOTIFY_ON_HALF_FULL
|
315 ((CRASH_BUFFER_SIZE
/ UNIT
- 1) << GUC_LOG_CRASH_SHIFT
) |
316 ((DPC_BUFFER_SIZE
/ UNIT
- 1) << GUC_LOG_DPC_SHIFT
) |
317 ((ISR_BUFFER_SIZE
/ UNIT
- 1) << GUC_LOG_ISR_SHIFT
) |
318 (offset
<< GUC_LOG_BUF_ADDR_SHIFT
);
327 * Initialise the GuC parameter block before starting the firmware
328 * transfer. These parameters are read by the firmware on startup
329 * and cannot be changed thereafter.
331 void intel_guc_init_params(struct intel_guc
*guc
)
333 struct drm_i915_private
*dev_priv
= guc_to_i915(guc
);
334 u32 params
[GUC_CTL_MAX_DWORDS
];
337 memset(params
, 0, sizeof(params
));
340 * GuC ARAT increment is 10 ns. GuC default scheduler quantum is one
341 * second. This ARAR is calculated by:
342 * Scheduler-Quantum-in-ns / ARAT-increment-in-ns = 1000000000 / 10
344 params
[GUC_CTL_ARAT_HIGH
] = 0;
345 params
[GUC_CTL_ARAT_LOW
] = 100000000;
347 params
[GUC_CTL_WA
] |= GUC_CTL_WA_UK_BY_DRIVER
;
349 params
[GUC_CTL_FEATURE
] = guc_ctl_feature_flags(guc
);
350 params
[GUC_CTL_LOG_PARAMS
] = guc_ctl_log_params_flags(guc
);
351 params
[GUC_CTL_DEBUG
] = guc_ctl_debug_flags(guc
);
352 params
[GUC_CTL_CTXINFO
] = guc_ctl_ctxinfo_flags(guc
);
354 for (i
= 0; i
< GUC_CTL_MAX_DWORDS
; i
++)
355 DRM_DEBUG_DRIVER("param[%2d] = %#x\n", i
, params
[i
]);
358 * All SOFT_SCRATCH registers are in FORCEWAKE_BLITTER domain and
359 * they are power context saved so it's ok to release forcewake
360 * when we are done here and take it again at xfer time.
362 intel_uncore_forcewake_get(dev_priv
, FORCEWAKE_BLITTER
);
364 I915_WRITE(SOFT_SCRATCH(0), 0);
366 for (i
= 0; i
< GUC_CTL_MAX_DWORDS
; i
++)
367 I915_WRITE(SOFT_SCRATCH(1 + i
), params
[i
]);
369 intel_uncore_forcewake_put(dev_priv
, FORCEWAKE_BLITTER
);
372 int intel_guc_send_nop(struct intel_guc
*guc
, const u32
*action
, u32 len
,
373 u32
*response_buf
, u32 response_buf_size
)
375 WARN(1, "Unexpected send: action=%#x\n", *action
);
379 void intel_guc_to_host_event_handler_nop(struct intel_guc
*guc
)
381 WARN(1, "Unexpected event: no suitable handler\n");
385 * This function implements the MMIO based host to GuC interface.
387 int intel_guc_send_mmio(struct intel_guc
*guc
, const u32
*action
, u32 len
,
388 u32
*response_buf
, u32 response_buf_size
)
390 struct drm_i915_private
*dev_priv
= guc_to_i915(guc
);
396 GEM_BUG_ON(len
> guc
->send_regs
.count
);
398 /* We expect only action code */
399 GEM_BUG_ON(*action
& ~INTEL_GUC_MSG_CODE_MASK
);
401 /* If CT is available, we expect to use MMIO only during init/fini */
402 GEM_BUG_ON(HAS_GUC_CT(dev_priv
) &&
403 *action
!= INTEL_GUC_ACTION_REGISTER_COMMAND_TRANSPORT_BUFFER
&&
404 *action
!= INTEL_GUC_ACTION_DEREGISTER_COMMAND_TRANSPORT_BUFFER
);
406 mutex_lock(&guc
->send_mutex
);
407 intel_uncore_forcewake_get(dev_priv
, guc
->send_regs
.fw_domains
);
409 for (i
= 0; i
< len
; i
++)
410 I915_WRITE(guc_send_reg(guc
, i
), action
[i
]);
412 POSTING_READ(guc_send_reg(guc
, i
- 1));
414 intel_guc_notify(guc
);
417 * No GuC command should ever take longer than 10ms.
418 * Fast commands should still complete in 10us.
420 ret
= __intel_wait_for_register_fw(dev_priv
,
421 guc_send_reg(guc
, 0),
422 INTEL_GUC_MSG_TYPE_MASK
,
423 INTEL_GUC_MSG_TYPE_RESPONSE
<<
424 INTEL_GUC_MSG_TYPE_SHIFT
,
426 /* If GuC explicitly returned an error, convert it to -EIO */
427 if (!ret
&& !INTEL_GUC_MSG_IS_RESPONSE_SUCCESS(status
))
431 DRM_ERROR("MMIO: GuC action %#x failed with error %d %#x\n",
432 action
[0], ret
, status
);
437 int count
= min(response_buf_size
, guc
->send_regs
.count
- 1);
439 for (i
= 0; i
< count
; i
++)
440 response_buf
[i
] = I915_READ(guc_send_reg(guc
, i
+ 1));
443 /* Use data from the GuC response as our return value */
444 ret
= INTEL_GUC_MSG_TO_DATA(status
);
447 intel_uncore_forcewake_put(dev_priv
, guc
->send_regs
.fw_domains
);
448 mutex_unlock(&guc
->send_mutex
);
453 void intel_guc_to_host_event_handler_mmio(struct intel_guc
*guc
)
455 struct drm_i915_private
*dev_priv
= guc_to_i915(guc
);
459 * Sample the log buffer flush related bits & clear them out now
460 * itself from the message identity register to minimize the
461 * probability of losing a flush interrupt, when there are back
462 * to back flush interrupts.
463 * There can be a new flush interrupt, for different log buffer
464 * type (like for ISR), whilst Host is handling one (for DPC).
465 * Since same bit is used in message register for ISR & DPC, it
466 * could happen that GuC sets the bit for 2nd interrupt but Host
467 * clears out the bit on handling the 1st interrupt.
469 disable_rpm_wakeref_asserts(dev_priv
);
470 spin_lock(&guc
->irq_lock
);
471 val
= I915_READ(SOFT_SCRATCH(15));
472 msg
= val
& guc
->msg_enabled_mask
;
473 I915_WRITE(SOFT_SCRATCH(15), val
& ~msg
);
474 spin_unlock(&guc
->irq_lock
);
475 enable_rpm_wakeref_asserts(dev_priv
);
477 intel_guc_to_host_process_recv_msg(guc
, msg
);
480 void intel_guc_to_host_process_recv_msg(struct intel_guc
*guc
, u32 msg
)
482 /* Make sure to handle only enabled messages */
483 msg
&= guc
->msg_enabled_mask
;
485 if (msg
& (INTEL_GUC_RECV_MSG_FLUSH_LOG_BUFFER
|
486 INTEL_GUC_RECV_MSG_CRASH_DUMP_POSTED
))
487 intel_guc_log_handle_flush_event(&guc
->log
);
490 int intel_guc_sample_forcewake(struct intel_guc
*guc
)
492 struct drm_i915_private
*dev_priv
= guc_to_i915(guc
);
495 action
[0] = INTEL_GUC_ACTION_SAMPLE_FORCEWAKE
;
496 /* WaRsDisableCoarsePowerGating:skl,cnl */
497 if (!HAS_RC6(dev_priv
) || NEEDS_WaRsDisableCoarsePowerGating(dev_priv
))
500 /* bit 0 and 1 are for Render and Media domain separately */
501 action
[1] = GUC_FORCEWAKE_RENDER
| GUC_FORCEWAKE_MEDIA
;
503 return intel_guc_send(guc
, action
, ARRAY_SIZE(action
));
507 * intel_guc_auth_huc() - Send action to GuC to authenticate HuC ucode
508 * @guc: intel_guc structure
509 * @rsa_offset: rsa offset w.r.t ggtt base of huc vma
511 * Triggers a HuC firmware authentication request to the GuC via intel_guc_send
512 * INTEL_GUC_ACTION_AUTHENTICATE_HUC interface. This function is invoked by
515 * Return: non-zero code on error
517 int intel_guc_auth_huc(struct intel_guc
*guc
, u32 rsa_offset
)
520 INTEL_GUC_ACTION_AUTHENTICATE_HUC
,
524 return intel_guc_send(guc
, action
, ARRAY_SIZE(action
));
528 * intel_guc_suspend() - notify GuC entering suspend state
531 int intel_guc_suspend(struct intel_guc
*guc
)
534 INTEL_GUC_ACTION_ENTER_S_STATE
,
535 GUC_POWER_D1
, /* any value greater than GUC_POWER_D0 */
536 intel_guc_ggtt_offset(guc
, guc
->shared_data
)
539 return intel_guc_send(guc
, data
, ARRAY_SIZE(data
));
543 * intel_guc_reset_engine() - ask GuC to reset an engine
544 * @guc: intel_guc structure
545 * @engine: engine to be reset
547 int intel_guc_reset_engine(struct intel_guc
*guc
,
548 struct intel_engine_cs
*engine
)
552 GEM_BUG_ON(!guc
->execbuf_client
);
554 data
[0] = INTEL_GUC_ACTION_REQUEST_ENGINE_RESET
;
555 data
[1] = engine
->guc_id
;
559 data
[5] = guc
->execbuf_client
->stage_id
;
560 data
[6] = intel_guc_ggtt_offset(guc
, guc
->shared_data
);
562 return intel_guc_send(guc
, data
, ARRAY_SIZE(data
));
566 * intel_guc_resume() - notify GuC resuming from suspend state
569 int intel_guc_resume(struct intel_guc
*guc
)
572 INTEL_GUC_ACTION_EXIT_S_STATE
,
574 intel_guc_ggtt_offset(guc
, guc
->shared_data
)
577 return intel_guc_send(guc
, data
, ARRAY_SIZE(data
));
581 * DOC: GuC Address Space
583 * The layout of GuC address space is shown below:
587 * +==============> +====================+ <== GUC_GGTT_TOP
594 * Address +========> +====================+ <== WOPCM Top
595 * Space ^ | HW contexts RSVD |
597 * | | +==> +--------------------+ <== GuC WOPCM Top
601 * | Bias WOPCM | WOPCM |
605 * +=====+=====+==> +====================+ <== GuC WOPCM Base
608 * +====================+ <== WOPCM Base
610 * The lower part of GuC Address Space [0, ggtt_pin_bias) is mapped to WOPCM
611 * while upper part of GuC Address Space [ggtt_pin_bias, GUC_GGTT_TOP) is mapped
612 * to DRAM. The value of the GuC ggtt_pin_bias is determined by WOPCM size and
613 * actual GuC WOPCM size.
617 * guc_init_ggtt_pin_bias() - Initialize the GuC ggtt_pin_bias value.
618 * @guc: intel_guc structure.
620 * This function will calculate and initialize the ggtt_pin_bias value based on
621 * overall WOPCM size and GuC WOPCM size.
623 static void guc_init_ggtt_pin_bias(struct intel_guc
*guc
)
625 struct drm_i915_private
*i915
= guc_to_i915(guc
);
627 GEM_BUG_ON(!i915
->wopcm
.size
);
628 GEM_BUG_ON(i915
->wopcm
.size
< i915
->wopcm
.guc
.base
);
630 guc
->ggtt_pin_bias
= i915
->wopcm
.size
- i915
->wopcm
.guc
.base
;
634 * intel_guc_allocate_vma() - Allocate a GGTT VMA for GuC usage
636 * @size: size of area to allocate (both virtual space and memory)
638 * This is a wrapper to create an object for use with the GuC. In order to
639 * use it inside the GuC, an object needs to be pinned lifetime, so we allocate
640 * both some backing storage and a range inside the Global GTT. We must pin
641 * it in the GGTT somewhere other than than [0, GUC ggtt_pin_bias) because that
642 * range is reserved inside GuC.
644 * Return: A i915_vma if successful, otherwise an ERR_PTR.
646 struct i915_vma
*intel_guc_allocate_vma(struct intel_guc
*guc
, u32 size
)
648 struct drm_i915_private
*dev_priv
= guc_to_i915(guc
);
649 struct drm_i915_gem_object
*obj
;
650 struct i915_vma
*vma
;
653 obj
= i915_gem_object_create(dev_priv
, size
);
655 return ERR_CAST(obj
);
657 vma
= i915_vma_instance(obj
, &dev_priv
->ggtt
.vm
, NULL
);
661 ret
= i915_vma_pin(vma
, 0, PAGE_SIZE
,
662 PIN_GLOBAL
| PIN_OFFSET_BIAS
| guc
->ggtt_pin_bias
);
671 i915_gem_object_put(obj
);