1/* SPDX-License-Identifier: MIT */
2/*
3 * Copyright © 2019 Intel Corporation
4 */
5
6#ifndef __INTEL_ENGINE_TYPES__
7#define __INTEL_ENGINE_TYPES__
8
9#include <linux/average.h>
10#include <linux/hashtable.h>
11#include <linux/irq_work.h>
12#include <linux/kref.h>
13#include <linux/list.h>
14#include <linux/llist.h>
15#include <linux/rbtree.h>
16#include <linux/timer.h>
17#include <linux/types.h>
18#include <linux/workqueue.h>
19
20#include "i915_gem.h"
21#include "i915_pmu.h"
22#include "i915_priolist_types.h"
23#include "i915_selftest.h"
24#include "intel_sseu.h"
25#include "intel_timeline_types.h"
26#include "intel_uncore.h"
27#include "intel_wakeref.h"
28#include "intel_workarounds_types.h"
29
30/* HW Engine class + instance */
31#define RENDER_CLASS 0
32#define VIDEO_DECODE_CLASS 1
33#define VIDEO_ENHANCEMENT_CLASS 2
34#define COPY_ENGINE_CLASS 3
35#define OTHER_CLASS 4
36#define COMPUTE_CLASS 5
37#define MAX_ENGINE_CLASS 5
38#define MAX_ENGINE_INSTANCE 8
39
40#define I915_MAX_SLICES 3
41#define I915_MAX_SUBSLICES 8
42
43#define I915_CMD_HASH_ORDER 9
44
45struct dma_fence;
46struct drm_i915_gem_object;
47struct drm_i915_reg_table;
48struct i915_gem_context;
49struct i915_request;
50struct i915_sched_attr;
51struct i915_sched_engine;
52struct intel_gt;
53struct intel_ring;
54struct intel_uncore;
55struct intel_breadcrumbs;
56struct intel_engine_cs;
57struct i915_perf_group;
58
59typedef u32 intel_engine_mask_t;
60#define ALL_ENGINES ((intel_engine_mask_t)~0ul)
61#define VIRTUAL_ENGINES BIT(BITS_PER_TYPE(intel_engine_mask_t) - 1)
62
63struct intel_hw_status_page {
64 struct list_head timelines;
65 struct i915_vma *vma;
66 u32 *addr;
67};
68
69struct intel_instdone {
70 u32 instdone;
71 /* The following exist only in the RCS engine */
72 u32 slice_common;
73 u32 slice_common_extra[2];
74 u32 sampler[GEN_MAX_GSLICES][I915_MAX_SUBSLICES];
75 u32 row[GEN_MAX_GSLICES][I915_MAX_SUBSLICES];
76
77 /* Added in XeHPG */
78 u32 geom_svg[GEN_MAX_GSLICES][I915_MAX_SUBSLICES];
79};
80
81/*
82 * we use a single page to load ctx workarounds so all of these
83 * values are referred in terms of dwords
84 *
85 * struct i915_wa_ctx_bb:
86 * offset: specifies batch starting position, also helpful in case
87 * if we want to have multiple batches at different offsets based on
88 * some criteria. It is not a requirement at the moment but provides
89 * an option for future use.
90 * size: size of the batch in DWORDS
91 */
92struct i915_ctx_workarounds {
93 struct i915_wa_ctx_bb {
94 u32 offset;
95 u32 size;
96 } indirect_ctx, per_ctx;
97 struct i915_vma *vma;
98};
99
100#define I915_MAX_VCS 8
101#define I915_MAX_VECS 4
102#define I915_MAX_SFC (I915_MAX_VCS / 2)
103#define I915_MAX_CCS 4
104#define I915_MAX_RCS 1
105#define I915_MAX_BCS 9
106
107/*
108 * Engine IDs definitions.
109 * Keep instances of the same type engine together.
110 */
111enum intel_engine_id {
112 RCS0 = 0,
113 BCS0,
114 BCS1,
115 BCS2,
116 BCS3,
117 BCS4,
118 BCS5,
119 BCS6,
120 BCS7,
121 BCS8,
122#define _BCS(n) (BCS0 + (n))
123 VCS0,
124 VCS1,
125 VCS2,
126 VCS3,
127 VCS4,
128 VCS5,
129 VCS6,
130 VCS7,
131#define _VCS(n) (VCS0 + (n))
132 VECS0,
133 VECS1,
134 VECS2,
135 VECS3,
136#define _VECS(n) (VECS0 + (n))
137 CCS0,
138 CCS1,
139 CCS2,
140 CCS3,
141#define _CCS(n) (CCS0 + (n))
142 GSC0,
143 I915_NUM_ENGINES
144#define INVALID_ENGINE ((enum intel_engine_id)-1)
145};
146
147/* A simple estimator for the round-trip latency of an engine */
148DECLARE_EWMA(_engine_latency, 6, 4)
149
150struct st_preempt_hang {
151 struct completion completion;
152 unsigned int count;
153};
154
155/**
156 * struct intel_engine_execlists - execlist submission queue and port state
157 *
158 * The struct intel_engine_execlists represents the combined logical state of
159 * driver and the hardware state for execlist mode of submission.
160 */
161struct intel_engine_execlists {
162 /**
163 * @timer: kick the current context if its timeslice expires
164 */
165 struct timer_list timer;
166
167 /**
168 * @preempt: reset the current context if it fails to give way
169 */
170 struct timer_list preempt;
171
172 /**
173 * @preempt_target: active request at the time of the preemption request
174 *
175 * We force a preemption to occur if the pending contexts have not
176 * been promoted to active upon receipt of the CS ack event within
177 * the timeout. This timeout maybe chosen based on the target,
178 * using a very short timeout if the context is no longer schedulable.
179 * That short timeout may not be applicable to other contexts, so
180 * if a context switch should happen within before the preemption
181 * timeout, we may shoot early at an innocent context. To prevent this,
182 * we record which context was active at the time of the preemption
183 * request and only reset that context upon the timeout.
184 */
185 const struct i915_request *preempt_target;
186
187 /**
188 * @ccid: identifier for contexts submitted to this engine
189 */
190 u32 ccid;
191
192 /**
193 * @yield: CCID at the time of the last semaphore-wait interrupt.
194 *
195 * Instead of leaving a semaphore busy-spinning on an engine, we would
196 * like to switch to another ready context, i.e. yielding the semaphore
197 * timeslice.
198 */
199 u32 yield;
200
201 /**
202 * @error_interrupt: CS Master EIR
203 *
204 * The CS generates an interrupt when it detects an error. We capture
205 * the first error interrupt, record the EIR and schedule the tasklet.
206 * In the tasklet, we process the pending CS events to ensure we have
207 * the guilty request, and then reset the engine.
208 *
209 * Low 16b are used by HW, with the upper 16b used as the enabling mask.
210 * Reserve the upper 16b for tracking internal errors.
211 */
212 u32 error_interrupt;
213#define ERROR_CSB BIT(31)
214#define ERROR_PREEMPT BIT(30)
215
216 /**
217 * @reset_ccid: Active CCID [EXECLISTS_STATUS_HI] at the time of reset
218 */
219 u32 reset_ccid;
220
221 /**
222 * @submit_reg: gen-specific execlist submission register
223 * set to the ExecList Submission Port (elsp) register pre-Gen11 and to
224 * the ExecList Submission Queue Contents register array for Gen11+
225 */
226 u32 __iomem *submit_reg;
227
228 /**
229 * @ctrl_reg: the enhanced execlists control register, used to load the
230 * submit queue on the HW and to request preemptions to idle
231 */
232 u32 __iomem *ctrl_reg;
233
234#define EXECLIST_MAX_PORTS 2
235 /**
236 * @active: the currently known context executing on HW
237 */
238 struct i915_request * const *active;
239 /**
240 * @inflight: the set of contexts submitted and acknowleged by HW
241 *
242 * The set of inflight contexts is managed by reading CS events
243 * from the HW. On a context-switch event (not preemption), we
244 * know the HW has transitioned from port0 to port1, and we
245 * advance our inflight/active tracking accordingly.
246 */
247 struct i915_request *inflight[EXECLIST_MAX_PORTS + 1 /* sentinel */];
248 /**
249 * @pending: the next set of contexts submitted to ELSP
250 *
251 * We store the array of contexts that we submit to HW (via ELSP) and
252 * promote them to the inflight array once HW has signaled the
253 * preemption or idle-to-active event.
254 */
255 struct i915_request *pending[EXECLIST_MAX_PORTS + 1];
256
257 /**
258 * @port_mask: number of execlist ports - 1
259 */
260 unsigned int port_mask;
261
262 /**
263 * @virtual: Queue of requets on a virtual engine, sorted by priority.
264 * Each RB entry is a struct i915_priolist containing a list of requests
265 * of the same priority.
266 */
267 struct rb_root_cached virtual;
268
269 /**
270 * @csb_write: control register for Context Switch buffer
271 *
272 * Note this register may be either mmio or HWSP shadow.
273 */
274 u32 *csb_write;
275
276 /**
277 * @csb_status: status array for Context Switch buffer
278 *
279 * Note these register may be either mmio or HWSP shadow.
280 */
281 u64 *csb_status;
282
283 /**
284 * @csb_size: context status buffer FIFO size
285 */
286 u8 csb_size;
287
288 /**
289 * @csb_head: context status buffer head
290 */
291 u8 csb_head;
292
293 /* private: selftest */
294 I915_SELFTEST_DECLARE(struct st_preempt_hang preempt_hang;)
295};
296
297#define INTEL_ENGINE_CS_MAX_NAME 8
298
299struct intel_engine_execlists_stats {
300 /**
301 * @active: Number of contexts currently scheduled in.
302 */
303 unsigned int active;
304
305 /**
306 * @lock: Lock protecting the below fields.
307 */
308 seqcount_t lock;
309
310 /**
311 * @total: Total time this engine was busy.
312 *
313 * Accumulated time not counting the most recent block in cases where
314 * engine is currently busy (active > 0).
315 */
316 ktime_t total;
317
318 /**
319 * @start: Timestamp of the last idle to active transition.
320 *
321 * Idle is defined as active == 0, active is active > 0.
322 */
323 ktime_t start;
324};
325
326struct intel_engine_guc_stats {
327 /**
328 * @running: Active state of the engine when busyness was last sampled.
329 */
330 bool running;
331
332 /**
333 * @prev_total: Previous value of total runtime clock cycles.
334 */
335 u32 prev_total;
336
337 /**
338 * @total_gt_clks: Total gt clock cycles this engine was busy.
339 */
340 u64 total_gt_clks;
341
342 /**
343 * @start_gt_clk: GT clock time of last idle to active transition.
344 */
345 u64 start_gt_clk;
346};
347
348union intel_engine_tlb_inv_reg {
349 i915_reg_t reg;
350 i915_mcr_reg_t mcr_reg;
351};
352
353struct intel_engine_tlb_inv {
354 bool mcr;
355 union intel_engine_tlb_inv_reg reg;
356 u32 request;
357 u32 done;
358};
359
360struct intel_engine_cs {
361 struct drm_i915_private *i915;
362 struct intel_gt *gt;
363 struct intel_uncore *uncore;
364 char name[INTEL_ENGINE_CS_MAX_NAME];
365
366 enum intel_engine_id id;
367 enum intel_engine_id legacy_idx;
368
369 unsigned int guc_id;
370
371 intel_engine_mask_t mask;
372 u32 reset_domain;
373 /**
374 * @logical_mask: logical mask of engine, reported to user space via
375 * query IOCTL and used to communicate with the GuC in logical space.
376 * The logical instance of a physical engine can change based on product
377 * and fusing.
378 */
379 intel_engine_mask_t logical_mask;
380
381 u8 class;
382 u8 instance;
383
384 u16 uabi_class;
385 u16 uabi_instance;
386
387 u32 uabi_capabilities;
388 u32 context_size;
389 u32 mmio_base;
390
391 struct intel_engine_tlb_inv tlb_inv;
392
393 /*
394 * Some w/a require forcewake to be held (which prevents RC6) while
395 * a particular engine is active. If so, we set fw_domain to which
396 * domains need to be held for the duration of request activity,
397 * and 0 if none. We try to limit the duration of the hold as much
398 * as possible.
399 */
400 enum forcewake_domains fw_domain;
401 unsigned int fw_active;
402
403 unsigned long context_tag;
404
405 /*
406 * The type evolves during initialization, see related comment for
407 * struct drm_i915_private's uabi_engines member.
408 */
409 union {
410 struct llist_node uabi_llist;
411 struct list_head uabi_list;
412 struct rb_node uabi_node;
413 };
414
415 struct intel_sseu sseu;
416
417 struct i915_sched_engine *sched_engine;
418
419 /* keep a request in reserve for a [pm] barrier under oom */
420 struct i915_request *request_pool;
421
422 struct intel_context *hung_ce;
423
424 struct llist_head barrier_tasks;
425
426 struct intel_context *kernel_context; /* pinned */
427 struct intel_context *bind_context; /* pinned, only for BCS0 */
428 /* mark the bind context's availability status */
429 bool bind_context_ready;
430
431 /**
432 * pinned_contexts_list: List of pinned contexts. This list is only
433 * assumed to be manipulated during driver load- or unload time and
434 * does therefore not have any additional protection.
435 */
436 struct list_head pinned_contexts_list;
437
438 intel_engine_mask_t saturated; /* submitting semaphores too late? */
439
440 struct {
441 struct delayed_work work;
442 struct i915_request *systole;
443 unsigned long blocked;
444 } heartbeat;
445
446 unsigned long serial;
447
448 unsigned long wakeref_serial;
449 intel_wakeref_t wakeref_track;
450 struct intel_wakeref wakeref;
451
452 struct file *default_state;
453
454 struct {
455 struct intel_ring *ring;
456 struct intel_timeline *timeline;
457 } legacy;
458
459 /*
460 * We track the average duration of the idle pulse on parking the
461 * engine to keep an estimate of the how the fast the engine is
462 * under ideal conditions.
463 */
464 struct ewma__engine_latency latency;
465
466 /* Keep track of all the seqno used, a trail of breadcrumbs */
467 struct intel_breadcrumbs *breadcrumbs;
468
469 struct intel_engine_pmu {
470 /**
471 * @enable: Bitmask of enable sample events on this engine.
472 *
473 * Bits correspond to sample event types, for instance
474 * I915_SAMPLE_QUEUED is bit 0 etc.
475 */
476 u32 enable;
477 /**
478 * @enable_count: Reference count for the enabled samplers.
479 *
480 * Index number corresponds to @enum drm_i915_pmu_engine_sample.
481 */
482 unsigned int enable_count[I915_ENGINE_SAMPLE_COUNT];
483 /**
484 * @sample: Counter values for sampling events.
485 *
486 * Our internal timer stores the current counters in this field.
487 *
488 * Index number corresponds to @enum drm_i915_pmu_engine_sample.
489 */
490 struct i915_pmu_sample sample[I915_ENGINE_SAMPLE_COUNT];
491 } pmu;
492
493 struct intel_hw_status_page status_page;
494 struct i915_ctx_workarounds wa_ctx;
495 struct i915_wa_list ctx_wa_list;
496 struct i915_wa_list wa_list;
497 struct i915_wa_list whitelist;
498
499 u32 irq_keep_mask; /* always keep these interrupts */
500 u32 irq_enable_mask; /* bitmask to enable ring interrupt */
501 void (*irq_enable)(struct intel_engine_cs *engine);
502 void (*irq_disable)(struct intel_engine_cs *engine);
503 void (*irq_handler)(struct intel_engine_cs *engine, u16 iir);
504
505 void (*sanitize)(struct intel_engine_cs *engine);
506 int (*resume)(struct intel_engine_cs *engine);
507
508 struct {
509 void (*prepare)(struct intel_engine_cs *engine);
510
511 void (*rewind)(struct intel_engine_cs *engine, bool stalled);
512 void (*cancel)(struct intel_engine_cs *engine);
513
514 void (*finish)(struct intel_engine_cs *engine);
515 } reset;
516
517 void (*park)(struct intel_engine_cs *engine);
518 void (*unpark)(struct intel_engine_cs *engine);
519
520 void (*bump_serial)(struct intel_engine_cs *engine);
521
522 void (*set_default_submission)(struct intel_engine_cs *engine);
523
524 const struct intel_context_ops *cops;
525
526 int (*request_alloc)(struct i915_request *rq);
527
528 int (*emit_flush)(struct i915_request *request, u32 mode);
529#define EMIT_INVALIDATE BIT(0)
530#define EMIT_FLUSH BIT(1)
531#define EMIT_BARRIER (EMIT_INVALIDATE | EMIT_FLUSH)
532 int (*emit_bb_start)(struct i915_request *rq,
533 u64 offset, u32 length,
534 unsigned int dispatch_flags);
535#define I915_DISPATCH_SECURE BIT(0)
536#define I915_DISPATCH_PINNED BIT(1)
537 int (*emit_init_breadcrumb)(struct i915_request *rq);
538 u32 *(*emit_fini_breadcrumb)(struct i915_request *rq,
539 u32 *cs);
540 unsigned int emit_fini_breadcrumb_dw;
541
542 /* Pass the request to the hardware queue (e.g. directly into
543 * the legacy ringbuffer or to the end of an execlist).
544 *
545 * This is called from an atomic context with irqs disabled; must
546 * be irq safe.
547 */
548 void (*submit_request)(struct i915_request *rq);
549
550 void (*release)(struct intel_engine_cs *engine);
551
552 /*
553 * Add / remove request from engine active tracking
554 */
555 void (*add_active_request)(struct i915_request *rq);
556 void (*remove_active_request)(struct i915_request *rq);
557
558 /*
559 * Get engine busyness and the time at which the busyness was sampled.
560 */
561 ktime_t (*busyness)(struct intel_engine_cs *engine,
562 ktime_t *now);
563
564 struct intel_engine_execlists execlists;
565
566 /*
567 * Keep track of completed timelines on this engine for early
568 * retirement with the goal of quickly enabling powersaving as
569 * soon as the engine is idle.
570 */
571 struct intel_timeline *retire;
572 struct work_struct retire_work;
573
574 /* status_notifier: list of callbacks for context-switch changes */
575 struct atomic_notifier_head context_status_notifier;
576
577#define I915_ENGINE_USING_CMD_PARSER BIT(0)
578#define I915_ENGINE_SUPPORTS_STATS BIT(1)
579#define I915_ENGINE_HAS_PREEMPTION BIT(2)
580#define I915_ENGINE_HAS_SEMAPHORES BIT(3)
581#define I915_ENGINE_HAS_TIMESLICES BIT(4)
582#define I915_ENGINE_IS_VIRTUAL BIT(5)
583#define I915_ENGINE_HAS_RELATIVE_MMIO BIT(6)
584#define I915_ENGINE_REQUIRES_CMD_PARSER BIT(7)
585#define I915_ENGINE_WANT_FORCED_PREEMPTION BIT(8)
586#define I915_ENGINE_HAS_RCS_REG_STATE BIT(9)
587#define I915_ENGINE_HAS_EU_PRIORITY BIT(10)
588#define I915_ENGINE_FIRST_RENDER_COMPUTE BIT(11)
589#define I915_ENGINE_USES_WA_HOLD_CCS_SWITCHOUT BIT(12)
590 unsigned int flags;
591
592 /*
593 * Table of commands the command parser needs to know about
594 * for this engine.
595 */
596 DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER);
597
598 /*
599 * Table of registers allowed in commands that read/write registers.
600 */
601 const struct drm_i915_reg_table *reg_tables;
602 int reg_table_count;
603
604 /*
605 * Returns the bitmask for the length field of the specified command.
606 * Return 0 for an unrecognized/invalid command.
607 *
608 * If the command parser finds an entry for a command in the engine's
609 * cmd_tables, it gets the command's length based on the table entry.
610 * If not, it calls this function to determine the per-engine length
611 * field encoding for the command (i.e. different opcode ranges use
612 * certain bits to encode the command length in the header).
613 */
614 u32 (*get_cmd_length_mask)(u32 cmd_header);
615
616 struct {
617 union {
618 struct intel_engine_execlists_stats execlists;
619 struct intel_engine_guc_stats guc;
620 };
621
622 /**
623 * @rps: Utilisation at last RPS sampling.
624 */
625 ktime_t rps;
626 } stats;
627
628 struct {
629 unsigned long heartbeat_interval_ms;
630 unsigned long max_busywait_duration_ns;
631 unsigned long preempt_timeout_ms;
632 unsigned long stop_timeout_ms;
633 unsigned long timeslice_duration_ms;
634 } props, defaults;
635
636 I915_SELFTEST_DECLARE(struct fault_attr reset_timeout);
637
638 /*
639 * The perf group maps to one OA unit which controls one OA buffer. All
640 * reports corresponding to this engine will be reported to this OA
641 * buffer. An engine will map to a single OA unit, but a single OA unit
642 * can generate reports for multiple engines.
643 */
644 struct i915_perf_group *oa_group;
645};
646
647static inline bool
648intel_engine_using_cmd_parser(const struct intel_engine_cs *engine)
649{
650 return engine->flags & I915_ENGINE_USING_CMD_PARSER;
651}
652
653static inline bool
654intel_engine_requires_cmd_parser(const struct intel_engine_cs *engine)
655{
656 return engine->flags & I915_ENGINE_REQUIRES_CMD_PARSER;
657}
658
659static inline bool
660intel_engine_supports_stats(const struct intel_engine_cs *engine)
661{
662 return engine->flags & I915_ENGINE_SUPPORTS_STATS;
663}
664
665static inline bool
666intel_engine_has_preemption(const struct intel_engine_cs *engine)
667{
668 return engine->flags & I915_ENGINE_HAS_PREEMPTION;
669}
670
671static inline bool
672intel_engine_has_semaphores(const struct intel_engine_cs *engine)
673{
674 return engine->flags & I915_ENGINE_HAS_SEMAPHORES;
675}
676
677static inline bool
678intel_engine_has_timeslices(const struct intel_engine_cs *engine)
679{
680 if (!CONFIG_DRM_I915_TIMESLICE_DURATION)
681 return false;
682
683 return engine->flags & I915_ENGINE_HAS_TIMESLICES;
684}
685
686static inline bool
687intel_engine_is_virtual(const struct intel_engine_cs *engine)
688{
689 return engine->flags & I915_ENGINE_IS_VIRTUAL;
690}
691
692static inline bool
693intel_engine_has_relative_mmio(const struct intel_engine_cs * const engine)
694{
695 return engine->flags & I915_ENGINE_HAS_RELATIVE_MMIO;
696}
697
698/* Wa_14014475959:dg2 */
699static inline bool
700intel_engine_uses_wa_hold_ccs_switchout(struct intel_engine_cs *engine)
701{
702 return engine->flags & I915_ENGINE_USES_WA_HOLD_CCS_SWITCHOUT;
703}
704
705#endif /* __INTEL_ENGINE_TYPES_H__ */
706

source code of linux/drivers/gpu/drm/i915/gt/intel_engine_types.h