1 /* SPDX-License-Identifier: MIT */
3 * Copyright © 2019 Intel Corporation
6 #ifndef __INTEL_ENGINE_TYPES__
7 #define __INTEL_ENGINE_TYPES__
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>
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"
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
36 #define COMPUTE_CLASS 5
37 #define MAX_ENGINE_CLASS 5
38 #define MAX_ENGINE_INSTANCE 8
40 #define I915_MAX_SLICES 3
41 #define I915_MAX_SUBSLICES 8
43 #define I915_CMD_HASH_ORDER 9
46 struct drm_i915_gem_object;
47 struct drm_i915_reg_table;
48 struct i915_gem_context;
50 struct i915_sched_attr;
51 struct i915_sched_engine;
55 struct intel_breadcrumbs;
56 struct intel_engine_cs;
57 struct i915_perf_group;
59 typedef 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)
63 struct intel_hw_status_page {
64 struct list_head timelines;
69 struct intel_instdone {
71 /* The following exist only in the RCS engine */
73 u32 slice_common_extra[2];
74 u32 sampler[GEN_MAX_GSLICES][I915_MAX_SUBSLICES];
75 u32 row[GEN_MAX_GSLICES][I915_MAX_SUBSLICES];
78 u32 geom_svg[GEN_MAX_GSLICES][I915_MAX_SUBSLICES];
82 * we use a single page to load ctx workarounds so all of these
83 * values are referred in terms of dwords
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
92 struct i915_ctx_workarounds {
93 struct i915_wa_ctx_bb {
96 } indirect_ctx, per_ctx;
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
108 * Engine IDs definitions.
109 * Keep instances of the same type engine together.
111 enum intel_engine_id {
122 #define _BCS(n) (BCS0 + (n))
131 #define _VCS(n) (VCS0 + (n))
136 #define _VECS(n) (VECS0 + (n))
141 #define _CCS(n) (CCS0 + (n))
144 #define INVALID_ENGINE ((enum intel_engine_id)-1)
147 /* A simple estimator for the round-trip latency of an engine */
148 DECLARE_EWMA(_engine_latency, 6, 4)
150 struct st_preempt_hang {
151 struct completion completion;
156 * struct intel_engine_execlists - execlist submission queue and port state
158 * The struct intel_engine_execlists represents the combined logical state of
159 * driver and the hardware state for execlist mode of submission.
161 struct intel_engine_execlists {
163 * @timer: kick the current context if its timeslice expires
165 struct timer_list timer;
168 * @preempt: reset the current context if it fails to give way
170 struct timer_list preempt;
173 * @preempt_target: active request at the time of the preemption request
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.
185 const struct i915_request *preempt_target;
188 * @ccid: identifier for contexts submitted to this engine
193 * @yield: CCID at the time of the last semaphore-wait interrupt.
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
202 * @error_interrupt: CS Master EIR
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.
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.
213 #define ERROR_CSB BIT(31)
214 #define ERROR_PREEMPT BIT(30)
217 * @reset_ccid: Active CCID [EXECLISTS_STATUS_HI] at the time of reset
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+
226 u32 __iomem *submit_reg;
229 * @ctrl_reg: the enhanced execlists control register, used to load the
230 * submit queue on the HW and to request preemptions to idle
232 u32 __iomem *ctrl_reg;
234 #define EXECLIST_MAX_PORTS 2
236 * @active: the currently known context executing on HW
238 struct i915_request * const *active;
240 * @inflight: the set of contexts submitted and acknowleged by HW
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.
247 struct i915_request *inflight[EXECLIST_MAX_PORTS + 1 /* sentinel */];
249 * @pending: the next set of contexts submitted to ELSP
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.
255 struct i915_request *pending[EXECLIST_MAX_PORTS + 1];
258 * @port_mask: number of execlist ports - 1
260 unsigned int port_mask;
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.
267 struct rb_root_cached virtual;
270 * @csb_write: control register for Context Switch buffer
272 * Note this register may be either mmio or HWSP shadow.
277 * @csb_status: status array for Context Switch buffer
279 * Note these register may be either mmio or HWSP shadow.
284 * @csb_size: context status buffer FIFO size
289 * @csb_head: context status buffer head
293 /* private: selftest */
294 I915_SELFTEST_DECLARE(struct st_preempt_hang preempt_hang;)
297 #define INTEL_ENGINE_CS_MAX_NAME 8
299 struct intel_engine_execlists_stats {
301 * @active: Number of contexts currently scheduled in.
306 * @lock: Lock protecting the below fields.
311 * @total: Total time this engine was busy.
313 * Accumulated time not counting the most recent block in cases where
314 * engine is currently busy (active > 0).
319 * @start: Timestamp of the last idle to active transition.
321 * Idle is defined as active == 0, active is active > 0.
326 struct intel_engine_guc_stats {
328 * @running: Active state of the engine when busyness was last sampled.
333 * @prev_total: Previous value of total runtime clock cycles.
338 * @total_gt_clks: Total gt clock cycles this engine was busy.
343 * @start_gt_clk: GT clock time of last idle to active transition.
348 union intel_engine_tlb_inv_reg {
350 i915_mcr_reg_t mcr_reg;
353 struct intel_engine_tlb_inv {
355 union intel_engine_tlb_inv_reg reg;
360 struct intel_engine_cs {
361 struct drm_i915_private *i915;
363 struct intel_uncore *uncore;
364 char name[INTEL_ENGINE_CS_MAX_NAME];
366 enum intel_engine_id id;
367 enum intel_engine_id legacy_idx;
371 intel_engine_mask_t mask;
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
379 intel_engine_mask_t logical_mask;
387 u32 uabi_capabilities;
391 struct intel_engine_tlb_inv tlb_inv;
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
400 enum forcewake_domains fw_domain;
401 unsigned int fw_active;
403 unsigned long context_tag;
405 struct rb_node uabi_node;
407 struct intel_sseu sseu;
409 struct i915_sched_engine *sched_engine;
411 /* keep a request in reserve for a [pm] barrier under oom */
412 struct i915_request *request_pool;
414 struct intel_context *hung_ce;
416 struct llist_head barrier_tasks;
418 struct intel_context *kernel_context; /* pinned */
421 * pinned_contexts_list: List of pinned contexts. This list is only
422 * assumed to be manipulated during driver load- or unload time and
423 * does therefore not have any additional protection.
425 struct list_head pinned_contexts_list;
427 intel_engine_mask_t saturated; /* submitting semaphores too late? */
430 struct delayed_work work;
431 struct i915_request *systole;
432 unsigned long blocked;
435 unsigned long serial;
437 unsigned long wakeref_serial;
438 struct intel_wakeref wakeref;
439 struct file *default_state;
442 struct intel_ring *ring;
443 struct intel_timeline *timeline;
447 * We track the average duration of the idle pulse on parking the
448 * engine to keep an estimate of the how the fast the engine is
449 * under ideal conditions.
451 struct ewma__engine_latency latency;
453 /* Keep track of all the seqno used, a trail of breadcrumbs */
454 struct intel_breadcrumbs *breadcrumbs;
456 struct intel_engine_pmu {
458 * @enable: Bitmask of enable sample events on this engine.
460 * Bits correspond to sample event types, for instance
461 * I915_SAMPLE_QUEUED is bit 0 etc.
465 * @enable_count: Reference count for the enabled samplers.
467 * Index number corresponds to @enum drm_i915_pmu_engine_sample.
469 unsigned int enable_count[I915_ENGINE_SAMPLE_COUNT];
471 * @sample: Counter values for sampling events.
473 * Our internal timer stores the current counters in this field.
475 * Index number corresponds to @enum drm_i915_pmu_engine_sample.
477 struct i915_pmu_sample sample[I915_ENGINE_SAMPLE_COUNT];
480 struct intel_hw_status_page status_page;
481 struct i915_ctx_workarounds wa_ctx;
482 struct i915_wa_list ctx_wa_list;
483 struct i915_wa_list wa_list;
484 struct i915_wa_list whitelist;
486 u32 irq_keep_mask; /* always keep these interrupts */
487 u32 irq_enable_mask; /* bitmask to enable ring interrupt */
488 void (*irq_enable)(struct intel_engine_cs *engine);
489 void (*irq_disable)(struct intel_engine_cs *engine);
490 void (*irq_handler)(struct intel_engine_cs *engine, u16 iir);
492 void (*sanitize)(struct intel_engine_cs *engine);
493 int (*resume)(struct intel_engine_cs *engine);
496 void (*prepare)(struct intel_engine_cs *engine);
498 void (*rewind)(struct intel_engine_cs *engine, bool stalled);
499 void (*cancel)(struct intel_engine_cs *engine);
501 void (*finish)(struct intel_engine_cs *engine);
504 void (*park)(struct intel_engine_cs *engine);
505 void (*unpark)(struct intel_engine_cs *engine);
507 void (*bump_serial)(struct intel_engine_cs *engine);
509 void (*set_default_submission)(struct intel_engine_cs *engine);
511 const struct intel_context_ops *cops;
513 int (*request_alloc)(struct i915_request *rq);
515 int (*emit_flush)(struct i915_request *request, u32 mode);
516 #define EMIT_INVALIDATE BIT(0)
517 #define EMIT_FLUSH BIT(1)
518 #define EMIT_BARRIER (EMIT_INVALIDATE | EMIT_FLUSH)
519 int (*emit_bb_start)(struct i915_request *rq,
520 u64 offset, u32 length,
521 unsigned int dispatch_flags);
522 #define I915_DISPATCH_SECURE BIT(0)
523 #define I915_DISPATCH_PINNED BIT(1)
524 int (*emit_init_breadcrumb)(struct i915_request *rq);
525 u32 *(*emit_fini_breadcrumb)(struct i915_request *rq,
527 unsigned int emit_fini_breadcrumb_dw;
529 /* Pass the request to the hardware queue (e.g. directly into
530 * the legacy ringbuffer or to the end of an execlist).
532 * This is called from an atomic context with irqs disabled; must
535 void (*submit_request)(struct i915_request *rq);
537 void (*release)(struct intel_engine_cs *engine);
540 * Add / remove request from engine active tracking
542 void (*add_active_request)(struct i915_request *rq);
543 void (*remove_active_request)(struct i915_request *rq);
546 * Get engine busyness and the time at which the busyness was sampled.
548 ktime_t (*busyness)(struct intel_engine_cs *engine,
551 struct intel_engine_execlists execlists;
554 * Keep track of completed timelines on this engine for early
555 * retirement with the goal of quickly enabling powersaving as
556 * soon as the engine is idle.
558 struct intel_timeline *retire;
559 struct work_struct retire_work;
561 /* status_notifier: list of callbacks for context-switch changes */
562 struct atomic_notifier_head context_status_notifier;
564 #define I915_ENGINE_USING_CMD_PARSER BIT(0)
565 #define I915_ENGINE_SUPPORTS_STATS BIT(1)
566 #define I915_ENGINE_HAS_PREEMPTION BIT(2)
567 #define I915_ENGINE_HAS_SEMAPHORES BIT(3)
568 #define I915_ENGINE_HAS_TIMESLICES BIT(4)
569 #define I915_ENGINE_IS_VIRTUAL BIT(5)
570 #define I915_ENGINE_HAS_RELATIVE_MMIO BIT(6)
571 #define I915_ENGINE_REQUIRES_CMD_PARSER BIT(7)
572 #define I915_ENGINE_WANT_FORCED_PREEMPTION BIT(8)
573 #define I915_ENGINE_HAS_RCS_REG_STATE BIT(9)
574 #define I915_ENGINE_HAS_EU_PRIORITY BIT(10)
575 #define I915_ENGINE_FIRST_RENDER_COMPUTE BIT(11)
576 #define I915_ENGINE_USES_WA_HOLD_CCS_SWITCHOUT BIT(12)
580 * Table of commands the command parser needs to know about
583 DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER);
586 * Table of registers allowed in commands that read/write registers.
588 const struct drm_i915_reg_table *reg_tables;
592 * Returns the bitmask for the length field of the specified command.
593 * Return 0 for an unrecognized/invalid command.
595 * If the command parser finds an entry for a command in the engine's
596 * cmd_tables, it gets the command's length based on the table entry.
597 * If not, it calls this function to determine the per-engine length
598 * field encoding for the command (i.e. different opcode ranges use
599 * certain bits to encode the command length in the header).
601 u32 (*get_cmd_length_mask)(u32 cmd_header);
605 struct intel_engine_execlists_stats execlists;
606 struct intel_engine_guc_stats guc;
610 * @rps: Utilisation at last RPS sampling.
616 unsigned long heartbeat_interval_ms;
617 unsigned long max_busywait_duration_ns;
618 unsigned long preempt_timeout_ms;
619 unsigned long stop_timeout_ms;
620 unsigned long timeslice_duration_ms;
623 I915_SELFTEST_DECLARE(struct fault_attr reset_timeout);
626 * The perf group maps to one OA unit which controls one OA buffer. All
627 * reports corresponding to this engine will be reported to this OA
628 * buffer. An engine will map to a single OA unit, but a single OA unit
629 * can generate reports for multiple engines.
631 struct i915_perf_group *oa_group;
635 intel_engine_using_cmd_parser(const struct intel_engine_cs *engine)
637 return engine->flags & I915_ENGINE_USING_CMD_PARSER;
641 intel_engine_requires_cmd_parser(const struct intel_engine_cs *engine)
643 return engine->flags & I915_ENGINE_REQUIRES_CMD_PARSER;
647 intel_engine_supports_stats(const struct intel_engine_cs *engine)
649 return engine->flags & I915_ENGINE_SUPPORTS_STATS;
653 intel_engine_has_preemption(const struct intel_engine_cs *engine)
655 return engine->flags & I915_ENGINE_HAS_PREEMPTION;
659 intel_engine_has_semaphores(const struct intel_engine_cs *engine)
661 return engine->flags & I915_ENGINE_HAS_SEMAPHORES;
665 intel_engine_has_timeslices(const struct intel_engine_cs *engine)
667 if (!CONFIG_DRM_I915_TIMESLICE_DURATION)
670 return engine->flags & I915_ENGINE_HAS_TIMESLICES;
674 intel_engine_is_virtual(const struct intel_engine_cs *engine)
676 return engine->flags & I915_ENGINE_IS_VIRTUAL;
680 intel_engine_has_relative_mmio(const struct intel_engine_cs * const engine)
682 return engine->flags & I915_ENGINE_HAS_RELATIVE_MMIO;
685 /* Wa_14014475959:dg2 */
687 intel_engine_uses_wa_hold_ccs_switchout(struct intel_engine_cs *engine)
689 return engine->flags & I915_ENGINE_USES_WA_HOLD_CCS_SWITCHOUT;
692 #endif /* __INTEL_ENGINE_TYPES_H__ */