1/* SPDX-License-Identifier: MIT */
2/*
3 * Copyright �� 2014-2019 Intel Corporation
4 */
5
6#ifndef _INTEL_GUC_H_
7#define _INTEL_GUC_H_
8
9#include <linux/delay.h>
10#include <linux/iosys-map.h>
11#include <linux/xarray.h>
12
13#include "intel_guc_ct.h"
14#include "intel_guc_fw.h"
15#include "intel_guc_fwif.h"
16#include "intel_guc_log.h"
17#include "intel_guc_reg.h"
18#include "intel_guc_slpc_types.h"
19#include "intel_uc_fw.h"
20#include "intel_uncore.h"
21#include "i915_utils.h"
22#include "i915_vma.h"
23
24struct __guc_ads_blob;
25struct intel_guc_state_capture;
26
27/**
28 * struct intel_guc - Top level structure of GuC.
29 *
30 * It handles firmware loading and manages client pool. intel_guc owns an
31 * i915_sched_engine for submission.
32 */
33struct intel_guc {
34	/** @fw: the GuC firmware */
35	struct intel_uc_fw fw;
36	/** @log: sub-structure containing GuC log related data and objects */
37	struct intel_guc_log log;
38	/** @ct: the command transport communication channel */
39	struct intel_guc_ct ct;
40	/** @slpc: sub-structure containing SLPC related data and objects */
41	struct intel_guc_slpc slpc;
42	/** @capture: the error-state-capture module's data and objects */
43	struct intel_guc_state_capture *capture;
44
45	/** @dbgfs_node: debugfs node */
46	struct dentry *dbgfs_node;
47
48	/** @sched_engine: Global engine used to submit requests to GuC */
49	struct i915_sched_engine *sched_engine;
50	/**
51	 * @stalled_request: if GuC can't process a request for any reason, we
52	 * save it until GuC restarts processing. No other request can be
53	 * submitted until the stalled request is processed.
54	 */
55	struct i915_request *stalled_request;
56	/**
57	 * @submission_stall_reason: reason why submission is stalled
58	 */
59	enum {
60		STALL_NONE,
61		STALL_REGISTER_CONTEXT,
62		STALL_MOVE_LRC_TAIL,
63		STALL_ADD_REQUEST,
64	} submission_stall_reason;
65
66	/* intel_guc_recv interrupt related state */
67	/** @irq_lock: protects GuC irq state */
68	spinlock_t irq_lock;
69	/**
70	 * @msg_enabled_mask: mask of events that are processed when receiving
71	 * an INTEL_GUC_ACTION_DEFAULT G2H message.
72	 */
73	unsigned int msg_enabled_mask;
74
75	/**
76	 * @outstanding_submission_g2h: number of outstanding GuC to Host
77	 * responses related to GuC submission, used to determine if the GT is
78	 * idle
79	 */
80	atomic_t outstanding_submission_g2h;
81
82	/** @tlb_lookup: xarray to store all pending TLB invalidation requests */
83	struct xarray tlb_lookup;
84
85	/**
86	 * @serial_slot: id to the initial waiter created in tlb_lookup,
87	 * which is used only when failed to allocate new waiter.
88	 */
89	u32 serial_slot;
90
91	/** @next_seqno: the next id (sequence number) to allocate. */
92	u32 next_seqno;
93
94	/** @interrupts: pointers to GuC interrupt-managing functions. */
95	struct {
96		bool enabled;
97		void (*reset)(struct intel_guc *guc);
98		void (*enable)(struct intel_guc *guc);
99		void (*disable)(struct intel_guc *guc);
100	} interrupts;
101
102	/**
103	 * @submission_state: sub-structure for submission state protected by
104	 * single lock
105	 */
106	struct {
107		/**
108		 * @submission_state.lock: protects everything in
109		 * submission_state, ce->guc_id.id, and ce->guc_id.ref
110		 * when transitioning in and out of zero
111		 */
112		spinlock_t lock;
113		/**
114		 * @submission_state.guc_ids: used to allocate new
115		 * guc_ids, single-lrc
116		 */
117		struct ida guc_ids;
118		/**
119		 * @submission_state.num_guc_ids: Number of guc_ids, selftest
120		 * feature to be able to reduce this number while testing.
121		 */
122		int num_guc_ids;
123		/**
124		 * @submission_state.guc_ids_bitmap: used to allocate
125		 * new guc_ids, multi-lrc
126		 */
127		unsigned long *guc_ids_bitmap;
128		/**
129		 * @submission_state.guc_id_list: list of intel_context
130		 * with valid guc_ids but no refs
131		 */
132		struct list_head guc_id_list;
133		/**
134		 * @submission_state.guc_ids_in_use: Number single-lrc
135		 * guc_ids in use
136		 */
137		unsigned int guc_ids_in_use;
138		/**
139		 * @submission_state.destroyed_contexts: list of contexts
140		 * waiting to be destroyed (deregistered with the GuC)
141		 */
142		struct list_head destroyed_contexts;
143		/**
144		 * @submission_state.destroyed_worker: worker to deregister
145		 * contexts, need as we need to take a GT PM reference and
146		 * can't from destroy function as it might be in an atomic
147		 * context (no sleeping)
148		 */
149		struct work_struct destroyed_worker;
150		/**
151		 * @submission_state.reset_fail_worker: worker to trigger
152		 * a GT reset after an engine reset fails
153		 */
154		struct work_struct reset_fail_worker;
155		/**
156		 * @submission_state.reset_fail_mask: mask of engines that
157		 * failed to reset
158		 */
159		intel_engine_mask_t reset_fail_mask;
160		/**
161		 * @submission_state.sched_disable_delay_ms: schedule
162		 * disable delay, in ms, for contexts
163		 */
164		unsigned int sched_disable_delay_ms;
165		/**
166		 * @submission_state.sched_disable_gucid_threshold:
167		 * threshold of min remaining available guc_ids before
168		 * we start bypassing the schedule disable delay
169		 */
170		unsigned int sched_disable_gucid_threshold;
171	} submission_state;
172
173	/**
174	 * @submission_supported: tracks whether we support GuC submission on
175	 * the current platform
176	 */
177	bool submission_supported;
178	/** @submission_selected: tracks whether the user enabled GuC submission */
179	bool submission_selected;
180	/** @submission_initialized: tracks whether GuC submission has been initialised */
181	bool submission_initialized;
182	/** @submission_version: Submission API version of the currently loaded firmware */
183	struct intel_uc_fw_ver submission_version;
184
185	/**
186	 * @rc_supported: tracks whether we support GuC rc on the current platform
187	 */
188	bool rc_supported;
189	/** @rc_selected: tracks whether the user enabled GuC rc */
190	bool rc_selected;
191
192	/** @ads_vma: object allocated to hold the GuC ADS */
193	struct i915_vma *ads_vma;
194	/** @ads_map: contents of the GuC ADS */
195	struct iosys_map ads_map;
196	/** @ads_regset_size: size of the save/restore regsets in the ADS */
197	u32 ads_regset_size;
198	/**
199	 * @ads_regset_count: number of save/restore registers in the ADS for
200	 * each engine
201	 */
202	u32 ads_regset_count[I915_NUM_ENGINES];
203	/** @ads_regset: save/restore regsets in the ADS */
204	struct guc_mmio_reg *ads_regset;
205	/** @ads_golden_ctxt_size: size of the golden contexts in the ADS */
206	u32 ads_golden_ctxt_size;
207	/** @ads_waklv_size: size of workaround KLVs */
208	u32 ads_waklv_size;
209	/** @ads_capture_size: size of register lists in the ADS used for error capture */
210	u32 ads_capture_size;
211
212	/** @lrc_desc_pool_v69: object allocated to hold the GuC LRC descriptor pool */
213	struct i915_vma *lrc_desc_pool_v69;
214	/** @lrc_desc_pool_vaddr_v69: contents of the GuC LRC descriptor pool */
215	void *lrc_desc_pool_vaddr_v69;
216
217	/**
218	 * @context_lookup: used to resolve intel_context from guc_id, if a
219	 * context is present in this structure it is registered with the GuC
220	 */
221	struct xarray context_lookup;
222
223	/** @params: Control params for fw initialization */
224	u32 params[GUC_CTL_MAX_DWORDS];
225
226	/** @send_regs: GuC's FW specific registers used for sending MMIO H2G */
227	struct {
228		u32 base;
229		unsigned int count;
230		enum forcewake_domains fw_domains;
231	} send_regs;
232
233	/** @notify_reg: register used to send interrupts to the GuC FW */
234	i915_reg_t notify_reg;
235
236	/**
237	 * @mmio_msg: notification bitmask that the GuC writes in one of its
238	 * registers when the CT channel is disabled, to be processed when the
239	 * channel is back up.
240	 */
241	u32 mmio_msg;
242
243	/** @send_mutex: used to serialize the intel_guc_send actions */
244	struct mutex send_mutex;
245
246	/**
247	 * @timestamp: GT timestamp object that stores a copy of the timestamp
248	 * and adjusts it for overflow using a worker.
249	 */
250	struct {
251		/**
252		 * @timestamp.lock: Lock protecting the below fields and
253		 * the engine stats.
254		 */
255		spinlock_t lock;
256
257		/**
258		 * @timestamp.gt_stamp: 64-bit extended value of the GT
259		 * timestamp.
260		 */
261		u64 gt_stamp;
262
263		/**
264		 * @timestamp.ping_delay: Period for polling the GT
265		 * timestamp for overflow.
266		 */
267		unsigned long ping_delay;
268
269		/**
270		 * @timestamp.work: Periodic work to adjust GT timestamp,
271		 * engine and context usage for overflows.
272		 */
273		struct delayed_work work;
274
275		/**
276		 * @timestamp.shift: Right shift value for the gpm timestamp
277		 */
278		u32 shift;
279
280		/**
281		 * @timestamp.last_stat_jiffies: jiffies at last actual
282		 * stats collection time. We use this timestamp to ensure
283		 * we don't oversample the stats because runtime power
284		 * management events can trigger stats collection at much
285		 * higher rates than required.
286		 */
287		unsigned long last_stat_jiffies;
288	} timestamp;
289
290	/**
291	 * @dead_guc_worker: Asynchronous worker thread for forcing a GuC reset.
292	 * Specifically used when the G2H handler wants to issue a reset. Resets
293	 * require flushing the G2H queue. So, the G2H processing itself must not
294	 * trigger a reset directly. Instead, go via this worker.
295	 */
296	struct work_struct dead_guc_worker;
297	/**
298	 * @last_dead_guc_jiffies: timestamp of previous 'dead guc' occurrance
299	 * used to prevent a fundamentally broken system from continuously
300	 * reloading the GuC.
301	 */
302	unsigned long last_dead_guc_jiffies;
303
304#ifdef CONFIG_DRM_I915_SELFTEST
305	/**
306	 * @number_guc_id_stolen: The number of guc_ids that have been stolen
307	 */
308	int number_guc_id_stolen;
309	/**
310	 * @fast_response_selftest: Backdoor to CT handler for fast response selftest
311	 */
312	u32 fast_response_selftest;
313#endif
314};
315
316struct intel_guc_tlb_wait {
317	struct wait_queue_head wq;
318	bool busy;
319};
320
321/*
322 * GuC version number components are only 8-bit, so converting to a 32bit 8.8.8
323 * integer works.
324 */
325#define MAKE_GUC_VER(maj, min, pat)	(((maj) << 16) | ((min) << 8) | (pat))
326#define MAKE_GUC_VER_STRUCT(ver)	MAKE_GUC_VER((ver).major, (ver).minor, (ver).patch)
327#define GUC_SUBMIT_VER(guc)		MAKE_GUC_VER_STRUCT((guc)->submission_version)
328#define GUC_FIRMWARE_VER(guc)		MAKE_GUC_VER_STRUCT((guc)->fw.file_selected.ver)
329
330static inline struct intel_guc *log_to_guc(struct intel_guc_log *log)
331{
332	return container_of(log, struct intel_guc, log);
333}
334
335static
336inline int intel_guc_send(struct intel_guc *guc, const u32 *action, u32 len)
337{
338	return intel_guc_ct_send(&guc->ct, action, len, NULL, 0, 0);
339}
340
341static
342inline int intel_guc_send_nb(struct intel_guc *guc, const u32 *action, u32 len,
343			     u32 g2h_len_dw)
344{
345	return intel_guc_ct_send(&guc->ct, action, len, NULL, 0,
346				 MAKE_SEND_FLAGS(g2h_len_dw));
347}
348
349static inline int
350intel_guc_send_and_receive(struct intel_guc *guc, const u32 *action, u32 len,
351			   u32 *response_buf, u32 response_buf_size)
352{
353	return intel_guc_ct_send(&guc->ct, action, len,
354				 response_buf, response_buf_size, 0);
355}
356
357static inline int intel_guc_send_busy_loop(struct intel_guc *guc,
358					   const u32 *action,
359					   u32 len,
360					   u32 g2h_len_dw,
361					   bool loop)
362{
363	int err;
364	unsigned int sleep_period_ms = 1;
365	bool not_atomic = !in_atomic() && !irqs_disabled();
366
367	/*
368	 * FIXME: Have caller pass in if we are in an atomic context to avoid
369	 * using in_atomic(). It is likely safe here as we check for irqs
370	 * disabled which basically all the spin locks in the i915 do but
371	 * regardless this should be cleaned up.
372	 */
373
374	/* No sleeping with spin locks, just busy loop */
375	might_sleep_if(loop && not_atomic);
376
377retry:
378	err = intel_guc_send_nb(guc, action, len, g2h_len_dw);
379	if (unlikely(err == -EBUSY && loop)) {
380		if (likely(not_atomic)) {
381			if (msleep_interruptible(sleep_period_ms))
382				return -EINTR;
383			sleep_period_ms = sleep_period_ms << 1;
384		} else {
385			cpu_relax();
386		}
387		goto retry;
388	}
389
390	return err;
391}
392
393/* Only call this from the interrupt handler code */
394static inline void intel_guc_to_host_event_handler(struct intel_guc *guc)
395{
396	if (guc->interrupts.enabled)
397		intel_guc_ct_event_handler(&guc->ct);
398}
399
400/* GuC addresses above GUC_GGTT_TOP also don't map through the GTT */
401#define GUC_GGTT_TOP	0xFEE00000
402
403/**
404 * intel_guc_ggtt_offset() - Get and validate the GGTT offset of @vma
405 * @guc: intel_guc structure.
406 * @vma: i915 graphics virtual memory area.
407 *
408 * GuC does not allow any gfx GGTT address that falls into range
409 * [0, ggtt.pin_bias), which is reserved for Boot ROM, SRAM and WOPCM.
410 * Currently, in order to exclude [0, ggtt.pin_bias) address space from
411 * GGTT, all gfx objects used by GuC are allocated with intel_guc_allocate_vma()
412 * and pinned with PIN_OFFSET_BIAS along with the value of ggtt.pin_bias.
413 *
414 * Return: GGTT offset of the @vma.
415 */
416static inline u32 intel_guc_ggtt_offset(struct intel_guc *guc,
417					struct i915_vma *vma)
418{
419	u32 offset = i915_ggtt_offset(vma);
420
421	GEM_BUG_ON(offset < i915_ggtt_pin_bias(vma));
422	GEM_BUG_ON(range_overflows_t(u64, offset, vma->size, GUC_GGTT_TOP));
423
424	return offset;
425}
426
427void intel_guc_init_early(struct intel_guc *guc);
428void intel_guc_init_late(struct intel_guc *guc);
429void intel_guc_init_send_regs(struct intel_guc *guc);
430void intel_guc_write_params(struct intel_guc *guc);
431int intel_guc_init(struct intel_guc *guc);
432void intel_guc_fini(struct intel_guc *guc);
433void intel_guc_notify(struct intel_guc *guc);
434int intel_guc_send_mmio(struct intel_guc *guc, const u32 *action, u32 len,
435			u32 *response_buf, u32 response_buf_size);
436int intel_guc_to_host_process_recv_msg(struct intel_guc *guc,
437				       const u32 *payload, u32 len);
438int intel_guc_auth_huc(struct intel_guc *guc, u32 rsa_offset);
439int intel_guc_suspend(struct intel_guc *guc);
440int intel_guc_resume(struct intel_guc *guc);
441struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size);
442int intel_guc_allocate_and_map_vma(struct intel_guc *guc, u32 size,
443				   struct i915_vma **out_vma, void **out_vaddr);
444int intel_guc_self_cfg32(struct intel_guc *guc, u16 key, u32 value);
445int intel_guc_self_cfg64(struct intel_guc *guc, u16 key, u64 value);
446
447static inline bool intel_guc_is_supported(struct intel_guc *guc)
448{
449	return intel_uc_fw_is_supported(&guc->fw);
450}
451
452static inline bool intel_guc_is_wanted(struct intel_guc *guc)
453{
454	return intel_uc_fw_is_enabled(&guc->fw);
455}
456
457static inline bool intel_guc_is_used(struct intel_guc *guc)
458{
459	GEM_BUG_ON(__intel_uc_fw_status(&guc->fw) == INTEL_UC_FIRMWARE_SELECTED);
460	return intel_uc_fw_is_available(&guc->fw);
461}
462
463static inline bool intel_guc_is_fw_running(struct intel_guc *guc)
464{
465	return intel_uc_fw_is_running(&guc->fw);
466}
467
468static inline bool intel_guc_is_ready(struct intel_guc *guc)
469{
470	return intel_guc_is_fw_running(guc) && intel_guc_ct_enabled(&guc->ct);
471}
472
473static inline void intel_guc_reset_interrupts(struct intel_guc *guc)
474{
475	guc->interrupts.reset(guc);
476}
477
478static inline void intel_guc_enable_interrupts(struct intel_guc *guc)
479{
480	guc->interrupts.enable(guc);
481}
482
483static inline void intel_guc_disable_interrupts(struct intel_guc *guc)
484{
485	guc->interrupts.disable(guc);
486}
487
488static inline int intel_guc_sanitize(struct intel_guc *guc)
489{
490	intel_uc_fw_sanitize(&guc->fw);
491	intel_guc_disable_interrupts(guc);
492	intel_guc_ct_sanitize(&guc->ct);
493	guc->mmio_msg = 0;
494
495	return 0;
496}
497
498static inline void intel_guc_enable_msg(struct intel_guc *guc, u32 mask)
499{
500	spin_lock_irq(&guc->irq_lock);
501	guc->msg_enabled_mask |= mask;
502	spin_unlock_irq(&guc->irq_lock);
503}
504
505static inline void intel_guc_disable_msg(struct intel_guc *guc, u32 mask)
506{
507	spin_lock_irq(&guc->irq_lock);
508	guc->msg_enabled_mask &= ~mask;
509	spin_unlock_irq(&guc->irq_lock);
510}
511
512int intel_guc_wait_for_idle(struct intel_guc *guc, long timeout);
513
514int intel_guc_deregister_done_process_msg(struct intel_guc *guc,
515					  const u32 *msg, u32 len);
516int intel_guc_sched_done_process_msg(struct intel_guc *guc,
517				     const u32 *msg, u32 len);
518int intel_guc_context_reset_process_msg(struct intel_guc *guc,
519					const u32 *msg, u32 len);
520int intel_guc_engine_failure_process_msg(struct intel_guc *guc,
521					 const u32 *msg, u32 len);
522int intel_guc_error_capture_process_msg(struct intel_guc *guc,
523					const u32 *msg, u32 len);
524int intel_guc_crash_process_msg(struct intel_guc *guc, u32 action);
525
526struct intel_engine_cs *
527intel_guc_lookup_engine(struct intel_guc *guc, u8 guc_class, u8 instance);
528
529void intel_guc_find_hung_context(struct intel_engine_cs *engine);
530
531int intel_guc_global_policies_update(struct intel_guc *guc);
532
533void intel_guc_context_ban(struct intel_context *ce, struct i915_request *rq);
534
535void intel_guc_submission_reset_prepare(struct intel_guc *guc);
536void intel_guc_submission_reset(struct intel_guc *guc, intel_engine_mask_t stalled);
537void intel_guc_submission_reset_finish(struct intel_guc *guc);
538void intel_guc_submission_cancel_requests(struct intel_guc *guc);
539
540void intel_guc_load_status(struct intel_guc *guc, struct drm_printer *p);
541
542void intel_guc_write_barrier(struct intel_guc *guc);
543
544void intel_guc_dump_time_info(struct intel_guc *guc, struct drm_printer *p);
545
546int intel_guc_sched_disable_gucid_threshold_max(struct intel_guc *guc);
547
548bool intel_guc_tlb_invalidation_is_available(struct intel_guc *guc);
549int intel_guc_invalidate_tlb_engines(struct intel_guc *guc);
550int intel_guc_invalidate_tlb_guc(struct intel_guc *guc);
551int intel_guc_tlb_invalidation_done(struct intel_guc *guc,
552				    const u32 *payload, u32 len);
553void wake_up_all_tlb_invalidate(struct intel_guc *guc);
554#endif
555