xref: /openbmc/linux/drivers/gpu/drm/i915/gt/uc/intel_guc.h (revision 3f58ff6b)
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 
24 struct __guc_ads_blob;
25 struct 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  */
33 struct 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 	/** @sched_engine: Global engine used to submit requests to GuC */
46 	struct i915_sched_engine *sched_engine;
47 	/**
48 	 * @stalled_request: if GuC can't process a request for any reason, we
49 	 * save it until GuC restarts processing. No other request can be
50 	 * submitted until the stalled request is processed.
51 	 */
52 	struct i915_request *stalled_request;
53 	/**
54 	 * @submission_stall_reason: reason why submission is stalled
55 	 */
56 	enum {
57 		STALL_NONE,
58 		STALL_REGISTER_CONTEXT,
59 		STALL_MOVE_LRC_TAIL,
60 		STALL_ADD_REQUEST,
61 	} submission_stall_reason;
62 
63 	/* intel_guc_recv interrupt related state */
64 	/** @irq_lock: protects GuC irq state */
65 	spinlock_t irq_lock;
66 	/**
67 	 * @msg_enabled_mask: mask of events that are processed when receiving
68 	 * an INTEL_GUC_ACTION_DEFAULT G2H message.
69 	 */
70 	unsigned int msg_enabled_mask;
71 
72 	/**
73 	 * @outstanding_submission_g2h: number of outstanding GuC to Host
74 	 * responses related to GuC submission, used to determine if the GT is
75 	 * idle
76 	 */
77 	atomic_t outstanding_submission_g2h;
78 
79 	/** @interrupts: pointers to GuC interrupt-managing functions. */
80 	struct {
81 		bool enabled;
82 		void (*reset)(struct intel_guc *guc);
83 		void (*enable)(struct intel_guc *guc);
84 		void (*disable)(struct intel_guc *guc);
85 	} interrupts;
86 
87 	/**
88 	 * @submission_state: sub-structure for submission state protected by
89 	 * single lock
90 	 */
91 	struct {
92 		/**
93 		 * @lock: protects everything in submission_state,
94 		 * ce->guc_id.id, and ce->guc_id.ref when transitioning in and
95 		 * out of zero
96 		 */
97 		spinlock_t lock;
98 		/**
99 		 * @guc_ids: used to allocate new guc_ids, single-lrc
100 		 */
101 		struct ida guc_ids;
102 		/**
103 		 * @num_guc_ids: Number of guc_ids, selftest feature to be able
104 		 * to reduce this number while testing.
105 		 */
106 		int num_guc_ids;
107 		/**
108 		 * @guc_ids_bitmap: used to allocate new guc_ids, multi-lrc
109 		 */
110 		unsigned long *guc_ids_bitmap;
111 		/**
112 		 * @guc_id_list: list of intel_context with valid guc_ids but no
113 		 * refs
114 		 */
115 		struct list_head guc_id_list;
116 		/**
117 		 * @guc_ids_in_use: Number single-lrc guc_ids in use
118 		 */
119 		unsigned int guc_ids_in_use;
120 		/**
121 		 * @destroyed_contexts: list of contexts waiting to be destroyed
122 		 * (deregistered with the GuC)
123 		 */
124 		struct list_head destroyed_contexts;
125 		/**
126 		 * @destroyed_worker: worker to deregister contexts, need as we
127 		 * need to take a GT PM reference and can't from destroy
128 		 * function as it might be in an atomic context (no sleeping)
129 		 */
130 		struct work_struct destroyed_worker;
131 		/**
132 		 * @reset_fail_worker: worker to trigger a GT reset after an
133 		 * engine reset fails
134 		 */
135 		struct work_struct reset_fail_worker;
136 		/**
137 		 * @reset_fail_mask: mask of engines that failed to reset
138 		 */
139 		intel_engine_mask_t reset_fail_mask;
140 		/**
141 		 * @sched_disable_delay_ms: schedule disable delay, in ms, for
142 		 * contexts
143 		 */
144 		unsigned int sched_disable_delay_ms;
145 		/**
146 		 * @sched_disable_gucid_threshold: threshold of min remaining available
147 		 * guc_ids before we start bypassing the schedule disable delay
148 		 */
149 		unsigned int sched_disable_gucid_threshold;
150 	} submission_state;
151 
152 	/**
153 	 * @submission_supported: tracks whether we support GuC submission on
154 	 * the current platform
155 	 */
156 	bool submission_supported;
157 	/** @submission_selected: tracks whether the user enabled GuC submission */
158 	bool submission_selected;
159 	/** @submission_initialized: tracks whether GuC submission has been initialised */
160 	bool submission_initialized;
161 	/**
162 	 * @rc_supported: tracks whether we support GuC rc on the current platform
163 	 */
164 	bool rc_supported;
165 	/** @rc_selected: tracks whether the user enabled GuC rc */
166 	bool rc_selected;
167 
168 	/** @ads_vma: object allocated to hold the GuC ADS */
169 	struct i915_vma *ads_vma;
170 	/** @ads_map: contents of the GuC ADS */
171 	struct iosys_map ads_map;
172 	/** @ads_regset_size: size of the save/restore regsets in the ADS */
173 	u32 ads_regset_size;
174 	/**
175 	 * @ads_regset_count: number of save/restore registers in the ADS for
176 	 * each engine
177 	 */
178 	u32 ads_regset_count[I915_NUM_ENGINES];
179 	/** @ads_regset: save/restore regsets in the ADS */
180 	struct guc_mmio_reg *ads_regset;
181 	/** @ads_golden_ctxt_size: size of the golden contexts in the ADS */
182 	u32 ads_golden_ctxt_size;
183 	/** @ads_capture_size: size of register lists in the ADS used for error capture */
184 	u32 ads_capture_size;
185 	/** @ads_engine_usage_size: size of engine usage in the ADS */
186 	u32 ads_engine_usage_size;
187 
188 	/** @lrc_desc_pool_v69: object allocated to hold the GuC LRC descriptor pool */
189 	struct i915_vma *lrc_desc_pool_v69;
190 	/** @lrc_desc_pool_vaddr_v69: contents of the GuC LRC descriptor pool */
191 	void *lrc_desc_pool_vaddr_v69;
192 
193 	/**
194 	 * @context_lookup: used to resolve intel_context from guc_id, if a
195 	 * context is present in this structure it is registered with the GuC
196 	 */
197 	struct xarray context_lookup;
198 
199 	/** @params: Control params for fw initialization */
200 	u32 params[GUC_CTL_MAX_DWORDS];
201 
202 	/** @send_regs: GuC's FW specific registers used for sending MMIO H2G */
203 	struct {
204 		u32 base;
205 		unsigned int count;
206 		enum forcewake_domains fw_domains;
207 	} send_regs;
208 
209 	/** @notify_reg: register used to send interrupts to the GuC FW */
210 	i915_reg_t notify_reg;
211 
212 	/**
213 	 * @mmio_msg: notification bitmask that the GuC writes in one of its
214 	 * registers when the CT channel is disabled, to be processed when the
215 	 * channel is back up.
216 	 */
217 	u32 mmio_msg;
218 
219 	/** @send_mutex: used to serialize the intel_guc_send actions */
220 	struct mutex send_mutex;
221 
222 	/**
223 	 * @timestamp: GT timestamp object that stores a copy of the timestamp
224 	 * and adjusts it for overflow using a worker.
225 	 */
226 	struct {
227 		/**
228 		 * @lock: Lock protecting the below fields and the engine stats.
229 		 */
230 		spinlock_t lock;
231 
232 		/**
233 		 * @gt_stamp: 64 bit extended value of the GT timestamp.
234 		 */
235 		u64 gt_stamp;
236 
237 		/**
238 		 * @ping_delay: Period for polling the GT timestamp for
239 		 * overflow.
240 		 */
241 		unsigned long ping_delay;
242 
243 		/**
244 		 * @work: Periodic work to adjust GT timestamp, engine and
245 		 * context usage for overflows.
246 		 */
247 		struct delayed_work work;
248 
249 		/**
250 		 * @shift: Right shift value for the gpm timestamp
251 		 */
252 		u32 shift;
253 
254 		/**
255 		 * @last_stat_jiffies: jiffies at last actual stats collection time
256 		 * We use this timestamp to ensure we don't oversample the
257 		 * stats because runtime power management events can trigger
258 		 * stats collection at much higher rates than required.
259 		 */
260 		unsigned long last_stat_jiffies;
261 	} timestamp;
262 
263 #ifdef CONFIG_DRM_I915_SELFTEST
264 	/**
265 	 * @number_guc_id_stolen: The number of guc_ids that have been stolen
266 	 */
267 	int number_guc_id_stolen;
268 #endif
269 };
270 
271 static inline struct intel_guc *log_to_guc(struct intel_guc_log *log)
272 {
273 	return container_of(log, struct intel_guc, log);
274 }
275 
276 static
277 inline int intel_guc_send(struct intel_guc *guc, const u32 *action, u32 len)
278 {
279 	return intel_guc_ct_send(&guc->ct, action, len, NULL, 0, 0);
280 }
281 
282 static
283 inline int intel_guc_send_nb(struct intel_guc *guc, const u32 *action, u32 len,
284 			     u32 g2h_len_dw)
285 {
286 	return intel_guc_ct_send(&guc->ct, action, len, NULL, 0,
287 				 MAKE_SEND_FLAGS(g2h_len_dw));
288 }
289 
290 static inline int
291 intel_guc_send_and_receive(struct intel_guc *guc, const u32 *action, u32 len,
292 			   u32 *response_buf, u32 response_buf_size)
293 {
294 	return intel_guc_ct_send(&guc->ct, action, len,
295 				 response_buf, response_buf_size, 0);
296 }
297 
298 static inline int intel_guc_send_busy_loop(struct intel_guc *guc,
299 					   const u32 *action,
300 					   u32 len,
301 					   u32 g2h_len_dw,
302 					   bool loop)
303 {
304 	int err;
305 	unsigned int sleep_period_ms = 1;
306 	bool not_atomic = !in_atomic() && !irqs_disabled();
307 
308 	/*
309 	 * FIXME: Have caller pass in if we are in an atomic context to avoid
310 	 * using in_atomic(). It is likely safe here as we check for irqs
311 	 * disabled which basically all the spin locks in the i915 do but
312 	 * regardless this should be cleaned up.
313 	 */
314 
315 	/* No sleeping with spin locks, just busy loop */
316 	might_sleep_if(loop && not_atomic);
317 
318 retry:
319 	err = intel_guc_send_nb(guc, action, len, g2h_len_dw);
320 	if (unlikely(err == -EBUSY && loop)) {
321 		if (likely(not_atomic)) {
322 			if (msleep_interruptible(sleep_period_ms))
323 				return -EINTR;
324 			sleep_period_ms = sleep_period_ms << 1;
325 		} else {
326 			cpu_relax();
327 		}
328 		goto retry;
329 	}
330 
331 	return err;
332 }
333 
334 /* Only call this from the interrupt handler code */
335 static inline void intel_guc_to_host_event_handler(struct intel_guc *guc)
336 {
337 	if (guc->interrupts.enabled)
338 		intel_guc_ct_event_handler(&guc->ct);
339 }
340 
341 /* GuC addresses above GUC_GGTT_TOP also don't map through the GTT */
342 #define GUC_GGTT_TOP	0xFEE00000
343 
344 /**
345  * intel_guc_ggtt_offset() - Get and validate the GGTT offset of @vma
346  * @guc: intel_guc structure.
347  * @vma: i915 graphics virtual memory area.
348  *
349  * GuC does not allow any gfx GGTT address that falls into range
350  * [0, ggtt.pin_bias), which is reserved for Boot ROM, SRAM and WOPCM.
351  * Currently, in order to exclude [0, ggtt.pin_bias) address space from
352  * GGTT, all gfx objects used by GuC are allocated with intel_guc_allocate_vma()
353  * and pinned with PIN_OFFSET_BIAS along with the value of ggtt.pin_bias.
354  *
355  * Return: GGTT offset of the @vma.
356  */
357 static inline u32 intel_guc_ggtt_offset(struct intel_guc *guc,
358 					struct i915_vma *vma)
359 {
360 	u32 offset = i915_ggtt_offset(vma);
361 
362 	GEM_BUG_ON(offset < i915_ggtt_pin_bias(vma));
363 	GEM_BUG_ON(range_overflows_t(u64, offset, vma->size, GUC_GGTT_TOP));
364 
365 	return offset;
366 }
367 
368 void intel_guc_init_early(struct intel_guc *guc);
369 void intel_guc_init_late(struct intel_guc *guc);
370 void intel_guc_init_send_regs(struct intel_guc *guc);
371 void intel_guc_write_params(struct intel_guc *guc);
372 int intel_guc_init(struct intel_guc *guc);
373 void intel_guc_fini(struct intel_guc *guc);
374 void intel_guc_notify(struct intel_guc *guc);
375 int intel_guc_send_mmio(struct intel_guc *guc, const u32 *action, u32 len,
376 			u32 *response_buf, u32 response_buf_size);
377 int intel_guc_to_host_process_recv_msg(struct intel_guc *guc,
378 				       const u32 *payload, u32 len);
379 int intel_guc_auth_huc(struct intel_guc *guc, u32 rsa_offset);
380 int intel_guc_suspend(struct intel_guc *guc);
381 int intel_guc_resume(struct intel_guc *guc);
382 struct i915_vma *intel_guc_allocate_vma(struct intel_guc *guc, u32 size);
383 int intel_guc_allocate_and_map_vma(struct intel_guc *guc, u32 size,
384 				   struct i915_vma **out_vma, void **out_vaddr);
385 int intel_guc_self_cfg32(struct intel_guc *guc, u16 key, u32 value);
386 int intel_guc_self_cfg64(struct intel_guc *guc, u16 key, u64 value);
387 
388 static inline bool intel_guc_is_supported(struct intel_guc *guc)
389 {
390 	return intel_uc_fw_is_supported(&guc->fw);
391 }
392 
393 static inline bool intel_guc_is_wanted(struct intel_guc *guc)
394 {
395 	return intel_uc_fw_is_enabled(&guc->fw);
396 }
397 
398 static inline bool intel_guc_is_used(struct intel_guc *guc)
399 {
400 	GEM_BUG_ON(__intel_uc_fw_status(&guc->fw) == INTEL_UC_FIRMWARE_SELECTED);
401 	return intel_uc_fw_is_available(&guc->fw);
402 }
403 
404 static inline bool intel_guc_is_fw_running(struct intel_guc *guc)
405 {
406 	return intel_uc_fw_is_running(&guc->fw);
407 }
408 
409 static inline bool intel_guc_is_ready(struct intel_guc *guc)
410 {
411 	return intel_guc_is_fw_running(guc) && intel_guc_ct_enabled(&guc->ct);
412 }
413 
414 static inline void intel_guc_reset_interrupts(struct intel_guc *guc)
415 {
416 	guc->interrupts.reset(guc);
417 }
418 
419 static inline void intel_guc_enable_interrupts(struct intel_guc *guc)
420 {
421 	guc->interrupts.enable(guc);
422 }
423 
424 static inline void intel_guc_disable_interrupts(struct intel_guc *guc)
425 {
426 	guc->interrupts.disable(guc);
427 }
428 
429 static inline int intel_guc_sanitize(struct intel_guc *guc)
430 {
431 	intel_uc_fw_sanitize(&guc->fw);
432 	intel_guc_disable_interrupts(guc);
433 	intel_guc_ct_sanitize(&guc->ct);
434 	guc->mmio_msg = 0;
435 
436 	return 0;
437 }
438 
439 static inline void intel_guc_enable_msg(struct intel_guc *guc, u32 mask)
440 {
441 	spin_lock_irq(&guc->irq_lock);
442 	guc->msg_enabled_mask |= mask;
443 	spin_unlock_irq(&guc->irq_lock);
444 }
445 
446 static inline void intel_guc_disable_msg(struct intel_guc *guc, u32 mask)
447 {
448 	spin_lock_irq(&guc->irq_lock);
449 	guc->msg_enabled_mask &= ~mask;
450 	spin_unlock_irq(&guc->irq_lock);
451 }
452 
453 int intel_guc_wait_for_idle(struct intel_guc *guc, long timeout);
454 
455 int intel_guc_deregister_done_process_msg(struct intel_guc *guc,
456 					  const u32 *msg, u32 len);
457 int intel_guc_sched_done_process_msg(struct intel_guc *guc,
458 				     const u32 *msg, u32 len);
459 int intel_guc_context_reset_process_msg(struct intel_guc *guc,
460 					const u32 *msg, u32 len);
461 int intel_guc_engine_failure_process_msg(struct intel_guc *guc,
462 					 const u32 *msg, u32 len);
463 int intel_guc_error_capture_process_msg(struct intel_guc *guc,
464 					const u32 *msg, u32 len);
465 
466 struct intel_engine_cs *
467 intel_guc_lookup_engine(struct intel_guc *guc, u8 guc_class, u8 instance);
468 
469 void intel_guc_find_hung_context(struct intel_engine_cs *engine);
470 
471 int intel_guc_global_policies_update(struct intel_guc *guc);
472 
473 void intel_guc_context_ban(struct intel_context *ce, struct i915_request *rq);
474 
475 void intel_guc_submission_reset_prepare(struct intel_guc *guc);
476 void intel_guc_submission_reset(struct intel_guc *guc, intel_engine_mask_t stalled);
477 void intel_guc_submission_reset_finish(struct intel_guc *guc);
478 void intel_guc_submission_cancel_requests(struct intel_guc *guc);
479 
480 void intel_guc_load_status(struct intel_guc *guc, struct drm_printer *p);
481 
482 void intel_guc_write_barrier(struct intel_guc *guc);
483 
484 void intel_guc_dump_time_info(struct intel_guc *guc, struct drm_printer *p);
485 
486 int intel_guc_sched_disable_gucid_threshold_max(struct intel_guc *guc);
487 
488 #endif
489