1 /* SPDX-License-Identifier: MIT */
2 #ifndef _INTEL_RINGBUFFER_H_
3 #define _INTEL_RINGBUFFER_H_
4
5 #include <asm/cacheflush.h>
6 #include <drm/drm_util.h>
7 #include <drm/drm_cache.h>
8
9 #include <linux/hashtable.h>
10 #include <linux/irq_work.h>
11 #include <linux/random.h>
12 #include <linux/seqlock.h>
13
14 #include "i915_pmu.h"
15 #include "i915_request.h"
16 #include "i915_selftest.h"
17 #include "intel_engine_types.h"
18 #include "intel_gt_types.h"
19 #include "intel_timeline.h"
20 #include "intel_workarounds.h"
21
22 struct drm_printer;
23 struct intel_context;
24 struct intel_gt;
25 struct lock_class_key;
26
27 /* Early gen2 devices have a cacheline of just 32 bytes, using 64 is overkill,
28 * but keeps the logic simple. Indeed, the whole purpose of this macro is just
29 * to give some inclination as to some of the magic values used in the various
30 * workarounds!
31 */
32 #define CACHELINE_BYTES 64
33 #define CACHELINE_DWORDS (CACHELINE_BYTES / sizeof(u32))
34
35 #define ENGINE_TRACE(e, fmt, ...) do { \
36 const struct intel_engine_cs *e__ __maybe_unused = (e); \
37 GEM_TRACE("%s %s: " fmt, \
38 dev_name(e__->i915->drm.dev), e__->name, \
39 ##__VA_ARGS__); \
40 } while (0)
41
42 /*
43 * The register defines to be used with the following macros need to accept a
44 * base param, e.g:
45 *
46 * REG_FOO(base) _MMIO((base) + <relative offset>)
47 * ENGINE_READ(engine, REG_FOO);
48 *
49 * register arrays are to be defined and accessed as follows:
50 *
51 * REG_BAR(base, i) _MMIO((base) + <relative offset> + (i) * <shift>)
52 * ENGINE_READ_IDX(engine, REG_BAR, i)
53 */
54
55 #define __ENGINE_REG_OP(op__, engine__, ...) \
56 intel_uncore_##op__((engine__)->uncore, __VA_ARGS__)
57
58 #define __ENGINE_READ_OP(op__, engine__, reg__) \
59 __ENGINE_REG_OP(op__, (engine__), reg__((engine__)->mmio_base))
60
61 #define ENGINE_READ16(...) __ENGINE_READ_OP(read16, __VA_ARGS__)
62 #define ENGINE_READ(...) __ENGINE_READ_OP(read, __VA_ARGS__)
63 #define ENGINE_READ_FW(...) __ENGINE_READ_OP(read_fw, __VA_ARGS__)
64 #define ENGINE_POSTING_READ(...) __ENGINE_READ_OP(posting_read_fw, __VA_ARGS__)
65 #define ENGINE_POSTING_READ16(...) __ENGINE_READ_OP(posting_read16, __VA_ARGS__)
66
67 #define ENGINE_READ64(engine__, lower_reg__, upper_reg__) \
68 __ENGINE_REG_OP(read64_2x32, (engine__), \
69 lower_reg__((engine__)->mmio_base), \
70 upper_reg__((engine__)->mmio_base))
71
72 #define ENGINE_READ_IDX(engine__, reg__, idx__) \
73 __ENGINE_REG_OP(read, (engine__), reg__((engine__)->mmio_base, (idx__)))
74
75 #define __ENGINE_WRITE_OP(op__, engine__, reg__, val__) \
76 __ENGINE_REG_OP(op__, (engine__), reg__((engine__)->mmio_base), (val__))
77
78 #define ENGINE_WRITE16(...) __ENGINE_WRITE_OP(write16, __VA_ARGS__)
79 #define ENGINE_WRITE(...) __ENGINE_WRITE_OP(write, __VA_ARGS__)
80 #define ENGINE_WRITE_FW(...) __ENGINE_WRITE_OP(write_fw, __VA_ARGS__)
81
82 #define __HAS_ENGINE(engine_mask, id) ((engine_mask) & BIT(id))
83 #define HAS_ENGINE(gt, id) __HAS_ENGINE((gt)->info.engine_mask, id)
84
85 #define __ENGINE_INSTANCES_MASK(mask, first, count) ({ \
86 unsigned int first__ = (first); \
87 unsigned int count__ = (count); \
88 ((mask) & GENMASK(first__ + count__ - 1, first__)) >> first__; \
89 })
90
91 #define ENGINE_INSTANCES_MASK(gt, first, count) \
92 __ENGINE_INSTANCES_MASK((gt)->info.engine_mask, first, count)
93
94 #define RCS_MASK(gt) \
95 ENGINE_INSTANCES_MASK(gt, RCS0, I915_MAX_RCS)
96 #define BCS_MASK(gt) \
97 ENGINE_INSTANCES_MASK(gt, BCS0, I915_MAX_BCS)
98 #define VDBOX_MASK(gt) \
99 ENGINE_INSTANCES_MASK(gt, VCS0, I915_MAX_VCS)
100 #define VEBOX_MASK(gt) \
101 ENGINE_INSTANCES_MASK(gt, VECS0, I915_MAX_VECS)
102 #define CCS_MASK(gt) \
103 ENGINE_INSTANCES_MASK(gt, CCS0, I915_MAX_CCS)
104
105 #define GEN6_RING_FAULT_REG_READ(engine__) \
106 intel_uncore_read((engine__)->uncore, RING_FAULT_REG(engine__))
107
108 #define GEN6_RING_FAULT_REG_POSTING_READ(engine__) \
109 intel_uncore_posting_read((engine__)->uncore, RING_FAULT_REG(engine__))
110
111 #define GEN6_RING_FAULT_REG_RMW(engine__, clear__, set__) \
112 ({ \
113 u32 __val; \
114 \
115 __val = intel_uncore_read((engine__)->uncore, \
116 RING_FAULT_REG(engine__)); \
117 __val &= ~(clear__); \
118 __val |= (set__); \
119 intel_uncore_write((engine__)->uncore, RING_FAULT_REG(engine__), \
120 __val); \
121 })
122
123 /* seqno size is actually only a uint32, but since we plan to use MI_FLUSH_DW to
124 * do the writes, and that must have qw aligned offsets, simply pretend it's 8b.
125 */
126
127 static inline unsigned int
execlists_num_ports(const struct intel_engine_execlists * const execlists)128 execlists_num_ports(const struct intel_engine_execlists * const execlists)
129 {
130 return execlists->port_mask + 1;
131 }
132
133 static inline struct i915_request *
execlists_active(const struct intel_engine_execlists * execlists)134 execlists_active(const struct intel_engine_execlists *execlists)
135 {
136 struct i915_request * const *cur, * const *old, *active;
137
138 cur = READ_ONCE(execlists->active);
139 smp_rmb(); /* pairs with overwrite protection in process_csb() */
140 do {
141 old = cur;
142
143 active = READ_ONCE(*cur);
144 cur = READ_ONCE(execlists->active);
145
146 smp_rmb(); /* and complete the seqlock retry */
147 } while (unlikely(cur != old));
148
149 return active;
150 }
151
152 static inline u32
intel_read_status_page(const struct intel_engine_cs * engine,int reg)153 intel_read_status_page(const struct intel_engine_cs *engine, int reg)
154 {
155 /* Ensure that the compiler doesn't optimize away the load. */
156 return READ_ONCE(engine->status_page.addr[reg]);
157 }
158
159 static inline void
intel_write_status_page(struct intel_engine_cs * engine,int reg,u32 value)160 intel_write_status_page(struct intel_engine_cs *engine, int reg, u32 value)
161 {
162 /* Writing into the status page should be done sparingly. Since
163 * we do when we are uncertain of the device state, we take a bit
164 * of extra paranoia to try and ensure that the HWS takes the value
165 * we give and that it doesn't end up trapped inside the CPU!
166 */
167 drm_clflush_virt_range(&engine->status_page.addr[reg], sizeof(value));
168 WRITE_ONCE(engine->status_page.addr[reg], value);
169 drm_clflush_virt_range(&engine->status_page.addr[reg], sizeof(value));
170 }
171
172 /*
173 * Reads a dword out of the status page, which is written to from the command
174 * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
175 * MI_STORE_DATA_IMM.
176 *
177 * The following dwords have a reserved meaning:
178 * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
179 * 0x04: ring 0 head pointer
180 * 0x05: ring 1 head pointer (915-class)
181 * 0x06: ring 2 head pointer (915-class)
182 * 0x10-0x1b: Context status DWords (GM45)
183 * 0x1f: Last written status offset. (GM45)
184 * 0x20-0x2f: Reserved (Gen6+)
185 *
186 * The area from dword 0x30 to 0x3ff is available for driver usage.
187 */
188 #define I915_GEM_HWS_PREEMPT 0x32
189 #define I915_GEM_HWS_PREEMPT_ADDR (I915_GEM_HWS_PREEMPT * sizeof(u32))
190 #define I915_GEM_HWS_SEQNO 0x40
191 #define I915_GEM_HWS_SEQNO_ADDR (I915_GEM_HWS_SEQNO * sizeof(u32))
192 #define I915_GEM_HWS_MIGRATE (0x42 * sizeof(u32))
193 #define I915_GEM_HWS_GGTT_BIND 0x46
194 #define I915_GEM_HWS_GGTT_BIND_ADDR (I915_GEM_HWS_GGTT_BIND * sizeof(u32))
195 #define I915_GEM_HWS_PXP 0x60
196 #define I915_GEM_HWS_PXP_ADDR (I915_GEM_HWS_PXP * sizeof(u32))
197 #define I915_GEM_HWS_GSC 0x62
198 #define I915_GEM_HWS_GSC_ADDR (I915_GEM_HWS_GSC * sizeof(u32))
199 #define I915_GEM_HWS_SCRATCH 0x80
200
201 #define I915_HWS_CSB_BUF0_INDEX 0x10
202 #define I915_HWS_CSB_WRITE_INDEX 0x1f
203 #define ICL_HWS_CSB_WRITE_INDEX 0x2f
204 #define INTEL_HWS_CSB_WRITE_INDEX(__i915) \
205 (GRAPHICS_VER(__i915) >= 11 ? ICL_HWS_CSB_WRITE_INDEX : I915_HWS_CSB_WRITE_INDEX)
206
207 void intel_engine_stop(struct intel_engine_cs *engine);
208 void intel_engine_cleanup(struct intel_engine_cs *engine);
209
210 int intel_engines_init_mmio(struct intel_gt *gt);
211 int intel_engines_init(struct intel_gt *gt);
212
213 void intel_engine_free_request_pool(struct intel_engine_cs *engine);
214
215 void intel_engines_release(struct intel_gt *gt);
216 void intel_engines_free(struct intel_gt *gt);
217
218 int intel_engine_init_common(struct intel_engine_cs *engine);
219 void intel_engine_cleanup_common(struct intel_engine_cs *engine);
220
221 int intel_engine_resume(struct intel_engine_cs *engine);
222
223 int intel_ring_submission_setup(struct intel_engine_cs *engine);
224
225 int intel_engine_stop_cs(struct intel_engine_cs *engine);
226 void intel_engine_cancel_stop_cs(struct intel_engine_cs *engine);
227
228 void intel_engine_wait_for_pending_mi_fw(struct intel_engine_cs *engine);
229
230 void intel_engine_set_hwsp_writemask(struct intel_engine_cs *engine, u32 mask);
231
232 u64 intel_engine_get_active_head(const struct intel_engine_cs *engine);
233 u64 intel_engine_get_last_batch_head(const struct intel_engine_cs *engine);
234
235 void intel_engine_get_instdone(const struct intel_engine_cs *engine,
236 struct intel_instdone *instdone);
237
238 void intel_engine_init_execlists(struct intel_engine_cs *engine);
239
240 bool intel_engine_irq_enable(struct intel_engine_cs *engine);
241 void intel_engine_irq_disable(struct intel_engine_cs *engine);
242
__intel_engine_reset(struct intel_engine_cs * engine,bool stalled)243 static inline void __intel_engine_reset(struct intel_engine_cs *engine,
244 bool stalled)
245 {
246 if (engine->reset.rewind)
247 engine->reset.rewind(engine, stalled);
248 engine->serial++; /* contexts lost */
249 }
250
251 bool intel_engines_are_idle(struct intel_gt *gt);
252 bool intel_engine_is_idle(struct intel_engine_cs *engine);
253
254 void __intel_engine_flush_submission(struct intel_engine_cs *engine, bool sync);
intel_engine_flush_submission(struct intel_engine_cs * engine)255 static inline void intel_engine_flush_submission(struct intel_engine_cs *engine)
256 {
257 __intel_engine_flush_submission(engine, true);
258 }
259
260 void intel_engines_reset_default_submission(struct intel_gt *gt);
261
262 bool intel_engine_can_store_dword(struct intel_engine_cs *engine);
263
264 __printf(3, 4)
265 void intel_engine_dump(struct intel_engine_cs *engine,
266 struct drm_printer *m,
267 const char *header, ...);
268 void intel_engine_dump_active_requests(struct list_head *requests,
269 struct i915_request *hung_rq,
270 struct drm_printer *m);
271
272 ktime_t intel_engine_get_busy_time(struct intel_engine_cs *engine,
273 ktime_t *now);
274
275 void intel_engine_get_hung_entity(struct intel_engine_cs *engine,
276 struct intel_context **ce, struct i915_request **rq);
277
278 u32 intel_engine_context_size(struct intel_gt *gt, u8 class);
279 struct intel_context *
280 intel_engine_create_pinned_context(struct intel_engine_cs *engine,
281 struct i915_address_space *vm,
282 unsigned int ring_size,
283 unsigned int hwsp,
284 struct lock_class_key *key,
285 const char *name);
286
287 void intel_engine_destroy_pinned_context(struct intel_context *ce);
288
289 void xehp_enable_ccs_engines(struct intel_engine_cs *engine);
290
291 #define ENGINE_PHYSICAL 0
292 #define ENGINE_MOCK 1
293 #define ENGINE_VIRTUAL 2
294
intel_engine_uses_guc(const struct intel_engine_cs * engine)295 static inline bool intel_engine_uses_guc(const struct intel_engine_cs *engine)
296 {
297 return engine->gt->submission_method >= INTEL_SUBMISSION_GUC;
298 }
299
300 static inline bool
intel_engine_has_preempt_reset(const struct intel_engine_cs * engine)301 intel_engine_has_preempt_reset(const struct intel_engine_cs *engine)
302 {
303 if (!CONFIG_DRM_I915_PREEMPT_TIMEOUT)
304 return false;
305
306 return intel_engine_has_preemption(engine);
307 }
308
309 #define FORCE_VIRTUAL BIT(0)
310 struct intel_context *
311 intel_engine_create_virtual(struct intel_engine_cs **siblings,
312 unsigned int count, unsigned long flags);
313
314 static inline struct intel_context *
intel_engine_create_parallel(struct intel_engine_cs ** engines,unsigned int num_engines,unsigned int width)315 intel_engine_create_parallel(struct intel_engine_cs **engines,
316 unsigned int num_engines,
317 unsigned int width)
318 {
319 GEM_BUG_ON(!engines[0]->cops->create_parallel);
320 return engines[0]->cops->create_parallel(engines, num_engines, width);
321 }
322
323 static inline bool
intel_virtual_engine_has_heartbeat(const struct intel_engine_cs * engine)324 intel_virtual_engine_has_heartbeat(const struct intel_engine_cs *engine)
325 {
326 /*
327 * For non-GuC submission we expect the back-end to look at the
328 * heartbeat status of the actual physical engine that the work
329 * has been (or is being) scheduled on, so we should only reach
330 * here with GuC submission enabled.
331 */
332 GEM_BUG_ON(!intel_engine_uses_guc(engine));
333
334 return intel_guc_virtual_engine_has_heartbeat(engine);
335 }
336
337 static inline bool
intel_engine_has_heartbeat(const struct intel_engine_cs * engine)338 intel_engine_has_heartbeat(const struct intel_engine_cs *engine)
339 {
340 if (!CONFIG_DRM_I915_HEARTBEAT_INTERVAL)
341 return false;
342
343 if (intel_engine_is_virtual(engine))
344 return intel_virtual_engine_has_heartbeat(engine);
345 else
346 return READ_ONCE(engine->props.heartbeat_interval_ms);
347 }
348
349 static inline struct intel_engine_cs *
intel_engine_get_sibling(struct intel_engine_cs * engine,unsigned int sibling)350 intel_engine_get_sibling(struct intel_engine_cs *engine, unsigned int sibling)
351 {
352 GEM_BUG_ON(!intel_engine_is_virtual(engine));
353 return engine->cops->get_sibling(engine, sibling);
354 }
355
356 static inline void
intel_engine_set_hung_context(struct intel_engine_cs * engine,struct intel_context * ce)357 intel_engine_set_hung_context(struct intel_engine_cs *engine,
358 struct intel_context *ce)
359 {
360 engine->hung_ce = ce;
361 }
362
363 static inline void
intel_engine_clear_hung_context(struct intel_engine_cs * engine)364 intel_engine_clear_hung_context(struct intel_engine_cs *engine)
365 {
366 intel_engine_set_hung_context(engine, NULL);
367 }
368
369 static inline struct intel_context *
intel_engine_get_hung_context(struct intel_engine_cs * engine)370 intel_engine_get_hung_context(struct intel_engine_cs *engine)
371 {
372 return engine->hung_ce;
373 }
374
375 u64 intel_clamp_heartbeat_interval_ms(struct intel_engine_cs *engine, u64 value);
376 u64 intel_clamp_max_busywait_duration_ns(struct intel_engine_cs *engine, u64 value);
377 u64 intel_clamp_preempt_timeout_ms(struct intel_engine_cs *engine, u64 value);
378 u64 intel_clamp_stop_timeout_ms(struct intel_engine_cs *engine, u64 value);
379 u64 intel_clamp_timeslice_duration_ms(struct intel_engine_cs *engine, u64 value);
380
381 #define rb_to_uabi_engine(rb) \
382 rb_entry_safe(rb, struct intel_engine_cs, uabi_node)
383
384 #define for_each_uabi_engine(engine__, i915__) \
385 for ((engine__) = rb_to_uabi_engine(rb_first(&(i915__)->uabi_engines));\
386 (engine__); \
387 (engine__) = rb_to_uabi_engine(rb_next(&(engine__)->uabi_node)))
388
389 #endif /* _INTEL_RINGBUFFER_H_ */
390