File: | dev/pci/drm/i915/gt/intel_engine_cs.c |
Warning: | line 1329, column 2 Value stored to 'x' is never read |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
1 | /* |
2 | * Copyright © 2016 Intel Corporation |
3 | * |
4 | * Permission is hereby granted, free of charge, to any person obtaining a |
5 | * copy of this software and associated documentation files (the "Software"), |
6 | * to deal in the Software without restriction, including without limitation |
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
8 | * and/or sell copies of the Software, and to permit persons to whom the |
9 | * Software is furnished to do so, subject to the following conditions: |
10 | * |
11 | * The above copyright notice and this permission notice (including the next |
12 | * paragraph) shall be included in all copies or substantial portions of the |
13 | * Software. |
14 | * |
15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
16 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
17 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
18 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
19 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING |
20 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS |
21 | * IN THE SOFTWARE. |
22 | * |
23 | */ |
24 | |
25 | #include <drm/drm_print.h> |
26 | |
27 | #include "gem/i915_gem_context.h" |
28 | |
29 | #include "i915_drv.h" |
30 | |
31 | #include "intel_breadcrumbs.h" |
32 | #include "intel_context.h" |
33 | #include "intel_engine.h" |
34 | #include "intel_engine_pm.h" |
35 | #include "intel_engine_user.h" |
36 | #include "intel_gt.h" |
37 | #include "intel_gt_requests.h" |
38 | #include "intel_gt_pm.h" |
39 | #include "intel_lrc.h" |
40 | #include "intel_reset.h" |
41 | #include "intel_ring.h" |
42 | |
43 | /* Haswell does have the CXT_SIZE register however it does not appear to be |
44 | * valid. Now, docs explain in dwords what is in the context object. The full |
45 | * size is 70720 bytes, however, the power context and execlist context will |
46 | * never be saved (power context is stored elsewhere, and execlists don't work |
47 | * on HSW) - so the final size, including the extra state required for the |
48 | * Resource Streamer, is 66944 bytes, which rounds to 17 pages. |
49 | */ |
50 | #define HSW_CXT_TOTAL_SIZE(17 * (1 << 12)) (17 * PAGE_SIZE(1 << 12)) |
51 | |
52 | #define DEFAULT_LR_CONTEXT_RENDER_SIZE(22 * (1 << 12)) (22 * PAGE_SIZE(1 << 12)) |
53 | #define GEN8_LR_CONTEXT_RENDER_SIZE(20 * (1 << 12)) (20 * PAGE_SIZE(1 << 12)) |
54 | #define GEN9_LR_CONTEXT_RENDER_SIZE(22 * (1 << 12)) (22 * PAGE_SIZE(1 << 12)) |
55 | #define GEN10_LR_CONTEXT_RENDER_SIZE(18 * (1 << 12)) (18 * PAGE_SIZE(1 << 12)) |
56 | #define GEN11_LR_CONTEXT_RENDER_SIZE(14 * (1 << 12)) (14 * PAGE_SIZE(1 << 12)) |
57 | |
58 | #define GEN8_LR_CONTEXT_OTHER_SIZE( 2 * (1 << 12)) ( 2 * PAGE_SIZE(1 << 12)) |
59 | |
60 | #define MAX_MMIO_BASES3 3 |
61 | struct engine_info { |
62 | unsigned int hw_id; |
63 | u8 class; |
64 | u8 instance; |
65 | /* mmio bases table *must* be sorted in reverse gen order */ |
66 | struct engine_mmio_base { |
67 | u32 gen : 8; |
68 | u32 base : 24; |
69 | } mmio_bases[MAX_MMIO_BASES3]; |
70 | }; |
71 | |
72 | static const struct engine_info intel_engines[] = { |
73 | [RCS0] = { |
74 | .hw_id = RCS0_HW0, |
75 | .class = RENDER_CLASS0, |
76 | .instance = 0, |
77 | .mmio_bases = { |
78 | { .gen = 1, .base = RENDER_RING_BASE0x02000 } |
79 | }, |
80 | }, |
81 | [BCS0] = { |
82 | .hw_id = BCS0_HW2, |
83 | .class = COPY_ENGINE_CLASS3, |
84 | .instance = 0, |
85 | .mmio_bases = { |
86 | { .gen = 6, .base = BLT_RING_BASE0x22000 } |
87 | }, |
88 | }, |
89 | [VCS0] = { |
90 | .hw_id = VCS0_HW1, |
91 | .class = VIDEO_DECODE_CLASS1, |
92 | .instance = 0, |
93 | .mmio_bases = { |
94 | { .gen = 11, .base = GEN11_BSD_RING_BASE0x1c0000 }, |
95 | { .gen = 6, .base = GEN6_BSD_RING_BASE0x12000 }, |
96 | { .gen = 4, .base = BSD_RING_BASE0x04000 } |
97 | }, |
98 | }, |
99 | [VCS1] = { |
100 | .hw_id = VCS1_HW4, |
101 | .class = VIDEO_DECODE_CLASS1, |
102 | .instance = 1, |
103 | .mmio_bases = { |
104 | { .gen = 11, .base = GEN11_BSD2_RING_BASE0x1c4000 }, |
105 | { .gen = 8, .base = GEN8_BSD2_RING_BASE0x1c000 } |
106 | }, |
107 | }, |
108 | [VCS2] = { |
109 | .hw_id = VCS2_HW6, |
110 | .class = VIDEO_DECODE_CLASS1, |
111 | .instance = 2, |
112 | .mmio_bases = { |
113 | { .gen = 11, .base = GEN11_BSD3_RING_BASE0x1d0000 } |
114 | }, |
115 | }, |
116 | [VCS3] = { |
117 | .hw_id = VCS3_HW7, |
118 | .class = VIDEO_DECODE_CLASS1, |
119 | .instance = 3, |
120 | .mmio_bases = { |
121 | { .gen = 11, .base = GEN11_BSD4_RING_BASE0x1d4000 } |
122 | }, |
123 | }, |
124 | [VECS0] = { |
125 | .hw_id = VECS0_HW3, |
126 | .class = VIDEO_ENHANCEMENT_CLASS2, |
127 | .instance = 0, |
128 | .mmio_bases = { |
129 | { .gen = 11, .base = GEN11_VEBOX_RING_BASE0x1c8000 }, |
130 | { .gen = 7, .base = VEBOX_RING_BASE0x1a000 } |
131 | }, |
132 | }, |
133 | [VECS1] = { |
134 | .hw_id = VECS1_HW12, |
135 | .class = VIDEO_ENHANCEMENT_CLASS2, |
136 | .instance = 1, |
137 | .mmio_bases = { |
138 | { .gen = 11, .base = GEN11_VEBOX2_RING_BASE0x1d8000 } |
139 | }, |
140 | }, |
141 | }; |
142 | |
143 | /** |
144 | * intel_engine_context_size() - return the size of the context for an engine |
145 | * @gt: the gt |
146 | * @class: engine class |
147 | * |
148 | * Each engine class may require a different amount of space for a context |
149 | * image. |
150 | * |
151 | * Return: size (in bytes) of an engine class specific context image |
152 | * |
153 | * Note: this size includes the HWSP, which is part of the context image |
154 | * in LRC mode, but does not include the "shared data page" used with |
155 | * GuC submission. The caller should account for this if using the GuC. |
156 | */ |
157 | u32 intel_engine_context_size(struct intel_gt *gt, u8 class) |
158 | { |
159 | struct intel_uncore *uncore = gt->uncore; |
160 | u32 cxt_size; |
161 | |
162 | BUILD_BUG_ON(I915_GTT_PAGE_SIZE != PAGE_SIZE)extern char _ctassert[(!((1ULL << (12)) != (1 << 12 ))) ? 1 : -1 ] __attribute__((__unused__)); |
163 | |
164 | switch (class) { |
165 | case RENDER_CLASS0: |
166 | switch (INTEL_GEN(gt->i915)((&(gt->i915)->__info)->gen)) { |
167 | default: |
168 | MISSING_CASE(INTEL_GEN(gt->i915))({ int __ret = !!(1); if (__ret) printf("Missing case (%s == %ld)\n" , "((&(gt->i915)->__info)->gen)", (long)(((& (gt->i915)->__info)->gen))); __builtin_expect(!!(__ret ), 0); }); |
169 | return DEFAULT_LR_CONTEXT_RENDER_SIZE(22 * (1 << 12)); |
170 | case 12: |
171 | case 11: |
172 | return GEN11_LR_CONTEXT_RENDER_SIZE(14 * (1 << 12)); |
173 | case 10: |
174 | return GEN10_LR_CONTEXT_RENDER_SIZE(18 * (1 << 12)); |
175 | case 9: |
176 | return GEN9_LR_CONTEXT_RENDER_SIZE(22 * (1 << 12)); |
177 | case 8: |
178 | return GEN8_LR_CONTEXT_RENDER_SIZE(20 * (1 << 12)); |
179 | case 7: |
180 | if (IS_HASWELL(gt->i915)IS_PLATFORM(gt->i915, INTEL_HASWELL)) |
181 | return HSW_CXT_TOTAL_SIZE(17 * (1 << 12)); |
182 | |
183 | cxt_size = intel_uncore_read(uncore, GEN7_CXT_SIZE((const i915_reg_t){ .reg = (0x21a8) })); |
184 | return round_up(GEN7_CXT_TOTAL_SIZE(cxt_size) * 64,((((((((cxt_size) >> 9) & 0x7f) + (((cxt_size) >> 0) & 0x3f)) * 64) + (((1 << 12)) - 1)) / ((1 << 12))) * ((1 << 12))) |
185 | PAGE_SIZE)((((((((cxt_size) >> 9) & 0x7f) + (((cxt_size) >> 0) & 0x3f)) * 64) + (((1 << 12)) - 1)) / ((1 << 12))) * ((1 << 12))); |
186 | case 6: |
187 | cxt_size = intel_uncore_read(uncore, CXT_SIZE((const i915_reg_t){ .reg = (0x21a0) })); |
188 | return round_up(GEN6_CXT_TOTAL_SIZE(cxt_size) * 64,((((((((cxt_size) >> 18) & 0x3f) + (((cxt_size) >> 6) & 0x3f) + (((cxt_size) >> 0) & 0x3f)) * 64) + (((1 << 12)) - 1)) / ((1 << 12))) * ((1 << 12))) |
189 | PAGE_SIZE)((((((((cxt_size) >> 18) & 0x3f) + (((cxt_size) >> 6) & 0x3f) + (((cxt_size) >> 0) & 0x3f)) * 64) + (((1 << 12)) - 1)) / ((1 << 12))) * ((1 << 12))); |
190 | case 5: |
191 | case 4: |
192 | /* |
193 | * There is a discrepancy here between the size reported |
194 | * by the register and the size of the context layout |
195 | * in the docs. Both are described as authorative! |
196 | * |
197 | * The discrepancy is on the order of a few cachelines, |
198 | * but the total is under one page (4k), which is our |
199 | * minimum allocation anyway so it should all come |
200 | * out in the wash. |
201 | */ |
202 | cxt_size = intel_uncore_read(uncore, CXT_SIZE((const i915_reg_t){ .reg = (0x21a0) })) + 1; |
203 | drm_dbg(>->i915->drm,drm_dev_dbg((>->i915->drm)->dev, DRM_UT_DRIVER , "gen%d CXT_SIZE = %d bytes [0x%08x]\n", ((&(gt->i915 )->__info)->gen), cxt_size * 64, cxt_size - 1) |
204 | "gen%d CXT_SIZE = %d bytes [0x%08x]\n",drm_dev_dbg((>->i915->drm)->dev, DRM_UT_DRIVER , "gen%d CXT_SIZE = %d bytes [0x%08x]\n", ((&(gt->i915 )->__info)->gen), cxt_size * 64, cxt_size - 1) |
205 | INTEL_GEN(gt->i915), cxt_size * 64,drm_dev_dbg((>->i915->drm)->dev, DRM_UT_DRIVER , "gen%d CXT_SIZE = %d bytes [0x%08x]\n", ((&(gt->i915 )->__info)->gen), cxt_size * 64, cxt_size - 1) |
206 | cxt_size - 1)drm_dev_dbg((>->i915->drm)->dev, DRM_UT_DRIVER , "gen%d CXT_SIZE = %d bytes [0x%08x]\n", ((&(gt->i915 )->__info)->gen), cxt_size * 64, cxt_size - 1); |
207 | return round_up(cxt_size * 64, PAGE_SIZE)((((cxt_size * 64) + (((1 << 12)) - 1)) / ((1 << 12 ))) * ((1 << 12))); |
208 | case 3: |
209 | case 2: |
210 | /* For the special day when i810 gets merged. */ |
211 | case 1: |
212 | return 0; |
213 | } |
214 | break; |
215 | default: |
216 | MISSING_CASE(class)({ int __ret = !!(1); if (__ret) printf("Missing case (%s == %ld)\n" , "class", (long)(class)); __builtin_expect(!!(__ret), 0); }); |
217 | fallthroughdo {} while (0); |
218 | case VIDEO_DECODE_CLASS1: |
219 | case VIDEO_ENHANCEMENT_CLASS2: |
220 | case COPY_ENGINE_CLASS3: |
221 | if (INTEL_GEN(gt->i915)((&(gt->i915)->__info)->gen) < 8) |
222 | return 0; |
223 | return GEN8_LR_CONTEXT_OTHER_SIZE( 2 * (1 << 12)); |
224 | } |
225 | } |
226 | |
227 | static u32 __engine_mmio_base(struct drm_i915_privateinteldrm_softc *i915, |
228 | const struct engine_mmio_base *bases) |
229 | { |
230 | int i; |
231 | |
232 | for (i = 0; i < MAX_MMIO_BASES3; i++) |
233 | if (INTEL_GEN(i915)((&(i915)->__info)->gen) >= bases[i].gen) |
234 | break; |
235 | |
236 | GEM_BUG_ON(i == MAX_MMIO_BASES)((void)0); |
237 | GEM_BUG_ON(!bases[i].base)((void)0); |
238 | |
239 | return bases[i].base; |
240 | } |
241 | |
242 | static void __sprint_engine_name(struct intel_engine_cs *engine) |
243 | { |
244 | /* |
245 | * Before we know what the uABI name for this engine will be, |
246 | * we still would like to keep track of this engine in the debug logs. |
247 | * We throw in a ' here as a reminder that this isn't its final name. |
248 | */ |
249 | GEM_WARN_ON(snprintf(engine->name, sizeof(engine->name), "%s'%u",({ __builtin_expect(!!(!!(snprintf(engine->name, sizeof(engine ->name), "%s'%u", intel_engine_class_repr(engine->class ), engine->instance) >= sizeof(engine->name))), 0); } ) |
250 | intel_engine_class_repr(engine->class),({ __builtin_expect(!!(!!(snprintf(engine->name, sizeof(engine ->name), "%s'%u", intel_engine_class_repr(engine->class ), engine->instance) >= sizeof(engine->name))), 0); } ) |
251 | engine->instance) >= sizeof(engine->name))({ __builtin_expect(!!(!!(snprintf(engine->name, sizeof(engine ->name), "%s'%u", intel_engine_class_repr(engine->class ), engine->instance) >= sizeof(engine->name))), 0); } ); |
252 | } |
253 | |
254 | void intel_engine_set_hwsp_writemask(struct intel_engine_cs *engine, u32 mask) |
255 | { |
256 | /* |
257 | * Though they added more rings on g4x/ilk, they did not add |
258 | * per-engine HWSTAM until gen6. |
259 | */ |
260 | if (INTEL_GEN(engine->i915)((&(engine->i915)->__info)->gen) < 6 && engine->class != RENDER_CLASS0) |
261 | return; |
262 | |
263 | if (INTEL_GEN(engine->i915)((&(engine->i915)->__info)->gen) >= 3) |
264 | ENGINE_WRITE(engine, RING_HWSTAM, mask)intel_uncore_write(((engine))->uncore, ((const i915_reg_t) { .reg = (((engine)->mmio_base) + 0x98) }), (mask)); |
265 | else |
266 | ENGINE_WRITE16(engine, RING_HWSTAM, mask)intel_uncore_write16(((engine))->uncore, ((const i915_reg_t ){ .reg = (((engine)->mmio_base) + 0x98) }), (mask)); |
267 | } |
268 | |
269 | static void intel_engine_sanitize_mmio(struct intel_engine_cs *engine) |
270 | { |
271 | /* Mask off all writes into the unknown HWSP */ |
272 | intel_engine_set_hwsp_writemask(engine, ~0u); |
273 | } |
274 | |
275 | static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id) |
276 | { |
277 | const struct engine_info *info = &intel_engines[id]; |
278 | struct drm_i915_privateinteldrm_softc *i915 = gt->i915; |
279 | struct intel_engine_cs *engine; |
280 | |
281 | BUILD_BUG_ON(MAX_ENGINE_CLASS >= BIT(GEN11_ENGINE_CLASS_WIDTH))extern char _ctassert[(!(4 >= (1UL << (3)))) ? 1 : - 1 ] __attribute__((__unused__)); |
282 | BUILD_BUG_ON(MAX_ENGINE_INSTANCE >= BIT(GEN11_ENGINE_INSTANCE_WIDTH))extern char _ctassert[(!(3 >= (1UL << (6)))) ? 1 : - 1 ] __attribute__((__unused__)); |
283 | |
284 | if (GEM_DEBUG_WARN_ON(id >= ARRAY_SIZE(gt->engine))({ ((void)0); 0; })) |
285 | return -EINVAL22; |
286 | |
287 | if (GEM_DEBUG_WARN_ON(info->class > MAX_ENGINE_CLASS)({ ((void)0); 0; })) |
288 | return -EINVAL22; |
289 | |
290 | if (GEM_DEBUG_WARN_ON(info->instance > MAX_ENGINE_INSTANCE)({ ((void)0); 0; })) |
291 | return -EINVAL22; |
292 | |
293 | if (GEM_DEBUG_WARN_ON(gt->engine_class[info->class][info->instance])({ ((void)0); 0; })) |
294 | return -EINVAL22; |
295 | |
296 | engine = kzalloc(sizeof(*engine), GFP_KERNEL(0x0001 | 0x0004)); |
297 | if (!engine) |
298 | return -ENOMEM12; |
299 | |
300 | BUILD_BUG_ON(BITS_PER_TYPE(engine->mask) < I915_NUM_ENGINES)extern char _ctassert[(!((8 * sizeof(engine->mask)) < I915_NUM_ENGINES )) ? 1 : -1 ] __attribute__((__unused__)); |
301 | |
302 | engine->id = id; |
303 | engine->legacy_idx = INVALID_ENGINE((enum intel_engine_id)-1); |
304 | engine->mask = BIT(id)(1UL << (id)); |
305 | engine->i915 = i915; |
306 | engine->gt = gt; |
307 | engine->uncore = gt->uncore; |
308 | engine->hw_id = engine->guc_id = info->hw_id; |
309 | engine->mmio_base = __engine_mmio_base(i915, info->mmio_bases); |
310 | |
311 | engine->class = info->class; |
312 | engine->instance = info->instance; |
313 | __sprint_engine_name(engine); |
314 | |
315 | engine->props.heartbeat_interval_ms = |
316 | CONFIG_DRM_I915_HEARTBEAT_INTERVAL2500; |
317 | engine->props.max_busywait_duration_ns = |
318 | CONFIG_DRM_I915_MAX_REQUEST_BUSYWAIT8000; |
319 | engine->props.preempt_timeout_ms = |
320 | CONFIG_DRM_I915_PREEMPT_TIMEOUT640; |
321 | engine->props.stop_timeout_ms = |
322 | CONFIG_DRM_I915_STOP_TIMEOUT100; |
323 | engine->props.timeslice_duration_ms = |
324 | CONFIG_DRM_I915_TIMESLICE_DURATION1; |
325 | |
326 | /* Override to uninterruptible for OpenCL workloads. */ |
327 | if (INTEL_GEN(i915)((&(i915)->__info)->gen) == 12 && engine->class == RENDER_CLASS0) |
328 | engine->props.preempt_timeout_ms = 0; |
329 | |
330 | engine->defaults = engine->props; /* never to change again */ |
331 | |
332 | engine->context_size = intel_engine_context_size(gt, engine->class); |
333 | if (WARN_ON(engine->context_size > BIT(20))({ int __ret = !!((engine->context_size > (1UL << (20)))); if (__ret) printf("%s", "WARN_ON(" "engine->context_size > (1UL << (20))" ")"); __builtin_expect(!!(__ret), 0); })) |
334 | engine->context_size = 0; |
335 | if (engine->context_size) |
336 | DRIVER_CAPS(i915)(&(i915)->caps)->has_logical_contexts = true1; |
337 | |
338 | /* Nothing to do here, execute in order of dependencies */ |
339 | engine->schedule = NULL((void *)0); |
340 | |
341 | ewma__engine_latency_init(&engine->latency); |
342 | seqlock_init(&engine->stats.lock, IPL_TTY0x9); |
343 | |
344 | ATOMIC_INIT_NOTIFIER_HEAD(&engine->context_status_notifier); |
345 | |
346 | /* Scrub mmio state on takeover */ |
347 | intel_engine_sanitize_mmio(engine); |
348 | |
349 | gt->engine_class[info->class][info->instance] = engine; |
350 | gt->engine[id] = engine; |
351 | |
352 | return 0; |
353 | } |
354 | |
355 | static void __setup_engine_capabilities(struct intel_engine_cs *engine) |
356 | { |
357 | struct drm_i915_privateinteldrm_softc *i915 = engine->i915; |
358 | |
359 | if (engine->class == VIDEO_DECODE_CLASS1) { |
360 | /* |
361 | * HEVC support is present on first engine instance |
362 | * before Gen11 and on all instances afterwards. |
363 | */ |
364 | if (INTEL_GEN(i915)((&(i915)->__info)->gen) >= 11 || |
365 | (INTEL_GEN(i915)((&(i915)->__info)->gen) >= 9 && engine->instance == 0)) |
366 | engine->uabi_capabilities |= |
367 | I915_VIDEO_CLASS_CAPABILITY_HEVC(1 << 0); |
368 | |
369 | /* |
370 | * SFC block is present only on even logical engine |
371 | * instances. |
372 | */ |
373 | if ((INTEL_GEN(i915)((&(i915)->__info)->gen) >= 11 && |
374 | (engine->gt->info.vdbox_sfc_access & |
375 | BIT(engine->instance)(1UL << (engine->instance)))) || |
376 | (INTEL_GEN(i915)((&(i915)->__info)->gen) >= 9 && engine->instance == 0)) |
377 | engine->uabi_capabilities |= |
378 | I915_VIDEO_AND_ENHANCE_CLASS_CAPABILITY_SFC(1 << 1); |
379 | } else if (engine->class == VIDEO_ENHANCEMENT_CLASS2) { |
380 | if (INTEL_GEN(i915)((&(i915)->__info)->gen) >= 9) |
381 | engine->uabi_capabilities |= |
382 | I915_VIDEO_AND_ENHANCE_CLASS_CAPABILITY_SFC(1 << 1); |
383 | } |
384 | } |
385 | |
386 | static void intel_setup_engine_capabilities(struct intel_gt *gt) |
387 | { |
388 | struct intel_engine_cs *engine; |
389 | enum intel_engine_id id; |
390 | |
391 | for_each_engine(engine, gt, id)for ((id) = 0; (id) < I915_NUM_ENGINES; (id)++) if (!((engine ) = (gt)->engine[(id)])) {} else |
392 | __setup_engine_capabilities(engine); |
393 | } |
394 | |
395 | /** |
396 | * intel_engines_release() - free the resources allocated for Command Streamers |
397 | * @gt: pointer to struct intel_gt |
398 | */ |
399 | void intel_engines_release(struct intel_gt *gt) |
400 | { |
401 | struct intel_engine_cs *engine; |
402 | enum intel_engine_id id; |
403 | |
404 | /* |
405 | * Before we release the resources held by engine, we must be certain |
406 | * that the HW is no longer accessing them -- having the GPU scribble |
407 | * to or read from a page being used for something else causes no end |
408 | * of fun. |
409 | * |
410 | * The GPU should be reset by this point, but assume the worst just |
411 | * in case we aborted before completely initialising the engines. |
412 | */ |
413 | GEM_BUG_ON(intel_gt_pm_is_awake(gt))((void)0); |
414 | if (!INTEL_INFO(gt->i915)(&(gt->i915)->__info)->gpu_reset_clobbers_display) |
415 | __intel_gt_reset(gt, ALL_ENGINES((intel_engine_mask_t)~0ul)); |
416 | |
417 | /* Decouple the backend; but keep the layout for late GPU resets */ |
418 | for_each_engine(engine, gt, id)for ((id) = 0; (id) < I915_NUM_ENGINES; (id)++) if (!((engine ) = (gt)->engine[(id)])) {} else { |
419 | if (!engine->release) |
420 | continue; |
421 | |
422 | intel_wakeref_wait_for_idle(&engine->wakeref); |
423 | GEM_BUG_ON(intel_engine_pm_is_awake(engine))((void)0); |
424 | |
425 | engine->release(engine); |
426 | engine->release = NULL((void *)0); |
427 | |
428 | memset(&engine->reset, 0, sizeof(engine->reset))__builtin_memset((&engine->reset), (0), (sizeof(engine ->reset))); |
429 | } |
430 | } |
431 | |
432 | void intel_engine_free_request_pool(struct intel_engine_cs *engine) |
433 | { |
434 | if (!engine->request_pool) |
435 | return; |
436 | |
437 | #ifdef __linux__ |
438 | kmem_cache_free(i915_request_slab_cache(), engine->request_pool); |
439 | #else |
440 | pool_put(i915_request_slab_cache(), engine->request_pool); |
441 | #endif |
442 | } |
443 | |
444 | void intel_engines_free(struct intel_gt *gt) |
445 | { |
446 | struct intel_engine_cs *engine; |
447 | enum intel_engine_id id; |
448 | |
449 | /* Free the requests! dma-resv keeps fences around for an eternity */ |
450 | rcu_barrier()__asm volatile("" : : : "memory"); |
451 | |
452 | for_each_engine(engine, gt, id)for ((id) = 0; (id) < I915_NUM_ENGINES; (id)++) if (!((engine ) = (gt)->engine[(id)])) {} else { |
453 | intel_engine_free_request_pool(engine); |
454 | kfree(engine); |
455 | gt->engine[id] = NULL((void *)0); |
456 | } |
457 | } |
458 | |
459 | /* |
460 | * Determine which engines are fused off in our particular hardware. |
461 | * Note that we have a catch-22 situation where we need to be able to access |
462 | * the blitter forcewake domain to read the engine fuses, but at the same time |
463 | * we need to know which engines are available on the system to know which |
464 | * forcewake domains are present. We solve this by intializing the forcewake |
465 | * domains based on the full engine mask in the platform capabilities before |
466 | * calling this function and pruning the domains for fused-off engines |
467 | * afterwards. |
468 | */ |
469 | static intel_engine_mask_t init_engine_mask(struct intel_gt *gt) |
470 | { |
471 | struct drm_i915_privateinteldrm_softc *i915 = gt->i915; |
472 | struct intel_gt_info *info = >->info; |
473 | struct intel_uncore *uncore = gt->uncore; |
474 | unsigned int logical_vdbox = 0; |
475 | unsigned int i; |
476 | u32 media_fuse; |
477 | u16 vdbox_mask; |
478 | u16 vebox_mask; |
479 | |
480 | info->engine_mask = INTEL_INFO(i915)(&(i915)->__info)->platform_engine_mask; |
481 | |
482 | if (INTEL_GEN(i915)((&(i915)->__info)->gen) < 11) |
483 | return info->engine_mask; |
484 | |
485 | media_fuse = ~intel_uncore_read(uncore, GEN11_GT_VEBOX_VDBOX_DISABLE((const i915_reg_t){ .reg = (0x9140) })); |
486 | |
487 | vdbox_mask = media_fuse & GEN11_GT_VDBOX_DISABLE_MASK0xff; |
488 | vebox_mask = (media_fuse & GEN11_GT_VEBOX_DISABLE_MASK(0x0f << 16)) >> |
489 | GEN11_GT_VEBOX_DISABLE_SHIFT16; |
490 | |
491 | for (i = 0; i < I915_MAX_VCS4; i++) { |
492 | if (!HAS_ENGINE(gt, _VCS(i))(((gt)->info.engine_mask) & (1UL << ((VCS0 + (i) ))))) { |
493 | vdbox_mask &= ~BIT(i)(1UL << (i)); |
494 | continue; |
495 | } |
496 | |
497 | if (!(BIT(i)(1UL << (i)) & vdbox_mask)) { |
498 | info->engine_mask &= ~BIT(_VCS(i))(1UL << ((VCS0 + (i)))); |
499 | drm_dbg(&i915->drm, "vcs%u fused off\n", i)drm_dev_dbg((&i915->drm)->dev, DRM_UT_DRIVER, "vcs%u fused off\n" , i); |
500 | continue; |
501 | } |
502 | |
503 | /* |
504 | * In Gen11, only even numbered logical VDBOXes are |
505 | * hooked up to an SFC (Scaler & Format Converter) unit. |
506 | * In TGL each VDBOX has access to an SFC. |
507 | */ |
508 | if (INTEL_GEN(i915)((&(i915)->__info)->gen) >= 12 || logical_vdbox++ % 2 == 0) |
509 | gt->info.vdbox_sfc_access |= BIT(i)(1UL << (i)); |
510 | } |
511 | drm_dbg(&i915->drm, "vdbox enable: %04x, instances: %04lx\n",drm_dev_dbg((&i915->drm)->dev, DRM_UT_DRIVER, "vdbox enable: %04x, instances: %04lx\n" , vdbox_mask, ({ unsigned int first__ = (VCS0); unsigned int count__ = (4); ((gt)->info.engine_mask & (((~0UL) >> (64 - (first__ + count__ - 1) - 1)) & ((~0UL) << (first__ )))) >> first__; })) |
512 | vdbox_mask, VDBOX_MASK(gt))drm_dev_dbg((&i915->drm)->dev, DRM_UT_DRIVER, "vdbox enable: %04x, instances: %04lx\n" , vdbox_mask, ({ unsigned int first__ = (VCS0); unsigned int count__ = (4); ((gt)->info.engine_mask & (((~0UL) >> (64 - (first__ + count__ - 1) - 1)) & ((~0UL) << (first__ )))) >> first__; })); |
513 | GEM_BUG_ON(vdbox_mask != VDBOX_MASK(gt))((void)0); |
514 | |
515 | for (i = 0; i < I915_MAX_VECS2; i++) { |
516 | if (!HAS_ENGINE(gt, _VECS(i))(((gt)->info.engine_mask) & (1UL << ((VECS0 + (i )))))) { |
517 | vebox_mask &= ~BIT(i)(1UL << (i)); |
518 | continue; |
519 | } |
520 | |
521 | if (!(BIT(i)(1UL << (i)) & vebox_mask)) { |
522 | info->engine_mask &= ~BIT(_VECS(i))(1UL << ((VECS0 + (i)))); |
523 | drm_dbg(&i915->drm, "vecs%u fused off\n", i)drm_dev_dbg((&i915->drm)->dev, DRM_UT_DRIVER, "vecs%u fused off\n" , i); |
524 | } |
525 | } |
526 | drm_dbg(&i915->drm, "vebox enable: %04x, instances: %04lx\n",drm_dev_dbg((&i915->drm)->dev, DRM_UT_DRIVER, "vebox enable: %04x, instances: %04lx\n" , vebox_mask, ({ unsigned int first__ = (VECS0); unsigned int count__ = (2); ((gt)->info.engine_mask & (((~0UL) >> (64 - (first__ + count__ - 1) - 1)) & ((~0UL) << ( first__)))) >> first__; })) |
527 | vebox_mask, VEBOX_MASK(gt))drm_dev_dbg((&i915->drm)->dev, DRM_UT_DRIVER, "vebox enable: %04x, instances: %04lx\n" , vebox_mask, ({ unsigned int first__ = (VECS0); unsigned int count__ = (2); ((gt)->info.engine_mask & (((~0UL) >> (64 - (first__ + count__ - 1) - 1)) & ((~0UL) << ( first__)))) >> first__; })); |
528 | GEM_BUG_ON(vebox_mask != VEBOX_MASK(gt))((void)0); |
529 | |
530 | return info->engine_mask; |
531 | } |
532 | |
533 | /** |
534 | * intel_engines_init_mmio() - allocate and prepare the Engine Command Streamers |
535 | * @gt: pointer to struct intel_gt |
536 | * |
537 | * Return: non-zero if the initialization failed. |
538 | */ |
539 | int intel_engines_init_mmio(struct intel_gt *gt) |
540 | { |
541 | struct drm_i915_privateinteldrm_softc *i915 = gt->i915; |
542 | const unsigned int engine_mask = init_engine_mask(gt); |
543 | unsigned int mask = 0; |
544 | unsigned int i; |
545 | int err; |
546 | |
547 | drm_WARN_ON(&i915->drm, engine_mask == 0)({ int __ret = !!((engine_mask == 0)); if (__ret) printf("%s %s: " "%s", dev_driver_string(((&i915->drm))->dev), "", "drm_WARN_ON(" "engine_mask == 0" ")"); __builtin_expect(!!(__ret), 0); }); |
548 | drm_WARN_ON(&i915->drm, engine_mask &({ int __ret = !!((engine_mask & (((~0UL) >> (64 - ( (8 * sizeof(mask)) - 1) - 1)) & ((~0UL) << (I915_NUM_ENGINES ))))); if (__ret) printf("%s %s: " "%s", dev_driver_string((( &i915->drm))->dev), "", "drm_WARN_ON(" "engine_mask & (((~0UL) >> (64 - ((8 * sizeof(mask)) - 1) - 1)) & ((~0UL) << (I915_NUM_ENGINES)))" ")"); __builtin_expect(!!(__ret), 0); }) |
549 | GENMASK(BITS_PER_TYPE(mask) - 1, I915_NUM_ENGINES))({ int __ret = !!((engine_mask & (((~0UL) >> (64 - ( (8 * sizeof(mask)) - 1) - 1)) & ((~0UL) << (I915_NUM_ENGINES ))))); if (__ret) printf("%s %s: " "%s", dev_driver_string((( &i915->drm))->dev), "", "drm_WARN_ON(" "engine_mask & (((~0UL) >> (64 - ((8 * sizeof(mask)) - 1) - 1)) & ((~0UL) << (I915_NUM_ENGINES)))" ")"); __builtin_expect(!!(__ret), 0); }); |
550 | |
551 | if (i915_inject_probe_failure(i915)({ ((void)0); 0; })) |
552 | return -ENODEV19; |
553 | |
554 | for (i = 0; i < ARRAY_SIZE(intel_engines)(sizeof((intel_engines)) / sizeof((intel_engines)[0])); i++) { |
555 | if (!HAS_ENGINE(gt, i)(((gt)->info.engine_mask) & (1UL << (i)))) |
556 | continue; |
557 | |
558 | err = intel_engine_setup(gt, i); |
559 | if (err) |
560 | goto cleanup; |
561 | |
562 | mask |= BIT(i)(1UL << (i)); |
563 | } |
564 | |
565 | /* |
566 | * Catch failures to update intel_engines table when the new engines |
567 | * are added to the driver by a warning and disabling the forgotten |
568 | * engines. |
569 | */ |
570 | if (drm_WARN_ON(&i915->drm, mask != engine_mask)({ int __ret = !!((mask != engine_mask)); if (__ret) printf("%s %s: " "%s", dev_driver_string(((&i915->drm))->dev), "", "drm_WARN_ON(" "mask != engine_mask" ")"); __builtin_expect(!!(__ret), 0); } )) |
571 | gt->info.engine_mask = mask; |
572 | |
573 | gt->info.num_engines = hweight32(mask); |
574 | |
575 | intel_gt_check_and_clear_faults(gt); |
576 | |
577 | intel_setup_engine_capabilities(gt); |
578 | |
579 | intel_uncore_prune_engine_fw_domains(gt->uncore, gt); |
580 | |
581 | return 0; |
582 | |
583 | cleanup: |
584 | intel_engines_free(gt); |
585 | return err; |
586 | } |
587 | |
588 | void intel_engine_init_execlists(struct intel_engine_cs *engine) |
589 | { |
590 | struct intel_engine_execlists * const execlists = &engine->execlists; |
591 | |
592 | execlists->port_mask = 1; |
593 | GEM_BUG_ON(!is_power_of_2(execlists_num_ports(execlists)))((void)0); |
594 | GEM_BUG_ON(execlists_num_ports(execlists) > EXECLIST_MAX_PORTS)((void)0); |
595 | |
596 | memset(execlists->pending, 0, sizeof(execlists->pending))__builtin_memset((execlists->pending), (0), (sizeof(execlists ->pending))); |
597 | execlists->active = |
598 | memset(execlists->inflight, 0, sizeof(execlists->inflight))__builtin_memset((execlists->inflight), (0), (sizeof(execlists ->inflight))); |
599 | |
600 | execlists->queue_priority_hint = INT_MIN(-0x7fffffff-1); |
601 | execlists->queue = RB_ROOT_CACHED(struct rb_root_cached) { ((void *)0) }; |
602 | } |
603 | |
604 | static void cleanup_status_page(struct intel_engine_cs *engine) |
605 | { |
606 | struct i915_vma *vma; |
607 | |
608 | /* Prevent writes into HWSP after returning the page to the system */ |
609 | intel_engine_set_hwsp_writemask(engine, ~0u); |
610 | |
611 | vma = fetch_and_zero(&engine->status_page.vma)({ typeof(*&engine->status_page.vma) __T = *(&engine ->status_page.vma); *(&engine->status_page.vma) = ( typeof(*&engine->status_page.vma))0; __T; }); |
612 | if (!vma) |
613 | return; |
614 | |
615 | if (!HWS_NEEDS_PHYSICAL(engine->i915)((&(engine->i915)->__info)->hws_needs_physical)) |
616 | i915_vma_unpin(vma); |
617 | |
618 | i915_gem_object_unpin_map(vma->obj); |
619 | i915_gem_object_put(vma->obj); |
620 | } |
621 | |
622 | static int pin_ggtt_status_page(struct intel_engine_cs *engine, |
623 | struct i915_vma *vma) |
624 | { |
625 | unsigned int flags; |
626 | |
627 | if (!HAS_LLC(engine->i915)((&(engine->i915)->__info)->has_llc) && i915_ggtt_has_aperture(engine->gt->ggtt)) |
628 | /* |
629 | * On g33, we cannot place HWS above 256MiB, so |
630 | * restrict its pinning to the low mappable arena. |
631 | * Though this restriction is not documented for |
632 | * gen4, gen5, or byt, they also behave similarly |
633 | * and hang if the HWS is placed at the top of the |
634 | * GTT. To generalise, it appears that all !llc |
635 | * platforms have issues with us placing the HWS |
636 | * above the mappable region (even though we never |
637 | * actually map it). |
638 | */ |
639 | flags = PIN_MAPPABLE(1ULL << (3)); |
640 | else |
641 | flags = PIN_HIGH(1ULL << (5)); |
642 | |
643 | return i915_ggtt_pin(vma, NULL((void *)0), 0, flags); |
644 | } |
645 | |
646 | static int init_status_page(struct intel_engine_cs *engine) |
647 | { |
648 | struct drm_i915_gem_object *obj; |
649 | struct i915_vma *vma; |
650 | void *vaddr; |
651 | int ret; |
652 | |
653 | /* |
654 | * Though the HWS register does support 36bit addresses, historically |
655 | * we have had hangs and corruption reported due to wild writes if |
656 | * the HWS is placed above 4G. We only allow objects to be allocated |
657 | * in GFP_DMA32 for i965, and no earlier physical address users had |
658 | * access to more than 4G. |
659 | */ |
660 | obj = i915_gem_object_create_internal(engine->i915, PAGE_SIZE(1 << 12)); |
661 | if (IS_ERR(obj)) { |
662 | drm_err(&engine->i915->drm,printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "Failed to allocate status page\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__) |
663 | "Failed to allocate status page\n")printf("drm:pid%d:%s *ERROR* " "[drm] " "*ERROR* " "Failed to allocate status page\n" , ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r" (__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self))); __ci;})->ci_curproc->p_p->ps_pid, __func__); |
664 | return PTR_ERR(obj); |
665 | } |
666 | |
667 | i915_gem_object_set_cache_coherency(obj, I915_CACHE_LLC); |
668 | |
669 | vma = i915_vma_instance(obj, &engine->gt->ggtt->vm, NULL((void *)0)); |
670 | if (IS_ERR(vma)) { |
671 | ret = PTR_ERR(vma); |
672 | goto err; |
673 | } |
674 | |
675 | vaddr = i915_gem_object_pin_map(obj, I915_MAP_WB); |
676 | if (IS_ERR(vaddr)) { |
677 | ret = PTR_ERR(vaddr); |
678 | goto err; |
679 | } |
680 | |
681 | engine->status_page.addr = memset(vaddr, 0, PAGE_SIZE)__builtin_memset((vaddr), (0), ((1 << 12))); |
682 | engine->status_page.vma = vma; |
683 | |
684 | if (!HWS_NEEDS_PHYSICAL(engine->i915)((&(engine->i915)->__info)->hws_needs_physical)) { |
685 | ret = pin_ggtt_status_page(engine, vma); |
686 | if (ret) |
687 | goto err_unpin; |
688 | } |
689 | |
690 | return 0; |
691 | |
692 | err_unpin: |
693 | i915_gem_object_unpin_map(obj); |
694 | err: |
695 | i915_gem_object_put(obj); |
696 | return ret; |
697 | } |
698 | |
699 | static int engine_setup_common(struct intel_engine_cs *engine) |
700 | { |
701 | int err; |
702 | |
703 | init_llist_head(&engine->barrier_tasks); |
704 | |
705 | err = init_status_page(engine); |
706 | if (err) |
707 | return err; |
708 | |
709 | engine->breadcrumbs = intel_breadcrumbs_create(engine); |
710 | if (!engine->breadcrumbs) { |
711 | err = -ENOMEM12; |
712 | goto err_status; |
713 | } |
714 | |
715 | err = intel_engine_init_cmd_parser(engine); |
716 | if (err) |
717 | goto err_cmd_parser; |
718 | |
719 | intel_engine_init_active(engine, ENGINE_PHYSICAL0); |
720 | intel_engine_init_execlists(engine); |
721 | intel_engine_init__pm(engine); |
722 | intel_engine_init_retire(engine); |
723 | |
724 | /* Use the whole device by default */ |
725 | engine->sseu = |
726 | intel_sseu_from_device_info(&engine->gt->info.sseu); |
727 | |
728 | intel_engine_init_workarounds(engine); |
729 | intel_engine_init_whitelist(engine); |
730 | intel_engine_init_ctx_wa(engine); |
731 | |
732 | return 0; |
733 | |
734 | err_cmd_parser: |
735 | intel_breadcrumbs_free(engine->breadcrumbs); |
736 | err_status: |
737 | cleanup_status_page(engine); |
738 | return err; |
739 | } |
740 | |
741 | struct measure_breadcrumb { |
742 | struct i915_request rq; |
743 | struct intel_ring ring; |
744 | u32 cs[2048]; |
745 | }; |
746 | |
747 | static int measure_breadcrumb_dw(struct intel_context *ce) |
748 | { |
749 | struct intel_engine_cs *engine = ce->engine; |
750 | struct measure_breadcrumb *frame; |
751 | int dw; |
752 | |
753 | GEM_BUG_ON(!engine->gt->scratch)((void)0); |
754 | |
755 | frame = kzalloc(sizeof(*frame), GFP_KERNEL(0x0001 | 0x0004)); |
756 | if (!frame) |
757 | return -ENOMEM12; |
758 | |
759 | frame->rq.engine = engine; |
760 | frame->rq.context = ce; |
761 | rcu_assign_pointer(frame->rq.timeline, ce->timeline)do { (frame->rq.timeline) = (ce->timeline); } while(0); |
762 | |
763 | frame->ring.vaddr = frame->cs; |
764 | frame->ring.size = sizeof(frame->cs); |
765 | frame->ring.wrap = |
766 | BITS_PER_TYPE(frame->ring.size)(8 * sizeof(frame->ring.size)) - ilog2(frame->ring.size)((sizeof(frame->ring.size) <= 4) ? (fls(frame->ring. size) - 1) : (flsl(frame->ring.size) - 1)); |
767 | frame->ring.effective_size = frame->ring.size; |
768 | intel_ring_update_space(&frame->ring); |
769 | frame->rq.ring = &frame->ring; |
770 | |
771 | mutex_lock(&ce->timeline->mutex)rw_enter_write(&ce->timeline->mutex); |
772 | spin_lock_irq(&engine->active.lock)mtx_enter(&engine->active.lock); |
773 | |
774 | dw = engine->emit_fini_breadcrumb(&frame->rq, frame->cs) - frame->cs; |
775 | |
776 | spin_unlock_irq(&engine->active.lock)mtx_leave(&engine->active.lock); |
777 | mutex_unlock(&ce->timeline->mutex)rw_exit_write(&ce->timeline->mutex); |
778 | |
779 | GEM_BUG_ON(dw & 1)((void)0); /* RING_TAIL must be qword aligned */ |
780 | |
781 | kfree(frame); |
782 | return dw; |
783 | } |
784 | |
785 | void |
786 | intel_engine_init_active(struct intel_engine_cs *engine, unsigned int subclass) |
787 | { |
788 | INIT_LIST_HEAD(&engine->active.requests); |
789 | INIT_LIST_HEAD(&engine->active.hold); |
790 | |
791 | mtx_init(&engine->active.lock, IPL_TTY)do { (void)(((void *)0)); (void)(0); __mtx_init((&engine-> active.lock), ((((0x9)) > 0x0 && ((0x9)) < 0x9) ? 0x9 : ((0x9)))); } while (0); |
792 | lockdep_set_subclass(&engine->active.lock, subclass); |
793 | |
794 | /* |
795 | * Due to an interesting quirk in lockdep's internal debug tracking, |
796 | * after setting a subclass we must ensure the lock is used. Otherwise, |
797 | * nr_unused_locks is incremented once too often. |
798 | */ |
799 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
800 | local_irq_disable()intr_disable(); |
801 | lock_map_acquire(&engine->active.lock.dep_map); |
802 | lock_map_release(&engine->active.lock.dep_map); |
803 | local_irq_enable()intr_enable(); |
804 | #endif |
805 | } |
806 | |
807 | static struct intel_context * |
808 | create_pinned_context(struct intel_engine_cs *engine, |
809 | unsigned int hwsp, |
810 | struct lock_class_key *key, |
811 | const char *name) |
812 | { |
813 | struct intel_context *ce; |
814 | int err; |
815 | |
816 | ce = intel_context_create(engine); |
817 | if (IS_ERR(ce)) |
818 | return ce; |
819 | |
820 | __set_bit(CONTEXT_BARRIER_BIT0, &ce->flags); |
821 | ce->timeline = page_pack_bits(NULL, hwsp)({ unsigned long __bits = (hwsp); ((void)0); ((typeof(((void * )0)))((unsigned long)(((void *)0)) | __bits)); }); |
822 | |
823 | err = intel_context_pin(ce); /* perma-pin so it is always available */ |
824 | if (err) { |
825 | intel_context_put(ce); |
826 | return ERR_PTR(err); |
827 | } |
828 | |
829 | /* |
830 | * Give our perma-pinned kernel timelines a separate lockdep class, |
831 | * so that we can use them from within the normal user timelines |
832 | * should we need to inject GPU operations during their request |
833 | * construction. |
834 | */ |
835 | lockdep_set_class_and_name(&ce->timeline->mutex, key, name); |
836 | |
837 | return ce; |
838 | } |
839 | |
840 | static struct intel_context * |
841 | create_kernel_context(struct intel_engine_cs *engine) |
842 | { |
843 | static struct lock_class_key kernel; |
844 | |
845 | return create_pinned_context(engine, I915_GEM_HWS_SEQNO_ADDR(0x40 * sizeof(u32)), |
846 | &kernel, "kernel_context"); |
847 | } |
848 | |
849 | /** |
850 | * intel_engines_init_common - initialize cengine state which might require hw access |
851 | * @engine: Engine to initialize. |
852 | * |
853 | * Initializes @engine@ structure members shared between legacy and execlists |
854 | * submission modes which do require hardware access. |
855 | * |
856 | * Typcally done at later stages of submission mode specific engine setup. |
857 | * |
858 | * Returns zero on success or an error code on failure. |
859 | */ |
860 | static int engine_init_common(struct intel_engine_cs *engine) |
861 | { |
862 | struct intel_context *ce; |
863 | int ret; |
864 | |
865 | engine->set_default_submission(engine); |
866 | |
867 | /* |
868 | * We may need to do things with the shrinker which |
869 | * require us to immediately switch back to the default |
870 | * context. This can cause a problem as pinning the |
871 | * default context also requires GTT space which may not |
872 | * be available. To avoid this we always pin the default |
873 | * context. |
874 | */ |
875 | ce = create_kernel_context(engine); |
876 | if (IS_ERR(ce)) |
877 | return PTR_ERR(ce); |
878 | |
879 | ret = measure_breadcrumb_dw(ce); |
880 | if (ret < 0) |
881 | goto err_context; |
882 | |
883 | engine->emit_fini_breadcrumb_dw = ret; |
884 | engine->kernel_context = ce; |
885 | |
886 | return 0; |
887 | |
888 | err_context: |
889 | intel_context_put(ce); |
890 | return ret; |
891 | } |
892 | |
893 | int intel_engines_init(struct intel_gt *gt) |
894 | { |
895 | int (*setup)(struct intel_engine_cs *engine); |
896 | struct intel_engine_cs *engine; |
897 | enum intel_engine_id id; |
898 | int err; |
899 | |
900 | if (HAS_EXECLISTS(gt->i915)((&(gt->i915)->__info)->has_logical_ring_contexts )) |
901 | setup = intel_execlists_submission_setup; |
902 | else |
903 | setup = intel_ring_submission_setup; |
904 | |
905 | for_each_engine(engine, gt, id)for ((id) = 0; (id) < I915_NUM_ENGINES; (id)++) if (!((engine ) = (gt)->engine[(id)])) {} else { |
906 | err = engine_setup_common(engine); |
907 | if (err) |
908 | return err; |
909 | |
910 | err = setup(engine); |
911 | if (err) |
912 | return err; |
913 | |
914 | err = engine_init_common(engine); |
915 | if (err) |
916 | return err; |
917 | |
918 | intel_engine_add_user(engine); |
919 | } |
920 | |
921 | return 0; |
922 | } |
923 | |
924 | /** |
925 | * intel_engines_cleanup_common - cleans up the engine state created by |
926 | * the common initiailizers. |
927 | * @engine: Engine to cleanup. |
928 | * |
929 | * This cleans up everything created by the common helpers. |
930 | */ |
931 | void intel_engine_cleanup_common(struct intel_engine_cs *engine) |
932 | { |
933 | GEM_BUG_ON(!list_empty(&engine->active.requests))((void)0); |
934 | tasklet_kill(&engine->execlists.tasklet); /* flush the callback */ |
935 | |
936 | cleanup_status_page(engine); |
937 | intel_breadcrumbs_free(engine->breadcrumbs); |
938 | |
939 | intel_engine_fini_retire(engine); |
940 | intel_engine_cleanup_cmd_parser(engine); |
941 | |
942 | if (engine->default_state) |
943 | uao_detach(engine->default_state); |
944 | |
945 | if (engine->kernel_context) { |
946 | intel_context_unpin(engine->kernel_context); |
947 | intel_context_put(engine->kernel_context); |
948 | } |
949 | GEM_BUG_ON(!llist_empty(&engine->barrier_tasks))((void)0); |
950 | |
951 | intel_wa_list_free(&engine->ctx_wa_list); |
952 | intel_wa_list_free(&engine->wa_list); |
953 | intel_wa_list_free(&engine->whitelist); |
954 | } |
955 | |
956 | /** |
957 | * intel_engine_resume - re-initializes the HW state of the engine |
958 | * @engine: Engine to resume. |
959 | * |
960 | * Returns zero on success or an error code on failure. |
961 | */ |
962 | int intel_engine_resume(struct intel_engine_cs *engine) |
963 | { |
964 | intel_engine_apply_workarounds(engine); |
965 | intel_engine_apply_whitelist(engine); |
966 | |
967 | return engine->resume(engine); |
968 | } |
969 | |
970 | u64 intel_engine_get_active_head(const struct intel_engine_cs *engine) |
971 | { |
972 | struct drm_i915_privateinteldrm_softc *i915 = engine->i915; |
973 | |
974 | u64 acthd; |
975 | |
976 | if (INTEL_GEN(i915)((&(i915)->__info)->gen) >= 8) |
977 | acthd = ENGINE_READ64(engine, RING_ACTHD, RING_ACTHD_UDW)intel_uncore_read64_2x32(((engine))->uncore, ((const i915_reg_t ){ .reg = (((engine)->mmio_base) + 0x74) }), ((const i915_reg_t ){ .reg = (((engine)->mmio_base) + 0x5c) })); |
978 | else if (INTEL_GEN(i915)((&(i915)->__info)->gen) >= 4) |
979 | acthd = ENGINE_READ(engine, RING_ACTHD)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x74) })); |
980 | else |
981 | acthd = ENGINE_READ(engine, ACTHD)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0xc8) })); |
982 | |
983 | return acthd; |
984 | } |
985 | |
986 | u64 intel_engine_get_last_batch_head(const struct intel_engine_cs *engine) |
987 | { |
988 | u64 bbaddr; |
989 | |
990 | if (INTEL_GEN(engine->i915)((&(engine->i915)->__info)->gen) >= 8) |
991 | bbaddr = ENGINE_READ64(engine, RING_BBADDR, RING_BBADDR_UDW)intel_uncore_read64_2x32(((engine))->uncore, ((const i915_reg_t ){ .reg = (((engine)->mmio_base) + 0x140) }), ((const i915_reg_t ){ .reg = (((engine)->mmio_base) + 0x168) })); |
992 | else |
993 | bbaddr = ENGINE_READ(engine, RING_BBADDR)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x140) })); |
994 | |
995 | return bbaddr; |
996 | } |
997 | |
998 | static unsigned long stop_timeout(const struct intel_engine_cs *engine) |
999 | { |
1000 | if (in_atomic()0 || irqs_disabled()) /* inside atomic preempt-reset? */ |
1001 | return 0; |
1002 | |
1003 | /* |
1004 | * If we are doing a normal GPU reset, we can take our time and allow |
1005 | * the engine to quiesce. We've stopped submission to the engine, and |
1006 | * if we wait long enough an innocent context should complete and |
1007 | * leave the engine idle. So they should not be caught unaware by |
1008 | * the forthcoming GPU reset (which usually follows the stop_cs)! |
1009 | */ |
1010 | return READ_ONCE(engine->props.stop_timeout_ms)({ typeof(engine->props.stop_timeout_ms) __tmp = *(volatile typeof(engine->props.stop_timeout_ms) *)&(engine-> props.stop_timeout_ms); membar_datadep_consumer(); __tmp; }); |
1011 | } |
1012 | |
1013 | int intel_engine_stop_cs(struct intel_engine_cs *engine) |
1014 | { |
1015 | struct intel_uncore *uncore = engine->uncore; |
1016 | const u32 base = engine->mmio_base; |
1017 | const i915_reg_t mode = RING_MI_MODE(base)((const i915_reg_t){ .reg = ((base) + 0x9c) }); |
1018 | int err; |
1019 | |
1020 | if (INTEL_GEN(engine->i915)((&(engine->i915)->__info)->gen) < 3) |
1021 | return -ENODEV19; |
1022 | |
1023 | ENGINE_TRACE(engine, "\n")do { const struct intel_engine_cs *e__ __attribute__((__unused__ )) = (engine); do { } while (0); } while (0); |
1024 | |
1025 | intel_uncore_write_fw(uncore, mode, _MASKED_BIT_ENABLE(STOP_RING))__raw_uncore_write32(uncore, mode, ({ typeof((1 << 8)) _a = ((1 << 8)); ({ if (__builtin_constant_p(_a)) do { } while (0); if (__builtin_constant_p(_a)) do { } while (0); if (__builtin_constant_p (_a) && __builtin_constant_p(_a)) do { } while (0); ( (_a) << 16 | (_a)); }); })); |
1026 | |
1027 | err = 0; |
1028 | if (__intel_wait_for_register_fw(uncore, |
1029 | mode, MODE_IDLE(1 << 9), MODE_IDLE(1 << 9), |
1030 | 1000, stop_timeout(engine), |
1031 | NULL((void *)0))) { |
1032 | ENGINE_TRACE(engine, "timed out on STOP_RING -> IDLE\n")do { const struct intel_engine_cs *e__ __attribute__((__unused__ )) = (engine); do { } while (0); } while (0); |
1033 | err = -ETIMEDOUT60; |
1034 | } |
1035 | |
1036 | /* A final mmio read to let GPU writes be hopefully flushed to memory */ |
1037 | intel_uncore_posting_read_fw(uncore, mode)((void)__raw_uncore_read32(uncore, mode)); |
1038 | |
1039 | return err; |
1040 | } |
1041 | |
1042 | void intel_engine_cancel_stop_cs(struct intel_engine_cs *engine) |
1043 | { |
1044 | ENGINE_TRACE(engine, "\n")do { const struct intel_engine_cs *e__ __attribute__((__unused__ )) = (engine); do { } while (0); } while (0); |
1045 | |
1046 | ENGINE_WRITE_FW(engine, RING_MI_MODE, _MASKED_BIT_DISABLE(STOP_RING))__raw_uncore_write32(((engine))->uncore, ((const i915_reg_t ){ .reg = (((engine)->mmio_base) + 0x9c) }), ((({ if (__builtin_constant_p (((1 << 8)))) do { } while (0); if (__builtin_constant_p (0)) do { } while (0); if (__builtin_constant_p(((1 << 8 ))) && __builtin_constant_p(0)) do { } while (0); ((( (1 << 8))) << 16 | (0)); })))); |
1047 | } |
1048 | |
1049 | const char *i915_cache_level_str(struct drm_i915_privateinteldrm_softc *i915, int type) |
1050 | { |
1051 | switch (type) { |
1052 | case I915_CACHE_NONE: return " uncached"; |
1053 | case I915_CACHE_LLC: return HAS_LLC(i915)((&(i915)->__info)->has_llc) ? " LLC" : " snooped"; |
1054 | case I915_CACHE_L3_LLC: return " L3+LLC"; |
1055 | case I915_CACHE_WT: return " WT"; |
1056 | default: return ""; |
1057 | } |
1058 | } |
1059 | |
1060 | static u32 |
1061 | read_subslice_reg(const struct intel_engine_cs *engine, |
1062 | int slice, int subslice, i915_reg_t reg) |
1063 | { |
1064 | struct drm_i915_privateinteldrm_softc *i915 = engine->i915; |
1065 | struct intel_uncore *uncore = engine->uncore; |
1066 | u32 mcr_mask, mcr_ss, mcr, old_mcr, val; |
1067 | enum forcewake_domains fw_domains; |
1068 | |
1069 | if (INTEL_GEN(i915)((&(i915)->__info)->gen) >= 11) { |
1070 | mcr_mask = GEN11_MCR_SLICE_MASK(((0xf) & 0xf) << 27) | GEN11_MCR_SUBSLICE_MASK(((0x7) & 0x7) << 24); |
1071 | mcr_ss = GEN11_MCR_SLICE(slice)(((slice) & 0xf) << 27) | GEN11_MCR_SUBSLICE(subslice)(((subslice) & 0x7) << 24); |
1072 | } else { |
1073 | mcr_mask = GEN8_MCR_SLICE_MASK(((3) & 3) << 26) | GEN8_MCR_SUBSLICE_MASK(((3) & 3) << 24); |
1074 | mcr_ss = GEN8_MCR_SLICE(slice)(((slice) & 3) << 26) | GEN8_MCR_SUBSLICE(subslice)(((subslice) & 3) << 24); |
1075 | } |
1076 | |
1077 | fw_domains = intel_uncore_forcewake_for_reg(uncore, reg, |
1078 | FW_REG_READ(1)); |
1079 | fw_domains |= intel_uncore_forcewake_for_reg(uncore, |
1080 | GEN8_MCR_SELECTOR((const i915_reg_t){ .reg = (0xfdc) }), |
1081 | FW_REG_READ(1) | FW_REG_WRITE(2)); |
1082 | |
1083 | spin_lock_irq(&uncore->lock)mtx_enter(&uncore->lock); |
1084 | intel_uncore_forcewake_get__locked(uncore, fw_domains); |
1085 | |
1086 | old_mcr = mcr = intel_uncore_read_fw(uncore, GEN8_MCR_SELECTOR)__raw_uncore_read32(uncore, ((const i915_reg_t){ .reg = (0xfdc ) })); |
1087 | |
1088 | mcr &= ~mcr_mask; |
1089 | mcr |= mcr_ss; |
1090 | intel_uncore_write_fw(uncore, GEN8_MCR_SELECTOR, mcr)__raw_uncore_write32(uncore, ((const i915_reg_t){ .reg = (0xfdc ) }), mcr); |
1091 | |
1092 | val = intel_uncore_read_fw(uncore, reg)__raw_uncore_read32(uncore, reg); |
1093 | |
1094 | mcr &= ~mcr_mask; |
1095 | mcr |= old_mcr & mcr_mask; |
1096 | |
1097 | intel_uncore_write_fw(uncore, GEN8_MCR_SELECTOR, mcr)__raw_uncore_write32(uncore, ((const i915_reg_t){ .reg = (0xfdc ) }), mcr); |
1098 | |
1099 | intel_uncore_forcewake_put__locked(uncore, fw_domains); |
1100 | spin_unlock_irq(&uncore->lock)mtx_leave(&uncore->lock); |
1101 | |
1102 | return val; |
1103 | } |
1104 | |
1105 | /* NB: please notice the memset */ |
1106 | void intel_engine_get_instdone(const struct intel_engine_cs *engine, |
1107 | struct intel_instdone *instdone) |
1108 | { |
1109 | struct drm_i915_privateinteldrm_softc *i915 = engine->i915; |
1110 | const struct sseu_dev_info *sseu = &engine->gt->info.sseu; |
1111 | struct intel_uncore *uncore = engine->uncore; |
1112 | u32 mmio_base = engine->mmio_base; |
1113 | int slice; |
1114 | int subslice; |
1115 | |
1116 | memset(instdone, 0, sizeof(*instdone))__builtin_memset((instdone), (0), (sizeof(*instdone))); |
1117 | |
1118 | switch (INTEL_GEN(i915)((&(i915)->__info)->gen)) { |
1119 | default: |
1120 | instdone->instdone = |
1121 | intel_uncore_read(uncore, RING_INSTDONE(mmio_base)((const i915_reg_t){ .reg = ((mmio_base) + 0x6c) })); |
1122 | |
1123 | if (engine->id != RCS0) |
1124 | break; |
1125 | |
1126 | instdone->slice_common = |
1127 | intel_uncore_read(uncore, GEN7_SC_INSTDONE((const i915_reg_t){ .reg = (0x7100) })); |
1128 | if (INTEL_GEN(i915)((&(i915)->__info)->gen) >= 12) { |
1129 | instdone->slice_common_extra[0] = |
1130 | intel_uncore_read(uncore, GEN12_SC_INSTDONE_EXTRA((const i915_reg_t){ .reg = (0x7104) })); |
1131 | instdone->slice_common_extra[1] = |
1132 | intel_uncore_read(uncore, GEN12_SC_INSTDONE_EXTRA2((const i915_reg_t){ .reg = (0x7108) })); |
1133 | } |
1134 | for_each_instdone_slice_subslice(i915, sseu, slice, subslice)for ((slice) = 0, (subslice) = 0; (slice) < 3; (subslice) = ((subslice) + 1) % 8, (slice) += ((subslice) == 0)) if (!((( ((0 + (&(i915)->__info)->gen == (7)) ? 1 : ((sseu)-> slice_mask)) & (1UL << (slice)))) && (((0 + (&(i915)->__info)->gen == (7)) ? (1 & (1UL << (subslice))) : intel_sseu_has_subslice(sseu, 0, subslice)))) ) {} else { |
1135 | instdone->sampler[slice][subslice] = |
1136 | read_subslice_reg(engine, slice, subslice, |
1137 | GEN7_SAMPLER_INSTDONE((const i915_reg_t){ .reg = (0xe160) })); |
1138 | instdone->row[slice][subslice] = |
1139 | read_subslice_reg(engine, slice, subslice, |
1140 | GEN7_ROW_INSTDONE((const i915_reg_t){ .reg = (0xe164) })); |
1141 | } |
1142 | break; |
1143 | case 7: |
1144 | instdone->instdone = |
1145 | intel_uncore_read(uncore, RING_INSTDONE(mmio_base)((const i915_reg_t){ .reg = ((mmio_base) + 0x6c) })); |
1146 | |
1147 | if (engine->id != RCS0) |
1148 | break; |
1149 | |
1150 | instdone->slice_common = |
1151 | intel_uncore_read(uncore, GEN7_SC_INSTDONE((const i915_reg_t){ .reg = (0x7100) })); |
1152 | instdone->sampler[0][0] = |
1153 | intel_uncore_read(uncore, GEN7_SAMPLER_INSTDONE((const i915_reg_t){ .reg = (0xe160) })); |
1154 | instdone->row[0][0] = |
1155 | intel_uncore_read(uncore, GEN7_ROW_INSTDONE((const i915_reg_t){ .reg = (0xe164) })); |
1156 | |
1157 | break; |
1158 | case 6: |
1159 | case 5: |
1160 | case 4: |
1161 | instdone->instdone = |
1162 | intel_uncore_read(uncore, RING_INSTDONE(mmio_base)((const i915_reg_t){ .reg = ((mmio_base) + 0x6c) })); |
1163 | if (engine->id == RCS0) |
1164 | /* HACK: Using the wrong struct member */ |
1165 | instdone->slice_common = |
1166 | intel_uncore_read(uncore, GEN4_INSTDONE1((const i915_reg_t){ .reg = (0x207c) })); |
1167 | break; |
1168 | case 3: |
1169 | case 2: |
1170 | instdone->instdone = intel_uncore_read(uncore, GEN2_INSTDONE((const i915_reg_t){ .reg = (0x2090) })); |
1171 | break; |
1172 | } |
1173 | } |
1174 | |
1175 | static bool_Bool ring_is_idle(struct intel_engine_cs *engine) |
1176 | { |
1177 | bool_Bool idle = true1; |
1178 | |
1179 | if (I915_SELFTEST_ONLY(!engine->mmio_base)0) |
1180 | return true1; |
1181 | |
1182 | if (!intel_engine_pm_get_if_awake(engine)) |
1183 | return true1; |
1184 | |
1185 | /* First check that no commands are left in the ring */ |
1186 | if ((ENGINE_READ(engine, RING_HEAD)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x34) })) & HEAD_ADDR0x001FFFFC) != |
1187 | (ENGINE_READ(engine, RING_TAIL)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x30) })) & TAIL_ADDR0x001FFFF8)) |
1188 | idle = false0; |
1189 | |
1190 | /* No bit for gen2, so assume the CS parser is idle */ |
1191 | if (INTEL_GEN(engine->i915)((&(engine->i915)->__info)->gen) > 2 && |
1192 | !(ENGINE_READ(engine, RING_MI_MODE)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x9c) })) & MODE_IDLE(1 << 9))) |
1193 | idle = false0; |
1194 | |
1195 | intel_engine_pm_put(engine); |
1196 | |
1197 | return idle; |
1198 | } |
1199 | |
1200 | void intel_engine_flush_submission(struct intel_engine_cs *engine) |
1201 | { |
1202 | struct tasklet_struct *t = &engine->execlists.tasklet; |
1203 | |
1204 | if (!t->func) |
1205 | return; |
1206 | |
1207 | /* Synchronise and wait for the tasklet on another CPU */ |
1208 | tasklet_kill(t); |
1209 | |
1210 | /* Having cancelled the tasklet, ensure that is run */ |
1211 | local_bh_disable(); |
1212 | if (tasklet_trylock(t)) { |
1213 | /* Must wait for any GPU reset in progress. */ |
1214 | if (__tasklet_is_enabled(t)) |
1215 | t->func(t->data); |
1216 | tasklet_unlock(t); |
1217 | } |
1218 | local_bh_enable(); |
1219 | } |
1220 | |
1221 | /** |
1222 | * intel_engine_is_idle() - Report if the engine has finished process all work |
1223 | * @engine: the intel_engine_cs |
1224 | * |
1225 | * Return true if there are no requests pending, nothing left to be submitted |
1226 | * to hardware, and that the engine is idle. |
1227 | */ |
1228 | bool_Bool intel_engine_is_idle(struct intel_engine_cs *engine) |
1229 | { |
1230 | /* More white lies, if wedged, hw state is inconsistent */ |
1231 | if (intel_gt_is_wedged(engine->gt)) |
1232 | return true1; |
1233 | |
1234 | if (!intel_engine_pm_is_awake(engine)) |
1235 | return true1; |
1236 | |
1237 | /* Waiting to drain ELSP? */ |
1238 | if (execlists_active(&engine->execlists)) { |
1239 | #ifdef __linux__ |
1240 | synchronize_hardirq(engine->i915->drm.pdev->irq); |
1241 | #else |
1242 | intr_barrier(engine->i915->irqh); |
1243 | #endif |
1244 | |
1245 | intel_engine_flush_submission(engine); |
1246 | |
1247 | if (execlists_active(&engine->execlists)) |
1248 | return false0; |
1249 | } |
1250 | |
1251 | /* ELSP is empty, but there are ready requests? E.g. after reset */ |
1252 | if (!RB_EMPTY_ROOT(&engine->execlists.queue.rb_root)((&engine->execlists.queue.rb_root)->rb_node == ((void *)0))) |
1253 | return false0; |
1254 | |
1255 | /* Ring stopped? */ |
1256 | return ring_is_idle(engine); |
1257 | } |
1258 | |
1259 | bool_Bool intel_engines_are_idle(struct intel_gt *gt) |
1260 | { |
1261 | struct intel_engine_cs *engine; |
1262 | enum intel_engine_id id; |
1263 | |
1264 | /* |
1265 | * If the driver is wedged, HW state may be very inconsistent and |
1266 | * report that it is still busy, even though we have stopped using it. |
1267 | */ |
1268 | if (intel_gt_is_wedged(gt)) |
1269 | return true1; |
1270 | |
1271 | /* Already parked (and passed an idleness test); must still be idle */ |
1272 | if (!READ_ONCE(gt->awake)({ typeof(gt->awake) __tmp = *(volatile typeof(gt->awake ) *)&(gt->awake); membar_datadep_consumer(); __tmp; })) |
1273 | return true1; |
1274 | |
1275 | for_each_engine(engine, gt, id)for ((id) = 0; (id) < I915_NUM_ENGINES; (id)++) if (!((engine ) = (gt)->engine[(id)])) {} else { |
1276 | if (!intel_engine_is_idle(engine)) |
1277 | return false0; |
1278 | } |
1279 | |
1280 | return true1; |
1281 | } |
1282 | |
1283 | void intel_engines_reset_default_submission(struct intel_gt *gt) |
1284 | { |
1285 | struct intel_engine_cs *engine; |
1286 | enum intel_engine_id id; |
1287 | |
1288 | for_each_engine(engine, gt, id)for ((id) = 0; (id) < I915_NUM_ENGINES; (id)++) if (!((engine ) = (gt)->engine[(id)])) {} else |
1289 | engine->set_default_submission(engine); |
1290 | } |
1291 | |
1292 | bool_Bool intel_engine_can_store_dword(struct intel_engine_cs *engine) |
1293 | { |
1294 | switch (INTEL_GEN(engine->i915)((&(engine->i915)->__info)->gen)) { |
1295 | case 2: |
1296 | return false0; /* uses physical not virtual addresses */ |
1297 | case 3: |
1298 | /* maybe only uses physical not virtual addresses */ |
1299 | return !(IS_I915G(engine->i915)IS_PLATFORM(engine->i915, INTEL_I915G) || IS_I915GM(engine->i915)IS_PLATFORM(engine->i915, INTEL_I915GM)); |
1300 | case 4: |
1301 | return !IS_I965G(engine->i915)IS_PLATFORM(engine->i915, INTEL_I965G); /* who knows! */ |
1302 | case 6: |
1303 | return engine->class != VIDEO_DECODE_CLASS1; /* b0rked */ |
1304 | default: |
1305 | return true1; |
1306 | } |
1307 | } |
1308 | |
1309 | static int print_sched_attr(const struct i915_sched_attr *attr, |
1310 | char *buf, int x, int len) |
1311 | { |
1312 | if (attr->priority == I915_PRIORITY_INVALID((-0x7fffffff-1) | (u8)((1UL << (0)) - 1))) |
1313 | return x; |
1314 | |
1315 | x += snprintf(buf + x, len - x, |
1316 | " prio=%d", attr->priority); |
1317 | |
1318 | return x; |
1319 | } |
1320 | |
1321 | static void print_request(struct drm_printer *m, |
1322 | struct i915_request *rq, |
1323 | const char *prefix) |
1324 | { |
1325 | const char *name = rq->fence.ops->get_timeline_name(&rq->fence); |
1326 | char buf[80] = ""; |
1327 | int x = 0; |
1328 | |
1329 | x = print_sched_attr(&rq->sched.attr, buf, x, sizeof(buf)); |
Value stored to 'x' is never read | |
1330 | |
1331 | drm_printf(m, "%s %llx:%llx%s%s %s @ %dms: %s\n", |
1332 | prefix, |
1333 | rq->fence.context, rq->fence.seqno, |
1334 | i915_request_completed(rq) ? "!" : |
1335 | i915_request_started(rq) ? "*" : |
1336 | "", |
1337 | test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, |
1338 | &rq->fence.flags) ? "+" : |
1339 | test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT, |
1340 | &rq->fence.flags) ? "-" : |
1341 | "", |
1342 | buf, |
1343 | jiffies_to_msecs(jiffies - rq->emitted_jiffies), |
1344 | name); |
1345 | } |
1346 | |
1347 | static struct intel_timeline *get_timeline(struct i915_request *rq) |
1348 | { |
1349 | struct intel_timeline *tl; |
1350 | |
1351 | /* |
1352 | * Even though we are holding the engine->active.lock here, there |
1353 | * is no control over the submission queue per-se and we are |
1354 | * inspecting the active state at a random point in time, with an |
1355 | * unknown queue. Play safe and make sure the timeline remains valid. |
1356 | * (Only being used for pretty printing, one extra kref shouldn't |
1357 | * cause a camel stampede!) |
1358 | */ |
1359 | rcu_read_lock(); |
1360 | tl = rcu_dereference(rq->timeline)(rq->timeline); |
1361 | if (!kref_get_unless_zero(&tl->kref)) |
1362 | tl = NULL((void *)0); |
1363 | rcu_read_unlock(); |
1364 | |
1365 | return tl; |
1366 | } |
1367 | |
1368 | static int print_ring(char *buf, int sz, struct i915_request *rq) |
1369 | { |
1370 | int len = 0; |
1371 | |
1372 | if (!i915_request_signaled(rq)) { |
1373 | struct intel_timeline *tl = get_timeline(rq); |
1374 | |
1375 | len = scnprintf(buf, sz,snprintf(buf, sz, "ring:{start:%08x, hwsp:%08x, seqno:%08x, runtime:%llums}, " , i915_ggtt_offset(rq->ring->vma), tl ? tl->hwsp_offset : 0, hwsp_seqno(rq), (((intel_context_get_total_runtime_ns(rq ->context)) + ((1000 * 1000) / 2)) / (1000 * 1000))) |
1376 | "ring:{start:%08x, hwsp:%08x, seqno:%08x, runtime:%llums}, ",snprintf(buf, sz, "ring:{start:%08x, hwsp:%08x, seqno:%08x, runtime:%llums}, " , i915_ggtt_offset(rq->ring->vma), tl ? tl->hwsp_offset : 0, hwsp_seqno(rq), (((intel_context_get_total_runtime_ns(rq ->context)) + ((1000 * 1000) / 2)) / (1000 * 1000))) |
1377 | i915_ggtt_offset(rq->ring->vma),snprintf(buf, sz, "ring:{start:%08x, hwsp:%08x, seqno:%08x, runtime:%llums}, " , i915_ggtt_offset(rq->ring->vma), tl ? tl->hwsp_offset : 0, hwsp_seqno(rq), (((intel_context_get_total_runtime_ns(rq ->context)) + ((1000 * 1000) / 2)) / (1000 * 1000))) |
1378 | tl ? tl->hwsp_offset : 0,snprintf(buf, sz, "ring:{start:%08x, hwsp:%08x, seqno:%08x, runtime:%llums}, " , i915_ggtt_offset(rq->ring->vma), tl ? tl->hwsp_offset : 0, hwsp_seqno(rq), (((intel_context_get_total_runtime_ns(rq ->context)) + ((1000 * 1000) / 2)) / (1000 * 1000))) |
1379 | hwsp_seqno(rq),snprintf(buf, sz, "ring:{start:%08x, hwsp:%08x, seqno:%08x, runtime:%llums}, " , i915_ggtt_offset(rq->ring->vma), tl ? tl->hwsp_offset : 0, hwsp_seqno(rq), (((intel_context_get_total_runtime_ns(rq ->context)) + ((1000 * 1000) / 2)) / (1000 * 1000))) |
1380 | DIV_ROUND_CLOSEST_ULL(intel_context_get_total_runtime_ns(rq->context),snprintf(buf, sz, "ring:{start:%08x, hwsp:%08x, seqno:%08x, runtime:%llums}, " , i915_ggtt_offset(rq->ring->vma), tl ? tl->hwsp_offset : 0, hwsp_seqno(rq), (((intel_context_get_total_runtime_ns(rq ->context)) + ((1000 * 1000) / 2)) / (1000 * 1000))) |
1381 | 1000 * 1000))snprintf(buf, sz, "ring:{start:%08x, hwsp:%08x, seqno:%08x, runtime:%llums}, " , i915_ggtt_offset(rq->ring->vma), tl ? tl->hwsp_offset : 0, hwsp_seqno(rq), (((intel_context_get_total_runtime_ns(rq ->context)) + ((1000 * 1000) / 2)) / (1000 * 1000))); |
1382 | |
1383 | if (tl) |
1384 | intel_timeline_put(tl); |
1385 | } |
1386 | |
1387 | return len; |
1388 | } |
1389 | |
1390 | static void hexdump(struct drm_printer *m, const void *buf, size_t len) |
1391 | { |
1392 | STUB()do { printf("%s: stub\n", __func__); } while(0); |
1393 | #ifdef notyet |
1394 | const size_t rowsize = 8 * sizeof(u32); |
1395 | const void *prev = NULL((void *)0); |
1396 | bool_Bool skip = false0; |
1397 | size_t pos; |
1398 | |
1399 | for (pos = 0; pos < len; pos += rowsize) { |
1400 | char line[128]; |
1401 | |
1402 | if (prev && !memcmp(prev, buf + pos, rowsize)__builtin_memcmp((prev), (buf + pos), (rowsize))) { |
1403 | if (!skip) { |
1404 | drm_printf(m, "*\n"); |
1405 | skip = true1; |
1406 | } |
1407 | continue; |
1408 | } |
1409 | |
1410 | WARN_ON_ONCE(hex_dump_to_buffer(buf + pos, len - pos,({ static int __warned; int __ret = !!((hex_dump_to_buffer(buf + pos, len - pos, rowsize, sizeof(u32), line, sizeof(line), 0 ) >= sizeof(line))); if (__ret && !__warned) { printf ("%s", "WARN_ON_ONCE(" "hex_dump_to_buffer(buf + pos, len - pos, rowsize, sizeof(u32), line, sizeof(line), 0) >= sizeof(line)" ")"); __warned = 1; } __builtin_expect(!!(__ret), 0); }) |
1411 | rowsize, sizeof(u32),({ static int __warned; int __ret = !!((hex_dump_to_buffer(buf + pos, len - pos, rowsize, sizeof(u32), line, sizeof(line), 0 ) >= sizeof(line))); if (__ret && !__warned) { printf ("%s", "WARN_ON_ONCE(" "hex_dump_to_buffer(buf + pos, len - pos, rowsize, sizeof(u32), line, sizeof(line), 0) >= sizeof(line)" ")"); __warned = 1; } __builtin_expect(!!(__ret), 0); }) |
1412 | line, sizeof(line),({ static int __warned; int __ret = !!((hex_dump_to_buffer(buf + pos, len - pos, rowsize, sizeof(u32), line, sizeof(line), 0 ) >= sizeof(line))); if (__ret && !__warned) { printf ("%s", "WARN_ON_ONCE(" "hex_dump_to_buffer(buf + pos, len - pos, rowsize, sizeof(u32), line, sizeof(line), 0) >= sizeof(line)" ")"); __warned = 1; } __builtin_expect(!!(__ret), 0); }) |
1413 | false) >= sizeof(line))({ static int __warned; int __ret = !!((hex_dump_to_buffer(buf + pos, len - pos, rowsize, sizeof(u32), line, sizeof(line), 0 ) >= sizeof(line))); if (__ret && !__warned) { printf ("%s", "WARN_ON_ONCE(" "hex_dump_to_buffer(buf + pos, len - pos, rowsize, sizeof(u32), line, sizeof(line), 0) >= sizeof(line)" ")"); __warned = 1; } __builtin_expect(!!(__ret), 0); }); |
1414 | drm_printf(m, "[%04zx] %s\n", pos, line); |
1415 | |
1416 | prev = buf + pos; |
1417 | skip = false0; |
1418 | } |
1419 | #endif |
1420 | } |
1421 | |
1422 | static const char *repr_timer(const struct timeout *t) |
1423 | { |
1424 | if (!READ_ONCE(t->to_time)({ typeof(t->to_time) __tmp = *(volatile typeof(t->to_time ) *)&(t->to_time); membar_datadep_consumer(); __tmp; } )) |
1425 | return "inactive"; |
1426 | |
1427 | if (timer_pending(t)(((t))->to_flags & 0x02)) |
1428 | return "active"; |
1429 | |
1430 | return "expired"; |
1431 | } |
1432 | |
1433 | static void intel_engine_print_registers(struct intel_engine_cs *engine, |
1434 | struct drm_printer *m) |
1435 | { |
1436 | struct drm_i915_privateinteldrm_softc *dev_priv = engine->i915; |
1437 | struct intel_engine_execlists * const execlists = &engine->execlists; |
1438 | u64 addr; |
1439 | |
1440 | if (engine->id == RENDER_CLASS0 && IS_GEN_RANGE(dev_priv, 4, 7)(!!((&(dev_priv)->__info)->gen_mask & ( 0 + 0 + (((~0UL) >> (64 - (((7)) - 1) - 1)) & ((~0UL) << (((4)) - 1))))))) |
1441 | drm_printf(m, "\tCCID: 0x%08x\n", ENGINE_READ(engine, CCID)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x180) }))); |
1442 | if (HAS_EXECLISTS(dev_priv)((&(dev_priv)->__info)->has_logical_ring_contexts)) { |
1443 | drm_printf(m, "\tEL_STAT_HI: 0x%08x\n", |
1444 | ENGINE_READ(engine, RING_EXECLIST_STATUS_HI)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x234 + 4) }))); |
1445 | drm_printf(m, "\tEL_STAT_LO: 0x%08x\n", |
1446 | ENGINE_READ(engine, RING_EXECLIST_STATUS_LO)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x234) }))); |
1447 | } |
1448 | drm_printf(m, "\tRING_START: 0x%08x\n", |
1449 | ENGINE_READ(engine, RING_START)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x38) }))); |
1450 | drm_printf(m, "\tRING_HEAD: 0x%08x\n", |
1451 | ENGINE_READ(engine, RING_HEAD)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x34) })) & HEAD_ADDR0x001FFFFC); |
1452 | drm_printf(m, "\tRING_TAIL: 0x%08x\n", |
1453 | ENGINE_READ(engine, RING_TAIL)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x30) })) & TAIL_ADDR0x001FFFF8); |
1454 | drm_printf(m, "\tRING_CTL: 0x%08x%s\n", |
1455 | ENGINE_READ(engine, RING_CTL)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x3c) })), |
1456 | ENGINE_READ(engine, RING_CTL)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x3c) })) & (RING_WAIT(1 << 11) | RING_WAIT_SEMAPHORE(1 << 10)) ? " [waiting]" : ""); |
1457 | if (INTEL_GEN(engine->i915)((&(engine->i915)->__info)->gen) > 2) { |
1458 | drm_printf(m, "\tRING_MODE: 0x%08x%s\n", |
1459 | ENGINE_READ(engine, RING_MI_MODE)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x9c) })), |
1460 | ENGINE_READ(engine, RING_MI_MODE)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x9c) })) & (MODE_IDLE(1 << 9)) ? " [idle]" : ""); |
1461 | } |
1462 | |
1463 | if (INTEL_GEN(dev_priv)((&(dev_priv)->__info)->gen) >= 6) { |
1464 | drm_printf(m, "\tRING_IMR: 0x%08x\n", |
1465 | ENGINE_READ(engine, RING_IMR)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0xa8) }))); |
1466 | drm_printf(m, "\tRING_ESR: 0x%08x\n", |
1467 | ENGINE_READ(engine, RING_ESR)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0xb8) }))); |
1468 | drm_printf(m, "\tRING_EMR: 0x%08x\n", |
1469 | ENGINE_READ(engine, RING_EMR)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0xb4) }))); |
1470 | drm_printf(m, "\tRING_EIR: 0x%08x\n", |
1471 | ENGINE_READ(engine, RING_EIR)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0xb0) }))); |
1472 | } |
1473 | |
1474 | addr = intel_engine_get_active_head(engine); |
1475 | drm_printf(m, "\tACTHD: 0x%08x_%08x\n", |
1476 | upper_32_bits(addr)((u32)(((addr) >> 16) >> 16)), lower_32_bits(addr)((u32)(addr))); |
1477 | addr = intel_engine_get_last_batch_head(engine); |
1478 | drm_printf(m, "\tBBADDR: 0x%08x_%08x\n", |
1479 | upper_32_bits(addr)((u32)(((addr) >> 16) >> 16)), lower_32_bits(addr)((u32)(addr))); |
1480 | if (INTEL_GEN(dev_priv)((&(dev_priv)->__info)->gen) >= 8) |
1481 | addr = ENGINE_READ64(engine, RING_DMA_FADD, RING_DMA_FADD_UDW)intel_uncore_read64_2x32(((engine))->uncore, ((const i915_reg_t ){ .reg = (((engine)->mmio_base) + 0x78) }), ((const i915_reg_t ){ .reg = (((engine)->mmio_base) + 0x60) })); |
1482 | else if (INTEL_GEN(dev_priv)((&(dev_priv)->__info)->gen) >= 4) |
1483 | addr = ENGINE_READ(engine, RING_DMA_FADD)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x78) })); |
1484 | else |
1485 | addr = ENGINE_READ(engine, DMA_FADD_I8XX)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0xd0) })); |
1486 | drm_printf(m, "\tDMA_FADDR: 0x%08x_%08x\n", |
1487 | upper_32_bits(addr)((u32)(((addr) >> 16) >> 16)), lower_32_bits(addr)((u32)(addr))); |
1488 | if (INTEL_GEN(dev_priv)((&(dev_priv)->__info)->gen) >= 4) { |
1489 | drm_printf(m, "\tIPEIR: 0x%08x\n", |
1490 | ENGINE_READ(engine, RING_IPEIR)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x64) }))); |
1491 | drm_printf(m, "\tIPEHR: 0x%08x\n", |
1492 | ENGINE_READ(engine, RING_IPEHR)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x68) }))); |
1493 | } else { |
1494 | drm_printf(m, "\tIPEIR: 0x%08x\n", ENGINE_READ(engine, IPEIR)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x88) }))); |
1495 | drm_printf(m, "\tIPEHR: 0x%08x\n", ENGINE_READ(engine, IPEHR)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x8c) }))); |
1496 | } |
1497 | |
1498 | if (HAS_EXECLISTS(dev_priv)((&(dev_priv)->__info)->has_logical_ring_contexts)) { |
1499 | struct i915_request * const *port, *rq; |
1500 | const u32 *hws = |
1501 | &engine->status_page.addr[I915_HWS_CSB_BUF0_INDEX0x10]; |
1502 | const u8 num_entries = execlists->csb_size; |
1503 | unsigned int idx; |
1504 | u8 read, write; |
1505 | |
1506 | drm_printf(m, "\tExeclist tasklet queued? %s (%s), preempt? %s, timeslice? %s\n", |
1507 | yesno(test_bit(TASKLET_STATE_SCHED1, |
1508 | &engine->execlists.tasklet.state)), |
1509 | enableddisabled(!atomic_read(&engine->execlists.tasklet.count)({ typeof(*(&engine->execlists.tasklet.count)) __tmp = *(volatile typeof(*(&engine->execlists.tasklet.count) ) *)&(*(&engine->execlists.tasklet.count)); membar_datadep_consumer (); __tmp; })), |
1510 | repr_timer(&engine->execlists.preempt), |
1511 | repr_timer(&engine->execlists.timer)); |
1512 | |
1513 | read = execlists->csb_head; |
1514 | write = READ_ONCE(*execlists->csb_write)({ typeof(*execlists->csb_write) __tmp = *(volatile typeof (*execlists->csb_write) *)&(*execlists->csb_write); membar_datadep_consumer(); __tmp; }); |
1515 | |
1516 | drm_printf(m, "\tExeclist status: 0x%08x %08x; CSB read:%d, write:%d, entries:%d\n", |
1517 | ENGINE_READ(engine, RING_EXECLIST_STATUS_LO)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x234) })), |
1518 | ENGINE_READ(engine, RING_EXECLIST_STATUS_HI)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x234 + 4) })), |
1519 | read, write, num_entries); |
1520 | |
1521 | if (read >= num_entries) |
1522 | read = 0; |
1523 | if (write >= num_entries) |
1524 | write = 0; |
1525 | if (read > write) |
1526 | write += num_entries; |
1527 | while (read < write) { |
1528 | idx = ++read % num_entries; |
1529 | drm_printf(m, "\tExeclist CSB[%d]: 0x%08x, context: %d\n", |
1530 | idx, hws[idx * 2], hws[idx * 2 + 1]); |
1531 | } |
1532 | |
1533 | execlists_active_lock_bh(execlists); |
1534 | rcu_read_lock(); |
1535 | for (port = execlists->active; (rq = *port); port++) { |
1536 | char hdr[160]; |
1537 | int len; |
1538 | |
1539 | len = scnprintf(hdr, sizeof(hdr),snprintf(hdr, sizeof(hdr), "\t\tActive[%d]: ccid:%08x%s%s, " , (int)(port - execlists->active), rq->context->lrc. ccid, intel_context_is_closed(rq->context) ? "!" : "", intel_context_is_banned (rq->context) ? "*" : "") |
1540 | "\t\tActive[%d]: ccid:%08x%s%s, ",snprintf(hdr, sizeof(hdr), "\t\tActive[%d]: ccid:%08x%s%s, " , (int)(port - execlists->active), rq->context->lrc. ccid, intel_context_is_closed(rq->context) ? "!" : "", intel_context_is_banned (rq->context) ? "*" : "") |
1541 | (int)(port - execlists->active),snprintf(hdr, sizeof(hdr), "\t\tActive[%d]: ccid:%08x%s%s, " , (int)(port - execlists->active), rq->context->lrc. ccid, intel_context_is_closed(rq->context) ? "!" : "", intel_context_is_banned (rq->context) ? "*" : "") |
1542 | rq->context->lrc.ccid,snprintf(hdr, sizeof(hdr), "\t\tActive[%d]: ccid:%08x%s%s, " , (int)(port - execlists->active), rq->context->lrc. ccid, intel_context_is_closed(rq->context) ? "!" : "", intel_context_is_banned (rq->context) ? "*" : "") |
1543 | intel_context_is_closed(rq->context) ? "!" : "",snprintf(hdr, sizeof(hdr), "\t\tActive[%d]: ccid:%08x%s%s, " , (int)(port - execlists->active), rq->context->lrc. ccid, intel_context_is_closed(rq->context) ? "!" : "", intel_context_is_banned (rq->context) ? "*" : "") |
1544 | intel_context_is_banned(rq->context) ? "*" : "")snprintf(hdr, sizeof(hdr), "\t\tActive[%d]: ccid:%08x%s%s, " , (int)(port - execlists->active), rq->context->lrc. ccid, intel_context_is_closed(rq->context) ? "!" : "", intel_context_is_banned (rq->context) ? "*" : ""); |
1545 | len += print_ring(hdr + len, sizeof(hdr) - len, rq); |
1546 | scnprintf(hdr + len, sizeof(hdr) - len, "rq: ")snprintf(hdr + len, sizeof(hdr) - len, "rq: "); |
1547 | print_request(m, rq, hdr); |
1548 | } |
1549 | for (port = execlists->pending; (rq = *port); port++) { |
1550 | char hdr[160]; |
1551 | int len; |
1552 | |
1553 | len = scnprintf(hdr, sizeof(hdr),snprintf(hdr, sizeof(hdr), "\t\tPending[%d]: ccid:%08x%s%s, " , (int)(port - execlists->pending), rq->context->lrc .ccid, intel_context_is_closed(rq->context) ? "!" : "", intel_context_is_banned (rq->context) ? "*" : "") |
1554 | "\t\tPending[%d]: ccid:%08x%s%s, ",snprintf(hdr, sizeof(hdr), "\t\tPending[%d]: ccid:%08x%s%s, " , (int)(port - execlists->pending), rq->context->lrc .ccid, intel_context_is_closed(rq->context) ? "!" : "", intel_context_is_banned (rq->context) ? "*" : "") |
1555 | (int)(port - execlists->pending),snprintf(hdr, sizeof(hdr), "\t\tPending[%d]: ccid:%08x%s%s, " , (int)(port - execlists->pending), rq->context->lrc .ccid, intel_context_is_closed(rq->context) ? "!" : "", intel_context_is_banned (rq->context) ? "*" : "") |
1556 | rq->context->lrc.ccid,snprintf(hdr, sizeof(hdr), "\t\tPending[%d]: ccid:%08x%s%s, " , (int)(port - execlists->pending), rq->context->lrc .ccid, intel_context_is_closed(rq->context) ? "!" : "", intel_context_is_banned (rq->context) ? "*" : "") |
1557 | intel_context_is_closed(rq->context) ? "!" : "",snprintf(hdr, sizeof(hdr), "\t\tPending[%d]: ccid:%08x%s%s, " , (int)(port - execlists->pending), rq->context->lrc .ccid, intel_context_is_closed(rq->context) ? "!" : "", intel_context_is_banned (rq->context) ? "*" : "") |
1558 | intel_context_is_banned(rq->context) ? "*" : "")snprintf(hdr, sizeof(hdr), "\t\tPending[%d]: ccid:%08x%s%s, " , (int)(port - execlists->pending), rq->context->lrc .ccid, intel_context_is_closed(rq->context) ? "!" : "", intel_context_is_banned (rq->context) ? "*" : ""); |
1559 | len += print_ring(hdr + len, sizeof(hdr) - len, rq); |
1560 | scnprintf(hdr + len, sizeof(hdr) - len, "rq: ")snprintf(hdr + len, sizeof(hdr) - len, "rq: "); |
1561 | print_request(m, rq, hdr); |
1562 | } |
1563 | rcu_read_unlock(); |
1564 | execlists_active_unlock_bh(execlists); |
1565 | } else if (INTEL_GEN(dev_priv)((&(dev_priv)->__info)->gen) > 6) { |
1566 | drm_printf(m, "\tPP_DIR_BASE: 0x%08x\n", |
1567 | ENGINE_READ(engine, RING_PP_DIR_BASE)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x228) }))); |
1568 | drm_printf(m, "\tPP_DIR_BASE_READ: 0x%08x\n", |
1569 | ENGINE_READ(engine, RING_PP_DIR_BASE_READ)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x518) }))); |
1570 | drm_printf(m, "\tPP_DIR_DCLV: 0x%08x\n", |
1571 | ENGINE_READ(engine, RING_PP_DIR_DCLV)intel_uncore_read(((engine))->uncore, ((const i915_reg_t){ .reg = (((engine)->mmio_base) + 0x220) }))); |
1572 | } |
1573 | } |
1574 | |
1575 | static void print_request_ring(struct drm_printer *m, struct i915_request *rq) |
1576 | { |
1577 | void *ring; |
1578 | int size; |
1579 | |
1580 | drm_printf(m, |
1581 | "[head %04x, postfix %04x, tail %04x, batch 0x%08x_%08x]:\n", |
1582 | rq->head, rq->postfix, rq->tail, |
1583 | rq->batch ? upper_32_bits(rq->batch->node.start)((u32)(((rq->batch->node.start) >> 16) >> 16 )) : ~0u, |
1584 | rq->batch ? lower_32_bits(rq->batch->node.start)((u32)(rq->batch->node.start)) : ~0u); |
1585 | |
1586 | size = rq->tail - rq->head; |
1587 | if (rq->tail < rq->head) |
1588 | size += rq->ring->size; |
1589 | |
1590 | ring = kmalloc(size, GFP_ATOMIC0x0002); |
1591 | if (ring) { |
1592 | const void *vaddr = rq->ring->vaddr; |
1593 | unsigned int head = rq->head; |
1594 | unsigned int len = 0; |
1595 | |
1596 | if (rq->tail < head) { |
1597 | len = rq->ring->size - head; |
1598 | memcpy(ring, vaddr + head, len)__builtin_memcpy((ring), (vaddr + head), (len)); |
1599 | head = 0; |
1600 | } |
1601 | memcpy(ring + len, vaddr + head, size - len)__builtin_memcpy((ring + len), (vaddr + head), (size - len)); |
1602 | |
1603 | hexdump(m, ring, size); |
1604 | kfree(ring); |
1605 | } |
1606 | } |
1607 | |
1608 | static unsigned long list_count(struct list_head *list) |
1609 | { |
1610 | struct list_head *pos; |
1611 | unsigned long count = 0; |
1612 | |
1613 | list_for_each(pos, list)for (pos = (list)->next; pos != list; pos = (pos)->next ) |
1614 | count++; |
1615 | |
1616 | return count; |
1617 | } |
1618 | |
1619 | void intel_engine_dump(struct intel_engine_cs *engine, |
1620 | struct drm_printer *m, |
1621 | const char *header, ...) |
1622 | { |
1623 | struct i915_gpu_error * const error = &engine->i915->gpu_error; |
1624 | struct i915_request *rq; |
1625 | intel_wakeref_t wakeref; |
1626 | unsigned long flags; |
1627 | ktime_t dummy; |
1628 | |
1629 | if (header) { |
1630 | va_list ap; |
1631 | |
1632 | va_start(ap, header)__builtin_va_start((ap), header); |
1633 | drm_vprintf(m, header, &ap); |
1634 | va_end(ap)__builtin_va_end((ap)); |
1635 | } |
1636 | |
1637 | if (intel_gt_is_wedged(engine->gt)) |
1638 | drm_printf(m, "*** WEDGED ***\n"); |
1639 | |
1640 | drm_printf(m, "\tAwake? %d\n", atomic_read(&engine->wakeref.count)({ typeof(*(&engine->wakeref.count)) __tmp = *(volatile typeof(*(&engine->wakeref.count)) *)&(*(&engine ->wakeref.count)); membar_datadep_consumer(); __tmp; })); |
1641 | drm_printf(m, "\tBarriers?: %s\n", |
1642 | yesno(!llist_empty(&engine->barrier_tasks))); |
1643 | drm_printf(m, "\tLatency: %luus\n", |
1644 | ewma__engine_latency_read(&engine->latency)); |
1645 | if (intel_engine_supports_stats(engine)) |
1646 | drm_printf(m, "\tRuntime: %llums\n", |
1647 | ktime_to_ms(intel_engine_get_busy_time(engine, |
1648 | &dummy))); |
1649 | drm_printf(m, "\tForcewake: %x domains, %d active\n", |
1650 | engine->fw_domain, atomic_read(&engine->fw_active)({ typeof(*(&engine->fw_active)) __tmp = *(volatile typeof (*(&engine->fw_active)) *)&(*(&engine->fw_active )); membar_datadep_consumer(); __tmp; })); |
1651 | |
1652 | rcu_read_lock(); |
1653 | rq = READ_ONCE(engine->heartbeat.systole)({ typeof(engine->heartbeat.systole) __tmp = *(volatile typeof (engine->heartbeat.systole) *)&(engine->heartbeat.systole ); membar_datadep_consumer(); __tmp; }); |
1654 | if (rq) |
1655 | drm_printf(m, "\tHeartbeat: %d ms ago\n", |
1656 | jiffies_to_msecs(jiffies - rq->emitted_jiffies)); |
1657 | rcu_read_unlock(); |
1658 | drm_printf(m, "\tReset count: %d (global %d)\n", |
1659 | i915_reset_engine_count(error, engine), |
1660 | i915_reset_count(error)); |
1661 | |
1662 | drm_printf(m, "\tRequests:\n"); |
1663 | |
1664 | spin_lock_irqsave(&engine->active.lock, flags)do { flags = 0; mtx_enter(&engine->active.lock); } while (0); |
1665 | rq = intel_engine_find_active_request(engine); |
1666 | if (rq) { |
1667 | struct intel_timeline *tl = get_timeline(rq); |
1668 | |
1669 | print_request(m, rq, "\t\tactive "); |
1670 | |
1671 | drm_printf(m, "\t\tring->start: 0x%08x\n", |
1672 | i915_ggtt_offset(rq->ring->vma)); |
1673 | drm_printf(m, "\t\tring->head: 0x%08x\n", |
1674 | rq->ring->head); |
1675 | drm_printf(m, "\t\tring->tail: 0x%08x\n", |
1676 | rq->ring->tail); |
1677 | drm_printf(m, "\t\tring->emit: 0x%08x\n", |
1678 | rq->ring->emit); |
1679 | drm_printf(m, "\t\tring->space: 0x%08x\n", |
1680 | rq->ring->space); |
1681 | |
1682 | if (tl) { |
1683 | drm_printf(m, "\t\tring->hwsp: 0x%08x\n", |
1684 | tl->hwsp_offset); |
1685 | intel_timeline_put(tl); |
1686 | } |
1687 | |
1688 | print_request_ring(m, rq); |
1689 | |
1690 | if (rq->context->lrc_reg_state) { |
1691 | drm_printf(m, "Logical Ring Context:\n"); |
1692 | hexdump(m, rq->context->lrc_reg_state, PAGE_SIZE(1 << 12)); |
1693 | } |
1694 | } |
1695 | drm_printf(m, "\tOn hold?: %lu\n", list_count(&engine->active.hold)); |
1696 | spin_unlock_irqrestore(&engine->active.lock, flags)do { (void)(flags); mtx_leave(&engine->active.lock); } while (0); |
1697 | |
1698 | drm_printf(m, "\tMMIO base: 0x%08x\n", engine->mmio_base); |
1699 | wakeref = intel_runtime_pm_get_if_in_use(engine->uncore->rpm); |
1700 | if (wakeref) { |
1701 | intel_engine_print_registers(engine, m); |
1702 | intel_runtime_pm_put(engine->uncore->rpm, wakeref); |
1703 | } else { |
1704 | drm_printf(m, "\tDevice is asleep; skipping register dump\n"); |
1705 | } |
1706 | |
1707 | intel_execlists_show_requests(engine, m, print_request, 8); |
1708 | |
1709 | drm_printf(m, "HWSP:\n"); |
1710 | hexdump(m, engine->status_page.addr, PAGE_SIZE(1 << 12)); |
1711 | |
1712 | drm_printf(m, "Idle? %s\n", yesno(intel_engine_is_idle(engine))); |
1713 | |
1714 | intel_engine_print_breadcrumbs(engine, m); |
1715 | } |
1716 | |
1717 | static ktime_t __intel_engine_get_busy_time(struct intel_engine_cs *engine, |
1718 | ktime_t *now) |
1719 | { |
1720 | ktime_t total = engine->stats.total; |
1721 | |
1722 | /* |
1723 | * If the engine is executing something at the moment |
1724 | * add it to the total. |
1725 | */ |
1726 | *now = ktime_get(); |
1727 | if (atomic_read(&engine->stats.active)({ typeof(*(&engine->stats.active)) __tmp = *(volatile typeof(*(&engine->stats.active)) *)&(*(&engine ->stats.active)); membar_datadep_consumer(); __tmp; })) |
1728 | total = ktime_add(total, ktime_sub(*now, engine->stats.start)); |
1729 | |
1730 | return total; |
1731 | } |
1732 | |
1733 | /** |
1734 | * intel_engine_get_busy_time() - Return current accumulated engine busyness |
1735 | * @engine: engine to report on |
1736 | * @now: monotonic timestamp of sampling |
1737 | * |
1738 | * Returns accumulated time @engine was busy since engine stats were enabled. |
1739 | */ |
1740 | ktime_t intel_engine_get_busy_time(struct intel_engine_cs *engine, ktime_t *now) |
1741 | { |
1742 | unsigned int seq; |
1743 | ktime_t total; |
1744 | |
1745 | do { |
1746 | seq = read_seqbegin(&engine->stats.lock); |
1747 | total = __intel_engine_get_busy_time(engine, now); |
1748 | } while (read_seqretry(&engine->stats.lock, seq)); |
1749 | |
1750 | return total; |
1751 | } |
1752 | |
1753 | static bool_Bool match_ring(struct i915_request *rq) |
1754 | { |
1755 | u32 ring = ENGINE_READ(rq->engine, RING_START)intel_uncore_read(((rq->engine))->uncore, ((const i915_reg_t ){ .reg = (((rq->engine)->mmio_base) + 0x38) })); |
1756 | |
1757 | return ring == i915_ggtt_offset(rq->ring->vma); |
1758 | } |
1759 | |
1760 | struct i915_request * |
1761 | intel_engine_find_active_request(struct intel_engine_cs *engine) |
1762 | { |
1763 | struct i915_request *request, *active = NULL((void *)0); |
1764 | |
1765 | /* |
1766 | * We are called by the error capture, reset and to dump engine |
1767 | * state at random points in time. In particular, note that neither is |
1768 | * crucially ordered with an interrupt. After a hang, the GPU is dead |
1769 | * and we assume that no more writes can happen (we waited long enough |
1770 | * for all writes that were in transaction to be flushed) - adding an |
1771 | * extra delay for a recent interrupt is pointless. Hence, we do |
1772 | * not need an engine->irq_seqno_barrier() before the seqno reads. |
1773 | * At all other times, we must assume the GPU is still running, but |
1774 | * we only care about the snapshot of this moment. |
1775 | */ |
1776 | lockdep_assert_held(&engine->active.lock)do { (void)(&engine->active.lock); } while(0); |
1777 | |
1778 | rcu_read_lock(); |
1779 | request = execlists_active(&engine->execlists); |
1780 | if (request) { |
1781 | struct intel_timeline *tl = request->context->timeline; |
1782 | |
1783 | list_for_each_entry_from_reverse(request, &tl->requests, link)for (; &request->link != (&tl->requests); request = ({ const __typeof( ((__typeof(*request) *)0)->link ) *__mptr = (request->link.prev); (__typeof(*request) *)( (char *)__mptr - __builtin_offsetof(__typeof(*request), link) );})) { |
1784 | if (i915_request_completed(request)) |
1785 | break; |
1786 | |
1787 | active = request; |
1788 | } |
1789 | } |
1790 | rcu_read_unlock(); |
1791 | if (active) |
1792 | return active; |
1793 | |
1794 | list_for_each_entry(request, &engine->active.requests, sched.link)for (request = ({ const __typeof( ((__typeof(*request) *)0)-> sched.link ) *__mptr = ((&engine->active.requests)-> next); (__typeof(*request) *)( (char *)__mptr - __builtin_offsetof (__typeof(*request), sched.link) );}); &request->sched .link != (&engine->active.requests); request = ({ const __typeof( ((__typeof(*request) *)0)->sched.link ) *__mptr = (request->sched.link.next); (__typeof(*request) *)( (char *)__mptr - __builtin_offsetof(__typeof(*request), sched.link ) );})) { |
1795 | if (i915_request_completed(request)) |
1796 | continue; |
1797 | |
1798 | if (!i915_request_started(request)) |
1799 | continue; |
1800 | |
1801 | /* More than one preemptible request may match! */ |
1802 | if (!match_ring(request)) |
1803 | continue; |
1804 | |
1805 | active = request; |
1806 | break; |
1807 | } |
1808 | |
1809 | return active; |
1810 | } |
1811 | |
1812 | #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)0 |
1813 | #include "mock_engine.c" |
1814 | #include "selftest_engine.c" |
1815 | #include "selftest_engine_cs.c" |
1816 | #endif |