]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright © 2014 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 | * Authors: | |
24 | * Ben Widawsky <ben@bwidawsk.net> | |
25 | * Michel Thierry <michel.thierry@intel.com> | |
26 | * Thomas Daniel <thomas.daniel@intel.com> | |
27 | * Oscar Mateo <oscar.mateo@intel.com> | |
28 | * | |
29 | */ | |
30 | ||
31 | /** | |
32 | * DOC: Logical Rings, Logical Ring Contexts and Execlists | |
33 | * | |
34 | * Motivation: | |
35 | * GEN8 brings an expansion of the HW contexts: "Logical Ring Contexts". | |
36 | * These expanded contexts enable a number of new abilities, especially | |
37 | * "Execlists" (also implemented in this file). | |
38 | * | |
39 | * One of the main differences with the legacy HW contexts is that logical | |
40 | * ring contexts incorporate many more things to the context's state, like | |
41 | * PDPs or ringbuffer control registers: | |
42 | * | |
43 | * The reason why PDPs are included in the context is straightforward: as | |
44 | * PPGTTs (per-process GTTs) are actually per-context, having the PDPs | |
45 | * contained there mean you don't need to do a ppgtt->switch_mm yourself, | |
46 | * instead, the GPU will do it for you on the context switch. | |
47 | * | |
48 | * But, what about the ringbuffer control registers (head, tail, etc..)? | |
49 | * shouldn't we just need a set of those per engine command streamer? This is | |
50 | * where the name "Logical Rings" starts to make sense: by virtualizing the | |
51 | * rings, the engine cs shifts to a new "ring buffer" with every context | |
52 | * switch. When you want to submit a workload to the GPU you: A) choose your | |
53 | * context, B) find its appropriate virtualized ring, C) write commands to it | |
54 | * and then, finally, D) tell the GPU to switch to that context. | |
55 | * | |
56 | * Instead of the legacy MI_SET_CONTEXT, the way you tell the GPU to switch | |
57 | * to a contexts is via a context execution list, ergo "Execlists". | |
58 | * | |
59 | * LRC implementation: | |
60 | * Regarding the creation of contexts, we have: | |
61 | * | |
62 | * - One global default context. | |
63 | * - One local default context for each opened fd. | |
64 | * - One local extra context for each context create ioctl call. | |
65 | * | |
66 | * Now that ringbuffers belong per-context (and not per-engine, like before) | |
67 | * and that contexts are uniquely tied to a given engine (and not reusable, | |
68 | * like before) we need: | |
69 | * | |
70 | * - One ringbuffer per-engine inside each context. | |
71 | * - One backing object per-engine inside each context. | |
72 | * | |
73 | * The global default context starts its life with these new objects fully | |
74 | * allocated and populated. The local default context for each opened fd is | |
75 | * more complex, because we don't know at creation time which engine is going | |
76 | * to use them. To handle this, we have implemented a deferred creation of LR | |
77 | * contexts: | |
78 | * | |
79 | * The local context starts its life as a hollow or blank holder, that only | |
80 | * gets populated for a given engine once we receive an execbuffer. If later | |
81 | * on we receive another execbuffer ioctl for the same context but a different | |
82 | * engine, we allocate/populate a new ringbuffer and context backing object and | |
83 | * so on. | |
84 | * | |
85 | * Finally, regarding local contexts created using the ioctl call: as they are | |
86 | * only allowed with the render ring, we can allocate & populate them right | |
87 | * away (no need to defer anything, at least for now). | |
88 | * | |
89 | * Execlists implementation: | |
90 | * Execlists are the new method by which, on gen8+ hardware, workloads are | |
91 | * submitted for execution (as opposed to the legacy, ringbuffer-based, method). | |
92 | * This method works as follows: | |
93 | * | |
94 | * When a request is committed, its commands (the BB start and any leading or | |
95 | * trailing commands, like the seqno breadcrumbs) are placed in the ringbuffer | |
96 | * for the appropriate context. The tail pointer in the hardware context is not | |
97 | * updated at this time, but instead, kept by the driver in the ringbuffer | |
98 | * structure. A structure representing this request is added to a request queue | |
99 | * for the appropriate engine: this structure contains a copy of the context's | |
100 | * tail after the request was written to the ring buffer and a pointer to the | |
101 | * context itself. | |
102 | * | |
103 | * If the engine's request queue was empty before the request was added, the | |
104 | * queue is processed immediately. Otherwise the queue will be processed during | |
105 | * a context switch interrupt. In any case, elements on the queue will get sent | |
106 | * (in pairs) to the GPU's ExecLists Submit Port (ELSP, for short) with a | |
107 | * globally unique 20-bits submission ID. | |
108 | * | |
109 | * When execution of a request completes, the GPU updates the context status | |
110 | * buffer with a context complete event and generates a context switch interrupt. | |
111 | * During the interrupt handling, the driver examines the events in the buffer: | |
112 | * for each context complete event, if the announced ID matches that on the head | |
113 | * of the request queue, then that request is retired and removed from the queue. | |
114 | * | |
115 | * After processing, if any requests were retired and the queue is not empty | |
116 | * then a new execution list can be submitted. The two requests at the front of | |
117 | * the queue are next to be submitted but since a context may not occur twice in | |
118 | * an execution list, if subsequent requests have the same ID as the first then | |
119 | * the two requests must be combined. This is done simply by discarding requests | |
120 | * at the head of the queue until either only one requests is left (in which case | |
121 | * we use a NULL second context) or the first two requests have unique IDs. | |
122 | * | |
123 | * By always executing the first two requests in the queue the driver ensures | |
124 | * that the GPU is kept as busy as possible. In the case where a single context | |
125 | * completes but a second context is still executing, the request for this second | |
126 | * context will be at the head of the queue when we remove the first one. This | |
127 | * request will then be resubmitted along with a new request for a different context, | |
128 | * which will cause the hardware to continue executing the second request and queue | |
129 | * the new request (the GPU detects the condition of a context getting preempted | |
130 | * with the same context and optimizes the context switch flow by not doing | |
131 | * preemption, but just sampling the new tail pointer). | |
132 | * | |
133 | */ | |
134 | ||
135 | #include <drm/drmP.h> | |
136 | #include <drm/i915_drm.h> | |
137 | #include "i915_drv.h" | |
138 | ||
139 | #define GEN9_LR_CONTEXT_RENDER_SIZE (22 * PAGE_SIZE) | |
140 | #define GEN8_LR_CONTEXT_RENDER_SIZE (20 * PAGE_SIZE) | |
141 | #define GEN8_LR_CONTEXT_OTHER_SIZE (2 * PAGE_SIZE) | |
142 | ||
143 | #define RING_EXECLIST_QFULL (1 << 0x2) | |
144 | #define RING_EXECLIST1_VALID (1 << 0x3) | |
145 | #define RING_EXECLIST0_VALID (1 << 0x4) | |
146 | #define RING_EXECLIST_ACTIVE_STATUS (3 << 0xE) | |
147 | #define RING_EXECLIST1_ACTIVE (1 << 0x11) | |
148 | #define RING_EXECLIST0_ACTIVE (1 << 0x12) | |
149 | ||
150 | #define GEN8_CTX_STATUS_IDLE_ACTIVE (1 << 0) | |
151 | #define GEN8_CTX_STATUS_PREEMPTED (1 << 1) | |
152 | #define GEN8_CTX_STATUS_ELEMENT_SWITCH (1 << 2) | |
153 | #define GEN8_CTX_STATUS_ACTIVE_IDLE (1 << 3) | |
154 | #define GEN8_CTX_STATUS_COMPLETE (1 << 4) | |
155 | #define GEN8_CTX_STATUS_LITE_RESTORE (1 << 15) | |
156 | ||
157 | #define CTX_LRI_HEADER_0 0x01 | |
158 | #define CTX_CONTEXT_CONTROL 0x02 | |
159 | #define CTX_RING_HEAD 0x04 | |
160 | #define CTX_RING_TAIL 0x06 | |
161 | #define CTX_RING_BUFFER_START 0x08 | |
162 | #define CTX_RING_BUFFER_CONTROL 0x0a | |
163 | #define CTX_BB_HEAD_U 0x0c | |
164 | #define CTX_BB_HEAD_L 0x0e | |
165 | #define CTX_BB_STATE 0x10 | |
166 | #define CTX_SECOND_BB_HEAD_U 0x12 | |
167 | #define CTX_SECOND_BB_HEAD_L 0x14 | |
168 | #define CTX_SECOND_BB_STATE 0x16 | |
169 | #define CTX_BB_PER_CTX_PTR 0x18 | |
170 | #define CTX_RCS_INDIRECT_CTX 0x1a | |
171 | #define CTX_RCS_INDIRECT_CTX_OFFSET 0x1c | |
172 | #define CTX_LRI_HEADER_1 0x21 | |
173 | #define CTX_CTX_TIMESTAMP 0x22 | |
174 | #define CTX_PDP3_UDW 0x24 | |
175 | #define CTX_PDP3_LDW 0x26 | |
176 | #define CTX_PDP2_UDW 0x28 | |
177 | #define CTX_PDP2_LDW 0x2a | |
178 | #define CTX_PDP1_UDW 0x2c | |
179 | #define CTX_PDP1_LDW 0x2e | |
180 | #define CTX_PDP0_UDW 0x30 | |
181 | #define CTX_PDP0_LDW 0x32 | |
182 | #define CTX_LRI_HEADER_2 0x41 | |
183 | #define CTX_R_PWR_CLK_STATE 0x42 | |
184 | #define CTX_GPGPU_CSR_BASE_ADDRESS 0x44 | |
185 | ||
186 | #define GEN8_CTX_VALID (1<<0) | |
187 | #define GEN8_CTX_FORCE_PD_RESTORE (1<<1) | |
188 | #define GEN8_CTX_FORCE_RESTORE (1<<2) | |
189 | #define GEN8_CTX_L3LLC_COHERENT (1<<5) | |
190 | #define GEN8_CTX_PRIVILEGE (1<<8) | |
191 | enum { | |
192 | ADVANCED_CONTEXT = 0, | |
193 | LEGACY_CONTEXT, | |
194 | ADVANCED_AD_CONTEXT, | |
195 | LEGACY_64B_CONTEXT | |
196 | }; | |
197 | #define GEN8_CTX_MODE_SHIFT 3 | |
198 | enum { | |
199 | FAULT_AND_HANG = 0, | |
200 | FAULT_AND_HALT, /* Debug only */ | |
201 | FAULT_AND_STREAM, | |
202 | FAULT_AND_CONTINUE /* Unsupported */ | |
203 | }; | |
204 | #define GEN8_CTX_ID_SHIFT 32 | |
205 | ||
206 | static int intel_lr_context_pin(struct intel_engine_cs *ring, | |
207 | struct intel_context *ctx); | |
208 | ||
209 | /** | |
210 | * intel_sanitize_enable_execlists() - sanitize i915.enable_execlists | |
211 | * @dev: DRM device. | |
212 | * @enable_execlists: value of i915.enable_execlists module parameter. | |
213 | * | |
214 | * Only certain platforms support Execlists (the prerequisites being | |
215 | * support for Logical Ring Contexts and Aliasing PPGTT or better). | |
216 | * | |
217 | * Return: 1 if Execlists is supported and has to be enabled. | |
218 | */ | |
219 | int intel_sanitize_enable_execlists(struct drm_device *dev, int enable_execlists) | |
220 | { | |
221 | WARN_ON(i915.enable_ppgtt == -1); | |
222 | ||
223 | if (INTEL_INFO(dev)->gen >= 9) | |
224 | return 1; | |
225 | ||
226 | if (enable_execlists == 0) | |
227 | return 0; | |
228 | ||
229 | if (HAS_LOGICAL_RING_CONTEXTS(dev) && USES_PPGTT(dev) && | |
230 | i915.use_mmio_flip >= 0) | |
231 | return 1; | |
232 | ||
233 | return 0; | |
234 | } | |
235 | ||
236 | /** | |
237 | * intel_execlists_ctx_id() - get the Execlists Context ID | |
238 | * @ctx_obj: Logical Ring Context backing object. | |
239 | * | |
240 | * Do not confuse with ctx->id! Unfortunately we have a name overload | |
241 | * here: the old context ID we pass to userspace as a handler so that | |
242 | * they can refer to a context, and the new context ID we pass to the | |
243 | * ELSP so that the GPU can inform us of the context status via | |
244 | * interrupts. | |
245 | * | |
246 | * Return: 20-bits globally unique context ID. | |
247 | */ | |
248 | u32 intel_execlists_ctx_id(struct drm_i915_gem_object *ctx_obj) | |
249 | { | |
250 | u32 lrca = i915_gem_obj_ggtt_offset(ctx_obj); | |
251 | ||
252 | /* LRCA is required to be 4K aligned so the more significant 20 bits | |
253 | * are globally unique */ | |
254 | return lrca >> 12; | |
255 | } | |
256 | ||
257 | static uint64_t execlists_ctx_descriptor(struct drm_i915_gem_object *ctx_obj) | |
258 | { | |
259 | uint64_t desc; | |
260 | uint64_t lrca = i915_gem_obj_ggtt_offset(ctx_obj); | |
261 | ||
262 | WARN_ON(lrca & 0xFFFFFFFF00000FFFULL); | |
263 | ||
264 | desc = GEN8_CTX_VALID; | |
265 | desc |= LEGACY_CONTEXT << GEN8_CTX_MODE_SHIFT; | |
266 | desc |= GEN8_CTX_L3LLC_COHERENT; | |
267 | desc |= GEN8_CTX_PRIVILEGE; | |
268 | desc |= lrca; | |
269 | desc |= (u64)intel_execlists_ctx_id(ctx_obj) << GEN8_CTX_ID_SHIFT; | |
270 | ||
271 | /* TODO: WaDisableLiteRestore when we start using semaphore | |
272 | * signalling between Command Streamers */ | |
273 | /* desc |= GEN8_CTX_FORCE_RESTORE; */ | |
274 | ||
275 | return desc; | |
276 | } | |
277 | ||
278 | static void execlists_elsp_write(struct intel_engine_cs *ring, | |
279 | struct drm_i915_gem_object *ctx_obj0, | |
280 | struct drm_i915_gem_object *ctx_obj1) | |
281 | { | |
282 | struct drm_device *dev = ring->dev; | |
283 | struct drm_i915_private *dev_priv = dev->dev_private; | |
284 | uint64_t temp = 0; | |
285 | uint32_t desc[4]; | |
286 | ||
287 | /* XXX: You must always write both descriptors in the order below. */ | |
288 | if (ctx_obj1) | |
289 | temp = execlists_ctx_descriptor(ctx_obj1); | |
290 | else | |
291 | temp = 0; | |
292 | desc[1] = (u32)(temp >> 32); | |
293 | desc[0] = (u32)temp; | |
294 | ||
295 | temp = execlists_ctx_descriptor(ctx_obj0); | |
296 | desc[3] = (u32)(temp >> 32); | |
297 | desc[2] = (u32)temp; | |
298 | ||
299 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); | |
300 | I915_WRITE(RING_ELSP(ring), desc[1]); | |
301 | I915_WRITE(RING_ELSP(ring), desc[0]); | |
302 | I915_WRITE(RING_ELSP(ring), desc[3]); | |
303 | ||
304 | /* The context is automatically loaded after the following */ | |
305 | I915_WRITE(RING_ELSP(ring), desc[2]); | |
306 | ||
307 | /* ELSP is a wo register, so use another nearby reg for posting instead */ | |
308 | POSTING_READ(RING_EXECLIST_STATUS(ring)); | |
309 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); | |
310 | } | |
311 | ||
312 | static int execlists_update_context(struct drm_i915_gem_object *ctx_obj, | |
313 | struct drm_i915_gem_object *ring_obj, | |
314 | u32 tail) | |
315 | { | |
316 | struct page *page; | |
317 | uint32_t *reg_state; | |
318 | ||
319 | page = i915_gem_object_get_page(ctx_obj, 1); | |
320 | reg_state = kmap_atomic(page); | |
321 | ||
322 | reg_state[CTX_RING_TAIL+1] = tail; | |
323 | reg_state[CTX_RING_BUFFER_START+1] = i915_gem_obj_ggtt_offset(ring_obj); | |
324 | ||
325 | kunmap_atomic(reg_state); | |
326 | ||
327 | return 0; | |
328 | } | |
329 | ||
330 | static void execlists_submit_contexts(struct intel_engine_cs *ring, | |
331 | struct intel_context *to0, u32 tail0, | |
332 | struct intel_context *to1, u32 tail1) | |
333 | { | |
334 | struct drm_i915_gem_object *ctx_obj0 = to0->engine[ring->id].state; | |
335 | struct intel_ringbuffer *ringbuf0 = to0->engine[ring->id].ringbuf; | |
336 | struct drm_i915_gem_object *ctx_obj1 = NULL; | |
337 | struct intel_ringbuffer *ringbuf1 = NULL; | |
338 | ||
339 | BUG_ON(!ctx_obj0); | |
340 | WARN_ON(!i915_gem_obj_is_pinned(ctx_obj0)); | |
341 | WARN_ON(!i915_gem_obj_is_pinned(ringbuf0->obj)); | |
342 | ||
343 | execlists_update_context(ctx_obj0, ringbuf0->obj, tail0); | |
344 | ||
345 | if (to1) { | |
346 | ringbuf1 = to1->engine[ring->id].ringbuf; | |
347 | ctx_obj1 = to1->engine[ring->id].state; | |
348 | BUG_ON(!ctx_obj1); | |
349 | WARN_ON(!i915_gem_obj_is_pinned(ctx_obj1)); | |
350 | WARN_ON(!i915_gem_obj_is_pinned(ringbuf1->obj)); | |
351 | ||
352 | execlists_update_context(ctx_obj1, ringbuf1->obj, tail1); | |
353 | } | |
354 | ||
355 | execlists_elsp_write(ring, ctx_obj0, ctx_obj1); | |
356 | } | |
357 | ||
358 | static void execlists_context_unqueue(struct intel_engine_cs *ring) | |
359 | { | |
360 | struct drm_i915_gem_request *req0 = NULL, *req1 = NULL; | |
361 | struct drm_i915_gem_request *cursor = NULL, *tmp = NULL; | |
362 | ||
363 | assert_spin_locked(&ring->execlist_lock); | |
364 | ||
365 | if (list_empty(&ring->execlist_queue)) | |
366 | return; | |
367 | ||
368 | /* Try to read in pairs */ | |
369 | list_for_each_entry_safe(cursor, tmp, &ring->execlist_queue, | |
370 | execlist_link) { | |
371 | if (!req0) { | |
372 | req0 = cursor; | |
373 | } else if (req0->ctx == cursor->ctx) { | |
374 | /* Same ctx: ignore first request, as second request | |
375 | * will update tail past first request's workload */ | |
376 | cursor->elsp_submitted = req0->elsp_submitted; | |
377 | list_del(&req0->execlist_link); | |
378 | list_add_tail(&req0->execlist_link, | |
379 | &ring->execlist_retired_req_list); | |
380 | req0 = cursor; | |
381 | } else { | |
382 | req1 = cursor; | |
383 | break; | |
384 | } | |
385 | } | |
386 | ||
387 | WARN_ON(req1 && req1->elsp_submitted); | |
388 | ||
389 | execlists_submit_contexts(ring, req0->ctx, req0->tail, | |
390 | req1 ? req1->ctx : NULL, | |
391 | req1 ? req1->tail : 0); | |
392 | ||
393 | req0->elsp_submitted++; | |
394 | if (req1) | |
395 | req1->elsp_submitted++; | |
396 | } | |
397 | ||
398 | static bool execlists_check_remove_request(struct intel_engine_cs *ring, | |
399 | u32 request_id) | |
400 | { | |
401 | struct drm_i915_gem_request *head_req; | |
402 | ||
403 | assert_spin_locked(&ring->execlist_lock); | |
404 | ||
405 | head_req = list_first_entry_or_null(&ring->execlist_queue, | |
406 | struct drm_i915_gem_request, | |
407 | execlist_link); | |
408 | ||
409 | if (head_req != NULL) { | |
410 | struct drm_i915_gem_object *ctx_obj = | |
411 | head_req->ctx->engine[ring->id].state; | |
412 | if (intel_execlists_ctx_id(ctx_obj) == request_id) { | |
413 | WARN(head_req->elsp_submitted == 0, | |
414 | "Never submitted head request\n"); | |
415 | ||
416 | if (--head_req->elsp_submitted <= 0) { | |
417 | list_del(&head_req->execlist_link); | |
418 | list_add_tail(&head_req->execlist_link, | |
419 | &ring->execlist_retired_req_list); | |
420 | return true; | |
421 | } | |
422 | } | |
423 | } | |
424 | ||
425 | return false; | |
426 | } | |
427 | ||
428 | /** | |
429 | * intel_lrc_irq_handler() - handle Context Switch interrupts | |
430 | * @ring: Engine Command Streamer to handle. | |
431 | * | |
432 | * Check the unread Context Status Buffers and manage the submission of new | |
433 | * contexts to the ELSP accordingly. | |
434 | */ | |
435 | void intel_lrc_irq_handler(struct intel_engine_cs *ring) | |
436 | { | |
437 | struct drm_i915_private *dev_priv = ring->dev->dev_private; | |
438 | u32 status_pointer; | |
439 | u8 read_pointer; | |
440 | u8 write_pointer; | |
441 | u32 status; | |
442 | u32 status_id; | |
443 | u32 submit_contexts = 0; | |
444 | ||
445 | status_pointer = I915_READ(RING_CONTEXT_STATUS_PTR(ring)); | |
446 | ||
447 | read_pointer = ring->next_context_status_buffer; | |
448 | write_pointer = status_pointer & 0x07; | |
449 | if (read_pointer > write_pointer) | |
450 | write_pointer += 6; | |
451 | ||
452 | spin_lock(&ring->execlist_lock); | |
453 | ||
454 | while (read_pointer < write_pointer) { | |
455 | read_pointer++; | |
456 | status = I915_READ(RING_CONTEXT_STATUS_BUF(ring) + | |
457 | (read_pointer % 6) * 8); | |
458 | status_id = I915_READ(RING_CONTEXT_STATUS_BUF(ring) + | |
459 | (read_pointer % 6) * 8 + 4); | |
460 | ||
461 | if (status & GEN8_CTX_STATUS_PREEMPTED) { | |
462 | if (status & GEN8_CTX_STATUS_LITE_RESTORE) { | |
463 | if (execlists_check_remove_request(ring, status_id)) | |
464 | WARN(1, "Lite Restored request removed from queue\n"); | |
465 | } else | |
466 | WARN(1, "Preemption without Lite Restore\n"); | |
467 | } | |
468 | ||
469 | if ((status & GEN8_CTX_STATUS_ACTIVE_IDLE) || | |
470 | (status & GEN8_CTX_STATUS_ELEMENT_SWITCH)) { | |
471 | if (execlists_check_remove_request(ring, status_id)) | |
472 | submit_contexts++; | |
473 | } | |
474 | } | |
475 | ||
476 | if (submit_contexts != 0) | |
477 | execlists_context_unqueue(ring); | |
478 | ||
479 | spin_unlock(&ring->execlist_lock); | |
480 | ||
481 | WARN(submit_contexts > 2, "More than two context complete events?\n"); | |
482 | ring->next_context_status_buffer = write_pointer % 6; | |
483 | ||
484 | I915_WRITE(RING_CONTEXT_STATUS_PTR(ring), | |
485 | ((u32)ring->next_context_status_buffer & 0x07) << 8); | |
486 | } | |
487 | ||
488 | static int execlists_context_queue(struct intel_engine_cs *ring, | |
489 | struct intel_context *to, | |
490 | u32 tail, | |
491 | struct drm_i915_gem_request *request) | |
492 | { | |
493 | struct drm_i915_gem_request *cursor; | |
494 | struct drm_i915_private *dev_priv = ring->dev->dev_private; | |
495 | unsigned long flags; | |
496 | int num_elements = 0; | |
497 | ||
498 | if (to != ring->default_context) | |
499 | intel_lr_context_pin(ring, to); | |
500 | ||
501 | if (!request) { | |
502 | /* | |
503 | * If there isn't a request associated with this submission, | |
504 | * create one as a temporary holder. | |
505 | */ | |
506 | WARN(1, "execlist context submission without request"); | |
507 | request = kzalloc(sizeof(*request), GFP_KERNEL); | |
508 | if (request == NULL) | |
509 | return -ENOMEM; | |
510 | request->ring = ring; | |
511 | request->ctx = to; | |
512 | } else { | |
513 | WARN_ON(to != request->ctx); | |
514 | } | |
515 | request->tail = tail; | |
516 | i915_gem_request_reference(request); | |
517 | i915_gem_context_reference(request->ctx); | |
518 | ||
519 | intel_runtime_pm_get(dev_priv); | |
520 | ||
521 | spin_lock_irqsave(&ring->execlist_lock, flags); | |
522 | ||
523 | list_for_each_entry(cursor, &ring->execlist_queue, execlist_link) | |
524 | if (++num_elements > 2) | |
525 | break; | |
526 | ||
527 | if (num_elements > 2) { | |
528 | struct drm_i915_gem_request *tail_req; | |
529 | ||
530 | tail_req = list_last_entry(&ring->execlist_queue, | |
531 | struct drm_i915_gem_request, | |
532 | execlist_link); | |
533 | ||
534 | if (to == tail_req->ctx) { | |
535 | WARN(tail_req->elsp_submitted != 0, | |
536 | "More than 2 already-submitted reqs queued\n"); | |
537 | list_del(&tail_req->execlist_link); | |
538 | list_add_tail(&tail_req->execlist_link, | |
539 | &ring->execlist_retired_req_list); | |
540 | } | |
541 | } | |
542 | ||
543 | list_add_tail(&request->execlist_link, &ring->execlist_queue); | |
544 | if (num_elements == 0) | |
545 | execlists_context_unqueue(ring); | |
546 | ||
547 | spin_unlock_irqrestore(&ring->execlist_lock, flags); | |
548 | ||
549 | return 0; | |
550 | } | |
551 | ||
552 | static int logical_ring_invalidate_all_caches(struct intel_ringbuffer *ringbuf, | |
553 | struct intel_context *ctx) | |
554 | { | |
555 | struct intel_engine_cs *ring = ringbuf->ring; | |
556 | uint32_t flush_domains; | |
557 | int ret; | |
558 | ||
559 | flush_domains = 0; | |
560 | if (ring->gpu_caches_dirty) | |
561 | flush_domains = I915_GEM_GPU_DOMAINS; | |
562 | ||
563 | ret = ring->emit_flush(ringbuf, ctx, | |
564 | I915_GEM_GPU_DOMAINS, flush_domains); | |
565 | if (ret) | |
566 | return ret; | |
567 | ||
568 | ring->gpu_caches_dirty = false; | |
569 | return 0; | |
570 | } | |
571 | ||
572 | static int execlists_move_to_gpu(struct intel_ringbuffer *ringbuf, | |
573 | struct intel_context *ctx, | |
574 | struct list_head *vmas) | |
575 | { | |
576 | struct intel_engine_cs *ring = ringbuf->ring; | |
577 | struct i915_vma *vma; | |
578 | uint32_t flush_domains = 0; | |
579 | bool flush_chipset = false; | |
580 | int ret; | |
581 | ||
582 | list_for_each_entry(vma, vmas, exec_list) { | |
583 | struct drm_i915_gem_object *obj = vma->obj; | |
584 | ||
585 | ret = i915_gem_object_sync(obj, ring); | |
586 | if (ret) | |
587 | return ret; | |
588 | ||
589 | if (obj->base.write_domain & I915_GEM_DOMAIN_CPU) | |
590 | flush_chipset |= i915_gem_clflush_object(obj, false); | |
591 | ||
592 | flush_domains |= obj->base.write_domain; | |
593 | } | |
594 | ||
595 | if (flush_domains & I915_GEM_DOMAIN_GTT) | |
596 | wmb(); | |
597 | ||
598 | /* Unconditionally invalidate gpu caches and ensure that we do flush | |
599 | * any residual writes from the previous batch. | |
600 | */ | |
601 | return logical_ring_invalidate_all_caches(ringbuf, ctx); | |
602 | } | |
603 | ||
604 | /** | |
605 | * execlists_submission() - submit a batchbuffer for execution, Execlists style | |
606 | * @dev: DRM device. | |
607 | * @file: DRM file. | |
608 | * @ring: Engine Command Streamer to submit to. | |
609 | * @ctx: Context to employ for this submission. | |
610 | * @args: execbuffer call arguments. | |
611 | * @vmas: list of vmas. | |
612 | * @batch_obj: the batchbuffer to submit. | |
613 | * @exec_start: batchbuffer start virtual address pointer. | |
614 | * @flags: translated execbuffer call flags. | |
615 | * | |
616 | * This is the evil twin version of i915_gem_ringbuffer_submission. It abstracts | |
617 | * away the submission details of the execbuffer ioctl call. | |
618 | * | |
619 | * Return: non-zero if the submission fails. | |
620 | */ | |
621 | int intel_execlists_submission(struct drm_device *dev, struct drm_file *file, | |
622 | struct intel_engine_cs *ring, | |
623 | struct intel_context *ctx, | |
624 | struct drm_i915_gem_execbuffer2 *args, | |
625 | struct list_head *vmas, | |
626 | struct drm_i915_gem_object *batch_obj, | |
627 | u64 exec_start, u32 flags) | |
628 | { | |
629 | struct drm_i915_private *dev_priv = dev->dev_private; | |
630 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; | |
631 | int instp_mode; | |
632 | u32 instp_mask; | |
633 | int ret; | |
634 | ||
635 | instp_mode = args->flags & I915_EXEC_CONSTANTS_MASK; | |
636 | instp_mask = I915_EXEC_CONSTANTS_MASK; | |
637 | switch (instp_mode) { | |
638 | case I915_EXEC_CONSTANTS_REL_GENERAL: | |
639 | case I915_EXEC_CONSTANTS_ABSOLUTE: | |
640 | case I915_EXEC_CONSTANTS_REL_SURFACE: | |
641 | if (instp_mode != 0 && ring != &dev_priv->ring[RCS]) { | |
642 | DRM_DEBUG("non-0 rel constants mode on non-RCS\n"); | |
643 | return -EINVAL; | |
644 | } | |
645 | ||
646 | if (instp_mode != dev_priv->relative_constants_mode) { | |
647 | if (instp_mode == I915_EXEC_CONSTANTS_REL_SURFACE) { | |
648 | DRM_DEBUG("rel surface constants mode invalid on gen5+\n"); | |
649 | return -EINVAL; | |
650 | } | |
651 | ||
652 | /* The HW changed the meaning on this bit on gen6 */ | |
653 | instp_mask &= ~I915_EXEC_CONSTANTS_REL_SURFACE; | |
654 | } | |
655 | break; | |
656 | default: | |
657 | DRM_DEBUG("execbuf with unknown constants: %d\n", instp_mode); | |
658 | return -EINVAL; | |
659 | } | |
660 | ||
661 | if (args->num_cliprects != 0) { | |
662 | DRM_DEBUG("clip rectangles are only valid on pre-gen5\n"); | |
663 | return -EINVAL; | |
664 | } else { | |
665 | if (args->DR4 == 0xffffffff) { | |
666 | DRM_DEBUG("UXA submitting garbage DR4, fixing up\n"); | |
667 | args->DR4 = 0; | |
668 | } | |
669 | ||
670 | if (args->DR1 || args->DR4 || args->cliprects_ptr) { | |
671 | DRM_DEBUG("0 cliprects but dirt in cliprects fields\n"); | |
672 | return -EINVAL; | |
673 | } | |
674 | } | |
675 | ||
676 | if (args->flags & I915_EXEC_GEN7_SOL_RESET) { | |
677 | DRM_DEBUG("sol reset is gen7 only\n"); | |
678 | return -EINVAL; | |
679 | } | |
680 | ||
681 | ret = execlists_move_to_gpu(ringbuf, ctx, vmas); | |
682 | if (ret) | |
683 | return ret; | |
684 | ||
685 | if (ring == &dev_priv->ring[RCS] && | |
686 | instp_mode != dev_priv->relative_constants_mode) { | |
687 | ret = intel_logical_ring_begin(ringbuf, ctx, 4); | |
688 | if (ret) | |
689 | return ret; | |
690 | ||
691 | intel_logical_ring_emit(ringbuf, MI_NOOP); | |
692 | intel_logical_ring_emit(ringbuf, MI_LOAD_REGISTER_IMM(1)); | |
693 | intel_logical_ring_emit(ringbuf, INSTPM); | |
694 | intel_logical_ring_emit(ringbuf, instp_mask << 16 | instp_mode); | |
695 | intel_logical_ring_advance(ringbuf); | |
696 | ||
697 | dev_priv->relative_constants_mode = instp_mode; | |
698 | } | |
699 | ||
700 | ret = ring->emit_bb_start(ringbuf, ctx, exec_start, flags); | |
701 | if (ret) | |
702 | return ret; | |
703 | ||
704 | i915_gem_execbuffer_move_to_active(vmas, ring); | |
705 | i915_gem_execbuffer_retire_commands(dev, file, ring, batch_obj); | |
706 | ||
707 | return 0; | |
708 | } | |
709 | ||
710 | void intel_execlists_retire_requests(struct intel_engine_cs *ring) | |
711 | { | |
712 | struct drm_i915_gem_request *req, *tmp; | |
713 | struct drm_i915_private *dev_priv = ring->dev->dev_private; | |
714 | unsigned long flags; | |
715 | struct list_head retired_list; | |
716 | ||
717 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); | |
718 | if (list_empty(&ring->execlist_retired_req_list)) | |
719 | return; | |
720 | ||
721 | INIT_LIST_HEAD(&retired_list); | |
722 | spin_lock_irqsave(&ring->execlist_lock, flags); | |
723 | list_replace_init(&ring->execlist_retired_req_list, &retired_list); | |
724 | spin_unlock_irqrestore(&ring->execlist_lock, flags); | |
725 | ||
726 | list_for_each_entry_safe(req, tmp, &retired_list, execlist_link) { | |
727 | struct intel_context *ctx = req->ctx; | |
728 | struct drm_i915_gem_object *ctx_obj = | |
729 | ctx->engine[ring->id].state; | |
730 | ||
731 | if (ctx_obj && (ctx != ring->default_context)) | |
732 | intel_lr_context_unpin(ring, ctx); | |
733 | intel_runtime_pm_put(dev_priv); | |
734 | i915_gem_context_unreference(ctx); | |
735 | list_del(&req->execlist_link); | |
736 | i915_gem_request_unreference(req); | |
737 | } | |
738 | } | |
739 | ||
740 | void intel_logical_ring_stop(struct intel_engine_cs *ring) | |
741 | { | |
742 | struct drm_i915_private *dev_priv = ring->dev->dev_private; | |
743 | int ret; | |
744 | ||
745 | if (!intel_ring_initialized(ring)) | |
746 | return; | |
747 | ||
748 | ret = intel_ring_idle(ring); | |
749 | if (ret && !i915_reset_in_progress(&to_i915(ring->dev)->gpu_error)) | |
750 | DRM_ERROR("failed to quiesce %s whilst cleaning up: %d\n", | |
751 | ring->name, ret); | |
752 | ||
753 | /* TODO: Is this correct with Execlists enabled? */ | |
754 | I915_WRITE_MODE(ring, _MASKED_BIT_ENABLE(STOP_RING)); | |
755 | if (wait_for_atomic((I915_READ_MODE(ring) & MODE_IDLE) != 0, 1000)) { | |
756 | DRM_ERROR("%s :timed out trying to stop ring\n", ring->name); | |
757 | return; | |
758 | } | |
759 | I915_WRITE_MODE(ring, _MASKED_BIT_DISABLE(STOP_RING)); | |
760 | } | |
761 | ||
762 | int logical_ring_flush_all_caches(struct intel_ringbuffer *ringbuf, | |
763 | struct intel_context *ctx) | |
764 | { | |
765 | struct intel_engine_cs *ring = ringbuf->ring; | |
766 | int ret; | |
767 | ||
768 | if (!ring->gpu_caches_dirty) | |
769 | return 0; | |
770 | ||
771 | ret = ring->emit_flush(ringbuf, ctx, 0, I915_GEM_GPU_DOMAINS); | |
772 | if (ret) | |
773 | return ret; | |
774 | ||
775 | ring->gpu_caches_dirty = false; | |
776 | return 0; | |
777 | } | |
778 | ||
779 | /** | |
780 | * intel_logical_ring_advance_and_submit() - advance the tail and submit the workload | |
781 | * @ringbuf: Logical Ringbuffer to advance. | |
782 | * | |
783 | * The tail is updated in our logical ringbuffer struct, not in the actual context. What | |
784 | * really happens during submission is that the context and current tail will be placed | |
785 | * on a queue waiting for the ELSP to be ready to accept a new context submission. At that | |
786 | * point, the tail *inside* the context is updated and the ELSP written to. | |
787 | */ | |
788 | void intel_logical_ring_advance_and_submit(struct intel_ringbuffer *ringbuf, | |
789 | struct intel_context *ctx, | |
790 | struct drm_i915_gem_request *request) | |
791 | { | |
792 | struct intel_engine_cs *ring = ringbuf->ring; | |
793 | ||
794 | intel_logical_ring_advance(ringbuf); | |
795 | ||
796 | if (intel_ring_stopped(ring)) | |
797 | return; | |
798 | ||
799 | execlists_context_queue(ring, ctx, ringbuf->tail, request); | |
800 | } | |
801 | ||
802 | static int intel_lr_context_pin(struct intel_engine_cs *ring, | |
803 | struct intel_context *ctx) | |
804 | { | |
805 | struct drm_i915_gem_object *ctx_obj = ctx->engine[ring->id].state; | |
806 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; | |
807 | int ret = 0; | |
808 | ||
809 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); | |
810 | if (ctx->engine[ring->id].pin_count++ == 0) { | |
811 | ret = i915_gem_obj_ggtt_pin(ctx_obj, | |
812 | GEN8_LR_CONTEXT_ALIGN, 0); | |
813 | if (ret) | |
814 | goto reset_pin_count; | |
815 | ||
816 | ret = intel_pin_and_map_ringbuffer_obj(ring->dev, ringbuf); | |
817 | if (ret) | |
818 | goto unpin_ctx_obj; | |
819 | } | |
820 | ||
821 | return ret; | |
822 | ||
823 | unpin_ctx_obj: | |
824 | i915_gem_object_ggtt_unpin(ctx_obj); | |
825 | reset_pin_count: | |
826 | ctx->engine[ring->id].pin_count = 0; | |
827 | ||
828 | return ret; | |
829 | } | |
830 | ||
831 | void intel_lr_context_unpin(struct intel_engine_cs *ring, | |
832 | struct intel_context *ctx) | |
833 | { | |
834 | struct drm_i915_gem_object *ctx_obj = ctx->engine[ring->id].state; | |
835 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; | |
836 | ||
837 | if (ctx_obj) { | |
838 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); | |
839 | if (--ctx->engine[ring->id].pin_count == 0) { | |
840 | intel_unpin_ringbuffer_obj(ringbuf); | |
841 | i915_gem_object_ggtt_unpin(ctx_obj); | |
842 | } | |
843 | } | |
844 | } | |
845 | ||
846 | static int logical_ring_alloc_request(struct intel_engine_cs *ring, | |
847 | struct intel_context *ctx) | |
848 | { | |
849 | struct drm_i915_gem_request *request; | |
850 | struct drm_i915_private *dev_private = ring->dev->dev_private; | |
851 | int ret; | |
852 | ||
853 | if (ring->outstanding_lazy_request) | |
854 | return 0; | |
855 | ||
856 | request = kzalloc(sizeof(*request), GFP_KERNEL); | |
857 | if (request == NULL) | |
858 | return -ENOMEM; | |
859 | ||
860 | if (ctx != ring->default_context) { | |
861 | ret = intel_lr_context_pin(ring, ctx); | |
862 | if (ret) { | |
863 | kfree(request); | |
864 | return ret; | |
865 | } | |
866 | } | |
867 | ||
868 | kref_init(&request->ref); | |
869 | request->ring = ring; | |
870 | request->uniq = dev_private->request_uniq++; | |
871 | ||
872 | ret = i915_gem_get_seqno(ring->dev, &request->seqno); | |
873 | if (ret) { | |
874 | intel_lr_context_unpin(ring, ctx); | |
875 | kfree(request); | |
876 | return ret; | |
877 | } | |
878 | ||
879 | /* Hold a reference to the context this request belongs to | |
880 | * (we will need it when the time comes to emit/retire the | |
881 | * request). | |
882 | */ | |
883 | request->ctx = ctx; | |
884 | i915_gem_context_reference(request->ctx); | |
885 | ||
886 | ring->outstanding_lazy_request = request; | |
887 | return 0; | |
888 | } | |
889 | ||
890 | static int logical_ring_wait_request(struct intel_ringbuffer *ringbuf, | |
891 | int bytes) | |
892 | { | |
893 | struct intel_engine_cs *ring = ringbuf->ring; | |
894 | struct drm_i915_gem_request *request; | |
895 | int ret; | |
896 | ||
897 | if (intel_ring_space(ringbuf) >= bytes) | |
898 | return 0; | |
899 | ||
900 | list_for_each_entry(request, &ring->request_list, list) { | |
901 | /* | |
902 | * The request queue is per-engine, so can contain requests | |
903 | * from multiple ringbuffers. Here, we must ignore any that | |
904 | * aren't from the ringbuffer we're considering. | |
905 | */ | |
906 | struct intel_context *ctx = request->ctx; | |
907 | if (ctx->engine[ring->id].ringbuf != ringbuf) | |
908 | continue; | |
909 | ||
910 | /* Would completion of this request free enough space? */ | |
911 | if (__intel_ring_space(request->tail, ringbuf->tail, | |
912 | ringbuf->size) >= bytes) { | |
913 | break; | |
914 | } | |
915 | } | |
916 | ||
917 | if (&request->list == &ring->request_list) | |
918 | return -ENOSPC; | |
919 | ||
920 | ret = i915_wait_request(request); | |
921 | if (ret) | |
922 | return ret; | |
923 | ||
924 | i915_gem_retire_requests_ring(ring); | |
925 | ||
926 | return intel_ring_space(ringbuf) >= bytes ? 0 : -ENOSPC; | |
927 | } | |
928 | ||
929 | static int logical_ring_wait_for_space(struct intel_ringbuffer *ringbuf, | |
930 | struct intel_context *ctx, | |
931 | int bytes) | |
932 | { | |
933 | struct intel_engine_cs *ring = ringbuf->ring; | |
934 | struct drm_device *dev = ring->dev; | |
935 | struct drm_i915_private *dev_priv = dev->dev_private; | |
936 | unsigned long end; | |
937 | int ret; | |
938 | ||
939 | ret = logical_ring_wait_request(ringbuf, bytes); | |
940 | if (ret != -ENOSPC) | |
941 | return ret; | |
942 | ||
943 | /* Force the context submission in case we have been skipping it */ | |
944 | intel_logical_ring_advance_and_submit(ringbuf, ctx, NULL); | |
945 | ||
946 | /* With GEM the hangcheck timer should kick us out of the loop, | |
947 | * leaving it early runs the risk of corrupting GEM state (due | |
948 | * to running on almost untested codepaths). But on resume | |
949 | * timers don't work yet, so prevent a complete hang in that | |
950 | * case by choosing an insanely large timeout. */ | |
951 | end = jiffies + 60 * HZ; | |
952 | ||
953 | ret = 0; | |
954 | do { | |
955 | if (intel_ring_space(ringbuf) >= bytes) | |
956 | break; | |
957 | ||
958 | msleep(1); | |
959 | ||
960 | if (dev_priv->mm.interruptible && signal_pending(current)) { | |
961 | ret = -ERESTARTSYS; | |
962 | break; | |
963 | } | |
964 | ||
965 | ret = i915_gem_check_wedge(&dev_priv->gpu_error, | |
966 | dev_priv->mm.interruptible); | |
967 | if (ret) | |
968 | break; | |
969 | ||
970 | if (time_after(jiffies, end)) { | |
971 | ret = -EBUSY; | |
972 | break; | |
973 | } | |
974 | } while (1); | |
975 | ||
976 | return ret; | |
977 | } | |
978 | ||
979 | static int logical_ring_wrap_buffer(struct intel_ringbuffer *ringbuf, | |
980 | struct intel_context *ctx) | |
981 | { | |
982 | uint32_t __iomem *virt; | |
983 | int rem = ringbuf->size - ringbuf->tail; | |
984 | ||
985 | if (ringbuf->space < rem) { | |
986 | int ret = logical_ring_wait_for_space(ringbuf, ctx, rem); | |
987 | ||
988 | if (ret) | |
989 | return ret; | |
990 | } | |
991 | ||
992 | virt = ringbuf->virtual_start + ringbuf->tail; | |
993 | rem /= 4; | |
994 | while (rem--) | |
995 | iowrite32(MI_NOOP, virt++); | |
996 | ||
997 | ringbuf->tail = 0; | |
998 | intel_ring_update_space(ringbuf); | |
999 | ||
1000 | return 0; | |
1001 | } | |
1002 | ||
1003 | static int logical_ring_prepare(struct intel_ringbuffer *ringbuf, | |
1004 | struct intel_context *ctx, int bytes) | |
1005 | { | |
1006 | int ret; | |
1007 | ||
1008 | if (unlikely(ringbuf->tail + bytes > ringbuf->effective_size)) { | |
1009 | ret = logical_ring_wrap_buffer(ringbuf, ctx); | |
1010 | if (unlikely(ret)) | |
1011 | return ret; | |
1012 | } | |
1013 | ||
1014 | if (unlikely(ringbuf->space < bytes)) { | |
1015 | ret = logical_ring_wait_for_space(ringbuf, ctx, bytes); | |
1016 | if (unlikely(ret)) | |
1017 | return ret; | |
1018 | } | |
1019 | ||
1020 | return 0; | |
1021 | } | |
1022 | ||
1023 | /** | |
1024 | * intel_logical_ring_begin() - prepare the logical ringbuffer to accept some commands | |
1025 | * | |
1026 | * @ringbuf: Logical ringbuffer. | |
1027 | * @num_dwords: number of DWORDs that we plan to write to the ringbuffer. | |
1028 | * | |
1029 | * The ringbuffer might not be ready to accept the commands right away (maybe it needs to | |
1030 | * be wrapped, or wait a bit for the tail to be updated). This function takes care of that | |
1031 | * and also preallocates a request (every workload submission is still mediated through | |
1032 | * requests, same as it did with legacy ringbuffer submission). | |
1033 | * | |
1034 | * Return: non-zero if the ringbuffer is not ready to be written to. | |
1035 | */ | |
1036 | int intel_logical_ring_begin(struct intel_ringbuffer *ringbuf, | |
1037 | struct intel_context *ctx, int num_dwords) | |
1038 | { | |
1039 | struct intel_engine_cs *ring = ringbuf->ring; | |
1040 | struct drm_device *dev = ring->dev; | |
1041 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1042 | int ret; | |
1043 | ||
1044 | ret = i915_gem_check_wedge(&dev_priv->gpu_error, | |
1045 | dev_priv->mm.interruptible); | |
1046 | if (ret) | |
1047 | return ret; | |
1048 | ||
1049 | ret = logical_ring_prepare(ringbuf, ctx, num_dwords * sizeof(uint32_t)); | |
1050 | if (ret) | |
1051 | return ret; | |
1052 | ||
1053 | /* Preallocate the olr before touching the ring */ | |
1054 | ret = logical_ring_alloc_request(ring, ctx); | |
1055 | if (ret) | |
1056 | return ret; | |
1057 | ||
1058 | ringbuf->space -= num_dwords * sizeof(uint32_t); | |
1059 | return 0; | |
1060 | } | |
1061 | ||
1062 | static int intel_logical_ring_workarounds_emit(struct intel_engine_cs *ring, | |
1063 | struct intel_context *ctx) | |
1064 | { | |
1065 | int ret, i; | |
1066 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; | |
1067 | struct drm_device *dev = ring->dev; | |
1068 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1069 | struct i915_workarounds *w = &dev_priv->workarounds; | |
1070 | ||
1071 | if (WARN_ON_ONCE(w->count == 0)) | |
1072 | return 0; | |
1073 | ||
1074 | ring->gpu_caches_dirty = true; | |
1075 | ret = logical_ring_flush_all_caches(ringbuf, ctx); | |
1076 | if (ret) | |
1077 | return ret; | |
1078 | ||
1079 | ret = intel_logical_ring_begin(ringbuf, ctx, w->count * 2 + 2); | |
1080 | if (ret) | |
1081 | return ret; | |
1082 | ||
1083 | intel_logical_ring_emit(ringbuf, MI_LOAD_REGISTER_IMM(w->count)); | |
1084 | for (i = 0; i < w->count; i++) { | |
1085 | intel_logical_ring_emit(ringbuf, w->reg[i].addr); | |
1086 | intel_logical_ring_emit(ringbuf, w->reg[i].value); | |
1087 | } | |
1088 | intel_logical_ring_emit(ringbuf, MI_NOOP); | |
1089 | ||
1090 | intel_logical_ring_advance(ringbuf); | |
1091 | ||
1092 | ring->gpu_caches_dirty = true; | |
1093 | ret = logical_ring_flush_all_caches(ringbuf, ctx); | |
1094 | if (ret) | |
1095 | return ret; | |
1096 | ||
1097 | return 0; | |
1098 | } | |
1099 | ||
1100 | static int gen8_init_common_ring(struct intel_engine_cs *ring) | |
1101 | { | |
1102 | struct drm_device *dev = ring->dev; | |
1103 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1104 | ||
1105 | I915_WRITE_IMR(ring, ~(ring->irq_enable_mask | ring->irq_keep_mask)); | |
1106 | I915_WRITE(RING_HWSTAM(ring->mmio_base), 0xffffffff); | |
1107 | ||
1108 | I915_WRITE(RING_MODE_GEN7(ring), | |
1109 | _MASKED_BIT_DISABLE(GFX_REPLAY_MODE) | | |
1110 | _MASKED_BIT_ENABLE(GFX_RUN_LIST_ENABLE)); | |
1111 | POSTING_READ(RING_MODE_GEN7(ring)); | |
1112 | ring->next_context_status_buffer = 0; | |
1113 | DRM_DEBUG_DRIVER("Execlists enabled for %s\n", ring->name); | |
1114 | ||
1115 | memset(&ring->hangcheck, 0, sizeof(ring->hangcheck)); | |
1116 | ||
1117 | return 0; | |
1118 | } | |
1119 | ||
1120 | static int gen8_init_render_ring(struct intel_engine_cs *ring) | |
1121 | { | |
1122 | struct drm_device *dev = ring->dev; | |
1123 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1124 | int ret; | |
1125 | ||
1126 | ret = gen8_init_common_ring(ring); | |
1127 | if (ret) | |
1128 | return ret; | |
1129 | ||
1130 | /* We need to disable the AsyncFlip performance optimisations in order | |
1131 | * to use MI_WAIT_FOR_EVENT within the CS. It should already be | |
1132 | * programmed to '1' on all products. | |
1133 | * | |
1134 | * WaDisableAsyncFlipPerfMode:snb,ivb,hsw,vlv,bdw,chv | |
1135 | */ | |
1136 | I915_WRITE(MI_MODE, _MASKED_BIT_ENABLE(ASYNC_FLIP_PERF_DISABLE)); | |
1137 | ||
1138 | I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_FORCE_ORDERING)); | |
1139 | ||
1140 | return init_workarounds_ring(ring); | |
1141 | } | |
1142 | ||
1143 | static int gen8_emit_bb_start(struct intel_ringbuffer *ringbuf, | |
1144 | struct intel_context *ctx, | |
1145 | u64 offset, unsigned flags) | |
1146 | { | |
1147 | bool ppgtt = !(flags & I915_DISPATCH_SECURE); | |
1148 | int ret; | |
1149 | ||
1150 | ret = intel_logical_ring_begin(ringbuf, ctx, 4); | |
1151 | if (ret) | |
1152 | return ret; | |
1153 | ||
1154 | /* FIXME(BDW): Address space and security selectors. */ | |
1155 | intel_logical_ring_emit(ringbuf, MI_BATCH_BUFFER_START_GEN8 | (ppgtt<<8)); | |
1156 | intel_logical_ring_emit(ringbuf, lower_32_bits(offset)); | |
1157 | intel_logical_ring_emit(ringbuf, upper_32_bits(offset)); | |
1158 | intel_logical_ring_emit(ringbuf, MI_NOOP); | |
1159 | intel_logical_ring_advance(ringbuf); | |
1160 | ||
1161 | return 0; | |
1162 | } | |
1163 | ||
1164 | static bool gen8_logical_ring_get_irq(struct intel_engine_cs *ring) | |
1165 | { | |
1166 | struct drm_device *dev = ring->dev; | |
1167 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1168 | unsigned long flags; | |
1169 | ||
1170 | if (WARN_ON(!intel_irqs_enabled(dev_priv))) | |
1171 | return false; | |
1172 | ||
1173 | spin_lock_irqsave(&dev_priv->irq_lock, flags); | |
1174 | if (ring->irq_refcount++ == 0) { | |
1175 | I915_WRITE_IMR(ring, ~(ring->irq_enable_mask | ring->irq_keep_mask)); | |
1176 | POSTING_READ(RING_IMR(ring->mmio_base)); | |
1177 | } | |
1178 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); | |
1179 | ||
1180 | return true; | |
1181 | } | |
1182 | ||
1183 | static void gen8_logical_ring_put_irq(struct intel_engine_cs *ring) | |
1184 | { | |
1185 | struct drm_device *dev = ring->dev; | |
1186 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1187 | unsigned long flags; | |
1188 | ||
1189 | spin_lock_irqsave(&dev_priv->irq_lock, flags); | |
1190 | if (--ring->irq_refcount == 0) { | |
1191 | I915_WRITE_IMR(ring, ~ring->irq_keep_mask); | |
1192 | POSTING_READ(RING_IMR(ring->mmio_base)); | |
1193 | } | |
1194 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); | |
1195 | } | |
1196 | ||
1197 | static int gen8_emit_flush(struct intel_ringbuffer *ringbuf, | |
1198 | struct intel_context *ctx, | |
1199 | u32 invalidate_domains, | |
1200 | u32 unused) | |
1201 | { | |
1202 | struct intel_engine_cs *ring = ringbuf->ring; | |
1203 | struct drm_device *dev = ring->dev; | |
1204 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1205 | uint32_t cmd; | |
1206 | int ret; | |
1207 | ||
1208 | ret = intel_logical_ring_begin(ringbuf, ctx, 4); | |
1209 | if (ret) | |
1210 | return ret; | |
1211 | ||
1212 | cmd = MI_FLUSH_DW + 1; | |
1213 | ||
1214 | if (ring == &dev_priv->ring[VCS]) { | |
1215 | if (invalidate_domains & I915_GEM_GPU_DOMAINS) | |
1216 | cmd |= MI_INVALIDATE_TLB | MI_INVALIDATE_BSD | | |
1217 | MI_FLUSH_DW_STORE_INDEX | | |
1218 | MI_FLUSH_DW_OP_STOREDW; | |
1219 | } else { | |
1220 | if (invalidate_domains & I915_GEM_DOMAIN_RENDER) | |
1221 | cmd |= MI_INVALIDATE_TLB | MI_FLUSH_DW_STORE_INDEX | | |
1222 | MI_FLUSH_DW_OP_STOREDW; | |
1223 | } | |
1224 | ||
1225 | intel_logical_ring_emit(ringbuf, cmd); | |
1226 | intel_logical_ring_emit(ringbuf, | |
1227 | I915_GEM_HWS_SCRATCH_ADDR | | |
1228 | MI_FLUSH_DW_USE_GTT); | |
1229 | intel_logical_ring_emit(ringbuf, 0); /* upper addr */ | |
1230 | intel_logical_ring_emit(ringbuf, 0); /* value */ | |
1231 | intel_logical_ring_advance(ringbuf); | |
1232 | ||
1233 | return 0; | |
1234 | } | |
1235 | ||
1236 | static int gen8_emit_flush_render(struct intel_ringbuffer *ringbuf, | |
1237 | struct intel_context *ctx, | |
1238 | u32 invalidate_domains, | |
1239 | u32 flush_domains) | |
1240 | { | |
1241 | struct intel_engine_cs *ring = ringbuf->ring; | |
1242 | u32 scratch_addr = ring->scratch.gtt_offset + 2 * CACHELINE_BYTES; | |
1243 | u32 flags = 0; | |
1244 | int ret; | |
1245 | ||
1246 | flags |= PIPE_CONTROL_CS_STALL; | |
1247 | ||
1248 | if (flush_domains) { | |
1249 | flags |= PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH; | |
1250 | flags |= PIPE_CONTROL_DEPTH_CACHE_FLUSH; | |
1251 | } | |
1252 | ||
1253 | if (invalidate_domains) { | |
1254 | flags |= PIPE_CONTROL_TLB_INVALIDATE; | |
1255 | flags |= PIPE_CONTROL_INSTRUCTION_CACHE_INVALIDATE; | |
1256 | flags |= PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE; | |
1257 | flags |= PIPE_CONTROL_VF_CACHE_INVALIDATE; | |
1258 | flags |= PIPE_CONTROL_CONST_CACHE_INVALIDATE; | |
1259 | flags |= PIPE_CONTROL_STATE_CACHE_INVALIDATE; | |
1260 | flags |= PIPE_CONTROL_QW_WRITE; | |
1261 | flags |= PIPE_CONTROL_GLOBAL_GTT_IVB; | |
1262 | } | |
1263 | ||
1264 | ret = intel_logical_ring_begin(ringbuf, ctx, 6); | |
1265 | if (ret) | |
1266 | return ret; | |
1267 | ||
1268 | intel_logical_ring_emit(ringbuf, GFX_OP_PIPE_CONTROL(6)); | |
1269 | intel_logical_ring_emit(ringbuf, flags); | |
1270 | intel_logical_ring_emit(ringbuf, scratch_addr); | |
1271 | intel_logical_ring_emit(ringbuf, 0); | |
1272 | intel_logical_ring_emit(ringbuf, 0); | |
1273 | intel_logical_ring_emit(ringbuf, 0); | |
1274 | intel_logical_ring_advance(ringbuf); | |
1275 | ||
1276 | return 0; | |
1277 | } | |
1278 | ||
1279 | static u32 gen8_get_seqno(struct intel_engine_cs *ring, bool lazy_coherency) | |
1280 | { | |
1281 | return intel_read_status_page(ring, I915_GEM_HWS_INDEX); | |
1282 | } | |
1283 | ||
1284 | static void gen8_set_seqno(struct intel_engine_cs *ring, u32 seqno) | |
1285 | { | |
1286 | intel_write_status_page(ring, I915_GEM_HWS_INDEX, seqno); | |
1287 | } | |
1288 | ||
1289 | static int gen8_emit_request(struct intel_ringbuffer *ringbuf, | |
1290 | struct drm_i915_gem_request *request) | |
1291 | { | |
1292 | struct intel_engine_cs *ring = ringbuf->ring; | |
1293 | u32 cmd; | |
1294 | int ret; | |
1295 | ||
1296 | ret = intel_logical_ring_begin(ringbuf, request->ctx, 6); | |
1297 | if (ret) | |
1298 | return ret; | |
1299 | ||
1300 | cmd = MI_STORE_DWORD_IMM_GEN4; | |
1301 | cmd |= MI_GLOBAL_GTT; | |
1302 | ||
1303 | intel_logical_ring_emit(ringbuf, cmd); | |
1304 | intel_logical_ring_emit(ringbuf, | |
1305 | (ring->status_page.gfx_addr + | |
1306 | (I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT))); | |
1307 | intel_logical_ring_emit(ringbuf, 0); | |
1308 | intel_logical_ring_emit(ringbuf, | |
1309 | i915_gem_request_get_seqno(ring->outstanding_lazy_request)); | |
1310 | intel_logical_ring_emit(ringbuf, MI_USER_INTERRUPT); | |
1311 | intel_logical_ring_emit(ringbuf, MI_NOOP); | |
1312 | intel_logical_ring_advance_and_submit(ringbuf, request->ctx, request); | |
1313 | ||
1314 | return 0; | |
1315 | } | |
1316 | ||
1317 | static int gen8_init_rcs_context(struct intel_engine_cs *ring, | |
1318 | struct intel_context *ctx) | |
1319 | { | |
1320 | int ret; | |
1321 | ||
1322 | ret = intel_logical_ring_workarounds_emit(ring, ctx); | |
1323 | if (ret) | |
1324 | return ret; | |
1325 | ||
1326 | return intel_lr_context_render_state_init(ring, ctx); | |
1327 | } | |
1328 | ||
1329 | /** | |
1330 | * intel_logical_ring_cleanup() - deallocate the Engine Command Streamer | |
1331 | * | |
1332 | * @ring: Engine Command Streamer. | |
1333 | * | |
1334 | */ | |
1335 | void intel_logical_ring_cleanup(struct intel_engine_cs *ring) | |
1336 | { | |
1337 | struct drm_i915_private *dev_priv; | |
1338 | ||
1339 | if (!intel_ring_initialized(ring)) | |
1340 | return; | |
1341 | ||
1342 | dev_priv = ring->dev->dev_private; | |
1343 | ||
1344 | intel_logical_ring_stop(ring); | |
1345 | WARN_ON((I915_READ_MODE(ring) & MODE_IDLE) == 0); | |
1346 | i915_gem_request_assign(&ring->outstanding_lazy_request, NULL); | |
1347 | ||
1348 | if (ring->cleanup) | |
1349 | ring->cleanup(ring); | |
1350 | ||
1351 | i915_cmd_parser_fini_ring(ring); | |
1352 | ||
1353 | if (ring->status_page.obj) { | |
1354 | kunmap(sg_page(ring->status_page.obj->pages->sgl)); | |
1355 | ring->status_page.obj = NULL; | |
1356 | } | |
1357 | } | |
1358 | ||
1359 | static int logical_ring_init(struct drm_device *dev, struct intel_engine_cs *ring) | |
1360 | { | |
1361 | int ret; | |
1362 | ||
1363 | /* Intentionally left blank. */ | |
1364 | ring->buffer = NULL; | |
1365 | ||
1366 | ring->dev = dev; | |
1367 | INIT_LIST_HEAD(&ring->active_list); | |
1368 | INIT_LIST_HEAD(&ring->request_list); | |
1369 | init_waitqueue_head(&ring->irq_queue); | |
1370 | ||
1371 | INIT_LIST_HEAD(&ring->execlist_queue); | |
1372 | INIT_LIST_HEAD(&ring->execlist_retired_req_list); | |
1373 | spin_lock_init(&ring->execlist_lock); | |
1374 | ||
1375 | ret = i915_cmd_parser_init_ring(ring); | |
1376 | if (ret) | |
1377 | return ret; | |
1378 | ||
1379 | ret = intel_lr_context_deferred_create(ring->default_context, ring); | |
1380 | ||
1381 | return ret; | |
1382 | } | |
1383 | ||
1384 | static int logical_render_ring_init(struct drm_device *dev) | |
1385 | { | |
1386 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1387 | struct intel_engine_cs *ring = &dev_priv->ring[RCS]; | |
1388 | int ret; | |
1389 | ||
1390 | ring->name = "render ring"; | |
1391 | ring->id = RCS; | |
1392 | ring->mmio_base = RENDER_RING_BASE; | |
1393 | ring->irq_enable_mask = | |
1394 | GT_RENDER_USER_INTERRUPT << GEN8_RCS_IRQ_SHIFT; | |
1395 | ring->irq_keep_mask = | |
1396 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_RCS_IRQ_SHIFT; | |
1397 | if (HAS_L3_DPF(dev)) | |
1398 | ring->irq_keep_mask |= GT_RENDER_L3_PARITY_ERROR_INTERRUPT; | |
1399 | ||
1400 | ring->init_hw = gen8_init_render_ring; | |
1401 | ring->init_context = gen8_init_rcs_context; | |
1402 | ring->cleanup = intel_fini_pipe_control; | |
1403 | ring->get_seqno = gen8_get_seqno; | |
1404 | ring->set_seqno = gen8_set_seqno; | |
1405 | ring->emit_request = gen8_emit_request; | |
1406 | ring->emit_flush = gen8_emit_flush_render; | |
1407 | ring->irq_get = gen8_logical_ring_get_irq; | |
1408 | ring->irq_put = gen8_logical_ring_put_irq; | |
1409 | ring->emit_bb_start = gen8_emit_bb_start; | |
1410 | ||
1411 | ring->dev = dev; | |
1412 | ret = logical_ring_init(dev, ring); | |
1413 | if (ret) | |
1414 | return ret; | |
1415 | ||
1416 | return intel_init_pipe_control(ring); | |
1417 | } | |
1418 | ||
1419 | static int logical_bsd_ring_init(struct drm_device *dev) | |
1420 | { | |
1421 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1422 | struct intel_engine_cs *ring = &dev_priv->ring[VCS]; | |
1423 | ||
1424 | ring->name = "bsd ring"; | |
1425 | ring->id = VCS; | |
1426 | ring->mmio_base = GEN6_BSD_RING_BASE; | |
1427 | ring->irq_enable_mask = | |
1428 | GT_RENDER_USER_INTERRUPT << GEN8_VCS1_IRQ_SHIFT; | |
1429 | ring->irq_keep_mask = | |
1430 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VCS1_IRQ_SHIFT; | |
1431 | ||
1432 | ring->init_hw = gen8_init_common_ring; | |
1433 | ring->get_seqno = gen8_get_seqno; | |
1434 | ring->set_seqno = gen8_set_seqno; | |
1435 | ring->emit_request = gen8_emit_request; | |
1436 | ring->emit_flush = gen8_emit_flush; | |
1437 | ring->irq_get = gen8_logical_ring_get_irq; | |
1438 | ring->irq_put = gen8_logical_ring_put_irq; | |
1439 | ring->emit_bb_start = gen8_emit_bb_start; | |
1440 | ||
1441 | return logical_ring_init(dev, ring); | |
1442 | } | |
1443 | ||
1444 | static int logical_bsd2_ring_init(struct drm_device *dev) | |
1445 | { | |
1446 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1447 | struct intel_engine_cs *ring = &dev_priv->ring[VCS2]; | |
1448 | ||
1449 | ring->name = "bds2 ring"; | |
1450 | ring->id = VCS2; | |
1451 | ring->mmio_base = GEN8_BSD2_RING_BASE; | |
1452 | ring->irq_enable_mask = | |
1453 | GT_RENDER_USER_INTERRUPT << GEN8_VCS2_IRQ_SHIFT; | |
1454 | ring->irq_keep_mask = | |
1455 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VCS2_IRQ_SHIFT; | |
1456 | ||
1457 | ring->init_hw = gen8_init_common_ring; | |
1458 | ring->get_seqno = gen8_get_seqno; | |
1459 | ring->set_seqno = gen8_set_seqno; | |
1460 | ring->emit_request = gen8_emit_request; | |
1461 | ring->emit_flush = gen8_emit_flush; | |
1462 | ring->irq_get = gen8_logical_ring_get_irq; | |
1463 | ring->irq_put = gen8_logical_ring_put_irq; | |
1464 | ring->emit_bb_start = gen8_emit_bb_start; | |
1465 | ||
1466 | return logical_ring_init(dev, ring); | |
1467 | } | |
1468 | ||
1469 | static int logical_blt_ring_init(struct drm_device *dev) | |
1470 | { | |
1471 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1472 | struct intel_engine_cs *ring = &dev_priv->ring[BCS]; | |
1473 | ||
1474 | ring->name = "blitter ring"; | |
1475 | ring->id = BCS; | |
1476 | ring->mmio_base = BLT_RING_BASE; | |
1477 | ring->irq_enable_mask = | |
1478 | GT_RENDER_USER_INTERRUPT << GEN8_BCS_IRQ_SHIFT; | |
1479 | ring->irq_keep_mask = | |
1480 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_BCS_IRQ_SHIFT; | |
1481 | ||
1482 | ring->init_hw = gen8_init_common_ring; | |
1483 | ring->get_seqno = gen8_get_seqno; | |
1484 | ring->set_seqno = gen8_set_seqno; | |
1485 | ring->emit_request = gen8_emit_request; | |
1486 | ring->emit_flush = gen8_emit_flush; | |
1487 | ring->irq_get = gen8_logical_ring_get_irq; | |
1488 | ring->irq_put = gen8_logical_ring_put_irq; | |
1489 | ring->emit_bb_start = gen8_emit_bb_start; | |
1490 | ||
1491 | return logical_ring_init(dev, ring); | |
1492 | } | |
1493 | ||
1494 | static int logical_vebox_ring_init(struct drm_device *dev) | |
1495 | { | |
1496 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1497 | struct intel_engine_cs *ring = &dev_priv->ring[VECS]; | |
1498 | ||
1499 | ring->name = "video enhancement ring"; | |
1500 | ring->id = VECS; | |
1501 | ring->mmio_base = VEBOX_RING_BASE; | |
1502 | ring->irq_enable_mask = | |
1503 | GT_RENDER_USER_INTERRUPT << GEN8_VECS_IRQ_SHIFT; | |
1504 | ring->irq_keep_mask = | |
1505 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VECS_IRQ_SHIFT; | |
1506 | ||
1507 | ring->init_hw = gen8_init_common_ring; | |
1508 | ring->get_seqno = gen8_get_seqno; | |
1509 | ring->set_seqno = gen8_set_seqno; | |
1510 | ring->emit_request = gen8_emit_request; | |
1511 | ring->emit_flush = gen8_emit_flush; | |
1512 | ring->irq_get = gen8_logical_ring_get_irq; | |
1513 | ring->irq_put = gen8_logical_ring_put_irq; | |
1514 | ring->emit_bb_start = gen8_emit_bb_start; | |
1515 | ||
1516 | return logical_ring_init(dev, ring); | |
1517 | } | |
1518 | ||
1519 | /** | |
1520 | * intel_logical_rings_init() - allocate, populate and init the Engine Command Streamers | |
1521 | * @dev: DRM device. | |
1522 | * | |
1523 | * This function inits the engines for an Execlists submission style (the equivalent in the | |
1524 | * legacy ringbuffer submission world would be i915_gem_init_rings). It does it only for | |
1525 | * those engines that are present in the hardware. | |
1526 | * | |
1527 | * Return: non-zero if the initialization failed. | |
1528 | */ | |
1529 | int intel_logical_rings_init(struct drm_device *dev) | |
1530 | { | |
1531 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1532 | int ret; | |
1533 | ||
1534 | ret = logical_render_ring_init(dev); | |
1535 | if (ret) | |
1536 | return ret; | |
1537 | ||
1538 | if (HAS_BSD(dev)) { | |
1539 | ret = logical_bsd_ring_init(dev); | |
1540 | if (ret) | |
1541 | goto cleanup_render_ring; | |
1542 | } | |
1543 | ||
1544 | if (HAS_BLT(dev)) { | |
1545 | ret = logical_blt_ring_init(dev); | |
1546 | if (ret) | |
1547 | goto cleanup_bsd_ring; | |
1548 | } | |
1549 | ||
1550 | if (HAS_VEBOX(dev)) { | |
1551 | ret = logical_vebox_ring_init(dev); | |
1552 | if (ret) | |
1553 | goto cleanup_blt_ring; | |
1554 | } | |
1555 | ||
1556 | if (HAS_BSD2(dev)) { | |
1557 | ret = logical_bsd2_ring_init(dev); | |
1558 | if (ret) | |
1559 | goto cleanup_vebox_ring; | |
1560 | } | |
1561 | ||
1562 | ret = i915_gem_set_seqno(dev, ((u32)~0 - 0x1000)); | |
1563 | if (ret) | |
1564 | goto cleanup_bsd2_ring; | |
1565 | ||
1566 | return 0; | |
1567 | ||
1568 | cleanup_bsd2_ring: | |
1569 | intel_logical_ring_cleanup(&dev_priv->ring[VCS2]); | |
1570 | cleanup_vebox_ring: | |
1571 | intel_logical_ring_cleanup(&dev_priv->ring[VECS]); | |
1572 | cleanup_blt_ring: | |
1573 | intel_logical_ring_cleanup(&dev_priv->ring[BCS]); | |
1574 | cleanup_bsd_ring: | |
1575 | intel_logical_ring_cleanup(&dev_priv->ring[VCS]); | |
1576 | cleanup_render_ring: | |
1577 | intel_logical_ring_cleanup(&dev_priv->ring[RCS]); | |
1578 | ||
1579 | return ret; | |
1580 | } | |
1581 | ||
1582 | int intel_lr_context_render_state_init(struct intel_engine_cs *ring, | |
1583 | struct intel_context *ctx) | |
1584 | { | |
1585 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; | |
1586 | struct render_state so; | |
1587 | struct drm_i915_file_private *file_priv = ctx->file_priv; | |
1588 | struct drm_file *file = file_priv ? file_priv->file : NULL; | |
1589 | int ret; | |
1590 | ||
1591 | ret = i915_gem_render_state_prepare(ring, &so); | |
1592 | if (ret) | |
1593 | return ret; | |
1594 | ||
1595 | if (so.rodata == NULL) | |
1596 | return 0; | |
1597 | ||
1598 | ret = ring->emit_bb_start(ringbuf, | |
1599 | ctx, | |
1600 | so.ggtt_offset, | |
1601 | I915_DISPATCH_SECURE); | |
1602 | if (ret) | |
1603 | goto out; | |
1604 | ||
1605 | i915_vma_move_to_active(i915_gem_obj_to_ggtt(so.obj), ring); | |
1606 | ||
1607 | ret = __i915_add_request(ring, file, so.obj); | |
1608 | /* intel_logical_ring_add_request moves object to inactive if it | |
1609 | * fails */ | |
1610 | out: | |
1611 | i915_gem_render_state_fini(&so); | |
1612 | return ret; | |
1613 | } | |
1614 | ||
1615 | static int | |
1616 | populate_lr_context(struct intel_context *ctx, struct drm_i915_gem_object *ctx_obj, | |
1617 | struct intel_engine_cs *ring, struct intel_ringbuffer *ringbuf) | |
1618 | { | |
1619 | struct drm_device *dev = ring->dev; | |
1620 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1621 | struct i915_hw_ppgtt *ppgtt = ctx->ppgtt; | |
1622 | struct page *page; | |
1623 | uint32_t *reg_state; | |
1624 | int ret; | |
1625 | ||
1626 | if (!ppgtt) | |
1627 | ppgtt = dev_priv->mm.aliasing_ppgtt; | |
1628 | ||
1629 | ret = i915_gem_object_set_to_cpu_domain(ctx_obj, true); | |
1630 | if (ret) { | |
1631 | DRM_DEBUG_DRIVER("Could not set to CPU domain\n"); | |
1632 | return ret; | |
1633 | } | |
1634 | ||
1635 | ret = i915_gem_object_get_pages(ctx_obj); | |
1636 | if (ret) { | |
1637 | DRM_DEBUG_DRIVER("Could not get object pages\n"); | |
1638 | return ret; | |
1639 | } | |
1640 | ||
1641 | i915_gem_object_pin_pages(ctx_obj); | |
1642 | ||
1643 | /* The second page of the context object contains some fields which must | |
1644 | * be set up prior to the first execution. */ | |
1645 | page = i915_gem_object_get_page(ctx_obj, 1); | |
1646 | reg_state = kmap_atomic(page); | |
1647 | ||
1648 | /* A context is actually a big batch buffer with several MI_LOAD_REGISTER_IMM | |
1649 | * commands followed by (reg, value) pairs. The values we are setting here are | |
1650 | * only for the first context restore: on a subsequent save, the GPU will | |
1651 | * recreate this batchbuffer with new values (including all the missing | |
1652 | * MI_LOAD_REGISTER_IMM commands that we are not initializing here). */ | |
1653 | if (ring->id == RCS) | |
1654 | reg_state[CTX_LRI_HEADER_0] = MI_LOAD_REGISTER_IMM(14); | |
1655 | else | |
1656 | reg_state[CTX_LRI_HEADER_0] = MI_LOAD_REGISTER_IMM(11); | |
1657 | reg_state[CTX_LRI_HEADER_0] |= MI_LRI_FORCE_POSTED; | |
1658 | reg_state[CTX_CONTEXT_CONTROL] = RING_CONTEXT_CONTROL(ring); | |
1659 | reg_state[CTX_CONTEXT_CONTROL+1] = | |
1660 | _MASKED_BIT_ENABLE((1<<3) | MI_RESTORE_INHIBIT); | |
1661 | reg_state[CTX_RING_HEAD] = RING_HEAD(ring->mmio_base); | |
1662 | reg_state[CTX_RING_HEAD+1] = 0; | |
1663 | reg_state[CTX_RING_TAIL] = RING_TAIL(ring->mmio_base); | |
1664 | reg_state[CTX_RING_TAIL+1] = 0; | |
1665 | reg_state[CTX_RING_BUFFER_START] = RING_START(ring->mmio_base); | |
1666 | /* Ring buffer start address is not known until the buffer is pinned. | |
1667 | * It is written to the context image in execlists_update_context() | |
1668 | */ | |
1669 | reg_state[CTX_RING_BUFFER_CONTROL] = RING_CTL(ring->mmio_base); | |
1670 | reg_state[CTX_RING_BUFFER_CONTROL+1] = | |
1671 | ((ringbuf->size - PAGE_SIZE) & RING_NR_PAGES) | RING_VALID; | |
1672 | reg_state[CTX_BB_HEAD_U] = ring->mmio_base + 0x168; | |
1673 | reg_state[CTX_BB_HEAD_U+1] = 0; | |
1674 | reg_state[CTX_BB_HEAD_L] = ring->mmio_base + 0x140; | |
1675 | reg_state[CTX_BB_HEAD_L+1] = 0; | |
1676 | reg_state[CTX_BB_STATE] = ring->mmio_base + 0x110; | |
1677 | reg_state[CTX_BB_STATE+1] = (1<<5); | |
1678 | reg_state[CTX_SECOND_BB_HEAD_U] = ring->mmio_base + 0x11c; | |
1679 | reg_state[CTX_SECOND_BB_HEAD_U+1] = 0; | |
1680 | reg_state[CTX_SECOND_BB_HEAD_L] = ring->mmio_base + 0x114; | |
1681 | reg_state[CTX_SECOND_BB_HEAD_L+1] = 0; | |
1682 | reg_state[CTX_SECOND_BB_STATE] = ring->mmio_base + 0x118; | |
1683 | reg_state[CTX_SECOND_BB_STATE+1] = 0; | |
1684 | if (ring->id == RCS) { | |
1685 | /* TODO: according to BSpec, the register state context | |
1686 | * for CHV does not have these. OTOH, these registers do | |
1687 | * exist in CHV. I'm waiting for a clarification */ | |
1688 | reg_state[CTX_BB_PER_CTX_PTR] = ring->mmio_base + 0x1c0; | |
1689 | reg_state[CTX_BB_PER_CTX_PTR+1] = 0; | |
1690 | reg_state[CTX_RCS_INDIRECT_CTX] = ring->mmio_base + 0x1c4; | |
1691 | reg_state[CTX_RCS_INDIRECT_CTX+1] = 0; | |
1692 | reg_state[CTX_RCS_INDIRECT_CTX_OFFSET] = ring->mmio_base + 0x1c8; | |
1693 | reg_state[CTX_RCS_INDIRECT_CTX_OFFSET+1] = 0; | |
1694 | } | |
1695 | reg_state[CTX_LRI_HEADER_1] = MI_LOAD_REGISTER_IMM(9); | |
1696 | reg_state[CTX_LRI_HEADER_1] |= MI_LRI_FORCE_POSTED; | |
1697 | reg_state[CTX_CTX_TIMESTAMP] = ring->mmio_base + 0x3a8; | |
1698 | reg_state[CTX_CTX_TIMESTAMP+1] = 0; | |
1699 | reg_state[CTX_PDP3_UDW] = GEN8_RING_PDP_UDW(ring, 3); | |
1700 | reg_state[CTX_PDP3_LDW] = GEN8_RING_PDP_LDW(ring, 3); | |
1701 | reg_state[CTX_PDP2_UDW] = GEN8_RING_PDP_UDW(ring, 2); | |
1702 | reg_state[CTX_PDP2_LDW] = GEN8_RING_PDP_LDW(ring, 2); | |
1703 | reg_state[CTX_PDP1_UDW] = GEN8_RING_PDP_UDW(ring, 1); | |
1704 | reg_state[CTX_PDP1_LDW] = GEN8_RING_PDP_LDW(ring, 1); | |
1705 | reg_state[CTX_PDP0_UDW] = GEN8_RING_PDP_UDW(ring, 0); | |
1706 | reg_state[CTX_PDP0_LDW] = GEN8_RING_PDP_LDW(ring, 0); | |
1707 | reg_state[CTX_PDP3_UDW+1] = upper_32_bits(ppgtt->pd_dma_addr[3]); | |
1708 | reg_state[CTX_PDP3_LDW+1] = lower_32_bits(ppgtt->pd_dma_addr[3]); | |
1709 | reg_state[CTX_PDP2_UDW+1] = upper_32_bits(ppgtt->pd_dma_addr[2]); | |
1710 | reg_state[CTX_PDP2_LDW+1] = lower_32_bits(ppgtt->pd_dma_addr[2]); | |
1711 | reg_state[CTX_PDP1_UDW+1] = upper_32_bits(ppgtt->pd_dma_addr[1]); | |
1712 | reg_state[CTX_PDP1_LDW+1] = lower_32_bits(ppgtt->pd_dma_addr[1]); | |
1713 | reg_state[CTX_PDP0_UDW+1] = upper_32_bits(ppgtt->pd_dma_addr[0]); | |
1714 | reg_state[CTX_PDP0_LDW+1] = lower_32_bits(ppgtt->pd_dma_addr[0]); | |
1715 | if (ring->id == RCS) { | |
1716 | reg_state[CTX_LRI_HEADER_2] = MI_LOAD_REGISTER_IMM(1); | |
1717 | reg_state[CTX_R_PWR_CLK_STATE] = 0x20c8; | |
1718 | reg_state[CTX_R_PWR_CLK_STATE+1] = 0; | |
1719 | } | |
1720 | ||
1721 | kunmap_atomic(reg_state); | |
1722 | ||
1723 | ctx_obj->dirty = 1; | |
1724 | set_page_dirty(page); | |
1725 | i915_gem_object_unpin_pages(ctx_obj); | |
1726 | ||
1727 | return 0; | |
1728 | } | |
1729 | ||
1730 | /** | |
1731 | * intel_lr_context_free() - free the LRC specific bits of a context | |
1732 | * @ctx: the LR context to free. | |
1733 | * | |
1734 | * The real context freeing is done in i915_gem_context_free: this only | |
1735 | * takes care of the bits that are LRC related: the per-engine backing | |
1736 | * objects and the logical ringbuffer. | |
1737 | */ | |
1738 | void intel_lr_context_free(struct intel_context *ctx) | |
1739 | { | |
1740 | int i; | |
1741 | ||
1742 | for (i = 0; i < I915_NUM_RINGS; i++) { | |
1743 | struct drm_i915_gem_object *ctx_obj = ctx->engine[i].state; | |
1744 | ||
1745 | if (ctx_obj) { | |
1746 | struct intel_ringbuffer *ringbuf = | |
1747 | ctx->engine[i].ringbuf; | |
1748 | struct intel_engine_cs *ring = ringbuf->ring; | |
1749 | ||
1750 | if (ctx == ring->default_context) { | |
1751 | intel_unpin_ringbuffer_obj(ringbuf); | |
1752 | i915_gem_object_ggtt_unpin(ctx_obj); | |
1753 | } | |
1754 | WARN_ON(ctx->engine[ring->id].pin_count); | |
1755 | intel_destroy_ringbuffer_obj(ringbuf); | |
1756 | kfree(ringbuf); | |
1757 | drm_gem_object_unreference(&ctx_obj->base); | |
1758 | } | |
1759 | } | |
1760 | } | |
1761 | ||
1762 | static uint32_t get_lr_context_size(struct intel_engine_cs *ring) | |
1763 | { | |
1764 | int ret = 0; | |
1765 | ||
1766 | WARN_ON(INTEL_INFO(ring->dev)->gen < 8); | |
1767 | ||
1768 | switch (ring->id) { | |
1769 | case RCS: | |
1770 | if (INTEL_INFO(ring->dev)->gen >= 9) | |
1771 | ret = GEN9_LR_CONTEXT_RENDER_SIZE; | |
1772 | else | |
1773 | ret = GEN8_LR_CONTEXT_RENDER_SIZE; | |
1774 | break; | |
1775 | case VCS: | |
1776 | case BCS: | |
1777 | case VECS: | |
1778 | case VCS2: | |
1779 | ret = GEN8_LR_CONTEXT_OTHER_SIZE; | |
1780 | break; | |
1781 | } | |
1782 | ||
1783 | return ret; | |
1784 | } | |
1785 | ||
1786 | static void lrc_setup_hardware_status_page(struct intel_engine_cs *ring, | |
1787 | struct drm_i915_gem_object *default_ctx_obj) | |
1788 | { | |
1789 | struct drm_i915_private *dev_priv = ring->dev->dev_private; | |
1790 | ||
1791 | /* The status page is offset 0 from the default context object | |
1792 | * in LRC mode. */ | |
1793 | ring->status_page.gfx_addr = i915_gem_obj_ggtt_offset(default_ctx_obj); | |
1794 | ring->status_page.page_addr = | |
1795 | kmap(sg_page(default_ctx_obj->pages->sgl)); | |
1796 | ring->status_page.obj = default_ctx_obj; | |
1797 | ||
1798 | I915_WRITE(RING_HWS_PGA(ring->mmio_base), | |
1799 | (u32)ring->status_page.gfx_addr); | |
1800 | POSTING_READ(RING_HWS_PGA(ring->mmio_base)); | |
1801 | } | |
1802 | ||
1803 | /** | |
1804 | * intel_lr_context_deferred_create() - create the LRC specific bits of a context | |
1805 | * @ctx: LR context to create. | |
1806 | * @ring: engine to be used with the context. | |
1807 | * | |
1808 | * This function can be called more than once, with different engines, if we plan | |
1809 | * to use the context with them. The context backing objects and the ringbuffers | |
1810 | * (specially the ringbuffer backing objects) suck a lot of memory up, and that's why | |
1811 | * the creation is a deferred call: it's better to make sure first that we need to use | |
1812 | * a given ring with the context. | |
1813 | * | |
1814 | * Return: non-zero on error. | |
1815 | */ | |
1816 | int intel_lr_context_deferred_create(struct intel_context *ctx, | |
1817 | struct intel_engine_cs *ring) | |
1818 | { | |
1819 | const bool is_global_default_ctx = (ctx == ring->default_context); | |
1820 | struct drm_device *dev = ring->dev; | |
1821 | struct drm_i915_gem_object *ctx_obj; | |
1822 | uint32_t context_size; | |
1823 | struct intel_ringbuffer *ringbuf; | |
1824 | int ret; | |
1825 | ||
1826 | WARN_ON(ctx->legacy_hw_ctx.rcs_state != NULL); | |
1827 | WARN_ON(ctx->engine[ring->id].state); | |
1828 | ||
1829 | context_size = round_up(get_lr_context_size(ring), 4096); | |
1830 | ||
1831 | ctx_obj = i915_gem_alloc_context_obj(dev, context_size); | |
1832 | if (IS_ERR(ctx_obj)) { | |
1833 | ret = PTR_ERR(ctx_obj); | |
1834 | DRM_DEBUG_DRIVER("Alloc LRC backing obj failed: %d\n", ret); | |
1835 | return ret; | |
1836 | } | |
1837 | ||
1838 | if (is_global_default_ctx) { | |
1839 | ret = i915_gem_obj_ggtt_pin(ctx_obj, GEN8_LR_CONTEXT_ALIGN, 0); | |
1840 | if (ret) { | |
1841 | DRM_DEBUG_DRIVER("Pin LRC backing obj failed: %d\n", | |
1842 | ret); | |
1843 | drm_gem_object_unreference(&ctx_obj->base); | |
1844 | return ret; | |
1845 | } | |
1846 | } | |
1847 | ||
1848 | ringbuf = kzalloc(sizeof(*ringbuf), GFP_KERNEL); | |
1849 | if (!ringbuf) { | |
1850 | DRM_DEBUG_DRIVER("Failed to allocate ringbuffer %s\n", | |
1851 | ring->name); | |
1852 | ret = -ENOMEM; | |
1853 | goto error_unpin_ctx; | |
1854 | } | |
1855 | ||
1856 | ringbuf->ring = ring; | |
1857 | ||
1858 | ringbuf->size = 32 * PAGE_SIZE; | |
1859 | ringbuf->effective_size = ringbuf->size; | |
1860 | ringbuf->head = 0; | |
1861 | ringbuf->tail = 0; | |
1862 | ringbuf->last_retired_head = -1; | |
1863 | intel_ring_update_space(ringbuf); | |
1864 | ||
1865 | if (ringbuf->obj == NULL) { | |
1866 | ret = intel_alloc_ringbuffer_obj(dev, ringbuf); | |
1867 | if (ret) { | |
1868 | DRM_DEBUG_DRIVER( | |
1869 | "Failed to allocate ringbuffer obj %s: %d\n", | |
1870 | ring->name, ret); | |
1871 | goto error_free_rbuf; | |
1872 | } | |
1873 | ||
1874 | if (is_global_default_ctx) { | |
1875 | ret = intel_pin_and_map_ringbuffer_obj(dev, ringbuf); | |
1876 | if (ret) { | |
1877 | DRM_ERROR( | |
1878 | "Failed to pin and map ringbuffer %s: %d\n", | |
1879 | ring->name, ret); | |
1880 | goto error_destroy_rbuf; | |
1881 | } | |
1882 | } | |
1883 | ||
1884 | } | |
1885 | ||
1886 | ret = populate_lr_context(ctx, ctx_obj, ring, ringbuf); | |
1887 | if (ret) { | |
1888 | DRM_DEBUG_DRIVER("Failed to populate LRC: %d\n", ret); | |
1889 | goto error; | |
1890 | } | |
1891 | ||
1892 | ctx->engine[ring->id].ringbuf = ringbuf; | |
1893 | ctx->engine[ring->id].state = ctx_obj; | |
1894 | ||
1895 | if (ctx == ring->default_context) | |
1896 | lrc_setup_hardware_status_page(ring, ctx_obj); | |
1897 | else if (ring->id == RCS && !ctx->rcs_initialized) { | |
1898 | if (ring->init_context) { | |
1899 | ret = ring->init_context(ring, ctx); | |
1900 | if (ret) { | |
1901 | DRM_ERROR("ring init context: %d\n", ret); | |
1902 | ctx->engine[ring->id].ringbuf = NULL; | |
1903 | ctx->engine[ring->id].state = NULL; | |
1904 | goto error; | |
1905 | } | |
1906 | } | |
1907 | ||
1908 | ctx->rcs_initialized = true; | |
1909 | } | |
1910 | ||
1911 | return 0; | |
1912 | ||
1913 | error: | |
1914 | if (is_global_default_ctx) | |
1915 | intel_unpin_ringbuffer_obj(ringbuf); | |
1916 | error_destroy_rbuf: | |
1917 | intel_destroy_ringbuffer_obj(ringbuf); | |
1918 | error_free_rbuf: | |
1919 | kfree(ringbuf); | |
1920 | error_unpin_ctx: | |
1921 | if (is_global_default_ctx) | |
1922 | i915_gem_object_ggtt_unpin(ctx_obj); | |
1923 | drm_gem_object_unreference(&ctx_obj->base); | |
1924 | return ret; | |
1925 | } |