Commit | Line | Data |
---|---|---|
b20385f1 OM |
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 | ||
73e4d07f OM |
31 | /** |
32 | * DOC: Logical Rings, Logical Ring Contexts and Execlists | |
33 | * | |
34 | * Motivation: | |
b20385f1 OM |
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 | * | |
73e4d07f OM |
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: | |
b20385f1 OM |
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). | |
73e4d07f OM |
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 | * | |
b20385f1 OM |
133 | */ |
134 | ||
135 | #include <drm/drmP.h> | |
136 | #include <drm/i915_drm.h> | |
137 | #include "i915_drv.h" | |
127f1003 | 138 | |
468c6816 | 139 | #define GEN9_LR_CONTEXT_RENDER_SIZE (22 * PAGE_SIZE) |
8c857917 OM |
140 | #define GEN8_LR_CONTEXT_RENDER_SIZE (20 * PAGE_SIZE) |
141 | #define GEN8_LR_CONTEXT_OTHER_SIZE (2 * PAGE_SIZE) | |
142 | ||
e981e7b1 TD |
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) | |
8670d6f9 OM |
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 | ||
84b790f8 BW |
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 | ||
7ba717cf TD |
206 | static int intel_lr_context_pin(struct intel_engine_cs *ring, |
207 | struct intel_context *ctx); | |
208 | ||
73e4d07f OM |
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 | |
27401d12 | 215 | * support for Logical Ring Contexts and Aliasing PPGTT or better). |
73e4d07f OM |
216 | * |
217 | * Return: 1 if Execlists is supported and has to be enabled. | |
218 | */ | |
127f1003 OM |
219 | int intel_sanitize_enable_execlists(struct drm_device *dev, int enable_execlists) |
220 | { | |
bd84b1e9 DV |
221 | WARN_ON(i915.enable_ppgtt == -1); |
222 | ||
70ee45e1 DL |
223 | if (INTEL_INFO(dev)->gen >= 9) |
224 | return 1; | |
225 | ||
127f1003 OM |
226 | if (enable_execlists == 0) |
227 | return 0; | |
228 | ||
14bf993e OM |
229 | if (HAS_LOGICAL_RING_CONTEXTS(dev) && USES_PPGTT(dev) && |
230 | i915.use_mmio_flip >= 0) | |
127f1003 OM |
231 | return 1; |
232 | ||
233 | return 0; | |
234 | } | |
ede7d42b | 235 | |
73e4d07f OM |
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 | */ | |
84b790f8 BW |
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 | ||
203a571b NH |
257 | static uint64_t execlists_ctx_descriptor(struct intel_engine_cs *ring, |
258 | struct drm_i915_gem_object *ctx_obj) | |
84b790f8 | 259 | { |
203a571b | 260 | struct drm_device *dev = ring->dev; |
84b790f8 BW |
261 | uint64_t desc; |
262 | uint64_t lrca = i915_gem_obj_ggtt_offset(ctx_obj); | |
acdd884a MT |
263 | |
264 | WARN_ON(lrca & 0xFFFFFFFF00000FFFULL); | |
84b790f8 BW |
265 | |
266 | desc = GEN8_CTX_VALID; | |
267 | desc |= LEGACY_CONTEXT << GEN8_CTX_MODE_SHIFT; | |
268 | desc |= GEN8_CTX_L3LLC_COHERENT; | |
269 | desc |= GEN8_CTX_PRIVILEGE; | |
270 | desc |= lrca; | |
271 | desc |= (u64)intel_execlists_ctx_id(ctx_obj) << GEN8_CTX_ID_SHIFT; | |
272 | ||
273 | /* TODO: WaDisableLiteRestore when we start using semaphore | |
274 | * signalling between Command Streamers */ | |
275 | /* desc |= GEN8_CTX_FORCE_RESTORE; */ | |
276 | ||
203a571b NH |
277 | /* WaEnableForceRestoreInCtxtDescForVCS:skl */ |
278 | if (IS_GEN9(dev) && | |
279 | INTEL_REVID(dev) <= SKL_REVID_B0 && | |
280 | (ring->id == BCS || ring->id == VCS || | |
281 | ring->id == VECS || ring->id == VCS2)) | |
282 | desc |= GEN8_CTX_FORCE_RESTORE; | |
283 | ||
84b790f8 BW |
284 | return desc; |
285 | } | |
286 | ||
287 | static void execlists_elsp_write(struct intel_engine_cs *ring, | |
288 | struct drm_i915_gem_object *ctx_obj0, | |
289 | struct drm_i915_gem_object *ctx_obj1) | |
290 | { | |
6e7cc470 TU |
291 | struct drm_device *dev = ring->dev; |
292 | struct drm_i915_private *dev_priv = dev->dev_private; | |
84b790f8 BW |
293 | uint64_t temp = 0; |
294 | uint32_t desc[4]; | |
295 | ||
296 | /* XXX: You must always write both descriptors in the order below. */ | |
297 | if (ctx_obj1) | |
203a571b | 298 | temp = execlists_ctx_descriptor(ring, ctx_obj1); |
84b790f8 BW |
299 | else |
300 | temp = 0; | |
301 | desc[1] = (u32)(temp >> 32); | |
302 | desc[0] = (u32)temp; | |
303 | ||
203a571b | 304 | temp = execlists_ctx_descriptor(ring, ctx_obj0); |
84b790f8 BW |
305 | desc[3] = (u32)(temp >> 32); |
306 | desc[2] = (u32)temp; | |
307 | ||
59bad947 | 308 | intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL); |
84b790f8 BW |
309 | I915_WRITE(RING_ELSP(ring), desc[1]); |
310 | I915_WRITE(RING_ELSP(ring), desc[0]); | |
311 | I915_WRITE(RING_ELSP(ring), desc[3]); | |
6daccb0b | 312 | |
84b790f8 BW |
313 | /* The context is automatically loaded after the following */ |
314 | I915_WRITE(RING_ELSP(ring), desc[2]); | |
315 | ||
316 | /* ELSP is a wo register, so use another nearby reg for posting instead */ | |
317 | POSTING_READ(RING_EXECLIST_STATUS(ring)); | |
59bad947 | 318 | intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); |
84b790f8 BW |
319 | } |
320 | ||
7ba717cf TD |
321 | static int execlists_update_context(struct drm_i915_gem_object *ctx_obj, |
322 | struct drm_i915_gem_object *ring_obj, | |
323 | u32 tail) | |
ae1250b9 OM |
324 | { |
325 | struct page *page; | |
326 | uint32_t *reg_state; | |
327 | ||
328 | page = i915_gem_object_get_page(ctx_obj, 1); | |
329 | reg_state = kmap_atomic(page); | |
330 | ||
331 | reg_state[CTX_RING_TAIL+1] = tail; | |
7ba717cf | 332 | reg_state[CTX_RING_BUFFER_START+1] = i915_gem_obj_ggtt_offset(ring_obj); |
ae1250b9 OM |
333 | |
334 | kunmap_atomic(reg_state); | |
335 | ||
336 | return 0; | |
337 | } | |
338 | ||
cd0707cb DG |
339 | static void execlists_submit_contexts(struct intel_engine_cs *ring, |
340 | struct intel_context *to0, u32 tail0, | |
341 | struct intel_context *to1, u32 tail1) | |
84b790f8 | 342 | { |
7ba717cf TD |
343 | struct drm_i915_gem_object *ctx_obj0 = to0->engine[ring->id].state; |
344 | struct intel_ringbuffer *ringbuf0 = to0->engine[ring->id].ringbuf; | |
84b790f8 | 345 | struct drm_i915_gem_object *ctx_obj1 = NULL; |
7ba717cf | 346 | struct intel_ringbuffer *ringbuf1 = NULL; |
84b790f8 | 347 | |
84b790f8 | 348 | BUG_ON(!ctx_obj0); |
acdd884a | 349 | WARN_ON(!i915_gem_obj_is_pinned(ctx_obj0)); |
7ba717cf | 350 | WARN_ON(!i915_gem_obj_is_pinned(ringbuf0->obj)); |
84b790f8 | 351 | |
7ba717cf | 352 | execlists_update_context(ctx_obj0, ringbuf0->obj, tail0); |
ae1250b9 | 353 | |
84b790f8 | 354 | if (to1) { |
7ba717cf | 355 | ringbuf1 = to1->engine[ring->id].ringbuf; |
84b790f8 BW |
356 | ctx_obj1 = to1->engine[ring->id].state; |
357 | BUG_ON(!ctx_obj1); | |
acdd884a | 358 | WARN_ON(!i915_gem_obj_is_pinned(ctx_obj1)); |
7ba717cf | 359 | WARN_ON(!i915_gem_obj_is_pinned(ringbuf1->obj)); |
ae1250b9 | 360 | |
7ba717cf | 361 | execlists_update_context(ctx_obj1, ringbuf1->obj, tail1); |
84b790f8 BW |
362 | } |
363 | ||
364 | execlists_elsp_write(ring, ctx_obj0, ctx_obj1); | |
84b790f8 BW |
365 | } |
366 | ||
acdd884a MT |
367 | static void execlists_context_unqueue(struct intel_engine_cs *ring) |
368 | { | |
6d3d8274 NH |
369 | struct drm_i915_gem_request *req0 = NULL, *req1 = NULL; |
370 | struct drm_i915_gem_request *cursor = NULL, *tmp = NULL; | |
e981e7b1 TD |
371 | |
372 | assert_spin_locked(&ring->execlist_lock); | |
acdd884a MT |
373 | |
374 | if (list_empty(&ring->execlist_queue)) | |
375 | return; | |
376 | ||
377 | /* Try to read in pairs */ | |
378 | list_for_each_entry_safe(cursor, tmp, &ring->execlist_queue, | |
379 | execlist_link) { | |
380 | if (!req0) { | |
381 | req0 = cursor; | |
6d3d8274 | 382 | } else if (req0->ctx == cursor->ctx) { |
acdd884a MT |
383 | /* Same ctx: ignore first request, as second request |
384 | * will update tail past first request's workload */ | |
e1fee72c | 385 | cursor->elsp_submitted = req0->elsp_submitted; |
acdd884a | 386 | list_del(&req0->execlist_link); |
c86ee3a9 TD |
387 | list_add_tail(&req0->execlist_link, |
388 | &ring->execlist_retired_req_list); | |
acdd884a MT |
389 | req0 = cursor; |
390 | } else { | |
391 | req1 = cursor; | |
392 | break; | |
393 | } | |
394 | } | |
395 | ||
53292cdb MT |
396 | if (IS_GEN8(ring->dev) || IS_GEN9(ring->dev)) { |
397 | /* | |
398 | * WaIdleLiteRestore: make sure we never cause a lite | |
399 | * restore with HEAD==TAIL | |
400 | */ | |
401 | if (req0 && req0->elsp_submitted) { | |
402 | /* | |
403 | * Apply the wa NOOPS to prevent ring:HEAD == req:TAIL | |
404 | * as we resubmit the request. See gen8_emit_request() | |
405 | * for where we prepare the padding after the end of the | |
406 | * request. | |
407 | */ | |
408 | struct intel_ringbuffer *ringbuf; | |
409 | ||
410 | ringbuf = req0->ctx->engine[ring->id].ringbuf; | |
411 | req0->tail += 8; | |
412 | req0->tail &= ringbuf->size - 1; | |
413 | } | |
414 | } | |
415 | ||
e1fee72c OM |
416 | WARN_ON(req1 && req1->elsp_submitted); |
417 | ||
6d3d8274 NH |
418 | execlists_submit_contexts(ring, req0->ctx, req0->tail, |
419 | req1 ? req1->ctx : NULL, | |
420 | req1 ? req1->tail : 0); | |
e1fee72c OM |
421 | |
422 | req0->elsp_submitted++; | |
423 | if (req1) | |
424 | req1->elsp_submitted++; | |
acdd884a MT |
425 | } |
426 | ||
e981e7b1 TD |
427 | static bool execlists_check_remove_request(struct intel_engine_cs *ring, |
428 | u32 request_id) | |
429 | { | |
6d3d8274 | 430 | struct drm_i915_gem_request *head_req; |
e981e7b1 TD |
431 | |
432 | assert_spin_locked(&ring->execlist_lock); | |
433 | ||
434 | head_req = list_first_entry_or_null(&ring->execlist_queue, | |
6d3d8274 | 435 | struct drm_i915_gem_request, |
e981e7b1 TD |
436 | execlist_link); |
437 | ||
438 | if (head_req != NULL) { | |
439 | struct drm_i915_gem_object *ctx_obj = | |
6d3d8274 | 440 | head_req->ctx->engine[ring->id].state; |
e981e7b1 | 441 | if (intel_execlists_ctx_id(ctx_obj) == request_id) { |
e1fee72c OM |
442 | WARN(head_req->elsp_submitted == 0, |
443 | "Never submitted head request\n"); | |
444 | ||
445 | if (--head_req->elsp_submitted <= 0) { | |
446 | list_del(&head_req->execlist_link); | |
c86ee3a9 TD |
447 | list_add_tail(&head_req->execlist_link, |
448 | &ring->execlist_retired_req_list); | |
e1fee72c OM |
449 | return true; |
450 | } | |
e981e7b1 TD |
451 | } |
452 | } | |
453 | ||
454 | return false; | |
455 | } | |
456 | ||
73e4d07f | 457 | /** |
3f7531c3 | 458 | * intel_lrc_irq_handler() - handle Context Switch interrupts |
73e4d07f OM |
459 | * @ring: Engine Command Streamer to handle. |
460 | * | |
461 | * Check the unread Context Status Buffers and manage the submission of new | |
462 | * contexts to the ELSP accordingly. | |
463 | */ | |
3f7531c3 | 464 | void intel_lrc_irq_handler(struct intel_engine_cs *ring) |
e981e7b1 TD |
465 | { |
466 | struct drm_i915_private *dev_priv = ring->dev->dev_private; | |
467 | u32 status_pointer; | |
468 | u8 read_pointer; | |
469 | u8 write_pointer; | |
470 | u32 status; | |
471 | u32 status_id; | |
472 | u32 submit_contexts = 0; | |
473 | ||
474 | status_pointer = I915_READ(RING_CONTEXT_STATUS_PTR(ring)); | |
475 | ||
476 | read_pointer = ring->next_context_status_buffer; | |
477 | write_pointer = status_pointer & 0x07; | |
478 | if (read_pointer > write_pointer) | |
479 | write_pointer += 6; | |
480 | ||
481 | spin_lock(&ring->execlist_lock); | |
482 | ||
483 | while (read_pointer < write_pointer) { | |
484 | read_pointer++; | |
485 | status = I915_READ(RING_CONTEXT_STATUS_BUF(ring) + | |
486 | (read_pointer % 6) * 8); | |
487 | status_id = I915_READ(RING_CONTEXT_STATUS_BUF(ring) + | |
488 | (read_pointer % 6) * 8 + 4); | |
489 | ||
e1fee72c OM |
490 | if (status & GEN8_CTX_STATUS_PREEMPTED) { |
491 | if (status & GEN8_CTX_STATUS_LITE_RESTORE) { | |
492 | if (execlists_check_remove_request(ring, status_id)) | |
493 | WARN(1, "Lite Restored request removed from queue\n"); | |
494 | } else | |
495 | WARN(1, "Preemption without Lite Restore\n"); | |
496 | } | |
497 | ||
498 | if ((status & GEN8_CTX_STATUS_ACTIVE_IDLE) || | |
499 | (status & GEN8_CTX_STATUS_ELEMENT_SWITCH)) { | |
e981e7b1 TD |
500 | if (execlists_check_remove_request(ring, status_id)) |
501 | submit_contexts++; | |
502 | } | |
503 | } | |
504 | ||
505 | if (submit_contexts != 0) | |
506 | execlists_context_unqueue(ring); | |
507 | ||
508 | spin_unlock(&ring->execlist_lock); | |
509 | ||
510 | WARN(submit_contexts > 2, "More than two context complete events?\n"); | |
511 | ring->next_context_status_buffer = write_pointer % 6; | |
512 | ||
513 | I915_WRITE(RING_CONTEXT_STATUS_PTR(ring), | |
514 | ((u32)ring->next_context_status_buffer & 0x07) << 8); | |
515 | } | |
516 | ||
acdd884a MT |
517 | static int execlists_context_queue(struct intel_engine_cs *ring, |
518 | struct intel_context *to, | |
2d12955a NH |
519 | u32 tail, |
520 | struct drm_i915_gem_request *request) | |
acdd884a | 521 | { |
6d3d8274 | 522 | struct drm_i915_gem_request *cursor; |
e981e7b1 | 523 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
acdd884a | 524 | unsigned long flags; |
f1ad5a1f | 525 | int num_elements = 0; |
acdd884a | 526 | |
7ba717cf TD |
527 | if (to != ring->default_context) |
528 | intel_lr_context_pin(ring, to); | |
529 | ||
2d12955a NH |
530 | if (!request) { |
531 | /* | |
532 | * If there isn't a request associated with this submission, | |
533 | * create one as a temporary holder. | |
534 | */ | |
2d12955a NH |
535 | request = kzalloc(sizeof(*request), GFP_KERNEL); |
536 | if (request == NULL) | |
537 | return -ENOMEM; | |
2d12955a | 538 | request->ring = ring; |
6d3d8274 | 539 | request->ctx = to; |
b3a38998 NH |
540 | kref_init(&request->ref); |
541 | request->uniq = dev_priv->request_uniq++; | |
542 | i915_gem_context_reference(request->ctx); | |
21076372 | 543 | } else { |
b3a38998 | 544 | i915_gem_request_reference(request); |
21076372 | 545 | WARN_ON(to != request->ctx); |
2d12955a | 546 | } |
72f95afa | 547 | request->tail = tail; |
2d12955a | 548 | |
e981e7b1 | 549 | intel_runtime_pm_get(dev_priv); |
acdd884a MT |
550 | |
551 | spin_lock_irqsave(&ring->execlist_lock, flags); | |
552 | ||
f1ad5a1f OM |
553 | list_for_each_entry(cursor, &ring->execlist_queue, execlist_link) |
554 | if (++num_elements > 2) | |
555 | break; | |
556 | ||
557 | if (num_elements > 2) { | |
6d3d8274 | 558 | struct drm_i915_gem_request *tail_req; |
f1ad5a1f OM |
559 | |
560 | tail_req = list_last_entry(&ring->execlist_queue, | |
6d3d8274 | 561 | struct drm_i915_gem_request, |
f1ad5a1f OM |
562 | execlist_link); |
563 | ||
6d3d8274 | 564 | if (to == tail_req->ctx) { |
f1ad5a1f | 565 | WARN(tail_req->elsp_submitted != 0, |
7ba717cf | 566 | "More than 2 already-submitted reqs queued\n"); |
f1ad5a1f | 567 | list_del(&tail_req->execlist_link); |
c86ee3a9 TD |
568 | list_add_tail(&tail_req->execlist_link, |
569 | &ring->execlist_retired_req_list); | |
f1ad5a1f OM |
570 | } |
571 | } | |
572 | ||
6d3d8274 | 573 | list_add_tail(&request->execlist_link, &ring->execlist_queue); |
f1ad5a1f | 574 | if (num_elements == 0) |
acdd884a MT |
575 | execlists_context_unqueue(ring); |
576 | ||
577 | spin_unlock_irqrestore(&ring->execlist_lock, flags); | |
578 | ||
579 | return 0; | |
580 | } | |
581 | ||
21076372 NH |
582 | static int logical_ring_invalidate_all_caches(struct intel_ringbuffer *ringbuf, |
583 | struct intel_context *ctx) | |
ba8b7ccb OM |
584 | { |
585 | struct intel_engine_cs *ring = ringbuf->ring; | |
586 | uint32_t flush_domains; | |
587 | int ret; | |
588 | ||
589 | flush_domains = 0; | |
590 | if (ring->gpu_caches_dirty) | |
591 | flush_domains = I915_GEM_GPU_DOMAINS; | |
592 | ||
21076372 NH |
593 | ret = ring->emit_flush(ringbuf, ctx, |
594 | I915_GEM_GPU_DOMAINS, flush_domains); | |
ba8b7ccb OM |
595 | if (ret) |
596 | return ret; | |
597 | ||
598 | ring->gpu_caches_dirty = false; | |
599 | return 0; | |
600 | } | |
601 | ||
602 | static int execlists_move_to_gpu(struct intel_ringbuffer *ringbuf, | |
21076372 | 603 | struct intel_context *ctx, |
ba8b7ccb OM |
604 | struct list_head *vmas) |
605 | { | |
606 | struct intel_engine_cs *ring = ringbuf->ring; | |
607 | struct i915_vma *vma; | |
608 | uint32_t flush_domains = 0; | |
609 | bool flush_chipset = false; | |
610 | int ret; | |
611 | ||
612 | list_for_each_entry(vma, vmas, exec_list) { | |
613 | struct drm_i915_gem_object *obj = vma->obj; | |
614 | ||
615 | ret = i915_gem_object_sync(obj, ring); | |
616 | if (ret) | |
617 | return ret; | |
618 | ||
619 | if (obj->base.write_domain & I915_GEM_DOMAIN_CPU) | |
620 | flush_chipset |= i915_gem_clflush_object(obj, false); | |
621 | ||
622 | flush_domains |= obj->base.write_domain; | |
623 | } | |
624 | ||
625 | if (flush_domains & I915_GEM_DOMAIN_GTT) | |
626 | wmb(); | |
627 | ||
628 | /* Unconditionally invalidate gpu caches and ensure that we do flush | |
629 | * any residual writes from the previous batch. | |
630 | */ | |
21076372 | 631 | return logical_ring_invalidate_all_caches(ringbuf, ctx); |
ba8b7ccb OM |
632 | } |
633 | ||
73e4d07f OM |
634 | /** |
635 | * execlists_submission() - submit a batchbuffer for execution, Execlists style | |
636 | * @dev: DRM device. | |
637 | * @file: DRM file. | |
638 | * @ring: Engine Command Streamer to submit to. | |
639 | * @ctx: Context to employ for this submission. | |
640 | * @args: execbuffer call arguments. | |
641 | * @vmas: list of vmas. | |
642 | * @batch_obj: the batchbuffer to submit. | |
643 | * @exec_start: batchbuffer start virtual address pointer. | |
8e004efc | 644 | * @dispatch_flags: translated execbuffer call flags. |
73e4d07f OM |
645 | * |
646 | * This is the evil twin version of i915_gem_ringbuffer_submission. It abstracts | |
647 | * away the submission details of the execbuffer ioctl call. | |
648 | * | |
649 | * Return: non-zero if the submission fails. | |
650 | */ | |
454afebd OM |
651 | int intel_execlists_submission(struct drm_device *dev, struct drm_file *file, |
652 | struct intel_engine_cs *ring, | |
653 | struct intel_context *ctx, | |
654 | struct drm_i915_gem_execbuffer2 *args, | |
655 | struct list_head *vmas, | |
656 | struct drm_i915_gem_object *batch_obj, | |
8e004efc | 657 | u64 exec_start, u32 dispatch_flags) |
454afebd | 658 | { |
ba8b7ccb OM |
659 | struct drm_i915_private *dev_priv = dev->dev_private; |
660 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; | |
661 | int instp_mode; | |
662 | u32 instp_mask; | |
663 | int ret; | |
664 | ||
665 | instp_mode = args->flags & I915_EXEC_CONSTANTS_MASK; | |
666 | instp_mask = I915_EXEC_CONSTANTS_MASK; | |
667 | switch (instp_mode) { | |
668 | case I915_EXEC_CONSTANTS_REL_GENERAL: | |
669 | case I915_EXEC_CONSTANTS_ABSOLUTE: | |
670 | case I915_EXEC_CONSTANTS_REL_SURFACE: | |
671 | if (instp_mode != 0 && ring != &dev_priv->ring[RCS]) { | |
672 | DRM_DEBUG("non-0 rel constants mode on non-RCS\n"); | |
673 | return -EINVAL; | |
674 | } | |
675 | ||
676 | if (instp_mode != dev_priv->relative_constants_mode) { | |
677 | if (instp_mode == I915_EXEC_CONSTANTS_REL_SURFACE) { | |
678 | DRM_DEBUG("rel surface constants mode invalid on gen5+\n"); | |
679 | return -EINVAL; | |
680 | } | |
681 | ||
682 | /* The HW changed the meaning on this bit on gen6 */ | |
683 | instp_mask &= ~I915_EXEC_CONSTANTS_REL_SURFACE; | |
684 | } | |
685 | break; | |
686 | default: | |
687 | DRM_DEBUG("execbuf with unknown constants: %d\n", instp_mode); | |
688 | return -EINVAL; | |
689 | } | |
690 | ||
691 | if (args->num_cliprects != 0) { | |
692 | DRM_DEBUG("clip rectangles are only valid on pre-gen5\n"); | |
693 | return -EINVAL; | |
694 | } else { | |
695 | if (args->DR4 == 0xffffffff) { | |
696 | DRM_DEBUG("UXA submitting garbage DR4, fixing up\n"); | |
697 | args->DR4 = 0; | |
698 | } | |
699 | ||
700 | if (args->DR1 || args->DR4 || args->cliprects_ptr) { | |
701 | DRM_DEBUG("0 cliprects but dirt in cliprects fields\n"); | |
702 | return -EINVAL; | |
703 | } | |
704 | } | |
705 | ||
706 | if (args->flags & I915_EXEC_GEN7_SOL_RESET) { | |
707 | DRM_DEBUG("sol reset is gen7 only\n"); | |
708 | return -EINVAL; | |
709 | } | |
710 | ||
21076372 | 711 | ret = execlists_move_to_gpu(ringbuf, ctx, vmas); |
ba8b7ccb OM |
712 | if (ret) |
713 | return ret; | |
714 | ||
715 | if (ring == &dev_priv->ring[RCS] && | |
716 | instp_mode != dev_priv->relative_constants_mode) { | |
21076372 | 717 | ret = intel_logical_ring_begin(ringbuf, ctx, 4); |
ba8b7ccb OM |
718 | if (ret) |
719 | return ret; | |
720 | ||
721 | intel_logical_ring_emit(ringbuf, MI_NOOP); | |
722 | intel_logical_ring_emit(ringbuf, MI_LOAD_REGISTER_IMM(1)); | |
723 | intel_logical_ring_emit(ringbuf, INSTPM); | |
724 | intel_logical_ring_emit(ringbuf, instp_mask << 16 | instp_mode); | |
725 | intel_logical_ring_advance(ringbuf); | |
726 | ||
727 | dev_priv->relative_constants_mode = instp_mode; | |
728 | } | |
729 | ||
8e004efc | 730 | ret = ring->emit_bb_start(ringbuf, ctx, exec_start, dispatch_flags); |
ba8b7ccb OM |
731 | if (ret) |
732 | return ret; | |
733 | ||
5e4be7bd JH |
734 | trace_i915_gem_ring_dispatch(intel_ring_get_request(ring), dispatch_flags); |
735 | ||
ba8b7ccb OM |
736 | i915_gem_execbuffer_move_to_active(vmas, ring); |
737 | i915_gem_execbuffer_retire_commands(dev, file, ring, batch_obj); | |
738 | ||
454afebd OM |
739 | return 0; |
740 | } | |
741 | ||
c86ee3a9 TD |
742 | void intel_execlists_retire_requests(struct intel_engine_cs *ring) |
743 | { | |
6d3d8274 | 744 | struct drm_i915_gem_request *req, *tmp; |
c86ee3a9 TD |
745 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
746 | unsigned long flags; | |
747 | struct list_head retired_list; | |
748 | ||
749 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); | |
750 | if (list_empty(&ring->execlist_retired_req_list)) | |
751 | return; | |
752 | ||
753 | INIT_LIST_HEAD(&retired_list); | |
754 | spin_lock_irqsave(&ring->execlist_lock, flags); | |
755 | list_replace_init(&ring->execlist_retired_req_list, &retired_list); | |
756 | spin_unlock_irqrestore(&ring->execlist_lock, flags); | |
757 | ||
758 | list_for_each_entry_safe(req, tmp, &retired_list, execlist_link) { | |
6d3d8274 | 759 | struct intel_context *ctx = req->ctx; |
7ba717cf TD |
760 | struct drm_i915_gem_object *ctx_obj = |
761 | ctx->engine[ring->id].state; | |
762 | ||
763 | if (ctx_obj && (ctx != ring->default_context)) | |
764 | intel_lr_context_unpin(ring, ctx); | |
c86ee3a9 | 765 | intel_runtime_pm_put(dev_priv); |
c86ee3a9 | 766 | list_del(&req->execlist_link); |
f8210795 | 767 | i915_gem_request_unreference(req); |
c86ee3a9 TD |
768 | } |
769 | } | |
770 | ||
454afebd OM |
771 | void intel_logical_ring_stop(struct intel_engine_cs *ring) |
772 | { | |
9832b9da OM |
773 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
774 | int ret; | |
775 | ||
776 | if (!intel_ring_initialized(ring)) | |
777 | return; | |
778 | ||
779 | ret = intel_ring_idle(ring); | |
780 | if (ret && !i915_reset_in_progress(&to_i915(ring->dev)->gpu_error)) | |
781 | DRM_ERROR("failed to quiesce %s whilst cleaning up: %d\n", | |
782 | ring->name, ret); | |
783 | ||
784 | /* TODO: Is this correct with Execlists enabled? */ | |
785 | I915_WRITE_MODE(ring, _MASKED_BIT_ENABLE(STOP_RING)); | |
786 | if (wait_for_atomic((I915_READ_MODE(ring) & MODE_IDLE) != 0, 1000)) { | |
787 | DRM_ERROR("%s :timed out trying to stop ring\n", ring->name); | |
788 | return; | |
789 | } | |
790 | I915_WRITE_MODE(ring, _MASKED_BIT_DISABLE(STOP_RING)); | |
454afebd OM |
791 | } |
792 | ||
21076372 NH |
793 | int logical_ring_flush_all_caches(struct intel_ringbuffer *ringbuf, |
794 | struct intel_context *ctx) | |
48e29f55 OM |
795 | { |
796 | struct intel_engine_cs *ring = ringbuf->ring; | |
797 | int ret; | |
798 | ||
799 | if (!ring->gpu_caches_dirty) | |
800 | return 0; | |
801 | ||
21076372 | 802 | ret = ring->emit_flush(ringbuf, ctx, 0, I915_GEM_GPU_DOMAINS); |
48e29f55 OM |
803 | if (ret) |
804 | return ret; | |
805 | ||
806 | ring->gpu_caches_dirty = false; | |
807 | return 0; | |
808 | } | |
809 | ||
183c9906 | 810 | /* |
73e4d07f OM |
811 | * intel_logical_ring_advance_and_submit() - advance the tail and submit the workload |
812 | * @ringbuf: Logical Ringbuffer to advance. | |
813 | * | |
814 | * The tail is updated in our logical ringbuffer struct, not in the actual context. What | |
815 | * really happens during submission is that the context and current tail will be placed | |
816 | * on a queue waiting for the ELSP to be ready to accept a new context submission. At that | |
817 | * point, the tail *inside* the context is updated and the ELSP written to. | |
818 | */ | |
183c9906 DL |
819 | static void |
820 | intel_logical_ring_advance_and_submit(struct intel_ringbuffer *ringbuf, | |
821 | struct intel_context *ctx, | |
822 | struct drm_i915_gem_request *request) | |
82e104cc | 823 | { |
84b790f8 | 824 | struct intel_engine_cs *ring = ringbuf->ring; |
84b790f8 | 825 | |
82e104cc OM |
826 | intel_logical_ring_advance(ringbuf); |
827 | ||
84b790f8 | 828 | if (intel_ring_stopped(ring)) |
82e104cc OM |
829 | return; |
830 | ||
2d12955a | 831 | execlists_context_queue(ring, ctx, ringbuf->tail, request); |
82e104cc OM |
832 | } |
833 | ||
dcb4c12a OM |
834 | static int intel_lr_context_pin(struct intel_engine_cs *ring, |
835 | struct intel_context *ctx) | |
836 | { | |
837 | struct drm_i915_gem_object *ctx_obj = ctx->engine[ring->id].state; | |
7ba717cf | 838 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; |
dcb4c12a OM |
839 | int ret = 0; |
840 | ||
841 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); | |
a7cbedec | 842 | if (ctx->engine[ring->id].pin_count++ == 0) { |
dcb4c12a OM |
843 | ret = i915_gem_obj_ggtt_pin(ctx_obj, |
844 | GEN8_LR_CONTEXT_ALIGN, 0); | |
845 | if (ret) | |
a7cbedec | 846 | goto reset_pin_count; |
7ba717cf TD |
847 | |
848 | ret = intel_pin_and_map_ringbuffer_obj(ring->dev, ringbuf); | |
849 | if (ret) | |
850 | goto unpin_ctx_obj; | |
dcb4c12a OM |
851 | } |
852 | ||
7ba717cf TD |
853 | return ret; |
854 | ||
855 | unpin_ctx_obj: | |
856 | i915_gem_object_ggtt_unpin(ctx_obj); | |
a7cbedec MK |
857 | reset_pin_count: |
858 | ctx->engine[ring->id].pin_count = 0; | |
7ba717cf | 859 | |
dcb4c12a OM |
860 | return ret; |
861 | } | |
862 | ||
863 | void intel_lr_context_unpin(struct intel_engine_cs *ring, | |
864 | struct intel_context *ctx) | |
865 | { | |
866 | struct drm_i915_gem_object *ctx_obj = ctx->engine[ring->id].state; | |
7ba717cf | 867 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; |
dcb4c12a OM |
868 | |
869 | if (ctx_obj) { | |
870 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); | |
a7cbedec | 871 | if (--ctx->engine[ring->id].pin_count == 0) { |
7ba717cf | 872 | intel_unpin_ringbuffer_obj(ringbuf); |
dcb4c12a | 873 | i915_gem_object_ggtt_unpin(ctx_obj); |
7ba717cf | 874 | } |
dcb4c12a OM |
875 | } |
876 | } | |
877 | ||
6259cead JH |
878 | static int logical_ring_alloc_request(struct intel_engine_cs *ring, |
879 | struct intel_context *ctx) | |
82e104cc | 880 | { |
9eba5d4a | 881 | struct drm_i915_gem_request *request; |
67e2937b | 882 | struct drm_i915_private *dev_private = ring->dev->dev_private; |
dcb4c12a OM |
883 | int ret; |
884 | ||
6259cead | 885 | if (ring->outstanding_lazy_request) |
9eba5d4a | 886 | return 0; |
82e104cc | 887 | |
aaeb1ba0 | 888 | request = kzalloc(sizeof(*request), GFP_KERNEL); |
9eba5d4a JH |
889 | if (request == NULL) |
890 | return -ENOMEM; | |
82e104cc | 891 | |
9eba5d4a JH |
892 | if (ctx != ring->default_context) { |
893 | ret = intel_lr_context_pin(ring, ctx); | |
894 | if (ret) { | |
895 | kfree(request); | |
896 | return ret; | |
dcb4c12a | 897 | } |
9eba5d4a | 898 | } |
dcb4c12a | 899 | |
abfe262a | 900 | kref_init(&request->ref); |
ff79e857 | 901 | request->ring = ring; |
67e2937b | 902 | request->uniq = dev_private->request_uniq++; |
abfe262a | 903 | |
6259cead | 904 | ret = i915_gem_get_seqno(ring->dev, &request->seqno); |
9eba5d4a JH |
905 | if (ret) { |
906 | intel_lr_context_unpin(ring, ctx); | |
907 | kfree(request); | |
908 | return ret; | |
82e104cc OM |
909 | } |
910 | ||
9eba5d4a JH |
911 | request->ctx = ctx; |
912 | i915_gem_context_reference(request->ctx); | |
98e1bd4a | 913 | request->ringbuf = ctx->engine[ring->id].ringbuf; |
9eba5d4a | 914 | |
6259cead | 915 | ring->outstanding_lazy_request = request; |
9eba5d4a | 916 | return 0; |
82e104cc OM |
917 | } |
918 | ||
919 | static int logical_ring_wait_request(struct intel_ringbuffer *ringbuf, | |
920 | int bytes) | |
921 | { | |
922 | struct intel_engine_cs *ring = ringbuf->ring; | |
923 | struct drm_i915_gem_request *request; | |
82e104cc OM |
924 | int ret; |
925 | ||
ebd0fd4b DG |
926 | if (intel_ring_space(ringbuf) >= bytes) |
927 | return 0; | |
82e104cc OM |
928 | |
929 | list_for_each_entry(request, &ring->request_list, list) { | |
57e21513 DG |
930 | /* |
931 | * The request queue is per-engine, so can contain requests | |
932 | * from multiple ringbuffers. Here, we must ignore any that | |
933 | * aren't from the ringbuffer we're considering. | |
934 | */ | |
935 | struct intel_context *ctx = request->ctx; | |
936 | if (ctx->engine[ring->id].ringbuf != ringbuf) | |
937 | continue; | |
938 | ||
939 | /* Would completion of this request free enough space? */ | |
82e104cc OM |
940 | if (__intel_ring_space(request->tail, ringbuf->tail, |
941 | ringbuf->size) >= bytes) { | |
82e104cc OM |
942 | break; |
943 | } | |
944 | } | |
945 | ||
a4b3a571 | 946 | if (&request->list == &ring->request_list) |
82e104cc OM |
947 | return -ENOSPC; |
948 | ||
a4b3a571 | 949 | ret = i915_wait_request(request); |
82e104cc OM |
950 | if (ret) |
951 | return ret; | |
952 | ||
82e104cc | 953 | i915_gem_retire_requests_ring(ring); |
82e104cc | 954 | |
ebd0fd4b | 955 | return intel_ring_space(ringbuf) >= bytes ? 0 : -ENOSPC; |
82e104cc OM |
956 | } |
957 | ||
958 | static int logical_ring_wait_for_space(struct intel_ringbuffer *ringbuf, | |
21076372 | 959 | struct intel_context *ctx, |
82e104cc OM |
960 | int bytes) |
961 | { | |
962 | struct intel_engine_cs *ring = ringbuf->ring; | |
963 | struct drm_device *dev = ring->dev; | |
964 | struct drm_i915_private *dev_priv = dev->dev_private; | |
965 | unsigned long end; | |
966 | int ret; | |
967 | ||
968 | ret = logical_ring_wait_request(ringbuf, bytes); | |
969 | if (ret != -ENOSPC) | |
970 | return ret; | |
971 | ||
972 | /* Force the context submission in case we have been skipping it */ | |
21076372 | 973 | intel_logical_ring_advance_and_submit(ringbuf, ctx, NULL); |
82e104cc OM |
974 | |
975 | /* With GEM the hangcheck timer should kick us out of the loop, | |
976 | * leaving it early runs the risk of corrupting GEM state (due | |
977 | * to running on almost untested codepaths). But on resume | |
978 | * timers don't work yet, so prevent a complete hang in that | |
979 | * case by choosing an insanely large timeout. */ | |
980 | end = jiffies + 60 * HZ; | |
981 | ||
ebd0fd4b | 982 | ret = 0; |
82e104cc | 983 | do { |
ebd0fd4b | 984 | if (intel_ring_space(ringbuf) >= bytes) |
82e104cc | 985 | break; |
82e104cc OM |
986 | |
987 | msleep(1); | |
988 | ||
989 | if (dev_priv->mm.interruptible && signal_pending(current)) { | |
990 | ret = -ERESTARTSYS; | |
991 | break; | |
992 | } | |
993 | ||
994 | ret = i915_gem_check_wedge(&dev_priv->gpu_error, | |
995 | dev_priv->mm.interruptible); | |
996 | if (ret) | |
997 | break; | |
998 | ||
999 | if (time_after(jiffies, end)) { | |
1000 | ret = -EBUSY; | |
1001 | break; | |
1002 | } | |
1003 | } while (1); | |
1004 | ||
1005 | return ret; | |
1006 | } | |
1007 | ||
21076372 NH |
1008 | static int logical_ring_wrap_buffer(struct intel_ringbuffer *ringbuf, |
1009 | struct intel_context *ctx) | |
82e104cc OM |
1010 | { |
1011 | uint32_t __iomem *virt; | |
1012 | int rem = ringbuf->size - ringbuf->tail; | |
1013 | ||
1014 | if (ringbuf->space < rem) { | |
21076372 | 1015 | int ret = logical_ring_wait_for_space(ringbuf, ctx, rem); |
82e104cc OM |
1016 | |
1017 | if (ret) | |
1018 | return ret; | |
1019 | } | |
1020 | ||
1021 | virt = ringbuf->virtual_start + ringbuf->tail; | |
1022 | rem /= 4; | |
1023 | while (rem--) | |
1024 | iowrite32(MI_NOOP, virt++); | |
1025 | ||
1026 | ringbuf->tail = 0; | |
ebd0fd4b | 1027 | intel_ring_update_space(ringbuf); |
82e104cc OM |
1028 | |
1029 | return 0; | |
1030 | } | |
1031 | ||
21076372 NH |
1032 | static int logical_ring_prepare(struct intel_ringbuffer *ringbuf, |
1033 | struct intel_context *ctx, int bytes) | |
82e104cc OM |
1034 | { |
1035 | int ret; | |
1036 | ||
1037 | if (unlikely(ringbuf->tail + bytes > ringbuf->effective_size)) { | |
21076372 | 1038 | ret = logical_ring_wrap_buffer(ringbuf, ctx); |
82e104cc OM |
1039 | if (unlikely(ret)) |
1040 | return ret; | |
1041 | } | |
1042 | ||
1043 | if (unlikely(ringbuf->space < bytes)) { | |
21076372 | 1044 | ret = logical_ring_wait_for_space(ringbuf, ctx, bytes); |
82e104cc OM |
1045 | if (unlikely(ret)) |
1046 | return ret; | |
1047 | } | |
1048 | ||
1049 | return 0; | |
1050 | } | |
1051 | ||
73e4d07f OM |
1052 | /** |
1053 | * intel_logical_ring_begin() - prepare the logical ringbuffer to accept some commands | |
1054 | * | |
1055 | * @ringbuf: Logical ringbuffer. | |
1056 | * @num_dwords: number of DWORDs that we plan to write to the ringbuffer. | |
1057 | * | |
1058 | * The ringbuffer might not be ready to accept the commands right away (maybe it needs to | |
1059 | * be wrapped, or wait a bit for the tail to be updated). This function takes care of that | |
1060 | * and also preallocates a request (every workload submission is still mediated through | |
1061 | * requests, same as it did with legacy ringbuffer submission). | |
1062 | * | |
1063 | * Return: non-zero if the ringbuffer is not ready to be written to. | |
1064 | */ | |
21076372 NH |
1065 | int intel_logical_ring_begin(struct intel_ringbuffer *ringbuf, |
1066 | struct intel_context *ctx, int num_dwords) | |
82e104cc OM |
1067 | { |
1068 | struct intel_engine_cs *ring = ringbuf->ring; | |
1069 | struct drm_device *dev = ring->dev; | |
1070 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1071 | int ret; | |
1072 | ||
1073 | ret = i915_gem_check_wedge(&dev_priv->gpu_error, | |
1074 | dev_priv->mm.interruptible); | |
1075 | if (ret) | |
1076 | return ret; | |
1077 | ||
21076372 | 1078 | ret = logical_ring_prepare(ringbuf, ctx, num_dwords * sizeof(uint32_t)); |
82e104cc OM |
1079 | if (ret) |
1080 | return ret; | |
1081 | ||
1082 | /* Preallocate the olr before touching the ring */ | |
21076372 | 1083 | ret = logical_ring_alloc_request(ring, ctx); |
82e104cc OM |
1084 | if (ret) |
1085 | return ret; | |
1086 | ||
1087 | ringbuf->space -= num_dwords * sizeof(uint32_t); | |
1088 | return 0; | |
1089 | } | |
1090 | ||
771b9a53 MT |
1091 | static int intel_logical_ring_workarounds_emit(struct intel_engine_cs *ring, |
1092 | struct intel_context *ctx) | |
1093 | { | |
1094 | int ret, i; | |
1095 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; | |
1096 | struct drm_device *dev = ring->dev; | |
1097 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1098 | struct i915_workarounds *w = &dev_priv->workarounds; | |
1099 | ||
e6c1abb7 | 1100 | if (WARN_ON_ONCE(w->count == 0)) |
771b9a53 MT |
1101 | return 0; |
1102 | ||
1103 | ring->gpu_caches_dirty = true; | |
21076372 | 1104 | ret = logical_ring_flush_all_caches(ringbuf, ctx); |
771b9a53 MT |
1105 | if (ret) |
1106 | return ret; | |
1107 | ||
21076372 | 1108 | ret = intel_logical_ring_begin(ringbuf, ctx, w->count * 2 + 2); |
771b9a53 MT |
1109 | if (ret) |
1110 | return ret; | |
1111 | ||
1112 | intel_logical_ring_emit(ringbuf, MI_LOAD_REGISTER_IMM(w->count)); | |
1113 | for (i = 0; i < w->count; i++) { | |
1114 | intel_logical_ring_emit(ringbuf, w->reg[i].addr); | |
1115 | intel_logical_ring_emit(ringbuf, w->reg[i].value); | |
1116 | } | |
1117 | intel_logical_ring_emit(ringbuf, MI_NOOP); | |
1118 | ||
1119 | intel_logical_ring_advance(ringbuf); | |
1120 | ||
1121 | ring->gpu_caches_dirty = true; | |
21076372 | 1122 | ret = logical_ring_flush_all_caches(ringbuf, ctx); |
771b9a53 MT |
1123 | if (ret) |
1124 | return ret; | |
1125 | ||
1126 | return 0; | |
1127 | } | |
1128 | ||
9b1136d5 OM |
1129 | static int gen8_init_common_ring(struct intel_engine_cs *ring) |
1130 | { | |
1131 | struct drm_device *dev = ring->dev; | |
1132 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1133 | ||
73d477f6 OM |
1134 | I915_WRITE_IMR(ring, ~(ring->irq_enable_mask | ring->irq_keep_mask)); |
1135 | I915_WRITE(RING_HWSTAM(ring->mmio_base), 0xffffffff); | |
1136 | ||
9b1136d5 OM |
1137 | I915_WRITE(RING_MODE_GEN7(ring), |
1138 | _MASKED_BIT_DISABLE(GFX_REPLAY_MODE) | | |
1139 | _MASKED_BIT_ENABLE(GFX_RUN_LIST_ENABLE)); | |
1140 | POSTING_READ(RING_MODE_GEN7(ring)); | |
c0a03a2e | 1141 | ring->next_context_status_buffer = 0; |
9b1136d5 OM |
1142 | DRM_DEBUG_DRIVER("Execlists enabled for %s\n", ring->name); |
1143 | ||
1144 | memset(&ring->hangcheck, 0, sizeof(ring->hangcheck)); | |
1145 | ||
1146 | return 0; | |
1147 | } | |
1148 | ||
1149 | static int gen8_init_render_ring(struct intel_engine_cs *ring) | |
1150 | { | |
1151 | struct drm_device *dev = ring->dev; | |
1152 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1153 | int ret; | |
1154 | ||
1155 | ret = gen8_init_common_ring(ring); | |
1156 | if (ret) | |
1157 | return ret; | |
1158 | ||
1159 | /* We need to disable the AsyncFlip performance optimisations in order | |
1160 | * to use MI_WAIT_FOR_EVENT within the CS. It should already be | |
1161 | * programmed to '1' on all products. | |
1162 | * | |
1163 | * WaDisableAsyncFlipPerfMode:snb,ivb,hsw,vlv,bdw,chv | |
1164 | */ | |
1165 | I915_WRITE(MI_MODE, _MASKED_BIT_ENABLE(ASYNC_FLIP_PERF_DISABLE)); | |
1166 | ||
9b1136d5 OM |
1167 | I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_FORCE_ORDERING)); |
1168 | ||
771b9a53 | 1169 | return init_workarounds_ring(ring); |
9b1136d5 OM |
1170 | } |
1171 | ||
82ef822e DL |
1172 | static int gen9_init_render_ring(struct intel_engine_cs *ring) |
1173 | { | |
1174 | int ret; | |
1175 | ||
1176 | ret = gen8_init_common_ring(ring); | |
1177 | if (ret) | |
1178 | return ret; | |
1179 | ||
1180 | return init_workarounds_ring(ring); | |
1181 | } | |
1182 | ||
15648585 | 1183 | static int gen8_emit_bb_start(struct intel_ringbuffer *ringbuf, |
21076372 | 1184 | struct intel_context *ctx, |
8e004efc | 1185 | u64 offset, unsigned dispatch_flags) |
15648585 | 1186 | { |
8e004efc | 1187 | bool ppgtt = !(dispatch_flags & I915_DISPATCH_SECURE); |
15648585 OM |
1188 | int ret; |
1189 | ||
21076372 | 1190 | ret = intel_logical_ring_begin(ringbuf, ctx, 4); |
15648585 OM |
1191 | if (ret) |
1192 | return ret; | |
1193 | ||
1194 | /* FIXME(BDW): Address space and security selectors. */ | |
1195 | intel_logical_ring_emit(ringbuf, MI_BATCH_BUFFER_START_GEN8 | (ppgtt<<8)); | |
1196 | intel_logical_ring_emit(ringbuf, lower_32_bits(offset)); | |
1197 | intel_logical_ring_emit(ringbuf, upper_32_bits(offset)); | |
1198 | intel_logical_ring_emit(ringbuf, MI_NOOP); | |
1199 | intel_logical_ring_advance(ringbuf); | |
1200 | ||
1201 | return 0; | |
1202 | } | |
1203 | ||
73d477f6 OM |
1204 | static bool gen8_logical_ring_get_irq(struct intel_engine_cs *ring) |
1205 | { | |
1206 | struct drm_device *dev = ring->dev; | |
1207 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1208 | unsigned long flags; | |
1209 | ||
7cd512f1 | 1210 | if (WARN_ON(!intel_irqs_enabled(dev_priv))) |
73d477f6 OM |
1211 | return false; |
1212 | ||
1213 | spin_lock_irqsave(&dev_priv->irq_lock, flags); | |
1214 | if (ring->irq_refcount++ == 0) { | |
1215 | I915_WRITE_IMR(ring, ~(ring->irq_enable_mask | ring->irq_keep_mask)); | |
1216 | POSTING_READ(RING_IMR(ring->mmio_base)); | |
1217 | } | |
1218 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); | |
1219 | ||
1220 | return true; | |
1221 | } | |
1222 | ||
1223 | static void gen8_logical_ring_put_irq(struct intel_engine_cs *ring) | |
1224 | { | |
1225 | struct drm_device *dev = ring->dev; | |
1226 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1227 | unsigned long flags; | |
1228 | ||
1229 | spin_lock_irqsave(&dev_priv->irq_lock, flags); | |
1230 | if (--ring->irq_refcount == 0) { | |
1231 | I915_WRITE_IMR(ring, ~ring->irq_keep_mask); | |
1232 | POSTING_READ(RING_IMR(ring->mmio_base)); | |
1233 | } | |
1234 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); | |
1235 | } | |
1236 | ||
4712274c | 1237 | static int gen8_emit_flush(struct intel_ringbuffer *ringbuf, |
21076372 | 1238 | struct intel_context *ctx, |
4712274c OM |
1239 | u32 invalidate_domains, |
1240 | u32 unused) | |
1241 | { | |
1242 | struct intel_engine_cs *ring = ringbuf->ring; | |
1243 | struct drm_device *dev = ring->dev; | |
1244 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1245 | uint32_t cmd; | |
1246 | int ret; | |
1247 | ||
21076372 | 1248 | ret = intel_logical_ring_begin(ringbuf, ctx, 4); |
4712274c OM |
1249 | if (ret) |
1250 | return ret; | |
1251 | ||
1252 | cmd = MI_FLUSH_DW + 1; | |
1253 | ||
f0a1fb10 CW |
1254 | /* We always require a command barrier so that subsequent |
1255 | * commands, such as breadcrumb interrupts, are strictly ordered | |
1256 | * wrt the contents of the write cache being flushed to memory | |
1257 | * (and thus being coherent from the CPU). | |
1258 | */ | |
1259 | cmd |= MI_FLUSH_DW_STORE_INDEX | MI_FLUSH_DW_OP_STOREDW; | |
1260 | ||
1261 | if (invalidate_domains & I915_GEM_GPU_DOMAINS) { | |
1262 | cmd |= MI_INVALIDATE_TLB; | |
1263 | if (ring == &dev_priv->ring[VCS]) | |
1264 | cmd |= MI_INVALIDATE_BSD; | |
4712274c OM |
1265 | } |
1266 | ||
1267 | intel_logical_ring_emit(ringbuf, cmd); | |
1268 | intel_logical_ring_emit(ringbuf, | |
1269 | I915_GEM_HWS_SCRATCH_ADDR | | |
1270 | MI_FLUSH_DW_USE_GTT); | |
1271 | intel_logical_ring_emit(ringbuf, 0); /* upper addr */ | |
1272 | intel_logical_ring_emit(ringbuf, 0); /* value */ | |
1273 | intel_logical_ring_advance(ringbuf); | |
1274 | ||
1275 | return 0; | |
1276 | } | |
1277 | ||
1278 | static int gen8_emit_flush_render(struct intel_ringbuffer *ringbuf, | |
21076372 | 1279 | struct intel_context *ctx, |
4712274c OM |
1280 | u32 invalidate_domains, |
1281 | u32 flush_domains) | |
1282 | { | |
1283 | struct intel_engine_cs *ring = ringbuf->ring; | |
1284 | u32 scratch_addr = ring->scratch.gtt_offset + 2 * CACHELINE_BYTES; | |
1285 | u32 flags = 0; | |
1286 | int ret; | |
1287 | ||
1288 | flags |= PIPE_CONTROL_CS_STALL; | |
1289 | ||
1290 | if (flush_domains) { | |
1291 | flags |= PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH; | |
1292 | flags |= PIPE_CONTROL_DEPTH_CACHE_FLUSH; | |
1293 | } | |
1294 | ||
1295 | if (invalidate_domains) { | |
1296 | flags |= PIPE_CONTROL_TLB_INVALIDATE; | |
1297 | flags |= PIPE_CONTROL_INSTRUCTION_CACHE_INVALIDATE; | |
1298 | flags |= PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE; | |
1299 | flags |= PIPE_CONTROL_VF_CACHE_INVALIDATE; | |
1300 | flags |= PIPE_CONTROL_CONST_CACHE_INVALIDATE; | |
1301 | flags |= PIPE_CONTROL_STATE_CACHE_INVALIDATE; | |
1302 | flags |= PIPE_CONTROL_QW_WRITE; | |
1303 | flags |= PIPE_CONTROL_GLOBAL_GTT_IVB; | |
1304 | } | |
1305 | ||
21076372 | 1306 | ret = intel_logical_ring_begin(ringbuf, ctx, 6); |
4712274c OM |
1307 | if (ret) |
1308 | return ret; | |
1309 | ||
1310 | intel_logical_ring_emit(ringbuf, GFX_OP_PIPE_CONTROL(6)); | |
1311 | intel_logical_ring_emit(ringbuf, flags); | |
1312 | intel_logical_ring_emit(ringbuf, scratch_addr); | |
1313 | intel_logical_ring_emit(ringbuf, 0); | |
1314 | intel_logical_ring_emit(ringbuf, 0); | |
1315 | intel_logical_ring_emit(ringbuf, 0); | |
1316 | intel_logical_ring_advance(ringbuf); | |
1317 | ||
1318 | return 0; | |
1319 | } | |
1320 | ||
e94e37ad OM |
1321 | static u32 gen8_get_seqno(struct intel_engine_cs *ring, bool lazy_coherency) |
1322 | { | |
1323 | return intel_read_status_page(ring, I915_GEM_HWS_INDEX); | |
1324 | } | |
1325 | ||
1326 | static void gen8_set_seqno(struct intel_engine_cs *ring, u32 seqno) | |
1327 | { | |
1328 | intel_write_status_page(ring, I915_GEM_HWS_INDEX, seqno); | |
1329 | } | |
1330 | ||
2d12955a NH |
1331 | static int gen8_emit_request(struct intel_ringbuffer *ringbuf, |
1332 | struct drm_i915_gem_request *request) | |
4da46e1e OM |
1333 | { |
1334 | struct intel_engine_cs *ring = ringbuf->ring; | |
1335 | u32 cmd; | |
1336 | int ret; | |
1337 | ||
53292cdb MT |
1338 | /* |
1339 | * Reserve space for 2 NOOPs at the end of each request to be | |
1340 | * used as a workaround for not being allowed to do lite | |
1341 | * restore with HEAD==TAIL (WaIdleLiteRestore). | |
1342 | */ | |
1343 | ret = intel_logical_ring_begin(ringbuf, request->ctx, 8); | |
4da46e1e OM |
1344 | if (ret) |
1345 | return ret; | |
1346 | ||
8edfbb8b | 1347 | cmd = MI_STORE_DWORD_IMM_GEN4; |
4da46e1e OM |
1348 | cmd |= MI_GLOBAL_GTT; |
1349 | ||
1350 | intel_logical_ring_emit(ringbuf, cmd); | |
1351 | intel_logical_ring_emit(ringbuf, | |
1352 | (ring->status_page.gfx_addr + | |
1353 | (I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT))); | |
1354 | intel_logical_ring_emit(ringbuf, 0); | |
6259cead JH |
1355 | intel_logical_ring_emit(ringbuf, |
1356 | i915_gem_request_get_seqno(ring->outstanding_lazy_request)); | |
4da46e1e OM |
1357 | intel_logical_ring_emit(ringbuf, MI_USER_INTERRUPT); |
1358 | intel_logical_ring_emit(ringbuf, MI_NOOP); | |
21076372 | 1359 | intel_logical_ring_advance_and_submit(ringbuf, request->ctx, request); |
4da46e1e | 1360 | |
53292cdb MT |
1361 | /* |
1362 | * Here we add two extra NOOPs as padding to avoid | |
1363 | * lite restore of a context with HEAD==TAIL. | |
1364 | */ | |
1365 | intel_logical_ring_emit(ringbuf, MI_NOOP); | |
1366 | intel_logical_ring_emit(ringbuf, MI_NOOP); | |
1367 | intel_logical_ring_advance(ringbuf); | |
1368 | ||
4da46e1e OM |
1369 | return 0; |
1370 | } | |
1371 | ||
cef437ad DL |
1372 | static int intel_lr_context_render_state_init(struct intel_engine_cs *ring, |
1373 | struct intel_context *ctx) | |
1374 | { | |
1375 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; | |
1376 | struct render_state so; | |
1377 | struct drm_i915_file_private *file_priv = ctx->file_priv; | |
1378 | struct drm_file *file = file_priv ? file_priv->file : NULL; | |
1379 | int ret; | |
1380 | ||
1381 | ret = i915_gem_render_state_prepare(ring, &so); | |
1382 | if (ret) | |
1383 | return ret; | |
1384 | ||
1385 | if (so.rodata == NULL) | |
1386 | return 0; | |
1387 | ||
1388 | ret = ring->emit_bb_start(ringbuf, | |
1389 | ctx, | |
1390 | so.ggtt_offset, | |
1391 | I915_DISPATCH_SECURE); | |
1392 | if (ret) | |
1393 | goto out; | |
1394 | ||
1395 | i915_vma_move_to_active(i915_gem_obj_to_ggtt(so.obj), ring); | |
1396 | ||
1397 | ret = __i915_add_request(ring, file, so.obj); | |
1398 | /* intel_logical_ring_add_request moves object to inactive if it | |
1399 | * fails */ | |
1400 | out: | |
1401 | i915_gem_render_state_fini(&so); | |
1402 | return ret; | |
1403 | } | |
1404 | ||
e7778be1 TD |
1405 | static int gen8_init_rcs_context(struct intel_engine_cs *ring, |
1406 | struct intel_context *ctx) | |
1407 | { | |
1408 | int ret; | |
1409 | ||
1410 | ret = intel_logical_ring_workarounds_emit(ring, ctx); | |
1411 | if (ret) | |
1412 | return ret; | |
1413 | ||
1414 | return intel_lr_context_render_state_init(ring, ctx); | |
1415 | } | |
1416 | ||
73e4d07f OM |
1417 | /** |
1418 | * intel_logical_ring_cleanup() - deallocate the Engine Command Streamer | |
1419 | * | |
1420 | * @ring: Engine Command Streamer. | |
1421 | * | |
1422 | */ | |
454afebd OM |
1423 | void intel_logical_ring_cleanup(struct intel_engine_cs *ring) |
1424 | { | |
6402c330 | 1425 | struct drm_i915_private *dev_priv; |
9832b9da | 1426 | |
48d82387 OM |
1427 | if (!intel_ring_initialized(ring)) |
1428 | return; | |
1429 | ||
6402c330 JH |
1430 | dev_priv = ring->dev->dev_private; |
1431 | ||
9832b9da OM |
1432 | intel_logical_ring_stop(ring); |
1433 | WARN_ON((I915_READ_MODE(ring) & MODE_IDLE) == 0); | |
6259cead | 1434 | i915_gem_request_assign(&ring->outstanding_lazy_request, NULL); |
48d82387 OM |
1435 | |
1436 | if (ring->cleanup) | |
1437 | ring->cleanup(ring); | |
1438 | ||
1439 | i915_cmd_parser_fini_ring(ring); | |
1440 | ||
1441 | if (ring->status_page.obj) { | |
1442 | kunmap(sg_page(ring->status_page.obj->pages->sgl)); | |
1443 | ring->status_page.obj = NULL; | |
1444 | } | |
454afebd OM |
1445 | } |
1446 | ||
1447 | static int logical_ring_init(struct drm_device *dev, struct intel_engine_cs *ring) | |
1448 | { | |
48d82387 | 1449 | int ret; |
48d82387 OM |
1450 | |
1451 | /* Intentionally left blank. */ | |
1452 | ring->buffer = NULL; | |
1453 | ||
1454 | ring->dev = dev; | |
1455 | INIT_LIST_HEAD(&ring->active_list); | |
1456 | INIT_LIST_HEAD(&ring->request_list); | |
1457 | init_waitqueue_head(&ring->irq_queue); | |
1458 | ||
acdd884a | 1459 | INIT_LIST_HEAD(&ring->execlist_queue); |
c86ee3a9 | 1460 | INIT_LIST_HEAD(&ring->execlist_retired_req_list); |
acdd884a MT |
1461 | spin_lock_init(&ring->execlist_lock); |
1462 | ||
48d82387 OM |
1463 | ret = i915_cmd_parser_init_ring(ring); |
1464 | if (ret) | |
1465 | return ret; | |
1466 | ||
564ddb2f OM |
1467 | ret = intel_lr_context_deferred_create(ring->default_context, ring); |
1468 | ||
1469 | return ret; | |
454afebd OM |
1470 | } |
1471 | ||
1472 | static int logical_render_ring_init(struct drm_device *dev) | |
1473 | { | |
1474 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1475 | struct intel_engine_cs *ring = &dev_priv->ring[RCS]; | |
99be1dfe | 1476 | int ret; |
454afebd OM |
1477 | |
1478 | ring->name = "render ring"; | |
1479 | ring->id = RCS; | |
1480 | ring->mmio_base = RENDER_RING_BASE; | |
1481 | ring->irq_enable_mask = | |
1482 | GT_RENDER_USER_INTERRUPT << GEN8_RCS_IRQ_SHIFT; | |
73d477f6 OM |
1483 | ring->irq_keep_mask = |
1484 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_RCS_IRQ_SHIFT; | |
1485 | if (HAS_L3_DPF(dev)) | |
1486 | ring->irq_keep_mask |= GT_RENDER_L3_PARITY_ERROR_INTERRUPT; | |
454afebd | 1487 | |
82ef822e DL |
1488 | if (INTEL_INFO(dev)->gen >= 9) |
1489 | ring->init_hw = gen9_init_render_ring; | |
1490 | else | |
1491 | ring->init_hw = gen8_init_render_ring; | |
e7778be1 | 1492 | ring->init_context = gen8_init_rcs_context; |
9b1136d5 | 1493 | ring->cleanup = intel_fini_pipe_control; |
e94e37ad OM |
1494 | ring->get_seqno = gen8_get_seqno; |
1495 | ring->set_seqno = gen8_set_seqno; | |
4da46e1e | 1496 | ring->emit_request = gen8_emit_request; |
4712274c | 1497 | ring->emit_flush = gen8_emit_flush_render; |
73d477f6 OM |
1498 | ring->irq_get = gen8_logical_ring_get_irq; |
1499 | ring->irq_put = gen8_logical_ring_put_irq; | |
15648585 | 1500 | ring->emit_bb_start = gen8_emit_bb_start; |
9b1136d5 | 1501 | |
99be1dfe DV |
1502 | ring->dev = dev; |
1503 | ret = logical_ring_init(dev, ring); | |
1504 | if (ret) | |
1505 | return ret; | |
1506 | ||
1507 | return intel_init_pipe_control(ring); | |
454afebd OM |
1508 | } |
1509 | ||
1510 | static int logical_bsd_ring_init(struct drm_device *dev) | |
1511 | { | |
1512 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1513 | struct intel_engine_cs *ring = &dev_priv->ring[VCS]; | |
1514 | ||
1515 | ring->name = "bsd ring"; | |
1516 | ring->id = VCS; | |
1517 | ring->mmio_base = GEN6_BSD_RING_BASE; | |
1518 | ring->irq_enable_mask = | |
1519 | GT_RENDER_USER_INTERRUPT << GEN8_VCS1_IRQ_SHIFT; | |
73d477f6 OM |
1520 | ring->irq_keep_mask = |
1521 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VCS1_IRQ_SHIFT; | |
454afebd | 1522 | |
ecfe00d8 | 1523 | ring->init_hw = gen8_init_common_ring; |
e94e37ad OM |
1524 | ring->get_seqno = gen8_get_seqno; |
1525 | ring->set_seqno = gen8_set_seqno; | |
4da46e1e | 1526 | ring->emit_request = gen8_emit_request; |
4712274c | 1527 | ring->emit_flush = gen8_emit_flush; |
73d477f6 OM |
1528 | ring->irq_get = gen8_logical_ring_get_irq; |
1529 | ring->irq_put = gen8_logical_ring_put_irq; | |
15648585 | 1530 | ring->emit_bb_start = gen8_emit_bb_start; |
9b1136d5 | 1531 | |
454afebd OM |
1532 | return logical_ring_init(dev, ring); |
1533 | } | |
1534 | ||
1535 | static int logical_bsd2_ring_init(struct drm_device *dev) | |
1536 | { | |
1537 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1538 | struct intel_engine_cs *ring = &dev_priv->ring[VCS2]; | |
1539 | ||
1540 | ring->name = "bds2 ring"; | |
1541 | ring->id = VCS2; | |
1542 | ring->mmio_base = GEN8_BSD2_RING_BASE; | |
1543 | ring->irq_enable_mask = | |
1544 | GT_RENDER_USER_INTERRUPT << GEN8_VCS2_IRQ_SHIFT; | |
73d477f6 OM |
1545 | ring->irq_keep_mask = |
1546 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VCS2_IRQ_SHIFT; | |
454afebd | 1547 | |
ecfe00d8 | 1548 | ring->init_hw = gen8_init_common_ring; |
e94e37ad OM |
1549 | ring->get_seqno = gen8_get_seqno; |
1550 | ring->set_seqno = gen8_set_seqno; | |
4da46e1e | 1551 | ring->emit_request = gen8_emit_request; |
4712274c | 1552 | ring->emit_flush = gen8_emit_flush; |
73d477f6 OM |
1553 | ring->irq_get = gen8_logical_ring_get_irq; |
1554 | ring->irq_put = gen8_logical_ring_put_irq; | |
15648585 | 1555 | ring->emit_bb_start = gen8_emit_bb_start; |
9b1136d5 | 1556 | |
454afebd OM |
1557 | return logical_ring_init(dev, ring); |
1558 | } | |
1559 | ||
1560 | static int logical_blt_ring_init(struct drm_device *dev) | |
1561 | { | |
1562 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1563 | struct intel_engine_cs *ring = &dev_priv->ring[BCS]; | |
1564 | ||
1565 | ring->name = "blitter ring"; | |
1566 | ring->id = BCS; | |
1567 | ring->mmio_base = BLT_RING_BASE; | |
1568 | ring->irq_enable_mask = | |
1569 | GT_RENDER_USER_INTERRUPT << GEN8_BCS_IRQ_SHIFT; | |
73d477f6 OM |
1570 | ring->irq_keep_mask = |
1571 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_BCS_IRQ_SHIFT; | |
454afebd | 1572 | |
ecfe00d8 | 1573 | ring->init_hw = gen8_init_common_ring; |
e94e37ad OM |
1574 | ring->get_seqno = gen8_get_seqno; |
1575 | ring->set_seqno = gen8_set_seqno; | |
4da46e1e | 1576 | ring->emit_request = gen8_emit_request; |
4712274c | 1577 | ring->emit_flush = gen8_emit_flush; |
73d477f6 OM |
1578 | ring->irq_get = gen8_logical_ring_get_irq; |
1579 | ring->irq_put = gen8_logical_ring_put_irq; | |
15648585 | 1580 | ring->emit_bb_start = gen8_emit_bb_start; |
9b1136d5 | 1581 | |
454afebd OM |
1582 | return logical_ring_init(dev, ring); |
1583 | } | |
1584 | ||
1585 | static int logical_vebox_ring_init(struct drm_device *dev) | |
1586 | { | |
1587 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1588 | struct intel_engine_cs *ring = &dev_priv->ring[VECS]; | |
1589 | ||
1590 | ring->name = "video enhancement ring"; | |
1591 | ring->id = VECS; | |
1592 | ring->mmio_base = VEBOX_RING_BASE; | |
1593 | ring->irq_enable_mask = | |
1594 | GT_RENDER_USER_INTERRUPT << GEN8_VECS_IRQ_SHIFT; | |
73d477f6 OM |
1595 | ring->irq_keep_mask = |
1596 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VECS_IRQ_SHIFT; | |
454afebd | 1597 | |
ecfe00d8 | 1598 | ring->init_hw = gen8_init_common_ring; |
e94e37ad OM |
1599 | ring->get_seqno = gen8_get_seqno; |
1600 | ring->set_seqno = gen8_set_seqno; | |
4da46e1e | 1601 | ring->emit_request = gen8_emit_request; |
4712274c | 1602 | ring->emit_flush = gen8_emit_flush; |
73d477f6 OM |
1603 | ring->irq_get = gen8_logical_ring_get_irq; |
1604 | ring->irq_put = gen8_logical_ring_put_irq; | |
15648585 | 1605 | ring->emit_bb_start = gen8_emit_bb_start; |
9b1136d5 | 1606 | |
454afebd OM |
1607 | return logical_ring_init(dev, ring); |
1608 | } | |
1609 | ||
73e4d07f OM |
1610 | /** |
1611 | * intel_logical_rings_init() - allocate, populate and init the Engine Command Streamers | |
1612 | * @dev: DRM device. | |
1613 | * | |
1614 | * This function inits the engines for an Execlists submission style (the equivalent in the | |
1615 | * legacy ringbuffer submission world would be i915_gem_init_rings). It does it only for | |
1616 | * those engines that are present in the hardware. | |
1617 | * | |
1618 | * Return: non-zero if the initialization failed. | |
1619 | */ | |
454afebd OM |
1620 | int intel_logical_rings_init(struct drm_device *dev) |
1621 | { | |
1622 | struct drm_i915_private *dev_priv = dev->dev_private; | |
1623 | int ret; | |
1624 | ||
1625 | ret = logical_render_ring_init(dev); | |
1626 | if (ret) | |
1627 | return ret; | |
1628 | ||
1629 | if (HAS_BSD(dev)) { | |
1630 | ret = logical_bsd_ring_init(dev); | |
1631 | if (ret) | |
1632 | goto cleanup_render_ring; | |
1633 | } | |
1634 | ||
1635 | if (HAS_BLT(dev)) { | |
1636 | ret = logical_blt_ring_init(dev); | |
1637 | if (ret) | |
1638 | goto cleanup_bsd_ring; | |
1639 | } | |
1640 | ||
1641 | if (HAS_VEBOX(dev)) { | |
1642 | ret = logical_vebox_ring_init(dev); | |
1643 | if (ret) | |
1644 | goto cleanup_blt_ring; | |
1645 | } | |
1646 | ||
1647 | if (HAS_BSD2(dev)) { | |
1648 | ret = logical_bsd2_ring_init(dev); | |
1649 | if (ret) | |
1650 | goto cleanup_vebox_ring; | |
1651 | } | |
1652 | ||
1653 | ret = i915_gem_set_seqno(dev, ((u32)~0 - 0x1000)); | |
1654 | if (ret) | |
1655 | goto cleanup_bsd2_ring; | |
1656 | ||
1657 | return 0; | |
1658 | ||
1659 | cleanup_bsd2_ring: | |
1660 | intel_logical_ring_cleanup(&dev_priv->ring[VCS2]); | |
1661 | cleanup_vebox_ring: | |
1662 | intel_logical_ring_cleanup(&dev_priv->ring[VECS]); | |
1663 | cleanup_blt_ring: | |
1664 | intel_logical_ring_cleanup(&dev_priv->ring[BCS]); | |
1665 | cleanup_bsd_ring: | |
1666 | intel_logical_ring_cleanup(&dev_priv->ring[VCS]); | |
1667 | cleanup_render_ring: | |
1668 | intel_logical_ring_cleanup(&dev_priv->ring[RCS]); | |
1669 | ||
1670 | return ret; | |
1671 | } | |
1672 | ||
0cea6502 JM |
1673 | static u32 |
1674 | make_rpcs(struct drm_device *dev) | |
1675 | { | |
1676 | u32 rpcs = 0; | |
1677 | ||
1678 | /* | |
1679 | * No explicit RPCS request is needed to ensure full | |
1680 | * slice/subslice/EU enablement prior to Gen9. | |
1681 | */ | |
1682 | if (INTEL_INFO(dev)->gen < 9) | |
1683 | return 0; | |
1684 | ||
1685 | /* | |
1686 | * Starting in Gen9, render power gating can leave | |
1687 | * slice/subslice/EU in a partially enabled state. We | |
1688 | * must make an explicit request through RPCS for full | |
1689 | * enablement. | |
1690 | */ | |
1691 | if (INTEL_INFO(dev)->has_slice_pg) { | |
1692 | rpcs |= GEN8_RPCS_S_CNT_ENABLE; | |
1693 | rpcs |= INTEL_INFO(dev)->slice_total << | |
1694 | GEN8_RPCS_S_CNT_SHIFT; | |
1695 | rpcs |= GEN8_RPCS_ENABLE; | |
1696 | } | |
1697 | ||
1698 | if (INTEL_INFO(dev)->has_subslice_pg) { | |
1699 | rpcs |= GEN8_RPCS_SS_CNT_ENABLE; | |
1700 | rpcs |= INTEL_INFO(dev)->subslice_per_slice << | |
1701 | GEN8_RPCS_SS_CNT_SHIFT; | |
1702 | rpcs |= GEN8_RPCS_ENABLE; | |
1703 | } | |
1704 | ||
1705 | if (INTEL_INFO(dev)->has_eu_pg) { | |
1706 | rpcs |= INTEL_INFO(dev)->eu_per_subslice << | |
1707 | GEN8_RPCS_EU_MIN_SHIFT; | |
1708 | rpcs |= INTEL_INFO(dev)->eu_per_subslice << | |
1709 | GEN8_RPCS_EU_MAX_SHIFT; | |
1710 | rpcs |= GEN8_RPCS_ENABLE; | |
1711 | } | |
1712 | ||
1713 | return rpcs; | |
1714 | } | |
1715 | ||
8670d6f9 OM |
1716 | static int |
1717 | populate_lr_context(struct intel_context *ctx, struct drm_i915_gem_object *ctx_obj, | |
1718 | struct intel_engine_cs *ring, struct intel_ringbuffer *ringbuf) | |
1719 | { | |
2d965536 TD |
1720 | struct drm_device *dev = ring->dev; |
1721 | struct drm_i915_private *dev_priv = dev->dev_private; | |
ae6c4806 | 1722 | struct i915_hw_ppgtt *ppgtt = ctx->ppgtt; |
8670d6f9 OM |
1723 | struct page *page; |
1724 | uint32_t *reg_state; | |
1725 | int ret; | |
1726 | ||
2d965536 TD |
1727 | if (!ppgtt) |
1728 | ppgtt = dev_priv->mm.aliasing_ppgtt; | |
1729 | ||
8670d6f9 OM |
1730 | ret = i915_gem_object_set_to_cpu_domain(ctx_obj, true); |
1731 | if (ret) { | |
1732 | DRM_DEBUG_DRIVER("Could not set to CPU domain\n"); | |
1733 | return ret; | |
1734 | } | |
1735 | ||
1736 | ret = i915_gem_object_get_pages(ctx_obj); | |
1737 | if (ret) { | |
1738 | DRM_DEBUG_DRIVER("Could not get object pages\n"); | |
1739 | return ret; | |
1740 | } | |
1741 | ||
1742 | i915_gem_object_pin_pages(ctx_obj); | |
1743 | ||
1744 | /* The second page of the context object contains some fields which must | |
1745 | * be set up prior to the first execution. */ | |
1746 | page = i915_gem_object_get_page(ctx_obj, 1); | |
1747 | reg_state = kmap_atomic(page); | |
1748 | ||
1749 | /* A context is actually a big batch buffer with several MI_LOAD_REGISTER_IMM | |
1750 | * commands followed by (reg, value) pairs. The values we are setting here are | |
1751 | * only for the first context restore: on a subsequent save, the GPU will | |
1752 | * recreate this batchbuffer with new values (including all the missing | |
1753 | * MI_LOAD_REGISTER_IMM commands that we are not initializing here). */ | |
1754 | if (ring->id == RCS) | |
1755 | reg_state[CTX_LRI_HEADER_0] = MI_LOAD_REGISTER_IMM(14); | |
1756 | else | |
1757 | reg_state[CTX_LRI_HEADER_0] = MI_LOAD_REGISTER_IMM(11); | |
1758 | reg_state[CTX_LRI_HEADER_0] |= MI_LRI_FORCE_POSTED; | |
1759 | reg_state[CTX_CONTEXT_CONTROL] = RING_CONTEXT_CONTROL(ring); | |
1760 | reg_state[CTX_CONTEXT_CONTROL+1] = | |
5baa22c5 ZW |
1761 | _MASKED_BIT_ENABLE(CTX_CTRL_INHIBIT_SYN_CTX_SWITCH | |
1762 | CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT); | |
8670d6f9 OM |
1763 | reg_state[CTX_RING_HEAD] = RING_HEAD(ring->mmio_base); |
1764 | reg_state[CTX_RING_HEAD+1] = 0; | |
1765 | reg_state[CTX_RING_TAIL] = RING_TAIL(ring->mmio_base); | |
1766 | reg_state[CTX_RING_TAIL+1] = 0; | |
1767 | reg_state[CTX_RING_BUFFER_START] = RING_START(ring->mmio_base); | |
7ba717cf TD |
1768 | /* Ring buffer start address is not known until the buffer is pinned. |
1769 | * It is written to the context image in execlists_update_context() | |
1770 | */ | |
8670d6f9 OM |
1771 | reg_state[CTX_RING_BUFFER_CONTROL] = RING_CTL(ring->mmio_base); |
1772 | reg_state[CTX_RING_BUFFER_CONTROL+1] = | |
1773 | ((ringbuf->size - PAGE_SIZE) & RING_NR_PAGES) | RING_VALID; | |
1774 | reg_state[CTX_BB_HEAD_U] = ring->mmio_base + 0x168; | |
1775 | reg_state[CTX_BB_HEAD_U+1] = 0; | |
1776 | reg_state[CTX_BB_HEAD_L] = ring->mmio_base + 0x140; | |
1777 | reg_state[CTX_BB_HEAD_L+1] = 0; | |
1778 | reg_state[CTX_BB_STATE] = ring->mmio_base + 0x110; | |
1779 | reg_state[CTX_BB_STATE+1] = (1<<5); | |
1780 | reg_state[CTX_SECOND_BB_HEAD_U] = ring->mmio_base + 0x11c; | |
1781 | reg_state[CTX_SECOND_BB_HEAD_U+1] = 0; | |
1782 | reg_state[CTX_SECOND_BB_HEAD_L] = ring->mmio_base + 0x114; | |
1783 | reg_state[CTX_SECOND_BB_HEAD_L+1] = 0; | |
1784 | reg_state[CTX_SECOND_BB_STATE] = ring->mmio_base + 0x118; | |
1785 | reg_state[CTX_SECOND_BB_STATE+1] = 0; | |
1786 | if (ring->id == RCS) { | |
1787 | /* TODO: according to BSpec, the register state context | |
1788 | * for CHV does not have these. OTOH, these registers do | |
1789 | * exist in CHV. I'm waiting for a clarification */ | |
1790 | reg_state[CTX_BB_PER_CTX_PTR] = ring->mmio_base + 0x1c0; | |
1791 | reg_state[CTX_BB_PER_CTX_PTR+1] = 0; | |
1792 | reg_state[CTX_RCS_INDIRECT_CTX] = ring->mmio_base + 0x1c4; | |
1793 | reg_state[CTX_RCS_INDIRECT_CTX+1] = 0; | |
1794 | reg_state[CTX_RCS_INDIRECT_CTX_OFFSET] = ring->mmio_base + 0x1c8; | |
1795 | reg_state[CTX_RCS_INDIRECT_CTX_OFFSET+1] = 0; | |
1796 | } | |
1797 | reg_state[CTX_LRI_HEADER_1] = MI_LOAD_REGISTER_IMM(9); | |
1798 | reg_state[CTX_LRI_HEADER_1] |= MI_LRI_FORCE_POSTED; | |
1799 | reg_state[CTX_CTX_TIMESTAMP] = ring->mmio_base + 0x3a8; | |
1800 | reg_state[CTX_CTX_TIMESTAMP+1] = 0; | |
1801 | reg_state[CTX_PDP3_UDW] = GEN8_RING_PDP_UDW(ring, 3); | |
1802 | reg_state[CTX_PDP3_LDW] = GEN8_RING_PDP_LDW(ring, 3); | |
1803 | reg_state[CTX_PDP2_UDW] = GEN8_RING_PDP_UDW(ring, 2); | |
1804 | reg_state[CTX_PDP2_LDW] = GEN8_RING_PDP_LDW(ring, 2); | |
1805 | reg_state[CTX_PDP1_UDW] = GEN8_RING_PDP_UDW(ring, 1); | |
1806 | reg_state[CTX_PDP1_LDW] = GEN8_RING_PDP_LDW(ring, 1); | |
1807 | reg_state[CTX_PDP0_UDW] = GEN8_RING_PDP_UDW(ring, 0); | |
1808 | reg_state[CTX_PDP0_LDW] = GEN8_RING_PDP_LDW(ring, 0); | |
06fda602 BW |
1809 | reg_state[CTX_PDP3_UDW+1] = upper_32_bits(ppgtt->pdp.page_directory[3]->daddr); |
1810 | reg_state[CTX_PDP3_LDW+1] = lower_32_bits(ppgtt->pdp.page_directory[3]->daddr); | |
1811 | reg_state[CTX_PDP2_UDW+1] = upper_32_bits(ppgtt->pdp.page_directory[2]->daddr); | |
1812 | reg_state[CTX_PDP2_LDW+1] = lower_32_bits(ppgtt->pdp.page_directory[2]->daddr); | |
1813 | reg_state[CTX_PDP1_UDW+1] = upper_32_bits(ppgtt->pdp.page_directory[1]->daddr); | |
1814 | reg_state[CTX_PDP1_LDW+1] = lower_32_bits(ppgtt->pdp.page_directory[1]->daddr); | |
1815 | reg_state[CTX_PDP0_UDW+1] = upper_32_bits(ppgtt->pdp.page_directory[0]->daddr); | |
1816 | reg_state[CTX_PDP0_LDW+1] = lower_32_bits(ppgtt->pdp.page_directory[0]->daddr); | |
8670d6f9 OM |
1817 | if (ring->id == RCS) { |
1818 | reg_state[CTX_LRI_HEADER_2] = MI_LOAD_REGISTER_IMM(1); | |
0cea6502 JM |
1819 | reg_state[CTX_R_PWR_CLK_STATE] = GEN8_R_PWR_CLK_STATE; |
1820 | reg_state[CTX_R_PWR_CLK_STATE+1] = make_rpcs(dev); | |
8670d6f9 OM |
1821 | } |
1822 | ||
1823 | kunmap_atomic(reg_state); | |
1824 | ||
1825 | ctx_obj->dirty = 1; | |
1826 | set_page_dirty(page); | |
1827 | i915_gem_object_unpin_pages(ctx_obj); | |
1828 | ||
1829 | return 0; | |
1830 | } | |
1831 | ||
73e4d07f OM |
1832 | /** |
1833 | * intel_lr_context_free() - free the LRC specific bits of a context | |
1834 | * @ctx: the LR context to free. | |
1835 | * | |
1836 | * The real context freeing is done in i915_gem_context_free: this only | |
1837 | * takes care of the bits that are LRC related: the per-engine backing | |
1838 | * objects and the logical ringbuffer. | |
1839 | */ | |
ede7d42b OM |
1840 | void intel_lr_context_free(struct intel_context *ctx) |
1841 | { | |
8c857917 OM |
1842 | int i; |
1843 | ||
1844 | for (i = 0; i < I915_NUM_RINGS; i++) { | |
1845 | struct drm_i915_gem_object *ctx_obj = ctx->engine[i].state; | |
84c2377f | 1846 | |
8c857917 | 1847 | if (ctx_obj) { |
dcb4c12a OM |
1848 | struct intel_ringbuffer *ringbuf = |
1849 | ctx->engine[i].ringbuf; | |
1850 | struct intel_engine_cs *ring = ringbuf->ring; | |
1851 | ||
7ba717cf TD |
1852 | if (ctx == ring->default_context) { |
1853 | intel_unpin_ringbuffer_obj(ringbuf); | |
1854 | i915_gem_object_ggtt_unpin(ctx_obj); | |
1855 | } | |
a7cbedec | 1856 | WARN_ON(ctx->engine[ring->id].pin_count); |
84c2377f OM |
1857 | intel_destroy_ringbuffer_obj(ringbuf); |
1858 | kfree(ringbuf); | |
8c857917 OM |
1859 | drm_gem_object_unreference(&ctx_obj->base); |
1860 | } | |
1861 | } | |
1862 | } | |
1863 | ||
1864 | static uint32_t get_lr_context_size(struct intel_engine_cs *ring) | |
1865 | { | |
1866 | int ret = 0; | |
1867 | ||
468c6816 | 1868 | WARN_ON(INTEL_INFO(ring->dev)->gen < 8); |
8c857917 OM |
1869 | |
1870 | switch (ring->id) { | |
1871 | case RCS: | |
468c6816 MN |
1872 | if (INTEL_INFO(ring->dev)->gen >= 9) |
1873 | ret = GEN9_LR_CONTEXT_RENDER_SIZE; | |
1874 | else | |
1875 | ret = GEN8_LR_CONTEXT_RENDER_SIZE; | |
8c857917 OM |
1876 | break; |
1877 | case VCS: | |
1878 | case BCS: | |
1879 | case VECS: | |
1880 | case VCS2: | |
1881 | ret = GEN8_LR_CONTEXT_OTHER_SIZE; | |
1882 | break; | |
1883 | } | |
1884 | ||
1885 | return ret; | |
ede7d42b OM |
1886 | } |
1887 | ||
70b0ea86 | 1888 | static void lrc_setup_hardware_status_page(struct intel_engine_cs *ring, |
1df06b75 TD |
1889 | struct drm_i915_gem_object *default_ctx_obj) |
1890 | { | |
1891 | struct drm_i915_private *dev_priv = ring->dev->dev_private; | |
1892 | ||
1893 | /* The status page is offset 0 from the default context object | |
1894 | * in LRC mode. */ | |
1895 | ring->status_page.gfx_addr = i915_gem_obj_ggtt_offset(default_ctx_obj); | |
1896 | ring->status_page.page_addr = | |
1897 | kmap(sg_page(default_ctx_obj->pages->sgl)); | |
1df06b75 TD |
1898 | ring->status_page.obj = default_ctx_obj; |
1899 | ||
1900 | I915_WRITE(RING_HWS_PGA(ring->mmio_base), | |
1901 | (u32)ring->status_page.gfx_addr); | |
1902 | POSTING_READ(RING_HWS_PGA(ring->mmio_base)); | |
1df06b75 TD |
1903 | } |
1904 | ||
73e4d07f OM |
1905 | /** |
1906 | * intel_lr_context_deferred_create() - create the LRC specific bits of a context | |
1907 | * @ctx: LR context to create. | |
1908 | * @ring: engine to be used with the context. | |
1909 | * | |
1910 | * This function can be called more than once, with different engines, if we plan | |
1911 | * to use the context with them. The context backing objects and the ringbuffers | |
1912 | * (specially the ringbuffer backing objects) suck a lot of memory up, and that's why | |
1913 | * the creation is a deferred call: it's better to make sure first that we need to use | |
1914 | * a given ring with the context. | |
1915 | * | |
32197aab | 1916 | * Return: non-zero on error. |
73e4d07f | 1917 | */ |
ede7d42b OM |
1918 | int intel_lr_context_deferred_create(struct intel_context *ctx, |
1919 | struct intel_engine_cs *ring) | |
1920 | { | |
dcb4c12a | 1921 | const bool is_global_default_ctx = (ctx == ring->default_context); |
8c857917 OM |
1922 | struct drm_device *dev = ring->dev; |
1923 | struct drm_i915_gem_object *ctx_obj; | |
1924 | uint32_t context_size; | |
84c2377f | 1925 | struct intel_ringbuffer *ringbuf; |
8c857917 OM |
1926 | int ret; |
1927 | ||
ede7d42b | 1928 | WARN_ON(ctx->legacy_hw_ctx.rcs_state != NULL); |
bfc882b4 | 1929 | WARN_ON(ctx->engine[ring->id].state); |
ede7d42b | 1930 | |
8c857917 OM |
1931 | context_size = round_up(get_lr_context_size(ring), 4096); |
1932 | ||
1933 | ctx_obj = i915_gem_alloc_context_obj(dev, context_size); | |
1934 | if (IS_ERR(ctx_obj)) { | |
1935 | ret = PTR_ERR(ctx_obj); | |
1936 | DRM_DEBUG_DRIVER("Alloc LRC backing obj failed: %d\n", ret); | |
1937 | return ret; | |
1938 | } | |
1939 | ||
dcb4c12a OM |
1940 | if (is_global_default_ctx) { |
1941 | ret = i915_gem_obj_ggtt_pin(ctx_obj, GEN8_LR_CONTEXT_ALIGN, 0); | |
1942 | if (ret) { | |
1943 | DRM_DEBUG_DRIVER("Pin LRC backing obj failed: %d\n", | |
1944 | ret); | |
1945 | drm_gem_object_unreference(&ctx_obj->base); | |
1946 | return ret; | |
1947 | } | |
8c857917 OM |
1948 | } |
1949 | ||
84c2377f OM |
1950 | ringbuf = kzalloc(sizeof(*ringbuf), GFP_KERNEL); |
1951 | if (!ringbuf) { | |
1952 | DRM_DEBUG_DRIVER("Failed to allocate ringbuffer %s\n", | |
1953 | ring->name); | |
84c2377f | 1954 | ret = -ENOMEM; |
7ba717cf | 1955 | goto error_unpin_ctx; |
84c2377f OM |
1956 | } |
1957 | ||
0c7dd53b | 1958 | ringbuf->ring = ring; |
582d67f0 | 1959 | |
84c2377f OM |
1960 | ringbuf->size = 32 * PAGE_SIZE; |
1961 | ringbuf->effective_size = ringbuf->size; | |
1962 | ringbuf->head = 0; | |
1963 | ringbuf->tail = 0; | |
84c2377f | 1964 | ringbuf->last_retired_head = -1; |
ebd0fd4b | 1965 | intel_ring_update_space(ringbuf); |
84c2377f | 1966 | |
7ba717cf TD |
1967 | if (ringbuf->obj == NULL) { |
1968 | ret = intel_alloc_ringbuffer_obj(dev, ringbuf); | |
1969 | if (ret) { | |
1970 | DRM_DEBUG_DRIVER( | |
1971 | "Failed to allocate ringbuffer obj %s: %d\n", | |
84c2377f | 1972 | ring->name, ret); |
7ba717cf TD |
1973 | goto error_free_rbuf; |
1974 | } | |
1975 | ||
1976 | if (is_global_default_ctx) { | |
1977 | ret = intel_pin_and_map_ringbuffer_obj(dev, ringbuf); | |
1978 | if (ret) { | |
1979 | DRM_ERROR( | |
1980 | "Failed to pin and map ringbuffer %s: %d\n", | |
1981 | ring->name, ret); | |
1982 | goto error_destroy_rbuf; | |
1983 | } | |
1984 | } | |
1985 | ||
8670d6f9 OM |
1986 | } |
1987 | ||
1988 | ret = populate_lr_context(ctx, ctx_obj, ring, ringbuf); | |
1989 | if (ret) { | |
1990 | DRM_DEBUG_DRIVER("Failed to populate LRC: %d\n", ret); | |
8670d6f9 | 1991 | goto error; |
84c2377f OM |
1992 | } |
1993 | ||
1994 | ctx->engine[ring->id].ringbuf = ringbuf; | |
8c857917 | 1995 | ctx->engine[ring->id].state = ctx_obj; |
ede7d42b | 1996 | |
70b0ea86 DV |
1997 | if (ctx == ring->default_context) |
1998 | lrc_setup_hardware_status_page(ring, ctx_obj); | |
e7778be1 | 1999 | else if (ring->id == RCS && !ctx->rcs_initialized) { |
771b9a53 MT |
2000 | if (ring->init_context) { |
2001 | ret = ring->init_context(ring, ctx); | |
e7778be1 | 2002 | if (ret) { |
771b9a53 | 2003 | DRM_ERROR("ring init context: %d\n", ret); |
e7778be1 TD |
2004 | ctx->engine[ring->id].ringbuf = NULL; |
2005 | ctx->engine[ring->id].state = NULL; | |
2006 | goto error; | |
2007 | } | |
771b9a53 MT |
2008 | } |
2009 | ||
564ddb2f OM |
2010 | ctx->rcs_initialized = true; |
2011 | } | |
2012 | ||
ede7d42b | 2013 | return 0; |
8670d6f9 OM |
2014 | |
2015 | error: | |
7ba717cf TD |
2016 | if (is_global_default_ctx) |
2017 | intel_unpin_ringbuffer_obj(ringbuf); | |
2018 | error_destroy_rbuf: | |
2019 | intel_destroy_ringbuffer_obj(ringbuf); | |
2020 | error_free_rbuf: | |
8670d6f9 | 2021 | kfree(ringbuf); |
7ba717cf | 2022 | error_unpin_ctx: |
dcb4c12a OM |
2023 | if (is_global_default_ctx) |
2024 | i915_gem_object_ggtt_unpin(ctx_obj); | |
8670d6f9 OM |
2025 | drm_gem_object_unreference(&ctx_obj->base); |
2026 | return ret; | |
ede7d42b | 2027 | } |
3e5b6f05 TD |
2028 | |
2029 | void intel_lr_context_reset(struct drm_device *dev, | |
2030 | struct intel_context *ctx) | |
2031 | { | |
2032 | struct drm_i915_private *dev_priv = dev->dev_private; | |
2033 | struct intel_engine_cs *ring; | |
2034 | int i; | |
2035 | ||
2036 | for_each_ring(ring, dev_priv, i) { | |
2037 | struct drm_i915_gem_object *ctx_obj = | |
2038 | ctx->engine[ring->id].state; | |
2039 | struct intel_ringbuffer *ringbuf = | |
2040 | ctx->engine[ring->id].ringbuf; | |
2041 | uint32_t *reg_state; | |
2042 | struct page *page; | |
2043 | ||
2044 | if (!ctx_obj) | |
2045 | continue; | |
2046 | ||
2047 | if (i915_gem_object_get_pages(ctx_obj)) { | |
2048 | WARN(1, "Failed get_pages for context obj\n"); | |
2049 | continue; | |
2050 | } | |
2051 | page = i915_gem_object_get_page(ctx_obj, 1); | |
2052 | reg_state = kmap_atomic(page); | |
2053 | ||
2054 | reg_state[CTX_RING_HEAD+1] = 0; | |
2055 | reg_state[CTX_RING_TAIL+1] = 0; | |
2056 | ||
2057 | kunmap_atomic(reg_state); | |
2058 | ||
2059 | ringbuf->head = 0; | |
2060 | ringbuf->tail = 0; | |
2061 | } | |
2062 | } |