Commit | Line | Data |
---|---|---|
62fdfeaf EA |
1 | /* |
2 | * Copyright © 2008-2010 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 | * Eric Anholt <eric@anholt.net> | |
25 | * Zou Nan hai <nanhai.zou@intel.com> | |
26 | * Xiang Hai hao<haihao.xiang@intel.com> | |
27 | * | |
28 | */ | |
29 | ||
760285e7 | 30 | #include <drm/drmP.h> |
62fdfeaf | 31 | #include "i915_drv.h" |
760285e7 | 32 | #include <drm/i915_drm.h> |
62fdfeaf | 33 | #include "i915_trace.h" |
881f47b6 | 34 | #include "intel_drv.h" |
62fdfeaf | 35 | |
8d315287 JB |
36 | /* |
37 | * 965+ support PIPE_CONTROL commands, which provide finer grained control | |
38 | * over cache flushing. | |
39 | */ | |
40 | struct pipe_control { | |
41 | struct drm_i915_gem_object *obj; | |
42 | volatile u32 *cpu_page; | |
43 | u32 gtt_offset; | |
44 | }; | |
45 | ||
c7dca47b CW |
46 | static inline int ring_space(struct intel_ring_buffer *ring) |
47 | { | |
48 | int space = (ring->head & HEAD_ADDR) - (ring->tail + 8); | |
49 | if (space < 0) | |
50 | space += ring->size; | |
51 | return space; | |
52 | } | |
53 | ||
b72f3acb | 54 | static int |
46f0f8d1 CW |
55 | gen2_render_ring_flush(struct intel_ring_buffer *ring, |
56 | u32 invalidate_domains, | |
57 | u32 flush_domains) | |
58 | { | |
59 | u32 cmd; | |
60 | int ret; | |
61 | ||
62 | cmd = MI_FLUSH; | |
31b14c9f | 63 | if (((invalidate_domains|flush_domains) & I915_GEM_DOMAIN_RENDER) == 0) |
46f0f8d1 CW |
64 | cmd |= MI_NO_WRITE_FLUSH; |
65 | ||
66 | if (invalidate_domains & I915_GEM_DOMAIN_SAMPLER) | |
67 | cmd |= MI_READ_FLUSH; | |
68 | ||
69 | ret = intel_ring_begin(ring, 2); | |
70 | if (ret) | |
71 | return ret; | |
72 | ||
73 | intel_ring_emit(ring, cmd); | |
74 | intel_ring_emit(ring, MI_NOOP); | |
75 | intel_ring_advance(ring); | |
76 | ||
77 | return 0; | |
78 | } | |
79 | ||
80 | static int | |
81 | gen4_render_ring_flush(struct intel_ring_buffer *ring, | |
82 | u32 invalidate_domains, | |
83 | u32 flush_domains) | |
62fdfeaf | 84 | { |
78501eac | 85 | struct drm_device *dev = ring->dev; |
6f392d54 | 86 | u32 cmd; |
b72f3acb | 87 | int ret; |
6f392d54 | 88 | |
36d527de CW |
89 | /* |
90 | * read/write caches: | |
91 | * | |
92 | * I915_GEM_DOMAIN_RENDER is always invalidated, but is | |
93 | * only flushed if MI_NO_WRITE_FLUSH is unset. On 965, it is | |
94 | * also flushed at 2d versus 3d pipeline switches. | |
95 | * | |
96 | * read-only caches: | |
97 | * | |
98 | * I915_GEM_DOMAIN_SAMPLER is flushed on pre-965 if | |
99 | * MI_READ_FLUSH is set, and is always flushed on 965. | |
100 | * | |
101 | * I915_GEM_DOMAIN_COMMAND may not exist? | |
102 | * | |
103 | * I915_GEM_DOMAIN_INSTRUCTION, which exists on 965, is | |
104 | * invalidated when MI_EXE_FLUSH is set. | |
105 | * | |
106 | * I915_GEM_DOMAIN_VERTEX, which exists on 965, is | |
107 | * invalidated with every MI_FLUSH. | |
108 | * | |
109 | * TLBs: | |
110 | * | |
111 | * On 965, TLBs associated with I915_GEM_DOMAIN_COMMAND | |
112 | * and I915_GEM_DOMAIN_CPU in are invalidated at PTE write and | |
113 | * I915_GEM_DOMAIN_RENDER and I915_GEM_DOMAIN_SAMPLER | |
114 | * are flushed at any MI_FLUSH. | |
115 | */ | |
116 | ||
117 | cmd = MI_FLUSH | MI_NO_WRITE_FLUSH; | |
46f0f8d1 | 118 | if ((invalidate_domains|flush_domains) & I915_GEM_DOMAIN_RENDER) |
36d527de | 119 | cmd &= ~MI_NO_WRITE_FLUSH; |
36d527de CW |
120 | if (invalidate_domains & I915_GEM_DOMAIN_INSTRUCTION) |
121 | cmd |= MI_EXE_FLUSH; | |
62fdfeaf | 122 | |
36d527de CW |
123 | if (invalidate_domains & I915_GEM_DOMAIN_COMMAND && |
124 | (IS_G4X(dev) || IS_GEN5(dev))) | |
125 | cmd |= MI_INVALIDATE_ISP; | |
70eac33e | 126 | |
36d527de CW |
127 | ret = intel_ring_begin(ring, 2); |
128 | if (ret) | |
129 | return ret; | |
b72f3acb | 130 | |
36d527de CW |
131 | intel_ring_emit(ring, cmd); |
132 | intel_ring_emit(ring, MI_NOOP); | |
133 | intel_ring_advance(ring); | |
b72f3acb CW |
134 | |
135 | return 0; | |
8187a2b7 ZN |
136 | } |
137 | ||
8d315287 JB |
138 | /** |
139 | * Emits a PIPE_CONTROL with a non-zero post-sync operation, for | |
140 | * implementing two workarounds on gen6. From section 1.4.7.1 | |
141 | * "PIPE_CONTROL" of the Sandy Bridge PRM volume 2 part 1: | |
142 | * | |
143 | * [DevSNB-C+{W/A}] Before any depth stall flush (including those | |
144 | * produced by non-pipelined state commands), software needs to first | |
145 | * send a PIPE_CONTROL with no bits set except Post-Sync Operation != | |
146 | * 0. | |
147 | * | |
148 | * [Dev-SNB{W/A}]: Before a PIPE_CONTROL with Write Cache Flush Enable | |
149 | * =1, a PIPE_CONTROL with any non-zero post-sync-op is required. | |
150 | * | |
151 | * And the workaround for these two requires this workaround first: | |
152 | * | |
153 | * [Dev-SNB{W/A}]: Pipe-control with CS-stall bit set must be sent | |
154 | * BEFORE the pipe-control with a post-sync op and no write-cache | |
155 | * flushes. | |
156 | * | |
157 | * And this last workaround is tricky because of the requirements on | |
158 | * that bit. From section 1.4.7.2.3 "Stall" of the Sandy Bridge PRM | |
159 | * volume 2 part 1: | |
160 | * | |
161 | * "1 of the following must also be set: | |
162 | * - Render Target Cache Flush Enable ([12] of DW1) | |
163 | * - Depth Cache Flush Enable ([0] of DW1) | |
164 | * - Stall at Pixel Scoreboard ([1] of DW1) | |
165 | * - Depth Stall ([13] of DW1) | |
166 | * - Post-Sync Operation ([13] of DW1) | |
167 | * - Notify Enable ([8] of DW1)" | |
168 | * | |
169 | * The cache flushes require the workaround flush that triggered this | |
170 | * one, so we can't use it. Depth stall would trigger the same. | |
171 | * Post-sync nonzero is what triggered this second workaround, so we | |
172 | * can't use that one either. Notify enable is IRQs, which aren't | |
173 | * really our business. That leaves only stall at scoreboard. | |
174 | */ | |
175 | static int | |
176 | intel_emit_post_sync_nonzero_flush(struct intel_ring_buffer *ring) | |
177 | { | |
178 | struct pipe_control *pc = ring->private; | |
179 | u32 scratch_addr = pc->gtt_offset + 128; | |
180 | int ret; | |
181 | ||
182 | ||
183 | ret = intel_ring_begin(ring, 6); | |
184 | if (ret) | |
185 | return ret; | |
186 | ||
187 | intel_ring_emit(ring, GFX_OP_PIPE_CONTROL(5)); | |
188 | intel_ring_emit(ring, PIPE_CONTROL_CS_STALL | | |
189 | PIPE_CONTROL_STALL_AT_SCOREBOARD); | |
190 | intel_ring_emit(ring, scratch_addr | PIPE_CONTROL_GLOBAL_GTT); /* address */ | |
191 | intel_ring_emit(ring, 0); /* low dword */ | |
192 | intel_ring_emit(ring, 0); /* high dword */ | |
193 | intel_ring_emit(ring, MI_NOOP); | |
194 | intel_ring_advance(ring); | |
195 | ||
196 | ret = intel_ring_begin(ring, 6); | |
197 | if (ret) | |
198 | return ret; | |
199 | ||
200 | intel_ring_emit(ring, GFX_OP_PIPE_CONTROL(5)); | |
201 | intel_ring_emit(ring, PIPE_CONTROL_QW_WRITE); | |
202 | intel_ring_emit(ring, scratch_addr | PIPE_CONTROL_GLOBAL_GTT); /* address */ | |
203 | intel_ring_emit(ring, 0); | |
204 | intel_ring_emit(ring, 0); | |
205 | intel_ring_emit(ring, MI_NOOP); | |
206 | intel_ring_advance(ring); | |
207 | ||
208 | return 0; | |
209 | } | |
210 | ||
211 | static int | |
212 | gen6_render_ring_flush(struct intel_ring_buffer *ring, | |
213 | u32 invalidate_domains, u32 flush_domains) | |
214 | { | |
215 | u32 flags = 0; | |
216 | struct pipe_control *pc = ring->private; | |
217 | u32 scratch_addr = pc->gtt_offset + 128; | |
218 | int ret; | |
219 | ||
b3111509 PZ |
220 | /* Force SNB workarounds for PIPE_CONTROL flushes */ |
221 | ret = intel_emit_post_sync_nonzero_flush(ring); | |
222 | if (ret) | |
223 | return ret; | |
224 | ||
8d315287 JB |
225 | /* Just flush everything. Experiments have shown that reducing the |
226 | * number of bits based on the write domains has little performance | |
227 | * impact. | |
228 | */ | |
7d54a904 CW |
229 | if (flush_domains) { |
230 | flags |= PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH; | |
231 | flags |= PIPE_CONTROL_DEPTH_CACHE_FLUSH; | |
232 | /* | |
233 | * Ensure that any following seqno writes only happen | |
234 | * when the render cache is indeed flushed. | |
235 | */ | |
97f209bc | 236 | flags |= PIPE_CONTROL_CS_STALL; |
7d54a904 CW |
237 | } |
238 | if (invalidate_domains) { | |
239 | flags |= PIPE_CONTROL_TLB_INVALIDATE; | |
240 | flags |= PIPE_CONTROL_INSTRUCTION_CACHE_INVALIDATE; | |
241 | flags |= PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE; | |
242 | flags |= PIPE_CONTROL_VF_CACHE_INVALIDATE; | |
243 | flags |= PIPE_CONTROL_CONST_CACHE_INVALIDATE; | |
244 | flags |= PIPE_CONTROL_STATE_CACHE_INVALIDATE; | |
245 | /* | |
246 | * TLB invalidate requires a post-sync write. | |
247 | */ | |
3ac78313 | 248 | flags |= PIPE_CONTROL_QW_WRITE | PIPE_CONTROL_CS_STALL; |
7d54a904 | 249 | } |
8d315287 | 250 | |
6c6cf5aa | 251 | ret = intel_ring_begin(ring, 4); |
8d315287 JB |
252 | if (ret) |
253 | return ret; | |
254 | ||
6c6cf5aa | 255 | intel_ring_emit(ring, GFX_OP_PIPE_CONTROL(4)); |
8d315287 JB |
256 | intel_ring_emit(ring, flags); |
257 | intel_ring_emit(ring, scratch_addr | PIPE_CONTROL_GLOBAL_GTT); | |
6c6cf5aa | 258 | intel_ring_emit(ring, 0); |
8d315287 JB |
259 | intel_ring_advance(ring); |
260 | ||
261 | return 0; | |
262 | } | |
263 | ||
f3987631 PZ |
264 | static int |
265 | gen7_render_ring_cs_stall_wa(struct intel_ring_buffer *ring) | |
266 | { | |
267 | int ret; | |
268 | ||
269 | ret = intel_ring_begin(ring, 4); | |
270 | if (ret) | |
271 | return ret; | |
272 | ||
273 | intel_ring_emit(ring, GFX_OP_PIPE_CONTROL(4)); | |
274 | intel_ring_emit(ring, PIPE_CONTROL_CS_STALL | | |
275 | PIPE_CONTROL_STALL_AT_SCOREBOARD); | |
276 | intel_ring_emit(ring, 0); | |
277 | intel_ring_emit(ring, 0); | |
278 | intel_ring_advance(ring); | |
279 | ||
280 | return 0; | |
281 | } | |
282 | ||
4772eaeb PZ |
283 | static int |
284 | gen7_render_ring_flush(struct intel_ring_buffer *ring, | |
285 | u32 invalidate_domains, u32 flush_domains) | |
286 | { | |
287 | u32 flags = 0; | |
288 | struct pipe_control *pc = ring->private; | |
289 | u32 scratch_addr = pc->gtt_offset + 128; | |
290 | int ret; | |
291 | ||
f3987631 PZ |
292 | /* |
293 | * Ensure that any following seqno writes only happen when the render | |
294 | * cache is indeed flushed. | |
295 | * | |
296 | * Workaround: 4th PIPE_CONTROL command (except the ones with only | |
297 | * read-cache invalidate bits set) must have the CS_STALL bit set. We | |
298 | * don't try to be clever and just set it unconditionally. | |
299 | */ | |
300 | flags |= PIPE_CONTROL_CS_STALL; | |
301 | ||
4772eaeb PZ |
302 | /* Just flush everything. Experiments have shown that reducing the |
303 | * number of bits based on the write domains has little performance | |
304 | * impact. | |
305 | */ | |
306 | if (flush_domains) { | |
307 | flags |= PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH; | |
308 | flags |= PIPE_CONTROL_DEPTH_CACHE_FLUSH; | |
4772eaeb PZ |
309 | } |
310 | if (invalidate_domains) { | |
311 | flags |= PIPE_CONTROL_TLB_INVALIDATE; | |
312 | flags |= PIPE_CONTROL_INSTRUCTION_CACHE_INVALIDATE; | |
313 | flags |= PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE; | |
314 | flags |= PIPE_CONTROL_VF_CACHE_INVALIDATE; | |
315 | flags |= PIPE_CONTROL_CONST_CACHE_INVALIDATE; | |
316 | flags |= PIPE_CONTROL_STATE_CACHE_INVALIDATE; | |
317 | /* | |
318 | * TLB invalidate requires a post-sync write. | |
319 | */ | |
320 | flags |= PIPE_CONTROL_QW_WRITE; | |
f3987631 PZ |
321 | |
322 | /* Workaround: we must issue a pipe_control with CS-stall bit | |
323 | * set before a pipe_control command that has the state cache | |
324 | * invalidate bit set. */ | |
325 | gen7_render_ring_cs_stall_wa(ring); | |
4772eaeb PZ |
326 | } |
327 | ||
328 | ret = intel_ring_begin(ring, 4); | |
329 | if (ret) | |
330 | return ret; | |
331 | ||
332 | intel_ring_emit(ring, GFX_OP_PIPE_CONTROL(4)); | |
333 | intel_ring_emit(ring, flags); | |
334 | intel_ring_emit(ring, scratch_addr | PIPE_CONTROL_GLOBAL_GTT); | |
335 | intel_ring_emit(ring, 0); | |
336 | intel_ring_advance(ring); | |
337 | ||
338 | return 0; | |
339 | } | |
340 | ||
78501eac | 341 | static void ring_write_tail(struct intel_ring_buffer *ring, |
297b0c5b | 342 | u32 value) |
d46eefa2 | 343 | { |
78501eac | 344 | drm_i915_private_t *dev_priv = ring->dev->dev_private; |
297b0c5b | 345 | I915_WRITE_TAIL(ring, value); |
d46eefa2 XH |
346 | } |
347 | ||
78501eac | 348 | u32 intel_ring_get_active_head(struct intel_ring_buffer *ring) |
8187a2b7 | 349 | { |
78501eac CW |
350 | drm_i915_private_t *dev_priv = ring->dev->dev_private; |
351 | u32 acthd_reg = INTEL_INFO(ring->dev)->gen >= 4 ? | |
3d281d8c | 352 | RING_ACTHD(ring->mmio_base) : ACTHD; |
8187a2b7 ZN |
353 | |
354 | return I915_READ(acthd_reg); | |
355 | } | |
356 | ||
78501eac | 357 | static int init_ring_common(struct intel_ring_buffer *ring) |
8187a2b7 | 358 | { |
b7884eb4 DV |
359 | struct drm_device *dev = ring->dev; |
360 | drm_i915_private_t *dev_priv = dev->dev_private; | |
05394f39 | 361 | struct drm_i915_gem_object *obj = ring->obj; |
b7884eb4 | 362 | int ret = 0; |
8187a2b7 | 363 | u32 head; |
8187a2b7 | 364 | |
b7884eb4 DV |
365 | if (HAS_FORCE_WAKE(dev)) |
366 | gen6_gt_force_wake_get(dev_priv); | |
367 | ||
8187a2b7 | 368 | /* Stop the ring if it's running. */ |
7f2ab699 | 369 | I915_WRITE_CTL(ring, 0); |
570ef608 | 370 | I915_WRITE_HEAD(ring, 0); |
78501eac | 371 | ring->write_tail(ring, 0); |
8187a2b7 | 372 | |
570ef608 | 373 | head = I915_READ_HEAD(ring) & HEAD_ADDR; |
8187a2b7 ZN |
374 | |
375 | /* G45 ring initialization fails to reset head to zero */ | |
376 | if (head != 0) { | |
6fd0d56e CW |
377 | DRM_DEBUG_KMS("%s head not reset to zero " |
378 | "ctl %08x head %08x tail %08x start %08x\n", | |
379 | ring->name, | |
380 | I915_READ_CTL(ring), | |
381 | I915_READ_HEAD(ring), | |
382 | I915_READ_TAIL(ring), | |
383 | I915_READ_START(ring)); | |
8187a2b7 | 384 | |
570ef608 | 385 | I915_WRITE_HEAD(ring, 0); |
8187a2b7 | 386 | |
6fd0d56e CW |
387 | if (I915_READ_HEAD(ring) & HEAD_ADDR) { |
388 | DRM_ERROR("failed to set %s head to zero " | |
389 | "ctl %08x head %08x tail %08x start %08x\n", | |
390 | ring->name, | |
391 | I915_READ_CTL(ring), | |
392 | I915_READ_HEAD(ring), | |
393 | I915_READ_TAIL(ring), | |
394 | I915_READ_START(ring)); | |
395 | } | |
8187a2b7 ZN |
396 | } |
397 | ||
0d8957c8 DV |
398 | /* Initialize the ring. This must happen _after_ we've cleared the ring |
399 | * registers with the above sequence (the readback of the HEAD registers | |
400 | * also enforces ordering), otherwise the hw might lose the new ring | |
401 | * register values. */ | |
402 | I915_WRITE_START(ring, obj->gtt_offset); | |
7f2ab699 | 403 | I915_WRITE_CTL(ring, |
ae69b42a | 404 | ((ring->size - PAGE_SIZE) & RING_NR_PAGES) |
5d031e5b | 405 | | RING_VALID); |
8187a2b7 | 406 | |
8187a2b7 | 407 | /* If the head is still not zero, the ring is dead */ |
f01db988 SP |
408 | if (wait_for((I915_READ_CTL(ring) & RING_VALID) != 0 && |
409 | I915_READ_START(ring) == obj->gtt_offset && | |
410 | (I915_READ_HEAD(ring) & HEAD_ADDR) == 0, 50)) { | |
e74cfed5 CW |
411 | DRM_ERROR("%s initialization failed " |
412 | "ctl %08x head %08x tail %08x start %08x\n", | |
413 | ring->name, | |
414 | I915_READ_CTL(ring), | |
415 | I915_READ_HEAD(ring), | |
416 | I915_READ_TAIL(ring), | |
417 | I915_READ_START(ring)); | |
b7884eb4 DV |
418 | ret = -EIO; |
419 | goto out; | |
8187a2b7 ZN |
420 | } |
421 | ||
78501eac CW |
422 | if (!drm_core_check_feature(ring->dev, DRIVER_MODESET)) |
423 | i915_kernel_lost_context(ring->dev); | |
8187a2b7 | 424 | else { |
c7dca47b | 425 | ring->head = I915_READ_HEAD(ring); |
870e86dd | 426 | ring->tail = I915_READ_TAIL(ring) & TAIL_ADDR; |
c7dca47b | 427 | ring->space = ring_space(ring); |
c3b20037 | 428 | ring->last_retired_head = -1; |
8187a2b7 | 429 | } |
1ec14ad3 | 430 | |
b7884eb4 DV |
431 | out: |
432 | if (HAS_FORCE_WAKE(dev)) | |
433 | gen6_gt_force_wake_put(dev_priv); | |
434 | ||
435 | return ret; | |
8187a2b7 ZN |
436 | } |
437 | ||
c6df541c CW |
438 | static int |
439 | init_pipe_control(struct intel_ring_buffer *ring) | |
440 | { | |
441 | struct pipe_control *pc; | |
442 | struct drm_i915_gem_object *obj; | |
443 | int ret; | |
444 | ||
445 | if (ring->private) | |
446 | return 0; | |
447 | ||
448 | pc = kmalloc(sizeof(*pc), GFP_KERNEL); | |
449 | if (!pc) | |
450 | return -ENOMEM; | |
451 | ||
452 | obj = i915_gem_alloc_object(ring->dev, 4096); | |
453 | if (obj == NULL) { | |
454 | DRM_ERROR("Failed to allocate seqno page\n"); | |
455 | ret = -ENOMEM; | |
456 | goto err; | |
457 | } | |
e4ffd173 CW |
458 | |
459 | i915_gem_object_set_cache_level(obj, I915_CACHE_LLC); | |
c6df541c | 460 | |
86a1ee26 | 461 | ret = i915_gem_object_pin(obj, 4096, true, false); |
c6df541c CW |
462 | if (ret) |
463 | goto err_unref; | |
464 | ||
465 | pc->gtt_offset = obj->gtt_offset; | |
9da3da66 | 466 | pc->cpu_page = kmap(sg_page(obj->pages->sgl)); |
c6df541c CW |
467 | if (pc->cpu_page == NULL) |
468 | goto err_unpin; | |
469 | ||
470 | pc->obj = obj; | |
471 | ring->private = pc; | |
472 | return 0; | |
473 | ||
474 | err_unpin: | |
475 | i915_gem_object_unpin(obj); | |
476 | err_unref: | |
477 | drm_gem_object_unreference(&obj->base); | |
478 | err: | |
479 | kfree(pc); | |
480 | return ret; | |
481 | } | |
482 | ||
483 | static void | |
484 | cleanup_pipe_control(struct intel_ring_buffer *ring) | |
485 | { | |
486 | struct pipe_control *pc = ring->private; | |
487 | struct drm_i915_gem_object *obj; | |
488 | ||
489 | if (!ring->private) | |
490 | return; | |
491 | ||
492 | obj = pc->obj; | |
9da3da66 CW |
493 | |
494 | kunmap(sg_page(obj->pages->sgl)); | |
c6df541c CW |
495 | i915_gem_object_unpin(obj); |
496 | drm_gem_object_unreference(&obj->base); | |
497 | ||
498 | kfree(pc); | |
499 | ring->private = NULL; | |
500 | } | |
501 | ||
78501eac | 502 | static int init_render_ring(struct intel_ring_buffer *ring) |
8187a2b7 | 503 | { |
78501eac | 504 | struct drm_device *dev = ring->dev; |
1ec14ad3 | 505 | struct drm_i915_private *dev_priv = dev->dev_private; |
78501eac | 506 | int ret = init_ring_common(ring); |
a69ffdbf | 507 | |
a6c45cf0 | 508 | if (INTEL_INFO(dev)->gen > 3) { |
6b26c86d | 509 | I915_WRITE(MI_MODE, _MASKED_BIT_ENABLE(VS_TIMER_DISPATCH)); |
b095cd0a JB |
510 | if (IS_GEN7(dev)) |
511 | I915_WRITE(GFX_MODE_GEN7, | |
6b26c86d DV |
512 | _MASKED_BIT_DISABLE(GFX_TLB_INVALIDATE_ALWAYS) | |
513 | _MASKED_BIT_ENABLE(GFX_REPLAY_MODE)); | |
8187a2b7 | 514 | } |
78501eac | 515 | |
8d315287 | 516 | if (INTEL_INFO(dev)->gen >= 5) { |
c6df541c CW |
517 | ret = init_pipe_control(ring); |
518 | if (ret) | |
519 | return ret; | |
520 | } | |
521 | ||
5e13a0c5 | 522 | if (IS_GEN6(dev)) { |
3a69ddd6 KG |
523 | /* From the Sandybridge PRM, volume 1 part 3, page 24: |
524 | * "If this bit is set, STCunit will have LRA as replacement | |
525 | * policy. [...] This bit must be reset. LRA replacement | |
526 | * policy is not supported." | |
527 | */ | |
528 | I915_WRITE(CACHE_MODE_0, | |
5e13a0c5 | 529 | _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB)); |
12b0286f BW |
530 | |
531 | /* This is not explicitly set for GEN6, so read the register. | |
532 | * see intel_ring_mi_set_context() for why we care. | |
533 | * TODO: consider explicitly setting the bit for GEN5 | |
534 | */ | |
535 | ring->itlb_before_ctx_switch = | |
536 | !!(I915_READ(GFX_MODE) & GFX_TLB_INVALIDATE_ALWAYS); | |
84f9f938 BW |
537 | } |
538 | ||
6b26c86d DV |
539 | if (INTEL_INFO(dev)->gen >= 6) |
540 | I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_FORCE_ORDERING)); | |
84f9f938 | 541 | |
e1ef7cc2 | 542 | if (HAS_L3_GPU_CACHE(dev)) |
15b9f80e BW |
543 | I915_WRITE_IMR(ring, ~GEN6_RENDER_L3_PARITY_ERROR); |
544 | ||
8187a2b7 ZN |
545 | return ret; |
546 | } | |
547 | ||
c6df541c CW |
548 | static void render_ring_cleanup(struct intel_ring_buffer *ring) |
549 | { | |
550 | if (!ring->private) | |
551 | return; | |
552 | ||
553 | cleanup_pipe_control(ring); | |
554 | } | |
555 | ||
1ec14ad3 | 556 | static void |
c8c99b0f | 557 | update_mboxes(struct intel_ring_buffer *ring, |
9d773091 | 558 | u32 mmio_offset) |
1ec14ad3 | 559 | { |
1c8b46fc | 560 | intel_ring_emit(ring, MI_LOAD_REGISTER_IMM(1)); |
c8c99b0f | 561 | intel_ring_emit(ring, mmio_offset); |
9d773091 | 562 | intel_ring_emit(ring, ring->outstanding_lazy_request); |
1ec14ad3 CW |
563 | } |
564 | ||
c8c99b0f BW |
565 | /** |
566 | * gen6_add_request - Update the semaphore mailbox registers | |
567 | * | |
568 | * @ring - ring that is adding a request | |
569 | * @seqno - return seqno stuck into the ring | |
570 | * | |
571 | * Update the mailbox registers in the *other* rings with the current seqno. | |
572 | * This acts like a signal in the canonical semaphore. | |
573 | */ | |
1ec14ad3 | 574 | static int |
9d773091 | 575 | gen6_add_request(struct intel_ring_buffer *ring) |
1ec14ad3 | 576 | { |
c8c99b0f BW |
577 | u32 mbox1_reg; |
578 | u32 mbox2_reg; | |
1ec14ad3 CW |
579 | int ret; |
580 | ||
581 | ret = intel_ring_begin(ring, 10); | |
582 | if (ret) | |
583 | return ret; | |
584 | ||
c8c99b0f BW |
585 | mbox1_reg = ring->signal_mbox[0]; |
586 | mbox2_reg = ring->signal_mbox[1]; | |
1ec14ad3 | 587 | |
9d773091 CW |
588 | update_mboxes(ring, mbox1_reg); |
589 | update_mboxes(ring, mbox2_reg); | |
1ec14ad3 CW |
590 | intel_ring_emit(ring, MI_STORE_DWORD_INDEX); |
591 | intel_ring_emit(ring, I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT); | |
9d773091 | 592 | intel_ring_emit(ring, ring->outstanding_lazy_request); |
1ec14ad3 CW |
593 | intel_ring_emit(ring, MI_USER_INTERRUPT); |
594 | intel_ring_advance(ring); | |
595 | ||
1ec14ad3 CW |
596 | return 0; |
597 | } | |
598 | ||
f72b3435 MK |
599 | static inline bool i915_gem_has_seqno_wrapped(struct drm_device *dev, |
600 | u32 seqno) | |
601 | { | |
602 | struct drm_i915_private *dev_priv = dev->dev_private; | |
603 | return dev_priv->last_seqno < seqno; | |
604 | } | |
605 | ||
c8c99b0f BW |
606 | /** |
607 | * intel_ring_sync - sync the waiter to the signaller on seqno | |
608 | * | |
609 | * @waiter - ring that is waiting | |
610 | * @signaller - ring which has, or will signal | |
611 | * @seqno - seqno which the waiter will block on | |
612 | */ | |
613 | static int | |
686cb5f9 DV |
614 | gen6_ring_sync(struct intel_ring_buffer *waiter, |
615 | struct intel_ring_buffer *signaller, | |
616 | u32 seqno) | |
1ec14ad3 CW |
617 | { |
618 | int ret; | |
c8c99b0f BW |
619 | u32 dw1 = MI_SEMAPHORE_MBOX | |
620 | MI_SEMAPHORE_COMPARE | | |
621 | MI_SEMAPHORE_REGISTER; | |
1ec14ad3 | 622 | |
1500f7ea BW |
623 | /* Throughout all of the GEM code, seqno passed implies our current |
624 | * seqno is >= the last seqno executed. However for hardware the | |
625 | * comparison is strictly greater than. | |
626 | */ | |
627 | seqno -= 1; | |
628 | ||
686cb5f9 DV |
629 | WARN_ON(signaller->semaphore_register[waiter->id] == |
630 | MI_SEMAPHORE_SYNC_INVALID); | |
631 | ||
c8c99b0f | 632 | ret = intel_ring_begin(waiter, 4); |
1ec14ad3 CW |
633 | if (ret) |
634 | return ret; | |
635 | ||
f72b3435 MK |
636 | /* If seqno wrap happened, omit the wait with no-ops */ |
637 | if (likely(!i915_gem_has_seqno_wrapped(waiter->dev, seqno))) { | |
638 | intel_ring_emit(waiter, | |
639 | dw1 | | |
640 | signaller->semaphore_register[waiter->id]); | |
641 | intel_ring_emit(waiter, seqno); | |
642 | intel_ring_emit(waiter, 0); | |
643 | intel_ring_emit(waiter, MI_NOOP); | |
644 | } else { | |
645 | intel_ring_emit(waiter, MI_NOOP); | |
646 | intel_ring_emit(waiter, MI_NOOP); | |
647 | intel_ring_emit(waiter, MI_NOOP); | |
648 | intel_ring_emit(waiter, MI_NOOP); | |
649 | } | |
c8c99b0f | 650 | intel_ring_advance(waiter); |
1ec14ad3 CW |
651 | |
652 | return 0; | |
653 | } | |
654 | ||
c6df541c CW |
655 | #define PIPE_CONTROL_FLUSH(ring__, addr__) \ |
656 | do { \ | |
fcbc34e4 KG |
657 | intel_ring_emit(ring__, GFX_OP_PIPE_CONTROL(4) | PIPE_CONTROL_QW_WRITE | \ |
658 | PIPE_CONTROL_DEPTH_STALL); \ | |
c6df541c CW |
659 | intel_ring_emit(ring__, (addr__) | PIPE_CONTROL_GLOBAL_GTT); \ |
660 | intel_ring_emit(ring__, 0); \ | |
661 | intel_ring_emit(ring__, 0); \ | |
662 | } while (0) | |
663 | ||
664 | static int | |
9d773091 | 665 | pc_render_add_request(struct intel_ring_buffer *ring) |
c6df541c | 666 | { |
c6df541c CW |
667 | struct pipe_control *pc = ring->private; |
668 | u32 scratch_addr = pc->gtt_offset + 128; | |
669 | int ret; | |
670 | ||
671 | /* For Ironlake, MI_USER_INTERRUPT was deprecated and apparently | |
672 | * incoherent with writes to memory, i.e. completely fubar, | |
673 | * so we need to use PIPE_NOTIFY instead. | |
674 | * | |
675 | * However, we also need to workaround the qword write | |
676 | * incoherence by flushing the 6 PIPE_NOTIFY buffers out to | |
677 | * memory before requesting an interrupt. | |
678 | */ | |
679 | ret = intel_ring_begin(ring, 32); | |
680 | if (ret) | |
681 | return ret; | |
682 | ||
fcbc34e4 | 683 | intel_ring_emit(ring, GFX_OP_PIPE_CONTROL(4) | PIPE_CONTROL_QW_WRITE | |
9d971b37 KG |
684 | PIPE_CONTROL_WRITE_FLUSH | |
685 | PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE); | |
c6df541c | 686 | intel_ring_emit(ring, pc->gtt_offset | PIPE_CONTROL_GLOBAL_GTT); |
9d773091 | 687 | intel_ring_emit(ring, ring->outstanding_lazy_request); |
c6df541c CW |
688 | intel_ring_emit(ring, 0); |
689 | PIPE_CONTROL_FLUSH(ring, scratch_addr); | |
690 | scratch_addr += 128; /* write to separate cachelines */ | |
691 | PIPE_CONTROL_FLUSH(ring, scratch_addr); | |
692 | scratch_addr += 128; | |
693 | PIPE_CONTROL_FLUSH(ring, scratch_addr); | |
694 | scratch_addr += 128; | |
695 | PIPE_CONTROL_FLUSH(ring, scratch_addr); | |
696 | scratch_addr += 128; | |
697 | PIPE_CONTROL_FLUSH(ring, scratch_addr); | |
698 | scratch_addr += 128; | |
699 | PIPE_CONTROL_FLUSH(ring, scratch_addr); | |
a71d8d94 | 700 | |
fcbc34e4 | 701 | intel_ring_emit(ring, GFX_OP_PIPE_CONTROL(4) | PIPE_CONTROL_QW_WRITE | |
9d971b37 KG |
702 | PIPE_CONTROL_WRITE_FLUSH | |
703 | PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE | | |
c6df541c CW |
704 | PIPE_CONTROL_NOTIFY); |
705 | intel_ring_emit(ring, pc->gtt_offset | PIPE_CONTROL_GLOBAL_GTT); | |
9d773091 | 706 | intel_ring_emit(ring, ring->outstanding_lazy_request); |
c6df541c CW |
707 | intel_ring_emit(ring, 0); |
708 | intel_ring_advance(ring); | |
709 | ||
c6df541c CW |
710 | return 0; |
711 | } | |
712 | ||
4cd53c0c | 713 | static u32 |
b2eadbc8 | 714 | gen6_ring_get_seqno(struct intel_ring_buffer *ring, bool lazy_coherency) |
4cd53c0c | 715 | { |
4cd53c0c DV |
716 | /* Workaround to force correct ordering between irq and seqno writes on |
717 | * ivb (and maybe also on snb) by reading from a CS register (like | |
718 | * ACTHD) before reading the status page. */ | |
b2eadbc8 | 719 | if (!lazy_coherency) |
4cd53c0c DV |
720 | intel_ring_get_active_head(ring); |
721 | return intel_read_status_page(ring, I915_GEM_HWS_INDEX); | |
722 | } | |
723 | ||
8187a2b7 | 724 | static u32 |
b2eadbc8 | 725 | ring_get_seqno(struct intel_ring_buffer *ring, bool lazy_coherency) |
8187a2b7 | 726 | { |
1ec14ad3 CW |
727 | return intel_read_status_page(ring, I915_GEM_HWS_INDEX); |
728 | } | |
729 | ||
b70ec5bf MK |
730 | static void |
731 | ring_set_seqno(struct intel_ring_buffer *ring, u32 seqno) | |
732 | { | |
733 | intel_write_status_page(ring, I915_GEM_HWS_INDEX, seqno); | |
734 | } | |
735 | ||
c6df541c | 736 | static u32 |
b2eadbc8 | 737 | pc_render_get_seqno(struct intel_ring_buffer *ring, bool lazy_coherency) |
c6df541c CW |
738 | { |
739 | struct pipe_control *pc = ring->private; | |
740 | return pc->cpu_page[0]; | |
741 | } | |
742 | ||
b70ec5bf MK |
743 | static void |
744 | pc_render_set_seqno(struct intel_ring_buffer *ring, u32 seqno) | |
745 | { | |
746 | struct pipe_control *pc = ring->private; | |
747 | pc->cpu_page[0] = seqno; | |
748 | } | |
749 | ||
e48d8634 DV |
750 | static bool |
751 | gen5_ring_get_irq(struct intel_ring_buffer *ring) | |
752 | { | |
753 | struct drm_device *dev = ring->dev; | |
754 | drm_i915_private_t *dev_priv = dev->dev_private; | |
7338aefa | 755 | unsigned long flags; |
e48d8634 DV |
756 | |
757 | if (!dev->irq_enabled) | |
758 | return false; | |
759 | ||
7338aefa | 760 | spin_lock_irqsave(&dev_priv->irq_lock, flags); |
f637fde4 DV |
761 | if (ring->irq_refcount++ == 0) { |
762 | dev_priv->gt_irq_mask &= ~ring->irq_enable_mask; | |
763 | I915_WRITE(GTIMR, dev_priv->gt_irq_mask); | |
764 | POSTING_READ(GTIMR); | |
765 | } | |
7338aefa | 766 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); |
e48d8634 DV |
767 | |
768 | return true; | |
769 | } | |
770 | ||
771 | static void | |
772 | gen5_ring_put_irq(struct intel_ring_buffer *ring) | |
773 | { | |
774 | struct drm_device *dev = ring->dev; | |
775 | drm_i915_private_t *dev_priv = dev->dev_private; | |
7338aefa | 776 | unsigned long flags; |
e48d8634 | 777 | |
7338aefa | 778 | spin_lock_irqsave(&dev_priv->irq_lock, flags); |
f637fde4 DV |
779 | if (--ring->irq_refcount == 0) { |
780 | dev_priv->gt_irq_mask |= ring->irq_enable_mask; | |
781 | I915_WRITE(GTIMR, dev_priv->gt_irq_mask); | |
782 | POSTING_READ(GTIMR); | |
783 | } | |
7338aefa | 784 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); |
e48d8634 DV |
785 | } |
786 | ||
b13c2b96 | 787 | static bool |
e3670319 | 788 | i9xx_ring_get_irq(struct intel_ring_buffer *ring) |
62fdfeaf | 789 | { |
78501eac | 790 | struct drm_device *dev = ring->dev; |
01a03331 | 791 | drm_i915_private_t *dev_priv = dev->dev_private; |
7338aefa | 792 | unsigned long flags; |
62fdfeaf | 793 | |
b13c2b96 CW |
794 | if (!dev->irq_enabled) |
795 | return false; | |
796 | ||
7338aefa | 797 | spin_lock_irqsave(&dev_priv->irq_lock, flags); |
f637fde4 DV |
798 | if (ring->irq_refcount++ == 0) { |
799 | dev_priv->irq_mask &= ~ring->irq_enable_mask; | |
800 | I915_WRITE(IMR, dev_priv->irq_mask); | |
801 | POSTING_READ(IMR); | |
802 | } | |
7338aefa | 803 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); |
b13c2b96 CW |
804 | |
805 | return true; | |
62fdfeaf EA |
806 | } |
807 | ||
8187a2b7 | 808 | static void |
e3670319 | 809 | i9xx_ring_put_irq(struct intel_ring_buffer *ring) |
62fdfeaf | 810 | { |
78501eac | 811 | struct drm_device *dev = ring->dev; |
01a03331 | 812 | drm_i915_private_t *dev_priv = dev->dev_private; |
7338aefa | 813 | unsigned long flags; |
62fdfeaf | 814 | |
7338aefa | 815 | spin_lock_irqsave(&dev_priv->irq_lock, flags); |
f637fde4 DV |
816 | if (--ring->irq_refcount == 0) { |
817 | dev_priv->irq_mask |= ring->irq_enable_mask; | |
818 | I915_WRITE(IMR, dev_priv->irq_mask); | |
819 | POSTING_READ(IMR); | |
820 | } | |
7338aefa | 821 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); |
62fdfeaf EA |
822 | } |
823 | ||
c2798b19 CW |
824 | static bool |
825 | i8xx_ring_get_irq(struct intel_ring_buffer *ring) | |
826 | { | |
827 | struct drm_device *dev = ring->dev; | |
828 | drm_i915_private_t *dev_priv = dev->dev_private; | |
7338aefa | 829 | unsigned long flags; |
c2798b19 CW |
830 | |
831 | if (!dev->irq_enabled) | |
832 | return false; | |
833 | ||
7338aefa | 834 | spin_lock_irqsave(&dev_priv->irq_lock, flags); |
c2798b19 CW |
835 | if (ring->irq_refcount++ == 0) { |
836 | dev_priv->irq_mask &= ~ring->irq_enable_mask; | |
837 | I915_WRITE16(IMR, dev_priv->irq_mask); | |
838 | POSTING_READ16(IMR); | |
839 | } | |
7338aefa | 840 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); |
c2798b19 CW |
841 | |
842 | return true; | |
843 | } | |
844 | ||
845 | static void | |
846 | i8xx_ring_put_irq(struct intel_ring_buffer *ring) | |
847 | { | |
848 | struct drm_device *dev = ring->dev; | |
849 | drm_i915_private_t *dev_priv = dev->dev_private; | |
7338aefa | 850 | unsigned long flags; |
c2798b19 | 851 | |
7338aefa | 852 | spin_lock_irqsave(&dev_priv->irq_lock, flags); |
c2798b19 CW |
853 | if (--ring->irq_refcount == 0) { |
854 | dev_priv->irq_mask |= ring->irq_enable_mask; | |
855 | I915_WRITE16(IMR, dev_priv->irq_mask); | |
856 | POSTING_READ16(IMR); | |
857 | } | |
7338aefa | 858 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); |
c2798b19 CW |
859 | } |
860 | ||
78501eac | 861 | void intel_ring_setup_status_page(struct intel_ring_buffer *ring) |
8187a2b7 | 862 | { |
4593010b | 863 | struct drm_device *dev = ring->dev; |
78501eac | 864 | drm_i915_private_t *dev_priv = ring->dev->dev_private; |
4593010b EA |
865 | u32 mmio = 0; |
866 | ||
867 | /* The ring status page addresses are no longer next to the rest of | |
868 | * the ring registers as of gen7. | |
869 | */ | |
870 | if (IS_GEN7(dev)) { | |
871 | switch (ring->id) { | |
96154f2f | 872 | case RCS: |
4593010b EA |
873 | mmio = RENDER_HWS_PGA_GEN7; |
874 | break; | |
96154f2f | 875 | case BCS: |
4593010b EA |
876 | mmio = BLT_HWS_PGA_GEN7; |
877 | break; | |
96154f2f | 878 | case VCS: |
4593010b EA |
879 | mmio = BSD_HWS_PGA_GEN7; |
880 | break; | |
881 | } | |
882 | } else if (IS_GEN6(ring->dev)) { | |
883 | mmio = RING_HWS_PGA_GEN6(ring->mmio_base); | |
884 | } else { | |
885 | mmio = RING_HWS_PGA(ring->mmio_base); | |
886 | } | |
887 | ||
78501eac CW |
888 | I915_WRITE(mmio, (u32)ring->status_page.gfx_addr); |
889 | POSTING_READ(mmio); | |
8187a2b7 ZN |
890 | } |
891 | ||
b72f3acb | 892 | static int |
78501eac CW |
893 | bsd_ring_flush(struct intel_ring_buffer *ring, |
894 | u32 invalidate_domains, | |
895 | u32 flush_domains) | |
d1b851fc | 896 | { |
b72f3acb CW |
897 | int ret; |
898 | ||
b72f3acb CW |
899 | ret = intel_ring_begin(ring, 2); |
900 | if (ret) | |
901 | return ret; | |
902 | ||
903 | intel_ring_emit(ring, MI_FLUSH); | |
904 | intel_ring_emit(ring, MI_NOOP); | |
905 | intel_ring_advance(ring); | |
906 | return 0; | |
d1b851fc ZN |
907 | } |
908 | ||
3cce469c | 909 | static int |
9d773091 | 910 | i9xx_add_request(struct intel_ring_buffer *ring) |
d1b851fc | 911 | { |
3cce469c CW |
912 | int ret; |
913 | ||
914 | ret = intel_ring_begin(ring, 4); | |
915 | if (ret) | |
916 | return ret; | |
6f392d54 | 917 | |
3cce469c CW |
918 | intel_ring_emit(ring, MI_STORE_DWORD_INDEX); |
919 | intel_ring_emit(ring, I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT); | |
9d773091 | 920 | intel_ring_emit(ring, ring->outstanding_lazy_request); |
3cce469c CW |
921 | intel_ring_emit(ring, MI_USER_INTERRUPT); |
922 | intel_ring_advance(ring); | |
d1b851fc | 923 | |
3cce469c | 924 | return 0; |
d1b851fc ZN |
925 | } |
926 | ||
0f46832f | 927 | static bool |
25c06300 | 928 | gen6_ring_get_irq(struct intel_ring_buffer *ring) |
0f46832f CW |
929 | { |
930 | struct drm_device *dev = ring->dev; | |
01a03331 | 931 | drm_i915_private_t *dev_priv = dev->dev_private; |
7338aefa | 932 | unsigned long flags; |
0f46832f CW |
933 | |
934 | if (!dev->irq_enabled) | |
935 | return false; | |
936 | ||
4cd53c0c DV |
937 | /* It looks like we need to prevent the gt from suspending while waiting |
938 | * for an notifiy irq, otherwise irqs seem to get lost on at least the | |
939 | * blt/bsd rings on ivb. */ | |
99ffa162 | 940 | gen6_gt_force_wake_get(dev_priv); |
4cd53c0c | 941 | |
7338aefa | 942 | spin_lock_irqsave(&dev_priv->irq_lock, flags); |
01a03331 | 943 | if (ring->irq_refcount++ == 0) { |
e1ef7cc2 | 944 | if (HAS_L3_GPU_CACHE(dev) && ring->id == RCS) |
15b9f80e BW |
945 | I915_WRITE_IMR(ring, ~(ring->irq_enable_mask | |
946 | GEN6_RENDER_L3_PARITY_ERROR)); | |
947 | else | |
948 | I915_WRITE_IMR(ring, ~ring->irq_enable_mask); | |
f637fde4 DV |
949 | dev_priv->gt_irq_mask &= ~ring->irq_enable_mask; |
950 | I915_WRITE(GTIMR, dev_priv->gt_irq_mask); | |
951 | POSTING_READ(GTIMR); | |
0f46832f | 952 | } |
7338aefa | 953 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); |
0f46832f CW |
954 | |
955 | return true; | |
956 | } | |
957 | ||
958 | static void | |
25c06300 | 959 | gen6_ring_put_irq(struct intel_ring_buffer *ring) |
0f46832f CW |
960 | { |
961 | struct drm_device *dev = ring->dev; | |
01a03331 | 962 | drm_i915_private_t *dev_priv = dev->dev_private; |
7338aefa | 963 | unsigned long flags; |
0f46832f | 964 | |
7338aefa | 965 | spin_lock_irqsave(&dev_priv->irq_lock, flags); |
01a03331 | 966 | if (--ring->irq_refcount == 0) { |
e1ef7cc2 | 967 | if (HAS_L3_GPU_CACHE(dev) && ring->id == RCS) |
15b9f80e BW |
968 | I915_WRITE_IMR(ring, ~GEN6_RENDER_L3_PARITY_ERROR); |
969 | else | |
970 | I915_WRITE_IMR(ring, ~0); | |
f637fde4 DV |
971 | dev_priv->gt_irq_mask |= ring->irq_enable_mask; |
972 | I915_WRITE(GTIMR, dev_priv->gt_irq_mask); | |
973 | POSTING_READ(GTIMR); | |
1ec14ad3 | 974 | } |
7338aefa | 975 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); |
4cd53c0c | 976 | |
99ffa162 | 977 | gen6_gt_force_wake_put(dev_priv); |
d1b851fc ZN |
978 | } |
979 | ||
d1b851fc | 980 | static int |
d7d4eedd CW |
981 | i965_dispatch_execbuffer(struct intel_ring_buffer *ring, |
982 | u32 offset, u32 length, | |
983 | unsigned flags) | |
d1b851fc | 984 | { |
e1f99ce6 | 985 | int ret; |
78501eac | 986 | |
e1f99ce6 CW |
987 | ret = intel_ring_begin(ring, 2); |
988 | if (ret) | |
989 | return ret; | |
990 | ||
78501eac | 991 | intel_ring_emit(ring, |
65f56876 CW |
992 | MI_BATCH_BUFFER_START | |
993 | MI_BATCH_GTT | | |
d7d4eedd | 994 | (flags & I915_DISPATCH_SECURE ? 0 : MI_BATCH_NON_SECURE_I965)); |
c4e7a414 | 995 | intel_ring_emit(ring, offset); |
78501eac CW |
996 | intel_ring_advance(ring); |
997 | ||
d1b851fc ZN |
998 | return 0; |
999 | } | |
1000 | ||
8187a2b7 | 1001 | static int |
fb3256da | 1002 | i830_dispatch_execbuffer(struct intel_ring_buffer *ring, |
d7d4eedd CW |
1003 | u32 offset, u32 len, |
1004 | unsigned flags) | |
62fdfeaf | 1005 | { |
c4e7a414 | 1006 | int ret; |
62fdfeaf | 1007 | |
fb3256da DV |
1008 | ret = intel_ring_begin(ring, 4); |
1009 | if (ret) | |
1010 | return ret; | |
62fdfeaf | 1011 | |
fb3256da | 1012 | intel_ring_emit(ring, MI_BATCH_BUFFER); |
d7d4eedd | 1013 | intel_ring_emit(ring, offset | (flags & I915_DISPATCH_SECURE ? 0 : MI_BATCH_NON_SECURE)); |
fb3256da DV |
1014 | intel_ring_emit(ring, offset + len - 8); |
1015 | intel_ring_emit(ring, 0); | |
1016 | intel_ring_advance(ring); | |
e1f99ce6 | 1017 | |
fb3256da DV |
1018 | return 0; |
1019 | } | |
1020 | ||
1021 | static int | |
1022 | i915_dispatch_execbuffer(struct intel_ring_buffer *ring, | |
d7d4eedd CW |
1023 | u32 offset, u32 len, |
1024 | unsigned flags) | |
fb3256da DV |
1025 | { |
1026 | int ret; | |
1027 | ||
1028 | ret = intel_ring_begin(ring, 2); | |
1029 | if (ret) | |
1030 | return ret; | |
1031 | ||
65f56876 | 1032 | intel_ring_emit(ring, MI_BATCH_BUFFER_START | MI_BATCH_GTT); |
d7d4eedd | 1033 | intel_ring_emit(ring, offset | (flags & I915_DISPATCH_SECURE ? 0 : MI_BATCH_NON_SECURE)); |
c4e7a414 | 1034 | intel_ring_advance(ring); |
62fdfeaf | 1035 | |
62fdfeaf EA |
1036 | return 0; |
1037 | } | |
1038 | ||
78501eac | 1039 | static void cleanup_status_page(struct intel_ring_buffer *ring) |
62fdfeaf | 1040 | { |
05394f39 | 1041 | struct drm_i915_gem_object *obj; |
62fdfeaf | 1042 | |
8187a2b7 ZN |
1043 | obj = ring->status_page.obj; |
1044 | if (obj == NULL) | |
62fdfeaf | 1045 | return; |
62fdfeaf | 1046 | |
9da3da66 | 1047 | kunmap(sg_page(obj->pages->sgl)); |
62fdfeaf | 1048 | i915_gem_object_unpin(obj); |
05394f39 | 1049 | drm_gem_object_unreference(&obj->base); |
8187a2b7 | 1050 | ring->status_page.obj = NULL; |
62fdfeaf EA |
1051 | } |
1052 | ||
78501eac | 1053 | static int init_status_page(struct intel_ring_buffer *ring) |
62fdfeaf | 1054 | { |
78501eac | 1055 | struct drm_device *dev = ring->dev; |
05394f39 | 1056 | struct drm_i915_gem_object *obj; |
62fdfeaf EA |
1057 | int ret; |
1058 | ||
62fdfeaf EA |
1059 | obj = i915_gem_alloc_object(dev, 4096); |
1060 | if (obj == NULL) { | |
1061 | DRM_ERROR("Failed to allocate status page\n"); | |
1062 | ret = -ENOMEM; | |
1063 | goto err; | |
1064 | } | |
e4ffd173 CW |
1065 | |
1066 | i915_gem_object_set_cache_level(obj, I915_CACHE_LLC); | |
62fdfeaf | 1067 | |
86a1ee26 | 1068 | ret = i915_gem_object_pin(obj, 4096, true, false); |
62fdfeaf | 1069 | if (ret != 0) { |
62fdfeaf EA |
1070 | goto err_unref; |
1071 | } | |
1072 | ||
05394f39 | 1073 | ring->status_page.gfx_addr = obj->gtt_offset; |
9da3da66 | 1074 | ring->status_page.page_addr = kmap(sg_page(obj->pages->sgl)); |
8187a2b7 | 1075 | if (ring->status_page.page_addr == NULL) { |
2e6c21ed | 1076 | ret = -ENOMEM; |
62fdfeaf EA |
1077 | goto err_unpin; |
1078 | } | |
8187a2b7 ZN |
1079 | ring->status_page.obj = obj; |
1080 | memset(ring->status_page.page_addr, 0, PAGE_SIZE); | |
62fdfeaf | 1081 | |
78501eac | 1082 | intel_ring_setup_status_page(ring); |
8187a2b7 ZN |
1083 | DRM_DEBUG_DRIVER("%s hws offset: 0x%08x\n", |
1084 | ring->name, ring->status_page.gfx_addr); | |
62fdfeaf EA |
1085 | |
1086 | return 0; | |
1087 | ||
1088 | err_unpin: | |
1089 | i915_gem_object_unpin(obj); | |
1090 | err_unref: | |
05394f39 | 1091 | drm_gem_object_unreference(&obj->base); |
62fdfeaf | 1092 | err: |
8187a2b7 | 1093 | return ret; |
62fdfeaf EA |
1094 | } |
1095 | ||
6b8294a4 CW |
1096 | static int init_phys_hws_pga(struct intel_ring_buffer *ring) |
1097 | { | |
1098 | struct drm_i915_private *dev_priv = ring->dev->dev_private; | |
1099 | u32 addr; | |
1100 | ||
1101 | if (!dev_priv->status_page_dmah) { | |
1102 | dev_priv->status_page_dmah = | |
1103 | drm_pci_alloc(ring->dev, PAGE_SIZE, PAGE_SIZE); | |
1104 | if (!dev_priv->status_page_dmah) | |
1105 | return -ENOMEM; | |
1106 | } | |
1107 | ||
1108 | addr = dev_priv->status_page_dmah->busaddr; | |
1109 | if (INTEL_INFO(ring->dev)->gen >= 4) | |
1110 | addr |= (dev_priv->status_page_dmah->busaddr >> 28) & 0xf0; | |
1111 | I915_WRITE(HWS_PGA, addr); | |
1112 | ||
1113 | ring->status_page.page_addr = dev_priv->status_page_dmah->vaddr; | |
1114 | memset(ring->status_page.page_addr, 0, PAGE_SIZE); | |
1115 | ||
1116 | return 0; | |
1117 | } | |
1118 | ||
c43b5634 BW |
1119 | static int intel_init_ring_buffer(struct drm_device *dev, |
1120 | struct intel_ring_buffer *ring) | |
62fdfeaf | 1121 | { |
05394f39 | 1122 | struct drm_i915_gem_object *obj; |
dd2757f8 | 1123 | struct drm_i915_private *dev_priv = dev->dev_private; |
dd785e35 CW |
1124 | int ret; |
1125 | ||
8187a2b7 | 1126 | ring->dev = dev; |
23bc5982 CW |
1127 | INIT_LIST_HEAD(&ring->active_list); |
1128 | INIT_LIST_HEAD(&ring->request_list); | |
dfc9ef2f | 1129 | ring->size = 32 * PAGE_SIZE; |
9d773091 | 1130 | memset(ring->sync_seqno, 0, sizeof(ring->sync_seqno)); |
0dc79fb2 | 1131 | |
b259f673 | 1132 | init_waitqueue_head(&ring->irq_queue); |
62fdfeaf | 1133 | |
8187a2b7 | 1134 | if (I915_NEED_GFX_HWS(dev)) { |
78501eac | 1135 | ret = init_status_page(ring); |
8187a2b7 ZN |
1136 | if (ret) |
1137 | return ret; | |
6b8294a4 CW |
1138 | } else { |
1139 | BUG_ON(ring->id != RCS); | |
1140 | ret = init_phys_hws_pga(ring); | |
1141 | if (ret) | |
1142 | return ret; | |
8187a2b7 | 1143 | } |
62fdfeaf | 1144 | |
ebc052e0 CW |
1145 | obj = NULL; |
1146 | if (!HAS_LLC(dev)) | |
1147 | obj = i915_gem_object_create_stolen(dev, ring->size); | |
1148 | if (obj == NULL) | |
1149 | obj = i915_gem_alloc_object(dev, ring->size); | |
62fdfeaf EA |
1150 | if (obj == NULL) { |
1151 | DRM_ERROR("Failed to allocate ringbuffer\n"); | |
8187a2b7 | 1152 | ret = -ENOMEM; |
dd785e35 | 1153 | goto err_hws; |
62fdfeaf | 1154 | } |
62fdfeaf | 1155 | |
05394f39 | 1156 | ring->obj = obj; |
8187a2b7 | 1157 | |
86a1ee26 | 1158 | ret = i915_gem_object_pin(obj, PAGE_SIZE, true, false); |
dd785e35 CW |
1159 | if (ret) |
1160 | goto err_unref; | |
62fdfeaf | 1161 | |
3eef8918 CW |
1162 | ret = i915_gem_object_set_to_gtt_domain(obj, true); |
1163 | if (ret) | |
1164 | goto err_unpin; | |
1165 | ||
dd2757f8 DV |
1166 | ring->virtual_start = |
1167 | ioremap_wc(dev_priv->mm.gtt->gma_bus_addr + obj->gtt_offset, | |
1168 | ring->size); | |
4225d0f2 | 1169 | if (ring->virtual_start == NULL) { |
62fdfeaf | 1170 | DRM_ERROR("Failed to map ringbuffer.\n"); |
8187a2b7 | 1171 | ret = -EINVAL; |
dd785e35 | 1172 | goto err_unpin; |
62fdfeaf EA |
1173 | } |
1174 | ||
78501eac | 1175 | ret = ring->init(ring); |
dd785e35 CW |
1176 | if (ret) |
1177 | goto err_unmap; | |
62fdfeaf | 1178 | |
55249baa CW |
1179 | /* Workaround an erratum on the i830 which causes a hang if |
1180 | * the TAIL pointer points to within the last 2 cachelines | |
1181 | * of the buffer. | |
1182 | */ | |
1183 | ring->effective_size = ring->size; | |
27c1cbd0 | 1184 | if (IS_I830(ring->dev) || IS_845G(ring->dev)) |
55249baa CW |
1185 | ring->effective_size -= 128; |
1186 | ||
f7e98ad4 MK |
1187 | intel_ring_init_seqno(ring, dev_priv->last_seqno); |
1188 | ||
c584fe47 | 1189 | return 0; |
dd785e35 CW |
1190 | |
1191 | err_unmap: | |
4225d0f2 | 1192 | iounmap(ring->virtual_start); |
dd785e35 CW |
1193 | err_unpin: |
1194 | i915_gem_object_unpin(obj); | |
1195 | err_unref: | |
05394f39 CW |
1196 | drm_gem_object_unreference(&obj->base); |
1197 | ring->obj = NULL; | |
dd785e35 | 1198 | err_hws: |
78501eac | 1199 | cleanup_status_page(ring); |
8187a2b7 | 1200 | return ret; |
62fdfeaf EA |
1201 | } |
1202 | ||
78501eac | 1203 | void intel_cleanup_ring_buffer(struct intel_ring_buffer *ring) |
62fdfeaf | 1204 | { |
33626e6a CW |
1205 | struct drm_i915_private *dev_priv; |
1206 | int ret; | |
1207 | ||
05394f39 | 1208 | if (ring->obj == NULL) |
62fdfeaf EA |
1209 | return; |
1210 | ||
33626e6a CW |
1211 | /* Disable the ring buffer. The ring must be idle at this point */ |
1212 | dev_priv = ring->dev->dev_private; | |
3e960501 | 1213 | ret = intel_ring_idle(ring); |
29ee3991 CW |
1214 | if (ret) |
1215 | DRM_ERROR("failed to quiesce %s whilst cleaning up: %d\n", | |
1216 | ring->name, ret); | |
1217 | ||
33626e6a CW |
1218 | I915_WRITE_CTL(ring, 0); |
1219 | ||
4225d0f2 | 1220 | iounmap(ring->virtual_start); |
62fdfeaf | 1221 | |
05394f39 CW |
1222 | i915_gem_object_unpin(ring->obj); |
1223 | drm_gem_object_unreference(&ring->obj->base); | |
1224 | ring->obj = NULL; | |
78501eac | 1225 | |
8d19215b ZN |
1226 | if (ring->cleanup) |
1227 | ring->cleanup(ring); | |
1228 | ||
78501eac | 1229 | cleanup_status_page(ring); |
62fdfeaf EA |
1230 | } |
1231 | ||
a71d8d94 CW |
1232 | static int intel_ring_wait_seqno(struct intel_ring_buffer *ring, u32 seqno) |
1233 | { | |
a71d8d94 CW |
1234 | int ret; |
1235 | ||
199b2bc2 | 1236 | ret = i915_wait_seqno(ring, seqno); |
b2da9fe5 BW |
1237 | if (!ret) |
1238 | i915_gem_retire_requests_ring(ring); | |
a71d8d94 CW |
1239 | |
1240 | return ret; | |
1241 | } | |
1242 | ||
1243 | static int intel_ring_wait_request(struct intel_ring_buffer *ring, int n) | |
1244 | { | |
1245 | struct drm_i915_gem_request *request; | |
1246 | u32 seqno = 0; | |
1247 | int ret; | |
1248 | ||
1249 | i915_gem_retire_requests_ring(ring); | |
1250 | ||
1251 | if (ring->last_retired_head != -1) { | |
1252 | ring->head = ring->last_retired_head; | |
1253 | ring->last_retired_head = -1; | |
1254 | ring->space = ring_space(ring); | |
1255 | if (ring->space >= n) | |
1256 | return 0; | |
1257 | } | |
1258 | ||
1259 | list_for_each_entry(request, &ring->request_list, list) { | |
1260 | int space; | |
1261 | ||
1262 | if (request->tail == -1) | |
1263 | continue; | |
1264 | ||
1265 | space = request->tail - (ring->tail + 8); | |
1266 | if (space < 0) | |
1267 | space += ring->size; | |
1268 | if (space >= n) { | |
1269 | seqno = request->seqno; | |
1270 | break; | |
1271 | } | |
1272 | ||
1273 | /* Consume this request in case we need more space than | |
1274 | * is available and so need to prevent a race between | |
1275 | * updating last_retired_head and direct reads of | |
1276 | * I915_RING_HEAD. It also provides a nice sanity check. | |
1277 | */ | |
1278 | request->tail = -1; | |
1279 | } | |
1280 | ||
1281 | if (seqno == 0) | |
1282 | return -ENOSPC; | |
1283 | ||
1284 | ret = intel_ring_wait_seqno(ring, seqno); | |
1285 | if (ret) | |
1286 | return ret; | |
1287 | ||
1288 | if (WARN_ON(ring->last_retired_head == -1)) | |
1289 | return -ENOSPC; | |
1290 | ||
1291 | ring->head = ring->last_retired_head; | |
1292 | ring->last_retired_head = -1; | |
1293 | ring->space = ring_space(ring); | |
1294 | if (WARN_ON(ring->space < n)) | |
1295 | return -ENOSPC; | |
1296 | ||
1297 | return 0; | |
1298 | } | |
1299 | ||
3e960501 | 1300 | static int ring_wait_for_space(struct intel_ring_buffer *ring, int n) |
62fdfeaf | 1301 | { |
78501eac | 1302 | struct drm_device *dev = ring->dev; |
cae5852d | 1303 | struct drm_i915_private *dev_priv = dev->dev_private; |
78501eac | 1304 | unsigned long end; |
a71d8d94 | 1305 | int ret; |
c7dca47b | 1306 | |
a71d8d94 CW |
1307 | ret = intel_ring_wait_request(ring, n); |
1308 | if (ret != -ENOSPC) | |
1309 | return ret; | |
1310 | ||
db53a302 | 1311 | trace_i915_ring_wait_begin(ring); |
63ed2cb2 DV |
1312 | /* With GEM the hangcheck timer should kick us out of the loop, |
1313 | * leaving it early runs the risk of corrupting GEM state (due | |
1314 | * to running on almost untested codepaths). But on resume | |
1315 | * timers don't work yet, so prevent a complete hang in that | |
1316 | * case by choosing an insanely large timeout. */ | |
1317 | end = jiffies + 60 * HZ; | |
e6bfaf85 | 1318 | |
8187a2b7 | 1319 | do { |
c7dca47b CW |
1320 | ring->head = I915_READ_HEAD(ring); |
1321 | ring->space = ring_space(ring); | |
62fdfeaf | 1322 | if (ring->space >= n) { |
db53a302 | 1323 | trace_i915_ring_wait_end(ring); |
62fdfeaf EA |
1324 | return 0; |
1325 | } | |
1326 | ||
1327 | if (dev->primary->master) { | |
1328 | struct drm_i915_master_private *master_priv = dev->primary->master->driver_priv; | |
1329 | if (master_priv->sarea_priv) | |
1330 | master_priv->sarea_priv->perf_boxes |= I915_BOX_WAIT; | |
1331 | } | |
d1b851fc | 1332 | |
e60a0b10 | 1333 | msleep(1); |
d6b2c790 DV |
1334 | |
1335 | ret = i915_gem_check_wedge(dev_priv, dev_priv->mm.interruptible); | |
1336 | if (ret) | |
1337 | return ret; | |
8187a2b7 | 1338 | } while (!time_after(jiffies, end)); |
db53a302 | 1339 | trace_i915_ring_wait_end(ring); |
8187a2b7 ZN |
1340 | return -EBUSY; |
1341 | } | |
62fdfeaf | 1342 | |
3e960501 CW |
1343 | static int intel_wrap_ring_buffer(struct intel_ring_buffer *ring) |
1344 | { | |
1345 | uint32_t __iomem *virt; | |
1346 | int rem = ring->size - ring->tail; | |
1347 | ||
1348 | if (ring->space < rem) { | |
1349 | int ret = ring_wait_for_space(ring, rem); | |
1350 | if (ret) | |
1351 | return ret; | |
1352 | } | |
1353 | ||
1354 | virt = ring->virtual_start + ring->tail; | |
1355 | rem /= 4; | |
1356 | while (rem--) | |
1357 | iowrite32(MI_NOOP, virt++); | |
1358 | ||
1359 | ring->tail = 0; | |
1360 | ring->space = ring_space(ring); | |
1361 | ||
1362 | return 0; | |
1363 | } | |
1364 | ||
1365 | int intel_ring_idle(struct intel_ring_buffer *ring) | |
1366 | { | |
1367 | u32 seqno; | |
1368 | int ret; | |
1369 | ||
1370 | /* We need to add any requests required to flush the objects and ring */ | |
1371 | if (ring->outstanding_lazy_request) { | |
1372 | ret = i915_add_request(ring, NULL, NULL); | |
1373 | if (ret) | |
1374 | return ret; | |
1375 | } | |
1376 | ||
1377 | /* Wait upon the last request to be completed */ | |
1378 | if (list_empty(&ring->request_list)) | |
1379 | return 0; | |
1380 | ||
1381 | seqno = list_entry(ring->request_list.prev, | |
1382 | struct drm_i915_gem_request, | |
1383 | list)->seqno; | |
1384 | ||
1385 | return i915_wait_seqno(ring, seqno); | |
1386 | } | |
1387 | ||
9d773091 CW |
1388 | static int |
1389 | intel_ring_alloc_seqno(struct intel_ring_buffer *ring) | |
1390 | { | |
1391 | if (ring->outstanding_lazy_request) | |
1392 | return 0; | |
1393 | ||
1394 | return i915_gem_get_seqno(ring->dev, &ring->outstanding_lazy_request); | |
1395 | } | |
1396 | ||
cbcc80df MK |
1397 | static int __intel_ring_begin(struct intel_ring_buffer *ring, |
1398 | int bytes) | |
1399 | { | |
1400 | int ret; | |
1401 | ||
1402 | if (unlikely(ring->tail + bytes > ring->effective_size)) { | |
1403 | ret = intel_wrap_ring_buffer(ring); | |
1404 | if (unlikely(ret)) | |
1405 | return ret; | |
1406 | } | |
1407 | ||
1408 | if (unlikely(ring->space < bytes)) { | |
1409 | ret = ring_wait_for_space(ring, bytes); | |
1410 | if (unlikely(ret)) | |
1411 | return ret; | |
1412 | } | |
1413 | ||
1414 | ring->space -= bytes; | |
1415 | return 0; | |
1416 | } | |
1417 | ||
e1f99ce6 CW |
1418 | int intel_ring_begin(struct intel_ring_buffer *ring, |
1419 | int num_dwords) | |
8187a2b7 | 1420 | { |
de2b9985 | 1421 | drm_i915_private_t *dev_priv = ring->dev->dev_private; |
e1f99ce6 | 1422 | int ret; |
78501eac | 1423 | |
de2b9985 DV |
1424 | ret = i915_gem_check_wedge(dev_priv, dev_priv->mm.interruptible); |
1425 | if (ret) | |
1426 | return ret; | |
21dd3734 | 1427 | |
9d773091 CW |
1428 | /* Preallocate the olr before touching the ring */ |
1429 | ret = intel_ring_alloc_seqno(ring); | |
1430 | if (ret) | |
1431 | return ret; | |
1432 | ||
cbcc80df | 1433 | return __intel_ring_begin(ring, num_dwords * sizeof(uint32_t)); |
8187a2b7 | 1434 | } |
62fdfeaf | 1435 | |
f7e98ad4 | 1436 | void intel_ring_init_seqno(struct intel_ring_buffer *ring, u32 seqno) |
498d2ac1 | 1437 | { |
f7e98ad4 | 1438 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
498d2ac1 MK |
1439 | |
1440 | BUG_ON(ring->outstanding_lazy_request); | |
1441 | ||
f7e98ad4 MK |
1442 | if (INTEL_INFO(ring->dev)->gen >= 6) { |
1443 | I915_WRITE(RING_SYNC_0(ring->mmio_base), 0); | |
1444 | I915_WRITE(RING_SYNC_1(ring->mmio_base), 0); | |
1445 | } | |
498d2ac1 | 1446 | |
f7e98ad4 | 1447 | ring->set_seqno(ring, seqno); |
498d2ac1 MK |
1448 | } |
1449 | ||
78501eac | 1450 | void intel_ring_advance(struct intel_ring_buffer *ring) |
8187a2b7 | 1451 | { |
e5eb3d63 DV |
1452 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
1453 | ||
d97ed339 | 1454 | ring->tail &= ring->size - 1; |
e5eb3d63 DV |
1455 | if (dev_priv->stop_rings & intel_ring_flag(ring)) |
1456 | return; | |
78501eac | 1457 | ring->write_tail(ring, ring->tail); |
8187a2b7 | 1458 | } |
62fdfeaf | 1459 | |
881f47b6 | 1460 | |
78501eac | 1461 | static void gen6_bsd_ring_write_tail(struct intel_ring_buffer *ring, |
297b0c5b | 1462 | u32 value) |
881f47b6 | 1463 | { |
0206e353 | 1464 | drm_i915_private_t *dev_priv = ring->dev->dev_private; |
881f47b6 XH |
1465 | |
1466 | /* Every tail move must follow the sequence below */ | |
12f55818 CW |
1467 | |
1468 | /* Disable notification that the ring is IDLE. The GT | |
1469 | * will then assume that it is busy and bring it out of rc6. | |
1470 | */ | |
0206e353 | 1471 | I915_WRITE(GEN6_BSD_SLEEP_PSMI_CONTROL, |
12f55818 CW |
1472 | _MASKED_BIT_ENABLE(GEN6_BSD_SLEEP_MSG_DISABLE)); |
1473 | ||
1474 | /* Clear the context id. Here be magic! */ | |
1475 | I915_WRITE64(GEN6_BSD_RNCID, 0x0); | |
0206e353 | 1476 | |
12f55818 | 1477 | /* Wait for the ring not to be idle, i.e. for it to wake up. */ |
0206e353 | 1478 | if (wait_for((I915_READ(GEN6_BSD_SLEEP_PSMI_CONTROL) & |
12f55818 CW |
1479 | GEN6_BSD_SLEEP_INDICATOR) == 0, |
1480 | 50)) | |
1481 | DRM_ERROR("timed out waiting for the BSD ring to wake up\n"); | |
0206e353 | 1482 | |
12f55818 | 1483 | /* Now that the ring is fully powered up, update the tail */ |
0206e353 | 1484 | I915_WRITE_TAIL(ring, value); |
12f55818 CW |
1485 | POSTING_READ(RING_TAIL(ring->mmio_base)); |
1486 | ||
1487 | /* Let the ring send IDLE messages to the GT again, | |
1488 | * and so let it sleep to conserve power when idle. | |
1489 | */ | |
0206e353 | 1490 | I915_WRITE(GEN6_BSD_SLEEP_PSMI_CONTROL, |
12f55818 | 1491 | _MASKED_BIT_DISABLE(GEN6_BSD_SLEEP_MSG_DISABLE)); |
881f47b6 XH |
1492 | } |
1493 | ||
b72f3acb | 1494 | static int gen6_ring_flush(struct intel_ring_buffer *ring, |
71a77e07 | 1495 | u32 invalidate, u32 flush) |
881f47b6 | 1496 | { |
71a77e07 | 1497 | uint32_t cmd; |
b72f3acb CW |
1498 | int ret; |
1499 | ||
b72f3acb CW |
1500 | ret = intel_ring_begin(ring, 4); |
1501 | if (ret) | |
1502 | return ret; | |
1503 | ||
71a77e07 | 1504 | cmd = MI_FLUSH_DW; |
9a289771 JB |
1505 | /* |
1506 | * Bspec vol 1c.5 - video engine command streamer: | |
1507 | * "If ENABLED, all TLBs will be invalidated once the flush | |
1508 | * operation is complete. This bit is only valid when the | |
1509 | * Post-Sync Operation field is a value of 1h or 3h." | |
1510 | */ | |
71a77e07 | 1511 | if (invalidate & I915_GEM_GPU_DOMAINS) |
9a289771 JB |
1512 | cmd |= MI_INVALIDATE_TLB | MI_INVALIDATE_BSD | |
1513 | MI_FLUSH_DW_STORE_INDEX | MI_FLUSH_DW_OP_STOREDW; | |
71a77e07 | 1514 | intel_ring_emit(ring, cmd); |
9a289771 | 1515 | intel_ring_emit(ring, I915_GEM_HWS_SCRATCH_ADDR | MI_FLUSH_DW_USE_GTT); |
b72f3acb | 1516 | intel_ring_emit(ring, 0); |
71a77e07 | 1517 | intel_ring_emit(ring, MI_NOOP); |
b72f3acb CW |
1518 | intel_ring_advance(ring); |
1519 | return 0; | |
881f47b6 XH |
1520 | } |
1521 | ||
d7d4eedd CW |
1522 | static int |
1523 | hsw_ring_dispatch_execbuffer(struct intel_ring_buffer *ring, | |
1524 | u32 offset, u32 len, | |
1525 | unsigned flags) | |
1526 | { | |
1527 | int ret; | |
1528 | ||
1529 | ret = intel_ring_begin(ring, 2); | |
1530 | if (ret) | |
1531 | return ret; | |
1532 | ||
1533 | intel_ring_emit(ring, | |
1534 | MI_BATCH_BUFFER_START | MI_BATCH_PPGTT_HSW | | |
1535 | (flags & I915_DISPATCH_SECURE ? 0 : MI_BATCH_NON_SECURE_HSW)); | |
1536 | /* bit0-7 is the length on GEN6+ */ | |
1537 | intel_ring_emit(ring, offset); | |
1538 | intel_ring_advance(ring); | |
1539 | ||
1540 | return 0; | |
1541 | } | |
1542 | ||
881f47b6 | 1543 | static int |
78501eac | 1544 | gen6_ring_dispatch_execbuffer(struct intel_ring_buffer *ring, |
d7d4eedd CW |
1545 | u32 offset, u32 len, |
1546 | unsigned flags) | |
881f47b6 | 1547 | { |
0206e353 | 1548 | int ret; |
ab6f8e32 | 1549 | |
0206e353 AJ |
1550 | ret = intel_ring_begin(ring, 2); |
1551 | if (ret) | |
1552 | return ret; | |
e1f99ce6 | 1553 | |
d7d4eedd CW |
1554 | intel_ring_emit(ring, |
1555 | MI_BATCH_BUFFER_START | | |
1556 | (flags & I915_DISPATCH_SECURE ? 0 : MI_BATCH_NON_SECURE_I965)); | |
0206e353 AJ |
1557 | /* bit0-7 is the length on GEN6+ */ |
1558 | intel_ring_emit(ring, offset); | |
1559 | intel_ring_advance(ring); | |
ab6f8e32 | 1560 | |
0206e353 | 1561 | return 0; |
881f47b6 XH |
1562 | } |
1563 | ||
549f7365 CW |
1564 | /* Blitter support (SandyBridge+) */ |
1565 | ||
b72f3acb | 1566 | static int blt_ring_flush(struct intel_ring_buffer *ring, |
71a77e07 | 1567 | u32 invalidate, u32 flush) |
8d19215b | 1568 | { |
71a77e07 | 1569 | uint32_t cmd; |
b72f3acb CW |
1570 | int ret; |
1571 | ||
6a233c78 | 1572 | ret = intel_ring_begin(ring, 4); |
b72f3acb CW |
1573 | if (ret) |
1574 | return ret; | |
1575 | ||
71a77e07 | 1576 | cmd = MI_FLUSH_DW; |
9a289771 JB |
1577 | /* |
1578 | * Bspec vol 1c.3 - blitter engine command streamer: | |
1579 | * "If ENABLED, all TLBs will be invalidated once the flush | |
1580 | * operation is complete. This bit is only valid when the | |
1581 | * Post-Sync Operation field is a value of 1h or 3h." | |
1582 | */ | |
71a77e07 | 1583 | if (invalidate & I915_GEM_DOMAIN_RENDER) |
9a289771 | 1584 | cmd |= MI_INVALIDATE_TLB | MI_FLUSH_DW_STORE_INDEX | |
b3fcabb1 | 1585 | MI_FLUSH_DW_OP_STOREDW; |
71a77e07 | 1586 | intel_ring_emit(ring, cmd); |
9a289771 | 1587 | intel_ring_emit(ring, I915_GEM_HWS_SCRATCH_ADDR | MI_FLUSH_DW_USE_GTT); |
b72f3acb | 1588 | intel_ring_emit(ring, 0); |
71a77e07 | 1589 | intel_ring_emit(ring, MI_NOOP); |
b72f3acb CW |
1590 | intel_ring_advance(ring); |
1591 | return 0; | |
8d19215b ZN |
1592 | } |
1593 | ||
5c1143bb XH |
1594 | int intel_init_render_ring_buffer(struct drm_device *dev) |
1595 | { | |
1596 | drm_i915_private_t *dev_priv = dev->dev_private; | |
1ec14ad3 | 1597 | struct intel_ring_buffer *ring = &dev_priv->ring[RCS]; |
5c1143bb | 1598 | |
59465b5f DV |
1599 | ring->name = "render ring"; |
1600 | ring->id = RCS; | |
1601 | ring->mmio_base = RENDER_RING_BASE; | |
1602 | ||
1ec14ad3 CW |
1603 | if (INTEL_INFO(dev)->gen >= 6) { |
1604 | ring->add_request = gen6_add_request; | |
4772eaeb | 1605 | ring->flush = gen7_render_ring_flush; |
6c6cf5aa | 1606 | if (INTEL_INFO(dev)->gen == 6) |
b3111509 | 1607 | ring->flush = gen6_render_ring_flush; |
25c06300 BW |
1608 | ring->irq_get = gen6_ring_get_irq; |
1609 | ring->irq_put = gen6_ring_put_irq; | |
6a848ccb | 1610 | ring->irq_enable_mask = GT_USER_INTERRUPT; |
4cd53c0c | 1611 | ring->get_seqno = gen6_ring_get_seqno; |
b70ec5bf | 1612 | ring->set_seqno = ring_set_seqno; |
686cb5f9 | 1613 | ring->sync_to = gen6_ring_sync; |
59465b5f DV |
1614 | ring->semaphore_register[0] = MI_SEMAPHORE_SYNC_INVALID; |
1615 | ring->semaphore_register[1] = MI_SEMAPHORE_SYNC_RV; | |
1616 | ring->semaphore_register[2] = MI_SEMAPHORE_SYNC_RB; | |
1617 | ring->signal_mbox[0] = GEN6_VRSYNC; | |
1618 | ring->signal_mbox[1] = GEN6_BRSYNC; | |
c6df541c CW |
1619 | } else if (IS_GEN5(dev)) { |
1620 | ring->add_request = pc_render_add_request; | |
46f0f8d1 | 1621 | ring->flush = gen4_render_ring_flush; |
c6df541c | 1622 | ring->get_seqno = pc_render_get_seqno; |
b70ec5bf | 1623 | ring->set_seqno = pc_render_set_seqno; |
e48d8634 DV |
1624 | ring->irq_get = gen5_ring_get_irq; |
1625 | ring->irq_put = gen5_ring_put_irq; | |
e3670319 | 1626 | ring->irq_enable_mask = GT_USER_INTERRUPT | GT_PIPE_NOTIFY; |
59465b5f | 1627 | } else { |
8620a3a9 | 1628 | ring->add_request = i9xx_add_request; |
46f0f8d1 CW |
1629 | if (INTEL_INFO(dev)->gen < 4) |
1630 | ring->flush = gen2_render_ring_flush; | |
1631 | else | |
1632 | ring->flush = gen4_render_ring_flush; | |
59465b5f | 1633 | ring->get_seqno = ring_get_seqno; |
b70ec5bf | 1634 | ring->set_seqno = ring_set_seqno; |
c2798b19 CW |
1635 | if (IS_GEN2(dev)) { |
1636 | ring->irq_get = i8xx_ring_get_irq; | |
1637 | ring->irq_put = i8xx_ring_put_irq; | |
1638 | } else { | |
1639 | ring->irq_get = i9xx_ring_get_irq; | |
1640 | ring->irq_put = i9xx_ring_put_irq; | |
1641 | } | |
e3670319 | 1642 | ring->irq_enable_mask = I915_USER_INTERRUPT; |
1ec14ad3 | 1643 | } |
59465b5f | 1644 | ring->write_tail = ring_write_tail; |
d7d4eedd CW |
1645 | if (IS_HASWELL(dev)) |
1646 | ring->dispatch_execbuffer = hsw_ring_dispatch_execbuffer; | |
1647 | else if (INTEL_INFO(dev)->gen >= 6) | |
fb3256da DV |
1648 | ring->dispatch_execbuffer = gen6_ring_dispatch_execbuffer; |
1649 | else if (INTEL_INFO(dev)->gen >= 4) | |
1650 | ring->dispatch_execbuffer = i965_dispatch_execbuffer; | |
1651 | else if (IS_I830(dev) || IS_845G(dev)) | |
1652 | ring->dispatch_execbuffer = i830_dispatch_execbuffer; | |
1653 | else | |
1654 | ring->dispatch_execbuffer = i915_dispatch_execbuffer; | |
59465b5f DV |
1655 | ring->init = init_render_ring; |
1656 | ring->cleanup = render_ring_cleanup; | |
1657 | ||
1ec14ad3 | 1658 | return intel_init_ring_buffer(dev, ring); |
5c1143bb XH |
1659 | } |
1660 | ||
e8616b6c CW |
1661 | int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size) |
1662 | { | |
1663 | drm_i915_private_t *dev_priv = dev->dev_private; | |
1664 | struct intel_ring_buffer *ring = &dev_priv->ring[RCS]; | |
6b8294a4 | 1665 | int ret; |
e8616b6c | 1666 | |
59465b5f DV |
1667 | ring->name = "render ring"; |
1668 | ring->id = RCS; | |
1669 | ring->mmio_base = RENDER_RING_BASE; | |
1670 | ||
e8616b6c | 1671 | if (INTEL_INFO(dev)->gen >= 6) { |
b4178f8a DV |
1672 | /* non-kms not supported on gen6+ */ |
1673 | return -ENODEV; | |
e8616b6c | 1674 | } |
28f0cbf7 DV |
1675 | |
1676 | /* Note: gem is not supported on gen5/ilk without kms (the corresponding | |
1677 | * gem_init ioctl returns with -ENODEV). Hence we do not need to set up | |
1678 | * the special gen5 functions. */ | |
1679 | ring->add_request = i9xx_add_request; | |
46f0f8d1 CW |
1680 | if (INTEL_INFO(dev)->gen < 4) |
1681 | ring->flush = gen2_render_ring_flush; | |
1682 | else | |
1683 | ring->flush = gen4_render_ring_flush; | |
28f0cbf7 | 1684 | ring->get_seqno = ring_get_seqno; |
b70ec5bf | 1685 | ring->set_seqno = ring_set_seqno; |
c2798b19 CW |
1686 | if (IS_GEN2(dev)) { |
1687 | ring->irq_get = i8xx_ring_get_irq; | |
1688 | ring->irq_put = i8xx_ring_put_irq; | |
1689 | } else { | |
1690 | ring->irq_get = i9xx_ring_get_irq; | |
1691 | ring->irq_put = i9xx_ring_put_irq; | |
1692 | } | |
28f0cbf7 | 1693 | ring->irq_enable_mask = I915_USER_INTERRUPT; |
59465b5f | 1694 | ring->write_tail = ring_write_tail; |
fb3256da DV |
1695 | if (INTEL_INFO(dev)->gen >= 4) |
1696 | ring->dispatch_execbuffer = i965_dispatch_execbuffer; | |
1697 | else if (IS_I830(dev) || IS_845G(dev)) | |
1698 | ring->dispatch_execbuffer = i830_dispatch_execbuffer; | |
1699 | else | |
1700 | ring->dispatch_execbuffer = i915_dispatch_execbuffer; | |
59465b5f DV |
1701 | ring->init = init_render_ring; |
1702 | ring->cleanup = render_ring_cleanup; | |
e8616b6c CW |
1703 | |
1704 | ring->dev = dev; | |
1705 | INIT_LIST_HEAD(&ring->active_list); | |
1706 | INIT_LIST_HEAD(&ring->request_list); | |
e8616b6c CW |
1707 | |
1708 | ring->size = size; | |
1709 | ring->effective_size = ring->size; | |
17f10fdc | 1710 | if (IS_I830(ring->dev) || IS_845G(ring->dev)) |
e8616b6c CW |
1711 | ring->effective_size -= 128; |
1712 | ||
4225d0f2 DV |
1713 | ring->virtual_start = ioremap_wc(start, size); |
1714 | if (ring->virtual_start == NULL) { | |
e8616b6c CW |
1715 | DRM_ERROR("can not ioremap virtual address for" |
1716 | " ring buffer\n"); | |
1717 | return -ENOMEM; | |
1718 | } | |
1719 | ||
6b8294a4 CW |
1720 | if (!I915_NEED_GFX_HWS(dev)) { |
1721 | ret = init_phys_hws_pga(ring); | |
1722 | if (ret) | |
1723 | return ret; | |
1724 | } | |
1725 | ||
e8616b6c CW |
1726 | return 0; |
1727 | } | |
1728 | ||
5c1143bb XH |
1729 | int intel_init_bsd_ring_buffer(struct drm_device *dev) |
1730 | { | |
1731 | drm_i915_private_t *dev_priv = dev->dev_private; | |
1ec14ad3 | 1732 | struct intel_ring_buffer *ring = &dev_priv->ring[VCS]; |
5c1143bb | 1733 | |
58fa3835 DV |
1734 | ring->name = "bsd ring"; |
1735 | ring->id = VCS; | |
1736 | ||
0fd2c201 | 1737 | ring->write_tail = ring_write_tail; |
58fa3835 DV |
1738 | if (IS_GEN6(dev) || IS_GEN7(dev)) { |
1739 | ring->mmio_base = GEN6_BSD_RING_BASE; | |
0fd2c201 DV |
1740 | /* gen6 bsd needs a special wa for tail updates */ |
1741 | if (IS_GEN6(dev)) | |
1742 | ring->write_tail = gen6_bsd_ring_write_tail; | |
58fa3835 DV |
1743 | ring->flush = gen6_ring_flush; |
1744 | ring->add_request = gen6_add_request; | |
1745 | ring->get_seqno = gen6_ring_get_seqno; | |
b70ec5bf | 1746 | ring->set_seqno = ring_set_seqno; |
58fa3835 DV |
1747 | ring->irq_enable_mask = GEN6_BSD_USER_INTERRUPT; |
1748 | ring->irq_get = gen6_ring_get_irq; | |
1749 | ring->irq_put = gen6_ring_put_irq; | |
1750 | ring->dispatch_execbuffer = gen6_ring_dispatch_execbuffer; | |
686cb5f9 | 1751 | ring->sync_to = gen6_ring_sync; |
58fa3835 DV |
1752 | ring->semaphore_register[0] = MI_SEMAPHORE_SYNC_VR; |
1753 | ring->semaphore_register[1] = MI_SEMAPHORE_SYNC_INVALID; | |
1754 | ring->semaphore_register[2] = MI_SEMAPHORE_SYNC_VB; | |
1755 | ring->signal_mbox[0] = GEN6_RVSYNC; | |
1756 | ring->signal_mbox[1] = GEN6_BVSYNC; | |
1757 | } else { | |
1758 | ring->mmio_base = BSD_RING_BASE; | |
58fa3835 | 1759 | ring->flush = bsd_ring_flush; |
8620a3a9 | 1760 | ring->add_request = i9xx_add_request; |
58fa3835 | 1761 | ring->get_seqno = ring_get_seqno; |
b70ec5bf | 1762 | ring->set_seqno = ring_set_seqno; |
e48d8634 | 1763 | if (IS_GEN5(dev)) { |
e3670319 | 1764 | ring->irq_enable_mask = GT_BSD_USER_INTERRUPT; |
e48d8634 DV |
1765 | ring->irq_get = gen5_ring_get_irq; |
1766 | ring->irq_put = gen5_ring_put_irq; | |
1767 | } else { | |
e3670319 | 1768 | ring->irq_enable_mask = I915_BSD_USER_INTERRUPT; |
e48d8634 DV |
1769 | ring->irq_get = i9xx_ring_get_irq; |
1770 | ring->irq_put = i9xx_ring_put_irq; | |
1771 | } | |
fb3256da | 1772 | ring->dispatch_execbuffer = i965_dispatch_execbuffer; |
58fa3835 DV |
1773 | } |
1774 | ring->init = init_ring_common; | |
1775 | ||
1ec14ad3 | 1776 | return intel_init_ring_buffer(dev, ring); |
5c1143bb | 1777 | } |
549f7365 CW |
1778 | |
1779 | int intel_init_blt_ring_buffer(struct drm_device *dev) | |
1780 | { | |
1781 | drm_i915_private_t *dev_priv = dev->dev_private; | |
1ec14ad3 | 1782 | struct intel_ring_buffer *ring = &dev_priv->ring[BCS]; |
549f7365 | 1783 | |
3535d9dd DV |
1784 | ring->name = "blitter ring"; |
1785 | ring->id = BCS; | |
1786 | ||
1787 | ring->mmio_base = BLT_RING_BASE; | |
1788 | ring->write_tail = ring_write_tail; | |
1789 | ring->flush = blt_ring_flush; | |
1790 | ring->add_request = gen6_add_request; | |
1791 | ring->get_seqno = gen6_ring_get_seqno; | |
b70ec5bf | 1792 | ring->set_seqno = ring_set_seqno; |
3535d9dd DV |
1793 | ring->irq_enable_mask = GEN6_BLITTER_USER_INTERRUPT; |
1794 | ring->irq_get = gen6_ring_get_irq; | |
1795 | ring->irq_put = gen6_ring_put_irq; | |
1796 | ring->dispatch_execbuffer = gen6_ring_dispatch_execbuffer; | |
686cb5f9 | 1797 | ring->sync_to = gen6_ring_sync; |
3535d9dd DV |
1798 | ring->semaphore_register[0] = MI_SEMAPHORE_SYNC_BR; |
1799 | ring->semaphore_register[1] = MI_SEMAPHORE_SYNC_BV; | |
1800 | ring->semaphore_register[2] = MI_SEMAPHORE_SYNC_INVALID; | |
1801 | ring->signal_mbox[0] = GEN6_RBSYNC; | |
1802 | ring->signal_mbox[1] = GEN6_VBSYNC; | |
1803 | ring->init = init_ring_common; | |
549f7365 | 1804 | |
1ec14ad3 | 1805 | return intel_init_ring_buffer(dev, ring); |
549f7365 | 1806 | } |
a7b9761d CW |
1807 | |
1808 | int | |
1809 | intel_ring_flush_all_caches(struct intel_ring_buffer *ring) | |
1810 | { | |
1811 | int ret; | |
1812 | ||
1813 | if (!ring->gpu_caches_dirty) | |
1814 | return 0; | |
1815 | ||
1816 | ret = ring->flush(ring, 0, I915_GEM_GPU_DOMAINS); | |
1817 | if (ret) | |
1818 | return ret; | |
1819 | ||
1820 | trace_i915_gem_ring_flush(ring, 0, I915_GEM_GPU_DOMAINS); | |
1821 | ||
1822 | ring->gpu_caches_dirty = false; | |
1823 | return 0; | |
1824 | } | |
1825 | ||
1826 | int | |
1827 | intel_ring_invalidate_all_caches(struct intel_ring_buffer *ring) | |
1828 | { | |
1829 | uint32_t flush_domains; | |
1830 | int ret; | |
1831 | ||
1832 | flush_domains = 0; | |
1833 | if (ring->gpu_caches_dirty) | |
1834 | flush_domains = I915_GEM_GPU_DOMAINS; | |
1835 | ||
1836 | ret = ring->flush(ring, I915_GEM_GPU_DOMAINS, flush_domains); | |
1837 | if (ret) | |
1838 | return ret; | |
1839 | ||
1840 | trace_i915_gem_ring_flush(ring, I915_GEM_GPU_DOMAINS, flush_domains); | |
1841 | ||
1842 | ring->gpu_caches_dirty = false; | |
1843 | return 0; | |
1844 | } |