Commit | Line | Data |
---|---|---|
7198e6b0 RC |
1 | /* |
2 | * Copyright (C) 2013 Red Hat | |
3 | * Author: Rob Clark <robdclark@gmail.com> | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify it | |
6 | * under the terms of the GNU General Public License version 2 as published by | |
7 | * the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, but WITHOUT | |
10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
12 | * more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License along with | |
15 | * this program. If not, see <http://www.gnu.org/licenses/>. | |
16 | */ | |
17 | ||
18 | #include "msm_gpu.h" | |
19 | #include "msm_gem.h" | |
871d812a | 20 | #include "msm_mmu.h" |
fde5de6c | 21 | #include "msm_fence.h" |
7198e6b0 RC |
22 | |
23 | ||
24 | /* | |
25 | * Power Management: | |
26 | */ | |
27 | ||
6490ad47 | 28 | #ifdef DOWNSTREAM_CONFIG_MSM_BUS_SCALING |
7198e6b0 | 29 | #include <mach/board.h> |
bf2b33af | 30 | static void bs_init(struct msm_gpu *gpu) |
7198e6b0 | 31 | { |
bf2b33af RC |
32 | if (gpu->bus_scale_table) { |
33 | gpu->bsc = msm_bus_scale_register_client(gpu->bus_scale_table); | |
7198e6b0 RC |
34 | DBG("bus scale client: %08x", gpu->bsc); |
35 | } | |
36 | } | |
37 | ||
38 | static void bs_fini(struct msm_gpu *gpu) | |
39 | { | |
40 | if (gpu->bsc) { | |
41 | msm_bus_scale_unregister_client(gpu->bsc); | |
42 | gpu->bsc = 0; | |
43 | } | |
44 | } | |
45 | ||
46 | static void bs_set(struct msm_gpu *gpu, int idx) | |
47 | { | |
48 | if (gpu->bsc) { | |
49 | DBG("set bus scaling: %d", idx); | |
50 | msm_bus_scale_client_update_request(gpu->bsc, idx); | |
51 | } | |
52 | } | |
53 | #else | |
bf2b33af | 54 | static void bs_init(struct msm_gpu *gpu) {} |
7198e6b0 RC |
55 | static void bs_fini(struct msm_gpu *gpu) {} |
56 | static void bs_set(struct msm_gpu *gpu, int idx) {} | |
57 | #endif | |
58 | ||
59 | static int enable_pwrrail(struct msm_gpu *gpu) | |
60 | { | |
61 | struct drm_device *dev = gpu->dev; | |
62 | int ret = 0; | |
63 | ||
64 | if (gpu->gpu_reg) { | |
65 | ret = regulator_enable(gpu->gpu_reg); | |
66 | if (ret) { | |
67 | dev_err(dev->dev, "failed to enable 'gpu_reg': %d\n", ret); | |
68 | return ret; | |
69 | } | |
70 | } | |
71 | ||
72 | if (gpu->gpu_cx) { | |
73 | ret = regulator_enable(gpu->gpu_cx); | |
74 | if (ret) { | |
75 | dev_err(dev->dev, "failed to enable 'gpu_cx': %d\n", ret); | |
76 | return ret; | |
77 | } | |
78 | } | |
79 | ||
80 | return 0; | |
81 | } | |
82 | ||
83 | static int disable_pwrrail(struct msm_gpu *gpu) | |
84 | { | |
85 | if (gpu->gpu_cx) | |
86 | regulator_disable(gpu->gpu_cx); | |
87 | if (gpu->gpu_reg) | |
88 | regulator_disable(gpu->gpu_reg); | |
89 | return 0; | |
90 | } | |
91 | ||
92 | static int enable_clk(struct msm_gpu *gpu) | |
93 | { | |
7198e6b0 RC |
94 | int i; |
95 | ||
98db803f JC |
96 | if (gpu->core_clk && gpu->fast_rate) |
97 | clk_set_rate(gpu->core_clk, gpu->fast_rate); | |
7198e6b0 | 98 | |
b5f103ab | 99 | /* Set the RBBM timer rate to 19.2Mhz */ |
98db803f JC |
100 | if (gpu->rbbmtimer_clk) |
101 | clk_set_rate(gpu->rbbmtimer_clk, 19200000); | |
b5f103ab | 102 | |
98db803f | 103 | for (i = gpu->nr_clocks - 1; i >= 0; i--) |
89d777a5 JC |
104 | if (gpu->grp_clks[i]) |
105 | clk_prepare(gpu->grp_clks[i]); | |
7198e6b0 | 106 | |
98db803f | 107 | for (i = gpu->nr_clocks - 1; i >= 0; i--) |
7198e6b0 RC |
108 | if (gpu->grp_clks[i]) |
109 | clk_enable(gpu->grp_clks[i]); | |
110 | ||
111 | return 0; | |
112 | } | |
113 | ||
114 | static int disable_clk(struct msm_gpu *gpu) | |
115 | { | |
7198e6b0 RC |
116 | int i; |
117 | ||
98db803f | 118 | for (i = gpu->nr_clocks - 1; i >= 0; i--) |
89d777a5 | 119 | if (gpu->grp_clks[i]) |
7198e6b0 | 120 | clk_disable(gpu->grp_clks[i]); |
7198e6b0 | 121 | |
98db803f | 122 | for (i = gpu->nr_clocks - 1; i >= 0; i--) |
7198e6b0 RC |
123 | if (gpu->grp_clks[i]) |
124 | clk_unprepare(gpu->grp_clks[i]); | |
125 | ||
bf5af4ae JC |
126 | /* |
127 | * Set the clock to a deliberately low rate. On older targets the clock | |
128 | * speed had to be non zero to avoid problems. On newer targets this | |
129 | * will be rounded down to zero anyway so it all works out. | |
130 | */ | |
98db803f JC |
131 | if (gpu->core_clk) |
132 | clk_set_rate(gpu->core_clk, 27000000); | |
89d777a5 | 133 | |
98db803f JC |
134 | if (gpu->rbbmtimer_clk) |
135 | clk_set_rate(gpu->rbbmtimer_clk, 0); | |
b5f103ab | 136 | |
7198e6b0 RC |
137 | return 0; |
138 | } | |
139 | ||
140 | static int enable_axi(struct msm_gpu *gpu) | |
141 | { | |
142 | if (gpu->ebi1_clk) | |
143 | clk_prepare_enable(gpu->ebi1_clk); | |
144 | if (gpu->bus_freq) | |
145 | bs_set(gpu, gpu->bus_freq); | |
146 | return 0; | |
147 | } | |
148 | ||
149 | static int disable_axi(struct msm_gpu *gpu) | |
150 | { | |
151 | if (gpu->ebi1_clk) | |
152 | clk_disable_unprepare(gpu->ebi1_clk); | |
153 | if (gpu->bus_freq) | |
154 | bs_set(gpu, 0); | |
155 | return 0; | |
156 | } | |
157 | ||
158 | int msm_gpu_pm_resume(struct msm_gpu *gpu) | |
159 | { | |
160 | int ret; | |
161 | ||
eeb75474 | 162 | DBG("%s", gpu->name); |
7198e6b0 RC |
163 | |
164 | ret = enable_pwrrail(gpu); | |
165 | if (ret) | |
166 | return ret; | |
167 | ||
168 | ret = enable_clk(gpu); | |
169 | if (ret) | |
170 | return ret; | |
171 | ||
172 | ret = enable_axi(gpu); | |
173 | if (ret) | |
174 | return ret; | |
175 | ||
eeb75474 RC |
176 | gpu->needs_hw_init = true; |
177 | ||
7198e6b0 RC |
178 | return 0; |
179 | } | |
180 | ||
181 | int msm_gpu_pm_suspend(struct msm_gpu *gpu) | |
182 | { | |
183 | int ret; | |
184 | ||
eeb75474 | 185 | DBG("%s", gpu->name); |
7198e6b0 RC |
186 | |
187 | ret = disable_axi(gpu); | |
188 | if (ret) | |
189 | return ret; | |
190 | ||
191 | ret = disable_clk(gpu); | |
192 | if (ret) | |
193 | return ret; | |
194 | ||
195 | ret = disable_pwrrail(gpu); | |
196 | if (ret) | |
197 | return ret; | |
198 | ||
199 | return 0; | |
200 | } | |
201 | ||
eeb75474 | 202 | int msm_gpu_hw_init(struct msm_gpu *gpu) |
37d77c3a | 203 | { |
eeb75474 | 204 | int ret; |
37d77c3a | 205 | |
cb1e3818 RC |
206 | WARN_ON(!mutex_is_locked(&gpu->dev->struct_mutex)); |
207 | ||
eeb75474 RC |
208 | if (!gpu->needs_hw_init) |
209 | return 0; | |
37d77c3a | 210 | |
eeb75474 RC |
211 | disable_irq(gpu->irq); |
212 | ret = gpu->funcs->hw_init(gpu); | |
213 | if (!ret) | |
214 | gpu->needs_hw_init = false; | |
215 | enable_irq(gpu->irq); | |
37d77c3a | 216 | |
eeb75474 | 217 | return ret; |
37d77c3a RC |
218 | } |
219 | ||
bd6f82d8 RC |
220 | /* |
221 | * Hangcheck detection for locked gpu: | |
222 | */ | |
223 | ||
b6295f9a | 224 | static void retire_submits(struct msm_gpu *gpu); |
1a370be9 | 225 | |
bd6f82d8 RC |
226 | static void recover_worker(struct work_struct *work) |
227 | { | |
228 | struct msm_gpu *gpu = container_of(work, struct msm_gpu, recover_work); | |
229 | struct drm_device *dev = gpu->dev; | |
4816b626 | 230 | struct msm_gem_submit *submit; |
b6295f9a | 231 | uint32_t fence = gpu->funcs->last_fence(gpu); |
bd6f82d8 | 232 | |
b6295f9a RC |
233 | msm_update_fence(gpu->fctx, fence + 1); |
234 | ||
bd6f82d8 | 235 | mutex_lock(&dev->struct_mutex); |
1a370be9 | 236 | |
4816b626 RC |
237 | dev_err(dev->dev, "%s: hangcheck recover!\n", gpu->name); |
238 | list_for_each_entry(submit, &gpu->submit_list, node) { | |
239 | if (submit->fence->seqno == (fence + 1)) { | |
240 | struct task_struct *task; | |
241 | ||
242 | rcu_read_lock(); | |
243 | task = pid_task(submit->pid, PIDTYPE_PID); | |
244 | if (task) { | |
245 | dev_err(dev->dev, "%s: offending task: %s\n", | |
246 | gpu->name, task->comm); | |
247 | } | |
248 | rcu_read_unlock(); | |
249 | break; | |
250 | } | |
251 | } | |
252 | ||
253 | if (msm_gpu_active(gpu)) { | |
1a370be9 | 254 | /* retire completed submits, plus the one that hung: */ |
b6295f9a | 255 | retire_submits(gpu); |
1a370be9 | 256 | |
eeb75474 | 257 | pm_runtime_get_sync(&gpu->pdev->dev); |
37d77c3a | 258 | gpu->funcs->recover(gpu); |
eeb75474 | 259 | pm_runtime_put_sync(&gpu->pdev->dev); |
1a370be9 RC |
260 | |
261 | /* replay the remaining submits after the one that hung: */ | |
262 | list_for_each_entry(submit, &gpu->submit_list, node) { | |
263 | gpu->funcs->submit(gpu, submit, NULL); | |
264 | } | |
37d77c3a | 265 | } |
4816b626 | 266 | |
bd6f82d8 RC |
267 | mutex_unlock(&dev->struct_mutex); |
268 | ||
269 | msm_gpu_retire(gpu); | |
270 | } | |
271 | ||
272 | static void hangcheck_timer_reset(struct msm_gpu *gpu) | |
273 | { | |
274 | DBG("%s", gpu->name); | |
275 | mod_timer(&gpu->hangcheck_timer, | |
276 | round_jiffies_up(jiffies + DRM_MSM_HANGCHECK_JIFFIES)); | |
277 | } | |
278 | ||
279 | static void hangcheck_handler(unsigned long data) | |
280 | { | |
281 | struct msm_gpu *gpu = (struct msm_gpu *)data; | |
6b8819c8 RC |
282 | struct drm_device *dev = gpu->dev; |
283 | struct msm_drm_private *priv = dev->dev_private; | |
bd6f82d8 RC |
284 | uint32_t fence = gpu->funcs->last_fence(gpu); |
285 | ||
286 | if (fence != gpu->hangcheck_fence) { | |
287 | /* some progress has been made.. ya! */ | |
288 | gpu->hangcheck_fence = fence; | |
ca762a8a | 289 | } else if (fence < gpu->fctx->last_fence) { |
bd6f82d8 | 290 | /* no progress and not done.. hung! */ |
bd6f82d8 | 291 | gpu->hangcheck_fence = fence; |
26791c48 RC |
292 | dev_err(dev->dev, "%s: hangcheck detected gpu lockup!\n", |
293 | gpu->name); | |
294 | dev_err(dev->dev, "%s: completed fence: %u\n", | |
295 | gpu->name, fence); | |
296 | dev_err(dev->dev, "%s: submitted fence: %u\n", | |
ca762a8a | 297 | gpu->name, gpu->fctx->last_fence); |
bd6f82d8 RC |
298 | queue_work(priv->wq, &gpu->recover_work); |
299 | } | |
300 | ||
301 | /* if still more pending work, reset the hangcheck timer: */ | |
ca762a8a | 302 | if (gpu->fctx->last_fence > gpu->hangcheck_fence) |
bd6f82d8 | 303 | hangcheck_timer_reset(gpu); |
6b8819c8 RC |
304 | |
305 | /* workaround for missing irq: */ | |
306 | queue_work(priv->wq, &gpu->retire_work); | |
bd6f82d8 RC |
307 | } |
308 | ||
70c70f09 RC |
309 | /* |
310 | * Performance Counters: | |
311 | */ | |
312 | ||
313 | /* called under perf_lock */ | |
314 | static int update_hw_cntrs(struct msm_gpu *gpu, uint32_t ncntrs, uint32_t *cntrs) | |
315 | { | |
316 | uint32_t current_cntrs[ARRAY_SIZE(gpu->last_cntrs)]; | |
317 | int i, n = min(ncntrs, gpu->num_perfcntrs); | |
318 | ||
319 | /* read current values: */ | |
320 | for (i = 0; i < gpu->num_perfcntrs; i++) | |
321 | current_cntrs[i] = gpu_read(gpu, gpu->perfcntrs[i].sample_reg); | |
322 | ||
323 | /* update cntrs: */ | |
324 | for (i = 0; i < n; i++) | |
325 | cntrs[i] = current_cntrs[i] - gpu->last_cntrs[i]; | |
326 | ||
327 | /* save current values: */ | |
328 | for (i = 0; i < gpu->num_perfcntrs; i++) | |
329 | gpu->last_cntrs[i] = current_cntrs[i]; | |
330 | ||
331 | return n; | |
332 | } | |
333 | ||
334 | static void update_sw_cntrs(struct msm_gpu *gpu) | |
335 | { | |
336 | ktime_t time; | |
337 | uint32_t elapsed; | |
338 | unsigned long flags; | |
339 | ||
340 | spin_lock_irqsave(&gpu->perf_lock, flags); | |
341 | if (!gpu->perfcntr_active) | |
342 | goto out; | |
343 | ||
344 | time = ktime_get(); | |
345 | elapsed = ktime_to_us(ktime_sub(time, gpu->last_sample.time)); | |
346 | ||
347 | gpu->totaltime += elapsed; | |
348 | if (gpu->last_sample.active) | |
349 | gpu->activetime += elapsed; | |
350 | ||
351 | gpu->last_sample.active = msm_gpu_active(gpu); | |
352 | gpu->last_sample.time = time; | |
353 | ||
354 | out: | |
355 | spin_unlock_irqrestore(&gpu->perf_lock, flags); | |
356 | } | |
357 | ||
358 | void msm_gpu_perfcntr_start(struct msm_gpu *gpu) | |
359 | { | |
360 | unsigned long flags; | |
361 | ||
eeb75474 RC |
362 | pm_runtime_get_sync(&gpu->pdev->dev); |
363 | ||
70c70f09 RC |
364 | spin_lock_irqsave(&gpu->perf_lock, flags); |
365 | /* we could dynamically enable/disable perfcntr registers too.. */ | |
366 | gpu->last_sample.active = msm_gpu_active(gpu); | |
367 | gpu->last_sample.time = ktime_get(); | |
368 | gpu->activetime = gpu->totaltime = 0; | |
369 | gpu->perfcntr_active = true; | |
370 | update_hw_cntrs(gpu, 0, NULL); | |
371 | spin_unlock_irqrestore(&gpu->perf_lock, flags); | |
372 | } | |
373 | ||
374 | void msm_gpu_perfcntr_stop(struct msm_gpu *gpu) | |
375 | { | |
376 | gpu->perfcntr_active = false; | |
eeb75474 | 377 | pm_runtime_put_sync(&gpu->pdev->dev); |
70c70f09 RC |
378 | } |
379 | ||
380 | /* returns -errno or # of cntrs sampled */ | |
381 | int msm_gpu_perfcntr_sample(struct msm_gpu *gpu, uint32_t *activetime, | |
382 | uint32_t *totaltime, uint32_t ncntrs, uint32_t *cntrs) | |
383 | { | |
384 | unsigned long flags; | |
385 | int ret; | |
386 | ||
387 | spin_lock_irqsave(&gpu->perf_lock, flags); | |
388 | ||
389 | if (!gpu->perfcntr_active) { | |
390 | ret = -EINVAL; | |
391 | goto out; | |
392 | } | |
393 | ||
394 | *activetime = gpu->activetime; | |
395 | *totaltime = gpu->totaltime; | |
396 | ||
397 | gpu->activetime = gpu->totaltime = 0; | |
398 | ||
399 | ret = update_hw_cntrs(gpu, ncntrs, cntrs); | |
400 | ||
401 | out: | |
402 | spin_unlock_irqrestore(&gpu->perf_lock, flags); | |
403 | ||
404 | return ret; | |
405 | } | |
406 | ||
7198e6b0 RC |
407 | /* |
408 | * Cmdstream submission/retirement: | |
409 | */ | |
410 | ||
7d12a279 RC |
411 | static void retire_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit) |
412 | { | |
413 | int i; | |
414 | ||
415 | for (i = 0; i < submit->nr_bos; i++) { | |
416 | struct msm_gem_object *msm_obj = submit->bos[i].obj; | |
417 | /* move to inactive: */ | |
418 | msm_gem_move_to_inactive(&msm_obj->base); | |
8bdcd949 | 419 | msm_gem_put_iova(&msm_obj->base, gpu->aspace); |
7d12a279 RC |
420 | drm_gem_object_unreference(&msm_obj->base); |
421 | } | |
422 | ||
eeb75474 RC |
423 | pm_runtime_mark_last_busy(&gpu->pdev->dev); |
424 | pm_runtime_put_autosuspend(&gpu->pdev->dev); | |
40e6815b | 425 | msm_gem_submit_free(submit); |
7d12a279 RC |
426 | } |
427 | ||
b6295f9a | 428 | static void retire_submits(struct msm_gpu *gpu) |
1a370be9 RC |
429 | { |
430 | struct drm_device *dev = gpu->dev; | |
431 | ||
432 | WARN_ON(!mutex_is_locked(&dev->struct_mutex)); | |
433 | ||
434 | while (!list_empty(&gpu->submit_list)) { | |
435 | struct msm_gem_submit *submit; | |
436 | ||
437 | submit = list_first_entry(&gpu->submit_list, | |
438 | struct msm_gem_submit, node); | |
439 | ||
f54d1867 | 440 | if (dma_fence_is_signaled(submit->fence)) { |
7d12a279 | 441 | retire_submit(gpu, submit); |
1a370be9 RC |
442 | } else { |
443 | break; | |
444 | } | |
445 | } | |
446 | } | |
447 | ||
7198e6b0 RC |
448 | static void retire_worker(struct work_struct *work) |
449 | { | |
450 | struct msm_gpu *gpu = container_of(work, struct msm_gpu, retire_work); | |
451 | struct drm_device *dev = gpu->dev; | |
452 | uint32_t fence = gpu->funcs->last_fence(gpu); | |
453 | ||
ca762a8a | 454 | msm_update_fence(gpu->fctx, fence); |
edd4fc63 | 455 | |
7198e6b0 | 456 | mutex_lock(&dev->struct_mutex); |
b6295f9a | 457 | retire_submits(gpu); |
7198e6b0 RC |
458 | mutex_unlock(&dev->struct_mutex); |
459 | } | |
460 | ||
461 | /* call from irq handler to schedule work to retire bo's */ | |
462 | void msm_gpu_retire(struct msm_gpu *gpu) | |
463 | { | |
464 | struct msm_drm_private *priv = gpu->dev->dev_private; | |
465 | queue_work(priv->wq, &gpu->retire_work); | |
70c70f09 | 466 | update_sw_cntrs(gpu); |
7198e6b0 RC |
467 | } |
468 | ||
469 | /* add bo's to gpu's ring, and kick gpu: */ | |
f44d32c7 | 470 | void msm_gpu_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit, |
7198e6b0 RC |
471 | struct msm_file_private *ctx) |
472 | { | |
473 | struct drm_device *dev = gpu->dev; | |
474 | struct msm_drm_private *priv = dev->dev_private; | |
f44d32c7 | 475 | int i; |
7198e6b0 | 476 | |
1a370be9 RC |
477 | WARN_ON(!mutex_is_locked(&dev->struct_mutex)); |
478 | ||
eeb75474 RC |
479 | pm_runtime_get_sync(&gpu->pdev->dev); |
480 | ||
481 | msm_gpu_hw_init(gpu); | |
37d77c3a | 482 | |
1a370be9 RC |
483 | list_add_tail(&submit->node, &gpu->submit_list); |
484 | ||
a7d3c950 RC |
485 | msm_rd_dump_submit(submit); |
486 | ||
70c70f09 RC |
487 | update_sw_cntrs(gpu); |
488 | ||
7198e6b0 RC |
489 | for (i = 0; i < submit->nr_bos; i++) { |
490 | struct msm_gem_object *msm_obj = submit->bos[i].obj; | |
78babc16 | 491 | uint64_t iova; |
7198e6b0 RC |
492 | |
493 | /* can't happen yet.. but when we add 2d support we'll have | |
494 | * to deal w/ cross-ring synchronization: | |
495 | */ | |
496 | WARN_ON(is_active(msm_obj) && (msm_obj->gpu != gpu)); | |
497 | ||
7d12a279 RC |
498 | /* submit takes a reference to the bo and iova until retired: */ |
499 | drm_gem_object_reference(&msm_obj->base); | |
0e08270a | 500 | msm_gem_get_iova(&msm_obj->base, |
8bdcd949 | 501 | submit->gpu->aspace, &iova); |
7198e6b0 | 502 | |
bf6811f3 RC |
503 | if (submit->bos[i].flags & MSM_SUBMIT_BO_WRITE) |
504 | msm_gem_move_to_active(&msm_obj->base, gpu, true, submit->fence); | |
b6295f9a RC |
505 | else if (submit->bos[i].flags & MSM_SUBMIT_BO_READ) |
506 | msm_gem_move_to_active(&msm_obj->base, gpu, false, submit->fence); | |
7198e6b0 | 507 | } |
1a370be9 | 508 | |
1193c3bc | 509 | gpu->funcs->submit(gpu, submit, ctx); |
1a370be9 RC |
510 | priv->lastctx = ctx; |
511 | ||
bd6f82d8 | 512 | hangcheck_timer_reset(gpu); |
7198e6b0 RC |
513 | } |
514 | ||
515 | /* | |
516 | * Init/Cleanup: | |
517 | */ | |
518 | ||
519 | static irqreturn_t irq_handler(int irq, void *data) | |
520 | { | |
521 | struct msm_gpu *gpu = data; | |
522 | return gpu->funcs->irq(gpu); | |
523 | } | |
524 | ||
98db803f JC |
525 | static struct clk *get_clock(struct device *dev, const char *name) |
526 | { | |
527 | struct clk *clk = devm_clk_get(dev, name); | |
528 | ||
529 | return IS_ERR(clk) ? NULL : clk; | |
530 | } | |
531 | ||
532 | static int get_clocks(struct platform_device *pdev, struct msm_gpu *gpu) | |
533 | { | |
534 | struct device *dev = &pdev->dev; | |
535 | struct property *prop; | |
536 | const char *name; | |
537 | int i = 0; | |
538 | ||
539 | gpu->nr_clocks = of_property_count_strings(dev->of_node, "clock-names"); | |
540 | if (gpu->nr_clocks < 1) { | |
541 | gpu->nr_clocks = 0; | |
542 | return 0; | |
543 | } | |
544 | ||
545 | gpu->grp_clks = devm_kcalloc(dev, sizeof(struct clk *), gpu->nr_clocks, | |
546 | GFP_KERNEL); | |
547 | if (!gpu->grp_clks) | |
548 | return -ENOMEM; | |
549 | ||
550 | of_property_for_each_string(dev->of_node, "clock-names", prop, name) { | |
551 | gpu->grp_clks[i] = get_clock(dev, name); | |
552 | ||
553 | /* Remember the key clocks that we need to control later */ | |
134ccada | 554 | if (!strcmp(name, "core") || !strcmp(name, "core_clk")) |
98db803f | 555 | gpu->core_clk = gpu->grp_clks[i]; |
134ccada | 556 | else if (!strcmp(name, "rbbmtimer") || !strcmp(name, "rbbmtimer_clk")) |
98db803f JC |
557 | gpu->rbbmtimer_clk = gpu->grp_clks[i]; |
558 | ||
559 | ++i; | |
560 | } | |
561 | ||
562 | return 0; | |
563 | } | |
7198e6b0 | 564 | |
1267a4df JC |
565 | static struct msm_gem_address_space * |
566 | msm_gpu_create_address_space(struct msm_gpu *gpu, struct platform_device *pdev, | |
567 | uint64_t va_start, uint64_t va_end) | |
568 | { | |
569 | struct iommu_domain *iommu; | |
570 | struct msm_gem_address_space *aspace; | |
571 | int ret; | |
572 | ||
573 | /* | |
574 | * Setup IOMMU.. eventually we will (I think) do this once per context | |
575 | * and have separate page tables per context. For now, to keep things | |
576 | * simple and to get something working, just use a single address space: | |
577 | */ | |
578 | iommu = iommu_domain_alloc(&platform_bus_type); | |
579 | if (!iommu) | |
580 | return NULL; | |
581 | ||
582 | iommu->geometry.aperture_start = va_start; | |
583 | iommu->geometry.aperture_end = va_end; | |
584 | ||
585 | dev_info(gpu->dev->dev, "%s: using IOMMU\n", gpu->name); | |
586 | ||
587 | aspace = msm_gem_address_space_create(&pdev->dev, iommu, "gpu"); | |
588 | if (IS_ERR(aspace)) { | |
589 | dev_err(gpu->dev->dev, "failed to init iommu: %ld\n", | |
590 | PTR_ERR(aspace)); | |
591 | iommu_domain_free(iommu); | |
592 | return ERR_CAST(aspace); | |
593 | } | |
594 | ||
595 | ret = aspace->mmu->funcs->attach(aspace->mmu, NULL, 0); | |
596 | if (ret) { | |
597 | msm_gem_address_space_put(aspace); | |
598 | return ERR_PTR(ret); | |
599 | } | |
600 | ||
601 | return aspace; | |
602 | } | |
603 | ||
7198e6b0 RC |
604 | int msm_gpu_init(struct drm_device *drm, struct platform_device *pdev, |
605 | struct msm_gpu *gpu, const struct msm_gpu_funcs *funcs, | |
5770fc7a | 606 | const char *name, struct msm_gpu_config *config) |
7198e6b0 | 607 | { |
98db803f | 608 | int ret; |
7198e6b0 | 609 | |
70c70f09 RC |
610 | if (WARN_ON(gpu->num_perfcntrs > ARRAY_SIZE(gpu->last_cntrs))) |
611 | gpu->num_perfcntrs = ARRAY_SIZE(gpu->last_cntrs); | |
612 | ||
7198e6b0 RC |
613 | gpu->dev = drm; |
614 | gpu->funcs = funcs; | |
615 | gpu->name = name; | |
ca762a8a RC |
616 | gpu->fctx = msm_fence_context_alloc(drm, name); |
617 | if (IS_ERR(gpu->fctx)) { | |
618 | ret = PTR_ERR(gpu->fctx); | |
619 | gpu->fctx = NULL; | |
620 | goto fail; | |
621 | } | |
7198e6b0 RC |
622 | |
623 | INIT_LIST_HEAD(&gpu->active_list); | |
624 | INIT_WORK(&gpu->retire_work, retire_worker); | |
bd6f82d8 RC |
625 | INIT_WORK(&gpu->recover_work, recover_worker); |
626 | ||
1a370be9 RC |
627 | INIT_LIST_HEAD(&gpu->submit_list); |
628 | ||
bd6f82d8 RC |
629 | setup_timer(&gpu->hangcheck_timer, hangcheck_handler, |
630 | (unsigned long)gpu); | |
7198e6b0 | 631 | |
70c70f09 RC |
632 | spin_lock_init(&gpu->perf_lock); |
633 | ||
7198e6b0 RC |
634 | |
635 | /* Map registers: */ | |
5770fc7a | 636 | gpu->mmio = msm_ioremap(pdev, config->ioname, name); |
7198e6b0 RC |
637 | if (IS_ERR(gpu->mmio)) { |
638 | ret = PTR_ERR(gpu->mmio); | |
639 | goto fail; | |
640 | } | |
641 | ||
642 | /* Get Interrupt: */ | |
5770fc7a | 643 | gpu->irq = platform_get_irq_byname(pdev, config->irqname); |
7198e6b0 RC |
644 | if (gpu->irq < 0) { |
645 | ret = gpu->irq; | |
646 | dev_err(drm->dev, "failed to get irq: %d\n", ret); | |
647 | goto fail; | |
648 | } | |
649 | ||
650 | ret = devm_request_irq(&pdev->dev, gpu->irq, irq_handler, | |
651 | IRQF_TRIGGER_HIGH, gpu->name, gpu); | |
652 | if (ret) { | |
653 | dev_err(drm->dev, "failed to request IRQ%u: %d\n", gpu->irq, ret); | |
654 | goto fail; | |
655 | } | |
656 | ||
98db803f JC |
657 | ret = get_clocks(pdev, gpu); |
658 | if (ret) | |
659 | goto fail; | |
7198e6b0 | 660 | |
720c3bb8 | 661 | gpu->ebi1_clk = msm_clk_get(pdev, "bus"); |
7198e6b0 RC |
662 | DBG("ebi1_clk: %p", gpu->ebi1_clk); |
663 | if (IS_ERR(gpu->ebi1_clk)) | |
664 | gpu->ebi1_clk = NULL; | |
665 | ||
666 | /* Acquire regulators: */ | |
667 | gpu->gpu_reg = devm_regulator_get(&pdev->dev, "vdd"); | |
668 | DBG("gpu_reg: %p", gpu->gpu_reg); | |
669 | if (IS_ERR(gpu->gpu_reg)) | |
670 | gpu->gpu_reg = NULL; | |
671 | ||
672 | gpu->gpu_cx = devm_regulator_get(&pdev->dev, "vddcx"); | |
673 | DBG("gpu_cx: %p", gpu->gpu_cx); | |
674 | if (IS_ERR(gpu->gpu_cx)) | |
675 | gpu->gpu_cx = NULL; | |
676 | ||
1267a4df JC |
677 | gpu->pdev = pdev; |
678 | platform_set_drvdata(pdev, gpu); | |
679 | ||
680 | bs_init(gpu); | |
5e921b19 | 681 | |
1267a4df JC |
682 | gpu->aspace = msm_gpu_create_address_space(gpu, pdev, |
683 | config->va_start, config->va_end); | |
684 | ||
685 | if (gpu->aspace == NULL) | |
871d812a | 686 | dev_info(drm->dev, "%s: no IOMMU, fallback to VRAM carveout!\n", name); |
1267a4df JC |
687 | else if (IS_ERR(gpu->aspace)) { |
688 | ret = PTR_ERR(gpu->aspace); | |
689 | goto fail; | |
7198e6b0 | 690 | } |
a1ad3523 | 691 | |
7198e6b0 | 692 | /* Create ringbuffer: */ |
5770fc7a | 693 | gpu->rb = msm_ringbuffer_new(gpu, config->ringsz); |
7198e6b0 RC |
694 | if (IS_ERR(gpu->rb)) { |
695 | ret = PTR_ERR(gpu->rb); | |
696 | gpu->rb = NULL; | |
697 | dev_err(drm->dev, "could not create ringbuffer: %d\n", ret); | |
698 | goto fail; | |
699 | } | |
700 | ||
7198e6b0 RC |
701 | return 0; |
702 | ||
703 | fail: | |
1267a4df | 704 | platform_set_drvdata(pdev, NULL); |
7198e6b0 RC |
705 | return ret; |
706 | } | |
707 | ||
708 | void msm_gpu_cleanup(struct msm_gpu *gpu) | |
709 | { | |
710 | DBG("%s", gpu->name); | |
711 | ||
712 | WARN_ON(!list_empty(&gpu->active_list)); | |
713 | ||
714 | bs_fini(gpu); | |
715 | ||
716 | if (gpu->rb) { | |
717 | if (gpu->rb_iova) | |
8bdcd949 | 718 | msm_gem_put_iova(gpu->rb->bo, gpu->aspace); |
7198e6b0 RC |
719 | msm_ringbuffer_destroy(gpu->rb); |
720 | } | |
6bd6ae2d RC |
721 | |
722 | if (!IS_ERR_OR_NULL(gpu->aspace)) { | |
1267a4df JC |
723 | gpu->aspace->mmu->funcs->detach(gpu->aspace->mmu, |
724 | NULL, 0); | |
725 | msm_gem_address_space_put(gpu->aspace); | |
726 | } | |
7198e6b0 | 727 | } |