diff options
Diffstat (limited to 'drivers/gpu/nvgpu/common/linux/module.c')
-rw-r--r-- | drivers/gpu/nvgpu/common/linux/module.c | 1277 |
1 files changed, 1277 insertions, 0 deletions
diff --git a/drivers/gpu/nvgpu/common/linux/module.c b/drivers/gpu/nvgpu/common/linux/module.c new file mode 100644 index 00000000..76b9d4db --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/module.c | |||
@@ -0,0 +1,1277 @@ | |||
1 | /* | ||
2 | * GK20A Graphics | ||
3 | * | ||
4 | * Copyright (c) 2011-2017, NVIDIA CORPORATION. All rights reserved. | ||
5 | * | ||
6 | * This program is free software; you can redistribute it and/or modify it | ||
7 | * under the terms and conditions of the GNU General Public License, | ||
8 | * version 2, as published by the Free Software Foundation. | ||
9 | * | ||
10 | * This program is distributed in the hope it will be useful, but WITHOUT | ||
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | ||
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | ||
13 | * more details. | ||
14 | * | ||
15 | * You should have received a copy of the GNU General Public License | ||
16 | * along with this program. If not, see <http://www.gnu.org/licenses/>. | ||
17 | */ | ||
18 | |||
19 | #include <linux/module.h> | ||
20 | #include <linux/of.h> | ||
21 | #include <linux/of_device.h> | ||
22 | #include <linux/of_platform.h> | ||
23 | #include <linux/interrupt.h> | ||
24 | #include <linux/pm_runtime.h> | ||
25 | #include <linux/reset.h> | ||
26 | #include <linux/platform/tegra/common.h> | ||
27 | #include <uapi/linux/nvgpu.h> | ||
28 | #include <dt-bindings/soc/gm20b-fuse.h> | ||
29 | #include <dt-bindings/soc/gp10b-fuse.h> | ||
30 | |||
31 | #include <soc/tegra/fuse.h> | ||
32 | |||
33 | #include <nvgpu/dma.h> | ||
34 | #include <nvgpu/kmem.h> | ||
35 | #include <nvgpu/nvgpu_common.h> | ||
36 | #include <nvgpu/soc.h> | ||
37 | #include <nvgpu/enabled.h> | ||
38 | #include <nvgpu/debug.h> | ||
39 | #include <nvgpu/ctxsw_trace.h> | ||
40 | |||
41 | #include "platform_gk20a.h" | ||
42 | #include "sysfs.h" | ||
43 | #include "vgpu/vgpu.h" | ||
44 | #include "scale.h" | ||
45 | #include "pci.h" | ||
46 | #include "module.h" | ||
47 | #include "intr.h" | ||
48 | #include "cde.h" | ||
49 | #include "ioctl.h" | ||
50 | #ifdef CONFIG_TEGRA_19x_GPU | ||
51 | #include "nvgpu_gpuid_t19x.h" | ||
52 | #ifdef CONFIG_TEGRA_GR_VIRTUALIZATION | ||
53 | #include "vgpu/vgpu_t19x.h" | ||
54 | #endif | ||
55 | #endif | ||
56 | |||
57 | #include "os_linux.h" | ||
58 | #include "cde_gm20b.h" | ||
59 | #include "cde_gp10b.h" | ||
60 | #include "ctxsw_trace.h" | ||
61 | #include "driver_common.h" | ||
62 | |||
63 | #define CLASS_NAME "nvidia-gpu" | ||
64 | /* TODO: Change to e.g. "nvidia-gpu%s" once we have symlinks in place. */ | ||
65 | |||
66 | #define GK20A_WAIT_FOR_IDLE_MS 2000 | ||
67 | |||
68 | #define CREATE_TRACE_POINTS | ||
69 | #include <trace/events/gk20a.h> | ||
70 | |||
71 | void gk20a_busy_noresume(struct gk20a *g) | ||
72 | { | ||
73 | pm_runtime_get_noresume(dev_from_gk20a(g)); | ||
74 | } | ||
75 | |||
76 | int gk20a_busy(struct gk20a *g) | ||
77 | { | ||
78 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
79 | int ret = 0; | ||
80 | struct device *dev; | ||
81 | |||
82 | if (!g) | ||
83 | return -ENODEV; | ||
84 | |||
85 | atomic_inc(&g->usage_count.atomic_var); | ||
86 | |||
87 | down_read(&l->busy_lock); | ||
88 | |||
89 | if (!gk20a_can_busy(g)) { | ||
90 | ret = -ENODEV; | ||
91 | atomic_dec(&g->usage_count.atomic_var); | ||
92 | goto fail; | ||
93 | } | ||
94 | |||
95 | dev = dev_from_gk20a(g); | ||
96 | |||
97 | if (pm_runtime_enabled(dev)) { | ||
98 | /* Increment usage count and attempt to resume device */ | ||
99 | ret = pm_runtime_get_sync(dev); | ||
100 | if (ret < 0) { | ||
101 | /* Mark suspended so runtime pm will retry later */ | ||
102 | pm_runtime_set_suspended(dev); | ||
103 | pm_runtime_put_noidle(dev); | ||
104 | atomic_dec(&g->usage_count.atomic_var); | ||
105 | goto fail; | ||
106 | } | ||
107 | } else { | ||
108 | nvgpu_mutex_acquire(&g->poweron_lock); | ||
109 | if (!g->power_on) { | ||
110 | ret = gk20a_gpu_is_virtual(dev) ? | ||
111 | vgpu_pm_finalize_poweron(dev) | ||
112 | : gk20a_pm_finalize_poweron(dev); | ||
113 | if (ret) { | ||
114 | atomic_dec(&g->usage_count.atomic_var); | ||
115 | nvgpu_mutex_release(&g->poweron_lock); | ||
116 | goto fail; | ||
117 | } | ||
118 | } | ||
119 | nvgpu_mutex_release(&g->poweron_lock); | ||
120 | } | ||
121 | |||
122 | fail: | ||
123 | up_read(&l->busy_lock); | ||
124 | |||
125 | return ret < 0 ? ret : 0; | ||
126 | } | ||
127 | |||
128 | void gk20a_idle_nosuspend(struct gk20a *g) | ||
129 | { | ||
130 | pm_runtime_put_noidle(dev_from_gk20a(g)); | ||
131 | } | ||
132 | |||
133 | void gk20a_idle(struct gk20a *g) | ||
134 | { | ||
135 | struct device *dev; | ||
136 | |||
137 | atomic_dec(&g->usage_count.atomic_var); | ||
138 | |||
139 | dev = dev_from_gk20a(g); | ||
140 | |||
141 | if (!(dev && gk20a_can_busy(g))) | ||
142 | return; | ||
143 | |||
144 | if (pm_runtime_enabled(dev)) { | ||
145 | pm_runtime_mark_last_busy(dev); | ||
146 | pm_runtime_put_sync_autosuspend(dev); | ||
147 | } | ||
148 | } | ||
149 | |||
150 | /* | ||
151 | * Undoes gk20a_lockout_registers(). | ||
152 | */ | ||
153 | static int gk20a_restore_registers(struct gk20a *g) | ||
154 | { | ||
155 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
156 | |||
157 | l->regs = l->regs_saved; | ||
158 | l->bar1 = l->bar1_saved; | ||
159 | |||
160 | #ifdef CONFIG_TEGRA_19x_GPU | ||
161 | t19x_restore_registers(g); | ||
162 | #endif | ||
163 | |||
164 | return 0; | ||
165 | } | ||
166 | |||
167 | static int nvgpu_init_os_linux_ops(struct nvgpu_os_linux *l) | ||
168 | { | ||
169 | struct gk20a *g = &l->g; | ||
170 | u32 ver = g->params.gpu_arch + g->params.gpu_impl; | ||
171 | |||
172 | switch (ver) { | ||
173 | case GK20A_GPUID_GM20B: | ||
174 | case GK20A_GPUID_GM20B_B: | ||
175 | l->ops.cde = gm20b_cde_ops.cde; | ||
176 | break; | ||
177 | case NVGPU_GPUID_GP10B: | ||
178 | l->ops.cde = gp10b_cde_ops.cde; | ||
179 | break; | ||
180 | default: | ||
181 | /* CDE is optional, so today ignoring unknown chip is fine */ | ||
182 | break; | ||
183 | } | ||
184 | |||
185 | return 0; | ||
186 | } | ||
187 | |||
188 | int gk20a_pm_finalize_poweron(struct device *dev) | ||
189 | { | ||
190 | struct gk20a *g = get_gk20a(dev); | ||
191 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
192 | struct gk20a_platform *platform = gk20a_get_platform(dev); | ||
193 | int err, nice_value; | ||
194 | |||
195 | gk20a_dbg_fn(""); | ||
196 | |||
197 | if (g->power_on) | ||
198 | return 0; | ||
199 | |||
200 | trace_gk20a_finalize_poweron(dev_name(dev)); | ||
201 | |||
202 | /* Increment platform power refcount */ | ||
203 | if (platform->busy) { | ||
204 | err = platform->busy(dev); | ||
205 | if (err < 0) { | ||
206 | nvgpu_err(g, "failed to poweron platform dependency"); | ||
207 | return err; | ||
208 | } | ||
209 | } | ||
210 | |||
211 | err = gk20a_restore_registers(g); | ||
212 | if (err) | ||
213 | return err; | ||
214 | |||
215 | nice_value = task_nice(current); | ||
216 | set_user_nice(current, -20); | ||
217 | |||
218 | /* Enable interrupt workqueue */ | ||
219 | if (!l->nonstall_work_queue) { | ||
220 | l->nonstall_work_queue = alloc_workqueue("%s", | ||
221 | WQ_HIGHPRI, 1, "mc_nonstall"); | ||
222 | INIT_WORK(&l->nonstall_fn_work, nvgpu_intr_nonstall_cb); | ||
223 | } | ||
224 | |||
225 | err = gk20a_finalize_poweron(g); | ||
226 | set_user_nice(current, nice_value); | ||
227 | if (err) | ||
228 | goto done; | ||
229 | |||
230 | trace_gk20a_finalize_poweron_done(dev_name(dev)); | ||
231 | |||
232 | err = nvgpu_init_os_linux_ops(l); | ||
233 | if (err) | ||
234 | goto done; | ||
235 | |||
236 | enable_irq(g->irq_stall); | ||
237 | if (g->irq_stall != g->irq_nonstall) | ||
238 | enable_irq(g->irq_nonstall); | ||
239 | g->irqs_enabled = 1; | ||
240 | |||
241 | gk20a_scale_resume(dev_from_gk20a(g)); | ||
242 | |||
243 | if (platform->has_cde) | ||
244 | gk20a_init_cde_support(l); | ||
245 | |||
246 | err = gk20a_sched_ctrl_init(g); | ||
247 | if (err) { | ||
248 | nvgpu_err(g, "failed to init sched control"); | ||
249 | return err; | ||
250 | } | ||
251 | |||
252 | done: | ||
253 | if (err) | ||
254 | g->power_on = false; | ||
255 | |||
256 | return err; | ||
257 | } | ||
258 | |||
259 | /* | ||
260 | * Locks out the driver from accessing GPU registers. This prevents access to | ||
261 | * thse registers after the GPU has been clock or power gated. This should help | ||
262 | * find annoying bugs where register reads and writes are silently dropped | ||
263 | * after the GPU has been turned off. On older chips these reads and writes can | ||
264 | * also lock the entire CPU up. | ||
265 | */ | ||
266 | static int gk20a_lockout_registers(struct gk20a *g) | ||
267 | { | ||
268 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
269 | |||
270 | l->regs = NULL; | ||
271 | l->bar1 = NULL; | ||
272 | |||
273 | #ifdef CONFIG_TEGRA_19x_GPU | ||
274 | t19x_lockout_registers(g); | ||
275 | #endif | ||
276 | |||
277 | return 0; | ||
278 | } | ||
279 | |||
280 | static int gk20a_pm_prepare_poweroff(struct device *dev) | ||
281 | { | ||
282 | struct gk20a *g = get_gk20a(dev); | ||
283 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
284 | int ret = 0; | ||
285 | struct gk20a_platform *platform = gk20a_get_platform(dev); | ||
286 | bool irqs_enabled; | ||
287 | |||
288 | gk20a_dbg_fn(""); | ||
289 | |||
290 | nvgpu_mutex_acquire(&g->poweroff_lock); | ||
291 | |||
292 | if (!g->power_on) | ||
293 | goto done; | ||
294 | |||
295 | /* disable IRQs and wait for completion */ | ||
296 | irqs_enabled = g->irqs_enabled; | ||
297 | if (irqs_enabled) { | ||
298 | disable_irq(g->irq_stall); | ||
299 | if (g->irq_stall != g->irq_nonstall) | ||
300 | disable_irq(g->irq_nonstall); | ||
301 | g->irqs_enabled = 0; | ||
302 | } | ||
303 | |||
304 | gk20a_scale_suspend(dev); | ||
305 | |||
306 | gk20a_cde_suspend(l); | ||
307 | |||
308 | ret = gk20a_prepare_poweroff(g); | ||
309 | if (ret) | ||
310 | goto error; | ||
311 | |||
312 | /* Decrement platform power refcount */ | ||
313 | if (platform->idle) | ||
314 | platform->idle(dev); | ||
315 | |||
316 | /* Stop CPU from accessing the GPU registers. */ | ||
317 | gk20a_lockout_registers(g); | ||
318 | |||
319 | nvgpu_mutex_release(&g->poweroff_lock); | ||
320 | return 0; | ||
321 | |||
322 | error: | ||
323 | /* re-enabled IRQs if previously enabled */ | ||
324 | if (irqs_enabled) { | ||
325 | enable_irq(g->irq_stall); | ||
326 | if (g->irq_stall != g->irq_nonstall) | ||
327 | enable_irq(g->irq_nonstall); | ||
328 | g->irqs_enabled = 1; | ||
329 | } | ||
330 | |||
331 | gk20a_scale_resume(dev); | ||
332 | done: | ||
333 | nvgpu_mutex_release(&g->poweroff_lock); | ||
334 | |||
335 | return ret; | ||
336 | } | ||
337 | |||
338 | static struct of_device_id tegra_gk20a_of_match[] = { | ||
339 | #ifdef CONFIG_TEGRA_GK20A | ||
340 | { .compatible = "nvidia,tegra210-gm20b", | ||
341 | .data = &gm20b_tegra_platform }, | ||
342 | #ifdef CONFIG_ARCH_TEGRA_18x_SOC | ||
343 | { .compatible = "nvidia,tegra186-gp10b", | ||
344 | .data = &gp10b_tegra_platform }, | ||
345 | #endif | ||
346 | #ifdef CONFIG_TEGRA_19x_GPU | ||
347 | { .compatible = TEGRA_19x_GPU_COMPAT_TEGRA, | ||
348 | .data = &t19x_gpu_tegra_platform }, | ||
349 | #ifdef CONFIG_TEGRA_GR_VIRTUALIZATION | ||
350 | { .compatible = TEGRA_19x_VGPU_COMPAT_TEGRA, | ||
351 | .data = &t19x_vgpu_tegra_platform }, | ||
352 | #endif | ||
353 | #endif | ||
354 | #ifdef CONFIG_TEGRA_GR_VIRTUALIZATION | ||
355 | { .compatible = "nvidia,tegra124-gk20a-vgpu", | ||
356 | .data = &vgpu_tegra_platform }, | ||
357 | #endif | ||
358 | #endif | ||
359 | |||
360 | { }, | ||
361 | }; | ||
362 | |||
363 | #ifdef CONFIG_PM | ||
364 | /** | ||
365 | * __gk20a_do_idle() - force the GPU to idle and railgate | ||
366 | * | ||
367 | * In success, this call MUST be balanced by caller with __gk20a_do_unidle() | ||
368 | * | ||
369 | * Acquires two locks : &l->busy_lock and &platform->railgate_lock | ||
370 | * In success, we hold these locks and return | ||
371 | * In failure, we release these locks and return | ||
372 | */ | ||
373 | int __gk20a_do_idle(struct gk20a *g, bool force_reset) | ||
374 | { | ||
375 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
376 | struct device *dev = dev_from_gk20a(g); | ||
377 | struct gk20a_platform *platform = dev_get_drvdata(dev); | ||
378 | struct nvgpu_timeout timeout; | ||
379 | int ref_cnt; | ||
380 | int target_ref_cnt = 0; | ||
381 | bool is_railgated; | ||
382 | int err = 0; | ||
383 | |||
384 | /* | ||
385 | * Hold back deterministic submits and changes to deterministic | ||
386 | * channels - this must be outside the power busy locks. | ||
387 | */ | ||
388 | gk20a_channel_deterministic_idle(g); | ||
389 | |||
390 | /* acquire busy lock to block other busy() calls */ | ||
391 | down_write(&l->busy_lock); | ||
392 | |||
393 | /* acquire railgate lock to prevent unrailgate in midst of do_idle() */ | ||
394 | nvgpu_mutex_acquire(&platform->railgate_lock); | ||
395 | |||
396 | /* check if it is already railgated ? */ | ||
397 | if (platform->is_railgated(dev)) | ||
398 | return 0; | ||
399 | |||
400 | /* | ||
401 | * release railgate_lock, prevent suspend by incrementing usage counter, | ||
402 | * re-acquire railgate_lock | ||
403 | */ | ||
404 | nvgpu_mutex_release(&platform->railgate_lock); | ||
405 | pm_runtime_get_sync(dev); | ||
406 | |||
407 | /* | ||
408 | * One refcount taken in this API | ||
409 | * If User disables rail gating, we take one more | ||
410 | * extra refcount | ||
411 | */ | ||
412 | if (g->user_railgate_disabled) | ||
413 | target_ref_cnt = 2; | ||
414 | else | ||
415 | target_ref_cnt = 1; | ||
416 | nvgpu_mutex_acquire(&platform->railgate_lock); | ||
417 | |||
418 | nvgpu_timeout_init(g, &timeout, GK20A_WAIT_FOR_IDLE_MS, | ||
419 | NVGPU_TIMER_CPU_TIMER); | ||
420 | |||
421 | /* check and wait until GPU is idle (with a timeout) */ | ||
422 | do { | ||
423 | nvgpu_usleep_range(1000, 1100); | ||
424 | ref_cnt = atomic_read(&dev->power.usage_count); | ||
425 | } while (ref_cnt != target_ref_cnt && !nvgpu_timeout_expired(&timeout)); | ||
426 | |||
427 | if (ref_cnt != target_ref_cnt) { | ||
428 | nvgpu_err(g, "failed to idle - refcount %d != target_ref_cnt", | ||
429 | ref_cnt); | ||
430 | goto fail_drop_usage_count; | ||
431 | } | ||
432 | |||
433 | /* check if global force_reset flag is set */ | ||
434 | force_reset |= platform->force_reset_in_do_idle; | ||
435 | |||
436 | nvgpu_timeout_init(g, &timeout, GK20A_WAIT_FOR_IDLE_MS, | ||
437 | NVGPU_TIMER_CPU_TIMER); | ||
438 | |||
439 | if (g->can_railgate && !force_reset) { | ||
440 | /* | ||
441 | * Case 1 : GPU railgate is supported | ||
442 | * | ||
443 | * if GPU is now idle, we will have only one ref count, | ||
444 | * drop this ref which will rail gate the GPU | ||
445 | */ | ||
446 | pm_runtime_put_sync(dev); | ||
447 | |||
448 | /* add sufficient delay to allow GPU to rail gate */ | ||
449 | nvgpu_msleep(g->railgate_delay); | ||
450 | |||
451 | /* check in loop if GPU is railgated or not */ | ||
452 | do { | ||
453 | nvgpu_usleep_range(1000, 1100); | ||
454 | is_railgated = platform->is_railgated(dev); | ||
455 | } while (!is_railgated && !nvgpu_timeout_expired(&timeout)); | ||
456 | |||
457 | if (is_railgated) { | ||
458 | return 0; | ||
459 | } else { | ||
460 | nvgpu_err(g, "failed to idle in timeout"); | ||
461 | goto fail_timeout; | ||
462 | } | ||
463 | } else { | ||
464 | /* | ||
465 | * Case 2 : GPU railgate is not supported or we explicitly | ||
466 | * do not want to depend on runtime PM | ||
467 | * | ||
468 | * if GPU is now idle, call prepare_poweroff() to save the | ||
469 | * state and then do explicit railgate | ||
470 | * | ||
471 | * __gk20a_do_unidle() needs to unrailgate, call | ||
472 | * finalize_poweron(), and then call pm_runtime_put_sync() | ||
473 | * to balance the GPU usage counter | ||
474 | */ | ||
475 | |||
476 | /* Save the GPU state */ | ||
477 | err = gk20a_pm_prepare_poweroff(dev); | ||
478 | if (err) | ||
479 | goto fail_drop_usage_count; | ||
480 | |||
481 | /* railgate GPU */ | ||
482 | platform->railgate(dev); | ||
483 | |||
484 | nvgpu_udelay(10); | ||
485 | |||
486 | g->forced_reset = true; | ||
487 | return 0; | ||
488 | } | ||
489 | |||
490 | fail_drop_usage_count: | ||
491 | pm_runtime_put_noidle(dev); | ||
492 | fail_timeout: | ||
493 | nvgpu_mutex_release(&platform->railgate_lock); | ||
494 | up_write(&l->busy_lock); | ||
495 | gk20a_channel_deterministic_unidle(g); | ||
496 | return -EBUSY; | ||
497 | } | ||
498 | |||
499 | /** | ||
500 | * gk20a_do_idle() - wrap up for __gk20a_do_idle() to be called | ||
501 | * from outside of GPU driver | ||
502 | * | ||
503 | * In success, this call MUST be balanced by caller with gk20a_do_unidle() | ||
504 | */ | ||
505 | static int gk20a_do_idle(void *_g) | ||
506 | { | ||
507 | struct gk20a *g = (struct gk20a *)_g; | ||
508 | |||
509 | return __gk20a_do_idle(g, true); | ||
510 | } | ||
511 | |||
512 | /** | ||
513 | * __gk20a_do_unidle() - unblock all the tasks blocked by __gk20a_do_idle() | ||
514 | */ | ||
515 | int __gk20a_do_unidle(struct gk20a *g) | ||
516 | { | ||
517 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
518 | struct device *dev = dev_from_gk20a(g); | ||
519 | struct gk20a_platform *platform = dev_get_drvdata(dev); | ||
520 | int err; | ||
521 | |||
522 | if (g->forced_reset) { | ||
523 | /* | ||
524 | * If we did a forced-reset/railgate | ||
525 | * then unrailgate the GPU here first | ||
526 | */ | ||
527 | platform->unrailgate(dev); | ||
528 | |||
529 | /* restore the GPU state */ | ||
530 | err = gk20a_pm_finalize_poweron(dev); | ||
531 | if (err) | ||
532 | return err; | ||
533 | |||
534 | /* balance GPU usage counter */ | ||
535 | pm_runtime_put_sync(dev); | ||
536 | |||
537 | g->forced_reset = false; | ||
538 | } | ||
539 | |||
540 | /* release the lock and open up all other busy() calls */ | ||
541 | nvgpu_mutex_release(&platform->railgate_lock); | ||
542 | up_write(&l->busy_lock); | ||
543 | |||
544 | gk20a_channel_deterministic_unidle(g); | ||
545 | |||
546 | return 0; | ||
547 | } | ||
548 | |||
549 | /** | ||
550 | * gk20a_do_unidle() - wrap up for __gk20a_do_unidle() | ||
551 | */ | ||
552 | static int gk20a_do_unidle(void *_g) | ||
553 | { | ||
554 | struct gk20a *g = (struct gk20a *)_g; | ||
555 | |||
556 | return __gk20a_do_unidle(g); | ||
557 | } | ||
558 | #endif | ||
559 | |||
560 | static void __iomem *gk20a_ioremap_resource(struct platform_device *dev, int i, | ||
561 | struct resource **out) | ||
562 | { | ||
563 | struct resource *r = platform_get_resource(dev, IORESOURCE_MEM, i); | ||
564 | |||
565 | if (!r) | ||
566 | return NULL; | ||
567 | if (out) | ||
568 | *out = r; | ||
569 | return devm_ioremap_resource(&dev->dev, r); | ||
570 | } | ||
571 | |||
572 | static irqreturn_t gk20a_intr_isr_stall(int irq, void *dev_id) | ||
573 | { | ||
574 | struct gk20a *g = dev_id; | ||
575 | |||
576 | return nvgpu_intr_stall(g); | ||
577 | } | ||
578 | |||
579 | static irqreturn_t gk20a_intr_isr_nonstall(int irq, void *dev_id) | ||
580 | { | ||
581 | struct gk20a *g = dev_id; | ||
582 | |||
583 | return nvgpu_intr_nonstall(g); | ||
584 | } | ||
585 | |||
586 | static irqreturn_t gk20a_intr_thread_stall(int irq, void *dev_id) | ||
587 | { | ||
588 | struct gk20a *g = dev_id; | ||
589 | |||
590 | return nvgpu_intr_thread_stall(g); | ||
591 | } | ||
592 | |||
593 | void gk20a_remove_support(struct gk20a *g) | ||
594 | { | ||
595 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
596 | |||
597 | tegra_unregister_idle_unidle(gk20a_do_idle); | ||
598 | |||
599 | nvgpu_kfree(g, g->dbg_regops_tmp_buf); | ||
600 | |||
601 | if (g->pmu.remove_support) | ||
602 | g->pmu.remove_support(&g->pmu); | ||
603 | |||
604 | if (g->gr.remove_support) | ||
605 | g->gr.remove_support(&g->gr); | ||
606 | |||
607 | if (g->mm.remove_ce_support) | ||
608 | g->mm.remove_ce_support(&g->mm); | ||
609 | |||
610 | if (g->fifo.remove_support) | ||
611 | g->fifo.remove_support(&g->fifo); | ||
612 | |||
613 | if (g->mm.remove_support) | ||
614 | g->mm.remove_support(&g->mm); | ||
615 | |||
616 | if (g->sim.remove_support) | ||
617 | g->sim.remove_support(&g->sim); | ||
618 | |||
619 | /* free mappings to registers, etc */ | ||
620 | |||
621 | if (l->regs) { | ||
622 | iounmap(l->regs); | ||
623 | l->regs = NULL; | ||
624 | } | ||
625 | if (l->bar1) { | ||
626 | iounmap(l->bar1); | ||
627 | l->bar1 = NULL; | ||
628 | } | ||
629 | |||
630 | #ifdef CONFIG_TEGRA_19x_GPU | ||
631 | t19x_remove_support(g); | ||
632 | #endif | ||
633 | } | ||
634 | |||
635 | static int gk20a_init_support(struct platform_device *dev) | ||
636 | { | ||
637 | int err = 0; | ||
638 | struct gk20a *g = get_gk20a(&dev->dev); | ||
639 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
640 | |||
641 | tegra_register_idle_unidle(gk20a_do_idle, gk20a_do_unidle, g); | ||
642 | |||
643 | l->regs = gk20a_ioremap_resource(dev, GK20A_BAR0_IORESOURCE_MEM, | ||
644 | &l->reg_mem); | ||
645 | if (IS_ERR(l->regs)) { | ||
646 | nvgpu_err(g, "failed to remap gk20a registers"); | ||
647 | err = PTR_ERR(l->regs); | ||
648 | goto fail; | ||
649 | } | ||
650 | |||
651 | l->bar1 = gk20a_ioremap_resource(dev, GK20A_BAR1_IORESOURCE_MEM, | ||
652 | &l->bar1_mem); | ||
653 | if (IS_ERR(l->bar1)) { | ||
654 | nvgpu_err(g, "failed to remap gk20a bar1"); | ||
655 | err = PTR_ERR(l->bar1); | ||
656 | goto fail; | ||
657 | } | ||
658 | |||
659 | if (nvgpu_platform_is_simulation(g)) { | ||
660 | g->sim.g = g; | ||
661 | g->sim.regs = gk20a_ioremap_resource(dev, | ||
662 | GK20A_SIM_IORESOURCE_MEM, | ||
663 | &g->sim.reg_mem); | ||
664 | if (IS_ERR(g->sim.regs)) { | ||
665 | nvgpu_err(g, "failed to remap gk20a sim regs"); | ||
666 | err = PTR_ERR(g->sim.regs); | ||
667 | goto fail; | ||
668 | } | ||
669 | |||
670 | err = gk20a_init_sim_support(g); | ||
671 | if (err) | ||
672 | goto fail; | ||
673 | } | ||
674 | |||
675 | #ifdef CONFIG_TEGRA_19x_GPU | ||
676 | t19x_init_support(g); | ||
677 | #endif | ||
678 | |||
679 | return 0; | ||
680 | |||
681 | fail: | ||
682 | return err; | ||
683 | } | ||
684 | |||
685 | static int gk20a_pm_railgate(struct device *dev) | ||
686 | { | ||
687 | struct gk20a_platform *platform = dev_get_drvdata(dev); | ||
688 | int ret = 0; | ||
689 | struct gk20a *g = get_gk20a(dev); | ||
690 | #ifdef CONFIG_DEBUG_FS | ||
691 | g->pstats.last_rail_gate_start = jiffies; | ||
692 | |||
693 | if (g->pstats.railgating_cycle_count >= 1) | ||
694 | g->pstats.total_rail_ungate_time_ms = | ||
695 | g->pstats.total_rail_ungate_time_ms + | ||
696 | jiffies_to_msecs(g->pstats.last_rail_gate_start - | ||
697 | g->pstats.last_rail_ungate_complete); | ||
698 | #endif | ||
699 | |||
700 | if (platform->railgate) | ||
701 | ret = platform->railgate(dev); | ||
702 | if (ret) { | ||
703 | nvgpu_err(g, "failed to railgate platform, err=%d", ret); | ||
704 | return ret; | ||
705 | } | ||
706 | |||
707 | #ifdef CONFIG_DEBUG_FS | ||
708 | g->pstats.last_rail_gate_complete = jiffies; | ||
709 | #endif | ||
710 | ret = tegra_fuse_clock_disable(); | ||
711 | if (ret) | ||
712 | nvgpu_err(g, "failed to disable tegra fuse clock, err=%d", ret); | ||
713 | |||
714 | return ret; | ||
715 | } | ||
716 | |||
717 | static int gk20a_pm_unrailgate(struct device *dev) | ||
718 | { | ||
719 | struct gk20a_platform *platform = dev_get_drvdata(dev); | ||
720 | int ret = 0; | ||
721 | struct gk20a *g = get_gk20a(dev); | ||
722 | |||
723 | ret = tegra_fuse_clock_enable(); | ||
724 | if (ret) { | ||
725 | nvgpu_err(g, "failed to enable tegra fuse clock, err=%d", ret); | ||
726 | return ret; | ||
727 | } | ||
728 | #ifdef CONFIG_DEBUG_FS | ||
729 | g->pstats.last_rail_ungate_start = jiffies; | ||
730 | if (g->pstats.railgating_cycle_count >= 1) | ||
731 | g->pstats.total_rail_gate_time_ms = | ||
732 | g->pstats.total_rail_gate_time_ms + | ||
733 | jiffies_to_msecs(g->pstats.last_rail_ungate_start - | ||
734 | g->pstats.last_rail_gate_complete); | ||
735 | |||
736 | g->pstats.railgating_cycle_count++; | ||
737 | #endif | ||
738 | |||
739 | trace_gk20a_pm_unrailgate(dev_name(dev)); | ||
740 | |||
741 | if (platform->unrailgate) { | ||
742 | nvgpu_mutex_acquire(&platform->railgate_lock); | ||
743 | ret = platform->unrailgate(dev); | ||
744 | nvgpu_mutex_release(&platform->railgate_lock); | ||
745 | } | ||
746 | |||
747 | #ifdef CONFIG_DEBUG_FS | ||
748 | g->pstats.last_rail_ungate_complete = jiffies; | ||
749 | #endif | ||
750 | |||
751 | return ret; | ||
752 | } | ||
753 | |||
754 | /* | ||
755 | * Remove association of the driver with OS interrupt handler | ||
756 | */ | ||
757 | void nvgpu_free_irq(struct gk20a *g) | ||
758 | { | ||
759 | struct device *dev = dev_from_gk20a(g); | ||
760 | |||
761 | devm_free_irq(dev, g->irq_stall, g); | ||
762 | if (g->irq_stall != g->irq_nonstall) | ||
763 | devm_free_irq(dev, g->irq_nonstall, g); | ||
764 | } | ||
765 | |||
766 | /* | ||
767 | * Idle the GPU in preparation of shutdown/remove. | ||
768 | * gk20a_driver_start_unload() does not idle the GPU, but instead changes the SW | ||
769 | * state to prevent further activity on the driver SW side. | ||
770 | * On driver removal quiesce() should be called after start_unload() | ||
771 | */ | ||
772 | int nvgpu_quiesce(struct gk20a *g) | ||
773 | { | ||
774 | int err; | ||
775 | struct device *dev = dev_from_gk20a(g); | ||
776 | |||
777 | if (g->power_on) { | ||
778 | err = gk20a_wait_for_idle(g); | ||
779 | if (err) { | ||
780 | nvgpu_err(g, "failed to idle GPU, err=%d", err); | ||
781 | return err; | ||
782 | } | ||
783 | |||
784 | err = gk20a_fifo_disable_all_engine_activity(g, true); | ||
785 | if (err) { | ||
786 | nvgpu_err(g, | ||
787 | "failed to disable engine activity, err=%d", | ||
788 | err); | ||
789 | return err; | ||
790 | } | ||
791 | |||
792 | err = gk20a_fifo_wait_engine_idle(g); | ||
793 | if (err) { | ||
794 | nvgpu_err(g, "failed to idle engines, err=%d", | ||
795 | err); | ||
796 | return err; | ||
797 | } | ||
798 | } | ||
799 | |||
800 | if (gk20a_gpu_is_virtual(dev)) | ||
801 | err = vgpu_pm_prepare_poweroff(dev); | ||
802 | else | ||
803 | err = gk20a_pm_prepare_poweroff(dev); | ||
804 | |||
805 | if (err) | ||
806 | nvgpu_err(g, "failed to prepare for poweroff, err=%d", | ||
807 | err); | ||
808 | |||
809 | return err; | ||
810 | } | ||
811 | |||
812 | static void gk20a_pm_shutdown(struct platform_device *pdev) | ||
813 | { | ||
814 | struct gk20a_platform *platform = platform_get_drvdata(pdev); | ||
815 | struct gk20a *g = platform->g; | ||
816 | int err; | ||
817 | |||
818 | nvgpu_info(g, "shutting down"); | ||
819 | |||
820 | /* vgpu has nothing to clean up currently */ | ||
821 | if (gk20a_gpu_is_virtual(&pdev->dev)) | ||
822 | return; | ||
823 | |||
824 | if (!g->power_on) | ||
825 | goto finish; | ||
826 | |||
827 | gk20a_driver_start_unload(g); | ||
828 | |||
829 | /* If GPU is already railgated, | ||
830 | * just prevent more requests, and return */ | ||
831 | if (platform->is_railgated && platform->is_railgated(&pdev->dev)) { | ||
832 | __pm_runtime_disable(&pdev->dev, false); | ||
833 | nvgpu_info(g, "already railgated, shut down complete"); | ||
834 | return; | ||
835 | } | ||
836 | |||
837 | /* Prevent more requests by disabling Runtime PM */ | ||
838 | __pm_runtime_disable(&pdev->dev, false); | ||
839 | |||
840 | err = nvgpu_quiesce(g); | ||
841 | if (err) | ||
842 | goto finish; | ||
843 | |||
844 | err = gk20a_pm_railgate(&pdev->dev); | ||
845 | if (err) | ||
846 | nvgpu_err(g, "failed to railgate, err=%d", err); | ||
847 | |||
848 | finish: | ||
849 | nvgpu_info(g, "shut down complete"); | ||
850 | } | ||
851 | |||
852 | #ifdef CONFIG_PM | ||
853 | static int gk20a_pm_runtime_resume(struct device *dev) | ||
854 | { | ||
855 | int err = 0; | ||
856 | |||
857 | err = gk20a_pm_unrailgate(dev); | ||
858 | if (err) | ||
859 | goto fail; | ||
860 | |||
861 | if (gk20a_gpu_is_virtual(dev)) | ||
862 | err = vgpu_pm_finalize_poweron(dev); | ||
863 | else | ||
864 | err = gk20a_pm_finalize_poweron(dev); | ||
865 | if (err) | ||
866 | goto fail_poweron; | ||
867 | |||
868 | return 0; | ||
869 | |||
870 | fail_poweron: | ||
871 | gk20a_pm_railgate(dev); | ||
872 | fail: | ||
873 | return err; | ||
874 | } | ||
875 | |||
876 | static int gk20a_pm_runtime_suspend(struct device *dev) | ||
877 | { | ||
878 | int err = 0; | ||
879 | |||
880 | if (gk20a_gpu_is_virtual(dev)) | ||
881 | err = vgpu_pm_prepare_poweroff(dev); | ||
882 | else | ||
883 | err = gk20a_pm_prepare_poweroff(dev); | ||
884 | if (err) | ||
885 | goto fail; | ||
886 | |||
887 | err = gk20a_pm_railgate(dev); | ||
888 | if (err) | ||
889 | goto fail_railgate; | ||
890 | |||
891 | return 0; | ||
892 | |||
893 | fail_railgate: | ||
894 | gk20a_pm_finalize_poweron(dev); | ||
895 | fail: | ||
896 | pm_runtime_mark_last_busy(dev); | ||
897 | return err; | ||
898 | } | ||
899 | |||
900 | static int gk20a_pm_suspend(struct device *dev) | ||
901 | { | ||
902 | struct gk20a_platform *platform = dev_get_drvdata(dev); | ||
903 | struct gk20a *g = get_gk20a(dev); | ||
904 | int ret = 0; | ||
905 | int idle_usage_count = g->user_railgate_disabled ? 1 : 0; | ||
906 | |||
907 | if (!g->power_on) | ||
908 | return 0; | ||
909 | |||
910 | if (nvgpu_atomic_read(&g->usage_count) > idle_usage_count) | ||
911 | return -EBUSY; | ||
912 | |||
913 | ret = gk20a_pm_runtime_suspend(dev); | ||
914 | if (ret) | ||
915 | return ret; | ||
916 | |||
917 | if (platform->suspend) | ||
918 | platform->suspend(dev); | ||
919 | |||
920 | g->suspended = true; | ||
921 | |||
922 | return 0; | ||
923 | } | ||
924 | |||
925 | static int gk20a_pm_resume(struct device *dev) | ||
926 | { | ||
927 | struct gk20a *g = get_gk20a(dev); | ||
928 | int ret = 0; | ||
929 | |||
930 | if (!g->suspended) | ||
931 | return 0; | ||
932 | |||
933 | ret = gk20a_pm_runtime_resume(dev); | ||
934 | |||
935 | g->suspended = false; | ||
936 | |||
937 | return ret; | ||
938 | } | ||
939 | |||
940 | static const struct dev_pm_ops gk20a_pm_ops = { | ||
941 | .runtime_resume = gk20a_pm_runtime_resume, | ||
942 | .runtime_suspend = gk20a_pm_runtime_suspend, | ||
943 | .resume = gk20a_pm_resume, | ||
944 | .suspend = gk20a_pm_suspend, | ||
945 | }; | ||
946 | #endif | ||
947 | |||
948 | static int gk20a_pm_init(struct device *dev) | ||
949 | { | ||
950 | struct gk20a *g = get_gk20a(dev); | ||
951 | int err = 0; | ||
952 | |||
953 | gk20a_dbg_fn(""); | ||
954 | |||
955 | /* Initialise pm runtime */ | ||
956 | if (g->railgate_delay) { | ||
957 | pm_runtime_set_autosuspend_delay(dev, | ||
958 | g->railgate_delay); | ||
959 | pm_runtime_use_autosuspend(dev); | ||
960 | } | ||
961 | |||
962 | if (g->can_railgate) { | ||
963 | pm_runtime_enable(dev); | ||
964 | if (!pm_runtime_enabled(dev)) | ||
965 | gk20a_pm_unrailgate(dev); | ||
966 | else | ||
967 | gk20a_pm_railgate(dev); | ||
968 | } else { | ||
969 | __pm_runtime_disable(dev, false); | ||
970 | gk20a_pm_unrailgate(dev); | ||
971 | } | ||
972 | |||
973 | return err; | ||
974 | } | ||
975 | |||
976 | /* | ||
977 | * Start the process for unloading the driver. Set NVGPU_DRIVER_IS_DYING. | ||
978 | */ | ||
979 | void gk20a_driver_start_unload(struct gk20a *g) | ||
980 | { | ||
981 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
982 | |||
983 | gk20a_dbg(gpu_dbg_shutdown, "Driver is now going down!\n"); | ||
984 | |||
985 | down_write(&l->busy_lock); | ||
986 | __nvgpu_set_enabled(g, NVGPU_DRIVER_IS_DYING, true); | ||
987 | /* GR SW ready needs to be invalidated at this time with the busy lock | ||
988 | * held to prevent a racing condition on the gr/mm code */ | ||
989 | g->gr.sw_ready = false; | ||
990 | up_write(&l->busy_lock); | ||
991 | |||
992 | if (g->is_virtual) | ||
993 | return; | ||
994 | |||
995 | gk20a_wait_for_idle(g); | ||
996 | |||
997 | nvgpu_wait_for_deferred_interrupts(g); | ||
998 | gk20a_channel_cancel_pending_sema_waits(g); | ||
999 | |||
1000 | if (l->nonstall_work_queue) { | ||
1001 | cancel_work_sync(&l->nonstall_fn_work); | ||
1002 | destroy_workqueue(l->nonstall_work_queue); | ||
1003 | l->nonstall_work_queue = NULL; | ||
1004 | } | ||
1005 | } | ||
1006 | |||
1007 | static inline void set_gk20a(struct platform_device *pdev, struct gk20a *gk20a) | ||
1008 | { | ||
1009 | gk20a_get_platform(&pdev->dev)->g = gk20a; | ||
1010 | } | ||
1011 | |||
1012 | static int nvgpu_read_fuse_overrides(struct gk20a *g) | ||
1013 | { | ||
1014 | struct device_node *np = dev_from_gk20a(g)->of_node; | ||
1015 | u32 *fuses; | ||
1016 | int count, i; | ||
1017 | |||
1018 | if (!np) /* may be pcie device */ | ||
1019 | return 0; | ||
1020 | |||
1021 | count = of_property_count_elems_of_size(np, "fuse-overrides", 8); | ||
1022 | if (count <= 0) | ||
1023 | return count; | ||
1024 | |||
1025 | fuses = nvgpu_kmalloc(g, sizeof(u32) * count * 2); | ||
1026 | if (!fuses) | ||
1027 | return -ENOMEM; | ||
1028 | of_property_read_u32_array(np, "fuse-overrides", fuses, count * 2); | ||
1029 | for (i = 0; i < count; i++) { | ||
1030 | u32 fuse, value; | ||
1031 | |||
1032 | fuse = fuses[2 * i]; | ||
1033 | value = fuses[2 * i + 1]; | ||
1034 | switch (fuse) { | ||
1035 | case GM20B_FUSE_OPT_TPC_DISABLE: | ||
1036 | g->tpc_fs_mask_user = ~value; | ||
1037 | break; | ||
1038 | #ifdef CONFIG_ARCH_TEGRA_18x_SOC | ||
1039 | case GP10B_FUSE_OPT_ECC_EN: | ||
1040 | g->gr.t18x.fecs_feature_override_ecc_val = value; | ||
1041 | break; | ||
1042 | #endif | ||
1043 | default: | ||
1044 | nvgpu_err(g, "ignore unknown fuse override %08x", fuse); | ||
1045 | break; | ||
1046 | } | ||
1047 | } | ||
1048 | |||
1049 | nvgpu_kfree(g, fuses); | ||
1050 | |||
1051 | return 0; | ||
1052 | } | ||
1053 | |||
1054 | static int gk20a_probe(struct platform_device *dev) | ||
1055 | { | ||
1056 | struct nvgpu_os_linux *l; | ||
1057 | struct gk20a *gk20a; | ||
1058 | int err; | ||
1059 | struct gk20a_platform *platform = NULL; | ||
1060 | |||
1061 | if (dev->dev.of_node) { | ||
1062 | const struct of_device_id *match; | ||
1063 | |||
1064 | match = of_match_device(tegra_gk20a_of_match, &dev->dev); | ||
1065 | if (match) | ||
1066 | platform = (struct gk20a_platform *)match->data; | ||
1067 | } else | ||
1068 | platform = (struct gk20a_platform *)dev->dev.platform_data; | ||
1069 | |||
1070 | if (!platform) { | ||
1071 | dev_err(&dev->dev, "no platform data\n"); | ||
1072 | return -ENODATA; | ||
1073 | } | ||
1074 | |||
1075 | gk20a_dbg_fn(""); | ||
1076 | |||
1077 | platform_set_drvdata(dev, platform); | ||
1078 | |||
1079 | if (gk20a_gpu_is_virtual(&dev->dev)) | ||
1080 | return vgpu_probe(dev); | ||
1081 | |||
1082 | l = kzalloc(sizeof(*l), GFP_KERNEL); | ||
1083 | if (!l) { | ||
1084 | dev_err(&dev->dev, "couldn't allocate gk20a support"); | ||
1085 | return -ENOMEM; | ||
1086 | } | ||
1087 | |||
1088 | gk20a = &l->g; | ||
1089 | nvgpu_init_gk20a(gk20a); | ||
1090 | set_gk20a(dev, gk20a); | ||
1091 | l->dev = &dev->dev; | ||
1092 | gk20a->log_mask = NVGPU_DEFAULT_DBG_MASK; | ||
1093 | |||
1094 | nvgpu_kmem_init(gk20a); | ||
1095 | |||
1096 | err = nvgpu_init_enabled_flags(gk20a); | ||
1097 | if (err) | ||
1098 | return err; | ||
1099 | |||
1100 | if (nvgpu_platform_is_simulation(gk20a)) | ||
1101 | __nvgpu_set_enabled(gk20a, NVGPU_IS_FMODEL, true); | ||
1102 | |||
1103 | gk20a->irq_stall = platform_get_irq(dev, 0); | ||
1104 | gk20a->irq_nonstall = platform_get_irq(dev, 1); | ||
1105 | if (gk20a->irq_stall < 0 || gk20a->irq_nonstall < 0) | ||
1106 | return -ENXIO; | ||
1107 | |||
1108 | err = devm_request_threaded_irq(&dev->dev, | ||
1109 | gk20a->irq_stall, | ||
1110 | gk20a_intr_isr_stall, | ||
1111 | gk20a_intr_thread_stall, | ||
1112 | 0, "gk20a_stall", gk20a); | ||
1113 | if (err) { | ||
1114 | dev_err(&dev->dev, | ||
1115 | "failed to request stall intr irq @ %d\n", | ||
1116 | gk20a->irq_stall); | ||
1117 | return err; | ||
1118 | } | ||
1119 | err = devm_request_irq(&dev->dev, | ||
1120 | gk20a->irq_nonstall, | ||
1121 | gk20a_intr_isr_nonstall, | ||
1122 | 0, "gk20a_nonstall", gk20a); | ||
1123 | if (err) { | ||
1124 | dev_err(&dev->dev, | ||
1125 | "failed to request non-stall intr irq @ %d\n", | ||
1126 | gk20a->irq_nonstall); | ||
1127 | return err; | ||
1128 | } | ||
1129 | disable_irq(gk20a->irq_stall); | ||
1130 | if (gk20a->irq_stall != gk20a->irq_nonstall) | ||
1131 | disable_irq(gk20a->irq_nonstall); | ||
1132 | |||
1133 | err = gk20a_init_support(dev); | ||
1134 | if (err) | ||
1135 | return err; | ||
1136 | |||
1137 | err = nvgpu_read_fuse_overrides(gk20a); | ||
1138 | |||
1139 | #ifdef CONFIG_RESET_CONTROLLER | ||
1140 | platform->reset_control = devm_reset_control_get(&dev->dev, NULL); | ||
1141 | if (IS_ERR(platform->reset_control)) | ||
1142 | platform->reset_control = NULL; | ||
1143 | #endif | ||
1144 | |||
1145 | err = nvgpu_probe(gk20a, "gpu.0", INTERFACE_NAME, &nvgpu_class); | ||
1146 | if (err) | ||
1147 | return err; | ||
1148 | |||
1149 | err = gk20a_pm_init(&dev->dev); | ||
1150 | if (err) { | ||
1151 | dev_err(&dev->dev, "pm init failed"); | ||
1152 | return err; | ||
1153 | } | ||
1154 | |||
1155 | gk20a->mm.has_physical_mode = !nvgpu_is_hypervisor_mode(gk20a); | ||
1156 | |||
1157 | return 0; | ||
1158 | } | ||
1159 | |||
1160 | int nvgpu_remove(struct device *dev, struct class *class) | ||
1161 | { | ||
1162 | struct gk20a *g = get_gk20a(dev); | ||
1163 | struct nvgpu_os_linux *l = nvgpu_os_linux_from_gk20a(g); | ||
1164 | struct gk20a_platform *platform = gk20a_get_platform(dev); | ||
1165 | int err; | ||
1166 | |||
1167 | gk20a_dbg_fn(""); | ||
1168 | |||
1169 | err = nvgpu_quiesce(g); | ||
1170 | WARN(err, "gpu failed to idle during driver removal"); | ||
1171 | |||
1172 | if (nvgpu_mem_is_valid(&g->syncpt_mem)) | ||
1173 | nvgpu_dma_free(g, &g->syncpt_mem); | ||
1174 | |||
1175 | if (platform->has_cde) | ||
1176 | gk20a_cde_destroy(l); | ||
1177 | |||
1178 | #ifdef CONFIG_GK20A_CTXSW_TRACE | ||
1179 | gk20a_ctxsw_trace_cleanup(g); | ||
1180 | #endif | ||
1181 | |||
1182 | gk20a_sched_ctrl_cleanup(g); | ||
1183 | |||
1184 | if (IS_ENABLED(CONFIG_GK20A_DEVFREQ)) | ||
1185 | gk20a_scale_exit(dev); | ||
1186 | |||
1187 | #ifdef CONFIG_ARCH_TEGRA_18x_SOC | ||
1188 | nvgpu_clk_arb_cleanup_arbiter(g); | ||
1189 | #endif | ||
1190 | |||
1191 | gk20a_user_deinit(dev, class); | ||
1192 | |||
1193 | gk20a_debug_deinit(g); | ||
1194 | |||
1195 | nvgpu_remove_sysfs(dev); | ||
1196 | |||
1197 | if (platform->secure_buffer.destroy) | ||
1198 | platform->secure_buffer.destroy(g, | ||
1199 | &platform->secure_buffer); | ||
1200 | |||
1201 | if (pm_runtime_enabled(dev)) | ||
1202 | pm_runtime_disable(dev); | ||
1203 | |||
1204 | if (platform->remove) | ||
1205 | platform->remove(dev); | ||
1206 | |||
1207 | gk20a_dbg_fn("removed"); | ||
1208 | |||
1209 | return err; | ||
1210 | } | ||
1211 | |||
1212 | static int __exit gk20a_remove(struct platform_device *pdev) | ||
1213 | { | ||
1214 | int err; | ||
1215 | struct device *dev = &pdev->dev; | ||
1216 | struct gk20a *g = get_gk20a(dev); | ||
1217 | |||
1218 | if (gk20a_gpu_is_virtual(dev)) | ||
1219 | return vgpu_remove(pdev); | ||
1220 | |||
1221 | err = nvgpu_remove(dev, &nvgpu_class); | ||
1222 | |||
1223 | set_gk20a(pdev, NULL); | ||
1224 | gk20a_put(g); | ||
1225 | |||
1226 | return err; | ||
1227 | } | ||
1228 | |||
1229 | static struct platform_driver gk20a_driver = { | ||
1230 | .probe = gk20a_probe, | ||
1231 | .remove = __exit_p(gk20a_remove), | ||
1232 | .shutdown = gk20a_pm_shutdown, | ||
1233 | .driver = { | ||
1234 | .owner = THIS_MODULE, | ||
1235 | .name = "gk20a", | ||
1236 | .probe_type = PROBE_PREFER_ASYNCHRONOUS, | ||
1237 | #ifdef CONFIG_OF | ||
1238 | .of_match_table = tegra_gk20a_of_match, | ||
1239 | #endif | ||
1240 | #ifdef CONFIG_PM | ||
1241 | .pm = &gk20a_pm_ops, | ||
1242 | #endif | ||
1243 | .suppress_bind_attrs = true, | ||
1244 | } | ||
1245 | }; | ||
1246 | |||
1247 | struct class nvgpu_class = { | ||
1248 | .owner = THIS_MODULE, | ||
1249 | .name = CLASS_NAME, | ||
1250 | }; | ||
1251 | |||
1252 | static int __init gk20a_init(void) | ||
1253 | { | ||
1254 | |||
1255 | int ret; | ||
1256 | |||
1257 | ret = class_register(&nvgpu_class); | ||
1258 | if (ret) | ||
1259 | return ret; | ||
1260 | |||
1261 | ret = nvgpu_pci_init(); | ||
1262 | if (ret) | ||
1263 | return ret; | ||
1264 | |||
1265 | return platform_driver_register(&gk20a_driver); | ||
1266 | } | ||
1267 | |||
1268 | static void __exit gk20a_exit(void) | ||
1269 | { | ||
1270 | nvgpu_pci_exit(); | ||
1271 | platform_driver_unregister(&gk20a_driver); | ||
1272 | class_unregister(&nvgpu_class); | ||
1273 | } | ||
1274 | |||
1275 | MODULE_LICENSE("GPL v2"); | ||
1276 | module_init(gk20a_init); | ||
1277 | module_exit(gk20a_exit); | ||