Loading...
Note: File does not exist in v3.1.
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 2020-2024 Intel Corporation
4 */
5
6#include <linux/highmem.h>
7#include <linux/moduleparam.h>
8#include <linux/pci.h>
9#include <linux/pm_runtime.h>
10#include <linux/reboot.h>
11
12#include "ivpu_coredump.h"
13#include "ivpu_drv.h"
14#include "ivpu_fw.h"
15#include "ivpu_fw_log.h"
16#include "ivpu_hw.h"
17#include "ivpu_ipc.h"
18#include "ivpu_job.h"
19#include "ivpu_jsm_msg.h"
20#include "ivpu_mmu.h"
21#include "ivpu_ms.h"
22#include "ivpu_pm.h"
23#include "ivpu_trace.h"
24#include "vpu_boot_api.h"
25
26static bool ivpu_disable_recovery;
27#if IS_ENABLED(CONFIG_DRM_ACCEL_IVPU_DEBUG)
28module_param_named_unsafe(disable_recovery, ivpu_disable_recovery, bool, 0644);
29MODULE_PARM_DESC(disable_recovery, "Disables recovery when NPU hang is detected");
30#endif
31
32static unsigned long ivpu_tdr_timeout_ms;
33module_param_named(tdr_timeout_ms, ivpu_tdr_timeout_ms, ulong, 0644);
34MODULE_PARM_DESC(tdr_timeout_ms, "Timeout for device hang detection, in milliseconds, 0 - default");
35
36#define PM_RESCHEDULE_LIMIT 5
37
38static void ivpu_pm_prepare_cold_boot(struct ivpu_device *vdev)
39{
40 struct ivpu_fw_info *fw = vdev->fw;
41
42 ivpu_cmdq_reset_all_contexts(vdev);
43 ivpu_ipc_reset(vdev);
44 ivpu_fw_log_reset(vdev);
45 ivpu_fw_load(vdev);
46 fw->entry_point = fw->cold_boot_entry_point;
47}
48
49static void ivpu_pm_prepare_warm_boot(struct ivpu_device *vdev)
50{
51 struct ivpu_fw_info *fw = vdev->fw;
52 struct vpu_boot_params *bp = ivpu_bo_vaddr(fw->mem);
53
54 if (!bp->save_restore_ret_address) {
55 ivpu_pm_prepare_cold_boot(vdev);
56 return;
57 }
58
59 ivpu_dbg(vdev, FW_BOOT, "Save/restore entry point %llx", bp->save_restore_ret_address);
60 fw->entry_point = bp->save_restore_ret_address;
61}
62
63static int ivpu_suspend(struct ivpu_device *vdev)
64{
65 int ret;
66
67 ivpu_prepare_for_reset(vdev);
68
69 ret = ivpu_shutdown(vdev);
70 if (ret)
71 ivpu_err(vdev, "Failed to shutdown NPU: %d\n", ret);
72
73 return ret;
74}
75
76static int ivpu_resume(struct ivpu_device *vdev)
77{
78 int ret;
79
80retry:
81 pci_set_power_state(to_pci_dev(vdev->drm.dev), PCI_D0);
82 pci_restore_state(to_pci_dev(vdev->drm.dev));
83
84 ret = ivpu_hw_power_up(vdev);
85 if (ret) {
86 ivpu_err(vdev, "Failed to power up HW: %d\n", ret);
87 goto err_power_down;
88 }
89
90 ret = ivpu_mmu_enable(vdev);
91 if (ret) {
92 ivpu_err(vdev, "Failed to resume MMU: %d\n", ret);
93 goto err_power_down;
94 }
95
96 ret = ivpu_boot(vdev);
97 if (ret)
98 goto err_mmu_disable;
99
100 return 0;
101
102err_mmu_disable:
103 ivpu_mmu_disable(vdev);
104err_power_down:
105 ivpu_hw_power_down(vdev);
106 pci_set_power_state(to_pci_dev(vdev->drm.dev), PCI_D3hot);
107
108 if (!ivpu_fw_is_cold_boot(vdev)) {
109 ivpu_pm_prepare_cold_boot(vdev);
110 goto retry;
111 } else {
112 ivpu_err(vdev, "Failed to resume the FW: %d\n", ret);
113 }
114
115 return ret;
116}
117
118static void ivpu_pm_reset_begin(struct ivpu_device *vdev)
119{
120 pm_runtime_disable(vdev->drm.dev);
121
122 atomic_inc(&vdev->pm->reset_counter);
123 atomic_set(&vdev->pm->reset_pending, 1);
124 down_write(&vdev->pm->reset_lock);
125}
126
127static void ivpu_pm_reset_complete(struct ivpu_device *vdev)
128{
129 int ret;
130
131 ivpu_pm_prepare_cold_boot(vdev);
132 ivpu_jobs_abort_all(vdev);
133 ivpu_ms_cleanup_all(vdev);
134
135 ret = ivpu_resume(vdev);
136 if (ret) {
137 ivpu_err(vdev, "Failed to resume NPU: %d\n", ret);
138 pm_runtime_set_suspended(vdev->drm.dev);
139 } else {
140 pm_runtime_set_active(vdev->drm.dev);
141 }
142
143 up_write(&vdev->pm->reset_lock);
144 atomic_set(&vdev->pm->reset_pending, 0);
145
146 pm_runtime_mark_last_busy(vdev->drm.dev);
147 pm_runtime_enable(vdev->drm.dev);
148}
149
150static void ivpu_pm_recovery_work(struct work_struct *work)
151{
152 struct ivpu_pm_info *pm = container_of(work, struct ivpu_pm_info, recovery_work);
153 struct ivpu_device *vdev = pm->vdev;
154 char *evt[2] = {"IVPU_PM_EVENT=IVPU_RECOVER", NULL};
155
156 ivpu_err(vdev, "Recovering the NPU (reset #%d)\n", atomic_read(&vdev->pm->reset_counter));
157
158 ivpu_pm_reset_begin(vdev);
159
160 if (!pm_runtime_status_suspended(vdev->drm.dev)) {
161 ivpu_jsm_state_dump(vdev);
162 ivpu_dev_coredump(vdev);
163 ivpu_suspend(vdev);
164 }
165
166 ivpu_pm_reset_complete(vdev);
167
168 kobject_uevent_env(&vdev->drm.dev->kobj, KOBJ_CHANGE, evt);
169}
170
171void ivpu_pm_trigger_recovery(struct ivpu_device *vdev, const char *reason)
172{
173 ivpu_err(vdev, "Recovery triggered by %s\n", reason);
174
175 if (ivpu_disable_recovery) {
176 ivpu_err(vdev, "Recovery not available when disable_recovery param is set\n");
177 return;
178 }
179
180 if (ivpu_is_fpga(vdev)) {
181 ivpu_err(vdev, "Recovery not available on FPGA\n");
182 return;
183 }
184
185 /* Trigger recovery if it's not in progress */
186 if (atomic_cmpxchg(&vdev->pm->reset_pending, 0, 1) == 0) {
187 ivpu_hw_diagnose_failure(vdev);
188 ivpu_hw_irq_disable(vdev); /* Disable IRQ early to protect from IRQ storm */
189 queue_work(system_long_wq, &vdev->pm->recovery_work);
190 }
191}
192
193static void ivpu_job_timeout_work(struct work_struct *work)
194{
195 struct ivpu_pm_info *pm = container_of(work, struct ivpu_pm_info, job_timeout_work.work);
196 struct ivpu_device *vdev = pm->vdev;
197
198 ivpu_pm_trigger_recovery(vdev, "TDR");
199}
200
201void ivpu_start_job_timeout_detection(struct ivpu_device *vdev)
202{
203 unsigned long timeout_ms = ivpu_tdr_timeout_ms ? ivpu_tdr_timeout_ms : vdev->timeout.tdr;
204
205 /* No-op if already queued */
206 queue_delayed_work(system_wq, &vdev->pm->job_timeout_work, msecs_to_jiffies(timeout_ms));
207}
208
209void ivpu_stop_job_timeout_detection(struct ivpu_device *vdev)
210{
211 cancel_delayed_work_sync(&vdev->pm->job_timeout_work);
212}
213
214int ivpu_pm_suspend_cb(struct device *dev)
215{
216 struct drm_device *drm = dev_get_drvdata(dev);
217 struct ivpu_device *vdev = to_ivpu_device(drm);
218 unsigned long timeout;
219
220 trace_pm("suspend");
221 ivpu_dbg(vdev, PM, "Suspend..\n");
222
223 timeout = jiffies + msecs_to_jiffies(vdev->timeout.tdr);
224 while (!ivpu_hw_is_idle(vdev)) {
225 cond_resched();
226 if (time_after_eq(jiffies, timeout)) {
227 ivpu_err(vdev, "Failed to enter idle on system suspend\n");
228 return -EBUSY;
229 }
230 }
231
232 ivpu_jsm_pwr_d0i3_enter(vdev);
233
234 ivpu_suspend(vdev);
235 ivpu_pm_prepare_warm_boot(vdev);
236
237 ivpu_dbg(vdev, PM, "Suspend done.\n");
238 trace_pm("suspend done");
239
240 return 0;
241}
242
243int ivpu_pm_resume_cb(struct device *dev)
244{
245 struct drm_device *drm = dev_get_drvdata(dev);
246 struct ivpu_device *vdev = to_ivpu_device(drm);
247 int ret;
248
249 trace_pm("resume");
250 ivpu_dbg(vdev, PM, "Resume..\n");
251
252 ret = ivpu_resume(vdev);
253 if (ret)
254 ivpu_err(vdev, "Failed to resume: %d\n", ret);
255
256 ivpu_dbg(vdev, PM, "Resume done.\n");
257 trace_pm("resume done");
258
259 return ret;
260}
261
262int ivpu_pm_runtime_suspend_cb(struct device *dev)
263{
264 struct drm_device *drm = dev_get_drvdata(dev);
265 struct ivpu_device *vdev = to_ivpu_device(drm);
266 int ret, ret_d0i3;
267 bool is_idle;
268
269 drm_WARN_ON(&vdev->drm, !xa_empty(&vdev->submitted_jobs_xa));
270 drm_WARN_ON(&vdev->drm, work_pending(&vdev->pm->recovery_work));
271
272 trace_pm("runtime suspend");
273 ivpu_dbg(vdev, PM, "Runtime suspend..\n");
274
275 ivpu_mmu_disable(vdev);
276
277 is_idle = ivpu_hw_is_idle(vdev) || vdev->pm->dct_active_percent;
278 if (!is_idle)
279 ivpu_err(vdev, "NPU is not idle before autosuspend\n");
280
281 ret_d0i3 = ivpu_jsm_pwr_d0i3_enter(vdev);
282 if (ret_d0i3)
283 ivpu_err(vdev, "Failed to prepare for d0i3: %d\n", ret_d0i3);
284
285 ret = ivpu_suspend(vdev);
286 if (ret)
287 ivpu_err(vdev, "Failed to suspend NPU: %d\n", ret);
288
289 if (!is_idle || ret_d0i3) {
290 ivpu_err(vdev, "Forcing cold boot due to previous errors\n");
291 atomic_inc(&vdev->pm->reset_counter);
292 ivpu_dev_coredump(vdev);
293 ivpu_pm_prepare_cold_boot(vdev);
294 } else {
295 ivpu_pm_prepare_warm_boot(vdev);
296 }
297
298 ivpu_dbg(vdev, PM, "Runtime suspend done.\n");
299 trace_pm("runtime suspend done");
300
301 return 0;
302}
303
304int ivpu_pm_runtime_resume_cb(struct device *dev)
305{
306 struct drm_device *drm = dev_get_drvdata(dev);
307 struct ivpu_device *vdev = to_ivpu_device(drm);
308 int ret;
309
310 trace_pm("runtime resume");
311 ivpu_dbg(vdev, PM, "Runtime resume..\n");
312
313 ret = ivpu_resume(vdev);
314 if (ret)
315 ivpu_err(vdev, "Failed to set RESUME state: %d\n", ret);
316
317 ivpu_dbg(vdev, PM, "Runtime resume done.\n");
318 trace_pm("runtime resume done");
319
320 return ret;
321}
322
323int ivpu_rpm_get(struct ivpu_device *vdev)
324{
325 int ret;
326
327 ret = pm_runtime_resume_and_get(vdev->drm.dev);
328 if (ret < 0) {
329 ivpu_err(vdev, "Failed to resume NPU: %d\n", ret);
330 pm_runtime_set_suspended(vdev->drm.dev);
331 }
332
333 return ret;
334}
335
336void ivpu_rpm_put(struct ivpu_device *vdev)
337{
338 pm_runtime_mark_last_busy(vdev->drm.dev);
339 pm_runtime_put_autosuspend(vdev->drm.dev);
340}
341
342void ivpu_pm_reset_prepare_cb(struct pci_dev *pdev)
343{
344 struct ivpu_device *vdev = pci_get_drvdata(pdev);
345
346 ivpu_dbg(vdev, PM, "Pre-reset..\n");
347
348 ivpu_pm_reset_begin(vdev);
349
350 if (!pm_runtime_status_suspended(vdev->drm.dev)) {
351 ivpu_prepare_for_reset(vdev);
352 ivpu_hw_reset(vdev);
353 }
354
355 ivpu_dbg(vdev, PM, "Pre-reset done.\n");
356}
357
358void ivpu_pm_reset_done_cb(struct pci_dev *pdev)
359{
360 struct ivpu_device *vdev = pci_get_drvdata(pdev);
361
362 ivpu_dbg(vdev, PM, "Post-reset..\n");
363
364 ivpu_pm_reset_complete(vdev);
365
366 ivpu_dbg(vdev, PM, "Post-reset done.\n");
367}
368
369void ivpu_pm_init(struct ivpu_device *vdev)
370{
371 struct device *dev = vdev->drm.dev;
372 struct ivpu_pm_info *pm = vdev->pm;
373 int delay;
374
375 pm->vdev = vdev;
376
377 init_rwsem(&pm->reset_lock);
378 atomic_set(&pm->reset_pending, 0);
379 atomic_set(&pm->reset_counter, 0);
380
381 INIT_WORK(&pm->recovery_work, ivpu_pm_recovery_work);
382 INIT_DELAYED_WORK(&pm->job_timeout_work, ivpu_job_timeout_work);
383
384 if (ivpu_disable_recovery)
385 delay = -1;
386 else
387 delay = vdev->timeout.autosuspend;
388
389 pm_runtime_use_autosuspend(dev);
390 pm_runtime_set_autosuspend_delay(dev, delay);
391 pm_runtime_set_active(dev);
392
393 ivpu_dbg(vdev, PM, "Autosuspend delay = %d\n", delay);
394}
395
396void ivpu_pm_cancel_recovery(struct ivpu_device *vdev)
397{
398 drm_WARN_ON(&vdev->drm, delayed_work_pending(&vdev->pm->job_timeout_work));
399 cancel_work_sync(&vdev->pm->recovery_work);
400}
401
402void ivpu_pm_enable(struct ivpu_device *vdev)
403{
404 struct device *dev = vdev->drm.dev;
405
406 pm_runtime_allow(dev);
407 pm_runtime_mark_last_busy(dev);
408 pm_runtime_put_autosuspend(dev);
409}
410
411void ivpu_pm_disable(struct ivpu_device *vdev)
412{
413 pm_runtime_get_noresume(vdev->drm.dev);
414 pm_runtime_forbid(vdev->drm.dev);
415}
416
417int ivpu_pm_dct_init(struct ivpu_device *vdev)
418{
419 if (vdev->pm->dct_active_percent)
420 return ivpu_pm_dct_enable(vdev, vdev->pm->dct_active_percent);
421
422 return 0;
423}
424
425int ivpu_pm_dct_enable(struct ivpu_device *vdev, u8 active_percent)
426{
427 u32 active_us, inactive_us;
428 int ret;
429
430 if (active_percent == 0 || active_percent > 100)
431 return -EINVAL;
432
433 active_us = (DCT_PERIOD_US * active_percent) / 100;
434 inactive_us = DCT_PERIOD_US - active_us;
435
436 ret = ivpu_jsm_dct_enable(vdev, active_us, inactive_us);
437 if (ret) {
438 ivpu_err_ratelimited(vdev, "Failed to enable DCT: %d\n", ret);
439 return ret;
440 }
441
442 vdev->pm->dct_active_percent = active_percent;
443
444 ivpu_dbg(vdev, PM, "DCT set to %u%% (D0: %uus, D0i2: %uus)\n",
445 active_percent, active_us, inactive_us);
446 return 0;
447}
448
449int ivpu_pm_dct_disable(struct ivpu_device *vdev)
450{
451 int ret;
452
453 ret = ivpu_jsm_dct_disable(vdev);
454 if (ret) {
455 ivpu_err_ratelimited(vdev, "Failed to disable DCT: %d\n", ret);
456 return ret;
457 }
458
459 vdev->pm->dct_active_percent = 0;
460
461 ivpu_dbg(vdev, PM, "DCT disabled\n");
462 return 0;
463}
464
465void ivpu_pm_dct_irq_thread_handler(struct ivpu_device *vdev)
466{
467 bool enable;
468 int ret;
469
470 if (ivpu_hw_btrs_dct_get_request(vdev, &enable))
471 return;
472
473 if (vdev->pm->dct_active_percent)
474 ret = ivpu_pm_dct_enable(vdev, DCT_DEFAULT_ACTIVE_PERCENT);
475 else
476 ret = ivpu_pm_dct_disable(vdev);
477
478 if (!ret)
479 ivpu_hw_btrs_dct_set_status(vdev, enable, vdev->pm->dct_active_percent);
480}