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 26 static bool ivpu_disable_recovery; 27 #if IS_ENABLED(CONFIG_DRM_ACCEL_IVPU_DEBUG) 28 module_param_named_unsafe(disable_recovery, ivpu_disable_recovery, bool, 0644); 29 MODULE_PARM_DESC(disable_recovery, "Disables recovery when NPU hang is detected"); 30 #endif 31 32 static unsigned long ivpu_tdr_timeout_ms; 33 module_param_named(tdr_timeout_ms, ivpu_tdr_timeout_ms, ulong, 0644); 34 MODULE_PARM_DESC(tdr_timeout_ms, "Timeout for device hang detection, in milliseconds, 0 - default"); 35 36 #define PM_RESCHEDULE_LIMIT 5 37 38 static 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 49 static 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 63 static 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 76 static int ivpu_resume(struct ivpu_device *vdev) 77 { 78 int ret; 79 80 retry: 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; 101 102 err_mmu_disable: 103 ivpu_mmu_disable(vdev); 104 err_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 } 117 118 static 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 } 126 127 static 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 } 149 150 static 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 } 170 171 void 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 /* Trigger recovery if it's not in progress */ 181 if (atomic_cmpxchg(&vdev->pm->reset_pending, 0, 1) == 0) { 182 ivpu_hw_diagnose_failure(vdev); 183 ivpu_hw_irq_disable(vdev); /* Disable IRQ early to protect from IRQ storm */ 184 queue_work(system_unbound_wq, &vdev->pm->recovery_work); 185 } 186 } 187 188 static void ivpu_job_timeout_work(struct work_struct *work) 189 { 190 struct ivpu_pm_info *pm = container_of(work, struct ivpu_pm_info, job_timeout_work.work); 191 struct ivpu_device *vdev = pm->vdev; 192 193 ivpu_pm_trigger_recovery(vdev, "TDR"); 194 } 195 196 void ivpu_start_job_timeout_detection(struct ivpu_device *vdev) 197 { 198 unsigned long timeout_ms = ivpu_tdr_timeout_ms ? ivpu_tdr_timeout_ms : vdev->timeout.tdr; 199 200 /* No-op if already queued */ 201 queue_delayed_work(system_wq, &vdev->pm->job_timeout_work, msecs_to_jiffies(timeout_ms)); 202 } 203 204 void ivpu_stop_job_timeout_detection(struct ivpu_device *vdev) 205 { 206 cancel_delayed_work_sync(&vdev->pm->job_timeout_work); 207 } 208 209 int ivpu_pm_suspend_cb(struct device *dev) 210 { 211 struct drm_device *drm = dev_get_drvdata(dev); 212 struct ivpu_device *vdev = to_ivpu_device(drm); 213 unsigned long timeout; 214 215 trace_pm("suspend"); 216 ivpu_dbg(vdev, PM, "Suspend..\n"); 217 218 timeout = jiffies + msecs_to_jiffies(vdev->timeout.tdr); 219 while (!ivpu_hw_is_idle(vdev)) { 220 cond_resched(); 221 if (time_after_eq(jiffies, timeout)) { 222 ivpu_err(vdev, "Failed to enter idle on system suspend\n"); 223 return -EBUSY; 224 } 225 } 226 227 ivpu_jsm_pwr_d0i3_enter(vdev); 228 229 ivpu_suspend(vdev); 230 ivpu_pm_prepare_warm_boot(vdev); 231 232 ivpu_dbg(vdev, PM, "Suspend done.\n"); 233 trace_pm("suspend done"); 234 235 return 0; 236 } 237 238 int ivpu_pm_resume_cb(struct device *dev) 239 { 240 struct drm_device *drm = dev_get_drvdata(dev); 241 struct ivpu_device *vdev = to_ivpu_device(drm); 242 int ret; 243 244 trace_pm("resume"); 245 ivpu_dbg(vdev, PM, "Resume..\n"); 246 247 ret = ivpu_resume(vdev); 248 if (ret) 249 ivpu_err(vdev, "Failed to resume: %d\n", ret); 250 251 ivpu_dbg(vdev, PM, "Resume done.\n"); 252 trace_pm("resume done"); 253 254 return ret; 255 } 256 257 int ivpu_pm_runtime_suspend_cb(struct device *dev) 258 { 259 struct drm_device *drm = dev_get_drvdata(dev); 260 struct ivpu_device *vdev = to_ivpu_device(drm); 261 int ret, ret_d0i3; 262 bool is_idle; 263 264 drm_WARN_ON(&vdev->drm, !xa_empty(&vdev->submitted_jobs_xa)); 265 drm_WARN_ON(&vdev->drm, work_pending(&vdev->pm->recovery_work)); 266 267 trace_pm("runtime suspend"); 268 ivpu_dbg(vdev, PM, "Runtime suspend..\n"); 269 270 ivpu_mmu_disable(vdev); 271 272 is_idle = ivpu_hw_is_idle(vdev) || vdev->pm->dct_active_percent; 273 if (!is_idle) 274 ivpu_err(vdev, "NPU is not idle before autosuspend\n"); 275 276 ret_d0i3 = ivpu_jsm_pwr_d0i3_enter(vdev); 277 if (ret_d0i3) 278 ivpu_err(vdev, "Failed to prepare for d0i3: %d\n", ret_d0i3); 279 280 ret = ivpu_suspend(vdev); 281 if (ret) 282 ivpu_err(vdev, "Failed to suspend NPU: %d\n", ret); 283 284 if (!is_idle || ret_d0i3) { 285 ivpu_err(vdev, "Forcing cold boot due to previous errors\n"); 286 atomic_inc(&vdev->pm->reset_counter); 287 ivpu_dev_coredump(vdev); 288 ivpu_pm_prepare_cold_boot(vdev); 289 } else { 290 ivpu_pm_prepare_warm_boot(vdev); 291 } 292 293 ivpu_dbg(vdev, PM, "Runtime suspend done.\n"); 294 trace_pm("runtime suspend done"); 295 296 return 0; 297 } 298 299 int ivpu_pm_runtime_resume_cb(struct device *dev) 300 { 301 struct drm_device *drm = dev_get_drvdata(dev); 302 struct ivpu_device *vdev = to_ivpu_device(drm); 303 int ret; 304 305 trace_pm("runtime resume"); 306 ivpu_dbg(vdev, PM, "Runtime resume..\n"); 307 308 ret = ivpu_resume(vdev); 309 if (ret) 310 ivpu_err(vdev, "Failed to set RESUME state: %d\n", ret); 311 312 ivpu_dbg(vdev, PM, "Runtime resume done.\n"); 313 trace_pm("runtime resume done"); 314 315 return ret; 316 } 317 318 int ivpu_rpm_get(struct ivpu_device *vdev) 319 { 320 int ret; 321 322 ret = pm_runtime_resume_and_get(vdev->drm.dev); 323 if (ret < 0) { 324 ivpu_err(vdev, "Failed to resume NPU: %d\n", ret); 325 pm_runtime_set_suspended(vdev->drm.dev); 326 } 327 328 return ret; 329 } 330 331 void ivpu_rpm_put(struct ivpu_device *vdev) 332 { 333 pm_runtime_mark_last_busy(vdev->drm.dev); 334 pm_runtime_put_autosuspend(vdev->drm.dev); 335 } 336 337 void ivpu_pm_reset_prepare_cb(struct pci_dev *pdev) 338 { 339 struct ivpu_device *vdev = pci_get_drvdata(pdev); 340 341 ivpu_dbg(vdev, PM, "Pre-reset..\n"); 342 343 ivpu_pm_reset_begin(vdev); 344 345 if (!pm_runtime_status_suspended(vdev->drm.dev)) { 346 ivpu_prepare_for_reset(vdev); 347 ivpu_hw_reset(vdev); 348 } 349 350 ivpu_dbg(vdev, PM, "Pre-reset done.\n"); 351 } 352 353 void ivpu_pm_reset_done_cb(struct pci_dev *pdev) 354 { 355 struct ivpu_device *vdev = pci_get_drvdata(pdev); 356 357 ivpu_dbg(vdev, PM, "Post-reset..\n"); 358 359 ivpu_pm_reset_complete(vdev); 360 361 ivpu_dbg(vdev, PM, "Post-reset done.\n"); 362 } 363 364 void ivpu_pm_init(struct ivpu_device *vdev) 365 { 366 struct device *dev = vdev->drm.dev; 367 struct ivpu_pm_info *pm = vdev->pm; 368 int delay; 369 370 pm->vdev = vdev; 371 372 init_rwsem(&pm->reset_lock); 373 atomic_set(&pm->reset_pending, 0); 374 atomic_set(&pm->reset_counter, 0); 375 376 INIT_WORK(&pm->recovery_work, ivpu_pm_recovery_work); 377 INIT_DELAYED_WORK(&pm->job_timeout_work, ivpu_job_timeout_work); 378 379 if (ivpu_disable_recovery) 380 delay = -1; 381 else 382 delay = vdev->timeout.autosuspend; 383 384 pm_runtime_use_autosuspend(dev); 385 pm_runtime_set_autosuspend_delay(dev, delay); 386 pm_runtime_set_active(dev); 387 388 ivpu_dbg(vdev, PM, "Autosuspend delay = %d\n", delay); 389 } 390 391 void ivpu_pm_cancel_recovery(struct ivpu_device *vdev) 392 { 393 drm_WARN_ON(&vdev->drm, delayed_work_pending(&vdev->pm->job_timeout_work)); 394 cancel_work_sync(&vdev->pm->recovery_work); 395 } 396 397 void ivpu_pm_enable(struct ivpu_device *vdev) 398 { 399 struct device *dev = vdev->drm.dev; 400 401 pm_runtime_allow(dev); 402 pm_runtime_mark_last_busy(dev); 403 pm_runtime_put_autosuspend(dev); 404 } 405 406 void ivpu_pm_disable(struct ivpu_device *vdev) 407 { 408 pm_runtime_get_noresume(vdev->drm.dev); 409 pm_runtime_forbid(vdev->drm.dev); 410 } 411 412 int ivpu_pm_dct_init(struct ivpu_device *vdev) 413 { 414 if (vdev->pm->dct_active_percent) 415 return ivpu_pm_dct_enable(vdev, vdev->pm->dct_active_percent); 416 417 return 0; 418 } 419 420 int ivpu_pm_dct_enable(struct ivpu_device *vdev, u8 active_percent) 421 { 422 u32 active_us, inactive_us; 423 int ret; 424 425 if (active_percent == 0 || active_percent > 100) 426 return -EINVAL; 427 428 active_us = (DCT_PERIOD_US * active_percent) / 100; 429 inactive_us = DCT_PERIOD_US - active_us; 430 431 vdev->pm->dct_active_percent = active_percent; 432 433 ivpu_dbg(vdev, PM, "DCT requested %u%% (D0: %uus, D0i2: %uus)\n", 434 active_percent, active_us, inactive_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 return 0; 443 } 444 445 int ivpu_pm_dct_disable(struct ivpu_device *vdev) 446 { 447 int ret; 448 449 vdev->pm->dct_active_percent = 0; 450 451 ivpu_dbg(vdev, PM, "DCT requested to be disabled\n"); 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 return 0; 460 } 461 462 void ivpu_pm_irq_dct_work_fn(struct work_struct *work) 463 { 464 struct ivpu_device *vdev = container_of(work, struct ivpu_device, irq_dct_work); 465 bool enable; 466 int ret; 467 468 if (ivpu_hw_btrs_dct_get_request(vdev, &enable)) 469 return; 470 471 if (enable) 472 ret = ivpu_pm_dct_enable(vdev, DCT_DEFAULT_ACTIVE_PERCENT); 473 else 474 ret = ivpu_pm_dct_disable(vdev); 475 476 if (!ret) 477 ivpu_hw_btrs_dct_set_status(vdev, enable, vdev->pm->dct_active_percent); 478 } 479