1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * PCI Express Downstream Port Containment services driver 4 * Author: Keith Busch <keith.busch@intel.com> 5 * 6 * Copyright (C) 2016 Intel Corp. 7 */ 8 9 #define dev_fmt(fmt) "DPC: " fmt 10 11 #include <linux/aer.h> 12 #include <linux/bitfield.h> 13 #include <linux/delay.h> 14 #include <linux/interrupt.h> 15 #include <linux/init.h> 16 #include <linux/pci.h> 17 18 #include "portdrv.h" 19 #include "../pci.h" 20 21 #define PCI_EXP_DPC_CTL_EN_MASK (PCI_EXP_DPC_CTL_EN_FATAL | \ 22 PCI_EXP_DPC_CTL_EN_NONFATAL) 23 24 static const char * const rp_pio_error_string[] = { 25 "Configuration Request received UR Completion", /* Bit Position 0 */ 26 "Configuration Request received CA Completion", /* Bit Position 1 */ 27 "Configuration Request Completion Timeout", /* Bit Position 2 */ 28 NULL, 29 NULL, 30 NULL, 31 NULL, 32 NULL, 33 "I/O Request received UR Completion", /* Bit Position 8 */ 34 "I/O Request received CA Completion", /* Bit Position 9 */ 35 "I/O Request Completion Timeout", /* Bit Position 10 */ 36 NULL, 37 NULL, 38 NULL, 39 NULL, 40 NULL, 41 "Memory Request received UR Completion", /* Bit Position 16 */ 42 "Memory Request received CA Completion", /* Bit Position 17 */ 43 "Memory Request Completion Timeout", /* Bit Position 18 */ 44 }; 45 46 void pci_save_dpc_state(struct pci_dev *dev) 47 { 48 struct pci_cap_saved_state *save_state; 49 u16 *cap; 50 51 if (!pci_is_pcie(dev)) 52 return; 53 54 save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_DPC); 55 if (!save_state) 56 return; 57 58 cap = (u16 *)&save_state->cap.data[0]; 59 pci_read_config_word(dev, dev->dpc_cap + PCI_EXP_DPC_CTL, cap); 60 } 61 62 void pci_restore_dpc_state(struct pci_dev *dev) 63 { 64 struct pci_cap_saved_state *save_state; 65 u16 *cap; 66 67 if (!pci_is_pcie(dev)) 68 return; 69 70 save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_DPC); 71 if (!save_state) 72 return; 73 74 cap = (u16 *)&save_state->cap.data[0]; 75 pci_write_config_word(dev, dev->dpc_cap + PCI_EXP_DPC_CTL, *cap); 76 } 77 78 static DECLARE_WAIT_QUEUE_HEAD(dpc_completed_waitqueue); 79 80 #ifdef CONFIG_HOTPLUG_PCI_PCIE 81 static bool dpc_completed(struct pci_dev *pdev) 82 { 83 u16 status; 84 85 pci_read_config_word(pdev, pdev->dpc_cap + PCI_EXP_DPC_STATUS, &status); 86 if ((!PCI_POSSIBLE_ERROR(status)) && (status & PCI_EXP_DPC_STATUS_TRIGGER)) 87 return false; 88 89 if (test_bit(PCI_DPC_RECOVERING, &pdev->priv_flags)) 90 return false; 91 92 return true; 93 } 94 95 /** 96 * pci_dpc_recovered - whether DPC triggered and has recovered successfully 97 * @pdev: PCI device 98 * 99 * Return true if DPC was triggered for @pdev and has recovered successfully. 100 * Wait for recovery if it hasn't completed yet. Called from the PCIe hotplug 101 * driver to recognize and ignore Link Down/Up events caused by DPC. 102 */ 103 bool pci_dpc_recovered(struct pci_dev *pdev) 104 { 105 struct pci_host_bridge *host; 106 107 if (!pdev->dpc_cap) 108 return false; 109 110 /* 111 * Synchronization between hotplug and DPC is not supported 112 * if DPC is owned by firmware and EDR is not enabled. 113 */ 114 host = pci_find_host_bridge(pdev->bus); 115 if (!host->native_dpc && !IS_ENABLED(CONFIG_PCIE_EDR)) 116 return false; 117 118 /* 119 * Need a timeout in case DPC never completes due to failure of 120 * dpc_wait_rp_inactive(). The spec doesn't mandate a time limit, 121 * but reports indicate that DPC completes within 4 seconds. 122 */ 123 wait_event_timeout(dpc_completed_waitqueue, dpc_completed(pdev), 124 msecs_to_jiffies(4000)); 125 126 return test_and_clear_bit(PCI_DPC_RECOVERED, &pdev->priv_flags); 127 } 128 #endif /* CONFIG_HOTPLUG_PCI_PCIE */ 129 130 static int dpc_wait_rp_inactive(struct pci_dev *pdev) 131 { 132 unsigned long timeout = jiffies + HZ; 133 u16 cap = pdev->dpc_cap, status; 134 135 pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status); 136 while (status & PCI_EXP_DPC_RP_BUSY && 137 !time_after(jiffies, timeout)) { 138 msleep(10); 139 pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status); 140 } 141 if (status & PCI_EXP_DPC_RP_BUSY) { 142 pci_warn(pdev, "root port still busy\n"); 143 return -EBUSY; 144 } 145 return 0; 146 } 147 148 pci_ers_result_t dpc_reset_link(struct pci_dev *pdev) 149 { 150 pci_ers_result_t ret; 151 u16 cap; 152 153 set_bit(PCI_DPC_RECOVERING, &pdev->priv_flags); 154 155 /* 156 * DPC disables the Link automatically in hardware, so it has 157 * already been reset by the time we get here. 158 */ 159 cap = pdev->dpc_cap; 160 161 /* 162 * Wait until the Link is inactive, then clear DPC Trigger Status 163 * to allow the Port to leave DPC. 164 */ 165 if (!pcie_wait_for_link(pdev, false)) 166 pci_info(pdev, "Data Link Layer Link Active not cleared in 1000 msec\n"); 167 168 if (pdev->dpc_rp_extensions && dpc_wait_rp_inactive(pdev)) { 169 clear_bit(PCI_DPC_RECOVERED, &pdev->priv_flags); 170 ret = PCI_ERS_RESULT_DISCONNECT; 171 goto out; 172 } 173 174 pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS, 175 PCI_EXP_DPC_STATUS_TRIGGER); 176 177 if (pci_bridge_wait_for_secondary_bus(pdev, "DPC")) { 178 clear_bit(PCI_DPC_RECOVERED, &pdev->priv_flags); 179 ret = PCI_ERS_RESULT_DISCONNECT; 180 } else { 181 set_bit(PCI_DPC_RECOVERED, &pdev->priv_flags); 182 ret = PCI_ERS_RESULT_RECOVERED; 183 } 184 out: 185 clear_bit(PCI_DPC_RECOVERING, &pdev->priv_flags); 186 wake_up_all(&dpc_completed_waitqueue); 187 return ret; 188 } 189 190 static void dpc_process_rp_pio_error(struct pci_dev *pdev) 191 { 192 u16 cap = pdev->dpc_cap, dpc_status, first_error; 193 u32 status, mask, sev, syserr, exc, log; 194 struct pcie_tlp_log tlp_log; 195 int i; 196 197 pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_STATUS, &status); 198 pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_MASK, &mask); 199 pci_err(pdev, "rp_pio_status: %#010x, rp_pio_mask: %#010x\n", 200 status, mask); 201 202 pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_SEVERITY, &sev); 203 pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_SYSERROR, &syserr); 204 pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_EXCEPTION, &exc); 205 pci_err(pdev, "RP PIO severity=%#010x, syserror=%#010x, exception=%#010x\n", 206 sev, syserr, exc); 207 208 /* Get First Error Pointer */ 209 pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &dpc_status); 210 first_error = FIELD_GET(PCI_EXP_DPC_RP_PIO_FEP, dpc_status); 211 212 for (i = 0; i < ARRAY_SIZE(rp_pio_error_string); i++) { 213 if ((status & ~mask) & (1 << i)) 214 pci_err(pdev, "[%2d] %s%s\n", i, rp_pio_error_string[i], 215 first_error == i ? " (First)" : ""); 216 } 217 218 if (pdev->dpc_rp_log_size < PCIE_STD_NUM_TLP_HEADERLOG) 219 goto clear_status; 220 pcie_read_tlp_log(pdev, cap + PCI_EXP_DPC_RP_PIO_HEADER_LOG, 221 cap + PCI_EXP_DPC_RP_PIO_TLPPREFIX_LOG, 222 dpc_tlp_log_len(pdev), 223 pdev->subordinate->flit_mode, 224 &tlp_log); 225 pcie_print_tlp_log(pdev, &tlp_log, dev_fmt("")); 226 227 if (pdev->dpc_rp_log_size < PCIE_STD_NUM_TLP_HEADERLOG + 1) 228 goto clear_status; 229 pci_read_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_IMPSPEC_LOG, &log); 230 pci_err(pdev, "RP PIO ImpSpec Log %#010x\n", log); 231 232 clear_status: 233 pci_write_config_dword(pdev, cap + PCI_EXP_DPC_RP_PIO_STATUS, status); 234 } 235 236 static int dpc_get_aer_uncorrect_severity(struct pci_dev *dev, 237 struct aer_err_info *info) 238 { 239 int pos = dev->aer_cap; 240 u32 status, mask, sev; 241 242 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status); 243 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, &mask); 244 status &= ~mask; 245 if (!status) 246 return 0; 247 248 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &sev); 249 status &= sev; 250 if (status) 251 info->severity = AER_FATAL; 252 else 253 info->severity = AER_NONFATAL; 254 255 return 1; 256 } 257 258 void dpc_process_error(struct pci_dev *pdev) 259 { 260 u16 cap = pdev->dpc_cap, status, source, reason, ext_reason; 261 struct aer_err_info info; 262 263 pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status); 264 pci_read_config_word(pdev, cap + PCI_EXP_DPC_SOURCE_ID, &source); 265 266 pci_info(pdev, "containment event, status:%#06x source:%#06x\n", 267 status, source); 268 269 reason = status & PCI_EXP_DPC_STATUS_TRIGGER_RSN; 270 ext_reason = status & PCI_EXP_DPC_STATUS_TRIGGER_RSN_EXT; 271 pci_warn(pdev, "%s detected\n", 272 (reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_UNCOR) ? 273 "unmasked uncorrectable error" : 274 (reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_NFE) ? 275 "ERR_NONFATAL" : 276 (reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_FE) ? 277 "ERR_FATAL" : 278 (ext_reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_RP_PIO) ? 279 "RP PIO error" : 280 (ext_reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_SW_TRIGGER) ? 281 "software trigger" : 282 "reserved error"); 283 284 /* show RP PIO error detail information */ 285 if (pdev->dpc_rp_extensions && 286 reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_IN_EXT && 287 ext_reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_RP_PIO) 288 dpc_process_rp_pio_error(pdev); 289 else if (reason == PCI_EXP_DPC_STATUS_TRIGGER_RSN_UNCOR && 290 dpc_get_aer_uncorrect_severity(pdev, &info) && 291 aer_get_device_error_info(pdev, &info)) { 292 aer_print_error(pdev, &info); 293 pci_aer_clear_nonfatal_status(pdev); 294 pci_aer_clear_fatal_status(pdev); 295 } 296 } 297 298 static void pci_clear_surpdn_errors(struct pci_dev *pdev) 299 { 300 if (pdev->dpc_rp_extensions) 301 pci_write_config_dword(pdev, pdev->dpc_cap + 302 PCI_EXP_DPC_RP_PIO_STATUS, ~0); 303 304 /* 305 * In practice, Surprise Down errors have been observed to also set 306 * error bits in the Status Register as well as the Fatal Error 307 * Detected bit in the Device Status Register. 308 */ 309 pci_write_config_word(pdev, PCI_STATUS, 0xffff); 310 311 pcie_capability_write_word(pdev, PCI_EXP_DEVSTA, PCI_EXP_DEVSTA_FED); 312 } 313 314 static void dpc_handle_surprise_removal(struct pci_dev *pdev) 315 { 316 if (!pcie_wait_for_link(pdev, false)) { 317 pci_info(pdev, "Data Link Layer Link Active not cleared in 1000 msec\n"); 318 goto out; 319 } 320 321 if (pdev->dpc_rp_extensions && dpc_wait_rp_inactive(pdev)) 322 goto out; 323 324 pci_aer_raw_clear_status(pdev); 325 pci_clear_surpdn_errors(pdev); 326 327 pci_write_config_word(pdev, pdev->dpc_cap + PCI_EXP_DPC_STATUS, 328 PCI_EXP_DPC_STATUS_TRIGGER); 329 330 out: 331 clear_bit(PCI_DPC_RECOVERED, &pdev->priv_flags); 332 wake_up_all(&dpc_completed_waitqueue); 333 } 334 335 static bool dpc_is_surprise_removal(struct pci_dev *pdev) 336 { 337 u16 status; 338 339 if (!pdev->is_hotplug_bridge) 340 return false; 341 342 if (pci_read_config_word(pdev, pdev->aer_cap + PCI_ERR_UNCOR_STATUS, 343 &status)) 344 return false; 345 346 return status & PCI_ERR_UNC_SURPDN; 347 } 348 349 static irqreturn_t dpc_handler(int irq, void *context) 350 { 351 struct pci_dev *pdev = context; 352 353 /* 354 * According to PCIe r6.0 sec 6.7.6, errors are an expected side effect 355 * of async removal and should be ignored by software. 356 */ 357 if (dpc_is_surprise_removal(pdev)) { 358 dpc_handle_surprise_removal(pdev); 359 return IRQ_HANDLED; 360 } 361 362 dpc_process_error(pdev); 363 364 /* We configure DPC so it only triggers on ERR_FATAL */ 365 pcie_do_recovery(pdev, pci_channel_io_frozen, dpc_reset_link); 366 367 return IRQ_HANDLED; 368 } 369 370 static irqreturn_t dpc_irq(int irq, void *context) 371 { 372 struct pci_dev *pdev = context; 373 u16 cap = pdev->dpc_cap, status; 374 375 pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status); 376 377 if (!(status & PCI_EXP_DPC_STATUS_INTERRUPT) || PCI_POSSIBLE_ERROR(status)) 378 return IRQ_NONE; 379 380 pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS, 381 PCI_EXP_DPC_STATUS_INTERRUPT); 382 if (status & PCI_EXP_DPC_STATUS_TRIGGER) 383 return IRQ_WAKE_THREAD; 384 return IRQ_HANDLED; 385 } 386 387 void pci_dpc_init(struct pci_dev *pdev) 388 { 389 u16 cap; 390 391 pdev->dpc_cap = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_DPC); 392 if (!pdev->dpc_cap) 393 return; 394 395 pci_read_config_word(pdev, pdev->dpc_cap + PCI_EXP_DPC_CAP, &cap); 396 if (!(cap & PCI_EXP_DPC_CAP_RP_EXT)) 397 return; 398 399 pdev->dpc_rp_extensions = true; 400 401 /* Quirks may set dpc_rp_log_size if device or firmware is buggy */ 402 if (!pdev->dpc_rp_log_size) { 403 u16 flags; 404 int ret; 405 406 ret = pcie_capability_read_word(pdev, PCI_EXP_FLAGS, &flags); 407 if (ret) 408 return; 409 410 pdev->dpc_rp_log_size = 411 FIELD_GET(PCI_EXP_DPC_RP_PIO_LOG_SIZE, cap); 412 if (FIELD_GET(PCI_EXP_FLAGS_FLIT, flags)) 413 pdev->dpc_rp_log_size += FIELD_GET(PCI_EXP_DPC_RP_PIO_LOG_SIZE4, 414 cap) << 4; 415 416 if (pdev->dpc_rp_log_size < PCIE_STD_NUM_TLP_HEADERLOG || 417 pdev->dpc_rp_log_size > PCIE_STD_MAX_TLP_HEADERLOG + 1) { 418 pci_err(pdev, "RP PIO log size %u is invalid\n", 419 pdev->dpc_rp_log_size); 420 pdev->dpc_rp_log_size = 0; 421 } 422 } 423 } 424 425 static void dpc_enable(struct pcie_device *dev) 426 { 427 struct pci_dev *pdev = dev->port; 428 int dpc = pdev->dpc_cap; 429 u16 ctl; 430 431 /* 432 * Clear DPC Interrupt Status so we don't get an interrupt for an 433 * old event when setting DPC Interrupt Enable. 434 */ 435 pci_write_config_word(pdev, dpc + PCI_EXP_DPC_STATUS, 436 PCI_EXP_DPC_STATUS_INTERRUPT); 437 438 pci_read_config_word(pdev, dpc + PCI_EXP_DPC_CTL, &ctl); 439 ctl &= ~PCI_EXP_DPC_CTL_EN_MASK; 440 ctl |= PCI_EXP_DPC_CTL_EN_FATAL | PCI_EXP_DPC_CTL_INT_EN; 441 pci_write_config_word(pdev, dpc + PCI_EXP_DPC_CTL, ctl); 442 } 443 444 static void dpc_disable(struct pcie_device *dev) 445 { 446 struct pci_dev *pdev = dev->port; 447 int dpc = pdev->dpc_cap; 448 u16 ctl; 449 450 /* Disable DPC triggering and DPC interrupts */ 451 pci_read_config_word(pdev, dpc + PCI_EXP_DPC_CTL, &ctl); 452 ctl &= ~(PCI_EXP_DPC_CTL_EN_FATAL | PCI_EXP_DPC_CTL_INT_EN); 453 pci_write_config_word(pdev, dpc + PCI_EXP_DPC_CTL, ctl); 454 } 455 456 #define FLAG(x, y) (((x) & (y)) ? '+' : '-') 457 static int dpc_probe(struct pcie_device *dev) 458 { 459 struct pci_dev *pdev = dev->port; 460 struct device *device = &dev->device; 461 int status; 462 u16 cap; 463 464 if (!pcie_aer_is_native(pdev) && !pcie_ports_dpc_native) 465 return -ENOTSUPP; 466 467 status = devm_request_threaded_irq(device, dev->irq, dpc_irq, 468 dpc_handler, IRQF_SHARED, 469 "pcie-dpc", pdev); 470 if (status) { 471 pci_warn(pdev, "request IRQ%d failed: %d\n", dev->irq, 472 status); 473 return status; 474 } 475 476 pci_read_config_word(pdev, pdev->dpc_cap + PCI_EXP_DPC_CAP, &cap); 477 dpc_enable(dev); 478 479 pci_info(pdev, "enabled with IRQ %d\n", dev->irq); 480 pci_info(pdev, "error containment capabilities: Int Msg #%d, RPExt%c PoisonedTLP%c SwTrigger%c RP PIO Log %d, DL_ActiveErr%c\n", 481 cap & PCI_EXP_DPC_IRQ, FLAG(cap, PCI_EXP_DPC_CAP_RP_EXT), 482 FLAG(cap, PCI_EXP_DPC_CAP_POISONED_TLP), 483 FLAG(cap, PCI_EXP_DPC_CAP_SW_TRIGGER), pdev->dpc_rp_log_size, 484 FLAG(cap, PCI_EXP_DPC_CAP_DL_ACTIVE)); 485 486 pci_add_ext_cap_save_buffer(pdev, PCI_EXT_CAP_ID_DPC, sizeof(u16)); 487 return status; 488 } 489 490 static int dpc_suspend(struct pcie_device *dev) 491 { 492 dpc_disable(dev); 493 return 0; 494 } 495 496 static int dpc_resume(struct pcie_device *dev) 497 { 498 dpc_enable(dev); 499 return 0; 500 } 501 502 static void dpc_remove(struct pcie_device *dev) 503 { 504 dpc_disable(dev); 505 } 506 507 static struct pcie_port_service_driver dpcdriver = { 508 .name = "dpc", 509 .port_type = PCIE_ANY_PORT, 510 .service = PCIE_PORT_SERVICE_DPC, 511 .probe = dpc_probe, 512 .suspend = dpc_suspend, 513 .resume = dpc_resume, 514 .remove = dpc_remove, 515 }; 516 517 int __init pcie_dpc_init(void) 518 { 519 return pcie_port_service_register(&dpcdriver); 520 } 521