1/* 2 * PCI Error Recovery Driver for RPA-compliant PPC64 platform. 3 * Copyright IBM Corp. 2004 2005 4 * Copyright Linas Vepstas <linas@linas.org> 2004, 2005 5 * 6 * All rights reserved. 7 * 8 * This program is free software; you can redistribute it and/or modify 9 * it under the terms of the GNU General Public License as published by 10 * the Free Software Foundation; either version 2 of the License, or (at 11 * your option) any later version. 12 * 13 * This program is distributed in the hope that it will be useful, but 14 * WITHOUT ANY WARRANTY; without even the implied warranty of 15 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or 16 * NON INFRINGEMENT. See the GNU General Public License for more 17 * details. 18 * 19 * You should have received a copy of the GNU General Public License 20 * along with this program; if not, write to the Free Software 21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 22 * 23 * Send comments and feedback to Linas Vepstas <linas@austin.ibm.com> 24 */ 25#include <linux/delay.h> 26#include <linux/interrupt.h> 27#include <linux/irq.h> 28#include <linux/pci.h> 29#include <asm/eeh.h> 30#include <asm/eeh_event.h> 31#include <asm/ppc-pci.h> 32#include <asm/pci-bridge.h> 33#include <asm/prom.h> 34#include <asm/rtas.h> 35 36 37static inline const char * pcid_name (struct pci_dev *pdev) 38{ 39 if (pdev && pdev->dev.driver) 40 return pdev->dev.driver->name; 41 return ""; 42} 43 44 45/** 46 * eeh_disable_irq - disable interrupt for the recovering device 47 */ 48static void eeh_disable_irq(struct pci_dev *dev) 49{ 50 struct device_node *dn = pci_device_to_OF_node(dev); 51 52 /* Don't disable MSI and MSI-X interrupts. They are 53 * effectively disabled by the DMA Stopped state 54 * when an EEH error occurs. 55 */ 56 if (dev->msi_enabled || dev->msix_enabled) 57 return; 58 59 if (!irq_has_action(dev->irq)) 60 return; 61 62 PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED; 63 disable_irq_nosync(dev->irq); 64} 65 66/** 67 * eeh_enable_irq - enable interrupt for the recovering device 68 */ 69static void eeh_enable_irq(struct pci_dev *dev) 70{ 71 struct device_node *dn = pci_device_to_OF_node(dev); 72 73 if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) { 74 PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED; 75 enable_irq(dev->irq); 76 } 77} 78 79/* ------------------------------------------------------- */ 80/** 81 * eeh_report_error - report pci error to each device driver 82 * 83 * Report an EEH error to each device driver, collect up and 84 * merge the device driver responses. Cumulative response 85 * passed back in "userdata". 86 */ 87 88static int eeh_report_error(struct pci_dev *dev, void *userdata) 89{ 90 enum pci_ers_result rc, *res = userdata; 91 struct pci_driver *driver = dev->driver; 92 93 dev->error_state = pci_channel_io_frozen; 94 95 if (!driver) 96 return 0; 97 98 eeh_disable_irq(dev); 99 100 if (!driver->err_handler || 101 !driver->err_handler->error_detected) 102 return 0; 103 104 rc = driver->err_handler->error_detected (dev, pci_channel_io_frozen); 105 106 /* A driver that needs a reset trumps all others */ 107 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; 108 if (*res == PCI_ERS_RESULT_NONE) *res = rc; 109 110 return 0; 111} 112 113/** 114 * eeh_report_mmio_enabled - tell drivers that MMIO has been enabled 115 * 116 * Tells each device driver that IO ports, MMIO and config space I/O 117 * are now enabled. Collects up and merges the device driver responses. 118 * Cumulative response passed back in "userdata". 119 */ 120 121static int eeh_report_mmio_enabled(struct pci_dev *dev, void *userdata) 122{ 123 enum pci_ers_result rc, *res = userdata; 124 struct pci_driver *driver = dev->driver; 125 126 if (!driver || 127 !driver->err_handler || 128 !driver->err_handler->mmio_enabled) 129 return 0; 130 131 rc = driver->err_handler->mmio_enabled (dev); 132 133 /* A driver that needs a reset trumps all others */ 134 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; 135 if (*res == PCI_ERS_RESULT_NONE) *res = rc; 136 137 return 0; 138} 139 140/** 141 * eeh_report_reset - tell device that slot has been reset 142 */ 143 144static int eeh_report_reset(struct pci_dev *dev, void *userdata) 145{ 146 enum pci_ers_result rc, *res = userdata; 147 struct pci_driver *driver = dev->driver; 148 149 if (!driver) 150 return 0; 151 152 dev->error_state = pci_channel_io_normal; 153 154 eeh_enable_irq(dev); 155 156 if (!driver->err_handler || 157 !driver->err_handler->slot_reset) 158 return 0; 159 160 rc = driver->err_handler->slot_reset(dev); 161 if ((*res == PCI_ERS_RESULT_NONE) || 162 (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc; 163 if (*res == PCI_ERS_RESULT_DISCONNECT && 164 rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; 165 166 return 0; 167} 168 169/** 170 * eeh_report_resume - tell device to resume normal operations 171 */ 172 173static int eeh_report_resume(struct pci_dev *dev, void *userdata) 174{ 175 struct pci_driver *driver = dev->driver; 176 177 dev->error_state = pci_channel_io_normal; 178 179 if (!driver) 180 return 0; 181 182 eeh_enable_irq(dev); 183 184 if (!driver->err_handler || 185 !driver->err_handler->resume) 186 return 0; 187 188 driver->err_handler->resume(dev); 189 190 return 0; 191} 192 193/** 194 * eeh_report_failure - tell device driver that device is dead. 195 * 196 * This informs the device driver that the device is permanently 197 * dead, and that no further recovery attempts will be made on it. 198 */ 199 200static int eeh_report_failure(struct pci_dev *dev, void *userdata) 201{ 202 struct pci_driver *driver = dev->driver; 203 204 dev->error_state = pci_channel_io_perm_failure; 205 206 if (!driver) 207 return 0; 208 209 eeh_disable_irq(dev); 210 211 if (!driver->err_handler || 212 !driver->err_handler->error_detected) 213 return 0; 214 215 driver->err_handler->error_detected(dev, pci_channel_io_perm_failure); 216 217 return 0; 218} 219 220/* ------------------------------------------------------- */ 221/** 222 * handle_eeh_events -- reset a PCI device after hard lockup. 223 * 224 * pSeries systems will isolate a PCI slot if the PCI-Host 225 * bridge detects address or data parity errors, DMA's 226 * occurring to wild addresses (which usually happen due to 227 * bugs in device drivers or in PCI adapter firmware). 228 * Slot isolations also occur if #SERR, #PERR or other misc 229 * PCI-related errors are detected. 230 * 231 * Recovery process consists of unplugging the device driver 232 * (which generated hotplug events to userspace), then issuing 233 * a PCI #RST to the device, then reconfiguring the PCI config 234 * space for all bridges & devices under this slot, and then 235 * finally restarting the device drivers (which cause a second 236 * set of hotplug events to go out to userspace). 237 */ 238 239/** 240 * eeh_reset_device() -- perform actual reset of a pci slot 241 * @bus: pointer to the pci bus structure corresponding 242 * to the isolated slot. A non-null value will 243 * cause all devices under the bus to be removed 244 * and then re-added. 245 * @pe_dn: pointer to a "Partionable Endpoint" device node. 246 * This is the top-level structure on which pci 247 * bus resets can be performed. 248 */ 249 250static int eeh_reset_device (struct pci_dn *pe_dn, struct pci_bus *bus) 251{ 252 struct device_node *dn; 253 int cnt, rc; 254 255 /* pcibios will clear the counter; save the value */ 256 cnt = pe_dn->eeh_freeze_count; 257 258 if (bus) 259 pcibios_remove_pci_devices(bus); 260 261 /* Reset the pci controller. (Asserts RST#; resets config space). 262 * Reconfigure bridges and devices. Don't try to bring the system 263 * up if the reset failed for some reason. */ 264 rc = rtas_set_slot_reset(pe_dn); 265 if (rc) 266 return rc; 267 268 /* Walk over all functions on this device. */ 269 dn = pe_dn->node; 270 if (!pcibios_find_pci_bus(dn) && PCI_DN(dn->parent)) 271 dn = dn->parent->child; 272 273 while (dn) { 274 struct pci_dn *ppe = PCI_DN(dn); 275 /* On Power4, always true because eeh_pe_config_addr=0 */ 276 if (pe_dn->eeh_pe_config_addr == ppe->eeh_pe_config_addr) { 277 rtas_configure_bridge(ppe); 278 eeh_restore_bars(ppe); 279 } 280 dn = dn->sibling; 281 } 282 283 /* Give the system 5 seconds to finish running the user-space 284 * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes, 285 * this is a hack, but if we don't do this, and try to bring 286 * the device up before the scripts have taken it down, 287 * potentially weird things happen. 288 */ 289 if (bus) { 290 ssleep (5); 291 pcibios_add_pci_devices(bus); 292 } 293 pe_dn->eeh_freeze_count = cnt; 294 295 return 0; 296} 297 298/* The longest amount of time to wait for a pci device 299 * to come back on line, in seconds. 300 */ 301#define MAX_WAIT_FOR_RECOVERY 150 302 303struct pci_dn * handle_eeh_events (struct eeh_event *event) 304{ 305 struct device_node *frozen_dn; 306 struct pci_dn *frozen_pdn; 307 struct pci_bus *frozen_bus; 308 int rc = 0; 309 enum pci_ers_result result = PCI_ERS_RESULT_NONE; 310 const char *location, *pci_str, *drv_str; 311 312 frozen_dn = find_device_pe(event->dn); 313 if (!frozen_dn) { 314 315 location = of_get_property(event->dn, "ibm,loc-code", NULL); 316 location = location ? location : "unknown"; 317 printk(KERN_ERR "EEH: Error: Cannot find partition endpoint " 318 "for location=%s pci addr=%s\n", 319 location, eeh_pci_name(event->dev)); 320 return NULL; 321 } 322 323 frozen_bus = pcibios_find_pci_bus(frozen_dn); 324 location = of_get_property(frozen_dn, "ibm,loc-code", NULL); 325 location = location ? location : "unknown"; 326 327 /* There are two different styles for coming up with the PE. 328 * In the old style, it was the highest EEH-capable device 329 * which was always an EADS pci bridge. In the new style, 330 * there might not be any EADS bridges, and even when there are, 331 * the firmware marks them as "EEH incapable". So another 332 * two-step is needed to find the pci bus.. */ 333 if (!frozen_bus) 334 frozen_bus = pcibios_find_pci_bus (frozen_dn->parent); 335 336 if (!frozen_bus) { 337 printk(KERN_ERR "EEH: Cannot find PCI bus " 338 "for location=%s dn=%s\n", 339 location, frozen_dn->full_name); 340 return NULL; 341 } 342 343 frozen_pdn = PCI_DN(frozen_dn); 344 frozen_pdn->eeh_freeze_count++; 345 346 if (frozen_pdn->pcidev) { 347 pci_str = pci_name (frozen_pdn->pcidev); 348 drv_str = pcid_name (frozen_pdn->pcidev); 349 } else { 350 pci_str = eeh_pci_name(event->dev); 351 drv_str = pcid_name (event->dev); 352 } 353 354 if (frozen_pdn->eeh_freeze_count > EEH_MAX_ALLOWED_FREEZES) 355 goto excess_failures; 356 357 printk(KERN_WARNING 358 "EEH: This PCI device has failed %d times in the last hour:\n", 359 frozen_pdn->eeh_freeze_count); 360 printk(KERN_WARNING 361 "EEH: location=%s driver=%s pci addr=%s\n", 362 location, drv_str, pci_str); 363 364 /* Walk the various device drivers attached to this slot through 365 * a reset sequence, giving each an opportunity to do what it needs 366 * to accomplish the reset. Each child gets a report of the 367 * status ... if any child can't handle the reset, then the entire 368 * slot is dlpar removed and added. 369 */ 370 pci_walk_bus(frozen_bus, eeh_report_error, &result); 371 372 /* Get the current PCI slot state. This can take a long time, 373 * sometimes over 3 seconds for certain systems. */ 374 rc = eeh_wait_for_slot_status (frozen_pdn, MAX_WAIT_FOR_RECOVERY*1000); 375 if (rc < 0) { 376 printk(KERN_WARNING "EEH: Permanent failure\n"); 377 goto hard_fail; 378 } 379 380 /* Since rtas may enable MMIO when posting the error log, 381 * don't post the error log until after all dev drivers 382 * have been informed. 383 */ 384 eeh_slot_error_detail(frozen_pdn, EEH_LOG_TEMP_FAILURE); 385 386 /* If all device drivers were EEH-unaware, then shut 387 * down all of the device drivers, and hope they 388 * go down willingly, without panicing the system. 389 */ 390 if (result == PCI_ERS_RESULT_NONE) { 391 rc = eeh_reset_device(frozen_pdn, frozen_bus); 392 if (rc) { 393 printk(KERN_WARNING "EEH: Unable to reset, rc=%d\n", rc); 394 goto hard_fail; 395 } 396 } 397 398 /* If all devices reported they can proceed, then re-enable MMIO */ 399 if (result == PCI_ERS_RESULT_CAN_RECOVER) { 400 rc = rtas_pci_enable(frozen_pdn, EEH_THAW_MMIO); 401 402 if (rc < 0) 403 goto hard_fail; 404 if (rc) { 405 result = PCI_ERS_RESULT_NEED_RESET; 406 } else { 407 result = PCI_ERS_RESULT_NONE; 408 pci_walk_bus(frozen_bus, eeh_report_mmio_enabled, &result); 409 } 410 } 411 412 /* If all devices reported they can proceed, then re-enable DMA */ 413 if (result == PCI_ERS_RESULT_CAN_RECOVER) { 414 rc = rtas_pci_enable(frozen_pdn, EEH_THAW_DMA); 415 416 if (rc < 0) 417 goto hard_fail; 418 if (rc) 419 result = PCI_ERS_RESULT_NEED_RESET; 420 else 421 result = PCI_ERS_RESULT_RECOVERED; 422 } 423 424 /* If any device has a hard failure, then shut off everything. */ 425 if (result == PCI_ERS_RESULT_DISCONNECT) { 426 printk(KERN_WARNING "EEH: Device driver gave up\n"); 427 goto hard_fail; 428 } 429 430 /* If any device called out for a reset, then reset the slot */ 431 if (result == PCI_ERS_RESULT_NEED_RESET) { 432 rc = eeh_reset_device(frozen_pdn, NULL); 433 if (rc) { 434 printk(KERN_WARNING "EEH: Cannot reset, rc=%d\n", rc); 435 goto hard_fail; 436 } 437 result = PCI_ERS_RESULT_NONE; 438 pci_walk_bus(frozen_bus, eeh_report_reset, &result); 439 } 440 441 /* All devices should claim they have recovered by now. */ 442 if ((result != PCI_ERS_RESULT_RECOVERED) && 443 (result != PCI_ERS_RESULT_NONE)) { 444 printk(KERN_WARNING "EEH: Not recovered\n"); 445 goto hard_fail; 446 } 447 448 /* Tell all device drivers that they can resume operations */ 449 pci_walk_bus(frozen_bus, eeh_report_resume, NULL); 450 451 return frozen_pdn; 452 453excess_failures: 454 /* 455 * About 90% of all real-life EEH failures in the field 456 * are due to poorly seated PCI cards. Only 10% or so are 457 * due to actual, failed cards. 458 */ 459 printk(KERN_ERR 460 "EEH: PCI device at location=%s driver=%s pci addr=%s\n" 461 "has failed %d times in the last hour " 462 "and has been permanently disabled.\n" 463 "Please try reseating this device or replacing it.\n", 464 location, drv_str, pci_str, frozen_pdn->eeh_freeze_count); 465 goto perm_error; 466 467hard_fail: 468 printk(KERN_ERR 469 "EEH: Unable to recover from failure of PCI device " 470 "at location=%s driver=%s pci addr=%s\n" 471 "Please try reseating this device or replacing it.\n", 472 location, drv_str, pci_str); 473 474perm_error: 475 eeh_slot_error_detail(frozen_pdn, EEH_LOG_PERM_FAILURE); 476 477 /* Notify all devices that they're about to go down. */ 478 pci_walk_bus(frozen_bus, eeh_report_failure, NULL); 479 480 /* Shut down the device drivers for good. */ 481 pcibios_remove_pci_devices(frozen_bus); 482 483 return NULL; 484} 485 486/* ---------- end of file ---------- */ 487