Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 1 | /* |
| 2 | * PCI Error Recovery Driver for RPA-compliant PPC64 platform. |
| 3 | * Copyright (C) 2004, 2005 Linas Vepstas <linas@linas.org> |
| 4 | * |
| 5 | * All rights reserved. |
| 6 | * |
| 7 | * This program is free software; you can redistribute it and/or modify |
| 8 | * it under the terms of the GNU General Public License as published by |
| 9 | * the Free Software Foundation; either version 2 of the License, or (at |
| 10 | * your option) any later version. |
| 11 | * |
| 12 | * This program is distributed in the hope that it will be useful, but |
| 13 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 | * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or |
| 15 | * NON INFRINGEMENT. See the GNU General Public License for more |
| 16 | * details. |
| 17 | * |
| 18 | * You should have received a copy of the GNU General Public License |
| 19 | * along with this program; if not, write to the Free Software |
| 20 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
| 21 | * |
| 22 | * Send feedback to <linas@us.ibm.com> |
| 23 | * |
| 24 | */ |
| 25 | #include <linux/delay.h> |
| 26 | #include <linux/irq.h> |
| 27 | #include <linux/interrupt.h> |
| 28 | #include <linux/notifier.h> |
| 29 | #include <linux/pci.h> |
| 30 | #include <asm/eeh.h> |
| 31 | #include <asm/eeh_event.h> |
| 32 | #include <asm/ppc-pci.h> |
| 33 | #include <asm/pci-bridge.h> |
| 34 | #include <asm/prom.h> |
| 35 | #include <asm/rtas.h> |
| 36 | |
| 37 | |
| 38 | static inline const char * pcid_name (struct pci_dev *pdev) |
| 39 | { |
| 40 | if (pdev->dev.driver) |
| 41 | return pdev->dev.driver->name; |
| 42 | return ""; |
| 43 | } |
| 44 | |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 45 | #ifdef DEBUG |
| 46 | static void print_device_node_tree (struct pci_dn *pdn, int dent) |
| 47 | { |
| 48 | int i; |
| 49 | if (!pdn) return; |
| 50 | for (i=0;i<dent; i++) |
| 51 | printk(" "); |
| 52 | printk("dn=%s mode=%x \tcfg_addr=%x pe_addr=%x \tfull=%s\n", |
| 53 | pdn->node->name, pdn->eeh_mode, pdn->eeh_config_addr, |
| 54 | pdn->eeh_pe_config_addr, pdn->node->full_name); |
| 55 | dent += 3; |
| 56 | struct device_node *pc = pdn->node->child; |
| 57 | while (pc) { |
| 58 | print_device_node_tree(PCI_DN(pc), dent); |
| 59 | pc = pc->sibling; |
| 60 | } |
| 61 | } |
| 62 | #endif |
| 63 | |
| 64 | /** |
| 65 | * irq_in_use - return true if this irq is being used |
| 66 | */ |
| 67 | static int irq_in_use(unsigned int irq) |
| 68 | { |
| 69 | int rc = 0; |
| 70 | unsigned long flags; |
| 71 | struct irq_desc *desc = irq_desc + irq; |
| 72 | |
| 73 | spin_lock_irqsave(&desc->lock, flags); |
| 74 | if (desc->action) |
| 75 | rc = 1; |
| 76 | spin_unlock_irqrestore(&desc->lock, flags); |
| 77 | return rc; |
| 78 | } |
| 79 | |
| 80 | /* ------------------------------------------------------- */ |
| 81 | /** eeh_report_error - report an EEH error to each device, |
| 82 | * collect up and merge the device responses. |
| 83 | */ |
| 84 | |
| 85 | static void eeh_report_error(struct pci_dev *dev, void *userdata) |
| 86 | { |
Paul Mackerras | 18eb3b3 | 2005-11-29 17:17:02 +1100 | [diff] [blame] | 87 | enum pci_ers_result rc, *res = userdata; |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 88 | struct pci_driver *driver = dev->driver; |
| 89 | |
| 90 | dev->error_state = pci_channel_io_frozen; |
| 91 | |
| 92 | if (!driver) |
| 93 | return; |
| 94 | |
| 95 | if (irq_in_use (dev->irq)) { |
| 96 | struct device_node *dn = pci_device_to_OF_node(dev); |
| 97 | PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED; |
| 98 | disable_irq_nosync(dev->irq); |
| 99 | } |
| 100 | if (!driver->err_handler) |
| 101 | return; |
| 102 | if (!driver->err_handler->error_detected) |
| 103 | return; |
| 104 | |
| 105 | rc = driver->err_handler->error_detected (dev, pci_channel_io_frozen); |
Paul Mackerras | 18eb3b3 | 2005-11-29 17:17:02 +1100 | [diff] [blame] | 106 | if (*res == PCI_ERS_RESULT_NONE) *res = rc; |
| 107 | if (*res == PCI_ERS_RESULT_NEED_RESET) return; |
| 108 | if (*res == PCI_ERS_RESULT_DISCONNECT && |
| 109 | rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 110 | } |
| 111 | |
| 112 | /** eeh_report_reset -- tell this device that the pci slot |
| 113 | * has been reset. |
| 114 | */ |
| 115 | |
| 116 | static void eeh_report_reset(struct pci_dev *dev, void *userdata) |
| 117 | { |
| 118 | struct pci_driver *driver = dev->driver; |
| 119 | struct device_node *dn = pci_device_to_OF_node(dev); |
| 120 | |
| 121 | if (!driver) |
| 122 | return; |
| 123 | |
| 124 | if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) { |
| 125 | PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED; |
| 126 | enable_irq(dev->irq); |
| 127 | } |
| 128 | if (!driver->err_handler) |
| 129 | return; |
| 130 | if (!driver->err_handler->slot_reset) |
| 131 | return; |
| 132 | |
| 133 | driver->err_handler->slot_reset(dev); |
| 134 | } |
| 135 | |
| 136 | static void eeh_report_resume(struct pci_dev *dev, void *userdata) |
| 137 | { |
| 138 | struct pci_driver *driver = dev->driver; |
| 139 | |
| 140 | dev->error_state = pci_channel_io_normal; |
| 141 | |
| 142 | if (!driver) |
| 143 | return; |
| 144 | if (!driver->err_handler) |
| 145 | return; |
| 146 | if (!driver->err_handler->resume) |
| 147 | return; |
| 148 | |
| 149 | driver->err_handler->resume(dev); |
| 150 | } |
| 151 | |
| 152 | static void eeh_report_failure(struct pci_dev *dev, void *userdata) |
| 153 | { |
| 154 | struct pci_driver *driver = dev->driver; |
| 155 | |
| 156 | dev->error_state = pci_channel_io_perm_failure; |
| 157 | |
| 158 | if (!driver) |
| 159 | return; |
| 160 | |
| 161 | if (irq_in_use (dev->irq)) { |
| 162 | struct device_node *dn = pci_device_to_OF_node(dev); |
| 163 | PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED; |
| 164 | disable_irq_nosync(dev->irq); |
| 165 | } |
| 166 | if (!driver->err_handler) |
| 167 | return; |
| 168 | if (!driver->err_handler->error_detected) |
| 169 | return; |
| 170 | driver->err_handler->error_detected(dev, pci_channel_io_perm_failure); |
| 171 | } |
| 172 | |
| 173 | /* ------------------------------------------------------- */ |
| 174 | /** |
| 175 | * handle_eeh_events -- reset a PCI device after hard lockup. |
| 176 | * |
| 177 | * pSeries systems will isolate a PCI slot if the PCI-Host |
| 178 | * bridge detects address or data parity errors, DMA's |
| 179 | * occuring to wild addresses (which usually happen due to |
| 180 | * bugs in device drivers or in PCI adapter firmware). |
| 181 | * Slot isolations also occur if #SERR, #PERR or other misc |
| 182 | * PCI-related errors are detected. |
| 183 | * |
| 184 | * Recovery process consists of unplugging the device driver |
| 185 | * (which generated hotplug events to userspace), then issuing |
| 186 | * a PCI #RST to the device, then reconfiguring the PCI config |
| 187 | * space for all bridges & devices under this slot, and then |
| 188 | * finally restarting the device drivers (which cause a second |
| 189 | * set of hotplug events to go out to userspace). |
| 190 | */ |
| 191 | |
| 192 | /** |
| 193 | * eeh_reset_device() -- perform actual reset of a pci slot |
| 194 | * Args: bus: pointer to the pci bus structure corresponding |
| 195 | * to the isolated slot. A non-null value will |
| 196 | * cause all devices under the bus to be removed |
| 197 | * and then re-added. |
| 198 | * pe_dn: pointer to a "Partionable Endpoint" device node. |
| 199 | * This is the top-level structure on which pci |
| 200 | * bus resets can be performed. |
| 201 | */ |
| 202 | |
Linas Vepstas | b6495c0 | 2005-11-03 18:54:54 -0600 | [diff] [blame] | 203 | static int eeh_reset_device (struct pci_dn *pe_dn, struct pci_bus *bus) |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 204 | { |
Linas Vepstas | b6495c0 | 2005-11-03 18:54:54 -0600 | [diff] [blame] | 205 | int rc; |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 206 | if (bus) |
| 207 | pcibios_remove_pci_devices(bus); |
| 208 | |
| 209 | /* Reset the pci controller. (Asserts RST#; resets config space). |
Linas Vepstas | b6495c0 | 2005-11-03 18:54:54 -0600 | [diff] [blame] | 210 | * Reconfigure bridges and devices. Don't try to bring the system |
| 211 | * up if the reset failed for some reason. */ |
| 212 | rc = rtas_set_slot_reset(pe_dn); |
| 213 | if (rc) |
| 214 | return rc; |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 215 | |
Linas Vepstas | 3914ac7 | 2005-11-03 18:55:01 -0600 | [diff] [blame] | 216 | /* New-style config addrs might be shared across multiple devices, |
| 217 | * Walk over all functions on this device */ |
| 218 | if (pe_dn->eeh_pe_config_addr) { |
| 219 | struct device_node *pe = pe_dn->node; |
| 220 | pe = pe->parent->child; |
| 221 | while (pe) { |
| 222 | struct pci_dn *ppe = PCI_DN(pe); |
| 223 | if (pe_dn->eeh_pe_config_addr == ppe->eeh_pe_config_addr) { |
| 224 | rtas_configure_bridge(ppe); |
| 225 | eeh_restore_bars(ppe); |
| 226 | } |
| 227 | pe = pe->sibling; |
| 228 | } |
| 229 | } else { |
| 230 | rtas_configure_bridge(pe_dn); |
| 231 | eeh_restore_bars(pe_dn); |
| 232 | } |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 233 | |
| 234 | /* Give the system 5 seconds to finish running the user-space |
| 235 | * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes, |
| 236 | * this is a hack, but if we don't do this, and try to bring |
| 237 | * the device up before the scripts have taken it down, |
| 238 | * potentially weird things happen. |
| 239 | */ |
| 240 | if (bus) { |
| 241 | ssleep (5); |
| 242 | pcibios_add_pci_devices(bus); |
| 243 | } |
Linas Vepstas | b6495c0 | 2005-11-03 18:54:54 -0600 | [diff] [blame] | 244 | |
| 245 | return 0; |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 246 | } |
| 247 | |
| 248 | /* The longest amount of time to wait for a pci device |
| 249 | * to come back on line, in seconds. |
| 250 | */ |
| 251 | #define MAX_WAIT_FOR_RECOVERY 15 |
| 252 | |
| 253 | void handle_eeh_events (struct eeh_event *event) |
| 254 | { |
| 255 | struct device_node *frozen_dn; |
| 256 | struct pci_dn *frozen_pdn; |
| 257 | struct pci_bus *frozen_bus; |
Linas Vepstas | b6495c0 | 2005-11-03 18:54:54 -0600 | [diff] [blame] | 258 | int rc = 0; |
Paul Mackerras | 18eb3b3 | 2005-11-29 17:17:02 +1100 | [diff] [blame] | 259 | enum pci_ers_result result = PCI_ERS_RESULT_NONE; |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 260 | |
| 261 | frozen_dn = find_device_pe(event->dn); |
| 262 | frozen_bus = pcibios_find_pci_bus(frozen_dn); |
| 263 | |
| 264 | if (!frozen_dn) { |
| 265 | printk(KERN_ERR "EEH: Error: Cannot find partition endpoint for %s\n", |
| 266 | pci_name(event->dev)); |
| 267 | return; |
| 268 | } |
| 269 | |
| 270 | /* There are two different styles for coming up with the PE. |
| 271 | * In the old style, it was the highest EEH-capable device |
| 272 | * which was always an EADS pci bridge. In the new style, |
| 273 | * there might not be any EADS bridges, and even when there are, |
| 274 | * the firmware marks them as "EEH incapable". So another |
| 275 | * two-step is needed to find the pci bus.. */ |
| 276 | if (!frozen_bus) |
| 277 | frozen_bus = pcibios_find_pci_bus (frozen_dn->parent); |
| 278 | |
| 279 | if (!frozen_bus) { |
| 280 | printk(KERN_ERR "EEH: Cannot find PCI bus for %s\n", |
| 281 | frozen_dn->full_name); |
| 282 | return; |
| 283 | } |
| 284 | |
| 285 | #if 0 |
| 286 | /* We may get "permanent failure" messages on empty slots. |
| 287 | * These are false alarms. Empty slots have no child dn. */ |
| 288 | if ((event->state == pci_channel_io_perm_failure) && (frozen_device == NULL)) |
| 289 | return; |
| 290 | #endif |
| 291 | |
| 292 | frozen_pdn = PCI_DN(frozen_dn); |
| 293 | frozen_pdn->eeh_freeze_count++; |
| 294 | |
| 295 | if (frozen_pdn->eeh_freeze_count > EEH_MAX_ALLOWED_FREEZES) |
Linas Vepstas | b6495c0 | 2005-11-03 18:54:54 -0600 | [diff] [blame] | 296 | goto hard_fail; |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 297 | |
| 298 | /* If the reset state is a '5' and the time to reset is 0 (infinity) |
| 299 | * or is more then 15 seconds, then mark this as a permanent failure. |
| 300 | */ |
| 301 | if ((event->state == pci_channel_io_perm_failure) && |
| 302 | ((event->time_unavail <= 0) || |
| 303 | (event->time_unavail > MAX_WAIT_FOR_RECOVERY*1000))) |
Linas Vepstas | b6495c0 | 2005-11-03 18:54:54 -0600 | [diff] [blame] | 304 | goto hard_fail; |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 305 | |
| 306 | eeh_slot_error_detail(frozen_pdn, 1 /* Temporary Error */); |
| 307 | printk(KERN_WARNING |
| 308 | "EEH: This PCI device has failed %d times since last reboot: %s - %s\n", |
| 309 | frozen_pdn->eeh_freeze_count, |
| 310 | pci_name (frozen_pdn->pcidev), |
| 311 | pcid_name(frozen_pdn->pcidev)); |
| 312 | |
| 313 | /* Walk the various device drivers attached to this slot through |
| 314 | * a reset sequence, giving each an opportunity to do what it needs |
| 315 | * to accomplish the reset. Each child gets a report of the |
| 316 | * status ... if any child can't handle the reset, then the entire |
| 317 | * slot is dlpar removed and added. |
| 318 | */ |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 319 | pci_walk_bus(frozen_bus, eeh_report_error, &result); |
| 320 | |
| 321 | /* If all device drivers were EEH-unaware, then shut |
| 322 | * down all of the device drivers, and hope they |
| 323 | * go down willingly, without panicing the system. |
| 324 | */ |
Paul Mackerras | 18eb3b3 | 2005-11-29 17:17:02 +1100 | [diff] [blame] | 325 | if (result == PCI_ERS_RESULT_NONE) { |
Linas Vepstas | b6495c0 | 2005-11-03 18:54:54 -0600 | [diff] [blame] | 326 | rc = eeh_reset_device(frozen_pdn, frozen_bus); |
| 327 | if (rc) |
| 328 | goto hard_fail; |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 329 | } |
| 330 | |
| 331 | /* If any device called out for a reset, then reset the slot */ |
Paul Mackerras | 18eb3b3 | 2005-11-29 17:17:02 +1100 | [diff] [blame] | 332 | if (result == PCI_ERS_RESULT_NEED_RESET) { |
Linas Vepstas | b6495c0 | 2005-11-03 18:54:54 -0600 | [diff] [blame] | 333 | rc = eeh_reset_device(frozen_pdn, NULL); |
| 334 | if (rc) |
| 335 | goto hard_fail; |
Al Viro | d04e4e1 | 2006-02-01 07:23:24 -0500 | [diff] [blame^] | 336 | pci_walk_bus(frozen_bus, eeh_report_reset, NULL); |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 337 | } |
| 338 | |
| 339 | /* If all devices reported they can proceed, the re-enable PIO */ |
Paul Mackerras | 18eb3b3 | 2005-11-29 17:17:02 +1100 | [diff] [blame] | 340 | if (result == PCI_ERS_RESULT_CAN_RECOVER) { |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 341 | /* XXX Not supported; we brute-force reset the device */ |
Linas Vepstas | b6495c0 | 2005-11-03 18:54:54 -0600 | [diff] [blame] | 342 | rc = eeh_reset_device(frozen_pdn, NULL); |
| 343 | if (rc) |
| 344 | goto hard_fail; |
Al Viro | d04e4e1 | 2006-02-01 07:23:24 -0500 | [diff] [blame^] | 345 | pci_walk_bus(frozen_bus, eeh_report_reset, NULL); |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 346 | } |
| 347 | |
| 348 | /* Tell all device drivers that they can resume operations */ |
Al Viro | d04e4e1 | 2006-02-01 07:23:24 -0500 | [diff] [blame^] | 349 | pci_walk_bus(frozen_bus, eeh_report_resume, NULL); |
Linas Vepstas | b6495c0 | 2005-11-03 18:54:54 -0600 | [diff] [blame] | 350 | |
| 351 | return; |
| 352 | |
| 353 | hard_fail: |
| 354 | /* |
| 355 | * About 90% of all real-life EEH failures in the field |
| 356 | * are due to poorly seated PCI cards. Only 10% or so are |
| 357 | * due to actual, failed cards. |
| 358 | */ |
| 359 | printk(KERN_ERR |
| 360 | "EEH: PCI device %s - %s has failed %d times \n" |
| 361 | "and has been permanently disabled. Please try reseating\n" |
| 362 | "this device or replacing it.\n", |
| 363 | pci_name (frozen_pdn->pcidev), |
| 364 | pcid_name(frozen_pdn->pcidev), |
| 365 | frozen_pdn->eeh_freeze_count); |
| 366 | |
| 367 | eeh_slot_error_detail(frozen_pdn, 2 /* Permanent Error */); |
| 368 | |
| 369 | /* Notify all devices that they're about to go down. */ |
Al Viro | d04e4e1 | 2006-02-01 07:23:24 -0500 | [diff] [blame^] | 370 | pci_walk_bus(frozen_bus, eeh_report_failure, NULL); |
Linas Vepstas | b6495c0 | 2005-11-03 18:54:54 -0600 | [diff] [blame] | 371 | |
| 372 | /* Shut down the device drivers for good. */ |
| 373 | pcibios_remove_pci_devices(frozen_bus); |
Linas Vepstas | 77bd741 | 2005-11-03 18:52:49 -0600 | [diff] [blame] | 374 | } |
| 375 | |
| 376 | /* ---------- end of file ---------- */ |