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
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.
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
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.
23 * Send comments and feedback to Linas Vepstas <linas@austin.ibm.com>
25 #include <linux/delay.h>
26 #include <linux/interrupt.h>
27 #include <linux/irq.h>
28 #include <linux/pci.h>
30 #include <asm/eeh_event.h>
31 #include <asm/ppc-pci.h>
32 #include <asm/pci-bridge.h>
37 static inline const char * pcid_name (struct pci_dev *pdev)
39 if (pdev && pdev->dev.driver)
40 return pdev->dev.driver->name;
45 static void print_device_node_tree(struct pci_dn *pdn, int dent)
48 struct device_node *pc;
52 for (i = 0; i < dent; i++)
54 printk("dn=%s mode=%x \tcfg_addr=%x pe_addr=%x \tfull=%s\n",
55 pdn->node->name, pdn->eeh_mode, pdn->eeh_config_addr,
56 pdn->eeh_pe_config_addr, pdn->node->full_name);
58 pc = pdn->node->child;
60 print_device_node_tree(PCI_DN(pc), dent);
67 * irq_in_use - return true if this irq is being used
69 static int irq_in_use(unsigned int irq)
73 struct irq_desc *desc = irq_desc + irq;
75 spin_lock_irqsave(&desc->lock, flags);
78 spin_unlock_irqrestore(&desc->lock, flags);
83 * eeh_disable_irq - disable interrupt for the recovering device
85 static void eeh_disable_irq(struct pci_dev *dev)
87 struct device_node *dn = pci_device_to_OF_node(dev);
89 /* Don't disable MSI and MSI-X interrupts. They are
90 * effectively disabled by the DMA Stopped state
91 * when an EEH error occurs.
93 if (dev->msi_enabled || dev->msix_enabled)
96 if (!irq_in_use(dev->irq))
99 PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED;
100 disable_irq_nosync(dev->irq);
104 * eeh_enable_irq - enable interrupt for the recovering device
106 static void eeh_enable_irq(struct pci_dev *dev)
108 struct device_node *dn = pci_device_to_OF_node(dev);
110 if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) {
111 PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED;
112 enable_irq(dev->irq);
116 /* ------------------------------------------------------- */
118 * eeh_report_error - report pci error to each device driver
120 * Report an EEH error to each device driver, collect up and
121 * merge the device driver responses. Cumulative response
122 * passed back in "userdata".
125 static void eeh_report_error(struct pci_dev *dev, void *userdata)
127 enum pci_ers_result rc, *res = userdata;
128 struct pci_driver *driver = dev->driver;
130 dev->error_state = pci_channel_io_frozen;
135 eeh_disable_irq(dev);
137 if (!driver->err_handler ||
138 !driver->err_handler->error_detected)
141 rc = driver->err_handler->error_detected (dev, pci_channel_io_frozen);
143 /* A driver that needs a reset trumps all others */
144 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
145 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
149 * eeh_report_mmio_enabled - tell drivers that MMIO has been enabled
151 * Tells each device driver that IO ports, MMIO and config space I/O
152 * are now enabled. Collects up and merges the device driver responses.
153 * Cumulative response passed back in "userdata".
156 static void eeh_report_mmio_enabled(struct pci_dev *dev, void *userdata)
158 enum pci_ers_result rc, *res = userdata;
159 struct pci_driver *driver = dev->driver;
162 !driver->err_handler ||
163 !driver->err_handler->mmio_enabled)
166 rc = driver->err_handler->mmio_enabled (dev);
168 /* A driver that needs a reset trumps all others */
169 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
170 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
174 * eeh_report_reset - tell device that slot has been reset
177 static void eeh_report_reset(struct pci_dev *dev, void *userdata)
179 enum pci_ers_result rc, *res = userdata;
180 struct pci_driver *driver = dev->driver;
185 dev->error_state = pci_channel_io_normal;
189 if (!driver->err_handler ||
190 !driver->err_handler->slot_reset)
193 rc = driver->err_handler->slot_reset(dev);
194 if ((*res == PCI_ERS_RESULT_NONE) ||
195 (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc;
196 if (*res == PCI_ERS_RESULT_DISCONNECT &&
197 rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
201 * eeh_report_resume - tell device to resume normal operations
204 static void eeh_report_resume(struct pci_dev *dev, void *userdata)
206 struct pci_driver *driver = dev->driver;
208 dev->error_state = pci_channel_io_normal;
215 if (!driver->err_handler ||
216 !driver->err_handler->resume)
219 driver->err_handler->resume(dev);
223 * eeh_report_failure - tell device driver that device is dead.
225 * This informs the device driver that the device is permanently
226 * dead, and that no further recovery attempts will be made on it.
229 static void eeh_report_failure(struct pci_dev *dev, void *userdata)
231 struct pci_driver *driver = dev->driver;
233 dev->error_state = pci_channel_io_perm_failure;
238 eeh_disable_irq(dev);
240 if (!driver->err_handler ||
241 !driver->err_handler->error_detected)
244 driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
247 /* ------------------------------------------------------- */
249 * handle_eeh_events -- reset a PCI device after hard lockup.
251 * pSeries systems will isolate a PCI slot if the PCI-Host
252 * bridge detects address or data parity errors, DMA's
253 * occurring to wild addresses (which usually happen due to
254 * bugs in device drivers or in PCI adapter firmware).
255 * Slot isolations also occur if #SERR, #PERR or other misc
256 * PCI-related errors are detected.
258 * Recovery process consists of unplugging the device driver
259 * (which generated hotplug events to userspace), then issuing
260 * a PCI #RST to the device, then reconfiguring the PCI config
261 * space for all bridges & devices under this slot, and then
262 * finally restarting the device drivers (which cause a second
263 * set of hotplug events to go out to userspace).
267 * eeh_reset_device() -- perform actual reset of a pci slot
268 * @bus: pointer to the pci bus structure corresponding
269 * to the isolated slot. A non-null value will
270 * cause all devices under the bus to be removed
272 * @pe_dn: pointer to a "Partionable Endpoint" device node.
273 * This is the top-level structure on which pci
274 * bus resets can be performed.
277 static int eeh_reset_device (struct pci_dn *pe_dn, struct pci_bus *bus)
279 struct device_node *dn;
282 /* pcibios will clear the counter; save the value */
283 cnt = pe_dn->eeh_freeze_count;
286 pcibios_remove_pci_devices(bus);
288 /* Reset the pci controller. (Asserts RST#; resets config space).
289 * Reconfigure bridges and devices. Don't try to bring the system
290 * up if the reset failed for some reason. */
291 rc = rtas_set_slot_reset(pe_dn);
295 /* Walk over all functions on this device. */
297 if (!pcibios_find_pci_bus(dn) && PCI_DN(dn->parent))
298 dn = dn->parent->child;
301 struct pci_dn *ppe = PCI_DN(dn);
302 /* On Power4, always true because eeh_pe_config_addr=0 */
303 if (pe_dn->eeh_pe_config_addr == ppe->eeh_pe_config_addr) {
304 rtas_configure_bridge(ppe);
305 eeh_restore_bars(ppe);
310 /* Give the system 5 seconds to finish running the user-space
311 * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes,
312 * this is a hack, but if we don't do this, and try to bring
313 * the device up before the scripts have taken it down,
314 * potentially weird things happen.
318 pcibios_add_pci_devices(bus);
320 pe_dn->eeh_freeze_count = cnt;
325 /* The longest amount of time to wait for a pci device
326 * to come back on line, in seconds.
328 #define MAX_WAIT_FOR_RECOVERY 150
330 struct pci_dn * handle_eeh_events (struct eeh_event *event)
332 struct device_node *frozen_dn;
333 struct pci_dn *frozen_pdn;
334 struct pci_bus *frozen_bus;
336 enum pci_ers_result result = PCI_ERS_RESULT_NONE;
337 const char *location, *pci_str, *drv_str;
339 frozen_dn = find_device_pe(event->dn);
342 location = of_get_property(event->dn, "ibm,loc-code", NULL);
343 location = location ? location : "unknown";
344 printk(KERN_ERR "EEH: Error: Cannot find partition endpoint "
345 "for location=%s pci addr=%s\n",
346 location, pci_name(event->dev));
350 frozen_bus = pcibios_find_pci_bus(frozen_dn);
351 location = of_get_property(frozen_dn, "ibm,loc-code", NULL);
352 location = location ? location : "unknown";
354 /* There are two different styles for coming up with the PE.
355 * In the old style, it was the highest EEH-capable device
356 * which was always an EADS pci bridge. In the new style,
357 * there might not be any EADS bridges, and even when there are,
358 * the firmware marks them as "EEH incapable". So another
359 * two-step is needed to find the pci bus.. */
361 frozen_bus = pcibios_find_pci_bus (frozen_dn->parent);
364 printk(KERN_ERR "EEH: Cannot find PCI bus "
365 "for location=%s dn=%s\n",
366 location, frozen_dn->full_name);
370 frozen_pdn = PCI_DN(frozen_dn);
371 frozen_pdn->eeh_freeze_count++;
373 if (frozen_pdn->pcidev) {
374 pci_str = pci_name (frozen_pdn->pcidev);
375 drv_str = pcid_name (frozen_pdn->pcidev);
377 pci_str = pci_name (event->dev);
378 drv_str = pcid_name (event->dev);
381 if (frozen_pdn->eeh_freeze_count > EEH_MAX_ALLOWED_FREEZES)
382 goto excess_failures;
385 "EEH: This PCI device has failed %d times in the last hour:\n",
386 frozen_pdn->eeh_freeze_count);
388 "EEH: location=%s driver=%s pci addr=%s\n",
389 location, drv_str, pci_str);
391 /* Walk the various device drivers attached to this slot through
392 * a reset sequence, giving each an opportunity to do what it needs
393 * to accomplish the reset. Each child gets a report of the
394 * status ... if any child can't handle the reset, then the entire
395 * slot is dlpar removed and added.
397 pci_walk_bus(frozen_bus, eeh_report_error, &result);
399 /* Get the current PCI slot state. This can take a long time,
400 * sometimes over 3 seconds for certain systems. */
401 rc = eeh_wait_for_slot_status (frozen_pdn, MAX_WAIT_FOR_RECOVERY*1000);
403 printk(KERN_WARNING "EEH: Permanent failure\n");
407 /* Since rtas may enable MMIO when posting the error log,
408 * don't post the error log until after all dev drivers
409 * have been informed.
411 eeh_slot_error_detail(frozen_pdn, EEH_LOG_TEMP_FAILURE);
413 /* If all device drivers were EEH-unaware, then shut
414 * down all of the device drivers, and hope they
415 * go down willingly, without panicing the system.
417 if (result == PCI_ERS_RESULT_NONE) {
418 rc = eeh_reset_device(frozen_pdn, frozen_bus);
420 printk(KERN_WARNING "EEH: Unable to reset, rc=%d\n", rc);
425 /* If all devices reported they can proceed, then re-enable MMIO */
426 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
427 rc = rtas_pci_enable(frozen_pdn, EEH_THAW_MMIO);
432 result = PCI_ERS_RESULT_NEED_RESET;
434 result = PCI_ERS_RESULT_NONE;
435 pci_walk_bus(frozen_bus, eeh_report_mmio_enabled, &result);
439 /* If all devices reported they can proceed, then re-enable DMA */
440 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
441 rc = rtas_pci_enable(frozen_pdn, EEH_THAW_DMA);
446 result = PCI_ERS_RESULT_NEED_RESET;
448 result = PCI_ERS_RESULT_RECOVERED;
451 /* If any device has a hard failure, then shut off everything. */
452 if (result == PCI_ERS_RESULT_DISCONNECT) {
453 printk(KERN_WARNING "EEH: Device driver gave up\n");
457 /* If any device called out for a reset, then reset the slot */
458 if (result == PCI_ERS_RESULT_NEED_RESET) {
459 rc = eeh_reset_device(frozen_pdn, NULL);
461 printk(KERN_WARNING "EEH: Cannot reset, rc=%d\n", rc);
464 result = PCI_ERS_RESULT_NONE;
465 pci_walk_bus(frozen_bus, eeh_report_reset, &result);
468 /* All devices should claim they have recovered by now. */
469 if ((result != PCI_ERS_RESULT_RECOVERED) &&
470 (result != PCI_ERS_RESULT_NONE)) {
471 printk(KERN_WARNING "EEH: Not recovered\n");
475 /* Tell all device drivers that they can resume operations */
476 pci_walk_bus(frozen_bus, eeh_report_resume, NULL);
482 * About 90% of all real-life EEH failures in the field
483 * are due to poorly seated PCI cards. Only 10% or so are
484 * due to actual, failed cards.
487 "EEH: PCI device at location=%s driver=%s pci addr=%s \n"
488 "has failed %d times in the last hour "
489 "and has been permanently disabled. \n"
490 "Please try reseating this device or replacing it.\n",
491 location, drv_str, pci_str, frozen_pdn->eeh_freeze_count);
496 "EEH: Unable to recover from failure of PCI device "
497 "at location=%s driver=%s pci addr=%s \n"
498 "Please try reseating this device or replacing it.\n",
499 location, drv_str, pci_str);
502 eeh_slot_error_detail(frozen_pdn, EEH_LOG_PERM_FAILURE);
504 /* Notify all devices that they're about to go down. */
505 pci_walk_bus(frozen_bus, eeh_report_failure, NULL);
507 /* Shut down the device drivers for good. */
508 pcibios_remove_pci_devices(frozen_bus);
513 /* ---------- end of file ---------- */