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)
51 printk("dn=%s mode=%x \tcfg_addr=%x pe_addr=%x \tfull=%s\n",
52 pdn->node->name, pdn->eeh_mode, pdn->eeh_config_addr,
53 pdn->eeh_pe_config_addr, pdn->node->full_name);
55 struct device_node *pc = pdn->node->child;
57 print_device_node_tree(PCI_DN(pc), dent);
64 * irq_in_use - return true if this irq is being used
66 static int irq_in_use(unsigned int irq)
70 struct irq_desc *desc = irq_desc + irq;
72 spin_lock_irqsave(&desc->lock, flags);
75 spin_unlock_irqrestore(&desc->lock, flags);
79 /* ------------------------------------------------------- */
81 * eeh_report_error - report pci error to each device driver
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".
88 static void eeh_report_error(struct pci_dev *dev, void *userdata)
90 enum pci_ers_result rc, *res = userdata;
91 struct pci_driver *driver = dev->driver;
93 dev->error_state = pci_channel_io_frozen;
98 if (irq_in_use (dev->irq)) {
99 struct device_node *dn = pci_device_to_OF_node(dev);
100 PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED;
101 disable_irq_nosync(dev->irq);
103 if (!driver->err_handler ||
104 !driver->err_handler->error_detected)
107 rc = driver->err_handler->error_detected (dev, pci_channel_io_frozen);
109 /* A driver that needs a reset trumps all others */
110 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
111 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
115 * eeh_report_mmio_enabled - tell drivers that MMIO has been enabled
117 * Tells each device driver that IO ports, MMIO and config space I/O
118 * are now enabled. Collects up and merges the device driver responses.
119 * Cumulative response passed back in "userdata".
122 static void eeh_report_mmio_enabled(struct pci_dev *dev, void *userdata)
124 enum pci_ers_result rc, *res = userdata;
125 struct pci_driver *driver = dev->driver;
128 !driver->err_handler ||
129 !driver->err_handler->mmio_enabled)
132 rc = driver->err_handler->mmio_enabled (dev);
134 /* A driver that needs a reset trumps all others */
135 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
136 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
140 * eeh_report_reset - tell device that slot has been reset
143 static void eeh_report_reset(struct pci_dev *dev, void *userdata)
145 enum pci_ers_result rc, *res = userdata;
146 struct pci_driver *driver = dev->driver;
147 struct device_node *dn = pci_device_to_OF_node(dev);
152 if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) {
153 PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED;
154 enable_irq(dev->irq);
156 if (!driver->err_handler ||
157 !driver->err_handler->slot_reset)
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;
168 * eeh_report_resume - tell device to resume normal operations
171 static void eeh_report_resume(struct pci_dev *dev, void *userdata)
173 struct pci_driver *driver = dev->driver;
174 struct device_node *dn = pci_device_to_OF_node(dev);
176 dev->error_state = pci_channel_io_normal;
181 if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) {
182 PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED;
183 enable_irq(dev->irq);
185 if (!driver->err_handler ||
186 !driver->err_handler->resume)
189 driver->err_handler->resume(dev);
193 * eeh_report_failure - tell device driver that device is dead.
195 * This informs the device driver that the device is permanently
196 * dead, and that no further recovery attempts will be made on it.
199 static void eeh_report_failure(struct pci_dev *dev, void *userdata)
201 struct pci_driver *driver = dev->driver;
203 dev->error_state = pci_channel_io_perm_failure;
208 if (irq_in_use (dev->irq)) {
209 struct device_node *dn = pci_device_to_OF_node(dev);
210 PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED;
211 disable_irq_nosync(dev->irq);
213 if (!driver->err_handler)
215 if (!driver->err_handler->error_detected)
217 driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
220 /* ------------------------------------------------------- */
222 * handle_eeh_events -- reset a PCI device after hard lockup.
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.
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).
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
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.
250 static int eeh_reset_device (struct pci_dn *pe_dn, struct pci_bus *bus)
252 struct device_node *dn;
255 /* pcibios will clear the counter; save the value */
256 cnt = pe_dn->eeh_freeze_count;
259 pcibios_remove_pci_devices(bus);
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);
268 /* Walk over all functions on this device. */
270 if (!pcibios_find_pci_bus(dn) && PCI_DN(dn->parent))
271 dn = dn->parent->child;
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);
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.
291 pcibios_add_pci_devices(bus);
293 pe_dn->eeh_freeze_count = cnt;
298 /* The longest amount of time to wait for a pci device
299 * to come back on line, in seconds.
301 #define MAX_WAIT_FOR_RECOVERY 150
303 struct pci_dn * handle_eeh_events (struct eeh_event *event)
305 struct device_node *frozen_dn;
306 struct pci_dn *frozen_pdn;
307 struct pci_bus *frozen_bus;
309 enum pci_ers_result result = PCI_ERS_RESULT_NONE;
310 const char *location, *pci_str, *drv_str;
312 frozen_dn = find_device_pe(event->dn);
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, pci_name(event->dev));
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";
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.. */
334 frozen_bus = pcibios_find_pci_bus (frozen_dn->parent);
337 printk(KERN_ERR "EEH: Cannot find PCI bus "
338 "for location=%s dn=%s\n",
339 location, frozen_dn->full_name);
343 frozen_pdn = PCI_DN(frozen_dn);
344 frozen_pdn->eeh_freeze_count++;
346 if (frozen_pdn->pcidev) {
347 pci_str = pci_name (frozen_pdn->pcidev);
348 drv_str = pcid_name (frozen_pdn->pcidev);
350 pci_str = pci_name (event->dev);
351 drv_str = pcid_name (event->dev);
354 if (frozen_pdn->eeh_freeze_count > EEH_MAX_ALLOWED_FREEZES)
355 goto excess_failures;
358 "EEH: This PCI device has failed %d times in the last hour:\n",
359 frozen_pdn->eeh_freeze_count);
361 "EEH: location=%s driver=%s pci addr=%s\n",
362 location, drv_str, pci_str);
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.
370 pci_walk_bus(frozen_bus, eeh_report_error, &result);
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);
376 printk(KERN_WARNING "EEH: Permanent failure\n");
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.
384 eeh_slot_error_detail(frozen_pdn, EEH_LOG_TEMP_FAILURE);
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.
390 if (result == PCI_ERS_RESULT_NONE) {
391 rc = eeh_reset_device(frozen_pdn, frozen_bus);
393 printk(KERN_WARNING "EEH: Unable to reset, rc=%d\n", rc);
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);
405 result = PCI_ERS_RESULT_NEED_RESET;
407 result = PCI_ERS_RESULT_NONE;
408 pci_walk_bus(frozen_bus, eeh_report_mmio_enabled, &result);
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);
419 result = PCI_ERS_RESULT_NEED_RESET;
421 result = PCI_ERS_RESULT_RECOVERED;
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");
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);
434 printk(KERN_WARNING "EEH: Cannot reset, rc=%d\n", rc);
437 result = PCI_ERS_RESULT_NONE;
438 pci_walk_bus(frozen_bus, eeh_report_reset, &result);
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");
448 /* Tell all device drivers that they can resume operations */
449 pci_walk_bus(frozen_bus, eeh_report_resume, NULL);
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.
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);
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);
475 eeh_slot_error_detail(frozen_pdn, EEH_LOG_PERM_FAILURE);
477 /* Notify all devices that they're about to go down. */
478 pci_walk_bus(frozen_bus, eeh_report_failure, NULL);
480 /* Shut down the device drivers for good. */
481 pcibios_remove_pci_devices(frozen_bus);
486 /* ---------- end of file ---------- */