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);
82 /* ------------------------------------------------------- */
84 * eeh_report_error - report pci error to each device driver
86 * Report an EEH error to each device driver, collect up and
87 * merge the device driver responses. Cumulative response
88 * passed back in "userdata".
91 static void eeh_report_error(struct pci_dev *dev, void *userdata)
93 enum pci_ers_result rc, *res = userdata;
94 struct pci_driver *driver = dev->driver;
96 dev->error_state = pci_channel_io_frozen;
101 if (irq_in_use (dev->irq)) {
102 struct device_node *dn = pci_device_to_OF_node(dev);
103 PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED;
104 disable_irq_nosync(dev->irq);
106 if (!driver->err_handler ||
107 !driver->err_handler->error_detected)
110 rc = driver->err_handler->error_detected (dev, pci_channel_io_frozen);
112 /* A driver that needs a reset trumps all others */
113 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
114 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
118 * eeh_report_mmio_enabled - tell drivers that MMIO has been enabled
120 * Tells each device driver that IO ports, MMIO and config space I/O
121 * are now enabled. Collects up and merges the device driver responses.
122 * Cumulative response passed back in "userdata".
125 static void eeh_report_mmio_enabled(struct pci_dev *dev, void *userdata)
127 enum pci_ers_result rc, *res = userdata;
128 struct pci_driver *driver = dev->driver;
131 !driver->err_handler ||
132 !driver->err_handler->mmio_enabled)
135 rc = driver->err_handler->mmio_enabled (dev);
137 /* A driver that needs a reset trumps all others */
138 if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
139 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
143 * eeh_report_reset - tell device that slot has been reset
146 static void eeh_report_reset(struct pci_dev *dev, void *userdata)
148 enum pci_ers_result rc, *res = userdata;
149 struct pci_driver *driver = dev->driver;
150 struct device_node *dn = pci_device_to_OF_node(dev);
155 if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) {
156 PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED;
157 enable_irq(dev->irq);
159 if (!driver->err_handler ||
160 !driver->err_handler->slot_reset)
163 rc = driver->err_handler->slot_reset(dev);
164 if ((*res == PCI_ERS_RESULT_NONE) ||
165 (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc;
166 if (*res == PCI_ERS_RESULT_DISCONNECT &&
167 rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
171 * eeh_report_resume - tell device to resume normal operations
174 static void eeh_report_resume(struct pci_dev *dev, void *userdata)
176 struct pci_driver *driver = dev->driver;
177 struct device_node *dn = pci_device_to_OF_node(dev);
179 dev->error_state = pci_channel_io_normal;
184 if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) {
185 PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED;
186 enable_irq(dev->irq);
188 if (!driver->err_handler ||
189 !driver->err_handler->resume)
192 driver->err_handler->resume(dev);
196 * eeh_report_failure - tell device driver that device is dead.
198 * This informs the device driver that the device is permanently
199 * dead, and that no further recovery attempts will be made on it.
202 static void eeh_report_failure(struct pci_dev *dev, void *userdata)
204 struct pci_driver *driver = dev->driver;
206 dev->error_state = pci_channel_io_perm_failure;
211 if (irq_in_use (dev->irq)) {
212 struct device_node *dn = pci_device_to_OF_node(dev);
213 PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED;
214 disable_irq_nosync(dev->irq);
216 if (!driver->err_handler)
218 if (!driver->err_handler->error_detected)
220 driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
223 /* ------------------------------------------------------- */
225 * handle_eeh_events -- reset a PCI device after hard lockup.
227 * pSeries systems will isolate a PCI slot if the PCI-Host
228 * bridge detects address or data parity errors, DMA's
229 * occurring to wild addresses (which usually happen due to
230 * bugs in device drivers or in PCI adapter firmware).
231 * Slot isolations also occur if #SERR, #PERR or other misc
232 * PCI-related errors are detected.
234 * Recovery process consists of unplugging the device driver
235 * (which generated hotplug events to userspace), then issuing
236 * a PCI #RST to the device, then reconfiguring the PCI config
237 * space for all bridges & devices under this slot, and then
238 * finally restarting the device drivers (which cause a second
239 * set of hotplug events to go out to userspace).
243 * eeh_reset_device() -- perform actual reset of a pci slot
244 * @bus: pointer to the pci bus structure corresponding
245 * to the isolated slot. A non-null value will
246 * cause all devices under the bus to be removed
248 * @pe_dn: pointer to a "Partionable Endpoint" device node.
249 * This is the top-level structure on which pci
250 * bus resets can be performed.
253 static int eeh_reset_device (struct pci_dn *pe_dn, struct pci_bus *bus)
255 struct device_node *dn;
258 /* pcibios will clear the counter; save the value */
259 cnt = pe_dn->eeh_freeze_count;
262 pcibios_remove_pci_devices(bus);
264 /* Reset the pci controller. (Asserts RST#; resets config space).
265 * Reconfigure bridges and devices. Don't try to bring the system
266 * up if the reset failed for some reason. */
267 rc = rtas_set_slot_reset(pe_dn);
271 /* Walk over all functions on this device. */
273 if (!pcibios_find_pci_bus(dn) && PCI_DN(dn->parent))
274 dn = dn->parent->child;
277 struct pci_dn *ppe = PCI_DN(dn);
278 /* On Power4, always true because eeh_pe_config_addr=0 */
279 if (pe_dn->eeh_pe_config_addr == ppe->eeh_pe_config_addr) {
280 rtas_configure_bridge(ppe);
281 eeh_restore_bars(ppe);
286 /* Give the system 5 seconds to finish running the user-space
287 * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes,
288 * this is a hack, but if we don't do this, and try to bring
289 * the device up before the scripts have taken it down,
290 * potentially weird things happen.
294 pcibios_add_pci_devices(bus);
296 pe_dn->eeh_freeze_count = cnt;
301 /* The longest amount of time to wait for a pci device
302 * to come back on line, in seconds.
304 #define MAX_WAIT_FOR_RECOVERY 150
306 struct pci_dn * handle_eeh_events (struct eeh_event *event)
308 struct device_node *frozen_dn;
309 struct pci_dn *frozen_pdn;
310 struct pci_bus *frozen_bus;
312 enum pci_ers_result result = PCI_ERS_RESULT_NONE;
313 const char *location, *pci_str, *drv_str;
315 frozen_dn = find_device_pe(event->dn);
318 location = of_get_property(event->dn, "ibm,loc-code", NULL);
319 location = location ? location : "unknown";
320 printk(KERN_ERR "EEH: Error: Cannot find partition endpoint "
321 "for location=%s pci addr=%s\n",
322 location, pci_name(event->dev));
326 frozen_bus = pcibios_find_pci_bus(frozen_dn);
327 location = of_get_property(frozen_dn, "ibm,loc-code", NULL);
328 location = location ? location : "unknown";
330 /* There are two different styles for coming up with the PE.
331 * In the old style, it was the highest EEH-capable device
332 * which was always an EADS pci bridge. In the new style,
333 * there might not be any EADS bridges, and even when there are,
334 * the firmware marks them as "EEH incapable". So another
335 * two-step is needed to find the pci bus.. */
337 frozen_bus = pcibios_find_pci_bus (frozen_dn->parent);
340 printk(KERN_ERR "EEH: Cannot find PCI bus "
341 "for location=%s dn=%s\n",
342 location, frozen_dn->full_name);
346 frozen_pdn = PCI_DN(frozen_dn);
347 frozen_pdn->eeh_freeze_count++;
349 if (frozen_pdn->pcidev) {
350 pci_str = pci_name (frozen_pdn->pcidev);
351 drv_str = pcid_name (frozen_pdn->pcidev);
353 pci_str = pci_name (event->dev);
354 drv_str = pcid_name (event->dev);
357 if (frozen_pdn->eeh_freeze_count > EEH_MAX_ALLOWED_FREEZES)
358 goto excess_failures;
361 "EEH: This PCI device has failed %d times in the last hour:\n",
362 frozen_pdn->eeh_freeze_count);
364 "EEH: location=%s driver=%s pci addr=%s\n",
365 location, drv_str, pci_str);
367 /* Walk the various device drivers attached to this slot through
368 * a reset sequence, giving each an opportunity to do what it needs
369 * to accomplish the reset. Each child gets a report of the
370 * status ... if any child can't handle the reset, then the entire
371 * slot is dlpar removed and added.
373 pci_walk_bus(frozen_bus, eeh_report_error, &result);
375 /* Get the current PCI slot state. This can take a long time,
376 * sometimes over 3 seconds for certain systems. */
377 rc = eeh_wait_for_slot_status (frozen_pdn, MAX_WAIT_FOR_RECOVERY*1000);
379 printk(KERN_WARNING "EEH: Permanent failure\n");
383 /* Since rtas may enable MMIO when posting the error log,
384 * don't post the error log until after all dev drivers
385 * have been informed.
387 eeh_slot_error_detail(frozen_pdn, EEH_LOG_TEMP_FAILURE);
389 /* If all device drivers were EEH-unaware, then shut
390 * down all of the device drivers, and hope they
391 * go down willingly, without panicing the system.
393 if (result == PCI_ERS_RESULT_NONE) {
394 rc = eeh_reset_device(frozen_pdn, frozen_bus);
396 printk(KERN_WARNING "EEH: Unable to reset, rc=%d\n", rc);
401 /* If all devices reported they can proceed, then re-enable MMIO */
402 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
403 rc = rtas_pci_enable(frozen_pdn, EEH_THAW_MMIO);
408 result = PCI_ERS_RESULT_NEED_RESET;
410 result = PCI_ERS_RESULT_NONE;
411 pci_walk_bus(frozen_bus, eeh_report_mmio_enabled, &result);
415 /* If all devices reported they can proceed, then re-enable DMA */
416 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
417 rc = rtas_pci_enable(frozen_pdn, EEH_THAW_DMA);
422 result = PCI_ERS_RESULT_NEED_RESET;
424 result = PCI_ERS_RESULT_RECOVERED;
427 /* If any device has a hard failure, then shut off everything. */
428 if (result == PCI_ERS_RESULT_DISCONNECT) {
429 printk(KERN_WARNING "EEH: Device driver gave up\n");
433 /* If any device called out for a reset, then reset the slot */
434 if (result == PCI_ERS_RESULT_NEED_RESET) {
435 rc = eeh_reset_device(frozen_pdn, NULL);
437 printk(KERN_WARNING "EEH: Cannot reset, rc=%d\n", rc);
440 result = PCI_ERS_RESULT_NONE;
441 pci_walk_bus(frozen_bus, eeh_report_reset, &result);
444 /* All devices should claim they have recovered by now. */
445 if ((result != PCI_ERS_RESULT_RECOVERED) &&
446 (result != PCI_ERS_RESULT_NONE)) {
447 printk(KERN_WARNING "EEH: Not recovered\n");
451 /* Tell all device drivers that they can resume operations */
452 pci_walk_bus(frozen_bus, eeh_report_resume, NULL);
458 * About 90% of all real-life EEH failures in the field
459 * are due to poorly seated PCI cards. Only 10% or so are
460 * due to actual, failed cards.
463 "EEH: PCI device at location=%s driver=%s pci addr=%s \n"
464 "has failed %d times in the last hour "
465 "and has been permanently disabled. \n"
466 "Please try reseating this device or replacing it.\n",
467 location, drv_str, pci_str, frozen_pdn->eeh_freeze_count);
472 "EEH: Unable to recover from failure of PCI device "
473 "at location=%s driver=%s pci addr=%s \n"
474 "Please try reseating this device or replacing it.\n",
475 location, drv_str, pci_str);
478 eeh_slot_error_detail(frozen_pdn, EEH_LOG_PERM_FAILURE);
480 /* Notify all devices that they're about to go down. */
481 pci_walk_bus(frozen_bus, eeh_report_failure, NULL);
483 /* Shut down the device drivers for good. */
484 pcibios_remove_pci_devices(frozen_bus);
489 /* ---------- end of file ---------- */