2 * PCI Error Recovery Driver for RPA-compliant PPC64 platform.
3 * Copyright (C) 2004, 2005 Linas Vepstas <linas@linas.org>
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.
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
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.
22 * Send feedback to <linas@us.ibm.com>
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>
31 #include <asm/eeh_event.h>
32 #include <asm/ppc-pci.h>
33 #include <asm/pci-bridge.h>
38 static inline const char * pcid_name (struct pci_dev *pdev)
41 return pdev->dev.driver->name;
46 * Return the "partitionable endpoint" (pe) under which this device lies
48 static struct device_node * find_device_pe(struct device_node *dn)
50 while ((dn->parent) && PCI_DN(dn->parent) &&
51 (PCI_DN(dn->parent)->eeh_mode & EEH_MODE_SUPPORTED)) {
59 static void print_device_node_tree (struct pci_dn *pdn, int dent)
65 printk("dn=%s mode=%x \tcfg_addr=%x pe_addr=%x \tfull=%s\n",
66 pdn->node->name, pdn->eeh_mode, pdn->eeh_config_addr,
67 pdn->eeh_pe_config_addr, pdn->node->full_name);
69 struct device_node *pc = pdn->node->child;
71 print_device_node_tree(PCI_DN(pc), dent);
78 * irq_in_use - return true if this irq is being used
80 static int irq_in_use(unsigned int irq)
84 struct irq_desc *desc = irq_desc + irq;
86 spin_lock_irqsave(&desc->lock, flags);
89 spin_unlock_irqrestore(&desc->lock, flags);
93 /* ------------------------------------------------------- */
94 /** eeh_report_error - report an EEH error to each device,
95 * collect up and merge the device responses.
98 static void eeh_report_error(struct pci_dev *dev, void *userdata)
100 enum pcierr_result rc, *res = userdata;
101 struct pci_driver *driver = dev->driver;
103 dev->error_state = pci_channel_io_frozen;
108 if (irq_in_use (dev->irq)) {
109 struct device_node *dn = pci_device_to_OF_node(dev);
110 PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED;
111 disable_irq_nosync(dev->irq);
113 if (!driver->err_handler)
115 if (!driver->err_handler->error_detected)
118 rc = driver->err_handler->error_detected (dev, pci_channel_io_frozen);
119 if (*res == PCIERR_RESULT_NONE) *res = rc;
120 if (*res == PCIERR_RESULT_NEED_RESET) return;
121 if (*res == PCIERR_RESULT_DISCONNECT &&
122 rc == PCIERR_RESULT_NEED_RESET) *res = rc;
125 /** eeh_report_reset -- tell this device that the pci slot
129 static void eeh_report_reset(struct pci_dev *dev, void *userdata)
131 struct pci_driver *driver = dev->driver;
132 struct device_node *dn = pci_device_to_OF_node(dev);
137 if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) {
138 PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED;
139 enable_irq(dev->irq);
141 if (!driver->err_handler)
143 if (!driver->err_handler->slot_reset)
146 driver->err_handler->slot_reset(dev);
149 static void eeh_report_resume(struct pci_dev *dev, void *userdata)
151 struct pci_driver *driver = dev->driver;
153 dev->error_state = pci_channel_io_normal;
157 if (!driver->err_handler)
159 if (!driver->err_handler->resume)
162 driver->err_handler->resume(dev);
165 static void eeh_report_failure(struct pci_dev *dev, void *userdata)
167 struct pci_driver *driver = dev->driver;
169 dev->error_state = pci_channel_io_perm_failure;
174 if (irq_in_use (dev->irq)) {
175 struct device_node *dn = pci_device_to_OF_node(dev);
176 PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED;
177 disable_irq_nosync(dev->irq);
179 if (!driver->err_handler)
181 if (!driver->err_handler->error_detected)
183 driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
186 /* ------------------------------------------------------- */
188 * handle_eeh_events -- reset a PCI device after hard lockup.
190 * pSeries systems will isolate a PCI slot if the PCI-Host
191 * bridge detects address or data parity errors, DMA's
192 * occuring to wild addresses (which usually happen due to
193 * bugs in device drivers or in PCI adapter firmware).
194 * Slot isolations also occur if #SERR, #PERR or other misc
195 * PCI-related errors are detected.
197 * Recovery process consists of unplugging the device driver
198 * (which generated hotplug events to userspace), then issuing
199 * a PCI #RST to the device, then reconfiguring the PCI config
200 * space for all bridges & devices under this slot, and then
201 * finally restarting the device drivers (which cause a second
202 * set of hotplug events to go out to userspace).
206 * eeh_reset_device() -- perform actual reset of a pci slot
207 * Args: bus: pointer to the pci bus structure corresponding
208 * to the isolated slot. A non-null value will
209 * cause all devices under the bus to be removed
211 * pe_dn: pointer to a "Partionable Endpoint" device node.
212 * This is the top-level structure on which pci
213 * bus resets can be performed.
216 static void eeh_reset_device (struct pci_dn *pe_dn, struct pci_bus *bus)
219 pcibios_remove_pci_devices(bus);
221 /* Reset the pci controller. (Asserts RST#; resets config space).
222 * Reconfigure bridges and devices */
223 rtas_set_slot_reset(pe_dn);
225 /* Walk over all functions on this device */
226 rtas_configure_bridge(pe_dn);
227 eeh_restore_bars(pe_dn);
229 /* Give the system 5 seconds to finish running the user-space
230 * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes,
231 * this is a hack, but if we don't do this, and try to bring
232 * the device up before the scripts have taken it down,
233 * potentially weird things happen.
237 pcibios_add_pci_devices(bus);
241 /* The longest amount of time to wait for a pci device
242 * to come back on line, in seconds.
244 #define MAX_WAIT_FOR_RECOVERY 15
246 void handle_eeh_events (struct eeh_event *event)
248 struct device_node *frozen_dn;
249 struct pci_dn *frozen_pdn;
250 struct pci_bus *frozen_bus;
251 int perm_failure = 0;
253 frozen_dn = find_device_pe(event->dn);
254 frozen_bus = pcibios_find_pci_bus(frozen_dn);
257 printk(KERN_ERR "EEH: Error: Cannot find partition endpoint for %s\n",
258 pci_name(event->dev));
262 /* There are two different styles for coming up with the PE.
263 * In the old style, it was the highest EEH-capable device
264 * which was always an EADS pci bridge. In the new style,
265 * there might not be any EADS bridges, and even when there are,
266 * the firmware marks them as "EEH incapable". So another
267 * two-step is needed to find the pci bus.. */
269 frozen_bus = pcibios_find_pci_bus (frozen_dn->parent);
272 printk(KERN_ERR "EEH: Cannot find PCI bus for %s\n",
273 frozen_dn->full_name);
278 /* We may get "permanent failure" messages on empty slots.
279 * These are false alarms. Empty slots have no child dn. */
280 if ((event->state == pci_channel_io_perm_failure) && (frozen_device == NULL))
284 frozen_pdn = PCI_DN(frozen_dn);
285 frozen_pdn->eeh_freeze_count++;
287 if (frozen_pdn->eeh_freeze_count > EEH_MAX_ALLOWED_FREEZES)
290 /* If the reset state is a '5' and the time to reset is 0 (infinity)
291 * or is more then 15 seconds, then mark this as a permanent failure.
293 if ((event->state == pci_channel_io_perm_failure) &&
294 ((event->time_unavail <= 0) ||
295 (event->time_unavail > MAX_WAIT_FOR_RECOVERY*1000)))
300 /* Log the error with the rtas logger. */
303 * About 90% of all real-life EEH failures in the field
304 * are due to poorly seated PCI cards. Only 10% or so are
305 * due to actual, failed cards.
308 "EEH: PCI device %s - %s has failed %d times \n"
309 "and has been permanently disabled. Please try reseating\n"
310 "this device or replacing it.\n",
311 pci_name (frozen_pdn->pcidev),
312 pcid_name(frozen_pdn->pcidev),
313 frozen_pdn->eeh_freeze_count);
315 eeh_slot_error_detail(frozen_pdn, 2 /* Permanent Error */);
317 /* Notify all devices that they're about to go down. */
318 pci_walk_bus(frozen_bus, eeh_report_failure, 0);
320 /* Shut down the device drivers for good. */
321 pcibios_remove_pci_devices(frozen_bus);
325 eeh_slot_error_detail(frozen_pdn, 1 /* Temporary Error */);
327 "EEH: This PCI device has failed %d times since last reboot: %s - %s\n",
328 frozen_pdn->eeh_freeze_count,
329 pci_name (frozen_pdn->pcidev),
330 pcid_name(frozen_pdn->pcidev));
332 /* Walk the various device drivers attached to this slot through
333 * a reset sequence, giving each an opportunity to do what it needs
334 * to accomplish the reset. Each child gets a report of the
335 * status ... if any child can't handle the reset, then the entire
336 * slot is dlpar removed and added.
338 enum pcierr_result result = PCIERR_RESULT_NONE;
339 pci_walk_bus(frozen_bus, eeh_report_error, &result);
341 /* If all device drivers were EEH-unaware, then shut
342 * down all of the device drivers, and hope they
343 * go down willingly, without panicing the system.
345 if (result == PCIERR_RESULT_NONE) {
346 eeh_reset_device(frozen_pdn, frozen_bus);
349 /* If any device called out for a reset, then reset the slot */
350 if (result == PCIERR_RESULT_NEED_RESET) {
351 eeh_reset_device(frozen_pdn, NULL);
352 pci_walk_bus(frozen_bus, eeh_report_reset, 0);
355 /* If all devices reported they can proceed, the re-enable PIO */
356 if (result == PCIERR_RESULT_CAN_RECOVER) {
357 /* XXX Not supported; we brute-force reset the device */
358 eeh_reset_device(frozen_pdn, NULL);
359 pci_walk_bus(frozen_bus, eeh_report_reset, 0);
362 /* Tell all device drivers that they can resume operations */
363 pci_walk_bus(frozen_bus, eeh_report_resume, 0);
366 /* ---------- end of file ---------- */