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/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 /* ------------------------------------------------------- */
80 /** eeh_report_error - report an EEH error to each device,
81 * collect up and merge the device responses.
84 static void eeh_report_error(struct pci_dev *dev, void *userdata)
86 enum pci_ers_result rc, *res = userdata;
87 struct pci_driver *driver = dev->driver;
89 dev->error_state = pci_channel_io_frozen;
94 if (irq_in_use (dev->irq)) {
95 struct device_node *dn = pci_device_to_OF_node(dev);
96 PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED;
97 disable_irq_nosync(dev->irq);
99 if (!driver->err_handler)
101 if (!driver->err_handler->error_detected)
104 rc = driver->err_handler->error_detected (dev, pci_channel_io_frozen);
105 if (*res == PCI_ERS_RESULT_NONE) *res = rc;
106 if (*res == PCI_ERS_RESULT_NEED_RESET) return;
107 if (*res == PCI_ERS_RESULT_DISCONNECT &&
108 rc == PCI_ERS_RESULT_NEED_RESET) *res = rc;
111 /** eeh_report_reset -- tell this device that the pci slot
115 static void eeh_report_reset(struct pci_dev *dev, void *userdata)
117 struct pci_driver *driver = dev->driver;
118 struct device_node *dn = pci_device_to_OF_node(dev);
123 if ((PCI_DN(dn)->eeh_mode) & EEH_MODE_IRQ_DISABLED) {
124 PCI_DN(dn)->eeh_mode &= ~EEH_MODE_IRQ_DISABLED;
125 enable_irq(dev->irq);
127 if (!driver->err_handler)
129 if (!driver->err_handler->slot_reset)
132 driver->err_handler->slot_reset(dev);
135 static void eeh_report_resume(struct pci_dev *dev, void *userdata)
137 struct pci_driver *driver = dev->driver;
139 dev->error_state = pci_channel_io_normal;
143 if (!driver->err_handler)
145 if (!driver->err_handler->resume)
148 driver->err_handler->resume(dev);
151 static void eeh_report_failure(struct pci_dev *dev, void *userdata)
153 struct pci_driver *driver = dev->driver;
155 dev->error_state = pci_channel_io_perm_failure;
160 if (irq_in_use (dev->irq)) {
161 struct device_node *dn = pci_device_to_OF_node(dev);
162 PCI_DN(dn)->eeh_mode |= EEH_MODE_IRQ_DISABLED;
163 disable_irq_nosync(dev->irq);
165 if (!driver->err_handler)
167 if (!driver->err_handler->error_detected)
169 driver->err_handler->error_detected(dev, pci_channel_io_perm_failure);
172 /* ------------------------------------------------------- */
174 * handle_eeh_events -- reset a PCI device after hard lockup.
176 * pSeries systems will isolate a PCI slot if the PCI-Host
177 * bridge detects address or data parity errors, DMA's
178 * occuring to wild addresses (which usually happen due to
179 * bugs in device drivers or in PCI adapter firmware).
180 * Slot isolations also occur if #SERR, #PERR or other misc
181 * PCI-related errors are detected.
183 * Recovery process consists of unplugging the device driver
184 * (which generated hotplug events to userspace), then issuing
185 * a PCI #RST to the device, then reconfiguring the PCI config
186 * space for all bridges & devices under this slot, and then
187 * finally restarting the device drivers (which cause a second
188 * set of hotplug events to go out to userspace).
192 * eeh_reset_device() -- perform actual reset of a pci slot
193 * Args: bus: pointer to the pci bus structure corresponding
194 * to the isolated slot. A non-null value will
195 * cause all devices under the bus to be removed
197 * pe_dn: pointer to a "Partionable Endpoint" device node.
198 * This is the top-level structure on which pci
199 * bus resets can be performed.
202 static int eeh_reset_device (struct pci_dn *pe_dn, struct pci_bus *bus)
206 pcibios_remove_pci_devices(bus);
208 /* Reset the pci controller. (Asserts RST#; resets config space).
209 * Reconfigure bridges and devices. Don't try to bring the system
210 * up if the reset failed for some reason. */
211 rc = rtas_set_slot_reset(pe_dn);
215 /* New-style config addrs might be shared across multiple devices,
216 * Walk over all functions on this device */
217 if (pe_dn->eeh_pe_config_addr) {
218 struct device_node *pe = pe_dn->node;
219 pe = pe->parent->child;
221 struct pci_dn *ppe = PCI_DN(pe);
222 if (pe_dn->eeh_pe_config_addr == ppe->eeh_pe_config_addr) {
223 rtas_configure_bridge(ppe);
224 eeh_restore_bars(ppe);
229 rtas_configure_bridge(pe_dn);
230 eeh_restore_bars(pe_dn);
233 /* Give the system 5 seconds to finish running the user-space
234 * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes,
235 * this is a hack, but if we don't do this, and try to bring
236 * the device up before the scripts have taken it down,
237 * potentially weird things happen.
241 pcibios_add_pci_devices(bus);
247 /* The longest amount of time to wait for a pci device
248 * to come back on line, in seconds.
250 #define MAX_WAIT_FOR_RECOVERY 15
252 struct pci_dn * handle_eeh_events (struct eeh_event *event)
254 struct device_node *frozen_dn;
255 struct pci_dn *frozen_pdn;
256 struct pci_bus *frozen_bus;
258 enum pci_ers_result result = PCI_ERS_RESULT_NONE;
259 const char *pci_str, *drv_str;
261 frozen_dn = find_device_pe(event->dn);
262 frozen_bus = pcibios_find_pci_bus(frozen_dn);
265 printk(KERN_ERR "EEH: Error: Cannot find partition endpoint for %s\n",
266 pci_name(event->dev));
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.. */
277 frozen_bus = pcibios_find_pci_bus (frozen_dn->parent);
280 printk(KERN_ERR "EEH: Cannot find PCI bus for %s\n",
281 frozen_dn->full_name);
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))
292 frozen_pdn = PCI_DN(frozen_dn);
293 frozen_pdn->eeh_freeze_count++;
295 if (frozen_pdn->pcidev) {
296 pci_str = pci_name (frozen_pdn->pcidev);
297 drv_str = pcid_name (frozen_pdn->pcidev);
299 pci_str = pci_name (event->dev);
300 drv_str = pcid_name (event->dev);
303 if (frozen_pdn->eeh_freeze_count > EEH_MAX_ALLOWED_FREEZES)
304 goto excess_failures;
306 /* If the reset state is a '5' and the time to reset is 0 (infinity)
307 * or is more then 15 seconds, then mark this as a permanent failure.
309 if ((event->state == pci_channel_io_perm_failure) &&
310 ((event->time_unavail <= 0) ||
311 (event->time_unavail > MAX_WAIT_FOR_RECOVERY*1000)))
314 eeh_slot_error_detail(frozen_pdn, 1 /* Temporary Error */);
316 "EEH: This PCI device has failed %d times since last reboot: %s - %s\n",
317 frozen_pdn->eeh_freeze_count, drv_str, pci_str);
319 /* Walk the various device drivers attached to this slot through
320 * a reset sequence, giving each an opportunity to do what it needs
321 * to accomplish the reset. Each child gets a report of the
322 * status ... if any child can't handle the reset, then the entire
323 * slot is dlpar removed and added.
325 pci_walk_bus(frozen_bus, eeh_report_error, &result);
327 /* If all device drivers were EEH-unaware, then shut
328 * down all of the device drivers, and hope they
329 * go down willingly, without panicing the system.
331 if (result == PCI_ERS_RESULT_NONE) {
332 rc = eeh_reset_device(frozen_pdn, frozen_bus);
337 /* If any device called out for a reset, then reset the slot */
338 if (result == PCI_ERS_RESULT_NEED_RESET) {
339 rc = eeh_reset_device(frozen_pdn, NULL);
342 pci_walk_bus(frozen_bus, eeh_report_reset, NULL);
345 /* If all devices reported they can proceed, the re-enable PIO */
346 if (result == PCI_ERS_RESULT_CAN_RECOVER) {
347 /* XXX Not supported; we brute-force reset the device */
348 rc = eeh_reset_device(frozen_pdn, NULL);
351 pci_walk_bus(frozen_bus, eeh_report_reset, NULL);
354 /* Tell all device drivers that they can resume operations */
355 pci_walk_bus(frozen_bus, eeh_report_resume, NULL);
361 * About 90% of all real-life EEH failures in the field
362 * are due to poorly seated PCI cards. Only 10% or so are
363 * due to actual, failed cards.
366 "EEH: PCI device %s - %s has failed %d times \n"
367 "and has been permanently disabled. Please try reseating\n"
368 "this device or replacing it.\n",
369 drv_str, pci_str, frozen_pdn->eeh_freeze_count);
374 "EEH: Unable to recover from failure of PCI device %s - %s\n"
375 "Please try reseating this device or replacing it.\n",
379 eeh_slot_error_detail(frozen_pdn, 2 /* Permanent Error */);
381 /* Notify all devices that they're about to go down. */
382 pci_walk_bus(frozen_bus, eeh_report_failure, NULL);
384 /* Shut down the device drivers for good. */
385 pcibios_remove_pci_devices(frozen_bus);
390 /* ---------- end of file ---------- */