1317f06deSGavin Shan /* 2317f06deSGavin Shan * PCI Error Recovery Driver for RPA-compliant PPC64 platform. 3317f06deSGavin Shan * Copyright IBM Corp. 2004 2005 4317f06deSGavin Shan * Copyright Linas Vepstas <linas@linas.org> 2004, 2005 5317f06deSGavin Shan * 6317f06deSGavin Shan * All rights reserved. 7317f06deSGavin Shan * 8317f06deSGavin Shan * This program is free software; you can redistribute it and/or modify 9317f06deSGavin Shan * it under the terms of the GNU General Public License as published by 10317f06deSGavin Shan * the Free Software Foundation; either version 2 of the License, or (at 11317f06deSGavin Shan * your option) any later version. 12317f06deSGavin Shan * 13317f06deSGavin Shan * This program is distributed in the hope that it will be useful, but 14317f06deSGavin Shan * WITHOUT ANY WARRANTY; without even the implied warranty of 15317f06deSGavin Shan * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or 16317f06deSGavin Shan * NON INFRINGEMENT. See the GNU General Public License for more 17317f06deSGavin Shan * details. 18317f06deSGavin Shan * 19317f06deSGavin Shan * You should have received a copy of the GNU General Public License 20317f06deSGavin Shan * along with this program; if not, write to the Free Software 21317f06deSGavin Shan * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 22317f06deSGavin Shan * 23317f06deSGavin Shan * Send comments and feedback to Linas Vepstas <linas@austin.ibm.com> 24317f06deSGavin Shan */ 25317f06deSGavin Shan #include <linux/delay.h> 26317f06deSGavin Shan #include <linux/interrupt.h> 27317f06deSGavin Shan #include <linux/irq.h> 28317f06deSGavin Shan #include <linux/module.h> 29317f06deSGavin Shan #include <linux/pci.h> 30317f06deSGavin Shan #include <asm/eeh.h> 31317f06deSGavin Shan #include <asm/eeh_event.h> 32317f06deSGavin Shan #include <asm/ppc-pci.h> 33317f06deSGavin Shan #include <asm/pci-bridge.h> 34317f06deSGavin Shan #include <asm/prom.h> 35317f06deSGavin Shan #include <asm/rtas.h> 36317f06deSGavin Shan 37317f06deSGavin Shan /** 38317f06deSGavin Shan * eeh_pcid_name - Retrieve name of PCI device driver 39317f06deSGavin Shan * @pdev: PCI device 40317f06deSGavin Shan * 41317f06deSGavin Shan * This routine is used to retrieve the name of PCI device driver 42317f06deSGavin Shan * if that's valid. 43317f06deSGavin Shan */ 44317f06deSGavin Shan static inline const char *eeh_pcid_name(struct pci_dev *pdev) 45317f06deSGavin Shan { 46317f06deSGavin Shan if (pdev && pdev->dev.driver) 47317f06deSGavin Shan return pdev->dev.driver->name; 48317f06deSGavin Shan return ""; 49317f06deSGavin Shan } 50317f06deSGavin Shan 51317f06deSGavin Shan /** 52317f06deSGavin Shan * eeh_pcid_get - Get the PCI device driver 53317f06deSGavin Shan * @pdev: PCI device 54317f06deSGavin Shan * 55317f06deSGavin Shan * The function is used to retrieve the PCI device driver for 56317f06deSGavin Shan * the indicated PCI device. Besides, we will increase the reference 57317f06deSGavin Shan * of the PCI device driver to prevent that being unloaded on 58317f06deSGavin Shan * the fly. Otherwise, kernel crash would be seen. 59317f06deSGavin Shan */ 60317f06deSGavin Shan static inline struct pci_driver *eeh_pcid_get(struct pci_dev *pdev) 61317f06deSGavin Shan { 62317f06deSGavin Shan if (!pdev || !pdev->driver) 63317f06deSGavin Shan return NULL; 64317f06deSGavin Shan 65317f06deSGavin Shan if (!try_module_get(pdev->driver->driver.owner)) 66317f06deSGavin Shan return NULL; 67317f06deSGavin Shan 68317f06deSGavin Shan return pdev->driver; 69317f06deSGavin Shan } 70317f06deSGavin Shan 71317f06deSGavin Shan /** 72317f06deSGavin Shan * eeh_pcid_put - Dereference on the PCI device driver 73317f06deSGavin Shan * @pdev: PCI device 74317f06deSGavin Shan * 75317f06deSGavin Shan * The function is called to do dereference on the PCI device 76317f06deSGavin Shan * driver of the indicated PCI device. 77317f06deSGavin Shan */ 78317f06deSGavin Shan static inline void eeh_pcid_put(struct pci_dev *pdev) 79317f06deSGavin Shan { 80317f06deSGavin Shan if (!pdev || !pdev->driver) 81317f06deSGavin Shan return; 82317f06deSGavin Shan 83317f06deSGavin Shan module_put(pdev->driver->driver.owner); 84317f06deSGavin Shan } 85317f06deSGavin Shan 86317f06deSGavin Shan #if 0 87317f06deSGavin Shan static void print_device_node_tree(struct pci_dn *pdn, int dent) 88317f06deSGavin Shan { 89317f06deSGavin Shan int i; 90317f06deSGavin Shan struct device_node *pc; 91317f06deSGavin Shan 92317f06deSGavin Shan if (!pdn) 93317f06deSGavin Shan return; 94317f06deSGavin Shan for (i = 0; i < dent; i++) 95317f06deSGavin Shan printk(" "); 96317f06deSGavin Shan printk("dn=%s mode=%x \tcfg_addr=%x pe_addr=%x \tfull=%s\n", 97317f06deSGavin Shan pdn->node->name, pdn->eeh_mode, pdn->eeh_config_addr, 98317f06deSGavin Shan pdn->eeh_pe_config_addr, pdn->node->full_name); 99317f06deSGavin Shan dent += 3; 100317f06deSGavin Shan pc = pdn->node->child; 101317f06deSGavin Shan while (pc) { 102317f06deSGavin Shan print_device_node_tree(PCI_DN(pc), dent); 103317f06deSGavin Shan pc = pc->sibling; 104317f06deSGavin Shan } 105317f06deSGavin Shan } 106317f06deSGavin Shan #endif 107317f06deSGavin Shan 108317f06deSGavin Shan /** 109317f06deSGavin Shan * eeh_disable_irq - Disable interrupt for the recovering device 110317f06deSGavin Shan * @dev: PCI device 111317f06deSGavin Shan * 112317f06deSGavin Shan * This routine must be called when reporting temporary or permanent 113317f06deSGavin Shan * error to the particular PCI device to disable interrupt of that 114317f06deSGavin Shan * device. If the device has enabled MSI or MSI-X interrupt, we needn't 115317f06deSGavin Shan * do real work because EEH should freeze DMA transfers for those PCI 116317f06deSGavin Shan * devices encountering EEH errors, which includes MSI or MSI-X. 117317f06deSGavin Shan */ 118317f06deSGavin Shan static void eeh_disable_irq(struct pci_dev *dev) 119317f06deSGavin Shan { 120317f06deSGavin Shan struct eeh_dev *edev = pci_dev_to_eeh_dev(dev); 121317f06deSGavin Shan 122317f06deSGavin Shan /* Don't disable MSI and MSI-X interrupts. They are 123317f06deSGavin Shan * effectively disabled by the DMA Stopped state 124317f06deSGavin Shan * when an EEH error occurs. 125317f06deSGavin Shan */ 126317f06deSGavin Shan if (dev->msi_enabled || dev->msix_enabled) 127317f06deSGavin Shan return; 128317f06deSGavin Shan 129317f06deSGavin Shan if (!irq_has_action(dev->irq)) 130317f06deSGavin Shan return; 131317f06deSGavin Shan 132317f06deSGavin Shan edev->mode |= EEH_DEV_IRQ_DISABLED; 133317f06deSGavin Shan disable_irq_nosync(dev->irq); 134317f06deSGavin Shan } 135317f06deSGavin Shan 136317f06deSGavin Shan /** 137317f06deSGavin Shan * eeh_enable_irq - Enable interrupt for the recovering device 138317f06deSGavin Shan * @dev: PCI device 139317f06deSGavin Shan * 140317f06deSGavin Shan * This routine must be called to enable interrupt while failed 141317f06deSGavin Shan * device could be resumed. 142317f06deSGavin Shan */ 143317f06deSGavin Shan static void eeh_enable_irq(struct pci_dev *dev) 144317f06deSGavin Shan { 145317f06deSGavin Shan struct eeh_dev *edev = pci_dev_to_eeh_dev(dev); 14691150af3SGavin Shan struct irq_desc *desc; 147317f06deSGavin Shan 148317f06deSGavin Shan if ((edev->mode) & EEH_DEV_IRQ_DISABLED) { 149317f06deSGavin Shan edev->mode &= ~EEH_DEV_IRQ_DISABLED; 15091150af3SGavin Shan 15191150af3SGavin Shan desc = irq_to_desc(dev->irq); 15291150af3SGavin Shan if (desc && desc->depth > 0) 153317f06deSGavin Shan enable_irq(dev->irq); 154317f06deSGavin Shan } 155317f06deSGavin Shan } 156317f06deSGavin Shan 157317f06deSGavin Shan /** 158317f06deSGavin Shan * eeh_report_error - Report pci error to each device driver 159317f06deSGavin Shan * @data: eeh device 160317f06deSGavin Shan * @userdata: return value 161317f06deSGavin Shan * 162317f06deSGavin Shan * Report an EEH error to each device driver, collect up and 163317f06deSGavin Shan * merge the device driver responses. Cumulative response 164317f06deSGavin Shan * passed back in "userdata". 165317f06deSGavin Shan */ 166317f06deSGavin Shan static void *eeh_report_error(void *data, void *userdata) 167317f06deSGavin Shan { 168317f06deSGavin Shan struct eeh_dev *edev = (struct eeh_dev *)data; 169317f06deSGavin Shan struct pci_dev *dev = eeh_dev_to_pci_dev(edev); 170317f06deSGavin Shan enum pci_ers_result rc, *res = userdata; 171317f06deSGavin Shan struct pci_driver *driver; 172317f06deSGavin Shan 173317f06deSGavin Shan /* We might not have the associated PCI device, 174317f06deSGavin Shan * then we should continue for next one. 175317f06deSGavin Shan */ 176317f06deSGavin Shan if (!dev) return NULL; 177317f06deSGavin Shan dev->error_state = pci_channel_io_frozen; 178317f06deSGavin Shan 179317f06deSGavin Shan driver = eeh_pcid_get(dev); 180317f06deSGavin Shan if (!driver) return NULL; 181317f06deSGavin Shan 182317f06deSGavin Shan eeh_disable_irq(dev); 183317f06deSGavin Shan 184317f06deSGavin Shan if (!driver->err_handler || 185317f06deSGavin Shan !driver->err_handler->error_detected) { 186317f06deSGavin Shan eeh_pcid_put(dev); 187317f06deSGavin Shan return NULL; 188317f06deSGavin Shan } 189317f06deSGavin Shan 190317f06deSGavin Shan rc = driver->err_handler->error_detected(dev, pci_channel_io_frozen); 191317f06deSGavin Shan 192317f06deSGavin Shan /* A driver that needs a reset trumps all others */ 193317f06deSGavin Shan if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; 194317f06deSGavin Shan if (*res == PCI_ERS_RESULT_NONE) *res = rc; 195317f06deSGavin Shan 196317f06deSGavin Shan eeh_pcid_put(dev); 197317f06deSGavin Shan return NULL; 198317f06deSGavin Shan } 199317f06deSGavin Shan 200317f06deSGavin Shan /** 201317f06deSGavin Shan * eeh_report_mmio_enabled - Tell drivers that MMIO has been enabled 202317f06deSGavin Shan * @data: eeh device 203317f06deSGavin Shan * @userdata: return value 204317f06deSGavin Shan * 205317f06deSGavin Shan * Tells each device driver that IO ports, MMIO and config space I/O 206317f06deSGavin Shan * are now enabled. Collects up and merges the device driver responses. 207317f06deSGavin Shan * Cumulative response passed back in "userdata". 208317f06deSGavin Shan */ 209317f06deSGavin Shan static void *eeh_report_mmio_enabled(void *data, void *userdata) 210317f06deSGavin Shan { 211317f06deSGavin Shan struct eeh_dev *edev = (struct eeh_dev *)data; 212317f06deSGavin Shan struct pci_dev *dev = eeh_dev_to_pci_dev(edev); 213317f06deSGavin Shan enum pci_ers_result rc, *res = userdata; 214317f06deSGavin Shan struct pci_driver *driver; 215317f06deSGavin Shan 216317f06deSGavin Shan driver = eeh_pcid_get(dev); 217317f06deSGavin Shan if (!driver) return NULL; 218317f06deSGavin Shan 219317f06deSGavin Shan if (!driver->err_handler || 220317f06deSGavin Shan !driver->err_handler->mmio_enabled) { 221317f06deSGavin Shan eeh_pcid_put(dev); 222317f06deSGavin Shan return NULL; 223317f06deSGavin Shan } 224317f06deSGavin Shan 225317f06deSGavin Shan rc = driver->err_handler->mmio_enabled(dev); 226317f06deSGavin Shan 227317f06deSGavin Shan /* A driver that needs a reset trumps all others */ 228317f06deSGavin Shan if (rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; 229317f06deSGavin Shan if (*res == PCI_ERS_RESULT_NONE) *res = rc; 230317f06deSGavin Shan 231317f06deSGavin Shan eeh_pcid_put(dev); 232317f06deSGavin Shan return NULL; 233317f06deSGavin Shan } 234317f06deSGavin Shan 235317f06deSGavin Shan /** 236317f06deSGavin Shan * eeh_report_reset - Tell device that slot has been reset 237317f06deSGavin Shan * @data: eeh device 238317f06deSGavin Shan * @userdata: return value 239317f06deSGavin Shan * 240317f06deSGavin Shan * This routine must be called while EEH tries to reset particular 241317f06deSGavin Shan * PCI device so that the associated PCI device driver could take 242317f06deSGavin Shan * some actions, usually to save data the driver needs so that the 243317f06deSGavin Shan * driver can work again while the device is recovered. 244317f06deSGavin Shan */ 245317f06deSGavin Shan static void *eeh_report_reset(void *data, void *userdata) 246317f06deSGavin Shan { 247317f06deSGavin Shan struct eeh_dev *edev = (struct eeh_dev *)data; 248317f06deSGavin Shan struct pci_dev *dev = eeh_dev_to_pci_dev(edev); 249317f06deSGavin Shan enum pci_ers_result rc, *res = userdata; 250317f06deSGavin Shan struct pci_driver *driver; 251317f06deSGavin Shan 252317f06deSGavin Shan if (!dev) return NULL; 253317f06deSGavin Shan dev->error_state = pci_channel_io_normal; 254317f06deSGavin Shan 255317f06deSGavin Shan driver = eeh_pcid_get(dev); 256317f06deSGavin Shan if (!driver) return NULL; 257317f06deSGavin Shan 258317f06deSGavin Shan eeh_enable_irq(dev); 259317f06deSGavin Shan 260317f06deSGavin Shan if (!driver->err_handler || 261317f06deSGavin Shan !driver->err_handler->slot_reset) { 262317f06deSGavin Shan eeh_pcid_put(dev); 263317f06deSGavin Shan return NULL; 264317f06deSGavin Shan } 265317f06deSGavin Shan 266317f06deSGavin Shan rc = driver->err_handler->slot_reset(dev); 267317f06deSGavin Shan if ((*res == PCI_ERS_RESULT_NONE) || 268317f06deSGavin Shan (*res == PCI_ERS_RESULT_RECOVERED)) *res = rc; 269317f06deSGavin Shan if (*res == PCI_ERS_RESULT_DISCONNECT && 270317f06deSGavin Shan rc == PCI_ERS_RESULT_NEED_RESET) *res = rc; 271317f06deSGavin Shan 272317f06deSGavin Shan eeh_pcid_put(dev); 273317f06deSGavin Shan return NULL; 274317f06deSGavin Shan } 275317f06deSGavin Shan 276317f06deSGavin Shan /** 277317f06deSGavin Shan * eeh_report_resume - Tell device to resume normal operations 278317f06deSGavin Shan * @data: eeh device 279317f06deSGavin Shan * @userdata: return value 280317f06deSGavin Shan * 281317f06deSGavin Shan * This routine must be called to notify the device driver that it 282317f06deSGavin Shan * could resume so that the device driver can do some initialization 283317f06deSGavin Shan * to make the recovered device work again. 284317f06deSGavin Shan */ 285317f06deSGavin Shan static void *eeh_report_resume(void *data, void *userdata) 286317f06deSGavin Shan { 287317f06deSGavin Shan struct eeh_dev *edev = (struct eeh_dev *)data; 288317f06deSGavin Shan struct pci_dev *dev = eeh_dev_to_pci_dev(edev); 289317f06deSGavin Shan struct pci_driver *driver; 290317f06deSGavin Shan 291317f06deSGavin Shan if (!dev) return NULL; 292317f06deSGavin Shan dev->error_state = pci_channel_io_normal; 293317f06deSGavin Shan 294317f06deSGavin Shan driver = eeh_pcid_get(dev); 295317f06deSGavin Shan if (!driver) return NULL; 296317f06deSGavin Shan 297317f06deSGavin Shan eeh_enable_irq(dev); 298317f06deSGavin Shan 299317f06deSGavin Shan if (!driver->err_handler || 300317f06deSGavin Shan !driver->err_handler->resume) { 301317f06deSGavin Shan eeh_pcid_put(dev); 302317f06deSGavin Shan return NULL; 303317f06deSGavin Shan } 304317f06deSGavin Shan 305317f06deSGavin Shan driver->err_handler->resume(dev); 306317f06deSGavin Shan 307317f06deSGavin Shan eeh_pcid_put(dev); 308317f06deSGavin Shan return NULL; 309317f06deSGavin Shan } 310317f06deSGavin Shan 311317f06deSGavin Shan /** 312317f06deSGavin Shan * eeh_report_failure - Tell device driver that device is dead. 313317f06deSGavin Shan * @data: eeh device 314317f06deSGavin Shan * @userdata: return value 315317f06deSGavin Shan * 316317f06deSGavin Shan * This informs the device driver that the device is permanently 317317f06deSGavin Shan * dead, and that no further recovery attempts will be made on it. 318317f06deSGavin Shan */ 319317f06deSGavin Shan static void *eeh_report_failure(void *data, void *userdata) 320317f06deSGavin Shan { 321317f06deSGavin Shan struct eeh_dev *edev = (struct eeh_dev *)data; 322317f06deSGavin Shan struct pci_dev *dev = eeh_dev_to_pci_dev(edev); 323317f06deSGavin Shan struct pci_driver *driver; 324317f06deSGavin Shan 325317f06deSGavin Shan if (!dev) return NULL; 326317f06deSGavin Shan dev->error_state = pci_channel_io_perm_failure; 327317f06deSGavin Shan 328317f06deSGavin Shan driver = eeh_pcid_get(dev); 329317f06deSGavin Shan if (!driver) return NULL; 330317f06deSGavin Shan 331317f06deSGavin Shan eeh_disable_irq(dev); 332317f06deSGavin Shan 333317f06deSGavin Shan if (!driver->err_handler || 334317f06deSGavin Shan !driver->err_handler->error_detected) { 335317f06deSGavin Shan eeh_pcid_put(dev); 336317f06deSGavin Shan return NULL; 337317f06deSGavin Shan } 338317f06deSGavin Shan 339317f06deSGavin Shan driver->err_handler->error_detected(dev, pci_channel_io_perm_failure); 340317f06deSGavin Shan 341317f06deSGavin Shan eeh_pcid_put(dev); 342317f06deSGavin Shan return NULL; 343317f06deSGavin Shan } 344317f06deSGavin Shan 345f5c57710SGavin Shan static void *eeh_rmv_device(void *data, void *userdata) 346f5c57710SGavin Shan { 347f5c57710SGavin Shan struct pci_driver *driver; 348f5c57710SGavin Shan struct eeh_dev *edev = (struct eeh_dev *)data; 349f5c57710SGavin Shan struct pci_dev *dev = eeh_dev_to_pci_dev(edev); 350f5c57710SGavin Shan int *removed = (int *)userdata; 351f5c57710SGavin Shan 352f5c57710SGavin Shan /* 353f5c57710SGavin Shan * Actually, we should remove the PCI bridges as well. 354f5c57710SGavin Shan * However, that's lots of complexity to do that, 355f5c57710SGavin Shan * particularly some of devices under the bridge might 356f5c57710SGavin Shan * support EEH. So we just care about PCI devices for 357f5c57710SGavin Shan * simplicity here. 358f5c57710SGavin Shan */ 359f5c57710SGavin Shan if (!dev || (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)) 360f5c57710SGavin Shan return NULL; 361f5c57710SGavin Shan driver = eeh_pcid_get(dev); 362f5c57710SGavin Shan if (driver && driver->err_handler) 363f5c57710SGavin Shan return NULL; 364f5c57710SGavin Shan 365f5c57710SGavin Shan /* Remove it from PCI subsystem */ 366f5c57710SGavin Shan pr_debug("EEH: Removing %s without EEH sensitive driver\n", 367f5c57710SGavin Shan pci_name(dev)); 368f5c57710SGavin Shan edev->bus = dev->bus; 369f5c57710SGavin Shan edev->mode |= EEH_DEV_DISCONNECTED; 370f5c57710SGavin Shan (*removed)++; 371f5c57710SGavin Shan 372f5c57710SGavin Shan pci_stop_and_remove_bus_device(dev); 373f5c57710SGavin Shan 374f5c57710SGavin Shan return NULL; 375f5c57710SGavin Shan } 376f5c57710SGavin Shan 377f5c57710SGavin Shan static void *eeh_pe_detach_dev(void *data, void *userdata) 378f5c57710SGavin Shan { 379f5c57710SGavin Shan struct eeh_pe *pe = (struct eeh_pe *)data; 380f5c57710SGavin Shan struct eeh_dev *edev, *tmp; 381f5c57710SGavin Shan 382f5c57710SGavin Shan eeh_pe_for_each_dev(pe, edev, tmp) { 383f5c57710SGavin Shan if (!(edev->mode & EEH_DEV_DISCONNECTED)) 384f5c57710SGavin Shan continue; 385f5c57710SGavin Shan 386f5c57710SGavin Shan edev->mode &= ~(EEH_DEV_DISCONNECTED | EEH_DEV_IRQ_DISABLED); 387f5c57710SGavin Shan eeh_rmv_from_parent_pe(edev); 388f5c57710SGavin Shan } 389f5c57710SGavin Shan 390f5c57710SGavin Shan return NULL; 391f5c57710SGavin Shan } 392f5c57710SGavin Shan 393317f06deSGavin Shan /** 394317f06deSGavin Shan * eeh_reset_device - Perform actual reset of a pci slot 395317f06deSGavin Shan * @pe: EEH PE 396317f06deSGavin Shan * @bus: PCI bus corresponding to the isolcated slot 397317f06deSGavin Shan * 398317f06deSGavin Shan * This routine must be called to do reset on the indicated PE. 399317f06deSGavin Shan * During the reset, udev might be invoked because those affected 400317f06deSGavin Shan * PCI devices will be removed and then added. 401317f06deSGavin Shan */ 402317f06deSGavin Shan static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus) 403317f06deSGavin Shan { 404f5c57710SGavin Shan struct pci_bus *frozen_bus = eeh_pe_bus_get(pe); 4055a71978eSGavin Shan struct timeval tstamp; 406f5c57710SGavin Shan int cnt, rc, removed = 0; 407317f06deSGavin Shan 408317f06deSGavin Shan /* pcibios will clear the counter; save the value */ 409317f06deSGavin Shan cnt = pe->freeze_count; 4105a71978eSGavin Shan tstamp = pe->tstamp; 411317f06deSGavin Shan 412317f06deSGavin Shan /* 413317f06deSGavin Shan * We don't remove the corresponding PE instances because 414317f06deSGavin Shan * we need the information afterwords. The attached EEH 415317f06deSGavin Shan * devices are expected to be attached soon when calling 416317f06deSGavin Shan * into pcibios_add_pci_devices(). 417317f06deSGavin Shan */ 418807a827dSGavin Shan eeh_pe_state_mark(pe, EEH_PE_KEEP); 419f5c57710SGavin Shan if (bus) 420807a827dSGavin Shan pcibios_remove_pci_devices(bus); 421f5c57710SGavin Shan else if (frozen_bus) 422f5c57710SGavin Shan eeh_pe_dev_traverse(pe, eeh_rmv_device, &removed); 423317f06deSGavin Shan 424317f06deSGavin Shan /* Reset the pci controller. (Asserts RST#; resets config space). 425317f06deSGavin Shan * Reconfigure bridges and devices. Don't try to bring the system 426317f06deSGavin Shan * up if the reset failed for some reason. 427317f06deSGavin Shan */ 428317f06deSGavin Shan rc = eeh_reset_pe(pe); 429317f06deSGavin Shan if (rc) 430317f06deSGavin Shan return rc; 431317f06deSGavin Shan 432317f06deSGavin Shan /* Restore PE */ 433317f06deSGavin Shan eeh_ops->configure_bridge(pe); 434317f06deSGavin Shan eeh_pe_restore_bars(pe); 435317f06deSGavin Shan 436317f06deSGavin Shan /* Give the system 5 seconds to finish running the user-space 437317f06deSGavin Shan * hotplug shutdown scripts, e.g. ifdown for ethernet. Yes, 438317f06deSGavin Shan * this is a hack, but if we don't do this, and try to bring 439317f06deSGavin Shan * the device up before the scripts have taken it down, 440317f06deSGavin Shan * potentially weird things happen. 441317f06deSGavin Shan */ 442317f06deSGavin Shan if (bus) { 443f5c57710SGavin Shan pr_info("EEH: Sleep 5s ahead of complete hotplug\n"); 444317f06deSGavin Shan ssleep(5); 445f5c57710SGavin Shan 446f5c57710SGavin Shan /* 447f5c57710SGavin Shan * The EEH device is still connected with its parent 448f5c57710SGavin Shan * PE. We should disconnect it so the binding can be 449f5c57710SGavin Shan * rebuilt when adding PCI devices. 450f5c57710SGavin Shan */ 451f5c57710SGavin Shan eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL); 452317f06deSGavin Shan pcibios_add_pci_devices(bus); 453f5c57710SGavin Shan } else if (frozen_bus && removed) { 454f5c57710SGavin Shan pr_info("EEH: Sleep 5s ahead of partial hotplug\n"); 455f5c57710SGavin Shan ssleep(5); 456f5c57710SGavin Shan 457f5c57710SGavin Shan eeh_pe_traverse(pe, eeh_pe_detach_dev, NULL); 458f5c57710SGavin Shan pcibios_add_pci_devices(frozen_bus); 459317f06deSGavin Shan } 460f5c57710SGavin Shan eeh_pe_state_clear(pe, EEH_PE_KEEP); 4615a71978eSGavin Shan 4625a71978eSGavin Shan pe->tstamp = tstamp; 463317f06deSGavin Shan pe->freeze_count = cnt; 464317f06deSGavin Shan 465317f06deSGavin Shan return 0; 466317f06deSGavin Shan } 467317f06deSGavin Shan 468317f06deSGavin Shan /* The longest amount of time to wait for a pci device 469317f06deSGavin Shan * to come back on line, in seconds. 470317f06deSGavin Shan */ 471fb48dc22SBrian King #define MAX_WAIT_FOR_RECOVERY 300 472317f06deSGavin Shan 4738a6b1bc7SGavin Shan static void eeh_handle_normal_event(struct eeh_pe *pe) 474317f06deSGavin Shan { 475317f06deSGavin Shan struct pci_bus *frozen_bus; 476317f06deSGavin Shan int rc = 0; 477317f06deSGavin Shan enum pci_ers_result result = PCI_ERS_RESULT_NONE; 478317f06deSGavin Shan 479317f06deSGavin Shan frozen_bus = eeh_pe_bus_get(pe); 480317f06deSGavin Shan if (!frozen_bus) { 481317f06deSGavin Shan pr_err("%s: Cannot find PCI bus for PHB#%d-PE#%x\n", 482317f06deSGavin Shan __func__, pe->phb->global_number, pe->addr); 483317f06deSGavin Shan return; 484317f06deSGavin Shan } 485317f06deSGavin Shan 4865a71978eSGavin Shan eeh_pe_update_time_stamp(pe); 487317f06deSGavin Shan pe->freeze_count++; 488317f06deSGavin Shan if (pe->freeze_count > EEH_MAX_ALLOWED_FREEZES) 489317f06deSGavin Shan goto excess_failures; 490317f06deSGavin Shan pr_warning("EEH: This PCI device has failed %d times in the last hour\n", 491317f06deSGavin Shan pe->freeze_count); 492317f06deSGavin Shan 493317f06deSGavin Shan /* Walk the various device drivers attached to this slot through 494317f06deSGavin Shan * a reset sequence, giving each an opportunity to do what it needs 495317f06deSGavin Shan * to accomplish the reset. Each child gets a report of the 496317f06deSGavin Shan * status ... if any child can't handle the reset, then the entire 497317f06deSGavin Shan * slot is dlpar removed and added. 498317f06deSGavin Shan */ 49956ca4fdeSGavin Shan pr_info("EEH: Notify device drivers to shutdown\n"); 500317f06deSGavin Shan eeh_pe_dev_traverse(pe, eeh_report_error, &result); 501317f06deSGavin Shan 502317f06deSGavin Shan /* Get the current PCI slot state. This can take a long time, 503317f06deSGavin Shan * sometimes over 3 seconds for certain systems. 504317f06deSGavin Shan */ 505317f06deSGavin Shan rc = eeh_ops->wait_state(pe, MAX_WAIT_FOR_RECOVERY*1000); 506317f06deSGavin Shan if (rc < 0 || rc == EEH_STATE_NOT_SUPPORT) { 50756ca4fdeSGavin Shan pr_warning("EEH: Permanent failure\n"); 508317f06deSGavin Shan goto hard_fail; 509317f06deSGavin Shan } 510317f06deSGavin Shan 511317f06deSGavin Shan /* Since rtas may enable MMIO when posting the error log, 512317f06deSGavin Shan * don't post the error log until after all dev drivers 513317f06deSGavin Shan * have been informed. 514317f06deSGavin Shan */ 51556ca4fdeSGavin Shan pr_info("EEH: Collect temporary log\n"); 516317f06deSGavin Shan eeh_slot_error_detail(pe, EEH_LOG_TEMP); 517317f06deSGavin Shan 518317f06deSGavin Shan /* If all device drivers were EEH-unaware, then shut 519317f06deSGavin Shan * down all of the device drivers, and hope they 520317f06deSGavin Shan * go down willingly, without panicing the system. 521317f06deSGavin Shan */ 522317f06deSGavin Shan if (result == PCI_ERS_RESULT_NONE) { 52356ca4fdeSGavin Shan pr_info("EEH: Reset with hotplug activity\n"); 524317f06deSGavin Shan rc = eeh_reset_device(pe, frozen_bus); 525317f06deSGavin Shan if (rc) { 52656ca4fdeSGavin Shan pr_warning("%s: Unable to reset, err=%d\n", 52756ca4fdeSGavin Shan __func__, rc); 528317f06deSGavin Shan goto hard_fail; 529317f06deSGavin Shan } 530317f06deSGavin Shan } 531317f06deSGavin Shan 532317f06deSGavin Shan /* If all devices reported they can proceed, then re-enable MMIO */ 533317f06deSGavin Shan if (result == PCI_ERS_RESULT_CAN_RECOVER) { 53456ca4fdeSGavin Shan pr_info("EEH: Enable I/O for affected devices\n"); 535317f06deSGavin Shan rc = eeh_pci_enable(pe, EEH_OPT_THAW_MMIO); 536317f06deSGavin Shan 537317f06deSGavin Shan if (rc < 0) 538317f06deSGavin Shan goto hard_fail; 539317f06deSGavin Shan if (rc) { 540317f06deSGavin Shan result = PCI_ERS_RESULT_NEED_RESET; 541317f06deSGavin Shan } else { 54256ca4fdeSGavin Shan pr_info("EEH: Notify device drivers to resume I/O\n"); 543317f06deSGavin Shan result = PCI_ERS_RESULT_NONE; 544317f06deSGavin Shan eeh_pe_dev_traverse(pe, eeh_report_mmio_enabled, &result); 545317f06deSGavin Shan } 546317f06deSGavin Shan } 547317f06deSGavin Shan 548317f06deSGavin Shan /* If all devices reported they can proceed, then re-enable DMA */ 549317f06deSGavin Shan if (result == PCI_ERS_RESULT_CAN_RECOVER) { 55056ca4fdeSGavin Shan pr_info("EEH: Enabled DMA for affected devices\n"); 551317f06deSGavin Shan rc = eeh_pci_enable(pe, EEH_OPT_THAW_DMA); 552317f06deSGavin Shan 553317f06deSGavin Shan if (rc < 0) 554317f06deSGavin Shan goto hard_fail; 555317f06deSGavin Shan if (rc) 556317f06deSGavin Shan result = PCI_ERS_RESULT_NEED_RESET; 557317f06deSGavin Shan else 558317f06deSGavin Shan result = PCI_ERS_RESULT_RECOVERED; 559317f06deSGavin Shan } 560317f06deSGavin Shan 561317f06deSGavin Shan /* If any device has a hard failure, then shut off everything. */ 562317f06deSGavin Shan if (result == PCI_ERS_RESULT_DISCONNECT) { 56356ca4fdeSGavin Shan pr_warning("EEH: Device driver gave up\n"); 564317f06deSGavin Shan goto hard_fail; 565317f06deSGavin Shan } 566317f06deSGavin Shan 567317f06deSGavin Shan /* If any device called out for a reset, then reset the slot */ 568317f06deSGavin Shan if (result == PCI_ERS_RESULT_NEED_RESET) { 56956ca4fdeSGavin Shan pr_info("EEH: Reset without hotplug activity\n"); 570317f06deSGavin Shan rc = eeh_reset_device(pe, NULL); 571317f06deSGavin Shan if (rc) { 57256ca4fdeSGavin Shan pr_warning("%s: Cannot reset, err=%d\n", 57356ca4fdeSGavin Shan __func__, rc); 574317f06deSGavin Shan goto hard_fail; 575317f06deSGavin Shan } 57656ca4fdeSGavin Shan 57756ca4fdeSGavin Shan pr_info("EEH: Notify device drivers " 57856ca4fdeSGavin Shan "the completion of reset\n"); 579317f06deSGavin Shan result = PCI_ERS_RESULT_NONE; 580317f06deSGavin Shan eeh_pe_dev_traverse(pe, eeh_report_reset, &result); 581317f06deSGavin Shan } 582317f06deSGavin Shan 583317f06deSGavin Shan /* All devices should claim they have recovered by now. */ 584317f06deSGavin Shan if ((result != PCI_ERS_RESULT_RECOVERED) && 585317f06deSGavin Shan (result != PCI_ERS_RESULT_NONE)) { 58656ca4fdeSGavin Shan pr_warning("EEH: Not recovered\n"); 587317f06deSGavin Shan goto hard_fail; 588317f06deSGavin Shan } 589317f06deSGavin Shan 590317f06deSGavin Shan /* Tell all device drivers that they can resume operations */ 59156ca4fdeSGavin Shan pr_info("EEH: Notify device driver to resume\n"); 592317f06deSGavin Shan eeh_pe_dev_traverse(pe, eeh_report_resume, NULL); 593317f06deSGavin Shan 594317f06deSGavin Shan return; 595317f06deSGavin Shan 596317f06deSGavin Shan excess_failures: 597317f06deSGavin Shan /* 598317f06deSGavin Shan * About 90% of all real-life EEH failures in the field 599317f06deSGavin Shan * are due to poorly seated PCI cards. Only 10% or so are 600317f06deSGavin Shan * due to actual, failed cards. 601317f06deSGavin Shan */ 602317f06deSGavin Shan pr_err("EEH: PHB#%d-PE#%x has failed %d times in the\n" 603317f06deSGavin Shan "last hour and has been permanently disabled.\n" 604317f06deSGavin Shan "Please try reseating or replacing it.\n", 605317f06deSGavin Shan pe->phb->global_number, pe->addr, 606317f06deSGavin Shan pe->freeze_count); 607317f06deSGavin Shan goto perm_error; 608317f06deSGavin Shan 609317f06deSGavin Shan hard_fail: 610317f06deSGavin Shan pr_err("EEH: Unable to recover from failure from PHB#%d-PE#%x.\n" 611317f06deSGavin Shan "Please try reseating or replacing it\n", 612317f06deSGavin Shan pe->phb->global_number, pe->addr); 613317f06deSGavin Shan 614317f06deSGavin Shan perm_error: 615317f06deSGavin Shan eeh_slot_error_detail(pe, EEH_LOG_PERM); 616317f06deSGavin Shan 617317f06deSGavin Shan /* Notify all devices that they're about to go down. */ 618317f06deSGavin Shan eeh_pe_dev_traverse(pe, eeh_report_failure, NULL); 619317f06deSGavin Shan 620317f06deSGavin Shan /* Shut down the device drivers for good. */ 621317f06deSGavin Shan if (frozen_bus) 622317f06deSGavin Shan pcibios_remove_pci_devices(frozen_bus); 623317f06deSGavin Shan } 6248a6b1bc7SGavin Shan 6258a6b1bc7SGavin Shan static void eeh_handle_special_event(void) 6268a6b1bc7SGavin Shan { 6278a6b1bc7SGavin Shan struct eeh_pe *pe, *phb_pe; 6288a6b1bc7SGavin Shan struct pci_bus *bus; 6298a6b1bc7SGavin Shan struct pci_controller *hose, *tmp; 6308a6b1bc7SGavin Shan unsigned long flags; 6318a6b1bc7SGavin Shan int rc = 0; 6328a6b1bc7SGavin Shan 6338a6b1bc7SGavin Shan /* 6348a6b1bc7SGavin Shan * The return value from next_error() has been classified as follows. 6358a6b1bc7SGavin Shan * It might be good to enumerate them. However, next_error() is only 6368a6b1bc7SGavin Shan * supported by PowerNV platform for now. So it would be fine to use 6378a6b1bc7SGavin Shan * integer directly: 6388a6b1bc7SGavin Shan * 6398a6b1bc7SGavin Shan * 4 - Dead IOC 3 - Dead PHB 6408a6b1bc7SGavin Shan * 2 - Fenced PHB 1 - Frozen PE 6418a6b1bc7SGavin Shan * 0 - No error found 6428a6b1bc7SGavin Shan * 6438a6b1bc7SGavin Shan */ 6448a6b1bc7SGavin Shan rc = eeh_ops->next_error(&pe); 6458a6b1bc7SGavin Shan if (rc <= 0) 6468a6b1bc7SGavin Shan return; 6478a6b1bc7SGavin Shan 6488a6b1bc7SGavin Shan switch (rc) { 6498a6b1bc7SGavin Shan case 4: 6508a6b1bc7SGavin Shan /* Mark all PHBs in dead state */ 6518a6b1bc7SGavin Shan eeh_serialize_lock(&flags); 6528a6b1bc7SGavin Shan list_for_each_entry_safe(hose, tmp, 6538a6b1bc7SGavin Shan &hose_list, list_node) { 6548a6b1bc7SGavin Shan phb_pe = eeh_phb_pe_get(hose); 6558a6b1bc7SGavin Shan if (!phb_pe) continue; 6568a6b1bc7SGavin Shan 6578a6b1bc7SGavin Shan eeh_pe_state_mark(phb_pe, 6588a6b1bc7SGavin Shan EEH_PE_ISOLATED | EEH_PE_PHB_DEAD); 6598a6b1bc7SGavin Shan } 6608a6b1bc7SGavin Shan eeh_serialize_unlock(flags); 6618a6b1bc7SGavin Shan 6628a6b1bc7SGavin Shan /* Purge all events */ 6638a6b1bc7SGavin Shan eeh_remove_event(NULL); 6648a6b1bc7SGavin Shan break; 6658a6b1bc7SGavin Shan case 3: 6668a6b1bc7SGavin Shan case 2: 6678a6b1bc7SGavin Shan case 1: 6688a6b1bc7SGavin Shan /* Mark the PE in fenced state */ 6698a6b1bc7SGavin Shan eeh_serialize_lock(&flags); 6708a6b1bc7SGavin Shan if (rc == 3) 6718a6b1bc7SGavin Shan eeh_pe_state_mark(pe, 6728a6b1bc7SGavin Shan EEH_PE_ISOLATED | EEH_PE_PHB_DEAD); 6738a6b1bc7SGavin Shan else 6748a6b1bc7SGavin Shan eeh_pe_state_mark(pe, 6758a6b1bc7SGavin Shan EEH_PE_ISOLATED | EEH_PE_RECOVERING); 6768a6b1bc7SGavin Shan eeh_serialize_unlock(flags); 6778a6b1bc7SGavin Shan 6788a6b1bc7SGavin Shan /* Purge all events of the PHB */ 6798a6b1bc7SGavin Shan eeh_remove_event(pe); 6808a6b1bc7SGavin Shan break; 6818a6b1bc7SGavin Shan default: 6828a6b1bc7SGavin Shan pr_err("%s: Invalid value %d from next_error()\n", 6838a6b1bc7SGavin Shan __func__, rc); 6848a6b1bc7SGavin Shan return; 6858a6b1bc7SGavin Shan } 6868a6b1bc7SGavin Shan 6878a6b1bc7SGavin Shan /* 6888a6b1bc7SGavin Shan * For fenced PHB and frozen PE, it's handled as normal 6898a6b1bc7SGavin Shan * event. We have to remove the affected PHBs for dead 6908a6b1bc7SGavin Shan * PHB and IOC 6918a6b1bc7SGavin Shan */ 6928a6b1bc7SGavin Shan if (rc == 2 || rc == 1) 6938a6b1bc7SGavin Shan eeh_handle_normal_event(pe); 6948a6b1bc7SGavin Shan else { 6958a6b1bc7SGavin Shan list_for_each_entry_safe(hose, tmp, 6968a6b1bc7SGavin Shan &hose_list, list_node) { 6978a6b1bc7SGavin Shan phb_pe = eeh_phb_pe_get(hose); 6988a6b1bc7SGavin Shan if (!phb_pe || !(phb_pe->state & EEH_PE_PHB_DEAD)) 6998a6b1bc7SGavin Shan continue; 7008a6b1bc7SGavin Shan 7018a6b1bc7SGavin Shan bus = eeh_pe_bus_get(phb_pe); 7028a6b1bc7SGavin Shan /* Notify all devices that they're about to go down. */ 7038a6b1bc7SGavin Shan eeh_pe_dev_traverse(pe, eeh_report_failure, NULL); 7048a6b1bc7SGavin Shan pcibios_remove_pci_devices(bus); 7058a6b1bc7SGavin Shan } 7068a6b1bc7SGavin Shan } 7078a6b1bc7SGavin Shan } 7088a6b1bc7SGavin Shan 7098a6b1bc7SGavin Shan /** 7108a6b1bc7SGavin Shan * eeh_handle_event - Reset a PCI device after hard lockup. 7118a6b1bc7SGavin Shan * @pe: EEH PE 7128a6b1bc7SGavin Shan * 7138a6b1bc7SGavin Shan * While PHB detects address or data parity errors on particular PCI 7148a6b1bc7SGavin Shan * slot, the associated PE will be frozen. Besides, DMA's occurring 7158a6b1bc7SGavin Shan * to wild addresses (which usually happen due to bugs in device 7168a6b1bc7SGavin Shan * drivers or in PCI adapter firmware) can cause EEH error. #SERR, 7178a6b1bc7SGavin Shan * #PERR or other misc PCI-related errors also can trigger EEH errors. 7188a6b1bc7SGavin Shan * 7198a6b1bc7SGavin Shan * Recovery process consists of unplugging the device driver (which 7208a6b1bc7SGavin Shan * generated hotplug events to userspace), then issuing a PCI #RST to 7218a6b1bc7SGavin Shan * the device, then reconfiguring the PCI config space for all bridges 7228a6b1bc7SGavin Shan * & devices under this slot, and then finally restarting the device 7238a6b1bc7SGavin Shan * drivers (which cause a second set of hotplug events to go out to 7248a6b1bc7SGavin Shan * userspace). 7258a6b1bc7SGavin Shan */ 7268a6b1bc7SGavin Shan void eeh_handle_event(struct eeh_pe *pe) 7278a6b1bc7SGavin Shan { 7288a6b1bc7SGavin Shan if (pe) 7298a6b1bc7SGavin Shan eeh_handle_normal_event(pe); 7308a6b1bc7SGavin Shan else 7318a6b1bc7SGavin Shan eeh_handle_special_event(); 7328a6b1bc7SGavin Shan } 733