1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * This file implements the error recovery as a core part of PCIe error 4 * reporting. When a PCIe error is delivered, an error message will be 5 * collected and printed to console, then, an error recovery procedure 6 * will be executed by following the PCI error recovery rules. 7 * 8 * Copyright (C) 2006 Intel Corp. 9 * Tom Long Nguyen (tom.l.nguyen@intel.com) 10 * Zhang Yanmin (yanmin.zhang@intel.com) 11 */ 12 13 #define dev_fmt(fmt) "AER: " fmt 14 15 #include <linux/pci.h> 16 #include <linux/module.h> 17 #include <linux/kernel.h> 18 #include <linux/errno.h> 19 #include <linux/aer.h> 20 #include "portdrv.h" 21 #include "../pci.h" 22 23 static pci_ers_result_t merge_result(enum pci_ers_result orig, 24 enum pci_ers_result new) 25 { 26 if (new == PCI_ERS_RESULT_NO_AER_DRIVER) 27 return PCI_ERS_RESULT_NO_AER_DRIVER; 28 29 if (new == PCI_ERS_RESULT_NONE) 30 return orig; 31 32 switch (orig) { 33 case PCI_ERS_RESULT_CAN_RECOVER: 34 case PCI_ERS_RESULT_RECOVERED: 35 orig = new; 36 break; 37 case PCI_ERS_RESULT_DISCONNECT: 38 if (new == PCI_ERS_RESULT_NEED_RESET) 39 orig = PCI_ERS_RESULT_NEED_RESET; 40 break; 41 default: 42 break; 43 } 44 45 return orig; 46 } 47 48 static int report_error_detected(struct pci_dev *dev, 49 pci_channel_state_t state, 50 enum pci_ers_result *result) 51 { 52 pci_ers_result_t vote; 53 const struct pci_error_handlers *err_handler; 54 55 device_lock(&dev->dev); 56 if (!pci_dev_set_io_state(dev, state) || 57 !dev->driver || 58 !dev->driver->err_handler || 59 !dev->driver->err_handler->error_detected) { 60 /* 61 * If any device in the subtree does not have an error_detected 62 * callback, PCI_ERS_RESULT_NO_AER_DRIVER prevents subsequent 63 * error callbacks of "any" device in the subtree, and will 64 * exit in the disconnected error state. 65 */ 66 if (dev->hdr_type != PCI_HEADER_TYPE_BRIDGE) { 67 vote = PCI_ERS_RESULT_NO_AER_DRIVER; 68 pci_info(dev, "can't recover (no error_detected callback)\n"); 69 } else { 70 vote = PCI_ERS_RESULT_NONE; 71 } 72 } else { 73 err_handler = dev->driver->err_handler; 74 vote = err_handler->error_detected(dev, state); 75 } 76 pci_uevent_ers(dev, vote); 77 *result = merge_result(*result, vote); 78 device_unlock(&dev->dev); 79 return 0; 80 } 81 82 static int report_frozen_detected(struct pci_dev *dev, void *data) 83 { 84 return report_error_detected(dev, pci_channel_io_frozen, data); 85 } 86 87 static int report_normal_detected(struct pci_dev *dev, void *data) 88 { 89 return report_error_detected(dev, pci_channel_io_normal, data); 90 } 91 92 static int report_mmio_enabled(struct pci_dev *dev, void *data) 93 { 94 pci_ers_result_t vote, *result = data; 95 const struct pci_error_handlers *err_handler; 96 97 device_lock(&dev->dev); 98 if (!dev->driver || 99 !dev->driver->err_handler || 100 !dev->driver->err_handler->mmio_enabled) 101 goto out; 102 103 err_handler = dev->driver->err_handler; 104 vote = err_handler->mmio_enabled(dev); 105 *result = merge_result(*result, vote); 106 out: 107 device_unlock(&dev->dev); 108 return 0; 109 } 110 111 static int report_slot_reset(struct pci_dev *dev, void *data) 112 { 113 pci_ers_result_t vote, *result = data; 114 const struct pci_error_handlers *err_handler; 115 116 device_lock(&dev->dev); 117 if (!dev->driver || 118 !dev->driver->err_handler || 119 !dev->driver->err_handler->slot_reset) 120 goto out; 121 122 err_handler = dev->driver->err_handler; 123 vote = err_handler->slot_reset(dev); 124 *result = merge_result(*result, vote); 125 out: 126 device_unlock(&dev->dev); 127 return 0; 128 } 129 130 static int report_resume(struct pci_dev *dev, void *data) 131 { 132 const struct pci_error_handlers *err_handler; 133 134 device_lock(&dev->dev); 135 if (!pci_dev_set_io_state(dev, pci_channel_io_normal) || 136 !dev->driver || 137 !dev->driver->err_handler || 138 !dev->driver->err_handler->resume) 139 goto out; 140 141 err_handler = dev->driver->err_handler; 142 err_handler->resume(dev); 143 out: 144 pci_uevent_ers(dev, PCI_ERS_RESULT_RECOVERED); 145 device_unlock(&dev->dev); 146 return 0; 147 } 148 149 /** 150 * pci_walk_bridge - walk bridges potentially AER affected 151 * @bridge: bridge which may be a Port, an RCEC, or an RCiEP 152 * @cb: callback to be called for each device found 153 * @userdata: arbitrary pointer to be passed to callback 154 * 155 * If the device provided is a bridge, walk the subordinate bus, including 156 * any bridged devices on buses under this bus. Call the provided callback 157 * on each device found. 158 * 159 * If the device provided has no subordinate bus, e.g., an RCEC or RCiEP, 160 * call the callback on the device itself. 161 */ 162 static void pci_walk_bridge(struct pci_dev *bridge, 163 int (*cb)(struct pci_dev *, void *), 164 void *userdata) 165 { 166 if (bridge->subordinate) 167 pci_walk_bus(bridge->subordinate, cb, userdata); 168 else 169 cb(bridge, userdata); 170 } 171 172 pci_ers_result_t pcie_do_recovery(struct pci_dev *dev, 173 pci_channel_state_t state, 174 pci_ers_result_t (*reset_subordinates)(struct pci_dev *pdev)) 175 { 176 int type = pci_pcie_type(dev); 177 struct pci_dev *bridge; 178 pci_ers_result_t status = PCI_ERS_RESULT_CAN_RECOVER; 179 struct pci_host_bridge *host = pci_find_host_bridge(dev->bus); 180 181 /* 182 * If the error was detected by a Root Port, Downstream Port, RCEC, 183 * or RCiEP, recovery runs on the device itself. For Ports, that 184 * also includes any subordinate devices. 185 * 186 * If it was detected by another device (Endpoint, etc), recovery 187 * runs on the device and anything else under the same Port, i.e., 188 * everything under "bridge". 189 */ 190 if (type == PCI_EXP_TYPE_ROOT_PORT || 191 type == PCI_EXP_TYPE_DOWNSTREAM || 192 type == PCI_EXP_TYPE_RC_EC || 193 type == PCI_EXP_TYPE_RC_END) 194 bridge = dev; 195 else 196 bridge = pci_upstream_bridge(dev); 197 198 pci_dbg(bridge, "broadcast error_detected message\n"); 199 if (state == pci_channel_io_frozen) { 200 pci_walk_bridge(bridge, report_frozen_detected, &status); 201 if (reset_subordinates(bridge) != PCI_ERS_RESULT_RECOVERED) { 202 pci_warn(bridge, "subordinate device reset failed\n"); 203 goto failed; 204 } 205 } else { 206 pci_walk_bridge(bridge, report_normal_detected, &status); 207 } 208 209 if (status == PCI_ERS_RESULT_CAN_RECOVER) { 210 status = PCI_ERS_RESULT_RECOVERED; 211 pci_dbg(bridge, "broadcast mmio_enabled message\n"); 212 pci_walk_bridge(bridge, report_mmio_enabled, &status); 213 } 214 215 if (status == PCI_ERS_RESULT_NEED_RESET) { 216 /* 217 * TODO: Should call platform-specific 218 * functions to reset slot before calling 219 * drivers' slot_reset callbacks? 220 */ 221 status = PCI_ERS_RESULT_RECOVERED; 222 pci_dbg(bridge, "broadcast slot_reset message\n"); 223 pci_walk_bridge(bridge, report_slot_reset, &status); 224 } 225 226 if (status != PCI_ERS_RESULT_RECOVERED) 227 goto failed; 228 229 pci_dbg(bridge, "broadcast resume message\n"); 230 pci_walk_bridge(bridge, report_resume, &status); 231 232 /* 233 * If we have native control of AER, clear error status in the device 234 * that detected the error. If the platform retained control of AER, 235 * it is responsible for clearing this status. In that case, the 236 * signaling device may not even be visible to the OS. 237 */ 238 if (host->native_aer || pcie_ports_native) { 239 pcie_clear_device_status(dev); 240 pci_aer_clear_nonfatal_status(dev); 241 } 242 pci_info(bridge, "device recovery successful\n"); 243 return status; 244 245 failed: 246 pci_uevent_ers(bridge, PCI_ERS_RESULT_DISCONNECT); 247 248 /* TODO: Should kernel panic here? */ 249 pci_info(bridge, "device recovery failed\n"); 250 251 return status; 252 } 253