drivers/pci/pcie/err.c

   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/pm_runtime.h>
  17 #include <linux/module.h>
  18 #include <linux/kernel.h>
  19 #include <linux/errno.h>
  20 #include <linux/aer.h>
  21 #include "portdrv.h"
  22 #include "../pci.h"
  23
  24 static pci_ers_result_t merge_result(enum pci_ers_result orig,
  25                                   enum pci_ers_result new)
  26 {
  27         if (new == PCI_ERS_RESULT_NO_AER_DRIVER)
  28                 return PCI_ERS_RESULT_NO_AER_DRIVER;
  29
  30         if (new == PCI_ERS_RESULT_NONE)
  31                 return orig;
  32
  33         switch (orig) {
  34         case PCI_ERS_RESULT_CAN_RECOVER:
  35         case PCI_ERS_RESULT_RECOVERED:
  36                 orig = new;
  37                 break;
  38         case PCI_ERS_RESULT_DISCONNECT:
  39                 if (new == PCI_ERS_RESULT_NEED_RESET)
  40                         orig = PCI_ERS_RESULT_NEED_RESET;
  41                 break;
  42         default:
  43                 break;
  44         }
  45
  46         return orig;
  47 }
  48
  49 static int report_error_detected(struct pci_dev *dev,
  50                                  pci_channel_state_t state,
  51                                  enum pci_ers_result *result)
  52 {
  53         struct pci_driver *pdrv;
  54         pci_ers_result_t vote;
  55         const struct pci_error_handlers *err_handler;
  56
  57         device_lock(&dev->dev);
  58         pdrv = dev->driver;
  59         if (pci_dev_is_disconnected(dev)) {
  60                 vote = PCI_ERS_RESULT_DISCONNECT;
  61         } else if (!pci_dev_set_io_state(dev, state)) {
  62                 pci_info(dev, "can't recover (state transition %u -> %u invalid)\n",
  63                         dev->error_state, state);
  64                 vote = PCI_ERS_RESULT_NONE;
  65         } else if (!pdrv || !pdrv->err_handler ||
  66                    !pdrv->err_handler->error_detected) {
  67                 /*
  68                  * If any device in the subtree does not have an error_detected
  69                  * callback, PCI_ERS_RESULT_NO_AER_DRIVER prevents subsequent
  70                  * error callbacks of "any" device in the subtree, and will
  71                  * exit in the disconnected error state.
  72                  */
  73                 if (dev->hdr_type != PCI_HEADER_TYPE_BRIDGE) {
  74                         vote = PCI_ERS_RESULT_NO_AER_DRIVER;
  75                         pci_info(dev, "can't recover (no error_detected callback)\n");
  76                 } else {
  77                         vote = PCI_ERS_RESULT_NONE;
  78                 }
  79         } else {
  80                 err_handler = pdrv->err_handler;
  81                 vote = err_handler->error_detected(dev, state);
  82         }
  83         pci_uevent_ers(dev, vote);
  84         *result = merge_result(*result, vote);
  85         device_unlock(&dev->dev);
  86         return 0;
  87 }
  88
  89 static int pci_pm_runtime_get_sync(struct pci_dev *pdev, void *data)
  90 {
  91         pm_runtime_get_sync(&pdev->dev);
  92         return 0;
  93 }
  94
  95 static int pci_pm_runtime_put(struct pci_dev *pdev, void *data)
  96 {
  97         pm_runtime_put(&pdev->dev);
  98         return 0;
  99 }
 100
 101 static int report_frozen_detected(struct pci_dev *dev, void *data)
 102 {
 103         return report_error_detected(dev, pci_channel_io_frozen, data);
 104 }
 105
 106 static int report_normal_detected(struct pci_dev *dev, void *data)
 107 {
 108         return report_error_detected(dev, pci_channel_io_normal, data);
 109 }
 110
 111 static int report_mmio_enabled(struct pci_dev *dev, void *data)
 112 {
 113         struct pci_driver *pdrv;
 114         pci_ers_result_t vote, *result = data;
 115         const struct pci_error_handlers *err_handler;
 116
 117         device_lock(&dev->dev);
 118         pdrv = dev->driver;
 119         if (!pdrv || !pdrv->err_handler || !pdrv->err_handler->mmio_enabled)
 120                 goto out;
 121
 122         err_handler = pdrv->err_handler;
 123         vote = err_handler->mmio_enabled(dev);
 124         *result = merge_result(*result, vote);
 125 out:
 126         device_unlock(&dev->dev);
 127         return 0;
 128 }
 129
 130 static int report_slot_reset(struct pci_dev *dev, void *data)
 131 {
 132         struct pci_driver *pdrv;
 133         pci_ers_result_t vote, *result = data;
 134         const struct pci_error_handlers *err_handler;
 135
 136         device_lock(&dev->dev);
 137         pdrv = dev->driver;
 138         if (!pdrv || !pdrv->err_handler || !pdrv->err_handler->slot_reset)
 139                 goto out;
 140
 141         err_handler = pdrv->err_handler;
 142         vote = err_handler->slot_reset(dev);
 143         *result = merge_result(*result, vote);
 144 out:
 145         device_unlock(&dev->dev);
 146         return 0;
 147 }
 148
 149 static int report_resume(struct pci_dev *dev, void *data)
 150 {
 151         struct pci_driver *pdrv;
 152         const struct pci_error_handlers *err_handler;
 153
 154         device_lock(&dev->dev);
 155         pdrv = dev->driver;
 156         if (!pci_dev_set_io_state(dev, pci_channel_io_normal) ||
 157             !pdrv || !pdrv->err_handler || !pdrv->err_handler->resume)
 158                 goto out;
 159
 160         err_handler = pdrv->err_handler;
 161         err_handler->resume(dev);
 162 out:
 163         pci_uevent_ers(dev, PCI_ERS_RESULT_RECOVERED);
 164         device_unlock(&dev->dev);
 165         return 0;
 166 }
 167
 168 /**
 169  * pci_walk_bridge - walk bridges potentially AER affected
 170  * @bridge:     bridge which may be a Port, an RCEC, or an RCiEP
 171  * @cb:         callback to be called for each device found
 172  * @userdata:   arbitrary pointer to be passed to callback
 173  *
 174  * If the device provided is a bridge, walk the subordinate bus, including
 175  * any bridged devices on buses under this bus.  Call the provided callback
 176  * on each device found.
 177  *
 178  * If the device provided has no subordinate bus, e.g., an RCEC or RCiEP,
 179  * call the callback on the device itself.
 180  */
 181 static void pci_walk_bridge(struct pci_dev *bridge,
 182                             int (*cb)(struct pci_dev *, void *),
 183                             void *userdata)
 184 {
 185         if (bridge->subordinate)
 186                 pci_walk_bus(bridge->subordinate, cb, userdata);
 187         else
 188                 cb(bridge, userdata);
 189 }
 190
 191 pci_ers_result_t pcie_do_recovery(struct pci_dev *dev,
 192                 pci_channel_state_t state,
 193                 pci_ers_result_t (*reset_subordinates)(struct pci_dev *pdev))
 194 {
 195         int type = pci_pcie_type(dev);
 196         struct pci_dev *bridge;
 197         pci_ers_result_t status = PCI_ERS_RESULT_CAN_RECOVER;
 198         struct pci_host_bridge *host = pci_find_host_bridge(dev->bus);
 199
 200         /*
 201          * If the error was detected by a Root Port, Downstream Port, RCEC,
 202          * or RCiEP, recovery runs on the device itself.  For Ports, that
 203          * also includes any subordinate devices.
 204          *
 205          * If it was detected by another device (Endpoint, etc), recovery
 206          * runs on the device and anything else under the same Port, i.e.,
 207          * everything under "bridge".
 208          */
 209         if (type == PCI_EXP_TYPE_ROOT_PORT ||
 210             type == PCI_EXP_TYPE_DOWNSTREAM ||
 211             type == PCI_EXP_TYPE_RC_EC ||
 212             type == PCI_EXP_TYPE_RC_END)
 213                 bridge = dev;
 214         else
 215                 bridge = pci_upstream_bridge(dev);
 216
 217         pci_walk_bridge(bridge, pci_pm_runtime_get_sync, NULL);
 218
 219         pci_dbg(bridge, "broadcast error_detected message\n");
 220         if (state == pci_channel_io_frozen) {
 221                 pci_walk_bridge(bridge, report_frozen_detected, &status);
 222                 if (reset_subordinates(bridge) != PCI_ERS_RESULT_RECOVERED) {
 223                         pci_warn(bridge, "subordinate device reset failed\n");
 224                         goto failed;
 225                 }
 226         } else {
 227                 pci_walk_bridge(bridge, report_normal_detected, &status);
 228         }
 229
 230         if (status == PCI_ERS_RESULT_CAN_RECOVER) {
 231                 status = PCI_ERS_RESULT_RECOVERED;
 232                 pci_dbg(bridge, "broadcast mmio_enabled message\n");
 233                 pci_walk_bridge(bridge, report_mmio_enabled, &status);
 234         }
 235
 236         if (status == PCI_ERS_RESULT_NEED_RESET) {
 237                 /*
 238                  * TODO: Should call platform-specific
 239                  * functions to reset slot before calling
 240                  * drivers' slot_reset callbacks?
 241                  */
 242                 status = PCI_ERS_RESULT_RECOVERED;
 243                 pci_dbg(bridge, "broadcast slot_reset message\n");
 244                 pci_walk_bridge(bridge, report_slot_reset, &status);
 245         }
 246
 247         if (status != PCI_ERS_RESULT_RECOVERED)
 248                 goto failed;
 249
 250         pci_dbg(bridge, "broadcast resume message\n");
 251         pci_walk_bridge(bridge, report_resume, &status);
 252
 253         /*
 254          * If we have native control of AER, clear error status in the device
 255          * that detected the error.  If the platform retained control of AER,
 256          * it is responsible for clearing this status.  In that case, the
 257          * signaling device may not even be visible to the OS.
 258          */
 259         if (host->native_aer || pcie_ports_native) {
 260                 pcie_clear_device_status(dev);
 261                 pci_aer_clear_nonfatal_status(dev);
 262         }
 263
 264         pci_walk_bridge(bridge, pci_pm_runtime_put, NULL);
 265
 266         pci_info(bridge, "device recovery successful\n");
 267         return status;
 268
 269 failed:
 270         pci_walk_bridge(bridge, pci_pm_runtime_put, NULL);
 271
 272         pci_uevent_ers(bridge, PCI_ERS_RESULT_DISCONNECT);
 273
 274         /* TODO: Should kernel panic here? */
 275         pci_info(bridge, "device recovery failed\n");
 276
 277         return status;
 278 }