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