drivers/crypto/qat/qat_common/adf_aer.c

   1 // SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0-only)
   2 /* Copyright(c) 2014 - 2020 Intel Corporation */
   3 #include <linux/kernel.h>
   4 #include <linux/pci.h>
   5 #include <linux/aer.h>
   6 #include <linux/completion.h>
   7 #include <linux/workqueue.h>
   8 #include <linux/delay.h>
   9 #include "adf_accel_devices.h"
  10 #include "adf_common_drv.h"
  11
  12 static struct workqueue_struct *device_reset_wq;
  13
  14 static pci_ers_result_t adf_error_detected(struct pci_dev *pdev,
  15                                            pci_channel_state_t state)
  16 {
  17         struct adf_accel_dev *accel_dev = adf_devmgr_pci_to_accel_dev(pdev);
  18
  19         dev_info(&pdev->dev, "Acceleration driver hardware error detected.\n");
  20         if (!accel_dev) {
  21                 dev_err(&pdev->dev, "Can't find acceleration device\n");
  22                 return PCI_ERS_RESULT_DISCONNECT;
  23         }
  24
  25         if (state == pci_channel_io_perm_failure) {
  26                 dev_err(&pdev->dev, "Can't recover from device error\n");
  27                 return PCI_ERS_RESULT_DISCONNECT;
  28         }
  29
  30         return PCI_ERS_RESULT_NEED_RESET;
  31 }
  32
  33 /* reset dev data */
  34 struct adf_reset_dev_data {
  35         int mode;
  36         struct adf_accel_dev *accel_dev;
  37         struct completion compl;
  38         struct work_struct reset_work;
  39 };
  40
  41 void adf_reset_sbr(struct adf_accel_dev *accel_dev)
  42 {
  43         struct pci_dev *pdev = accel_to_pci_dev(accel_dev);
  44         struct pci_dev *parent = pdev->bus->self;
  45         u16 bridge_ctl = 0;
  46
  47         if (!parent)
  48                 parent = pdev;
  49
  50         if (!pci_wait_for_pending_transaction(pdev))
  51                 dev_info(&GET_DEV(accel_dev),
  52                          "Transaction still in progress. Proceeding\n");
  53
  54         dev_info(&GET_DEV(accel_dev), "Secondary bus reset\n");
  55
  56         pci_read_config_word(parent, PCI_BRIDGE_CONTROL, &bridge_ctl);
  57         bridge_ctl |= PCI_BRIDGE_CTL_BUS_RESET;
  58         pci_write_config_word(parent, PCI_BRIDGE_CONTROL, bridge_ctl);
  59         msleep(100);
  60         bridge_ctl &= ~PCI_BRIDGE_CTL_BUS_RESET;
  61         pci_write_config_word(parent, PCI_BRIDGE_CONTROL, bridge_ctl);
  62         msleep(100);
  63 }
  64 EXPORT_SYMBOL_GPL(adf_reset_sbr);
  65
  66 void adf_reset_flr(struct adf_accel_dev *accel_dev)
  67 {
  68         pcie_flr(accel_to_pci_dev(accel_dev));
  69 }
  70 EXPORT_SYMBOL_GPL(adf_reset_flr);
  71
  72 void adf_dev_restore(struct adf_accel_dev *accel_dev)
  73 {
  74         struct adf_hw_device_data *hw_device = accel_dev->hw_device;
  75         struct pci_dev *pdev = accel_to_pci_dev(accel_dev);
  76
  77         if (hw_device->reset_device) {
  78                 dev_info(&GET_DEV(accel_dev), "Resetting device qat_dev%d\n",
  79                          accel_dev->accel_id);
  80                 hw_device->reset_device(accel_dev);
  81                 pci_restore_state(pdev);
  82                 pci_save_state(pdev);
  83         }
  84 }
  85
  86 static void adf_device_reset_worker(struct work_struct *work)
  87 {
  88         struct adf_reset_dev_data *reset_data =
  89                   container_of(work, struct adf_reset_dev_data, reset_work);
  90         struct adf_accel_dev *accel_dev = reset_data->accel_dev;
  91
  92         adf_dev_restarting_notify(accel_dev);
  93         adf_dev_stop(accel_dev);
  94         adf_dev_shutdown(accel_dev);
  95         if (adf_dev_init(accel_dev) || adf_dev_start(accel_dev)) {
  96                 /* The device hanged and we can't restart it so stop here */
  97                 dev_err(&GET_DEV(accel_dev), "Restart device failed\n");
  98                 if (reset_data->mode == ADF_DEV_RESET_ASYNC ||
  99                     completion_done(&reset_data->compl))
 100                         kfree(reset_data);
 101                 WARN(1, "QAT: device restart failed. Device is unusable\n");
 102                 return;
 103         }
 104         adf_dev_restarted_notify(accel_dev);
 105         clear_bit(ADF_STATUS_RESTARTING, &accel_dev->status);
 106
 107         /*
 108          * The dev is back alive. Notify the caller if in sync mode
 109          *
 110          * If device restart will take a more time than expected,
 111          * the schedule_reset() function can timeout and exit. This can be
 112          * detected by calling the completion_done() function. In this case
 113          * the reset_data structure needs to be freed here.
 114          */
 115         if (reset_data->mode == ADF_DEV_RESET_ASYNC ||
 116             completion_done(&reset_data->compl))
 117                 kfree(reset_data);
 118         else
 119                 complete(&reset_data->compl);
 120 }
 121
 122 static int adf_dev_aer_schedule_reset(struct adf_accel_dev *accel_dev,
 123                                       enum adf_dev_reset_mode mode)
 124 {
 125         struct adf_reset_dev_data *reset_data;
 126
 127         if (!adf_dev_started(accel_dev) ||
 128             test_bit(ADF_STATUS_RESTARTING, &accel_dev->status))
 129                 return 0;
 130
 131         set_bit(ADF_STATUS_RESTARTING, &accel_dev->status);
 132         reset_data = kzalloc(sizeof(*reset_data), GFP_KERNEL);
 133         if (!reset_data)
 134                 return -ENOMEM;
 135         reset_data->accel_dev = accel_dev;
 136         init_completion(&reset_data->compl);
 137         reset_data->mode = mode;
 138         INIT_WORK(&reset_data->reset_work, adf_device_reset_worker);
 139         queue_work(device_reset_wq, &reset_data->reset_work);
 140
 141         /* If in sync mode wait for the result */
 142         if (mode == ADF_DEV_RESET_SYNC) {
 143                 int ret = 0;
 144                 /* Maximum device reset time is 10 seconds */
 145                 unsigned long wait_jiffies = msecs_to_jiffies(10000);
 146                 unsigned long timeout = wait_for_completion_timeout(
 147                                    &reset_data->compl, wait_jiffies);
 148                 if (!timeout) {
 149                         dev_err(&GET_DEV(accel_dev),
 150                                 "Reset device timeout expired\n");
 151                         ret = -EFAULT;
 152                 } else {
 153                         kfree(reset_data);
 154                 }
 155                 return ret;
 156         }
 157         return 0;
 158 }
 159
 160 static pci_ers_result_t adf_slot_reset(struct pci_dev *pdev)
 161 {
 162         struct adf_accel_dev *accel_dev = adf_devmgr_pci_to_accel_dev(pdev);
 163
 164         if (!accel_dev) {
 165                 pr_err("QAT: Can't find acceleration device\n");
 166                 return PCI_ERS_RESULT_DISCONNECT;
 167         }
 168         if (adf_dev_aer_schedule_reset(accel_dev, ADF_DEV_RESET_SYNC))
 169                 return PCI_ERS_RESULT_DISCONNECT;
 170
 171         return PCI_ERS_RESULT_RECOVERED;
 172 }
 173
 174 static void adf_resume(struct pci_dev *pdev)
 175 {
 176         dev_info(&pdev->dev, "Acceleration driver reset completed\n");
 177         dev_info(&pdev->dev, "Device is up and running\n");
 178 }
 179
 180 static const struct pci_error_handlers adf_err_handler = {
 181         .error_detected = adf_error_detected,
 182         .slot_reset = adf_slot_reset,
 183         .resume = adf_resume,
 184 };
 185
 186 /**
 187  * adf_enable_aer() - Enable Advance Error Reporting for acceleration device
 188  * @accel_dev:  Pointer to acceleration device.
 189  *
 190  * Function enables PCI Advance Error Reporting for the
 191  * QAT acceleration device accel_dev.
 192  * To be used by QAT device specific drivers.
 193  *
 194  * Return: 0 on success, error code otherwise.
 195  */
 196 int adf_enable_aer(struct adf_accel_dev *accel_dev)
 197 {
 198         struct pci_dev *pdev = accel_to_pci_dev(accel_dev);
 199         struct pci_driver *pdrv = pdev->driver;
 200
 201         pdrv->err_handler = &adf_err_handler;
 202         pci_enable_pcie_error_reporting(pdev);
 203         return 0;
 204 }
 205 EXPORT_SYMBOL_GPL(adf_enable_aer);
 206
 207 /**
 208  * adf_disable_aer() - Enable Advance Error Reporting for acceleration device
 209  * @accel_dev:  Pointer to acceleration device.
 210  *
 211  * Function disables PCI Advance Error Reporting for the
 212  * QAT acceleration device accel_dev.
 213  * To be used by QAT device specific drivers.
 214  *
 215  * Return: void
 216  */
 217 void adf_disable_aer(struct adf_accel_dev *accel_dev)
 218 {
 219         struct pci_dev *pdev = accel_to_pci_dev(accel_dev);
 220
 221         pci_disable_pcie_error_reporting(pdev);
 222 }
 223 EXPORT_SYMBOL_GPL(adf_disable_aer);
 224
 225 int adf_init_aer(void)
 226 {
 227         device_reset_wq = alloc_workqueue("qat_device_reset_wq",
 228                                           WQ_MEM_RECLAIM, 0);
 229         return !device_reset_wq ? -EFAULT : 0;
 230 }
 231
 232 void adf_exit_aer(void)
 233 {
 234         if (device_reset_wq)
 235                 destroy_workqueue(device_reset_wq);
 236         device_reset_wq = NULL;
 237 }