1 // SPDX-License-Identifier: (BSD-3-Clause OR GPL-2.0-only)
2 /* Copyright(c) 2014 - 2020 Intel Corporation */
3 #include <linux/kernel.h>
5 #include <linux/completion.h>
6 #include <linux/workqueue.h>
7 #include <linux/delay.h>
8 #include "adf_accel_devices.h"
9 #include "adf_common_drv.h"
11 static struct workqueue_struct *device_reset_wq;
13 static pci_ers_result_t adf_error_detected(struct pci_dev *pdev,
14 pci_channel_state_t state)
16 struct adf_accel_dev *accel_dev = adf_devmgr_pci_to_accel_dev(pdev);
18 dev_info(&pdev->dev, "Acceleration driver hardware error detected.\n");
20 dev_err(&pdev->dev, "Can't find acceleration device\n");
21 return PCI_ERS_RESULT_DISCONNECT;
24 if (state == pci_channel_io_perm_failure) {
25 dev_err(&pdev->dev, "Can't recover from device error\n");
26 return PCI_ERS_RESULT_DISCONNECT;
29 return PCI_ERS_RESULT_NEED_RESET;
33 struct adf_reset_dev_data {
35 struct adf_accel_dev *accel_dev;
36 struct completion compl;
37 struct work_struct reset_work;
40 void adf_reset_sbr(struct adf_accel_dev *accel_dev)
42 struct pci_dev *pdev = accel_to_pci_dev(accel_dev);
43 struct pci_dev *parent = pdev->bus->self;
49 if (!pci_wait_for_pending_transaction(pdev))
50 dev_info(&GET_DEV(accel_dev),
51 "Transaction still in progress. Proceeding\n");
53 dev_info(&GET_DEV(accel_dev), "Secondary bus reset\n");
55 pci_read_config_word(parent, PCI_BRIDGE_CONTROL, &bridge_ctl);
56 bridge_ctl |= PCI_BRIDGE_CTL_BUS_RESET;
57 pci_write_config_word(parent, PCI_BRIDGE_CONTROL, bridge_ctl);
59 bridge_ctl &= ~PCI_BRIDGE_CTL_BUS_RESET;
60 pci_write_config_word(parent, PCI_BRIDGE_CONTROL, bridge_ctl);
63 EXPORT_SYMBOL_GPL(adf_reset_sbr);
65 void adf_reset_flr(struct adf_accel_dev *accel_dev)
67 pcie_flr(accel_to_pci_dev(accel_dev));
69 EXPORT_SYMBOL_GPL(adf_reset_flr);
71 void adf_dev_restore(struct adf_accel_dev *accel_dev)
73 struct adf_hw_device_data *hw_device = accel_dev->hw_device;
74 struct pci_dev *pdev = accel_to_pci_dev(accel_dev);
76 if (hw_device->reset_device) {
77 dev_info(&GET_DEV(accel_dev), "Resetting device qat_dev%d\n",
79 hw_device->reset_device(accel_dev);
80 pci_restore_state(pdev);
85 static void adf_device_reset_worker(struct work_struct *work)
87 struct adf_reset_dev_data *reset_data =
88 container_of(work, struct adf_reset_dev_data, reset_work);
89 struct adf_accel_dev *accel_dev = reset_data->accel_dev;
91 adf_dev_restarting_notify(accel_dev);
92 if (adf_dev_restart(accel_dev)) {
93 /* The device hanged and we can't restart it so stop here */
94 dev_err(&GET_DEV(accel_dev), "Restart device failed\n");
96 WARN(1, "QAT: device restart failed. Device is unusable\n");
99 adf_dev_restarted_notify(accel_dev);
100 clear_bit(ADF_STATUS_RESTARTING, &accel_dev->status);
102 /* The dev is back alive. Notify the caller if in sync mode */
103 if (reset_data->mode == ADF_DEV_RESET_SYNC)
104 complete(&reset_data->compl);
109 static int adf_dev_aer_schedule_reset(struct adf_accel_dev *accel_dev,
110 enum adf_dev_reset_mode mode)
112 struct adf_reset_dev_data *reset_data;
114 if (!adf_dev_started(accel_dev) ||
115 test_bit(ADF_STATUS_RESTARTING, &accel_dev->status))
118 set_bit(ADF_STATUS_RESTARTING, &accel_dev->status);
119 reset_data = kzalloc(sizeof(*reset_data), GFP_KERNEL);
122 reset_data->accel_dev = accel_dev;
123 init_completion(&reset_data->compl);
124 reset_data->mode = mode;
125 INIT_WORK(&reset_data->reset_work, adf_device_reset_worker);
126 queue_work(device_reset_wq, &reset_data->reset_work);
128 /* If in sync mode wait for the result */
129 if (mode == ADF_DEV_RESET_SYNC) {
131 /* Maximum device reset time is 10 seconds */
132 unsigned long wait_jiffies = msecs_to_jiffies(10000);
133 unsigned long timeout = wait_for_completion_timeout(
134 &reset_data->compl, wait_jiffies);
136 dev_err(&GET_DEV(accel_dev),
137 "Reset device timeout expired\n");
146 static pci_ers_result_t adf_slot_reset(struct pci_dev *pdev)
148 struct adf_accel_dev *accel_dev = adf_devmgr_pci_to_accel_dev(pdev);
151 pr_err("QAT: Can't find acceleration device\n");
152 return PCI_ERS_RESULT_DISCONNECT;
154 if (adf_dev_aer_schedule_reset(accel_dev, ADF_DEV_RESET_SYNC))
155 return PCI_ERS_RESULT_DISCONNECT;
157 return PCI_ERS_RESULT_RECOVERED;
160 static void adf_resume(struct pci_dev *pdev)
162 dev_info(&pdev->dev, "Acceleration driver reset completed\n");
163 dev_info(&pdev->dev, "Device is up and running\n");
166 const struct pci_error_handlers adf_err_handler = {
167 .error_detected = adf_error_detected,
168 .slot_reset = adf_slot_reset,
169 .resume = adf_resume,
171 EXPORT_SYMBOL_GPL(adf_err_handler);
173 int adf_init_aer(void)
175 device_reset_wq = alloc_workqueue("qat_device_reset_wq",
177 return !device_reset_wq ? -EFAULT : 0;
180 void adf_exit_aer(void)
183 destroy_workqueue(device_reset_wq);
184 device_reset_wq = NULL;