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
adf_error_detected(struct pci_dev * pdev,pci_channel_state_t state)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
adf_reset_sbr(struct adf_accel_dev * accel_dev)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
adf_reset_flr(struct adf_accel_dev * accel_dev)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
adf_dev_restore(struct adf_accel_dev * accel_dev)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
adf_device_reset_worker(struct work_struct * work)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 kfree(reset_data);
99 WARN(1, "QAT: device restart failed. Device is unusable\n");
100 return;
101 }
102 adf_dev_restarted_notify(accel_dev);
103 clear_bit(ADF_STATUS_RESTARTING, &accel_dev->status);
104
105 /* The dev is back alive. Notify the caller if in sync mode */
106 if (reset_data->mode == ADF_DEV_RESET_SYNC)
107 complete(&reset_data->compl);
108 else
109 kfree(reset_data);
110 }
111
adf_dev_aer_schedule_reset(struct adf_accel_dev * accel_dev,enum adf_dev_reset_mode mode)112 static int adf_dev_aer_schedule_reset(struct adf_accel_dev *accel_dev,
113 enum adf_dev_reset_mode mode)
114 {
115 struct adf_reset_dev_data *reset_data;
116
117 if (!adf_dev_started(accel_dev) ||
118 test_bit(ADF_STATUS_RESTARTING, &accel_dev->status))
119 return 0;
120
121 set_bit(ADF_STATUS_RESTARTING, &accel_dev->status);
122 reset_data = kzalloc(sizeof(*reset_data), GFP_KERNEL);
123 if (!reset_data)
124 return -ENOMEM;
125 reset_data->accel_dev = accel_dev;
126 init_completion(&reset_data->compl);
127 reset_data->mode = mode;
128 INIT_WORK(&reset_data->reset_work, adf_device_reset_worker);
129 queue_work(device_reset_wq, &reset_data->reset_work);
130
131 /* If in sync mode wait for the result */
132 if (mode == ADF_DEV_RESET_SYNC) {
133 int ret = 0;
134 /* Maximum device reset time is 10 seconds */
135 unsigned long wait_jiffies = msecs_to_jiffies(10000);
136 unsigned long timeout = wait_for_completion_timeout(
137 &reset_data->compl, wait_jiffies);
138 if (!timeout) {
139 dev_err(&GET_DEV(accel_dev),
140 "Reset device timeout expired\n");
141 ret = -EFAULT;
142 }
143 kfree(reset_data);
144 return ret;
145 }
146 return 0;
147 }
148
adf_slot_reset(struct pci_dev * pdev)149 static pci_ers_result_t adf_slot_reset(struct pci_dev *pdev)
150 {
151 struct adf_accel_dev *accel_dev = adf_devmgr_pci_to_accel_dev(pdev);
152
153 if (!accel_dev) {
154 pr_err("QAT: Can't find acceleration device\n");
155 return PCI_ERS_RESULT_DISCONNECT;
156 }
157 if (adf_dev_aer_schedule_reset(accel_dev, ADF_DEV_RESET_SYNC))
158 return PCI_ERS_RESULT_DISCONNECT;
159
160 return PCI_ERS_RESULT_RECOVERED;
161 }
162
adf_resume(struct pci_dev * pdev)163 static void adf_resume(struct pci_dev *pdev)
164 {
165 dev_info(&pdev->dev, "Acceleration driver reset completed\n");
166 dev_info(&pdev->dev, "Device is up and running\n");
167 }
168
169 static const struct pci_error_handlers adf_err_handler = {
170 .error_detected = adf_error_detected,
171 .slot_reset = adf_slot_reset,
172 .resume = adf_resume,
173 };
174
175 /**
176 * adf_enable_aer() - Enable Advance Error Reporting for acceleration device
177 * @accel_dev: Pointer to acceleration device.
178 *
179 * Function enables PCI Advance Error Reporting for the
180 * QAT acceleration device accel_dev.
181 * To be used by QAT device specific drivers.
182 *
183 * Return: 0 on success, error code otherwise.
184 */
adf_enable_aer(struct adf_accel_dev * accel_dev)185 int adf_enable_aer(struct adf_accel_dev *accel_dev)
186 {
187 struct pci_dev *pdev = accel_to_pci_dev(accel_dev);
188 struct pci_driver *pdrv = pdev->driver;
189
190 pdrv->err_handler = &adf_err_handler;
191 pci_enable_pcie_error_reporting(pdev);
192 return 0;
193 }
194 EXPORT_SYMBOL_GPL(adf_enable_aer);
195
196 /**
197 * adf_disable_aer() - Enable Advance Error Reporting for acceleration device
198 * @accel_dev: Pointer to acceleration device.
199 *
200 * Function disables PCI Advance Error Reporting for the
201 * QAT acceleration device accel_dev.
202 * To be used by QAT device specific drivers.
203 *
204 * Return: void
205 */
adf_disable_aer(struct adf_accel_dev * accel_dev)206 void adf_disable_aer(struct adf_accel_dev *accel_dev)
207 {
208 struct pci_dev *pdev = accel_to_pci_dev(accel_dev);
209
210 pci_disable_pcie_error_reporting(pdev);
211 }
212 EXPORT_SYMBOL_GPL(adf_disable_aer);
213
adf_init_aer(void)214 int adf_init_aer(void)
215 {
216 device_reset_wq = alloc_workqueue("qat_device_reset_wq",
217 WQ_MEM_RECLAIM, 0);
218 return !device_reset_wq ? -EFAULT : 0;
219 }
220
adf_exit_aer(void)221 void adf_exit_aer(void)
222 {
223 if (device_reset_wq)
224 destroy_workqueue(device_reset_wq);
225 device_reset_wq = NULL;
226 }
227