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/completion.h>
6 #include <linux/workqueue.h>
7 #include <linux/delay.h>
8 #include "adf_accel_devices.h"
9 #include "adf_common_drv.h"
10 #include "adf_pfvf_pf_msg.h"
11
12 struct adf_fatal_error_data {
13 struct adf_accel_dev *accel_dev;
14 struct work_struct work;
15 };
16
17 static struct workqueue_struct *device_reset_wq;
18 static struct workqueue_struct *device_sriov_wq;
19
adf_error_detected(struct pci_dev * pdev,pci_channel_state_t state)20 static pci_ers_result_t adf_error_detected(struct pci_dev *pdev,
21 pci_channel_state_t state)
22 {
23 struct adf_accel_dev *accel_dev = adf_devmgr_pci_to_accel_dev(pdev);
24
25 dev_info(&pdev->dev, "Acceleration driver hardware error detected.\n");
26 if (!accel_dev) {
27 dev_err(&pdev->dev, "Can't find acceleration device\n");
28 return PCI_ERS_RESULT_DISCONNECT;
29 }
30
31 if (state == pci_channel_io_perm_failure) {
32 dev_err(&pdev->dev, "Can't recover from device error\n");
33 return PCI_ERS_RESULT_DISCONNECT;
34 }
35
36 set_bit(ADF_STATUS_RESTARTING, &accel_dev->status);
37 if (accel_dev->hw_device->exit_arb) {
38 dev_dbg(&pdev->dev, "Disabling arbitration\n");
39 accel_dev->hw_device->exit_arb(accel_dev);
40 }
41 adf_error_notifier(accel_dev);
42 adf_pf2vf_notify_fatal_error(accel_dev);
43 adf_dev_restarting_notify(accel_dev);
44 adf_pf2vf_notify_restarting(accel_dev);
45 adf_pf2vf_wait_for_restarting_complete(accel_dev);
46 pci_clear_master(pdev);
47 adf_dev_down(accel_dev);
48
49 return PCI_ERS_RESULT_NEED_RESET;
50 }
51
52 /* reset dev data */
53 struct adf_reset_dev_data {
54 int mode;
55 struct adf_accel_dev *accel_dev;
56 struct completion compl;
57 struct work_struct reset_work;
58 };
59
60 /* sriov dev data */
61 struct adf_sriov_dev_data {
62 struct adf_accel_dev *accel_dev;
63 struct completion compl;
64 struct work_struct sriov_work;
65 };
66
adf_reset_sbr(struct adf_accel_dev * accel_dev)67 void adf_reset_sbr(struct adf_accel_dev *accel_dev)
68 {
69 struct pci_dev *pdev = accel_to_pci_dev(accel_dev);
70 struct pci_dev *parent = pdev->bus->self;
71 u16 bridge_ctl = 0;
72
73 if (!parent)
74 parent = pdev;
75
76 if (!pci_wait_for_pending_transaction(pdev))
77 dev_info(&GET_DEV(accel_dev),
78 "Transaction still in progress. Proceeding\n");
79
80 dev_info(&GET_DEV(accel_dev), "Secondary bus reset\n");
81
82 pci_read_config_word(parent, PCI_BRIDGE_CONTROL, &bridge_ctl);
83 bridge_ctl |= PCI_BRIDGE_CTL_BUS_RESET;
84 pci_write_config_word(parent, PCI_BRIDGE_CONTROL, bridge_ctl);
85 msleep(100);
86 bridge_ctl &= ~PCI_BRIDGE_CTL_BUS_RESET;
87 pci_write_config_word(parent, PCI_BRIDGE_CONTROL, bridge_ctl);
88 msleep(100);
89 }
90 EXPORT_SYMBOL_GPL(adf_reset_sbr);
91
adf_reset_flr(struct adf_accel_dev * accel_dev)92 void adf_reset_flr(struct adf_accel_dev *accel_dev)
93 {
94 pcie_flr(accel_to_pci_dev(accel_dev));
95 }
96 EXPORT_SYMBOL_GPL(adf_reset_flr);
97
adf_dev_restore(struct adf_accel_dev * accel_dev)98 void adf_dev_restore(struct adf_accel_dev *accel_dev)
99 {
100 struct adf_hw_device_data *hw_device = accel_dev->hw_device;
101 struct pci_dev *pdev = accel_to_pci_dev(accel_dev);
102
103 if (hw_device->reset_device) {
104 dev_info(&GET_DEV(accel_dev), "Resetting device qat_dev%d\n",
105 accel_dev->accel_id);
106 hw_device->reset_device(accel_dev);
107 pci_restore_state(pdev);
108 pci_save_state(pdev);
109 }
110 }
111
adf_device_sriov_worker(struct work_struct * work)112 static void adf_device_sriov_worker(struct work_struct *work)
113 {
114 struct adf_sriov_dev_data *sriov_data =
115 container_of(work, struct adf_sriov_dev_data, sriov_work);
116
117 adf_reenable_sriov(sriov_data->accel_dev);
118 complete(&sriov_data->compl);
119 }
120
adf_device_reset_worker(struct work_struct * work)121 static void adf_device_reset_worker(struct work_struct *work)
122 {
123 struct adf_reset_dev_data *reset_data =
124 container_of(work, struct adf_reset_dev_data, reset_work);
125 struct adf_accel_dev *accel_dev = reset_data->accel_dev;
126 unsigned long wait_jiffies = msecs_to_jiffies(10000);
127 struct adf_sriov_dev_data sriov_data;
128
129 adf_dev_restarting_notify(accel_dev);
130 if (adf_dev_restart(accel_dev)) {
131 /* The device hanged and we can't restart it so stop here */
132 dev_err(&GET_DEV(accel_dev), "Restart device failed\n");
133 if (reset_data->mode == ADF_DEV_RESET_ASYNC)
134 kfree(reset_data);
135 WARN(1, "QAT: device restart failed. Device is unusable\n");
136 return;
137 }
138
139 sriov_data.accel_dev = accel_dev;
140 init_completion(&sriov_data.compl);
141 INIT_WORK(&sriov_data.sriov_work, adf_device_sriov_worker);
142 queue_work(device_sriov_wq, &sriov_data.sriov_work);
143 if (wait_for_completion_timeout(&sriov_data.compl, wait_jiffies))
144 adf_pf2vf_notify_restarted(accel_dev);
145
146 adf_dev_restarted_notify(accel_dev);
147 clear_bit(ADF_STATUS_RESTARTING, &accel_dev->status);
148
149 /* The dev is back alive. Notify the caller if in sync mode */
150 if (reset_data->mode == ADF_DEV_RESET_ASYNC)
151 kfree(reset_data);
152 else
153 complete(&reset_data->compl);
154 }
155
adf_dev_aer_schedule_reset(struct adf_accel_dev * accel_dev,enum adf_dev_reset_mode mode)156 static int adf_dev_aer_schedule_reset(struct adf_accel_dev *accel_dev,
157 enum adf_dev_reset_mode mode)
158 {
159 struct adf_reset_dev_data *reset_data;
160
161 if (!adf_dev_started(accel_dev) ||
162 test_bit(ADF_STATUS_RESTARTING, &accel_dev->status))
163 return 0;
164
165 set_bit(ADF_STATUS_RESTARTING, &accel_dev->status);
166 reset_data = kzalloc(sizeof(*reset_data), GFP_KERNEL);
167 if (!reset_data)
168 return -ENOMEM;
169 reset_data->accel_dev = accel_dev;
170 init_completion(&reset_data->compl);
171 reset_data->mode = mode;
172 INIT_WORK(&reset_data->reset_work, adf_device_reset_worker);
173 queue_work(device_reset_wq, &reset_data->reset_work);
174
175 /* If in sync mode wait for the result */
176 if (mode == ADF_DEV_RESET_SYNC) {
177 int ret = 0;
178 /* Maximum device reset time is 10 seconds */
179 unsigned long wait_jiffies = msecs_to_jiffies(10000);
180 unsigned long timeout = wait_for_completion_timeout(
181 &reset_data->compl, wait_jiffies);
182 if (!timeout) {
183 dev_err(&GET_DEV(accel_dev),
184 "Reset device timeout expired\n");
185 cancel_work_sync(&reset_data->reset_work);
186 ret = -EFAULT;
187 }
188 kfree(reset_data);
189 return ret;
190 }
191 return 0;
192 }
193
adf_slot_reset(struct pci_dev * pdev)194 static pci_ers_result_t adf_slot_reset(struct pci_dev *pdev)
195 {
196 struct adf_accel_dev *accel_dev = adf_devmgr_pci_to_accel_dev(pdev);
197 int res = 0;
198
199 if (!accel_dev) {
200 pr_err("QAT: Can't find acceleration device\n");
201 return PCI_ERS_RESULT_DISCONNECT;
202 }
203
204 if (!pdev->is_busmaster)
205 pci_set_master(pdev);
206 pci_restore_state(pdev);
207 pci_save_state(pdev);
208 res = adf_dev_up(accel_dev, false);
209 if (res && res != -EALREADY)
210 return PCI_ERS_RESULT_DISCONNECT;
211
212 adf_reenable_sriov(accel_dev);
213 adf_pf2vf_notify_restarted(accel_dev);
214 adf_dev_restarted_notify(accel_dev);
215 clear_bit(ADF_STATUS_RESTARTING, &accel_dev->status);
216 return PCI_ERS_RESULT_RECOVERED;
217 }
218
adf_resume(struct pci_dev * pdev)219 static void adf_resume(struct pci_dev *pdev)
220 {
221 dev_info(&pdev->dev, "Acceleration driver reset completed\n");
222 dev_info(&pdev->dev, "Device is up and running\n");
223 }
224
225 const struct pci_error_handlers adf_err_handler = {
226 .error_detected = adf_error_detected,
227 .slot_reset = adf_slot_reset,
228 .resume = adf_resume,
229 };
230 EXPORT_SYMBOL_GPL(adf_err_handler);
231
adf_dev_autoreset(struct adf_accel_dev * accel_dev)232 int adf_dev_autoreset(struct adf_accel_dev *accel_dev)
233 {
234 if (accel_dev->autoreset_on_error)
235 return adf_dev_aer_schedule_reset(accel_dev, ADF_DEV_RESET_ASYNC);
236
237 return 0;
238 }
239
adf_notify_fatal_error_worker(struct work_struct * work)240 static void adf_notify_fatal_error_worker(struct work_struct *work)
241 {
242 struct adf_fatal_error_data *wq_data =
243 container_of(work, struct adf_fatal_error_data, work);
244 struct adf_accel_dev *accel_dev = wq_data->accel_dev;
245 struct adf_hw_device_data *hw_device = accel_dev->hw_device;
246
247 adf_error_notifier(accel_dev);
248
249 if (!accel_dev->is_vf) {
250 /* Disable arbitration to stop processing of new requests */
251 if (accel_dev->autoreset_on_error && hw_device->exit_arb)
252 hw_device->exit_arb(accel_dev);
253 if (accel_dev->pf.vf_info)
254 adf_pf2vf_notify_fatal_error(accel_dev);
255 adf_dev_autoreset(accel_dev);
256 }
257
258 kfree(wq_data);
259 }
260
adf_notify_fatal_error(struct adf_accel_dev * accel_dev)261 int adf_notify_fatal_error(struct adf_accel_dev *accel_dev)
262 {
263 struct adf_fatal_error_data *wq_data;
264
265 wq_data = kzalloc(sizeof(*wq_data), GFP_ATOMIC);
266 if (!wq_data)
267 return -ENOMEM;
268
269 wq_data->accel_dev = accel_dev;
270 INIT_WORK(&wq_data->work, adf_notify_fatal_error_worker);
271 adf_misc_wq_queue_work(&wq_data->work);
272
273 return 0;
274 }
275
adf_init_aer(void)276 int adf_init_aer(void)
277 {
278 device_reset_wq = alloc_workqueue("qat_device_reset_wq",
279 WQ_MEM_RECLAIM, 0);
280 if (!device_reset_wq)
281 return -EFAULT;
282
283 device_sriov_wq = alloc_workqueue("qat_device_sriov_wq", 0, 0);
284 if (!device_sriov_wq) {
285 destroy_workqueue(device_reset_wq);
286 device_reset_wq = NULL;
287 return -EFAULT;
288 }
289
290 return 0;
291 }
292
adf_exit_aer(void)293 void adf_exit_aer(void)
294 {
295 if (device_reset_wq)
296 destroy_workqueue(device_reset_wq);
297 device_reset_wq = NULL;
298
299 if (device_sriov_wq)
300 destroy_workqueue(device_sriov_wq);
301 device_sriov_wq = NULL;
302 }
303