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 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 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 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 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 } 109 } 110 111 static void adf_device_sriov_worker(struct work_struct *work) 112 { 113 struct adf_sriov_dev_data *sriov_data = 114 container_of(work, struct adf_sriov_dev_data, sriov_work); 115 116 adf_reenable_sriov(sriov_data->accel_dev); 117 complete(&sriov_data->compl); 118 } 119 120 static void adf_device_reset_worker(struct work_struct *work) 121 { 122 struct adf_reset_dev_data *reset_data = 123 container_of(work, struct adf_reset_dev_data, reset_work); 124 struct adf_accel_dev *accel_dev = reset_data->accel_dev; 125 unsigned long wait_jiffies = msecs_to_jiffies(10000); 126 struct adf_sriov_dev_data sriov_data; 127 128 adf_dev_restarting_notify(accel_dev); 129 if (adf_dev_restart(accel_dev)) { 130 /* The device hanged and we can't restart it so stop here */ 131 dev_err(&GET_DEV(accel_dev), "Restart device failed\n"); 132 if (reset_data->mode == ADF_DEV_RESET_ASYNC) 133 kfree(reset_data); 134 WARN(1, "QAT: device restart failed. Device is unusable\n"); 135 return; 136 } 137 138 sriov_data.accel_dev = accel_dev; 139 init_completion(&sriov_data.compl); 140 INIT_WORK(&sriov_data.sriov_work, adf_device_sriov_worker); 141 queue_work(device_sriov_wq, &sriov_data.sriov_work); 142 if (wait_for_completion_timeout(&sriov_data.compl, wait_jiffies)) 143 adf_pf2vf_notify_restarted(accel_dev); 144 145 adf_dev_restarted_notify(accel_dev); 146 clear_bit(ADF_STATUS_RESTARTING, &accel_dev->status); 147 148 /* The dev is back alive. Notify the caller if in sync mode */ 149 if (reset_data->mode == ADF_DEV_RESET_ASYNC) 150 kfree(reset_data); 151 else 152 complete(&reset_data->compl); 153 } 154 155 static int adf_dev_aer_schedule_reset(struct adf_accel_dev *accel_dev, 156 enum adf_dev_reset_mode mode) 157 { 158 struct adf_reset_dev_data *reset_data; 159 160 if (!adf_dev_started(accel_dev) || 161 test_bit(ADF_STATUS_RESTARTING, &accel_dev->status)) 162 return 0; 163 164 set_bit(ADF_STATUS_RESTARTING, &accel_dev->status); 165 reset_data = kzalloc(sizeof(*reset_data), GFP_KERNEL); 166 if (!reset_data) 167 return -ENOMEM; 168 reset_data->accel_dev = accel_dev; 169 init_completion(&reset_data->compl); 170 reset_data->mode = mode; 171 INIT_WORK(&reset_data->reset_work, adf_device_reset_worker); 172 queue_work(device_reset_wq, &reset_data->reset_work); 173 174 /* If in sync mode wait for the result */ 175 if (mode == ADF_DEV_RESET_SYNC) { 176 int ret = 0; 177 /* Maximum device reset time is 10 seconds */ 178 unsigned long wait_jiffies = msecs_to_jiffies(10000); 179 unsigned long timeout = wait_for_completion_timeout( 180 &reset_data->compl, wait_jiffies); 181 if (!timeout) { 182 dev_err(&GET_DEV(accel_dev), 183 "Reset device timeout expired\n"); 184 cancel_work_sync(&reset_data->reset_work); 185 ret = -EFAULT; 186 } 187 kfree(reset_data); 188 return ret; 189 } 190 return 0; 191 } 192 193 static pci_ers_result_t adf_slot_reset(struct pci_dev *pdev) 194 { 195 struct adf_accel_dev *accel_dev = adf_devmgr_pci_to_accel_dev(pdev); 196 int res = 0; 197 198 if (!accel_dev) { 199 pr_err("QAT: Can't find acceleration device\n"); 200 return PCI_ERS_RESULT_DISCONNECT; 201 } 202 203 if (!pdev->is_busmaster) 204 pci_set_master(pdev); 205 pci_restore_state(pdev); 206 res = adf_dev_up(accel_dev, false); 207 if (res && res != -EALREADY) 208 return PCI_ERS_RESULT_DISCONNECT; 209 210 adf_reenable_sriov(accel_dev); 211 adf_pf2vf_notify_restarted(accel_dev); 212 adf_dev_restarted_notify(accel_dev); 213 clear_bit(ADF_STATUS_RESTARTING, &accel_dev->status); 214 return PCI_ERS_RESULT_RECOVERED; 215 } 216 217 static void adf_resume(struct pci_dev *pdev) 218 { 219 dev_info(&pdev->dev, "Acceleration driver reset completed\n"); 220 dev_info(&pdev->dev, "Device is up and running\n"); 221 } 222 223 const struct pci_error_handlers adf_err_handler = { 224 .error_detected = adf_error_detected, 225 .slot_reset = adf_slot_reset, 226 .resume = adf_resume, 227 }; 228 EXPORT_SYMBOL_GPL(adf_err_handler); 229 230 static int adf_dev_autoreset(struct adf_accel_dev *accel_dev) 231 { 232 if (accel_dev->autoreset_on_error) 233 return adf_dev_aer_schedule_reset(accel_dev, ADF_DEV_RESET_ASYNC); 234 235 return 0; 236 } 237 238 static void adf_notify_fatal_error_worker(struct work_struct *work) 239 { 240 struct adf_fatal_error_data *wq_data = 241 container_of(work, struct adf_fatal_error_data, work); 242 struct adf_accel_dev *accel_dev = wq_data->accel_dev; 243 struct adf_hw_device_data *hw_device = accel_dev->hw_device; 244 245 adf_error_notifier(accel_dev); 246 247 if (!accel_dev->is_vf) { 248 /* Disable arbitration to stop processing of new requests */ 249 if (accel_dev->autoreset_on_error && hw_device->exit_arb) 250 hw_device->exit_arb(accel_dev); 251 if (accel_dev->pf.vf_info) 252 adf_pf2vf_notify_fatal_error(accel_dev); 253 adf_dev_autoreset(accel_dev); 254 } 255 256 kfree(wq_data); 257 } 258 259 int adf_notify_fatal_error(struct adf_accel_dev *accel_dev) 260 { 261 struct adf_fatal_error_data *wq_data; 262 263 wq_data = kzalloc(sizeof(*wq_data), GFP_ATOMIC); 264 if (!wq_data) 265 return -ENOMEM; 266 267 wq_data->accel_dev = accel_dev; 268 INIT_WORK(&wq_data->work, adf_notify_fatal_error_worker); 269 adf_misc_wq_queue_work(&wq_data->work); 270 271 return 0; 272 } 273 274 int adf_init_aer(void) 275 { 276 device_reset_wq = alloc_workqueue("qat_device_reset_wq", 277 WQ_MEM_RECLAIM | WQ_PERCPU, 0); 278 if (!device_reset_wq) 279 return -EFAULT; 280 281 device_sriov_wq = alloc_workqueue("qat_device_sriov_wq", WQ_PERCPU, 0); 282 if (!device_sriov_wq) { 283 destroy_workqueue(device_reset_wq); 284 device_reset_wq = NULL; 285 return -EFAULT; 286 } 287 288 return 0; 289 } 290 291 void adf_exit_aer(void) 292 { 293 if (device_reset_wq) 294 destroy_workqueue(device_reset_wq); 295 device_reset_wq = NULL; 296 297 if (device_sriov_wq) 298 destroy_workqueue(device_sriov_wq); 299 device_sriov_wq = NULL; 300 } 301