1 // SPDX-License-Identifier: GPL-2.0-only 2 /* Copyright(c) 2022 Intel Corporation. */ 3 4 #include <linux/cpu.h> 5 #include <linux/delay.h> 6 #include <linux/fs.h> 7 #include <linux/nmi.h> 8 #include <linux/slab.h> 9 #include <linux/stop_machine.h> 10 11 #include "ifs.h" 12 13 /* 14 * Note all code and data in this file is protected by 15 * ifs_sem. On HT systems all threads on a core will 16 * execute together, but only the first thread on the 17 * core will update results of the test. 18 */ 19 20 #define CREATE_TRACE_POINTS 21 #include <trace/events/intel_ifs.h> 22 23 /* Max retries on the same chunk */ 24 #define MAX_IFS_RETRIES 5 25 26 struct run_params { 27 struct ifs_data *ifsd; 28 union ifs_scan *activate; 29 union ifs_status status; 30 }; 31 32 /* 33 * Number of TSC cycles that a logical CPU will wait for the other 34 * logical CPU on the core in the WRMSR(ACTIVATE_SCAN). 35 */ 36 #define IFS_THREAD_WAIT 100000 37 38 enum ifs_status_err_code { 39 IFS_NO_ERROR = 0, 40 IFS_OTHER_THREAD_COULD_NOT_JOIN = 1, 41 IFS_INTERRUPTED_BEFORE_RENDEZVOUS = 2, 42 IFS_POWER_MGMT_INADEQUATE_FOR_SCAN = 3, 43 IFS_INVALID_CHUNK_RANGE = 4, 44 IFS_MISMATCH_ARGUMENTS_BETWEEN_THREADS = 5, 45 IFS_CORE_NOT_CAPABLE_CURRENTLY = 6, 46 IFS_UNASSIGNED_ERROR_CODE = 7, 47 IFS_EXCEED_NUMBER_OF_THREADS_CONCURRENT = 8, 48 IFS_INTERRUPTED_DURING_EXECUTION = 9, 49 IFS_UNASSIGNED_ERROR_CODE_0xA = 0xA, 50 IFS_CORRUPTED_CHUNK = 0xB, 51 }; 52 53 static const char * const scan_test_status[] = { 54 [IFS_NO_ERROR] = "SCAN no error", 55 [IFS_OTHER_THREAD_COULD_NOT_JOIN] = "Other thread could not join.", 56 [IFS_INTERRUPTED_BEFORE_RENDEZVOUS] = "Interrupt occurred prior to SCAN coordination.", 57 [IFS_POWER_MGMT_INADEQUATE_FOR_SCAN] = 58 "Core Abort SCAN Response due to power management condition.", 59 [IFS_INVALID_CHUNK_RANGE] = "Non valid chunks in the range", 60 [IFS_MISMATCH_ARGUMENTS_BETWEEN_THREADS] = "Mismatch in arguments between threads T0/T1.", 61 [IFS_CORE_NOT_CAPABLE_CURRENTLY] = "Core not capable of performing SCAN currently", 62 [IFS_UNASSIGNED_ERROR_CODE] = "Unassigned error code 0x7", 63 [IFS_EXCEED_NUMBER_OF_THREADS_CONCURRENT] = 64 "Exceeded number of Logical Processors (LP) allowed to run Scan-At-Field concurrently", 65 [IFS_INTERRUPTED_DURING_EXECUTION] = "Interrupt occurred prior to SCAN start", 66 [IFS_UNASSIGNED_ERROR_CODE_0xA] = "Unassigned error code 0xA", 67 [IFS_CORRUPTED_CHUNK] = "Scan operation aborted due to corrupted image. Try reloading", 68 }; 69 70 static void message_not_tested(struct device *dev, int cpu, union ifs_status status) 71 { 72 if (status.error_code < ARRAY_SIZE(scan_test_status)) { 73 dev_info(dev, "CPU(s) %*pbl: SCAN operation did not start. %s\n", 74 cpumask_pr_args(cpu_smt_mask(cpu)), 75 scan_test_status[status.error_code]); 76 } else if (status.error_code == IFS_SW_TIMEOUT) { 77 dev_info(dev, "CPU(s) %*pbl: software timeout during scan\n", 78 cpumask_pr_args(cpu_smt_mask(cpu))); 79 } else if (status.error_code == IFS_SW_PARTIAL_COMPLETION) { 80 dev_info(dev, "CPU(s) %*pbl: %s\n", 81 cpumask_pr_args(cpu_smt_mask(cpu)), 82 "Not all scan chunks were executed. Maximum forward progress retries exceeded"); 83 } else { 84 dev_info(dev, "CPU(s) %*pbl: SCAN unknown status %llx\n", 85 cpumask_pr_args(cpu_smt_mask(cpu)), status.data); 86 } 87 } 88 89 static void message_fail(struct device *dev, int cpu, union ifs_status status) 90 { 91 struct ifs_data *ifsd = ifs_get_data(dev); 92 93 /* 94 * control_error is set when the microcode runs into a problem 95 * loading the image from the reserved BIOS memory, or it has 96 * been corrupted. Reloading the image may fix this issue. 97 */ 98 if (status.control_error) { 99 dev_err(dev, "CPU(s) %*pbl: could not execute from loaded scan image. Batch: %02x version: 0x%x\n", 100 cpumask_pr_args(cpu_smt_mask(cpu)), ifsd->cur_batch, ifsd->loaded_version); 101 } 102 103 /* 104 * signature_error is set when the output from the scan chains does not 105 * match the expected signature. This might be a transient problem (e.g. 106 * due to a bit flip from an alpha particle or neutron). If the problem 107 * repeats on a subsequent test, then it indicates an actual problem in 108 * the core being tested. 109 */ 110 if (status.signature_error) { 111 dev_err(dev, "CPU(s) %*pbl: test signature incorrect. Batch: %02x version: 0x%x\n", 112 cpumask_pr_args(cpu_smt_mask(cpu)), ifsd->cur_batch, ifsd->loaded_version); 113 } 114 } 115 116 static bool can_restart(union ifs_status status) 117 { 118 enum ifs_status_err_code err_code = status.error_code; 119 120 /* Signature for chunk is bad, or scan test failed */ 121 if (status.signature_error || status.control_error) 122 return false; 123 124 switch (err_code) { 125 case IFS_NO_ERROR: 126 case IFS_OTHER_THREAD_COULD_NOT_JOIN: 127 case IFS_INTERRUPTED_BEFORE_RENDEZVOUS: 128 case IFS_POWER_MGMT_INADEQUATE_FOR_SCAN: 129 case IFS_EXCEED_NUMBER_OF_THREADS_CONCURRENT: 130 case IFS_INTERRUPTED_DURING_EXECUTION: 131 return true; 132 case IFS_INVALID_CHUNK_RANGE: 133 case IFS_MISMATCH_ARGUMENTS_BETWEEN_THREADS: 134 case IFS_CORE_NOT_CAPABLE_CURRENTLY: 135 case IFS_UNASSIGNED_ERROR_CODE: 136 case IFS_UNASSIGNED_ERROR_CODE_0xA: 137 case IFS_CORRUPTED_CHUNK: 138 break; 139 } 140 return false; 141 } 142 143 #define SPINUNIT 100 /* 100 nsec */ 144 static atomic_t array_cpus_in; 145 static atomic_t scan_cpus_in; 146 147 /* 148 * Simplified cpu sibling rendezvous loop based on microcode loader __wait_for_cpus() 149 */ 150 static void wait_for_sibling_cpu(atomic_t *t, long long timeout) 151 { 152 int cpu = smp_processor_id(); 153 const struct cpumask *smt_mask = cpu_smt_mask(cpu); 154 int all_cpus = cpumask_weight(smt_mask); 155 156 atomic_inc(t); 157 while (atomic_read(t) < all_cpus) { 158 if (timeout < SPINUNIT) 159 return; 160 ndelay(SPINUNIT); 161 timeout -= SPINUNIT; 162 touch_nmi_watchdog(); 163 } 164 } 165 166 /* 167 * Execute the scan. Called "simultaneously" on all threads of a core 168 * at high priority using the stop_cpus mechanism. 169 */ 170 static int doscan(void *data) 171 { 172 int cpu = smp_processor_id(), start, stop; 173 struct run_params *params = data; 174 union ifs_status status; 175 struct ifs_data *ifsd; 176 int first; 177 178 ifsd = params->ifsd; 179 180 if (ifsd->generation) { 181 start = params->activate->gen2.start; 182 stop = params->activate->gen2.stop; 183 } else { 184 start = params->activate->gen0.start; 185 stop = params->activate->gen0.stop; 186 } 187 188 /* Only the first logical CPU on a core reports result */ 189 first = cpumask_first(cpu_smt_mask(cpu)); 190 191 wait_for_sibling_cpu(&scan_cpus_in, NSEC_PER_SEC); 192 193 /* 194 * This WRMSR will wait for other HT threads to also write 195 * to this MSR (at most for activate.delay cycles). Then it 196 * starts scan of each requested chunk. The core scan happens 197 * during the "execution" of the WRMSR. This instruction can 198 * take up to 200 milliseconds (in the case where all chunks 199 * are processed in a single pass) before it retires. 200 */ 201 wrmsrl(MSR_ACTIVATE_SCAN, params->activate->data); 202 rdmsrl(MSR_SCAN_STATUS, status.data); 203 204 trace_ifs_status(ifsd->cur_batch, start, stop, status.data); 205 206 /* Pass back the result of the scan */ 207 if (cpu == first) 208 params->status = status; 209 210 return 0; 211 } 212 213 /* 214 * Use stop_core_cpuslocked() to synchronize writing to MSR_ACTIVATE_SCAN 215 * on all threads of the core to be tested. Loop if necessary to complete 216 * run of all chunks. Include some defensive tests to make sure forward 217 * progress is made, and that the whole test completes in a reasonable time. 218 */ 219 static void ifs_test_core(int cpu, struct device *dev) 220 { 221 union ifs_scan activate; 222 union ifs_status status; 223 unsigned long timeout; 224 struct ifs_data *ifsd; 225 int to_start, to_stop; 226 int status_chunk; 227 struct run_params params; 228 int retries; 229 230 ifsd = ifs_get_data(dev); 231 232 activate.gen0.rsvd = 0; 233 activate.delay = IFS_THREAD_WAIT; 234 activate.sigmce = 0; 235 to_start = 0; 236 to_stop = ifsd->valid_chunks - 1; 237 238 params.ifsd = ifs_get_data(dev); 239 240 if (ifsd->generation) { 241 activate.gen2.start = to_start; 242 activate.gen2.stop = to_stop; 243 } else { 244 activate.gen0.start = to_start; 245 activate.gen0.stop = to_stop; 246 } 247 248 timeout = jiffies + HZ / 2; 249 retries = MAX_IFS_RETRIES; 250 251 while (to_start <= to_stop) { 252 if (time_after(jiffies, timeout)) { 253 status.error_code = IFS_SW_TIMEOUT; 254 break; 255 } 256 257 params.activate = &activate; 258 atomic_set(&scan_cpus_in, 0); 259 stop_core_cpuslocked(cpu, doscan, ¶ms); 260 261 status = params.status; 262 263 /* Some cases can be retried, give up for others */ 264 if (!can_restart(status)) 265 break; 266 267 status_chunk = ifsd->generation ? status.gen2.chunk_num : status.gen0.chunk_num; 268 if (status_chunk == to_start) { 269 /* Check for forward progress */ 270 if (--retries == 0) { 271 if (status.error_code == IFS_NO_ERROR) 272 status.error_code = IFS_SW_PARTIAL_COMPLETION; 273 break; 274 } 275 } else { 276 retries = MAX_IFS_RETRIES; 277 if (ifsd->generation) 278 activate.gen2.start = status_chunk; 279 else 280 activate.gen0.start = status_chunk; 281 to_start = status_chunk; 282 } 283 } 284 285 /* Update status for this core */ 286 ifsd->scan_details = status.data; 287 288 if (status.control_error || status.signature_error) { 289 ifsd->status = SCAN_TEST_FAIL; 290 message_fail(dev, cpu, status); 291 } else if (status.error_code) { 292 ifsd->status = SCAN_NOT_TESTED; 293 message_not_tested(dev, cpu, status); 294 } else { 295 ifsd->status = SCAN_TEST_PASS; 296 } 297 } 298 299 static int do_array_test(void *data) 300 { 301 union ifs_array *command = data; 302 int cpu = smp_processor_id(); 303 int first; 304 305 wait_for_sibling_cpu(&array_cpus_in, NSEC_PER_SEC); 306 307 /* 308 * Only one logical CPU on a core needs to trigger the Array test via MSR write. 309 */ 310 first = cpumask_first(cpu_smt_mask(cpu)); 311 312 if (cpu == first) { 313 wrmsrl(MSR_ARRAY_BIST, command->data); 314 /* Pass back the result of the test */ 315 rdmsrl(MSR_ARRAY_BIST, command->data); 316 } 317 318 return 0; 319 } 320 321 static void ifs_array_test_core(int cpu, struct device *dev) 322 { 323 union ifs_array command = {}; 324 bool timed_out = false; 325 struct ifs_data *ifsd; 326 unsigned long timeout; 327 328 ifsd = ifs_get_data(dev); 329 330 command.array_bitmask = ~0U; 331 timeout = jiffies + HZ / 2; 332 333 do { 334 if (time_after(jiffies, timeout)) { 335 timed_out = true; 336 break; 337 } 338 atomic_set(&array_cpus_in, 0); 339 stop_core_cpuslocked(cpu, do_array_test, &command); 340 341 if (command.ctrl_result) 342 break; 343 } while (command.array_bitmask); 344 345 ifsd->scan_details = command.data; 346 347 if (command.ctrl_result) 348 ifsd->status = SCAN_TEST_FAIL; 349 else if (timed_out || command.array_bitmask) 350 ifsd->status = SCAN_NOT_TESTED; 351 else 352 ifsd->status = SCAN_TEST_PASS; 353 } 354 355 #define ARRAY_GEN1_TEST_ALL_ARRAYS 0x0ULL 356 #define ARRAY_GEN1_STATUS_FAIL 0x1ULL 357 358 static int do_array_test_gen1(void *status) 359 { 360 int cpu = smp_processor_id(); 361 int first; 362 363 first = cpumask_first(cpu_smt_mask(cpu)); 364 365 if (cpu == first) { 366 wrmsrl(MSR_ARRAY_TRIGGER, ARRAY_GEN1_TEST_ALL_ARRAYS); 367 rdmsrl(MSR_ARRAY_STATUS, *((u64 *)status)); 368 } 369 370 return 0; 371 } 372 373 static void ifs_array_test_gen1(int cpu, struct device *dev) 374 { 375 struct ifs_data *ifsd = ifs_get_data(dev); 376 u64 status = 0; 377 378 stop_core_cpuslocked(cpu, do_array_test_gen1, &status); 379 ifsd->scan_details = status; 380 381 if (status & ARRAY_GEN1_STATUS_FAIL) 382 ifsd->status = SCAN_TEST_FAIL; 383 else 384 ifsd->status = SCAN_TEST_PASS; 385 } 386 387 /* 388 * Initiate per core test. It wakes up work queue threads on the target cpu and 389 * its sibling cpu. Once all sibling threads wake up, the scan test gets executed and 390 * wait for all sibling threads to finish the scan test. 391 */ 392 int do_core_test(int cpu, struct device *dev) 393 { 394 const struct ifs_test_caps *test = ifs_get_test_caps(dev); 395 struct ifs_data *ifsd = ifs_get_data(dev); 396 int ret = 0; 397 398 /* Prevent CPUs from being taken offline during the scan test */ 399 cpus_read_lock(); 400 401 if (!cpu_online(cpu)) { 402 dev_info(dev, "cannot test on the offline cpu %d\n", cpu); 403 ret = -EINVAL; 404 goto out; 405 } 406 407 switch (test->test_num) { 408 case IFS_TYPE_SAF: 409 if (!ifsd->loaded) 410 ret = -EPERM; 411 else 412 ifs_test_core(cpu, dev); 413 break; 414 case IFS_TYPE_ARRAY_BIST: 415 if (ifsd->array_gen == ARRAY_GEN0) 416 ifs_array_test_core(cpu, dev); 417 else 418 ifs_array_test_gen1(cpu, dev); 419 break; 420 default: 421 ret = -EINVAL; 422 } 423 out: 424 cpus_read_unlock(); 425 return ret; 426 } 427