1 /* 2 * PowerNV OPAL high level interfaces 3 * 4 * Copyright 2011 IBM Corp. 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License 8 * as published by the Free Software Foundation; either version 9 * 2 of the License, or (at your option) any later version. 10 */ 11 12 #undef DEBUG 13 14 #include <linux/types.h> 15 #include <linux/of.h> 16 #include <linux/of_fdt.h> 17 #include <linux/of_platform.h> 18 #include <linux/interrupt.h> 19 #include <linux/notifier.h> 20 #include <linux/slab.h> 21 #include <linux/sched.h> 22 #include <linux/kobject.h> 23 #include <linux/delay.h> 24 #include <asm/opal.h> 25 #include <asm/firmware.h> 26 #include <asm/mce.h> 27 28 #include "powernv.h" 29 30 /* /sys/firmware/opal */ 31 struct kobject *opal_kobj; 32 33 struct opal { 34 u64 base; 35 u64 entry; 36 } opal; 37 38 static struct device_node *opal_node; 39 static DEFINE_SPINLOCK(opal_write_lock); 40 extern u64 opal_mc_secondary_handler[]; 41 static unsigned int *opal_irqs; 42 static unsigned int opal_irq_count; 43 static ATOMIC_NOTIFIER_HEAD(opal_notifier_head); 44 static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX]; 45 static DEFINE_SPINLOCK(opal_notifier_lock); 46 static uint64_t last_notified_mask = 0x0ul; 47 static atomic_t opal_notifier_hold = ATOMIC_INIT(0); 48 49 int __init early_init_dt_scan_opal(unsigned long node, 50 const char *uname, int depth, void *data) 51 { 52 const void *basep, *entryp; 53 unsigned long basesz, entrysz; 54 55 if (depth != 1 || strcmp(uname, "ibm,opal") != 0) 56 return 0; 57 58 basep = of_get_flat_dt_prop(node, "opal-base-address", &basesz); 59 entryp = of_get_flat_dt_prop(node, "opal-entry-address", &entrysz); 60 61 if (!basep || !entryp) 62 return 1; 63 64 opal.base = of_read_number(basep, basesz/4); 65 opal.entry = of_read_number(entryp, entrysz/4); 66 67 pr_debug("OPAL Base = 0x%llx (basep=%p basesz=%ld)\n", 68 opal.base, basep, basesz); 69 pr_debug("OPAL Entry = 0x%llx (entryp=%p basesz=%ld)\n", 70 opal.entry, entryp, entrysz); 71 72 powerpc_firmware_features |= FW_FEATURE_OPAL; 73 if (of_flat_dt_is_compatible(node, "ibm,opal-v3")) { 74 powerpc_firmware_features |= FW_FEATURE_OPALv2; 75 powerpc_firmware_features |= FW_FEATURE_OPALv3; 76 printk("OPAL V3 detected !\n"); 77 } else if (of_flat_dt_is_compatible(node, "ibm,opal-v2")) { 78 powerpc_firmware_features |= FW_FEATURE_OPALv2; 79 printk("OPAL V2 detected !\n"); 80 } else { 81 printk("OPAL V1 detected !\n"); 82 } 83 84 return 1; 85 } 86 87 static int __init opal_register_exception_handlers(void) 88 { 89 #ifdef __BIG_ENDIAN__ 90 u64 glue; 91 92 if (!(powerpc_firmware_features & FW_FEATURE_OPAL)) 93 return -ENODEV; 94 95 /* Hookup some exception handlers except machine check. We use the 96 * fwnmi area at 0x7000 to provide the glue space to OPAL 97 */ 98 glue = 0x7000; 99 opal_register_exception_handler(OPAL_HYPERVISOR_MAINTENANCE_HANDLER, 100 0, glue); 101 glue += 128; 102 opal_register_exception_handler(OPAL_SOFTPATCH_HANDLER, 0, glue); 103 #endif 104 105 return 0; 106 } 107 108 early_initcall(opal_register_exception_handlers); 109 110 int opal_notifier_register(struct notifier_block *nb) 111 { 112 if (!nb) { 113 pr_warning("%s: Invalid argument (%p)\n", 114 __func__, nb); 115 return -EINVAL; 116 } 117 118 atomic_notifier_chain_register(&opal_notifier_head, nb); 119 return 0; 120 } 121 122 static void opal_do_notifier(uint64_t events) 123 { 124 unsigned long flags; 125 uint64_t changed_mask; 126 127 if (atomic_read(&opal_notifier_hold)) 128 return; 129 130 spin_lock_irqsave(&opal_notifier_lock, flags); 131 changed_mask = last_notified_mask ^ events; 132 last_notified_mask = events; 133 spin_unlock_irqrestore(&opal_notifier_lock, flags); 134 135 /* 136 * We feed with the event bits and changed bits for 137 * enough information to the callback. 138 */ 139 atomic_notifier_call_chain(&opal_notifier_head, 140 events, (void *)changed_mask); 141 } 142 143 void opal_notifier_update_evt(uint64_t evt_mask, 144 uint64_t evt_val) 145 { 146 unsigned long flags; 147 148 spin_lock_irqsave(&opal_notifier_lock, flags); 149 last_notified_mask &= ~evt_mask; 150 last_notified_mask |= evt_val; 151 spin_unlock_irqrestore(&opal_notifier_lock, flags); 152 } 153 154 void opal_notifier_enable(void) 155 { 156 int64_t rc; 157 uint64_t evt = 0; 158 159 atomic_set(&opal_notifier_hold, 0); 160 161 /* Process pending events */ 162 rc = opal_poll_events(&evt); 163 if (rc == OPAL_SUCCESS && evt) 164 opal_do_notifier(evt); 165 } 166 167 void opal_notifier_disable(void) 168 { 169 atomic_set(&opal_notifier_hold, 1); 170 } 171 172 /* 173 * Opal message notifier based on message type. Allow subscribers to get 174 * notified for specific messgae type. 175 */ 176 int opal_message_notifier_register(enum OpalMessageType msg_type, 177 struct notifier_block *nb) 178 { 179 if (!nb) { 180 pr_warning("%s: Invalid argument (%p)\n", 181 __func__, nb); 182 return -EINVAL; 183 } 184 if (msg_type > OPAL_MSG_TYPE_MAX) { 185 pr_warning("%s: Invalid message type argument (%d)\n", 186 __func__, msg_type); 187 return -EINVAL; 188 } 189 return atomic_notifier_chain_register( 190 &opal_msg_notifier_head[msg_type], nb); 191 } 192 193 static void opal_message_do_notify(uint32_t msg_type, void *msg) 194 { 195 /* notify subscribers */ 196 atomic_notifier_call_chain(&opal_msg_notifier_head[msg_type], 197 msg_type, msg); 198 } 199 200 static void opal_handle_message(void) 201 { 202 s64 ret; 203 /* 204 * TODO: pre-allocate a message buffer depending on opal-msg-size 205 * value in /proc/device-tree. 206 */ 207 static struct opal_msg msg; 208 209 ret = opal_get_msg(__pa(&msg), sizeof(msg)); 210 /* No opal message pending. */ 211 if (ret == OPAL_RESOURCE) 212 return; 213 214 /* check for errors. */ 215 if (ret) { 216 pr_warning("%s: Failed to retrive opal message, err=%lld\n", 217 __func__, ret); 218 return; 219 } 220 221 /* Sanity check */ 222 if (msg.msg_type > OPAL_MSG_TYPE_MAX) { 223 pr_warning("%s: Unknown message type: %u\n", 224 __func__, msg.msg_type); 225 return; 226 } 227 opal_message_do_notify(msg.msg_type, (void *)&msg); 228 } 229 230 static int opal_message_notify(struct notifier_block *nb, 231 unsigned long events, void *change) 232 { 233 if (events & OPAL_EVENT_MSG_PENDING) 234 opal_handle_message(); 235 return 0; 236 } 237 238 static struct notifier_block opal_message_nb = { 239 .notifier_call = opal_message_notify, 240 .next = NULL, 241 .priority = 0, 242 }; 243 244 static int __init opal_message_init(void) 245 { 246 int ret, i; 247 248 for (i = 0; i < OPAL_MSG_TYPE_MAX; i++) 249 ATOMIC_INIT_NOTIFIER_HEAD(&opal_msg_notifier_head[i]); 250 251 ret = opal_notifier_register(&opal_message_nb); 252 if (ret) { 253 pr_err("%s: Can't register OPAL event notifier (%d)\n", 254 __func__, ret); 255 return ret; 256 } 257 return 0; 258 } 259 early_initcall(opal_message_init); 260 261 int opal_get_chars(uint32_t vtermno, char *buf, int count) 262 { 263 s64 rc; 264 __be64 evt, len; 265 266 if (!opal.entry) 267 return -ENODEV; 268 opal_poll_events(&evt); 269 if ((be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_INPUT) == 0) 270 return 0; 271 len = cpu_to_be64(count); 272 rc = opal_console_read(vtermno, &len, buf); 273 if (rc == OPAL_SUCCESS) 274 return be64_to_cpu(len); 275 return 0; 276 } 277 278 int opal_put_chars(uint32_t vtermno, const char *data, int total_len) 279 { 280 int written = 0; 281 __be64 olen; 282 s64 len, rc; 283 unsigned long flags; 284 __be64 evt; 285 286 if (!opal.entry) 287 return -ENODEV; 288 289 /* We want put_chars to be atomic to avoid mangling of hvsi 290 * packets. To do that, we first test for room and return 291 * -EAGAIN if there isn't enough. 292 * 293 * Unfortunately, opal_console_write_buffer_space() doesn't 294 * appear to work on opal v1, so we just assume there is 295 * enough room and be done with it 296 */ 297 spin_lock_irqsave(&opal_write_lock, flags); 298 if (firmware_has_feature(FW_FEATURE_OPALv2)) { 299 rc = opal_console_write_buffer_space(vtermno, &olen); 300 len = be64_to_cpu(olen); 301 if (rc || len < total_len) { 302 spin_unlock_irqrestore(&opal_write_lock, flags); 303 /* Closed -> drop characters */ 304 if (rc) 305 return total_len; 306 opal_poll_events(NULL); 307 return -EAGAIN; 308 } 309 } 310 311 /* We still try to handle partial completions, though they 312 * should no longer happen. 313 */ 314 rc = OPAL_BUSY; 315 while(total_len > 0 && (rc == OPAL_BUSY || 316 rc == OPAL_BUSY_EVENT || rc == OPAL_SUCCESS)) { 317 olen = cpu_to_be64(total_len); 318 rc = opal_console_write(vtermno, &olen, data); 319 len = be64_to_cpu(olen); 320 321 /* Closed or other error drop */ 322 if (rc != OPAL_SUCCESS && rc != OPAL_BUSY && 323 rc != OPAL_BUSY_EVENT) { 324 written = total_len; 325 break; 326 } 327 if (rc == OPAL_SUCCESS) { 328 total_len -= len; 329 data += len; 330 written += len; 331 } 332 /* This is a bit nasty but we need that for the console to 333 * flush when there aren't any interrupts. We will clean 334 * things a bit later to limit that to synchronous path 335 * such as the kernel console and xmon/udbg 336 */ 337 do 338 opal_poll_events(&evt); 339 while(rc == OPAL_SUCCESS && 340 (be64_to_cpu(evt) & OPAL_EVENT_CONSOLE_OUTPUT)); 341 } 342 spin_unlock_irqrestore(&opal_write_lock, flags); 343 return written; 344 } 345 346 static int opal_recover_mce(struct pt_regs *regs, 347 struct machine_check_event *evt) 348 { 349 int recovered = 0; 350 uint64_t ea = get_mce_fault_addr(evt); 351 352 if (!(regs->msr & MSR_RI)) { 353 /* If MSR_RI isn't set, we cannot recover */ 354 recovered = 0; 355 } else if (evt->disposition == MCE_DISPOSITION_RECOVERED) { 356 /* Platform corrected itself */ 357 recovered = 1; 358 } else if (ea && !is_kernel_addr(ea)) { 359 /* 360 * Faulting address is not in kernel text. We should be fine. 361 * We need to find which process uses this address. 362 * For now, kill the task if we have received exception when 363 * in userspace. 364 * 365 * TODO: Queue up this address for hwpoisioning later. 366 */ 367 if (user_mode(regs) && !is_global_init(current)) { 368 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip); 369 recovered = 1; 370 } else 371 recovered = 0; 372 } else if (user_mode(regs) && !is_global_init(current) && 373 evt->severity == MCE_SEV_ERROR_SYNC) { 374 /* 375 * If we have received a synchronous error when in userspace 376 * kill the task. 377 */ 378 _exception(SIGBUS, regs, BUS_MCEERR_AR, regs->nip); 379 recovered = 1; 380 } 381 return recovered; 382 } 383 384 int opal_machine_check(struct pt_regs *regs) 385 { 386 struct machine_check_event evt; 387 388 if (!get_mce_event(&evt, MCE_EVENT_RELEASE)) 389 return 0; 390 391 /* Print things out */ 392 if (evt.version != MCE_V1) { 393 pr_err("Machine Check Exception, Unknown event version %d !\n", 394 evt.version); 395 return 0; 396 } 397 machine_check_print_event_info(&evt); 398 399 if (opal_recover_mce(regs, &evt)) 400 return 1; 401 return 0; 402 } 403 404 static irqreturn_t opal_interrupt(int irq, void *data) 405 { 406 __be64 events; 407 408 opal_handle_interrupt(virq_to_hw(irq), &events); 409 410 opal_do_notifier(events); 411 412 return IRQ_HANDLED; 413 } 414 415 static int opal_sysfs_init(void) 416 { 417 opal_kobj = kobject_create_and_add("opal", firmware_kobj); 418 if (!opal_kobj) { 419 pr_warn("kobject_create_and_add opal failed\n"); 420 return -ENOMEM; 421 } 422 423 return 0; 424 } 425 426 static int __init opal_init(void) 427 { 428 struct device_node *np, *consoles; 429 const __be32 *irqs; 430 int rc, i, irqlen; 431 432 opal_node = of_find_node_by_path("/ibm,opal"); 433 if (!opal_node) { 434 pr_warn("opal: Node not found\n"); 435 return -ENODEV; 436 } 437 438 /* Register OPAL consoles if any ports */ 439 if (firmware_has_feature(FW_FEATURE_OPALv2)) 440 consoles = of_find_node_by_path("/ibm,opal/consoles"); 441 else 442 consoles = of_node_get(opal_node); 443 if (consoles) { 444 for_each_child_of_node(consoles, np) { 445 if (strcmp(np->name, "serial")) 446 continue; 447 of_platform_device_create(np, NULL, NULL); 448 } 449 of_node_put(consoles); 450 } 451 452 /* Find all OPAL interrupts and request them */ 453 irqs = of_get_property(opal_node, "opal-interrupts", &irqlen); 454 pr_debug("opal: Found %d interrupts reserved for OPAL\n", 455 irqs ? (irqlen / 4) : 0); 456 opal_irq_count = irqlen / 4; 457 opal_irqs = kzalloc(opal_irq_count * sizeof(unsigned int), GFP_KERNEL); 458 for (i = 0; irqs && i < (irqlen / 4); i++, irqs++) { 459 unsigned int hwirq = be32_to_cpup(irqs); 460 unsigned int irq = irq_create_mapping(NULL, hwirq); 461 if (irq == NO_IRQ) { 462 pr_warning("opal: Failed to map irq 0x%x\n", hwirq); 463 continue; 464 } 465 rc = request_irq(irq, opal_interrupt, 0, "opal", NULL); 466 if (rc) 467 pr_warning("opal: Error %d requesting irq %d" 468 " (0x%x)\n", rc, irq, hwirq); 469 opal_irqs[i] = irq; 470 } 471 472 /* Create "opal" kobject under /sys/firmware */ 473 rc = opal_sysfs_init(); 474 if (rc == 0) { 475 /* Setup code update interface */ 476 opal_flash_init(); 477 } 478 479 return 0; 480 } 481 subsys_initcall(opal_init); 482 483 void opal_shutdown(void) 484 { 485 unsigned int i; 486 long rc = OPAL_BUSY; 487 488 /* First free interrupts, which will also mask them */ 489 for (i = 0; i < opal_irq_count; i++) { 490 if (opal_irqs[i]) 491 free_irq(opal_irqs[i], NULL); 492 opal_irqs[i] = 0; 493 } 494 495 /* 496 * Then sync with OPAL which ensure anything that can 497 * potentially write to our memory has completed such 498 * as an ongoing dump retrieval 499 */ 500 while (rc == OPAL_BUSY || rc == OPAL_BUSY_EVENT) { 501 rc = opal_sync_host_reboot(); 502 if (rc == OPAL_BUSY) 503 opal_poll_events(NULL); 504 else 505 mdelay(10); 506 } 507 } 508