1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 /* 28 * PSMI 1.1 extensions are supported only in 2.6 and later versions. 29 * PSMI 1.2 extensions are supported only in 2.7 and later versions. 30 * PSMI 1.3 and 1.4 extensions are supported in Solaris 10. 31 * PSMI 1.5 extensions are supported in Solaris Nevada. 32 * PSMI 1.6 extensions are supported in Solaris Nevada. 33 */ 34 #define PSMI_1_6 35 36 #include <sys/processor.h> 37 #include <sys/time.h> 38 #include <sys/psm.h> 39 #include <sys/smp_impldefs.h> 40 #include <sys/cram.h> 41 #include <sys/acpi/acpi.h> 42 #include <sys/acpica.h> 43 #include <sys/psm_common.h> 44 #include <sys/apic.h> 45 #include <sys/pit.h> 46 #include <sys/ddi.h> 47 #include <sys/sunddi.h> 48 #include <sys/ddi_impldefs.h> 49 #include <sys/pci.h> 50 #include <sys/promif.h> 51 #include <sys/x86_archext.h> 52 #include <sys/cpc_impl.h> 53 #include <sys/uadmin.h> 54 #include <sys/panic.h> 55 #include <sys/debug.h> 56 #include <sys/archsystm.h> 57 #include <sys/trap.h> 58 #include <sys/machsystm.h> 59 #include <sys/sysmacros.h> 60 #include <sys/cpuvar.h> 61 #include <sys/rm_platter.h> 62 #include <sys/privregs.h> 63 #include <sys/note.h> 64 #include <sys/pci_intr_lib.h> 65 #include <sys/spl.h> 66 #include <sys/clock.h> 67 #include <sys/dditypes.h> 68 #include <sys/sunddi.h> 69 #include <sys/x_call.h> 70 #include <sys/reboot.h> 71 #include <sys/hpet.h> 72 73 /* 74 * Local Function Prototypes 75 */ 76 static void apic_init_intr(); 77 static void apic_nmi_intr(caddr_t arg, struct regs *rp); 78 79 /* 80 * standard MP entries 81 */ 82 static int apic_probe(); 83 static int apic_clkinit(); 84 static int apic_getclkirq(int ipl); 85 static uint_t apic_calibrate(volatile uint32_t *addr, 86 uint16_t *pit_ticks_adj); 87 static hrtime_t apic_gettime(); 88 static hrtime_t apic_gethrtime(); 89 static void apic_init(); 90 static void apic_picinit(void); 91 static int apic_cpu_start(processorid_t, caddr_t); 92 static int apic_post_cpu_start(void); 93 static void apic_send_ipi(int cpun, int ipl); 94 static void apic_set_idlecpu(processorid_t cpun); 95 static void apic_unset_idlecpu(processorid_t cpun); 96 static int apic_intr_enter(int ipl, int *vect); 97 static void apic_setspl(int ipl); 98 static void x2apic_setspl(int ipl); 99 static int apic_addspl(int ipl, int vector, int min_ipl, int max_ipl); 100 static int apic_delspl(int ipl, int vector, int min_ipl, int max_ipl); 101 static void apic_shutdown(int cmd, int fcn); 102 static void apic_preshutdown(int cmd, int fcn); 103 static int apic_disable_intr(processorid_t cpun); 104 static void apic_enable_intr(processorid_t cpun); 105 static processorid_t apic_get_next_processorid(processorid_t cpun); 106 static int apic_get_ipivect(int ipl, int type); 107 static void apic_timer_reprogram(hrtime_t time); 108 static void apic_timer_enable(void); 109 static void apic_timer_disable(void); 110 static void apic_post_cyclic_setup(void *arg); 111 static void apic_intrr_init(int apic_mode); 112 static void apic_record_ioapic_rdt(apic_irq_t *irq_ptr, ioapic_rdt_t *irdt); 113 static void apic_record_msi(apic_irq_t *irq_ptr, msi_regs_t *mregs); 114 115 static int apic_oneshot = 0; 116 int apic_oneshot_enable = 1; /* to allow disabling one-shot capability */ 117 118 /* Now the ones for Dynamic Interrupt distribution */ 119 int apic_enable_dynamic_migration = 0; 120 121 extern int apic_have_32bit_cr8; 122 123 /* 124 * These variables are frequently accessed in apic_intr_enter(), 125 * apic_intr_exit and apic_setspl, so group them together 126 */ 127 volatile uint32_t *apicadr = NULL; /* virtual addr of local APIC */ 128 int apic_setspl_delay = 1; /* apic_setspl - delay enable */ 129 int apic_clkvect; 130 131 /* vector at which error interrupts come in */ 132 int apic_errvect; 133 int apic_enable_error_intr = 1; 134 int apic_error_display_delay = 100; 135 136 /* vector at which performance counter overflow interrupts come in */ 137 int apic_cpcovf_vect; 138 int apic_enable_cpcovf_intr = 1; 139 140 /* vector at which CMCI interrupts come in */ 141 int apic_cmci_vect; 142 extern int cmi_enable_cmci; 143 extern void cmi_cmci_trap(void); 144 145 static kmutex_t cmci_cpu_setup_lock; /* protects cmci_cpu_setup_registered */ 146 static int cmci_cpu_setup_registered; 147 148 /* 149 * The following vector assignments influence the value of ipltopri and 150 * vectortoipl. Note that vectors 0 - 0x1f are not used. We can program 151 * idle to 0 and IPL 0 to 0xf to differentiate idle in case 152 * we care to do so in future. Note some IPLs which are rarely used 153 * will share the vector ranges and heavily used IPLs (5 and 6) have 154 * a wide range. 155 * 156 * This array is used to initialize apic_ipls[] (in apic_init()). 157 * 158 * IPL Vector range. as passed to intr_enter 159 * 0 none. 160 * 1,2,3 0x20-0x2f 0x0-0xf 161 * 4 0x30-0x3f 0x10-0x1f 162 * 5 0x40-0x5f 0x20-0x3f 163 * 6 0x60-0x7f 0x40-0x5f 164 * 7,8,9 0x80-0x8f 0x60-0x6f 165 * 10 0x90-0x9f 0x70-0x7f 166 * 11 0xa0-0xaf 0x80-0x8f 167 * ... ... 168 * 15 0xe0-0xef 0xc0-0xcf 169 * 15 0xf0-0xff 0xd0-0xdf 170 */ 171 uchar_t apic_vectortoipl[APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL] = { 172 3, 4, 5, 5, 6, 6, 9, 10, 11, 12, 13, 14, 15, 15 173 }; 174 /* 175 * The ipl of an ISR at vector X is apic_vectortoipl[X>>4] 176 * NOTE that this is vector as passed into intr_enter which is 177 * programmed vector - 0x20 (APIC_BASE_VECT) 178 */ 179 180 uchar_t apic_ipltopri[MAXIPL + 1]; /* unix ipl to apic pri */ 181 /* The taskpri to be programmed into apic to mask given ipl */ 182 183 #if defined(__amd64) 184 uchar_t apic_cr8pri[MAXIPL + 1]; /* unix ipl to cr8 pri */ 185 #endif 186 187 /* 188 * Correlation of the hardware vector to the IPL in use, initialized 189 * from apic_vectortoipl[] in apic_init(). The final IPLs may not correlate 190 * to the IPLs in apic_vectortoipl on some systems that share interrupt lines 191 * connected to errata-stricken IOAPICs 192 */ 193 uchar_t apic_ipls[APIC_AVAIL_VECTOR]; 194 195 /* 196 * Patchable global variables. 197 */ 198 int apic_forceload = 0; 199 200 int apic_coarse_hrtime = 1; /* 0 - use accurate slow gethrtime() */ 201 /* 1 - use gettime() for performance */ 202 int apic_flat_model = 0; /* 0 - clustered. 1 - flat */ 203 int apic_enable_hwsoftint = 0; /* 0 - disable, 1 - enable */ 204 int apic_enable_bind_log = 1; /* 1 - display interrupt binding log */ 205 int apic_panic_on_nmi = 0; 206 int apic_panic_on_apic_error = 0; 207 208 int apic_verbose = 0; 209 210 /* minimum number of timer ticks to program to */ 211 int apic_min_timer_ticks = 1; 212 /* 213 * Local static data 214 */ 215 static struct psm_ops apic_ops = { 216 apic_probe, 217 218 apic_init, 219 apic_picinit, 220 apic_intr_enter, 221 apic_intr_exit, 222 apic_setspl, 223 apic_addspl, 224 apic_delspl, 225 apic_disable_intr, 226 apic_enable_intr, 227 (int (*)(int))NULL, /* psm_softlvl_to_irq */ 228 (void (*)(int))NULL, /* psm_set_softintr */ 229 230 apic_set_idlecpu, 231 apic_unset_idlecpu, 232 233 apic_clkinit, 234 apic_getclkirq, 235 (void (*)(void))NULL, /* psm_hrtimeinit */ 236 apic_gethrtime, 237 238 apic_get_next_processorid, 239 apic_cpu_start, 240 apic_post_cpu_start, 241 apic_shutdown, 242 apic_get_ipivect, 243 apic_send_ipi, 244 245 (int (*)(dev_info_t *, int))NULL, /* psm_translate_irq */ 246 (void (*)(int, char *))NULL, /* psm_notify_error */ 247 (void (*)(int))NULL, /* psm_notify_func */ 248 apic_timer_reprogram, 249 apic_timer_enable, 250 apic_timer_disable, 251 apic_post_cyclic_setup, 252 apic_preshutdown, 253 apic_intr_ops, /* Advanced DDI Interrupt framework */ 254 apic_state, /* save, restore apic state for S3 */ 255 }; 256 257 258 static struct psm_info apic_psm_info = { 259 PSM_INFO_VER01_6, /* version */ 260 PSM_OWN_EXCLUSIVE, /* ownership */ 261 (struct psm_ops *)&apic_ops, /* operation */ 262 APIC_PCPLUSMP_NAME, /* machine name */ 263 "pcplusmp v1.4 compatible", 264 }; 265 266 static void *apic_hdlp; 267 268 #ifdef DEBUG 269 int apic_debug = 0; 270 int apic_restrict_vector = 0; 271 272 int apic_debug_msgbuf[APIC_DEBUG_MSGBUFSIZE]; 273 int apic_debug_msgbufindex = 0; 274 275 #endif /* DEBUG */ 276 277 apic_cpus_info_t *apic_cpus; 278 279 cpuset_t apic_cpumask; 280 uint_t apic_picinit_called; 281 282 /* Flag to indicate that we need to shut down all processors */ 283 static uint_t apic_shutdown_processors; 284 285 uint_t apic_nsec_per_intr = 0; 286 287 /* 288 * apic_let_idle_redistribute can have the following values: 289 * 0 - If clock decremented it from 1 to 0, clock has to call redistribute. 290 * apic_redistribute_lock prevents multiple idle cpus from redistributing 291 */ 292 int apic_num_idle_redistributions = 0; 293 static int apic_let_idle_redistribute = 0; 294 static uint_t apic_nticks = 0; 295 static uint_t apic_skipped_redistribute = 0; 296 297 /* to gather intr data and redistribute */ 298 static void apic_redistribute_compute(void); 299 300 static uint_t last_count_read = 0; 301 static lock_t apic_gethrtime_lock; 302 volatile int apic_hrtime_stamp = 0; 303 volatile hrtime_t apic_nsec_since_boot = 0; 304 static uint_t apic_hertz_count; 305 306 uint64_t apic_ticks_per_SFnsecs; /* # of ticks in SF nsecs */ 307 308 static hrtime_t apic_nsec_max; 309 310 static hrtime_t apic_last_hrtime = 0; 311 int apic_hrtime_error = 0; 312 int apic_remote_hrterr = 0; 313 int apic_num_nmis = 0; 314 int apic_apic_error = 0; 315 int apic_num_apic_errors = 0; 316 int apic_num_cksum_errors = 0; 317 318 int apic_error = 0; 319 static int apic_cmos_ssb_set = 0; 320 321 /* use to make sure only one cpu handles the nmi */ 322 static lock_t apic_nmi_lock; 323 /* use to make sure only one cpu handles the error interrupt */ 324 static lock_t apic_error_lock; 325 326 static struct { 327 uchar_t cntl; 328 uchar_t data; 329 } aspen_bmc[] = { 330 { CC_SMS_WR_START, 0x18 }, /* NetFn/LUN */ 331 { CC_SMS_WR_NEXT, 0x24 }, /* Cmd SET_WATCHDOG_TIMER */ 332 { CC_SMS_WR_NEXT, 0x84 }, /* DataByte 1: SMS/OS no log */ 333 { CC_SMS_WR_NEXT, 0x2 }, /* DataByte 2: Power Down */ 334 { CC_SMS_WR_NEXT, 0x0 }, /* DataByte 3: no pre-timeout */ 335 { CC_SMS_WR_NEXT, 0x0 }, /* DataByte 4: timer expir. */ 336 { CC_SMS_WR_NEXT, 0xa }, /* DataByte 5: init countdown */ 337 { CC_SMS_WR_END, 0x0 }, /* DataByte 6: init countdown */ 338 339 { CC_SMS_WR_START, 0x18 }, /* NetFn/LUN */ 340 { CC_SMS_WR_END, 0x22 } /* Cmd RESET_WATCHDOG_TIMER */ 341 }; 342 343 static struct { 344 int port; 345 uchar_t data; 346 } sitka_bmc[] = { 347 { SMS_COMMAND_REGISTER, SMS_WRITE_START }, 348 { SMS_DATA_REGISTER, 0x18 }, /* NetFn/LUN */ 349 { SMS_DATA_REGISTER, 0x24 }, /* Cmd SET_WATCHDOG_TIMER */ 350 { SMS_DATA_REGISTER, 0x84 }, /* DataByte 1: SMS/OS no log */ 351 { SMS_DATA_REGISTER, 0x2 }, /* DataByte 2: Power Down */ 352 { SMS_DATA_REGISTER, 0x0 }, /* DataByte 3: no pre-timeout */ 353 { SMS_DATA_REGISTER, 0x0 }, /* DataByte 4: timer expir. */ 354 { SMS_DATA_REGISTER, 0xa }, /* DataByte 5: init countdown */ 355 { SMS_COMMAND_REGISTER, SMS_WRITE_END }, 356 { SMS_DATA_REGISTER, 0x0 }, /* DataByte 6: init countdown */ 357 358 { SMS_COMMAND_REGISTER, SMS_WRITE_START }, 359 { SMS_DATA_REGISTER, 0x18 }, /* NetFn/LUN */ 360 { SMS_COMMAND_REGISTER, SMS_WRITE_END }, 361 { SMS_DATA_REGISTER, 0x22 } /* Cmd RESET_WATCHDOG_TIMER */ 362 }; 363 364 /* Patchable global variables. */ 365 int apic_kmdb_on_nmi = 0; /* 0 - no, 1 - yes enter kmdb */ 366 uint32_t apic_divide_reg_init = 0; /* 0 - divide by 2 */ 367 368 /* default apic ops without interrupt remapping */ 369 static apic_intrr_ops_t apic_nointrr_ops = { 370 (int (*)(int))return_instr, 371 (void (*)(void))return_instr, 372 (void (*)(apic_irq_t *))return_instr, 373 (void (*)(apic_irq_t *, void *))return_instr, 374 (void (*)(apic_irq_t *))return_instr, 375 apic_record_ioapic_rdt, 376 apic_record_msi, 377 }; 378 379 apic_intrr_ops_t *apic_vt_ops = &apic_nointrr_ops; 380 381 /* 382 * This is the loadable module wrapper 383 */ 384 385 int 386 _init(void) 387 { 388 if (apic_coarse_hrtime) 389 apic_ops.psm_gethrtime = &apic_gettime; 390 return (psm_mod_init(&apic_hdlp, &apic_psm_info)); 391 } 392 393 int 394 _fini(void) 395 { 396 return (psm_mod_fini(&apic_hdlp, &apic_psm_info)); 397 } 398 399 int 400 _info(struct modinfo *modinfop) 401 { 402 return (psm_mod_info(&apic_hdlp, &apic_psm_info, modinfop)); 403 } 404 405 406 static int 407 apic_probe() 408 { 409 return (apic_probe_common(apic_psm_info.p_mach_idstring)); 410 } 411 412 void 413 apic_init() 414 { 415 int i; 416 int j = 1; 417 418 apic_ipltopri[0] = APIC_VECTOR_PER_IPL; /* leave 0 for idle */ 419 for (i = 0; i < (APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL); i++) { 420 if ((i < ((APIC_AVAIL_VECTOR / APIC_VECTOR_PER_IPL) - 1)) && 421 (apic_vectortoipl[i + 1] == apic_vectortoipl[i])) 422 /* get to highest vector at the same ipl */ 423 continue; 424 for (; j <= apic_vectortoipl[i]; j++) { 425 apic_ipltopri[j] = (i << APIC_IPL_SHIFT) + 426 APIC_BASE_VECT; 427 } 428 } 429 for (; j < MAXIPL + 1; j++) 430 /* fill up any empty ipltopri slots */ 431 apic_ipltopri[j] = (i << APIC_IPL_SHIFT) + APIC_BASE_VECT; 432 apic_init_common(); 433 #if defined(__amd64) 434 /* 435 * Make cpu-specific interrupt info point to cr8pri vector 436 */ 437 for (i = 0; i <= MAXIPL; i++) 438 apic_cr8pri[i] = apic_ipltopri[i] >> APIC_IPL_SHIFT; 439 CPU->cpu_pri_data = apic_cr8pri; 440 #else 441 if (cpuid_have_cr8access(CPU)) 442 apic_have_32bit_cr8 = 1; 443 #endif /* __amd64 */ 444 } 445 446 /* 447 * handler for APIC Error interrupt. Just print a warning and continue 448 */ 449 static int 450 apic_error_intr() 451 { 452 uint_t error0, error1, error; 453 uint_t i; 454 455 /* 456 * We need to write before read as per 7.4.17 of system prog manual. 457 * We do both and or the results to be safe 458 */ 459 error0 = apic_reg_ops->apic_read(APIC_ERROR_STATUS); 460 apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0); 461 error1 = apic_reg_ops->apic_read(APIC_ERROR_STATUS); 462 error = error0 | error1; 463 464 /* 465 * Clear the APIC error status (do this on all cpus that enter here) 466 * (two writes are required due to the semantics of accessing the 467 * error status register.) 468 */ 469 apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0); 470 apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0); 471 472 /* 473 * Prevent more than 1 CPU from handling error interrupt causing 474 * double printing (interleave of characters from multiple 475 * CPU's when using prom_printf) 476 */ 477 if (lock_try(&apic_error_lock) == 0) 478 return (error ? DDI_INTR_CLAIMED : DDI_INTR_UNCLAIMED); 479 if (error) { 480 #if DEBUG 481 if (apic_debug) 482 debug_enter("pcplusmp: APIC Error interrupt received"); 483 #endif /* DEBUG */ 484 if (apic_panic_on_apic_error) 485 cmn_err(CE_PANIC, 486 "APIC Error interrupt on CPU %d. Status = %x\n", 487 psm_get_cpu_id(), error); 488 else { 489 if ((error & ~APIC_CS_ERRORS) == 0) { 490 /* cksum error only */ 491 apic_error |= APIC_ERR_APIC_ERROR; 492 apic_apic_error |= error; 493 apic_num_apic_errors++; 494 apic_num_cksum_errors++; 495 } else { 496 /* 497 * prom_printf is the best shot we have of 498 * something which is problem free from 499 * high level/NMI type of interrupts 500 */ 501 prom_printf("APIC Error interrupt on CPU %d. " 502 "Status 0 = %x, Status 1 = %x\n", 503 psm_get_cpu_id(), error0, error1); 504 apic_error |= APIC_ERR_APIC_ERROR; 505 apic_apic_error |= error; 506 apic_num_apic_errors++; 507 for (i = 0; i < apic_error_display_delay; i++) { 508 tenmicrosec(); 509 } 510 /* 511 * provide more delay next time limited to 512 * roughly 1 clock tick time 513 */ 514 if (apic_error_display_delay < 500) 515 apic_error_display_delay *= 2; 516 } 517 } 518 lock_clear(&apic_error_lock); 519 return (DDI_INTR_CLAIMED); 520 } else { 521 lock_clear(&apic_error_lock); 522 return (DDI_INTR_UNCLAIMED); 523 } 524 /* NOTREACHED */ 525 } 526 527 /* 528 * Turn off the mask bit in the performance counter Local Vector Table entry. 529 */ 530 static void 531 apic_cpcovf_mask_clear(void) 532 { 533 apic_reg_ops->apic_write(APIC_PCINT_VECT, 534 (apic_reg_ops->apic_read(APIC_PCINT_VECT) & ~APIC_LVT_MASK)); 535 } 536 537 /*ARGSUSED*/ 538 static int 539 apic_cmci_enable(xc_arg_t arg1, xc_arg_t arg2, xc_arg_t arg3) 540 { 541 apic_reg_ops->apic_write(APIC_CMCI_VECT, apic_cmci_vect); 542 return (0); 543 } 544 545 /*ARGSUSED*/ 546 static int 547 apic_cmci_disable(xc_arg_t arg1, xc_arg_t arg2, xc_arg_t arg3) 548 { 549 apic_reg_ops->apic_write(APIC_CMCI_VECT, apic_cmci_vect | AV_MASK); 550 return (0); 551 } 552 553 /*ARGSUSED*/ 554 static int 555 cmci_cpu_setup(cpu_setup_t what, int cpuid, void *arg) 556 { 557 cpuset_t cpu_set; 558 559 CPUSET_ONLY(cpu_set, cpuid); 560 561 switch (what) { 562 case CPU_ON: 563 xc_call(NULL, NULL, NULL, CPUSET2BV(cpu_set), 564 (xc_func_t)apic_cmci_enable); 565 break; 566 567 case CPU_OFF: 568 xc_call(NULL, NULL, NULL, CPUSET2BV(cpu_set), 569 (xc_func_t)apic_cmci_disable); 570 break; 571 572 default: 573 break; 574 } 575 576 return (0); 577 } 578 579 static void 580 apic_init_intr() 581 { 582 processorid_t cpun = psm_get_cpu_id(); 583 uint_t nlvt; 584 uint32_t svr = AV_UNIT_ENABLE | APIC_SPUR_INTR; 585 586 apic_reg_ops->apic_write_task_reg(APIC_MASK_ALL); 587 588 if (apic_mode == LOCAL_APIC) { 589 /* 590 * We are running APIC in MMIO mode. 591 */ 592 if (apic_flat_model) { 593 apic_reg_ops->apic_write(APIC_FORMAT_REG, 594 APIC_FLAT_MODEL); 595 } else { 596 apic_reg_ops->apic_write(APIC_FORMAT_REG, 597 APIC_CLUSTER_MODEL); 598 } 599 600 apic_reg_ops->apic_write(APIC_DEST_REG, 601 AV_HIGH_ORDER >> cpun); 602 } 603 604 if (apic_direct_EOI) { 605 /* 606 * Set 12th bit in Spurious Interrupt Vector 607 * Register to support level triggered interrupt 608 * directed EOI. 609 */ 610 svr |= (0x1 << APIC_SVR); 611 } 612 613 /* need to enable APIC before unmasking NMI */ 614 apic_reg_ops->apic_write(APIC_SPUR_INT_REG, svr); 615 616 /* 617 * Presence of an invalid vector with delivery mode AV_FIXED can 618 * cause an error interrupt, even if the entry is masked...so 619 * write a valid vector to LVT entries along with the mask bit 620 */ 621 622 /* All APICs have timer and LINT0/1 */ 623 apic_reg_ops->apic_write(APIC_LOCAL_TIMER, AV_MASK|APIC_RESV_IRQ); 624 apic_reg_ops->apic_write(APIC_INT_VECT0, AV_MASK|APIC_RESV_IRQ); 625 apic_reg_ops->apic_write(APIC_INT_VECT1, AV_NMI); /* enable NMI */ 626 627 /* 628 * On integrated APICs, the number of LVT entries is 629 * 'Max LVT entry' + 1; on 82489DX's (non-integrated 630 * APICs), nlvt is "3" (LINT0, LINT1, and timer) 631 */ 632 633 if (apic_cpus[cpun].aci_local_ver < APIC_INTEGRATED_VERS) { 634 nlvt = 3; 635 } else { 636 nlvt = ((apic_reg_ops->apic_read(APIC_VERS_REG) >> 16) & 637 0xFF) + 1; 638 } 639 640 if (nlvt >= 5) { 641 /* Enable performance counter overflow interrupt */ 642 643 if ((x86_feature & X86_MSR) != X86_MSR) 644 apic_enable_cpcovf_intr = 0; 645 if (apic_enable_cpcovf_intr) { 646 if (apic_cpcovf_vect == 0) { 647 int ipl = APIC_PCINT_IPL; 648 int irq = apic_get_ipivect(ipl, -1); 649 650 ASSERT(irq != -1); 651 apic_cpcovf_vect = 652 apic_irq_table[irq]->airq_vector; 653 ASSERT(apic_cpcovf_vect); 654 (void) add_avintr(NULL, ipl, 655 (avfunc)kcpc_hw_overflow_intr, 656 "apic pcint", irq, NULL, NULL, NULL, NULL); 657 kcpc_hw_overflow_intr_installed = 1; 658 kcpc_hw_enable_cpc_intr = 659 apic_cpcovf_mask_clear; 660 } 661 apic_reg_ops->apic_write(APIC_PCINT_VECT, 662 apic_cpcovf_vect); 663 } 664 } 665 666 if (nlvt >= 6) { 667 /* Only mask TM intr if the BIOS apparently doesn't use it */ 668 669 uint32_t lvtval; 670 671 lvtval = apic_reg_ops->apic_read(APIC_THERM_VECT); 672 if (((lvtval & AV_MASK) == AV_MASK) || 673 ((lvtval & AV_DELIV_MODE) != AV_SMI)) { 674 apic_reg_ops->apic_write(APIC_THERM_VECT, 675 AV_MASK|APIC_RESV_IRQ); 676 } 677 } 678 679 /* Enable error interrupt */ 680 681 if (nlvt >= 4 && apic_enable_error_intr) { 682 if (apic_errvect == 0) { 683 int ipl = 0xf; /* get highest priority intr */ 684 int irq = apic_get_ipivect(ipl, -1); 685 686 ASSERT(irq != -1); 687 apic_errvect = apic_irq_table[irq]->airq_vector; 688 ASSERT(apic_errvect); 689 /* 690 * Not PSMI compliant, but we are going to merge 691 * with ON anyway 692 */ 693 (void) add_avintr((void *)NULL, ipl, 694 (avfunc)apic_error_intr, "apic error intr", 695 irq, NULL, NULL, NULL, NULL); 696 } 697 apic_reg_ops->apic_write(APIC_ERR_VECT, apic_errvect); 698 apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0); 699 apic_reg_ops->apic_write(APIC_ERROR_STATUS, 0); 700 } 701 702 /* Enable CMCI interrupt */ 703 if (cmi_enable_cmci) { 704 705 mutex_enter(&cmci_cpu_setup_lock); 706 if (cmci_cpu_setup_registered == 0) { 707 mutex_enter(&cpu_lock); 708 register_cpu_setup_func(cmci_cpu_setup, NULL); 709 mutex_exit(&cpu_lock); 710 cmci_cpu_setup_registered = 1; 711 } 712 mutex_exit(&cmci_cpu_setup_lock); 713 714 if (apic_cmci_vect == 0) { 715 int ipl = 0x2; 716 int irq = apic_get_ipivect(ipl, -1); 717 718 ASSERT(irq != -1); 719 apic_cmci_vect = apic_irq_table[irq]->airq_vector; 720 ASSERT(apic_cmci_vect); 721 722 (void) add_avintr(NULL, ipl, 723 (avfunc)cmi_cmci_trap, 724 "apic cmci intr", irq, NULL, NULL, NULL, NULL); 725 } 726 apic_reg_ops->apic_write(APIC_CMCI_VECT, apic_cmci_vect); 727 } 728 } 729 730 static void 731 apic_disable_local_apic() 732 { 733 apic_reg_ops->apic_write_task_reg(APIC_MASK_ALL); 734 apic_reg_ops->apic_write(APIC_LOCAL_TIMER, AV_MASK); 735 736 /* local intr reg 0 */ 737 apic_reg_ops->apic_write(APIC_INT_VECT0, AV_MASK); 738 739 /* disable NMI */ 740 apic_reg_ops->apic_write(APIC_INT_VECT1, AV_MASK); 741 742 /* and error interrupt */ 743 apic_reg_ops->apic_write(APIC_ERR_VECT, AV_MASK); 744 745 /* and perf counter intr */ 746 apic_reg_ops->apic_write(APIC_PCINT_VECT, AV_MASK); 747 748 apic_reg_ops->apic_write(APIC_SPUR_INT_REG, APIC_SPUR_INTR); 749 } 750 751 static void 752 apic_picinit(void) 753 { 754 int i, j; 755 uint_t isr; 756 uint32_t ver; 757 758 /* 759 * initialize interrupt remapping before apic 760 * hardware initialization 761 */ 762 apic_intrr_init(apic_mode); 763 764 /* 765 * On UniSys Model 6520, the BIOS leaves vector 0x20 isr 766 * bit on without clearing it with EOI. Since softint 767 * uses vector 0x20 to interrupt itself, so softint will 768 * not work on this machine. In order to fix this problem 769 * a check is made to verify all the isr bits are clear. 770 * If not, EOIs are issued to clear the bits. 771 */ 772 for (i = 7; i >= 1; i--) { 773 isr = apic_reg_ops->apic_read(APIC_ISR_REG + (i * 4)); 774 if (isr != 0) 775 for (j = 0; ((j < 32) && (isr != 0)); j++) 776 if (isr & (1 << j)) { 777 apic_reg_ops->apic_write( 778 APIC_EOI_REG, 0); 779 isr &= ~(1 << j); 780 apic_error |= APIC_ERR_BOOT_EOI; 781 } 782 } 783 784 /* set a flag so we know we have run apic_picinit() */ 785 apic_picinit_called = 1; 786 LOCK_INIT_CLEAR(&apic_gethrtime_lock); 787 LOCK_INIT_CLEAR(&apic_ioapic_lock); 788 LOCK_INIT_CLEAR(&apic_error_lock); 789 790 picsetup(); /* initialise the 8259 */ 791 792 /* add nmi handler - least priority nmi handler */ 793 LOCK_INIT_CLEAR(&apic_nmi_lock); 794 795 if (!psm_add_nmintr(0, (avfunc) apic_nmi_intr, 796 "pcplusmp NMI handler", (caddr_t)NULL)) 797 cmn_err(CE_WARN, "pcplusmp: Unable to add nmi handler"); 798 799 ver = apic_reg_ops->apic_read(APIC_VERS_REG); 800 /* 801 * In order to determine support for Directed EOI capability, 802 * we check for 24th bit in Local APIC Version Register. 803 */ 804 if (ver & (0x1 << APIC_DIRECTED_EOI)) { 805 apic_direct_EOI = 1; 806 apic_change_eoi(); 807 } 808 809 apic_init_intr(); 810 811 /* enable apic mode if imcr present */ 812 if (apic_imcrp) { 813 outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT); 814 outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_APIC); 815 } 816 817 ioapic_init_intr(IOAPIC_MASK); 818 } 819 820 821 /*ARGSUSED1*/ 822 static int 823 apic_cpu_start(processorid_t cpun, caddr_t arg) 824 { 825 int loop_count; 826 uint32_t vector; 827 uint_t cpu_id; 828 ulong_t iflag; 829 830 cpu_id = apic_cpus[cpun].aci_local_id; 831 832 apic_cmos_ssb_set = 1; 833 834 /* 835 * Interrupts on BSP cpu will be disabled during these startup 836 * steps in order to avoid unwanted side effects from 837 * executing interrupt handlers on a problematic BIOS. 838 */ 839 840 iflag = intr_clear(); 841 outb(CMOS_ADDR, SSB); 842 outb(CMOS_DATA, BIOS_SHUTDOWN); 843 844 /* 845 * According to X2APIC specification in section '2.3.5.1' of 846 * Interrupt Command Register Semantics, the semantics of 847 * programming the Interrupt Command Register to dispatch an interrupt 848 * is simplified. A single MSR write to the 64-bit ICR is required 849 * for dispatching an interrupt. Specifically, with the 64-bit MSR 850 * interface to ICR, system software is not required to check the 851 * status of the delivery status bit prior to writing to the ICR 852 * to send an IPI. With the removal of the Delivery Status bit, 853 * system software no longer has a reason to read the ICR. It remains 854 * readable only to aid in debugging. 855 */ 856 #ifdef DEBUG 857 APIC_AV_PENDING_SET(); 858 #else 859 if (apic_mode == LOCAL_APIC) { 860 APIC_AV_PENDING_SET(); 861 } 862 #endif /* DEBUG */ 863 864 /* for integrated - make sure there is one INIT IPI in buffer */ 865 /* for external - it will wake up the cpu */ 866 apic_reg_ops->apic_write_int_cmd(cpu_id, AV_ASSERT | AV_RESET); 867 868 /* If only 1 CPU is installed, PENDING bit will not go low */ 869 for (loop_count = 0x1000; loop_count; loop_count--) { 870 if (apic_mode == LOCAL_APIC && 871 apic_reg_ops->apic_read(APIC_INT_CMD1) & AV_PENDING) 872 apic_ret(); 873 else 874 break; 875 } 876 877 apic_reg_ops->apic_write_int_cmd(cpu_id, AV_DEASSERT | AV_RESET); 878 879 drv_usecwait(20000); /* 20 milli sec */ 880 881 if (apic_cpus[cpun].aci_local_ver >= APIC_INTEGRATED_VERS) { 882 /* integrated apic */ 883 884 vector = (rm_platter_pa >> MMU_PAGESHIFT) & 885 (APIC_VECTOR_MASK | APIC_IPL_MASK); 886 887 /* to offset the INIT IPI queue up in the buffer */ 888 apic_reg_ops->apic_write_int_cmd(cpu_id, vector | AV_STARTUP); 889 890 drv_usecwait(200); /* 20 micro sec */ 891 892 apic_reg_ops->apic_write_int_cmd(cpu_id, vector | AV_STARTUP); 893 894 drv_usecwait(200); /* 20 micro sec */ 895 } 896 intr_restore(iflag); 897 return (0); 898 } 899 900 901 #ifdef DEBUG 902 int apic_break_on_cpu = 9; 903 int apic_stretch_interrupts = 0; 904 int apic_stretch_ISR = 1 << 3; /* IPL of 3 matches nothing now */ 905 906 void 907 apic_break() 908 { 909 } 910 #endif /* DEBUG */ 911 912 /* 913 * platform_intr_enter 914 * 915 * Called at the beginning of the interrupt service routine to 916 * mask all level equal to and below the interrupt priority 917 * of the interrupting vector. An EOI should be given to 918 * the interrupt controller to enable other HW interrupts. 919 * 920 * Return -1 for spurious interrupts 921 * 922 */ 923 /*ARGSUSED*/ 924 static int 925 apic_intr_enter(int ipl, int *vectorp) 926 { 927 uchar_t vector; 928 int nipl; 929 int irq; 930 ulong_t iflag; 931 apic_cpus_info_t *cpu_infop; 932 933 /* 934 * The real vector delivered is (*vectorp + 0x20), but our caller 935 * subtracts 0x20 from the vector before passing it to us. 936 * (That's why APIC_BASE_VECT is 0x20.) 937 */ 938 vector = (uchar_t)*vectorp; 939 940 /* if interrupted by the clock, increment apic_nsec_since_boot */ 941 if (vector == apic_clkvect) { 942 if (!apic_oneshot) { 943 /* NOTE: this is not MT aware */ 944 apic_hrtime_stamp++; 945 apic_nsec_since_boot += apic_nsec_per_intr; 946 apic_hrtime_stamp++; 947 last_count_read = apic_hertz_count; 948 apic_redistribute_compute(); 949 } 950 951 /* We will avoid all the book keeping overhead for clock */ 952 nipl = apic_ipls[vector]; 953 954 *vectorp = apic_vector_to_irq[vector + APIC_BASE_VECT]; 955 if (apic_mode == LOCAL_APIC) { 956 #if defined(__amd64) 957 setcr8((ulong_t)(apic_ipltopri[nipl] >> 958 APIC_IPL_SHIFT)); 959 #else 960 if (apic_have_32bit_cr8) 961 setcr8((ulong_t)(apic_ipltopri[nipl] >> 962 APIC_IPL_SHIFT)); 963 else 964 LOCAL_APIC_WRITE_REG(APIC_TASK_REG, 965 (uint32_t)apic_ipltopri[nipl]); 966 #endif 967 LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0); 968 } else { 969 X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[nipl]); 970 X2APIC_WRITE(APIC_EOI_REG, 0); 971 } 972 973 return (nipl); 974 } 975 976 cpu_infop = &apic_cpus[psm_get_cpu_id()]; 977 978 if (vector == (APIC_SPUR_INTR - APIC_BASE_VECT)) { 979 cpu_infop->aci_spur_cnt++; 980 return (APIC_INT_SPURIOUS); 981 } 982 983 /* Check if the vector we got is really what we need */ 984 if (apic_revector_pending) { 985 /* 986 * Disable interrupts for the duration of 987 * the vector translation to prevent a self-race for 988 * the apic_revector_lock. This cannot be done 989 * in apic_xlate_vector because it is recursive and 990 * we want the vector translation to be atomic with 991 * respect to other (higher-priority) interrupts. 992 */ 993 iflag = intr_clear(); 994 vector = apic_xlate_vector(vector + APIC_BASE_VECT) - 995 APIC_BASE_VECT; 996 intr_restore(iflag); 997 } 998 999 nipl = apic_ipls[vector]; 1000 *vectorp = irq = apic_vector_to_irq[vector + APIC_BASE_VECT]; 1001 1002 if (apic_mode == LOCAL_APIC) { 1003 #if defined(__amd64) 1004 setcr8((ulong_t)(apic_ipltopri[nipl] >> APIC_IPL_SHIFT)); 1005 #else 1006 if (apic_have_32bit_cr8) 1007 setcr8((ulong_t)(apic_ipltopri[nipl] >> 1008 APIC_IPL_SHIFT)); 1009 else 1010 LOCAL_APIC_WRITE_REG(APIC_TASK_REG, 1011 (uint32_t)apic_ipltopri[nipl]); 1012 #endif 1013 } else { 1014 X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[nipl]); 1015 } 1016 1017 cpu_infop->aci_current[nipl] = (uchar_t)irq; 1018 cpu_infop->aci_curipl = (uchar_t)nipl; 1019 cpu_infop->aci_ISR_in_progress |= 1 << nipl; 1020 1021 /* 1022 * apic_level_intr could have been assimilated into the irq struct. 1023 * but, having it as a character array is more efficient in terms of 1024 * cache usage. So, we leave it as is. 1025 */ 1026 if (!apic_level_intr[irq]) { 1027 if (apic_mode == LOCAL_APIC) { 1028 LOCAL_APIC_WRITE_REG(APIC_EOI_REG, 0); 1029 } else { 1030 X2APIC_WRITE(APIC_EOI_REG, 0); 1031 } 1032 } 1033 1034 #ifdef DEBUG 1035 APIC_DEBUG_BUF_PUT(vector); 1036 APIC_DEBUG_BUF_PUT(irq); 1037 APIC_DEBUG_BUF_PUT(nipl); 1038 APIC_DEBUG_BUF_PUT(psm_get_cpu_id()); 1039 if ((apic_stretch_interrupts) && (apic_stretch_ISR & (1 << nipl))) 1040 drv_usecwait(apic_stretch_interrupts); 1041 1042 if (apic_break_on_cpu == psm_get_cpu_id()) 1043 apic_break(); 1044 #endif /* DEBUG */ 1045 return (nipl); 1046 } 1047 1048 /* 1049 * This macro is a common code used by MMIO local apic and X2APIC 1050 * local apic. 1051 */ 1052 #define APIC_INTR_EXIT() \ 1053 { \ 1054 cpu_infop = &apic_cpus[psm_get_cpu_id()]; \ 1055 if (apic_level_intr[irq]) \ 1056 apic_reg_ops->apic_send_eoi(irq); \ 1057 cpu_infop->aci_curipl = (uchar_t)prev_ipl; \ 1058 /* ISR above current pri could not be in progress */ \ 1059 cpu_infop->aci_ISR_in_progress &= (2 << prev_ipl) - 1; \ 1060 } 1061 1062 /* 1063 * Any changes made to this function must also change X2APIC 1064 * version of intr_exit. 1065 */ 1066 void 1067 apic_intr_exit(int prev_ipl, int irq) 1068 { 1069 apic_cpus_info_t *cpu_infop; 1070 1071 #if defined(__amd64) 1072 setcr8((ulong_t)apic_cr8pri[prev_ipl]); 1073 #else 1074 if (apic_have_32bit_cr8) 1075 setcr8((ulong_t)(apic_ipltopri[prev_ipl] >> APIC_IPL_SHIFT)); 1076 else 1077 apicadr[APIC_TASK_REG] = apic_ipltopri[prev_ipl]; 1078 #endif 1079 1080 APIC_INTR_EXIT(); 1081 } 1082 1083 /* 1084 * Same as apic_intr_exit() except it uses MSR rather than MMIO 1085 * to access local apic registers. 1086 */ 1087 void 1088 x2apic_intr_exit(int prev_ipl, int irq) 1089 { 1090 apic_cpus_info_t *cpu_infop; 1091 1092 X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[prev_ipl]); 1093 APIC_INTR_EXIT(); 1094 } 1095 1096 intr_exit_fn_t 1097 psm_intr_exit_fn(void) 1098 { 1099 if (apic_mode == LOCAL_X2APIC) 1100 return (x2apic_intr_exit); 1101 1102 return (apic_intr_exit); 1103 } 1104 1105 /* 1106 * Mask all interrupts below or equal to the given IPL. 1107 * Any changes made to this function must also change X2APIC 1108 * version of setspl. 1109 */ 1110 static void 1111 apic_setspl(int ipl) 1112 { 1113 #if defined(__amd64) 1114 setcr8((ulong_t)apic_cr8pri[ipl]); 1115 #else 1116 if (apic_have_32bit_cr8) 1117 setcr8((ulong_t)(apic_ipltopri[ipl] >> APIC_IPL_SHIFT)); 1118 else 1119 apicadr[APIC_TASK_REG] = apic_ipltopri[ipl]; 1120 #endif 1121 1122 /* interrupts at ipl above this cannot be in progress */ 1123 apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1; 1124 /* 1125 * this is a patch fix for the ALR QSMP P5 machine, so that interrupts 1126 * have enough time to come in before the priority is raised again 1127 * during the idle() loop. 1128 */ 1129 if (apic_setspl_delay) 1130 (void) apic_reg_ops->apic_get_pri(); 1131 } 1132 1133 /* 1134 * X2APIC version of setspl. 1135 * Mask all interrupts below or equal to the given IPL 1136 */ 1137 static void 1138 x2apic_setspl(int ipl) 1139 { 1140 X2APIC_WRITE(APIC_TASK_REG, apic_ipltopri[ipl]); 1141 1142 /* interrupts at ipl above this cannot be in progress */ 1143 apic_cpus[psm_get_cpu_id()].aci_ISR_in_progress &= (2 << ipl) - 1; 1144 } 1145 1146 /* 1147 * generates an interprocessor interrupt to another CPU. Any changes made to 1148 * this routine must be accompanied by similar changes to 1149 * apic_common_send_ipi(). 1150 */ 1151 static void 1152 apic_send_ipi(int cpun, int ipl) 1153 { 1154 int vector; 1155 ulong_t flag; 1156 1157 vector = apic_resv_vector[ipl]; 1158 1159 ASSERT((vector >= APIC_BASE_VECT) && (vector <= APIC_SPUR_INTR)); 1160 1161 flag = intr_clear(); 1162 1163 APIC_AV_PENDING_SET(); 1164 1165 apic_reg_ops->apic_write_int_cmd(apic_cpus[cpun].aci_local_id, 1166 vector); 1167 1168 intr_restore(flag); 1169 } 1170 1171 1172 /*ARGSUSED*/ 1173 static void 1174 apic_set_idlecpu(processorid_t cpun) 1175 { 1176 } 1177 1178 /*ARGSUSED*/ 1179 static void 1180 apic_unset_idlecpu(processorid_t cpun) 1181 { 1182 } 1183 1184 1185 void 1186 apic_ret() 1187 { 1188 } 1189 1190 /* 1191 * If apic_coarse_time == 1, then apic_gettime() is used instead of 1192 * apic_gethrtime(). This is used for performance instead of accuracy. 1193 */ 1194 1195 static hrtime_t 1196 apic_gettime() 1197 { 1198 int old_hrtime_stamp; 1199 hrtime_t temp; 1200 1201 /* 1202 * In one-shot mode, we do not keep time, so if anyone 1203 * calls psm_gettime() directly, we vector over to 1204 * gethrtime(). 1205 * one-shot mode MUST NOT be enabled if this psm is the source of 1206 * hrtime. 1207 */ 1208 1209 if (apic_oneshot) 1210 return (gethrtime()); 1211 1212 1213 gettime_again: 1214 while ((old_hrtime_stamp = apic_hrtime_stamp) & 1) 1215 apic_ret(); 1216 1217 temp = apic_nsec_since_boot; 1218 1219 if (apic_hrtime_stamp != old_hrtime_stamp) { /* got an interrupt */ 1220 goto gettime_again; 1221 } 1222 return (temp); 1223 } 1224 1225 /* 1226 * Here we return the number of nanoseconds since booting. Note every 1227 * clock interrupt increments apic_nsec_since_boot by the appropriate 1228 * amount. 1229 */ 1230 static hrtime_t 1231 apic_gethrtime() 1232 { 1233 int curr_timeval, countval, elapsed_ticks; 1234 int old_hrtime_stamp, status; 1235 hrtime_t temp; 1236 uint32_t cpun; 1237 ulong_t oflags; 1238 1239 /* 1240 * In one-shot mode, we do not keep time, so if anyone 1241 * calls psm_gethrtime() directly, we vector over to 1242 * gethrtime(). 1243 * one-shot mode MUST NOT be enabled if this psm is the source of 1244 * hrtime. 1245 */ 1246 1247 if (apic_oneshot) 1248 return (gethrtime()); 1249 1250 oflags = intr_clear(); /* prevent migration */ 1251 1252 cpun = apic_reg_ops->apic_read(APIC_LID_REG); 1253 if (apic_mode == LOCAL_APIC) 1254 cpun >>= APIC_ID_BIT_OFFSET; 1255 1256 lock_set(&apic_gethrtime_lock); 1257 1258 gethrtime_again: 1259 while ((old_hrtime_stamp = apic_hrtime_stamp) & 1) 1260 apic_ret(); 1261 1262 /* 1263 * Check to see which CPU we are on. Note the time is kept on 1264 * the local APIC of CPU 0. If on CPU 0, simply read the current 1265 * counter. If on another CPU, issue a remote read command to CPU 0. 1266 */ 1267 if (cpun == apic_cpus[0].aci_local_id) { 1268 countval = apic_reg_ops->apic_read(APIC_CURR_COUNT); 1269 } else { 1270 #ifdef DEBUG 1271 APIC_AV_PENDING_SET(); 1272 #else 1273 if (apic_mode == LOCAL_APIC) 1274 APIC_AV_PENDING_SET(); 1275 #endif /* DEBUG */ 1276 1277 apic_reg_ops->apic_write_int_cmd( 1278 apic_cpus[0].aci_local_id, APIC_CURR_ADD | AV_REMOTE); 1279 1280 while ((status = apic_reg_ops->apic_read(APIC_INT_CMD1)) 1281 & AV_READ_PENDING) { 1282 apic_ret(); 1283 } 1284 1285 if (status & AV_REMOTE_STATUS) /* 1 = valid */ 1286 countval = apic_reg_ops->apic_read(APIC_REMOTE_READ); 1287 else { /* 0 = invalid */ 1288 apic_remote_hrterr++; 1289 /* 1290 * return last hrtime right now, will need more 1291 * testing if change to retry 1292 */ 1293 temp = apic_last_hrtime; 1294 1295 lock_clear(&apic_gethrtime_lock); 1296 1297 intr_restore(oflags); 1298 1299 return (temp); 1300 } 1301 } 1302 if (countval > last_count_read) 1303 countval = 0; 1304 else 1305 last_count_read = countval; 1306 1307 elapsed_ticks = apic_hertz_count - countval; 1308 1309 curr_timeval = APIC_TICKS_TO_NSECS(elapsed_ticks); 1310 temp = apic_nsec_since_boot + curr_timeval; 1311 1312 if (apic_hrtime_stamp != old_hrtime_stamp) { /* got an interrupt */ 1313 /* we might have clobbered last_count_read. Restore it */ 1314 last_count_read = apic_hertz_count; 1315 goto gethrtime_again; 1316 } 1317 1318 if (temp < apic_last_hrtime) { 1319 /* return last hrtime if error occurs */ 1320 apic_hrtime_error++; 1321 temp = apic_last_hrtime; 1322 } 1323 else 1324 apic_last_hrtime = temp; 1325 1326 lock_clear(&apic_gethrtime_lock); 1327 intr_restore(oflags); 1328 1329 return (temp); 1330 } 1331 1332 /* apic NMI handler */ 1333 /*ARGSUSED*/ 1334 static void 1335 apic_nmi_intr(caddr_t arg, struct regs *rp) 1336 { 1337 if (apic_shutdown_processors) { 1338 apic_disable_local_apic(); 1339 return; 1340 } 1341 1342 apic_error |= APIC_ERR_NMI; 1343 1344 if (!lock_try(&apic_nmi_lock)) 1345 return; 1346 apic_num_nmis++; 1347 1348 if (apic_kmdb_on_nmi && psm_debugger()) { 1349 debug_enter("NMI received: entering kmdb\n"); 1350 } else if (apic_panic_on_nmi) { 1351 /* Keep panic from entering kmdb. */ 1352 nopanicdebug = 1; 1353 panic("NMI received\n"); 1354 } else { 1355 /* 1356 * prom_printf is the best shot we have of something which is 1357 * problem free from high level/NMI type of interrupts 1358 */ 1359 prom_printf("NMI received\n"); 1360 } 1361 1362 lock_clear(&apic_nmi_lock); 1363 } 1364 1365 /*ARGSUSED*/ 1366 static int 1367 apic_addspl(int irqno, int ipl, int min_ipl, int max_ipl) 1368 { 1369 return (apic_addspl_common(irqno, ipl, min_ipl, max_ipl)); 1370 } 1371 1372 static int 1373 apic_delspl(int irqno, int ipl, int min_ipl, int max_ipl) 1374 { 1375 return (apic_delspl_common(irqno, ipl, min_ipl, max_ipl)); 1376 } 1377 1378 static int 1379 apic_post_cpu_start() 1380 { 1381 int cpun; 1382 static int cpus_started = 1; 1383 struct psm_ops *pops = &apic_ops; 1384 1385 /* We know this CPU + BSP started successfully. */ 1386 cpus_started++; 1387 1388 /* 1389 * On BSP we would have enabled X2APIC, if supported by processor, 1390 * in acpi_probe(), but on AP we do it here. 1391 * 1392 * We enable X2APIC mode only if BSP is running in X2APIC & the 1393 * local APIC mode of the current CPU is MMIO (xAPIC). 1394 */ 1395 if (apic_mode == LOCAL_X2APIC && apic_detect_x2apic() && 1396 apic_local_mode() == LOCAL_APIC) { 1397 apic_enable_x2apic(); 1398 } 1399 1400 /* 1401 * We change psm_send_ipi and send_dirintf only if Solaris 1402 * is booted in kmdb & the current CPU is the last CPU being 1403 * brought up. We don't need to do anything if Solaris is running 1404 * in MMIO mode (xAPIC). 1405 */ 1406 if ((boothowto & RB_DEBUG) && 1407 (cpus_started == boot_ncpus || cpus_started == apic_nproc) && 1408 apic_mode == LOCAL_X2APIC) { 1409 /* 1410 * We no longer need help from apic_common_send_ipi() 1411 * since we will not start any more CPUs. 1412 * 1413 * We will need to revisit this if we start supporting 1414 * hot-plugging of CPUs. 1415 */ 1416 pops->psm_send_ipi = x2apic_send_ipi; 1417 send_dirintf = pops->psm_send_ipi; 1418 } 1419 1420 splx(ipltospl(LOCK_LEVEL)); 1421 apic_init_intr(); 1422 1423 /* 1424 * since some systems don't enable the internal cache on the non-boot 1425 * cpus, so we have to enable them here 1426 */ 1427 setcr0(getcr0() & ~(CR0_CD | CR0_NW)); 1428 1429 #ifdef DEBUG 1430 APIC_AV_PENDING_SET(); 1431 #else 1432 if (apic_mode == LOCAL_APIC) 1433 APIC_AV_PENDING_SET(); 1434 #endif /* DEBUG */ 1435 1436 /* 1437 * We may be booting, or resuming from suspend; aci_status will 1438 * be APIC_CPU_INTR_ENABLE if coming from suspend, so we add the 1439 * APIC_CPU_ONLINE flag here rather than setting aci_status completely. 1440 */ 1441 cpun = psm_get_cpu_id(); 1442 apic_cpus[cpun].aci_status |= APIC_CPU_ONLINE; 1443 1444 apic_reg_ops->apic_write(APIC_DIVIDE_REG, apic_divide_reg_init); 1445 return (PSM_SUCCESS); 1446 } 1447 1448 processorid_t 1449 apic_get_next_processorid(processorid_t cpu_id) 1450 { 1451 1452 int i; 1453 1454 if (cpu_id == -1) 1455 return ((processorid_t)0); 1456 1457 for (i = cpu_id + 1; i < NCPU; i++) { 1458 if (CPU_IN_SET(apic_cpumask, i)) 1459 return (i); 1460 } 1461 1462 return ((processorid_t)-1); 1463 } 1464 1465 1466 /* 1467 * type == -1 indicates it is an internal request. Do not change 1468 * resv_vector for these requests 1469 */ 1470 static int 1471 apic_get_ipivect(int ipl, int type) 1472 { 1473 uchar_t vector; 1474 int irq; 1475 1476 if (irq = apic_allocate_irq(APIC_VECTOR(ipl))) { 1477 if (vector = apic_allocate_vector(ipl, irq, 1)) { 1478 apic_irq_table[irq]->airq_mps_intr_index = 1479 RESERVE_INDEX; 1480 apic_irq_table[irq]->airq_vector = vector; 1481 if (type != -1) { 1482 apic_resv_vector[ipl] = vector; 1483 } 1484 return (irq); 1485 } 1486 } 1487 apic_error |= APIC_ERR_GET_IPIVECT_FAIL; 1488 return (-1); /* shouldn't happen */ 1489 } 1490 1491 static int 1492 apic_getclkirq(int ipl) 1493 { 1494 int irq; 1495 1496 if ((irq = apic_get_ipivect(ipl, -1)) == -1) 1497 return (-1); 1498 /* 1499 * Note the vector in apic_clkvect for per clock handling. 1500 */ 1501 apic_clkvect = apic_irq_table[irq]->airq_vector - APIC_BASE_VECT; 1502 APIC_VERBOSE_IOAPIC((CE_NOTE, "get_clkirq: vector = %x\n", 1503 apic_clkvect)); 1504 return (irq); 1505 } 1506 1507 1508 /* 1509 * Return the number of APIC clock ticks elapsed for 8245 to decrement 1510 * (APIC_TIME_COUNT + pit_ticks_adj) ticks. 1511 */ 1512 static uint_t 1513 apic_calibrate(volatile uint32_t *addr, uint16_t *pit_ticks_adj) 1514 { 1515 uint8_t pit_tick_lo; 1516 uint16_t pit_tick, target_pit_tick; 1517 uint32_t start_apic_tick, end_apic_tick; 1518 ulong_t iflag; 1519 uint32_t reg; 1520 1521 reg = addr + APIC_CURR_COUNT - apicadr; 1522 1523 iflag = intr_clear(); 1524 1525 do { 1526 pit_tick_lo = inb(PITCTR0_PORT); 1527 pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo; 1528 } while (pit_tick < APIC_TIME_MIN || 1529 pit_tick_lo <= APIC_LB_MIN || pit_tick_lo >= APIC_LB_MAX); 1530 1531 /* 1532 * Wait for the 8254 to decrement by 5 ticks to ensure 1533 * we didn't start in the middle of a tick. 1534 * Compare with 0x10 for the wrap around case. 1535 */ 1536 target_pit_tick = pit_tick - 5; 1537 do { 1538 pit_tick_lo = inb(PITCTR0_PORT); 1539 pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo; 1540 } while (pit_tick > target_pit_tick || pit_tick_lo < 0x10); 1541 1542 start_apic_tick = apic_reg_ops->apic_read(reg); 1543 1544 /* 1545 * Wait for the 8254 to decrement by 1546 * (APIC_TIME_COUNT + pit_ticks_adj) ticks 1547 */ 1548 target_pit_tick = pit_tick - APIC_TIME_COUNT; 1549 do { 1550 pit_tick_lo = inb(PITCTR0_PORT); 1551 pit_tick = (inb(PITCTR0_PORT) << 8) | pit_tick_lo; 1552 } while (pit_tick > target_pit_tick || pit_tick_lo < 0x10); 1553 1554 end_apic_tick = apic_reg_ops->apic_read(reg); 1555 1556 *pit_ticks_adj = target_pit_tick - pit_tick; 1557 1558 intr_restore(iflag); 1559 1560 return (start_apic_tick - end_apic_tick); 1561 } 1562 1563 /* 1564 * Initialise the APIC timer on the local APIC of CPU 0 to the desired 1565 * frequency. Note at this stage in the boot sequence, the boot processor 1566 * is the only active processor. 1567 * hertz value of 0 indicates a one-shot mode request. In this case 1568 * the function returns the resolution (in nanoseconds) for the hardware 1569 * timer interrupt. If one-shot mode capability is not available, 1570 * the return value will be 0. apic_enable_oneshot is a global switch 1571 * for disabling the functionality. 1572 * A non-zero positive value for hertz indicates a periodic mode request. 1573 * In this case the hardware will be programmed to generate clock interrupts 1574 * at hertz frequency and returns the resolution of interrupts in 1575 * nanosecond. 1576 */ 1577 1578 static int 1579 apic_clkinit(int hertz) 1580 { 1581 uint_t apic_ticks = 0; 1582 uint_t pit_ticks; 1583 int ret; 1584 uint16_t pit_ticks_adj; 1585 static int firsttime = 1; 1586 1587 if (firsttime) { 1588 /* first time calibrate on CPU0 only */ 1589 1590 apic_reg_ops->apic_write(APIC_DIVIDE_REG, apic_divide_reg_init); 1591 apic_reg_ops->apic_write(APIC_INIT_COUNT, APIC_MAXVAL); 1592 apic_ticks = apic_calibrate(apicadr, &pit_ticks_adj); 1593 1594 /* total number of PIT ticks corresponding to apic_ticks */ 1595 pit_ticks = APIC_TIME_COUNT + pit_ticks_adj; 1596 1597 /* 1598 * Determine the number of nanoseconds per APIC clock tick 1599 * and then determine how many APIC ticks to interrupt at the 1600 * desired frequency 1601 * apic_ticks / (pitticks / PIT_HZ) = apic_ticks_per_s 1602 * (apic_ticks * PIT_HZ) / pitticks = apic_ticks_per_s 1603 * apic_ticks_per_ns = (apic_ticks * PIT_HZ) / (pitticks * 10^9) 1604 * pic_ticks_per_SFns = 1605 * (SF * apic_ticks * PIT_HZ) / (pitticks * 10^9) 1606 */ 1607 apic_ticks_per_SFnsecs = 1608 ((SF * apic_ticks * PIT_HZ) / 1609 ((uint64_t)pit_ticks * NANOSEC)); 1610 1611 /* the interval timer initial count is 32 bit max */ 1612 apic_nsec_max = APIC_TICKS_TO_NSECS(APIC_MAXVAL); 1613 firsttime = 0; 1614 } 1615 1616 if (hertz != 0) { 1617 /* periodic */ 1618 apic_nsec_per_intr = NANOSEC / hertz; 1619 apic_hertz_count = APIC_NSECS_TO_TICKS(apic_nsec_per_intr); 1620 } 1621 1622 apic_int_busy_mark = (apic_int_busy_mark * 1623 apic_sample_factor_redistribution) / 100; 1624 apic_int_free_mark = (apic_int_free_mark * 1625 apic_sample_factor_redistribution) / 100; 1626 apic_diff_for_redistribution = (apic_diff_for_redistribution * 1627 apic_sample_factor_redistribution) / 100; 1628 1629 if (hertz == 0) { 1630 /* requested one_shot */ 1631 if (!tsc_gethrtime_enable || !apic_oneshot_enable) 1632 return (0); 1633 apic_oneshot = 1; 1634 ret = (int)APIC_TICKS_TO_NSECS(1); 1635 } else { 1636 /* program the local APIC to interrupt at the given frequency */ 1637 apic_reg_ops->apic_write(APIC_INIT_COUNT, apic_hertz_count); 1638 apic_reg_ops->apic_write(APIC_LOCAL_TIMER, 1639 (apic_clkvect + APIC_BASE_VECT) | AV_TIME); 1640 apic_oneshot = 0; 1641 ret = NANOSEC / hertz; 1642 } 1643 1644 return (ret); 1645 1646 } 1647 1648 /* 1649 * apic_preshutdown: 1650 * Called early in shutdown whilst we can still access filesystems to do 1651 * things like loading modules which will be required to complete shutdown 1652 * after filesystems are all unmounted. 1653 */ 1654 static void 1655 apic_preshutdown(int cmd, int fcn) 1656 { 1657 APIC_VERBOSE_POWEROFF(("apic_preshutdown(%d,%d); m=%d a=%d\n", 1658 cmd, fcn, apic_poweroff_method, apic_enable_acpi)); 1659 1660 if ((cmd != A_SHUTDOWN) || (fcn != AD_POWEROFF)) { 1661 return; 1662 } 1663 } 1664 1665 static void 1666 apic_shutdown(int cmd, int fcn) 1667 { 1668 int restarts, attempts; 1669 int i; 1670 uchar_t byte; 1671 ulong_t iflag; 1672 1673 hpet_acpi_fini(); 1674 1675 /* Send NMI to all CPUs except self to do per processor shutdown */ 1676 iflag = intr_clear(); 1677 #ifdef DEBUG 1678 APIC_AV_PENDING_SET(); 1679 #else 1680 if (apic_mode == LOCAL_APIC) 1681 APIC_AV_PENDING_SET(); 1682 #endif /* DEBUG */ 1683 apic_shutdown_processors = 1; 1684 apic_reg_ops->apic_write(APIC_INT_CMD1, 1685 AV_NMI | AV_LEVEL | AV_SH_ALL_EXCSELF); 1686 1687 /* restore cmos shutdown byte before reboot */ 1688 if (apic_cmos_ssb_set) { 1689 outb(CMOS_ADDR, SSB); 1690 outb(CMOS_DATA, 0); 1691 } 1692 1693 ioapic_disable_redirection(); 1694 1695 /* disable apic mode if imcr present */ 1696 if (apic_imcrp) { 1697 outb(APIC_IMCR_P1, (uchar_t)APIC_IMCR_SELECT); 1698 outb(APIC_IMCR_P2, (uchar_t)APIC_IMCR_PIC); 1699 } 1700 1701 apic_disable_local_apic(); 1702 1703 intr_restore(iflag); 1704 1705 /* remainder of function is for shutdown cases only */ 1706 if (cmd != A_SHUTDOWN) 1707 return; 1708 1709 /* 1710 * Switch system back into Legacy-Mode if using ACPI and 1711 * not powering-off. Some BIOSes need to remain in ACPI-mode 1712 * for power-off to succeed (Dell Dimension 4600) 1713 * Do not disable ACPI while doing fastreboot 1714 */ 1715 if (apic_enable_acpi && fcn != AD_POWEROFF && fcn != AD_FASTREBOOT) 1716 (void) AcpiDisable(); 1717 1718 if (fcn == AD_FASTREBOOT) { 1719 apic_reg_ops->apic_write(APIC_INT_CMD1, 1720 AV_ASSERT | AV_RESET | AV_SH_ALL_EXCSELF); 1721 } 1722 1723 /* remainder of function is for shutdown+poweroff case only */ 1724 if (fcn != AD_POWEROFF) 1725 return; 1726 1727 switch (apic_poweroff_method) { 1728 case APIC_POWEROFF_VIA_RTC: 1729 1730 /* select the extended NVRAM bank in the RTC */ 1731 outb(CMOS_ADDR, RTC_REGA); 1732 byte = inb(CMOS_DATA); 1733 outb(CMOS_DATA, (byte | EXT_BANK)); 1734 1735 outb(CMOS_ADDR, PFR_REG); 1736 1737 /* for Predator must toggle the PAB bit */ 1738 byte = inb(CMOS_DATA); 1739 1740 /* 1741 * clear power active bar, wakeup alarm and 1742 * kickstart 1743 */ 1744 byte &= ~(PAB_CBIT | WF_FLAG | KS_FLAG); 1745 outb(CMOS_DATA, byte); 1746 1747 /* delay before next write */ 1748 drv_usecwait(1000); 1749 1750 /* for S40 the following would suffice */ 1751 byte = inb(CMOS_DATA); 1752 1753 /* power active bar control bit */ 1754 byte |= PAB_CBIT; 1755 outb(CMOS_DATA, byte); 1756 1757 break; 1758 1759 case APIC_POWEROFF_VIA_ASPEN_BMC: 1760 restarts = 0; 1761 restart_aspen_bmc: 1762 if (++restarts == 3) 1763 break; 1764 attempts = 0; 1765 do { 1766 byte = inb(MISMIC_FLAG_REGISTER); 1767 byte &= MISMIC_BUSY_MASK; 1768 if (byte != 0) { 1769 drv_usecwait(1000); 1770 if (attempts >= 3) 1771 goto restart_aspen_bmc; 1772 ++attempts; 1773 } 1774 } while (byte != 0); 1775 outb(MISMIC_CNTL_REGISTER, CC_SMS_GET_STATUS); 1776 byte = inb(MISMIC_FLAG_REGISTER); 1777 byte |= 0x1; 1778 outb(MISMIC_FLAG_REGISTER, byte); 1779 i = 0; 1780 for (; i < (sizeof (aspen_bmc)/sizeof (aspen_bmc[0])); 1781 i++) { 1782 attempts = 0; 1783 do { 1784 byte = inb(MISMIC_FLAG_REGISTER); 1785 byte &= MISMIC_BUSY_MASK; 1786 if (byte != 0) { 1787 drv_usecwait(1000); 1788 if (attempts >= 3) 1789 goto restart_aspen_bmc; 1790 ++attempts; 1791 } 1792 } while (byte != 0); 1793 outb(MISMIC_CNTL_REGISTER, aspen_bmc[i].cntl); 1794 outb(MISMIC_DATA_REGISTER, aspen_bmc[i].data); 1795 byte = inb(MISMIC_FLAG_REGISTER); 1796 byte |= 0x1; 1797 outb(MISMIC_FLAG_REGISTER, byte); 1798 } 1799 break; 1800 1801 case APIC_POWEROFF_VIA_SITKA_BMC: 1802 restarts = 0; 1803 restart_sitka_bmc: 1804 if (++restarts == 3) 1805 break; 1806 attempts = 0; 1807 do { 1808 byte = inb(SMS_STATUS_REGISTER); 1809 byte &= SMS_STATE_MASK; 1810 if ((byte == SMS_READ_STATE) || 1811 (byte == SMS_WRITE_STATE)) { 1812 drv_usecwait(1000); 1813 if (attempts >= 3) 1814 goto restart_sitka_bmc; 1815 ++attempts; 1816 } 1817 } while ((byte == SMS_READ_STATE) || 1818 (byte == SMS_WRITE_STATE)); 1819 outb(SMS_COMMAND_REGISTER, SMS_GET_STATUS); 1820 i = 0; 1821 for (; i < (sizeof (sitka_bmc)/sizeof (sitka_bmc[0])); 1822 i++) { 1823 attempts = 0; 1824 do { 1825 byte = inb(SMS_STATUS_REGISTER); 1826 byte &= SMS_IBF_MASK; 1827 if (byte != 0) { 1828 drv_usecwait(1000); 1829 if (attempts >= 3) 1830 goto restart_sitka_bmc; 1831 ++attempts; 1832 } 1833 } while (byte != 0); 1834 outb(sitka_bmc[i].port, sitka_bmc[i].data); 1835 } 1836 break; 1837 1838 case APIC_POWEROFF_NONE: 1839 1840 /* If no APIC direct method, we will try using ACPI */ 1841 if (apic_enable_acpi) { 1842 if (acpi_poweroff() == 1) 1843 return; 1844 } else 1845 return; 1846 1847 break; 1848 } 1849 /* 1850 * Wait a limited time here for power to go off. 1851 * If the power does not go off, then there was a 1852 * problem and we should continue to the halt which 1853 * prints a message for the user to press a key to 1854 * reboot. 1855 */ 1856 drv_usecwait(7000000); /* wait seven seconds */ 1857 1858 } 1859 1860 /* 1861 * Try and disable all interrupts. We just assign interrupts to other 1862 * processors based on policy. If any were bound by user request, we 1863 * let them continue and return failure. We do not bother to check 1864 * for cache affinity while rebinding. 1865 */ 1866 1867 static int 1868 apic_disable_intr(processorid_t cpun) 1869 { 1870 int bind_cpu = 0, i, hardbound = 0; 1871 apic_irq_t *irq_ptr; 1872 ulong_t iflag; 1873 1874 iflag = intr_clear(); 1875 lock_set(&apic_ioapic_lock); 1876 1877 for (i = 0; i <= APIC_MAX_VECTOR; i++) { 1878 if (apic_reprogram_info[i].done == B_FALSE) { 1879 if (apic_reprogram_info[i].bindcpu == cpun) { 1880 /* 1881 * CPU is busy -- it's the target of 1882 * a pending reprogramming attempt 1883 */ 1884 lock_clear(&apic_ioapic_lock); 1885 intr_restore(iflag); 1886 return (PSM_FAILURE); 1887 } 1888 } 1889 } 1890 1891 apic_cpus[cpun].aci_status &= ~APIC_CPU_INTR_ENABLE; 1892 1893 apic_cpus[cpun].aci_curipl = 0; 1894 1895 i = apic_min_device_irq; 1896 for (; i <= apic_max_device_irq; i++) { 1897 /* 1898 * If there are bound interrupts on this cpu, then 1899 * rebind them to other processors. 1900 */ 1901 if ((irq_ptr = apic_irq_table[i]) != NULL) { 1902 ASSERT((irq_ptr->airq_temp_cpu == IRQ_UNBOUND) || 1903 (irq_ptr->airq_temp_cpu == IRQ_UNINIT) || 1904 ((irq_ptr->airq_temp_cpu & ~IRQ_USER_BOUND) < 1905 apic_nproc)); 1906 1907 if (irq_ptr->airq_temp_cpu == (cpun | IRQ_USER_BOUND)) { 1908 hardbound = 1; 1909 continue; 1910 } 1911 1912 if (irq_ptr->airq_temp_cpu == cpun) { 1913 do { 1914 bind_cpu = apic_next_bind_cpu++; 1915 if (bind_cpu >= apic_nproc) { 1916 apic_next_bind_cpu = 1; 1917 bind_cpu = 0; 1918 1919 } 1920 } while (apic_rebind_all(irq_ptr, bind_cpu)); 1921 } 1922 } 1923 } 1924 1925 lock_clear(&apic_ioapic_lock); 1926 intr_restore(iflag); 1927 1928 if (hardbound) { 1929 cmn_err(CE_WARN, "Could not disable interrupts on %d" 1930 "due to user bound interrupts", cpun); 1931 return (PSM_FAILURE); 1932 } 1933 else 1934 return (PSM_SUCCESS); 1935 } 1936 1937 /* 1938 * Bind interrupts to the CPU's local APIC. 1939 * Interrupts should not be bound to a CPU's local APIC until the CPU 1940 * is ready to receive interrupts. 1941 */ 1942 static void 1943 apic_enable_intr(processorid_t cpun) 1944 { 1945 int i; 1946 apic_irq_t *irq_ptr; 1947 ulong_t iflag; 1948 1949 iflag = intr_clear(); 1950 lock_set(&apic_ioapic_lock); 1951 1952 apic_cpus[cpun].aci_status |= APIC_CPU_INTR_ENABLE; 1953 1954 i = apic_min_device_irq; 1955 for (i = apic_min_device_irq; i <= apic_max_device_irq; i++) { 1956 if ((irq_ptr = apic_irq_table[i]) != NULL) { 1957 if ((irq_ptr->airq_cpu & ~IRQ_USER_BOUND) == cpun) { 1958 (void) apic_rebind_all(irq_ptr, 1959 irq_ptr->airq_cpu); 1960 } 1961 } 1962 } 1963 1964 lock_clear(&apic_ioapic_lock); 1965 intr_restore(iflag); 1966 } 1967 1968 1969 /* 1970 * This function will reprogram the timer. 1971 * 1972 * When in oneshot mode the argument is the absolute time in future to 1973 * generate the interrupt at. 1974 * 1975 * When in periodic mode, the argument is the interval at which the 1976 * interrupts should be generated. There is no need to support the periodic 1977 * mode timer change at this time. 1978 */ 1979 static void 1980 apic_timer_reprogram(hrtime_t time) 1981 { 1982 hrtime_t now; 1983 uint_t ticks; 1984 int64_t delta; 1985 1986 /* 1987 * We should be called from high PIL context (CBE_HIGH_PIL), 1988 * so kpreempt is disabled. 1989 */ 1990 1991 if (!apic_oneshot) { 1992 /* time is the interval for periodic mode */ 1993 ticks = APIC_NSECS_TO_TICKS(time); 1994 } else { 1995 /* one shot mode */ 1996 1997 now = gethrtime(); 1998 delta = time - now; 1999 2000 if (delta <= 0) { 2001 /* 2002 * requested to generate an interrupt in the past 2003 * generate an interrupt as soon as possible 2004 */ 2005 ticks = apic_min_timer_ticks; 2006 } else if (delta > apic_nsec_max) { 2007 /* 2008 * requested to generate an interrupt at a time 2009 * further than what we are capable of. Set to max 2010 * the hardware can handle 2011 */ 2012 2013 ticks = APIC_MAXVAL; 2014 #ifdef DEBUG 2015 cmn_err(CE_CONT, "apic_timer_reprogram, request at" 2016 " %lld too far in future, current time" 2017 " %lld \n", time, now); 2018 #endif 2019 } else 2020 ticks = APIC_NSECS_TO_TICKS(delta); 2021 } 2022 2023 if (ticks < apic_min_timer_ticks) 2024 ticks = apic_min_timer_ticks; 2025 2026 apic_reg_ops->apic_write(APIC_INIT_COUNT, ticks); 2027 } 2028 2029 /* 2030 * This function will enable timer interrupts. 2031 */ 2032 static void 2033 apic_timer_enable(void) 2034 { 2035 /* 2036 * We should be Called from high PIL context (CBE_HIGH_PIL), 2037 * so kpreempt is disabled. 2038 */ 2039 2040 if (!apic_oneshot) { 2041 apic_reg_ops->apic_write(APIC_LOCAL_TIMER, 2042 (apic_clkvect + APIC_BASE_VECT) | AV_TIME); 2043 } else { 2044 /* one shot */ 2045 apic_reg_ops->apic_write(APIC_LOCAL_TIMER, 2046 (apic_clkvect + APIC_BASE_VECT)); 2047 } 2048 } 2049 2050 /* 2051 * This function will disable timer interrupts. 2052 */ 2053 static void 2054 apic_timer_disable(void) 2055 { 2056 /* 2057 * We should be Called from high PIL context (CBE_HIGH_PIL), 2058 * so kpreempt is disabled. 2059 */ 2060 apic_reg_ops->apic_write(APIC_LOCAL_TIMER, 2061 (apic_clkvect + APIC_BASE_VECT) | AV_MASK); 2062 } 2063 2064 /* 2065 * Set timer far into the future and return timer 2066 * current Count in nanoseconds. 2067 */ 2068 hrtime_t 2069 apic_timer_stop_count(void) 2070 { 2071 hrtime_t ns_val; 2072 int enable_val, count_val; 2073 2074 /* 2075 * Should be called with interrupts disabled. 2076 */ 2077 ASSERT(!interrupts_enabled()); 2078 2079 enable_val = apic_reg_ops->apic_read(APIC_LOCAL_TIMER); 2080 if ((enable_val & AV_MASK) == AV_MASK) 2081 return ((hrtime_t)-1); /* timer is disabled */ 2082 2083 count_val = apic_reg_ops->apic_read(APIC_CURR_COUNT); 2084 ns_val = APIC_TICKS_TO_NSECS(count_val); 2085 2086 apic_reg_ops->apic_write(APIC_INIT_COUNT, APIC_MAXVAL); 2087 2088 return (ns_val); 2089 } 2090 2091 /* 2092 * Reprogram timer after Deep C-State. 2093 */ 2094 void 2095 apic_timer_restart(hrtime_t time) 2096 { 2097 apic_timer_reprogram(time); 2098 } 2099 2100 ddi_periodic_t apic_periodic_id; 2101 2102 /* 2103 * If this module needs a periodic handler for the interrupt distribution, it 2104 * can be added here. The argument to the periodic handler is not currently 2105 * used, but is reserved for future. 2106 */ 2107 static void 2108 apic_post_cyclic_setup(void *arg) 2109 { 2110 _NOTE(ARGUNUSED(arg)) 2111 /* cpu_lock is held */ 2112 /* set up a periodic handler for intr redistribution */ 2113 2114 /* 2115 * In peridoc mode intr redistribution processing is done in 2116 * apic_intr_enter during clk intr processing 2117 */ 2118 if (!apic_oneshot) 2119 return; 2120 /* 2121 * Register a periodical handler for the redistribution processing. 2122 * On X86, CY_LOW_LEVEL is mapped to the level 2 interrupt, so 2123 * DDI_IPL_2 should be passed to ddi_periodic_add() here. 2124 */ 2125 apic_periodic_id = ddi_periodic_add( 2126 (void (*)(void *))apic_redistribute_compute, NULL, 2127 apic_redistribute_sample_interval, DDI_IPL_2); 2128 } 2129 2130 static void 2131 apic_redistribute_compute(void) 2132 { 2133 int i, j, max_busy; 2134 2135 if (apic_enable_dynamic_migration) { 2136 if (++apic_nticks == apic_sample_factor_redistribution) { 2137 /* 2138 * Time to call apic_intr_redistribute(). 2139 * reset apic_nticks. This will cause max_busy 2140 * to be calculated below and if it is more than 2141 * apic_int_busy, we will do the whole thing 2142 */ 2143 apic_nticks = 0; 2144 } 2145 max_busy = 0; 2146 for (i = 0; i < apic_nproc; i++) { 2147 2148 /* 2149 * Check if curipl is non zero & if ISR is in 2150 * progress 2151 */ 2152 if (((j = apic_cpus[i].aci_curipl) != 0) && 2153 (apic_cpus[i].aci_ISR_in_progress & (1 << j))) { 2154 2155 int irq; 2156 apic_cpus[i].aci_busy++; 2157 irq = apic_cpus[i].aci_current[j]; 2158 apic_irq_table[irq]->airq_busy++; 2159 } 2160 2161 if (!apic_nticks && 2162 (apic_cpus[i].aci_busy > max_busy)) 2163 max_busy = apic_cpus[i].aci_busy; 2164 } 2165 if (!apic_nticks) { 2166 if (max_busy > apic_int_busy_mark) { 2167 /* 2168 * We could make the following check be 2169 * skipped > 1 in which case, we get a 2170 * redistribution at half the busy mark (due to 2171 * double interval). Need to be able to collect 2172 * more empirical data to decide if that is a 2173 * good strategy. Punt for now. 2174 */ 2175 if (apic_skipped_redistribute) { 2176 apic_cleanup_busy(); 2177 apic_skipped_redistribute = 0; 2178 } else { 2179 apic_intr_redistribute(); 2180 } 2181 } else 2182 apic_skipped_redistribute++; 2183 } 2184 } 2185 } 2186 2187 2188 /* 2189 * The following functions are in the platform specific file so that they 2190 * can be different functions depending on whether we are running on 2191 * bare metal or a hypervisor. 2192 */ 2193 2194 /* 2195 * map an apic for memory-mapped access 2196 */ 2197 uint32_t * 2198 mapin_apic(uint32_t addr, size_t len, int flags) 2199 { 2200 /*LINTED: pointer cast may result in improper alignment */ 2201 return ((uint32_t *)psm_map_phys(addr, len, flags)); 2202 } 2203 2204 uint32_t * 2205 mapin_ioapic(uint32_t addr, size_t len, int flags) 2206 { 2207 return (mapin_apic(addr, len, flags)); 2208 } 2209 2210 /* 2211 * unmap an apic 2212 */ 2213 void 2214 mapout_apic(caddr_t addr, size_t len) 2215 { 2216 psm_unmap_phys(addr, len); 2217 } 2218 2219 void 2220 mapout_ioapic(caddr_t addr, size_t len) 2221 { 2222 mapout_apic(addr, len); 2223 } 2224 2225 /* 2226 * Check to make sure there are enough irq slots 2227 */ 2228 int 2229 apic_check_free_irqs(int count) 2230 { 2231 int i, avail; 2232 2233 avail = 0; 2234 for (i = APIC_FIRST_FREE_IRQ; i < APIC_RESV_IRQ; i++) { 2235 if ((apic_irq_table[i] == NULL) || 2236 apic_irq_table[i]->airq_mps_intr_index == FREE_INDEX) { 2237 if (++avail >= count) 2238 return (PSM_SUCCESS); 2239 } 2240 } 2241 return (PSM_FAILURE); 2242 } 2243 2244 /* 2245 * This function allocates "count" MSI vector(s) for the given "dip/pri/type" 2246 */ 2247 int 2248 apic_alloc_msi_vectors(dev_info_t *dip, int inum, int count, int pri, 2249 int behavior) 2250 { 2251 int rcount, i; 2252 uchar_t start, irqno; 2253 uint32_t cpu; 2254 major_t major; 2255 apic_irq_t *irqptr; 2256 2257 DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: dip=0x%p " 2258 "inum=0x%x pri=0x%x count=0x%x behavior=%d\n", 2259 (void *)dip, inum, pri, count, behavior)); 2260 2261 if (count > 1) { 2262 if (behavior == DDI_INTR_ALLOC_STRICT && 2263 apic_multi_msi_enable == 0) 2264 return (0); 2265 if (apic_multi_msi_enable == 0) 2266 count = 1; 2267 } 2268 2269 if ((rcount = apic_navail_vector(dip, pri)) > count) 2270 rcount = count; 2271 else if (rcount == 0 || (rcount < count && 2272 behavior == DDI_INTR_ALLOC_STRICT)) 2273 return (0); 2274 2275 /* if not ISP2, then round it down */ 2276 if (!ISP2(rcount)) 2277 rcount = 1 << (highbit(rcount) - 1); 2278 2279 mutex_enter(&airq_mutex); 2280 2281 for (start = 0; rcount > 0; rcount >>= 1) { 2282 if ((start = apic_find_multi_vectors(pri, rcount)) != 0 || 2283 behavior == DDI_INTR_ALLOC_STRICT) 2284 break; 2285 } 2286 2287 if (start == 0) { 2288 /* no vector available */ 2289 mutex_exit(&airq_mutex); 2290 return (0); 2291 } 2292 2293 if (apic_check_free_irqs(rcount) == PSM_FAILURE) { 2294 /* not enough free irq slots available */ 2295 mutex_exit(&airq_mutex); 2296 return (0); 2297 } 2298 2299 major = (dip != NULL) ? ddi_driver_major(dip) : 0; 2300 for (i = 0; i < rcount; i++) { 2301 if ((irqno = apic_allocate_irq(apic_first_avail_irq)) == 2302 (uchar_t)-1) { 2303 /* 2304 * shouldn't happen because of the 2305 * apic_check_free_irqs() check earlier 2306 */ 2307 mutex_exit(&airq_mutex); 2308 DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: " 2309 "apic_allocate_irq failed\n")); 2310 return (i); 2311 } 2312 apic_max_device_irq = max(irqno, apic_max_device_irq); 2313 apic_min_device_irq = min(irqno, apic_min_device_irq); 2314 irqptr = apic_irq_table[irqno]; 2315 #ifdef DEBUG 2316 if (apic_vector_to_irq[start + i] != APIC_RESV_IRQ) 2317 DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: " 2318 "apic_vector_to_irq is not APIC_RESV_IRQ\n")); 2319 #endif 2320 apic_vector_to_irq[start + i] = (uchar_t)irqno; 2321 2322 irqptr->airq_vector = (uchar_t)(start + i); 2323 irqptr->airq_ioapicindex = (uchar_t)inum; /* start */ 2324 irqptr->airq_intin_no = (uchar_t)rcount; 2325 irqptr->airq_ipl = pri; 2326 irqptr->airq_vector = start + i; 2327 irqptr->airq_origirq = (uchar_t)(inum + i); 2328 irqptr->airq_share_id = 0; 2329 irqptr->airq_mps_intr_index = MSI_INDEX; 2330 irqptr->airq_dip = dip; 2331 irqptr->airq_major = major; 2332 if (i == 0) /* they all bound to the same cpu */ 2333 cpu = irqptr->airq_cpu = apic_bind_intr(dip, irqno, 2334 0xff, 0xff); 2335 else 2336 irqptr->airq_cpu = cpu; 2337 DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msi_vectors: irq=0x%x " 2338 "dip=0x%p vector=0x%x origirq=0x%x pri=0x%x\n", irqno, 2339 (void *)irqptr->airq_dip, irqptr->airq_vector, 2340 irqptr->airq_origirq, pri)); 2341 } 2342 mutex_exit(&airq_mutex); 2343 return (rcount); 2344 } 2345 2346 /* 2347 * This function allocates "count" MSI-X vector(s) for the given "dip/pri/type" 2348 */ 2349 int 2350 apic_alloc_msix_vectors(dev_info_t *dip, int inum, int count, int pri, 2351 int behavior) 2352 { 2353 int rcount, i; 2354 major_t major; 2355 2356 mutex_enter(&airq_mutex); 2357 2358 if ((rcount = apic_navail_vector(dip, pri)) > count) 2359 rcount = count; 2360 else if (rcount == 0 || (rcount < count && 2361 behavior == DDI_INTR_ALLOC_STRICT)) { 2362 rcount = 0; 2363 goto out; 2364 } 2365 2366 if (apic_check_free_irqs(rcount) == PSM_FAILURE) { 2367 /* not enough free irq slots available */ 2368 rcount = 0; 2369 goto out; 2370 } 2371 2372 major = (dip != NULL) ? ddi_driver_major(dip) : 0; 2373 for (i = 0; i < rcount; i++) { 2374 uchar_t vector, irqno; 2375 apic_irq_t *irqptr; 2376 2377 if ((irqno = apic_allocate_irq(apic_first_avail_irq)) == 2378 (uchar_t)-1) { 2379 /* 2380 * shouldn't happen because of the 2381 * apic_check_free_irqs() check earlier 2382 */ 2383 DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msix_vectors: " 2384 "apic_allocate_irq failed\n")); 2385 rcount = i; 2386 goto out; 2387 } 2388 if ((vector = apic_allocate_vector(pri, irqno, 1)) == 0) { 2389 /* 2390 * shouldn't happen because of the 2391 * apic_navail_vector() call earlier 2392 */ 2393 DDI_INTR_IMPLDBG((CE_CONT, "apic_alloc_msix_vectors: " 2394 "apic_allocate_vector failed\n")); 2395 rcount = i; 2396 goto out; 2397 } 2398 apic_max_device_irq = max(irqno, apic_max_device_irq); 2399 apic_min_device_irq = min(irqno, apic_min_device_irq); 2400 irqptr = apic_irq_table[irqno]; 2401 irqptr->airq_vector = (uchar_t)vector; 2402 irqptr->airq_ipl = pri; 2403 irqptr->airq_origirq = (uchar_t)(inum + i); 2404 irqptr->airq_share_id = 0; 2405 irqptr->airq_mps_intr_index = MSIX_INDEX; 2406 irqptr->airq_dip = dip; 2407 irqptr->airq_major = major; 2408 irqptr->airq_cpu = apic_bind_intr(dip, irqno, 0xff, 0xff); 2409 } 2410 out: 2411 mutex_exit(&airq_mutex); 2412 return (rcount); 2413 } 2414 2415 /* 2416 * Allocate a free vector for irq at ipl. Takes care of merging of multiple 2417 * IPLs into a single APIC level as well as stretching some IPLs onto multiple 2418 * levels. APIC_HI_PRI_VECTS interrupts are reserved for high priority 2419 * requests and allocated only when pri is set. 2420 */ 2421 uchar_t 2422 apic_allocate_vector(int ipl, int irq, int pri) 2423 { 2424 int lowest, highest, i; 2425 2426 highest = apic_ipltopri[ipl] + APIC_VECTOR_MASK; 2427 lowest = apic_ipltopri[ipl - 1] + APIC_VECTOR_PER_IPL; 2428 2429 if (highest < lowest) /* Both ipl and ipl - 1 map to same pri */ 2430 lowest -= APIC_VECTOR_PER_IPL; 2431 2432 #ifdef DEBUG 2433 if (apic_restrict_vector) /* for testing shared interrupt logic */ 2434 highest = lowest + apic_restrict_vector + APIC_HI_PRI_VECTS; 2435 #endif /* DEBUG */ 2436 if (pri == 0) 2437 highest -= APIC_HI_PRI_VECTS; 2438 2439 for (i = lowest; i < highest; i++) { 2440 if (APIC_CHECK_RESERVE_VECTORS(i)) 2441 continue; 2442 if (apic_vector_to_irq[i] == APIC_RESV_IRQ) { 2443 apic_vector_to_irq[i] = (uchar_t)irq; 2444 return (i); 2445 } 2446 } 2447 2448 return (0); 2449 } 2450 2451 /* Mark vector as not being used by any irq */ 2452 void 2453 apic_free_vector(uchar_t vector) 2454 { 2455 apic_vector_to_irq[vector] = APIC_RESV_IRQ; 2456 } 2457 2458 uint32_t 2459 ioapic_read(int ioapic_ix, uint32_t reg) 2460 { 2461 volatile uint32_t *ioapic; 2462 2463 ioapic = apicioadr[ioapic_ix]; 2464 ioapic[APIC_IO_REG] = reg; 2465 return (ioapic[APIC_IO_DATA]); 2466 } 2467 2468 void 2469 ioapic_write(int ioapic_ix, uint32_t reg, uint32_t value) 2470 { 2471 volatile uint32_t *ioapic; 2472 2473 ioapic = apicioadr[ioapic_ix]; 2474 ioapic[APIC_IO_REG] = reg; 2475 ioapic[APIC_IO_DATA] = value; 2476 } 2477 2478 void 2479 ioapic_write_eoi(int ioapic_ix, uint32_t value) 2480 { 2481 volatile uint32_t *ioapic; 2482 2483 ioapic = apicioadr[ioapic_ix]; 2484 ioapic[APIC_IO_EOI] = value; 2485 } 2486 2487 static processorid_t 2488 apic_find_cpu(int flag) 2489 { 2490 processorid_t acid = 0; 2491 int i; 2492 2493 /* Find the first CPU with the passed-in flag set */ 2494 for (i = 0; i < apic_nproc; i++) { 2495 if (apic_cpus[i].aci_status & flag) { 2496 acid = i; 2497 break; 2498 } 2499 } 2500 2501 ASSERT((apic_cpus[acid].aci_status & flag) != 0); 2502 return (acid); 2503 } 2504 2505 /* 2506 * Call rebind to do the actual programming. 2507 * Must be called with interrupts disabled and apic_ioapic_lock held 2508 * 'p' is polymorphic -- if this function is called to process a deferred 2509 * reprogramming, p is of type 'struct ioapic_reprogram_data *', from which 2510 * the irq pointer is retrieved. If not doing deferred reprogramming, 2511 * p is of the type 'apic_irq_t *'. 2512 * 2513 * apic_ioapic_lock must be held across this call, as it protects apic_rebind 2514 * and it protects apic_find_cpu() from a race in which a CPU can be taken 2515 * offline after a cpu is selected, but before apic_rebind is called to 2516 * bind interrupts to it. 2517 */ 2518 int 2519 apic_setup_io_intr(void *p, int irq, boolean_t deferred) 2520 { 2521 apic_irq_t *irqptr; 2522 struct ioapic_reprogram_data *drep = NULL; 2523 int rv; 2524 2525 if (deferred) { 2526 drep = (struct ioapic_reprogram_data *)p; 2527 ASSERT(drep != NULL); 2528 irqptr = drep->irqp; 2529 } else 2530 irqptr = (apic_irq_t *)p; 2531 2532 ASSERT(irqptr != NULL); 2533 2534 rv = apic_rebind(irqptr, apic_irq_table[irq]->airq_cpu, drep); 2535 if (rv) { 2536 /* 2537 * CPU is not up or interrupts are disabled. Fall back to 2538 * the first available CPU 2539 */ 2540 rv = apic_rebind(irqptr, apic_find_cpu(APIC_CPU_INTR_ENABLE), 2541 drep); 2542 } 2543 2544 return (rv); 2545 } 2546 2547 2548 uchar_t 2549 apic_modify_vector(uchar_t vector, int irq) 2550 { 2551 apic_vector_to_irq[vector] = (uchar_t)irq; 2552 return (vector); 2553 } 2554 2555 char * 2556 apic_get_apic_type() 2557 { 2558 return (apic_psm_info.p_mach_idstring); 2559 } 2560 2561 void 2562 x2apic_update_psm() 2563 { 2564 struct psm_ops *pops = &apic_ops; 2565 2566 ASSERT(pops != NULL); 2567 2568 /* 2569 * We don't need to do any magic if one of the following 2570 * conditions is true : 2571 * - Not being run under kernel debugger. 2572 * - MP is not set. 2573 * - Booted with one CPU only. 2574 * - One CPU configured. 2575 * 2576 * We set apic_common_send_ipi() since kernel debuggers 2577 * attempt to send IPIs to other slave CPUs during 2578 * entry (exit) from (to) debugger. 2579 */ 2580 if (!(boothowto & RB_DEBUG) || use_mp == 0 || 2581 apic_nproc == 1 || boot_ncpus == 1) { 2582 pops->psm_send_ipi = x2apic_send_ipi; 2583 } else { 2584 pops->psm_send_ipi = apic_common_send_ipi; 2585 } 2586 2587 pops->psm_intr_exit = x2apic_intr_exit; 2588 pops->psm_setspl = x2apic_setspl; 2589 2590 send_dirintf = pops->psm_send_ipi; 2591 2592 apic_mode = LOCAL_X2APIC; 2593 apic_change_ops(); 2594 } 2595 2596 static void 2597 apic_intrr_init(int apic_mode) 2598 { 2599 if (psm_vt_ops != NULL) { 2600 if (((apic_intrr_ops_t *)psm_vt_ops)->apic_intrr_init(apic_mode) 2601 == DDI_SUCCESS) { 2602 apic_vt_ops = psm_vt_ops; 2603 apic_vt_ops->apic_intrr_enable(); 2604 } 2605 } 2606 } 2607 2608 /*ARGSUSED*/ 2609 static void 2610 apic_record_ioapic_rdt(apic_irq_t *irq_ptr, ioapic_rdt_t *irdt) 2611 { 2612 irdt->ir_hi <<= APIC_ID_BIT_OFFSET; 2613 } 2614 2615 /*ARGSUSED*/ 2616 static void 2617 apic_record_msi(apic_irq_t *irq_ptr, msi_regs_t *mregs) 2618 { 2619 mregs->mr_addr = MSI_ADDR_HDR | 2620 (MSI_ADDR_RH_FIXED << MSI_ADDR_RH_SHIFT) | 2621 (MSI_ADDR_DM_PHYSICAL << MSI_ADDR_DM_SHIFT) | 2622 (mregs->mr_addr << MSI_ADDR_DEST_SHIFT); 2623 mregs->mr_data = (MSI_DATA_TM_EDGE << MSI_DATA_TM_SHIFT) | 2624 mregs->mr_data; 2625 } 2626