1 /* 2 * Copyright (c) 2004 Poul-Henning Kamp 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 * $FreeBSD: head/sys/kern/subr_unit.c 255057 2013-08-30 07:37:45Z kib $ 27 */ 28 /* 29 * This file and its contents are supplied under the terms of the 30 * Common Development and Distribution License ("CDDL"), version 1.0. 31 * You may only use this file in accordance with the terms of version 32 * 1.0 of the CDDL. 33 * 34 * A full copy of the text of the CDDL should have accompanied this 35 * source. A copy of the CDDL is also available via the Internet at 36 * http://www.illumos.org/license/CDDL. 37 * 38 * Copyright 2014 Pluribus Networks Inc. 39 * Copyright 2019 Joyent, Inc. 40 * Copyright 2020 Oxide Computer Company 41 */ 42 43 #include <sys/types.h> 44 #include <sys/archsystm.h> 45 #include <sys/cpuset.h> 46 #include <sys/fp.h> 47 #include <sys/kmem.h> 48 #include <sys/queue.h> 49 #include <sys/spl.h> 50 #include <sys/systm.h> 51 #include <sys/ddidmareq.h> 52 #include <sys/id_space.h> 53 #include <sys/psm_defs.h> 54 #include <sys/smp_impldefs.h> 55 #include <sys/modhash.h> 56 #include <sys/hma.h> 57 58 #include <sys/x86_archext.h> 59 60 #include <machine/cpufunc.h> 61 #include <machine/md_var.h> 62 #include <machine/specialreg.h> 63 #include <machine/vmm.h> 64 #include <machine/vmparam.h> 65 #include <sys/vmm_impl.h> 66 #include <sys/kernel.h> 67 68 #include <vm/as.h> 69 #include <vm/seg_kmem.h> 70 71 SET_DECLARE(sysinit_set, struct sysinit); 72 73 void 74 sysinit(void) 75 { 76 struct sysinit **si; 77 78 SET_FOREACH(si, sysinit_set) 79 (*si)->func((*si)->data); 80 } 81 82 uint8_t const bin2bcd_data[] = { 83 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 84 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 85 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 86 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 87 0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, 88 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, 89 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, 90 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 91 0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89, 92 0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98, 0x99 93 }; 94 95 void 96 invalidate_cache_all(void) 97 { 98 cpuset_t cpuset; 99 100 kpreempt_disable(); 101 cpuset_all_but(&cpuset, CPU->cpu_id); 102 xc_call((xc_arg_t)NULL, (xc_arg_t)NULL, (xc_arg_t)NULL, 103 CPUSET2BV(cpuset), (xc_func_t)invalidate_cache); 104 invalidate_cache(); 105 kpreempt_enable(); 106 } 107 108 vm_paddr_t 109 vtophys(void *va) 110 { 111 pfn_t pfn; 112 113 /* 114 * Since hat_getpfnum() may block on an htable mutex, this is not at 115 * all safe to run from a critical_enter/kpreempt_disable context. 116 * The FreeBSD analog does not have the same locking constraints, so 117 * close attention must be paid wherever this is called. 118 */ 119 ASSERT(curthread->t_preempt == 0); 120 121 pfn = hat_getpfnum(kas.a_hat, (caddr_t)va); 122 ASSERT(pfn != PFN_INVALID); 123 return (pfn << PAGE_SHIFT) | ((uintptr_t)va & PAGE_MASK); 124 } 125 126 int 127 cpusetobj_ffs(const cpuset_t *set) 128 { 129 uint_t large, small; 130 131 /* 132 * Rather than reaching into the cpuset_t ourselves, leave that task to 133 * cpuset_bounds(). The simplicity is worth the extra wasted work to 134 * find the upper bound. 135 */ 136 cpuset_bounds(set, &small, &large); 137 138 if (small == CPUSET_NOTINSET) { 139 /* The FreeBSD version returns 0 if it find nothing */ 140 return (0); 141 } 142 143 ASSERT3U(small, <=, INT_MAX); 144 145 /* Least significant bit index starts at 1 for valid results */ 146 return (small + 1); 147 } 148 149 struct vmm_ptp_item { 150 void *vpi_vaddr; 151 }; 152 static kmutex_t vmm_ptp_lock; 153 154 static mod_hash_t *vmm_ptp_hash; 155 uint_t vmm_ptp_hash_nchains = 16381; 156 uint_t vmm_ptp_hash_size = PAGESIZE; 157 158 static void 159 vmm_ptp_hash_valdtor(mod_hash_val_t val) 160 { 161 struct vmm_ptp_item *i = (struct vmm_ptp_item *)val; 162 163 kmem_free(i->vpi_vaddr, PAGE_SIZE); 164 kmem_free(i, sizeof (*i)); 165 } 166 167 static void 168 vmm_ptp_init(void) 169 { 170 vmm_ptp_hash = mod_hash_create_ptrhash("vmm_ptp_hash", 171 vmm_ptp_hash_nchains, vmm_ptp_hash_valdtor, vmm_ptp_hash_size); 172 173 VERIFY(vmm_ptp_hash != NULL); 174 } 175 176 static uint_t 177 vmm_ptp_check(mod_hash_key_t key, mod_hash_val_t *val, void *unused) 178 { 179 struct vmm_ptp_item *i = (struct vmm_ptp_item *)val; 180 181 cmn_err(CE_PANIC, "!vmm_ptp_check: hash not empty: %p", i->vpi_vaddr); 182 183 return (MH_WALK_TERMINATE); 184 } 185 186 static void 187 vmm_ptp_cleanup(void) 188 { 189 mod_hash_walk(vmm_ptp_hash, vmm_ptp_check, NULL); 190 mod_hash_destroy_ptrhash(vmm_ptp_hash); 191 } 192 193 /* 194 * The logic in VT-d uses both kernel-virtual and direct-mapped addresses when 195 * freeing PTP pages. Until the consuming code is improved to better track the 196 * pages it allocates, we keep the kernel-virtual addresses to those pages in a 197 * hash table for when they are freed. 198 */ 199 void * 200 vmm_ptp_alloc(void) 201 { 202 void *p; 203 struct vmm_ptp_item *i; 204 205 p = kmem_zalloc(PAGE_SIZE, KM_SLEEP); 206 i = kmem_alloc(sizeof (struct vmm_ptp_item), KM_SLEEP); 207 i->vpi_vaddr = p; 208 209 mutex_enter(&vmm_ptp_lock); 210 VERIFY(mod_hash_insert(vmm_ptp_hash, 211 (mod_hash_key_t)PHYS_TO_DMAP(vtophys(p)), (mod_hash_val_t)i) == 0); 212 mutex_exit(&vmm_ptp_lock); 213 214 return (p); 215 } 216 217 void 218 vmm_ptp_free(void *addr) 219 { 220 mutex_enter(&vmm_ptp_lock); 221 VERIFY(mod_hash_destroy(vmm_ptp_hash, 222 (mod_hash_key_t)PHYS_TO_DMAP(vtophys(addr))) == 0); 223 mutex_exit(&vmm_ptp_lock); 224 } 225 226 /* Reach into i86pc/os/ddi_impl.c for these */ 227 extern void *contig_alloc(size_t, ddi_dma_attr_t *, uintptr_t, int); 228 extern void contig_free(void *, size_t); 229 230 void * 231 vmm_contig_alloc(size_t size) 232 { 233 ddi_dma_attr_t attr = { 234 /* Using fastboot_dma_attr as a guide... */ 235 .dma_attr_version = DMA_ATTR_V0, 236 .dma_attr_addr_lo = 0, 237 .dma_attr_addr_hi = ~0UL, 238 .dma_attr_count_max = 0x00000000FFFFFFFFULL, 239 .dma_attr_align = PAGE_SIZE, 240 .dma_attr_burstsizes = 1, 241 .dma_attr_minxfer = 1, 242 .dma_attr_maxxfer = 0x00000000FFFFFFFFULL, 243 .dma_attr_seg = 0x00000000FFFFFFFFULL, /* any */ 244 .dma_attr_sgllen = 1, 245 .dma_attr_granular = PAGE_SIZE, 246 .dma_attr_flags = 0, 247 }; 248 void *res; 249 250 res = contig_alloc(size, &attr, PAGE_SIZE, 1); 251 if (res != NULL) { 252 bzero(res, size); 253 } 254 255 return (res); 256 } 257 258 void 259 vmm_contig_free(void *addr, size_t size) 260 { 261 contig_free(addr, size); 262 } 263 264 void 265 critical_enter(void) 266 { 267 kpreempt_disable(); 268 } 269 270 void 271 critical_exit(void) 272 { 273 kpreempt_enable(); 274 } 275 276 277 static void 278 vmm_glue_callout_handler(void *arg) 279 { 280 struct callout *c = arg; 281 282 if (callout_active(c)) { 283 /* 284 * Record the handler fire time so that callout_pending() is 285 * able to detect if the callout becomes rescheduled during the 286 * course of the handler. 287 */ 288 c->c_fired = gethrtime(); 289 (c->c_func)(c->c_arg); 290 } 291 } 292 293 void 294 vmm_glue_callout_init(struct callout *c, int mpsafe) 295 { 296 cyc_handler_t hdlr; 297 cyc_time_t when; 298 299 hdlr.cyh_level = CY_LOW_LEVEL; 300 hdlr.cyh_func = vmm_glue_callout_handler; 301 hdlr.cyh_arg = c; 302 when.cyt_when = CY_INFINITY; 303 when.cyt_interval = CY_INFINITY; 304 bzero(c, sizeof (*c)); 305 306 mutex_enter(&cpu_lock); 307 c->c_cyc_id = cyclic_add(&hdlr, &when); 308 mutex_exit(&cpu_lock); 309 } 310 311 void 312 callout_reset_hrtime(struct callout *c, hrtime_t target, void (*func)(void *), 313 void *arg, int flags) 314 { 315 ASSERT(c->c_cyc_id != CYCLIC_NONE); 316 317 if ((flags & C_ABSOLUTE) == 0) { 318 target += gethrtime(); 319 } 320 321 c->c_func = func; 322 c->c_arg = arg; 323 c->c_target = target; 324 (void) cyclic_reprogram(c->c_cyc_id, target); 325 } 326 327 void 328 vmm_glue_callout_stop(struct callout *c) 329 { 330 ASSERT(c->c_cyc_id != CYCLIC_NONE); 331 332 c->c_target = 0; 333 (void) cyclic_reprogram(c->c_cyc_id, CY_INFINITY); 334 } 335 336 void 337 vmm_glue_callout_drain(struct callout *c) 338 { 339 ASSERT(c->c_cyc_id != CYCLIC_NONE); 340 341 c->c_target = 0; 342 mutex_enter(&cpu_lock); 343 cyclic_remove(c->c_cyc_id); 344 c->c_cyc_id = CYCLIC_NONE; 345 mutex_exit(&cpu_lock); 346 } 347 348 void 349 vmm_glue_callout_localize(struct callout *c) 350 { 351 mutex_enter(&cpu_lock); 352 cyclic_move_here(c->c_cyc_id); 353 mutex_exit(&cpu_lock); 354 } 355 356 /* 357 * Given an interval (in ns) and a frequency (in hz), calculate the number of 358 * "ticks" at that frequency which cover the interval. 359 */ 360 uint64_t 361 hrt_freq_count(hrtime_t interval, uint32_t freq) 362 { 363 ASSERT3S(interval, >=, 0); 364 const uint64_t sec = interval / NANOSEC; 365 const uint64_t nsec = interval % NANOSEC; 366 367 return ((sec * freq) + ((nsec * freq) / NANOSEC)); 368 } 369 370 /* 371 * Given a frequency (in hz) and number of "ticks", calculate the interval 372 * (in ns) which would be covered by those ticks. 373 */ 374 hrtime_t 375 hrt_freq_interval(uint32_t freq, uint64_t count) 376 { 377 const uint64_t sec = count / freq; 378 const uint64_t frac = count % freq; 379 380 return ((NANOSEC * sec) + ((frac * NANOSEC) / freq)); 381 } 382 383 384 uint_t cpu_high; /* Highest arg to CPUID */ 385 uint_t cpu_exthigh; /* Highest arg to extended CPUID */ 386 uint_t cpu_id; /* Stepping ID */ 387 char cpu_vendor[20]; /* CPU Origin code */ 388 389 static void 390 vmm_cpuid_init(void) 391 { 392 uint_t regs[4]; 393 394 do_cpuid(0, regs); 395 cpu_high = regs[0]; 396 ((uint_t *)&cpu_vendor)[0] = regs[1]; 397 ((uint_t *)&cpu_vendor)[1] = regs[3]; 398 ((uint_t *)&cpu_vendor)[2] = regs[2]; 399 cpu_vendor[12] = '\0'; 400 401 do_cpuid(1, regs); 402 cpu_id = regs[0]; 403 404 do_cpuid(0x80000000, regs); 405 cpu_exthigh = regs[0]; 406 } 407 408 void 409 vmm_sol_glue_init(void) 410 { 411 vmm_ptp_init(); 412 vmm_cpuid_init(); 413 } 414 415 void 416 vmm_sol_glue_cleanup(void) 417 { 418 vmm_ptp_cleanup(); 419 } 420 421 422 /* From FreeBSD's sys/kern/subr_clock.c */ 423 424 /*- 425 * Copyright (c) 1988 University of Utah. 426 * Copyright (c) 1982, 1990, 1993 427 * The Regents of the University of California. All rights reserved. 428 * 429 * This code is derived from software contributed to Berkeley by 430 * the Systems Programming Group of the University of Utah Computer 431 * Science Department. 432 * 433 * Redistribution and use in source and binary forms, with or without 434 * modification, are permitted provided that the following conditions 435 * are met: 436 * 1. Redistributions of source code must retain the above copyright 437 * notice, this list of conditions and the following disclaimer. 438 * 2. Redistributions in binary form must reproduce the above copyright 439 * notice, this list of conditions and the following disclaimer in the 440 * documentation and/or other materials provided with the distribution. 441 * 4. Neither the name of the University nor the names of its contributors 442 * may be used to endorse or promote products derived from this software 443 * without specific prior written permission. 444 * 445 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 446 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 447 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 448 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 449 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 450 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 451 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 452 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 453 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 454 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 455 * SUCH DAMAGE. 456 * 457 * from: Utah $Hdr: clock.c 1.18 91/01/21$ 458 * from: @(#)clock.c 8.2 (Berkeley) 1/12/94 459 * from: NetBSD: clock_subr.c,v 1.6 2001/07/07 17:04:02 thorpej Exp 460 * and 461 * from: src/sys/i386/isa/clock.c,v 1.176 2001/09/04 462 */ 463 464 #include <sys/clock.h> 465 466 /* 467 * Generic routines to convert between a POSIX date 468 * (seconds since 1/1/1970) and yr/mo/day/hr/min/sec 469 * Derived from NetBSD arch/hp300/hp300/clock.c 470 */ 471 472 #define FEBRUARY 2 473 #define days_in_year(y) (leapyear(y) ? 366 : 365) 474 #define days_in_month(y, m) \ 475 (month_days[(m) - 1] + (m == FEBRUARY ? leapyear(y) : 0)) 476 /* Day of week. Days are counted from 1/1/1970, which was a Thursday */ 477 #define day_of_week(days) (((days) + 4) % 7) 478 479 static const int month_days[12] = { 480 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 481 }; 482 483 484 /* 485 * This inline avoids some unnecessary modulo operations 486 * as compared with the usual macro: 487 * ( ((year % 4) == 0 && 488 * (year % 100) != 0) || 489 * ((year % 400) == 0) ) 490 * It is otherwise equivalent. 491 */ 492 static int 493 leapyear(int year) 494 { 495 int rv = 0; 496 497 if ((year & 3) == 0) { 498 rv = 1; 499 if ((year % 100) == 0) { 500 rv = 0; 501 if ((year % 400) == 0) 502 rv = 1; 503 } 504 } 505 return (rv); 506 } 507 508 int 509 clock_ct_to_ts(struct clocktime *ct, struct timespec *ts) 510 { 511 int i, year, days; 512 513 year = ct->year; 514 515 #ifdef __FreeBSD__ 516 if (ct_debug) { 517 printf("ct_to_ts("); 518 print_ct(ct); 519 printf(")"); 520 } 521 #endif 522 523 /* Sanity checks. */ 524 if (ct->mon < 1 || ct->mon > 12 || ct->day < 1 || 525 ct->day > days_in_month(year, ct->mon) || 526 ct->hour > 23 || ct->min > 59 || ct->sec > 59 || 527 (sizeof (time_t) == 4 && year > 2037)) { /* time_t overflow */ 528 #ifdef __FreeBSD__ 529 if (ct_debug) 530 printf(" = EINVAL\n"); 531 #endif 532 return (EINVAL); 533 } 534 535 /* 536 * Compute days since start of time 537 * First from years, then from months. 538 */ 539 days = 0; 540 for (i = POSIX_BASE_YEAR; i < year; i++) 541 days += days_in_year(i); 542 543 /* Months */ 544 for (i = 1; i < ct->mon; i++) 545 days += days_in_month(year, i); 546 days += (ct->day - 1); 547 548 ts->tv_sec = (((time_t)days * 24 + ct->hour) * 60 + ct->min) * 60 + 549 ct->sec; 550 ts->tv_nsec = ct->nsec; 551 552 #ifdef __FreeBSD__ 553 if (ct_debug) 554 printf(" = %ld.%09ld\n", (long)ts->tv_sec, (long)ts->tv_nsec); 555 #endif 556 return (0); 557 } 558 559 void 560 clock_ts_to_ct(struct timespec *ts, struct clocktime *ct) 561 { 562 int i, year, days; 563 time_t rsec; /* remainder seconds */ 564 time_t secs; 565 566 secs = ts->tv_sec; 567 days = secs / SECDAY; 568 rsec = secs % SECDAY; 569 570 ct->dow = day_of_week(days); 571 572 /* Subtract out whole years, counting them in i. */ 573 for (year = POSIX_BASE_YEAR; days >= days_in_year(year); year++) 574 days -= days_in_year(year); 575 ct->year = year; 576 577 /* Subtract out whole months, counting them in i. */ 578 for (i = 1; days >= days_in_month(year, i); i++) 579 days -= days_in_month(year, i); 580 ct->mon = i; 581 582 /* Days are what is left over (+1) from all that. */ 583 ct->day = days + 1; 584 585 /* Hours, minutes, seconds are easy */ 586 ct->hour = rsec / 3600; 587 rsec = rsec % 3600; 588 ct->min = rsec / 60; 589 rsec = rsec % 60; 590 ct->sec = rsec; 591 ct->nsec = ts->tv_nsec; 592 #ifdef __FreeBSD__ 593 if (ct_debug) { 594 printf("ts_to_ct(%ld.%09ld) = ", 595 (long)ts->tv_sec, (long)ts->tv_nsec); 596 print_ct(ct); 597 printf("\n"); 598 } 599 #endif 600 } 601 602 /* Equivalent to the FreeBSD rdtsc(), but with any necessary per-cpu offset */ 603 uint64_t 604 rdtsc_offset(void) 605 { 606 /* 607 * The timestamp logic will decide if a delta need be applied to the 608 * unscaled hrtime reading (effectively rdtsc), but we do require it be 609 * backed by the TSC itself. 610 */ 611 extern hrtime_t (*gethrtimeunscaledf)(void); 612 extern hrtime_t tsc_gethrtimeunscaled(void); 613 extern hrtime_t tsc_gethrtimeunscaled_delta(void); 614 615 ASSERT(*gethrtimeunscaledf == tsc_gethrtimeunscaled || 616 *gethrtimeunscaledf == tsc_gethrtimeunscaled_delta); 617 return ((uint64_t)gethrtimeunscaledf()); 618 } 619