1 /* $NetBSD: cpu.c,v 1.210 2024/04/22 23:07:47 andvar Exp $ */
2
3 /*
4 * Copyright (c) 2000-2020 NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Bill Sommerfeld of RedBack Networks Inc, and by Andrew Doran.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32 /*
33 * Copyright (c) 1999 Stefan Grefen
34 *
35 * Redistribution and use in source and binary forms, with or without
36 * modification, are permitted provided that the following conditions
37 * are met:
38 * 1. Redistributions of source code must retain the above copyright
39 * notice, this list of conditions and the following disclaimer.
40 * 2. Redistributions in binary form must reproduce the above copyright
41 * notice, this list of conditions and the following disclaimer in the
42 * documentation and/or other materials provided with the distribution.
43 * 3. All advertising materials mentioning features or use of this software
44 * must display the following acknowledgement:
45 * This product includes software developed by the NetBSD
46 * Foundation, Inc. and its contributors.
47 * 4. Neither the name of The NetBSD Foundation nor the names of its
48 * contributors may be used to endorse or promote products derived
49 * from this software without specific prior written permission.
50 *
51 * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND ANY
52 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
53 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
54 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR AND CONTRIBUTORS BE LIABLE
55 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
56 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
57 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
58 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
59 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
60 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
61 * SUCH DAMAGE.
62 */
63
64 #include <sys/cdefs.h>
65 __KERNEL_RCSID(0, "$NetBSD: cpu.c,v 1.210 2024/04/22 23:07:47 andvar Exp $");
66
67 #include "opt_ddb.h"
68 #include "opt_mpbios.h" /* for MPDEBUG */
69 #include "opt_mtrr.h"
70 #include "opt_multiprocessor.h"
71 #include "opt_svs.h"
72
73 #include "lapic.h"
74 #include "ioapic.h"
75 #include "acpica.h"
76 #include "hpet.h"
77
78 #include <sys/param.h>
79 #include <sys/proc.h>
80 #include <sys/systm.h>
81 #include <sys/device.h>
82 #include <sys/cpu.h>
83 #include <sys/cpufreq.h>
84 #include <sys/idle.h>
85 #include <sys/atomic.h>
86 #include <sys/reboot.h>
87 #include <sys/csan.h>
88
89 #include <uvm/uvm.h>
90
91 #include "acpica.h" /* for NACPICA, for mp_verbose */
92
93 #include <x86/machdep.h>
94 #include <machine/cpufunc.h>
95 #include <machine/cpuvar.h>
96 #include <machine/pmap.h>
97 #include <machine/vmparam.h>
98 #if defined(MULTIPROCESSOR)
99 #include <machine/mpbiosvar.h>
100 #endif
101 #include <machine/mpconfig.h> /* for mp_verbose */
102 #include <machine/pcb.h>
103 #include <machine/specialreg.h>
104 #include <machine/segments.h>
105 #include <machine/gdt.h>
106 #include <machine/mtrr.h>
107 #include <machine/pio.h>
108 #include <machine/cpu_counter.h>
109 #include <machine/pmap_private.h>
110
111 #include <x86/fpu.h>
112
113 #if NACPICA > 0
114 #include <dev/acpi/acpi_srat.h>
115 #endif
116
117 #if NLAPIC > 0
118 #include <machine/apicvar.h>
119 #include <machine/i82489reg.h>
120 #include <machine/i82489var.h>
121 #endif
122
123 #include <dev/ic/mc146818reg.h>
124 #include <dev/ic/hpetvar.h>
125 #include <i386/isa/nvram.h>
126 #include <dev/isa/isareg.h>
127
128 #include "tsc.h"
129
130 #ifndef XENPV
131 #include "hyperv.h"
132 #if NHYPERV > 0
133 #include <x86/x86/hypervvar.h>
134 #endif
135 #endif
136
137 #ifdef XEN
138 #include <xen/hypervisor.h>
139 #endif
140
141 static int cpu_match(device_t, cfdata_t, void *);
142 static void cpu_attach(device_t, device_t, void *);
143 static void cpu_defer(device_t);
144 static int cpu_rescan(device_t, const char *, const int *);
145 static void cpu_childdetached(device_t, device_t);
146 static bool cpu_stop(device_t);
147 static bool cpu_suspend(device_t, const pmf_qual_t *);
148 static bool cpu_resume(device_t, const pmf_qual_t *);
149 static bool cpu_shutdown(device_t, int);
150
151 struct cpu_softc {
152 device_t sc_dev; /* device tree glue */
153 struct cpu_info *sc_info; /* pointer to CPU info */
154 bool sc_wasonline;
155 };
156
157 #ifdef MULTIPROCESSOR
158 int mp_cpu_start(struct cpu_info *, paddr_t);
159 void mp_cpu_start_cleanup(struct cpu_info *);
160 const struct cpu_functions mp_cpu_funcs = { mp_cpu_start, NULL,
161 mp_cpu_start_cleanup };
162 #endif
163
164
165 CFATTACH_DECL2_NEW(cpu, sizeof(struct cpu_softc),
166 cpu_match, cpu_attach, NULL, NULL, cpu_rescan, cpu_childdetached);
167
168 /*
169 * Statically-allocated CPU info for the primary CPU (or the only
170 * CPU, on uniprocessors). The CPU info list is initialized to
171 * point at it.
172 */
173 struct cpu_info cpu_info_primary __aligned(CACHE_LINE_SIZE) = {
174 .ci_dev = 0,
175 .ci_self = &cpu_info_primary,
176 .ci_idepth = -1,
177 .ci_curlwp = &lwp0,
178 .ci_curldt = -1,
179 .ci_kfpu_spl = -1,
180 };
181
182 struct cpu_info *cpu_info_list = &cpu_info_primary;
183
184 #ifdef i386
185 void cpu_set_tss_gates(struct cpu_info *);
186 #endif
187
188 static void cpu_init_idle_lwp(struct cpu_info *);
189
190 uint32_t cpu_feature[7] __read_mostly; /* X86 CPUID feature bits */
191 /* [0] basic features cpuid.1:%edx
192 * [1] basic features cpuid.1:%ecx (CPUID2_xxx bits)
193 * [2] extended features cpuid:80000001:%edx
194 * [3] extended features cpuid:80000001:%ecx
195 * [4] VIA padlock features
196 * [5] structured extended features cpuid.7:%ebx
197 * [6] structured extended features cpuid.7:%ecx
198 */
199
200 #ifdef MULTIPROCESSOR
201 bool x86_mp_online;
202 paddr_t mp_trampoline_paddr = MP_TRAMPOLINE;
203 #endif
204 #if NLAPIC > 0
205 static vaddr_t cmos_data_mapping;
206 #endif
207 struct cpu_info *cpu_starting;
208
209 #ifdef MULTIPROCESSOR
210 void cpu_hatch(void *);
211 static void cpu_boot_secondary(struct cpu_info *ci);
212 static void cpu_start_secondary(struct cpu_info *ci);
213 #if NLAPIC > 0
214 static void cpu_copy_trampoline(paddr_t);
215 #endif
216 #endif /* MULTIPROCESSOR */
217
218 /*
219 * Runs once per boot once multiprocessor goo has been detected and
220 * the local APIC on the boot processor has been mapped.
221 *
222 * Called from lapic_boot_init() (from mpbios_scan()).
223 */
224 #if NLAPIC > 0
225 void
cpu_init_first(void)226 cpu_init_first(void)
227 {
228
229 cpu_info_primary.ci_cpuid = lapic_cpu_number();
230
231 cmos_data_mapping = uvm_km_alloc(kernel_map, PAGE_SIZE, 0, UVM_KMF_VAONLY);
232 if (cmos_data_mapping == 0)
233 panic("No KVA for page 0");
234 pmap_kenter_pa(cmos_data_mapping, 0, VM_PROT_READ|VM_PROT_WRITE, 0);
235 pmap_update(pmap_kernel());
236 }
237 #endif
238
239 static int
cpu_match(device_t parent,cfdata_t match,void * aux)240 cpu_match(device_t parent, cfdata_t match, void *aux)
241 {
242
243 return 1;
244 }
245
246 #ifdef __HAVE_PCPU_AREA
247 void
cpu_pcpuarea_init(struct cpu_info * ci)248 cpu_pcpuarea_init(struct cpu_info *ci)
249 {
250 struct vm_page *pg;
251 size_t i, npages;
252 vaddr_t base, va;
253 paddr_t pa;
254
255 CTASSERT(sizeof(struct pcpu_entry) % PAGE_SIZE == 0);
256
257 npages = sizeof(struct pcpu_entry) / PAGE_SIZE;
258 base = (vaddr_t)&pcpuarea->ent[cpu_index(ci)];
259
260 for (i = 0; i < npages; i++) {
261 pg = uvm_pagealloc(NULL, 0, NULL, UVM_PGA_ZERO);
262 if (pg == NULL) {
263 panic("failed to allocate pcpu PA");
264 }
265
266 va = base + i * PAGE_SIZE;
267 pa = VM_PAGE_TO_PHYS(pg);
268
269 pmap_kenter_pa(va, pa, VM_PROT_READ|VM_PROT_WRITE, 0);
270 }
271
272 pmap_update(pmap_kernel());
273 }
274 #endif
275
276 static void
cpu_vm_init(struct cpu_info * ci)277 cpu_vm_init(struct cpu_info *ci)
278 {
279 unsigned int ncolors = 2;
280
281 /*
282 * XXX: for AP's the cache info has not been initialized yet
283 * but that does not matter because uvm only pays attention at
284 * the maximum only. We should fix it once cpus have different
285 * cache sizes.
286 */
287 for (unsigned int i = CAI_ICACHE; i <= CAI_L2CACHE; i++) {
288 struct x86_cache_info *cai;
289 unsigned int tcolors;
290
291 cai = &ci->ci_cinfo[i];
292
293 tcolors = atop(cai->cai_totalsize);
294 switch (cai->cai_associativity) {
295 case 0xff:
296 tcolors = 1; /* fully associative */
297 break;
298 case 0:
299 case 1:
300 break;
301 default:
302 tcolors /= cai->cai_associativity;
303 }
304 if (tcolors <= ncolors)
305 continue;
306 ncolors = tcolors;
307 }
308
309 /*
310 * If the desired number of colors is not a power of
311 * two, it won't be good. Find the greatest power of
312 * two which is an even divisor of the number of colors,
313 * to preserve even coloring of pages.
314 */
315 if (ncolors & (ncolors - 1) ) {
316 unsigned int try, picked = 1;
317 for (try = 1; try < ncolors; try *= 2) {
318 if (ncolors % try == 0) picked = try;
319 }
320 if (picked == 1) {
321 panic("desired number of cache colors %u is "
322 " > 1, but not even!", ncolors);
323 }
324 ncolors = picked;
325 }
326
327 /*
328 * Knowing the size of the largest cache on this CPU, potentially
329 * re-color our pages.
330 */
331 aprint_debug_dev(ci->ci_dev, "%d page colors\n", ncolors);
332 uvm_page_recolor(ncolors);
333
334 pmap_tlb_cpu_init(ci);
335 #ifndef __HAVE_DIRECT_MAP
336 pmap_vpage_cpu_init(ci);
337 #endif
338 }
339
340 static void
cpu_attach(device_t parent,device_t self,void * aux)341 cpu_attach(device_t parent, device_t self, void *aux)
342 {
343 struct cpu_softc *sc = device_private(self);
344 struct cpu_attach_args *caa = aux;
345 struct cpu_info *ci;
346 uintptr_t ptr;
347 #if NLAPIC > 0
348 int cpunum = caa->cpu_number;
349 #endif
350 static bool again;
351
352 sc->sc_dev = self;
353
354 if (ncpu > maxcpus) {
355 #ifndef _LP64
356 aprint_error(": too many CPUs, please use NetBSD/amd64\n");
357 #else
358 aprint_error(": too many CPUs\n");
359 #endif
360 return;
361 }
362
363 /*
364 * If we're an Application Processor, allocate a cpu_info
365 * structure, otherwise use the primary's.
366 */
367 if (caa->cpu_role == CPU_ROLE_AP) {
368 if ((boothowto & RB_MD1) != 0) {
369 aprint_error(": multiprocessor boot disabled\n");
370 if (!pmf_device_register(self, NULL, NULL))
371 aprint_error_dev(self,
372 "couldn't establish power handler\n");
373 return;
374 }
375 aprint_naive(": Application Processor\n");
376 ptr = (uintptr_t)uvm_km_alloc(kernel_map,
377 sizeof(*ci) + CACHE_LINE_SIZE - 1, 0,
378 UVM_KMF_WIRED|UVM_KMF_ZERO);
379 ci = (struct cpu_info *)roundup2(ptr, CACHE_LINE_SIZE);
380 ci->ci_curldt = -1;
381 } else {
382 aprint_naive(": %s Processor\n",
383 caa->cpu_role == CPU_ROLE_SP ? "Single" : "Boot");
384 ci = &cpu_info_primary;
385 #if NLAPIC > 0
386 if (cpunum != lapic_cpu_number()) {
387 /* XXX should be done earlier. */
388 uint32_t reg;
389 aprint_verbose("\n");
390 aprint_verbose_dev(self, "running CPU at apic %d"
391 " instead of at expected %d", lapic_cpu_number(),
392 cpunum);
393 reg = lapic_readreg(LAPIC_ID);
394 lapic_writereg(LAPIC_ID, (reg & ~LAPIC_ID_MASK) |
395 (cpunum << LAPIC_ID_SHIFT));
396 }
397 if (cpunum != lapic_cpu_number()) {
398 aprint_error_dev(self, "unable to reset apic id\n");
399 }
400 #endif
401 }
402
403 ci->ci_self = ci;
404 sc->sc_info = ci;
405 ci->ci_dev = self;
406 ci->ci_acpiid = caa->cpu_id;
407 ci->ci_cpuid = caa->cpu_number;
408 ci->ci_func = caa->cpu_func;
409 ci->ci_kfpu_spl = -1;
410 aprint_normal("\n");
411
412 /* Must be before mi_cpu_attach(). */
413 cpu_vm_init(ci);
414
415 if (caa->cpu_role == CPU_ROLE_AP) {
416 int error;
417
418 error = mi_cpu_attach(ci);
419 if (error != 0) {
420 aprint_error_dev(self,
421 "mi_cpu_attach failed with %d\n", error);
422 return;
423 }
424 #ifdef __HAVE_PCPU_AREA
425 cpu_pcpuarea_init(ci);
426 #endif
427 cpu_init_tss(ci);
428 } else {
429 KASSERT(ci->ci_data.cpu_idlelwp != NULL);
430 #if NACPICA > 0
431 /* Parse out NUMA info for cpu_identify(). */
432 acpisrat_init();
433 #endif
434 }
435
436 #ifdef SVS
437 cpu_svs_init(ci);
438 #endif
439
440 pmap_reference(pmap_kernel());
441 ci->ci_pmap = pmap_kernel();
442 ci->ci_tlbstate = TLBSTATE_STALE;
443
444 /*
445 * Boot processor may not be attached first, but the below
446 * must be done to allow booting other processors.
447 */
448 if (!again) {
449 /* Make sure DELAY() (likely i8254_delay()) is initialized. */
450 DELAY(1);
451
452 /*
453 * Basic init. Compute an approximate frequency for the TSC
454 * using the i8254. If there's a HPET we'll redo it later.
455 */
456 atomic_or_32(&ci->ci_flags, CPUF_PRESENT | CPUF_PRIMARY);
457 cpu_intr_init(ci);
458 tsc_setfunc(ci);
459 cpu_get_tsc_freq(ci);
460 cpu_init(ci);
461 #ifdef i386
462 cpu_set_tss_gates(ci);
463 #endif
464 pmap_cpu_init_late(ci);
465 #if NLAPIC > 0
466 if (caa->cpu_role != CPU_ROLE_SP) {
467 /* Enable lapic. */
468 lapic_enable();
469 lapic_set_lvt();
470 if (!vm_guest_is_xenpvh_or_pvhvm())
471 lapic_calibrate_timer(false);
472 }
473 #endif
474 kcsan_cpu_init(ci);
475 again = true;
476 }
477
478 /* further PCB init done later. */
479
480 switch (caa->cpu_role) {
481 case CPU_ROLE_SP:
482 atomic_or_32(&ci->ci_flags, CPUF_SP);
483 cpu_identify(ci);
484 x86_errata();
485 x86_cpu_idle_init();
486 #ifdef XENPVHVM
487 xen_hvm_init_cpu(ci);
488 #endif
489 break;
490
491 case CPU_ROLE_BP:
492 atomic_or_32(&ci->ci_flags, CPUF_BSP);
493 cpu_identify(ci);
494 x86_errata();
495 x86_cpu_idle_init();
496 #ifdef XENPVHVM
497 xen_hvm_init_cpu(ci);
498 #endif
499 break;
500
501 #ifdef MULTIPROCESSOR
502 case CPU_ROLE_AP:
503 /*
504 * report on an AP
505 */
506 cpu_intr_init(ci);
507 idt_vec_init_cpu_md(&ci->ci_idtvec, cpu_index(ci));
508 gdt_alloc_cpu(ci);
509 #ifdef i386
510 cpu_set_tss_gates(ci);
511 #endif
512 pmap_cpu_init_late(ci);
513 cpu_start_secondary(ci);
514 if (ci->ci_flags & CPUF_PRESENT) {
515 struct cpu_info *tmp;
516
517 cpu_identify(ci);
518 tmp = cpu_info_list;
519 while (tmp->ci_next)
520 tmp = tmp->ci_next;
521
522 tmp->ci_next = ci;
523 }
524 break;
525 #endif
526
527 default:
528 panic("unknown processor type??\n");
529 }
530
531 pat_init(ci);
532
533 if (!pmf_device_register1(self, cpu_suspend, cpu_resume, cpu_shutdown))
534 aprint_error_dev(self, "couldn't establish power handler\n");
535
536 #ifdef MULTIPROCESSOR
537 if (mp_verbose) {
538 struct lwp *l = ci->ci_data.cpu_idlelwp;
539 struct pcb *pcb = lwp_getpcb(l);
540
541 aprint_verbose_dev(self,
542 "idle lwp at %p, idle sp at %p\n",
543 l,
544 #ifdef i386
545 (void *)pcb->pcb_esp
546 #else
547 (void *)pcb->pcb_rsp
548 #endif
549 );
550 }
551 #endif
552
553 /*
554 * Postpone the "cpufeaturebus" scan.
555 * It is safe to scan the pseudo-bus
556 * only after all CPUs have attached.
557 */
558 (void)config_defer(self, cpu_defer);
559 }
560
561 static void
cpu_defer(device_t self)562 cpu_defer(device_t self)
563 {
564 cpu_rescan(self, NULL, NULL);
565 }
566
567 static int
cpu_rescan(device_t self,const char * ifattr,const int * locators)568 cpu_rescan(device_t self, const char *ifattr, const int *locators)
569 {
570 struct cpu_softc *sc = device_private(self);
571 struct cpufeature_attach_args cfaa;
572 struct cpu_info *ci = sc->sc_info;
573
574 /*
575 * If we booted with RB_MD1 to disable multiprocessor, the
576 * auto-configuration data still contains the additional
577 * CPUs. But their initialization was mostly bypassed
578 * during attach, so we have to make sure we don't look at
579 * their featurebus info, since it wasn't retrieved.
580 */
581 if (ci == NULL)
582 return 0;
583
584 memset(&cfaa, 0, sizeof(cfaa));
585 cfaa.ci = ci;
586
587 if (ifattr_match(ifattr, "cpufeaturebus")) {
588 if (ci->ci_frequency == NULL) {
589 cfaa.name = "frequency";
590 ci->ci_frequency =
591 config_found(self, &cfaa, NULL,
592 CFARGS(.iattr = "cpufeaturebus"));
593 }
594
595 if (ci->ci_padlock == NULL) {
596 cfaa.name = "padlock";
597 ci->ci_padlock =
598 config_found(self, &cfaa, NULL,
599 CFARGS(.iattr = "cpufeaturebus"));
600 }
601
602 if (ci->ci_temperature == NULL) {
603 cfaa.name = "temperature";
604 ci->ci_temperature =
605 config_found(self, &cfaa, NULL,
606 CFARGS(.iattr = "cpufeaturebus"));
607 }
608
609 if (ci->ci_vm == NULL) {
610 cfaa.name = "vm";
611 ci->ci_vm =
612 config_found(self, &cfaa, NULL,
613 CFARGS(.iattr = "cpufeaturebus"));
614 }
615 }
616
617 return 0;
618 }
619
620 static void
cpu_childdetached(device_t self,device_t child)621 cpu_childdetached(device_t self, device_t child)
622 {
623 struct cpu_softc *sc = device_private(self);
624 struct cpu_info *ci = sc->sc_info;
625
626 if (ci->ci_frequency == child)
627 ci->ci_frequency = NULL;
628
629 if (ci->ci_padlock == child)
630 ci->ci_padlock = NULL;
631
632 if (ci->ci_temperature == child)
633 ci->ci_temperature = NULL;
634
635 if (ci->ci_vm == child)
636 ci->ci_vm = NULL;
637 }
638
639 /*
640 * Initialize the processor appropriately.
641 */
642
643 void
cpu_init(struct cpu_info * ci)644 cpu_init(struct cpu_info *ci)
645 {
646 extern int x86_fpu_save;
647 uint32_t cr4 = 0;
648
649 lcr0(rcr0() | CR0_WP);
650
651 /* If global TLB caching is supported, enable it */
652 if (cpu_feature[0] & CPUID_PGE)
653 cr4 |= CR4_PGE;
654
655 /*
656 * If we have FXSAVE/FXRESTOR, use them.
657 */
658 if (cpu_feature[0] & CPUID_FXSR) {
659 cr4 |= CR4_OSFXSR;
660
661 /*
662 * If we have SSE/SSE2, enable XMM exceptions.
663 */
664 if (cpu_feature[0] & (CPUID_SSE|CPUID_SSE2))
665 cr4 |= CR4_OSXMMEXCPT;
666 }
667
668 /* If xsave is supported, enable it */
669 if (cpu_feature[1] & CPUID2_XSAVE)
670 cr4 |= CR4_OSXSAVE;
671
672 /* If SMEP is supported, enable it */
673 if (cpu_feature[5] & CPUID_SEF_SMEP)
674 cr4 |= CR4_SMEP;
675
676 /* If SMAP is supported, enable it */
677 if (cpu_feature[5] & CPUID_SEF_SMAP)
678 cr4 |= CR4_SMAP;
679
680 #ifdef SVS
681 /* If PCID is supported, enable it */
682 if (svs_pcid)
683 cr4 |= CR4_PCIDE;
684 #endif
685
686 if (cr4) {
687 cr4 |= rcr4();
688 lcr4(cr4);
689 }
690
691 /*
692 * Changing CR4 register may change cpuid values. For example, setting
693 * CR4_OSXSAVE sets CPUID2_OSXSAVE. The CPUID2_OSXSAVE is in
694 * ci_feat_val[1], so update it.
695 * XXX Other than ci_feat_val[1] might be changed.
696 */
697 if (cpuid_level >= 1) {
698 u_int descs[4];
699
700 x86_cpuid(1, descs);
701 ci->ci_feat_val[1] = descs[2];
702 }
703
704 if (CPU_IS_PRIMARY(ci) &&
705 x86_fpu_save >= FPU_SAVE_FXSAVE) {
706 fpuinit_mxcsr_mask();
707 }
708
709 /* If xsave is enabled, enable all fpu features */
710 if (cr4 & CR4_OSXSAVE)
711 wrxcr(0, x86_xsave_features & XCR0_FPU);
712
713 #ifdef MTRR
714 /*
715 * On a P6 or above, initialize MTRR's if the hardware supports them.
716 */
717 if (cpu_feature[0] & CPUID_MTRR) {
718 if ((ci->ci_flags & CPUF_AP) == 0)
719 i686_mtrr_init_first();
720 mtrr_init_cpu(ci);
721 }
722
723 #ifdef i386
724 if (strcmp((char *)(ci->ci_vendor), "AuthenticAMD") == 0) {
725 /*
726 * Must be a K6-2 Step >= 7 or a K6-III.
727 */
728 if (CPUID_TO_FAMILY(ci->ci_signature) == 5) {
729 if (CPUID_TO_MODEL(ci->ci_signature) > 8 ||
730 (CPUID_TO_MODEL(ci->ci_signature) == 8 &&
731 CPUID_TO_STEPPING(ci->ci_signature) >= 7)) {
732 mtrr_funcs = &k6_mtrr_funcs;
733 k6_mtrr_init_first();
734 mtrr_init_cpu(ci);
735 }
736 }
737 }
738 #endif /* i386 */
739 #endif /* MTRR */
740
741 if (ci != &cpu_info_primary) {
742 /* Synchronize TSC */
743 atomic_or_32(&ci->ci_flags, CPUF_RUNNING);
744 tsc_sync_ap(ci);
745 } else {
746 atomic_or_32(&ci->ci_flags, CPUF_RUNNING);
747 }
748 }
749
750 #ifdef MULTIPROCESSOR
751 void
cpu_boot_secondary_processors(void)752 cpu_boot_secondary_processors(void)
753 {
754 struct cpu_info *ci;
755 kcpuset_t *cpus;
756 u_long i;
757
758 /* Now that we know the number of CPUs, patch the text segment. */
759 x86_patch(false);
760
761 #if NACPICA > 0
762 /* Finished with NUMA info for now. */
763 acpisrat_exit();
764 #endif
765
766 kcpuset_create(&cpus, true);
767 kcpuset_set(cpus, cpu_index(curcpu()));
768 for (i = 0; i < maxcpus; i++) {
769 ci = cpu_lookup(i);
770 if (ci == NULL)
771 continue;
772 if (ci->ci_data.cpu_idlelwp == NULL)
773 continue;
774 if ((ci->ci_flags & CPUF_PRESENT) == 0)
775 continue;
776 if (ci->ci_flags & (CPUF_BSP|CPUF_SP|CPUF_PRIMARY))
777 continue;
778 cpu_boot_secondary(ci);
779 kcpuset_set(cpus, cpu_index(ci));
780 }
781 while (!kcpuset_match(cpus, kcpuset_running))
782 ;
783 kcpuset_destroy(cpus);
784
785 x86_mp_online = true;
786
787 /* Now that we know about the TSC, attach the timecounter. */
788 tsc_tc_init();
789 }
790 #endif
791
792 static void
cpu_init_idle_lwp(struct cpu_info * ci)793 cpu_init_idle_lwp(struct cpu_info *ci)
794 {
795 struct lwp *l = ci->ci_data.cpu_idlelwp;
796 struct pcb *pcb = lwp_getpcb(l);
797
798 pcb->pcb_cr0 = rcr0();
799 }
800
801 void
cpu_init_idle_lwps(void)802 cpu_init_idle_lwps(void)
803 {
804 struct cpu_info *ci;
805 u_long i;
806
807 for (i = 0; i < maxcpus; i++) {
808 ci = cpu_lookup(i);
809 if (ci == NULL)
810 continue;
811 if (ci->ci_data.cpu_idlelwp == NULL)
812 continue;
813 if ((ci->ci_flags & CPUF_PRESENT) == 0)
814 continue;
815 cpu_init_idle_lwp(ci);
816 }
817 }
818
819 #ifdef MULTIPROCESSOR
820 void
cpu_start_secondary(struct cpu_info * ci)821 cpu_start_secondary(struct cpu_info *ci)
822 {
823 u_long psl;
824 int i;
825
826 #if NLAPIC > 0
827 paddr_t mp_pdirpa;
828 mp_pdirpa = pmap_init_tmp_pgtbl(mp_trampoline_paddr);
829 cpu_copy_trampoline(mp_pdirpa);
830 #endif
831
832 atomic_or_32(&ci->ci_flags, CPUF_AP);
833 ci->ci_curlwp = ci->ci_data.cpu_idlelwp;
834 if (CPU_STARTUP(ci, mp_trampoline_paddr) != 0) {
835 return;
836 }
837
838 /*
839 * Wait for it to become ready. Setting cpu_starting opens the
840 * initial gate and allows the AP to start soft initialization.
841 */
842 KASSERT(cpu_starting == NULL);
843 cpu_starting = ci;
844 for (i = 100000; (!(ci->ci_flags & CPUF_PRESENT)) && i > 0; i--) {
845 delay_func(10);
846 }
847
848 if ((ci->ci_flags & CPUF_PRESENT) == 0) {
849 aprint_error_dev(ci->ci_dev, "failed to become ready\n");
850 #if defined(MPDEBUG) && defined(DDB)
851 printf("dropping into debugger; continue from here to resume boot\n");
852 Debugger();
853 #endif
854 } else {
855 /*
856 * Synchronize time stamp counters. Invalidate cache and do
857 * twice (in tsc_sync_bp) to minimize possible cache effects.
858 * Disable interrupts to try and rule out any external
859 * interference.
860 */
861 psl = x86_read_psl();
862 x86_disable_intr();
863 tsc_sync_bp(ci);
864 x86_write_psl(psl);
865 }
866
867 CPU_START_CLEANUP(ci);
868 cpu_starting = NULL;
869 }
870
871 void
cpu_boot_secondary(struct cpu_info * ci)872 cpu_boot_secondary(struct cpu_info *ci)
873 {
874 int64_t drift;
875 u_long psl;
876 int i;
877
878 atomic_or_32(&ci->ci_flags, CPUF_GO);
879 for (i = 100000; (!(ci->ci_flags & CPUF_RUNNING)) && i > 0; i--) {
880 delay_func(10);
881 }
882 if ((ci->ci_flags & CPUF_RUNNING) == 0) {
883 aprint_error_dev(ci->ci_dev, "failed to start\n");
884 #if defined(MPDEBUG) && defined(DDB)
885 printf("dropping into debugger; continue from here to resume boot\n");
886 Debugger();
887 #endif
888 } else {
889 /* Synchronize TSC again, check for drift. */
890 drift = ci->ci_data.cpu_cc_skew;
891 psl = x86_read_psl();
892 x86_disable_intr();
893 tsc_sync_bp(ci);
894 x86_write_psl(psl);
895 drift -= ci->ci_data.cpu_cc_skew;
896 aprint_debug_dev(ci->ci_dev, "TSC skew=%lld drift=%lld\n",
897 (long long)ci->ci_data.cpu_cc_skew, (long long)drift);
898 tsc_sync_drift(drift);
899 }
900 }
901
902 /*
903 * The CPU ends up here when it's ready to run.
904 * This is called from code in mptramp.s; at this point, we are running
905 * in the idle pcb/idle stack of the new CPU. When this function returns,
906 * this processor will enter the idle loop and start looking for work.
907 */
908 void
cpu_hatch(void * v)909 cpu_hatch(void *v)
910 {
911 struct cpu_info *ci = (struct cpu_info *)v;
912 struct pcb *pcb;
913 int s, i;
914
915 /* ------------------------------------------------------------- */
916
917 /*
918 * This section of code must be compiled with SSP disabled, to
919 * prevent a race against cpu0. See sys/conf/ssp.mk.
920 */
921
922 /*
923 * Initialize MSRs on this CPU:
924 *
925 * - On amd64: Enables SYSCALL/SYSRET.
926 *
927 * - On amd64: Sets up %fs and %gs so that %gs points to the
928 * current struct cpu_info as needed for CPUVAR(...),
929 * curcpu(), and curlwp.
930 *
931 * (On i386, CPUVAR(...), curcpu(), and curlwp are made to
932 * work first by the conifguration of segment descriptors in
933 * the Global Descriptor Table (GDT) in initgdt.)
934 *
935 * - Enables the no-execute bit if supported.
936 *
937 * Thus, after this point, CPUVAR(...), curcpu(), and curlwp
938 * will work on this CPU.
939 *
940 * Note: The call to cpu_init_msrs for cpu0 happens in
941 * init386/init_x86_64.
942 */
943 cpu_init_msrs(ci, true);
944
945 cpu_probe(ci);
946 cpu_speculation_init(ci);
947 #if NHYPERV > 0
948 hyperv_init_cpu(ci);
949 #endif
950
951 ci->ci_data.cpu_cc_freq = cpu_info_primary.ci_data.cpu_cc_freq;
952 /* cpu_get_tsc_freq(ci); */
953
954 KDASSERT((ci->ci_flags & CPUF_PRESENT) == 0);
955
956 /*
957 * Synchronize the TSC for the first time. Note that interrupts are
958 * off at this point.
959 */
960 atomic_or_32(&ci->ci_flags, CPUF_PRESENT);
961 tsc_sync_ap(ci);
962
963 /* ------------------------------------------------------------- */
964
965 /*
966 * Wait to be brought online.
967 *
968 * Use MONITOR/MWAIT if available. These instructions put the CPU in
969 * a low consumption mode (C-state), and if the TSC is not invariant,
970 * this causes the TSC to drift. We want this to happen, so that we
971 * can later detect (in tsc_tc_init) any abnormal drift with invariant
972 * TSCs. That's just for safety; by definition such drifts should
973 * never occur with invariant TSCs.
974 *
975 * If not available, try PAUSE. We'd like to use HLT, but we have
976 * interrupts off.
977 */
978 while ((ci->ci_flags & CPUF_GO) == 0) {
979 if ((cpu_feature[1] & CPUID2_MONITOR) != 0) {
980 x86_monitor(&ci->ci_flags, 0, 0);
981 if ((ci->ci_flags & CPUF_GO) != 0) {
982 continue;
983 }
984 x86_mwait(0, 0);
985 } else {
986 /*
987 * XXX The loop repetition count could be a lot higher, but
988 * XXX currently qemu emulator takes a _very_long_time_ to
989 * XXX execute the pause instruction. So for now, use a low
990 * XXX value to allow the cpu to hatch before timing out.
991 */
992 for (i = 50; i != 0; i--) {
993 x86_pause();
994 }
995 }
996 }
997
998 /* Because the text may have been patched in x86_patch(). */
999 wbinvd();
1000 x86_flush();
1001 tlbflushg();
1002
1003 KASSERT((ci->ci_flags & CPUF_RUNNING) == 0);
1004
1005 #ifdef PAE
1006 pd_entry_t * l3_pd = ci->ci_pae_l3_pdir;
1007 for (i = 0 ; i < PDP_SIZE; i++) {
1008 l3_pd[i] = pmap_kernel()->pm_pdirpa[i] | PTE_P;
1009 }
1010 lcr3(ci->ci_pae_l3_pdirpa);
1011 #else
1012 lcr3(pmap_pdirpa(pmap_kernel(), 0));
1013 #endif
1014
1015 pcb = lwp_getpcb(curlwp);
1016 pcb->pcb_cr3 = rcr3();
1017 pcb = lwp_getpcb(ci->ci_data.cpu_idlelwp);
1018 lcr0(pcb->pcb_cr0);
1019
1020 cpu_init_idt(ci);
1021 gdt_init_cpu(ci);
1022 #if NLAPIC > 0
1023 lapic_enable();
1024 lapic_set_lvt();
1025 #endif
1026
1027 fpuinit(ci);
1028 lldt(GSYSSEL(GLDT_SEL, SEL_KPL));
1029 ltr(ci->ci_tss_sel);
1030
1031 /*
1032 * cpu_init will re-synchronize the TSC, and will detect any abnormal
1033 * drift that would have been caused by the use of MONITOR/MWAIT
1034 * above.
1035 */
1036 cpu_init(ci);
1037 #ifdef XENPVHVM
1038 xen_hvm_init_cpu(ci);
1039 #endif
1040 (*x86_initclock_func)();
1041 cpu_get_tsc_freq(ci);
1042
1043 s = splhigh();
1044 #if NLAPIC > 0
1045 lapic_write_tpri(0);
1046 #endif
1047 x86_enable_intr();
1048 splx(s);
1049 x86_errata();
1050
1051 aprint_debug_dev(ci->ci_dev, "running\n");
1052
1053 kcsan_cpu_init(ci);
1054
1055 idle_loop(NULL);
1056 KASSERT(false);
1057 }
1058 #endif
1059
1060 #if defined(DDB)
1061
1062 #include <ddb/db_output.h>
1063 #include <machine/db_machdep.h>
1064
1065 /*
1066 * Dump CPU information from ddb.
1067 */
1068 void
cpu_debug_dump(void)1069 cpu_debug_dump(void)
1070 {
1071 struct cpu_info *ci;
1072 CPU_INFO_ITERATOR cii;
1073 const char sixtyfour64space[] =
1074 #ifdef _LP64
1075 " "
1076 #endif
1077 "";
1078
1079 db_printf("addr %sdev id flags ipis spl curlwp "
1080 "\n", sixtyfour64space);
1081 for (CPU_INFO_FOREACH(cii, ci)) {
1082 db_printf("%p %s %ld %x %x %d %10p\n",
1083 ci,
1084 ci->ci_dev == NULL ? "BOOT" : device_xname(ci->ci_dev),
1085 (long)ci->ci_cpuid,
1086 ci->ci_flags, ci->ci_ipis, ci->ci_ilevel,
1087 ci->ci_curlwp);
1088 }
1089 }
1090 #endif
1091
1092 #ifdef MULTIPROCESSOR
1093 #if NLAPIC > 0
1094 static void
cpu_copy_trampoline(paddr_t pdir_pa)1095 cpu_copy_trampoline(paddr_t pdir_pa)
1096 {
1097 extern uint32_t nox_flag;
1098 extern u_char cpu_spinup_trampoline[];
1099 extern u_char cpu_spinup_trampoline_end[];
1100 vaddr_t mp_trampoline_vaddr;
1101 struct {
1102 uint32_t large;
1103 uint32_t nox;
1104 uint32_t pdir;
1105 } smp_data;
1106 CTASSERT(sizeof(smp_data) == 3 * 4);
1107
1108 smp_data.large = (pmap_largepages != 0);
1109 smp_data.nox = nox_flag;
1110 smp_data.pdir = (uint32_t)(pdir_pa & 0xFFFFFFFF);
1111
1112 /* Enter the physical address */
1113 mp_trampoline_vaddr = uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
1114 UVM_KMF_VAONLY);
1115 pmap_kenter_pa(mp_trampoline_vaddr, mp_trampoline_paddr,
1116 VM_PROT_READ | VM_PROT_WRITE, 0);
1117 pmap_update(pmap_kernel());
1118
1119 /* Copy boot code */
1120 memcpy((void *)mp_trampoline_vaddr,
1121 cpu_spinup_trampoline,
1122 cpu_spinup_trampoline_end - cpu_spinup_trampoline);
1123
1124 /* Copy smp_data at the end */
1125 memcpy((void *)(mp_trampoline_vaddr + PAGE_SIZE - sizeof(smp_data)),
1126 &smp_data, sizeof(smp_data));
1127
1128 pmap_kremove(mp_trampoline_vaddr, PAGE_SIZE);
1129 pmap_update(pmap_kernel());
1130 uvm_km_free(kernel_map, mp_trampoline_vaddr, PAGE_SIZE, UVM_KMF_VAONLY);
1131 }
1132 #endif
1133
1134 int
mp_cpu_start(struct cpu_info * ci,paddr_t target)1135 mp_cpu_start(struct cpu_info *ci, paddr_t target)
1136 {
1137 #if NLAPIC > 0
1138 int error;
1139
1140 /*
1141 * Bootstrap code must be addressable in real mode
1142 * and it must be page aligned.
1143 */
1144 KASSERT(target < 0x10000 && target % PAGE_SIZE == 0);
1145
1146 /*
1147 * "The BSP must initialize CMOS shutdown code to 0Ah ..."
1148 */
1149
1150 outb(IO_RTC, NVRAM_RESET);
1151 outb(IO_RTC+1, NVRAM_RESET_JUMP);
1152
1153 /*
1154 * "and the warm reset vector (DWORD based at 40:67) to point
1155 * to the AP startup code ..."
1156 */
1157 unsigned short dwordptr[2];
1158 dwordptr[0] = 0;
1159 dwordptr[1] = target >> 4;
1160
1161 memcpy((uint8_t *)cmos_data_mapping + 0x467, dwordptr, 4);
1162
1163 if ((cpu_feature[0] & CPUID_APIC) == 0) {
1164 aprint_error("mp_cpu_start: CPU does not have APIC\n");
1165 return ENODEV;
1166 }
1167
1168 /*
1169 * ... prior to executing the following sequence:". We'll also add in
1170 * local cache flush, in case the BIOS has left the AP with its cache
1171 * disabled. It may not be able to cope with MP coherency.
1172 */
1173 wbinvd();
1174
1175 if (ci->ci_flags & CPUF_AP) {
1176 error = x86_ipi_init(ci->ci_cpuid);
1177 if (error != 0) {
1178 aprint_error_dev(ci->ci_dev, "%s: IPI not taken (1)\n",
1179 __func__);
1180 return error;
1181 }
1182 delay_func(10000);
1183
1184 error = x86_ipi_startup(ci->ci_cpuid, target / PAGE_SIZE);
1185 if (error != 0) {
1186 aprint_error_dev(ci->ci_dev, "%s: IPI not taken (2)\n",
1187 __func__);
1188 return error;
1189 }
1190 delay_func(200);
1191
1192 error = x86_ipi_startup(ci->ci_cpuid, target / PAGE_SIZE);
1193 if (error != 0) {
1194 aprint_error_dev(ci->ci_dev, "%s: IPI not taken (3)\n",
1195 __func__);
1196 return error;
1197 }
1198 delay_func(200);
1199 }
1200
1201 return 0;
1202 #else
1203 return ENODEV;
1204 #endif /* NLAPIC > 0 */
1205 }
1206
1207 void
mp_cpu_start_cleanup(struct cpu_info * ci)1208 mp_cpu_start_cleanup(struct cpu_info *ci)
1209 {
1210 /*
1211 * Ensure the NVRAM reset byte contains something vaguely sane.
1212 */
1213
1214 outb(IO_RTC, NVRAM_RESET);
1215 outb(IO_RTC+1, NVRAM_RESET_RST);
1216 }
1217 #endif
1218
1219 #ifdef __x86_64__
1220 typedef void (vector)(void);
1221 extern vector Xsyscall, Xsyscall32, Xsyscall_svs;
1222 #endif
1223
1224 /*
1225 * cpu_init_msrs(ci, full)
1226 *
1227 * Initialize some Model-Specific Registers (MSRs) on the current
1228 * CPU, whose struct cpu_info pointer is ci, for:
1229 *
1230 * - SYSCALL/SYSRET.
1231 * - %fs/%gs on amd64 if `full' is true; needed to make
1232 * CPUVAR(...), curcpu(), and curlwp work. (We do this at boot,
1233 * but skip it on ACPI wakeup.)
1234 * - No-execute bit, if supported.
1235 *
1236 * References:
1237 *
1238 * - Intel 64 and IA-32 Architectures Software Developer's Manual,
1239 * Volume 3: System Programming Guide, Order Number 325384,
1240 * April 2022, Sec. 5.8.8 `Fast System Calls in 64-Bit Mode',
1241 * pp. 5-22 through 5-23.
1242 *
1243 * - Intel 64 and IA-32 Architectures Software Developer's Manual,
1244 * Volume 4: Model-Specific Registers, Order Number 335592,
1245 * April 2022, Sec. 2.1 `Architectural MSRs', Table 2-2,
1246 * pp. 2-60 through 2-61.
1247 */
1248 void
cpu_init_msrs(struct cpu_info * ci,bool full)1249 cpu_init_msrs(struct cpu_info *ci, bool full)
1250 {
1251 #ifdef __x86_64__
1252 /*
1253 * On amd64, set up the syscall target address registers
1254 * for SYSCALL/SYSRET:
1255 *
1256 * - IA32_STAR, c000_0081h (MSR_STAR): System Call Target
1257 * Address. Code and stack segment selectors for SYSRET
1258 * (bits 48:63) and SYSCALL (bits 32:47).
1259 *
1260 * - IA32_LSTAR, c000_0082h (MSR_LSTAR): IA-32e Mode System
1261 * Call Target Address. Target rip for SYSCALL when executed
1262 * in 64-bit mode.
1263 *
1264 * - IA32_CSTAR, c000_0083h (MSR_CSTAR): IA-32e Mode System
1265 * Call Target Address. Target rip for SYSCALL when executed
1266 * in compatibility mode. (XXX Manual says this is `[n]ot
1267 * used, as the SYSCALL instruction is not recognized in
1268 * compatibility mode', so why do we set it?)
1269 *
1270 * - IA32_FMASK, c000_0084h (MSR_SFMASK): System Call Flag
1271 * Mask. Mask for the RFLAGS register on SYSCALL.
1272 */
1273 wrmsr(MSR_STAR,
1274 ((uint64_t)GSEL(GCODE_SEL, SEL_KPL) << 32) |
1275 ((uint64_t)LSEL(LSYSRETBASE_SEL, SEL_UPL) << 48));
1276 wrmsr(MSR_LSTAR, (uint64_t)Xsyscall);
1277 wrmsr(MSR_CSTAR, (uint64_t)Xsyscall32);
1278 wrmsr(MSR_SFMASK, PSL_NT|PSL_T|PSL_I|PSL_C|PSL_D|PSL_AC);
1279
1280 #ifdef SVS
1281 if (svs_enabled)
1282 wrmsr(MSR_LSTAR, (uint64_t)Xsyscall_svs);
1283 #endif
1284
1285 /*
1286 * On amd64 if `full' is true -- used at boot, but not on ACPI
1287 * wakeup -- then additionally set up %fs and %gs:
1288 *
1289 * - IA32_FS_BASE, c000_0100h (MSR_FSBASE): Base address of
1290 * %fs. Not used in NetBSD kernel, so zero it.
1291 *
1292 * - IA32_GS_BASE, c000_0101h (MSR_GSBASE): Base address of
1293 * %gs. Used in NetBSD kernel by CPUVAR(...), curcpu(), and
1294 * curlwp for access to the CPU-local area, so set it to ci.
1295 *
1296 * - IA32_KERNEL_GS_BASE, c000_0102h (MSR_KERNELGSBASE): Base
1297 * address of what swapgs will leave in %gs when switching to
1298 * userland. Zero for now; will be set to pcb->pcb_gs in
1299 * cpu_switchto for user threads.
1300 */
1301 if (full) {
1302 wrmsr(MSR_FSBASE, 0);
1303 wrmsr(MSR_GSBASE, (uint64_t)ci);
1304 wrmsr(MSR_KERNELGSBASE, 0);
1305 }
1306 #endif /* __x86_64__ */
1307
1308 /*
1309 * If the no-execute bit is supported, enable it in:
1310 *
1311 * - IA32_EFER, c000_0080h (MSR_EFER): Extended Feature
1312 * Enables.
1313 */
1314 if (cpu_feature[2] & CPUID_NOX)
1315 wrmsr(MSR_EFER, rdmsr(MSR_EFER) | EFER_NXE);
1316 }
1317
1318 void
cpu_offline_md(void)1319 cpu_offline_md(void)
1320 {
1321 return;
1322 }
1323
1324 /* XXX joerg restructure and restart CPUs individually */
1325 static bool
cpu_stop(device_t dv)1326 cpu_stop(device_t dv)
1327 {
1328 struct cpu_softc *sc = device_private(dv);
1329 struct cpu_info *ci = sc->sc_info;
1330 int err;
1331
1332 KASSERT((ci->ci_flags & CPUF_PRESENT) != 0);
1333
1334 if (CPU_IS_PRIMARY(ci))
1335 return true;
1336
1337 if (ci->ci_data.cpu_idlelwp == NULL)
1338 return true;
1339
1340 sc->sc_wasonline = !(ci->ci_schedstate.spc_flags & SPCF_OFFLINE);
1341
1342 if (sc->sc_wasonline) {
1343 mutex_enter(&cpu_lock);
1344 err = cpu_setstate(ci, false);
1345 mutex_exit(&cpu_lock);
1346
1347 if (err != 0)
1348 return false;
1349 }
1350
1351 return true;
1352 }
1353
1354 static bool
cpu_suspend(device_t dv,const pmf_qual_t * qual)1355 cpu_suspend(device_t dv, const pmf_qual_t *qual)
1356 {
1357 struct cpu_softc *sc = device_private(dv);
1358 struct cpu_info *ci = sc->sc_info;
1359
1360 if ((ci->ci_flags & CPUF_PRESENT) == 0)
1361 return true;
1362 else {
1363 cpufreq_suspend(ci);
1364 }
1365
1366 return cpu_stop(dv);
1367 }
1368
1369 static bool
cpu_resume(device_t dv,const pmf_qual_t * qual)1370 cpu_resume(device_t dv, const pmf_qual_t *qual)
1371 {
1372 struct cpu_softc *sc = device_private(dv);
1373 struct cpu_info *ci = sc->sc_info;
1374 int err = 0;
1375
1376 if ((ci->ci_flags & CPUF_PRESENT) == 0)
1377 return true;
1378
1379 if (CPU_IS_PRIMARY(ci))
1380 goto out;
1381
1382 if (ci->ci_data.cpu_idlelwp == NULL)
1383 goto out;
1384
1385 if (sc->sc_wasonline) {
1386 mutex_enter(&cpu_lock);
1387 err = cpu_setstate(ci, true);
1388 mutex_exit(&cpu_lock);
1389 }
1390
1391 out:
1392 if (err != 0)
1393 return false;
1394
1395 cpufreq_resume(ci);
1396
1397 return true;
1398 }
1399
1400 static bool
cpu_shutdown(device_t dv,int how)1401 cpu_shutdown(device_t dv, int how)
1402 {
1403 struct cpu_softc *sc = device_private(dv);
1404 struct cpu_info *ci = sc->sc_info;
1405
1406 if ((ci->ci_flags & CPUF_BSP) != 0)
1407 return false;
1408
1409 if ((ci->ci_flags & CPUF_PRESENT) == 0)
1410 return true;
1411
1412 return cpu_stop(dv);
1413 }
1414
1415 /* Get the TSC frequency and set it to ci->ci_data.cpu_cc_freq. */
1416 void
cpu_get_tsc_freq(struct cpu_info * ci)1417 cpu_get_tsc_freq(struct cpu_info *ci)
1418 {
1419 uint64_t freq = 0, freq_from_cpuid, t0, t1;
1420 int64_t overhead;
1421
1422 if (CPU_IS_PRIMARY(ci) && cpu_hascounter()) {
1423 /*
1424 * If it's the first call of this function, try to get TSC
1425 * freq from CPUID by calling cpu_tsc_freq_cpuid().
1426 * The function also set lapic_per_second variable if it's
1427 * known. This is required for Intel's Comet Lake and newer
1428 * processors to set LAPIC timer correctly.
1429 */
1430 if (ci->ci_data.cpu_cc_freq == 0)
1431 freq = freq_from_cpuid = cpu_tsc_freq_cpuid(ci);
1432 if (freq != 0)
1433 aprint_debug_dev(ci->ci_dev, "TSC freq "
1434 "from CPUID %" PRIu64 " Hz\n", freq);
1435 #if NHPET > 0
1436 if (freq == 0) {
1437 freq = hpet_tsc_freq();
1438 if (freq != 0)
1439 aprint_debug_dev(ci->ci_dev, "TSC freq "
1440 "from HPET %" PRIu64 " Hz\n", freq);
1441 }
1442 #endif
1443 if (freq == 0) {
1444 /*
1445 * Work out the approximate overhead involved below.
1446 * Discard the result of the first go around the
1447 * loop.
1448 */
1449 overhead = 0;
1450 for (int i = 0; i <= 8; i++) {
1451 const int s = splhigh();
1452 t0 = cpu_counter();
1453 delay_func(0);
1454 t1 = cpu_counter();
1455 splx(s);
1456 if (i > 0) {
1457 overhead += (t1 - t0);
1458 }
1459 }
1460 overhead >>= 3;
1461
1462 /*
1463 * Now do the calibration.
1464 */
1465 freq = 0;
1466 for (int i = 0; i < 1000; i++) {
1467 const int s = splhigh();
1468 t0 = cpu_counter();
1469 delay_func(100);
1470 t1 = cpu_counter();
1471 splx(s);
1472 freq += t1 - t0 - overhead;
1473 }
1474 freq = freq * 10;
1475
1476 aprint_debug_dev(ci->ci_dev, "TSC freq "
1477 "from delay %" PRIu64 " Hz\n", freq);
1478 }
1479 if (ci->ci_data.cpu_cc_freq != 0) {
1480 freq_from_cpuid = cpu_tsc_freq_cpuid(ci);
1481 if ((freq_from_cpuid != 0)
1482 && (freq != freq_from_cpuid))
1483 aprint_verbose_dev(ci->ci_dev, "TSC freq "
1484 "calibrated %" PRIu64 " Hz\n", freq);
1485 }
1486 } else {
1487 freq = cpu_info_primary.ci_data.cpu_cc_freq;
1488 }
1489
1490 ci->ci_data.cpu_cc_freq = freq;
1491 }
1492
1493 void
x86_cpu_idle_mwait(void)1494 x86_cpu_idle_mwait(void)
1495 {
1496 struct cpu_info *ci = curcpu();
1497
1498 KASSERT(ci->ci_ilevel == IPL_NONE);
1499
1500 x86_monitor(&ci->ci_want_resched, 0, 0);
1501 if (__predict_false(ci->ci_want_resched)) {
1502 return;
1503 }
1504 x86_mwait(0, 0);
1505 }
1506
1507 void
x86_cpu_idle_halt(void)1508 x86_cpu_idle_halt(void)
1509 {
1510 struct cpu_info *ci = curcpu();
1511
1512 KASSERT(ci->ci_ilevel == IPL_NONE);
1513
1514 x86_disable_intr();
1515 if (!__predict_false(ci->ci_want_resched)) {
1516 x86_stihlt();
1517 } else {
1518 x86_enable_intr();
1519 }
1520 }
1521
1522 /*
1523 * Loads pmap for the current CPU.
1524 */
1525 void
cpu_load_pmap(struct pmap * pmap,struct pmap * oldpmap)1526 cpu_load_pmap(struct pmap *pmap, struct pmap *oldpmap)
1527 {
1528
1529 KASSERT(kpreempt_disabled());
1530
1531 #ifdef SVS
1532 if (svs_enabled && pmap_is_user(pmap)) {
1533 svs_pdir_switch(pmap);
1534 }
1535 #endif
1536
1537 #ifdef PAE
1538 struct cpu_info *ci = curcpu();
1539 bool interrupts_enabled;
1540 pd_entry_t *l3_pd = ci->ci_pae_l3_pdir;
1541 int i;
1542
1543 /*
1544 * disable interrupts to block TLB shootdowns, which can reload cr3.
1545 * while this doesn't block NMIs, it's probably ok as NMIs unlikely
1546 * reload cr3.
1547 */
1548 interrupts_enabled = (x86_read_flags() & PSL_I) != 0;
1549 if (interrupts_enabled)
1550 x86_disable_intr();
1551
1552 for (i = 0 ; i < PDP_SIZE; i++) {
1553 l3_pd[i] = pmap->pm_pdirpa[i] | PTE_P;
1554 }
1555
1556 if (interrupts_enabled)
1557 x86_enable_intr();
1558 tlbflush();
1559 #else
1560 lcr3(pmap_pdirpa(pmap, 0));
1561 #endif
1562 }
1563
1564 /*
1565 * Notify all other cpus to halt.
1566 */
1567
1568 void
cpu_broadcast_halt(void)1569 cpu_broadcast_halt(void)
1570 {
1571 x86_broadcast_ipi(X86_IPI_HALT);
1572 }
1573
1574 /*
1575 * Send a dummy ipi to a cpu to force it to run splraise()/spllower(),
1576 * and trigger an AST on the running LWP.
1577 */
1578
1579 void
cpu_kick(struct cpu_info * ci)1580 cpu_kick(struct cpu_info *ci)
1581 {
1582 x86_send_ipi(ci, X86_IPI_AST);
1583 }
1584