1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 1996, by Steve Passe
5 * Copyright (c) 2003, by Peter Wemm
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. The name of the developer may NOT be used to endorse or promote products
14 * derived from this software without specific prior written permission.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD: releng/12.0/sys/amd64/amd64/mp_machdep.c 338356 2018-08-28 18:47:02Z kib $");
31
32 #include "opt_cpu.h"
33 #include "opt_ddb.h"
34 #include "opt_kstack_pages.h"
35 #include "opt_sched.h"
36 #include "opt_smp.h"
37
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/bus.h>
41 #include <sys/cpuset.h>
42 #ifdef GPROF
43 #include <sys/gmon.h>
44 #endif
45 #include <sys/kernel.h>
46 #include <sys/ktr.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/memrange.h>
50 #include <sys/mutex.h>
51 #include <sys/pcpu.h>
52 #include <sys/proc.h>
53 #include <sys/sched.h>
54 #include <sys/smp.h>
55 #include <sys/sysctl.h>
56
57 #include <vm/vm.h>
58 #include <vm/vm_param.h>
59 #include <vm/pmap.h>
60 #include <vm/vm_kern.h>
61 #include <vm/vm_extern.h>
62
63 #include <x86/apicreg.h>
64 #include <machine/clock.h>
65 #include <machine/cputypes.h>
66 #include <machine/cpufunc.h>
67 #include <x86/mca.h>
68 #include <machine/md_var.h>
69 #include <machine/pcb.h>
70 #include <machine/psl.h>
71 #include <machine/smp.h>
72 #include <machine/specialreg.h>
73 #include <machine/tss.h>
74 #include <x86/ucode.h>
75 #include <machine/cpu.h>
76 #include <x86/init.h>
77
78 #define WARMBOOT_TARGET 0
79 #define WARMBOOT_OFF (KERNBASE + 0x0467)
80 #define WARMBOOT_SEG (KERNBASE + 0x0469)
81
82 #define CMOS_REG (0x70)
83 #define CMOS_DATA (0x71)
84 #define BIOS_RESET (0x0f)
85 #define BIOS_WARM (0x0a)
86
87 #define GiB(v) (v ## ULL << 30)
88
89 #define AP_BOOTPT_SZ (PAGE_SIZE * 3)
90
91 extern struct pcpu __pcpu[];
92
93 /* Temporary variables for init_secondary() */
94 char *doublefault_stack;
95 char *mce_stack;
96 char *nmi_stack;
97 char *dbg_stack;
98
99 /*
100 * Local data and functions.
101 */
102
103 static int start_ap(int apic_id);
104
105 static bool
106 is_kernel_paddr(vm_paddr_t pa)
107 {
108
109 return (pa >= trunc_2mpage(btext - KERNBASE) &&
110 pa < round_page(_end - KERNBASE));
111 }
112
113 static bool
114 is_mpboot_good(vm_paddr_t start, vm_paddr_t end)
115 {
116
117 return (start + AP_BOOTPT_SZ <= GiB(4) && atop(end) < Maxmem);
118 }
119
120 /*
121 * Calculate usable address in base memory for AP trampoline code.
122 */
123 void
124 mp_bootaddress(vm_paddr_t *physmap, unsigned int *physmap_idx)
125 {
126 vm_paddr_t start, end;
127 unsigned int i;
128 bool allocated;
129
130 alloc_ap_trampoline(physmap, physmap_idx);
131
132 /*
133 * Find a memory region big enough below the 4GB boundary to
134 * store the initial page tables. Region must be mapped by
135 * the direct map.
136 *
137 * Note that it needs to be aligned to a page boundary.
138 */
139 allocated = false;
140 for (i = *physmap_idx; i <= *physmap_idx; i -= 2) {
141 /*
142 * First, try to chomp at the start of the physmap region.
143 * Kernel binary might claim it already.
144 */
145 start = round_page(physmap[i]);
146 end = start + AP_BOOTPT_SZ;
147 if (start < end && end <= physmap[i + 1] &&
148 is_mpboot_good(start, end) &&
149 !is_kernel_paddr(start) && !is_kernel_paddr(end - 1)) {
150 allocated = true;
151 physmap[i] = end;
152 break;
153 }
154
155 /*
156 * Second, try to chomp at the end. Again, check
157 * against kernel.
158 */
159 end = trunc_page(physmap[i + 1]);
160 start = end - AP_BOOTPT_SZ;
161 if (start < end && start >= physmap[i] &&
162 is_mpboot_good(start, end) &&
163 !is_kernel_paddr(start) && !is_kernel_paddr(end - 1)) {
164 allocated = true;
165 physmap[i + 1] = start;
166 break;
167 }
168 }
169 if (allocated) {
170 mptramp_pagetables = start;
171 if (physmap[i] == physmap[i + 1] && *physmap_idx != 0) {
172 memmove(&physmap[i], &physmap[i + 2],
173 sizeof(*physmap) * (*physmap_idx - i + 2));
174 *physmap_idx -= 2;
175 }
176 } else {
177 mptramp_pagetables = trunc_page(boot_address) - AP_BOOTPT_SZ;
178 if (bootverbose)
179 printf(
180 "Cannot find enough space for the initial AP page tables, placing them at %#x",
181 mptramp_pagetables);
182 }
183 }
184
185 /*
186 * Initialize the IPI handlers and start up the AP's.
187 */
188 void
189 cpu_mp_start(void)
190 {
191 int i;
192
193 /* Initialize the logical ID to APIC ID table. */
194 for (i = 0; i < MAXCPU; i++) {
195 cpu_apic_ids[i] = -1;
196 cpu_ipi_pending[i] = 0;
197 }
198
199 /* Install an inter-CPU IPI for TLB invalidation */
200 if (pmap_pcid_enabled) {
201 if (invpcid_works) {
202 setidt(IPI_INVLTLB, pti ?
203 IDTVEC(invltlb_invpcid_pti_pti) :
204 IDTVEC(invltlb_invpcid_nopti), SDT_SYSIGT,
205 SEL_KPL, 0);
206 setidt(IPI_INVLPG, pti ? IDTVEC(invlpg_invpcid_pti) :
207 IDTVEC(invlpg_invpcid), SDT_SYSIGT, SEL_KPL, 0);
208 setidt(IPI_INVLRNG, pti ? IDTVEC(invlrng_invpcid_pti) :
209 IDTVEC(invlrng_invpcid), SDT_SYSIGT, SEL_KPL, 0);
210 } else {
211 setidt(IPI_INVLTLB, pti ? IDTVEC(invltlb_pcid_pti) :
212 IDTVEC(invltlb_pcid), SDT_SYSIGT, SEL_KPL, 0);
213 setidt(IPI_INVLPG, pti ? IDTVEC(invlpg_pcid_pti) :
214 IDTVEC(invlpg_pcid), SDT_SYSIGT, SEL_KPL, 0);
215 setidt(IPI_INVLRNG, pti ? IDTVEC(invlrng_pcid_pti) :
216 IDTVEC(invlrng_pcid), SDT_SYSIGT, SEL_KPL, 0);
217 }
218 } else {
219 setidt(IPI_INVLTLB, pti ? IDTVEC(invltlb_pti) : IDTVEC(invltlb),
220 SDT_SYSIGT, SEL_KPL, 0);
221 setidt(IPI_INVLPG, pti ? IDTVEC(invlpg_pti) : IDTVEC(invlpg),
222 SDT_SYSIGT, SEL_KPL, 0);
223 setidt(IPI_INVLRNG, pti ? IDTVEC(invlrng_pti) : IDTVEC(invlrng),
224 SDT_SYSIGT, SEL_KPL, 0);
225 }
226
227 /* Install an inter-CPU IPI for cache invalidation. */
228 setidt(IPI_INVLCACHE, pti ? IDTVEC(invlcache_pti) : IDTVEC(invlcache),
229 SDT_SYSIGT, SEL_KPL, 0);
230
231 /* Install an inter-CPU IPI for all-CPU rendezvous */
232 setidt(IPI_RENDEZVOUS, pti ? IDTVEC(rendezvous_pti) :
233 IDTVEC(rendezvous), SDT_SYSIGT, SEL_KPL, 0);
234
235 /* Install generic inter-CPU IPI handler */
236 setidt(IPI_BITMAP_VECTOR, pti ? IDTVEC(ipi_intr_bitmap_handler_pti) :
237 IDTVEC(ipi_intr_bitmap_handler), SDT_SYSIGT, SEL_KPL, 0);
238
239 /* Install an inter-CPU IPI for CPU stop/restart */
240 setidt(IPI_STOP, pti ? IDTVEC(cpustop_pti) : IDTVEC(cpustop),
241 SDT_SYSIGT, SEL_KPL, 0);
242
243 /* Install an inter-CPU IPI for CPU suspend/resume */
244 setidt(IPI_SUSPEND, pti ? IDTVEC(cpususpend_pti) : IDTVEC(cpususpend),
245 SDT_SYSIGT, SEL_KPL, 0);
246
247 /* Set boot_cpu_id if needed. */
248 if (boot_cpu_id == -1) {
249 boot_cpu_id = PCPU_GET(apic_id);
250 cpu_info[boot_cpu_id].cpu_bsp = 1;
251 } else
252 KASSERT(boot_cpu_id == PCPU_GET(apic_id),
253 ("BSP's APIC ID doesn't match boot_cpu_id"));
254
255 /* Probe logical/physical core configuration. */
256 topo_probe();
257
258 assign_cpu_ids();
259
260 /* Start each Application Processor */
261 init_ops.start_all_aps();
262
263 set_interrupt_apic_ids();
264 }
265
266
267 /*
268 * AP CPU's call this to initialize themselves.
269 */
270 void
271 init_secondary(void)
272 {
273 struct pcpu *pc;
274 struct nmi_pcpu *np;
275 u_int64_t cr0;
276 int cpu, gsel_tss, x;
277 struct region_descriptor ap_gdt;
278
279 /* Set by the startup code for us to use */
280 cpu = bootAP;
281
282 /* Update microcode before doing anything else. */
283 ucode_load_ap(cpu);
284
285 /* Init tss */
286 common_tss[cpu] = common_tss[0];
287 common_tss[cpu].tss_iobase = sizeof(struct amd64tss) +
288 IOPERM_BITMAP_SIZE;
289 common_tss[cpu].tss_ist1 = (long)&doublefault_stack[PAGE_SIZE];
290
291 /* The NMI stack runs on IST2. */
292 np = ((struct nmi_pcpu *) &nmi_stack[PAGE_SIZE]) - 1;
293 common_tss[cpu].tss_ist2 = (long) np;
294
295 /* The MC# stack runs on IST3. */
296 np = ((struct nmi_pcpu *) &mce_stack[PAGE_SIZE]) - 1;
297 common_tss[cpu].tss_ist3 = (long) np;
298
299 /* The DB# stack runs on IST4. */
300 np = ((struct nmi_pcpu *) &dbg_stack[PAGE_SIZE]) - 1;
301 common_tss[cpu].tss_ist4 = (long) np;
302
303 /* Prepare private GDT */
304 gdt_segs[GPROC0_SEL].ssd_base = (long) &common_tss[cpu];
305 for (x = 0; x < NGDT; x++) {
306 if (x != GPROC0_SEL && x != (GPROC0_SEL + 1) &&
307 x != GUSERLDT_SEL && x != (GUSERLDT_SEL + 1))
308 ssdtosd(&gdt_segs[x], &gdt[NGDT * cpu + x]);
309 }
310 ssdtosyssd(&gdt_segs[GPROC0_SEL],
311 (struct system_segment_descriptor *)&gdt[NGDT * cpu + GPROC0_SEL]);
312 ap_gdt.rd_limit = NGDT * sizeof(gdt[0]) - 1;
313 ap_gdt.rd_base = (long) &gdt[NGDT * cpu];
314 lgdt(&ap_gdt); /* does magic intra-segment return */
315
316 /* Get per-cpu data */
317 pc = &__pcpu[cpu];
318
319 /* prime data page for it to use */
320 pcpu_init(pc, cpu, sizeof(struct pcpu));
321 dpcpu_init(dpcpu, cpu);
322 pc->pc_apic_id = cpu_apic_ids[cpu];
323 pc->pc_prvspace = pc;
324 pc->pc_curthread = 0;
325 pc->pc_tssp = &common_tss[cpu];
326 pc->pc_commontssp = &common_tss[cpu];
327 pc->pc_rsp0 = 0;
328 pc->pc_pti_rsp0 = (((vm_offset_t)&pc->pc_pti_stack +
329 PC_PTI_STACK_SZ * sizeof(uint64_t)) & ~0xful);
330 pc->pc_tss = (struct system_segment_descriptor *)&gdt[NGDT * cpu +
331 GPROC0_SEL];
332 pc->pc_fs32p = &gdt[NGDT * cpu + GUFS32_SEL];
333 pc->pc_gs32p = &gdt[NGDT * cpu + GUGS32_SEL];
334 pc->pc_ldt = (struct system_segment_descriptor *)&gdt[NGDT * cpu +
335 GUSERLDT_SEL];
336 /* See comment in pmap_bootstrap(). */
337 pc->pc_pcid_next = PMAP_PCID_KERN + 2;
338 pc->pc_pcid_gen = 1;
339 common_tss[cpu].tss_rsp0 = 0;
340
341 /* Save the per-cpu pointer for use by the NMI handler. */
342 np = ((struct nmi_pcpu *) &nmi_stack[PAGE_SIZE]) - 1;
343 np->np_pcpu = (register_t) pc;
344
345 /* Save the per-cpu pointer for use by the MC# handler. */
346 np = ((struct nmi_pcpu *) &mce_stack[PAGE_SIZE]) - 1;
347 np->np_pcpu = (register_t) pc;
348
349 /* Save the per-cpu pointer for use by the DB# handler. */
350 np = ((struct nmi_pcpu *) &dbg_stack[PAGE_SIZE]) - 1;
351 np->np_pcpu = (register_t) pc;
352
353 wrmsr(MSR_FSBASE, 0); /* User value */
354 wrmsr(MSR_GSBASE, (u_int64_t)pc);
355 wrmsr(MSR_KGSBASE, (u_int64_t)pc); /* XXX User value while we're in the kernel */
356 fix_cpuid();
357
358 lidt(&r_idt);
359
360 gsel_tss = GSEL(GPROC0_SEL, SEL_KPL);
361 ltr(gsel_tss);
362
363 /*
364 * Set to a known state:
365 * Set by mpboot.s: CR0_PG, CR0_PE
366 * Set by cpu_setregs: CR0_NE, CR0_MP, CR0_TS, CR0_WP, CR0_AM
367 */
368 cr0 = rcr0();
369 cr0 &= ~(CR0_CD | CR0_NW | CR0_EM);
370 load_cr0(cr0);
371
372 amd64_conf_fast_syscall();
373
374 /* signal our startup to the BSP. */
375 mp_naps++;
376
377 /* Spin until the BSP releases the AP's. */
378 while (atomic_load_acq_int(&aps_ready) == 0)
379 ia32_pause();
380
381 init_secondary_tail();
382 }
383
384 /*******************************************************************
385 * local functions and data
386 */
387
388 /*
389 * start each AP in our list
390 */
391 int
392 native_start_all_aps(void)
393 {
394 u_int64_t *pt4, *pt3, *pt2;
395 u_int32_t mpbioswarmvec;
396 int apic_id, cpu, i;
397 u_char mpbiosreason;
398
399 mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN);
400
401 /* copy the AP 1st level boot code */
402 bcopy(mptramp_start, (void *)PHYS_TO_DMAP(boot_address), bootMP_size);
403
404 /* Locate the page tables, they'll be below the trampoline */
405 pt4 = (uint64_t *)PHYS_TO_DMAP(mptramp_pagetables);
406 pt3 = pt4 + (PAGE_SIZE) / sizeof(u_int64_t);
407 pt2 = pt3 + (PAGE_SIZE) / sizeof(u_int64_t);
408
409 /* Create the initial 1GB replicated page tables */
410 for (i = 0; i < 512; i++) {
411 /* Each slot of the level 4 pages points to the same level 3 page */
412 pt4[i] = (u_int64_t)(uintptr_t)(mptramp_pagetables + PAGE_SIZE);
413 pt4[i] |= PG_V | PG_RW | PG_U;
414
415 /* Each slot of the level 3 pages points to the same level 2 page */
416 pt3[i] = (u_int64_t)(uintptr_t)(mptramp_pagetables + (2 * PAGE_SIZE));
417 pt3[i] |= PG_V | PG_RW | PG_U;
418
419 /* The level 2 page slots are mapped with 2MB pages for 1GB. */
420 pt2[i] = i * (2 * 1024 * 1024);
421 pt2[i] |= PG_V | PG_RW | PG_PS | PG_U;
422 }
423
424 /* save the current value of the warm-start vector */
425 mpbioswarmvec = *((u_int32_t *) WARMBOOT_OFF);
426 outb(CMOS_REG, BIOS_RESET);
427 mpbiosreason = inb(CMOS_DATA);
428
429 /* setup a vector to our boot code */
430 *((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET;
431 *((volatile u_short *) WARMBOOT_SEG) = (boot_address >> 4);
432 outb(CMOS_REG, BIOS_RESET);
433 outb(CMOS_DATA, BIOS_WARM); /* 'warm-start' */
434
435 /* start each AP */
436 for (cpu = 1; cpu < mp_ncpus; cpu++) {
437 apic_id = cpu_apic_ids[cpu];
438
439 /* allocate and set up an idle stack data page */
440 bootstacks[cpu] = (void *)kmem_malloc(kstack_pages * PAGE_SIZE,
441 M_WAITOK | M_ZERO);
442 doublefault_stack = (char *)kmem_malloc(PAGE_SIZE, M_WAITOK |
443 M_ZERO);
444 mce_stack = (char *)kmem_malloc(PAGE_SIZE, M_WAITOK | M_ZERO);
445 nmi_stack = (char *)kmem_malloc(PAGE_SIZE, M_WAITOK | M_ZERO);
446 dbg_stack = (char *)kmem_malloc(PAGE_SIZE, M_WAITOK | M_ZERO);
447 dpcpu = (void *)kmem_malloc(DPCPU_SIZE, M_WAITOK | M_ZERO);
448
449 bootSTK = (char *)bootstacks[cpu] + kstack_pages * PAGE_SIZE - 8;
450 bootAP = cpu;
451
452 /* attempt to start the Application Processor */
453 if (!start_ap(apic_id)) {
454 /* restore the warmstart vector */
455 *(u_int32_t *) WARMBOOT_OFF = mpbioswarmvec;
456 panic("AP #%d (PHY# %d) failed!", cpu, apic_id);
457 }
458
459 CPU_SET(cpu, &all_cpus); /* record AP in CPU map */
460 }
461
462 /* restore the warmstart vector */
463 *(u_int32_t *) WARMBOOT_OFF = mpbioswarmvec;
464
465 outb(CMOS_REG, BIOS_RESET);
466 outb(CMOS_DATA, mpbiosreason);
467
468 /* number of APs actually started */
469 return mp_naps;
470 }
471
472
473 /*
474 * This function starts the AP (application processor) identified
475 * by the APIC ID 'physicalCpu'. It does quite a "song and dance"
476 * to accomplish this. This is necessary because of the nuances
477 * of the different hardware we might encounter. It isn't pretty,
478 * but it seems to work.
479 */
480 static int
481 start_ap(int apic_id)
482 {
483 int vector, ms;
484 int cpus;
485
486 /* calculate the vector */
487 vector = (boot_address >> 12) & 0xff;
488
489 /* used as a watchpoint to signal AP startup */
490 cpus = mp_naps;
491
492 ipi_startup(apic_id, vector);
493
494 /* Wait up to 5 seconds for it to start. */
495 for (ms = 0; ms < 5000; ms++) {
496 if (mp_naps > cpus)
497 return 1; /* return SUCCESS */
498 DELAY(1000);
499 }
500 return 0; /* return FAILURE */
501 }
502
503 void
504 invltlb_invpcid_handler(void)
505 {
506 struct invpcid_descr d;
507 uint32_t generation;
508
509 #ifdef COUNT_XINVLTLB_HITS
510 xhits_gbl[PCPU_GET(cpuid)]++;
511 #endif /* COUNT_XINVLTLB_HITS */
512 #ifdef COUNT_IPIS
513 (*ipi_invltlb_counts[PCPU_GET(cpuid)])++;
514 #endif /* COUNT_IPIS */
515
516 generation = smp_tlb_generation;
517 d.pcid = smp_tlb_pmap->pm_pcids[PCPU_GET(cpuid)].pm_pcid;
518 d.pad = 0;
519 d.addr = 0;
520 invpcid(&d, smp_tlb_pmap == kernel_pmap ? INVPCID_CTXGLOB :
521 INVPCID_CTX);
522 PCPU_SET(smp_tlb_done, generation);
523 }
524
525 void
526 invltlb_invpcid_pti_handler(void)
527 {
528 struct invpcid_descr d;
529 uint32_t generation;
530
531 #ifdef COUNT_XINVLTLB_HITS
532 xhits_gbl[PCPU_GET(cpuid)]++;
533 #endif /* COUNT_XINVLTLB_HITS */
534 #ifdef COUNT_IPIS
535 (*ipi_invltlb_counts[PCPU_GET(cpuid)])++;
536 #endif /* COUNT_IPIS */
537
538 generation = smp_tlb_generation;
539 d.pcid = smp_tlb_pmap->pm_pcids[PCPU_GET(cpuid)].pm_pcid;
540 d.pad = 0;
541 d.addr = 0;
542 if (smp_tlb_pmap == kernel_pmap) {
543 /*
544 * This invalidation actually needs to clear kernel
545 * mappings from the TLB in the current pmap, but
546 * since we were asked for the flush in the kernel
547 * pmap, achieve it by performing global flush.
548 */
549 invpcid(&d, INVPCID_CTXGLOB);
550 } else {
551 invpcid(&d, INVPCID_CTX);
552 d.pcid |= PMAP_PCID_USER_PT;
553 invpcid(&d, INVPCID_CTX);
554 }
555 PCPU_SET(smp_tlb_done, generation);
556 }
557
558 void
559 invltlb_pcid_handler(void)
560 {
561 uint64_t kcr3, ucr3;
562 uint32_t generation, pcid;
563
564 #ifdef COUNT_XINVLTLB_HITS
565 xhits_gbl[PCPU_GET(cpuid)]++;
566 #endif /* COUNT_XINVLTLB_HITS */
567 #ifdef COUNT_IPIS
568 (*ipi_invltlb_counts[PCPU_GET(cpuid)])++;
569 #endif /* COUNT_IPIS */
570
571 generation = smp_tlb_generation; /* Overlap with serialization */
572 if (smp_tlb_pmap == kernel_pmap) {
573 invltlb_glob();
574 } else {
575 /*
576 * The current pmap might not be equal to
577 * smp_tlb_pmap. The clearing of the pm_gen in
578 * pmap_invalidate_all() takes care of TLB
579 * invalidation when switching to the pmap on this
580 * CPU.
581 */
582 if (PCPU_GET(curpmap) == smp_tlb_pmap) {
583 pcid = smp_tlb_pmap->pm_pcids[PCPU_GET(cpuid)].pm_pcid;
584 kcr3 = smp_tlb_pmap->pm_cr3 | pcid;
585 ucr3 = smp_tlb_pmap->pm_ucr3;
586 if (ucr3 != PMAP_NO_CR3) {
587 ucr3 |= PMAP_PCID_USER_PT | pcid;
588 pmap_pti_pcid_invalidate(ucr3, kcr3);
589 } else
590 load_cr3(kcr3);
591 }
592 }
593 PCPU_SET(smp_tlb_done, generation);
594 }
595
596 void
597 invlpg_invpcid_handler(void)
598 {
599 struct invpcid_descr d;
600 uint32_t generation;
601
602 #ifdef COUNT_XINVLTLB_HITS
603 xhits_pg[PCPU_GET(cpuid)]++;
604 #endif /* COUNT_XINVLTLB_HITS */
605 #ifdef COUNT_IPIS
606 (*ipi_invlpg_counts[PCPU_GET(cpuid)])++;
607 #endif /* COUNT_IPIS */
608
609 generation = smp_tlb_generation; /* Overlap with serialization */
610 invlpg(smp_tlb_addr1);
611 if (smp_tlb_pmap->pm_ucr3 != PMAP_NO_CR3) {
612 d.pcid = smp_tlb_pmap->pm_pcids[PCPU_GET(cpuid)].pm_pcid |
613 PMAP_PCID_USER_PT;
614 d.pad = 0;
615 d.addr = smp_tlb_addr1;
616 invpcid(&d, INVPCID_ADDR);
617 }
618 PCPU_SET(smp_tlb_done, generation);
619 }
620
621 void
622 invlpg_pcid_handler(void)
623 {
624 uint64_t kcr3, ucr3;
625 uint32_t generation;
626 uint32_t pcid;
627
628 #ifdef COUNT_XINVLTLB_HITS
629 xhits_pg[PCPU_GET(cpuid)]++;
630 #endif /* COUNT_XINVLTLB_HITS */
631 #ifdef COUNT_IPIS
632 (*ipi_invlpg_counts[PCPU_GET(cpuid)])++;
633 #endif /* COUNT_IPIS */
634
635 generation = smp_tlb_generation; /* Overlap with serialization */
636 invlpg(smp_tlb_addr1);
637 if (smp_tlb_pmap == PCPU_GET(curpmap) &&
638 (ucr3 = smp_tlb_pmap->pm_ucr3) != PMAP_NO_CR3) {
639 pcid = smp_tlb_pmap->pm_pcids[PCPU_GET(cpuid)].pm_pcid;
640 kcr3 = smp_tlb_pmap->pm_cr3 | pcid | CR3_PCID_SAVE;
641 ucr3 |= pcid | PMAP_PCID_USER_PT | CR3_PCID_SAVE;
642 pmap_pti_pcid_invlpg(ucr3, kcr3, smp_tlb_addr1);
643 }
644 PCPU_SET(smp_tlb_done, generation);
645 }
646
647 void
648 invlrng_invpcid_handler(void)
649 {
650 struct invpcid_descr d;
651 vm_offset_t addr, addr2;
652 uint32_t generation;
653
654 #ifdef COUNT_XINVLTLB_HITS
655 xhits_rng[PCPU_GET(cpuid)]++;
656 #endif /* COUNT_XINVLTLB_HITS */
657 #ifdef COUNT_IPIS
658 (*ipi_invlrng_counts[PCPU_GET(cpuid)])++;
659 #endif /* COUNT_IPIS */
660
661 addr = smp_tlb_addr1;
662 addr2 = smp_tlb_addr2;
663 generation = smp_tlb_generation; /* Overlap with serialization */
664 do {
665 invlpg(addr);
666 addr += PAGE_SIZE;
667 } while (addr < addr2);
668 if (smp_tlb_pmap->pm_ucr3 != PMAP_NO_CR3) {
669 d.pcid = smp_tlb_pmap->pm_pcids[PCPU_GET(cpuid)].pm_pcid |
670 PMAP_PCID_USER_PT;
671 d.pad = 0;
672 d.addr = smp_tlb_addr1;
673 do {
674 invpcid(&d, INVPCID_ADDR);
675 d.addr += PAGE_SIZE;
676 } while (d.addr < addr2);
677 }
678 PCPU_SET(smp_tlb_done, generation);
679 }
680
681 void
682 invlrng_pcid_handler(void)
683 {
684 vm_offset_t addr, addr2;
685 uint64_t kcr3, ucr3;
686 uint32_t generation;
687 uint32_t pcid;
688
689 #ifdef COUNT_XINVLTLB_HITS
690 xhits_rng[PCPU_GET(cpuid)]++;
691 #endif /* COUNT_XINVLTLB_HITS */
692 #ifdef COUNT_IPIS
693 (*ipi_invlrng_counts[PCPU_GET(cpuid)])++;
694 #endif /* COUNT_IPIS */
695
696 addr = smp_tlb_addr1;
697 addr2 = smp_tlb_addr2;
698 generation = smp_tlb_generation; /* Overlap with serialization */
699 do {
700 invlpg(addr);
701 addr += PAGE_SIZE;
702 } while (addr < addr2);
703 if (smp_tlb_pmap == PCPU_GET(curpmap) &&
704 (ucr3 = smp_tlb_pmap->pm_ucr3) != PMAP_NO_CR3) {
705 pcid = smp_tlb_pmap->pm_pcids[PCPU_GET(cpuid)].pm_pcid;
706 kcr3 = smp_tlb_pmap->pm_cr3 | pcid | CR3_PCID_SAVE;
707 ucr3 |= pcid | PMAP_PCID_USER_PT | CR3_PCID_SAVE;
708 pmap_pti_pcid_invlrng(ucr3, kcr3, smp_tlb_addr1, addr2);
709 }
710 PCPU_SET(smp_tlb_done, generation);
711 }
Cache object: 03f1925ec627f7fa2a0ba60daf31a704
|