xref: /freebsd-src/sys/i386/i386/mp_machdep.c (revision 2eb4d8dc723da3cf7d735a3226ae49da4c8c5dbc)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1996, by Steve Passe
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. The name of the developer may NOT be used to endorse or promote products
13  *    derived from this software without specific prior written permission.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_acpi.h"
32 #include "opt_apic.h"
33 #include "opt_cpu.h"
34 #include "opt_kstack_pages.h"
35 #include "opt_pmap.h"
36 #include "opt_sched.h"
37 #include "opt_smp.h"
38 
39 #if !defined(lint)
40 #if !defined(SMP)
41 #error How did you get here?
42 #endif
43 
44 #ifndef DEV_APIC
45 #error The apic device is required for SMP, add "device apic" to your config file.
46 #endif
47 #endif /* not lint */
48 
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/bus.h>
52 #include <sys/cons.h>	/* cngetc() */
53 #include <sys/cpuset.h>
54 #ifdef GPROF
55 #include <sys/gmon.h>
56 #endif
57 #include <sys/kdb.h>
58 #include <sys/kernel.h>
59 #include <sys/ktr.h>
60 #include <sys/lock.h>
61 #include <sys/malloc.h>
62 #include <sys/memrange.h>
63 #include <sys/mutex.h>
64 #include <sys/pcpu.h>
65 #include <sys/proc.h>
66 #include <sys/sched.h>
67 #include <sys/smp.h>
68 #include <sys/sysctl.h>
69 
70 #include <vm/vm.h>
71 #include <vm/vm_param.h>
72 #include <vm/pmap.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_extern.h>
75 
76 #include <x86/apicreg.h>
77 #include <machine/clock.h>
78 #include <machine/cpu.h>
79 #include <machine/cputypes.h>
80 #include <x86/mca.h>
81 #include <machine/md_var.h>
82 #include <machine/pcb.h>
83 #include <machine/psl.h>
84 #include <machine/smp.h>
85 #include <machine/specialreg.h>
86 #include <x86/ucode.h>
87 
88 #ifdef DEV_ACPI
89 #include <contrib/dev/acpica/include/acpi.h>
90 #include <dev/acpica/acpivar.h>
91 #endif
92 
93 #define WARMBOOT_TARGET		0
94 #define WARMBOOT_OFF		(PMAP_MAP_LOW + 0x0467)
95 #define WARMBOOT_SEG		(PMAP_MAP_LOW + 0x0469)
96 
97 #define CMOS_REG		(0x70)
98 #define CMOS_DATA		(0x71)
99 #define BIOS_RESET		(0x0f)
100 #define BIOS_WARM		(0x0a)
101 
102 /*
103  * this code MUST be enabled here and in mpboot.s.
104  * it follows the very early stages of AP boot by placing values in CMOS ram.
105  * it NORMALLY will never be needed and thus the primitive method for enabling.
106  *
107 #define CHECK_POINTS
108  */
109 
110 #if defined(CHECK_POINTS)
111 #define CHECK_READ(A)	 (outb(CMOS_REG, (A)), inb(CMOS_DATA))
112 #define CHECK_WRITE(A,D) (outb(CMOS_REG, (A)), outb(CMOS_DATA, (D)))
113 
114 #define CHECK_INIT(D);				\
115 	CHECK_WRITE(0x34, (D));			\
116 	CHECK_WRITE(0x35, (D));			\
117 	CHECK_WRITE(0x36, (D));			\
118 	CHECK_WRITE(0x37, (D));			\
119 	CHECK_WRITE(0x38, (D));			\
120 	CHECK_WRITE(0x39, (D));
121 
122 #define CHECK_PRINT(S);				\
123 	printf("%s: %d, %d, %d, %d, %d, %d\n",	\
124 	   (S),					\
125 	   CHECK_READ(0x34),			\
126 	   CHECK_READ(0x35),			\
127 	   CHECK_READ(0x36),			\
128 	   CHECK_READ(0x37),			\
129 	   CHECK_READ(0x38),			\
130 	   CHECK_READ(0x39));
131 
132 #else				/* CHECK_POINTS */
133 
134 #define CHECK_INIT(D)
135 #define CHECK_PRINT(S)
136 #define CHECK_WRITE(A, D)
137 
138 #endif				/* CHECK_POINTS */
139 
140 /*
141  * Local data and functions.
142  */
143 
144 static void	install_ap_tramp(void);
145 static int	start_all_aps(void);
146 static int	start_ap(int apic_id);
147 
148 static char *ap_copyout_buf;
149 static char *ap_tramp_stack_base;
150 /*
151  * Initialize the IPI handlers and start up the AP's.
152  */
153 void
154 cpu_mp_start(void)
155 {
156 	int i;
157 
158 	/* Initialize the logical ID to APIC ID table. */
159 	for (i = 0; i < MAXCPU; i++) {
160 		cpu_apic_ids[i] = -1;
161 	}
162 
163 	/* Install an inter-CPU IPI for TLB invalidation */
164 	setidt(IPI_INVLTLB, IDTVEC(invltlb),
165 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
166 	setidt(IPI_INVLPG, IDTVEC(invlpg),
167 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
168 	setidt(IPI_INVLRNG, IDTVEC(invlrng),
169 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
170 
171 	/* Install an inter-CPU IPI for cache invalidation. */
172 	setidt(IPI_INVLCACHE, IDTVEC(invlcache),
173 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
174 
175 	/* Install an inter-CPU IPI for all-CPU rendezvous */
176 	setidt(IPI_RENDEZVOUS, IDTVEC(rendezvous),
177 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
178 
179 	/* Install generic inter-CPU IPI handler */
180 	setidt(IPI_BITMAP_VECTOR, IDTVEC(ipi_intr_bitmap_handler),
181 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
182 
183 	/* Install an inter-CPU IPI for CPU stop/restart */
184 	setidt(IPI_STOP, IDTVEC(cpustop),
185 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
186 
187 	/* Install an inter-CPU IPI for CPU suspend/resume */
188 	setidt(IPI_SUSPEND, IDTVEC(cpususpend),
189 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
190 
191 	/* Install an IPI for calling delayed SWI */
192 	setidt(IPI_SWI, IDTVEC(ipi_swi),
193 	       SDT_SYS386IGT, SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
194 
195 	/* Set boot_cpu_id if needed. */
196 	if (boot_cpu_id == -1) {
197 		boot_cpu_id = PCPU_GET(apic_id);
198 		cpu_info[boot_cpu_id].cpu_bsp = 1;
199 	} else
200 		KASSERT(boot_cpu_id == PCPU_GET(apic_id),
201 		    ("BSP's APIC ID doesn't match boot_cpu_id"));
202 
203 	/* Probe logical/physical core configuration. */
204 	topo_probe();
205 
206 	assign_cpu_ids();
207 
208 	/* Start each Application Processor */
209 	start_all_aps();
210 
211 	set_interrupt_apic_ids();
212 
213 #if defined(DEV_ACPI) && MAXMEMDOM > 1
214 	acpi_pxm_set_cpu_locality();
215 #endif
216 }
217 
218 /*
219  * AP CPU's call this to initialize themselves.
220  */
221 void
222 init_secondary(void)
223 {
224 	struct pcpu *pc;
225 	struct i386tss *common_tssp;
226 	struct region_descriptor r_gdt, r_idt;
227 	int gsel_tss, myid, x;
228 	u_int cr0;
229 
230 	/* bootAP is set in start_ap() to our ID. */
231 	myid = bootAP;
232 
233 	/* Update microcode before doing anything else. */
234 	ucode_load_ap(myid);
235 
236 	/* Get per-cpu data */
237 	pc = &__pcpu[myid];
238 
239 	/* prime data page for it to use */
240 	pcpu_init(pc, myid, sizeof(struct pcpu));
241 	dpcpu_init(dpcpu, myid);
242 	pc->pc_apic_id = cpu_apic_ids[myid];
243 	pc->pc_prvspace = pc;
244 	pc->pc_curthread = 0;
245 	pc->pc_common_tssp = common_tssp = &(__pcpu[0].pc_common_tssp)[myid];
246 
247 	fix_cpuid();
248 
249 	gdt_segs[GPRIV_SEL].ssd_base = (int)pc;
250 	gdt_segs[GPROC0_SEL].ssd_base = (int)common_tssp;
251 	gdt_segs[GLDT_SEL].ssd_base = (int)ldt;
252 
253 	for (x = 0; x < NGDT; x++) {
254 		ssdtosd(&gdt_segs[x], &gdt[myid * NGDT + x].sd);
255 	}
256 
257 	r_gdt.rd_limit = NGDT * sizeof(gdt[0]) - 1;
258 	r_gdt.rd_base = (int) &gdt[myid * NGDT];
259 	lgdt(&r_gdt);			/* does magic intra-segment return */
260 
261 	r_idt.rd_limit = sizeof(struct gate_descriptor) * NIDT - 1;
262 	r_idt.rd_base = (int)idt;
263 	lidt(&r_idt);
264 
265 	lldt(_default_ldt);
266 	PCPU_SET(currentldt, _default_ldt);
267 
268 	PCPU_SET(trampstk, (uintptr_t)ap_tramp_stack_base + TRAMP_STACK_SZ -
269 	    VM86_STACK_SPACE);
270 
271 	gsel_tss = GSEL(GPROC0_SEL, SEL_KPL);
272 	gdt[myid * NGDT + GPROC0_SEL].sd.sd_type = SDT_SYS386TSS;
273 	common_tssp->tss_esp0 = PCPU_GET(trampstk);
274 	common_tssp->tss_ss0 = GSEL(GDATA_SEL, SEL_KPL);
275 	common_tssp->tss_ioopt = sizeof(struct i386tss) << 16;
276 	PCPU_SET(tss_gdt, &gdt[myid * NGDT + GPROC0_SEL].sd);
277 	PCPU_SET(common_tssd, *PCPU_GET(tss_gdt));
278 	ltr(gsel_tss);
279 
280 	PCPU_SET(fsgs_gdt, &gdt[myid * NGDT + GUFS_SEL].sd);
281 	PCPU_SET(copyout_buf, ap_copyout_buf);
282 
283 	/*
284 	 * Set to a known state:
285 	 * Set by mpboot.s: CR0_PG, CR0_PE
286 	 * Set by cpu_setregs: CR0_NE, CR0_MP, CR0_TS, CR0_WP, CR0_AM
287 	 */
288 	cr0 = rcr0();
289 	cr0 &= ~(CR0_CD | CR0_NW | CR0_EM);
290 	load_cr0(cr0);
291 	CHECK_WRITE(0x38, 5);
292 
293 	/* signal our startup to the BSP. */
294 	mp_naps++;
295 	CHECK_WRITE(0x39, 6);
296 
297 	/* Spin until the BSP releases the AP's. */
298 	while (atomic_load_acq_int(&aps_ready) == 0)
299 		ia32_pause();
300 
301 	/* BSP may have changed PTD while we were waiting */
302 	invltlb();
303 
304 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
305 	lidt(&r_idt);
306 #endif
307 
308 	init_secondary_tail();
309 }
310 
311 /*
312  * start each AP in our list
313  */
314 #define TMPMAP_START 1
315 static int
316 start_all_aps(void)
317 {
318 	u_char mpbiosreason;
319 	u_int32_t mpbioswarmvec;
320 	int apic_id, cpu;
321 
322 	mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN);
323 
324 	pmap_remap_lower(true);
325 
326 	/* install the AP 1st level boot code */
327 	install_ap_tramp();
328 
329 	/* save the current value of the warm-start vector */
330 	mpbioswarmvec = *((u_int32_t *) WARMBOOT_OFF);
331 	outb(CMOS_REG, BIOS_RESET);
332 	mpbiosreason = inb(CMOS_DATA);
333 
334 	/* take advantage of the P==V mapping for PTD[0] for AP boot */
335 
336 	/* start each AP */
337 	for (cpu = 1; cpu < mp_ncpus; cpu++) {
338 		apic_id = cpu_apic_ids[cpu];
339 
340 		/* allocate and set up a boot stack data page */
341 		bootstacks[cpu] = (char *)kmem_malloc(kstack_pages * PAGE_SIZE,
342 		    M_WAITOK | M_ZERO);
343 		dpcpu = (void *)kmem_malloc(DPCPU_SIZE, M_WAITOK | M_ZERO);
344 		/* setup a vector to our boot code */
345 		*((volatile u_short *) WARMBOOT_OFF) = WARMBOOT_TARGET;
346 		*((volatile u_short *) WARMBOOT_SEG) = (boot_address >> 4);
347 		outb(CMOS_REG, BIOS_RESET);
348 		outb(CMOS_DATA, BIOS_WARM);	/* 'warm-start' */
349 
350 		bootSTK = (char *)bootstacks[cpu] + kstack_pages *
351 		    PAGE_SIZE - 4;
352 		bootAP = cpu;
353 
354 		ap_tramp_stack_base = pmap_trm_alloc(TRAMP_STACK_SZ, M_NOWAIT);
355 		ap_copyout_buf = pmap_trm_alloc(TRAMP_COPYOUT_SZ, M_NOWAIT);
356 
357 		/* attempt to start the Application Processor */
358 		CHECK_INIT(99);	/* setup checkpoints */
359 		if (!start_ap(apic_id)) {
360 			printf("AP #%d (PHY# %d) failed!\n", cpu, apic_id);
361 			CHECK_PRINT("trace");	/* show checkpoints */
362 			/* better panic as the AP may be running loose */
363 			printf("panic y/n? [y] ");
364 			if (cngetc() != 'n')
365 				panic("bye-bye");
366 		}
367 		CHECK_PRINT("trace");		/* show checkpoints */
368 
369 		CPU_SET(cpu, &all_cpus);	/* record AP in CPU map */
370 	}
371 
372 	pmap_remap_lower(false);
373 
374 	/* restore the warmstart vector */
375 	*(u_int32_t *) WARMBOOT_OFF = mpbioswarmvec;
376 
377 	outb(CMOS_REG, BIOS_RESET);
378 	outb(CMOS_DATA, mpbiosreason);
379 
380 	/* number of APs actually started */
381 	return mp_naps;
382 }
383 
384 /*
385  * load the 1st level AP boot code into base memory.
386  */
387 
388 /* targets for relocation */
389 extern void bigJump(void);
390 extern void bootCodeSeg(void);
391 extern void bootDataSeg(void);
392 extern void MPentry(void);
393 extern u_int MP_GDT;
394 extern u_int mp_gdtbase;
395 
396 static void
397 install_ap_tramp(void)
398 {
399 	int     x;
400 	int     size = *(int *) ((u_long) & bootMP_size);
401 	vm_offset_t va = boot_address;
402 	u_char *src = (u_char *) ((u_long) bootMP);
403 	u_char *dst = (u_char *) va;
404 	u_int   boot_base = (u_int) bootMP;
405 	u_int8_t *dst8;
406 	u_int16_t *dst16;
407 	u_int32_t *dst32;
408 
409 	KASSERT (size <= PAGE_SIZE,
410 	    ("'size' do not fit into PAGE_SIZE, as expected."));
411 	pmap_kenter(va, boot_address);
412 	pmap_invalidate_page (kernel_pmap, va);
413 	for (x = 0; x < size; ++x)
414 		*dst++ = *src++;
415 
416 	/*
417 	 * modify addresses in code we just moved to basemem. unfortunately we
418 	 * need fairly detailed info about mpboot.s for this to work.  changes
419 	 * to mpboot.s might require changes here.
420 	 */
421 
422 	/* boot code is located in KERNEL space */
423 	dst = (u_char *) va;
424 
425 	/* modify the lgdt arg */
426 	dst32 = (u_int32_t *) (dst + ((u_int) & mp_gdtbase - boot_base));
427 	*dst32 = boot_address + ((u_int) & MP_GDT - boot_base);
428 
429 	/* modify the ljmp target for MPentry() */
430 	dst32 = (u_int32_t *) (dst + ((u_int) bigJump - boot_base) + 1);
431 	*dst32 = (u_int)MPentry;
432 
433 	/* modify the target for boot code segment */
434 	dst16 = (u_int16_t *) (dst + ((u_int) bootCodeSeg - boot_base));
435 	dst8 = (u_int8_t *) (dst16 + 1);
436 	*dst16 = (u_int) boot_address & 0xffff;
437 	*dst8 = ((u_int) boot_address >> 16) & 0xff;
438 
439 	/* modify the target for boot data segment */
440 	dst16 = (u_int16_t *) (dst + ((u_int) bootDataSeg - boot_base));
441 	dst8 = (u_int8_t *) (dst16 + 1);
442 	*dst16 = (u_int) boot_address & 0xffff;
443 	*dst8 = ((u_int) boot_address >> 16) & 0xff;
444 }
445 
446 /*
447  * This function starts the AP (application processor) identified
448  * by the APIC ID 'physicalCpu'.  It does quite a "song and dance"
449  * to accomplish this.  This is necessary because of the nuances
450  * of the different hardware we might encounter.  It isn't pretty,
451  * but it seems to work.
452  */
453 static int
454 start_ap(int apic_id)
455 {
456 	int vector, ms;
457 	int cpus;
458 
459 	/* calculate the vector */
460 	vector = (boot_address >> 12) & 0xff;
461 
462 	/* used as a watchpoint to signal AP startup */
463 	cpus = mp_naps;
464 
465 	ipi_startup(apic_id, vector);
466 
467 	/* Wait up to 5 seconds for it to start. */
468 	for (ms = 0; ms < 5000; ms++) {
469 		if (mp_naps > cpus)
470 			return 1;	/* return SUCCESS */
471 		DELAY(1000);
472 	}
473 	return 0;		/* return FAILURE */
474 }
475 
476 /*
477  * Flush the TLB on other CPU's
478  */
479 
480 /* Variables needed for SMP tlb shootdown. */
481 vm_offset_t smp_tlb_addr1, smp_tlb_addr2;
482 pmap_t smp_tlb_pmap;
483 volatile uint32_t smp_tlb_generation;
484 
485 /*
486  * Used by pmap to request cache or TLB invalidation on local and
487  * remote processors.  Mask provides the set of remote CPUs which are
488  * to be signalled with the invalidation IPI.  Vector specifies which
489  * invalidation IPI is used.  As an optimization, the curcpu_cb
490  * callback is invoked on the calling CPU while waiting for remote
491  * CPUs to complete the operation.
492  *
493  * The callback function is called unconditionally on the caller's
494  * underlying processor, even when this processor is not set in the
495  * mask.  So, the callback function must be prepared to handle such
496  * spurious invocations.
497  */
498 static void
499 smp_targeted_tlb_shootdown(cpuset_t mask, u_int vector, pmap_t pmap,
500     vm_offset_t addr1, vm_offset_t addr2, smp_invl_cb_t curcpu_cb)
501 {
502 	cpuset_t other_cpus;
503 	volatile uint32_t *p_cpudone;
504 	uint32_t generation;
505 	int cpu;
506 
507 	/*
508 	 * It is not necessary to signal other CPUs while booting or
509 	 * when in the debugger.
510 	 */
511 	if (kdb_active || KERNEL_PANICKED() || !smp_started) {
512 		curcpu_cb(pmap, addr1, addr2);
513 		return;
514 	}
515 
516 	sched_pin();
517 
518 	/*
519 	 * Check for other cpus.  Return if none.
520 	 */
521 	if (CPU_ISFULLSET(&mask)) {
522 		if (mp_ncpus <= 1)
523 			goto nospinexit;
524 	} else {
525 		CPU_CLR(PCPU_GET(cpuid), &mask);
526 		if (CPU_EMPTY(&mask))
527 			goto nospinexit;
528 	}
529 
530 	KASSERT((read_eflags() & PSL_I) != 0,
531 	    ("smp_targeted_tlb_shootdown: interrupts disabled"));
532 	mtx_lock_spin(&smp_ipi_mtx);
533 	smp_tlb_addr1 = addr1;
534 	smp_tlb_addr2 = addr2;
535 	smp_tlb_pmap = pmap;
536 	generation = ++smp_tlb_generation;
537 	if (CPU_ISFULLSET(&mask)) {
538 		ipi_all_but_self(vector);
539 		other_cpus = all_cpus;
540 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
541 	} else {
542 		other_cpus = mask;
543 		ipi_selected(mask, vector);
544 	}
545 	curcpu_cb(pmap, addr1, addr2);
546 	while ((cpu = CPU_FFS(&other_cpus)) != 0) {
547 		cpu--;
548 		CPU_CLR(cpu, &other_cpus);
549 		p_cpudone = &cpuid_to_pcpu[cpu]->pc_smp_tlb_done;
550 		while (*p_cpudone != generation)
551 			ia32_pause();
552 	}
553 	mtx_unlock_spin(&smp_ipi_mtx);
554 	sched_unpin();
555 	return;
556 
557 nospinexit:
558 	curcpu_cb(pmap, addr1, addr2);
559 	sched_unpin();
560 }
561 
562 void
563 smp_masked_invltlb(cpuset_t mask, pmap_t pmap, smp_invl_cb_t curcpu_cb)
564 {
565 	smp_targeted_tlb_shootdown(mask, IPI_INVLTLB, pmap, 0, 0, curcpu_cb);
566 #ifdef COUNT_XINVLTLB_HITS
567 	ipi_global++;
568 #endif
569 }
570 
571 void
572 smp_masked_invlpg(cpuset_t mask, vm_offset_t addr, pmap_t pmap,
573     smp_invl_cb_t curcpu_cb)
574 {
575 	smp_targeted_tlb_shootdown(mask, IPI_INVLPG, pmap, addr, 0, curcpu_cb);
576 #ifdef COUNT_XINVLTLB_HITS
577 	ipi_page++;
578 #endif
579 }
580 
581 void
582 smp_masked_invlpg_range(cpuset_t mask, vm_offset_t addr1, vm_offset_t addr2,
583     pmap_t pmap, smp_invl_cb_t curcpu_cb)
584 {
585 	smp_targeted_tlb_shootdown(mask, IPI_INVLRNG, pmap, addr1, addr2,
586 	    curcpu_cb);
587 #ifdef COUNT_XINVLTLB_HITS
588 	ipi_range++;
589 	ipi_range_size += (addr2 - addr1) / PAGE_SIZE;
590 #endif
591 }
592 
593 void
594 smp_cache_flush(smp_invl_cb_t curcpu_cb)
595 {
596 	smp_targeted_tlb_shootdown(all_cpus, IPI_INVLCACHE, NULL, 0, 0,
597 	    curcpu_cb);
598 }
599 
600 /*
601  * Handlers for TLB related IPIs
602  */
603 void
604 invltlb_handler(void)
605 {
606 	uint32_t generation;
607 
608 #ifdef COUNT_XINVLTLB_HITS
609 	xhits_gbl[PCPU_GET(cpuid)]++;
610 #endif /* COUNT_XINVLTLB_HITS */
611 #ifdef COUNT_IPIS
612 	(*ipi_invltlb_counts[PCPU_GET(cpuid)])++;
613 #endif /* COUNT_IPIS */
614 
615 	/*
616 	 * Reading the generation here allows greater parallelism
617 	 * since invalidating the TLB is a serializing operation.
618 	 */
619 	generation = smp_tlb_generation;
620 	if (smp_tlb_pmap == kernel_pmap)
621 		invltlb_glob();
622 	PCPU_SET(smp_tlb_done, generation);
623 }
624 
625 void
626 invlpg_handler(void)
627 {
628 	uint32_t generation;
629 
630 #ifdef COUNT_XINVLTLB_HITS
631 	xhits_pg[PCPU_GET(cpuid)]++;
632 #endif /* COUNT_XINVLTLB_HITS */
633 #ifdef COUNT_IPIS
634 	(*ipi_invlpg_counts[PCPU_GET(cpuid)])++;
635 #endif /* COUNT_IPIS */
636 
637 	generation = smp_tlb_generation;	/* Overlap with serialization */
638 	if (smp_tlb_pmap == kernel_pmap)
639 		invlpg(smp_tlb_addr1);
640 	PCPU_SET(smp_tlb_done, generation);
641 }
642 
643 void
644 invlrng_handler(void)
645 {
646 	vm_offset_t addr, addr2;
647 	uint32_t generation;
648 
649 #ifdef COUNT_XINVLTLB_HITS
650 	xhits_rng[PCPU_GET(cpuid)]++;
651 #endif /* COUNT_XINVLTLB_HITS */
652 #ifdef COUNT_IPIS
653 	(*ipi_invlrng_counts[PCPU_GET(cpuid)])++;
654 #endif /* COUNT_IPIS */
655 
656 	addr = smp_tlb_addr1;
657 	addr2 = smp_tlb_addr2;
658 	generation = smp_tlb_generation;	/* Overlap with serialization */
659 	if (smp_tlb_pmap == kernel_pmap) {
660 		do {
661 			invlpg(addr);
662 			addr += PAGE_SIZE;
663 		} while (addr < addr2);
664 	}
665 
666 	PCPU_SET(smp_tlb_done, generation);
667 }
668 
669 void
670 invlcache_handler(void)
671 {
672 	uint32_t generation;
673 
674 #ifdef COUNT_IPIS
675 	(*ipi_invlcache_counts[PCPU_GET(cpuid)])++;
676 #endif /* COUNT_IPIS */
677 
678 	/*
679 	 * Reading the generation here allows greater parallelism
680 	 * since wbinvd is a serializing instruction.  Without the
681 	 * temporary, we'd wait for wbinvd to complete, then the read
682 	 * would execute, then the dependent write, which must then
683 	 * complete before return from interrupt.
684 	 */
685 	generation = smp_tlb_generation;
686 	wbinvd();
687 	PCPU_SET(smp_tlb_done, generation);
688 }
689