2 * Copyright (c) 1982, 1986 The Regents of the University of California.
3 * Copyright (c) 1989, 1990 William Jolitz
4 * Copyright (c) 1994 John Dyson
5 * Copyright (c) 2008-2018 The DragonFly Project.
8 * This code is derived from software contributed to Berkeley by
9 * the Systems Programming Group of the University of Utah Computer
10 * Science Department, and William Jolitz.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. All advertising materials mentioning features or use of this software
21 * must display the following acknowledgement:
22 * This product includes software developed by the University of
23 * California, Berkeley and its contributors.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91
41 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
42 * $FreeBSD: src/sys/i386/i386/vm_machdep.c,v 1.132.2.9 2003/01/25 19:02:23 dillon Exp $
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/malloc.h>
50 #include <sys/interrupt.h>
51 #include <sys/vnode.h>
52 #include <sys/vmmeter.h>
53 #include <sys/kernel.h>
54 #include <sys/sysctl.h>
55 #include <sys/unistd.h>
58 #include <machine/clock.h>
59 #include <machine/cpu.h>
60 #include <machine/md_var.h>
61 #include <machine/smp.h>
62 #include <machine/pcb.h>
63 #include <machine/pcb_ext.h>
64 #include <machine/segments.h>
65 #include <machine/globaldata.h> /* npxthread */
66 #include <machine/vmm.h>
69 #include <vm/vm_param.h>
71 #include <vm/vm_kern.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_map.h>
74 #include <vm/vm_extern.h>
76 #include <sys/thread2.h>
77 #include <sys/mplock2.h>
79 #include <bus/isa/isa.h>
81 static void cpu_reset_real (void);
83 int spectre_mitigation
= -1;
86 * Finish a fork operation, with lwp lp2 nearly set up.
87 * Copy and update the pcb, set up the stack so that the child
88 * ready to run and return to user mode.
91 cpu_fork(struct lwp
*lp1
, struct lwp
*lp2
, int flags
)
96 if ((flags
& RFPROC
) == 0) {
97 if ((flags
& RFMEM
) == 0) {
99 * Unshare user LDT. > 1 test is MPSAFE. While
100 * it can potentially race a 2->1 transition, the
101 * worst that happens is that we do an unnecessary
104 struct pcb
*pcb1
= lp1
->lwp_thread
->td_pcb
;
105 struct pcb_ldt
*pcb_ldt
= pcb1
->pcb_ldt
;
107 if (pcb_ldt
&& pcb_ldt
->ldt_refcnt
> 1) {
108 pcb_ldt
= user_ldt_alloc(pcb1
,pcb_ldt
->ldt_len
);
110 pcb1
->pcb_ldt
= pcb_ldt
;
117 /* Ensure that lp1's pcb is up to date. */
118 if (mdcpu
->gd_npxthread
== lp1
->lwp_thread
)
119 npxsave(lp1
->lwp_thread
->td_savefpu
);
122 * Copy lp1's PCB. This really only applies to the
123 * debug registers and FP state, but its faster to just copy the
124 * whole thing. Because we only save the PCB at switchout time,
125 * the register state may not be current.
127 pcb2
= lp2
->lwp_thread
->td_pcb
;
128 *pcb2
= *lp1
->lwp_thread
->td_pcb
;
131 * Create a new fresh stack for the new process.
132 * Copy the trap frame for the return to user mode as if from a
133 * syscall. This copies the user mode register values.
135 * pcb_rsp must allocate an additional call-return pointer below
136 * the trap frame which will be restored by cpu_heavy_restore from
137 * PCB_RIP, and the thread's td_sp pointer must allocate an
138 * additonal two quadwords below the pcb_rsp call-return pointer to
139 * hold the LWKT restore function pointer and rflags.
141 * The LWKT restore function pointer must be set to cpu_heavy_restore,
142 * which is our standard heavy-weight process switch-in function.
143 * YYY eventually we should shortcut fork_return and fork_trampoline
144 * to use the LWKT restore function directly so we can get rid of
145 * all the extra crap we are setting up.
147 lp2
->lwp_md
.md_regs
= (struct trapframe
*)pcb2
- 1;
148 bcopy(lp1
->lwp_md
.md_regs
, lp2
->lwp_md
.md_regs
, sizeof(*lp2
->lwp_md
.md_regs
));
151 * Set registers for trampoline to user mode. Leave space for the
152 * return address on stack. These are the kernel mode register values.
154 * Set the new pmap CR3. If the new process uses isolated VM spaces,
155 * also set the isolated CR3.
157 pmap2
= vmspace_pmap(lp2
->lwp_proc
->p_vmspace
);
158 pcb2
->pcb_cr3
= vtophys(pmap2
->pm_pml4
);
159 if ((pcb2
->pcb_flags
& PCB_ISOMMU
) && pmap2
->pm_pmlpv_iso
) {
160 pcb2
->pcb_cr3_iso
= vtophys(pmap2
->pm_pml4_iso
);
162 pcb2
->pcb_flags
&= ~PCB_ISOMMU
;
163 pcb2
->pcb_cr3_iso
= 0;
168 * Per-process spectre mitigation (future)
170 pcb2
->pcb_flags
&= ~(PCB_IBRS1
| PCB_IBRS2
);
171 switch (spectre_mitigation
) {
173 pcb2
->pcb_flags
|= PCB_IBRS1
;
176 pcb2
->pcb_flags
|= PCB_IBRS2
;
183 pcb2
->pcb_rbx
= (unsigned long)fork_return
; /* fork_trampoline argument */
185 pcb2
->pcb_rsp
= (unsigned long)lp2
->lwp_md
.md_regs
- sizeof(void *);
186 pcb2
->pcb_r12
= (unsigned long)lp2
; /* fork_trampoline argument */
190 pcb2
->pcb_rip
= (unsigned long)fork_trampoline
;
191 lp2
->lwp_thread
->td_sp
= (char *)(pcb2
->pcb_rsp
- sizeof(void *));
192 *(u_int64_t
*)lp2
->lwp_thread
->td_sp
= PSL_USER
;
193 lp2
->lwp_thread
->td_sp
-= sizeof(void *);
194 *(void **)lp2
->lwp_thread
->td_sp
= (void *)cpu_heavy_restore
;
197 * pcb2->pcb_ldt: duplicated below, if necessary.
198 * pcb2->pcb_savefpu: cloned above.
199 * pcb2->pcb_flags: cloned above
200 * pcb2->pcb_onfault: cloned above (always NULL here).
201 * pcb2->pcb_onfault_sp:cloned above (dont care)
205 * XXX don't copy the i/o pages. this should probably be fixed.
207 pcb2
->pcb_ext
= NULL
;
209 /* Copy the LDT, if necessary. */
210 if (pcb2
->pcb_ldt
!= NULL
) {
212 atomic_add_int(&pcb2
->pcb_ldt
->ldt_refcnt
, 1);
214 pcb2
->pcb_ldt
= user_ldt_alloc(pcb2
,
215 pcb2
->pcb_ldt
->ldt_len
);
218 bcopy(&lp1
->lwp_thread
->td_tls
, &lp2
->lwp_thread
->td_tls
,
219 sizeof(lp2
->lwp_thread
->td_tls
));
221 * Now, cpu_switch() can schedule the new lwp.
222 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
223 * containing the return address when exiting cpu_switch.
224 * This will normally be to fork_trampoline(), which will have
225 * %rbx loaded with the new lwp's pointer. fork_trampoline()
226 * will set up a stack to call fork_return(lp, frame); to complete
227 * the return to user-mode.
232 * Prepare new lwp to return to the address specified in params.
235 cpu_prepare_lwp(struct lwp
*lp
, struct lwp_params
*params
)
237 struct trapframe
*regs
= lp
->lwp_md
.md_regs
;
238 void *bad_return
= NULL
;
241 regs
->tf_rip
= (long)params
->lwp_func
;
242 regs
->tf_rsp
= (long)params
->lwp_stack
;
243 /* Set up argument for function call */
244 regs
->tf_rdi
= (long)params
->lwp_arg
;
247 * Set up fake return address. As the lwp function may never return,
248 * we simply copy out a NULL pointer and force the lwp to receive
249 * a SIGSEGV if it returns anyways.
251 regs
->tf_rsp
-= sizeof(void *);
252 error
= copyout(&bad_return
, (void *)regs
->tf_rsp
, sizeof(bad_return
));
256 if (lp
->lwp_proc
->p_vmm
) {
257 lp
->lwp_thread
->td_pcb
->pcb_cr3
= KPML4phys
;
258 cpu_set_fork_handler(lp
,
259 (void (*)(void *, struct trapframe
*))vmm_lwp_return
, lp
);
261 cpu_set_fork_handler(lp
,
262 (void (*)(void *, struct trapframe
*))generic_lwp_return
, lp
);
268 * Intercept the return address from a freshly forked process that has NOT
269 * been scheduled yet.
271 * This is needed to make kernel threads stay in kernel mode.
274 cpu_set_fork_handler(struct lwp
*lp
, void (*func
)(void *, struct trapframe
*),
278 * Note that the trap frame follows the args, so the function
279 * is really called like this: func(arg, frame);
281 lp
->lwp_thread
->td_pcb
->pcb_rbx
= (long)func
; /* function */
282 lp
->lwp_thread
->td_pcb
->pcb_r12
= (long)arg
; /* first arg */
286 cpu_set_thread_handler(thread_t td
, void (*rfunc
)(void), void *func
, void *arg
)
288 td
->td_pcb
->pcb_rbx
= (long)func
;
289 td
->td_pcb
->pcb_r12
= (long)arg
;
290 td
->td_switch
= cpu_lwkt_switch
;
291 td
->td_sp
-= sizeof(void *);
292 *(void **)td
->td_sp
= rfunc
; /* exit function on return */
293 td
->td_sp
-= sizeof(void *);
294 *(void **)td
->td_sp
= cpu_kthread_restore
;
300 struct thread
*td
= curthread
;
305 /* Some x86 functionality was dropped */
306 KKASSERT(pcb
->pcb_ext
== NULL
);
309 * disable all hardware breakpoints
311 if (pcb
->pcb_flags
& PCB_DBREGS
) {
313 pcb
->pcb_flags
&= ~PCB_DBREGS
;
315 td
->td_gd
->gd_cnt
.v_swtch
++;
317 crit_enter_quick(td
);
318 if (td
->td_flags
& TDF_TSLEEPQ
)
320 lwkt_deschedule_self(td
);
321 lwkt_remove_tdallq(td
);
326 * Terminate the current thread. The caller must have already acquired
327 * the thread's rwlock and placed it on a reap list or otherwise notified
328 * a reaper of its existance. We set a special assembly switch function which
329 * releases td_rwlock after it has cleaned up the MMU state and switched
332 * Must be caller from a critical section and with the thread descheduled.
335 cpu_thread_exit(void)
338 curthread
->td_switch
= cpu_exit_switch
;
339 curthread
->td_flags
|= TDF_EXITING
;
341 panic("cpu_thread_exit: lwkt_switch() unexpectedly returned");
354 * Attempt to do a CPU reset via the keyboard controller,
355 * do not turn off the GateA20, as any machine that fails
356 * to do the reset here would then end up in no man's land.
359 #if !defined(BROKEN_KEYBOARD_RESET)
360 outb(IO_KBD
+ 4, 0xFE);
361 DELAY(500000); /* wait 0.5 sec to see if that did it */
362 kprintf("Keyboard reset did not work, attempting CPU shutdown\n");
363 DELAY(1000000); /* wait 1 sec for kprintf to complete */
366 /* force a shutdown by unmapping entire address space ! */
367 bzero((caddr_t
) PTD
, PAGE_SIZE
);
370 /* "good night, sweet prince .... <THUNK!>" */
377 * Convert kernel VA to physical address
384 pa
= pmap_kextract((vm_offset_t
)addr
);
386 panic("kvtop: zero page frame");
391 swi_vm(void *arg
, void *frame
)
393 if (busdma_swi_pending
!= 0)
398 swi_vm_setup(void *arg
)
400 register_swi_mp(SWI_VM
, swi_vm
, NULL
, "swi_vm", NULL
, 0);
403 SYSINIT(swi_vm_setup
, SI_BOOT2_MACHDEP
, SI_ORDER_ANY
, swi_vm_setup
, NULL
);
406 * NOTE: This routine is also called after a successful microcode
409 void spectre_vm_setup(void *arg
);
412 * Check for IBRS support
416 spectre_check_support(void)
424 cpuid_count(7, 0, p
);
425 if ((p
[3] & 0x0C000000U
) == 0x0C000000U
) {
428 * SPEC_CTRL (bit 26) and STIBP support (bit 27)
430 * 0x80000008 p[0] bit 12 indicates IBPB support
436 do_cpuid(0x80000008U
, p
);
437 if (p
[0] & 0x00001000)
444 * Iterate CPUs and adjust MSR for global operations, since
445 * the KMMU* code won't do it if spectre_mitigation is 0 or 2.
449 spectre_sysctl_changed(int old_value
)
451 globaldata_t save_gd
;
455 for (n
= 0; n
< ncpus
; ++n
) {
456 lwkt_setcpu_self(globaldata_find(n
));
458 pscpu
->trampoline
.tr_pcb_gflags
&= ~(PCB_IBRS1
| PCB_IBRS2
);
460 switch(spectre_mitigation
) {
466 pscpu
->trampoline
.tr_pcb_gflags
|= PCB_IBRS1
;
470 pscpu
->trampoline
.tr_pcb_gflags
|= PCB_IBRS2
;
475 if (save_gd
!= mycpu
)
476 lwkt_setcpu_self(save_gd
);
480 * User changes sysctl value
483 sysctl_spectre_mitigation(SYSCTL_HANDLER_ARGS
)
489 old_spectre
= spectre_mitigation
;
490 new_spectre
= old_spectre
;
491 error
= sysctl_handle_int(oidp
, &new_spectre
, 0, req
);
492 if (error
|| req
->newptr
== NULL
)
494 spectre_mitigation
= new_spectre
;
495 spectre_sysctl_changed(old_spectre
);
500 SYSCTL_PROC(_machdep
, OID_AUTO
, spectre_mitigation
, CTLTYPE_INT
| CTLFLAG_RW
,
501 0, 0, sysctl_spectre_mitigation
, "I", "Spectre exploit mitigation");
504 spectre_vm_setup(void *arg
)
506 int inconsistent
= 0;
507 int old_value
= spectre_mitigation
;
509 if (spectre_mitigation
< 0) {
510 TUNABLE_INT_FETCH("machdep.spectre_mitigation",
511 &spectre_mitigation
);
514 if (cpu_vendor_id
== CPU_VENDOR_INTEL
) {
515 if (spectre_check_support()) {
517 * Must be supported on all cpus before we
518 * can enable it. Returns silently if it
521 * NOTE! arg != NULL indicates we were called
522 * from cpuctl after a successful microcode
526 globaldata_t save_gd
;
530 for (n
= 0; n
< ncpus
; ++n
) {
531 lwkt_setcpu_self(globaldata_find(n
));
532 if (spectre_check_support() == 0) {
537 if (save_gd
!= mycpu
)
538 lwkt_setcpu_self(save_gd
);
540 if (inconsistent
== 0) {
541 if (spectre_mitigation
< 0)
542 spectre_mitigation
= 1;
544 spectre_mitigation
= -1;
547 spectre_mitigation
= -1;
550 spectre_mitigation
= -1; /* no support */
554 * Be silent while microcode is being loaded on various CPUs,
561 * Disallow sysctl changes when there is no support (otherwise
562 * the wrmsr will cause a protection fault).
564 switch(spectre_mitigation
) {
566 sysctl___machdep_spectre_mitigation
.oid_kind
|= CTLFLAG_WR
;
567 kprintf("machdep.spectre_mitigation available but disabled\n");
570 sysctl___machdep_spectre_mitigation
.oid_kind
|= CTLFLAG_WR
;
571 kprintf("machdep.spectre_mitigation available, system call\n"
572 "performance and kernel operation will be impacted\n");
575 sysctl___machdep_spectre_mitigation
.oid_kind
|= CTLFLAG_WR
;
576 kprintf("machdep.spectre_mitigation available, whole machine\n"
577 "performance will be impacted\n");
580 sysctl___machdep_spectre_mitigation
.oid_kind
&= ~CTLFLAG_WR
;
581 if (cpu_vendor_id
== CPU_VENDOR_INTEL
)
582 kprintf("no microcode spectre mitigation available\n");
585 spectre_sysctl_changed(old_value
);
588 SYSINIT(spectre_vm_setup
, SI_BOOT2_MACHDEP
, SI_ORDER_ANY
,
589 spectre_vm_setup
, NULL
);
592 * platform-specific vmspace initialization (nothing for x86_64)
595 cpu_vmspace_alloc(struct vmspace
*vm __unused
)
600 cpu_vmspace_free(struct vmspace
*vm __unused
)
605 kvm_access_check(vm_offset_t saddr
, vm_offset_t eaddr
, int prot
)
609 if (saddr
< KvaStart
)
613 for (addr
= saddr
; addr
< eaddr
; addr
+= PAGE_SIZE
) {
614 if (pmap_kextract(addr
) == 0)
617 if (!kernacc((caddr_t
)saddr
, eaddr
- saddr
, prot
))
624 void _test_frame_enter(struct trapframe
*frame
);
625 void _test_frame_exit(struct trapframe
*frame
);
628 _test_frame_enter(struct trapframe
*frame
)
630 thread_t td
= curthread
;
632 if (ISPL(frame
->tf_cs
) == SEL_UPL
) {
633 KKASSERT(td
->td_lwp
);
634 KASSERT(td
->td_lwp
->lwp_md
.md_regs
== frame
,
635 ("_test_frame_exit: Frame mismatch %p %p",
636 td
->td_lwp
->lwp_md
.md_regs
, frame
));
637 td
->td_lwp
->lwp_saveusp
= (void *)frame
->tf_rsp
;
638 td
->td_lwp
->lwp_saveupc
= (void *)frame
->tf_rip
;
640 if ((char *)frame
< td
->td_kstack
||
641 (char *)frame
> td
->td_kstack
+ td
->td_kstack_size
) {
642 panic("_test_frame_exit: frame not on kstack %p kstack=%p",
643 frame
, td
->td_kstack
);
648 _test_frame_exit(struct trapframe
*frame
)
650 thread_t td
= curthread
;
652 if (ISPL(frame
->tf_cs
) == SEL_UPL
) {
653 KKASSERT(td
->td_lwp
);
654 KASSERT(td
->td_lwp
->lwp_md
.md_regs
== frame
,
655 ("_test_frame_exit: Frame mismatch %p %p",
656 td
->td_lwp
->lwp_md
.md_regs
, frame
));
657 if (td
->td_lwp
->lwp_saveusp
!= (void *)frame
->tf_rsp
) {
658 kprintf("_test_frame_exit: %s:%d usp mismatch %p/%p\n",
659 td
->td_comm
, td
->td_proc
->p_pid
,
660 td
->td_lwp
->lwp_saveusp
,
661 (void *)frame
->tf_rsp
);
663 if (td
->td_lwp
->lwp_saveupc
!= (void *)frame
->tf_rip
) {
664 kprintf("_test_frame_exit: %s:%d upc mismatch %p/%p\n",
665 td
->td_comm
, td
->td_proc
->p_pid
,
666 td
->td_lwp
->lwp_saveupc
,
667 (void *)frame
->tf_rip
);
671 * adulterate the fields to catch entries that
672 * don't run through test_frame_enter
674 td
->td_lwp
->lwp_saveusp
=
675 (void *)~(intptr_t)td
->td_lwp
->lwp_saveusp
;
676 td
->td_lwp
->lwp_saveupc
=
677 (void *)~(intptr_t)td
->td_lwp
->lwp_saveupc
;
679 if ((char *)frame
< td
->td_kstack
||
680 (char *)frame
> td
->td_kstack
+ td
->td_kstack_size
) {
681 panic("_test_frame_exit: frame not on kstack %p kstack=%p",
682 frame
, td
->td_kstack
);