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
7 * This code is derived from software contributed to Berkeley by
8 * the Systems Programming Group of the University of Utah Computer
9 * Science Department, and William Jolitz.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91
40 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
48 #include "opt_compat.h"
50 #include <sys/param.h>
51 #include <sys/systm.h>
54 #include <sys/kernel.h>
57 #include <sys/malloc.h>
59 #include <sys/mutex.h>
60 #include <sys/pioctl.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
65 #include <sys/unistd.h>
66 #include <sys/vnode.h>
67 #include <sys/vmmeter.h>
69 #include <machine/cpu.h>
70 #include <machine/md_var.h>
71 #include <machine/pcb.h>
72 #include <machine/smp.h>
73 #include <machine/specialreg.h>
74 #include <machine/tss.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_param.h>
83 #include <isa/isareg.h>
85 static void cpu_reset_real(void);
87 static void cpu_reset_proxy(void);
88 static u_int cpu_reset_proxyid;
89 static volatile u_int cpu_reset_proxy_active;
92 _Static_assert(OFFSETOF_CURTHREAD == offsetof(struct pcpu, pc_curthread),
93 "OFFSETOF_CURTHREAD does not correspond with offset of pc_curthread.");
94 _Static_assert(OFFSETOF_CURPCB == offsetof(struct pcpu, pc_curpcb),
95 "OFFSETOF_CURPCB does not correspond with offset of pc_curpcb.");
96 _Static_assert(OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf),
97 "OFFSETOF_MONINORBUF does not correspond with offset of pc_monitorbuf.");
100 get_pcb_user_save_td(struct thread *td)
104 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
105 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
106 KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area"));
107 return ((struct savefpu *)p);
111 get_pcb_user_save_pcb(struct pcb *pcb)
115 p = (vm_offset_t)(pcb + 1);
116 return ((struct savefpu *)p);
120 get_pcb_td(struct thread *td)
124 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
125 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) -
127 return ((struct pcb *)p);
131 alloc_fpusave(int flags)
134 struct savefpu_ymm *sf;
136 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
138 sf = (struct savefpu_ymm *)res;
139 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
140 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
146 * Finish a fork operation, with process p2 nearly set up.
147 * Copy and update the pcb, set up the stack so that the child
148 * ready to run and return to user mode.
151 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
155 struct mdproc *mdp1, *mdp2;
156 struct proc_ldt *pldt;
159 if ((flags & RFPROC) == 0) {
160 if ((flags & RFMEM) == 0) {
161 /* unshare user LDT */
164 if ((pldt = mdp1->md_ldt) != NULL &&
165 pldt->ldt_refcnt > 1 &&
166 user_ldt_alloc(p1, 1) == NULL)
167 panic("could not copy LDT");
168 mtx_unlock(&dt_lock);
173 /* Ensure that td1's pcb is up to date. */
175 update_pcb_bases(td1->td_pcb);
177 /* Point the pcb to the top of the stack */
178 pcb2 = get_pcb_td(td2);
182 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
184 /* Properly initialize pcb_save */
185 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
186 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
187 cpu_max_ext_state_size);
189 /* Point mdproc and then copy over td1's contents */
191 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
194 * Create a new fresh stack for the new process.
195 * Copy the trap frame for the return to user mode as if from a
196 * syscall. This copies most of the user mode register values.
198 td2->td_frame = (struct trapframe *)td2->td_pcb - 1;
199 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
201 td2->td_frame->tf_rax = 0; /* Child returns zero */
202 td2->td_frame->tf_rflags &= ~PSL_C; /* success */
203 td2->td_frame->tf_rdx = 1;
206 * If the parent process has the trap bit set (i.e. a debugger had
207 * single stepped the process to the system call), we need to clear
208 * the trap flag from the new frame unless the debugger had set PF_FORK
209 * on the parent. Otherwise, the child will receive a (likely
210 * unexpected) SIGTRAP when it executes the first instruction after
211 * returning to userland.
213 if ((p1->p_pfsflags & PF_FORK) == 0)
214 td2->td_frame->tf_rflags &= ~PSL_T;
217 * Set registers for trampoline to user mode. Leave space for the
218 * return address on stack. These are the kernel mode register values.
220 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */
222 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *);
223 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */
224 pcb2->pcb_rip = (register_t)fork_trampoline;
226 * pcb2->pcb_dr*: cloned above.
227 * pcb2->pcb_savefpu: cloned above.
228 * pcb2->pcb_flags: cloned above.
229 * pcb2->pcb_onfault: cloned above (always NULL here?).
230 * pcb2->pcb_[fg]sbase: cloned above
233 /* Setup to release spin count in fork_exit(). */
234 td2->td_md.md_spinlock_count = 1;
235 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
236 td2->td_md.md_invl_gen.gen = 0;
238 /* As an i386, do not copy io permission bitmap. */
239 pcb2->pcb_tssp = NULL;
241 /* New segment registers. */
242 set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
244 /* Copy the LDT, if necessary. */
245 mdp1 = &td1->td_proc->p_md;
248 if (mdp1->md_ldt != NULL) {
250 mdp1->md_ldt->ldt_refcnt++;
251 mdp2->md_ldt = mdp1->md_ldt;
252 bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct
253 system_segment_descriptor));
256 mdp2->md_ldt = user_ldt_alloc(p2, 0);
257 if (mdp2->md_ldt == NULL)
258 panic("could not copy LDT");
259 amd64_set_ldt_data(td2, 0, max_ldt_segment,
260 (struct user_segment_descriptor *)
261 mdp1->md_ldt->ldt_base);
265 mtx_unlock(&dt_lock);
268 * Now, cpu_switch() can schedule the new process.
269 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
270 * containing the return address when exiting cpu_switch.
271 * This will normally be to fork_trampoline(), which will have
272 * %ebx loaded with the new proc's pointer. fork_trampoline()
273 * will set up a stack to call fork_return(p, frame); to complete
274 * the return to user-mode.
279 * Intercept the return address from a freshly forked process that has NOT
280 * been scheduled yet.
282 * This is needed to make kernel threads stay in kernel mode.
285 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
288 * Note that the trap frame follows the args, so the function
289 * is really called like this: func(arg, frame);
291 td->td_pcb->pcb_r12 = (long) func; /* function */
292 td->td_pcb->pcb_rbx = (long) arg; /* first arg */
296 cpu_exit(struct thread *td)
300 * If this process has a custom LDT, release it.
302 if (td->td_proc->p_md.md_ldt != NULL)
307 cpu_thread_exit(struct thread *td)
312 if (td == PCPU_GET(fpcurthread))
318 /* Disable any hardware breakpoints. */
319 if (pcb->pcb_flags & PCB_DBREGS) {
321 clear_pcb_flags(pcb, PCB_DBREGS);
326 cpu_thread_clean(struct thread *td)
335 if (pcb->pcb_tssp != NULL) {
336 kmem_free(kernel_arena, (vm_offset_t)pcb->pcb_tssp,
338 pcb->pcb_tssp = NULL;
343 cpu_thread_swapin(struct thread *td)
348 cpu_thread_swapout(struct thread *td)
353 cpu_thread_alloc(struct thread *td)
356 struct xstate_hdr *xhdr;
358 td->td_pcb = pcb = get_pcb_td(td);
359 td->td_frame = (struct trapframe *)pcb - 1;
360 pcb->pcb_save = get_pcb_user_save_pcb(pcb);
362 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
363 bzero(xhdr, sizeof(*xhdr));
364 xhdr->xstate_bv = xsave_mask;
369 cpu_thread_free(struct thread *td)
372 cpu_thread_clean(td);
376 cpu_set_syscall_retval(struct thread *td, int error)
381 td->td_frame->tf_rax = td->td_retval[0];
382 td->td_frame->tf_rdx = td->td_retval[1];
383 td->td_frame->tf_rflags &= ~PSL_C;
388 * Reconstruct pc, we know that 'syscall' is 2 bytes,
389 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
390 * We saved this in tf_err.
391 * %r10 (which was holding the value of %rcx) is restored
392 * for the next iteration.
393 * %r10 restore is only required for freebsd/amd64 processes,
394 * but shall be innocent for any ia32 ABI.
396 * Require full context restore to get the arguments
397 * in the registers reloaded at return to usermode.
399 td->td_frame->tf_rip -= td->td_frame->tf_err;
400 td->td_frame->tf_r10 = td->td_frame->tf_rcx;
401 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
408 td->td_frame->tf_rax = SV_ABI_ERRNO(td->td_proc, error);
409 td->td_frame->tf_rflags |= PSL_C;
415 * Initialize machine state, mostly pcb and trap frame for a new
416 * thread, about to return to userspace. Put enough state in the new
417 * thread's PCB to get it to go back to the fork_return(), which
418 * finalizes the thread state and handles peculiarities of the first
419 * return to userspace for the new thread.
422 cpu_copy_thread(struct thread *td, struct thread *td0)
426 /* Point the pcb to the top of the stack. */
430 * Copy the upcall pcb. This loads kernel regs.
431 * Those not loaded individually below get their default
434 update_pcb_bases(td0->td_pcb);
435 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
436 clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE |
438 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
439 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
440 cpu_max_ext_state_size);
441 set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
444 * Create a new fresh stack for the new thread.
446 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
448 /* If the current thread has the trap bit set (i.e. a debugger had
449 * single stepped the process to the system call), we need to clear
450 * the trap flag from the new frame. Otherwise, the new thread will
451 * receive a (likely unexpected) SIGTRAP when it executes the first
452 * instruction after returning to userland.
454 td->td_frame->tf_rflags &= ~PSL_T;
457 * Set registers for trampoline to user mode. Leave space for the
458 * return address on stack. These are the kernel mode register values.
460 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */
462 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */
463 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */
464 pcb2->pcb_rip = (register_t)fork_trampoline;
466 * If we didn't copy the pcb, we'd need to do the following registers:
467 * pcb2->pcb_dr*: cloned above.
468 * pcb2->pcb_savefpu: cloned above.
469 * pcb2->pcb_onfault: cloned above (always NULL here?).
470 * pcb2->pcb_[fg]sbase: cloned above
473 /* Setup to release spin count in fork_exit(). */
474 td->td_md.md_spinlock_count = 1;
475 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
479 * Set that machine state for performing an upcall that starts
480 * the entry function with the given argument.
483 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
488 * Do any extra cleaning that needs to be done.
489 * The thread may have optional components
490 * that are not present in a fresh thread.
491 * This may be a recycled thread so make it look
492 * as though it's newly allocated.
494 cpu_thread_clean(td);
496 #ifdef COMPAT_FREEBSD32
497 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
499 * Set the trap frame to point at the beginning of the entry
502 td->td_frame->tf_rbp = 0;
503 td->td_frame->tf_rsp =
504 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
505 td->td_frame->tf_rip = (uintptr_t)entry;
507 /* Pass the argument to the entry point. */
508 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)),
509 (uint32_t)(uintptr_t)arg);
516 * Set the trap frame to point at the beginning of the uts
519 td->td_frame->tf_rbp = 0;
520 td->td_frame->tf_rsp =
521 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f;
522 td->td_frame->tf_rsp -= 8;
523 td->td_frame->tf_rip = (register_t)entry;
524 td->td_frame->tf_ds = _udatasel;
525 td->td_frame->tf_es = _udatasel;
526 td->td_frame->tf_fs = _ufssel;
527 td->td_frame->tf_gs = _ugssel;
528 td->td_frame->tf_flags = TF_HASSEGS;
530 /* Pass the argument to the entry point. */
531 td->td_frame->tf_rdi = (register_t)arg;
535 cpu_set_user_tls(struct thread *td, void *tls_base)
539 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS)
543 set_pcb_flags(pcb, PCB_FULL_IRET);
544 #ifdef COMPAT_FREEBSD32
545 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
546 pcb->pcb_gsbase = (register_t)tls_base;
550 pcb->pcb_fsbase = (register_t)tls_base;
560 cpu_reset_proxy_active = 1;
561 while (cpu_reset_proxy_active == 1)
562 ia32_pause(); /* Wait for other cpu to see that we've started */
564 CPU_SETOF(cpu_reset_proxyid, &tcrp);
566 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
581 CPU_CLR(PCPU_GET(cpuid), &map);
582 CPU_NAND(&map, &stopped_cpus);
583 if (!CPU_EMPTY(&map)) {
584 printf("cpu_reset: Stopping other CPUs\n");
588 if (PCPU_GET(cpuid) != 0) {
589 cpu_reset_proxyid = PCPU_GET(cpuid);
590 cpustop_restartfunc = cpu_reset_proxy;
591 cpu_reset_proxy_active = 0;
592 printf("cpu_reset: Restarting BSP\n");
594 /* Restart CPU #0. */
595 CPU_SETOF(0, &started_cpus);
599 while (cpu_reset_proxy_active == 0 && cnt < 10000000) {
601 cnt++; /* Wait for BSP to announce restart */
603 if (cpu_reset_proxy_active == 0)
604 printf("cpu_reset: Failed to restart BSP\n");
606 cpu_reset_proxy_active = 2;
623 struct region_descriptor null_idt;
629 * Attempt to do a CPU reset via the keyboard controller,
630 * do not turn off GateA20, as any machine that fails
631 * to do the reset here would then end up in no man's land.
633 outb(IO_KBD + 4, 0xFE);
634 DELAY(500000); /* wait 0.5 sec to see if that did it */
637 * Attempt to force a reset via the Reset Control register at
638 * I/O port 0xcf9. Bit 2 forces a system reset when it
639 * transitions from 0 to 1. Bit 1 selects the type of reset
640 * to attempt: 0 selects a "soft" reset, and 1 selects a
641 * "hard" reset. We try a "hard" reset. The first write sets
642 * bit 1 to select a "hard" reset and clears bit 2. The
643 * second write forces a 0 -> 1 transition in bit 2 to trigger
648 DELAY(500000); /* wait 0.5 sec to see if that did it */
651 * Attempt to force a reset via the Fast A20 and Init register
652 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate.
653 * Bit 0 asserts INIT# when set to 1. We are careful to only
654 * preserve bit 1 while setting bit 0. We also must clear bit
655 * 0 before setting it if it isn't already clear.
660 outb(0x92, b & 0xfe);
662 DELAY(500000); /* wait 0.5 sec to see if that did it */
665 printf("No known reset method worked, attempting CPU shutdown\n");
666 DELAY(1000000); /* wait 1 sec for printf to complete */
669 null_idt.rd_limit = 0;
670 null_idt.rd_base = 0;
673 /* "good night, sweet prince .... <THUNK!>" */
681 * Software interrupt handler for queued VM system processing.
686 if (busdma_swi_pending != 0)
691 * Tell whether this address is in some physical memory region.
692 * Currently used by the kernel coredump code in order to avoid
693 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
694 * or other unpredictable behaviour.
698 is_physical_memory(vm_paddr_t addr)
702 /* The ISA ``memory hole''. */
703 if (addr >= 0xa0000 && addr < 0x100000)
708 * stuff other tests for known memory-mapped devices (PCI?)