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 cpu_max_ext_state_size;
106 KASSERT((p % 64) == 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 cpu_max_ext_state_size - sizeof(struct pcb);
126 return ((struct pcb *)p);
130 alloc_fpusave(int flags)
133 struct savefpu_ymm *sf;
135 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
137 sf = (struct savefpu_ymm *)res;
138 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
139 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
145 * Finish a fork operation, with process p2 nearly set up.
146 * Copy and update the pcb, set up the stack so that the child
147 * ready to run and return to user mode.
150 cpu_fork(td1, p2, td2, flags)
151 register struct thread *td1;
152 register struct proc *p2;
156 register struct proc *p1;
158 struct mdproc *mdp1, *mdp2;
159 struct proc_ldt *pldt;
162 if ((flags & RFPROC) == 0) {
163 if ((flags & RFMEM) == 0) {
164 /* unshare user LDT */
167 if ((pldt = mdp1->md_ldt) != NULL &&
168 pldt->ldt_refcnt > 1 &&
169 user_ldt_alloc(p1, 1) == NULL)
170 panic("could not copy LDT");
171 mtx_unlock(&dt_lock);
176 /* Ensure that td1's pcb is up to date. */
179 /* Point the pcb to the top of the stack */
180 pcb2 = get_pcb_td(td2);
184 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
186 /* Properly initialize pcb_save */
187 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
188 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
189 cpu_max_ext_state_size);
191 /* Point mdproc and then copy over td1's contents */
193 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
196 * Create a new fresh stack for the new process.
197 * Copy the trap frame for the return to user mode as if from a
198 * syscall. This copies most of the user mode register values.
200 td2->td_frame = (struct trapframe *)td2->td_pcb - 1;
201 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
203 td2->td_frame->tf_rax = 0; /* Child returns zero */
204 td2->td_frame->tf_rflags &= ~PSL_C; /* success */
205 td2->td_frame->tf_rdx = 1;
208 * If the parent process has the trap bit set (i.e. a debugger had
209 * single stepped the process to the system call), we need to clear
210 * the trap flag from the new frame unless the debugger had set PF_FORK
211 * on the parent. Otherwise, the child will receive a (likely
212 * unexpected) SIGTRAP when it executes the first instruction after
213 * returning to userland.
215 if ((p1->p_pfsflags & PF_FORK) == 0)
216 td2->td_frame->tf_rflags &= ~PSL_T;
219 * Set registers for trampoline to user mode. Leave space for the
220 * return address on stack. These are the kernel mode register values.
222 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */
224 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *);
225 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */
226 pcb2->pcb_rip = (register_t)fork_trampoline;
228 * pcb2->pcb_dr*: cloned above.
229 * pcb2->pcb_savefpu: cloned above.
230 * pcb2->pcb_flags: cloned above.
231 * pcb2->pcb_onfault: cloned above (always NULL here?).
232 * pcb2->pcb_[fg]sbase: cloned above
235 /* Setup to release spin count in fork_exit(). */
236 td2->td_md.md_spinlock_count = 1;
237 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
239 /* As an i386, do not copy io permission bitmap. */
240 pcb2->pcb_tssp = NULL;
242 /* New segment registers. */
243 set_pcb_flags(pcb2, PCB_FULL_IRET);
245 /* Copy the LDT, if necessary. */
246 mdp1 = &td1->td_proc->p_md;
249 if (mdp1->md_ldt != NULL) {
251 mdp1->md_ldt->ldt_refcnt++;
252 mdp2->md_ldt = mdp1->md_ldt;
253 bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct
254 system_segment_descriptor));
257 mdp2->md_ldt = user_ldt_alloc(p2, 0);
258 if (mdp2->md_ldt == NULL)
259 panic("could not copy LDT");
260 amd64_set_ldt_data(td2, 0, max_ldt_segment,
261 (struct user_segment_descriptor *)
262 mdp1->md_ldt->ldt_base);
266 mtx_unlock(&dt_lock);
269 * Now, cpu_switch() can schedule the new process.
270 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
271 * containing the return address when exiting cpu_switch.
272 * This will normally be to fork_trampoline(), which will have
273 * %ebx loaded with the new proc's pointer. fork_trampoline()
274 * will set up a stack to call fork_return(p, frame); to complete
275 * the return to user-mode.
280 * Intercept the return address from a freshly forked process that has NOT
281 * been scheduled yet.
283 * This is needed to make kernel threads stay in kernel mode.
286 cpu_set_fork_handler(td, func, arg)
288 void (*func)(void *);
292 * Note that the trap frame follows the args, so the function
293 * is really called like this: func(arg, frame);
295 td->td_pcb->pcb_r12 = (long) func; /* function */
296 td->td_pcb->pcb_rbx = (long) arg; /* first arg */
300 cpu_exit(struct thread *td)
304 * If this process has a custom LDT, release it.
307 if (td->td_proc->p_md.md_ldt != 0)
310 mtx_unlock(&dt_lock);
314 cpu_thread_exit(struct thread *td)
319 if (td == PCPU_GET(fpcurthread))
325 /* Disable any hardware breakpoints. */
326 if (pcb->pcb_flags & PCB_DBREGS) {
328 clear_pcb_flags(pcb, PCB_DBREGS);
333 cpu_thread_clean(struct thread *td)
342 if (pcb->pcb_tssp != NULL) {
343 kmem_free(kernel_arena, (vm_offset_t)pcb->pcb_tssp,
345 pcb->pcb_tssp = NULL;
350 cpu_thread_swapin(struct thread *td)
355 cpu_thread_swapout(struct thread *td)
360 cpu_thread_alloc(struct thread *td)
363 struct xstate_hdr *xhdr;
365 td->td_pcb = pcb = get_pcb_td(td);
366 td->td_frame = (struct trapframe *)pcb - 1;
367 pcb->pcb_save = get_pcb_user_save_pcb(pcb);
369 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
370 bzero(xhdr, sizeof(*xhdr));
371 xhdr->xstate_bv = xsave_mask;
376 cpu_thread_free(struct thread *td)
379 cpu_thread_clean(td);
383 cpu_set_syscall_retval(struct thread *td, int error)
388 td->td_frame->tf_rax = td->td_retval[0];
389 td->td_frame->tf_rdx = td->td_retval[1];
390 td->td_frame->tf_rflags &= ~PSL_C;
395 * Reconstruct pc, we know that 'syscall' is 2 bytes,
396 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
397 * We saved this in tf_err.
398 * %r10 (which was holding the value of %rcx) is restored
399 * for the next iteration.
400 * %r10 restore is only required for freebsd/amd64 processes,
401 * but shall be innocent for any ia32 ABI.
403 * Require full context restore to get the arguments
404 * in the registers reloaded at return to usermode.
406 td->td_frame->tf_rip -= td->td_frame->tf_err;
407 td->td_frame->tf_r10 = td->td_frame->tf_rcx;
408 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
415 if (td->td_proc->p_sysent->sv_errsize) {
416 if (error >= td->td_proc->p_sysent->sv_errsize)
417 error = -1; /* XXX */
419 error = td->td_proc->p_sysent->sv_errtbl[error];
421 td->td_frame->tf_rax = error;
422 td->td_frame->tf_rflags |= PSL_C;
428 * Initialize machine state (pcb and trap frame) for a new thread about to
429 * upcall. Put enough state in the new thread's PCB to get it to go back
430 * userret(), where we can intercept it again to set the return (upcall)
431 * Address and stack, along with those from upcals that are from other sources
432 * such as those generated in thread_userret() itself.
435 cpu_set_upcall(struct thread *td, struct thread *td0)
439 /* Point the pcb to the top of the stack. */
443 * Copy the upcall pcb. This loads kernel regs.
444 * Those not loaded individually below get their default
447 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
448 clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE |
450 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
451 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
452 cpu_max_ext_state_size);
453 set_pcb_flags(pcb2, PCB_FULL_IRET);
456 * Create a new fresh stack for the new thread.
458 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
460 /* If the current thread has the trap bit set (i.e. a debugger had
461 * single stepped the process to the system call), we need to clear
462 * the trap flag from the new frame. Otherwise, the new thread will
463 * receive a (likely unexpected) SIGTRAP when it executes the first
464 * instruction after returning to userland.
466 td->td_frame->tf_rflags &= ~PSL_T;
469 * Set registers for trampoline to user mode. Leave space for the
470 * return address on stack. These are the kernel mode register values.
472 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */
474 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */
475 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */
476 pcb2->pcb_rip = (register_t)fork_trampoline;
478 * If we didn't copy the pcb, we'd need to do the following registers:
479 * pcb2->pcb_dr*: cloned above.
480 * pcb2->pcb_savefpu: cloned above.
481 * pcb2->pcb_onfault: cloned above (always NULL here?).
482 * pcb2->pcb_[fg]sbase: cloned above
485 /* Setup to release spin count in fork_exit(). */
486 td->td_md.md_spinlock_count = 1;
487 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
491 * Set that machine state for performing an upcall that has to
492 * be done in thread_userret() so that those upcalls generated
493 * in thread_userret() itself can be done as well.
496 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
501 * Do any extra cleaning that needs to be done.
502 * The thread may have optional components
503 * that are not present in a fresh thread.
504 * This may be a recycled thread so make it look
505 * as though it's newly allocated.
507 cpu_thread_clean(td);
509 #ifdef COMPAT_FREEBSD32
510 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
512 * Set the trap frame to point at the beginning of the uts
515 td->td_frame->tf_rbp = 0;
516 td->td_frame->tf_rsp =
517 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
518 td->td_frame->tf_rip = (uintptr_t)entry;
521 * Pass the address of the mailbox for this kse to the uts
522 * function as a parameter on the stack.
524 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)),
525 (uint32_t)(uintptr_t)arg);
532 * Set the trap frame to point at the beginning of the uts
535 td->td_frame->tf_rbp = 0;
536 td->td_frame->tf_rsp =
537 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f;
538 td->td_frame->tf_rsp -= 8;
539 td->td_frame->tf_rip = (register_t)entry;
540 td->td_frame->tf_ds = _udatasel;
541 td->td_frame->tf_es = _udatasel;
542 td->td_frame->tf_fs = _ufssel;
543 td->td_frame->tf_gs = _ugssel;
544 td->td_frame->tf_flags = TF_HASSEGS;
547 * Pass the address of the mailbox for this kse to the uts
548 * function as a parameter on the stack.
550 td->td_frame->tf_rdi = (register_t)arg;
554 cpu_set_user_tls(struct thread *td, void *tls_base)
558 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS)
562 set_pcb_flags(pcb, PCB_FULL_IRET);
563 #ifdef COMPAT_FREEBSD32
564 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
565 pcb->pcb_gsbase = (register_t)tls_base;
569 pcb->pcb_fsbase = (register_t)tls_base;
579 cpu_reset_proxy_active = 1;
580 while (cpu_reset_proxy_active == 1)
581 ia32_pause(); /* Wait for other cpu to see that we've started */
583 CPU_SETOF(cpu_reset_proxyid, &tcrp);
585 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
600 CPU_CLR(PCPU_GET(cpuid), &map);
601 CPU_NAND(&map, &stopped_cpus);
602 if (!CPU_EMPTY(&map)) {
603 printf("cpu_reset: Stopping other CPUs\n");
607 if (PCPU_GET(cpuid) != 0) {
608 cpu_reset_proxyid = PCPU_GET(cpuid);
609 cpustop_restartfunc = cpu_reset_proxy;
610 cpu_reset_proxy_active = 0;
611 printf("cpu_reset: Restarting BSP\n");
613 /* Restart CPU #0. */
614 CPU_SETOF(0, &started_cpus);
618 while (cpu_reset_proxy_active == 0 && cnt < 10000000) {
620 cnt++; /* Wait for BSP to announce restart */
622 if (cpu_reset_proxy_active == 0)
623 printf("cpu_reset: Failed to restart BSP\n");
625 cpu_reset_proxy_active = 2;
642 struct region_descriptor null_idt;
648 * Attempt to do a CPU reset via the keyboard controller,
649 * do not turn off GateA20, as any machine that fails
650 * to do the reset here would then end up in no man's land.
652 outb(IO_KBD + 4, 0xFE);
653 DELAY(500000); /* wait 0.5 sec to see if that did it */
656 * Attempt to force a reset via the Reset Control register at
657 * I/O port 0xcf9. Bit 2 forces a system reset when it
658 * transitions from 0 to 1. Bit 1 selects the type of reset
659 * to attempt: 0 selects a "soft" reset, and 1 selects a
660 * "hard" reset. We try a "hard" reset. The first write sets
661 * bit 1 to select a "hard" reset and clears bit 2. The
662 * second write forces a 0 -> 1 transition in bit 2 to trigger
667 DELAY(500000); /* wait 0.5 sec to see if that did it */
670 * Attempt to force a reset via the Fast A20 and Init register
671 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate.
672 * Bit 0 asserts INIT# when set to 1. We are careful to only
673 * preserve bit 1 while setting bit 0. We also must clear bit
674 * 0 before setting it if it isn't already clear.
679 outb(0x92, b & 0xfe);
681 DELAY(500000); /* wait 0.5 sec to see if that did it */
684 printf("No known reset method worked, attempting CPU shutdown\n");
685 DELAY(1000000); /* wait 1 sec for printf to complete */
688 null_idt.rd_limit = 0;
689 null_idt.rd_base = 0;
692 /* "good night, sweet prince .... <THUNK!>" */
700 * Software interrupt handler for queued VM system processing.
705 if (busdma_swi_pending != 0)
710 * Tell whether this address is in some physical memory region.
711 * Currently used by the kernel coredump code in order to avoid
712 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
713 * or other unpredictable behaviour.
717 is_physical_memory(vm_paddr_t addr)
721 /* The ISA ``memory hole''. */
722 if (addr >= 0xa0000 && addr < 0x100000)
727 * stuff other tests for known memory-mapped devices (PCI?)