2 * SPDX-License-Identifier: BSD-4-Clause
4 * Copyright (c) 1982, 1986 The Regents of the University of California.
5 * Copyright (c) 1989, 1990 William Jolitz
6 * Copyright (c) 1994 John Dyson
9 * This code is derived from software contributed to Berkeley by
10 * the Systems Programming Group of the University of Utah Computer
11 * Science Department, and William Jolitz.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. All advertising materials mentioning features or use of this software
22 * must display the following acknowledgement:
23 * This product includes software developed by the University of
24 * California, Berkeley and its contributors.
25 * 4. Neither the name of the University nor the names of its contributors
26 * may be used to endorse or promote products derived from this software
27 * without specific prior written permission.
29 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
30 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
31 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
32 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
33 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
35 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
36 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
37 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
38 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
41 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91
42 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
50 #include "opt_compat.h"
52 #include <sys/param.h>
53 #include <sys/systm.h>
56 #include <sys/kernel.h>
59 #include <sys/malloc.h>
61 #include <sys/mutex.h>
62 #include <sys/pioctl.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/unistd.h>
68 #include <sys/vnode.h>
69 #include <sys/vmmeter.h>
71 #include <machine/cpu.h>
72 #include <machine/md_var.h>
73 #include <machine/pcb.h>
74 #include <machine/smp.h>
75 #include <machine/specialreg.h>
76 #include <machine/tss.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_kern.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_param.h>
85 #include <isa/isareg.h>
87 static void cpu_reset_real(void);
89 static void cpu_reset_proxy(void);
90 static u_int cpu_reset_proxyid;
91 static volatile u_int cpu_reset_proxy_active;
94 _Static_assert(OFFSETOF_CURTHREAD == offsetof(struct pcpu, pc_curthread),
95 "OFFSETOF_CURTHREAD does not correspond with offset of pc_curthread.");
96 _Static_assert(OFFSETOF_CURPCB == offsetof(struct pcpu, pc_curpcb),
97 "OFFSETOF_CURPCB does not correspond with offset of pc_curpcb.");
98 _Static_assert(OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf),
99 "OFFSETOF_MONINORBUF does not correspond with offset of pc_monitorbuf.");
102 get_pcb_user_save_td(struct thread *td)
106 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
107 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
108 KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area"));
109 return ((struct savefpu *)p);
113 get_pcb_user_save_pcb(struct pcb *pcb)
117 p = (vm_offset_t)(pcb + 1);
118 return ((struct savefpu *)p);
122 get_pcb_td(struct thread *td)
126 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
127 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) -
129 return ((struct pcb *)p);
133 alloc_fpusave(int flags)
136 struct savefpu_ymm *sf;
138 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
140 sf = (struct savefpu_ymm *)res;
141 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
142 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
148 * Finish a fork operation, with process p2 nearly set up.
149 * Copy and update the pcb, set up the stack so that the child
150 * ready to run and return to user mode.
153 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
157 struct mdproc *mdp1, *mdp2;
158 struct proc_ldt *pldt;
161 if ((flags & RFPROC) == 0) {
162 if ((flags & RFMEM) == 0) {
163 /* unshare user LDT */
166 if ((pldt = mdp1->md_ldt) != NULL &&
167 pldt->ldt_refcnt > 1 &&
168 user_ldt_alloc(p1, 1) == NULL)
169 panic("could not copy LDT");
170 mtx_unlock(&dt_lock);
175 /* Ensure that td1's pcb is up to date. */
177 update_pcb_bases(td1->td_pcb);
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;
238 td2->td_md.md_invl_gen.gen = 0;
240 /* As an i386, do not copy io permission bitmap. */
241 pcb2->pcb_tssp = NULL;
243 /* New segment registers. */
244 set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
246 /* Copy the LDT, if necessary. */
247 mdp1 = &td1->td_proc->p_md;
249 if (mdp1->md_ldt == NULL) {
254 if (mdp1->md_ldt != NULL) {
256 mdp1->md_ldt->ldt_refcnt++;
257 mdp2->md_ldt = mdp1->md_ldt;
258 bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct
259 system_segment_descriptor));
262 mdp2->md_ldt = user_ldt_alloc(p2, 0);
263 if (mdp2->md_ldt == NULL)
264 panic("could not copy LDT");
265 amd64_set_ldt_data(td2, 0, max_ldt_segment,
266 (struct user_segment_descriptor *)
267 mdp1->md_ldt->ldt_base);
271 mtx_unlock(&dt_lock);
274 * Now, cpu_switch() can schedule the new process.
275 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
276 * containing the return address when exiting cpu_switch.
277 * This will normally be to fork_trampoline(), which will have
278 * %ebx loaded with the new proc's pointer. fork_trampoline()
279 * will set up a stack to call fork_return(p, frame); to complete
280 * the return to user-mode.
285 * Intercept the return address from a freshly forked process that has NOT
286 * been scheduled yet.
288 * This is needed to make kernel threads stay in kernel mode.
291 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
294 * Note that the trap frame follows the args, so the function
295 * is really called like this: func(arg, frame);
297 td->td_pcb->pcb_r12 = (long) func; /* function */
298 td->td_pcb->pcb_rbx = (long) arg; /* first arg */
302 cpu_exit(struct thread *td)
306 * If this process has a custom LDT, release it.
308 if (td->td_proc->p_md.md_ldt != NULL)
313 cpu_thread_exit(struct thread *td)
318 if (td == PCPU_GET(fpcurthread))
324 /* Disable any hardware breakpoints. */
325 if (pcb->pcb_flags & PCB_DBREGS) {
327 clear_pcb_flags(pcb, PCB_DBREGS);
332 cpu_thread_clean(struct thread *td)
341 if (pcb->pcb_tssp != NULL) {
342 kmem_free(kernel_arena, (vm_offset_t)pcb->pcb_tssp,
344 pcb->pcb_tssp = NULL;
349 cpu_thread_swapin(struct thread *td)
354 cpu_thread_swapout(struct thread *td)
359 cpu_thread_alloc(struct thread *td)
362 struct xstate_hdr *xhdr;
364 td->td_pcb = pcb = get_pcb_td(td);
365 td->td_frame = (struct trapframe *)pcb - 1;
366 pcb->pcb_save = get_pcb_user_save_pcb(pcb);
368 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
369 bzero(xhdr, sizeof(*xhdr));
370 xhdr->xstate_bv = xsave_mask;
375 cpu_thread_free(struct thread *td)
378 cpu_thread_clean(td);
382 cpu_set_syscall_retval(struct thread *td, int error)
387 td->td_frame->tf_rax = td->td_retval[0];
388 td->td_frame->tf_rdx = td->td_retval[1];
389 td->td_frame->tf_rflags &= ~PSL_C;
394 * Reconstruct pc, we know that 'syscall' is 2 bytes,
395 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
396 * We saved this in tf_err.
397 * %r10 (which was holding the value of %rcx) is restored
398 * for the next iteration.
399 * %r10 restore is only required for freebsd/amd64 processes,
400 * but shall be innocent for any ia32 ABI.
402 * Require full context restore to get the arguments
403 * in the registers reloaded at return to usermode.
405 td->td_frame->tf_rip -= td->td_frame->tf_err;
406 td->td_frame->tf_r10 = td->td_frame->tf_rcx;
407 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
414 td->td_frame->tf_rax = SV_ABI_ERRNO(td->td_proc, error);
415 td->td_frame->tf_rflags |= PSL_C;
421 * Initialize machine state, mostly pcb and trap frame for a new
422 * thread, about to return to userspace. Put enough state in the new
423 * thread's PCB to get it to go back to the fork_return(), which
424 * finalizes the thread state and handles peculiarities of the first
425 * return to userspace for the new thread.
428 cpu_copy_thread(struct thread *td, struct thread *td0)
432 /* Point the pcb to the top of the stack. */
436 * Copy the upcall pcb. This loads kernel regs.
437 * Those not loaded individually below get their default
440 update_pcb_bases(td0->td_pcb);
441 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
442 clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE |
444 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
445 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
446 cpu_max_ext_state_size);
447 set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
450 * Create a new fresh stack for the new thread.
452 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
454 /* If the current thread has the trap bit set (i.e. a debugger had
455 * single stepped the process to the system call), we need to clear
456 * the trap flag from the new frame. Otherwise, the new thread will
457 * receive a (likely unexpected) SIGTRAP when it executes the first
458 * instruction after returning to userland.
460 td->td_frame->tf_rflags &= ~PSL_T;
463 * Set registers for trampoline to user mode. Leave space for the
464 * return address on stack. These are the kernel mode register values.
466 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */
468 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */
469 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */
470 pcb2->pcb_rip = (register_t)fork_trampoline;
472 * If we didn't copy the pcb, we'd need to do the following registers:
473 * pcb2->pcb_dr*: cloned above.
474 * pcb2->pcb_savefpu: cloned above.
475 * pcb2->pcb_onfault: cloned above (always NULL here?).
476 * pcb2->pcb_[fg]sbase: cloned above
479 /* Setup to release spin count in fork_exit(). */
480 td->td_md.md_spinlock_count = 1;
481 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
485 * Set that machine state for performing an upcall that starts
486 * the entry function with the given argument.
489 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
494 * Do any extra cleaning that needs to be done.
495 * The thread may have optional components
496 * that are not present in a fresh thread.
497 * This may be a recycled thread so make it look
498 * as though it's newly allocated.
500 cpu_thread_clean(td);
502 #ifdef COMPAT_FREEBSD32
503 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
505 * Set the trap frame to point at the beginning of the entry
508 td->td_frame->tf_rbp = 0;
509 td->td_frame->tf_rsp =
510 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
511 td->td_frame->tf_rip = (uintptr_t)entry;
513 /* Return address sentinel value to stop stack unwinding. */
514 suword32((void *)td->td_frame->tf_rsp, 0);
516 /* Pass the argument to the entry point. */
517 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)),
518 (uint32_t)(uintptr_t)arg);
525 * Set the trap frame to point at the beginning of the uts
528 td->td_frame->tf_rbp = 0;
529 td->td_frame->tf_rsp =
530 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f;
531 td->td_frame->tf_rsp -= 8;
532 td->td_frame->tf_rip = (register_t)entry;
533 td->td_frame->tf_ds = _udatasel;
534 td->td_frame->tf_es = _udatasel;
535 td->td_frame->tf_fs = _ufssel;
536 td->td_frame->tf_gs = _ugssel;
537 td->td_frame->tf_flags = TF_HASSEGS;
539 /* Return address sentinel value to stop stack unwinding. */
540 suword((void *)td->td_frame->tf_rsp, 0);
542 /* Pass the argument to the entry point. */
543 td->td_frame->tf_rdi = (register_t)arg;
547 cpu_set_user_tls(struct thread *td, void *tls_base)
551 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS)
555 set_pcb_flags(pcb, PCB_FULL_IRET);
556 #ifdef COMPAT_FREEBSD32
557 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
558 pcb->pcb_gsbase = (register_t)tls_base;
562 pcb->pcb_fsbase = (register_t)tls_base;
572 cpu_reset_proxy_active = 1;
573 while (cpu_reset_proxy_active == 1)
574 ia32_pause(); /* Wait for other cpu to see that we've started */
576 CPU_SETOF(cpu_reset_proxyid, &tcrp);
578 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
593 CPU_CLR(PCPU_GET(cpuid), &map);
594 CPU_NAND(&map, &stopped_cpus);
595 if (!CPU_EMPTY(&map)) {
596 printf("cpu_reset: Stopping other CPUs\n");
600 if (PCPU_GET(cpuid) != 0) {
601 cpu_reset_proxyid = PCPU_GET(cpuid);
602 cpustop_restartfunc = cpu_reset_proxy;
603 cpu_reset_proxy_active = 0;
604 printf("cpu_reset: Restarting BSP\n");
606 /* Restart CPU #0. */
607 CPU_SETOF(0, &started_cpus);
611 while (cpu_reset_proxy_active == 0 && cnt < 10000000) {
613 cnt++; /* Wait for BSP to announce restart */
615 if (cpu_reset_proxy_active == 0)
616 printf("cpu_reset: Failed to restart BSP\n");
618 cpu_reset_proxy_active = 2;
635 struct region_descriptor null_idt;
641 * Attempt to do a CPU reset via the keyboard controller,
642 * do not turn off GateA20, as any machine that fails
643 * to do the reset here would then end up in no man's land.
645 outb(IO_KBD + 4, 0xFE);
646 DELAY(500000); /* wait 0.5 sec to see if that did it */
649 * Attempt to force a reset via the Reset Control register at
650 * I/O port 0xcf9. Bit 2 forces a system reset when it
651 * transitions from 0 to 1. Bit 1 selects the type of reset
652 * to attempt: 0 selects a "soft" reset, and 1 selects a
653 * "hard" reset. We try a "hard" reset. The first write sets
654 * bit 1 to select a "hard" reset and clears bit 2. The
655 * second write forces a 0 -> 1 transition in bit 2 to trigger
660 DELAY(500000); /* wait 0.5 sec to see if that did it */
663 * Attempt to force a reset via the Fast A20 and Init register
664 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate.
665 * Bit 0 asserts INIT# when set to 1. We are careful to only
666 * preserve bit 1 while setting bit 0. We also must clear bit
667 * 0 before setting it if it isn't already clear.
672 outb(0x92, b & 0xfe);
674 DELAY(500000); /* wait 0.5 sec to see if that did it */
677 printf("No known reset method worked, attempting CPU shutdown\n");
678 DELAY(1000000); /* wait 1 sec for printf to complete */
681 null_idt.rd_limit = 0;
682 null_idt.rd_base = 0;
685 /* "good night, sweet prince .... <THUNK!>" */
693 * Software interrupt handler for queued VM system processing.
698 if (busdma_swi_pending != 0)
703 * Tell whether this address is in some physical memory region.
704 * Currently used by the kernel coredump code in order to avoid
705 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
706 * or other unpredictable behaviour.
710 is_physical_memory(vm_paddr_t addr)
714 /* The ISA ``memory hole''. */
715 if (addr >= 0xa0000 && addr < 0x100000)
720 * stuff other tests for known memory-mapped devices (PCI?)