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>
62 #include <sys/sf_buf.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
66 #include <sys/unistd.h>
67 #include <sys/vnode.h>
68 #include <sys/vmmeter.h>
70 #include <machine/cpu.h>
71 #include <machine/md_var.h>
72 #include <machine/pcb.h>
73 #include <machine/smp.h>
74 #include <machine/specialreg.h>
75 #include <machine/tss.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_kern.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_param.h>
84 #include <x86/isa/isa.h>
86 static void cpu_reset_real(void);
88 static void cpu_reset_proxy(void);
89 static u_int cpu_reset_proxyid;
90 static volatile u_int cpu_reset_proxy_active;
93 CTASSERT((struct thread **)OFFSETOF_CURTHREAD ==
94 &((struct pcpu *)NULL)->pc_curthread);
95 CTASSERT((struct pcb **)OFFSETOF_CURPCB == &((struct pcpu *)NULL)->pc_curpcb);
98 get_pcb_user_save_td(struct thread *td)
102 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
103 cpu_max_ext_state_size;
104 KASSERT((p % 64) == 0, ("Unaligned pcb_user_save area"));
105 return ((struct savefpu *)p);
109 get_pcb_user_save_pcb(struct pcb *pcb)
113 p = (vm_offset_t)(pcb + 1);
114 return ((struct savefpu *)p);
118 get_pcb_td(struct thread *td)
122 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
123 cpu_max_ext_state_size - sizeof(struct pcb);
124 return ((struct pcb *)p);
128 alloc_fpusave(int flags)
131 struct savefpu_ymm *sf;
133 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
135 sf = (struct savefpu_ymm *)res;
136 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
137 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
143 * Finish a fork operation, with process p2 nearly set up.
144 * Copy and update the pcb, set up the stack so that the child
145 * ready to run and return to user mode.
148 cpu_fork(td1, p2, td2, flags)
149 register struct thread *td1;
150 register struct proc *p2;
154 register struct proc *p1;
156 struct mdproc *mdp1, *mdp2;
157 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. */
178 /* Point the pcb to the top of the stack */
179 pcb2 = get_pcb_td(td2);
183 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
185 /* Properly initialize pcb_save */
186 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
187 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
188 cpu_max_ext_state_size);
190 /* Point mdproc and then copy over td1's contents */
192 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
195 * Create a new fresh stack for the new process.
196 * Copy the trap frame for the return to user mode as if from a
197 * syscall. This copies most of the user mode register values.
199 td2->td_frame = (struct trapframe *)td2->td_pcb - 1;
200 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
202 td2->td_frame->tf_rax = 0; /* Child returns zero */
203 td2->td_frame->tf_rflags &= ~PSL_C; /* success */
204 td2->td_frame->tf_rdx = 1;
207 * If the parent process has the trap bit set (i.e. a debugger had
208 * single stepped the process to the system call), we need to clear
209 * the trap flag from the new frame unless the debugger had set PF_FORK
210 * on the parent. Otherwise, the child will receive a (likely
211 * unexpected) SIGTRAP when it executes the first instruction after
212 * returning to userland.
214 if ((p1->p_pfsflags & PF_FORK) == 0)
215 td2->td_frame->tf_rflags &= ~PSL_T;
218 * Set registers for trampoline to user mode. Leave space for the
219 * return address on stack. These are the kernel mode register values.
221 pmap2 = vmspace_pmap(p2->p_vmspace);
222 pcb2->pcb_cr3 = DMAP_TO_PHYS((vm_offset_t)pmap2->pm_pml4);
223 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */
225 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *);
226 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */
227 pcb2->pcb_rip = (register_t)fork_trampoline;
229 * pcb2->pcb_dr*: cloned above.
230 * pcb2->pcb_savefpu: cloned above.
231 * pcb2->pcb_flags: cloned above.
232 * pcb2->pcb_onfault: cloned above (always NULL here?).
233 * pcb2->pcb_[fg]sbase: cloned above
236 /* Setup to release spin count in fork_exit(). */
237 td2->td_md.md_spinlock_count = 1;
238 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
240 /* As an i386, do not copy io permission bitmap. */
241 pcb2->pcb_tssp = NULL;
243 /* New segment registers. */
244 set_pcb_flags(pcb2, PCB_FULL_IRET);
246 /* Copy the LDT, if necessary. */
247 mdp1 = &td1->td_proc->p_md;
250 if (mdp1->md_ldt != NULL) {
252 mdp1->md_ldt->ldt_refcnt++;
253 mdp2->md_ldt = mdp1->md_ldt;
254 bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct
255 system_segment_descriptor));
258 mdp2->md_ldt = user_ldt_alloc(p2, 0);
259 if (mdp2->md_ldt == NULL)
260 panic("could not copy LDT");
261 amd64_set_ldt_data(td2, 0, max_ldt_segment,
262 (struct user_segment_descriptor *)
263 mdp1->md_ldt->ldt_base);
267 mtx_unlock(&dt_lock);
270 * Now, cpu_switch() can schedule the new process.
271 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
272 * containing the return address when exiting cpu_switch.
273 * This will normally be to fork_trampoline(), which will have
274 * %ebx loaded with the new proc's pointer. fork_trampoline()
275 * will set up a stack to call fork_return(p, frame); to complete
276 * the return to user-mode.
281 * Intercept the return address from a freshly forked process that has NOT
282 * been scheduled yet.
284 * This is needed to make kernel threads stay in kernel mode.
287 cpu_set_fork_handler(td, func, arg)
289 void (*func)(void *);
293 * Note that the trap frame follows the args, so the function
294 * is really called like this: func(arg, frame);
296 td->td_pcb->pcb_r12 = (long) func; /* function */
297 td->td_pcb->pcb_rbx = (long) arg; /* first arg */
301 cpu_exit(struct thread *td)
305 * If this process has a custom LDT, release it.
308 if (td->td_proc->p_md.md_ldt != 0)
311 mtx_unlock(&dt_lock);
315 cpu_thread_exit(struct thread *td)
320 if (td == PCPU_GET(fpcurthread))
326 /* Disable any hardware breakpoints. */
327 if (pcb->pcb_flags & PCB_DBREGS) {
329 clear_pcb_flags(pcb, PCB_DBREGS);
334 cpu_thread_clean(struct thread *td)
343 if (pcb->pcb_tssp != NULL) {
344 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_tssp,
346 pcb->pcb_tssp = NULL;
351 cpu_thread_swapin(struct thread *td)
356 cpu_thread_swapout(struct thread *td)
361 cpu_thread_alloc(struct thread *td)
364 struct xstate_hdr *xhdr;
366 td->td_pcb = pcb = get_pcb_td(td);
367 td->td_frame = (struct trapframe *)pcb - 1;
368 pcb->pcb_save = get_pcb_user_save_pcb(pcb);
370 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
371 bzero(xhdr, sizeof(*xhdr));
372 xhdr->xstate_bv = xsave_mask;
377 cpu_thread_free(struct thread *td)
380 cpu_thread_clean(td);
384 cpu_set_syscall_retval(struct thread *td, int error)
389 td->td_frame->tf_rax = td->td_retval[0];
390 td->td_frame->tf_rdx = td->td_retval[1];
391 td->td_frame->tf_rflags &= ~PSL_C;
396 * Reconstruct pc, we know that 'syscall' is 2 bytes,
397 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
398 * We saved this in tf_err.
399 * %r10 (which was holding the value of %rcx) is restored
400 * for the next iteration.
401 * %r10 restore is only required for freebsd/amd64 processes,
402 * but shall be innocent for any ia32 ABI.
404 * Require full context restore to get the arguments
405 * in the registers reloaded at return to usermode.
407 td->td_frame->tf_rip -= td->td_frame->tf_err;
408 td->td_frame->tf_r10 = td->td_frame->tf_rcx;
409 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
416 if (td->td_proc->p_sysent->sv_errsize) {
417 if (error >= td->td_proc->p_sysent->sv_errsize)
418 error = -1; /* XXX */
420 error = td->td_proc->p_sysent->sv_errtbl[error];
422 td->td_frame->tf_rax = error;
423 td->td_frame->tf_rflags |= PSL_C;
429 * Initialize machine state (pcb and trap frame) for a new thread about to
430 * upcall. Put enough state in the new thread's PCB to get it to go back
431 * userret(), where we can intercept it again to set the return (upcall)
432 * Address and stack, along with those from upcals that are from other sources
433 * such as those generated in thread_userret() itself.
436 cpu_set_upcall(struct thread *td, struct thread *td0)
440 /* Point the pcb to the top of the stack. */
444 * Copy the upcall pcb. This loads kernel regs.
445 * Those not loaded individually below get their default
448 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
449 clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE |
451 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
452 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
453 cpu_max_ext_state_size);
454 set_pcb_flags(pcb2, PCB_FULL_IRET);
457 * Create a new fresh stack for the new thread.
459 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
461 /* If the current thread has the trap bit set (i.e. a debugger had
462 * single stepped the process to the system call), we need to clear
463 * the trap flag from the new frame. Otherwise, the new thread will
464 * receive a (likely unexpected) SIGTRAP when it executes the first
465 * instruction after returning to userland.
467 td->td_frame->tf_rflags &= ~PSL_T;
470 * Set registers for trampoline to user mode. Leave space for the
471 * return address on stack. These are the kernel mode register values.
473 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */
475 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */
476 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */
477 pcb2->pcb_rip = (register_t)fork_trampoline;
479 * If we didn't copy the pcb, we'd need to do the following registers:
480 * pcb2->pcb_cr3: cloned above.
481 * pcb2->pcb_dr*: cloned above.
482 * pcb2->pcb_savefpu: cloned above.
483 * pcb2->pcb_onfault: cloned above (always NULL here?).
484 * pcb2->pcb_[fg]sbase: cloned above
487 /* Setup to release spin count in fork_exit(). */
488 td->td_md.md_spinlock_count = 1;
489 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
493 * Set that machine state for performing an upcall that has to
494 * be done in thread_userret() so that those upcalls generated
495 * in thread_userret() itself can be done as well.
498 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
503 * Do any extra cleaning that needs to be done.
504 * The thread may have optional components
505 * that are not present in a fresh thread.
506 * This may be a recycled thread so make it look
507 * as though it's newly allocated.
509 cpu_thread_clean(td);
511 #ifdef COMPAT_FREEBSD32
512 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
514 * Set the trap frame to point at the beginning of the uts
517 td->td_frame->tf_rbp = 0;
518 td->td_frame->tf_rsp =
519 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
520 td->td_frame->tf_rip = (uintptr_t)entry;
523 * Pass the address of the mailbox for this kse to the uts
524 * function as a parameter on the stack.
526 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)),
527 (uint32_t)(uintptr_t)arg);
534 * Set the trap frame to point at the beginning of the uts
537 td->td_frame->tf_rbp = 0;
538 td->td_frame->tf_rsp =
539 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f;
540 td->td_frame->tf_rsp -= 8;
541 td->td_frame->tf_rip = (register_t)entry;
542 td->td_frame->tf_ds = _udatasel;
543 td->td_frame->tf_es = _udatasel;
544 td->td_frame->tf_fs = _ufssel;
545 td->td_frame->tf_gs = _ugssel;
546 td->td_frame->tf_flags = TF_HASSEGS;
549 * Pass the address of the mailbox for this kse to the uts
550 * function as a parameter on the stack.
552 td->td_frame->tf_rdi = (register_t)arg;
556 cpu_set_user_tls(struct thread *td, void *tls_base)
560 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS)
564 set_pcb_flags(pcb, PCB_FULL_IRET);
565 #ifdef COMPAT_FREEBSD32
566 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
567 pcb->pcb_gsbase = (register_t)tls_base;
571 pcb->pcb_fsbase = (register_t)tls_base;
581 cpu_reset_proxy_active = 1;
582 while (cpu_reset_proxy_active == 1)
583 ; /* Wait for other cpu to see that we've started */
584 CPU_SETOF(cpu_reset_proxyid, &tcrp);
586 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
601 CPU_CLR(PCPU_GET(cpuid), &map);
602 CPU_NAND(&map, &stopped_cpus);
603 if (!CPU_EMPTY(&map)) {
604 printf("cpu_reset: Stopping other CPUs\n");
608 if (PCPU_GET(cpuid) != 0) {
609 cpu_reset_proxyid = PCPU_GET(cpuid);
610 cpustop_restartfunc = cpu_reset_proxy;
611 cpu_reset_proxy_active = 0;
612 printf("cpu_reset: Restarting BSP\n");
614 /* Restart CPU #0. */
615 CPU_SETOF(0, &started_cpus);
619 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
620 cnt++; /* Wait for BSP to announce restart */
621 if (cpu_reset_proxy_active == 0)
622 printf("cpu_reset: Failed to restart BSP\n");
624 cpu_reset_proxy_active = 2;
640 struct region_descriptor null_idt;
646 * Attempt to do a CPU reset via the keyboard controller,
647 * do not turn off GateA20, as any machine that fails
648 * to do the reset here would then end up in no man's land.
650 outb(IO_KBD + 4, 0xFE);
651 DELAY(500000); /* wait 0.5 sec to see if that did it */
654 * Attempt to force a reset via the Reset Control register at
655 * I/O port 0xcf9. Bit 2 forces a system reset when it
656 * transitions from 0 to 1. Bit 1 selects the type of reset
657 * to attempt: 0 selects a "soft" reset, and 1 selects a
658 * "hard" reset. We try a "hard" reset. The first write sets
659 * bit 1 to select a "hard" reset and clears bit 2. The
660 * second write forces a 0 -> 1 transition in bit 2 to trigger
665 DELAY(500000); /* wait 0.5 sec to see if that did it */
668 * Attempt to force a reset via the Fast A20 and Init register
669 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate.
670 * Bit 0 asserts INIT# when set to 1. We are careful to only
671 * preserve bit 1 while setting bit 0. We also must clear bit
672 * 0 before setting it if it isn't already clear.
677 outb(0x92, b & 0xfe);
679 DELAY(500000); /* wait 0.5 sec to see if that did it */
682 printf("No known reset method worked, attempting CPU shutdown\n");
683 DELAY(1000000); /* wait 1 sec for printf to complete */
686 null_idt.rd_limit = 0;
687 null_idt.rd_base = 0;
690 /* "good night, sweet prince .... <THUNK!>" */
698 * Allocate an sf_buf for the given vm_page. On this machine, however, there
699 * is no sf_buf object. Instead, an opaque pointer to the given vm_page is
703 sf_buf_alloc(struct vm_page *m, int pri)
706 return ((struct sf_buf *)m);
710 * Free the sf_buf. In fact, do nothing because there are no resources
711 * associated with the sf_buf.
714 sf_buf_free(struct sf_buf *sf)
719 * Software interrupt handler for queued VM system processing.
724 if (busdma_swi_pending != 0)
729 * Tell whether this address is in some physical memory region.
730 * Currently used by the kernel coredump code in order to avoid
731 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
732 * or other unpredictable behaviour.
736 is_physical_memory(vm_paddr_t addr)
740 /* The ISA ``memory hole''. */
741 if (addr >= 0xa0000 && addr < 0x100000)
746 * stuff other tests for known memory-mapped devices (PCI?)