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_reset.h"
53 #include <sys/param.h>
54 #include <sys/systm.h>
57 #include <sys/kernel.h>
60 #include <sys/malloc.h>
62 #include <sys/mutex.h>
63 #include <sys/pioctl.h>
65 #include <sys/sysent.h>
66 #include <sys/sf_buf.h>
68 #include <sys/sched.h>
69 #include <sys/sysctl.h>
70 #include <sys/unistd.h>
71 #include <sys/vnode.h>
72 #include <sys/vmmeter.h>
74 #include <machine/cpu.h>
75 #include <machine/cputypes.h>
76 #include <machine/md_var.h>
77 #include <machine/pcb.h>
78 #include <machine/pcb_ext.h>
79 #include <machine/smp.h>
80 #include <machine/vm86.h>
83 #include <vm/vm_extern.h>
84 #include <vm/vm_kern.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_map.h>
87 #include <vm/vm_param.h>
90 #define NSFBUFS (512 + maxusers * 16)
93 _Static_assert(__OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf),
94 "__OFFSETOF_MONITORBUF does not correspond with offset of pc_monitorbuf.");
97 get_pcb_user_save_td(struct thread *td)
101 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
102 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
103 KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area"));
104 return ((union savefpu *)p);
108 get_pcb_user_save_pcb(struct pcb *pcb)
112 p = (vm_offset_t)(pcb + 1);
113 return ((union savefpu *)p);
117 get_pcb_td(struct thread *td)
121 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
122 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) -
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;
142 * Finish a fork operation, with process p2 nearly set up.
143 * Copy and update the pcb, set up the stack so that the child
144 * ready to run and return to user mode.
147 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
154 if ((flags & RFPROC) == 0) {
155 if ((flags & RFMEM) == 0) {
156 /* unshare user LDT */
157 struct mdproc *mdp1 = &p1->p_md;
158 struct proc_ldt *pldt, *pldt1;
160 mtx_lock_spin(&dt_lock);
161 if ((pldt1 = mdp1->md_ldt) != NULL &&
162 pldt1->ldt_refcnt > 1) {
163 pldt = user_ldt_alloc(mdp1, pldt1->ldt_len);
165 panic("could not copy LDT");
168 user_ldt_deref(pldt1);
170 mtx_unlock_spin(&dt_lock);
175 /* Ensure that td1's pcb is up to date. */
176 if (td1 == curthread)
177 td1->td_pcb->pcb_gs = rgs();
179 if (PCPU_GET(fpcurthread) == td1)
180 npxsave(td1->td_pcb->pcb_save);
183 /* Point the pcb to the top of the stack */
184 pcb2 = get_pcb_td(td2);
188 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
190 /* Properly initialize pcb_save */
191 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
192 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
193 cpu_max_ext_state_size);
195 /* Point mdproc and then copy over td1's contents */
197 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
200 * Create a new fresh stack for the new process.
201 * Copy the trap frame for the return to user mode as if from a
202 * syscall. This copies most of the user mode register values.
203 * The -VM86_STACK_SPACE (-16) is so we can expand the trapframe
206 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb -
207 VM86_STACK_SPACE) - 1;
208 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
210 td2->td_frame->tf_eax = 0; /* Child returns zero */
211 td2->td_frame->tf_eflags &= ~PSL_C; /* success */
212 td2->td_frame->tf_edx = 1;
215 * If the parent process has the trap bit set (i.e. a debugger had
216 * single stepped the process to the system call), we need to clear
217 * the trap flag from the new frame unless the debugger had set PF_FORK
218 * on the parent. Otherwise, the child will receive a (likely
219 * unexpected) SIGTRAP when it executes the first instruction after
220 * returning to userland.
222 if ((p1->p_pfsflags & PF_FORK) == 0)
223 td2->td_frame->tf_eflags &= ~PSL_T;
226 * Set registers for trampoline to user mode. Leave space for the
227 * return address on stack. These are the kernel mode register values.
229 pcb2->pcb_cr3 = pmap_get_cr3(vmspace_pmap(p2->p_vmspace));
231 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
233 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
234 pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */
235 pcb2->pcb_eip = (int)fork_trampoline + setidt_disp;
237 * pcb2->pcb_dr*: cloned above.
238 * pcb2->pcb_savefpu: cloned above.
239 * pcb2->pcb_flags: cloned above.
240 * pcb2->pcb_onfault: cloned above (always NULL here?).
241 * pcb2->pcb_gs: cloned above.
242 * pcb2->pcb_ext: cleared below.
246 * XXX don't copy the i/o pages. this should probably be fixed.
250 /* Copy the LDT, if necessary. */
251 mtx_lock_spin(&dt_lock);
252 if (mdp2->md_ldt != NULL) {
254 mdp2->md_ldt->ldt_refcnt++;
256 mdp2->md_ldt = user_ldt_alloc(mdp2,
257 mdp2->md_ldt->ldt_len);
258 if (mdp2->md_ldt == NULL)
259 panic("could not copy LDT");
262 mtx_unlock_spin(&dt_lock);
264 /* Setup to release spin count in fork_exit(). */
265 td2->td_md.md_spinlock_count = 1;
266 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
269 * Now, cpu_switch() can schedule the new process.
270 * pcb_esp 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_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
289 * Note that the trap frame follows the args, so the function
290 * is really called like this: func(arg, frame);
292 td->td_pcb->pcb_esi = (int) func; /* function */
293 td->td_pcb->pcb_ebx = (int) arg; /* first arg */
297 cpu_exit(struct thread *td)
301 * If this process has a custom LDT, release it. Reset pc->pcb_gs
302 * and %gs before we free it in case they refer to an LDT entry.
304 mtx_lock_spin(&dt_lock);
305 if (td->td_proc->p_md.md_ldt) {
306 td->td_pcb->pcb_gs = _udatasel;
310 mtx_unlock_spin(&dt_lock);
314 cpu_thread_exit(struct thread *td)
318 if (td == PCPU_GET(fpcurthread))
322 /* Disable any hardware breakpoints. */
323 if (td->td_pcb->pcb_flags & PCB_DBREGS) {
325 td->td_pcb->pcb_flags &= ~PCB_DBREGS;
330 cpu_thread_clean(struct thread *td)
335 if (pcb->pcb_ext != NULL) {
336 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
338 * XXX do we need to move the TSS off the allocated pages
339 * before freeing them? (not done here)
341 pmap_trm_free(pcb->pcb_ext, ctob(IOPAGES + 1));
347 cpu_thread_swapin(struct thread *td)
352 cpu_thread_swapout(struct thread *td)
357 cpu_thread_alloc(struct thread *td)
360 struct xstate_hdr *xhdr;
362 td->td_pcb = pcb = get_pcb_td(td);
363 td->td_frame = (struct trapframe *)((caddr_t)pcb -
364 VM86_STACK_SPACE) - 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_exec_vmspace_reuse(struct proc *p __unused, vm_map_t map __unused)
389 cpu_procctl(struct thread *td __unused, int idtype __unused, id_t id __unused,
390 int com __unused, void *data __unused)
397 cpu_set_syscall_retval(struct thread *td, int error)
402 td->td_frame->tf_eax = td->td_retval[0];
403 td->td_frame->tf_edx = td->td_retval[1];
404 td->td_frame->tf_eflags &= ~PSL_C;
409 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int
410 * 0x80 is 2 bytes. We saved this in tf_err.
412 td->td_frame->tf_eip -= td->td_frame->tf_err;
419 td->td_frame->tf_eax = SV_ABI_ERRNO(td->td_proc, error);
420 td->td_frame->tf_eflags |= PSL_C;
426 * Initialize machine state, mostly pcb and trap frame for a new
427 * thread, about to return to userspace. Put enough state in the new
428 * thread's PCB to get it to go back to the fork_return(), which
429 * finalizes the thread state and handles peculiarities of the first
430 * return to userspace for the new thread.
433 cpu_copy_thread(struct thread *td, struct thread *td0)
437 /* Point the pcb to the top of the stack. */
441 * Copy the upcall pcb. This loads kernel regs.
442 * Those not loaded individually below get their default
445 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
446 pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE |
448 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
449 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
450 cpu_max_ext_state_size);
453 * Create a new fresh stack for the new thread.
455 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
457 /* If the current thread has the trap bit set (i.e. a debugger had
458 * single stepped the process to the system call), we need to clear
459 * the trap flag from the new frame. Otherwise, the new thread will
460 * receive a (likely unexpected) SIGTRAP when it executes the first
461 * instruction after returning to userland.
463 td->td_frame->tf_eflags &= ~PSL_T;
466 * Set registers for trampoline to user mode. Leave space for the
467 * return address on stack. These are the kernel mode register values.
470 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */
472 pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
473 pcb2->pcb_ebx = (int)td; /* trampoline arg */
474 pcb2->pcb_eip = (int)fork_trampoline + setidt_disp;
475 pcb2->pcb_gs = rgs();
477 * If we didn't copy the pcb, we'd need to do the following registers:
478 * pcb2->pcb_cr3: cloned above.
479 * pcb2->pcb_dr*: cloned above.
480 * pcb2->pcb_savefpu: cloned above.
481 * pcb2->pcb_flags: cloned above.
482 * pcb2->pcb_onfault: cloned above (always NULL here?).
483 * pcb2->pcb_gs: cloned above.
484 * pcb2->pcb_ext: cleared below.
486 pcb2->pcb_ext = NULL;
488 /* Setup to release spin count in fork_exit(). */
489 td->td_md.md_spinlock_count = 1;
490 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
494 * Set that machine state for performing an upcall that starts
495 * the entry function with the given argument.
498 cpu_set_upcall(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);
512 * Set the trap frame to point at the beginning of the entry
515 td->td_frame->tf_ebp = 0;
516 td->td_frame->tf_esp =
517 (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
518 td->td_frame->tf_eip = (int)entry;
520 /* Return address sentinel value to stop stack unwinding. */
521 suword((void *)td->td_frame->tf_esp, 0);
523 /* Pass the argument to the entry point. */
524 suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
529 cpu_set_user_tls(struct thread *td, void *tls_base)
531 struct segment_descriptor sd;
535 * Construct a descriptor and store it in the pcb for
536 * the next context switch. Also store it in the gdt
537 * so that the load of tf_fs into %fs will activate it
538 * at return to userland.
540 base = (uint32_t)tls_base;
541 sd.sd_lobase = base & 0xffffff;
542 sd.sd_hibase = (base >> 24) & 0xff;
543 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
545 sd.sd_type = SDT_MEMRWA;
553 td->td_pcb->pcb_gsd = sd;
554 if (td == curthread) {
555 PCPU_GET(fsgs_gdt)[1] = sd;
556 load_gs(GSEL(GUGS_SEL, SEL_UPL));
563 * Convert kernel VA to physical address
570 pa = pmap_kextract((vm_offset_t)addr);
572 panic("kvtop: zero page frame");
577 * Get an sf_buf from the freelist. May block if none are available.
580 sf_buf_map(struct sf_buf *sf, int flags)
585 sf_buf_shootdown(sf, flags);
591 sf_buf_shootdown(struct sf_buf *sf, int flags)
597 cpuid = PCPU_GET(cpuid);
598 if (!CPU_ISSET(cpuid, &sf->cpumask)) {
599 CPU_SET(cpuid, &sf->cpumask);
602 if ((flags & SFB_CPUPRIVATE) == 0) {
603 other_cpus = all_cpus;
604 CPU_CLR(cpuid, &other_cpus);
605 CPU_NAND(&other_cpus, &sf->cpumask);
606 if (!CPU_EMPTY(&other_cpus)) {
607 CPU_OR(&sf->cpumask, &other_cpus);
608 smp_masked_invlpg(other_cpus, sf->kva, kernel_pmap);
616 * MD part of sf_buf_free().
619 sf_buf_unmap(struct sf_buf *sf)
626 sf_buf_invalidate(struct sf_buf *sf)
631 * Use pmap_qenter to update the pte for
632 * existing mapping, in particular, the PAT
633 * settings are recalculated.
635 pmap_qenter(sf->kva, &m, 1);
636 pmap_invalidate_cache_range(sf->kva, sf->kva + PAGE_SIZE);
640 * Invalidate the cache lines that may belong to the page, if
641 * (possibly old) mapping of the page by sf buffer exists. Returns
642 * TRUE when mapping was found and cache invalidated.
645 sf_buf_invalidate_cache(vm_page_t m)
648 return (sf_buf_process_page(m, sf_buf_invalidate));
652 * Software interrupt handler for queued VM system processing.
657 if (busdma_swi_pending != 0)
662 * Tell whether this address is in some physical memory region.
663 * Currently used by the kernel coredump code in order to avoid
664 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
665 * or other unpredictable behaviour.
669 is_physical_memory(vm_paddr_t addr)
673 /* The ISA ``memory hole''. */
674 if (addr >= 0xa0000 && addr < 0x100000)
679 * stuff other tests for known memory-mapped devices (PCI?)