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("$FreeBSD$");
51 #include <sys/param.h>
52 #include <sys/systm.h>
55 #include <sys/kernel.h>
58 #include <sys/malloc.h>
60 #include <sys/mutex.h>
61 #include <sys/pioctl.h>
64 #include <sys/procctl.h>
66 #include <sys/sysctl.h>
67 #include <sys/sysent.h>
68 #include <sys/unistd.h>
69 #include <sys/vnode.h>
70 #include <sys/vmmeter.h>
73 #include <machine/cpu.h>
74 #include <machine/md_var.h>
75 #include <machine/pcb.h>
76 #include <machine/smp.h>
77 #include <machine/specialreg.h>
78 #include <machine/tss.h>
81 #include <vm/vm_extern.h>
82 #include <vm/vm_kern.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_param.h>
87 _Static_assert(OFFSETOF_CURTHREAD == offsetof(struct pcpu, pc_curthread),
88 "OFFSETOF_CURTHREAD does not correspond with offset of pc_curthread.");
89 _Static_assert(OFFSETOF_CURPCB == offsetof(struct pcpu, pc_curpcb),
90 "OFFSETOF_CURPCB does not correspond with offset of pc_curpcb.");
91 _Static_assert(OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf),
92 "OFFSETOF_MONITORBUF does not correspond with offset of pc_monitorbuf.");
95 get_pcb_user_save_td(struct thread *td)
99 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
100 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
101 KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area"));
102 return ((struct savefpu *)p);
106 get_pcb_user_save_pcb(struct pcb *pcb)
110 p = (vm_offset_t)(pcb + 1);
111 return ((struct savefpu *)p);
115 get_pcb_td(struct thread *td)
119 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
120 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) -
122 return ((struct pcb *)p);
126 alloc_fpusave(int flags)
129 struct savefpu_ymm *sf;
131 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
133 sf = (struct savefpu_ymm *)res;
134 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
135 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
141 * Finish a fork operation, with process p2 nearly set up.
142 * Copy and update the pcb, set up the stack so that the child
143 * ready to run and return to user mode.
146 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
150 struct mdproc *mdp1, *mdp2;
151 struct proc_ldt *pldt;
154 if ((flags & RFPROC) == 0) {
155 if ((flags & RFMEM) == 0) {
156 /* unshare user LDT */
159 if ((pldt = mdp1->md_ldt) != NULL &&
160 pldt->ldt_refcnt > 1 &&
161 user_ldt_alloc(p1, 1) == NULL)
162 panic("could not copy LDT");
163 mtx_unlock(&dt_lock);
168 /* Ensure that td1's pcb is up to date. */
170 update_pcb_bases(td1->td_pcb);
172 /* Point the pcb to the top of the stack */
173 pcb2 = get_pcb_td(td2);
177 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
179 /* Properly initialize pcb_save */
180 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
181 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
182 cpu_max_ext_state_size);
184 /* Point mdproc and then copy over td1's contents */
186 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
189 * Create a new fresh stack for the new process.
190 * Copy the trap frame for the return to user mode as if from a
191 * syscall. This copies most of the user mode register values.
193 td2->td_frame = (struct trapframe *)td2->td_pcb - 1;
194 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
196 td2->td_frame->tf_rax = 0; /* Child returns zero */
197 td2->td_frame->tf_rflags &= ~PSL_C; /* success */
198 td2->td_frame->tf_rdx = 1;
201 * If the parent process has the trap bit set (i.e. a debugger had
202 * single stepped the process to the system call), we need to clear
203 * the trap flag from the new frame unless the debugger had set PF_FORK
204 * on the parent. Otherwise, the child will receive a (likely
205 * unexpected) SIGTRAP when it executes the first instruction after
206 * returning to userland.
208 if ((p1->p_pfsflags & PF_FORK) == 0)
209 td2->td_frame->tf_rflags &= ~PSL_T;
212 * Set registers for trampoline to user mode. Leave space for the
213 * return address on stack. These are the kernel mode register values.
215 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */
217 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *);
218 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */
219 pcb2->pcb_rip = (register_t)fork_trampoline;
221 * pcb2->pcb_dr*: cloned above.
222 * pcb2->pcb_savefpu: cloned above.
223 * pcb2->pcb_flags: cloned above.
224 * pcb2->pcb_onfault: cloned above (always NULL here?).
225 * pcb2->pcb_[fg]sbase: cloned above
228 /* Setup to release spin count in fork_exit(). */
229 td2->td_md.md_spinlock_count = 1;
230 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
231 td2->td_md.md_invl_gen.gen = 0;
233 /* As an i386, do not copy io permission bitmap. */
234 pcb2->pcb_tssp = NULL;
236 /* New segment registers. */
237 set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
239 /* Copy the LDT, if necessary. */
240 mdp1 = &td1->td_proc->p_md;
242 if (mdp1->md_ldt == NULL) {
247 if (mdp1->md_ldt != NULL) {
249 mdp1->md_ldt->ldt_refcnt++;
250 mdp2->md_ldt = mdp1->md_ldt;
251 bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct
252 system_segment_descriptor));
255 mdp2->md_ldt = user_ldt_alloc(p2, 0);
256 if (mdp2->md_ldt == NULL)
257 panic("could not copy LDT");
258 amd64_set_ldt_data(td2, 0, max_ldt_segment,
259 (struct user_segment_descriptor *)
260 mdp1->md_ldt->ldt_base);
264 mtx_unlock(&dt_lock);
267 * Now, cpu_switch() can schedule the new process.
268 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
269 * containing the return address when exiting cpu_switch.
270 * This will normally be to fork_trampoline(), which will have
271 * %ebx loaded with the new proc's pointer. fork_trampoline()
272 * will set up a stack to call fork_return(p, frame); to complete
273 * the return to user-mode.
278 * Intercept the return address from a freshly forked process that has NOT
279 * been scheduled yet.
281 * This is needed to make kernel threads stay in kernel mode.
284 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
287 * Note that the trap frame follows the args, so the function
288 * is really called like this: func(arg, frame);
290 td->td_pcb->pcb_r12 = (long) func; /* function */
291 td->td_pcb->pcb_rbx = (long) arg; /* first arg */
295 cpu_exit(struct thread *td)
299 * If this process has a custom LDT, release it.
301 if (td->td_proc->p_md.md_ldt != NULL)
306 cpu_thread_exit(struct thread *td)
311 if (td == PCPU_GET(fpcurthread))
317 /* Disable any hardware breakpoints. */
318 if (pcb->pcb_flags & PCB_DBREGS) {
320 clear_pcb_flags(pcb, PCB_DBREGS);
325 cpu_thread_clean(struct thread *td)
334 if (pcb->pcb_tssp != NULL) {
335 pmap_pti_remove_kva((vm_offset_t)pcb->pcb_tssp,
336 (vm_offset_t)pcb->pcb_tssp + ctob(IOPAGES + 1));
337 kmem_free((vm_offset_t)pcb->pcb_tssp, ctob(IOPAGES + 1));
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_exec_vmspace_reuse(struct proc *p, vm_map_t map)
379 return (((curproc->p_md.md_flags & P_MD_KPTI) != 0) ==
380 (vm_map_pmap(map)->pm_ucr3 != PMAP_NO_CR3));
384 cpu_procctl_kpti(struct proc *p, int com, int *val)
387 if (com == PROC_KPTI_CTL) {
388 if (pti && *val == PROC_KPTI_CTL_ENABLE_ON_EXEC)
389 p->p_md.md_flags |= P_MD_KPTI;
390 if (*val == PROC_KPTI_CTL_DISABLE_ON_EXEC)
391 p->p_md.md_flags &= ~P_MD_KPTI;
392 } else /* PROC_KPTI_STATUS */ {
393 *val = (p->p_md.md_flags & P_MD_KPTI) != 0 ?
394 PROC_KPTI_CTL_ENABLE_ON_EXEC:
395 PROC_KPTI_CTL_DISABLE_ON_EXEC;
396 if (vmspace_pmap(p->p_vmspace)->pm_ucr3 != PMAP_NO_CR3)
397 *val |= PROC_KPTI_STATUS_ACTIVE;
402 cpu_procctl(struct thread *td, int idtype, id_t id, int com, void *data)
409 case PROC_KPTI_STATUS:
410 if (idtype != P_PID) {
414 if (com == PROC_KPTI_CTL) {
415 /* sad but true and not a joke */
416 error = priv_check(td, PRIV_IO);
419 error = copyin(data, &val, sizeof(val));
422 if (val != PROC_KPTI_CTL_ENABLE_ON_EXEC &&
423 val != PROC_KPTI_CTL_DISABLE_ON_EXEC) {
428 error = pget(id, PGET_CANSEE | PGET_NOTWEXIT | PGET_NOTID, &p);
430 cpu_procctl_kpti(p, com, &val);
432 if (com == PROC_KPTI_STATUS)
433 error = copyout(&val, data, sizeof(val));
444 cpu_set_syscall_retval(struct thread *td, int error)
446 struct trapframe *frame;
448 frame = td->td_frame;
449 if (__predict_true(error == 0)) {
450 frame->tf_rax = td->td_retval[0];
451 frame->tf_rdx = td->td_retval[1];
452 frame->tf_rflags &= ~PSL_C;
459 * Reconstruct pc, we know that 'syscall' is 2 bytes,
460 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
461 * We saved this in tf_err.
462 * %r10 (which was holding the value of %rcx) is restored
463 * for the next iteration.
464 * %r10 restore is only required for freebsd/amd64 processes,
465 * but shall be innocent for any ia32 ABI.
467 * Require full context restore to get the arguments
468 * in the registers reloaded at return to usermode.
470 frame->tf_rip -= frame->tf_err;
471 frame->tf_r10 = frame->tf_rcx;
472 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
479 frame->tf_rax = SV_ABI_ERRNO(td->td_proc, error);
480 frame->tf_rflags |= PSL_C;
486 * Initialize machine state, mostly pcb and trap frame for a new
487 * thread, about to return to userspace. Put enough state in the new
488 * thread's PCB to get it to go back to the fork_return(), which
489 * finalizes the thread state and handles peculiarities of the first
490 * return to userspace for the new thread.
493 cpu_copy_thread(struct thread *td, struct thread *td0)
497 /* Point the pcb to the top of the stack. */
501 * Copy the upcall pcb. This loads kernel regs.
502 * Those not loaded individually below get their default
505 update_pcb_bases(td0->td_pcb);
506 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
507 clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE |
509 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
510 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
511 cpu_max_ext_state_size);
512 set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
515 * Create a new fresh stack for the new thread.
517 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
519 /* If the current thread has the trap bit set (i.e. a debugger had
520 * single stepped the process to the system call), we need to clear
521 * the trap flag from the new frame. Otherwise, the new thread will
522 * receive a (likely unexpected) SIGTRAP when it executes the first
523 * instruction after returning to userland.
525 td->td_frame->tf_rflags &= ~PSL_T;
528 * Set registers for trampoline to user mode. Leave space for the
529 * return address on stack. These are the kernel mode register values.
531 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */
533 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */
534 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */
535 pcb2->pcb_rip = (register_t)fork_trampoline;
537 * If we didn't copy the pcb, we'd need to do the following registers:
538 * pcb2->pcb_dr*: cloned above.
539 * pcb2->pcb_savefpu: cloned above.
540 * pcb2->pcb_onfault: cloned above (always NULL here?).
541 * pcb2->pcb_[fg]sbase: cloned above
544 /* Setup to release spin count in fork_exit(). */
545 td->td_md.md_spinlock_count = 1;
546 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
550 * Set that machine state for performing an upcall that starts
551 * the entry function with the given argument.
554 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
559 * Do any extra cleaning that needs to be done.
560 * The thread may have optional components
561 * that are not present in a fresh thread.
562 * This may be a recycled thread so make it look
563 * as though it's newly allocated.
565 cpu_thread_clean(td);
567 #ifdef COMPAT_FREEBSD32
568 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
570 * Set the trap frame to point at the beginning of the entry
573 td->td_frame->tf_rbp = 0;
574 td->td_frame->tf_rsp =
575 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
576 td->td_frame->tf_rip = (uintptr_t)entry;
578 /* Return address sentinel value to stop stack unwinding. */
579 suword32((void *)td->td_frame->tf_rsp, 0);
581 /* Pass the argument to the entry point. */
582 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)),
583 (uint32_t)(uintptr_t)arg);
590 * Set the trap frame to point at the beginning of the uts
593 td->td_frame->tf_rbp = 0;
594 td->td_frame->tf_rsp =
595 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f;
596 td->td_frame->tf_rsp -= 8;
597 td->td_frame->tf_rip = (register_t)entry;
598 td->td_frame->tf_ds = _udatasel;
599 td->td_frame->tf_es = _udatasel;
600 td->td_frame->tf_fs = _ufssel;
601 td->td_frame->tf_gs = _ugssel;
602 td->td_frame->tf_flags = TF_HASSEGS;
604 /* Return address sentinel value to stop stack unwinding. */
605 suword((void *)td->td_frame->tf_rsp, 0);
607 /* Pass the argument to the entry point. */
608 td->td_frame->tf_rdi = (register_t)arg;
612 cpu_set_user_tls(struct thread *td, void *tls_base)
616 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS)
620 set_pcb_flags(pcb, PCB_FULL_IRET);
621 #ifdef COMPAT_FREEBSD32
622 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
623 pcb->pcb_gsbase = (register_t)tls_base;
627 pcb->pcb_fsbase = (register_t)tls_base;
632 * Software interrupt handler for queued VM system processing.
637 if (busdma_swi_pending != 0)
642 * Tell whether this address is in some physical memory region.
643 * Currently used by the kernel coredump code in order to avoid
644 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
645 * or other unpredictable behaviour.
649 is_physical_memory(vm_paddr_t addr)
653 /* The ISA ``memory hole''. */
654 if (addr >= 0xa0000 && addr < 0x100000)
659 * stuff other tests for known memory-mapped devices (PCI?)