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>
63 #include <sys/procctl.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>
72 #include <machine/cpu.h>
73 #include <machine/md_var.h>
74 #include <machine/pcb.h>
75 #include <machine/smp.h>
76 #include <machine/specialreg.h>
77 #include <machine/tss.h>
80 #include <vm/vm_extern.h>
81 #include <vm/vm_kern.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_param.h>
86 _Static_assert(OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf),
87 "OFFSETOF_MONITORBUF does not correspond with offset of pc_monitorbuf.");
90 set_top_of_stack_td(struct thread *td)
92 td->td_md.md_stack_base = td->td_kstack +
93 td->td_kstack_pages * PAGE_SIZE -
94 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
98 get_pcb_user_save_td(struct thread *td)
102 p = td->td_md.md_stack_base;
103 KASSERT((p % XSAVE_AREA_ALIGN) == 0,
104 ("Unaligned pcb_user_save area ptr %#lx td %p", p, td));
105 return ((struct savefpu *)p);
109 get_pcb_td(struct thread *td)
112 return (&td->td_md.md_pcb);
116 get_pcb_user_save_pcb(struct pcb *pcb)
120 td = __containerof(pcb, struct thread, td_md.md_pcb);
121 return (get_pcb_user_save_td(td));
125 alloc_fpusave(int flags)
128 struct savefpu_ymm *sf;
130 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
132 sf = (struct savefpu_ymm *)res;
133 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
134 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
140 * Finish a fork operation, with process p2 nearly set up.
141 * Copy and update the pcb, set up the stack so that the child
142 * ready to run and return to user mode.
145 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
149 struct mdproc *mdp1, *mdp2;
150 struct proc_ldt *pldt;
153 if ((flags & RFPROC) == 0) {
154 if ((flags & RFMEM) == 0) {
155 /* unshare user LDT */
158 if ((pldt = mdp1->md_ldt) != NULL &&
159 pldt->ldt_refcnt > 1 &&
160 user_ldt_alloc(p1, 1) == NULL)
161 panic("could not copy LDT");
162 mtx_unlock(&dt_lock);
167 /* Ensure that td1's pcb is up to date. */
169 update_pcb_bases(td1->td_pcb);
171 /* Point the stack and pcb to the actual location */
172 set_top_of_stack_td(td2);
173 td2->td_pcb = pcb2 = get_pcb_td(td2);
176 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
178 /* Properly initialize pcb_save */
179 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
180 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
181 cpu_max_ext_state_size);
183 /* Point mdproc and then copy over td1's contents */
185 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
188 * Create a new fresh stack for the new process.
189 * Copy the trap frame for the return to user mode as if from a
190 * syscall. This copies most of the user mode register values.
192 td2->td_frame = (struct trapframe *)td2->td_md.md_stack_base - 1;
193 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
195 td2->td_frame->tf_rax = 0; /* Child returns zero */
196 td2->td_frame->tf_rflags &= ~PSL_C; /* success */
197 td2->td_frame->tf_rdx = 1;
200 * If the parent process has the trap bit set (i.e. a debugger
201 * had single stepped the process to the system call), we need
202 * to clear the trap flag from the new frame.
204 td2->td_frame->tf_rflags &= ~PSL_T;
207 * Set registers for trampoline to user mode. Leave space for the
208 * return address on stack. These are the kernel mode register values.
210 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */
212 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *);
213 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */
214 pcb2->pcb_rip = (register_t)fork_trampoline;
216 * pcb2->pcb_dr*: cloned above.
217 * pcb2->pcb_savefpu: cloned above.
218 * pcb2->pcb_flags: cloned above.
219 * pcb2->pcb_onfault: cloned above (always NULL here?).
220 * pcb2->pcb_[fg]sbase: cloned above
223 /* Setup to release spin count in fork_exit(). */
224 td2->td_md.md_spinlock_count = 1;
225 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
226 pmap_thread_init_invl_gen(td2);
228 /* As an i386, do not copy io permission bitmap. */
229 pcb2->pcb_tssp = NULL;
231 /* New segment registers. */
232 set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
234 /* Copy the LDT, if necessary. */
235 mdp1 = &td1->td_proc->p_md;
237 if (mdp1->md_ldt == NULL) {
242 if (mdp1->md_ldt != NULL) {
244 mdp1->md_ldt->ldt_refcnt++;
245 mdp2->md_ldt = mdp1->md_ldt;
246 bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct
247 system_segment_descriptor));
250 mdp2->md_ldt = user_ldt_alloc(p2, 0);
251 if (mdp2->md_ldt == NULL)
252 panic("could not copy LDT");
253 amd64_set_ldt_data(td2, 0, max_ldt_segment,
254 (struct user_segment_descriptor *)
255 mdp1->md_ldt->ldt_base);
259 mtx_unlock(&dt_lock);
262 * Now, cpu_switch() can schedule the new process.
263 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
264 * containing the return address when exiting cpu_switch.
265 * This will normally be to fork_trampoline(), which will have
266 * %ebx loaded with the new proc's pointer. fork_trampoline()
267 * will set up a stack to call fork_return(p, frame); to complete
268 * the return to user-mode.
273 * Intercept the return address from a freshly forked process that has NOT
274 * been scheduled yet.
276 * This is needed to make kernel threads stay in kernel mode.
279 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
282 * Note that the trap frame follows the args, so the function
283 * is really called like this: func(arg, frame);
285 td->td_pcb->pcb_r12 = (long) func; /* function */
286 td->td_pcb->pcb_rbx = (long) arg; /* first arg */
290 cpu_exit(struct thread *td)
294 * If this process has a custom LDT, release it.
296 if (td->td_proc->p_md.md_ldt != NULL)
301 cpu_thread_exit(struct thread *td)
306 if (td == PCPU_GET(fpcurthread))
312 /* Disable any hardware breakpoints. */
313 if (pcb->pcb_flags & PCB_DBREGS) {
315 clear_pcb_flags(pcb, PCB_DBREGS);
320 cpu_thread_clean(struct thread *td)
329 if (pcb->pcb_tssp != NULL) {
330 pmap_pti_remove_kva((vm_offset_t)pcb->pcb_tssp,
331 (vm_offset_t)pcb->pcb_tssp + ctob(IOPAGES + 1));
332 kmem_free((vm_offset_t)pcb->pcb_tssp, ctob(IOPAGES + 1));
333 pcb->pcb_tssp = NULL;
338 cpu_thread_swapin(struct thread *td)
343 cpu_thread_swapout(struct thread *td)
348 cpu_thread_alloc(struct thread *td)
351 struct xstate_hdr *xhdr;
353 set_top_of_stack_td(td);
354 td->td_pcb = pcb = get_pcb_td(td);
355 td->td_frame = (struct trapframe *)td->td_md.md_stack_base - 1;
356 pcb->pcb_save = get_pcb_user_save_pcb(pcb);
358 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
359 bzero(xhdr, sizeof(*xhdr));
360 xhdr->xstate_bv = xsave_mask;
365 cpu_thread_free(struct thread *td)
368 cpu_thread_clean(td);
372 cpu_exec_vmspace_reuse(struct proc *p, vm_map_t map)
375 return (((curproc->p_md.md_flags & P_MD_KPTI) != 0) ==
376 (vm_map_pmap(map)->pm_ucr3 != PMAP_NO_CR3));
380 cpu_procctl_kpti_ctl(struct proc *p, int val)
383 if (pti && val == PROC_KPTI_CTL_ENABLE_ON_EXEC)
384 p->p_md.md_flags |= P_MD_KPTI;
385 if (val == PROC_KPTI_CTL_DISABLE_ON_EXEC)
386 p->p_md.md_flags &= ~P_MD_KPTI;
390 cpu_procctl_kpti_status(struct proc *p, int *val)
392 *val = (p->p_md.md_flags & P_MD_KPTI) != 0 ?
393 PROC_KPTI_CTL_ENABLE_ON_EXEC:
394 PROC_KPTI_CTL_DISABLE_ON_EXEC;
395 if (vmspace_pmap(p->p_vmspace)->pm_ucr3 != PMAP_NO_CR3)
396 *val |= PROC_KPTI_STATUS_ACTIVE;
400 cpu_procctl_la_ctl(struct proc *p, int val)
406 case PROC_LA_CTL_LA48_ON_EXEC:
407 p->p_md.md_flags |= P_MD_LA48;
408 p->p_md.md_flags &= ~P_MD_LA57;
410 case PROC_LA_CTL_LA57_ON_EXEC:
412 p->p_md.md_flags &= ~P_MD_LA48;
413 p->p_md.md_flags |= P_MD_LA57;
418 case PROC_LA_CTL_DEFAULT_ON_EXEC:
419 p->p_md.md_flags &= ~(P_MD_LA48 | P_MD_LA57);
426 cpu_procctl_la_status(struct proc *p, int *val)
430 if ((p->p_md.md_flags & P_MD_LA48) != 0)
431 res = PROC_LA_CTL_LA48_ON_EXEC;
432 else if ((p->p_md.md_flags & P_MD_LA57) != 0)
433 res = PROC_LA_CTL_LA57_ON_EXEC;
435 res = PROC_LA_CTL_DEFAULT_ON_EXEC;
436 if (p->p_sysent->sv_maxuser == VM_MAXUSER_ADDRESS_LA48)
437 res |= PROC_LA_STATUS_LA48;
439 res |= PROC_LA_STATUS_LA57;
444 cpu_procctl(struct thread *td, int idtype, id_t id, int com, void *data)
451 case PROC_KPTI_STATUS:
454 if (idtype != P_PID) {
458 if (com == PROC_KPTI_CTL) {
459 /* sad but true and not a joke */
460 error = priv_check(td, PRIV_IO);
464 if (com == PROC_KPTI_CTL || com == PROC_LA_CTL) {
465 error = copyin(data, &val, sizeof(val));
469 if (com == PROC_KPTI_CTL &&
470 val != PROC_KPTI_CTL_ENABLE_ON_EXEC &&
471 val != PROC_KPTI_CTL_DISABLE_ON_EXEC) {
475 if (com == PROC_LA_CTL &&
476 val != PROC_LA_CTL_LA48_ON_EXEC &&
477 val != PROC_LA_CTL_LA57_ON_EXEC &&
478 val != PROC_LA_CTL_DEFAULT_ON_EXEC) {
482 error = pget(id, PGET_CANSEE | PGET_NOTWEXIT | PGET_NOTID, &p);
487 cpu_procctl_kpti_ctl(p, val);
489 case PROC_KPTI_STATUS:
490 cpu_procctl_kpti_status(p, &val);
493 error = cpu_procctl_la_ctl(p, val);
496 cpu_procctl_la_status(p, &val);
500 if (com == PROC_KPTI_STATUS || com == PROC_LA_STATUS)
501 error = copyout(&val, data, sizeof(val));
511 cpu_set_syscall_retval(struct thread *td, int error)
513 struct trapframe *frame;
515 frame = td->td_frame;
516 if (__predict_true(error == 0)) {
517 frame->tf_rax = td->td_retval[0];
518 frame->tf_rdx = td->td_retval[1];
519 frame->tf_rflags &= ~PSL_C;
526 * Reconstruct pc, we know that 'syscall' is 2 bytes,
527 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
528 * We saved this in tf_err.
529 * %r10 (which was holding the value of %rcx) is restored
530 * for the next iteration.
531 * %r10 restore is only required for freebsd/amd64 processes,
532 * but shall be innocent for any ia32 ABI.
534 * Require full context restore to get the arguments
535 * in the registers reloaded at return to usermode.
537 frame->tf_rip -= frame->tf_err;
538 frame->tf_r10 = frame->tf_rcx;
539 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
546 frame->tf_rax = error;
547 frame->tf_rflags |= PSL_C;
553 * Initialize machine state, mostly pcb and trap frame for a new
554 * thread, about to return to userspace. Put enough state in the new
555 * thread's PCB to get it to go back to the fork_return(), which
556 * finalizes the thread state and handles peculiarities of the first
557 * return to userspace for the new thread.
560 cpu_copy_thread(struct thread *td, struct thread *td0)
567 * Copy the upcall pcb. This loads kernel regs.
568 * Those not loaded individually below get their default
571 update_pcb_bases(td0->td_pcb);
572 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
573 clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE |
575 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
576 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
577 cpu_max_ext_state_size);
578 set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
581 * Create a new fresh stack for the new thread.
583 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
585 /* If the current thread has the trap bit set (i.e. a debugger had
586 * single stepped the process to the system call), we need to clear
587 * the trap flag from the new frame. Otherwise, the new thread will
588 * receive a (likely unexpected) SIGTRAP when it executes the first
589 * instruction after returning to userland.
591 td->td_frame->tf_rflags &= ~PSL_T;
594 * Set registers for trampoline to user mode. Leave space for the
595 * return address on stack. These are the kernel mode register values.
597 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */
599 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */
600 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */
601 pcb2->pcb_rip = (register_t)fork_trampoline;
603 * If we didn't copy the pcb, we'd need to do the following registers:
604 * pcb2->pcb_dr*: cloned above.
605 * pcb2->pcb_savefpu: cloned above.
606 * pcb2->pcb_onfault: cloned above (always NULL here?).
607 * pcb2->pcb_[fg]sbase: cloned above
610 /* Setup to release spin count in fork_exit(). */
611 td->td_md.md_spinlock_count = 1;
612 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
613 pmap_thread_init_invl_gen(td);
617 * Set that machine state for performing an upcall that starts
618 * the entry function with the given argument.
621 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
626 * Do any extra cleaning that needs to be done.
627 * The thread may have optional components
628 * that are not present in a fresh thread.
629 * This may be a recycled thread so make it look
630 * as though it's newly allocated.
632 cpu_thread_clean(td);
634 #ifdef COMPAT_FREEBSD32
635 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
637 * Set the trap frame to point at the beginning of the entry
640 td->td_frame->tf_rbp = 0;
641 td->td_frame->tf_rsp =
642 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
643 td->td_frame->tf_rip = (uintptr_t)entry;
645 /* Return address sentinel value to stop stack unwinding. */
646 suword32((void *)td->td_frame->tf_rsp, 0);
648 /* Pass the argument to the entry point. */
649 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)),
650 (uint32_t)(uintptr_t)arg);
657 * Set the trap frame to point at the beginning of the uts
660 td->td_frame->tf_rbp = 0;
661 td->td_frame->tf_rsp =
662 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f;
663 td->td_frame->tf_rsp -= 8;
664 td->td_frame->tf_rip = (register_t)entry;
665 td->td_frame->tf_ds = _udatasel;
666 td->td_frame->tf_es = _udatasel;
667 td->td_frame->tf_fs = _ufssel;
668 td->td_frame->tf_gs = _ugssel;
669 td->td_frame->tf_flags = TF_HASSEGS;
671 /* Return address sentinel value to stop stack unwinding. */
672 suword((void *)td->td_frame->tf_rsp, 0);
674 /* Pass the argument to the entry point. */
675 td->td_frame->tf_rdi = (register_t)arg;
679 cpu_set_user_tls(struct thread *td, void *tls_base)
683 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS)
687 set_pcb_flags(pcb, PCB_FULL_IRET);
688 #ifdef COMPAT_FREEBSD32
689 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
690 pcb->pcb_gsbase = (register_t)tls_base;
694 pcb->pcb_fsbase = (register_t)tls_base;
699 * Software interrupt handler for queued VM system processing.
704 if (busdma_swi_pending != 0)
709 * Tell whether this address is in some physical memory region.
710 * Currently used by the kernel coredump code in order to avoid
711 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
712 * or other unpredictable behaviour.
716 is_physical_memory(vm_paddr_t addr)
720 /* The ISA ``memory hole''. */
721 if (addr >= 0xa0000 && addr < 0x100000)
726 * stuff other tests for known memory-mapped devices (PCI?)