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 * Common code shared between cpu_fork() and cpu_copy_thread() for
141 * initializing a thread.
144 copy_thread(struct thread *td1, struct thread *td2)
150 /* Ensure that td1's pcb is up to date for user threads. */
151 if ((td2->td_pflags & TDP_KTHREAD) == 0) {
152 MPASS(td1 == curthread);
154 update_pcb_bases(td1->td_pcb);
158 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
160 /* Properly initialize pcb_save */
161 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
163 /* Kernel threads start with clean FPU and segment bases. */
164 if ((td2->td_pflags & TDP_KTHREAD) != 0) {
165 pcb2->pcb_fsbase = 0;
166 pcb2->pcb_gsbase = 0;
167 clear_pcb_flags(pcb2, PCB_FPUINITDONE | PCB_USERFPUINITDONE |
168 PCB_KERNFPU | PCB_KERNFPU_THR);
170 MPASS((pcb2->pcb_flags & (PCB_KERNFPU | PCB_KERNFPU_THR)) == 0);
171 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
172 cpu_max_ext_state_size);
176 * Set registers for trampoline to user mode. Leave space for the
177 * return address on stack. These are the kernel mode register values.
179 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */
181 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *);
182 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */
183 pcb2->pcb_rip = (register_t)fork_trampoline;
185 * pcb2->pcb_dr*: cloned above.
186 * pcb2->pcb_savefpu: cloned above.
187 * pcb2->pcb_flags: cloned above.
188 * pcb2->pcb_onfault: cloned above (always NULL here?).
189 * pcb2->pcb_[fg]sbase: cloned above
192 pcb2->pcb_tssp = NULL;
194 /* Setup to release spin count in fork_exit(). */
195 td2->td_md.md_spinlock_count = 1;
196 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
197 pmap_thread_init_invl_gen(td2);
201 * Finish a fork operation, with process p2 nearly set up.
202 * Copy and update the pcb, set up the stack so that the child
203 * ready to run and return to user mode.
206 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
210 struct mdproc *mdp1, *mdp2;
211 struct proc_ldt *pldt;
214 if ((flags & RFPROC) == 0) {
215 if ((flags & RFMEM) == 0) {
216 /* unshare user LDT */
219 if ((pldt = mdp1->md_ldt) != NULL &&
220 pldt->ldt_refcnt > 1 &&
221 user_ldt_alloc(p1, 1) == NULL)
222 panic("could not copy LDT");
223 mtx_unlock(&dt_lock);
228 /* Point the stack and pcb to the actual location */
229 set_top_of_stack_td(td2);
230 td2->td_pcb = pcb2 = get_pcb_td(td2);
232 copy_thread(td1, td2);
234 /* Reset debug registers in the new process */
235 x86_clear_dbregs(pcb2);
237 /* Point mdproc and then copy over p1's contents */
239 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
242 * Copy the trap frame for the return to user mode as if from a
243 * syscall. This copies most of the user mode register values.
245 td2->td_frame = (struct trapframe *)td2->td_md.md_stack_base - 1;
246 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
248 td2->td_frame->tf_rax = 0; /* Child returns zero */
249 td2->td_frame->tf_rflags &= ~PSL_C; /* success */
250 td2->td_frame->tf_rdx = 1;
253 * If the parent process has the trap bit set (i.e. a debugger
254 * had single stepped the process to the system call), we need
255 * to clear the trap flag from the new frame.
257 td2->td_frame->tf_rflags &= ~PSL_T;
259 /* As on i386, do not copy io permission bitmap. */
260 pcb2->pcb_tssp = NULL;
262 /* New segment registers. */
263 set_pcb_flags_raw(pcb2, PCB_FULL_IRET);
265 /* Copy the LDT, if necessary. */
266 mdp1 = &td1->td_proc->p_md;
268 if (mdp1->md_ldt == NULL) {
273 if (mdp1->md_ldt != NULL) {
275 mdp1->md_ldt->ldt_refcnt++;
276 mdp2->md_ldt = mdp1->md_ldt;
277 bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct
278 system_segment_descriptor));
281 mdp2->md_ldt = user_ldt_alloc(p2, 0);
282 if (mdp2->md_ldt == NULL)
283 panic("could not copy LDT");
284 amd64_set_ldt_data(td2, 0, max_ldt_segment,
285 (struct user_segment_descriptor *)
286 mdp1->md_ldt->ldt_base);
290 mtx_unlock(&dt_lock);
293 * Now, cpu_switch() can schedule the new process.
294 * pcb_rsp is loaded pointing to the cpu_switch() stack frame
295 * containing the return address when exiting cpu_switch.
296 * This will normally be to fork_trampoline(), which will have
297 * %rbx loaded with the new proc's pointer. fork_trampoline()
298 * will set up a stack to call fork_return(p, frame); to complete
299 * the return to user-mode.
304 * Intercept the return address from a freshly forked process that has NOT
305 * been scheduled yet.
307 * This is needed to make kernel threads stay in kernel mode.
310 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
313 * Note that the trap frame follows the args, so the function
314 * is really called like this: func(arg, frame);
316 td->td_pcb->pcb_r12 = (long) func; /* function */
317 td->td_pcb->pcb_rbx = (long) arg; /* first arg */
321 cpu_exit(struct thread *td)
325 * If this process has a custom LDT, release it.
327 if (td->td_proc->p_md.md_ldt != NULL)
332 cpu_thread_exit(struct thread *td)
337 if (td == PCPU_GET(fpcurthread))
343 /* Disable any hardware breakpoints. */
344 if (pcb->pcb_flags & PCB_DBREGS) {
346 clear_pcb_flags(pcb, PCB_DBREGS);
351 cpu_thread_clean(struct thread *td)
360 if (pcb->pcb_tssp != NULL) {
361 pmap_pti_remove_kva((vm_offset_t)pcb->pcb_tssp,
362 (vm_offset_t)pcb->pcb_tssp + ctob(IOPAGES + 1));
363 kmem_free((vm_offset_t)pcb->pcb_tssp, ctob(IOPAGES + 1));
364 pcb->pcb_tssp = NULL;
369 cpu_thread_swapin(struct thread *td)
374 cpu_thread_swapout(struct thread *td)
379 cpu_thread_alloc(struct thread *td)
382 struct xstate_hdr *xhdr;
384 set_top_of_stack_td(td);
385 td->td_pcb = pcb = get_pcb_td(td);
386 td->td_frame = (struct trapframe *)td->td_md.md_stack_base - 1;
387 pcb->pcb_save = get_pcb_user_save_pcb(pcb);
389 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
390 bzero(xhdr, sizeof(*xhdr));
391 xhdr->xstate_bv = xsave_mask;
396 cpu_thread_free(struct thread *td)
399 cpu_thread_clean(td);
403 cpu_exec_vmspace_reuse(struct proc *p, vm_map_t map)
406 return (((curproc->p_md.md_flags & P_MD_KPTI) != 0) ==
407 (vm_map_pmap(map)->pm_ucr3 != PMAP_NO_CR3));
411 cpu_procctl_kpti_ctl(struct proc *p, int val)
414 if (pti && val == PROC_KPTI_CTL_ENABLE_ON_EXEC)
415 p->p_md.md_flags |= P_MD_KPTI;
416 if (val == PROC_KPTI_CTL_DISABLE_ON_EXEC)
417 p->p_md.md_flags &= ~P_MD_KPTI;
421 cpu_procctl_kpti_status(struct proc *p, int *val)
423 *val = (p->p_md.md_flags & P_MD_KPTI) != 0 ?
424 PROC_KPTI_CTL_ENABLE_ON_EXEC:
425 PROC_KPTI_CTL_DISABLE_ON_EXEC;
426 if (vmspace_pmap(p->p_vmspace)->pm_ucr3 != PMAP_NO_CR3)
427 *val |= PROC_KPTI_STATUS_ACTIVE;
431 cpu_procctl_la_ctl(struct proc *p, int val)
437 case PROC_LA_CTL_LA48_ON_EXEC:
438 p->p_md.md_flags |= P_MD_LA48;
439 p->p_md.md_flags &= ~P_MD_LA57;
441 case PROC_LA_CTL_LA57_ON_EXEC:
443 p->p_md.md_flags &= ~P_MD_LA48;
444 p->p_md.md_flags |= P_MD_LA57;
449 case PROC_LA_CTL_DEFAULT_ON_EXEC:
450 p->p_md.md_flags &= ~(P_MD_LA48 | P_MD_LA57);
457 cpu_procctl_la_status(struct proc *p, int *val)
461 if ((p->p_md.md_flags & P_MD_LA48) != 0)
462 res = PROC_LA_CTL_LA48_ON_EXEC;
463 else if ((p->p_md.md_flags & P_MD_LA57) != 0)
464 res = PROC_LA_CTL_LA57_ON_EXEC;
466 res = PROC_LA_CTL_DEFAULT_ON_EXEC;
467 if (p->p_sysent->sv_maxuser == VM_MAXUSER_ADDRESS_LA48)
468 res |= PROC_LA_STATUS_LA48;
470 res |= PROC_LA_STATUS_LA57;
475 cpu_procctl(struct thread *td, int idtype, id_t id, int com, void *data)
482 case PROC_KPTI_STATUS:
485 if (idtype != P_PID) {
489 if (com == PROC_KPTI_CTL) {
490 /* sad but true and not a joke */
491 error = priv_check(td, PRIV_IO);
495 if (com == PROC_KPTI_CTL || com == PROC_LA_CTL) {
496 error = copyin(data, &val, sizeof(val));
500 if (com == PROC_KPTI_CTL &&
501 val != PROC_KPTI_CTL_ENABLE_ON_EXEC &&
502 val != PROC_KPTI_CTL_DISABLE_ON_EXEC) {
506 if (com == PROC_LA_CTL &&
507 val != PROC_LA_CTL_LA48_ON_EXEC &&
508 val != PROC_LA_CTL_LA57_ON_EXEC &&
509 val != PROC_LA_CTL_DEFAULT_ON_EXEC) {
513 error = pget(id, PGET_CANSEE | PGET_NOTWEXIT | PGET_NOTID, &p);
518 cpu_procctl_kpti_ctl(p, val);
520 case PROC_KPTI_STATUS:
521 cpu_procctl_kpti_status(p, &val);
524 error = cpu_procctl_la_ctl(p, val);
527 cpu_procctl_la_status(p, &val);
531 if (com == PROC_KPTI_STATUS || com == PROC_LA_STATUS)
532 error = copyout(&val, data, sizeof(val));
542 cpu_set_syscall_retval(struct thread *td, int error)
544 struct trapframe *frame;
546 frame = td->td_frame;
547 if (__predict_true(error == 0)) {
548 frame->tf_rax = td->td_retval[0];
549 frame->tf_rdx = td->td_retval[1];
550 frame->tf_rflags &= ~PSL_C;
557 * Reconstruct pc, we know that 'syscall' is 2 bytes,
558 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
559 * We saved this in tf_err.
560 * %r10 (which was holding the value of %rcx) is restored
561 * for the next iteration.
562 * %r10 restore is only required for freebsd/amd64 processes,
563 * but shall be innocent for any ia32 ABI.
565 * Require full context restore to get the arguments
566 * in the registers reloaded at return to usermode.
568 frame->tf_rip -= frame->tf_err;
569 frame->tf_r10 = frame->tf_rcx;
570 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
577 frame->tf_rax = error;
578 frame->tf_rflags |= PSL_C;
584 * Initialize machine state, mostly pcb and trap frame for a new
585 * thread, about to return to userspace. Put enough state in the new
586 * thread's PCB to get it to go back to the fork_return(), which
587 * finalizes the thread state and handles peculiarities of the first
588 * return to userspace for the new thread.
591 cpu_copy_thread(struct thread *td, struct thread *td0)
593 copy_thread(td0, td);
596 * Copy user general-purpose registers.
598 * Some of these registers are rewritten by cpu_set_upcall()
599 * and linux_set_upcall().
601 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
603 /* If the current thread has the trap bit set (i.e. a debugger had
604 * single stepped the process to the system call), we need to clear
605 * the trap flag from the new frame. Otherwise, the new thread will
606 * receive a (likely unexpected) SIGTRAP when it executes the first
607 * instruction after returning to userland.
609 td->td_frame->tf_rflags &= ~PSL_T;
611 set_pcb_flags_raw(td->td_pcb, PCB_FULL_IRET);
615 * Set that machine state for performing an upcall that starts
616 * the entry function with the given argument.
619 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
624 * Do any extra cleaning that needs to be done.
625 * The thread may have optional components
626 * that are not present in a fresh thread.
627 * This may be a recycled thread so make it look
628 * as though it's newly allocated.
630 cpu_thread_clean(td);
632 #ifdef COMPAT_FREEBSD32
633 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
635 * Set the trap frame to point at the beginning of the entry
638 td->td_frame->tf_rbp = 0;
639 td->td_frame->tf_rsp =
640 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
641 td->td_frame->tf_rip = (uintptr_t)entry;
643 /* Return address sentinel value to stop stack unwinding. */
644 suword32((void *)td->td_frame->tf_rsp, 0);
646 /* Pass the argument to the entry point. */
647 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)),
648 (uint32_t)(uintptr_t)arg);
655 * Set the trap frame to point at the beginning of the uts
658 td->td_frame->tf_rbp = 0;
659 td->td_frame->tf_rsp =
660 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f;
661 td->td_frame->tf_rsp -= 8;
662 td->td_frame->tf_rip = (register_t)entry;
663 td->td_frame->tf_ds = _udatasel;
664 td->td_frame->tf_es = _udatasel;
665 td->td_frame->tf_fs = _ufssel;
666 td->td_frame->tf_gs = _ugssel;
667 td->td_frame->tf_flags = TF_HASSEGS;
669 /* Return address sentinel value to stop stack unwinding. */
670 suword((void *)td->td_frame->tf_rsp, 0);
672 /* Pass the argument to the entry point. */
673 td->td_frame->tf_rdi = (register_t)arg;
677 cpu_set_user_tls(struct thread *td, void *tls_base)
681 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS)
685 set_pcb_flags(pcb, PCB_FULL_IRET);
686 #ifdef COMPAT_FREEBSD32
687 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
688 pcb->pcb_gsbase = (register_t)tls_base;
692 pcb->pcb_fsbase = (register_t)tls_base;
697 * Software interrupt handler for queued VM system processing.
702 if (busdma_swi_pending != 0)
707 * Tell whether this address is in some physical memory region.
708 * Currently used by the kernel coredump code in order to avoid
709 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
710 * or other unpredictable behaviour.
714 is_physical_memory(vm_paddr_t addr)
718 /* The ISA ``memory hole''. */
719 if (addr >= 0xa0000 && addr < 0x100000)
724 * stuff other tests for known memory-mapped devices (PCI?)