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>
64 #include <sys/sysent.h>
65 #include <sys/sf_buf.h>
67 #include <sys/sched.h>
68 #include <sys/sysctl.h>
69 #include <sys/unistd.h>
70 #include <sys/vnode.h>
71 #include <sys/vmmeter.h>
73 #include <machine/cpu.h>
74 #include <machine/cputypes.h>
75 #include <machine/md_var.h>
76 #include <machine/pcb.h>
77 #include <machine/pcb_ext.h>
78 #include <machine/smp.h>
79 #include <machine/vm86.h>
82 #include <vm/vm_extern.h>
83 #include <vm/vm_kern.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_param.h>
88 _Static_assert(__OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf),
89 "__OFFSETOF_MONITORBUF does not correspond with offset of pc_monitorbuf.");
92 get_pcb_user_save_td(struct thread *td)
96 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
97 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
98 KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area"));
99 return ((union savefpu *)p);
103 get_pcb_user_save_pcb(struct pcb *pcb)
107 p = (vm_offset_t)(pcb + 1);
108 return ((union savefpu *)p);
112 get_pcb_td(struct thread *td)
116 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
117 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) -
119 return ((struct pcb *)p);
123 alloc_fpusave(int flags)
126 struct savefpu_ymm *sf;
128 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
130 sf = (struct savefpu_ymm *)res;
131 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
132 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
138 * Common code shared between cpu_fork() and cpu_copy_thread() for
139 * initializing a thread.
142 copy_thread(struct thread *td1, struct thread *td2)
148 /* Ensure that td1's pcb is up to date for user threads. */
149 if ((td2->td_pflags & TDP_KTHREAD) == 0) {
150 MPASS(td1 == curthread);
151 td1->td_pcb->pcb_gs = rgs();
153 if (PCPU_GET(fpcurthread) == td1)
154 npxsave(td1->td_pcb->pcb_save);
159 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
161 /* Properly initialize pcb_save */
162 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
164 /* Kernel threads start with clean NPX and segment bases. */
165 if ((td2->td_pflags & TDP_KTHREAD) != 0) {
166 pcb2->pcb_gs = _udatasel;
169 pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE |
170 PCB_KERNNPX | PCB_KERNNPX_THR);
172 MPASS((pcb2->pcb_flags & (PCB_KERNNPX | PCB_KERNNPX_THR)) == 0);
173 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
174 cpu_max_ext_state_size);
178 * Set registers for trampoline to user mode. Leave space for the
179 * return address on stack. These are the kernel mode register values.
182 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */
184 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *); /* trampoline arg */
185 pcb2->pcb_ebx = (int)td2; /* trampoline arg */
186 pcb2->pcb_eip = (int)fork_trampoline + setidt_disp;
188 * If we didn't copy the pcb, we'd need to do the following registers:
189 * pcb2->pcb_cr3: cloned above.
190 * pcb2->pcb_dr*: cloned above.
191 * pcb2->pcb_savefpu: cloned above.
192 * pcb2->pcb_flags: cloned above.
193 * pcb2->pcb_onfault: cloned above (always NULL here?).
194 * pcb2->pcb_gs: cloned above.
195 * pcb2->pcb_ext: cleared below.
197 pcb2->pcb_ext = NULL;
199 /* Setup to release spin count in fork_exit(). */
200 td2->td_md.md_spinlock_count = 1;
201 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
205 * Finish a fork operation, with process p2 nearly set up.
206 * Copy and update the pcb, set up the stack so that the child
207 * ready to run and return to user mode.
210 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
217 if ((flags & RFPROC) == 0) {
218 if ((flags & RFMEM) == 0) {
219 /* unshare user LDT */
220 struct mdproc *mdp1 = &p1->p_md;
221 struct proc_ldt *pldt, *pldt1;
223 mtx_lock_spin(&dt_lock);
224 if ((pldt1 = mdp1->md_ldt) != NULL &&
225 pldt1->ldt_refcnt > 1) {
226 pldt = user_ldt_alloc(mdp1, pldt1->ldt_len);
228 panic("could not copy LDT");
231 user_ldt_deref(pldt1);
233 mtx_unlock_spin(&dt_lock);
238 /* Point the pcb to the top of the stack */
239 pcb2 = get_pcb_td(td2);
242 copy_thread(td1, td2);
244 /* Point mdproc and then copy over td1's contents */
246 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
249 * Copy the trap frame for the return to user mode as if from a
250 * syscall. This copies most of the user mode register values.
251 * The -VM86_STACK_SPACE (-16) is so we can expand the trapframe
254 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb -
255 VM86_STACK_SPACE) - 1;
256 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
258 td2->td_frame->tf_eax = 0; /* Child returns zero */
259 td2->td_frame->tf_eflags &= ~PSL_C; /* success */
260 td2->td_frame->tf_edx = 1;
263 * If the parent process has the trap bit set (i.e. a debugger
264 * had single stepped the process to the system call), we need
265 * to clear the trap flag from the new frame.
267 td2->td_frame->tf_eflags &= ~PSL_T;
269 /* Set cr3 for the new process. */
270 pcb2->pcb_cr3 = pmap_get_cr3(vmspace_pmap(p2->p_vmspace));
273 * XXX don't copy the i/o pages. this should probably be fixed.
275 pcb2->pcb_ext = NULL;
277 /* Copy the LDT, if necessary. */
278 mtx_lock_spin(&dt_lock);
279 if (mdp2->md_ldt != NULL) {
281 mdp2->md_ldt->ldt_refcnt++;
283 mdp2->md_ldt = user_ldt_alloc(mdp2,
284 mdp2->md_ldt->ldt_len);
285 if (mdp2->md_ldt == NULL)
286 panic("could not copy LDT");
289 mtx_unlock_spin(&dt_lock);
292 * Now, cpu_switch() can schedule the new process.
293 * pcb_esp is loaded pointing to the cpu_switch() stack frame
294 * containing the return address when exiting cpu_switch.
295 * This will normally be to fork_trampoline(), which will have
296 * %ebx loaded with the new proc's pointer. fork_trampoline()
297 * will set up a stack to call fork_return(p, frame); to complete
298 * the return to user-mode.
303 * Intercept the return address from a freshly forked process that has NOT
304 * been scheduled yet.
306 * This is needed to make kernel threads stay in kernel mode.
309 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
312 * Note that the trap frame follows the args, so the function
313 * is really called like this: func(arg, frame);
315 td->td_pcb->pcb_esi = (int) func; /* function */
316 td->td_pcb->pcb_ebx = (int) arg; /* first arg */
320 cpu_exit(struct thread *td)
324 * If this process has a custom LDT, release it. Reset pc->pcb_gs
325 * and %gs before we free it in case they refer to an LDT entry.
327 mtx_lock_spin(&dt_lock);
328 if (td->td_proc->p_md.md_ldt) {
329 td->td_pcb->pcb_gs = _udatasel;
333 mtx_unlock_spin(&dt_lock);
337 cpu_thread_exit(struct thread *td)
341 if (td == PCPU_GET(fpcurthread))
345 /* Disable any hardware breakpoints. */
346 if (td->td_pcb->pcb_flags & PCB_DBREGS) {
348 td->td_pcb->pcb_flags &= ~PCB_DBREGS;
353 cpu_thread_clean(struct thread *td)
358 if (pcb->pcb_ext != NULL) {
359 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
361 * XXX do we need to move the TSS off the allocated pages
362 * before freeing them? (not done here)
364 pmap_trm_free(pcb->pcb_ext, ctob(IOPAGES + 1));
370 cpu_thread_swapin(struct thread *td)
375 cpu_thread_swapout(struct thread *td)
380 cpu_thread_alloc(struct thread *td)
383 struct xstate_hdr *xhdr;
385 td->td_pcb = pcb = get_pcb_td(td);
386 td->td_frame = (struct trapframe *)((caddr_t)pcb -
387 VM86_STACK_SPACE) - 1;
389 pcb->pcb_save = get_pcb_user_save_pcb(pcb);
391 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
392 bzero(xhdr, sizeof(*xhdr));
393 xhdr->xstate_bv = xsave_mask;
398 cpu_thread_free(struct thread *td)
401 cpu_thread_clean(td);
405 cpu_exec_vmspace_reuse(struct proc *p __unused, vm_map_t map __unused)
412 cpu_procctl(struct thread *td __unused, int idtype __unused, id_t id __unused,
413 int com __unused, void *data __unused)
420 cpu_set_syscall_retval(struct thread *td, int error)
425 td->td_frame->tf_eax = td->td_retval[0];
426 td->td_frame->tf_edx = td->td_retval[1];
427 td->td_frame->tf_eflags &= ~PSL_C;
432 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int
433 * 0x80 is 2 bytes. We saved this in tf_err.
435 td->td_frame->tf_eip -= td->td_frame->tf_err;
442 td->td_frame->tf_eax = error;
443 td->td_frame->tf_eflags |= PSL_C;
449 * Initialize machine state, mostly pcb and trap frame for a new
450 * thread, about to return to userspace. Put enough state in the new
451 * thread's PCB to get it to go back to the fork_return(), which
452 * finalizes the thread state and handles peculiarities of the first
453 * return to userspace for the new thread.
456 cpu_copy_thread(struct thread *td, struct thread *td0)
458 copy_thread(td0, td);
461 * Copy user general-purpose registers.
463 * Some of these registers are rewritten by cpu_set_upcall()
464 * and linux_set_upcall().
466 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
468 /* If the current thread has the trap bit set (i.e. a debugger had
469 * single stepped the process to the system call), we need to clear
470 * the trap flag from the new frame. Otherwise, the new thread will
471 * receive a (likely unexpected) SIGTRAP when it executes the first
472 * instruction after returning to userland.
474 td->td_frame->tf_eflags &= ~PSL_T;
478 * Set that machine state for performing an upcall that starts
479 * the entry function with the given argument.
482 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
487 * Do any extra cleaning that needs to be done.
488 * The thread may have optional components
489 * that are not present in a fresh thread.
490 * This may be a recycled thread so make it look
491 * as though it's newly allocated.
493 cpu_thread_clean(td);
496 * Set the trap frame to point at the beginning of the entry
499 td->td_frame->tf_ebp = 0;
500 td->td_frame->tf_esp =
501 (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
502 td->td_frame->tf_eip = (int)entry;
504 /* Return address sentinel value to stop stack unwinding. */
505 suword((void *)td->td_frame->tf_esp, 0);
507 /* Pass the argument to the entry point. */
508 suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
513 cpu_set_user_tls(struct thread *td, void *tls_base)
515 struct segment_descriptor sd;
519 * Construct a descriptor and store it in the pcb for
520 * the next context switch. Also store it in the gdt
521 * so that the load of tf_fs into %fs will activate it
522 * at return to userland.
524 base = (uint32_t)tls_base;
525 sd.sd_lobase = base & 0xffffff;
526 sd.sd_hibase = (base >> 24) & 0xff;
527 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
529 sd.sd_type = SDT_MEMRWA;
537 td->td_pcb->pcb_gsd = sd;
538 if (td == curthread) {
539 PCPU_GET(fsgs_gdt)[1] = sd;
540 load_gs(GSEL(GUGS_SEL, SEL_UPL));
547 * Convert kernel VA to physical address
554 pa = pmap_kextract((vm_offset_t)addr);
556 panic("kvtop: zero page frame");
561 * Get an sf_buf from the freelist. May block if none are available.
564 sf_buf_map(struct sf_buf *sf, int flags)
569 sf_buf_shootdown(sf, flags);
575 sf_buf_shootdown_curcpu_cb(pmap_t pmap __unused,
576 vm_offset_t addr1 __unused, vm_offset_t addr2 __unused)
581 sf_buf_shootdown(struct sf_buf *sf, int flags)
587 cpuid = PCPU_GET(cpuid);
588 if (!CPU_ISSET(cpuid, &sf->cpumask)) {
589 CPU_SET(cpuid, &sf->cpumask);
592 if ((flags & SFB_CPUPRIVATE) == 0) {
593 other_cpus = all_cpus;
594 CPU_CLR(cpuid, &other_cpus);
595 CPU_ANDNOT(&other_cpus, &sf->cpumask);
596 if (!CPU_EMPTY(&other_cpus)) {
597 CPU_OR(&sf->cpumask, &other_cpus);
598 smp_masked_invlpg(other_cpus, sf->kva, kernel_pmap,
599 sf_buf_shootdown_curcpu_cb);
607 * MD part of sf_buf_free().
610 sf_buf_unmap(struct sf_buf *sf)
617 sf_buf_invalidate(struct sf_buf *sf)
622 * Use pmap_qenter to update the pte for
623 * existing mapping, in particular, the PAT
624 * settings are recalculated.
626 pmap_qenter(sf->kva, &m, 1);
627 pmap_invalidate_cache_range(sf->kva, sf->kva + PAGE_SIZE);
631 * Invalidate the cache lines that may belong to the page, if
632 * (possibly old) mapping of the page by sf buffer exists. Returns
633 * TRUE when mapping was found and cache invalidated.
636 sf_buf_invalidate_cache(vm_page_t m)
639 return (sf_buf_process_page(m, sf_buf_invalidate));
643 * Software interrupt handler for queued VM system processing.
648 if (busdma_swi_pending != 0)
653 * Tell whether this address is in some physical memory region.
654 * Currently used by the kernel coredump code in order to avoid
655 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
656 * or other unpredictable behaviour.
660 is_physical_memory(vm_paddr_t addr)
664 /* The ISA ``memory hole''. */
665 if (addr >= 0xa0000 && addr < 0x100000)
670 * stuff other tests for known memory-mapped devices (PCI?)