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_CURTHREAD == offsetof(struct pcpu, pc_curthread),
94 "OFFSETOF_CURTHREAD does not correspond with offset of pc_curthread.");
95 _Static_assert(OFFSETOF_CURPCB == offsetof(struct pcpu, pc_curpcb),
96 "OFFSETOF_CURPCB does not correspond with offset of pc_curpcb.");
97 _Static_assert(__OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf),
98 "__OFFSETOF_MONINORBUF does not correspond with offset of pc_monitorbuf.");
101 get_pcb_user_save_td(struct thread *td)
105 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
106 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
107 KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area"));
108 return ((union savefpu *)p);
112 get_pcb_user_save_pcb(struct pcb *pcb)
116 p = (vm_offset_t)(pcb + 1);
117 return ((union savefpu *)p);
121 get_pcb_td(struct thread *td)
125 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
126 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) -
128 return ((struct pcb *)p);
132 alloc_fpusave(int flags)
135 struct savefpu_ymm *sf;
137 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
139 sf = (struct savefpu_ymm *)res;
140 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
141 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
146 * Finish a fork operation, with process p2 nearly set up.
147 * Copy and update the pcb, set up the stack so that the child
148 * ready to run and return to user mode.
151 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
158 if ((flags & RFPROC) == 0) {
159 if ((flags & RFMEM) == 0) {
160 /* unshare user LDT */
161 struct mdproc *mdp1 = &p1->p_md;
162 struct proc_ldt *pldt, *pldt1;
164 mtx_lock_spin(&dt_lock);
165 if ((pldt1 = mdp1->md_ldt) != NULL &&
166 pldt1->ldt_refcnt > 1) {
167 pldt = user_ldt_alloc(mdp1, pldt1->ldt_len);
169 panic("could not copy LDT");
172 user_ldt_deref(pldt1);
174 mtx_unlock_spin(&dt_lock);
179 /* Ensure that td1's pcb is up to date. */
180 if (td1 == curthread)
181 td1->td_pcb->pcb_gs = rgs();
183 if (PCPU_GET(fpcurthread) == td1)
184 npxsave(td1->td_pcb->pcb_save);
187 /* Point the pcb to the top of the stack */
188 pcb2 = get_pcb_td(td2);
192 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
194 /* Properly initialize pcb_save */
195 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
196 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
197 cpu_max_ext_state_size);
199 /* Point mdproc and then copy over td1's contents */
201 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
204 * Create a new fresh stack for the new process.
205 * Copy the trap frame for the return to user mode as if from a
206 * syscall. This copies most of the user mode register values.
207 * The -16 is so we can expand the trapframe if we go to vm86.
209 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
210 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
212 td2->td_frame->tf_eax = 0; /* Child returns zero */
213 td2->td_frame->tf_eflags &= ~PSL_C; /* success */
214 td2->td_frame->tf_edx = 1;
217 * If the parent process has the trap bit set (i.e. a debugger had
218 * single stepped the process to the system call), we need to clear
219 * the trap flag from the new frame unless the debugger had set PF_FORK
220 * on the parent. Otherwise, the child will receive a (likely
221 * unexpected) SIGTRAP when it executes the first instruction after
222 * returning to userland.
224 if ((p1->p_pfsflags & PF_FORK) == 0)
225 td2->td_frame->tf_eflags &= ~PSL_T;
228 * Set registers for trampoline to user mode. Leave space for the
229 * return address on stack. These are the kernel mode register values.
231 #if defined(PAE) || defined(PAE_TABLES)
232 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
234 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
237 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
239 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
240 pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */
241 pcb2->pcb_eip = (int)fork_trampoline;
243 * pcb2->pcb_dr*: cloned above.
244 * pcb2->pcb_savefpu: cloned above.
245 * pcb2->pcb_flags: cloned above.
246 * pcb2->pcb_onfault: cloned above (always NULL here?).
247 * pcb2->pcb_gs: cloned above.
248 * pcb2->pcb_ext: cleared below.
252 * XXX don't copy the i/o pages. this should probably be fixed.
256 /* Copy the LDT, if necessary. */
257 mtx_lock_spin(&dt_lock);
258 if (mdp2->md_ldt != NULL) {
260 mdp2->md_ldt->ldt_refcnt++;
262 mdp2->md_ldt = user_ldt_alloc(mdp2,
263 mdp2->md_ldt->ldt_len);
264 if (mdp2->md_ldt == NULL)
265 panic("could not copy LDT");
268 mtx_unlock_spin(&dt_lock);
270 /* Setup to release spin count in fork_exit(). */
271 td2->td_md.md_spinlock_count = 1;
272 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
275 * Now, cpu_switch() can schedule the new process.
276 * pcb_esp is loaded pointing to the cpu_switch() stack frame
277 * containing the return address when exiting cpu_switch.
278 * This will normally be to fork_trampoline(), which will have
279 * %ebx loaded with the new proc's pointer. fork_trampoline()
280 * will set up a stack to call fork_return(p, frame); to complete
281 * the return to user-mode.
286 * Intercept the return address from a freshly forked process that has NOT
287 * been scheduled yet.
289 * This is needed to make kernel threads stay in kernel mode.
292 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
295 * Note that the trap frame follows the args, so the function
296 * is really called like this: func(arg, frame);
298 td->td_pcb->pcb_esi = (int) func; /* function */
299 td->td_pcb->pcb_ebx = (int) arg; /* first arg */
303 cpu_exit(struct thread *td)
307 * If this process has a custom LDT, release it. Reset pc->pcb_gs
308 * and %gs before we free it in case they refer to an LDT entry.
310 mtx_lock_spin(&dt_lock);
311 if (td->td_proc->p_md.md_ldt) {
312 td->td_pcb->pcb_gs = _udatasel;
316 mtx_unlock_spin(&dt_lock);
320 cpu_thread_exit(struct thread *td)
324 if (td == PCPU_GET(fpcurthread))
328 /* Disable any hardware breakpoints. */
329 if (td->td_pcb->pcb_flags & PCB_DBREGS) {
331 td->td_pcb->pcb_flags &= ~PCB_DBREGS;
336 cpu_thread_clean(struct thread *td)
341 if (pcb->pcb_ext != NULL) {
342 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
344 * XXX do we need to move the TSS off the allocated pages
345 * before freeing them? (not done here)
347 kmem_free(kernel_arena, (vm_offset_t)pcb->pcb_ext,
354 cpu_thread_swapin(struct thread *td)
359 cpu_thread_swapout(struct thread *td)
364 cpu_thread_alloc(struct thread *td)
367 struct xstate_hdr *xhdr;
369 td->td_pcb = pcb = get_pcb_td(td);
370 td->td_frame = (struct trapframe *)((caddr_t)pcb - 16) - 1;
372 pcb->pcb_save = get_pcb_user_save_pcb(pcb);
374 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
375 bzero(xhdr, sizeof(*xhdr));
376 xhdr->xstate_bv = xsave_mask;
381 cpu_thread_free(struct thread *td)
384 cpu_thread_clean(td);
388 cpu_set_syscall_retval(struct thread *td, int error)
393 td->td_frame->tf_eax = td->td_retval[0];
394 td->td_frame->tf_edx = td->td_retval[1];
395 td->td_frame->tf_eflags &= ~PSL_C;
400 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int
401 * 0x80 is 2 bytes. We saved this in tf_err.
403 td->td_frame->tf_eip -= td->td_frame->tf_err;
410 td->td_frame->tf_eax = SV_ABI_ERRNO(td->td_proc, error);
411 td->td_frame->tf_eflags |= PSL_C;
417 * Initialize machine state, mostly pcb and trap frame for a new
418 * thread, about to return to userspace. Put enough state in the new
419 * thread's PCB to get it to go back to the fork_return(), which
420 * finalizes the thread state and handles peculiarities of the first
421 * return to userspace for the new thread.
424 cpu_copy_thread(struct thread *td, struct thread *td0)
428 /* Point the pcb to the top of the stack. */
432 * Copy the upcall pcb. This loads kernel regs.
433 * Those not loaded individually below get their default
436 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
437 pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE |
439 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
440 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
441 cpu_max_ext_state_size);
444 * Create a new fresh stack for the new thread.
446 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
448 /* If the current thread has the trap bit set (i.e. a debugger had
449 * single stepped the process to the system call), we need to clear
450 * the trap flag from the new frame. Otherwise, the new thread will
451 * receive a (likely unexpected) SIGTRAP when it executes the first
452 * instruction after returning to userland.
454 td->td_frame->tf_eflags &= ~PSL_T;
457 * Set registers for trampoline to user mode. Leave space for the
458 * return address on stack. These are the kernel mode register values.
461 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */
463 pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
464 pcb2->pcb_ebx = (int)td; /* trampoline arg */
465 pcb2->pcb_eip = (int)fork_trampoline;
466 pcb2->pcb_gs = rgs();
468 * If we didn't copy the pcb, we'd need to do the following registers:
469 * pcb2->pcb_cr3: cloned above.
470 * pcb2->pcb_dr*: cloned above.
471 * pcb2->pcb_savefpu: cloned above.
472 * pcb2->pcb_flags: cloned above.
473 * pcb2->pcb_onfault: cloned above (always NULL here?).
474 * pcb2->pcb_gs: cloned above.
475 * pcb2->pcb_ext: cleared below.
477 pcb2->pcb_ext = NULL;
479 /* Setup to release spin count in fork_exit(). */
480 td->td_md.md_spinlock_count = 1;
481 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
485 * Set that machine state for performing an upcall that starts
486 * the entry function with the given argument.
489 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
494 * Do any extra cleaning that needs to be done.
495 * The thread may have optional components
496 * that are not present in a fresh thread.
497 * This may be a recycled thread so make it look
498 * as though it's newly allocated.
500 cpu_thread_clean(td);
503 * Set the trap frame to point at the beginning of the entry
506 td->td_frame->tf_ebp = 0;
507 td->td_frame->tf_esp =
508 (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
509 td->td_frame->tf_eip = (int)entry;
511 /* Return address sentinel value to stop stack unwinding. */
512 suword((void *)td->td_frame->tf_esp, 0);
514 /* Pass the argument to the entry point. */
515 suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
520 cpu_set_user_tls(struct thread *td, void *tls_base)
522 struct segment_descriptor sd;
526 * Construct a descriptor and store it in the pcb for
527 * the next context switch. Also store it in the gdt
528 * so that the load of tf_fs into %fs will activate it
529 * at return to userland.
531 base = (uint32_t)tls_base;
532 sd.sd_lobase = base & 0xffffff;
533 sd.sd_hibase = (base >> 24) & 0xff;
534 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
536 sd.sd_type = SDT_MEMRWA;
544 td->td_pcb->pcb_gsd = sd;
545 if (td == curthread) {
546 PCPU_GET(fsgs_gdt)[1] = sd;
547 load_gs(GSEL(GUGS_SEL, SEL_UPL));
554 * Convert kernel VA to physical address
561 pa = pmap_kextract((vm_offset_t)addr);
563 panic("kvtop: zero page frame");
568 * Get an sf_buf from the freelist. May block if none are available.
571 sf_buf_map(struct sf_buf *sf, int flags)
573 pt_entry_t opte, *ptep;
576 * Update the sf_buf's virtual-to-physical mapping, flushing the
577 * virtual address from the TLB. Since the reference count for
578 * the sf_buf's old mapping was zero, that mapping is not
579 * currently in use. Consequently, there is no need to exchange
580 * the old and new PTEs atomically, even under PAE.
582 ptep = vtopte(sf->kva);
584 *ptep = VM_PAGE_TO_PHYS(sf->m) | pgeflag | PG_RW | PG_V |
585 pmap_cache_bits(sf->m->md.pat_mode, 0);
588 * Avoid unnecessary TLB invalidations: If the sf_buf's old
589 * virtual-to-physical mapping was not used, then any processor
590 * that has invalidated the sf_buf's virtual address from its TLB
591 * since the last used mapping need not invalidate again.
594 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A))
595 CPU_ZERO(&sf->cpumask);
597 sf_buf_shootdown(sf, flags);
599 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A))
600 pmap_invalidate_page(kernel_pmap, sf->kva);
606 sf_buf_shootdown(struct sf_buf *sf, int flags)
612 cpuid = PCPU_GET(cpuid);
613 if (!CPU_ISSET(cpuid, &sf->cpumask)) {
614 CPU_SET(cpuid, &sf->cpumask);
617 if ((flags & SFB_CPUPRIVATE) == 0) {
618 other_cpus = all_cpus;
619 CPU_CLR(cpuid, &other_cpus);
620 CPU_NAND(&other_cpus, &sf->cpumask);
621 if (!CPU_EMPTY(&other_cpus)) {
622 CPU_OR(&sf->cpumask, &other_cpus);
623 smp_masked_invlpg(other_cpus, sf->kva, kernel_pmap);
631 * MD part of sf_buf_free().
634 sf_buf_unmap(struct sf_buf *sf)
641 sf_buf_invalidate(struct sf_buf *sf)
646 * Use pmap_qenter to update the pte for
647 * existing mapping, in particular, the PAT
648 * settings are recalculated.
650 pmap_qenter(sf->kva, &m, 1);
651 pmap_invalidate_cache_range(sf->kva, sf->kva + PAGE_SIZE, FALSE);
655 * Invalidate the cache lines that may belong to the page, if
656 * (possibly old) mapping of the page by sf buffer exists. Returns
657 * TRUE when mapping was found and cache invalidated.
660 sf_buf_invalidate_cache(vm_page_t m)
663 return (sf_buf_process_page(m, sf_buf_invalidate));
667 * Software interrupt handler for queued VM system processing.
672 if (busdma_swi_pending != 0)
677 * Tell whether this address is in some physical memory region.
678 * Currently used by the kernel coredump code in order to avoid
679 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
680 * or other unpredictable behaviour.
684 is_physical_memory(vm_paddr_t addr)
688 /* The ISA ``memory hole''. */
689 if (addr >= 0xa0000 && addr < 0x100000)
694 * stuff other tests for known memory-mapped devices (PCI?)