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 -VM86_STACK_SPACE (-16) is so we can expand the trapframe
210 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb -
211 VM86_STACK_SPACE) - 1;
212 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
214 td2->td_frame->tf_eax = 0; /* Child returns zero */
215 td2->td_frame->tf_eflags &= ~PSL_C; /* success */
216 td2->td_frame->tf_edx = 1;
219 * If the parent process has the trap bit set (i.e. a debugger had
220 * single stepped the process to the system call), we need to clear
221 * the trap flag from the new frame unless the debugger had set PF_FORK
222 * on the parent. Otherwise, the child will receive a (likely
223 * unexpected) SIGTRAP when it executes the first instruction after
224 * returning to userland.
226 if ((p1->p_pfsflags & PF_FORK) == 0)
227 td2->td_frame->tf_eflags &= ~PSL_T;
230 * Set registers for trampoline to user mode. Leave space for the
231 * return address on stack. These are the kernel mode register values.
233 #if defined(PAE) || defined(PAE_TABLES)
234 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
236 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
239 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
241 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
242 pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */
243 pcb2->pcb_eip = (int)fork_trampoline + setidt_disp;
245 * pcb2->pcb_dr*: cloned above.
246 * pcb2->pcb_savefpu: cloned above.
247 * pcb2->pcb_flags: cloned above.
248 * pcb2->pcb_onfault: cloned above (always NULL here?).
249 * pcb2->pcb_gs: cloned above.
250 * pcb2->pcb_ext: cleared below.
254 * XXX don't copy the i/o pages. this should probably be fixed.
258 /* Copy the LDT, if necessary. */
259 mtx_lock_spin(&dt_lock);
260 if (mdp2->md_ldt != NULL) {
262 mdp2->md_ldt->ldt_refcnt++;
264 mdp2->md_ldt = user_ldt_alloc(mdp2,
265 mdp2->md_ldt->ldt_len);
266 if (mdp2->md_ldt == NULL)
267 panic("could not copy LDT");
270 mtx_unlock_spin(&dt_lock);
272 /* Setup to release spin count in fork_exit(). */
273 td2->td_md.md_spinlock_count = 1;
274 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
277 * Now, cpu_switch() can schedule the new process.
278 * pcb_esp is loaded pointing to the cpu_switch() stack frame
279 * containing the return address when exiting cpu_switch.
280 * This will normally be to fork_trampoline(), which will have
281 * %ebx loaded with the new proc's pointer. fork_trampoline()
282 * will set up a stack to call fork_return(p, frame); to complete
283 * the return to user-mode.
288 * Intercept the return address from a freshly forked process that has NOT
289 * been scheduled yet.
291 * This is needed to make kernel threads stay in kernel mode.
294 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
297 * Note that the trap frame follows the args, so the function
298 * is really called like this: func(arg, frame);
300 td->td_pcb->pcb_esi = (int) func; /* function */
301 td->td_pcb->pcb_ebx = (int) arg; /* first arg */
305 cpu_exit(struct thread *td)
309 * If this process has a custom LDT, release it. Reset pc->pcb_gs
310 * and %gs before we free it in case they refer to an LDT entry.
312 mtx_lock_spin(&dt_lock);
313 if (td->td_proc->p_md.md_ldt) {
314 td->td_pcb->pcb_gs = _udatasel;
318 mtx_unlock_spin(&dt_lock);
322 cpu_thread_exit(struct thread *td)
326 if (td == PCPU_GET(fpcurthread))
330 /* Disable any hardware breakpoints. */
331 if (td->td_pcb->pcb_flags & PCB_DBREGS) {
333 td->td_pcb->pcb_flags &= ~PCB_DBREGS;
338 cpu_thread_clean(struct thread *td)
343 if (pcb->pcb_ext != NULL) {
344 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
346 * XXX do we need to move the TSS off the allocated pages
347 * before freeing them? (not done here)
349 pmap_trm_free(pcb->pcb_ext, ctob(IOPAGES + 1));
355 cpu_thread_swapin(struct thread *td)
360 cpu_thread_swapout(struct thread *td)
365 cpu_thread_alloc(struct thread *td)
368 struct xstate_hdr *xhdr;
370 td->td_pcb = pcb = get_pcb_td(td);
371 td->td_frame = (struct trapframe *)((caddr_t)pcb -
372 VM86_STACK_SPACE) - 1;
374 pcb->pcb_save = get_pcb_user_save_pcb(pcb);
376 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
377 bzero(xhdr, sizeof(*xhdr));
378 xhdr->xstate_bv = xsave_mask;
383 cpu_thread_free(struct thread *td)
386 cpu_thread_clean(td);
390 cpu_set_syscall_retval(struct thread *td, int error)
395 td->td_frame->tf_eax = td->td_retval[0];
396 td->td_frame->tf_edx = td->td_retval[1];
397 td->td_frame->tf_eflags &= ~PSL_C;
402 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int
403 * 0x80 is 2 bytes. We saved this in tf_err.
405 td->td_frame->tf_eip -= td->td_frame->tf_err;
412 td->td_frame->tf_eax = SV_ABI_ERRNO(td->td_proc, error);
413 td->td_frame->tf_eflags |= PSL_C;
419 * Initialize machine state, mostly pcb and trap frame for a new
420 * thread, about to return to userspace. Put enough state in the new
421 * thread's PCB to get it to go back to the fork_return(), which
422 * finalizes the thread state and handles peculiarities of the first
423 * return to userspace for the new thread.
426 cpu_copy_thread(struct thread *td, struct thread *td0)
430 /* Point the pcb to the top of the stack. */
434 * Copy the upcall pcb. This loads kernel regs.
435 * Those not loaded individually below get their default
438 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
439 pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE |
441 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
442 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
443 cpu_max_ext_state_size);
446 * Create a new fresh stack for the new thread.
448 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
450 /* If the current thread has the trap bit set (i.e. a debugger had
451 * single stepped the process to the system call), we need to clear
452 * the trap flag from the new frame. Otherwise, the new thread will
453 * receive a (likely unexpected) SIGTRAP when it executes the first
454 * instruction after returning to userland.
456 td->td_frame->tf_eflags &= ~PSL_T;
459 * Set registers for trampoline to user mode. Leave space for the
460 * return address on stack. These are the kernel mode register values.
463 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */
465 pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
466 pcb2->pcb_ebx = (int)td; /* trampoline arg */
467 pcb2->pcb_eip = (int)fork_trampoline + setidt_disp;
468 pcb2->pcb_gs = rgs();
470 * If we didn't copy the pcb, we'd need to do the following registers:
471 * pcb2->pcb_cr3: cloned above.
472 * pcb2->pcb_dr*: cloned above.
473 * pcb2->pcb_savefpu: cloned above.
474 * pcb2->pcb_flags: cloned above.
475 * pcb2->pcb_onfault: cloned above (always NULL here?).
476 * pcb2->pcb_gs: cloned above.
477 * pcb2->pcb_ext: cleared below.
479 pcb2->pcb_ext = NULL;
481 /* Setup to release spin count in fork_exit(). */
482 td->td_md.md_spinlock_count = 1;
483 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
487 * Set that machine state for performing an upcall that starts
488 * the entry function with the given argument.
491 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
496 * Do any extra cleaning that needs to be done.
497 * The thread may have optional components
498 * that are not present in a fresh thread.
499 * This may be a recycled thread so make it look
500 * as though it's newly allocated.
502 cpu_thread_clean(td);
505 * Set the trap frame to point at the beginning of the entry
508 td->td_frame->tf_ebp = 0;
509 td->td_frame->tf_esp =
510 (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
511 td->td_frame->tf_eip = (int)entry;
513 /* Return address sentinel value to stop stack unwinding. */
514 suword((void *)td->td_frame->tf_esp, 0);
516 /* Pass the argument to the entry point. */
517 suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
522 cpu_set_user_tls(struct thread *td, void *tls_base)
524 struct segment_descriptor sd;
528 * Construct a descriptor and store it in the pcb for
529 * the next context switch. Also store it in the gdt
530 * so that the load of tf_fs into %fs will activate it
531 * at return to userland.
533 base = (uint32_t)tls_base;
534 sd.sd_lobase = base & 0xffffff;
535 sd.sd_hibase = (base >> 24) & 0xff;
536 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
538 sd.sd_type = SDT_MEMRWA;
546 td->td_pcb->pcb_gsd = sd;
547 if (td == curthread) {
548 PCPU_GET(fsgs_gdt)[1] = sd;
549 load_gs(GSEL(GUGS_SEL, SEL_UPL));
556 * Convert kernel VA to physical address
563 pa = pmap_kextract((vm_offset_t)addr);
565 panic("kvtop: zero page frame");
570 * Get an sf_buf from the freelist. May block if none are available.
573 sf_buf_map(struct sf_buf *sf, int flags)
575 pt_entry_t opte, *ptep;
578 * Update the sf_buf's virtual-to-physical mapping, flushing the
579 * virtual address from the TLB. Since the reference count for
580 * the sf_buf's old mapping was zero, that mapping is not
581 * currently in use. Consequently, there is no need to exchange
582 * the old and new PTEs atomically, even under PAE.
584 ptep = vtopte(sf->kva);
586 *ptep = VM_PAGE_TO_PHYS(sf->m) | PG_RW | PG_V |
587 pmap_cache_bits(kernel_pmap, sf->m->md.pat_mode, 0);
590 * Avoid unnecessary TLB invalidations: If the sf_buf's old
591 * virtual-to-physical mapping was not used, then any processor
592 * that has invalidated the sf_buf's virtual address from its TLB
593 * since the last used mapping need not invalidate again.
596 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A))
597 CPU_ZERO(&sf->cpumask);
599 sf_buf_shootdown(sf, flags);
601 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A))
602 pmap_invalidate_page(kernel_pmap, sf->kva);
608 sf_buf_shootdown(struct sf_buf *sf, int flags)
614 cpuid = PCPU_GET(cpuid);
615 if (!CPU_ISSET(cpuid, &sf->cpumask)) {
616 CPU_SET(cpuid, &sf->cpumask);
619 if ((flags & SFB_CPUPRIVATE) == 0) {
620 other_cpus = all_cpus;
621 CPU_CLR(cpuid, &other_cpus);
622 CPU_NAND(&other_cpus, &sf->cpumask);
623 if (!CPU_EMPTY(&other_cpus)) {
624 CPU_OR(&sf->cpumask, &other_cpus);
625 smp_masked_invlpg(other_cpus, sf->kva, kernel_pmap);
633 * MD part of sf_buf_free().
636 sf_buf_unmap(struct sf_buf *sf)
643 sf_buf_invalidate(struct sf_buf *sf)
648 * Use pmap_qenter to update the pte for
649 * existing mapping, in particular, the PAT
650 * settings are recalculated.
652 pmap_qenter(sf->kva, &m, 1);
653 pmap_invalidate_cache_range(sf->kva, sf->kva + PAGE_SIZE, FALSE);
657 * Invalidate the cache lines that may belong to the page, if
658 * (possibly old) mapping of the page by sf buffer exists. Returns
659 * TRUE when mapping was found and cache invalidated.
662 sf_buf_invalidate_cache(vm_page_t m)
665 return (sf_buf_process_page(m, sf_buf_invalidate));
669 * Software interrupt handler for queued VM system processing.
674 if (busdma_swi_pending != 0)
679 * Tell whether this address is in some physical memory region.
680 * Currently used by the kernel coredump code in order to avoid
681 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
682 * or other unpredictable behaviour.
686 is_physical_memory(vm_paddr_t addr)
690 /* The ISA ``memory hole''. */
691 if (addr >= 0xa0000 && addr < 0x100000)
696 * stuff other tests for known memory-mapped devices (PCI?)