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 <machine/elan_mmcr.h>
87 #include <vm/vm_extern.h>
88 #include <vm/vm_kern.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_param.h>
93 #include <isa/isareg.h>
96 #define NSFBUFS (512 + maxusers * 16)
99 _Static_assert(OFFSETOF_CURTHREAD == offsetof(struct pcpu, pc_curthread),
100 "OFFSETOF_CURTHREAD does not correspond with offset of pc_curthread.");
101 _Static_assert(OFFSETOF_CURPCB == offsetof(struct pcpu, pc_curpcb),
102 "OFFSETOF_CURPCB does not correspond with offset of pc_curpcb.");
103 _Static_assert(__OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf),
104 "__OFFSETOF_MONINORBUF does not correspond with offset of pc_monitorbuf.");
107 get_pcb_user_save_td(struct thread *td)
111 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
112 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
113 KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area"));
114 return ((union savefpu *)p);
118 get_pcb_user_save_pcb(struct pcb *pcb)
122 p = (vm_offset_t)(pcb + 1);
123 return ((union savefpu *)p);
127 get_pcb_td(struct thread *td)
131 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
132 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) -
134 return ((struct pcb *)p);
138 alloc_fpusave(int flags)
141 struct savefpu_ymm *sf;
143 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
145 sf = (struct savefpu_ymm *)res;
146 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
147 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
152 * Finish a fork operation, with process p2 nearly set up.
153 * Copy and update the pcb, set up the stack so that the child
154 * ready to run and return to user mode.
157 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
164 if ((flags & RFPROC) == 0) {
165 if ((flags & RFMEM) == 0) {
166 /* unshare user LDT */
167 struct mdproc *mdp1 = &p1->p_md;
168 struct proc_ldt *pldt, *pldt1;
170 mtx_lock_spin(&dt_lock);
171 if ((pldt1 = mdp1->md_ldt) != NULL &&
172 pldt1->ldt_refcnt > 1) {
173 pldt = user_ldt_alloc(mdp1, pldt1->ldt_len);
175 panic("could not copy LDT");
178 user_ldt_deref(pldt1);
180 mtx_unlock_spin(&dt_lock);
185 /* Ensure that td1's pcb is up to date. */
186 if (td1 == curthread)
187 td1->td_pcb->pcb_gs = rgs();
189 if (PCPU_GET(fpcurthread) == td1)
190 npxsave(td1->td_pcb->pcb_save);
193 /* Point the pcb to the top of the stack */
194 pcb2 = get_pcb_td(td2);
198 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
200 /* Properly initialize pcb_save */
201 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
202 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
203 cpu_max_ext_state_size);
205 /* Point mdproc and then copy over td1's contents */
207 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
210 * Create a new fresh stack for the new process.
211 * Copy the trap frame for the return to user mode as if from a
212 * syscall. This copies most of the user mode register values.
213 * The -16 is so we can expand the trapframe if we go to vm86.
215 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
216 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
218 td2->td_frame->tf_eax = 0; /* Child returns zero */
219 td2->td_frame->tf_eflags &= ~PSL_C; /* success */
220 td2->td_frame->tf_edx = 1;
223 * If the parent process has the trap bit set (i.e. a debugger had
224 * single stepped the process to the system call), we need to clear
225 * the trap flag from the new frame unless the debugger had set PF_FORK
226 * on the parent. Otherwise, the child will receive a (likely
227 * unexpected) SIGTRAP when it executes the first instruction after
228 * returning to userland.
230 if ((p1->p_pfsflags & PF_FORK) == 0)
231 td2->td_frame->tf_eflags &= ~PSL_T;
234 * Set registers for trampoline to user mode. Leave space for the
235 * return address on stack. These are the kernel mode register values.
237 #if defined(PAE) || defined(PAE_TABLES)
238 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
240 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
243 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
245 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
246 pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */
247 pcb2->pcb_eip = (int)fork_trampoline;
249 * pcb2->pcb_dr*: cloned above.
250 * pcb2->pcb_savefpu: cloned above.
251 * pcb2->pcb_flags: cloned above.
252 * pcb2->pcb_onfault: cloned above (always NULL here?).
253 * pcb2->pcb_gs: cloned above.
254 * pcb2->pcb_ext: cleared below.
258 * XXX don't copy the i/o pages. this should probably be fixed.
262 /* Copy the LDT, if necessary. */
263 mtx_lock_spin(&dt_lock);
264 if (mdp2->md_ldt != NULL) {
266 mdp2->md_ldt->ldt_refcnt++;
268 mdp2->md_ldt = user_ldt_alloc(mdp2,
269 mdp2->md_ldt->ldt_len);
270 if (mdp2->md_ldt == NULL)
271 panic("could not copy LDT");
274 mtx_unlock_spin(&dt_lock);
276 /* Setup to release spin count in fork_exit(). */
277 td2->td_md.md_spinlock_count = 1;
278 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
281 * Now, cpu_switch() can schedule the new process.
282 * pcb_esp is loaded pointing to the cpu_switch() stack frame
283 * containing the return address when exiting cpu_switch.
284 * This will normally be to fork_trampoline(), which will have
285 * %ebx loaded with the new proc's pointer. fork_trampoline()
286 * will set up a stack to call fork_return(p, frame); to complete
287 * the return to user-mode.
292 * Intercept the return address from a freshly forked process that has NOT
293 * been scheduled yet.
295 * This is needed to make kernel threads stay in kernel mode.
298 cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg)
301 * Note that the trap frame follows the args, so the function
302 * is really called like this: func(arg, frame);
304 td->td_pcb->pcb_esi = (int) func; /* function */
305 td->td_pcb->pcb_ebx = (int) arg; /* first arg */
309 cpu_exit(struct thread *td)
313 * If this process has a custom LDT, release it. Reset pc->pcb_gs
314 * and %gs before we free it in case they refer to an LDT entry.
316 mtx_lock_spin(&dt_lock);
317 if (td->td_proc->p_md.md_ldt) {
318 td->td_pcb->pcb_gs = _udatasel;
322 mtx_unlock_spin(&dt_lock);
326 cpu_thread_exit(struct thread *td)
330 if (td == PCPU_GET(fpcurthread))
334 /* Disable any hardware breakpoints. */
335 if (td->td_pcb->pcb_flags & PCB_DBREGS) {
337 td->td_pcb->pcb_flags &= ~PCB_DBREGS;
342 cpu_thread_clean(struct thread *td)
347 if (pcb->pcb_ext != NULL) {
348 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
350 * XXX do we need to move the TSS off the allocated pages
351 * before freeing them? (not done here)
353 kmem_free(kernel_arena, (vm_offset_t)pcb->pcb_ext,
360 cpu_thread_swapin(struct thread *td)
365 cpu_thread_swapout(struct thread *td)
370 cpu_thread_alloc(struct thread *td)
373 struct xstate_hdr *xhdr;
375 td->td_pcb = pcb = get_pcb_td(td);
376 td->td_frame = (struct trapframe *)((caddr_t)pcb - 16) - 1;
378 pcb->pcb_save = get_pcb_user_save_pcb(pcb);
380 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
381 bzero(xhdr, sizeof(*xhdr));
382 xhdr->xstate_bv = xsave_mask;
387 cpu_thread_free(struct thread *td)
390 cpu_thread_clean(td);
394 cpu_set_syscall_retval(struct thread *td, int error)
399 td->td_frame->tf_eax = td->td_retval[0];
400 td->td_frame->tf_edx = td->td_retval[1];
401 td->td_frame->tf_eflags &= ~PSL_C;
406 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int
407 * 0x80 is 2 bytes. We saved this in tf_err.
409 td->td_frame->tf_eip -= td->td_frame->tf_err;
416 td->td_frame->tf_eax = SV_ABI_ERRNO(td->td_proc, error);
417 td->td_frame->tf_eflags |= PSL_C;
423 * Initialize machine state, mostly pcb and trap frame for a new
424 * thread, about to return to userspace. Put enough state in the new
425 * thread's PCB to get it to go back to the fork_return(), which
426 * finalizes the thread state and handles peculiarities of the first
427 * return to userspace for the new thread.
430 cpu_copy_thread(struct thread *td, struct thread *td0)
434 /* Point the pcb to the top of the stack. */
438 * Copy the upcall pcb. This loads kernel regs.
439 * Those not loaded individually below get their default
442 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
443 pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE |
445 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
446 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
447 cpu_max_ext_state_size);
450 * Create a new fresh stack for the new thread.
452 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
454 /* If the current thread has the trap bit set (i.e. a debugger had
455 * single stepped the process to the system call), we need to clear
456 * the trap flag from the new frame. Otherwise, the new thread will
457 * receive a (likely unexpected) SIGTRAP when it executes the first
458 * instruction after returning to userland.
460 td->td_frame->tf_eflags &= ~PSL_T;
463 * Set registers for trampoline to user mode. Leave space for the
464 * return address on stack. These are the kernel mode register values.
467 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */
469 pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
470 pcb2->pcb_ebx = (int)td; /* trampoline arg */
471 pcb2->pcb_eip = (int)fork_trampoline;
472 pcb2->pcb_gs = rgs();
474 * If we didn't copy the pcb, we'd need to do the following registers:
475 * pcb2->pcb_cr3: cloned above.
476 * pcb2->pcb_dr*: cloned above.
477 * pcb2->pcb_savefpu: cloned above.
478 * pcb2->pcb_flags: cloned above.
479 * pcb2->pcb_onfault: cloned above (always NULL here?).
480 * pcb2->pcb_gs: cloned above.
481 * pcb2->pcb_ext: cleared below.
483 pcb2->pcb_ext = NULL;
485 /* Setup to release spin count in fork_exit(). */
486 td->td_md.md_spinlock_count = 1;
487 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
491 * Set that machine state for performing an upcall that starts
492 * the entry function with the given argument.
495 cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg,
500 * Do any extra cleaning that needs to be done.
501 * The thread may have optional components
502 * that are not present in a fresh thread.
503 * This may be a recycled thread so make it look
504 * as though it's newly allocated.
506 cpu_thread_clean(td);
509 * Set the trap frame to point at the beginning of the entry
512 td->td_frame->tf_ebp = 0;
513 td->td_frame->tf_esp =
514 (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
515 td->td_frame->tf_eip = (int)entry;
517 /* Return address sentinel value to stop stack unwinding. */
518 suword((void *)td->td_frame->tf_esp, 0);
520 /* Pass the argument to the entry point. */
521 suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
526 cpu_set_user_tls(struct thread *td, void *tls_base)
528 struct segment_descriptor sd;
532 * Construct a descriptor and store it in the pcb for
533 * the next context switch. Also store it in the gdt
534 * so that the load of tf_fs into %fs will activate it
535 * at return to userland.
537 base = (uint32_t)tls_base;
538 sd.sd_lobase = base & 0xffffff;
539 sd.sd_hibase = (base >> 24) & 0xff;
540 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
542 sd.sd_type = SDT_MEMRWA;
550 td->td_pcb->pcb_gsd = sd;
551 if (td == curthread) {
552 PCPU_GET(fsgs_gdt)[1] = sd;
553 load_gs(GSEL(GUGS_SEL, SEL_UPL));
560 * Convert kernel VA to physical address
567 pa = pmap_kextract((vm_offset_t)addr);
569 panic("kvtop: zero page frame");
574 * Get an sf_buf from the freelist. May block if none are available.
577 sf_buf_map(struct sf_buf *sf, int flags)
579 pt_entry_t opte, *ptep;
582 * Update the sf_buf's virtual-to-physical mapping, flushing the
583 * virtual address from the TLB. Since the reference count for
584 * the sf_buf's old mapping was zero, that mapping is not
585 * currently in use. Consequently, there is no need to exchange
586 * the old and new PTEs atomically, even under PAE.
588 ptep = vtopte(sf->kva);
590 *ptep = VM_PAGE_TO_PHYS(sf->m) | pgeflag | PG_RW | PG_V |
591 pmap_cache_bits(sf->m->md.pat_mode, 0);
594 * Avoid unnecessary TLB invalidations: If the sf_buf's old
595 * virtual-to-physical mapping was not used, then any processor
596 * that has invalidated the sf_buf's virtual address from its TLB
597 * since the last used mapping need not invalidate again.
600 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A))
601 CPU_ZERO(&sf->cpumask);
603 sf_buf_shootdown(sf, flags);
605 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A))
606 pmap_invalidate_page(kernel_pmap, sf->kva);
612 sf_buf_shootdown(struct sf_buf *sf, int flags)
618 cpuid = PCPU_GET(cpuid);
619 if (!CPU_ISSET(cpuid, &sf->cpumask)) {
620 CPU_SET(cpuid, &sf->cpumask);
623 if ((flags & SFB_CPUPRIVATE) == 0) {
624 other_cpus = all_cpus;
625 CPU_CLR(cpuid, &other_cpus);
626 CPU_NAND(&other_cpus, &sf->cpumask);
627 if (!CPU_EMPTY(&other_cpus)) {
628 CPU_OR(&sf->cpumask, &other_cpus);
629 smp_masked_invlpg(other_cpus, sf->kva, kernel_pmap);
637 * MD part of sf_buf_free().
640 sf_buf_unmap(struct sf_buf *sf)
647 sf_buf_invalidate(struct sf_buf *sf)
652 * Use pmap_qenter to update the pte for
653 * existing mapping, in particular, the PAT
654 * settings are recalculated.
656 pmap_qenter(sf->kva, &m, 1);
657 pmap_invalidate_cache_range(sf->kva, sf->kva + PAGE_SIZE, FALSE);
661 * Invalidate the cache lines that may belong to the page, if
662 * (possibly old) mapping of the page by sf buffer exists. Returns
663 * TRUE when mapping was found and cache invalidated.
666 sf_buf_invalidate_cache(vm_page_t m)
669 return (sf_buf_process_page(m, sf_buf_invalidate));
673 * Software interrupt handler for queued VM system processing.
678 if (busdma_swi_pending != 0)
683 * Tell whether this address is in some physical memory region.
684 * Currently used by the kernel coredump code in order to avoid
685 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
686 * or other unpredictable behaviour.
690 is_physical_memory(vm_paddr_t addr)
694 /* The ISA ``memory hole''. */
695 if (addr >= 0xa0000 && addr < 0x100000)
700 * stuff other tests for known memory-mapped devices (PCI?)