2 * Copyright (c) 1982, 1986 The Regents of the University of California.
3 * Copyright (c) 1989, 1990 William Jolitz
4 * Copyright (c) 1994 John Dyson
7 * This code is derived from software contributed to Berkeley by
8 * the Systems Programming Group of the University of Utah Computer
9 * Science Department, and William Jolitz.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91
40 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
48 #include "opt_reset.h"
52 #include <sys/param.h>
53 #include <sys/systm.h>
56 #include <sys/kernel.h>
59 #include <sys/malloc.h>
61 #include <sys/mutex.h>
62 #include <sys/pioctl.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 <machine/elan_mmcr.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_kern.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_param.h>
93 #include <xen/hypervisor.h>
96 #include <pc98/cbus/cbus.h>
98 #include <i386/isa/isa.h>
102 #include <machine/xbox.h>
106 #define NSFBUFS (512 + maxusers * 16)
109 static void cpu_reset_real(void);
111 static void cpu_reset_proxy(void);
112 static u_int cpu_reset_proxyid;
113 static volatile u_int cpu_reset_proxy_active;
115 static void sf_buf_init(void *arg);
116 SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL);
118 LIST_HEAD(sf_head, sf_buf);
121 * A hash table of active sendfile(2) buffers
123 static struct sf_head *sf_buf_active;
124 static u_long sf_buf_hashmask;
126 #define SF_BUF_HASH(m) (((m) - vm_page_array) & sf_buf_hashmask)
128 static TAILQ_HEAD(, sf_buf) sf_buf_freelist;
129 static u_int sf_buf_alloc_want;
132 * A lock used to synchronize access to the hash table and free list
134 static struct mtx sf_buf_lock;
136 extern int _ucodesel, _udatasel;
139 * Finish a fork operation, with process p2 nearly set up.
140 * Copy and update the pcb, set up the stack so that the child
141 * ready to run and return to user mode.
144 cpu_fork(td1, p2, td2, flags)
145 register struct thread *td1;
146 register struct proc *p2;
150 register struct proc *p1;
155 if ((flags & RFPROC) == 0) {
156 if ((flags & RFMEM) == 0) {
157 /* unshare user LDT */
158 struct mdproc *mdp1 = &p1->p_md;
159 struct proc_ldt *pldt, *pldt1;
161 mtx_lock_spin(&dt_lock);
162 if ((pldt1 = mdp1->md_ldt) != NULL &&
163 pldt1->ldt_refcnt > 1) {
164 pldt = user_ldt_alloc(mdp1, pldt1->ldt_len);
166 panic("could not copy LDT");
169 user_ldt_deref(pldt1);
171 mtx_unlock_spin(&dt_lock);
176 /* Ensure that td1's pcb is up to date. */
177 if (td1 == curthread)
178 td1->td_pcb->pcb_gs = rgs();
181 if (PCPU_GET(fpcurthread) == td1)
182 npxsave(td1->td_pcb->pcb_save);
186 /* Point the pcb to the top of the stack */
187 pcb2 = (struct pcb *)(td2->td_kstack +
188 td2->td_kstack_pages * PAGE_SIZE) - 1;
192 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
194 /* Properly initialize pcb_save */
195 pcb2->pcb_save = &pcb2->pcb_user_save;
197 /* Point mdproc and then copy over td1's contents */
199 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
202 * Create a new fresh stack for the new process.
203 * Copy the trap frame for the return to user mode as if from a
204 * syscall. This copies most of the user mode register values.
205 * The -16 is so we can expand the trapframe if we go to vm86.
207 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
208 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
210 td2->td_frame->tf_eax = 0; /* Child returns zero */
211 td2->td_frame->tf_eflags &= ~PSL_C; /* success */
212 td2->td_frame->tf_edx = 1;
215 * If the parent process has the trap bit set (i.e. a debugger had
216 * single stepped the process to the system call), we need to clear
217 * the trap flag from the new frame unless the debugger had set PF_FORK
218 * on the parent. Otherwise, the child will receive a (likely
219 * unexpected) SIGTRAP when it executes the first instruction after
220 * returning to userland.
222 if ((p1->p_pfsflags & PF_FORK) == 0)
223 td2->td_frame->tf_eflags &= ~PSL_T;
226 * Set registers for trampoline to user mode. Leave space for the
227 * return address on stack. These are the kernel mode register values.
230 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
232 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
235 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
237 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
238 pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */
239 pcb2->pcb_eip = (int)fork_trampoline;
240 pcb2->pcb_psl = PSL_KERNEL; /* ints disabled */
242 * pcb2->pcb_dr*: cloned above.
243 * pcb2->pcb_savefpu: cloned above.
244 * pcb2->pcb_flags: cloned above.
245 * pcb2->pcb_onfault: cloned above (always NULL here?).
246 * pcb2->pcb_gs: cloned above.
247 * pcb2->pcb_ext: cleared below.
251 * XXX don't copy the i/o pages. this should probably be fixed.
255 /* Copy the LDT, if necessary. */
256 mtx_lock_spin(&dt_lock);
257 if (mdp2->md_ldt != NULL) {
259 mdp2->md_ldt->ldt_refcnt++;
261 mdp2->md_ldt = user_ldt_alloc(mdp2,
262 mdp2->md_ldt->ldt_len);
263 if (mdp2->md_ldt == NULL)
264 panic("could not copy LDT");
267 mtx_unlock_spin(&dt_lock);
269 /* Setup to release spin count in fork_exit(). */
270 td2->td_md.md_spinlock_count = 1;
272 * XXX XEN need to check on PSL_USER is handled
274 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
276 * Now, cpu_switch() can schedule the new process.
277 * pcb_esp is loaded pointing to the cpu_switch() stack frame
278 * containing the return address when exiting cpu_switch.
279 * This will normally be to fork_trampoline(), which will have
280 * %ebx loaded with the new proc's pointer. fork_trampoline()
281 * will set up a stack to call fork_return(p, frame); to complete
282 * the return to user-mode.
287 * Intercept the return address from a freshly forked process that has NOT
288 * been scheduled yet.
290 * This is needed to make kernel threads stay in kernel mode.
293 cpu_set_fork_handler(td, func, arg)
295 void (*func)(void *);
299 * Note that the trap frame follows the args, so the function
300 * is really called like this: func(arg, frame);
302 td->td_pcb->pcb_esi = (int) func; /* function */
303 td->td_pcb->pcb_ebx = (int) arg; /* first arg */
307 cpu_exit(struct thread *td)
311 * If this process has a custom LDT, release it. Reset pc->pcb_gs
312 * and %gs before we free it in case they refer to an LDT entry.
314 mtx_lock_spin(&dt_lock);
315 if (td->td_proc->p_md.md_ldt) {
316 td->td_pcb->pcb_gs = _udatasel;
320 mtx_unlock_spin(&dt_lock);
324 cpu_thread_exit(struct thread *td)
329 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_map, (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 td->td_pcb = (struct pcb *)(td->td_kstack +
374 td->td_kstack_pages * PAGE_SIZE) - 1;
375 td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb - 16) - 1;
376 td->td_pcb->pcb_ext = NULL;
377 td->td_pcb->pcb_save = &td->td_pcb->pcb_user_save;
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 if (td->td_proc->p_sysent->sv_errsize) {
411 if (error >= td->td_proc->p_sysent->sv_errsize)
412 error = -1; /* XXX */
414 error = td->td_proc->p_sysent->sv_errtbl[error];
416 td->td_frame->tf_eax = error;
417 td->td_frame->tf_eflags |= PSL_C;
423 * Initialize machine state (pcb and trap frame) for a new thread about to
424 * upcall. Put enough state in the new thread's PCB to get it to go back
425 * userret(), where we can intercept it again to set the return (upcall)
426 * Address and stack, along with those from upcals that are from other sources
427 * such as those generated in thread_userret() itself.
430 cpu_set_upcall(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);
444 pcb2->pcb_save = &pcb2->pcb_user_save;
447 * Create a new fresh stack for the new thread.
449 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
451 /* If the current thread has the trap bit set (i.e. a debugger had
452 * single stepped the process to the system call), we need to clear
453 * the trap flag from the new frame. Otherwise, the new thread will
454 * receive a (likely unexpected) SIGTRAP when it executes the first
455 * instruction after returning to userland.
457 td->td_frame->tf_eflags &= ~PSL_T;
460 * Set registers for trampoline to user mode. Leave space for the
461 * return address on stack. These are the kernel mode register values.
464 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */
466 pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
467 pcb2->pcb_ebx = (int)td; /* trampoline arg */
468 pcb2->pcb_eip = (int)fork_trampoline;
469 pcb2->pcb_psl &= ~(PSL_I); /* interrupts must be disabled */
470 pcb2->pcb_gs = rgs();
472 * If we didn't copy the pcb, we'd need to do the following registers:
473 * pcb2->pcb_cr3: cloned above.
474 * pcb2->pcb_dr*: cloned above.
475 * pcb2->pcb_savefpu: cloned above.
476 * pcb2->pcb_flags: cloned above.
477 * pcb2->pcb_onfault: cloned above (always NULL here?).
478 * pcb2->pcb_gs: cloned above.
479 * pcb2->pcb_ext: cleared below.
481 pcb2->pcb_ext = NULL;
483 /* Setup to release spin count in fork_exit(). */
484 td->td_md.md_spinlock_count = 1;
485 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
489 * Set that machine state for performing an upcall that has to
490 * be done in thread_userret() so that those upcalls generated
491 * in thread_userret() itself can be done as well.
494 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
499 * Do any extra cleaning that needs to be done.
500 * The thread may have optional components
501 * that are not present in a fresh thread.
502 * This may be a recycled thread so make it look
503 * as though it's newly allocated.
505 cpu_thread_clean(td);
508 * Set the trap frame to point at the beginning of the uts
511 td->td_frame->tf_ebp = 0;
512 td->td_frame->tf_esp =
513 (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
514 td->td_frame->tf_eip = (int)entry;
517 * Pass the address of the mailbox for this kse to the uts
518 * function as a parameter on the stack.
520 suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
525 cpu_set_user_tls(struct thread *td, void *tls_base)
527 struct segment_descriptor sd;
531 * Construct a descriptor and store it in the pcb for
532 * the next context switch. Also store it in the gdt
533 * so that the load of tf_fs into %fs will activate it
534 * at return to userland.
536 base = (uint32_t)tls_base;
537 sd.sd_lobase = base & 0xffffff;
538 sd.sd_hibase = (base >> 24) & 0xff;
539 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
541 sd.sd_type = SDT_MEMRWA;
549 td->td_pcb->pcb_gsd = sd;
550 if (td == curthread) {
551 PCPU_GET(fsgs_gdt)[1] = sd;
552 load_gs(GSEL(GUGS_SEL, SEL_UPL));
559 * Convert kernel VA to physical address
566 pa = pmap_kextract((vm_offset_t)addr);
568 panic("kvtop: zero page frame");
577 cpu_reset_proxy_active = 1;
578 while (cpu_reset_proxy_active == 1)
579 ; /* Wait for other cpu to see that we've started */
580 stop_cpus((1<<cpu_reset_proxyid));
581 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
591 if (arch_i386_is_xbox) {
592 /* Kick the PIC16L, it can reboot the box */
603 map = PCPU_GET(other_cpus) & ~stopped_cpus;
605 printf("cpu_reset: Stopping other CPUs\n");
609 if (PCPU_GET(cpuid) != 0) {
610 cpu_reset_proxyid = PCPU_GET(cpuid);
611 cpustop_restartfunc = cpu_reset_proxy;
612 cpu_reset_proxy_active = 0;
613 printf("cpu_reset: Restarting BSP\n");
615 /* Restart CPU #0. */
616 /* XXX: restart_cpus(1 << 0); */
617 atomic_store_rel_int(&started_cpus, (1 << 0));
620 while (cpu_reset_proxy_active == 0 && cnt < 10000000)
621 cnt++; /* Wait for BSP to announce restart */
622 if (cpu_reset_proxy_active == 0)
623 printf("cpu_reset: Failed to restart BSP\n");
625 cpu_reset_proxy_active = 2;
641 struct region_descriptor null_idt;
648 if (smp_processor_id() == 0)
649 HYPERVISOR_shutdown(SHUTDOWN_reboot);
651 HYPERVISOR_shutdown(SHUTDOWN_poweroff);
654 if (elan_mmcr != NULL)
655 elan_mmcr->RESCFG = 1;
658 if (cpu == CPU_GEODE1100) {
659 /* Attempt Geode's own reset */
660 outl(0xcf8, 0x80009044ul);
666 * Attempt to do a CPU reset via CPU reset port.
668 if ((inb(0x35) & 0xa0) != 0xa0) {
669 outb(0x37, 0x0f); /* SHUT0 = 0. */
670 outb(0x37, 0x0b); /* SHUT1 = 0. */
672 outb(0xf0, 0x00); /* Reset. */
674 #if !defined(BROKEN_KEYBOARD_RESET)
676 * Attempt to do a CPU reset via the keyboard controller,
677 * do not turn off GateA20, as any machine that fails
678 * to do the reset here would then end up in no man's land.
680 outb(IO_KBD + 4, 0xFE);
681 DELAY(500000); /* wait 0.5 sec to see if that did it */
685 * Attempt to force a reset via the Reset Control register at
686 * I/O port 0xcf9. Bit 2 forces a system reset when it
687 * transitions from 0 to 1. Bit 1 selects the type of reset
688 * to attempt: 0 selects a "soft" reset, and 1 selects a
689 * "hard" reset. We try a "hard" reset. The first write sets
690 * bit 1 to select a "hard" reset and clears bit 2. The
691 * second write forces a 0 -> 1 transition in bit 2 to trigger
696 DELAY(500000); /* wait 0.5 sec to see if that did it */
699 * Attempt to force a reset via the Fast A20 and Init register
700 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate.
701 * Bit 0 asserts INIT# when set to 1. We are careful to only
702 * preserve bit 1 while setting bit 0. We also must clear bit
703 * 0 before setting it if it isn't already clear.
708 outb(0x92, b & 0xfe);
710 DELAY(500000); /* wait 0.5 sec to see if that did it */
714 printf("No known reset method worked, attempting CPU shutdown\n");
715 DELAY(1000000); /* wait 1 sec for printf to complete */
718 null_idt.rd_limit = 0;
719 null_idt.rd_base = 0;
722 /* "good night, sweet prince .... <THUNK!>" */
730 * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
733 sf_buf_init(void *arg)
735 struct sf_buf *sf_bufs;
740 TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs);
742 sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask);
743 TAILQ_INIT(&sf_buf_freelist);
744 sf_base = kmem_alloc_nofault(kernel_map, nsfbufs * PAGE_SIZE);
745 sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP,
747 for (i = 0; i < nsfbufs; i++) {
748 sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
749 TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry);
751 sf_buf_alloc_want = 0;
752 mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF);
756 * Invalidate the cache lines that may belong to the page, if
757 * (possibly old) mapping of the page by sf buffer exists. Returns
758 * TRUE when mapping was found and cache invalidated.
761 sf_buf_invalidate_cache(vm_page_t m)
763 struct sf_head *hash_list;
767 hash_list = &sf_buf_active[SF_BUF_HASH(m)];
769 mtx_lock(&sf_buf_lock);
770 LIST_FOREACH(sf, hash_list, list_entry) {
773 * Use pmap_qenter to update the pte for
774 * existing mapping, in particular, the PAT
775 * settings are recalculated.
777 pmap_qenter(sf->kva, &m, 1);
778 pmap_invalidate_cache_range(sf->kva, sf->kva +
784 mtx_unlock(&sf_buf_lock);
789 * Get an sf_buf from the freelist. May block if none are available.
792 sf_buf_alloc(struct vm_page *m, int flags)
794 pt_entry_t opte, *ptep;
795 struct sf_head *hash_list;
798 cpumask_t cpumask, other_cpus;
802 KASSERT(curthread->td_pinned > 0 || (flags & SFB_CPUPRIVATE) == 0,
803 ("sf_buf_alloc(SFB_CPUPRIVATE): curthread not pinned"));
804 hash_list = &sf_buf_active[SF_BUF_HASH(m)];
805 mtx_lock(&sf_buf_lock);
806 LIST_FOREACH(sf, hash_list, list_entry) {
809 if (sf->ref_count == 1) {
810 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
812 nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
821 while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) {
822 if (flags & SFB_NOWAIT)
825 mbstat.sf_allocwait++;
826 error = msleep(&sf_buf_freelist, &sf_buf_lock,
827 (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0);
831 * If we got a signal, don't risk going back to sleep.
836 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
838 LIST_REMOVE(sf, list_entry);
839 LIST_INSERT_HEAD(hash_list, sf, list_entry);
843 nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
846 * Update the sf_buf's virtual-to-physical mapping, flushing the
847 * virtual address from the TLB. Since the reference count for
848 * the sf_buf's old mapping was zero, that mapping is not
849 * currently in use. Consequently, there is no need to exchange
850 * the old and new PTEs atomically, even under PAE.
852 ptep = vtopte(sf->kva);
855 PT_SET_MA(sf->kva, xpmap_ptom(VM_PAGE_TO_PHYS(m)) | pgeflag
856 | PG_RW | PG_V | pmap_cache_bits(m->md.pat_mode, 0));
858 *ptep = VM_PAGE_TO_PHYS(m) | pgeflag | PG_RW | PG_V |
859 pmap_cache_bits(m->md.pat_mode, 0);
863 * Avoid unnecessary TLB invalidations: If the sf_buf's old
864 * virtual-to-physical mapping was not used, then any processor
865 * that has invalidated the sf_buf's virtual address from its TLB
866 * since the last used mapping need not invalidate again.
869 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A))
873 cpumask = PCPU_GET(cpumask);
874 if ((sf->cpumask & cpumask) == 0) {
875 sf->cpumask |= cpumask;
878 if ((flags & SFB_CPUPRIVATE) == 0) {
879 other_cpus = PCPU_GET(other_cpus) & ~sf->cpumask;
880 if (other_cpus != 0) {
881 sf->cpumask |= other_cpus;
882 smp_masked_invlpg(other_cpus, sf->kva);
887 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A))
888 pmap_invalidate_page(kernel_pmap, sf->kva);
891 mtx_unlock(&sf_buf_lock);
896 * Remove a reference from the given sf_buf, adding it to the free
897 * list when its reference count reaches zero. A freed sf_buf still,
898 * however, retains its virtual-to-physical mapping until it is
899 * recycled or reactivated by sf_buf_alloc(9).
902 sf_buf_free(struct sf_buf *sf)
905 mtx_lock(&sf_buf_lock);
907 if (sf->ref_count == 0) {
908 TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry);
912 * Xen doesn't like having dangling R/W mappings
914 pmap_qremove(sf->kva, 1);
916 LIST_REMOVE(sf, list_entry);
918 if (sf_buf_alloc_want > 0)
919 wakeup_one(&sf_buf_freelist);
921 mtx_unlock(&sf_buf_lock);
925 * Software interrupt handler for queued VM system processing.
930 if (busdma_swi_pending != 0)
935 * Tell whether this address is in some physical memory region.
936 * Currently used by the kernel coredump code in order to avoid
937 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
938 * or other unpredictable behaviour.
942 is_physical_memory(vm_paddr_t addr)
946 /* The ISA ``memory hole''. */
947 if (addr >= 0xa0000 && addr < 0x100000)
952 * stuff other tests for known memory-mapped devices (PCI?)