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 <x86/isa/isa.h>
102 #include <machine/xbox.h>
106 #define NSFBUFS (512 + maxusers * 16)
109 #if !defined(CPU_DISABLE_SSE) && defined(I686_CPU)
110 #define CPU_ENABLE_SSE
113 _Static_assert(OFFSETOF_CURTHREAD == offsetof(struct pcpu, pc_curthread),
114 "OFFSETOF_CURTHREAD does not correspond with offset of pc_curthread.");
115 _Static_assert(OFFSETOF_CURPCB == offsetof(struct pcpu, pc_curpcb),
116 "OFFSETOF_CURPCB does not correspond with offset of pc_curpcb.");
118 static void cpu_reset_real(void);
120 static void cpu_reset_proxy(void);
121 static u_int cpu_reset_proxyid;
122 static volatile u_int cpu_reset_proxy_active;
126 static int nsfbufspeak;
127 static int nsfbufsused;
129 SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufs, CTLFLAG_RDTUN, &nsfbufs, 0,
130 "Maximum number of sendfile(2) sf_bufs available");
131 SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufspeak, CTLFLAG_RD, &nsfbufspeak, 0,
132 "Number of sendfile(2) sf_bufs at peak usage");
133 SYSCTL_INT(_kern_ipc, OID_AUTO, nsfbufsused, CTLFLAG_RD, &nsfbufsused, 0,
134 "Number of sendfile(2) sf_bufs in use");
136 static void sf_buf_init(void *arg);
137 SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL);
139 LIST_HEAD(sf_head, sf_buf);
142 * A hash table of active sendfile(2) buffers
144 static struct sf_head *sf_buf_active;
145 static u_long sf_buf_hashmask;
147 #define SF_BUF_HASH(m) (((m) - vm_page_array) & sf_buf_hashmask)
149 static TAILQ_HEAD(, sf_buf) sf_buf_freelist;
150 static u_int sf_buf_alloc_want;
153 * A lock used to synchronize access to the hash table and free list
155 static struct mtx sf_buf_lock;
158 get_pcb_user_save_td(struct thread *td)
161 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
162 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN);
163 KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area"));
164 return ((union savefpu *)p);
168 get_pcb_user_save_pcb(struct pcb *pcb)
172 p = (vm_offset_t)(pcb + 1);
173 return ((union savefpu *)p);
177 get_pcb_td(struct thread *td)
181 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE -
182 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) -
184 return ((struct pcb *)p);
188 alloc_fpusave(int flags)
191 #ifdef CPU_ENABLE_SSE
192 struct savefpu_ymm *sf;
195 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags);
196 #ifdef CPU_ENABLE_SSE
198 sf = (struct savefpu_ymm *)res;
199 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd));
200 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask;
207 * Finish a fork operation, with process p2 nearly set up.
208 * Copy and update the pcb, set up the stack so that the child
209 * ready to run and return to user mode.
212 cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags)
214 register struct proc *p1;
219 if ((flags & RFPROC) == 0) {
220 if ((flags & RFMEM) == 0) {
221 /* unshare user LDT */
222 struct mdproc *mdp1 = &p1->p_md;
223 struct proc_ldt *pldt, *pldt1;
225 mtx_lock_spin(&dt_lock);
226 if ((pldt1 = mdp1->md_ldt) != NULL &&
227 pldt1->ldt_refcnt > 1) {
228 pldt = user_ldt_alloc(mdp1, pldt1->ldt_len);
230 panic("could not copy LDT");
233 user_ldt_deref(pldt1);
235 mtx_unlock_spin(&dt_lock);
240 /* Ensure that td1's pcb is up to date. */
241 if (td1 == curthread)
242 td1->td_pcb->pcb_gs = rgs();
245 if (PCPU_GET(fpcurthread) == td1)
246 npxsave(td1->td_pcb->pcb_save);
250 /* Point the pcb to the top of the stack */
251 pcb2 = get_pcb_td(td2);
255 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
257 /* Properly initialize pcb_save */
258 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
259 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2),
260 cpu_max_ext_state_size);
262 /* Point mdproc and then copy over td1's contents */
264 bcopy(&p1->p_md, mdp2, sizeof(*mdp2));
267 * Create a new fresh stack for the new process.
268 * Copy the trap frame for the return to user mode as if from a
269 * syscall. This copies most of the user mode register values.
270 * The -16 is so we can expand the trapframe if we go to vm86.
272 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1;
273 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe));
275 td2->td_frame->tf_eax = 0; /* Child returns zero */
276 td2->td_frame->tf_eflags &= ~PSL_C; /* success */
277 td2->td_frame->tf_edx = 1;
280 * If the parent process has the trap bit set (i.e. a debugger had
281 * single stepped the process to the system call), we need to clear
282 * the trap flag from the new frame unless the debugger had set PF_FORK
283 * on the parent. Otherwise, the child will receive a (likely
284 * unexpected) SIGTRAP when it executes the first instruction after
285 * returning to userland.
287 if ((p1->p_pfsflags & PF_FORK) == 0)
288 td2->td_frame->tf_eflags &= ~PSL_T;
291 * Set registers for trampoline to user mode. Leave space for the
292 * return address on stack. These are the kernel mode register values.
294 #if defined(PAE) || defined(PAE_TABLES)
295 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt);
297 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir);
300 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */
302 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *);
303 pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */
304 pcb2->pcb_eip = (int)fork_trampoline;
305 pcb2->pcb_psl = PSL_KERNEL; /* ints disabled */
307 * pcb2->pcb_dr*: cloned above.
308 * pcb2->pcb_savefpu: cloned above.
309 * pcb2->pcb_flags: cloned above.
310 * pcb2->pcb_onfault: cloned above (always NULL here?).
311 * pcb2->pcb_gs: cloned above.
312 * pcb2->pcb_ext: cleared below.
316 * XXX don't copy the i/o pages. this should probably be fixed.
320 /* Copy the LDT, if necessary. */
321 mtx_lock_spin(&dt_lock);
322 if (mdp2->md_ldt != NULL) {
324 mdp2->md_ldt->ldt_refcnt++;
326 mdp2->md_ldt = user_ldt_alloc(mdp2,
327 mdp2->md_ldt->ldt_len);
328 if (mdp2->md_ldt == NULL)
329 panic("could not copy LDT");
332 mtx_unlock_spin(&dt_lock);
334 /* Setup to release spin count in fork_exit(). */
335 td2->td_md.md_spinlock_count = 1;
337 * XXX XEN need to check on PSL_USER is handled
339 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
341 * Now, cpu_switch() can schedule the new process.
342 * pcb_esp is loaded pointing to the cpu_switch() stack frame
343 * containing the return address when exiting cpu_switch.
344 * This will normally be to fork_trampoline(), which will have
345 * %ebx loaded with the new proc's pointer. fork_trampoline()
346 * will set up a stack to call fork_return(p, frame); to complete
347 * the return to user-mode.
352 * Intercept the return address from a freshly forked process that has NOT
353 * been scheduled yet.
355 * This is needed to make kernel threads stay in kernel mode.
358 cpu_set_fork_handler(td, func, arg)
360 void (*func)(void *);
364 * Note that the trap frame follows the args, so the function
365 * is really called like this: func(arg, frame);
367 td->td_pcb->pcb_esi = (int) func; /* function */
368 td->td_pcb->pcb_ebx = (int) arg; /* first arg */
372 cpu_exit(struct thread *td)
376 * If this process has a custom LDT, release it. Reset pc->pcb_gs
377 * and %gs before we free it in case they refer to an LDT entry.
379 mtx_lock_spin(&dt_lock);
380 if (td->td_proc->p_md.md_ldt) {
381 td->td_pcb->pcb_gs = _udatasel;
385 mtx_unlock_spin(&dt_lock);
389 cpu_thread_exit(struct thread *td)
394 if (td == PCPU_GET(fpcurthread))
399 /* Disable any hardware breakpoints. */
400 if (td->td_pcb->pcb_flags & PCB_DBREGS) {
402 td->td_pcb->pcb_flags &= ~PCB_DBREGS;
407 cpu_thread_clean(struct thread *td)
412 if (pcb->pcb_ext != NULL) {
413 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */
415 * XXX do we need to move the TSS off the allocated pages
416 * before freeing them? (not done here)
418 kmem_free(kernel_arena, (vm_offset_t)pcb->pcb_ext,
425 cpu_thread_swapin(struct thread *td)
430 cpu_thread_swapout(struct thread *td)
435 cpu_thread_alloc(struct thread *td)
438 #ifdef CPU_ENABLE_SSE
439 struct xstate_hdr *xhdr;
442 td->td_pcb = pcb = get_pcb_td(td);
443 td->td_frame = (struct trapframe *)((caddr_t)pcb - 16) - 1;
445 pcb->pcb_save = get_pcb_user_save_pcb(pcb);
446 #ifdef CPU_ENABLE_SSE
448 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1);
449 bzero(xhdr, sizeof(*xhdr));
450 xhdr->xstate_bv = xsave_mask;
456 cpu_thread_free(struct thread *td)
459 cpu_thread_clean(td);
463 cpu_set_syscall_retval(struct thread *td, int error)
468 td->td_frame->tf_eax = td->td_retval[0];
469 td->td_frame->tf_edx = td->td_retval[1];
470 td->td_frame->tf_eflags &= ~PSL_C;
475 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int
476 * 0x80 is 2 bytes. We saved this in tf_err.
478 td->td_frame->tf_eip -= td->td_frame->tf_err;
485 td->td_frame->tf_eax = SV_ABI_ERRNO(td->td_proc, error);
486 td->td_frame->tf_eflags |= PSL_C;
492 * Initialize machine state (pcb and trap frame) for a new thread about to
493 * upcall. Put enough state in the new thread's PCB to get it to go back
494 * userret(), where we can intercept it again to set the return (upcall)
495 * Address and stack, along with those from upcals that are from other sources
496 * such as those generated in thread_userret() itself.
499 cpu_set_upcall(struct thread *td, struct thread *td0)
503 /* Point the pcb to the top of the stack. */
507 * Copy the upcall pcb. This loads kernel regs.
508 * Those not loaded individually below get their default
511 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2));
512 pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE |
514 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2);
515 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save,
516 cpu_max_ext_state_size);
519 * Create a new fresh stack for the new thread.
521 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
523 /* If the current thread has the trap bit set (i.e. a debugger had
524 * single stepped the process to the system call), we need to clear
525 * the trap flag from the new frame. Otherwise, the new thread will
526 * receive a (likely unexpected) SIGTRAP when it executes the first
527 * instruction after returning to userland.
529 td->td_frame->tf_eflags &= ~PSL_T;
532 * Set registers for trampoline to user mode. Leave space for the
533 * return address on stack. These are the kernel mode register values.
536 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */
538 pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */
539 pcb2->pcb_ebx = (int)td; /* trampoline arg */
540 pcb2->pcb_eip = (int)fork_trampoline;
541 pcb2->pcb_psl &= ~(PSL_I); /* interrupts must be disabled */
542 pcb2->pcb_gs = rgs();
544 * If we didn't copy the pcb, we'd need to do the following registers:
545 * pcb2->pcb_cr3: cloned above.
546 * pcb2->pcb_dr*: cloned above.
547 * pcb2->pcb_savefpu: cloned above.
548 * pcb2->pcb_flags: cloned above.
549 * pcb2->pcb_onfault: cloned above (always NULL here?).
550 * pcb2->pcb_gs: cloned above.
551 * pcb2->pcb_ext: cleared below.
553 pcb2->pcb_ext = NULL;
555 /* Setup to release spin count in fork_exit(). */
556 td->td_md.md_spinlock_count = 1;
557 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I;
561 * Set that machine state for performing an upcall that has to
562 * be done in thread_userret() so that those upcalls generated
563 * in thread_userret() itself can be done as well.
566 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg,
571 * Do any extra cleaning that needs to be done.
572 * The thread may have optional components
573 * that are not present in a fresh thread.
574 * This may be a recycled thread so make it look
575 * as though it's newly allocated.
577 cpu_thread_clean(td);
580 * Set the trap frame to point at the beginning of the uts
583 td->td_frame->tf_ebp = 0;
584 td->td_frame->tf_esp =
585 (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4;
586 td->td_frame->tf_eip = (int)entry;
589 * Pass the address of the mailbox for this kse to the uts
590 * function as a parameter on the stack.
592 suword((void *)(td->td_frame->tf_esp + sizeof(void *)),
597 cpu_set_user_tls(struct thread *td, void *tls_base)
599 struct segment_descriptor sd;
603 * Construct a descriptor and store it in the pcb for
604 * the next context switch. Also store it in the gdt
605 * so that the load of tf_fs into %fs will activate it
606 * at return to userland.
608 base = (uint32_t)tls_base;
609 sd.sd_lobase = base & 0xffffff;
610 sd.sd_hibase = (base >> 24) & 0xff;
611 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */
613 sd.sd_type = SDT_MEMRWA;
621 td->td_pcb->pcb_gsd = sd;
622 if (td == curthread) {
623 PCPU_GET(fsgs_gdt)[1] = sd;
624 load_gs(GSEL(GUGS_SEL, SEL_UPL));
631 * Convert kernel VA to physical address
638 pa = pmap_kextract((vm_offset_t)addr);
640 panic("kvtop: zero page frame");
649 cpu_reset_proxy_active = 1;
650 while (cpu_reset_proxy_active == 1)
651 ia32_pause(); /* Wait for other cpu to see that we've started */
653 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid);
663 if (arch_i386_is_xbox) {
664 /* Kick the PIC16L, it can reboot the box */
676 CPU_CLR(PCPU_GET(cpuid), &map);
677 CPU_NAND(&map, &stopped_cpus);
678 if (!CPU_EMPTY(&map)) {
679 printf("cpu_reset: Stopping other CPUs\n");
683 if (PCPU_GET(cpuid) != 0) {
684 cpu_reset_proxyid = PCPU_GET(cpuid);
685 cpustop_restartfunc = cpu_reset_proxy;
686 cpu_reset_proxy_active = 0;
687 printf("cpu_reset: Restarting BSP\n");
689 /* Restart CPU #0. */
690 CPU_SETOF(0, &started_cpus);
694 while (cpu_reset_proxy_active == 0 && cnt < 10000000) {
696 cnt++; /* Wait for BSP to announce restart */
698 if (cpu_reset_proxy_active == 0) {
699 printf("cpu_reset: Failed to restart BSP\n");
701 cpu_reset_proxy_active = 2;
718 struct region_descriptor null_idt;
725 if (smp_processor_id() == 0)
726 HYPERVISOR_shutdown(SHUTDOWN_reboot);
728 HYPERVISOR_shutdown(SHUTDOWN_poweroff);
731 if (elan_mmcr != NULL)
732 elan_mmcr->RESCFG = 1;
735 if (cpu == CPU_GEODE1100) {
736 /* Attempt Geode's own reset */
737 outl(0xcf8, 0x80009044ul);
743 * Attempt to do a CPU reset via CPU reset port.
745 if ((inb(0x35) & 0xa0) != 0xa0) {
746 outb(0x37, 0x0f); /* SHUT0 = 0. */
747 outb(0x37, 0x0b); /* SHUT1 = 0. */
749 outb(0xf0, 0x00); /* Reset. */
751 #if !defined(BROKEN_KEYBOARD_RESET)
753 * Attempt to do a CPU reset via the keyboard controller,
754 * do not turn off GateA20, as any machine that fails
755 * to do the reset here would then end up in no man's land.
757 outb(IO_KBD + 4, 0xFE);
758 DELAY(500000); /* wait 0.5 sec to see if that did it */
762 * Attempt to force a reset via the Reset Control register at
763 * I/O port 0xcf9. Bit 2 forces a system reset when it
764 * transitions from 0 to 1. Bit 1 selects the type of reset
765 * to attempt: 0 selects a "soft" reset, and 1 selects a
766 * "hard" reset. We try a "hard" reset. The first write sets
767 * bit 1 to select a "hard" reset and clears bit 2. The
768 * second write forces a 0 -> 1 transition in bit 2 to trigger
773 DELAY(500000); /* wait 0.5 sec to see if that did it */
776 * Attempt to force a reset via the Fast A20 and Init register
777 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate.
778 * Bit 0 asserts INIT# when set to 1. We are careful to only
779 * preserve bit 1 while setting bit 0. We also must clear bit
780 * 0 before setting it if it isn't already clear.
785 outb(0x92, b & 0xfe);
787 DELAY(500000); /* wait 0.5 sec to see if that did it */
791 printf("No known reset method worked, attempting CPU shutdown\n");
792 DELAY(1000000); /* wait 1 sec for printf to complete */
795 null_idt.rd_limit = 0;
796 null_idt.rd_base = 0;
799 /* "good night, sweet prince .... <THUNK!>" */
807 * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
810 sf_buf_init(void *arg)
812 struct sf_buf *sf_bufs;
817 TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs);
819 sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask);
820 TAILQ_INIT(&sf_buf_freelist);
821 sf_base = kva_alloc(nsfbufs * PAGE_SIZE);
822 sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP,
824 for (i = 0; i < nsfbufs; i++) {
825 sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
826 TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry);
828 sf_buf_alloc_want = 0;
829 mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF);
833 * Invalidate the cache lines that may belong to the page, if
834 * (possibly old) mapping of the page by sf buffer exists. Returns
835 * TRUE when mapping was found and cache invalidated.
838 sf_buf_invalidate_cache(vm_page_t m)
840 struct sf_head *hash_list;
844 hash_list = &sf_buf_active[SF_BUF_HASH(m)];
846 mtx_lock(&sf_buf_lock);
847 LIST_FOREACH(sf, hash_list, list_entry) {
850 * Use pmap_qenter to update the pte for
851 * existing mapping, in particular, the PAT
852 * settings are recalculated.
854 pmap_qenter(sf->kva, &m, 1);
855 pmap_invalidate_cache_range(sf->kva, sf->kva +
861 mtx_unlock(&sf_buf_lock);
866 * Get an sf_buf from the freelist. May block if none are available.
869 sf_buf_alloc(struct vm_page *m, int flags)
871 pt_entry_t opte, *ptep;
872 struct sf_head *hash_list;
880 KASSERT(curthread->td_pinned > 0 || (flags & SFB_CPUPRIVATE) == 0,
881 ("sf_buf_alloc(SFB_CPUPRIVATE): curthread not pinned"));
882 hash_list = &sf_buf_active[SF_BUF_HASH(m)];
883 mtx_lock(&sf_buf_lock);
884 LIST_FOREACH(sf, hash_list, list_entry) {
887 if (sf->ref_count == 1) {
888 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
890 nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
899 while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) {
900 if (flags & SFB_NOWAIT)
903 SFSTAT_INC(sf_allocwait);
904 error = msleep(&sf_buf_freelist, &sf_buf_lock,
905 (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0);
909 * If we got a signal, don't risk going back to sleep.
914 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry);
916 LIST_REMOVE(sf, list_entry);
917 LIST_INSERT_HEAD(hash_list, sf, list_entry);
921 nsfbufspeak = imax(nsfbufspeak, nsfbufsused);
924 * Update the sf_buf's virtual-to-physical mapping, flushing the
925 * virtual address from the TLB. Since the reference count for
926 * the sf_buf's old mapping was zero, that mapping is not
927 * currently in use. Consequently, there is no need to exchange
928 * the old and new PTEs atomically, even under PAE.
930 ptep = vtopte(sf->kva);
933 PT_SET_MA(sf->kva, xpmap_ptom(VM_PAGE_TO_PHYS(m)) | pgeflag
934 | PG_RW | PG_V | pmap_cache_bits(m->md.pat_mode, 0));
936 *ptep = VM_PAGE_TO_PHYS(m) | pgeflag | PG_RW | PG_V |
937 pmap_cache_bits(m->md.pat_mode, 0);
941 * Avoid unnecessary TLB invalidations: If the sf_buf's old
942 * virtual-to-physical mapping was not used, then any processor
943 * that has invalidated the sf_buf's virtual address from its TLB
944 * since the last used mapping need not invalidate again.
947 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A))
948 CPU_ZERO(&sf->cpumask);
951 cpuid = PCPU_GET(cpuid);
952 if (!CPU_ISSET(cpuid, &sf->cpumask)) {
953 CPU_SET(cpuid, &sf->cpumask);
956 if ((flags & SFB_CPUPRIVATE) == 0) {
957 other_cpus = all_cpus;
958 CPU_CLR(cpuid, &other_cpus);
959 CPU_NAND(&other_cpus, &sf->cpumask);
960 if (!CPU_EMPTY(&other_cpus)) {
961 CPU_OR(&sf->cpumask, &other_cpus);
962 smp_masked_invlpg(other_cpus, sf->kva);
967 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A))
968 pmap_invalidate_page(kernel_pmap, sf->kva);
971 mtx_unlock(&sf_buf_lock);
976 * Remove a reference from the given sf_buf, adding it to the free
977 * list when its reference count reaches zero. A freed sf_buf still,
978 * however, retains its virtual-to-physical mapping until it is
979 * recycled or reactivated by sf_buf_alloc(9).
982 sf_buf_free(struct sf_buf *sf)
985 mtx_lock(&sf_buf_lock);
987 if (sf->ref_count == 0) {
988 TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry);
992 * Xen doesn't like having dangling R/W mappings
994 pmap_qremove(sf->kva, 1);
996 LIST_REMOVE(sf, list_entry);
998 if (sf_buf_alloc_want > 0)
999 wakeup(&sf_buf_freelist);
1001 mtx_unlock(&sf_buf_lock);
1005 * Software interrupt handler for queued VM system processing.
1010 if (busdma_swi_pending != 0)
1015 * Tell whether this address is in some physical memory region.
1016 * Currently used by the kernel coredump code in order to avoid
1017 * dumping the ``ISA memory hole'' which could cause indefinite hangs,
1018 * or other unpredictable behaviour.
1022 is_physical_memory(vm_paddr_t addr)
1026 /* The ISA ``memory hole''. */
1027 if (addr >= 0xa0000 && addr < 0x100000)
1032 * stuff other tests for known memory-mapped devices (PCI?)