vm_machdep.c revision 314210
1/*- 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 5 * All rights reserved. 6 * 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. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 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. 26 * 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 37 * SUCH DAMAGE. 38 * 39 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91 40 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ 41 */ 42 43#include <sys/cdefs.h> 44__FBSDID("$FreeBSD: stable/11/sys/i386/i386/vm_machdep.c 314210 2017-02-24 16:02:01Z kib $"); 45 46#include "opt_isa.h" 47#include "opt_npx.h" 48#include "opt_reset.h" 49#include "opt_cpu.h" 50#include "opt_xbox.h" 51 52#include <sys/param.h> 53#include <sys/systm.h> 54#include <sys/bio.h> 55#include <sys/buf.h> 56#include <sys/kernel.h> 57#include <sys/ktr.h> 58#include <sys/lock.h> 59#include <sys/malloc.h> 60#include <sys/mbuf.h> 61#include <sys/mutex.h> 62#include <sys/pioctl.h> 63#include <sys/proc.h> 64#include <sys/sysent.h> 65#include <sys/sf_buf.h> 66#include <sys/smp.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> 72 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> 80 81#ifdef CPU_ELAN 82#include <machine/elan_mmcr.h> 83#endif 84 85#include <vm/vm.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> 91 92#ifdef PC98 93#include <pc98/cbus/cbus.h> 94#else 95#include <isa/isareg.h> 96#endif 97 98#ifdef XBOX 99#include <machine/xbox.h> 100#endif 101 102#ifndef NSFBUFS 103#define NSFBUFS (512 + maxusers * 16) 104#endif 105 106_Static_assert(OFFSETOF_CURTHREAD == offsetof(struct pcpu, pc_curthread), 107 "OFFSETOF_CURTHREAD does not correspond with offset of pc_curthread."); 108_Static_assert(OFFSETOF_CURPCB == offsetof(struct pcpu, pc_curpcb), 109 "OFFSETOF_CURPCB does not correspond with offset of pc_curpcb."); 110_Static_assert(__OFFSETOF_MONITORBUF == offsetof(struct pcpu, pc_monitorbuf), 111 "__OFFSETOF_MONINORBUF does not correspond with offset of pc_monitorbuf."); 112 113static void cpu_reset_real(void); 114#ifdef SMP 115static void cpu_reset_proxy(void); 116static u_int cpu_reset_proxyid; 117static volatile u_int cpu_reset_proxy_active; 118#endif 119 120union savefpu * 121get_pcb_user_save_td(struct thread *td) 122{ 123 vm_offset_t p; 124 125 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE - 126 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN); 127 KASSERT((p % XSAVE_AREA_ALIGN) == 0, ("Unaligned pcb_user_save area")); 128 return ((union savefpu *)p); 129} 130 131union savefpu * 132get_pcb_user_save_pcb(struct pcb *pcb) 133{ 134 vm_offset_t p; 135 136 p = (vm_offset_t)(pcb + 1); 137 return ((union savefpu *)p); 138} 139 140struct pcb * 141get_pcb_td(struct thread *td) 142{ 143 vm_offset_t p; 144 145 p = td->td_kstack + td->td_kstack_pages * PAGE_SIZE - 146 roundup2(cpu_max_ext_state_size, XSAVE_AREA_ALIGN) - 147 sizeof(struct pcb); 148 return ((struct pcb *)p); 149} 150 151void * 152alloc_fpusave(int flags) 153{ 154 void *res; 155 struct savefpu_ymm *sf; 156 157 res = malloc(cpu_max_ext_state_size, M_DEVBUF, flags); 158 if (use_xsave) { 159 sf = (struct savefpu_ymm *)res; 160 bzero(&sf->sv_xstate.sx_hd, sizeof(sf->sv_xstate.sx_hd)); 161 sf->sv_xstate.sx_hd.xstate_bv = xsave_mask; 162 } 163 return (res); 164} 165/* 166 * Finish a fork operation, with process p2 nearly set up. 167 * Copy and update the pcb, set up the stack so that the child 168 * ready to run and return to user mode. 169 */ 170void 171cpu_fork(struct thread *td1, struct proc *p2, struct thread *td2, int flags) 172{ 173 register struct proc *p1; 174 struct pcb *pcb2; 175 struct mdproc *mdp2; 176 177 p1 = td1->td_proc; 178 if ((flags & RFPROC) == 0) { 179 if ((flags & RFMEM) == 0) { 180 /* unshare user LDT */ 181 struct mdproc *mdp1 = &p1->p_md; 182 struct proc_ldt *pldt, *pldt1; 183 184 mtx_lock_spin(&dt_lock); 185 if ((pldt1 = mdp1->md_ldt) != NULL && 186 pldt1->ldt_refcnt > 1) { 187 pldt = user_ldt_alloc(mdp1, pldt1->ldt_len); 188 if (pldt == NULL) 189 panic("could not copy LDT"); 190 mdp1->md_ldt = pldt; 191 set_user_ldt(mdp1); 192 user_ldt_deref(pldt1); 193 } else 194 mtx_unlock_spin(&dt_lock); 195 } 196 return; 197 } 198 199 /* Ensure that td1's pcb is up to date. */ 200 if (td1 == curthread) 201 td1->td_pcb->pcb_gs = rgs(); 202 critical_enter(); 203 if (PCPU_GET(fpcurthread) == td1) 204 npxsave(td1->td_pcb->pcb_save); 205 critical_exit(); 206 207 /* Point the pcb to the top of the stack */ 208 pcb2 = get_pcb_td(td2); 209 td2->td_pcb = pcb2; 210 211 /* Copy td1's pcb */ 212 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 213 214 /* Properly initialize pcb_save */ 215 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2); 216 bcopy(get_pcb_user_save_td(td1), get_pcb_user_save_pcb(pcb2), 217 cpu_max_ext_state_size); 218 219 /* Point mdproc and then copy over td1's contents */ 220 mdp2 = &p2->p_md; 221 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 222 223 /* 224 * Create a new fresh stack for the new process. 225 * Copy the trap frame for the return to user mode as if from a 226 * syscall. This copies most of the user mode register values. 227 * The -16 is so we can expand the trapframe if we go to vm86. 228 */ 229 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1; 230 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 231 232 td2->td_frame->tf_eax = 0; /* Child returns zero */ 233 td2->td_frame->tf_eflags &= ~PSL_C; /* success */ 234 td2->td_frame->tf_edx = 1; 235 236 /* 237 * If the parent process has the trap bit set (i.e. a debugger had 238 * single stepped the process to the system call), we need to clear 239 * the trap flag from the new frame unless the debugger had set PF_FORK 240 * on the parent. Otherwise, the child will receive a (likely 241 * unexpected) SIGTRAP when it executes the first instruction after 242 * returning to userland. 243 */ 244 if ((p1->p_pfsflags & PF_FORK) == 0) 245 td2->td_frame->tf_eflags &= ~PSL_T; 246 247 /* 248 * Set registers for trampoline to user mode. Leave space for the 249 * return address on stack. These are the kernel mode register values. 250 */ 251#if defined(PAE) || defined(PAE_TABLES) 252 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt); 253#else 254 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir); 255#endif 256 pcb2->pcb_edi = 0; 257 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */ 258 pcb2->pcb_ebp = 0; 259 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *); 260 pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */ 261 pcb2->pcb_eip = (int)fork_trampoline; 262 pcb2->pcb_psl = PSL_KERNEL; /* ints disabled */ 263 /*- 264 * pcb2->pcb_dr*: cloned above. 265 * pcb2->pcb_savefpu: cloned above. 266 * pcb2->pcb_flags: cloned above. 267 * pcb2->pcb_onfault: cloned above (always NULL here?). 268 * pcb2->pcb_gs: cloned above. 269 * pcb2->pcb_ext: cleared below. 270 */ 271 272 /* 273 * XXX don't copy the i/o pages. this should probably be fixed. 274 */ 275 pcb2->pcb_ext = 0; 276 277 /* Copy the LDT, if necessary. */ 278 mtx_lock_spin(&dt_lock); 279 if (mdp2->md_ldt != NULL) { 280 if (flags & RFMEM) { 281 mdp2->md_ldt->ldt_refcnt++; 282 } else { 283 mdp2->md_ldt = user_ldt_alloc(mdp2, 284 mdp2->md_ldt->ldt_len); 285 if (mdp2->md_ldt == NULL) 286 panic("could not copy LDT"); 287 } 288 } 289 mtx_unlock_spin(&dt_lock); 290 291 /* Setup to release spin count in fork_exit(). */ 292 td2->td_md.md_spinlock_count = 1; 293 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 294 295 /* 296 * Now, cpu_switch() can schedule the new process. 297 * pcb_esp is loaded pointing to the cpu_switch() stack frame 298 * containing the return address when exiting cpu_switch. 299 * This will normally be to fork_trampoline(), which will have 300 * %ebx loaded with the new proc's pointer. fork_trampoline() 301 * will set up a stack to call fork_return(p, frame); to complete 302 * the return to user-mode. 303 */ 304} 305 306/* 307 * Intercept the return address from a freshly forked process that has NOT 308 * been scheduled yet. 309 * 310 * This is needed to make kernel threads stay in kernel mode. 311 */ 312void 313cpu_fork_kthread_handler(struct thread *td, void (*func)(void *), void *arg) 314{ 315 /* 316 * Note that the trap frame follows the args, so the function 317 * is really called like this: func(arg, frame); 318 */ 319 td->td_pcb->pcb_esi = (int) func; /* function */ 320 td->td_pcb->pcb_ebx = (int) arg; /* first arg */ 321} 322 323void 324cpu_exit(struct thread *td) 325{ 326 327 /* 328 * If this process has a custom LDT, release it. Reset pc->pcb_gs 329 * and %gs before we free it in case they refer to an LDT entry. 330 */ 331 mtx_lock_spin(&dt_lock); 332 if (td->td_proc->p_md.md_ldt) { 333 td->td_pcb->pcb_gs = _udatasel; 334 load_gs(_udatasel); 335 user_ldt_free(td); 336 } else 337 mtx_unlock_spin(&dt_lock); 338} 339 340void 341cpu_thread_exit(struct thread *td) 342{ 343 344 critical_enter(); 345 if (td == PCPU_GET(fpcurthread)) 346 npxdrop(); 347 critical_exit(); 348 349 /* Disable any hardware breakpoints. */ 350 if (td->td_pcb->pcb_flags & PCB_DBREGS) { 351 reset_dbregs(); 352 td->td_pcb->pcb_flags &= ~PCB_DBREGS; 353 } 354} 355 356void 357cpu_thread_clean(struct thread *td) 358{ 359 struct pcb *pcb; 360 361 pcb = td->td_pcb; 362 if (pcb->pcb_ext != NULL) { 363 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */ 364 /* 365 * XXX do we need to move the TSS off the allocated pages 366 * before freeing them? (not done here) 367 */ 368 kmem_free(kernel_arena, (vm_offset_t)pcb->pcb_ext, 369 ctob(IOPAGES + 1)); 370 pcb->pcb_ext = NULL; 371 } 372} 373 374void 375cpu_thread_swapin(struct thread *td) 376{ 377} 378 379void 380cpu_thread_swapout(struct thread *td) 381{ 382} 383 384void 385cpu_thread_alloc(struct thread *td) 386{ 387 struct pcb *pcb; 388 struct xstate_hdr *xhdr; 389 390 td->td_pcb = pcb = get_pcb_td(td); 391 td->td_frame = (struct trapframe *)((caddr_t)pcb - 16) - 1; 392 pcb->pcb_ext = NULL; 393 pcb->pcb_save = get_pcb_user_save_pcb(pcb); 394 if (use_xsave) { 395 xhdr = (struct xstate_hdr *)(pcb->pcb_save + 1); 396 bzero(xhdr, sizeof(*xhdr)); 397 xhdr->xstate_bv = xsave_mask; 398 } 399} 400 401void 402cpu_thread_free(struct thread *td) 403{ 404 405 cpu_thread_clean(td); 406} 407 408void 409cpu_set_syscall_retval(struct thread *td, int error) 410{ 411 412 switch (error) { 413 case 0: 414 td->td_frame->tf_eax = td->td_retval[0]; 415 td->td_frame->tf_edx = td->td_retval[1]; 416 td->td_frame->tf_eflags &= ~PSL_C; 417 break; 418 419 case ERESTART: 420 /* 421 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int 422 * 0x80 is 2 bytes. We saved this in tf_err. 423 */ 424 td->td_frame->tf_eip -= td->td_frame->tf_err; 425 break; 426 427 case EJUSTRETURN: 428 break; 429 430 default: 431 td->td_frame->tf_eax = SV_ABI_ERRNO(td->td_proc, error); 432 td->td_frame->tf_eflags |= PSL_C; 433 break; 434 } 435} 436 437/* 438 * Initialize machine state, mostly pcb and trap frame for a new 439 * thread, about to return to userspace. Put enough state in the new 440 * thread's PCB to get it to go back to the fork_return(), which 441 * finalizes the thread state and handles peculiarities of the first 442 * return to userspace for the new thread. 443 */ 444void 445cpu_copy_thread(struct thread *td, struct thread *td0) 446{ 447 struct pcb *pcb2; 448 449 /* Point the pcb to the top of the stack. */ 450 pcb2 = td->td_pcb; 451 452 /* 453 * Copy the upcall pcb. This loads kernel regs. 454 * Those not loaded individually below get their default 455 * values here. 456 */ 457 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 458 pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE | 459 PCB_KERNNPX); 460 pcb2->pcb_save = get_pcb_user_save_pcb(pcb2); 461 bcopy(get_pcb_user_save_td(td0), pcb2->pcb_save, 462 cpu_max_ext_state_size); 463 464 /* 465 * Create a new fresh stack for the new thread. 466 */ 467 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 468 469 /* If the current thread has the trap bit set (i.e. a debugger had 470 * single stepped the process to the system call), we need to clear 471 * the trap flag from the new frame. Otherwise, the new thread will 472 * receive a (likely unexpected) SIGTRAP when it executes the first 473 * instruction after returning to userland. 474 */ 475 td->td_frame->tf_eflags &= ~PSL_T; 476 477 /* 478 * Set registers for trampoline to user mode. Leave space for the 479 * return address on stack. These are the kernel mode register values. 480 */ 481 pcb2->pcb_edi = 0; 482 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */ 483 pcb2->pcb_ebp = 0; 484 pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */ 485 pcb2->pcb_ebx = (int)td; /* trampoline arg */ 486 pcb2->pcb_eip = (int)fork_trampoline; 487 pcb2->pcb_psl &= ~(PSL_I); /* interrupts must be disabled */ 488 pcb2->pcb_gs = rgs(); 489 /* 490 * If we didn't copy the pcb, we'd need to do the following registers: 491 * pcb2->pcb_cr3: cloned above. 492 * pcb2->pcb_dr*: cloned above. 493 * pcb2->pcb_savefpu: cloned above. 494 * pcb2->pcb_flags: cloned above. 495 * pcb2->pcb_onfault: cloned above (always NULL here?). 496 * pcb2->pcb_gs: cloned above. 497 * pcb2->pcb_ext: cleared below. 498 */ 499 pcb2->pcb_ext = NULL; 500 501 /* Setup to release spin count in fork_exit(). */ 502 td->td_md.md_spinlock_count = 1; 503 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 504} 505 506/* 507 * Set that machine state for performing an upcall that starts 508 * the entry function with the given argument. 509 */ 510void 511cpu_set_upcall(struct thread *td, void (*entry)(void *), void *arg, 512 stack_t *stack) 513{ 514 515 /* 516 * Do any extra cleaning that needs to be done. 517 * The thread may have optional components 518 * that are not present in a fresh thread. 519 * This may be a recycled thread so make it look 520 * as though it's newly allocated. 521 */ 522 cpu_thread_clean(td); 523 524 /* 525 * Set the trap frame to point at the beginning of the entry 526 * function. 527 */ 528 td->td_frame->tf_ebp = 0; 529 td->td_frame->tf_esp = 530 (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4; 531 td->td_frame->tf_eip = (int)entry; 532 533 /* Pass the argument to the entry point. */ 534 suword((void *)(td->td_frame->tf_esp + sizeof(void *)), 535 (int)arg); 536} 537 538int 539cpu_set_user_tls(struct thread *td, void *tls_base) 540{ 541 struct segment_descriptor sd; 542 uint32_t base; 543 544 /* 545 * Construct a descriptor and store it in the pcb for 546 * the next context switch. Also store it in the gdt 547 * so that the load of tf_fs into %fs will activate it 548 * at return to userland. 549 */ 550 base = (uint32_t)tls_base; 551 sd.sd_lobase = base & 0xffffff; 552 sd.sd_hibase = (base >> 24) & 0xff; 553 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */ 554 sd.sd_hilimit = 0xf; 555 sd.sd_type = SDT_MEMRWA; 556 sd.sd_dpl = SEL_UPL; 557 sd.sd_p = 1; 558 sd.sd_xx = 0; 559 sd.sd_def32 = 1; 560 sd.sd_gran = 1; 561 critical_enter(); 562 /* set %gs */ 563 td->td_pcb->pcb_gsd = sd; 564 if (td == curthread) { 565 PCPU_GET(fsgs_gdt)[1] = sd; 566 load_gs(GSEL(GUGS_SEL, SEL_UPL)); 567 } 568 critical_exit(); 569 return (0); 570} 571 572/* 573 * Convert kernel VA to physical address 574 */ 575vm_paddr_t 576kvtop(void *addr) 577{ 578 vm_paddr_t pa; 579 580 pa = pmap_kextract((vm_offset_t)addr); 581 if (pa == 0) 582 panic("kvtop: zero page frame"); 583 return (pa); 584} 585 586#ifdef SMP 587static void 588cpu_reset_proxy() 589{ 590 cpuset_t tcrp; 591 592 cpu_reset_proxy_active = 1; 593 while (cpu_reset_proxy_active == 1) 594 ; /* Wait for other cpu to see that we've started */ 595 CPU_SETOF(cpu_reset_proxyid, &tcrp); 596 stop_cpus(tcrp); 597 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 598 DELAY(1000000); 599 cpu_reset_real(); 600} 601#endif 602 603void 604cpu_reset() 605{ 606#ifdef XBOX 607 if (arch_i386_is_xbox) { 608 /* Kick the PIC16L, it can reboot the box */ 609 pic16l_reboot(); 610 for (;;); 611 } 612#endif 613 614#ifdef SMP 615 cpuset_t map; 616 u_int cnt; 617 618 if (smp_started) { 619 map = all_cpus; 620 CPU_CLR(PCPU_GET(cpuid), &map); 621 CPU_NAND(&map, &stopped_cpus); 622 if (!CPU_EMPTY(&map)) { 623 printf("cpu_reset: Stopping other CPUs\n"); 624 stop_cpus(map); 625 } 626 627 if (PCPU_GET(cpuid) != 0) { 628 cpu_reset_proxyid = PCPU_GET(cpuid); 629 cpustop_restartfunc = cpu_reset_proxy; 630 cpu_reset_proxy_active = 0; 631 printf("cpu_reset: Restarting BSP\n"); 632 633 /* Restart CPU #0. */ 634 /* XXX: restart_cpus(1 << 0); */ 635 CPU_SETOF(0, &started_cpus); 636 wmb(); 637 638 cnt = 0; 639 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 640 cnt++; /* Wait for BSP to announce restart */ 641 if (cpu_reset_proxy_active == 0) 642 printf("cpu_reset: Failed to restart BSP\n"); 643 enable_intr(); 644 cpu_reset_proxy_active = 2; 645 646 while (1); 647 /* NOTREACHED */ 648 } 649 650 DELAY(1000000); 651 } 652#endif 653 cpu_reset_real(); 654 /* NOTREACHED */ 655} 656 657static void 658cpu_reset_real() 659{ 660 struct region_descriptor null_idt; 661#ifndef PC98 662 int b; 663#endif 664 665 disable_intr(); 666#ifdef CPU_ELAN 667 if (elan_mmcr != NULL) 668 elan_mmcr->RESCFG = 1; 669#endif 670 671 if (cpu == CPU_GEODE1100) { 672 /* Attempt Geode's own reset */ 673 outl(0xcf8, 0x80009044ul); 674 outl(0xcfc, 0xf); 675 } 676 677#ifdef PC98 678 /* 679 * Attempt to do a CPU reset via CPU reset port. 680 */ 681 if ((inb(0x35) & 0xa0) != 0xa0) { 682 outb(0x37, 0x0f); /* SHUT0 = 0. */ 683 outb(0x37, 0x0b); /* SHUT1 = 0. */ 684 } 685 outb(0xf0, 0x00); /* Reset. */ 686#else 687#if !defined(BROKEN_KEYBOARD_RESET) 688 /* 689 * Attempt to do a CPU reset via the keyboard controller, 690 * do not turn off GateA20, as any machine that fails 691 * to do the reset here would then end up in no man's land. 692 */ 693 outb(IO_KBD + 4, 0xFE); 694 DELAY(500000); /* wait 0.5 sec to see if that did it */ 695#endif 696 697 /* 698 * Attempt to force a reset via the Reset Control register at 699 * I/O port 0xcf9. Bit 2 forces a system reset when it 700 * transitions from 0 to 1. Bit 1 selects the type of reset 701 * to attempt: 0 selects a "soft" reset, and 1 selects a 702 * "hard" reset. We try a "hard" reset. The first write sets 703 * bit 1 to select a "hard" reset and clears bit 2. The 704 * second write forces a 0 -> 1 transition in bit 2 to trigger 705 * a reset. 706 */ 707 outb(0xcf9, 0x2); 708 outb(0xcf9, 0x6); 709 DELAY(500000); /* wait 0.5 sec to see if that did it */ 710 711 /* 712 * Attempt to force a reset via the Fast A20 and Init register 713 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate. 714 * Bit 0 asserts INIT# when set to 1. We are careful to only 715 * preserve bit 1 while setting bit 0. We also must clear bit 716 * 0 before setting it if it isn't already clear. 717 */ 718 b = inb(0x92); 719 if (b != 0xff) { 720 if ((b & 0x1) != 0) 721 outb(0x92, b & 0xfe); 722 outb(0x92, b | 0x1); 723 DELAY(500000); /* wait 0.5 sec to see if that did it */ 724 } 725#endif /* PC98 */ 726 727 printf("No known reset method worked, attempting CPU shutdown\n"); 728 DELAY(1000000); /* wait 1 sec for printf to complete */ 729 730 /* Wipe the IDT. */ 731 null_idt.rd_limit = 0; 732 null_idt.rd_base = 0; 733 lidt(&null_idt); 734 735 /* "good night, sweet prince .... <THUNK!>" */ 736 breakpoint(); 737 738 /* NOTREACHED */ 739 while(1); 740} 741 742/* 743 * Get an sf_buf from the freelist. May block if none are available. 744 */ 745void 746sf_buf_map(struct sf_buf *sf, int flags) 747{ 748 pt_entry_t opte, *ptep; 749 750 /* 751 * Update the sf_buf's virtual-to-physical mapping, flushing the 752 * virtual address from the TLB. Since the reference count for 753 * the sf_buf's old mapping was zero, that mapping is not 754 * currently in use. Consequently, there is no need to exchange 755 * the old and new PTEs atomically, even under PAE. 756 */ 757 ptep = vtopte(sf->kva); 758 opte = *ptep; 759 *ptep = VM_PAGE_TO_PHYS(sf->m) | pgeflag | PG_RW | PG_V | 760 pmap_cache_bits(sf->m->md.pat_mode, 0); 761 762 /* 763 * Avoid unnecessary TLB invalidations: If the sf_buf's old 764 * virtual-to-physical mapping was not used, then any processor 765 * that has invalidated the sf_buf's virtual address from its TLB 766 * since the last used mapping need not invalidate again. 767 */ 768#ifdef SMP 769 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A)) 770 CPU_ZERO(&sf->cpumask); 771 772 sf_buf_shootdown(sf, flags); 773#else 774 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A)) 775 pmap_invalidate_page(kernel_pmap, sf->kva); 776#endif 777} 778 779#ifdef SMP 780void 781sf_buf_shootdown(struct sf_buf *sf, int flags) 782{ 783 cpuset_t other_cpus; 784 u_int cpuid; 785 786 sched_pin(); 787 cpuid = PCPU_GET(cpuid); 788 if (!CPU_ISSET(cpuid, &sf->cpumask)) { 789 CPU_SET(cpuid, &sf->cpumask); 790 invlpg(sf->kva); 791 } 792 if ((flags & SFB_CPUPRIVATE) == 0) { 793 other_cpus = all_cpus; 794 CPU_CLR(cpuid, &other_cpus); 795 CPU_NAND(&other_cpus, &sf->cpumask); 796 if (!CPU_EMPTY(&other_cpus)) { 797 CPU_OR(&sf->cpumask, &other_cpus); 798 smp_masked_invlpg(other_cpus, sf->kva); 799 } 800 } 801 sched_unpin(); 802} 803#endif 804 805/* 806 * MD part of sf_buf_free(). 807 */ 808int 809sf_buf_unmap(struct sf_buf *sf) 810{ 811 812 return (0); 813} 814 815static void 816sf_buf_invalidate(struct sf_buf *sf) 817{ 818 vm_page_t m = sf->m; 819 820 /* 821 * Use pmap_qenter to update the pte for 822 * existing mapping, in particular, the PAT 823 * settings are recalculated. 824 */ 825 pmap_qenter(sf->kva, &m, 1); 826 pmap_invalidate_cache_range(sf->kva, sf->kva + PAGE_SIZE, FALSE); 827} 828 829/* 830 * Invalidate the cache lines that may belong to the page, if 831 * (possibly old) mapping of the page by sf buffer exists. Returns 832 * TRUE when mapping was found and cache invalidated. 833 */ 834boolean_t 835sf_buf_invalidate_cache(vm_page_t m) 836{ 837 838 return (sf_buf_process_page(m, sf_buf_invalidate)); 839} 840 841/* 842 * Software interrupt handler for queued VM system processing. 843 */ 844void 845swi_vm(void *dummy) 846{ 847 if (busdma_swi_pending != 0) 848 busdma_swi(); 849} 850 851/* 852 * Tell whether this address is in some physical memory region. 853 * Currently used by the kernel coredump code in order to avoid 854 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 855 * or other unpredictable behaviour. 856 */ 857 858int 859is_physical_memory(vm_paddr_t addr) 860{ 861 862#ifdef DEV_ISA 863 /* The ISA ``memory hole''. */ 864 if (addr >= 0xa0000 && addr < 0x100000) 865 return 0; 866#endif 867 868 /* 869 * stuff other tests for known memory-mapped devices (PCI?) 870 * here 871 */ 872 873 return 1; 874} 875