vm_machdep.c revision 43387
11558Srgrimes/*- 21558Srgrimes * Copyright (c) 1982, 1986 The Regents of the University of California. 31558Srgrimes * Copyright (c) 1989, 1990 William Jolitz 41558Srgrimes * Copyright (c) 1994 John Dyson 51558Srgrimes * All rights reserved. 61558Srgrimes * 71558Srgrimes * This code is derived from software contributed to Berkeley by 81558Srgrimes * the Systems Programming Group of the University of Utah Computer 91558Srgrimes * Science Department, and William Jolitz. 101558Srgrimes * 111558Srgrimes * Redistribution and use in source and binary forms, with or without 121558Srgrimes * modification, are permitted provided that the following conditions 131558Srgrimes * are met: 141558Srgrimes * 1. Redistributions of source code must retain the above copyright 151558Srgrimes * notice, this list of conditions and the following disclaimer. 161558Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 171558Srgrimes * notice, this list of conditions and the following disclaimer in the 181558Srgrimes * documentation and/or other materials provided with the distribution. 191558Srgrimes * 3. All advertising materials mentioning features or use of this software 201558Srgrimes * must display the following acknowledgement: 211558Srgrimes * This product includes software developed by the University of 221558Srgrimes * California, Berkeley and its contributors. 231558Srgrimes * 4. Neither the name of the University nor the names of its contributors 241558Srgrimes * may be used to endorse or promote products derived from this software 251558Srgrimes * without specific prior written permission. 261558Srgrimes * 271558Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 281558Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 291558Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 301558Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 311558Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 321558Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 331558Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 341558Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 351558Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 361558Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 371558Srgrimes * SUCH DAMAGE. 381558Srgrimes * 391558Srgrimes * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91 401558Srgrimes * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ 411558Srgrimes * $Id: vm_machdep.c,v 1.115 1999/01/06 23:05:37 julian Exp $ 421558Srgrimes */ 431558Srgrimes 441558Srgrimes#include "npx.h" 451558Srgrimes#include "opt_user_ldt.h" 461558Srgrimes#include "opt_vm86.h" 471558Srgrimes#ifdef PC98 481558Srgrimes#include "opt_pc98.h" 491558Srgrimes#endif 501558Srgrimes 511558Srgrimes#include <sys/param.h> 521558Srgrimes#include <sys/systm.h> 531558Srgrimes#include <sys/proc.h> 541558Srgrimes#include <sys/malloc.h> 551558Srgrimes#include <sys/buf.h> 561558Srgrimes#include <sys/vnode.h> 571558Srgrimes#include <sys/vmmeter.h> 581558Srgrimes#include <sys/kernel.h> 591558Srgrimes#include <sys/sysctl.h> 601558Srgrimes 611558Srgrimes#include <machine/clock.h> 621558Srgrimes#include <machine/cpu.h> 631558Srgrimes#include <machine/md_var.h> 641558Srgrimes#ifdef SMP 651558Srgrimes#include <machine/smp.h> 661558Srgrimes#endif 671558Srgrimes#ifdef VM86 681558Srgrimes#include <machine/pcb_ext.h> 691558Srgrimes#include <machine/vm86.h> 701558Srgrimes#endif 711558Srgrimes 721558Srgrimes#include <vm/vm.h> 731558Srgrimes#include <vm/vm_param.h> 741558Srgrimes#include <vm/vm_prot.h> 751558Srgrimes#include <sys/lock.h> 761558Srgrimes#include <vm/vm_kern.h> 771558Srgrimes#include <vm/vm_page.h> 781558Srgrimes#include <vm/vm_map.h> 791558Srgrimes#include <vm/vm_extern.h> 801558Srgrimes 811558Srgrimes#include <sys/user.h> 821558Srgrimes 831558Srgrimes#ifdef PC98 841558Srgrimes#include <pc98/pc98/pc98.h> 851558Srgrimes#else 861558Srgrimes#include <i386/isa/isa.h> 871558Srgrimes#endif 881558Srgrimes 891558Srgrimesstatic void cpu_reset_real __P((void)); 901558Srgrimes#ifdef SMP 911558Srgrimesstatic void cpu_reset_proxy __P((void)); 921558Srgrimesstatic u_int cpu_reset_proxyid; 931558Srgrimesstatic volatile u_int cpu_reset_proxy_active; 941558Srgrimes#endif 951558Srgrimes 961558Srgrimes/* 971558Srgrimes * quick version of vm_fault 981558Srgrimes */ 991558Srgrimesvoid 1001558Srgrimesvm_fault_quick(v, prot) 1011558Srgrimes caddr_t v; 1021558Srgrimes int prot; 1031558Srgrimes{ 1041558Srgrimes if (prot & VM_PROT_WRITE) 1051558Srgrimes subyte(v, fubyte(v)); 1061558Srgrimes else 1071558Srgrimes fubyte(v); 1081558Srgrimes} 1091558Srgrimes 1101558Srgrimes/* 1111558Srgrimes * Finish a fork operation, with process p2 nearly set up. 1121558Srgrimes * Copy and update the pcb, set up the stack so that the child 1131558Srgrimes * ready to run and return to user mode. 1141558Srgrimes */ 1151558Srgrimesvoid 1161558Srgrimescpu_fork(p1, p2) 1171558Srgrimes register struct proc *p1, *p2; 1181558Srgrimes{ 1191558Srgrimes struct pcb *pcb2 = &p2->p_addr->u_pcb; 1201558Srgrimes 1211558Srgrimes#if NNPX > 0 1221558Srgrimes /* Ensure that p1's pcb is up to date. */ 1231558Srgrimes if (npxproc == p1) 1241558Srgrimes npxsave(&p1->p_addr->u_pcb.pcb_savefpu); 1251558Srgrimes#endif 1261558Srgrimes 1271558Srgrimes /* Copy p1's pcb. */ 1281558Srgrimes p2->p_addr->u_pcb = p1->p_addr->u_pcb; 1291558Srgrimes 1301558Srgrimes /* 1311558Srgrimes * Create a new fresh stack for the new process. 1321558Srgrimes * Copy the trap frame for the return to user mode as if from a 1331558Srgrimes * syscall. This copies the user mode register values. 1341558Srgrimes */ 1351558Srgrimes p2->p_md.md_regs = (struct trapframe *) 1361558Srgrimes#ifdef VM86 1371558Srgrimes ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1; 1381558Srgrimes#else 1391558Srgrimes ((int)p2->p_addr + UPAGES * PAGE_SIZE) - 1; 1401558Srgrimes#endif /* VM86 */ 1411558Srgrimes *p2->p_md.md_regs = *p1->p_md.md_regs; 1421558Srgrimes 1431558Srgrimes /* 1441558Srgrimes * Set registers for trampoline to user mode. Leave space for the 1451558Srgrimes * return address on stack. These are the kernel mode register values. 1461558Srgrimes */ 1471558Srgrimes pcb2->pcb_cr3 = vtophys(p2->p_vmspace->vm_pmap.pm_pdir); 1481558Srgrimes pcb2->pcb_edi = p2->p_md.md_regs->tf_edi; 1491558Srgrimes pcb2->pcb_esi = (int)fork_return; 1501558Srgrimes pcb2->pcb_ebp = p2->p_md.md_regs->tf_ebp; 1511558Srgrimes pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *); 1521558Srgrimes pcb2->pcb_ebx = (int)p2; 1531558Srgrimes pcb2->pcb_eip = (int)fork_trampoline; 1541558Srgrimes /* 1551558Srgrimes * pcb2->pcb_ldt: duplicated below, if necessary. 1561558Srgrimes * pcb2->pcb_ldt_len: cloned above. 1571558Srgrimes * pcb2->pcb_savefpu: cloned above. 1581558Srgrimes * pcb2->pcb_flags: cloned above (always 0 here?). 1591558Srgrimes * pcb2->pcb_onfault: cloned above (always NULL here?). 1601558Srgrimes */ 1611558Srgrimes 1621558Srgrimes#ifdef SMP 1631558Srgrimes pcb2->pcb_mpnest = 1; 1641558Srgrimes#endif 1651558Srgrimes#ifdef VM86 1661558Srgrimes /* 1671558Srgrimes * XXX don't copy the i/o pages. this should probably be fixed. 1681558Srgrimes */ 1691558Srgrimes pcb2->pcb_ext = 0; 1701558Srgrimes#endif 1711558Srgrimes 1721558Srgrimes#ifdef USER_LDT 1731558Srgrimes /* Copy the LDT, if necessary. */ 1741558Srgrimes if (pcb2->pcb_ldt != 0) { 1751558Srgrimes union descriptor *new_ldt; 1761558Srgrimes size_t len = pcb2->pcb_ldt_len * sizeof(union descriptor); 1771558Srgrimes 1781558Srgrimes new_ldt = (union descriptor *)kmem_alloc(kernel_map, len); 1791558Srgrimes bcopy(pcb2->pcb_ldt, new_ldt, len); 1801558Srgrimes pcb2->pcb_ldt = (caddr_t)new_ldt; 1811558Srgrimes } 1821558Srgrimes#endif 1831558Srgrimes 1841558Srgrimes /* 1851558Srgrimes * Now, cpu_switch() can schedule the new process. 1861558Srgrimes * pcb_esp is loaded pointing to the cpu_switch() stack frame 1871558Srgrimes * containing the return address when exiting cpu_switch. 1881558Srgrimes * This will normally be to proc_trampoline(), which will have 1891558Srgrimes * %ebx loaded with the new proc's pointer. proc_trampoline() 1901558Srgrimes * will set up a stack to call fork_return(p, frame); to complete 1911558Srgrimes * the return to user-mode. 1921558Srgrimes */ 1931558Srgrimes} 1941558Srgrimes 1951558Srgrimes/* 1961558Srgrimes * Intercept the return address from a freshly forked process that has NOT 1971558Srgrimes * been scheduled yet. 1981558Srgrimes * 1991558Srgrimes * This is needed to make kernel threads stay in kernel mode. 2001558Srgrimes */ 2011558Srgrimesvoid 2021558Srgrimescpu_set_fork_handler(p, func, arg) 2031558Srgrimes struct proc *p; 2041558Srgrimes void (*func) __P((const void *)); 2051558Srgrimes const void *arg; 2061558Srgrimes{ 2071558Srgrimes /* 2081558Srgrimes * Note that the trap frame follows the args, so the function 2091558Srgrimes * is really called like this: func(arg, frame); 2101558Srgrimes */ 2111558Srgrimes p->p_addr->u_pcb.pcb_esi = (int) func; /* function */ 2121558Srgrimes p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */ 2131558Srgrimes} 2141558Srgrimes 2151558Srgrimesvoid 2161558Srgrimescpu_exit(p) 2171558Srgrimes register struct proc *p; 2181558Srgrimes{ 2191558Srgrimes#if defined(USER_LDT) || defined(VM86) 2201558Srgrimes struct pcb *pcb = &p->p_addr->u_pcb; 2211558Srgrimes#endif 2221558Srgrimes 2231558Srgrimes#if NNPX > 0 2241558Srgrimes npxexit(p); 2251558Srgrimes#endif /* NNPX */ 2261558Srgrimes#ifdef VM86 2271558Srgrimes if (pcb->pcb_ext != 0) { 2281558Srgrimes /* 2291558Srgrimes * XXX do we need to move the TSS off the allocated pages 2301558Srgrimes * before freeing them? (not done here) 2311558Srgrimes */ 2321558Srgrimes kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 2331558Srgrimes ctob(IOPAGES + 1)); 2341558Srgrimes pcb->pcb_ext = 0; 2351558Srgrimes } 2361558Srgrimes#endif 2371558Srgrimes#ifdef USER_LDT 2381558Srgrimes if (pcb->pcb_ldt != 0) { 2391558Srgrimes if (pcb == curpcb) { 2401558Srgrimes lldt(_default_ldt); 2411558Srgrimes currentldt = _default_ldt; 2421558Srgrimes } 2431558Srgrimes kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ldt, 2441558Srgrimes pcb->pcb_ldt_len * sizeof(union descriptor)); 2451558Srgrimes pcb->pcb_ldt_len = (int)pcb->pcb_ldt = 0; 2461558Srgrimes } 2471558Srgrimes#endif 2481558Srgrimes cnt.v_swtch++; 2491558Srgrimes cpu_switch(p); 2501558Srgrimes panic("cpu_exit"); 2511558Srgrimes} 2521558Srgrimes 2531558Srgrimesvoid 2541558Srgrimescpu_wait(p) 2551558Srgrimes struct proc *p; 2561558Srgrimes{ 2571558Srgrimes /* drop per-process resources */ 2581558Srgrimes pmap_dispose_proc(p); 2591558Srgrimes 2601558Srgrimes /* and clean-out the vmspace */ 2611558Srgrimes vmspace_free(p->p_vmspace); 2621558Srgrimes} 2631558Srgrimes 2641558Srgrimes/* 2651558Srgrimes * Dump the machine specific header information at the start of a core dump. 2661558Srgrimes */ 2671558Srgrimesint 2681558Srgrimescpu_coredump(p, vp, cred) 2691558Srgrimes struct proc *p; 2701558Srgrimes struct vnode *vp; 2711558Srgrimes struct ucred *cred; 2721558Srgrimes{ 2731558Srgrimes int error; 2741558Srgrimes caddr_t tempuser; 2751558Srgrimes 2761558Srgrimes tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK); 2771558Srgrimes if (!tempuser) 2781558Srgrimes return EINVAL; 2791558Srgrimes 2801558Srgrimes bzero(tempuser, ctob(UPAGES)); 2811558Srgrimes bcopy(p->p_addr, tempuser, sizeof(struct user)); 2821558Srgrimes bcopy(p->p_md.md_regs, 2831558Srgrimes tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr), 2841558Srgrimes sizeof(struct trapframe)); 2851558Srgrimes 2861558Srgrimes error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, 2871558Srgrimes ctob(UPAGES), 2881558Srgrimes (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, 2891558Srgrimes cred, (int *)NULL, p); 2901558Srgrimes 2911558Srgrimes free(tempuser, M_TEMP); 2921558Srgrimes 2931558Srgrimes return error; 2941558Srgrimes} 2951558Srgrimes 2961558Srgrimes#ifdef notyet 2971558Srgrimesstatic void 2981558Srgrimessetredzone(pte, vaddr) 2991558Srgrimes u_short *pte; 3001558Srgrimes caddr_t vaddr; 3011558Srgrimes{ 3021558Srgrimes/* eventually do this by setting up an expand-down stack segment 3031558Srgrimes for ss0: selector, allowing stack access down to top of u. 3041558Srgrimes this means though that protection violations need to be handled 3051558Srgrimes thru a double fault exception that must do an integral task 3061558Srgrimes switch to a known good context, within which a dump can be 3071558Srgrimes taken. a sensible scheme might be to save the initial context 3081558Srgrimes used by sched (that has physical memory mapped 1:1 at bottom) 3091558Srgrimes and take the dump while still in mapped mode */ 3101558Srgrimes} 31117261Sjkh#endif 3121558Srgrimes 3131558Srgrimes/* 3141558Srgrimes * Convert kernel VA to physical address 3151558Srgrimes */ 3161558Srgrimesu_long 3171558Srgrimeskvtop(void *addr) 31817261Sjkh{ 31917261Sjkh vm_offset_t va; 3201558Srgrimes 3211558Srgrimes va = pmap_kextract((vm_offset_t)addr); 3221558Srgrimes if (va == 0) 3231558Srgrimes panic("kvtop: zero page frame"); 3241558Srgrimes return((int)va); 3251558Srgrimes} 3261558Srgrimes 3271558Srgrimes/* 3281558Srgrimes * Map an IO request into kernel virtual address space. 3291558Srgrimes * 3301558Srgrimes * All requests are (re)mapped into kernel VA space. 3311558Srgrimes * Notice that we use b_bufsize for the size of the buffer 3321558Srgrimes * to be mapped. b_bcount might be modified by the driver. 3331558Srgrimes */ 3341558Srgrimesvoid 3351558Srgrimesvmapbuf(bp) 3361558Srgrimes register struct buf *bp; 3371558Srgrimes{ 3381558Srgrimes register caddr_t addr, v, kva; 3391558Srgrimes vm_offset_t pa; 3401558Srgrimes 3411558Srgrimes if ((bp->b_flags & B_PHYS) == 0) 3421558Srgrimes panic("vmapbuf"); 3431558Srgrimes 3441558Srgrimes for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 3451558Srgrimes addr < bp->b_data + bp->b_bufsize; 3461558Srgrimes addr += PAGE_SIZE, v += PAGE_SIZE) { 3471558Srgrimes /* 348 * Do the vm_fault if needed; do the copy-on-write thing 349 * when reading stuff off device into memory. 350 */ 351 vm_fault_quick(addr, 352 (bp->b_flags&B_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 353 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 354 if (pa == 0) 355 panic("vmapbuf: page not present"); 356 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 357 pmap_kenter((vm_offset_t) v, pa); 358 } 359 360 kva = bp->b_saveaddr; 361 bp->b_saveaddr = bp->b_data; 362 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 363} 364 365/* 366 * Free the io map PTEs associated with this IO operation. 367 * We also invalidate the TLB entries and restore the original b_addr. 368 */ 369void 370vunmapbuf(bp) 371 register struct buf *bp; 372{ 373 register caddr_t addr; 374 vm_offset_t pa; 375 376 if ((bp->b_flags & B_PHYS) == 0) 377 panic("vunmapbuf"); 378 379 for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 380 addr < bp->b_data + bp->b_bufsize; 381 addr += PAGE_SIZE) { 382 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 383 pmap_kremove((vm_offset_t) addr); 384 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 385 } 386 387 bp->b_data = bp->b_saveaddr; 388} 389 390/* 391 * Force reset the processor by invalidating the entire address space! 392 */ 393 394#ifdef SMP 395static void 396cpu_reset_proxy() 397{ 398 u_int saved_mp_lock; 399 400 cpu_reset_proxy_active = 1; 401 while (cpu_reset_proxy_active == 1) 402 ; /* Wait for other cpu to disable interupts */ 403 saved_mp_lock = mp_lock; 404 mp_lock = 1; 405 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n"); 406 cpu_reset_proxy_active = 3; 407 while (cpu_reset_proxy_active == 3) 408 ; /* Wait for other cpu to enable interrupts */ 409 stop_cpus((1<<cpu_reset_proxyid)); 410 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 411 DELAY(1000000); 412 cpu_reset_real(); 413} 414#endif 415 416void 417cpu_reset() 418{ 419#ifdef SMP 420 if (smp_active == 0) { 421 cpu_reset_real(); 422 /* NOTREACHED */ 423 } else { 424 425 u_int map; 426 int cnt; 427 printf("cpu_reset called on cpu#%d\n",cpuid); 428 429 map = other_cpus & ~ stopped_cpus; 430 431 if (map != 0) { 432 printf("cpu_reset: Stopping other CPUs\n"); 433 stop_cpus(map); /* Stop all other CPUs */ 434 } 435 436 if (cpuid == 0) { 437 DELAY(1000000); 438 cpu_reset_real(); 439 /* NOTREACHED */ 440 } else { 441 /* We are not BSP (CPU #0) */ 442 443 cpu_reset_proxyid = cpuid; 444 cpustop_restartfunc = cpu_reset_proxy; 445 printf("cpu_reset: Restarting BSP\n"); 446 started_cpus = (1<<0); /* Restart CPU #0 */ 447 448 cnt = 0; 449 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 450 cnt++; /* Wait for BSP to announce restart */ 451 if (cpu_reset_proxy_active == 0) 452 printf("cpu_reset: Failed to restart BSP\n"); 453 __asm __volatile("cli" : : : "memory"); 454 cpu_reset_proxy_active = 2; 455 cnt = 0; 456 while (cpu_reset_proxy_active == 2 && cnt < 10000000) 457 cnt++; /* Do nothing */ 458 if (cpu_reset_proxy_active == 2) { 459 printf("cpu_reset: BSP did not grab mp lock\n"); 460 cpu_reset_real(); /* XXX: Bogus ? */ 461 } 462 cpu_reset_proxy_active = 4; 463 __asm __volatile("sti" : : : "memory"); 464 while (1); 465 /* NOTREACHED */ 466 } 467 } 468#else 469 cpu_reset_real(); 470#endif 471} 472 473static void 474cpu_reset_real() 475{ 476 477#ifdef PC98 478 /* 479 * Attempt to do a CPU reset via CPU reset port. 480 */ 481 disable_intr(); 482 if ((inb(0x35) & 0xa0) != 0xa0) { 483 outb(0x37, 0x0f); /* SHUT0 = 0. */ 484 outb(0x37, 0x0b); /* SHUT1 = 0. */ 485 } 486 outb(0xf0, 0x00); /* Reset. */ 487#else 488 /* 489 * Attempt to do a CPU reset via the keyboard controller, 490 * do not turn of the GateA20, as any machine that fails 491 * to do the reset here would then end up in no man's land. 492 */ 493 494#if !defined(BROKEN_KEYBOARD_RESET) 495 outb(IO_KBD + 4, 0xFE); 496 DELAY(500000); /* wait 0.5 sec to see if that did it */ 497 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 498 DELAY(1000000); /* wait 1 sec for printf to complete */ 499#endif 500#endif /* PC98 */ 501 /* force a shutdown by unmapping entire address space ! */ 502 bzero((caddr_t) PTD, PAGE_SIZE); 503 504 /* "good night, sweet prince .... <THUNK!>" */ 505 invltlb(); 506 /* NOTREACHED */ 507 while(1); 508} 509 510#ifndef VM_STACK 511/* 512 * Grow the user stack to allow for 'sp'. This version grows the stack in 513 * chunks of SGROWSIZ. 514 */ 515int 516grow(p, sp) 517 struct proc *p; 518 u_int sp; 519{ 520 unsigned int nss; 521 caddr_t v; 522 struct vmspace *vm = p->p_vmspace; 523 524 if ((caddr_t)sp <= vm->vm_maxsaddr || sp >= USRSTACK) 525 return (1); 526 527 nss = roundup(USRSTACK - sp, PAGE_SIZE); 528 529 if (nss > p->p_rlimit[RLIMIT_STACK].rlim_cur) 530 return (0); 531 532 if (vm->vm_ssize && roundup(vm->vm_ssize << PAGE_SHIFT, 533 SGROWSIZ) < nss) { 534 int grow_amount; 535 /* 536 * If necessary, grow the VM that the stack occupies 537 * to allow for the rlimit. This allows us to not have 538 * to allocate all of the VM up-front in execve (which 539 * is expensive). 540 * Grow the VM by the amount requested rounded up to 541 * the nearest SGROWSIZ to provide for some hysteresis. 542 */ 543 grow_amount = roundup((nss - (vm->vm_ssize << PAGE_SHIFT)), SGROWSIZ); 544 v = (char *)USRSTACK - roundup(vm->vm_ssize << PAGE_SHIFT, 545 SGROWSIZ) - grow_amount; 546 /* 547 * If there isn't enough room to extend by SGROWSIZ, then 548 * just extend to the maximum size 549 */ 550 if (v < vm->vm_maxsaddr) { 551 v = vm->vm_maxsaddr; 552 grow_amount = MAXSSIZ - (vm->vm_ssize << PAGE_SHIFT); 553 } 554 if ((grow_amount == 0) || (vm_map_find(&vm->vm_map, NULL, 0, (vm_offset_t *)&v, 555 grow_amount, FALSE, VM_PROT_ALL, VM_PROT_ALL, 0) != KERN_SUCCESS)) { 556 return (0); 557 } 558 vm->vm_ssize += grow_amount >> PAGE_SHIFT; 559 } 560 561 return (1); 562} 563#else 564int 565grow_stack(p, sp) 566 struct proc *p; 567 u_int sp; 568{ 569 int rv; 570 571 rv = vm_map_growstack (p, sp); 572 if (rv != KERN_SUCCESS) 573 return (0); 574 575 return (1); 576} 577#endif 578 579 580static int cnt_prezero; 581 582SYSCTL_INT(_vm_stats_misc, OID_AUTO, 583 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, ""); 584 585/* 586 * Implement the pre-zeroed page mechanism. 587 * This routine is called from the idle loop. 588 */ 589int 590vm_page_zero_idle() 591{ 592 static int free_rover; 593 vm_page_t m; 594 int s; 595 596 /* 597 * XXX 598 * We stop zeroing pages when there are sufficent prezeroed pages. 599 * This threshold isn't really needed, except we want to 600 * bypass unneeded calls to vm_page_list_find, and the 601 * associated cache flush and latency. The pre-zero will 602 * still be called when there are significantly more 603 * non-prezeroed pages than zeroed pages. The threshold 604 * of half the number of reserved pages is arbitrary, but 605 * approximately the right amount. Eventually, we should 606 * perhaps interrupt the zero operation when a process 607 * is found to be ready to run. 608 */ 609 if (cnt.v_free_count - vm_page_zero_count <= cnt.v_free_reserved / 2) 610 return (0); 611#ifdef SMP 612 if (try_mplock()) { 613#endif 614 s = splvm(); 615 __asm __volatile("sti" : : : "memory"); 616 m = vm_page_list_find(PQ_FREE, free_rover); 617 if (m != NULL) { 618 --(*vm_page_queues[m->queue].lcnt); 619 TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq); 620 m->queue = PQ_NONE; 621 splx(s); 622#if 0 623 rel_mplock(); 624#endif 625 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 626#if 0 627 get_mplock(); 628#endif 629 (void)splvm(); 630 m->queue = PQ_ZERO + m->pc; 631 ++(*vm_page_queues[m->queue].lcnt); 632 TAILQ_INSERT_HEAD(vm_page_queues[m->queue].pl, m, 633 pageq); 634 free_rover = (free_rover + PQ_PRIME3) & PQ_L2_MASK; 635 ++vm_page_zero_count; 636 ++cnt_prezero; 637 } 638 splx(s); 639 __asm __volatile("cli" : : : "memory"); 640#ifdef SMP 641 rel_mplock(); 642#endif 643 return (1); 644#ifdef SMP 645 } 646#endif 647 return (0); 648} 649 650/* 651 * Software interrupt handler for queued VM system processing. 652 */ 653void 654swi_vm() 655{ 656 if (busdma_swi_pending != 0) 657 busdma_swi(); 658} 659 660/* 661 * Tell whether this address is in some physical memory region. 662 * Currently used by the kernel coredump code in order to avoid 663 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 664 * or other unpredictable behaviour. 665 */ 666 667#include "isa.h" 668 669int 670is_physical_memory(addr) 671 vm_offset_t addr; 672{ 673 674#if NISA > 0 675 /* The ISA ``memory hole''. */ 676 if (addr >= 0xa0000 && addr < 0x100000) 677 return 0; 678#endif 679 680 /* 681 * stuff other tests for known memory-mapped devices (PCI?) 682 * here 683 */ 684 685 return 1; 686} 687