vm_machdep.c revision 39648
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 * $Id: vm_machdep.c,v 1.109 1998/08/18 07:46:58 msmith Exp $ 42 */ 43 44#include "npx.h" 45#include "opt_user_ldt.h" 46#include "opt_vm86.h" 47#ifdef PC98 48#include "opt_pc98.h" 49#endif 50 51#include <sys/param.h> 52#include <sys/systm.h> 53#include <sys/proc.h> 54#include <sys/malloc.h> 55#include <sys/buf.h> 56#include <sys/vnode.h> 57#include <sys/vmmeter.h> 58#include <sys/kernel.h> 59#include <sys/sysctl.h> 60 61#include <machine/clock.h> 62#include <machine/cpu.h> 63#include <machine/md_var.h> 64#ifdef SMP 65#include <machine/smp.h> 66#endif 67#ifdef VM86 68#include <machine/pcb_ext.h> 69#include <machine/vm86.h> 70#endif 71 72#include <vm/vm.h> 73#include <vm/vm_param.h> 74#include <vm/vm_prot.h> 75#include <sys/lock.h> 76#include <vm/vm_kern.h> 77#include <vm/vm_page.h> 78#include <vm/vm_map.h> 79#include <vm/vm_extern.h> 80 81#include <sys/user.h> 82 83#ifdef PC98 84#include <pc98/pc98/pc98.h> 85#else 86#include <i386/isa/isa.h> 87#endif 88 89static void cpu_reset_real __P((void)); 90#ifdef SMP 91static void cpu_reset_proxy __P((void)); 92static u_int cpu_reset_proxyid; 93static volatile u_int cpu_reset_proxy_active; 94#endif 95 96/* 97 * quick version of vm_fault 98 */ 99void 100vm_fault_quick(v, prot) 101 caddr_t v; 102 int prot; 103{ 104 if (prot & VM_PROT_WRITE) 105 subyte(v, fubyte(v)); 106 else 107 fubyte(v); 108} 109 110/* 111 * Finish a fork operation, with process p2 nearly set up. 112 * Copy and update the pcb, set up the stack so that the child 113 * ready to run and return to user mode. 114 */ 115void 116cpu_fork(p1, p2) 117 register struct proc *p1, *p2; 118{ 119 struct pcb *pcb2 = &p2->p_addr->u_pcb; 120 121#if NNPX > 0 122 /* Ensure that p1's pcb is up to date. */ 123 if (npxproc == p1) 124 npxsave(&p1->p_addr->u_pcb.pcb_savefpu); 125#endif 126 127 /* Copy p1's pcb. */ 128 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 129 130 /* 131 * Create a new fresh stack for the new process. 132 * Copy the trap frame for the return to user mode as if from a 133 * syscall. This copies the user mode register values. 134 */ 135 p2->p_md.md_regs = (struct trapframe *) 136#ifdef VM86 137 ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1; 138#else 139 ((int)p2->p_addr + UPAGES * PAGE_SIZE) - 1; 140#endif /* VM86 */ 141 *p2->p_md.md_regs = *p1->p_md.md_regs; 142 143 /* 144 * Set registers for trampoline to user mode. Leave space for the 145 * return address on stack. These are the kernel mode register values. 146 */ 147 pcb2->pcb_cr3 = vtophys(p2->p_vmspace->vm_pmap.pm_pdir); 148 pcb2->pcb_edi = p2->p_md.md_regs->tf_edi; 149 pcb2->pcb_esi = (int)fork_return; 150 pcb2->pcb_ebp = p2->p_md.md_regs->tf_ebp; 151 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *); 152 pcb2->pcb_ebx = (int)p2; 153 pcb2->pcb_eip = (int)fork_trampoline; 154 /* 155 * pcb2->pcb_ldt: duplicated below, if necessary. 156 * pcb2->pcb_ldt_len: cloned above. 157 * pcb2->pcb_savefpu: cloned above. 158 * pcb2->pcb_flags: cloned above (always 0 here?). 159 * pcb2->pcb_onfault: cloned above (always NULL here?). 160 */ 161 162#ifdef VM86 163 /* 164 * XXX don't copy the i/o pages. this should probably be fixed. 165 */ 166 pcb2->pcb_ext = 0; 167#endif 168 169#ifdef USER_LDT 170 /* Copy the LDT, if necessary. */ 171 if (pcb2->pcb_ldt != 0) { 172 union descriptor *new_ldt; 173 size_t len = pcb2->pcb_ldt_len * sizeof(union descriptor); 174 175 new_ldt = (union descriptor *)kmem_alloc(kernel_map, len); 176 bcopy(pcb2->pcb_ldt, new_ldt, len); 177 pcb2->pcb_ldt = (caddr_t)new_ldt; 178 } 179#endif 180 181 /* 182 * Now, cpu_switch() can schedule the new process. 183 * pcb_esp is loaded pointing to the cpu_switch() stack frame 184 * containing the return address when exiting cpu_switch. 185 * This will normally be to proc_trampoline(), which will have 186 * %ebx loaded with the new proc's pointer. proc_trampoline() 187 * will set up a stack to call fork_return(p, frame); to complete 188 * the return to user-mode. 189 */ 190} 191 192/* 193 * Intercept the return address from a freshly forked process that has NOT 194 * been scheduled yet. 195 * 196 * This is needed to make kernel threads stay in kernel mode. 197 */ 198void 199cpu_set_fork_handler(p, func, arg) 200 struct proc *p; 201 void (*func) __P((void *)); 202 void *arg; 203{ 204 /* 205 * Note that the trap frame follows the args, so the function 206 * is really called like this: func(arg, frame); 207 */ 208 p->p_addr->u_pcb.pcb_esi = (int) func; /* function */ 209 p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */ 210} 211 212void 213cpu_exit(p) 214 register struct proc *p; 215{ 216#if defined(USER_LDT) || defined(VM86) 217 struct pcb *pcb = &p->p_addr->u_pcb; 218#endif 219 220#if NNPX > 0 221 npxexit(p); 222#endif /* NNPX */ 223#ifdef VM86 224 if (pcb->pcb_ext != 0) { 225 /* 226 * XXX do we need to move the TSS off the allocated pages 227 * before freeing them? (not done here) 228 */ 229 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 230 ctob(IOPAGES + 1)); 231 pcb->pcb_ext = 0; 232 } 233#endif 234#ifdef USER_LDT 235 if (pcb->pcb_ldt != 0) { 236 if (pcb == curpcb) { 237 lldt(_default_ldt); 238 currentldt = _default_ldt; 239 } 240 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ldt, 241 pcb->pcb_ldt_len * sizeof(union descriptor)); 242 pcb->pcb_ldt_len = (int)pcb->pcb_ldt = 0; 243 } 244#endif 245 cnt.v_swtch++; 246 cpu_switch(p); 247 panic("cpu_exit"); 248} 249 250void 251cpu_wait(p) 252 struct proc *p; 253{ 254 /* drop per-process resources */ 255 pmap_dispose_proc(p); 256 257 /* and clean-out the vmspace */ 258 vmspace_free(p->p_vmspace); 259} 260 261/* 262 * Dump the machine specific header information at the start of a core dump. 263 */ 264int 265cpu_coredump(p, vp, cred) 266 struct proc *p; 267 struct vnode *vp; 268 struct ucred *cred; 269{ 270 int error; 271 caddr_t tempuser; 272 273 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK); 274 if (!tempuser) 275 return EINVAL; 276 277 bzero(tempuser, ctob(UPAGES)); 278 bcopy(p->p_addr, tempuser, sizeof(struct user)); 279 bcopy(p->p_md.md_regs, 280 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr), 281 sizeof(struct trapframe)); 282 283 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, 284 ctob(UPAGES), 285 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, 286 cred, (int *)NULL, p); 287 288 free(tempuser, M_TEMP); 289 290 return error; 291} 292 293#ifdef notyet 294static void 295setredzone(pte, vaddr) 296 u_short *pte; 297 caddr_t vaddr; 298{ 299/* eventually do this by setting up an expand-down stack segment 300 for ss0: selector, allowing stack access down to top of u. 301 this means though that protection violations need to be handled 302 thru a double fault exception that must do an integral task 303 switch to a known good context, within which a dump can be 304 taken. a sensible scheme might be to save the initial context 305 used by sched (that has physical memory mapped 1:1 at bottom) 306 and take the dump while still in mapped mode */ 307} 308#endif 309 310/* 311 * Convert kernel VA to physical address 312 */ 313u_long 314kvtop(void *addr) 315{ 316 vm_offset_t va; 317 318 va = pmap_kextract((vm_offset_t)addr); 319 if (va == 0) 320 panic("kvtop: zero page frame"); 321 return((int)va); 322} 323 324/* 325 * Map an IO request into kernel virtual address space. 326 * 327 * All requests are (re)mapped into kernel VA space. 328 * Notice that we use b_bufsize for the size of the buffer 329 * to be mapped. b_bcount might be modified by the driver. 330 */ 331void 332vmapbuf(bp) 333 register struct buf *bp; 334{ 335 register caddr_t addr, v, kva; 336 vm_offset_t pa; 337 338 if ((bp->b_flags & B_PHYS) == 0) 339 panic("vmapbuf"); 340 341 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page(bp->b_data); 342 addr < bp->b_data + bp->b_bufsize; 343 addr += PAGE_SIZE, v += PAGE_SIZE) { 344 /* 345 * Do the vm_fault if needed; do the copy-on-write thing 346 * when reading stuff off device into memory. 347 */ 348 vm_fault_quick(addr, 349 (bp->b_flags&B_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 350 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 351 if (pa == 0) 352 panic("vmapbuf: page not present"); 353 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 354 pmap_kenter((vm_offset_t) v, pa); 355 } 356 357 kva = bp->b_saveaddr; 358 bp->b_saveaddr = bp->b_data; 359 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 360} 361 362/* 363 * Free the io map PTEs associated with this IO operation. 364 * We also invalidate the TLB entries and restore the original b_addr. 365 */ 366void 367vunmapbuf(bp) 368 register struct buf *bp; 369{ 370 register caddr_t addr; 371 vm_offset_t pa; 372 373 if ((bp->b_flags & B_PHYS) == 0) 374 panic("vunmapbuf"); 375 376 for (addr = (caddr_t)trunc_page(bp->b_data); 377 addr < bp->b_data + bp->b_bufsize; 378 addr += PAGE_SIZE) { 379 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 380 pmap_kremove((vm_offset_t) addr); 381 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 382 } 383 384 bp->b_data = bp->b_saveaddr; 385} 386 387/* 388 * Force reset the processor by invalidating the entire address space! 389 */ 390 391#ifdef SMP 392static void 393cpu_reset_proxy() 394{ 395 u_int saved_mp_lock; 396 397 cpu_reset_proxy_active = 1; 398 while (cpu_reset_proxy_active == 1) 399 ; /* Wait for other cpu to disable interupts */ 400 saved_mp_lock = mp_lock; 401 mp_lock = 1; 402 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n"); 403 cpu_reset_proxy_active = 3; 404 while (cpu_reset_proxy_active == 3) 405 ; /* Wait for other cpu to enable interrupts */ 406 stop_cpus((1<<cpu_reset_proxyid)); 407 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 408 DELAY(1000000); 409 cpu_reset_real(); 410} 411#endif 412 413void 414cpu_reset() 415{ 416#ifdef SMP 417 if (smp_active == 0) { 418 cpu_reset_real(); 419 /* NOTREACHED */ 420 } else { 421 422 u_int map; 423 int cnt; 424 printf("cpu_reset called on cpu#%d\n",cpuid); 425 426 map = other_cpus & ~ stopped_cpus; 427 428 if (map != 0) { 429 printf("cpu_reset: Stopping other CPUs\n"); 430 stop_cpus(map); /* Stop all other CPUs */ 431 } 432 433 if (cpuid == 0) { 434 DELAY(1000000); 435 cpu_reset_real(); 436 /* NOTREACHED */ 437 } else { 438 /* We are not BSP (CPU #0) */ 439 440 cpu_reset_proxyid = cpuid; 441 cpustop_restartfunc = cpu_reset_proxy; 442 printf("cpu_reset: Restarting BSP\n"); 443 started_cpus = (1<<0); /* Restart CPU #0 */ 444 445 cnt = 0; 446 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 447 cnt++; /* Wait for BSP to announce restart */ 448 if (cpu_reset_proxy_active == 0) 449 printf("cpu_reset: Failed to restart BSP\n"); 450 __asm __volatile("cli" : : : "memory"); 451 cpu_reset_proxy_active = 2; 452 cnt = 0; 453 while (cpu_reset_proxy_active == 2 && cnt < 10000000) 454 cnt++; /* Do nothing */ 455 if (cpu_reset_proxy_active == 2) { 456 printf("cpu_reset: BSP did not grab mp lock\n"); 457 cpu_reset_real(); /* XXX: Bogus ? */ 458 } 459 cpu_reset_proxy_active = 4; 460 __asm __volatile("sti" : : : "memory"); 461 while (1); 462 /* NOTREACHED */ 463 } 464 } 465#else 466 cpu_reset_real(); 467#endif 468} 469 470static void 471cpu_reset_real() 472{ 473 474#ifdef PC98 475 /* 476 * Attempt to do a CPU reset via CPU reset port. 477 */ 478 disable_intr(); 479 if ((inb(0x35) & 0xa0) != 0xa0) { 480 outb(0x37, 0x0f); /* SHUT0 = 0. */ 481 outb(0x37, 0x0b); /* SHUT1 = 0. */ 482 } 483 outb(0xf0, 0x00); /* Reset. */ 484#else 485 /* 486 * Attempt to do a CPU reset via the keyboard controller, 487 * do not turn of the GateA20, as any machine that fails 488 * to do the reset here would then end up in no man's land. 489 */ 490 491#if !defined(BROKEN_KEYBOARD_RESET) 492 outb(IO_KBD + 4, 0xFE); 493 DELAY(500000); /* wait 0.5 sec to see if that did it */ 494 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 495 DELAY(1000000); /* wait 1 sec for printf to complete */ 496#endif 497#endif /* PC98 */ 498 /* force a shutdown by unmapping entire address space ! */ 499 bzero((caddr_t) PTD, PAGE_SIZE); 500 501 /* "good night, sweet prince .... <THUNK!>" */ 502 invltlb(); 503 /* NOTREACHED */ 504 while(1); 505} 506 507/* 508 * Grow the user stack to allow for 'sp'. This version grows the stack in 509 * chunks of SGROWSIZ. 510 */ 511int 512grow(p, sp) 513 struct proc *p; 514 u_int sp; 515{ 516 unsigned int nss; 517 caddr_t v; 518 struct vmspace *vm = p->p_vmspace; 519 520 if ((caddr_t)sp <= vm->vm_maxsaddr || (unsigned)sp >= (unsigned)USRSTACK) 521 return (1); 522 523 nss = roundup(USRSTACK - (unsigned)sp, PAGE_SIZE); 524 525 if (nss > p->p_rlimit[RLIMIT_STACK].rlim_cur) 526 return (0); 527 528 if (vm->vm_ssize && roundup(vm->vm_ssize << PAGE_SHIFT, 529 SGROWSIZ) < nss) { 530 int grow_amount; 531 /* 532 * If necessary, grow the VM that the stack occupies 533 * to allow for the rlimit. This allows us to not have 534 * to allocate all of the VM up-front in execve (which 535 * is expensive). 536 * Grow the VM by the amount requested rounded up to 537 * the nearest SGROWSIZ to provide for some hysteresis. 538 */ 539 grow_amount = roundup((nss - (vm->vm_ssize << PAGE_SHIFT)), SGROWSIZ); 540 v = (char *)USRSTACK - roundup(vm->vm_ssize << PAGE_SHIFT, 541 SGROWSIZ) - grow_amount; 542 /* 543 * If there isn't enough room to extend by SGROWSIZ, then 544 * just extend to the maximum size 545 */ 546 if (v < vm->vm_maxsaddr) { 547 v = vm->vm_maxsaddr; 548 grow_amount = MAXSSIZ - (vm->vm_ssize << PAGE_SHIFT); 549 } 550 if ((grow_amount == 0) || (vm_map_find(&vm->vm_map, NULL, 0, (vm_offset_t *)&v, 551 grow_amount, FALSE, VM_PROT_ALL, VM_PROT_ALL, 0) != KERN_SUCCESS)) { 552 return (0); 553 } 554 vm->vm_ssize += grow_amount >> PAGE_SHIFT; 555 } 556 557 return (1); 558} 559 560static int cnt_prezero; 561 562SYSCTL_INT(_machdep, OID_AUTO, cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, ""); 563 564/* 565 * Implement the pre-zeroed page mechanism. 566 * This routine is called from the idle loop. 567 */ 568int 569vm_page_zero_idle() 570{ 571 static int free_rover; 572 vm_page_t m; 573 int s; 574 575 /* 576 * XXX 577 * We stop zeroing pages when there are sufficent prezeroed pages. 578 * This threshold isn't really needed, except we want to 579 * bypass unneeded calls to vm_page_list_find, and the 580 * associated cache flush and latency. The pre-zero will 581 * still be called when there are significantly more 582 * non-prezeroed pages than zeroed pages. The threshold 583 * of half the number of reserved pages is arbitrary, but 584 * approximately the right amount. Eventually, we should 585 * perhaps interrupt the zero operation when a process 586 * is found to be ready to run. 587 */ 588 if (cnt.v_free_count - vm_page_zero_count <= cnt.v_free_reserved / 2) 589 return (0); 590#ifdef SMP 591 if (try_mplock()) { 592#endif 593 s = splvm(); 594 __asm __volatile("sti" : : : "memory"); 595 m = vm_page_list_find(PQ_FREE, free_rover); 596 if (m != NULL) { 597 --(*vm_page_queues[m->queue].lcnt); 598 TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq); 599 m->queue = PQ_NONE; 600 splx(s); 601#if 0 602 rel_mplock(); 603#endif 604 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 605#if 0 606 get_mplock(); 607#endif 608 (void)splvm(); 609 m->queue = PQ_ZERO + m->pc; 610 ++(*vm_page_queues[m->queue].lcnt); 611 TAILQ_INSERT_HEAD(vm_page_queues[m->queue].pl, m, 612 pageq); 613 free_rover = (free_rover + PQ_PRIME3) & PQ_L2_MASK; 614 ++vm_page_zero_count; 615 ++cnt_prezero; 616 } 617 splx(s); 618 __asm __volatile("cli" : : : "memory"); 619#ifdef SMP 620 rel_mplock(); 621#endif 622 return (1); 623#ifdef SMP 624 } 625#endif 626 return (0); 627} 628 629/* 630 * Software interrupt handler for queued VM system processing. 631 */ 632void 633swi_vm() 634{ 635 if (busdma_swi_pending != 0) 636 busdma_swi(); 637} 638 639/* 640 * Tell whether this address is in some physical memory region. 641 * Currently used by the kernel coredump code in order to avoid 642 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 643 * or other unpredictable behaviour. 644 */ 645 646#include "isa.h" 647 648int 649is_physical_memory(addr) 650 vm_offset_t addr; 651{ 652 653#if NISA > 0 654 /* The ISA ``memory hole''. */ 655 if (addr >= 0xa0000 && addr < 0x100000) 656 return 0; 657#endif 658 659 /* 660 * stuff other tests for known memory-mapped devices (PCI?) 661 * here 662 */ 663 664 return 1; 665} 666