vm_machdep.c revision 64529
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 * $FreeBSD: head/sys/amd64/amd64/vm_machdep.c 64529 2000-08-11 09:05:12Z peter $ 42 */ 43 44#include "npx.h" 45#include "opt_user_ldt.h" 46#ifdef PC98 47#include "opt_pc98.h" 48#endif 49#include "opt_reset.h" 50 51#include <sys/param.h> 52#include <sys/systm.h> 53#include <sys/malloc.h> 54#include <sys/proc.h> 55#include <sys/bio.h> 56#include <sys/buf.h> 57#include <sys/vnode.h> 58#include <sys/vmmeter.h> 59#include <sys/kernel.h> 60#include <sys/sysctl.h> 61#include <sys/unistd.h> 62 63#include <machine/clock.h> 64#include <machine/cpu.h> 65#include <machine/md_var.h> 66#ifdef SMP 67#include <machine/smp.h> 68#endif 69#include <machine/pcb.h> 70#include <machine/pcb_ext.h> 71#include <machine/vm86.h> 72 73#include <vm/vm.h> 74#include <vm/vm_param.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 95extern int _ucodesel, _udatasel; 96 97/* 98 * quick version of vm_fault 99 */ 100int 101vm_fault_quick(v, prot) 102 caddr_t v; 103 int prot; 104{ 105 int r; 106 107 if (prot & VM_PROT_WRITE) 108 r = subyte(v, fubyte(v)); 109 else 110 r = fubyte(v); 111 return(r); 112} 113 114/* 115 * Finish a fork operation, with process p2 nearly set up. 116 * Copy and update the pcb, set up the stack so that the child 117 * ready to run and return to user mode. 118 */ 119void 120cpu_fork(p1, p2, flags) 121 register struct proc *p1, *p2; 122 int flags; 123{ 124 struct pcb *pcb2; 125 126 if ((flags & RFPROC) == 0) { 127#ifdef USER_LDT 128 if ((flags & RFMEM) == 0) { 129 /* unshare user LDT */ 130 struct pcb *pcb1 = &p1->p_addr->u_pcb; 131 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt; 132 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) { 133 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len); 134 user_ldt_free(pcb1); 135 pcb1->pcb_ldt = pcb_ldt; 136 set_user_ldt(pcb1); 137 } 138 } 139#endif 140 return; 141 } 142 143#if NNPX > 0 144 /* Ensure that p1's pcb is up to date. */ 145 if (npxproc == p1) 146 npxsave(&p1->p_addr->u_pcb.pcb_savefpu); 147#endif 148 149 /* Copy p1's pcb. */ 150 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 151 pcb2 = &p2->p_addr->u_pcb; 152 153 /* 154 * Create a new fresh stack for the new process. 155 * Copy the trap frame for the return to user mode as if from a 156 * syscall. This copies the user mode register values. 157 */ 158 p2->p_md.md_regs = (struct trapframe *) 159 ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1; 160 bcopy(p1->p_md.md_regs, p2->p_md.md_regs, sizeof(*p2->p_md.md_regs)); 161 162 /* 163 * Set registers for trampoline to user mode. Leave space for the 164 * return address on stack. These are the kernel mode register values. 165 */ 166 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir); 167 pcb2->pcb_edi = 0; 168 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */ 169 pcb2->pcb_ebp = 0; 170 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *); 171 pcb2->pcb_ebx = (int)p2; /* fork_trampoline argument */ 172 pcb2->pcb_eip = (int)fork_trampoline; 173 /* 174 * pcb2->pcb_ldt: duplicated below, if necessary. 175 * pcb2->pcb_savefpu: cloned above. 176 * pcb2->pcb_flags: cloned above (always 0 here?). 177 * pcb2->pcb_onfault: cloned above (always NULL here?). 178 */ 179 180#ifdef SMP 181 pcb2->pcb_mpnest = 1; 182#endif 183 /* 184 * XXX don't copy the i/o pages. this should probably be fixed. 185 */ 186 pcb2->pcb_ext = 0; 187 188#ifdef USER_LDT 189 /* Copy the LDT, if necessary. */ 190 if (pcb2->pcb_ldt != 0) { 191 if (flags & RFMEM) { 192 pcb2->pcb_ldt->ldt_refcnt++; 193 } else { 194 pcb2->pcb_ldt = user_ldt_alloc(pcb2, 195 pcb2->pcb_ldt->ldt_len); 196 } 197 } 198#endif 199 200 /* 201 * Now, cpu_switch() can schedule the new process. 202 * pcb_esp is loaded pointing to the cpu_switch() stack frame 203 * containing the return address when exiting cpu_switch. 204 * This will normally be to fork_trampoline(), which will have 205 * %ebx loaded with the new proc's pointer. fork_trampoline() 206 * will set up a stack to call fork_return(p, frame); to complete 207 * the return to user-mode. 208 */ 209} 210 211/* 212 * Intercept the return address from a freshly forked process that has NOT 213 * been scheduled yet. 214 * 215 * This is needed to make kernel threads stay in kernel mode. 216 */ 217void 218cpu_set_fork_handler(p, func, arg) 219 struct proc *p; 220 void (*func) __P((void *)); 221 void *arg; 222{ 223 /* 224 * Note that the trap frame follows the args, so the function 225 * is really called like this: func(arg, frame); 226 */ 227 p->p_addr->u_pcb.pcb_esi = (int) func; /* function */ 228 p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */ 229} 230 231void 232cpu_exit(p) 233 register struct proc *p; 234{ 235 struct pcb *pcb = &p->p_addr->u_pcb; 236 237#if NNPX > 0 238 npxexit(p); 239#endif /* NNPX */ 240 if (pcb->pcb_ext != 0) { 241 /* 242 * XXX do we need to move the TSS off the allocated pages 243 * before freeing them? (not done here) 244 */ 245 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 246 ctob(IOPAGES + 1)); 247 pcb->pcb_ext = 0; 248 } 249#ifdef USER_LDT 250 user_ldt_free(pcb); 251#endif 252 if (pcb->pcb_flags & PCB_DBREGS) { 253 /* 254 * disable all hardware breakpoints 255 */ 256 reset_dbregs(); 257 pcb->pcb_flags &= ~PCB_DBREGS; 258 } 259 cnt.v_swtch++; 260 cpu_switch(p); 261 panic("cpu_exit"); 262} 263 264void 265cpu_wait(p) 266 struct proc *p; 267{ 268 /* drop per-process resources */ 269 pmap_dispose_proc(p); 270 271 /* and clean-out the vmspace */ 272 vmspace_free(p->p_vmspace); 273} 274 275/* 276 * Dump the machine specific header information at the start of a core dump. 277 */ 278int 279cpu_coredump(p, vp, cred) 280 struct proc *p; 281 struct vnode *vp; 282 struct ucred *cred; 283{ 284 int error; 285 caddr_t tempuser; 286 287 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK); 288 if (!tempuser) 289 return EINVAL; 290 291 bzero(tempuser, ctob(UPAGES)); 292 bcopy(p->p_addr, tempuser, sizeof(struct user)); 293 bcopy(p->p_md.md_regs, 294 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr), 295 sizeof(struct trapframe)); 296 297 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, 298 ctob(UPAGES), 299 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, 300 cred, (int *)NULL, p); 301 302 free(tempuser, M_TEMP); 303 304 return error; 305} 306 307#ifdef notyet 308static void 309setredzone(pte, vaddr) 310 u_short *pte; 311 caddr_t vaddr; 312{ 313/* eventually do this by setting up an expand-down stack segment 314 for ss0: selector, allowing stack access down to top of u. 315 this means though that protection violations need to be handled 316 thru a double fault exception that must do an integral task 317 switch to a known good context, within which a dump can be 318 taken. a sensible scheme might be to save the initial context 319 used by sched (that has physical memory mapped 1:1 at bottom) 320 and take the dump while still in mapped mode */ 321} 322#endif 323 324/* 325 * Convert kernel VA to physical address 326 */ 327u_long 328kvtop(void *addr) 329{ 330 vm_offset_t va; 331 332 va = pmap_kextract((vm_offset_t)addr); 333 if (va == 0) 334 panic("kvtop: zero page frame"); 335 return((int)va); 336} 337 338/* 339 * Map an IO request into kernel virtual address space. 340 * 341 * All requests are (re)mapped into kernel VA space. 342 * Notice that we use b_bufsize for the size of the buffer 343 * to be mapped. b_bcount might be modified by the driver. 344 */ 345void 346vmapbuf(bp) 347 register struct buf *bp; 348{ 349 register caddr_t addr, v, kva; 350 vm_offset_t pa; 351 352 if ((bp->b_flags & B_PHYS) == 0) 353 panic("vmapbuf"); 354 355 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 356 addr < bp->b_data + bp->b_bufsize; 357 addr += PAGE_SIZE, v += PAGE_SIZE) { 358 /* 359 * Do the vm_fault if needed; do the copy-on-write thing 360 * when reading stuff off device into memory. 361 */ 362 vm_fault_quick(addr, 363 (bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 364 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 365 if (pa == 0) 366 panic("vmapbuf: page not present"); 367 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 368 pmap_kenter((vm_offset_t) v, pa); 369 } 370 371 kva = bp->b_saveaddr; 372 bp->b_saveaddr = bp->b_data; 373 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 374} 375 376/* 377 * Free the io map PTEs associated with this IO operation. 378 * We also invalidate the TLB entries and restore the original b_addr. 379 */ 380void 381vunmapbuf(bp) 382 register struct buf *bp; 383{ 384 register caddr_t addr; 385 vm_offset_t pa; 386 387 if ((bp->b_flags & B_PHYS) == 0) 388 panic("vunmapbuf"); 389 390 for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 391 addr < bp->b_data + bp->b_bufsize; 392 addr += PAGE_SIZE) { 393 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 394 pmap_kremove((vm_offset_t) addr); 395 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 396 } 397 398 bp->b_data = bp->b_saveaddr; 399} 400 401/* 402 * Force reset the processor by invalidating the entire address space! 403 */ 404 405#ifdef SMP 406static void 407cpu_reset_proxy() 408{ 409 u_int saved_mp_lock; 410 411 cpu_reset_proxy_active = 1; 412 while (cpu_reset_proxy_active == 1) 413 ; /* Wait for other cpu to disable interupts */ 414 saved_mp_lock = mp_lock; 415 mp_lock = 1; 416 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n"); 417 cpu_reset_proxy_active = 3; 418 while (cpu_reset_proxy_active == 3) 419 ; /* Wait for other cpu to enable interrupts */ 420 stop_cpus((1<<cpu_reset_proxyid)); 421 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 422 DELAY(1000000); 423 cpu_reset_real(); 424} 425#endif 426 427void 428cpu_reset() 429{ 430#ifdef SMP 431 if (smp_active == 0) { 432 cpu_reset_real(); 433 /* NOTREACHED */ 434 } else { 435 436 u_int map; 437 int cnt; 438 printf("cpu_reset called on cpu#%d\n",cpuid); 439 440 map = other_cpus & ~ stopped_cpus; 441 442 if (map != 0) { 443 printf("cpu_reset: Stopping other CPUs\n"); 444 stop_cpus(map); /* Stop all other CPUs */ 445 } 446 447 if (cpuid == 0) { 448 DELAY(1000000); 449 cpu_reset_real(); 450 /* NOTREACHED */ 451 } else { 452 /* We are not BSP (CPU #0) */ 453 454 cpu_reset_proxyid = cpuid; 455 cpustop_restartfunc = cpu_reset_proxy; 456 printf("cpu_reset: Restarting BSP\n"); 457 started_cpus = (1<<0); /* Restart CPU #0 */ 458 459 cnt = 0; 460 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 461 cnt++; /* Wait for BSP to announce restart */ 462 if (cpu_reset_proxy_active == 0) 463 printf("cpu_reset: Failed to restart BSP\n"); 464 __asm __volatile("cli" : : : "memory"); 465 cpu_reset_proxy_active = 2; 466 cnt = 0; 467 while (cpu_reset_proxy_active == 2 && cnt < 10000000) 468 cnt++; /* Do nothing */ 469 if (cpu_reset_proxy_active == 2) { 470 printf("cpu_reset: BSP did not grab mp lock\n"); 471 cpu_reset_real(); /* XXX: Bogus ? */ 472 } 473 cpu_reset_proxy_active = 4; 474 __asm __volatile("sti" : : : "memory"); 475 while (1); 476 /* NOTREACHED */ 477 } 478 } 479#else 480 cpu_reset_real(); 481#endif 482} 483 484static void 485cpu_reset_real() 486{ 487 488#ifdef PC98 489 /* 490 * Attempt to do a CPU reset via CPU reset port. 491 */ 492 disable_intr(); 493 if ((inb(0x35) & 0xa0) != 0xa0) { 494 outb(0x37, 0x0f); /* SHUT0 = 0. */ 495 outb(0x37, 0x0b); /* SHUT1 = 0. */ 496 } 497 outb(0xf0, 0x00); /* Reset. */ 498#else 499 /* 500 * Attempt to do a CPU reset via the keyboard controller, 501 * do not turn of the GateA20, as any machine that fails 502 * to do the reset here would then end up in no man's land. 503 */ 504 505#if !defined(BROKEN_KEYBOARD_RESET) 506 outb(IO_KBD + 4, 0xFE); 507 DELAY(500000); /* wait 0.5 sec to see if that did it */ 508 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 509 DELAY(1000000); /* wait 1 sec for printf to complete */ 510#endif 511#endif /* PC98 */ 512 /* force a shutdown by unmapping entire address space ! */ 513 bzero((caddr_t) PTD, PAGE_SIZE); 514 515 /* "good night, sweet prince .... <THUNK!>" */ 516 invltlb(); 517 /* NOTREACHED */ 518 while(1); 519} 520 521int 522grow_stack(p, sp) 523 struct proc *p; 524 u_int sp; 525{ 526 int rv; 527 528 rv = vm_map_growstack (p, sp); 529 if (rv != KERN_SUCCESS) 530 return (0); 531 532 return (1); 533} 534 535SYSCTL_DECL(_vm_stats_misc); 536 537static int cnt_prezero; 538 539SYSCTL_INT(_vm_stats_misc, OID_AUTO, 540 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, ""); 541 542/* 543 * Implement the pre-zeroed page mechanism. 544 * This routine is called from the idle loop. 545 */ 546 547#define ZIDLE_LO(v) ((v) * 2 / 3) 548#define ZIDLE_HI(v) ((v) * 4 / 5) 549 550int 551vm_page_zero_idle() 552{ 553 static int free_rover; 554 static int zero_state; 555 vm_page_t m; 556 int s; 557 558 /* 559 * Attempt to maintain approximately 1/2 of our free pages in a 560 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid 561 * generally zeroing a page when the system is near steady-state. 562 * Otherwise we might get 'flutter' during disk I/O / IPC or 563 * fast sleeps. We also do not want to be continuously zeroing 564 * pages because doing so may flush our L1 and L2 caches too much. 565 */ 566 567 if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count)) 568 return(0); 569 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 570 return(0); 571 572#ifdef SMP 573 if (try_mplock()) { 574#endif 575 s = splvm(); 576 __asm __volatile("sti" : : : "memory"); 577 zero_state = 0; 578 m = vm_page_list_find(PQ_FREE, free_rover, FALSE); 579 if (m != NULL && (m->flags & PG_ZERO) == 0) { 580 vm_page_queues[m->queue].lcnt--; 581 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq); 582 m->queue = PQ_NONE; 583 splx(s); 584 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 585 (void)splvm(); 586 vm_page_flag_set(m, PG_ZERO); 587 m->queue = PQ_FREE + m->pc; 588 vm_page_queues[m->queue].lcnt++; 589 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m, 590 pageq); 591 ++vm_page_zero_count; 592 ++cnt_prezero; 593 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 594 zero_state = 1; 595 } 596 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK; 597 splx(s); 598 __asm __volatile("cli" : : : "memory"); 599#ifdef SMP 600 rel_mplock(); 601#endif 602 return (1); 603#ifdef SMP 604 } 605#endif 606 /* 607 * We have to enable interrupts for a moment if the try_mplock fails 608 * in order to potentially take an IPI. XXX this should be in 609 * swtch.s 610 */ 611 __asm __volatile("sti; nop; cli" : : : "memory"); 612 return (0); 613} 614 615/* 616 * Software interrupt handler for queued VM system processing. 617 */ 618void 619swi_vm() 620{ 621 if (busdma_swi_pending != 0) 622 busdma_swi(); 623} 624 625/* 626 * Tell whether this address is in some physical memory region. 627 * Currently used by the kernel coredump code in order to avoid 628 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 629 * or other unpredictable behaviour. 630 */ 631 632#include "isa.h" 633 634int 635is_physical_memory(addr) 636 vm_offset_t addr; 637{ 638 639#if NISA > 0 640 /* The ISA ``memory hole''. */ 641 if (addr >= 0xa0000 && addr < 0x100000) 642 return 0; 643#endif 644 645 /* 646 * stuff other tests for known memory-mapped devices (PCI?) 647 * here 648 */ 649 650 return 1; 651} 652