vm_machdep.c revision 54188
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 54188 1999-12-06 04:53:08Z luoqi $ 42 */ 43 44#include "npx.h" 45#include "opt_user_ldt.h" 46#ifdef PC98 47#include "opt_pc98.h" 48#endif 49 50#include <sys/param.h> 51#include <sys/systm.h> 52#include <sys/malloc.h> 53#include <sys/proc.h> 54#include <sys/buf.h> 55#include <sys/vnode.h> 56#include <sys/vmmeter.h> 57#include <sys/kernel.h> 58#include <sys/sysctl.h> 59#include <sys/unistd.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#include <machine/pcb.h> 68#include <machine/pcb_ext.h> 69#include <machine/vm86.h> 70 71#include <vm/vm.h> 72#include <vm/vm_param.h> 73#include <sys/lock.h> 74#include <vm/vm_kern.h> 75#include <vm/vm_page.h> 76#include <vm/vm_map.h> 77#include <vm/vm_extern.h> 78 79#include <sys/user.h> 80 81#ifdef PC98 82#include <pc98/pc98/pc98.h> 83#else 84#include <i386/isa/isa.h> 85#endif 86 87static void cpu_reset_real __P((void)); 88#ifdef SMP 89static void cpu_reset_proxy __P((void)); 90static u_int cpu_reset_proxyid; 91static volatile u_int cpu_reset_proxy_active; 92#endif 93 94/* 95 * quick version of vm_fault 96 */ 97int 98vm_fault_quick(v, prot) 99 caddr_t v; 100 int prot; 101{ 102 int r; 103 104 if (prot & VM_PROT_WRITE) 105 r = subyte(v, fubyte(v)); 106 else 107 r = fubyte(v); 108 return(r); 109} 110 111/* 112 * Finish a fork operation, with process p2 nearly set up. 113 * Copy and update the pcb, set up the stack so that the child 114 * ready to run and return to user mode. 115 */ 116void 117cpu_fork(p1, p2, flags) 118 register struct proc *p1, *p2; 119 int flags; 120{ 121 struct pcb *pcb2; 122 123 if ((flags & RFPROC) == 0) { 124#ifdef USER_LDT 125 if ((flags & RFMEM) == 0) { 126 /* unshare user LDT */ 127 struct pcb *pcb1 = &p1->p_addr->u_pcb; 128 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt; 129 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) { 130 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len); 131 user_ldt_free(pcb1); 132 pcb1->pcb_ldt = pcb_ldt; 133 set_user_ldt(pcb1); 134 } 135 } 136#endif 137 return; 138 } 139 140#if NNPX > 0 141 /* Ensure that p1's pcb is up to date. */ 142 if (npxproc == p1) 143 npxsave(&p1->p_addr->u_pcb.pcb_savefpu); 144#endif 145 146 /* Copy p1's pcb. */ 147 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 148 pcb2 = &p2->p_addr->u_pcb; 149 150 /* 151 * Create a new fresh stack for the new process. 152 * Copy the trap frame for the return to user mode as if from a 153 * syscall. This copies the user mode register values. 154 */ 155 p2->p_md.md_regs = (struct trapframe *) 156 ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1; 157 *p2->p_md.md_regs = *p1->p_md.md_regs; 158 159 /* 160 * Set registers for trampoline to user mode. Leave space for the 161 * return address on stack. These are the kernel mode register values. 162 */ 163 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir); 164 pcb2->pcb_edi = p2->p_md.md_regs->tf_edi; 165 pcb2->pcb_esi = (int)fork_return; 166 pcb2->pcb_ebp = p2->p_md.md_regs->tf_ebp; 167 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *); 168 pcb2->pcb_ebx = (int)p2; 169 pcb2->pcb_eip = (int)fork_trampoline; 170 /* 171 * pcb2->pcb_ldt: duplicated below, if necessary. 172 * pcb2->pcb_savefpu: cloned above. 173 * pcb2->pcb_flags: cloned above (always 0 here?). 174 * pcb2->pcb_onfault: cloned above (always NULL here?). 175 */ 176 177#ifdef SMP 178 pcb2->pcb_mpnest = 1; 179#endif 180 /* 181 * XXX don't copy the i/o pages. this should probably be fixed. 182 */ 183 pcb2->pcb_ext = 0; 184 185#ifdef USER_LDT 186 /* Copy the LDT, if necessary. */ 187 if (pcb2->pcb_ldt != 0) { 188 if (flags & RFMEM) { 189 pcb2->pcb_ldt->ldt_refcnt++; 190 } else { 191 pcb2->pcb_ldt = user_ldt_alloc(pcb2, 192 pcb2->pcb_ldt->ldt_len); 193 } 194 } 195#endif 196 197 /* 198 * Now, cpu_switch() can schedule the new process. 199 * pcb_esp is loaded pointing to the cpu_switch() stack frame 200 * containing the return address when exiting cpu_switch. 201 * This will normally be to proc_trampoline(), which will have 202 * %ebx loaded with the new proc's pointer. proc_trampoline() 203 * will set up a stack to call fork_return(p, frame); to complete 204 * the return to user-mode. 205 */ 206} 207 208/* 209 * Intercept the return address from a freshly forked process that has NOT 210 * been scheduled yet. 211 * 212 * This is needed to make kernel threads stay in kernel mode. 213 */ 214void 215cpu_set_fork_handler(p, func, arg) 216 struct proc *p; 217 void (*func) __P((void *)); 218 void *arg; 219{ 220 /* 221 * Note that the trap frame follows the args, so the function 222 * is really called like this: func(arg, frame); 223 */ 224 p->p_addr->u_pcb.pcb_esi = (int) func; /* function */ 225 p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */ 226} 227 228void 229cpu_exit(p) 230 register struct proc *p; 231{ 232 struct pcb *pcb = &p->p_addr->u_pcb; 233 234#if NNPX > 0 235 npxexit(p); 236#endif /* NNPX */ 237 if (pcb->pcb_ext != 0) { 238 /* 239 * XXX do we need to move the TSS off the allocated pages 240 * before freeing them? (not done here) 241 */ 242 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 243 ctob(IOPAGES + 1)); 244 pcb->pcb_ext = 0; 245 } 246#ifdef USER_LDT 247 user_ldt_free(pcb); 248#endif 249 cnt.v_swtch++; 250 cpu_switch(p); 251 panic("cpu_exit"); 252} 253 254void 255cpu_wait(p) 256 struct proc *p; 257{ 258 /* drop per-process resources */ 259 pmap_dispose_proc(p); 260 261 /* and clean-out the vmspace */ 262 vmspace_free(p->p_vmspace); 263} 264 265/* 266 * Dump the machine specific header information at the start of a core dump. 267 */ 268int 269cpu_coredump(p, vp, cred) 270 struct proc *p; 271 struct vnode *vp; 272 struct ucred *cred; 273{ 274 int error; 275 caddr_t tempuser; 276 277 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK); 278 if (!tempuser) 279 return EINVAL; 280 281 bzero(tempuser, ctob(UPAGES)); 282 bcopy(p->p_addr, tempuser, sizeof(struct user)); 283 bcopy(p->p_md.md_regs, 284 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr), 285 sizeof(struct trapframe)); 286 287 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, 288 ctob(UPAGES), 289 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, 290 cred, (int *)NULL, p); 291 292 free(tempuser, M_TEMP); 293 294 return error; 295} 296 297#ifdef notyet 298static void 299setredzone(pte, vaddr) 300 u_short *pte; 301 caddr_t vaddr; 302{ 303/* eventually do this by setting up an expand-down stack segment 304 for ss0: selector, allowing stack access down to top of u. 305 this means though that protection violations need to be handled 306 thru a double fault exception that must do an integral task 307 switch to a known good context, within which a dump can be 308 taken. a sensible scheme might be to save the initial context 309 used by sched (that has physical memory mapped 1:1 at bottom) 310 and take the dump while still in mapped mode */ 311} 312#endif 313 314/* 315 * Convert kernel VA to physical address 316 */ 317u_long 318kvtop(void *addr) 319{ 320 vm_offset_t va; 321 322 va = pmap_kextract((vm_offset_t)addr); 323 if (va == 0) 324 panic("kvtop: zero page frame"); 325 return((int)va); 326} 327 328/* 329 * Map an IO request into kernel virtual address space. 330 * 331 * All requests are (re)mapped into kernel VA space. 332 * Notice that we use b_bufsize for the size of the buffer 333 * to be mapped. b_bcount might be modified by the driver. 334 */ 335void 336vmapbuf(bp) 337 register struct buf *bp; 338{ 339 register caddr_t addr, v, kva; 340 vm_offset_t pa; 341 342 if ((bp->b_flags & B_PHYS) == 0) 343 panic("vmapbuf"); 344 345 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 346 addr < bp->b_data + bp->b_bufsize; 347 addr += PAGE_SIZE, v += PAGE_SIZE) { 348 /* 349 * Do the vm_fault if needed; do the copy-on-write thing 350 * when reading stuff off device into memory. 351 */ 352 vm_fault_quick(addr, 353 (bp->b_flags&B_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 354 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 355 if (pa == 0) 356 panic("vmapbuf: page not present"); 357 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 358 pmap_kenter((vm_offset_t) v, pa); 359 } 360 361 kva = bp->b_saveaddr; 362 bp->b_saveaddr = bp->b_data; 363 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 364} 365 366/* 367 * Free the io map PTEs associated with this IO operation. 368 * We also invalidate the TLB entries and restore the original b_addr. 369 */ 370void 371vunmapbuf(bp) 372 register struct buf *bp; 373{ 374 register caddr_t addr; 375 vm_offset_t pa; 376 377 if ((bp->b_flags & B_PHYS) == 0) 378 panic("vunmapbuf"); 379 380 for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 381 addr < bp->b_data + bp->b_bufsize; 382 addr += PAGE_SIZE) { 383 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 384 pmap_kremove((vm_offset_t) addr); 385 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 386 } 387 388 bp->b_data = bp->b_saveaddr; 389} 390 391/* 392 * Force reset the processor by invalidating the entire address space! 393 */ 394 395#ifdef SMP 396static void 397cpu_reset_proxy() 398{ 399 u_int saved_mp_lock; 400 401 cpu_reset_proxy_active = 1; 402 while (cpu_reset_proxy_active == 1) 403 ; /* Wait for other cpu to disable interupts */ 404 saved_mp_lock = mp_lock; 405 mp_lock = 1; 406 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n"); 407 cpu_reset_proxy_active = 3; 408 while (cpu_reset_proxy_active == 3) 409 ; /* Wait for other cpu to enable interrupts */ 410 stop_cpus((1<<cpu_reset_proxyid)); 411 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 412 DELAY(1000000); 413 cpu_reset_real(); 414} 415#endif 416 417void 418cpu_reset() 419{ 420#ifdef SMP 421 if (smp_active == 0) { 422 cpu_reset_real(); 423 /* NOTREACHED */ 424 } else { 425 426 u_int map; 427 int cnt; 428 printf("cpu_reset called on cpu#%d\n",cpuid); 429 430 map = other_cpus & ~ stopped_cpus; 431 432 if (map != 0) { 433 printf("cpu_reset: Stopping other CPUs\n"); 434 stop_cpus(map); /* Stop all other CPUs */ 435 } 436 437 if (cpuid == 0) { 438 DELAY(1000000); 439 cpu_reset_real(); 440 /* NOTREACHED */ 441 } else { 442 /* We are not BSP (CPU #0) */ 443 444 cpu_reset_proxyid = cpuid; 445 cpustop_restartfunc = cpu_reset_proxy; 446 printf("cpu_reset: Restarting BSP\n"); 447 started_cpus = (1<<0); /* Restart CPU #0 */ 448 449 cnt = 0; 450 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 451 cnt++; /* Wait for BSP to announce restart */ 452 if (cpu_reset_proxy_active == 0) 453 printf("cpu_reset: Failed to restart BSP\n"); 454 __asm __volatile("cli" : : : "memory"); 455 cpu_reset_proxy_active = 2; 456 cnt = 0; 457 while (cpu_reset_proxy_active == 2 && cnt < 10000000) 458 cnt++; /* Do nothing */ 459 if (cpu_reset_proxy_active == 2) { 460 printf("cpu_reset: BSP did not grab mp lock\n"); 461 cpu_reset_real(); /* XXX: Bogus ? */ 462 } 463 cpu_reset_proxy_active = 4; 464 __asm __volatile("sti" : : : "memory"); 465 while (1); 466 /* NOTREACHED */ 467 } 468 } 469#else 470 cpu_reset_real(); 471#endif 472} 473 474static void 475cpu_reset_real() 476{ 477 478#ifdef PC98 479 /* 480 * Attempt to do a CPU reset via CPU reset port. 481 */ 482 disable_intr(); 483 if ((inb(0x35) & 0xa0) != 0xa0) { 484 outb(0x37, 0x0f); /* SHUT0 = 0. */ 485 outb(0x37, 0x0b); /* SHUT1 = 0. */ 486 } 487 outb(0xf0, 0x00); /* Reset. */ 488#else 489 /* 490 * Attempt to do a CPU reset via the keyboard controller, 491 * do not turn of the GateA20, as any machine that fails 492 * to do the reset here would then end up in no man's land. 493 */ 494 495#if !defined(BROKEN_KEYBOARD_RESET) 496 outb(IO_KBD + 4, 0xFE); 497 DELAY(500000); /* wait 0.5 sec to see if that did it */ 498 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 499 DELAY(1000000); /* wait 1 sec for printf to complete */ 500#endif 501#endif /* PC98 */ 502 /* force a shutdown by unmapping entire address space ! */ 503 bzero((caddr_t) PTD, PAGE_SIZE); 504 505 /* "good night, sweet prince .... <THUNK!>" */ 506 invltlb(); 507 /* NOTREACHED */ 508 while(1); 509} 510 511int 512grow_stack(p, sp) 513 struct proc *p; 514 u_int sp; 515{ 516 int rv; 517 518 rv = vm_map_growstack (p, sp); 519 if (rv != KERN_SUCCESS) 520 return (0); 521 522 return (1); 523} 524 525SYSCTL_DECL(_vm_stats_misc); 526 527static int cnt_prezero; 528 529SYSCTL_INT(_vm_stats_misc, OID_AUTO, 530 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, ""); 531 532/* 533 * Implement the pre-zeroed page mechanism. 534 * This routine is called from the idle loop. 535 */ 536 537#define ZIDLE_LO(v) ((v) * 2 / 3) 538#define ZIDLE_HI(v) ((v) * 4 / 5) 539 540int 541vm_page_zero_idle() 542{ 543 static int free_rover; 544 static int zero_state; 545 vm_page_t m; 546 int s; 547 548 /* 549 * Attempt to maintain approximately 1/2 of our free pages in a 550 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid 551 * generally zeroing a page when the system is near steady-state. 552 * Otherwise we might get 'flutter' during disk I/O / IPC or 553 * fast sleeps. We also do not want to be continuously zeroing 554 * pages because doing so may flush our L1 and L2 caches too much. 555 */ 556 557 if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count)) 558 return(0); 559 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 560 return(0); 561 562#ifdef SMP 563 if (try_mplock()) { 564#endif 565 s = splvm(); 566 __asm __volatile("sti" : : : "memory"); 567 zero_state = 0; 568 m = vm_page_list_find(PQ_FREE, free_rover, FALSE); 569 if (m != NULL && (m->flags & PG_ZERO) == 0) { 570 vm_page_queues[m->queue].lcnt--; 571 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq); 572 m->queue = PQ_NONE; 573 splx(s); 574#if 0 575 rel_mplock(); 576#endif 577 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 578#if 0 579 get_mplock(); 580#endif 581 (void)splvm(); 582 vm_page_flag_set(m, PG_ZERO); 583 m->queue = PQ_FREE + m->pc; 584 vm_page_queues[m->queue].lcnt++; 585 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m, 586 pageq); 587 ++vm_page_zero_count; 588 ++cnt_prezero; 589 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 590 zero_state = 1; 591 } 592 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK; 593 splx(s); 594 __asm __volatile("cli" : : : "memory"); 595#ifdef SMP 596 rel_mplock(); 597#endif 598 return (1); 599#ifdef SMP 600 } 601#endif 602 return (0); 603} 604 605/* 606 * Software interrupt handler for queued VM system processing. 607 */ 608void 609swi_vm() 610{ 611 if (busdma_swi_pending != 0) 612 busdma_swi(); 613} 614 615/* 616 * Tell whether this address is in some physical memory region. 617 * Currently used by the kernel coredump code in order to avoid 618 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 619 * or other unpredictable behaviour. 620 */ 621 622#include "isa.h" 623 624int 625is_physical_memory(addr) 626 vm_offset_t addr; 627{ 628 629#if NISA > 0 630 /* The ISA ``memory hole''. */ 631 if (addr >= 0xa0000 && addr < 0x100000) 632 return 0; 633#endif 634 635 /* 636 * stuff other tests for known memory-mapped devices (PCI?) 637 * here 638 */ 639 640 return 1; 641} 642