vm_machdep.c revision 60041
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 60041 2000-05-05 09:59:14Z phk $ 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/bio.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#include <sys/unistd.h> 61 62#include <machine/clock.h> 63#include <machine/cpu.h> 64#include <machine/md_var.h> 65#ifdef SMP 66#include <machine/smp.h> 67#endif 68#include <machine/pcb.h> 69#include <machine/pcb_ext.h> 70#include <machine/vm86.h> 71 72#include <vm/vm.h> 73#include <vm/vm_param.h> 74#include <sys/lock.h> 75#include <vm/vm_kern.h> 76#include <vm/vm_page.h> 77#include <vm/vm_map.h> 78#include <vm/vm_extern.h> 79 80#include <sys/user.h> 81 82#ifdef PC98 83#include <pc98/pc98/pc98.h> 84#else 85#include <i386/isa/isa.h> 86#endif 87 88static void cpu_reset_real __P((void)); 89#ifdef SMP 90static void cpu_reset_proxy __P((void)); 91static u_int cpu_reset_proxyid; 92static volatile u_int cpu_reset_proxy_active; 93#endif 94 95/* 96 * quick version of vm_fault 97 */ 98int 99vm_fault_quick(v, prot) 100 caddr_t v; 101 int prot; 102{ 103 int r; 104 105 if (prot & VM_PROT_WRITE) 106 r = subyte(v, fubyte(v)); 107 else 108 r = fubyte(v); 109 return(r); 110} 111 112/* 113 * Finish a fork operation, with process p2 nearly set up. 114 * Copy and update the pcb, set up the stack so that the child 115 * ready to run and return to user mode. 116 */ 117void 118cpu_fork(p1, p2, flags) 119 register struct proc *p1, *p2; 120 int flags; 121{ 122 struct pcb *pcb2; 123 124 if ((flags & RFPROC) == 0) { 125#ifdef USER_LDT 126 if ((flags & RFMEM) == 0) { 127 /* unshare user LDT */ 128 struct pcb *pcb1 = &p1->p_addr->u_pcb; 129 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt; 130 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) { 131 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len); 132 user_ldt_free(pcb1); 133 pcb1->pcb_ldt = pcb_ldt; 134 set_user_ldt(pcb1); 135 } 136 } 137#endif 138 return; 139 } 140 141#if NNPX > 0 142 /* Ensure that p1's pcb is up to date. */ 143 if (npxproc == p1) 144 npxsave(&p1->p_addr->u_pcb.pcb_savefpu); 145#endif 146 147 /* Copy p1's pcb. */ 148 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 149 pcb2 = &p2->p_addr->u_pcb; 150 151 /* 152 * Create a new fresh stack for the new process. 153 * Copy the trap frame for the return to user mode as if from a 154 * syscall. This copies the user mode register values. 155 */ 156 p2->p_md.md_regs = (struct trapframe *) 157 ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1; 158 *p2->p_md.md_regs = *p1->p_md.md_regs; 159 160 /* 161 * Set registers for trampoline to user mode. Leave space for the 162 * return address on stack. These are the kernel mode register values. 163 */ 164 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir); 165 pcb2->pcb_edi = p2->p_md.md_regs->tf_edi; 166 pcb2->pcb_esi = (int)fork_return; 167 pcb2->pcb_ebp = p2->p_md.md_regs->tf_ebp; 168 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *); 169 pcb2->pcb_ebx = (int)p2; 170 pcb2->pcb_eip = (int)fork_trampoline; 171 /* 172 * pcb2->pcb_ldt: duplicated below, if necessary. 173 * pcb2->pcb_savefpu: cloned above. 174 * pcb2->pcb_flags: cloned above (always 0 here?). 175 * pcb2->pcb_onfault: cloned above (always NULL here?). 176 */ 177 178#ifdef SMP 179 pcb2->pcb_mpnest = 1; 180#endif 181 /* 182 * XXX don't copy the i/o pages. this should probably be fixed. 183 */ 184 pcb2->pcb_ext = 0; 185 186#ifdef USER_LDT 187 /* Copy the LDT, if necessary. */ 188 if (pcb2->pcb_ldt != 0) { 189 if (flags & RFMEM) { 190 pcb2->pcb_ldt->ldt_refcnt++; 191 } else { 192 pcb2->pcb_ldt = user_ldt_alloc(pcb2, 193 pcb2->pcb_ldt->ldt_len); 194 } 195 } 196#endif 197 198 /* 199 * Now, cpu_switch() can schedule the new process. 200 * pcb_esp is loaded pointing to the cpu_switch() stack frame 201 * containing the return address when exiting cpu_switch. 202 * This will normally be to proc_trampoline(), which will have 203 * %ebx loaded with the new proc's pointer. proc_trampoline() 204 * will set up a stack to call fork_return(p, frame); to complete 205 * the return to user-mode. 206 */ 207} 208 209/* 210 * Intercept the return address from a freshly forked process that has NOT 211 * been scheduled yet. 212 * 213 * This is needed to make kernel threads stay in kernel mode. 214 */ 215void 216cpu_set_fork_handler(p, func, arg) 217 struct proc *p; 218 void (*func) __P((void *)); 219 void *arg; 220{ 221 /* 222 * Note that the trap frame follows the args, so the function 223 * is really called like this: func(arg, frame); 224 */ 225 p->p_addr->u_pcb.pcb_esi = (int) func; /* function */ 226 p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */ 227} 228 229void 230cpu_exit(p) 231 register struct proc *p; 232{ 233 struct pcb *pcb = &p->p_addr->u_pcb; 234 235#if NNPX > 0 236 npxexit(p); 237#endif /* NNPX */ 238 if (pcb->pcb_ext != 0) { 239 /* 240 * XXX do we need to move the TSS off the allocated pages 241 * before freeing them? (not done here) 242 */ 243 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 244 ctob(IOPAGES + 1)); 245 pcb->pcb_ext = 0; 246 } 247#ifdef USER_LDT 248 user_ldt_free(pcb); 249#endif 250 if (pcb->pcb_flags & PCB_DBREGS) { 251 /* 252 * disable all hardware breakpoints 253 */ 254 reset_dbregs(); 255 pcb->pcb_flags &= ~PCB_DBREGS; 256 } 257 cnt.v_swtch++; 258 cpu_switch(p); 259 panic("cpu_exit"); 260} 261 262void 263cpu_wait(p) 264 struct proc *p; 265{ 266 /* drop per-process resources */ 267 pmap_dispose_proc(p); 268 269 /* and clean-out the vmspace */ 270 vmspace_free(p->p_vmspace); 271} 272 273/* 274 * Dump the machine specific header information at the start of a core dump. 275 */ 276int 277cpu_coredump(p, vp, cred) 278 struct proc *p; 279 struct vnode *vp; 280 struct ucred *cred; 281{ 282 int error; 283 caddr_t tempuser; 284 285 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK); 286 if (!tempuser) 287 return EINVAL; 288 289 bzero(tempuser, ctob(UPAGES)); 290 bcopy(p->p_addr, tempuser, sizeof(struct user)); 291 bcopy(p->p_md.md_regs, 292 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr), 293 sizeof(struct trapframe)); 294 295 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, 296 ctob(UPAGES), 297 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, 298 cred, (int *)NULL, p); 299 300 free(tempuser, M_TEMP); 301 302 return error; 303} 304 305#ifdef notyet 306static void 307setredzone(pte, vaddr) 308 u_short *pte; 309 caddr_t vaddr; 310{ 311/* eventually do this by setting up an expand-down stack segment 312 for ss0: selector, allowing stack access down to top of u. 313 this means though that protection violations need to be handled 314 thru a double fault exception that must do an integral task 315 switch to a known good context, within which a dump can be 316 taken. a sensible scheme might be to save the initial context 317 used by sched (that has physical memory mapped 1:1 at bottom) 318 and take the dump while still in mapped mode */ 319} 320#endif 321 322/* 323 * Convert kernel VA to physical address 324 */ 325u_long 326kvtop(void *addr) 327{ 328 vm_offset_t va; 329 330 va = pmap_kextract((vm_offset_t)addr); 331 if (va == 0) 332 panic("kvtop: zero page frame"); 333 return((int)va); 334} 335 336/* 337 * Map an IO request into kernel virtual address space. 338 * 339 * All requests are (re)mapped into kernel VA space. 340 * Notice that we use b_bufsize for the size of the buffer 341 * to be mapped. b_bcount might be modified by the driver. 342 */ 343void 344vmapbuf(bp) 345 register struct buf *bp; 346{ 347 register caddr_t addr, v, kva; 348 vm_offset_t pa; 349 350 if ((bp->b_flags & B_PHYS) == 0) 351 panic("vmapbuf"); 352 353 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 354 addr < bp->b_data + bp->b_bufsize; 355 addr += PAGE_SIZE, v += PAGE_SIZE) { 356 /* 357 * Do the vm_fault if needed; do the copy-on-write thing 358 * when reading stuff off device into memory. 359 */ 360 vm_fault_quick(addr, 361 (bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 362 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 363 if (pa == 0) 364 panic("vmapbuf: page not present"); 365 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 366 pmap_kenter((vm_offset_t) v, pa); 367 } 368 369 kva = bp->b_saveaddr; 370 bp->b_saveaddr = bp->b_data; 371 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 372} 373 374/* 375 * Free the io map PTEs associated with this IO operation. 376 * We also invalidate the TLB entries and restore the original b_addr. 377 */ 378void 379vunmapbuf(bp) 380 register struct buf *bp; 381{ 382 register caddr_t addr; 383 vm_offset_t pa; 384 385 if ((bp->b_flags & B_PHYS) == 0) 386 panic("vunmapbuf"); 387 388 for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 389 addr < bp->b_data + bp->b_bufsize; 390 addr += PAGE_SIZE) { 391 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 392 pmap_kremove((vm_offset_t) addr); 393 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 394 } 395 396 bp->b_data = bp->b_saveaddr; 397} 398 399/* 400 * Force reset the processor by invalidating the entire address space! 401 */ 402 403#ifdef SMP 404static void 405cpu_reset_proxy() 406{ 407 u_int saved_mp_lock; 408 409 cpu_reset_proxy_active = 1; 410 while (cpu_reset_proxy_active == 1) 411 ; /* Wait for other cpu to disable interupts */ 412 saved_mp_lock = mp_lock; 413 mp_lock = 1; 414 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n"); 415 cpu_reset_proxy_active = 3; 416 while (cpu_reset_proxy_active == 3) 417 ; /* Wait for other cpu to enable interrupts */ 418 stop_cpus((1<<cpu_reset_proxyid)); 419 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 420 DELAY(1000000); 421 cpu_reset_real(); 422} 423#endif 424 425void 426cpu_reset() 427{ 428#ifdef SMP 429 if (smp_active == 0) { 430 cpu_reset_real(); 431 /* NOTREACHED */ 432 } else { 433 434 u_int map; 435 int cnt; 436 printf("cpu_reset called on cpu#%d\n",cpuid); 437 438 map = other_cpus & ~ stopped_cpus; 439 440 if (map != 0) { 441 printf("cpu_reset: Stopping other CPUs\n"); 442 stop_cpus(map); /* Stop all other CPUs */ 443 } 444 445 if (cpuid == 0) { 446 DELAY(1000000); 447 cpu_reset_real(); 448 /* NOTREACHED */ 449 } else { 450 /* We are not BSP (CPU #0) */ 451 452 cpu_reset_proxyid = cpuid; 453 cpustop_restartfunc = cpu_reset_proxy; 454 printf("cpu_reset: Restarting BSP\n"); 455 started_cpus = (1<<0); /* Restart CPU #0 */ 456 457 cnt = 0; 458 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 459 cnt++; /* Wait for BSP to announce restart */ 460 if (cpu_reset_proxy_active == 0) 461 printf("cpu_reset: Failed to restart BSP\n"); 462 __asm __volatile("cli" : : : "memory"); 463 cpu_reset_proxy_active = 2; 464 cnt = 0; 465 while (cpu_reset_proxy_active == 2 && cnt < 10000000) 466 cnt++; /* Do nothing */ 467 if (cpu_reset_proxy_active == 2) { 468 printf("cpu_reset: BSP did not grab mp lock\n"); 469 cpu_reset_real(); /* XXX: Bogus ? */ 470 } 471 cpu_reset_proxy_active = 4; 472 __asm __volatile("sti" : : : "memory"); 473 while (1); 474 /* NOTREACHED */ 475 } 476 } 477#else 478 cpu_reset_real(); 479#endif 480} 481 482static void 483cpu_reset_real() 484{ 485 486#ifdef PC98 487 /* 488 * Attempt to do a CPU reset via CPU reset port. 489 */ 490 disable_intr(); 491 if ((inb(0x35) & 0xa0) != 0xa0) { 492 outb(0x37, 0x0f); /* SHUT0 = 0. */ 493 outb(0x37, 0x0b); /* SHUT1 = 0. */ 494 } 495 outb(0xf0, 0x00); /* Reset. */ 496#else 497 /* 498 * Attempt to do a CPU reset via the keyboard controller, 499 * do not turn of the GateA20, as any machine that fails 500 * to do the reset here would then end up in no man's land. 501 */ 502 503#if !defined(BROKEN_KEYBOARD_RESET) 504 outb(IO_KBD + 4, 0xFE); 505 DELAY(500000); /* wait 0.5 sec to see if that did it */ 506 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 507 DELAY(1000000); /* wait 1 sec for printf to complete */ 508#endif 509#endif /* PC98 */ 510 /* force a shutdown by unmapping entire address space ! */ 511 bzero((caddr_t) PTD, PAGE_SIZE); 512 513 /* "good night, sweet prince .... <THUNK!>" */ 514 invltlb(); 515 /* NOTREACHED */ 516 while(1); 517} 518 519int 520grow_stack(p, sp) 521 struct proc *p; 522 u_int sp; 523{ 524 int rv; 525 526 rv = vm_map_growstack (p, sp); 527 if (rv != KERN_SUCCESS) 528 return (0); 529 530 return (1); 531} 532 533SYSCTL_DECL(_vm_stats_misc); 534 535static int cnt_prezero; 536 537SYSCTL_INT(_vm_stats_misc, OID_AUTO, 538 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, ""); 539 540/* 541 * Implement the pre-zeroed page mechanism. 542 * This routine is called from the idle loop. 543 */ 544 545#define ZIDLE_LO(v) ((v) * 2 / 3) 546#define ZIDLE_HI(v) ((v) * 4 / 5) 547 548int 549vm_page_zero_idle() 550{ 551 static int free_rover; 552 static int zero_state; 553 vm_page_t m; 554 int s; 555 556 /* 557 * Attempt to maintain approximately 1/2 of our free pages in a 558 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid 559 * generally zeroing a page when the system is near steady-state. 560 * Otherwise we might get 'flutter' during disk I/O / IPC or 561 * fast sleeps. We also do not want to be continuously zeroing 562 * pages because doing so may flush our L1 and L2 caches too much. 563 */ 564 565 if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count)) 566 return(0); 567 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 568 return(0); 569 570#ifdef SMP 571 if (try_mplock()) { 572#endif 573 s = splvm(); 574 __asm __volatile("sti" : : : "memory"); 575 zero_state = 0; 576 m = vm_page_list_find(PQ_FREE, free_rover, FALSE); 577 if (m != NULL && (m->flags & PG_ZERO) == 0) { 578 vm_page_queues[m->queue].lcnt--; 579 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq); 580 m->queue = PQ_NONE; 581 splx(s); 582 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 583 (void)splvm(); 584 vm_page_flag_set(m, PG_ZERO); 585 m->queue = PQ_FREE + m->pc; 586 vm_page_queues[m->queue].lcnt++; 587 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m, 588 pageq); 589 ++vm_page_zero_count; 590 ++cnt_prezero; 591 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 592 zero_state = 1; 593 } 594 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK; 595 splx(s); 596 __asm __volatile("cli" : : : "memory"); 597#ifdef SMP 598 rel_mplock(); 599#endif 600 return (1); 601#ifdef SMP 602 } 603#endif 604 /* 605 * We have to enable interrupts for a moment if the try_mplock fails 606 * in order to potentially take an IPI. XXX this should be in 607 * swtch.s 608 */ 609 __asm __volatile("sti; nop; cli" : : : "memory"); 610 return (0); 611} 612 613/* 614 * Software interrupt handler for queued VM system processing. 615 */ 616void 617swi_vm() 618{ 619 if (busdma_swi_pending != 0) 620 busdma_swi(); 621} 622 623/* 624 * Tell whether this address is in some physical memory region. 625 * Currently used by the kernel coredump code in order to avoid 626 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 627 * or other unpredictable behaviour. 628 */ 629 630#include "isa.h" 631 632int 633is_physical_memory(addr) 634 vm_offset_t addr; 635{ 636 637#if NISA > 0 638 /* The ISA ``memory hole''. */ 639 if (addr >= 0xa0000 && addr < 0x100000) 640 return 0; 641#endif 642 643 /* 644 * stuff other tests for known memory-mapped devices (PCI?) 645 * here 646 */ 647 648 return 1; 649} 650