vm_machdep.c revision 52121
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 52121 1999-10-11 14:50:03Z peter $ 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/proc.h> 53#include <sys/malloc.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 60#include <machine/clock.h> 61#include <machine/cpu.h> 62#include <machine/md_var.h> 63#ifdef SMP 64#include <machine/smp.h> 65#endif 66 67#include <vm/vm.h> 68#include <vm/vm_param.h> 69#include <vm/vm_prot.h> 70#include <sys/lock.h> 71#include <vm/vm_kern.h> 72#include <vm/vm_page.h> 73#include <vm/vm_map.h> 74#include <vm/vm_extern.h> 75 76#include <sys/user.h> 77 78#ifdef PC98 79#include <pc98/pc98/pc98.h> 80#else 81#include <i386/isa/isa.h> 82#endif 83 84static void cpu_reset_real __P((void)); 85#ifdef SMP 86static void cpu_reset_proxy __P((void)); 87static u_int cpu_reset_proxyid; 88static volatile u_int cpu_reset_proxy_active; 89#endif 90 91/* 92 * quick version of vm_fault 93 */ 94int 95vm_fault_quick(v, prot) 96 caddr_t v; 97 int prot; 98{ 99 int r; 100 101 if (prot & VM_PROT_WRITE) 102 r = subyte(v, fubyte(v)); 103 else 104 r = fubyte(v); 105 return(r); 106} 107 108/* 109 * Finish a fork operation, with process p2 nearly set up. 110 * Copy and update the pcb, set up the stack so that the child 111 * ready to run and return to user mode. 112 */ 113void 114cpu_fork(p1, p2) 115 register struct proc *p1, *p2; 116{ 117 struct pcb *pcb2 = &p2->p_addr->u_pcb; 118 119#if NNPX > 0 120 /* Ensure that p1's pcb is up to date. */ 121 if (npxproc == p1) 122 npxsave(&p1->p_addr->u_pcb.pcb_savefpu); 123#endif 124 125 /* Copy p1's pcb. */ 126 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 127 128 /* 129 * Create a new fresh stack for the new process. 130 * Copy the trap frame for the return to user mode as if from a 131 * syscall. This copies the user mode register values. 132 */ 133 p2->p_md.md_regs = (struct trapframe *) 134 ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1; 135 *p2->p_md.md_regs = *p1->p_md.md_regs; 136 137 /* 138 * Set registers for trampoline to user mode. Leave space for the 139 * return address on stack. These are the kernel mode register values. 140 */ 141 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir); 142 pcb2->pcb_edi = p2->p_md.md_regs->tf_edi; 143 pcb2->pcb_esi = (int)fork_return; 144 pcb2->pcb_ebp = p2->p_md.md_regs->tf_ebp; 145 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *); 146 pcb2->pcb_ebx = (int)p2; 147 pcb2->pcb_eip = (int)fork_trampoline; 148 /* 149 * pcb2->pcb_ldt: duplicated below, if necessary. 150 * pcb2->pcb_ldt_len: cloned above. 151 * pcb2->pcb_savefpu: cloned above. 152 * pcb2->pcb_flags: cloned above (always 0 here?). 153 * pcb2->pcb_onfault: cloned above (always NULL here?). 154 */ 155 156#ifdef SMP 157 pcb2->pcb_mpnest = 1; 158#endif 159 /* 160 * XXX don't copy the i/o pages. this should probably be fixed. 161 */ 162 pcb2->pcb_ext = 0; 163 164#ifdef USER_LDT 165 /* Copy the LDT, if necessary. */ 166 if (pcb2->pcb_ldt != 0) { 167 union descriptor *new_ldt; 168 size_t len = pcb2->pcb_ldt_len * sizeof(union descriptor); 169 170 new_ldt = (union descriptor *)kmem_alloc(kernel_map, len); 171 bcopy(pcb2->pcb_ldt, new_ldt, len); 172 pcb2->pcb_ldt = (caddr_t)new_ldt; 173 } 174#endif 175 176 /* 177 * Now, cpu_switch() can schedule the new process. 178 * pcb_esp is loaded pointing to the cpu_switch() stack frame 179 * containing the return address when exiting cpu_switch. 180 * This will normally be to proc_trampoline(), which will have 181 * %ebx loaded with the new proc's pointer. proc_trampoline() 182 * will set up a stack to call fork_return(p, frame); to complete 183 * the return to user-mode. 184 */ 185} 186 187/* 188 * Intercept the return address from a freshly forked process that has NOT 189 * been scheduled yet. 190 * 191 * This is needed to make kernel threads stay in kernel mode. 192 */ 193void 194cpu_set_fork_handler(p, func, arg) 195 struct proc *p; 196 void (*func) __P((void *)); 197 void *arg; 198{ 199 /* 200 * Note that the trap frame follows the args, so the function 201 * is really called like this: func(arg, frame); 202 */ 203 p->p_addr->u_pcb.pcb_esi = (int) func; /* function */ 204 p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */ 205} 206 207void 208cpu_exit(p) 209 register struct proc *p; 210{ 211 struct pcb *pcb = &p->p_addr->u_pcb; 212 213#if NNPX > 0 214 npxexit(p); 215#endif /* NNPX */ 216 if (pcb->pcb_ext != 0) { 217 /* 218 * XXX do we need to move the TSS off the allocated pages 219 * before freeing them? (not done here) 220 */ 221 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 222 ctob(IOPAGES + 1)); 223 pcb->pcb_ext = 0; 224 } 225#ifdef USER_LDT 226 if (pcb->pcb_ldt != 0) { 227 if (pcb == curpcb) { 228 lldt(_default_ldt); 229 currentldt = _default_ldt; 230 } 231 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ldt, 232 pcb->pcb_ldt_len * sizeof(union descriptor)); 233 pcb->pcb_ldt_len = (int)pcb->pcb_ldt = 0; 234 } 235#endif 236 cnt.v_swtch++; 237 cpu_switch(p); 238 panic("cpu_exit"); 239} 240 241void 242cpu_wait(p) 243 struct proc *p; 244{ 245 /* drop per-process resources */ 246 pmap_dispose_proc(p); 247 248 /* and clean-out the vmspace */ 249 vmspace_free(p->p_vmspace); 250} 251 252/* 253 * Dump the machine specific header information at the start of a core dump. 254 */ 255int 256cpu_coredump(p, vp, cred) 257 struct proc *p; 258 struct vnode *vp; 259 struct ucred *cred; 260{ 261 int error; 262 caddr_t tempuser; 263 264 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK); 265 if (!tempuser) 266 return EINVAL; 267 268 bzero(tempuser, ctob(UPAGES)); 269 bcopy(p->p_addr, tempuser, sizeof(struct user)); 270 bcopy(p->p_md.md_regs, 271 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr), 272 sizeof(struct trapframe)); 273 274 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, 275 ctob(UPAGES), 276 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, 277 cred, (int *)NULL, p); 278 279 free(tempuser, M_TEMP); 280 281 return error; 282} 283 284#ifdef notyet 285static void 286setredzone(pte, vaddr) 287 u_short *pte; 288 caddr_t vaddr; 289{ 290/* eventually do this by setting up an expand-down stack segment 291 for ss0: selector, allowing stack access down to top of u. 292 this means though that protection violations need to be handled 293 thru a double fault exception that must do an integral task 294 switch to a known good context, within which a dump can be 295 taken. a sensible scheme might be to save the initial context 296 used by sched (that has physical memory mapped 1:1 at bottom) 297 and take the dump while still in mapped mode */ 298} 299#endif 300 301/* 302 * Convert kernel VA to physical address 303 */ 304u_long 305kvtop(void *addr) 306{ 307 vm_offset_t va; 308 309 va = pmap_kextract((vm_offset_t)addr); 310 if (va == 0) 311 panic("kvtop: zero page frame"); 312 return((int)va); 313} 314 315/* 316 * Map an IO request into kernel virtual address space. 317 * 318 * All requests are (re)mapped into kernel VA space. 319 * Notice that we use b_bufsize for the size of the buffer 320 * to be mapped. b_bcount might be modified by the driver. 321 */ 322void 323vmapbuf(bp) 324 register struct buf *bp; 325{ 326 register caddr_t addr, v, kva; 327 vm_offset_t pa; 328 329 if ((bp->b_flags & B_PHYS) == 0) 330 panic("vmapbuf"); 331 332 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 333 addr < bp->b_data + bp->b_bufsize; 334 addr += PAGE_SIZE, v += PAGE_SIZE) { 335 /* 336 * Do the vm_fault if needed; do the copy-on-write thing 337 * when reading stuff off device into memory. 338 */ 339 vm_fault_quick(addr, 340 (bp->b_flags&B_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 341 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 342 if (pa == 0) 343 panic("vmapbuf: page not present"); 344 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 345 pmap_kenter((vm_offset_t) v, pa); 346 } 347 348 kva = bp->b_saveaddr; 349 bp->b_saveaddr = bp->b_data; 350 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 351} 352 353/* 354 * Free the io map PTEs associated with this IO operation. 355 * We also invalidate the TLB entries and restore the original b_addr. 356 */ 357void 358vunmapbuf(bp) 359 register struct buf *bp; 360{ 361 register caddr_t addr; 362 vm_offset_t pa; 363 364 if ((bp->b_flags & B_PHYS) == 0) 365 panic("vunmapbuf"); 366 367 for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 368 addr < bp->b_data + bp->b_bufsize; 369 addr += PAGE_SIZE) { 370 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 371 pmap_kremove((vm_offset_t) addr); 372 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 373 } 374 375 bp->b_data = bp->b_saveaddr; 376} 377 378/* 379 * Force reset the processor by invalidating the entire address space! 380 */ 381 382#ifdef SMP 383static void 384cpu_reset_proxy() 385{ 386 u_int saved_mp_lock; 387 388 cpu_reset_proxy_active = 1; 389 while (cpu_reset_proxy_active == 1) 390 ; /* Wait for other cpu to disable interupts */ 391 saved_mp_lock = mp_lock; 392 mp_lock = 1; 393 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n"); 394 cpu_reset_proxy_active = 3; 395 while (cpu_reset_proxy_active == 3) 396 ; /* Wait for other cpu to enable interrupts */ 397 stop_cpus((1<<cpu_reset_proxyid)); 398 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 399 DELAY(1000000); 400 cpu_reset_real(); 401} 402#endif 403 404void 405cpu_reset() 406{ 407#ifdef SMP 408 if (smp_active == 0) { 409 cpu_reset_real(); 410 /* NOTREACHED */ 411 } else { 412 413 u_int map; 414 int cnt; 415 printf("cpu_reset called on cpu#%d\n",cpuid); 416 417 map = other_cpus & ~ stopped_cpus; 418 419 if (map != 0) { 420 printf("cpu_reset: Stopping other CPUs\n"); 421 stop_cpus(map); /* Stop all other CPUs */ 422 } 423 424 if (cpuid == 0) { 425 DELAY(1000000); 426 cpu_reset_real(); 427 /* NOTREACHED */ 428 } else { 429 /* We are not BSP (CPU #0) */ 430 431 cpu_reset_proxyid = cpuid; 432 cpustop_restartfunc = cpu_reset_proxy; 433 printf("cpu_reset: Restarting BSP\n"); 434 started_cpus = (1<<0); /* Restart CPU #0 */ 435 436 cnt = 0; 437 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 438 cnt++; /* Wait for BSP to announce restart */ 439 if (cpu_reset_proxy_active == 0) 440 printf("cpu_reset: Failed to restart BSP\n"); 441 __asm __volatile("cli" : : : "memory"); 442 cpu_reset_proxy_active = 2; 443 cnt = 0; 444 while (cpu_reset_proxy_active == 2 && cnt < 10000000) 445 cnt++; /* Do nothing */ 446 if (cpu_reset_proxy_active == 2) { 447 printf("cpu_reset: BSP did not grab mp lock\n"); 448 cpu_reset_real(); /* XXX: Bogus ? */ 449 } 450 cpu_reset_proxy_active = 4; 451 __asm __volatile("sti" : : : "memory"); 452 while (1); 453 /* NOTREACHED */ 454 } 455 } 456#else 457 cpu_reset_real(); 458#endif 459} 460 461static void 462cpu_reset_real() 463{ 464 465#ifdef PC98 466 /* 467 * Attempt to do a CPU reset via CPU reset port. 468 */ 469 disable_intr(); 470 if ((inb(0x35) & 0xa0) != 0xa0) { 471 outb(0x37, 0x0f); /* SHUT0 = 0. */ 472 outb(0x37, 0x0b); /* SHUT1 = 0. */ 473 } 474 outb(0xf0, 0x00); /* Reset. */ 475#else 476 /* 477 * Attempt to do a CPU reset via the keyboard controller, 478 * do not turn of the GateA20, as any machine that fails 479 * to do the reset here would then end up in no man's land. 480 */ 481 482#if !defined(BROKEN_KEYBOARD_RESET) 483 outb(IO_KBD + 4, 0xFE); 484 DELAY(500000); /* wait 0.5 sec to see if that did it */ 485 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 486 DELAY(1000000); /* wait 1 sec for printf to complete */ 487#endif 488#endif /* PC98 */ 489 /* force a shutdown by unmapping entire address space ! */ 490 bzero((caddr_t) PTD, PAGE_SIZE); 491 492 /* "good night, sweet prince .... <THUNK!>" */ 493 invltlb(); 494 /* NOTREACHED */ 495 while(1); 496} 497 498int 499grow_stack(p, sp) 500 struct proc *p; 501 u_int sp; 502{ 503 int rv; 504 505 rv = vm_map_growstack (p, sp); 506 if (rv != KERN_SUCCESS) 507 return (0); 508 509 return (1); 510} 511 512SYSCTL_DECL(_vm_stats_misc); 513 514static int cnt_prezero; 515 516SYSCTL_INT(_vm_stats_misc, OID_AUTO, 517 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, ""); 518 519/* 520 * Implement the pre-zeroed page mechanism. 521 * This routine is called from the idle loop. 522 */ 523 524#define ZIDLE_LO(v) ((v) * 2 / 3) 525#define ZIDLE_HI(v) ((v) * 4 / 5) 526 527int 528vm_page_zero_idle() 529{ 530 static int free_rover; 531 static int zero_state; 532 vm_page_t m; 533 int s; 534 535 /* 536 * Attempt to maintain approximately 1/2 of our free pages in a 537 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid 538 * generally zeroing a page when the system is near steady-state. 539 * Otherwise we might get 'flutter' during disk I/O / IPC or 540 * fast sleeps. We also do not want to be continuously zeroing 541 * pages because doing so may flush our L1 and L2 caches too much. 542 */ 543 544 if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count)) 545 return(0); 546 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 547 return(0); 548 549#ifdef SMP 550 if (try_mplock()) { 551#endif 552 s = splvm(); 553 __asm __volatile("sti" : : : "memory"); 554 zero_state = 0; 555 m = vm_page_list_find(PQ_FREE, free_rover, FALSE); 556 if (m != NULL && (m->flags & PG_ZERO) == 0) { 557 vm_page_queues[m->queue].lcnt--; 558 TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq); 559 m->queue = PQ_NONE; 560 splx(s); 561#if 0 562 rel_mplock(); 563#endif 564 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 565#if 0 566 get_mplock(); 567#endif 568 (void)splvm(); 569 vm_page_flag_set(m, PG_ZERO); 570 m->queue = PQ_FREE + m->pc; 571 vm_page_queues[m->queue].lcnt++; 572 TAILQ_INSERT_TAIL(vm_page_queues[m->queue].pl, m, 573 pageq); 574 ++vm_page_zero_count; 575 ++cnt_prezero; 576 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 577 zero_state = 1; 578 } 579 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK; 580 splx(s); 581 __asm __volatile("cli" : : : "memory"); 582#ifdef SMP 583 rel_mplock(); 584#endif 585 return (1); 586#ifdef SMP 587 } 588#endif 589 return (0); 590} 591 592/* 593 * Software interrupt handler for queued VM system processing. 594 */ 595void 596swi_vm() 597{ 598 if (busdma_swi_pending != 0) 599 busdma_swi(); 600} 601 602/* 603 * Tell whether this address is in some physical memory region. 604 * Currently used by the kernel coredump code in order to avoid 605 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 606 * or other unpredictable behaviour. 607 */ 608 609#include "isa.h" 610 611int 612is_physical_memory(addr) 613 vm_offset_t addr; 614{ 615 616#if NISA > 0 617 /* The ISA ``memory hole''. */ 618 if (addr >= 0xa0000 && addr < 0x100000) 619 return 0; 620#endif 621 622 /* 623 * stuff other tests for known memory-mapped devices (PCI?) 624 * here 625 */ 626 627 return 1; 628} 629