vm_machdep.c revision 51474
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 51474 1999-09-20 19:08:48Z dillon $ 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#include <machine/pcb_ext.h> 67#include <machine/vm86.h> 68 69#include <vm/vm.h> 70#include <vm/vm_param.h> 71#include <vm/vm_prot.h> 72#include <sys/lock.h> 73#include <vm/vm_kern.h> 74#include <vm/vm_page.h> 75#include <vm/vm_map.h> 76#include <vm/vm_extern.h> 77 78#include <sys/user.h> 79 80#ifdef PC98 81#include <pc98/pc98/pc98.h> 82#else 83#include <i386/isa/isa.h> 84#endif 85 86static void cpu_reset_real __P((void)); 87#ifdef SMP 88static void cpu_reset_proxy __P((void)); 89static u_int cpu_reset_proxyid; 90static volatile u_int cpu_reset_proxy_active; 91#endif 92 93/* 94 * quick version of vm_fault 95 */ 96int 97vm_fault_quick(v, prot) 98 caddr_t v; 99 int prot; 100{ 101 int r; 102 103 if (prot & VM_PROT_WRITE) 104 r = subyte(v, fubyte(v)); 105 else 106 r = fubyte(v); 107 return(r); 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 ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1; 137 *p2->p_md.md_regs = *p1->p_md.md_regs; 138 139 /* 140 * Set registers for trampoline to user mode. Leave space for the 141 * return address on stack. These are the kernel mode register values. 142 */ 143 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir); 144 pcb2->pcb_edi = p2->p_md.md_regs->tf_edi; 145 pcb2->pcb_esi = (int)fork_return; 146 pcb2->pcb_ebp = p2->p_md.md_regs->tf_ebp; 147 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *); 148 pcb2->pcb_ebx = (int)p2; 149 pcb2->pcb_eip = (int)fork_trampoline; 150 /* 151 * pcb2->pcb_ldt: duplicated below, if necessary. 152 * pcb2->pcb_ldt_len: cloned above. 153 * pcb2->pcb_savefpu: cloned above. 154 * pcb2->pcb_flags: cloned above (always 0 here?). 155 * pcb2->pcb_onfault: cloned above (always NULL here?). 156 */ 157 158#ifdef SMP 159 pcb2->pcb_mpnest = 1; 160#endif 161 /* 162 * XXX don't copy the i/o pages. this should probably be fixed. 163 */ 164 pcb2->pcb_ext = 0; 165 166#ifdef USER_LDT 167 /* Copy the LDT, if necessary. */ 168 if (pcb2->pcb_ldt != 0) { 169 union descriptor *new_ldt; 170 size_t len = pcb2->pcb_ldt_len * sizeof(union descriptor); 171 172 new_ldt = (union descriptor *)kmem_alloc(kernel_map, len); 173 bcopy(pcb2->pcb_ldt, new_ldt, len); 174 pcb2->pcb_ldt = (caddr_t)new_ldt; 175 } 176#endif 177 178 /* 179 * Now, cpu_switch() can schedule the new process. 180 * pcb_esp is loaded pointing to the cpu_switch() stack frame 181 * containing the return address when exiting cpu_switch. 182 * This will normally be to proc_trampoline(), which will have 183 * %ebx loaded with the new proc's pointer. proc_trampoline() 184 * will set up a stack to call fork_return(p, frame); to complete 185 * the return to user-mode. 186 */ 187} 188 189/* 190 * Intercept the return address from a freshly forked process that has NOT 191 * been scheduled yet. 192 * 193 * This is needed to make kernel threads stay in kernel mode. 194 */ 195void 196cpu_set_fork_handler(p, func, arg) 197 struct proc *p; 198 void (*func) __P((void *)); 199 void *arg; 200{ 201 /* 202 * Note that the trap frame follows the args, so the function 203 * is really called like this: func(arg, frame); 204 */ 205 p->p_addr->u_pcb.pcb_esi = (int) func; /* function */ 206 p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */ 207} 208 209void 210cpu_exit(p) 211 register struct proc *p; 212{ 213 struct pcb *pcb = &p->p_addr->u_pcb; 214 215#if NNPX > 0 216 npxexit(p); 217#endif /* NNPX */ 218 if (pcb->pcb_ext != 0) { 219 /* 220 * XXX do we need to move the TSS off the allocated pages 221 * before freeing them? (not done here) 222 */ 223 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 224 ctob(IOPAGES + 1)); 225 pcb->pcb_ext = 0; 226 } 227#ifdef USER_LDT 228 if (pcb->pcb_ldt != 0) { 229 if (pcb == curpcb) { 230 lldt(_default_ldt); 231 currentldt = _default_ldt; 232 } 233 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ldt, 234 pcb->pcb_ldt_len * sizeof(union descriptor)); 235 pcb->pcb_ldt_len = (int)pcb->pcb_ldt = 0; 236 } 237#endif 238 cnt.v_swtch++; 239 cpu_switch(p); 240 panic("cpu_exit"); 241} 242 243void 244cpu_wait(p) 245 struct proc *p; 246{ 247 /* drop per-process resources */ 248 pmap_dispose_proc(p); 249 250 /* and clean-out the vmspace */ 251 vmspace_free(p->p_vmspace); 252} 253 254/* 255 * Dump the machine specific header information at the start of a core dump. 256 */ 257int 258cpu_coredump(p, vp, cred) 259 struct proc *p; 260 struct vnode *vp; 261 struct ucred *cred; 262{ 263 int error; 264 caddr_t tempuser; 265 266 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK); 267 if (!tempuser) 268 return EINVAL; 269 270 bzero(tempuser, ctob(UPAGES)); 271 bcopy(p->p_addr, tempuser, sizeof(struct user)); 272 bcopy(p->p_md.md_regs, 273 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr), 274 sizeof(struct trapframe)); 275 276 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, 277 ctob(UPAGES), 278 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, 279 cred, (int *)NULL, p); 280 281 free(tempuser, M_TEMP); 282 283 return error; 284} 285 286#ifdef notyet 287static void 288setredzone(pte, vaddr) 289 u_short *pte; 290 caddr_t vaddr; 291{ 292/* eventually do this by setting up an expand-down stack segment 293 for ss0: selector, allowing stack access down to top of u. 294 this means though that protection violations need to be handled 295 thru a double fault exception that must do an integral task 296 switch to a known good context, within which a dump can be 297 taken. a sensible scheme might be to save the initial context 298 used by sched (that has physical memory mapped 1:1 at bottom) 299 and take the dump while still in mapped mode */ 300} 301#endif 302 303/* 304 * Convert kernel VA to physical address 305 */ 306u_long 307kvtop(void *addr) 308{ 309 vm_offset_t va; 310 311 va = pmap_kextract((vm_offset_t)addr); 312 if (va == 0) 313 panic("kvtop: zero page frame"); 314 return((int)va); 315} 316 317/* 318 * Map an IO request into kernel virtual address space. 319 * 320 * All requests are (re)mapped into kernel VA space. 321 * Notice that we use b_bufsize for the size of the buffer 322 * to be mapped. b_bcount might be modified by the driver. 323 */ 324void 325vmapbuf(bp) 326 register struct buf *bp; 327{ 328 register caddr_t addr, v, kva; 329 vm_offset_t pa; 330 331 if ((bp->b_flags & B_PHYS) == 0) 332 panic("vmapbuf"); 333 334 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 335 addr < bp->b_data + bp->b_bufsize; 336 addr += PAGE_SIZE, v += PAGE_SIZE) { 337 /* 338 * Do the vm_fault if needed; do the copy-on-write thing 339 * when reading stuff off device into memory. 340 */ 341 vm_fault_quick(addr, 342 (bp->b_flags&B_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 343 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 344 if (pa == 0) 345 panic("vmapbuf: page not present"); 346 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 347 pmap_kenter((vm_offset_t) v, pa); 348 } 349 350 kva = bp->b_saveaddr; 351 bp->b_saveaddr = bp->b_data; 352 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 353} 354 355/* 356 * Free the io map PTEs associated with this IO operation. 357 * We also invalidate the TLB entries and restore the original b_addr. 358 */ 359void 360vunmapbuf(bp) 361 register struct buf *bp; 362{ 363 register caddr_t addr; 364 vm_offset_t pa; 365 366 if ((bp->b_flags & B_PHYS) == 0) 367 panic("vunmapbuf"); 368 369 for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 370 addr < bp->b_data + bp->b_bufsize; 371 addr += PAGE_SIZE) { 372 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 373 pmap_kremove((vm_offset_t) addr); 374 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 375 } 376 377 bp->b_data = bp->b_saveaddr; 378} 379 380/* 381 * Force reset the processor by invalidating the entire address space! 382 */ 383 384#ifdef SMP 385static void 386cpu_reset_proxy() 387{ 388 u_int saved_mp_lock; 389 390 cpu_reset_proxy_active = 1; 391 while (cpu_reset_proxy_active == 1) 392 ; /* Wait for other cpu to disable interupts */ 393 saved_mp_lock = mp_lock; 394 mp_lock = 1; 395 printf("cpu_reset_proxy: Grabbed mp lock for BSP\n"); 396 cpu_reset_proxy_active = 3; 397 while (cpu_reset_proxy_active == 3) 398 ; /* Wait for other cpu to enable interrupts */ 399 stop_cpus((1<<cpu_reset_proxyid)); 400 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 401 DELAY(1000000); 402 cpu_reset_real(); 403} 404#endif 405 406void 407cpu_reset() 408{ 409#ifdef SMP 410 if (smp_active == 0) { 411 cpu_reset_real(); 412 /* NOTREACHED */ 413 } else { 414 415 u_int map; 416 int cnt; 417 printf("cpu_reset called on cpu#%d\n",cpuid); 418 419 map = other_cpus & ~ stopped_cpus; 420 421 if (map != 0) { 422 printf("cpu_reset: Stopping other CPUs\n"); 423 stop_cpus(map); /* Stop all other CPUs */ 424 } 425 426 if (cpuid == 0) { 427 DELAY(1000000); 428 cpu_reset_real(); 429 /* NOTREACHED */ 430 } else { 431 /* We are not BSP (CPU #0) */ 432 433 cpu_reset_proxyid = cpuid; 434 cpustop_restartfunc = cpu_reset_proxy; 435 printf("cpu_reset: Restarting BSP\n"); 436 started_cpus = (1<<0); /* Restart CPU #0 */ 437 438 cnt = 0; 439 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 440 cnt++; /* Wait for BSP to announce restart */ 441 if (cpu_reset_proxy_active == 0) 442 printf("cpu_reset: Failed to restart BSP\n"); 443 __asm __volatile("cli" : : : "memory"); 444 cpu_reset_proxy_active = 2; 445 cnt = 0; 446 while (cpu_reset_proxy_active == 2 && cnt < 10000000) 447 cnt++; /* Do nothing */ 448 if (cpu_reset_proxy_active == 2) { 449 printf("cpu_reset: BSP did not grab mp lock\n"); 450 cpu_reset_real(); /* XXX: Bogus ? */ 451 } 452 cpu_reset_proxy_active = 4; 453 __asm __volatile("sti" : : : "memory"); 454 while (1); 455 /* NOTREACHED */ 456 } 457 } 458#else 459 cpu_reset_real(); 460#endif 461} 462 463static void 464cpu_reset_real() 465{ 466 467#ifdef PC98 468 /* 469 * Attempt to do a CPU reset via CPU reset port. 470 */ 471 disable_intr(); 472 if ((inb(0x35) & 0xa0) != 0xa0) { 473 outb(0x37, 0x0f); /* SHUT0 = 0. */ 474 outb(0x37, 0x0b); /* SHUT1 = 0. */ 475 } 476 outb(0xf0, 0x00); /* Reset. */ 477#else 478 /* 479 * Attempt to do a CPU reset via the keyboard controller, 480 * do not turn of the GateA20, as any machine that fails 481 * to do the reset here would then end up in no man's land. 482 */ 483 484#if !defined(BROKEN_KEYBOARD_RESET) 485 outb(IO_KBD + 4, 0xFE); 486 DELAY(500000); /* wait 0.5 sec to see if that did it */ 487 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 488 DELAY(1000000); /* wait 1 sec for printf to complete */ 489#endif 490#endif /* PC98 */ 491 /* force a shutdown by unmapping entire address space ! */ 492 bzero((caddr_t) PTD, PAGE_SIZE); 493 494 /* "good night, sweet prince .... <THUNK!>" */ 495 invltlb(); 496 /* NOTREACHED */ 497 while(1); 498} 499 500int 501grow_stack(p, sp) 502 struct proc *p; 503 u_int sp; 504{ 505 int rv; 506 507 rv = vm_map_growstack (p, sp); 508 if (rv != KERN_SUCCESS) 509 return (0); 510 511 return (1); 512} 513 514SYSCTL_DECL(_vm_stats_misc); 515 516static int cnt_prezero; 517 518SYSCTL_INT(_vm_stats_misc, OID_AUTO, 519 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, ""); 520 521/* 522 * Implement the pre-zeroed page mechanism. 523 * This routine is called from the idle loop. 524 */ 525 526#define ZIDLE_LO(v) ((v) * 2 / 3) 527#define ZIDLE_HI(v) ((v) * 4 / 5) 528 529int 530vm_page_zero_idle() 531{ 532 static int free_rover; 533 static int zero_state; 534 vm_page_t m; 535 int s; 536 537 /* 538 * Attempt to maintain approximately 1/2 of our free pages in a 539 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid 540 * generally zeroing a page when the system is near steady-state. 541 * Otherwise we might get 'flutter' during disk I/O / IPC or 542 * fast sleeps. We also do not want to be continuously zeroing 543 * pages because doing so may flush our L1 and L2 caches too much. 544 */ 545 546 if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count)) 547 return(0); 548 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 549 return(0); 550 551#ifdef SMP 552 if (try_mplock()) { 553#endif 554 s = splvm(); 555 __asm __volatile("sti" : : : "memory"); 556 zero_state = 0; 557 m = vm_page_list_find(PQ_FREE, free_rover, FALSE); 558 if (m != NULL && (m->flags & PG_ZERO) == 0) { 559 vm_page_queues[m->queue].lcnt--; 560 TAILQ_REMOVE(vm_page_queues[m->queue].pl, m, pageq); 561 m->queue = PQ_NONE; 562 splx(s); 563#if 0 564 rel_mplock(); 565#endif 566 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 567#if 0 568 get_mplock(); 569#endif 570 (void)splvm(); 571 vm_page_flag_set(m, PG_ZERO); 572 m->queue = PQ_FREE + m->pc; 573 vm_page_queues[m->queue].lcnt++; 574 TAILQ_INSERT_TAIL(vm_page_queues[m->queue].pl, m, 575 pageq); 576 ++vm_page_zero_count; 577 ++cnt_prezero; 578 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 579 zero_state = 1; 580 } 581 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK; 582 splx(s); 583 __asm __volatile("cli" : : : "memory"); 584#ifdef SMP 585 rel_mplock(); 586#endif 587 return (1); 588#ifdef SMP 589 } 590#endif 591 return (0); 592} 593 594/* 595 * Software interrupt handler for queued VM system processing. 596 */ 597void 598swi_vm() 599{ 600 if (busdma_swi_pending != 0) 601 busdma_swi(); 602} 603 604/* 605 * Tell whether this address is in some physical memory region. 606 * Currently used by the kernel coredump code in order to avoid 607 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 608 * or other unpredictable behaviour. 609 */ 610 611#include "isa.h" 612 613int 614is_physical_memory(addr) 615 vm_offset_t addr; 616{ 617 618#if NISA > 0 619 /* The ISA ``memory hole''. */ 620 if (addr >= 0xa0000 && addr < 0x100000) 621 return 0; 622#endif 623 624 /* 625 * stuff other tests for known memory-mapped devices (PCI?) 626 * here 627 */ 628 629 return 1; 630} 631