vm_machdep.c revision 76546
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 76546 2001-05-13 07:44:14Z bde $ 42 */ 43 44#include "opt_npx.h" 45#ifdef PC98 46#include "opt_pc98.h" 47#endif 48#include "opt_reset.h" 49#include "opt_isa.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/ktr.h> 61#include <sys/mutex.h> 62#include <sys/smp.h> 63#include <sys/sysctl.h> 64#include <sys/unistd.h> 65 66#include <machine/cpu.h> 67#include <machine/md_var.h> 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 94extern int _ucodesel, _udatasel; 95 96/* 97 * quick version of vm_fault 98 */ 99int 100vm_fault_quick(v, prot) 101 caddr_t v; 102 int prot; 103{ 104 int r; 105 106 if (prot & VM_PROT_WRITE) 107 r = subyte(v, fubyte(v)); 108 else 109 r = fubyte(v); 110 return(r); 111} 112 113/* 114 * Finish a fork operation, with process p2 nearly set up. 115 * Copy and update the pcb, set up the stack so that the child 116 * ready to run and return to user mode. 117 */ 118void 119cpu_fork(p1, p2, flags) 120 register struct proc *p1, *p2; 121 int flags; 122{ 123 struct pcb *pcb2; 124#ifdef DEV_NPX 125 int savecrit; 126#endif 127 128 if ((flags & RFPROC) == 0) { 129 if ((flags & RFMEM) == 0) { 130 /* unshare user LDT */ 131 struct pcb *pcb1 = &p1->p_addr->u_pcb; 132 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt; 133 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) { 134 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len); 135 user_ldt_free(pcb1); 136 pcb1->pcb_ldt = pcb_ldt; 137 set_user_ldt(pcb1); 138 } 139 } 140 return; 141 } 142 143 /* Ensure that p1's pcb is up to date. */ 144#ifdef DEV_NPX 145 if (p1 == curproc) 146 p1->p_addr->u_pcb.pcb_gs = rgs(); 147 savecrit = critical_enter(); 148 if (PCPU_GET(npxproc) == p1) 149 npxsave(&p1->p_addr->u_pcb.pcb_savefpu); 150 critical_exit(savecrit); 151#endif 152 153 /* Copy p1's pcb. */ 154 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 155 pcb2 = &p2->p_addr->u_pcb; 156 157 /* 158 * Create a new fresh stack for the new process. 159 * Copy the trap frame for the return to user mode as if from a 160 * syscall. This copies most of the user mode register values. 161 */ 162 p2->p_md.md_regs = (struct trapframe *) 163 ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1; 164 bcopy(p1->p_md.md_regs, p2->p_md.md_regs, sizeof(*p2->p_md.md_regs)); 165 166 p2->p_md.md_regs->tf_eax = 0; /* Child returns zero */ 167 p2->p_md.md_regs->tf_eflags &= ~PSL_C; /* success */ 168 p2->p_md.md_regs->tf_edx = 1; 169 170 /* 171 * Set registers for trampoline to user mode. Leave space for the 172 * return address on stack. These are the kernel mode register values. 173 */ 174 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir); 175 pcb2->pcb_edi = 0; 176 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */ 177 pcb2->pcb_ebp = 0; 178 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *); 179 pcb2->pcb_ebx = (int)p2; /* fork_trampoline argument */ 180 pcb2->pcb_eip = (int)fork_trampoline; 181 /*- 182 * pcb2->pcb_dr*: cloned above. 183 * pcb2->pcb_ldt: duplicated below, if necessary. 184 * pcb2->pcb_savefpu: cloned above. 185 * pcb2->pcb_flags: cloned above. 186 * pcb2->pcb_onfault: cloned above (always NULL here?). 187 * pcb2->pcb_gs: cloned above. 188 * pcb2->pcb_ext: cleared below. 189 */ 190 191 /* 192 * XXX don't copy the i/o pages. this should probably be fixed. 193 */ 194 pcb2->pcb_ext = 0; 195 196 /* Copy the LDT, if necessary. */ 197 mtx_lock_spin(&sched_lock); 198 if (pcb2->pcb_ldt != 0) { 199 if (flags & RFMEM) { 200 pcb2->pcb_ldt->ldt_refcnt++; 201 } else { 202 pcb2->pcb_ldt = user_ldt_alloc(pcb2, 203 pcb2->pcb_ldt->ldt_len); 204 if (pcb2->pcb_ldt == NULL) 205 panic("could not copy LDT"); 206 } 207 } 208 mtx_unlock_spin(&sched_lock); 209 210 /* 211 * Now, cpu_switch() can schedule the new process. 212 * pcb_esp is loaded pointing to the cpu_switch() stack frame 213 * containing the return address when exiting cpu_switch. 214 * This will normally be to fork_trampoline(), which will have 215 * %ebx loaded with the new proc's pointer. fork_trampoline() 216 * will set up a stack to call fork_return(p, frame); to complete 217 * the return to user-mode. 218 */ 219} 220 221/* 222 * Intercept the return address from a freshly forked process that has NOT 223 * been scheduled yet. 224 * 225 * This is needed to make kernel threads stay in kernel mode. 226 */ 227void 228cpu_set_fork_handler(p, func, arg) 229 struct proc *p; 230 void (*func) __P((void *)); 231 void *arg; 232{ 233 /* 234 * Note that the trap frame follows the args, so the function 235 * is really called like this: func(arg, frame); 236 */ 237 p->p_addr->u_pcb.pcb_esi = (int) func; /* function */ 238 p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */ 239} 240 241void 242cpu_exit(p) 243 register struct proc *p; 244{ 245 struct pcb *pcb = &p->p_addr->u_pcb; 246 247#ifdef DEV_NPX 248 npxexit(p); 249#endif 250 if (pcb->pcb_ext != 0) { 251 /* 252 * XXX do we need to move the TSS off the allocated pages 253 * before freeing them? (not done here) 254 */ 255 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 256 ctob(IOPAGES + 1)); 257 pcb->pcb_ext = 0; 258 } 259 if (pcb->pcb_ldt) 260 user_ldt_free(pcb); 261 if (pcb->pcb_flags & PCB_DBREGS) { 262 /* 263 * disable all hardware breakpoints 264 */ 265 reset_dbregs(); 266 pcb->pcb_flags &= ~PCB_DBREGS; 267 } 268 PROC_LOCK(p); 269 mtx_lock_spin(&sched_lock); 270 mtx_unlock_flags(&Giant, MTX_NOSWITCH); 271 mtx_assert(&Giant, MA_NOTOWNED); 272 273 /* 274 * We have to wait until after releasing all locks before 275 * changing p_stat. If we block on a mutex then we will be 276 * back at SRUN when we resume and our parent will never 277 * harvest us. 278 */ 279 p->p_stat = SZOMB; 280 281 wakeup(p->p_pptr); 282 PROC_UNLOCK_NOSWITCH(p); 283 284 cnt.v_swtch++; 285 cpu_throw(); 286 panic("cpu_exit"); 287} 288 289void 290cpu_wait(p) 291 struct proc *p; 292{ 293 /* drop per-process resources */ 294 pmap_dispose_proc(p); 295 296 /* and clean-out the vmspace */ 297 vmspace_free(p->p_vmspace); 298} 299 300/* 301 * Dump the machine specific header information at the start of a core dump. 302 */ 303int 304cpu_coredump(p, vp, cred) 305 struct proc *p; 306 struct vnode *vp; 307 struct ucred *cred; 308{ 309 int error; 310 caddr_t tempuser; 311 312 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK | M_ZERO); 313 if (!tempuser) 314 return EINVAL; 315 316 bcopy(p->p_addr, tempuser, sizeof(struct user)); 317 bcopy(p->p_md.md_regs, 318 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr), 319 sizeof(struct trapframe)); 320 321 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, 322 ctob(UPAGES), 323 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, 324 cred, (int *)NULL, p); 325 326 free(tempuser, M_TEMP); 327 328 return error; 329} 330 331#ifdef notyet 332static void 333setredzone(pte, vaddr) 334 u_short *pte; 335 caddr_t vaddr; 336{ 337/* eventually do this by setting up an expand-down stack segment 338 for ss0: selector, allowing stack access down to top of u. 339 this means though that protection violations need to be handled 340 thru a double fault exception that must do an integral task 341 switch to a known good context, within which a dump can be 342 taken. a sensible scheme might be to save the initial context 343 used by sched (that has physical memory mapped 1:1 at bottom) 344 and take the dump while still in mapped mode */ 345} 346#endif 347 348/* 349 * Convert kernel VA to physical address 350 */ 351u_long 352kvtop(void *addr) 353{ 354 vm_offset_t va; 355 356 va = pmap_kextract((vm_offset_t)addr); 357 if (va == 0) 358 panic("kvtop: zero page frame"); 359 return((int)va); 360} 361 362/* 363 * Map an IO request into kernel virtual address space. 364 * 365 * All requests are (re)mapped into kernel VA space. 366 * Notice that we use b_bufsize for the size of the buffer 367 * to be mapped. b_bcount might be modified by the driver. 368 */ 369void 370vmapbuf(bp) 371 register struct buf *bp; 372{ 373 register caddr_t addr, v, kva; 374 vm_offset_t pa; 375 376 if ((bp->b_flags & B_PHYS) == 0) 377 panic("vmapbuf"); 378 379 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 380 addr < bp->b_data + bp->b_bufsize; 381 addr += PAGE_SIZE, v += PAGE_SIZE) { 382 /* 383 * Do the vm_fault if needed; do the copy-on-write thing 384 * when reading stuff off device into memory. 385 */ 386 vm_fault_quick(addr, 387 (bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 388 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 389 if (pa == 0) 390 panic("vmapbuf: page not present"); 391 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 392 pmap_kenter((vm_offset_t) v, pa); 393 } 394 395 kva = bp->b_saveaddr; 396 bp->b_saveaddr = bp->b_data; 397 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 398} 399 400/* 401 * Free the io map PTEs associated with this IO operation. 402 * We also invalidate the TLB entries and restore the original b_addr. 403 */ 404void 405vunmapbuf(bp) 406 register struct buf *bp; 407{ 408 register caddr_t addr; 409 vm_offset_t pa; 410 411 if ((bp->b_flags & B_PHYS) == 0) 412 panic("vunmapbuf"); 413 414 for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 415 addr < bp->b_data + bp->b_bufsize; 416 addr += PAGE_SIZE) { 417 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 418 pmap_kremove((vm_offset_t) addr); 419 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 420 } 421 422 bp->b_data = bp->b_saveaddr; 423} 424 425/* 426 * Force reset the processor by invalidating the entire address space! 427 */ 428 429#ifdef SMP 430static void 431cpu_reset_proxy() 432{ 433 434 cpu_reset_proxy_active = 1; 435 while (cpu_reset_proxy_active == 1) 436 ; /* Wait for other cpu to see that we've started */ 437 stop_cpus((1<<cpu_reset_proxyid)); 438 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 439 DELAY(1000000); 440 cpu_reset_real(); 441} 442#endif 443 444void 445cpu_reset() 446{ 447#ifdef SMP 448 if (smp_active == 0) { 449 cpu_reset_real(); 450 /* NOTREACHED */ 451 } else { 452 453 u_int map; 454 int cnt; 455 printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid)); 456 457 map = PCPU_GET(other_cpus) & ~ stopped_cpus; 458 459 if (map != 0) { 460 printf("cpu_reset: Stopping other CPUs\n"); 461 stop_cpus(map); /* Stop all other CPUs */ 462 } 463 464 if (PCPU_GET(cpuid) == 0) { 465 DELAY(1000000); 466 cpu_reset_real(); 467 /* NOTREACHED */ 468 } else { 469 /* We are not BSP (CPU #0) */ 470 471 cpu_reset_proxyid = PCPU_GET(cpuid); 472 cpustop_restartfunc = cpu_reset_proxy; 473 cpu_reset_proxy_active = 0; 474 printf("cpu_reset: Restarting BSP\n"); 475 started_cpus = (1<<0); /* Restart CPU #0 */ 476 477 cnt = 0; 478 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 479 cnt++; /* Wait for BSP to announce restart */ 480 if (cpu_reset_proxy_active == 0) 481 printf("cpu_reset: Failed to restart BSP\n"); 482 enable_intr(); 483 cpu_reset_proxy_active = 2; 484 485 while (1); 486 /* NOTREACHED */ 487 } 488 } 489#else 490 cpu_reset_real(); 491#endif 492} 493 494static void 495cpu_reset_real() 496{ 497 498#ifdef PC98 499 /* 500 * Attempt to do a CPU reset via CPU reset port. 501 */ 502 disable_intr(); 503 if ((inb(0x35) & 0xa0) != 0xa0) { 504 outb(0x37, 0x0f); /* SHUT0 = 0. */ 505 outb(0x37, 0x0b); /* SHUT1 = 0. */ 506 } 507 outb(0xf0, 0x00); /* Reset. */ 508#else 509 /* 510 * Attempt to do a CPU reset via the keyboard controller, 511 * do not turn of the GateA20, as any machine that fails 512 * to do the reset here would then end up in no man's land. 513 */ 514 515#if !defined(BROKEN_KEYBOARD_RESET) 516 outb(IO_KBD + 4, 0xFE); 517 DELAY(500000); /* wait 0.5 sec to see if that did it */ 518 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 519 DELAY(1000000); /* wait 1 sec for printf to complete */ 520#endif 521#endif /* PC98 */ 522 /* force a shutdown by unmapping entire address space ! */ 523 bzero((caddr_t) PTD, PAGE_SIZE); 524 525 /* "good night, sweet prince .... <THUNK!>" */ 526 invltlb(); 527 /* NOTREACHED */ 528 while(1); 529} 530 531int 532grow_stack(p, sp) 533 struct proc *p; 534 u_int sp; 535{ 536 int rv; 537 538 rv = vm_map_growstack (p, sp); 539 if (rv != KERN_SUCCESS) 540 return (0); 541 542 return (1); 543} 544 545SYSCTL_DECL(_vm_stats_misc); 546 547static int cnt_prezero; 548 549SYSCTL_INT(_vm_stats_misc, OID_AUTO, 550 cnt_prezero, CTLFLAG_RD, &cnt_prezero, 0, ""); 551 552/* 553 * Implement the pre-zeroed page mechanism. 554 * This routine is called from the idle loop. 555 */ 556 557#define ZIDLE_LO(v) ((v) * 2 / 3) 558#define ZIDLE_HI(v) ((v) * 4 / 5) 559 560int 561vm_page_zero_idle() 562{ 563 static int free_rover; 564 static int zero_state; 565 vm_page_t m; 566 int s; 567 568 /* 569 * Attempt to maintain approximately 1/2 of our free pages in a 570 * PG_ZERO'd state. Add some hysteresis to (attempt to) avoid 571 * generally zeroing a page when the system is near steady-state. 572 * Otherwise we might get 'flutter' during disk I/O / IPC or 573 * fast sleeps. We also do not want to be continuously zeroing 574 * pages because doing so may flush our L1 and L2 caches too much. 575 */ 576 577 if (zero_state && vm_page_zero_count >= ZIDLE_LO(cnt.v_free_count)) 578 return(0); 579 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 580 return(0); 581 582 if (mtx_trylock(&Giant)) { 583 s = splvm(); 584 zero_state = 0; 585 m = vm_page_list_find(PQ_FREE, free_rover, FALSE); 586 if (m != NULL && (m->flags & PG_ZERO) == 0) { 587 vm_page_queues[m->queue].lcnt--; 588 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq); 589 m->queue = PQ_NONE; 590 splx(s); 591 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 592 (void)splvm(); 593 vm_page_flag_set(m, PG_ZERO); 594 m->queue = PQ_FREE + m->pc; 595 vm_page_queues[m->queue].lcnt++; 596 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m, 597 pageq); 598 ++vm_page_zero_count; 599 ++cnt_prezero; 600 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 601 zero_state = 1; 602 } 603 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK; 604 splx(s); 605 mtx_unlock(&Giant); 606 return (1); 607 } 608 return (0); 609} 610 611/* 612 * Software interrupt handler for queued VM system processing. 613 */ 614void 615swi_vm(void *dummy) 616{ 617 if (busdma_swi_pending != 0) 618 busdma_swi(); 619} 620 621/* 622 * Tell whether this address is in some physical memory region. 623 * Currently used by the kernel coredump code in order to avoid 624 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 625 * or other unpredictable behaviour. 626 */ 627 628int 629is_physical_memory(addr) 630 vm_offset_t addr; 631{ 632 633#ifdef DEV_ISA 634 /* The ISA ``memory hole''. */ 635 if (addr >= 0xa0000 && addr < 0x100000) 636 return 0; 637#endif 638 639 /* 640 * stuff other tests for known memory-mapped devices (PCI?) 641 * here 642 */ 643 644 return 1; 645} 646