vm_machdep.c revision 73922
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 73922 2001-03-07 03:20:15Z jhb $ 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/sysctl.h> 63#include <sys/unistd.h> 64 65#include <machine/cpu.h> 66#include <machine/md_var.h> 67#ifdef SMP 68#include <machine/smp.h> 69#endif 70#include <machine/pcb.h> 71#include <machine/pcb_ext.h> 72#include <machine/vm86.h> 73 74#include <vm/vm.h> 75#include <vm/vm_param.h> 76#include <sys/lock.h> 77#include <vm/vm_kern.h> 78#include <vm/vm_page.h> 79#include <vm/vm_map.h> 80#include <vm/vm_extern.h> 81 82#include <sys/user.h> 83 84#ifdef PC98 85#include <pc98/pc98/pc98.h> 86#else 87#include <i386/isa/isa.h> 88#endif 89 90static void cpu_reset_real __P((void)); 91#ifdef SMP 92static void cpu_reset_proxy __P((void)); 93static u_int cpu_reset_proxyid; 94static volatile u_int cpu_reset_proxy_active; 95#endif 96extern int _ucodesel, _udatasel; 97 98/* 99 * quick version of vm_fault 100 */ 101int 102vm_fault_quick(v, prot) 103 caddr_t v; 104 int prot; 105{ 106 int r; 107 108 if (prot & VM_PROT_WRITE) 109 r = subyte(v, fubyte(v)); 110 else 111 r = fubyte(v); 112 return(r); 113} 114 115/* 116 * Finish a fork operation, with process p2 nearly set up. 117 * Copy and update the pcb, set up the stack so that the child 118 * ready to run and return to user mode. 119 */ 120void 121cpu_fork(p1, p2, flags) 122 register struct proc *p1, *p2; 123 int flags; 124{ 125 struct pcb *pcb2; 126 127 if ((flags & RFPROC) == 0) { 128 if ((flags & RFMEM) == 0) { 129 /* unshare user LDT */ 130 struct pcb *pcb1 = &p1->p_addr->u_pcb; 131 struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt; 132 if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) { 133 pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len); 134 user_ldt_free(pcb1); 135 pcb1->pcb_ldt = pcb_ldt; 136 set_user_ldt(pcb1); 137 } 138 } 139 return; 140 } 141 142#ifdef DEV_NPX 143 /* Ensure that p1's pcb is up to date. */ 144 if (PCPU_GET(npxproc) == p1) 145 npxsave(&p1->p_addr->u_pcb.pcb_savefpu); 146#endif 147 148 /* Copy p1's pcb. */ 149 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 150 pcb2 = &p2->p_addr->u_pcb; 151 152 /* 153 * Create a new fresh stack for the new process. 154 * Copy the trap frame for the return to user mode as if from a 155 * syscall. This copies the user mode register values. 156 */ 157 p2->p_md.md_regs = (struct trapframe *) 158 ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1; 159 bcopy(p1->p_md.md_regs, p2->p_md.md_regs, sizeof(*p2->p_md.md_regs)); 160 161 p2->p_md.md_regs->tf_eax = 0; /* Child returns zero */ 162 p2->p_md.md_regs->tf_eflags &= ~PSL_C; /* success */ 163 p2->p_md.md_regs->tf_edx = 1; 164 165 /* 166 * Set registers for trampoline to user mode. Leave space for the 167 * return address on stack. These are the kernel mode register values. 168 */ 169 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir); 170 pcb2->pcb_edi = 0; 171 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */ 172 pcb2->pcb_ebp = 0; 173 pcb2->pcb_esp = (int)p2->p_md.md_regs - sizeof(void *); 174 pcb2->pcb_ebx = (int)p2; /* fork_trampoline argument */ 175 pcb2->pcb_eip = (int)fork_trampoline; 176 /* 177 * pcb2->pcb_ldt: duplicated below, if necessary. 178 * pcb2->pcb_savefpu: cloned above. 179 * pcb2->pcb_flags: cloned above (always 0 here?). 180 * pcb2->pcb_onfault: cloned above (always NULL here?). 181 */ 182 183 /* 184 * XXX don't copy the i/o pages. this should probably be fixed. 185 */ 186 pcb2->pcb_ext = 0; 187 188 /* Copy the LDT, if necessary. */ 189 if (pcb2->pcb_ldt != 0) { 190 if (flags & RFMEM) { 191 pcb2->pcb_ldt->ldt_refcnt++; 192 } else { 193 pcb2->pcb_ldt = user_ldt_alloc(pcb2, 194 pcb2->pcb_ldt->ldt_len); 195 } 196 } 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 fork_trampoline(), which will have 203 * %ebx loaded with the new proc's pointer. fork_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#ifdef DEV_NPX 236 npxexit(p); 237#endif 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 if (pcb->pcb_ldt) 248 user_ldt_free(pcb); 249 if (pcb->pcb_flags & PCB_DBREGS) { 250 /* 251 * disable all hardware breakpoints 252 */ 253 reset_dbregs(); 254 pcb->pcb_flags &= ~PCB_DBREGS; 255 } 256 PROC_LOCK(p); 257 mtx_lock_spin(&sched_lock); 258 mtx_unlock_flags(&Giant, MTX_NOSWITCH); 259 mtx_assert(&Giant, MA_NOTOWNED); 260 261 /* 262 * We have to wait until after releasing all locks before 263 * changing p_stat. If we block on a mutex then we will be 264 * back at SRUN when we resume and our parent will never 265 * harvest us. 266 */ 267 p->p_stat = SZOMB; 268 269 wakeup(p->p_pptr); 270 PROC_UNLOCK_NOSWITCH(p); 271 272 cnt.v_swtch++; 273 cpu_throw(); 274 panic("cpu_exit"); 275} 276 277void 278cpu_wait(p) 279 struct proc *p; 280{ 281 /* drop per-process resources */ 282 pmap_dispose_proc(p); 283 284 /* and clean-out the vmspace */ 285 vmspace_free(p->p_vmspace); 286} 287 288/* 289 * Dump the machine specific header information at the start of a core dump. 290 */ 291int 292cpu_coredump(p, vp, cred) 293 struct proc *p; 294 struct vnode *vp; 295 struct ucred *cred; 296{ 297 int error; 298 caddr_t tempuser; 299 300 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK | M_ZERO); 301 if (!tempuser) 302 return EINVAL; 303 304 bcopy(p->p_addr, tempuser, sizeof(struct user)); 305 bcopy(p->p_md.md_regs, 306 tempuser + ((caddr_t) p->p_md.md_regs - (caddr_t) p->p_addr), 307 sizeof(struct trapframe)); 308 309 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, 310 ctob(UPAGES), 311 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, 312 cred, (int *)NULL, p); 313 314 free(tempuser, M_TEMP); 315 316 return error; 317} 318 319#ifdef notyet 320static void 321setredzone(pte, vaddr) 322 u_short *pte; 323 caddr_t vaddr; 324{ 325/* eventually do this by setting up an expand-down stack segment 326 for ss0: selector, allowing stack access down to top of u. 327 this means though that protection violations need to be handled 328 thru a double fault exception that must do an integral task 329 switch to a known good context, within which a dump can be 330 taken. a sensible scheme might be to save the initial context 331 used by sched (that has physical memory mapped 1:1 at bottom) 332 and take the dump while still in mapped mode */ 333} 334#endif 335 336/* 337 * Convert kernel VA to physical address 338 */ 339u_long 340kvtop(void *addr) 341{ 342 vm_offset_t va; 343 344 va = pmap_kextract((vm_offset_t)addr); 345 if (va == 0) 346 panic("kvtop: zero page frame"); 347 return((int)va); 348} 349 350/* 351 * Map an IO request into kernel virtual address space. 352 * 353 * All requests are (re)mapped into kernel VA space. 354 * Notice that we use b_bufsize for the size of the buffer 355 * to be mapped. b_bcount might be modified by the driver. 356 */ 357void 358vmapbuf(bp) 359 register struct buf *bp; 360{ 361 register caddr_t addr, v, kva; 362 vm_offset_t pa; 363 364 if ((bp->b_flags & B_PHYS) == 0) 365 panic("vmapbuf"); 366 367 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 368 addr < bp->b_data + bp->b_bufsize; 369 addr += PAGE_SIZE, v += PAGE_SIZE) { 370 /* 371 * Do the vm_fault if needed; do the copy-on-write thing 372 * when reading stuff off device into memory. 373 */ 374 vm_fault_quick(addr, 375 (bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 376 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 377 if (pa == 0) 378 panic("vmapbuf: page not present"); 379 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 380 pmap_kenter((vm_offset_t) v, pa); 381 } 382 383 kva = bp->b_saveaddr; 384 bp->b_saveaddr = bp->b_data; 385 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 386} 387 388/* 389 * Free the io map PTEs associated with this IO operation. 390 * We also invalidate the TLB entries and restore the original b_addr. 391 */ 392void 393vunmapbuf(bp) 394 register struct buf *bp; 395{ 396 register caddr_t addr; 397 vm_offset_t pa; 398 399 if ((bp->b_flags & B_PHYS) == 0) 400 panic("vunmapbuf"); 401 402 for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 403 addr < bp->b_data + bp->b_bufsize; 404 addr += PAGE_SIZE) { 405 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 406 pmap_kremove((vm_offset_t) addr); 407 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 408 } 409 410 bp->b_data = bp->b_saveaddr; 411} 412 413/* 414 * Force reset the processor by invalidating the entire address space! 415 */ 416 417#ifdef SMP 418static void 419cpu_reset_proxy() 420{ 421 422 cpu_reset_proxy_active = 1; 423 while (cpu_reset_proxy_active == 1) 424 ; /* Wait for other cpu to see that we've started */ 425 stop_cpus((1<<cpu_reset_proxyid)); 426 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 427 DELAY(1000000); 428 cpu_reset_real(); 429} 430#endif 431 432void 433cpu_reset() 434{ 435#ifdef SMP 436 if (smp_active == 0) { 437 cpu_reset_real(); 438 /* NOTREACHED */ 439 } else { 440 441 u_int map; 442 int cnt; 443 printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid)); 444 445 map = PCPU_GET(other_cpus) & ~ stopped_cpus; 446 447 if (map != 0) { 448 printf("cpu_reset: Stopping other CPUs\n"); 449 stop_cpus(map); /* Stop all other CPUs */ 450 } 451 452 if (PCPU_GET(cpuid) == 0) { 453 DELAY(1000000); 454 cpu_reset_real(); 455 /* NOTREACHED */ 456 } else { 457 /* We are not BSP (CPU #0) */ 458 459 cpu_reset_proxyid = PCPU_GET(cpuid); 460 cpustop_restartfunc = cpu_reset_proxy; 461 cpu_reset_proxy_active = 0; 462 printf("cpu_reset: Restarting BSP\n"); 463 started_cpus = (1<<0); /* Restart CPU #0 */ 464 465 cnt = 0; 466 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 467 cnt++; /* Wait for BSP to announce restart */ 468 if (cpu_reset_proxy_active == 0) 469 printf("cpu_reset: Failed to restart BSP\n"); 470 enable_intr(); 471 cpu_reset_proxy_active = 2; 472 473 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 if (mtx_trylock(&Giant)) { 571 s = splvm(); 572 zero_state = 0; 573 m = vm_page_list_find(PQ_FREE, free_rover, FALSE); 574 if (m != NULL && (m->flags & PG_ZERO) == 0) { 575 vm_page_queues[m->queue].lcnt--; 576 TAILQ_REMOVE(&vm_page_queues[m->queue].pl, m, pageq); 577 m->queue = PQ_NONE; 578 splx(s); 579 pmap_zero_page(VM_PAGE_TO_PHYS(m)); 580 (void)splvm(); 581 vm_page_flag_set(m, PG_ZERO); 582 m->queue = PQ_FREE + m->pc; 583 vm_page_queues[m->queue].lcnt++; 584 TAILQ_INSERT_TAIL(&vm_page_queues[m->queue].pl, m, 585 pageq); 586 ++vm_page_zero_count; 587 ++cnt_prezero; 588 if (vm_page_zero_count >= ZIDLE_HI(cnt.v_free_count)) 589 zero_state = 1; 590 } 591 free_rover = (free_rover + PQ_PRIME2) & PQ_L2_MASK; 592 splx(s); 593 mtx_unlock(&Giant); 594 return (1); 595 } 596 return (0); 597} 598 599/* 600 * Software interrupt handler for queued VM system processing. 601 */ 602void 603swi_vm(void *dummy) 604{ 605 if (busdma_swi_pending != 0) 606 busdma_swi(); 607} 608 609/* 610 * Tell whether this address is in some physical memory region. 611 * Currently used by the kernel coredump code in order to avoid 612 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 613 * or other unpredictable behaviour. 614 */ 615 616int 617is_physical_memory(addr) 618 vm_offset_t addr; 619{ 620 621#ifdef DEV_ISA 622 /* The ISA ``memory hole''. */ 623 if (addr >= 0xa0000 && addr < 0x100000) 624 return 0; 625#endif 626 627 /* 628 * stuff other tests for known memory-mapped devices (PCI?) 629 * here 630 */ 631 632 return 1; 633} 634