vm_machdep.c revision 82309
1139825Simp/*- 21541Srgrimes * Copyright (c) 1982, 1986 The Regents of the University of California. 31541Srgrimes * Copyright (c) 1989, 1990 William Jolitz 41541Srgrimes * Copyright (c) 1994 John Dyson 51541Srgrimes * All rights reserved. 61541Srgrimes * 71541Srgrimes * This code is derived from software contributed to Berkeley by 81541Srgrimes * the Systems Programming Group of the University of Utah Computer 91541Srgrimes * Science Department, and William Jolitz. 101541Srgrimes * 111541Srgrimes * Redistribution and use in source and binary forms, with or without 121541Srgrimes * modification, are permitted provided that the following conditions 131541Srgrimes * are met: 141541Srgrimes * 1. Redistributions of source code must retain the above copyright 151541Srgrimes * notice, this list of conditions and the following disclaimer. 161541Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 171541Srgrimes * notice, this list of conditions and the following disclaimer in the 181541Srgrimes * documentation and/or other materials provided with the distribution. 191541Srgrimes * 3. All advertising materials mentioning features or use of this software 201541Srgrimes * must display the following acknowledgement: 211541Srgrimes * This product includes software developed by the University of 221541Srgrimes * California, Berkeley and its contributors. 231541Srgrimes * 4. Neither the name of the University nor the names of its contributors 241541Srgrimes * may be used to endorse or promote products derived from this software 251541Srgrimes * without specific prior written permission. 261541Srgrimes * 271541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 281541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 291541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 3050477Speter * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 311541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 321541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 331541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 341541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 351541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 36103731Swollman * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3734319Sdufault * SUCH DAMAGE. 38103731Swollman * 39103731Swollman * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91 40103731Swollman * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ 41103731Swollman * $FreeBSD: head/sys/amd64/amd64/vm_machdep.c 82309 2001-08-25 02:20:02Z peter $ 42103731Swollman */ 43103731Swollman 44103731Swollman#include "opt_npx.h" 45103731Swollman#ifdef PC98 46103731Swollman#include "opt_pc98.h" 47103731Swollman#endif 48103731Swollman#include "opt_reset.h" 49103731Swollman#include "opt_isa.h" 50103731Swollman#include "opt_upages.h" 51103731Swollman 52103731Swollman#include <sys/param.h> 53103731Swollman#include <sys/systm.h> 54103731Swollman#include <sys/malloc.h> 55103731Swollman#include <sys/proc.h> 56103731Swollman#include <sys/bio.h> 57103731Swollman#include <sys/buf.h> 58106055Swollman#include <sys/vnode.h> 59103731Swollman#include <sys/vmmeter.h> 60103731Swollman#include <sys/kernel.h> 61103731Swollman#include <sys/ktr.h> 62103731Swollman#include <sys/mutex.h> 63103731Swollman#include <sys/smp.h> 64153006Sdavidxu#include <sys/sysctl.h> 65151659Swollman#include <sys/unistd.h> 66103731Swollman 67103731Swollman#include <machine/cpu.h> 68103731Swollman#include <machine/md_var.h> 69103731Swollman#include <machine/pcb.h> 70151578Sdavidxu#include <machine/pcb_ext.h> 71103731Swollman#include <machine/vm86.h> 72103731Swollman 73103731Swollman#include <vm/vm.h> 74103731Swollman#include <vm/vm_param.h> 75153004Sdavidxu#include <sys/lock.h> 76151872Sdavidxu#include <vm/vm_kern.h> 77103731Swollman#include <vm/vm_page.h> 78103731Swollman#include <vm/vm_map.h> 791541Srgrimes#include <vm/vm_extern.h> 80103731Swollman 81103731Swollman#include <sys/user.h> 82103731Swollman 83103731Swollman#ifdef PC98 84103731Swollman#include <pc98/pc98/pc98.h> 851541Srgrimes#else 8620640Sbde#include <i386/isa/isa.h> 8720640Sbde#endif 8820640Sbde 89103731Swollmanstatic void cpu_reset_real __P((void)); 90103731Swollman#ifdef SMP 911541Srgrimesstatic void cpu_reset_proxy __P((void)); 9220640Sbdestatic u_int cpu_reset_proxyid; 93103731Swollmanstatic volatile u_int cpu_reset_proxy_active; 9420640Sbde#endif 951541Srgrimesextern int _ucodesel, _udatasel; 96105045Smike 97105045Smike/* 98105045Smike * quick version of vm_fault 991541Srgrimes */ 1001541Srgrimesint 1011541Srgrimesvm_fault_quick(v, prot) 1021541Srgrimes caddr_t v; 1031541Srgrimes int prot; 1041541Srgrimes{ 1051541Srgrimes int r; 106105789Sphk 1071541Srgrimes if (prot & VM_PROT_WRITE) 1081541Srgrimes r = subyte(v, fubyte(v)); 1091541Srgrimes else 110168397Spjd r = fubyte(v); 111168397Spjd return(r); 112105789Sphk} 1131541Srgrimes 1141541Srgrimes/* 1151541Srgrimes * Finish a fork operation, with process p2 nearly set up. 1161541Srgrimes * Copy and update the pcb, set up the stack so that the child 1171541Srgrimes * ready to run and return to user mode. 1181541Srgrimes */ 1191541Srgrimesvoid 1201541Srgrimescpu_fork(p1, p2, flags) 1211541Srgrimes register struct proc *p1, *p2; 1221541Srgrimes int flags; 1231541Srgrimes{ 1241541Srgrimes struct pcb *pcb2; 1251541Srgrimes#ifdef DEV_NPX 1261541Srgrimes int savecrit; 1271541Srgrimes#endif 1281541Srgrimes 1291541Srgrimes if ((flags & RFPROC) == 0) { 1301541Srgrimes if ((flags & RFMEM) == 0) { 1311541Srgrimes /* unshare user LDT */ 132103731Swollman struct pcb *pcb1 = &p1->p_addr->u_pcb; 133103768Sbde struct pcb_ldt *pcb_ldt = pcb1->pcb_ldt; 134103768Sbde if (pcb_ldt && pcb_ldt->ldt_refcnt > 1) { 135103768Sbde pcb_ldt = user_ldt_alloc(pcb1,pcb_ldt->ldt_len); 136103731Swollman if (pcb_ldt == NULL) 13734030Sdufault panic("could not copy LDT"); 138103731Swollman pcb1->pcb_ldt = pcb_ldt; 139103731Swollman set_user_ldt(pcb1); 140103731Swollman user_ldt_free(pcb1); 141103731Swollman } 142103731Swollman } 143103731Swollman return; 144103731Swollman } 145106055Swollman 146103731Swollman /* Ensure that p1's pcb is up to date. */ 14734030Sdufault#ifdef DEV_NPX 148103731Swollman if (p1 == curproc) 149105572Srwatson p1->p_addr->u_pcb.pcb_gs = rgs(); 150105572Srwatson savecrit = critical_enter(); 151105572Srwatson if (PCPU_GET(npxproc) == p1) 152105572Srwatson npxsave(&p1->p_addr->u_pcb.pcb_save); 153105572Srwatson critical_exit(savecrit); 154103731Swollman#endif 155101071Srwatson 156168397Spjd /* Copy p1's pcb. */ 157168397Spjd p2->p_addr->u_pcb = p1->p_addr->u_pcb; 158168397Spjd pcb2 = &p2->p_addr->u_pcb; 159103731Swollman 16014222Speter /* 16114222Speter * Create a new fresh stack for the new process. 16214222Speter * Copy the trap frame for the return to user mode as if from a 16365557Sjasone * syscall. This copies most of the user mode register values. 16414222Speter */ 165103768Sbde p2->p_frame = (struct trapframe *) 166103768Sbde ((int)p2->p_addr + UPAGES * PAGE_SIZE - 16) - 1; 167103768Sbde bcopy(p1->p_frame, p2->p_frame, sizeof(struct trapframe)); 168103768Sbde 169103768Sbde p2->p_frame->tf_eax = 0; /* Child returns zero */ 170103768Sbde p2->p_frame->tf_eflags &= ~PSL_C; /* success */ 171103768Sbde p2->p_frame->tf_edx = 1; 172103768Sbde 173103768Sbde /* 174103768Sbde * Set registers for trampoline to user mode. Leave space for the 175103731Swollman * return address on stack. These are the kernel mode register values. 176103731Swollman */ 177103768Sbde pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir); 178103731Swollman pcb2->pcb_edi = 0; 179103768Sbde pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */ 180103768Sbde pcb2->pcb_ebp = 0; 181103768Sbde pcb2->pcb_esp = (int)p2->p_frame - sizeof(void *); 18226671Sdyson pcb2->pcb_ebx = (int)p2; /* fork_trampoline argument */ 183103731Swollman pcb2->pcb_eip = (int)fork_trampoline; 18414222Speter /*- 1851541Srgrimes * pcb2->pcb_dr*: cloned above. 186 * pcb2->pcb_ldt: duplicated below, if necessary. 187 * pcb2->pcb_savefpu: cloned above. 188 * pcb2->pcb_flags: cloned above. 189 * pcb2->pcb_onfault: cloned above (always NULL here?). 190 * pcb2->pcb_gs: cloned above. 191 * pcb2->pcb_ext: cleared below. 192 */ 193 194 /* 195 * XXX don't copy the i/o pages. this should probably be fixed. 196 */ 197 pcb2->pcb_ext = 0; 198 199 /* Copy the LDT, if necessary. */ 200 mtx_lock_spin(&sched_lock); 201 if (pcb2->pcb_ldt != 0) { 202 if (flags & RFMEM) { 203 pcb2->pcb_ldt->ldt_refcnt++; 204 } else { 205 pcb2->pcb_ldt = user_ldt_alloc(pcb2, 206 pcb2->pcb_ldt->ldt_len); 207 if (pcb2->pcb_ldt == NULL) 208 panic("could not copy LDT"); 209 } 210 } 211 mtx_unlock_spin(&sched_lock); 212 213 /* 214 * Now, cpu_switch() can schedule the new process. 215 * pcb_esp is loaded pointing to the cpu_switch() stack frame 216 * containing the return address when exiting cpu_switch. 217 * This will normally be to fork_trampoline(), which will have 218 * %ebx loaded with the new proc's pointer. fork_trampoline() 219 * will set up a stack to call fork_return(p, frame); to complete 220 * the return to user-mode. 221 */ 222} 223 224/* 225 * Intercept the return address from a freshly forked process that has NOT 226 * been scheduled yet. 227 * 228 * This is needed to make kernel threads stay in kernel mode. 229 */ 230void 231cpu_set_fork_handler(p, func, arg) 232 struct proc *p; 233 void (*func) __P((void *)); 234 void *arg; 235{ 236 /* 237 * Note that the trap frame follows the args, so the function 238 * is really called like this: func(arg, frame); 239 */ 240 p->p_addr->u_pcb.pcb_esi = (int) func; /* function */ 241 p->p_addr->u_pcb.pcb_ebx = (int) arg; /* first arg */ 242} 243 244void 245cpu_exit(p) 246 register struct proc *p; 247{ 248 struct pcb *pcb = &p->p_addr->u_pcb; 249 250#ifdef DEV_NPX 251 npxexit(p); 252#endif 253 if (pcb->pcb_ext != 0) { 254 /* 255 * XXX do we need to move the TSS off the allocated pages 256 * before freeing them? (not done here) 257 */ 258 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 259 ctob(IOPAGES + 1)); 260 pcb->pcb_ext = 0; 261 } 262 if (pcb->pcb_ldt) 263 user_ldt_free(pcb); 264 if (pcb->pcb_flags & PCB_DBREGS) { 265 /* 266 * disable all hardware breakpoints 267 */ 268 reset_dbregs(); 269 pcb->pcb_flags &= ~PCB_DBREGS; 270 } 271 PROC_LOCK(p); 272 mtx_lock_spin(&sched_lock); 273 while (mtx_owned(&Giant)) 274 mtx_unlock_flags(&Giant, MTX_NOSWITCH); 275 276 /* 277 * We have to wait until after releasing all locks before 278 * changing p_stat. If we block on a mutex then we will be 279 * back at SRUN when we resume and our parent will never 280 * harvest us. 281 */ 282 p->p_stat = SZOMB; 283 284 wakeup(p->p_pptr); 285 PROC_UNLOCK_NOSWITCH(p); 286 287 cnt.v_swtch++; 288 cpu_throw(); 289 panic("cpu_exit"); 290} 291 292void 293cpu_wait(p) 294 struct proc *p; 295{ 296 GIANT_REQUIRED; 297 298 /* drop per-process resources */ 299 pmap_dispose_proc(p); 300 301 /* and clean-out the vmspace */ 302 vmspace_free(p->p_vmspace); 303} 304 305/* 306 * Dump the machine specific header information at the start of a core dump. 307 */ 308int 309cpu_coredump(p, vp, cred) 310 struct proc *p; 311 struct vnode *vp; 312 struct ucred *cred; 313{ 314 int error; 315 caddr_t tempuser; 316 317 tempuser = malloc(ctob(UPAGES), M_TEMP, M_WAITOK | M_ZERO); 318 if (!tempuser) 319 return EINVAL; 320 321 bcopy(p->p_addr, tempuser, sizeof(struct user)); 322 bcopy(p->p_frame, 323 tempuser + ((caddr_t) p->p_frame - (caddr_t) p->p_addr), 324 sizeof(struct trapframe)); 325 326 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, 327 ctob(UPAGES), 328 (off_t)0, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, 329 cred, (int *)NULL, p); 330 331 free(tempuser, M_TEMP); 332 333 return error; 334} 335 336#ifdef notyet 337static void 338setredzone(pte, vaddr) 339 u_short *pte; 340 caddr_t vaddr; 341{ 342/* eventually do this by setting up an expand-down stack segment 343 for ss0: selector, allowing stack access down to top of u. 344 this means though that protection violations need to be handled 345 thru a double fault exception that must do an integral task 346 switch to a known good context, within which a dump can be 347 taken. a sensible scheme might be to save the initial context 348 used by sched (that has physical memory mapped 1:1 at bottom) 349 and take the dump while still in mapped mode */ 350} 351#endif 352 353/* 354 * Convert kernel VA to physical address 355 */ 356u_long 357kvtop(void *addr) 358{ 359 vm_offset_t va; 360 361 va = pmap_kextract((vm_offset_t)addr); 362 if (va == 0) 363 panic("kvtop: zero page frame"); 364 return((int)va); 365} 366 367/* 368 * Map an IO request into kernel virtual address space. 369 * 370 * All requests are (re)mapped into kernel VA space. 371 * Notice that we use b_bufsize for the size of the buffer 372 * to be mapped. b_bcount might be modified by the driver. 373 */ 374void 375vmapbuf(bp) 376 register struct buf *bp; 377{ 378 register caddr_t addr, v, kva; 379 vm_offset_t pa; 380 381 GIANT_REQUIRED; 382 383 if ((bp->b_flags & B_PHYS) == 0) 384 panic("vmapbuf"); 385 386 for (v = bp->b_saveaddr, addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 387 addr < bp->b_data + bp->b_bufsize; 388 addr += PAGE_SIZE, v += PAGE_SIZE) { 389 /* 390 * Do the vm_fault if needed; do the copy-on-write thing 391 * when reading stuff off device into memory. 392 */ 393 vm_fault_quick(addr, 394 (bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 395 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 396 if (pa == 0) 397 panic("vmapbuf: page not present"); 398 vm_page_hold(PHYS_TO_VM_PAGE(pa)); 399 pmap_kenter((vm_offset_t) v, pa); 400 } 401 402 kva = bp->b_saveaddr; 403 bp->b_saveaddr = bp->b_data; 404 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 405} 406 407/* 408 * Free the io map PTEs associated with this IO operation. 409 * We also invalidate the TLB entries and restore the original b_addr. 410 */ 411void 412vunmapbuf(bp) 413 register struct buf *bp; 414{ 415 register caddr_t addr; 416 vm_offset_t pa; 417 418 GIANT_REQUIRED; 419 420 if ((bp->b_flags & B_PHYS) == 0) 421 panic("vunmapbuf"); 422 423 for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data); 424 addr < bp->b_data + bp->b_bufsize; 425 addr += PAGE_SIZE) { 426 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 427 pmap_kremove((vm_offset_t) addr); 428 vm_page_unhold(PHYS_TO_VM_PAGE(pa)); 429 } 430 431 bp->b_data = bp->b_saveaddr; 432} 433 434/* 435 * Force reset the processor by invalidating the entire address space! 436 */ 437 438#ifdef SMP 439static void 440cpu_reset_proxy() 441{ 442 443 cpu_reset_proxy_active = 1; 444 while (cpu_reset_proxy_active == 1) 445 ; /* Wait for other cpu to see that we've started */ 446 stop_cpus((1<<cpu_reset_proxyid)); 447 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 448 DELAY(1000000); 449 cpu_reset_real(); 450} 451#endif 452 453void 454cpu_reset() 455{ 456#ifdef SMP 457 if (smp_active == 0) { 458 cpu_reset_real(); 459 /* NOTREACHED */ 460 } else { 461 462 u_int map; 463 int cnt; 464 printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid)); 465 466 map = PCPU_GET(other_cpus) & ~ stopped_cpus; 467 468 if (map != 0) { 469 printf("cpu_reset: Stopping other CPUs\n"); 470 stop_cpus(map); /* Stop all other CPUs */ 471 } 472 473 if (PCPU_GET(cpuid) == 0) { 474 DELAY(1000000); 475 cpu_reset_real(); 476 /* NOTREACHED */ 477 } else { 478 /* We are not BSP (CPU #0) */ 479 480 cpu_reset_proxyid = PCPU_GET(cpuid); 481 cpustop_restartfunc = cpu_reset_proxy; 482 cpu_reset_proxy_active = 0; 483 printf("cpu_reset: Restarting BSP\n"); 484 started_cpus = (1<<0); /* Restart CPU #0 */ 485 486 cnt = 0; 487 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 488 cnt++; /* Wait for BSP to announce restart */ 489 if (cpu_reset_proxy_active == 0) 490 printf("cpu_reset: Failed to restart BSP\n"); 491 enable_intr(); 492 cpu_reset_proxy_active = 2; 493 494 while (1); 495 /* NOTREACHED */ 496 } 497 } 498#else 499 cpu_reset_real(); 500#endif 501} 502 503static void 504cpu_reset_real() 505{ 506 507#ifdef PC98 508 /* 509 * Attempt to do a CPU reset via CPU reset port. 510 */ 511 disable_intr(); 512 if ((inb(0x35) & 0xa0) != 0xa0) { 513 outb(0x37, 0x0f); /* SHUT0 = 0. */ 514 outb(0x37, 0x0b); /* SHUT1 = 0. */ 515 } 516 outb(0xf0, 0x00); /* Reset. */ 517#else 518 /* 519 * Attempt to do a CPU reset via the keyboard controller, 520 * do not turn of the GateA20, as any machine that fails 521 * to do the reset here would then end up in no man's land. 522 */ 523 524#if !defined(BROKEN_KEYBOARD_RESET) 525 outb(IO_KBD + 4, 0xFE); 526 DELAY(500000); /* wait 0.5 sec to see if that did it */ 527 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 528 DELAY(1000000); /* wait 1 sec for printf to complete */ 529#endif 530#endif /* PC98 */ 531 /* force a shutdown by unmapping entire address space ! */ 532 bzero((caddr_t) PTD, PAGE_SIZE); 533 534 /* "good night, sweet prince .... <THUNK!>" */ 535 invltlb(); 536 /* NOTREACHED */ 537 while(1); 538} 539 540int 541grow_stack(p, sp) 542 struct proc *p; 543 u_int sp; 544{ 545 int rv; 546 547 rv = vm_map_growstack (p, sp); 548 if (rv != KERN_SUCCESS) 549 return (0); 550 551 return (1); 552} 553 554/* 555 * Software interrupt handler for queued VM system processing. 556 */ 557void 558swi_vm(void *dummy) 559{ 560 if (busdma_swi_pending != 0) 561 busdma_swi(); 562} 563 564/* 565 * Tell whether this address is in some physical memory region. 566 * Currently used by the kernel coredump code in order to avoid 567 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 568 * or other unpredictable behaviour. 569 */ 570 571int 572is_physical_memory(addr) 573 vm_offset_t addr; 574{ 575 576#ifdef DEV_ISA 577 /* The ISA ``memory hole''. */ 578 if (addr >= 0xa0000 && addr < 0x100000) 579 return 0; 580#endif 581 582 /* 583 * stuff other tests for known memory-mapped devices (PCI?) 584 * here 585 */ 586 587 return 1; 588} 589