vm_machdep.c revision 103049
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 103049 2002-09-07 01:26:34Z peter $ 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#include "opt_kstack_pages.h" 51 52#include <sys/param.h> 53#include <sys/systm.h> 54#include <sys/malloc.h> 55#include <sys/proc.h> 56#include <sys/kse.h> 57#include <sys/bio.h> 58#include <sys/buf.h> 59#include <sys/vnode.h> 60#include <sys/vmmeter.h> 61#include <sys/kernel.h> 62#include <sys/ktr.h> 63#include <sys/mutex.h> 64#include <sys/smp.h> 65#include <sys/sysctl.h> 66#include <sys/unistd.h> 67 68#include <machine/cpu.h> 69#include <machine/md_var.h> 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(void); 91#ifdef SMP 92static void cpu_reset_proxy(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(td1, p2, td2, flags) 122 register struct thread *td1; 123 register struct proc *p2; 124 struct thread *td2; 125 int flags; 126{ 127 register struct proc *p1; 128 struct pcb *pcb2; 129 struct mdproc *mdp2; 130#ifdef DEV_NPX 131 register_t savecrit; 132#endif 133 134 p1 = td1->td_proc; 135 if ((flags & RFPROC) == 0) { 136 if ((flags & RFMEM) == 0) { 137 /* unshare user LDT */ 138 struct mdproc *mdp1 = &p1->p_md; 139 struct proc_ldt *pldt = mdp1->md_ldt; 140 if (pldt && pldt->ldt_refcnt > 1) { 141 pldt = user_ldt_alloc(mdp1, pldt->ldt_len); 142 if (pldt == NULL) 143 panic("could not copy LDT"); 144 mdp1->md_ldt = pldt; 145 set_user_ldt(mdp1); 146 user_ldt_free(td1); 147 } 148 } 149 return; 150 } 151 152 /* Ensure that p1's pcb is up to date. */ 153#ifdef DEV_NPX 154 if (td1 == curthread) 155 td1->td_pcb->pcb_gs = rgs(); 156 savecrit = intr_disable(); 157 if (PCPU_GET(fpcurthread) == td1) 158 npxsave(&td1->td_pcb->pcb_save); 159 intr_restore(savecrit); 160#endif 161 162 /* Point the pcb to the top of the stack */ 163 pcb2 = (struct pcb *)(td2->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1; 164 td2->td_pcb = pcb2; 165 166 /* Copy p1's pcb */ 167 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 168 169 /* Point mdproc and then copy over td1's contents */ 170 mdp2 = &p2->p_md; 171 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 172 173 /* 174 * Create a new fresh stack for the new process. 175 * Copy the trap frame for the return to user mode as if from a 176 * syscall. This copies most of the user mode register values. 177 * The -16 is so we can expand the trapframe if we go to vm86. 178 */ 179 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1; 180 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 181 182 td2->td_frame->tf_eax = 0; /* Child returns zero */ 183 td2->td_frame->tf_eflags &= ~PSL_C; /* success */ 184 td2->td_frame->tf_edx = 1; 185 186 /* 187 * Set registers for trampoline to user mode. Leave space for the 188 * return address on stack. These are the kernel mode register values. 189 */ 190 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir); 191 pcb2->pcb_edi = 0; 192 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */ 193 pcb2->pcb_ebp = 0; 194 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *); 195 pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */ 196 pcb2->pcb_eip = (int)fork_trampoline; 197 pcb2->pcb_psl = td2->td_frame->tf_eflags & ~PSL_I; /* ints disabled */ 198 /*- 199 * pcb2->pcb_dr*: cloned above. 200 * pcb2->pcb_savefpu: cloned above. 201 * pcb2->pcb_flags: cloned above. 202 * pcb2->pcb_onfault: cloned above (always NULL here?). 203 * pcb2->pcb_gs: cloned above. 204 * pcb2->pcb_ext: cleared below. 205 */ 206 207 /* 208 * XXX don't copy the i/o pages. this should probably be fixed. 209 */ 210 pcb2->pcb_ext = 0; 211 212 /* Copy the LDT, if necessary. */ 213 mtx_lock_spin(&sched_lock); 214 if (mdp2->md_ldt != 0) { 215 if (flags & RFMEM) { 216 mdp2->md_ldt->ldt_refcnt++; 217 } else { 218 mdp2->md_ldt = user_ldt_alloc(mdp2, 219 mdp2->md_ldt->ldt_len); 220 if (mdp2->md_ldt == NULL) 221 panic("could not copy LDT"); 222 } 223 } 224 mtx_unlock_spin(&sched_lock); 225 226 /* 227 * Now, cpu_switch() can schedule the new process. 228 * pcb_esp is loaded pointing to the cpu_switch() stack frame 229 * containing the return address when exiting cpu_switch. 230 * This will normally be to fork_trampoline(), which will have 231 * %ebx loaded with the new proc's pointer. fork_trampoline() 232 * will set up a stack to call fork_return(p, frame); to complete 233 * the return to user-mode. 234 */ 235} 236 237/* 238 * Intercept the return address from a freshly forked process that has NOT 239 * been scheduled yet. 240 * 241 * This is needed to make kernel threads stay in kernel mode. 242 */ 243void 244cpu_set_fork_handler(td, func, arg) 245 struct thread *td; 246 void (*func)(void *); 247 void *arg; 248{ 249 /* 250 * Note that the trap frame follows the args, so the function 251 * is really called like this: func(arg, frame); 252 */ 253 td->td_pcb->pcb_esi = (int) func; /* function */ 254 td->td_pcb->pcb_ebx = (int) arg; /* first arg */ 255} 256 257void 258cpu_exit(struct thread *td) 259{ 260 struct mdproc *mdp; 261 262 mdp = &td->td_proc->p_md; 263 if (mdp->md_ldt) 264 user_ldt_free(td); 265 reset_dbregs(); 266} 267 268void 269cpu_thread_exit(struct thread *td) 270{ 271 struct pcb *pcb = td->td_pcb; 272#ifdef DEV_NPX 273 npxexit(td); 274#endif 275 if (pcb->pcb_ext != 0) { 276 /* XXXKSE XXXSMP not SMP SAFE.. what locks do we have? */ 277 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */ 278 /* 279 * XXX do we need to move the TSS off the allocated pages 280 * before freeing them? (not done here) 281 */ 282 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 283 ctob(IOPAGES + 1)); 284 pcb->pcb_ext = 0; 285 } 286 if (pcb->pcb_flags & PCB_DBREGS) { 287 /* 288 * disable all hardware breakpoints 289 */ 290 reset_dbregs(); 291 pcb->pcb_flags &= ~PCB_DBREGS; 292 } 293} 294 295void 296cpu_sched_exit(td) 297 register struct thread *td; 298{ 299} 300 301void 302cpu_thread_setup(struct thread *td) 303{ 304 305 td->td_pcb = 306 (struct pcb *)(td->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1; 307 td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb - 16) - 1; 308} 309 310struct md_store { 311 struct pcb mds_pcb; 312 struct trapframe mds_frame; 313}; 314 315void 316cpu_save_upcall(struct thread *td, struct kse *newkse) 317{ 318 struct trapframe *tf; 319 320 newkse->ke_mdstorage = malloc(sizeof(struct md_store), M_TEMP, 321 M_WAITOK); 322 /* Note: use of M_WAITOK means it won't fail. */ 323 /* set up shortcuts in MI section */ 324 newkse->ke_pcb = 325 &(((struct md_store *)(newkse->ke_mdstorage))->mds_pcb); 326 newkse->ke_frame = 327 &(((struct md_store *)(newkse->ke_mdstorage))->mds_frame); 328 tf = newkse->ke_frame; 329 330 /* Copy the upcall pcb. Kernel mode & fp regs are here. */ 331 /* XXXKSE this may be un-needed */ 332 bcopy(td->td_pcb, newkse->ke_pcb, sizeof(struct pcb)); 333 334 /* 335 * This initialises most of the user mode register values 336 * to good values. Eventually set them explicitly to know values 337 */ 338 bcopy(td->td_frame, newkse->ke_frame, sizeof(struct trapframe)); 339 tf->tf_edi = 0; 340 tf->tf_esi = 0; /* trampoline arg */ 341 tf->tf_ebp = 0; 342 tf->tf_esp = (int)newkse->ke_stackbase + newkse->ke_stacksize - 16; 343 tf->tf_ebx = 0; /* trampoline arg */ 344 tf->tf_eip = (int)newkse->ke_upcall; 345} 346 347void 348cpu_set_upcall(struct thread *td, void *pcb) 349{ 350 struct pcb *pcb2; 351 352 td->td_flags |= TDF_UPCALLING; 353 354 /* Point the pcb to the top of the stack. */ 355 pcb2 = td->td_pcb; 356 357 /* 358 * Copy the upcall pcb. This loads kernel regs. 359 * Those not loaded individually below get their default 360 * values here. 361 * 362 * XXXKSE It might be a good idea to simply skip this as 363 * the values of the other registers may be unimportant. 364 * This would remove any requirement for knowing the KSE 365 * at this time (see the matching comment below for 366 * more analysis) (need a good safe default). 367 */ 368 bcopy(pcb, pcb2, sizeof(*pcb2)); 369 370 /* 371 * Create a new fresh stack for the new thread. 372 * The -16 is so we can expand the trapframe if we go to vm86. 373 * Don't forget to set this stack value into whatever supplies 374 * the address for the fault handlers. 375 * The contexts are filled in at the time we actually DO the 376 * upcall as only then do we know which KSE we got. 377 */ 378 td->td_frame = (struct trapframe *)((caddr_t)pcb2 - 16) - 1; 379 380 /* 381 * Set registers for trampoline to user mode. Leave space for the 382 * return address on stack. These are the kernel mode register values. 383 */ 384 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdir); 385 pcb2->pcb_edi = 0; 386 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */ 387 pcb2->pcb_ebp = 0; 388 pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */ 389 pcb2->pcb_ebx = (int)td; /* trampoline arg */ 390 pcb2->pcb_eip = (int)fork_trampoline; 391 pcb2->pcb_psl &= ~(PSL_I); /* interrupts must be disabled */ 392 /* 393 * If we didn't copy the pcb, we'd need to do the following registers: 394 * pcb2->pcb_dr*: cloned above. 395 * pcb2->pcb_savefpu: cloned above. 396 * pcb2->pcb_flags: cloned above. 397 * pcb2->pcb_onfault: cloned above (always NULL here?). 398 * pcb2->pcb_gs: cloned above. XXXKSE ??? 399 * pcb2->pcb_ext: cleared below. 400 */ 401 pcb2->pcb_ext = NULL; 402} 403 404void 405cpu_set_args(struct thread *td, struct kse *ke) 406{ 407 suword((void *)(ke->ke_frame->tf_esp + sizeof(void *)), 408 (int)ke->ke_mailbox); 409} 410 411void 412cpu_free_kse_mdstorage(struct kse *kse) 413{ 414 415 free(kse->ke_mdstorage, M_TEMP); 416 kse->ke_mdstorage = NULL; 417 kse->ke_pcb = NULL; 418 kse->ke_frame = NULL; 419} 420 421int 422cpu_export_context(struct thread *td) 423{ 424 struct trapframe *frame; 425 struct thread_mailbox *tm; 426 struct trapframe *uframe; 427 int error; 428 429 frame = td->td_frame; 430 tm = td->td_mailbox; 431 uframe = &tm->ctx.tfrm.tf_tf; 432 error = copyout(frame, uframe, sizeof(*frame)); 433 /* 434 * "What about the fp regs?" I hear you ask.... XXXKSE 435 * Don't know where gs and "onstack" come from. 436 * May need to fiddle a few other values too. 437 */ 438 return (error); 439} 440 441void 442cpu_wait(p) 443 struct proc *p; 444{ 445} 446 447/* 448 * Convert kernel VA to physical address 449 */ 450u_long 451kvtop(void *addr) 452{ 453 vm_offset_t va; 454 455 va = pmap_kextract((vm_offset_t)addr); 456 if (va == 0) 457 panic("kvtop: zero page frame"); 458 return((int)va); 459} 460 461/* 462 * Map an IO request into kernel virtual address space. 463 * 464 * All requests are (re)mapped into kernel VA space. 465 * Notice that we use b_bufsize for the size of the buffer 466 * to be mapped. b_bcount might be modified by the driver. 467 */ 468void 469vmapbuf(bp) 470 register struct buf *bp; 471{ 472 register caddr_t addr, kva; 473 vm_offset_t pa; 474 int pidx; 475 struct vm_page *m; 476 477 GIANT_REQUIRED; 478 479 if ((bp->b_flags & B_PHYS) == 0) 480 panic("vmapbuf"); 481 482 for (addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data), pidx = 0; 483 addr < bp->b_data + bp->b_bufsize; 484 addr += PAGE_SIZE, pidx++) { 485 /* 486 * Do the vm_fault if needed; do the copy-on-write thing 487 * when reading stuff off device into memory. 488 */ 489 vm_fault_quick((addr >= bp->b_data) ? addr : bp->b_data, 490 (bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 491 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 492 if (pa == 0) 493 panic("vmapbuf: page not present"); 494 m = PHYS_TO_VM_PAGE(pa); 495 vm_page_hold(m); 496 bp->b_pages[pidx] = m; 497 } 498 if (pidx > btoc(MAXPHYS)) 499 panic("vmapbuf: mapped more than MAXPHYS"); 500 pmap_qenter((vm_offset_t)bp->b_saveaddr, bp->b_pages, pidx); 501 502 kva = bp->b_saveaddr; 503 bp->b_npages = pidx; 504 bp->b_saveaddr = bp->b_data; 505 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 506} 507 508/* 509 * Free the io map PTEs associated with this IO operation. 510 * We also invalidate the TLB entries and restore the original b_addr. 511 */ 512void 513vunmapbuf(bp) 514 register struct buf *bp; 515{ 516 int pidx; 517 int npages; 518 vm_page_t *m; 519 520 GIANT_REQUIRED; 521 522 if ((bp->b_flags & B_PHYS) == 0) 523 panic("vunmapbuf"); 524 525 npages = bp->b_npages; 526 pmap_qremove(trunc_page((vm_offset_t)bp->b_data), 527 npages); 528 m = bp->b_pages; 529 for (pidx = 0; pidx < npages; pidx++) 530 vm_page_unhold(*m++); 531 532 bp->b_data = bp->b_saveaddr; 533} 534 535/* 536 * Force reset the processor by invalidating the entire address space! 537 */ 538 539#ifdef SMP 540static void 541cpu_reset_proxy() 542{ 543 544 cpu_reset_proxy_active = 1; 545 while (cpu_reset_proxy_active == 1) 546 ; /* Wait for other cpu to see that we've started */ 547 stop_cpus((1<<cpu_reset_proxyid)); 548 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 549 DELAY(1000000); 550 cpu_reset_real(); 551} 552#endif 553 554void 555cpu_reset() 556{ 557#ifdef SMP 558 if (smp_active == 0) { 559 cpu_reset_real(); 560 /* NOTREACHED */ 561 } else { 562 563 u_int map; 564 int cnt; 565 printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid)); 566 567 map = PCPU_GET(other_cpus) & ~ stopped_cpus; 568 569 if (map != 0) { 570 printf("cpu_reset: Stopping other CPUs\n"); 571 stop_cpus(map); /* Stop all other CPUs */ 572 } 573 574 if (PCPU_GET(cpuid) == 0) { 575 DELAY(1000000); 576 cpu_reset_real(); 577 /* NOTREACHED */ 578 } else { 579 /* We are not BSP (CPU #0) */ 580 581 cpu_reset_proxyid = PCPU_GET(cpuid); 582 cpustop_restartfunc = cpu_reset_proxy; 583 cpu_reset_proxy_active = 0; 584 printf("cpu_reset: Restarting BSP\n"); 585 started_cpus = (1<<0); /* Restart CPU #0 */ 586 587 cnt = 0; 588 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 589 cnt++; /* Wait for BSP to announce restart */ 590 if (cpu_reset_proxy_active == 0) 591 printf("cpu_reset: Failed to restart BSP\n"); 592 enable_intr(); 593 cpu_reset_proxy_active = 2; 594 595 while (1); 596 /* NOTREACHED */ 597 } 598 } 599#else 600 cpu_reset_real(); 601#endif 602} 603 604static void 605cpu_reset_real() 606{ 607 608#ifdef PC98 609 /* 610 * Attempt to do a CPU reset via CPU reset port. 611 */ 612 disable_intr(); 613 if ((inb(0x35) & 0xa0) != 0xa0) { 614 outb(0x37, 0x0f); /* SHUT0 = 0. */ 615 outb(0x37, 0x0b); /* SHUT1 = 0. */ 616 } 617 outb(0xf0, 0x00); /* Reset. */ 618#else 619 /* 620 * Attempt to do a CPU reset via the keyboard controller, 621 * do not turn of the GateA20, as any machine that fails 622 * to do the reset here would then end up in no man's land. 623 */ 624 625#if !defined(BROKEN_KEYBOARD_RESET) 626 outb(IO_KBD + 4, 0xFE); 627 DELAY(500000); /* wait 0.5 sec to see if that did it */ 628 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 629 DELAY(1000000); /* wait 1 sec for printf to complete */ 630#endif 631#endif /* PC98 */ 632 /* force a shutdown by unmapping entire address space ! */ 633 bzero((caddr_t) PTD, PAGE_SIZE); 634 635 /* "good night, sweet prince .... <THUNK!>" */ 636 invltlb(); 637 /* NOTREACHED */ 638 while(1); 639} 640 641/* 642 * Software interrupt handler for queued VM system processing. 643 */ 644void 645swi_vm(void *dummy) 646{ 647 if (busdma_swi_pending != 0) 648 busdma_swi(); 649} 650 651/* 652 * Tell whether this address is in some physical memory region. 653 * Currently used by the kernel coredump code in order to avoid 654 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 655 * or other unpredictable behaviour. 656 */ 657 658int 659is_physical_memory(addr) 660 vm_offset_t addr; 661{ 662 663#ifdef DEV_ISA 664 /* The ISA ``memory hole''. */ 665 if (addr >= 0xa0000 && addr < 0x100000) 666 return 0; 667#endif 668 669 /* 670 * stuff other tests for known memory-mapped devices (PCI?) 671 * here 672 */ 673 674 return 1; 675} 676