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 */ 42 43 #include <sys/cdefs.h> 44 __FBSDID("$FreeBSD$"); 45 46 #include "opt_isa.h" 47 #include "opt_npx.h" 48 #include "opt_reset.h" 49 #include "opt_cpu.h" 50 #include "opt_xbox.h" 51 52 #include <sys/param.h> 53 #include <sys/systm.h> 54 #include <sys/bio.h> 55 #include <sys/buf.h> 56 #include <sys/kernel.h> 57 #include <sys/ktr.h> 58 #include <sys/lock.h> 59 #include <sys/malloc.h> 60 #include <sys/mbuf.h> 61 #include <sys/mutex.h> 62 #include <sys/pioctl.h> 63 #include <sys/proc.h> 64 #include <sys/sysent.h> 65 #include <sys/sf_buf.h> 66 #include <sys/smp.h> 67 #include <sys/sched.h> 68 #include <sys/sysctl.h> 69 #include <sys/unistd.h> 70 #include <sys/vnode.h> 71 #include <sys/vmmeter.h> 72 73 #include <machine/cpu.h> 74 #include <machine/cputypes.h> 75 #include <machine/md_var.h> 76 #include <machine/pcb.h> 77 #include <machine/pcb_ext.h> 78 #include <machine/smp.h> 79 #include <machine/vm86.h> 80 81 #ifdef CPU_ELAN 82 #include <machine/elan_mmcr.h> 83 #endif 84 85 #include <vm/vm.h> 86 #include <vm/vm_extern.h> 87 #include <vm/vm_kern.h> 88 #include <vm/vm_page.h> 89 #include <vm/vm_map.h> 90 #include <vm/vm_param.h> 91 92 #ifdef XEN 93 #include <xen/hypervisor.h> 94 #endif 95 #ifdef PC98 96 #include <pc98/cbus/cbus.h> 97 #else 98 #include <x86/isa/isa.h> 99 #endif 100 101 #ifdef XBOX 102 #include <machine/xbox.h> 103 #endif 104 105 #ifndef NSFBUFS 106 #define NSFBUFS (512 + maxusers * 16) 107 #endif 108 109 CTASSERT((struct thread **)OFFSETOF_CURTHREAD == 110 &((struct pcpu *)NULL)->pc_curthread); 111 CTASSERT((struct pcb **)OFFSETOF_CURPCB == &((struct pcpu *)NULL)->pc_curpcb); 112 113 static void cpu_reset_real(void); 114 #ifdef SMP 115 static void cpu_reset_proxy(void); 116 static u_int cpu_reset_proxyid; 117 static volatile u_int cpu_reset_proxy_active; 118 #endif 119 static void sf_buf_init(void *arg); 120 SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL); 121 122 LIST_HEAD(sf_head, sf_buf); 123 124 /* 125 * A hash table of active sendfile(2) buffers 126 */ 127 static struct sf_head *sf_buf_active; 128 static u_long sf_buf_hashmask; 129 130 #define SF_BUF_HASH(m) (((m) - vm_page_array) & sf_buf_hashmask) 131 132 static TAILQ_HEAD(, sf_buf) sf_buf_freelist; 133 static u_int sf_buf_alloc_want; 134 135 /* 136 * A lock used to synchronize access to the hash table and free list 137 */ 138 static struct mtx sf_buf_lock; 139 140 extern int _ucodesel, _udatasel; 141 142 /* 143 * Finish a fork operation, with process p2 nearly set up. 144 * Copy and update the pcb, set up the stack so that the child 145 * ready to run and return to user mode. 146 */ 147 void 148 cpu_fork(td1, p2, td2, flags) 149 register struct thread *td1; 150 register struct proc *p2; 151 struct thread *td2; 152 int flags; 153 { 154 register struct proc *p1; 155 struct pcb *pcb2; 156 struct mdproc *mdp2; 157 158 p1 = td1->td_proc; 159 if ((flags & RFPROC) == 0) { 160 if ((flags & RFMEM) == 0) { 161 /* unshare user LDT */ 162 struct mdproc *mdp1 = &p1->p_md; 163 struct proc_ldt *pldt, *pldt1; 164 165 mtx_lock_spin(&dt_lock); 166 if ((pldt1 = mdp1->md_ldt) != NULL && 167 pldt1->ldt_refcnt > 1) { 168 pldt = user_ldt_alloc(mdp1, pldt1->ldt_len); 169 if (pldt == NULL) 170 panic("could not copy LDT"); 171 mdp1->md_ldt = pldt; 172 set_user_ldt(mdp1); 173 user_ldt_deref(pldt1); 174 } else 175 mtx_unlock_spin(&dt_lock); 176 } 177 return; 178 } 179 180 /* Ensure that td1's pcb is up to date. */ 181 if (td1 == curthread) 182 td1->td_pcb->pcb_gs = rgs(); 183 #ifdef DEV_NPX 184 critical_enter(); 185 if (PCPU_GET(fpcurthread) == td1) 186 npxsave(td1->td_pcb->pcb_save); 187 critical_exit(); 188 #endif 189 190 /* Point the pcb to the top of the stack */ 191 pcb2 = (struct pcb *)(td2->td_kstack + 192 td2->td_kstack_pages * PAGE_SIZE) - 1; 193 td2->td_pcb = pcb2; 194 195 /* Copy td1's pcb */ 196 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 197 198 /* Properly initialize pcb_save */ 199 pcb2->pcb_save = &pcb2->pcb_user_save; 200 201 /* Point mdproc and then copy over td1's contents */ 202 mdp2 = &p2->p_md; 203 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 204 205 /* 206 * Create a new fresh stack for the new process. 207 * Copy the trap frame for the return to user mode as if from a 208 * syscall. This copies most of the user mode register values. 209 * The -16 is so we can expand the trapframe if we go to vm86. 210 */ 211 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1; 212 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 213 214 td2->td_frame->tf_eax = 0; /* Child returns zero */ 215 td2->td_frame->tf_eflags &= ~PSL_C; /* success */ 216 td2->td_frame->tf_edx = 1; 217 218 /* 219 * If the parent process has the trap bit set (i.e. a debugger had 220 * single stepped the process to the system call), we need to clear 221 * the trap flag from the new frame unless the debugger had set PF_FORK 222 * on the parent. Otherwise, the child will receive a (likely 223 * unexpected) SIGTRAP when it executes the first instruction after 224 * returning to userland. 225 */ 226 if ((p1->p_pfsflags & PF_FORK) == 0) 227 td2->td_frame->tf_eflags &= ~PSL_T; 228 229 /* 230 * Set registers for trampoline to user mode. Leave space for the 231 * return address on stack. These are the kernel mode register values. 232 */ 233 #ifdef PAE 234 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt); 235 #else 236 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir); 237 #endif 238 pcb2->pcb_edi = 0; 239 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */ 240 pcb2->pcb_ebp = 0; 241 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *); 242 pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */ 243 pcb2->pcb_eip = (int)fork_trampoline; 244 pcb2->pcb_psl = PSL_KERNEL; /* ints disabled */ 245 /*- 246 * pcb2->pcb_dr*: cloned above. 247 * pcb2->pcb_savefpu: cloned above. 248 * pcb2->pcb_flags: cloned above. 249 * pcb2->pcb_onfault: cloned above (always NULL here?). 250 * pcb2->pcb_gs: cloned above. 251 * pcb2->pcb_ext: cleared below. 252 */ 253 254 /* 255 * XXX don't copy the i/o pages. this should probably be fixed. 256 */ 257 pcb2->pcb_ext = 0; 258 259 /* Copy the LDT, if necessary. */ 260 mtx_lock_spin(&dt_lock); 261 if (mdp2->md_ldt != NULL) { 262 if (flags & RFMEM) { 263 mdp2->md_ldt->ldt_refcnt++; 264 } else { 265 mdp2->md_ldt = user_ldt_alloc(mdp2, 266 mdp2->md_ldt->ldt_len); 267 if (mdp2->md_ldt == NULL) 268 panic("could not copy LDT"); 269 } 270 } 271 mtx_unlock_spin(&dt_lock); 272 273 /* Setup to release spin count in fork_exit(). */ 274 td2->td_md.md_spinlock_count = 1; 275 /* 276 * XXX XEN need to check on PSL_USER is handled 277 */ 278 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 279 /* 280 * Now, cpu_switch() can schedule the new process. 281 * pcb_esp is loaded pointing to the cpu_switch() stack frame 282 * containing the return address when exiting cpu_switch. 283 * This will normally be to fork_trampoline(), which will have 284 * %ebx loaded with the new proc's pointer. fork_trampoline() 285 * will set up a stack to call fork_return(p, frame); to complete 286 * the return to user-mode. 287 */ 288 } 289 290 /* 291 * Intercept the return address from a freshly forked process that has NOT 292 * been scheduled yet. 293 * 294 * This is needed to make kernel threads stay in kernel mode. 295 */ 296 void 297 cpu_set_fork_handler(td, func, arg) 298 struct thread *td; 299 void (*func)(void *); 300 void *arg; 301 { 302 /* 303 * Note that the trap frame follows the args, so the function 304 * is really called like this: func(arg, frame); 305 */ 306 td->td_pcb->pcb_esi = (int) func; /* function */ 307 td->td_pcb->pcb_ebx = (int) arg; /* first arg */ 308 } 309 310 void 311 cpu_exit(struct thread *td) 312 { 313 314 /* 315 * If this process has a custom LDT, release it. Reset pc->pcb_gs 316 * and %gs before we free it in case they refer to an LDT entry. 317 */ 318 mtx_lock_spin(&dt_lock); 319 if (td->td_proc->p_md.md_ldt) { 320 td->td_pcb->pcb_gs = _udatasel; 321 load_gs(_udatasel); 322 user_ldt_free(td); 323 } else 324 mtx_unlock_spin(&dt_lock); 325 } 326 327 void 328 cpu_thread_exit(struct thread *td) 329 { 330 331 #ifdef DEV_NPX 332 critical_enter(); 333 if (td == PCPU_GET(fpcurthread)) 334 npxdrop(); 335 critical_exit(); 336 #endif 337 338 /* Disable any hardware breakpoints. */ 339 if (td->td_pcb->pcb_flags & PCB_DBREGS) { 340 reset_dbregs(); 341 td->td_pcb->pcb_flags &= ~PCB_DBREGS; 342 } 343 } 344 345 void 346 cpu_thread_clean(struct thread *td) 347 { 348 struct pcb *pcb; 349 350 pcb = td->td_pcb; 351 if (pcb->pcb_ext != NULL) { 352 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */ 353 /* 354 * XXX do we need to move the TSS off the allocated pages 355 * before freeing them? (not done here) 356 */ 357 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 358 ctob(IOPAGES + 1)); 359 pcb->pcb_ext = NULL; 360 } 361 } 362 363 void 364 cpu_thread_swapin(struct thread *td) 365 { 366 } 367 368 void 369 cpu_thread_swapout(struct thread *td) 370 { 371 } 372 373 void 374 cpu_thread_alloc(struct thread *td) 375 { 376 377 td->td_pcb = (struct pcb *)(td->td_kstack + 378 td->td_kstack_pages * PAGE_SIZE) - 1; 379 td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb - 16) - 1; 380 td->td_pcb->pcb_ext = NULL; 381 td->td_pcb->pcb_save = &td->td_pcb->pcb_user_save; 382 } 383 384 void 385 cpu_thread_free(struct thread *td) 386 { 387 388 cpu_thread_clean(td); 389 } 390 391 void 392 cpu_set_syscall_retval(struct thread *td, int error) 393 { 394 395 switch (error) { 396 case 0: 397 td->td_frame->tf_eax = td->td_retval[0]; 398 td->td_frame->tf_edx = td->td_retval[1]; 399 td->td_frame->tf_eflags &= ~PSL_C; 400 break; 401 402 case ERESTART: 403 /* 404 * Reconstruct pc, assuming lcall $X,y is 7 bytes, int 405 * 0x80 is 2 bytes. We saved this in tf_err. 406 */ 407 td->td_frame->tf_eip -= td->td_frame->tf_err; 408 break; 409 410 case EJUSTRETURN: 411 break; 412 413 default: 414 if (td->td_proc->p_sysent->sv_errsize) { 415 if (error >= td->td_proc->p_sysent->sv_errsize) 416 error = -1; /* XXX */ 417 else 418 error = td->td_proc->p_sysent->sv_errtbl[error]; 419 } 420 td->td_frame->tf_eax = error; 421 td->td_frame->tf_eflags |= PSL_C; 422 break; 423 } 424 } 425 426 /* 427 * Initialize machine state (pcb and trap frame) for a new thread about to 428 * upcall. Put enough state in the new thread's PCB to get it to go back 429 * userret(), where we can intercept it again to set the return (upcall) 430 * Address and stack, along with those from upcals that are from other sources 431 * such as those generated in thread_userret() itself. 432 */ 433 void 434 cpu_set_upcall(struct thread *td, struct thread *td0) 435 { 436 struct pcb *pcb2; 437 438 /* Point the pcb to the top of the stack. */ 439 pcb2 = td->td_pcb; 440 441 /* 442 * Copy the upcall pcb. This loads kernel regs. 443 * Those not loaded individually below get their default 444 * values here. 445 */ 446 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 447 pcb2->pcb_flags &= ~(PCB_NPXINITDONE | PCB_NPXUSERINITDONE); 448 pcb2->pcb_save = &pcb2->pcb_user_save; 449 450 /* 451 * Create a new fresh stack for the new thread. 452 */ 453 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 454 455 /* If the current thread has the trap bit set (i.e. a debugger had 456 * single stepped the process to the system call), we need to clear 457 * the trap flag from the new frame. Otherwise, the new thread will 458 * receive a (likely unexpected) SIGTRAP when it executes the first 459 * instruction after returning to userland. 460 */ 461 td->td_frame->tf_eflags &= ~PSL_T; 462 463 /* 464 * Set registers for trampoline to user mode. Leave space for the 465 * return address on stack. These are the kernel mode register values. 466 */ 467 pcb2->pcb_edi = 0; 468 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */ 469 pcb2->pcb_ebp = 0; 470 pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */ 471 pcb2->pcb_ebx = (int)td; /* trampoline arg */ 472 pcb2->pcb_eip = (int)fork_trampoline; 473 pcb2->pcb_psl &= ~(PSL_I); /* interrupts must be disabled */ 474 pcb2->pcb_gs = rgs(); 475 /* 476 * If we didn't copy the pcb, we'd need to do the following registers: 477 * pcb2->pcb_cr3: cloned above. 478 * pcb2->pcb_dr*: cloned above. 479 * pcb2->pcb_savefpu: cloned above. 480 * pcb2->pcb_flags: cloned above. 481 * pcb2->pcb_onfault: cloned above (always NULL here?). 482 * pcb2->pcb_gs: cloned above. 483 * pcb2->pcb_ext: cleared below. 484 */ 485 pcb2->pcb_ext = NULL; 486 487 /* Setup to release spin count in fork_exit(). */ 488 td->td_md.md_spinlock_count = 1; 489 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 490 } 491 492 /* 493 * Set that machine state for performing an upcall that has to 494 * be done in thread_userret() so that those upcalls generated 495 * in thread_userret() itself can be done as well. 496 */ 497 void 498 cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg, 499 stack_t *stack) 500 { 501 502 /* 503 * Do any extra cleaning that needs to be done. 504 * The thread may have optional components 505 * that are not present in a fresh thread. 506 * This may be a recycled thread so make it look 507 * as though it's newly allocated. 508 */ 509 cpu_thread_clean(td); 510 511 /* 512 * Set the trap frame to point at the beginning of the uts 513 * function. 514 */ 515 td->td_frame->tf_ebp = 0; 516 td->td_frame->tf_esp = 517 (((int)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4; 518 td->td_frame->tf_eip = (int)entry; 519 520 /* 521 * Pass the address of the mailbox for this kse to the uts 522 * function as a parameter on the stack. 523 */ 524 suword((void *)(td->td_frame->tf_esp + sizeof(void *)), 525 (int)arg); 526 } 527 528 int 529 cpu_set_user_tls(struct thread *td, void *tls_base) 530 { 531 struct segment_descriptor sd; 532 uint32_t base; 533 534 /* 535 * Construct a descriptor and store it in the pcb for 536 * the next context switch. Also store it in the gdt 537 * so that the load of tf_fs into %fs will activate it 538 * at return to userland. 539 */ 540 base = (uint32_t)tls_base; 541 sd.sd_lobase = base & 0xffffff; 542 sd.sd_hibase = (base >> 24) & 0xff; 543 sd.sd_lolimit = 0xffff; /* 4GB limit, wraps around */ 544 sd.sd_hilimit = 0xf; 545 sd.sd_type = SDT_MEMRWA; 546 sd.sd_dpl = SEL_UPL; 547 sd.sd_p = 1; 548 sd.sd_xx = 0; 549 sd.sd_def32 = 1; 550 sd.sd_gran = 1; 551 critical_enter(); 552 /* set %gs */ 553 td->td_pcb->pcb_gsd = sd; 554 if (td == curthread) { 555 PCPU_GET(fsgs_gdt)[1] = sd; 556 load_gs(GSEL(GUGS_SEL, SEL_UPL)); 557 } 558 critical_exit(); 559 return (0); 560 } 561 562 /* 563 * Convert kernel VA to physical address 564 */ 565 vm_paddr_t 566 kvtop(void *addr) 567 { 568 vm_paddr_t pa; 569 570 pa = pmap_kextract((vm_offset_t)addr); 571 if (pa == 0) 572 panic("kvtop: zero page frame"); 573 return (pa); 574 } 575 576 #ifdef SMP 577 static void 578 cpu_reset_proxy() 579 { 580 cpuset_t tcrp; 581 582 cpu_reset_proxy_active = 1; 583 while (cpu_reset_proxy_active == 1) 584 ; /* Wait for other cpu to see that we've started */ 585 CPU_SETOF(cpu_reset_proxyid, &tcrp); 586 stop_cpus(tcrp); 587 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 588 DELAY(1000000); 589 cpu_reset_real(); 590 } 591 #endif 592 593 void 594 cpu_reset() 595 { 596 #ifdef XBOX 597 if (arch_i386_is_xbox) { 598 /* Kick the PIC16L, it can reboot the box */ 599 pic16l_reboot(); 600 for (;;); 601 } 602 #endif 603 604 #ifdef SMP 605 cpuset_t map; 606 u_int cnt; 607 608 if (smp_active) { 609 map = all_cpus; 610 CPU_CLR(PCPU_GET(cpuid), &map); 611 CPU_NAND(&map, &stopped_cpus); 612 if (!CPU_EMPTY(&map)) { 613 printf("cpu_reset: Stopping other CPUs\n"); 614 stop_cpus(map); 615 } 616 617 if (PCPU_GET(cpuid) != 0) { 618 cpu_reset_proxyid = PCPU_GET(cpuid); 619 cpustop_restartfunc = cpu_reset_proxy; 620 cpu_reset_proxy_active = 0; 621 printf("cpu_reset: Restarting BSP\n"); 622 623 /* Restart CPU #0. */ 624 /* XXX: restart_cpus(1 << 0); */ 625 CPU_SETOF(0, &started_cpus); 626 wmb(); 627 628 cnt = 0; 629 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 630 cnt++; /* Wait for BSP to announce restart */ 631 if (cpu_reset_proxy_active == 0) 632 printf("cpu_reset: Failed to restart BSP\n"); 633 enable_intr(); 634 cpu_reset_proxy_active = 2; 635 636 while (1); 637 /* NOTREACHED */ 638 } 639 640 DELAY(1000000); 641 } 642 #endif 643 cpu_reset_real(); 644 /* NOTREACHED */ 645 } 646 647 static void 648 cpu_reset_real() 649 { 650 struct region_descriptor null_idt; 651 #ifndef PC98 652 int b; 653 #endif 654 655 disable_intr(); 656 #ifdef XEN 657 if (smp_processor_id() == 0) 658 HYPERVISOR_shutdown(SHUTDOWN_reboot); 659 else 660 HYPERVISOR_shutdown(SHUTDOWN_poweroff); 661 #endif 662 #ifdef CPU_ELAN 663 if (elan_mmcr != NULL) 664 elan_mmcr->RESCFG = 1; 665 #endif 666 667 if (cpu == CPU_GEODE1100) { 668 /* Attempt Geode's own reset */ 669 outl(0xcf8, 0x80009044ul); 670 outl(0xcfc, 0xf); 671 } 672 673 #ifdef PC98 674 /* 675 * Attempt to do a CPU reset via CPU reset port. 676 */ 677 if ((inb(0x35) & 0xa0) != 0xa0) { 678 outb(0x37, 0x0f); /* SHUT0 = 0. */ 679 outb(0x37, 0x0b); /* SHUT1 = 0. */ 680 } 681 outb(0xf0, 0x00); /* Reset. */ 682 #else 683 #if !defined(BROKEN_KEYBOARD_RESET) 684 /* 685 * Attempt to do a CPU reset via the keyboard controller, 686 * do not turn off GateA20, as any machine that fails 687 * to do the reset here would then end up in no man's land. 688 */ 689 outb(IO_KBD + 4, 0xFE); 690 DELAY(500000); /* wait 0.5 sec to see if that did it */ 691 #endif 692 693 /* 694 * Attempt to force a reset via the Reset Control register at 695 * I/O port 0xcf9. Bit 2 forces a system reset when it 696 * transitions from 0 to 1. Bit 1 selects the type of reset 697 * to attempt: 0 selects a "soft" reset, and 1 selects a 698 * "hard" reset. We try a "hard" reset. The first write sets 699 * bit 1 to select a "hard" reset and clears bit 2. The 700 * second write forces a 0 -> 1 transition in bit 2 to trigger 701 * a reset. 702 */ 703 outb(0xcf9, 0x2); 704 outb(0xcf9, 0x6); 705 DELAY(500000); /* wait 0.5 sec to see if that did it */ 706 707 /* 708 * Attempt to force a reset via the Fast A20 and Init register 709 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate. 710 * Bit 0 asserts INIT# when set to 1. We are careful to only 711 * preserve bit 1 while setting bit 0. We also must clear bit 712 * 0 before setting it if it isn't already clear. 713 */ 714 b = inb(0x92); 715 if (b != 0xff) { 716 if ((b & 0x1) != 0) 717 outb(0x92, b & 0xfe); 718 outb(0x92, b | 0x1); 719 DELAY(500000); /* wait 0.5 sec to see if that did it */ 720 } 721 #endif /* PC98 */ 722 723 printf("No known reset method worked, attempting CPU shutdown\n"); 724 DELAY(1000000); /* wait 1 sec for printf to complete */ 725 726 /* Wipe the IDT. */ 727 null_idt.rd_limit = 0; 728 null_idt.rd_base = 0; 729 lidt(&null_idt); 730 731 /* "good night, sweet prince .... <THUNK!>" */ 732 breakpoint(); 733 734 /* NOTREACHED */ 735 while(1); 736 } 737 738 /* 739 * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-)) 740 */ 741 static void 742 sf_buf_init(void *arg) 743 { 744 struct sf_buf *sf_bufs; 745 vm_offset_t sf_base; 746 int i; 747 748 nsfbufs = NSFBUFS; 749 TUNABLE_INT_FETCH("kern.ipc.nsfbufs", &nsfbufs); 750 751 sf_buf_active = hashinit(nsfbufs, M_TEMP, &sf_buf_hashmask); 752 TAILQ_INIT(&sf_buf_freelist); 753 sf_base = kmem_alloc_nofault(kernel_map, nsfbufs * PAGE_SIZE); 754 sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP, 755 M_NOWAIT | M_ZERO); 756 for (i = 0; i < nsfbufs; i++) { 757 sf_bufs[i].kva = sf_base + i * PAGE_SIZE; 758 TAILQ_INSERT_TAIL(&sf_buf_freelist, &sf_bufs[i], free_entry); 759 } 760 sf_buf_alloc_want = 0; 761 mtx_init(&sf_buf_lock, "sf_buf", NULL, MTX_DEF); 762 } 763 764 /* 765 * Invalidate the cache lines that may belong to the page, if 766 * (possibly old) mapping of the page by sf buffer exists. Returns 767 * TRUE when mapping was found and cache invalidated. 768 */ 769 boolean_t 770 sf_buf_invalidate_cache(vm_page_t m) 771 { 772 struct sf_head *hash_list; 773 struct sf_buf *sf; 774 boolean_t ret; 775 776 hash_list = &sf_buf_active[SF_BUF_HASH(m)]; 777 ret = FALSE; 778 mtx_lock(&sf_buf_lock); 779 LIST_FOREACH(sf, hash_list, list_entry) { 780 if (sf->m == m) { 781 /* 782 * Use pmap_qenter to update the pte for 783 * existing mapping, in particular, the PAT 784 * settings are recalculated. 785 */ 786 pmap_qenter(sf->kva, &m, 1); 787 pmap_invalidate_cache_range(sf->kva, sf->kva + 788 PAGE_SIZE); 789 ret = TRUE; 790 break; 791 } 792 } 793 mtx_unlock(&sf_buf_lock); 794 return (ret); 795 } 796 797 /* 798 * Get an sf_buf from the freelist. May block if none are available. 799 */ 800 struct sf_buf * 801 sf_buf_alloc(struct vm_page *m, int flags) 802 { 803 pt_entry_t opte, *ptep; 804 struct sf_head *hash_list; 805 struct sf_buf *sf; 806 #ifdef SMP 807 cpuset_t other_cpus; 808 u_int cpuid; 809 #endif 810 int error; 811 812 KASSERT(curthread->td_pinned > 0 || (flags & SFB_CPUPRIVATE) == 0, 813 ("sf_buf_alloc(SFB_CPUPRIVATE): curthread not pinned")); 814 hash_list = &sf_buf_active[SF_BUF_HASH(m)]; 815 mtx_lock(&sf_buf_lock); 816 LIST_FOREACH(sf, hash_list, list_entry) { 817 if (sf->m == m) { 818 sf->ref_count++; 819 if (sf->ref_count == 1) { 820 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry); 821 nsfbufsused++; 822 nsfbufspeak = imax(nsfbufspeak, nsfbufsused); 823 } 824 #ifdef SMP 825 goto shootdown; 826 #else 827 goto done; 828 #endif 829 } 830 } 831 while ((sf = TAILQ_FIRST(&sf_buf_freelist)) == NULL) { 832 if (flags & SFB_NOWAIT) 833 goto done; 834 sf_buf_alloc_want++; 835 mbstat.sf_allocwait++; 836 error = msleep(&sf_buf_freelist, &sf_buf_lock, 837 (flags & SFB_CATCH) ? PCATCH | PVM : PVM, "sfbufa", 0); 838 sf_buf_alloc_want--; 839 840 /* 841 * If we got a signal, don't risk going back to sleep. 842 */ 843 if (error) 844 goto done; 845 } 846 TAILQ_REMOVE(&sf_buf_freelist, sf, free_entry); 847 if (sf->m != NULL) 848 LIST_REMOVE(sf, list_entry); 849 LIST_INSERT_HEAD(hash_list, sf, list_entry); 850 sf->ref_count = 1; 851 sf->m = m; 852 nsfbufsused++; 853 nsfbufspeak = imax(nsfbufspeak, nsfbufsused); 854 855 /* 856 * Update the sf_buf's virtual-to-physical mapping, flushing the 857 * virtual address from the TLB. Since the reference count for 858 * the sf_buf's old mapping was zero, that mapping is not 859 * currently in use. Consequently, there is no need to exchange 860 * the old and new PTEs atomically, even under PAE. 861 */ 862 ptep = vtopte(sf->kva); 863 opte = *ptep; 864 #ifdef XEN 865 PT_SET_MA(sf->kva, xpmap_ptom(VM_PAGE_TO_PHYS(m)) | pgeflag 866 | PG_RW | PG_V | pmap_cache_bits(m->md.pat_mode, 0)); 867 #else 868 *ptep = VM_PAGE_TO_PHYS(m) | pgeflag | PG_RW | PG_V | 869 pmap_cache_bits(m->md.pat_mode, 0); 870 #endif 871 872 /* 873 * Avoid unnecessary TLB invalidations: If the sf_buf's old 874 * virtual-to-physical mapping was not used, then any processor 875 * that has invalidated the sf_buf's virtual address from its TLB 876 * since the last used mapping need not invalidate again. 877 */ 878 #ifdef SMP 879 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A)) 880 CPU_ZERO(&sf->cpumask); 881 shootdown: 882 sched_pin(); 883 cpuid = PCPU_GET(cpuid); 884 if (!CPU_ISSET(cpuid, &sf->cpumask)) { 885 CPU_SET(cpuid, &sf->cpumask); 886 invlpg(sf->kva); 887 } 888 if ((flags & SFB_CPUPRIVATE) == 0) { 889 other_cpus = all_cpus; 890 CPU_CLR(cpuid, &other_cpus); 891 CPU_NAND(&other_cpus, &sf->cpumask); 892 if (!CPU_EMPTY(&other_cpus)) { 893 CPU_OR(&sf->cpumask, &other_cpus); 894 smp_masked_invlpg(other_cpus, sf->kva); 895 } 896 } 897 sched_unpin(); 898 #else 899 if ((opte & (PG_V | PG_A)) == (PG_V | PG_A)) 900 pmap_invalidate_page(kernel_pmap, sf->kva); 901 #endif 902 done: 903 mtx_unlock(&sf_buf_lock); 904 return (sf); 905 } 906 907 /* 908 * Remove a reference from the given sf_buf, adding it to the free 909 * list when its reference count reaches zero. A freed sf_buf still, 910 * however, retains its virtual-to-physical mapping until it is 911 * recycled or reactivated by sf_buf_alloc(9). 912 */ 913 void 914 sf_buf_free(struct sf_buf *sf) 915 { 916 917 mtx_lock(&sf_buf_lock); 918 sf->ref_count--; 919 if (sf->ref_count == 0) { 920 TAILQ_INSERT_TAIL(&sf_buf_freelist, sf, free_entry); 921 nsfbufsused--; 922 #ifdef XEN 923 /* 924 * Xen doesn't like having dangling R/W mappings 925 */ 926 pmap_qremove(sf->kva, 1); 927 sf->m = NULL; 928 LIST_REMOVE(sf, list_entry); 929 #endif 930 if (sf_buf_alloc_want > 0) 931 wakeup(&sf_buf_freelist); 932 } 933 mtx_unlock(&sf_buf_lock); 934 } 935 936 /* 937 * Software interrupt handler for queued VM system processing. 938 */ 939 void 940 swi_vm(void *dummy) 941 { 942 if (busdma_swi_pending != 0) 943 busdma_swi(); 944 } 945 946 /* 947 * Tell whether this address is in some physical memory region. 948 * Currently used by the kernel coredump code in order to avoid 949 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 950 * or other unpredictable behaviour. 951 */ 952 953 int 954 is_physical_memory(vm_paddr_t addr) 955 { 956 957 #ifdef DEV_ISA 958 /* The ISA ``memory hole''. */ 959 if (addr >= 0xa0000 && addr < 0x100000) 960 return 0; 961 #endif 962 963 /* 964 * stuff other tests for known memory-mapped devices (PCI?) 965 * here 966 */ 967 968 return 1; 969 } 970