vm_machdep.c revision 199135
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: head/sys/amd64/amd64/vm_machdep.c 199135 2009-11-10 11:43:07Z kib $"); 45 46#include "opt_isa.h" 47#include "opt_cpu.h" 48#include "opt_compat.h" 49 50#include <sys/param.h> 51#include <sys/systm.h> 52#include <sys/bio.h> 53#include <sys/buf.h> 54#include <sys/kernel.h> 55#include <sys/ktr.h> 56#include <sys/lock.h> 57#include <sys/malloc.h> 58#include <sys/mbuf.h> 59#include <sys/mutex.h> 60#include <sys/pioctl.h> 61#include <sys/proc.h> 62#include <sys/sf_buf.h> 63#include <sys/smp.h> 64#include <sys/sysctl.h> 65#include <sys/sysent.h> 66#include <sys/unistd.h> 67#include <sys/vnode.h> 68#include <sys/vmmeter.h> 69 70#include <machine/cpu.h> 71#include <machine/md_var.h> 72#include <machine/pcb.h> 73#include <machine/specialreg.h> 74#include <machine/tss.h> 75 76#include <vm/vm.h> 77#include <vm/vm_extern.h> 78#include <vm/vm_kern.h> 79#include <vm/vm_page.h> 80#include <vm/vm_map.h> 81#include <vm/vm_param.h> 82 83#include <amd64/isa/isa.h> 84 85static void cpu_reset_real(void); 86#ifdef SMP 87static void cpu_reset_proxy(void); 88static u_int cpu_reset_proxyid; 89static volatile u_int cpu_reset_proxy_active; 90#endif 91 92/* 93 * Finish a fork operation, with process p2 nearly set up. 94 * Copy and update the pcb, set up the stack so that the child 95 * ready to run and return to user mode. 96 */ 97void 98cpu_fork(td1, p2, td2, flags) 99 register struct thread *td1; 100 register struct proc *p2; 101 struct thread *td2; 102 int flags; 103{ 104 register struct proc *p1; 105 struct pcb *pcb2; 106 struct mdproc *mdp1, *mdp2; 107 struct proc_ldt *pldt; 108 pmap_t pmap2; 109 110 p1 = td1->td_proc; 111 if ((flags & RFPROC) == 0) { 112 if ((flags & RFMEM) == 0) { 113 /* unshare user LDT */ 114 mdp1 = &p1->p_md; 115 mtx_lock(&dt_lock); 116 if ((pldt = mdp1->md_ldt) != NULL && 117 pldt->ldt_refcnt > 1 && 118 user_ldt_alloc(p1, 1) == NULL) 119 panic("could not copy LDT"); 120 mtx_unlock(&dt_lock); 121 } 122 return; 123 } 124 125 /* Ensure that p1's pcb is up to date. */ 126 fpuexit(td1); 127 128 /* Point the pcb to the top of the stack */ 129 pcb2 = (struct pcb *)(td2->td_kstack + 130 td2->td_kstack_pages * PAGE_SIZE) - 1; 131 td2->td_pcb = pcb2; 132 133 /* Copy p1's pcb */ 134 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 135 136 /* Point mdproc and then copy over td1's contents */ 137 mdp2 = &p2->p_md; 138 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 139 140 /* 141 * Create a new fresh stack for the new process. 142 * Copy the trap frame for the return to user mode as if from a 143 * syscall. This copies most of the user mode register values. 144 */ 145 td2->td_frame = (struct trapframe *)td2->td_pcb - 1; 146 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 147 148 td2->td_frame->tf_rax = 0; /* Child returns zero */ 149 td2->td_frame->tf_rflags &= ~PSL_C; /* success */ 150 td2->td_frame->tf_rdx = 1; 151 152 /* 153 * If the parent process has the trap bit set (i.e. a debugger had 154 * single stepped the process to the system call), we need to clear 155 * the trap flag from the new frame unless the debugger had set PF_FORK 156 * on the parent. Otherwise, the child will receive a (likely 157 * unexpected) SIGTRAP when it executes the first instruction after 158 * returning to userland. 159 */ 160 if ((p1->p_pfsflags & PF_FORK) == 0) 161 td2->td_frame->tf_rflags &= ~PSL_T; 162 163 /* 164 * Set registers for trampoline to user mode. Leave space for the 165 * return address on stack. These are the kernel mode register values. 166 */ 167 pmap2 = vmspace_pmap(p2->p_vmspace); 168 pcb2->pcb_cr3 = DMAP_TO_PHYS((vm_offset_t)pmap2->pm_pml4); 169 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */ 170 pcb2->pcb_rbp = 0; 171 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *); 172 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */ 173 pcb2->pcb_rip = (register_t)fork_trampoline; 174 /*- 175 * pcb2->pcb_dr*: cloned above. 176 * pcb2->pcb_savefpu: cloned above. 177 * pcb2->pcb_flags: cloned above. 178 * pcb2->pcb_onfault: cloned above (always NULL here?). 179 * pcb2->pcb_[fg]sbase: cloned above 180 */ 181 182 /* Setup to release spin count in fork_exit(). */ 183 td2->td_md.md_spinlock_count = 1; 184 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 185 186 /* As an i386, do not copy io permission bitmap. */ 187 pcb2->pcb_tssp = NULL; 188 189 /* New segment registers. */ 190 pcb2->pcb_full_iret = 1; 191 192 /* Copy the LDT, if necessary. */ 193 mdp1 = &td1->td_proc->p_md; 194 mdp2 = &p2->p_md; 195 mtx_lock(&dt_lock); 196 if (mdp1->md_ldt != NULL) { 197 if (flags & RFMEM) { 198 mdp1->md_ldt->ldt_refcnt++; 199 mdp2->md_ldt = mdp1->md_ldt; 200 bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct 201 system_segment_descriptor)); 202 } else { 203 mdp2->md_ldt = NULL; 204 mdp2->md_ldt = user_ldt_alloc(p2, 0); 205 if (mdp2->md_ldt == NULL) 206 panic("could not copy LDT"); 207 amd64_set_ldt_data(td2, 0, max_ldt_segment, 208 (struct user_segment_descriptor *) 209 mdp1->md_ldt->ldt_base); 210 } 211 } else 212 mdp2->md_ldt = NULL; 213 mtx_unlock(&dt_lock); 214 215 /* 216 * Now, cpu_switch() can schedule the new process. 217 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 218 * containing the return address when exiting cpu_switch. 219 * This will normally be to fork_trampoline(), which will have 220 * %ebx loaded with the new proc's pointer. fork_trampoline() 221 * will set up a stack to call fork_return(p, frame); to complete 222 * the return to user-mode. 223 */ 224} 225 226/* 227 * Intercept the return address from a freshly forked process that has NOT 228 * been scheduled yet. 229 * 230 * This is needed to make kernel threads stay in kernel mode. 231 */ 232void 233cpu_set_fork_handler(td, func, arg) 234 struct thread *td; 235 void (*func)(void *); 236 void *arg; 237{ 238 /* 239 * Note that the trap frame follows the args, so the function 240 * is really called like this: func(arg, frame); 241 */ 242 td->td_pcb->pcb_r12 = (long) func; /* function */ 243 td->td_pcb->pcb_rbx = (long) arg; /* first arg */ 244} 245 246void 247cpu_exit(struct thread *td) 248{ 249 250 /* 251 * If this process has a custom LDT, release it. 252 */ 253 mtx_lock(&dt_lock); 254 if (td->td_proc->p_md.md_ldt != 0) 255 user_ldt_free(td); 256 else 257 mtx_unlock(&dt_lock); 258} 259 260void 261cpu_thread_exit(struct thread *td) 262{ 263 struct pcb *pcb; 264 265 if (td == PCPU_GET(fpcurthread)) 266 fpudrop(); 267 268 pcb = td->td_pcb; 269 270 /* Disable any hardware breakpoints. */ 271 if (pcb->pcb_flags & PCB_DBREGS) { 272 reset_dbregs(); 273 pcb->pcb_flags &= ~PCB_DBREGS; 274 } 275} 276 277void 278cpu_thread_clean(struct thread *td) 279{ 280 struct pcb *pcb; 281 282 pcb = td->td_pcb; 283 284 /* 285 * Clean TSS/iomap 286 */ 287 if (pcb->pcb_tssp != NULL) { 288 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_tssp, 289 ctob(IOPAGES + 1)); 290 pcb->pcb_tssp = NULL; 291 } 292} 293 294void 295cpu_thread_swapin(struct thread *td) 296{ 297} 298 299void 300cpu_thread_swapout(struct thread *td) 301{ 302} 303 304void 305cpu_thread_alloc(struct thread *td) 306{ 307 308 td->td_pcb = (struct pcb *)(td->td_kstack + 309 td->td_kstack_pages * PAGE_SIZE) - 1; 310 td->td_frame = (struct trapframe *)td->td_pcb - 1; 311} 312 313void 314cpu_thread_free(struct thread *td) 315{ 316 317 cpu_thread_clean(td); 318} 319 320void 321cpu_set_syscall_retval(struct thread *td, int error) 322{ 323 324 switch (error) { 325 case 0: 326 td->td_frame->tf_rax = td->td_retval[0]; 327 td->td_frame->tf_rdx = td->td_retval[1]; 328 td->td_frame->tf_rflags &= ~PSL_C; 329 break; 330 331 case ERESTART: 332 /* 333 * Reconstruct pc, we know that 'syscall' is 2 bytes. 334 * We have to do a full context restore so that %r10 335 * (which was holding the value of %rcx) is restored 336 * for the next iteration. 337 */ 338 td->td_frame->tf_rip -= td->td_frame->tf_err; 339 td->td_frame->tf_r10 = td->td_frame->tf_rcx; 340 td->td_pcb->pcb_flags |= PCB_FULLCTX; 341 break; 342 343 case EJUSTRETURN: 344 break; 345 346 default: 347 if (td->td_proc->p_sysent->sv_errsize) { 348 if (error >= td->td_proc->p_sysent->sv_errsize) 349 error = -1; /* XXX */ 350 else 351 error = td->td_proc->p_sysent->sv_errtbl[error]; 352 } 353 td->td_frame->tf_rax = error; 354 td->td_frame->tf_rflags |= PSL_C; 355 break; 356 } 357} 358 359/* 360 * Initialize machine state (pcb and trap frame) for a new thread about to 361 * upcall. Put enough state in the new thread's PCB to get it to go back 362 * userret(), where we can intercept it again to set the return (upcall) 363 * Address and stack, along with those from upcals that are from other sources 364 * such as those generated in thread_userret() itself. 365 */ 366void 367cpu_set_upcall(struct thread *td, struct thread *td0) 368{ 369 struct pcb *pcb2; 370 371 /* Point the pcb to the top of the stack. */ 372 pcb2 = td->td_pcb; 373 374 /* 375 * Copy the upcall pcb. This loads kernel regs. 376 * Those not loaded individually below get their default 377 * values here. 378 */ 379 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 380 pcb2->pcb_flags &= ~PCB_FPUINITDONE; 381 pcb2->pcb_full_iret = 1; 382 383 /* 384 * Create a new fresh stack for the new thread. 385 */ 386 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 387 388 /* If the current thread has the trap bit set (i.e. a debugger had 389 * single stepped the process to the system call), we need to clear 390 * the trap flag from the new frame. Otherwise, the new thread will 391 * receive a (likely unexpected) SIGTRAP when it executes the first 392 * instruction after returning to userland. 393 */ 394 td->td_frame->tf_rflags &= ~PSL_T; 395 396 /* 397 * Set registers for trampoline to user mode. Leave space for the 398 * return address on stack. These are the kernel mode register values. 399 */ 400 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */ 401 pcb2->pcb_rbp = 0; 402 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */ 403 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */ 404 pcb2->pcb_rip = (register_t)fork_trampoline; 405 /* 406 * If we didn't copy the pcb, we'd need to do the following registers: 407 * pcb2->pcb_cr3: cloned above. 408 * pcb2->pcb_dr*: cloned above. 409 * pcb2->pcb_savefpu: cloned above. 410 * pcb2->pcb_onfault: cloned above (always NULL here?). 411 * pcb2->pcb_[fg]sbase: cloned above 412 */ 413 414 /* Setup to release spin count in fork_exit(). */ 415 td->td_md.md_spinlock_count = 1; 416 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 417} 418 419/* 420 * Set that machine state for performing an upcall that has to 421 * be done in thread_userret() so that those upcalls generated 422 * in thread_userret() itself can be done as well. 423 */ 424void 425cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg, 426 stack_t *stack) 427{ 428 429 /* 430 * Do any extra cleaning that needs to be done. 431 * The thread may have optional components 432 * that are not present in a fresh thread. 433 * This may be a recycled thread so make it look 434 * as though it's newly allocated. 435 */ 436 cpu_thread_clean(td); 437 438#ifdef COMPAT_IA32 439 if (td->td_proc->p_sysent->sv_flags & SV_ILP32) { 440 /* 441 * Set the trap frame to point at the beginning of the uts 442 * function. 443 */ 444 td->td_frame->tf_rbp = 0; 445 td->td_frame->tf_rsp = 446 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4; 447 td->td_frame->tf_rip = (uintptr_t)entry; 448 449 /* 450 * Pass the address of the mailbox for this kse to the uts 451 * function as a parameter on the stack. 452 */ 453 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)), 454 (uint32_t)(uintptr_t)arg); 455 456 return; 457 } 458#endif 459 460 /* 461 * Set the trap frame to point at the beginning of the uts 462 * function. 463 */ 464 td->td_frame->tf_rbp = 0; 465 td->td_frame->tf_rsp = 466 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f; 467 td->td_frame->tf_rsp -= 8; 468 td->td_frame->tf_rip = (register_t)entry; 469 td->td_frame->tf_ds = _udatasel; 470 td->td_frame->tf_es = _udatasel; 471 td->td_frame->tf_fs = _ufssel; 472 td->td_frame->tf_gs = _ugssel; 473 td->td_frame->tf_flags = TF_HASSEGS; 474 475 /* 476 * Pass the address of the mailbox for this kse to the uts 477 * function as a parameter on the stack. 478 */ 479 td->td_frame->tf_rdi = (register_t)arg; 480} 481 482int 483cpu_set_user_tls(struct thread *td, void *tls_base) 484{ 485 486 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS) 487 return (EINVAL); 488 489#ifdef COMPAT_IA32 490 if (td->td_proc->p_sysent->sv_flags & SV_ILP32) { 491 td->td_pcb->pcb_gsbase = (register_t)tls_base; 492 return (0); 493 } 494#endif 495 td->td_pcb->pcb_fsbase = (register_t)tls_base; 496 td->td_pcb->pcb_full_iret = 1; 497 return (0); 498} 499 500#ifdef SMP 501static void 502cpu_reset_proxy() 503{ 504 505 cpu_reset_proxy_active = 1; 506 while (cpu_reset_proxy_active == 1) 507 ; /* Wait for other cpu to see that we've started */ 508 stop_cpus((1<<cpu_reset_proxyid)); 509 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 510 DELAY(1000000); 511 cpu_reset_real(); 512} 513#endif 514 515void 516cpu_reset() 517{ 518#ifdef SMP 519 u_int cnt, map; 520 521 if (smp_active) { 522 map = PCPU_GET(other_cpus) & ~stopped_cpus; 523 if (map != 0) { 524 printf("cpu_reset: Stopping other CPUs\n"); 525 stop_cpus(map); 526 } 527 528 if (PCPU_GET(cpuid) != 0) { 529 cpu_reset_proxyid = PCPU_GET(cpuid); 530 cpustop_restartfunc = cpu_reset_proxy; 531 cpu_reset_proxy_active = 0; 532 printf("cpu_reset: Restarting BSP\n"); 533 534 /* Restart CPU #0. */ 535 atomic_store_rel_int(&started_cpus, 1 << 0); 536 537 cnt = 0; 538 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 539 cnt++; /* Wait for BSP to announce restart */ 540 if (cpu_reset_proxy_active == 0) 541 printf("cpu_reset: Failed to restart BSP\n"); 542 enable_intr(); 543 cpu_reset_proxy_active = 2; 544 545 while (1); 546 /* NOTREACHED */ 547 } 548 549 DELAY(1000000); 550 } 551#endif 552 cpu_reset_real(); 553 /* NOTREACHED */ 554} 555 556static void 557cpu_reset_real() 558{ 559 struct region_descriptor null_idt; 560 int b; 561 562 disable_intr(); 563 564 /* 565 * Attempt to do a CPU reset via the keyboard controller, 566 * do not turn off GateA20, as any machine that fails 567 * to do the reset here would then end up in no man's land. 568 */ 569 outb(IO_KBD + 4, 0xFE); 570 DELAY(500000); /* wait 0.5 sec to see if that did it */ 571 572 /* 573 * Attempt to force a reset via the Reset Control register at 574 * I/O port 0xcf9. Bit 2 forces a system reset when it 575 * transitions from 0 to 1. Bit 1 selects the type of reset 576 * to attempt: 0 selects a "soft" reset, and 1 selects a 577 * "hard" reset. We try a "hard" reset. The first write sets 578 * bit 1 to select a "hard" reset and clears bit 2. The 579 * second write forces a 0 -> 1 transition in bit 2 to trigger 580 * a reset. 581 */ 582 outb(0xcf9, 0x2); 583 outb(0xcf9, 0x6); 584 DELAY(500000); /* wait 0.5 sec to see if that did it */ 585 586 /* 587 * Attempt to force a reset via the Fast A20 and Init register 588 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate. 589 * Bit 0 asserts INIT# when set to 1. We are careful to only 590 * preserve bit 1 while setting bit 0. We also must clear bit 591 * 0 before setting it if it isn't already clear. 592 */ 593 b = inb(0x92); 594 if (b != 0xff) { 595 if ((b & 0x1) != 0) 596 outb(0x92, b & 0xfe); 597 outb(0x92, b | 0x1); 598 DELAY(500000); /* wait 0.5 sec to see if that did it */ 599 } 600 601 printf("No known reset method worked, attempting CPU shutdown\n"); 602 DELAY(1000000); /* wait 1 sec for printf to complete */ 603 604 /* Wipe the IDT. */ 605 null_idt.rd_limit = 0; 606 null_idt.rd_base = 0; 607 lidt(&null_idt); 608 609 /* "good night, sweet prince .... <THUNK!>" */ 610 breakpoint(); 611 612 /* NOTREACHED */ 613 while(1); 614} 615 616/* 617 * Allocate an sf_buf for the given vm_page. On this machine, however, there 618 * is no sf_buf object. Instead, an opaque pointer to the given vm_page is 619 * returned. 620 */ 621struct sf_buf * 622sf_buf_alloc(struct vm_page *m, int pri) 623{ 624 625 return ((struct sf_buf *)m); 626} 627 628/* 629 * Free the sf_buf. In fact, do nothing because there are no resources 630 * associated with the sf_buf. 631 */ 632void 633sf_buf_free(struct sf_buf *sf) 634{ 635} 636 637/* 638 * Software interrupt handler for queued VM system processing. 639 */ 640void 641swi_vm(void *dummy) 642{ 643 if (busdma_swi_pending != 0) 644 busdma_swi(); 645} 646 647/* 648 * Tell whether this address is in some physical memory region. 649 * Currently used by the kernel coredump code in order to avoid 650 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 651 * or other unpredictable behaviour. 652 */ 653 654int 655is_physical_memory(vm_paddr_t addr) 656{ 657 658#ifdef DEV_ISA 659 /* The ISA ``memory hole''. */ 660 if (addr >= 0xa0000 && addr < 0x100000) 661 return 0; 662#endif 663 664 /* 665 * stuff other tests for known memory-mapped devices (PCI?) 666 * here 667 */ 668 669 return 1; 670} 671