vm_machdep.c revision 190620
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 190620 2009-04-01 13:09:26Z 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 /* Copy the LDT, if necessary. */ 190 mdp1 = &td1->td_proc->p_md; 191 mdp2 = &p2->p_md; 192 mtx_lock(&dt_lock); 193 if (mdp1->md_ldt != NULL) { 194 if (flags & RFMEM) { 195 mdp1->md_ldt->ldt_refcnt++; 196 mdp2->md_ldt = mdp1->md_ldt; 197 bcopy(&mdp1->md_ldt_sd, &mdp2->md_ldt_sd, sizeof(struct 198 system_segment_descriptor)); 199 } else { 200 mdp2->md_ldt = NULL; 201 mdp2->md_ldt = user_ldt_alloc(p2, 0); 202 if (mdp2->md_ldt == NULL) 203 panic("could not copy LDT"); 204 amd64_set_ldt_data(td2, 0, max_ldt_segment, 205 (struct user_segment_descriptor *) 206 mdp1->md_ldt->ldt_base); 207 } 208 } else 209 mdp2->md_ldt = NULL; 210 mtx_unlock(&dt_lock); 211 212 /* 213 * Now, cpu_switch() can schedule the new process. 214 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 215 * containing the return address when exiting cpu_switch. 216 * This will normally be to fork_trampoline(), which will have 217 * %ebx loaded with the new proc's pointer. fork_trampoline() 218 * will set up a stack to call fork_return(p, frame); to complete 219 * the return to user-mode. 220 */ 221} 222 223/* 224 * Intercept the return address from a freshly forked process that has NOT 225 * been scheduled yet. 226 * 227 * This is needed to make kernel threads stay in kernel mode. 228 */ 229void 230cpu_set_fork_handler(td, func, arg) 231 struct thread *td; 232 void (*func)(void *); 233 void *arg; 234{ 235 /* 236 * Note that the trap frame follows the args, so the function 237 * is really called like this: func(arg, frame); 238 */ 239 td->td_pcb->pcb_r12 = (long) func; /* function */ 240 td->td_pcb->pcb_rbx = (long) arg; /* first arg */ 241} 242 243void 244cpu_exit(struct thread *td) 245{ 246 247 /* 248 * If this process has a custom LDT, release it. 249 */ 250 mtx_lock(&dt_lock); 251 if (td->td_proc->p_md.md_ldt != 0) 252 user_ldt_free(td); 253 else 254 mtx_unlock(&dt_lock); 255} 256 257void 258cpu_thread_exit(struct thread *td) 259{ 260 struct pcb *pcb; 261 262 if (td == PCPU_GET(fpcurthread)) 263 fpudrop(); 264 265 pcb = td->td_pcb; 266 267 /* Disable any hardware breakpoints. */ 268 if (pcb->pcb_flags & PCB_DBREGS) { 269 reset_dbregs(); 270 pcb->pcb_flags &= ~PCB_DBREGS; 271 } 272} 273 274void 275cpu_thread_clean(struct thread *td) 276{ 277 struct pcb *pcb; 278 279 pcb = td->td_pcb; 280 281 /* 282 * Clean TSS/iomap 283 */ 284 if (pcb->pcb_tssp != NULL) { 285 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_tssp, 286 ctob(IOPAGES + 1)); 287 pcb->pcb_tssp = NULL; 288 } 289} 290 291void 292cpu_thread_swapin(struct thread *td) 293{ 294} 295 296void 297cpu_thread_swapout(struct thread *td) 298{ 299} 300 301void 302cpu_thread_alloc(struct thread *td) 303{ 304 305 td->td_pcb = (struct pcb *)(td->td_kstack + 306 td->td_kstack_pages * PAGE_SIZE) - 1; 307 td->td_frame = (struct trapframe *)td->td_pcb - 1; 308} 309 310void 311cpu_thread_free(struct thread *td) 312{ 313 314 cpu_thread_clean(td); 315} 316 317/* 318 * Initialize machine state (pcb and trap frame) for a new thread about to 319 * upcall. Put enough state in the new thread's PCB to get it to go back 320 * userret(), where we can intercept it again to set the return (upcall) 321 * Address and stack, along with those from upcals that are from other sources 322 * such as those generated in thread_userret() itself. 323 */ 324void 325cpu_set_upcall(struct thread *td, struct thread *td0) 326{ 327 struct pcb *pcb2; 328 329 /* Point the pcb to the top of the stack. */ 330 pcb2 = td->td_pcb; 331 332 /* 333 * Copy the upcall pcb. This loads kernel regs. 334 * Those not loaded individually below get their default 335 * values here. 336 */ 337 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 338 pcb2->pcb_flags &= ~PCB_FPUINITDONE; 339 340 /* 341 * Create a new fresh stack for the new thread. 342 */ 343 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 344 345 /* If the current thread has the trap bit set (i.e. a debugger had 346 * single stepped the process to the system call), we need to clear 347 * the trap flag from the new frame. Otherwise, the new thread will 348 * receive a (likely unexpected) SIGTRAP when it executes the first 349 * instruction after returning to userland. 350 */ 351 td->td_frame->tf_rflags &= ~PSL_T; 352 353 /* 354 * Set registers for trampoline to user mode. Leave space for the 355 * return address on stack. These are the kernel mode register values. 356 */ 357 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */ 358 pcb2->pcb_rbp = 0; 359 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */ 360 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */ 361 pcb2->pcb_rip = (register_t)fork_trampoline; 362 /* 363 * If we didn't copy the pcb, we'd need to do the following registers: 364 * pcb2->pcb_cr3: cloned above. 365 * pcb2->pcb_dr*: cloned above. 366 * pcb2->pcb_savefpu: cloned above. 367 * pcb2->pcb_onfault: cloned above (always NULL here?). 368 * pcb2->pcb_[fg]sbase: cloned above 369 */ 370 371 /* Setup to release spin count in fork_exit(). */ 372 td->td_md.md_spinlock_count = 1; 373 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 374} 375 376/* 377 * Set that machine state for performing an upcall that has to 378 * be done in thread_userret() so that those upcalls generated 379 * in thread_userret() itself can be done as well. 380 */ 381void 382cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg, 383 stack_t *stack) 384{ 385 386 /* 387 * Do any extra cleaning that needs to be done. 388 * The thread may have optional components 389 * that are not present in a fresh thread. 390 * This may be a recycled thread so make it look 391 * as though it's newly allocated. 392 */ 393 cpu_thread_clean(td); 394 395#ifdef COMPAT_IA32 396 if (td->td_proc->p_sysent->sv_flags & SV_ILP32) { 397 /* 398 * Set the trap frame to point at the beginning of the uts 399 * function. 400 */ 401 td->td_frame->tf_rbp = 0; 402 td->td_frame->tf_rsp = 403 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4; 404 td->td_frame->tf_rip = (uintptr_t)entry; 405 406 /* 407 * Pass the address of the mailbox for this kse to the uts 408 * function as a parameter on the stack. 409 */ 410 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)), 411 (uint32_t)(uintptr_t)arg); 412 413 return; 414 } 415#endif 416 417 /* 418 * Set the trap frame to point at the beginning of the uts 419 * function. 420 */ 421 td->td_frame->tf_rbp = 0; 422 td->td_frame->tf_rsp = 423 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f; 424 td->td_frame->tf_rsp -= 8; 425 td->td_frame->tf_rip = (register_t)entry; 426 td->td_frame->tf_ds = _udatasel; 427 td->td_frame->tf_es = _udatasel; 428 td->td_frame->tf_fs = _ufssel; 429 td->td_frame->tf_gs = _ugssel; 430 td->td_frame->tf_flags = TF_HASSEGS; 431 432 /* 433 * Pass the address of the mailbox for this kse to the uts 434 * function as a parameter on the stack. 435 */ 436 td->td_frame->tf_rdi = (register_t)arg; 437} 438 439int 440cpu_set_user_tls(struct thread *td, void *tls_base) 441{ 442 443 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS) 444 return (EINVAL); 445 446#ifdef COMPAT_IA32 447 if (td->td_proc->p_sysent->sv_flags & SV_ILP32) { 448 td->td_pcb->pcb_gsbase = (register_t)tls_base; 449 return (0); 450 } 451#endif 452 td->td_pcb->pcb_fsbase = (register_t)tls_base; 453 return (0); 454} 455 456#ifdef SMP 457static void 458cpu_reset_proxy() 459{ 460 461 cpu_reset_proxy_active = 1; 462 while (cpu_reset_proxy_active == 1) 463 ; /* Wait for other cpu to see that we've started */ 464 stop_cpus((1<<cpu_reset_proxyid)); 465 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 466 DELAY(1000000); 467 cpu_reset_real(); 468} 469#endif 470 471void 472cpu_reset() 473{ 474#ifdef SMP 475 u_int cnt, map; 476 477 if (smp_active) { 478 map = PCPU_GET(other_cpus) & ~stopped_cpus; 479 if (map != 0) { 480 printf("cpu_reset: Stopping other CPUs\n"); 481 stop_cpus(map); 482 } 483 484 if (PCPU_GET(cpuid) != 0) { 485 cpu_reset_proxyid = PCPU_GET(cpuid); 486 cpustop_restartfunc = cpu_reset_proxy; 487 cpu_reset_proxy_active = 0; 488 printf("cpu_reset: Restarting BSP\n"); 489 490 /* Restart CPU #0. */ 491 atomic_store_rel_int(&started_cpus, 1 << 0); 492 493 cnt = 0; 494 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 495 cnt++; /* Wait for BSP to announce restart */ 496 if (cpu_reset_proxy_active == 0) 497 printf("cpu_reset: Failed to restart BSP\n"); 498 enable_intr(); 499 cpu_reset_proxy_active = 2; 500 501 while (1); 502 /* NOTREACHED */ 503 } 504 505 DELAY(1000000); 506 } 507#endif 508 cpu_reset_real(); 509 /* NOTREACHED */ 510} 511 512static void 513cpu_reset_real() 514{ 515 struct region_descriptor null_idt; 516 int b; 517 518 disable_intr(); 519 520 /* 521 * Attempt to do a CPU reset via the keyboard controller, 522 * do not turn off GateA20, as any machine that fails 523 * to do the reset here would then end up in no man's land. 524 */ 525 outb(IO_KBD + 4, 0xFE); 526 DELAY(500000); /* wait 0.5 sec to see if that did it */ 527 528 /* 529 * Attempt to force a reset via the Reset Control register at 530 * I/O port 0xcf9. Bit 2 forces a system reset when it 531 * transitions from 0 to 1. Bit 1 selects the type of reset 532 * to attempt: 0 selects a "soft" reset, and 1 selects a 533 * "hard" reset. We try a "hard" reset. The first write sets 534 * bit 1 to select a "hard" reset and clears bit 2. The 535 * second write forces a 0 -> 1 transition in bit 2 to trigger 536 * a reset. 537 */ 538 outb(0xcf9, 0x2); 539 outb(0xcf9, 0x6); 540 DELAY(500000); /* wait 0.5 sec to see if that did it */ 541 542 /* 543 * Attempt to force a reset via the Fast A20 and Init register 544 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate. 545 * Bit 0 asserts INIT# when set to 1. We are careful to only 546 * preserve bit 1 while setting bit 0. We also must clear bit 547 * 0 before setting it if it isn't already clear. 548 */ 549 b = inb(0x92); 550 if (b != 0xff) { 551 if ((b & 0x1) != 0) 552 outb(0x92, b & 0xfe); 553 outb(0x92, b | 0x1); 554 DELAY(500000); /* wait 0.5 sec to see if that did it */ 555 } 556 557 printf("No known reset method worked, attempting CPU shutdown\n"); 558 DELAY(1000000); /* wait 1 sec for printf to complete */ 559 560 /* Wipe the IDT. */ 561 null_idt.rd_limit = 0; 562 null_idt.rd_base = 0; 563 lidt(&null_idt); 564 565 /* "good night, sweet prince .... <THUNK!>" */ 566 breakpoint(); 567 568 /* NOTREACHED */ 569 while(1); 570} 571 572/* 573 * Allocate an sf_buf for the given vm_page. On this machine, however, there 574 * is no sf_buf object. Instead, an opaque pointer to the given vm_page is 575 * returned. 576 */ 577struct sf_buf * 578sf_buf_alloc(struct vm_page *m, int pri) 579{ 580 581 return ((struct sf_buf *)m); 582} 583 584/* 585 * Free the sf_buf. In fact, do nothing because there are no resources 586 * associated with the sf_buf. 587 */ 588void 589sf_buf_free(struct sf_buf *sf) 590{ 591} 592 593/* 594 * Software interrupt handler for queued VM system processing. 595 */ 596void 597swi_vm(void *dummy) 598{ 599 if (busdma_swi_pending != 0) 600 busdma_swi(); 601} 602 603/* 604 * Tell whether this address is in some physical memory region. 605 * Currently used by the kernel coredump code in order to avoid 606 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 607 * or other unpredictable behaviour. 608 */ 609 610int 611is_physical_memory(vm_paddr_t addr) 612{ 613 614#ifdef DEV_ISA 615 /* The ISA ``memory hole''. */ 616 if (addr >= 0xa0000 && addr < 0x100000) 617 return 0; 618#endif 619 620 /* 621 * stuff other tests for known memory-mapped devices (PCI?) 622 * here 623 */ 624 625 return 1; 626} 627