vm_machdep.c revision 183615
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 183615 2008-10-05 02:03:54Z davidxu $"); 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/unistd.h> 66#include <sys/vnode.h> 67#include <sys/vmmeter.h> 68 69#include <machine/cpu.h> 70#include <machine/md_var.h> 71#include <machine/pcb.h> 72#include <machine/specialreg.h> 73 74#include <vm/vm.h> 75#include <vm/vm_extern.h> 76#include <vm/vm_kern.h> 77#include <vm/vm_page.h> 78#include <vm/vm_map.h> 79#include <vm/vm_param.h> 80 81#include <amd64/isa/isa.h> 82 83#ifdef COMPAT_IA32 84 85extern struct sysentvec ia32_freebsd_sysvec; 86 87#endif 88 89static void cpu_reset_real(void); 90#ifdef SMP 91static void cpu_reset_proxy(void); 92static u_int cpu_reset_proxyid; 93static volatile u_int cpu_reset_proxy_active; 94#endif 95 96/* 97 * Finish a fork operation, with process p2 nearly set up. 98 * Copy and update the pcb, set up the stack so that the child 99 * ready to run and return to user mode. 100 */ 101void 102cpu_fork(td1, p2, td2, flags) 103 register struct thread *td1; 104 register struct proc *p2; 105 struct thread *td2; 106 int flags; 107{ 108 register struct proc *p1; 109 struct pcb *pcb2; 110 struct mdproc *mdp2; 111 112 p1 = td1->td_proc; 113 if ((flags & RFPROC) == 0) 114 return; 115 116 /* Ensure that p1's pcb is up to date. */ 117 fpuexit(td1); 118 119 /* Point the pcb to the top of the stack */ 120 pcb2 = (struct pcb *)(td2->td_kstack + 121 td2->td_kstack_pages * PAGE_SIZE) - 1; 122 td2->td_pcb = pcb2; 123 124 /* Copy p1's pcb */ 125 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 126 127 /* Point mdproc and then copy over td1's contents */ 128 mdp2 = &p2->p_md; 129 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 130 131 /* 132 * Create a new fresh stack for the new process. 133 * Copy the trap frame for the return to user mode as if from a 134 * syscall. This copies most of the user mode register values. 135 */ 136 td2->td_frame = (struct trapframe *)td2->td_pcb - 1; 137 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 138 139 td2->td_frame->tf_rax = 0; /* Child returns zero */ 140 td2->td_frame->tf_rflags &= ~PSL_C; /* success */ 141 td2->td_frame->tf_rdx = 1; 142 143 /* 144 * If the parent process has the trap bit set (i.e. a debugger had 145 * single stepped the process to the system call), we need to clear 146 * the trap flag from the new frame unless the debugger had set PF_FORK 147 * on the parent. Otherwise, the child will receive a (likely 148 * unexpected) SIGTRAP when it executes the first instruction after 149 * returning to userland. 150 */ 151 if ((p1->p_pfsflags & PF_FORK) == 0) 152 td2->td_frame->tf_rflags &= ~PSL_T; 153 154 /* 155 * Set registers for trampoline to user mode. Leave space for the 156 * return address on stack. These are the kernel mode register values. 157 */ 158 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pml4); 159 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */ 160 pcb2->pcb_rbp = 0; 161 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *); 162 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */ 163 pcb2->pcb_rip = (register_t)fork_trampoline; 164 /*- 165 * pcb2->pcb_dr*: cloned above. 166 * pcb2->pcb_savefpu: cloned above. 167 * pcb2->pcb_flags: cloned above. 168 * pcb2->pcb_onfault: cloned above (always NULL here?). 169 * pcb2->pcb_[fg]sbase: cloned above 170 */ 171 172 /* Setup to release spin count in fork_exit(). */ 173 td2->td_md.md_spinlock_count = 1; 174 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 175 176 /* 177 * Now, cpu_switch() can schedule the new process. 178 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 179 * containing the return address when exiting cpu_switch. 180 * This will normally be to fork_trampoline(), which will have 181 * %ebx loaded with the new proc's pointer. fork_trampoline() 182 * will set up a stack to call fork_return(p, frame); to complete 183 * the return to user-mode. 184 */ 185} 186 187/* 188 * Intercept the return address from a freshly forked process that has NOT 189 * been scheduled yet. 190 * 191 * This is needed to make kernel threads stay in kernel mode. 192 */ 193void 194cpu_set_fork_handler(td, func, arg) 195 struct thread *td; 196 void (*func)(void *); 197 void *arg; 198{ 199 /* 200 * Note that the trap frame follows the args, so the function 201 * is really called like this: func(arg, frame); 202 */ 203 td->td_pcb->pcb_r12 = (long) func; /* function */ 204 td->td_pcb->pcb_rbx = (long) arg; /* first arg */ 205} 206 207void 208cpu_exit(struct thread *td) 209{ 210} 211 212void 213cpu_thread_exit(struct thread *td) 214{ 215 216 if (td == PCPU_GET(fpcurthread)) 217 fpudrop(); 218 219 /* Disable any hardware breakpoints. */ 220 if (td->td_pcb->pcb_flags & PCB_DBREGS) { 221 reset_dbregs(); 222 td->td_pcb->pcb_flags &= ~PCB_DBREGS; 223 } 224} 225 226void 227cpu_thread_clean(struct thread *td) 228{ 229} 230 231void 232cpu_thread_swapin(struct thread *td) 233{ 234} 235 236void 237cpu_thread_swapout(struct thread *td) 238{ 239} 240 241void 242cpu_thread_alloc(struct thread *td) 243{ 244 245 td->td_pcb = (struct pcb *)(td->td_kstack + 246 td->td_kstack_pages * PAGE_SIZE) - 1; 247 td->td_frame = (struct trapframe *)td->td_pcb - 1; 248} 249 250void 251cpu_thread_free(struct thread *td) 252{ 253} 254 255/* 256 * Initialize machine state (pcb and trap frame) for a new thread about to 257 * upcall. Put enough state in the new thread's PCB to get it to go back 258 * userret(), where we can intercept it again to set the return (upcall) 259 * Address and stack, along with those from upcals that are from other sources 260 * such as those generated in thread_userret() itself. 261 */ 262void 263cpu_set_upcall(struct thread *td, struct thread *td0) 264{ 265 struct pcb *pcb2; 266 267 /* Point the pcb to the top of the stack. */ 268 pcb2 = td->td_pcb; 269 270 /* 271 * Copy the upcall pcb. This loads kernel regs. 272 * Those not loaded individually below get their default 273 * values here. 274 */ 275 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 276 pcb2->pcb_flags &= ~PCB_FPUINITDONE; 277 278 /* 279 * Create a new fresh stack for the new thread. 280 */ 281 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 282 283 /* If the current thread has the trap bit set (i.e. a debugger had 284 * single stepped the process to the system call), we need to clear 285 * the trap flag from the new frame. Otherwise, the new thread will 286 * receive a (likely unexpected) SIGTRAP when it executes the first 287 * instruction after returning to userland. 288 */ 289 td->td_frame->tf_rflags &= ~PSL_T; 290 291 /* 292 * Set registers for trampoline to user mode. Leave space for the 293 * return address on stack. These are the kernel mode register values. 294 */ 295 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pml4); 296 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */ 297 pcb2->pcb_rbp = 0; 298 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */ 299 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */ 300 pcb2->pcb_rip = (register_t)fork_trampoline; 301 /* 302 * If we didn't copy the pcb, we'd need to do the following registers: 303 * pcb2->pcb_dr*: cloned above. 304 * pcb2->pcb_savefpu: cloned above. 305 * pcb2->pcb_onfault: cloned above (always NULL here?). 306 * pcb2->pcb_[fg]sbase: cloned above 307 */ 308 309 /* Setup to release spin count in fork_exit(). */ 310 td->td_md.md_spinlock_count = 1; 311 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 312} 313 314/* 315 * Set that machine state for performing an upcall that has to 316 * be done in thread_userret() so that those upcalls generated 317 * in thread_userret() itself can be done as well. 318 */ 319void 320cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg, 321 stack_t *stack) 322{ 323 324 /* 325 * Do any extra cleaning that needs to be done. 326 * The thread may have optional components 327 * that are not present in a fresh thread. 328 * This may be a recycled thread so make it look 329 * as though it's newly allocated. 330 */ 331 cpu_thread_clean(td); 332 333#ifdef COMPAT_IA32 334 if (td->td_proc->p_sysent == &ia32_freebsd_sysvec) { 335 /* 336 * Set the trap frame to point at the beginning of the uts 337 * function. 338 */ 339 td->td_frame->tf_rbp = 0; 340 td->td_frame->tf_rsp = 341 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4; 342 td->td_frame->tf_rip = (uintptr_t)entry; 343 344 /* 345 * Pass the address of the mailbox for this kse to the uts 346 * function as a parameter on the stack. 347 */ 348 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)), 349 (uint32_t)(uintptr_t)arg); 350 351 return; 352 } 353#endif 354 355 /* 356 * Set the trap frame to point at the beginning of the uts 357 * function. 358 */ 359 td->td_frame->tf_rbp = 0; 360 td->td_frame->tf_rsp = 361 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f; 362 td->td_frame->tf_rsp -= 8; 363 td->td_frame->tf_rip = (register_t)entry; 364 365 /* 366 * Pass the address of the mailbox for this kse to the uts 367 * function as a parameter on the stack. 368 */ 369 td->td_frame->tf_rdi = (register_t)arg; 370} 371 372int 373cpu_set_user_tls(struct thread *td, void *tls_base) 374{ 375 376 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS) 377 return (EINVAL); 378 379#ifdef COMPAT_IA32 380 if (td->td_proc->p_sysent == &ia32_freebsd_sysvec) { 381 if (td == curthread) { 382 critical_enter(); 383 td->td_pcb->pcb_gsbase = (register_t)tls_base; 384 wrmsr(MSR_KGSBASE, td->td_pcb->pcb_gsbase); 385 critical_exit(); 386 } else { 387 td->td_pcb->pcb_gsbase = (register_t)tls_base; 388 } 389 return (0); 390 } 391#endif 392 if (td == curthread) { 393 critical_enter(); 394 td->td_pcb->pcb_fsbase = (register_t)tls_base; 395 wrmsr(MSR_FSBASE, td->td_pcb->pcb_fsbase); 396 critical_exit(); 397 } else { 398 td->td_pcb->pcb_fsbase = (register_t)tls_base; 399 } 400 return (0); 401} 402 403#ifdef SMP 404static void 405cpu_reset_proxy() 406{ 407 408 cpu_reset_proxy_active = 1; 409 while (cpu_reset_proxy_active == 1) 410 ; /* Wait for other cpu to see that we've started */ 411 stop_cpus((1<<cpu_reset_proxyid)); 412 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 413 DELAY(1000000); 414 cpu_reset_real(); 415} 416#endif 417 418void 419cpu_reset() 420{ 421#ifdef SMP 422 u_int cnt, map; 423 424 if (smp_active) { 425 map = PCPU_GET(other_cpus) & ~stopped_cpus; 426 if (map != 0) { 427 printf("cpu_reset: Stopping other CPUs\n"); 428 stop_cpus(map); 429 } 430 431 if (PCPU_GET(cpuid) != 0) { 432 cpu_reset_proxyid = PCPU_GET(cpuid); 433 cpustop_restartfunc = cpu_reset_proxy; 434 cpu_reset_proxy_active = 0; 435 printf("cpu_reset: Restarting BSP\n"); 436 437 /* Restart CPU #0. */ 438 atomic_store_rel_int(&started_cpus, 1 << 0); 439 440 cnt = 0; 441 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 442 cnt++; /* Wait for BSP to announce restart */ 443 if (cpu_reset_proxy_active == 0) 444 printf("cpu_reset: Failed to restart BSP\n"); 445 enable_intr(); 446 cpu_reset_proxy_active = 2; 447 448 while (1); 449 /* NOTREACHED */ 450 } 451 452 DELAY(1000000); 453 } 454#endif 455 cpu_reset_real(); 456 /* NOTREACHED */ 457} 458 459static void 460cpu_reset_real() 461{ 462 struct region_descriptor null_idt; 463 int b; 464 465 disable_intr(); 466 467 /* 468 * Attempt to do a CPU reset via the keyboard controller, 469 * do not turn off GateA20, as any machine that fails 470 * to do the reset here would then end up in no man's land. 471 */ 472 outb(IO_KBD + 4, 0xFE); 473 DELAY(500000); /* wait 0.5 sec to see if that did it */ 474 475 /* 476 * Attempt to force a reset via the Reset Control register at 477 * I/O port 0xcf9. Bit 2 forces a system reset when it 478 * transitions from 0 to 1. Bit 1 selects the type of reset 479 * to attempt: 0 selects a "soft" reset, and 1 selects a 480 * "hard" reset. We try a "hard" reset. The first write sets 481 * bit 1 to select a "hard" reset and clears bit 2. The 482 * second write forces a 0 -> 1 transition in bit 2 to trigger 483 * a reset. 484 */ 485 outb(0xcf9, 0x2); 486 outb(0xcf9, 0x6); 487 DELAY(500000); /* wait 0.5 sec to see if that did it */ 488 489 /* 490 * Attempt to force a reset via the Fast A20 and Init register 491 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate. 492 * Bit 0 asserts INIT# when set to 1. We are careful to only 493 * preserve bit 1 while setting bit 0. We also must clear bit 494 * 0 before setting it if it isn't already clear. 495 */ 496 b = inb(0x92); 497 if (b != 0xff) { 498 if ((b & 0x1) != 0) 499 outb(0x92, b & 0xfe); 500 outb(0x92, b | 0x1); 501 DELAY(500000); /* wait 0.5 sec to see if that did it */ 502 } 503 504 printf("No known reset method worked, attempting CPU shutdown\n"); 505 DELAY(1000000); /* wait 1 sec for printf to complete */ 506 507 /* Wipe the IDT. */ 508 null_idt.rd_limit = 0; 509 null_idt.rd_base = 0; 510 lidt(&null_idt); 511 512 /* "good night, sweet prince .... <THUNK!>" */ 513 breakpoint(); 514 515 /* NOTREACHED */ 516 while(1); 517} 518 519/* 520 * Allocate an sf_buf for the given vm_page. On this machine, however, there 521 * is no sf_buf object. Instead, an opaque pointer to the given vm_page is 522 * returned. 523 */ 524struct sf_buf * 525sf_buf_alloc(struct vm_page *m, int pri) 526{ 527 528 return ((struct sf_buf *)m); 529} 530 531/* 532 * Free the sf_buf. In fact, do nothing because there are no resources 533 * associated with the sf_buf. 534 */ 535void 536sf_buf_free(struct sf_buf *sf) 537{ 538} 539 540/* 541 * Software interrupt handler for queued VM system processing. 542 */ 543void 544swi_vm(void *dummy) 545{ 546 if (busdma_swi_pending != 0) 547 busdma_swi(); 548} 549 550/* 551 * Tell whether this address is in some physical memory region. 552 * Currently used by the kernel coredump code in order to avoid 553 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 554 * or other unpredictable behaviour. 555 */ 556 557int 558is_physical_memory(vm_paddr_t addr) 559{ 560 561#ifdef DEV_ISA 562 /* The ISA ``memory hole''. */ 563 if (addr >= 0xa0000 && addr < 0x100000) 564 return 0; 565#endif 566 567 /* 568 * stuff other tests for known memory-mapped devices (PCI?) 569 * here 570 */ 571 572 return 1; 573} 574