vm_machdep.c revision 169029
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 169029 2007-04-24 20:06:36Z jhb $"); 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/kse.h> 55#include <sys/kernel.h> 56#include <sys/ktr.h> 57#include <sys/lock.h> 58#include <sys/malloc.h> 59#include <sys/mbuf.h> 60#include <sys/mutex.h> 61#include <sys/pioctl.h> 62#include <sys/proc.h> 63#include <sys/sf_buf.h> 64#include <sys/smp.h> 65#include <sys/sysctl.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 75#include <vm/vm.h> 76#include <vm/vm_extern.h> 77#include <vm/vm_kern.h> 78#include <vm/vm_page.h> 79#include <vm/vm_map.h> 80#include <vm/vm_param.h> 81 82#include <amd64/isa/isa.h> 83 84#ifdef COMPAT_IA32 85 86extern struct sysentvec ia32_freebsd_sysvec; 87 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 96 97/* 98 * Finish a fork operation, with process p2 nearly set up. 99 * Copy and update the pcb, set up the stack so that the child 100 * ready to run and return to user mode. 101 */ 102void 103cpu_fork(td1, p2, td2, flags) 104 register struct thread *td1; 105 register struct proc *p2; 106 struct thread *td2; 107 int flags; 108{ 109 register struct proc *p1; 110 struct pcb *pcb2; 111 struct mdproc *mdp2; 112 113 p1 = td1->td_proc; 114 if ((flags & RFPROC) == 0) 115 return; 116 117 /* Ensure that p1's pcb is up to date. */ 118 fpuexit(td1); 119 120 /* Point the pcb to the top of the stack */ 121 pcb2 = (struct pcb *)(td2->td_kstack + 122 td2->td_kstack_pages * PAGE_SIZE) - 1; 123 td2->td_pcb = pcb2; 124 125 /* Copy p1's pcb */ 126 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 127 128 /* Point mdproc and then copy over td1's contents */ 129 mdp2 = &p2->p_md; 130 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 131 132 /* 133 * Create a new fresh stack for the new process. 134 * Copy the trap frame for the return to user mode as if from a 135 * syscall. This copies most of the user mode register values. 136 */ 137 td2->td_frame = (struct trapframe *)td2->td_pcb - 1; 138 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 139 140 td2->td_frame->tf_rax = 0; /* Child returns zero */ 141 td2->td_frame->tf_rflags &= ~PSL_C; /* success */ 142 td2->td_frame->tf_rdx = 1; 143 144 /* 145 * If the parent process has the trap bit set (i.e. a debugger had 146 * single stepped the process to the system call), we need to clear 147 * the trap flag from the new frame unless the debugger had set PF_FORK 148 * on the parent. Otherwise, the child will receive a (likely 149 * unexpected) SIGTRAP when it executes the first instruction after 150 * returning to userland. 151 */ 152 if ((p1->p_pfsflags & PF_FORK) == 0) 153 td2->td_frame->tf_rflags &= ~PSL_T; 154 155 /* 156 * Set registers for trampoline to user mode. Leave space for the 157 * return address on stack. These are the kernel mode register values. 158 */ 159 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pml4); 160 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */ 161 pcb2->pcb_rbp = 0; 162 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *); 163 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */ 164 pcb2->pcb_rip = (register_t)fork_trampoline; 165 /*- 166 * pcb2->pcb_dr*: cloned above. 167 * pcb2->pcb_savefpu: cloned above. 168 * pcb2->pcb_flags: cloned above. 169 * pcb2->pcb_onfault: cloned above (always NULL here?). 170 * pcb2->pcb_[fg]sbase: cloned above 171 */ 172 173 /* Setup to release sched_lock in fork_exit(). */ 174 td2->td_md.md_spinlock_count = 1; 175 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 176 177 /* 178 * Now, cpu_switch() can schedule the new process. 179 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 180 * containing the return address when exiting cpu_switch. 181 * This will normally be to fork_trampoline(), which will have 182 * %ebx loaded with the new proc's pointer. fork_trampoline() 183 * will set up a stack to call fork_return(p, frame); to complete 184 * the return to user-mode. 185 */ 186} 187 188/* 189 * Intercept the return address from a freshly forked process that has NOT 190 * been scheduled yet. 191 * 192 * This is needed to make kernel threads stay in kernel mode. 193 */ 194void 195cpu_set_fork_handler(td, func, arg) 196 struct thread *td; 197 void (*func)(void *); 198 void *arg; 199{ 200 /* 201 * Note that the trap frame follows the args, so the function 202 * is really called like this: func(arg, frame); 203 */ 204 td->td_pcb->pcb_r12 = (long) func; /* function */ 205 td->td_pcb->pcb_rbx = (long) arg; /* first arg */ 206} 207 208void 209cpu_exit(struct thread *td) 210{ 211} 212 213void 214cpu_thread_exit(struct thread *td) 215{ 216 217 if (td == PCPU_GET(fpcurthread)) 218 fpudrop(); 219 220 /* Disable any hardware breakpoints. */ 221 if (td->td_pcb->pcb_flags & PCB_DBREGS) { 222 reset_dbregs(); 223 td->td_pcb->pcb_flags &= ~PCB_DBREGS; 224 } 225} 226 227void 228cpu_thread_clean(struct thread *td) 229{ 230} 231 232void 233cpu_thread_swapin(struct thread *td) 234{ 235} 236 237void 238cpu_thread_swapout(struct thread *td) 239{ 240} 241 242void 243cpu_thread_setup(struct thread *td) 244{ 245 246 td->td_pcb = (struct pcb *)(td->td_kstack + 247 td->td_kstack_pages * PAGE_SIZE) - 1; 248 td->td_frame = (struct trapframe *)td->td_pcb - 1; 249} 250 251/* 252 * Initialize machine state (pcb and trap frame) for a new thread about to 253 * upcall. Put enough state in the new thread's PCB to get it to go back 254 * userret(), where we can intercept it again to set the return (upcall) 255 * Address and stack, along with those from upcals that are from other sources 256 * such as those generated in thread_userret() itself. 257 */ 258void 259cpu_set_upcall(struct thread *td, struct thread *td0) 260{ 261 struct pcb *pcb2; 262 263 /* Point the pcb to the top of the stack. */ 264 pcb2 = td->td_pcb; 265 266 /* 267 * Copy the upcall pcb. This loads kernel regs. 268 * Those not loaded individually below get their default 269 * values here. 270 * 271 * XXXKSE It might be a good idea to simply skip this as 272 * the values of the other registers may be unimportant. 273 * This would remove any requirement for knowing the KSE 274 * at this time (see the matching comment below for 275 * more analysis) (need a good safe default). 276 */ 277 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 278 pcb2->pcb_flags &= ~PCB_FPUINITDONE; 279 280 /* 281 * Create a new fresh stack for the new thread. 282 * Don't forget to set this stack value into whatever supplies 283 * the address for the fault handlers. 284 * The contexts are filled in at the time we actually DO the 285 * upcall as only then do we know which KSE we got. 286 */ 287 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 288 289 /* 290 * Set registers for trampoline to user mode. Leave space for the 291 * return address on stack. These are the kernel mode register values. 292 */ 293 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pml4); 294 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */ 295 pcb2->pcb_rbp = 0; 296 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */ 297 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */ 298 pcb2->pcb_rip = (register_t)fork_trampoline; 299 /* 300 * If we didn't copy the pcb, we'd need to do the following registers: 301 * pcb2->pcb_dr*: cloned above. 302 * pcb2->pcb_savefpu: cloned above. 303 * pcb2->pcb_onfault: cloned above (always NULL here?). 304 * pcb2->pcb_[fg]sbase: cloned above 305 */ 306 307 /* Setup to release sched_lock in fork_exit(). */ 308 td->td_md.md_spinlock_count = 1; 309 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 310} 311 312/* 313 * Set that machine state for performing an upcall that has to 314 * be done in thread_userret() so that those upcalls generated 315 * in thread_userret() itself can be done as well. 316 */ 317void 318cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg, 319 stack_t *stack) 320{ 321 322 /* 323 * Do any extra cleaning that needs to be done. 324 * The thread may have optional components 325 * that are not present in a fresh thread. 326 * This may be a recycled thread so make it look 327 * as though it's newly allocated. 328 */ 329 cpu_thread_clean(td); 330 331#ifdef COMPAT_IA32 332 if (td->td_proc->p_sysent == &ia32_freebsd_sysvec) { 333 /* 334 * Set the trap frame to point at the beginning of the uts 335 * function. 336 */ 337 td->td_frame->tf_rbp = 0; 338 td->td_frame->tf_rsp = 339 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4; 340 td->td_frame->tf_rip = (uintptr_t)entry; 341 342 /* 343 * Pass the address of the mailbox for this kse to the uts 344 * function as a parameter on the stack. 345 */ 346 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)), 347 (uint32_t)(uintptr_t)arg); 348 349 return; 350 } 351#endif 352 353 /* 354 * Set the trap frame to point at the beginning of the uts 355 * function. 356 */ 357 td->td_frame->tf_rbp = 0; 358 td->td_frame->tf_rsp = 359 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f; 360 td->td_frame->tf_rsp -= 8; 361 td->td_frame->tf_rip = (register_t)entry; 362 363 /* 364 * Pass the address of the mailbox for this kse to the uts 365 * function as a parameter on the stack. 366 */ 367 td->td_frame->tf_rdi = (register_t)arg; 368} 369 370int 371cpu_set_user_tls(struct thread *td, void *tls_base) 372{ 373 374 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS) 375 return (EINVAL); 376 377#ifdef COMPAT_IA32 378 if (td->td_proc->p_sysent == &ia32_freebsd_sysvec) { 379 if (td == curthread) { 380 critical_enter(); 381 td->td_pcb->pcb_gsbase = (register_t)tls_base; 382 wrmsr(MSR_KGSBASE, td->td_pcb->pcb_gsbase); 383 critical_exit(); 384 } else { 385 td->td_pcb->pcb_gsbase = (register_t)tls_base; 386 } 387 return (0); 388 } 389#endif 390 if (td == curthread) { 391 critical_enter(); 392 td->td_pcb->pcb_fsbase = (register_t)tls_base; 393 wrmsr(MSR_FSBASE, td->td_pcb->pcb_fsbase); 394 critical_exit(); 395 } else { 396 td->td_pcb->pcb_fsbase = (register_t)tls_base; 397 } 398 return (0); 399} 400 401#ifdef SMP 402static void 403cpu_reset_proxy() 404{ 405 406 cpu_reset_proxy_active = 1; 407 while (cpu_reset_proxy_active == 1) 408 ; /* Wait for other cpu to see that we've started */ 409 stop_cpus((1<<cpu_reset_proxyid)); 410 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 411 DELAY(1000000); 412 cpu_reset_real(); 413} 414#endif 415 416void 417cpu_reset() 418{ 419#ifdef SMP 420 u_int cnt, map; 421 422 if (smp_active) { 423 map = PCPU_GET(other_cpus) & ~stopped_cpus; 424 if (map != 0) { 425 printf("cpu_reset: Stopping other CPUs\n"); 426 stop_cpus(map); 427 } 428 429 if (PCPU_GET(cpuid) != 0) { 430 cpu_reset_proxyid = PCPU_GET(cpuid); 431 cpustop_restartfunc = cpu_reset_proxy; 432 cpu_reset_proxy_active = 0; 433 printf("cpu_reset: Restarting BSP\n"); 434 435 /* Restart CPU #0. */ 436 atomic_store_rel_int(&started_cpus, 1 << 0); 437 438 cnt = 0; 439 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 440 cnt++; /* Wait for BSP to announce restart */ 441 if (cpu_reset_proxy_active == 0) 442 printf("cpu_reset: Failed to restart BSP\n"); 443 enable_intr(); 444 cpu_reset_proxy_active = 2; 445 446 while (1); 447 /* NOTREACHED */ 448 } 449 450 DELAY(1000000); 451 } 452#endif 453 cpu_reset_real(); 454 /* NOTREACHED */ 455} 456 457static void 458cpu_reset_real() 459{ 460 int b; 461 462 disable_intr(); 463 464 /* 465 * Attempt to do a CPU reset via the keyboard controller, 466 * do not turn off GateA20, as any machine that fails 467 * to do the reset here would then end up in no man's land. 468 */ 469 outb(IO_KBD + 4, 0xFE); 470 DELAY(500000); /* wait 0.5 sec to see if that did it */ 471 472 /* 473 * Attempt to force a reset via the Reset Control register at 474 * I/O port 0xcf9. Bit 2 forces a system reset when it is 475 * written as 1. Bit 1 selects the type of reset to attempt: 476 * 0 selects a "soft" reset, and 1 selects a "hard" reset. We 477 * try to do a "soft" reset first, and then a "hard" reset. 478 */ 479 outb(0xcf9, 0x2); 480 outb(0xcf9, 0x6); 481 DELAY(500000); /* wait 0.5 sec to see if that did it */ 482 483 /* 484 * Attempt to force a reset via the Fast A20 and Init register 485 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate. 486 * Bit 0 asserts INIT# when set to 1. We are careful to only 487 * preserve bit 1 while setting bit 0. We also must clear bit 488 * 0 before setting it if it isn't already clear. 489 */ 490 b = inb(0x92); 491 if (b != 0xff) { 492 if ((b & 0x1) != 0) 493 outb(0x92, b & 0xfe); 494 outb(0x92, b | 0x1); 495 DELAY(500000); /* wait 0.5 sec to see if that did it */ 496 } 497 printf("No known reset method worked, attempting CPU shutdown\n"); 498 DELAY(1000000); /* wait 1 sec for printf to complete */ 499 500 /* Force a shutdown by unmapping entire address space. */ 501 bzero((caddr_t)PML4map, PAGE_SIZE); 502 503 /* "good night, sweet prince .... <THUNK!>" */ 504 invltlb(); 505 /* NOTREACHED */ 506 while(1); 507} 508 509/* 510 * Allocate an sf_buf for the given vm_page. On this machine, however, there 511 * is no sf_buf object. Instead, an opaque pointer to the given vm_page is 512 * returned. 513 */ 514struct sf_buf * 515sf_buf_alloc(struct vm_page *m, int pri) 516{ 517 518 return ((struct sf_buf *)m); 519} 520 521/* 522 * Free the sf_buf. In fact, do nothing because there are no resources 523 * associated with the sf_buf. 524 */ 525void 526sf_buf_free(struct sf_buf *sf) 527{ 528} 529 530/* 531 * Software interrupt handler for queued VM system processing. 532 */ 533void 534swi_vm(void *dummy) 535{ 536 if (busdma_swi_pending != 0) 537 busdma_swi(); 538} 539 540/* 541 * Tell whether this address is in some physical memory region. 542 * Currently used by the kernel coredump code in order to avoid 543 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 544 * or other unpredictable behaviour. 545 */ 546 547int 548is_physical_memory(vm_paddr_t addr) 549{ 550 551#ifdef DEV_ISA 552 /* The ISA ``memory hole''. */ 553 if (addr >= 0xa0000 && addr < 0x100000) 554 return 0; 555#endif 556 557 /* 558 * stuff other tests for known memory-mapped devices (PCI?) 559 * here 560 */ 561 562 return 1; 563} 564