vm_machdep.c revision 190237
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 190237 2009-03-22 02:33:48Z alc $"); 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 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 84static void cpu_reset_real(void); 85#ifdef SMP 86static void cpu_reset_proxy(void); 87static u_int cpu_reset_proxyid; 88static volatile u_int cpu_reset_proxy_active; 89#endif 90 91/* 92 * Finish a fork operation, with process p2 nearly set up. 93 * Copy and update the pcb, set up the stack so that the child 94 * ready to run and return to user mode. 95 */ 96void 97cpu_fork(td1, p2, td2, flags) 98 register struct thread *td1; 99 register struct proc *p2; 100 struct thread *td2; 101 int flags; 102{ 103 register struct proc *p1; 104 struct pcb *pcb2; 105 struct mdproc *mdp2; 106 107 p1 = td1->td_proc; 108 if ((flags & RFPROC) == 0) 109 return; 110 111 /* Ensure that p1's pcb is up to date. */ 112 fpuexit(td1); 113 114 /* Point the pcb to the top of the stack */ 115 pcb2 = (struct pcb *)(td2->td_kstack + 116 td2->td_kstack_pages * PAGE_SIZE) - 1; 117 td2->td_pcb = pcb2; 118 119 /* Copy p1's pcb */ 120 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 121 122 /* Point mdproc and then copy over td1's contents */ 123 mdp2 = &p2->p_md; 124 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 125 126 /* 127 * Create a new fresh stack for the new process. 128 * Copy the trap frame for the return to user mode as if from a 129 * syscall. This copies most of the user mode register values. 130 */ 131 td2->td_frame = (struct trapframe *)td2->td_pcb - 1; 132 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 133 134 td2->td_frame->tf_rax = 0; /* Child returns zero */ 135 td2->td_frame->tf_rflags &= ~PSL_C; /* success */ 136 td2->td_frame->tf_rdx = 1; 137 138 /* 139 * If the parent process has the trap bit set (i.e. a debugger had 140 * single stepped the process to the system call), we need to clear 141 * the trap flag from the new frame unless the debugger had set PF_FORK 142 * on the parent. Otherwise, the child will receive a (likely 143 * unexpected) SIGTRAP when it executes the first instruction after 144 * returning to userland. 145 */ 146 if ((p1->p_pfsflags & PF_FORK) == 0) 147 td2->td_frame->tf_rflags &= ~PSL_T; 148 149 /* 150 * Set registers for trampoline to user mode. Leave space for the 151 * return address on stack. These are the kernel mode register values. 152 */ 153 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pml4); 154 pcb2->pcb_r12 = (register_t)fork_return; /* fork_trampoline argument */ 155 pcb2->pcb_rbp = 0; 156 pcb2->pcb_rsp = (register_t)td2->td_frame - sizeof(void *); 157 pcb2->pcb_rbx = (register_t)td2; /* fork_trampoline argument */ 158 pcb2->pcb_rip = (register_t)fork_trampoline; 159 /*- 160 * pcb2->pcb_dr*: cloned above. 161 * pcb2->pcb_savefpu: cloned above. 162 * pcb2->pcb_flags: cloned above. 163 * pcb2->pcb_onfault: cloned above (always NULL here?). 164 * pcb2->pcb_[fg]sbase: cloned above 165 */ 166 167 /* Setup to release spin count in fork_exit(). */ 168 td2->td_md.md_spinlock_count = 1; 169 td2->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 170 171 /* 172 * Now, cpu_switch() can schedule the new process. 173 * pcb_rsp is loaded pointing to the cpu_switch() stack frame 174 * containing the return address when exiting cpu_switch. 175 * This will normally be to fork_trampoline(), which will have 176 * %ebx loaded with the new proc's pointer. fork_trampoline() 177 * will set up a stack to call fork_return(p, frame); to complete 178 * the return to user-mode. 179 */ 180} 181 182/* 183 * Intercept the return address from a freshly forked process that has NOT 184 * been scheduled yet. 185 * 186 * This is needed to make kernel threads stay in kernel mode. 187 */ 188void 189cpu_set_fork_handler(td, func, arg) 190 struct thread *td; 191 void (*func)(void *); 192 void *arg; 193{ 194 /* 195 * Note that the trap frame follows the args, so the function 196 * is really called like this: func(arg, frame); 197 */ 198 td->td_pcb->pcb_r12 = (long) func; /* function */ 199 td->td_pcb->pcb_rbx = (long) arg; /* first arg */ 200} 201 202void 203cpu_exit(struct thread *td) 204{ 205} 206 207void 208cpu_thread_exit(struct thread *td) 209{ 210 211 if (td == PCPU_GET(fpcurthread)) 212 fpudrop(); 213 214 /* Disable any hardware breakpoints. */ 215 if (td->td_pcb->pcb_flags & PCB_DBREGS) { 216 reset_dbregs(); 217 td->td_pcb->pcb_flags &= ~PCB_DBREGS; 218 } 219} 220 221void 222cpu_thread_clean(struct thread *td) 223{ 224} 225 226void 227cpu_thread_swapin(struct thread *td) 228{ 229} 230 231void 232cpu_thread_swapout(struct thread *td) 233{ 234} 235 236void 237cpu_thread_alloc(struct thread *td) 238{ 239 240 td->td_pcb = (struct pcb *)(td->td_kstack + 241 td->td_kstack_pages * PAGE_SIZE) - 1; 242 td->td_frame = (struct trapframe *)td->td_pcb - 1; 243} 244 245void 246cpu_thread_free(struct thread *td) 247{ 248} 249 250/* 251 * Initialize machine state (pcb and trap frame) for a new thread about to 252 * upcall. Put enough state in the new thread's PCB to get it to go back 253 * userret(), where we can intercept it again to set the return (upcall) 254 * Address and stack, along with those from upcals that are from other sources 255 * such as those generated in thread_userret() itself. 256 */ 257void 258cpu_set_upcall(struct thread *td, struct thread *td0) 259{ 260 struct pcb *pcb2; 261 262 /* Point the pcb to the top of the stack. */ 263 pcb2 = td->td_pcb; 264 265 /* 266 * Copy the upcall pcb. This loads kernel regs. 267 * Those not loaded individually below get their default 268 * values here. 269 */ 270 bcopy(td0->td_pcb, pcb2, sizeof(*pcb2)); 271 pcb2->pcb_flags &= ~PCB_FPUINITDONE; 272 273 /* 274 * Create a new fresh stack for the new thread. 275 */ 276 bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe)); 277 278 /* If the current thread has the trap bit set (i.e. a debugger had 279 * single stepped the process to the system call), we need to clear 280 * the trap flag from the new frame. Otherwise, the new thread will 281 * receive a (likely unexpected) SIGTRAP when it executes the first 282 * instruction after returning to userland. 283 */ 284 td->td_frame->tf_rflags &= ~PSL_T; 285 286 /* 287 * Set registers for trampoline to user mode. Leave space for the 288 * return address on stack. These are the kernel mode register values. 289 */ 290 pcb2->pcb_r12 = (register_t)fork_return; /* trampoline arg */ 291 pcb2->pcb_rbp = 0; 292 pcb2->pcb_rsp = (register_t)td->td_frame - sizeof(void *); /* trampoline arg */ 293 pcb2->pcb_rbx = (register_t)td; /* trampoline arg */ 294 pcb2->pcb_rip = (register_t)fork_trampoline; 295 /* 296 * If we didn't copy the pcb, we'd need to do the following registers: 297 * pcb2->pcb_cr3: cloned above. 298 * pcb2->pcb_dr*: cloned above. 299 * pcb2->pcb_savefpu: cloned above. 300 * pcb2->pcb_onfault: cloned above (always NULL here?). 301 * pcb2->pcb_[fg]sbase: cloned above 302 */ 303 304 /* Setup to release spin count in fork_exit(). */ 305 td->td_md.md_spinlock_count = 1; 306 td->td_md.md_saved_flags = PSL_KERNEL | PSL_I; 307} 308 309/* 310 * Set that machine state for performing an upcall that has to 311 * be done in thread_userret() so that those upcalls generated 312 * in thread_userret() itself can be done as well. 313 */ 314void 315cpu_set_upcall_kse(struct thread *td, void (*entry)(void *), void *arg, 316 stack_t *stack) 317{ 318 319 /* 320 * Do any extra cleaning that needs to be done. 321 * The thread may have optional components 322 * that are not present in a fresh thread. 323 * This may be a recycled thread so make it look 324 * as though it's newly allocated. 325 */ 326 cpu_thread_clean(td); 327 328#ifdef COMPAT_IA32 329 if (td->td_proc->p_sysent->sv_flags & SV_ILP32) { 330 /* 331 * Set the trap frame to point at the beginning of the uts 332 * function. 333 */ 334 td->td_frame->tf_rbp = 0; 335 td->td_frame->tf_rsp = 336 (((uintptr_t)stack->ss_sp + stack->ss_size - 4) & ~0x0f) - 4; 337 td->td_frame->tf_rip = (uintptr_t)entry; 338 339 /* 340 * Pass the address of the mailbox for this kse to the uts 341 * function as a parameter on the stack. 342 */ 343 suword32((void *)(td->td_frame->tf_rsp + sizeof(int32_t)), 344 (uint32_t)(uintptr_t)arg); 345 346 return; 347 } 348#endif 349 350 /* 351 * Set the trap frame to point at the beginning of the uts 352 * function. 353 */ 354 td->td_frame->tf_rbp = 0; 355 td->td_frame->tf_rsp = 356 ((register_t)stack->ss_sp + stack->ss_size) & ~0x0f; 357 td->td_frame->tf_rsp -= 8; 358 td->td_frame->tf_rip = (register_t)entry; 359 360 /* 361 * Pass the address of the mailbox for this kse to the uts 362 * function as a parameter on the stack. 363 */ 364 td->td_frame->tf_rdi = (register_t)arg; 365} 366 367int 368cpu_set_user_tls(struct thread *td, void *tls_base) 369{ 370 371 if ((u_int64_t)tls_base >= VM_MAXUSER_ADDRESS) 372 return (EINVAL); 373 374#ifdef COMPAT_IA32 375 if (td->td_proc->p_sysent->sv_flags & SV_ILP32) { 376 if (td == curthread) { 377 critical_enter(); 378 td->td_pcb->pcb_gsbase = (register_t)tls_base; 379 wrmsr(MSR_KGSBASE, td->td_pcb->pcb_gsbase); 380 critical_exit(); 381 } else { 382 td->td_pcb->pcb_gsbase = (register_t)tls_base; 383 } 384 return (0); 385 } 386#endif 387 if (td == curthread) { 388 critical_enter(); 389 td->td_pcb->pcb_fsbase = (register_t)tls_base; 390 wrmsr(MSR_FSBASE, td->td_pcb->pcb_fsbase); 391 critical_exit(); 392 } else { 393 td->td_pcb->pcb_fsbase = (register_t)tls_base; 394 } 395 return (0); 396} 397 398#ifdef SMP 399static void 400cpu_reset_proxy() 401{ 402 403 cpu_reset_proxy_active = 1; 404 while (cpu_reset_proxy_active == 1) 405 ; /* Wait for other cpu to see that we've started */ 406 stop_cpus((1<<cpu_reset_proxyid)); 407 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 408 DELAY(1000000); 409 cpu_reset_real(); 410} 411#endif 412 413void 414cpu_reset() 415{ 416#ifdef SMP 417 u_int cnt, map; 418 419 if (smp_active) { 420 map = PCPU_GET(other_cpus) & ~stopped_cpus; 421 if (map != 0) { 422 printf("cpu_reset: Stopping other CPUs\n"); 423 stop_cpus(map); 424 } 425 426 if (PCPU_GET(cpuid) != 0) { 427 cpu_reset_proxyid = PCPU_GET(cpuid); 428 cpustop_restartfunc = cpu_reset_proxy; 429 cpu_reset_proxy_active = 0; 430 printf("cpu_reset: Restarting BSP\n"); 431 432 /* Restart CPU #0. */ 433 atomic_store_rel_int(&started_cpus, 1 << 0); 434 435 cnt = 0; 436 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 437 cnt++; /* Wait for BSP to announce restart */ 438 if (cpu_reset_proxy_active == 0) 439 printf("cpu_reset: Failed to restart BSP\n"); 440 enable_intr(); 441 cpu_reset_proxy_active = 2; 442 443 while (1); 444 /* NOTREACHED */ 445 } 446 447 DELAY(1000000); 448 } 449#endif 450 cpu_reset_real(); 451 /* NOTREACHED */ 452} 453 454static void 455cpu_reset_real() 456{ 457 struct region_descriptor null_idt; 458 int b; 459 460 disable_intr(); 461 462 /* 463 * Attempt to do a CPU reset via the keyboard controller, 464 * do not turn off GateA20, as any machine that fails 465 * to do the reset here would then end up in no man's land. 466 */ 467 outb(IO_KBD + 4, 0xFE); 468 DELAY(500000); /* wait 0.5 sec to see if that did it */ 469 470 /* 471 * Attempt to force a reset via the Reset Control register at 472 * I/O port 0xcf9. Bit 2 forces a system reset when it 473 * transitions from 0 to 1. Bit 1 selects the type of reset 474 * to attempt: 0 selects a "soft" reset, and 1 selects a 475 * "hard" reset. We try a "hard" reset. The first write sets 476 * bit 1 to select a "hard" reset and clears bit 2. The 477 * second write forces a 0 -> 1 transition in bit 2 to trigger 478 * a reset. 479 */ 480 outb(0xcf9, 0x2); 481 outb(0xcf9, 0x6); 482 DELAY(500000); /* wait 0.5 sec to see if that did it */ 483 484 /* 485 * Attempt to force a reset via the Fast A20 and Init register 486 * at I/O port 0x92. Bit 1 serves as an alternate A20 gate. 487 * Bit 0 asserts INIT# when set to 1. We are careful to only 488 * preserve bit 1 while setting bit 0. We also must clear bit 489 * 0 before setting it if it isn't already clear. 490 */ 491 b = inb(0x92); 492 if (b != 0xff) { 493 if ((b & 0x1) != 0) 494 outb(0x92, b & 0xfe); 495 outb(0x92, b | 0x1); 496 DELAY(500000); /* wait 0.5 sec to see if that did it */ 497 } 498 499 printf("No known reset method worked, attempting CPU shutdown\n"); 500 DELAY(1000000); /* wait 1 sec for printf to complete */ 501 502 /* Wipe the IDT. */ 503 null_idt.rd_limit = 0; 504 null_idt.rd_base = 0; 505 lidt(&null_idt); 506 507 /* "good night, sweet prince .... <THUNK!>" */ 508 breakpoint(); 509 510 /* NOTREACHED */ 511 while(1); 512} 513 514/* 515 * Allocate an sf_buf for the given vm_page. On this machine, however, there 516 * is no sf_buf object. Instead, an opaque pointer to the given vm_page is 517 * returned. 518 */ 519struct sf_buf * 520sf_buf_alloc(struct vm_page *m, int pri) 521{ 522 523 return ((struct sf_buf *)m); 524} 525 526/* 527 * Free the sf_buf. In fact, do nothing because there are no resources 528 * associated with the sf_buf. 529 */ 530void 531sf_buf_free(struct sf_buf *sf) 532{ 533} 534 535/* 536 * Software interrupt handler for queued VM system processing. 537 */ 538void 539swi_vm(void *dummy) 540{ 541 if (busdma_swi_pending != 0) 542 busdma_swi(); 543} 544 545/* 546 * Tell whether this address is in some physical memory region. 547 * Currently used by the kernel coredump code in order to avoid 548 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 549 * or other unpredictable behaviour. 550 */ 551 552int 553is_physical_memory(vm_paddr_t addr) 554{ 555 556#ifdef DEV_ISA 557 /* The ISA ``memory hole''. */ 558 if (addr >= 0xa0000 && addr < 0x100000) 559 return 0; 560#endif 561 562 /* 563 * stuff other tests for known memory-mapped devices (PCI?) 564 * here 565 */ 566 567 return 1; 568} 569