vm_machdep.c revision 113796
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 * $FreeBSD: head/sys/amd64/amd64/vm_machdep.c 113796 2003-04-21 15:05:05Z davidxu $ 42 */ 43 44#include "opt_npx.h" 45#ifdef PC98 46#include "opt_pc98.h" 47#endif 48#include "opt_reset.h" 49#include "opt_isa.h" 50#include "opt_kstack_pages.h" 51 52#include <sys/param.h> 53#include <sys/systm.h> 54#include <sys/malloc.h> 55#include <sys/proc.h> 56#include <sys/kse.h> 57#include <sys/bio.h> 58#include <sys/buf.h> 59#include <sys/vnode.h> 60#include <sys/vmmeter.h> 61#include <sys/kernel.h> 62#include <sys/ktr.h> 63#include <sys/mutex.h> 64#include <sys/smp.h> 65#include <sys/sysctl.h> 66#include <sys/unistd.h> 67 68#include <machine/cpu.h> 69#include <machine/md_var.h> 70#include <machine/pcb.h> 71#include <machine/pcb_ext.h> 72#include <machine/vm86.h> 73 74#include <vm/vm.h> 75#include <vm/vm_param.h> 76#include <sys/lock.h> 77#include <vm/vm_kern.h> 78#include <vm/vm_page.h> 79#include <vm/vm_map.h> 80#include <vm/vm_extern.h> 81 82#include <sys/user.h> 83 84#ifdef PC98 85#include <pc98/pc98/pc98.h> 86#else 87#include <i386/isa/isa.h> 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 96extern int _ucodesel, _udatasel; 97 98/* 99 * Finish a fork operation, with process p2 nearly set up. 100 * Copy and update the pcb, set up the stack so that the child 101 * ready to run and return to user mode. 102 */ 103void 104cpu_fork(td1, p2, td2, flags) 105 register struct thread *td1; 106 register struct proc *p2; 107 struct thread *td2; 108 int flags; 109{ 110 register struct proc *p1; 111 struct pcb *pcb2; 112 struct mdproc *mdp2; 113#ifdef DEV_NPX 114 register_t savecrit; 115#endif 116 117 p1 = td1->td_proc; 118 if ((flags & RFPROC) == 0) { 119 if ((flags & RFMEM) == 0) { 120 /* unshare user LDT */ 121 struct mdproc *mdp1 = &p1->p_md; 122 struct proc_ldt *pldt = mdp1->md_ldt; 123 if (pldt && pldt->ldt_refcnt > 1) { 124 pldt = user_ldt_alloc(mdp1, pldt->ldt_len); 125 if (pldt == NULL) 126 panic("could not copy LDT"); 127 mdp1->md_ldt = pldt; 128 set_user_ldt(mdp1); 129 user_ldt_free(td1); 130 } 131 } 132 return; 133 } 134 135 /* Ensure that p1's pcb is up to date. */ 136#ifdef DEV_NPX 137 if (td1 == curthread) 138 td1->td_pcb->pcb_gs = rgs(); 139 savecrit = intr_disable(); 140 if (PCPU_GET(fpcurthread) == td1) 141 npxsave(&td1->td_pcb->pcb_save); 142 intr_restore(savecrit); 143#endif 144 145 /* Point the pcb to the top of the stack */ 146 pcb2 = (struct pcb *)(td2->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1; 147 td2->td_pcb = pcb2; 148 149 /* Copy p1's pcb */ 150 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 151 152 /* Point mdproc and then copy over td1's contents */ 153 mdp2 = &p2->p_md; 154 bcopy(&p1->p_md, mdp2, sizeof(*mdp2)); 155 156 /* 157 * Create a new fresh stack for the new process. 158 * Copy the trap frame for the return to user mode as if from a 159 * syscall. This copies most of the user mode register values. 160 * The -16 is so we can expand the trapframe if we go to vm86. 161 */ 162 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1; 163 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 164 165 td2->td_frame->tf_eax = 0; /* Child returns zero */ 166 td2->td_frame->tf_eflags &= ~PSL_C; /* success */ 167 td2->td_frame->tf_edx = 1; 168 169 /* 170 * Set registers for trampoline to user mode. Leave space for the 171 * return address on stack. These are the kernel mode register values. 172 */ 173#ifdef PAE 174 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdpt); 175#else 176 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir); 177#endif 178 pcb2->pcb_edi = 0; 179 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */ 180 pcb2->pcb_ebp = 0; 181 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *); 182 pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */ 183 pcb2->pcb_eip = (int)fork_trampoline; 184 pcb2->pcb_psl = td2->td_frame->tf_eflags & ~PSL_I; /* ints disabled */ 185 pcb2->pcb_gs = rgs(); 186 /*- 187 * pcb2->pcb_dr*: cloned above. 188 * pcb2->pcb_savefpu: cloned above. 189 * pcb2->pcb_flags: cloned above. 190 * pcb2->pcb_onfault: cloned above (always NULL here?). 191 * pcb2->pcb_gs: cloned above. 192 * pcb2->pcb_ext: cleared below. 193 */ 194 195 /* 196 * XXX don't copy the i/o pages. this should probably be fixed. 197 */ 198 pcb2->pcb_ext = 0; 199 200 /* Copy the LDT, if necessary. */ 201 mtx_lock_spin(&sched_lock); 202 if (mdp2->md_ldt != 0) { 203 if (flags & RFMEM) { 204 mdp2->md_ldt->ldt_refcnt++; 205 } else { 206 mdp2->md_ldt = user_ldt_alloc(mdp2, 207 mdp2->md_ldt->ldt_len); 208 if (mdp2->md_ldt == NULL) 209 panic("could not copy LDT"); 210 } 211 } 212 mtx_unlock_spin(&sched_lock); 213 214 /* 215 * Now, cpu_switch() can schedule the new process. 216 * pcb_esp is loaded pointing to the cpu_switch() stack frame 217 * containing the return address when exiting cpu_switch. 218 * This will normally be to fork_trampoline(), which will have 219 * %ebx loaded with the new proc's pointer. fork_trampoline() 220 * will set up a stack to call fork_return(p, frame); to complete 221 * the return to user-mode. 222 */ 223} 224 225/* 226 * Intercept the return address from a freshly forked process that has NOT 227 * been scheduled yet. 228 * 229 * This is needed to make kernel threads stay in kernel mode. 230 */ 231void 232cpu_set_fork_handler(td, func, arg) 233 struct thread *td; 234 void (*func)(void *); 235 void *arg; 236{ 237 /* 238 * Note that the trap frame follows the args, so the function 239 * is really called like this: func(arg, frame); 240 */ 241 td->td_pcb->pcb_esi = (int) func; /* function */ 242 td->td_pcb->pcb_ebx = (int) arg; /* first arg */ 243} 244 245void 246cpu_exit(struct thread *td) 247{ 248 struct mdproc *mdp; 249 250 /* Reset pc->pcb_gs and %gs before possibly invalidating it. */ 251 mdp = &td->td_proc->p_md; 252 if (mdp->md_ldt) { 253 td->td_pcb->pcb_gs = _udatasel; 254 load_gs(_udatasel); 255 user_ldt_free(td); 256 } 257 reset_dbregs(); 258} 259 260void 261cpu_thread_exit(struct thread *td) 262{ 263 struct pcb *pcb = td->td_pcb; 264#ifdef DEV_NPX 265 npxexit(td); 266#endif 267 if (pcb->pcb_flags & PCB_DBREGS) { 268 /* 269 * disable all hardware breakpoints 270 */ 271 reset_dbregs(); 272 pcb->pcb_flags &= ~PCB_DBREGS; 273 } 274} 275 276void 277cpu_thread_clean(struct thread *td) 278{ 279 struct pcb *pcb; 280 281 pcb = td->td_pcb; 282 if (pcb->pcb_ext != 0) { 283 /* XXXKSE XXXSMP not SMP SAFE.. what locks do we have? */ 284 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */ 285 /* 286 * XXX do we need to move the TSS off the allocated pages 287 * before freeing them? (not done here) 288 */ 289 mtx_lock(&Giant); 290 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 291 ctob(IOPAGES + 1)); 292 mtx_unlock(&Giant); 293 pcb->pcb_ext = 0; 294 } 295} 296 297void 298cpu_sched_exit(td) 299 register struct thread *td; 300{ 301} 302 303void 304cpu_thread_setup(struct thread *td) 305{ 306 307 td->td_pcb = 308 (struct pcb *)(td->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1; 309 td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb - 16) - 1; 310} 311 312/* 313 * Initialize machine state (pcb and trap frame) for a new thread about to 314 * upcall. Pu t enough state in the new thread's PCB to get it to go back 315 * userret(), where we can intercept it again to set the return (upcall) 316 * Address and stack, along with those from upcals that are from other sources 317 * such as those generated in thread_userret() itself. 318 */ 319void 320cpu_set_upcall(struct thread *td, void *pcb) 321{ 322 struct pcb *pcb2; 323 324 /* Point the pcb to the top of the stack. */ 325 pcb2 = td->td_pcb; 326 327 /* 328 * Copy the upcall pcb. This loads kernel regs. 329 * Those not loaded individually below get their default 330 * values here. 331 * 332 * XXXKSE It might be a good idea to simply skip this as 333 * the values of the other registers may be unimportant. 334 * This would remove any requirement for knowing the KSE 335 * at this time (see the matching comment below for 336 * more analysis) (need a good safe default). 337 */ 338 bcopy(pcb, pcb2, sizeof(*pcb2)); 339 340 /* 341 * Create a new fresh stack for the new thread. 342 * The -16 is so we can expand the trapframe if we go to vm86. 343 * Don't forget to set this stack value into whatever supplies 344 * the address for the fault handlers. 345 * The contexts are filled in at the time we actually DO the 346 * upcall as only then do we know which KSE we got. 347 */ 348 td->td_frame = (struct trapframe *)((caddr_t)pcb2 - 16) - 1; 349 350 /* 351 * Set registers for trampoline to user mode. Leave space for the 352 * return address on stack. These are the kernel mode register values. 353 */ 354#ifdef PAE 355 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdpt); 356#else 357 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdir); 358#endif 359 pcb2->pcb_edi = 0; 360 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */ 361 pcb2->pcb_ebp = 0; 362 pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */ 363 pcb2->pcb_ebx = (int)td; /* trampoline arg */ 364 pcb2->pcb_eip = (int)fork_trampoline; 365 pcb2->pcb_psl &= ~(PSL_I); /* interrupts must be disabled */ 366 pcb2->pcb_gs = rgs(); 367 /* 368 * If we didn't copy the pcb, we'd need to do the following registers: 369 * pcb2->pcb_dr*: cloned above. 370 * pcb2->pcb_savefpu: cloned above. 371 * pcb2->pcb_flags: cloned above. 372 * pcb2->pcb_onfault: cloned above (always NULL here?). 373 * pcb2->pcb_gs: cloned above. XXXKSE ??? 374 * pcb2->pcb_ext: cleared below. 375 */ 376 pcb2->pcb_ext = NULL; 377} 378 379/* 380 * Set that machine state for performing an upcall that has to 381 * be done in thread_userret() so that those upcalls generated 382 * in thread_userret() itself can be done as well. 383 */ 384void 385cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku) 386{ 387 388 /* 389 * Do any extra cleaning that needs to be done. 390 * The thread may have optional components 391 * that are not present in a fresh thread. 392 * This may be a recycled thread so make it look 393 * as though it's newly allocated. 394 */ 395 cpu_thread_clean(td); 396 397 /* 398 * Set the trap frame to point at the beginning of the uts 399 * function. 400 */ 401 td->td_frame->tf_esp = 402 (int)ku->ku_stack.ss_sp + ku->ku_stack.ss_size - 16; 403 td->td_frame->tf_eip = (int)ku->ku_func; 404 405 /* 406 * Pass the address of the mailbox for this kse to the uts 407 * function as a parameter on the stack. 408 */ 409 suword((void *)(td->td_frame->tf_esp + sizeof(void *)), 410 (int)ku->ku_mailbox); 411} 412 413void 414cpu_wait(p) 415 struct proc *p; 416{ 417} 418 419/* 420 * Convert kernel VA to physical address 421 */ 422vm_paddr_t 423kvtop(void *addr) 424{ 425 vm_paddr_t pa; 426 427 pa = pmap_kextract((vm_offset_t)addr); 428 if (pa == 0) 429 panic("kvtop: zero page frame"); 430 return (pa); 431} 432 433/* 434 * Force reset the processor by invalidating the entire address space! 435 */ 436 437#ifdef SMP 438static void 439cpu_reset_proxy() 440{ 441 442 cpu_reset_proxy_active = 1; 443 while (cpu_reset_proxy_active == 1) 444 ; /* Wait for other cpu to see that we've started */ 445 stop_cpus((1<<cpu_reset_proxyid)); 446 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 447 DELAY(1000000); 448 cpu_reset_real(); 449} 450#endif 451 452void 453cpu_reset() 454{ 455#ifdef SMP 456 if (smp_active == 0) { 457 cpu_reset_real(); 458 /* NOTREACHED */ 459 } else { 460 461 u_int map; 462 int cnt; 463 printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid)); 464 465 map = PCPU_GET(other_cpus) & ~ stopped_cpus; 466 467 if (map != 0) { 468 printf("cpu_reset: Stopping other CPUs\n"); 469 stop_cpus(map); /* Stop all other CPUs */ 470 } 471 472 if (PCPU_GET(cpuid) == 0) { 473 DELAY(1000000); 474 cpu_reset_real(); 475 /* NOTREACHED */ 476 } else { 477 /* We are not BSP (CPU #0) */ 478 479 cpu_reset_proxyid = PCPU_GET(cpuid); 480 cpustop_restartfunc = cpu_reset_proxy; 481 cpu_reset_proxy_active = 0; 482 printf("cpu_reset: Restarting BSP\n"); 483 started_cpus = (1<<0); /* Restart CPU #0 */ 484 485 cnt = 0; 486 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 487 cnt++; /* Wait for BSP to announce restart */ 488 if (cpu_reset_proxy_active == 0) 489 printf("cpu_reset: Failed to restart BSP\n"); 490 enable_intr(); 491 cpu_reset_proxy_active = 2; 492 493 while (1); 494 /* NOTREACHED */ 495 } 496 } 497#else 498 cpu_reset_real(); 499#endif 500} 501 502static void 503cpu_reset_real() 504{ 505 506#ifdef PC98 507 /* 508 * Attempt to do a CPU reset via CPU reset port. 509 */ 510 disable_intr(); 511 if ((inb(0x35) & 0xa0) != 0xa0) { 512 outb(0x37, 0x0f); /* SHUT0 = 0. */ 513 outb(0x37, 0x0b); /* SHUT1 = 0. */ 514 } 515 outb(0xf0, 0x00); /* Reset. */ 516#else 517 /* 518 * Attempt to do a CPU reset via the keyboard controller, 519 * do not turn of the GateA20, as any machine that fails 520 * to do the reset here would then end up in no man's land. 521 */ 522 523#if !defined(BROKEN_KEYBOARD_RESET) 524 outb(IO_KBD + 4, 0xFE); 525 DELAY(500000); /* wait 0.5 sec to see if that did it */ 526 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 527 DELAY(1000000); /* wait 1 sec for printf to complete */ 528#endif 529#endif /* PC98 */ 530 /* force a shutdown by unmapping entire address space ! */ 531 bzero((caddr_t)PTD, NBPTD); 532 533 /* "good night, sweet prince .... <THUNK!>" */ 534 invltlb(); 535 /* NOTREACHED */ 536 while(1); 537} 538 539/* 540 * Software interrupt handler for queued VM system processing. 541 */ 542void 543swi_vm(void *dummy) 544{ 545 if (busdma_swi_pending != 0) 546 busdma_swi(); 547} 548 549/* 550 * Tell whether this address is in some physical memory region. 551 * Currently used by the kernel coredump code in order to avoid 552 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 553 * or other unpredictable behaviour. 554 */ 555 556int 557is_physical_memory(addr) 558 vm_offset_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