vm_machdep.c revision 112841
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 112841 2003-03-30 05:24:52Z jake $ 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 /*- 186 * pcb2->pcb_dr*: cloned above. 187 * pcb2->pcb_savefpu: cloned above. 188 * pcb2->pcb_flags: cloned above. 189 * pcb2->pcb_onfault: cloned above (always NULL here?). 190 * pcb2->pcb_gs: cloned above. 191 * pcb2->pcb_ext: cleared below. 192 */ 193 194 /* 195 * XXX don't copy the i/o pages. this should probably be fixed. 196 */ 197 pcb2->pcb_ext = 0; 198 199 /* Copy the LDT, if necessary. */ 200 mtx_lock_spin(&sched_lock); 201 if (mdp2->md_ldt != 0) { 202 if (flags & RFMEM) { 203 mdp2->md_ldt->ldt_refcnt++; 204 } else { 205 mdp2->md_ldt = user_ldt_alloc(mdp2, 206 mdp2->md_ldt->ldt_len); 207 if (mdp2->md_ldt == NULL) 208 panic("could not copy LDT"); 209 } 210 } 211 mtx_unlock_spin(&sched_lock); 212 213 /* 214 * Now, cpu_switch() can schedule the new process. 215 * pcb_esp is loaded pointing to the cpu_switch() stack frame 216 * containing the return address when exiting cpu_switch. 217 * This will normally be to fork_trampoline(), which will have 218 * %ebx loaded with the new proc's pointer. fork_trampoline() 219 * will set up a stack to call fork_return(p, frame); to complete 220 * the return to user-mode. 221 */ 222} 223 224/* 225 * Intercept the return address from a freshly forked process that has NOT 226 * been scheduled yet. 227 * 228 * This is needed to make kernel threads stay in kernel mode. 229 */ 230void 231cpu_set_fork_handler(td, func, arg) 232 struct thread *td; 233 void (*func)(void *); 234 void *arg; 235{ 236 /* 237 * Note that the trap frame follows the args, so the function 238 * is really called like this: func(arg, frame); 239 */ 240 td->td_pcb->pcb_esi = (int) func; /* function */ 241 td->td_pcb->pcb_ebx = (int) arg; /* first arg */ 242} 243 244void 245cpu_exit(struct thread *td) 246{ 247 struct mdproc *mdp; 248 249 mdp = &td->td_proc->p_md; 250 if (mdp->md_ldt) 251 user_ldt_free(td); 252 reset_dbregs(); 253} 254 255void 256cpu_thread_exit(struct thread *td) 257{ 258 struct pcb *pcb = td->td_pcb; 259#ifdef DEV_NPX 260 npxexit(td); 261#endif 262 if (pcb->pcb_flags & PCB_DBREGS) { 263 /* 264 * disable all hardware breakpoints 265 */ 266 reset_dbregs(); 267 pcb->pcb_flags &= ~PCB_DBREGS; 268 } 269} 270 271void 272cpu_thread_clean(struct thread *td) 273{ 274 struct pcb *pcb; 275 276 pcb = td->td_pcb; 277 if (pcb->pcb_ext != 0) { 278 /* XXXKSE XXXSMP not SMP SAFE.. what locks do we have? */ 279 /* if (pcb->pcb_ext->ext_refcount-- == 1) ?? */ 280 /* 281 * XXX do we need to move the TSS off the allocated pages 282 * before freeing them? (not done here) 283 */ 284 mtx_lock(&Giant); 285 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 286 ctob(IOPAGES + 1)); 287 mtx_unlock(&Giant); 288 pcb->pcb_ext = 0; 289 } 290} 291 292void 293cpu_sched_exit(td) 294 register struct thread *td; 295{ 296} 297 298void 299cpu_thread_setup(struct thread *td) 300{ 301 302 td->td_pcb = 303 (struct pcb *)(td->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1; 304 td->td_frame = (struct trapframe *)((caddr_t)td->td_pcb - 16) - 1; 305} 306 307/* 308 * Initialize machine state (pcb and trap frame) for a new thread about to 309 * upcall. Pu t enough state in the new thread's PCB to get it to go back 310 * userret(), where we can intercept it again to set the return (upcall) 311 * Address and stack, along with those from upcals that are from other sources 312 * such as those generated in thread_userret() itself. 313 */ 314void 315cpu_set_upcall(struct thread *td, void *pcb) 316{ 317 struct pcb *pcb2; 318 319 /* Point the pcb to the top of the stack. */ 320 pcb2 = td->td_pcb; 321 322 /* 323 * Copy the upcall pcb. This loads kernel regs. 324 * Those not loaded individually below get their default 325 * values here. 326 * 327 * XXXKSE It might be a good idea to simply skip this as 328 * the values of the other registers may be unimportant. 329 * This would remove any requirement for knowing the KSE 330 * at this time (see the matching comment below for 331 * more analysis) (need a good safe default). 332 */ 333 bcopy(pcb, pcb2, sizeof(*pcb2)); 334 335 /* 336 * Create a new fresh stack for the new thread. 337 * The -16 is so we can expand the trapframe if we go to vm86. 338 * Don't forget to set this stack value into whatever supplies 339 * the address for the fault handlers. 340 * The contexts are filled in at the time we actually DO the 341 * upcall as only then do we know which KSE we got. 342 */ 343 td->td_frame = (struct trapframe *)((caddr_t)pcb2 - 16) - 1; 344 345 /* 346 * Set registers for trampoline to user mode. Leave space for the 347 * return address on stack. These are the kernel mode register values. 348 */ 349#ifdef PAE 350 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdpt); 351#else 352 pcb2->pcb_cr3 = vtophys(vmspace_pmap(td->td_proc->p_vmspace)->pm_pdir); 353#endif 354 pcb2->pcb_edi = 0; 355 pcb2->pcb_esi = (int)fork_return; /* trampoline arg */ 356 pcb2->pcb_ebp = 0; 357 pcb2->pcb_esp = (int)td->td_frame - sizeof(void *); /* trampoline arg */ 358 pcb2->pcb_ebx = (int)td; /* trampoline arg */ 359 pcb2->pcb_eip = (int)fork_trampoline; 360 pcb2->pcb_psl &= ~(PSL_I); /* interrupts must be disabled */ 361 /* 362 * If we didn't copy the pcb, we'd need to do the following registers: 363 * pcb2->pcb_dr*: cloned above. 364 * pcb2->pcb_savefpu: cloned above. 365 * pcb2->pcb_flags: cloned above. 366 * pcb2->pcb_onfault: cloned above (always NULL here?). 367 * pcb2->pcb_gs: cloned above. XXXKSE ??? 368 * pcb2->pcb_ext: cleared below. 369 */ 370 pcb2->pcb_ext = NULL; 371} 372 373/* 374 * Set that machine state for performing an upcall that has to 375 * be done in thread_userret() so that those upcalls generated 376 * in thread_userret() itself can be done as well. 377 */ 378void 379cpu_set_upcall_kse(struct thread *td, struct kse_upcall *ku) 380{ 381 382 /* 383 * Do any extra cleaning that needs to be done. 384 * The thread may have optional components 385 * that are not present in a fresh thread. 386 * This may be a recycled thread so make it look 387 * as though it's newly allocated. 388 */ 389 cpu_thread_clean(td); 390 391 /* 392 * Set the trap frame to point at the beginning of the uts 393 * function. 394 */ 395 td->td_frame->tf_esp = 396 (int)ku->ku_stack.ss_sp + ku->ku_stack.ss_size - 16; 397 td->td_frame->tf_eip = (int)ku->ku_func; 398 399 /* 400 * Pass the address of the mailbox for this kse to the uts 401 * function as a parameter on the stack. 402 */ 403 suword((void *)(td->td_frame->tf_esp + sizeof(void *)), 404 (int)ku->ku_mailbox); 405} 406 407void 408cpu_wait(p) 409 struct proc *p; 410{ 411} 412 413/* 414 * Convert kernel VA to physical address 415 */ 416vm_paddr_t 417kvtop(void *addr) 418{ 419 vm_paddr_t pa; 420 421 pa = pmap_kextract((vm_offset_t)addr); 422 if (pa == 0) 423 panic("kvtop: zero page frame"); 424 return (pa); 425} 426 427/* 428 * Force reset the processor by invalidating the entire address space! 429 */ 430 431#ifdef SMP 432static void 433cpu_reset_proxy() 434{ 435 436 cpu_reset_proxy_active = 1; 437 while (cpu_reset_proxy_active == 1) 438 ; /* Wait for other cpu to see that we've started */ 439 stop_cpus((1<<cpu_reset_proxyid)); 440 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 441 DELAY(1000000); 442 cpu_reset_real(); 443} 444#endif 445 446void 447cpu_reset() 448{ 449#ifdef SMP 450 if (smp_active == 0) { 451 cpu_reset_real(); 452 /* NOTREACHED */ 453 } else { 454 455 u_int map; 456 int cnt; 457 printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid)); 458 459 map = PCPU_GET(other_cpus) & ~ stopped_cpus; 460 461 if (map != 0) { 462 printf("cpu_reset: Stopping other CPUs\n"); 463 stop_cpus(map); /* Stop all other CPUs */ 464 } 465 466 if (PCPU_GET(cpuid) == 0) { 467 DELAY(1000000); 468 cpu_reset_real(); 469 /* NOTREACHED */ 470 } else { 471 /* We are not BSP (CPU #0) */ 472 473 cpu_reset_proxyid = PCPU_GET(cpuid); 474 cpustop_restartfunc = cpu_reset_proxy; 475 cpu_reset_proxy_active = 0; 476 printf("cpu_reset: Restarting BSP\n"); 477 started_cpus = (1<<0); /* Restart CPU #0 */ 478 479 cnt = 0; 480 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 481 cnt++; /* Wait for BSP to announce restart */ 482 if (cpu_reset_proxy_active == 0) 483 printf("cpu_reset: Failed to restart BSP\n"); 484 enable_intr(); 485 cpu_reset_proxy_active = 2; 486 487 while (1); 488 /* NOTREACHED */ 489 } 490 } 491#else 492 cpu_reset_real(); 493#endif 494} 495 496static void 497cpu_reset_real() 498{ 499 500#ifdef PC98 501 /* 502 * Attempt to do a CPU reset via CPU reset port. 503 */ 504 disable_intr(); 505 if ((inb(0x35) & 0xa0) != 0xa0) { 506 outb(0x37, 0x0f); /* SHUT0 = 0. */ 507 outb(0x37, 0x0b); /* SHUT1 = 0. */ 508 } 509 outb(0xf0, 0x00); /* Reset. */ 510#else 511 /* 512 * Attempt to do a CPU reset via the keyboard controller, 513 * do not turn of the GateA20, as any machine that fails 514 * to do the reset here would then end up in no man's land. 515 */ 516 517#if !defined(BROKEN_KEYBOARD_RESET) 518 outb(IO_KBD + 4, 0xFE); 519 DELAY(500000); /* wait 0.5 sec to see if that did it */ 520 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 521 DELAY(1000000); /* wait 1 sec for printf to complete */ 522#endif 523#endif /* PC98 */ 524 /* force a shutdown by unmapping entire address space ! */ 525 bzero((caddr_t)PTD, NBPTD); 526 527 /* "good night, sweet prince .... <THUNK!>" */ 528 invltlb(); 529 /* NOTREACHED */ 530 while(1); 531} 532 533/* 534 * Software interrupt handler for queued VM system processing. 535 */ 536void 537swi_vm(void *dummy) 538{ 539 if (busdma_swi_pending != 0) 540 busdma_swi(); 541} 542 543/* 544 * Tell whether this address is in some physical memory region. 545 * Currently used by the kernel coredump code in order to avoid 546 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 547 * or other unpredictable behaviour. 548 */ 549 550int 551is_physical_memory(addr) 552 vm_offset_t addr; 553{ 554 555#ifdef DEV_ISA 556 /* The ISA ``memory hole''. */ 557 if (addr >= 0xa0000 && addr < 0x100000) 558 return 0; 559#endif 560 561 /* 562 * stuff other tests for known memory-mapped devices (PCI?) 563 * here 564 */ 565 566 return 1; 567} 568