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 * $FreeBSD$ 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/bio.h> 57 #include <sys/buf.h> 58 #include <sys/vnode.h> 59 #include <sys/vmmeter.h> 60 #include <sys/kernel.h> 61 #include <sys/ktr.h> 62 #include <sys/mutex.h> 63 #include <sys/smp.h> 64 #include <sys/sysctl.h> 65 #include <sys/unistd.h> 66 67 #include <machine/cpu.h> 68 #include <machine/md_var.h> 69 #include <machine/pcb.h> 70 #include <machine/pcb_ext.h> 71 #include <machine/vm86.h> 72 73 #include <vm/vm.h> 74 #include <vm/vm_param.h> 75 #include <sys/lock.h> 76 #include <vm/vm_kern.h> 77 #include <vm/vm_page.h> 78 #include <vm/vm_map.h> 79 #include <vm/vm_extern.h> 80 81 #include <sys/user.h> 82 83 #ifdef PC98 84 #include <pc98/pc98/pc98.h> 85 #else 86 #include <i386/isa/isa.h> 87 #endif 88 89 static void cpu_reset_real __P((void)); 90 #ifdef SMP 91 static void cpu_reset_proxy __P((void)); 92 static u_int cpu_reset_proxyid; 93 static volatile u_int cpu_reset_proxy_active; 94 #endif 95 extern int _ucodesel, _udatasel; 96 97 /* 98 * quick version of vm_fault 99 */ 100 int 101 vm_fault_quick(v, prot) 102 caddr_t v; 103 int prot; 104 { 105 int r; 106 107 if (prot & VM_PROT_WRITE) 108 r = subyte(v, fubyte(v)); 109 else 110 r = fubyte(v); 111 return(r); 112 } 113 114 /* 115 * Finish a fork operation, with process p2 nearly set up. 116 * Copy and update the pcb, set up the stack so that the child 117 * ready to run and return to user mode. 118 */ 119 void 120 cpu_fork(td1, p2, flags) 121 register struct thread *td1; 122 register struct proc *p2; 123 int flags; 124 { 125 register struct proc *p1; 126 struct thread *td2; 127 struct pcb *pcb2; 128 struct mdproc *mdp2; 129 #ifdef DEV_NPX 130 int savecrit; 131 #endif 132 133 p1 = td1->td_proc; 134 td2 = &p2->p_thread; 135 if ((flags & RFPROC) == 0) { 136 if ((flags & RFMEM) == 0) { 137 /* unshare user LDT */ 138 struct mdproc *mdp1 = &td1->td_proc->p_md; 139 struct proc_ldt *pldt = mdp1->md_ldt; 140 if (pldt && pldt->ldt_refcnt > 1) { 141 pldt = user_ldt_alloc(mdp1, pldt->ldt_len); 142 if (pldt == NULL) 143 panic("could not copy LDT"); 144 mdp1->md_ldt = pldt; 145 set_user_ldt(mdp1); 146 user_ldt_free(td1); 147 } 148 } 149 return; 150 } 151 152 /* Ensure that p1's pcb is up to date. */ 153 #ifdef DEV_NPX 154 if (td1 == curthread) 155 td1->td_pcb->pcb_gs = rgs(); 156 savecrit = critical_enter(); 157 if (PCPU_GET(npxthread) == td1) 158 npxsave(&td1->td_pcb->pcb_save); 159 critical_exit(savecrit); 160 #endif 161 162 /* Point the pcb to the top of the stack */ 163 pcb2 = (struct pcb *)(td2->td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1; 164 td2->td_pcb = pcb2; 165 166 /* Copy p1's pcb */ 167 bcopy(td1->td_pcb, pcb2, sizeof(*pcb2)); 168 169 /* Point mdproc and then copy over td1's contents */ 170 mdp2 = &td2->td_proc->p_md; 171 bcopy(&td1->td_proc->p_md, mdp2, sizeof(*mdp2)); 172 173 /* 174 * Create a new fresh stack for the new process. 175 * Copy the trap frame for the return to user mode as if from a 176 * syscall. This copies most of the user mode register values. 177 * The -16 is so we can expand the trapframe if we go to vm86. 178 */ 179 td2->td_frame = (struct trapframe *)((caddr_t)td2->td_pcb - 16) - 1; 180 bcopy(td1->td_frame, td2->td_frame, sizeof(struct trapframe)); 181 182 td2->td_frame->tf_eax = 0; /* Child returns zero */ 183 td2->td_frame->tf_eflags &= ~PSL_C; /* success */ 184 td2->td_frame->tf_edx = 1; 185 186 /* 187 * Set registers for trampoline to user mode. Leave space for the 188 * return address on stack. These are the kernel mode register values. 189 */ 190 pcb2->pcb_cr3 = vtophys(vmspace_pmap(p2->p_vmspace)->pm_pdir); 191 pcb2->pcb_edi = 0; 192 pcb2->pcb_esi = (int)fork_return; /* fork_trampoline argument */ 193 pcb2->pcb_ebp = 0; 194 pcb2->pcb_esp = (int)td2->td_frame - sizeof(void *); 195 pcb2->pcb_ebx = (int)td2; /* fork_trampoline argument */ 196 pcb2->pcb_eip = (int)fork_trampoline; 197 /*- 198 * pcb2->pcb_dr*: cloned above. 199 * pcb2->pcb_savefpu: cloned above. 200 * pcb2->pcb_flags: cloned above. 201 * pcb2->pcb_onfault: cloned above (always NULL here?). 202 * pcb2->pcb_gs: cloned above. 203 * pcb2->pcb_ext: cleared below. 204 */ 205 206 /* 207 * XXX don't copy the i/o pages. this should probably be fixed. 208 */ 209 pcb2->pcb_ext = 0; 210 211 /* Copy the LDT, if necessary. */ 212 mtx_lock_spin(&sched_lock); 213 if (mdp2->md_ldt != 0) { 214 if (flags & RFMEM) { 215 mdp2->md_ldt->ldt_refcnt++; 216 } else { 217 mdp2->md_ldt = user_ldt_alloc(mdp2, 218 mdp2->md_ldt->ldt_len); 219 if (mdp2->md_ldt == NULL) 220 panic("could not copy LDT"); 221 } 222 } 223 mtx_unlock_spin(&sched_lock); 224 225 /* 226 * Now, cpu_switch() can schedule the new process. 227 * pcb_esp is loaded pointing to the cpu_switch() stack frame 228 * containing the return address when exiting cpu_switch. 229 * This will normally be to fork_trampoline(), which will have 230 * %ebx loaded with the new proc's pointer. fork_trampoline() 231 * will set up a stack to call fork_return(p, frame); to complete 232 * the return to user-mode. 233 */ 234 } 235 236 /* 237 * Intercept the return address from a freshly forked process that has NOT 238 * been scheduled yet. 239 * 240 * This is needed to make kernel threads stay in kernel mode. 241 */ 242 void 243 cpu_set_fork_handler(td, func, arg) 244 struct thread *td; 245 void (*func) __P((void *)); 246 void *arg; 247 { 248 /* 249 * Note that the trap frame follows the args, so the function 250 * is really called like this: func(arg, frame); 251 */ 252 td->td_pcb->pcb_esi = (int) func; /* function */ 253 td->td_pcb->pcb_ebx = (int) arg; /* first arg */ 254 } 255 256 void 257 cpu_exit(td) 258 register struct thread *td; 259 { 260 struct pcb *pcb = td->td_pcb; 261 struct mdproc *mdp = &td->td_proc->p_md; 262 #ifdef DEV_NPX 263 npxexit(td); 264 #endif 265 if (pcb->pcb_ext != 0) { 266 /* 267 * XXX do we need to move the TSS off the allocated pages 268 * before freeing them? (not done here) 269 */ 270 kmem_free(kernel_map, (vm_offset_t)pcb->pcb_ext, 271 ctob(IOPAGES + 1)); 272 pcb->pcb_ext = 0; 273 } 274 if (mdp->md_ldt) 275 user_ldt_free(td); 276 if (pcb->pcb_flags & PCB_DBREGS) { 277 /* 278 * disable all hardware breakpoints 279 */ 280 reset_dbregs(); 281 pcb->pcb_flags &= ~PCB_DBREGS; 282 } 283 } 284 285 void 286 cpu_wait(p) 287 struct proc *p; 288 { 289 } 290 291 /* 292 * Dump the machine specific header information at the start of a core dump. 293 */ 294 int 295 cpu_coredump(td, vp, cred) 296 struct thread *td; 297 struct vnode *vp; 298 struct ucred *cred; 299 { 300 struct proc *p = td->td_proc; 301 int error; 302 caddr_t tempuser; 303 304 tempuser = malloc(ctob(UAREA_PAGES + KSTACK_PAGES), M_TEMP, M_WAITOK | M_ZERO); 305 if (!tempuser) 306 return EINVAL; 307 308 bcopy(p->p_uarea, tempuser, sizeof(struct user)); 309 #if 0 /* XXXKSE - broken, fixme!!!!! td_frame is in kstack! */ 310 bcopy(td->td_frame, 311 tempuser + ((caddr_t) td->td_frame - (caddr_t) p->p_uarea), 312 sizeof(struct trapframe)); 313 #endif 314 315 error = vn_rdwr(UIO_WRITE, vp, (caddr_t) tempuser, 316 ctob(UAREA_PAGES + KSTACK_PAGES), 317 (off_t)0, UIO_SYSSPACE, IO_UNIT, cred, (int *)NULL, td); 318 319 free(tempuser, M_TEMP); 320 321 return error; 322 } 323 324 /* 325 * Convert kernel VA to physical address 326 */ 327 u_long 328 kvtop(void *addr) 329 { 330 vm_offset_t va; 331 332 va = pmap_kextract((vm_offset_t)addr); 333 if (va == 0) 334 panic("kvtop: zero page frame"); 335 return((int)va); 336 } 337 338 /* 339 * Map an IO request into kernel virtual address space. 340 * 341 * All requests are (re)mapped into kernel VA space. 342 * Notice that we use b_bufsize for the size of the buffer 343 * to be mapped. b_bcount might be modified by the driver. 344 */ 345 void 346 vmapbuf(bp) 347 register struct buf *bp; 348 { 349 register caddr_t addr, v, kva; 350 vm_offset_t pa; 351 int pidx; 352 struct vm_page *m; 353 354 GIANT_REQUIRED; 355 356 if ((bp->b_flags & B_PHYS) == 0) 357 panic("vmapbuf"); 358 359 for (v = bp->b_saveaddr, 360 addr = (caddr_t)trunc_page((vm_offset_t)bp->b_data), 361 pidx = 0; 362 addr < bp->b_data + bp->b_bufsize; 363 addr += PAGE_SIZE, v += PAGE_SIZE, pidx++) { 364 /* 365 * Do the vm_fault if needed; do the copy-on-write thing 366 * when reading stuff off device into memory. 367 */ 368 vm_fault_quick((addr >= bp->b_data) ? addr : bp->b_data, 369 (bp->b_iocmd == BIO_READ)?(VM_PROT_READ|VM_PROT_WRITE):VM_PROT_READ); 370 pa = trunc_page(pmap_kextract((vm_offset_t) addr)); 371 if (pa == 0) 372 panic("vmapbuf: page not present"); 373 m = PHYS_TO_VM_PAGE(pa); 374 vm_page_hold(m); 375 bp->b_pages[pidx] = m; 376 } 377 if (pidx > btoc(MAXPHYS)) 378 panic("vmapbuf: mapped more than MAXPHYS"); 379 pmap_qenter((vm_offset_t)bp->b_saveaddr, bp->b_pages, pidx); 380 381 kva = bp->b_saveaddr; 382 bp->b_npages = pidx; 383 bp->b_saveaddr = bp->b_data; 384 bp->b_data = kva + (((vm_offset_t) bp->b_data) & PAGE_MASK); 385 } 386 387 /* 388 * Free the io map PTEs associated with this IO operation. 389 * We also invalidate the TLB entries and restore the original b_addr. 390 */ 391 void 392 vunmapbuf(bp) 393 register struct buf *bp; 394 { 395 int pidx; 396 int npages; 397 vm_page_t *m; 398 399 GIANT_REQUIRED; 400 401 if ((bp->b_flags & B_PHYS) == 0) 402 panic("vunmapbuf"); 403 404 npages = bp->b_npages; 405 pmap_qremove(trunc_page((vm_offset_t)bp->b_data), 406 npages); 407 m = bp->b_pages; 408 for (pidx = 0; pidx < npages; pidx++) 409 vm_page_unhold(*m++); 410 411 bp->b_data = bp->b_saveaddr; 412 } 413 414 /* 415 * Force reset the processor by invalidating the entire address space! 416 */ 417 418 #ifdef SMP 419 static void 420 cpu_reset_proxy() 421 { 422 423 cpu_reset_proxy_active = 1; 424 while (cpu_reset_proxy_active == 1) 425 ; /* Wait for other cpu to see that we've started */ 426 stop_cpus((1<<cpu_reset_proxyid)); 427 printf("cpu_reset_proxy: Stopped CPU %d\n", cpu_reset_proxyid); 428 DELAY(1000000); 429 cpu_reset_real(); 430 } 431 #endif 432 433 void 434 cpu_reset() 435 { 436 #ifdef SMP 437 if (smp_active == 0) { 438 cpu_reset_real(); 439 /* NOTREACHED */ 440 } else { 441 442 u_int map; 443 int cnt; 444 printf("cpu_reset called on cpu#%d\n", PCPU_GET(cpuid)); 445 446 map = PCPU_GET(other_cpus) & ~ stopped_cpus; 447 448 if (map != 0) { 449 printf("cpu_reset: Stopping other CPUs\n"); 450 stop_cpus(map); /* Stop all other CPUs */ 451 } 452 453 if (PCPU_GET(cpuid) == 0) { 454 DELAY(1000000); 455 cpu_reset_real(); 456 /* NOTREACHED */ 457 } else { 458 /* We are not BSP (CPU #0) */ 459 460 cpu_reset_proxyid = PCPU_GET(cpuid); 461 cpustop_restartfunc = cpu_reset_proxy; 462 cpu_reset_proxy_active = 0; 463 printf("cpu_reset: Restarting BSP\n"); 464 started_cpus = (1<<0); /* Restart CPU #0 */ 465 466 cnt = 0; 467 while (cpu_reset_proxy_active == 0 && cnt < 10000000) 468 cnt++; /* Wait for BSP to announce restart */ 469 if (cpu_reset_proxy_active == 0) 470 printf("cpu_reset: Failed to restart BSP\n"); 471 enable_intr(); 472 cpu_reset_proxy_active = 2; 473 474 while (1); 475 /* NOTREACHED */ 476 } 477 } 478 #else 479 cpu_reset_real(); 480 #endif 481 } 482 483 static void 484 cpu_reset_real() 485 { 486 487 #ifdef PC98 488 /* 489 * Attempt to do a CPU reset via CPU reset port. 490 */ 491 disable_intr(); 492 if ((inb(0x35) & 0xa0) != 0xa0) { 493 outb(0x37, 0x0f); /* SHUT0 = 0. */ 494 outb(0x37, 0x0b); /* SHUT1 = 0. */ 495 } 496 outb(0xf0, 0x00); /* Reset. */ 497 #else 498 /* 499 * Attempt to do a CPU reset via the keyboard controller, 500 * do not turn of the GateA20, as any machine that fails 501 * to do the reset here would then end up in no man's land. 502 */ 503 504 #if !defined(BROKEN_KEYBOARD_RESET) 505 outb(IO_KBD + 4, 0xFE); 506 DELAY(500000); /* wait 0.5 sec to see if that did it */ 507 printf("Keyboard reset did not work, attempting CPU shutdown\n"); 508 DELAY(1000000); /* wait 1 sec for printf to complete */ 509 #endif 510 #endif /* PC98 */ 511 /* force a shutdown by unmapping entire address space ! */ 512 bzero((caddr_t) PTD, PAGE_SIZE); 513 514 /* "good night, sweet prince .... <THUNK!>" */ 515 invltlb(); 516 /* NOTREACHED */ 517 while(1); 518 } 519 520 int 521 grow_stack(p, sp) 522 struct proc *p; 523 u_int sp; 524 { 525 int rv; 526 527 rv = vm_map_growstack (p, sp); 528 if (rv != KERN_SUCCESS) 529 return (0); 530 531 return (1); 532 } 533 534 /* 535 * Software interrupt handler for queued VM system processing. 536 */ 537 void 538 swi_vm(void *dummy) 539 { 540 if (busdma_swi_pending != 0) 541 busdma_swi(); 542 } 543 544 /* 545 * Tell whether this address is in some physical memory region. 546 * Currently used by the kernel coredump code in order to avoid 547 * dumping the ``ISA memory hole'' which could cause indefinite hangs, 548 * or other unpredictable behaviour. 549 */ 550 551 int 552 is_physical_memory(addr) 553 vm_offset_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