| 1 | /*- |
| 2 | * Copyright (c) 1990 The Regents of the University of California. |
| 3 | * All rights reserved. |
| 4 | * |
| 5 | * This code is derived from software contributed to Berkeley by |
| 6 | * the University of Utah, and William Jolitz. |
| 7 | * |
| 8 | * %sccs.include.386.c% |
| 9 | * |
| 10 | * @(#)vm_machdep.c 5.4 (Berkeley) %G% |
| 11 | */ |
| 12 | |
| 13 | /* |
| 14 | * Copyright (c) 1989, 1990 William F. Jolitz |
| 15 | */ |
| 16 | |
| 17 | /* |
| 18 | * Copyright (c) 1988 University of Utah. |
| 19 | * All rights reserved. The Utah Software License Agreement |
| 20 | * specifies the terms and conditions for redistribution. |
| 21 | * |
| 22 | * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ |
| 23 | */ |
| 24 | /* |
| 25 | * Copyright (c) 1982, 1986 Regents of the University of California. |
| 26 | * All rights reserved. The Berkeley software License Agreement |
| 27 | * specifies the terms and conditions for redistribution. |
| 28 | * |
| 29 | * @(#)vm_machdep.c 7.1 (Berkeley) 6/5/86 |
| 30 | */ |
| 31 | |
| 32 | #include "pte.h" |
| 33 | |
| 34 | #include "param.h" |
| 35 | #include "systm.h" |
| 36 | #include "dir.h" |
| 37 | #include "user.h" |
| 38 | #include "proc.h" |
| 39 | #include "cmap.h" |
| 40 | #include "mount.h" |
| 41 | #include "vm.h" |
| 42 | #include "text.h" |
| 43 | |
| 44 | #include "buf.h" |
| 45 | |
| 46 | /* |
| 47 | * Set a red zone in the kernel stack after the u. area. |
| 48 | */ |
| 49 | setredzone(pte, vaddr) |
| 50 | u_short *pte; |
| 51 | caddr_t vaddr; |
| 52 | { |
| 53 | /* eventually do this by setting up an expand-down stack segment |
| 54 | for ss0: selector, allowing stack access down to top of u. |
| 55 | this means though that protection violations need to be handled |
| 56 | thru a double fault exception that must do an integral task |
| 57 | switch to a known good context, within which a dump can be |
| 58 | taken. a sensible scheme might be to save the initial context |
| 59 | used by sched (that has physical memory mapped 1:1 at bottom) |
| 60 | and take the dump while still in mapped mode */ |
| 61 | } |
| 62 | |
| 63 | /* |
| 64 | * Check for valid program size |
| 65 | * NB - Check data and data growth separately as they may overflow |
| 66 | * when summed together. |
| 67 | */ |
| 68 | chksize(ts, ids, uds, ss) |
| 69 | unsigned ts, ids, uds, ss; |
| 70 | { |
| 71 | extern unsigned maxtsize; |
| 72 | |
| 73 | if (ctob(ts) > maxtsize || |
| 74 | ctob(ids) > u.u_rlimit[RLIMIT_DATA].rlim_cur || |
| 75 | ctob(uds) > u.u_rlimit[RLIMIT_DATA].rlim_cur || |
| 76 | ctob(ids + uds) > u.u_rlimit[RLIMIT_DATA].rlim_cur || |
| 77 | ctob(ss) > u.u_rlimit[RLIMIT_STACK].rlim_cur) { |
| 78 | u.u_error = ENOMEM; |
| 79 | return (1); |
| 80 | } |
| 81 | return (0); |
| 82 | } |
| 83 | |
| 84 | /*ARGSUSED*/ |
| 85 | newptes(pte, v, size) |
| 86 | struct pte *pte; |
| 87 | u_int v; |
| 88 | register int size; |
| 89 | { |
| 90 | register caddr_t a; |
| 91 | |
| 92 | #ifdef lint |
| 93 | pte = pte; |
| 94 | #endif |
| 95 | tlbflush(); |
| 96 | } |
| 97 | |
| 98 | /* |
| 99 | * Change protection codes of text segment. |
| 100 | * Have to flush translation buffer since this |
| 101 | * affect virtual memory mapping of current process. |
| 102 | */ |
| 103 | chgprot(addr, tprot) |
| 104 | caddr_t addr; |
| 105 | long tprot; |
| 106 | { |
| 107 | unsigned v; |
| 108 | int tp; |
| 109 | register struct pte *pte; |
| 110 | register struct cmap *c; |
| 111 | |
| 112 | v = clbase(btop(addr)); |
| 113 | if (!isatsv(u.u_procp, v)) { |
| 114 | u.u_error = EFAULT; |
| 115 | return (0); |
| 116 | } |
| 117 | tp = vtotp(u.u_procp, v); |
| 118 | pte = tptopte(u.u_procp, tp); |
| 119 | if (pte->pg_fod == 0 && pte->pg_pfnum) { |
| 120 | c = &cmap[pgtocm(pte->pg_pfnum)]; |
| 121 | if (c->c_blkno && c->c_mdev != MSWAPX) |
| 122 | munhash(mount[c->c_mdev].m_dev, |
| 123 | (daddr_t)(u_long)c->c_blkno); |
| 124 | } |
| 125 | *(u_int *)pte &= ~PG_PROT; |
| 126 | *(u_int *)pte |= tprot; |
| 127 | tlbflush(); |
| 128 | return (1); |
| 129 | } |
| 130 | |
| 131 | settprot(tprot) |
| 132 | long tprot; |
| 133 | { |
| 134 | register u_int *ptaddr, i; |
| 135 | |
| 136 | ptaddr = (u_int *)u.u_procp->p_p0br; |
| 137 | for (i = 0; i < u.u_tsize; i++) { |
| 138 | ptaddr[i] &= ~PG_PROT; |
| 139 | ptaddr[i] |= tprot; |
| 140 | } |
| 141 | tlbflush(); |
| 142 | } |
| 143 | |
| 144 | /* |
| 145 | * Simulate effect of VAX region length registers. |
| 146 | * The one case where we must do anything is if a region has shrunk. |
| 147 | * In that case we must invalidate all the PTEs for the no longer valid VAs. |
| 148 | */ |
| 149 | setptlr(region, nlen) |
| 150 | int nlen; |
| 151 | { |
| 152 | register struct pte *pte; |
| 153 | register int change; |
| 154 | int olen; |
| 155 | |
| 156 | if (region == 0) { |
| 157 | olen = u.u_pcb.pcb_p0lr; |
| 158 | u.u_pcb.pcb_p0lr = nlen; |
| 159 | } else { |
| 160 | olen = P1PAGES - u.u_pcb.pcb_p1lr; |
| 161 | u.u_pcb.pcb_p1lr = nlen; |
| 162 | nlen = P1PAGES - nlen; |
| 163 | } |
| 164 | if ((change = olen - nlen) <= 0) |
| 165 | return; |
| 166 | if (region == 0) |
| 167 | pte = u.u_pcb.pcb_p0br + u.u_pcb.pcb_p0lr; |
| 168 | else |
| 169 | pte = u.u_pcb.pcb_p1br + u.u_pcb.pcb_p1lr - change; |
| 170 | do { |
| 171 | *(u_int *)pte++ = 0; |
| 172 | } while (--change); |
| 173 | /* short cut newptes */ |
| 174 | tlbflush(); |
| 175 | } |
| 176 | |
| 177 | /* |
| 178 | * Map `size' bytes of physical memory starting at `paddr' into |
| 179 | * kernel VA space using PTEs starting at `pte'. Read/write and |
| 180 | * cache-inhibit status are specified by `prot'. |
| 181 | */ |
| 182 | physaccess(pte, paddr, size, prot) |
| 183 | register struct pte *pte; |
| 184 | caddr_t paddr; |
| 185 | register int size; |
| 186 | { |
| 187 | register u_int page; |
| 188 | |
| 189 | page = (u_int)paddr & PG_FRAME; |
| 190 | for (size = btoc(size); size; size--) { |
| 191 | *(int *)pte = PG_V | prot | page; |
| 192 | page += NBPG; |
| 193 | pte++; |
| 194 | } |
| 195 | tlbflush(); |
| 196 | } |
| 197 | |
| 198 | /* |
| 199 | * Move pages from one kernel virtual address to another. |
| 200 | * Both addresses are assumed to reside in the Sysmap, |
| 201 | * and size must be a multiple of CLSIZE. |
| 202 | */ |
| 203 | pagemove(from, to, size) |
| 204 | register caddr_t from, to; |
| 205 | int size; |
| 206 | { |
| 207 | register struct pte *fpte, *tpte; |
| 208 | |
| 209 | if (size % CLBYTES) |
| 210 | panic("pagemove"); |
| 211 | fpte = &Sysmap[btop(from-0xfe000000)]; |
| 212 | tpte = &Sysmap[btop(to-0xfe000000)]; |
| 213 | while (size > 0) { |
| 214 | *tpte++ = *fpte; |
| 215 | *(int *)fpte++ = 0; |
| 216 | from += NBPG; |
| 217 | to += NBPG; |
| 218 | size -= NBPG; |
| 219 | } |
| 220 | tlbflush(); |
| 221 | } |
| 222 | |
| 223 | /* |
| 224 | * The probe[rw] routines should probably be redone in assembler |
| 225 | * for efficiency. |
| 226 | */ |
| 227 | prober(addr) |
| 228 | register u_int addr; |
| 229 | { |
| 230 | register int page; |
| 231 | register struct proc *p; |
| 232 | |
| 233 | if (addr >= USRSTACK) |
| 234 | return(0); |
| 235 | p = u.u_procp; |
| 236 | page = btop(addr); |
| 237 | if (page < dptov(p, p->p_dsize) || page > sptov(p, p->p_ssize)) |
| 238 | return(1); |
| 239 | return(0); |
| 240 | } |
| 241 | |
| 242 | probew(addr) |
| 243 | register u_int addr; |
| 244 | { |
| 245 | register int page; |
| 246 | register struct proc *p; |
| 247 | |
| 248 | if (addr >= USRSTACK) |
| 249 | return(0); |
| 250 | p = u.u_procp; |
| 251 | page = btop(addr); |
| 252 | if (page < dptov(p, p->p_dsize) || page > sptov(p, p->p_ssize)) |
| 253 | return((*(int *)vtopte(p, page) & PG_PROT) == PG_UW); |
| 254 | return(0); |
| 255 | } |
| 256 | |
| 257 | /* |
| 258 | * NB: assumes a physically contiguous kernel page table |
| 259 | * (makes life a LOT simpler). |
| 260 | */ |
| 261 | kernacc(addr, count, rw) |
| 262 | register u_int addr; |
| 263 | int count, rw; |
| 264 | { |
| 265 | register struct pde *pde; |
| 266 | register struct pte *pte; |
| 267 | register int ix, cnt; |
| 268 | extern long Syssize; |
| 269 | |
| 270 | if (count <= 0) |
| 271 | return(0); |
| 272 | pde = (struct pde *)((u_int)u.u_procp->p_p0br + u.u_procp->p_szpt * NBPG); |
| 273 | ix = (addr & PD_MASK) >> PD_SHIFT; |
| 274 | cnt = ((addr + count + (1 << PD_SHIFT) - 1) & PD_MASK) >> PD_SHIFT; |
| 275 | cnt -= ix; |
| 276 | for (pde += ix; cnt; cnt--, pde++) |
| 277 | if (pde->pd_v == 0) |
| 278 | return(0); |
| 279 | ix = btop(addr-0xfe000000); |
| 280 | cnt = btop(addr-0xfe000000+count+NBPG-1); |
| 281 | if (cnt > (int)&Syssize) |
| 282 | return(0); |
| 283 | cnt -= ix; |
| 284 | for (pte = &Sysmap[ix]; cnt; cnt--, pte++) |
| 285 | if (pte->pg_v == 0 /*|| (rw == B_WRITE && pte->pg_prot == 1)*/) |
| 286 | return(0); |
| 287 | return(1); |
| 288 | } |
| 289 | |
| 290 | useracc(addr, count, rw) |
| 291 | register u_int addr; |
| 292 | int count, rw; |
| 293 | { |
| 294 | register int (*func)(); |
| 295 | register u_int addr2; |
| 296 | extern int prober(), probew(); |
| 297 | |
| 298 | if (count <= 0) |
| 299 | return(0); |
| 300 | addr2 = addr; |
| 301 | addr += count; |
| 302 | func = (rw == B_READ) ? prober : probew; |
| 303 | do { |
| 304 | if ((*func)(addr2) == 0) |
| 305 | return(0); |
| 306 | addr2 = (addr2 + NBPG) & ~PGOFSET; |
| 307 | } while (addr2 < addr); |
| 308 | return(1); |
| 309 | } |
| 310 | |
| 311 | /* |
| 312 | * Convert kernel VA to physical address |
| 313 | */ |
| 314 | kvtop(addr) |
| 315 | register u_int addr; |
| 316 | { |
| 317 | register int pf; |
| 318 | |
| 319 | pf = Sysmap[btop(addr-0xfe000000)].pg_pfnum; |
| 320 | if (pf == 0) |
| 321 | panic("kvtop: zero page frame"); |
| 322 | return((u_int)ptob(pf) + (addr & PGOFSET)); |
| 323 | } |
| 324 | |
| 325 | struct pde * |
| 326 | vtopde(p, va) |
| 327 | register struct proc *p; |
| 328 | register u_int va; |
| 329 | { |
| 330 | register struct pde *pde; |
| 331 | |
| 332 | pde = (struct pde *)((u_int)p->p_p0br + p->p_szpt * NBPG); |
| 333 | return(pde + ((va & PD_MASK) >> PD_SHIFT)); |
| 334 | } |
| 335 | |
| 336 | |
| 337 | initcr3(p) |
| 338 | register struct proc *p; |
| 339 | { |
| 340 | return(ctob(Usrptmap[btokmx(p->p_p0br+p->p_szpt*NPTEPG)].pg_pfnum)); |
| 341 | /*return((int)Usrptmap[btokmx(p->p_p0br) + p->p_szpt].pg_pfnum);*/ |
| 342 | } |
| 343 | |
| 344 | /* |
| 345 | * Initialize page directory table to reflect PTEs in Usrptmap. |
| 346 | * Page directory table address is given by Usrptmap index of p_szpt. |
| 347 | * [used by vgetpt for kernal mode entries, and ptexpand for user mode entries] |
| 348 | */ |
| 349 | initpdt(p) |
| 350 | register struct proc *p; |
| 351 | { |
| 352 | register int i, k, sz; |
| 353 | register struct pde *pde, *toppde; |
| 354 | extern struct pde *vtopde(); |
| 355 | extern Sysbase; |
| 356 | |
| 357 | /* clear entire map */ |
| 358 | pde = vtopde(p, 0); |
| 359 | /*bzero(pde, NBPG); */ |
| 360 | /* map kernel */ |
| 361 | pde = vtopde(p, &Sysbase); |
| 362 | for (i = 0; i < 5; i++, pde++) { |
| 363 | *(int *)pde = PG_UW | PG_V; |
| 364 | pde->pd_pfnum = btoc((unsigned) Sysmap & ~0xfe000000)+i; |
| 365 | } |
| 366 | /* map u dot */ |
| 367 | pde = vtopde(p, &u); |
| 368 | *(int *)pde = PG_UW | PG_V; |
| 369 | pde->pd_pfnum = Usrptmap[btokmx(p->p_addr)].pg_pfnum; |
| 370 | |
| 371 | /* otherwise, fill in user map */ |
| 372 | k = btokmx(p->p_p0br); |
| 373 | pde = vtopde(p, 0); |
| 374 | toppde = vtopde(p, &u); |
| 375 | |
| 376 | /* text and data */ |
| 377 | sz = ctopt(p->p_tsize + p->p_dsize); |
| 378 | for (i = 0; i < sz; i++, pde++) { |
| 379 | *(int *)pde = PG_UW | PG_V; |
| 380 | pde->pd_pfnum = Usrptmap[k++].pg_pfnum; |
| 381 | } |
| 382 | /* |
| 383 | * Bogus! The kernelmap may map unused PT pages |
| 384 | * (since we don't shrink PTs) so we need to skip over |
| 385 | * those PDEs. We should really free the unused PT |
| 386 | * pages in expand(). |
| 387 | */ |
| 388 | sz += ctopt(p->p_ssize+UPAGES); |
| 389 | if (sz < p->p_szpt) |
| 390 | k += p->p_szpt - sz; |
| 391 | /* hole */ |
| 392 | sz = NPTEPG - ctopt(p->p_ssize + UPAGES + btoc(&Sysbase)); |
| 393 | for ( ; i < sz; i++, pde++) |
| 394 | *(int *)pde = 0; |
| 395 | /* stack and u-area */ |
| 396 | sz = NPTEPG - ctopt(UPAGES + btoc(&Sysbase)); |
| 397 | for ( ; i < sz; i++, pde++) { |
| 398 | *(int *)pde = PG_UW | PG_V; |
| 399 | pde->pd_pfnum = Usrptmap[k++].pg_pfnum; |
| 400 | } |
| 401 | return(initcr3(p)); |
| 402 | } |
| 403 | |
| 404 | #ifdef notdef |
| 405 | /* |
| 406 | * Allocate wired-down, non-paged, cache-inhibited pages in kernel |
| 407 | * virtual memory and clear them |
| 408 | */ |
| 409 | caddr_t |
| 410 | cimemall(n) |
| 411 | int n; |
| 412 | { |
| 413 | register int npg, a; |
| 414 | register struct pte *pte; |
| 415 | extern struct map *kernelmap; |
| 416 | |
| 417 | npg = clrnd(btoc(n)); |
| 418 | a = rmalloc(kernelmap, (long)npg); |
| 419 | if (a == 0) |
| 420 | return ((caddr_t)0); |
| 421 | pte = &Usrptmap[a]; |
| 422 | (void) vmemall(pte, npg, &proc[0], CSYS); |
| 423 | while (--npg >= 0) { |
| 424 | *(int *)pte |= (PG_V|PG_KW|PG_CI); |
| 425 | clearseg((unsigned)pte->pg_pfnum); |
| 426 | pte++; |
| 427 | } |
| 428 | TBIAS(); |
| 429 | return ((caddr_t)kmxtob(a)); |
| 430 | } |
| 431 | #endif |
| 432 | |
| 433 | extern char usrio[]; |
| 434 | extern struct pte Usriomap[]; |
| 435 | struct map *useriomap; |
| 436 | int usriowanted; |
| 437 | |
| 438 | /* |
| 439 | * Map an IO request into kernel virtual address space. Requests fall into |
| 440 | * one of five catagories: |
| 441 | * |
| 442 | * B_PHYS|B_UAREA: User u-area swap. |
| 443 | * Address is relative to start of u-area (p_addr). |
| 444 | * B_PHYS|B_PAGET: User page table swap. |
| 445 | * Address is a kernel VA in usrpt (Usrptmap). |
| 446 | * B_PHYS|B_DIRTY: Dirty page push. |
| 447 | * Address is a VA in proc2's address space. |
| 448 | * B_PHYS|B_PGIN: Kernel pagein of user pages. |
| 449 | * Address is VA in user's address space. |
| 450 | * B_PHYS: User "raw" IO request. |
| 451 | * Address is VA in user's address space. |
| 452 | * |
| 453 | * All requests are (re)mapped into kernel VA space via the useriomap |
| 454 | * (a name with only slightly more meaning than "kernelmap") |
| 455 | */ |
| 456 | vmapbuf(bp) |
| 457 | register struct buf *bp; |
| 458 | { |
| 459 | register int npf, a; |
| 460 | register caddr_t addr; |
| 461 | register struct pte *pte, *iopte; |
| 462 | register long flags = bp->b_flags; |
| 463 | struct proc *p; |
| 464 | int off, s; |
| 465 | |
| 466 | if ((flags & B_PHYS) == 0) |
| 467 | panic("vmapbuf"); |
| 468 | /* |
| 469 | * Find PTEs for the area to be mapped |
| 470 | */ |
| 471 | p = flags&B_DIRTY ? &proc[2] : bp->b_proc; |
| 472 | addr = bp->b_un.b_addr; |
| 473 | if (flags & B_UAREA) |
| 474 | pte = &p->p_addr[btop(addr)]; |
| 475 | else if (flags & B_PAGET) |
| 476 | pte = &Usrptmap[btokmx((struct pte *)addr)]; |
| 477 | else |
| 478 | pte = vtopte(p, btop(addr)); |
| 479 | /* |
| 480 | * Allocate some kernel PTEs and load them |
| 481 | */ |
| 482 | off = (int)addr & PGOFSET; |
| 483 | npf = btoc(bp->b_bcount + off); |
| 484 | s = splbio(); |
| 485 | while ((a = rmalloc(useriomap, npf)) == 0) { |
| 486 | usriowanted = 1; |
| 487 | sleep((caddr_t)useriomap, PSWP); |
| 488 | } |
| 489 | splx(s); |
| 490 | iopte = &Usriomap[a]; |
| 491 | addr = bp->b_un.b_addr = (caddr_t)(usrio + (a << PGSHIFT)) + off; |
| 492 | a = btop(addr); |
| 493 | while (npf--) { |
| 494 | mapin(iopte, a, pte->pg_pfnum, PG_V); |
| 495 | iopte++, pte++; |
| 496 | a++; |
| 497 | } |
| 498 | tlbflush(); |
| 499 | } |
| 500 | |
| 501 | /* |
| 502 | * Free the io map PTEs associated with this IO operation. |
| 503 | * We also invalidate the TLB entries. |
| 504 | */ |
| 505 | vunmapbuf(bp) |
| 506 | register struct buf *bp; |
| 507 | { |
| 508 | register int a, npf; |
| 509 | register caddr_t addr = bp->b_un.b_addr; |
| 510 | register struct pte *pte; |
| 511 | int s; |
| 512 | |
| 513 | if ((bp->b_flags & B_PHYS) == 0) |
| 514 | panic("vunmapbuf"); |
| 515 | a = (int)(addr - usrio) >> PGSHIFT; |
| 516 | npf = btoc(bp->b_bcount + ((int)addr & PGOFSET)); |
| 517 | s = splbio(); |
| 518 | rmfree(useriomap, npf, a); |
| 519 | if (usriowanted) { |
| 520 | usriowanted = 0; |
| 521 | wakeup((caddr_t)useriomap); |
| 522 | } |
| 523 | splx(s); |
| 524 | pte = &Usriomap[a]; |
| 525 | while (npf--) { |
| 526 | *(int *)pte = 0; |
| 527 | addr += NBPG; |
| 528 | pte++; |
| 529 | } |
| 530 | /* |
| 531 | * If we just completed a dirty page push, we must reconstruct |
| 532 | * the original b_addr since cleanup() needs it. |
| 533 | */ |
| 534 | if (bp->b_flags & B_DIRTY) { |
| 535 | a = ((bp - swbuf) * CLSIZE) * KLMAX; |
| 536 | bp->b_un.b_addr = (caddr_t)ctob(dptov(&proc[2], a)); |
| 537 | } |
| 538 | tlbflush(); |
| 539 | } |