pmap.c revision 1.13
1 /* $NetBSD: pmap.c,v 1.13 2003/08/12 05:06:57 matt Exp $ */ 2 /*- 3 * Copyright (c) 2001 The NetBSD Foundation, Inc. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to The NetBSD Foundation 7 * by Matt Thomas <matt (at) 3am-software.com> of Allegro Networks, Inc. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the NetBSD 20 * Foundation, Inc. and its contributors. 21 * 4. Neither the name of The NetBSD Foundation nor the names of its 22 * contributors may be used to endorse or promote products derived 23 * from this software without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 26 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 27 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 28 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 29 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 30 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 31 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 32 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 33 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 34 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 35 * POSSIBILITY OF SUCH DAMAGE. 36 */ 37 38 /* 39 * Copyright (C) 1995, 1996 Wolfgang Solfrank. 40 * Copyright (C) 1995, 1996 TooLs GmbH. 41 * All rights reserved. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. All advertising materials mentioning features or use of this software 52 * must display the following acknowledgement: 53 * This product includes software developed by TooLs GmbH. 54 * 4. The name of TooLs GmbH may not be used to endorse or promote products 55 * derived from this software without specific prior written permission. 56 * 57 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR 58 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 59 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 60 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 61 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 62 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 63 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 64 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 65 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 66 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.13 2003/08/12 05:06:57 matt Exp $"); 71 72 #include "opt_altivec.h" 73 #include "opt_pmap.h" 74 #include <sys/param.h> 75 #include <sys/malloc.h> 76 #include <sys/proc.h> 77 #include <sys/user.h> 78 #include <sys/pool.h> 79 #include <sys/queue.h> 80 #include <sys/device.h> /* for evcnt */ 81 #include <sys/systm.h> 82 83 #if __NetBSD_Version__ < 105010000 84 #include <vm/vm.h> 85 #include <vm/vm_kern.h> 86 #define splvm() splimp() 87 #endif 88 89 #include <uvm/uvm.h> 90 91 #include <machine/pcb.h> 92 #include <machine/powerpc.h> 93 #include <powerpc/spr.h> 94 #include <powerpc/oea/sr_601.h> 95 #if __NetBSD_Version__ > 105010000 96 #include <powerpc/oea/bat.h> 97 #else 98 #include <powerpc/bat.h> 99 #endif 100 101 #if defined(DEBUG) || defined(PMAPCHECK) 102 #define STATIC 103 #else 104 #define STATIC static 105 #endif 106 107 #ifdef ALTIVEC 108 int pmap_use_altivec; 109 #endif 110 111 volatile struct pteg *pmap_pteg_table; 112 unsigned int pmap_pteg_cnt; 113 unsigned int pmap_pteg_mask; 114 paddr_t pmap_memlimit = -PAGE_SIZE; /* there is no limit */ 115 116 struct pmap kernel_pmap_; 117 unsigned int pmap_pages_stolen; 118 u_long pmap_pte_valid; 119 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 120 u_long pmap_pvo_enter_depth; 121 u_long pmap_pvo_remove_depth; 122 #endif 123 124 int physmem; 125 #ifndef MSGBUFADDR 126 extern paddr_t msgbuf_paddr; 127 #endif 128 129 static struct mem_region *mem, *avail; 130 static u_int mem_cnt, avail_cnt; 131 132 #ifdef __HAVE_PMAP_PHYSSEG 133 /* 134 * This is a cache of referenced/modified bits. 135 * Bits herein are shifted by ATTRSHFT. 136 */ 137 #define ATTR_SHFT 4 138 struct pmap_physseg pmap_physseg; 139 #endif 140 141 /* 142 * The following structure is exactly 32 bytes long (one cacheline). 143 */ 144 struct pvo_entry { 145 LIST_ENTRY(pvo_entry) pvo_vlink; /* Link to common virt page */ 146 TAILQ_ENTRY(pvo_entry) pvo_olink; /* Link to overflow entry */ 147 struct pte pvo_pte; /* Prebuilt PTE */ 148 pmap_t pvo_pmap; /* ptr to owning pmap */ 149 vaddr_t pvo_vaddr; /* VA of entry */ 150 #define PVO_PTEGIDX_MASK 0x0007 /* which PTEG slot */ 151 #define PVO_PTEGIDX_VALID 0x0008 /* slot is valid */ 152 #define PVO_WIRED 0x0010 /* PVO entry is wired */ 153 #define PVO_MANAGED 0x0020 /* PVO e. for managed page */ 154 #define PVO_EXECUTABLE 0x0040 /* PVO e. for executable page */ 155 #define PVO_ENTER_INSERT 0 /* PVO has been removed */ 156 #define PVO_SPILL_UNSET 1 /* PVO has been evicted */ 157 #define PVO_SPILL_SET 2 /* PVO has been spilled */ 158 #define PVO_SPILL_INSERT 3 /* PVO has been inserted */ 159 #define PVO_PMAP_PAGE_PROTECT 4 /* PVO has changed */ 160 #define PVO_PMAP_PROTECT 5 /* PVO has changed */ 161 #define PVO_REMOVE 6 /* PVO has been removed */ 162 #define PVO_WHERE_MASK 15 163 #define PVO_WHERE_SHFT 8 164 }; 165 #define PVO_VADDR(pvo) ((pvo)->pvo_vaddr & ~ADDR_POFF) 166 #define PVO_ISEXECUTABLE(pvo) ((pvo)->pvo_vaddr & PVO_EXECUTABLE) 167 #define PVO_PTEGIDX_GET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK) 168 #define PVO_PTEGIDX_ISSET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID) 169 #define PVO_PTEGIDX_CLR(pvo) \ 170 ((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK))) 171 #define PVO_PTEGIDX_SET(pvo,i) \ 172 ((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID)) 173 #define PVO_WHERE(pvo,w) \ 174 ((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \ 175 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT)) 176 177 TAILQ_HEAD(pvo_tqhead, pvo_entry); 178 struct pvo_tqhead *pmap_pvo_table; /* pvo entries by ptegroup index */ 179 struct pvo_head pmap_pvo_kunmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged); /* list of unmanaged pages */ 180 struct pvo_head pmap_pvo_unmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged); /* list of unmanaged pages */ 181 182 struct pool pmap_pool; /* pool for pmap structures */ 183 struct pool pmap_upvo_pool; /* pool for pvo entries for unmanaged pages */ 184 struct pool pmap_mpvo_pool; /* pool for pvo entries for managed pages */ 185 186 /* 187 * We keep a cache of unmanaged pages to be used for pvo entries for 188 * unmanaged pages. 189 */ 190 struct pvo_page { 191 SIMPLEQ_ENTRY(pvo_page) pvop_link; 192 }; 193 SIMPLEQ_HEAD(pvop_head, pvo_page); 194 struct pvop_head pmap_upvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_upvop_head); 195 struct pvop_head pmap_mpvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_mpvop_head); 196 u_long pmap_upvop_free; 197 u_long pmap_upvop_maxfree; 198 u_long pmap_mpvop_free; 199 u_long pmap_mpvop_maxfree; 200 201 STATIC void *pmap_pool_ualloc(struct pool *, int); 202 STATIC void *pmap_pool_malloc(struct pool *, int); 203 204 STATIC void pmap_pool_ufree(struct pool *, void *); 205 STATIC void pmap_pool_mfree(struct pool *, void *); 206 207 static struct pool_allocator pmap_pool_mallocator = { 208 pmap_pool_malloc, pmap_pool_mfree, 0, 209 }; 210 211 static struct pool_allocator pmap_pool_uallocator = { 212 pmap_pool_ualloc, pmap_pool_ufree, 0, 213 }; 214 215 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 216 void pmap_pte_print(volatile struct pte *); 217 #endif 218 219 #ifdef DDB 220 void pmap_pteg_check(void); 221 void pmap_pteg_dist(void); 222 void pmap_print_pte(pmap_t, vaddr_t); 223 void pmap_print_mmuregs(void); 224 #endif 225 226 #if defined(DEBUG) || defined(PMAPCHECK) 227 #ifdef PMAPCHECK 228 int pmapcheck = 1; 229 #else 230 int pmapcheck = 0; 231 #endif 232 void pmap_pvo_verify(void); 233 STATIC void pmap_pvo_check(const struct pvo_entry *); 234 #define PMAP_PVO_CHECK(pvo) \ 235 do { \ 236 if (pmapcheck) \ 237 pmap_pvo_check(pvo); \ 238 } while (0) 239 #else 240 #define PMAP_PVO_CHECK(pvo) do { } while (/*CONSTCOND*/0) 241 #endif 242 STATIC int pmap_pte_insert(int, struct pte *); 243 STATIC int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *, 244 vaddr_t, paddr_t, register_t, int); 245 STATIC void pmap_pvo_remove(struct pvo_entry *, int); 246 STATIC struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *); 247 STATIC volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int); 248 249 STATIC void tlbia(void); 250 251 STATIC void pmap_release(pmap_t); 252 STATIC void *pmap_boot_find_memory(psize_t, psize_t, int); 253 254 #define VSID_NBPW (sizeof(uint32_t) * 8) 255 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW]; 256 257 static int pmap_initialized; 258 259 #if defined(DEBUG) || defined(PMAPDEBUG) 260 #define PMAPDEBUG_BOOT 0x0001 261 #define PMAPDEBUG_PTE 0x0002 262 #define PMAPDEBUG_EXEC 0x0008 263 #define PMAPDEBUG_PVOENTER 0x0010 264 #define PMAPDEBUG_PVOREMOVE 0x0020 265 #define PMAPDEBUG_ACTIVATE 0x0100 266 #define PMAPDEBUG_CREATE 0x0200 267 #define PMAPDEBUG_ENTER 0x1000 268 #define PMAPDEBUG_KENTER 0x2000 269 #define PMAPDEBUG_KREMOVE 0x4000 270 #define PMAPDEBUG_REMOVE 0x8000 271 unsigned int pmapdebug = 0; 272 # define DPRINTF(x) printf x 273 # define DPRINTFN(n, x) if (pmapdebug & PMAPDEBUG_ ## n) printf x 274 #else 275 # define DPRINTF(x) 276 # define DPRINTFN(n, x) 277 #endif 278 279 280 #ifdef PMAPCOUNTERS 281 #define PMAPCOUNT(ev) ((pmap_evcnt_ ## ev).ev_count++) 282 #define PMAPCOUNT2(ev) ((ev).ev_count++) 283 284 struct evcnt pmap_evcnt_mappings = 285 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 286 "pmap", "pages mapped"); 287 struct evcnt pmap_evcnt_unmappings = 288 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 289 "pmap", "pages unmapped"); 290 291 struct evcnt pmap_evcnt_kernel_mappings = 292 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 293 "pmap", "kernel pages mapped"); 294 struct evcnt pmap_evcnt_kernel_unmappings = 295 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_kernel_mappings, 296 "pmap", "kernel pages unmapped"); 297 298 struct evcnt pmap_evcnt_mappings_replaced = 299 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 300 "pmap", "page mappings replaced"); 301 302 struct evcnt pmap_evcnt_exec_mappings = 303 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 304 "pmap", "exec pages mapped"); 305 struct evcnt pmap_evcnt_exec_cached = 306 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 307 "pmap", "exec pages cached"); 308 309 struct evcnt pmap_evcnt_exec_synced = 310 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 311 "pmap", "exec pages synced"); 312 struct evcnt pmap_evcnt_exec_synced_clear_modify = 313 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 314 "pmap", "exec pages synced (CM)"); 315 316 struct evcnt pmap_evcnt_exec_uncached_page_protect = 317 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 318 "pmap", "exec pages uncached (PP)"); 319 struct evcnt pmap_evcnt_exec_uncached_clear_modify = 320 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 321 "pmap", "exec pages uncached (CM)"); 322 struct evcnt pmap_evcnt_exec_uncached_zero_page = 323 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 324 "pmap", "exec pages uncached (ZP)"); 325 struct evcnt pmap_evcnt_exec_uncached_copy_page = 326 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 327 "pmap", "exec pages uncached (CP)"); 328 329 struct evcnt pmap_evcnt_updates = 330 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 331 "pmap", "updates"); 332 struct evcnt pmap_evcnt_collects = 333 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 334 "pmap", "collects"); 335 struct evcnt pmap_evcnt_copies = 336 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 337 "pmap", "copies"); 338 339 struct evcnt pmap_evcnt_ptes_spilled = 340 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 341 "pmap", "ptes spilled from overflow"); 342 struct evcnt pmap_evcnt_ptes_unspilled = 343 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 344 "pmap", "ptes not spilled"); 345 struct evcnt pmap_evcnt_ptes_evicted = 346 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 347 "pmap", "ptes evicted"); 348 349 struct evcnt pmap_evcnt_ptes_primary[8] = { 350 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 351 "pmap", "ptes added at primary[0]"), 352 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 353 "pmap", "ptes added at primary[1]"), 354 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 355 "pmap", "ptes added at primary[2]"), 356 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 357 "pmap", "ptes added at primary[3]"), 358 359 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 360 "pmap", "ptes added at primary[4]"), 361 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 362 "pmap", "ptes added at primary[5]"), 363 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 364 "pmap", "ptes added at primary[6]"), 365 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 366 "pmap", "ptes added at primary[7]"), 367 }; 368 struct evcnt pmap_evcnt_ptes_secondary[8] = { 369 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 370 "pmap", "ptes added at secondary[0]"), 371 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 372 "pmap", "ptes added at secondary[1]"), 373 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 374 "pmap", "ptes added at secondary[2]"), 375 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 376 "pmap", "ptes added at secondary[3]"), 377 378 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 379 "pmap", "ptes added at secondary[4]"), 380 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 381 "pmap", "ptes added at secondary[5]"), 382 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 383 "pmap", "ptes added at secondary[6]"), 384 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 385 "pmap", "ptes added at secondary[7]"), 386 }; 387 struct evcnt pmap_evcnt_ptes_removed = 388 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 389 "pmap", "ptes removed"); 390 struct evcnt pmap_evcnt_ptes_changed = 391 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 392 "pmap", "ptes changed"); 393 394 /* 395 * From pmap_subr.c 396 */ 397 extern struct evcnt pmap_evcnt_zeroed_pages; 398 extern struct evcnt pmap_evcnt_copied_pages; 399 extern struct evcnt pmap_evcnt_idlezeroed_pages; 400 #else 401 #define PMAPCOUNT(ev) ((void) 0) 402 #define PMAPCOUNT2(ev) ((void) 0) 403 #endif 404 405 #define TLBIE(va) __asm __volatile("tlbie %0" :: "r"(va)) 406 #define TLBSYNC() __asm __volatile("tlbsync") 407 #define SYNC() __asm __volatile("sync") 408 #define EIEIO() __asm __volatile("eieio") 409 #define MFMSR() mfmsr() 410 #define MTMSR(psl) mtmsr(psl) 411 #define MFPVR() mfpvr() 412 #define MFSRIN(va) mfsrin(va) 413 #define MFTB() mfrtcltbl() 414 415 static __inline register_t 416 mfsrin(vaddr_t va) 417 { 418 register_t sr; 419 __asm __volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va)); 420 return sr; 421 } 422 423 static __inline register_t 424 pmap_interrupts_off(void) 425 { 426 register_t msr = MFMSR(); 427 if (msr & PSL_EE) 428 MTMSR(msr & ~PSL_EE); 429 return msr; 430 } 431 432 static void 433 pmap_interrupts_restore(register_t msr) 434 { 435 if (msr & PSL_EE) 436 MTMSR(msr); 437 } 438 439 static __inline u_int32_t 440 mfrtcltbl(void) 441 { 442 443 if ((MFPVR() >> 16) == MPC601) 444 return (mfrtcl() >> 7); 445 else 446 return (mftbl()); 447 } 448 449 /* 450 * These small routines may have to be replaced, 451 * if/when we support processors other that the 604. 452 */ 453 454 void 455 tlbia(void) 456 { 457 caddr_t i; 458 459 SYNC(); 460 /* 461 * Why not use "tlbia"? Because not all processors implement it. 462 * 463 * This needs to be a per-cpu callback to do the appropriate thing 464 * for the CPU. XXX 465 */ 466 for (i = 0; i < (caddr_t)0x00040000; i += 0x00001000) { 467 TLBIE(i); 468 EIEIO(); 469 SYNC(); 470 } 471 TLBSYNC(); 472 SYNC(); 473 } 474 475 static __inline register_t 476 va_to_vsid(const struct pmap *pm, vaddr_t addr) 477 { 478 return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID); 479 } 480 481 static __inline register_t 482 va_to_pteg(const struct pmap *pm, vaddr_t addr) 483 { 484 register_t hash; 485 486 hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT); 487 return hash & pmap_pteg_mask; 488 } 489 490 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 491 /* 492 * Given a PTE in the page table, calculate the VADDR that hashes to it. 493 * The only bit of magic is that the top 4 bits of the address doesn't 494 * technically exist in the PTE. But we know we reserved 4 bits of the 495 * VSID for it so that's how we get it. 496 */ 497 static vaddr_t 498 pmap_pte_to_va(volatile const struct pte *pt) 499 { 500 vaddr_t va; 501 uintptr_t ptaddr = (uintptr_t) pt; 502 503 if (pt->pte_hi & PTE_HID) 504 ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg)); 505 506 /* PPC Bits 10-19 */ 507 va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff; 508 va <<= ADDR_PIDX_SHFT; 509 510 /* PPC Bits 4-9 */ 511 va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT; 512 513 /* PPC Bits 0-3 */ 514 va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; 515 516 return va; 517 } 518 #endif 519 520 static __inline struct pvo_head * 521 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p) 522 { 523 #ifdef __HAVE_VM_PAGE_MD 524 struct vm_page *pg; 525 526 pg = PHYS_TO_VM_PAGE(pa); 527 if (pg_p != NULL) 528 *pg_p = pg; 529 if (pg == NULL) 530 return &pmap_pvo_unmanaged; 531 return &pg->mdpage.mdpg_pvoh; 532 #endif 533 #ifdef __HAVE_PMAP_PHYSSEG 534 int bank, pg; 535 536 bank = vm_physseg_find(atop(pa), &pg); 537 if (pg_p != NULL) 538 *pg_p = pg; 539 if (bank == -1) 540 return &pmap_pvo_unmanaged; 541 return &vm_physmem[bank].pmseg.pvoh[pg]; 542 #endif 543 } 544 545 static __inline struct pvo_head * 546 vm_page_to_pvoh(struct vm_page *pg) 547 { 548 #ifdef __HAVE_VM_PAGE_MD 549 return &pg->mdpage.mdpg_pvoh; 550 #endif 551 #ifdef __HAVE_PMAP_PHYSSEG 552 return pa_to_pvoh(VM_PAGE_TO_PHYS(pg), NULL); 553 #endif 554 } 555 556 557 #ifdef __HAVE_PMAP_PHYSSEG 558 static __inline char * 559 pa_to_attr(paddr_t pa) 560 { 561 int bank, pg; 562 563 bank = vm_physseg_find(atop(pa), &pg); 564 if (bank == -1) 565 return NULL; 566 return &vm_physmem[bank].pmseg.attrs[pg]; 567 } 568 #endif 569 570 static __inline void 571 pmap_attr_clear(struct vm_page *pg, int ptebit) 572 { 573 #ifdef __HAVE_PMAP_PHYSSEG 574 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) &= ~(ptebit >> ATTR_SHFT); 575 #endif 576 #ifdef __HAVE_VM_PAGE_MD 577 pg->mdpage.mdpg_attrs &= ~ptebit; 578 #endif 579 } 580 581 static __inline int 582 pmap_attr_fetch(struct vm_page *pg) 583 { 584 #ifdef __HAVE_PMAP_PHYSSEG 585 return *pa_to_attr(VM_PAGE_TO_PHYS(pg)) << ATTR_SHFT; 586 #endif 587 #ifdef __HAVE_VM_PAGE_MD 588 return pg->mdpage.mdpg_attrs; 589 #endif 590 } 591 592 static __inline void 593 pmap_attr_save(struct vm_page *pg, int ptebit) 594 { 595 #ifdef __HAVE_PMAP_PHYSSEG 596 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) |= (ptebit >> ATTR_SHFT); 597 #endif 598 #ifdef __HAVE_VM_PAGE_MD 599 pg->mdpage.mdpg_attrs |= ptebit; 600 #endif 601 } 602 603 static __inline int 604 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt) 605 { 606 if (pt->pte_hi == pvo_pt->pte_hi 607 #if 0 608 && ((pt->pte_lo ^ pvo_pt->pte_lo) & 609 ~(PTE_REF|PTE_CHG)) == 0 610 #endif 611 ) 612 return 1; 613 return 0; 614 } 615 616 static __inline void 617 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo) 618 { 619 /* 620 * Construct the PTE. Default to IMB initially. Valid bit 621 * only gets set when the real pte is set in memory. 622 * 623 * Note: Don't set the valid bit for correct operation of tlb update. 624 */ 625 pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT) 626 | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API); 627 pt->pte_lo = pte_lo; 628 } 629 630 static __inline void 631 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt) 632 { 633 pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG); 634 } 635 636 static __inline void 637 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit) 638 { 639 /* 640 * As shown in Section 7.6.3.2.3 641 */ 642 pt->pte_lo &= ~ptebit; 643 TLBIE(va); 644 SYNC(); 645 EIEIO(); 646 TLBSYNC(); 647 SYNC(); 648 } 649 650 static __inline void 651 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt) 652 { 653 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 654 if (pvo_pt->pte_hi & PTE_VALID) 655 panic("pte_set: setting an already valid pte %p", pvo_pt); 656 #endif 657 pvo_pt->pte_hi |= PTE_VALID; 658 /* 659 * Update the PTE as defined in section 7.6.3.1 660 * Note that the REF/CHG bits are from pvo_pt and thus should 661 * have been saved so this routine can restore them (if desired). 662 */ 663 pt->pte_lo = pvo_pt->pte_lo; 664 EIEIO(); 665 pt->pte_hi = pvo_pt->pte_hi; 666 SYNC(); 667 pmap_pte_valid++; 668 } 669 670 static __inline void 671 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 672 { 673 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 674 if ((pvo_pt->pte_hi & PTE_VALID) == 0) 675 panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt); 676 if ((pt->pte_hi & PTE_VALID) == 0) 677 panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt); 678 #endif 679 680 pvo_pt->pte_hi &= ~PTE_VALID; 681 /* 682 * Force the ref & chg bits back into the PTEs. 683 */ 684 SYNC(); 685 /* 686 * Invalidate the pte ... (Section 7.6.3.3) 687 */ 688 pt->pte_hi &= ~PTE_VALID; 689 SYNC(); 690 TLBIE(va); 691 SYNC(); 692 EIEIO(); 693 TLBSYNC(); 694 SYNC(); 695 /* 696 * Save the ref & chg bits ... 697 */ 698 pmap_pte_synch(pt, pvo_pt); 699 pmap_pte_valid--; 700 } 701 702 static __inline void 703 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 704 { 705 /* 706 * Invalidate the PTE 707 */ 708 pmap_pte_unset(pt, pvo_pt, va); 709 pmap_pte_set(pt, pvo_pt); 710 } 711 712 /* 713 * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx 714 * (either primary or secondary location). 715 * 716 * Note: both the destination and source PTEs must not have PTE_VALID set. 717 */ 718 719 STATIC int 720 pmap_pte_insert(int ptegidx, struct pte *pvo_pt) 721 { 722 volatile struct pte *pt; 723 int i; 724 725 #if defined(DEBUG) 726 DPRINTFN(PTE, ("pmap_pte_insert: idx 0x%x, pte 0x%lx 0x%lx\n", 727 ptegidx, pvo_pt->pte_hi, pvo_pt->pte_lo)); 728 #endif 729 /* 730 * First try primary hash. 731 */ 732 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 733 if ((pt->pte_hi & PTE_VALID) == 0) { 734 pvo_pt->pte_hi &= ~PTE_HID; 735 pmap_pte_set(pt, pvo_pt); 736 return i; 737 } 738 } 739 740 /* 741 * Now try secondary hash. 742 */ 743 ptegidx ^= pmap_pteg_mask; 744 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 745 if ((pt->pte_hi & PTE_VALID) == 0) { 746 pvo_pt->pte_hi |= PTE_HID; 747 pmap_pte_set(pt, pvo_pt); 748 return i; 749 } 750 } 751 return -1; 752 } 753 754 /* 755 * Spill handler. 756 * 757 * Tries to spill a page table entry from the overflow area. 758 * This runs in either real mode (if dealing with a exception spill) 759 * or virtual mode when dealing with manually spilling one of the 760 * kernel's pte entries. In either case, interrupts are already 761 * disabled. 762 */ 763 int 764 pmap_pte_spill(struct pmap *pm, vaddr_t addr) 765 { 766 struct pvo_entry *source_pvo, *victim_pvo, *next_pvo; 767 struct pvo_entry *pvo; 768 struct pvo_tqhead *pvoh, *vpvoh; 769 int ptegidx, i, j; 770 volatile struct pteg *pteg; 771 volatile struct pte *pt; 772 773 ptegidx = va_to_pteg(pm, addr); 774 775 /* 776 * Have to substitute some entry. Use the primary hash for this. 777 * Use low bits of timebase as random generator. Make sure we are 778 * not picking a kernel pte for replacement. 779 */ 780 pteg = &pmap_pteg_table[ptegidx]; 781 i = MFTB() & 7; 782 for (j = 0; j < 8; j++) { 783 pt = &pteg->pt[i]; 784 if ((pt->pte_hi & PTE_VALID) == 0 || 785 VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT) 786 != KERNEL_VSIDBITS) 787 break; 788 i = (i + 1) & 7; 789 } 790 KASSERT(j < 8); 791 792 source_pvo = NULL; 793 victim_pvo = NULL; 794 pvoh = &pmap_pvo_table[ptegidx]; 795 TAILQ_FOREACH(pvo, pvoh, pvo_olink) { 796 797 /* 798 * We need to find pvo entry for this address... 799 */ 800 PMAP_PVO_CHECK(pvo); /* sanity check */ 801 802 /* 803 * If we haven't found the source and we come to a PVO with 804 * a valid PTE, then we know we can't find it because all 805 * evicted PVOs always are first in the list. 806 */ 807 if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID)) 808 break; 809 if (source_pvo == NULL && pm == pvo->pvo_pmap && 810 addr == PVO_VADDR(pvo)) { 811 812 /* 813 * Now we have found the entry to be spilled into the 814 * pteg. Attempt to insert it into the page table. 815 */ 816 j = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 817 if (j >= 0) { 818 PVO_PTEGIDX_SET(pvo, j); 819 PMAP_PVO_CHECK(pvo); /* sanity check */ 820 PVO_WHERE(pvo, SPILL_INSERT); 821 pvo->pvo_pmap->pm_evictions--; 822 PMAPCOUNT(ptes_spilled); 823 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 824 ? pmap_evcnt_ptes_secondary 825 : pmap_evcnt_ptes_primary)[j]); 826 827 /* 828 * Since we keep the evicted entries at the 829 * from of the PVO list, we need move this 830 * (now resident) PVO after the evicted 831 * entries. 832 */ 833 next_pvo = TAILQ_NEXT(pvo, pvo_olink); 834 835 /* 836 * If we don't have to move (either we were the 837 * last entry or the next entry was valid), 838 * don't change our position. Otherwise 839 * move ourselves to the tail of the queue. 840 */ 841 if (next_pvo != NULL && 842 !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) { 843 TAILQ_REMOVE(pvoh, pvo, pvo_olink); 844 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 845 } 846 return 1; 847 } 848 source_pvo = pvo; 849 if (victim_pvo != NULL) 850 break; 851 } 852 853 /* 854 * We also need the pvo entry of the victim we are replacing 855 * so save the R & C bits of the PTE. 856 */ 857 if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL && 858 pmap_pte_compare(pt, &pvo->pvo_pte)) { 859 vpvoh = pvoh; 860 victim_pvo = pvo; 861 if (source_pvo != NULL) 862 break; 863 } 864 } 865 866 if (source_pvo == NULL) { 867 PMAPCOUNT(ptes_unspilled); 868 return 0; 869 } 870 871 if (victim_pvo == NULL) { 872 if ((pt->pte_hi & PTE_HID) == 0) 873 panic("pmap_pte_spill: victim p-pte (%p) has " 874 "no pvo entry!", pt); 875 876 /* 877 * If this is a secondary PTE, we need to search 878 * its primary pvo bucket for the matching PVO. 879 */ 880 vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; 881 TAILQ_FOREACH(pvo, vpvoh, pvo_olink) { 882 PMAP_PVO_CHECK(pvo); /* sanity check */ 883 884 /* 885 * We also need the pvo entry of the victim we are 886 * replacing so save the R & C bits of the PTE. 887 */ 888 if (pmap_pte_compare(pt, &pvo->pvo_pte)) { 889 victim_pvo = pvo; 890 break; 891 } 892 } 893 if (victim_pvo == NULL) 894 panic("pmap_pte_spill: victim s-pte (%p) has " 895 "no pvo entry!", pt); 896 } 897 898 /* 899 * The victim should be not be a kernel PVO/PTE entry. 900 */ 901 KASSERT(victim_pvo->pvo_pmap != pmap_kernel()); 902 KASSERT(PVO_PTEGIDX_ISSET(victim_pvo)); 903 KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i); 904 905 /* 906 * We are invalidating the TLB entry for the EA for the 907 * we are replacing even though its valid; If we don't 908 * we lose any ref/chg bit changes contained in the TLB 909 * entry. 910 */ 911 source_pvo->pvo_pte.pte_hi &= ~PTE_HID; 912 913 /* 914 * To enforce the PVO list ordering constraint that all 915 * evicted entries should come before all valid entries, 916 * move the source PVO to the tail of its list and the 917 * victim PVO to the head of its list (which might not be 918 * the same list, if the victim was using the secondary hash). 919 */ 920 TAILQ_REMOVE(pvoh, source_pvo, pvo_olink); 921 TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink); 922 TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink); 923 TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink); 924 pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr); 925 pmap_pte_set(pt, &source_pvo->pvo_pte); 926 victim_pvo->pvo_pmap->pm_evictions++; 927 source_pvo->pvo_pmap->pm_evictions--; 928 PVO_WHERE(victim_pvo, SPILL_UNSET); 929 PVO_WHERE(source_pvo, SPILL_SET); 930 931 PVO_PTEGIDX_CLR(victim_pvo); 932 PVO_PTEGIDX_SET(source_pvo, i); 933 PMAPCOUNT2(pmap_evcnt_ptes_primary[i]); 934 PMAPCOUNT(ptes_spilled); 935 PMAPCOUNT(ptes_evicted); 936 PMAPCOUNT(ptes_removed); 937 938 PMAP_PVO_CHECK(victim_pvo); 939 PMAP_PVO_CHECK(source_pvo); 940 return 1; 941 } 942 943 /* 944 * Restrict given range to physical memory 945 */ 946 void 947 pmap_real_memory(paddr_t *start, psize_t *size) 948 { 949 struct mem_region *mp; 950 951 for (mp = mem; mp->size; mp++) { 952 if (*start + *size > mp->start 953 && *start < mp->start + mp->size) { 954 if (*start < mp->start) { 955 *size -= mp->start - *start; 956 *start = mp->start; 957 } 958 if (*start + *size > mp->start + mp->size) 959 *size = mp->start + mp->size - *start; 960 return; 961 } 962 } 963 *size = 0; 964 } 965 966 /* 967 * Initialize anything else for pmap handling. 968 * Called during vm_init(). 969 */ 970 void 971 pmap_init(void) 972 { 973 int s; 974 #ifdef __HAVE_PMAP_PHYSSEG 975 struct pvo_tqhead *pvoh; 976 int bank; 977 long sz; 978 char *attr; 979 980 s = splvm(); 981 pvoh = pmap_physseg.pvoh; 982 attr = pmap_physseg.attrs; 983 for (bank = 0; bank < vm_nphysseg; bank++) { 984 sz = vm_physmem[bank].end - vm_physmem[bank].start; 985 vm_physmem[bank].pmseg.pvoh = pvoh; 986 vm_physmem[bank].pmseg.attrs = attr; 987 for (; sz > 0; sz--, pvoh++, attr++) { 988 TAILQ_INIT(pvoh); 989 *attr = 0; 990 } 991 } 992 splx(s); 993 #endif 994 995 s = splvm(); 996 pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry), 997 sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl", 998 &pmap_pool_mallocator); 999 1000 pool_setlowat(&pmap_mpvo_pool, 1008); 1001 1002 pmap_initialized = 1; 1003 splx(s); 1004 1005 #ifdef PMAPCOUNTERS 1006 evcnt_attach_static(&pmap_evcnt_mappings); 1007 evcnt_attach_static(&pmap_evcnt_mappings_replaced); 1008 evcnt_attach_static(&pmap_evcnt_unmappings); 1009 1010 evcnt_attach_static(&pmap_evcnt_kernel_mappings); 1011 evcnt_attach_static(&pmap_evcnt_kernel_unmappings); 1012 1013 evcnt_attach_static(&pmap_evcnt_exec_mappings); 1014 evcnt_attach_static(&pmap_evcnt_exec_cached); 1015 evcnt_attach_static(&pmap_evcnt_exec_synced); 1016 evcnt_attach_static(&pmap_evcnt_exec_synced_clear_modify); 1017 1018 evcnt_attach_static(&pmap_evcnt_exec_uncached_page_protect); 1019 evcnt_attach_static(&pmap_evcnt_exec_uncached_clear_modify); 1020 evcnt_attach_static(&pmap_evcnt_exec_uncached_zero_page); 1021 evcnt_attach_static(&pmap_evcnt_exec_uncached_copy_page); 1022 1023 evcnt_attach_static(&pmap_evcnt_zeroed_pages); 1024 evcnt_attach_static(&pmap_evcnt_copied_pages); 1025 evcnt_attach_static(&pmap_evcnt_idlezeroed_pages); 1026 1027 evcnt_attach_static(&pmap_evcnt_updates); 1028 evcnt_attach_static(&pmap_evcnt_collects); 1029 evcnt_attach_static(&pmap_evcnt_copies); 1030 1031 evcnt_attach_static(&pmap_evcnt_ptes_spilled); 1032 evcnt_attach_static(&pmap_evcnt_ptes_unspilled); 1033 evcnt_attach_static(&pmap_evcnt_ptes_evicted); 1034 evcnt_attach_static(&pmap_evcnt_ptes_removed); 1035 evcnt_attach_static(&pmap_evcnt_ptes_changed); 1036 evcnt_attach_static(&pmap_evcnt_ptes_primary[0]); 1037 evcnt_attach_static(&pmap_evcnt_ptes_primary[1]); 1038 evcnt_attach_static(&pmap_evcnt_ptes_primary[2]); 1039 evcnt_attach_static(&pmap_evcnt_ptes_primary[3]); 1040 evcnt_attach_static(&pmap_evcnt_ptes_primary[4]); 1041 evcnt_attach_static(&pmap_evcnt_ptes_primary[5]); 1042 evcnt_attach_static(&pmap_evcnt_ptes_primary[6]); 1043 evcnt_attach_static(&pmap_evcnt_ptes_primary[7]); 1044 evcnt_attach_static(&pmap_evcnt_ptes_secondary[0]); 1045 evcnt_attach_static(&pmap_evcnt_ptes_secondary[1]); 1046 evcnt_attach_static(&pmap_evcnt_ptes_secondary[2]); 1047 evcnt_attach_static(&pmap_evcnt_ptes_secondary[3]); 1048 evcnt_attach_static(&pmap_evcnt_ptes_secondary[4]); 1049 evcnt_attach_static(&pmap_evcnt_ptes_secondary[5]); 1050 evcnt_attach_static(&pmap_evcnt_ptes_secondary[6]); 1051 evcnt_attach_static(&pmap_evcnt_ptes_secondary[7]); 1052 #endif 1053 } 1054 1055 /* 1056 * How much virtual space does the kernel get? 1057 */ 1058 void 1059 pmap_virtual_space(vaddr_t *start, vaddr_t *end) 1060 { 1061 /* 1062 * For now, reserve one segment (minus some overhead) for kernel 1063 * virtual memory 1064 */ 1065 *start = VM_MIN_KERNEL_ADDRESS; 1066 *end = VM_MAX_KERNEL_ADDRESS; 1067 } 1068 1069 /* 1070 * Allocate, initialize, and return a new physical map. 1071 */ 1072 pmap_t 1073 pmap_create(void) 1074 { 1075 pmap_t pm; 1076 1077 pm = pool_get(&pmap_pool, PR_WAITOK); 1078 memset((caddr_t)pm, 0, sizeof *pm); 1079 pmap_pinit(pm); 1080 1081 DPRINTFN(CREATE,("pmap_create: pm %p:\n" 1082 "\t%06lx %06lx %06lx %06lx %06lx %06lx %06lx %06lx\n" 1083 "\t%06lx %06lx %06lx %06lx %06lx %06lx %06lx %06lx\n", pm, 1084 pm->pm_sr[0], pm->pm_sr[1], pm->pm_sr[2], pm->pm_sr[3], 1085 pm->pm_sr[4], pm->pm_sr[5], pm->pm_sr[6], pm->pm_sr[7], 1086 pm->pm_sr[8], pm->pm_sr[9], pm->pm_sr[10], pm->pm_sr[11], 1087 pm->pm_sr[12], pm->pm_sr[13], pm->pm_sr[14], pm->pm_sr[15])); 1088 return pm; 1089 } 1090 1091 /* 1092 * Initialize a preallocated and zeroed pmap structure. 1093 */ 1094 void 1095 pmap_pinit(pmap_t pm) 1096 { 1097 register_t entropy = MFTB(); 1098 register_t mask; 1099 int i; 1100 1101 /* 1102 * Allocate some segment registers for this pmap. 1103 */ 1104 pm->pm_refs = 1; 1105 for (i = 0; i < NPMAPS; i += VSID_NBPW) { 1106 static register_t pmap_vsidcontext; 1107 register_t hash; 1108 unsigned int n; 1109 1110 /* Create a new value by multiplying by a prime adding in 1111 * entropy from the timebase register. This is to make the 1112 * VSID more random so that the PT Hash function collides 1113 * less often. (note that the prime causes gcc to do shifts 1114 * instead of a multiply) 1115 */ 1116 pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy; 1117 hash = pmap_vsidcontext & (NPMAPS - 1); 1118 if (hash == 0) /* 0 is special, avoid it */ 1119 continue; 1120 n = hash >> 5; 1121 mask = 1L << (hash & (VSID_NBPW-1)); 1122 hash = pmap_vsidcontext; 1123 if (pmap_vsid_bitmap[n] & mask) { /* collision? */ 1124 /* anything free in this bucket? */ 1125 if (~pmap_vsid_bitmap[n] == 0) { 1126 entropy = hash >> PTE_VSID_SHFT; 1127 continue; 1128 } 1129 i = ffs(~pmap_vsid_bitmap[n]) - 1; 1130 mask = 1L << i; 1131 hash &= ~(VSID_NBPW-1); 1132 hash |= i; 1133 } 1134 /* 1135 * Make sure clear out SR_KEY_LEN bits because we put our 1136 * our data in those bits (to identify the segment). 1137 */ 1138 hash &= PTE_VSID >> (PTE_VSID_SHFT + SR_KEY_LEN); 1139 pmap_vsid_bitmap[n] |= mask; 1140 for (i = 0; i < 16; i++) 1141 pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY; 1142 return; 1143 } 1144 panic("pmap_pinit: out of segments"); 1145 } 1146 1147 /* 1148 * Add a reference to the given pmap. 1149 */ 1150 void 1151 pmap_reference(pmap_t pm) 1152 { 1153 pm->pm_refs++; 1154 } 1155 1156 /* 1157 * Retire the given pmap from service. 1158 * Should only be called if the map contains no valid mappings. 1159 */ 1160 void 1161 pmap_destroy(pmap_t pm) 1162 { 1163 if (--pm->pm_refs == 0) { 1164 pmap_release(pm); 1165 pool_put(&pmap_pool, pm); 1166 } 1167 } 1168 1169 /* 1170 * Release any resources held by the given physical map. 1171 * Called when a pmap initialized by pmap_pinit is being released. 1172 */ 1173 void 1174 pmap_release(pmap_t pm) 1175 { 1176 int idx, mask; 1177 1178 if (pm->pm_sr[0] == 0) 1179 panic("pmap_release"); 1180 idx = VSID_TO_HASH(pm->pm_sr[0]) & (NPMAPS-1); 1181 mask = 1 << (idx % VSID_NBPW); 1182 idx /= VSID_NBPW; 1183 pmap_vsid_bitmap[idx] &= ~mask; 1184 } 1185 1186 /* 1187 * Copy the range specified by src_addr/len 1188 * from the source map to the range dst_addr/len 1189 * in the destination map. 1190 * 1191 * This routine is only advisory and need not do anything. 1192 */ 1193 void 1194 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr, 1195 vsize_t len, vaddr_t src_addr) 1196 { 1197 PMAPCOUNT(copies); 1198 } 1199 1200 /* 1201 * Require that all active physical maps contain no 1202 * incorrect entries NOW. 1203 */ 1204 void 1205 pmap_update(struct pmap *pmap) 1206 { 1207 PMAPCOUNT(updates); 1208 TLBSYNC(); 1209 } 1210 1211 /* 1212 * Garbage collects the physical map system for 1213 * pages which are no longer used. 1214 * Success need not be guaranteed -- that is, there 1215 * may well be pages which are not referenced, but 1216 * others may be collected. 1217 * Called by the pageout daemon when pages are scarce. 1218 */ 1219 void 1220 pmap_collect(pmap_t pm) 1221 { 1222 PMAPCOUNT(collects); 1223 } 1224 1225 static __inline int 1226 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx) 1227 { 1228 int pteidx; 1229 /* 1230 * We can find the actual pte entry without searching by 1231 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9] 1232 * and by noticing the HID bit. 1233 */ 1234 pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo); 1235 if (pvo->pvo_pte.pte_hi & PTE_HID) 1236 pteidx ^= pmap_pteg_mask * 8; 1237 return pteidx; 1238 } 1239 1240 volatile struct pte * 1241 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx) 1242 { 1243 volatile struct pte *pt; 1244 1245 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1246 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) 1247 return NULL; 1248 #endif 1249 1250 /* 1251 * If we haven't been supplied the ptegidx, calculate it. 1252 */ 1253 if (pteidx == -1) { 1254 int ptegidx; 1255 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1256 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1257 } 1258 1259 pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7]; 1260 1261 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1262 return pt; 1263 #else 1264 if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) { 1265 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1266 "pvo but no valid pte index", pvo); 1267 } 1268 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) { 1269 panic("pmap_pvo_to_pte: pvo %p: has valid pte index in " 1270 "pvo but no valid pte", pvo); 1271 } 1272 1273 if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) { 1274 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) { 1275 #if defined(DEBUG) || defined(PMAPCHECK) 1276 pmap_pte_print(pt); 1277 #endif 1278 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1279 "pmap_pteg_table %p but invalid in pvo", 1280 pvo, pt); 1281 } 1282 if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) { 1283 #if defined(DEBUG) || defined(PMAPCHECK) 1284 pmap_pte_print(pt); 1285 #endif 1286 panic("pmap_pvo_to_pte: pvo %p: pvo pte does " 1287 "not match pte %p in pmap_pteg_table", 1288 pvo, pt); 1289 } 1290 return pt; 1291 } 1292 1293 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1294 #if defined(DEBUG) || defined(PMAPCHECK) 1295 pmap_pte_print(pt); 1296 #endif 1297 panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in " 1298 "pmap_pteg_table but valid in pvo", pvo, pt); 1299 } 1300 return NULL; 1301 #endif /* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */ 1302 } 1303 1304 struct pvo_entry * 1305 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p) 1306 { 1307 struct pvo_entry *pvo; 1308 int ptegidx; 1309 1310 va &= ~ADDR_POFF; 1311 ptegidx = va_to_pteg(pm, va); 1312 1313 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1314 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1315 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 1316 panic("pmap_pvo_find_va: invalid pvo %p on " 1317 "list %#x (%p)", pvo, ptegidx, 1318 &pmap_pvo_table[ptegidx]); 1319 #endif 1320 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1321 if (pteidx_p) 1322 *pteidx_p = pmap_pvo_pte_index(pvo, ptegidx); 1323 return pvo; 1324 } 1325 } 1326 return NULL; 1327 } 1328 1329 #if defined(DEBUG) || defined(PMAPCHECK) 1330 void 1331 pmap_pvo_check(const struct pvo_entry *pvo) 1332 { 1333 struct pvo_head *pvo_head; 1334 struct pvo_entry *pvo0; 1335 volatile struct pte *pt; 1336 int failed = 0; 1337 1338 if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH) 1339 panic("pmap_pvo_check: pvo %p: invalid address", pvo); 1340 1341 if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) { 1342 printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n", 1343 pvo, pvo->pvo_pmap); 1344 failed = 1; 1345 } 1346 1347 if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH || 1348 (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) { 1349 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1350 pvo, TAILQ_NEXT(pvo, pvo_olink)); 1351 failed = 1; 1352 } 1353 1354 if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH || 1355 (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) { 1356 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1357 pvo, LIST_NEXT(pvo, pvo_vlink)); 1358 failed = 1; 1359 } 1360 1361 if (pvo->pvo_vaddr & PVO_MANAGED) { 1362 pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL); 1363 } else { 1364 if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) { 1365 printf("pmap_pvo_check: pvo %p: non kernel address " 1366 "on kernel unmanaged list\n", pvo); 1367 failed = 1; 1368 } 1369 pvo_head = &pmap_pvo_kunmanaged; 1370 } 1371 LIST_FOREACH(pvo0, pvo_head, pvo_vlink) { 1372 if (pvo0 == pvo) 1373 break; 1374 } 1375 if (pvo0 == NULL) { 1376 printf("pmap_pvo_check: pvo %p: not present " 1377 "on its vlist head %p\n", pvo, pvo_head); 1378 failed = 1; 1379 } 1380 if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) { 1381 printf("pmap_pvo_check: pvo %p: not present " 1382 "on its olist head\n", pvo); 1383 failed = 1; 1384 } 1385 pt = pmap_pvo_to_pte(pvo, -1); 1386 if (pt == NULL) { 1387 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1388 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1389 "no PTE\n", pvo); 1390 failed = 1; 1391 } 1392 } else { 1393 if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] || 1394 (uintptr_t) pt >= 1395 (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) { 1396 printf("pmap_pvo_check: pvo %p: pte %p not in " 1397 "pteg table\n", pvo, pt); 1398 failed = 1; 1399 } 1400 if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) { 1401 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1402 "no PTE\n", pvo); 1403 failed = 1; 1404 } 1405 if (pvo->pvo_pte.pte_hi != pt->pte_hi) { 1406 printf("pmap_pvo_check: pvo %p: pte_hi differ: " 1407 "%#lx/%#lx\n", pvo, pvo->pvo_pte.pte_hi, pt->pte_hi); 1408 failed = 1; 1409 } 1410 if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) & 1411 (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) { 1412 printf("pmap_pvo_check: pvo %p: pte_lo differ: " 1413 "%#lx/%#lx\n", pvo, 1414 pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN), 1415 pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)); 1416 failed = 1; 1417 } 1418 if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) { 1419 printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#lx" 1420 " doesn't not match PVO's VA %#lx\n", 1421 pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo)); 1422 failed = 1; 1423 } 1424 if (failed) 1425 pmap_pte_print(pt); 1426 } 1427 if (failed) 1428 panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo, 1429 pvo->pvo_pmap); 1430 } 1431 #endif /* DEBUG || PMAPCHECK */ 1432 1433 /* 1434 * This returns whether this is the first mapping of a page. 1435 */ 1436 int 1437 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head, 1438 vaddr_t va, paddr_t pa, register_t pte_lo, int flags) 1439 { 1440 struct pvo_entry *pvo; 1441 struct pvo_tqhead *pvoh; 1442 register_t msr; 1443 int ptegidx; 1444 int i; 1445 int poolflags = PR_NOWAIT; 1446 1447 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1448 if (pmap_pvo_remove_depth > 0) 1449 panic("pmap_pvo_enter: called while pmap_pvo_remove active!"); 1450 if (++pmap_pvo_enter_depth > 1) 1451 panic("pmap_pvo_enter: called recursively!"); 1452 #endif 1453 1454 /* 1455 * Compute the PTE Group index. 1456 */ 1457 va &= ~ADDR_POFF; 1458 ptegidx = va_to_pteg(pm, va); 1459 1460 msr = pmap_interrupts_off(); 1461 /* 1462 * Remove any existing mapping for this page. Reuse the 1463 * pvo entry if there a mapping. 1464 */ 1465 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1466 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1467 #ifdef DEBUG 1468 if ((pmapdebug & PMAPDEBUG_PVOENTER) && 1469 ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) & 1470 ~(PTE_REF|PTE_CHG)) == 0 && 1471 va < VM_MIN_KERNEL_ADDRESS) { 1472 printf("pmap_pvo_enter: pvo %p: dup %#lx/%#lx\n", 1473 pvo, pvo->pvo_pte.pte_lo, pte_lo|pa); 1474 printf("pmap_pvo_enter: pte_hi=%#lx sr=%#lx\n", 1475 pvo->pvo_pte.pte_hi, 1476 pm->pm_sr[va >> ADDR_SR_SHFT]); 1477 pmap_pte_print(pmap_pvo_to_pte(pvo, -1)); 1478 #ifdef DDBX 1479 Debugger(); 1480 #endif 1481 } 1482 #endif 1483 PMAPCOUNT(mappings_replaced); 1484 pmap_pvo_remove(pvo, -1); 1485 break; 1486 } 1487 } 1488 1489 /* 1490 * If we aren't overwriting an mapping, try to allocate 1491 */ 1492 pmap_interrupts_restore(msr); 1493 pvo = pool_get(pl, poolflags); 1494 msr = pmap_interrupts_off(); 1495 if (pvo == NULL) { 1496 #if 0 1497 pvo = pmap_pvo_reclaim(pm); 1498 if (pvo == NULL) { 1499 #endif 1500 if ((flags & PMAP_CANFAIL) == 0) 1501 panic("pmap_pvo_enter: failed"); 1502 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1503 pmap_pvo_enter_depth--; 1504 #endif 1505 pmap_interrupts_restore(msr); 1506 return ENOMEM; 1507 #if 0 1508 } 1509 #endif 1510 } 1511 pvo->pvo_vaddr = va; 1512 pvo->pvo_pmap = pm; 1513 pvo->pvo_vaddr &= ~ADDR_POFF; 1514 if (flags & VM_PROT_EXECUTE) { 1515 PMAPCOUNT(exec_mappings); 1516 pvo->pvo_vaddr |= PVO_EXECUTABLE; 1517 } 1518 if (flags & PMAP_WIRED) 1519 pvo->pvo_vaddr |= PVO_WIRED; 1520 if (pvo_head != &pmap_pvo_kunmanaged) { 1521 pvo->pvo_vaddr |= PVO_MANAGED; 1522 PMAPCOUNT(mappings); 1523 } else { 1524 PMAPCOUNT(kernel_mappings); 1525 } 1526 pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo); 1527 1528 LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink); 1529 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1530 pvo->pvo_pmap->pm_stats.wired_count++; 1531 pvo->pvo_pmap->pm_stats.resident_count++; 1532 #if defined(DEBUG) 1533 if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) 1534 DPRINTFN(PVOENTER, 1535 ("pmap_pvo_enter: pvo %p: pm %p va %#lx pa %#lx\n", 1536 pvo, pm, va, pa)); 1537 #endif 1538 1539 /* 1540 * We hope this succeeds but it isn't required. 1541 */ 1542 pvoh = &pmap_pvo_table[ptegidx]; 1543 i = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 1544 if (i >= 0) { 1545 PVO_PTEGIDX_SET(pvo, i); 1546 PVO_WHERE(pvo, ENTER_INSERT); 1547 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 1548 ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]); 1549 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 1550 } else { 1551 /* 1552 * Since we didn't have room for this entry (which makes it 1553 * and evicted entry), place it at the head of the list. 1554 */ 1555 TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink); 1556 PMAPCOUNT(ptes_evicted); 1557 pm->pm_evictions++; 1558 /* 1559 * If this is a kernel page, make sure it's active. 1560 */ 1561 if (pm == pmap_kernel()) { 1562 i = pmap_pte_spill(pm, va); 1563 KASSERT(i); 1564 } 1565 } 1566 PMAP_PVO_CHECK(pvo); /* sanity check */ 1567 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1568 pmap_pvo_enter_depth--; 1569 #endif 1570 pmap_interrupts_restore(msr); 1571 return 0; 1572 } 1573 1574 void 1575 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx) 1576 { 1577 volatile struct pte *pt; 1578 int ptegidx; 1579 1580 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1581 if (++pmap_pvo_remove_depth > 1) 1582 panic("pmap_pvo_remove: called recursively!"); 1583 #endif 1584 1585 /* 1586 * If we haven't been supplied the ptegidx, calculate it. 1587 */ 1588 if (pteidx == -1) { 1589 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1590 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1591 } else { 1592 ptegidx = pteidx >> 3; 1593 if (pvo->pvo_pte.pte_hi & PTE_HID) 1594 ptegidx ^= pmap_pteg_mask; 1595 } 1596 PMAP_PVO_CHECK(pvo); /* sanity check */ 1597 1598 /* 1599 * If there is an active pte entry, we need to deactivate it 1600 * (and save the ref & chg bits). 1601 */ 1602 pt = pmap_pvo_to_pte(pvo, pteidx); 1603 if (pt != NULL) { 1604 pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 1605 PVO_WHERE(pvo, REMOVE); 1606 PVO_PTEGIDX_CLR(pvo); 1607 PMAPCOUNT(ptes_removed); 1608 } else { 1609 KASSERT(pvo->pvo_pmap->pm_evictions > 0); 1610 pvo->pvo_pmap->pm_evictions--; 1611 } 1612 1613 /* 1614 * Update our statistics 1615 */ 1616 pvo->pvo_pmap->pm_stats.resident_count--; 1617 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1618 pvo->pvo_pmap->pm_stats.wired_count--; 1619 1620 /* 1621 * Save the REF/CHG bits into their cache if the page is managed. 1622 */ 1623 if (pvo->pvo_vaddr & PVO_MANAGED) { 1624 register_t ptelo = pvo->pvo_pte.pte_lo; 1625 struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN); 1626 1627 if (pg != NULL) { 1628 pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG)); 1629 } 1630 PMAPCOUNT(unmappings); 1631 } else { 1632 PMAPCOUNT(kernel_unmappings); 1633 } 1634 1635 /* 1636 * Remove the PVO from its lists and return it to the pool. 1637 */ 1638 LIST_REMOVE(pvo, pvo_vlink); 1639 TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink); 1640 pool_put(pvo->pvo_vaddr & PVO_MANAGED 1641 ? &pmap_mpvo_pool : &pmap_upvo_pool, pvo); 1642 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1643 pmap_pvo_remove_depth--; 1644 #endif 1645 } 1646 1647 /* 1648 * Insert physical page at pa into the given pmap at virtual address va. 1649 */ 1650 int 1651 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags) 1652 { 1653 struct mem_region *mp; 1654 struct pvo_head *pvo_head; 1655 struct vm_page *pg; 1656 struct pool *pl; 1657 register_t pte_lo; 1658 int s; 1659 int error; 1660 u_int pvo_flags; 1661 u_int was_exec = 0; 1662 1663 if (__predict_false(!pmap_initialized)) { 1664 pvo_head = &pmap_pvo_kunmanaged; 1665 pl = &pmap_upvo_pool; 1666 pvo_flags = 0; 1667 pg = NULL; 1668 was_exec = PTE_EXEC; 1669 } else { 1670 pvo_head = pa_to_pvoh(pa, &pg); 1671 pl = &pmap_mpvo_pool; 1672 pvo_flags = PVO_MANAGED; 1673 } 1674 1675 DPRINTFN(ENTER, 1676 ("pmap_enter(%p, 0x%lx, 0x%lx, 0x%x, 0x%x):", 1677 pm, va, pa, prot, flags)); 1678 1679 /* 1680 * If this is a managed page, and it's the first reference to the 1681 * page clear the execness of the page. Otherwise fetch the execness. 1682 */ 1683 if (pg != NULL) 1684 was_exec = pmap_attr_fetch(pg) & PTE_EXEC; 1685 1686 DPRINTFN(ENTER, (" was_exec=%d", was_exec)); 1687 1688 /* 1689 * Assume the page is cache inhibited and access is guarded unless 1690 * it's in our available memory array. If it is in the memory array, 1691 * asssume it's in memory coherent memory. 1692 */ 1693 pte_lo = PTE_IG; 1694 if ((flags & PMAP_NC) == 0) { 1695 for (mp = mem; mp->size; mp++) { 1696 if (pa >= mp->start && pa < mp->start + mp->size) { 1697 pte_lo = PTE_M; 1698 break; 1699 } 1700 } 1701 } 1702 1703 if (prot & VM_PROT_WRITE) 1704 pte_lo |= PTE_BW; 1705 else 1706 pte_lo |= PTE_BR; 1707 1708 /* 1709 * If this was in response to a fault, "pre-fault" the PTE's 1710 * changed/referenced bit appropriately. 1711 */ 1712 if (flags & VM_PROT_WRITE) 1713 pte_lo |= PTE_CHG; 1714 if (flags & (VM_PROT_READ|VM_PROT_WRITE)) 1715 pte_lo |= PTE_REF; 1716 1717 #if 0 1718 if (pm == pmap_kernel()) { 1719 if ((prot & (VM_PROT_READ|VM_PROT_WRITE)) == VM_PROT_READ) 1720 printf("pmap_pvo_enter: Kernel RO va %#lx pa %#lx\n", 1721 va, pa); 1722 if ((prot & (VM_PROT_READ|VM_PROT_WRITE)) == VM_PROT_NONE) 1723 printf("pmap_pvo_enter: Kernel N/A va %#lx pa %#lx\n", 1724 va, pa); 1725 } 1726 #endif 1727 1728 /* 1729 * We need to know if this page can be executable 1730 */ 1731 flags |= (prot & VM_PROT_EXECUTE); 1732 1733 /* 1734 * Record mapping for later back-translation and pte spilling. 1735 * This will overwrite any existing mapping. 1736 */ 1737 s = splvm(); 1738 error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags); 1739 splx(s); 1740 1741 /* 1742 * Flush the real page from the instruction cache if this page is 1743 * mapped executable and cacheable and has not been flushed since 1744 * the last time it was modified. 1745 */ 1746 if (error == 0 && 1747 (flags & VM_PROT_EXECUTE) && 1748 (pte_lo & PTE_I) == 0 && 1749 was_exec == 0) { 1750 DPRINTFN(ENTER, (" syncicache")); 1751 PMAPCOUNT(exec_synced); 1752 pmap_syncicache(pa, PAGE_SIZE); 1753 if (pg != NULL) { 1754 pmap_attr_save(pg, PTE_EXEC); 1755 PMAPCOUNT(exec_cached); 1756 #if defined(DEBUG) || defined(PMAPDEBUG) 1757 if (pmapdebug & PMAPDEBUG_ENTER) 1758 printf(" marked-as-exec"); 1759 else if (pmapdebug & PMAPDEBUG_EXEC) 1760 printf("[pmap_enter: %#lx: marked-as-exec]\n", 1761 pg->phys_addr); 1762 1763 #endif 1764 } 1765 } 1766 1767 DPRINTFN(ENTER, (": error=%d\n", error)); 1768 1769 return error; 1770 } 1771 1772 void 1773 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot) 1774 { 1775 struct mem_region *mp; 1776 register_t pte_lo; 1777 register_t msr; 1778 int error; 1779 int s; 1780 1781 if (va < VM_MIN_KERNEL_ADDRESS) 1782 panic("pmap_kenter_pa: attempt to enter " 1783 "non-kernel address %#lx!", va); 1784 1785 DPRINTFN(KENTER, 1786 ("pmap_kenter_pa(%#lx,%#lx,%#x)\n", va, pa, prot)); 1787 1788 /* 1789 * Assume the page is cache inhibited and access is guarded unless 1790 * it's in our available memory array. If it is in the memory array, 1791 * asssume it's in memory coherent memory. 1792 */ 1793 pte_lo = PTE_IG; 1794 if ((prot & PMAP_NC) == 0) { 1795 for (mp = mem; mp->size; mp++) { 1796 if (pa >= mp->start && pa < mp->start + mp->size) { 1797 pte_lo = PTE_M; 1798 break; 1799 } 1800 } 1801 } 1802 1803 if (prot & VM_PROT_WRITE) 1804 pte_lo |= PTE_BW; 1805 else 1806 pte_lo |= PTE_BR; 1807 1808 /* 1809 * We don't care about REF/CHG on PVOs on the unmanaged list. 1810 */ 1811 s = splvm(); 1812 msr = pmap_interrupts_off(); 1813 error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool, 1814 &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED); 1815 pmap_interrupts_restore(msr); 1816 splx(s); 1817 1818 if (error != 0) 1819 panic("pmap_kenter_pa: failed to enter va %#lx pa %#lx: %d", 1820 va, pa, error); 1821 } 1822 1823 void 1824 pmap_kremove(vaddr_t va, vsize_t len) 1825 { 1826 if (va < VM_MIN_KERNEL_ADDRESS) 1827 panic("pmap_kremove: attempt to remove " 1828 "non-kernel address %#lx!", va); 1829 1830 DPRINTFN(KREMOVE,("pmap_kremove(%#lx,%#lx)\n", va, len)); 1831 pmap_remove(pmap_kernel(), va, va + len); 1832 } 1833 1834 /* 1835 * Remove the given range of mapping entries. 1836 */ 1837 void 1838 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva) 1839 { 1840 struct pvo_entry *pvo; 1841 register_t msr; 1842 int pteidx; 1843 int s; 1844 1845 for (; va < endva; va += PAGE_SIZE) { 1846 s = splvm(); 1847 msr = pmap_interrupts_off(); 1848 pvo = pmap_pvo_find_va(pm, va, &pteidx); 1849 if (pvo != NULL) { 1850 pmap_pvo_remove(pvo, pteidx); 1851 } 1852 pmap_interrupts_restore(msr); 1853 splx(s); 1854 } 1855 } 1856 1857 /* 1858 * Get the physical page address for the given pmap/virtual address. 1859 */ 1860 boolean_t 1861 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap) 1862 { 1863 struct pvo_entry *pvo; 1864 register_t msr; 1865 int s; 1866 1867 /* 1868 * If this is a kernel pmap lookup, also check the battable 1869 * and if we get a hit, translate the VA to a PA using the 1870 * BAT entries. Don't check for VM_MAX_KENREL_ADDRESS is 1871 * that will wrap back to 0. 1872 */ 1873 if (pm == pmap_kernel() && 1874 (va < VM_MIN_KERNEL_ADDRESS || 1875 (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) { 1876 register_t batu = battable[va >> ADDR_SR_SHFT].batu; 1877 KASSERT((va >> ADDR_SR_SHFT) != USER_SR); 1878 if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) { 1879 register_t batl = battable[va >> ADDR_SR_SHFT].batl; 1880 register_t mask = (~(batu & BAT_BL) << 15) & ~0x1ffffL; 1881 *pap = (batl & mask) | (va & ~mask); 1882 return TRUE; 1883 } 1884 return FALSE; 1885 } 1886 1887 s = splvm(); 1888 msr = pmap_interrupts_off(); 1889 pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL); 1890 if (pvo != NULL) { 1891 PMAP_PVO_CHECK(pvo); /* sanity check */ 1892 *pap = (pvo->pvo_pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF); 1893 } 1894 pmap_interrupts_restore(msr); 1895 splx(s); 1896 return pvo != NULL; 1897 } 1898 1899 /* 1900 * Lower the protection on the specified range of this pmap. 1901 * 1902 * There are only two cases: either the protection is going to 0, 1903 * or it is going to read-only. 1904 */ 1905 void 1906 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot) 1907 { 1908 struct pvo_entry *pvo; 1909 volatile struct pte *pt; 1910 register_t msr; 1911 int s; 1912 int pteidx; 1913 1914 /* 1915 * Since this routine only downgrades protection, we should 1916 * always be called without WRITE permisison. 1917 */ 1918 KASSERT((prot & VM_PROT_WRITE) == 0); 1919 1920 /* 1921 * If there is no protection, this is equivalent to 1922 * remove the pmap from the pmap. 1923 */ 1924 if ((prot & VM_PROT_READ) == 0) { 1925 pmap_remove(pm, va, endva); 1926 return; 1927 } 1928 1929 s = splvm(); 1930 msr = pmap_interrupts_off(); 1931 1932 for (; va < endva; va += PAGE_SIZE) { 1933 pvo = pmap_pvo_find_va(pm, va, &pteidx); 1934 if (pvo == NULL) 1935 continue; 1936 PMAP_PVO_CHECK(pvo); /* sanity check */ 1937 1938 /* 1939 * Revoke executable if asked to do so. 1940 */ 1941 if ((prot & VM_PROT_EXECUTE) == 0) 1942 pvo->pvo_vaddr &= ~PVO_EXECUTABLE; 1943 1944 #if 0 1945 /* 1946 * If the page is already read-only, no change 1947 * needs to be made. 1948 */ 1949 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) 1950 continue; 1951 #endif 1952 /* 1953 * Grab the PTE pointer before we diddle with 1954 * the cached PTE copy. 1955 */ 1956 pt = pmap_pvo_to_pte(pvo, pteidx); 1957 /* 1958 * Change the protection of the page. 1959 */ 1960 pvo->pvo_pte.pte_lo &= ~PTE_PP; 1961 pvo->pvo_pte.pte_lo |= PTE_BR; 1962 1963 /* 1964 * If the PVO is in the page table, update 1965 * that pte at well. 1966 */ 1967 if (pt != NULL) { 1968 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 1969 PVO_WHERE(pvo, PMAP_PROTECT); 1970 PMAPCOUNT(ptes_changed); 1971 } 1972 1973 PMAP_PVO_CHECK(pvo); /* sanity check */ 1974 } 1975 1976 pmap_interrupts_restore(msr); 1977 splx(s); 1978 } 1979 1980 void 1981 pmap_unwire(pmap_t pm, vaddr_t va) 1982 { 1983 struct pvo_entry *pvo; 1984 register_t msr; 1985 int s; 1986 1987 s = splvm(); 1988 msr = pmap_interrupts_off(); 1989 1990 pvo = pmap_pvo_find_va(pm, va, NULL); 1991 if (pvo != NULL) { 1992 if (pvo->pvo_vaddr & PVO_WIRED) { 1993 pvo->pvo_vaddr &= ~PVO_WIRED; 1994 pm->pm_stats.wired_count--; 1995 } 1996 PMAP_PVO_CHECK(pvo); /* sanity check */ 1997 } 1998 1999 pmap_interrupts_restore(msr); 2000 splx(s); 2001 } 2002 2003 /* 2004 * Lower the protection on the specified physical page. 2005 * 2006 * There are only two cases: either the protection is going to 0, 2007 * or it is going to read-only. 2008 */ 2009 void 2010 pmap_page_protect(struct vm_page *pg, vm_prot_t prot) 2011 { 2012 struct pvo_head *pvo_head; 2013 struct pvo_entry *pvo, *next_pvo; 2014 volatile struct pte *pt; 2015 register_t msr; 2016 int s; 2017 2018 /* 2019 * Since this routine only downgrades protection, if the 2020 * maximal protection is desired, there isn't any change 2021 * to be made. 2022 */ 2023 KASSERT((prot & VM_PROT_WRITE) == 0); 2024 if ((prot & (VM_PROT_READ|VM_PROT_WRITE)) == (VM_PROT_READ|VM_PROT_WRITE)) 2025 return; 2026 2027 s = splvm(); 2028 msr = pmap_interrupts_off(); 2029 2030 /* 2031 * When UVM reuses a page, it does a pmap_page_protect with 2032 * VM_PROT_NONE. At that point, we can clear the exec flag 2033 * since we know the page will have different contents. 2034 */ 2035 if ((prot & VM_PROT_READ) == 0) { 2036 DPRINTFN(EXEC, ("[pmap_page_protect: %#lx: clear-exec]\n", 2037 pg->phys_addr)); 2038 if (pmap_attr_fetch(pg) & PTE_EXEC) { 2039 PMAPCOUNT(exec_uncached_page_protect); 2040 pmap_attr_clear(pg, PTE_EXEC); 2041 } 2042 } 2043 2044 pvo_head = vm_page_to_pvoh(pg); 2045 for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) { 2046 next_pvo = LIST_NEXT(pvo, pvo_vlink); 2047 PMAP_PVO_CHECK(pvo); /* sanity check */ 2048 2049 /* 2050 * Downgrading to no mapping at all, we just remove the entry. 2051 */ 2052 if ((prot & VM_PROT_READ) == 0) { 2053 pmap_pvo_remove(pvo, -1); 2054 continue; 2055 } 2056 2057 /* 2058 * If EXEC permission is being revoked, just clear the 2059 * flag in the PVO. 2060 */ 2061 if ((prot & VM_PROT_EXECUTE) == 0) 2062 pvo->pvo_vaddr &= ~PVO_EXECUTABLE; 2063 2064 /* 2065 * If this entry is already RO, don't diddle with the 2066 * page table. 2067 */ 2068 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) { 2069 PMAP_PVO_CHECK(pvo); 2070 continue; 2071 } 2072 2073 /* 2074 * Grab the PTE before the we diddle the bits so 2075 * pvo_to_pte can verify the pte contents are as 2076 * expected. 2077 */ 2078 pt = pmap_pvo_to_pte(pvo, -1); 2079 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2080 pvo->pvo_pte.pte_lo |= PTE_BR; 2081 if (pt != NULL) { 2082 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2083 PVO_WHERE(pvo, PMAP_PAGE_PROTECT); 2084 PMAPCOUNT(ptes_changed); 2085 } 2086 PMAP_PVO_CHECK(pvo); /* sanity check */ 2087 } 2088 2089 pmap_interrupts_restore(msr); 2090 splx(s); 2091 } 2092 2093 /* 2094 * Activate the address space for the specified process. If the process 2095 * is the current process, load the new MMU context. 2096 */ 2097 void 2098 pmap_activate(struct lwp *l) 2099 { 2100 struct pcb *pcb = &l->l_addr->u_pcb; 2101 pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap; 2102 2103 DPRINTFN(ACTIVATE, 2104 ("pmap_activate: lwp %p (curlwp %p)\n", l, curlwp)); 2105 2106 /* 2107 * XXX Normally performed in cpu_fork(). 2108 */ 2109 pcb->pcb_pm = pmap; 2110 } 2111 2112 /* 2113 * Deactivate the specified process's address space. 2114 */ 2115 void 2116 pmap_deactivate(struct lwp *l) 2117 { 2118 } 2119 2120 boolean_t 2121 pmap_query_bit(struct vm_page *pg, int ptebit) 2122 { 2123 struct pvo_entry *pvo; 2124 volatile struct pte *pt; 2125 register_t msr; 2126 int s; 2127 2128 if (pmap_attr_fetch(pg) & ptebit) 2129 return TRUE; 2130 s = splvm(); 2131 msr = pmap_interrupts_off(); 2132 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2133 PMAP_PVO_CHECK(pvo); /* sanity check */ 2134 /* 2135 * See if we saved the bit off. If so cache, it and return 2136 * success. 2137 */ 2138 if (pvo->pvo_pte.pte_lo & ptebit) { 2139 pmap_attr_save(pg, ptebit); 2140 PMAP_PVO_CHECK(pvo); /* sanity check */ 2141 pmap_interrupts_restore(msr); 2142 splx(s); 2143 return TRUE; 2144 } 2145 } 2146 /* 2147 * No luck, now go thru the hard part of looking at the ptes 2148 * themselves. Sync so any pending REF/CHG bits are flushed 2149 * to the PTEs. 2150 */ 2151 SYNC(); 2152 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2153 PMAP_PVO_CHECK(pvo); /* sanity check */ 2154 /* 2155 * See if this pvo have a valid PTE. If so, fetch the 2156 * REF/CHG bits from the valid PTE. If the appropriate 2157 * ptebit is set, cache, it and return success. 2158 */ 2159 pt = pmap_pvo_to_pte(pvo, -1); 2160 if (pt != NULL) { 2161 pmap_pte_synch(pt, &pvo->pvo_pte); 2162 if (pvo->pvo_pte.pte_lo & ptebit) { 2163 pmap_attr_save(pg, ptebit); 2164 PMAP_PVO_CHECK(pvo); /* sanity check */ 2165 pmap_interrupts_restore(msr); 2166 splx(s); 2167 return TRUE; 2168 } 2169 } 2170 } 2171 pmap_interrupts_restore(msr); 2172 splx(s); 2173 return FALSE; 2174 } 2175 2176 boolean_t 2177 pmap_clear_bit(struct vm_page *pg, int ptebit) 2178 { 2179 struct pvo_head *pvoh = vm_page_to_pvoh(pg); 2180 struct pvo_entry *pvo; 2181 volatile struct pte *pt; 2182 register_t msr; 2183 int rv = 0; 2184 int s; 2185 2186 s = splvm(); 2187 msr = pmap_interrupts_off(); 2188 2189 /* 2190 * Fetch the cache value 2191 */ 2192 rv |= pmap_attr_fetch(pg); 2193 2194 /* 2195 * Clear the cached value. 2196 */ 2197 pmap_attr_clear(pg, ptebit); 2198 2199 /* 2200 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we 2201 * can reset the right ones). Note that since the pvo entries and 2202 * list heads are accessed via BAT0 and are never placed in the 2203 * page table, we don't have to worry about further accesses setting 2204 * the REF/CHG bits. 2205 */ 2206 SYNC(); 2207 2208 /* 2209 * For each pvo entry, clear pvo's ptebit. If this pvo have a 2210 * valid PTE. If so, clear the ptebit from the valid PTE. 2211 */ 2212 LIST_FOREACH(pvo, pvoh, pvo_vlink) { 2213 PMAP_PVO_CHECK(pvo); /* sanity check */ 2214 pt = pmap_pvo_to_pte(pvo, -1); 2215 if (pt != NULL) { 2216 /* 2217 * Only sync the PTE if the bit we are looking 2218 * for is not already set. 2219 */ 2220 if ((pvo->pvo_pte.pte_lo & ptebit) == 0) 2221 pmap_pte_synch(pt, &pvo->pvo_pte); 2222 /* 2223 * If the bit we are looking for was already set, 2224 * clear that bit in the pte. 2225 */ 2226 if (pvo->pvo_pte.pte_lo & ptebit) 2227 pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit); 2228 } 2229 rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF); 2230 pvo->pvo_pte.pte_lo &= ~ptebit; 2231 PMAP_PVO_CHECK(pvo); /* sanity check */ 2232 } 2233 pmap_interrupts_restore(msr); 2234 splx(s); 2235 /* 2236 * If we are clearing the modify bit and this page was marked EXEC 2237 * and the user of the page thinks the page was modified, then we 2238 * need to clean it from the icache if it's mapped or clear the EXEC 2239 * bit if it's not mapped. The page itself might not have the CHG 2240 * bit set if the modification was done via DMA to the page. 2241 */ 2242 if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) { 2243 if (LIST_EMPTY(pvoh)) { 2244 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: clear-exec]\n", 2245 pg->phys_addr)); 2246 pmap_attr_clear(pg, PTE_EXEC); 2247 PMAPCOUNT(exec_uncached_clear_modify); 2248 } else { 2249 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: syncicache]\n", 2250 pg->phys_addr)); 2251 pmap_syncicache(pg->phys_addr, PAGE_SIZE); 2252 PMAPCOUNT(exec_synced_clear_modify); 2253 } 2254 } 2255 return (rv & ptebit) != 0; 2256 } 2257 2258 void 2259 pmap_procwr(struct proc *p, vaddr_t va, size_t len) 2260 { 2261 struct pvo_entry *pvo; 2262 size_t offset = va & ADDR_POFF; 2263 int s; 2264 2265 s = splvm(); 2266 while (len > 0) { 2267 size_t seglen = PAGE_SIZE - offset; 2268 if (seglen > len) 2269 seglen = len; 2270 pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL); 2271 if (pvo != NULL && PVO_ISEXECUTABLE(pvo)) { 2272 pmap_syncicache( 2273 (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen); 2274 PMAP_PVO_CHECK(pvo); 2275 } 2276 va += seglen; 2277 len -= seglen; 2278 offset = 0; 2279 } 2280 splx(s); 2281 } 2282 2283 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 2284 void 2285 pmap_pte_print(volatile struct pte *pt) 2286 { 2287 printf("PTE %p: ", pt); 2288 /* High word: */ 2289 printf("0x%08lx: [", pt->pte_hi); 2290 printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i'); 2291 printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-'); 2292 printf("0x%06lx 0x%02lx", 2293 (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT, 2294 pt->pte_hi & PTE_API); 2295 printf(" (va 0x%08lx)] ", pmap_pte_to_va(pt)); 2296 /* Low word: */ 2297 printf(" 0x%08lx: [", pt->pte_lo); 2298 printf("0x%05lx... ", pt->pte_lo >> 12); 2299 printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u'); 2300 printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n'); 2301 printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.'); 2302 printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.'); 2303 printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.'); 2304 printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.'); 2305 switch (pt->pte_lo & PTE_PP) { 2306 case PTE_BR: printf("br]\n"); break; 2307 case PTE_BW: printf("bw]\n"); break; 2308 case PTE_SO: printf("so]\n"); break; 2309 case PTE_SW: printf("sw]\n"); break; 2310 } 2311 } 2312 #endif 2313 2314 #if defined(DDB) 2315 void 2316 pmap_pteg_check(void) 2317 { 2318 volatile struct pte *pt; 2319 int i; 2320 int ptegidx; 2321 u_int p_valid = 0; 2322 u_int s_valid = 0; 2323 u_int invalid = 0; 2324 2325 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2326 for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) { 2327 if (pt->pte_hi & PTE_VALID) { 2328 if (pt->pte_hi & PTE_HID) 2329 s_valid++; 2330 else 2331 p_valid++; 2332 } else 2333 invalid++; 2334 } 2335 } 2336 printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n", 2337 p_valid, p_valid, s_valid, s_valid, 2338 invalid, invalid); 2339 } 2340 2341 void 2342 pmap_print_mmuregs(void) 2343 { 2344 int i; 2345 u_int cpuvers; 2346 vaddr_t addr; 2347 register_t soft_sr[16]; 2348 struct bat soft_ibat[4]; 2349 struct bat soft_dbat[4]; 2350 register_t sdr1; 2351 2352 cpuvers = MFPVR() >> 16; 2353 2354 __asm __volatile ("mfsdr1 %0" : "=r"(sdr1)); 2355 for (i=0; i<16; i++) { 2356 soft_sr[i] = MFSRIN(addr); 2357 addr += (1 << ADDR_SR_SHFT); 2358 } 2359 2360 /* read iBAT (601: uBAT) registers */ 2361 __asm __volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu)); 2362 __asm __volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl)); 2363 __asm __volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu)); 2364 __asm __volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl)); 2365 __asm __volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu)); 2366 __asm __volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl)); 2367 __asm __volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu)); 2368 __asm __volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl)); 2369 2370 2371 if (cpuvers != MPC601) { 2372 /* read dBAT registers */ 2373 __asm __volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu)); 2374 __asm __volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl)); 2375 __asm __volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu)); 2376 __asm __volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl)); 2377 __asm __volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu)); 2378 __asm __volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl)); 2379 __asm __volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu)); 2380 __asm __volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl)); 2381 } 2382 2383 printf("SDR1:\t%#lx\n", sdr1); 2384 printf("SR[]:\t"); 2385 addr = 0; 2386 for (i=0; i<4; i++) 2387 printf("0x%08lx, ", soft_sr[i]); 2388 printf("\n\t"); 2389 for ( ; i<8; i++) 2390 printf("0x%08lx, ", soft_sr[i]); 2391 printf("\n\t"); 2392 for ( ; i<12; i++) 2393 printf("0x%08lx, ", soft_sr[i]); 2394 printf("\n\t"); 2395 for ( ; i<16; i++) 2396 printf("0x%08lx, ", soft_sr[i]); 2397 printf("\n"); 2398 2399 printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i'); 2400 for (i=0; i<4; i++) { 2401 printf("0x%08lx 0x%08lx, ", 2402 soft_ibat[i].batu, soft_ibat[i].batl); 2403 if (i == 1) 2404 printf("\n\t"); 2405 } 2406 if (cpuvers != MPC601) { 2407 printf("\ndBAT[]:\t"); 2408 for (i=0; i<4; i++) { 2409 printf("0x%08lx 0x%08lx, ", 2410 soft_dbat[i].batu, soft_dbat[i].batl); 2411 if (i == 1) 2412 printf("\n\t"); 2413 } 2414 } 2415 printf("\n"); 2416 } 2417 2418 void 2419 pmap_print_pte(pmap_t pm, vaddr_t va) 2420 { 2421 struct pvo_entry *pvo; 2422 volatile struct pte *pt; 2423 int pteidx; 2424 2425 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2426 if (pvo != NULL) { 2427 pt = pmap_pvo_to_pte(pvo, pteidx); 2428 if (pt != NULL) { 2429 printf("VA %#lx -> %p -> %s %#lx, %#lx\n", 2430 va, pt, 2431 pt->pte_hi & PTE_HID ? "(sec)" : "(pri)", 2432 pt->pte_hi, pt->pte_lo); 2433 } else { 2434 printf("No valid PTE found\n"); 2435 } 2436 } else { 2437 printf("Address not in pmap\n"); 2438 } 2439 } 2440 2441 void 2442 pmap_pteg_dist(void) 2443 { 2444 struct pvo_entry *pvo; 2445 int ptegidx; 2446 int depth; 2447 int max_depth = 0; 2448 unsigned int depths[64]; 2449 2450 memset(depths, 0, sizeof(depths)); 2451 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2452 depth = 0; 2453 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2454 depth++; 2455 } 2456 if (depth > max_depth) 2457 max_depth = depth; 2458 if (depth > 63) 2459 depth = 63; 2460 depths[depth]++; 2461 } 2462 2463 for (depth = 0; depth < 64; depth++) { 2464 printf(" [%2d]: %8u", depth, depths[depth]); 2465 if ((depth & 3) == 3) 2466 printf("\n"); 2467 if (depth == max_depth) 2468 break; 2469 } 2470 if ((depth & 3) != 3) 2471 printf("\n"); 2472 printf("Max depth found was %d\n", max_depth); 2473 } 2474 #endif /* DEBUG */ 2475 2476 #if defined(PMAPCHECK) || defined(DEBUG) 2477 void 2478 pmap_pvo_verify(void) 2479 { 2480 int ptegidx; 2481 int s; 2482 2483 s = splvm(); 2484 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2485 struct pvo_entry *pvo; 2486 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2487 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 2488 panic("pmap_pvo_verify: invalid pvo %p " 2489 "on list %#x", pvo, ptegidx); 2490 pmap_pvo_check(pvo); 2491 } 2492 } 2493 splx(s); 2494 } 2495 #endif /* PMAPCHECK */ 2496 2497 2498 void * 2499 pmap_pool_ualloc(struct pool *pp, int flags) 2500 { 2501 struct pvo_page *pvop; 2502 2503 pvop = SIMPLEQ_FIRST(&pmap_upvop_head); 2504 if (pvop != NULL) { 2505 pmap_upvop_free--; 2506 SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link); 2507 return pvop; 2508 } 2509 if (uvm.page_init_done != TRUE) { 2510 return (void *) uvm_pageboot_alloc(PAGE_SIZE); 2511 } 2512 return pmap_pool_malloc(pp, flags); 2513 } 2514 2515 void * 2516 pmap_pool_malloc(struct pool *pp, int flags) 2517 { 2518 struct pvo_page *pvop; 2519 struct vm_page *pg; 2520 2521 pvop = SIMPLEQ_FIRST(&pmap_mpvop_head); 2522 if (pvop != NULL) { 2523 pmap_mpvop_free--; 2524 SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link); 2525 return pvop; 2526 } 2527 again: 2528 pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE, 2529 UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256); 2530 if (__predict_false(pg == NULL)) { 2531 if (flags & PR_WAITOK) { 2532 uvm_wait("plpg"); 2533 goto again; 2534 } else { 2535 return (0); 2536 } 2537 } 2538 return (void *) VM_PAGE_TO_PHYS(pg); 2539 } 2540 2541 void 2542 pmap_pool_ufree(struct pool *pp, void *va) 2543 { 2544 struct pvo_page *pvop; 2545 #if 0 2546 if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) { 2547 pmap_pool_mfree(va, size, tag); 2548 return; 2549 } 2550 #endif 2551 pvop = va; 2552 SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link); 2553 pmap_upvop_free++; 2554 if (pmap_upvop_free > pmap_upvop_maxfree) 2555 pmap_upvop_maxfree = pmap_upvop_free; 2556 } 2557 2558 void 2559 pmap_pool_mfree(struct pool *pp, void *va) 2560 { 2561 struct pvo_page *pvop; 2562 2563 pvop = va; 2564 SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link); 2565 pmap_mpvop_free++; 2566 if (pmap_mpvop_free > pmap_mpvop_maxfree) 2567 pmap_mpvop_maxfree = pmap_mpvop_free; 2568 #if 0 2569 uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va)); 2570 #endif 2571 } 2572 2573 /* 2574 * This routine in bootstraping to steal to-be-managed memory (which will 2575 * then be unmanaged). We use it to grab from the first 256MB for our 2576 * pmap needs and above 256MB for other stuff. 2577 */ 2578 vaddr_t 2579 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp) 2580 { 2581 vsize_t size; 2582 vaddr_t va; 2583 paddr_t pa = 0; 2584 int npgs, bank; 2585 struct vm_physseg *ps; 2586 2587 if (uvm.page_init_done == TRUE) 2588 panic("pmap_steal_memory: called _after_ bootstrap"); 2589 2590 *vstartp = VM_MIN_KERNEL_ADDRESS; 2591 *vendp = VM_MAX_KERNEL_ADDRESS; 2592 2593 size = round_page(vsize); 2594 npgs = atop(size); 2595 2596 /* 2597 * PA 0 will never be among those given to UVM so we can use it 2598 * to indicate we couldn't steal any memory. 2599 */ 2600 for (ps = vm_physmem, bank = 0; bank < vm_nphysseg; bank++, ps++) { 2601 if (ps->free_list == VM_FREELIST_FIRST256 && 2602 ps->avail_end - ps->avail_start >= npgs) { 2603 pa = ptoa(ps->avail_start); 2604 break; 2605 } 2606 } 2607 2608 if (pa == 0) 2609 panic("pmap_steal_memory: no approriate memory to steal!"); 2610 2611 ps->avail_start += npgs; 2612 ps->start += npgs; 2613 2614 /* 2615 * If we've used up all the pages in the segment, remove it and 2616 * compact the list. 2617 */ 2618 if (ps->avail_start == ps->end) { 2619 /* 2620 * If this was the last one, then a very bad thing has occurred 2621 */ 2622 if (--vm_nphysseg == 0) 2623 panic("pmap_steal_memory: out of memory!"); 2624 2625 printf("pmap_steal_memory: consumed bank %d\n", bank); 2626 for (; bank < vm_nphysseg; bank++, ps++) { 2627 ps[0] = ps[1]; 2628 } 2629 } 2630 2631 va = (vaddr_t) pa; 2632 memset((caddr_t) va, 0, size); 2633 pmap_pages_stolen += npgs; 2634 #ifdef DEBUG 2635 if (pmapdebug && npgs > 1) { 2636 u_int cnt = 0; 2637 for (bank = 0, ps = vm_physmem; bank < vm_nphysseg; bank++, ps++) 2638 cnt += ps->avail_end - ps->avail_start; 2639 printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n", 2640 npgs, pmap_pages_stolen, cnt); 2641 } 2642 #endif 2643 2644 return va; 2645 } 2646 2647 /* 2648 * Find a chuck of memory with right size and alignment. 2649 */ 2650 void * 2651 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end) 2652 { 2653 struct mem_region *mp; 2654 paddr_t s, e; 2655 int i, j; 2656 2657 size = round_page(size); 2658 2659 DPRINTFN(BOOT, 2660 ("pmap_boot_find_memory: size=%lx, alignment=%lx, at_end=%d", 2661 size, alignment, at_end)); 2662 2663 if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0) 2664 panic("pmap_boot_find_memory: invalid alignment %lx", 2665 alignment); 2666 2667 if (at_end) { 2668 if (alignment != PAGE_SIZE) 2669 panic("pmap_boot_find_memory: invalid ending " 2670 "alignment %lx", alignment); 2671 2672 for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) { 2673 s = mp->start + mp->size - size; 2674 if (s >= mp->start && mp->size >= size) { 2675 DPRINTFN(BOOT,(": %lx\n", s)); 2676 DPRINTFN(BOOT, 2677 ("pmap_boot_find_memory: b-avail[%d] start " 2678 "0x%lx size 0x%lx\n", mp - avail, 2679 mp->start, mp->size)); 2680 mp->size -= size; 2681 DPRINTFN(BOOT, 2682 ("pmap_boot_find_memory: a-avail[%d] start " 2683 "0x%lx size 0x%lx\n", mp - avail, 2684 mp->start, mp->size)); 2685 return (void *) s; 2686 } 2687 } 2688 panic("pmap_boot_find_memory: no available memory"); 2689 } 2690 2691 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2692 s = (mp->start + alignment - 1) & ~(alignment-1); 2693 e = s + size; 2694 2695 /* 2696 * Is the calculated region entirely within the region? 2697 */ 2698 if (s < mp->start || e > mp->start + mp->size) 2699 continue; 2700 2701 DPRINTFN(BOOT,(": %lx\n", s)); 2702 if (s == mp->start) { 2703 /* 2704 * If the block starts at the beginning of region, 2705 * adjust the size & start. (the region may now be 2706 * zero in length) 2707 */ 2708 DPRINTFN(BOOT, 2709 ("pmap_boot_find_memory: b-avail[%d] start " 2710 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2711 mp->start += size; 2712 mp->size -= size; 2713 DPRINTFN(BOOT, 2714 ("pmap_boot_find_memory: a-avail[%d] start " 2715 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2716 } else if (e == mp->start + mp->size) { 2717 /* 2718 * If the block starts at the beginning of region, 2719 * adjust only the size. 2720 */ 2721 DPRINTFN(BOOT, 2722 ("pmap_boot_find_memory: b-avail[%d] start " 2723 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2724 mp->size -= size; 2725 DPRINTFN(BOOT, 2726 ("pmap_boot_find_memory: a-avail[%d] start " 2727 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2728 } else { 2729 /* 2730 * Block is in the middle of the region, so we 2731 * have to split it in two. 2732 */ 2733 for (j = avail_cnt; j > i + 1; j--) { 2734 avail[j] = avail[j-1]; 2735 } 2736 DPRINTFN(BOOT, 2737 ("pmap_boot_find_memory: b-avail[%d] start " 2738 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2739 mp[1].start = e; 2740 mp[1].size = mp[0].start + mp[0].size - e; 2741 mp[0].size = s - mp[0].start; 2742 avail_cnt++; 2743 for (; i < avail_cnt; i++) { 2744 DPRINTFN(BOOT, 2745 ("pmap_boot_find_memory: a-avail[%d] " 2746 "start 0x%lx size 0x%lx\n", i, 2747 avail[i].start, avail[i].size)); 2748 } 2749 } 2750 return (void *) s; 2751 } 2752 panic("pmap_boot_find_memory: not enough memory for " 2753 "%lx/%lx allocation?", size, alignment); 2754 } 2755 2756 /* 2757 * This is not part of the defined PMAP interface and is specific to the 2758 * PowerPC architecture. This is called during initppc, before the system 2759 * is really initialized. 2760 */ 2761 void 2762 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend) 2763 { 2764 struct mem_region *mp, tmp; 2765 paddr_t s, e; 2766 psize_t size; 2767 int i, j; 2768 2769 /* 2770 * Get memory. 2771 */ 2772 mem_regions(&mem, &avail); 2773 #if defined(DEBUG) 2774 if (pmapdebug & PMAPDEBUG_BOOT) { 2775 printf("pmap_bootstrap: memory configuration:\n"); 2776 for (mp = mem; mp->size; mp++) { 2777 printf("pmap_bootstrap: mem start 0x%lx size 0x%lx\n", 2778 mp->start, mp->size); 2779 } 2780 for (mp = avail; mp->size; mp++) { 2781 printf("pmap_bootstrap: avail start 0x%lx size 0x%lx\n", 2782 mp->start, mp->size); 2783 } 2784 } 2785 #endif 2786 2787 /* 2788 * Find out how much physical memory we have and in how many chunks. 2789 */ 2790 for (mem_cnt = 0, mp = mem; mp->size; mp++) { 2791 if (mp->start >= pmap_memlimit) 2792 continue; 2793 if (mp->start + mp->size > pmap_memlimit) { 2794 size = pmap_memlimit - mp->start; 2795 physmem += btoc(size); 2796 } else { 2797 physmem += btoc(mp->size); 2798 } 2799 mem_cnt++; 2800 } 2801 2802 /* 2803 * Count the number of available entries. 2804 */ 2805 for (avail_cnt = 0, mp = avail; mp->size; mp++) 2806 avail_cnt++; 2807 2808 /* 2809 * Page align all regions. 2810 */ 2811 kernelstart = trunc_page(kernelstart); 2812 kernelend = round_page(kernelend); 2813 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2814 s = round_page(mp->start); 2815 mp->size -= (s - mp->start); 2816 mp->size = trunc_page(mp->size); 2817 mp->start = s; 2818 e = mp->start + mp->size; 2819 2820 DPRINTFN(BOOT, 2821 ("pmap_bootstrap: b-avail[%d] start 0x%lx size 0x%lx\n", 2822 i, mp->start, mp->size)); 2823 2824 /* 2825 * Don't allow the end to run beyond our artificial limit 2826 */ 2827 if (e > pmap_memlimit) 2828 e = pmap_memlimit; 2829 2830 /* 2831 * Is this region empty or strange? skip it. 2832 */ 2833 if (e <= s) { 2834 mp->start = 0; 2835 mp->size = 0; 2836 continue; 2837 } 2838 2839 /* 2840 * Does this overlap the beginning of kernel? 2841 * Does extend past the end of the kernel? 2842 */ 2843 else if (s < kernelstart && e > kernelstart) { 2844 if (e > kernelend) { 2845 avail[avail_cnt].start = kernelend; 2846 avail[avail_cnt].size = e - kernelend; 2847 avail_cnt++; 2848 } 2849 mp->size = kernelstart - s; 2850 } 2851 /* 2852 * Check whether this region overlaps the end of the kernel. 2853 */ 2854 else if (s < kernelend && e > kernelend) { 2855 mp->start = kernelend; 2856 mp->size = e - kernelend; 2857 } 2858 /* 2859 * Look whether this regions is completely inside the kernel. 2860 * Nuke it if it does. 2861 */ 2862 else if (s >= kernelstart && e <= kernelend) { 2863 mp->start = 0; 2864 mp->size = 0; 2865 } 2866 /* 2867 * If the user imposed a memory limit, enforce it. 2868 */ 2869 else if (s >= pmap_memlimit) { 2870 mp->start = -PAGE_SIZE; /* let's know why */ 2871 mp->size = 0; 2872 } 2873 else { 2874 mp->start = s; 2875 mp->size = e - s; 2876 } 2877 DPRINTFN(BOOT, 2878 ("pmap_bootstrap: a-avail[%d] start 0x%lx size 0x%lx\n", 2879 i, mp->start, mp->size)); 2880 } 2881 2882 /* 2883 * Move (and uncount) all the null return to the end. 2884 */ 2885 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2886 if (mp->size == 0) { 2887 tmp = avail[i]; 2888 avail[i] = avail[--avail_cnt]; 2889 avail[avail_cnt] = avail[i]; 2890 } 2891 } 2892 2893 /* 2894 * (Bubble)sort them into asecnding order. 2895 */ 2896 for (i = 0; i < avail_cnt; i++) { 2897 for (j = i + 1; j < avail_cnt; j++) { 2898 if (avail[i].start > avail[j].start) { 2899 tmp = avail[i]; 2900 avail[i] = avail[j]; 2901 avail[j] = tmp; 2902 } 2903 } 2904 } 2905 2906 /* 2907 * Make sure they don't overlap. 2908 */ 2909 for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) { 2910 if (mp[0].start + mp[0].size > mp[1].start) { 2911 mp[0].size = mp[1].start - mp[0].start; 2912 } 2913 DPRINTFN(BOOT, 2914 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 2915 i, mp->start, mp->size)); 2916 } 2917 DPRINTFN(BOOT, 2918 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 2919 i, mp->start, mp->size)); 2920 2921 #ifdef PTEGCOUNT 2922 pmap_pteg_cnt = PTEGCOUNT; 2923 #else /* PTEGCOUNT */ 2924 pmap_pteg_cnt = 0x1000; 2925 2926 while (pmap_pteg_cnt < physmem) 2927 pmap_pteg_cnt <<= 1; 2928 2929 pmap_pteg_cnt >>= 1; 2930 #endif /* PTEGCOUNT */ 2931 2932 /* 2933 * Find suitably aligned memory for PTEG hash table. 2934 */ 2935 size = pmap_pteg_cnt * sizeof(struct pteg); 2936 pmap_pteg_table = pmap_boot_find_memory(size, size, 0); 2937 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 2938 if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH) 2939 panic("pmap_bootstrap: pmap_pteg_table end (%p + %lx) > 256MB", 2940 pmap_pteg_table, size); 2941 #endif 2942 2943 memset((void *)pmap_pteg_table, 0, pmap_pteg_cnt * sizeof(struct pteg)); 2944 pmap_pteg_mask = pmap_pteg_cnt - 1; 2945 2946 /* 2947 * We cannot do pmap_steal_memory here since UVM hasn't been loaded 2948 * with pages. So we just steal them before giving them to UVM. 2949 */ 2950 size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt; 2951 pmap_pvo_table = pmap_boot_find_memory(size, PAGE_SIZE, 0); 2952 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 2953 if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH) 2954 panic("pmap_bootstrap: pmap_pvo_table end (%p + %lx) > 256MB", 2955 pmap_pvo_table, size); 2956 #endif 2957 2958 for (i = 0; i < pmap_pteg_cnt; i++) 2959 TAILQ_INIT(&pmap_pvo_table[i]); 2960 2961 #ifndef MSGBUFADDR 2962 /* 2963 * Allocate msgbuf in high memory. 2964 */ 2965 msgbuf_paddr = 2966 (paddr_t) pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1); 2967 #endif 2968 2969 #ifdef __HAVE_PMAP_PHYSSEG 2970 { 2971 u_int npgs = 0; 2972 for (i = 0, mp = avail; i < avail_cnt; i++, mp++) 2973 npgs += btoc(mp->size); 2974 size = (sizeof(struct pvo_head) + 1) * npgs; 2975 pmap_physseg.pvoh = pmap_boot_find_memory(size, PAGE_SIZE, 0); 2976 pmap_physseg.attrs = (char *) &pmap_physseg.pvoh[npgs]; 2977 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 2978 if ((uintptr_t)pmap_physseg.pvoh + size > SEGMENT_LENGTH) 2979 panic("pmap_bootstrap: PVO list end (%p + %lx) > 256MB", 2980 pmap_physseg.pvoh, size); 2981 #endif 2982 } 2983 #endif 2984 2985 for (mp = avail, i = 0; i < avail_cnt; mp++, i++) { 2986 paddr_t pfstart = atop(mp->start); 2987 paddr_t pfend = atop(mp->start + mp->size); 2988 if (mp->size == 0) 2989 continue; 2990 if (mp->start + mp->size <= SEGMENT_LENGTH) { 2991 uvm_page_physload(pfstart, pfend, pfstart, pfend, 2992 VM_FREELIST_FIRST256); 2993 } else if (mp->start >= SEGMENT_LENGTH) { 2994 uvm_page_physload(pfstart, pfend, pfstart, pfend, 2995 VM_FREELIST_DEFAULT); 2996 } else { 2997 pfend = atop(SEGMENT_LENGTH); 2998 uvm_page_physload(pfstart, pfend, pfstart, pfend, 2999 VM_FREELIST_FIRST256); 3000 pfstart = atop(SEGMENT_LENGTH); 3001 pfend = atop(mp->start + mp->size); 3002 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3003 VM_FREELIST_DEFAULT); 3004 } 3005 } 3006 3007 /* 3008 * Make sure kernel vsid is allocated as well as VSID 0. 3009 */ 3010 pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW] 3011 |= 1 << (KERNEL_VSIDBITS % VSID_NBPW); 3012 pmap_vsid_bitmap[0] |= 1; 3013 3014 /* 3015 * Initialize kernel pmap and hardware. 3016 */ 3017 for (i = 0; i < 16; i++) { 3018 pmap_kernel()->pm_sr[i] = EMPTY_SEGMENT; 3019 __asm __volatile ("mtsrin %0,%1" 3020 :: "r"(EMPTY_SEGMENT), "r"(i << ADDR_SR_SHFT)); 3021 } 3022 3023 pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY; 3024 __asm __volatile ("mtsr %0,%1" 3025 :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT)); 3026 #ifdef KERNEL2_SR 3027 pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY; 3028 __asm __volatile ("mtsr %0,%1" 3029 :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT)); 3030 #endif 3031 for (i = 0; i < 16; i++) { 3032 if (iosrtable[i] & SR601_T) { 3033 pmap_kernel()->pm_sr[i] = iosrtable[i]; 3034 __asm __volatile ("mtsrin %0,%1" 3035 :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT)); 3036 } 3037 } 3038 3039 __asm __volatile ("sync; mtsdr1 %0; isync" 3040 :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10))); 3041 tlbia(); 3042 3043 #ifdef ALTIVEC 3044 pmap_use_altivec = cpu_altivec; 3045 #endif 3046 3047 #ifdef DEBUG 3048 if (pmapdebug & PMAPDEBUG_BOOT) { 3049 u_int cnt; 3050 int bank; 3051 char pbuf[9]; 3052 for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) { 3053 cnt += vm_physmem[bank].avail_end - vm_physmem[bank].avail_start; 3054 printf("pmap_bootstrap: vm_physmem[%d]=%#lx-%#lx/%#lx\n", 3055 bank, 3056 ptoa(vm_physmem[bank].avail_start), 3057 ptoa(vm_physmem[bank].avail_end), 3058 ptoa(vm_physmem[bank].avail_end - vm_physmem[bank].avail_start)); 3059 } 3060 format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt)); 3061 printf("pmap_bootstrap: UVM memory = %s (%u pages)\n", 3062 pbuf, cnt); 3063 } 3064 #endif 3065 3066 pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry), 3067 sizeof(struct pvo_entry), 0, 0, "pmap_upvopl", 3068 &pmap_pool_uallocator); 3069 3070 pool_setlowat(&pmap_upvo_pool, 252); 3071 3072 pool_init(&pmap_pool, sizeof(struct pmap), 3073 sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator); 3074 } 3075
Indexes created Fri Sep 26 08:10:20 GMT 2025