1 /* $NetBSD: pmap.c,v 1.35 2005/12/24 20:07:28 perry 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@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 1.35 2005/12/24 20:07:28 perry Exp $"); 71 72 #include "opt_ppcarch.h" 73 #include "opt_altivec.h" 74 #include "opt_pmap.h" 75 #include <sys/param.h> 76 #include <sys/malloc.h> 77 #include <sys/proc.h> 78 #include <sys/user.h> 79 #include <sys/pool.h> 80 #include <sys/queue.h> 81 #include <sys/device.h> /* for evcnt */ 82 #include <sys/systm.h> 83 84 #if __NetBSD_Version__ < 105010000 85 #include <vm/vm.h> 86 #include <vm/vm_kern.h> 87 #define splvm() splimp() 88 #endif 89 90 #include <uvm/uvm.h> 91 92 #include <machine/pcb.h> 93 #include <machine/powerpc.h> 94 #include <powerpc/spr.h> 95 #include <powerpc/oea/sr_601.h> 96 #include <powerpc/bat.h> 97 98 #if defined(DEBUG) || defined(PMAPCHECK) 99 #define STATIC 100 #else 101 #define STATIC static 102 #endif 103 104 #ifdef ALTIVEC 105 int pmap_use_altivec; 106 #endif 107 108 volatile struct pteg *pmap_pteg_table; 109 unsigned int pmap_pteg_cnt; 110 unsigned int pmap_pteg_mask; 111 #ifdef PMAP_MEMLIMIT 112 paddr_t pmap_memlimit = PMAP_MEMLIMIT; 113 #else 114 paddr_t pmap_memlimit = -PAGE_SIZE; /* there is no limit */ 115 #endif 116 117 struct pmap kernel_pmap_; 118 unsigned int pmap_pages_stolen; 119 u_long pmap_pte_valid; 120 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 121 u_long pmap_pvo_enter_depth; 122 u_long pmap_pvo_remove_depth; 123 #endif 124 125 int physmem; 126 #ifndef MSGBUFADDR 127 extern paddr_t msgbuf_paddr; 128 #endif 129 130 static struct mem_region *mem, *avail; 131 static u_int mem_cnt, avail_cnt; 132 133 #ifdef __HAVE_PMAP_PHYSSEG 134 /* 135 * This is a cache of referenced/modified bits. 136 * Bits herein are shifted by ATTRSHFT. 137 */ 138 #define ATTR_SHFT 4 139 struct pmap_physseg pmap_physseg; 140 #endif 141 142 /* 143 * The following structure is exactly 32 bytes long (one cacheline). 144 */ 145 struct pvo_entry { 146 LIST_ENTRY(pvo_entry) pvo_vlink; /* Link to common virt page */ 147 TAILQ_ENTRY(pvo_entry) pvo_olink; /* Link to overflow entry */ 148 struct pte pvo_pte; /* Prebuilt PTE */ 149 pmap_t pvo_pmap; /* ptr to owning pmap */ 150 vaddr_t pvo_vaddr; /* VA of entry */ 151 #define PVO_PTEGIDX_MASK 0x0007 /* which PTEG slot */ 152 #define PVO_PTEGIDX_VALID 0x0008 /* slot is valid */ 153 #define PVO_WIRED 0x0010 /* PVO entry is wired */ 154 #define PVO_MANAGED 0x0020 /* PVO e. for managed page */ 155 #define PVO_EXECUTABLE 0x0040 /* PVO e. for executable page */ 156 #define PVO_ENTER_INSERT 0 /* PVO has been removed */ 157 #define PVO_SPILL_UNSET 1 /* PVO has been evicted */ 158 #define PVO_SPILL_SET 2 /* PVO has been spilled */ 159 #define PVO_SPILL_INSERT 3 /* PVO has been inserted */ 160 #define PVO_PMAP_PAGE_PROTECT 4 /* PVO has changed */ 161 #define PVO_PMAP_PROTECT 5 /* PVO has changed */ 162 #define PVO_REMOVE 6 /* PVO has been removed */ 163 #define PVO_WHERE_MASK 15 164 #define PVO_WHERE_SHFT 8 165 }; 166 #define PVO_VADDR(pvo) ((pvo)->pvo_vaddr & ~ADDR_POFF) 167 #define PVO_ISEXECUTABLE(pvo) ((pvo)->pvo_vaddr & PVO_EXECUTABLE) 168 #define PVO_PTEGIDX_GET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK) 169 #define PVO_PTEGIDX_ISSET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID) 170 #define PVO_PTEGIDX_CLR(pvo) \ 171 ((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK))) 172 #define PVO_PTEGIDX_SET(pvo,i) \ 173 ((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID)) 174 #define PVO_WHERE(pvo,w) \ 175 ((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \ 176 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT)) 177 178 TAILQ_HEAD(pvo_tqhead, pvo_entry); 179 struct pvo_tqhead *pmap_pvo_table; /* pvo entries by ptegroup index */ 180 struct pvo_head pmap_pvo_kunmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged); /* list of unmanaged pages */ 181 struct pvo_head pmap_pvo_unmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged); /* list of unmanaged pages */ 182 183 struct pool pmap_pool; /* pool for pmap structures */ 184 struct pool pmap_upvo_pool; /* pool for pvo entries for unmanaged pages */ 185 struct pool pmap_mpvo_pool; /* pool for pvo entries for managed pages */ 186 187 /* 188 * We keep a cache of unmanaged pages to be used for pvo entries for 189 * unmanaged pages. 190 */ 191 struct pvo_page { 192 SIMPLEQ_ENTRY(pvo_page) pvop_link; 193 }; 194 SIMPLEQ_HEAD(pvop_head, pvo_page); 195 struct pvop_head pmap_upvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_upvop_head); 196 struct pvop_head pmap_mpvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_mpvop_head); 197 u_long pmap_upvop_free; 198 u_long pmap_upvop_maxfree; 199 u_long pmap_mpvop_free; 200 u_long pmap_mpvop_maxfree; 201 202 STATIC void *pmap_pool_ualloc(struct pool *, int); 203 STATIC void *pmap_pool_malloc(struct pool *, int); 204 205 STATIC void pmap_pool_ufree(struct pool *, void *); 206 STATIC void pmap_pool_mfree(struct pool *, void *); 207 208 static struct pool_allocator pmap_pool_mallocator = { 209 pmap_pool_malloc, pmap_pool_mfree, 0, 210 }; 211 212 static struct pool_allocator pmap_pool_uallocator = { 213 pmap_pool_ualloc, pmap_pool_ufree, 0, 214 }; 215 216 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 217 void pmap_pte_print(volatile struct pte *); 218 #endif 219 220 #ifdef DDB 221 void pmap_pteg_check(void); 222 void pmap_pteg_dist(void); 223 void pmap_print_pte(pmap_t, vaddr_t); 224 void pmap_print_mmuregs(void); 225 #endif 226 227 #if defined(DEBUG) || defined(PMAPCHECK) 228 #ifdef PMAPCHECK 229 int pmapcheck = 1; 230 #else 231 int pmapcheck = 0; 232 #endif 233 void pmap_pvo_verify(void); 234 STATIC void pmap_pvo_check(const struct pvo_entry *); 235 #define PMAP_PVO_CHECK(pvo) \ 236 do { \ 237 if (pmapcheck) \ 238 pmap_pvo_check(pvo); \ 239 } while (0) 240 #else 241 #define PMAP_PVO_CHECK(pvo) do { } while (/*CONSTCOND*/0) 242 #endif 243 STATIC int pmap_pte_insert(int, struct pte *); 244 STATIC int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *, 245 vaddr_t, paddr_t, register_t, int); 246 STATIC void pmap_pvo_remove(struct pvo_entry *, int, struct pvo_head *); 247 STATIC void pmap_pvo_free(struct pvo_entry *); 248 STATIC void pmap_pvo_free_list(struct pvo_head *); 249 STATIC struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *); 250 STATIC volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int); 251 STATIC struct pvo_entry *pmap_pvo_reclaim(struct pmap *); 252 STATIC void pvo_set_exec(struct pvo_entry *); 253 STATIC void pvo_clear_exec(struct pvo_entry *); 254 255 STATIC void tlbia(void); 256 257 STATIC void pmap_release(pmap_t); 258 STATIC void *pmap_boot_find_memory(psize_t, psize_t, int); 259 260 static uint32_t pmap_pvo_reclaim_nextidx; 261 #ifdef DEBUG 262 static int pmap_pvo_reclaim_debugctr; 263 #endif 264 265 #define VSID_NBPW (sizeof(uint32_t) * 8) 266 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW]; 267 268 static int pmap_initialized; 269 270 #if defined(DEBUG) || defined(PMAPDEBUG) 271 #define PMAPDEBUG_BOOT 0x0001 272 #define PMAPDEBUG_PTE 0x0002 273 #define PMAPDEBUG_EXEC 0x0008 274 #define PMAPDEBUG_PVOENTER 0x0010 275 #define PMAPDEBUG_PVOREMOVE 0x0020 276 #define PMAPDEBUG_ACTIVATE 0x0100 277 #define PMAPDEBUG_CREATE 0x0200 278 #define PMAPDEBUG_ENTER 0x1000 279 #define PMAPDEBUG_KENTER 0x2000 280 #define PMAPDEBUG_KREMOVE 0x4000 281 #define PMAPDEBUG_REMOVE 0x8000 282 unsigned int pmapdebug = 0; 283 # define DPRINTF(x) printf x 284 # define DPRINTFN(n, x) if (pmapdebug & PMAPDEBUG_ ## n) printf x 285 #else 286 # define DPRINTF(x) 287 # define DPRINTFN(n, x) 288 #endif 289 290 291 #ifdef PMAPCOUNTERS 292 #define PMAPCOUNT(ev) ((pmap_evcnt_ ## ev).ev_count++) 293 #define PMAPCOUNT2(ev) ((ev).ev_count++) 294 295 struct evcnt pmap_evcnt_mappings = 296 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 297 "pmap", "pages mapped"); 298 struct evcnt pmap_evcnt_unmappings = 299 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 300 "pmap", "pages unmapped"); 301 302 struct evcnt pmap_evcnt_kernel_mappings = 303 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 304 "pmap", "kernel pages mapped"); 305 struct evcnt pmap_evcnt_kernel_unmappings = 306 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_kernel_mappings, 307 "pmap", "kernel pages unmapped"); 308 309 struct evcnt pmap_evcnt_mappings_replaced = 310 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 311 "pmap", "page mappings replaced"); 312 313 struct evcnt pmap_evcnt_exec_mappings = 314 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 315 "pmap", "exec pages mapped"); 316 struct evcnt pmap_evcnt_exec_cached = 317 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 318 "pmap", "exec pages cached"); 319 320 struct evcnt pmap_evcnt_exec_synced = 321 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 322 "pmap", "exec pages synced"); 323 struct evcnt pmap_evcnt_exec_synced_clear_modify = 324 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 325 "pmap", "exec pages synced (CM)"); 326 327 struct evcnt pmap_evcnt_exec_uncached_page_protect = 328 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 329 "pmap", "exec pages uncached (PP)"); 330 struct evcnt pmap_evcnt_exec_uncached_clear_modify = 331 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 332 "pmap", "exec pages uncached (CM)"); 333 struct evcnt pmap_evcnt_exec_uncached_zero_page = 334 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 335 "pmap", "exec pages uncached (ZP)"); 336 struct evcnt pmap_evcnt_exec_uncached_copy_page = 337 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 338 "pmap", "exec pages uncached (CP)"); 339 340 struct evcnt pmap_evcnt_updates = 341 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 342 "pmap", "updates"); 343 struct evcnt pmap_evcnt_collects = 344 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 345 "pmap", "collects"); 346 struct evcnt pmap_evcnt_copies = 347 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 348 "pmap", "copies"); 349 350 struct evcnt pmap_evcnt_ptes_spilled = 351 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 352 "pmap", "ptes spilled from overflow"); 353 struct evcnt pmap_evcnt_ptes_unspilled = 354 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 355 "pmap", "ptes not spilled"); 356 struct evcnt pmap_evcnt_ptes_evicted = 357 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 358 "pmap", "ptes evicted"); 359 360 struct evcnt pmap_evcnt_ptes_primary[8] = { 361 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 362 "pmap", "ptes added at primary[0]"), 363 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 364 "pmap", "ptes added at primary[1]"), 365 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 366 "pmap", "ptes added at primary[2]"), 367 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 368 "pmap", "ptes added at primary[3]"), 369 370 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 371 "pmap", "ptes added at primary[4]"), 372 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 373 "pmap", "ptes added at primary[5]"), 374 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 375 "pmap", "ptes added at primary[6]"), 376 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 377 "pmap", "ptes added at primary[7]"), 378 }; 379 struct evcnt pmap_evcnt_ptes_secondary[8] = { 380 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 381 "pmap", "ptes added at secondary[0]"), 382 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 383 "pmap", "ptes added at secondary[1]"), 384 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 385 "pmap", "ptes added at secondary[2]"), 386 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 387 "pmap", "ptes added at secondary[3]"), 388 389 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 390 "pmap", "ptes added at secondary[4]"), 391 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 392 "pmap", "ptes added at secondary[5]"), 393 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 394 "pmap", "ptes added at secondary[6]"), 395 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 396 "pmap", "ptes added at secondary[7]"), 397 }; 398 struct evcnt pmap_evcnt_ptes_removed = 399 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 400 "pmap", "ptes removed"); 401 struct evcnt pmap_evcnt_ptes_changed = 402 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 403 "pmap", "ptes changed"); 404 struct evcnt pmap_evcnt_pvos_reclaimed = 405 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 406 "pmap", "pvos reclaimed"); 407 struct evcnt pmap_evcnt_pvos_failed = 408 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 409 "pmap", "pvo allocation failures"); 410 411 /* 412 * From pmap_subr.c 413 */ 414 extern struct evcnt pmap_evcnt_zeroed_pages; 415 extern struct evcnt pmap_evcnt_copied_pages; 416 extern struct evcnt pmap_evcnt_idlezeroed_pages; 417 418 EVCNT_ATTACH_STATIC(pmap_evcnt_mappings); 419 EVCNT_ATTACH_STATIC(pmap_evcnt_mappings_replaced); 420 EVCNT_ATTACH_STATIC(pmap_evcnt_unmappings); 421 422 EVCNT_ATTACH_STATIC(pmap_evcnt_kernel_mappings); 423 EVCNT_ATTACH_STATIC(pmap_evcnt_kernel_unmappings); 424 425 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_mappings); 426 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_cached); 427 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_synced); 428 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_synced_clear_modify); 429 430 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_page_protect); 431 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_clear_modify); 432 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_zero_page); 433 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_copy_page); 434 435 EVCNT_ATTACH_STATIC(pmap_evcnt_zeroed_pages); 436 EVCNT_ATTACH_STATIC(pmap_evcnt_copied_pages); 437 EVCNT_ATTACH_STATIC(pmap_evcnt_idlezeroed_pages); 438 439 EVCNT_ATTACH_STATIC(pmap_evcnt_updates); 440 EVCNT_ATTACH_STATIC(pmap_evcnt_collects); 441 EVCNT_ATTACH_STATIC(pmap_evcnt_copies); 442 443 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_spilled); 444 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_unspilled); 445 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_evicted); 446 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_removed); 447 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_changed); 448 449 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 0); 450 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 1); 451 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 2); 452 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 3); 453 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 4); 454 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 5); 455 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 6); 456 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 7); 457 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 0); 458 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 1); 459 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 2); 460 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 3); 461 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 4); 462 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 5); 463 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 6); 464 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 7); 465 466 EVCNT_ATTACH_STATIC(pmap_evcnt_pvos_reclaimed); 467 EVCNT_ATTACH_STATIC(pmap_evcnt_pvos_failed); 468 #else 469 #define PMAPCOUNT(ev) ((void) 0) 470 #define PMAPCOUNT2(ev) ((void) 0) 471 #endif 472 473 #define TLBIE(va) __asm volatile("tlbie %0" :: "r"(va)) 474 #define TLBSYNC() __asm volatile("tlbsync") 475 #define SYNC() __asm volatile("sync") 476 #define EIEIO() __asm volatile("eieio") 477 #define MFMSR() mfmsr() 478 #define MTMSR(psl) mtmsr(psl) 479 #define MFPVR() mfpvr() 480 #define MFSRIN(va) mfsrin(va) 481 #define MFTB() mfrtcltbl() 482 483 #ifndef PPC_OEA64 484 static inline register_t 485 mfsrin(vaddr_t va) 486 { 487 register_t sr; 488 __asm volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va)); 489 return sr; 490 } 491 #endif /* PPC_OEA64 */ 492 493 static inline register_t 494 pmap_interrupts_off(void) 495 { 496 register_t msr = MFMSR(); 497 if (msr & PSL_EE) 498 MTMSR(msr & ~PSL_EE); 499 return msr; 500 } 501 502 static void 503 pmap_interrupts_restore(register_t msr) 504 { 505 if (msr & PSL_EE) 506 MTMSR(msr); 507 } 508 509 static inline u_int32_t 510 mfrtcltbl(void) 511 { 512 513 if ((MFPVR() >> 16) == MPC601) 514 return (mfrtcl() >> 7); 515 else 516 return (mftbl()); 517 } 518 519 /* 520 * These small routines may have to be replaced, 521 * if/when we support processors other that the 604. 522 */ 523 524 void 525 tlbia(void) 526 { 527 caddr_t i; 528 529 SYNC(); 530 /* 531 * Why not use "tlbia"? Because not all processors implement it. 532 * 533 * This needs to be a per-CPU callback to do the appropriate thing 534 * for the CPU. XXX 535 */ 536 for (i = 0; i < (caddr_t)0x00040000; i += 0x00001000) { 537 TLBIE(i); 538 EIEIO(); 539 SYNC(); 540 } 541 TLBSYNC(); 542 SYNC(); 543 } 544 545 static inline register_t 546 va_to_vsid(const struct pmap *pm, vaddr_t addr) 547 { 548 #ifdef PPC_OEA64 549 #if 0 550 const struct ste *ste; 551 register_t hash; 552 int i; 553 554 hash = (addr >> ADDR_ESID_SHFT) & ADDR_ESID_HASH; 555 556 /* 557 * Try the primary group first 558 */ 559 ste = pm->pm_stes[hash].stes; 560 for (i = 0; i < 8; i++, ste++) { 561 if (ste->ste_hi & STE_V) && 562 (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID)) 563 return ste; 564 } 565 566 /* 567 * Then the secondary group. 568 */ 569 ste = pm->pm_stes[hash ^ ADDR_ESID_HASH].stes; 570 for (i = 0; i < 8; i++, ste++) { 571 if (ste->ste_hi & STE_V) && 572 (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID)) 573 return addr; 574 } 575 576 return NULL; 577 #else 578 /* 579 * Rather than searching the STE groups for the VSID, we know 580 * how we generate that from the ESID and so do that. 581 */ 582 return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT; 583 #endif 584 #else 585 return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID) >> SR_VSID_SHFT; 586 #endif 587 } 588 589 static inline register_t 590 va_to_pteg(const struct pmap *pm, vaddr_t addr) 591 { 592 register_t hash; 593 594 hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT); 595 return hash & pmap_pteg_mask; 596 } 597 598 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 599 /* 600 * Given a PTE in the page table, calculate the VADDR that hashes to it. 601 * The only bit of magic is that the top 4 bits of the address doesn't 602 * technically exist in the PTE. But we know we reserved 4 bits of the 603 * VSID for it so that's how we get it. 604 */ 605 static vaddr_t 606 pmap_pte_to_va(volatile const struct pte *pt) 607 { 608 vaddr_t va; 609 uintptr_t ptaddr = (uintptr_t) pt; 610 611 if (pt->pte_hi & PTE_HID) 612 ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg)); 613 614 /* PPC Bits 10-19 PPC64 Bits 42-51 */ 615 va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff; 616 va <<= ADDR_PIDX_SHFT; 617 618 /* PPC Bits 4-9 PPC64 Bits 36-41 */ 619 va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT; 620 621 #ifdef PPC_OEA64 622 /* PPC63 Bits 0-35 */ 623 /* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */ 624 #endif 625 #ifdef PPC_OEA 626 /* PPC Bits 0-3 */ 627 va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; 628 #endif 629 630 return va; 631 } 632 #endif 633 634 static inline struct pvo_head * 635 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p) 636 { 637 #ifdef __HAVE_VM_PAGE_MD 638 struct vm_page *pg; 639 640 pg = PHYS_TO_VM_PAGE(pa); 641 if (pg_p != NULL) 642 *pg_p = pg; 643 if (pg == NULL) 644 return &pmap_pvo_unmanaged; 645 return &pg->mdpage.mdpg_pvoh; 646 #endif 647 #ifdef __HAVE_PMAP_PHYSSEG 648 int bank, pg; 649 650 bank = vm_physseg_find(atop(pa), &pg); 651 if (pg_p != NULL) 652 *pg_p = pg; 653 if (bank == -1) 654 return &pmap_pvo_unmanaged; 655 return &vm_physmem[bank].pmseg.pvoh[pg]; 656 #endif 657 } 658 659 static inline struct pvo_head * 660 vm_page_to_pvoh(struct vm_page *pg) 661 { 662 #ifdef __HAVE_VM_PAGE_MD 663 return &pg->mdpage.mdpg_pvoh; 664 #endif 665 #ifdef __HAVE_PMAP_PHYSSEG 666 return pa_to_pvoh(VM_PAGE_TO_PHYS(pg), NULL); 667 #endif 668 } 669 670 671 #ifdef __HAVE_PMAP_PHYSSEG 672 static inline char * 673 pa_to_attr(paddr_t pa) 674 { 675 int bank, pg; 676 677 bank = vm_physseg_find(atop(pa), &pg); 678 if (bank == -1) 679 return NULL; 680 return &vm_physmem[bank].pmseg.attrs[pg]; 681 } 682 #endif 683 684 static inline void 685 pmap_attr_clear(struct vm_page *pg, int ptebit) 686 { 687 #ifdef __HAVE_PMAP_PHYSSEG 688 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) &= ~(ptebit >> ATTR_SHFT); 689 #endif 690 #ifdef __HAVE_VM_PAGE_MD 691 pg->mdpage.mdpg_attrs &= ~ptebit; 692 #endif 693 } 694 695 static inline int 696 pmap_attr_fetch(struct vm_page *pg) 697 { 698 #ifdef __HAVE_PMAP_PHYSSEG 699 return *pa_to_attr(VM_PAGE_TO_PHYS(pg)) << ATTR_SHFT; 700 #endif 701 #ifdef __HAVE_VM_PAGE_MD 702 return pg->mdpage.mdpg_attrs; 703 #endif 704 } 705 706 static inline void 707 pmap_attr_save(struct vm_page *pg, int ptebit) 708 { 709 #ifdef __HAVE_PMAP_PHYSSEG 710 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) |= (ptebit >> ATTR_SHFT); 711 #endif 712 #ifdef __HAVE_VM_PAGE_MD 713 pg->mdpage.mdpg_attrs |= ptebit; 714 #endif 715 } 716 717 static inline int 718 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt) 719 { 720 if (pt->pte_hi == pvo_pt->pte_hi 721 #if 0 722 && ((pt->pte_lo ^ pvo_pt->pte_lo) & 723 ~(PTE_REF|PTE_CHG)) == 0 724 #endif 725 ) 726 return 1; 727 return 0; 728 } 729 730 static inline void 731 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo) 732 { 733 /* 734 * Construct the PTE. Default to IMB initially. Valid bit 735 * only gets set when the real pte is set in memory. 736 * 737 * Note: Don't set the valid bit for correct operation of tlb update. 738 */ 739 pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT) 740 | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API); 741 pt->pte_lo = pte_lo; 742 } 743 744 static inline void 745 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt) 746 { 747 pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG); 748 } 749 750 static inline void 751 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit) 752 { 753 /* 754 * As shown in Section 7.6.3.2.3 755 */ 756 pt->pte_lo &= ~ptebit; 757 TLBIE(va); 758 SYNC(); 759 EIEIO(); 760 TLBSYNC(); 761 SYNC(); 762 } 763 764 static inline void 765 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt) 766 { 767 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 768 if (pvo_pt->pte_hi & PTE_VALID) 769 panic("pte_set: setting an already valid pte %p", pvo_pt); 770 #endif 771 pvo_pt->pte_hi |= PTE_VALID; 772 /* 773 * Update the PTE as defined in section 7.6.3.1 774 * Note that the REF/CHG bits are from pvo_pt and thus should 775 * have been saved so this routine can restore them (if desired). 776 */ 777 pt->pte_lo = pvo_pt->pte_lo; 778 EIEIO(); 779 pt->pte_hi = pvo_pt->pte_hi; 780 SYNC(); 781 pmap_pte_valid++; 782 } 783 784 static inline void 785 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 786 { 787 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 788 if ((pvo_pt->pte_hi & PTE_VALID) == 0) 789 panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt); 790 if ((pt->pte_hi & PTE_VALID) == 0) 791 panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt); 792 #endif 793 794 pvo_pt->pte_hi &= ~PTE_VALID; 795 /* 796 * Force the ref & chg bits back into the PTEs. 797 */ 798 SYNC(); 799 /* 800 * Invalidate the pte ... (Section 7.6.3.3) 801 */ 802 pt->pte_hi &= ~PTE_VALID; 803 SYNC(); 804 TLBIE(va); 805 SYNC(); 806 EIEIO(); 807 TLBSYNC(); 808 SYNC(); 809 /* 810 * Save the ref & chg bits ... 811 */ 812 pmap_pte_synch(pt, pvo_pt); 813 pmap_pte_valid--; 814 } 815 816 static inline void 817 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 818 { 819 /* 820 * Invalidate the PTE 821 */ 822 pmap_pte_unset(pt, pvo_pt, va); 823 pmap_pte_set(pt, pvo_pt); 824 } 825 826 /* 827 * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx 828 * (either primary or secondary location). 829 * 830 * Note: both the destination and source PTEs must not have PTE_VALID set. 831 */ 832 833 STATIC int 834 pmap_pte_insert(int ptegidx, struct pte *pvo_pt) 835 { 836 volatile struct pte *pt; 837 int i; 838 839 #if defined(DEBUG) 840 DPRINTFN(PTE, ("pmap_pte_insert: idx 0x%x, pte 0x%x 0x%x\n", 841 ptegidx, (unsigned int) pvo_pt->pte_hi, (unsigned int) pvo_pt->pte_lo)); 842 #endif 843 /* 844 * First try primary hash. 845 */ 846 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 847 if ((pt->pte_hi & PTE_VALID) == 0) { 848 pvo_pt->pte_hi &= ~PTE_HID; 849 pmap_pte_set(pt, pvo_pt); 850 return i; 851 } 852 } 853 854 /* 855 * Now try secondary hash. 856 */ 857 ptegidx ^= pmap_pteg_mask; 858 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 859 if ((pt->pte_hi & PTE_VALID) == 0) { 860 pvo_pt->pte_hi |= PTE_HID; 861 pmap_pte_set(pt, pvo_pt); 862 return i; 863 } 864 } 865 return -1; 866 } 867 868 /* 869 * Spill handler. 870 * 871 * Tries to spill a page table entry from the overflow area. 872 * This runs in either real mode (if dealing with a exception spill) 873 * or virtual mode when dealing with manually spilling one of the 874 * kernel's pte entries. In either case, interrupts are already 875 * disabled. 876 */ 877 878 int 879 pmap_pte_spill(struct pmap *pm, vaddr_t addr, boolean_t exec) 880 { 881 struct pvo_entry *source_pvo, *victim_pvo, *next_pvo; 882 struct pvo_entry *pvo; 883 /* XXX: gcc -- vpvoh is always set at either *1* or *2* */ 884 struct pvo_tqhead *pvoh, *vpvoh = NULL; 885 int ptegidx, i, j; 886 volatile struct pteg *pteg; 887 volatile struct pte *pt; 888 889 ptegidx = va_to_pteg(pm, addr); 890 891 /* 892 * Have to substitute some entry. Use the primary hash for this. 893 * Use low bits of timebase as random generator. Make sure we are 894 * not picking a kernel pte for replacement. 895 */ 896 pteg = &pmap_pteg_table[ptegidx]; 897 i = MFTB() & 7; 898 for (j = 0; j < 8; j++) { 899 pt = &pteg->pt[i]; 900 if ((pt->pte_hi & PTE_VALID) == 0 || 901 VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT) 902 != KERNEL_VSIDBITS) 903 break; 904 i = (i + 1) & 7; 905 } 906 KASSERT(j < 8); 907 908 source_pvo = NULL; 909 victim_pvo = NULL; 910 pvoh = &pmap_pvo_table[ptegidx]; 911 TAILQ_FOREACH(pvo, pvoh, pvo_olink) { 912 913 /* 914 * We need to find pvo entry for this address... 915 */ 916 PMAP_PVO_CHECK(pvo); /* sanity check */ 917 918 /* 919 * If we haven't found the source and we come to a PVO with 920 * a valid PTE, then we know we can't find it because all 921 * evicted PVOs always are first in the list. 922 */ 923 if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID)) 924 break; 925 if (source_pvo == NULL && pm == pvo->pvo_pmap && 926 addr == PVO_VADDR(pvo)) { 927 928 /* 929 * Now we have found the entry to be spilled into the 930 * pteg. Attempt to insert it into the page table. 931 */ 932 j = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 933 if (j >= 0) { 934 PVO_PTEGIDX_SET(pvo, j); 935 PMAP_PVO_CHECK(pvo); /* sanity check */ 936 PVO_WHERE(pvo, SPILL_INSERT); 937 pvo->pvo_pmap->pm_evictions--; 938 PMAPCOUNT(ptes_spilled); 939 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 940 ? pmap_evcnt_ptes_secondary 941 : pmap_evcnt_ptes_primary)[j]); 942 943 /* 944 * Since we keep the evicted entries at the 945 * from of the PVO list, we need move this 946 * (now resident) PVO after the evicted 947 * entries. 948 */ 949 next_pvo = TAILQ_NEXT(pvo, pvo_olink); 950 951 /* 952 * If we don't have to move (either we were the 953 * last entry or the next entry was valid), 954 * don't change our position. Otherwise 955 * move ourselves to the tail of the queue. 956 */ 957 if (next_pvo != NULL && 958 !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) { 959 TAILQ_REMOVE(pvoh, pvo, pvo_olink); 960 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 961 } 962 return 1; 963 } 964 source_pvo = pvo; 965 if (exec && !PVO_ISEXECUTABLE(source_pvo)) { 966 return 0; 967 } 968 if (victim_pvo != NULL) 969 break; 970 } 971 972 /* 973 * We also need the pvo entry of the victim we are replacing 974 * so save the R & C bits of the PTE. 975 */ 976 if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL && 977 pmap_pte_compare(pt, &pvo->pvo_pte)) { 978 vpvoh = pvoh; /* *1* */ 979 victim_pvo = pvo; 980 if (source_pvo != NULL) 981 break; 982 } 983 } 984 985 if (source_pvo == NULL) { 986 PMAPCOUNT(ptes_unspilled); 987 return 0; 988 } 989 990 if (victim_pvo == NULL) { 991 if ((pt->pte_hi & PTE_HID) == 0) 992 panic("pmap_pte_spill: victim p-pte (%p) has " 993 "no pvo entry!", pt); 994 995 /* 996 * If this is a secondary PTE, we need to search 997 * its primary pvo bucket for the matching PVO. 998 */ 999 vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; /* *2* */ 1000 TAILQ_FOREACH(pvo, vpvoh, pvo_olink) { 1001 PMAP_PVO_CHECK(pvo); /* sanity check */ 1002 1003 /* 1004 * We also need the pvo entry of the victim we are 1005 * replacing so save the R & C bits of the PTE. 1006 */ 1007 if (pmap_pte_compare(pt, &pvo->pvo_pte)) { 1008 victim_pvo = pvo; 1009 break; 1010 } 1011 } 1012 if (victim_pvo == NULL) 1013 panic("pmap_pte_spill: victim s-pte (%p) has " 1014 "no pvo entry!", pt); 1015 } 1016 1017 /* 1018 * The victim should be not be a kernel PVO/PTE entry. 1019 */ 1020 KASSERT(victim_pvo->pvo_pmap != pmap_kernel()); 1021 KASSERT(PVO_PTEGIDX_ISSET(victim_pvo)); 1022 KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i); 1023 1024 /* 1025 * We are invalidating the TLB entry for the EA for the 1026 * we are replacing even though its valid; If we don't 1027 * we lose any ref/chg bit changes contained in the TLB 1028 * entry. 1029 */ 1030 source_pvo->pvo_pte.pte_hi &= ~PTE_HID; 1031 1032 /* 1033 * To enforce the PVO list ordering constraint that all 1034 * evicted entries should come before all valid entries, 1035 * move the source PVO to the tail of its list and the 1036 * victim PVO to the head of its list (which might not be 1037 * the same list, if the victim was using the secondary hash). 1038 */ 1039 TAILQ_REMOVE(pvoh, source_pvo, pvo_olink); 1040 TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink); 1041 TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink); 1042 TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink); 1043 pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr); 1044 pmap_pte_set(pt, &source_pvo->pvo_pte); 1045 victim_pvo->pvo_pmap->pm_evictions++; 1046 source_pvo->pvo_pmap->pm_evictions--; 1047 PVO_WHERE(victim_pvo, SPILL_UNSET); 1048 PVO_WHERE(source_pvo, SPILL_SET); 1049 1050 PVO_PTEGIDX_CLR(victim_pvo); 1051 PVO_PTEGIDX_SET(source_pvo, i); 1052 PMAPCOUNT2(pmap_evcnt_ptes_primary[i]); 1053 PMAPCOUNT(ptes_spilled); 1054 PMAPCOUNT(ptes_evicted); 1055 PMAPCOUNT(ptes_removed); 1056 1057 PMAP_PVO_CHECK(victim_pvo); 1058 PMAP_PVO_CHECK(source_pvo); 1059 return 1; 1060 } 1061 1062 /* 1063 * Restrict given range to physical memory 1064 */ 1065 void 1066 pmap_real_memory(paddr_t *start, psize_t *size) 1067 { 1068 struct mem_region *mp; 1069 1070 for (mp = mem; mp->size; mp++) { 1071 if (*start + *size > mp->start 1072 && *start < mp->start + mp->size) { 1073 if (*start < mp->start) { 1074 *size -= mp->start - *start; 1075 *start = mp->start; 1076 } 1077 if (*start + *size > mp->start + mp->size) 1078 *size = mp->start + mp->size - *start; 1079 return; 1080 } 1081 } 1082 *size = 0; 1083 } 1084 1085 /* 1086 * Initialize anything else for pmap handling. 1087 * Called during vm_init(). 1088 */ 1089 void 1090 pmap_init(void) 1091 { 1092 #ifdef __HAVE_PMAP_PHYSSEG 1093 struct pvo_tqhead *pvoh; 1094 int bank; 1095 long sz; 1096 char *attr; 1097 1098 pvoh = pmap_physseg.pvoh; 1099 attr = pmap_physseg.attrs; 1100 for (bank = 0; bank < vm_nphysseg; bank++) { 1101 sz = vm_physmem[bank].end - vm_physmem[bank].start; 1102 vm_physmem[bank].pmseg.pvoh = pvoh; 1103 vm_physmem[bank].pmseg.attrs = attr; 1104 for (; sz > 0; sz--, pvoh++, attr++) { 1105 TAILQ_INIT(pvoh); 1106 *attr = 0; 1107 } 1108 } 1109 #endif 1110 1111 pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry), 1112 sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl", 1113 &pmap_pool_mallocator); 1114 1115 pool_setlowat(&pmap_mpvo_pool, 1008); 1116 1117 pmap_initialized = 1; 1118 1119 } 1120 1121 /* 1122 * How much virtual space does the kernel get? 1123 */ 1124 void 1125 pmap_virtual_space(vaddr_t *start, vaddr_t *end) 1126 { 1127 /* 1128 * For now, reserve one segment (minus some overhead) for kernel 1129 * virtual memory 1130 */ 1131 *start = VM_MIN_KERNEL_ADDRESS; 1132 *end = VM_MAX_KERNEL_ADDRESS; 1133 } 1134 1135 /* 1136 * Allocate, initialize, and return a new physical map. 1137 */ 1138 pmap_t 1139 pmap_create(void) 1140 { 1141 pmap_t pm; 1142 1143 pm = pool_get(&pmap_pool, PR_WAITOK); 1144 memset((caddr_t)pm, 0, sizeof *pm); 1145 pmap_pinit(pm); 1146 1147 DPRINTFN(CREATE,("pmap_create: pm %p:\n" 1148 "\t%06x %06x %06x %06x %06x %06x %06x %06x\n" 1149 "\t%06x %06x %06x %06x %06x %06x %06x %06x\n", pm, 1150 (unsigned int) pm->pm_sr[0], (unsigned int) pm->pm_sr[1], 1151 (unsigned int) pm->pm_sr[2], (unsigned int) pm->pm_sr[3], 1152 (unsigned int) pm->pm_sr[4], (unsigned int) pm->pm_sr[5], 1153 (unsigned int) pm->pm_sr[6], (unsigned int) pm->pm_sr[7], 1154 (unsigned int) pm->pm_sr[8], (unsigned int) pm->pm_sr[9], 1155 (unsigned int) pm->pm_sr[10], (unsigned int) pm->pm_sr[11], 1156 (unsigned int) pm->pm_sr[12], (unsigned int) pm->pm_sr[13], 1157 (unsigned int) pm->pm_sr[14], (unsigned int) pm->pm_sr[15])); 1158 return pm; 1159 } 1160 1161 /* 1162 * Initialize a preallocated and zeroed pmap structure. 1163 */ 1164 void 1165 pmap_pinit(pmap_t pm) 1166 { 1167 register_t entropy = MFTB(); 1168 register_t mask; 1169 int i; 1170 1171 /* 1172 * Allocate some segment registers for this pmap. 1173 */ 1174 pm->pm_refs = 1; 1175 for (i = 0; i < NPMAPS; i += VSID_NBPW) { 1176 static register_t pmap_vsidcontext; 1177 register_t hash; 1178 unsigned int n; 1179 1180 /* Create a new value by multiplying by a prime adding in 1181 * entropy from the timebase register. This is to make the 1182 * VSID more random so that the PT Hash function collides 1183 * less often. (note that the prime causes gcc to do shifts 1184 * instead of a multiply) 1185 */ 1186 pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy; 1187 hash = pmap_vsidcontext & (NPMAPS - 1); 1188 if (hash == 0) { /* 0 is special, avoid it */ 1189 entropy += 0xbadf00d; 1190 continue; 1191 } 1192 n = hash >> 5; 1193 mask = 1L << (hash & (VSID_NBPW-1)); 1194 hash = pmap_vsidcontext; 1195 if (pmap_vsid_bitmap[n] & mask) { /* collision? */ 1196 /* anything free in this bucket? */ 1197 if (~pmap_vsid_bitmap[n] == 0) { 1198 entropy = hash ^ (hash >> 16); 1199 continue; 1200 } 1201 i = ffs(~pmap_vsid_bitmap[n]) - 1; 1202 mask = 1L << i; 1203 hash &= ~(VSID_NBPW-1); 1204 hash |= i; 1205 } 1206 hash &= PTE_VSID >> PTE_VSID_SHFT; 1207 pmap_vsid_bitmap[n] |= mask; 1208 pm->pm_vsid = hash; 1209 #ifndef PPC_OEA64 1210 for (i = 0; i < 16; i++) 1211 pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY | 1212 SR_NOEXEC; 1213 #endif 1214 return; 1215 } 1216 panic("pmap_pinit: out of segments"); 1217 } 1218 1219 /* 1220 * Add a reference to the given pmap. 1221 */ 1222 void 1223 pmap_reference(pmap_t pm) 1224 { 1225 pm->pm_refs++; 1226 } 1227 1228 /* 1229 * Retire the given pmap from service. 1230 * Should only be called if the map contains no valid mappings. 1231 */ 1232 void 1233 pmap_destroy(pmap_t pm) 1234 { 1235 if (--pm->pm_refs == 0) { 1236 pmap_release(pm); 1237 pool_put(&pmap_pool, pm); 1238 } 1239 } 1240 1241 /* 1242 * Release any resources held by the given physical map. 1243 * Called when a pmap initialized by pmap_pinit is being released. 1244 */ 1245 void 1246 pmap_release(pmap_t pm) 1247 { 1248 int idx, mask; 1249 1250 if (pm->pm_sr[0] == 0) 1251 panic("pmap_release"); 1252 idx = pm->pm_vsid & (NPMAPS-1); 1253 mask = 1 << (idx % VSID_NBPW); 1254 idx /= VSID_NBPW; 1255 1256 KASSERT(pmap_vsid_bitmap[idx] & mask); 1257 pmap_vsid_bitmap[idx] &= ~mask; 1258 } 1259 1260 /* 1261 * Copy the range specified by src_addr/len 1262 * from the source map to the range dst_addr/len 1263 * in the destination map. 1264 * 1265 * This routine is only advisory and need not do anything. 1266 */ 1267 void 1268 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr, 1269 vsize_t len, vaddr_t src_addr) 1270 { 1271 PMAPCOUNT(copies); 1272 } 1273 1274 /* 1275 * Require that all active physical maps contain no 1276 * incorrect entries NOW. 1277 */ 1278 void 1279 pmap_update(struct pmap *pmap) 1280 { 1281 PMAPCOUNT(updates); 1282 TLBSYNC(); 1283 } 1284 1285 /* 1286 * Garbage collects the physical map system for 1287 * pages which are no longer used. 1288 * Success need not be guaranteed -- that is, there 1289 * may well be pages which are not referenced, but 1290 * others may be collected. 1291 * Called by the pageout daemon when pages are scarce. 1292 */ 1293 void 1294 pmap_collect(pmap_t pm) 1295 { 1296 PMAPCOUNT(collects); 1297 } 1298 1299 static inline int 1300 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx) 1301 { 1302 int pteidx; 1303 /* 1304 * We can find the actual pte entry without searching by 1305 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9] 1306 * and by noticing the HID bit. 1307 */ 1308 pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo); 1309 if (pvo->pvo_pte.pte_hi & PTE_HID) 1310 pteidx ^= pmap_pteg_mask * 8; 1311 return pteidx; 1312 } 1313 1314 volatile struct pte * 1315 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx) 1316 { 1317 volatile struct pte *pt; 1318 1319 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1320 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) 1321 return NULL; 1322 #endif 1323 1324 /* 1325 * If we haven't been supplied the ptegidx, calculate it. 1326 */ 1327 if (pteidx == -1) { 1328 int ptegidx; 1329 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1330 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1331 } 1332 1333 pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7]; 1334 1335 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1336 return pt; 1337 #else 1338 if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) { 1339 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1340 "pvo but no valid pte index", pvo); 1341 } 1342 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) { 1343 panic("pmap_pvo_to_pte: pvo %p: has valid pte index in " 1344 "pvo but no valid pte", pvo); 1345 } 1346 1347 if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) { 1348 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) { 1349 #if defined(DEBUG) || defined(PMAPCHECK) 1350 pmap_pte_print(pt); 1351 #endif 1352 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1353 "pmap_pteg_table %p but invalid in pvo", 1354 pvo, pt); 1355 } 1356 if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) { 1357 #if defined(DEBUG) || defined(PMAPCHECK) 1358 pmap_pte_print(pt); 1359 #endif 1360 panic("pmap_pvo_to_pte: pvo %p: pvo pte does " 1361 "not match pte %p in pmap_pteg_table", 1362 pvo, pt); 1363 } 1364 return pt; 1365 } 1366 1367 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1368 #if defined(DEBUG) || defined(PMAPCHECK) 1369 pmap_pte_print(pt); 1370 #endif 1371 panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in " 1372 "pmap_pteg_table but valid in pvo", pvo, pt); 1373 } 1374 return NULL; 1375 #endif /* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */ 1376 } 1377 1378 struct pvo_entry * 1379 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p) 1380 { 1381 struct pvo_entry *pvo; 1382 int ptegidx; 1383 1384 va &= ~ADDR_POFF; 1385 ptegidx = va_to_pteg(pm, va); 1386 1387 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1388 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1389 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 1390 panic("pmap_pvo_find_va: invalid pvo %p on " 1391 "list %#x (%p)", pvo, ptegidx, 1392 &pmap_pvo_table[ptegidx]); 1393 #endif 1394 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1395 if (pteidx_p) 1396 *pteidx_p = pmap_pvo_pte_index(pvo, ptegidx); 1397 return pvo; 1398 } 1399 } 1400 return NULL; 1401 } 1402 1403 #if defined(DEBUG) || defined(PMAPCHECK) 1404 void 1405 pmap_pvo_check(const struct pvo_entry *pvo) 1406 { 1407 struct pvo_head *pvo_head; 1408 struct pvo_entry *pvo0; 1409 volatile struct pte *pt; 1410 int failed = 0; 1411 1412 if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH) 1413 panic("pmap_pvo_check: pvo %p: invalid address", pvo); 1414 1415 if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) { 1416 printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n", 1417 pvo, pvo->pvo_pmap); 1418 failed = 1; 1419 } 1420 1421 if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH || 1422 (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) { 1423 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1424 pvo, TAILQ_NEXT(pvo, pvo_olink)); 1425 failed = 1; 1426 } 1427 1428 if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH || 1429 (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) { 1430 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1431 pvo, LIST_NEXT(pvo, pvo_vlink)); 1432 failed = 1; 1433 } 1434 1435 if (pvo->pvo_vaddr & PVO_MANAGED) { 1436 pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL); 1437 } else { 1438 if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) { 1439 printf("pmap_pvo_check: pvo %p: non kernel address " 1440 "on kernel unmanaged list\n", pvo); 1441 failed = 1; 1442 } 1443 pvo_head = &pmap_pvo_kunmanaged; 1444 } 1445 LIST_FOREACH(pvo0, pvo_head, pvo_vlink) { 1446 if (pvo0 == pvo) 1447 break; 1448 } 1449 if (pvo0 == NULL) { 1450 printf("pmap_pvo_check: pvo %p: not present " 1451 "on its vlist head %p\n", pvo, pvo_head); 1452 failed = 1; 1453 } 1454 if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) { 1455 printf("pmap_pvo_check: pvo %p: not present " 1456 "on its olist head\n", pvo); 1457 failed = 1; 1458 } 1459 pt = pmap_pvo_to_pte(pvo, -1); 1460 if (pt == NULL) { 1461 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1462 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1463 "no PTE\n", pvo); 1464 failed = 1; 1465 } 1466 } else { 1467 if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] || 1468 (uintptr_t) pt >= 1469 (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) { 1470 printf("pmap_pvo_check: pvo %p: pte %p not in " 1471 "pteg table\n", pvo, pt); 1472 failed = 1; 1473 } 1474 if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) { 1475 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1476 "no PTE\n", pvo); 1477 failed = 1; 1478 } 1479 if (pvo->pvo_pte.pte_hi != pt->pte_hi) { 1480 printf("pmap_pvo_check: pvo %p: pte_hi differ: " 1481 "%#x/%#x\n", pvo, (unsigned int) pvo->pvo_pte.pte_hi, (unsigned int) pt->pte_hi); 1482 failed = 1; 1483 } 1484 if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) & 1485 (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) { 1486 printf("pmap_pvo_check: pvo %p: pte_lo differ: " 1487 "%#x/%#x\n", pvo, 1488 (unsigned int) (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)), 1489 (unsigned int) (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN))); 1490 failed = 1; 1491 } 1492 if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) { 1493 printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#lx" 1494 " doesn't not match PVO's VA %#lx\n", 1495 pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo)); 1496 failed = 1; 1497 } 1498 if (failed) 1499 pmap_pte_print(pt); 1500 } 1501 if (failed) 1502 panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo, 1503 pvo->pvo_pmap); 1504 } 1505 #endif /* DEBUG || PMAPCHECK */ 1506 1507 /* 1508 * Search the PVO table looking for a non-wired entry. 1509 * If we find one, remove it and return it. 1510 */ 1511 1512 struct pvo_entry * 1513 pmap_pvo_reclaim(struct pmap *pm) 1514 { 1515 struct pvo_tqhead *pvoh; 1516 struct pvo_entry *pvo; 1517 uint32_t idx, endidx; 1518 1519 endidx = pmap_pvo_reclaim_nextidx; 1520 for (idx = (endidx + 1) & pmap_pteg_mask; idx != endidx; 1521 idx = (idx + 1) & pmap_pteg_mask) { 1522 pvoh = &pmap_pvo_table[idx]; 1523 TAILQ_FOREACH(pvo, pvoh, pvo_olink) { 1524 if ((pvo->pvo_vaddr & PVO_WIRED) == 0) { 1525 pmap_pvo_remove(pvo, -1, NULL); 1526 pmap_pvo_reclaim_nextidx = idx; 1527 PMAPCOUNT(pvos_reclaimed); 1528 return pvo; 1529 } 1530 } 1531 } 1532 return NULL; 1533 } 1534 1535 /* 1536 * This returns whether this is the first mapping of a page. 1537 */ 1538 int 1539 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head, 1540 vaddr_t va, paddr_t pa, register_t pte_lo, int flags) 1541 { 1542 struct pvo_entry *pvo; 1543 struct pvo_tqhead *pvoh; 1544 register_t msr; 1545 int ptegidx; 1546 int i; 1547 int poolflags = PR_NOWAIT; 1548 1549 /* 1550 * Compute the PTE Group index. 1551 */ 1552 va &= ~ADDR_POFF; 1553 ptegidx = va_to_pteg(pm, va); 1554 1555 msr = pmap_interrupts_off(); 1556 1557 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1558 if (pmap_pvo_remove_depth > 0) 1559 panic("pmap_pvo_enter: called while pmap_pvo_remove active!"); 1560 if (++pmap_pvo_enter_depth > 1) 1561 panic("pmap_pvo_enter: called recursively!"); 1562 #endif 1563 1564 /* 1565 * Remove any existing mapping for this page. Reuse the 1566 * pvo entry if there a mapping. 1567 */ 1568 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1569 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1570 #ifdef DEBUG 1571 if ((pmapdebug & PMAPDEBUG_PVOENTER) && 1572 ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) & 1573 ~(PTE_REF|PTE_CHG)) == 0 && 1574 va < VM_MIN_KERNEL_ADDRESS) { 1575 printf("pmap_pvo_enter: pvo %p: dup %#x/%#lx\n", 1576 pvo, (unsigned int) pvo->pvo_pte.pte_lo, (unsigned int) pte_lo|pa); 1577 printf("pmap_pvo_enter: pte_hi=%#x sr=%#x\n", 1578 (unsigned int) pvo->pvo_pte.pte_hi, 1579 (unsigned int) pm->pm_sr[va >> ADDR_SR_SHFT]); 1580 pmap_pte_print(pmap_pvo_to_pte(pvo, -1)); 1581 #ifdef DDBX 1582 Debugger(); 1583 #endif 1584 } 1585 #endif 1586 PMAPCOUNT(mappings_replaced); 1587 pmap_pvo_remove(pvo, -1, NULL); 1588 break; 1589 } 1590 } 1591 1592 /* 1593 * If we aren't overwriting an mapping, try to allocate 1594 */ 1595 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1596 --pmap_pvo_enter_depth; 1597 #endif 1598 pmap_interrupts_restore(msr); 1599 if (pvo) { 1600 pmap_pvo_free(pvo); 1601 } 1602 pvo = pool_get(pl, poolflags); 1603 1604 #ifdef DEBUG 1605 /* 1606 * Exercise pmap_pvo_reclaim() a little. 1607 */ 1608 if (pvo && (flags & PMAP_CANFAIL) != 0 && 1609 pmap_pvo_reclaim_debugctr++ > 0x1000 && 1610 (pmap_pvo_reclaim_debugctr & 0xff) == 0) { 1611 pool_put(pl, pvo); 1612 pvo = NULL; 1613 } 1614 #endif 1615 1616 msr = pmap_interrupts_off(); 1617 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1618 ++pmap_pvo_enter_depth; 1619 #endif 1620 if (pvo == NULL) { 1621 pvo = pmap_pvo_reclaim(pm); 1622 if (pvo == NULL) { 1623 if ((flags & PMAP_CANFAIL) == 0) 1624 panic("pmap_pvo_enter: failed"); 1625 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1626 pmap_pvo_enter_depth--; 1627 #endif 1628 PMAPCOUNT(pvos_failed); 1629 pmap_interrupts_restore(msr); 1630 return ENOMEM; 1631 } 1632 } 1633 1634 pvo->pvo_vaddr = va; 1635 pvo->pvo_pmap = pm; 1636 pvo->pvo_vaddr &= ~ADDR_POFF; 1637 if (flags & VM_PROT_EXECUTE) { 1638 PMAPCOUNT(exec_mappings); 1639 pvo_set_exec(pvo); 1640 } 1641 if (flags & PMAP_WIRED) 1642 pvo->pvo_vaddr |= PVO_WIRED; 1643 if (pvo_head != &pmap_pvo_kunmanaged) { 1644 pvo->pvo_vaddr |= PVO_MANAGED; 1645 PMAPCOUNT(mappings); 1646 } else { 1647 PMAPCOUNT(kernel_mappings); 1648 } 1649 pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo); 1650 1651 LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink); 1652 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1653 pvo->pvo_pmap->pm_stats.wired_count++; 1654 pvo->pvo_pmap->pm_stats.resident_count++; 1655 #if defined(DEBUG) 1656 if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) 1657 DPRINTFN(PVOENTER, 1658 ("pmap_pvo_enter: pvo %p: pm %p va %#lx pa %#lx\n", 1659 pvo, pm, va, pa)); 1660 #endif 1661 1662 /* 1663 * We hope this succeeds but it isn't required. 1664 */ 1665 pvoh = &pmap_pvo_table[ptegidx]; 1666 i = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 1667 if (i >= 0) { 1668 PVO_PTEGIDX_SET(pvo, i); 1669 PVO_WHERE(pvo, ENTER_INSERT); 1670 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 1671 ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]); 1672 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 1673 } else { 1674 /* 1675 * Since we didn't have room for this entry (which makes it 1676 * and evicted entry), place it at the head of the list. 1677 */ 1678 TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink); 1679 PMAPCOUNT(ptes_evicted); 1680 pm->pm_evictions++; 1681 /* 1682 * If this is a kernel page, make sure it's active. 1683 */ 1684 if (pm == pmap_kernel()) { 1685 i = pmap_pte_spill(pm, va, FALSE); 1686 KASSERT(i); 1687 } 1688 } 1689 PMAP_PVO_CHECK(pvo); /* sanity check */ 1690 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1691 pmap_pvo_enter_depth--; 1692 #endif 1693 pmap_interrupts_restore(msr); 1694 return 0; 1695 } 1696 1697 void 1698 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx, struct pvo_head *pvol) 1699 { 1700 volatile struct pte *pt; 1701 int ptegidx; 1702 1703 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1704 if (++pmap_pvo_remove_depth > 1) 1705 panic("pmap_pvo_remove: called recursively!"); 1706 #endif 1707 1708 /* 1709 * If we haven't been supplied the ptegidx, calculate it. 1710 */ 1711 if (pteidx == -1) { 1712 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1713 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1714 } else { 1715 ptegidx = pteidx >> 3; 1716 if (pvo->pvo_pte.pte_hi & PTE_HID) 1717 ptegidx ^= pmap_pteg_mask; 1718 } 1719 PMAP_PVO_CHECK(pvo); /* sanity check */ 1720 1721 /* 1722 * If there is an active pte entry, we need to deactivate it 1723 * (and save the ref & chg bits). 1724 */ 1725 pt = pmap_pvo_to_pte(pvo, pteidx); 1726 if (pt != NULL) { 1727 pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 1728 PVO_WHERE(pvo, REMOVE); 1729 PVO_PTEGIDX_CLR(pvo); 1730 PMAPCOUNT(ptes_removed); 1731 } else { 1732 KASSERT(pvo->pvo_pmap->pm_evictions > 0); 1733 pvo->pvo_pmap->pm_evictions--; 1734 } 1735 1736 /* 1737 * Account for executable mappings. 1738 */ 1739 if (PVO_ISEXECUTABLE(pvo)) 1740 pvo_clear_exec(pvo); 1741 1742 /* 1743 * Update our statistics. 1744 */ 1745 pvo->pvo_pmap->pm_stats.resident_count--; 1746 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1747 pvo->pvo_pmap->pm_stats.wired_count--; 1748 1749 /* 1750 * Save the REF/CHG bits into their cache if the page is managed. 1751 */ 1752 if (pvo->pvo_vaddr & PVO_MANAGED) { 1753 register_t ptelo = pvo->pvo_pte.pte_lo; 1754 struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN); 1755 1756 if (pg != NULL) { 1757 pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG)); 1758 } 1759 PMAPCOUNT(unmappings); 1760 } else { 1761 PMAPCOUNT(kernel_unmappings); 1762 } 1763 1764 /* 1765 * Remove the PVO from its lists and return it to the pool. 1766 */ 1767 LIST_REMOVE(pvo, pvo_vlink); 1768 TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink); 1769 if (pvol) { 1770 LIST_INSERT_HEAD(pvol, pvo, pvo_vlink); 1771 } 1772 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1773 pmap_pvo_remove_depth--; 1774 #endif 1775 } 1776 1777 void 1778 pmap_pvo_free(struct pvo_entry *pvo) 1779 { 1780 1781 pool_put(pvo->pvo_vaddr & PVO_MANAGED ? &pmap_mpvo_pool : 1782 &pmap_upvo_pool, pvo); 1783 } 1784 1785 void 1786 pmap_pvo_free_list(struct pvo_head *pvol) 1787 { 1788 struct pvo_entry *pvo, *npvo; 1789 1790 for (pvo = LIST_FIRST(pvol); pvo != NULL; pvo = npvo) { 1791 npvo = LIST_NEXT(pvo, pvo_vlink); 1792 LIST_REMOVE(pvo, pvo_vlink); 1793 pmap_pvo_free(pvo); 1794 } 1795 } 1796 1797 /* 1798 * Mark a mapping as executable. 1799 * If this is the first executable mapping in the segment, 1800 * clear the noexec flag. 1801 */ 1802 STATIC void 1803 pvo_set_exec(struct pvo_entry *pvo) 1804 { 1805 struct pmap *pm = pvo->pvo_pmap; 1806 1807 if (pm == pmap_kernel() || PVO_ISEXECUTABLE(pvo)) { 1808 return; 1809 } 1810 pvo->pvo_vaddr |= PVO_EXECUTABLE; 1811 #ifdef PPC_OEA 1812 { 1813 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1814 if (pm->pm_exec[sr]++ == 0) { 1815 pm->pm_sr[sr] &= ~SR_NOEXEC; 1816 } 1817 } 1818 #endif 1819 } 1820 1821 /* 1822 * Mark a mapping as non-executable. 1823 * If this was the last executable mapping in the segment, 1824 * set the noexec flag. 1825 */ 1826 STATIC void 1827 pvo_clear_exec(struct pvo_entry *pvo) 1828 { 1829 struct pmap *pm = pvo->pvo_pmap; 1830 1831 if (pm == pmap_kernel() || !PVO_ISEXECUTABLE(pvo)) { 1832 return; 1833 } 1834 pvo->pvo_vaddr &= ~PVO_EXECUTABLE; 1835 #ifdef PPC_OEA 1836 { 1837 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1838 if (--pm->pm_exec[sr] == 0) { 1839 pm->pm_sr[sr] |= SR_NOEXEC; 1840 } 1841 } 1842 #endif 1843 } 1844 1845 /* 1846 * Insert physical page at pa into the given pmap at virtual address va. 1847 */ 1848 int 1849 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags) 1850 { 1851 struct mem_region *mp; 1852 struct pvo_head *pvo_head; 1853 struct vm_page *pg; 1854 struct pool *pl; 1855 register_t pte_lo; 1856 int error; 1857 u_int pvo_flags; 1858 u_int was_exec = 0; 1859 1860 if (__predict_false(!pmap_initialized)) { 1861 pvo_head = &pmap_pvo_kunmanaged; 1862 pl = &pmap_upvo_pool; 1863 pvo_flags = 0; 1864 pg = NULL; 1865 was_exec = PTE_EXEC; 1866 } else { 1867 pvo_head = pa_to_pvoh(pa, &pg); 1868 pl = &pmap_mpvo_pool; 1869 pvo_flags = PVO_MANAGED; 1870 } 1871 1872 DPRINTFN(ENTER, 1873 ("pmap_enter(%p, 0x%lx, 0x%lx, 0x%x, 0x%x):", 1874 pm, va, pa, prot, flags)); 1875 1876 /* 1877 * If this is a managed page, and it's the first reference to the 1878 * page clear the execness of the page. Otherwise fetch the execness. 1879 */ 1880 if (pg != NULL) 1881 was_exec = pmap_attr_fetch(pg) & PTE_EXEC; 1882 1883 DPRINTFN(ENTER, (" was_exec=%d", was_exec)); 1884 1885 /* 1886 * Assume the page is cache inhibited and access is guarded unless 1887 * it's in our available memory array. If it is in the memory array, 1888 * asssume it's in memory coherent memory. 1889 */ 1890 pte_lo = PTE_IG; 1891 if ((flags & PMAP_NC) == 0) { 1892 for (mp = mem; mp->size; mp++) { 1893 if (pa >= mp->start && pa < mp->start + mp->size) { 1894 pte_lo = PTE_M; 1895 break; 1896 } 1897 } 1898 } 1899 1900 if (prot & VM_PROT_WRITE) 1901 pte_lo |= PTE_BW; 1902 else 1903 pte_lo |= PTE_BR; 1904 1905 /* 1906 * If this was in response to a fault, "pre-fault" the PTE's 1907 * changed/referenced bit appropriately. 1908 */ 1909 if (flags & VM_PROT_WRITE) 1910 pte_lo |= PTE_CHG; 1911 if (flags & VM_PROT_ALL) 1912 pte_lo |= PTE_REF; 1913 1914 /* 1915 * We need to know if this page can be executable 1916 */ 1917 flags |= (prot & VM_PROT_EXECUTE); 1918 1919 /* 1920 * Record mapping for later back-translation and pte spilling. 1921 * This will overwrite any existing mapping. 1922 */ 1923 error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags); 1924 1925 /* 1926 * Flush the real page from the instruction cache if this page is 1927 * mapped executable and cacheable and has not been flushed since 1928 * the last time it was modified. 1929 */ 1930 if (error == 0 && 1931 (flags & VM_PROT_EXECUTE) && 1932 (pte_lo & PTE_I) == 0 && 1933 was_exec == 0) { 1934 DPRINTFN(ENTER, (" syncicache")); 1935 PMAPCOUNT(exec_synced); 1936 pmap_syncicache(pa, PAGE_SIZE); 1937 if (pg != NULL) { 1938 pmap_attr_save(pg, PTE_EXEC); 1939 PMAPCOUNT(exec_cached); 1940 #if defined(DEBUG) || defined(PMAPDEBUG) 1941 if (pmapdebug & PMAPDEBUG_ENTER) 1942 printf(" marked-as-exec"); 1943 else if (pmapdebug & PMAPDEBUG_EXEC) 1944 printf("[pmap_enter: %#lx: marked-as-exec]\n", 1945 VM_PAGE_TO_PHYS(pg)); 1946 1947 #endif 1948 } 1949 } 1950 1951 DPRINTFN(ENTER, (": error=%d\n", error)); 1952 1953 return error; 1954 } 1955 1956 void 1957 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot) 1958 { 1959 struct mem_region *mp; 1960 register_t pte_lo; 1961 int error; 1962 1963 if (va < VM_MIN_KERNEL_ADDRESS) 1964 panic("pmap_kenter_pa: attempt to enter " 1965 "non-kernel address %#lx!", va); 1966 1967 DPRINTFN(KENTER, 1968 ("pmap_kenter_pa(%#lx,%#lx,%#x)\n", va, pa, prot)); 1969 1970 /* 1971 * Assume the page is cache inhibited and access is guarded unless 1972 * it's in our available memory array. If it is in the memory array, 1973 * asssume it's in memory coherent memory. 1974 */ 1975 pte_lo = PTE_IG; 1976 if ((prot & PMAP_NC) == 0) { 1977 for (mp = mem; mp->size; mp++) { 1978 if (pa >= mp->start && pa < mp->start + mp->size) { 1979 pte_lo = PTE_M; 1980 break; 1981 } 1982 } 1983 } 1984 1985 if (prot & VM_PROT_WRITE) 1986 pte_lo |= PTE_BW; 1987 else 1988 pte_lo |= PTE_BR; 1989 1990 /* 1991 * We don't care about REF/CHG on PVOs on the unmanaged list. 1992 */ 1993 error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool, 1994 &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED); 1995 1996 if (error != 0) 1997 panic("pmap_kenter_pa: failed to enter va %#lx pa %#lx: %d", 1998 va, pa, error); 1999 } 2000 2001 void 2002 pmap_kremove(vaddr_t va, vsize_t len) 2003 { 2004 if (va < VM_MIN_KERNEL_ADDRESS) 2005 panic("pmap_kremove: attempt to remove " 2006 "non-kernel address %#lx!", va); 2007 2008 DPRINTFN(KREMOVE,("pmap_kremove(%#lx,%#lx)\n", va, len)); 2009 pmap_remove(pmap_kernel(), va, va + len); 2010 } 2011 2012 /* 2013 * Remove the given range of mapping entries. 2014 */ 2015 void 2016 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva) 2017 { 2018 struct pvo_head pvol; 2019 struct pvo_entry *pvo; 2020 register_t msr; 2021 int pteidx; 2022 2023 LIST_INIT(&pvol); 2024 msr = pmap_interrupts_off(); 2025 for (; va < endva; va += PAGE_SIZE) { 2026 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2027 if (pvo != NULL) { 2028 pmap_pvo_remove(pvo, pteidx, &pvol); 2029 } 2030 } 2031 pmap_interrupts_restore(msr); 2032 pmap_pvo_free_list(&pvol); 2033 } 2034 2035 /* 2036 * Get the physical page address for the given pmap/virtual address. 2037 */ 2038 boolean_t 2039 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap) 2040 { 2041 struct pvo_entry *pvo; 2042 register_t msr; 2043 2044 /* 2045 * If this is a kernel pmap lookup, also check the battable 2046 * and if we get a hit, translate the VA to a PA using the 2047 * BAT entries. Don't check for VM_MAX_KENREL_ADDRESS is 2048 * that will wrap back to 0. 2049 */ 2050 if (pm == pmap_kernel() && 2051 (va < VM_MIN_KERNEL_ADDRESS || 2052 (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) { 2053 KASSERT((va >> ADDR_SR_SHFT) != USER_SR); 2054 if ((MFPVR() >> 16) != MPC601) { 2055 register_t batu = battable[va >> ADDR_SR_SHFT].batu; 2056 if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) { 2057 register_t batl = 2058 battable[va >> ADDR_SR_SHFT].batl; 2059 register_t mask = 2060 (~(batu & BAT_BL) << 15) & ~0x1ffffL; 2061 if (pap) 2062 *pap = (batl & mask) | (va & ~mask); 2063 return TRUE; 2064 } 2065 } else { 2066 register_t batu = battable[va >> 23].batu; 2067 register_t batl = battable[va >> 23].batl; 2068 register_t sr = iosrtable[va >> ADDR_SR_SHFT]; 2069 if (BAT601_VALID_P(batl) && 2070 BAT601_VA_MATCH_P(batu, batl, va)) { 2071 register_t mask = 2072 (~(batl & BAT601_BSM) << 17) & ~0x1ffffL; 2073 if (pap) 2074 *pap = (batl & mask) | (va & ~mask); 2075 return TRUE; 2076 } else if (SR601_VALID_P(sr) && 2077 SR601_PA_MATCH_P(sr, va)) { 2078 if (pap) 2079 *pap = va; 2080 return TRUE; 2081 } 2082 } 2083 return FALSE; 2084 } 2085 2086 msr = pmap_interrupts_off(); 2087 pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL); 2088 if (pvo != NULL) { 2089 PMAP_PVO_CHECK(pvo); /* sanity check */ 2090 if (pap) 2091 *pap = (pvo->pvo_pte.pte_lo & PTE_RPGN) 2092 | (va & ADDR_POFF); 2093 } 2094 pmap_interrupts_restore(msr); 2095 return pvo != NULL; 2096 } 2097 2098 /* 2099 * Lower the protection on the specified range of this pmap. 2100 */ 2101 void 2102 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot) 2103 { 2104 struct pvo_entry *pvo; 2105 volatile struct pte *pt; 2106 register_t msr; 2107 int pteidx; 2108 2109 /* 2110 * Since this routine only downgrades protection, we should 2111 * always be called with at least one bit not set. 2112 */ 2113 KASSERT(prot != VM_PROT_ALL); 2114 2115 /* 2116 * If there is no protection, this is equivalent to 2117 * remove the pmap from the pmap. 2118 */ 2119 if ((prot & VM_PROT_READ) == 0) { 2120 pmap_remove(pm, va, endva); 2121 return; 2122 } 2123 2124 msr = pmap_interrupts_off(); 2125 for (; va < endva; va += PAGE_SIZE) { 2126 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2127 if (pvo == NULL) 2128 continue; 2129 PMAP_PVO_CHECK(pvo); /* sanity check */ 2130 2131 /* 2132 * Revoke executable if asked to do so. 2133 */ 2134 if ((prot & VM_PROT_EXECUTE) == 0) 2135 pvo_clear_exec(pvo); 2136 2137 #if 0 2138 /* 2139 * If the page is already read-only, no change 2140 * needs to be made. 2141 */ 2142 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) 2143 continue; 2144 #endif 2145 /* 2146 * Grab the PTE pointer before we diddle with 2147 * the cached PTE copy. 2148 */ 2149 pt = pmap_pvo_to_pte(pvo, pteidx); 2150 /* 2151 * Change the protection of the page. 2152 */ 2153 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2154 pvo->pvo_pte.pte_lo |= PTE_BR; 2155 2156 /* 2157 * If the PVO is in the page table, update 2158 * that pte at well. 2159 */ 2160 if (pt != NULL) { 2161 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2162 PVO_WHERE(pvo, PMAP_PROTECT); 2163 PMAPCOUNT(ptes_changed); 2164 } 2165 2166 PMAP_PVO_CHECK(pvo); /* sanity check */ 2167 } 2168 pmap_interrupts_restore(msr); 2169 } 2170 2171 void 2172 pmap_unwire(pmap_t pm, vaddr_t va) 2173 { 2174 struct pvo_entry *pvo; 2175 register_t msr; 2176 2177 msr = pmap_interrupts_off(); 2178 pvo = pmap_pvo_find_va(pm, va, NULL); 2179 if (pvo != NULL) { 2180 if (pvo->pvo_vaddr & PVO_WIRED) { 2181 pvo->pvo_vaddr &= ~PVO_WIRED; 2182 pm->pm_stats.wired_count--; 2183 } 2184 PMAP_PVO_CHECK(pvo); /* sanity check */ 2185 } 2186 pmap_interrupts_restore(msr); 2187 } 2188 2189 /* 2190 * Lower the protection on the specified physical page. 2191 */ 2192 void 2193 pmap_page_protect(struct vm_page *pg, vm_prot_t prot) 2194 { 2195 struct pvo_head *pvo_head, pvol; 2196 struct pvo_entry *pvo, *next_pvo; 2197 volatile struct pte *pt; 2198 register_t msr; 2199 2200 KASSERT(prot != VM_PROT_ALL); 2201 LIST_INIT(&pvol); 2202 msr = pmap_interrupts_off(); 2203 2204 /* 2205 * When UVM reuses a page, it does a pmap_page_protect with 2206 * VM_PROT_NONE. At that point, we can clear the exec flag 2207 * since we know the page will have different contents. 2208 */ 2209 if ((prot & VM_PROT_READ) == 0) { 2210 DPRINTFN(EXEC, ("[pmap_page_protect: %#lx: clear-exec]\n", 2211 VM_PAGE_TO_PHYS(pg))); 2212 if (pmap_attr_fetch(pg) & PTE_EXEC) { 2213 PMAPCOUNT(exec_uncached_page_protect); 2214 pmap_attr_clear(pg, PTE_EXEC); 2215 } 2216 } 2217 2218 pvo_head = vm_page_to_pvoh(pg); 2219 for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) { 2220 next_pvo = LIST_NEXT(pvo, pvo_vlink); 2221 PMAP_PVO_CHECK(pvo); /* sanity check */ 2222 2223 /* 2224 * Downgrading to no mapping at all, we just remove the entry. 2225 */ 2226 if ((prot & VM_PROT_READ) == 0) { 2227 pmap_pvo_remove(pvo, -1, &pvol); 2228 continue; 2229 } 2230 2231 /* 2232 * If EXEC permission is being revoked, just clear the 2233 * flag in the PVO. 2234 */ 2235 if ((prot & VM_PROT_EXECUTE) == 0) 2236 pvo_clear_exec(pvo); 2237 2238 /* 2239 * If this entry is already RO, don't diddle with the 2240 * page table. 2241 */ 2242 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) { 2243 PMAP_PVO_CHECK(pvo); 2244 continue; 2245 } 2246 2247 /* 2248 * Grab the PTE before the we diddle the bits so 2249 * pvo_to_pte can verify the pte contents are as 2250 * expected. 2251 */ 2252 pt = pmap_pvo_to_pte(pvo, -1); 2253 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2254 pvo->pvo_pte.pte_lo |= PTE_BR; 2255 if (pt != NULL) { 2256 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2257 PVO_WHERE(pvo, PMAP_PAGE_PROTECT); 2258 PMAPCOUNT(ptes_changed); 2259 } 2260 PMAP_PVO_CHECK(pvo); /* sanity check */ 2261 } 2262 pmap_interrupts_restore(msr); 2263 pmap_pvo_free_list(&pvol); 2264 } 2265 2266 /* 2267 * Activate the address space for the specified process. If the process 2268 * is the current process, load the new MMU context. 2269 */ 2270 void 2271 pmap_activate(struct lwp *l) 2272 { 2273 struct pcb *pcb = &l->l_addr->u_pcb; 2274 pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap; 2275 2276 DPRINTFN(ACTIVATE, 2277 ("pmap_activate: lwp %p (curlwp %p)\n", l, curlwp)); 2278 2279 /* 2280 * XXX Normally performed in cpu_fork(). 2281 */ 2282 pcb->pcb_pm = pmap; 2283 2284 /* 2285 * In theory, the SR registers need only be valid on return 2286 * to user space wait to do them there. 2287 */ 2288 if (l == curlwp) { 2289 /* Store pointer to new current pmap. */ 2290 curpm = pmap; 2291 } 2292 } 2293 2294 /* 2295 * Deactivate the specified process's address space. 2296 */ 2297 void 2298 pmap_deactivate(struct lwp *l) 2299 { 2300 } 2301 2302 boolean_t 2303 pmap_query_bit(struct vm_page *pg, int ptebit) 2304 { 2305 struct pvo_entry *pvo; 2306 volatile struct pte *pt; 2307 register_t msr; 2308 2309 if (pmap_attr_fetch(pg) & ptebit) 2310 return TRUE; 2311 2312 msr = pmap_interrupts_off(); 2313 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2314 PMAP_PVO_CHECK(pvo); /* sanity check */ 2315 /* 2316 * See if we saved the bit off. If so cache, it and return 2317 * success. 2318 */ 2319 if (pvo->pvo_pte.pte_lo & ptebit) { 2320 pmap_attr_save(pg, ptebit); 2321 PMAP_PVO_CHECK(pvo); /* sanity check */ 2322 pmap_interrupts_restore(msr); 2323 return TRUE; 2324 } 2325 } 2326 /* 2327 * No luck, now go thru the hard part of looking at the ptes 2328 * themselves. Sync so any pending REF/CHG bits are flushed 2329 * to the PTEs. 2330 */ 2331 SYNC(); 2332 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2333 PMAP_PVO_CHECK(pvo); /* sanity check */ 2334 /* 2335 * See if this pvo have a valid PTE. If so, fetch the 2336 * REF/CHG bits from the valid PTE. If the appropriate 2337 * ptebit is set, cache, it and return success. 2338 */ 2339 pt = pmap_pvo_to_pte(pvo, -1); 2340 if (pt != NULL) { 2341 pmap_pte_synch(pt, &pvo->pvo_pte); 2342 if (pvo->pvo_pte.pte_lo & ptebit) { 2343 pmap_attr_save(pg, ptebit); 2344 PMAP_PVO_CHECK(pvo); /* sanity check */ 2345 pmap_interrupts_restore(msr); 2346 return TRUE; 2347 } 2348 } 2349 } 2350 pmap_interrupts_restore(msr); 2351 return FALSE; 2352 } 2353 2354 boolean_t 2355 pmap_clear_bit(struct vm_page *pg, int ptebit) 2356 { 2357 struct pvo_head *pvoh = vm_page_to_pvoh(pg); 2358 struct pvo_entry *pvo; 2359 volatile struct pte *pt; 2360 register_t msr; 2361 int rv = 0; 2362 2363 msr = pmap_interrupts_off(); 2364 2365 /* 2366 * Fetch the cache value 2367 */ 2368 rv |= pmap_attr_fetch(pg); 2369 2370 /* 2371 * Clear the cached value. 2372 */ 2373 pmap_attr_clear(pg, ptebit); 2374 2375 /* 2376 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we 2377 * can reset the right ones). Note that since the pvo entries and 2378 * list heads are accessed via BAT0 and are never placed in the 2379 * page table, we don't have to worry about further accesses setting 2380 * the REF/CHG bits. 2381 */ 2382 SYNC(); 2383 2384 /* 2385 * For each pvo entry, clear pvo's ptebit. If this pvo have a 2386 * valid PTE. If so, clear the ptebit from the valid PTE. 2387 */ 2388 LIST_FOREACH(pvo, pvoh, pvo_vlink) { 2389 PMAP_PVO_CHECK(pvo); /* sanity check */ 2390 pt = pmap_pvo_to_pte(pvo, -1); 2391 if (pt != NULL) { 2392 /* 2393 * Only sync the PTE if the bit we are looking 2394 * for is not already set. 2395 */ 2396 if ((pvo->pvo_pte.pte_lo & ptebit) == 0) 2397 pmap_pte_synch(pt, &pvo->pvo_pte); 2398 /* 2399 * If the bit we are looking for was already set, 2400 * clear that bit in the pte. 2401 */ 2402 if (pvo->pvo_pte.pte_lo & ptebit) 2403 pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit); 2404 } 2405 rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF); 2406 pvo->pvo_pte.pte_lo &= ~ptebit; 2407 PMAP_PVO_CHECK(pvo); /* sanity check */ 2408 } 2409 pmap_interrupts_restore(msr); 2410 2411 /* 2412 * If we are clearing the modify bit and this page was marked EXEC 2413 * and the user of the page thinks the page was modified, then we 2414 * need to clean it from the icache if it's mapped or clear the EXEC 2415 * bit if it's not mapped. The page itself might not have the CHG 2416 * bit set if the modification was done via DMA to the page. 2417 */ 2418 if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) { 2419 if (LIST_EMPTY(pvoh)) { 2420 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: clear-exec]\n", 2421 VM_PAGE_TO_PHYS(pg))); 2422 pmap_attr_clear(pg, PTE_EXEC); 2423 PMAPCOUNT(exec_uncached_clear_modify); 2424 } else { 2425 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: syncicache]\n", 2426 VM_PAGE_TO_PHYS(pg))); 2427 pmap_syncicache(VM_PAGE_TO_PHYS(pg), PAGE_SIZE); 2428 PMAPCOUNT(exec_synced_clear_modify); 2429 } 2430 } 2431 return (rv & ptebit) != 0; 2432 } 2433 2434 void 2435 pmap_procwr(struct proc *p, vaddr_t va, size_t len) 2436 { 2437 struct pvo_entry *pvo; 2438 size_t offset = va & ADDR_POFF; 2439 int s; 2440 2441 s = splvm(); 2442 while (len > 0) { 2443 size_t seglen = PAGE_SIZE - offset; 2444 if (seglen > len) 2445 seglen = len; 2446 pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL); 2447 if (pvo != NULL && PVO_ISEXECUTABLE(pvo)) { 2448 pmap_syncicache( 2449 (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen); 2450 PMAP_PVO_CHECK(pvo); 2451 } 2452 va += seglen; 2453 len -= seglen; 2454 offset = 0; 2455 } 2456 splx(s); 2457 } 2458 2459 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 2460 void 2461 pmap_pte_print(volatile struct pte *pt) 2462 { 2463 printf("PTE %p: ", pt); 2464 /* High word: */ 2465 printf("0x%08lx: [", pt->pte_hi); 2466 printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i'); 2467 printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-'); 2468 printf("0x%06lx 0x%02lx", 2469 (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT, 2470 pt->pte_hi & PTE_API); 2471 printf(" (va 0x%08lx)] ", pmap_pte_to_va(pt)); 2472 /* Low word: */ 2473 printf(" 0x%08lx: [", pt->pte_lo); 2474 printf("0x%05lx... ", pt->pte_lo >> 12); 2475 printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u'); 2476 printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n'); 2477 printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.'); 2478 printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.'); 2479 printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.'); 2480 printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.'); 2481 switch (pt->pte_lo & PTE_PP) { 2482 case PTE_BR: printf("br]\n"); break; 2483 case PTE_BW: printf("bw]\n"); break; 2484 case PTE_SO: printf("so]\n"); break; 2485 case PTE_SW: printf("sw]\n"); break; 2486 } 2487 } 2488 #endif 2489 2490 #if defined(DDB) 2491 void 2492 pmap_pteg_check(void) 2493 { 2494 volatile struct pte *pt; 2495 int i; 2496 int ptegidx; 2497 u_int p_valid = 0; 2498 u_int s_valid = 0; 2499 u_int invalid = 0; 2500 2501 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2502 for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) { 2503 if (pt->pte_hi & PTE_VALID) { 2504 if (pt->pte_hi & PTE_HID) 2505 s_valid++; 2506 else 2507 p_valid++; 2508 } else 2509 invalid++; 2510 } 2511 } 2512 printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n", 2513 p_valid, p_valid, s_valid, s_valid, 2514 invalid, invalid); 2515 } 2516 2517 void 2518 pmap_print_mmuregs(void) 2519 { 2520 int i; 2521 u_int cpuvers; 2522 #ifndef PPC_OEA64 2523 vaddr_t addr; 2524 register_t soft_sr[16]; 2525 #endif 2526 struct bat soft_ibat[4]; 2527 struct bat soft_dbat[4]; 2528 register_t sdr1; 2529 2530 cpuvers = MFPVR() >> 16; 2531 __asm volatile ("mfsdr1 %0" : "=r"(sdr1)); 2532 #ifndef PPC_OEA64 2533 addr = 0; 2534 for (i = 0; i < 16; i++) { 2535 soft_sr[i] = MFSRIN(addr); 2536 addr += (1 << ADDR_SR_SHFT); 2537 } 2538 #endif 2539 2540 /* read iBAT (601: uBAT) registers */ 2541 __asm volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu)); 2542 __asm volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl)); 2543 __asm volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu)); 2544 __asm volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl)); 2545 __asm volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu)); 2546 __asm volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl)); 2547 __asm volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu)); 2548 __asm volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl)); 2549 2550 2551 if (cpuvers != MPC601) { 2552 /* read dBAT registers */ 2553 __asm volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu)); 2554 __asm volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl)); 2555 __asm volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu)); 2556 __asm volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl)); 2557 __asm volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu)); 2558 __asm volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl)); 2559 __asm volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu)); 2560 __asm volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl)); 2561 } 2562 2563 printf("SDR1:\t0x%lx\n", (long) sdr1); 2564 #ifndef PPC_OEA64 2565 printf("SR[]:\t"); 2566 for (i = 0; i < 4; i++) 2567 printf("0x%08lx, ", soft_sr[i]); 2568 printf("\n\t"); 2569 for ( ; i < 8; i++) 2570 printf("0x%08lx, ", soft_sr[i]); 2571 printf("\n\t"); 2572 for ( ; i < 12; i++) 2573 printf("0x%08lx, ", soft_sr[i]); 2574 printf("\n\t"); 2575 for ( ; i < 16; i++) 2576 printf("0x%08lx, ", soft_sr[i]); 2577 printf("\n"); 2578 #endif 2579 2580 printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i'); 2581 for (i = 0; i < 4; i++) { 2582 printf("0x%08lx 0x%08lx, ", 2583 soft_ibat[i].batu, soft_ibat[i].batl); 2584 if (i == 1) 2585 printf("\n\t"); 2586 } 2587 if (cpuvers != MPC601) { 2588 printf("\ndBAT[]:\t"); 2589 for (i = 0; i < 4; i++) { 2590 printf("0x%08lx 0x%08lx, ", 2591 soft_dbat[i].batu, soft_dbat[i].batl); 2592 if (i == 1) 2593 printf("\n\t"); 2594 } 2595 } 2596 printf("\n"); 2597 } 2598 2599 void 2600 pmap_print_pte(pmap_t pm, vaddr_t va) 2601 { 2602 struct pvo_entry *pvo; 2603 volatile struct pte *pt; 2604 int pteidx; 2605 2606 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2607 if (pvo != NULL) { 2608 pt = pmap_pvo_to_pte(pvo, pteidx); 2609 if (pt != NULL) { 2610 printf("VA %#lx -> %p -> %s %#lx, %#lx\n", 2611 va, pt, 2612 pt->pte_hi & PTE_HID ? "(sec)" : "(pri)", 2613 pt->pte_hi, pt->pte_lo); 2614 } else { 2615 printf("No valid PTE found\n"); 2616 } 2617 } else { 2618 printf("Address not in pmap\n"); 2619 } 2620 } 2621 2622 void 2623 pmap_pteg_dist(void) 2624 { 2625 struct pvo_entry *pvo; 2626 int ptegidx; 2627 int depth; 2628 int max_depth = 0; 2629 unsigned int depths[64]; 2630 2631 memset(depths, 0, sizeof(depths)); 2632 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2633 depth = 0; 2634 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2635 depth++; 2636 } 2637 if (depth > max_depth) 2638 max_depth = depth; 2639 if (depth > 63) 2640 depth = 63; 2641 depths[depth]++; 2642 } 2643 2644 for (depth = 0; depth < 64; depth++) { 2645 printf(" [%2d]: %8u", depth, depths[depth]); 2646 if ((depth & 3) == 3) 2647 printf("\n"); 2648 if (depth == max_depth) 2649 break; 2650 } 2651 if ((depth & 3) != 3) 2652 printf("\n"); 2653 printf("Max depth found was %d\n", max_depth); 2654 } 2655 #endif /* DEBUG */ 2656 2657 #if defined(PMAPCHECK) || defined(DEBUG) 2658 void 2659 pmap_pvo_verify(void) 2660 { 2661 int ptegidx; 2662 int s; 2663 2664 s = splvm(); 2665 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2666 struct pvo_entry *pvo; 2667 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2668 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 2669 panic("pmap_pvo_verify: invalid pvo %p " 2670 "on list %#x", pvo, ptegidx); 2671 pmap_pvo_check(pvo); 2672 } 2673 } 2674 splx(s); 2675 } 2676 #endif /* PMAPCHECK */ 2677 2678 2679 void * 2680 pmap_pool_ualloc(struct pool *pp, int flags) 2681 { 2682 struct pvo_page *pvop; 2683 2684 pvop = SIMPLEQ_FIRST(&pmap_upvop_head); 2685 if (pvop != NULL) { 2686 pmap_upvop_free--; 2687 SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link); 2688 return pvop; 2689 } 2690 if (uvm.page_init_done != TRUE) { 2691 return (void *) uvm_pageboot_alloc(PAGE_SIZE); 2692 } 2693 return pmap_pool_malloc(pp, flags); 2694 } 2695 2696 void * 2697 pmap_pool_malloc(struct pool *pp, int flags) 2698 { 2699 struct pvo_page *pvop; 2700 struct vm_page *pg; 2701 2702 pvop = SIMPLEQ_FIRST(&pmap_mpvop_head); 2703 if (pvop != NULL) { 2704 pmap_mpvop_free--; 2705 SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link); 2706 return pvop; 2707 } 2708 again: 2709 pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE, 2710 UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256); 2711 if (__predict_false(pg == NULL)) { 2712 if (flags & PR_WAITOK) { 2713 uvm_wait("plpg"); 2714 goto again; 2715 } else { 2716 return (0); 2717 } 2718 } 2719 return (void *) VM_PAGE_TO_PHYS(pg); 2720 } 2721 2722 void 2723 pmap_pool_ufree(struct pool *pp, void *va) 2724 { 2725 struct pvo_page *pvop; 2726 #if 0 2727 if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) { 2728 pmap_pool_mfree(va, size, tag); 2729 return; 2730 } 2731 #endif 2732 pvop = va; 2733 SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link); 2734 pmap_upvop_free++; 2735 if (pmap_upvop_free > pmap_upvop_maxfree) 2736 pmap_upvop_maxfree = pmap_upvop_free; 2737 } 2738 2739 void 2740 pmap_pool_mfree(struct pool *pp, void *va) 2741 { 2742 struct pvo_page *pvop; 2743 2744 pvop = va; 2745 SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link); 2746 pmap_mpvop_free++; 2747 if (pmap_mpvop_free > pmap_mpvop_maxfree) 2748 pmap_mpvop_maxfree = pmap_mpvop_free; 2749 #if 0 2750 uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va)); 2751 #endif 2752 } 2753 2754 /* 2755 * This routine in bootstraping to steal to-be-managed memory (which will 2756 * then be unmanaged). We use it to grab from the first 256MB for our 2757 * pmap needs and above 256MB for other stuff. 2758 */ 2759 vaddr_t 2760 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp) 2761 { 2762 vsize_t size; 2763 vaddr_t va; 2764 paddr_t pa = 0; 2765 int npgs, bank; 2766 struct vm_physseg *ps; 2767 2768 if (uvm.page_init_done == TRUE) 2769 panic("pmap_steal_memory: called _after_ bootstrap"); 2770 2771 *vstartp = VM_MIN_KERNEL_ADDRESS; 2772 *vendp = VM_MAX_KERNEL_ADDRESS; 2773 2774 size = round_page(vsize); 2775 npgs = atop(size); 2776 2777 /* 2778 * PA 0 will never be among those given to UVM so we can use it 2779 * to indicate we couldn't steal any memory. 2780 */ 2781 for (ps = vm_physmem, bank = 0; bank < vm_nphysseg; bank++, ps++) { 2782 if (ps->free_list == VM_FREELIST_FIRST256 && 2783 ps->avail_end - ps->avail_start >= npgs) { 2784 pa = ptoa(ps->avail_start); 2785 break; 2786 } 2787 } 2788 2789 if (pa == 0) 2790 panic("pmap_steal_memory: no approriate memory to steal!"); 2791 2792 ps->avail_start += npgs; 2793 ps->start += npgs; 2794 2795 /* 2796 * If we've used up all the pages in the segment, remove it and 2797 * compact the list. 2798 */ 2799 if (ps->avail_start == ps->end) { 2800 /* 2801 * If this was the last one, then a very bad thing has occurred 2802 */ 2803 if (--vm_nphysseg == 0) 2804 panic("pmap_steal_memory: out of memory!"); 2805 2806 printf("pmap_steal_memory: consumed bank %d\n", bank); 2807 for (; bank < vm_nphysseg; bank++, ps++) { 2808 ps[0] = ps[1]; 2809 } 2810 } 2811 2812 va = (vaddr_t) pa; 2813 memset((caddr_t) va, 0, size); 2814 pmap_pages_stolen += npgs; 2815 #ifdef DEBUG 2816 if (pmapdebug && npgs > 1) { 2817 u_int cnt = 0; 2818 for (bank = 0, ps = vm_physmem; bank < vm_nphysseg; bank++, ps++) 2819 cnt += ps->avail_end - ps->avail_start; 2820 printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n", 2821 npgs, pmap_pages_stolen, cnt); 2822 } 2823 #endif 2824 2825 return va; 2826 } 2827 2828 /* 2829 * Find a chuck of memory with right size and alignment. 2830 */ 2831 void * 2832 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end) 2833 { 2834 struct mem_region *mp; 2835 paddr_t s, e; 2836 int i, j; 2837 2838 size = round_page(size); 2839 2840 DPRINTFN(BOOT, 2841 ("pmap_boot_find_memory: size=%lx, alignment=%lx, at_end=%d", 2842 size, alignment, at_end)); 2843 2844 if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0) 2845 panic("pmap_boot_find_memory: invalid alignment %lx", 2846 alignment); 2847 2848 if (at_end) { 2849 if (alignment != PAGE_SIZE) 2850 panic("pmap_boot_find_memory: invalid ending " 2851 "alignment %lx", alignment); 2852 2853 for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) { 2854 s = mp->start + mp->size - size; 2855 if (s >= mp->start && mp->size >= size) { 2856 DPRINTFN(BOOT,(": %lx\n", s)); 2857 DPRINTFN(BOOT, 2858 ("pmap_boot_find_memory: b-avail[%d] start " 2859 "0x%lx size 0x%lx\n", mp - avail, 2860 mp->start, mp->size)); 2861 mp->size -= size; 2862 DPRINTFN(BOOT, 2863 ("pmap_boot_find_memory: a-avail[%d] start " 2864 "0x%lx size 0x%lx\n", mp - avail, 2865 mp->start, mp->size)); 2866 return (void *) s; 2867 } 2868 } 2869 panic("pmap_boot_find_memory: no available memory"); 2870 } 2871 2872 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2873 s = (mp->start + alignment - 1) & ~(alignment-1); 2874 e = s + size; 2875 2876 /* 2877 * Is the calculated region entirely within the region? 2878 */ 2879 if (s < mp->start || e > mp->start + mp->size) 2880 continue; 2881 2882 DPRINTFN(BOOT,(": %lx\n", s)); 2883 if (s == mp->start) { 2884 /* 2885 * If the block starts at the beginning of region, 2886 * adjust the size & start. (the region may now be 2887 * zero in length) 2888 */ 2889 DPRINTFN(BOOT, 2890 ("pmap_boot_find_memory: b-avail[%d] start " 2891 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2892 mp->start += size; 2893 mp->size -= size; 2894 DPRINTFN(BOOT, 2895 ("pmap_boot_find_memory: a-avail[%d] start " 2896 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2897 } else if (e == mp->start + mp->size) { 2898 /* 2899 * If the block starts at the beginning of region, 2900 * adjust only the size. 2901 */ 2902 DPRINTFN(BOOT, 2903 ("pmap_boot_find_memory: b-avail[%d] start " 2904 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2905 mp->size -= size; 2906 DPRINTFN(BOOT, 2907 ("pmap_boot_find_memory: a-avail[%d] start " 2908 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2909 } else { 2910 /* 2911 * Block is in the middle of the region, so we 2912 * have to split it in two. 2913 */ 2914 for (j = avail_cnt; j > i + 1; j--) { 2915 avail[j] = avail[j-1]; 2916 } 2917 DPRINTFN(BOOT, 2918 ("pmap_boot_find_memory: b-avail[%d] start " 2919 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2920 mp[1].start = e; 2921 mp[1].size = mp[0].start + mp[0].size - e; 2922 mp[0].size = s - mp[0].start; 2923 avail_cnt++; 2924 for (; i < avail_cnt; i++) { 2925 DPRINTFN(BOOT, 2926 ("pmap_boot_find_memory: a-avail[%d] " 2927 "start 0x%lx size 0x%lx\n", i, 2928 avail[i].start, avail[i].size)); 2929 } 2930 } 2931 return (void *) s; 2932 } 2933 panic("pmap_boot_find_memory: not enough memory for " 2934 "%lx/%lx allocation?", size, alignment); 2935 } 2936 2937 /* 2938 * This is not part of the defined PMAP interface and is specific to the 2939 * PowerPC architecture. This is called during initppc, before the system 2940 * is really initialized. 2941 */ 2942 void 2943 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend) 2944 { 2945 struct mem_region *mp, tmp; 2946 paddr_t s, e; 2947 psize_t size; 2948 int i, j; 2949 2950 /* 2951 * Get memory. 2952 */ 2953 mem_regions(&mem, &avail); 2954 #if defined(DEBUG) 2955 if (pmapdebug & PMAPDEBUG_BOOT) { 2956 printf("pmap_bootstrap: memory configuration:\n"); 2957 for (mp = mem; mp->size; mp++) { 2958 printf("pmap_bootstrap: mem start 0x%lx size 0x%lx\n", 2959 mp->start, mp->size); 2960 } 2961 for (mp = avail; mp->size; mp++) { 2962 printf("pmap_bootstrap: avail start 0x%lx size 0x%lx\n", 2963 mp->start, mp->size); 2964 } 2965 } 2966 #endif 2967 2968 /* 2969 * Find out how much physical memory we have and in how many chunks. 2970 */ 2971 for (mem_cnt = 0, mp = mem; mp->size; mp++) { 2972 if (mp->start >= pmap_memlimit) 2973 continue; 2974 if (mp->start + mp->size > pmap_memlimit) { 2975 size = pmap_memlimit - mp->start; 2976 physmem += btoc(size); 2977 } else { 2978 physmem += btoc(mp->size); 2979 } 2980 mem_cnt++; 2981 } 2982 2983 /* 2984 * Count the number of available entries. 2985 */ 2986 for (avail_cnt = 0, mp = avail; mp->size; mp++) 2987 avail_cnt++; 2988 2989 /* 2990 * Page align all regions. 2991 */ 2992 kernelstart = trunc_page(kernelstart); 2993 kernelend = round_page(kernelend); 2994 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2995 s = round_page(mp->start); 2996 mp->size -= (s - mp->start); 2997 mp->size = trunc_page(mp->size); 2998 mp->start = s; 2999 e = mp->start + mp->size; 3000 3001 DPRINTFN(BOOT, 3002 ("pmap_bootstrap: b-avail[%d] start 0x%lx size 0x%lx\n", 3003 i, mp->start, mp->size)); 3004 3005 /* 3006 * Don't allow the end to run beyond our artificial limit 3007 */ 3008 if (e > pmap_memlimit) 3009 e = pmap_memlimit; 3010 3011 /* 3012 * Is this region empty or strange? skip it. 3013 */ 3014 if (e <= s) { 3015 mp->start = 0; 3016 mp->size = 0; 3017 continue; 3018 } 3019 3020 /* 3021 * Does this overlap the beginning of kernel? 3022 * Does extend past the end of the kernel? 3023 */ 3024 else if (s < kernelstart && e > kernelstart) { 3025 if (e > kernelend) { 3026 avail[avail_cnt].start = kernelend; 3027 avail[avail_cnt].size = e - kernelend; 3028 avail_cnt++; 3029 } 3030 mp->size = kernelstart - s; 3031 } 3032 /* 3033 * Check whether this region overlaps the end of the kernel. 3034 */ 3035 else if (s < kernelend && e > kernelend) { 3036 mp->start = kernelend; 3037 mp->size = e - kernelend; 3038 } 3039 /* 3040 * Look whether this regions is completely inside the kernel. 3041 * Nuke it if it does. 3042 */ 3043 else if (s >= kernelstart && e <= kernelend) { 3044 mp->start = 0; 3045 mp->size = 0; 3046 } 3047 /* 3048 * If the user imposed a memory limit, enforce it. 3049 */ 3050 else if (s >= pmap_memlimit) { 3051 mp->start = -PAGE_SIZE; /* let's know why */ 3052 mp->size = 0; 3053 } 3054 else { 3055 mp->start = s; 3056 mp->size = e - s; 3057 } 3058 DPRINTFN(BOOT, 3059 ("pmap_bootstrap: a-avail[%d] start 0x%lx size 0x%lx\n", 3060 i, mp->start, mp->size)); 3061 } 3062 3063 /* 3064 * Move (and uncount) all the null return to the end. 3065 */ 3066 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 3067 if (mp->size == 0) { 3068 tmp = avail[i]; 3069 avail[i] = avail[--avail_cnt]; 3070 avail[avail_cnt] = avail[i]; 3071 } 3072 } 3073 3074 /* 3075 * (Bubble)sort them into asecnding order. 3076 */ 3077 for (i = 0; i < avail_cnt; i++) { 3078 for (j = i + 1; j < avail_cnt; j++) { 3079 if (avail[i].start > avail[j].start) { 3080 tmp = avail[i]; 3081 avail[i] = avail[j]; 3082 avail[j] = tmp; 3083 } 3084 } 3085 } 3086 3087 /* 3088 * Make sure they don't overlap. 3089 */ 3090 for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) { 3091 if (mp[0].start + mp[0].size > mp[1].start) { 3092 mp[0].size = mp[1].start - mp[0].start; 3093 } 3094 DPRINTFN(BOOT, 3095 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 3096 i, mp->start, mp->size)); 3097 } 3098 DPRINTFN(BOOT, 3099 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 3100 i, mp->start, mp->size)); 3101 3102 #ifdef PTEGCOUNT 3103 pmap_pteg_cnt = PTEGCOUNT; 3104 #else /* PTEGCOUNT */ 3105 pmap_pteg_cnt = 0x1000; 3106 3107 while (pmap_pteg_cnt < physmem) 3108 pmap_pteg_cnt <<= 1; 3109 3110 pmap_pteg_cnt >>= 1; 3111 #endif /* PTEGCOUNT */ 3112 3113 /* 3114 * Find suitably aligned memory for PTEG hash table. 3115 */ 3116 size = pmap_pteg_cnt * sizeof(struct pteg); 3117 pmap_pteg_table = pmap_boot_find_memory(size, size, 0); 3118 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3119 if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH) 3120 panic("pmap_bootstrap: pmap_pteg_table end (%p + %lx) > 256MB", 3121 pmap_pteg_table, size); 3122 #endif 3123 3124 memset(__UNVOLATILE(pmap_pteg_table), 0, 3125 pmap_pteg_cnt * sizeof(struct pteg)); 3126 pmap_pteg_mask = pmap_pteg_cnt - 1; 3127 3128 /* 3129 * We cannot do pmap_steal_memory here since UVM hasn't been loaded 3130 * with pages. So we just steal them before giving them to UVM. 3131 */ 3132 size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt; 3133 pmap_pvo_table = pmap_boot_find_memory(size, PAGE_SIZE, 0); 3134 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3135 if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH) 3136 panic("pmap_bootstrap: pmap_pvo_table end (%p + %lx) > 256MB", 3137 pmap_pvo_table, size); 3138 #endif 3139 3140 for (i = 0; i < pmap_pteg_cnt; i++) 3141 TAILQ_INIT(&pmap_pvo_table[i]); 3142 3143 #ifndef MSGBUFADDR 3144 /* 3145 * Allocate msgbuf in high memory. 3146 */ 3147 msgbuf_paddr = 3148 (paddr_t) pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1); 3149 #endif 3150 3151 #ifdef __HAVE_PMAP_PHYSSEG 3152 { 3153 u_int npgs = 0; 3154 for (i = 0, mp = avail; i < avail_cnt; i++, mp++) 3155 npgs += btoc(mp->size); 3156 size = (sizeof(struct pvo_head) + 1) * npgs; 3157 pmap_physseg.pvoh = pmap_boot_find_memory(size, PAGE_SIZE, 0); 3158 pmap_physseg.attrs = (char *) &pmap_physseg.pvoh[npgs]; 3159 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3160 if ((uintptr_t)pmap_physseg.pvoh + size > SEGMENT_LENGTH) 3161 panic("pmap_bootstrap: PVO list end (%p + %lx) > 256MB", 3162 pmap_physseg.pvoh, size); 3163 #endif 3164 } 3165 #endif 3166 3167 for (mp = avail, i = 0; i < avail_cnt; mp++, i++) { 3168 paddr_t pfstart = atop(mp->start); 3169 paddr_t pfend = atop(mp->start + mp->size); 3170 if (mp->size == 0) 3171 continue; 3172 if (mp->start + mp->size <= SEGMENT_LENGTH) { 3173 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3174 VM_FREELIST_FIRST256); 3175 } else if (mp->start >= SEGMENT_LENGTH) { 3176 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3177 VM_FREELIST_DEFAULT); 3178 } else { 3179 pfend = atop(SEGMENT_LENGTH); 3180 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3181 VM_FREELIST_FIRST256); 3182 pfstart = atop(SEGMENT_LENGTH); 3183 pfend = atop(mp->start + mp->size); 3184 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3185 VM_FREELIST_DEFAULT); 3186 } 3187 } 3188 3189 /* 3190 * Make sure kernel vsid is allocated as well as VSID 0. 3191 */ 3192 pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW] 3193 |= 1 << (KERNEL_VSIDBITS % VSID_NBPW); 3194 pmap_vsid_bitmap[0] |= 1; 3195 3196 /* 3197 * Initialize kernel pmap and hardware. 3198 */ 3199 #ifndef PPC_OEA64 3200 for (i = 0; i < 16; i++) { 3201 pmap_kernel()->pm_sr[i] = EMPTY_SEGMENT; 3202 __asm volatile ("mtsrin %0,%1" 3203 :: "r"(EMPTY_SEGMENT), "r"(i << ADDR_SR_SHFT)); 3204 } 3205 3206 pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY; 3207 __asm volatile ("mtsr %0,%1" 3208 :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT)); 3209 #ifdef KERNEL2_SR 3210 pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY; 3211 __asm volatile ("mtsr %0,%1" 3212 :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT)); 3213 #endif 3214 for (i = 0; i < 16; i++) { 3215 if (iosrtable[i] & SR601_T) { 3216 pmap_kernel()->pm_sr[i] = iosrtable[i]; 3217 __asm volatile ("mtsrin %0,%1" 3218 :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT)); 3219 } 3220 } 3221 #endif /* !PPC_OEA64 */ 3222 3223 __asm volatile ("sync; mtsdr1 %0; isync" 3224 :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10))); 3225 tlbia(); 3226 3227 #ifdef ALTIVEC 3228 pmap_use_altivec = cpu_altivec; 3229 #endif 3230 3231 #ifdef DEBUG 3232 if (pmapdebug & PMAPDEBUG_BOOT) { 3233 u_int cnt; 3234 int bank; 3235 char pbuf[9]; 3236 for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) { 3237 cnt += vm_physmem[bank].avail_end - vm_physmem[bank].avail_start; 3238 printf("pmap_bootstrap: vm_physmem[%d]=%#lx-%#lx/%#lx\n", 3239 bank, 3240 ptoa(vm_physmem[bank].avail_start), 3241 ptoa(vm_physmem[bank].avail_end), 3242 ptoa(vm_physmem[bank].avail_end - vm_physmem[bank].avail_start)); 3243 } 3244 format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt)); 3245 printf("pmap_bootstrap: UVM memory = %s (%u pages)\n", 3246 pbuf, cnt); 3247 } 3248 #endif 3249 3250 pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry), 3251 sizeof(struct pvo_entry), 0, 0, "pmap_upvopl", 3252 &pmap_pool_uallocator); 3253 3254 pool_setlowat(&pmap_upvo_pool, 252); 3255 3256 pool_init(&pmap_pool, sizeof(struct pmap), 3257 sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator); 3258 } 3259