pmap.c revision 1.32.2.1
1 /* $NetBSD: pmap.c,v 1.32.2.1 2006/06/21 14:55:03 yamt Exp $ */ 2 /*- 3 * Copyright (c) 2001 The NetBSD Foundation, Inc. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to The NetBSD Foundation 7 * by Matt Thomas <matt (at) 3am-software.com> of Allegro Networks, Inc. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the NetBSD 20 * Foundation, Inc. and its contributors. 21 * 4. Neither the name of The NetBSD Foundation nor the names of its 22 * contributors may be used to endorse or promote products derived 23 * from this software without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 26 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 27 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 28 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 29 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 30 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 31 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 32 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 33 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 34 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 35 * POSSIBILITY OF SUCH DAMAGE. 36 */ 37 38 /* 39 * Copyright (C) 1995, 1996 Wolfgang Solfrank. 40 * Copyright (C) 1995, 1996 TooLs GmbH. 41 * All rights reserved. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. All advertising materials mentioning features or use of this software 52 * must display the following acknowledgement: 53 * This product includes software developed by TooLs GmbH. 54 * 4. The name of TooLs GmbH may not be used to endorse or promote products 55 * derived from this software without specific prior written permission. 56 * 57 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR 58 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 59 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 60 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 61 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 62 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 63 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 64 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 65 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 66 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.32.2.1 2006/06/21 14:55:03 yamt 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 struct evcnt pmap_evcnt_exec_synced_pvo_remove = 327 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 328 "pmap", "exec pages synced (PR)"); 329 330 struct evcnt pmap_evcnt_exec_uncached_page_protect = 331 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 332 "pmap", "exec pages uncached (PP)"); 333 struct evcnt pmap_evcnt_exec_uncached_clear_modify = 334 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 335 "pmap", "exec pages uncached (CM)"); 336 struct evcnt pmap_evcnt_exec_uncached_zero_page = 337 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 338 "pmap", "exec pages uncached (ZP)"); 339 struct evcnt pmap_evcnt_exec_uncached_copy_page = 340 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 341 "pmap", "exec pages uncached (CP)"); 342 struct evcnt pmap_evcnt_exec_uncached_pvo_remove = 343 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 344 "pmap", "exec pages uncached (PR)"); 345 346 struct evcnt pmap_evcnt_updates = 347 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 348 "pmap", "updates"); 349 struct evcnt pmap_evcnt_collects = 350 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 351 "pmap", "collects"); 352 struct evcnt pmap_evcnt_copies = 353 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 354 "pmap", "copies"); 355 356 struct evcnt pmap_evcnt_ptes_spilled = 357 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 358 "pmap", "ptes spilled from overflow"); 359 struct evcnt pmap_evcnt_ptes_unspilled = 360 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 361 "pmap", "ptes not spilled"); 362 struct evcnt pmap_evcnt_ptes_evicted = 363 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 364 "pmap", "ptes evicted"); 365 366 struct evcnt pmap_evcnt_ptes_primary[8] = { 367 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 368 "pmap", "ptes added at primary[0]"), 369 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 370 "pmap", "ptes added at primary[1]"), 371 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 372 "pmap", "ptes added at primary[2]"), 373 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 374 "pmap", "ptes added at primary[3]"), 375 376 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 377 "pmap", "ptes added at primary[4]"), 378 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 379 "pmap", "ptes added at primary[5]"), 380 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 381 "pmap", "ptes added at primary[6]"), 382 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 383 "pmap", "ptes added at primary[7]"), 384 }; 385 struct evcnt pmap_evcnt_ptes_secondary[8] = { 386 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 387 "pmap", "ptes added at secondary[0]"), 388 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 389 "pmap", "ptes added at secondary[1]"), 390 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 391 "pmap", "ptes added at secondary[2]"), 392 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 393 "pmap", "ptes added at secondary[3]"), 394 395 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 396 "pmap", "ptes added at secondary[4]"), 397 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 398 "pmap", "ptes added at secondary[5]"), 399 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 400 "pmap", "ptes added at secondary[6]"), 401 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 402 "pmap", "ptes added at secondary[7]"), 403 }; 404 struct evcnt pmap_evcnt_ptes_removed = 405 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 406 "pmap", "ptes removed"); 407 struct evcnt pmap_evcnt_ptes_changed = 408 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 409 "pmap", "ptes changed"); 410 struct evcnt pmap_evcnt_pvos_reclaimed = 411 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 412 "pmap", "pvos reclaimed"); 413 struct evcnt pmap_evcnt_pvos_failed = 414 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 415 "pmap", "pvo allocation failures"); 416 417 /* 418 * From pmap_subr.c 419 */ 420 extern struct evcnt pmap_evcnt_zeroed_pages; 421 extern struct evcnt pmap_evcnt_copied_pages; 422 extern struct evcnt pmap_evcnt_idlezeroed_pages; 423 424 EVCNT_ATTACH_STATIC(pmap_evcnt_mappings); 425 EVCNT_ATTACH_STATIC(pmap_evcnt_mappings_replaced); 426 EVCNT_ATTACH_STATIC(pmap_evcnt_unmappings); 427 428 EVCNT_ATTACH_STATIC(pmap_evcnt_kernel_mappings); 429 EVCNT_ATTACH_STATIC(pmap_evcnt_kernel_unmappings); 430 431 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_mappings); 432 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_cached); 433 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_synced); 434 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_synced_clear_modify); 435 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_synced_pvo_remove); 436 437 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_page_protect); 438 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_clear_modify); 439 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_zero_page); 440 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_copy_page); 441 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_pvo_remove); 442 443 EVCNT_ATTACH_STATIC(pmap_evcnt_zeroed_pages); 444 EVCNT_ATTACH_STATIC(pmap_evcnt_copied_pages); 445 EVCNT_ATTACH_STATIC(pmap_evcnt_idlezeroed_pages); 446 447 EVCNT_ATTACH_STATIC(pmap_evcnt_updates); 448 EVCNT_ATTACH_STATIC(pmap_evcnt_collects); 449 EVCNT_ATTACH_STATIC(pmap_evcnt_copies); 450 451 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_spilled); 452 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_unspilled); 453 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_evicted); 454 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_removed); 455 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_changed); 456 457 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 0); 458 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 1); 459 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 2); 460 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 3); 461 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 4); 462 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 5); 463 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 6); 464 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 7); 465 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 0); 466 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 1); 467 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 2); 468 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 3); 469 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 4); 470 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 5); 471 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 6); 472 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 7); 473 474 EVCNT_ATTACH_STATIC(pmap_evcnt_pvos_reclaimed); 475 EVCNT_ATTACH_STATIC(pmap_evcnt_pvos_failed); 476 #else 477 #define PMAPCOUNT(ev) ((void) 0) 478 #define PMAPCOUNT2(ev) ((void) 0) 479 #endif 480 481 #define TLBIE(va) __asm volatile("tlbie %0" :: "r"(va)) 482 #define TLBSYNC() __asm volatile("tlbsync") 483 #define SYNC() __asm volatile("sync") 484 #define EIEIO() __asm volatile("eieio") 485 #define MFMSR() mfmsr() 486 #define MTMSR(psl) mtmsr(psl) 487 #define MFPVR() mfpvr() 488 #define MFSRIN(va) mfsrin(va) 489 #define MFTB() mfrtcltbl() 490 491 #ifndef PPC_OEA64 492 static inline register_t 493 mfsrin(vaddr_t va) 494 { 495 register_t sr; 496 __asm volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va)); 497 return sr; 498 } 499 #endif /* PPC_OEA64 */ 500 501 static inline register_t 502 pmap_interrupts_off(void) 503 { 504 register_t msr = MFMSR(); 505 if (msr & PSL_EE) 506 MTMSR(msr & ~PSL_EE); 507 return msr; 508 } 509 510 static void 511 pmap_interrupts_restore(register_t msr) 512 { 513 if (msr & PSL_EE) 514 MTMSR(msr); 515 } 516 517 static inline u_int32_t 518 mfrtcltbl(void) 519 { 520 521 if ((MFPVR() >> 16) == MPC601) 522 return (mfrtcl() >> 7); 523 else 524 return (mftbl()); 525 } 526 527 /* 528 * These small routines may have to be replaced, 529 * if/when we support processors other that the 604. 530 */ 531 532 void 533 tlbia(void) 534 { 535 caddr_t i; 536 537 SYNC(); 538 /* 539 * Why not use "tlbia"? Because not all processors implement it. 540 * 541 * This needs to be a per-CPU callback to do the appropriate thing 542 * for the CPU. XXX 543 */ 544 for (i = 0; i < (caddr_t)0x00040000; i += 0x00001000) { 545 TLBIE(i); 546 EIEIO(); 547 SYNC(); 548 } 549 TLBSYNC(); 550 SYNC(); 551 } 552 553 static inline register_t 554 va_to_vsid(const struct pmap *pm, vaddr_t addr) 555 { 556 #ifdef PPC_OEA64 557 #if 0 558 const struct ste *ste; 559 register_t hash; 560 int i; 561 562 hash = (addr >> ADDR_ESID_SHFT) & ADDR_ESID_HASH; 563 564 /* 565 * Try the primary group first 566 */ 567 ste = pm->pm_stes[hash].stes; 568 for (i = 0; i < 8; i++, ste++) { 569 if (ste->ste_hi & STE_V) && 570 (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID)) 571 return ste; 572 } 573 574 /* 575 * Then the secondary group. 576 */ 577 ste = pm->pm_stes[hash ^ ADDR_ESID_HASH].stes; 578 for (i = 0; i < 8; i++, ste++) { 579 if (ste->ste_hi & STE_V) && 580 (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID)) 581 return addr; 582 } 583 584 return NULL; 585 #else 586 /* 587 * Rather than searching the STE groups for the VSID, we know 588 * how we generate that from the ESID and so do that. 589 */ 590 return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT; 591 #endif 592 #else 593 return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID) >> SR_VSID_SHFT; 594 #endif 595 } 596 597 static inline register_t 598 va_to_pteg(const struct pmap *pm, vaddr_t addr) 599 { 600 register_t hash; 601 602 hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT); 603 return hash & pmap_pteg_mask; 604 } 605 606 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 607 /* 608 * Given a PTE in the page table, calculate the VADDR that hashes to it. 609 * The only bit of magic is that the top 4 bits of the address doesn't 610 * technically exist in the PTE. But we know we reserved 4 bits of the 611 * VSID for it so that's how we get it. 612 */ 613 static vaddr_t 614 pmap_pte_to_va(volatile const struct pte *pt) 615 { 616 vaddr_t va; 617 uintptr_t ptaddr = (uintptr_t) pt; 618 619 if (pt->pte_hi & PTE_HID) 620 ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg)); 621 622 /* PPC Bits 10-19 PPC64 Bits 42-51 */ 623 va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff; 624 va <<= ADDR_PIDX_SHFT; 625 626 /* PPC Bits 4-9 PPC64 Bits 36-41 */ 627 va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT; 628 629 #ifdef PPC_OEA64 630 /* PPC63 Bits 0-35 */ 631 /* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */ 632 #endif 633 #ifdef PPC_OEA 634 /* PPC Bits 0-3 */ 635 va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; 636 #endif 637 638 return va; 639 } 640 #endif 641 642 static inline struct pvo_head * 643 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p) 644 { 645 #ifdef __HAVE_VM_PAGE_MD 646 struct vm_page *pg; 647 648 pg = PHYS_TO_VM_PAGE(pa); 649 if (pg_p != NULL) 650 *pg_p = pg; 651 if (pg == NULL) 652 return &pmap_pvo_unmanaged; 653 return &pg->mdpage.mdpg_pvoh; 654 #endif 655 #ifdef __HAVE_PMAP_PHYSSEG 656 int bank, pg; 657 658 bank = vm_physseg_find(atop(pa), &pg); 659 if (pg_p != NULL) 660 *pg_p = pg; 661 if (bank == -1) 662 return &pmap_pvo_unmanaged; 663 return &vm_physmem[bank].pmseg.pvoh[pg]; 664 #endif 665 } 666 667 static inline struct pvo_head * 668 vm_page_to_pvoh(struct vm_page *pg) 669 { 670 #ifdef __HAVE_VM_PAGE_MD 671 return &pg->mdpage.mdpg_pvoh; 672 #endif 673 #ifdef __HAVE_PMAP_PHYSSEG 674 return pa_to_pvoh(VM_PAGE_TO_PHYS(pg), NULL); 675 #endif 676 } 677 678 679 #ifdef __HAVE_PMAP_PHYSSEG 680 static inline char * 681 pa_to_attr(paddr_t pa) 682 { 683 int bank, pg; 684 685 bank = vm_physseg_find(atop(pa), &pg); 686 if (bank == -1) 687 return NULL; 688 return &vm_physmem[bank].pmseg.attrs[pg]; 689 } 690 #endif 691 692 static inline void 693 pmap_attr_clear(struct vm_page *pg, int ptebit) 694 { 695 #ifdef __HAVE_PMAP_PHYSSEG 696 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) &= ~(ptebit >> ATTR_SHFT); 697 #endif 698 #ifdef __HAVE_VM_PAGE_MD 699 pg->mdpage.mdpg_attrs &= ~ptebit; 700 #endif 701 } 702 703 static inline int 704 pmap_attr_fetch(struct vm_page *pg) 705 { 706 #ifdef __HAVE_PMAP_PHYSSEG 707 return *pa_to_attr(VM_PAGE_TO_PHYS(pg)) << ATTR_SHFT; 708 #endif 709 #ifdef __HAVE_VM_PAGE_MD 710 return pg->mdpage.mdpg_attrs; 711 #endif 712 } 713 714 static inline void 715 pmap_attr_save(struct vm_page *pg, int ptebit) 716 { 717 #ifdef __HAVE_PMAP_PHYSSEG 718 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) |= (ptebit >> ATTR_SHFT); 719 #endif 720 #ifdef __HAVE_VM_PAGE_MD 721 pg->mdpage.mdpg_attrs |= ptebit; 722 #endif 723 } 724 725 static inline int 726 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt) 727 { 728 if (pt->pte_hi == pvo_pt->pte_hi 729 #if 0 730 && ((pt->pte_lo ^ pvo_pt->pte_lo) & 731 ~(PTE_REF|PTE_CHG)) == 0 732 #endif 733 ) 734 return 1; 735 return 0; 736 } 737 738 static inline void 739 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo) 740 { 741 /* 742 * Construct the PTE. Default to IMB initially. Valid bit 743 * only gets set when the real pte is set in memory. 744 * 745 * Note: Don't set the valid bit for correct operation of tlb update. 746 */ 747 pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT) 748 | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API); 749 pt->pte_lo = pte_lo; 750 } 751 752 static inline void 753 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt) 754 { 755 pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG); 756 } 757 758 static inline void 759 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit) 760 { 761 /* 762 * As shown in Section 7.6.3.2.3 763 */ 764 pt->pte_lo &= ~ptebit; 765 TLBIE(va); 766 SYNC(); 767 EIEIO(); 768 TLBSYNC(); 769 SYNC(); 770 } 771 772 static inline void 773 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt) 774 { 775 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 776 if (pvo_pt->pte_hi & PTE_VALID) 777 panic("pte_set: setting an already valid pte %p", pvo_pt); 778 #endif 779 pvo_pt->pte_hi |= PTE_VALID; 780 /* 781 * Update the PTE as defined in section 7.6.3.1 782 * Note that the REF/CHG bits are from pvo_pt and thus should 783 * have been saved so this routine can restore them (if desired). 784 */ 785 pt->pte_lo = pvo_pt->pte_lo; 786 EIEIO(); 787 pt->pte_hi = pvo_pt->pte_hi; 788 SYNC(); 789 pmap_pte_valid++; 790 } 791 792 static inline void 793 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 794 { 795 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 796 if ((pvo_pt->pte_hi & PTE_VALID) == 0) 797 panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt); 798 if ((pt->pte_hi & PTE_VALID) == 0) 799 panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt); 800 #endif 801 802 pvo_pt->pte_hi &= ~PTE_VALID; 803 /* 804 * Force the ref & chg bits back into the PTEs. 805 */ 806 SYNC(); 807 /* 808 * Invalidate the pte ... (Section 7.6.3.3) 809 */ 810 pt->pte_hi &= ~PTE_VALID; 811 SYNC(); 812 TLBIE(va); 813 SYNC(); 814 EIEIO(); 815 TLBSYNC(); 816 SYNC(); 817 /* 818 * Save the ref & chg bits ... 819 */ 820 pmap_pte_synch(pt, pvo_pt); 821 pmap_pte_valid--; 822 } 823 824 static inline void 825 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 826 { 827 /* 828 * Invalidate the PTE 829 */ 830 pmap_pte_unset(pt, pvo_pt, va); 831 pmap_pte_set(pt, pvo_pt); 832 } 833 834 /* 835 * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx 836 * (either primary or secondary location). 837 * 838 * Note: both the destination and source PTEs must not have PTE_VALID set. 839 */ 840 841 STATIC int 842 pmap_pte_insert(int ptegidx, struct pte *pvo_pt) 843 { 844 volatile struct pte *pt; 845 int i; 846 847 #if defined(DEBUG) 848 DPRINTFN(PTE, ("pmap_pte_insert: idx 0x%x, pte 0x%x 0x%x\n", 849 ptegidx, (unsigned int) pvo_pt->pte_hi, (unsigned int) pvo_pt->pte_lo)); 850 #endif 851 /* 852 * First try primary hash. 853 */ 854 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 855 if ((pt->pte_hi & PTE_VALID) == 0) { 856 pvo_pt->pte_hi &= ~PTE_HID; 857 pmap_pte_set(pt, pvo_pt); 858 return i; 859 } 860 } 861 862 /* 863 * Now try secondary hash. 864 */ 865 ptegidx ^= pmap_pteg_mask; 866 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 867 if ((pt->pte_hi & PTE_VALID) == 0) { 868 pvo_pt->pte_hi |= PTE_HID; 869 pmap_pte_set(pt, pvo_pt); 870 return i; 871 } 872 } 873 return -1; 874 } 875 876 /* 877 * Spill handler. 878 * 879 * Tries to spill a page table entry from the overflow area. 880 * This runs in either real mode (if dealing with a exception spill) 881 * or virtual mode when dealing with manually spilling one of the 882 * kernel's pte entries. In either case, interrupts are already 883 * disabled. 884 */ 885 886 int 887 pmap_pte_spill(struct pmap *pm, vaddr_t addr, boolean_t exec) 888 { 889 struct pvo_entry *source_pvo, *victim_pvo, *next_pvo; 890 struct pvo_entry *pvo; 891 /* XXX: gcc -- vpvoh is always set at either *1* or *2* */ 892 struct pvo_tqhead *pvoh, *vpvoh = NULL; 893 int ptegidx, i, j; 894 volatile struct pteg *pteg; 895 volatile struct pte *pt; 896 897 ptegidx = va_to_pteg(pm, addr); 898 899 /* 900 * Have to substitute some entry. Use the primary hash for this. 901 * Use low bits of timebase as random generator. Make sure we are 902 * not picking a kernel pte for replacement. 903 */ 904 pteg = &pmap_pteg_table[ptegidx]; 905 i = MFTB() & 7; 906 for (j = 0; j < 8; j++) { 907 pt = &pteg->pt[i]; 908 if ((pt->pte_hi & PTE_VALID) == 0 || 909 VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT) 910 != KERNEL_VSIDBITS) 911 break; 912 i = (i + 1) & 7; 913 } 914 KASSERT(j < 8); 915 916 source_pvo = NULL; 917 victim_pvo = NULL; 918 pvoh = &pmap_pvo_table[ptegidx]; 919 TAILQ_FOREACH(pvo, pvoh, pvo_olink) { 920 921 /* 922 * We need to find pvo entry for this address... 923 */ 924 PMAP_PVO_CHECK(pvo); /* sanity check */ 925 926 /* 927 * If we haven't found the source and we come to a PVO with 928 * a valid PTE, then we know we can't find it because all 929 * evicted PVOs always are first in the list. 930 */ 931 if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID)) 932 break; 933 if (source_pvo == NULL && pm == pvo->pvo_pmap && 934 addr == PVO_VADDR(pvo)) { 935 936 /* 937 * Now we have found the entry to be spilled into the 938 * pteg. Attempt to insert it into the page table. 939 */ 940 j = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 941 if (j >= 0) { 942 PVO_PTEGIDX_SET(pvo, j); 943 PMAP_PVO_CHECK(pvo); /* sanity check */ 944 PVO_WHERE(pvo, SPILL_INSERT); 945 pvo->pvo_pmap->pm_evictions--; 946 PMAPCOUNT(ptes_spilled); 947 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 948 ? pmap_evcnt_ptes_secondary 949 : pmap_evcnt_ptes_primary)[j]); 950 951 /* 952 * Since we keep the evicted entries at the 953 * from of the PVO list, we need move this 954 * (now resident) PVO after the evicted 955 * entries. 956 */ 957 next_pvo = TAILQ_NEXT(pvo, pvo_olink); 958 959 /* 960 * If we don't have to move (either we were the 961 * last entry or the next entry was valid), 962 * don't change our position. Otherwise 963 * move ourselves to the tail of the queue. 964 */ 965 if (next_pvo != NULL && 966 !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) { 967 TAILQ_REMOVE(pvoh, pvo, pvo_olink); 968 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 969 } 970 return 1; 971 } 972 source_pvo = pvo; 973 if (exec && !PVO_ISEXECUTABLE(source_pvo)) { 974 return 0; 975 } 976 if (victim_pvo != NULL) 977 break; 978 } 979 980 /* 981 * We also need the pvo entry of the victim we are replacing 982 * so save the R & C bits of the PTE. 983 */ 984 if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL && 985 pmap_pte_compare(pt, &pvo->pvo_pte)) { 986 vpvoh = pvoh; /* *1* */ 987 victim_pvo = pvo; 988 if (source_pvo != NULL) 989 break; 990 } 991 } 992 993 if (source_pvo == NULL) { 994 PMAPCOUNT(ptes_unspilled); 995 return 0; 996 } 997 998 if (victim_pvo == NULL) { 999 if ((pt->pte_hi & PTE_HID) == 0) 1000 panic("pmap_pte_spill: victim p-pte (%p) has " 1001 "no pvo entry!", pt); 1002 1003 /* 1004 * If this is a secondary PTE, we need to search 1005 * its primary pvo bucket for the matching PVO. 1006 */ 1007 vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; /* *2* */ 1008 TAILQ_FOREACH(pvo, vpvoh, pvo_olink) { 1009 PMAP_PVO_CHECK(pvo); /* sanity check */ 1010 1011 /* 1012 * We also need the pvo entry of the victim we are 1013 * replacing so save the R & C bits of the PTE. 1014 */ 1015 if (pmap_pte_compare(pt, &pvo->pvo_pte)) { 1016 victim_pvo = pvo; 1017 break; 1018 } 1019 } 1020 if (victim_pvo == NULL) 1021 panic("pmap_pte_spill: victim s-pte (%p) has " 1022 "no pvo entry!", pt); 1023 } 1024 1025 /* 1026 * The victim should be not be a kernel PVO/PTE entry. 1027 */ 1028 KASSERT(victim_pvo->pvo_pmap != pmap_kernel()); 1029 KASSERT(PVO_PTEGIDX_ISSET(victim_pvo)); 1030 KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i); 1031 1032 /* 1033 * We are invalidating the TLB entry for the EA for the 1034 * we are replacing even though its valid; If we don't 1035 * we lose any ref/chg bit changes contained in the TLB 1036 * entry. 1037 */ 1038 source_pvo->pvo_pte.pte_hi &= ~PTE_HID; 1039 1040 /* 1041 * To enforce the PVO list ordering constraint that all 1042 * evicted entries should come before all valid entries, 1043 * move the source PVO to the tail of its list and the 1044 * victim PVO to the head of its list (which might not be 1045 * the same list, if the victim was using the secondary hash). 1046 */ 1047 TAILQ_REMOVE(pvoh, source_pvo, pvo_olink); 1048 TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink); 1049 TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink); 1050 TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink); 1051 pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr); 1052 pmap_pte_set(pt, &source_pvo->pvo_pte); 1053 victim_pvo->pvo_pmap->pm_evictions++; 1054 source_pvo->pvo_pmap->pm_evictions--; 1055 PVO_WHERE(victim_pvo, SPILL_UNSET); 1056 PVO_WHERE(source_pvo, SPILL_SET); 1057 1058 PVO_PTEGIDX_CLR(victim_pvo); 1059 PVO_PTEGIDX_SET(source_pvo, i); 1060 PMAPCOUNT2(pmap_evcnt_ptes_primary[i]); 1061 PMAPCOUNT(ptes_spilled); 1062 PMAPCOUNT(ptes_evicted); 1063 PMAPCOUNT(ptes_removed); 1064 1065 PMAP_PVO_CHECK(victim_pvo); 1066 PMAP_PVO_CHECK(source_pvo); 1067 return 1; 1068 } 1069 1070 /* 1071 * Restrict given range to physical memory 1072 */ 1073 void 1074 pmap_real_memory(paddr_t *start, psize_t *size) 1075 { 1076 struct mem_region *mp; 1077 1078 for (mp = mem; mp->size; mp++) { 1079 if (*start + *size > mp->start 1080 && *start < mp->start + mp->size) { 1081 if (*start < mp->start) { 1082 *size -= mp->start - *start; 1083 *start = mp->start; 1084 } 1085 if (*start + *size > mp->start + mp->size) 1086 *size = mp->start + mp->size - *start; 1087 return; 1088 } 1089 } 1090 *size = 0; 1091 } 1092 1093 /* 1094 * Initialize anything else for pmap handling. 1095 * Called during vm_init(). 1096 */ 1097 void 1098 pmap_init(void) 1099 { 1100 #ifdef __HAVE_PMAP_PHYSSEG 1101 struct pvo_tqhead *pvoh; 1102 int bank; 1103 long sz; 1104 char *attr; 1105 1106 pvoh = pmap_physseg.pvoh; 1107 attr = pmap_physseg.attrs; 1108 for (bank = 0; bank < vm_nphysseg; bank++) { 1109 sz = vm_physmem[bank].end - vm_physmem[bank].start; 1110 vm_physmem[bank].pmseg.pvoh = pvoh; 1111 vm_physmem[bank].pmseg.attrs = attr; 1112 for (; sz > 0; sz--, pvoh++, attr++) { 1113 TAILQ_INIT(pvoh); 1114 *attr = 0; 1115 } 1116 } 1117 #endif 1118 1119 pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry), 1120 sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl", 1121 &pmap_pool_mallocator); 1122 1123 pool_setlowat(&pmap_mpvo_pool, 1008); 1124 1125 pmap_initialized = 1; 1126 1127 } 1128 1129 /* 1130 * How much virtual space does the kernel get? 1131 */ 1132 void 1133 pmap_virtual_space(vaddr_t *start, vaddr_t *end) 1134 { 1135 /* 1136 * For now, reserve one segment (minus some overhead) for kernel 1137 * virtual memory 1138 */ 1139 *start = VM_MIN_KERNEL_ADDRESS; 1140 *end = VM_MAX_KERNEL_ADDRESS; 1141 } 1142 1143 /* 1144 * Allocate, initialize, and return a new physical map. 1145 */ 1146 pmap_t 1147 pmap_create(void) 1148 { 1149 pmap_t pm; 1150 1151 pm = pool_get(&pmap_pool, PR_WAITOK); 1152 memset((caddr_t)pm, 0, sizeof *pm); 1153 pmap_pinit(pm); 1154 1155 DPRINTFN(CREATE,("pmap_create: pm %p:\n" 1156 "\t%06x %06x %06x %06x %06x %06x %06x %06x\n" 1157 "\t%06x %06x %06x %06x %06x %06x %06x %06x\n", pm, 1158 (unsigned int) pm->pm_sr[0], (unsigned int) pm->pm_sr[1], 1159 (unsigned int) pm->pm_sr[2], (unsigned int) pm->pm_sr[3], 1160 (unsigned int) pm->pm_sr[4], (unsigned int) pm->pm_sr[5], 1161 (unsigned int) pm->pm_sr[6], (unsigned int) pm->pm_sr[7], 1162 (unsigned int) pm->pm_sr[8], (unsigned int) pm->pm_sr[9], 1163 (unsigned int) pm->pm_sr[10], (unsigned int) pm->pm_sr[11], 1164 (unsigned int) pm->pm_sr[12], (unsigned int) pm->pm_sr[13], 1165 (unsigned int) pm->pm_sr[14], (unsigned int) pm->pm_sr[15])); 1166 return pm; 1167 } 1168 1169 /* 1170 * Initialize a preallocated and zeroed pmap structure. 1171 */ 1172 void 1173 pmap_pinit(pmap_t pm) 1174 { 1175 register_t entropy = MFTB(); 1176 register_t mask; 1177 int i; 1178 1179 /* 1180 * Allocate some segment registers for this pmap. 1181 */ 1182 pm->pm_refs = 1; 1183 for (i = 0; i < NPMAPS; i += VSID_NBPW) { 1184 static register_t pmap_vsidcontext; 1185 register_t hash; 1186 unsigned int n; 1187 1188 /* Create a new value by multiplying by a prime adding in 1189 * entropy from the timebase register. This is to make the 1190 * VSID more random so that the PT Hash function collides 1191 * less often. (note that the prime causes gcc to do shifts 1192 * instead of a multiply) 1193 */ 1194 pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy; 1195 hash = pmap_vsidcontext & (NPMAPS - 1); 1196 if (hash == 0) { /* 0 is special, avoid it */ 1197 entropy += 0xbadf00d; 1198 continue; 1199 } 1200 n = hash >> 5; 1201 mask = 1L << (hash & (VSID_NBPW-1)); 1202 hash = pmap_vsidcontext; 1203 if (pmap_vsid_bitmap[n] & mask) { /* collision? */ 1204 /* anything free in this bucket? */ 1205 if (~pmap_vsid_bitmap[n] == 0) { 1206 entropy = hash ^ (hash >> 16); 1207 continue; 1208 } 1209 i = ffs(~pmap_vsid_bitmap[n]) - 1; 1210 mask = 1L << i; 1211 hash &= ~(VSID_NBPW-1); 1212 hash |= i; 1213 } 1214 hash &= PTE_VSID >> PTE_VSID_SHFT; 1215 pmap_vsid_bitmap[n] |= mask; 1216 pm->pm_vsid = hash; 1217 #ifndef PPC_OEA64 1218 for (i = 0; i < 16; i++) 1219 pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY | 1220 SR_NOEXEC; 1221 #endif 1222 return; 1223 } 1224 panic("pmap_pinit: out of segments"); 1225 } 1226 1227 /* 1228 * Add a reference to the given pmap. 1229 */ 1230 void 1231 pmap_reference(pmap_t pm) 1232 { 1233 pm->pm_refs++; 1234 } 1235 1236 /* 1237 * Retire the given pmap from service. 1238 * Should only be called if the map contains no valid mappings. 1239 */ 1240 void 1241 pmap_destroy(pmap_t pm) 1242 { 1243 if (--pm->pm_refs == 0) { 1244 pmap_release(pm); 1245 pool_put(&pmap_pool, pm); 1246 } 1247 } 1248 1249 /* 1250 * Release any resources held by the given physical map. 1251 * Called when a pmap initialized by pmap_pinit is being released. 1252 */ 1253 void 1254 pmap_release(pmap_t pm) 1255 { 1256 int idx, mask; 1257 1258 if (pm->pm_sr[0] == 0) 1259 panic("pmap_release"); 1260 idx = pm->pm_vsid & (NPMAPS-1); 1261 mask = 1 << (idx % VSID_NBPW); 1262 idx /= VSID_NBPW; 1263 1264 KASSERT(pmap_vsid_bitmap[idx] & mask); 1265 pmap_vsid_bitmap[idx] &= ~mask; 1266 } 1267 1268 /* 1269 * Copy the range specified by src_addr/len 1270 * from the source map to the range dst_addr/len 1271 * in the destination map. 1272 * 1273 * This routine is only advisory and need not do anything. 1274 */ 1275 void 1276 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr, 1277 vsize_t len, vaddr_t src_addr) 1278 { 1279 PMAPCOUNT(copies); 1280 } 1281 1282 /* 1283 * Require that all active physical maps contain no 1284 * incorrect entries NOW. 1285 */ 1286 void 1287 pmap_update(struct pmap *pmap) 1288 { 1289 PMAPCOUNT(updates); 1290 TLBSYNC(); 1291 } 1292 1293 /* 1294 * Garbage collects the physical map system for 1295 * pages which are no longer used. 1296 * Success need not be guaranteed -- that is, there 1297 * may well be pages which are not referenced, but 1298 * others may be collected. 1299 * Called by the pageout daemon when pages are scarce. 1300 */ 1301 void 1302 pmap_collect(pmap_t pm) 1303 { 1304 PMAPCOUNT(collects); 1305 } 1306 1307 static inline int 1308 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx) 1309 { 1310 int pteidx; 1311 /* 1312 * We can find the actual pte entry without searching by 1313 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9] 1314 * and by noticing the HID bit. 1315 */ 1316 pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo); 1317 if (pvo->pvo_pte.pte_hi & PTE_HID) 1318 pteidx ^= pmap_pteg_mask * 8; 1319 return pteidx; 1320 } 1321 1322 volatile struct pte * 1323 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx) 1324 { 1325 volatile struct pte *pt; 1326 1327 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1328 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) 1329 return NULL; 1330 #endif 1331 1332 /* 1333 * If we haven't been supplied the ptegidx, calculate it. 1334 */ 1335 if (pteidx == -1) { 1336 int ptegidx; 1337 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1338 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1339 } 1340 1341 pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7]; 1342 1343 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1344 return pt; 1345 #else 1346 if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) { 1347 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1348 "pvo but no valid pte index", pvo); 1349 } 1350 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) { 1351 panic("pmap_pvo_to_pte: pvo %p: has valid pte index in " 1352 "pvo but no valid pte", pvo); 1353 } 1354 1355 if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) { 1356 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) { 1357 #if defined(DEBUG) || defined(PMAPCHECK) 1358 pmap_pte_print(pt); 1359 #endif 1360 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1361 "pmap_pteg_table %p but invalid in pvo", 1362 pvo, pt); 1363 } 1364 if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) { 1365 #if defined(DEBUG) || defined(PMAPCHECK) 1366 pmap_pte_print(pt); 1367 #endif 1368 panic("pmap_pvo_to_pte: pvo %p: pvo pte does " 1369 "not match pte %p in pmap_pteg_table", 1370 pvo, pt); 1371 } 1372 return pt; 1373 } 1374 1375 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1376 #if defined(DEBUG) || defined(PMAPCHECK) 1377 pmap_pte_print(pt); 1378 #endif 1379 panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in " 1380 "pmap_pteg_table but valid in pvo", pvo, pt); 1381 } 1382 return NULL; 1383 #endif /* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */ 1384 } 1385 1386 struct pvo_entry * 1387 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p) 1388 { 1389 struct pvo_entry *pvo; 1390 int ptegidx; 1391 1392 va &= ~ADDR_POFF; 1393 ptegidx = va_to_pteg(pm, va); 1394 1395 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1396 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1397 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 1398 panic("pmap_pvo_find_va: invalid pvo %p on " 1399 "list %#x (%p)", pvo, ptegidx, 1400 &pmap_pvo_table[ptegidx]); 1401 #endif 1402 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1403 if (pteidx_p) 1404 *pteidx_p = pmap_pvo_pte_index(pvo, ptegidx); 1405 return pvo; 1406 } 1407 } 1408 return NULL; 1409 } 1410 1411 #if defined(DEBUG) || defined(PMAPCHECK) 1412 void 1413 pmap_pvo_check(const struct pvo_entry *pvo) 1414 { 1415 struct pvo_head *pvo_head; 1416 struct pvo_entry *pvo0; 1417 volatile struct pte *pt; 1418 int failed = 0; 1419 1420 if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH) 1421 panic("pmap_pvo_check: pvo %p: invalid address", pvo); 1422 1423 if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) { 1424 printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n", 1425 pvo, pvo->pvo_pmap); 1426 failed = 1; 1427 } 1428 1429 if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH || 1430 (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) { 1431 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1432 pvo, TAILQ_NEXT(pvo, pvo_olink)); 1433 failed = 1; 1434 } 1435 1436 if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH || 1437 (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) { 1438 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1439 pvo, LIST_NEXT(pvo, pvo_vlink)); 1440 failed = 1; 1441 } 1442 1443 if (pvo->pvo_vaddr & PVO_MANAGED) { 1444 pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL); 1445 } else { 1446 if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) { 1447 printf("pmap_pvo_check: pvo %p: non kernel address " 1448 "on kernel unmanaged list\n", pvo); 1449 failed = 1; 1450 } 1451 pvo_head = &pmap_pvo_kunmanaged; 1452 } 1453 LIST_FOREACH(pvo0, pvo_head, pvo_vlink) { 1454 if (pvo0 == pvo) 1455 break; 1456 } 1457 if (pvo0 == NULL) { 1458 printf("pmap_pvo_check: pvo %p: not present " 1459 "on its vlist head %p\n", pvo, pvo_head); 1460 failed = 1; 1461 } 1462 if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) { 1463 printf("pmap_pvo_check: pvo %p: not present " 1464 "on its olist head\n", pvo); 1465 failed = 1; 1466 } 1467 pt = pmap_pvo_to_pte(pvo, -1); 1468 if (pt == NULL) { 1469 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1470 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1471 "no PTE\n", pvo); 1472 failed = 1; 1473 } 1474 } else { 1475 if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] || 1476 (uintptr_t) pt >= 1477 (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) { 1478 printf("pmap_pvo_check: pvo %p: pte %p not in " 1479 "pteg table\n", pvo, pt); 1480 failed = 1; 1481 } 1482 if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) { 1483 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1484 "no PTE\n", pvo); 1485 failed = 1; 1486 } 1487 if (pvo->pvo_pte.pte_hi != pt->pte_hi) { 1488 printf("pmap_pvo_check: pvo %p: pte_hi differ: " 1489 "%#x/%#x\n", pvo, (unsigned int) pvo->pvo_pte.pte_hi, (unsigned int) pt->pte_hi); 1490 failed = 1; 1491 } 1492 if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) & 1493 (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) { 1494 printf("pmap_pvo_check: pvo %p: pte_lo differ: " 1495 "%#x/%#x\n", pvo, 1496 (unsigned int) (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)), 1497 (unsigned int) (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN))); 1498 failed = 1; 1499 } 1500 if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) { 1501 printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#lx" 1502 " doesn't not match PVO's VA %#lx\n", 1503 pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo)); 1504 failed = 1; 1505 } 1506 if (failed) 1507 pmap_pte_print(pt); 1508 } 1509 if (failed) 1510 panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo, 1511 pvo->pvo_pmap); 1512 } 1513 #endif /* DEBUG || PMAPCHECK */ 1514 1515 /* 1516 * Search the PVO table looking for a non-wired entry. 1517 * If we find one, remove it and return it. 1518 */ 1519 1520 struct pvo_entry * 1521 pmap_pvo_reclaim(struct pmap *pm) 1522 { 1523 struct pvo_tqhead *pvoh; 1524 struct pvo_entry *pvo; 1525 uint32_t idx, endidx; 1526 1527 endidx = pmap_pvo_reclaim_nextidx; 1528 for (idx = (endidx + 1) & pmap_pteg_mask; idx != endidx; 1529 idx = (idx + 1) & pmap_pteg_mask) { 1530 pvoh = &pmap_pvo_table[idx]; 1531 TAILQ_FOREACH(pvo, pvoh, pvo_olink) { 1532 if ((pvo->pvo_vaddr & PVO_WIRED) == 0) { 1533 pmap_pvo_remove(pvo, -1, NULL); 1534 pmap_pvo_reclaim_nextidx = idx; 1535 PMAPCOUNT(pvos_reclaimed); 1536 return pvo; 1537 } 1538 } 1539 } 1540 return NULL; 1541 } 1542 1543 /* 1544 * This returns whether this is the first mapping of a page. 1545 */ 1546 int 1547 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head, 1548 vaddr_t va, paddr_t pa, register_t pte_lo, int flags) 1549 { 1550 struct pvo_entry *pvo; 1551 struct pvo_tqhead *pvoh; 1552 register_t msr; 1553 int ptegidx; 1554 int i; 1555 int poolflags = PR_NOWAIT; 1556 1557 /* 1558 * Compute the PTE Group index. 1559 */ 1560 va &= ~ADDR_POFF; 1561 ptegidx = va_to_pteg(pm, va); 1562 1563 msr = pmap_interrupts_off(); 1564 1565 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1566 if (pmap_pvo_remove_depth > 0) 1567 panic("pmap_pvo_enter: called while pmap_pvo_remove active!"); 1568 if (++pmap_pvo_enter_depth > 1) 1569 panic("pmap_pvo_enter: called recursively!"); 1570 #endif 1571 1572 /* 1573 * Remove any existing mapping for this page. Reuse the 1574 * pvo entry if there a mapping. 1575 */ 1576 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1577 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1578 #ifdef DEBUG 1579 if ((pmapdebug & PMAPDEBUG_PVOENTER) && 1580 ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) & 1581 ~(PTE_REF|PTE_CHG)) == 0 && 1582 va < VM_MIN_KERNEL_ADDRESS) { 1583 printf("pmap_pvo_enter: pvo %p: dup %#x/%#lx\n", 1584 pvo, (unsigned int) pvo->pvo_pte.pte_lo, (unsigned int) pte_lo|pa); 1585 printf("pmap_pvo_enter: pte_hi=%#x sr=%#x\n", 1586 (unsigned int) pvo->pvo_pte.pte_hi, 1587 (unsigned int) pm->pm_sr[va >> ADDR_SR_SHFT]); 1588 pmap_pte_print(pmap_pvo_to_pte(pvo, -1)); 1589 #ifdef DDBX 1590 Debugger(); 1591 #endif 1592 } 1593 #endif 1594 PMAPCOUNT(mappings_replaced); 1595 pmap_pvo_remove(pvo, -1, NULL); 1596 break; 1597 } 1598 } 1599 1600 /* 1601 * If we aren't overwriting an mapping, try to allocate 1602 */ 1603 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1604 --pmap_pvo_enter_depth; 1605 #endif 1606 pmap_interrupts_restore(msr); 1607 if (pvo) { 1608 pmap_pvo_free(pvo); 1609 } 1610 pvo = pool_get(pl, poolflags); 1611 1612 #ifdef DEBUG 1613 /* 1614 * Exercise pmap_pvo_reclaim() a little. 1615 */ 1616 if (pvo && (flags & PMAP_CANFAIL) != 0 && 1617 pmap_pvo_reclaim_debugctr++ > 0x1000 && 1618 (pmap_pvo_reclaim_debugctr & 0xff) == 0) { 1619 pool_put(pl, pvo); 1620 pvo = NULL; 1621 } 1622 #endif 1623 1624 msr = pmap_interrupts_off(); 1625 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1626 ++pmap_pvo_enter_depth; 1627 #endif 1628 if (pvo == NULL) { 1629 pvo = pmap_pvo_reclaim(pm); 1630 if (pvo == NULL) { 1631 if ((flags & PMAP_CANFAIL) == 0) 1632 panic("pmap_pvo_enter: failed"); 1633 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1634 pmap_pvo_enter_depth--; 1635 #endif 1636 PMAPCOUNT(pvos_failed); 1637 pmap_interrupts_restore(msr); 1638 return ENOMEM; 1639 } 1640 } 1641 1642 pvo->pvo_vaddr = va; 1643 pvo->pvo_pmap = pm; 1644 pvo->pvo_vaddr &= ~ADDR_POFF; 1645 if (flags & VM_PROT_EXECUTE) { 1646 PMAPCOUNT(exec_mappings); 1647 pvo_set_exec(pvo); 1648 } 1649 if (flags & PMAP_WIRED) 1650 pvo->pvo_vaddr |= PVO_WIRED; 1651 if (pvo_head != &pmap_pvo_kunmanaged) { 1652 pvo->pvo_vaddr |= PVO_MANAGED; 1653 PMAPCOUNT(mappings); 1654 } else { 1655 PMAPCOUNT(kernel_mappings); 1656 } 1657 pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo); 1658 1659 LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink); 1660 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1661 pvo->pvo_pmap->pm_stats.wired_count++; 1662 pvo->pvo_pmap->pm_stats.resident_count++; 1663 #if defined(DEBUG) 1664 if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) 1665 DPRINTFN(PVOENTER, 1666 ("pmap_pvo_enter: pvo %p: pm %p va %#lx pa %#lx\n", 1667 pvo, pm, va, pa)); 1668 #endif 1669 1670 /* 1671 * We hope this succeeds but it isn't required. 1672 */ 1673 pvoh = &pmap_pvo_table[ptegidx]; 1674 i = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 1675 if (i >= 0) { 1676 PVO_PTEGIDX_SET(pvo, i); 1677 PVO_WHERE(pvo, ENTER_INSERT); 1678 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 1679 ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]); 1680 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 1681 } else { 1682 /* 1683 * Since we didn't have room for this entry (which makes it 1684 * and evicted entry), place it at the head of the list. 1685 */ 1686 TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink); 1687 PMAPCOUNT(ptes_evicted); 1688 pm->pm_evictions++; 1689 /* 1690 * If this is a kernel page, make sure it's active. 1691 */ 1692 if (pm == pmap_kernel()) { 1693 i = pmap_pte_spill(pm, va, FALSE); 1694 KASSERT(i); 1695 } 1696 } 1697 PMAP_PVO_CHECK(pvo); /* sanity check */ 1698 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1699 pmap_pvo_enter_depth--; 1700 #endif 1701 pmap_interrupts_restore(msr); 1702 return 0; 1703 } 1704 1705 void 1706 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx, struct pvo_head *pvol) 1707 { 1708 volatile struct pte *pt; 1709 int ptegidx; 1710 1711 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1712 if (++pmap_pvo_remove_depth > 1) 1713 panic("pmap_pvo_remove: called recursively!"); 1714 #endif 1715 1716 /* 1717 * If we haven't been supplied the ptegidx, calculate it. 1718 */ 1719 if (pteidx == -1) { 1720 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1721 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1722 } else { 1723 ptegidx = pteidx >> 3; 1724 if (pvo->pvo_pte.pte_hi & PTE_HID) 1725 ptegidx ^= pmap_pteg_mask; 1726 } 1727 PMAP_PVO_CHECK(pvo); /* sanity check */ 1728 1729 /* 1730 * If there is an active pte entry, we need to deactivate it 1731 * (and save the ref & chg bits). 1732 */ 1733 pt = pmap_pvo_to_pte(pvo, pteidx); 1734 if (pt != NULL) { 1735 pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 1736 PVO_WHERE(pvo, REMOVE); 1737 PVO_PTEGIDX_CLR(pvo); 1738 PMAPCOUNT(ptes_removed); 1739 } else { 1740 KASSERT(pvo->pvo_pmap->pm_evictions > 0); 1741 pvo->pvo_pmap->pm_evictions--; 1742 } 1743 1744 /* 1745 * Account for executable mappings. 1746 */ 1747 if (PVO_ISEXECUTABLE(pvo)) 1748 pvo_clear_exec(pvo); 1749 1750 /* 1751 * Update our statistics. 1752 */ 1753 pvo->pvo_pmap->pm_stats.resident_count--; 1754 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1755 pvo->pvo_pmap->pm_stats.wired_count--; 1756 1757 /* 1758 * Save the REF/CHG bits into their cache if the page is managed. 1759 */ 1760 if (pvo->pvo_vaddr & PVO_MANAGED) { 1761 register_t ptelo = pvo->pvo_pte.pte_lo; 1762 struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN); 1763 1764 if (pg != NULL) { 1765 /* 1766 * If this page was changed and it is mapped exec, 1767 * invalidate it. 1768 */ 1769 if ((ptelo & PTE_CHG) && 1770 (pmap_attr_fetch(pg) & PTE_EXEC)) { 1771 struct pvo_head *pvoh = vm_page_to_pvoh(pg); 1772 if (LIST_EMPTY(pvoh)) { 1773 DPRINTFN(EXEC, ("[pmap_pvo_remove: " 1774 "%#lx: clear-exec]\n", 1775 VM_PAGE_TO_PHYS(pg))); 1776 pmap_attr_clear(pg, PTE_EXEC); 1777 PMAPCOUNT(exec_uncached_pvo_remove); 1778 } else { 1779 DPRINTFN(EXEC, ("[pmap_pvo_remove: " 1780 "%#lx: syncicache]\n", 1781 VM_PAGE_TO_PHYS(pg))); 1782 pmap_syncicache(VM_PAGE_TO_PHYS(pg), 1783 PAGE_SIZE); 1784 PMAPCOUNT(exec_synced_pvo_remove); 1785 } 1786 } 1787 1788 pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG)); 1789 } 1790 PMAPCOUNT(unmappings); 1791 } else { 1792 PMAPCOUNT(kernel_unmappings); 1793 } 1794 1795 /* 1796 * Remove the PVO from its lists and return it to the pool. 1797 */ 1798 LIST_REMOVE(pvo, pvo_vlink); 1799 TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink); 1800 if (pvol) { 1801 LIST_INSERT_HEAD(pvol, pvo, pvo_vlink); 1802 } 1803 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1804 pmap_pvo_remove_depth--; 1805 #endif 1806 } 1807 1808 void 1809 pmap_pvo_free(struct pvo_entry *pvo) 1810 { 1811 1812 pool_put(pvo->pvo_vaddr & PVO_MANAGED ? &pmap_mpvo_pool : 1813 &pmap_upvo_pool, pvo); 1814 } 1815 1816 void 1817 pmap_pvo_free_list(struct pvo_head *pvol) 1818 { 1819 struct pvo_entry *pvo, *npvo; 1820 1821 for (pvo = LIST_FIRST(pvol); pvo != NULL; pvo = npvo) { 1822 npvo = LIST_NEXT(pvo, pvo_vlink); 1823 LIST_REMOVE(pvo, pvo_vlink); 1824 pmap_pvo_free(pvo); 1825 } 1826 } 1827 1828 /* 1829 * Mark a mapping as executable. 1830 * If this is the first executable mapping in the segment, 1831 * clear the noexec flag. 1832 */ 1833 STATIC void 1834 pvo_set_exec(struct pvo_entry *pvo) 1835 { 1836 struct pmap *pm = pvo->pvo_pmap; 1837 1838 if (pm == pmap_kernel() || PVO_ISEXECUTABLE(pvo)) { 1839 return; 1840 } 1841 pvo->pvo_vaddr |= PVO_EXECUTABLE; 1842 #ifdef PPC_OEA 1843 { 1844 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1845 if (pm->pm_exec[sr]++ == 0) { 1846 pm->pm_sr[sr] &= ~SR_NOEXEC; 1847 } 1848 } 1849 #endif 1850 } 1851 1852 /* 1853 * Mark a mapping as non-executable. 1854 * If this was the last executable mapping in the segment, 1855 * set the noexec flag. 1856 */ 1857 STATIC void 1858 pvo_clear_exec(struct pvo_entry *pvo) 1859 { 1860 struct pmap *pm = pvo->pvo_pmap; 1861 1862 if (pm == pmap_kernel() || !PVO_ISEXECUTABLE(pvo)) { 1863 return; 1864 } 1865 pvo->pvo_vaddr &= ~PVO_EXECUTABLE; 1866 #ifdef PPC_OEA 1867 { 1868 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1869 if (--pm->pm_exec[sr] == 0) { 1870 pm->pm_sr[sr] |= SR_NOEXEC; 1871 } 1872 } 1873 #endif 1874 } 1875 1876 /* 1877 * Insert physical page at pa into the given pmap at virtual address va. 1878 */ 1879 int 1880 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags) 1881 { 1882 struct mem_region *mp; 1883 struct pvo_head *pvo_head; 1884 struct vm_page *pg; 1885 struct pool *pl; 1886 register_t pte_lo; 1887 int error; 1888 u_int pvo_flags; 1889 u_int was_exec = 0; 1890 1891 if (__predict_false(!pmap_initialized)) { 1892 pvo_head = &pmap_pvo_kunmanaged; 1893 pl = &pmap_upvo_pool; 1894 pvo_flags = 0; 1895 pg = NULL; 1896 was_exec = PTE_EXEC; 1897 } else { 1898 pvo_head = pa_to_pvoh(pa, &pg); 1899 pl = &pmap_mpvo_pool; 1900 pvo_flags = PVO_MANAGED; 1901 } 1902 1903 DPRINTFN(ENTER, 1904 ("pmap_enter(%p, 0x%lx, 0x%lx, 0x%x, 0x%x):", 1905 pm, va, pa, prot, flags)); 1906 1907 /* 1908 * If this is a managed page, and it's the first reference to the 1909 * page clear the execness of the page. Otherwise fetch the execness. 1910 */ 1911 if (pg != NULL) 1912 was_exec = pmap_attr_fetch(pg) & PTE_EXEC; 1913 1914 DPRINTFN(ENTER, (" was_exec=%d", was_exec)); 1915 1916 /* 1917 * Assume the page is cache inhibited and access is guarded unless 1918 * it's in our available memory array. If it is in the memory array, 1919 * asssume it's in memory coherent memory. 1920 */ 1921 pte_lo = PTE_IG; 1922 if ((flags & PMAP_NC) == 0) { 1923 for (mp = mem; mp->size; mp++) { 1924 if (pa >= mp->start && pa < mp->start + mp->size) { 1925 pte_lo = PTE_M; 1926 break; 1927 } 1928 } 1929 } 1930 1931 if (prot & VM_PROT_WRITE) 1932 pte_lo |= PTE_BW; 1933 else 1934 pte_lo |= PTE_BR; 1935 1936 /* 1937 * If this was in response to a fault, "pre-fault" the PTE's 1938 * changed/referenced bit appropriately. 1939 */ 1940 if (flags & VM_PROT_WRITE) 1941 pte_lo |= PTE_CHG; 1942 if (flags & VM_PROT_ALL) 1943 pte_lo |= PTE_REF; 1944 1945 /* 1946 * We need to know if this page can be executable 1947 */ 1948 flags |= (prot & VM_PROT_EXECUTE); 1949 1950 /* 1951 * Record mapping for later back-translation and pte spilling. 1952 * This will overwrite any existing mapping. 1953 */ 1954 error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags); 1955 1956 /* 1957 * Flush the real page from the instruction cache if this page is 1958 * mapped executable and cacheable and has not been flushed since 1959 * the last time it was modified. 1960 */ 1961 if (error == 0 && 1962 (flags & VM_PROT_EXECUTE) && 1963 (pte_lo & PTE_I) == 0 && 1964 was_exec == 0) { 1965 DPRINTFN(ENTER, (" syncicache")); 1966 PMAPCOUNT(exec_synced); 1967 pmap_syncicache(pa, PAGE_SIZE); 1968 if (pg != NULL) { 1969 pmap_attr_save(pg, PTE_EXEC); 1970 PMAPCOUNT(exec_cached); 1971 #if defined(DEBUG) || defined(PMAPDEBUG) 1972 if (pmapdebug & PMAPDEBUG_ENTER) 1973 printf(" marked-as-exec"); 1974 else if (pmapdebug & PMAPDEBUG_EXEC) 1975 printf("[pmap_enter: %#lx: marked-as-exec]\n", 1976 VM_PAGE_TO_PHYS(pg)); 1977 1978 #endif 1979 } 1980 } 1981 1982 DPRINTFN(ENTER, (": error=%d\n", error)); 1983 1984 return error; 1985 } 1986 1987 void 1988 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot) 1989 { 1990 struct mem_region *mp; 1991 register_t pte_lo; 1992 int error; 1993 1994 if (va < VM_MIN_KERNEL_ADDRESS) 1995 panic("pmap_kenter_pa: attempt to enter " 1996 "non-kernel address %#lx!", va); 1997 1998 DPRINTFN(KENTER, 1999 ("pmap_kenter_pa(%#lx,%#lx,%#x)\n", va, pa, prot)); 2000 2001 /* 2002 * Assume the page is cache inhibited and access is guarded unless 2003 * it's in our available memory array. If it is in the memory array, 2004 * asssume it's in memory coherent memory. 2005 */ 2006 pte_lo = PTE_IG; 2007 if ((prot & PMAP_NC) == 0) { 2008 for (mp = mem; mp->size; mp++) { 2009 if (pa >= mp->start && pa < mp->start + mp->size) { 2010 pte_lo = PTE_M; 2011 break; 2012 } 2013 } 2014 } 2015 2016 if (prot & VM_PROT_WRITE) 2017 pte_lo |= PTE_BW; 2018 else 2019 pte_lo |= PTE_BR; 2020 2021 /* 2022 * We don't care about REF/CHG on PVOs on the unmanaged list. 2023 */ 2024 error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool, 2025 &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED); 2026 2027 if (error != 0) 2028 panic("pmap_kenter_pa: failed to enter va %#lx pa %#lx: %d", 2029 va, pa, error); 2030 } 2031 2032 void 2033 pmap_kremove(vaddr_t va, vsize_t len) 2034 { 2035 if (va < VM_MIN_KERNEL_ADDRESS) 2036 panic("pmap_kremove: attempt to remove " 2037 "non-kernel address %#lx!", va); 2038 2039 DPRINTFN(KREMOVE,("pmap_kremove(%#lx,%#lx)\n", va, len)); 2040 pmap_remove(pmap_kernel(), va, va + len); 2041 } 2042 2043 /* 2044 * Remove the given range of mapping entries. 2045 */ 2046 void 2047 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva) 2048 { 2049 struct pvo_head pvol; 2050 struct pvo_entry *pvo; 2051 register_t msr; 2052 int pteidx; 2053 2054 LIST_INIT(&pvol); 2055 msr = pmap_interrupts_off(); 2056 for (; va < endva; va += PAGE_SIZE) { 2057 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2058 if (pvo != NULL) { 2059 pmap_pvo_remove(pvo, pteidx, &pvol); 2060 } 2061 } 2062 pmap_interrupts_restore(msr); 2063 pmap_pvo_free_list(&pvol); 2064 } 2065 2066 /* 2067 * Get the physical page address for the given pmap/virtual address. 2068 */ 2069 boolean_t 2070 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap) 2071 { 2072 struct pvo_entry *pvo; 2073 register_t msr; 2074 2075 /* 2076 * If this is a kernel pmap lookup, also check the battable 2077 * and if we get a hit, translate the VA to a PA using the 2078 * BAT entries. Don't check for VM_MAX_KERNEL_ADDRESS is 2079 * that will wrap back to 0. 2080 */ 2081 if (pm == pmap_kernel() && 2082 (va < VM_MIN_KERNEL_ADDRESS || 2083 (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) { 2084 KASSERT((va >> ADDR_SR_SHFT) != USER_SR); 2085 if ((MFPVR() >> 16) != MPC601) { 2086 register_t batu = battable[va >> ADDR_SR_SHFT].batu; 2087 if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) { 2088 register_t batl = 2089 battable[va >> ADDR_SR_SHFT].batl; 2090 register_t mask = 2091 (~(batu & BAT_BL) << 15) & ~0x1ffffL; 2092 if (pap) 2093 *pap = (batl & mask) | (va & ~mask); 2094 return TRUE; 2095 } 2096 } else { 2097 register_t batu = battable[va >> 23].batu; 2098 register_t batl = battable[va >> 23].batl; 2099 register_t sr = iosrtable[va >> ADDR_SR_SHFT]; 2100 if (BAT601_VALID_P(batl) && 2101 BAT601_VA_MATCH_P(batu, batl, va)) { 2102 register_t mask = 2103 (~(batl & BAT601_BSM) << 17) & ~0x1ffffL; 2104 if (pap) 2105 *pap = (batl & mask) | (va & ~mask); 2106 return TRUE; 2107 } else if (SR601_VALID_P(sr) && 2108 SR601_PA_MATCH_P(sr, va)) { 2109 if (pap) 2110 *pap = va; 2111 return TRUE; 2112 } 2113 } 2114 return FALSE; 2115 } 2116 2117 msr = pmap_interrupts_off(); 2118 pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL); 2119 if (pvo != NULL) { 2120 PMAP_PVO_CHECK(pvo); /* sanity check */ 2121 if (pap) 2122 *pap = (pvo->pvo_pte.pte_lo & PTE_RPGN) 2123 | (va & ADDR_POFF); 2124 } 2125 pmap_interrupts_restore(msr); 2126 return pvo != NULL; 2127 } 2128 2129 /* 2130 * Lower the protection on the specified range of this pmap. 2131 */ 2132 void 2133 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot) 2134 { 2135 struct pvo_entry *pvo; 2136 volatile struct pte *pt; 2137 register_t msr; 2138 int pteidx; 2139 2140 /* 2141 * Since this routine only downgrades protection, we should 2142 * always be called with at least one bit not set. 2143 */ 2144 KASSERT(prot != VM_PROT_ALL); 2145 2146 /* 2147 * If there is no protection, this is equivalent to 2148 * remove the pmap from the pmap. 2149 */ 2150 if ((prot & VM_PROT_READ) == 0) { 2151 pmap_remove(pm, va, endva); 2152 return; 2153 } 2154 2155 msr = pmap_interrupts_off(); 2156 for (; va < endva; va += PAGE_SIZE) { 2157 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2158 if (pvo == NULL) 2159 continue; 2160 PMAP_PVO_CHECK(pvo); /* sanity check */ 2161 2162 /* 2163 * Revoke executable if asked to do so. 2164 */ 2165 if ((prot & VM_PROT_EXECUTE) == 0) 2166 pvo_clear_exec(pvo); 2167 2168 #if 0 2169 /* 2170 * If the page is already read-only, no change 2171 * needs to be made. 2172 */ 2173 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) 2174 continue; 2175 #endif 2176 /* 2177 * Grab the PTE pointer before we diddle with 2178 * the cached PTE copy. 2179 */ 2180 pt = pmap_pvo_to_pte(pvo, pteidx); 2181 /* 2182 * Change the protection of the page. 2183 */ 2184 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2185 pvo->pvo_pte.pte_lo |= PTE_BR; 2186 2187 /* 2188 * If the PVO is in the page table, update 2189 * that pte at well. 2190 */ 2191 if (pt != NULL) { 2192 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2193 PVO_WHERE(pvo, PMAP_PROTECT); 2194 PMAPCOUNT(ptes_changed); 2195 } 2196 2197 PMAP_PVO_CHECK(pvo); /* sanity check */ 2198 } 2199 pmap_interrupts_restore(msr); 2200 } 2201 2202 void 2203 pmap_unwire(pmap_t pm, vaddr_t va) 2204 { 2205 struct pvo_entry *pvo; 2206 register_t msr; 2207 2208 msr = pmap_interrupts_off(); 2209 pvo = pmap_pvo_find_va(pm, va, NULL); 2210 if (pvo != NULL) { 2211 if (pvo->pvo_vaddr & PVO_WIRED) { 2212 pvo->pvo_vaddr &= ~PVO_WIRED; 2213 pm->pm_stats.wired_count--; 2214 } 2215 PMAP_PVO_CHECK(pvo); /* sanity check */ 2216 } 2217 pmap_interrupts_restore(msr); 2218 } 2219 2220 /* 2221 * Lower the protection on the specified physical page. 2222 */ 2223 void 2224 pmap_page_protect(struct vm_page *pg, vm_prot_t prot) 2225 { 2226 struct pvo_head *pvo_head, pvol; 2227 struct pvo_entry *pvo, *next_pvo; 2228 volatile struct pte *pt; 2229 register_t msr; 2230 2231 KASSERT(prot != VM_PROT_ALL); 2232 LIST_INIT(&pvol); 2233 msr = pmap_interrupts_off(); 2234 2235 /* 2236 * When UVM reuses a page, it does a pmap_page_protect with 2237 * VM_PROT_NONE. At that point, we can clear the exec flag 2238 * since we know the page will have different contents. 2239 */ 2240 if ((prot & VM_PROT_READ) == 0) { 2241 DPRINTFN(EXEC, ("[pmap_page_protect: %#lx: clear-exec]\n", 2242 VM_PAGE_TO_PHYS(pg))); 2243 if (pmap_attr_fetch(pg) & PTE_EXEC) { 2244 PMAPCOUNT(exec_uncached_page_protect); 2245 pmap_attr_clear(pg, PTE_EXEC); 2246 } 2247 } 2248 2249 pvo_head = vm_page_to_pvoh(pg); 2250 for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) { 2251 next_pvo = LIST_NEXT(pvo, pvo_vlink); 2252 PMAP_PVO_CHECK(pvo); /* sanity check */ 2253 2254 /* 2255 * Downgrading to no mapping at all, we just remove the entry. 2256 */ 2257 if ((prot & VM_PROT_READ) == 0) { 2258 pmap_pvo_remove(pvo, -1, &pvol); 2259 continue; 2260 } 2261 2262 /* 2263 * If EXEC permission is being revoked, just clear the 2264 * flag in the PVO. 2265 */ 2266 if ((prot & VM_PROT_EXECUTE) == 0) 2267 pvo_clear_exec(pvo); 2268 2269 /* 2270 * If this entry is already RO, don't diddle with the 2271 * page table. 2272 */ 2273 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) { 2274 PMAP_PVO_CHECK(pvo); 2275 continue; 2276 } 2277 2278 /* 2279 * Grab the PTE before the we diddle the bits so 2280 * pvo_to_pte can verify the pte contents are as 2281 * expected. 2282 */ 2283 pt = pmap_pvo_to_pte(pvo, -1); 2284 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2285 pvo->pvo_pte.pte_lo |= PTE_BR; 2286 if (pt != NULL) { 2287 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2288 PVO_WHERE(pvo, PMAP_PAGE_PROTECT); 2289 PMAPCOUNT(ptes_changed); 2290 } 2291 PMAP_PVO_CHECK(pvo); /* sanity check */ 2292 } 2293 pmap_interrupts_restore(msr); 2294 pmap_pvo_free_list(&pvol); 2295 } 2296 2297 /* 2298 * Activate the address space for the specified process. If the process 2299 * is the current process, load the new MMU context. 2300 */ 2301 void 2302 pmap_activate(struct lwp *l) 2303 { 2304 struct pcb *pcb = &l->l_addr->u_pcb; 2305 pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap; 2306 2307 DPRINTFN(ACTIVATE, 2308 ("pmap_activate: lwp %p (curlwp %p)\n", l, curlwp)); 2309 2310 /* 2311 * XXX Normally performed in cpu_fork(). 2312 */ 2313 pcb->pcb_pm = pmap; 2314 2315 /* 2316 * In theory, the SR registers need only be valid on return 2317 * to user space wait to do them there. 2318 */ 2319 if (l == curlwp) { 2320 /* Store pointer to new current pmap. */ 2321 curpm = pmap; 2322 } 2323 } 2324 2325 /* 2326 * Deactivate the specified process's address space. 2327 */ 2328 void 2329 pmap_deactivate(struct lwp *l) 2330 { 2331 } 2332 2333 boolean_t 2334 pmap_query_bit(struct vm_page *pg, int ptebit) 2335 { 2336 struct pvo_entry *pvo; 2337 volatile struct pte *pt; 2338 register_t msr; 2339 2340 if (pmap_attr_fetch(pg) & ptebit) 2341 return TRUE; 2342 2343 msr = pmap_interrupts_off(); 2344 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2345 PMAP_PVO_CHECK(pvo); /* sanity check */ 2346 /* 2347 * See if we saved the bit off. If so cache, it and return 2348 * success. 2349 */ 2350 if (pvo->pvo_pte.pte_lo & ptebit) { 2351 pmap_attr_save(pg, ptebit); 2352 PMAP_PVO_CHECK(pvo); /* sanity check */ 2353 pmap_interrupts_restore(msr); 2354 return TRUE; 2355 } 2356 } 2357 /* 2358 * No luck, now go thru the hard part of looking at the ptes 2359 * themselves. Sync so any pending REF/CHG bits are flushed 2360 * to the PTEs. 2361 */ 2362 SYNC(); 2363 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2364 PMAP_PVO_CHECK(pvo); /* sanity check */ 2365 /* 2366 * See if this pvo have a valid PTE. If so, fetch the 2367 * REF/CHG bits from the valid PTE. If the appropriate 2368 * ptebit is set, cache, it and return success. 2369 */ 2370 pt = pmap_pvo_to_pte(pvo, -1); 2371 if (pt != NULL) { 2372 pmap_pte_synch(pt, &pvo->pvo_pte); 2373 if (pvo->pvo_pte.pte_lo & ptebit) { 2374 pmap_attr_save(pg, ptebit); 2375 PMAP_PVO_CHECK(pvo); /* sanity check */ 2376 pmap_interrupts_restore(msr); 2377 return TRUE; 2378 } 2379 } 2380 } 2381 pmap_interrupts_restore(msr); 2382 return FALSE; 2383 } 2384 2385 boolean_t 2386 pmap_clear_bit(struct vm_page *pg, int ptebit) 2387 { 2388 struct pvo_head *pvoh = vm_page_to_pvoh(pg); 2389 struct pvo_entry *pvo; 2390 volatile struct pte *pt; 2391 register_t msr; 2392 int rv = 0; 2393 2394 msr = pmap_interrupts_off(); 2395 2396 /* 2397 * Fetch the cache value 2398 */ 2399 rv |= pmap_attr_fetch(pg); 2400 2401 /* 2402 * Clear the cached value. 2403 */ 2404 pmap_attr_clear(pg, ptebit); 2405 2406 /* 2407 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we 2408 * can reset the right ones). Note that since the pvo entries and 2409 * list heads are accessed via BAT0 and are never placed in the 2410 * page table, we don't have to worry about further accesses setting 2411 * the REF/CHG bits. 2412 */ 2413 SYNC(); 2414 2415 /* 2416 * For each pvo entry, clear pvo's ptebit. If this pvo have a 2417 * valid PTE. If so, clear the ptebit from the valid PTE. 2418 */ 2419 LIST_FOREACH(pvo, pvoh, pvo_vlink) { 2420 PMAP_PVO_CHECK(pvo); /* sanity check */ 2421 pt = pmap_pvo_to_pte(pvo, -1); 2422 if (pt != NULL) { 2423 /* 2424 * Only sync the PTE if the bit we are looking 2425 * for is not already set. 2426 */ 2427 if ((pvo->pvo_pte.pte_lo & ptebit) == 0) 2428 pmap_pte_synch(pt, &pvo->pvo_pte); 2429 /* 2430 * If the bit we are looking for was already set, 2431 * clear that bit in the pte. 2432 */ 2433 if (pvo->pvo_pte.pte_lo & ptebit) 2434 pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit); 2435 } 2436 rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF); 2437 pvo->pvo_pte.pte_lo &= ~ptebit; 2438 PMAP_PVO_CHECK(pvo); /* sanity check */ 2439 } 2440 pmap_interrupts_restore(msr); 2441 2442 /* 2443 * If we are clearing the modify bit and this page was marked EXEC 2444 * and the user of the page thinks the page was modified, then we 2445 * need to clean it from the icache if it's mapped or clear the EXEC 2446 * bit if it's not mapped. The page itself might not have the CHG 2447 * bit set if the modification was done via DMA to the page. 2448 */ 2449 if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) { 2450 if (LIST_EMPTY(pvoh)) { 2451 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: clear-exec]\n", 2452 VM_PAGE_TO_PHYS(pg))); 2453 pmap_attr_clear(pg, PTE_EXEC); 2454 PMAPCOUNT(exec_uncached_clear_modify); 2455 } else { 2456 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: syncicache]\n", 2457 VM_PAGE_TO_PHYS(pg))); 2458 pmap_syncicache(VM_PAGE_TO_PHYS(pg), PAGE_SIZE); 2459 PMAPCOUNT(exec_synced_clear_modify); 2460 } 2461 } 2462 return (rv & ptebit) != 0; 2463 } 2464 2465 void 2466 pmap_procwr(struct proc *p, vaddr_t va, size_t len) 2467 { 2468 struct pvo_entry *pvo; 2469 size_t offset = va & ADDR_POFF; 2470 int s; 2471 2472 s = splvm(); 2473 while (len > 0) { 2474 size_t seglen = PAGE_SIZE - offset; 2475 if (seglen > len) 2476 seglen = len; 2477 pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL); 2478 if (pvo != NULL && PVO_ISEXECUTABLE(pvo)) { 2479 pmap_syncicache( 2480 (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen); 2481 PMAP_PVO_CHECK(pvo); 2482 } 2483 va += seglen; 2484 len -= seglen; 2485 offset = 0; 2486 } 2487 splx(s); 2488 } 2489 2490 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 2491 void 2492 pmap_pte_print(volatile struct pte *pt) 2493 { 2494 printf("PTE %p: ", pt); 2495 /* High word: */ 2496 printf("0x%08lx: [", pt->pte_hi); 2497 printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i'); 2498 printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-'); 2499 printf("0x%06lx 0x%02lx", 2500 (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT, 2501 pt->pte_hi & PTE_API); 2502 printf(" (va 0x%08lx)] ", pmap_pte_to_va(pt)); 2503 /* Low word: */ 2504 printf(" 0x%08lx: [", pt->pte_lo); 2505 printf("0x%05lx... ", pt->pte_lo >> 12); 2506 printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u'); 2507 printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n'); 2508 printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.'); 2509 printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.'); 2510 printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.'); 2511 printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.'); 2512 switch (pt->pte_lo & PTE_PP) { 2513 case PTE_BR: printf("br]\n"); break; 2514 case PTE_BW: printf("bw]\n"); break; 2515 case PTE_SO: printf("so]\n"); break; 2516 case PTE_SW: printf("sw]\n"); break; 2517 } 2518 } 2519 #endif 2520 2521 #if defined(DDB) 2522 void 2523 pmap_pteg_check(void) 2524 { 2525 volatile struct pte *pt; 2526 int i; 2527 int ptegidx; 2528 u_int p_valid = 0; 2529 u_int s_valid = 0; 2530 u_int invalid = 0; 2531 2532 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2533 for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) { 2534 if (pt->pte_hi & PTE_VALID) { 2535 if (pt->pte_hi & PTE_HID) 2536 s_valid++; 2537 else 2538 p_valid++; 2539 } else 2540 invalid++; 2541 } 2542 } 2543 printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n", 2544 p_valid, p_valid, s_valid, s_valid, 2545 invalid, invalid); 2546 } 2547 2548 void 2549 pmap_print_mmuregs(void) 2550 { 2551 int i; 2552 u_int cpuvers; 2553 #ifndef PPC_OEA64 2554 vaddr_t addr; 2555 register_t soft_sr[16]; 2556 #endif 2557 struct bat soft_ibat[4]; 2558 struct bat soft_dbat[4]; 2559 register_t sdr1; 2560 2561 cpuvers = MFPVR() >> 16; 2562 __asm volatile ("mfsdr1 %0" : "=r"(sdr1)); 2563 #ifndef PPC_OEA64 2564 addr = 0; 2565 for (i = 0; i < 16; i++) { 2566 soft_sr[i] = MFSRIN(addr); 2567 addr += (1 << ADDR_SR_SHFT); 2568 } 2569 #endif 2570 2571 /* read iBAT (601: uBAT) registers */ 2572 __asm volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu)); 2573 __asm volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl)); 2574 __asm volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu)); 2575 __asm volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl)); 2576 __asm volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu)); 2577 __asm volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl)); 2578 __asm volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu)); 2579 __asm volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl)); 2580 2581 2582 if (cpuvers != MPC601) { 2583 /* read dBAT registers */ 2584 __asm volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu)); 2585 __asm volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl)); 2586 __asm volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu)); 2587 __asm volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl)); 2588 __asm volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu)); 2589 __asm volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl)); 2590 __asm volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu)); 2591 __asm volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl)); 2592 } 2593 2594 printf("SDR1:\t0x%lx\n", (long) sdr1); 2595 #ifndef PPC_OEA64 2596 printf("SR[]:\t"); 2597 for (i = 0; i < 4; i++) 2598 printf("0x%08lx, ", soft_sr[i]); 2599 printf("\n\t"); 2600 for ( ; i < 8; i++) 2601 printf("0x%08lx, ", soft_sr[i]); 2602 printf("\n\t"); 2603 for ( ; i < 12; i++) 2604 printf("0x%08lx, ", soft_sr[i]); 2605 printf("\n\t"); 2606 for ( ; i < 16; i++) 2607 printf("0x%08lx, ", soft_sr[i]); 2608 printf("\n"); 2609 #endif 2610 2611 printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i'); 2612 for (i = 0; i < 4; i++) { 2613 printf("0x%08lx 0x%08lx, ", 2614 soft_ibat[i].batu, soft_ibat[i].batl); 2615 if (i == 1) 2616 printf("\n\t"); 2617 } 2618 if (cpuvers != MPC601) { 2619 printf("\ndBAT[]:\t"); 2620 for (i = 0; i < 4; i++) { 2621 printf("0x%08lx 0x%08lx, ", 2622 soft_dbat[i].batu, soft_dbat[i].batl); 2623 if (i == 1) 2624 printf("\n\t"); 2625 } 2626 } 2627 printf("\n"); 2628 } 2629 2630 void 2631 pmap_print_pte(pmap_t pm, vaddr_t va) 2632 { 2633 struct pvo_entry *pvo; 2634 volatile struct pte *pt; 2635 int pteidx; 2636 2637 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2638 if (pvo != NULL) { 2639 pt = pmap_pvo_to_pte(pvo, pteidx); 2640 if (pt != NULL) { 2641 printf("VA %#lx -> %p -> %s %#lx, %#lx\n", 2642 va, pt, 2643 pt->pte_hi & PTE_HID ? "(sec)" : "(pri)", 2644 pt->pte_hi, pt->pte_lo); 2645 } else { 2646 printf("No valid PTE found\n"); 2647 } 2648 } else { 2649 printf("Address not in pmap\n"); 2650 } 2651 } 2652 2653 void 2654 pmap_pteg_dist(void) 2655 { 2656 struct pvo_entry *pvo; 2657 int ptegidx; 2658 int depth; 2659 int max_depth = 0; 2660 unsigned int depths[64]; 2661 2662 memset(depths, 0, sizeof(depths)); 2663 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2664 depth = 0; 2665 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2666 depth++; 2667 } 2668 if (depth > max_depth) 2669 max_depth = depth; 2670 if (depth > 63) 2671 depth = 63; 2672 depths[depth]++; 2673 } 2674 2675 for (depth = 0; depth < 64; depth++) { 2676 printf(" [%2d]: %8u", depth, depths[depth]); 2677 if ((depth & 3) == 3) 2678 printf("\n"); 2679 if (depth == max_depth) 2680 break; 2681 } 2682 if ((depth & 3) != 3) 2683 printf("\n"); 2684 printf("Max depth found was %d\n", max_depth); 2685 } 2686 #endif /* DEBUG */ 2687 2688 #if defined(PMAPCHECK) || defined(DEBUG) 2689 void 2690 pmap_pvo_verify(void) 2691 { 2692 int ptegidx; 2693 int s; 2694 2695 s = splvm(); 2696 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2697 struct pvo_entry *pvo; 2698 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2699 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 2700 panic("pmap_pvo_verify: invalid pvo %p " 2701 "on list %#x", pvo, ptegidx); 2702 pmap_pvo_check(pvo); 2703 } 2704 } 2705 splx(s); 2706 } 2707 #endif /* PMAPCHECK */ 2708 2709 2710 void * 2711 pmap_pool_ualloc(struct pool *pp, int flags) 2712 { 2713 struct pvo_page *pvop; 2714 2715 pvop = SIMPLEQ_FIRST(&pmap_upvop_head); 2716 if (pvop != NULL) { 2717 pmap_upvop_free--; 2718 SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link); 2719 return pvop; 2720 } 2721 if (uvm.page_init_done != TRUE) { 2722 return (void *) uvm_pageboot_alloc(PAGE_SIZE); 2723 } 2724 return pmap_pool_malloc(pp, flags); 2725 } 2726 2727 void * 2728 pmap_pool_malloc(struct pool *pp, int flags) 2729 { 2730 struct pvo_page *pvop; 2731 struct vm_page *pg; 2732 2733 pvop = SIMPLEQ_FIRST(&pmap_mpvop_head); 2734 if (pvop != NULL) { 2735 pmap_mpvop_free--; 2736 SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link); 2737 return pvop; 2738 } 2739 again: 2740 pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE, 2741 UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256); 2742 if (__predict_false(pg == NULL)) { 2743 if (flags & PR_WAITOK) { 2744 uvm_wait("plpg"); 2745 goto again; 2746 } else { 2747 return (0); 2748 } 2749 } 2750 return (void *) VM_PAGE_TO_PHYS(pg); 2751 } 2752 2753 void 2754 pmap_pool_ufree(struct pool *pp, void *va) 2755 { 2756 struct pvo_page *pvop; 2757 #if 0 2758 if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) { 2759 pmap_pool_mfree(va, size, tag); 2760 return; 2761 } 2762 #endif 2763 pvop = va; 2764 SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link); 2765 pmap_upvop_free++; 2766 if (pmap_upvop_free > pmap_upvop_maxfree) 2767 pmap_upvop_maxfree = pmap_upvop_free; 2768 } 2769 2770 void 2771 pmap_pool_mfree(struct pool *pp, void *va) 2772 { 2773 struct pvo_page *pvop; 2774 2775 pvop = va; 2776 SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link); 2777 pmap_mpvop_free++; 2778 if (pmap_mpvop_free > pmap_mpvop_maxfree) 2779 pmap_mpvop_maxfree = pmap_mpvop_free; 2780 #if 0 2781 uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va)); 2782 #endif 2783 } 2784 2785 /* 2786 * This routine in bootstraping to steal to-be-managed memory (which will 2787 * then be unmanaged). We use it to grab from the first 256MB for our 2788 * pmap needs and above 256MB for other stuff. 2789 */ 2790 vaddr_t 2791 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp) 2792 { 2793 vsize_t size; 2794 vaddr_t va; 2795 paddr_t pa = 0; 2796 int npgs, bank; 2797 struct vm_physseg *ps; 2798 2799 if (uvm.page_init_done == TRUE) 2800 panic("pmap_steal_memory: called _after_ bootstrap"); 2801 2802 *vstartp = VM_MIN_KERNEL_ADDRESS; 2803 *vendp = VM_MAX_KERNEL_ADDRESS; 2804 2805 size = round_page(vsize); 2806 npgs = atop(size); 2807 2808 /* 2809 * PA 0 will never be among those given to UVM so we can use it 2810 * to indicate we couldn't steal any memory. 2811 */ 2812 for (ps = vm_physmem, bank = 0; bank < vm_nphysseg; bank++, ps++) { 2813 if (ps->free_list == VM_FREELIST_FIRST256 && 2814 ps->avail_end - ps->avail_start >= npgs) { 2815 pa = ptoa(ps->avail_start); 2816 break; 2817 } 2818 } 2819 2820 if (pa == 0) 2821 panic("pmap_steal_memory: no approriate memory to steal!"); 2822 2823 ps->avail_start += npgs; 2824 ps->start += npgs; 2825 2826 /* 2827 * If we've used up all the pages in the segment, remove it and 2828 * compact the list. 2829 */ 2830 if (ps->avail_start == ps->end) { 2831 /* 2832 * If this was the last one, then a very bad thing has occurred 2833 */ 2834 if (--vm_nphysseg == 0) 2835 panic("pmap_steal_memory: out of memory!"); 2836 2837 printf("pmap_steal_memory: consumed bank %d\n", bank); 2838 for (; bank < vm_nphysseg; bank++, ps++) { 2839 ps[0] = ps[1]; 2840 } 2841 } 2842 2843 va = (vaddr_t) pa; 2844 memset((caddr_t) va, 0, size); 2845 pmap_pages_stolen += npgs; 2846 #ifdef DEBUG 2847 if (pmapdebug && npgs > 1) { 2848 u_int cnt = 0; 2849 for (bank = 0, ps = vm_physmem; bank < vm_nphysseg; bank++, ps++) 2850 cnt += ps->avail_end - ps->avail_start; 2851 printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n", 2852 npgs, pmap_pages_stolen, cnt); 2853 } 2854 #endif 2855 2856 return va; 2857 } 2858 2859 /* 2860 * Find a chuck of memory with right size and alignment. 2861 */ 2862 void * 2863 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end) 2864 { 2865 struct mem_region *mp; 2866 paddr_t s, e; 2867 int i, j; 2868 2869 size = round_page(size); 2870 2871 DPRINTFN(BOOT, 2872 ("pmap_boot_find_memory: size=%lx, alignment=%lx, at_end=%d", 2873 size, alignment, at_end)); 2874 2875 if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0) 2876 panic("pmap_boot_find_memory: invalid alignment %lx", 2877 alignment); 2878 2879 if (at_end) { 2880 if (alignment != PAGE_SIZE) 2881 panic("pmap_boot_find_memory: invalid ending " 2882 "alignment %lx", alignment); 2883 2884 for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) { 2885 s = mp->start + mp->size - size; 2886 if (s >= mp->start && mp->size >= size) { 2887 DPRINTFN(BOOT,(": %lx\n", s)); 2888 DPRINTFN(BOOT, 2889 ("pmap_boot_find_memory: b-avail[%d] start " 2890 "0x%lx size 0x%lx\n", mp - avail, 2891 mp->start, mp->size)); 2892 mp->size -= size; 2893 DPRINTFN(BOOT, 2894 ("pmap_boot_find_memory: a-avail[%d] start " 2895 "0x%lx size 0x%lx\n", mp - avail, 2896 mp->start, mp->size)); 2897 return (void *) s; 2898 } 2899 } 2900 panic("pmap_boot_find_memory: no available memory"); 2901 } 2902 2903 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2904 s = (mp->start + alignment - 1) & ~(alignment-1); 2905 e = s + size; 2906 2907 /* 2908 * Is the calculated region entirely within the region? 2909 */ 2910 if (s < mp->start || e > mp->start + mp->size) 2911 continue; 2912 2913 DPRINTFN(BOOT,(": %lx\n", s)); 2914 if (s == mp->start) { 2915 /* 2916 * If the block starts at the beginning of region, 2917 * adjust the size & start. (the region may now be 2918 * zero in length) 2919 */ 2920 DPRINTFN(BOOT, 2921 ("pmap_boot_find_memory: b-avail[%d] start " 2922 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2923 mp->start += size; 2924 mp->size -= size; 2925 DPRINTFN(BOOT, 2926 ("pmap_boot_find_memory: a-avail[%d] start " 2927 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2928 } else if (e == mp->start + mp->size) { 2929 /* 2930 * If the block starts at the beginning of region, 2931 * adjust only the size. 2932 */ 2933 DPRINTFN(BOOT, 2934 ("pmap_boot_find_memory: b-avail[%d] start " 2935 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2936 mp->size -= size; 2937 DPRINTFN(BOOT, 2938 ("pmap_boot_find_memory: a-avail[%d] start " 2939 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2940 } else { 2941 /* 2942 * Block is in the middle of the region, so we 2943 * have to split it in two. 2944 */ 2945 for (j = avail_cnt; j > i + 1; j--) { 2946 avail[j] = avail[j-1]; 2947 } 2948 DPRINTFN(BOOT, 2949 ("pmap_boot_find_memory: b-avail[%d] start " 2950 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2951 mp[1].start = e; 2952 mp[1].size = mp[0].start + mp[0].size - e; 2953 mp[0].size = s - mp[0].start; 2954 avail_cnt++; 2955 for (; i < avail_cnt; i++) { 2956 DPRINTFN(BOOT, 2957 ("pmap_boot_find_memory: a-avail[%d] " 2958 "start 0x%lx size 0x%lx\n", i, 2959 avail[i].start, avail[i].size)); 2960 } 2961 } 2962 return (void *) s; 2963 } 2964 panic("pmap_boot_find_memory: not enough memory for " 2965 "%lx/%lx allocation?", size, alignment); 2966 } 2967 2968 /* 2969 * This is not part of the defined PMAP interface and is specific to the 2970 * PowerPC architecture. This is called during initppc, before the system 2971 * is really initialized. 2972 */ 2973 void 2974 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend) 2975 { 2976 struct mem_region *mp, tmp; 2977 paddr_t s, e; 2978 psize_t size; 2979 int i, j; 2980 2981 /* 2982 * Get memory. 2983 */ 2984 mem_regions(&mem, &avail); 2985 #if defined(DEBUG) 2986 if (pmapdebug & PMAPDEBUG_BOOT) { 2987 printf("pmap_bootstrap: memory configuration:\n"); 2988 for (mp = mem; mp->size; mp++) { 2989 printf("pmap_bootstrap: mem start 0x%lx size 0x%lx\n", 2990 mp->start, mp->size); 2991 } 2992 for (mp = avail; mp->size; mp++) { 2993 printf("pmap_bootstrap: avail start 0x%lx size 0x%lx\n", 2994 mp->start, mp->size); 2995 } 2996 } 2997 #endif 2998 2999 /* 3000 * Find out how much physical memory we have and in how many chunks. 3001 */ 3002 for (mem_cnt = 0, mp = mem; mp->size; mp++) { 3003 if (mp->start >= pmap_memlimit) 3004 continue; 3005 if (mp->start + mp->size > pmap_memlimit) { 3006 size = pmap_memlimit - mp->start; 3007 physmem += btoc(size); 3008 } else { 3009 physmem += btoc(mp->size); 3010 } 3011 mem_cnt++; 3012 } 3013 3014 /* 3015 * Count the number of available entries. 3016 */ 3017 for (avail_cnt = 0, mp = avail; mp->size; mp++) 3018 avail_cnt++; 3019 3020 /* 3021 * Page align all regions. 3022 */ 3023 kernelstart = trunc_page(kernelstart); 3024 kernelend = round_page(kernelend); 3025 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 3026 s = round_page(mp->start); 3027 mp->size -= (s - mp->start); 3028 mp->size = trunc_page(mp->size); 3029 mp->start = s; 3030 e = mp->start + mp->size; 3031 3032 DPRINTFN(BOOT, 3033 ("pmap_bootstrap: b-avail[%d] start 0x%lx size 0x%lx\n", 3034 i, mp->start, mp->size)); 3035 3036 /* 3037 * Don't allow the end to run beyond our artificial limit 3038 */ 3039 if (e > pmap_memlimit) 3040 e = pmap_memlimit; 3041 3042 /* 3043 * Is this region empty or strange? skip it. 3044 */ 3045 if (e <= s) { 3046 mp->start = 0; 3047 mp->size = 0; 3048 continue; 3049 } 3050 3051 /* 3052 * Does this overlap the beginning of kernel? 3053 * Does extend past the end of the kernel? 3054 */ 3055 else if (s < kernelstart && e > kernelstart) { 3056 if (e > kernelend) { 3057 avail[avail_cnt].start = kernelend; 3058 avail[avail_cnt].size = e - kernelend; 3059 avail_cnt++; 3060 } 3061 mp->size = kernelstart - s; 3062 } 3063 /* 3064 * Check whether this region overlaps the end of the kernel. 3065 */ 3066 else if (s < kernelend && e > kernelend) { 3067 mp->start = kernelend; 3068 mp->size = e - kernelend; 3069 } 3070 /* 3071 * Look whether this regions is completely inside the kernel. 3072 * Nuke it if it does. 3073 */ 3074 else if (s >= kernelstart && e <= kernelend) { 3075 mp->start = 0; 3076 mp->size = 0; 3077 } 3078 /* 3079 * If the user imposed a memory limit, enforce it. 3080 */ 3081 else if (s >= pmap_memlimit) { 3082 mp->start = -PAGE_SIZE; /* let's know why */ 3083 mp->size = 0; 3084 } 3085 else { 3086 mp->start = s; 3087 mp->size = e - s; 3088 } 3089 DPRINTFN(BOOT, 3090 ("pmap_bootstrap: a-avail[%d] start 0x%lx size 0x%lx\n", 3091 i, mp->start, mp->size)); 3092 } 3093 3094 /* 3095 * Move (and uncount) all the null return to the end. 3096 */ 3097 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 3098 if (mp->size == 0) { 3099 tmp = avail[i]; 3100 avail[i] = avail[--avail_cnt]; 3101 avail[avail_cnt] = avail[i]; 3102 } 3103 } 3104 3105 /* 3106 * (Bubble)sort them into asecnding order. 3107 */ 3108 for (i = 0; i < avail_cnt; i++) { 3109 for (j = i + 1; j < avail_cnt; j++) { 3110 if (avail[i].start > avail[j].start) { 3111 tmp = avail[i]; 3112 avail[i] = avail[j]; 3113 avail[j] = tmp; 3114 } 3115 } 3116 } 3117 3118 /* 3119 * Make sure they don't overlap. 3120 */ 3121 for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) { 3122 if (mp[0].start + mp[0].size > mp[1].start) { 3123 mp[0].size = mp[1].start - mp[0].start; 3124 } 3125 DPRINTFN(BOOT, 3126 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 3127 i, mp->start, mp->size)); 3128 } 3129 DPRINTFN(BOOT, 3130 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 3131 i, mp->start, mp->size)); 3132 3133 #ifdef PTEGCOUNT 3134 pmap_pteg_cnt = PTEGCOUNT; 3135 #else /* PTEGCOUNT */ 3136 pmap_pteg_cnt = 0x1000; 3137 3138 while (pmap_pteg_cnt < physmem) 3139 pmap_pteg_cnt <<= 1; 3140 3141 pmap_pteg_cnt >>= 1; 3142 #endif /* PTEGCOUNT */ 3143 3144 /* 3145 * Find suitably aligned memory for PTEG hash table. 3146 */ 3147 size = pmap_pteg_cnt * sizeof(struct pteg); 3148 pmap_pteg_table = pmap_boot_find_memory(size, size, 0); 3149 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3150 if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH) 3151 panic("pmap_bootstrap: pmap_pteg_table end (%p + %lx) > 256MB", 3152 pmap_pteg_table, size); 3153 #endif 3154 3155 memset(__UNVOLATILE(pmap_pteg_table), 0, 3156 pmap_pteg_cnt * sizeof(struct pteg)); 3157 pmap_pteg_mask = pmap_pteg_cnt - 1; 3158 3159 /* 3160 * We cannot do pmap_steal_memory here since UVM hasn't been loaded 3161 * with pages. So we just steal them before giving them to UVM. 3162 */ 3163 size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt; 3164 pmap_pvo_table = pmap_boot_find_memory(size, PAGE_SIZE, 0); 3165 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3166 if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH) 3167 panic("pmap_bootstrap: pmap_pvo_table end (%p + %lx) > 256MB", 3168 pmap_pvo_table, size); 3169 #endif 3170 3171 for (i = 0; i < pmap_pteg_cnt; i++) 3172 TAILQ_INIT(&pmap_pvo_table[i]); 3173 3174 #ifndef MSGBUFADDR 3175 /* 3176 * Allocate msgbuf in high memory. 3177 */ 3178 msgbuf_paddr = 3179 (paddr_t) pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1); 3180 #endif 3181 3182 #ifdef __HAVE_PMAP_PHYSSEG 3183 { 3184 u_int npgs = 0; 3185 for (i = 0, mp = avail; i < avail_cnt; i++, mp++) 3186 npgs += btoc(mp->size); 3187 size = (sizeof(struct pvo_head) + 1) * npgs; 3188 pmap_physseg.pvoh = pmap_boot_find_memory(size, PAGE_SIZE, 0); 3189 pmap_physseg.attrs = (char *) &pmap_physseg.pvoh[npgs]; 3190 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3191 if ((uintptr_t)pmap_physseg.pvoh + size > SEGMENT_LENGTH) 3192 panic("pmap_bootstrap: PVO list end (%p + %lx) > 256MB", 3193 pmap_physseg.pvoh, size); 3194 #endif 3195 } 3196 #endif 3197 3198 for (mp = avail, i = 0; i < avail_cnt; mp++, i++) { 3199 paddr_t pfstart = atop(mp->start); 3200 paddr_t pfend = atop(mp->start + mp->size); 3201 if (mp->size == 0) 3202 continue; 3203 if (mp->start + mp->size <= SEGMENT_LENGTH) { 3204 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3205 VM_FREELIST_FIRST256); 3206 } else if (mp->start >= SEGMENT_LENGTH) { 3207 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3208 VM_FREELIST_DEFAULT); 3209 } else { 3210 pfend = atop(SEGMENT_LENGTH); 3211 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3212 VM_FREELIST_FIRST256); 3213 pfstart = atop(SEGMENT_LENGTH); 3214 pfend = atop(mp->start + mp->size); 3215 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3216 VM_FREELIST_DEFAULT); 3217 } 3218 } 3219 3220 /* 3221 * Make sure kernel vsid is allocated as well as VSID 0. 3222 */ 3223 pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW] 3224 |= 1 << (KERNEL_VSIDBITS % VSID_NBPW); 3225 pmap_vsid_bitmap[0] |= 1; 3226 3227 /* 3228 * Initialize kernel pmap and hardware. 3229 */ 3230 #ifndef PPC_OEA64 3231 for (i = 0; i < 16; i++) { 3232 pmap_kernel()->pm_sr[i] = EMPTY_SEGMENT; 3233 __asm volatile ("mtsrin %0,%1" 3234 :: "r"(EMPTY_SEGMENT), "r"(i << ADDR_SR_SHFT)); 3235 } 3236 3237 pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY; 3238 __asm volatile ("mtsr %0,%1" 3239 :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT)); 3240 #ifdef KERNEL2_SR 3241 pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY; 3242 __asm volatile ("mtsr %0,%1" 3243 :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT)); 3244 #endif 3245 for (i = 0; i < 16; i++) { 3246 if (iosrtable[i] & SR601_T) { 3247 pmap_kernel()->pm_sr[i] = iosrtable[i]; 3248 __asm volatile ("mtsrin %0,%1" 3249 :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT)); 3250 } 3251 } 3252 #endif /* !PPC_OEA64 */ 3253 3254 __asm volatile ("sync; mtsdr1 %0; isync" 3255 :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10))); 3256 tlbia(); 3257 3258 #ifdef ALTIVEC 3259 pmap_use_altivec = cpu_altivec; 3260 #endif 3261 3262 #ifdef DEBUG 3263 if (pmapdebug & PMAPDEBUG_BOOT) { 3264 u_int cnt; 3265 int bank; 3266 char pbuf[9]; 3267 for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) { 3268 cnt += vm_physmem[bank].avail_end - vm_physmem[bank].avail_start; 3269 printf("pmap_bootstrap: vm_physmem[%d]=%#lx-%#lx/%#lx\n", 3270 bank, 3271 ptoa(vm_physmem[bank].avail_start), 3272 ptoa(vm_physmem[bank].avail_end), 3273 ptoa(vm_physmem[bank].avail_end - vm_physmem[bank].avail_start)); 3274 } 3275 format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt)); 3276 printf("pmap_bootstrap: UVM memory = %s (%u pages)\n", 3277 pbuf, cnt); 3278 } 3279 #endif 3280 3281 pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry), 3282 sizeof(struct pvo_entry), 0, 0, "pmap_upvopl", 3283 &pmap_pool_uallocator); 3284 3285 pool_setlowat(&pmap_upvo_pool, 252); 3286 3287 pool_init(&pmap_pool, sizeof(struct pmap), 3288 sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator); 3289 } 3290
Indexes created Sat Oct 25 07:10:08 GMT 2025