pmap.c revision 1.48.30.3
1 /* $NetBSD: pmap.c,v 1.48.30.3 2008/01/19 12:14:41 bouyer 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 * Support for PPC64 Bridge mode added by Sanjay Lal <sanjayl (at) kymasys.com> 10 * of Kyma Systems LLC. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. All advertising materials mentioning features or use of this software 21 * must display the following acknowledgement: 22 * This product includes software developed by the NetBSD 23 * Foundation, Inc. and its contributors. 24 * 4. Neither the name of The NetBSD Foundation nor the names of its 25 * contributors may be used to endorse or promote products derived 26 * from this software without specific prior written permission. 27 * 28 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 29 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 30 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 31 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 32 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 33 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 34 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 35 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 36 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 37 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 38 * POSSIBILITY OF SUCH DAMAGE. 39 */ 40 41 /* 42 * Copyright (C) 1995, 1996 Wolfgang Solfrank. 43 * Copyright (C) 1995, 1996 TooLs GmbH. 44 * All rights reserved. 45 * 46 * Redistribution and use in source and binary forms, with or without 47 * modification, are permitted provided that the following conditions 48 * are met: 49 * 1. Redistributions of source code must retain the above copyright 50 * notice, this list of conditions and the following disclaimer. 51 * 2. Redistributions in binary form must reproduce the above copyright 52 * notice, this list of conditions and the following disclaimer in the 53 * documentation and/or other materials provided with the distribution. 54 * 3. All advertising materials mentioning features or use of this software 55 * must display the following acknowledgement: 56 * This product includes software developed by TooLs GmbH. 57 * 4. The name of TooLs GmbH may not be used to endorse or promote products 58 * derived from this software without specific prior written permission. 59 * 60 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR 61 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 62 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 63 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 64 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 65 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 66 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 67 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 68 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 69 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 70 */ 71 72 #include <sys/cdefs.h> 73 __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.48.30.3 2008/01/19 12:14:41 bouyer Exp $"); 74 75 #include "opt_ppcarch.h" 76 #include "opt_altivec.h" 77 #include "opt_pmap.h" 78 #include <sys/param.h> 79 #include <sys/malloc.h> 80 #include <sys/proc.h> 81 #include <sys/user.h> 82 #include <sys/pool.h> 83 #include <sys/queue.h> 84 #include <sys/device.h> /* for evcnt */ 85 #include <sys/systm.h> 86 #include <sys/atomic.h> 87 88 #if __NetBSD_Version__ < 105010000 89 #include <vm/vm.h> 90 #include <vm/vm_kern.h> 91 #define splvm() splimp() 92 #endif 93 94 #include <uvm/uvm.h> 95 96 #include <machine/pcb.h> 97 #include <machine/powerpc.h> 98 #include <powerpc/spr.h> 99 #include <powerpc/oea/sr_601.h> 100 #include <powerpc/bat.h> 101 #include <powerpc/stdarg.h> 102 103 #if defined(DEBUG) || defined(PMAPCHECK) 104 #define STATIC 105 #else 106 #define STATIC static 107 #endif 108 109 #ifdef ALTIVEC 110 int pmap_use_altivec; 111 #endif 112 113 volatile struct pteg *pmap_pteg_table; 114 unsigned int pmap_pteg_cnt; 115 unsigned int pmap_pteg_mask; 116 #ifdef PMAP_MEMLIMIT 117 paddr_t pmap_memlimit = PMAP_MEMLIMIT; 118 #else 119 paddr_t pmap_memlimit = -PAGE_SIZE; /* there is no limit */ 120 #endif 121 122 struct pmap kernel_pmap_; 123 unsigned int pmap_pages_stolen; 124 u_long pmap_pte_valid; 125 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 126 u_long pmap_pvo_enter_depth; 127 u_long pmap_pvo_remove_depth; 128 #endif 129 130 int physmem; 131 #ifndef MSGBUFADDR 132 extern paddr_t msgbuf_paddr; 133 #endif 134 135 static struct mem_region *mem, *avail; 136 static u_int mem_cnt, avail_cnt; 137 138 #ifdef __HAVE_PMAP_PHYSSEG 139 /* 140 * This is a cache of referenced/modified bits. 141 * Bits herein are shifted by ATTRSHFT. 142 */ 143 #define ATTR_SHFT 4 144 struct pmap_physseg pmap_physseg; 145 #endif 146 147 /* 148 * The following structure is aligned to 32 bytes 149 */ 150 struct pvo_entry { 151 LIST_ENTRY(pvo_entry) pvo_vlink; /* Link to common virt page */ 152 TAILQ_ENTRY(pvo_entry) pvo_olink; /* Link to overflow entry */ 153 struct pte pvo_pte; /* Prebuilt PTE */ 154 pmap_t pvo_pmap; /* ptr to owning pmap */ 155 vaddr_t pvo_vaddr; /* VA of entry */ 156 #define PVO_PTEGIDX_MASK 0x0007 /* which PTEG slot */ 157 #define PVO_PTEGIDX_VALID 0x0008 /* slot is valid */ 158 #define PVO_WIRED 0x0010 /* PVO entry is wired */ 159 #define PVO_MANAGED 0x0020 /* PVO e. for managed page */ 160 #define PVO_EXECUTABLE 0x0040 /* PVO e. for executable page */ 161 #define PVO_WIRED_P(pvo) ((pvo)->pvo_vaddr & PVO_WIRED) 162 #define PVO_MANAGED_P(pvo) ((pvo)->pvo_vaddr & PVO_MANAGED) 163 #define PVO_EXECUTABLE_P(pvo) ((pvo)->pvo_vaddr & PVO_EXECUTABLE) 164 #define PVO_ENTER_INSERT 0 /* PVO has been removed */ 165 #define PVO_SPILL_UNSET 1 /* PVO has been evicted */ 166 #define PVO_SPILL_SET 2 /* PVO has been spilled */ 167 #define PVO_SPILL_INSERT 3 /* PVO has been inserted */ 168 #define PVO_PMAP_PAGE_PROTECT 4 /* PVO has changed */ 169 #define PVO_PMAP_PROTECT 5 /* PVO has changed */ 170 #define PVO_REMOVE 6 /* PVO has been removed */ 171 #define PVO_WHERE_MASK 15 172 #define PVO_WHERE_SHFT 8 173 } __attribute__ ((aligned (32))); 174 #define PVO_VADDR(pvo) ((pvo)->pvo_vaddr & ~ADDR_POFF) 175 #define PVO_PTEGIDX_GET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK) 176 #define PVO_PTEGIDX_ISSET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID) 177 #define PVO_PTEGIDX_CLR(pvo) \ 178 ((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK))) 179 #define PVO_PTEGIDX_SET(pvo,i) \ 180 ((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID)) 181 #define PVO_WHERE(pvo,w) \ 182 ((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \ 183 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT)) 184 185 TAILQ_HEAD(pvo_tqhead, pvo_entry); 186 struct pvo_tqhead *pmap_pvo_table; /* pvo entries by ptegroup index */ 187 struct pvo_head pmap_pvo_kunmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged); /* list of unmanaged pages */ 188 struct pvo_head pmap_pvo_unmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged); /* list of unmanaged pages */ 189 190 struct pool pmap_pool; /* pool for pmap structures */ 191 struct pool pmap_upvo_pool; /* pool for pvo entries for unmanaged pages */ 192 struct pool pmap_mpvo_pool; /* pool for pvo entries for managed pages */ 193 194 /* 195 * We keep a cache of unmanaged pages to be used for pvo entries for 196 * unmanaged pages. 197 */ 198 struct pvo_page { 199 SIMPLEQ_ENTRY(pvo_page) pvop_link; 200 }; 201 SIMPLEQ_HEAD(pvop_head, pvo_page); 202 struct pvop_head pmap_upvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_upvop_head); 203 struct pvop_head pmap_mpvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_mpvop_head); 204 u_long pmap_upvop_free; 205 u_long pmap_upvop_maxfree; 206 u_long pmap_mpvop_free; 207 u_long pmap_mpvop_maxfree; 208 209 STATIC void *pmap_pool_ualloc(struct pool *, int); 210 STATIC void *pmap_pool_malloc(struct pool *, int); 211 212 STATIC void pmap_pool_ufree(struct pool *, void *); 213 STATIC void pmap_pool_mfree(struct pool *, void *); 214 215 static struct pool_allocator pmap_pool_mallocator = { 216 .pa_alloc = pmap_pool_malloc, 217 .pa_free = pmap_pool_mfree, 218 .pa_pagesz = 0, 219 }; 220 221 static struct pool_allocator pmap_pool_uallocator = { 222 .pa_alloc = pmap_pool_ualloc, 223 .pa_free = pmap_pool_ufree, 224 .pa_pagesz = 0, 225 }; 226 227 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 228 void pmap_pte_print(volatile struct pte *); 229 void pmap_pteg_check(void); 230 void pmap_pteg_dist(void); 231 void pmap_print_pte(pmap_t, vaddr_t); 232 void pmap_print_mmuregs(void); 233 #endif 234 235 #if defined(DEBUG) || defined(PMAPCHECK) 236 #ifdef PMAPCHECK 237 int pmapcheck = 1; 238 #else 239 int pmapcheck = 0; 240 #endif 241 void pmap_pvo_verify(void); 242 STATIC void pmap_pvo_check(const struct pvo_entry *); 243 #define PMAP_PVO_CHECK(pvo) \ 244 do { \ 245 if (pmapcheck) \ 246 pmap_pvo_check(pvo); \ 247 } while (0) 248 #else 249 #define PMAP_PVO_CHECK(pvo) do { } while (/*CONSTCOND*/0) 250 #endif 251 STATIC int pmap_pte_insert(int, struct pte *); 252 STATIC int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *, 253 vaddr_t, paddr_t, register_t, int); 254 STATIC void pmap_pvo_remove(struct pvo_entry *, int, struct pvo_head *); 255 STATIC void pmap_pvo_free(struct pvo_entry *); 256 STATIC void pmap_pvo_free_list(struct pvo_head *); 257 STATIC struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *); 258 STATIC volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int); 259 STATIC struct pvo_entry *pmap_pvo_reclaim(struct pmap *); 260 STATIC void pvo_set_exec(struct pvo_entry *); 261 STATIC void pvo_clear_exec(struct pvo_entry *); 262 263 STATIC void tlbia(void); 264 265 STATIC void pmap_release(pmap_t); 266 STATIC void *pmap_boot_find_memory(psize_t, psize_t, int); 267 268 static uint32_t pmap_pvo_reclaim_nextidx; 269 #ifdef DEBUG 270 static int pmap_pvo_reclaim_debugctr; 271 #endif 272 273 #define VSID_NBPW (sizeof(uint32_t) * 8) 274 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW]; 275 276 static int pmap_initialized; 277 278 #if defined(DEBUG) || defined(PMAPDEBUG) 279 #define PMAPDEBUG_BOOT 0x0001 280 #define PMAPDEBUG_PTE 0x0002 281 #define PMAPDEBUG_EXEC 0x0008 282 #define PMAPDEBUG_PVOENTER 0x0010 283 #define PMAPDEBUG_PVOREMOVE 0x0020 284 #define PMAPDEBUG_ACTIVATE 0x0100 285 #define PMAPDEBUG_CREATE 0x0200 286 #define PMAPDEBUG_ENTER 0x1000 287 #define PMAPDEBUG_KENTER 0x2000 288 #define PMAPDEBUG_KREMOVE 0x4000 289 #define PMAPDEBUG_REMOVE 0x8000 290 291 unsigned int pmapdebug = 0; 292 293 # define DPRINTF(x) printf x 294 # define DPRINTFN(n, x) if (pmapdebug & PMAPDEBUG_ ## n) printf x 295 #else 296 # define DPRINTF(x) 297 # define DPRINTFN(n, x) 298 #endif 299 300 301 #ifdef PMAPCOUNTERS 302 #define PMAPCOUNT(ev) ((pmap_evcnt_ ## ev).ev_count++) 303 #define PMAPCOUNT2(ev) ((ev).ev_count++) 304 305 struct evcnt pmap_evcnt_mappings = 306 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 307 "pmap", "pages mapped"); 308 struct evcnt pmap_evcnt_unmappings = 309 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 310 "pmap", "pages unmapped"); 311 312 struct evcnt pmap_evcnt_kernel_mappings = 313 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 314 "pmap", "kernel pages mapped"); 315 struct evcnt pmap_evcnt_kernel_unmappings = 316 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_kernel_mappings, 317 "pmap", "kernel pages unmapped"); 318 319 struct evcnt pmap_evcnt_mappings_replaced = 320 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 321 "pmap", "page mappings replaced"); 322 323 struct evcnt pmap_evcnt_exec_mappings = 324 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 325 "pmap", "exec pages mapped"); 326 struct evcnt pmap_evcnt_exec_cached = 327 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 328 "pmap", "exec pages cached"); 329 330 struct evcnt pmap_evcnt_exec_synced = 331 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 332 "pmap", "exec pages synced"); 333 struct evcnt pmap_evcnt_exec_synced_clear_modify = 334 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 335 "pmap", "exec pages synced (CM)"); 336 struct evcnt pmap_evcnt_exec_synced_pvo_remove = 337 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 338 "pmap", "exec pages synced (PR)"); 339 340 struct evcnt pmap_evcnt_exec_uncached_page_protect = 341 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 342 "pmap", "exec pages uncached (PP)"); 343 struct evcnt pmap_evcnt_exec_uncached_clear_modify = 344 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 345 "pmap", "exec pages uncached (CM)"); 346 struct evcnt pmap_evcnt_exec_uncached_zero_page = 347 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 348 "pmap", "exec pages uncached (ZP)"); 349 struct evcnt pmap_evcnt_exec_uncached_copy_page = 350 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 351 "pmap", "exec pages uncached (CP)"); 352 struct evcnt pmap_evcnt_exec_uncached_pvo_remove = 353 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 354 "pmap", "exec pages uncached (PR)"); 355 356 struct evcnt pmap_evcnt_updates = 357 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 358 "pmap", "updates"); 359 struct evcnt pmap_evcnt_collects = 360 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 361 "pmap", "collects"); 362 struct evcnt pmap_evcnt_copies = 363 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 364 "pmap", "copies"); 365 366 struct evcnt pmap_evcnt_ptes_spilled = 367 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 368 "pmap", "ptes spilled from overflow"); 369 struct evcnt pmap_evcnt_ptes_unspilled = 370 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 371 "pmap", "ptes not spilled"); 372 struct evcnt pmap_evcnt_ptes_evicted = 373 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 374 "pmap", "ptes evicted"); 375 376 struct evcnt pmap_evcnt_ptes_primary[8] = { 377 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 378 "pmap", "ptes added at primary[0]"), 379 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 380 "pmap", "ptes added at primary[1]"), 381 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 382 "pmap", "ptes added at primary[2]"), 383 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 384 "pmap", "ptes added at primary[3]"), 385 386 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 387 "pmap", "ptes added at primary[4]"), 388 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 389 "pmap", "ptes added at primary[5]"), 390 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 391 "pmap", "ptes added at primary[6]"), 392 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 393 "pmap", "ptes added at primary[7]"), 394 }; 395 struct evcnt pmap_evcnt_ptes_secondary[8] = { 396 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 397 "pmap", "ptes added at secondary[0]"), 398 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 399 "pmap", "ptes added at secondary[1]"), 400 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 401 "pmap", "ptes added at secondary[2]"), 402 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 403 "pmap", "ptes added at secondary[3]"), 404 405 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 406 "pmap", "ptes added at secondary[4]"), 407 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 408 "pmap", "ptes added at secondary[5]"), 409 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 410 "pmap", "ptes added at secondary[6]"), 411 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 412 "pmap", "ptes added at secondary[7]"), 413 }; 414 struct evcnt pmap_evcnt_ptes_removed = 415 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 416 "pmap", "ptes removed"); 417 struct evcnt pmap_evcnt_ptes_changed = 418 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 419 "pmap", "ptes changed"); 420 struct evcnt pmap_evcnt_pvos_reclaimed = 421 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 422 "pmap", "pvos reclaimed"); 423 struct evcnt pmap_evcnt_pvos_failed = 424 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 425 "pmap", "pvo allocation failures"); 426 427 /* 428 * From pmap_subr.c 429 */ 430 extern struct evcnt pmap_evcnt_zeroed_pages; 431 extern struct evcnt pmap_evcnt_copied_pages; 432 extern struct evcnt pmap_evcnt_idlezeroed_pages; 433 434 EVCNT_ATTACH_STATIC(pmap_evcnt_mappings); 435 EVCNT_ATTACH_STATIC(pmap_evcnt_mappings_replaced); 436 EVCNT_ATTACH_STATIC(pmap_evcnt_unmappings); 437 438 EVCNT_ATTACH_STATIC(pmap_evcnt_kernel_mappings); 439 EVCNT_ATTACH_STATIC(pmap_evcnt_kernel_unmappings); 440 441 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_mappings); 442 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_cached); 443 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_synced); 444 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_synced_clear_modify); 445 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_synced_pvo_remove); 446 447 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_page_protect); 448 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_clear_modify); 449 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_zero_page); 450 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_copy_page); 451 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_pvo_remove); 452 453 EVCNT_ATTACH_STATIC(pmap_evcnt_zeroed_pages); 454 EVCNT_ATTACH_STATIC(pmap_evcnt_copied_pages); 455 EVCNT_ATTACH_STATIC(pmap_evcnt_idlezeroed_pages); 456 457 EVCNT_ATTACH_STATIC(pmap_evcnt_updates); 458 EVCNT_ATTACH_STATIC(pmap_evcnt_collects); 459 EVCNT_ATTACH_STATIC(pmap_evcnt_copies); 460 461 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_spilled); 462 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_unspilled); 463 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_evicted); 464 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_removed); 465 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_changed); 466 467 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 0); 468 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 1); 469 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 2); 470 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 3); 471 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 4); 472 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 5); 473 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 6); 474 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 7); 475 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 0); 476 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 1); 477 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 2); 478 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 3); 479 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 4); 480 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 5); 481 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 6); 482 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 7); 483 484 EVCNT_ATTACH_STATIC(pmap_evcnt_pvos_reclaimed); 485 EVCNT_ATTACH_STATIC(pmap_evcnt_pvos_failed); 486 #else 487 #define PMAPCOUNT(ev) ((void) 0) 488 #define PMAPCOUNT2(ev) ((void) 0) 489 #endif 490 491 #define TLBIE(va) __asm volatile("tlbie %0" :: "r"(va)) 492 493 /* XXXSL: this needs to be moved to assembler */ 494 #define TLBIEL(va) __asm __volatile("tlbie %0" :: "r"(va)) 495 496 #define TLBSYNC() __asm volatile("tlbsync") 497 #define SYNC() __asm volatile("sync") 498 #define EIEIO() __asm volatile("eieio") 499 #define MFMSR() mfmsr() 500 #define MTMSR(psl) mtmsr(psl) 501 #define MFPVR() mfpvr() 502 #define MFSRIN(va) mfsrin(va) 503 #define MFTB() mfrtcltbl() 504 505 #if defined (PPC_OEA) || defined (PPC_OEA64_BRIDGE) 506 static inline register_t 507 mfsrin(vaddr_t va) 508 { 509 register_t sr; 510 __asm volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va)); 511 return sr; 512 } 513 #endif /* PPC_OEA*/ 514 515 #if defined (PPC_OEA64_BRIDGE) 516 extern void mfmsr64 (register64_t *result); 517 #endif /* PPC_OEA64_BRIDGE */ 518 519 #define PMAP_LOCK() KERNEL_LOCK(1, NULL) 520 #define PMAP_UNLOCK() KERNEL_UNLOCK_ONE(NULL) 521 522 static inline register_t 523 pmap_interrupts_off(void) 524 { 525 register_t msr = MFMSR(); 526 if (msr & PSL_EE) 527 MTMSR(msr & ~PSL_EE); 528 return msr; 529 } 530 531 static void 532 pmap_interrupts_restore(register_t msr) 533 { 534 if (msr & PSL_EE) 535 MTMSR(msr); 536 } 537 538 static inline u_int32_t 539 mfrtcltbl(void) 540 { 541 542 if ((MFPVR() >> 16) == MPC601) 543 return (mfrtcl() >> 7); 544 else 545 return (mftbl()); 546 } 547 548 /* 549 * These small routines may have to be replaced, 550 * if/when we support processors other that the 604. 551 */ 552 553 void 554 tlbia(void) 555 { 556 char *i; 557 558 SYNC(); 559 #if defined(PPC_OEA) 560 /* 561 * Why not use "tlbia"? Because not all processors implement it. 562 * 563 * This needs to be a per-CPU callback to do the appropriate thing 564 * for the CPU. XXX 565 */ 566 for (i = 0; i < (char *)0x00040000; i += 0x00001000) { 567 TLBIE(i); 568 EIEIO(); 569 SYNC(); 570 } 571 #elif defined (PPC_OEA64) || defined (PPC_OEA64_BRIDGE) 572 printf("Invalidating ALL TLB entries......\n"); 573 /* This is specifically for the 970, 970UM v1.6 pp. 140. */ 574 for (i = 0; i <= (char *)0xFF000; i += 0x00001000) { 575 TLBIEL(i); 576 EIEIO(); 577 SYNC(); 578 } 579 #endif 580 TLBSYNC(); 581 SYNC(); 582 } 583 584 static inline register_t 585 va_to_vsid(const struct pmap *pm, vaddr_t addr) 586 { 587 #if defined (PPC_OEA) || defined (PPC_OEA64_BRIDGE) 588 return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID) >> SR_VSID_SHFT; 589 #else /* PPC_OEA64 */ 590 #if 0 591 const struct ste *ste; 592 register_t hash; 593 int i; 594 595 hash = (addr >> ADDR_ESID_SHFT) & ADDR_ESID_HASH; 596 597 /* 598 * Try the primary group first 599 */ 600 ste = pm->pm_stes[hash].stes; 601 for (i = 0; i < 8; i++, ste++) { 602 if (ste->ste_hi & STE_V) && 603 (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID)) 604 return ste; 605 } 606 607 /* 608 * Then the secondary group. 609 */ 610 ste = pm->pm_stes[hash ^ ADDR_ESID_HASH].stes; 611 for (i = 0; i < 8; i++, ste++) { 612 if (ste->ste_hi & STE_V) && 613 (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID)) 614 return addr; 615 } 616 617 return NULL; 618 #else 619 /* 620 * Rather than searching the STE groups for the VSID, we know 621 * how we generate that from the ESID and so do that. 622 */ 623 return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT; 624 #endif 625 #endif /* PPC_OEA */ 626 } 627 628 static inline register_t 629 va_to_pteg(const struct pmap *pm, vaddr_t addr) 630 { 631 register_t hash; 632 633 hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT); 634 return hash & pmap_pteg_mask; 635 } 636 637 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 638 /* 639 * Given a PTE in the page table, calculate the VADDR that hashes to it. 640 * The only bit of magic is that the top 4 bits of the address doesn't 641 * technically exist in the PTE. But we know we reserved 4 bits of the 642 * VSID for it so that's how we get it. 643 */ 644 static vaddr_t 645 pmap_pte_to_va(volatile const struct pte *pt) 646 { 647 vaddr_t va; 648 uintptr_t ptaddr = (uintptr_t) pt; 649 650 if (pt->pte_hi & PTE_HID) 651 ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg)); 652 653 /* PPC Bits 10-19 PPC64 Bits 42-51 */ 654 #if defined(PPC_OEA) 655 va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff; 656 #elif defined (PPC_OEA64) || defined (PPC_OEA64_BRIDGE) 657 va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x7ff; 658 #endif 659 va <<= ADDR_PIDX_SHFT; 660 661 /* PPC Bits 4-9 PPC64 Bits 36-41 */ 662 va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT; 663 664 #if defined(PPC_OEA64) 665 /* PPC63 Bits 0-35 */ 666 /* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */ 667 #elif defined(PPC_OEA) || defined(PPC_OEA64_BRIDGE) 668 /* PPC Bits 0-3 */ 669 va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; 670 #endif 671 672 return va; 673 } 674 #endif 675 676 static inline struct pvo_head * 677 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p) 678 { 679 #ifdef __HAVE_VM_PAGE_MD 680 struct vm_page *pg; 681 682 pg = PHYS_TO_VM_PAGE(pa); 683 if (pg_p != NULL) 684 *pg_p = pg; 685 if (pg == NULL) 686 return &pmap_pvo_unmanaged; 687 return &pg->mdpage.mdpg_pvoh; 688 #endif 689 #ifdef __HAVE_PMAP_PHYSSEG 690 int bank, pg; 691 692 bank = vm_physseg_find(atop(pa), &pg); 693 if (pg_p != NULL) 694 *pg_p = pg; 695 if (bank == -1) 696 return &pmap_pvo_unmanaged; 697 return &vm_physmem[bank].pmseg.pvoh[pg]; 698 #endif 699 } 700 701 static inline struct pvo_head * 702 vm_page_to_pvoh(struct vm_page *pg) 703 { 704 #ifdef __HAVE_VM_PAGE_MD 705 return &pg->mdpage.mdpg_pvoh; 706 #endif 707 #ifdef __HAVE_PMAP_PHYSSEG 708 return pa_to_pvoh(VM_PAGE_TO_PHYS(pg), NULL); 709 #endif 710 } 711 712 713 #ifdef __HAVE_PMAP_PHYSSEG 714 static inline char * 715 pa_to_attr(paddr_t pa) 716 { 717 int bank, pg; 718 719 bank = vm_physseg_find(atop(pa), &pg); 720 if (bank == -1) 721 return NULL; 722 return &vm_physmem[bank].pmseg.attrs[pg]; 723 } 724 #endif 725 726 static inline void 727 pmap_attr_clear(struct vm_page *pg, int ptebit) 728 { 729 #ifdef __HAVE_PMAP_PHYSSEG 730 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) &= ~(ptebit >> ATTR_SHFT); 731 #endif 732 #ifdef __HAVE_VM_PAGE_MD 733 pg->mdpage.mdpg_attrs &= ~ptebit; 734 #endif 735 } 736 737 static inline int 738 pmap_attr_fetch(struct vm_page *pg) 739 { 740 #ifdef __HAVE_PMAP_PHYSSEG 741 return *pa_to_attr(VM_PAGE_TO_PHYS(pg)) << ATTR_SHFT; 742 #endif 743 #ifdef __HAVE_VM_PAGE_MD 744 return pg->mdpage.mdpg_attrs; 745 #endif 746 } 747 748 static inline void 749 pmap_attr_save(struct vm_page *pg, int ptebit) 750 { 751 #ifdef __HAVE_PMAP_PHYSSEG 752 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) |= (ptebit >> ATTR_SHFT); 753 #endif 754 #ifdef __HAVE_VM_PAGE_MD 755 pg->mdpage.mdpg_attrs |= ptebit; 756 #endif 757 } 758 759 static inline int 760 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt) 761 { 762 if (pt->pte_hi == pvo_pt->pte_hi 763 #if 0 764 && ((pt->pte_lo ^ pvo_pt->pte_lo) & 765 ~(PTE_REF|PTE_CHG)) == 0 766 #endif 767 ) 768 return 1; 769 return 0; 770 } 771 772 static inline void 773 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo) 774 { 775 /* 776 * Construct the PTE. Default to IMB initially. Valid bit 777 * only gets set when the real pte is set in memory. 778 * 779 * Note: Don't set the valid bit for correct operation of tlb update. 780 */ 781 #if defined(PPC_OEA) 782 pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT) 783 | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API); 784 pt->pte_lo = pte_lo; 785 #elif defined (PPC_OEA64_BRIDGE) 786 pt->pte_hi = ((u_int64_t)va_to_vsid(pm, va) << PTE_VSID_SHFT) 787 | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API); 788 pt->pte_lo = (u_int64_t) pte_lo; 789 #elif defined (PPC_OEA64) 790 #error PPC_OEA64 not supported 791 #endif /* PPC_OEA */ 792 } 793 794 static inline void 795 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt) 796 { 797 pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG); 798 } 799 800 static inline void 801 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit) 802 { 803 /* 804 * As shown in Section 7.6.3.2.3 805 */ 806 pt->pte_lo &= ~ptebit; 807 TLBIE(va); 808 SYNC(); 809 EIEIO(); 810 TLBSYNC(); 811 SYNC(); 812 } 813 814 static inline void 815 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt) 816 { 817 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 818 if (pvo_pt->pte_hi & PTE_VALID) 819 panic("pte_set: setting an already valid pte %p", pvo_pt); 820 #endif 821 pvo_pt->pte_hi |= PTE_VALID; 822 823 /* 824 * Update the PTE as defined in section 7.6.3.1 825 * Note that the REF/CHG bits are from pvo_pt and thus should 826 * have been saved so this routine can restore them (if desired). 827 */ 828 pt->pte_lo = pvo_pt->pte_lo; 829 EIEIO(); 830 pt->pte_hi = pvo_pt->pte_hi; 831 TLBSYNC(); 832 SYNC(); 833 pmap_pte_valid++; 834 } 835 836 static inline void 837 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 838 { 839 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 840 if ((pvo_pt->pte_hi & PTE_VALID) == 0) 841 panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt); 842 if ((pt->pte_hi & PTE_VALID) == 0) 843 panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt); 844 #endif 845 846 pvo_pt->pte_hi &= ~PTE_VALID; 847 /* 848 * Force the ref & chg bits back into the PTEs. 849 */ 850 SYNC(); 851 /* 852 * Invalidate the pte ... (Section 7.6.3.3) 853 */ 854 pt->pte_hi &= ~PTE_VALID; 855 SYNC(); 856 TLBIE(va); 857 SYNC(); 858 EIEIO(); 859 TLBSYNC(); 860 SYNC(); 861 /* 862 * Save the ref & chg bits ... 863 */ 864 pmap_pte_synch(pt, pvo_pt); 865 pmap_pte_valid--; 866 } 867 868 static inline void 869 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 870 { 871 /* 872 * Invalidate the PTE 873 */ 874 pmap_pte_unset(pt, pvo_pt, va); 875 pmap_pte_set(pt, pvo_pt); 876 } 877 878 /* 879 * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx 880 * (either primary or secondary location). 881 * 882 * Note: both the destination and source PTEs must not have PTE_VALID set. 883 */ 884 885 STATIC int 886 pmap_pte_insert(int ptegidx, struct pte *pvo_pt) 887 { 888 volatile struct pte *pt; 889 int i; 890 891 #if defined(DEBUG) 892 #if defined (PPC_OEA) 893 DPRINTFN(PTE, ("pmap_pte_insert: idx 0x%x, pte 0x%x 0x%x\n", 894 ptegidx, (unsigned int) pvo_pt->pte_hi, (unsigned int) pvo_pt->pte_lo)); 895 #elif defined (PPC_OEA64_BRIDGE) 896 DPRINTFN(PTE, ("pmap_pte_insert: idx 0x%x, pte 0x%016llx 0x%016llx\n", 897 ptegidx, (unsigned long long) pvo_pt->pte_hi, 898 (unsigned long long) pvo_pt->pte_lo)); 899 900 #endif 901 #endif 902 /* 903 * First try primary hash. 904 */ 905 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 906 if ((pt->pte_hi & PTE_VALID) == 0) { 907 pvo_pt->pte_hi &= ~PTE_HID; 908 pmap_pte_set(pt, pvo_pt); 909 return i; 910 } 911 } 912 913 /* 914 * Now try secondary hash. 915 */ 916 ptegidx ^= pmap_pteg_mask; 917 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 918 if ((pt->pte_hi & PTE_VALID) == 0) { 919 pvo_pt->pte_hi |= PTE_HID; 920 pmap_pte_set(pt, pvo_pt); 921 return i; 922 } 923 } 924 return -1; 925 } 926 927 /* 928 * Spill handler. 929 * 930 * Tries to spill a page table entry from the overflow area. 931 * This runs in either real mode (if dealing with a exception spill) 932 * or virtual mode when dealing with manually spilling one of the 933 * kernel's pte entries. In either case, interrupts are already 934 * disabled. 935 */ 936 937 int 938 pmap_pte_spill(struct pmap *pm, vaddr_t addr, bool exec) 939 { 940 struct pvo_entry *source_pvo, *victim_pvo, *next_pvo; 941 struct pvo_entry *pvo; 942 /* XXX: gcc -- vpvoh is always set at either *1* or *2* */ 943 struct pvo_tqhead *pvoh, *vpvoh = NULL; 944 int ptegidx, i, j; 945 volatile struct pteg *pteg; 946 volatile struct pte *pt; 947 948 PMAP_LOCK(); 949 950 ptegidx = va_to_pteg(pm, addr); 951 952 /* 953 * Have to substitute some entry. Use the primary hash for this. 954 * Use low bits of timebase as random generator. Make sure we are 955 * not picking a kernel pte for replacement. 956 */ 957 pteg = &pmap_pteg_table[ptegidx]; 958 i = MFTB() & 7; 959 for (j = 0; j < 8; j++) { 960 pt = &pteg->pt[i]; 961 if ((pt->pte_hi & PTE_VALID) == 0 || 962 VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT) 963 != KERNEL_VSIDBITS) 964 break; 965 i = (i + 1) & 7; 966 } 967 KASSERT(j < 8); 968 969 source_pvo = NULL; 970 victim_pvo = NULL; 971 pvoh = &pmap_pvo_table[ptegidx]; 972 TAILQ_FOREACH(pvo, pvoh, pvo_olink) { 973 974 /* 975 * We need to find pvo entry for this address... 976 */ 977 PMAP_PVO_CHECK(pvo); /* sanity check */ 978 979 /* 980 * If we haven't found the source and we come to a PVO with 981 * a valid PTE, then we know we can't find it because all 982 * evicted PVOs always are first in the list. 983 */ 984 if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID)) 985 break; 986 if (source_pvo == NULL && pm == pvo->pvo_pmap && 987 addr == PVO_VADDR(pvo)) { 988 989 /* 990 * Now we have found the entry to be spilled into the 991 * pteg. Attempt to insert it into the page table. 992 */ 993 j = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 994 if (j >= 0) { 995 PVO_PTEGIDX_SET(pvo, j); 996 PMAP_PVO_CHECK(pvo); /* sanity check */ 997 PVO_WHERE(pvo, SPILL_INSERT); 998 pvo->pvo_pmap->pm_evictions--; 999 PMAPCOUNT(ptes_spilled); 1000 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 1001 ? pmap_evcnt_ptes_secondary 1002 : pmap_evcnt_ptes_primary)[j]); 1003 1004 /* 1005 * Since we keep the evicted entries at the 1006 * from of the PVO list, we need move this 1007 * (now resident) PVO after the evicted 1008 * entries. 1009 */ 1010 next_pvo = TAILQ_NEXT(pvo, pvo_olink); 1011 1012 /* 1013 * If we don't have to move (either we were the 1014 * last entry or the next entry was valid), 1015 * don't change our position. Otherwise 1016 * move ourselves to the tail of the queue. 1017 */ 1018 if (next_pvo != NULL && 1019 !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) { 1020 TAILQ_REMOVE(pvoh, pvo, pvo_olink); 1021 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 1022 } 1023 PMAP_UNLOCK(); 1024 return 1; 1025 } 1026 source_pvo = pvo; 1027 if (exec && !PVO_EXECUTABLE_P(source_pvo)) { 1028 return 0; 1029 } 1030 if (victim_pvo != NULL) 1031 break; 1032 } 1033 1034 /* 1035 * We also need the pvo entry of the victim we are replacing 1036 * so save the R & C bits of the PTE. 1037 */ 1038 if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL && 1039 pmap_pte_compare(pt, &pvo->pvo_pte)) { 1040 vpvoh = pvoh; /* *1* */ 1041 victim_pvo = pvo; 1042 if (source_pvo != NULL) 1043 break; 1044 } 1045 } 1046 1047 if (source_pvo == NULL) { 1048 PMAPCOUNT(ptes_unspilled); 1049 PMAP_UNLOCK(); 1050 return 0; 1051 } 1052 1053 if (victim_pvo == NULL) { 1054 if ((pt->pte_hi & PTE_HID) == 0) 1055 panic("pmap_pte_spill: victim p-pte (%p) has " 1056 "no pvo entry!", pt); 1057 1058 /* 1059 * If this is a secondary PTE, we need to search 1060 * its primary pvo bucket for the matching PVO. 1061 */ 1062 vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; /* *2* */ 1063 TAILQ_FOREACH(pvo, vpvoh, pvo_olink) { 1064 PMAP_PVO_CHECK(pvo); /* sanity check */ 1065 1066 /* 1067 * We also need the pvo entry of the victim we are 1068 * replacing so save the R & C bits of the PTE. 1069 */ 1070 if (pmap_pte_compare(pt, &pvo->pvo_pte)) { 1071 victim_pvo = pvo; 1072 break; 1073 } 1074 } 1075 if (victim_pvo == NULL) 1076 panic("pmap_pte_spill: victim s-pte (%p) has " 1077 "no pvo entry!", pt); 1078 } 1079 1080 /* 1081 * The victim should be not be a kernel PVO/PTE entry. 1082 */ 1083 KASSERT(victim_pvo->pvo_pmap != pmap_kernel()); 1084 KASSERT(PVO_PTEGIDX_ISSET(victim_pvo)); 1085 KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i); 1086 1087 /* 1088 * We are invalidating the TLB entry for the EA for the 1089 * we are replacing even though its valid; If we don't 1090 * we lose any ref/chg bit changes contained in the TLB 1091 * entry. 1092 */ 1093 source_pvo->pvo_pte.pte_hi &= ~PTE_HID; 1094 1095 /* 1096 * To enforce the PVO list ordering constraint that all 1097 * evicted entries should come before all valid entries, 1098 * move the source PVO to the tail of its list and the 1099 * victim PVO to the head of its list (which might not be 1100 * the same list, if the victim was using the secondary hash). 1101 */ 1102 TAILQ_REMOVE(pvoh, source_pvo, pvo_olink); 1103 TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink); 1104 TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink); 1105 TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink); 1106 pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr); 1107 pmap_pte_set(pt, &source_pvo->pvo_pte); 1108 victim_pvo->pvo_pmap->pm_evictions++; 1109 source_pvo->pvo_pmap->pm_evictions--; 1110 PVO_WHERE(victim_pvo, SPILL_UNSET); 1111 PVO_WHERE(source_pvo, SPILL_SET); 1112 1113 PVO_PTEGIDX_CLR(victim_pvo); 1114 PVO_PTEGIDX_SET(source_pvo, i); 1115 PMAPCOUNT2(pmap_evcnt_ptes_primary[i]); 1116 PMAPCOUNT(ptes_spilled); 1117 PMAPCOUNT(ptes_evicted); 1118 PMAPCOUNT(ptes_removed); 1119 1120 PMAP_PVO_CHECK(victim_pvo); 1121 PMAP_PVO_CHECK(source_pvo); 1122 1123 PMAP_UNLOCK(); 1124 return 1; 1125 } 1126 1127 /* 1128 * Restrict given range to physical memory 1129 */ 1130 void 1131 pmap_real_memory(paddr_t *start, psize_t *size) 1132 { 1133 struct mem_region *mp; 1134 1135 for (mp = mem; mp->size; mp++) { 1136 if (*start + *size > mp->start 1137 && *start < mp->start + mp->size) { 1138 if (*start < mp->start) { 1139 *size -= mp->start - *start; 1140 *start = mp->start; 1141 } 1142 if (*start + *size > mp->start + mp->size) 1143 *size = mp->start + mp->size - *start; 1144 return; 1145 } 1146 } 1147 *size = 0; 1148 } 1149 1150 /* 1151 * Initialize anything else for pmap handling. 1152 * Called during vm_init(). 1153 */ 1154 void 1155 pmap_init(void) 1156 { 1157 #ifdef __HAVE_PMAP_PHYSSEG 1158 struct pvo_tqhead *pvoh; 1159 int bank; 1160 long sz; 1161 char *attr; 1162 1163 pvoh = pmap_physseg.pvoh; 1164 attr = pmap_physseg.attrs; 1165 for (bank = 0; bank < vm_nphysseg; bank++) { 1166 sz = vm_physmem[bank].end - vm_physmem[bank].start; 1167 vm_physmem[bank].pmseg.pvoh = pvoh; 1168 vm_physmem[bank].pmseg.attrs = attr; 1169 for (; sz > 0; sz--, pvoh++, attr++) { 1170 TAILQ_INIT(pvoh); 1171 *attr = 0; 1172 } 1173 } 1174 #endif 1175 1176 pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry), 1177 sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl", 1178 &pmap_pool_mallocator, IPL_NONE); 1179 1180 pool_setlowat(&pmap_mpvo_pool, 1008); 1181 1182 pmap_initialized = 1; 1183 1184 } 1185 1186 /* 1187 * How much virtual space does the kernel get? 1188 */ 1189 void 1190 pmap_virtual_space(vaddr_t *start, vaddr_t *end) 1191 { 1192 /* 1193 * For now, reserve one segment (minus some overhead) for kernel 1194 * virtual memory 1195 */ 1196 *start = VM_MIN_KERNEL_ADDRESS; 1197 *end = VM_MAX_KERNEL_ADDRESS; 1198 } 1199 1200 /* 1201 * Allocate, initialize, and return a new physical map. 1202 */ 1203 pmap_t 1204 pmap_create(void) 1205 { 1206 pmap_t pm; 1207 1208 pm = pool_get(&pmap_pool, PR_WAITOK); 1209 memset((void *)pm, 0, sizeof *pm); 1210 pmap_pinit(pm); 1211 1212 DPRINTFN(CREATE,("pmap_create: pm %p:\n" 1213 "\t%06x %06x %06x %06x %06x %06x %06x %06x\n" 1214 "\t%06x %06x %06x %06x %06x %06x %06x %06x\n", pm, 1215 (unsigned int) pm->pm_sr[0], (unsigned int) pm->pm_sr[1], 1216 (unsigned int) pm->pm_sr[2], (unsigned int) pm->pm_sr[3], 1217 (unsigned int) pm->pm_sr[4], (unsigned int) pm->pm_sr[5], 1218 (unsigned int) pm->pm_sr[6], (unsigned int) pm->pm_sr[7], 1219 (unsigned int) pm->pm_sr[8], (unsigned int) pm->pm_sr[9], 1220 (unsigned int) pm->pm_sr[10], (unsigned int) pm->pm_sr[11], 1221 (unsigned int) pm->pm_sr[12], (unsigned int) pm->pm_sr[13], 1222 (unsigned int) pm->pm_sr[14], (unsigned int) pm->pm_sr[15])); 1223 return pm; 1224 } 1225 1226 /* 1227 * Initialize a preallocated and zeroed pmap structure. 1228 */ 1229 void 1230 pmap_pinit(pmap_t pm) 1231 { 1232 register_t entropy = MFTB(); 1233 register_t mask; 1234 int i; 1235 1236 /* 1237 * Allocate some segment registers for this pmap. 1238 */ 1239 pm->pm_refs = 1; 1240 PMAP_LOCK(); 1241 for (i = 0; i < NPMAPS; i += VSID_NBPW) { 1242 static register_t pmap_vsidcontext; 1243 register_t hash; 1244 unsigned int n; 1245 1246 /* Create a new value by multiplying by a prime adding in 1247 * entropy from the timebase register. This is to make the 1248 * VSID more random so that the PT Hash function collides 1249 * less often. (note that the prime causes gcc to do shifts 1250 * instead of a multiply) 1251 */ 1252 pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy; 1253 hash = pmap_vsidcontext & (NPMAPS - 1); 1254 if (hash == 0) { /* 0 is special, avoid it */ 1255 entropy += 0xbadf00d; 1256 continue; 1257 } 1258 n = hash >> 5; 1259 mask = 1L << (hash & (VSID_NBPW-1)); 1260 hash = pmap_vsidcontext; 1261 if (pmap_vsid_bitmap[n] & mask) { /* collision? */ 1262 /* anything free in this bucket? */ 1263 if (~pmap_vsid_bitmap[n] == 0) { 1264 entropy = hash ^ (hash >> 16); 1265 continue; 1266 } 1267 i = ffs(~pmap_vsid_bitmap[n]) - 1; 1268 mask = 1L << i; 1269 hash &= ~(VSID_NBPW-1); 1270 hash |= i; 1271 } 1272 hash &= PTE_VSID >> PTE_VSID_SHFT; 1273 pmap_vsid_bitmap[n] |= mask; 1274 pm->pm_vsid = hash; 1275 #if defined (PPC_OEA) || defined (PPC_OEA64_BRIDGE) 1276 for (i = 0; i < 16; i++) 1277 pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY | 1278 SR_NOEXEC; 1279 #endif 1280 PMAP_UNLOCK(); 1281 return; 1282 } 1283 PMAP_UNLOCK(); 1284 panic("pmap_pinit: out of segments"); 1285 } 1286 1287 /* 1288 * Add a reference to the given pmap. 1289 */ 1290 void 1291 pmap_reference(pmap_t pm) 1292 { 1293 atomic_inc_uint(&pm->pm_refs); 1294 } 1295 1296 /* 1297 * Retire the given pmap from service. 1298 * Should only be called if the map contains no valid mappings. 1299 */ 1300 void 1301 pmap_destroy(pmap_t pm) 1302 { 1303 if (atomic_dec_uint_nv(&pm->pm_refs) == 0) { 1304 pmap_release(pm); 1305 pool_put(&pmap_pool, pm); 1306 } 1307 } 1308 1309 /* 1310 * Release any resources held by the given physical map. 1311 * Called when a pmap initialized by pmap_pinit is being released. 1312 */ 1313 void 1314 pmap_release(pmap_t pm) 1315 { 1316 int idx, mask; 1317 1318 KASSERT(pm->pm_stats.resident_count == 0); 1319 KASSERT(pm->pm_stats.wired_count == 0); 1320 1321 PMAP_LOCK(); 1322 if (pm->pm_sr[0] == 0) 1323 panic("pmap_release"); 1324 idx = pm->pm_vsid & (NPMAPS-1); 1325 mask = 1 << (idx % VSID_NBPW); 1326 idx /= VSID_NBPW; 1327 1328 KASSERT(pmap_vsid_bitmap[idx] & mask); 1329 pmap_vsid_bitmap[idx] &= ~mask; 1330 PMAP_UNLOCK(); 1331 } 1332 1333 /* 1334 * Copy the range specified by src_addr/len 1335 * from the source map to the range dst_addr/len 1336 * in the destination map. 1337 * 1338 * This routine is only advisory and need not do anything. 1339 */ 1340 void 1341 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr, 1342 vsize_t len, vaddr_t src_addr) 1343 { 1344 PMAPCOUNT(copies); 1345 } 1346 1347 /* 1348 * Require that all active physical maps contain no 1349 * incorrect entries NOW. 1350 */ 1351 void 1352 pmap_update(struct pmap *pmap) 1353 { 1354 PMAPCOUNT(updates); 1355 TLBSYNC(); 1356 } 1357 1358 /* 1359 * Garbage collects the physical map system for 1360 * pages which are no longer used. 1361 * Success need not be guaranteed -- that is, there 1362 * may well be pages which are not referenced, but 1363 * others may be collected. 1364 * Called by the pageout daemon when pages are scarce. 1365 */ 1366 void 1367 pmap_collect(pmap_t pm) 1368 { 1369 PMAPCOUNT(collects); 1370 } 1371 1372 static inline int 1373 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx) 1374 { 1375 int pteidx; 1376 /* 1377 * We can find the actual pte entry without searching by 1378 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9] 1379 * and by noticing the HID bit. 1380 */ 1381 pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo); 1382 if (pvo->pvo_pte.pte_hi & PTE_HID) 1383 pteidx ^= pmap_pteg_mask * 8; 1384 return pteidx; 1385 } 1386 1387 volatile struct pte * 1388 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx) 1389 { 1390 volatile struct pte *pt; 1391 1392 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1393 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) 1394 return NULL; 1395 #endif 1396 1397 /* 1398 * If we haven't been supplied the ptegidx, calculate it. 1399 */ 1400 if (pteidx == -1) { 1401 int ptegidx; 1402 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1403 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1404 } 1405 1406 pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7]; 1407 1408 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1409 return pt; 1410 #else 1411 if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) { 1412 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1413 "pvo but no valid pte index", pvo); 1414 } 1415 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) { 1416 panic("pmap_pvo_to_pte: pvo %p: has valid pte index in " 1417 "pvo but no valid pte", pvo); 1418 } 1419 1420 if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) { 1421 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) { 1422 #if defined(DEBUG) || defined(PMAPCHECK) 1423 pmap_pte_print(pt); 1424 #endif 1425 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1426 "pmap_pteg_table %p but invalid in pvo", 1427 pvo, pt); 1428 } 1429 if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) { 1430 #if defined(DEBUG) || defined(PMAPCHECK) 1431 pmap_pte_print(pt); 1432 #endif 1433 panic("pmap_pvo_to_pte: pvo %p: pvo pte does " 1434 "not match pte %p in pmap_pteg_table", 1435 pvo, pt); 1436 } 1437 return pt; 1438 } 1439 1440 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1441 #if defined(DEBUG) || defined(PMAPCHECK) 1442 pmap_pte_print(pt); 1443 #endif 1444 panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in " 1445 "pmap_pteg_table but valid in pvo", pvo, pt); 1446 } 1447 return NULL; 1448 #endif /* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */ 1449 } 1450 1451 struct pvo_entry * 1452 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p) 1453 { 1454 struct pvo_entry *pvo; 1455 int ptegidx; 1456 1457 va &= ~ADDR_POFF; 1458 ptegidx = va_to_pteg(pm, va); 1459 1460 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1461 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1462 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 1463 panic("pmap_pvo_find_va: invalid pvo %p on " 1464 "list %#x (%p)", pvo, ptegidx, 1465 &pmap_pvo_table[ptegidx]); 1466 #endif 1467 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1468 if (pteidx_p) 1469 *pteidx_p = pmap_pvo_pte_index(pvo, ptegidx); 1470 return pvo; 1471 } 1472 } 1473 if ((pm == pmap_kernel()) && (va < SEGMENT_LENGTH)) 1474 panic("%s: returning NULL for %s pmap, va: 0x%08lx\n", __func__, 1475 (pm == pmap_kernel() ? "kernel" : "user"), va); 1476 return NULL; 1477 } 1478 1479 #if defined(DEBUG) || defined(PMAPCHECK) 1480 void 1481 pmap_pvo_check(const struct pvo_entry *pvo) 1482 { 1483 struct pvo_head *pvo_head; 1484 struct pvo_entry *pvo0; 1485 volatile struct pte *pt; 1486 int failed = 0; 1487 1488 PMAP_LOCK(); 1489 1490 if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH) 1491 panic("pmap_pvo_check: pvo %p: invalid address", pvo); 1492 1493 if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) { 1494 printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n", 1495 pvo, pvo->pvo_pmap); 1496 failed = 1; 1497 } 1498 1499 if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH || 1500 (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) { 1501 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1502 pvo, TAILQ_NEXT(pvo, pvo_olink)); 1503 failed = 1; 1504 } 1505 1506 if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH || 1507 (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) { 1508 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1509 pvo, LIST_NEXT(pvo, pvo_vlink)); 1510 failed = 1; 1511 } 1512 1513 if (PVO_MANAGED_P(pvo)) { 1514 pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL); 1515 } else { 1516 if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) { 1517 printf("pmap_pvo_check: pvo %p: non kernel address " 1518 "on kernel unmanaged list\n", pvo); 1519 failed = 1; 1520 } 1521 pvo_head = &pmap_pvo_kunmanaged; 1522 } 1523 LIST_FOREACH(pvo0, pvo_head, pvo_vlink) { 1524 if (pvo0 == pvo) 1525 break; 1526 } 1527 if (pvo0 == NULL) { 1528 printf("pmap_pvo_check: pvo %p: not present " 1529 "on its vlist head %p\n", pvo, pvo_head); 1530 failed = 1; 1531 } 1532 if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) { 1533 printf("pmap_pvo_check: pvo %p: not present " 1534 "on its olist head\n", pvo); 1535 failed = 1; 1536 } 1537 pt = pmap_pvo_to_pte(pvo, -1); 1538 if (pt == NULL) { 1539 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1540 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1541 "no PTE\n", pvo); 1542 failed = 1; 1543 } 1544 } else { 1545 if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] || 1546 (uintptr_t) pt >= 1547 (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) { 1548 printf("pmap_pvo_check: pvo %p: pte %p not in " 1549 "pteg table\n", pvo, pt); 1550 failed = 1; 1551 } 1552 if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) { 1553 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1554 "no PTE\n", pvo); 1555 failed = 1; 1556 } 1557 if (pvo->pvo_pte.pte_hi != pt->pte_hi) { 1558 printf("pmap_pvo_check: pvo %p: pte_hi differ: " 1559 "%#x/%#x\n", pvo, (unsigned int) pvo->pvo_pte.pte_hi, (unsigned int) pt->pte_hi); 1560 failed = 1; 1561 } 1562 if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) & 1563 (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) { 1564 printf("pmap_pvo_check: pvo %p: pte_lo differ: " 1565 "%#x/%#x\n", pvo, 1566 (unsigned int) (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)), 1567 (unsigned int) (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN))); 1568 failed = 1; 1569 } 1570 if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) { 1571 printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#lx" 1572 " doesn't not match PVO's VA %#lx\n", 1573 pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo)); 1574 failed = 1; 1575 } 1576 if (failed) 1577 pmap_pte_print(pt); 1578 } 1579 if (failed) 1580 panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo, 1581 pvo->pvo_pmap); 1582 1583 PMAP_UNLOCK(); 1584 } 1585 #endif /* DEBUG || PMAPCHECK */ 1586 1587 /* 1588 * Search the PVO table looking for a non-wired entry. 1589 * If we find one, remove it and return it. 1590 */ 1591 1592 struct pvo_entry * 1593 pmap_pvo_reclaim(struct pmap *pm) 1594 { 1595 struct pvo_tqhead *pvoh; 1596 struct pvo_entry *pvo; 1597 uint32_t idx, endidx; 1598 1599 endidx = pmap_pvo_reclaim_nextidx; 1600 for (idx = (endidx + 1) & pmap_pteg_mask; idx != endidx; 1601 idx = (idx + 1) & pmap_pteg_mask) { 1602 pvoh = &pmap_pvo_table[idx]; 1603 TAILQ_FOREACH(pvo, pvoh, pvo_olink) { 1604 if (!PVO_WIRED_P(pvo)) { 1605 pmap_pvo_remove(pvo, -1, NULL); 1606 pmap_pvo_reclaim_nextidx = idx; 1607 PMAPCOUNT(pvos_reclaimed); 1608 return pvo; 1609 } 1610 } 1611 } 1612 return NULL; 1613 } 1614 1615 /* 1616 * This returns whether this is the first mapping of a page. 1617 */ 1618 int 1619 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head, 1620 vaddr_t va, paddr_t pa, register_t pte_lo, int flags) 1621 { 1622 struct pvo_entry *pvo; 1623 struct pvo_tqhead *pvoh; 1624 register_t msr; 1625 int ptegidx; 1626 int i; 1627 int poolflags = PR_NOWAIT; 1628 1629 /* 1630 * Compute the PTE Group index. 1631 */ 1632 va &= ~ADDR_POFF; 1633 ptegidx = va_to_pteg(pm, va); 1634 1635 msr = pmap_interrupts_off(); 1636 1637 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1638 if (pmap_pvo_remove_depth > 0) 1639 panic("pmap_pvo_enter: called while pmap_pvo_remove active!"); 1640 if (++pmap_pvo_enter_depth > 1) 1641 panic("pmap_pvo_enter: called recursively!"); 1642 #endif 1643 1644 /* 1645 * Remove any existing mapping for this page. Reuse the 1646 * pvo entry if there a mapping. 1647 */ 1648 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1649 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1650 #ifdef DEBUG 1651 if ((pmapdebug & PMAPDEBUG_PVOENTER) && 1652 ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) & 1653 ~(PTE_REF|PTE_CHG)) == 0 && 1654 va < VM_MIN_KERNEL_ADDRESS) { 1655 printf("pmap_pvo_enter: pvo %p: dup %#x/%#lx\n", 1656 pvo, (unsigned int) pvo->pvo_pte.pte_lo, (unsigned int) pte_lo|pa); 1657 printf("pmap_pvo_enter: pte_hi=%#x sr=%#x\n", 1658 (unsigned int) pvo->pvo_pte.pte_hi, 1659 (unsigned int) pm->pm_sr[va >> ADDR_SR_SHFT]); 1660 pmap_pte_print(pmap_pvo_to_pte(pvo, -1)); 1661 #ifdef DDBX 1662 Debugger(); 1663 #endif 1664 } 1665 #endif 1666 PMAPCOUNT(mappings_replaced); 1667 pmap_pvo_remove(pvo, -1, NULL); 1668 break; 1669 } 1670 } 1671 1672 /* 1673 * If we aren't overwriting an mapping, try to allocate 1674 */ 1675 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1676 --pmap_pvo_enter_depth; 1677 #endif 1678 pmap_interrupts_restore(msr); 1679 if (pvo) { 1680 pmap_pvo_free(pvo); 1681 } 1682 pvo = pool_get(pl, poolflags); 1683 1684 #ifdef DEBUG 1685 /* 1686 * Exercise pmap_pvo_reclaim() a little. 1687 */ 1688 if (pvo && (flags & PMAP_CANFAIL) != 0 && 1689 pmap_pvo_reclaim_debugctr++ > 0x1000 && 1690 (pmap_pvo_reclaim_debugctr & 0xff) == 0) { 1691 pool_put(pl, pvo); 1692 pvo = NULL; 1693 } 1694 #endif 1695 1696 msr = pmap_interrupts_off(); 1697 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1698 ++pmap_pvo_enter_depth; 1699 #endif 1700 if (pvo == NULL) { 1701 pvo = pmap_pvo_reclaim(pm); 1702 if (pvo == NULL) { 1703 if ((flags & PMAP_CANFAIL) == 0) 1704 panic("pmap_pvo_enter: failed"); 1705 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1706 pmap_pvo_enter_depth--; 1707 #endif 1708 PMAPCOUNT(pvos_failed); 1709 pmap_interrupts_restore(msr); 1710 return ENOMEM; 1711 } 1712 } 1713 1714 pvo->pvo_vaddr = va; 1715 pvo->pvo_pmap = pm; 1716 pvo->pvo_vaddr &= ~ADDR_POFF; 1717 if (flags & VM_PROT_EXECUTE) { 1718 PMAPCOUNT(exec_mappings); 1719 pvo_set_exec(pvo); 1720 } 1721 if (flags & PMAP_WIRED) 1722 pvo->pvo_vaddr |= PVO_WIRED; 1723 if (pvo_head != &pmap_pvo_kunmanaged) { 1724 pvo->pvo_vaddr |= PVO_MANAGED; 1725 PMAPCOUNT(mappings); 1726 } else { 1727 PMAPCOUNT(kernel_mappings); 1728 } 1729 pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo); 1730 1731 LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink); 1732 if (PVO_WIRED_P(pvo)) 1733 pvo->pvo_pmap->pm_stats.wired_count++; 1734 pvo->pvo_pmap->pm_stats.resident_count++; 1735 #if defined(DEBUG) 1736 /* if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) */ 1737 DPRINTFN(PVOENTER, 1738 ("pmap_pvo_enter: pvo %p: pm %p va %#lx pa %#lx\n", 1739 pvo, pm, va, pa)); 1740 #endif 1741 1742 /* 1743 * We hope this succeeds but it isn't required. 1744 */ 1745 pvoh = &pmap_pvo_table[ptegidx]; 1746 i = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 1747 if (i >= 0) { 1748 PVO_PTEGIDX_SET(pvo, i); 1749 PVO_WHERE(pvo, ENTER_INSERT); 1750 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 1751 ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]); 1752 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 1753 1754 } else { 1755 /* 1756 * Since we didn't have room for this entry (which makes it 1757 * and evicted entry), place it at the head of the list. 1758 */ 1759 TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink); 1760 PMAPCOUNT(ptes_evicted); 1761 pm->pm_evictions++; 1762 /* 1763 * If this is a kernel page, make sure it's active. 1764 */ 1765 if (pm == pmap_kernel()) { 1766 i = pmap_pte_spill(pm, va, false); 1767 KASSERT(i); 1768 } 1769 } 1770 PMAP_PVO_CHECK(pvo); /* sanity check */ 1771 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1772 pmap_pvo_enter_depth--; 1773 #endif 1774 pmap_interrupts_restore(msr); 1775 return 0; 1776 } 1777 1778 void 1779 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx, struct pvo_head *pvol) 1780 { 1781 volatile struct pte *pt; 1782 int ptegidx; 1783 1784 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1785 if (++pmap_pvo_remove_depth > 1) 1786 panic("pmap_pvo_remove: called recursively!"); 1787 #endif 1788 1789 /* 1790 * If we haven't been supplied the ptegidx, calculate it. 1791 */ 1792 if (pteidx == -1) { 1793 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1794 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1795 } else { 1796 ptegidx = pteidx >> 3; 1797 if (pvo->pvo_pte.pte_hi & PTE_HID) 1798 ptegidx ^= pmap_pteg_mask; 1799 } 1800 PMAP_PVO_CHECK(pvo); /* sanity check */ 1801 1802 /* 1803 * If there is an active pte entry, we need to deactivate it 1804 * (and save the ref & chg bits). 1805 */ 1806 pt = pmap_pvo_to_pte(pvo, pteidx); 1807 if (pt != NULL) { 1808 pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 1809 PVO_WHERE(pvo, REMOVE); 1810 PVO_PTEGIDX_CLR(pvo); 1811 PMAPCOUNT(ptes_removed); 1812 } else { 1813 KASSERT(pvo->pvo_pmap->pm_evictions > 0); 1814 pvo->pvo_pmap->pm_evictions--; 1815 } 1816 1817 /* 1818 * Account for executable mappings. 1819 */ 1820 if (PVO_EXECUTABLE_P(pvo)) 1821 pvo_clear_exec(pvo); 1822 1823 /* 1824 * Update our statistics. 1825 */ 1826 pvo->pvo_pmap->pm_stats.resident_count--; 1827 if (PVO_WIRED_P(pvo)) 1828 pvo->pvo_pmap->pm_stats.wired_count--; 1829 1830 /* 1831 * Save the REF/CHG bits into their cache if the page is managed. 1832 */ 1833 if (PVO_MANAGED_P(pvo)) { 1834 register_t ptelo = pvo->pvo_pte.pte_lo; 1835 struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN); 1836 1837 if (pg != NULL) { 1838 /* 1839 * If this page was changed and it is mapped exec, 1840 * invalidate it. 1841 */ 1842 if ((ptelo & PTE_CHG) && 1843 (pmap_attr_fetch(pg) & PTE_EXEC)) { 1844 struct pvo_head *pvoh = vm_page_to_pvoh(pg); 1845 if (LIST_EMPTY(pvoh)) { 1846 DPRINTFN(EXEC, ("[pmap_pvo_remove: " 1847 "%#lx: clear-exec]\n", 1848 VM_PAGE_TO_PHYS(pg))); 1849 pmap_attr_clear(pg, PTE_EXEC); 1850 PMAPCOUNT(exec_uncached_pvo_remove); 1851 } else { 1852 DPRINTFN(EXEC, ("[pmap_pvo_remove: " 1853 "%#lx: syncicache]\n", 1854 VM_PAGE_TO_PHYS(pg))); 1855 pmap_syncicache(VM_PAGE_TO_PHYS(pg), 1856 PAGE_SIZE); 1857 PMAPCOUNT(exec_synced_pvo_remove); 1858 } 1859 } 1860 1861 pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG)); 1862 } 1863 PMAPCOUNT(unmappings); 1864 } else { 1865 PMAPCOUNT(kernel_unmappings); 1866 } 1867 1868 /* 1869 * Remove the PVO from its lists and return it to the pool. 1870 */ 1871 LIST_REMOVE(pvo, pvo_vlink); 1872 TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink); 1873 if (pvol) { 1874 LIST_INSERT_HEAD(pvol, pvo, pvo_vlink); 1875 } 1876 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1877 pmap_pvo_remove_depth--; 1878 #endif 1879 } 1880 1881 void 1882 pmap_pvo_free(struct pvo_entry *pvo) 1883 { 1884 1885 pool_put(PVO_MANAGED_P(pvo) ? &pmap_mpvo_pool : &pmap_upvo_pool, pvo); 1886 } 1887 1888 void 1889 pmap_pvo_free_list(struct pvo_head *pvol) 1890 { 1891 struct pvo_entry *pvo, *npvo; 1892 1893 for (pvo = LIST_FIRST(pvol); pvo != NULL; pvo = npvo) { 1894 npvo = LIST_NEXT(pvo, pvo_vlink); 1895 LIST_REMOVE(pvo, pvo_vlink); 1896 pmap_pvo_free(pvo); 1897 } 1898 } 1899 1900 /* 1901 * Mark a mapping as executable. 1902 * If this is the first executable mapping in the segment, 1903 * clear the noexec flag. 1904 */ 1905 STATIC void 1906 pvo_set_exec(struct pvo_entry *pvo) 1907 { 1908 struct pmap *pm = pvo->pvo_pmap; 1909 1910 if (pm == pmap_kernel() || PVO_EXECUTABLE_P(pvo)) { 1911 return; 1912 } 1913 pvo->pvo_vaddr |= PVO_EXECUTABLE; 1914 #if defined (PPC_OEA) || defined (PPC_OEA64_BRIDGE) 1915 { 1916 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1917 if (pm->pm_exec[sr]++ == 0) { 1918 pm->pm_sr[sr] &= ~SR_NOEXEC; 1919 } 1920 } 1921 #endif 1922 } 1923 1924 /* 1925 * Mark a mapping as non-executable. 1926 * If this was the last executable mapping in the segment, 1927 * set the noexec flag. 1928 */ 1929 STATIC void 1930 pvo_clear_exec(struct pvo_entry *pvo) 1931 { 1932 struct pmap *pm = pvo->pvo_pmap; 1933 1934 if (pm == pmap_kernel() || !PVO_EXECUTABLE_P(pvo)) { 1935 return; 1936 } 1937 pvo->pvo_vaddr &= ~PVO_EXECUTABLE; 1938 #if defined (PPC_OEA) || defined (PPC_OEA64_BRIDGE) 1939 { 1940 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1941 if (--pm->pm_exec[sr] == 0) { 1942 pm->pm_sr[sr] |= SR_NOEXEC; 1943 } 1944 } 1945 #endif 1946 } 1947 1948 /* 1949 * Insert physical page at pa into the given pmap at virtual address va. 1950 */ 1951 int 1952 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags) 1953 { 1954 struct mem_region *mp; 1955 struct pvo_head *pvo_head; 1956 struct vm_page *pg; 1957 struct pool *pl; 1958 register_t pte_lo; 1959 int error; 1960 u_int pvo_flags; 1961 u_int was_exec = 0; 1962 1963 PMAP_LOCK(); 1964 1965 if (__predict_false(!pmap_initialized)) { 1966 pvo_head = &pmap_pvo_kunmanaged; 1967 pl = &pmap_upvo_pool; 1968 pvo_flags = 0; 1969 pg = NULL; 1970 was_exec = PTE_EXEC; 1971 } else { 1972 pvo_head = pa_to_pvoh(pa, &pg); 1973 pl = &pmap_mpvo_pool; 1974 pvo_flags = PVO_MANAGED; 1975 } 1976 1977 DPRINTFN(ENTER, 1978 ("pmap_enter(%p, 0x%lx, 0x%lx, 0x%x, 0x%x):", 1979 pm, va, pa, prot, flags)); 1980 1981 /* 1982 * If this is a managed page, and it's the first reference to the 1983 * page clear the execness of the page. Otherwise fetch the execness. 1984 */ 1985 if (pg != NULL) 1986 was_exec = pmap_attr_fetch(pg) & PTE_EXEC; 1987 1988 DPRINTFN(ENTER, (" was_exec=%d", was_exec)); 1989 1990 /* 1991 * Assume the page is cache inhibited and access is guarded unless 1992 * it's in our available memory array. If it is in the memory array, 1993 * asssume it's in memory coherent memory. 1994 */ 1995 pte_lo = PTE_IG; 1996 if ((flags & PMAP_NC) == 0) { 1997 for (mp = mem; mp->size; mp++) { 1998 if (pa >= mp->start && pa < mp->start + mp->size) { 1999 pte_lo = PTE_M; 2000 break; 2001 } 2002 } 2003 } 2004 2005 if (prot & VM_PROT_WRITE) 2006 pte_lo |= PTE_BW; 2007 else 2008 pte_lo |= PTE_BR; 2009 2010 /* 2011 * If this was in response to a fault, "pre-fault" the PTE's 2012 * changed/referenced bit appropriately. 2013 */ 2014 if (flags & VM_PROT_WRITE) 2015 pte_lo |= PTE_CHG; 2016 if (flags & VM_PROT_ALL) 2017 pte_lo |= PTE_REF; 2018 2019 /* 2020 * We need to know if this page can be executable 2021 */ 2022 flags |= (prot & VM_PROT_EXECUTE); 2023 2024 /* 2025 * Record mapping for later back-translation and pte spilling. 2026 * This will overwrite any existing mapping. 2027 */ 2028 error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags); 2029 2030 /* 2031 * Flush the real page from the instruction cache if this page is 2032 * mapped executable and cacheable and has not been flushed since 2033 * the last time it was modified. 2034 */ 2035 if (error == 0 && 2036 (flags & VM_PROT_EXECUTE) && 2037 (pte_lo & PTE_I) == 0 && 2038 was_exec == 0) { 2039 DPRINTFN(ENTER, (" syncicache")); 2040 PMAPCOUNT(exec_synced); 2041 pmap_syncicache(pa, PAGE_SIZE); 2042 if (pg != NULL) { 2043 pmap_attr_save(pg, PTE_EXEC); 2044 PMAPCOUNT(exec_cached); 2045 #if defined(DEBUG) || defined(PMAPDEBUG) 2046 if (pmapdebug & PMAPDEBUG_ENTER) 2047 printf(" marked-as-exec"); 2048 else if (pmapdebug & PMAPDEBUG_EXEC) 2049 printf("[pmap_enter: %#lx: marked-as-exec]\n", 2050 VM_PAGE_TO_PHYS(pg)); 2051 2052 #endif 2053 } 2054 } 2055 2056 DPRINTFN(ENTER, (": error=%d\n", error)); 2057 2058 PMAP_UNLOCK(); 2059 2060 return error; 2061 } 2062 2063 void 2064 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot) 2065 { 2066 struct mem_region *mp; 2067 register_t pte_lo; 2068 int error; 2069 2070 #if defined (PPC_OEA64_BRIDGE) 2071 if (va < VM_MIN_KERNEL_ADDRESS) 2072 panic("pmap_kenter_pa: attempt to enter " 2073 "non-kernel address %#lx!", va); 2074 #endif 2075 2076 DPRINTFN(KENTER, 2077 ("pmap_kenter_pa(%#lx,%#lx,%#x)\n", va, pa, prot)); 2078 2079 PMAP_LOCK(); 2080 2081 /* 2082 * Assume the page is cache inhibited and access is guarded unless 2083 * it's in our available memory array. If it is in the memory array, 2084 * asssume it's in memory coherent memory. 2085 */ 2086 pte_lo = PTE_IG; 2087 if ((prot & PMAP_NC) == 0) { 2088 for (mp = mem; mp->size; mp++) { 2089 if (pa >= mp->start && pa < mp->start + mp->size) { 2090 pte_lo = PTE_M; 2091 break; 2092 } 2093 } 2094 } 2095 2096 if (prot & VM_PROT_WRITE) 2097 pte_lo |= PTE_BW; 2098 else 2099 pte_lo |= PTE_BR; 2100 2101 /* 2102 * We don't care about REF/CHG on PVOs on the unmanaged list. 2103 */ 2104 error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool, 2105 &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED); 2106 2107 if (error != 0) 2108 panic("pmap_kenter_pa: failed to enter va %#lx pa %#lx: %d", 2109 va, pa, error); 2110 2111 PMAP_UNLOCK(); 2112 } 2113 2114 void 2115 pmap_kremove(vaddr_t va, vsize_t len) 2116 { 2117 if (va < VM_MIN_KERNEL_ADDRESS) 2118 panic("pmap_kremove: attempt to remove " 2119 "non-kernel address %#lx!", va); 2120 2121 DPRINTFN(KREMOVE,("pmap_kremove(%#lx,%#lx)\n", va, len)); 2122 pmap_remove(pmap_kernel(), va, va + len); 2123 } 2124 2125 /* 2126 * Remove the given range of mapping entries. 2127 */ 2128 void 2129 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva) 2130 { 2131 struct pvo_head pvol; 2132 struct pvo_entry *pvo; 2133 register_t msr; 2134 int pteidx; 2135 2136 PMAP_LOCK(); 2137 LIST_INIT(&pvol); 2138 msr = pmap_interrupts_off(); 2139 for (; va < endva; va += PAGE_SIZE) { 2140 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2141 if (pvo != NULL) { 2142 pmap_pvo_remove(pvo, pteidx, &pvol); 2143 } 2144 } 2145 pmap_interrupts_restore(msr); 2146 pmap_pvo_free_list(&pvol); 2147 PMAP_UNLOCK(); 2148 } 2149 2150 /* 2151 * Get the physical page address for the given pmap/virtual address. 2152 */ 2153 bool 2154 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap) 2155 { 2156 struct pvo_entry *pvo; 2157 register_t msr; 2158 2159 PMAP_LOCK(); 2160 2161 /* 2162 * If this is a kernel pmap lookup, also check the battable 2163 * and if we get a hit, translate the VA to a PA using the 2164 * BAT entries. Don't check for VM_MAX_KERNEL_ADDRESS is 2165 * that will wrap back to 0. 2166 */ 2167 if (pm == pmap_kernel() && 2168 (va < VM_MIN_KERNEL_ADDRESS || 2169 (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) { 2170 KASSERT((va >> ADDR_SR_SHFT) != USER_SR); 2171 #if defined (PPC_OEA) 2172 if ((MFPVR() >> 16) != MPC601) { 2173 register_t batu = battable[va >> ADDR_SR_SHFT].batu; 2174 if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) { 2175 register_t batl = 2176 battable[va >> ADDR_SR_SHFT].batl; 2177 register_t mask = 2178 (~(batu & BAT_BL) << 15) & ~0x1ffffL; 2179 if (pap) 2180 *pap = (batl & mask) | (va & ~mask); 2181 PMAP_UNLOCK(); 2182 return true; 2183 } 2184 } else { 2185 register_t batu = battable[va >> 23].batu; 2186 register_t batl = battable[va >> 23].batl; 2187 register_t sr = iosrtable[va >> ADDR_SR_SHFT]; 2188 if (BAT601_VALID_P(batl) && 2189 BAT601_VA_MATCH_P(batu, batl, va)) { 2190 register_t mask = 2191 (~(batl & BAT601_BSM) << 17) & ~0x1ffffL; 2192 if (pap) 2193 *pap = (batl & mask) | (va & ~mask); 2194 PMAP_UNLOCK(); 2195 return true; 2196 } else if (SR601_VALID_P(sr) && 2197 SR601_PA_MATCH_P(sr, va)) { 2198 if (pap) 2199 *pap = va; 2200 PMAP_UNLOCK(); 2201 return true; 2202 } 2203 } 2204 PMAP_UNLOCK(); 2205 return false; 2206 #elif defined (PPC_OEA64_BRIDGE) 2207 if (va >= SEGMENT_LENGTH) 2208 panic("%s: pm: %s va >= SEGMENT_LENGTH, va: 0x%08lx\n", 2209 __func__, (pm == pmap_kernel() ? "kernel" : "user"), va); 2210 else { 2211 if (pap) 2212 *pap = va; 2213 PMAP_UNLOCK(); 2214 return true; 2215 } 2216 #elif defined (PPC_OEA64) 2217 #error PPC_OEA64 not supported 2218 #endif /* PPC_OEA */ 2219 } 2220 2221 msr = pmap_interrupts_off(); 2222 pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL); 2223 if (pvo != NULL) { 2224 PMAP_PVO_CHECK(pvo); /* sanity check */ 2225 if (pap) 2226 *pap = (pvo->pvo_pte.pte_lo & PTE_RPGN) 2227 | (va & ADDR_POFF); 2228 } 2229 pmap_interrupts_restore(msr); 2230 PMAP_UNLOCK(); 2231 return pvo != NULL; 2232 } 2233 2234 /* 2235 * Lower the protection on the specified range of this pmap. 2236 */ 2237 void 2238 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot) 2239 { 2240 struct pvo_entry *pvo; 2241 volatile struct pte *pt; 2242 register_t msr; 2243 int pteidx; 2244 2245 /* 2246 * Since this routine only downgrades protection, we should 2247 * always be called with at least one bit not set. 2248 */ 2249 KASSERT(prot != VM_PROT_ALL); 2250 2251 /* 2252 * If there is no protection, this is equivalent to 2253 * remove the pmap from the pmap. 2254 */ 2255 if ((prot & VM_PROT_READ) == 0) { 2256 pmap_remove(pm, va, endva); 2257 return; 2258 } 2259 2260 PMAP_LOCK(); 2261 2262 msr = pmap_interrupts_off(); 2263 for (; va < endva; va += PAGE_SIZE) { 2264 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2265 if (pvo == NULL) 2266 continue; 2267 PMAP_PVO_CHECK(pvo); /* sanity check */ 2268 2269 /* 2270 * Revoke executable if asked to do so. 2271 */ 2272 if ((prot & VM_PROT_EXECUTE) == 0) 2273 pvo_clear_exec(pvo); 2274 2275 #if 0 2276 /* 2277 * If the page is already read-only, no change 2278 * needs to be made. 2279 */ 2280 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) 2281 continue; 2282 #endif 2283 /* 2284 * Grab the PTE pointer before we diddle with 2285 * the cached PTE copy. 2286 */ 2287 pt = pmap_pvo_to_pte(pvo, pteidx); 2288 /* 2289 * Change the protection of the page. 2290 */ 2291 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2292 pvo->pvo_pte.pte_lo |= PTE_BR; 2293 2294 /* 2295 * If the PVO is in the page table, update 2296 * that pte at well. 2297 */ 2298 if (pt != NULL) { 2299 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2300 PVO_WHERE(pvo, PMAP_PROTECT); 2301 PMAPCOUNT(ptes_changed); 2302 } 2303 2304 PMAP_PVO_CHECK(pvo); /* sanity check */ 2305 } 2306 pmap_interrupts_restore(msr); 2307 PMAP_UNLOCK(); 2308 } 2309 2310 void 2311 pmap_unwire(pmap_t pm, vaddr_t va) 2312 { 2313 struct pvo_entry *pvo; 2314 register_t msr; 2315 2316 PMAP_LOCK(); 2317 msr = pmap_interrupts_off(); 2318 pvo = pmap_pvo_find_va(pm, va, NULL); 2319 if (pvo != NULL) { 2320 if (PVO_WIRED_P(pvo)) { 2321 pvo->pvo_vaddr &= ~PVO_WIRED; 2322 pm->pm_stats.wired_count--; 2323 } 2324 PMAP_PVO_CHECK(pvo); /* sanity check */ 2325 } 2326 pmap_interrupts_restore(msr); 2327 PMAP_UNLOCK(); 2328 } 2329 2330 /* 2331 * Lower the protection on the specified physical page. 2332 */ 2333 void 2334 pmap_page_protect(struct vm_page *pg, vm_prot_t prot) 2335 { 2336 struct pvo_head *pvo_head, pvol; 2337 struct pvo_entry *pvo, *next_pvo; 2338 volatile struct pte *pt; 2339 register_t msr; 2340 2341 PMAP_LOCK(); 2342 2343 KASSERT(prot != VM_PROT_ALL); 2344 LIST_INIT(&pvol); 2345 msr = pmap_interrupts_off(); 2346 2347 /* 2348 * When UVM reuses a page, it does a pmap_page_protect with 2349 * VM_PROT_NONE. At that point, we can clear the exec flag 2350 * since we know the page will have different contents. 2351 */ 2352 if ((prot & VM_PROT_READ) == 0) { 2353 DPRINTFN(EXEC, ("[pmap_page_protect: %#lx: clear-exec]\n", 2354 VM_PAGE_TO_PHYS(pg))); 2355 if (pmap_attr_fetch(pg) & PTE_EXEC) { 2356 PMAPCOUNT(exec_uncached_page_protect); 2357 pmap_attr_clear(pg, PTE_EXEC); 2358 } 2359 } 2360 2361 pvo_head = vm_page_to_pvoh(pg); 2362 for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) { 2363 next_pvo = LIST_NEXT(pvo, pvo_vlink); 2364 PMAP_PVO_CHECK(pvo); /* sanity check */ 2365 2366 /* 2367 * Downgrading to no mapping at all, we just remove the entry. 2368 */ 2369 if ((prot & VM_PROT_READ) == 0) { 2370 pmap_pvo_remove(pvo, -1, &pvol); 2371 continue; 2372 } 2373 2374 /* 2375 * If EXEC permission is being revoked, just clear the 2376 * flag in the PVO. 2377 */ 2378 if ((prot & VM_PROT_EXECUTE) == 0) 2379 pvo_clear_exec(pvo); 2380 2381 /* 2382 * If this entry is already RO, don't diddle with the 2383 * page table. 2384 */ 2385 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) { 2386 PMAP_PVO_CHECK(pvo); 2387 continue; 2388 } 2389 2390 /* 2391 * Grab the PTE before the we diddle the bits so 2392 * pvo_to_pte can verify the pte contents are as 2393 * expected. 2394 */ 2395 pt = pmap_pvo_to_pte(pvo, -1); 2396 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2397 pvo->pvo_pte.pte_lo |= PTE_BR; 2398 if (pt != NULL) { 2399 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2400 PVO_WHERE(pvo, PMAP_PAGE_PROTECT); 2401 PMAPCOUNT(ptes_changed); 2402 } 2403 PMAP_PVO_CHECK(pvo); /* sanity check */ 2404 } 2405 pmap_interrupts_restore(msr); 2406 pmap_pvo_free_list(&pvol); 2407 2408 PMAP_UNLOCK(); 2409 } 2410 2411 /* 2412 * Activate the address space for the specified process. If the process 2413 * is the current process, load the new MMU context. 2414 */ 2415 void 2416 pmap_activate(struct lwp *l) 2417 { 2418 struct pcb *pcb = &l->l_addr->u_pcb; 2419 pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap; 2420 2421 DPRINTFN(ACTIVATE, 2422 ("pmap_activate: lwp %p (curlwp %p)\n", l, curlwp)); 2423 2424 /* 2425 * XXX Normally performed in cpu_fork(). 2426 */ 2427 pcb->pcb_pm = pmap; 2428 2429 /* 2430 * In theory, the SR registers need only be valid on return 2431 * to user space wait to do them there. 2432 */ 2433 if (l == curlwp) { 2434 /* Store pointer to new current pmap. */ 2435 curpm = pmap; 2436 } 2437 } 2438 2439 /* 2440 * Deactivate the specified process's address space. 2441 */ 2442 void 2443 pmap_deactivate(struct lwp *l) 2444 { 2445 } 2446 2447 bool 2448 pmap_query_bit(struct vm_page *pg, int ptebit) 2449 { 2450 struct pvo_entry *pvo; 2451 volatile struct pte *pt; 2452 register_t msr; 2453 2454 PMAP_LOCK(); 2455 2456 if (pmap_attr_fetch(pg) & ptebit) { 2457 PMAP_UNLOCK(); 2458 return true; 2459 } 2460 2461 msr = pmap_interrupts_off(); 2462 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2463 PMAP_PVO_CHECK(pvo); /* sanity check */ 2464 /* 2465 * See if we saved the bit off. If so cache, it and return 2466 * success. 2467 */ 2468 if (pvo->pvo_pte.pte_lo & ptebit) { 2469 pmap_attr_save(pg, ptebit); 2470 PMAP_PVO_CHECK(pvo); /* sanity check */ 2471 pmap_interrupts_restore(msr); 2472 PMAP_UNLOCK(); 2473 return true; 2474 } 2475 } 2476 /* 2477 * No luck, now go thru the hard part of looking at the ptes 2478 * themselves. Sync so any pending REF/CHG bits are flushed 2479 * to the PTEs. 2480 */ 2481 SYNC(); 2482 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2483 PMAP_PVO_CHECK(pvo); /* sanity check */ 2484 /* 2485 * See if this pvo have a valid PTE. If so, fetch the 2486 * REF/CHG bits from the valid PTE. If the appropriate 2487 * ptebit is set, cache, it and return success. 2488 */ 2489 pt = pmap_pvo_to_pte(pvo, -1); 2490 if (pt != NULL) { 2491 pmap_pte_synch(pt, &pvo->pvo_pte); 2492 if (pvo->pvo_pte.pte_lo & ptebit) { 2493 pmap_attr_save(pg, ptebit); 2494 PMAP_PVO_CHECK(pvo); /* sanity check */ 2495 pmap_interrupts_restore(msr); 2496 PMAP_UNLOCK(); 2497 return true; 2498 } 2499 } 2500 } 2501 pmap_interrupts_restore(msr); 2502 PMAP_UNLOCK(); 2503 return false; 2504 } 2505 2506 bool 2507 pmap_clear_bit(struct vm_page *pg, int ptebit) 2508 { 2509 struct pvo_head *pvoh = vm_page_to_pvoh(pg); 2510 struct pvo_entry *pvo; 2511 volatile struct pte *pt; 2512 register_t msr; 2513 int rv = 0; 2514 2515 PMAP_LOCK(); 2516 msr = pmap_interrupts_off(); 2517 2518 /* 2519 * Fetch the cache value 2520 */ 2521 rv |= pmap_attr_fetch(pg); 2522 2523 /* 2524 * Clear the cached value. 2525 */ 2526 pmap_attr_clear(pg, ptebit); 2527 2528 /* 2529 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we 2530 * can reset the right ones). Note that since the pvo entries and 2531 * list heads are accessed via BAT0 and are never placed in the 2532 * page table, we don't have to worry about further accesses setting 2533 * the REF/CHG bits. 2534 */ 2535 SYNC(); 2536 2537 /* 2538 * For each pvo entry, clear pvo's ptebit. If this pvo have a 2539 * valid PTE. If so, clear the ptebit from the valid PTE. 2540 */ 2541 LIST_FOREACH(pvo, pvoh, pvo_vlink) { 2542 PMAP_PVO_CHECK(pvo); /* sanity check */ 2543 pt = pmap_pvo_to_pte(pvo, -1); 2544 if (pt != NULL) { 2545 /* 2546 * Only sync the PTE if the bit we are looking 2547 * for is not already set. 2548 */ 2549 if ((pvo->pvo_pte.pte_lo & ptebit) == 0) 2550 pmap_pte_synch(pt, &pvo->pvo_pte); 2551 /* 2552 * If the bit we are looking for was already set, 2553 * clear that bit in the pte. 2554 */ 2555 if (pvo->pvo_pte.pte_lo & ptebit) 2556 pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit); 2557 } 2558 rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF); 2559 pvo->pvo_pte.pte_lo &= ~ptebit; 2560 PMAP_PVO_CHECK(pvo); /* sanity check */ 2561 } 2562 pmap_interrupts_restore(msr); 2563 2564 /* 2565 * If we are clearing the modify bit and this page was marked EXEC 2566 * and the user of the page thinks the page was modified, then we 2567 * need to clean it from the icache if it's mapped or clear the EXEC 2568 * bit if it's not mapped. The page itself might not have the CHG 2569 * bit set if the modification was done via DMA to the page. 2570 */ 2571 if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) { 2572 if (LIST_EMPTY(pvoh)) { 2573 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: clear-exec]\n", 2574 VM_PAGE_TO_PHYS(pg))); 2575 pmap_attr_clear(pg, PTE_EXEC); 2576 PMAPCOUNT(exec_uncached_clear_modify); 2577 } else { 2578 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: syncicache]\n", 2579 VM_PAGE_TO_PHYS(pg))); 2580 pmap_syncicache(VM_PAGE_TO_PHYS(pg), PAGE_SIZE); 2581 PMAPCOUNT(exec_synced_clear_modify); 2582 } 2583 } 2584 PMAP_UNLOCK(); 2585 return (rv & ptebit) != 0; 2586 } 2587 2588 void 2589 pmap_procwr(struct proc *p, vaddr_t va, size_t len) 2590 { 2591 struct pvo_entry *pvo; 2592 size_t offset = va & ADDR_POFF; 2593 int s; 2594 2595 PMAP_LOCK(); 2596 s = splvm(); 2597 while (len > 0) { 2598 size_t seglen = PAGE_SIZE - offset; 2599 if (seglen > len) 2600 seglen = len; 2601 pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL); 2602 if (pvo != NULL && PVO_EXECUTABLE_P(pvo)) { 2603 pmap_syncicache( 2604 (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen); 2605 PMAP_PVO_CHECK(pvo); 2606 } 2607 va += seglen; 2608 len -= seglen; 2609 offset = 0; 2610 } 2611 splx(s); 2612 PMAP_UNLOCK(); 2613 } 2614 2615 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 2616 void 2617 pmap_pte_print(volatile struct pte *pt) 2618 { 2619 printf("PTE %p: ", pt); 2620 2621 #if defined(PPC_OEA) 2622 /* High word: */ 2623 printf("0x%08lx: [", pt->pte_hi); 2624 #elif defined (PPC_OEA64_BRIDGE) 2625 printf("0x%016llx: [", pt->pte_hi); 2626 #else /* PPC_OEA64 */ 2627 printf("0x%016lx: [", pt->pte_hi); 2628 #endif /* PPC_OEA */ 2629 2630 printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i'); 2631 printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-'); 2632 2633 #if defined (PPC_OEA) 2634 printf("0x%06lx 0x%02lx", 2635 (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT, 2636 pt->pte_hi & PTE_API); 2637 printf(" (va 0x%08lx)] ", pmap_pte_to_va(pt)); 2638 #elif defined (PPC_OEA64) 2639 printf("0x%06lx 0x%02lx", 2640 (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT, 2641 pt->pte_hi & PTE_API); 2642 printf(" (va 0x%016lx)] ", pmap_pte_to_va(pt)); 2643 #else 2644 /* PPC_OEA64_BRIDGE */ 2645 printf("0x%06llx 0x%02llx", 2646 (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT, 2647 pt->pte_hi & PTE_API); 2648 printf(" (va 0x%08lx)] ", pmap_pte_to_va(pt)); 2649 #endif /* PPC_OEA */ 2650 2651 /* Low word: */ 2652 #if defined (PPC_OEA) 2653 printf(" 0x%08lx: [", pt->pte_lo); 2654 printf("0x%05lx... ", pt->pte_lo >> 12); 2655 #elif defined (PPC_OEA64) 2656 printf(" 0x%016lx: [", pt->pte_lo); 2657 printf("0x%012lx... ", pt->pte_lo >> 12); 2658 #else /* PPC_OEA64_BRIDGE */ 2659 printf(" 0x%016llx: [", pt->pte_lo); 2660 printf("0x%012llx... ", pt->pte_lo >> 12); 2661 #endif 2662 printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u'); 2663 printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n'); 2664 printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.'); 2665 printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.'); 2666 printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.'); 2667 printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.'); 2668 switch (pt->pte_lo & PTE_PP) { 2669 case PTE_BR: printf("br]\n"); break; 2670 case PTE_BW: printf("bw]\n"); break; 2671 case PTE_SO: printf("so]\n"); break; 2672 case PTE_SW: printf("sw]\n"); break; 2673 } 2674 } 2675 #endif 2676 2677 #if defined(DDB) 2678 void 2679 pmap_pteg_check(void) 2680 { 2681 volatile struct pte *pt; 2682 int i; 2683 int ptegidx; 2684 u_int p_valid = 0; 2685 u_int s_valid = 0; 2686 u_int invalid = 0; 2687 2688 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2689 for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) { 2690 if (pt->pte_hi & PTE_VALID) { 2691 if (pt->pte_hi & PTE_HID) 2692 s_valid++; 2693 else 2694 { 2695 p_valid++; 2696 } 2697 } else 2698 invalid++; 2699 } 2700 } 2701 printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n", 2702 p_valid, p_valid, s_valid, s_valid, 2703 invalid, invalid); 2704 } 2705 2706 void 2707 pmap_print_mmuregs(void) 2708 { 2709 int i; 2710 u_int cpuvers; 2711 #ifndef PPC_OEA64 2712 vaddr_t addr; 2713 register_t soft_sr[16]; 2714 #endif 2715 #if defined (PPC_OEA) && !defined (PPC_OEA64) && !defined (PPC_OEA64_BRIDGE) 2716 struct bat soft_ibat[4]; 2717 struct bat soft_dbat[4]; 2718 #endif 2719 register_t sdr1; 2720 2721 cpuvers = MFPVR() >> 16; 2722 __asm volatile ("mfsdr1 %0" : "=r"(sdr1)); 2723 #ifndef PPC_OEA64 2724 addr = 0; 2725 for (i = 0; i < 16; i++) { 2726 soft_sr[i] = MFSRIN(addr); 2727 addr += (1 << ADDR_SR_SHFT); 2728 } 2729 #endif 2730 2731 #if defined(PPC_OEA) && !defined (PPC_OEA64) && !defined (PPC_OEA64_BRIDGE) 2732 /* read iBAT (601: uBAT) registers */ 2733 __asm volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu)); 2734 __asm volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl)); 2735 __asm volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu)); 2736 __asm volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl)); 2737 __asm volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu)); 2738 __asm volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl)); 2739 __asm volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu)); 2740 __asm volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl)); 2741 2742 2743 if (cpuvers != MPC601) { 2744 /* read dBAT registers */ 2745 __asm volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu)); 2746 __asm volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl)); 2747 __asm volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu)); 2748 __asm volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl)); 2749 __asm volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu)); 2750 __asm volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl)); 2751 __asm volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu)); 2752 __asm volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl)); 2753 } 2754 #endif 2755 2756 printf("SDR1:\t0x%lx\n", (long) sdr1); 2757 #ifndef PPC_OEA64 2758 printf("SR[]:\t"); 2759 for (i = 0; i < 4; i++) 2760 printf("0x%08lx, ", (long) soft_sr[i]); 2761 printf("\n\t"); 2762 for ( ; i < 8; i++) 2763 printf("0x%08lx, ", (long) soft_sr[i]); 2764 printf("\n\t"); 2765 for ( ; i < 12; i++) 2766 printf("0x%08lx, ", (long) soft_sr[i]); 2767 printf("\n\t"); 2768 for ( ; i < 16; i++) 2769 printf("0x%08lx, ", (long) soft_sr[i]); 2770 printf("\n"); 2771 #endif 2772 2773 #if defined(PPC_OEA) && !defined (PPC_OEA64) && !defined (PPC_OEA64_BRIDGE) 2774 printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i'); 2775 for (i = 0; i < 4; i++) { 2776 printf("0x%08lx 0x%08lx, ", 2777 soft_ibat[i].batu, soft_ibat[i].batl); 2778 if (i == 1) 2779 printf("\n\t"); 2780 } 2781 if (cpuvers != MPC601) { 2782 printf("\ndBAT[]:\t"); 2783 for (i = 0; i < 4; i++) { 2784 printf("0x%08lx 0x%08lx, ", 2785 soft_dbat[i].batu, soft_dbat[i].batl); 2786 if (i == 1) 2787 printf("\n\t"); 2788 } 2789 } 2790 printf("\n"); 2791 #endif /* PPC_OEA... */ 2792 } 2793 2794 void 2795 pmap_print_pte(pmap_t pm, vaddr_t va) 2796 { 2797 struct pvo_entry *pvo; 2798 volatile struct pte *pt; 2799 int pteidx; 2800 2801 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2802 if (pvo != NULL) { 2803 pt = pmap_pvo_to_pte(pvo, pteidx); 2804 if (pt != NULL) { 2805 #if defined (PPC_OEA) || defined (PPC_OEA64) 2806 printf("VA %#lx -> %p -> %s %#lx, %#lx\n", 2807 va, pt, 2808 pt->pte_hi & PTE_HID ? "(sec)" : "(pri)", 2809 pt->pte_hi, pt->pte_lo); 2810 #else /* PPC_OEA64_BRIDGE */ 2811 printf("VA %#lx -> %p -> %s %#llx, %#llx\n", 2812 va, pt, 2813 pt->pte_hi & PTE_HID ? "(sec)" : "(pri)", 2814 pt->pte_hi, pt->pte_lo); 2815 #endif 2816 } else { 2817 printf("No valid PTE found\n"); 2818 } 2819 } else { 2820 printf("Address not in pmap\n"); 2821 } 2822 } 2823 2824 void 2825 pmap_pteg_dist(void) 2826 { 2827 struct pvo_entry *pvo; 2828 int ptegidx; 2829 int depth; 2830 int max_depth = 0; 2831 unsigned int depths[64]; 2832 2833 memset(depths, 0, sizeof(depths)); 2834 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2835 depth = 0; 2836 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2837 depth++; 2838 } 2839 if (depth > max_depth) 2840 max_depth = depth; 2841 if (depth > 63) 2842 depth = 63; 2843 depths[depth]++; 2844 } 2845 2846 for (depth = 0; depth < 64; depth++) { 2847 printf(" [%2d]: %8u", depth, depths[depth]); 2848 if ((depth & 3) == 3) 2849 printf("\n"); 2850 if (depth == max_depth) 2851 break; 2852 } 2853 if ((depth & 3) != 3) 2854 printf("\n"); 2855 printf("Max depth found was %d\n", max_depth); 2856 } 2857 #endif /* DEBUG */ 2858 2859 #if defined(PMAPCHECK) || defined(DEBUG) 2860 void 2861 pmap_pvo_verify(void) 2862 { 2863 int ptegidx; 2864 int s; 2865 2866 s = splvm(); 2867 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2868 struct pvo_entry *pvo; 2869 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2870 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 2871 panic("pmap_pvo_verify: invalid pvo %p " 2872 "on list %#x", pvo, ptegidx); 2873 pmap_pvo_check(pvo); 2874 } 2875 } 2876 splx(s); 2877 } 2878 #endif /* PMAPCHECK */ 2879 2880 2881 void * 2882 pmap_pool_ualloc(struct pool *pp, int flags) 2883 { 2884 struct pvo_page *pvop; 2885 2886 if (uvm.page_init_done != true) { 2887 return (void *) uvm_pageboot_alloc(PAGE_SIZE); 2888 } 2889 2890 PMAP_LOCK(); 2891 pvop = SIMPLEQ_FIRST(&pmap_upvop_head); 2892 if (pvop != NULL) { 2893 pmap_upvop_free--; 2894 SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link); 2895 PMAP_UNLOCK(); 2896 return pvop; 2897 } 2898 PMAP_UNLOCK(); 2899 return pmap_pool_malloc(pp, flags); 2900 } 2901 2902 void * 2903 pmap_pool_malloc(struct pool *pp, int flags) 2904 { 2905 struct pvo_page *pvop; 2906 struct vm_page *pg; 2907 2908 PMAP_LOCK(); 2909 pvop = SIMPLEQ_FIRST(&pmap_mpvop_head); 2910 if (pvop != NULL) { 2911 pmap_mpvop_free--; 2912 SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link); 2913 PMAP_UNLOCK(); 2914 return pvop; 2915 } 2916 PMAP_UNLOCK(); 2917 again: 2918 pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE, 2919 UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256); 2920 if (__predict_false(pg == NULL)) { 2921 if (flags & PR_WAITOK) { 2922 uvm_wait("plpg"); 2923 goto again; 2924 } else { 2925 return (0); 2926 } 2927 } 2928 return (void *) VM_PAGE_TO_PHYS(pg); 2929 } 2930 2931 void 2932 pmap_pool_ufree(struct pool *pp, void *va) 2933 { 2934 struct pvo_page *pvop; 2935 #if 0 2936 if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) { 2937 pmap_pool_mfree(va, size, tag); 2938 return; 2939 } 2940 #endif 2941 PMAP_LOCK(); 2942 pvop = va; 2943 SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link); 2944 pmap_upvop_free++; 2945 if (pmap_upvop_free > pmap_upvop_maxfree) 2946 pmap_upvop_maxfree = pmap_upvop_free; 2947 PMAP_UNLOCK(); 2948 } 2949 2950 void 2951 pmap_pool_mfree(struct pool *pp, void *va) 2952 { 2953 struct pvo_page *pvop; 2954 2955 PMAP_LOCK(); 2956 pvop = va; 2957 SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link); 2958 pmap_mpvop_free++; 2959 if (pmap_mpvop_free > pmap_mpvop_maxfree) 2960 pmap_mpvop_maxfree = pmap_mpvop_free; 2961 PMAP_UNLOCK(); 2962 #if 0 2963 uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va)); 2964 #endif 2965 } 2966 2967 /* 2968 * This routine in bootstraping to steal to-be-managed memory (which will 2969 * then be unmanaged). We use it to grab from the first 256MB for our 2970 * pmap needs and above 256MB for other stuff. 2971 */ 2972 vaddr_t 2973 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp) 2974 { 2975 vsize_t size; 2976 vaddr_t va; 2977 paddr_t pa = 0; 2978 int npgs, bank; 2979 struct vm_physseg *ps; 2980 2981 if (uvm.page_init_done == true) 2982 panic("pmap_steal_memory: called _after_ bootstrap"); 2983 2984 *vstartp = VM_MIN_KERNEL_ADDRESS; 2985 *vendp = VM_MAX_KERNEL_ADDRESS; 2986 2987 size = round_page(vsize); 2988 npgs = atop(size); 2989 2990 /* 2991 * PA 0 will never be among those given to UVM so we can use it 2992 * to indicate we couldn't steal any memory. 2993 */ 2994 for (ps = vm_physmem, bank = 0; bank < vm_nphysseg; bank++, ps++) { 2995 if (ps->free_list == VM_FREELIST_FIRST256 && 2996 ps->avail_end - ps->avail_start >= npgs) { 2997 pa = ptoa(ps->avail_start); 2998 break; 2999 } 3000 } 3001 3002 if (pa == 0) 3003 panic("pmap_steal_memory: no approriate memory to steal!"); 3004 3005 ps->avail_start += npgs; 3006 ps->start += npgs; 3007 3008 /* 3009 * If we've used up all the pages in the segment, remove it and 3010 * compact the list. 3011 */ 3012 if (ps->avail_start == ps->end) { 3013 /* 3014 * If this was the last one, then a very bad thing has occurred 3015 */ 3016 if (--vm_nphysseg == 0) 3017 panic("pmap_steal_memory: out of memory!"); 3018 3019 printf("pmap_steal_memory: consumed bank %d\n", bank); 3020 for (; bank < vm_nphysseg; bank++, ps++) { 3021 ps[0] = ps[1]; 3022 } 3023 } 3024 3025 va = (vaddr_t) pa; 3026 memset((void *) va, 0, size); 3027 pmap_pages_stolen += npgs; 3028 #ifdef DEBUG 3029 if (pmapdebug && npgs > 1) { 3030 u_int cnt = 0; 3031 for (bank = 0, ps = vm_physmem; bank < vm_nphysseg; bank++, ps++) 3032 cnt += ps->avail_end - ps->avail_start; 3033 printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n", 3034 npgs, pmap_pages_stolen, cnt); 3035 } 3036 #endif 3037 3038 return va; 3039 } 3040 3041 /* 3042 * Find a chuck of memory with right size and alignment. 3043 */ 3044 void * 3045 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end) 3046 { 3047 struct mem_region *mp; 3048 paddr_t s, e; 3049 int i, j; 3050 3051 size = round_page(size); 3052 3053 DPRINTFN(BOOT, 3054 ("pmap_boot_find_memory: size=%lx, alignment=%lx, at_end=%d", 3055 size, alignment, at_end)); 3056 3057 if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0) 3058 panic("pmap_boot_find_memory: invalid alignment %lx", 3059 alignment); 3060 3061 if (at_end) { 3062 if (alignment != PAGE_SIZE) 3063 panic("pmap_boot_find_memory: invalid ending " 3064 "alignment %lx", alignment); 3065 3066 for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) { 3067 s = mp->start + mp->size - size; 3068 if (s >= mp->start && mp->size >= size) { 3069 DPRINTFN(BOOT,(": %lx\n", s)); 3070 DPRINTFN(BOOT, 3071 ("pmap_boot_find_memory: b-avail[%d] start " 3072 "0x%lx size 0x%lx\n", mp - avail, 3073 mp->start, mp->size)); 3074 mp->size -= size; 3075 DPRINTFN(BOOT, 3076 ("pmap_boot_find_memory: a-avail[%d] start " 3077 "0x%lx size 0x%lx\n", mp - avail, 3078 mp->start, mp->size)); 3079 return (void *) s; 3080 } 3081 } 3082 panic("pmap_boot_find_memory: no available memory"); 3083 } 3084 3085 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 3086 s = (mp->start + alignment - 1) & ~(alignment-1); 3087 e = s + size; 3088 3089 /* 3090 * Is the calculated region entirely within the region? 3091 */ 3092 if (s < mp->start || e > mp->start + mp->size) 3093 continue; 3094 3095 DPRINTFN(BOOT,(": %lx\n", s)); 3096 if (s == mp->start) { 3097 /* 3098 * If the block starts at the beginning of region, 3099 * adjust the size & start. (the region may now be 3100 * zero in length) 3101 */ 3102 DPRINTFN(BOOT, 3103 ("pmap_boot_find_memory: b-avail[%d] start " 3104 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 3105 mp->start += size; 3106 mp->size -= size; 3107 DPRINTFN(BOOT, 3108 ("pmap_boot_find_memory: a-avail[%d] start " 3109 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 3110 } else if (e == mp->start + mp->size) { 3111 /* 3112 * If the block starts at the beginning of region, 3113 * adjust only the size. 3114 */ 3115 DPRINTFN(BOOT, 3116 ("pmap_boot_find_memory: b-avail[%d] start " 3117 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 3118 mp->size -= size; 3119 DPRINTFN(BOOT, 3120 ("pmap_boot_find_memory: a-avail[%d] start " 3121 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 3122 } else { 3123 /* 3124 * Block is in the middle of the region, so we 3125 * have to split it in two. 3126 */ 3127 for (j = avail_cnt; j > i + 1; j--) { 3128 avail[j] = avail[j-1]; 3129 } 3130 DPRINTFN(BOOT, 3131 ("pmap_boot_find_memory: b-avail[%d] start " 3132 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 3133 mp[1].start = e; 3134 mp[1].size = mp[0].start + mp[0].size - e; 3135 mp[0].size = s - mp[0].start; 3136 avail_cnt++; 3137 for (; i < avail_cnt; i++) { 3138 DPRINTFN(BOOT, 3139 ("pmap_boot_find_memory: a-avail[%d] " 3140 "start 0x%lx size 0x%lx\n", i, 3141 avail[i].start, avail[i].size)); 3142 } 3143 } 3144 return (void *) s; 3145 } 3146 panic("pmap_boot_find_memory: not enough memory for " 3147 "%lx/%lx allocation?", size, alignment); 3148 } 3149 3150 /* XXXSL: we dont have any BATs to do this, map in Segment 0 1:1 using page tables */ 3151 #if defined (PPC_OEA64_BRIDGE) 3152 int 3153 pmap_setup_segment0_map(int use_large_pages, ...) 3154 { 3155 vaddr_t va; 3156 3157 register_t pte_lo = 0x0; 3158 int ptegidx = 0, i = 0; 3159 struct pte pte; 3160 va_list ap; 3161 3162 /* Coherent + Supervisor RW, no user access */ 3163 pte_lo = PTE_M; 3164 3165 /* XXXSL 3166 * Map in 1st segment 1:1, we'll be careful not to spill kernel entries later, 3167 * these have to take priority. 3168 */ 3169 for (va = 0x0; va < SEGMENT_LENGTH; va += 0x1000) { 3170 ptegidx = va_to_pteg(pmap_kernel(), va); 3171 pmap_pte_create(&pte, pmap_kernel(), va, va | pte_lo); 3172 i = pmap_pte_insert(ptegidx, &pte); 3173 } 3174 3175 va_start(ap, use_large_pages); 3176 while (1) { 3177 paddr_t pa; 3178 size_t size; 3179 3180 va = va_arg(ap, vaddr_t); 3181 3182 if (va == 0) 3183 break; 3184 3185 pa = va_arg(ap, paddr_t); 3186 size = va_arg(ap, size_t); 3187 3188 for (; va < (va + size); va += 0x1000, pa += 0x1000) { 3189 #if 0 3190 printf("%s: Inserting: va: 0x%08lx, pa: 0x%08lx\n", __func__, va, pa); 3191 #endif 3192 ptegidx = va_to_pteg(pmap_kernel(), va); 3193 pmap_pte_create(&pte, pmap_kernel(), va, pa | pte_lo); 3194 i = pmap_pte_insert(ptegidx, &pte); 3195 } 3196 } 3197 3198 TLBSYNC(); 3199 SYNC(); 3200 return (0); 3201 } 3202 #endif /* PPC_OEA64_BRIDGE */ 3203 3204 /* 3205 * This is not part of the defined PMAP interface and is specific to the 3206 * PowerPC architecture. This is called during initppc, before the system 3207 * is really initialized. 3208 */ 3209 void 3210 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend) 3211 { 3212 struct mem_region *mp, tmp; 3213 paddr_t s, e; 3214 psize_t size; 3215 int i, j; 3216 3217 /* 3218 * Get memory. 3219 */ 3220 mem_regions(&mem, &avail); 3221 #if defined(DEBUG) 3222 if (pmapdebug & PMAPDEBUG_BOOT) { 3223 printf("pmap_bootstrap: memory configuration:\n"); 3224 for (mp = mem; mp->size; mp++) { 3225 printf("pmap_bootstrap: mem start 0x%lx size 0x%lx\n", 3226 mp->start, mp->size); 3227 } 3228 for (mp = avail; mp->size; mp++) { 3229 printf("pmap_bootstrap: avail start 0x%lx size 0x%lx\n", 3230 mp->start, mp->size); 3231 } 3232 } 3233 #endif 3234 3235 /* 3236 * Find out how much physical memory we have and in how many chunks. 3237 */ 3238 for (mem_cnt = 0, mp = mem; mp->size; mp++) { 3239 if (mp->start >= pmap_memlimit) 3240 continue; 3241 if (mp->start + mp->size > pmap_memlimit) { 3242 size = pmap_memlimit - mp->start; 3243 physmem += btoc(size); 3244 } else { 3245 physmem += btoc(mp->size); 3246 } 3247 mem_cnt++; 3248 } 3249 3250 /* 3251 * Count the number of available entries. 3252 */ 3253 for (avail_cnt = 0, mp = avail; mp->size; mp++) 3254 avail_cnt++; 3255 3256 /* 3257 * Page align all regions. 3258 */ 3259 kernelstart = trunc_page(kernelstart); 3260 kernelend = round_page(kernelend); 3261 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 3262 s = round_page(mp->start); 3263 mp->size -= (s - mp->start); 3264 mp->size = trunc_page(mp->size); 3265 mp->start = s; 3266 e = mp->start + mp->size; 3267 3268 DPRINTFN(BOOT, 3269 ("pmap_bootstrap: b-avail[%d] start 0x%lx size 0x%lx\n", 3270 i, mp->start, mp->size)); 3271 3272 /* 3273 * Don't allow the end to run beyond our artificial limit 3274 */ 3275 if (e > pmap_memlimit) 3276 e = pmap_memlimit; 3277 3278 /* 3279 * Is this region empty or strange? skip it. 3280 */ 3281 if (e <= s) { 3282 mp->start = 0; 3283 mp->size = 0; 3284 continue; 3285 } 3286 3287 /* 3288 * Does this overlap the beginning of kernel? 3289 * Does extend past the end of the kernel? 3290 */ 3291 else if (s < kernelstart && e > kernelstart) { 3292 if (e > kernelend) { 3293 avail[avail_cnt].start = kernelend; 3294 avail[avail_cnt].size = e - kernelend; 3295 avail_cnt++; 3296 } 3297 mp->size = kernelstart - s; 3298 } 3299 /* 3300 * Check whether this region overlaps the end of the kernel. 3301 */ 3302 else if (s < kernelend && e > kernelend) { 3303 mp->start = kernelend; 3304 mp->size = e - kernelend; 3305 } 3306 /* 3307 * Look whether this regions is completely inside the kernel. 3308 * Nuke it if it does. 3309 */ 3310 else if (s >= kernelstart && e <= kernelend) { 3311 mp->start = 0; 3312 mp->size = 0; 3313 } 3314 /* 3315 * If the user imposed a memory limit, enforce it. 3316 */ 3317 else if (s >= pmap_memlimit) { 3318 mp->start = -PAGE_SIZE; /* let's know why */ 3319 mp->size = 0; 3320 } 3321 else { 3322 mp->start = s; 3323 mp->size = e - s; 3324 } 3325 DPRINTFN(BOOT, 3326 ("pmap_bootstrap: a-avail[%d] start 0x%lx size 0x%lx\n", 3327 i, mp->start, mp->size)); 3328 } 3329 3330 /* 3331 * Move (and uncount) all the null return to the end. 3332 */ 3333 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 3334 if (mp->size == 0) { 3335 tmp = avail[i]; 3336 avail[i] = avail[--avail_cnt]; 3337 avail[avail_cnt] = avail[i]; 3338 } 3339 } 3340 3341 /* 3342 * (Bubble)sort them into asecnding order. 3343 */ 3344 for (i = 0; i < avail_cnt; i++) { 3345 for (j = i + 1; j < avail_cnt; j++) { 3346 if (avail[i].start > avail[j].start) { 3347 tmp = avail[i]; 3348 avail[i] = avail[j]; 3349 avail[j] = tmp; 3350 } 3351 } 3352 } 3353 3354 /* 3355 * Make sure they don't overlap. 3356 */ 3357 for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) { 3358 if (mp[0].start + mp[0].size > mp[1].start) { 3359 mp[0].size = mp[1].start - mp[0].start; 3360 } 3361 DPRINTFN(BOOT, 3362 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 3363 i, mp->start, mp->size)); 3364 } 3365 DPRINTFN(BOOT, 3366 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 3367 i, mp->start, mp->size)); 3368 3369 #ifdef PTEGCOUNT 3370 pmap_pteg_cnt = PTEGCOUNT; 3371 #else /* PTEGCOUNT */ 3372 3373 pmap_pteg_cnt = 0x1000; 3374 3375 while (pmap_pteg_cnt < physmem) 3376 pmap_pteg_cnt <<= 1; 3377 3378 pmap_pteg_cnt >>= 1; 3379 #endif /* PTEGCOUNT */ 3380 3381 #ifdef DEBUG 3382 DPRINTFN(BOOT, 3383 ("pmap_pteg_cnt: 0x%x\n", pmap_pteg_cnt)); 3384 #endif 3385 3386 /* 3387 * Find suitably aligned memory for PTEG hash table. 3388 */ 3389 size = pmap_pteg_cnt * sizeof(struct pteg); 3390 pmap_pteg_table = pmap_boot_find_memory(size, size, 0); 3391 3392 #ifdef DEBUG 3393 DPRINTFN(BOOT, 3394 ("PTEG cnt: 0x%x HTAB size: 0x%08x bytes, address: %p\n", pmap_pteg_cnt, (unsigned int)size, pmap_pteg_table)); 3395 #endif 3396 3397 3398 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3399 if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH) 3400 panic("pmap_bootstrap: pmap_pteg_table end (%p + %lx) > 256MB", 3401 pmap_pteg_table, size); 3402 #endif 3403 3404 memset(__UNVOLATILE(pmap_pteg_table), 0, 3405 pmap_pteg_cnt * sizeof(struct pteg)); 3406 pmap_pteg_mask = pmap_pteg_cnt - 1; 3407 3408 /* 3409 * We cannot do pmap_steal_memory here since UVM hasn't been loaded 3410 * with pages. So we just steal them before giving them to UVM. 3411 */ 3412 size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt; 3413 pmap_pvo_table = pmap_boot_find_memory(size, PAGE_SIZE, 0); 3414 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3415 if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH) 3416 panic("pmap_bootstrap: pmap_pvo_table end (%p + %lx) > 256MB", 3417 pmap_pvo_table, size); 3418 #endif 3419 3420 for (i = 0; i < pmap_pteg_cnt; i++) 3421 TAILQ_INIT(&pmap_pvo_table[i]); 3422 3423 #ifndef MSGBUFADDR 3424 /* 3425 * Allocate msgbuf in high memory. 3426 */ 3427 msgbuf_paddr = 3428 (paddr_t) pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1); 3429 #endif 3430 3431 #ifdef __HAVE_PMAP_PHYSSEG 3432 { 3433 u_int npgs = 0; 3434 for (i = 0, mp = avail; i < avail_cnt; i++, mp++) 3435 npgs += btoc(mp->size); 3436 size = (sizeof(struct pvo_head) + 1) * npgs; 3437 pmap_physseg.pvoh = pmap_boot_find_memory(size, PAGE_SIZE, 0); 3438 pmap_physseg.attrs = (char *) &pmap_physseg.pvoh[npgs]; 3439 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3440 if ((uintptr_t)pmap_physseg.pvoh + size > SEGMENT_LENGTH) 3441 panic("pmap_bootstrap: PVO list end (%p + %lx) > 256MB", 3442 pmap_physseg.pvoh, size); 3443 #endif 3444 } 3445 #endif 3446 3447 3448 for (mp = avail, i = 0; i < avail_cnt; mp++, i++) { 3449 paddr_t pfstart = atop(mp->start); 3450 paddr_t pfend = atop(mp->start + mp->size); 3451 if (mp->size == 0) 3452 continue; 3453 if (mp->start + mp->size <= SEGMENT_LENGTH) { 3454 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3455 VM_FREELIST_FIRST256); 3456 } else if (mp->start >= SEGMENT_LENGTH) { 3457 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3458 VM_FREELIST_DEFAULT); 3459 } else { 3460 pfend = atop(SEGMENT_LENGTH); 3461 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3462 VM_FREELIST_FIRST256); 3463 pfstart = atop(SEGMENT_LENGTH); 3464 pfend = atop(mp->start + mp->size); 3465 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3466 VM_FREELIST_DEFAULT); 3467 } 3468 } 3469 3470 /* 3471 * Make sure kernel vsid is allocated as well as VSID 0. 3472 */ 3473 pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW] 3474 |= 1 << (KERNEL_VSIDBITS % VSID_NBPW); 3475 pmap_vsid_bitmap[0] |= 1; 3476 3477 /* 3478 * Initialize kernel pmap and hardware. 3479 */ 3480 3481 /* PPC_OEA64_BRIDGE does support these instructions */ 3482 #if defined (PPC_OEA) || defined (PPC_OEA64_BRIDGE) 3483 for (i = 0; i < 16; i++) { 3484 pmap_kernel()->pm_sr[i] = KERNELN_SEGMENT(i)|SR_PRKEY; 3485 __asm volatile ("mtsrin %0,%1" 3486 :: "r"(KERNELN_SEGMENT(i)|SR_PRKEY), "r"(i << ADDR_SR_SHFT)); 3487 } 3488 3489 pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY; 3490 __asm volatile ("mtsr %0,%1" 3491 :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT)); 3492 #ifdef KERNEL2_SR 3493 pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY; 3494 __asm volatile ("mtsr %0,%1" 3495 :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT)); 3496 #endif 3497 for (i = 0; i < 16; i++) { 3498 if (iosrtable[i] & SR601_T) { 3499 pmap_kernel()->pm_sr[i] = iosrtable[i]; 3500 __asm volatile ("mtsrin %0,%1" 3501 :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT)); 3502 } 3503 } 3504 #endif /* PPC_OEA || PPC_OEA64_BRIDGE */ 3505 #if defined (PPC_OEA) 3506 __asm volatile ("sync; mtsdr1 %0; isync" 3507 :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10))); 3508 #elif defined (PPC_OEA64) || defined (PPC_OEA64_BRIDGE) 3509 __asm __volatile ("sync; mtsdr1 %0; isync" 3510 :: "r"((uintptr_t)pmap_pteg_table | (32 - cntlzw(pmap_pteg_mask >> 11)))); 3511 #endif 3512 tlbia(); 3513 3514 #ifdef ALTIVEC 3515 pmap_use_altivec = cpu_altivec; 3516 #endif 3517 3518 #ifdef DEBUG 3519 if (pmapdebug & PMAPDEBUG_BOOT) { 3520 u_int cnt; 3521 int bank; 3522 char pbuf[9]; 3523 for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) { 3524 cnt += vm_physmem[bank].avail_end - vm_physmem[bank].avail_start; 3525 printf("pmap_bootstrap: vm_physmem[%d]=%#lx-%#lx/%#lx\n", 3526 bank, 3527 ptoa(vm_physmem[bank].avail_start), 3528 ptoa(vm_physmem[bank].avail_end), 3529 ptoa(vm_physmem[bank].avail_end - vm_physmem[bank].avail_start)); 3530 } 3531 format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt)); 3532 printf("pmap_bootstrap: UVM memory = %s (%u pages)\n", 3533 pbuf, cnt); 3534 } 3535 #endif 3536 3537 pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry), 3538 sizeof(struct pvo_entry), 0, 0, "pmap_upvopl", 3539 &pmap_pool_uallocator, IPL_NONE); 3540 3541 pool_setlowat(&pmap_upvo_pool, 252); 3542 3543 pool_init(&pmap_pool, sizeof(struct pmap), 3544 sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator, 3545 IPL_NONE); 3546 3547 #if defined(PMAP_NEED_MAPKERNEL) 3548 { 3549 extern int etext[], kernel_text[]; 3550 vaddr_t va, va_etext = (paddr_t) etext; 3551 paddr_t pa; 3552 register_t sr; 3553 3554 sr = KERNELN_SEGMENT(kernelstart >> ADDR_SR_SHFT) 3555 |SR_SUKEY|SR_PRKEY; 3556 3557 va = (vaddr_t) kernel_text; 3558 3559 for (pa = kernelstart; va < va_etext; 3560 pa += PAGE_SIZE, va += PAGE_SIZE) 3561 pmap_enter(pmap_kernel(), va, pa, 3562 VM_PROT_READ|VM_PROT_EXECUTE, 0); 3563 3564 for (; pa < kernelend; 3565 pa += PAGE_SIZE, va += PAGE_SIZE) 3566 pmap_enter(pmap_kernel(), va, pa, 3567 VM_PROT_READ|VM_PROT_WRITE, 0); 3568 3569 pmap_kernel()->pm_sr[kernelstart >> ADDR_SR_SHFT] = sr; 3570 __asm volatile ("mtsrin %0,%1" 3571 :: "r"(sr), "r"(kernelstart)); 3572 } 3573 #endif 3574 } 3575
Indexes created Mon Oct 27 20:09:55 GMT 2025