1 /* $NetBSD: pmap.c,v 1.32 2005/06/02 14:35:08 he Exp $ */ 2 /*- 3 * Copyright (c) 2001 The NetBSD Foundation, Inc. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to The NetBSD Foundation 7 * by Matt Thomas <matt@3am-software.com> of Allegro Networks, Inc. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the NetBSD 20 * Foundation, Inc. and its contributors. 21 * 4. Neither the name of The NetBSD Foundation nor the names of its 22 * contributors may be used to endorse or promote products derived 23 * from this software without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 26 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 27 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 28 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 29 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 30 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 31 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 32 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 33 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 34 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 35 * POSSIBILITY OF SUCH DAMAGE. 36 */ 37 38 /* 39 * Copyright (C) 1995, 1996 Wolfgang Solfrank. 40 * Copyright (C) 1995, 1996 TooLs GmbH. 41 * All rights reserved. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. All advertising materials mentioning features or use of this software 52 * must display the following acknowledgement: 53 * This product includes software developed by TooLs GmbH. 54 * 4. The name of TooLs GmbH may not be used to endorse or promote products 55 * derived from this software without specific prior written permission. 56 * 57 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR 58 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 59 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 60 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 61 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 62 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 63 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 64 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 65 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 66 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.32 2005/06/02 14:35:08 he Exp $"); 71 72 #include "opt_ppcarch.h" 73 #include "opt_altivec.h" 74 #include "opt_pmap.h" 75 #include <sys/param.h> 76 #include <sys/malloc.h> 77 #include <sys/proc.h> 78 #include <sys/user.h> 79 #include <sys/pool.h> 80 #include <sys/queue.h> 81 #include <sys/device.h> /* for evcnt */ 82 #include <sys/systm.h> 83 84 #if __NetBSD_Version__ < 105010000 85 #include <vm/vm.h> 86 #include <vm/vm_kern.h> 87 #define splvm() splimp() 88 #endif 89 90 #include <uvm/uvm.h> 91 92 #include <machine/pcb.h> 93 #include <machine/powerpc.h> 94 #include <powerpc/spr.h> 95 #include <powerpc/oea/sr_601.h> 96 #include <powerpc/bat.h> 97 98 #if defined(DEBUG) || defined(PMAPCHECK) 99 #define STATIC 100 #else 101 #define STATIC static 102 #endif 103 104 #ifdef ALTIVEC 105 int pmap_use_altivec; 106 #endif 107 108 volatile struct pteg *pmap_pteg_table; 109 unsigned int pmap_pteg_cnt; 110 unsigned int pmap_pteg_mask; 111 #ifdef PMAP_MEMLIMIT 112 paddr_t pmap_memlimit = PMAP_MEMLIMIT; 113 #else 114 paddr_t pmap_memlimit = -PAGE_SIZE; /* there is no limit */ 115 #endif 116 117 struct pmap kernel_pmap_; 118 unsigned int pmap_pages_stolen; 119 u_long pmap_pte_valid; 120 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 121 u_long pmap_pvo_enter_depth; 122 u_long pmap_pvo_remove_depth; 123 #endif 124 125 int physmem; 126 #ifndef MSGBUFADDR 127 extern paddr_t msgbuf_paddr; 128 #endif 129 130 static struct mem_region *mem, *avail; 131 static u_int mem_cnt, avail_cnt; 132 133 #ifdef __HAVE_PMAP_PHYSSEG 134 /* 135 * This is a cache of referenced/modified bits. 136 * Bits herein are shifted by ATTRSHFT. 137 */ 138 #define ATTR_SHFT 4 139 struct pmap_physseg pmap_physseg; 140 #endif 141 142 /* 143 * The following structure is exactly 32 bytes long (one cacheline). 144 */ 145 struct pvo_entry { 146 LIST_ENTRY(pvo_entry) pvo_vlink; /* Link to common virt page */ 147 TAILQ_ENTRY(pvo_entry) pvo_olink; /* Link to overflow entry */ 148 struct pte pvo_pte; /* Prebuilt PTE */ 149 pmap_t pvo_pmap; /* ptr to owning pmap */ 150 vaddr_t pvo_vaddr; /* VA of entry */ 151 #define PVO_PTEGIDX_MASK 0x0007 /* which PTEG slot */ 152 #define PVO_PTEGIDX_VALID 0x0008 /* slot is valid */ 153 #define PVO_WIRED 0x0010 /* PVO entry is wired */ 154 #define PVO_MANAGED 0x0020 /* PVO e. for managed page */ 155 #define PVO_EXECUTABLE 0x0040 /* PVO e. for executable page */ 156 #define PVO_ENTER_INSERT 0 /* PVO has been removed */ 157 #define PVO_SPILL_UNSET 1 /* PVO has been evicted */ 158 #define PVO_SPILL_SET 2 /* PVO has been spilled */ 159 #define PVO_SPILL_INSERT 3 /* PVO has been inserted */ 160 #define PVO_PMAP_PAGE_PROTECT 4 /* PVO has changed */ 161 #define PVO_PMAP_PROTECT 5 /* PVO has changed */ 162 #define PVO_REMOVE 6 /* PVO has been removed */ 163 #define PVO_WHERE_MASK 15 164 #define PVO_WHERE_SHFT 8 165 }; 166 #define PVO_VADDR(pvo) ((pvo)->pvo_vaddr & ~ADDR_POFF) 167 #define PVO_ISEXECUTABLE(pvo) ((pvo)->pvo_vaddr & PVO_EXECUTABLE) 168 #define PVO_PTEGIDX_GET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK) 169 #define PVO_PTEGIDX_ISSET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID) 170 #define PVO_PTEGIDX_CLR(pvo) \ 171 ((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK))) 172 #define PVO_PTEGIDX_SET(pvo,i) \ 173 ((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID)) 174 #define PVO_WHERE(pvo,w) \ 175 ((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \ 176 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT)) 177 178 TAILQ_HEAD(pvo_tqhead, pvo_entry); 179 struct pvo_tqhead *pmap_pvo_table; /* pvo entries by ptegroup index */ 180 struct pvo_head pmap_pvo_kunmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged); /* list of unmanaged pages */ 181 struct pvo_head pmap_pvo_unmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged); /* list of unmanaged pages */ 182 183 struct pool pmap_pool; /* pool for pmap structures */ 184 struct pool pmap_upvo_pool; /* pool for pvo entries for unmanaged pages */ 185 struct pool pmap_mpvo_pool; /* pool for pvo entries for managed pages */ 186 187 /* 188 * We keep a cache of unmanaged pages to be used for pvo entries for 189 * unmanaged pages. 190 */ 191 struct pvo_page { 192 SIMPLEQ_ENTRY(pvo_page) pvop_link; 193 }; 194 SIMPLEQ_HEAD(pvop_head, pvo_page); 195 struct pvop_head pmap_upvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_upvop_head); 196 struct pvop_head pmap_mpvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_mpvop_head); 197 u_long pmap_upvop_free; 198 u_long pmap_upvop_maxfree; 199 u_long pmap_mpvop_free; 200 u_long pmap_mpvop_maxfree; 201 202 STATIC void *pmap_pool_ualloc(struct pool *, int); 203 STATIC void *pmap_pool_malloc(struct pool *, int); 204 205 STATIC void pmap_pool_ufree(struct pool *, void *); 206 STATIC void pmap_pool_mfree(struct pool *, void *); 207 208 static struct pool_allocator pmap_pool_mallocator = { 209 pmap_pool_malloc, pmap_pool_mfree, 0, 210 }; 211 212 static struct pool_allocator pmap_pool_uallocator = { 213 pmap_pool_ualloc, pmap_pool_ufree, 0, 214 }; 215 216 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 217 void pmap_pte_print(volatile struct pte *); 218 #endif 219 220 #ifdef DDB 221 void pmap_pteg_check(void); 222 void pmap_pteg_dist(void); 223 void pmap_print_pte(pmap_t, vaddr_t); 224 void pmap_print_mmuregs(void); 225 #endif 226 227 #if defined(DEBUG) || defined(PMAPCHECK) 228 #ifdef PMAPCHECK 229 int pmapcheck = 1; 230 #else 231 int pmapcheck = 0; 232 #endif 233 void pmap_pvo_verify(void); 234 STATIC void pmap_pvo_check(const struct pvo_entry *); 235 #define PMAP_PVO_CHECK(pvo) \ 236 do { \ 237 if (pmapcheck) \ 238 pmap_pvo_check(pvo); \ 239 } while (0) 240 #else 241 #define PMAP_PVO_CHECK(pvo) do { } while (/*CONSTCOND*/0) 242 #endif 243 STATIC int pmap_pte_insert(int, struct pte *); 244 STATIC int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *, 245 vaddr_t, paddr_t, register_t, int); 246 STATIC void pmap_pvo_remove(struct pvo_entry *, int, boolean_t); 247 STATIC struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *); 248 STATIC volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int); 249 STATIC struct pvo_entry *pmap_pvo_reclaim(struct pmap *); 250 STATIC void pvo_set_exec(struct pvo_entry *); 251 STATIC void pvo_clear_exec(struct pvo_entry *); 252 253 STATIC void tlbia(void); 254 255 STATIC void pmap_release(pmap_t); 256 STATIC void *pmap_boot_find_memory(psize_t, psize_t, int); 257 258 static uint32_t pmap_pvo_reclaim_nextidx; 259 #ifdef DEBUG 260 static int pmap_pvo_reclaim_debugctr; 261 #endif 262 263 #define VSID_NBPW (sizeof(uint32_t) * 8) 264 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW]; 265 266 static int pmap_initialized; 267 268 #if defined(DEBUG) || defined(PMAPDEBUG) 269 #define PMAPDEBUG_BOOT 0x0001 270 #define PMAPDEBUG_PTE 0x0002 271 #define PMAPDEBUG_EXEC 0x0008 272 #define PMAPDEBUG_PVOENTER 0x0010 273 #define PMAPDEBUG_PVOREMOVE 0x0020 274 #define PMAPDEBUG_ACTIVATE 0x0100 275 #define PMAPDEBUG_CREATE 0x0200 276 #define PMAPDEBUG_ENTER 0x1000 277 #define PMAPDEBUG_KENTER 0x2000 278 #define PMAPDEBUG_KREMOVE 0x4000 279 #define PMAPDEBUG_REMOVE 0x8000 280 unsigned int pmapdebug = 0; 281 # define DPRINTF(x) printf x 282 # define DPRINTFN(n, x) if (pmapdebug & PMAPDEBUG_ ## n) printf x 283 #else 284 # define DPRINTF(x) 285 # define DPRINTFN(n, x) 286 #endif 287 288 289 #ifdef PMAPCOUNTERS 290 #define PMAPCOUNT(ev) ((pmap_evcnt_ ## ev).ev_count++) 291 #define PMAPCOUNT2(ev) ((ev).ev_count++) 292 293 struct evcnt pmap_evcnt_mappings = 294 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 295 "pmap", "pages mapped"); 296 struct evcnt pmap_evcnt_unmappings = 297 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 298 "pmap", "pages unmapped"); 299 300 struct evcnt pmap_evcnt_kernel_mappings = 301 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 302 "pmap", "kernel pages mapped"); 303 struct evcnt pmap_evcnt_kernel_unmappings = 304 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_kernel_mappings, 305 "pmap", "kernel pages unmapped"); 306 307 struct evcnt pmap_evcnt_mappings_replaced = 308 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 309 "pmap", "page mappings replaced"); 310 311 struct evcnt pmap_evcnt_exec_mappings = 312 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 313 "pmap", "exec pages mapped"); 314 struct evcnt pmap_evcnt_exec_cached = 315 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 316 "pmap", "exec pages cached"); 317 318 struct evcnt pmap_evcnt_exec_synced = 319 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 320 "pmap", "exec pages synced"); 321 struct evcnt pmap_evcnt_exec_synced_clear_modify = 322 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 323 "pmap", "exec pages synced (CM)"); 324 325 struct evcnt pmap_evcnt_exec_uncached_page_protect = 326 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 327 "pmap", "exec pages uncached (PP)"); 328 struct evcnt pmap_evcnt_exec_uncached_clear_modify = 329 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 330 "pmap", "exec pages uncached (CM)"); 331 struct evcnt pmap_evcnt_exec_uncached_zero_page = 332 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 333 "pmap", "exec pages uncached (ZP)"); 334 struct evcnt pmap_evcnt_exec_uncached_copy_page = 335 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 336 "pmap", "exec pages uncached (CP)"); 337 338 struct evcnt pmap_evcnt_updates = 339 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 340 "pmap", "updates"); 341 struct evcnt pmap_evcnt_collects = 342 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 343 "pmap", "collects"); 344 struct evcnt pmap_evcnt_copies = 345 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 346 "pmap", "copies"); 347 348 struct evcnt pmap_evcnt_ptes_spilled = 349 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 350 "pmap", "ptes spilled from overflow"); 351 struct evcnt pmap_evcnt_ptes_unspilled = 352 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 353 "pmap", "ptes not spilled"); 354 struct evcnt pmap_evcnt_ptes_evicted = 355 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 356 "pmap", "ptes evicted"); 357 358 struct evcnt pmap_evcnt_ptes_primary[8] = { 359 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 360 "pmap", "ptes added at primary[0]"), 361 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 362 "pmap", "ptes added at primary[1]"), 363 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 364 "pmap", "ptes added at primary[2]"), 365 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 366 "pmap", "ptes added at primary[3]"), 367 368 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 369 "pmap", "ptes added at primary[4]"), 370 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 371 "pmap", "ptes added at primary[5]"), 372 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 373 "pmap", "ptes added at primary[6]"), 374 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 375 "pmap", "ptes added at primary[7]"), 376 }; 377 struct evcnt pmap_evcnt_ptes_secondary[8] = { 378 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 379 "pmap", "ptes added at secondary[0]"), 380 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 381 "pmap", "ptes added at secondary[1]"), 382 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 383 "pmap", "ptes added at secondary[2]"), 384 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 385 "pmap", "ptes added at secondary[3]"), 386 387 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 388 "pmap", "ptes added at secondary[4]"), 389 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 390 "pmap", "ptes added at secondary[5]"), 391 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 392 "pmap", "ptes added at secondary[6]"), 393 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 394 "pmap", "ptes added at secondary[7]"), 395 }; 396 struct evcnt pmap_evcnt_ptes_removed = 397 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 398 "pmap", "ptes removed"); 399 struct evcnt pmap_evcnt_ptes_changed = 400 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 401 "pmap", "ptes changed"); 402 struct evcnt pmap_evcnt_pvos_reclaimed = 403 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 404 "pmap", "pvos reclaimed"); 405 struct evcnt pmap_evcnt_pvos_failed = 406 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 407 "pmap", "pvo allocation failures"); 408 409 /* 410 * From pmap_subr.c 411 */ 412 extern struct evcnt pmap_evcnt_zeroed_pages; 413 extern struct evcnt pmap_evcnt_copied_pages; 414 extern struct evcnt pmap_evcnt_idlezeroed_pages; 415 416 EVCNT_ATTACH_STATIC(pmap_evcnt_mappings); 417 EVCNT_ATTACH_STATIC(pmap_evcnt_mappings_replaced); 418 EVCNT_ATTACH_STATIC(pmap_evcnt_unmappings); 419 420 EVCNT_ATTACH_STATIC(pmap_evcnt_kernel_mappings); 421 EVCNT_ATTACH_STATIC(pmap_evcnt_kernel_unmappings); 422 423 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_mappings); 424 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_cached); 425 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_synced); 426 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_synced_clear_modify); 427 428 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_page_protect); 429 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_clear_modify); 430 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_zero_page); 431 EVCNT_ATTACH_STATIC(pmap_evcnt_exec_uncached_copy_page); 432 433 EVCNT_ATTACH_STATIC(pmap_evcnt_zeroed_pages); 434 EVCNT_ATTACH_STATIC(pmap_evcnt_copied_pages); 435 EVCNT_ATTACH_STATIC(pmap_evcnt_idlezeroed_pages); 436 437 EVCNT_ATTACH_STATIC(pmap_evcnt_updates); 438 EVCNT_ATTACH_STATIC(pmap_evcnt_collects); 439 EVCNT_ATTACH_STATIC(pmap_evcnt_copies); 440 441 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_spilled); 442 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_unspilled); 443 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_evicted); 444 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_removed); 445 EVCNT_ATTACH_STATIC(pmap_evcnt_ptes_changed); 446 447 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 0); 448 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 1); 449 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 2); 450 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 3); 451 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 4); 452 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 5); 453 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 6); 454 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_primary, 7); 455 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 0); 456 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 1); 457 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 2); 458 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 3); 459 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 4); 460 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 5); 461 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 6); 462 EVCNT_ATTACH_STATIC2(pmap_evcnt_ptes_secondary, 7); 463 464 EVCNT_ATTACH_STATIC(pmap_evcnt_pvos_reclaimed); 465 EVCNT_ATTACH_STATIC(pmap_evcnt_pvos_failed); 466 #else 467 #define PMAPCOUNT(ev) ((void) 0) 468 #define PMAPCOUNT2(ev) ((void) 0) 469 #endif 470 471 #define TLBIE(va) __asm __volatile("tlbie %0" :: "r"(va)) 472 #define TLBSYNC() __asm __volatile("tlbsync") 473 #define SYNC() __asm __volatile("sync") 474 #define EIEIO() __asm __volatile("eieio") 475 #define MFMSR() mfmsr() 476 #define MTMSR(psl) mtmsr(psl) 477 #define MFPVR() mfpvr() 478 #define MFSRIN(va) mfsrin(va) 479 #define MFTB() mfrtcltbl() 480 481 #ifndef PPC_OEA64 482 static __inline register_t 483 mfsrin(vaddr_t va) 484 { 485 register_t sr; 486 __asm __volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va)); 487 return sr; 488 } 489 #endif /* PPC_OEA64 */ 490 491 static __inline register_t 492 pmap_interrupts_off(void) 493 { 494 register_t msr = MFMSR(); 495 if (msr & PSL_EE) 496 MTMSR(msr & ~PSL_EE); 497 return msr; 498 } 499 500 static void 501 pmap_interrupts_restore(register_t msr) 502 { 503 if (msr & PSL_EE) 504 MTMSR(msr); 505 } 506 507 static __inline u_int32_t 508 mfrtcltbl(void) 509 { 510 511 if ((MFPVR() >> 16) == MPC601) 512 return (mfrtcl() >> 7); 513 else 514 return (mftbl()); 515 } 516 517 /* 518 * These small routines may have to be replaced, 519 * if/when we support processors other that the 604. 520 */ 521 522 void 523 tlbia(void) 524 { 525 caddr_t i; 526 527 SYNC(); 528 /* 529 * Why not use "tlbia"? Because not all processors implement it. 530 * 531 * This needs to be a per-CPU callback to do the appropriate thing 532 * for the CPU. XXX 533 */ 534 for (i = 0; i < (caddr_t)0x00040000; i += 0x00001000) { 535 TLBIE(i); 536 EIEIO(); 537 SYNC(); 538 } 539 TLBSYNC(); 540 SYNC(); 541 } 542 543 static __inline register_t 544 va_to_vsid(const struct pmap *pm, vaddr_t addr) 545 { 546 #ifdef PPC_OEA64 547 #if 0 548 const struct ste *ste; 549 register_t hash; 550 int i; 551 552 hash = (addr >> ADDR_ESID_SHFT) & ADDR_ESID_HASH; 553 554 /* 555 * Try the primary group first 556 */ 557 ste = pm->pm_stes[hash].stes; 558 for (i = 0; i < 8; i++, ste++) { 559 if (ste->ste_hi & STE_V) && 560 (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID)) 561 return ste; 562 } 563 564 /* 565 * Then the secondary group. 566 */ 567 ste = pm->pm_stes[hash ^ ADDR_ESID_HASH].stes; 568 for (i = 0; i < 8; i++, ste++) { 569 if (ste->ste_hi & STE_V) && 570 (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID)) 571 return addr; 572 } 573 574 return NULL; 575 #else 576 /* 577 * Rather than searching the STE groups for the VSID, we know 578 * how we generate that from the ESID and so do that. 579 */ 580 return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT; 581 #endif 582 #else 583 return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID) >> SR_VSID_SHFT; 584 #endif 585 } 586 587 static __inline register_t 588 va_to_pteg(const struct pmap *pm, vaddr_t addr) 589 { 590 register_t hash; 591 592 hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT); 593 return hash & pmap_pteg_mask; 594 } 595 596 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 597 /* 598 * Given a PTE in the page table, calculate the VADDR that hashes to it. 599 * The only bit of magic is that the top 4 bits of the address doesn't 600 * technically exist in the PTE. But we know we reserved 4 bits of the 601 * VSID for it so that's how we get it. 602 */ 603 static vaddr_t 604 pmap_pte_to_va(volatile const struct pte *pt) 605 { 606 vaddr_t va; 607 uintptr_t ptaddr = (uintptr_t) pt; 608 609 if (pt->pte_hi & PTE_HID) 610 ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg)); 611 612 /* PPC Bits 10-19 PPC64 Bits 42-51 */ 613 va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff; 614 va <<= ADDR_PIDX_SHFT; 615 616 /* PPC Bits 4-9 PPC64 Bits 36-41 */ 617 va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT; 618 619 #ifdef PPC_OEA64 620 /* PPC63 Bits 0-35 */ 621 /* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */ 622 #endif 623 #ifdef PPC_OEA 624 /* PPC Bits 0-3 */ 625 va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; 626 #endif 627 628 return va; 629 } 630 #endif 631 632 static __inline struct pvo_head * 633 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p) 634 { 635 #ifdef __HAVE_VM_PAGE_MD 636 struct vm_page *pg; 637 638 pg = PHYS_TO_VM_PAGE(pa); 639 if (pg_p != NULL) 640 *pg_p = pg; 641 if (pg == NULL) 642 return &pmap_pvo_unmanaged; 643 return &pg->mdpage.mdpg_pvoh; 644 #endif 645 #ifdef __HAVE_PMAP_PHYSSEG 646 int bank, pg; 647 648 bank = vm_physseg_find(atop(pa), &pg); 649 if (pg_p != NULL) 650 *pg_p = pg; 651 if (bank == -1) 652 return &pmap_pvo_unmanaged; 653 return &vm_physmem[bank].pmseg.pvoh[pg]; 654 #endif 655 } 656 657 static __inline struct pvo_head * 658 vm_page_to_pvoh(struct vm_page *pg) 659 { 660 #ifdef __HAVE_VM_PAGE_MD 661 return &pg->mdpage.mdpg_pvoh; 662 #endif 663 #ifdef __HAVE_PMAP_PHYSSEG 664 return pa_to_pvoh(VM_PAGE_TO_PHYS(pg), NULL); 665 #endif 666 } 667 668 669 #ifdef __HAVE_PMAP_PHYSSEG 670 static __inline char * 671 pa_to_attr(paddr_t pa) 672 { 673 int bank, pg; 674 675 bank = vm_physseg_find(atop(pa), &pg); 676 if (bank == -1) 677 return NULL; 678 return &vm_physmem[bank].pmseg.attrs[pg]; 679 } 680 #endif 681 682 static __inline void 683 pmap_attr_clear(struct vm_page *pg, int ptebit) 684 { 685 #ifdef __HAVE_PMAP_PHYSSEG 686 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) &= ~(ptebit >> ATTR_SHFT); 687 #endif 688 #ifdef __HAVE_VM_PAGE_MD 689 pg->mdpage.mdpg_attrs &= ~ptebit; 690 #endif 691 } 692 693 static __inline int 694 pmap_attr_fetch(struct vm_page *pg) 695 { 696 #ifdef __HAVE_PMAP_PHYSSEG 697 return *pa_to_attr(VM_PAGE_TO_PHYS(pg)) << ATTR_SHFT; 698 #endif 699 #ifdef __HAVE_VM_PAGE_MD 700 return pg->mdpage.mdpg_attrs; 701 #endif 702 } 703 704 static __inline void 705 pmap_attr_save(struct vm_page *pg, int ptebit) 706 { 707 #ifdef __HAVE_PMAP_PHYSSEG 708 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) |= (ptebit >> ATTR_SHFT); 709 #endif 710 #ifdef __HAVE_VM_PAGE_MD 711 pg->mdpage.mdpg_attrs |= ptebit; 712 #endif 713 } 714 715 static __inline int 716 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt) 717 { 718 if (pt->pte_hi == pvo_pt->pte_hi 719 #if 0 720 && ((pt->pte_lo ^ pvo_pt->pte_lo) & 721 ~(PTE_REF|PTE_CHG)) == 0 722 #endif 723 ) 724 return 1; 725 return 0; 726 } 727 728 static __inline void 729 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo) 730 { 731 /* 732 * Construct the PTE. Default to IMB initially. Valid bit 733 * only gets set when the real pte is set in memory. 734 * 735 * Note: Don't set the valid bit for correct operation of tlb update. 736 */ 737 pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT) 738 | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API); 739 pt->pte_lo = pte_lo; 740 } 741 742 static __inline void 743 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt) 744 { 745 pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG); 746 } 747 748 static __inline void 749 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit) 750 { 751 /* 752 * As shown in Section 7.6.3.2.3 753 */ 754 pt->pte_lo &= ~ptebit; 755 TLBIE(va); 756 SYNC(); 757 EIEIO(); 758 TLBSYNC(); 759 SYNC(); 760 } 761 762 static __inline void 763 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt) 764 { 765 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 766 if (pvo_pt->pte_hi & PTE_VALID) 767 panic("pte_set: setting an already valid pte %p", pvo_pt); 768 #endif 769 pvo_pt->pte_hi |= PTE_VALID; 770 /* 771 * Update the PTE as defined in section 7.6.3.1 772 * Note that the REF/CHG bits are from pvo_pt and thus should 773 * have been saved so this routine can restore them (if desired). 774 */ 775 pt->pte_lo = pvo_pt->pte_lo; 776 EIEIO(); 777 pt->pte_hi = pvo_pt->pte_hi; 778 SYNC(); 779 pmap_pte_valid++; 780 } 781 782 static __inline void 783 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 784 { 785 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 786 if ((pvo_pt->pte_hi & PTE_VALID) == 0) 787 panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt); 788 if ((pt->pte_hi & PTE_VALID) == 0) 789 panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt); 790 #endif 791 792 pvo_pt->pte_hi &= ~PTE_VALID; 793 /* 794 * Force the ref & chg bits back into the PTEs. 795 */ 796 SYNC(); 797 /* 798 * Invalidate the pte ... (Section 7.6.3.3) 799 */ 800 pt->pte_hi &= ~PTE_VALID; 801 SYNC(); 802 TLBIE(va); 803 SYNC(); 804 EIEIO(); 805 TLBSYNC(); 806 SYNC(); 807 /* 808 * Save the ref & chg bits ... 809 */ 810 pmap_pte_synch(pt, pvo_pt); 811 pmap_pte_valid--; 812 } 813 814 static __inline void 815 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 816 { 817 /* 818 * Invalidate the PTE 819 */ 820 pmap_pte_unset(pt, pvo_pt, va); 821 pmap_pte_set(pt, pvo_pt); 822 } 823 824 /* 825 * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx 826 * (either primary or secondary location). 827 * 828 * Note: both the destination and source PTEs must not have PTE_VALID set. 829 */ 830 831 STATIC int 832 pmap_pte_insert(int ptegidx, struct pte *pvo_pt) 833 { 834 volatile struct pte *pt; 835 int i; 836 837 #if defined(DEBUG) 838 DPRINTFN(PTE, ("pmap_pte_insert: idx 0x%x, pte 0x%x 0x%x\n", 839 ptegidx, (unsigned int) pvo_pt->pte_hi, (unsigned int) pvo_pt->pte_lo)); 840 #endif 841 /* 842 * First try primary hash. 843 */ 844 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 845 if ((pt->pte_hi & PTE_VALID) == 0) { 846 pvo_pt->pte_hi &= ~PTE_HID; 847 pmap_pte_set(pt, pvo_pt); 848 return i; 849 } 850 } 851 852 /* 853 * Now try secondary hash. 854 */ 855 ptegidx ^= pmap_pteg_mask; 856 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 857 if ((pt->pte_hi & PTE_VALID) == 0) { 858 pvo_pt->pte_hi |= PTE_HID; 859 pmap_pte_set(pt, pvo_pt); 860 return i; 861 } 862 } 863 return -1; 864 } 865 866 /* 867 * Spill handler. 868 * 869 * Tries to spill a page table entry from the overflow area. 870 * This runs in either real mode (if dealing with a exception spill) 871 * or virtual mode when dealing with manually spilling one of the 872 * kernel's pte entries. In either case, interrupts are already 873 * disabled. 874 */ 875 876 int 877 pmap_pte_spill(struct pmap *pm, vaddr_t addr, boolean_t exec) 878 { 879 struct pvo_entry *source_pvo, *victim_pvo, *next_pvo; 880 struct pvo_entry *pvo; 881 /* XXX: gcc -- vpvoh is always set at either *1* or *2* */ 882 struct pvo_tqhead *pvoh, *vpvoh = NULL; 883 int ptegidx, i, j; 884 volatile struct pteg *pteg; 885 volatile struct pte *pt; 886 887 ptegidx = va_to_pteg(pm, addr); 888 889 /* 890 * Have to substitute some entry. Use the primary hash for this. 891 * Use low bits of timebase as random generator. Make sure we are 892 * not picking a kernel pte for replacement. 893 */ 894 pteg = &pmap_pteg_table[ptegidx]; 895 i = MFTB() & 7; 896 for (j = 0; j < 8; j++) { 897 pt = &pteg->pt[i]; 898 if ((pt->pte_hi & PTE_VALID) == 0 || 899 VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT) 900 != KERNEL_VSIDBITS) 901 break; 902 i = (i + 1) & 7; 903 } 904 KASSERT(j < 8); 905 906 source_pvo = NULL; 907 victim_pvo = NULL; 908 pvoh = &pmap_pvo_table[ptegidx]; 909 TAILQ_FOREACH(pvo, pvoh, pvo_olink) { 910 911 /* 912 * We need to find pvo entry for this address... 913 */ 914 PMAP_PVO_CHECK(pvo); /* sanity check */ 915 916 /* 917 * If we haven't found the source and we come to a PVO with 918 * a valid PTE, then we know we can't find it because all 919 * evicted PVOs always are first in the list. 920 */ 921 if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID)) 922 break; 923 if (source_pvo == NULL && pm == pvo->pvo_pmap && 924 addr == PVO_VADDR(pvo)) { 925 926 /* 927 * Now we have found the entry to be spilled into the 928 * pteg. Attempt to insert it into the page table. 929 */ 930 j = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 931 if (j >= 0) { 932 PVO_PTEGIDX_SET(pvo, j); 933 PMAP_PVO_CHECK(pvo); /* sanity check */ 934 PVO_WHERE(pvo, SPILL_INSERT); 935 pvo->pvo_pmap->pm_evictions--; 936 PMAPCOUNT(ptes_spilled); 937 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 938 ? pmap_evcnt_ptes_secondary 939 : pmap_evcnt_ptes_primary)[j]); 940 941 /* 942 * Since we keep the evicted entries at the 943 * from of the PVO list, we need move this 944 * (now resident) PVO after the evicted 945 * entries. 946 */ 947 next_pvo = TAILQ_NEXT(pvo, pvo_olink); 948 949 /* 950 * If we don't have to move (either we were the 951 * last entry or the next entry was valid), 952 * don't change our position. Otherwise 953 * move ourselves to the tail of the queue. 954 */ 955 if (next_pvo != NULL && 956 !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) { 957 TAILQ_REMOVE(pvoh, pvo, pvo_olink); 958 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 959 } 960 return 1; 961 } 962 source_pvo = pvo; 963 if (exec && !PVO_ISEXECUTABLE(source_pvo)) { 964 return 0; 965 } 966 if (victim_pvo != NULL) 967 break; 968 } 969 970 /* 971 * We also need the pvo entry of the victim we are replacing 972 * so save the R & C bits of the PTE. 973 */ 974 if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL && 975 pmap_pte_compare(pt, &pvo->pvo_pte)) { 976 vpvoh = pvoh; /* *1* */ 977 victim_pvo = pvo; 978 if (source_pvo != NULL) 979 break; 980 } 981 } 982 983 if (source_pvo == NULL) { 984 PMAPCOUNT(ptes_unspilled); 985 return 0; 986 } 987 988 if (victim_pvo == NULL) { 989 if ((pt->pte_hi & PTE_HID) == 0) 990 panic("pmap_pte_spill: victim p-pte (%p) has " 991 "no pvo entry!", pt); 992 993 /* 994 * If this is a secondary PTE, we need to search 995 * its primary pvo bucket for the matching PVO. 996 */ 997 vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; /* *2* */ 998 TAILQ_FOREACH(pvo, vpvoh, pvo_olink) { 999 PMAP_PVO_CHECK(pvo); /* sanity check */ 1000 1001 /* 1002 * We also need the pvo entry of the victim we are 1003 * replacing so save the R & C bits of the PTE. 1004 */ 1005 if (pmap_pte_compare(pt, &pvo->pvo_pte)) { 1006 victim_pvo = pvo; 1007 break; 1008 } 1009 } 1010 if (victim_pvo == NULL) 1011 panic("pmap_pte_spill: victim s-pte (%p) has " 1012 "no pvo entry!", pt); 1013 } 1014 1015 /* 1016 * The victim should be not be a kernel PVO/PTE entry. 1017 */ 1018 KASSERT(victim_pvo->pvo_pmap != pmap_kernel()); 1019 KASSERT(PVO_PTEGIDX_ISSET(victim_pvo)); 1020 KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i); 1021 1022 /* 1023 * We are invalidating the TLB entry for the EA for the 1024 * we are replacing even though its valid; If we don't 1025 * we lose any ref/chg bit changes contained in the TLB 1026 * entry. 1027 */ 1028 source_pvo->pvo_pte.pte_hi &= ~PTE_HID; 1029 1030 /* 1031 * To enforce the PVO list ordering constraint that all 1032 * evicted entries should come before all valid entries, 1033 * move the source PVO to the tail of its list and the 1034 * victim PVO to the head of its list (which might not be 1035 * the same list, if the victim was using the secondary hash). 1036 */ 1037 TAILQ_REMOVE(pvoh, source_pvo, pvo_olink); 1038 TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink); 1039 TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink); 1040 TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink); 1041 pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr); 1042 pmap_pte_set(pt, &source_pvo->pvo_pte); 1043 victim_pvo->pvo_pmap->pm_evictions++; 1044 source_pvo->pvo_pmap->pm_evictions--; 1045 PVO_WHERE(victim_pvo, SPILL_UNSET); 1046 PVO_WHERE(source_pvo, SPILL_SET); 1047 1048 PVO_PTEGIDX_CLR(victim_pvo); 1049 PVO_PTEGIDX_SET(source_pvo, i); 1050 PMAPCOUNT2(pmap_evcnt_ptes_primary[i]); 1051 PMAPCOUNT(ptes_spilled); 1052 PMAPCOUNT(ptes_evicted); 1053 PMAPCOUNT(ptes_removed); 1054 1055 PMAP_PVO_CHECK(victim_pvo); 1056 PMAP_PVO_CHECK(source_pvo); 1057 return 1; 1058 } 1059 1060 /* 1061 * Restrict given range to physical memory 1062 */ 1063 void 1064 pmap_real_memory(paddr_t *start, psize_t *size) 1065 { 1066 struct mem_region *mp; 1067 1068 for (mp = mem; mp->size; mp++) { 1069 if (*start + *size > mp->start 1070 && *start < mp->start + mp->size) { 1071 if (*start < mp->start) { 1072 *size -= mp->start - *start; 1073 *start = mp->start; 1074 } 1075 if (*start + *size > mp->start + mp->size) 1076 *size = mp->start + mp->size - *start; 1077 return; 1078 } 1079 } 1080 *size = 0; 1081 } 1082 1083 /* 1084 * Initialize anything else for pmap handling. 1085 * Called during vm_init(). 1086 */ 1087 void 1088 pmap_init(void) 1089 { 1090 #ifdef __HAVE_PMAP_PHYSSEG 1091 struct pvo_tqhead *pvoh; 1092 int bank; 1093 long sz; 1094 char *attr; 1095 1096 pvoh = pmap_physseg.pvoh; 1097 attr = pmap_physseg.attrs; 1098 for (bank = 0; bank < vm_nphysseg; bank++) { 1099 sz = vm_physmem[bank].end - vm_physmem[bank].start; 1100 vm_physmem[bank].pmseg.pvoh = pvoh; 1101 vm_physmem[bank].pmseg.attrs = attr; 1102 for (; sz > 0; sz--, pvoh++, attr++) { 1103 TAILQ_INIT(pvoh); 1104 *attr = 0; 1105 } 1106 } 1107 #endif 1108 1109 pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry), 1110 sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl", 1111 &pmap_pool_mallocator); 1112 1113 pool_setlowat(&pmap_mpvo_pool, 1008); 1114 1115 pmap_initialized = 1; 1116 1117 } 1118 1119 /* 1120 * How much virtual space does the kernel get? 1121 */ 1122 void 1123 pmap_virtual_space(vaddr_t *start, vaddr_t *end) 1124 { 1125 /* 1126 * For now, reserve one segment (minus some overhead) for kernel 1127 * virtual memory 1128 */ 1129 *start = VM_MIN_KERNEL_ADDRESS; 1130 *end = VM_MAX_KERNEL_ADDRESS; 1131 } 1132 1133 /* 1134 * Allocate, initialize, and return a new physical map. 1135 */ 1136 pmap_t 1137 pmap_create(void) 1138 { 1139 pmap_t pm; 1140 1141 pm = pool_get(&pmap_pool, PR_WAITOK); 1142 memset((caddr_t)pm, 0, sizeof *pm); 1143 pmap_pinit(pm); 1144 1145 DPRINTFN(CREATE,("pmap_create: pm %p:\n" 1146 "\t%06x %06x %06x %06x %06x %06x %06x %06x\n" 1147 "\t%06x %06x %06x %06x %06x %06x %06x %06x\n", pm, 1148 (unsigned int) pm->pm_sr[0], (unsigned int) pm->pm_sr[1], 1149 (unsigned int) pm->pm_sr[2], (unsigned int) pm->pm_sr[3], 1150 (unsigned int) pm->pm_sr[4], (unsigned int) pm->pm_sr[5], 1151 (unsigned int) pm->pm_sr[6], (unsigned int) pm->pm_sr[7], 1152 (unsigned int) pm->pm_sr[8], (unsigned int) pm->pm_sr[9], 1153 (unsigned int) pm->pm_sr[10], (unsigned int) pm->pm_sr[11], 1154 (unsigned int) pm->pm_sr[12], (unsigned int) pm->pm_sr[13], 1155 (unsigned int) pm->pm_sr[14], (unsigned int) pm->pm_sr[15])); 1156 return pm; 1157 } 1158 1159 /* 1160 * Initialize a preallocated and zeroed pmap structure. 1161 */ 1162 void 1163 pmap_pinit(pmap_t pm) 1164 { 1165 register_t entropy = MFTB(); 1166 register_t mask; 1167 int i; 1168 1169 /* 1170 * Allocate some segment registers for this pmap. 1171 */ 1172 pm->pm_refs = 1; 1173 for (i = 0; i < NPMAPS; i += VSID_NBPW) { 1174 static register_t pmap_vsidcontext; 1175 register_t hash; 1176 unsigned int n; 1177 1178 /* Create a new value by multiplying by a prime adding in 1179 * entropy from the timebase register. This is to make the 1180 * VSID more random so that the PT Hash function collides 1181 * less often. (note that the prime causes gcc to do shifts 1182 * instead of a multiply) 1183 */ 1184 pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy; 1185 hash = pmap_vsidcontext & (NPMAPS - 1); 1186 if (hash == 0) { /* 0 is special, avoid it */ 1187 entropy += 0xbadf00d; 1188 continue; 1189 } 1190 n = hash >> 5; 1191 mask = 1L << (hash & (VSID_NBPW-1)); 1192 hash = pmap_vsidcontext; 1193 if (pmap_vsid_bitmap[n] & mask) { /* collision? */ 1194 /* anything free in this bucket? */ 1195 if (~pmap_vsid_bitmap[n] == 0) { 1196 entropy = hash ^ (hash >> 16); 1197 continue; 1198 } 1199 i = ffs(~pmap_vsid_bitmap[n]) - 1; 1200 mask = 1L << i; 1201 hash &= ~(VSID_NBPW-1); 1202 hash |= i; 1203 } 1204 hash &= PTE_VSID >> PTE_VSID_SHFT; 1205 pmap_vsid_bitmap[n] |= mask; 1206 pm->pm_vsid = hash; 1207 #ifndef PPC_OEA64 1208 for (i = 0; i < 16; i++) 1209 pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY | 1210 SR_NOEXEC; 1211 #endif 1212 return; 1213 } 1214 panic("pmap_pinit: out of segments"); 1215 } 1216 1217 /* 1218 * Add a reference to the given pmap. 1219 */ 1220 void 1221 pmap_reference(pmap_t pm) 1222 { 1223 pm->pm_refs++; 1224 } 1225 1226 /* 1227 * Retire the given pmap from service. 1228 * Should only be called if the map contains no valid mappings. 1229 */ 1230 void 1231 pmap_destroy(pmap_t pm) 1232 { 1233 if (--pm->pm_refs == 0) { 1234 pmap_release(pm); 1235 pool_put(&pmap_pool, pm); 1236 } 1237 } 1238 1239 /* 1240 * Release any resources held by the given physical map. 1241 * Called when a pmap initialized by pmap_pinit is being released. 1242 */ 1243 void 1244 pmap_release(pmap_t pm) 1245 { 1246 int idx, mask; 1247 1248 if (pm->pm_sr[0] == 0) 1249 panic("pmap_release"); 1250 idx = pm->pm_vsid & (NPMAPS-1); 1251 mask = 1 << (idx % VSID_NBPW); 1252 idx /= VSID_NBPW; 1253 1254 KASSERT(pmap_vsid_bitmap[idx] & mask); 1255 pmap_vsid_bitmap[idx] &= ~mask; 1256 } 1257 1258 /* 1259 * Copy the range specified by src_addr/len 1260 * from the source map to the range dst_addr/len 1261 * in the destination map. 1262 * 1263 * This routine is only advisory and need not do anything. 1264 */ 1265 void 1266 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr, 1267 vsize_t len, vaddr_t src_addr) 1268 { 1269 PMAPCOUNT(copies); 1270 } 1271 1272 /* 1273 * Require that all active physical maps contain no 1274 * incorrect entries NOW. 1275 */ 1276 void 1277 pmap_update(struct pmap *pmap) 1278 { 1279 PMAPCOUNT(updates); 1280 TLBSYNC(); 1281 } 1282 1283 /* 1284 * Garbage collects the physical map system for 1285 * pages which are no longer used. 1286 * Success need not be guaranteed -- that is, there 1287 * may well be pages which are not referenced, but 1288 * others may be collected. 1289 * Called by the pageout daemon when pages are scarce. 1290 */ 1291 void 1292 pmap_collect(pmap_t pm) 1293 { 1294 PMAPCOUNT(collects); 1295 } 1296 1297 static __inline int 1298 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx) 1299 { 1300 int pteidx; 1301 /* 1302 * We can find the actual pte entry without searching by 1303 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9] 1304 * and by noticing the HID bit. 1305 */ 1306 pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo); 1307 if (pvo->pvo_pte.pte_hi & PTE_HID) 1308 pteidx ^= pmap_pteg_mask * 8; 1309 return pteidx; 1310 } 1311 1312 volatile struct pte * 1313 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx) 1314 { 1315 volatile struct pte *pt; 1316 1317 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1318 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) 1319 return NULL; 1320 #endif 1321 1322 /* 1323 * If we haven't been supplied the ptegidx, calculate it. 1324 */ 1325 if (pteidx == -1) { 1326 int ptegidx; 1327 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1328 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1329 } 1330 1331 pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7]; 1332 1333 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1334 return pt; 1335 #else 1336 if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) { 1337 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1338 "pvo but no valid pte index", pvo); 1339 } 1340 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) { 1341 panic("pmap_pvo_to_pte: pvo %p: has valid pte index in " 1342 "pvo but no valid pte", pvo); 1343 } 1344 1345 if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) { 1346 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) { 1347 #if defined(DEBUG) || defined(PMAPCHECK) 1348 pmap_pte_print(pt); 1349 #endif 1350 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1351 "pmap_pteg_table %p but invalid in pvo", 1352 pvo, pt); 1353 } 1354 if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) { 1355 #if defined(DEBUG) || defined(PMAPCHECK) 1356 pmap_pte_print(pt); 1357 #endif 1358 panic("pmap_pvo_to_pte: pvo %p: pvo pte does " 1359 "not match pte %p in pmap_pteg_table", 1360 pvo, pt); 1361 } 1362 return pt; 1363 } 1364 1365 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1366 #if defined(DEBUG) || defined(PMAPCHECK) 1367 pmap_pte_print(pt); 1368 #endif 1369 panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in " 1370 "pmap_pteg_table but valid in pvo", pvo, pt); 1371 } 1372 return NULL; 1373 #endif /* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */ 1374 } 1375 1376 struct pvo_entry * 1377 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p) 1378 { 1379 struct pvo_entry *pvo; 1380 int ptegidx; 1381 1382 va &= ~ADDR_POFF; 1383 ptegidx = va_to_pteg(pm, va); 1384 1385 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1386 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1387 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 1388 panic("pmap_pvo_find_va: invalid pvo %p on " 1389 "list %#x (%p)", pvo, ptegidx, 1390 &pmap_pvo_table[ptegidx]); 1391 #endif 1392 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1393 if (pteidx_p) 1394 *pteidx_p = pmap_pvo_pte_index(pvo, ptegidx); 1395 return pvo; 1396 } 1397 } 1398 return NULL; 1399 } 1400 1401 #if defined(DEBUG) || defined(PMAPCHECK) 1402 void 1403 pmap_pvo_check(const struct pvo_entry *pvo) 1404 { 1405 struct pvo_head *pvo_head; 1406 struct pvo_entry *pvo0; 1407 volatile struct pte *pt; 1408 int failed = 0; 1409 1410 if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH) 1411 panic("pmap_pvo_check: pvo %p: invalid address", pvo); 1412 1413 if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) { 1414 printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n", 1415 pvo, pvo->pvo_pmap); 1416 failed = 1; 1417 } 1418 1419 if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH || 1420 (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) { 1421 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1422 pvo, TAILQ_NEXT(pvo, pvo_olink)); 1423 failed = 1; 1424 } 1425 1426 if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH || 1427 (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) { 1428 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1429 pvo, LIST_NEXT(pvo, pvo_vlink)); 1430 failed = 1; 1431 } 1432 1433 if (pvo->pvo_vaddr & PVO_MANAGED) { 1434 pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL); 1435 } else { 1436 if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) { 1437 printf("pmap_pvo_check: pvo %p: non kernel address " 1438 "on kernel unmanaged list\n", pvo); 1439 failed = 1; 1440 } 1441 pvo_head = &pmap_pvo_kunmanaged; 1442 } 1443 LIST_FOREACH(pvo0, pvo_head, pvo_vlink) { 1444 if (pvo0 == pvo) 1445 break; 1446 } 1447 if (pvo0 == NULL) { 1448 printf("pmap_pvo_check: pvo %p: not present " 1449 "on its vlist head %p\n", pvo, pvo_head); 1450 failed = 1; 1451 } 1452 if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) { 1453 printf("pmap_pvo_check: pvo %p: not present " 1454 "on its olist head\n", pvo); 1455 failed = 1; 1456 } 1457 pt = pmap_pvo_to_pte(pvo, -1); 1458 if (pt == NULL) { 1459 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1460 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1461 "no PTE\n", pvo); 1462 failed = 1; 1463 } 1464 } else { 1465 if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] || 1466 (uintptr_t) pt >= 1467 (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) { 1468 printf("pmap_pvo_check: pvo %p: pte %p not in " 1469 "pteg table\n", pvo, pt); 1470 failed = 1; 1471 } 1472 if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) { 1473 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1474 "no PTE\n", pvo); 1475 failed = 1; 1476 } 1477 if (pvo->pvo_pte.pte_hi != pt->pte_hi) { 1478 printf("pmap_pvo_check: pvo %p: pte_hi differ: " 1479 "%#x/%#x\n", pvo, (unsigned int) pvo->pvo_pte.pte_hi, (unsigned int) pt->pte_hi); 1480 failed = 1; 1481 } 1482 if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) & 1483 (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) { 1484 printf("pmap_pvo_check: pvo %p: pte_lo differ: " 1485 "%#x/%#x\n", pvo, 1486 (unsigned int) (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)), 1487 (unsigned int) (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN))); 1488 failed = 1; 1489 } 1490 if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) { 1491 printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#lx" 1492 " doesn't not match PVO's VA %#lx\n", 1493 pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo)); 1494 failed = 1; 1495 } 1496 if (failed) 1497 pmap_pte_print(pt); 1498 } 1499 if (failed) 1500 panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo, 1501 pvo->pvo_pmap); 1502 } 1503 #endif /* DEBUG || PMAPCHECK */ 1504 1505 /* 1506 * Search the PVO table looking for a non-wired entry. 1507 * If we find one, remove it and return it. 1508 */ 1509 1510 struct pvo_entry * 1511 pmap_pvo_reclaim(struct pmap *pm) 1512 { 1513 struct pvo_tqhead *pvoh; 1514 struct pvo_entry *pvo; 1515 uint32_t idx, endidx; 1516 1517 endidx = pmap_pvo_reclaim_nextidx; 1518 for (idx = (endidx + 1) & pmap_pteg_mask; idx != endidx; 1519 idx = (idx + 1) & pmap_pteg_mask) { 1520 pvoh = &pmap_pvo_table[idx]; 1521 TAILQ_FOREACH(pvo, pvoh, pvo_olink) { 1522 if ((pvo->pvo_vaddr & PVO_WIRED) == 0) { 1523 pmap_pvo_remove(pvo, -1, FALSE); 1524 pmap_pvo_reclaim_nextidx = idx; 1525 PMAPCOUNT(pvos_reclaimed); 1526 return pvo; 1527 } 1528 } 1529 } 1530 return NULL; 1531 } 1532 1533 /* 1534 * This returns whether this is the first mapping of a page. 1535 */ 1536 int 1537 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head, 1538 vaddr_t va, paddr_t pa, register_t pte_lo, int flags) 1539 { 1540 struct pvo_entry *pvo; 1541 struct pvo_tqhead *pvoh; 1542 register_t msr; 1543 int ptegidx; 1544 int i; 1545 int poolflags = PR_NOWAIT; 1546 1547 /* 1548 * Compute the PTE Group index. 1549 */ 1550 va &= ~ADDR_POFF; 1551 ptegidx = va_to_pteg(pm, va); 1552 1553 msr = pmap_interrupts_off(); 1554 1555 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1556 if (pmap_pvo_remove_depth > 0) 1557 panic("pmap_pvo_enter: called while pmap_pvo_remove active!"); 1558 if (++pmap_pvo_enter_depth > 1) 1559 panic("pmap_pvo_enter: called recursively!"); 1560 #endif 1561 1562 /* 1563 * Remove any existing mapping for this page. Reuse the 1564 * pvo entry if there a mapping. 1565 */ 1566 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1567 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1568 #ifdef DEBUG 1569 if ((pmapdebug & PMAPDEBUG_PVOENTER) && 1570 ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) & 1571 ~(PTE_REF|PTE_CHG)) == 0 && 1572 va < VM_MIN_KERNEL_ADDRESS) { 1573 printf("pmap_pvo_enter: pvo %p: dup %#x/%#lx\n", 1574 pvo, (unsigned int) pvo->pvo_pte.pte_lo, (unsigned int) pte_lo|pa); 1575 printf("pmap_pvo_enter: pte_hi=%#x sr=%#x\n", 1576 (unsigned int) pvo->pvo_pte.pte_hi, 1577 (unsigned int) pm->pm_sr[va >> ADDR_SR_SHFT]); 1578 pmap_pte_print(pmap_pvo_to_pte(pvo, -1)); 1579 #ifdef DDBX 1580 Debugger(); 1581 #endif 1582 } 1583 #endif 1584 PMAPCOUNT(mappings_replaced); 1585 pmap_pvo_remove(pvo, -1, TRUE); 1586 break; 1587 } 1588 } 1589 1590 /* 1591 * If we aren't overwriting an mapping, try to allocate 1592 */ 1593 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1594 --pmap_pvo_enter_depth; 1595 #endif 1596 pmap_interrupts_restore(msr); 1597 pvo = pool_get(pl, poolflags); 1598 1599 #ifdef DEBUG 1600 /* 1601 * Exercise pmap_pvo_reclaim() a little. 1602 */ 1603 if (pvo && (flags & PMAP_CANFAIL) != 0 && 1604 pmap_pvo_reclaim_debugctr++ > 0x1000 && 1605 (pmap_pvo_reclaim_debugctr & 0xff) == 0) { 1606 pool_put(pl, pvo); 1607 pvo = NULL; 1608 } 1609 #endif 1610 1611 msr = pmap_interrupts_off(); 1612 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1613 ++pmap_pvo_enter_depth; 1614 #endif 1615 if (pvo == NULL) { 1616 pvo = pmap_pvo_reclaim(pm); 1617 if (pvo == NULL) { 1618 if ((flags & PMAP_CANFAIL) == 0) 1619 panic("pmap_pvo_enter: failed"); 1620 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1621 pmap_pvo_enter_depth--; 1622 #endif 1623 PMAPCOUNT(pvos_failed); 1624 pmap_interrupts_restore(msr); 1625 return ENOMEM; 1626 } 1627 } 1628 1629 pvo->pvo_vaddr = va; 1630 pvo->pvo_pmap = pm; 1631 pvo->pvo_vaddr &= ~ADDR_POFF; 1632 if (flags & VM_PROT_EXECUTE) { 1633 PMAPCOUNT(exec_mappings); 1634 pvo_set_exec(pvo); 1635 } 1636 if (flags & PMAP_WIRED) 1637 pvo->pvo_vaddr |= PVO_WIRED; 1638 if (pvo_head != &pmap_pvo_kunmanaged) { 1639 pvo->pvo_vaddr |= PVO_MANAGED; 1640 PMAPCOUNT(mappings); 1641 } else { 1642 PMAPCOUNT(kernel_mappings); 1643 } 1644 pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo); 1645 1646 LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink); 1647 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1648 pvo->pvo_pmap->pm_stats.wired_count++; 1649 pvo->pvo_pmap->pm_stats.resident_count++; 1650 #if defined(DEBUG) 1651 if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) 1652 DPRINTFN(PVOENTER, 1653 ("pmap_pvo_enter: pvo %p: pm %p va %#lx pa %#lx\n", 1654 pvo, pm, va, pa)); 1655 #endif 1656 1657 /* 1658 * We hope this succeeds but it isn't required. 1659 */ 1660 pvoh = &pmap_pvo_table[ptegidx]; 1661 i = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 1662 if (i >= 0) { 1663 PVO_PTEGIDX_SET(pvo, i); 1664 PVO_WHERE(pvo, ENTER_INSERT); 1665 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 1666 ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]); 1667 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 1668 } else { 1669 /* 1670 * Since we didn't have room for this entry (which makes it 1671 * and evicted entry), place it at the head of the list. 1672 */ 1673 TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink); 1674 PMAPCOUNT(ptes_evicted); 1675 pm->pm_evictions++; 1676 /* 1677 * If this is a kernel page, make sure it's active. 1678 */ 1679 if (pm == pmap_kernel()) { 1680 i = pmap_pte_spill(pm, va, FALSE); 1681 KASSERT(i); 1682 } 1683 } 1684 PMAP_PVO_CHECK(pvo); /* sanity check */ 1685 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1686 pmap_pvo_enter_depth--; 1687 #endif 1688 pmap_interrupts_restore(msr); 1689 return 0; 1690 } 1691 1692 void 1693 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx, boolean_t do_free) 1694 { 1695 volatile struct pte *pt; 1696 int ptegidx; 1697 1698 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1699 if (++pmap_pvo_remove_depth > 1) 1700 panic("pmap_pvo_remove: called recursively!"); 1701 #endif 1702 1703 /* 1704 * If we haven't been supplied the ptegidx, calculate it. 1705 */ 1706 if (pteidx == -1) { 1707 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1708 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1709 } else { 1710 ptegidx = pteidx >> 3; 1711 if (pvo->pvo_pte.pte_hi & PTE_HID) 1712 ptegidx ^= pmap_pteg_mask; 1713 } 1714 PMAP_PVO_CHECK(pvo); /* sanity check */ 1715 1716 /* 1717 * If there is an active pte entry, we need to deactivate it 1718 * (and save the ref & chg bits). 1719 */ 1720 pt = pmap_pvo_to_pte(pvo, pteidx); 1721 if (pt != NULL) { 1722 pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 1723 PVO_WHERE(pvo, REMOVE); 1724 PVO_PTEGIDX_CLR(pvo); 1725 PMAPCOUNT(ptes_removed); 1726 } else { 1727 KASSERT(pvo->pvo_pmap->pm_evictions > 0); 1728 pvo->pvo_pmap->pm_evictions--; 1729 } 1730 1731 /* 1732 * Account for executable mappings. 1733 */ 1734 if (PVO_ISEXECUTABLE(pvo)) 1735 pvo_clear_exec(pvo); 1736 1737 /* 1738 * Update our statistics. 1739 */ 1740 pvo->pvo_pmap->pm_stats.resident_count--; 1741 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1742 pvo->pvo_pmap->pm_stats.wired_count--; 1743 1744 /* 1745 * Save the REF/CHG bits into their cache if the page is managed. 1746 */ 1747 if (pvo->pvo_vaddr & PVO_MANAGED) { 1748 register_t ptelo = pvo->pvo_pte.pte_lo; 1749 struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN); 1750 1751 if (pg != NULL) { 1752 pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG)); 1753 } 1754 PMAPCOUNT(unmappings); 1755 } else { 1756 PMAPCOUNT(kernel_unmappings); 1757 } 1758 1759 /* 1760 * Remove the PVO from its lists and return it to the pool. 1761 */ 1762 LIST_REMOVE(pvo, pvo_vlink); 1763 TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink); 1764 if (do_free) { 1765 pool_put(pvo->pvo_vaddr & PVO_MANAGED ? &pmap_mpvo_pool : 1766 &pmap_upvo_pool, pvo); 1767 } 1768 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1769 pmap_pvo_remove_depth--; 1770 #endif 1771 } 1772 1773 /* 1774 * Mark a mapping as executable. 1775 * If this is the first executable mapping in the segment, 1776 * clear the noexec flag. 1777 */ 1778 STATIC void 1779 pvo_set_exec(struct pvo_entry *pvo) 1780 { 1781 struct pmap *pm = pvo->pvo_pmap; 1782 1783 if (pm == pmap_kernel() || PVO_ISEXECUTABLE(pvo)) { 1784 return; 1785 } 1786 pvo->pvo_vaddr |= PVO_EXECUTABLE; 1787 #ifdef PPC_OEA 1788 { 1789 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1790 if (pm->pm_exec[sr]++ == 0) { 1791 pm->pm_sr[sr] &= ~SR_NOEXEC; 1792 } 1793 } 1794 #endif 1795 } 1796 1797 /* 1798 * Mark a mapping as non-executable. 1799 * If this was the last executable mapping in the segment, 1800 * set the noexec flag. 1801 */ 1802 STATIC void 1803 pvo_clear_exec(struct pvo_entry *pvo) 1804 { 1805 struct pmap *pm = pvo->pvo_pmap; 1806 1807 if (pm == pmap_kernel() || !PVO_ISEXECUTABLE(pvo)) { 1808 return; 1809 } 1810 pvo->pvo_vaddr &= ~PVO_EXECUTABLE; 1811 #ifdef PPC_OEA 1812 { 1813 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1814 if (--pm->pm_exec[sr] == 0) { 1815 pm->pm_sr[sr] |= SR_NOEXEC; 1816 } 1817 } 1818 #endif 1819 } 1820 1821 /* 1822 * Insert physical page at pa into the given pmap at virtual address va. 1823 */ 1824 int 1825 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags) 1826 { 1827 struct mem_region *mp; 1828 struct pvo_head *pvo_head; 1829 struct vm_page *pg; 1830 struct pool *pl; 1831 register_t pte_lo; 1832 int error; 1833 u_int pvo_flags; 1834 u_int was_exec = 0; 1835 1836 if (__predict_false(!pmap_initialized)) { 1837 pvo_head = &pmap_pvo_kunmanaged; 1838 pl = &pmap_upvo_pool; 1839 pvo_flags = 0; 1840 pg = NULL; 1841 was_exec = PTE_EXEC; 1842 } else { 1843 pvo_head = pa_to_pvoh(pa, &pg); 1844 pl = &pmap_mpvo_pool; 1845 pvo_flags = PVO_MANAGED; 1846 } 1847 1848 DPRINTFN(ENTER, 1849 ("pmap_enter(%p, 0x%lx, 0x%lx, 0x%x, 0x%x):", 1850 pm, va, pa, prot, flags)); 1851 1852 /* 1853 * If this is a managed page, and it's the first reference to the 1854 * page clear the execness of the page. Otherwise fetch the execness. 1855 */ 1856 if (pg != NULL) 1857 was_exec = pmap_attr_fetch(pg) & PTE_EXEC; 1858 1859 DPRINTFN(ENTER, (" was_exec=%d", was_exec)); 1860 1861 /* 1862 * Assume the page is cache inhibited and access is guarded unless 1863 * it's in our available memory array. If it is in the memory array, 1864 * asssume it's in memory coherent memory. 1865 */ 1866 pte_lo = PTE_IG; 1867 if ((flags & PMAP_NC) == 0) { 1868 for (mp = mem; mp->size; mp++) { 1869 if (pa >= mp->start && pa < mp->start + mp->size) { 1870 pte_lo = PTE_M; 1871 break; 1872 } 1873 } 1874 } 1875 1876 if (prot & VM_PROT_WRITE) 1877 pte_lo |= PTE_BW; 1878 else 1879 pte_lo |= PTE_BR; 1880 1881 /* 1882 * If this was in response to a fault, "pre-fault" the PTE's 1883 * changed/referenced bit appropriately. 1884 */ 1885 if (flags & VM_PROT_WRITE) 1886 pte_lo |= PTE_CHG; 1887 if (flags & VM_PROT_ALL) 1888 pte_lo |= PTE_REF; 1889 1890 /* 1891 * We need to know if this page can be executable 1892 */ 1893 flags |= (prot & VM_PROT_EXECUTE); 1894 1895 /* 1896 * Record mapping for later back-translation and pte spilling. 1897 * This will overwrite any existing mapping. 1898 */ 1899 error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags); 1900 1901 /* 1902 * Flush the real page from the instruction cache if this page is 1903 * mapped executable and cacheable and has not been flushed since 1904 * the last time it was modified. 1905 */ 1906 if (error == 0 && 1907 (flags & VM_PROT_EXECUTE) && 1908 (pte_lo & PTE_I) == 0 && 1909 was_exec == 0) { 1910 DPRINTFN(ENTER, (" syncicache")); 1911 PMAPCOUNT(exec_synced); 1912 pmap_syncicache(pa, PAGE_SIZE); 1913 if (pg != NULL) { 1914 pmap_attr_save(pg, PTE_EXEC); 1915 PMAPCOUNT(exec_cached); 1916 #if defined(DEBUG) || defined(PMAPDEBUG) 1917 if (pmapdebug & PMAPDEBUG_ENTER) 1918 printf(" marked-as-exec"); 1919 else if (pmapdebug & PMAPDEBUG_EXEC) 1920 printf("[pmap_enter: %#lx: marked-as-exec]\n", 1921 pg->phys_addr); 1922 1923 #endif 1924 } 1925 } 1926 1927 DPRINTFN(ENTER, (": error=%d\n", error)); 1928 1929 return error; 1930 } 1931 1932 void 1933 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot) 1934 { 1935 struct mem_region *mp; 1936 register_t pte_lo; 1937 int error; 1938 1939 if (va < VM_MIN_KERNEL_ADDRESS) 1940 panic("pmap_kenter_pa: attempt to enter " 1941 "non-kernel address %#lx!", va); 1942 1943 DPRINTFN(KENTER, 1944 ("pmap_kenter_pa(%#lx,%#lx,%#x)\n", va, pa, prot)); 1945 1946 /* 1947 * Assume the page is cache inhibited and access is guarded unless 1948 * it's in our available memory array. If it is in the memory array, 1949 * asssume it's in memory coherent memory. 1950 */ 1951 pte_lo = PTE_IG; 1952 if ((prot & PMAP_NC) == 0) { 1953 for (mp = mem; mp->size; mp++) { 1954 if (pa >= mp->start && pa < mp->start + mp->size) { 1955 pte_lo = PTE_M; 1956 break; 1957 } 1958 } 1959 } 1960 1961 if (prot & VM_PROT_WRITE) 1962 pte_lo |= PTE_BW; 1963 else 1964 pte_lo |= PTE_BR; 1965 1966 /* 1967 * We don't care about REF/CHG on PVOs on the unmanaged list. 1968 */ 1969 error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool, 1970 &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED); 1971 1972 if (error != 0) 1973 panic("pmap_kenter_pa: failed to enter va %#lx pa %#lx: %d", 1974 va, pa, error); 1975 } 1976 1977 void 1978 pmap_kremove(vaddr_t va, vsize_t len) 1979 { 1980 if (va < VM_MIN_KERNEL_ADDRESS) 1981 panic("pmap_kremove: attempt to remove " 1982 "non-kernel address %#lx!", va); 1983 1984 DPRINTFN(KREMOVE,("pmap_kremove(%#lx,%#lx)\n", va, len)); 1985 pmap_remove(pmap_kernel(), va, va + len); 1986 } 1987 1988 /* 1989 * Remove the given range of mapping entries. 1990 */ 1991 void 1992 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva) 1993 { 1994 struct pvo_entry *pvo; 1995 register_t msr; 1996 int pteidx; 1997 1998 msr = pmap_interrupts_off(); 1999 for (; va < endva; va += PAGE_SIZE) { 2000 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2001 if (pvo != NULL) { 2002 pmap_pvo_remove(pvo, pteidx, TRUE); 2003 } 2004 } 2005 pmap_interrupts_restore(msr); 2006 } 2007 2008 /* 2009 * Get the physical page address for the given pmap/virtual address. 2010 */ 2011 boolean_t 2012 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap) 2013 { 2014 struct pvo_entry *pvo; 2015 register_t msr; 2016 2017 /* 2018 * If this is a kernel pmap lookup, also check the battable 2019 * and if we get a hit, translate the VA to a PA using the 2020 * BAT entries. Don't check for VM_MAX_KENREL_ADDRESS is 2021 * that will wrap back to 0. 2022 */ 2023 if (pm == pmap_kernel() && 2024 (va < VM_MIN_KERNEL_ADDRESS || 2025 (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) { 2026 KASSERT((va >> ADDR_SR_SHFT) != USER_SR); 2027 if ((MFPVR() >> 16) != MPC601) { 2028 register_t batu = battable[va >> ADDR_SR_SHFT].batu; 2029 if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) { 2030 register_t batl = 2031 battable[va >> ADDR_SR_SHFT].batl; 2032 register_t mask = 2033 (~(batu & BAT_BL) << 15) & ~0x1ffffL; 2034 if (pap) 2035 *pap = (batl & mask) | (va & ~mask); 2036 return TRUE; 2037 } 2038 } else { 2039 register_t batu = battable[va >> 23].batu; 2040 register_t batl = battable[va >> 23].batl; 2041 register_t sr = iosrtable[va >> ADDR_SR_SHFT]; 2042 if (BAT601_VALID_P(batl) && 2043 BAT601_VA_MATCH_P(batu, batl, va)) { 2044 register_t mask = 2045 (~(batl & BAT601_BSM) << 17) & ~0x1ffffL; 2046 if (pap) 2047 *pap = (batl & mask) | (va & ~mask); 2048 return TRUE; 2049 } else if (SR601_VALID_P(sr) && 2050 SR601_PA_MATCH_P(sr, va)) { 2051 if (pap) 2052 *pap = va; 2053 return TRUE; 2054 } 2055 } 2056 return FALSE; 2057 } 2058 2059 msr = pmap_interrupts_off(); 2060 pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL); 2061 if (pvo != NULL) { 2062 PMAP_PVO_CHECK(pvo); /* sanity check */ 2063 if (pap) 2064 *pap = (pvo->pvo_pte.pte_lo & PTE_RPGN) 2065 | (va & ADDR_POFF); 2066 } 2067 pmap_interrupts_restore(msr); 2068 return pvo != NULL; 2069 } 2070 2071 /* 2072 * Lower the protection on the specified range of this pmap. 2073 */ 2074 void 2075 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot) 2076 { 2077 struct pvo_entry *pvo; 2078 volatile struct pte *pt; 2079 register_t msr; 2080 int pteidx; 2081 2082 /* 2083 * Since this routine only downgrades protection, we should 2084 * always be called with at least one bit not set. 2085 */ 2086 KASSERT(prot != VM_PROT_ALL); 2087 2088 /* 2089 * If there is no protection, this is equivalent to 2090 * remove the pmap from the pmap. 2091 */ 2092 if ((prot & VM_PROT_READ) == 0) { 2093 pmap_remove(pm, va, endva); 2094 return; 2095 } 2096 2097 msr = pmap_interrupts_off(); 2098 for (; va < endva; va += PAGE_SIZE) { 2099 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2100 if (pvo == NULL) 2101 continue; 2102 PMAP_PVO_CHECK(pvo); /* sanity check */ 2103 2104 /* 2105 * Revoke executable if asked to do so. 2106 */ 2107 if ((prot & VM_PROT_EXECUTE) == 0) 2108 pvo_clear_exec(pvo); 2109 2110 #if 0 2111 /* 2112 * If the page is already read-only, no change 2113 * needs to be made. 2114 */ 2115 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) 2116 continue; 2117 #endif 2118 /* 2119 * Grab the PTE pointer before we diddle with 2120 * the cached PTE copy. 2121 */ 2122 pt = pmap_pvo_to_pte(pvo, pteidx); 2123 /* 2124 * Change the protection of the page. 2125 */ 2126 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2127 pvo->pvo_pte.pte_lo |= PTE_BR; 2128 2129 /* 2130 * If the PVO is in the page table, update 2131 * that pte at well. 2132 */ 2133 if (pt != NULL) { 2134 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2135 PVO_WHERE(pvo, PMAP_PROTECT); 2136 PMAPCOUNT(ptes_changed); 2137 } 2138 2139 PMAP_PVO_CHECK(pvo); /* sanity check */ 2140 } 2141 pmap_interrupts_restore(msr); 2142 } 2143 2144 void 2145 pmap_unwire(pmap_t pm, vaddr_t va) 2146 { 2147 struct pvo_entry *pvo; 2148 register_t msr; 2149 2150 msr = pmap_interrupts_off(); 2151 pvo = pmap_pvo_find_va(pm, va, NULL); 2152 if (pvo != NULL) { 2153 if (pvo->pvo_vaddr & PVO_WIRED) { 2154 pvo->pvo_vaddr &= ~PVO_WIRED; 2155 pm->pm_stats.wired_count--; 2156 } 2157 PMAP_PVO_CHECK(pvo); /* sanity check */ 2158 } 2159 pmap_interrupts_restore(msr); 2160 } 2161 2162 /* 2163 * Lower the protection on the specified physical page. 2164 */ 2165 void 2166 pmap_page_protect(struct vm_page *pg, vm_prot_t prot) 2167 { 2168 struct pvo_head *pvo_head; 2169 struct pvo_entry *pvo, *next_pvo; 2170 volatile struct pte *pt; 2171 register_t msr; 2172 2173 KASSERT(prot != VM_PROT_ALL); 2174 msr = pmap_interrupts_off(); 2175 2176 /* 2177 * When UVM reuses a page, it does a pmap_page_protect with 2178 * VM_PROT_NONE. At that point, we can clear the exec flag 2179 * since we know the page will have different contents. 2180 */ 2181 if ((prot & VM_PROT_READ) == 0) { 2182 DPRINTFN(EXEC, ("[pmap_page_protect: %#lx: clear-exec]\n", 2183 pg->phys_addr)); 2184 if (pmap_attr_fetch(pg) & PTE_EXEC) { 2185 PMAPCOUNT(exec_uncached_page_protect); 2186 pmap_attr_clear(pg, PTE_EXEC); 2187 } 2188 } 2189 2190 pvo_head = vm_page_to_pvoh(pg); 2191 for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) { 2192 next_pvo = LIST_NEXT(pvo, pvo_vlink); 2193 PMAP_PVO_CHECK(pvo); /* sanity check */ 2194 2195 /* 2196 * Downgrading to no mapping at all, we just remove the entry. 2197 */ 2198 if ((prot & VM_PROT_READ) == 0) { 2199 pmap_pvo_remove(pvo, -1, TRUE); 2200 continue; 2201 } 2202 2203 /* 2204 * If EXEC permission is being revoked, just clear the 2205 * flag in the PVO. 2206 */ 2207 if ((prot & VM_PROT_EXECUTE) == 0) 2208 pvo_clear_exec(pvo); 2209 2210 /* 2211 * If this entry is already RO, don't diddle with the 2212 * page table. 2213 */ 2214 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) { 2215 PMAP_PVO_CHECK(pvo); 2216 continue; 2217 } 2218 2219 /* 2220 * Grab the PTE before the we diddle the bits so 2221 * pvo_to_pte can verify the pte contents are as 2222 * expected. 2223 */ 2224 pt = pmap_pvo_to_pte(pvo, -1); 2225 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2226 pvo->pvo_pte.pte_lo |= PTE_BR; 2227 if (pt != NULL) { 2228 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2229 PVO_WHERE(pvo, PMAP_PAGE_PROTECT); 2230 PMAPCOUNT(ptes_changed); 2231 } 2232 PMAP_PVO_CHECK(pvo); /* sanity check */ 2233 } 2234 pmap_interrupts_restore(msr); 2235 } 2236 2237 /* 2238 * Activate the address space for the specified process. If the process 2239 * is the current process, load the new MMU context. 2240 */ 2241 void 2242 pmap_activate(struct lwp *l) 2243 { 2244 struct pcb *pcb = &l->l_addr->u_pcb; 2245 pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap; 2246 2247 DPRINTFN(ACTIVATE, 2248 ("pmap_activate: lwp %p (curlwp %p)\n", l, curlwp)); 2249 2250 /* 2251 * XXX Normally performed in cpu_fork(). 2252 */ 2253 pcb->pcb_pm = pmap; 2254 2255 /* 2256 * In theory, the SR registers need only be valid on return 2257 * to user space wait to do them there. 2258 */ 2259 if (l == curlwp) { 2260 /* Store pointer to new current pmap. */ 2261 curpm = pmap; 2262 } 2263 } 2264 2265 /* 2266 * Deactivate the specified process's address space. 2267 */ 2268 void 2269 pmap_deactivate(struct lwp *l) 2270 { 2271 } 2272 2273 boolean_t 2274 pmap_query_bit(struct vm_page *pg, int ptebit) 2275 { 2276 struct pvo_entry *pvo; 2277 volatile struct pte *pt; 2278 register_t msr; 2279 2280 if (pmap_attr_fetch(pg) & ptebit) 2281 return TRUE; 2282 2283 msr = pmap_interrupts_off(); 2284 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2285 PMAP_PVO_CHECK(pvo); /* sanity check */ 2286 /* 2287 * See if we saved the bit off. If so cache, it and return 2288 * success. 2289 */ 2290 if (pvo->pvo_pte.pte_lo & ptebit) { 2291 pmap_attr_save(pg, ptebit); 2292 PMAP_PVO_CHECK(pvo); /* sanity check */ 2293 pmap_interrupts_restore(msr); 2294 return TRUE; 2295 } 2296 } 2297 /* 2298 * No luck, now go thru the hard part of looking at the ptes 2299 * themselves. Sync so any pending REF/CHG bits are flushed 2300 * to the PTEs. 2301 */ 2302 SYNC(); 2303 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2304 PMAP_PVO_CHECK(pvo); /* sanity check */ 2305 /* 2306 * See if this pvo have a valid PTE. If so, fetch the 2307 * REF/CHG bits from the valid PTE. If the appropriate 2308 * ptebit is set, cache, it and return success. 2309 */ 2310 pt = pmap_pvo_to_pte(pvo, -1); 2311 if (pt != NULL) { 2312 pmap_pte_synch(pt, &pvo->pvo_pte); 2313 if (pvo->pvo_pte.pte_lo & ptebit) { 2314 pmap_attr_save(pg, ptebit); 2315 PMAP_PVO_CHECK(pvo); /* sanity check */ 2316 pmap_interrupts_restore(msr); 2317 return TRUE; 2318 } 2319 } 2320 } 2321 pmap_interrupts_restore(msr); 2322 return FALSE; 2323 } 2324 2325 boolean_t 2326 pmap_clear_bit(struct vm_page *pg, int ptebit) 2327 { 2328 struct pvo_head *pvoh = vm_page_to_pvoh(pg); 2329 struct pvo_entry *pvo; 2330 volatile struct pte *pt; 2331 register_t msr; 2332 int rv = 0; 2333 2334 msr = pmap_interrupts_off(); 2335 2336 /* 2337 * Fetch the cache value 2338 */ 2339 rv |= pmap_attr_fetch(pg); 2340 2341 /* 2342 * Clear the cached value. 2343 */ 2344 pmap_attr_clear(pg, ptebit); 2345 2346 /* 2347 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we 2348 * can reset the right ones). Note that since the pvo entries and 2349 * list heads are accessed via BAT0 and are never placed in the 2350 * page table, we don't have to worry about further accesses setting 2351 * the REF/CHG bits. 2352 */ 2353 SYNC(); 2354 2355 /* 2356 * For each pvo entry, clear pvo's ptebit. If this pvo have a 2357 * valid PTE. If so, clear the ptebit from the valid PTE. 2358 */ 2359 LIST_FOREACH(pvo, pvoh, pvo_vlink) { 2360 PMAP_PVO_CHECK(pvo); /* sanity check */ 2361 pt = pmap_pvo_to_pte(pvo, -1); 2362 if (pt != NULL) { 2363 /* 2364 * Only sync the PTE if the bit we are looking 2365 * for is not already set. 2366 */ 2367 if ((pvo->pvo_pte.pte_lo & ptebit) == 0) 2368 pmap_pte_synch(pt, &pvo->pvo_pte); 2369 /* 2370 * If the bit we are looking for was already set, 2371 * clear that bit in the pte. 2372 */ 2373 if (pvo->pvo_pte.pte_lo & ptebit) 2374 pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit); 2375 } 2376 rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF); 2377 pvo->pvo_pte.pte_lo &= ~ptebit; 2378 PMAP_PVO_CHECK(pvo); /* sanity check */ 2379 } 2380 pmap_interrupts_restore(msr); 2381 2382 /* 2383 * If we are clearing the modify bit and this page was marked EXEC 2384 * and the user of the page thinks the page was modified, then we 2385 * need to clean it from the icache if it's mapped or clear the EXEC 2386 * bit if it's not mapped. The page itself might not have the CHG 2387 * bit set if the modification was done via DMA to the page. 2388 */ 2389 if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) { 2390 if (LIST_EMPTY(pvoh)) { 2391 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: clear-exec]\n", 2392 pg->phys_addr)); 2393 pmap_attr_clear(pg, PTE_EXEC); 2394 PMAPCOUNT(exec_uncached_clear_modify); 2395 } else { 2396 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: syncicache]\n", 2397 pg->phys_addr)); 2398 pmap_syncicache(pg->phys_addr, PAGE_SIZE); 2399 PMAPCOUNT(exec_synced_clear_modify); 2400 } 2401 } 2402 return (rv & ptebit) != 0; 2403 } 2404 2405 void 2406 pmap_procwr(struct proc *p, vaddr_t va, size_t len) 2407 { 2408 struct pvo_entry *pvo; 2409 size_t offset = va & ADDR_POFF; 2410 int s; 2411 2412 s = splvm(); 2413 while (len > 0) { 2414 size_t seglen = PAGE_SIZE - offset; 2415 if (seglen > len) 2416 seglen = len; 2417 pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL); 2418 if (pvo != NULL && PVO_ISEXECUTABLE(pvo)) { 2419 pmap_syncicache( 2420 (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen); 2421 PMAP_PVO_CHECK(pvo); 2422 } 2423 va += seglen; 2424 len -= seglen; 2425 offset = 0; 2426 } 2427 splx(s); 2428 } 2429 2430 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 2431 void 2432 pmap_pte_print(volatile struct pte *pt) 2433 { 2434 printf("PTE %p: ", pt); 2435 /* High word: */ 2436 printf("0x%08lx: [", pt->pte_hi); 2437 printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i'); 2438 printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-'); 2439 printf("0x%06lx 0x%02lx", 2440 (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT, 2441 pt->pte_hi & PTE_API); 2442 printf(" (va 0x%08lx)] ", pmap_pte_to_va(pt)); 2443 /* Low word: */ 2444 printf(" 0x%08lx: [", pt->pte_lo); 2445 printf("0x%05lx... ", pt->pte_lo >> 12); 2446 printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u'); 2447 printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n'); 2448 printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.'); 2449 printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.'); 2450 printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.'); 2451 printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.'); 2452 switch (pt->pte_lo & PTE_PP) { 2453 case PTE_BR: printf("br]\n"); break; 2454 case PTE_BW: printf("bw]\n"); break; 2455 case PTE_SO: printf("so]\n"); break; 2456 case PTE_SW: printf("sw]\n"); break; 2457 } 2458 } 2459 #endif 2460 2461 #if defined(DDB) 2462 void 2463 pmap_pteg_check(void) 2464 { 2465 volatile struct pte *pt; 2466 int i; 2467 int ptegidx; 2468 u_int p_valid = 0; 2469 u_int s_valid = 0; 2470 u_int invalid = 0; 2471 2472 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2473 for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) { 2474 if (pt->pte_hi & PTE_VALID) { 2475 if (pt->pte_hi & PTE_HID) 2476 s_valid++; 2477 else 2478 p_valid++; 2479 } else 2480 invalid++; 2481 } 2482 } 2483 printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n", 2484 p_valid, p_valid, s_valid, s_valid, 2485 invalid, invalid); 2486 } 2487 2488 void 2489 pmap_print_mmuregs(void) 2490 { 2491 int i; 2492 u_int cpuvers; 2493 #ifndef PPC_OEA64 2494 vaddr_t addr; 2495 register_t soft_sr[16]; 2496 #endif 2497 struct bat soft_ibat[4]; 2498 struct bat soft_dbat[4]; 2499 register_t sdr1; 2500 2501 cpuvers = MFPVR() >> 16; 2502 __asm __volatile ("mfsdr1 %0" : "=r"(sdr1)); 2503 #ifndef PPC_OEA64 2504 addr = 0; 2505 for (i = 0; i < 16; i++) { 2506 soft_sr[i] = MFSRIN(addr); 2507 addr += (1 << ADDR_SR_SHFT); 2508 } 2509 #endif 2510 2511 /* read iBAT (601: uBAT) registers */ 2512 __asm __volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu)); 2513 __asm __volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl)); 2514 __asm __volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu)); 2515 __asm __volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl)); 2516 __asm __volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu)); 2517 __asm __volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl)); 2518 __asm __volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu)); 2519 __asm __volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl)); 2520 2521 2522 if (cpuvers != MPC601) { 2523 /* read dBAT registers */ 2524 __asm __volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu)); 2525 __asm __volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl)); 2526 __asm __volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu)); 2527 __asm __volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl)); 2528 __asm __volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu)); 2529 __asm __volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl)); 2530 __asm __volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu)); 2531 __asm __volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl)); 2532 } 2533 2534 printf("SDR1:\t0x%lx\n", (long) sdr1); 2535 #ifndef PPC_OEA64 2536 printf("SR[]:\t"); 2537 for (i = 0; i < 4; i++) 2538 printf("0x%08lx, ", soft_sr[i]); 2539 printf("\n\t"); 2540 for ( ; i < 8; i++) 2541 printf("0x%08lx, ", soft_sr[i]); 2542 printf("\n\t"); 2543 for ( ; i < 12; i++) 2544 printf("0x%08lx, ", soft_sr[i]); 2545 printf("\n\t"); 2546 for ( ; i < 16; i++) 2547 printf("0x%08lx, ", soft_sr[i]); 2548 printf("\n"); 2549 #endif 2550 2551 printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i'); 2552 for (i = 0; i < 4; i++) { 2553 printf("0x%08lx 0x%08lx, ", 2554 soft_ibat[i].batu, soft_ibat[i].batl); 2555 if (i == 1) 2556 printf("\n\t"); 2557 } 2558 if (cpuvers != MPC601) { 2559 printf("\ndBAT[]:\t"); 2560 for (i = 0; i < 4; i++) { 2561 printf("0x%08lx 0x%08lx, ", 2562 soft_dbat[i].batu, soft_dbat[i].batl); 2563 if (i == 1) 2564 printf("\n\t"); 2565 } 2566 } 2567 printf("\n"); 2568 } 2569 2570 void 2571 pmap_print_pte(pmap_t pm, vaddr_t va) 2572 { 2573 struct pvo_entry *pvo; 2574 volatile struct pte *pt; 2575 int pteidx; 2576 2577 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2578 if (pvo != NULL) { 2579 pt = pmap_pvo_to_pte(pvo, pteidx); 2580 if (pt != NULL) { 2581 printf("VA %#lx -> %p -> %s %#lx, %#lx\n", 2582 va, pt, 2583 pt->pte_hi & PTE_HID ? "(sec)" : "(pri)", 2584 pt->pte_hi, pt->pte_lo); 2585 } else { 2586 printf("No valid PTE found\n"); 2587 } 2588 } else { 2589 printf("Address not in pmap\n"); 2590 } 2591 } 2592 2593 void 2594 pmap_pteg_dist(void) 2595 { 2596 struct pvo_entry *pvo; 2597 int ptegidx; 2598 int depth; 2599 int max_depth = 0; 2600 unsigned int depths[64]; 2601 2602 memset(depths, 0, sizeof(depths)); 2603 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2604 depth = 0; 2605 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2606 depth++; 2607 } 2608 if (depth > max_depth) 2609 max_depth = depth; 2610 if (depth > 63) 2611 depth = 63; 2612 depths[depth]++; 2613 } 2614 2615 for (depth = 0; depth < 64; depth++) { 2616 printf(" [%2d]: %8u", depth, depths[depth]); 2617 if ((depth & 3) == 3) 2618 printf("\n"); 2619 if (depth == max_depth) 2620 break; 2621 } 2622 if ((depth & 3) != 3) 2623 printf("\n"); 2624 printf("Max depth found was %d\n", max_depth); 2625 } 2626 #endif /* DEBUG */ 2627 2628 #if defined(PMAPCHECK) || defined(DEBUG) 2629 void 2630 pmap_pvo_verify(void) 2631 { 2632 int ptegidx; 2633 int s; 2634 2635 s = splvm(); 2636 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2637 struct pvo_entry *pvo; 2638 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2639 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 2640 panic("pmap_pvo_verify: invalid pvo %p " 2641 "on list %#x", pvo, ptegidx); 2642 pmap_pvo_check(pvo); 2643 } 2644 } 2645 splx(s); 2646 } 2647 #endif /* PMAPCHECK */ 2648 2649 2650 void * 2651 pmap_pool_ualloc(struct pool *pp, int flags) 2652 { 2653 struct pvo_page *pvop; 2654 2655 pvop = SIMPLEQ_FIRST(&pmap_upvop_head); 2656 if (pvop != NULL) { 2657 pmap_upvop_free--; 2658 SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link); 2659 return pvop; 2660 } 2661 if (uvm.page_init_done != TRUE) { 2662 return (void *) uvm_pageboot_alloc(PAGE_SIZE); 2663 } 2664 return pmap_pool_malloc(pp, flags); 2665 } 2666 2667 void * 2668 pmap_pool_malloc(struct pool *pp, int flags) 2669 { 2670 struct pvo_page *pvop; 2671 struct vm_page *pg; 2672 2673 pvop = SIMPLEQ_FIRST(&pmap_mpvop_head); 2674 if (pvop != NULL) { 2675 pmap_mpvop_free--; 2676 SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link); 2677 return pvop; 2678 } 2679 again: 2680 pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE, 2681 UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256); 2682 if (__predict_false(pg == NULL)) { 2683 if (flags & PR_WAITOK) { 2684 uvm_wait("plpg"); 2685 goto again; 2686 } else { 2687 return (0); 2688 } 2689 } 2690 return (void *) VM_PAGE_TO_PHYS(pg); 2691 } 2692 2693 void 2694 pmap_pool_ufree(struct pool *pp, void *va) 2695 { 2696 struct pvo_page *pvop; 2697 #if 0 2698 if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) { 2699 pmap_pool_mfree(va, size, tag); 2700 return; 2701 } 2702 #endif 2703 pvop = va; 2704 SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link); 2705 pmap_upvop_free++; 2706 if (pmap_upvop_free > pmap_upvop_maxfree) 2707 pmap_upvop_maxfree = pmap_upvop_free; 2708 } 2709 2710 void 2711 pmap_pool_mfree(struct pool *pp, void *va) 2712 { 2713 struct pvo_page *pvop; 2714 2715 pvop = va; 2716 SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link); 2717 pmap_mpvop_free++; 2718 if (pmap_mpvop_free > pmap_mpvop_maxfree) 2719 pmap_mpvop_maxfree = pmap_mpvop_free; 2720 #if 0 2721 uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va)); 2722 #endif 2723 } 2724 2725 /* 2726 * This routine in bootstraping to steal to-be-managed memory (which will 2727 * then be unmanaged). We use it to grab from the first 256MB for our 2728 * pmap needs and above 256MB for other stuff. 2729 */ 2730 vaddr_t 2731 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp) 2732 { 2733 vsize_t size; 2734 vaddr_t va; 2735 paddr_t pa = 0; 2736 int npgs, bank; 2737 struct vm_physseg *ps; 2738 2739 if (uvm.page_init_done == TRUE) 2740 panic("pmap_steal_memory: called _after_ bootstrap"); 2741 2742 *vstartp = VM_MIN_KERNEL_ADDRESS; 2743 *vendp = VM_MAX_KERNEL_ADDRESS; 2744 2745 size = round_page(vsize); 2746 npgs = atop(size); 2747 2748 /* 2749 * PA 0 will never be among those given to UVM so we can use it 2750 * to indicate we couldn't steal any memory. 2751 */ 2752 for (ps = vm_physmem, bank = 0; bank < vm_nphysseg; bank++, ps++) { 2753 if (ps->free_list == VM_FREELIST_FIRST256 && 2754 ps->avail_end - ps->avail_start >= npgs) { 2755 pa = ptoa(ps->avail_start); 2756 break; 2757 } 2758 } 2759 2760 if (pa == 0) 2761 panic("pmap_steal_memory: no approriate memory to steal!"); 2762 2763 ps->avail_start += npgs; 2764 ps->start += npgs; 2765 2766 /* 2767 * If we've used up all the pages in the segment, remove it and 2768 * compact the list. 2769 */ 2770 if (ps->avail_start == ps->end) { 2771 /* 2772 * If this was the last one, then a very bad thing has occurred 2773 */ 2774 if (--vm_nphysseg == 0) 2775 panic("pmap_steal_memory: out of memory!"); 2776 2777 printf("pmap_steal_memory: consumed bank %d\n", bank); 2778 for (; bank < vm_nphysseg; bank++, ps++) { 2779 ps[0] = ps[1]; 2780 } 2781 } 2782 2783 va = (vaddr_t) pa; 2784 memset((caddr_t) va, 0, size); 2785 pmap_pages_stolen += npgs; 2786 #ifdef DEBUG 2787 if (pmapdebug && npgs > 1) { 2788 u_int cnt = 0; 2789 for (bank = 0, ps = vm_physmem; bank < vm_nphysseg; bank++, ps++) 2790 cnt += ps->avail_end - ps->avail_start; 2791 printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n", 2792 npgs, pmap_pages_stolen, cnt); 2793 } 2794 #endif 2795 2796 return va; 2797 } 2798 2799 /* 2800 * Find a chuck of memory with right size and alignment. 2801 */ 2802 void * 2803 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end) 2804 { 2805 struct mem_region *mp; 2806 paddr_t s, e; 2807 int i, j; 2808 2809 size = round_page(size); 2810 2811 DPRINTFN(BOOT, 2812 ("pmap_boot_find_memory: size=%lx, alignment=%lx, at_end=%d", 2813 size, alignment, at_end)); 2814 2815 if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0) 2816 panic("pmap_boot_find_memory: invalid alignment %lx", 2817 alignment); 2818 2819 if (at_end) { 2820 if (alignment != PAGE_SIZE) 2821 panic("pmap_boot_find_memory: invalid ending " 2822 "alignment %lx", alignment); 2823 2824 for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) { 2825 s = mp->start + mp->size - size; 2826 if (s >= mp->start && mp->size >= size) { 2827 DPRINTFN(BOOT,(": %lx\n", s)); 2828 DPRINTFN(BOOT, 2829 ("pmap_boot_find_memory: b-avail[%d] start " 2830 "0x%lx size 0x%lx\n", mp - avail, 2831 mp->start, mp->size)); 2832 mp->size -= size; 2833 DPRINTFN(BOOT, 2834 ("pmap_boot_find_memory: a-avail[%d] start " 2835 "0x%lx size 0x%lx\n", mp - avail, 2836 mp->start, mp->size)); 2837 return (void *) s; 2838 } 2839 } 2840 panic("pmap_boot_find_memory: no available memory"); 2841 } 2842 2843 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2844 s = (mp->start + alignment - 1) & ~(alignment-1); 2845 e = s + size; 2846 2847 /* 2848 * Is the calculated region entirely within the region? 2849 */ 2850 if (s < mp->start || e > mp->start + mp->size) 2851 continue; 2852 2853 DPRINTFN(BOOT,(": %lx\n", s)); 2854 if (s == mp->start) { 2855 /* 2856 * If the block starts at the beginning of region, 2857 * adjust the size & start. (the region may now be 2858 * zero in length) 2859 */ 2860 DPRINTFN(BOOT, 2861 ("pmap_boot_find_memory: b-avail[%d] start " 2862 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2863 mp->start += size; 2864 mp->size -= size; 2865 DPRINTFN(BOOT, 2866 ("pmap_boot_find_memory: a-avail[%d] start " 2867 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2868 } else if (e == mp->start + mp->size) { 2869 /* 2870 * If the block starts at the beginning of region, 2871 * adjust only the size. 2872 */ 2873 DPRINTFN(BOOT, 2874 ("pmap_boot_find_memory: b-avail[%d] start " 2875 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2876 mp->size -= size; 2877 DPRINTFN(BOOT, 2878 ("pmap_boot_find_memory: a-avail[%d] start " 2879 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2880 } else { 2881 /* 2882 * Block is in the middle of the region, so we 2883 * have to split it in two. 2884 */ 2885 for (j = avail_cnt; j > i + 1; j--) { 2886 avail[j] = avail[j-1]; 2887 } 2888 DPRINTFN(BOOT, 2889 ("pmap_boot_find_memory: b-avail[%d] start " 2890 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2891 mp[1].start = e; 2892 mp[1].size = mp[0].start + mp[0].size - e; 2893 mp[0].size = s - mp[0].start; 2894 avail_cnt++; 2895 for (; i < avail_cnt; i++) { 2896 DPRINTFN(BOOT, 2897 ("pmap_boot_find_memory: a-avail[%d] " 2898 "start 0x%lx size 0x%lx\n", i, 2899 avail[i].start, avail[i].size)); 2900 } 2901 } 2902 return (void *) s; 2903 } 2904 panic("pmap_boot_find_memory: not enough memory for " 2905 "%lx/%lx allocation?", size, alignment); 2906 } 2907 2908 /* 2909 * This is not part of the defined PMAP interface and is specific to the 2910 * PowerPC architecture. This is called during initppc, before the system 2911 * is really initialized. 2912 */ 2913 void 2914 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend) 2915 { 2916 struct mem_region *mp, tmp; 2917 paddr_t s, e; 2918 psize_t size; 2919 int i, j; 2920 2921 /* 2922 * Get memory. 2923 */ 2924 mem_regions(&mem, &avail); 2925 #if defined(DEBUG) 2926 if (pmapdebug & PMAPDEBUG_BOOT) { 2927 printf("pmap_bootstrap: memory configuration:\n"); 2928 for (mp = mem; mp->size; mp++) { 2929 printf("pmap_bootstrap: mem start 0x%lx size 0x%lx\n", 2930 mp->start, mp->size); 2931 } 2932 for (mp = avail; mp->size; mp++) { 2933 printf("pmap_bootstrap: avail start 0x%lx size 0x%lx\n", 2934 mp->start, mp->size); 2935 } 2936 } 2937 #endif 2938 2939 /* 2940 * Find out how much physical memory we have and in how many chunks. 2941 */ 2942 for (mem_cnt = 0, mp = mem; mp->size; mp++) { 2943 if (mp->start >= pmap_memlimit) 2944 continue; 2945 if (mp->start + mp->size > pmap_memlimit) { 2946 size = pmap_memlimit - mp->start; 2947 physmem += btoc(size); 2948 } else { 2949 physmem += btoc(mp->size); 2950 } 2951 mem_cnt++; 2952 } 2953 2954 /* 2955 * Count the number of available entries. 2956 */ 2957 for (avail_cnt = 0, mp = avail; mp->size; mp++) 2958 avail_cnt++; 2959 2960 /* 2961 * Page align all regions. 2962 */ 2963 kernelstart = trunc_page(kernelstart); 2964 kernelend = round_page(kernelend); 2965 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2966 s = round_page(mp->start); 2967 mp->size -= (s - mp->start); 2968 mp->size = trunc_page(mp->size); 2969 mp->start = s; 2970 e = mp->start + mp->size; 2971 2972 DPRINTFN(BOOT, 2973 ("pmap_bootstrap: b-avail[%d] start 0x%lx size 0x%lx\n", 2974 i, mp->start, mp->size)); 2975 2976 /* 2977 * Don't allow the end to run beyond our artificial limit 2978 */ 2979 if (e > pmap_memlimit) 2980 e = pmap_memlimit; 2981 2982 /* 2983 * Is this region empty or strange? skip it. 2984 */ 2985 if (e <= s) { 2986 mp->start = 0; 2987 mp->size = 0; 2988 continue; 2989 } 2990 2991 /* 2992 * Does this overlap the beginning of kernel? 2993 * Does extend past the end of the kernel? 2994 */ 2995 else if (s < kernelstart && e > kernelstart) { 2996 if (e > kernelend) { 2997 avail[avail_cnt].start = kernelend; 2998 avail[avail_cnt].size = e - kernelend; 2999 avail_cnt++; 3000 } 3001 mp->size = kernelstart - s; 3002 } 3003 /* 3004 * Check whether this region overlaps the end of the kernel. 3005 */ 3006 else if (s < kernelend && e > kernelend) { 3007 mp->start = kernelend; 3008 mp->size = e - kernelend; 3009 } 3010 /* 3011 * Look whether this regions is completely inside the kernel. 3012 * Nuke it if it does. 3013 */ 3014 else if (s >= kernelstart && e <= kernelend) { 3015 mp->start = 0; 3016 mp->size = 0; 3017 } 3018 /* 3019 * If the user imposed a memory limit, enforce it. 3020 */ 3021 else if (s >= pmap_memlimit) { 3022 mp->start = -PAGE_SIZE; /* let's know why */ 3023 mp->size = 0; 3024 } 3025 else { 3026 mp->start = s; 3027 mp->size = e - s; 3028 } 3029 DPRINTFN(BOOT, 3030 ("pmap_bootstrap: a-avail[%d] start 0x%lx size 0x%lx\n", 3031 i, mp->start, mp->size)); 3032 } 3033 3034 /* 3035 * Move (and uncount) all the null return to the end. 3036 */ 3037 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 3038 if (mp->size == 0) { 3039 tmp = avail[i]; 3040 avail[i] = avail[--avail_cnt]; 3041 avail[avail_cnt] = avail[i]; 3042 } 3043 } 3044 3045 /* 3046 * (Bubble)sort them into asecnding order. 3047 */ 3048 for (i = 0; i < avail_cnt; i++) { 3049 for (j = i + 1; j < avail_cnt; j++) { 3050 if (avail[i].start > avail[j].start) { 3051 tmp = avail[i]; 3052 avail[i] = avail[j]; 3053 avail[j] = tmp; 3054 } 3055 } 3056 } 3057 3058 /* 3059 * Make sure they don't overlap. 3060 */ 3061 for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) { 3062 if (mp[0].start + mp[0].size > mp[1].start) { 3063 mp[0].size = mp[1].start - mp[0].start; 3064 } 3065 DPRINTFN(BOOT, 3066 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 3067 i, mp->start, mp->size)); 3068 } 3069 DPRINTFN(BOOT, 3070 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 3071 i, mp->start, mp->size)); 3072 3073 #ifdef PTEGCOUNT 3074 pmap_pteg_cnt = PTEGCOUNT; 3075 #else /* PTEGCOUNT */ 3076 pmap_pteg_cnt = 0x1000; 3077 3078 while (pmap_pteg_cnt < physmem) 3079 pmap_pteg_cnt <<= 1; 3080 3081 pmap_pteg_cnt >>= 1; 3082 #endif /* PTEGCOUNT */ 3083 3084 /* 3085 * Find suitably aligned memory for PTEG hash table. 3086 */ 3087 size = pmap_pteg_cnt * sizeof(struct pteg); 3088 pmap_pteg_table = pmap_boot_find_memory(size, size, 0); 3089 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3090 if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH) 3091 panic("pmap_bootstrap: pmap_pteg_table end (%p + %lx) > 256MB", 3092 pmap_pteg_table, size); 3093 #endif 3094 3095 memset(__UNVOLATILE(pmap_pteg_table), 0, 3096 pmap_pteg_cnt * sizeof(struct pteg)); 3097 pmap_pteg_mask = pmap_pteg_cnt - 1; 3098 3099 /* 3100 * We cannot do pmap_steal_memory here since UVM hasn't been loaded 3101 * with pages. So we just steal them before giving them to UVM. 3102 */ 3103 size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt; 3104 pmap_pvo_table = pmap_boot_find_memory(size, PAGE_SIZE, 0); 3105 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3106 if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH) 3107 panic("pmap_bootstrap: pmap_pvo_table end (%p + %lx) > 256MB", 3108 pmap_pvo_table, size); 3109 #endif 3110 3111 for (i = 0; i < pmap_pteg_cnt; i++) 3112 TAILQ_INIT(&pmap_pvo_table[i]); 3113 3114 #ifndef MSGBUFADDR 3115 /* 3116 * Allocate msgbuf in high memory. 3117 */ 3118 msgbuf_paddr = 3119 (paddr_t) pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1); 3120 #endif 3121 3122 #ifdef __HAVE_PMAP_PHYSSEG 3123 { 3124 u_int npgs = 0; 3125 for (i = 0, mp = avail; i < avail_cnt; i++, mp++) 3126 npgs += btoc(mp->size); 3127 size = (sizeof(struct pvo_head) + 1) * npgs; 3128 pmap_physseg.pvoh = pmap_boot_find_memory(size, PAGE_SIZE, 0); 3129 pmap_physseg.attrs = (char *) &pmap_physseg.pvoh[npgs]; 3130 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3131 if ((uintptr_t)pmap_physseg.pvoh + size > SEGMENT_LENGTH) 3132 panic("pmap_bootstrap: PVO list end (%p + %lx) > 256MB", 3133 pmap_physseg.pvoh, size); 3134 #endif 3135 } 3136 #endif 3137 3138 for (mp = avail, i = 0; i < avail_cnt; mp++, i++) { 3139 paddr_t pfstart = atop(mp->start); 3140 paddr_t pfend = atop(mp->start + mp->size); 3141 if (mp->size == 0) 3142 continue; 3143 if (mp->start + mp->size <= SEGMENT_LENGTH) { 3144 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3145 VM_FREELIST_FIRST256); 3146 } else if (mp->start >= SEGMENT_LENGTH) { 3147 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3148 VM_FREELIST_DEFAULT); 3149 } else { 3150 pfend = atop(SEGMENT_LENGTH); 3151 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3152 VM_FREELIST_FIRST256); 3153 pfstart = atop(SEGMENT_LENGTH); 3154 pfend = atop(mp->start + mp->size); 3155 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3156 VM_FREELIST_DEFAULT); 3157 } 3158 } 3159 3160 /* 3161 * Make sure kernel vsid is allocated as well as VSID 0. 3162 */ 3163 pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW] 3164 |= 1 << (KERNEL_VSIDBITS % VSID_NBPW); 3165 pmap_vsid_bitmap[0] |= 1; 3166 3167 /* 3168 * Initialize kernel pmap and hardware. 3169 */ 3170 #ifndef PPC_OEA64 3171 for (i = 0; i < 16; i++) { 3172 pmap_kernel()->pm_sr[i] = EMPTY_SEGMENT; 3173 __asm __volatile ("mtsrin %0,%1" 3174 :: "r"(EMPTY_SEGMENT), "r"(i << ADDR_SR_SHFT)); 3175 } 3176 3177 pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY; 3178 __asm __volatile ("mtsr %0,%1" 3179 :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT)); 3180 #ifdef KERNEL2_SR 3181 pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY; 3182 __asm __volatile ("mtsr %0,%1" 3183 :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT)); 3184 #endif 3185 for (i = 0; i < 16; i++) { 3186 if (iosrtable[i] & SR601_T) { 3187 pmap_kernel()->pm_sr[i] = iosrtable[i]; 3188 __asm __volatile ("mtsrin %0,%1" 3189 :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT)); 3190 } 3191 } 3192 #endif /* !PPC_OEA64 */ 3193 3194 __asm __volatile ("sync; mtsdr1 %0; isync" 3195 :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10))); 3196 tlbia(); 3197 3198 #ifdef ALTIVEC 3199 pmap_use_altivec = cpu_altivec; 3200 #endif 3201 3202 #ifdef DEBUG 3203 if (pmapdebug & PMAPDEBUG_BOOT) { 3204 u_int cnt; 3205 int bank; 3206 char pbuf[9]; 3207 for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) { 3208 cnt += vm_physmem[bank].avail_end - vm_physmem[bank].avail_start; 3209 printf("pmap_bootstrap: vm_physmem[%d]=%#lx-%#lx/%#lx\n", 3210 bank, 3211 ptoa(vm_physmem[bank].avail_start), 3212 ptoa(vm_physmem[bank].avail_end), 3213 ptoa(vm_physmem[bank].avail_end - vm_physmem[bank].avail_start)); 3214 } 3215 format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt)); 3216 printf("pmap_bootstrap: UVM memory = %s (%u pages)\n", 3217 pbuf, cnt); 3218 } 3219 #endif 3220 3221 pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry), 3222 sizeof(struct pvo_entry), 0, 0, "pmap_upvopl", 3223 &pmap_pool_uallocator); 3224 3225 pool_setlowat(&pmap_upvo_pool, 252); 3226 3227 pool_init(&pmap_pool, sizeof(struct pmap), 3228 sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator); 3229 } 3230