pmap.c revision 1.22
1 /* $NetBSD: pmap.c,v 1.22 2004/03/21 10:25:59 aymeric Exp $ */ 2 /*- 3 * Copyright (c) 2001 The NetBSD Foundation, Inc. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to The NetBSD Foundation 7 * by Matt Thomas <matt (at) 3am-software.com> of Allegro Networks, Inc. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the NetBSD 20 * Foundation, Inc. and its contributors. 21 * 4. Neither the name of The NetBSD Foundation nor the names of its 22 * contributors may be used to endorse or promote products derived 23 * from this software without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 26 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 27 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 28 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 29 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 30 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 31 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 32 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 33 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 34 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 35 * POSSIBILITY OF SUCH DAMAGE. 36 */ 37 38 /* 39 * Copyright (C) 1995, 1996 Wolfgang Solfrank. 40 * Copyright (C) 1995, 1996 TooLs GmbH. 41 * All rights reserved. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. All advertising materials mentioning features or use of this software 52 * must display the following acknowledgement: 53 * This product includes software developed by TooLs GmbH. 54 * 4. The name of TooLs GmbH may not be used to endorse or promote products 55 * derived from this software without specific prior written permission. 56 * 57 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR 58 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 59 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 60 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 61 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 62 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 63 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 64 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 65 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 66 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.22 2004/03/21 10:25:59 aymeric 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); 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 #define pmap_pvo_reclaim(pm) NULL 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 #define VSID_NBPW (sizeof(uint32_t) * 8) 259 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW]; 260 261 static int pmap_initialized; 262 263 #if defined(DEBUG) || defined(PMAPDEBUG) 264 #define PMAPDEBUG_BOOT 0x0001 265 #define PMAPDEBUG_PTE 0x0002 266 #define PMAPDEBUG_EXEC 0x0008 267 #define PMAPDEBUG_PVOENTER 0x0010 268 #define PMAPDEBUG_PVOREMOVE 0x0020 269 #define PMAPDEBUG_ACTIVATE 0x0100 270 #define PMAPDEBUG_CREATE 0x0200 271 #define PMAPDEBUG_ENTER 0x1000 272 #define PMAPDEBUG_KENTER 0x2000 273 #define PMAPDEBUG_KREMOVE 0x4000 274 #define PMAPDEBUG_REMOVE 0x8000 275 unsigned int pmapdebug = 0; 276 # define DPRINTF(x) printf x 277 # define DPRINTFN(n, x) if (pmapdebug & PMAPDEBUG_ ## n) printf x 278 #else 279 # define DPRINTF(x) 280 # define DPRINTFN(n, x) 281 #endif 282 283 284 #ifdef PMAPCOUNTERS 285 #define PMAPCOUNT(ev) ((pmap_evcnt_ ## ev).ev_count++) 286 #define PMAPCOUNT2(ev) ((ev).ev_count++) 287 288 struct evcnt pmap_evcnt_mappings = 289 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 290 "pmap", "pages mapped"); 291 struct evcnt pmap_evcnt_unmappings = 292 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 293 "pmap", "pages unmapped"); 294 295 struct evcnt pmap_evcnt_kernel_mappings = 296 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 297 "pmap", "kernel pages mapped"); 298 struct evcnt pmap_evcnt_kernel_unmappings = 299 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_kernel_mappings, 300 "pmap", "kernel pages unmapped"); 301 302 struct evcnt pmap_evcnt_mappings_replaced = 303 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 304 "pmap", "page mappings replaced"); 305 306 struct evcnt pmap_evcnt_exec_mappings = 307 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 308 "pmap", "exec pages mapped"); 309 struct evcnt pmap_evcnt_exec_cached = 310 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 311 "pmap", "exec pages cached"); 312 313 struct evcnt pmap_evcnt_exec_synced = 314 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 315 "pmap", "exec pages synced"); 316 struct evcnt pmap_evcnt_exec_synced_clear_modify = 317 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 318 "pmap", "exec pages synced (CM)"); 319 320 struct evcnt pmap_evcnt_exec_uncached_page_protect = 321 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 322 "pmap", "exec pages uncached (PP)"); 323 struct evcnt pmap_evcnt_exec_uncached_clear_modify = 324 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 325 "pmap", "exec pages uncached (CM)"); 326 struct evcnt pmap_evcnt_exec_uncached_zero_page = 327 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 328 "pmap", "exec pages uncached (ZP)"); 329 struct evcnt pmap_evcnt_exec_uncached_copy_page = 330 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 331 "pmap", "exec pages uncached (CP)"); 332 333 struct evcnt pmap_evcnt_updates = 334 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 335 "pmap", "updates"); 336 struct evcnt pmap_evcnt_collects = 337 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 338 "pmap", "collects"); 339 struct evcnt pmap_evcnt_copies = 340 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 341 "pmap", "copies"); 342 343 struct evcnt pmap_evcnt_ptes_spilled = 344 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 345 "pmap", "ptes spilled from overflow"); 346 struct evcnt pmap_evcnt_ptes_unspilled = 347 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 348 "pmap", "ptes not spilled"); 349 struct evcnt pmap_evcnt_ptes_evicted = 350 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 351 "pmap", "ptes evicted"); 352 353 struct evcnt pmap_evcnt_ptes_primary[8] = { 354 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 355 "pmap", "ptes added at primary[0]"), 356 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 357 "pmap", "ptes added at primary[1]"), 358 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 359 "pmap", "ptes added at primary[2]"), 360 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 361 "pmap", "ptes added at primary[3]"), 362 363 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 364 "pmap", "ptes added at primary[4]"), 365 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 366 "pmap", "ptes added at primary[5]"), 367 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 368 "pmap", "ptes added at primary[6]"), 369 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 370 "pmap", "ptes added at primary[7]"), 371 }; 372 struct evcnt pmap_evcnt_ptes_secondary[8] = { 373 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 374 "pmap", "ptes added at secondary[0]"), 375 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 376 "pmap", "ptes added at secondary[1]"), 377 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 378 "pmap", "ptes added at secondary[2]"), 379 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 380 "pmap", "ptes added at secondary[3]"), 381 382 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 383 "pmap", "ptes added at secondary[4]"), 384 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 385 "pmap", "ptes added at secondary[5]"), 386 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 387 "pmap", "ptes added at secondary[6]"), 388 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 389 "pmap", "ptes added at secondary[7]"), 390 }; 391 struct evcnt pmap_evcnt_ptes_removed = 392 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 393 "pmap", "ptes removed"); 394 struct evcnt pmap_evcnt_ptes_changed = 395 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 396 "pmap", "ptes changed"); 397 398 /* 399 * From pmap_subr.c 400 */ 401 extern struct evcnt pmap_evcnt_zeroed_pages; 402 extern struct evcnt pmap_evcnt_copied_pages; 403 extern struct evcnt pmap_evcnt_idlezeroed_pages; 404 #else 405 #define PMAPCOUNT(ev) ((void) 0) 406 #define PMAPCOUNT2(ev) ((void) 0) 407 #endif 408 409 #define TLBIE(va) __asm __volatile("tlbie %0" :: "r"(va)) 410 #define TLBSYNC() __asm __volatile("tlbsync") 411 #define SYNC() __asm __volatile("sync") 412 #define EIEIO() __asm __volatile("eieio") 413 #define MFMSR() mfmsr() 414 #define MTMSR(psl) mtmsr(psl) 415 #define MFPVR() mfpvr() 416 #define MFSRIN(va) mfsrin(va) 417 #define MFTB() mfrtcltbl() 418 419 #ifndef PPC_OEA64 420 static __inline register_t 421 mfsrin(vaddr_t va) 422 { 423 register_t sr; 424 __asm __volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va)); 425 return sr; 426 } 427 #endif /* PPC_OEA64 */ 428 429 static __inline register_t 430 pmap_interrupts_off(void) 431 { 432 register_t msr = MFMSR(); 433 if (msr & PSL_EE) 434 MTMSR(msr & ~PSL_EE); 435 return msr; 436 } 437 438 static void 439 pmap_interrupts_restore(register_t msr) 440 { 441 if (msr & PSL_EE) 442 MTMSR(msr); 443 } 444 445 static __inline u_int32_t 446 mfrtcltbl(void) 447 { 448 449 if ((MFPVR() >> 16) == MPC601) 450 return (mfrtcl() >> 7); 451 else 452 return (mftbl()); 453 } 454 455 /* 456 * These small routines may have to be replaced, 457 * if/when we support processors other that the 604. 458 */ 459 460 void 461 tlbia(void) 462 { 463 caddr_t i; 464 465 SYNC(); 466 /* 467 * Why not use "tlbia"? Because not all processors implement it. 468 * 469 * This needs to be a per-CPU callback to do the appropriate thing 470 * for the CPU. XXX 471 */ 472 for (i = 0; i < (caddr_t)0x00040000; i += 0x00001000) { 473 TLBIE(i); 474 EIEIO(); 475 SYNC(); 476 } 477 TLBSYNC(); 478 SYNC(); 479 } 480 481 static __inline register_t 482 va_to_vsid(const struct pmap *pm, vaddr_t addr) 483 { 484 #ifdef PPC_OEA64 485 #if 0 486 const struct ste *ste; 487 register_t hash; 488 int i; 489 490 hash = (addr >> ADDR_ESID_SHFT) & ADDR_ESID_HASH; 491 492 /* 493 * Try the primary group first 494 */ 495 ste = pm->pm_stes[hash].stes; 496 for (i = 0; i < 8; i++, ste++) { 497 if (ste->ste_hi & STE_V) && 498 (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID)) 499 return ste; 500 } 501 502 /* 503 * Then the secondary group. 504 */ 505 ste = pm->pm_stes[hash ^ ADDR_ESID_HASH].stes; 506 for (i = 0; i < 8; i++, ste++) { 507 if (ste->ste_hi & STE_V) && 508 (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID)) 509 return addr; 510 } 511 512 return NULL; 513 #else 514 /* 515 * Rather than searching the STE groups for the VSID, we know 516 * how we generate that from the ESID and so do that. 517 */ 518 return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT; 519 #endif 520 #else 521 return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID) >> SR_VSID_SHFT; 522 #endif 523 } 524 525 static __inline register_t 526 va_to_pteg(const struct pmap *pm, vaddr_t addr) 527 { 528 register_t hash; 529 530 hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT); 531 return hash & pmap_pteg_mask; 532 } 533 534 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 535 /* 536 * Given a PTE in the page table, calculate the VADDR that hashes to it. 537 * The only bit of magic is that the top 4 bits of the address doesn't 538 * technically exist in the PTE. But we know we reserved 4 bits of the 539 * VSID for it so that's how we get it. 540 */ 541 static vaddr_t 542 pmap_pte_to_va(volatile const struct pte *pt) 543 { 544 vaddr_t va; 545 uintptr_t ptaddr = (uintptr_t) pt; 546 547 if (pt->pte_hi & PTE_HID) 548 ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg)); 549 550 /* PPC Bits 10-19 PPC64 Bits 42-51 */ 551 va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff; 552 va <<= ADDR_PIDX_SHFT; 553 554 /* PPC Bits 4-9 PPC64 Bits 36-41 */ 555 va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT; 556 557 #ifdef PPC_OEA64 558 /* PPC63 Bits 0-35 */ 559 /* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */ 560 #endif 561 #ifdef PPC_OEA 562 /* PPC Bits 0-3 */ 563 va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; 564 #endif 565 566 return va; 567 } 568 #endif 569 570 static __inline struct pvo_head * 571 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p) 572 { 573 #ifdef __HAVE_VM_PAGE_MD 574 struct vm_page *pg; 575 576 pg = PHYS_TO_VM_PAGE(pa); 577 if (pg_p != NULL) 578 *pg_p = pg; 579 if (pg == NULL) 580 return &pmap_pvo_unmanaged; 581 return &pg->mdpage.mdpg_pvoh; 582 #endif 583 #ifdef __HAVE_PMAP_PHYSSEG 584 int bank, pg; 585 586 bank = vm_physseg_find(atop(pa), &pg); 587 if (pg_p != NULL) 588 *pg_p = pg; 589 if (bank == -1) 590 return &pmap_pvo_unmanaged; 591 return &vm_physmem[bank].pmseg.pvoh[pg]; 592 #endif 593 } 594 595 static __inline struct pvo_head * 596 vm_page_to_pvoh(struct vm_page *pg) 597 { 598 #ifdef __HAVE_VM_PAGE_MD 599 return &pg->mdpage.mdpg_pvoh; 600 #endif 601 #ifdef __HAVE_PMAP_PHYSSEG 602 return pa_to_pvoh(VM_PAGE_TO_PHYS(pg), NULL); 603 #endif 604 } 605 606 607 #ifdef __HAVE_PMAP_PHYSSEG 608 static __inline char * 609 pa_to_attr(paddr_t pa) 610 { 611 int bank, pg; 612 613 bank = vm_physseg_find(atop(pa), &pg); 614 if (bank == -1) 615 return NULL; 616 return &vm_physmem[bank].pmseg.attrs[pg]; 617 } 618 #endif 619 620 static __inline void 621 pmap_attr_clear(struct vm_page *pg, int ptebit) 622 { 623 #ifdef __HAVE_PMAP_PHYSSEG 624 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) &= ~(ptebit >> ATTR_SHFT); 625 #endif 626 #ifdef __HAVE_VM_PAGE_MD 627 pg->mdpage.mdpg_attrs &= ~ptebit; 628 #endif 629 } 630 631 static __inline int 632 pmap_attr_fetch(struct vm_page *pg) 633 { 634 #ifdef __HAVE_PMAP_PHYSSEG 635 return *pa_to_attr(VM_PAGE_TO_PHYS(pg)) << ATTR_SHFT; 636 #endif 637 #ifdef __HAVE_VM_PAGE_MD 638 return pg->mdpage.mdpg_attrs; 639 #endif 640 } 641 642 static __inline void 643 pmap_attr_save(struct vm_page *pg, int ptebit) 644 { 645 #ifdef __HAVE_PMAP_PHYSSEG 646 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) |= (ptebit >> ATTR_SHFT); 647 #endif 648 #ifdef __HAVE_VM_PAGE_MD 649 pg->mdpage.mdpg_attrs |= ptebit; 650 #endif 651 } 652 653 static __inline int 654 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt) 655 { 656 if (pt->pte_hi == pvo_pt->pte_hi 657 #if 0 658 && ((pt->pte_lo ^ pvo_pt->pte_lo) & 659 ~(PTE_REF|PTE_CHG)) == 0 660 #endif 661 ) 662 return 1; 663 return 0; 664 } 665 666 static __inline void 667 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo) 668 { 669 /* 670 * Construct the PTE. Default to IMB initially. Valid bit 671 * only gets set when the real pte is set in memory. 672 * 673 * Note: Don't set the valid bit for correct operation of tlb update. 674 */ 675 pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT) 676 | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API); 677 pt->pte_lo = pte_lo; 678 } 679 680 static __inline void 681 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt) 682 { 683 pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG); 684 } 685 686 static __inline void 687 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit) 688 { 689 /* 690 * As shown in Section 7.6.3.2.3 691 */ 692 pt->pte_lo &= ~ptebit; 693 TLBIE(va); 694 SYNC(); 695 EIEIO(); 696 TLBSYNC(); 697 SYNC(); 698 } 699 700 static __inline void 701 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt) 702 { 703 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 704 if (pvo_pt->pte_hi & PTE_VALID) 705 panic("pte_set: setting an already valid pte %p", pvo_pt); 706 #endif 707 pvo_pt->pte_hi |= PTE_VALID; 708 /* 709 * Update the PTE as defined in section 7.6.3.1 710 * Note that the REF/CHG bits are from pvo_pt and thus should 711 * have been saved so this routine can restore them (if desired). 712 */ 713 pt->pte_lo = pvo_pt->pte_lo; 714 EIEIO(); 715 pt->pte_hi = pvo_pt->pte_hi; 716 SYNC(); 717 pmap_pte_valid++; 718 } 719 720 static __inline void 721 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 722 { 723 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 724 if ((pvo_pt->pte_hi & PTE_VALID) == 0) 725 panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt); 726 if ((pt->pte_hi & PTE_VALID) == 0) 727 panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt); 728 #endif 729 730 pvo_pt->pte_hi &= ~PTE_VALID; 731 /* 732 * Force the ref & chg bits back into the PTEs. 733 */ 734 SYNC(); 735 /* 736 * Invalidate the pte ... (Section 7.6.3.3) 737 */ 738 pt->pte_hi &= ~PTE_VALID; 739 SYNC(); 740 TLBIE(va); 741 SYNC(); 742 EIEIO(); 743 TLBSYNC(); 744 SYNC(); 745 /* 746 * Save the ref & chg bits ... 747 */ 748 pmap_pte_synch(pt, pvo_pt); 749 pmap_pte_valid--; 750 } 751 752 static __inline void 753 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 754 { 755 /* 756 * Invalidate the PTE 757 */ 758 pmap_pte_unset(pt, pvo_pt, va); 759 pmap_pte_set(pt, pvo_pt); 760 } 761 762 /* 763 * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx 764 * (either primary or secondary location). 765 * 766 * Note: both the destination and source PTEs must not have PTE_VALID set. 767 */ 768 769 STATIC int 770 pmap_pte_insert(int ptegidx, struct pte *pvo_pt) 771 { 772 volatile struct pte *pt; 773 int i; 774 775 #if defined(DEBUG) 776 DPRINTFN(PTE, ("pmap_pte_insert: idx 0x%x, pte 0x%x 0x%x\n", 777 ptegidx, (unsigned int) pvo_pt->pte_hi, (unsigned int) pvo_pt->pte_lo)); 778 #endif 779 /* 780 * First try primary hash. 781 */ 782 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 783 if ((pt->pte_hi & PTE_VALID) == 0) { 784 pvo_pt->pte_hi &= ~PTE_HID; 785 pmap_pte_set(pt, pvo_pt); 786 return i; 787 } 788 } 789 790 /* 791 * Now try secondary hash. 792 */ 793 ptegidx ^= pmap_pteg_mask; 794 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 795 if ((pt->pte_hi & PTE_VALID) == 0) { 796 pvo_pt->pte_hi |= PTE_HID; 797 pmap_pte_set(pt, pvo_pt); 798 return i; 799 } 800 } 801 return -1; 802 } 803 804 /* 805 * Spill handler. 806 * 807 * Tries to spill a page table entry from the overflow area. 808 * This runs in either real mode (if dealing with a exception spill) 809 * or virtual mode when dealing with manually spilling one of the 810 * kernel's pte entries. In either case, interrupts are already 811 * disabled. 812 */ 813 814 int 815 pmap_pte_spill(struct pmap *pm, vaddr_t addr, boolean_t exec) 816 { 817 struct pvo_entry *source_pvo, *victim_pvo, *next_pvo; 818 struct pvo_entry *pvo; 819 /* XXX: gcc -- vpvoh is always set at either *1* or *2* */ 820 struct pvo_tqhead *pvoh, *vpvoh = NULL; 821 int ptegidx, i, j; 822 volatile struct pteg *pteg; 823 volatile struct pte *pt; 824 825 ptegidx = va_to_pteg(pm, addr); 826 827 /* 828 * Have to substitute some entry. Use the primary hash for this. 829 * Use low bits of timebase as random generator. Make sure we are 830 * not picking a kernel pte for replacement. 831 */ 832 pteg = &pmap_pteg_table[ptegidx]; 833 i = MFTB() & 7; 834 for (j = 0; j < 8; j++) { 835 pt = &pteg->pt[i]; 836 if ((pt->pte_hi & PTE_VALID) == 0 || 837 VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT) 838 != KERNEL_VSIDBITS) 839 break; 840 i = (i + 1) & 7; 841 } 842 KASSERT(j < 8); 843 844 source_pvo = NULL; 845 victim_pvo = NULL; 846 pvoh = &pmap_pvo_table[ptegidx]; 847 TAILQ_FOREACH(pvo, pvoh, pvo_olink) { 848 849 /* 850 * We need to find pvo entry for this address... 851 */ 852 PMAP_PVO_CHECK(pvo); /* sanity check */ 853 854 /* 855 * If we haven't found the source and we come to a PVO with 856 * a valid PTE, then we know we can't find it because all 857 * evicted PVOs always are first in the list. 858 */ 859 if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID)) 860 break; 861 if (source_pvo == NULL && pm == pvo->pvo_pmap && 862 addr == PVO_VADDR(pvo)) { 863 864 /* 865 * Now we have found the entry to be spilled into the 866 * pteg. Attempt to insert it into the page table. 867 */ 868 j = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 869 if (j >= 0) { 870 PVO_PTEGIDX_SET(pvo, j); 871 PMAP_PVO_CHECK(pvo); /* sanity check */ 872 PVO_WHERE(pvo, SPILL_INSERT); 873 pvo->pvo_pmap->pm_evictions--; 874 PMAPCOUNT(ptes_spilled); 875 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 876 ? pmap_evcnt_ptes_secondary 877 : pmap_evcnt_ptes_primary)[j]); 878 879 /* 880 * Since we keep the evicted entries at the 881 * from of the PVO list, we need move this 882 * (now resident) PVO after the evicted 883 * entries. 884 */ 885 next_pvo = TAILQ_NEXT(pvo, pvo_olink); 886 887 /* 888 * If we don't have to move (either we were the 889 * last entry or the next entry was valid), 890 * don't change our position. Otherwise 891 * move ourselves to the tail of the queue. 892 */ 893 if (next_pvo != NULL && 894 !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) { 895 TAILQ_REMOVE(pvoh, pvo, pvo_olink); 896 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 897 } 898 return 1; 899 } 900 source_pvo = pvo; 901 if (exec && !PVO_ISEXECUTABLE(source_pvo)) { 902 return 0; 903 } 904 if (victim_pvo != NULL) 905 break; 906 } 907 908 /* 909 * We also need the pvo entry of the victim we are replacing 910 * so save the R & C bits of the PTE. 911 */ 912 if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL && 913 pmap_pte_compare(pt, &pvo->pvo_pte)) { 914 vpvoh = pvoh; /* *1* */ 915 victim_pvo = pvo; 916 if (source_pvo != NULL) 917 break; 918 } 919 } 920 921 if (source_pvo == NULL) { 922 PMAPCOUNT(ptes_unspilled); 923 return 0; 924 } 925 926 if (victim_pvo == NULL) { 927 if ((pt->pte_hi & PTE_HID) == 0) 928 panic("pmap_pte_spill: victim p-pte (%p) has " 929 "no pvo entry!", pt); 930 931 /* 932 * If this is a secondary PTE, we need to search 933 * its primary pvo bucket for the matching PVO. 934 */ 935 vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; /* *2* */ 936 TAILQ_FOREACH(pvo, vpvoh, pvo_olink) { 937 PMAP_PVO_CHECK(pvo); /* sanity check */ 938 939 /* 940 * We also need the pvo entry of the victim we are 941 * replacing so save the R & C bits of the PTE. 942 */ 943 if (pmap_pte_compare(pt, &pvo->pvo_pte)) { 944 victim_pvo = pvo; 945 break; 946 } 947 } 948 if (victim_pvo == NULL) 949 panic("pmap_pte_spill: victim s-pte (%p) has " 950 "no pvo entry!", pt); 951 } 952 953 /* 954 * The victim should be not be a kernel PVO/PTE entry. 955 */ 956 KASSERT(victim_pvo->pvo_pmap != pmap_kernel()); 957 KASSERT(PVO_PTEGIDX_ISSET(victim_pvo)); 958 KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i); 959 960 /* 961 * We are invalidating the TLB entry for the EA for the 962 * we are replacing even though its valid; If we don't 963 * we lose any ref/chg bit changes contained in the TLB 964 * entry. 965 */ 966 source_pvo->pvo_pte.pte_hi &= ~PTE_HID; 967 968 /* 969 * To enforce the PVO list ordering constraint that all 970 * evicted entries should come before all valid entries, 971 * move the source PVO to the tail of its list and the 972 * victim PVO to the head of its list (which might not be 973 * the same list, if the victim was using the secondary hash). 974 */ 975 TAILQ_REMOVE(pvoh, source_pvo, pvo_olink); 976 TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink); 977 TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink); 978 TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink); 979 pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr); 980 pmap_pte_set(pt, &source_pvo->pvo_pte); 981 victim_pvo->pvo_pmap->pm_evictions++; 982 source_pvo->pvo_pmap->pm_evictions--; 983 PVO_WHERE(victim_pvo, SPILL_UNSET); 984 PVO_WHERE(source_pvo, SPILL_SET); 985 986 PVO_PTEGIDX_CLR(victim_pvo); 987 PVO_PTEGIDX_SET(source_pvo, i); 988 PMAPCOUNT2(pmap_evcnt_ptes_primary[i]); 989 PMAPCOUNT(ptes_spilled); 990 PMAPCOUNT(ptes_evicted); 991 PMAPCOUNT(ptes_removed); 992 993 PMAP_PVO_CHECK(victim_pvo); 994 PMAP_PVO_CHECK(source_pvo); 995 return 1; 996 } 997 998 /* 999 * Restrict given range to physical memory 1000 */ 1001 void 1002 pmap_real_memory(paddr_t *start, psize_t *size) 1003 { 1004 struct mem_region *mp; 1005 1006 for (mp = mem; mp->size; mp++) { 1007 if (*start + *size > mp->start 1008 && *start < mp->start + mp->size) { 1009 if (*start < mp->start) { 1010 *size -= mp->start - *start; 1011 *start = mp->start; 1012 } 1013 if (*start + *size > mp->start + mp->size) 1014 *size = mp->start + mp->size - *start; 1015 return; 1016 } 1017 } 1018 *size = 0; 1019 } 1020 1021 /* 1022 * Initialize anything else for pmap handling. 1023 * Called during vm_init(). 1024 */ 1025 void 1026 pmap_init(void) 1027 { 1028 #ifdef __HAVE_PMAP_PHYSSEG 1029 struct pvo_tqhead *pvoh; 1030 int bank; 1031 long sz; 1032 char *attr; 1033 1034 pvoh = pmap_physseg.pvoh; 1035 attr = pmap_physseg.attrs; 1036 for (bank = 0; bank < vm_nphysseg; bank++) { 1037 sz = vm_physmem[bank].end - vm_physmem[bank].start; 1038 vm_physmem[bank].pmseg.pvoh = pvoh; 1039 vm_physmem[bank].pmseg.attrs = attr; 1040 for (; sz > 0; sz--, pvoh++, attr++) { 1041 TAILQ_INIT(pvoh); 1042 *attr = 0; 1043 } 1044 } 1045 #endif 1046 1047 pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry), 1048 sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl", 1049 &pmap_pool_mallocator); 1050 1051 pool_setlowat(&pmap_mpvo_pool, 1008); 1052 1053 pmap_initialized = 1; 1054 1055 #ifdef PMAPCOUNTERS 1056 evcnt_attach_static(&pmap_evcnt_mappings); 1057 evcnt_attach_static(&pmap_evcnt_mappings_replaced); 1058 evcnt_attach_static(&pmap_evcnt_unmappings); 1059 1060 evcnt_attach_static(&pmap_evcnt_kernel_mappings); 1061 evcnt_attach_static(&pmap_evcnt_kernel_unmappings); 1062 1063 evcnt_attach_static(&pmap_evcnt_exec_mappings); 1064 evcnt_attach_static(&pmap_evcnt_exec_cached); 1065 evcnt_attach_static(&pmap_evcnt_exec_synced); 1066 evcnt_attach_static(&pmap_evcnt_exec_synced_clear_modify); 1067 1068 evcnt_attach_static(&pmap_evcnt_exec_uncached_page_protect); 1069 evcnt_attach_static(&pmap_evcnt_exec_uncached_clear_modify); 1070 evcnt_attach_static(&pmap_evcnt_exec_uncached_zero_page); 1071 evcnt_attach_static(&pmap_evcnt_exec_uncached_copy_page); 1072 1073 evcnt_attach_static(&pmap_evcnt_zeroed_pages); 1074 evcnt_attach_static(&pmap_evcnt_copied_pages); 1075 evcnt_attach_static(&pmap_evcnt_idlezeroed_pages); 1076 1077 evcnt_attach_static(&pmap_evcnt_updates); 1078 evcnt_attach_static(&pmap_evcnt_collects); 1079 evcnt_attach_static(&pmap_evcnt_copies); 1080 1081 evcnt_attach_static(&pmap_evcnt_ptes_spilled); 1082 evcnt_attach_static(&pmap_evcnt_ptes_unspilled); 1083 evcnt_attach_static(&pmap_evcnt_ptes_evicted); 1084 evcnt_attach_static(&pmap_evcnt_ptes_removed); 1085 evcnt_attach_static(&pmap_evcnt_ptes_changed); 1086 evcnt_attach_static(&pmap_evcnt_ptes_primary[0]); 1087 evcnt_attach_static(&pmap_evcnt_ptes_primary[1]); 1088 evcnt_attach_static(&pmap_evcnt_ptes_primary[2]); 1089 evcnt_attach_static(&pmap_evcnt_ptes_primary[3]); 1090 evcnt_attach_static(&pmap_evcnt_ptes_primary[4]); 1091 evcnt_attach_static(&pmap_evcnt_ptes_primary[5]); 1092 evcnt_attach_static(&pmap_evcnt_ptes_primary[6]); 1093 evcnt_attach_static(&pmap_evcnt_ptes_primary[7]); 1094 evcnt_attach_static(&pmap_evcnt_ptes_secondary[0]); 1095 evcnt_attach_static(&pmap_evcnt_ptes_secondary[1]); 1096 evcnt_attach_static(&pmap_evcnt_ptes_secondary[2]); 1097 evcnt_attach_static(&pmap_evcnt_ptes_secondary[3]); 1098 evcnt_attach_static(&pmap_evcnt_ptes_secondary[4]); 1099 evcnt_attach_static(&pmap_evcnt_ptes_secondary[5]); 1100 evcnt_attach_static(&pmap_evcnt_ptes_secondary[6]); 1101 evcnt_attach_static(&pmap_evcnt_ptes_secondary[7]); 1102 #endif 1103 } 1104 1105 /* 1106 * How much virtual space does the kernel get? 1107 */ 1108 void 1109 pmap_virtual_space(vaddr_t *start, vaddr_t *end) 1110 { 1111 /* 1112 * For now, reserve one segment (minus some overhead) for kernel 1113 * virtual memory 1114 */ 1115 *start = VM_MIN_KERNEL_ADDRESS; 1116 *end = VM_MAX_KERNEL_ADDRESS; 1117 } 1118 1119 /* 1120 * Allocate, initialize, and return a new physical map. 1121 */ 1122 pmap_t 1123 pmap_create(void) 1124 { 1125 pmap_t pm; 1126 1127 pm = pool_get(&pmap_pool, PR_WAITOK); 1128 memset((caddr_t)pm, 0, sizeof *pm); 1129 pmap_pinit(pm); 1130 1131 DPRINTFN(CREATE,("pmap_create: pm %p:\n" 1132 "\t%06x %06x %06x %06x %06x %06x %06x %06x\n" 1133 "\t%06x %06x %06x %06x %06x %06x %06x %06x\n", pm, 1134 (unsigned int) pm->pm_sr[0], (unsigned int) pm->pm_sr[1], 1135 (unsigned int) pm->pm_sr[2], (unsigned int) pm->pm_sr[3], 1136 (unsigned int) pm->pm_sr[4], (unsigned int) pm->pm_sr[5], 1137 (unsigned int) pm->pm_sr[6], (unsigned int) pm->pm_sr[7], 1138 (unsigned int) pm->pm_sr[8], (unsigned int) pm->pm_sr[9], 1139 (unsigned int) pm->pm_sr[10], (unsigned int) pm->pm_sr[11], 1140 (unsigned int) pm->pm_sr[12], (unsigned int) pm->pm_sr[13], 1141 (unsigned int) pm->pm_sr[14], (unsigned int) pm->pm_sr[15])); 1142 return pm; 1143 } 1144 1145 /* 1146 * Initialize a preallocated and zeroed pmap structure. 1147 */ 1148 void 1149 pmap_pinit(pmap_t pm) 1150 { 1151 register_t entropy = MFTB(); 1152 register_t mask; 1153 int i; 1154 1155 /* 1156 * Allocate some segment registers for this pmap. 1157 */ 1158 pm->pm_refs = 1; 1159 for (i = 0; i < NPMAPS; i += VSID_NBPW) { 1160 static register_t pmap_vsidcontext; 1161 register_t hash; 1162 unsigned int n; 1163 1164 /* Create a new value by multiplying by a prime adding in 1165 * entropy from the timebase register. This is to make the 1166 * VSID more random so that the PT Hash function collides 1167 * less often. (note that the prime causes gcc to do shifts 1168 * instead of a multiply) 1169 */ 1170 pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy; 1171 hash = pmap_vsidcontext & (NPMAPS - 1); 1172 if (hash == 0) /* 0 is special, avoid it */ 1173 continue; 1174 n = hash >> 5; 1175 mask = 1L << (hash & (VSID_NBPW-1)); 1176 hash = pmap_vsidcontext; 1177 if (pmap_vsid_bitmap[n] & mask) { /* collision? */ 1178 /* anything free in this bucket? */ 1179 if (~pmap_vsid_bitmap[n] == 0) { 1180 entropy = hash >> PTE_VSID_SHFT; 1181 continue; 1182 } 1183 i = ffs(~pmap_vsid_bitmap[n]) - 1; 1184 mask = 1L << i; 1185 hash &= ~(VSID_NBPW-1); 1186 hash |= i; 1187 } 1188 hash &= PTE_VSID >> PTE_VSID_SHFT; 1189 pmap_vsid_bitmap[n] |= mask; 1190 pm->pm_vsid = hash; 1191 #ifndef PPC_OEA64 1192 for (i = 0; i < 16; i++) 1193 pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY | 1194 SR_NOEXEC; 1195 #endif 1196 return; 1197 } 1198 panic("pmap_pinit: out of segments"); 1199 } 1200 1201 /* 1202 * Add a reference to the given pmap. 1203 */ 1204 void 1205 pmap_reference(pmap_t pm) 1206 { 1207 pm->pm_refs++; 1208 } 1209 1210 /* 1211 * Retire the given pmap from service. 1212 * Should only be called if the map contains no valid mappings. 1213 */ 1214 void 1215 pmap_destroy(pmap_t pm) 1216 { 1217 if (--pm->pm_refs == 0) { 1218 pmap_release(pm); 1219 pool_put(&pmap_pool, pm); 1220 } 1221 } 1222 1223 /* 1224 * Release any resources held by the given physical map. 1225 * Called when a pmap initialized by pmap_pinit is being released. 1226 */ 1227 void 1228 pmap_release(pmap_t pm) 1229 { 1230 int idx, mask; 1231 1232 if (pm->pm_sr[0] == 0) 1233 panic("pmap_release"); 1234 idx = pm->pm_vsid & (NPMAPS-1); 1235 mask = 1 << (idx % VSID_NBPW); 1236 idx /= VSID_NBPW; 1237 1238 KASSERT(pmap_vsid_bitmap[idx] & mask); 1239 pmap_vsid_bitmap[idx] &= ~mask; 1240 } 1241 1242 /* 1243 * Copy the range specified by src_addr/len 1244 * from the source map to the range dst_addr/len 1245 * in the destination map. 1246 * 1247 * This routine is only advisory and need not do anything. 1248 */ 1249 void 1250 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr, 1251 vsize_t len, vaddr_t src_addr) 1252 { 1253 PMAPCOUNT(copies); 1254 } 1255 1256 /* 1257 * Require that all active physical maps contain no 1258 * incorrect entries NOW. 1259 */ 1260 void 1261 pmap_update(struct pmap *pmap) 1262 { 1263 PMAPCOUNT(updates); 1264 TLBSYNC(); 1265 } 1266 1267 /* 1268 * Garbage collects the physical map system for 1269 * pages which are no longer used. 1270 * Success need not be guaranteed -- that is, there 1271 * may well be pages which are not referenced, but 1272 * others may be collected. 1273 * Called by the pageout daemon when pages are scarce. 1274 */ 1275 void 1276 pmap_collect(pmap_t pm) 1277 { 1278 PMAPCOUNT(collects); 1279 } 1280 1281 static __inline int 1282 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx) 1283 { 1284 int pteidx; 1285 /* 1286 * We can find the actual pte entry without searching by 1287 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9] 1288 * and by noticing the HID bit. 1289 */ 1290 pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo); 1291 if (pvo->pvo_pte.pte_hi & PTE_HID) 1292 pteidx ^= pmap_pteg_mask * 8; 1293 return pteidx; 1294 } 1295 1296 volatile struct pte * 1297 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx) 1298 { 1299 volatile struct pte *pt; 1300 1301 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1302 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) 1303 return NULL; 1304 #endif 1305 1306 /* 1307 * If we haven't been supplied the ptegidx, calculate it. 1308 */ 1309 if (pteidx == -1) { 1310 int ptegidx; 1311 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1312 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1313 } 1314 1315 pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7]; 1316 1317 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1318 return pt; 1319 #else 1320 if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) { 1321 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1322 "pvo but no valid pte index", pvo); 1323 } 1324 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) { 1325 panic("pmap_pvo_to_pte: pvo %p: has valid pte index in " 1326 "pvo but no valid pte", pvo); 1327 } 1328 1329 if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) { 1330 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) { 1331 #if defined(DEBUG) || defined(PMAPCHECK) 1332 pmap_pte_print(pt); 1333 #endif 1334 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1335 "pmap_pteg_table %p but invalid in pvo", 1336 pvo, pt); 1337 } 1338 if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) { 1339 #if defined(DEBUG) || defined(PMAPCHECK) 1340 pmap_pte_print(pt); 1341 #endif 1342 panic("pmap_pvo_to_pte: pvo %p: pvo pte does " 1343 "not match pte %p in pmap_pteg_table", 1344 pvo, pt); 1345 } 1346 return pt; 1347 } 1348 1349 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1350 #if defined(DEBUG) || defined(PMAPCHECK) 1351 pmap_pte_print(pt); 1352 #endif 1353 panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in " 1354 "pmap_pteg_table but valid in pvo", pvo, pt); 1355 } 1356 return NULL; 1357 #endif /* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */ 1358 } 1359 1360 struct pvo_entry * 1361 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p) 1362 { 1363 struct pvo_entry *pvo; 1364 int ptegidx; 1365 1366 va &= ~ADDR_POFF; 1367 ptegidx = va_to_pteg(pm, va); 1368 1369 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1370 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1371 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 1372 panic("pmap_pvo_find_va: invalid pvo %p on " 1373 "list %#x (%p)", pvo, ptegidx, 1374 &pmap_pvo_table[ptegidx]); 1375 #endif 1376 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1377 if (pteidx_p) 1378 *pteidx_p = pmap_pvo_pte_index(pvo, ptegidx); 1379 return pvo; 1380 } 1381 } 1382 return NULL; 1383 } 1384 1385 #if defined(DEBUG) || defined(PMAPCHECK) 1386 void 1387 pmap_pvo_check(const struct pvo_entry *pvo) 1388 { 1389 struct pvo_head *pvo_head; 1390 struct pvo_entry *pvo0; 1391 volatile struct pte *pt; 1392 int failed = 0; 1393 1394 if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH) 1395 panic("pmap_pvo_check: pvo %p: invalid address", pvo); 1396 1397 if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) { 1398 printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n", 1399 pvo, pvo->pvo_pmap); 1400 failed = 1; 1401 } 1402 1403 if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH || 1404 (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) { 1405 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1406 pvo, TAILQ_NEXT(pvo, pvo_olink)); 1407 failed = 1; 1408 } 1409 1410 if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH || 1411 (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) { 1412 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1413 pvo, LIST_NEXT(pvo, pvo_vlink)); 1414 failed = 1; 1415 } 1416 1417 if (pvo->pvo_vaddr & PVO_MANAGED) { 1418 pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL); 1419 } else { 1420 if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) { 1421 printf("pmap_pvo_check: pvo %p: non kernel address " 1422 "on kernel unmanaged list\n", pvo); 1423 failed = 1; 1424 } 1425 pvo_head = &pmap_pvo_kunmanaged; 1426 } 1427 LIST_FOREACH(pvo0, pvo_head, pvo_vlink) { 1428 if (pvo0 == pvo) 1429 break; 1430 } 1431 if (pvo0 == NULL) { 1432 printf("pmap_pvo_check: pvo %p: not present " 1433 "on its vlist head %p\n", pvo, pvo_head); 1434 failed = 1; 1435 } 1436 if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) { 1437 printf("pmap_pvo_check: pvo %p: not present " 1438 "on its olist head\n", pvo); 1439 failed = 1; 1440 } 1441 pt = pmap_pvo_to_pte(pvo, -1); 1442 if (pt == NULL) { 1443 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1444 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1445 "no PTE\n", pvo); 1446 failed = 1; 1447 } 1448 } else { 1449 if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] || 1450 (uintptr_t) pt >= 1451 (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) { 1452 printf("pmap_pvo_check: pvo %p: pte %p not in " 1453 "pteg table\n", pvo, pt); 1454 failed = 1; 1455 } 1456 if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) { 1457 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1458 "no PTE\n", pvo); 1459 failed = 1; 1460 } 1461 if (pvo->pvo_pte.pte_hi != pt->pte_hi) { 1462 printf("pmap_pvo_check: pvo %p: pte_hi differ: " 1463 "%#x/%#x\n", pvo, (unsigned int) pvo->pvo_pte.pte_hi, (unsigned int) pt->pte_hi); 1464 failed = 1; 1465 } 1466 if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) & 1467 (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) { 1468 printf("pmap_pvo_check: pvo %p: pte_lo differ: " 1469 "%#x/%#x\n", pvo, 1470 (unsigned int) (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)), 1471 (unsigned int) (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN))); 1472 failed = 1; 1473 } 1474 if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) { 1475 printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#lx" 1476 " doesn't not match PVO's VA %#lx\n", 1477 pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo)); 1478 failed = 1; 1479 } 1480 if (failed) 1481 pmap_pte_print(pt); 1482 } 1483 if (failed) 1484 panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo, 1485 pvo->pvo_pmap); 1486 } 1487 #endif /* DEBUG || PMAPCHECK */ 1488 1489 /* 1490 * This returns whether this is the first mapping of a page. 1491 */ 1492 int 1493 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head, 1494 vaddr_t va, paddr_t pa, register_t pte_lo, int flags) 1495 { 1496 struct pvo_entry *pvo; 1497 struct pvo_tqhead *pvoh; 1498 register_t msr; 1499 int ptegidx; 1500 int i; 1501 int poolflags = PR_NOWAIT; 1502 1503 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1504 if (pmap_pvo_remove_depth > 0) 1505 panic("pmap_pvo_enter: called while pmap_pvo_remove active!"); 1506 if (++pmap_pvo_enter_depth > 1) 1507 panic("pmap_pvo_enter: called recursively!"); 1508 #endif 1509 1510 /* 1511 * Compute the PTE Group index. 1512 */ 1513 va &= ~ADDR_POFF; 1514 ptegidx = va_to_pteg(pm, va); 1515 1516 msr = pmap_interrupts_off(); 1517 /* 1518 * Remove any existing mapping for this page. Reuse the 1519 * pvo entry if there a mapping. 1520 */ 1521 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1522 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1523 #ifdef DEBUG 1524 if ((pmapdebug & PMAPDEBUG_PVOENTER) && 1525 ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) & 1526 ~(PTE_REF|PTE_CHG)) == 0 && 1527 va < VM_MIN_KERNEL_ADDRESS) { 1528 printf("pmap_pvo_enter: pvo %p: dup %#x/%#lx\n", 1529 pvo, (unsigned int) pvo->pvo_pte.pte_lo, (unsigned int) pte_lo|pa); 1530 printf("pmap_pvo_enter: pte_hi=%#x sr=%#x\n", 1531 (unsigned int) pvo->pvo_pte.pte_hi, 1532 (unsigned int) pm->pm_sr[va >> ADDR_SR_SHFT]); 1533 pmap_pte_print(pmap_pvo_to_pte(pvo, -1)); 1534 #ifdef DDBX 1535 Debugger(); 1536 #endif 1537 } 1538 #endif 1539 PMAPCOUNT(mappings_replaced); 1540 pmap_pvo_remove(pvo, -1); 1541 break; 1542 } 1543 } 1544 1545 /* 1546 * If we aren't overwriting an mapping, try to allocate 1547 */ 1548 pmap_interrupts_restore(msr); 1549 pvo = pool_get(pl, poolflags); 1550 msr = pmap_interrupts_off(); 1551 if (pvo == NULL) { 1552 pvo = pmap_pvo_reclaim(pm); 1553 if (pvo == NULL) { 1554 if ((flags & PMAP_CANFAIL) == 0) 1555 panic("pmap_pvo_enter: failed"); 1556 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1557 pmap_pvo_enter_depth--; 1558 #endif 1559 pmap_interrupts_restore(msr); 1560 return ENOMEM; 1561 } 1562 } 1563 pvo->pvo_vaddr = va; 1564 pvo->pvo_pmap = pm; 1565 pvo->pvo_vaddr &= ~ADDR_POFF; 1566 if (flags & VM_PROT_EXECUTE) { 1567 PMAPCOUNT(exec_mappings); 1568 pvo_set_exec(pvo); 1569 } 1570 if (flags & PMAP_WIRED) 1571 pvo->pvo_vaddr |= PVO_WIRED; 1572 if (pvo_head != &pmap_pvo_kunmanaged) { 1573 pvo->pvo_vaddr |= PVO_MANAGED; 1574 PMAPCOUNT(mappings); 1575 } else { 1576 PMAPCOUNT(kernel_mappings); 1577 } 1578 pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo); 1579 1580 LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink); 1581 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1582 pvo->pvo_pmap->pm_stats.wired_count++; 1583 pvo->pvo_pmap->pm_stats.resident_count++; 1584 #if defined(DEBUG) 1585 if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) 1586 DPRINTFN(PVOENTER, 1587 ("pmap_pvo_enter: pvo %p: pm %p va %#lx pa %#lx\n", 1588 pvo, pm, va, pa)); 1589 #endif 1590 1591 /* 1592 * We hope this succeeds but it isn't required. 1593 */ 1594 pvoh = &pmap_pvo_table[ptegidx]; 1595 i = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 1596 if (i >= 0) { 1597 PVO_PTEGIDX_SET(pvo, i); 1598 PVO_WHERE(pvo, ENTER_INSERT); 1599 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 1600 ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]); 1601 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 1602 } else { 1603 /* 1604 * Since we didn't have room for this entry (which makes it 1605 * and evicted entry), place it at the head of the list. 1606 */ 1607 TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink); 1608 PMAPCOUNT(ptes_evicted); 1609 pm->pm_evictions++; 1610 /* 1611 * If this is a kernel page, make sure it's active. 1612 */ 1613 if (pm == pmap_kernel()) { 1614 i = pmap_pte_spill(pm, va, FALSE); 1615 KASSERT(i); 1616 } 1617 } 1618 PMAP_PVO_CHECK(pvo); /* sanity check */ 1619 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1620 pmap_pvo_enter_depth--; 1621 #endif 1622 pmap_interrupts_restore(msr); 1623 return 0; 1624 } 1625 1626 void 1627 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx) 1628 { 1629 volatile struct pte *pt; 1630 int ptegidx; 1631 1632 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1633 if (++pmap_pvo_remove_depth > 1) 1634 panic("pmap_pvo_remove: called recursively!"); 1635 #endif 1636 1637 /* 1638 * If we haven't been supplied the ptegidx, calculate it. 1639 */ 1640 if (pteidx == -1) { 1641 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1642 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1643 } else { 1644 ptegidx = pteidx >> 3; 1645 if (pvo->pvo_pte.pte_hi & PTE_HID) 1646 ptegidx ^= pmap_pteg_mask; 1647 } 1648 PMAP_PVO_CHECK(pvo); /* sanity check */ 1649 1650 /* 1651 * If there is an active pte entry, we need to deactivate it 1652 * (and save the ref & chg bits). 1653 */ 1654 pt = pmap_pvo_to_pte(pvo, pteidx); 1655 if (pt != NULL) { 1656 pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 1657 PVO_WHERE(pvo, REMOVE); 1658 PVO_PTEGIDX_CLR(pvo); 1659 PMAPCOUNT(ptes_removed); 1660 } else { 1661 KASSERT(pvo->pvo_pmap->pm_evictions > 0); 1662 pvo->pvo_pmap->pm_evictions--; 1663 } 1664 1665 /* 1666 * Account for executable mappings. 1667 */ 1668 if (PVO_ISEXECUTABLE(pvo)) 1669 pvo_clear_exec(pvo); 1670 1671 /* 1672 * Update our statistics. 1673 */ 1674 pvo->pvo_pmap->pm_stats.resident_count--; 1675 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1676 pvo->pvo_pmap->pm_stats.wired_count--; 1677 1678 /* 1679 * Save the REF/CHG bits into their cache if the page is managed. 1680 */ 1681 if (pvo->pvo_vaddr & PVO_MANAGED) { 1682 register_t ptelo = pvo->pvo_pte.pte_lo; 1683 struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN); 1684 1685 if (pg != NULL) { 1686 pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG)); 1687 } 1688 PMAPCOUNT(unmappings); 1689 } else { 1690 PMAPCOUNT(kernel_unmappings); 1691 } 1692 1693 /* 1694 * Remove the PVO from its lists and return it to the pool. 1695 */ 1696 LIST_REMOVE(pvo, pvo_vlink); 1697 TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink); 1698 pool_put(pvo->pvo_vaddr & PVO_MANAGED 1699 ? &pmap_mpvo_pool : &pmap_upvo_pool, pvo); 1700 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1701 pmap_pvo_remove_depth--; 1702 #endif 1703 } 1704 1705 /* 1706 * Mark a mapping as executable. 1707 * If this is the first executable mapping in the segment, 1708 * clear the noexec flag. 1709 */ 1710 STATIC void 1711 pvo_set_exec(struct pvo_entry *pvo) 1712 { 1713 struct pmap *pm = pvo->pvo_pmap; 1714 1715 if (pm == pmap_kernel() || PVO_ISEXECUTABLE(pvo)) { 1716 return; 1717 } 1718 pvo->pvo_vaddr |= PVO_EXECUTABLE; 1719 #ifdef PPC_OEA 1720 { 1721 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1722 if (pm->pm_exec[sr]++ == 0) { 1723 pm->pm_sr[sr] &= ~SR_NOEXEC; 1724 } 1725 } 1726 #endif 1727 } 1728 1729 /* 1730 * Mark a mapping as non-executable. 1731 * If this was the last executable mapping in the segment, 1732 * set the noexec flag. 1733 */ 1734 STATIC void 1735 pvo_clear_exec(struct pvo_entry *pvo) 1736 { 1737 struct pmap *pm = pvo->pvo_pmap; 1738 1739 if (pm == pmap_kernel() || !PVO_ISEXECUTABLE(pvo)) { 1740 return; 1741 } 1742 pvo->pvo_vaddr &= ~PVO_EXECUTABLE; 1743 #ifdef PPC_OEA 1744 { 1745 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1746 if (--pm->pm_exec[sr] == 0) { 1747 pm->pm_sr[sr] |= SR_NOEXEC; 1748 } 1749 } 1750 #endif 1751 } 1752 1753 /* 1754 * Insert physical page at pa into the given pmap at virtual address va. 1755 */ 1756 int 1757 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags) 1758 { 1759 struct mem_region *mp; 1760 struct pvo_head *pvo_head; 1761 struct vm_page *pg; 1762 struct pool *pl; 1763 register_t pte_lo; 1764 int error; 1765 u_int pvo_flags; 1766 u_int was_exec = 0; 1767 1768 if (__predict_false(!pmap_initialized)) { 1769 pvo_head = &pmap_pvo_kunmanaged; 1770 pl = &pmap_upvo_pool; 1771 pvo_flags = 0; 1772 pg = NULL; 1773 was_exec = PTE_EXEC; 1774 } else { 1775 pvo_head = pa_to_pvoh(pa, &pg); 1776 pl = &pmap_mpvo_pool; 1777 pvo_flags = PVO_MANAGED; 1778 } 1779 1780 DPRINTFN(ENTER, 1781 ("pmap_enter(%p, 0x%lx, 0x%lx, 0x%x, 0x%x):", 1782 pm, va, pa, prot, flags)); 1783 1784 /* 1785 * If this is a managed page, and it's the first reference to the 1786 * page clear the execness of the page. Otherwise fetch the execness. 1787 */ 1788 if (pg != NULL) 1789 was_exec = pmap_attr_fetch(pg) & PTE_EXEC; 1790 1791 DPRINTFN(ENTER, (" was_exec=%d", was_exec)); 1792 1793 /* 1794 * Assume the page is cache inhibited and access is guarded unless 1795 * it's in our available memory array. If it is in the memory array, 1796 * asssume it's in memory coherent memory. 1797 */ 1798 pte_lo = PTE_IG; 1799 if ((flags & PMAP_NC) == 0) { 1800 for (mp = mem; mp->size; mp++) { 1801 if (pa >= mp->start && pa < mp->start + mp->size) { 1802 pte_lo = PTE_M; 1803 break; 1804 } 1805 } 1806 } 1807 1808 if (prot & VM_PROT_WRITE) 1809 pte_lo |= PTE_BW; 1810 else 1811 pte_lo |= PTE_BR; 1812 1813 /* 1814 * If this was in response to a fault, "pre-fault" the PTE's 1815 * changed/referenced bit appropriately. 1816 */ 1817 if (flags & VM_PROT_WRITE) 1818 pte_lo |= PTE_CHG; 1819 if (flags & (VM_PROT_READ|VM_PROT_WRITE)) 1820 pte_lo |= PTE_REF; 1821 1822 /* 1823 * We need to know if this page can be executable 1824 */ 1825 flags |= (prot & VM_PROT_EXECUTE); 1826 1827 /* 1828 * Record mapping for later back-translation and pte spilling. 1829 * This will overwrite any existing mapping. 1830 */ 1831 error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags); 1832 1833 /* 1834 * Flush the real page from the instruction cache if this page is 1835 * mapped executable and cacheable and has not been flushed since 1836 * the last time it was modified. 1837 */ 1838 if (error == 0 && 1839 (flags & VM_PROT_EXECUTE) && 1840 (pte_lo & PTE_I) == 0 && 1841 was_exec == 0) { 1842 DPRINTFN(ENTER, (" syncicache")); 1843 PMAPCOUNT(exec_synced); 1844 pmap_syncicache(pa, PAGE_SIZE); 1845 if (pg != NULL) { 1846 pmap_attr_save(pg, PTE_EXEC); 1847 PMAPCOUNT(exec_cached); 1848 #if defined(DEBUG) || defined(PMAPDEBUG) 1849 if (pmapdebug & PMAPDEBUG_ENTER) 1850 printf(" marked-as-exec"); 1851 else if (pmapdebug & PMAPDEBUG_EXEC) 1852 printf("[pmap_enter: %#lx: marked-as-exec]\n", 1853 pg->phys_addr); 1854 1855 #endif 1856 } 1857 } 1858 1859 DPRINTFN(ENTER, (": error=%d\n", error)); 1860 1861 return error; 1862 } 1863 1864 void 1865 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot) 1866 { 1867 struct mem_region *mp; 1868 register_t pte_lo; 1869 int error; 1870 1871 if (va < VM_MIN_KERNEL_ADDRESS) 1872 panic("pmap_kenter_pa: attempt to enter " 1873 "non-kernel address %#lx!", va); 1874 1875 DPRINTFN(KENTER, 1876 ("pmap_kenter_pa(%#lx,%#lx,%#x)\n", va, pa, prot)); 1877 1878 /* 1879 * Assume the page is cache inhibited and access is guarded unless 1880 * it's in our available memory array. If it is in the memory array, 1881 * asssume it's in memory coherent memory. 1882 */ 1883 pte_lo = PTE_IG; 1884 if ((prot & PMAP_NC) == 0) { 1885 for (mp = mem; mp->size; mp++) { 1886 if (pa >= mp->start && pa < mp->start + mp->size) { 1887 pte_lo = PTE_M; 1888 break; 1889 } 1890 } 1891 } 1892 1893 if (prot & VM_PROT_WRITE) 1894 pte_lo |= PTE_BW; 1895 else 1896 pte_lo |= PTE_BR; 1897 1898 /* 1899 * We don't care about REF/CHG on PVOs on the unmanaged list. 1900 */ 1901 error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool, 1902 &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED); 1903 1904 if (error != 0) 1905 panic("pmap_kenter_pa: failed to enter va %#lx pa %#lx: %d", 1906 va, pa, error); 1907 } 1908 1909 void 1910 pmap_kremove(vaddr_t va, vsize_t len) 1911 { 1912 if (va < VM_MIN_KERNEL_ADDRESS) 1913 panic("pmap_kremove: attempt to remove " 1914 "non-kernel address %#lx!", va); 1915 1916 DPRINTFN(KREMOVE,("pmap_kremove(%#lx,%#lx)\n", va, len)); 1917 pmap_remove(pmap_kernel(), va, va + len); 1918 } 1919 1920 /* 1921 * Remove the given range of mapping entries. 1922 */ 1923 void 1924 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva) 1925 { 1926 struct pvo_entry *pvo; 1927 register_t msr; 1928 int pteidx; 1929 1930 msr = pmap_interrupts_off(); 1931 for (; va < endva; va += PAGE_SIZE) { 1932 pvo = pmap_pvo_find_va(pm, va, &pteidx); 1933 if (pvo != NULL) { 1934 pmap_pvo_remove(pvo, pteidx); 1935 } 1936 } 1937 pmap_interrupts_restore(msr); 1938 } 1939 1940 /* 1941 * Get the physical page address for the given pmap/virtual address. 1942 */ 1943 boolean_t 1944 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap) 1945 { 1946 struct pvo_entry *pvo; 1947 register_t msr; 1948 1949 /* 1950 * If this is a kernel pmap lookup, also check the battable 1951 * and if we get a hit, translate the VA to a PA using the 1952 * BAT entries. Don't check for VM_MAX_KENREL_ADDRESS is 1953 * that will wrap back to 0. 1954 */ 1955 if (pm == pmap_kernel() && 1956 (va < VM_MIN_KERNEL_ADDRESS || 1957 (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) { 1958 register_t batu = battable[va >> ADDR_SR_SHFT].batu; 1959 KASSERT((va >> ADDR_SR_SHFT) != USER_SR); 1960 if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) { 1961 register_t batl = battable[va >> ADDR_SR_SHFT].batl; 1962 register_t mask = (~(batu & BAT_BL) << 15) & ~0x1ffffL; 1963 *pap = (batl & mask) | (va & ~mask); 1964 return TRUE; 1965 } 1966 return FALSE; 1967 } 1968 1969 msr = pmap_interrupts_off(); 1970 pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL); 1971 if (pvo != NULL) { 1972 PMAP_PVO_CHECK(pvo); /* sanity check */ 1973 *pap = (pvo->pvo_pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF); 1974 } 1975 pmap_interrupts_restore(msr); 1976 return pvo != NULL; 1977 } 1978 1979 /* 1980 * Lower the protection on the specified range of this pmap. 1981 */ 1982 void 1983 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot) 1984 { 1985 struct pvo_entry *pvo; 1986 volatile struct pte *pt; 1987 register_t msr; 1988 int pteidx; 1989 1990 /* 1991 * Since this routine only downgrades protection, we should 1992 * always be called with at least one bit not set. 1993 */ 1994 KASSERT(prot != VM_PROT_ALL); 1995 1996 /* 1997 * If there is no protection, this is equivalent to 1998 * remove the pmap from the pmap. 1999 */ 2000 if ((prot & VM_PROT_READ) == 0) { 2001 pmap_remove(pm, va, endva); 2002 return; 2003 } 2004 2005 msr = pmap_interrupts_off(); 2006 for (; va < endva; va += PAGE_SIZE) { 2007 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2008 if (pvo == NULL) 2009 continue; 2010 PMAP_PVO_CHECK(pvo); /* sanity check */ 2011 2012 /* 2013 * Revoke executable if asked to do so. 2014 */ 2015 if ((prot & VM_PROT_EXECUTE) == 0) 2016 pvo_clear_exec(pvo); 2017 2018 #if 0 2019 /* 2020 * If the page is already read-only, no change 2021 * needs to be made. 2022 */ 2023 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) 2024 continue; 2025 #endif 2026 /* 2027 * Grab the PTE pointer before we diddle with 2028 * the cached PTE copy. 2029 */ 2030 pt = pmap_pvo_to_pte(pvo, pteidx); 2031 /* 2032 * Change the protection of the page. 2033 */ 2034 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2035 pvo->pvo_pte.pte_lo |= PTE_BR; 2036 2037 /* 2038 * If the PVO is in the page table, update 2039 * that pte at well. 2040 */ 2041 if (pt != NULL) { 2042 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2043 PVO_WHERE(pvo, PMAP_PROTECT); 2044 PMAPCOUNT(ptes_changed); 2045 } 2046 2047 PMAP_PVO_CHECK(pvo); /* sanity check */ 2048 } 2049 pmap_interrupts_restore(msr); 2050 } 2051 2052 void 2053 pmap_unwire(pmap_t pm, vaddr_t va) 2054 { 2055 struct pvo_entry *pvo; 2056 register_t msr; 2057 2058 msr = pmap_interrupts_off(); 2059 pvo = pmap_pvo_find_va(pm, va, NULL); 2060 if (pvo != NULL) { 2061 if (pvo->pvo_vaddr & PVO_WIRED) { 2062 pvo->pvo_vaddr &= ~PVO_WIRED; 2063 pm->pm_stats.wired_count--; 2064 } 2065 PMAP_PVO_CHECK(pvo); /* sanity check */ 2066 } 2067 pmap_interrupts_restore(msr); 2068 } 2069 2070 /* 2071 * Lower the protection on the specified physical page. 2072 */ 2073 void 2074 pmap_page_protect(struct vm_page *pg, vm_prot_t prot) 2075 { 2076 struct pvo_head *pvo_head; 2077 struct pvo_entry *pvo, *next_pvo; 2078 volatile struct pte *pt; 2079 register_t msr; 2080 2081 KASSERT(prot != VM_PROT_ALL); 2082 msr = pmap_interrupts_off(); 2083 2084 /* 2085 * When UVM reuses a page, it does a pmap_page_protect with 2086 * VM_PROT_NONE. At that point, we can clear the exec flag 2087 * since we know the page will have different contents. 2088 */ 2089 if ((prot & VM_PROT_READ) == 0) { 2090 DPRINTFN(EXEC, ("[pmap_page_protect: %#lx: clear-exec]\n", 2091 pg->phys_addr)); 2092 if (pmap_attr_fetch(pg) & PTE_EXEC) { 2093 PMAPCOUNT(exec_uncached_page_protect); 2094 pmap_attr_clear(pg, PTE_EXEC); 2095 } 2096 } 2097 2098 pvo_head = vm_page_to_pvoh(pg); 2099 for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) { 2100 next_pvo = LIST_NEXT(pvo, pvo_vlink); 2101 PMAP_PVO_CHECK(pvo); /* sanity check */ 2102 2103 /* 2104 * Downgrading to no mapping at all, we just remove the entry. 2105 */ 2106 if ((prot & VM_PROT_READ) == 0) { 2107 pmap_pvo_remove(pvo, -1); 2108 continue; 2109 } 2110 2111 /* 2112 * If EXEC permission is being revoked, just clear the 2113 * flag in the PVO. 2114 */ 2115 if ((prot & VM_PROT_EXECUTE) == 0) 2116 pvo_clear_exec(pvo); 2117 2118 /* 2119 * If this entry is already RO, don't diddle with the 2120 * page table. 2121 */ 2122 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) { 2123 PMAP_PVO_CHECK(pvo); 2124 continue; 2125 } 2126 2127 /* 2128 * Grab the PTE before the we diddle the bits so 2129 * pvo_to_pte can verify the pte contents are as 2130 * expected. 2131 */ 2132 pt = pmap_pvo_to_pte(pvo, -1); 2133 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2134 pvo->pvo_pte.pte_lo |= PTE_BR; 2135 if (pt != NULL) { 2136 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2137 PVO_WHERE(pvo, PMAP_PAGE_PROTECT); 2138 PMAPCOUNT(ptes_changed); 2139 } 2140 PMAP_PVO_CHECK(pvo); /* sanity check */ 2141 } 2142 pmap_interrupts_restore(msr); 2143 } 2144 2145 /* 2146 * Activate the address space for the specified process. If the process 2147 * is the current process, load the new MMU context. 2148 */ 2149 void 2150 pmap_activate(struct lwp *l) 2151 { 2152 struct pcb *pcb = &l->l_addr->u_pcb; 2153 pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap; 2154 2155 DPRINTFN(ACTIVATE, 2156 ("pmap_activate: lwp %p (curlwp %p)\n", l, curlwp)); 2157 2158 /* 2159 * XXX Normally performed in cpu_fork(). 2160 */ 2161 pcb->pcb_pm = pmap; 2162 2163 /* 2164 * In theory, the SR registers need only be valid on return 2165 * to user space wait to do them there. 2166 */ 2167 if (l == curlwp) { 2168 /* Store pointer to new current pmap. */ 2169 curpm = pmap; 2170 } 2171 } 2172 2173 /* 2174 * Deactivate the specified process's address space. 2175 */ 2176 void 2177 pmap_deactivate(struct lwp *l) 2178 { 2179 } 2180 2181 boolean_t 2182 pmap_query_bit(struct vm_page *pg, int ptebit) 2183 { 2184 struct pvo_entry *pvo; 2185 volatile struct pte *pt; 2186 register_t msr; 2187 2188 if (pmap_attr_fetch(pg) & ptebit) 2189 return TRUE; 2190 2191 msr = pmap_interrupts_off(); 2192 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2193 PMAP_PVO_CHECK(pvo); /* sanity check */ 2194 /* 2195 * See if we saved the bit off. If so cache, it and return 2196 * success. 2197 */ 2198 if (pvo->pvo_pte.pte_lo & ptebit) { 2199 pmap_attr_save(pg, ptebit); 2200 PMAP_PVO_CHECK(pvo); /* sanity check */ 2201 pmap_interrupts_restore(msr); 2202 return TRUE; 2203 } 2204 } 2205 /* 2206 * No luck, now go thru the hard part of looking at the ptes 2207 * themselves. Sync so any pending REF/CHG bits are flushed 2208 * to the PTEs. 2209 */ 2210 SYNC(); 2211 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2212 PMAP_PVO_CHECK(pvo); /* sanity check */ 2213 /* 2214 * See if this pvo have a valid PTE. If so, fetch the 2215 * REF/CHG bits from the valid PTE. If the appropriate 2216 * ptebit is set, cache, it and return success. 2217 */ 2218 pt = pmap_pvo_to_pte(pvo, -1); 2219 if (pt != NULL) { 2220 pmap_pte_synch(pt, &pvo->pvo_pte); 2221 if (pvo->pvo_pte.pte_lo & ptebit) { 2222 pmap_attr_save(pg, ptebit); 2223 PMAP_PVO_CHECK(pvo); /* sanity check */ 2224 pmap_interrupts_restore(msr); 2225 return TRUE; 2226 } 2227 } 2228 } 2229 pmap_interrupts_restore(msr); 2230 return FALSE; 2231 } 2232 2233 boolean_t 2234 pmap_clear_bit(struct vm_page *pg, int ptebit) 2235 { 2236 struct pvo_head *pvoh = vm_page_to_pvoh(pg); 2237 struct pvo_entry *pvo; 2238 volatile struct pte *pt; 2239 register_t msr; 2240 int rv = 0; 2241 2242 msr = pmap_interrupts_off(); 2243 2244 /* 2245 * Fetch the cache value 2246 */ 2247 rv |= pmap_attr_fetch(pg); 2248 2249 /* 2250 * Clear the cached value. 2251 */ 2252 pmap_attr_clear(pg, ptebit); 2253 2254 /* 2255 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we 2256 * can reset the right ones). Note that since the pvo entries and 2257 * list heads are accessed via BAT0 and are never placed in the 2258 * page table, we don't have to worry about further accesses setting 2259 * the REF/CHG bits. 2260 */ 2261 SYNC(); 2262 2263 /* 2264 * For each pvo entry, clear pvo's ptebit. If this pvo have a 2265 * valid PTE. If so, clear the ptebit from the valid PTE. 2266 */ 2267 LIST_FOREACH(pvo, pvoh, pvo_vlink) { 2268 PMAP_PVO_CHECK(pvo); /* sanity check */ 2269 pt = pmap_pvo_to_pte(pvo, -1); 2270 if (pt != NULL) { 2271 /* 2272 * Only sync the PTE if the bit we are looking 2273 * for is not already set. 2274 */ 2275 if ((pvo->pvo_pte.pte_lo & ptebit) == 0) 2276 pmap_pte_synch(pt, &pvo->pvo_pte); 2277 /* 2278 * If the bit we are looking for was already set, 2279 * clear that bit in the pte. 2280 */ 2281 if (pvo->pvo_pte.pte_lo & ptebit) 2282 pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit); 2283 } 2284 rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF); 2285 pvo->pvo_pte.pte_lo &= ~ptebit; 2286 PMAP_PVO_CHECK(pvo); /* sanity check */ 2287 } 2288 pmap_interrupts_restore(msr); 2289 2290 /* 2291 * If we are clearing the modify bit and this page was marked EXEC 2292 * and the user of the page thinks the page was modified, then we 2293 * need to clean it from the icache if it's mapped or clear the EXEC 2294 * bit if it's not mapped. The page itself might not have the CHG 2295 * bit set if the modification was done via DMA to the page. 2296 */ 2297 if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) { 2298 if (LIST_EMPTY(pvoh)) { 2299 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: clear-exec]\n", 2300 pg->phys_addr)); 2301 pmap_attr_clear(pg, PTE_EXEC); 2302 PMAPCOUNT(exec_uncached_clear_modify); 2303 } else { 2304 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: syncicache]\n", 2305 pg->phys_addr)); 2306 pmap_syncicache(pg->phys_addr, PAGE_SIZE); 2307 PMAPCOUNT(exec_synced_clear_modify); 2308 } 2309 } 2310 return (rv & ptebit) != 0; 2311 } 2312 2313 void 2314 pmap_procwr(struct proc *p, vaddr_t va, size_t len) 2315 { 2316 struct pvo_entry *pvo; 2317 size_t offset = va & ADDR_POFF; 2318 int s; 2319 2320 s = splvm(); 2321 while (len > 0) { 2322 size_t seglen = PAGE_SIZE - offset; 2323 if (seglen > len) 2324 seglen = len; 2325 pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL); 2326 if (pvo != NULL && PVO_ISEXECUTABLE(pvo)) { 2327 pmap_syncicache( 2328 (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen); 2329 PMAP_PVO_CHECK(pvo); 2330 } 2331 va += seglen; 2332 len -= seglen; 2333 offset = 0; 2334 } 2335 splx(s); 2336 } 2337 2338 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 2339 void 2340 pmap_pte_print(volatile struct pte *pt) 2341 { 2342 printf("PTE %p: ", pt); 2343 /* High word: */ 2344 printf("0x%08lx: [", pt->pte_hi); 2345 printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i'); 2346 printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-'); 2347 printf("0x%06lx 0x%02lx", 2348 (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT, 2349 pt->pte_hi & PTE_API); 2350 printf(" (va 0x%08lx)] ", pmap_pte_to_va(pt)); 2351 /* Low word: */ 2352 printf(" 0x%08lx: [", pt->pte_lo); 2353 printf("0x%05lx... ", pt->pte_lo >> 12); 2354 printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u'); 2355 printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n'); 2356 printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.'); 2357 printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.'); 2358 printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.'); 2359 printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.'); 2360 switch (pt->pte_lo & PTE_PP) { 2361 case PTE_BR: printf("br]\n"); break; 2362 case PTE_BW: printf("bw]\n"); break; 2363 case PTE_SO: printf("so]\n"); break; 2364 case PTE_SW: printf("sw]\n"); break; 2365 } 2366 } 2367 #endif 2368 2369 #if defined(DDB) 2370 void 2371 pmap_pteg_check(void) 2372 { 2373 volatile struct pte *pt; 2374 int i; 2375 int ptegidx; 2376 u_int p_valid = 0; 2377 u_int s_valid = 0; 2378 u_int invalid = 0; 2379 2380 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2381 for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) { 2382 if (pt->pte_hi & PTE_VALID) { 2383 if (pt->pte_hi & PTE_HID) 2384 s_valid++; 2385 else 2386 p_valid++; 2387 } else 2388 invalid++; 2389 } 2390 } 2391 printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n", 2392 p_valid, p_valid, s_valid, s_valid, 2393 invalid, invalid); 2394 } 2395 2396 void 2397 pmap_print_mmuregs(void) 2398 { 2399 int i; 2400 u_int cpuvers; 2401 #ifndef PPC_OEA64 2402 vaddr_t addr; 2403 register_t soft_sr[16]; 2404 #endif 2405 struct bat soft_ibat[4]; 2406 struct bat soft_dbat[4]; 2407 register_t sdr1; 2408 2409 cpuvers = MFPVR() >> 16; 2410 2411 __asm __volatile ("mfsdr1 %0" : "=r"(sdr1)); 2412 #ifndef PPC_OEA64 2413 addr = 0; 2414 for (i=0; i<16; i++) { 2415 soft_sr[i] = MFSRIN(addr); 2416 addr += (1 << ADDR_SR_SHFT); 2417 } 2418 #endif 2419 2420 /* read iBAT (601: uBAT) registers */ 2421 __asm __volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu)); 2422 __asm __volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl)); 2423 __asm __volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu)); 2424 __asm __volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl)); 2425 __asm __volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu)); 2426 __asm __volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl)); 2427 __asm __volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu)); 2428 __asm __volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl)); 2429 2430 2431 if (cpuvers != MPC601) { 2432 /* read dBAT registers */ 2433 __asm __volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu)); 2434 __asm __volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl)); 2435 __asm __volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu)); 2436 __asm __volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl)); 2437 __asm __volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu)); 2438 __asm __volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl)); 2439 __asm __volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu)); 2440 __asm __volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl)); 2441 } 2442 2443 printf("SDR1:\t0x%lx\n", (long) sdr1); 2444 #ifndef PPC_OEA64 2445 printf("SR[]:\t"); 2446 for (i=0; i<4; i++) 2447 printf("0x%08lx, ", soft_sr[i]); 2448 printf("\n\t"); 2449 for ( ; i<8; i++) 2450 printf("0x%08lx, ", soft_sr[i]); 2451 printf("\n\t"); 2452 for ( ; i<12; i++) 2453 printf("0x%08lx, ", soft_sr[i]); 2454 printf("\n\t"); 2455 for ( ; i<16; i++) 2456 printf("0x%08lx, ", soft_sr[i]); 2457 printf("\n"); 2458 #endif 2459 2460 printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i'); 2461 for (i=0; i<4; i++) { 2462 printf("0x%08lx 0x%08lx, ", 2463 soft_ibat[i].batu, soft_ibat[i].batl); 2464 if (i == 1) 2465 printf("\n\t"); 2466 } 2467 if (cpuvers != MPC601) { 2468 printf("\ndBAT[]:\t"); 2469 for (i=0; i<4; i++) { 2470 printf("0x%08lx 0x%08lx, ", 2471 soft_dbat[i].batu, soft_dbat[i].batl); 2472 if (i == 1) 2473 printf("\n\t"); 2474 } 2475 } 2476 printf("\n"); 2477 } 2478 2479 void 2480 pmap_print_pte(pmap_t pm, vaddr_t va) 2481 { 2482 struct pvo_entry *pvo; 2483 volatile struct pte *pt; 2484 int pteidx; 2485 2486 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2487 if (pvo != NULL) { 2488 pt = pmap_pvo_to_pte(pvo, pteidx); 2489 if (pt != NULL) { 2490 printf("VA %#lx -> %p -> %s %#lx, %#lx\n", 2491 va, pt, 2492 pt->pte_hi & PTE_HID ? "(sec)" : "(pri)", 2493 pt->pte_hi, pt->pte_lo); 2494 } else { 2495 printf("No valid PTE found\n"); 2496 } 2497 } else { 2498 printf("Address not in pmap\n"); 2499 } 2500 } 2501 2502 void 2503 pmap_pteg_dist(void) 2504 { 2505 struct pvo_entry *pvo; 2506 int ptegidx; 2507 int depth; 2508 int max_depth = 0; 2509 unsigned int depths[64]; 2510 2511 memset(depths, 0, sizeof(depths)); 2512 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2513 depth = 0; 2514 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2515 depth++; 2516 } 2517 if (depth > max_depth) 2518 max_depth = depth; 2519 if (depth > 63) 2520 depth = 63; 2521 depths[depth]++; 2522 } 2523 2524 for (depth = 0; depth < 64; depth++) { 2525 printf(" [%2d]: %8u", depth, depths[depth]); 2526 if ((depth & 3) == 3) 2527 printf("\n"); 2528 if (depth == max_depth) 2529 break; 2530 } 2531 if ((depth & 3) != 3) 2532 printf("\n"); 2533 printf("Max depth found was %d\n", max_depth); 2534 } 2535 #endif /* DEBUG */ 2536 2537 #if defined(PMAPCHECK) || defined(DEBUG) 2538 void 2539 pmap_pvo_verify(void) 2540 { 2541 int ptegidx; 2542 int s; 2543 2544 s = splvm(); 2545 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2546 struct pvo_entry *pvo; 2547 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2548 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 2549 panic("pmap_pvo_verify: invalid pvo %p " 2550 "on list %#x", pvo, ptegidx); 2551 pmap_pvo_check(pvo); 2552 } 2553 } 2554 splx(s); 2555 } 2556 #endif /* PMAPCHECK */ 2557 2558 2559 void * 2560 pmap_pool_ualloc(struct pool *pp, int flags) 2561 { 2562 struct pvo_page *pvop; 2563 2564 pvop = SIMPLEQ_FIRST(&pmap_upvop_head); 2565 if (pvop != NULL) { 2566 pmap_upvop_free--; 2567 SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link); 2568 return pvop; 2569 } 2570 if (uvm.page_init_done != TRUE) { 2571 return (void *) uvm_pageboot_alloc(PAGE_SIZE); 2572 } 2573 return pmap_pool_malloc(pp, flags); 2574 } 2575 2576 void * 2577 pmap_pool_malloc(struct pool *pp, int flags) 2578 { 2579 struct pvo_page *pvop; 2580 struct vm_page *pg; 2581 2582 pvop = SIMPLEQ_FIRST(&pmap_mpvop_head); 2583 if (pvop != NULL) { 2584 pmap_mpvop_free--; 2585 SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link); 2586 return pvop; 2587 } 2588 again: 2589 pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE, 2590 UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256); 2591 if (__predict_false(pg == NULL)) { 2592 if (flags & PR_WAITOK) { 2593 uvm_wait("plpg"); 2594 goto again; 2595 } else { 2596 return (0); 2597 } 2598 } 2599 return (void *) VM_PAGE_TO_PHYS(pg); 2600 } 2601 2602 void 2603 pmap_pool_ufree(struct pool *pp, void *va) 2604 { 2605 struct pvo_page *pvop; 2606 #if 0 2607 if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) { 2608 pmap_pool_mfree(va, size, tag); 2609 return; 2610 } 2611 #endif 2612 pvop = va; 2613 SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link); 2614 pmap_upvop_free++; 2615 if (pmap_upvop_free > pmap_upvop_maxfree) 2616 pmap_upvop_maxfree = pmap_upvop_free; 2617 } 2618 2619 void 2620 pmap_pool_mfree(struct pool *pp, void *va) 2621 { 2622 struct pvo_page *pvop; 2623 2624 pvop = va; 2625 SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link); 2626 pmap_mpvop_free++; 2627 if (pmap_mpvop_free > pmap_mpvop_maxfree) 2628 pmap_mpvop_maxfree = pmap_mpvop_free; 2629 #if 0 2630 uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va)); 2631 #endif 2632 } 2633 2634 /* 2635 * This routine in bootstraping to steal to-be-managed memory (which will 2636 * then be unmanaged). We use it to grab from the first 256MB for our 2637 * pmap needs and above 256MB for other stuff. 2638 */ 2639 vaddr_t 2640 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp) 2641 { 2642 vsize_t size; 2643 vaddr_t va; 2644 paddr_t pa = 0; 2645 int npgs, bank; 2646 struct vm_physseg *ps; 2647 2648 if (uvm.page_init_done == TRUE) 2649 panic("pmap_steal_memory: called _after_ bootstrap"); 2650 2651 *vstartp = VM_MIN_KERNEL_ADDRESS; 2652 *vendp = VM_MAX_KERNEL_ADDRESS; 2653 2654 size = round_page(vsize); 2655 npgs = atop(size); 2656 2657 /* 2658 * PA 0 will never be among those given to UVM so we can use it 2659 * to indicate we couldn't steal any memory. 2660 */ 2661 for (ps = vm_physmem, bank = 0; bank < vm_nphysseg; bank++, ps++) { 2662 if (ps->free_list == VM_FREELIST_FIRST256 && 2663 ps->avail_end - ps->avail_start >= npgs) { 2664 pa = ptoa(ps->avail_start); 2665 break; 2666 } 2667 } 2668 2669 if (pa == 0) 2670 panic("pmap_steal_memory: no approriate memory to steal!"); 2671 2672 ps->avail_start += npgs; 2673 ps->start += npgs; 2674 2675 /* 2676 * If we've used up all the pages in the segment, remove it and 2677 * compact the list. 2678 */ 2679 if (ps->avail_start == ps->end) { 2680 /* 2681 * If this was the last one, then a very bad thing has occurred 2682 */ 2683 if (--vm_nphysseg == 0) 2684 panic("pmap_steal_memory: out of memory!"); 2685 2686 printf("pmap_steal_memory: consumed bank %d\n", bank); 2687 for (; bank < vm_nphysseg; bank++, ps++) { 2688 ps[0] = ps[1]; 2689 } 2690 } 2691 2692 va = (vaddr_t) pa; 2693 memset((caddr_t) va, 0, size); 2694 pmap_pages_stolen += npgs; 2695 #ifdef DEBUG 2696 if (pmapdebug && npgs > 1) { 2697 u_int cnt = 0; 2698 for (bank = 0, ps = vm_physmem; bank < vm_nphysseg; bank++, ps++) 2699 cnt += ps->avail_end - ps->avail_start; 2700 printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n", 2701 npgs, pmap_pages_stolen, cnt); 2702 } 2703 #endif 2704 2705 return va; 2706 } 2707 2708 /* 2709 * Find a chuck of memory with right size and alignment. 2710 */ 2711 void * 2712 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end) 2713 { 2714 struct mem_region *mp; 2715 paddr_t s, e; 2716 int i, j; 2717 2718 size = round_page(size); 2719 2720 DPRINTFN(BOOT, 2721 ("pmap_boot_find_memory: size=%lx, alignment=%lx, at_end=%d", 2722 size, alignment, at_end)); 2723 2724 if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0) 2725 panic("pmap_boot_find_memory: invalid alignment %lx", 2726 alignment); 2727 2728 if (at_end) { 2729 if (alignment != PAGE_SIZE) 2730 panic("pmap_boot_find_memory: invalid ending " 2731 "alignment %lx", alignment); 2732 2733 for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) { 2734 s = mp->start + mp->size - size; 2735 if (s >= mp->start && mp->size >= size) { 2736 DPRINTFN(BOOT,(": %lx\n", s)); 2737 DPRINTFN(BOOT, 2738 ("pmap_boot_find_memory: b-avail[%d] start " 2739 "0x%lx size 0x%lx\n", mp - avail, 2740 mp->start, mp->size)); 2741 mp->size -= size; 2742 DPRINTFN(BOOT, 2743 ("pmap_boot_find_memory: a-avail[%d] start " 2744 "0x%lx size 0x%lx\n", mp - avail, 2745 mp->start, mp->size)); 2746 return (void *) s; 2747 } 2748 } 2749 panic("pmap_boot_find_memory: no available memory"); 2750 } 2751 2752 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2753 s = (mp->start + alignment - 1) & ~(alignment-1); 2754 e = s + size; 2755 2756 /* 2757 * Is the calculated region entirely within the region? 2758 */ 2759 if (s < mp->start || e > mp->start + mp->size) 2760 continue; 2761 2762 DPRINTFN(BOOT,(": %lx\n", s)); 2763 if (s == mp->start) { 2764 /* 2765 * If the block starts at the beginning of region, 2766 * adjust the size & start. (the region may now be 2767 * zero in length) 2768 */ 2769 DPRINTFN(BOOT, 2770 ("pmap_boot_find_memory: b-avail[%d] start " 2771 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2772 mp->start += size; 2773 mp->size -= size; 2774 DPRINTFN(BOOT, 2775 ("pmap_boot_find_memory: a-avail[%d] start " 2776 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2777 } else if (e == mp->start + mp->size) { 2778 /* 2779 * If the block starts at the beginning of region, 2780 * adjust only the size. 2781 */ 2782 DPRINTFN(BOOT, 2783 ("pmap_boot_find_memory: b-avail[%d] start " 2784 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2785 mp->size -= size; 2786 DPRINTFN(BOOT, 2787 ("pmap_boot_find_memory: a-avail[%d] start " 2788 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2789 } else { 2790 /* 2791 * Block is in the middle of the region, so we 2792 * have to split it in two. 2793 */ 2794 for (j = avail_cnt; j > i + 1; j--) { 2795 avail[j] = avail[j-1]; 2796 } 2797 DPRINTFN(BOOT, 2798 ("pmap_boot_find_memory: b-avail[%d] start " 2799 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2800 mp[1].start = e; 2801 mp[1].size = mp[0].start + mp[0].size - e; 2802 mp[0].size = s - mp[0].start; 2803 avail_cnt++; 2804 for (; i < avail_cnt; i++) { 2805 DPRINTFN(BOOT, 2806 ("pmap_boot_find_memory: a-avail[%d] " 2807 "start 0x%lx size 0x%lx\n", i, 2808 avail[i].start, avail[i].size)); 2809 } 2810 } 2811 return (void *) s; 2812 } 2813 panic("pmap_boot_find_memory: not enough memory for " 2814 "%lx/%lx allocation?", size, alignment); 2815 } 2816 2817 /* 2818 * This is not part of the defined PMAP interface and is specific to the 2819 * PowerPC architecture. This is called during initppc, before the system 2820 * is really initialized. 2821 */ 2822 void 2823 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend) 2824 { 2825 struct mem_region *mp, tmp; 2826 paddr_t s, e; 2827 psize_t size; 2828 int i, j; 2829 2830 /* 2831 * Get memory. 2832 */ 2833 mem_regions(&mem, &avail); 2834 #if defined(DEBUG) 2835 if (pmapdebug & PMAPDEBUG_BOOT) { 2836 printf("pmap_bootstrap: memory configuration:\n"); 2837 for (mp = mem; mp->size; mp++) { 2838 printf("pmap_bootstrap: mem start 0x%lx size 0x%lx\n", 2839 mp->start, mp->size); 2840 } 2841 for (mp = avail; mp->size; mp++) { 2842 printf("pmap_bootstrap: avail start 0x%lx size 0x%lx\n", 2843 mp->start, mp->size); 2844 } 2845 } 2846 #endif 2847 2848 /* 2849 * Find out how much physical memory we have and in how many chunks. 2850 */ 2851 for (mem_cnt = 0, mp = mem; mp->size; mp++) { 2852 if (mp->start >= pmap_memlimit) 2853 continue; 2854 if (mp->start + mp->size > pmap_memlimit) { 2855 size = pmap_memlimit - mp->start; 2856 physmem += btoc(size); 2857 } else { 2858 physmem += btoc(mp->size); 2859 } 2860 mem_cnt++; 2861 } 2862 2863 /* 2864 * Count the number of available entries. 2865 */ 2866 for (avail_cnt = 0, mp = avail; mp->size; mp++) 2867 avail_cnt++; 2868 2869 /* 2870 * Page align all regions. 2871 */ 2872 kernelstart = trunc_page(kernelstart); 2873 kernelend = round_page(kernelend); 2874 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2875 s = round_page(mp->start); 2876 mp->size -= (s - mp->start); 2877 mp->size = trunc_page(mp->size); 2878 mp->start = s; 2879 e = mp->start + mp->size; 2880 2881 DPRINTFN(BOOT, 2882 ("pmap_bootstrap: b-avail[%d] start 0x%lx size 0x%lx\n", 2883 i, mp->start, mp->size)); 2884 2885 /* 2886 * Don't allow the end to run beyond our artificial limit 2887 */ 2888 if (e > pmap_memlimit) 2889 e = pmap_memlimit; 2890 2891 /* 2892 * Is this region empty or strange? skip it. 2893 */ 2894 if (e <= s) { 2895 mp->start = 0; 2896 mp->size = 0; 2897 continue; 2898 } 2899 2900 /* 2901 * Does this overlap the beginning of kernel? 2902 * Does extend past the end of the kernel? 2903 */ 2904 else if (s < kernelstart && e > kernelstart) { 2905 if (e > kernelend) { 2906 avail[avail_cnt].start = kernelend; 2907 avail[avail_cnt].size = e - kernelend; 2908 avail_cnt++; 2909 } 2910 mp->size = kernelstart - s; 2911 } 2912 /* 2913 * Check whether this region overlaps the end of the kernel. 2914 */ 2915 else if (s < kernelend && e > kernelend) { 2916 mp->start = kernelend; 2917 mp->size = e - kernelend; 2918 } 2919 /* 2920 * Look whether this regions is completely inside the kernel. 2921 * Nuke it if it does. 2922 */ 2923 else if (s >= kernelstart && e <= kernelend) { 2924 mp->start = 0; 2925 mp->size = 0; 2926 } 2927 /* 2928 * If the user imposed a memory limit, enforce it. 2929 */ 2930 else if (s >= pmap_memlimit) { 2931 mp->start = -PAGE_SIZE; /* let's know why */ 2932 mp->size = 0; 2933 } 2934 else { 2935 mp->start = s; 2936 mp->size = e - s; 2937 } 2938 DPRINTFN(BOOT, 2939 ("pmap_bootstrap: a-avail[%d] start 0x%lx size 0x%lx\n", 2940 i, mp->start, mp->size)); 2941 } 2942 2943 /* 2944 * Move (and uncount) all the null return to the end. 2945 */ 2946 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2947 if (mp->size == 0) { 2948 tmp = avail[i]; 2949 avail[i] = avail[--avail_cnt]; 2950 avail[avail_cnt] = avail[i]; 2951 } 2952 } 2953 2954 /* 2955 * (Bubble)sort them into asecnding order. 2956 */ 2957 for (i = 0; i < avail_cnt; i++) { 2958 for (j = i + 1; j < avail_cnt; j++) { 2959 if (avail[i].start > avail[j].start) { 2960 tmp = avail[i]; 2961 avail[i] = avail[j]; 2962 avail[j] = tmp; 2963 } 2964 } 2965 } 2966 2967 /* 2968 * Make sure they don't overlap. 2969 */ 2970 for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) { 2971 if (mp[0].start + mp[0].size > mp[1].start) { 2972 mp[0].size = mp[1].start - mp[0].start; 2973 } 2974 DPRINTFN(BOOT, 2975 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 2976 i, mp->start, mp->size)); 2977 } 2978 DPRINTFN(BOOT, 2979 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 2980 i, mp->start, mp->size)); 2981 2982 #ifdef PTEGCOUNT 2983 pmap_pteg_cnt = PTEGCOUNT; 2984 #else /* PTEGCOUNT */ 2985 pmap_pteg_cnt = 0x1000; 2986 2987 while (pmap_pteg_cnt < physmem) 2988 pmap_pteg_cnt <<= 1; 2989 2990 pmap_pteg_cnt >>= 1; 2991 #endif /* PTEGCOUNT */ 2992 2993 /* 2994 * Find suitably aligned memory for PTEG hash table. 2995 */ 2996 size = pmap_pteg_cnt * sizeof(struct pteg); 2997 pmap_pteg_table = pmap_boot_find_memory(size, size, 0); 2998 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 2999 if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH) 3000 panic("pmap_bootstrap: pmap_pteg_table end (%p + %lx) > 256MB", 3001 pmap_pteg_table, size); 3002 #endif 3003 3004 memset((void *)pmap_pteg_table, 0, pmap_pteg_cnt * sizeof(struct pteg)); 3005 pmap_pteg_mask = pmap_pteg_cnt - 1; 3006 3007 /* 3008 * We cannot do pmap_steal_memory here since UVM hasn't been loaded 3009 * with pages. So we just steal them before giving them to UVM. 3010 */ 3011 size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt; 3012 pmap_pvo_table = pmap_boot_find_memory(size, PAGE_SIZE, 0); 3013 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3014 if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH) 3015 panic("pmap_bootstrap: pmap_pvo_table end (%p + %lx) > 256MB", 3016 pmap_pvo_table, size); 3017 #endif 3018 3019 for (i = 0; i < pmap_pteg_cnt; i++) 3020 TAILQ_INIT(&pmap_pvo_table[i]); 3021 3022 #ifndef MSGBUFADDR 3023 /* 3024 * Allocate msgbuf in high memory. 3025 */ 3026 msgbuf_paddr = 3027 (paddr_t) pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1); 3028 #endif 3029 3030 #ifdef __HAVE_PMAP_PHYSSEG 3031 { 3032 u_int npgs = 0; 3033 for (i = 0, mp = avail; i < avail_cnt; i++, mp++) 3034 npgs += btoc(mp->size); 3035 size = (sizeof(struct pvo_head) + 1) * npgs; 3036 pmap_physseg.pvoh = pmap_boot_find_memory(size, PAGE_SIZE, 0); 3037 pmap_physseg.attrs = (char *) &pmap_physseg.pvoh[npgs]; 3038 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3039 if ((uintptr_t)pmap_physseg.pvoh + size > SEGMENT_LENGTH) 3040 panic("pmap_bootstrap: PVO list end (%p + %lx) > 256MB", 3041 pmap_physseg.pvoh, size); 3042 #endif 3043 } 3044 #endif 3045 3046 for (mp = avail, i = 0; i < avail_cnt; mp++, i++) { 3047 paddr_t pfstart = atop(mp->start); 3048 paddr_t pfend = atop(mp->start + mp->size); 3049 if (mp->size == 0) 3050 continue; 3051 if (mp->start + mp->size <= SEGMENT_LENGTH) { 3052 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3053 VM_FREELIST_FIRST256); 3054 } else if (mp->start >= SEGMENT_LENGTH) { 3055 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3056 VM_FREELIST_DEFAULT); 3057 } else { 3058 pfend = atop(SEGMENT_LENGTH); 3059 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3060 VM_FREELIST_FIRST256); 3061 pfstart = atop(SEGMENT_LENGTH); 3062 pfend = atop(mp->start + mp->size); 3063 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3064 VM_FREELIST_DEFAULT); 3065 } 3066 } 3067 3068 /* 3069 * Make sure kernel vsid is allocated as well as VSID 0. 3070 */ 3071 pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW] 3072 |= 1 << (KERNEL_VSIDBITS % VSID_NBPW); 3073 pmap_vsid_bitmap[0] |= 1; 3074 3075 /* 3076 * Initialize kernel pmap and hardware. 3077 */ 3078 #ifndef PPC_OEA64 3079 for (i = 0; i < 16; i++) { 3080 pmap_kernel()->pm_sr[i] = EMPTY_SEGMENT; 3081 __asm __volatile ("mtsrin %0,%1" 3082 :: "r"(EMPTY_SEGMENT), "r"(i << ADDR_SR_SHFT)); 3083 } 3084 3085 pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY; 3086 __asm __volatile ("mtsr %0,%1" 3087 :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT)); 3088 #ifdef KERNEL2_SR 3089 pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY; 3090 __asm __volatile ("mtsr %0,%1" 3091 :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT)); 3092 #endif 3093 for (i = 0; i < 16; i++) { 3094 if (iosrtable[i] & SR601_T) { 3095 pmap_kernel()->pm_sr[i] = iosrtable[i]; 3096 __asm __volatile ("mtsrin %0,%1" 3097 :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT)); 3098 } 3099 } 3100 #endif /* !PPC_OEA64 */ 3101 3102 __asm __volatile ("sync; mtsdr1 %0; isync" 3103 :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10))); 3104 tlbia(); 3105 3106 #ifdef ALTIVEC 3107 pmap_use_altivec = cpu_altivec; 3108 #endif 3109 3110 #ifdef DEBUG 3111 if (pmapdebug & PMAPDEBUG_BOOT) { 3112 u_int cnt; 3113 int bank; 3114 char pbuf[9]; 3115 for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) { 3116 cnt += vm_physmem[bank].avail_end - vm_physmem[bank].avail_start; 3117 printf("pmap_bootstrap: vm_physmem[%d]=%#lx-%#lx/%#lx\n", 3118 bank, 3119 ptoa(vm_physmem[bank].avail_start), 3120 ptoa(vm_physmem[bank].avail_end), 3121 ptoa(vm_physmem[bank].avail_end - vm_physmem[bank].avail_start)); 3122 } 3123 format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt)); 3124 printf("pmap_bootstrap: UVM memory = %s (%u pages)\n", 3125 pbuf, cnt); 3126 } 3127 #endif 3128 3129 pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry), 3130 sizeof(struct pvo_entry), 0, 0, "pmap_upvopl", 3131 &pmap_pool_uallocator); 3132 3133 pool_setlowat(&pmap_upvo_pool, 252); 3134 3135 pool_init(&pmap_pool, sizeof(struct pmap), 3136 sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator); 3137 } 3138
Indexes created Mon Oct 27 08:10:08 GMT 2025