pmap.c revision 1.19
1 /* $NetBSD: pmap.c,v 1.19 2003/12/27 13:35:52 mjl 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 1.19 2003/12/27 13:35:52 mjl 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 paddr_t pmap_memlimit = -PAGE_SIZE; /* there is no limit */ 112 113 struct pmap kernel_pmap_; 114 unsigned int pmap_pages_stolen; 115 u_long pmap_pte_valid; 116 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 117 u_long pmap_pvo_enter_depth; 118 u_long pmap_pvo_remove_depth; 119 #endif 120 121 int physmem; 122 #ifndef MSGBUFADDR 123 extern paddr_t msgbuf_paddr; 124 #endif 125 126 static struct mem_region *mem, *avail; 127 static u_int mem_cnt, avail_cnt; 128 129 #ifdef __HAVE_PMAP_PHYSSEG 130 /* 131 * This is a cache of referenced/modified bits. 132 * Bits herein are shifted by ATTRSHFT. 133 */ 134 #define ATTR_SHFT 4 135 struct pmap_physseg pmap_physseg; 136 #endif 137 138 /* 139 * The following structure is exactly 32 bytes long (one cacheline). 140 */ 141 struct pvo_entry { 142 LIST_ENTRY(pvo_entry) pvo_vlink; /* Link to common virt page */ 143 TAILQ_ENTRY(pvo_entry) pvo_olink; /* Link to overflow entry */ 144 struct pte pvo_pte; /* Prebuilt PTE */ 145 pmap_t pvo_pmap; /* ptr to owning pmap */ 146 vaddr_t pvo_vaddr; /* VA of entry */ 147 #define PVO_PTEGIDX_MASK 0x0007 /* which PTEG slot */ 148 #define PVO_PTEGIDX_VALID 0x0008 /* slot is valid */ 149 #define PVO_WIRED 0x0010 /* PVO entry is wired */ 150 #define PVO_MANAGED 0x0020 /* PVO e. for managed page */ 151 #define PVO_EXECUTABLE 0x0040 /* PVO e. for executable page */ 152 #define PVO_ENTER_INSERT 0 /* PVO has been removed */ 153 #define PVO_SPILL_UNSET 1 /* PVO has been evicted */ 154 #define PVO_SPILL_SET 2 /* PVO has been spilled */ 155 #define PVO_SPILL_INSERT 3 /* PVO has been inserted */ 156 #define PVO_PMAP_PAGE_PROTECT 4 /* PVO has changed */ 157 #define PVO_PMAP_PROTECT 5 /* PVO has changed */ 158 #define PVO_REMOVE 6 /* PVO has been removed */ 159 #define PVO_WHERE_MASK 15 160 #define PVO_WHERE_SHFT 8 161 }; 162 #define PVO_VADDR(pvo) ((pvo)->pvo_vaddr & ~ADDR_POFF) 163 #define PVO_ISEXECUTABLE(pvo) ((pvo)->pvo_vaddr & PVO_EXECUTABLE) 164 #define PVO_PTEGIDX_GET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK) 165 #define PVO_PTEGIDX_ISSET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID) 166 #define PVO_PTEGIDX_CLR(pvo) \ 167 ((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK))) 168 #define PVO_PTEGIDX_SET(pvo,i) \ 169 ((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID)) 170 #define PVO_WHERE(pvo,w) \ 171 ((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \ 172 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT)) 173 174 TAILQ_HEAD(pvo_tqhead, pvo_entry); 175 struct pvo_tqhead *pmap_pvo_table; /* pvo entries by ptegroup index */ 176 struct pvo_head pmap_pvo_kunmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged); /* list of unmanaged pages */ 177 struct pvo_head pmap_pvo_unmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged); /* list of unmanaged pages */ 178 179 struct pool pmap_pool; /* pool for pmap structures */ 180 struct pool pmap_upvo_pool; /* pool for pvo entries for unmanaged pages */ 181 struct pool pmap_mpvo_pool; /* pool for pvo entries for managed pages */ 182 183 /* 184 * We keep a cache of unmanaged pages to be used for pvo entries for 185 * unmanaged pages. 186 */ 187 struct pvo_page { 188 SIMPLEQ_ENTRY(pvo_page) pvop_link; 189 }; 190 SIMPLEQ_HEAD(pvop_head, pvo_page); 191 struct pvop_head pmap_upvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_upvop_head); 192 struct pvop_head pmap_mpvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_mpvop_head); 193 u_long pmap_upvop_free; 194 u_long pmap_upvop_maxfree; 195 u_long pmap_mpvop_free; 196 u_long pmap_mpvop_maxfree; 197 198 STATIC void *pmap_pool_ualloc(struct pool *, int); 199 STATIC void *pmap_pool_malloc(struct pool *, int); 200 201 STATIC void pmap_pool_ufree(struct pool *, void *); 202 STATIC void pmap_pool_mfree(struct pool *, void *); 203 204 static struct pool_allocator pmap_pool_mallocator = { 205 pmap_pool_malloc, pmap_pool_mfree, 0, 206 }; 207 208 static struct pool_allocator pmap_pool_uallocator = { 209 pmap_pool_ualloc, pmap_pool_ufree, 0, 210 }; 211 212 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 213 void pmap_pte_print(volatile struct pte *); 214 #endif 215 216 #ifdef DDB 217 void pmap_pteg_check(void); 218 void pmap_pteg_dist(void); 219 void pmap_print_pte(pmap_t, vaddr_t); 220 void pmap_print_mmuregs(void); 221 #endif 222 223 #if defined(DEBUG) || defined(PMAPCHECK) 224 #ifdef PMAPCHECK 225 int pmapcheck = 1; 226 #else 227 int pmapcheck = 0; 228 #endif 229 void pmap_pvo_verify(void); 230 STATIC void pmap_pvo_check(const struct pvo_entry *); 231 #define PMAP_PVO_CHECK(pvo) \ 232 do { \ 233 if (pmapcheck) \ 234 pmap_pvo_check(pvo); \ 235 } while (0) 236 #else 237 #define PMAP_PVO_CHECK(pvo) do { } while (/*CONSTCOND*/0) 238 #endif 239 STATIC int pmap_pte_insert(int, struct pte *); 240 STATIC int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *, 241 vaddr_t, paddr_t, register_t, int); 242 STATIC void pmap_pvo_remove(struct pvo_entry *, int); 243 STATIC struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *); 244 STATIC volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int); 245 #define pmap_pvo_reclaim(pm) NULL 246 STATIC void pvo_set_exec(struct pvo_entry *); 247 STATIC void pvo_clear_exec(struct pvo_entry *); 248 249 STATIC void tlbia(void); 250 251 STATIC void pmap_release(pmap_t); 252 STATIC void *pmap_boot_find_memory(psize_t, psize_t, int); 253 254 #define VSID_NBPW (sizeof(uint32_t) * 8) 255 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW]; 256 257 static int pmap_initialized; 258 259 #if defined(DEBUG) || defined(PMAPDEBUG) 260 #define PMAPDEBUG_BOOT 0x0001 261 #define PMAPDEBUG_PTE 0x0002 262 #define PMAPDEBUG_EXEC 0x0008 263 #define PMAPDEBUG_PVOENTER 0x0010 264 #define PMAPDEBUG_PVOREMOVE 0x0020 265 #define PMAPDEBUG_ACTIVATE 0x0100 266 #define PMAPDEBUG_CREATE 0x0200 267 #define PMAPDEBUG_ENTER 0x1000 268 #define PMAPDEBUG_KENTER 0x2000 269 #define PMAPDEBUG_KREMOVE 0x4000 270 #define PMAPDEBUG_REMOVE 0x8000 271 unsigned int pmapdebug = 0; 272 # define DPRINTF(x) printf x 273 # define DPRINTFN(n, x) if (pmapdebug & PMAPDEBUG_ ## n) printf x 274 #else 275 # define DPRINTF(x) 276 # define DPRINTFN(n, x) 277 #endif 278 279 280 #ifdef PMAPCOUNTERS 281 #define PMAPCOUNT(ev) ((pmap_evcnt_ ## ev).ev_count++) 282 #define PMAPCOUNT2(ev) ((ev).ev_count++) 283 284 struct evcnt pmap_evcnt_mappings = 285 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 286 "pmap", "pages mapped"); 287 struct evcnt pmap_evcnt_unmappings = 288 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 289 "pmap", "pages unmapped"); 290 291 struct evcnt pmap_evcnt_kernel_mappings = 292 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 293 "pmap", "kernel pages mapped"); 294 struct evcnt pmap_evcnt_kernel_unmappings = 295 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_kernel_mappings, 296 "pmap", "kernel pages unmapped"); 297 298 struct evcnt pmap_evcnt_mappings_replaced = 299 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 300 "pmap", "page mappings replaced"); 301 302 struct evcnt pmap_evcnt_exec_mappings = 303 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 304 "pmap", "exec pages mapped"); 305 struct evcnt pmap_evcnt_exec_cached = 306 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 307 "pmap", "exec pages cached"); 308 309 struct evcnt pmap_evcnt_exec_synced = 310 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 311 "pmap", "exec pages synced"); 312 struct evcnt pmap_evcnt_exec_synced_clear_modify = 313 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 314 "pmap", "exec pages synced (CM)"); 315 316 struct evcnt pmap_evcnt_exec_uncached_page_protect = 317 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 318 "pmap", "exec pages uncached (PP)"); 319 struct evcnt pmap_evcnt_exec_uncached_clear_modify = 320 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 321 "pmap", "exec pages uncached (CM)"); 322 struct evcnt pmap_evcnt_exec_uncached_zero_page = 323 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 324 "pmap", "exec pages uncached (ZP)"); 325 struct evcnt pmap_evcnt_exec_uncached_copy_page = 326 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 327 "pmap", "exec pages uncached (CP)"); 328 329 struct evcnt pmap_evcnt_updates = 330 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 331 "pmap", "updates"); 332 struct evcnt pmap_evcnt_collects = 333 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 334 "pmap", "collects"); 335 struct evcnt pmap_evcnt_copies = 336 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 337 "pmap", "copies"); 338 339 struct evcnt pmap_evcnt_ptes_spilled = 340 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 341 "pmap", "ptes spilled from overflow"); 342 struct evcnt pmap_evcnt_ptes_unspilled = 343 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 344 "pmap", "ptes not spilled"); 345 struct evcnt pmap_evcnt_ptes_evicted = 346 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 347 "pmap", "ptes evicted"); 348 349 struct evcnt pmap_evcnt_ptes_primary[8] = { 350 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 351 "pmap", "ptes added at primary[0]"), 352 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 353 "pmap", "ptes added at primary[1]"), 354 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 355 "pmap", "ptes added at primary[2]"), 356 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 357 "pmap", "ptes added at primary[3]"), 358 359 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 360 "pmap", "ptes added at primary[4]"), 361 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 362 "pmap", "ptes added at primary[5]"), 363 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 364 "pmap", "ptes added at primary[6]"), 365 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 366 "pmap", "ptes added at primary[7]"), 367 }; 368 struct evcnt pmap_evcnt_ptes_secondary[8] = { 369 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 370 "pmap", "ptes added at secondary[0]"), 371 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 372 "pmap", "ptes added at secondary[1]"), 373 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 374 "pmap", "ptes added at secondary[2]"), 375 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 376 "pmap", "ptes added at secondary[3]"), 377 378 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 379 "pmap", "ptes added at secondary[4]"), 380 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 381 "pmap", "ptes added at secondary[5]"), 382 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 383 "pmap", "ptes added at secondary[6]"), 384 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 385 "pmap", "ptes added at secondary[7]"), 386 }; 387 struct evcnt pmap_evcnt_ptes_removed = 388 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 389 "pmap", "ptes removed"); 390 struct evcnt pmap_evcnt_ptes_changed = 391 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 392 "pmap", "ptes changed"); 393 394 /* 395 * From pmap_subr.c 396 */ 397 extern struct evcnt pmap_evcnt_zeroed_pages; 398 extern struct evcnt pmap_evcnt_copied_pages; 399 extern struct evcnt pmap_evcnt_idlezeroed_pages; 400 #else 401 #define PMAPCOUNT(ev) ((void) 0) 402 #define PMAPCOUNT2(ev) ((void) 0) 403 #endif 404 405 #define TLBIE(va) __asm __volatile("tlbie %0" :: "r"(va)) 406 #define TLBSYNC() __asm __volatile("tlbsync") 407 #define SYNC() __asm __volatile("sync") 408 #define EIEIO() __asm __volatile("eieio") 409 #define MFMSR() mfmsr() 410 #define MTMSR(psl) mtmsr(psl) 411 #define MFPVR() mfpvr() 412 #define MFSRIN(va) mfsrin(va) 413 #define MFTB() mfrtcltbl() 414 415 #ifndef PPC_OEA64 416 static __inline register_t 417 mfsrin(vaddr_t va) 418 { 419 register_t sr; 420 __asm __volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va)); 421 return sr; 422 } 423 #endif /* PPC_OEA64 */ 424 425 static __inline register_t 426 pmap_interrupts_off(void) 427 { 428 register_t msr = MFMSR(); 429 if (msr & PSL_EE) 430 MTMSR(msr & ~PSL_EE); 431 return msr; 432 } 433 434 static void 435 pmap_interrupts_restore(register_t msr) 436 { 437 if (msr & PSL_EE) 438 MTMSR(msr); 439 } 440 441 static __inline u_int32_t 442 mfrtcltbl(void) 443 { 444 445 if ((MFPVR() >> 16) == MPC601) 446 return (mfrtcl() >> 7); 447 else 448 return (mftbl()); 449 } 450 451 /* 452 * These small routines may have to be replaced, 453 * if/when we support processors other that the 604. 454 */ 455 456 void 457 tlbia(void) 458 { 459 caddr_t i; 460 461 SYNC(); 462 /* 463 * Why not use "tlbia"? Because not all processors implement it. 464 * 465 * This needs to be a per-cpu callback to do the appropriate thing 466 * for the CPU. XXX 467 */ 468 for (i = 0; i < (caddr_t)0x00040000; i += 0x00001000) { 469 TLBIE(i); 470 EIEIO(); 471 SYNC(); 472 } 473 TLBSYNC(); 474 SYNC(); 475 } 476 477 static __inline register_t 478 va_to_vsid(const struct pmap *pm, vaddr_t addr) 479 { 480 #ifdef PPC_OEA64 481 #if 0 482 const struct ste *ste; 483 register_t hash; 484 int i; 485 486 hash = (addr >> ADDR_ESID_SHFT) & ADDR_ESID_HASH; 487 488 /* 489 * Try the primary group first 490 */ 491 ste = pm->pm_stes[hash].stes; 492 for (i = 0; i < 8; i++, ste++) { 493 if (ste->ste_hi & STE_V) && 494 (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID)) 495 return ste; 496 } 497 498 /* 499 * Then the secondary group. 500 */ 501 ste = pm->pm_stes[hash ^ ADDR_ESID_HASH].stes; 502 for (i = 0; i < 8; i++, ste++) { 503 if (ste->ste_hi & STE_V) && 504 (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID)) 505 return addr; 506 } 507 508 return NULL; 509 #else 510 /* 511 * Rather than searching the STE groups for the VSID, we know 512 * how we generate that from the ESID and so do that. 513 */ 514 return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT; 515 #endif 516 #else 517 return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID) >> SR_VSID_SHFT; 518 #endif 519 } 520 521 static __inline register_t 522 va_to_pteg(const struct pmap *pm, vaddr_t addr) 523 { 524 register_t hash; 525 526 hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT); 527 return hash & pmap_pteg_mask; 528 } 529 530 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 531 /* 532 * Given a PTE in the page table, calculate the VADDR that hashes to it. 533 * The only bit of magic is that the top 4 bits of the address doesn't 534 * technically exist in the PTE. But we know we reserved 4 bits of the 535 * VSID for it so that's how we get it. 536 */ 537 static vaddr_t 538 pmap_pte_to_va(volatile const struct pte *pt) 539 { 540 vaddr_t va; 541 uintptr_t ptaddr = (uintptr_t) pt; 542 543 if (pt->pte_hi & PTE_HID) 544 ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg)); 545 546 /* PPC Bits 10-19 PPC64 Bits 42-51 */ 547 va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff; 548 va <<= ADDR_PIDX_SHFT; 549 550 /* PPC Bits 4-9 PPC64 Bits 36-41 */ 551 va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT; 552 553 #ifdef PPC_OEA64 554 /* PPC63 Bits 0-35 */ 555 /* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */ 556 #endif 557 #ifdef PPC_OEA 558 /* PPC Bits 0-3 */ 559 va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; 560 #endif 561 562 return va; 563 } 564 #endif 565 566 static __inline struct pvo_head * 567 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p) 568 { 569 #ifdef __HAVE_VM_PAGE_MD 570 struct vm_page *pg; 571 572 pg = PHYS_TO_VM_PAGE(pa); 573 if (pg_p != NULL) 574 *pg_p = pg; 575 if (pg == NULL) 576 return &pmap_pvo_unmanaged; 577 return &pg->mdpage.mdpg_pvoh; 578 #endif 579 #ifdef __HAVE_PMAP_PHYSSEG 580 int bank, pg; 581 582 bank = vm_physseg_find(atop(pa), &pg); 583 if (pg_p != NULL) 584 *pg_p = pg; 585 if (bank == -1) 586 return &pmap_pvo_unmanaged; 587 return &vm_physmem[bank].pmseg.pvoh[pg]; 588 #endif 589 } 590 591 static __inline struct pvo_head * 592 vm_page_to_pvoh(struct vm_page *pg) 593 { 594 #ifdef __HAVE_VM_PAGE_MD 595 return &pg->mdpage.mdpg_pvoh; 596 #endif 597 #ifdef __HAVE_PMAP_PHYSSEG 598 return pa_to_pvoh(VM_PAGE_TO_PHYS(pg), NULL); 599 #endif 600 } 601 602 603 #ifdef __HAVE_PMAP_PHYSSEG 604 static __inline char * 605 pa_to_attr(paddr_t pa) 606 { 607 int bank, pg; 608 609 bank = vm_physseg_find(atop(pa), &pg); 610 if (bank == -1) 611 return NULL; 612 return &vm_physmem[bank].pmseg.attrs[pg]; 613 } 614 #endif 615 616 static __inline void 617 pmap_attr_clear(struct vm_page *pg, int ptebit) 618 { 619 #ifdef __HAVE_PMAP_PHYSSEG 620 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) &= ~(ptebit >> ATTR_SHFT); 621 #endif 622 #ifdef __HAVE_VM_PAGE_MD 623 pg->mdpage.mdpg_attrs &= ~ptebit; 624 #endif 625 } 626 627 static __inline int 628 pmap_attr_fetch(struct vm_page *pg) 629 { 630 #ifdef __HAVE_PMAP_PHYSSEG 631 return *pa_to_attr(VM_PAGE_TO_PHYS(pg)) << ATTR_SHFT; 632 #endif 633 #ifdef __HAVE_VM_PAGE_MD 634 return pg->mdpage.mdpg_attrs; 635 #endif 636 } 637 638 static __inline void 639 pmap_attr_save(struct vm_page *pg, int ptebit) 640 { 641 #ifdef __HAVE_PMAP_PHYSSEG 642 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) |= (ptebit >> ATTR_SHFT); 643 #endif 644 #ifdef __HAVE_VM_PAGE_MD 645 pg->mdpage.mdpg_attrs |= ptebit; 646 #endif 647 } 648 649 static __inline int 650 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt) 651 { 652 if (pt->pte_hi == pvo_pt->pte_hi 653 #if 0 654 && ((pt->pte_lo ^ pvo_pt->pte_lo) & 655 ~(PTE_REF|PTE_CHG)) == 0 656 #endif 657 ) 658 return 1; 659 return 0; 660 } 661 662 static __inline void 663 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo) 664 { 665 /* 666 * Construct the PTE. Default to IMB initially. Valid bit 667 * only gets set when the real pte is set in memory. 668 * 669 * Note: Don't set the valid bit for correct operation of tlb update. 670 */ 671 pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT) 672 | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API); 673 pt->pte_lo = pte_lo; 674 } 675 676 static __inline void 677 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt) 678 { 679 pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG); 680 } 681 682 static __inline void 683 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit) 684 { 685 /* 686 * As shown in Section 7.6.3.2.3 687 */ 688 pt->pte_lo &= ~ptebit; 689 TLBIE(va); 690 SYNC(); 691 EIEIO(); 692 TLBSYNC(); 693 SYNC(); 694 } 695 696 static __inline void 697 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt) 698 { 699 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 700 if (pvo_pt->pte_hi & PTE_VALID) 701 panic("pte_set: setting an already valid pte %p", pvo_pt); 702 #endif 703 pvo_pt->pte_hi |= PTE_VALID; 704 /* 705 * Update the PTE as defined in section 7.6.3.1 706 * Note that the REF/CHG bits are from pvo_pt and thus should 707 * have been saved so this routine can restore them (if desired). 708 */ 709 pt->pte_lo = pvo_pt->pte_lo; 710 EIEIO(); 711 pt->pte_hi = pvo_pt->pte_hi; 712 SYNC(); 713 pmap_pte_valid++; 714 } 715 716 static __inline void 717 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 718 { 719 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 720 if ((pvo_pt->pte_hi & PTE_VALID) == 0) 721 panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt); 722 if ((pt->pte_hi & PTE_VALID) == 0) 723 panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt); 724 #endif 725 726 pvo_pt->pte_hi &= ~PTE_VALID; 727 /* 728 * Force the ref & chg bits back into the PTEs. 729 */ 730 SYNC(); 731 /* 732 * Invalidate the pte ... (Section 7.6.3.3) 733 */ 734 pt->pte_hi &= ~PTE_VALID; 735 SYNC(); 736 TLBIE(va); 737 SYNC(); 738 EIEIO(); 739 TLBSYNC(); 740 SYNC(); 741 /* 742 * Save the ref & chg bits ... 743 */ 744 pmap_pte_synch(pt, pvo_pt); 745 pmap_pte_valid--; 746 } 747 748 static __inline void 749 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 750 { 751 /* 752 * Invalidate the PTE 753 */ 754 pmap_pte_unset(pt, pvo_pt, va); 755 pmap_pte_set(pt, pvo_pt); 756 } 757 758 /* 759 * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx 760 * (either primary or secondary location). 761 * 762 * Note: both the destination and source PTEs must not have PTE_VALID set. 763 */ 764 765 STATIC int 766 pmap_pte_insert(int ptegidx, struct pte *pvo_pt) 767 { 768 volatile struct pte *pt; 769 int i; 770 771 #if defined(DEBUG) 772 DPRINTFN(PTE, ("pmap_pte_insert: idx 0x%x, pte 0x%x 0x%x\n", 773 ptegidx, (unsigned int) pvo_pt->pte_hi, (unsigned int) pvo_pt->pte_lo)); 774 #endif 775 /* 776 * First try primary hash. 777 */ 778 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 779 if ((pt->pte_hi & PTE_VALID) == 0) { 780 pvo_pt->pte_hi &= ~PTE_HID; 781 pmap_pte_set(pt, pvo_pt); 782 return i; 783 } 784 } 785 786 /* 787 * Now try secondary hash. 788 */ 789 ptegidx ^= pmap_pteg_mask; 790 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 791 if ((pt->pte_hi & PTE_VALID) == 0) { 792 pvo_pt->pte_hi |= PTE_HID; 793 pmap_pte_set(pt, pvo_pt); 794 return i; 795 } 796 } 797 return -1; 798 } 799 800 /* 801 * Spill handler. 802 * 803 * Tries to spill a page table entry from the overflow area. 804 * This runs in either real mode (if dealing with a exception spill) 805 * or virtual mode when dealing with manually spilling one of the 806 * kernel's pte entries. In either case, interrupts are already 807 * disabled. 808 */ 809 810 int 811 pmap_pte_spill(struct pmap *pm, vaddr_t addr, boolean_t exec) 812 { 813 struct pvo_entry *source_pvo, *victim_pvo, *next_pvo; 814 struct pvo_entry *pvo; 815 /* XXX: gcc -- vpvoh is always set at either *1* or *2* */ 816 struct pvo_tqhead *pvoh, *vpvoh = NULL; 817 int ptegidx, i, j; 818 volatile struct pteg *pteg; 819 volatile struct pte *pt; 820 821 ptegidx = va_to_pteg(pm, addr); 822 823 /* 824 * Have to substitute some entry. Use the primary hash for this. 825 * Use low bits of timebase as random generator. Make sure we are 826 * not picking a kernel pte for replacement. 827 */ 828 pteg = &pmap_pteg_table[ptegidx]; 829 i = MFTB() & 7; 830 for (j = 0; j < 8; j++) { 831 pt = &pteg->pt[i]; 832 if ((pt->pte_hi & PTE_VALID) == 0 || 833 VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT) 834 != KERNEL_VSIDBITS) 835 break; 836 i = (i + 1) & 7; 837 } 838 KASSERT(j < 8); 839 840 source_pvo = NULL; 841 victim_pvo = NULL; 842 pvoh = &pmap_pvo_table[ptegidx]; 843 TAILQ_FOREACH(pvo, pvoh, pvo_olink) { 844 845 /* 846 * We need to find pvo entry for this address... 847 */ 848 PMAP_PVO_CHECK(pvo); /* sanity check */ 849 850 /* 851 * If we haven't found the source and we come to a PVO with 852 * a valid PTE, then we know we can't find it because all 853 * evicted PVOs always are first in the list. 854 */ 855 if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID)) 856 break; 857 if (source_pvo == NULL && pm == pvo->pvo_pmap && 858 addr == PVO_VADDR(pvo)) { 859 860 /* 861 * Now we have found the entry to be spilled into the 862 * pteg. Attempt to insert it into the page table. 863 */ 864 j = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 865 if (j >= 0) { 866 PVO_PTEGIDX_SET(pvo, j); 867 PMAP_PVO_CHECK(pvo); /* sanity check */ 868 PVO_WHERE(pvo, SPILL_INSERT); 869 pvo->pvo_pmap->pm_evictions--; 870 PMAPCOUNT(ptes_spilled); 871 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 872 ? pmap_evcnt_ptes_secondary 873 : pmap_evcnt_ptes_primary)[j]); 874 875 /* 876 * Since we keep the evicted entries at the 877 * from of the PVO list, we need move this 878 * (now resident) PVO after the evicted 879 * entries. 880 */ 881 next_pvo = TAILQ_NEXT(pvo, pvo_olink); 882 883 /* 884 * If we don't have to move (either we were the 885 * last entry or the next entry was valid), 886 * don't change our position. Otherwise 887 * move ourselves to the tail of the queue. 888 */ 889 if (next_pvo != NULL && 890 !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) { 891 TAILQ_REMOVE(pvoh, pvo, pvo_olink); 892 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 893 } 894 return 1; 895 } 896 source_pvo = pvo; 897 if (exec && !PVO_ISEXECUTABLE(source_pvo)) { 898 return 0; 899 } 900 if (victim_pvo != NULL) 901 break; 902 } 903 904 /* 905 * We also need the pvo entry of the victim we are replacing 906 * so save the R & C bits of the PTE. 907 */ 908 if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL && 909 pmap_pte_compare(pt, &pvo->pvo_pte)) { 910 vpvoh = pvoh; /* *1* */ 911 victim_pvo = pvo; 912 if (source_pvo != NULL) 913 break; 914 } 915 } 916 917 if (source_pvo == NULL) { 918 PMAPCOUNT(ptes_unspilled); 919 return 0; 920 } 921 922 if (victim_pvo == NULL) { 923 if ((pt->pte_hi & PTE_HID) == 0) 924 panic("pmap_pte_spill: victim p-pte (%p) has " 925 "no pvo entry!", pt); 926 927 /* 928 * If this is a secondary PTE, we need to search 929 * its primary pvo bucket for the matching PVO. 930 */ 931 vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; /* *2* */ 932 TAILQ_FOREACH(pvo, vpvoh, pvo_olink) { 933 PMAP_PVO_CHECK(pvo); /* sanity check */ 934 935 /* 936 * We also need the pvo entry of the victim we are 937 * replacing so save the R & C bits of the PTE. 938 */ 939 if (pmap_pte_compare(pt, &pvo->pvo_pte)) { 940 victim_pvo = pvo; 941 break; 942 } 943 } 944 if (victim_pvo == NULL) 945 panic("pmap_pte_spill: victim s-pte (%p) has " 946 "no pvo entry!", pt); 947 } 948 949 /* 950 * The victim should be not be a kernel PVO/PTE entry. 951 */ 952 KASSERT(victim_pvo->pvo_pmap != pmap_kernel()); 953 KASSERT(PVO_PTEGIDX_ISSET(victim_pvo)); 954 KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i); 955 956 /* 957 * We are invalidating the TLB entry for the EA for the 958 * we are replacing even though its valid; If we don't 959 * we lose any ref/chg bit changes contained in the TLB 960 * entry. 961 */ 962 source_pvo->pvo_pte.pte_hi &= ~PTE_HID; 963 964 /* 965 * To enforce the PVO list ordering constraint that all 966 * evicted entries should come before all valid entries, 967 * move the source PVO to the tail of its list and the 968 * victim PVO to the head of its list (which might not be 969 * the same list, if the victim was using the secondary hash). 970 */ 971 TAILQ_REMOVE(pvoh, source_pvo, pvo_olink); 972 TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink); 973 TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink); 974 TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink); 975 pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr); 976 pmap_pte_set(pt, &source_pvo->pvo_pte); 977 victim_pvo->pvo_pmap->pm_evictions++; 978 source_pvo->pvo_pmap->pm_evictions--; 979 PVO_WHERE(victim_pvo, SPILL_UNSET); 980 PVO_WHERE(source_pvo, SPILL_SET); 981 982 PVO_PTEGIDX_CLR(victim_pvo); 983 PVO_PTEGIDX_SET(source_pvo, i); 984 PMAPCOUNT2(pmap_evcnt_ptes_primary[i]); 985 PMAPCOUNT(ptes_spilled); 986 PMAPCOUNT(ptes_evicted); 987 PMAPCOUNT(ptes_removed); 988 989 PMAP_PVO_CHECK(victim_pvo); 990 PMAP_PVO_CHECK(source_pvo); 991 return 1; 992 } 993 994 /* 995 * Restrict given range to physical memory 996 */ 997 void 998 pmap_real_memory(paddr_t *start, psize_t *size) 999 { 1000 struct mem_region *mp; 1001 1002 for (mp = mem; mp->size; mp++) { 1003 if (*start + *size > mp->start 1004 && *start < mp->start + mp->size) { 1005 if (*start < mp->start) { 1006 *size -= mp->start - *start; 1007 *start = mp->start; 1008 } 1009 if (*start + *size > mp->start + mp->size) 1010 *size = mp->start + mp->size - *start; 1011 return; 1012 } 1013 } 1014 *size = 0; 1015 } 1016 1017 /* 1018 * Initialize anything else for pmap handling. 1019 * Called during vm_init(). 1020 */ 1021 void 1022 pmap_init(void) 1023 { 1024 #ifdef __HAVE_PMAP_PHYSSEG 1025 struct pvo_tqhead *pvoh; 1026 int bank; 1027 long sz; 1028 char *attr; 1029 1030 pvoh = pmap_physseg.pvoh; 1031 attr = pmap_physseg.attrs; 1032 for (bank = 0; bank < vm_nphysseg; bank++) { 1033 sz = vm_physmem[bank].end - vm_physmem[bank].start; 1034 vm_physmem[bank].pmseg.pvoh = pvoh; 1035 vm_physmem[bank].pmseg.attrs = attr; 1036 for (; sz > 0; sz--, pvoh++, attr++) { 1037 TAILQ_INIT(pvoh); 1038 *attr = 0; 1039 } 1040 } 1041 #endif 1042 1043 pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry), 1044 sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl", 1045 &pmap_pool_mallocator); 1046 1047 pool_setlowat(&pmap_mpvo_pool, 1008); 1048 1049 pmap_initialized = 1; 1050 1051 #ifdef PMAPCOUNTERS 1052 evcnt_attach_static(&pmap_evcnt_mappings); 1053 evcnt_attach_static(&pmap_evcnt_mappings_replaced); 1054 evcnt_attach_static(&pmap_evcnt_unmappings); 1055 1056 evcnt_attach_static(&pmap_evcnt_kernel_mappings); 1057 evcnt_attach_static(&pmap_evcnt_kernel_unmappings); 1058 1059 evcnt_attach_static(&pmap_evcnt_exec_mappings); 1060 evcnt_attach_static(&pmap_evcnt_exec_cached); 1061 evcnt_attach_static(&pmap_evcnt_exec_synced); 1062 evcnt_attach_static(&pmap_evcnt_exec_synced_clear_modify); 1063 1064 evcnt_attach_static(&pmap_evcnt_exec_uncached_page_protect); 1065 evcnt_attach_static(&pmap_evcnt_exec_uncached_clear_modify); 1066 evcnt_attach_static(&pmap_evcnt_exec_uncached_zero_page); 1067 evcnt_attach_static(&pmap_evcnt_exec_uncached_copy_page); 1068 1069 evcnt_attach_static(&pmap_evcnt_zeroed_pages); 1070 evcnt_attach_static(&pmap_evcnt_copied_pages); 1071 evcnt_attach_static(&pmap_evcnt_idlezeroed_pages); 1072 1073 evcnt_attach_static(&pmap_evcnt_updates); 1074 evcnt_attach_static(&pmap_evcnt_collects); 1075 evcnt_attach_static(&pmap_evcnt_copies); 1076 1077 evcnt_attach_static(&pmap_evcnt_ptes_spilled); 1078 evcnt_attach_static(&pmap_evcnt_ptes_unspilled); 1079 evcnt_attach_static(&pmap_evcnt_ptes_evicted); 1080 evcnt_attach_static(&pmap_evcnt_ptes_removed); 1081 evcnt_attach_static(&pmap_evcnt_ptes_changed); 1082 evcnt_attach_static(&pmap_evcnt_ptes_primary[0]); 1083 evcnt_attach_static(&pmap_evcnt_ptes_primary[1]); 1084 evcnt_attach_static(&pmap_evcnt_ptes_primary[2]); 1085 evcnt_attach_static(&pmap_evcnt_ptes_primary[3]); 1086 evcnt_attach_static(&pmap_evcnt_ptes_primary[4]); 1087 evcnt_attach_static(&pmap_evcnt_ptes_primary[5]); 1088 evcnt_attach_static(&pmap_evcnt_ptes_primary[6]); 1089 evcnt_attach_static(&pmap_evcnt_ptes_primary[7]); 1090 evcnt_attach_static(&pmap_evcnt_ptes_secondary[0]); 1091 evcnt_attach_static(&pmap_evcnt_ptes_secondary[1]); 1092 evcnt_attach_static(&pmap_evcnt_ptes_secondary[2]); 1093 evcnt_attach_static(&pmap_evcnt_ptes_secondary[3]); 1094 evcnt_attach_static(&pmap_evcnt_ptes_secondary[4]); 1095 evcnt_attach_static(&pmap_evcnt_ptes_secondary[5]); 1096 evcnt_attach_static(&pmap_evcnt_ptes_secondary[6]); 1097 evcnt_attach_static(&pmap_evcnt_ptes_secondary[7]); 1098 #endif 1099 } 1100 1101 /* 1102 * How much virtual space does the kernel get? 1103 */ 1104 void 1105 pmap_virtual_space(vaddr_t *start, vaddr_t *end) 1106 { 1107 /* 1108 * For now, reserve one segment (minus some overhead) for kernel 1109 * virtual memory 1110 */ 1111 *start = VM_MIN_KERNEL_ADDRESS; 1112 *end = VM_MAX_KERNEL_ADDRESS; 1113 } 1114 1115 /* 1116 * Allocate, initialize, and return a new physical map. 1117 */ 1118 pmap_t 1119 pmap_create(void) 1120 { 1121 pmap_t pm; 1122 1123 pm = pool_get(&pmap_pool, PR_WAITOK); 1124 memset((caddr_t)pm, 0, sizeof *pm); 1125 pmap_pinit(pm); 1126 1127 DPRINTFN(CREATE,("pmap_create: pm %p:\n" 1128 "\t%06x %06x %06x %06x %06x %06x %06x %06x\n" 1129 "\t%06x %06x %06x %06x %06x %06x %06x %06x\n", pm, 1130 (unsigned int) pm->pm_sr[0], (unsigned int) pm->pm_sr[1], 1131 (unsigned int) pm->pm_sr[2], (unsigned int) pm->pm_sr[3], 1132 (unsigned int) pm->pm_sr[4], (unsigned int) pm->pm_sr[5], 1133 (unsigned int) pm->pm_sr[6], (unsigned int) pm->pm_sr[7], 1134 (unsigned int) pm->pm_sr[8], (unsigned int) pm->pm_sr[9], 1135 (unsigned int) pm->pm_sr[10], (unsigned int) pm->pm_sr[11], 1136 (unsigned int) pm->pm_sr[12], (unsigned int) pm->pm_sr[13], 1137 (unsigned int) pm->pm_sr[14], (unsigned int) pm->pm_sr[15])); 1138 return pm; 1139 } 1140 1141 /* 1142 * Initialize a preallocated and zeroed pmap structure. 1143 */ 1144 void 1145 pmap_pinit(pmap_t pm) 1146 { 1147 register_t entropy = MFTB(); 1148 register_t mask; 1149 int i; 1150 1151 /* 1152 * Allocate some segment registers for this pmap. 1153 */ 1154 pm->pm_refs = 1; 1155 for (i = 0; i < NPMAPS; i += VSID_NBPW) { 1156 static register_t pmap_vsidcontext; 1157 register_t hash; 1158 unsigned int n; 1159 1160 /* Create a new value by multiplying by a prime adding in 1161 * entropy from the timebase register. This is to make the 1162 * VSID more random so that the PT Hash function collides 1163 * less often. (note that the prime causes gcc to do shifts 1164 * instead of a multiply) 1165 */ 1166 pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy; 1167 hash = pmap_vsidcontext & (NPMAPS - 1); 1168 if (hash == 0) /* 0 is special, avoid it */ 1169 continue; 1170 n = hash >> 5; 1171 mask = 1L << (hash & (VSID_NBPW-1)); 1172 hash = pmap_vsidcontext; 1173 if (pmap_vsid_bitmap[n] & mask) { /* collision? */ 1174 /* anything free in this bucket? */ 1175 if (~pmap_vsid_bitmap[n] == 0) { 1176 entropy = hash >> PTE_VSID_SHFT; 1177 continue; 1178 } 1179 i = ffs(~pmap_vsid_bitmap[n]) - 1; 1180 mask = 1L << i; 1181 hash &= ~(VSID_NBPW-1); 1182 hash |= i; 1183 } 1184 hash &= PTE_VSID >> PTE_VSID_SHFT; 1185 pmap_vsid_bitmap[n] |= mask; 1186 pm->pm_vsid = hash; 1187 #ifndef PPC_OEA64 1188 for (i = 0; i < 16; i++) 1189 pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY | 1190 SR_NOEXEC; 1191 #endif 1192 return; 1193 } 1194 panic("pmap_pinit: out of segments"); 1195 } 1196 1197 /* 1198 * Add a reference to the given pmap. 1199 */ 1200 void 1201 pmap_reference(pmap_t pm) 1202 { 1203 pm->pm_refs++; 1204 } 1205 1206 /* 1207 * Retire the given pmap from service. 1208 * Should only be called if the map contains no valid mappings. 1209 */ 1210 void 1211 pmap_destroy(pmap_t pm) 1212 { 1213 if (--pm->pm_refs == 0) { 1214 pmap_release(pm); 1215 pool_put(&pmap_pool, pm); 1216 } 1217 } 1218 1219 /* 1220 * Release any resources held by the given physical map. 1221 * Called when a pmap initialized by pmap_pinit is being released. 1222 */ 1223 void 1224 pmap_release(pmap_t pm) 1225 { 1226 int idx, mask; 1227 1228 if (pm->pm_sr[0] == 0) 1229 panic("pmap_release"); 1230 idx = VSID_TO_HASH(pm->pm_vsid) & (NPMAPS-1); 1231 mask = 1 << (idx % VSID_NBPW); 1232 idx /= VSID_NBPW; 1233 pmap_vsid_bitmap[idx] &= ~mask; 1234 } 1235 1236 /* 1237 * Copy the range specified by src_addr/len 1238 * from the source map to the range dst_addr/len 1239 * in the destination map. 1240 * 1241 * This routine is only advisory and need not do anything. 1242 */ 1243 void 1244 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr, 1245 vsize_t len, vaddr_t src_addr) 1246 { 1247 PMAPCOUNT(copies); 1248 } 1249 1250 /* 1251 * Require that all active physical maps contain no 1252 * incorrect entries NOW. 1253 */ 1254 void 1255 pmap_update(struct pmap *pmap) 1256 { 1257 PMAPCOUNT(updates); 1258 TLBSYNC(); 1259 } 1260 1261 /* 1262 * Garbage collects the physical map system for 1263 * pages which are no longer used. 1264 * Success need not be guaranteed -- that is, there 1265 * may well be pages which are not referenced, but 1266 * others may be collected. 1267 * Called by the pageout daemon when pages are scarce. 1268 */ 1269 void 1270 pmap_collect(pmap_t pm) 1271 { 1272 PMAPCOUNT(collects); 1273 } 1274 1275 static __inline int 1276 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx) 1277 { 1278 int pteidx; 1279 /* 1280 * We can find the actual pte entry without searching by 1281 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9] 1282 * and by noticing the HID bit. 1283 */ 1284 pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo); 1285 if (pvo->pvo_pte.pte_hi & PTE_HID) 1286 pteidx ^= pmap_pteg_mask * 8; 1287 return pteidx; 1288 } 1289 1290 volatile struct pte * 1291 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx) 1292 { 1293 volatile struct pte *pt; 1294 1295 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1296 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) 1297 return NULL; 1298 #endif 1299 1300 /* 1301 * If we haven't been supplied the ptegidx, calculate it. 1302 */ 1303 if (pteidx == -1) { 1304 int ptegidx; 1305 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1306 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1307 } 1308 1309 pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7]; 1310 1311 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1312 return pt; 1313 #else 1314 if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) { 1315 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1316 "pvo but no valid pte index", pvo); 1317 } 1318 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) { 1319 panic("pmap_pvo_to_pte: pvo %p: has valid pte index in " 1320 "pvo but no valid pte", pvo); 1321 } 1322 1323 if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) { 1324 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) { 1325 #if defined(DEBUG) || defined(PMAPCHECK) 1326 pmap_pte_print(pt); 1327 #endif 1328 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1329 "pmap_pteg_table %p but invalid in pvo", 1330 pvo, pt); 1331 } 1332 if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) { 1333 #if defined(DEBUG) || defined(PMAPCHECK) 1334 pmap_pte_print(pt); 1335 #endif 1336 panic("pmap_pvo_to_pte: pvo %p: pvo pte does " 1337 "not match pte %p in pmap_pteg_table", 1338 pvo, pt); 1339 } 1340 return pt; 1341 } 1342 1343 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1344 #if defined(DEBUG) || defined(PMAPCHECK) 1345 pmap_pte_print(pt); 1346 #endif 1347 panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in " 1348 "pmap_pteg_table but valid in pvo", pvo, pt); 1349 } 1350 return NULL; 1351 #endif /* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */ 1352 } 1353 1354 struct pvo_entry * 1355 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p) 1356 { 1357 struct pvo_entry *pvo; 1358 int ptegidx; 1359 1360 va &= ~ADDR_POFF; 1361 ptegidx = va_to_pteg(pm, va); 1362 1363 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1364 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1365 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 1366 panic("pmap_pvo_find_va: invalid pvo %p on " 1367 "list %#x (%p)", pvo, ptegidx, 1368 &pmap_pvo_table[ptegidx]); 1369 #endif 1370 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1371 if (pteidx_p) 1372 *pteidx_p = pmap_pvo_pte_index(pvo, ptegidx); 1373 return pvo; 1374 } 1375 } 1376 return NULL; 1377 } 1378 1379 #if defined(DEBUG) || defined(PMAPCHECK) 1380 void 1381 pmap_pvo_check(const struct pvo_entry *pvo) 1382 { 1383 struct pvo_head *pvo_head; 1384 struct pvo_entry *pvo0; 1385 volatile struct pte *pt; 1386 int failed = 0; 1387 1388 if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH) 1389 panic("pmap_pvo_check: pvo %p: invalid address", pvo); 1390 1391 if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) { 1392 printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n", 1393 pvo, pvo->pvo_pmap); 1394 failed = 1; 1395 } 1396 1397 if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH || 1398 (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) { 1399 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1400 pvo, TAILQ_NEXT(pvo, pvo_olink)); 1401 failed = 1; 1402 } 1403 1404 if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH || 1405 (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) { 1406 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1407 pvo, LIST_NEXT(pvo, pvo_vlink)); 1408 failed = 1; 1409 } 1410 1411 if (pvo->pvo_vaddr & PVO_MANAGED) { 1412 pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL); 1413 } else { 1414 if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) { 1415 printf("pmap_pvo_check: pvo %p: non kernel address " 1416 "on kernel unmanaged list\n", pvo); 1417 failed = 1; 1418 } 1419 pvo_head = &pmap_pvo_kunmanaged; 1420 } 1421 LIST_FOREACH(pvo0, pvo_head, pvo_vlink) { 1422 if (pvo0 == pvo) 1423 break; 1424 } 1425 if (pvo0 == NULL) { 1426 printf("pmap_pvo_check: pvo %p: not present " 1427 "on its vlist head %p\n", pvo, pvo_head); 1428 failed = 1; 1429 } 1430 if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) { 1431 printf("pmap_pvo_check: pvo %p: not present " 1432 "on its olist head\n", pvo); 1433 failed = 1; 1434 } 1435 pt = pmap_pvo_to_pte(pvo, -1); 1436 if (pt == NULL) { 1437 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1438 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1439 "no PTE\n", pvo); 1440 failed = 1; 1441 } 1442 } else { 1443 if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] || 1444 (uintptr_t) pt >= 1445 (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) { 1446 printf("pmap_pvo_check: pvo %p: pte %p not in " 1447 "pteg table\n", pvo, pt); 1448 failed = 1; 1449 } 1450 if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) { 1451 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1452 "no PTE\n", pvo); 1453 failed = 1; 1454 } 1455 if (pvo->pvo_pte.pte_hi != pt->pte_hi) { 1456 printf("pmap_pvo_check: pvo %p: pte_hi differ: " 1457 "%#x/%#x\n", pvo, (unsigned int) pvo->pvo_pte.pte_hi, (unsigned int) pt->pte_hi); 1458 failed = 1; 1459 } 1460 if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) & 1461 (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) { 1462 printf("pmap_pvo_check: pvo %p: pte_lo differ: " 1463 "%#x/%#x\n", pvo, 1464 (unsigned int) (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)), 1465 (unsigned int) (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN))); 1466 failed = 1; 1467 } 1468 if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) { 1469 printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#lx" 1470 " doesn't not match PVO's VA %#lx\n", 1471 pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo)); 1472 failed = 1; 1473 } 1474 if (failed) 1475 pmap_pte_print(pt); 1476 } 1477 if (failed) 1478 panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo, 1479 pvo->pvo_pmap); 1480 } 1481 #endif /* DEBUG || PMAPCHECK */ 1482 1483 /* 1484 * This returns whether this is the first mapping of a page. 1485 */ 1486 int 1487 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head, 1488 vaddr_t va, paddr_t pa, register_t pte_lo, int flags) 1489 { 1490 struct pvo_entry *pvo; 1491 struct pvo_tqhead *pvoh; 1492 register_t msr; 1493 int ptegidx; 1494 int i; 1495 int poolflags = PR_NOWAIT; 1496 1497 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1498 if (pmap_pvo_remove_depth > 0) 1499 panic("pmap_pvo_enter: called while pmap_pvo_remove active!"); 1500 if (++pmap_pvo_enter_depth > 1) 1501 panic("pmap_pvo_enter: called recursively!"); 1502 #endif 1503 1504 /* 1505 * Compute the PTE Group index. 1506 */ 1507 va &= ~ADDR_POFF; 1508 ptegidx = va_to_pteg(pm, va); 1509 1510 msr = pmap_interrupts_off(); 1511 /* 1512 * Remove any existing mapping for this page. Reuse the 1513 * pvo entry if there a mapping. 1514 */ 1515 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1516 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1517 #ifdef DEBUG 1518 if ((pmapdebug & PMAPDEBUG_PVOENTER) && 1519 ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) & 1520 ~(PTE_REF|PTE_CHG)) == 0 && 1521 va < VM_MIN_KERNEL_ADDRESS) { 1522 printf("pmap_pvo_enter: pvo %p: dup %#x/%#lx\n", 1523 pvo, (unsigned int) pvo->pvo_pte.pte_lo, (unsigned int) pte_lo|pa); 1524 printf("pmap_pvo_enter: pte_hi=%#x sr=%#x\n", 1525 (unsigned int) pvo->pvo_pte.pte_hi, 1526 (unsigned int) pm->pm_sr[va >> ADDR_SR_SHFT]); 1527 pmap_pte_print(pmap_pvo_to_pte(pvo, -1)); 1528 #ifdef DDBX 1529 Debugger(); 1530 #endif 1531 } 1532 #endif 1533 PMAPCOUNT(mappings_replaced); 1534 pmap_pvo_remove(pvo, -1); 1535 break; 1536 } 1537 } 1538 1539 /* 1540 * If we aren't overwriting an mapping, try to allocate 1541 */ 1542 pmap_interrupts_restore(msr); 1543 pvo = pool_get(pl, poolflags); 1544 msr = pmap_interrupts_off(); 1545 if (pvo == NULL) { 1546 pvo = pmap_pvo_reclaim(pm); 1547 if (pvo == NULL) { 1548 if ((flags & PMAP_CANFAIL) == 0) 1549 panic("pmap_pvo_enter: failed"); 1550 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1551 pmap_pvo_enter_depth--; 1552 #endif 1553 pmap_interrupts_restore(msr); 1554 return ENOMEM; 1555 } 1556 } 1557 pvo->pvo_vaddr = va; 1558 pvo->pvo_pmap = pm; 1559 pvo->pvo_vaddr &= ~ADDR_POFF; 1560 if (flags & VM_PROT_EXECUTE) { 1561 PMAPCOUNT(exec_mappings); 1562 pvo_set_exec(pvo); 1563 } 1564 if (flags & PMAP_WIRED) 1565 pvo->pvo_vaddr |= PVO_WIRED; 1566 if (pvo_head != &pmap_pvo_kunmanaged) { 1567 pvo->pvo_vaddr |= PVO_MANAGED; 1568 PMAPCOUNT(mappings); 1569 } else { 1570 PMAPCOUNT(kernel_mappings); 1571 } 1572 pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo); 1573 1574 LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink); 1575 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1576 pvo->pvo_pmap->pm_stats.wired_count++; 1577 pvo->pvo_pmap->pm_stats.resident_count++; 1578 #if defined(DEBUG) 1579 if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) 1580 DPRINTFN(PVOENTER, 1581 ("pmap_pvo_enter: pvo %p: pm %p va %#lx pa %#lx\n", 1582 pvo, pm, va, pa)); 1583 #endif 1584 1585 /* 1586 * We hope this succeeds but it isn't required. 1587 */ 1588 pvoh = &pmap_pvo_table[ptegidx]; 1589 i = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 1590 if (i >= 0) { 1591 PVO_PTEGIDX_SET(pvo, i); 1592 PVO_WHERE(pvo, ENTER_INSERT); 1593 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 1594 ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]); 1595 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 1596 } else { 1597 /* 1598 * Since we didn't have room for this entry (which makes it 1599 * and evicted entry), place it at the head of the list. 1600 */ 1601 TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink); 1602 PMAPCOUNT(ptes_evicted); 1603 pm->pm_evictions++; 1604 /* 1605 * If this is a kernel page, make sure it's active. 1606 */ 1607 if (pm == pmap_kernel()) { 1608 i = pmap_pte_spill(pm, va, FALSE); 1609 KASSERT(i); 1610 } 1611 } 1612 PMAP_PVO_CHECK(pvo); /* sanity check */ 1613 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1614 pmap_pvo_enter_depth--; 1615 #endif 1616 pmap_interrupts_restore(msr); 1617 return 0; 1618 } 1619 1620 void 1621 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx) 1622 { 1623 volatile struct pte *pt; 1624 int ptegidx; 1625 1626 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1627 if (++pmap_pvo_remove_depth > 1) 1628 panic("pmap_pvo_remove: called recursively!"); 1629 #endif 1630 1631 /* 1632 * If we haven't been supplied the ptegidx, calculate it. 1633 */ 1634 if (pteidx == -1) { 1635 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1636 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1637 } else { 1638 ptegidx = pteidx >> 3; 1639 if (pvo->pvo_pte.pte_hi & PTE_HID) 1640 ptegidx ^= pmap_pteg_mask; 1641 } 1642 PMAP_PVO_CHECK(pvo); /* sanity check */ 1643 1644 /* 1645 * If there is an active pte entry, we need to deactivate it 1646 * (and save the ref & chg bits). 1647 */ 1648 pt = pmap_pvo_to_pte(pvo, pteidx); 1649 if (pt != NULL) { 1650 pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 1651 PVO_WHERE(pvo, REMOVE); 1652 PVO_PTEGIDX_CLR(pvo); 1653 PMAPCOUNT(ptes_removed); 1654 } else { 1655 KASSERT(pvo->pvo_pmap->pm_evictions > 0); 1656 pvo->pvo_pmap->pm_evictions--; 1657 } 1658 1659 /* 1660 * Account for executable mappings. 1661 */ 1662 if (PVO_ISEXECUTABLE(pvo)) 1663 pvo_clear_exec(pvo); 1664 1665 /* 1666 * Update our statistics. 1667 */ 1668 pvo->pvo_pmap->pm_stats.resident_count--; 1669 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1670 pvo->pvo_pmap->pm_stats.wired_count--; 1671 1672 /* 1673 * Save the REF/CHG bits into their cache if the page is managed. 1674 */ 1675 if (pvo->pvo_vaddr & PVO_MANAGED) { 1676 register_t ptelo = pvo->pvo_pte.pte_lo; 1677 struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN); 1678 1679 if (pg != NULL) { 1680 pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG)); 1681 } 1682 PMAPCOUNT(unmappings); 1683 } else { 1684 PMAPCOUNT(kernel_unmappings); 1685 } 1686 1687 /* 1688 * Remove the PVO from its lists and return it to the pool. 1689 */ 1690 LIST_REMOVE(pvo, pvo_vlink); 1691 TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink); 1692 pool_put(pvo->pvo_vaddr & PVO_MANAGED 1693 ? &pmap_mpvo_pool : &pmap_upvo_pool, pvo); 1694 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1695 pmap_pvo_remove_depth--; 1696 #endif 1697 } 1698 1699 /* 1700 * Mark a mapping as executable. 1701 * If this is the first executable mapping in the segment, 1702 * clear the noexec flag. 1703 */ 1704 STATIC void 1705 pvo_set_exec(struct pvo_entry *pvo) 1706 { 1707 struct pmap *pm = pvo->pvo_pmap; 1708 1709 if (pm == pmap_kernel() || PVO_ISEXECUTABLE(pvo)) { 1710 return; 1711 } 1712 pvo->pvo_vaddr |= PVO_EXECUTABLE; 1713 #ifdef PPC_OEA 1714 { 1715 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1716 if (pm->pm_exec[sr]++ == 0) { 1717 pm->pm_sr[sr] &= ~SR_NOEXEC; 1718 } 1719 } 1720 #endif 1721 } 1722 1723 /* 1724 * Mark a mapping as non-executable. 1725 * If this was the last executable mapping in the segment, 1726 * set the noexec flag. 1727 */ 1728 STATIC void 1729 pvo_clear_exec(struct pvo_entry *pvo) 1730 { 1731 struct pmap *pm = pvo->pvo_pmap; 1732 1733 if (pm == pmap_kernel() || !PVO_ISEXECUTABLE(pvo)) { 1734 return; 1735 } 1736 pvo->pvo_vaddr &= ~PVO_EXECUTABLE; 1737 #ifdef PPC_OEA 1738 { 1739 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1740 if (--pm->pm_exec[sr] == 0) { 1741 pm->pm_sr[sr] |= SR_NOEXEC; 1742 } 1743 } 1744 #endif 1745 } 1746 1747 /* 1748 * Insert physical page at pa into the given pmap at virtual address va. 1749 */ 1750 int 1751 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags) 1752 { 1753 struct mem_region *mp; 1754 struct pvo_head *pvo_head; 1755 struct vm_page *pg; 1756 struct pool *pl; 1757 register_t pte_lo; 1758 int error; 1759 u_int pvo_flags; 1760 u_int was_exec = 0; 1761 1762 if (__predict_false(!pmap_initialized)) { 1763 pvo_head = &pmap_pvo_kunmanaged; 1764 pl = &pmap_upvo_pool; 1765 pvo_flags = 0; 1766 pg = NULL; 1767 was_exec = PTE_EXEC; 1768 } else { 1769 pvo_head = pa_to_pvoh(pa, &pg); 1770 pl = &pmap_mpvo_pool; 1771 pvo_flags = PVO_MANAGED; 1772 } 1773 1774 DPRINTFN(ENTER, 1775 ("pmap_enter(%p, 0x%lx, 0x%lx, 0x%x, 0x%x):", 1776 pm, va, pa, prot, flags)); 1777 1778 /* 1779 * If this is a managed page, and it's the first reference to the 1780 * page clear the execness of the page. Otherwise fetch the execness. 1781 */ 1782 if (pg != NULL) 1783 was_exec = pmap_attr_fetch(pg) & PTE_EXEC; 1784 1785 DPRINTFN(ENTER, (" was_exec=%d", was_exec)); 1786 1787 /* 1788 * Assume the page is cache inhibited and access is guarded unless 1789 * it's in our available memory array. If it is in the memory array, 1790 * asssume it's in memory coherent memory. 1791 */ 1792 pte_lo = PTE_IG; 1793 if ((flags & PMAP_NC) == 0) { 1794 for (mp = mem; mp->size; mp++) { 1795 if (pa >= mp->start && pa < mp->start + mp->size) { 1796 pte_lo = PTE_M; 1797 break; 1798 } 1799 } 1800 } 1801 1802 if (prot & VM_PROT_WRITE) 1803 pte_lo |= PTE_BW; 1804 else 1805 pte_lo |= PTE_BR; 1806 1807 /* 1808 * If this was in response to a fault, "pre-fault" the PTE's 1809 * changed/referenced bit appropriately. 1810 */ 1811 if (flags & VM_PROT_WRITE) 1812 pte_lo |= PTE_CHG; 1813 if (flags & (VM_PROT_READ|VM_PROT_WRITE)) 1814 pte_lo |= PTE_REF; 1815 1816 /* 1817 * We need to know if this page can be executable 1818 */ 1819 flags |= (prot & VM_PROT_EXECUTE); 1820 1821 /* 1822 * Record mapping for later back-translation and pte spilling. 1823 * This will overwrite any existing mapping. 1824 */ 1825 error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags); 1826 1827 /* 1828 * Flush the real page from the instruction cache if this page is 1829 * mapped executable and cacheable and has not been flushed since 1830 * the last time it was modified. 1831 */ 1832 if (error == 0 && 1833 (flags & VM_PROT_EXECUTE) && 1834 (pte_lo & PTE_I) == 0 && 1835 was_exec == 0) { 1836 DPRINTFN(ENTER, (" syncicache")); 1837 PMAPCOUNT(exec_synced); 1838 pmap_syncicache(pa, PAGE_SIZE); 1839 if (pg != NULL) { 1840 pmap_attr_save(pg, PTE_EXEC); 1841 PMAPCOUNT(exec_cached); 1842 #if defined(DEBUG) || defined(PMAPDEBUG) 1843 if (pmapdebug & PMAPDEBUG_ENTER) 1844 printf(" marked-as-exec"); 1845 else if (pmapdebug & PMAPDEBUG_EXEC) 1846 printf("[pmap_enter: %#lx: marked-as-exec]\n", 1847 pg->phys_addr); 1848 1849 #endif 1850 } 1851 } 1852 1853 DPRINTFN(ENTER, (": error=%d\n", error)); 1854 1855 return error; 1856 } 1857 1858 void 1859 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot) 1860 { 1861 struct mem_region *mp; 1862 register_t pte_lo; 1863 int error; 1864 1865 if (va < VM_MIN_KERNEL_ADDRESS) 1866 panic("pmap_kenter_pa: attempt to enter " 1867 "non-kernel address %#lx!", va); 1868 1869 DPRINTFN(KENTER, 1870 ("pmap_kenter_pa(%#lx,%#lx,%#x)\n", va, pa, prot)); 1871 1872 /* 1873 * Assume the page is cache inhibited and access is guarded unless 1874 * it's in our available memory array. If it is in the memory array, 1875 * asssume it's in memory coherent memory. 1876 */ 1877 pte_lo = PTE_IG; 1878 if ((prot & PMAP_NC) == 0) { 1879 for (mp = mem; mp->size; mp++) { 1880 if (pa >= mp->start && pa < mp->start + mp->size) { 1881 pte_lo = PTE_M; 1882 break; 1883 } 1884 } 1885 } 1886 1887 if (prot & VM_PROT_WRITE) 1888 pte_lo |= PTE_BW; 1889 else 1890 pte_lo |= PTE_BR; 1891 1892 /* 1893 * We don't care about REF/CHG on PVOs on the unmanaged list. 1894 */ 1895 error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool, 1896 &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED); 1897 1898 if (error != 0) 1899 panic("pmap_kenter_pa: failed to enter va %#lx pa %#lx: %d", 1900 va, pa, error); 1901 } 1902 1903 void 1904 pmap_kremove(vaddr_t va, vsize_t len) 1905 { 1906 if (va < VM_MIN_KERNEL_ADDRESS) 1907 panic("pmap_kremove: attempt to remove " 1908 "non-kernel address %#lx!", va); 1909 1910 DPRINTFN(KREMOVE,("pmap_kremove(%#lx,%#lx)\n", va, len)); 1911 pmap_remove(pmap_kernel(), va, va + len); 1912 } 1913 1914 /* 1915 * Remove the given range of mapping entries. 1916 */ 1917 void 1918 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva) 1919 { 1920 struct pvo_entry *pvo; 1921 register_t msr; 1922 int pteidx; 1923 1924 msr = pmap_interrupts_off(); 1925 for (; va < endva; va += PAGE_SIZE) { 1926 pvo = pmap_pvo_find_va(pm, va, &pteidx); 1927 if (pvo != NULL) { 1928 pmap_pvo_remove(pvo, pteidx); 1929 } 1930 } 1931 pmap_interrupts_restore(msr); 1932 } 1933 1934 /* 1935 * Get the physical page address for the given pmap/virtual address. 1936 */ 1937 boolean_t 1938 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap) 1939 { 1940 struct pvo_entry *pvo; 1941 register_t msr; 1942 1943 /* 1944 * If this is a kernel pmap lookup, also check the battable 1945 * and if we get a hit, translate the VA to a PA using the 1946 * BAT entries. Don't check for VM_MAX_KENREL_ADDRESS is 1947 * that will wrap back to 0. 1948 */ 1949 if (pm == pmap_kernel() && 1950 (va < VM_MIN_KERNEL_ADDRESS || 1951 (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) { 1952 register_t batu = battable[va >> ADDR_SR_SHFT].batu; 1953 KASSERT((va >> ADDR_SR_SHFT) != USER_SR); 1954 if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) { 1955 register_t batl = battable[va >> ADDR_SR_SHFT].batl; 1956 register_t mask = (~(batu & BAT_BL) << 15) & ~0x1ffffL; 1957 *pap = (batl & mask) | (va & ~mask); 1958 return TRUE; 1959 } 1960 return FALSE; 1961 } 1962 1963 msr = pmap_interrupts_off(); 1964 pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL); 1965 if (pvo != NULL) { 1966 PMAP_PVO_CHECK(pvo); /* sanity check */ 1967 *pap = (pvo->pvo_pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF); 1968 } 1969 pmap_interrupts_restore(msr); 1970 return pvo != NULL; 1971 } 1972 1973 /* 1974 * Lower the protection on the specified range of this pmap. 1975 */ 1976 void 1977 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot) 1978 { 1979 struct pvo_entry *pvo; 1980 volatile struct pte *pt; 1981 register_t msr; 1982 int pteidx; 1983 1984 /* 1985 * Since this routine only downgrades protection, we should 1986 * always be called with at least one bit not set. 1987 */ 1988 KASSERT(prot != VM_PROT_ALL); 1989 1990 /* 1991 * If there is no protection, this is equivalent to 1992 * remove the pmap from the pmap. 1993 */ 1994 if ((prot & VM_PROT_READ) == 0) { 1995 pmap_remove(pm, va, endva); 1996 return; 1997 } 1998 1999 msr = pmap_interrupts_off(); 2000 for (; va < endva; va += PAGE_SIZE) { 2001 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2002 if (pvo == NULL) 2003 continue; 2004 PMAP_PVO_CHECK(pvo); /* sanity check */ 2005 2006 /* 2007 * Revoke executable if asked to do so. 2008 */ 2009 if ((prot & VM_PROT_EXECUTE) == 0) 2010 pvo_clear_exec(pvo); 2011 2012 #if 0 2013 /* 2014 * If the page is already read-only, no change 2015 * needs to be made. 2016 */ 2017 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) 2018 continue; 2019 #endif 2020 /* 2021 * Grab the PTE pointer before we diddle with 2022 * the cached PTE copy. 2023 */ 2024 pt = pmap_pvo_to_pte(pvo, pteidx); 2025 /* 2026 * Change the protection of the page. 2027 */ 2028 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2029 pvo->pvo_pte.pte_lo |= PTE_BR; 2030 2031 /* 2032 * If the PVO is in the page table, update 2033 * that pte at well. 2034 */ 2035 if (pt != NULL) { 2036 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2037 PVO_WHERE(pvo, PMAP_PROTECT); 2038 PMAPCOUNT(ptes_changed); 2039 } 2040 2041 PMAP_PVO_CHECK(pvo); /* sanity check */ 2042 } 2043 pmap_interrupts_restore(msr); 2044 } 2045 2046 void 2047 pmap_unwire(pmap_t pm, vaddr_t va) 2048 { 2049 struct pvo_entry *pvo; 2050 register_t msr; 2051 2052 msr = pmap_interrupts_off(); 2053 pvo = pmap_pvo_find_va(pm, va, NULL); 2054 if (pvo != NULL) { 2055 if (pvo->pvo_vaddr & PVO_WIRED) { 2056 pvo->pvo_vaddr &= ~PVO_WIRED; 2057 pm->pm_stats.wired_count--; 2058 } 2059 PMAP_PVO_CHECK(pvo); /* sanity check */ 2060 } 2061 pmap_interrupts_restore(msr); 2062 } 2063 2064 /* 2065 * Lower the protection on the specified physical page. 2066 */ 2067 void 2068 pmap_page_protect(struct vm_page *pg, vm_prot_t prot) 2069 { 2070 struct pvo_head *pvo_head; 2071 struct pvo_entry *pvo, *next_pvo; 2072 volatile struct pte *pt; 2073 register_t msr; 2074 2075 KASSERT(prot != VM_PROT_ALL); 2076 msr = pmap_interrupts_off(); 2077 2078 /* 2079 * When UVM reuses a page, it does a pmap_page_protect with 2080 * VM_PROT_NONE. At that point, we can clear the exec flag 2081 * since we know the page will have different contents. 2082 */ 2083 if ((prot & VM_PROT_READ) == 0) { 2084 DPRINTFN(EXEC, ("[pmap_page_protect: %#lx: clear-exec]\n", 2085 pg->phys_addr)); 2086 if (pmap_attr_fetch(pg) & PTE_EXEC) { 2087 PMAPCOUNT(exec_uncached_page_protect); 2088 pmap_attr_clear(pg, PTE_EXEC); 2089 } 2090 } 2091 2092 pvo_head = vm_page_to_pvoh(pg); 2093 for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) { 2094 next_pvo = LIST_NEXT(pvo, pvo_vlink); 2095 PMAP_PVO_CHECK(pvo); /* sanity check */ 2096 2097 /* 2098 * Downgrading to no mapping at all, we just remove the entry. 2099 */ 2100 if ((prot & VM_PROT_READ) == 0) { 2101 pmap_pvo_remove(pvo, -1); 2102 continue; 2103 } 2104 2105 /* 2106 * If EXEC permission is being revoked, just clear the 2107 * flag in the PVO. 2108 */ 2109 if ((prot & VM_PROT_EXECUTE) == 0) 2110 pvo_clear_exec(pvo); 2111 2112 /* 2113 * If this entry is already RO, don't diddle with the 2114 * page table. 2115 */ 2116 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) { 2117 PMAP_PVO_CHECK(pvo); 2118 continue; 2119 } 2120 2121 /* 2122 * Grab the PTE before the we diddle the bits so 2123 * pvo_to_pte can verify the pte contents are as 2124 * expected. 2125 */ 2126 pt = pmap_pvo_to_pte(pvo, -1); 2127 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2128 pvo->pvo_pte.pte_lo |= PTE_BR; 2129 if (pt != NULL) { 2130 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2131 PVO_WHERE(pvo, PMAP_PAGE_PROTECT); 2132 PMAPCOUNT(ptes_changed); 2133 } 2134 PMAP_PVO_CHECK(pvo); /* sanity check */ 2135 } 2136 pmap_interrupts_restore(msr); 2137 } 2138 2139 /* 2140 * Activate the address space for the specified process. If the process 2141 * is the current process, load the new MMU context. 2142 */ 2143 void 2144 pmap_activate(struct lwp *l) 2145 { 2146 struct pcb *pcb = &l->l_addr->u_pcb; 2147 pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap; 2148 2149 DPRINTFN(ACTIVATE, 2150 ("pmap_activate: lwp %p (curlwp %p)\n", l, curlwp)); 2151 2152 /* 2153 * XXX Normally performed in cpu_fork(). 2154 */ 2155 pcb->pcb_pm = pmap; 2156 2157 /* 2158 * In theory, the SR registers need only be valid on return 2159 * to user space wait to do them there. 2160 */ 2161 if (l == curlwp) { 2162 /* Store pointer to new current pmap. */ 2163 curpm = pmap; 2164 } 2165 } 2166 2167 /* 2168 * Deactivate the specified process's address space. 2169 */ 2170 void 2171 pmap_deactivate(struct lwp *l) 2172 { 2173 } 2174 2175 boolean_t 2176 pmap_query_bit(struct vm_page *pg, int ptebit) 2177 { 2178 struct pvo_entry *pvo; 2179 volatile struct pte *pt; 2180 register_t msr; 2181 2182 if (pmap_attr_fetch(pg) & ptebit) 2183 return TRUE; 2184 2185 msr = pmap_interrupts_off(); 2186 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2187 PMAP_PVO_CHECK(pvo); /* sanity check */ 2188 /* 2189 * See if we saved the bit off. If so cache, it and return 2190 * success. 2191 */ 2192 if (pvo->pvo_pte.pte_lo & ptebit) { 2193 pmap_attr_save(pg, ptebit); 2194 PMAP_PVO_CHECK(pvo); /* sanity check */ 2195 pmap_interrupts_restore(msr); 2196 return TRUE; 2197 } 2198 } 2199 /* 2200 * No luck, now go thru the hard part of looking at the ptes 2201 * themselves. Sync so any pending REF/CHG bits are flushed 2202 * to the PTEs. 2203 */ 2204 SYNC(); 2205 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2206 PMAP_PVO_CHECK(pvo); /* sanity check */ 2207 /* 2208 * See if this pvo have a valid PTE. If so, fetch the 2209 * REF/CHG bits from the valid PTE. If the appropriate 2210 * ptebit is set, cache, it and return success. 2211 */ 2212 pt = pmap_pvo_to_pte(pvo, -1); 2213 if (pt != NULL) { 2214 pmap_pte_synch(pt, &pvo->pvo_pte); 2215 if (pvo->pvo_pte.pte_lo & ptebit) { 2216 pmap_attr_save(pg, ptebit); 2217 PMAP_PVO_CHECK(pvo); /* sanity check */ 2218 pmap_interrupts_restore(msr); 2219 return TRUE; 2220 } 2221 } 2222 } 2223 pmap_interrupts_restore(msr); 2224 return FALSE; 2225 } 2226 2227 boolean_t 2228 pmap_clear_bit(struct vm_page *pg, int ptebit) 2229 { 2230 struct pvo_head *pvoh = vm_page_to_pvoh(pg); 2231 struct pvo_entry *pvo; 2232 volatile struct pte *pt; 2233 register_t msr; 2234 int rv = 0; 2235 2236 msr = pmap_interrupts_off(); 2237 2238 /* 2239 * Fetch the cache value 2240 */ 2241 rv |= pmap_attr_fetch(pg); 2242 2243 /* 2244 * Clear the cached value. 2245 */ 2246 pmap_attr_clear(pg, ptebit); 2247 2248 /* 2249 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we 2250 * can reset the right ones). Note that since the pvo entries and 2251 * list heads are accessed via BAT0 and are never placed in the 2252 * page table, we don't have to worry about further accesses setting 2253 * the REF/CHG bits. 2254 */ 2255 SYNC(); 2256 2257 /* 2258 * For each pvo entry, clear pvo's ptebit. If this pvo have a 2259 * valid PTE. If so, clear the ptebit from the valid PTE. 2260 */ 2261 LIST_FOREACH(pvo, pvoh, pvo_vlink) { 2262 PMAP_PVO_CHECK(pvo); /* sanity check */ 2263 pt = pmap_pvo_to_pte(pvo, -1); 2264 if (pt != NULL) { 2265 /* 2266 * Only sync the PTE if the bit we are looking 2267 * for is not already set. 2268 */ 2269 if ((pvo->pvo_pte.pte_lo & ptebit) == 0) 2270 pmap_pte_synch(pt, &pvo->pvo_pte); 2271 /* 2272 * If the bit we are looking for was already set, 2273 * clear that bit in the pte. 2274 */ 2275 if (pvo->pvo_pte.pte_lo & ptebit) 2276 pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit); 2277 } 2278 rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF); 2279 pvo->pvo_pte.pte_lo &= ~ptebit; 2280 PMAP_PVO_CHECK(pvo); /* sanity check */ 2281 } 2282 pmap_interrupts_restore(msr); 2283 2284 /* 2285 * If we are clearing the modify bit and this page was marked EXEC 2286 * and the user of the page thinks the page was modified, then we 2287 * need to clean it from the icache if it's mapped or clear the EXEC 2288 * bit if it's not mapped. The page itself might not have the CHG 2289 * bit set if the modification was done via DMA to the page. 2290 */ 2291 if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) { 2292 if (LIST_EMPTY(pvoh)) { 2293 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: clear-exec]\n", 2294 pg->phys_addr)); 2295 pmap_attr_clear(pg, PTE_EXEC); 2296 PMAPCOUNT(exec_uncached_clear_modify); 2297 } else { 2298 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: syncicache]\n", 2299 pg->phys_addr)); 2300 pmap_syncicache(pg->phys_addr, PAGE_SIZE); 2301 PMAPCOUNT(exec_synced_clear_modify); 2302 } 2303 } 2304 return (rv & ptebit) != 0; 2305 } 2306 2307 void 2308 pmap_procwr(struct proc *p, vaddr_t va, size_t len) 2309 { 2310 struct pvo_entry *pvo; 2311 size_t offset = va & ADDR_POFF; 2312 int s; 2313 2314 s = splvm(); 2315 while (len > 0) { 2316 size_t seglen = PAGE_SIZE - offset; 2317 if (seglen > len) 2318 seglen = len; 2319 pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL); 2320 if (pvo != NULL && PVO_ISEXECUTABLE(pvo)) { 2321 pmap_syncicache( 2322 (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen); 2323 PMAP_PVO_CHECK(pvo); 2324 } 2325 va += seglen; 2326 len -= seglen; 2327 offset = 0; 2328 } 2329 splx(s); 2330 } 2331 2332 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 2333 void 2334 pmap_pte_print(volatile struct pte *pt) 2335 { 2336 printf("PTE %p: ", pt); 2337 /* High word: */ 2338 printf("0x%08lx: [", pt->pte_hi); 2339 printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i'); 2340 printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-'); 2341 printf("0x%06lx 0x%02lx", 2342 (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT, 2343 pt->pte_hi & PTE_API); 2344 printf(" (va 0x%08lx)] ", pmap_pte_to_va(pt)); 2345 /* Low word: */ 2346 printf(" 0x%08lx: [", pt->pte_lo); 2347 printf("0x%05lx... ", pt->pte_lo >> 12); 2348 printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u'); 2349 printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n'); 2350 printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.'); 2351 printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.'); 2352 printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.'); 2353 printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.'); 2354 switch (pt->pte_lo & PTE_PP) { 2355 case PTE_BR: printf("br]\n"); break; 2356 case PTE_BW: printf("bw]\n"); break; 2357 case PTE_SO: printf("so]\n"); break; 2358 case PTE_SW: printf("sw]\n"); break; 2359 } 2360 } 2361 #endif 2362 2363 #if defined(DDB) 2364 void 2365 pmap_pteg_check(void) 2366 { 2367 volatile struct pte *pt; 2368 int i; 2369 int ptegidx; 2370 u_int p_valid = 0; 2371 u_int s_valid = 0; 2372 u_int invalid = 0; 2373 2374 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2375 for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) { 2376 if (pt->pte_hi & PTE_VALID) { 2377 if (pt->pte_hi & PTE_HID) 2378 s_valid++; 2379 else 2380 p_valid++; 2381 } else 2382 invalid++; 2383 } 2384 } 2385 printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n", 2386 p_valid, p_valid, s_valid, s_valid, 2387 invalid, invalid); 2388 } 2389 2390 void 2391 pmap_print_mmuregs(void) 2392 { 2393 int i; 2394 u_int cpuvers; 2395 #ifndef PPC_OEA64 2396 vaddr_t addr; 2397 register_t soft_sr[16]; 2398 #endif 2399 struct bat soft_ibat[4]; 2400 struct bat soft_dbat[4]; 2401 register_t sdr1; 2402 2403 cpuvers = MFPVR() >> 16; 2404 2405 __asm __volatile ("mfsdr1 %0" : "=r"(sdr1)); 2406 #ifndef PPC_OEA64 2407 addr = 0; 2408 for (i=0; i<16; i++) { 2409 soft_sr[i] = MFSRIN(addr); 2410 addr += (1 << ADDR_SR_SHFT); 2411 } 2412 #endif 2413 2414 /* read iBAT (601: uBAT) registers */ 2415 __asm __volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu)); 2416 __asm __volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl)); 2417 __asm __volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu)); 2418 __asm __volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl)); 2419 __asm __volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu)); 2420 __asm __volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl)); 2421 __asm __volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu)); 2422 __asm __volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl)); 2423 2424 2425 if (cpuvers != MPC601) { 2426 /* read dBAT registers */ 2427 __asm __volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu)); 2428 __asm __volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl)); 2429 __asm __volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu)); 2430 __asm __volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl)); 2431 __asm __volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu)); 2432 __asm __volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl)); 2433 __asm __volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu)); 2434 __asm __volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl)); 2435 } 2436 2437 printf("SDR1:\t0x%lx\n", (long) sdr1); 2438 #ifndef PPC_OEA64 2439 printf("SR[]:\t"); 2440 for (i=0; i<4; i++) 2441 printf("0x%08lx, ", soft_sr[i]); 2442 printf("\n\t"); 2443 for ( ; i<8; i++) 2444 printf("0x%08lx, ", soft_sr[i]); 2445 printf("\n\t"); 2446 for ( ; i<12; i++) 2447 printf("0x%08lx, ", soft_sr[i]); 2448 printf("\n\t"); 2449 for ( ; i<16; i++) 2450 printf("0x%08lx, ", soft_sr[i]); 2451 printf("\n"); 2452 #endif 2453 2454 printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i'); 2455 for (i=0; i<4; i++) { 2456 printf("0x%08lx 0x%08lx, ", 2457 soft_ibat[i].batu, soft_ibat[i].batl); 2458 if (i == 1) 2459 printf("\n\t"); 2460 } 2461 if (cpuvers != MPC601) { 2462 printf("\ndBAT[]:\t"); 2463 for (i=0; i<4; i++) { 2464 printf("0x%08lx 0x%08lx, ", 2465 soft_dbat[i].batu, soft_dbat[i].batl); 2466 if (i == 1) 2467 printf("\n\t"); 2468 } 2469 } 2470 printf("\n"); 2471 } 2472 2473 void 2474 pmap_print_pte(pmap_t pm, vaddr_t va) 2475 { 2476 struct pvo_entry *pvo; 2477 volatile struct pte *pt; 2478 int pteidx; 2479 2480 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2481 if (pvo != NULL) { 2482 pt = pmap_pvo_to_pte(pvo, pteidx); 2483 if (pt != NULL) { 2484 printf("VA %#lx -> %p -> %s %#lx, %#lx\n", 2485 va, pt, 2486 pt->pte_hi & PTE_HID ? "(sec)" : "(pri)", 2487 pt->pte_hi, pt->pte_lo); 2488 } else { 2489 printf("No valid PTE found\n"); 2490 } 2491 } else { 2492 printf("Address not in pmap\n"); 2493 } 2494 } 2495 2496 void 2497 pmap_pteg_dist(void) 2498 { 2499 struct pvo_entry *pvo; 2500 int ptegidx; 2501 int depth; 2502 int max_depth = 0; 2503 unsigned int depths[64]; 2504 2505 memset(depths, 0, sizeof(depths)); 2506 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2507 depth = 0; 2508 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2509 depth++; 2510 } 2511 if (depth > max_depth) 2512 max_depth = depth; 2513 if (depth > 63) 2514 depth = 63; 2515 depths[depth]++; 2516 } 2517 2518 for (depth = 0; depth < 64; depth++) { 2519 printf(" [%2d]: %8u", depth, depths[depth]); 2520 if ((depth & 3) == 3) 2521 printf("\n"); 2522 if (depth == max_depth) 2523 break; 2524 } 2525 if ((depth & 3) != 3) 2526 printf("\n"); 2527 printf("Max depth found was %d\n", max_depth); 2528 } 2529 #endif /* DEBUG */ 2530 2531 #if defined(PMAPCHECK) || defined(DEBUG) 2532 void 2533 pmap_pvo_verify(void) 2534 { 2535 int ptegidx; 2536 int s; 2537 2538 s = splvm(); 2539 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2540 struct pvo_entry *pvo; 2541 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2542 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 2543 panic("pmap_pvo_verify: invalid pvo %p " 2544 "on list %#x", pvo, ptegidx); 2545 pmap_pvo_check(pvo); 2546 } 2547 } 2548 splx(s); 2549 } 2550 #endif /* PMAPCHECK */ 2551 2552 2553 void * 2554 pmap_pool_ualloc(struct pool *pp, int flags) 2555 { 2556 struct pvo_page *pvop; 2557 2558 pvop = SIMPLEQ_FIRST(&pmap_upvop_head); 2559 if (pvop != NULL) { 2560 pmap_upvop_free--; 2561 SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link); 2562 return pvop; 2563 } 2564 if (uvm.page_init_done != TRUE) { 2565 return (void *) uvm_pageboot_alloc(PAGE_SIZE); 2566 } 2567 return pmap_pool_malloc(pp, flags); 2568 } 2569 2570 void * 2571 pmap_pool_malloc(struct pool *pp, int flags) 2572 { 2573 struct pvo_page *pvop; 2574 struct vm_page *pg; 2575 2576 pvop = SIMPLEQ_FIRST(&pmap_mpvop_head); 2577 if (pvop != NULL) { 2578 pmap_mpvop_free--; 2579 SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link); 2580 return pvop; 2581 } 2582 again: 2583 pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE, 2584 UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256); 2585 if (__predict_false(pg == NULL)) { 2586 if (flags & PR_WAITOK) { 2587 uvm_wait("plpg"); 2588 goto again; 2589 } else { 2590 return (0); 2591 } 2592 } 2593 return (void *) VM_PAGE_TO_PHYS(pg); 2594 } 2595 2596 void 2597 pmap_pool_ufree(struct pool *pp, void *va) 2598 { 2599 struct pvo_page *pvop; 2600 #if 0 2601 if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) { 2602 pmap_pool_mfree(va, size, tag); 2603 return; 2604 } 2605 #endif 2606 pvop = va; 2607 SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link); 2608 pmap_upvop_free++; 2609 if (pmap_upvop_free > pmap_upvop_maxfree) 2610 pmap_upvop_maxfree = pmap_upvop_free; 2611 } 2612 2613 void 2614 pmap_pool_mfree(struct pool *pp, void *va) 2615 { 2616 struct pvo_page *pvop; 2617 2618 pvop = va; 2619 SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link); 2620 pmap_mpvop_free++; 2621 if (pmap_mpvop_free > pmap_mpvop_maxfree) 2622 pmap_mpvop_maxfree = pmap_mpvop_free; 2623 #if 0 2624 uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va)); 2625 #endif 2626 } 2627 2628 /* 2629 * This routine in bootstraping to steal to-be-managed memory (which will 2630 * then be unmanaged). We use it to grab from the first 256MB for our 2631 * pmap needs and above 256MB for other stuff. 2632 */ 2633 vaddr_t 2634 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp) 2635 { 2636 vsize_t size; 2637 vaddr_t va; 2638 paddr_t pa = 0; 2639 int npgs, bank; 2640 struct vm_physseg *ps; 2641 2642 if (uvm.page_init_done == TRUE) 2643 panic("pmap_steal_memory: called _after_ bootstrap"); 2644 2645 *vstartp = VM_MIN_KERNEL_ADDRESS; 2646 *vendp = VM_MAX_KERNEL_ADDRESS; 2647 2648 size = round_page(vsize); 2649 npgs = atop(size); 2650 2651 /* 2652 * PA 0 will never be among those given to UVM so we can use it 2653 * to indicate we couldn't steal any memory. 2654 */ 2655 for (ps = vm_physmem, bank = 0; bank < vm_nphysseg; bank++, ps++) { 2656 if (ps->free_list == VM_FREELIST_FIRST256 && 2657 ps->avail_end - ps->avail_start >= npgs) { 2658 pa = ptoa(ps->avail_start); 2659 break; 2660 } 2661 } 2662 2663 if (pa == 0) 2664 panic("pmap_steal_memory: no approriate memory to steal!"); 2665 2666 ps->avail_start += npgs; 2667 ps->start += npgs; 2668 2669 /* 2670 * If we've used up all the pages in the segment, remove it and 2671 * compact the list. 2672 */ 2673 if (ps->avail_start == ps->end) { 2674 /* 2675 * If this was the last one, then a very bad thing has occurred 2676 */ 2677 if (--vm_nphysseg == 0) 2678 panic("pmap_steal_memory: out of memory!"); 2679 2680 printf("pmap_steal_memory: consumed bank %d\n", bank); 2681 for (; bank < vm_nphysseg; bank++, ps++) { 2682 ps[0] = ps[1]; 2683 } 2684 } 2685 2686 va = (vaddr_t) pa; 2687 memset((caddr_t) va, 0, size); 2688 pmap_pages_stolen += npgs; 2689 #ifdef DEBUG 2690 if (pmapdebug && npgs > 1) { 2691 u_int cnt = 0; 2692 for (bank = 0, ps = vm_physmem; bank < vm_nphysseg; bank++, ps++) 2693 cnt += ps->avail_end - ps->avail_start; 2694 printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n", 2695 npgs, pmap_pages_stolen, cnt); 2696 } 2697 #endif 2698 2699 return va; 2700 } 2701 2702 /* 2703 * Find a chuck of memory with right size and alignment. 2704 */ 2705 void * 2706 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end) 2707 { 2708 struct mem_region *mp; 2709 paddr_t s, e; 2710 int i, j; 2711 2712 size = round_page(size); 2713 2714 DPRINTFN(BOOT, 2715 ("pmap_boot_find_memory: size=%lx, alignment=%lx, at_end=%d", 2716 size, alignment, at_end)); 2717 2718 if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0) 2719 panic("pmap_boot_find_memory: invalid alignment %lx", 2720 alignment); 2721 2722 if (at_end) { 2723 if (alignment != PAGE_SIZE) 2724 panic("pmap_boot_find_memory: invalid ending " 2725 "alignment %lx", alignment); 2726 2727 for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) { 2728 s = mp->start + mp->size - size; 2729 if (s >= mp->start && mp->size >= size) { 2730 DPRINTFN(BOOT,(": %lx\n", s)); 2731 DPRINTFN(BOOT, 2732 ("pmap_boot_find_memory: b-avail[%d] start " 2733 "0x%lx size 0x%lx\n", mp - avail, 2734 mp->start, mp->size)); 2735 mp->size -= size; 2736 DPRINTFN(BOOT, 2737 ("pmap_boot_find_memory: a-avail[%d] start " 2738 "0x%lx size 0x%lx\n", mp - avail, 2739 mp->start, mp->size)); 2740 return (void *) s; 2741 } 2742 } 2743 panic("pmap_boot_find_memory: no available memory"); 2744 } 2745 2746 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2747 s = (mp->start + alignment - 1) & ~(alignment-1); 2748 e = s + size; 2749 2750 /* 2751 * Is the calculated region entirely within the region? 2752 */ 2753 if (s < mp->start || e > mp->start + mp->size) 2754 continue; 2755 2756 DPRINTFN(BOOT,(": %lx\n", s)); 2757 if (s == mp->start) { 2758 /* 2759 * If the block starts at the beginning of region, 2760 * adjust the size & start. (the region may now be 2761 * zero in length) 2762 */ 2763 DPRINTFN(BOOT, 2764 ("pmap_boot_find_memory: b-avail[%d] start " 2765 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2766 mp->start += size; 2767 mp->size -= size; 2768 DPRINTFN(BOOT, 2769 ("pmap_boot_find_memory: a-avail[%d] start " 2770 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2771 } else if (e == mp->start + mp->size) { 2772 /* 2773 * If the block starts at the beginning of region, 2774 * adjust only the size. 2775 */ 2776 DPRINTFN(BOOT, 2777 ("pmap_boot_find_memory: b-avail[%d] start " 2778 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2779 mp->size -= size; 2780 DPRINTFN(BOOT, 2781 ("pmap_boot_find_memory: a-avail[%d] start " 2782 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2783 } else { 2784 /* 2785 * Block is in the middle of the region, so we 2786 * have to split it in two. 2787 */ 2788 for (j = avail_cnt; j > i + 1; j--) { 2789 avail[j] = avail[j-1]; 2790 } 2791 DPRINTFN(BOOT, 2792 ("pmap_boot_find_memory: b-avail[%d] start " 2793 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2794 mp[1].start = e; 2795 mp[1].size = mp[0].start + mp[0].size - e; 2796 mp[0].size = s - mp[0].start; 2797 avail_cnt++; 2798 for (; i < avail_cnt; i++) { 2799 DPRINTFN(BOOT, 2800 ("pmap_boot_find_memory: a-avail[%d] " 2801 "start 0x%lx size 0x%lx\n", i, 2802 avail[i].start, avail[i].size)); 2803 } 2804 } 2805 return (void *) s; 2806 } 2807 panic("pmap_boot_find_memory: not enough memory for " 2808 "%lx/%lx allocation?", size, alignment); 2809 } 2810 2811 /* 2812 * This is not part of the defined PMAP interface and is specific to the 2813 * PowerPC architecture. This is called during initppc, before the system 2814 * is really initialized. 2815 */ 2816 void 2817 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend) 2818 { 2819 struct mem_region *mp, tmp; 2820 paddr_t s, e; 2821 psize_t size; 2822 int i, j; 2823 2824 /* 2825 * Get memory. 2826 */ 2827 mem_regions(&mem, &avail); 2828 #if defined(DEBUG) 2829 if (pmapdebug & PMAPDEBUG_BOOT) { 2830 printf("pmap_bootstrap: memory configuration:\n"); 2831 for (mp = mem; mp->size; mp++) { 2832 printf("pmap_bootstrap: mem start 0x%lx size 0x%lx\n", 2833 mp->start, mp->size); 2834 } 2835 for (mp = avail; mp->size; mp++) { 2836 printf("pmap_bootstrap: avail start 0x%lx size 0x%lx\n", 2837 mp->start, mp->size); 2838 } 2839 } 2840 #endif 2841 2842 /* 2843 * Find out how much physical memory we have and in how many chunks. 2844 */ 2845 for (mem_cnt = 0, mp = mem; mp->size; mp++) { 2846 if (mp->start >= pmap_memlimit) 2847 continue; 2848 if (mp->start + mp->size > pmap_memlimit) { 2849 size = pmap_memlimit - mp->start; 2850 physmem += btoc(size); 2851 } else { 2852 physmem += btoc(mp->size); 2853 } 2854 mem_cnt++; 2855 } 2856 2857 /* 2858 * Count the number of available entries. 2859 */ 2860 for (avail_cnt = 0, mp = avail; mp->size; mp++) 2861 avail_cnt++; 2862 2863 /* 2864 * Page align all regions. 2865 */ 2866 kernelstart = trunc_page(kernelstart); 2867 kernelend = round_page(kernelend); 2868 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2869 s = round_page(mp->start); 2870 mp->size -= (s - mp->start); 2871 mp->size = trunc_page(mp->size); 2872 mp->start = s; 2873 e = mp->start + mp->size; 2874 2875 DPRINTFN(BOOT, 2876 ("pmap_bootstrap: b-avail[%d] start 0x%lx size 0x%lx\n", 2877 i, mp->start, mp->size)); 2878 2879 /* 2880 * Don't allow the end to run beyond our artificial limit 2881 */ 2882 if (e > pmap_memlimit) 2883 e = pmap_memlimit; 2884 2885 /* 2886 * Is this region empty or strange? skip it. 2887 */ 2888 if (e <= s) { 2889 mp->start = 0; 2890 mp->size = 0; 2891 continue; 2892 } 2893 2894 /* 2895 * Does this overlap the beginning of kernel? 2896 * Does extend past the end of the kernel? 2897 */ 2898 else if (s < kernelstart && e > kernelstart) { 2899 if (e > kernelend) { 2900 avail[avail_cnt].start = kernelend; 2901 avail[avail_cnt].size = e - kernelend; 2902 avail_cnt++; 2903 } 2904 mp->size = kernelstart - s; 2905 } 2906 /* 2907 * Check whether this region overlaps the end of the kernel. 2908 */ 2909 else if (s < kernelend && e > kernelend) { 2910 mp->start = kernelend; 2911 mp->size = e - kernelend; 2912 } 2913 /* 2914 * Look whether this regions is completely inside the kernel. 2915 * Nuke it if it does. 2916 */ 2917 else if (s >= kernelstart && e <= kernelend) { 2918 mp->start = 0; 2919 mp->size = 0; 2920 } 2921 /* 2922 * If the user imposed a memory limit, enforce it. 2923 */ 2924 else if (s >= pmap_memlimit) { 2925 mp->start = -PAGE_SIZE; /* let's know why */ 2926 mp->size = 0; 2927 } 2928 else { 2929 mp->start = s; 2930 mp->size = e - s; 2931 } 2932 DPRINTFN(BOOT, 2933 ("pmap_bootstrap: a-avail[%d] start 0x%lx size 0x%lx\n", 2934 i, mp->start, mp->size)); 2935 } 2936 2937 /* 2938 * Move (and uncount) all the null return to the end. 2939 */ 2940 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2941 if (mp->size == 0) { 2942 tmp = avail[i]; 2943 avail[i] = avail[--avail_cnt]; 2944 avail[avail_cnt] = avail[i]; 2945 } 2946 } 2947 2948 /* 2949 * (Bubble)sort them into asecnding order. 2950 */ 2951 for (i = 0; i < avail_cnt; i++) { 2952 for (j = i + 1; j < avail_cnt; j++) { 2953 if (avail[i].start > avail[j].start) { 2954 tmp = avail[i]; 2955 avail[i] = avail[j]; 2956 avail[j] = tmp; 2957 } 2958 } 2959 } 2960 2961 /* 2962 * Make sure they don't overlap. 2963 */ 2964 for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) { 2965 if (mp[0].start + mp[0].size > mp[1].start) { 2966 mp[0].size = mp[1].start - mp[0].start; 2967 } 2968 DPRINTFN(BOOT, 2969 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 2970 i, mp->start, mp->size)); 2971 } 2972 DPRINTFN(BOOT, 2973 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 2974 i, mp->start, mp->size)); 2975 2976 #ifdef PTEGCOUNT 2977 pmap_pteg_cnt = PTEGCOUNT; 2978 #else /* PTEGCOUNT */ 2979 pmap_pteg_cnt = 0x1000; 2980 2981 while (pmap_pteg_cnt < physmem) 2982 pmap_pteg_cnt <<= 1; 2983 2984 pmap_pteg_cnt >>= 1; 2985 #endif /* PTEGCOUNT */ 2986 2987 /* 2988 * Find suitably aligned memory for PTEG hash table. 2989 */ 2990 size = pmap_pteg_cnt * sizeof(struct pteg); 2991 pmap_pteg_table = pmap_boot_find_memory(size, size, 0); 2992 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 2993 if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH) 2994 panic("pmap_bootstrap: pmap_pteg_table end (%p + %lx) > 256MB", 2995 pmap_pteg_table, size); 2996 #endif 2997 2998 memset((void *)pmap_pteg_table, 0, pmap_pteg_cnt * sizeof(struct pteg)); 2999 pmap_pteg_mask = pmap_pteg_cnt - 1; 3000 3001 /* 3002 * We cannot do pmap_steal_memory here since UVM hasn't been loaded 3003 * with pages. So we just steal them before giving them to UVM. 3004 */ 3005 size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt; 3006 pmap_pvo_table = pmap_boot_find_memory(size, PAGE_SIZE, 0); 3007 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3008 if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH) 3009 panic("pmap_bootstrap: pmap_pvo_table end (%p + %lx) > 256MB", 3010 pmap_pvo_table, size); 3011 #endif 3012 3013 for (i = 0; i < pmap_pteg_cnt; i++) 3014 TAILQ_INIT(&pmap_pvo_table[i]); 3015 3016 #ifndef MSGBUFADDR 3017 /* 3018 * Allocate msgbuf in high memory. 3019 */ 3020 msgbuf_paddr = 3021 (paddr_t) pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1); 3022 #endif 3023 3024 #ifdef __HAVE_PMAP_PHYSSEG 3025 { 3026 u_int npgs = 0; 3027 for (i = 0, mp = avail; i < avail_cnt; i++, mp++) 3028 npgs += btoc(mp->size); 3029 size = (sizeof(struct pvo_head) + 1) * npgs; 3030 pmap_physseg.pvoh = pmap_boot_find_memory(size, PAGE_SIZE, 0); 3031 pmap_physseg.attrs = (char *) &pmap_physseg.pvoh[npgs]; 3032 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3033 if ((uintptr_t)pmap_physseg.pvoh + size > SEGMENT_LENGTH) 3034 panic("pmap_bootstrap: PVO list end (%p + %lx) > 256MB", 3035 pmap_physseg.pvoh, size); 3036 #endif 3037 } 3038 #endif 3039 3040 for (mp = avail, i = 0; i < avail_cnt; mp++, i++) { 3041 paddr_t pfstart = atop(mp->start); 3042 paddr_t pfend = atop(mp->start + mp->size); 3043 if (mp->size == 0) 3044 continue; 3045 if (mp->start + mp->size <= SEGMENT_LENGTH) { 3046 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3047 VM_FREELIST_FIRST256); 3048 } else if (mp->start >= SEGMENT_LENGTH) { 3049 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3050 VM_FREELIST_DEFAULT); 3051 } else { 3052 pfend = atop(SEGMENT_LENGTH); 3053 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3054 VM_FREELIST_FIRST256); 3055 pfstart = atop(SEGMENT_LENGTH); 3056 pfend = atop(mp->start + mp->size); 3057 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3058 VM_FREELIST_DEFAULT); 3059 } 3060 } 3061 3062 /* 3063 * Make sure kernel vsid is allocated as well as VSID 0. 3064 */ 3065 pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW] 3066 |= 1 << (KERNEL_VSIDBITS % VSID_NBPW); 3067 pmap_vsid_bitmap[0] |= 1; 3068 3069 /* 3070 * Initialize kernel pmap and hardware. 3071 */ 3072 #ifndef PPC_OEA64 3073 for (i = 0; i < 16; i++) { 3074 pmap_kernel()->pm_sr[i] = EMPTY_SEGMENT; 3075 __asm __volatile ("mtsrin %0,%1" 3076 :: "r"(EMPTY_SEGMENT), "r"(i << ADDR_SR_SHFT)); 3077 } 3078 3079 pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY; 3080 __asm __volatile ("mtsr %0,%1" 3081 :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT)); 3082 #ifdef KERNEL2_SR 3083 pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY; 3084 __asm __volatile ("mtsr %0,%1" 3085 :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT)); 3086 #endif 3087 for (i = 0; i < 16; i++) { 3088 if (iosrtable[i] & SR601_T) { 3089 pmap_kernel()->pm_sr[i] = iosrtable[i]; 3090 __asm __volatile ("mtsrin %0,%1" 3091 :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT)); 3092 } 3093 } 3094 #endif /* !PPC_OEA64 */ 3095 3096 __asm __volatile ("sync; mtsdr1 %0; isync" 3097 :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10))); 3098 tlbia(); 3099 3100 #ifdef ALTIVEC 3101 pmap_use_altivec = cpu_altivec; 3102 #endif 3103 3104 #ifdef DEBUG 3105 if (pmapdebug & PMAPDEBUG_BOOT) { 3106 u_int cnt; 3107 int bank; 3108 char pbuf[9]; 3109 for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) { 3110 cnt += vm_physmem[bank].avail_end - vm_physmem[bank].avail_start; 3111 printf("pmap_bootstrap: vm_physmem[%d]=%#lx-%#lx/%#lx\n", 3112 bank, 3113 ptoa(vm_physmem[bank].avail_start), 3114 ptoa(vm_physmem[bank].avail_end), 3115 ptoa(vm_physmem[bank].avail_end - vm_physmem[bank].avail_start)); 3116 } 3117 format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt)); 3118 printf("pmap_bootstrap: UVM memory = %s (%u pages)\n", 3119 pbuf, cnt); 3120 } 3121 #endif 3122 3123 pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry), 3124 sizeof(struct pvo_entry), 0, 0, "pmap_upvopl", 3125 &pmap_pool_uallocator); 3126 3127 pool_setlowat(&pmap_upvo_pool, 252); 3128 3129 pool_init(&pmap_pool, sizeof(struct pmap), 3130 sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator); 3131 } 3132
Indexes created Mon Sep 29 21:09:56 GMT 2025