pmap.c revision 1.21
1 /* $NetBSD: pmap.c,v 1.21 2004/03/17 14:14:02 aymeric Exp $ */ 2 /*- 3 * Copyright (c) 2001 The NetBSD Foundation, Inc. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to The NetBSD Foundation 7 * by Matt Thomas <matt (at) 3am-software.com> of Allegro Networks, Inc. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by the NetBSD 20 * Foundation, Inc. and its contributors. 21 * 4. Neither the name of The NetBSD Foundation nor the names of its 22 * contributors may be used to endorse or promote products derived 23 * from this software without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 26 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 27 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 28 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 29 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 30 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 31 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 32 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 33 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 34 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 35 * POSSIBILITY OF SUCH DAMAGE. 36 */ 37 38 /* 39 * Copyright (C) 1995, 1996 Wolfgang Solfrank. 40 * Copyright (C) 1995, 1996 TooLs GmbH. 41 * All rights reserved. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. All advertising materials mentioning features or use of this software 52 * must display the following acknowledgement: 53 * This product includes software developed by TooLs GmbH. 54 * 4. The name of TooLs GmbH may not be used to endorse or promote products 55 * derived from this software without specific prior written permission. 56 * 57 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR 58 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 59 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 60 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 61 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 62 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 63 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 64 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 65 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 66 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.21 2004/03/17 14:14:02 aymeric Exp $"); 71 72 #include "opt_ppcarch.h" 73 #include "opt_altivec.h" 74 #include "opt_pmap.h" 75 #include <sys/param.h> 76 #include <sys/malloc.h> 77 #include <sys/proc.h> 78 #include <sys/user.h> 79 #include <sys/pool.h> 80 #include <sys/queue.h> 81 #include <sys/device.h> /* for evcnt */ 82 #include <sys/systm.h> 83 84 #if __NetBSD_Version__ < 105010000 85 #include <vm/vm.h> 86 #include <vm/vm_kern.h> 87 #define splvm() splimp() 88 #endif 89 90 #include <uvm/uvm.h> 91 92 #include <machine/pcb.h> 93 #include <machine/powerpc.h> 94 #include <powerpc/spr.h> 95 #include <powerpc/oea/sr_601.h> 96 #include <powerpc/bat.h> 97 98 #if defined(DEBUG) || defined(PMAPCHECK) 99 #define STATIC 100 #else 101 #define STATIC static 102 #endif 103 104 #ifdef ALTIVEC 105 int pmap_use_altivec; 106 #endif 107 108 volatile struct pteg *pmap_pteg_table; 109 unsigned int pmap_pteg_cnt; 110 unsigned int pmap_pteg_mask; 111 #ifdef PMAP_MEMLIMIT 112 paddr_t pmap_memlimit = PMAP_MEMLIMIT; 113 #else 114 paddr_t pmap_memlimit = -PAGE_SIZE; /* there is no limit */ 115 #endif 116 117 struct pmap kernel_pmap_; 118 unsigned int pmap_pages_stolen; 119 u_long pmap_pte_valid; 120 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 121 u_long pmap_pvo_enter_depth; 122 u_long pmap_pvo_remove_depth; 123 #endif 124 125 int physmem; 126 #ifndef MSGBUFADDR 127 extern paddr_t msgbuf_paddr; 128 #endif 129 130 static struct mem_region *mem, *avail; 131 static u_int mem_cnt, avail_cnt; 132 133 #ifdef __HAVE_PMAP_PHYSSEG 134 /* 135 * This is a cache of referenced/modified bits. 136 * Bits herein are shifted by ATTRSHFT. 137 */ 138 #define ATTR_SHFT 4 139 struct pmap_physseg pmap_physseg; 140 #endif 141 142 /* 143 * The following structure is exactly 32 bytes long (one cacheline). 144 */ 145 struct pvo_entry { 146 LIST_ENTRY(pvo_entry) pvo_vlink; /* Link to common virt page */ 147 TAILQ_ENTRY(pvo_entry) pvo_olink; /* Link to overflow entry */ 148 struct pte pvo_pte; /* Prebuilt PTE */ 149 pmap_t pvo_pmap; /* ptr to owning pmap */ 150 vaddr_t pvo_vaddr; /* VA of entry */ 151 #define PVO_PTEGIDX_MASK 0x0007 /* which PTEG slot */ 152 #define PVO_PTEGIDX_VALID 0x0008 /* slot is valid */ 153 #define PVO_WIRED 0x0010 /* PVO entry is wired */ 154 #define PVO_MANAGED 0x0020 /* PVO e. for managed page */ 155 #define PVO_EXECUTABLE 0x0040 /* PVO e. for executable page */ 156 #define PVO_ENTER_INSERT 0 /* PVO has been removed */ 157 #define PVO_SPILL_UNSET 1 /* PVO has been evicted */ 158 #define PVO_SPILL_SET 2 /* PVO has been spilled */ 159 #define PVO_SPILL_INSERT 3 /* PVO has been inserted */ 160 #define PVO_PMAP_PAGE_PROTECT 4 /* PVO has changed */ 161 #define PVO_PMAP_PROTECT 5 /* PVO has changed */ 162 #define PVO_REMOVE 6 /* PVO has been removed */ 163 #define PVO_WHERE_MASK 15 164 #define PVO_WHERE_SHFT 8 165 }; 166 #define PVO_VADDR(pvo) ((pvo)->pvo_vaddr & ~ADDR_POFF) 167 #define PVO_ISEXECUTABLE(pvo) ((pvo)->pvo_vaddr & PVO_EXECUTABLE) 168 #define PVO_PTEGIDX_GET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK) 169 #define PVO_PTEGIDX_ISSET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID) 170 #define PVO_PTEGIDX_CLR(pvo) \ 171 ((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK))) 172 #define PVO_PTEGIDX_SET(pvo,i) \ 173 ((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID)) 174 #define PVO_WHERE(pvo,w) \ 175 ((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \ 176 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT)) 177 178 TAILQ_HEAD(pvo_tqhead, pvo_entry); 179 struct pvo_tqhead *pmap_pvo_table; /* pvo entries by ptegroup index */ 180 struct pvo_head pmap_pvo_kunmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged); /* list of unmanaged pages */ 181 struct pvo_head pmap_pvo_unmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged); /* list of unmanaged pages */ 182 183 struct pool pmap_pool; /* pool for pmap structures */ 184 struct pool pmap_upvo_pool; /* pool for pvo entries for unmanaged pages */ 185 struct pool pmap_mpvo_pool; /* pool for pvo entries for managed pages */ 186 187 /* 188 * We keep a cache of unmanaged pages to be used for pvo entries for 189 * unmanaged pages. 190 */ 191 struct pvo_page { 192 SIMPLEQ_ENTRY(pvo_page) pvop_link; 193 }; 194 SIMPLEQ_HEAD(pvop_head, pvo_page); 195 struct pvop_head pmap_upvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_upvop_head); 196 struct pvop_head pmap_mpvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_mpvop_head); 197 u_long pmap_upvop_free; 198 u_long pmap_upvop_maxfree; 199 u_long pmap_mpvop_free; 200 u_long pmap_mpvop_maxfree; 201 202 STATIC void *pmap_pool_ualloc(struct pool *, int); 203 STATIC void *pmap_pool_malloc(struct pool *, int); 204 205 STATIC void pmap_pool_ufree(struct pool *, void *); 206 STATIC void pmap_pool_mfree(struct pool *, void *); 207 208 static struct pool_allocator pmap_pool_mallocator = { 209 pmap_pool_malloc, pmap_pool_mfree, 0, 210 }; 211 212 static struct pool_allocator pmap_pool_uallocator = { 213 pmap_pool_ualloc, pmap_pool_ufree, 0, 214 }; 215 216 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 217 void pmap_pte_print(volatile struct pte *); 218 #endif 219 220 #ifdef DDB 221 void pmap_pteg_check(void); 222 void pmap_pteg_dist(void); 223 void pmap_print_pte(pmap_t, vaddr_t); 224 void pmap_print_mmuregs(void); 225 #endif 226 227 #if defined(DEBUG) || defined(PMAPCHECK) 228 #ifdef PMAPCHECK 229 int pmapcheck = 1; 230 #else 231 int pmapcheck = 0; 232 #endif 233 void pmap_pvo_verify(void); 234 STATIC void pmap_pvo_check(const struct pvo_entry *); 235 #define PMAP_PVO_CHECK(pvo) \ 236 do { \ 237 if (pmapcheck) \ 238 pmap_pvo_check(pvo); \ 239 } while (0) 240 #else 241 #define PMAP_PVO_CHECK(pvo) do { } while (/*CONSTCOND*/0) 242 #endif 243 STATIC int pmap_pte_insert(int, struct pte *); 244 STATIC int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *, 245 vaddr_t, paddr_t, register_t, int); 246 STATIC void pmap_pvo_remove(struct pvo_entry *, int); 247 STATIC struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *); 248 STATIC volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int); 249 #define pmap_pvo_reclaim(pm) NULL 250 STATIC void pvo_set_exec(struct pvo_entry *); 251 STATIC void pvo_clear_exec(struct pvo_entry *); 252 253 STATIC void tlbia(void); 254 255 STATIC void pmap_release(pmap_t); 256 STATIC void *pmap_boot_find_memory(psize_t, psize_t, int); 257 258 #define VSID_NBPW (sizeof(uint32_t) * 8) 259 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW]; 260 261 static int pmap_initialized; 262 263 #if defined(DEBUG) || defined(PMAPDEBUG) 264 #define PMAPDEBUG_BOOT 0x0001 265 #define PMAPDEBUG_PTE 0x0002 266 #define PMAPDEBUG_EXEC 0x0008 267 #define PMAPDEBUG_PVOENTER 0x0010 268 #define PMAPDEBUG_PVOREMOVE 0x0020 269 #define PMAPDEBUG_ACTIVATE 0x0100 270 #define PMAPDEBUG_CREATE 0x0200 271 #define PMAPDEBUG_ENTER 0x1000 272 #define PMAPDEBUG_KENTER 0x2000 273 #define PMAPDEBUG_KREMOVE 0x4000 274 #define PMAPDEBUG_REMOVE 0x8000 275 unsigned int pmapdebug = 0; 276 # define DPRINTF(x) printf x 277 # define DPRINTFN(n, x) if (pmapdebug & PMAPDEBUG_ ## n) printf x 278 #else 279 # define DPRINTF(x) 280 # define DPRINTFN(n, x) 281 #endif 282 283 284 #ifdef PMAPCOUNTERS 285 #define PMAPCOUNT(ev) ((pmap_evcnt_ ## ev).ev_count++) 286 #define PMAPCOUNT2(ev) ((ev).ev_count++) 287 288 struct evcnt pmap_evcnt_mappings = 289 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 290 "pmap", "pages mapped"); 291 struct evcnt pmap_evcnt_unmappings = 292 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 293 "pmap", "pages unmapped"); 294 295 struct evcnt pmap_evcnt_kernel_mappings = 296 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 297 "pmap", "kernel pages mapped"); 298 struct evcnt pmap_evcnt_kernel_unmappings = 299 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_kernel_mappings, 300 "pmap", "kernel pages unmapped"); 301 302 struct evcnt pmap_evcnt_mappings_replaced = 303 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 304 "pmap", "page mappings replaced"); 305 306 struct evcnt pmap_evcnt_exec_mappings = 307 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 308 "pmap", "exec pages mapped"); 309 struct evcnt pmap_evcnt_exec_cached = 310 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 311 "pmap", "exec pages cached"); 312 313 struct evcnt pmap_evcnt_exec_synced = 314 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 315 "pmap", "exec pages synced"); 316 struct evcnt pmap_evcnt_exec_synced_clear_modify = 317 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 318 "pmap", "exec pages synced (CM)"); 319 320 struct evcnt pmap_evcnt_exec_uncached_page_protect = 321 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 322 "pmap", "exec pages uncached (PP)"); 323 struct evcnt pmap_evcnt_exec_uncached_clear_modify = 324 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 325 "pmap", "exec pages uncached (CM)"); 326 struct evcnt pmap_evcnt_exec_uncached_zero_page = 327 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 328 "pmap", "exec pages uncached (ZP)"); 329 struct evcnt pmap_evcnt_exec_uncached_copy_page = 330 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 331 "pmap", "exec pages uncached (CP)"); 332 333 struct evcnt pmap_evcnt_updates = 334 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 335 "pmap", "updates"); 336 struct evcnt pmap_evcnt_collects = 337 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 338 "pmap", "collects"); 339 struct evcnt pmap_evcnt_copies = 340 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 341 "pmap", "copies"); 342 343 struct evcnt pmap_evcnt_ptes_spilled = 344 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 345 "pmap", "ptes spilled from overflow"); 346 struct evcnt pmap_evcnt_ptes_unspilled = 347 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 348 "pmap", "ptes not spilled"); 349 struct evcnt pmap_evcnt_ptes_evicted = 350 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 351 "pmap", "ptes evicted"); 352 353 struct evcnt pmap_evcnt_ptes_primary[8] = { 354 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 355 "pmap", "ptes added at primary[0]"), 356 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 357 "pmap", "ptes added at primary[1]"), 358 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 359 "pmap", "ptes added at primary[2]"), 360 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 361 "pmap", "ptes added at primary[3]"), 362 363 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 364 "pmap", "ptes added at primary[4]"), 365 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 366 "pmap", "ptes added at primary[5]"), 367 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 368 "pmap", "ptes added at primary[6]"), 369 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 370 "pmap", "ptes added at primary[7]"), 371 }; 372 struct evcnt pmap_evcnt_ptes_secondary[8] = { 373 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 374 "pmap", "ptes added at secondary[0]"), 375 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 376 "pmap", "ptes added at secondary[1]"), 377 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 378 "pmap", "ptes added at secondary[2]"), 379 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 380 "pmap", "ptes added at secondary[3]"), 381 382 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 383 "pmap", "ptes added at secondary[4]"), 384 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 385 "pmap", "ptes added at secondary[5]"), 386 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 387 "pmap", "ptes added at secondary[6]"), 388 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 389 "pmap", "ptes added at secondary[7]"), 390 }; 391 struct evcnt pmap_evcnt_ptes_removed = 392 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 393 "pmap", "ptes removed"); 394 struct evcnt pmap_evcnt_ptes_changed = 395 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 396 "pmap", "ptes changed"); 397 398 /* 399 * From pmap_subr.c 400 */ 401 extern struct evcnt pmap_evcnt_zeroed_pages; 402 extern struct evcnt pmap_evcnt_copied_pages; 403 extern struct evcnt pmap_evcnt_idlezeroed_pages; 404 #else 405 #define PMAPCOUNT(ev) ((void) 0) 406 #define PMAPCOUNT2(ev) ((void) 0) 407 #endif 408 409 #define TLBIE(va) __asm __volatile("tlbie %0" :: "r"(va)) 410 #define TLBSYNC() __asm __volatile("tlbsync") 411 #define SYNC() __asm __volatile("sync") 412 #define EIEIO() __asm __volatile("eieio") 413 #define MFMSR() mfmsr() 414 #define MTMSR(psl) mtmsr(psl) 415 #define MFPVR() mfpvr() 416 #define MFSRIN(va) mfsrin(va) 417 #define MFTB() mfrtcltbl() 418 419 #ifndef PPC_OEA64 420 static __inline register_t 421 mfsrin(vaddr_t va) 422 { 423 register_t sr; 424 __asm __volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va)); 425 return sr; 426 } 427 #endif /* PPC_OEA64 */ 428 429 static __inline register_t 430 pmap_interrupts_off(void) 431 { 432 register_t msr = MFMSR(); 433 if (msr & PSL_EE) 434 MTMSR(msr & ~PSL_EE); 435 return msr; 436 } 437 438 static void 439 pmap_interrupts_restore(register_t msr) 440 { 441 if (msr & PSL_EE) 442 MTMSR(msr); 443 } 444 445 static __inline u_int32_t 446 mfrtcltbl(void) 447 { 448 449 if ((MFPVR() >> 16) == MPC601) 450 return (mfrtcl() >> 7); 451 else 452 return (mftbl()); 453 } 454 455 /* 456 * These small routines may have to be replaced, 457 * if/when we support processors other that the 604. 458 */ 459 460 void 461 tlbia(void) 462 { 463 caddr_t i; 464 465 SYNC(); 466 /* 467 * Why not use "tlbia"? Because not all processors implement it. 468 * 469 * This needs to be a per-CPU callback to do the appropriate thing 470 * for the CPU. XXX 471 */ 472 for (i = 0; i < (caddr_t)0x00040000; i += 0x00001000) { 473 TLBIE(i); 474 EIEIO(); 475 SYNC(); 476 } 477 TLBSYNC(); 478 SYNC(); 479 } 480 481 static __inline register_t 482 va_to_vsid(const struct pmap *pm, vaddr_t addr) 483 { 484 #ifdef PPC_OEA64 485 #if 0 486 const struct ste *ste; 487 register_t hash; 488 int i; 489 490 hash = (addr >> ADDR_ESID_SHFT) & ADDR_ESID_HASH; 491 492 /* 493 * Try the primary group first 494 */ 495 ste = pm->pm_stes[hash].stes; 496 for (i = 0; i < 8; i++, ste++) { 497 if (ste->ste_hi & STE_V) && 498 (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID)) 499 return ste; 500 } 501 502 /* 503 * Then the secondary group. 504 */ 505 ste = pm->pm_stes[hash ^ ADDR_ESID_HASH].stes; 506 for (i = 0; i < 8; i++, ste++) { 507 if (ste->ste_hi & STE_V) && 508 (addr & ~(ADDR_POFF|ADDR_PIDX)) == (ste->ste_hi & STE_ESID)) 509 return addr; 510 } 511 512 return NULL; 513 #else 514 /* 515 * Rather than searching the STE groups for the VSID, we know 516 * how we generate that from the ESID and so do that. 517 */ 518 return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT; 519 #endif 520 #else 521 return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID) >> SR_VSID_SHFT; 522 #endif 523 } 524 525 static __inline register_t 526 va_to_pteg(const struct pmap *pm, vaddr_t addr) 527 { 528 register_t hash; 529 530 hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT); 531 return hash & pmap_pteg_mask; 532 } 533 534 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 535 /* 536 * Given a PTE in the page table, calculate the VADDR that hashes to it. 537 * The only bit of magic is that the top 4 bits of the address doesn't 538 * technically exist in the PTE. But we know we reserved 4 bits of the 539 * VSID for it so that's how we get it. 540 */ 541 static vaddr_t 542 pmap_pte_to_va(volatile const struct pte *pt) 543 { 544 vaddr_t va; 545 uintptr_t ptaddr = (uintptr_t) pt; 546 547 if (pt->pte_hi & PTE_HID) 548 ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg)); 549 550 /* PPC Bits 10-19 PPC64 Bits 42-51 */ 551 va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff; 552 va <<= ADDR_PIDX_SHFT; 553 554 /* PPC Bits 4-9 PPC64 Bits 36-41 */ 555 va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT; 556 557 #ifdef PPC_OEA64 558 /* PPC63 Bits 0-35 */ 559 /* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */ 560 #endif 561 #ifdef PPC_OEA 562 /* PPC Bits 0-3 */ 563 va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; 564 #endif 565 566 return va; 567 } 568 #endif 569 570 static __inline struct pvo_head * 571 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p) 572 { 573 #ifdef __HAVE_VM_PAGE_MD 574 struct vm_page *pg; 575 576 pg = PHYS_TO_VM_PAGE(pa); 577 if (pg_p != NULL) 578 *pg_p = pg; 579 if (pg == NULL) 580 return &pmap_pvo_unmanaged; 581 return &pg->mdpage.mdpg_pvoh; 582 #endif 583 #ifdef __HAVE_PMAP_PHYSSEG 584 int bank, pg; 585 586 bank = vm_physseg_find(atop(pa), &pg); 587 if (pg_p != NULL) 588 *pg_p = pg; 589 if (bank == -1) 590 return &pmap_pvo_unmanaged; 591 return &vm_physmem[bank].pmseg.pvoh[pg]; 592 #endif 593 } 594 595 static __inline struct pvo_head * 596 vm_page_to_pvoh(struct vm_page *pg) 597 { 598 #ifdef __HAVE_VM_PAGE_MD 599 return &pg->mdpage.mdpg_pvoh; 600 #endif 601 #ifdef __HAVE_PMAP_PHYSSEG 602 return pa_to_pvoh(VM_PAGE_TO_PHYS(pg), NULL); 603 #endif 604 } 605 606 607 #ifdef __HAVE_PMAP_PHYSSEG 608 static __inline char * 609 pa_to_attr(paddr_t pa) 610 { 611 int bank, pg; 612 613 bank = vm_physseg_find(atop(pa), &pg); 614 if (bank == -1) 615 return NULL; 616 return &vm_physmem[bank].pmseg.attrs[pg]; 617 } 618 #endif 619 620 static __inline void 621 pmap_attr_clear(struct vm_page *pg, int ptebit) 622 { 623 #ifdef __HAVE_PMAP_PHYSSEG 624 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) &= ~(ptebit >> ATTR_SHFT); 625 #endif 626 #ifdef __HAVE_VM_PAGE_MD 627 pg->mdpage.mdpg_attrs &= ~ptebit; 628 #endif 629 } 630 631 static __inline int 632 pmap_attr_fetch(struct vm_page *pg) 633 { 634 #ifdef __HAVE_PMAP_PHYSSEG 635 return *pa_to_attr(VM_PAGE_TO_PHYS(pg)) << ATTR_SHFT; 636 #endif 637 #ifdef __HAVE_VM_PAGE_MD 638 return pg->mdpage.mdpg_attrs; 639 #endif 640 } 641 642 static __inline void 643 pmap_attr_save(struct vm_page *pg, int ptebit) 644 { 645 #ifdef __HAVE_PMAP_PHYSSEG 646 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) |= (ptebit >> ATTR_SHFT); 647 #endif 648 #ifdef __HAVE_VM_PAGE_MD 649 pg->mdpage.mdpg_attrs |= ptebit; 650 #endif 651 } 652 653 static __inline int 654 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt) 655 { 656 if (pt->pte_hi == pvo_pt->pte_hi 657 #if 0 658 && ((pt->pte_lo ^ pvo_pt->pte_lo) & 659 ~(PTE_REF|PTE_CHG)) == 0 660 #endif 661 ) 662 return 1; 663 return 0; 664 } 665 666 static __inline void 667 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo) 668 { 669 /* 670 * Construct the PTE. Default to IMB initially. Valid bit 671 * only gets set when the real pte is set in memory. 672 * 673 * Note: Don't set the valid bit for correct operation of tlb update. 674 */ 675 pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT) 676 | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API); 677 pt->pte_lo = pte_lo; 678 } 679 680 static __inline void 681 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt) 682 { 683 pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG); 684 } 685 686 static __inline void 687 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit) 688 { 689 /* 690 * As shown in Section 7.6.3.2.3 691 */ 692 pt->pte_lo &= ~ptebit; 693 TLBIE(va); 694 SYNC(); 695 EIEIO(); 696 TLBSYNC(); 697 SYNC(); 698 } 699 700 static __inline void 701 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt) 702 { 703 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 704 if (pvo_pt->pte_hi & PTE_VALID) 705 panic("pte_set: setting an already valid pte %p", pvo_pt); 706 #endif 707 pvo_pt->pte_hi |= PTE_VALID; 708 /* 709 * Update the PTE as defined in section 7.6.3.1 710 * Note that the REF/CHG bits are from pvo_pt and thus should 711 * have been saved so this routine can restore them (if desired). 712 */ 713 pt->pte_lo = pvo_pt->pte_lo; 714 EIEIO(); 715 pt->pte_hi = pvo_pt->pte_hi; 716 SYNC(); 717 pmap_pte_valid++; 718 } 719 720 static __inline void 721 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 722 { 723 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 724 if ((pvo_pt->pte_hi & PTE_VALID) == 0) 725 panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt); 726 if ((pt->pte_hi & PTE_VALID) == 0) 727 panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt); 728 #endif 729 730 pvo_pt->pte_hi &= ~PTE_VALID; 731 /* 732 * Force the ref & chg bits back into the PTEs. 733 */ 734 SYNC(); 735 /* 736 * Invalidate the pte ... (Section 7.6.3.3) 737 */ 738 pt->pte_hi &= ~PTE_VALID; 739 SYNC(); 740 TLBIE(va); 741 SYNC(); 742 EIEIO(); 743 TLBSYNC(); 744 SYNC(); 745 /* 746 * Save the ref & chg bits ... 747 */ 748 pmap_pte_synch(pt, pvo_pt); 749 pmap_pte_valid--; 750 } 751 752 static __inline void 753 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 754 { 755 /* 756 * Invalidate the PTE 757 */ 758 pmap_pte_unset(pt, pvo_pt, va); 759 pmap_pte_set(pt, pvo_pt); 760 } 761 762 /* 763 * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx 764 * (either primary or secondary location). 765 * 766 * Note: both the destination and source PTEs must not have PTE_VALID set. 767 */ 768 769 STATIC int 770 pmap_pte_insert(int ptegidx, struct pte *pvo_pt) 771 { 772 volatile struct pte *pt; 773 int i; 774 775 #if defined(DEBUG) 776 DPRINTFN(PTE, ("pmap_pte_insert: idx 0x%x, pte 0x%x 0x%x\n", 777 ptegidx, (unsigned int) pvo_pt->pte_hi, (unsigned int) pvo_pt->pte_lo)); 778 #endif 779 /* 780 * First try primary hash. 781 */ 782 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 783 if ((pt->pte_hi & PTE_VALID) == 0) { 784 pvo_pt->pte_hi &= ~PTE_HID; 785 pmap_pte_set(pt, pvo_pt); 786 return i; 787 } 788 } 789 790 /* 791 * Now try secondary hash. 792 */ 793 ptegidx ^= pmap_pteg_mask; 794 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 795 if ((pt->pte_hi & PTE_VALID) == 0) { 796 pvo_pt->pte_hi |= PTE_HID; 797 pmap_pte_set(pt, pvo_pt); 798 return i; 799 } 800 } 801 return -1; 802 } 803 804 /* 805 * Spill handler. 806 * 807 * Tries to spill a page table entry from the overflow area. 808 * This runs in either real mode (if dealing with a exception spill) 809 * or virtual mode when dealing with manually spilling one of the 810 * kernel's pte entries. In either case, interrupts are already 811 * disabled. 812 */ 813 814 int 815 pmap_pte_spill(struct pmap *pm, vaddr_t addr, boolean_t exec) 816 { 817 struct pvo_entry *source_pvo, *victim_pvo, *next_pvo; 818 struct pvo_entry *pvo; 819 /* XXX: gcc -- vpvoh is always set at either *1* or *2* */ 820 struct pvo_tqhead *pvoh, *vpvoh = NULL; 821 int ptegidx, i, j; 822 volatile struct pteg *pteg; 823 volatile struct pte *pt; 824 825 ptegidx = va_to_pteg(pm, addr); 826 827 /* 828 * Have to substitute some entry. Use the primary hash for this. 829 * Use low bits of timebase as random generator. Make sure we are 830 * not picking a kernel pte for replacement. 831 */ 832 pteg = &pmap_pteg_table[ptegidx]; 833 i = MFTB() & 7; 834 for (j = 0; j < 8; j++) { 835 pt = &pteg->pt[i]; 836 if ((pt->pte_hi & PTE_VALID) == 0 || 837 VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT) 838 != KERNEL_VSIDBITS) 839 break; 840 i = (i + 1) & 7; 841 } 842 KASSERT(j < 8); 843 844 source_pvo = NULL; 845 victim_pvo = NULL; 846 pvoh = &pmap_pvo_table[ptegidx]; 847 TAILQ_FOREACH(pvo, pvoh, pvo_olink) { 848 849 /* 850 * We need to find pvo entry for this address... 851 */ 852 PMAP_PVO_CHECK(pvo); /* sanity check */ 853 854 /* 855 * If we haven't found the source and we come to a PVO with 856 * a valid PTE, then we know we can't find it because all 857 * evicted PVOs always are first in the list. 858 */ 859 if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID)) 860 break; 861 if (source_pvo == NULL && pm == pvo->pvo_pmap && 862 addr == PVO_VADDR(pvo)) { 863 864 /* 865 * Now we have found the entry to be spilled into the 866 * pteg. Attempt to insert it into the page table. 867 */ 868 j = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 869 if (j >= 0) { 870 PVO_PTEGIDX_SET(pvo, j); 871 PMAP_PVO_CHECK(pvo); /* sanity check */ 872 PVO_WHERE(pvo, SPILL_INSERT); 873 pvo->pvo_pmap->pm_evictions--; 874 PMAPCOUNT(ptes_spilled); 875 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 876 ? pmap_evcnt_ptes_secondary 877 : pmap_evcnt_ptes_primary)[j]); 878 879 /* 880 * Since we keep the evicted entries at the 881 * from of the PVO list, we need move this 882 * (now resident) PVO after the evicted 883 * entries. 884 */ 885 next_pvo = TAILQ_NEXT(pvo, pvo_olink); 886 887 /* 888 * If we don't have to move (either we were the 889 * last entry or the next entry was valid), 890 * don't change our position. Otherwise 891 * move ourselves to the tail of the queue. 892 */ 893 if (next_pvo != NULL && 894 !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) { 895 TAILQ_REMOVE(pvoh, pvo, pvo_olink); 896 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 897 } 898 return 1; 899 } 900 source_pvo = pvo; 901 if (exec && !PVO_ISEXECUTABLE(source_pvo)) { 902 return 0; 903 } 904 if (victim_pvo != NULL) 905 break; 906 } 907 908 /* 909 * We also need the pvo entry of the victim we are replacing 910 * so save the R & C bits of the PTE. 911 */ 912 if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL && 913 pmap_pte_compare(pt, &pvo->pvo_pte)) { 914 vpvoh = pvoh; /* *1* */ 915 victim_pvo = pvo; 916 if (source_pvo != NULL) 917 break; 918 } 919 } 920 921 if (source_pvo == NULL) { 922 PMAPCOUNT(ptes_unspilled); 923 return 0; 924 } 925 926 if (victim_pvo == NULL) { 927 if ((pt->pte_hi & PTE_HID) == 0) 928 panic("pmap_pte_spill: victim p-pte (%p) has " 929 "no pvo entry!", pt); 930 931 /* 932 * If this is a secondary PTE, we need to search 933 * its primary pvo bucket for the matching PVO. 934 */ 935 vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; /* *2* */ 936 TAILQ_FOREACH(pvo, vpvoh, pvo_olink) { 937 PMAP_PVO_CHECK(pvo); /* sanity check */ 938 939 /* 940 * We also need the pvo entry of the victim we are 941 * replacing so save the R & C bits of the PTE. 942 */ 943 if (pmap_pte_compare(pt, &pvo->pvo_pte)) { 944 victim_pvo = pvo; 945 break; 946 } 947 } 948 if (victim_pvo == NULL) 949 panic("pmap_pte_spill: victim s-pte (%p) has " 950 "no pvo entry!", pt); 951 } 952 953 /* 954 * The victim should be not be a kernel PVO/PTE entry. 955 */ 956 KASSERT(victim_pvo->pvo_pmap != pmap_kernel()); 957 KASSERT(PVO_PTEGIDX_ISSET(victim_pvo)); 958 KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i); 959 960 /* 961 * We are invalidating the TLB entry for the EA for the 962 * we are replacing even though its valid; If we don't 963 * we lose any ref/chg bit changes contained in the TLB 964 * entry. 965 */ 966 source_pvo->pvo_pte.pte_hi &= ~PTE_HID; 967 968 /* 969 * To enforce the PVO list ordering constraint that all 970 * evicted entries should come before all valid entries, 971 * move the source PVO to the tail of its list and the 972 * victim PVO to the head of its list (which might not be 973 * the same list, if the victim was using the secondary hash). 974 */ 975 TAILQ_REMOVE(pvoh, source_pvo, pvo_olink); 976 TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink); 977 TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink); 978 TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink); 979 pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr); 980 pmap_pte_set(pt, &source_pvo->pvo_pte); 981 victim_pvo->pvo_pmap->pm_evictions++; 982 source_pvo->pvo_pmap->pm_evictions--; 983 PVO_WHERE(victim_pvo, SPILL_UNSET); 984 PVO_WHERE(source_pvo, SPILL_SET); 985 986 PVO_PTEGIDX_CLR(victim_pvo); 987 PVO_PTEGIDX_SET(source_pvo, i); 988 PMAPCOUNT2(pmap_evcnt_ptes_primary[i]); 989 PMAPCOUNT(ptes_spilled); 990 PMAPCOUNT(ptes_evicted); 991 PMAPCOUNT(ptes_removed); 992 993 PMAP_PVO_CHECK(victim_pvo); 994 PMAP_PVO_CHECK(source_pvo); 995 return 1; 996 } 997 998 /* 999 * Restrict given range to physical memory 1000 */ 1001 void 1002 pmap_real_memory(paddr_t *start, psize_t *size) 1003 { 1004 struct mem_region *mp; 1005 1006 for (mp = mem; mp->size; mp++) { 1007 if (*start + *size > mp->start 1008 && *start < mp->start + mp->size) { 1009 if (*start < mp->start) { 1010 *size -= mp->start - *start; 1011 *start = mp->start; 1012 } 1013 if (*start + *size > mp->start + mp->size) 1014 *size = mp->start + mp->size - *start; 1015 return; 1016 } 1017 } 1018 *size = 0; 1019 } 1020 1021 /* 1022 * Initialize anything else for pmap handling. 1023 * Called during vm_init(). 1024 */ 1025 void 1026 pmap_init(void) 1027 { 1028 #ifdef __HAVE_PMAP_PHYSSEG 1029 struct pvo_tqhead *pvoh; 1030 int bank; 1031 long sz; 1032 char *attr; 1033 1034 pvoh = pmap_physseg.pvoh; 1035 attr = pmap_physseg.attrs; 1036 for (bank = 0; bank < vm_nphysseg; bank++) { 1037 sz = vm_physmem[bank].end - vm_physmem[bank].start; 1038 vm_physmem[bank].pmseg.pvoh = pvoh; 1039 vm_physmem[bank].pmseg.attrs = attr; 1040 for (; sz > 0; sz--, pvoh++, attr++) { 1041 TAILQ_INIT(pvoh); 1042 *attr = 0; 1043 } 1044 } 1045 #endif 1046 1047 pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry), 1048 sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl", 1049 &pmap_pool_mallocator); 1050 1051 pool_setlowat(&pmap_mpvo_pool, 1008); 1052 1053 pmap_initialized = 1; 1054 1055 #ifdef PMAPCOUNTERS 1056 evcnt_attach_static(&pmap_evcnt_mappings); 1057 evcnt_attach_static(&pmap_evcnt_mappings_replaced); 1058 evcnt_attach_static(&pmap_evcnt_unmappings); 1059 1060 evcnt_attach_static(&pmap_evcnt_kernel_mappings); 1061 evcnt_attach_static(&pmap_evcnt_kernel_unmappings); 1062 1063 evcnt_attach_static(&pmap_evcnt_exec_mappings); 1064 evcnt_attach_static(&pmap_evcnt_exec_cached); 1065 evcnt_attach_static(&pmap_evcnt_exec_synced); 1066 evcnt_attach_static(&pmap_evcnt_exec_synced_clear_modify); 1067 1068 evcnt_attach_static(&pmap_evcnt_exec_uncached_page_protect); 1069 evcnt_attach_static(&pmap_evcnt_exec_uncached_clear_modify); 1070 evcnt_attach_static(&pmap_evcnt_exec_uncached_zero_page); 1071 evcnt_attach_static(&pmap_evcnt_exec_uncached_copy_page); 1072 1073 evcnt_attach_static(&pmap_evcnt_zeroed_pages); 1074 evcnt_attach_static(&pmap_evcnt_copied_pages); 1075 evcnt_attach_static(&pmap_evcnt_idlezeroed_pages); 1076 1077 evcnt_attach_static(&pmap_evcnt_updates); 1078 evcnt_attach_static(&pmap_evcnt_collects); 1079 evcnt_attach_static(&pmap_evcnt_copies); 1080 1081 evcnt_attach_static(&pmap_evcnt_ptes_spilled); 1082 evcnt_attach_static(&pmap_evcnt_ptes_unspilled); 1083 evcnt_attach_static(&pmap_evcnt_ptes_evicted); 1084 evcnt_attach_static(&pmap_evcnt_ptes_removed); 1085 evcnt_attach_static(&pmap_evcnt_ptes_changed); 1086 evcnt_attach_static(&pmap_evcnt_ptes_primary[0]); 1087 evcnt_attach_static(&pmap_evcnt_ptes_primary[1]); 1088 evcnt_attach_static(&pmap_evcnt_ptes_primary[2]); 1089 evcnt_attach_static(&pmap_evcnt_ptes_primary[3]); 1090 evcnt_attach_static(&pmap_evcnt_ptes_primary[4]); 1091 evcnt_attach_static(&pmap_evcnt_ptes_primary[5]); 1092 evcnt_attach_static(&pmap_evcnt_ptes_primary[6]); 1093 evcnt_attach_static(&pmap_evcnt_ptes_primary[7]); 1094 evcnt_attach_static(&pmap_evcnt_ptes_secondary[0]); 1095 evcnt_attach_static(&pmap_evcnt_ptes_secondary[1]); 1096 evcnt_attach_static(&pmap_evcnt_ptes_secondary[2]); 1097 evcnt_attach_static(&pmap_evcnt_ptes_secondary[3]); 1098 evcnt_attach_static(&pmap_evcnt_ptes_secondary[4]); 1099 evcnt_attach_static(&pmap_evcnt_ptes_secondary[5]); 1100 evcnt_attach_static(&pmap_evcnt_ptes_secondary[6]); 1101 evcnt_attach_static(&pmap_evcnt_ptes_secondary[7]); 1102 #endif 1103 } 1104 1105 /* 1106 * How much virtual space does the kernel get? 1107 */ 1108 void 1109 pmap_virtual_space(vaddr_t *start, vaddr_t *end) 1110 { 1111 /* 1112 * For now, reserve one segment (minus some overhead) for kernel 1113 * virtual memory 1114 */ 1115 *start = VM_MIN_KERNEL_ADDRESS; 1116 *end = VM_MAX_KERNEL_ADDRESS; 1117 } 1118 1119 /* 1120 * Allocate, initialize, and return a new physical map. 1121 */ 1122 pmap_t 1123 pmap_create(void) 1124 { 1125 pmap_t pm; 1126 1127 pm = pool_get(&pmap_pool, PR_WAITOK); 1128 memset((caddr_t)pm, 0, sizeof *pm); 1129 pmap_pinit(pm); 1130 1131 DPRINTFN(CREATE,("pmap_create: pm %p:\n" 1132 "\t%06x %06x %06x %06x %06x %06x %06x %06x\n" 1133 "\t%06x %06x %06x %06x %06x %06x %06x %06x\n", pm, 1134 (unsigned int) pm->pm_sr[0], (unsigned int) pm->pm_sr[1], 1135 (unsigned int) pm->pm_sr[2], (unsigned int) pm->pm_sr[3], 1136 (unsigned int) pm->pm_sr[4], (unsigned int) pm->pm_sr[5], 1137 (unsigned int) pm->pm_sr[6], (unsigned int) pm->pm_sr[7], 1138 (unsigned int) pm->pm_sr[8], (unsigned int) pm->pm_sr[9], 1139 (unsigned int) pm->pm_sr[10], (unsigned int) pm->pm_sr[11], 1140 (unsigned int) pm->pm_sr[12], (unsigned int) pm->pm_sr[13], 1141 (unsigned int) pm->pm_sr[14], (unsigned int) pm->pm_sr[15])); 1142 return pm; 1143 } 1144 1145 /* 1146 * Initialize a preallocated and zeroed pmap structure. 1147 */ 1148 void 1149 pmap_pinit(pmap_t pm) 1150 { 1151 register_t entropy = MFTB(); 1152 register_t mask; 1153 int i; 1154 1155 /* 1156 * Allocate some segment registers for this pmap. 1157 */ 1158 pm->pm_refs = 1; 1159 for (i = 0; i < NPMAPS; i += VSID_NBPW) { 1160 static register_t pmap_vsidcontext; 1161 register_t hash; 1162 unsigned int n; 1163 1164 /* Create a new value by multiplying by a prime adding in 1165 * entropy from the timebase register. This is to make the 1166 * VSID more random so that the PT Hash function collides 1167 * less often. (note that the prime causes gcc to do shifts 1168 * instead of a multiply) 1169 */ 1170 pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy; 1171 hash = pmap_vsidcontext & (NPMAPS - 1); 1172 if (hash == 0) /* 0 is special, avoid it */ 1173 continue; 1174 n = hash >> 5; 1175 mask = 1L << (hash & (VSID_NBPW-1)); 1176 hash = pmap_vsidcontext; 1177 if (pmap_vsid_bitmap[n] & mask) { /* collision? */ 1178 /* anything free in this bucket? */ 1179 if (~pmap_vsid_bitmap[n] == 0) { 1180 entropy = hash >> PTE_VSID_SHFT; 1181 continue; 1182 } 1183 i = ffs(~pmap_vsid_bitmap[n]) - 1; 1184 mask = 1L << i; 1185 hash &= ~(VSID_NBPW-1); 1186 hash |= i; 1187 } 1188 hash &= PTE_VSID >> PTE_VSID_SHFT; 1189 pmap_vsid_bitmap[n] |= mask; 1190 pm->pm_vsid = hash; 1191 #ifndef PPC_OEA64 1192 for (i = 0; i < 16; i++) 1193 pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY | 1194 SR_NOEXEC; 1195 #endif 1196 return; 1197 } 1198 panic("pmap_pinit: out of segments"); 1199 } 1200 1201 /* 1202 * Add a reference to the given pmap. 1203 */ 1204 void 1205 pmap_reference(pmap_t pm) 1206 { 1207 pm->pm_refs++; 1208 } 1209 1210 /* 1211 * Retire the given pmap from service. 1212 * Should only be called if the map contains no valid mappings. 1213 */ 1214 void 1215 pmap_destroy(pmap_t pm) 1216 { 1217 if (--pm->pm_refs == 0) { 1218 pmap_release(pm); 1219 pool_put(&pmap_pool, pm); 1220 } 1221 } 1222 1223 /* 1224 * Release any resources held by the given physical map. 1225 * Called when a pmap initialized by pmap_pinit is being released. 1226 */ 1227 void 1228 pmap_release(pmap_t pm) 1229 { 1230 int idx, mask; 1231 1232 if (pm->pm_sr[0] == 0) 1233 panic("pmap_release"); 1234 idx = VSID_TO_HASH(pm->pm_vsid) & (NPMAPS-1); 1235 mask = 1 << (idx % VSID_NBPW); 1236 idx /= VSID_NBPW; 1237 pmap_vsid_bitmap[idx] &= ~mask; 1238 } 1239 1240 /* 1241 * Copy the range specified by src_addr/len 1242 * from the source map to the range dst_addr/len 1243 * in the destination map. 1244 * 1245 * This routine is only advisory and need not do anything. 1246 */ 1247 void 1248 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr, 1249 vsize_t len, vaddr_t src_addr) 1250 { 1251 PMAPCOUNT(copies); 1252 } 1253 1254 /* 1255 * Require that all active physical maps contain no 1256 * incorrect entries NOW. 1257 */ 1258 void 1259 pmap_update(struct pmap *pmap) 1260 { 1261 PMAPCOUNT(updates); 1262 TLBSYNC(); 1263 } 1264 1265 /* 1266 * Garbage collects the physical map system for 1267 * pages which are no longer used. 1268 * Success need not be guaranteed -- that is, there 1269 * may well be pages which are not referenced, but 1270 * others may be collected. 1271 * Called by the pageout daemon when pages are scarce. 1272 */ 1273 void 1274 pmap_collect(pmap_t pm) 1275 { 1276 PMAPCOUNT(collects); 1277 } 1278 1279 static __inline int 1280 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx) 1281 { 1282 int pteidx; 1283 /* 1284 * We can find the actual pte entry without searching by 1285 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9] 1286 * and by noticing the HID bit. 1287 */ 1288 pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo); 1289 if (pvo->pvo_pte.pte_hi & PTE_HID) 1290 pteidx ^= pmap_pteg_mask * 8; 1291 return pteidx; 1292 } 1293 1294 volatile struct pte * 1295 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx) 1296 { 1297 volatile struct pte *pt; 1298 1299 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1300 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) 1301 return NULL; 1302 #endif 1303 1304 /* 1305 * If we haven't been supplied the ptegidx, calculate it. 1306 */ 1307 if (pteidx == -1) { 1308 int ptegidx; 1309 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1310 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1311 } 1312 1313 pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7]; 1314 1315 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1316 return pt; 1317 #else 1318 if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) { 1319 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1320 "pvo but no valid pte index", pvo); 1321 } 1322 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) { 1323 panic("pmap_pvo_to_pte: pvo %p: has valid pte index in " 1324 "pvo but no valid pte", pvo); 1325 } 1326 1327 if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) { 1328 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) { 1329 #if defined(DEBUG) || defined(PMAPCHECK) 1330 pmap_pte_print(pt); 1331 #endif 1332 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1333 "pmap_pteg_table %p but invalid in pvo", 1334 pvo, pt); 1335 } 1336 if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) { 1337 #if defined(DEBUG) || defined(PMAPCHECK) 1338 pmap_pte_print(pt); 1339 #endif 1340 panic("pmap_pvo_to_pte: pvo %p: pvo pte does " 1341 "not match pte %p in pmap_pteg_table", 1342 pvo, pt); 1343 } 1344 return pt; 1345 } 1346 1347 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1348 #if defined(DEBUG) || defined(PMAPCHECK) 1349 pmap_pte_print(pt); 1350 #endif 1351 panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in " 1352 "pmap_pteg_table but valid in pvo", pvo, pt); 1353 } 1354 return NULL; 1355 #endif /* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */ 1356 } 1357 1358 struct pvo_entry * 1359 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p) 1360 { 1361 struct pvo_entry *pvo; 1362 int ptegidx; 1363 1364 va &= ~ADDR_POFF; 1365 ptegidx = va_to_pteg(pm, va); 1366 1367 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1368 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1369 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 1370 panic("pmap_pvo_find_va: invalid pvo %p on " 1371 "list %#x (%p)", pvo, ptegidx, 1372 &pmap_pvo_table[ptegidx]); 1373 #endif 1374 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1375 if (pteidx_p) 1376 *pteidx_p = pmap_pvo_pte_index(pvo, ptegidx); 1377 return pvo; 1378 } 1379 } 1380 return NULL; 1381 } 1382 1383 #if defined(DEBUG) || defined(PMAPCHECK) 1384 void 1385 pmap_pvo_check(const struct pvo_entry *pvo) 1386 { 1387 struct pvo_head *pvo_head; 1388 struct pvo_entry *pvo0; 1389 volatile struct pte *pt; 1390 int failed = 0; 1391 1392 if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH) 1393 panic("pmap_pvo_check: pvo %p: invalid address", pvo); 1394 1395 if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) { 1396 printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n", 1397 pvo, pvo->pvo_pmap); 1398 failed = 1; 1399 } 1400 1401 if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH || 1402 (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) { 1403 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1404 pvo, TAILQ_NEXT(pvo, pvo_olink)); 1405 failed = 1; 1406 } 1407 1408 if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH || 1409 (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) { 1410 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1411 pvo, LIST_NEXT(pvo, pvo_vlink)); 1412 failed = 1; 1413 } 1414 1415 if (pvo->pvo_vaddr & PVO_MANAGED) { 1416 pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL); 1417 } else { 1418 if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) { 1419 printf("pmap_pvo_check: pvo %p: non kernel address " 1420 "on kernel unmanaged list\n", pvo); 1421 failed = 1; 1422 } 1423 pvo_head = &pmap_pvo_kunmanaged; 1424 } 1425 LIST_FOREACH(pvo0, pvo_head, pvo_vlink) { 1426 if (pvo0 == pvo) 1427 break; 1428 } 1429 if (pvo0 == NULL) { 1430 printf("pmap_pvo_check: pvo %p: not present " 1431 "on its vlist head %p\n", pvo, pvo_head); 1432 failed = 1; 1433 } 1434 if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) { 1435 printf("pmap_pvo_check: pvo %p: not present " 1436 "on its olist head\n", pvo); 1437 failed = 1; 1438 } 1439 pt = pmap_pvo_to_pte(pvo, -1); 1440 if (pt == NULL) { 1441 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1442 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1443 "no PTE\n", pvo); 1444 failed = 1; 1445 } 1446 } else { 1447 if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] || 1448 (uintptr_t) pt >= 1449 (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) { 1450 printf("pmap_pvo_check: pvo %p: pte %p not in " 1451 "pteg table\n", pvo, pt); 1452 failed = 1; 1453 } 1454 if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) { 1455 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1456 "no PTE\n", pvo); 1457 failed = 1; 1458 } 1459 if (pvo->pvo_pte.pte_hi != pt->pte_hi) { 1460 printf("pmap_pvo_check: pvo %p: pte_hi differ: " 1461 "%#x/%#x\n", pvo, (unsigned int) pvo->pvo_pte.pte_hi, (unsigned int) pt->pte_hi); 1462 failed = 1; 1463 } 1464 if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) & 1465 (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) { 1466 printf("pmap_pvo_check: pvo %p: pte_lo differ: " 1467 "%#x/%#x\n", pvo, 1468 (unsigned int) (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)), 1469 (unsigned int) (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN))); 1470 failed = 1; 1471 } 1472 if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) { 1473 printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#lx" 1474 " doesn't not match PVO's VA %#lx\n", 1475 pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo)); 1476 failed = 1; 1477 } 1478 if (failed) 1479 pmap_pte_print(pt); 1480 } 1481 if (failed) 1482 panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo, 1483 pvo->pvo_pmap); 1484 } 1485 #endif /* DEBUG || PMAPCHECK */ 1486 1487 /* 1488 * This returns whether this is the first mapping of a page. 1489 */ 1490 int 1491 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head, 1492 vaddr_t va, paddr_t pa, register_t pte_lo, int flags) 1493 { 1494 struct pvo_entry *pvo; 1495 struct pvo_tqhead *pvoh; 1496 register_t msr; 1497 int ptegidx; 1498 int i; 1499 int poolflags = PR_NOWAIT; 1500 1501 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1502 if (pmap_pvo_remove_depth > 0) 1503 panic("pmap_pvo_enter: called while pmap_pvo_remove active!"); 1504 if (++pmap_pvo_enter_depth > 1) 1505 panic("pmap_pvo_enter: called recursively!"); 1506 #endif 1507 1508 /* 1509 * Compute the PTE Group index. 1510 */ 1511 va &= ~ADDR_POFF; 1512 ptegidx = va_to_pteg(pm, va); 1513 1514 msr = pmap_interrupts_off(); 1515 /* 1516 * Remove any existing mapping for this page. Reuse the 1517 * pvo entry if there a mapping. 1518 */ 1519 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1520 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1521 #ifdef DEBUG 1522 if ((pmapdebug & PMAPDEBUG_PVOENTER) && 1523 ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) & 1524 ~(PTE_REF|PTE_CHG)) == 0 && 1525 va < VM_MIN_KERNEL_ADDRESS) { 1526 printf("pmap_pvo_enter: pvo %p: dup %#x/%#lx\n", 1527 pvo, (unsigned int) pvo->pvo_pte.pte_lo, (unsigned int) pte_lo|pa); 1528 printf("pmap_pvo_enter: pte_hi=%#x sr=%#x\n", 1529 (unsigned int) pvo->pvo_pte.pte_hi, 1530 (unsigned int) pm->pm_sr[va >> ADDR_SR_SHFT]); 1531 pmap_pte_print(pmap_pvo_to_pte(pvo, -1)); 1532 #ifdef DDBX 1533 Debugger(); 1534 #endif 1535 } 1536 #endif 1537 PMAPCOUNT(mappings_replaced); 1538 pmap_pvo_remove(pvo, -1); 1539 break; 1540 } 1541 } 1542 1543 /* 1544 * If we aren't overwriting an mapping, try to allocate 1545 */ 1546 pmap_interrupts_restore(msr); 1547 pvo = pool_get(pl, poolflags); 1548 msr = pmap_interrupts_off(); 1549 if (pvo == NULL) { 1550 pvo = pmap_pvo_reclaim(pm); 1551 if (pvo == NULL) { 1552 if ((flags & PMAP_CANFAIL) == 0) 1553 panic("pmap_pvo_enter: failed"); 1554 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1555 pmap_pvo_enter_depth--; 1556 #endif 1557 pmap_interrupts_restore(msr); 1558 return ENOMEM; 1559 } 1560 } 1561 pvo->pvo_vaddr = va; 1562 pvo->pvo_pmap = pm; 1563 pvo->pvo_vaddr &= ~ADDR_POFF; 1564 if (flags & VM_PROT_EXECUTE) { 1565 PMAPCOUNT(exec_mappings); 1566 pvo_set_exec(pvo); 1567 } 1568 if (flags & PMAP_WIRED) 1569 pvo->pvo_vaddr |= PVO_WIRED; 1570 if (pvo_head != &pmap_pvo_kunmanaged) { 1571 pvo->pvo_vaddr |= PVO_MANAGED; 1572 PMAPCOUNT(mappings); 1573 } else { 1574 PMAPCOUNT(kernel_mappings); 1575 } 1576 pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo); 1577 1578 LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink); 1579 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1580 pvo->pvo_pmap->pm_stats.wired_count++; 1581 pvo->pvo_pmap->pm_stats.resident_count++; 1582 #if defined(DEBUG) 1583 if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) 1584 DPRINTFN(PVOENTER, 1585 ("pmap_pvo_enter: pvo %p: pm %p va %#lx pa %#lx\n", 1586 pvo, pm, va, pa)); 1587 #endif 1588 1589 /* 1590 * We hope this succeeds but it isn't required. 1591 */ 1592 pvoh = &pmap_pvo_table[ptegidx]; 1593 i = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 1594 if (i >= 0) { 1595 PVO_PTEGIDX_SET(pvo, i); 1596 PVO_WHERE(pvo, ENTER_INSERT); 1597 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 1598 ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]); 1599 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 1600 } else { 1601 /* 1602 * Since we didn't have room for this entry (which makes it 1603 * and evicted entry), place it at the head of the list. 1604 */ 1605 TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink); 1606 PMAPCOUNT(ptes_evicted); 1607 pm->pm_evictions++; 1608 /* 1609 * If this is a kernel page, make sure it's active. 1610 */ 1611 if (pm == pmap_kernel()) { 1612 i = pmap_pte_spill(pm, va, FALSE); 1613 KASSERT(i); 1614 } 1615 } 1616 PMAP_PVO_CHECK(pvo); /* sanity check */ 1617 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1618 pmap_pvo_enter_depth--; 1619 #endif 1620 pmap_interrupts_restore(msr); 1621 return 0; 1622 } 1623 1624 void 1625 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx) 1626 { 1627 volatile struct pte *pt; 1628 int ptegidx; 1629 1630 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1631 if (++pmap_pvo_remove_depth > 1) 1632 panic("pmap_pvo_remove: called recursively!"); 1633 #endif 1634 1635 /* 1636 * If we haven't been supplied the ptegidx, calculate it. 1637 */ 1638 if (pteidx == -1) { 1639 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1640 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1641 } else { 1642 ptegidx = pteidx >> 3; 1643 if (pvo->pvo_pte.pte_hi & PTE_HID) 1644 ptegidx ^= pmap_pteg_mask; 1645 } 1646 PMAP_PVO_CHECK(pvo); /* sanity check */ 1647 1648 /* 1649 * If there is an active pte entry, we need to deactivate it 1650 * (and save the ref & chg bits). 1651 */ 1652 pt = pmap_pvo_to_pte(pvo, pteidx); 1653 if (pt != NULL) { 1654 pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 1655 PVO_WHERE(pvo, REMOVE); 1656 PVO_PTEGIDX_CLR(pvo); 1657 PMAPCOUNT(ptes_removed); 1658 } else { 1659 KASSERT(pvo->pvo_pmap->pm_evictions > 0); 1660 pvo->pvo_pmap->pm_evictions--; 1661 } 1662 1663 /* 1664 * Account for executable mappings. 1665 */ 1666 if (PVO_ISEXECUTABLE(pvo)) 1667 pvo_clear_exec(pvo); 1668 1669 /* 1670 * Update our statistics. 1671 */ 1672 pvo->pvo_pmap->pm_stats.resident_count--; 1673 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1674 pvo->pvo_pmap->pm_stats.wired_count--; 1675 1676 /* 1677 * Save the REF/CHG bits into their cache if the page is managed. 1678 */ 1679 if (pvo->pvo_vaddr & PVO_MANAGED) { 1680 register_t ptelo = pvo->pvo_pte.pte_lo; 1681 struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN); 1682 1683 if (pg != NULL) { 1684 pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG)); 1685 } 1686 PMAPCOUNT(unmappings); 1687 } else { 1688 PMAPCOUNT(kernel_unmappings); 1689 } 1690 1691 /* 1692 * Remove the PVO from its lists and return it to the pool. 1693 */ 1694 LIST_REMOVE(pvo, pvo_vlink); 1695 TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink); 1696 pool_put(pvo->pvo_vaddr & PVO_MANAGED 1697 ? &pmap_mpvo_pool : &pmap_upvo_pool, pvo); 1698 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1699 pmap_pvo_remove_depth--; 1700 #endif 1701 } 1702 1703 /* 1704 * Mark a mapping as executable. 1705 * If this is the first executable mapping in the segment, 1706 * clear the noexec flag. 1707 */ 1708 STATIC void 1709 pvo_set_exec(struct pvo_entry *pvo) 1710 { 1711 struct pmap *pm = pvo->pvo_pmap; 1712 1713 if (pm == pmap_kernel() || PVO_ISEXECUTABLE(pvo)) { 1714 return; 1715 } 1716 pvo->pvo_vaddr |= PVO_EXECUTABLE; 1717 #ifdef PPC_OEA 1718 { 1719 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1720 if (pm->pm_exec[sr]++ == 0) { 1721 pm->pm_sr[sr] &= ~SR_NOEXEC; 1722 } 1723 } 1724 #endif 1725 } 1726 1727 /* 1728 * Mark a mapping as non-executable. 1729 * If this was the last executable mapping in the segment, 1730 * set the noexec flag. 1731 */ 1732 STATIC void 1733 pvo_clear_exec(struct pvo_entry *pvo) 1734 { 1735 struct pmap *pm = pvo->pvo_pmap; 1736 1737 if (pm == pmap_kernel() || !PVO_ISEXECUTABLE(pvo)) { 1738 return; 1739 } 1740 pvo->pvo_vaddr &= ~PVO_EXECUTABLE; 1741 #ifdef PPC_OEA 1742 { 1743 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1744 if (--pm->pm_exec[sr] == 0) { 1745 pm->pm_sr[sr] |= SR_NOEXEC; 1746 } 1747 } 1748 #endif 1749 } 1750 1751 /* 1752 * Insert physical page at pa into the given pmap at virtual address va. 1753 */ 1754 int 1755 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags) 1756 { 1757 struct mem_region *mp; 1758 struct pvo_head *pvo_head; 1759 struct vm_page *pg; 1760 struct pool *pl; 1761 register_t pte_lo; 1762 int error; 1763 u_int pvo_flags; 1764 u_int was_exec = 0; 1765 1766 if (__predict_false(!pmap_initialized)) { 1767 pvo_head = &pmap_pvo_kunmanaged; 1768 pl = &pmap_upvo_pool; 1769 pvo_flags = 0; 1770 pg = NULL; 1771 was_exec = PTE_EXEC; 1772 } else { 1773 pvo_head = pa_to_pvoh(pa, &pg); 1774 pl = &pmap_mpvo_pool; 1775 pvo_flags = PVO_MANAGED; 1776 } 1777 1778 DPRINTFN(ENTER, 1779 ("pmap_enter(%p, 0x%lx, 0x%lx, 0x%x, 0x%x):", 1780 pm, va, pa, prot, flags)); 1781 1782 /* 1783 * If this is a managed page, and it's the first reference to the 1784 * page clear the execness of the page. Otherwise fetch the execness. 1785 */ 1786 if (pg != NULL) 1787 was_exec = pmap_attr_fetch(pg) & PTE_EXEC; 1788 1789 DPRINTFN(ENTER, (" was_exec=%d", was_exec)); 1790 1791 /* 1792 * Assume the page is cache inhibited and access is guarded unless 1793 * it's in our available memory array. If it is in the memory array, 1794 * asssume it's in memory coherent memory. 1795 */ 1796 pte_lo = PTE_IG; 1797 if ((flags & PMAP_NC) == 0) { 1798 for (mp = mem; mp->size; mp++) { 1799 if (pa >= mp->start && pa < mp->start + mp->size) { 1800 pte_lo = PTE_M; 1801 break; 1802 } 1803 } 1804 } 1805 1806 if (prot & VM_PROT_WRITE) 1807 pte_lo |= PTE_BW; 1808 else 1809 pte_lo |= PTE_BR; 1810 1811 /* 1812 * If this was in response to a fault, "pre-fault" the PTE's 1813 * changed/referenced bit appropriately. 1814 */ 1815 if (flags & VM_PROT_WRITE) 1816 pte_lo |= PTE_CHG; 1817 if (flags & (VM_PROT_READ|VM_PROT_WRITE)) 1818 pte_lo |= PTE_REF; 1819 1820 /* 1821 * We need to know if this page can be executable 1822 */ 1823 flags |= (prot & VM_PROT_EXECUTE); 1824 1825 /* 1826 * Record mapping for later back-translation and pte spilling. 1827 * This will overwrite any existing mapping. 1828 */ 1829 error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags); 1830 1831 /* 1832 * Flush the real page from the instruction cache if this page is 1833 * mapped executable and cacheable and has not been flushed since 1834 * the last time it was modified. 1835 */ 1836 if (error == 0 && 1837 (flags & VM_PROT_EXECUTE) && 1838 (pte_lo & PTE_I) == 0 && 1839 was_exec == 0) { 1840 DPRINTFN(ENTER, (" syncicache")); 1841 PMAPCOUNT(exec_synced); 1842 pmap_syncicache(pa, PAGE_SIZE); 1843 if (pg != NULL) { 1844 pmap_attr_save(pg, PTE_EXEC); 1845 PMAPCOUNT(exec_cached); 1846 #if defined(DEBUG) || defined(PMAPDEBUG) 1847 if (pmapdebug & PMAPDEBUG_ENTER) 1848 printf(" marked-as-exec"); 1849 else if (pmapdebug & PMAPDEBUG_EXEC) 1850 printf("[pmap_enter: %#lx: marked-as-exec]\n", 1851 pg->phys_addr); 1852 1853 #endif 1854 } 1855 } 1856 1857 DPRINTFN(ENTER, (": error=%d\n", error)); 1858 1859 return error; 1860 } 1861 1862 void 1863 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot) 1864 { 1865 struct mem_region *mp; 1866 register_t pte_lo; 1867 int error; 1868 1869 if (va < VM_MIN_KERNEL_ADDRESS) 1870 panic("pmap_kenter_pa: attempt to enter " 1871 "non-kernel address %#lx!", va); 1872 1873 DPRINTFN(KENTER, 1874 ("pmap_kenter_pa(%#lx,%#lx,%#x)\n", va, pa, prot)); 1875 1876 /* 1877 * Assume the page is cache inhibited and access is guarded unless 1878 * it's in our available memory array. If it is in the memory array, 1879 * asssume it's in memory coherent memory. 1880 */ 1881 pte_lo = PTE_IG; 1882 if ((prot & PMAP_NC) == 0) { 1883 for (mp = mem; mp->size; mp++) { 1884 if (pa >= mp->start && pa < mp->start + mp->size) { 1885 pte_lo = PTE_M; 1886 break; 1887 } 1888 } 1889 } 1890 1891 if (prot & VM_PROT_WRITE) 1892 pte_lo |= PTE_BW; 1893 else 1894 pte_lo |= PTE_BR; 1895 1896 /* 1897 * We don't care about REF/CHG on PVOs on the unmanaged list. 1898 */ 1899 error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool, 1900 &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED); 1901 1902 if (error != 0) 1903 panic("pmap_kenter_pa: failed to enter va %#lx pa %#lx: %d", 1904 va, pa, error); 1905 } 1906 1907 void 1908 pmap_kremove(vaddr_t va, vsize_t len) 1909 { 1910 if (va < VM_MIN_KERNEL_ADDRESS) 1911 panic("pmap_kremove: attempt to remove " 1912 "non-kernel address %#lx!", va); 1913 1914 DPRINTFN(KREMOVE,("pmap_kremove(%#lx,%#lx)\n", va, len)); 1915 pmap_remove(pmap_kernel(), va, va + len); 1916 } 1917 1918 /* 1919 * Remove the given range of mapping entries. 1920 */ 1921 void 1922 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva) 1923 { 1924 struct pvo_entry *pvo; 1925 register_t msr; 1926 int pteidx; 1927 1928 msr = pmap_interrupts_off(); 1929 for (; va < endva; va += PAGE_SIZE) { 1930 pvo = pmap_pvo_find_va(pm, va, &pteidx); 1931 if (pvo != NULL) { 1932 pmap_pvo_remove(pvo, pteidx); 1933 } 1934 } 1935 pmap_interrupts_restore(msr); 1936 } 1937 1938 /* 1939 * Get the physical page address for the given pmap/virtual address. 1940 */ 1941 boolean_t 1942 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap) 1943 { 1944 struct pvo_entry *pvo; 1945 register_t msr; 1946 1947 /* 1948 * If this is a kernel pmap lookup, also check the battable 1949 * and if we get a hit, translate the VA to a PA using the 1950 * BAT entries. Don't check for VM_MAX_KENREL_ADDRESS is 1951 * that will wrap back to 0. 1952 */ 1953 if (pm == pmap_kernel() && 1954 (va < VM_MIN_KERNEL_ADDRESS || 1955 (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) { 1956 register_t batu = battable[va >> ADDR_SR_SHFT].batu; 1957 KASSERT((va >> ADDR_SR_SHFT) != USER_SR); 1958 if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) { 1959 register_t batl = battable[va >> ADDR_SR_SHFT].batl; 1960 register_t mask = (~(batu & BAT_BL) << 15) & ~0x1ffffL; 1961 *pap = (batl & mask) | (va & ~mask); 1962 return TRUE; 1963 } 1964 return FALSE; 1965 } 1966 1967 msr = pmap_interrupts_off(); 1968 pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL); 1969 if (pvo != NULL) { 1970 PMAP_PVO_CHECK(pvo); /* sanity check */ 1971 *pap = (pvo->pvo_pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF); 1972 } 1973 pmap_interrupts_restore(msr); 1974 return pvo != NULL; 1975 } 1976 1977 /* 1978 * Lower the protection on the specified range of this pmap. 1979 */ 1980 void 1981 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot) 1982 { 1983 struct pvo_entry *pvo; 1984 volatile struct pte *pt; 1985 register_t msr; 1986 int pteidx; 1987 1988 /* 1989 * Since this routine only downgrades protection, we should 1990 * always be called with at least one bit not set. 1991 */ 1992 KASSERT(prot != VM_PROT_ALL); 1993 1994 /* 1995 * If there is no protection, this is equivalent to 1996 * remove the pmap from the pmap. 1997 */ 1998 if ((prot & VM_PROT_READ) == 0) { 1999 pmap_remove(pm, va, endva); 2000 return; 2001 } 2002 2003 msr = pmap_interrupts_off(); 2004 for (; va < endva; va += PAGE_SIZE) { 2005 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2006 if (pvo == NULL) 2007 continue; 2008 PMAP_PVO_CHECK(pvo); /* sanity check */ 2009 2010 /* 2011 * Revoke executable if asked to do so. 2012 */ 2013 if ((prot & VM_PROT_EXECUTE) == 0) 2014 pvo_clear_exec(pvo); 2015 2016 #if 0 2017 /* 2018 * If the page is already read-only, no change 2019 * needs to be made. 2020 */ 2021 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) 2022 continue; 2023 #endif 2024 /* 2025 * Grab the PTE pointer before we diddle with 2026 * the cached PTE copy. 2027 */ 2028 pt = pmap_pvo_to_pte(pvo, pteidx); 2029 /* 2030 * Change the protection of the page. 2031 */ 2032 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2033 pvo->pvo_pte.pte_lo |= PTE_BR; 2034 2035 /* 2036 * If the PVO is in the page table, update 2037 * that pte at well. 2038 */ 2039 if (pt != NULL) { 2040 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2041 PVO_WHERE(pvo, PMAP_PROTECT); 2042 PMAPCOUNT(ptes_changed); 2043 } 2044 2045 PMAP_PVO_CHECK(pvo); /* sanity check */ 2046 } 2047 pmap_interrupts_restore(msr); 2048 } 2049 2050 void 2051 pmap_unwire(pmap_t pm, vaddr_t va) 2052 { 2053 struct pvo_entry *pvo; 2054 register_t msr; 2055 2056 msr = pmap_interrupts_off(); 2057 pvo = pmap_pvo_find_va(pm, va, NULL); 2058 if (pvo != NULL) { 2059 if (pvo->pvo_vaddr & PVO_WIRED) { 2060 pvo->pvo_vaddr &= ~PVO_WIRED; 2061 pm->pm_stats.wired_count--; 2062 } 2063 PMAP_PVO_CHECK(pvo); /* sanity check */ 2064 } 2065 pmap_interrupts_restore(msr); 2066 } 2067 2068 /* 2069 * Lower the protection on the specified physical page. 2070 */ 2071 void 2072 pmap_page_protect(struct vm_page *pg, vm_prot_t prot) 2073 { 2074 struct pvo_head *pvo_head; 2075 struct pvo_entry *pvo, *next_pvo; 2076 volatile struct pte *pt; 2077 register_t msr; 2078 2079 KASSERT(prot != VM_PROT_ALL); 2080 msr = pmap_interrupts_off(); 2081 2082 /* 2083 * When UVM reuses a page, it does a pmap_page_protect with 2084 * VM_PROT_NONE. At that point, we can clear the exec flag 2085 * since we know the page will have different contents. 2086 */ 2087 if ((prot & VM_PROT_READ) == 0) { 2088 DPRINTFN(EXEC, ("[pmap_page_protect: %#lx: clear-exec]\n", 2089 pg->phys_addr)); 2090 if (pmap_attr_fetch(pg) & PTE_EXEC) { 2091 PMAPCOUNT(exec_uncached_page_protect); 2092 pmap_attr_clear(pg, PTE_EXEC); 2093 } 2094 } 2095 2096 pvo_head = vm_page_to_pvoh(pg); 2097 for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) { 2098 next_pvo = LIST_NEXT(pvo, pvo_vlink); 2099 PMAP_PVO_CHECK(pvo); /* sanity check */ 2100 2101 /* 2102 * Downgrading to no mapping at all, we just remove the entry. 2103 */ 2104 if ((prot & VM_PROT_READ) == 0) { 2105 pmap_pvo_remove(pvo, -1); 2106 continue; 2107 } 2108 2109 /* 2110 * If EXEC permission is being revoked, just clear the 2111 * flag in the PVO. 2112 */ 2113 if ((prot & VM_PROT_EXECUTE) == 0) 2114 pvo_clear_exec(pvo); 2115 2116 /* 2117 * If this entry is already RO, don't diddle with the 2118 * page table. 2119 */ 2120 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) { 2121 PMAP_PVO_CHECK(pvo); 2122 continue; 2123 } 2124 2125 /* 2126 * Grab the PTE before the we diddle the bits so 2127 * pvo_to_pte can verify the pte contents are as 2128 * expected. 2129 */ 2130 pt = pmap_pvo_to_pte(pvo, -1); 2131 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2132 pvo->pvo_pte.pte_lo |= PTE_BR; 2133 if (pt != NULL) { 2134 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2135 PVO_WHERE(pvo, PMAP_PAGE_PROTECT); 2136 PMAPCOUNT(ptes_changed); 2137 } 2138 PMAP_PVO_CHECK(pvo); /* sanity check */ 2139 } 2140 pmap_interrupts_restore(msr); 2141 } 2142 2143 /* 2144 * Activate the address space for the specified process. If the process 2145 * is the current process, load the new MMU context. 2146 */ 2147 void 2148 pmap_activate(struct lwp *l) 2149 { 2150 struct pcb *pcb = &l->l_addr->u_pcb; 2151 pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap; 2152 2153 DPRINTFN(ACTIVATE, 2154 ("pmap_activate: lwp %p (curlwp %p)\n", l, curlwp)); 2155 2156 /* 2157 * XXX Normally performed in cpu_fork(). 2158 */ 2159 pcb->pcb_pm = pmap; 2160 2161 /* 2162 * In theory, the SR registers need only be valid on return 2163 * to user space wait to do them there. 2164 */ 2165 if (l == curlwp) { 2166 /* Store pointer to new current pmap. */ 2167 curpm = pmap; 2168 } 2169 } 2170 2171 /* 2172 * Deactivate the specified process's address space. 2173 */ 2174 void 2175 pmap_deactivate(struct lwp *l) 2176 { 2177 } 2178 2179 boolean_t 2180 pmap_query_bit(struct vm_page *pg, int ptebit) 2181 { 2182 struct pvo_entry *pvo; 2183 volatile struct pte *pt; 2184 register_t msr; 2185 2186 if (pmap_attr_fetch(pg) & ptebit) 2187 return TRUE; 2188 2189 msr = pmap_interrupts_off(); 2190 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2191 PMAP_PVO_CHECK(pvo); /* sanity check */ 2192 /* 2193 * See if we saved the bit off. If so cache, it and return 2194 * success. 2195 */ 2196 if (pvo->pvo_pte.pte_lo & ptebit) { 2197 pmap_attr_save(pg, ptebit); 2198 PMAP_PVO_CHECK(pvo); /* sanity check */ 2199 pmap_interrupts_restore(msr); 2200 return TRUE; 2201 } 2202 } 2203 /* 2204 * No luck, now go thru the hard part of looking at the ptes 2205 * themselves. Sync so any pending REF/CHG bits are flushed 2206 * to the PTEs. 2207 */ 2208 SYNC(); 2209 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2210 PMAP_PVO_CHECK(pvo); /* sanity check */ 2211 /* 2212 * See if this pvo have a valid PTE. If so, fetch the 2213 * REF/CHG bits from the valid PTE. If the appropriate 2214 * ptebit is set, cache, it and return success. 2215 */ 2216 pt = pmap_pvo_to_pte(pvo, -1); 2217 if (pt != NULL) { 2218 pmap_pte_synch(pt, &pvo->pvo_pte); 2219 if (pvo->pvo_pte.pte_lo & ptebit) { 2220 pmap_attr_save(pg, ptebit); 2221 PMAP_PVO_CHECK(pvo); /* sanity check */ 2222 pmap_interrupts_restore(msr); 2223 return TRUE; 2224 } 2225 } 2226 } 2227 pmap_interrupts_restore(msr); 2228 return FALSE; 2229 } 2230 2231 boolean_t 2232 pmap_clear_bit(struct vm_page *pg, int ptebit) 2233 { 2234 struct pvo_head *pvoh = vm_page_to_pvoh(pg); 2235 struct pvo_entry *pvo; 2236 volatile struct pte *pt; 2237 register_t msr; 2238 int rv = 0; 2239 2240 msr = pmap_interrupts_off(); 2241 2242 /* 2243 * Fetch the cache value 2244 */ 2245 rv |= pmap_attr_fetch(pg); 2246 2247 /* 2248 * Clear the cached value. 2249 */ 2250 pmap_attr_clear(pg, ptebit); 2251 2252 /* 2253 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we 2254 * can reset the right ones). Note that since the pvo entries and 2255 * list heads are accessed via BAT0 and are never placed in the 2256 * page table, we don't have to worry about further accesses setting 2257 * the REF/CHG bits. 2258 */ 2259 SYNC(); 2260 2261 /* 2262 * For each pvo entry, clear pvo's ptebit. If this pvo have a 2263 * valid PTE. If so, clear the ptebit from the valid PTE. 2264 */ 2265 LIST_FOREACH(pvo, pvoh, pvo_vlink) { 2266 PMAP_PVO_CHECK(pvo); /* sanity check */ 2267 pt = pmap_pvo_to_pte(pvo, -1); 2268 if (pt != NULL) { 2269 /* 2270 * Only sync the PTE if the bit we are looking 2271 * for is not already set. 2272 */ 2273 if ((pvo->pvo_pte.pte_lo & ptebit) == 0) 2274 pmap_pte_synch(pt, &pvo->pvo_pte); 2275 /* 2276 * If the bit we are looking for was already set, 2277 * clear that bit in the pte. 2278 */ 2279 if (pvo->pvo_pte.pte_lo & ptebit) 2280 pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit); 2281 } 2282 rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF); 2283 pvo->pvo_pte.pte_lo &= ~ptebit; 2284 PMAP_PVO_CHECK(pvo); /* sanity check */ 2285 } 2286 pmap_interrupts_restore(msr); 2287 2288 /* 2289 * If we are clearing the modify bit and this page was marked EXEC 2290 * and the user of the page thinks the page was modified, then we 2291 * need to clean it from the icache if it's mapped or clear the EXEC 2292 * bit if it's not mapped. The page itself might not have the CHG 2293 * bit set if the modification was done via DMA to the page. 2294 */ 2295 if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) { 2296 if (LIST_EMPTY(pvoh)) { 2297 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: clear-exec]\n", 2298 pg->phys_addr)); 2299 pmap_attr_clear(pg, PTE_EXEC); 2300 PMAPCOUNT(exec_uncached_clear_modify); 2301 } else { 2302 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: syncicache]\n", 2303 pg->phys_addr)); 2304 pmap_syncicache(pg->phys_addr, PAGE_SIZE); 2305 PMAPCOUNT(exec_synced_clear_modify); 2306 } 2307 } 2308 return (rv & ptebit) != 0; 2309 } 2310 2311 void 2312 pmap_procwr(struct proc *p, vaddr_t va, size_t len) 2313 { 2314 struct pvo_entry *pvo; 2315 size_t offset = va & ADDR_POFF; 2316 int s; 2317 2318 s = splvm(); 2319 while (len > 0) { 2320 size_t seglen = PAGE_SIZE - offset; 2321 if (seglen > len) 2322 seglen = len; 2323 pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL); 2324 if (pvo != NULL && PVO_ISEXECUTABLE(pvo)) { 2325 pmap_syncicache( 2326 (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen); 2327 PMAP_PVO_CHECK(pvo); 2328 } 2329 va += seglen; 2330 len -= seglen; 2331 offset = 0; 2332 } 2333 splx(s); 2334 } 2335 2336 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 2337 void 2338 pmap_pte_print(volatile struct pte *pt) 2339 { 2340 printf("PTE %p: ", pt); 2341 /* High word: */ 2342 printf("0x%08lx: [", pt->pte_hi); 2343 printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i'); 2344 printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-'); 2345 printf("0x%06lx 0x%02lx", 2346 (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT, 2347 pt->pte_hi & PTE_API); 2348 printf(" (va 0x%08lx)] ", pmap_pte_to_va(pt)); 2349 /* Low word: */ 2350 printf(" 0x%08lx: [", pt->pte_lo); 2351 printf("0x%05lx... ", pt->pte_lo >> 12); 2352 printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u'); 2353 printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n'); 2354 printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.'); 2355 printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.'); 2356 printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.'); 2357 printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.'); 2358 switch (pt->pte_lo & PTE_PP) { 2359 case PTE_BR: printf("br]\n"); break; 2360 case PTE_BW: printf("bw]\n"); break; 2361 case PTE_SO: printf("so]\n"); break; 2362 case PTE_SW: printf("sw]\n"); break; 2363 } 2364 } 2365 #endif 2366 2367 #if defined(DDB) 2368 void 2369 pmap_pteg_check(void) 2370 { 2371 volatile struct pte *pt; 2372 int i; 2373 int ptegidx; 2374 u_int p_valid = 0; 2375 u_int s_valid = 0; 2376 u_int invalid = 0; 2377 2378 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2379 for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) { 2380 if (pt->pte_hi & PTE_VALID) { 2381 if (pt->pte_hi & PTE_HID) 2382 s_valid++; 2383 else 2384 p_valid++; 2385 } else 2386 invalid++; 2387 } 2388 } 2389 printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n", 2390 p_valid, p_valid, s_valid, s_valid, 2391 invalid, invalid); 2392 } 2393 2394 void 2395 pmap_print_mmuregs(void) 2396 { 2397 int i; 2398 u_int cpuvers; 2399 #ifndef PPC_OEA64 2400 vaddr_t addr; 2401 register_t soft_sr[16]; 2402 #endif 2403 struct bat soft_ibat[4]; 2404 struct bat soft_dbat[4]; 2405 register_t sdr1; 2406 2407 cpuvers = MFPVR() >> 16; 2408 2409 __asm __volatile ("mfsdr1 %0" : "=r"(sdr1)); 2410 #ifndef PPC_OEA64 2411 addr = 0; 2412 for (i=0; i<16; i++) { 2413 soft_sr[i] = MFSRIN(addr); 2414 addr += (1 << ADDR_SR_SHFT); 2415 } 2416 #endif 2417 2418 /* read iBAT (601: uBAT) registers */ 2419 __asm __volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu)); 2420 __asm __volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl)); 2421 __asm __volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu)); 2422 __asm __volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl)); 2423 __asm __volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu)); 2424 __asm __volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl)); 2425 __asm __volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu)); 2426 __asm __volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl)); 2427 2428 2429 if (cpuvers != MPC601) { 2430 /* read dBAT registers */ 2431 __asm __volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu)); 2432 __asm __volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl)); 2433 __asm __volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu)); 2434 __asm __volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl)); 2435 __asm __volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu)); 2436 __asm __volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl)); 2437 __asm __volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu)); 2438 __asm __volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl)); 2439 } 2440 2441 printf("SDR1:\t0x%lx\n", (long) sdr1); 2442 #ifndef PPC_OEA64 2443 printf("SR[]:\t"); 2444 for (i=0; i<4; i++) 2445 printf("0x%08lx, ", soft_sr[i]); 2446 printf("\n\t"); 2447 for ( ; i<8; i++) 2448 printf("0x%08lx, ", soft_sr[i]); 2449 printf("\n\t"); 2450 for ( ; i<12; i++) 2451 printf("0x%08lx, ", soft_sr[i]); 2452 printf("\n\t"); 2453 for ( ; i<16; i++) 2454 printf("0x%08lx, ", soft_sr[i]); 2455 printf("\n"); 2456 #endif 2457 2458 printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i'); 2459 for (i=0; i<4; i++) { 2460 printf("0x%08lx 0x%08lx, ", 2461 soft_ibat[i].batu, soft_ibat[i].batl); 2462 if (i == 1) 2463 printf("\n\t"); 2464 } 2465 if (cpuvers != MPC601) { 2466 printf("\ndBAT[]:\t"); 2467 for (i=0; i<4; i++) { 2468 printf("0x%08lx 0x%08lx, ", 2469 soft_dbat[i].batu, soft_dbat[i].batl); 2470 if (i == 1) 2471 printf("\n\t"); 2472 } 2473 } 2474 printf("\n"); 2475 } 2476 2477 void 2478 pmap_print_pte(pmap_t pm, vaddr_t va) 2479 { 2480 struct pvo_entry *pvo; 2481 volatile struct pte *pt; 2482 int pteidx; 2483 2484 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2485 if (pvo != NULL) { 2486 pt = pmap_pvo_to_pte(pvo, pteidx); 2487 if (pt != NULL) { 2488 printf("VA %#lx -> %p -> %s %#lx, %#lx\n", 2489 va, pt, 2490 pt->pte_hi & PTE_HID ? "(sec)" : "(pri)", 2491 pt->pte_hi, pt->pte_lo); 2492 } else { 2493 printf("No valid PTE found\n"); 2494 } 2495 } else { 2496 printf("Address not in pmap\n"); 2497 } 2498 } 2499 2500 void 2501 pmap_pteg_dist(void) 2502 { 2503 struct pvo_entry *pvo; 2504 int ptegidx; 2505 int depth; 2506 int max_depth = 0; 2507 unsigned int depths[64]; 2508 2509 memset(depths, 0, sizeof(depths)); 2510 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2511 depth = 0; 2512 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2513 depth++; 2514 } 2515 if (depth > max_depth) 2516 max_depth = depth; 2517 if (depth > 63) 2518 depth = 63; 2519 depths[depth]++; 2520 } 2521 2522 for (depth = 0; depth < 64; depth++) { 2523 printf(" [%2d]: %8u", depth, depths[depth]); 2524 if ((depth & 3) == 3) 2525 printf("\n"); 2526 if (depth == max_depth) 2527 break; 2528 } 2529 if ((depth & 3) != 3) 2530 printf("\n"); 2531 printf("Max depth found was %d\n", max_depth); 2532 } 2533 #endif /* DEBUG */ 2534 2535 #if defined(PMAPCHECK) || defined(DEBUG) 2536 void 2537 pmap_pvo_verify(void) 2538 { 2539 int ptegidx; 2540 int s; 2541 2542 s = splvm(); 2543 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2544 struct pvo_entry *pvo; 2545 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2546 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 2547 panic("pmap_pvo_verify: invalid pvo %p " 2548 "on list %#x", pvo, ptegidx); 2549 pmap_pvo_check(pvo); 2550 } 2551 } 2552 splx(s); 2553 } 2554 #endif /* PMAPCHECK */ 2555 2556 2557 void * 2558 pmap_pool_ualloc(struct pool *pp, int flags) 2559 { 2560 struct pvo_page *pvop; 2561 2562 pvop = SIMPLEQ_FIRST(&pmap_upvop_head); 2563 if (pvop != NULL) { 2564 pmap_upvop_free--; 2565 SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link); 2566 return pvop; 2567 } 2568 if (uvm.page_init_done != TRUE) { 2569 return (void *) uvm_pageboot_alloc(PAGE_SIZE); 2570 } 2571 return pmap_pool_malloc(pp, flags); 2572 } 2573 2574 void * 2575 pmap_pool_malloc(struct pool *pp, int flags) 2576 { 2577 struct pvo_page *pvop; 2578 struct vm_page *pg; 2579 2580 pvop = SIMPLEQ_FIRST(&pmap_mpvop_head); 2581 if (pvop != NULL) { 2582 pmap_mpvop_free--; 2583 SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link); 2584 return pvop; 2585 } 2586 again: 2587 pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE, 2588 UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256); 2589 if (__predict_false(pg == NULL)) { 2590 if (flags & PR_WAITOK) { 2591 uvm_wait("plpg"); 2592 goto again; 2593 } else { 2594 return (0); 2595 } 2596 } 2597 return (void *) VM_PAGE_TO_PHYS(pg); 2598 } 2599 2600 void 2601 pmap_pool_ufree(struct pool *pp, void *va) 2602 { 2603 struct pvo_page *pvop; 2604 #if 0 2605 if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) { 2606 pmap_pool_mfree(va, size, tag); 2607 return; 2608 } 2609 #endif 2610 pvop = va; 2611 SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link); 2612 pmap_upvop_free++; 2613 if (pmap_upvop_free > pmap_upvop_maxfree) 2614 pmap_upvop_maxfree = pmap_upvop_free; 2615 } 2616 2617 void 2618 pmap_pool_mfree(struct pool *pp, void *va) 2619 { 2620 struct pvo_page *pvop; 2621 2622 pvop = va; 2623 SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link); 2624 pmap_mpvop_free++; 2625 if (pmap_mpvop_free > pmap_mpvop_maxfree) 2626 pmap_mpvop_maxfree = pmap_mpvop_free; 2627 #if 0 2628 uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va)); 2629 #endif 2630 } 2631 2632 /* 2633 * This routine in bootstraping to steal to-be-managed memory (which will 2634 * then be unmanaged). We use it to grab from the first 256MB for our 2635 * pmap needs and above 256MB for other stuff. 2636 */ 2637 vaddr_t 2638 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp) 2639 { 2640 vsize_t size; 2641 vaddr_t va; 2642 paddr_t pa = 0; 2643 int npgs, bank; 2644 struct vm_physseg *ps; 2645 2646 if (uvm.page_init_done == TRUE) 2647 panic("pmap_steal_memory: called _after_ bootstrap"); 2648 2649 *vstartp = VM_MIN_KERNEL_ADDRESS; 2650 *vendp = VM_MAX_KERNEL_ADDRESS; 2651 2652 size = round_page(vsize); 2653 npgs = atop(size); 2654 2655 /* 2656 * PA 0 will never be among those given to UVM so we can use it 2657 * to indicate we couldn't steal any memory. 2658 */ 2659 for (ps = vm_physmem, bank = 0; bank < vm_nphysseg; bank++, ps++) { 2660 if (ps->free_list == VM_FREELIST_FIRST256 && 2661 ps->avail_end - ps->avail_start >= npgs) { 2662 pa = ptoa(ps->avail_start); 2663 break; 2664 } 2665 } 2666 2667 if (pa == 0) 2668 panic("pmap_steal_memory: no approriate memory to steal!"); 2669 2670 ps->avail_start += npgs; 2671 ps->start += npgs; 2672 2673 /* 2674 * If we've used up all the pages in the segment, remove it and 2675 * compact the list. 2676 */ 2677 if (ps->avail_start == ps->end) { 2678 /* 2679 * If this was the last one, then a very bad thing has occurred 2680 */ 2681 if (--vm_nphysseg == 0) 2682 panic("pmap_steal_memory: out of memory!"); 2683 2684 printf("pmap_steal_memory: consumed bank %d\n", bank); 2685 for (; bank < vm_nphysseg; bank++, ps++) { 2686 ps[0] = ps[1]; 2687 } 2688 } 2689 2690 va = (vaddr_t) pa; 2691 memset((caddr_t) va, 0, size); 2692 pmap_pages_stolen += npgs; 2693 #ifdef DEBUG 2694 if (pmapdebug && npgs > 1) { 2695 u_int cnt = 0; 2696 for (bank = 0, ps = vm_physmem; bank < vm_nphysseg; bank++, ps++) 2697 cnt += ps->avail_end - ps->avail_start; 2698 printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n", 2699 npgs, pmap_pages_stolen, cnt); 2700 } 2701 #endif 2702 2703 return va; 2704 } 2705 2706 /* 2707 * Find a chuck of memory with right size and alignment. 2708 */ 2709 void * 2710 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end) 2711 { 2712 struct mem_region *mp; 2713 paddr_t s, e; 2714 int i, j; 2715 2716 size = round_page(size); 2717 2718 DPRINTFN(BOOT, 2719 ("pmap_boot_find_memory: size=%lx, alignment=%lx, at_end=%d", 2720 size, alignment, at_end)); 2721 2722 if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0) 2723 panic("pmap_boot_find_memory: invalid alignment %lx", 2724 alignment); 2725 2726 if (at_end) { 2727 if (alignment != PAGE_SIZE) 2728 panic("pmap_boot_find_memory: invalid ending " 2729 "alignment %lx", alignment); 2730 2731 for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) { 2732 s = mp->start + mp->size - size; 2733 if (s >= mp->start && mp->size >= size) { 2734 DPRINTFN(BOOT,(": %lx\n", s)); 2735 DPRINTFN(BOOT, 2736 ("pmap_boot_find_memory: b-avail[%d] start " 2737 "0x%lx size 0x%lx\n", mp - avail, 2738 mp->start, mp->size)); 2739 mp->size -= size; 2740 DPRINTFN(BOOT, 2741 ("pmap_boot_find_memory: a-avail[%d] start " 2742 "0x%lx size 0x%lx\n", mp - avail, 2743 mp->start, mp->size)); 2744 return (void *) s; 2745 } 2746 } 2747 panic("pmap_boot_find_memory: no available memory"); 2748 } 2749 2750 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2751 s = (mp->start + alignment - 1) & ~(alignment-1); 2752 e = s + size; 2753 2754 /* 2755 * Is the calculated region entirely within the region? 2756 */ 2757 if (s < mp->start || e > mp->start + mp->size) 2758 continue; 2759 2760 DPRINTFN(BOOT,(": %lx\n", s)); 2761 if (s == mp->start) { 2762 /* 2763 * If the block starts at the beginning of region, 2764 * adjust the size & start. (the region may now be 2765 * zero in length) 2766 */ 2767 DPRINTFN(BOOT, 2768 ("pmap_boot_find_memory: b-avail[%d] start " 2769 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2770 mp->start += size; 2771 mp->size -= size; 2772 DPRINTFN(BOOT, 2773 ("pmap_boot_find_memory: a-avail[%d] start " 2774 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2775 } else if (e == mp->start + mp->size) { 2776 /* 2777 * If the block starts at the beginning of region, 2778 * adjust only the size. 2779 */ 2780 DPRINTFN(BOOT, 2781 ("pmap_boot_find_memory: b-avail[%d] start " 2782 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2783 mp->size -= size; 2784 DPRINTFN(BOOT, 2785 ("pmap_boot_find_memory: a-avail[%d] start " 2786 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2787 } else { 2788 /* 2789 * Block is in the middle of the region, so we 2790 * have to split it in two. 2791 */ 2792 for (j = avail_cnt; j > i + 1; j--) { 2793 avail[j] = avail[j-1]; 2794 } 2795 DPRINTFN(BOOT, 2796 ("pmap_boot_find_memory: b-avail[%d] start " 2797 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2798 mp[1].start = e; 2799 mp[1].size = mp[0].start + mp[0].size - e; 2800 mp[0].size = s - mp[0].start; 2801 avail_cnt++; 2802 for (; i < avail_cnt; i++) { 2803 DPRINTFN(BOOT, 2804 ("pmap_boot_find_memory: a-avail[%d] " 2805 "start 0x%lx size 0x%lx\n", i, 2806 avail[i].start, avail[i].size)); 2807 } 2808 } 2809 return (void *) s; 2810 } 2811 panic("pmap_boot_find_memory: not enough memory for " 2812 "%lx/%lx allocation?", size, alignment); 2813 } 2814 2815 /* 2816 * This is not part of the defined PMAP interface and is specific to the 2817 * PowerPC architecture. This is called during initppc, before the system 2818 * is really initialized. 2819 */ 2820 void 2821 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend) 2822 { 2823 struct mem_region *mp, tmp; 2824 paddr_t s, e; 2825 psize_t size; 2826 int i, j; 2827 2828 /* 2829 * Get memory. 2830 */ 2831 mem_regions(&mem, &avail); 2832 #if defined(DEBUG) 2833 if (pmapdebug & PMAPDEBUG_BOOT) { 2834 printf("pmap_bootstrap: memory configuration:\n"); 2835 for (mp = mem; mp->size; mp++) { 2836 printf("pmap_bootstrap: mem start 0x%lx size 0x%lx\n", 2837 mp->start, mp->size); 2838 } 2839 for (mp = avail; mp->size; mp++) { 2840 printf("pmap_bootstrap: avail start 0x%lx size 0x%lx\n", 2841 mp->start, mp->size); 2842 } 2843 } 2844 #endif 2845 2846 /* 2847 * Find out how much physical memory we have and in how many chunks. 2848 */ 2849 for (mem_cnt = 0, mp = mem; mp->size; mp++) { 2850 if (mp->start >= pmap_memlimit) 2851 continue; 2852 if (mp->start + mp->size > pmap_memlimit) { 2853 size = pmap_memlimit - mp->start; 2854 physmem += btoc(size); 2855 } else { 2856 physmem += btoc(mp->size); 2857 } 2858 mem_cnt++; 2859 } 2860 2861 /* 2862 * Count the number of available entries. 2863 */ 2864 for (avail_cnt = 0, mp = avail; mp->size; mp++) 2865 avail_cnt++; 2866 2867 /* 2868 * Page align all regions. 2869 */ 2870 kernelstart = trunc_page(kernelstart); 2871 kernelend = round_page(kernelend); 2872 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2873 s = round_page(mp->start); 2874 mp->size -= (s - mp->start); 2875 mp->size = trunc_page(mp->size); 2876 mp->start = s; 2877 e = mp->start + mp->size; 2878 2879 DPRINTFN(BOOT, 2880 ("pmap_bootstrap: b-avail[%d] start 0x%lx size 0x%lx\n", 2881 i, mp->start, mp->size)); 2882 2883 /* 2884 * Don't allow the end to run beyond our artificial limit 2885 */ 2886 if (e > pmap_memlimit) 2887 e = pmap_memlimit; 2888 2889 /* 2890 * Is this region empty or strange? skip it. 2891 */ 2892 if (e <= s) { 2893 mp->start = 0; 2894 mp->size = 0; 2895 continue; 2896 } 2897 2898 /* 2899 * Does this overlap the beginning of kernel? 2900 * Does extend past the end of the kernel? 2901 */ 2902 else if (s < kernelstart && e > kernelstart) { 2903 if (e > kernelend) { 2904 avail[avail_cnt].start = kernelend; 2905 avail[avail_cnt].size = e - kernelend; 2906 avail_cnt++; 2907 } 2908 mp->size = kernelstart - s; 2909 } 2910 /* 2911 * Check whether this region overlaps the end of the kernel. 2912 */ 2913 else if (s < kernelend && e > kernelend) { 2914 mp->start = kernelend; 2915 mp->size = e - kernelend; 2916 } 2917 /* 2918 * Look whether this regions is completely inside the kernel. 2919 * Nuke it if it does. 2920 */ 2921 else if (s >= kernelstart && e <= kernelend) { 2922 mp->start = 0; 2923 mp->size = 0; 2924 } 2925 /* 2926 * If the user imposed a memory limit, enforce it. 2927 */ 2928 else if (s >= pmap_memlimit) { 2929 mp->start = -PAGE_SIZE; /* let's know why */ 2930 mp->size = 0; 2931 } 2932 else { 2933 mp->start = s; 2934 mp->size = e - s; 2935 } 2936 DPRINTFN(BOOT, 2937 ("pmap_bootstrap: a-avail[%d] start 0x%lx size 0x%lx\n", 2938 i, mp->start, mp->size)); 2939 } 2940 2941 /* 2942 * Move (and uncount) all the null return to the end. 2943 */ 2944 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2945 if (mp->size == 0) { 2946 tmp = avail[i]; 2947 avail[i] = avail[--avail_cnt]; 2948 avail[avail_cnt] = avail[i]; 2949 } 2950 } 2951 2952 /* 2953 * (Bubble)sort them into asecnding order. 2954 */ 2955 for (i = 0; i < avail_cnt; i++) { 2956 for (j = i + 1; j < avail_cnt; j++) { 2957 if (avail[i].start > avail[j].start) { 2958 tmp = avail[i]; 2959 avail[i] = avail[j]; 2960 avail[j] = tmp; 2961 } 2962 } 2963 } 2964 2965 /* 2966 * Make sure they don't overlap. 2967 */ 2968 for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) { 2969 if (mp[0].start + mp[0].size > mp[1].start) { 2970 mp[0].size = mp[1].start - mp[0].start; 2971 } 2972 DPRINTFN(BOOT, 2973 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 2974 i, mp->start, mp->size)); 2975 } 2976 DPRINTFN(BOOT, 2977 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 2978 i, mp->start, mp->size)); 2979 2980 #ifdef PTEGCOUNT 2981 pmap_pteg_cnt = PTEGCOUNT; 2982 #else /* PTEGCOUNT */ 2983 pmap_pteg_cnt = 0x1000; 2984 2985 while (pmap_pteg_cnt < physmem) 2986 pmap_pteg_cnt <<= 1; 2987 2988 pmap_pteg_cnt >>= 1; 2989 #endif /* PTEGCOUNT */ 2990 2991 /* 2992 * Find suitably aligned memory for PTEG hash table. 2993 */ 2994 size = pmap_pteg_cnt * sizeof(struct pteg); 2995 pmap_pteg_table = pmap_boot_find_memory(size, size, 0); 2996 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 2997 if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH) 2998 panic("pmap_bootstrap: pmap_pteg_table end (%p + %lx) > 256MB", 2999 pmap_pteg_table, size); 3000 #endif 3001 3002 memset((void *)pmap_pteg_table, 0, pmap_pteg_cnt * sizeof(struct pteg)); 3003 pmap_pteg_mask = pmap_pteg_cnt - 1; 3004 3005 /* 3006 * We cannot do pmap_steal_memory here since UVM hasn't been loaded 3007 * with pages. So we just steal them before giving them to UVM. 3008 */ 3009 size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt; 3010 pmap_pvo_table = pmap_boot_find_memory(size, PAGE_SIZE, 0); 3011 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3012 if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH) 3013 panic("pmap_bootstrap: pmap_pvo_table end (%p + %lx) > 256MB", 3014 pmap_pvo_table, size); 3015 #endif 3016 3017 for (i = 0; i < pmap_pteg_cnt; i++) 3018 TAILQ_INIT(&pmap_pvo_table[i]); 3019 3020 #ifndef MSGBUFADDR 3021 /* 3022 * Allocate msgbuf in high memory. 3023 */ 3024 msgbuf_paddr = 3025 (paddr_t) pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1); 3026 #endif 3027 3028 #ifdef __HAVE_PMAP_PHYSSEG 3029 { 3030 u_int npgs = 0; 3031 for (i = 0, mp = avail; i < avail_cnt; i++, mp++) 3032 npgs += btoc(mp->size); 3033 size = (sizeof(struct pvo_head) + 1) * npgs; 3034 pmap_physseg.pvoh = pmap_boot_find_memory(size, PAGE_SIZE, 0); 3035 pmap_physseg.attrs = (char *) &pmap_physseg.pvoh[npgs]; 3036 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 3037 if ((uintptr_t)pmap_physseg.pvoh + size > SEGMENT_LENGTH) 3038 panic("pmap_bootstrap: PVO list end (%p + %lx) > 256MB", 3039 pmap_physseg.pvoh, size); 3040 #endif 3041 } 3042 #endif 3043 3044 for (mp = avail, i = 0; i < avail_cnt; mp++, i++) { 3045 paddr_t pfstart = atop(mp->start); 3046 paddr_t pfend = atop(mp->start + mp->size); 3047 if (mp->size == 0) 3048 continue; 3049 if (mp->start + mp->size <= SEGMENT_LENGTH) { 3050 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3051 VM_FREELIST_FIRST256); 3052 } else if (mp->start >= SEGMENT_LENGTH) { 3053 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3054 VM_FREELIST_DEFAULT); 3055 } else { 3056 pfend = atop(SEGMENT_LENGTH); 3057 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3058 VM_FREELIST_FIRST256); 3059 pfstart = atop(SEGMENT_LENGTH); 3060 pfend = atop(mp->start + mp->size); 3061 uvm_page_physload(pfstart, pfend, pfstart, pfend, 3062 VM_FREELIST_DEFAULT); 3063 } 3064 } 3065 3066 /* 3067 * Make sure kernel vsid is allocated as well as VSID 0. 3068 */ 3069 pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW] 3070 |= 1 << (KERNEL_VSIDBITS % VSID_NBPW); 3071 pmap_vsid_bitmap[0] |= 1; 3072 3073 /* 3074 * Initialize kernel pmap and hardware. 3075 */ 3076 #ifndef PPC_OEA64 3077 for (i = 0; i < 16; i++) { 3078 pmap_kernel()->pm_sr[i] = EMPTY_SEGMENT; 3079 __asm __volatile ("mtsrin %0,%1" 3080 :: "r"(EMPTY_SEGMENT), "r"(i << ADDR_SR_SHFT)); 3081 } 3082 3083 pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY; 3084 __asm __volatile ("mtsr %0,%1" 3085 :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT)); 3086 #ifdef KERNEL2_SR 3087 pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY; 3088 __asm __volatile ("mtsr %0,%1" 3089 :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT)); 3090 #endif 3091 for (i = 0; i < 16; i++) { 3092 if (iosrtable[i] & SR601_T) { 3093 pmap_kernel()->pm_sr[i] = iosrtable[i]; 3094 __asm __volatile ("mtsrin %0,%1" 3095 :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT)); 3096 } 3097 } 3098 #endif /* !PPC_OEA64 */ 3099 3100 __asm __volatile ("sync; mtsdr1 %0; isync" 3101 :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10))); 3102 tlbia(); 3103 3104 #ifdef ALTIVEC 3105 pmap_use_altivec = cpu_altivec; 3106 #endif 3107 3108 #ifdef DEBUG 3109 if (pmapdebug & PMAPDEBUG_BOOT) { 3110 u_int cnt; 3111 int bank; 3112 char pbuf[9]; 3113 for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) { 3114 cnt += vm_physmem[bank].avail_end - vm_physmem[bank].avail_start; 3115 printf("pmap_bootstrap: vm_physmem[%d]=%#lx-%#lx/%#lx\n", 3116 bank, 3117 ptoa(vm_physmem[bank].avail_start), 3118 ptoa(vm_physmem[bank].avail_end), 3119 ptoa(vm_physmem[bank].avail_end - vm_physmem[bank].avail_start)); 3120 } 3121 format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt)); 3122 printf("pmap_bootstrap: UVM memory = %s (%u pages)\n", 3123 pbuf, cnt); 3124 } 3125 #endif 3126 3127 pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry), 3128 sizeof(struct pvo_entry), 0, 0, "pmap_upvopl", 3129 &pmap_pool_uallocator); 3130 3131 pool_setlowat(&pmap_upvo_pool, 252); 3132 3133 pool_init(&pmap_pool, sizeof(struct pmap), 3134 sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator); 3135 } 3136
Indexes created Fri Sep 26 20:09:58 GMT 2025