pmap.c revision 1.14
1 /* $NetBSD: pmap.c,v 1.14 2003/08/24 17:52:35 chs 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 1.14 2003/08/24 17:52:35 chs Exp $"); 71 72 #include "opt_altivec.h" 73 #include "opt_pmap.h" 74 #include <sys/param.h> 75 #include <sys/malloc.h> 76 #include <sys/proc.h> 77 #include <sys/user.h> 78 #include <sys/pool.h> 79 #include <sys/queue.h> 80 #include <sys/device.h> /* for evcnt */ 81 #include <sys/systm.h> 82 83 #if __NetBSD_Version__ < 105010000 84 #include <vm/vm.h> 85 #include <vm/vm_kern.h> 86 #define splvm() splimp() 87 #endif 88 89 #include <uvm/uvm.h> 90 91 #include <machine/pcb.h> 92 #include <machine/powerpc.h> 93 #include <powerpc/spr.h> 94 #include <powerpc/oea/sr_601.h> 95 #if __NetBSD_Version__ > 105010000 96 #include <powerpc/oea/bat.h> 97 #else 98 #include <powerpc/bat.h> 99 #endif 100 101 #if defined(DEBUG) || defined(PMAPCHECK) 102 #define STATIC 103 #else 104 #define STATIC static 105 #endif 106 107 #ifdef ALTIVEC 108 int pmap_use_altivec; 109 #endif 110 111 volatile struct pteg *pmap_pteg_table; 112 unsigned int pmap_pteg_cnt; 113 unsigned int pmap_pteg_mask; 114 paddr_t pmap_memlimit = -PAGE_SIZE; /* there is no limit */ 115 116 struct pmap kernel_pmap_; 117 unsigned int pmap_pages_stolen; 118 u_long pmap_pte_valid; 119 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 120 u_long pmap_pvo_enter_depth; 121 u_long pmap_pvo_remove_depth; 122 #endif 123 124 int physmem; 125 #ifndef MSGBUFADDR 126 extern paddr_t msgbuf_paddr; 127 #endif 128 129 static struct mem_region *mem, *avail; 130 static u_int mem_cnt, avail_cnt; 131 132 #ifdef __HAVE_PMAP_PHYSSEG 133 /* 134 * This is a cache of referenced/modified bits. 135 * Bits herein are shifted by ATTRSHFT. 136 */ 137 #define ATTR_SHFT 4 138 struct pmap_physseg pmap_physseg; 139 #endif 140 141 /* 142 * The following structure is exactly 32 bytes long (one cacheline). 143 */ 144 struct pvo_entry { 145 LIST_ENTRY(pvo_entry) pvo_vlink; /* Link to common virt page */ 146 TAILQ_ENTRY(pvo_entry) pvo_olink; /* Link to overflow entry */ 147 struct pte pvo_pte; /* Prebuilt PTE */ 148 pmap_t pvo_pmap; /* ptr to owning pmap */ 149 vaddr_t pvo_vaddr; /* VA of entry */ 150 #define PVO_PTEGIDX_MASK 0x0007 /* which PTEG slot */ 151 #define PVO_PTEGIDX_VALID 0x0008 /* slot is valid */ 152 #define PVO_WIRED 0x0010 /* PVO entry is wired */ 153 #define PVO_MANAGED 0x0020 /* PVO e. for managed page */ 154 #define PVO_EXECUTABLE 0x0040 /* PVO e. for executable page */ 155 #define PVO_ENTER_INSERT 0 /* PVO has been removed */ 156 #define PVO_SPILL_UNSET 1 /* PVO has been evicted */ 157 #define PVO_SPILL_SET 2 /* PVO has been spilled */ 158 #define PVO_SPILL_INSERT 3 /* PVO has been inserted */ 159 #define PVO_PMAP_PAGE_PROTECT 4 /* PVO has changed */ 160 #define PVO_PMAP_PROTECT 5 /* PVO has changed */ 161 #define PVO_REMOVE 6 /* PVO has been removed */ 162 #define PVO_WHERE_MASK 15 163 #define PVO_WHERE_SHFT 8 164 }; 165 #define PVO_VADDR(pvo) ((pvo)->pvo_vaddr & ~ADDR_POFF) 166 #define PVO_ISEXECUTABLE(pvo) ((pvo)->pvo_vaddr & PVO_EXECUTABLE) 167 #define PVO_PTEGIDX_GET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK) 168 #define PVO_PTEGIDX_ISSET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID) 169 #define PVO_PTEGIDX_CLR(pvo) \ 170 ((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK))) 171 #define PVO_PTEGIDX_SET(pvo,i) \ 172 ((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID)) 173 #define PVO_WHERE(pvo,w) \ 174 ((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \ 175 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT)) 176 177 TAILQ_HEAD(pvo_tqhead, pvo_entry); 178 struct pvo_tqhead *pmap_pvo_table; /* pvo entries by ptegroup index */ 179 struct pvo_head pmap_pvo_kunmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_kunmanaged); /* list of unmanaged pages */ 180 struct pvo_head pmap_pvo_unmanaged = LIST_HEAD_INITIALIZER(pmap_pvo_unmanaged); /* list of unmanaged pages */ 181 182 struct pool pmap_pool; /* pool for pmap structures */ 183 struct pool pmap_upvo_pool; /* pool for pvo entries for unmanaged pages */ 184 struct pool pmap_mpvo_pool; /* pool for pvo entries for managed pages */ 185 186 /* 187 * We keep a cache of unmanaged pages to be used for pvo entries for 188 * unmanaged pages. 189 */ 190 struct pvo_page { 191 SIMPLEQ_ENTRY(pvo_page) pvop_link; 192 }; 193 SIMPLEQ_HEAD(pvop_head, pvo_page); 194 struct pvop_head pmap_upvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_upvop_head); 195 struct pvop_head pmap_mpvop_head = SIMPLEQ_HEAD_INITIALIZER(pmap_mpvop_head); 196 u_long pmap_upvop_free; 197 u_long pmap_upvop_maxfree; 198 u_long pmap_mpvop_free; 199 u_long pmap_mpvop_maxfree; 200 201 STATIC void *pmap_pool_ualloc(struct pool *, int); 202 STATIC void *pmap_pool_malloc(struct pool *, int); 203 204 STATIC void pmap_pool_ufree(struct pool *, void *); 205 STATIC void pmap_pool_mfree(struct pool *, void *); 206 207 static struct pool_allocator pmap_pool_mallocator = { 208 pmap_pool_malloc, pmap_pool_mfree, 0, 209 }; 210 211 static struct pool_allocator pmap_pool_uallocator = { 212 pmap_pool_ualloc, pmap_pool_ufree, 0, 213 }; 214 215 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 216 void pmap_pte_print(volatile struct pte *); 217 #endif 218 219 #ifdef DDB 220 void pmap_pteg_check(void); 221 void pmap_pteg_dist(void); 222 void pmap_print_pte(pmap_t, vaddr_t); 223 void pmap_print_mmuregs(void); 224 #endif 225 226 #if defined(DEBUG) || defined(PMAPCHECK) 227 #ifdef PMAPCHECK 228 int pmapcheck = 1; 229 #else 230 int pmapcheck = 0; 231 #endif 232 void pmap_pvo_verify(void); 233 STATIC void pmap_pvo_check(const struct pvo_entry *); 234 #define PMAP_PVO_CHECK(pvo) \ 235 do { \ 236 if (pmapcheck) \ 237 pmap_pvo_check(pvo); \ 238 } while (0) 239 #else 240 #define PMAP_PVO_CHECK(pvo) do { } while (/*CONSTCOND*/0) 241 #endif 242 STATIC int pmap_pte_insert(int, struct pte *); 243 STATIC int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *, 244 vaddr_t, paddr_t, register_t, int); 245 STATIC void pmap_pvo_remove(struct pvo_entry *, int); 246 STATIC struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *); 247 STATIC volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int); 248 #define pmap_pvo_reclaim(pm) NULL 249 STATIC void pvo_set_exec(struct pvo_entry *); 250 STATIC void pvo_clear_exec(struct pvo_entry *); 251 252 STATIC void tlbia(void); 253 254 STATIC void pmap_release(pmap_t); 255 STATIC void *pmap_boot_find_memory(psize_t, psize_t, int); 256 257 #define VSID_NBPW (sizeof(uint32_t) * 8) 258 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW]; 259 260 static int pmap_initialized; 261 262 #if defined(DEBUG) || defined(PMAPDEBUG) 263 #define PMAPDEBUG_BOOT 0x0001 264 #define PMAPDEBUG_PTE 0x0002 265 #define PMAPDEBUG_EXEC 0x0008 266 #define PMAPDEBUG_PVOENTER 0x0010 267 #define PMAPDEBUG_PVOREMOVE 0x0020 268 #define PMAPDEBUG_ACTIVATE 0x0100 269 #define PMAPDEBUG_CREATE 0x0200 270 #define PMAPDEBUG_ENTER 0x1000 271 #define PMAPDEBUG_KENTER 0x2000 272 #define PMAPDEBUG_KREMOVE 0x4000 273 #define PMAPDEBUG_REMOVE 0x8000 274 unsigned int pmapdebug = 0; 275 # define DPRINTF(x) printf x 276 # define DPRINTFN(n, x) if (pmapdebug & PMAPDEBUG_ ## n) printf x 277 #else 278 # define DPRINTF(x) 279 # define DPRINTFN(n, x) 280 #endif 281 282 283 #ifdef PMAPCOUNTERS 284 #define PMAPCOUNT(ev) ((pmap_evcnt_ ## ev).ev_count++) 285 #define PMAPCOUNT2(ev) ((ev).ev_count++) 286 287 struct evcnt pmap_evcnt_mappings = 288 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 289 "pmap", "pages mapped"); 290 struct evcnt pmap_evcnt_unmappings = 291 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 292 "pmap", "pages unmapped"); 293 294 struct evcnt pmap_evcnt_kernel_mappings = 295 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 296 "pmap", "kernel pages mapped"); 297 struct evcnt pmap_evcnt_kernel_unmappings = 298 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_kernel_mappings, 299 "pmap", "kernel pages unmapped"); 300 301 struct evcnt pmap_evcnt_mappings_replaced = 302 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 303 "pmap", "page mappings replaced"); 304 305 struct evcnt pmap_evcnt_exec_mappings = 306 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 307 "pmap", "exec pages mapped"); 308 struct evcnt pmap_evcnt_exec_cached = 309 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_mappings, 310 "pmap", "exec pages cached"); 311 312 struct evcnt pmap_evcnt_exec_synced = 313 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 314 "pmap", "exec pages synced"); 315 struct evcnt pmap_evcnt_exec_synced_clear_modify = 316 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 317 "pmap", "exec pages synced (CM)"); 318 319 struct evcnt pmap_evcnt_exec_uncached_page_protect = 320 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 321 "pmap", "exec pages uncached (PP)"); 322 struct evcnt pmap_evcnt_exec_uncached_clear_modify = 323 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 324 "pmap", "exec pages uncached (CM)"); 325 struct evcnt pmap_evcnt_exec_uncached_zero_page = 326 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 327 "pmap", "exec pages uncached (ZP)"); 328 struct evcnt pmap_evcnt_exec_uncached_copy_page = 329 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &pmap_evcnt_exec_mappings, 330 "pmap", "exec pages uncached (CP)"); 331 332 struct evcnt pmap_evcnt_updates = 333 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 334 "pmap", "updates"); 335 struct evcnt pmap_evcnt_collects = 336 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 337 "pmap", "collects"); 338 struct evcnt pmap_evcnt_copies = 339 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 340 "pmap", "copies"); 341 342 struct evcnt pmap_evcnt_ptes_spilled = 343 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 344 "pmap", "ptes spilled from overflow"); 345 struct evcnt pmap_evcnt_ptes_unspilled = 346 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 347 "pmap", "ptes not spilled"); 348 struct evcnt pmap_evcnt_ptes_evicted = 349 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 350 "pmap", "ptes evicted"); 351 352 struct evcnt pmap_evcnt_ptes_primary[8] = { 353 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 354 "pmap", "ptes added at primary[0]"), 355 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 356 "pmap", "ptes added at primary[1]"), 357 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 358 "pmap", "ptes added at primary[2]"), 359 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 360 "pmap", "ptes added at primary[3]"), 361 362 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 363 "pmap", "ptes added at primary[4]"), 364 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 365 "pmap", "ptes added at primary[5]"), 366 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 367 "pmap", "ptes added at primary[6]"), 368 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 369 "pmap", "ptes added at primary[7]"), 370 }; 371 struct evcnt pmap_evcnt_ptes_secondary[8] = { 372 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 373 "pmap", "ptes added at secondary[0]"), 374 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 375 "pmap", "ptes added at secondary[1]"), 376 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 377 "pmap", "ptes added at secondary[2]"), 378 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 379 "pmap", "ptes added at secondary[3]"), 380 381 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 382 "pmap", "ptes added at secondary[4]"), 383 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 384 "pmap", "ptes added at secondary[5]"), 385 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 386 "pmap", "ptes added at secondary[6]"), 387 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 388 "pmap", "ptes added at secondary[7]"), 389 }; 390 struct evcnt pmap_evcnt_ptes_removed = 391 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 392 "pmap", "ptes removed"); 393 struct evcnt pmap_evcnt_ptes_changed = 394 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, 395 "pmap", "ptes changed"); 396 397 /* 398 * From pmap_subr.c 399 */ 400 extern struct evcnt pmap_evcnt_zeroed_pages; 401 extern struct evcnt pmap_evcnt_copied_pages; 402 extern struct evcnt pmap_evcnt_idlezeroed_pages; 403 #else 404 #define PMAPCOUNT(ev) ((void) 0) 405 #define PMAPCOUNT2(ev) ((void) 0) 406 #endif 407 408 #define TLBIE(va) __asm __volatile("tlbie %0" :: "r"(va)) 409 #define TLBSYNC() __asm __volatile("tlbsync") 410 #define SYNC() __asm __volatile("sync") 411 #define EIEIO() __asm __volatile("eieio") 412 #define MFMSR() mfmsr() 413 #define MTMSR(psl) mtmsr(psl) 414 #define MFPVR() mfpvr() 415 #define MFSRIN(va) mfsrin(va) 416 #define MFTB() mfrtcltbl() 417 418 static __inline register_t 419 mfsrin(vaddr_t va) 420 { 421 register_t sr; 422 __asm __volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va)); 423 return sr; 424 } 425 426 static __inline register_t 427 pmap_interrupts_off(void) 428 { 429 register_t msr = MFMSR(); 430 if (msr & PSL_EE) 431 MTMSR(msr & ~PSL_EE); 432 return msr; 433 } 434 435 static void 436 pmap_interrupts_restore(register_t msr) 437 { 438 if (msr & PSL_EE) 439 MTMSR(msr); 440 } 441 442 static __inline u_int32_t 443 mfrtcltbl(void) 444 { 445 446 if ((MFPVR() >> 16) == MPC601) 447 return (mfrtcl() >> 7); 448 else 449 return (mftbl()); 450 } 451 452 /* 453 * These small routines may have to be replaced, 454 * if/when we support processors other that the 604. 455 */ 456 457 void 458 tlbia(void) 459 { 460 caddr_t i; 461 462 SYNC(); 463 /* 464 * Why not use "tlbia"? Because not all processors implement it. 465 * 466 * This needs to be a per-cpu callback to do the appropriate thing 467 * for the CPU. XXX 468 */ 469 for (i = 0; i < (caddr_t)0x00040000; i += 0x00001000) { 470 TLBIE(i); 471 EIEIO(); 472 SYNC(); 473 } 474 TLBSYNC(); 475 SYNC(); 476 } 477 478 static __inline register_t 479 va_to_vsid(const struct pmap *pm, vaddr_t addr) 480 { 481 return (pm->pm_sr[addr >> ADDR_SR_SHFT] & SR_VSID); 482 } 483 484 static __inline register_t 485 va_to_pteg(const struct pmap *pm, vaddr_t addr) 486 { 487 register_t hash; 488 489 hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT); 490 return hash & pmap_pteg_mask; 491 } 492 493 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 494 /* 495 * Given a PTE in the page table, calculate the VADDR that hashes to it. 496 * The only bit of magic is that the top 4 bits of the address doesn't 497 * technically exist in the PTE. But we know we reserved 4 bits of the 498 * VSID for it so that's how we get it. 499 */ 500 static vaddr_t 501 pmap_pte_to_va(volatile const struct pte *pt) 502 { 503 vaddr_t va; 504 uintptr_t ptaddr = (uintptr_t) pt; 505 506 if (pt->pte_hi & PTE_HID) 507 ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg)); 508 509 /* PPC Bits 10-19 */ 510 va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff; 511 va <<= ADDR_PIDX_SHFT; 512 513 /* PPC Bits 4-9 */ 514 va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT; 515 516 /* PPC Bits 0-3 */ 517 va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; 518 519 return va; 520 } 521 #endif 522 523 static __inline struct pvo_head * 524 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p) 525 { 526 #ifdef __HAVE_VM_PAGE_MD 527 struct vm_page *pg; 528 529 pg = PHYS_TO_VM_PAGE(pa); 530 if (pg_p != NULL) 531 *pg_p = pg; 532 if (pg == NULL) 533 return &pmap_pvo_unmanaged; 534 return &pg->mdpage.mdpg_pvoh; 535 #endif 536 #ifdef __HAVE_PMAP_PHYSSEG 537 int bank, pg; 538 539 bank = vm_physseg_find(atop(pa), &pg); 540 if (pg_p != NULL) 541 *pg_p = pg; 542 if (bank == -1) 543 return &pmap_pvo_unmanaged; 544 return &vm_physmem[bank].pmseg.pvoh[pg]; 545 #endif 546 } 547 548 static __inline struct pvo_head * 549 vm_page_to_pvoh(struct vm_page *pg) 550 { 551 #ifdef __HAVE_VM_PAGE_MD 552 return &pg->mdpage.mdpg_pvoh; 553 #endif 554 #ifdef __HAVE_PMAP_PHYSSEG 555 return pa_to_pvoh(VM_PAGE_TO_PHYS(pg), NULL); 556 #endif 557 } 558 559 560 #ifdef __HAVE_PMAP_PHYSSEG 561 static __inline char * 562 pa_to_attr(paddr_t pa) 563 { 564 int bank, pg; 565 566 bank = vm_physseg_find(atop(pa), &pg); 567 if (bank == -1) 568 return NULL; 569 return &vm_physmem[bank].pmseg.attrs[pg]; 570 } 571 #endif 572 573 static __inline void 574 pmap_attr_clear(struct vm_page *pg, int ptebit) 575 { 576 #ifdef __HAVE_PMAP_PHYSSEG 577 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) &= ~(ptebit >> ATTR_SHFT); 578 #endif 579 #ifdef __HAVE_VM_PAGE_MD 580 pg->mdpage.mdpg_attrs &= ~ptebit; 581 #endif 582 } 583 584 static __inline int 585 pmap_attr_fetch(struct vm_page *pg) 586 { 587 #ifdef __HAVE_PMAP_PHYSSEG 588 return *pa_to_attr(VM_PAGE_TO_PHYS(pg)) << ATTR_SHFT; 589 #endif 590 #ifdef __HAVE_VM_PAGE_MD 591 return pg->mdpage.mdpg_attrs; 592 #endif 593 } 594 595 static __inline void 596 pmap_attr_save(struct vm_page *pg, int ptebit) 597 { 598 #ifdef __HAVE_PMAP_PHYSSEG 599 *pa_to_attr(VM_PAGE_TO_PHYS(pg)) |= (ptebit >> ATTR_SHFT); 600 #endif 601 #ifdef __HAVE_VM_PAGE_MD 602 pg->mdpage.mdpg_attrs |= ptebit; 603 #endif 604 } 605 606 static __inline int 607 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt) 608 { 609 if (pt->pte_hi == pvo_pt->pte_hi 610 #if 0 611 && ((pt->pte_lo ^ pvo_pt->pte_lo) & 612 ~(PTE_REF|PTE_CHG)) == 0 613 #endif 614 ) 615 return 1; 616 return 0; 617 } 618 619 static __inline void 620 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo) 621 { 622 /* 623 * Construct the PTE. Default to IMB initially. Valid bit 624 * only gets set when the real pte is set in memory. 625 * 626 * Note: Don't set the valid bit for correct operation of tlb update. 627 */ 628 pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT) 629 | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API); 630 pt->pte_lo = pte_lo; 631 } 632 633 static __inline void 634 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt) 635 { 636 pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG); 637 } 638 639 static __inline void 640 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit) 641 { 642 /* 643 * As shown in Section 7.6.3.2.3 644 */ 645 pt->pte_lo &= ~ptebit; 646 TLBIE(va); 647 SYNC(); 648 EIEIO(); 649 TLBSYNC(); 650 SYNC(); 651 } 652 653 static __inline void 654 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt) 655 { 656 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 657 if (pvo_pt->pte_hi & PTE_VALID) 658 panic("pte_set: setting an already valid pte %p", pvo_pt); 659 #endif 660 pvo_pt->pte_hi |= PTE_VALID; 661 /* 662 * Update the PTE as defined in section 7.6.3.1 663 * Note that the REF/CHG bits are from pvo_pt and thus should 664 * have been saved so this routine can restore them (if desired). 665 */ 666 pt->pte_lo = pvo_pt->pte_lo; 667 EIEIO(); 668 pt->pte_hi = pvo_pt->pte_hi; 669 SYNC(); 670 pmap_pte_valid++; 671 } 672 673 static __inline void 674 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 675 { 676 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 677 if ((pvo_pt->pte_hi & PTE_VALID) == 0) 678 panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt); 679 if ((pt->pte_hi & PTE_VALID) == 0) 680 panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt); 681 #endif 682 683 pvo_pt->pte_hi &= ~PTE_VALID; 684 /* 685 * Force the ref & chg bits back into the PTEs. 686 */ 687 SYNC(); 688 /* 689 * Invalidate the pte ... (Section 7.6.3.3) 690 */ 691 pt->pte_hi &= ~PTE_VALID; 692 SYNC(); 693 TLBIE(va); 694 SYNC(); 695 EIEIO(); 696 TLBSYNC(); 697 SYNC(); 698 /* 699 * Save the ref & chg bits ... 700 */ 701 pmap_pte_synch(pt, pvo_pt); 702 pmap_pte_valid--; 703 } 704 705 static __inline void 706 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va) 707 { 708 /* 709 * Invalidate the PTE 710 */ 711 pmap_pte_unset(pt, pvo_pt, va); 712 pmap_pte_set(pt, pvo_pt); 713 } 714 715 /* 716 * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx 717 * (either primary or secondary location). 718 * 719 * Note: both the destination and source PTEs must not have PTE_VALID set. 720 */ 721 722 STATIC int 723 pmap_pte_insert(int ptegidx, struct pte *pvo_pt) 724 { 725 volatile struct pte *pt; 726 int i; 727 728 #if defined(DEBUG) 729 DPRINTFN(PTE, ("pmap_pte_insert: idx 0x%x, pte 0x%lx 0x%lx\n", 730 ptegidx, pvo_pt->pte_hi, pvo_pt->pte_lo)); 731 #endif 732 /* 733 * First try primary hash. 734 */ 735 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 736 if ((pt->pte_hi & PTE_VALID) == 0) { 737 pvo_pt->pte_hi &= ~PTE_HID; 738 pmap_pte_set(pt, pvo_pt); 739 return i; 740 } 741 } 742 743 /* 744 * Now try secondary hash. 745 */ 746 ptegidx ^= pmap_pteg_mask; 747 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 748 if ((pt->pte_hi & PTE_VALID) == 0) { 749 pvo_pt->pte_hi |= PTE_HID; 750 pmap_pte_set(pt, pvo_pt); 751 return i; 752 } 753 } 754 return -1; 755 } 756 757 /* 758 * Spill handler. 759 * 760 * Tries to spill a page table entry from the overflow area. 761 * This runs in either real mode (if dealing with a exception spill) 762 * or virtual mode when dealing with manually spilling one of the 763 * kernel's pte entries. In either case, interrupts are already 764 * disabled. 765 */ 766 767 int 768 pmap_pte_spill(struct pmap *pm, vaddr_t addr, boolean_t exec) 769 { 770 struct pvo_entry *source_pvo, *victim_pvo, *next_pvo; 771 struct pvo_entry *pvo; 772 struct pvo_tqhead *pvoh, *vpvoh; 773 int ptegidx, i, j; 774 volatile struct pteg *pteg; 775 volatile struct pte *pt; 776 777 ptegidx = va_to_pteg(pm, addr); 778 779 /* 780 * Have to substitute some entry. Use the primary hash for this. 781 * Use low bits of timebase as random generator. Make sure we are 782 * not picking a kernel pte for replacement. 783 */ 784 pteg = &pmap_pteg_table[ptegidx]; 785 i = MFTB() & 7; 786 for (j = 0; j < 8; j++) { 787 pt = &pteg->pt[i]; 788 if ((pt->pte_hi & PTE_VALID) == 0 || 789 VSID_TO_HASH((pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT) 790 != KERNEL_VSIDBITS) 791 break; 792 i = (i + 1) & 7; 793 } 794 KASSERT(j < 8); 795 796 source_pvo = NULL; 797 victim_pvo = NULL; 798 pvoh = &pmap_pvo_table[ptegidx]; 799 TAILQ_FOREACH(pvo, pvoh, pvo_olink) { 800 801 /* 802 * We need to find pvo entry for this address... 803 */ 804 PMAP_PVO_CHECK(pvo); /* sanity check */ 805 806 /* 807 * If we haven't found the source and we come to a PVO with 808 * a valid PTE, then we know we can't find it because all 809 * evicted PVOs always are first in the list. 810 */ 811 if (source_pvo == NULL && (pvo->pvo_pte.pte_hi & PTE_VALID)) 812 break; 813 if (source_pvo == NULL && pm == pvo->pvo_pmap && 814 addr == PVO_VADDR(pvo)) { 815 816 /* 817 * Now we have found the entry to be spilled into the 818 * pteg. Attempt to insert it into the page table. 819 */ 820 j = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 821 if (j >= 0) { 822 PVO_PTEGIDX_SET(pvo, j); 823 PMAP_PVO_CHECK(pvo); /* sanity check */ 824 PVO_WHERE(pvo, SPILL_INSERT); 825 pvo->pvo_pmap->pm_evictions--; 826 PMAPCOUNT(ptes_spilled); 827 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 828 ? pmap_evcnt_ptes_secondary 829 : pmap_evcnt_ptes_primary)[j]); 830 831 /* 832 * Since we keep the evicted entries at the 833 * from of the PVO list, we need move this 834 * (now resident) PVO after the evicted 835 * entries. 836 */ 837 next_pvo = TAILQ_NEXT(pvo, pvo_olink); 838 839 /* 840 * If we don't have to move (either we were the 841 * last entry or the next entry was valid), 842 * don't change our position. Otherwise 843 * move ourselves to the tail of the queue. 844 */ 845 if (next_pvo != NULL && 846 !(next_pvo->pvo_pte.pte_hi & PTE_VALID)) { 847 TAILQ_REMOVE(pvoh, pvo, pvo_olink); 848 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 849 } 850 return 1; 851 } 852 source_pvo = pvo; 853 if (exec && !PVO_ISEXECUTABLE(source_pvo)) { 854 return 0; 855 } 856 if (victim_pvo != NULL) 857 break; 858 } 859 860 /* 861 * We also need the pvo entry of the victim we are replacing 862 * so save the R & C bits of the PTE. 863 */ 864 if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL && 865 pmap_pte_compare(pt, &pvo->pvo_pte)) { 866 vpvoh = pvoh; 867 victim_pvo = pvo; 868 if (source_pvo != NULL) 869 break; 870 } 871 } 872 873 if (source_pvo == NULL) { 874 PMAPCOUNT(ptes_unspilled); 875 return 0; 876 } 877 878 if (victim_pvo == NULL) { 879 if ((pt->pte_hi & PTE_HID) == 0) 880 panic("pmap_pte_spill: victim p-pte (%p) has " 881 "no pvo entry!", pt); 882 883 /* 884 * If this is a secondary PTE, we need to search 885 * its primary pvo bucket for the matching PVO. 886 */ 887 vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask]; 888 TAILQ_FOREACH(pvo, vpvoh, pvo_olink) { 889 PMAP_PVO_CHECK(pvo); /* sanity check */ 890 891 /* 892 * We also need the pvo entry of the victim we are 893 * replacing so save the R & C bits of the PTE. 894 */ 895 if (pmap_pte_compare(pt, &pvo->pvo_pte)) { 896 victim_pvo = pvo; 897 break; 898 } 899 } 900 if (victim_pvo == NULL) 901 panic("pmap_pte_spill: victim s-pte (%p) has " 902 "no pvo entry!", pt); 903 } 904 905 /* 906 * The victim should be not be a kernel PVO/PTE entry. 907 */ 908 KASSERT(victim_pvo->pvo_pmap != pmap_kernel()); 909 KASSERT(PVO_PTEGIDX_ISSET(victim_pvo)); 910 KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i); 911 912 /* 913 * We are invalidating the TLB entry for the EA for the 914 * we are replacing even though its valid; If we don't 915 * we lose any ref/chg bit changes contained in the TLB 916 * entry. 917 */ 918 source_pvo->pvo_pte.pte_hi &= ~PTE_HID; 919 920 /* 921 * To enforce the PVO list ordering constraint that all 922 * evicted entries should come before all valid entries, 923 * move the source PVO to the tail of its list and the 924 * victim PVO to the head of its list (which might not be 925 * the same list, if the victim was using the secondary hash). 926 */ 927 TAILQ_REMOVE(pvoh, source_pvo, pvo_olink); 928 TAILQ_INSERT_TAIL(pvoh, source_pvo, pvo_olink); 929 TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink); 930 TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink); 931 pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr); 932 pmap_pte_set(pt, &source_pvo->pvo_pte); 933 victim_pvo->pvo_pmap->pm_evictions++; 934 source_pvo->pvo_pmap->pm_evictions--; 935 PVO_WHERE(victim_pvo, SPILL_UNSET); 936 PVO_WHERE(source_pvo, SPILL_SET); 937 938 PVO_PTEGIDX_CLR(victim_pvo); 939 PVO_PTEGIDX_SET(source_pvo, i); 940 PMAPCOUNT2(pmap_evcnt_ptes_primary[i]); 941 PMAPCOUNT(ptes_spilled); 942 PMAPCOUNT(ptes_evicted); 943 PMAPCOUNT(ptes_removed); 944 945 PMAP_PVO_CHECK(victim_pvo); 946 PMAP_PVO_CHECK(source_pvo); 947 return 1; 948 } 949 950 /* 951 * Restrict given range to physical memory 952 */ 953 void 954 pmap_real_memory(paddr_t *start, psize_t *size) 955 { 956 struct mem_region *mp; 957 958 for (mp = mem; mp->size; mp++) { 959 if (*start + *size > mp->start 960 && *start < mp->start + mp->size) { 961 if (*start < mp->start) { 962 *size -= mp->start - *start; 963 *start = mp->start; 964 } 965 if (*start + *size > mp->start + mp->size) 966 *size = mp->start + mp->size - *start; 967 return; 968 } 969 } 970 *size = 0; 971 } 972 973 /* 974 * Initialize anything else for pmap handling. 975 * Called during vm_init(). 976 */ 977 void 978 pmap_init(void) 979 { 980 #ifdef __HAVE_PMAP_PHYSSEG 981 struct pvo_tqhead *pvoh; 982 int bank; 983 long sz; 984 char *attr; 985 986 pvoh = pmap_physseg.pvoh; 987 attr = pmap_physseg.attrs; 988 for (bank = 0; bank < vm_nphysseg; bank++) { 989 sz = vm_physmem[bank].end - vm_physmem[bank].start; 990 vm_physmem[bank].pmseg.pvoh = pvoh; 991 vm_physmem[bank].pmseg.attrs = attr; 992 for (; sz > 0; sz--, pvoh++, attr++) { 993 TAILQ_INIT(pvoh); 994 *attr = 0; 995 } 996 } 997 #endif 998 999 pool_init(&pmap_mpvo_pool, sizeof(struct pvo_entry), 1000 sizeof(struct pvo_entry), 0, 0, "pmap_mpvopl", 1001 &pmap_pool_mallocator); 1002 1003 pool_setlowat(&pmap_mpvo_pool, 1008); 1004 1005 pmap_initialized = 1; 1006 1007 #ifdef PMAPCOUNTERS 1008 evcnt_attach_static(&pmap_evcnt_mappings); 1009 evcnt_attach_static(&pmap_evcnt_mappings_replaced); 1010 evcnt_attach_static(&pmap_evcnt_unmappings); 1011 1012 evcnt_attach_static(&pmap_evcnt_kernel_mappings); 1013 evcnt_attach_static(&pmap_evcnt_kernel_unmappings); 1014 1015 evcnt_attach_static(&pmap_evcnt_exec_mappings); 1016 evcnt_attach_static(&pmap_evcnt_exec_cached); 1017 evcnt_attach_static(&pmap_evcnt_exec_synced); 1018 evcnt_attach_static(&pmap_evcnt_exec_synced_clear_modify); 1019 1020 evcnt_attach_static(&pmap_evcnt_exec_uncached_page_protect); 1021 evcnt_attach_static(&pmap_evcnt_exec_uncached_clear_modify); 1022 evcnt_attach_static(&pmap_evcnt_exec_uncached_zero_page); 1023 evcnt_attach_static(&pmap_evcnt_exec_uncached_copy_page); 1024 1025 evcnt_attach_static(&pmap_evcnt_zeroed_pages); 1026 evcnt_attach_static(&pmap_evcnt_copied_pages); 1027 evcnt_attach_static(&pmap_evcnt_idlezeroed_pages); 1028 1029 evcnt_attach_static(&pmap_evcnt_updates); 1030 evcnt_attach_static(&pmap_evcnt_collects); 1031 evcnt_attach_static(&pmap_evcnt_copies); 1032 1033 evcnt_attach_static(&pmap_evcnt_ptes_spilled); 1034 evcnt_attach_static(&pmap_evcnt_ptes_unspilled); 1035 evcnt_attach_static(&pmap_evcnt_ptes_evicted); 1036 evcnt_attach_static(&pmap_evcnt_ptes_removed); 1037 evcnt_attach_static(&pmap_evcnt_ptes_changed); 1038 evcnt_attach_static(&pmap_evcnt_ptes_primary[0]); 1039 evcnt_attach_static(&pmap_evcnt_ptes_primary[1]); 1040 evcnt_attach_static(&pmap_evcnt_ptes_primary[2]); 1041 evcnt_attach_static(&pmap_evcnt_ptes_primary[3]); 1042 evcnt_attach_static(&pmap_evcnt_ptes_primary[4]); 1043 evcnt_attach_static(&pmap_evcnt_ptes_primary[5]); 1044 evcnt_attach_static(&pmap_evcnt_ptes_primary[6]); 1045 evcnt_attach_static(&pmap_evcnt_ptes_primary[7]); 1046 evcnt_attach_static(&pmap_evcnt_ptes_secondary[0]); 1047 evcnt_attach_static(&pmap_evcnt_ptes_secondary[1]); 1048 evcnt_attach_static(&pmap_evcnt_ptes_secondary[2]); 1049 evcnt_attach_static(&pmap_evcnt_ptes_secondary[3]); 1050 evcnt_attach_static(&pmap_evcnt_ptes_secondary[4]); 1051 evcnt_attach_static(&pmap_evcnt_ptes_secondary[5]); 1052 evcnt_attach_static(&pmap_evcnt_ptes_secondary[6]); 1053 evcnt_attach_static(&pmap_evcnt_ptes_secondary[7]); 1054 #endif 1055 } 1056 1057 /* 1058 * How much virtual space does the kernel get? 1059 */ 1060 void 1061 pmap_virtual_space(vaddr_t *start, vaddr_t *end) 1062 { 1063 /* 1064 * For now, reserve one segment (minus some overhead) for kernel 1065 * virtual memory 1066 */ 1067 *start = VM_MIN_KERNEL_ADDRESS; 1068 *end = VM_MAX_KERNEL_ADDRESS; 1069 } 1070 1071 /* 1072 * Allocate, initialize, and return a new physical map. 1073 */ 1074 pmap_t 1075 pmap_create(void) 1076 { 1077 pmap_t pm; 1078 1079 pm = pool_get(&pmap_pool, PR_WAITOK); 1080 memset((caddr_t)pm, 0, sizeof *pm); 1081 pmap_pinit(pm); 1082 1083 DPRINTFN(CREATE,("pmap_create: pm %p:\n" 1084 "\t%06lx %06lx %06lx %06lx %06lx %06lx %06lx %06lx\n" 1085 "\t%06lx %06lx %06lx %06lx %06lx %06lx %06lx %06lx\n", pm, 1086 pm->pm_sr[0], pm->pm_sr[1], pm->pm_sr[2], pm->pm_sr[3], 1087 pm->pm_sr[4], pm->pm_sr[5], pm->pm_sr[6], pm->pm_sr[7], 1088 pm->pm_sr[8], pm->pm_sr[9], pm->pm_sr[10], pm->pm_sr[11], 1089 pm->pm_sr[12], pm->pm_sr[13], pm->pm_sr[14], pm->pm_sr[15])); 1090 return pm; 1091 } 1092 1093 /* 1094 * Initialize a preallocated and zeroed pmap structure. 1095 */ 1096 void 1097 pmap_pinit(pmap_t pm) 1098 { 1099 register_t entropy = MFTB(); 1100 register_t mask; 1101 int i; 1102 1103 /* 1104 * Allocate some segment registers for this pmap. 1105 */ 1106 pm->pm_refs = 1; 1107 for (i = 0; i < NPMAPS; i += VSID_NBPW) { 1108 static register_t pmap_vsidcontext; 1109 register_t hash; 1110 unsigned int n; 1111 1112 /* Create a new value by multiplying by a prime adding in 1113 * entropy from the timebase register. This is to make the 1114 * VSID more random so that the PT Hash function collides 1115 * less often. (note that the prime causes gcc to do shifts 1116 * instead of a multiply) 1117 */ 1118 pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy; 1119 hash = pmap_vsidcontext & (NPMAPS - 1); 1120 if (hash == 0) /* 0 is special, avoid it */ 1121 continue; 1122 n = hash >> 5; 1123 mask = 1L << (hash & (VSID_NBPW-1)); 1124 hash = pmap_vsidcontext; 1125 if (pmap_vsid_bitmap[n] & mask) { /* collision? */ 1126 /* anything free in this bucket? */ 1127 if (~pmap_vsid_bitmap[n] == 0) { 1128 entropy = hash >> PTE_VSID_SHFT; 1129 continue; 1130 } 1131 i = ffs(~pmap_vsid_bitmap[n]) - 1; 1132 mask = 1L << i; 1133 hash &= ~(VSID_NBPW-1); 1134 hash |= i; 1135 } 1136 /* 1137 * Make sure clear out SR_KEY_LEN bits because we put our 1138 * our data in those bits (to identify the segment). 1139 */ 1140 hash &= PTE_VSID >> (PTE_VSID_SHFT + SR_KEY_LEN); 1141 pmap_vsid_bitmap[n] |= mask; 1142 for (i = 0; i < 16; i++) 1143 pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY | 1144 SR_NOEXEC; 1145 return; 1146 } 1147 panic("pmap_pinit: out of segments"); 1148 } 1149 1150 /* 1151 * Add a reference to the given pmap. 1152 */ 1153 void 1154 pmap_reference(pmap_t pm) 1155 { 1156 pm->pm_refs++; 1157 } 1158 1159 /* 1160 * Retire the given pmap from service. 1161 * Should only be called if the map contains no valid mappings. 1162 */ 1163 void 1164 pmap_destroy(pmap_t pm) 1165 { 1166 if (--pm->pm_refs == 0) { 1167 pmap_release(pm); 1168 pool_put(&pmap_pool, pm); 1169 } 1170 } 1171 1172 /* 1173 * Release any resources held by the given physical map. 1174 * Called when a pmap initialized by pmap_pinit is being released. 1175 */ 1176 void 1177 pmap_release(pmap_t pm) 1178 { 1179 int idx, mask; 1180 1181 if (pm->pm_sr[0] == 0) 1182 panic("pmap_release"); 1183 idx = VSID_TO_HASH(pm->pm_sr[0]) & (NPMAPS-1); 1184 mask = 1 << (idx % VSID_NBPW); 1185 idx /= VSID_NBPW; 1186 pmap_vsid_bitmap[idx] &= ~mask; 1187 } 1188 1189 /* 1190 * Copy the range specified by src_addr/len 1191 * from the source map to the range dst_addr/len 1192 * in the destination map. 1193 * 1194 * This routine is only advisory and need not do anything. 1195 */ 1196 void 1197 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr, 1198 vsize_t len, vaddr_t src_addr) 1199 { 1200 PMAPCOUNT(copies); 1201 } 1202 1203 /* 1204 * Require that all active physical maps contain no 1205 * incorrect entries NOW. 1206 */ 1207 void 1208 pmap_update(struct pmap *pmap) 1209 { 1210 PMAPCOUNT(updates); 1211 TLBSYNC(); 1212 } 1213 1214 /* 1215 * Garbage collects the physical map system for 1216 * pages which are no longer used. 1217 * Success need not be guaranteed -- that is, there 1218 * may well be pages which are not referenced, but 1219 * others may be collected. 1220 * Called by the pageout daemon when pages are scarce. 1221 */ 1222 void 1223 pmap_collect(pmap_t pm) 1224 { 1225 PMAPCOUNT(collects); 1226 } 1227 1228 static __inline int 1229 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx) 1230 { 1231 int pteidx; 1232 /* 1233 * We can find the actual pte entry without searching by 1234 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9] 1235 * and by noticing the HID bit. 1236 */ 1237 pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo); 1238 if (pvo->pvo_pte.pte_hi & PTE_HID) 1239 pteidx ^= pmap_pteg_mask * 8; 1240 return pteidx; 1241 } 1242 1243 volatile struct pte * 1244 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx) 1245 { 1246 volatile struct pte *pt; 1247 1248 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1249 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) 1250 return NULL; 1251 #endif 1252 1253 /* 1254 * If we haven't been supplied the ptegidx, calculate it. 1255 */ 1256 if (pteidx == -1) { 1257 int ptegidx; 1258 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1259 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1260 } 1261 1262 pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7]; 1263 1264 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK) 1265 return pt; 1266 #else 1267 if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) { 1268 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1269 "pvo but no valid pte index", pvo); 1270 } 1271 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) { 1272 panic("pmap_pvo_to_pte: pvo %p: has valid pte index in " 1273 "pvo but no valid pte", pvo); 1274 } 1275 1276 if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) { 1277 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) { 1278 #if defined(DEBUG) || defined(PMAPCHECK) 1279 pmap_pte_print(pt); 1280 #endif 1281 panic("pmap_pvo_to_pte: pvo %p: has valid pte in " 1282 "pmap_pteg_table %p but invalid in pvo", 1283 pvo, pt); 1284 } 1285 if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) { 1286 #if defined(DEBUG) || defined(PMAPCHECK) 1287 pmap_pte_print(pt); 1288 #endif 1289 panic("pmap_pvo_to_pte: pvo %p: pvo pte does " 1290 "not match pte %p in pmap_pteg_table", 1291 pvo, pt); 1292 } 1293 return pt; 1294 } 1295 1296 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1297 #if defined(DEBUG) || defined(PMAPCHECK) 1298 pmap_pte_print(pt); 1299 #endif 1300 panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in " 1301 "pmap_pteg_table but valid in pvo", pvo, pt); 1302 } 1303 return NULL; 1304 #endif /* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */ 1305 } 1306 1307 struct pvo_entry * 1308 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p) 1309 { 1310 struct pvo_entry *pvo; 1311 int ptegidx; 1312 1313 va &= ~ADDR_POFF; 1314 ptegidx = va_to_pteg(pm, va); 1315 1316 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1317 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1318 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 1319 panic("pmap_pvo_find_va: invalid pvo %p on " 1320 "list %#x (%p)", pvo, ptegidx, 1321 &pmap_pvo_table[ptegidx]); 1322 #endif 1323 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1324 if (pteidx_p) 1325 *pteidx_p = pmap_pvo_pte_index(pvo, ptegidx); 1326 return pvo; 1327 } 1328 } 1329 return NULL; 1330 } 1331 1332 #if defined(DEBUG) || defined(PMAPCHECK) 1333 void 1334 pmap_pvo_check(const struct pvo_entry *pvo) 1335 { 1336 struct pvo_head *pvo_head; 1337 struct pvo_entry *pvo0; 1338 volatile struct pte *pt; 1339 int failed = 0; 1340 1341 if ((uintptr_t)(pvo+1) >= SEGMENT_LENGTH) 1342 panic("pmap_pvo_check: pvo %p: invalid address", pvo); 1343 1344 if ((uintptr_t)(pvo->pvo_pmap+1) >= SEGMENT_LENGTH) { 1345 printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n", 1346 pvo, pvo->pvo_pmap); 1347 failed = 1; 1348 } 1349 1350 if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= SEGMENT_LENGTH || 1351 (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) { 1352 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1353 pvo, TAILQ_NEXT(pvo, pvo_olink)); 1354 failed = 1; 1355 } 1356 1357 if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= SEGMENT_LENGTH || 1358 (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) { 1359 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n", 1360 pvo, LIST_NEXT(pvo, pvo_vlink)); 1361 failed = 1; 1362 } 1363 1364 if (pvo->pvo_vaddr & PVO_MANAGED) { 1365 pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL); 1366 } else { 1367 if (pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS) { 1368 printf("pmap_pvo_check: pvo %p: non kernel address " 1369 "on kernel unmanaged list\n", pvo); 1370 failed = 1; 1371 } 1372 pvo_head = &pmap_pvo_kunmanaged; 1373 } 1374 LIST_FOREACH(pvo0, pvo_head, pvo_vlink) { 1375 if (pvo0 == pvo) 1376 break; 1377 } 1378 if (pvo0 == NULL) { 1379 printf("pmap_pvo_check: pvo %p: not present " 1380 "on its vlist head %p\n", pvo, pvo_head); 1381 failed = 1; 1382 } 1383 if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) { 1384 printf("pmap_pvo_check: pvo %p: not present " 1385 "on its olist head\n", pvo); 1386 failed = 1; 1387 } 1388 pt = pmap_pvo_to_pte(pvo, -1); 1389 if (pt == NULL) { 1390 if (pvo->pvo_pte.pte_hi & PTE_VALID) { 1391 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1392 "no PTE\n", pvo); 1393 failed = 1; 1394 } 1395 } else { 1396 if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] || 1397 (uintptr_t) pt >= 1398 (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) { 1399 printf("pmap_pvo_check: pvo %p: pte %p not in " 1400 "pteg table\n", pvo, pt); 1401 failed = 1; 1402 } 1403 if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) { 1404 printf("pmap_pvo_check: pvo %p: pte_hi VALID but " 1405 "no PTE\n", pvo); 1406 failed = 1; 1407 } 1408 if (pvo->pvo_pte.pte_hi != pt->pte_hi) { 1409 printf("pmap_pvo_check: pvo %p: pte_hi differ: " 1410 "%#lx/%#lx\n", pvo, pvo->pvo_pte.pte_hi, pt->pte_hi); 1411 failed = 1; 1412 } 1413 if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) & 1414 (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) { 1415 printf("pmap_pvo_check: pvo %p: pte_lo differ: " 1416 "%#lx/%#lx\n", pvo, 1417 pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN), 1418 pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)); 1419 failed = 1; 1420 } 1421 if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) { 1422 printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#lx" 1423 " doesn't not match PVO's VA %#lx\n", 1424 pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo)); 1425 failed = 1; 1426 } 1427 if (failed) 1428 pmap_pte_print(pt); 1429 } 1430 if (failed) 1431 panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo, 1432 pvo->pvo_pmap); 1433 } 1434 #endif /* DEBUG || PMAPCHECK */ 1435 1436 /* 1437 * This returns whether this is the first mapping of a page. 1438 */ 1439 int 1440 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head, 1441 vaddr_t va, paddr_t pa, register_t pte_lo, int flags) 1442 { 1443 struct pvo_entry *pvo; 1444 struct pvo_tqhead *pvoh; 1445 register_t msr; 1446 int ptegidx; 1447 int i; 1448 int poolflags = PR_NOWAIT; 1449 1450 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1451 if (pmap_pvo_remove_depth > 0) 1452 panic("pmap_pvo_enter: called while pmap_pvo_remove active!"); 1453 if (++pmap_pvo_enter_depth > 1) 1454 panic("pmap_pvo_enter: called recursively!"); 1455 #endif 1456 1457 /* 1458 * Compute the PTE Group index. 1459 */ 1460 va &= ~ADDR_POFF; 1461 ptegidx = va_to_pteg(pm, va); 1462 1463 msr = pmap_interrupts_off(); 1464 /* 1465 * Remove any existing mapping for this page. Reuse the 1466 * pvo entry if there a mapping. 1467 */ 1468 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 1469 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 1470 #ifdef DEBUG 1471 if ((pmapdebug & PMAPDEBUG_PVOENTER) && 1472 ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) & 1473 ~(PTE_REF|PTE_CHG)) == 0 && 1474 va < VM_MIN_KERNEL_ADDRESS) { 1475 printf("pmap_pvo_enter: pvo %p: dup %#lx/%#lx\n", 1476 pvo, pvo->pvo_pte.pte_lo, pte_lo|pa); 1477 printf("pmap_pvo_enter: pte_hi=%#lx sr=%#lx\n", 1478 pvo->pvo_pte.pte_hi, 1479 pm->pm_sr[va >> ADDR_SR_SHFT]); 1480 pmap_pte_print(pmap_pvo_to_pte(pvo, -1)); 1481 #ifdef DDBX 1482 Debugger(); 1483 #endif 1484 } 1485 #endif 1486 PMAPCOUNT(mappings_replaced); 1487 pmap_pvo_remove(pvo, -1); 1488 break; 1489 } 1490 } 1491 1492 /* 1493 * If we aren't overwriting an mapping, try to allocate 1494 */ 1495 pmap_interrupts_restore(msr); 1496 pvo = pool_get(pl, poolflags); 1497 msr = pmap_interrupts_off(); 1498 if (pvo == NULL) { 1499 pvo = pmap_pvo_reclaim(pm); 1500 if (pvo == NULL) { 1501 if ((flags & PMAP_CANFAIL) == 0) 1502 panic("pmap_pvo_enter: failed"); 1503 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1504 pmap_pvo_enter_depth--; 1505 #endif 1506 pmap_interrupts_restore(msr); 1507 return ENOMEM; 1508 } 1509 } 1510 pvo->pvo_vaddr = va; 1511 pvo->pvo_pmap = pm; 1512 pvo->pvo_vaddr &= ~ADDR_POFF; 1513 if (flags & VM_PROT_EXECUTE) { 1514 PMAPCOUNT(exec_mappings); 1515 pvo_set_exec(pvo); 1516 } 1517 if (flags & PMAP_WIRED) 1518 pvo->pvo_vaddr |= PVO_WIRED; 1519 if (pvo_head != &pmap_pvo_kunmanaged) { 1520 pvo->pvo_vaddr |= PVO_MANAGED; 1521 PMAPCOUNT(mappings); 1522 } else { 1523 PMAPCOUNT(kernel_mappings); 1524 } 1525 pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo); 1526 1527 LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink); 1528 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1529 pvo->pvo_pmap->pm_stats.wired_count++; 1530 pvo->pvo_pmap->pm_stats.resident_count++; 1531 #if defined(DEBUG) 1532 if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) 1533 DPRINTFN(PVOENTER, 1534 ("pmap_pvo_enter: pvo %p: pm %p va %#lx pa %#lx\n", 1535 pvo, pm, va, pa)); 1536 #endif 1537 1538 /* 1539 * We hope this succeeds but it isn't required. 1540 */ 1541 pvoh = &pmap_pvo_table[ptegidx]; 1542 i = pmap_pte_insert(ptegidx, &pvo->pvo_pte); 1543 if (i >= 0) { 1544 PVO_PTEGIDX_SET(pvo, i); 1545 PVO_WHERE(pvo, ENTER_INSERT); 1546 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) 1547 ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]); 1548 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink); 1549 } else { 1550 /* 1551 * Since we didn't have room for this entry (which makes it 1552 * and evicted entry), place it at the head of the list. 1553 */ 1554 TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink); 1555 PMAPCOUNT(ptes_evicted); 1556 pm->pm_evictions++; 1557 /* 1558 * If this is a kernel page, make sure it's active. 1559 */ 1560 if (pm == pmap_kernel()) { 1561 i = pmap_pte_spill(pm, va, FALSE); 1562 KASSERT(i); 1563 } 1564 } 1565 PMAP_PVO_CHECK(pvo); /* sanity check */ 1566 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1567 pmap_pvo_enter_depth--; 1568 #endif 1569 pmap_interrupts_restore(msr); 1570 return 0; 1571 } 1572 1573 void 1574 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx) 1575 { 1576 volatile struct pte *pt; 1577 int ptegidx; 1578 1579 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1580 if (++pmap_pvo_remove_depth > 1) 1581 panic("pmap_pvo_remove: called recursively!"); 1582 #endif 1583 1584 /* 1585 * If we haven't been supplied the ptegidx, calculate it. 1586 */ 1587 if (pteidx == -1) { 1588 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr); 1589 pteidx = pmap_pvo_pte_index(pvo, ptegidx); 1590 } else { 1591 ptegidx = pteidx >> 3; 1592 if (pvo->pvo_pte.pte_hi & PTE_HID) 1593 ptegidx ^= pmap_pteg_mask; 1594 } 1595 PMAP_PVO_CHECK(pvo); /* sanity check */ 1596 1597 /* 1598 * If there is an active pte entry, we need to deactivate it 1599 * (and save the ref & chg bits). 1600 */ 1601 pt = pmap_pvo_to_pte(pvo, pteidx); 1602 if (pt != NULL) { 1603 pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 1604 PVO_WHERE(pvo, REMOVE); 1605 PVO_PTEGIDX_CLR(pvo); 1606 PMAPCOUNT(ptes_removed); 1607 } else { 1608 KASSERT(pvo->pvo_pmap->pm_evictions > 0); 1609 pvo->pvo_pmap->pm_evictions--; 1610 } 1611 1612 /* 1613 * Account for executable mappings. 1614 */ 1615 if (PVO_ISEXECUTABLE(pvo)) 1616 pvo_clear_exec(pvo); 1617 1618 /* 1619 * Update our statistics. 1620 */ 1621 pvo->pvo_pmap->pm_stats.resident_count--; 1622 if (pvo->pvo_pte.pte_lo & PVO_WIRED) 1623 pvo->pvo_pmap->pm_stats.wired_count--; 1624 1625 /* 1626 * Save the REF/CHG bits into their cache if the page is managed. 1627 */ 1628 if (pvo->pvo_vaddr & PVO_MANAGED) { 1629 register_t ptelo = pvo->pvo_pte.pte_lo; 1630 struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN); 1631 1632 if (pg != NULL) { 1633 pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG)); 1634 } 1635 PMAPCOUNT(unmappings); 1636 } else { 1637 PMAPCOUNT(kernel_unmappings); 1638 } 1639 1640 /* 1641 * Remove the PVO from its lists and return it to the pool. 1642 */ 1643 LIST_REMOVE(pvo, pvo_vlink); 1644 TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink); 1645 pool_put(pvo->pvo_vaddr & PVO_MANAGED 1646 ? &pmap_mpvo_pool : &pmap_upvo_pool, pvo); 1647 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 1648 pmap_pvo_remove_depth--; 1649 #endif 1650 } 1651 1652 /* 1653 * Mark a mapping as executable. 1654 * If this is the first executable mapping in the segment, 1655 * clear the noexec flag. 1656 */ 1657 STATIC void 1658 pvo_set_exec(struct pvo_entry *pvo) 1659 { 1660 struct pmap *pm = pvo->pvo_pmap; 1661 int sr; 1662 1663 if (pm == pmap_kernel() || PVO_ISEXECUTABLE(pvo)) { 1664 return; 1665 } 1666 pvo->pvo_vaddr |= PVO_EXECUTABLE; 1667 sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1668 if (pm->pm_exec[sr]++ == 0) { 1669 pm->pm_sr[sr] &= ~SR_NOEXEC; 1670 } 1671 } 1672 1673 /* 1674 * Mark a mapping as non-executable. 1675 * If this was the last executable mapping in the segment, 1676 * set the noexec flag. 1677 */ 1678 STATIC void 1679 pvo_clear_exec(struct pvo_entry *pvo) 1680 { 1681 struct pmap *pm = pvo->pvo_pmap; 1682 int sr; 1683 1684 if (pm == pmap_kernel() || !PVO_ISEXECUTABLE(pvo)) { 1685 return; 1686 } 1687 pvo->pvo_vaddr &= ~PVO_EXECUTABLE; 1688 sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT; 1689 if (--pm->pm_exec[sr] == 0) { 1690 pm->pm_sr[sr] |= SR_NOEXEC; 1691 } 1692 } 1693 1694 /* 1695 * Insert physical page at pa into the given pmap at virtual address va. 1696 */ 1697 int 1698 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, int flags) 1699 { 1700 struct mem_region *mp; 1701 struct pvo_head *pvo_head; 1702 struct vm_page *pg; 1703 struct pool *pl; 1704 register_t pte_lo; 1705 int error; 1706 u_int pvo_flags; 1707 u_int was_exec = 0; 1708 1709 if (__predict_false(!pmap_initialized)) { 1710 pvo_head = &pmap_pvo_kunmanaged; 1711 pl = &pmap_upvo_pool; 1712 pvo_flags = 0; 1713 pg = NULL; 1714 was_exec = PTE_EXEC; 1715 } else { 1716 pvo_head = pa_to_pvoh(pa, &pg); 1717 pl = &pmap_mpvo_pool; 1718 pvo_flags = PVO_MANAGED; 1719 } 1720 1721 DPRINTFN(ENTER, 1722 ("pmap_enter(%p, 0x%lx, 0x%lx, 0x%x, 0x%x):", 1723 pm, va, pa, prot, flags)); 1724 1725 /* 1726 * If this is a managed page, and it's the first reference to the 1727 * page clear the execness of the page. Otherwise fetch the execness. 1728 */ 1729 if (pg != NULL) 1730 was_exec = pmap_attr_fetch(pg) & PTE_EXEC; 1731 1732 DPRINTFN(ENTER, (" was_exec=%d", was_exec)); 1733 1734 /* 1735 * Assume the page is cache inhibited and access is guarded unless 1736 * it's in our available memory array. If it is in the memory array, 1737 * asssume it's in memory coherent memory. 1738 */ 1739 pte_lo = PTE_IG; 1740 if ((flags & PMAP_NC) == 0) { 1741 for (mp = mem; mp->size; mp++) { 1742 if (pa >= mp->start && pa < mp->start + mp->size) { 1743 pte_lo = PTE_M; 1744 break; 1745 } 1746 } 1747 } 1748 1749 if (prot & VM_PROT_WRITE) 1750 pte_lo |= PTE_BW; 1751 else 1752 pte_lo |= PTE_BR; 1753 1754 /* 1755 * If this was in response to a fault, "pre-fault" the PTE's 1756 * changed/referenced bit appropriately. 1757 */ 1758 if (flags & VM_PROT_WRITE) 1759 pte_lo |= PTE_CHG; 1760 if (flags & (VM_PROT_READ|VM_PROT_WRITE)) 1761 pte_lo |= PTE_REF; 1762 1763 /* 1764 * We need to know if this page can be executable 1765 */ 1766 flags |= (prot & VM_PROT_EXECUTE); 1767 1768 /* 1769 * Record mapping for later back-translation and pte spilling. 1770 * This will overwrite any existing mapping. 1771 */ 1772 error = pmap_pvo_enter(pm, pl, pvo_head, va, pa, pte_lo, flags); 1773 1774 /* 1775 * Flush the real page from the instruction cache if this page is 1776 * mapped executable and cacheable and has not been flushed since 1777 * the last time it was modified. 1778 */ 1779 if (error == 0 && 1780 (flags & VM_PROT_EXECUTE) && 1781 (pte_lo & PTE_I) == 0 && 1782 was_exec == 0) { 1783 DPRINTFN(ENTER, (" syncicache")); 1784 PMAPCOUNT(exec_synced); 1785 pmap_syncicache(pa, PAGE_SIZE); 1786 if (pg != NULL) { 1787 pmap_attr_save(pg, PTE_EXEC); 1788 PMAPCOUNT(exec_cached); 1789 #if defined(DEBUG) || defined(PMAPDEBUG) 1790 if (pmapdebug & PMAPDEBUG_ENTER) 1791 printf(" marked-as-exec"); 1792 else if (pmapdebug & PMAPDEBUG_EXEC) 1793 printf("[pmap_enter: %#lx: marked-as-exec]\n", 1794 pg->phys_addr); 1795 1796 #endif 1797 } 1798 } 1799 1800 DPRINTFN(ENTER, (": error=%d\n", error)); 1801 1802 return error; 1803 } 1804 1805 void 1806 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot) 1807 { 1808 struct mem_region *mp; 1809 register_t pte_lo; 1810 int error; 1811 1812 if (va < VM_MIN_KERNEL_ADDRESS) 1813 panic("pmap_kenter_pa: attempt to enter " 1814 "non-kernel address %#lx!", va); 1815 1816 DPRINTFN(KENTER, 1817 ("pmap_kenter_pa(%#lx,%#lx,%#x)\n", va, pa, prot)); 1818 1819 /* 1820 * Assume the page is cache inhibited and access is guarded unless 1821 * it's in our available memory array. If it is in the memory array, 1822 * asssume it's in memory coherent memory. 1823 */ 1824 pte_lo = PTE_IG; 1825 if ((prot & PMAP_NC) == 0) { 1826 for (mp = mem; mp->size; mp++) { 1827 if (pa >= mp->start && pa < mp->start + mp->size) { 1828 pte_lo = PTE_M; 1829 break; 1830 } 1831 } 1832 } 1833 1834 if (prot & VM_PROT_WRITE) 1835 pte_lo |= PTE_BW; 1836 else 1837 pte_lo |= PTE_BR; 1838 1839 /* 1840 * We don't care about REF/CHG on PVOs on the unmanaged list. 1841 */ 1842 error = pmap_pvo_enter(pmap_kernel(), &pmap_upvo_pool, 1843 &pmap_pvo_kunmanaged, va, pa, pte_lo, prot|PMAP_WIRED); 1844 1845 if (error != 0) 1846 panic("pmap_kenter_pa: failed to enter va %#lx pa %#lx: %d", 1847 va, pa, error); 1848 } 1849 1850 void 1851 pmap_kremove(vaddr_t va, vsize_t len) 1852 { 1853 if (va < VM_MIN_KERNEL_ADDRESS) 1854 panic("pmap_kremove: attempt to remove " 1855 "non-kernel address %#lx!", va); 1856 1857 DPRINTFN(KREMOVE,("pmap_kremove(%#lx,%#lx)\n", va, len)); 1858 pmap_remove(pmap_kernel(), va, va + len); 1859 } 1860 1861 /* 1862 * Remove the given range of mapping entries. 1863 */ 1864 void 1865 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva) 1866 { 1867 struct pvo_entry *pvo; 1868 register_t msr; 1869 int pteidx; 1870 1871 msr = pmap_interrupts_off(); 1872 for (; va < endva; va += PAGE_SIZE) { 1873 pvo = pmap_pvo_find_va(pm, va, &pteidx); 1874 if (pvo != NULL) { 1875 pmap_pvo_remove(pvo, pteidx); 1876 } 1877 } 1878 pmap_interrupts_restore(msr); 1879 } 1880 1881 /* 1882 * Get the physical page address for the given pmap/virtual address. 1883 */ 1884 boolean_t 1885 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap) 1886 { 1887 struct pvo_entry *pvo; 1888 register_t msr; 1889 1890 /* 1891 * If this is a kernel pmap lookup, also check the battable 1892 * and if we get a hit, translate the VA to a PA using the 1893 * BAT entries. Don't check for VM_MAX_KENREL_ADDRESS is 1894 * that will wrap back to 0. 1895 */ 1896 if (pm == pmap_kernel() && 1897 (va < VM_MIN_KERNEL_ADDRESS || 1898 (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) { 1899 register_t batu = battable[va >> ADDR_SR_SHFT].batu; 1900 KASSERT((va >> ADDR_SR_SHFT) != USER_SR); 1901 if (BAT_VALID_P(batu,0) && BAT_VA_MATCH_P(batu,va)) { 1902 register_t batl = battable[va >> ADDR_SR_SHFT].batl; 1903 register_t mask = (~(batu & BAT_BL) << 15) & ~0x1ffffL; 1904 *pap = (batl & mask) | (va & ~mask); 1905 return TRUE; 1906 } 1907 return FALSE; 1908 } 1909 1910 msr = pmap_interrupts_off(); 1911 pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL); 1912 if (pvo != NULL) { 1913 PMAP_PVO_CHECK(pvo); /* sanity check */ 1914 *pap = (pvo->pvo_pte.pte_lo & PTE_RPGN) | (va & ADDR_POFF); 1915 } 1916 pmap_interrupts_restore(msr); 1917 return pvo != NULL; 1918 } 1919 1920 /* 1921 * Lower the protection on the specified range of this pmap. 1922 */ 1923 void 1924 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot) 1925 { 1926 struct pvo_entry *pvo; 1927 volatile struct pte *pt; 1928 register_t msr; 1929 int pteidx; 1930 1931 /* 1932 * Since this routine only downgrades protection, we should 1933 * always be called with at least one bit not set. 1934 */ 1935 KASSERT(prot != VM_PROT_ALL); 1936 1937 /* 1938 * If there is no protection, this is equivalent to 1939 * remove the pmap from the pmap. 1940 */ 1941 if ((prot & VM_PROT_READ) == 0) { 1942 pmap_remove(pm, va, endva); 1943 return; 1944 } 1945 1946 msr = pmap_interrupts_off(); 1947 for (; va < endva; va += PAGE_SIZE) { 1948 pvo = pmap_pvo_find_va(pm, va, &pteidx); 1949 if (pvo == NULL) 1950 continue; 1951 PMAP_PVO_CHECK(pvo); /* sanity check */ 1952 1953 /* 1954 * Revoke executable if asked to do so. 1955 */ 1956 if ((prot & VM_PROT_EXECUTE) == 0) 1957 pvo_clear_exec(pvo); 1958 1959 #if 0 1960 /* 1961 * If the page is already read-only, no change 1962 * needs to be made. 1963 */ 1964 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) 1965 continue; 1966 #endif 1967 /* 1968 * Grab the PTE pointer before we diddle with 1969 * the cached PTE copy. 1970 */ 1971 pt = pmap_pvo_to_pte(pvo, pteidx); 1972 /* 1973 * Change the protection of the page. 1974 */ 1975 pvo->pvo_pte.pte_lo &= ~PTE_PP; 1976 pvo->pvo_pte.pte_lo |= PTE_BR; 1977 1978 /* 1979 * If the PVO is in the page table, update 1980 * that pte at well. 1981 */ 1982 if (pt != NULL) { 1983 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 1984 PVO_WHERE(pvo, PMAP_PROTECT); 1985 PMAPCOUNT(ptes_changed); 1986 } 1987 1988 PMAP_PVO_CHECK(pvo); /* sanity check */ 1989 } 1990 pmap_interrupts_restore(msr); 1991 } 1992 1993 void 1994 pmap_unwire(pmap_t pm, vaddr_t va) 1995 { 1996 struct pvo_entry *pvo; 1997 register_t msr; 1998 1999 msr = pmap_interrupts_off(); 2000 pvo = pmap_pvo_find_va(pm, va, NULL); 2001 if (pvo != NULL) { 2002 if (pvo->pvo_vaddr & PVO_WIRED) { 2003 pvo->pvo_vaddr &= ~PVO_WIRED; 2004 pm->pm_stats.wired_count--; 2005 } 2006 PMAP_PVO_CHECK(pvo); /* sanity check */ 2007 } 2008 pmap_interrupts_restore(msr); 2009 } 2010 2011 /* 2012 * Lower the protection on the specified physical page. 2013 */ 2014 void 2015 pmap_page_protect(struct vm_page *pg, vm_prot_t prot) 2016 { 2017 struct pvo_head *pvo_head; 2018 struct pvo_entry *pvo, *next_pvo; 2019 volatile struct pte *pt; 2020 register_t msr; 2021 2022 KASSERT(prot != VM_PROT_ALL); 2023 msr = pmap_interrupts_off(); 2024 2025 /* 2026 * When UVM reuses a page, it does a pmap_page_protect with 2027 * VM_PROT_NONE. At that point, we can clear the exec flag 2028 * since we know the page will have different contents. 2029 */ 2030 if ((prot & VM_PROT_READ) == 0) { 2031 DPRINTFN(EXEC, ("[pmap_page_protect: %#lx: clear-exec]\n", 2032 pg->phys_addr)); 2033 if (pmap_attr_fetch(pg) & PTE_EXEC) { 2034 PMAPCOUNT(exec_uncached_page_protect); 2035 pmap_attr_clear(pg, PTE_EXEC); 2036 } 2037 } 2038 2039 pvo_head = vm_page_to_pvoh(pg); 2040 for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) { 2041 next_pvo = LIST_NEXT(pvo, pvo_vlink); 2042 PMAP_PVO_CHECK(pvo); /* sanity check */ 2043 2044 /* 2045 * Downgrading to no mapping at all, we just remove the entry. 2046 */ 2047 if ((prot & VM_PROT_READ) == 0) { 2048 pmap_pvo_remove(pvo, -1); 2049 continue; 2050 } 2051 2052 /* 2053 * If EXEC permission is being revoked, just clear the 2054 * flag in the PVO. 2055 */ 2056 if ((prot & VM_PROT_EXECUTE) == 0) 2057 pvo_clear_exec(pvo); 2058 2059 /* 2060 * If this entry is already RO, don't diddle with the 2061 * page table. 2062 */ 2063 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) { 2064 PMAP_PVO_CHECK(pvo); 2065 continue; 2066 } 2067 2068 /* 2069 * Grab the PTE before the we diddle the bits so 2070 * pvo_to_pte can verify the pte contents are as 2071 * expected. 2072 */ 2073 pt = pmap_pvo_to_pte(pvo, -1); 2074 pvo->pvo_pte.pte_lo &= ~PTE_PP; 2075 pvo->pvo_pte.pte_lo |= PTE_BR; 2076 if (pt != NULL) { 2077 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr); 2078 PVO_WHERE(pvo, PMAP_PAGE_PROTECT); 2079 PMAPCOUNT(ptes_changed); 2080 } 2081 PMAP_PVO_CHECK(pvo); /* sanity check */ 2082 } 2083 pmap_interrupts_restore(msr); 2084 } 2085 2086 /* 2087 * Activate the address space for the specified process. If the process 2088 * is the current process, load the new MMU context. 2089 */ 2090 void 2091 pmap_activate(struct lwp *l) 2092 { 2093 struct pcb *pcb = &l->l_addr->u_pcb; 2094 pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap; 2095 2096 DPRINTFN(ACTIVATE, 2097 ("pmap_activate: lwp %p (curlwp %p)\n", l, curlwp)); 2098 2099 /* 2100 * XXX Normally performed in cpu_fork(). 2101 */ 2102 pcb->pcb_pm = pmap; 2103 } 2104 2105 /* 2106 * Deactivate the specified process's address space. 2107 */ 2108 void 2109 pmap_deactivate(struct lwp *l) 2110 { 2111 } 2112 2113 boolean_t 2114 pmap_query_bit(struct vm_page *pg, int ptebit) 2115 { 2116 struct pvo_entry *pvo; 2117 volatile struct pte *pt; 2118 register_t msr; 2119 2120 if (pmap_attr_fetch(pg) & ptebit) 2121 return TRUE; 2122 2123 msr = pmap_interrupts_off(); 2124 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2125 PMAP_PVO_CHECK(pvo); /* sanity check */ 2126 /* 2127 * See if we saved the bit off. If so cache, it and return 2128 * success. 2129 */ 2130 if (pvo->pvo_pte.pte_lo & ptebit) { 2131 pmap_attr_save(pg, ptebit); 2132 PMAP_PVO_CHECK(pvo); /* sanity check */ 2133 pmap_interrupts_restore(msr); 2134 return TRUE; 2135 } 2136 } 2137 /* 2138 * No luck, now go thru the hard part of looking at the ptes 2139 * themselves. Sync so any pending REF/CHG bits are flushed 2140 * to the PTEs. 2141 */ 2142 SYNC(); 2143 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) { 2144 PMAP_PVO_CHECK(pvo); /* sanity check */ 2145 /* 2146 * See if this pvo have a valid PTE. If so, fetch the 2147 * REF/CHG bits from the valid PTE. If the appropriate 2148 * ptebit is set, cache, it and return success. 2149 */ 2150 pt = pmap_pvo_to_pte(pvo, -1); 2151 if (pt != NULL) { 2152 pmap_pte_synch(pt, &pvo->pvo_pte); 2153 if (pvo->pvo_pte.pte_lo & ptebit) { 2154 pmap_attr_save(pg, ptebit); 2155 PMAP_PVO_CHECK(pvo); /* sanity check */ 2156 pmap_interrupts_restore(msr); 2157 return TRUE; 2158 } 2159 } 2160 } 2161 pmap_interrupts_restore(msr); 2162 return FALSE; 2163 } 2164 2165 boolean_t 2166 pmap_clear_bit(struct vm_page *pg, int ptebit) 2167 { 2168 struct pvo_head *pvoh = vm_page_to_pvoh(pg); 2169 struct pvo_entry *pvo; 2170 volatile struct pte *pt; 2171 register_t msr; 2172 int rv = 0; 2173 2174 msr = pmap_interrupts_off(); 2175 2176 /* 2177 * Fetch the cache value 2178 */ 2179 rv |= pmap_attr_fetch(pg); 2180 2181 /* 2182 * Clear the cached value. 2183 */ 2184 pmap_attr_clear(pg, ptebit); 2185 2186 /* 2187 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we 2188 * can reset the right ones). Note that since the pvo entries and 2189 * list heads are accessed via BAT0 and are never placed in the 2190 * page table, we don't have to worry about further accesses setting 2191 * the REF/CHG bits. 2192 */ 2193 SYNC(); 2194 2195 /* 2196 * For each pvo entry, clear pvo's ptebit. If this pvo have a 2197 * valid PTE. If so, clear the ptebit from the valid PTE. 2198 */ 2199 LIST_FOREACH(pvo, pvoh, pvo_vlink) { 2200 PMAP_PVO_CHECK(pvo); /* sanity check */ 2201 pt = pmap_pvo_to_pte(pvo, -1); 2202 if (pt != NULL) { 2203 /* 2204 * Only sync the PTE if the bit we are looking 2205 * for is not already set. 2206 */ 2207 if ((pvo->pvo_pte.pte_lo & ptebit) == 0) 2208 pmap_pte_synch(pt, &pvo->pvo_pte); 2209 /* 2210 * If the bit we are looking for was already set, 2211 * clear that bit in the pte. 2212 */ 2213 if (pvo->pvo_pte.pte_lo & ptebit) 2214 pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit); 2215 } 2216 rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF); 2217 pvo->pvo_pte.pte_lo &= ~ptebit; 2218 PMAP_PVO_CHECK(pvo); /* sanity check */ 2219 } 2220 pmap_interrupts_restore(msr); 2221 2222 /* 2223 * If we are clearing the modify bit and this page was marked EXEC 2224 * and the user of the page thinks the page was modified, then we 2225 * need to clean it from the icache if it's mapped or clear the EXEC 2226 * bit if it's not mapped. The page itself might not have the CHG 2227 * bit set if the modification was done via DMA to the page. 2228 */ 2229 if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) { 2230 if (LIST_EMPTY(pvoh)) { 2231 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: clear-exec]\n", 2232 pg->phys_addr)); 2233 pmap_attr_clear(pg, PTE_EXEC); 2234 PMAPCOUNT(exec_uncached_clear_modify); 2235 } else { 2236 DPRINTFN(EXEC, ("[pmap_clear_bit: %#lx: syncicache]\n", 2237 pg->phys_addr)); 2238 pmap_syncicache(pg->phys_addr, PAGE_SIZE); 2239 PMAPCOUNT(exec_synced_clear_modify); 2240 } 2241 } 2242 return (rv & ptebit) != 0; 2243 } 2244 2245 void 2246 pmap_procwr(struct proc *p, vaddr_t va, size_t len) 2247 { 2248 struct pvo_entry *pvo; 2249 size_t offset = va & ADDR_POFF; 2250 int s; 2251 2252 s = splvm(); 2253 while (len > 0) { 2254 size_t seglen = PAGE_SIZE - offset; 2255 if (seglen > len) 2256 seglen = len; 2257 pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL); 2258 if (pvo != NULL && PVO_ISEXECUTABLE(pvo)) { 2259 pmap_syncicache( 2260 (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen); 2261 PMAP_PVO_CHECK(pvo); 2262 } 2263 va += seglen; 2264 len -= seglen; 2265 offset = 0; 2266 } 2267 splx(s); 2268 } 2269 2270 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB) 2271 void 2272 pmap_pte_print(volatile struct pte *pt) 2273 { 2274 printf("PTE %p: ", pt); 2275 /* High word: */ 2276 printf("0x%08lx: [", pt->pte_hi); 2277 printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i'); 2278 printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-'); 2279 printf("0x%06lx 0x%02lx", 2280 (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT, 2281 pt->pte_hi & PTE_API); 2282 printf(" (va 0x%08lx)] ", pmap_pte_to_va(pt)); 2283 /* Low word: */ 2284 printf(" 0x%08lx: [", pt->pte_lo); 2285 printf("0x%05lx... ", pt->pte_lo >> 12); 2286 printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u'); 2287 printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n'); 2288 printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.'); 2289 printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.'); 2290 printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.'); 2291 printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.'); 2292 switch (pt->pte_lo & PTE_PP) { 2293 case PTE_BR: printf("br]\n"); break; 2294 case PTE_BW: printf("bw]\n"); break; 2295 case PTE_SO: printf("so]\n"); break; 2296 case PTE_SW: printf("sw]\n"); break; 2297 } 2298 } 2299 #endif 2300 2301 #if defined(DDB) 2302 void 2303 pmap_pteg_check(void) 2304 { 2305 volatile struct pte *pt; 2306 int i; 2307 int ptegidx; 2308 u_int p_valid = 0; 2309 u_int s_valid = 0; 2310 u_int invalid = 0; 2311 2312 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2313 for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) { 2314 if (pt->pte_hi & PTE_VALID) { 2315 if (pt->pte_hi & PTE_HID) 2316 s_valid++; 2317 else 2318 p_valid++; 2319 } else 2320 invalid++; 2321 } 2322 } 2323 printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n", 2324 p_valid, p_valid, s_valid, s_valid, 2325 invalid, invalid); 2326 } 2327 2328 void 2329 pmap_print_mmuregs(void) 2330 { 2331 int i; 2332 u_int cpuvers; 2333 vaddr_t addr; 2334 register_t soft_sr[16]; 2335 struct bat soft_ibat[4]; 2336 struct bat soft_dbat[4]; 2337 register_t sdr1; 2338 2339 cpuvers = MFPVR() >> 16; 2340 2341 __asm __volatile ("mfsdr1 %0" : "=r"(sdr1)); 2342 for (i=0; i<16; i++) { 2343 soft_sr[i] = MFSRIN(addr); 2344 addr += (1 << ADDR_SR_SHFT); 2345 } 2346 2347 /* read iBAT (601: uBAT) registers */ 2348 __asm __volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu)); 2349 __asm __volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl)); 2350 __asm __volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu)); 2351 __asm __volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl)); 2352 __asm __volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu)); 2353 __asm __volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl)); 2354 __asm __volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu)); 2355 __asm __volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl)); 2356 2357 2358 if (cpuvers != MPC601) { 2359 /* read dBAT registers */ 2360 __asm __volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu)); 2361 __asm __volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl)); 2362 __asm __volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu)); 2363 __asm __volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl)); 2364 __asm __volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu)); 2365 __asm __volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl)); 2366 __asm __volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu)); 2367 __asm __volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl)); 2368 } 2369 2370 printf("SDR1:\t%#lx\n", sdr1); 2371 printf("SR[]:\t"); 2372 addr = 0; 2373 for (i=0; i<4; i++) 2374 printf("0x%08lx, ", soft_sr[i]); 2375 printf("\n\t"); 2376 for ( ; i<8; i++) 2377 printf("0x%08lx, ", soft_sr[i]); 2378 printf("\n\t"); 2379 for ( ; i<12; i++) 2380 printf("0x%08lx, ", soft_sr[i]); 2381 printf("\n\t"); 2382 for ( ; i<16; i++) 2383 printf("0x%08lx, ", soft_sr[i]); 2384 printf("\n"); 2385 2386 printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i'); 2387 for (i=0; i<4; i++) { 2388 printf("0x%08lx 0x%08lx, ", 2389 soft_ibat[i].batu, soft_ibat[i].batl); 2390 if (i == 1) 2391 printf("\n\t"); 2392 } 2393 if (cpuvers != MPC601) { 2394 printf("\ndBAT[]:\t"); 2395 for (i=0; i<4; i++) { 2396 printf("0x%08lx 0x%08lx, ", 2397 soft_dbat[i].batu, soft_dbat[i].batl); 2398 if (i == 1) 2399 printf("\n\t"); 2400 } 2401 } 2402 printf("\n"); 2403 } 2404 2405 void 2406 pmap_print_pte(pmap_t pm, vaddr_t va) 2407 { 2408 struct pvo_entry *pvo; 2409 volatile struct pte *pt; 2410 int pteidx; 2411 2412 pvo = pmap_pvo_find_va(pm, va, &pteidx); 2413 if (pvo != NULL) { 2414 pt = pmap_pvo_to_pte(pvo, pteidx); 2415 if (pt != NULL) { 2416 printf("VA %#lx -> %p -> %s %#lx, %#lx\n", 2417 va, pt, 2418 pt->pte_hi & PTE_HID ? "(sec)" : "(pri)", 2419 pt->pte_hi, pt->pte_lo); 2420 } else { 2421 printf("No valid PTE found\n"); 2422 } 2423 } else { 2424 printf("Address not in pmap\n"); 2425 } 2426 } 2427 2428 void 2429 pmap_pteg_dist(void) 2430 { 2431 struct pvo_entry *pvo; 2432 int ptegidx; 2433 int depth; 2434 int max_depth = 0; 2435 unsigned int depths[64]; 2436 2437 memset(depths, 0, sizeof(depths)); 2438 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2439 depth = 0; 2440 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2441 depth++; 2442 } 2443 if (depth > max_depth) 2444 max_depth = depth; 2445 if (depth > 63) 2446 depth = 63; 2447 depths[depth]++; 2448 } 2449 2450 for (depth = 0; depth < 64; depth++) { 2451 printf(" [%2d]: %8u", depth, depths[depth]); 2452 if ((depth & 3) == 3) 2453 printf("\n"); 2454 if (depth == max_depth) 2455 break; 2456 } 2457 if ((depth & 3) != 3) 2458 printf("\n"); 2459 printf("Max depth found was %d\n", max_depth); 2460 } 2461 #endif /* DEBUG */ 2462 2463 #if defined(PMAPCHECK) || defined(DEBUG) 2464 void 2465 pmap_pvo_verify(void) 2466 { 2467 int ptegidx; 2468 int s; 2469 2470 s = splvm(); 2471 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) { 2472 struct pvo_entry *pvo; 2473 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) { 2474 if ((uintptr_t) pvo >= SEGMENT_LENGTH) 2475 panic("pmap_pvo_verify: invalid pvo %p " 2476 "on list %#x", pvo, ptegidx); 2477 pmap_pvo_check(pvo); 2478 } 2479 } 2480 splx(s); 2481 } 2482 #endif /* PMAPCHECK */ 2483 2484 2485 void * 2486 pmap_pool_ualloc(struct pool *pp, int flags) 2487 { 2488 struct pvo_page *pvop; 2489 2490 pvop = SIMPLEQ_FIRST(&pmap_upvop_head); 2491 if (pvop != NULL) { 2492 pmap_upvop_free--; 2493 SIMPLEQ_REMOVE_HEAD(&pmap_upvop_head, pvop_link); 2494 return pvop; 2495 } 2496 if (uvm.page_init_done != TRUE) { 2497 return (void *) uvm_pageboot_alloc(PAGE_SIZE); 2498 } 2499 return pmap_pool_malloc(pp, flags); 2500 } 2501 2502 void * 2503 pmap_pool_malloc(struct pool *pp, int flags) 2504 { 2505 struct pvo_page *pvop; 2506 struct vm_page *pg; 2507 2508 pvop = SIMPLEQ_FIRST(&pmap_mpvop_head); 2509 if (pvop != NULL) { 2510 pmap_mpvop_free--; 2511 SIMPLEQ_REMOVE_HEAD(&pmap_mpvop_head, pvop_link); 2512 return pvop; 2513 } 2514 again: 2515 pg = uvm_pagealloc_strat(NULL, 0, NULL, UVM_PGA_USERESERVE, 2516 UVM_PGA_STRAT_ONLY, VM_FREELIST_FIRST256); 2517 if (__predict_false(pg == NULL)) { 2518 if (flags & PR_WAITOK) { 2519 uvm_wait("plpg"); 2520 goto again; 2521 } else { 2522 return (0); 2523 } 2524 } 2525 return (void *) VM_PAGE_TO_PHYS(pg); 2526 } 2527 2528 void 2529 pmap_pool_ufree(struct pool *pp, void *va) 2530 { 2531 struct pvo_page *pvop; 2532 #if 0 2533 if (PHYS_TO_VM_PAGE((paddr_t) va) != NULL) { 2534 pmap_pool_mfree(va, size, tag); 2535 return; 2536 } 2537 #endif 2538 pvop = va; 2539 SIMPLEQ_INSERT_HEAD(&pmap_upvop_head, pvop, pvop_link); 2540 pmap_upvop_free++; 2541 if (pmap_upvop_free > pmap_upvop_maxfree) 2542 pmap_upvop_maxfree = pmap_upvop_free; 2543 } 2544 2545 void 2546 pmap_pool_mfree(struct pool *pp, void *va) 2547 { 2548 struct pvo_page *pvop; 2549 2550 pvop = va; 2551 SIMPLEQ_INSERT_HEAD(&pmap_mpvop_head, pvop, pvop_link); 2552 pmap_mpvop_free++; 2553 if (pmap_mpvop_free > pmap_mpvop_maxfree) 2554 pmap_mpvop_maxfree = pmap_mpvop_free; 2555 #if 0 2556 uvm_pagefree(PHYS_TO_VM_PAGE((paddr_t) va)); 2557 #endif 2558 } 2559 2560 /* 2561 * This routine in bootstraping to steal to-be-managed memory (which will 2562 * then be unmanaged). We use it to grab from the first 256MB for our 2563 * pmap needs and above 256MB for other stuff. 2564 */ 2565 vaddr_t 2566 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp) 2567 { 2568 vsize_t size; 2569 vaddr_t va; 2570 paddr_t pa = 0; 2571 int npgs, bank; 2572 struct vm_physseg *ps; 2573 2574 if (uvm.page_init_done == TRUE) 2575 panic("pmap_steal_memory: called _after_ bootstrap"); 2576 2577 *vstartp = VM_MIN_KERNEL_ADDRESS; 2578 *vendp = VM_MAX_KERNEL_ADDRESS; 2579 2580 size = round_page(vsize); 2581 npgs = atop(size); 2582 2583 /* 2584 * PA 0 will never be among those given to UVM so we can use it 2585 * to indicate we couldn't steal any memory. 2586 */ 2587 for (ps = vm_physmem, bank = 0; bank < vm_nphysseg; bank++, ps++) { 2588 if (ps->free_list == VM_FREELIST_FIRST256 && 2589 ps->avail_end - ps->avail_start >= npgs) { 2590 pa = ptoa(ps->avail_start); 2591 break; 2592 } 2593 } 2594 2595 if (pa == 0) 2596 panic("pmap_steal_memory: no approriate memory to steal!"); 2597 2598 ps->avail_start += npgs; 2599 ps->start += npgs; 2600 2601 /* 2602 * If we've used up all the pages in the segment, remove it and 2603 * compact the list. 2604 */ 2605 if (ps->avail_start == ps->end) { 2606 /* 2607 * If this was the last one, then a very bad thing has occurred 2608 */ 2609 if (--vm_nphysseg == 0) 2610 panic("pmap_steal_memory: out of memory!"); 2611 2612 printf("pmap_steal_memory: consumed bank %d\n", bank); 2613 for (; bank < vm_nphysseg; bank++, ps++) { 2614 ps[0] = ps[1]; 2615 } 2616 } 2617 2618 va = (vaddr_t) pa; 2619 memset((caddr_t) va, 0, size); 2620 pmap_pages_stolen += npgs; 2621 #ifdef DEBUG 2622 if (pmapdebug && npgs > 1) { 2623 u_int cnt = 0; 2624 for (bank = 0, ps = vm_physmem; bank < vm_nphysseg; bank++, ps++) 2625 cnt += ps->avail_end - ps->avail_start; 2626 printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n", 2627 npgs, pmap_pages_stolen, cnt); 2628 } 2629 #endif 2630 2631 return va; 2632 } 2633 2634 /* 2635 * Find a chuck of memory with right size and alignment. 2636 */ 2637 void * 2638 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end) 2639 { 2640 struct mem_region *mp; 2641 paddr_t s, e; 2642 int i, j; 2643 2644 size = round_page(size); 2645 2646 DPRINTFN(BOOT, 2647 ("pmap_boot_find_memory: size=%lx, alignment=%lx, at_end=%d", 2648 size, alignment, at_end)); 2649 2650 if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0) 2651 panic("pmap_boot_find_memory: invalid alignment %lx", 2652 alignment); 2653 2654 if (at_end) { 2655 if (alignment != PAGE_SIZE) 2656 panic("pmap_boot_find_memory: invalid ending " 2657 "alignment %lx", alignment); 2658 2659 for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) { 2660 s = mp->start + mp->size - size; 2661 if (s >= mp->start && mp->size >= size) { 2662 DPRINTFN(BOOT,(": %lx\n", s)); 2663 DPRINTFN(BOOT, 2664 ("pmap_boot_find_memory: b-avail[%d] start " 2665 "0x%lx size 0x%lx\n", mp - avail, 2666 mp->start, mp->size)); 2667 mp->size -= size; 2668 DPRINTFN(BOOT, 2669 ("pmap_boot_find_memory: a-avail[%d] start " 2670 "0x%lx size 0x%lx\n", mp - avail, 2671 mp->start, mp->size)); 2672 return (void *) s; 2673 } 2674 } 2675 panic("pmap_boot_find_memory: no available memory"); 2676 } 2677 2678 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2679 s = (mp->start + alignment - 1) & ~(alignment-1); 2680 e = s + size; 2681 2682 /* 2683 * Is the calculated region entirely within the region? 2684 */ 2685 if (s < mp->start || e > mp->start + mp->size) 2686 continue; 2687 2688 DPRINTFN(BOOT,(": %lx\n", s)); 2689 if (s == mp->start) { 2690 /* 2691 * If the block starts at the beginning of region, 2692 * adjust the size & start. (the region may now be 2693 * zero in length) 2694 */ 2695 DPRINTFN(BOOT, 2696 ("pmap_boot_find_memory: b-avail[%d] start " 2697 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2698 mp->start += size; 2699 mp->size -= size; 2700 DPRINTFN(BOOT, 2701 ("pmap_boot_find_memory: a-avail[%d] start " 2702 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2703 } else if (e == mp->start + mp->size) { 2704 /* 2705 * If the block starts at the beginning of region, 2706 * adjust only the size. 2707 */ 2708 DPRINTFN(BOOT, 2709 ("pmap_boot_find_memory: b-avail[%d] start " 2710 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2711 mp->size -= size; 2712 DPRINTFN(BOOT, 2713 ("pmap_boot_find_memory: a-avail[%d] start " 2714 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2715 } else { 2716 /* 2717 * Block is in the middle of the region, so we 2718 * have to split it in two. 2719 */ 2720 for (j = avail_cnt; j > i + 1; j--) { 2721 avail[j] = avail[j-1]; 2722 } 2723 DPRINTFN(BOOT, 2724 ("pmap_boot_find_memory: b-avail[%d] start " 2725 "0x%lx size 0x%lx\n", i, mp->start, mp->size)); 2726 mp[1].start = e; 2727 mp[1].size = mp[0].start + mp[0].size - e; 2728 mp[0].size = s - mp[0].start; 2729 avail_cnt++; 2730 for (; i < avail_cnt; i++) { 2731 DPRINTFN(BOOT, 2732 ("pmap_boot_find_memory: a-avail[%d] " 2733 "start 0x%lx size 0x%lx\n", i, 2734 avail[i].start, avail[i].size)); 2735 } 2736 } 2737 return (void *) s; 2738 } 2739 panic("pmap_boot_find_memory: not enough memory for " 2740 "%lx/%lx allocation?", size, alignment); 2741 } 2742 2743 /* 2744 * This is not part of the defined PMAP interface and is specific to the 2745 * PowerPC architecture. This is called during initppc, before the system 2746 * is really initialized. 2747 */ 2748 void 2749 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend) 2750 { 2751 struct mem_region *mp, tmp; 2752 paddr_t s, e; 2753 psize_t size; 2754 int i, j; 2755 2756 /* 2757 * Get memory. 2758 */ 2759 mem_regions(&mem, &avail); 2760 #if defined(DEBUG) 2761 if (pmapdebug & PMAPDEBUG_BOOT) { 2762 printf("pmap_bootstrap: memory configuration:\n"); 2763 for (mp = mem; mp->size; mp++) { 2764 printf("pmap_bootstrap: mem start 0x%lx size 0x%lx\n", 2765 mp->start, mp->size); 2766 } 2767 for (mp = avail; mp->size; mp++) { 2768 printf("pmap_bootstrap: avail start 0x%lx size 0x%lx\n", 2769 mp->start, mp->size); 2770 } 2771 } 2772 #endif 2773 2774 /* 2775 * Find out how much physical memory we have and in how many chunks. 2776 */ 2777 for (mem_cnt = 0, mp = mem; mp->size; mp++) { 2778 if (mp->start >= pmap_memlimit) 2779 continue; 2780 if (mp->start + mp->size > pmap_memlimit) { 2781 size = pmap_memlimit - mp->start; 2782 physmem += btoc(size); 2783 } else { 2784 physmem += btoc(mp->size); 2785 } 2786 mem_cnt++; 2787 } 2788 2789 /* 2790 * Count the number of available entries. 2791 */ 2792 for (avail_cnt = 0, mp = avail; mp->size; mp++) 2793 avail_cnt++; 2794 2795 /* 2796 * Page align all regions. 2797 */ 2798 kernelstart = trunc_page(kernelstart); 2799 kernelend = round_page(kernelend); 2800 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2801 s = round_page(mp->start); 2802 mp->size -= (s - mp->start); 2803 mp->size = trunc_page(mp->size); 2804 mp->start = s; 2805 e = mp->start + mp->size; 2806 2807 DPRINTFN(BOOT, 2808 ("pmap_bootstrap: b-avail[%d] start 0x%lx size 0x%lx\n", 2809 i, mp->start, mp->size)); 2810 2811 /* 2812 * Don't allow the end to run beyond our artificial limit 2813 */ 2814 if (e > pmap_memlimit) 2815 e = pmap_memlimit; 2816 2817 /* 2818 * Is this region empty or strange? skip it. 2819 */ 2820 if (e <= s) { 2821 mp->start = 0; 2822 mp->size = 0; 2823 continue; 2824 } 2825 2826 /* 2827 * Does this overlap the beginning of kernel? 2828 * Does extend past the end of the kernel? 2829 */ 2830 else if (s < kernelstart && e > kernelstart) { 2831 if (e > kernelend) { 2832 avail[avail_cnt].start = kernelend; 2833 avail[avail_cnt].size = e - kernelend; 2834 avail_cnt++; 2835 } 2836 mp->size = kernelstart - s; 2837 } 2838 /* 2839 * Check whether this region overlaps the end of the kernel. 2840 */ 2841 else if (s < kernelend && e > kernelend) { 2842 mp->start = kernelend; 2843 mp->size = e - kernelend; 2844 } 2845 /* 2846 * Look whether this regions is completely inside the kernel. 2847 * Nuke it if it does. 2848 */ 2849 else if (s >= kernelstart && e <= kernelend) { 2850 mp->start = 0; 2851 mp->size = 0; 2852 } 2853 /* 2854 * If the user imposed a memory limit, enforce it. 2855 */ 2856 else if (s >= pmap_memlimit) { 2857 mp->start = -PAGE_SIZE; /* let's know why */ 2858 mp->size = 0; 2859 } 2860 else { 2861 mp->start = s; 2862 mp->size = e - s; 2863 } 2864 DPRINTFN(BOOT, 2865 ("pmap_bootstrap: a-avail[%d] start 0x%lx size 0x%lx\n", 2866 i, mp->start, mp->size)); 2867 } 2868 2869 /* 2870 * Move (and uncount) all the null return to the end. 2871 */ 2872 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) { 2873 if (mp->size == 0) { 2874 tmp = avail[i]; 2875 avail[i] = avail[--avail_cnt]; 2876 avail[avail_cnt] = avail[i]; 2877 } 2878 } 2879 2880 /* 2881 * (Bubble)sort them into asecnding order. 2882 */ 2883 for (i = 0; i < avail_cnt; i++) { 2884 for (j = i + 1; j < avail_cnt; j++) { 2885 if (avail[i].start > avail[j].start) { 2886 tmp = avail[i]; 2887 avail[i] = avail[j]; 2888 avail[j] = tmp; 2889 } 2890 } 2891 } 2892 2893 /* 2894 * Make sure they don't overlap. 2895 */ 2896 for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) { 2897 if (mp[0].start + mp[0].size > mp[1].start) { 2898 mp[0].size = mp[1].start - mp[0].start; 2899 } 2900 DPRINTFN(BOOT, 2901 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 2902 i, mp->start, mp->size)); 2903 } 2904 DPRINTFN(BOOT, 2905 ("pmap_bootstrap: avail[%d] start 0x%lx size 0x%lx\n", 2906 i, mp->start, mp->size)); 2907 2908 #ifdef PTEGCOUNT 2909 pmap_pteg_cnt = PTEGCOUNT; 2910 #else /* PTEGCOUNT */ 2911 pmap_pteg_cnt = 0x1000; 2912 2913 while (pmap_pteg_cnt < physmem) 2914 pmap_pteg_cnt <<= 1; 2915 2916 pmap_pteg_cnt >>= 1; 2917 #endif /* PTEGCOUNT */ 2918 2919 /* 2920 * Find suitably aligned memory for PTEG hash table. 2921 */ 2922 size = pmap_pteg_cnt * sizeof(struct pteg); 2923 pmap_pteg_table = pmap_boot_find_memory(size, size, 0); 2924 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 2925 if ( (uintptr_t) pmap_pteg_table + size > SEGMENT_LENGTH) 2926 panic("pmap_bootstrap: pmap_pteg_table end (%p + %lx) > 256MB", 2927 pmap_pteg_table, size); 2928 #endif 2929 2930 memset((void *)pmap_pteg_table, 0, pmap_pteg_cnt * sizeof(struct pteg)); 2931 pmap_pteg_mask = pmap_pteg_cnt - 1; 2932 2933 /* 2934 * We cannot do pmap_steal_memory here since UVM hasn't been loaded 2935 * with pages. So we just steal them before giving them to UVM. 2936 */ 2937 size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt; 2938 pmap_pvo_table = pmap_boot_find_memory(size, PAGE_SIZE, 0); 2939 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 2940 if ( (uintptr_t) pmap_pvo_table + size > SEGMENT_LENGTH) 2941 panic("pmap_bootstrap: pmap_pvo_table end (%p + %lx) > 256MB", 2942 pmap_pvo_table, size); 2943 #endif 2944 2945 for (i = 0; i < pmap_pteg_cnt; i++) 2946 TAILQ_INIT(&pmap_pvo_table[i]); 2947 2948 #ifndef MSGBUFADDR 2949 /* 2950 * Allocate msgbuf in high memory. 2951 */ 2952 msgbuf_paddr = 2953 (paddr_t) pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1); 2954 #endif 2955 2956 #ifdef __HAVE_PMAP_PHYSSEG 2957 { 2958 u_int npgs = 0; 2959 for (i = 0, mp = avail; i < avail_cnt; i++, mp++) 2960 npgs += btoc(mp->size); 2961 size = (sizeof(struct pvo_head) + 1) * npgs; 2962 pmap_physseg.pvoh = pmap_boot_find_memory(size, PAGE_SIZE, 0); 2963 pmap_physseg.attrs = (char *) &pmap_physseg.pvoh[npgs]; 2964 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK) 2965 if ((uintptr_t)pmap_physseg.pvoh + size > SEGMENT_LENGTH) 2966 panic("pmap_bootstrap: PVO list end (%p + %lx) > 256MB", 2967 pmap_physseg.pvoh, size); 2968 #endif 2969 } 2970 #endif 2971 2972 for (mp = avail, i = 0; i < avail_cnt; mp++, i++) { 2973 paddr_t pfstart = atop(mp->start); 2974 paddr_t pfend = atop(mp->start + mp->size); 2975 if (mp->size == 0) 2976 continue; 2977 if (mp->start + mp->size <= SEGMENT_LENGTH) { 2978 uvm_page_physload(pfstart, pfend, pfstart, pfend, 2979 VM_FREELIST_FIRST256); 2980 } else if (mp->start >= SEGMENT_LENGTH) { 2981 uvm_page_physload(pfstart, pfend, pfstart, pfend, 2982 VM_FREELIST_DEFAULT); 2983 } else { 2984 pfend = atop(SEGMENT_LENGTH); 2985 uvm_page_physload(pfstart, pfend, pfstart, pfend, 2986 VM_FREELIST_FIRST256); 2987 pfstart = atop(SEGMENT_LENGTH); 2988 pfend = atop(mp->start + mp->size); 2989 uvm_page_physload(pfstart, pfend, pfstart, pfend, 2990 VM_FREELIST_DEFAULT); 2991 } 2992 } 2993 2994 /* 2995 * Make sure kernel vsid is allocated as well as VSID 0. 2996 */ 2997 pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW] 2998 |= 1 << (KERNEL_VSIDBITS % VSID_NBPW); 2999 pmap_vsid_bitmap[0] |= 1; 3000 3001 /* 3002 * Initialize kernel pmap and hardware. 3003 */ 3004 for (i = 0; i < 16; i++) { 3005 pmap_kernel()->pm_sr[i] = EMPTY_SEGMENT; 3006 __asm __volatile ("mtsrin %0,%1" 3007 :: "r"(EMPTY_SEGMENT), "r"(i << ADDR_SR_SHFT)); 3008 } 3009 3010 pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY; 3011 __asm __volatile ("mtsr %0,%1" 3012 :: "n"(KERNEL_SR), "r"(KERNEL_SEGMENT)); 3013 #ifdef KERNEL2_SR 3014 pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY; 3015 __asm __volatile ("mtsr %0,%1" 3016 :: "n"(KERNEL2_SR), "r"(KERNEL2_SEGMENT)); 3017 #endif 3018 for (i = 0; i < 16; i++) { 3019 if (iosrtable[i] & SR601_T) { 3020 pmap_kernel()->pm_sr[i] = iosrtable[i]; 3021 __asm __volatile ("mtsrin %0,%1" 3022 :: "r"(iosrtable[i]), "r"(i << ADDR_SR_SHFT)); 3023 } 3024 } 3025 3026 __asm __volatile ("sync; mtsdr1 %0; isync" 3027 :: "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10))); 3028 tlbia(); 3029 3030 #ifdef ALTIVEC 3031 pmap_use_altivec = cpu_altivec; 3032 #endif 3033 3034 #ifdef DEBUG 3035 if (pmapdebug & PMAPDEBUG_BOOT) { 3036 u_int cnt; 3037 int bank; 3038 char pbuf[9]; 3039 for (cnt = 0, bank = 0; bank < vm_nphysseg; bank++) { 3040 cnt += vm_physmem[bank].avail_end - vm_physmem[bank].avail_start; 3041 printf("pmap_bootstrap: vm_physmem[%d]=%#lx-%#lx/%#lx\n", 3042 bank, 3043 ptoa(vm_physmem[bank].avail_start), 3044 ptoa(vm_physmem[bank].avail_end), 3045 ptoa(vm_physmem[bank].avail_end - vm_physmem[bank].avail_start)); 3046 } 3047 format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt)); 3048 printf("pmap_bootstrap: UVM memory = %s (%u pages)\n", 3049 pbuf, cnt); 3050 } 3051 #endif 3052 3053 pool_init(&pmap_upvo_pool, sizeof(struct pvo_entry), 3054 sizeof(struct pvo_entry), 0, 0, "pmap_upvopl", 3055 &pmap_pool_uallocator); 3056 3057 pool_setlowat(&pmap_upvo_pool, 252); 3058 3059 pool_init(&pmap_pool, sizeof(struct pmap), 3060 sizeof(void *), 0, 0, "pmap_pl", &pmap_pool_uallocator); 3061 } 3062
Indexes created Thu Sep 25 16:09:42 GMT 2025