1 /*- 2 * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com> 3 * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com> 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN 18 * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 19 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED 20 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 21 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 22 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 23 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 24 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 * 26 * Some hw specific parts of this pmap were derived or influenced 27 * by NetBSD's ibm4xx pmap module. More generic code is shared with 28 * a few other pmap modules from the FreeBSD tree. 29 */ 30 31 /* 32 * VM layout notes: 33 * 34 * Kernel and user threads run within one common virtual address space 35 * defined by AS=0. 36 * 37 * Virtual address space layout: 38 * ----------------------------- 39 * 0x0000_0000 - 0xafff_ffff : user process 40 * 0xb000_0000 - 0xbfff_ffff : pmap_mapdev()-ed area (PCI/PCIE etc.) 41 * 0xc000_0000 - 0xc0ff_ffff : kernel reserved 42 * 0xc000_0000 - data_end : kernel code+data, env, metadata etc. 43 * 0xc100_0000 - 0xfeef_ffff : KVA 44 * 0xc100_0000 - 0xc100_3fff : reserved for page zero/copy 45 * 0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs 46 * 0xc200_4000 - 0xc200_8fff : guard page + kstack0 47 * 0xc200_9000 - 0xfeef_ffff : actual free KVA space 48 * 0xfef0_0000 - 0xffff_ffff : I/O devices region 49 */ 50 51 #include <sys/cdefs.h> 52 __FBSDID("$FreeBSD$"); 53 54 #include <sys/types.h> 55 #include <sys/param.h> 56 #include <sys/malloc.h> 57 #include <sys/ktr.h> 58 #include <sys/proc.h> 59 #include <sys/user.h> 60 #include <sys/queue.h> 61 #include <sys/systm.h> 62 #include <sys/kernel.h> 63 #include <sys/linker.h> 64 #include <sys/msgbuf.h> 65 #include <sys/lock.h> 66 #include <sys/mutex.h> 67 #include <sys/sched.h> 68 #include <sys/smp.h> 69 #include <sys/vmmeter.h> 70 71 #include <vm/vm.h> 72 #include <vm/vm_page.h> 73 #include <vm/vm_kern.h> 74 #include <vm/vm_pageout.h> 75 #include <vm/vm_extern.h> 76 #include <vm/vm_object.h> 77 #include <vm/vm_param.h> 78 #include <vm/vm_map.h> 79 #include <vm/vm_pager.h> 80 #include <vm/uma.h> 81 82 #include <machine/cpu.h> 83 #include <machine/pcb.h> 84 #include <machine/platform.h> 85 86 #include <machine/tlb.h> 87 #include <machine/spr.h> 88 #include <machine/vmparam.h> 89 #include <machine/md_var.h> 90 #include <machine/mmuvar.h> 91 #include <machine/pmap.h> 92 #include <machine/pte.h> 93 94 #include "mmu_if.h" 95 96 #ifdef DEBUG 97 #define debugf(fmt, args...) printf(fmt, ##args) 98 #else 99 #define debugf(fmt, args...) 100 #endif 101 102 #define TODO panic("%s: not implemented", __func__); 103 104 extern struct mtx sched_lock; 105 106 extern int dumpsys_minidump; 107 108 extern unsigned char _etext[]; 109 extern unsigned char _end[]; 110 111 extern uint32_t *bootinfo; 112 113 #ifdef SMP 114 extern uint32_t bp_kernload; 115 #endif 116 117 vm_paddr_t kernload; 118 vm_offset_t kernstart; 119 vm_size_t kernsize; 120 121 /* Message buffer and tables. */ 122 static vm_offset_t data_start; 123 static vm_size_t data_end; 124 125 /* Phys/avail memory regions. */ 126 static struct mem_region *availmem_regions; 127 static int availmem_regions_sz; 128 static struct mem_region *physmem_regions; 129 static int physmem_regions_sz; 130 131 /* Reserved KVA space and mutex for mmu_booke_zero_page. */ 132 static vm_offset_t zero_page_va; 133 static struct mtx zero_page_mutex; 134 135 static struct mtx tlbivax_mutex; 136 137 /* 138 * Reserved KVA space for mmu_booke_zero_page_idle. This is used 139 * by idle thred only, no lock required. 140 */ 141 static vm_offset_t zero_page_idle_va; 142 143 /* Reserved KVA space and mutex for mmu_booke_copy_page. */ 144 static vm_offset_t copy_page_src_va; 145 static vm_offset_t copy_page_dst_va; 146 static struct mtx copy_page_mutex; 147 148 /**************************************************************************/ 149 /* PMAP */ 150 /**************************************************************************/ 151 152 static void mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t, 153 vm_prot_t, boolean_t); 154 155 unsigned int kptbl_min; /* Index of the first kernel ptbl. */ 156 unsigned int kernel_ptbls; /* Number of KVA ptbls. */ 157 158 /* 159 * If user pmap is processed with mmu_booke_remove and the resident count 160 * drops to 0, there are no more pages to remove, so we need not continue. 161 */ 162 #define PMAP_REMOVE_DONE(pmap) \ 163 ((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0) 164 165 extern void tid_flush(tlbtid_t); 166 167 /**************************************************************************/ 168 /* TLB and TID handling */ 169 /**************************************************************************/ 170 171 /* Translation ID busy table */ 172 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1]; 173 174 /* 175 * TLB0 capabilities (entry, way numbers etc.). These can vary between e500 176 * core revisions and should be read from h/w registers during early config. 177 */ 178 uint32_t tlb0_entries; 179 uint32_t tlb0_ways; 180 uint32_t tlb0_entries_per_way; 181 182 #define TLB0_ENTRIES (tlb0_entries) 183 #define TLB0_WAYS (tlb0_ways) 184 #define TLB0_ENTRIES_PER_WAY (tlb0_entries_per_way) 185 186 #define TLB1_ENTRIES 16 187 188 /* In-ram copy of the TLB1 */ 189 static tlb_entry_t tlb1[TLB1_ENTRIES]; 190 191 /* Next free entry in the TLB1 */ 192 static unsigned int tlb1_idx; 193 194 static tlbtid_t tid_alloc(struct pmap *); 195 196 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t); 197 198 static int tlb1_set_entry(vm_offset_t, vm_offset_t, vm_size_t, uint32_t); 199 static void tlb1_write_entry(unsigned int); 200 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *); 201 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t); 202 203 static vm_size_t tsize2size(unsigned int); 204 static unsigned int size2tsize(vm_size_t); 205 static unsigned int ilog2(unsigned int); 206 207 static void set_mas4_defaults(void); 208 209 static inline void tlb0_flush_entry(vm_offset_t); 210 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int); 211 212 /**************************************************************************/ 213 /* Page table management */ 214 /**************************************************************************/ 215 216 /* Data for the pv entry allocation mechanism */ 217 static uma_zone_t pvzone; 218 static struct vm_object pvzone_obj; 219 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0; 220 221 #define PV_ENTRY_ZONE_MIN 2048 /* min pv entries in uma zone */ 222 223 #ifndef PMAP_SHPGPERPROC 224 #define PMAP_SHPGPERPROC 200 225 #endif 226 227 static void ptbl_init(void); 228 static struct ptbl_buf *ptbl_buf_alloc(void); 229 static void ptbl_buf_free(struct ptbl_buf *); 230 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *); 231 232 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int); 233 static void ptbl_free(mmu_t, pmap_t, unsigned int); 234 static void ptbl_hold(mmu_t, pmap_t, unsigned int); 235 static int ptbl_unhold(mmu_t, pmap_t, unsigned int); 236 237 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t); 238 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t); 239 static void pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t); 240 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t); 241 242 static pv_entry_t pv_alloc(void); 243 static void pv_free(pv_entry_t); 244 static void pv_insert(pmap_t, vm_offset_t, vm_page_t); 245 static void pv_remove(pmap_t, vm_offset_t, vm_page_t); 246 247 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */ 248 #define PTBL_BUFS (128 * 16) 249 250 struct ptbl_buf { 251 TAILQ_ENTRY(ptbl_buf) link; /* list link */ 252 vm_offset_t kva; /* va of mapping */ 253 }; 254 255 /* ptbl free list and a lock used for access synchronization. */ 256 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist; 257 static struct mtx ptbl_buf_freelist_lock; 258 259 /* Base address of kva space allocated fot ptbl bufs. */ 260 static vm_offset_t ptbl_buf_pool_vabase; 261 262 /* Pointer to ptbl_buf structures. */ 263 static struct ptbl_buf *ptbl_bufs; 264 265 void pmap_bootstrap_ap(volatile uint32_t *); 266 267 /* 268 * Kernel MMU interface 269 */ 270 static void mmu_booke_change_wiring(mmu_t, pmap_t, vm_offset_t, boolean_t); 271 static void mmu_booke_clear_modify(mmu_t, vm_page_t); 272 static void mmu_booke_clear_reference(mmu_t, vm_page_t); 273 static void mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t, 274 vm_size_t, vm_offset_t); 275 static void mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t); 276 static void mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t, 277 vm_prot_t, boolean_t); 278 static void mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t, 279 vm_page_t, vm_prot_t); 280 static void mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t, 281 vm_prot_t); 282 static vm_paddr_t mmu_booke_extract(mmu_t, pmap_t, vm_offset_t); 283 static vm_page_t mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t, 284 vm_prot_t); 285 static void mmu_booke_init(mmu_t); 286 static boolean_t mmu_booke_is_modified(mmu_t, vm_page_t); 287 static boolean_t mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t); 288 static boolean_t mmu_booke_is_referenced(mmu_t, vm_page_t); 289 static int mmu_booke_ts_referenced(mmu_t, vm_page_t); 290 static vm_offset_t mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t, 291 int); 292 static int mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t, 293 vm_paddr_t *); 294 static void mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t, 295 vm_object_t, vm_pindex_t, vm_size_t); 296 static boolean_t mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t); 297 static void mmu_booke_page_init(mmu_t, vm_page_t); 298 static int mmu_booke_page_wired_mappings(mmu_t, vm_page_t); 299 static void mmu_booke_pinit(mmu_t, pmap_t); 300 static void mmu_booke_pinit0(mmu_t, pmap_t); 301 static void mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t, 302 vm_prot_t); 303 static void mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int); 304 static void mmu_booke_qremove(mmu_t, vm_offset_t, int); 305 static void mmu_booke_release(mmu_t, pmap_t); 306 static void mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t); 307 static void mmu_booke_remove_all(mmu_t, vm_page_t); 308 static void mmu_booke_remove_write(mmu_t, vm_page_t); 309 static void mmu_booke_zero_page(mmu_t, vm_page_t); 310 static void mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int); 311 static void mmu_booke_zero_page_idle(mmu_t, vm_page_t); 312 static void mmu_booke_activate(mmu_t, struct thread *); 313 static void mmu_booke_deactivate(mmu_t, struct thread *); 314 static void mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t); 315 static void *mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t); 316 static void mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t); 317 static vm_paddr_t mmu_booke_kextract(mmu_t, vm_offset_t); 318 static void mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t); 319 static void mmu_booke_kremove(mmu_t, vm_offset_t); 320 static boolean_t mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t); 321 static void mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t, 322 vm_size_t); 323 static vm_offset_t mmu_booke_dumpsys_map(mmu_t, struct pmap_md *, 324 vm_size_t, vm_size_t *); 325 static void mmu_booke_dumpsys_unmap(mmu_t, struct pmap_md *, 326 vm_size_t, vm_offset_t); 327 static struct pmap_md *mmu_booke_scan_md(mmu_t, struct pmap_md *); 328 329 static mmu_method_t mmu_booke_methods[] = { 330 /* pmap dispatcher interface */ 331 MMUMETHOD(mmu_change_wiring, mmu_booke_change_wiring), 332 MMUMETHOD(mmu_clear_modify, mmu_booke_clear_modify), 333 MMUMETHOD(mmu_clear_reference, mmu_booke_clear_reference), 334 MMUMETHOD(mmu_copy, mmu_booke_copy), 335 MMUMETHOD(mmu_copy_page, mmu_booke_copy_page), 336 MMUMETHOD(mmu_enter, mmu_booke_enter), 337 MMUMETHOD(mmu_enter_object, mmu_booke_enter_object), 338 MMUMETHOD(mmu_enter_quick, mmu_booke_enter_quick), 339 MMUMETHOD(mmu_extract, mmu_booke_extract), 340 MMUMETHOD(mmu_extract_and_hold, mmu_booke_extract_and_hold), 341 MMUMETHOD(mmu_init, mmu_booke_init), 342 MMUMETHOD(mmu_is_modified, mmu_booke_is_modified), 343 MMUMETHOD(mmu_is_prefaultable, mmu_booke_is_prefaultable), 344 MMUMETHOD(mmu_is_referenced, mmu_booke_is_referenced), 345 MMUMETHOD(mmu_ts_referenced, mmu_booke_ts_referenced), 346 MMUMETHOD(mmu_map, mmu_booke_map), 347 MMUMETHOD(mmu_mincore, mmu_booke_mincore), 348 MMUMETHOD(mmu_object_init_pt, mmu_booke_object_init_pt), 349 MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick), 350 MMUMETHOD(mmu_page_init, mmu_booke_page_init), 351 MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings), 352 MMUMETHOD(mmu_pinit, mmu_booke_pinit), 353 MMUMETHOD(mmu_pinit0, mmu_booke_pinit0), 354 MMUMETHOD(mmu_protect, mmu_booke_protect), 355 MMUMETHOD(mmu_qenter, mmu_booke_qenter), 356 MMUMETHOD(mmu_qremove, mmu_booke_qremove), 357 MMUMETHOD(mmu_release, mmu_booke_release), 358 MMUMETHOD(mmu_remove, mmu_booke_remove), 359 MMUMETHOD(mmu_remove_all, mmu_booke_remove_all), 360 MMUMETHOD(mmu_remove_write, mmu_booke_remove_write), 361 MMUMETHOD(mmu_sync_icache, mmu_booke_sync_icache), 362 MMUMETHOD(mmu_zero_page, mmu_booke_zero_page), 363 MMUMETHOD(mmu_zero_page_area, mmu_booke_zero_page_area), 364 MMUMETHOD(mmu_zero_page_idle, mmu_booke_zero_page_idle), 365 MMUMETHOD(mmu_activate, mmu_booke_activate), 366 MMUMETHOD(mmu_deactivate, mmu_booke_deactivate), 367 368 /* Internal interfaces */ 369 MMUMETHOD(mmu_bootstrap, mmu_booke_bootstrap), 370 MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped), 371 MMUMETHOD(mmu_mapdev, mmu_booke_mapdev), 372 MMUMETHOD(mmu_kenter, mmu_booke_kenter), 373 MMUMETHOD(mmu_kextract, mmu_booke_kextract), 374 /* MMUMETHOD(mmu_kremove, mmu_booke_kremove), */ 375 MMUMETHOD(mmu_unmapdev, mmu_booke_unmapdev), 376 377 /* dumpsys() support */ 378 MMUMETHOD(mmu_dumpsys_map, mmu_booke_dumpsys_map), 379 MMUMETHOD(mmu_dumpsys_unmap, mmu_booke_dumpsys_unmap), 380 MMUMETHOD(mmu_scan_md, mmu_booke_scan_md), 381 382 { 0, 0 } 383 }; 384 385 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0); 386 387 static inline void 388 tlb_miss_lock(void) 389 { 390 #ifdef SMP 391 struct pcpu *pc; 392 393 if (!smp_started) 394 return; 395 396 STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) { 397 if (pc != pcpup) { 398 399 CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, " 400 "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock); 401 402 KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)), 403 ("tlb_miss_lock: tried to lock self")); 404 405 tlb_lock(pc->pc_booke_tlb_lock); 406 407 CTR1(KTR_PMAP, "%s: locked", __func__); 408 } 409 } 410 #endif 411 } 412 413 static inline void 414 tlb_miss_unlock(void) 415 { 416 #ifdef SMP 417 struct pcpu *pc; 418 419 if (!smp_started) 420 return; 421 422 STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) { 423 if (pc != pcpup) { 424 CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d", 425 __func__, pc->pc_cpuid); 426 427 tlb_unlock(pc->pc_booke_tlb_lock); 428 429 CTR1(KTR_PMAP, "%s: unlocked", __func__); 430 } 431 } 432 #endif 433 } 434 435 /* Return number of entries in TLB0. */ 436 static __inline void 437 tlb0_get_tlbconf(void) 438 { 439 uint32_t tlb0_cfg; 440 441 tlb0_cfg = mfspr(SPR_TLB0CFG); 442 tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK; 443 tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT; 444 tlb0_entries_per_way = tlb0_entries / tlb0_ways; 445 } 446 447 /* Initialize pool of kva ptbl buffers. */ 448 static void 449 ptbl_init(void) 450 { 451 int i; 452 453 CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__, 454 (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS); 455 CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)", 456 __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE); 457 458 mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF); 459 TAILQ_INIT(&ptbl_buf_freelist); 460 461 for (i = 0; i < PTBL_BUFS; i++) { 462 ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE; 463 TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link); 464 } 465 } 466 467 /* Get a ptbl_buf from the freelist. */ 468 static struct ptbl_buf * 469 ptbl_buf_alloc(void) 470 { 471 struct ptbl_buf *buf; 472 473 mtx_lock(&ptbl_buf_freelist_lock); 474 buf = TAILQ_FIRST(&ptbl_buf_freelist); 475 if (buf != NULL) 476 TAILQ_REMOVE(&ptbl_buf_freelist, buf, link); 477 mtx_unlock(&ptbl_buf_freelist_lock); 478 479 CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf); 480 481 return (buf); 482 } 483 484 /* Return ptbl buff to free pool. */ 485 static void 486 ptbl_buf_free(struct ptbl_buf *buf) 487 { 488 489 CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf); 490 491 mtx_lock(&ptbl_buf_freelist_lock); 492 TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link); 493 mtx_unlock(&ptbl_buf_freelist_lock); 494 } 495 496 /* 497 * Search the list of allocated ptbl bufs and find on list of allocated ptbls 498 */ 499 static void 500 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl) 501 { 502 struct ptbl_buf *pbuf; 503 504 CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl); 505 506 PMAP_LOCK_ASSERT(pmap, MA_OWNED); 507 508 TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link) 509 if (pbuf->kva == (vm_offset_t)ptbl) { 510 /* Remove from pmap ptbl buf list. */ 511 TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link); 512 513 /* Free corresponding ptbl buf. */ 514 ptbl_buf_free(pbuf); 515 break; 516 } 517 } 518 519 /* Allocate page table. */ 520 static pte_t * 521 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx) 522 { 523 vm_page_t mtbl[PTBL_PAGES]; 524 vm_page_t m; 525 struct ptbl_buf *pbuf; 526 unsigned int pidx; 527 pte_t *ptbl; 528 int i; 529 530 CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap, 531 (pmap == kernel_pmap), pdir_idx); 532 533 KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)), 534 ("ptbl_alloc: invalid pdir_idx")); 535 KASSERT((pmap->pm_pdir[pdir_idx] == NULL), 536 ("pte_alloc: valid ptbl entry exists!")); 537 538 pbuf = ptbl_buf_alloc(); 539 if (pbuf == NULL) 540 panic("pte_alloc: couldn't alloc kernel virtual memory"); 541 542 ptbl = (pte_t *)pbuf->kva; 543 544 CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl); 545 546 /* Allocate ptbl pages, this will sleep! */ 547 for (i = 0; i < PTBL_PAGES; i++) { 548 pidx = (PTBL_PAGES * pdir_idx) + i; 549 while ((m = vm_page_alloc(NULL, pidx, 550 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) { 551 552 PMAP_UNLOCK(pmap); 553 vm_page_unlock_queues(); 554 VM_WAIT; 555 vm_page_lock_queues(); 556 PMAP_LOCK(pmap); 557 } 558 mtbl[i] = m; 559 } 560 561 /* Map allocated pages into kernel_pmap. */ 562 mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES); 563 564 /* Zero whole ptbl. */ 565 bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE); 566 567 /* Add pbuf to the pmap ptbl bufs list. */ 568 TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link); 569 570 return (ptbl); 571 } 572 573 /* Free ptbl pages and invalidate pdir entry. */ 574 static void 575 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx) 576 { 577 pte_t *ptbl; 578 vm_paddr_t pa; 579 vm_offset_t va; 580 vm_page_t m; 581 int i; 582 583 CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap, 584 (pmap == kernel_pmap), pdir_idx); 585 586 KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)), 587 ("ptbl_free: invalid pdir_idx")); 588 589 ptbl = pmap->pm_pdir[pdir_idx]; 590 591 CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl); 592 593 KASSERT((ptbl != NULL), ("ptbl_free: null ptbl")); 594 595 /* 596 * Invalidate the pdir entry as soon as possible, so that other CPUs 597 * don't attempt to look up the page tables we are releasing. 598 */ 599 mtx_lock_spin(&tlbivax_mutex); 600 tlb_miss_lock(); 601 602 pmap->pm_pdir[pdir_idx] = NULL; 603 604 tlb_miss_unlock(); 605 mtx_unlock_spin(&tlbivax_mutex); 606 607 for (i = 0; i < PTBL_PAGES; i++) { 608 va = ((vm_offset_t)ptbl + (i * PAGE_SIZE)); 609 pa = pte_vatopa(mmu, kernel_pmap, va); 610 m = PHYS_TO_VM_PAGE(pa); 611 vm_page_free_zero(m); 612 atomic_subtract_int(&cnt.v_wire_count, 1); 613 mmu_booke_kremove(mmu, va); 614 } 615 616 ptbl_free_pmap_ptbl(pmap, ptbl); 617 } 618 619 /* 620 * Decrement ptbl pages hold count and attempt to free ptbl pages. 621 * Called when removing pte entry from ptbl. 622 * 623 * Return 1 if ptbl pages were freed. 624 */ 625 static int 626 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx) 627 { 628 pte_t *ptbl; 629 vm_paddr_t pa; 630 vm_page_t m; 631 int i; 632 633 CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap, 634 (pmap == kernel_pmap), pdir_idx); 635 636 KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)), 637 ("ptbl_unhold: invalid pdir_idx")); 638 KASSERT((pmap != kernel_pmap), 639 ("ptbl_unhold: unholding kernel ptbl!")); 640 641 ptbl = pmap->pm_pdir[pdir_idx]; 642 643 //debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl); 644 KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS), 645 ("ptbl_unhold: non kva ptbl")); 646 647 /* decrement hold count */ 648 for (i = 0; i < PTBL_PAGES; i++) { 649 pa = pte_vatopa(mmu, kernel_pmap, 650 (vm_offset_t)ptbl + (i * PAGE_SIZE)); 651 m = PHYS_TO_VM_PAGE(pa); 652 m->wire_count--; 653 } 654 655 /* 656 * Free ptbl pages if there are no pte etries in this ptbl. 657 * wire_count has the same value for all ptbl pages, so check the last 658 * page. 659 */ 660 if (m->wire_count == 0) { 661 ptbl_free(mmu, pmap, pdir_idx); 662 663 //debugf("ptbl_unhold: e (freed ptbl)\n"); 664 return (1); 665 } 666 667 return (0); 668 } 669 670 /* 671 * Increment hold count for ptbl pages. This routine is used when a new pte 672 * entry is being inserted into the ptbl. 673 */ 674 static void 675 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx) 676 { 677 vm_paddr_t pa; 678 pte_t *ptbl; 679 vm_page_t m; 680 int i; 681 682 CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap, 683 pdir_idx); 684 685 KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)), 686 ("ptbl_hold: invalid pdir_idx")); 687 KASSERT((pmap != kernel_pmap), 688 ("ptbl_hold: holding kernel ptbl!")); 689 690 ptbl = pmap->pm_pdir[pdir_idx]; 691 692 KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl")); 693 694 for (i = 0; i < PTBL_PAGES; i++) { 695 pa = pte_vatopa(mmu, kernel_pmap, 696 (vm_offset_t)ptbl + (i * PAGE_SIZE)); 697 m = PHYS_TO_VM_PAGE(pa); 698 m->wire_count++; 699 } 700 } 701 702 /* Allocate pv_entry structure. */ 703 pv_entry_t 704 pv_alloc(void) 705 { 706 pv_entry_t pv; 707 708 pv_entry_count++; 709 if (pv_entry_count > pv_entry_high_water) 710 pagedaemon_wakeup(); 711 pv = uma_zalloc(pvzone, M_NOWAIT); 712 713 return (pv); 714 } 715 716 /* Free pv_entry structure. */ 717 static __inline void 718 pv_free(pv_entry_t pve) 719 { 720 721 pv_entry_count--; 722 uma_zfree(pvzone, pve); 723 } 724 725 726 /* Allocate and initialize pv_entry structure. */ 727 static void 728 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m) 729 { 730 pv_entry_t pve; 731 732 //int su = (pmap == kernel_pmap); 733 //debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su, 734 // (u_int32_t)pmap, va, (u_int32_t)m); 735 736 pve = pv_alloc(); 737 if (pve == NULL) 738 panic("pv_insert: no pv entries!"); 739 740 pve->pv_pmap = pmap; 741 pve->pv_va = va; 742 743 /* add to pv_list */ 744 PMAP_LOCK_ASSERT(pmap, MA_OWNED); 745 mtx_assert(&vm_page_queue_mtx, MA_OWNED); 746 747 TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link); 748 749 //debugf("pv_insert: e\n"); 750 } 751 752 /* Destroy pv entry. */ 753 static void 754 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m) 755 { 756 pv_entry_t pve; 757 758 //int su = (pmap == kernel_pmap); 759 //debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va); 760 761 PMAP_LOCK_ASSERT(pmap, MA_OWNED); 762 mtx_assert(&vm_page_queue_mtx, MA_OWNED); 763 764 /* find pv entry */ 765 TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) { 766 if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) { 767 /* remove from pv_list */ 768 TAILQ_REMOVE(&m->md.pv_list, pve, pv_link); 769 if (TAILQ_EMPTY(&m->md.pv_list)) 770 vm_page_aflag_clear(m, PGA_WRITEABLE); 771 772 /* free pv entry struct */ 773 pv_free(pve); 774 break; 775 } 776 } 777 778 //debugf("pv_remove: e\n"); 779 } 780 781 /* 782 * Clean pte entry, try to free page table page if requested. 783 * 784 * Return 1 if ptbl pages were freed, otherwise return 0. 785 */ 786 static int 787 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags) 788 { 789 unsigned int pdir_idx = PDIR_IDX(va); 790 unsigned int ptbl_idx = PTBL_IDX(va); 791 vm_page_t m; 792 pte_t *ptbl; 793 pte_t *pte; 794 795 //int su = (pmap == kernel_pmap); 796 //debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n", 797 // su, (u_int32_t)pmap, va, flags); 798 799 ptbl = pmap->pm_pdir[pdir_idx]; 800 KASSERT(ptbl, ("pte_remove: null ptbl")); 801 802 pte = &ptbl[ptbl_idx]; 803 804 if (pte == NULL || !PTE_ISVALID(pte)) 805 return (0); 806 807 if (PTE_ISWIRED(pte)) 808 pmap->pm_stats.wired_count--; 809 810 /* Handle managed entry. */ 811 if (PTE_ISMANAGED(pte)) { 812 /* Get vm_page_t for mapped pte. */ 813 m = PHYS_TO_VM_PAGE(PTE_PA(pte)); 814 815 if (PTE_ISMODIFIED(pte)) 816 vm_page_dirty(m); 817 818 if (PTE_ISREFERENCED(pte)) 819 vm_page_aflag_set(m, PGA_REFERENCED); 820 821 pv_remove(pmap, va, m); 822 } 823 824 mtx_lock_spin(&tlbivax_mutex); 825 tlb_miss_lock(); 826 827 tlb0_flush_entry(va); 828 pte->flags = 0; 829 pte->rpn = 0; 830 831 tlb_miss_unlock(); 832 mtx_unlock_spin(&tlbivax_mutex); 833 834 pmap->pm_stats.resident_count--; 835 836 if (flags & PTBL_UNHOLD) { 837 //debugf("pte_remove: e (unhold)\n"); 838 return (ptbl_unhold(mmu, pmap, pdir_idx)); 839 } 840 841 //debugf("pte_remove: e\n"); 842 return (0); 843 } 844 845 /* 846 * Insert PTE for a given page and virtual address. 847 */ 848 static void 849 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags) 850 { 851 unsigned int pdir_idx = PDIR_IDX(va); 852 unsigned int ptbl_idx = PTBL_IDX(va); 853 pte_t *ptbl, *pte; 854 855 CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__, 856 pmap == kernel_pmap, pmap, va); 857 858 /* Get the page table pointer. */ 859 ptbl = pmap->pm_pdir[pdir_idx]; 860 861 if (ptbl == NULL) { 862 /* Allocate page table pages. */ 863 ptbl = ptbl_alloc(mmu, pmap, pdir_idx); 864 } else { 865 /* 866 * Check if there is valid mapping for requested 867 * va, if there is, remove it. 868 */ 869 pte = &pmap->pm_pdir[pdir_idx][ptbl_idx]; 870 if (PTE_ISVALID(pte)) { 871 pte_remove(mmu, pmap, va, PTBL_HOLD); 872 } else { 873 /* 874 * pte is not used, increment hold count 875 * for ptbl pages. 876 */ 877 if (pmap != kernel_pmap) 878 ptbl_hold(mmu, pmap, pdir_idx); 879 } 880 } 881 882 /* 883 * Insert pv_entry into pv_list for mapped page if part of managed 884 * memory. 885 */ 886 if ((m->oflags & VPO_UNMANAGED) == 0) { 887 flags |= PTE_MANAGED; 888 889 /* Create and insert pv entry. */ 890 pv_insert(pmap, va, m); 891 } 892 893 pmap->pm_stats.resident_count++; 894 895 mtx_lock_spin(&tlbivax_mutex); 896 tlb_miss_lock(); 897 898 tlb0_flush_entry(va); 899 if (pmap->pm_pdir[pdir_idx] == NULL) { 900 /* 901 * If we just allocated a new page table, hook it in 902 * the pdir. 903 */ 904 pmap->pm_pdir[pdir_idx] = ptbl; 905 } 906 pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]); 907 pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK; 908 pte->flags |= (PTE_VALID | flags); 909 910 tlb_miss_unlock(); 911 mtx_unlock_spin(&tlbivax_mutex); 912 } 913 914 /* Return the pa for the given pmap/va. */ 915 static vm_paddr_t 916 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va) 917 { 918 vm_paddr_t pa = 0; 919 pte_t *pte; 920 921 pte = pte_find(mmu, pmap, va); 922 if ((pte != NULL) && PTE_ISVALID(pte)) 923 pa = (PTE_PA(pte) | (va & PTE_PA_MASK)); 924 return (pa); 925 } 926 927 /* Get a pointer to a PTE in a page table. */ 928 static pte_t * 929 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va) 930 { 931 unsigned int pdir_idx = PDIR_IDX(va); 932 unsigned int ptbl_idx = PTBL_IDX(va); 933 934 KASSERT((pmap != NULL), ("pte_find: invalid pmap")); 935 936 if (pmap->pm_pdir[pdir_idx]) 937 return (&(pmap->pm_pdir[pdir_idx][ptbl_idx])); 938 939 return (NULL); 940 } 941 942 /**************************************************************************/ 943 /* PMAP related */ 944 /**************************************************************************/ 945 946 /* 947 * This is called during booke_init, before the system is really initialized. 948 */ 949 static void 950 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend) 951 { 952 vm_offset_t phys_kernelend; 953 struct mem_region *mp, *mp1; 954 int cnt, i, j; 955 u_int s, e, sz; 956 u_int phys_avail_count; 957 vm_size_t physsz, hwphyssz, kstack0_sz; 958 vm_offset_t kernel_pdir, kstack0, va; 959 vm_paddr_t kstack0_phys; 960 void *dpcpu; 961 pte_t *pte; 962 963 debugf("mmu_booke_bootstrap: entered\n"); 964 965 #ifdef SMP 966 bp_kernload = kernload; 967 #endif 968 969 /* Initialize invalidation mutex */ 970 mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN); 971 972 /* Read TLB0 size and associativity. */ 973 tlb0_get_tlbconf(); 974 975 /* 976 * Align kernel start and end address (kernel image). 977 * Note that kernel end does not necessarily relate to kernsize. 978 * kernsize is the size of the kernel that is actually mapped. 979 * Also note that "start - 1" is deliberate. With SMP, the 980 * entry point is exactly a page from the actual load address. 981 * As such, trunc_page() has no effect and we're off by a page. 982 * Since we always have the ELF header between the load address 983 * and the entry point, we can safely subtract 1 to compensate. 984 */ 985 kernstart = trunc_page(start - 1); 986 data_start = round_page(kernelend); 987 data_end = data_start; 988 989 /* 990 * Addresses of preloaded modules (like file systems) use 991 * physical addresses. Make sure we relocate those into 992 * virtual addresses. 993 */ 994 preload_addr_relocate = kernstart - kernload; 995 996 /* Allocate the dynamic per-cpu area. */ 997 dpcpu = (void *)data_end; 998 data_end += DPCPU_SIZE; 999 1000 /* Allocate space for the message buffer. */ 1001 msgbufp = (struct msgbuf *)data_end; 1002 data_end += msgbufsize; 1003 debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp, 1004 data_end); 1005 1006 data_end = round_page(data_end); 1007 1008 /* Allocate space for ptbl_bufs. */ 1009 ptbl_bufs = (struct ptbl_buf *)data_end; 1010 data_end += sizeof(struct ptbl_buf) * PTBL_BUFS; 1011 debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs, 1012 data_end); 1013 1014 data_end = round_page(data_end); 1015 1016 /* Allocate PTE tables for kernel KVA. */ 1017 kernel_pdir = data_end; 1018 kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS + 1019 PDIR_SIZE - 1) / PDIR_SIZE; 1020 data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE; 1021 debugf(" kernel ptbls: %d\n", kernel_ptbls); 1022 debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end); 1023 1024 debugf(" data_end: 0x%08x\n", data_end); 1025 if (data_end - kernstart > kernsize) { 1026 kernsize += tlb1_mapin_region(kernstart + kernsize, 1027 kernload + kernsize, (data_end - kernstart) - kernsize); 1028 } 1029 data_end = kernstart + kernsize; 1030 debugf(" updated data_end: 0x%08x\n", data_end); 1031 1032 /* 1033 * Clear the structures - note we can only do it safely after the 1034 * possible additional TLB1 translations are in place (above) so that 1035 * all range up to the currently calculated 'data_end' is covered. 1036 */ 1037 dpcpu_init(dpcpu, 0); 1038 memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE); 1039 memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE); 1040 1041 /*******************************************************/ 1042 /* Set the start and end of kva. */ 1043 /*******************************************************/ 1044 virtual_avail = round_page(data_end); 1045 virtual_end = VM_MAX_KERNEL_ADDRESS; 1046 1047 /* Allocate KVA space for page zero/copy operations. */ 1048 zero_page_va = virtual_avail; 1049 virtual_avail += PAGE_SIZE; 1050 zero_page_idle_va = virtual_avail; 1051 virtual_avail += PAGE_SIZE; 1052 copy_page_src_va = virtual_avail; 1053 virtual_avail += PAGE_SIZE; 1054 copy_page_dst_va = virtual_avail; 1055 virtual_avail += PAGE_SIZE; 1056 debugf("zero_page_va = 0x%08x\n", zero_page_va); 1057 debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va); 1058 debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va); 1059 debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va); 1060 1061 /* Initialize page zero/copy mutexes. */ 1062 mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF); 1063 mtx_init(©_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF); 1064 1065 /* Allocate KVA space for ptbl bufs. */ 1066 ptbl_buf_pool_vabase = virtual_avail; 1067 virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE; 1068 debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n", 1069 ptbl_buf_pool_vabase, virtual_avail); 1070 1071 /* Calculate corresponding physical addresses for the kernel region. */ 1072 phys_kernelend = kernload + kernsize; 1073 debugf("kernel image and allocated data:\n"); 1074 debugf(" kernload = 0x%08x\n", kernload); 1075 debugf(" kernstart = 0x%08x\n", kernstart); 1076 debugf(" kernsize = 0x%08x\n", kernsize); 1077 1078 if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz) 1079 panic("mmu_booke_bootstrap: phys_avail too small"); 1080 1081 /* 1082 * Remove kernel physical address range from avail regions list. Page 1083 * align all regions. Non-page aligned memory isn't very interesting 1084 * to us. Also, sort the entries for ascending addresses. 1085 */ 1086 1087 /* Retrieve phys/avail mem regions */ 1088 mem_regions(&physmem_regions, &physmem_regions_sz, 1089 &availmem_regions, &availmem_regions_sz); 1090 sz = 0; 1091 cnt = availmem_regions_sz; 1092 debugf("processing avail regions:\n"); 1093 for (mp = availmem_regions; mp->mr_size; mp++) { 1094 s = mp->mr_start; 1095 e = mp->mr_start + mp->mr_size; 1096 debugf(" %08x-%08x -> ", s, e); 1097 /* Check whether this region holds all of the kernel. */ 1098 if (s < kernload && e > phys_kernelend) { 1099 availmem_regions[cnt].mr_start = phys_kernelend; 1100 availmem_regions[cnt++].mr_size = e - phys_kernelend; 1101 e = kernload; 1102 } 1103 /* Look whether this regions starts within the kernel. */ 1104 if (s >= kernload && s < phys_kernelend) { 1105 if (e <= phys_kernelend) 1106 goto empty; 1107 s = phys_kernelend; 1108 } 1109 /* Now look whether this region ends within the kernel. */ 1110 if (e > kernload && e <= phys_kernelend) { 1111 if (s >= kernload) 1112 goto empty; 1113 e = kernload; 1114 } 1115 /* Now page align the start and size of the region. */ 1116 s = round_page(s); 1117 e = trunc_page(e); 1118 if (e < s) 1119 e = s; 1120 sz = e - s; 1121 debugf("%08x-%08x = %x\n", s, e, sz); 1122 1123 /* Check whether some memory is left here. */ 1124 if (sz == 0) { 1125 empty: 1126 memmove(mp, mp + 1, 1127 (cnt - (mp - availmem_regions)) * sizeof(*mp)); 1128 cnt--; 1129 mp--; 1130 continue; 1131 } 1132 1133 /* Do an insertion sort. */ 1134 for (mp1 = availmem_regions; mp1 < mp; mp1++) 1135 if (s < mp1->mr_start) 1136 break; 1137 if (mp1 < mp) { 1138 memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1); 1139 mp1->mr_start = s; 1140 mp1->mr_size = sz; 1141 } else { 1142 mp->mr_start = s; 1143 mp->mr_size = sz; 1144 } 1145 } 1146 availmem_regions_sz = cnt; 1147 1148 /*******************************************************/ 1149 /* Steal physical memory for kernel stack from the end */ 1150 /* of the first avail region */ 1151 /*******************************************************/ 1152 kstack0_sz = KSTACK_PAGES * PAGE_SIZE; 1153 kstack0_phys = availmem_regions[0].mr_start + 1154 availmem_regions[0].mr_size; 1155 kstack0_phys -= kstack0_sz; 1156 availmem_regions[0].mr_size -= kstack0_sz; 1157 1158 /*******************************************************/ 1159 /* Fill in phys_avail table, based on availmem_regions */ 1160 /*******************************************************/ 1161 phys_avail_count = 0; 1162 physsz = 0; 1163 hwphyssz = 0; 1164 TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz); 1165 1166 debugf("fill in phys_avail:\n"); 1167 for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) { 1168 1169 debugf(" region: 0x%08x - 0x%08x (0x%08x)\n", 1170 availmem_regions[i].mr_start, 1171 availmem_regions[i].mr_start + 1172 availmem_regions[i].mr_size, 1173 availmem_regions[i].mr_size); 1174 1175 if (hwphyssz != 0 && 1176 (physsz + availmem_regions[i].mr_size) >= hwphyssz) { 1177 debugf(" hw.physmem adjust\n"); 1178 if (physsz < hwphyssz) { 1179 phys_avail[j] = availmem_regions[i].mr_start; 1180 phys_avail[j + 1] = 1181 availmem_regions[i].mr_start + 1182 hwphyssz - physsz; 1183 physsz = hwphyssz; 1184 phys_avail_count++; 1185 } 1186 break; 1187 } 1188 1189 phys_avail[j] = availmem_regions[i].mr_start; 1190 phys_avail[j + 1] = availmem_regions[i].mr_start + 1191 availmem_regions[i].mr_size; 1192 phys_avail_count++; 1193 physsz += availmem_regions[i].mr_size; 1194 } 1195 physmem = btoc(physsz); 1196 1197 /* Calculate the last available physical address. */ 1198 for (i = 0; phys_avail[i + 2] != 0; i += 2) 1199 ; 1200 Maxmem = powerpc_btop(phys_avail[i + 1]); 1201 1202 debugf("Maxmem = 0x%08lx\n", Maxmem); 1203 debugf("phys_avail_count = %d\n", phys_avail_count); 1204 debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem, 1205 physmem); 1206 1207 /*******************************************************/ 1208 /* Initialize (statically allocated) kernel pmap. */ 1209 /*******************************************************/ 1210 PMAP_LOCK_INIT(kernel_pmap); 1211 kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE; 1212 1213 debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap); 1214 debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls); 1215 debugf("kernel pdir range: 0x%08x - 0x%08x\n", 1216 kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1); 1217 1218 /* Initialize kernel pdir */ 1219 for (i = 0; i < kernel_ptbls; i++) 1220 kernel_pmap->pm_pdir[kptbl_min + i] = 1221 (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES)); 1222 1223 for (i = 0; i < MAXCPU; i++) { 1224 kernel_pmap->pm_tid[i] = TID_KERNEL; 1225 1226 /* Initialize each CPU's tidbusy entry 0 with kernel_pmap */ 1227 tidbusy[i][0] = kernel_pmap; 1228 } 1229 1230 /* 1231 * Fill in PTEs covering kernel code and data. They are not required 1232 * for address translation, as this area is covered by static TLB1 1233 * entries, but for pte_vatopa() to work correctly with kernel area 1234 * addresses. 1235 */ 1236 for (va = kernstart; va < data_end; va += PAGE_SIZE) { 1237 pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]); 1238 pte->rpn = kernload + (va - kernstart); 1239 pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | 1240 PTE_VALID; 1241 } 1242 /* Mark kernel_pmap active on all CPUs */ 1243 CPU_FILL(&kernel_pmap->pm_active); 1244 1245 /*******************************************************/ 1246 /* Final setup */ 1247 /*******************************************************/ 1248 1249 /* Enter kstack0 into kernel map, provide guard page */ 1250 kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE; 1251 thread0.td_kstack = kstack0; 1252 thread0.td_kstack_pages = KSTACK_PAGES; 1253 1254 debugf("kstack_sz = 0x%08x\n", kstack0_sz); 1255 debugf("kstack0_phys at 0x%08x - 0x%08x\n", 1256 kstack0_phys, kstack0_phys + kstack0_sz); 1257 debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz); 1258 1259 virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz; 1260 for (i = 0; i < KSTACK_PAGES; i++) { 1261 mmu_booke_kenter(mmu, kstack0, kstack0_phys); 1262 kstack0 += PAGE_SIZE; 1263 kstack0_phys += PAGE_SIZE; 1264 } 1265 1266 debugf("virtual_avail = %08x\n", virtual_avail); 1267 debugf("virtual_end = %08x\n", virtual_end); 1268 1269 debugf("mmu_booke_bootstrap: exit\n"); 1270 } 1271 1272 void 1273 pmap_bootstrap_ap(volatile uint32_t *trcp __unused) 1274 { 1275 int i; 1276 1277 /* 1278 * Finish TLB1 configuration: the BSP already set up its TLB1 and we 1279 * have the snapshot of its contents in the s/w tlb1[] table, so use 1280 * these values directly to (re)program AP's TLB1 hardware. 1281 */ 1282 for (i = 0; i < tlb1_idx; i ++) { 1283 /* Skip invalid entries */ 1284 if (!(tlb1[i].mas1 & MAS1_VALID)) 1285 continue; 1286 1287 tlb1_write_entry(i); 1288 } 1289 1290 set_mas4_defaults(); 1291 } 1292 1293 /* 1294 * Get the physical page address for the given pmap/virtual address. 1295 */ 1296 static vm_paddr_t 1297 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va) 1298 { 1299 vm_paddr_t pa; 1300 1301 PMAP_LOCK(pmap); 1302 pa = pte_vatopa(mmu, pmap, va); 1303 PMAP_UNLOCK(pmap); 1304 1305 return (pa); 1306 } 1307 1308 /* 1309 * Extract the physical page address associated with the given 1310 * kernel virtual address. 1311 */ 1312 static vm_paddr_t 1313 mmu_booke_kextract(mmu_t mmu, vm_offset_t va) 1314 { 1315 1316 return (pte_vatopa(mmu, kernel_pmap, va)); 1317 } 1318 1319 /* 1320 * Initialize the pmap module. 1321 * Called by vm_init, to initialize any structures that the pmap 1322 * system needs to map virtual memory. 1323 */ 1324 static void 1325 mmu_booke_init(mmu_t mmu) 1326 { 1327 int shpgperproc = PMAP_SHPGPERPROC; 1328 1329 /* 1330 * Initialize the address space (zone) for the pv entries. Set a 1331 * high water mark so that the system can recover from excessive 1332 * numbers of pv entries. 1333 */ 1334 pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL, 1335 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE); 1336 1337 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc); 1338 pv_entry_max = shpgperproc * maxproc + cnt.v_page_count; 1339 1340 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max); 1341 pv_entry_high_water = 9 * (pv_entry_max / 10); 1342 1343 uma_zone_set_obj(pvzone, &pvzone_obj, pv_entry_max); 1344 1345 /* Pre-fill pvzone with initial number of pv entries. */ 1346 uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN); 1347 1348 /* Initialize ptbl allocation. */ 1349 ptbl_init(); 1350 } 1351 1352 /* 1353 * Map a list of wired pages into kernel virtual address space. This is 1354 * intended for temporary mappings which do not need page modification or 1355 * references recorded. Existing mappings in the region are overwritten. 1356 */ 1357 static void 1358 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count) 1359 { 1360 vm_offset_t va; 1361 1362 va = sva; 1363 while (count-- > 0) { 1364 mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m)); 1365 va += PAGE_SIZE; 1366 m++; 1367 } 1368 } 1369 1370 /* 1371 * Remove page mappings from kernel virtual address space. Intended for 1372 * temporary mappings entered by mmu_booke_qenter. 1373 */ 1374 static void 1375 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count) 1376 { 1377 vm_offset_t va; 1378 1379 va = sva; 1380 while (count-- > 0) { 1381 mmu_booke_kremove(mmu, va); 1382 va += PAGE_SIZE; 1383 } 1384 } 1385 1386 /* 1387 * Map a wired page into kernel virtual address space. 1388 */ 1389 static void 1390 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa) 1391 { 1392 unsigned int pdir_idx = PDIR_IDX(va); 1393 unsigned int ptbl_idx = PTBL_IDX(va); 1394 uint32_t flags; 1395 pte_t *pte; 1396 1397 KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) && 1398 (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va")); 1399 1400 flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID; 1401 1402 pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]); 1403 1404 mtx_lock_spin(&tlbivax_mutex); 1405 tlb_miss_lock(); 1406 1407 if (PTE_ISVALID(pte)) { 1408 1409 CTR1(KTR_PMAP, "%s: replacing entry!", __func__); 1410 1411 /* Flush entry from TLB0 */ 1412 tlb0_flush_entry(va); 1413 } 1414 1415 pte->rpn = pa & ~PTE_PA_MASK; 1416 pte->flags = flags; 1417 1418 //debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x " 1419 // "pa=0x%08x rpn=0x%08x flags=0x%08x\n", 1420 // pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags); 1421 1422 /* Flush the real memory from the instruction cache. */ 1423 if ((flags & (PTE_I | PTE_G)) == 0) { 1424 __syncicache((void *)va, PAGE_SIZE); 1425 } 1426 1427 tlb_miss_unlock(); 1428 mtx_unlock_spin(&tlbivax_mutex); 1429 } 1430 1431 /* 1432 * Remove a page from kernel page table. 1433 */ 1434 static void 1435 mmu_booke_kremove(mmu_t mmu, vm_offset_t va) 1436 { 1437 unsigned int pdir_idx = PDIR_IDX(va); 1438 unsigned int ptbl_idx = PTBL_IDX(va); 1439 pte_t *pte; 1440 1441 // CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va)); 1442 1443 KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) && 1444 (va <= VM_MAX_KERNEL_ADDRESS)), 1445 ("mmu_booke_kremove: invalid va")); 1446 1447 pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]); 1448 1449 if (!PTE_ISVALID(pte)) { 1450 1451 CTR1(KTR_PMAP, "%s: invalid pte", __func__); 1452 1453 return; 1454 } 1455 1456 mtx_lock_spin(&tlbivax_mutex); 1457 tlb_miss_lock(); 1458 1459 /* Invalidate entry in TLB0, update PTE. */ 1460 tlb0_flush_entry(va); 1461 pte->flags = 0; 1462 pte->rpn = 0; 1463 1464 tlb_miss_unlock(); 1465 mtx_unlock_spin(&tlbivax_mutex); 1466 } 1467 1468 /* 1469 * Initialize pmap associated with process 0. 1470 */ 1471 static void 1472 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap) 1473 { 1474 1475 mmu_booke_pinit(mmu, pmap); 1476 PCPU_SET(curpmap, pmap); 1477 } 1478 1479 /* 1480 * Initialize a preallocated and zeroed pmap structure, 1481 * such as one in a vmspace structure. 1482 */ 1483 static void 1484 mmu_booke_pinit(mmu_t mmu, pmap_t pmap) 1485 { 1486 int i; 1487 1488 CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap, 1489 curthread->td_proc->p_pid, curthread->td_proc->p_comm); 1490 1491 KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap")); 1492 1493 PMAP_LOCK_INIT(pmap); 1494 for (i = 0; i < MAXCPU; i++) 1495 pmap->pm_tid[i] = TID_NONE; 1496 CPU_ZERO(&kernel_pmap->pm_active); 1497 bzero(&pmap->pm_stats, sizeof(pmap->pm_stats)); 1498 bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES); 1499 TAILQ_INIT(&pmap->pm_ptbl_list); 1500 } 1501 1502 /* 1503 * Release any resources held by the given physical map. 1504 * Called when a pmap initialized by mmu_booke_pinit is being released. 1505 * Should only be called if the map contains no valid mappings. 1506 */ 1507 static void 1508 mmu_booke_release(mmu_t mmu, pmap_t pmap) 1509 { 1510 1511 KASSERT(pmap->pm_stats.resident_count == 0, 1512 ("pmap_release: pmap resident count %ld != 0", 1513 pmap->pm_stats.resident_count)); 1514 1515 PMAP_LOCK_DESTROY(pmap); 1516 } 1517 1518 /* 1519 * Insert the given physical page at the specified virtual address in the 1520 * target physical map with the protection requested. If specified the page 1521 * will be wired down. 1522 */ 1523 static void 1524 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m, 1525 vm_prot_t prot, boolean_t wired) 1526 { 1527 1528 vm_page_lock_queues(); 1529 PMAP_LOCK(pmap); 1530 mmu_booke_enter_locked(mmu, pmap, va, m, prot, wired); 1531 vm_page_unlock_queues(); 1532 PMAP_UNLOCK(pmap); 1533 } 1534 1535 static void 1536 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m, 1537 vm_prot_t prot, boolean_t wired) 1538 { 1539 pte_t *pte; 1540 vm_paddr_t pa; 1541 uint32_t flags; 1542 int su, sync; 1543 1544 pa = VM_PAGE_TO_PHYS(m); 1545 su = (pmap == kernel_pmap); 1546 sync = 0; 1547 1548 //debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x " 1549 // "pa=0x%08x prot=0x%08x wired=%d)\n", 1550 // (u_int32_t)pmap, su, pmap->pm_tid, 1551 // (u_int32_t)m, va, pa, prot, wired); 1552 1553 if (su) { 1554 KASSERT(((va >= virtual_avail) && 1555 (va <= VM_MAX_KERNEL_ADDRESS)), 1556 ("mmu_booke_enter_locked: kernel pmap, non kernel va")); 1557 } else { 1558 KASSERT((va <= VM_MAXUSER_ADDRESS), 1559 ("mmu_booke_enter_locked: user pmap, non user va")); 1560 } 1561 KASSERT((m->oflags & (VPO_UNMANAGED | VPO_BUSY)) != 0 || 1562 VM_OBJECT_LOCKED(m->object), 1563 ("mmu_booke_enter_locked: page %p is not busy", m)); 1564 1565 PMAP_LOCK_ASSERT(pmap, MA_OWNED); 1566 1567 /* 1568 * If there is an existing mapping, and the physical address has not 1569 * changed, must be protection or wiring change. 1570 */ 1571 if (((pte = pte_find(mmu, pmap, va)) != NULL) && 1572 (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) { 1573 1574 /* 1575 * Before actually updating pte->flags we calculate and 1576 * prepare its new value in a helper var. 1577 */ 1578 flags = pte->flags; 1579 flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED); 1580 1581 /* Wiring change, just update stats. */ 1582 if (wired) { 1583 if (!PTE_ISWIRED(pte)) { 1584 flags |= PTE_WIRED; 1585 pmap->pm_stats.wired_count++; 1586 } 1587 } else { 1588 if (PTE_ISWIRED(pte)) { 1589 flags &= ~PTE_WIRED; 1590 pmap->pm_stats.wired_count--; 1591 } 1592 } 1593 1594 if (prot & VM_PROT_WRITE) { 1595 /* Add write permissions. */ 1596 flags |= PTE_SW; 1597 if (!su) 1598 flags |= PTE_UW; 1599 1600 if ((flags & PTE_MANAGED) != 0) 1601 vm_page_aflag_set(m, PGA_WRITEABLE); 1602 } else { 1603 /* Handle modified pages, sense modify status. */ 1604 1605 /* 1606 * The PTE_MODIFIED flag could be set by underlying 1607 * TLB misses since we last read it (above), possibly 1608 * other CPUs could update it so we check in the PTE 1609 * directly rather than rely on that saved local flags 1610 * copy. 1611 */ 1612 if (PTE_ISMODIFIED(pte)) 1613 vm_page_dirty(m); 1614 } 1615 1616 if (prot & VM_PROT_EXECUTE) { 1617 flags |= PTE_SX; 1618 if (!su) 1619 flags |= PTE_UX; 1620 1621 /* 1622 * Check existing flags for execute permissions: if we 1623 * are turning execute permissions on, icache should 1624 * be flushed. 1625 */ 1626 if ((pte->flags & (PTE_UX | PTE_SX)) == 0) 1627 sync++; 1628 } 1629 1630 flags &= ~PTE_REFERENCED; 1631 1632 /* 1633 * The new flags value is all calculated -- only now actually 1634 * update the PTE. 1635 */ 1636 mtx_lock_spin(&tlbivax_mutex); 1637 tlb_miss_lock(); 1638 1639 tlb0_flush_entry(va); 1640 pte->flags = flags; 1641 1642 tlb_miss_unlock(); 1643 mtx_unlock_spin(&tlbivax_mutex); 1644 1645 } else { 1646 /* 1647 * If there is an existing mapping, but it's for a different 1648 * physical address, pte_enter() will delete the old mapping. 1649 */ 1650 //if ((pte != NULL) && PTE_ISVALID(pte)) 1651 // debugf("mmu_booke_enter_locked: replace\n"); 1652 //else 1653 // debugf("mmu_booke_enter_locked: new\n"); 1654 1655 /* Now set up the flags and install the new mapping. */ 1656 flags = (PTE_SR | PTE_VALID); 1657 flags |= PTE_M; 1658 1659 if (!su) 1660 flags |= PTE_UR; 1661 1662 if (prot & VM_PROT_WRITE) { 1663 flags |= PTE_SW; 1664 if (!su) 1665 flags |= PTE_UW; 1666 1667 if ((m->oflags & VPO_UNMANAGED) == 0) 1668 vm_page_aflag_set(m, PGA_WRITEABLE); 1669 } 1670 1671 if (prot & VM_PROT_EXECUTE) { 1672 flags |= PTE_SX; 1673 if (!su) 1674 flags |= PTE_UX; 1675 } 1676 1677 /* If its wired update stats. */ 1678 if (wired) { 1679 pmap->pm_stats.wired_count++; 1680 flags |= PTE_WIRED; 1681 } 1682 1683 pte_enter(mmu, pmap, m, va, flags); 1684 1685 /* Flush the real memory from the instruction cache. */ 1686 if (prot & VM_PROT_EXECUTE) 1687 sync++; 1688 } 1689 1690 if (sync && (su || pmap == PCPU_GET(curpmap))) { 1691 __syncicache((void *)va, PAGE_SIZE); 1692 sync = 0; 1693 } 1694 } 1695 1696 /* 1697 * Maps a sequence of resident pages belonging to the same object. 1698 * The sequence begins with the given page m_start. This page is 1699 * mapped at the given virtual address start. Each subsequent page is 1700 * mapped at a virtual address that is offset from start by the same 1701 * amount as the page is offset from m_start within the object. The 1702 * last page in the sequence is the page with the largest offset from 1703 * m_start that can be mapped at a virtual address less than the given 1704 * virtual address end. Not every virtual page between start and end 1705 * is mapped; only those for which a resident page exists with the 1706 * corresponding offset from m_start are mapped. 1707 */ 1708 static void 1709 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start, 1710 vm_offset_t end, vm_page_t m_start, vm_prot_t prot) 1711 { 1712 vm_page_t m; 1713 vm_pindex_t diff, psize; 1714 1715 psize = atop(end - start); 1716 m = m_start; 1717 vm_page_lock_queues(); 1718 PMAP_LOCK(pmap); 1719 while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) { 1720 mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m, 1721 prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE); 1722 m = TAILQ_NEXT(m, listq); 1723 } 1724 vm_page_unlock_queues(); 1725 PMAP_UNLOCK(pmap); 1726 } 1727 1728 static void 1729 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m, 1730 vm_prot_t prot) 1731 { 1732 1733 vm_page_lock_queues(); 1734 PMAP_LOCK(pmap); 1735 mmu_booke_enter_locked(mmu, pmap, va, m, 1736 prot & (VM_PROT_READ | VM_PROT_EXECUTE), FALSE); 1737 vm_page_unlock_queues(); 1738 PMAP_UNLOCK(pmap); 1739 } 1740 1741 /* 1742 * Remove the given range of addresses from the specified map. 1743 * 1744 * It is assumed that the start and end are properly rounded to the page size. 1745 */ 1746 static void 1747 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva) 1748 { 1749 pte_t *pte; 1750 uint8_t hold_flag; 1751 1752 int su = (pmap == kernel_pmap); 1753 1754 //debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n", 1755 // su, (u_int32_t)pmap, pmap->pm_tid, va, endva); 1756 1757 if (su) { 1758 KASSERT(((va >= virtual_avail) && 1759 (va <= VM_MAX_KERNEL_ADDRESS)), 1760 ("mmu_booke_remove: kernel pmap, non kernel va")); 1761 } else { 1762 KASSERT((va <= VM_MAXUSER_ADDRESS), 1763 ("mmu_booke_remove: user pmap, non user va")); 1764 } 1765 1766 if (PMAP_REMOVE_DONE(pmap)) { 1767 //debugf("mmu_booke_remove: e (empty)\n"); 1768 return; 1769 } 1770 1771 hold_flag = PTBL_HOLD_FLAG(pmap); 1772 //debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag); 1773 1774 vm_page_lock_queues(); 1775 PMAP_LOCK(pmap); 1776 for (; va < endva; va += PAGE_SIZE) { 1777 pte = pte_find(mmu, pmap, va); 1778 if ((pte != NULL) && PTE_ISVALID(pte)) 1779 pte_remove(mmu, pmap, va, hold_flag); 1780 } 1781 PMAP_UNLOCK(pmap); 1782 vm_page_unlock_queues(); 1783 1784 //debugf("mmu_booke_remove: e\n"); 1785 } 1786 1787 /* 1788 * Remove physical page from all pmaps in which it resides. 1789 */ 1790 static void 1791 mmu_booke_remove_all(mmu_t mmu, vm_page_t m) 1792 { 1793 pv_entry_t pv, pvn; 1794 uint8_t hold_flag; 1795 1796 vm_page_lock_queues(); 1797 for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) { 1798 pvn = TAILQ_NEXT(pv, pv_link); 1799 1800 PMAP_LOCK(pv->pv_pmap); 1801 hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap); 1802 pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag); 1803 PMAP_UNLOCK(pv->pv_pmap); 1804 } 1805 vm_page_aflag_clear(m, PGA_WRITEABLE); 1806 vm_page_unlock_queues(); 1807 } 1808 1809 /* 1810 * Map a range of physical addresses into kernel virtual address space. 1811 */ 1812 static vm_offset_t 1813 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start, 1814 vm_paddr_t pa_end, int prot) 1815 { 1816 vm_offset_t sva = *virt; 1817 vm_offset_t va = sva; 1818 1819 //debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n", 1820 // sva, pa_start, pa_end); 1821 1822 while (pa_start < pa_end) { 1823 mmu_booke_kenter(mmu, va, pa_start); 1824 va += PAGE_SIZE; 1825 pa_start += PAGE_SIZE; 1826 } 1827 *virt = va; 1828 1829 //debugf("mmu_booke_map: e (va = 0x%08x)\n", va); 1830 return (sva); 1831 } 1832 1833 /* 1834 * The pmap must be activated before it's address space can be accessed in any 1835 * way. 1836 */ 1837 static void 1838 mmu_booke_activate(mmu_t mmu, struct thread *td) 1839 { 1840 pmap_t pmap; 1841 u_int cpuid; 1842 1843 pmap = &td->td_proc->p_vmspace->vm_pmap; 1844 1845 CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)", 1846 __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap); 1847 1848 KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!")); 1849 1850 mtx_lock_spin(&sched_lock); 1851 1852 cpuid = PCPU_GET(cpuid); 1853 CPU_SET_ATOMIC(cpuid, &pmap->pm_active); 1854 PCPU_SET(curpmap, pmap); 1855 1856 if (pmap->pm_tid[cpuid] == TID_NONE) 1857 tid_alloc(pmap); 1858 1859 /* Load PID0 register with pmap tid value. */ 1860 mtspr(SPR_PID0, pmap->pm_tid[cpuid]); 1861 __asm __volatile("isync"); 1862 1863 mtx_unlock_spin(&sched_lock); 1864 1865 CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__, 1866 pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm); 1867 } 1868 1869 /* 1870 * Deactivate the specified process's address space. 1871 */ 1872 static void 1873 mmu_booke_deactivate(mmu_t mmu, struct thread *td) 1874 { 1875 pmap_t pmap; 1876 1877 pmap = &td->td_proc->p_vmspace->vm_pmap; 1878 1879 CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x", 1880 __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap); 1881 1882 CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active); 1883 PCPU_SET(curpmap, NULL); 1884 } 1885 1886 /* 1887 * Copy the range specified by src_addr/len 1888 * from the source map to the range dst_addr/len 1889 * in the destination map. 1890 * 1891 * This routine is only advisory and need not do anything. 1892 */ 1893 static void 1894 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap, 1895 vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr) 1896 { 1897 1898 } 1899 1900 /* 1901 * Set the physical protection on the specified range of this map as requested. 1902 */ 1903 static void 1904 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva, 1905 vm_prot_t prot) 1906 { 1907 vm_offset_t va; 1908 vm_page_t m; 1909 pte_t *pte; 1910 1911 if ((prot & VM_PROT_READ) == VM_PROT_NONE) { 1912 mmu_booke_remove(mmu, pmap, sva, eva); 1913 return; 1914 } 1915 1916 if (prot & VM_PROT_WRITE) 1917 return; 1918 1919 PMAP_LOCK(pmap); 1920 for (va = sva; va < eva; va += PAGE_SIZE) { 1921 if ((pte = pte_find(mmu, pmap, va)) != NULL) { 1922 if (PTE_ISVALID(pte)) { 1923 m = PHYS_TO_VM_PAGE(PTE_PA(pte)); 1924 1925 mtx_lock_spin(&tlbivax_mutex); 1926 tlb_miss_lock(); 1927 1928 /* Handle modified pages. */ 1929 if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte)) 1930 vm_page_dirty(m); 1931 1932 tlb0_flush_entry(va); 1933 pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED); 1934 1935 tlb_miss_unlock(); 1936 mtx_unlock_spin(&tlbivax_mutex); 1937 } 1938 } 1939 } 1940 PMAP_UNLOCK(pmap); 1941 } 1942 1943 /* 1944 * Clear the write and modified bits in each of the given page's mappings. 1945 */ 1946 static void 1947 mmu_booke_remove_write(mmu_t mmu, vm_page_t m) 1948 { 1949 pv_entry_t pv; 1950 pte_t *pte; 1951 1952 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 1953 ("mmu_booke_remove_write: page %p is not managed", m)); 1954 1955 /* 1956 * If the page is not VPO_BUSY, then PGA_WRITEABLE cannot be set by 1957 * another thread while the object is locked. Thus, if PGA_WRITEABLE 1958 * is clear, no page table entries need updating. 1959 */ 1960 VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 1961 if ((m->oflags & VPO_BUSY) == 0 && 1962 (m->aflags & PGA_WRITEABLE) == 0) 1963 return; 1964 vm_page_lock_queues(); 1965 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) { 1966 PMAP_LOCK(pv->pv_pmap); 1967 if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) { 1968 if (PTE_ISVALID(pte)) { 1969 m = PHYS_TO_VM_PAGE(PTE_PA(pte)); 1970 1971 mtx_lock_spin(&tlbivax_mutex); 1972 tlb_miss_lock(); 1973 1974 /* Handle modified pages. */ 1975 if (PTE_ISMODIFIED(pte)) 1976 vm_page_dirty(m); 1977 1978 /* Flush mapping from TLB0. */ 1979 pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED); 1980 1981 tlb_miss_unlock(); 1982 mtx_unlock_spin(&tlbivax_mutex); 1983 } 1984 } 1985 PMAP_UNLOCK(pv->pv_pmap); 1986 } 1987 vm_page_aflag_clear(m, PGA_WRITEABLE); 1988 vm_page_unlock_queues(); 1989 } 1990 1991 static void 1992 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz) 1993 { 1994 pte_t *pte; 1995 pmap_t pmap; 1996 vm_page_t m; 1997 vm_offset_t addr; 1998 vm_paddr_t pa; 1999 int active, valid; 2000 2001 va = trunc_page(va); 2002 sz = round_page(sz); 2003 2004 vm_page_lock_queues(); 2005 pmap = PCPU_GET(curpmap); 2006 active = (pm == kernel_pmap || pm == pmap) ? 1 : 0; 2007 while (sz > 0) { 2008 PMAP_LOCK(pm); 2009 pte = pte_find(mmu, pm, va); 2010 valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0; 2011 if (valid) 2012 pa = PTE_PA(pte); 2013 PMAP_UNLOCK(pm); 2014 if (valid) { 2015 if (!active) { 2016 /* Create a mapping in the active pmap. */ 2017 addr = 0; 2018 m = PHYS_TO_VM_PAGE(pa); 2019 PMAP_LOCK(pmap); 2020 pte_enter(mmu, pmap, m, addr, 2021 PTE_SR | PTE_VALID | PTE_UR); 2022 __syncicache((void *)addr, PAGE_SIZE); 2023 pte_remove(mmu, pmap, addr, PTBL_UNHOLD); 2024 PMAP_UNLOCK(pmap); 2025 } else 2026 __syncicache((void *)va, PAGE_SIZE); 2027 } 2028 va += PAGE_SIZE; 2029 sz -= PAGE_SIZE; 2030 } 2031 vm_page_unlock_queues(); 2032 } 2033 2034 /* 2035 * Atomically extract and hold the physical page with the given 2036 * pmap and virtual address pair if that mapping permits the given 2037 * protection. 2038 */ 2039 static vm_page_t 2040 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va, 2041 vm_prot_t prot) 2042 { 2043 pte_t *pte; 2044 vm_page_t m; 2045 uint32_t pte_wbit; 2046 vm_paddr_t pa; 2047 2048 m = NULL; 2049 pa = 0; 2050 PMAP_LOCK(pmap); 2051 retry: 2052 pte = pte_find(mmu, pmap, va); 2053 if ((pte != NULL) && PTE_ISVALID(pte)) { 2054 if (pmap == kernel_pmap) 2055 pte_wbit = PTE_SW; 2056 else 2057 pte_wbit = PTE_UW; 2058 2059 if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) { 2060 if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa)) 2061 goto retry; 2062 m = PHYS_TO_VM_PAGE(PTE_PA(pte)); 2063 vm_page_hold(m); 2064 } 2065 } 2066 2067 PA_UNLOCK_COND(pa); 2068 PMAP_UNLOCK(pmap); 2069 return (m); 2070 } 2071 2072 /* 2073 * Initialize a vm_page's machine-dependent fields. 2074 */ 2075 static void 2076 mmu_booke_page_init(mmu_t mmu, vm_page_t m) 2077 { 2078 2079 TAILQ_INIT(&m->md.pv_list); 2080 } 2081 2082 /* 2083 * mmu_booke_zero_page_area zeros the specified hardware page by 2084 * mapping it into virtual memory and using bzero to clear 2085 * its contents. 2086 * 2087 * off and size must reside within a single page. 2088 */ 2089 static void 2090 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size) 2091 { 2092 vm_offset_t va; 2093 2094 /* XXX KASSERT off and size are within a single page? */ 2095 2096 mtx_lock(&zero_page_mutex); 2097 va = zero_page_va; 2098 2099 mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m)); 2100 bzero((caddr_t)va + off, size); 2101 mmu_booke_kremove(mmu, va); 2102 2103 mtx_unlock(&zero_page_mutex); 2104 } 2105 2106 /* 2107 * mmu_booke_zero_page zeros the specified hardware page. 2108 */ 2109 static void 2110 mmu_booke_zero_page(mmu_t mmu, vm_page_t m) 2111 { 2112 2113 mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE); 2114 } 2115 2116 /* 2117 * mmu_booke_copy_page copies the specified (machine independent) page by 2118 * mapping the page into virtual memory and using memcopy to copy the page, 2119 * one machine dependent page at a time. 2120 */ 2121 static void 2122 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm) 2123 { 2124 vm_offset_t sva, dva; 2125 2126 sva = copy_page_src_va; 2127 dva = copy_page_dst_va; 2128 2129 mtx_lock(©_page_mutex); 2130 mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm)); 2131 mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm)); 2132 memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE); 2133 mmu_booke_kremove(mmu, dva); 2134 mmu_booke_kremove(mmu, sva); 2135 mtx_unlock(©_page_mutex); 2136 } 2137 2138 /* 2139 * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it 2140 * into virtual memory and using bzero to clear its contents. This is intended 2141 * to be called from the vm_pagezero process only and outside of Giant. No 2142 * lock is required. 2143 */ 2144 static void 2145 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m) 2146 { 2147 vm_offset_t va; 2148 2149 va = zero_page_idle_va; 2150 mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m)); 2151 bzero((caddr_t)va, PAGE_SIZE); 2152 mmu_booke_kremove(mmu, va); 2153 } 2154 2155 /* 2156 * Return whether or not the specified physical page was modified 2157 * in any of physical maps. 2158 */ 2159 static boolean_t 2160 mmu_booke_is_modified(mmu_t mmu, vm_page_t m) 2161 { 2162 pte_t *pte; 2163 pv_entry_t pv; 2164 boolean_t rv; 2165 2166 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 2167 ("mmu_booke_is_modified: page %p is not managed", m)); 2168 rv = FALSE; 2169 2170 /* 2171 * If the page is not VPO_BUSY, then PGA_WRITEABLE cannot be 2172 * concurrently set while the object is locked. Thus, if PGA_WRITEABLE 2173 * is clear, no PTEs can be modified. 2174 */ 2175 VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 2176 if ((m->oflags & VPO_BUSY) == 0 && 2177 (m->aflags & PGA_WRITEABLE) == 0) 2178 return (rv); 2179 vm_page_lock_queues(); 2180 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) { 2181 PMAP_LOCK(pv->pv_pmap); 2182 if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL && 2183 PTE_ISVALID(pte)) { 2184 if (PTE_ISMODIFIED(pte)) 2185 rv = TRUE; 2186 } 2187 PMAP_UNLOCK(pv->pv_pmap); 2188 if (rv) 2189 break; 2190 } 2191 vm_page_unlock_queues(); 2192 return (rv); 2193 } 2194 2195 /* 2196 * Return whether or not the specified virtual address is eligible 2197 * for prefault. 2198 */ 2199 static boolean_t 2200 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr) 2201 { 2202 2203 return (FALSE); 2204 } 2205 2206 /* 2207 * Return whether or not the specified physical page was referenced 2208 * in any physical maps. 2209 */ 2210 static boolean_t 2211 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m) 2212 { 2213 pte_t *pte; 2214 pv_entry_t pv; 2215 boolean_t rv; 2216 2217 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 2218 ("mmu_booke_is_referenced: page %p is not managed", m)); 2219 rv = FALSE; 2220 vm_page_lock_queues(); 2221 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) { 2222 PMAP_LOCK(pv->pv_pmap); 2223 if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL && 2224 PTE_ISVALID(pte)) { 2225 if (PTE_ISREFERENCED(pte)) 2226 rv = TRUE; 2227 } 2228 PMAP_UNLOCK(pv->pv_pmap); 2229 if (rv) 2230 break; 2231 } 2232 vm_page_unlock_queues(); 2233 return (rv); 2234 } 2235 2236 /* 2237 * Clear the modify bits on the specified physical page. 2238 */ 2239 static void 2240 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m) 2241 { 2242 pte_t *pte; 2243 pv_entry_t pv; 2244 2245 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 2246 ("mmu_booke_clear_modify: page %p is not managed", m)); 2247 VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED); 2248 KASSERT((m->oflags & VPO_BUSY) == 0, 2249 ("mmu_booke_clear_modify: page %p is busy", m)); 2250 2251 /* 2252 * If the page is not PG_AWRITEABLE, then no PTEs can be modified. 2253 * If the object containing the page is locked and the page is not 2254 * VPO_BUSY, then PG_AWRITEABLE cannot be concurrently set. 2255 */ 2256 if ((m->aflags & PGA_WRITEABLE) == 0) 2257 return; 2258 vm_page_lock_queues(); 2259 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) { 2260 PMAP_LOCK(pv->pv_pmap); 2261 if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL && 2262 PTE_ISVALID(pte)) { 2263 mtx_lock_spin(&tlbivax_mutex); 2264 tlb_miss_lock(); 2265 2266 if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) { 2267 tlb0_flush_entry(pv->pv_va); 2268 pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED | 2269 PTE_REFERENCED); 2270 } 2271 2272 tlb_miss_unlock(); 2273 mtx_unlock_spin(&tlbivax_mutex); 2274 } 2275 PMAP_UNLOCK(pv->pv_pmap); 2276 } 2277 vm_page_unlock_queues(); 2278 } 2279 2280 /* 2281 * Return a count of reference bits for a page, clearing those bits. 2282 * It is not necessary for every reference bit to be cleared, but it 2283 * is necessary that 0 only be returned when there are truly no 2284 * reference bits set. 2285 * 2286 * XXX: The exact number of bits to check and clear is a matter that 2287 * should be tested and standardized at some point in the future for 2288 * optimal aging of shared pages. 2289 */ 2290 static int 2291 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m) 2292 { 2293 pte_t *pte; 2294 pv_entry_t pv; 2295 int count; 2296 2297 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 2298 ("mmu_booke_ts_referenced: page %p is not managed", m)); 2299 count = 0; 2300 vm_page_lock_queues(); 2301 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) { 2302 PMAP_LOCK(pv->pv_pmap); 2303 if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL && 2304 PTE_ISVALID(pte)) { 2305 if (PTE_ISREFERENCED(pte)) { 2306 mtx_lock_spin(&tlbivax_mutex); 2307 tlb_miss_lock(); 2308 2309 tlb0_flush_entry(pv->pv_va); 2310 pte->flags &= ~PTE_REFERENCED; 2311 2312 tlb_miss_unlock(); 2313 mtx_unlock_spin(&tlbivax_mutex); 2314 2315 if (++count > 4) { 2316 PMAP_UNLOCK(pv->pv_pmap); 2317 break; 2318 } 2319 } 2320 } 2321 PMAP_UNLOCK(pv->pv_pmap); 2322 } 2323 vm_page_unlock_queues(); 2324 return (count); 2325 } 2326 2327 /* 2328 * Clear the reference bit on the specified physical page. 2329 */ 2330 static void 2331 mmu_booke_clear_reference(mmu_t mmu, vm_page_t m) 2332 { 2333 pte_t *pte; 2334 pv_entry_t pv; 2335 2336 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 2337 ("mmu_booke_clear_reference: page %p is not managed", m)); 2338 vm_page_lock_queues(); 2339 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) { 2340 PMAP_LOCK(pv->pv_pmap); 2341 if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL && 2342 PTE_ISVALID(pte)) { 2343 if (PTE_ISREFERENCED(pte)) { 2344 mtx_lock_spin(&tlbivax_mutex); 2345 tlb_miss_lock(); 2346 2347 tlb0_flush_entry(pv->pv_va); 2348 pte->flags &= ~PTE_REFERENCED; 2349 2350 tlb_miss_unlock(); 2351 mtx_unlock_spin(&tlbivax_mutex); 2352 } 2353 } 2354 PMAP_UNLOCK(pv->pv_pmap); 2355 } 2356 vm_page_unlock_queues(); 2357 } 2358 2359 /* 2360 * Change wiring attribute for a map/virtual-address pair. 2361 */ 2362 static void 2363 mmu_booke_change_wiring(mmu_t mmu, pmap_t pmap, vm_offset_t va, boolean_t wired) 2364 { 2365 pte_t *pte; 2366 2367 PMAP_LOCK(pmap); 2368 if ((pte = pte_find(mmu, pmap, va)) != NULL) { 2369 if (wired) { 2370 if (!PTE_ISWIRED(pte)) { 2371 pte->flags |= PTE_WIRED; 2372 pmap->pm_stats.wired_count++; 2373 } 2374 } else { 2375 if (PTE_ISWIRED(pte)) { 2376 pte->flags &= ~PTE_WIRED; 2377 pmap->pm_stats.wired_count--; 2378 } 2379 } 2380 } 2381 PMAP_UNLOCK(pmap); 2382 } 2383 2384 /* 2385 * Return true if the pmap's pv is one of the first 16 pvs linked to from this 2386 * page. This count may be changed upwards or downwards in the future; it is 2387 * only necessary that true be returned for a small subset of pmaps for proper 2388 * page aging. 2389 */ 2390 static boolean_t 2391 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m) 2392 { 2393 pv_entry_t pv; 2394 int loops; 2395 boolean_t rv; 2396 2397 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 2398 ("mmu_booke_page_exists_quick: page %p is not managed", m)); 2399 loops = 0; 2400 rv = FALSE; 2401 vm_page_lock_queues(); 2402 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) { 2403 if (pv->pv_pmap == pmap) { 2404 rv = TRUE; 2405 break; 2406 } 2407 if (++loops >= 16) 2408 break; 2409 } 2410 vm_page_unlock_queues(); 2411 return (rv); 2412 } 2413 2414 /* 2415 * Return the number of managed mappings to the given physical page that are 2416 * wired. 2417 */ 2418 static int 2419 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m) 2420 { 2421 pv_entry_t pv; 2422 pte_t *pte; 2423 int count = 0; 2424 2425 if ((m->oflags & VPO_UNMANAGED) != 0) 2426 return (count); 2427 vm_page_lock_queues(); 2428 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) { 2429 PMAP_LOCK(pv->pv_pmap); 2430 if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) 2431 if (PTE_ISVALID(pte) && PTE_ISWIRED(pte)) 2432 count++; 2433 PMAP_UNLOCK(pv->pv_pmap); 2434 } 2435 vm_page_unlock_queues(); 2436 return (count); 2437 } 2438 2439 static int 2440 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size) 2441 { 2442 int i; 2443 vm_offset_t va; 2444 2445 /* 2446 * This currently does not work for entries that 2447 * overlap TLB1 entries. 2448 */ 2449 for (i = 0; i < tlb1_idx; i ++) { 2450 if (tlb1_iomapped(i, pa, size, &va) == 0) 2451 return (0); 2452 } 2453 2454 return (EFAULT); 2455 } 2456 2457 vm_offset_t 2458 mmu_booke_dumpsys_map(mmu_t mmu, struct pmap_md *md, vm_size_t ofs, 2459 vm_size_t *sz) 2460 { 2461 vm_paddr_t pa, ppa; 2462 vm_offset_t va; 2463 vm_size_t gran; 2464 2465 /* Raw physical memory dumps don't have a virtual address. */ 2466 if (md->md_vaddr == ~0UL) { 2467 /* We always map a 256MB page at 256M. */ 2468 gran = 256 * 1024 * 1024; 2469 pa = md->md_paddr + ofs; 2470 ppa = pa & ~(gran - 1); 2471 ofs = pa - ppa; 2472 va = gran; 2473 tlb1_set_entry(va, ppa, gran, _TLB_ENTRY_IO); 2474 if (*sz > (gran - ofs)) 2475 *sz = gran - ofs; 2476 return (va + ofs); 2477 } 2478 2479 /* Minidumps are based on virtual memory addresses. */ 2480 va = md->md_vaddr + ofs; 2481 if (va >= kernstart + kernsize) { 2482 gran = PAGE_SIZE - (va & PAGE_MASK); 2483 if (*sz > gran) 2484 *sz = gran; 2485 } 2486 return (va); 2487 } 2488 2489 void 2490 mmu_booke_dumpsys_unmap(mmu_t mmu, struct pmap_md *md, vm_size_t ofs, 2491 vm_offset_t va) 2492 { 2493 2494 /* Raw physical memory dumps don't have a virtual address. */ 2495 if (md->md_vaddr == ~0UL) { 2496 tlb1_idx--; 2497 tlb1[tlb1_idx].mas1 = 0; 2498 tlb1[tlb1_idx].mas2 = 0; 2499 tlb1[tlb1_idx].mas3 = 0; 2500 tlb1_write_entry(tlb1_idx); 2501 return; 2502 } 2503 2504 /* Minidumps are based on virtual memory addresses. */ 2505 /* Nothing to do... */ 2506 } 2507 2508 struct pmap_md * 2509 mmu_booke_scan_md(mmu_t mmu, struct pmap_md *prev) 2510 { 2511 static struct pmap_md md; 2512 pte_t *pte; 2513 vm_offset_t va; 2514 2515 if (dumpsys_minidump) { 2516 md.md_paddr = ~0UL; /* Minidumps use virtual addresses. */ 2517 if (prev == NULL) { 2518 /* 1st: kernel .data and .bss. */ 2519 md.md_index = 1; 2520 md.md_vaddr = trunc_page((uintptr_t)_etext); 2521 md.md_size = round_page((uintptr_t)_end) - md.md_vaddr; 2522 return (&md); 2523 } 2524 switch (prev->md_index) { 2525 case 1: 2526 /* 2nd: msgbuf and tables (see pmap_bootstrap()). */ 2527 md.md_index = 2; 2528 md.md_vaddr = data_start; 2529 md.md_size = data_end - data_start; 2530 break; 2531 case 2: 2532 /* 3rd: kernel VM. */ 2533 va = prev->md_vaddr + prev->md_size; 2534 /* Find start of next chunk (from va). */ 2535 while (va < virtual_end) { 2536 /* Don't dump the buffer cache. */ 2537 if (va >= kmi.buffer_sva && 2538 va < kmi.buffer_eva) { 2539 va = kmi.buffer_eva; 2540 continue; 2541 } 2542 pte = pte_find(mmu, kernel_pmap, va); 2543 if (pte != NULL && PTE_ISVALID(pte)) 2544 break; 2545 va += PAGE_SIZE; 2546 } 2547 if (va < virtual_end) { 2548 md.md_vaddr = va; 2549 va += PAGE_SIZE; 2550 /* Find last page in chunk. */ 2551 while (va < virtual_end) { 2552 /* Don't run into the buffer cache. */ 2553 if (va == kmi.buffer_sva) 2554 break; 2555 pte = pte_find(mmu, kernel_pmap, va); 2556 if (pte == NULL || !PTE_ISVALID(pte)) 2557 break; 2558 va += PAGE_SIZE; 2559 } 2560 md.md_size = va - md.md_vaddr; 2561 break; 2562 } 2563 md.md_index = 3; 2564 /* FALLTHROUGH */ 2565 default: 2566 return (NULL); 2567 } 2568 } else { /* minidumps */ 2569 mem_regions(&physmem_regions, &physmem_regions_sz, 2570 &availmem_regions, &availmem_regions_sz); 2571 2572 if (prev == NULL) { 2573 /* first physical chunk. */ 2574 md.md_paddr = physmem_regions[0].mr_start; 2575 md.md_size = physmem_regions[0].mr_size; 2576 md.md_vaddr = ~0UL; 2577 md.md_index = 1; 2578 } else if (md.md_index < physmem_regions_sz) { 2579 md.md_paddr = physmem_regions[md.md_index].mr_start; 2580 md.md_size = physmem_regions[md.md_index].mr_size; 2581 md.md_vaddr = ~0UL; 2582 md.md_index++; 2583 } else { 2584 /* There's no next physical chunk. */ 2585 return (NULL); 2586 } 2587 } 2588 2589 return (&md); 2590 } 2591 2592 /* 2593 * Map a set of physical memory pages into the kernel virtual address space. 2594 * Return a pointer to where it is mapped. This routine is intended to be used 2595 * for mapping device memory, NOT real memory. 2596 */ 2597 static void * 2598 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size) 2599 { 2600 void *res; 2601 uintptr_t va; 2602 vm_size_t sz; 2603 2604 va = (pa >= 0x80000000) ? pa : (0xe2000000 + pa); 2605 res = (void *)va; 2606 2607 do { 2608 sz = 1 << (ilog2(size) & ~1); 2609 if (bootverbose) 2610 printf("Wiring VA=%x to PA=%x (size=%x), " 2611 "using TLB1[%d]\n", va, pa, sz, tlb1_idx); 2612 tlb1_set_entry(va, pa, sz, _TLB_ENTRY_IO); 2613 size -= sz; 2614 pa += sz; 2615 va += sz; 2616 } while (size > 0); 2617 2618 return (res); 2619 } 2620 2621 /* 2622 * 'Unmap' a range mapped by mmu_booke_mapdev(). 2623 */ 2624 static void 2625 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size) 2626 { 2627 vm_offset_t base, offset; 2628 2629 /* 2630 * Unmap only if this is inside kernel virtual space. 2631 */ 2632 if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) { 2633 base = trunc_page(va); 2634 offset = va & PAGE_MASK; 2635 size = roundup(offset + size, PAGE_SIZE); 2636 kmem_free(kernel_map, base, size); 2637 } 2638 } 2639 2640 /* 2641 * mmu_booke_object_init_pt preloads the ptes for a given object into the 2642 * specified pmap. This eliminates the blast of soft faults on process startup 2643 * and immediately after an mmap. 2644 */ 2645 static void 2646 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr, 2647 vm_object_t object, vm_pindex_t pindex, vm_size_t size) 2648 { 2649 2650 VM_OBJECT_LOCK_ASSERT(object, MA_OWNED); 2651 KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG, 2652 ("mmu_booke_object_init_pt: non-device object")); 2653 } 2654 2655 /* 2656 * Perform the pmap work for mincore. 2657 */ 2658 static int 2659 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr, 2660 vm_paddr_t *locked_pa) 2661 { 2662 2663 TODO; 2664 return (0); 2665 } 2666 2667 /**************************************************************************/ 2668 /* TID handling */ 2669 /**************************************************************************/ 2670 2671 /* 2672 * Allocate a TID. If necessary, steal one from someone else. 2673 * The new TID is flushed from the TLB before returning. 2674 */ 2675 static tlbtid_t 2676 tid_alloc(pmap_t pmap) 2677 { 2678 tlbtid_t tid; 2679 int thiscpu; 2680 2681 KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap")); 2682 2683 CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap); 2684 2685 thiscpu = PCPU_GET(cpuid); 2686 2687 tid = PCPU_GET(tid_next); 2688 if (tid > TID_MAX) 2689 tid = TID_MIN; 2690 PCPU_SET(tid_next, tid + 1); 2691 2692 /* If we are stealing TID then clear the relevant pmap's field */ 2693 if (tidbusy[thiscpu][tid] != NULL) { 2694 2695 CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid); 2696 2697 tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE; 2698 2699 /* Flush all entries from TLB0 matching this TID. */ 2700 tid_flush(tid); 2701 } 2702 2703 tidbusy[thiscpu][tid] = pmap; 2704 pmap->pm_tid[thiscpu] = tid; 2705 __asm __volatile("msync; isync"); 2706 2707 CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid, 2708 PCPU_GET(tid_next)); 2709 2710 return (tid); 2711 } 2712 2713 /**************************************************************************/ 2714 /* TLB0 handling */ 2715 /**************************************************************************/ 2716 2717 static void 2718 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3, 2719 uint32_t mas7) 2720 { 2721 int as; 2722 char desc[3]; 2723 tlbtid_t tid; 2724 vm_size_t size; 2725 unsigned int tsize; 2726 2727 desc[2] = '\0'; 2728 if (mas1 & MAS1_VALID) 2729 desc[0] = 'V'; 2730 else 2731 desc[0] = ' '; 2732 2733 if (mas1 & MAS1_IPROT) 2734 desc[1] = 'P'; 2735 else 2736 desc[1] = ' '; 2737 2738 as = (mas1 & MAS1_TS_MASK) ? 1 : 0; 2739 tid = MAS1_GETTID(mas1); 2740 2741 tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT; 2742 size = 0; 2743 if (tsize) 2744 size = tsize2size(tsize); 2745 2746 debugf("%3d: (%s) [AS=%d] " 2747 "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x " 2748 "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n", 2749 i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7); 2750 } 2751 2752 /* Convert TLB0 va and way number to tlb0[] table index. */ 2753 static inline unsigned int 2754 tlb0_tableidx(vm_offset_t va, unsigned int way) 2755 { 2756 unsigned int idx; 2757 2758 idx = (way * TLB0_ENTRIES_PER_WAY); 2759 idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT; 2760 return (idx); 2761 } 2762 2763 /* 2764 * Invalidate TLB0 entry. 2765 */ 2766 static inline void 2767 tlb0_flush_entry(vm_offset_t va) 2768 { 2769 2770 CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va); 2771 2772 mtx_assert(&tlbivax_mutex, MA_OWNED); 2773 2774 __asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK)); 2775 __asm __volatile("isync; msync"); 2776 __asm __volatile("tlbsync; msync"); 2777 2778 CTR1(KTR_PMAP, "%s: e", __func__); 2779 } 2780 2781 /* Print out contents of the MAS registers for each TLB0 entry */ 2782 void 2783 tlb0_print_tlbentries(void) 2784 { 2785 uint32_t mas0, mas1, mas2, mas3, mas7; 2786 int entryidx, way, idx; 2787 2788 debugf("TLB0 entries:\n"); 2789 for (way = 0; way < TLB0_WAYS; way ++) 2790 for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) { 2791 2792 mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way); 2793 mtspr(SPR_MAS0, mas0); 2794 __asm __volatile("isync"); 2795 2796 mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT; 2797 mtspr(SPR_MAS2, mas2); 2798 2799 __asm __volatile("isync; tlbre"); 2800 2801 mas1 = mfspr(SPR_MAS1); 2802 mas2 = mfspr(SPR_MAS2); 2803 mas3 = mfspr(SPR_MAS3); 2804 mas7 = mfspr(SPR_MAS7); 2805 2806 idx = tlb0_tableidx(mas2, way); 2807 tlb_print_entry(idx, mas1, mas2, mas3, mas7); 2808 } 2809 } 2810 2811 /**************************************************************************/ 2812 /* TLB1 handling */ 2813 /**************************************************************************/ 2814 2815 /* 2816 * TLB1 mapping notes: 2817 * 2818 * TLB1[0] CCSRBAR 2819 * TLB1[1] Kernel text and data. 2820 * TLB1[2-15] Additional kernel text and data mappings (if required), PCI 2821 * windows, other devices mappings. 2822 */ 2823 2824 /* 2825 * Write given entry to TLB1 hardware. 2826 * Use 32 bit pa, clear 4 high-order bits of RPN (mas7). 2827 */ 2828 static void 2829 tlb1_write_entry(unsigned int idx) 2830 { 2831 uint32_t mas0, mas7; 2832 2833 //debugf("tlb1_write_entry: s\n"); 2834 2835 /* Clear high order RPN bits */ 2836 mas7 = 0; 2837 2838 /* Select entry */ 2839 mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx); 2840 //debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0); 2841 2842 mtspr(SPR_MAS0, mas0); 2843 __asm __volatile("isync"); 2844 mtspr(SPR_MAS1, tlb1[idx].mas1); 2845 __asm __volatile("isync"); 2846 mtspr(SPR_MAS2, tlb1[idx].mas2); 2847 __asm __volatile("isync"); 2848 mtspr(SPR_MAS3, tlb1[idx].mas3); 2849 __asm __volatile("isync"); 2850 mtspr(SPR_MAS7, mas7); 2851 __asm __volatile("isync; tlbwe; isync; msync"); 2852 2853 //debugf("tlb1_write_entry: e\n"); 2854 } 2855 2856 /* 2857 * Return the largest uint value log such that 2^log <= num. 2858 */ 2859 static unsigned int 2860 ilog2(unsigned int num) 2861 { 2862 int lz; 2863 2864 __asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num)); 2865 return (31 - lz); 2866 } 2867 2868 /* 2869 * Convert TLB TSIZE value to mapped region size. 2870 */ 2871 static vm_size_t 2872 tsize2size(unsigned int tsize) 2873 { 2874 2875 /* 2876 * size = 4^tsize KB 2877 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10) 2878 */ 2879 2880 return ((1 << (2 * tsize)) * 1024); 2881 } 2882 2883 /* 2884 * Convert region size (must be power of 4) to TLB TSIZE value. 2885 */ 2886 static unsigned int 2887 size2tsize(vm_size_t size) 2888 { 2889 2890 return (ilog2(size) / 2 - 5); 2891 } 2892 2893 /* 2894 * Register permanent kernel mapping in TLB1. 2895 * 2896 * Entries are created starting from index 0 (current free entry is 2897 * kept in tlb1_idx) and are not supposed to be invalidated. 2898 */ 2899 static int 2900 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size, 2901 uint32_t flags) 2902 { 2903 uint32_t ts, tid; 2904 int tsize; 2905 2906 if (tlb1_idx >= TLB1_ENTRIES) { 2907 printf("tlb1_set_entry: TLB1 full!\n"); 2908 return (-1); 2909 } 2910 2911 /* Convert size to TSIZE */ 2912 tsize = size2tsize(size); 2913 2914 tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK; 2915 /* XXX TS is hard coded to 0 for now as we only use single address space */ 2916 ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK; 2917 2918 /* XXX LOCK tlb1[] */ 2919 2920 tlb1[tlb1_idx].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid; 2921 tlb1[tlb1_idx].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK); 2922 tlb1[tlb1_idx].mas2 = (va & MAS2_EPN_MASK) | flags; 2923 2924 /* Set supervisor RWX permission bits */ 2925 tlb1[tlb1_idx].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX; 2926 2927 tlb1_write_entry(tlb1_idx++); 2928 2929 /* XXX UNLOCK tlb1[] */ 2930 2931 /* 2932 * XXX in general TLB1 updates should be propagated between CPUs, 2933 * since current design assumes to have the same TLB1 set-up on all 2934 * cores. 2935 */ 2936 return (0); 2937 } 2938 2939 /* 2940 * Map in contiguous RAM region into the TLB1 using maximum of 2941 * KERNEL_REGION_MAX_TLB_ENTRIES entries. 2942 * 2943 * If necessary round up last entry size and return total size 2944 * used by all allocated entries. 2945 */ 2946 vm_size_t 2947 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size) 2948 { 2949 vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES]; 2950 vm_size_t mapped, pgsz, base, mask; 2951 int idx, nents; 2952 2953 /* Round up to the next 1M */ 2954 size = (size + (1 << 20) - 1) & ~((1 << 20) - 1); 2955 2956 mapped = 0; 2957 idx = 0; 2958 base = va; 2959 pgsz = 64*1024*1024; 2960 while (mapped < size) { 2961 while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) { 2962 while (pgsz > (size - mapped)) 2963 pgsz >>= 2; 2964 pgs[idx++] = pgsz; 2965 mapped += pgsz; 2966 } 2967 2968 /* We under-map. Correct for this. */ 2969 if (mapped < size) { 2970 while (pgs[idx - 1] == pgsz) { 2971 idx--; 2972 mapped -= pgsz; 2973 } 2974 /* XXX We may increase beyond out starting point. */ 2975 pgsz <<= 2; 2976 pgs[idx++] = pgsz; 2977 mapped += pgsz; 2978 } 2979 } 2980 2981 nents = idx; 2982 mask = pgs[0] - 1; 2983 /* Align address to the boundary */ 2984 if (va & mask) { 2985 va = (va + mask) & ~mask; 2986 pa = (pa + mask) & ~mask; 2987 } 2988 2989 for (idx = 0; idx < nents; idx++) { 2990 pgsz = pgs[idx]; 2991 debugf("%u: %x -> %x, size=%x\n", idx, pa, va, pgsz); 2992 tlb1_set_entry(va, pa, pgsz, _TLB_ENTRY_MEM); 2993 pa += pgsz; 2994 va += pgsz; 2995 } 2996 2997 mapped = (va - base); 2998 debugf("mapped size 0x%08x (wasted space 0x%08x)\n", 2999 mapped, mapped - size); 3000 return (mapped); 3001 } 3002 3003 /* 3004 * TLB1 initialization routine, to be called after the very first 3005 * assembler level setup done in locore.S. 3006 */ 3007 void 3008 tlb1_init(vm_offset_t ccsrbar) 3009 { 3010 uint32_t mas0, mas1, mas3; 3011 uint32_t tsz; 3012 u_int i; 3013 3014 if (bootinfo != NULL && bootinfo[0] != 1) { 3015 tlb1_idx = *((uint16_t *)(bootinfo + 8)); 3016 } else 3017 tlb1_idx = 1; 3018 3019 /* The first entry/entries are used to map the kernel. */ 3020 for (i = 0; i < tlb1_idx; i++) { 3021 mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i); 3022 mtspr(SPR_MAS0, mas0); 3023 __asm __volatile("isync; tlbre"); 3024 3025 mas1 = mfspr(SPR_MAS1); 3026 if ((mas1 & MAS1_VALID) == 0) 3027 continue; 3028 3029 mas3 = mfspr(SPR_MAS3); 3030 3031 tlb1[i].mas1 = mas1; 3032 tlb1[i].mas2 = mfspr(SPR_MAS2); 3033 tlb1[i].mas3 = mas3; 3034 3035 if (i == 0) 3036 kernload = mas3 & MAS3_RPN; 3037 3038 tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT; 3039 kernsize += (tsz > 0) ? tsize2size(tsz) : 0; 3040 } 3041 3042 /* Map in CCSRBAR. */ 3043 tlb1_set_entry(CCSRBAR_VA, ccsrbar, CCSRBAR_SIZE, _TLB_ENTRY_IO); 3044 3045 /* Purge the remaining entries */ 3046 for (i = tlb1_idx; i < TLB1_ENTRIES; i++) 3047 tlb1_write_entry(i); 3048 3049 /* Setup TLB miss defaults */ 3050 set_mas4_defaults(); 3051 } 3052 3053 /* 3054 * Setup MAS4 defaults. 3055 * These values are loaded to MAS0-2 on a TLB miss. 3056 */ 3057 static void 3058 set_mas4_defaults(void) 3059 { 3060 uint32_t mas4; 3061 3062 /* Defaults: TLB0, PID0, TSIZED=4K */ 3063 mas4 = MAS4_TLBSELD0; 3064 mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK; 3065 #ifdef SMP 3066 mas4 |= MAS4_MD; 3067 #endif 3068 mtspr(SPR_MAS4, mas4); 3069 __asm __volatile("isync"); 3070 } 3071 3072 /* 3073 * Print out contents of the MAS registers for each TLB1 entry 3074 */ 3075 void 3076 tlb1_print_tlbentries(void) 3077 { 3078 uint32_t mas0, mas1, mas2, mas3, mas7; 3079 int i; 3080 3081 debugf("TLB1 entries:\n"); 3082 for (i = 0; i < TLB1_ENTRIES; i++) { 3083 3084 mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i); 3085 mtspr(SPR_MAS0, mas0); 3086 3087 __asm __volatile("isync; tlbre"); 3088 3089 mas1 = mfspr(SPR_MAS1); 3090 mas2 = mfspr(SPR_MAS2); 3091 mas3 = mfspr(SPR_MAS3); 3092 mas7 = mfspr(SPR_MAS7); 3093 3094 tlb_print_entry(i, mas1, mas2, mas3, mas7); 3095 } 3096 } 3097 3098 /* 3099 * Print out contents of the in-ram tlb1 table. 3100 */ 3101 void 3102 tlb1_print_entries(void) 3103 { 3104 int i; 3105 3106 debugf("tlb1[] table entries:\n"); 3107 for (i = 0; i < TLB1_ENTRIES; i++) 3108 tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0); 3109 } 3110 3111 /* 3112 * Return 0 if the physical IO range is encompassed by one of the 3113 * the TLB1 entries, otherwise return related error code. 3114 */ 3115 static int 3116 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va) 3117 { 3118 uint32_t prot; 3119 vm_paddr_t pa_start; 3120 vm_paddr_t pa_end; 3121 unsigned int entry_tsize; 3122 vm_size_t entry_size; 3123 3124 *va = (vm_offset_t)NULL; 3125 3126 /* Skip invalid entries */ 3127 if (!(tlb1[i].mas1 & MAS1_VALID)) 3128 return (EINVAL); 3129 3130 /* 3131 * The entry must be cache-inhibited, guarded, and r/w 3132 * so it can function as an i/o page 3133 */ 3134 prot = tlb1[i].mas2 & (MAS2_I | MAS2_G); 3135 if (prot != (MAS2_I | MAS2_G)) 3136 return (EPERM); 3137 3138 prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW); 3139 if (prot != (MAS3_SR | MAS3_SW)) 3140 return (EPERM); 3141 3142 /* The address should be within the entry range. */ 3143 entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT; 3144 KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize")); 3145 3146 entry_size = tsize2size(entry_tsize); 3147 pa_start = tlb1[i].mas3 & MAS3_RPN; 3148 pa_end = pa_start + entry_size - 1; 3149 3150 if ((pa < pa_start) || ((pa + size) > pa_end)) 3151 return (ERANGE); 3152 3153 /* Return virtual address of this mapping. */ 3154 *va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start); 3155 return (0); 3156 } 3157