1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD AND BSD-4-Clause 3 * 4 * Copyright (c) 2001 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Matt Thomas <matt@3am-software.com> of Allegro Networks, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 /*- 32 * Copyright (C) 1995, 1996 Wolfgang Solfrank. 33 * Copyright (C) 1995, 1996 TooLs GmbH. 34 * All rights reserved. 35 * 36 * Redistribution and use in source and binary forms, with or without 37 * modification, are permitted provided that the following conditions 38 * are met: 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. All advertising materials mentioning features or use of this software 45 * must display the following acknowledgement: 46 * This product includes software developed by TooLs GmbH. 47 * 4. The name of TooLs GmbH may not be used to endorse or promote products 48 * derived from this software without specific prior written permission. 49 * 50 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR 51 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 52 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 53 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 54 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 55 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 56 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 57 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 58 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 59 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 60 * 61 * $NetBSD: pmap.c,v 1.28 2000/03/26 20:42:36 kleink Exp $ 62 */ 63 /*- 64 * Copyright (C) 2001 Benno Rice. 65 * All rights reserved. 66 * 67 * Redistribution and use in source and binary forms, with or without 68 * modification, are permitted provided that the following conditions 69 * are met: 70 * 1. Redistributions of source code must retain the above copyright 71 * notice, this list of conditions and the following disclaimer. 72 * 2. Redistributions in binary form must reproduce the above copyright 73 * notice, this list of conditions and the following disclaimer in the 74 * documentation and/or other materials provided with the distribution. 75 * 76 * THIS SOFTWARE IS PROVIDED BY Benno Rice ``AS IS'' AND ANY EXPRESS OR 77 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 78 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 79 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 80 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 81 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 82 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 83 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 84 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 85 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 86 */ 87 88 #include <sys/cdefs.h> 89 __FBSDID("$FreeBSD$"); 90 91 /* 92 * Manages physical address maps. 93 * 94 * Since the information managed by this module is also stored by the 95 * logical address mapping module, this module may throw away valid virtual 96 * to physical mappings at almost any time. However, invalidations of 97 * mappings must be done as requested. 98 * 99 * In order to cope with hardware architectures which make virtual to 100 * physical map invalidates expensive, this module may delay invalidate 101 * reduced protection operations until such time as they are actually 102 * necessary. This module is given full information as to which processors 103 * are currently using which maps, and to when physical maps must be made 104 * correct. 105 */ 106 107 #include "opt_kstack_pages.h" 108 109 #include <sys/param.h> 110 #include <sys/kernel.h> 111 #include <sys/conf.h> 112 #include <sys/queue.h> 113 #include <sys/cpuset.h> 114 #include <sys/kerneldump.h> 115 #include <sys/ktr.h> 116 #include <sys/lock.h> 117 #include <sys/mman.h> 118 #include <sys/msgbuf.h> 119 #include <sys/mutex.h> 120 #include <sys/proc.h> 121 #include <sys/rwlock.h> 122 #include <sys/sched.h> 123 #include <sys/sysctl.h> 124 #include <sys/systm.h> 125 #include <sys/vmmeter.h> 126 127 #include <dev/ofw/openfirm.h> 128 129 #include <vm/vm.h> 130 #include <vm/pmap.h> 131 #include <vm/vm_param.h> 132 #include <vm/vm_kern.h> 133 #include <vm/vm_page.h> 134 #include <vm/vm_map.h> 135 #include <vm/vm_object.h> 136 #include <vm/vm_extern.h> 137 #include <vm/vm_page.h> 138 #include <vm/vm_phys.h> 139 #include <vm/vm_pageout.h> 140 #include <vm/uma.h> 141 142 #include <machine/cpu.h> 143 #include <machine/platform.h> 144 #include <machine/bat.h> 145 #include <machine/frame.h> 146 #include <machine/md_var.h> 147 #include <machine/psl.h> 148 #include <machine/pte.h> 149 #include <machine/smp.h> 150 #include <machine/sr.h> 151 #include <machine/mmuvar.h> 152 #include <machine/trap.h> 153 154 #define MOEA_DEBUG 155 156 #define TODO panic("%s: not implemented", __func__); 157 158 #define VSID_MAKE(sr, hash) ((sr) | (((hash) & 0xfffff) << 4)) 159 #define VSID_TO_SR(vsid) ((vsid) & 0xf) 160 #define VSID_TO_HASH(vsid) (((vsid) >> 4) & 0xfffff) 161 162 /* Get physical address from PVO. */ 163 #define PVO_PADDR(pvo) ((pvo)->pvo_pte.pte.pte_lo & PTE_RPGN) 164 165 struct ofw_map { 166 vm_offset_t om_va; 167 vm_size_t om_len; 168 vm_offset_t om_pa; 169 u_int om_mode; 170 }; 171 172 extern unsigned char _etext[]; 173 extern unsigned char _end[]; 174 175 /* 176 * Map of physical memory regions. 177 */ 178 static struct mem_region *regions; 179 static struct mem_region *pregions; 180 static u_int phys_avail_count; 181 static int regions_sz, pregions_sz; 182 static struct ofw_map *translations; 183 184 /* 185 * Lock for the pteg and pvo tables. 186 */ 187 struct mtx moea_table_mutex; 188 struct mtx moea_vsid_mutex; 189 190 /* tlbie instruction synchronization */ 191 static struct mtx tlbie_mtx; 192 193 /* 194 * PTEG data. 195 */ 196 static struct pteg *moea_pteg_table; 197 u_int moea_pteg_count; 198 u_int moea_pteg_mask; 199 200 /* 201 * PVO data. 202 */ 203 struct pvo_head *moea_pvo_table; /* pvo entries by pteg index */ 204 struct pvo_head moea_pvo_kunmanaged = 205 LIST_HEAD_INITIALIZER(moea_pvo_kunmanaged); /* list of unmanaged pages */ 206 207 static struct rwlock_padalign pvh_global_lock; 208 209 uma_zone_t moea_upvo_zone; /* zone for pvo entries for unmanaged pages */ 210 uma_zone_t moea_mpvo_zone; /* zone for pvo entries for managed pages */ 211 212 #define BPVO_POOL_SIZE 32768 213 static struct pvo_entry *moea_bpvo_pool; 214 static int moea_bpvo_pool_index = 0; 215 216 #define VSID_NBPW (sizeof(u_int32_t) * 8) 217 static u_int moea_vsid_bitmap[NPMAPS / VSID_NBPW]; 218 219 static boolean_t moea_initialized = FALSE; 220 221 /* 222 * Statistics. 223 */ 224 u_int moea_pte_valid = 0; 225 u_int moea_pte_overflow = 0; 226 u_int moea_pte_replacements = 0; 227 u_int moea_pvo_entries = 0; 228 u_int moea_pvo_enter_calls = 0; 229 u_int moea_pvo_remove_calls = 0; 230 u_int moea_pte_spills = 0; 231 SYSCTL_INT(_machdep, OID_AUTO, moea_pte_valid, CTLFLAG_RD, &moea_pte_valid, 232 0, ""); 233 SYSCTL_INT(_machdep, OID_AUTO, moea_pte_overflow, CTLFLAG_RD, 234 &moea_pte_overflow, 0, ""); 235 SYSCTL_INT(_machdep, OID_AUTO, moea_pte_replacements, CTLFLAG_RD, 236 &moea_pte_replacements, 0, ""); 237 SYSCTL_INT(_machdep, OID_AUTO, moea_pvo_entries, CTLFLAG_RD, &moea_pvo_entries, 238 0, ""); 239 SYSCTL_INT(_machdep, OID_AUTO, moea_pvo_enter_calls, CTLFLAG_RD, 240 &moea_pvo_enter_calls, 0, ""); 241 SYSCTL_INT(_machdep, OID_AUTO, moea_pvo_remove_calls, CTLFLAG_RD, 242 &moea_pvo_remove_calls, 0, ""); 243 SYSCTL_INT(_machdep, OID_AUTO, moea_pte_spills, CTLFLAG_RD, 244 &moea_pte_spills, 0, ""); 245 246 /* 247 * Allocate physical memory for use in moea_bootstrap. 248 */ 249 static vm_offset_t moea_bootstrap_alloc(vm_size_t, u_int); 250 251 /* 252 * PTE calls. 253 */ 254 static int moea_pte_insert(u_int, struct pte *); 255 256 /* 257 * PVO calls. 258 */ 259 static int moea_pvo_enter(pmap_t, uma_zone_t, struct pvo_head *, 260 vm_offset_t, vm_paddr_t, u_int, int); 261 static void moea_pvo_remove(struct pvo_entry *, int); 262 static struct pvo_entry *moea_pvo_find_va(pmap_t, vm_offset_t, int *); 263 static struct pte *moea_pvo_to_pte(const struct pvo_entry *, int); 264 265 /* 266 * Utility routines. 267 */ 268 static int moea_enter_locked(pmap_t, vm_offset_t, vm_page_t, 269 vm_prot_t, u_int, int8_t); 270 static void moea_syncicache(vm_paddr_t, vm_size_t); 271 static boolean_t moea_query_bit(vm_page_t, int); 272 static u_int moea_clear_bit(vm_page_t, int); 273 static void moea_kremove(vm_offset_t); 274 int moea_pte_spill(vm_offset_t); 275 276 /* 277 * Kernel MMU interface 278 */ 279 void moea_clear_modify(vm_page_t); 280 void moea_copy_page(vm_page_t, vm_page_t); 281 void moea_copy_pages(vm_page_t *ma, vm_offset_t a_offset, 282 vm_page_t *mb, vm_offset_t b_offset, int xfersize); 283 int moea_enter(pmap_t, vm_offset_t, vm_page_t, vm_prot_t, u_int, 284 int8_t); 285 void moea_enter_object(pmap_t, vm_offset_t, vm_offset_t, vm_page_t, 286 vm_prot_t); 287 void moea_enter_quick(pmap_t, vm_offset_t, vm_page_t, vm_prot_t); 288 vm_paddr_t moea_extract(pmap_t, vm_offset_t); 289 vm_page_t moea_extract_and_hold(pmap_t, vm_offset_t, vm_prot_t); 290 void moea_init(void); 291 boolean_t moea_is_modified(vm_page_t); 292 boolean_t moea_is_prefaultable(pmap_t, vm_offset_t); 293 boolean_t moea_is_referenced(vm_page_t); 294 int moea_ts_referenced(vm_page_t); 295 vm_offset_t moea_map(vm_offset_t *, vm_paddr_t, vm_paddr_t, int); 296 static int moea_mincore(pmap_t, vm_offset_t, vm_paddr_t *); 297 boolean_t moea_page_exists_quick(pmap_t, vm_page_t); 298 void moea_page_init(vm_page_t); 299 int moea_page_wired_mappings(vm_page_t); 300 int moea_pinit(pmap_t); 301 void moea_pinit0(pmap_t); 302 void moea_protect(pmap_t, vm_offset_t, vm_offset_t, vm_prot_t); 303 void moea_qenter(vm_offset_t, vm_page_t *, int); 304 void moea_qremove(vm_offset_t, int); 305 void moea_release(pmap_t); 306 void moea_remove(pmap_t, vm_offset_t, vm_offset_t); 307 void moea_remove_all(vm_page_t); 308 void moea_remove_write(vm_page_t); 309 void moea_unwire(pmap_t, vm_offset_t, vm_offset_t); 310 void moea_zero_page(vm_page_t); 311 void moea_zero_page_area(vm_page_t, int, int); 312 void moea_activate(struct thread *); 313 void moea_deactivate(struct thread *); 314 void moea_cpu_bootstrap(int); 315 void moea_bootstrap(vm_offset_t, vm_offset_t); 316 void *moea_mapdev(vm_paddr_t, vm_size_t); 317 void *moea_mapdev_attr(vm_paddr_t, vm_size_t, vm_memattr_t); 318 void moea_unmapdev(void *, vm_size_t); 319 vm_paddr_t moea_kextract(vm_offset_t); 320 void moea_kenter_attr(vm_offset_t, vm_paddr_t, vm_memattr_t); 321 void moea_kenter(vm_offset_t, vm_paddr_t); 322 void moea_page_set_memattr(vm_page_t m, vm_memattr_t ma); 323 boolean_t moea_dev_direct_mapped(vm_paddr_t, vm_size_t); 324 static void moea_sync_icache(pmap_t, vm_offset_t, vm_size_t); 325 void moea_dumpsys_map(vm_paddr_t pa, size_t sz, void **va); 326 void moea_scan_init(void); 327 vm_offset_t moea_quick_enter_page(vm_page_t m); 328 void moea_quick_remove_page(vm_offset_t addr); 329 boolean_t moea_page_is_mapped(vm_page_t m); 330 bool moea_ps_enabled(pmap_t pmap); 331 static int moea_map_user_ptr(pmap_t pm, 332 volatile const void *uaddr, void **kaddr, size_t ulen, size_t *klen); 333 static int moea_decode_kernel_ptr(vm_offset_t addr, 334 int *is_user, vm_offset_t *decoded_addr); 335 336 static struct pmap_funcs moea_methods = { 337 .clear_modify = moea_clear_modify, 338 .copy_page = moea_copy_page, 339 .copy_pages = moea_copy_pages, 340 .enter = moea_enter, 341 .enter_object = moea_enter_object, 342 .enter_quick = moea_enter_quick, 343 .extract = moea_extract, 344 .extract_and_hold = moea_extract_and_hold, 345 .init = moea_init, 346 .is_modified = moea_is_modified, 347 .is_prefaultable = moea_is_prefaultable, 348 .is_referenced = moea_is_referenced, 349 .ts_referenced = moea_ts_referenced, 350 .map = moea_map, 351 .page_exists_quick = moea_page_exists_quick, 352 .page_init = moea_page_init, 353 .page_wired_mappings = moea_page_wired_mappings, 354 .pinit = moea_pinit, 355 .pinit0 = moea_pinit0, 356 .protect = moea_protect, 357 .qenter = moea_qenter, 358 .qremove = moea_qremove, 359 .release = moea_release, 360 .remove = moea_remove, 361 .remove_all = moea_remove_all, 362 .mincore = moea_mincore, 363 .remove_write = moea_remove_write, 364 .sync_icache = moea_sync_icache, 365 .unwire = moea_unwire, 366 .zero_page = moea_zero_page, 367 .zero_page_area = moea_zero_page_area, 368 .activate = moea_activate, 369 .deactivate = moea_deactivate, 370 .page_set_memattr = moea_page_set_memattr, 371 .quick_enter_page = moea_quick_enter_page, 372 .quick_remove_page = moea_quick_remove_page, 373 .page_is_mapped = moea_page_is_mapped, 374 .ps_enabled = moea_ps_enabled, 375 376 /* Internal interfaces */ 377 .bootstrap = moea_bootstrap, 378 .cpu_bootstrap = moea_cpu_bootstrap, 379 .mapdev_attr = moea_mapdev_attr, 380 .mapdev = moea_mapdev, 381 .unmapdev = moea_unmapdev, 382 .kextract = moea_kextract, 383 .kenter = moea_kenter, 384 .kenter_attr = moea_kenter_attr, 385 .dev_direct_mapped = moea_dev_direct_mapped, 386 .dumpsys_pa_init = moea_scan_init, 387 .dumpsys_map_chunk = moea_dumpsys_map, 388 .map_user_ptr = moea_map_user_ptr, 389 .decode_kernel_ptr = moea_decode_kernel_ptr, 390 }; 391 392 MMU_DEF(oea_mmu, MMU_TYPE_OEA, moea_methods); 393 394 static __inline uint32_t 395 moea_calc_wimg(vm_paddr_t pa, vm_memattr_t ma) 396 { 397 uint32_t pte_lo; 398 int i; 399 400 if (ma != VM_MEMATTR_DEFAULT) { 401 switch (ma) { 402 case VM_MEMATTR_UNCACHEABLE: 403 return (PTE_I | PTE_G); 404 case VM_MEMATTR_CACHEABLE: 405 return (PTE_M); 406 case VM_MEMATTR_WRITE_COMBINING: 407 case VM_MEMATTR_WRITE_BACK: 408 case VM_MEMATTR_PREFETCHABLE: 409 return (PTE_I); 410 case VM_MEMATTR_WRITE_THROUGH: 411 return (PTE_W | PTE_M); 412 } 413 } 414 415 /* 416 * Assume the page is cache inhibited and access is guarded unless 417 * it's in our available memory array. 418 */ 419 pte_lo = PTE_I | PTE_G; 420 for (i = 0; i < pregions_sz; i++) { 421 if ((pa >= pregions[i].mr_start) && 422 (pa < (pregions[i].mr_start + pregions[i].mr_size))) { 423 pte_lo = PTE_M; 424 break; 425 } 426 } 427 428 return pte_lo; 429 } 430 431 /* 432 * Translate OFW translations into VM attributes. 433 */ 434 static __inline vm_memattr_t 435 moea_bootstrap_convert_wimg(uint32_t mode) 436 { 437 438 switch (mode) { 439 case (PTE_I | PTE_G): 440 /* PCI device memory */ 441 return VM_MEMATTR_UNCACHEABLE; 442 case (PTE_M): 443 /* Explicitly coherent */ 444 return VM_MEMATTR_CACHEABLE; 445 case 0: /* Default claim */ 446 case 2: /* Alternate PP bits set by OF for the original payload */ 447 /* "Normal" memory. */ 448 return VM_MEMATTR_DEFAULT; 449 450 default: 451 /* Err on the side of caution for unknowns */ 452 /* XXX should we panic instead? */ 453 return VM_MEMATTR_UNCACHEABLE; 454 } 455 } 456 457 static void 458 tlbie(vm_offset_t va) 459 { 460 461 mtx_lock_spin(&tlbie_mtx); 462 __asm __volatile("ptesync"); 463 __asm __volatile("tlbie %0" :: "r"(va)); 464 __asm __volatile("eieio; tlbsync; ptesync"); 465 mtx_unlock_spin(&tlbie_mtx); 466 } 467 468 static void 469 tlbia(void) 470 { 471 vm_offset_t va; 472 473 for (va = 0; va < 0x00040000; va += 0x00001000) { 474 __asm __volatile("tlbie %0" :: "r"(va)); 475 powerpc_sync(); 476 } 477 __asm __volatile("tlbsync"); 478 powerpc_sync(); 479 } 480 481 static __inline int 482 va_to_sr(u_int *sr, vm_offset_t va) 483 { 484 return (sr[(uintptr_t)va >> ADDR_SR_SHFT]); 485 } 486 487 static __inline u_int 488 va_to_pteg(u_int sr, vm_offset_t addr) 489 { 490 u_int hash; 491 492 hash = (sr & SR_VSID_MASK) ^ (((u_int)addr & ADDR_PIDX) >> 493 ADDR_PIDX_SHFT); 494 return (hash & moea_pteg_mask); 495 } 496 497 static __inline struct pvo_head * 498 vm_page_to_pvoh(vm_page_t m) 499 { 500 501 return (&m->md.mdpg_pvoh); 502 } 503 504 static __inline void 505 moea_attr_clear(vm_page_t m, int ptebit) 506 { 507 508 rw_assert(&pvh_global_lock, RA_WLOCKED); 509 m->md.mdpg_attrs &= ~ptebit; 510 } 511 512 static __inline int 513 moea_attr_fetch(vm_page_t m) 514 { 515 516 return (m->md.mdpg_attrs); 517 } 518 519 static __inline void 520 moea_attr_save(vm_page_t m, int ptebit) 521 { 522 523 rw_assert(&pvh_global_lock, RA_WLOCKED); 524 m->md.mdpg_attrs |= ptebit; 525 } 526 527 static __inline int 528 moea_pte_compare(const struct pte *pt, const struct pte *pvo_pt) 529 { 530 if (pt->pte_hi == pvo_pt->pte_hi) 531 return (1); 532 533 return (0); 534 } 535 536 static __inline int 537 moea_pte_match(struct pte *pt, u_int sr, vm_offset_t va, int which) 538 { 539 return (pt->pte_hi & ~PTE_VALID) == 540 (((sr & SR_VSID_MASK) << PTE_VSID_SHFT) | 541 ((va >> ADDR_API_SHFT) & PTE_API) | which); 542 } 543 544 static __inline void 545 moea_pte_create(struct pte *pt, u_int sr, vm_offset_t va, u_int pte_lo) 546 { 547 548 mtx_assert(&moea_table_mutex, MA_OWNED); 549 550 /* 551 * Construct a PTE. Default to IMB initially. Valid bit only gets 552 * set when the real pte is set in memory. 553 * 554 * Note: Don't set the valid bit for correct operation of tlb update. 555 */ 556 pt->pte_hi = ((sr & SR_VSID_MASK) << PTE_VSID_SHFT) | 557 (((va & ADDR_PIDX) >> ADDR_API_SHFT) & PTE_API); 558 pt->pte_lo = pte_lo; 559 } 560 561 static __inline void 562 moea_pte_synch(struct pte *pt, struct pte *pvo_pt) 563 { 564 565 mtx_assert(&moea_table_mutex, MA_OWNED); 566 pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF | PTE_CHG); 567 } 568 569 static __inline void 570 moea_pte_clear(struct pte *pt, vm_offset_t va, int ptebit) 571 { 572 573 mtx_assert(&moea_table_mutex, MA_OWNED); 574 575 /* 576 * As shown in Section 7.6.3.2.3 577 */ 578 pt->pte_lo &= ~ptebit; 579 tlbie(va); 580 } 581 582 static __inline void 583 moea_pte_set(struct pte *pt, struct pte *pvo_pt) 584 { 585 586 mtx_assert(&moea_table_mutex, MA_OWNED); 587 pvo_pt->pte_hi |= PTE_VALID; 588 589 /* 590 * Update the PTE as defined in section 7.6.3.1. 591 * Note that the REF/CHG bits are from pvo_pt and thus should have 592 * been saved so this routine can restore them (if desired). 593 */ 594 pt->pte_lo = pvo_pt->pte_lo; 595 powerpc_sync(); 596 pt->pte_hi = pvo_pt->pte_hi; 597 powerpc_sync(); 598 moea_pte_valid++; 599 } 600 601 static __inline void 602 moea_pte_unset(struct pte *pt, struct pte *pvo_pt, vm_offset_t va) 603 { 604 605 mtx_assert(&moea_table_mutex, MA_OWNED); 606 pvo_pt->pte_hi &= ~PTE_VALID; 607 608 /* 609 * Force the reg & chg bits back into the PTEs. 610 */ 611 powerpc_sync(); 612 613 /* 614 * Invalidate the pte. 615 */ 616 pt->pte_hi &= ~PTE_VALID; 617 618 tlbie(va); 619 620 /* 621 * Save the reg & chg bits. 622 */ 623 moea_pte_synch(pt, pvo_pt); 624 moea_pte_valid--; 625 } 626 627 static __inline void 628 moea_pte_change(struct pte *pt, struct pte *pvo_pt, vm_offset_t va) 629 { 630 631 /* 632 * Invalidate the PTE 633 */ 634 moea_pte_unset(pt, pvo_pt, va); 635 moea_pte_set(pt, pvo_pt); 636 } 637 638 /* 639 * Quick sort callout for comparing memory regions. 640 */ 641 static int om_cmp(const void *a, const void *b); 642 643 static int 644 om_cmp(const void *a, const void *b) 645 { 646 const struct ofw_map *mapa; 647 const struct ofw_map *mapb; 648 649 mapa = a; 650 mapb = b; 651 if (mapa->om_pa < mapb->om_pa) 652 return (-1); 653 else if (mapa->om_pa > mapb->om_pa) 654 return (1); 655 else 656 return (0); 657 } 658 659 void 660 moea_cpu_bootstrap(int ap) 661 { 662 u_int sdr; 663 int i; 664 665 if (ap) { 666 powerpc_sync(); 667 __asm __volatile("mtdbatu 0,%0" :: "r"(battable[0].batu)); 668 __asm __volatile("mtdbatl 0,%0" :: "r"(battable[0].batl)); 669 isync(); 670 __asm __volatile("mtibatu 0,%0" :: "r"(battable[0].batu)); 671 __asm __volatile("mtibatl 0,%0" :: "r"(battable[0].batl)); 672 isync(); 673 } 674 675 __asm __volatile("mtdbatu 1,%0" :: "r"(battable[8].batu)); 676 __asm __volatile("mtdbatl 1,%0" :: "r"(battable[8].batl)); 677 isync(); 678 679 __asm __volatile("mtibatu 1,%0" :: "r"(0)); 680 __asm __volatile("mtdbatu 2,%0" :: "r"(0)); 681 __asm __volatile("mtibatu 2,%0" :: "r"(0)); 682 __asm __volatile("mtdbatu 3,%0" :: "r"(0)); 683 __asm __volatile("mtibatu 3,%0" :: "r"(0)); 684 isync(); 685 686 for (i = 0; i < 16; i++) 687 mtsrin(i << ADDR_SR_SHFT, kernel_pmap->pm_sr[i]); 688 powerpc_sync(); 689 690 sdr = (u_int)moea_pteg_table | (moea_pteg_mask >> 10); 691 __asm __volatile("mtsdr1 %0" :: "r"(sdr)); 692 isync(); 693 694 tlbia(); 695 } 696 697 void 698 moea_bootstrap(vm_offset_t kernelstart, vm_offset_t kernelend) 699 { 700 ihandle_t mmui; 701 phandle_t chosen, mmu; 702 int sz; 703 int i, j; 704 vm_size_t size, physsz, hwphyssz; 705 vm_offset_t pa, va, off; 706 void *dpcpu; 707 708 /* 709 * Map PCI memory space. 710 */ 711 battable[0x8].batl = BATL(0x80000000, BAT_I|BAT_G, BAT_PP_RW); 712 battable[0x8].batu = BATU(0x80000000, BAT_BL_256M, BAT_Vs); 713 714 battable[0x9].batl = BATL(0x90000000, BAT_I|BAT_G, BAT_PP_RW); 715 battable[0x9].batu = BATU(0x90000000, BAT_BL_256M, BAT_Vs); 716 717 battable[0xa].batl = BATL(0xa0000000, BAT_I|BAT_G, BAT_PP_RW); 718 battable[0xa].batu = BATU(0xa0000000, BAT_BL_256M, BAT_Vs); 719 720 battable[0xb].batl = BATL(0xb0000000, BAT_I|BAT_G, BAT_PP_RW); 721 battable[0xb].batu = BATU(0xb0000000, BAT_BL_256M, BAT_Vs); 722 723 powerpc_sync(); 724 725 /* map pci space */ 726 __asm __volatile("mtdbatu 1,%0" :: "r"(battable[8].batu)); 727 __asm __volatile("mtdbatl 1,%0" :: "r"(battable[8].batl)); 728 isync(); 729 730 /* set global direct map flag */ 731 hw_direct_map = 1; 732 733 mem_regions(&pregions, &pregions_sz, ®ions, ®ions_sz); 734 CTR0(KTR_PMAP, "moea_bootstrap: physical memory"); 735 736 for (i = 0; i < pregions_sz; i++) { 737 vm_offset_t pa; 738 vm_offset_t end; 739 740 CTR3(KTR_PMAP, "physregion: %#x - %#x (%#x)", 741 pregions[i].mr_start, 742 pregions[i].mr_start + pregions[i].mr_size, 743 pregions[i].mr_size); 744 /* 745 * Install entries into the BAT table to allow all 746 * of physmem to be convered by on-demand BAT entries. 747 * The loop will sometimes set the same battable element 748 * twice, but that's fine since they won't be used for 749 * a while yet. 750 */ 751 pa = pregions[i].mr_start & 0xf0000000; 752 end = pregions[i].mr_start + pregions[i].mr_size; 753 do { 754 u_int n = pa >> ADDR_SR_SHFT; 755 756 battable[n].batl = BATL(pa, BAT_M, BAT_PP_RW); 757 battable[n].batu = BATU(pa, BAT_BL_256M, BAT_Vs); 758 pa += SEGMENT_LENGTH; 759 } while (pa < end); 760 } 761 762 if (PHYS_AVAIL_ENTRIES < regions_sz) 763 panic("moea_bootstrap: phys_avail too small"); 764 765 phys_avail_count = 0; 766 physsz = 0; 767 hwphyssz = 0; 768 TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz); 769 for (i = 0, j = 0; i < regions_sz; i++, j += 2) { 770 CTR3(KTR_PMAP, "region: %#x - %#x (%#x)", regions[i].mr_start, 771 regions[i].mr_start + regions[i].mr_size, 772 regions[i].mr_size); 773 if (hwphyssz != 0 && 774 (physsz + regions[i].mr_size) >= hwphyssz) { 775 if (physsz < hwphyssz) { 776 phys_avail[j] = regions[i].mr_start; 777 phys_avail[j + 1] = regions[i].mr_start + 778 hwphyssz - physsz; 779 physsz = hwphyssz; 780 phys_avail_count++; 781 } 782 break; 783 } 784 phys_avail[j] = regions[i].mr_start; 785 phys_avail[j + 1] = regions[i].mr_start + regions[i].mr_size; 786 phys_avail_count++; 787 physsz += regions[i].mr_size; 788 } 789 790 /* Check for overlap with the kernel and exception vectors */ 791 for (j = 0; j < 2*phys_avail_count; j+=2) { 792 if (phys_avail[j] < EXC_LAST) 793 phys_avail[j] += EXC_LAST; 794 795 if (kernelstart >= phys_avail[j] && 796 kernelstart < phys_avail[j+1]) { 797 if (kernelend < phys_avail[j+1]) { 798 phys_avail[2*phys_avail_count] = 799 (kernelend & ~PAGE_MASK) + PAGE_SIZE; 800 phys_avail[2*phys_avail_count + 1] = 801 phys_avail[j+1]; 802 phys_avail_count++; 803 } 804 805 phys_avail[j+1] = kernelstart & ~PAGE_MASK; 806 } 807 808 if (kernelend >= phys_avail[j] && 809 kernelend < phys_avail[j+1]) { 810 if (kernelstart > phys_avail[j]) { 811 phys_avail[2*phys_avail_count] = phys_avail[j]; 812 phys_avail[2*phys_avail_count + 1] = 813 kernelstart & ~PAGE_MASK; 814 phys_avail_count++; 815 } 816 817 phys_avail[j] = (kernelend & ~PAGE_MASK) + PAGE_SIZE; 818 } 819 } 820 821 physmem = btoc(physsz); 822 823 /* 824 * Allocate PTEG table. 825 */ 826 #ifdef PTEGCOUNT 827 moea_pteg_count = PTEGCOUNT; 828 #else 829 moea_pteg_count = 0x1000; 830 831 while (moea_pteg_count < physmem) 832 moea_pteg_count <<= 1; 833 834 moea_pteg_count >>= 1; 835 #endif /* PTEGCOUNT */ 836 837 size = moea_pteg_count * sizeof(struct pteg); 838 CTR2(KTR_PMAP, "moea_bootstrap: %d PTEGs, %d bytes", moea_pteg_count, 839 size); 840 moea_pteg_table = (struct pteg *)moea_bootstrap_alloc(size, size); 841 CTR1(KTR_PMAP, "moea_bootstrap: PTEG table at %p", moea_pteg_table); 842 bzero((void *)moea_pteg_table, moea_pteg_count * sizeof(struct pteg)); 843 moea_pteg_mask = moea_pteg_count - 1; 844 845 /* 846 * Allocate pv/overflow lists. 847 */ 848 size = sizeof(struct pvo_head) * moea_pteg_count; 849 moea_pvo_table = (struct pvo_head *)moea_bootstrap_alloc(size, 850 PAGE_SIZE); 851 CTR1(KTR_PMAP, "moea_bootstrap: PVO table at %p", moea_pvo_table); 852 for (i = 0; i < moea_pteg_count; i++) 853 LIST_INIT(&moea_pvo_table[i]); 854 855 /* 856 * Initialize the lock that synchronizes access to the pteg and pvo 857 * tables. 858 */ 859 mtx_init(&moea_table_mutex, "pmap table", NULL, MTX_DEF | 860 MTX_RECURSE); 861 mtx_init(&moea_vsid_mutex, "VSID table", NULL, MTX_DEF); 862 863 mtx_init(&tlbie_mtx, "tlbie", NULL, MTX_SPIN); 864 865 /* 866 * Initialise the unmanaged pvo pool. 867 */ 868 moea_bpvo_pool = (struct pvo_entry *)moea_bootstrap_alloc( 869 BPVO_POOL_SIZE*sizeof(struct pvo_entry), 0); 870 moea_bpvo_pool_index = 0; 871 872 /* 873 * Make sure kernel vsid is allocated as well as VSID 0. 874 */ 875 moea_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS - 1)) / VSID_NBPW] 876 |= 1 << (KERNEL_VSIDBITS % VSID_NBPW); 877 moea_vsid_bitmap[0] |= 1; 878 879 /* 880 * Initialize the kernel pmap (which is statically allocated). 881 */ 882 PMAP_LOCK_INIT(kernel_pmap); 883 for (i = 0; i < 16; i++) 884 kernel_pmap->pm_sr[i] = EMPTY_SEGMENT + i; 885 CPU_FILL(&kernel_pmap->pm_active); 886 RB_INIT(&kernel_pmap->pmap_pvo); 887 888 /* 889 * Initialize the global pv list lock. 890 */ 891 rw_init(&pvh_global_lock, "pmap pv global"); 892 893 /* 894 * Set up the Open Firmware mappings 895 */ 896 chosen = OF_finddevice("/chosen"); 897 if (chosen != -1 && OF_getprop(chosen, "mmu", &mmui, 4) != -1 && 898 (mmu = OF_instance_to_package(mmui)) != -1 && 899 (sz = OF_getproplen(mmu, "translations")) != -1) { 900 translations = NULL; 901 for (i = 0; phys_avail[i] != 0; i += 2) { 902 if (phys_avail[i + 1] >= sz) { 903 translations = (struct ofw_map *)phys_avail[i]; 904 break; 905 } 906 } 907 if (translations == NULL) 908 panic("moea_bootstrap: no space to copy translations"); 909 bzero(translations, sz); 910 if (OF_getprop(mmu, "translations", translations, sz) == -1) 911 panic("moea_bootstrap: can't get ofw translations"); 912 CTR0(KTR_PMAP, "moea_bootstrap: translations"); 913 sz /= sizeof(*translations); 914 qsort(translations, sz, sizeof (*translations), om_cmp); 915 for (i = 0; i < sz; i++) { 916 CTR3(KTR_PMAP, "translation: pa=%#x va=%#x len=%#x", 917 translations[i].om_pa, translations[i].om_va, 918 translations[i].om_len); 919 920 /* 921 * If the mapping is 1:1, let the RAM and device 922 * on-demand BAT tables take care of the translation. 923 * 924 * However, always enter mappings for segment 16, 925 * which is mixed-protection and therefore not 926 * compatible with a BAT entry. 927 */ 928 if ((translations[i].om_va >> ADDR_SR_SHFT) != 0xf && 929 translations[i].om_va == translations[i].om_pa) 930 continue; 931 932 /* Enter the pages */ 933 for (off = 0; off < translations[i].om_len; 934 off += PAGE_SIZE) 935 moea_kenter_attr(translations[i].om_va + off, 936 translations[i].om_pa + off, 937 moea_bootstrap_convert_wimg(translations[i].om_mode)); 938 } 939 } 940 941 /* 942 * Calculate the last available physical address. 943 */ 944 for (i = 0; phys_avail[i + 2] != 0; i += 2) 945 ; 946 Maxmem = powerpc_btop(phys_avail[i + 1]); 947 948 moea_cpu_bootstrap(0); 949 mtmsr(mfmsr() | PSL_DR | PSL_IR); 950 pmap_bootstrapped++; 951 952 /* 953 * Set the start and end of kva. 954 */ 955 virtual_avail = VM_MIN_KERNEL_ADDRESS; 956 virtual_end = VM_MAX_SAFE_KERNEL_ADDRESS; 957 958 /* 959 * Allocate a kernel stack with a guard page for thread0 and map it 960 * into the kernel page map. 961 */ 962 pa = moea_bootstrap_alloc(kstack_pages * PAGE_SIZE, PAGE_SIZE); 963 va = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE; 964 virtual_avail = va + kstack_pages * PAGE_SIZE; 965 CTR2(KTR_PMAP, "moea_bootstrap: kstack0 at %#x (%#x)", pa, va); 966 thread0.td_kstack = va; 967 thread0.td_kstack_pages = kstack_pages; 968 for (i = 0; i < kstack_pages; i++) { 969 moea_kenter(va, pa); 970 pa += PAGE_SIZE; 971 va += PAGE_SIZE; 972 } 973 974 /* 975 * Allocate virtual address space for the message buffer. 976 */ 977 pa = msgbuf_phys = moea_bootstrap_alloc(msgbufsize, PAGE_SIZE); 978 msgbufp = (struct msgbuf *)virtual_avail; 979 va = virtual_avail; 980 virtual_avail += round_page(msgbufsize); 981 while (va < virtual_avail) { 982 moea_kenter(va, pa); 983 pa += PAGE_SIZE; 984 va += PAGE_SIZE; 985 } 986 987 /* 988 * Allocate virtual address space for the dynamic percpu area. 989 */ 990 pa = moea_bootstrap_alloc(DPCPU_SIZE, PAGE_SIZE); 991 dpcpu = (void *)virtual_avail; 992 va = virtual_avail; 993 virtual_avail += DPCPU_SIZE; 994 while (va < virtual_avail) { 995 moea_kenter(va, pa); 996 pa += PAGE_SIZE; 997 va += PAGE_SIZE; 998 } 999 dpcpu_init(dpcpu, 0); 1000 } 1001 1002 /* 1003 * Activate a user pmap. The pmap must be activated before it's address 1004 * space can be accessed in any way. 1005 */ 1006 void 1007 moea_activate(struct thread *td) 1008 { 1009 pmap_t pm, pmr; 1010 1011 /* 1012 * Load all the data we need up front to encourage the compiler to 1013 * not issue any loads while we have interrupts disabled below. 1014 */ 1015 pm = &td->td_proc->p_vmspace->vm_pmap; 1016 pmr = pm->pmap_phys; 1017 1018 CPU_SET(PCPU_GET(cpuid), &pm->pm_active); 1019 PCPU_SET(curpmap, pmr); 1020 1021 mtsrin(USER_SR << ADDR_SR_SHFT, td->td_pcb->pcb_cpu.aim.usr_vsid); 1022 } 1023 1024 void 1025 moea_deactivate(struct thread *td) 1026 { 1027 pmap_t pm; 1028 1029 pm = &td->td_proc->p_vmspace->vm_pmap; 1030 CPU_CLR(PCPU_GET(cpuid), &pm->pm_active); 1031 PCPU_SET(curpmap, NULL); 1032 } 1033 1034 void 1035 moea_unwire(pmap_t pm, vm_offset_t sva, vm_offset_t eva) 1036 { 1037 struct pvo_entry key, *pvo; 1038 1039 PMAP_LOCK(pm); 1040 key.pvo_vaddr = sva; 1041 for (pvo = RB_NFIND(pvo_tree, &pm->pmap_pvo, &key); 1042 pvo != NULL && PVO_VADDR(pvo) < eva; 1043 pvo = RB_NEXT(pvo_tree, &pm->pmap_pvo, pvo)) { 1044 if ((pvo->pvo_vaddr & PVO_WIRED) == 0) 1045 panic("moea_unwire: pvo %p is missing PVO_WIRED", pvo); 1046 pvo->pvo_vaddr &= ~PVO_WIRED; 1047 pm->pm_stats.wired_count--; 1048 } 1049 PMAP_UNLOCK(pm); 1050 } 1051 1052 void 1053 moea_copy_page(vm_page_t msrc, vm_page_t mdst) 1054 { 1055 vm_offset_t dst; 1056 vm_offset_t src; 1057 1058 dst = VM_PAGE_TO_PHYS(mdst); 1059 src = VM_PAGE_TO_PHYS(msrc); 1060 1061 bcopy((void *)src, (void *)dst, PAGE_SIZE); 1062 } 1063 1064 void 1065 moea_copy_pages(vm_page_t *ma, vm_offset_t a_offset, 1066 vm_page_t *mb, vm_offset_t b_offset, int xfersize) 1067 { 1068 void *a_cp, *b_cp; 1069 vm_offset_t a_pg_offset, b_pg_offset; 1070 int cnt; 1071 1072 while (xfersize > 0) { 1073 a_pg_offset = a_offset & PAGE_MASK; 1074 cnt = min(xfersize, PAGE_SIZE - a_pg_offset); 1075 a_cp = (char *)VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]) + 1076 a_pg_offset; 1077 b_pg_offset = b_offset & PAGE_MASK; 1078 cnt = min(cnt, PAGE_SIZE - b_pg_offset); 1079 b_cp = (char *)VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]) + 1080 b_pg_offset; 1081 bcopy(a_cp, b_cp, cnt); 1082 a_offset += cnt; 1083 b_offset += cnt; 1084 xfersize -= cnt; 1085 } 1086 } 1087 1088 /* 1089 * Zero a page of physical memory by temporarily mapping it into the tlb. 1090 */ 1091 void 1092 moea_zero_page(vm_page_t m) 1093 { 1094 vm_offset_t off, pa = VM_PAGE_TO_PHYS(m); 1095 1096 for (off = 0; off < PAGE_SIZE; off += cacheline_size) 1097 __asm __volatile("dcbz 0,%0" :: "r"(pa + off)); 1098 } 1099 1100 void 1101 moea_zero_page_area(vm_page_t m, int off, int size) 1102 { 1103 vm_offset_t pa = VM_PAGE_TO_PHYS(m); 1104 void *va = (void *)(pa + off); 1105 1106 bzero(va, size); 1107 } 1108 1109 vm_offset_t 1110 moea_quick_enter_page(vm_page_t m) 1111 { 1112 1113 return (VM_PAGE_TO_PHYS(m)); 1114 } 1115 1116 void 1117 moea_quick_remove_page(vm_offset_t addr) 1118 { 1119 } 1120 1121 boolean_t 1122 moea_page_is_mapped(vm_page_t m) 1123 { 1124 return (!LIST_EMPTY(&(m)->md.mdpg_pvoh)); 1125 } 1126 1127 bool 1128 moea_ps_enabled(pmap_t pmap __unused) 1129 { 1130 return (false); 1131 } 1132 1133 /* 1134 * Map the given physical page at the specified virtual address in the 1135 * target pmap with the protection requested. If specified the page 1136 * will be wired down. 1137 */ 1138 int 1139 moea_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot, 1140 u_int flags, int8_t psind) 1141 { 1142 int error; 1143 1144 for (;;) { 1145 rw_wlock(&pvh_global_lock); 1146 PMAP_LOCK(pmap); 1147 error = moea_enter_locked(pmap, va, m, prot, flags, psind); 1148 rw_wunlock(&pvh_global_lock); 1149 PMAP_UNLOCK(pmap); 1150 if (error != ENOMEM) 1151 return (KERN_SUCCESS); 1152 if ((flags & PMAP_ENTER_NOSLEEP) != 0) 1153 return (KERN_RESOURCE_SHORTAGE); 1154 VM_OBJECT_ASSERT_UNLOCKED(m->object); 1155 vm_wait(NULL); 1156 } 1157 } 1158 1159 /* 1160 * Map the given physical page at the specified virtual address in the 1161 * target pmap with the protection requested. If specified the page 1162 * will be wired down. 1163 * 1164 * The global pvh and pmap must be locked. 1165 */ 1166 static int 1167 moea_enter_locked(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot, 1168 u_int flags, int8_t psind __unused) 1169 { 1170 struct pvo_head *pvo_head; 1171 uma_zone_t zone; 1172 u_int pte_lo, pvo_flags; 1173 int error; 1174 1175 if (pmap_bootstrapped) 1176 rw_assert(&pvh_global_lock, RA_WLOCKED); 1177 PMAP_LOCK_ASSERT(pmap, MA_OWNED); 1178 if ((m->oflags & VPO_UNMANAGED) == 0) { 1179 if ((flags & PMAP_ENTER_QUICK_LOCKED) == 0) 1180 VM_PAGE_OBJECT_BUSY_ASSERT(m); 1181 else 1182 VM_OBJECT_ASSERT_LOCKED(m->object); 1183 } 1184 1185 if ((m->oflags & VPO_UNMANAGED) != 0 || !moea_initialized) { 1186 pvo_head = &moea_pvo_kunmanaged; 1187 zone = moea_upvo_zone; 1188 pvo_flags = 0; 1189 } else { 1190 pvo_head = vm_page_to_pvoh(m); 1191 zone = moea_mpvo_zone; 1192 pvo_flags = PVO_MANAGED; 1193 } 1194 1195 pte_lo = moea_calc_wimg(VM_PAGE_TO_PHYS(m), pmap_page_get_memattr(m)); 1196 1197 if (prot & VM_PROT_WRITE) { 1198 pte_lo |= PTE_BW; 1199 if (pmap_bootstrapped && 1200 (m->oflags & VPO_UNMANAGED) == 0) 1201 vm_page_aflag_set(m, PGA_WRITEABLE); 1202 } else 1203 pte_lo |= PTE_BR; 1204 1205 if ((flags & PMAP_ENTER_WIRED) != 0) 1206 pvo_flags |= PVO_WIRED; 1207 1208 error = moea_pvo_enter(pmap, zone, pvo_head, va, VM_PAGE_TO_PHYS(m), 1209 pte_lo, pvo_flags); 1210 1211 /* 1212 * Flush the real page from the instruction cache. This has be done 1213 * for all user mappings to prevent information leakage via the 1214 * instruction cache. moea_pvo_enter() returns ENOENT for the first 1215 * mapping for a page. 1216 */ 1217 if (pmap != kernel_pmap && error == ENOENT && 1218 (pte_lo & (PTE_I | PTE_G)) == 0) 1219 moea_syncicache(VM_PAGE_TO_PHYS(m), PAGE_SIZE); 1220 1221 return (error); 1222 } 1223 1224 /* 1225 * Maps a sequence of resident pages belonging to the same object. 1226 * The sequence begins with the given page m_start. This page is 1227 * mapped at the given virtual address start. Each subsequent page is 1228 * mapped at a virtual address that is offset from start by the same 1229 * amount as the page is offset from m_start within the object. The 1230 * last page in the sequence is the page with the largest offset from 1231 * m_start that can be mapped at a virtual address less than the given 1232 * virtual address end. Not every virtual page between start and end 1233 * is mapped; only those for which a resident page exists with the 1234 * corresponding offset from m_start are mapped. 1235 */ 1236 void 1237 moea_enter_object(pmap_t pm, vm_offset_t start, vm_offset_t end, 1238 vm_page_t m_start, vm_prot_t prot) 1239 { 1240 vm_page_t m; 1241 vm_pindex_t diff, psize; 1242 1243 VM_OBJECT_ASSERT_LOCKED(m_start->object); 1244 1245 psize = atop(end - start); 1246 m = m_start; 1247 rw_wlock(&pvh_global_lock); 1248 PMAP_LOCK(pm); 1249 while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) { 1250 moea_enter_locked(pm, start + ptoa(diff), m, prot & 1251 (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_QUICK_LOCKED, 1252 0); 1253 m = TAILQ_NEXT(m, listq); 1254 } 1255 rw_wunlock(&pvh_global_lock); 1256 PMAP_UNLOCK(pm); 1257 } 1258 1259 void 1260 moea_enter_quick(pmap_t pm, vm_offset_t va, vm_page_t m, 1261 vm_prot_t prot) 1262 { 1263 1264 rw_wlock(&pvh_global_lock); 1265 PMAP_LOCK(pm); 1266 moea_enter_locked(pm, va, m, prot & (VM_PROT_READ | VM_PROT_EXECUTE), 1267 PMAP_ENTER_QUICK_LOCKED, 0); 1268 rw_wunlock(&pvh_global_lock); 1269 PMAP_UNLOCK(pm); 1270 } 1271 1272 vm_paddr_t 1273 moea_extract(pmap_t pm, vm_offset_t va) 1274 { 1275 struct pvo_entry *pvo; 1276 vm_paddr_t pa; 1277 1278 PMAP_LOCK(pm); 1279 pvo = moea_pvo_find_va(pm, va & ~ADDR_POFF, NULL); 1280 if (pvo == NULL) 1281 pa = 0; 1282 else 1283 pa = PVO_PADDR(pvo) | (va & ADDR_POFF); 1284 PMAP_UNLOCK(pm); 1285 return (pa); 1286 } 1287 1288 /* 1289 * Atomically extract and hold the physical page with the given 1290 * pmap and virtual address pair if that mapping permits the given 1291 * protection. 1292 */ 1293 vm_page_t 1294 moea_extract_and_hold(pmap_t pmap, vm_offset_t va, vm_prot_t prot) 1295 { 1296 struct pvo_entry *pvo; 1297 vm_page_t m; 1298 1299 m = NULL; 1300 PMAP_LOCK(pmap); 1301 pvo = moea_pvo_find_va(pmap, va & ~ADDR_POFF, NULL); 1302 if (pvo != NULL && (pvo->pvo_pte.pte.pte_hi & PTE_VALID) && 1303 ((pvo->pvo_pte.pte.pte_lo & PTE_PP) == PTE_RW || 1304 (prot & VM_PROT_WRITE) == 0)) { 1305 m = PHYS_TO_VM_PAGE(PVO_PADDR(pvo)); 1306 if (!vm_page_wire_mapped(m)) 1307 m = NULL; 1308 } 1309 PMAP_UNLOCK(pmap); 1310 return (m); 1311 } 1312 1313 void 1314 moea_init() 1315 { 1316 1317 moea_upvo_zone = uma_zcreate("UPVO entry", sizeof (struct pvo_entry), 1318 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1319 UMA_ZONE_VM | UMA_ZONE_NOFREE); 1320 moea_mpvo_zone = uma_zcreate("MPVO entry", sizeof(struct pvo_entry), 1321 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 1322 UMA_ZONE_VM | UMA_ZONE_NOFREE); 1323 moea_initialized = TRUE; 1324 } 1325 1326 boolean_t 1327 moea_is_referenced(vm_page_t m) 1328 { 1329 boolean_t rv; 1330 1331 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 1332 ("moea_is_referenced: page %p is not managed", m)); 1333 rw_wlock(&pvh_global_lock); 1334 rv = moea_query_bit(m, PTE_REF); 1335 rw_wunlock(&pvh_global_lock); 1336 return (rv); 1337 } 1338 1339 boolean_t 1340 moea_is_modified(vm_page_t m) 1341 { 1342 boolean_t rv; 1343 1344 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 1345 ("moea_is_modified: page %p is not managed", m)); 1346 1347 /* 1348 * If the page is not busied then this check is racy. 1349 */ 1350 if (!pmap_page_is_write_mapped(m)) 1351 return (FALSE); 1352 1353 rw_wlock(&pvh_global_lock); 1354 rv = moea_query_bit(m, PTE_CHG); 1355 rw_wunlock(&pvh_global_lock); 1356 return (rv); 1357 } 1358 1359 boolean_t 1360 moea_is_prefaultable(pmap_t pmap, vm_offset_t va) 1361 { 1362 struct pvo_entry *pvo; 1363 boolean_t rv; 1364 1365 PMAP_LOCK(pmap); 1366 pvo = moea_pvo_find_va(pmap, va & ~ADDR_POFF, NULL); 1367 rv = pvo == NULL || (pvo->pvo_pte.pte.pte_hi & PTE_VALID) == 0; 1368 PMAP_UNLOCK(pmap); 1369 return (rv); 1370 } 1371 1372 void 1373 moea_clear_modify(vm_page_t m) 1374 { 1375 1376 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 1377 ("moea_clear_modify: page %p is not managed", m)); 1378 vm_page_assert_busied(m); 1379 1380 if (!pmap_page_is_write_mapped(m)) 1381 return; 1382 rw_wlock(&pvh_global_lock); 1383 moea_clear_bit(m, PTE_CHG); 1384 rw_wunlock(&pvh_global_lock); 1385 } 1386 1387 /* 1388 * Clear the write and modified bits in each of the given page's mappings. 1389 */ 1390 void 1391 moea_remove_write(vm_page_t m) 1392 { 1393 struct pvo_entry *pvo; 1394 struct pte *pt; 1395 pmap_t pmap; 1396 u_int lo; 1397 1398 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 1399 ("moea_remove_write: page %p is not managed", m)); 1400 vm_page_assert_busied(m); 1401 1402 if (!pmap_page_is_write_mapped(m)) 1403 return; 1404 rw_wlock(&pvh_global_lock); 1405 lo = moea_attr_fetch(m); 1406 powerpc_sync(); 1407 LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) { 1408 pmap = pvo->pvo_pmap; 1409 PMAP_LOCK(pmap); 1410 if ((pvo->pvo_pte.pte.pte_lo & PTE_PP) != PTE_BR) { 1411 pt = moea_pvo_to_pte(pvo, -1); 1412 pvo->pvo_pte.pte.pte_lo &= ~PTE_PP; 1413 pvo->pvo_pte.pte.pte_lo |= PTE_BR; 1414 if (pt != NULL) { 1415 moea_pte_synch(pt, &pvo->pvo_pte.pte); 1416 lo |= pvo->pvo_pte.pte.pte_lo; 1417 pvo->pvo_pte.pte.pte_lo &= ~PTE_CHG; 1418 moea_pte_change(pt, &pvo->pvo_pte.pte, 1419 pvo->pvo_vaddr); 1420 mtx_unlock(&moea_table_mutex); 1421 } 1422 } 1423 PMAP_UNLOCK(pmap); 1424 } 1425 if ((lo & PTE_CHG) != 0) { 1426 moea_attr_clear(m, PTE_CHG); 1427 vm_page_dirty(m); 1428 } 1429 vm_page_aflag_clear(m, PGA_WRITEABLE); 1430 rw_wunlock(&pvh_global_lock); 1431 } 1432 1433 /* 1434 * moea_ts_referenced: 1435 * 1436 * Return a count of reference bits for a page, clearing those bits. 1437 * It is not necessary for every reference bit to be cleared, but it 1438 * is necessary that 0 only be returned when there are truly no 1439 * reference bits set. 1440 * 1441 * XXX: The exact number of bits to check and clear is a matter that 1442 * should be tested and standardized at some point in the future for 1443 * optimal aging of shared pages. 1444 */ 1445 int 1446 moea_ts_referenced(vm_page_t m) 1447 { 1448 int count; 1449 1450 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 1451 ("moea_ts_referenced: page %p is not managed", m)); 1452 rw_wlock(&pvh_global_lock); 1453 count = moea_clear_bit(m, PTE_REF); 1454 rw_wunlock(&pvh_global_lock); 1455 return (count); 1456 } 1457 1458 /* 1459 * Modify the WIMG settings of all mappings for a page. 1460 */ 1461 void 1462 moea_page_set_memattr(vm_page_t m, vm_memattr_t ma) 1463 { 1464 struct pvo_entry *pvo; 1465 struct pvo_head *pvo_head; 1466 struct pte *pt; 1467 pmap_t pmap; 1468 u_int lo; 1469 1470 if ((m->oflags & VPO_UNMANAGED) != 0) { 1471 m->md.mdpg_cache_attrs = ma; 1472 return; 1473 } 1474 1475 rw_wlock(&pvh_global_lock); 1476 pvo_head = vm_page_to_pvoh(m); 1477 lo = moea_calc_wimg(VM_PAGE_TO_PHYS(m), ma); 1478 1479 LIST_FOREACH(pvo, pvo_head, pvo_vlink) { 1480 pmap = pvo->pvo_pmap; 1481 PMAP_LOCK(pmap); 1482 pt = moea_pvo_to_pte(pvo, -1); 1483 pvo->pvo_pte.pte.pte_lo &= ~PTE_WIMG; 1484 pvo->pvo_pte.pte.pte_lo |= lo; 1485 if (pt != NULL) { 1486 moea_pte_change(pt, &pvo->pvo_pte.pte, 1487 pvo->pvo_vaddr); 1488 if (pvo->pvo_pmap == kernel_pmap) 1489 isync(); 1490 } 1491 mtx_unlock(&moea_table_mutex); 1492 PMAP_UNLOCK(pmap); 1493 } 1494 m->md.mdpg_cache_attrs = ma; 1495 rw_wunlock(&pvh_global_lock); 1496 } 1497 1498 /* 1499 * Map a wired page into kernel virtual address space. 1500 */ 1501 void 1502 moea_kenter(vm_offset_t va, vm_paddr_t pa) 1503 { 1504 1505 moea_kenter_attr(va, pa, VM_MEMATTR_DEFAULT); 1506 } 1507 1508 void 1509 moea_kenter_attr(vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma) 1510 { 1511 u_int pte_lo; 1512 int error; 1513 1514 #if 0 1515 if (va < VM_MIN_KERNEL_ADDRESS) 1516 panic("moea_kenter: attempt to enter non-kernel address %#x", 1517 va); 1518 #endif 1519 1520 pte_lo = moea_calc_wimg(pa, ma); 1521 1522 PMAP_LOCK(kernel_pmap); 1523 error = moea_pvo_enter(kernel_pmap, moea_upvo_zone, 1524 &moea_pvo_kunmanaged, va, pa, pte_lo, PVO_WIRED); 1525 1526 if (error != 0 && error != ENOENT) 1527 panic("moea_kenter: failed to enter va %#x pa %#x: %d", va, 1528 pa, error); 1529 1530 PMAP_UNLOCK(kernel_pmap); 1531 } 1532 1533 /* 1534 * Extract the physical page address associated with the given kernel virtual 1535 * address. 1536 */ 1537 vm_paddr_t 1538 moea_kextract(vm_offset_t va) 1539 { 1540 struct pvo_entry *pvo; 1541 vm_paddr_t pa; 1542 1543 /* 1544 * Allow direct mappings on 32-bit OEA 1545 */ 1546 if (va < VM_MIN_KERNEL_ADDRESS) { 1547 return (va); 1548 } 1549 1550 PMAP_LOCK(kernel_pmap); 1551 pvo = moea_pvo_find_va(kernel_pmap, va & ~ADDR_POFF, NULL); 1552 KASSERT(pvo != NULL, ("moea_kextract: no addr found")); 1553 pa = PVO_PADDR(pvo) | (va & ADDR_POFF); 1554 PMAP_UNLOCK(kernel_pmap); 1555 return (pa); 1556 } 1557 1558 /* 1559 * Remove a wired page from kernel virtual address space. 1560 */ 1561 void 1562 moea_kremove(vm_offset_t va) 1563 { 1564 1565 moea_remove(kernel_pmap, va, va + PAGE_SIZE); 1566 } 1567 1568 /* 1569 * Provide a kernel pointer corresponding to a given userland pointer. 1570 * The returned pointer is valid until the next time this function is 1571 * called in this thread. This is used internally in copyin/copyout. 1572 */ 1573 int 1574 moea_map_user_ptr(pmap_t pm, volatile const void *uaddr, 1575 void **kaddr, size_t ulen, size_t *klen) 1576 { 1577 size_t l; 1578 register_t vsid; 1579 1580 *kaddr = (char *)USER_ADDR + ((uintptr_t)uaddr & ~SEGMENT_MASK); 1581 l = ((char *)USER_ADDR + SEGMENT_LENGTH) - (char *)(*kaddr); 1582 if (l > ulen) 1583 l = ulen; 1584 if (klen) 1585 *klen = l; 1586 else if (l != ulen) 1587 return (EFAULT); 1588 1589 vsid = va_to_vsid(pm, (vm_offset_t)uaddr); 1590 1591 /* Mark segment no-execute */ 1592 vsid |= SR_N; 1593 1594 /* If we have already set this VSID, we can just return */ 1595 if (curthread->td_pcb->pcb_cpu.aim.usr_vsid == vsid) 1596 return (0); 1597 1598 __asm __volatile("isync"); 1599 curthread->td_pcb->pcb_cpu.aim.usr_segm = 1600 (uintptr_t)uaddr >> ADDR_SR_SHFT; 1601 curthread->td_pcb->pcb_cpu.aim.usr_vsid = vsid; 1602 __asm __volatile("mtsr %0,%1; isync" :: "n"(USER_SR), "r"(vsid)); 1603 1604 return (0); 1605 } 1606 1607 /* 1608 * Figure out where a given kernel pointer (usually in a fault) points 1609 * to from the VM's perspective, potentially remapping into userland's 1610 * address space. 1611 */ 1612 static int 1613 moea_decode_kernel_ptr(vm_offset_t addr, int *is_user, 1614 vm_offset_t *decoded_addr) 1615 { 1616 vm_offset_t user_sr; 1617 1618 if ((addr >> ADDR_SR_SHFT) == (USER_ADDR >> ADDR_SR_SHFT)) { 1619 user_sr = curthread->td_pcb->pcb_cpu.aim.usr_segm; 1620 addr &= ADDR_PIDX | ADDR_POFF; 1621 addr |= user_sr << ADDR_SR_SHFT; 1622 *decoded_addr = addr; 1623 *is_user = 1; 1624 } else { 1625 *decoded_addr = addr; 1626 *is_user = 0; 1627 } 1628 1629 return (0); 1630 } 1631 1632 /* 1633 * Map a range of physical addresses into kernel virtual address space. 1634 * 1635 * The value passed in *virt is a suggested virtual address for the mapping. 1636 * Architectures which can support a direct-mapped physical to virtual region 1637 * can return the appropriate address within that region, leaving '*virt' 1638 * unchanged. We cannot and therefore do not; *virt is updated with the 1639 * first usable address after the mapped region. 1640 */ 1641 vm_offset_t 1642 moea_map(vm_offset_t *virt, vm_paddr_t pa_start, 1643 vm_paddr_t pa_end, int prot) 1644 { 1645 vm_offset_t sva, va; 1646 1647 sva = *virt; 1648 va = sva; 1649 for (; pa_start < pa_end; pa_start += PAGE_SIZE, va += PAGE_SIZE) 1650 moea_kenter(va, pa_start); 1651 *virt = va; 1652 return (sva); 1653 } 1654 1655 /* 1656 * Returns true if the pmap's pv is one of the first 1657 * 16 pvs linked to from this page. This count may 1658 * be changed upwards or downwards in the future; it 1659 * is only necessary that true be returned for a small 1660 * subset of pmaps for proper page aging. 1661 */ 1662 boolean_t 1663 moea_page_exists_quick(pmap_t pmap, vm_page_t m) 1664 { 1665 int loops; 1666 struct pvo_entry *pvo; 1667 boolean_t rv; 1668 1669 KASSERT((m->oflags & VPO_UNMANAGED) == 0, 1670 ("moea_page_exists_quick: page %p is not managed", m)); 1671 loops = 0; 1672 rv = FALSE; 1673 rw_wlock(&pvh_global_lock); 1674 LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) { 1675 if (pvo->pvo_pmap == pmap) { 1676 rv = TRUE; 1677 break; 1678 } 1679 if (++loops >= 16) 1680 break; 1681 } 1682 rw_wunlock(&pvh_global_lock); 1683 return (rv); 1684 } 1685 1686 void 1687 moea_page_init(vm_page_t m) 1688 { 1689 1690 m->md.mdpg_attrs = 0; 1691 m->md.mdpg_cache_attrs = VM_MEMATTR_DEFAULT; 1692 LIST_INIT(&m->md.mdpg_pvoh); 1693 } 1694 1695 /* 1696 * Return the number of managed mappings to the given physical page 1697 * that are wired. 1698 */ 1699 int 1700 moea_page_wired_mappings(vm_page_t m) 1701 { 1702 struct pvo_entry *pvo; 1703 int count; 1704 1705 count = 0; 1706 if ((m->oflags & VPO_UNMANAGED) != 0) 1707 return (count); 1708 rw_wlock(&pvh_global_lock); 1709 LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) 1710 if ((pvo->pvo_vaddr & PVO_WIRED) != 0) 1711 count++; 1712 rw_wunlock(&pvh_global_lock); 1713 return (count); 1714 } 1715 1716 static u_int moea_vsidcontext; 1717 1718 int 1719 moea_pinit(pmap_t pmap) 1720 { 1721 int i, mask; 1722 u_int entropy; 1723 1724 RB_INIT(&pmap->pmap_pvo); 1725 1726 entropy = 0; 1727 __asm __volatile("mftb %0" : "=r"(entropy)); 1728 1729 if ((pmap->pmap_phys = (pmap_t)moea_kextract((vm_offset_t)pmap)) 1730 == NULL) { 1731 pmap->pmap_phys = pmap; 1732 } 1733 1734 mtx_lock(&moea_vsid_mutex); 1735 /* 1736 * Allocate some segment registers for this pmap. 1737 */ 1738 for (i = 0; i < NPMAPS; i += VSID_NBPW) { 1739 u_int hash, n; 1740 1741 /* 1742 * Create a new value by multiplying by a prime and adding in 1743 * entropy from the timebase register. This is to make the 1744 * VSID more random so that the PT hash function collides 1745 * less often. (Note that the prime casues gcc to do shifts 1746 * instead of a multiply.) 1747 */ 1748 moea_vsidcontext = (moea_vsidcontext * 0x1105) + entropy; 1749 hash = moea_vsidcontext & (NPMAPS - 1); 1750 if (hash == 0) /* 0 is special, avoid it */ 1751 continue; 1752 n = hash >> 5; 1753 mask = 1 << (hash & (VSID_NBPW - 1)); 1754 hash = (moea_vsidcontext & 0xfffff); 1755 if (moea_vsid_bitmap[n] & mask) { /* collision? */ 1756 /* anything free in this bucket? */ 1757 if (moea_vsid_bitmap[n] == 0xffffffff) { 1758 entropy = (moea_vsidcontext >> 20); 1759 continue; 1760 } 1761 i = ffs(~moea_vsid_bitmap[n]) - 1; 1762 mask = 1 << i; 1763 hash &= rounddown2(0xfffff, VSID_NBPW); 1764 hash |= i; 1765 } 1766 KASSERT(!(moea_vsid_bitmap[n] & mask), 1767 ("Allocating in-use VSID group %#x\n", hash)); 1768 moea_vsid_bitmap[n] |= mask; 1769 for (i = 0; i < 16; i++) 1770 pmap->pm_sr[i] = VSID_MAKE(i, hash); 1771 mtx_unlock(&moea_vsid_mutex); 1772 return (1); 1773 } 1774 1775 mtx_unlock(&moea_vsid_mutex); 1776 panic("moea_pinit: out of segments"); 1777 } 1778 1779 /* 1780 * Initialize the pmap associated with process 0. 1781 */ 1782 void 1783 moea_pinit0(pmap_t pm) 1784 { 1785 1786 PMAP_LOCK_INIT(pm); 1787 moea_pinit(pm); 1788 bzero(&pm->pm_stats, sizeof(pm->pm_stats)); 1789 } 1790 1791 /* 1792 * Set the physical protection on the specified range of this map as requested. 1793 */ 1794 void 1795 moea_protect(pmap_t pm, vm_offset_t sva, vm_offset_t eva, 1796 vm_prot_t prot) 1797 { 1798 struct pvo_entry *pvo, *tpvo, key; 1799 struct pte *pt; 1800 1801 KASSERT(pm == &curproc->p_vmspace->vm_pmap || pm == kernel_pmap, 1802 ("moea_protect: non current pmap")); 1803 1804 if ((prot & VM_PROT_READ) == VM_PROT_NONE) { 1805 moea_remove(pm, sva, eva); 1806 return; 1807 } 1808 1809 rw_wlock(&pvh_global_lock); 1810 PMAP_LOCK(pm); 1811 key.pvo_vaddr = sva; 1812 for (pvo = RB_NFIND(pvo_tree, &pm->pmap_pvo, &key); 1813 pvo != NULL && PVO_VADDR(pvo) < eva; pvo = tpvo) { 1814 tpvo = RB_NEXT(pvo_tree, &pm->pmap_pvo, pvo); 1815 1816 /* 1817 * Grab the PTE pointer before we diddle with the cached PTE 1818 * copy. 1819 */ 1820 pt = moea_pvo_to_pte(pvo, -1); 1821 /* 1822 * Change the protection of the page. 1823 */ 1824 pvo->pvo_pte.pte.pte_lo &= ~PTE_PP; 1825 pvo->pvo_pte.pte.pte_lo |= PTE_BR; 1826 1827 /* 1828 * If the PVO is in the page table, update that pte as well. 1829 */ 1830 if (pt != NULL) { 1831 moea_pte_change(pt, &pvo->pvo_pte.pte, pvo->pvo_vaddr); 1832 mtx_unlock(&moea_table_mutex); 1833 } 1834 } 1835 rw_wunlock(&pvh_global_lock); 1836 PMAP_UNLOCK(pm); 1837 } 1838 1839 /* 1840 * Map a list of wired pages into kernel virtual address space. This is 1841 * intended for temporary mappings which do not need page modification or 1842 * references recorded. Existing mappings in the region are overwritten. 1843 */ 1844 void 1845 moea_qenter(vm_offset_t sva, vm_page_t *m, int count) 1846 { 1847 vm_offset_t va; 1848 1849 va = sva; 1850 while (count-- > 0) { 1851 moea_kenter(va, VM_PAGE_TO_PHYS(*m)); 1852 va += PAGE_SIZE; 1853 m++; 1854 } 1855 } 1856 1857 /* 1858 * Remove page mappings from kernel virtual address space. Intended for 1859 * temporary mappings entered by moea_qenter. 1860 */ 1861 void 1862 moea_qremove(vm_offset_t sva, int count) 1863 { 1864 vm_offset_t va; 1865 1866 va = sva; 1867 while (count-- > 0) { 1868 moea_kremove(va); 1869 va += PAGE_SIZE; 1870 } 1871 } 1872 1873 void 1874 moea_release(pmap_t pmap) 1875 { 1876 int idx, mask; 1877 1878 /* 1879 * Free segment register's VSID 1880 */ 1881 if (pmap->pm_sr[0] == 0) 1882 panic("moea_release"); 1883 1884 mtx_lock(&moea_vsid_mutex); 1885 idx = VSID_TO_HASH(pmap->pm_sr[0]) & (NPMAPS-1); 1886 mask = 1 << (idx % VSID_NBPW); 1887 idx /= VSID_NBPW; 1888 moea_vsid_bitmap[idx] &= ~mask; 1889 mtx_unlock(&moea_vsid_mutex); 1890 } 1891 1892 /* 1893 * Remove the given range of addresses from the specified map. 1894 */ 1895 void 1896 moea_remove(pmap_t pm, vm_offset_t sva, vm_offset_t eva) 1897 { 1898 struct pvo_entry *pvo, *tpvo, key; 1899 1900 rw_wlock(&pvh_global_lock); 1901 PMAP_LOCK(pm); 1902 key.pvo_vaddr = sva; 1903 for (pvo = RB_NFIND(pvo_tree, &pm->pmap_pvo, &key); 1904 pvo != NULL && PVO_VADDR(pvo) < eva; pvo = tpvo) { 1905 tpvo = RB_NEXT(pvo_tree, &pm->pmap_pvo, pvo); 1906 moea_pvo_remove(pvo, -1); 1907 } 1908 PMAP_UNLOCK(pm); 1909 rw_wunlock(&pvh_global_lock); 1910 } 1911 1912 /* 1913 * Remove physical page from all pmaps in which it resides. moea_pvo_remove() 1914 * will reflect changes in pte's back to the vm_page. 1915 */ 1916 void 1917 moea_remove_all(vm_page_t m) 1918 { 1919 struct pvo_head *pvo_head; 1920 struct pvo_entry *pvo, *next_pvo; 1921 pmap_t pmap; 1922 1923 rw_wlock(&pvh_global_lock); 1924 pvo_head = vm_page_to_pvoh(m); 1925 for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) { 1926 next_pvo = LIST_NEXT(pvo, pvo_vlink); 1927 1928 pmap = pvo->pvo_pmap; 1929 PMAP_LOCK(pmap); 1930 moea_pvo_remove(pvo, -1); 1931 PMAP_UNLOCK(pmap); 1932 } 1933 if ((m->a.flags & PGA_WRITEABLE) && moea_query_bit(m, PTE_CHG)) { 1934 moea_attr_clear(m, PTE_CHG); 1935 vm_page_dirty(m); 1936 } 1937 vm_page_aflag_clear(m, PGA_WRITEABLE); 1938 rw_wunlock(&pvh_global_lock); 1939 } 1940 1941 static int 1942 moea_mincore(pmap_t pm, vm_offset_t va, vm_paddr_t *pap) 1943 { 1944 struct pvo_entry *pvo; 1945 vm_paddr_t pa; 1946 vm_page_t m; 1947 int val; 1948 bool managed; 1949 1950 PMAP_LOCK(pm); 1951 1952 pvo = moea_pvo_find_va(pm, va & ~ADDR_POFF, NULL); 1953 if (pvo != NULL) { 1954 pa = PVO_PADDR(pvo); 1955 m = PHYS_TO_VM_PAGE(pa); 1956 managed = (pvo->pvo_vaddr & PVO_MANAGED) == PVO_MANAGED; 1957 val = MINCORE_INCORE; 1958 } else { 1959 PMAP_UNLOCK(pm); 1960 return (0); 1961 } 1962 1963 PMAP_UNLOCK(pm); 1964 1965 if (m == NULL) 1966 return (0); 1967 1968 if (managed) { 1969 if (moea_is_modified(m)) 1970 val |= MINCORE_MODIFIED | MINCORE_MODIFIED_OTHER; 1971 1972 if (moea_is_referenced(m)) 1973 val |= MINCORE_REFERENCED | MINCORE_REFERENCED_OTHER; 1974 } 1975 1976 if ((val & (MINCORE_MODIFIED_OTHER | MINCORE_REFERENCED_OTHER)) != 1977 (MINCORE_MODIFIED_OTHER | MINCORE_REFERENCED_OTHER) && 1978 managed) { 1979 *pap = pa; 1980 } 1981 1982 return (val); 1983 } 1984 1985 /* 1986 * Allocate a physical page of memory directly from the phys_avail map. 1987 * Can only be called from moea_bootstrap before avail start and end are 1988 * calculated. 1989 */ 1990 static vm_offset_t 1991 moea_bootstrap_alloc(vm_size_t size, u_int align) 1992 { 1993 vm_offset_t s, e; 1994 int i, j; 1995 1996 size = round_page(size); 1997 for (i = 0; phys_avail[i + 1] != 0; i += 2) { 1998 if (align != 0) 1999 s = roundup2(phys_avail[i], align); 2000 else 2001 s = phys_avail[i]; 2002 e = s + size; 2003 2004 if (s < phys_avail[i] || e > phys_avail[i + 1]) 2005 continue; 2006 2007 if (s == phys_avail[i]) { 2008 phys_avail[i] += size; 2009 } else if (e == phys_avail[i + 1]) { 2010 phys_avail[i + 1] -= size; 2011 } else { 2012 for (j = phys_avail_count * 2; j > i; j -= 2) { 2013 phys_avail[j] = phys_avail[j - 2]; 2014 phys_avail[j + 1] = phys_avail[j - 1]; 2015 } 2016 2017 phys_avail[i + 3] = phys_avail[i + 1]; 2018 phys_avail[i + 1] = s; 2019 phys_avail[i + 2] = e; 2020 phys_avail_count++; 2021 } 2022 2023 return (s); 2024 } 2025 panic("moea_bootstrap_alloc: could not allocate memory"); 2026 } 2027 2028 static void 2029 moea_syncicache(vm_paddr_t pa, vm_size_t len) 2030 { 2031 __syncicache((void *)pa, len); 2032 } 2033 2034 static int 2035 moea_pvo_enter(pmap_t pm, uma_zone_t zone, struct pvo_head *pvo_head, 2036 vm_offset_t va, vm_paddr_t pa, u_int pte_lo, int flags) 2037 { 2038 struct pvo_entry *pvo; 2039 u_int sr; 2040 int first; 2041 u_int ptegidx; 2042 int i; 2043 int bootstrap; 2044 2045 moea_pvo_enter_calls++; 2046 first = 0; 2047 bootstrap = 0; 2048 2049 /* 2050 * Compute the PTE Group index. 2051 */ 2052 va &= ~ADDR_POFF; 2053 sr = va_to_sr(pm->pm_sr, va); 2054 ptegidx = va_to_pteg(sr, va); 2055 2056 /* 2057 * Remove any existing mapping for this page. Reuse the pvo entry if 2058 * there is a mapping. 2059 */ 2060 mtx_lock(&moea_table_mutex); 2061 LIST_FOREACH(pvo, &moea_pvo_table[ptegidx], pvo_olink) { 2062 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 2063 if (PVO_PADDR(pvo) == pa && 2064 (pvo->pvo_pte.pte.pte_lo & PTE_PP) == 2065 (pte_lo & PTE_PP)) { 2066 /* 2067 * The PTE is not changing. Instead, this may 2068 * be a request to change the mapping's wired 2069 * attribute. 2070 */ 2071 mtx_unlock(&moea_table_mutex); 2072 if ((flags & PVO_WIRED) != 0 && 2073 (pvo->pvo_vaddr & PVO_WIRED) == 0) { 2074 pvo->pvo_vaddr |= PVO_WIRED; 2075 pm->pm_stats.wired_count++; 2076 } else if ((flags & PVO_WIRED) == 0 && 2077 (pvo->pvo_vaddr & PVO_WIRED) != 0) { 2078 pvo->pvo_vaddr &= ~PVO_WIRED; 2079 pm->pm_stats.wired_count--; 2080 } 2081 return (0); 2082 } 2083 moea_pvo_remove(pvo, -1); 2084 break; 2085 } 2086 } 2087 2088 /* 2089 * If we aren't overwriting a mapping, try to allocate. 2090 */ 2091 if (moea_initialized) { 2092 pvo = uma_zalloc(zone, M_NOWAIT); 2093 } else { 2094 if (moea_bpvo_pool_index >= BPVO_POOL_SIZE) { 2095 panic("moea_enter: bpvo pool exhausted, %d, %d, %d", 2096 moea_bpvo_pool_index, BPVO_POOL_SIZE, 2097 BPVO_POOL_SIZE * sizeof(struct pvo_entry)); 2098 } 2099 pvo = &moea_bpvo_pool[moea_bpvo_pool_index]; 2100 moea_bpvo_pool_index++; 2101 bootstrap = 1; 2102 } 2103 2104 if (pvo == NULL) { 2105 mtx_unlock(&moea_table_mutex); 2106 return (ENOMEM); 2107 } 2108 2109 moea_pvo_entries++; 2110 pvo->pvo_vaddr = va; 2111 pvo->pvo_pmap = pm; 2112 LIST_INSERT_HEAD(&moea_pvo_table[ptegidx], pvo, pvo_olink); 2113 pvo->pvo_vaddr &= ~ADDR_POFF; 2114 if (flags & PVO_WIRED) 2115 pvo->pvo_vaddr |= PVO_WIRED; 2116 if (pvo_head != &moea_pvo_kunmanaged) 2117 pvo->pvo_vaddr |= PVO_MANAGED; 2118 if (bootstrap) 2119 pvo->pvo_vaddr |= PVO_BOOTSTRAP; 2120 2121 moea_pte_create(&pvo->pvo_pte.pte, sr, va, pa | pte_lo); 2122 2123 /* 2124 * Add to pmap list 2125 */ 2126 RB_INSERT(pvo_tree, &pm->pmap_pvo, pvo); 2127 2128 /* 2129 * Remember if the list was empty and therefore will be the first 2130 * item. 2131 */ 2132 if (LIST_FIRST(pvo_head) == NULL) 2133 first = 1; 2134 LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink); 2135 2136 if (pvo->pvo_vaddr & PVO_WIRED) 2137 pm->pm_stats.wired_count++; 2138 pm->pm_stats.resident_count++; 2139 2140 i = moea_pte_insert(ptegidx, &pvo->pvo_pte.pte); 2141 KASSERT(i < 8, ("Invalid PTE index")); 2142 if (i >= 0) { 2143 PVO_PTEGIDX_SET(pvo, i); 2144 } else { 2145 panic("moea_pvo_enter: overflow"); 2146 moea_pte_overflow++; 2147 } 2148 mtx_unlock(&moea_table_mutex); 2149 2150 return (first ? ENOENT : 0); 2151 } 2152 2153 static void 2154 moea_pvo_remove(struct pvo_entry *pvo, int pteidx) 2155 { 2156 struct pte *pt; 2157 2158 /* 2159 * If there is an active pte entry, we need to deactivate it (and 2160 * save the ref & cfg bits). 2161 */ 2162 pt = moea_pvo_to_pte(pvo, pteidx); 2163 if (pt != NULL) { 2164 moea_pte_unset(pt, &pvo->pvo_pte.pte, pvo->pvo_vaddr); 2165 mtx_unlock(&moea_table_mutex); 2166 PVO_PTEGIDX_CLR(pvo); 2167 } else { 2168 moea_pte_overflow--; 2169 } 2170 2171 /* 2172 * Update our statistics. 2173 */ 2174 pvo->pvo_pmap->pm_stats.resident_count--; 2175 if (pvo->pvo_vaddr & PVO_WIRED) 2176 pvo->pvo_pmap->pm_stats.wired_count--; 2177 2178 /* 2179 * Remove this PVO from the PV and pmap lists. 2180 */ 2181 LIST_REMOVE(pvo, pvo_vlink); 2182 RB_REMOVE(pvo_tree, &pvo->pvo_pmap->pmap_pvo, pvo); 2183 2184 /* 2185 * Save the REF/CHG bits into their cache if the page is managed. 2186 * Clear PGA_WRITEABLE if all mappings of the page have been removed. 2187 */ 2188 if ((pvo->pvo_vaddr & PVO_MANAGED) == PVO_MANAGED) { 2189 struct vm_page *pg; 2190 2191 pg = PHYS_TO_VM_PAGE(PVO_PADDR(pvo)); 2192 if (pg != NULL) { 2193 moea_attr_save(pg, pvo->pvo_pte.pte.pte_lo & 2194 (PTE_REF | PTE_CHG)); 2195 if (LIST_EMPTY(&pg->md.mdpg_pvoh)) 2196 vm_page_aflag_clear(pg, PGA_WRITEABLE); 2197 } 2198 } 2199 2200 /* 2201 * Remove this from the overflow list and return it to the pool 2202 * if we aren't going to reuse it. 2203 */ 2204 LIST_REMOVE(pvo, pvo_olink); 2205 if (!(pvo->pvo_vaddr & PVO_BOOTSTRAP)) 2206 uma_zfree(pvo->pvo_vaddr & PVO_MANAGED ? moea_mpvo_zone : 2207 moea_upvo_zone, pvo); 2208 moea_pvo_entries--; 2209 moea_pvo_remove_calls++; 2210 } 2211 2212 static __inline int 2213 moea_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx) 2214 { 2215 int pteidx; 2216 2217 /* 2218 * We can find the actual pte entry without searching by grabbing 2219 * the PTEG index from 3 unused bits in pte_lo[11:9] and by 2220 * noticing the HID bit. 2221 */ 2222 pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo); 2223 if (pvo->pvo_pte.pte.pte_hi & PTE_HID) 2224 pteidx ^= moea_pteg_mask * 8; 2225 2226 return (pteidx); 2227 } 2228 2229 static struct pvo_entry * 2230 moea_pvo_find_va(pmap_t pm, vm_offset_t va, int *pteidx_p) 2231 { 2232 struct pvo_entry *pvo; 2233 int ptegidx; 2234 u_int sr; 2235 2236 va &= ~ADDR_POFF; 2237 sr = va_to_sr(pm->pm_sr, va); 2238 ptegidx = va_to_pteg(sr, va); 2239 2240 mtx_lock(&moea_table_mutex); 2241 LIST_FOREACH(pvo, &moea_pvo_table[ptegidx], pvo_olink) { 2242 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) { 2243 if (pteidx_p) 2244 *pteidx_p = moea_pvo_pte_index(pvo, ptegidx); 2245 break; 2246 } 2247 } 2248 mtx_unlock(&moea_table_mutex); 2249 2250 return (pvo); 2251 } 2252 2253 static struct pte * 2254 moea_pvo_to_pte(const struct pvo_entry *pvo, int pteidx) 2255 { 2256 struct pte *pt; 2257 2258 /* 2259 * If we haven't been supplied the ptegidx, calculate it. 2260 */ 2261 if (pteidx == -1) { 2262 int ptegidx; 2263 u_int sr; 2264 2265 sr = va_to_sr(pvo->pvo_pmap->pm_sr, pvo->pvo_vaddr); 2266 ptegidx = va_to_pteg(sr, pvo->pvo_vaddr); 2267 pteidx = moea_pvo_pte_index(pvo, ptegidx); 2268 } 2269 2270 pt = &moea_pteg_table[pteidx >> 3].pt[pteidx & 7]; 2271 mtx_lock(&moea_table_mutex); 2272 2273 if ((pvo->pvo_pte.pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) { 2274 panic("moea_pvo_to_pte: pvo %p has valid pte in pvo but no " 2275 "valid pte index", pvo); 2276 } 2277 2278 if ((pvo->pvo_pte.pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) { 2279 panic("moea_pvo_to_pte: pvo %p has valid pte index in pvo " 2280 "pvo but no valid pte", pvo); 2281 } 2282 2283 if ((pt->pte_hi ^ (pvo->pvo_pte.pte.pte_hi & ~PTE_VALID)) == PTE_VALID) { 2284 if ((pvo->pvo_pte.pte.pte_hi & PTE_VALID) == 0) { 2285 panic("moea_pvo_to_pte: pvo %p has valid pte in " 2286 "moea_pteg_table %p but invalid in pvo", pvo, pt); 2287 } 2288 2289 if (((pt->pte_lo ^ pvo->pvo_pte.pte.pte_lo) & ~(PTE_CHG|PTE_REF)) 2290 != 0) { 2291 panic("moea_pvo_to_pte: pvo %p pte does not match " 2292 "pte %p in moea_pteg_table", pvo, pt); 2293 } 2294 2295 mtx_assert(&moea_table_mutex, MA_OWNED); 2296 return (pt); 2297 } 2298 2299 if (pvo->pvo_pte.pte.pte_hi & PTE_VALID) { 2300 panic("moea_pvo_to_pte: pvo %p has invalid pte %p in " 2301 "moea_pteg_table but valid in pvo: %8x, %8x", pvo, pt, pvo->pvo_pte.pte.pte_hi, pt->pte_hi); 2302 } 2303 2304 mtx_unlock(&moea_table_mutex); 2305 return (NULL); 2306 } 2307 2308 /* 2309 * XXX: THIS STUFF SHOULD BE IN pte.c? 2310 */ 2311 int 2312 moea_pte_spill(vm_offset_t addr) 2313 { 2314 struct pvo_entry *source_pvo, *victim_pvo; 2315 struct pvo_entry *pvo; 2316 int ptegidx, i, j; 2317 u_int sr; 2318 struct pteg *pteg; 2319 struct pte *pt; 2320 2321 moea_pte_spills++; 2322 2323 sr = mfsrin(addr); 2324 ptegidx = va_to_pteg(sr, addr); 2325 2326 /* 2327 * Have to substitute some entry. Use the primary hash for this. 2328 * Use low bits of timebase as random generator. 2329 */ 2330 pteg = &moea_pteg_table[ptegidx]; 2331 mtx_lock(&moea_table_mutex); 2332 __asm __volatile("mftb %0" : "=r"(i)); 2333 i &= 7; 2334 pt = &pteg->pt[i]; 2335 2336 source_pvo = NULL; 2337 victim_pvo = NULL; 2338 LIST_FOREACH(pvo, &moea_pvo_table[ptegidx], pvo_olink) { 2339 /* 2340 * We need to find a pvo entry for this address. 2341 */ 2342 if (source_pvo == NULL && 2343 moea_pte_match(&pvo->pvo_pte.pte, sr, addr, 2344 pvo->pvo_pte.pte.pte_hi & PTE_HID)) { 2345 /* 2346 * Now found an entry to be spilled into the pteg. 2347 * The PTE is now valid, so we know it's active. 2348 */ 2349 j = moea_pte_insert(ptegidx, &pvo->pvo_pte.pte); 2350 2351 if (j >= 0) { 2352 PVO_PTEGIDX_SET(pvo, j); 2353 moea_pte_overflow--; 2354 mtx_unlock(&moea_table_mutex); 2355 return (1); 2356 } 2357 2358 source_pvo = pvo; 2359 2360 if (victim_pvo != NULL) 2361 break; 2362 } 2363 2364 /* 2365 * We also need the pvo entry of the victim we are replacing 2366 * so save the R & C bits of the PTE. 2367 */ 2368 if ((pt->pte_hi & PTE_HID) == 0 && victim_pvo == NULL && 2369 moea_pte_compare(pt, &pvo->pvo_pte.pte)) { 2370 victim_pvo = pvo; 2371 if (source_pvo != NULL) 2372 break; 2373 } 2374 } 2375 2376 if (source_pvo == NULL) { 2377 mtx_unlock(&moea_table_mutex); 2378 return (0); 2379 } 2380 2381 if (victim_pvo == NULL) { 2382 if ((pt->pte_hi & PTE_HID) == 0) 2383 panic("moea_pte_spill: victim p-pte (%p) has no pvo" 2384 "entry", pt); 2385 2386 /* 2387 * If this is a secondary PTE, we need to search it's primary 2388 * pvo bucket for the matching PVO. 2389 */ 2390 LIST_FOREACH(pvo, &moea_pvo_table[ptegidx ^ moea_pteg_mask], 2391 pvo_olink) { 2392 /* 2393 * We also need the pvo entry of the victim we are 2394 * replacing so save the R & C bits of the PTE. 2395 */ 2396 if (moea_pte_compare(pt, &pvo->pvo_pte.pte)) { 2397 victim_pvo = pvo; 2398 break; 2399 } 2400 } 2401 2402 if (victim_pvo == NULL) 2403 panic("moea_pte_spill: victim s-pte (%p) has no pvo" 2404 "entry", pt); 2405 } 2406 2407 /* 2408 * We are invalidating the TLB entry for the EA we are replacing even 2409 * though it's valid. If we don't, we lose any ref/chg bit changes 2410 * contained in the TLB entry. 2411 */ 2412 source_pvo->pvo_pte.pte.pte_hi &= ~PTE_HID; 2413 2414 moea_pte_unset(pt, &victim_pvo->pvo_pte.pte, victim_pvo->pvo_vaddr); 2415 moea_pte_set(pt, &source_pvo->pvo_pte.pte); 2416 2417 PVO_PTEGIDX_CLR(victim_pvo); 2418 PVO_PTEGIDX_SET(source_pvo, i); 2419 moea_pte_replacements++; 2420 2421 mtx_unlock(&moea_table_mutex); 2422 return (1); 2423 } 2424 2425 static __inline struct pvo_entry * 2426 moea_pte_spillable_ident(u_int ptegidx) 2427 { 2428 struct pte *pt; 2429 struct pvo_entry *pvo_walk, *pvo = NULL; 2430 2431 LIST_FOREACH(pvo_walk, &moea_pvo_table[ptegidx], pvo_olink) { 2432 if (pvo_walk->pvo_vaddr & PVO_WIRED) 2433 continue; 2434 2435 if (!(pvo_walk->pvo_pte.pte.pte_hi & PTE_VALID)) 2436 continue; 2437 2438 pt = moea_pvo_to_pte(pvo_walk, -1); 2439 2440 if (pt == NULL) 2441 continue; 2442 2443 pvo = pvo_walk; 2444 2445 mtx_unlock(&moea_table_mutex); 2446 if (!(pt->pte_lo & PTE_REF)) 2447 return (pvo_walk); 2448 } 2449 2450 return (pvo); 2451 } 2452 2453 static int 2454 moea_pte_insert(u_int ptegidx, struct pte *pvo_pt) 2455 { 2456 struct pte *pt; 2457 struct pvo_entry *victim_pvo; 2458 int i; 2459 int victim_idx; 2460 u_int pteg_bkpidx = ptegidx; 2461 2462 mtx_assert(&moea_table_mutex, MA_OWNED); 2463 2464 /* 2465 * First try primary hash. 2466 */ 2467 for (pt = moea_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 2468 if ((pt->pte_hi & PTE_VALID) == 0) { 2469 pvo_pt->pte_hi &= ~PTE_HID; 2470 moea_pte_set(pt, pvo_pt); 2471 return (i); 2472 } 2473 } 2474 2475 /* 2476 * Now try secondary hash. 2477 */ 2478 ptegidx ^= moea_pteg_mask; 2479 2480 for (pt = moea_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) { 2481 if ((pt->pte_hi & PTE_VALID) == 0) { 2482 pvo_pt->pte_hi |= PTE_HID; 2483 moea_pte_set(pt, pvo_pt); 2484 return (i); 2485 } 2486 } 2487 2488 /* Try again, but this time try to force a PTE out. */ 2489 ptegidx = pteg_bkpidx; 2490 2491 victim_pvo = moea_pte_spillable_ident(ptegidx); 2492 if (victim_pvo == NULL) { 2493 ptegidx ^= moea_pteg_mask; 2494 victim_pvo = moea_pte_spillable_ident(ptegidx); 2495 } 2496 2497 if (victim_pvo == NULL) { 2498 panic("moea_pte_insert: overflow"); 2499 return (-1); 2500 } 2501 2502 victim_idx = moea_pvo_pte_index(victim_pvo, ptegidx); 2503 2504 if (pteg_bkpidx == ptegidx) 2505 pvo_pt->pte_hi &= ~PTE_HID; 2506 else 2507 pvo_pt->pte_hi |= PTE_HID; 2508 2509 /* 2510 * Synchronize the sacrifice PTE with its PVO, then mark both 2511 * invalid. The PVO will be reused when/if the VM system comes 2512 * here after a fault. 2513 */ 2514 pt = &moea_pteg_table[victim_idx >> 3].pt[victim_idx & 7]; 2515 2516 if (pt->pte_hi != victim_pvo->pvo_pte.pte.pte_hi) 2517 panic("Victim PVO doesn't match PTE! PVO: %8x, PTE: %8x", victim_pvo->pvo_pte.pte.pte_hi, pt->pte_hi); 2518 2519 /* 2520 * Set the new PTE. 2521 */ 2522 moea_pte_unset(pt, &victim_pvo->pvo_pte.pte, victim_pvo->pvo_vaddr); 2523 PVO_PTEGIDX_CLR(victim_pvo); 2524 moea_pte_overflow++; 2525 moea_pte_set(pt, pvo_pt); 2526 2527 return (victim_idx & 7); 2528 } 2529 2530 static boolean_t 2531 moea_query_bit(vm_page_t m, int ptebit) 2532 { 2533 struct pvo_entry *pvo; 2534 struct pte *pt; 2535 2536 rw_assert(&pvh_global_lock, RA_WLOCKED); 2537 if (moea_attr_fetch(m) & ptebit) 2538 return (TRUE); 2539 2540 LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) { 2541 /* 2542 * See if we saved the bit off. If so, cache it and return 2543 * success. 2544 */ 2545 if (pvo->pvo_pte.pte.pte_lo & ptebit) { 2546 moea_attr_save(m, ptebit); 2547 return (TRUE); 2548 } 2549 } 2550 2551 /* 2552 * No luck, now go through the hard part of looking at the PTEs 2553 * themselves. Sync so that any pending REF/CHG bits are flushed to 2554 * the PTEs. 2555 */ 2556 powerpc_sync(); 2557 LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) { 2558 /* 2559 * See if this pvo has a valid PTE. if so, fetch the 2560 * REF/CHG bits from the valid PTE. If the appropriate 2561 * ptebit is set, cache it and return success. 2562 */ 2563 pt = moea_pvo_to_pte(pvo, -1); 2564 if (pt != NULL) { 2565 moea_pte_synch(pt, &pvo->pvo_pte.pte); 2566 mtx_unlock(&moea_table_mutex); 2567 if (pvo->pvo_pte.pte.pte_lo & ptebit) { 2568 moea_attr_save(m, ptebit); 2569 return (TRUE); 2570 } 2571 } 2572 } 2573 2574 return (FALSE); 2575 } 2576 2577 static u_int 2578 moea_clear_bit(vm_page_t m, int ptebit) 2579 { 2580 u_int count; 2581 struct pvo_entry *pvo; 2582 struct pte *pt; 2583 2584 rw_assert(&pvh_global_lock, RA_WLOCKED); 2585 2586 /* 2587 * Clear the cached value. 2588 */ 2589 moea_attr_clear(m, ptebit); 2590 2591 /* 2592 * Sync so that any pending REF/CHG bits are flushed to the PTEs (so 2593 * we can reset the right ones). note that since the pvo entries and 2594 * list heads are accessed via BAT0 and are never placed in the page 2595 * table, we don't have to worry about further accesses setting the 2596 * REF/CHG bits. 2597 */ 2598 powerpc_sync(); 2599 2600 /* 2601 * For each pvo entry, clear the pvo's ptebit. If this pvo has a 2602 * valid pte clear the ptebit from the valid pte. 2603 */ 2604 count = 0; 2605 LIST_FOREACH(pvo, vm_page_to_pvoh(m), pvo_vlink) { 2606 pt = moea_pvo_to_pte(pvo, -1); 2607 if (pt != NULL) { 2608 moea_pte_synch(pt, &pvo->pvo_pte.pte); 2609 if (pvo->pvo_pte.pte.pte_lo & ptebit) { 2610 count++; 2611 moea_pte_clear(pt, PVO_VADDR(pvo), ptebit); 2612 } 2613 mtx_unlock(&moea_table_mutex); 2614 } 2615 pvo->pvo_pte.pte.pte_lo &= ~ptebit; 2616 } 2617 2618 return (count); 2619 } 2620 2621 /* 2622 * Return true if the physical range is encompassed by the battable[idx] 2623 */ 2624 static int 2625 moea_bat_mapped(int idx, vm_paddr_t pa, vm_size_t size) 2626 { 2627 u_int prot; 2628 u_int32_t start; 2629 u_int32_t end; 2630 u_int32_t bat_ble; 2631 2632 /* 2633 * Return immediately if not a valid mapping 2634 */ 2635 if (!(battable[idx].batu & BAT_Vs)) 2636 return (EINVAL); 2637 2638 /* 2639 * The BAT entry must be cache-inhibited, guarded, and r/w 2640 * so it can function as an i/o page 2641 */ 2642 prot = battable[idx].batl & (BAT_I|BAT_G|BAT_PP_RW); 2643 if (prot != (BAT_I|BAT_G|BAT_PP_RW)) 2644 return (EPERM); 2645 2646 /* 2647 * The address should be within the BAT range. Assume that the 2648 * start address in the BAT has the correct alignment (thus 2649 * not requiring masking) 2650 */ 2651 start = battable[idx].batl & BAT_PBS; 2652 bat_ble = (battable[idx].batu & ~(BAT_EBS)) | 0x03; 2653 end = start | (bat_ble << 15) | 0x7fff; 2654 2655 if ((pa < start) || ((pa + size) > end)) 2656 return (ERANGE); 2657 2658 return (0); 2659 } 2660 2661 boolean_t 2662 moea_dev_direct_mapped(vm_paddr_t pa, vm_size_t size) 2663 { 2664 int i; 2665 2666 /* 2667 * This currently does not work for entries that 2668 * overlap 256M BAT segments. 2669 */ 2670 2671 for(i = 0; i < 16; i++) 2672 if (moea_bat_mapped(i, pa, size) == 0) 2673 return (0); 2674 2675 return (EFAULT); 2676 } 2677 2678 /* 2679 * Map a set of physical memory pages into the kernel virtual 2680 * address space. Return a pointer to where it is mapped. This 2681 * routine is intended to be used for mapping device memory, 2682 * NOT real memory. 2683 */ 2684 void * 2685 moea_mapdev(vm_paddr_t pa, vm_size_t size) 2686 { 2687 2688 return (moea_mapdev_attr(pa, size, VM_MEMATTR_DEFAULT)); 2689 } 2690 2691 void * 2692 moea_mapdev_attr(vm_paddr_t pa, vm_size_t size, vm_memattr_t ma) 2693 { 2694 vm_offset_t va, tmpva, ppa, offset; 2695 int i; 2696 2697 ppa = trunc_page(pa); 2698 offset = pa & PAGE_MASK; 2699 size = roundup(offset + size, PAGE_SIZE); 2700 2701 /* 2702 * If the physical address lies within a valid BAT table entry, 2703 * return the 1:1 mapping. This currently doesn't work 2704 * for regions that overlap 256M BAT segments. 2705 */ 2706 for (i = 0; i < 16; i++) { 2707 if (moea_bat_mapped(i, pa, size) == 0) 2708 return ((void *) pa); 2709 } 2710 2711 va = kva_alloc(size); 2712 if (!va) 2713 panic("moea_mapdev: Couldn't alloc kernel virtual memory"); 2714 2715 for (tmpva = va; size > 0;) { 2716 moea_kenter_attr(tmpva, ppa, ma); 2717 tlbie(tmpva); 2718 size -= PAGE_SIZE; 2719 tmpva += PAGE_SIZE; 2720 ppa += PAGE_SIZE; 2721 } 2722 2723 return ((void *)(va + offset)); 2724 } 2725 2726 void 2727 moea_unmapdev(void *p, vm_size_t size) 2728 { 2729 vm_offset_t base, offset, va; 2730 2731 /* 2732 * If this is outside kernel virtual space, then it's a 2733 * battable entry and doesn't require unmapping 2734 */ 2735 va = (vm_offset_t)p; 2736 if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= virtual_end)) { 2737 base = trunc_page(va); 2738 offset = va & PAGE_MASK; 2739 size = roundup(offset + size, PAGE_SIZE); 2740 moea_qremove(base, atop(size)); 2741 kva_free(base, size); 2742 } 2743 } 2744 2745 static void 2746 moea_sync_icache(pmap_t pm, vm_offset_t va, vm_size_t sz) 2747 { 2748 struct pvo_entry *pvo; 2749 vm_offset_t lim; 2750 vm_paddr_t pa; 2751 vm_size_t len; 2752 2753 PMAP_LOCK(pm); 2754 while (sz > 0) { 2755 lim = round_page(va + 1); 2756 len = MIN(lim - va, sz); 2757 pvo = moea_pvo_find_va(pm, va & ~ADDR_POFF, NULL); 2758 if (pvo != NULL) { 2759 pa = PVO_PADDR(pvo) | (va & ADDR_POFF); 2760 moea_syncicache(pa, len); 2761 } 2762 va += len; 2763 sz -= len; 2764 } 2765 PMAP_UNLOCK(pm); 2766 } 2767 2768 void 2769 moea_dumpsys_map(vm_paddr_t pa, size_t sz, void **va) 2770 { 2771 2772 *va = (void *)pa; 2773 } 2774 2775 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1]; 2776 2777 void 2778 moea_scan_init() 2779 { 2780 struct pvo_entry *pvo; 2781 vm_offset_t va; 2782 int i; 2783 2784 if (!do_minidump) { 2785 /* Initialize phys. segments for dumpsys(). */ 2786 memset(&dump_map, 0, sizeof(dump_map)); 2787 mem_regions(&pregions, &pregions_sz, ®ions, ®ions_sz); 2788 for (i = 0; i < pregions_sz; i++) { 2789 dump_map[i].pa_start = pregions[i].mr_start; 2790 dump_map[i].pa_size = pregions[i].mr_size; 2791 } 2792 return; 2793 } 2794 2795 /* Virtual segments for minidumps: */ 2796 memset(&dump_map, 0, sizeof(dump_map)); 2797 2798 /* 1st: kernel .data and .bss. */ 2799 dump_map[0].pa_start = trunc_page((uintptr_t)_etext); 2800 dump_map[0].pa_size = 2801 round_page((uintptr_t)_end) - dump_map[0].pa_start; 2802 2803 /* 2nd: msgbuf and tables (see pmap_bootstrap()). */ 2804 dump_map[1].pa_start = (vm_paddr_t)msgbufp->msg_ptr; 2805 dump_map[1].pa_size = round_page(msgbufp->msg_size); 2806 2807 /* 3rd: kernel VM. */ 2808 va = dump_map[1].pa_start + dump_map[1].pa_size; 2809 /* Find start of next chunk (from va). */ 2810 while (va < virtual_end) { 2811 /* Don't dump the buffer cache. */ 2812 if (va >= kmi.buffer_sva && va < kmi.buffer_eva) { 2813 va = kmi.buffer_eva; 2814 continue; 2815 } 2816 pvo = moea_pvo_find_va(kernel_pmap, va & ~ADDR_POFF, NULL); 2817 if (pvo != NULL && (pvo->pvo_pte.pte.pte_hi & PTE_VALID)) 2818 break; 2819 va += PAGE_SIZE; 2820 } 2821 if (va < virtual_end) { 2822 dump_map[2].pa_start = va; 2823 va += PAGE_SIZE; 2824 /* Find last page in chunk. */ 2825 while (va < virtual_end) { 2826 /* Don't run into the buffer cache. */ 2827 if (va == kmi.buffer_sva) 2828 break; 2829 pvo = moea_pvo_find_va(kernel_pmap, va & ~ADDR_POFF, 2830 NULL); 2831 if (pvo == NULL || 2832 !(pvo->pvo_pte.pte.pte_hi & PTE_VALID)) 2833 break; 2834 va += PAGE_SIZE; 2835 } 2836 dump_map[2].pa_size = va - dump_map[2].pa_start; 2837 } 2838 } 2839