1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* 27 * UNIX machine dependent virtual memory support. 28 */ 29 30 #ifndef _VM_DEP_H 31 #define _VM_DEP_H 32 33 #ifdef __cplusplus 34 extern "C" { 35 #endif 36 37 #include <vm/hat_sfmmu.h> 38 #include <sys/archsystm.h> 39 #include <sys/memnode.h> 40 41 #define GETTICK() gettick() 42 43 /* #define for keeping code architecturally neutral */ 44 #define randtick() gettick() 45 46 /* 47 * Per page size free lists. Allocated dynamically. 48 */ 49 #define MAX_MEM_TYPES 2 /* 0 = reloc, 1 = noreloc */ 50 #define MTYPE_RELOC 0 51 #define MTYPE_NORELOC 1 52 53 #define PP_2_MTYPE(pp) (PP_ISNORELOC(pp) ? MTYPE_NORELOC : MTYPE_RELOC) 54 55 #define MTYPE_INIT(mtype, vp, vaddr, flags, pgsz) \ 56 mtype = (flags & PG_NORELOC) ? MTYPE_NORELOC : MTYPE_RELOC; 57 58 /* mtype init for page_get_replacement_page */ 59 #define MTYPE_PGR_INIT(mtype, flags, pp, mnode, pgcnt) \ 60 mtype = (flags & PG_NORELOC) ? MTYPE_NORELOC : MTYPE_RELOC; 61 62 #define MNODETYPE_2_PFN(mnode, mtype, pfnlo, pfnhi) \ 63 ASSERT(mtype != MTYPE_NORELOC); \ 64 pfnlo = mem_node_config[mnode].physbase; \ 65 pfnhi = mem_node_config[mnode].physmax; 66 67 /* 68 * candidate counters in vm_pagelist.c are indexed by color and range 69 */ 70 #define MAX_MNODE_MRANGES MAX_MEM_TYPES 71 #define MNODE_RANGE_CNT(mnode) MAX_MNODE_MRANGES 72 #define MNODE_MAX_MRANGE(mnode) (MAX_MEM_TYPES - 1) 73 #define MTYPE_2_MRANGE(mnode, mtype) (mtype) 74 75 /* 76 * Internal PG_ flags. 77 */ 78 #define PGI_RELOCONLY 0x10000 /* acts in the opposite sense to PG_NORELOC */ 79 #define PGI_NOCAGE 0x20000 /* indicates Cage is disabled */ 80 #define PGI_PGCPHIPRI 0x40000 /* page_get_contig_page priority allocation */ 81 #define PGI_PGCPSZC0 0x80000 /* relocate base pagesize page */ 82 83 /* 84 * PGI mtype flags - should not overlap PGI flags 85 */ 86 #define PGI_MT_RANGE 0x1000000 /* mtype range */ 87 #define PGI_MT_NEXT 0x2000000 /* get next mtype */ 88 89 extern page_t ***page_freelists[MMU_PAGE_SIZES][MAX_MEM_TYPES]; 90 extern page_t ***page_cachelists[MAX_MEM_TYPES]; 91 92 #define PAGE_FREELISTS(mnode, szc, color, mtype) \ 93 (*(page_freelists[szc][mtype][mnode] + (color))) 94 95 #define PAGE_CACHELISTS(mnode, color, mtype) \ 96 (*(page_cachelists[mtype][mnode] + (color))) 97 98 /* 99 * There are 'page_colors' colors/bins. Spread them out under a 100 * couple of locks. There are mutexes for both the page freelist 101 * and the page cachelist. We want enough locks to make contention 102 * reasonable, but not too many -- otherwise page_freelist_lock() gets 103 * so expensive that it becomes the bottleneck! 104 */ 105 #define NPC_MUTEX 16 106 107 extern kmutex_t *fpc_mutex[NPC_MUTEX]; 108 extern kmutex_t *cpc_mutex[NPC_MUTEX]; 109 110 /* 111 * Iterator provides the info needed to convert RA to PA. 112 * MEM_NODE_ITERATOR_INIT() should be called before 113 * PAGE_NEXT_PFN_FOR_COLOR() if pfn was not obtained via a previous 114 * PAGE_NEXT_PFN_FOR_COLOR() call. Iterator caches color 2 hash 115 * translations requiring initializer call if color or ceq_mask changes, 116 * even if pfn doesn't. MEM_NODE_ITERATOR_INIT() must also be called before 117 * PFN_2_COLOR() that uses a valid iterator argument. 118 */ 119 #ifdef sun4v 120 121 typedef struct mem_node_iterator { 122 uint_t mi_mnode; /* mnode in which to iterate */ 123 int mi_init; /* set to 1 when first init */ 124 int mi_last_mblock; /* last mblock visited */ 125 uint_t mi_hash_ceq_mask; /* cached copy of ceq_mask */ 126 uint_t mi_hash_color; /* cached copy of color */ 127 uint_t mi_mnode_mask; /* number of mask bits */ 128 uint_t mi_mnode_pfn_shift; /* mnode position in pfn */ 129 pfn_t mi_mblock_base; /* first valid pfn in current mblock */ 130 pfn_t mi_mblock_end; /* last valid pfn in current mblock */ 131 pfn_t mi_ra_to_pa; /* ra adjustment for current mblock */ 132 pfn_t mi_mnode_pfn_mask; /* mask to obtain mnode id bits */ 133 } mem_node_iterator_t; 134 135 #define MEM_NODE_ITERATOR_DECL(it) \ 136 mem_node_iterator_t it 137 #define MEM_NODE_ITERATOR_INIT(pfn, mnode, szc, it) \ 138 (pfn) = plat_mem_node_iterator_init((pfn), (mnode), (szc), (it), 1) 139 140 extern pfn_t plat_mem_node_iterator_init(pfn_t, int, uchar_t, 141 mem_node_iterator_t *, int); 142 extern pfn_t plat_rapfn_to_papfn(pfn_t); 143 extern int interleaved_mnodes; 144 145 #else /* sun4v */ 146 147 #define MEM_NODE_ITERATOR_DECL(it) \ 148 void *it = NULL 149 #define MEM_NODE_ITERATOR_INIT(pfn, mnode, szc, it) 150 151 #endif /* sun4v */ 152 153 /* 154 * Return the mnode limits so that hpc_counters length and base 155 * index can be determined. When interleaved_mnodes is set, we 156 * create an array only for the first mnode that exists. All other 157 * mnodes will share the array in this case. 158 * If interleaved_mnodes is not set, simply return the limits for 159 * the given mnode. 160 */ 161 #define HPM_COUNTERS_LIMITS(mnode, physbase, physmax, first) \ 162 if (!interleaved_mnodes) { \ 163 (physbase) = mem_node_config[(mnode)].physbase; \ 164 (physmax) = mem_node_config[(mnode)].physmax; \ 165 (first) = (mnode); \ 166 } else if ((first) < 0) { \ 167 mem_node_max_range(&(physbase), &(physmax)); \ 168 (first) = (mnode); \ 169 } 170 171 #define PAGE_CTRS_WRITE_LOCK(mnode) \ 172 if (!interleaved_mnodes) { \ 173 rw_enter(&page_ctrs_rwlock[(mnode)], RW_WRITER); \ 174 page_freelist_lock(mnode); \ 175 } else { \ 176 /* changing shared hpm_counters */ \ 177 int _i; \ 178 for (_i = 0; _i < max_mem_nodes; _i++) { \ 179 rw_enter(&page_ctrs_rwlock[_i], RW_WRITER); \ 180 page_freelist_lock(_i); \ 181 } \ 182 } 183 184 #define PAGE_CTRS_WRITE_UNLOCK(mnode) \ 185 if (!interleaved_mnodes) { \ 186 page_freelist_unlock(mnode); \ 187 rw_exit(&page_ctrs_rwlock[(mnode)]); \ 188 } else { \ 189 int _i; \ 190 for (_i = 0; _i < max_mem_nodes; _i++) { \ 191 page_freelist_unlock(_i); \ 192 rw_exit(&page_ctrs_rwlock[_i]); \ 193 } \ 194 } 195 196 /* 197 * cpu specific color conversion functions 198 */ 199 extern uint_t page_get_nsz_color_mask_cpu(uchar_t, uint_t); 200 #pragma weak page_get_nsz_color_mask_cpu 201 202 extern uint_t page_get_nsz_color_cpu(uchar_t, uint_t); 203 #pragma weak page_get_nsz_color_cpu 204 205 extern uint_t page_get_color_shift_cpu(uchar_t, uchar_t); 206 #pragma weak page_get_color_shift_cpu 207 208 extern uint_t page_convert_color_cpu(uint_t, uchar_t, uchar_t); 209 #pragma weak page_convert_color_cpu 210 211 extern pfn_t page_next_pfn_for_color_cpu(pfn_t, 212 uchar_t, uint_t, uint_t, uint_t, void *); 213 #pragma weak page_next_pfn_for_color_cpu 214 215 extern uint_t page_pfn_2_color_cpu(pfn_t, uchar_t, void *); 216 #pragma weak page_pfn_2_color_cpu 217 218 #define PAGE_GET_COLOR_SHIFT(szc, nszc) \ 219 ((&page_get_color_shift_cpu != NULL) ? \ 220 page_get_color_shift_cpu(szc, nszc) : \ 221 (hw_page_array[(nszc)].hp_shift - \ 222 hw_page_array[(szc)].hp_shift)) 223 224 #define PAGE_CONVERT_COLOR(ncolor, szc, nszc) \ 225 ((&page_convert_color_cpu != NULL) ? \ 226 page_convert_color_cpu(ncolor, szc, nszc) : \ 227 ((ncolor) << PAGE_GET_COLOR_SHIFT((szc), (nszc)))) 228 229 #define PFN_2_COLOR(pfn, szc, it) \ 230 ((&page_pfn_2_color_cpu != NULL) ? \ 231 page_pfn_2_color_cpu(pfn, szc, it) : \ 232 ((pfn & (hw_page_array[0].hp_colors - 1)) >> \ 233 (hw_page_array[szc].hp_shift - \ 234 hw_page_array[0].hp_shift))) 235 236 #define PNUM_SIZE(szc) \ 237 (hw_page_array[(szc)].hp_pgcnt) 238 #define PNUM_SHIFT(szc) \ 239 (hw_page_array[(szc)].hp_shift - hw_page_array[0].hp_shift) 240 #define PAGE_GET_SHIFT(szc) \ 241 (hw_page_array[(szc)].hp_shift) 242 #define PAGE_GET_PAGECOLORS(szc) \ 243 (hw_page_array[(szc)].hp_colors) 244 245 /* 246 * This macro calculates the next sequential pfn with the specified 247 * color using color equivalency mask 248 */ 249 #define PAGE_NEXT_PFN_FOR_COLOR(pfn, szc, color, ceq_mask, color_mask, it) \ 250 { \ 251 ASSERT(((color) & ~(ceq_mask)) == 0); \ 252 if (&page_next_pfn_for_color_cpu == NULL) { \ 253 uint_t pfn_shift = PAGE_BSZS_SHIFT(szc); \ 254 pfn_t spfn = pfn >> pfn_shift; \ 255 pfn_t stride = (ceq_mask) + 1; \ 256 ASSERT((((ceq_mask) + 1) & (ceq_mask)) == 0); \ 257 if (((spfn ^ (color)) & (ceq_mask)) == 0) { \ 258 pfn += stride << pfn_shift; \ 259 } else { \ 260 pfn = (spfn & ~(pfn_t)(ceq_mask)) | (color); \ 261 pfn = (pfn > spfn ? pfn : pfn + stride) << \ 262 pfn_shift; \ 263 } \ 264 } else { \ 265 pfn = page_next_pfn_for_color_cpu(pfn, szc, color, \ 266 ceq_mask, color_mask, it); \ 267 } \ 268 } 269 270 /* get the color equivalency mask for the next szc */ 271 #define PAGE_GET_NSZ_MASK(szc, mask) \ 272 ((&page_get_nsz_color_mask_cpu == NULL) ? \ 273 ((mask) >> (PAGE_GET_SHIFT((szc) + 1) - PAGE_GET_SHIFT(szc))) : \ 274 page_get_nsz_color_mask_cpu(szc, mask)) 275 276 /* get the color of the next szc */ 277 #define PAGE_GET_NSZ_COLOR(szc, color) \ 278 ((&page_get_nsz_color_cpu == NULL) ? \ 279 ((color) >> (PAGE_GET_SHIFT((szc) + 1) - PAGE_GET_SHIFT(szc))) : \ 280 page_get_nsz_color_cpu(szc, color)) 281 282 /* Find the bin for the given page if it was of size szc */ 283 #define PP_2_BIN_SZC(pp, szc) (PFN_2_COLOR(pp->p_pagenum, szc, (void *)(-1))) 284 285 #define PP_2_BIN(pp) (PP_2_BIN_SZC(pp, pp->p_szc)) 286 287 #define PP_2_MEM_NODE(pp) (PFN_2_MEM_NODE(pp->p_pagenum)) 288 289 #define PC_BIN_MUTEX(mnode, bin, flags) ((flags & PG_FREE_LIST) ? \ 290 &fpc_mutex[(bin) & (NPC_MUTEX - 1)][mnode] : \ 291 &cpc_mutex[(bin) & (NPC_MUTEX - 1)][mnode]) 292 293 #define FPC_MUTEX(mnode, i) (&fpc_mutex[i][mnode]) 294 #define CPC_MUTEX(mnode, i) (&cpc_mutex[i][mnode]) 295 296 #define PFN_BASE(pfnum, szc) (pfnum & ~((1 << PAGE_BSZS_SHIFT(szc)) - 1)) 297 298 /* 299 * this structure is used for walking free page lists 300 * controls when to split large pages into smaller pages, 301 * and when to coalesce smaller pages into larger pages 302 */ 303 typedef struct page_list_walker { 304 uint_t plw_colors; /* num of colors for szc */ 305 uint_t plw_color_mask; /* colors-1 */ 306 uint_t plw_bin_step; /* next bin: 1 or 2 */ 307 uint_t plw_count; /* loop count */ 308 uint_t plw_bin0; /* starting bin */ 309 uint_t plw_bin_marker; /* bin after initial jump */ 310 uint_t plw_bin_split_prev; /* last bin we tried to split */ 311 uint_t plw_do_split; /* set if OK to split */ 312 uint_t plw_split_next; /* next bin to split */ 313 uint_t plw_ceq_dif; /* number of different color groups */ 314 /* to check */ 315 uint_t plw_ceq_mask[MMU_PAGE_SIZES + 1]; /* color equiv mask */ 316 uint_t plw_bins[MMU_PAGE_SIZES + 1]; /* num of bins */ 317 } page_list_walker_t; 318 319 void page_list_walk_init(uchar_t szc, uint_t flags, uint_t bin, 320 int can_split, int use_ceq, page_list_walker_t *plw); 321 322 typedef char hpmctr_t; 323 324 #ifdef DEBUG 325 #define CHK_LPG(pp, szc) chk_lpg(pp, szc) 326 extern void chk_lpg(page_t *, uchar_t); 327 #else 328 #define CHK_LPG(pp, szc) 329 #endif 330 331 /* 332 * page list count per mnode and type. 333 */ 334 typedef struct { 335 pgcnt_t plc_mt_pgmax; /* max page cnt */ 336 pgcnt_t plc_mt_clpgcnt; /* cache list cnt */ 337 pgcnt_t plc_mt_flpgcnt; /* free list cnt - small pages */ 338 pgcnt_t plc_mt_lgpgcnt; /* free list cnt - large pages */ 339 #ifdef DEBUG 340 struct { 341 pgcnt_t plc_mts_pgcnt; /* per page size count */ 342 int plc_mts_colors; 343 pgcnt_t *plc_mtsc_pgcnt; /* per color bin count */ 344 } plc_mts[MMU_PAGE_SIZES]; 345 #endif 346 } plcnt_t[MAX_MEM_NODES][MAX_MEM_TYPES]; 347 348 #ifdef DEBUG 349 350 #define PLCNT_SZ(ctrs_sz) { \ 351 int szc; \ 352 for (szc = 0; szc < mmu_page_sizes; szc++) { \ 353 int colors = page_get_pagecolors(szc); \ 354 ctrs_sz += (max_mem_nodes * MAX_MEM_TYPES * \ 355 colors * sizeof (pgcnt_t)); \ 356 } \ 357 } 358 359 #define PLCNT_INIT(base) { \ 360 int mn, mt, szc, colors; \ 361 for (szc = 0; szc < mmu_page_sizes; szc++) { \ 362 colors = page_get_pagecolors(szc); \ 363 for (mn = 0; mn < max_mem_nodes; mn++) { \ 364 for (mt = 0; mt < MAX_MEM_TYPES; mt++) { \ 365 plcnt[mn][mt].plc_mts[szc]. \ 366 plc_mts_colors = colors; \ 367 plcnt[mn][mt].plc_mts[szc]. \ 368 plc_mtsc_pgcnt = (pgcnt_t *)base; \ 369 base += (colors * sizeof (pgcnt_t)); \ 370 } \ 371 } \ 372 } \ 373 } 374 375 #define PLCNT_DO(pp, mn, mtype, szc, cnt, flags) { \ 376 int bin = PP_2_BIN(pp); \ 377 if (flags & PG_CACHE_LIST) \ 378 atomic_add_long(&plcnt[mn][mtype].plc_mt_clpgcnt, cnt); \ 379 else if (szc) \ 380 atomic_add_long(&plcnt[mn][mtype].plc_mt_lgpgcnt, cnt); \ 381 else \ 382 atomic_add_long(&plcnt[mn][mtype].plc_mt_flpgcnt, cnt); \ 383 atomic_add_long(&plcnt[mn][mtype].plc_mts[szc].plc_mts_pgcnt, \ 384 cnt); \ 385 atomic_add_long(&plcnt[mn][mtype].plc_mts[szc]. \ 386 plc_mtsc_pgcnt[bin], cnt); \ 387 } 388 389 #else 390 391 #define PLCNT_SZ(ctrs_sz) 392 393 #define PLCNT_INIT(base) 394 395 /* PG_FREE_LIST may not be explicitly set in flags for large pages */ 396 397 #define PLCNT_DO(pp, mn, mtype, szc, cnt, flags) { \ 398 if (flags & PG_CACHE_LIST) \ 399 atomic_add_long(&plcnt[mn][mtype].plc_mt_clpgcnt, cnt); \ 400 else if (szc) \ 401 atomic_add_long(&plcnt[mn][mtype].plc_mt_lgpgcnt, cnt); \ 402 else \ 403 atomic_add_long(&plcnt[mn][mtype].plc_mt_flpgcnt, cnt); \ 404 } 405 406 #endif 407 408 #define PLCNT_INCR(pp, mn, mtype, szc, flags) { \ 409 long cnt = (1 << PAGE_BSZS_SHIFT(szc)); \ 410 PLCNT_DO(pp, mn, mtype, szc, cnt, flags); \ 411 } 412 413 #define PLCNT_DECR(pp, mn, mtype, szc, flags) { \ 414 long cnt = ((-1) << PAGE_BSZS_SHIFT(szc)); \ 415 PLCNT_DO(pp, mn, mtype, szc, cnt, flags); \ 416 } 417 418 /* 419 * macros to update page list max counts - done when pages transferred 420 * from RELOC to NORELOC mtype (kcage_init or kcage_assimilate_page). 421 */ 422 423 #define PLCNT_XFER_NORELOC(pp) { \ 424 long cnt = (1 << PAGE_BSZS_SHIFT((pp)->p_szc)); \ 425 int mn = PP_2_MEM_NODE(pp); \ 426 atomic_add_long(&plcnt[mn][MTYPE_NORELOC].plc_mt_pgmax, cnt); \ 427 atomic_add_long(&plcnt[mn][MTYPE_RELOC].plc_mt_pgmax, -cnt); \ 428 } 429 430 /* 431 * macro to modify the page list max counts when memory is added to 432 * the page lists during startup (add_physmem) or during a DR operation 433 * when memory is added (kphysm_add_memory_dynamic) or deleted 434 * (kphysm_del_cleanup). 435 */ 436 #define PLCNT_MODIFY_MAX(pfn, cnt) { \ 437 spgcnt_t _cnt = (spgcnt_t)(cnt); \ 438 pgcnt_t _acnt = ABS(_cnt); \ 439 int _mn; \ 440 pgcnt_t _np; \ 441 if (&plat_mem_node_intersect_range != NULL) { \ 442 for (_mn = 0; _mn < max_mem_nodes; _mn++) { \ 443 plat_mem_node_intersect_range((pfn), _acnt, _mn, &_np);\ 444 if (_np == 0) \ 445 continue; \ 446 atomic_add_long(&plcnt[_mn][MTYPE_RELOC].plc_mt_pgmax, \ 447 (_cnt < 0) ? -_np : _np); \ 448 } \ 449 } else { \ 450 pfn_t _pfn = (pfn); \ 451 pfn_t _endpfn = _pfn + _acnt; \ 452 while (_pfn < _endpfn) { \ 453 _mn = PFN_2_MEM_NODE(_pfn); \ 454 _np = MIN(_endpfn, mem_node_config[_mn].physmax + 1) - \ 455 _pfn; \ 456 _pfn += _np; \ 457 atomic_add_long(&plcnt[_mn][MTYPE_RELOC].plc_mt_pgmax, \ 458 (_cnt < 0) ? -_np : _np); \ 459 } \ 460 } \ 461 } 462 463 extern plcnt_t plcnt; 464 465 #define MNODE_PGCNT(mn) \ 466 (plcnt[mn][MTYPE_RELOC].plc_mt_clpgcnt + \ 467 plcnt[mn][MTYPE_NORELOC].plc_mt_clpgcnt + \ 468 plcnt[mn][MTYPE_RELOC].plc_mt_flpgcnt + \ 469 plcnt[mn][MTYPE_NORELOC].plc_mt_flpgcnt + \ 470 plcnt[mn][MTYPE_RELOC].plc_mt_lgpgcnt + \ 471 plcnt[mn][MTYPE_NORELOC].plc_mt_lgpgcnt) 472 473 #define MNODETYPE_PGCNT(mn, mtype) \ 474 (plcnt[mn][mtype].plc_mt_clpgcnt + \ 475 plcnt[mn][mtype].plc_mt_flpgcnt + \ 476 plcnt[mn][mtype].plc_mt_lgpgcnt) 477 478 /* 479 * macros to loop through the mtype range - MTYPE_START returns -1 in 480 * mtype if no pages in mnode/mtype and possibly NEXT mtype. 481 */ 482 #define MTYPE_START(mnode, mtype, flags) { \ 483 if (plcnt[mnode][mtype].plc_mt_pgmax == 0) { \ 484 ASSERT(mtype == MTYPE_RELOC || \ 485 MNODETYPE_PGCNT(mnode, mtype) == 0 || \ 486 plcnt[mnode][mtype].plc_mt_pgmax != 0); \ 487 MTYPE_NEXT(mnode, mtype, flags); \ 488 } \ 489 } 490 491 /* 492 * if allocation from the RELOC pool failed and there is sufficient cage 493 * memory, attempt to allocate from the NORELOC pool. 494 */ 495 #define MTYPE_NEXT(mnode, mtype, flags) { \ 496 if (!(flags & (PG_NORELOC | PGI_NOCAGE | PGI_RELOCONLY)) && \ 497 (kcage_freemem >= kcage_lotsfree)) { \ 498 if (plcnt[mnode][MTYPE_NORELOC].plc_mt_pgmax == 0) { \ 499 ASSERT(MNODETYPE_PGCNT(mnode, MTYPE_NORELOC) == 0 || \ 500 plcnt[mnode][MTYPE_NORELOC].plc_mt_pgmax != 0); \ 501 mtype = -1; \ 502 } else { \ 503 mtype = MTYPE_NORELOC; \ 504 flags |= PG_NORELOC; \ 505 } \ 506 } else { \ 507 mtype = -1; \ 508 } \ 509 } 510 511 /* 512 * get the ecache setsize for the current cpu. 513 */ 514 #define CPUSETSIZE() (cpunodes[CPU->cpu_id].ecache_setsize) 515 516 extern struct cpu cpu0; 517 #define CPU0 &cpu0 518 519 #define PAGE_BSZS_SHIFT(szc) TTE_BSZS_SHIFT(szc) 520 /* 521 * For sfmmu each larger page is 8 times the size of the previous 522 * size page. 523 */ 524 #define FULL_REGION_CNT(rg_szc) (8) 525 526 /* 527 * The counter base must be per page_counter element to prevent 528 * races when re-indexing, and the base page size element should 529 * be aligned on a boundary of the given region size. 530 * 531 * We also round up the number of pages spanned by the counters 532 * for a given region to PC_BASE_ALIGN in certain situations to simplify 533 * the coding for some non-performance critical routines. 534 */ 535 #define PC_BASE_ALIGN ((pfn_t)1 << PAGE_BSZS_SHIFT(mmu_page_sizes-1)) 536 #define PC_BASE_ALIGN_MASK (PC_BASE_ALIGN - 1) 537 538 extern int ecache_alignsize; 539 #define L2CACHE_ALIGN ecache_alignsize 540 #define L2CACHE_ALIGN_MAX 512 541 542 extern int update_proc_pgcolorbase_after_fork; 543 extern int consistent_coloring; 544 extern uint_t vac_colors_mask; 545 extern int vac_size; 546 extern int vac_shift; 547 548 /* 549 * Kernel mem segment in 64-bit space 550 */ 551 extern caddr_t kmem64_base, kmem64_end, kmem64_aligned_end; 552 extern int kmem64_alignsize, kmem64_szc; 553 extern uint64_t kmem64_pabase; 554 extern int max_bootlp_tteszc; 555 556 /* 557 * Maximum and default values for user heap, stack, private and shared 558 * anonymous memory, and user text and initialized data. 559 * 560 * Initial values are defined in architecture specific mach_vm_dep.c file. 561 * Used by map_pgsz*() routines. 562 */ 563 extern size_t max_uheap_lpsize; 564 extern size_t default_uheap_lpsize; 565 extern size_t max_ustack_lpsize; 566 extern size_t default_ustack_lpsize; 567 extern size_t max_privmap_lpsize; 568 extern size_t max_uidata_lpsize; 569 extern size_t max_utext_lpsize; 570 extern size_t max_shm_lpsize; 571 572 /* 573 * For adjusting the default lpsize, for DTLB-limited page sizes. 574 */ 575 extern void adjust_data_maxlpsize(size_t ismpagesize); 576 577 /* 578 * Sanity control. Don't use large pages regardless of user 579 * settings if there's less than priv or shm_lpg_min_physmem memory installed. 580 * The units for this variable are 8K pages. 581 */ 582 extern pgcnt_t privm_lpg_min_physmem; 583 extern pgcnt_t shm_lpg_min_physmem; 584 585 /* 586 * AS_2_BIN macro controls the page coloring policy. 587 * 0 (default) uses various vaddr bits 588 * 1 virtual=paddr 589 * 2 bin hopping 590 */ 591 #define AS_2_BIN(as, seg, vp, addr, bin, szc) \ 592 switch (consistent_coloring) { \ 593 default: \ 594 cmn_err(CE_WARN, \ 595 "AS_2_BIN: bad consistent coloring value"); \ 596 /* assume default algorithm -> continue */ \ 597 case 0: { \ 598 uint32_t ndx, new; \ 599 int slew = 0; \ 600 pfn_t pfn; \ 601 \ 602 if (vp != NULL && IS_SWAPVP(vp) && \ 603 seg->s_ops == &segvn_ops) \ 604 slew = as_color_bin(as); \ 605 \ 606 pfn = ((uintptr_t)addr >> MMU_PAGESHIFT) + \ 607 (((uintptr_t)addr >> page_coloring_shift) << \ 608 (vac_shift - MMU_PAGESHIFT)); \ 609 if ((szc) == 0 || &page_pfn_2_color_cpu == NULL) { \ 610 pfn += slew; \ 611 bin = PFN_2_COLOR(pfn, szc, NULL); \ 612 } else { \ 613 bin = PFN_2_COLOR(pfn, szc, NULL); \ 614 bin += slew >> (vac_shift - MMU_PAGESHIFT); \ 615 bin &= hw_page_array[(szc)].hp_colors - 1; \ 616 } \ 617 break; \ 618 } \ 619 case 1: \ 620 bin = PFN_2_COLOR(((uintptr_t)addr >> MMU_PAGESHIFT), \ 621 szc, NULL); \ 622 break; \ 623 case 2: { \ 624 int cnt = as_color_bin(as); \ 625 uint_t color_mask = page_get_pagecolors(0) - 1; \ 626 \ 627 /* make sure physical color aligns with vac color */ \ 628 while ((cnt & vac_colors_mask) != \ 629 addr_to_vcolor(addr)) { \ 630 cnt++; \ 631 } \ 632 bin = cnt = cnt & color_mask; \ 633 bin >>= PAGE_GET_COLOR_SHIFT(0, szc); \ 634 /* update per as page coloring fields */ \ 635 cnt = (cnt + 1) & color_mask; \ 636 if (cnt == (as_color_start(as) & color_mask)) { \ 637 cnt = as_color_start(as) = as_color_start(as) + \ 638 PGCLR_LOOPFACTOR; \ 639 } \ 640 as_color_bin(as) = cnt & color_mask; \ 641 break; \ 642 } \ 643 } \ 644 ASSERT(bin < page_get_pagecolors(szc)); 645 646 /* 647 * cpu private vm data - accessed thru CPU->cpu_vm_data 648 * vc_pnum_memseg: tracks last memseg visited in page_numtopp_nolock() 649 * vc_pnext_memseg: tracks last memseg visited in page_nextn() 650 * vc_kmptr: unaligned kmem pointer for this vm_cpu_data_t 651 * vc_kmsize: orignal kmem size for this vm_cpu_data_t 652 */ 653 654 typedef struct { 655 struct memseg *vc_pnum_memseg; 656 struct memseg *vc_pnext_memseg; 657 void *vc_kmptr; 658 size_t vc_kmsize; 659 } vm_cpu_data_t; 660 661 /* allocation size to ensure vm_cpu_data_t resides in its own cache line */ 662 #define VM_CPU_DATA_PADSIZE \ 663 (P2ROUNDUP(sizeof (vm_cpu_data_t), L2CACHE_ALIGN_MAX)) 664 665 /* for boot cpu before kmem is initialized */ 666 extern char vm_cpu_data0[]; 667 668 /* 669 * Function to get an ecache color bin: F(as, cnt, vcolor). 670 * the goal of this function is to: 671 * - to spread a processes' physical pages across the entire ecache to 672 * maximize its use. 673 * - to minimize vac flushes caused when we reuse a physical page on a 674 * different vac color than it was previously used. 675 * - to prevent all processes to use the same exact colors and trash each 676 * other. 677 * 678 * cnt is a bin ptr kept on a per as basis. As we page_create we increment 679 * the ptr so we spread out the physical pages to cover the entire ecache. 680 * The virtual color is made a subset of the physical color in order to 681 * in minimize virtual cache flushing. 682 * We add in the as to spread out different as. This happens when we 683 * initialize the start count value. 684 * sizeof(struct as) is 60 so we shift by 3 to get into the bit range 685 * that will tend to change. For example, on spitfire based machines 686 * (vcshft == 1) contigous as are spread bu ~6 bins. 687 * vcshft provides for proper virtual color alignment. 688 * In theory cnt should be updated using cas only but if we are off by one 689 * or 2 it is no big deal. 690 * We also keep a start value which is used to randomize on what bin we 691 * start counting when it is time to start another loop. This avoids 692 * contigous allocations of ecache size to point to the same bin. 693 * Why 3? Seems work ok. Better than 7 or anything larger. 694 */ 695 #define PGCLR_LOOPFACTOR 3 696 697 /* 698 * When a bin is empty, and we can't satisfy a color request correctly, 699 * we scan. If we assume that the programs have reasonable spatial 700 * behavior, then it will not be a good idea to use the adjacent color. 701 * Using the adjacent color would result in virtually adjacent addresses 702 * mapping into the same spot in the cache. So, if we stumble across 703 * an empty bin, skip a bunch before looking. After the first skip, 704 * then just look one bin at a time so we don't miss our cache on 705 * every look. Be sure to check every bin. Page_create() will panic 706 * if we miss a page. 707 * 708 * This also explains the `<=' in the for loops in both page_get_freelist() 709 * and page_get_cachelist(). Since we checked the target bin, skipped 710 * a bunch, then continued one a time, we wind up checking the target bin 711 * twice to make sure we get all of them bins. 712 */ 713 #define BIN_STEP 20 714 715 #ifdef VM_STATS 716 struct vmm_vmstats_str { 717 ulong_t pgf_alloc[MMU_PAGE_SIZES]; /* page_get_freelist */ 718 ulong_t pgf_allocok[MMU_PAGE_SIZES]; 719 ulong_t pgf_allocokrem[MMU_PAGE_SIZES]; 720 ulong_t pgf_allocfailed[MMU_PAGE_SIZES]; 721 ulong_t pgf_allocdeferred; 722 ulong_t pgf_allocretry[MMU_PAGE_SIZES]; 723 ulong_t pgc_alloc; /* page_get_cachelist */ 724 ulong_t pgc_allocok; 725 ulong_t pgc_allocokrem; 726 ulong_t pgc_allocokdeferred; 727 ulong_t pgc_allocfailed; 728 ulong_t pgcp_alloc[MMU_PAGE_SIZES]; /* page_get_contig_pages */ 729 ulong_t pgcp_allocfailed[MMU_PAGE_SIZES]; 730 ulong_t pgcp_allocempty[MMU_PAGE_SIZES]; 731 ulong_t pgcp_allocok[MMU_PAGE_SIZES]; 732 ulong_t ptcp[MMU_PAGE_SIZES]; /* page_trylock_contig_pages */ 733 ulong_t ptcpfreethresh[MMU_PAGE_SIZES]; 734 ulong_t ptcpfailexcl[MMU_PAGE_SIZES]; 735 ulong_t ptcpfailszc[MMU_PAGE_SIZES]; 736 ulong_t ptcpfailcage[MMU_PAGE_SIZES]; 737 ulong_t ptcpok[MMU_PAGE_SIZES]; 738 ulong_t pgmf_alloc[MMU_PAGE_SIZES]; /* page_get_mnode_freelist */ 739 ulong_t pgmf_allocfailed[MMU_PAGE_SIZES]; 740 ulong_t pgmf_allocempty[MMU_PAGE_SIZES]; 741 ulong_t pgmf_allocok[MMU_PAGE_SIZES]; 742 ulong_t pgmc_alloc; /* page_get_mnode_cachelist */ 743 ulong_t pgmc_allocfailed; 744 ulong_t pgmc_allocempty; 745 ulong_t pgmc_allocok; 746 ulong_t pladd_free[MMU_PAGE_SIZES]; /* page_list_add/sub */ 747 ulong_t plsub_free[MMU_PAGE_SIZES]; 748 ulong_t pladd_cache; 749 ulong_t plsub_cache; 750 ulong_t plsubpages_szcbig; 751 ulong_t plsubpages_szc0; 752 ulong_t pfs_req[MMU_PAGE_SIZES]; /* page_freelist_split */ 753 ulong_t pfs_demote[MMU_PAGE_SIZES]; 754 ulong_t pfc_coalok[MMU_PAGE_SIZES][MAX_MNODE_MRANGES]; 755 ulong_t ppr_reloc[MMU_PAGE_SIZES]; /* page_relocate */ 756 ulong_t ppr_relocok[MMU_PAGE_SIZES]; 757 ulong_t ppr_relocnoroot[MMU_PAGE_SIZES]; 758 ulong_t ppr_reloc_replnoroot[MMU_PAGE_SIZES]; 759 ulong_t ppr_relocnolock[MMU_PAGE_SIZES]; 760 ulong_t ppr_relocnomem[MMU_PAGE_SIZES]; 761 ulong_t ppr_krelocfail[MMU_PAGE_SIZES]; 762 ulong_t ppr_copyfail; 763 /* page coalesce counter */ 764 ulong_t page_ctrs_coalesce[MMU_PAGE_SIZES][MAX_MNODE_MRANGES]; 765 /* candidates useful */ 766 ulong_t page_ctrs_cands_skip[MMU_PAGE_SIZES][MAX_MNODE_MRANGES]; 767 /* ctrs changed after locking */ 768 ulong_t page_ctrs_changed[MMU_PAGE_SIZES][MAX_MNODE_MRANGES]; 769 /* page_freelist_coalesce failed */ 770 ulong_t page_ctrs_failed[MMU_PAGE_SIZES][MAX_MNODE_MRANGES]; 771 ulong_t page_ctrs_coalesce_all; /* page coalesce all counter */ 772 ulong_t page_ctrs_cands_skip_all; /* candidates useful for all func */ 773 }; 774 extern struct vmm_vmstats_str vmm_vmstats; 775 #endif /* VM_STATS */ 776 777 /* 778 * Used to hold off page relocations into the cage until OBP has completed 779 * its boot-time handoff of its resources to the kernel. 780 */ 781 extern int page_relocate_ready; 782 783 /* 784 * cpu/mmu-dependent vm variables may be reset at bootup. 785 */ 786 extern uint_t mmu_page_sizes; 787 extern uint_t max_mmu_page_sizes; 788 extern uint_t mmu_hashcnt; 789 extern uint_t max_mmu_hashcnt; 790 extern size_t mmu_ism_pagesize; 791 extern int mmu_exported_pagesize_mask; 792 extern uint_t mmu_exported_page_sizes; 793 extern uint_t szc_2_userszc[]; 794 extern uint_t userszc_2_szc[]; 795 796 #define mmu_legacy_page_sizes mmu_exported_page_sizes 797 #define USERSZC_2_SZC(userszc) (userszc_2_szc[userszc]) 798 #define SZC_2_USERSZC(szc) (szc_2_userszc[szc]) 799 800 /* 801 * Platform specific page routines 802 */ 803 extern void mach_page_add(page_t **, page_t *); 804 extern void mach_page_sub(page_t **, page_t *); 805 extern uint_t page_get_pagecolors(uint_t); 806 extern void ppcopy_kernel__relocatable(page_t *, page_t *); 807 #define ppcopy_kernel(p1, p2) ppcopy_kernel__relocatable(p1, p2) 808 809 /* 810 * platform specific large pages for kernel heap support 811 */ 812 extern size_t get_segkmem_lpsize(size_t lpsize); 813 extern size_t mmu_get_kernel_lpsize(size_t lpsize); 814 extern void mmu_init_kernel_pgsz(struct hat *hat); 815 extern void mmu_init_kcontext(); 816 extern uint64_t kcontextreg; 817 818 /* 819 * Nucleus data page allocator routines 820 */ 821 extern void ndata_alloc_init(struct memlist *, uintptr_t, uintptr_t); 822 extern void *ndata_alloc(struct memlist *, size_t, size_t); 823 extern void *ndata_extra_base(struct memlist *, size_t, caddr_t); 824 extern size_t ndata_maxsize(struct memlist *); 825 extern size_t ndata_spare(struct memlist *, size_t, size_t); 826 827 /* 828 * Platform specific support for non-coherent I-cache and soft exec 829 */ 830 extern uint_t icache_is_coherent; 831 extern uint_t force_sync_icache_after_bcopy; 832 extern uint_t force_sync_icache_after_dma; 833 834 extern void mach_setup_icache(uint_t); 835 #pragma weak mach_setup_icache 836 837 #ifdef __cplusplus 838 } 839 #endif 840 841 #endif /* _VM_DEP_H */ 842