xref: /titanic_52/usr/src/uts/i86pc/vm/vm_dep.h (revision 18c2aff776a775d34a4c9893a4c72e0434d68e36)
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 2006 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 #pragma ident	"%Z%%M%	%I%	%E% SMI"
34 
35 #ifdef	__cplusplus
36 extern "C" {
37 #endif
38 
39 #include <sys/clock.h>
40 #include <vm/hat_pte.h>
41 
42 /*
43  * WARNING: vm_dep.h is included by files in common. As such, macros
44  * dependent upon PTE36 such as LARGEPAGESIZE cannot be used in this file.
45  */
46 
47 #define	GETTICK()	tsc_read()
48 
49 /* memranges in descending order */
50 extern pfn_t		*memranges;
51 
52 #define	MEMRANGEHI(mtype)						\
53 	((mtype > 0) ? memranges[mtype - 1] - 1: physmax)
54 #define	MEMRANGELO(mtype)	(memranges[mtype])
55 
56 #define	MTYPE_FREEMEM(mt)						\
57 	(mnoderanges[mt].mnr_mt_clpgcnt +				\
58 	    mnoderanges[mt].mnr_mt_flpgcnt +				\
59 	    mnoderanges[mt].mnr_mt_lgpgcnt)
60 
61 /*
62  * combined memory ranges from mnode and memranges[] to manage single
63  * mnode/mtype dimension in the page lists.
64  */
65 typedef struct {
66 	pfn_t	mnr_pfnlo;
67 	pfn_t	mnr_pfnhi;
68 	int	mnr_mnode;
69 	int	mnr_memrange;		/* index into memranges[] */
70 	/* maintain page list stats */
71 	pgcnt_t	mnr_mt_pgmax;		/* mnode/mtype max page cnt */
72 	pgcnt_t	mnr_mt_clpgcnt;		/* cache list cnt */
73 	pgcnt_t	mnr_mt_flpgcnt;		/* free list cnt - small pages */
74 	pgcnt_t	mnr_mt_lgpgcnt;		/* free list cnt - large pages */
75 #ifdef DEBUG
76 	struct mnr_mts {		/* mnode/mtype szc stats */
77 		pgcnt_t	mnr_mts_pgcnt;
78 		int	mnr_mts_colors;
79 		pgcnt_t *mnr_mtsc_pgcnt;
80 	} 	*mnr_mts;
81 #endif
82 } mnoderange_t;
83 
84 #ifdef DEBUG
85 #define	PLCNT_SZ(ctrs_sz) {						\
86 	int	szc, colors;						\
87 	ctrs_sz += mnoderangecnt * sizeof (struct mnr_mts) *		\
88 	    mmu_page_sizes;						\
89 	for (szc = 0; szc < mmu_page_sizes; szc++) {			\
90 		colors = page_get_pagecolors(szc);			\
91 		ctrs_sz += mnoderangecnt * sizeof (pgcnt_t) * colors;	\
92 	}								\
93 }
94 
95 #define	PLCNT_INIT(addr) {						\
96 	int	mt, szc, colors;					\
97 	for (mt = 0; mt < mnoderangecnt; mt++) {			\
98 		mnoderanges[mt].mnr_mts = (struct mnr_mts *)addr;	\
99 		addr += (sizeof (struct mnr_mts) * mmu_page_sizes);	\
100 		for (szc = 0; szc < mmu_page_sizes; szc++) {		\
101 			colors = page_get_pagecolors(szc);		\
102 			mnoderanges[mt].mnr_mts[szc].mnr_mts_colors =	\
103 			    colors;					\
104 			mnoderanges[mt].mnr_mts[szc].mnr_mtsc_pgcnt =	\
105 			    (pgcnt_t *)addr;				\
106 			addr += (sizeof (pgcnt_t) * colors);		\
107 		}							\
108 	}								\
109 }
110 #define	PLCNT_DO(pp, mtype, szc, cnt, flags) {				\
111 	int	bin = PP_2_BIN(pp);					\
112 	if (flags & PG_CACHE_LIST)					\
113 		atomic_add_long(&mnoderanges[mtype].			\
114 		    mnr_mt_clpgcnt, cnt);				\
115 	else if (szc)							\
116 		atomic_add_long(&mnoderanges[mtype].			\
117 		    mnr_mt_lgpgcnt, cnt);				\
118 	else								\
119 		atomic_add_long(&mnoderanges[mtype].			\
120 		    mnr_mt_flpgcnt, cnt);				\
121 	atomic_add_long(&mnoderanges[mtype].mnr_mts[szc].		\
122 	    mnr_mts_pgcnt, cnt);					\
123 	atomic_add_long(&mnoderanges[mtype].mnr_mts[szc].		\
124 	    mnr_mtsc_pgcnt[bin], cnt);					\
125 }
126 #else
127 #define	PLCNT_SZ(ctrs_sz)
128 #define	PLCNT_INIT(base)
129 #define	PLCNT_DO(pp, mtype, szc, cnt, flags) {				\
130 	if (flags & PG_CACHE_LIST)					\
131 		atomic_add_long(&mnoderanges[mtype].			\
132 		    mnr_mt_clpgcnt, cnt);				\
133 	else if (szc)							\
134 		atomic_add_long(&mnoderanges[mtype].			\
135 		    mnr_mt_lgpgcnt, cnt);				\
136 	else								\
137 		atomic_add_long(&mnoderanges[mtype].			\
138 		    mnr_mt_flpgcnt, cnt);				\
139 }
140 #endif
141 
142 #define	PLCNT_INCR(pp, mnode, mtype, szc, flags) {			\
143 	long	cnt = (1 << PAGE_BSZS_SHIFT(szc));			\
144 	ASSERT(mtype == PP_2_MTYPE(pp));				\
145 	if (physmax4g && mtype <= mtype4g)				\
146 		atomic_add_long(&freemem4g, cnt);			\
147 	PLCNT_DO(pp, mtype, szc, cnt, flags);				\
148 }
149 
150 #define	PLCNT_DECR(pp, mnode, mtype, szc, flags) {			\
151 	long	cnt = ((-1) << PAGE_BSZS_SHIFT(szc));			\
152 	ASSERT(mtype == PP_2_MTYPE(pp));				\
153 	if (physmax4g && mtype <= mtype4g)				\
154 		atomic_add_long(&freemem4g, cnt);			\
155 	PLCNT_DO(pp, mtype, szc, cnt, flags);				\
156 }
157 
158 /*
159  * macros to update page list max counts.  no-op on x86.
160  */
161 #define	PLCNT_XFER_NORELOC(pp)
162 
163 #define	PLCNT_MODIFY_MAX(pfn, cnt)	mtype_modify_max(pfn, (pgcnt_t)cnt)
164 
165 extern mnoderange_t	*mnoderanges;
166 extern int		mnoderangecnt;
167 extern int		mtype4g;
168 
169 /*
170  * 4g memory management variables for systems with more than 4g of memory:
171  *
172  * physical memory below 4g is required for 32bit dma devices and, currently,
173  * for kmem memory. On systems with more than 4g of memory, the pool of memory
174  * below 4g can be depleted without any paging activity given that there is
175  * likely to be sufficient memory above 4g.
176  *
177  * physmax4g is set true if the largest pfn is over 4g. The rest of the
178  * 4g memory management code is enabled only when physmax4g is true.
179  *
180  * maxmem4g is the count of the maximum number of pages on the page lists
181  * with physical addresses below 4g. It can be a lot less then 4g given that
182  * BIOS may reserve large chunks of space below 4g for hot plug pci devices,
183  * agp aperture etc.
184  *
185  * freemem4g maintains the count of the number of available pages on the
186  * page lists with physical addresses below 4g.
187  *
188  * DESFREE4G specifies the desired amount of below 4g memory. It defaults to
189  * 6% (desfree4gshift = 4) of maxmem4g.
190  *
191  * RESTRICT4G_ALLOC returns true if freemem4g falls below DESFREE4G
192  * and the amount of physical memory above 4g is greater than freemem4g.
193  * In this case, page_get_* routines will restrict below 4g allocations
194  * for requests that don't specifically require it.
195  */
196 
197 extern int		physmax4g;
198 extern pgcnt_t		maxmem4g;
199 extern pgcnt_t		freemem4g;
200 extern int		lotsfree4gshift;
201 extern int		desfree4gshift;
202 #define	LOTSFREE4G	(maxmem4g >> lotsfree4gshift)
203 #define	DESFREE4G	(maxmem4g >> desfree4gshift)
204 
205 #define	RESTRICT4G_ALLOC					\
206 	(physmax4g && (freemem4g < DESFREE4G) && ((freemem4g << 1) < freemem))
207 
208 /*
209  * 16m memory management:
210  *
211  * reserve some amount of physical memory below 16m for legacy devices.
212  *
213  * RESTRICT16M_ALLOC returns true if an there are sufficient free pages above
214  * 16m or if the 16m pool drops below DESFREE16M.
215  *
216  * In this case, general page allocations via page_get_{free,cache}list
217  * routines will be restricted from allocating from the 16m pool. Allocations
218  * that require specific pfn ranges (page_get_anylist) and PG_PANIC allocations
219  * are not restricted.
220  */
221 
222 #define	FREEMEM16M	MTYPE_FREEMEM(0)
223 #define	DESFREE16M	desfree16m
224 #define	RESTRICT16M_ALLOC(freemem, pgcnt, flags)		\
225 	((freemem != 0) && ((flags & PG_PANIC) == 0) &&		\
226 	    ((freemem >= (FREEMEM16M)) ||			\
227 	    (FREEMEM16M  < (DESFREE16M + pgcnt))))
228 extern pgcnt_t		desfree16m;
229 
230 extern int		restricted_kmemalloc;
231 extern int		memrange_num(pfn_t);
232 extern int		pfn_2_mtype(pfn_t);
233 extern int		mtype_func(int, int, uint_t);
234 extern void		mtype_modify_max(pfn_t, long);
235 extern int		mnode_pgcnt(int);
236 
237 #define	NUM_MEM_RANGES	4		/* memory range types */
238 
239 /*
240  * Per page size free lists. Allocated dynamically.
241  * dimensions [mtype][mmu_page_sizes][colors]
242  *
243  * mtype specifies a physical memory range with a unique mnode.
244  */
245 
246 extern page_t ****page_freelists;
247 
248 #define	PAGE_FREELISTS(mnode, szc, color, mtype)		\
249 	(*(page_freelists[mtype][szc] + (color)))
250 
251 /*
252  * For now there is only a single size cache list. Allocated dynamically.
253  * dimensions [mtype][colors]
254  *
255  * mtype specifies a physical memory range with a unique mnode.
256  */
257 extern page_t ***page_cachelists;
258 
259 #define	PAGE_CACHELISTS(mnode, color, mtype) 		\
260 	(*(page_cachelists[mtype] + (color)))
261 
262 /*
263  * There are mutexes for both the page freelist
264  * and the page cachelist.  We want enough locks to make contention
265  * reasonable, but not too many -- otherwise page_freelist_lock() gets
266  * so expensive that it becomes the bottleneck!
267  */
268 
269 #define	NPC_MUTEX	16
270 
271 extern kmutex_t	*fpc_mutex[NPC_MUTEX];
272 extern kmutex_t	*cpc_mutex[NPC_MUTEX];
273 
274 extern page_t *page_get_mnode_freelist(int, uint_t, int, uchar_t, uint_t);
275 extern page_t *page_get_mnode_cachelist(uint_t, uint_t, int, int);
276 
277 /* Find the bin for the given page if it was of size szc */
278 #define	PP_2_BIN_SZC(pp, szc)						\
279 	(((pp->p_pagenum) & page_colors_mask) >>			\
280 	(hw_page_array[szc].hp_shift - hw_page_array[0].hp_shift))
281 
282 #define	PP_2_BIN(pp)		(PP_2_BIN_SZC(pp, pp->p_szc))
283 
284 #define	PP_2_MEM_NODE(pp)	(PFN_2_MEM_NODE(pp->p_pagenum))
285 #define	PP_2_MTYPE(pp)		(pfn_2_mtype(pp->p_pagenum))
286 #define	PP_2_SZC(pp)		(pp->p_szc)
287 
288 #define	SZCPAGES(szc)		(1 << PAGE_BSZS_SHIFT(szc))
289 #define	PFN_BASE(pfnum, szc)	(pfnum & ~(SZCPAGES(szc) - 1))
290 
291 extern struct cpu	cpus[];
292 #define	CPU0		cpus
293 
294 #if defined(__amd64)
295 
296 /*
297  * set the mtype range (called from page_get_{free,cache}list)
298  *   - set range to above 4g if the system has more than 4g of memory and the
299  *   amount of memory below 4g runs low. If not, set range to above 16m if
300  *   16m threshold is reached otherwise set range to all of memory
301  *   starting from the hi pfns.
302  *
303  * page_get_anylist gets its mtype range from the specified ddi_dma_attr_t.
304  */
305 #define	MTYPE_INIT(mtype, vp, vaddr, flags, pgsz) {			\
306 	mtype = mnoderangecnt - 1;					\
307 	if (RESTRICT4G_ALLOC) {						\
308 		VM_STAT_ADD(vmm_vmstats.restrict4gcnt);			\
309 		/* here only for > 4g systems */			\
310 		flags |= PGI_MT_RANGE4G;				\
311 	} else if (RESTRICT16M_ALLOC(freemem, btop(pgsz), flags)) {	\
312 		flags |= PGI_MT_RANGE16M;				\
313 	} else {							\
314 		VM_STAT_ADD(vmm_vmstats.unrestrict16mcnt);		\
315 		VM_STAT_COND_ADD((flags & PG_PANIC),			\
316 		    vmm_vmstats.pgpanicalloc);				\
317 		flags |= PGI_MT_RANGE0;					\
318 	}								\
319 }
320 
321 #elif defined(__i386)
322 
323 /*
324  * set the mtype range
325  *   - kmem requests needs to be below 4g if restricted_kmemalloc is set.
326  *   - for non kmem requests, set range to above 4g if the amount of memory
327  *   below 4g runs low.
328  */
329 
330 #define	MTYPE_INIT(mtype, vp, vaddr, flags, pgsz) {			\
331 	if (restricted_kmemalloc && (vp) == &kvp &&			\
332 	    (caddr_t)(vaddr) >= kernelheap &&				\
333 	    (caddr_t)(vaddr) < ekernelheap) {				\
334 		ASSERT(physmax4g);					\
335 		mtype = mtype4g;					\
336 		if (RESTRICT16M_ALLOC(freemem4g - btop(pgsz),		\
337 		    btop(pgsz), flags)) {				\
338 			flags |= PGI_MT_RANGE16M;			\
339 		} else {						\
340 			VM_STAT_ADD(vmm_vmstats.unrestrict16mcnt);	\
341 			VM_STAT_COND_ADD((flags & PG_PANIC),		\
342 			    vmm_vmstats.pgpanicalloc);			\
343 			flags |= PGI_MT_RANGE0;				\
344 		}							\
345 	} else {							\
346 		mtype = mnoderangecnt - 1;				\
347 		if (RESTRICT4G_ALLOC) {					\
348 			VM_STAT_ADD(vmm_vmstats.restrict4gcnt);		\
349 			/* here only for > 4g systems */		\
350 			flags |= PGI_MT_RANGE4G;			\
351 		} else if (RESTRICT16M_ALLOC(freemem, btop(pgsz),	\
352 		    flags)) {						\
353 			flags |= PGI_MT_RANGE16M;			\
354 		} else {						\
355 			VM_STAT_ADD(vmm_vmstats.unrestrict16mcnt);	\
356 			VM_STAT_COND_ADD((flags & PG_PANIC),		\
357 			    vmm_vmstats.pgpanicalloc);			\
358 			flags |= PGI_MT_RANGE0;				\
359 		}							\
360 	}								\
361 }
362 
363 #endif	/* __i386 */
364 
365 /*
366  * macros to loop through the mtype range (page_get_mnode_{free,cache,any}list,
367  * and page_get_contig_pages)
368  *
369  * MTYPE_START sets the initial mtype. -1 if the mtype range specified does
370  * not contain mnode.
371  *
372  * MTYPE_NEXT sets the next mtype. -1 if there are no more valid
373  * mtype in the range.
374  */
375 
376 #define	MTYPE_START(mnode, mtype, flags)				\
377 	(mtype = mtype_func(mnode, mtype, flags))
378 
379 #define	MTYPE_NEXT(mnode, mtype, flags) {				\
380 	if (flags & PGI_MT_RANGE) {					\
381 		mtype = mtype_func(mnode, mtype, flags | PGI_MT_NEXT);	\
382 	} else {							\
383 		mtype = -1;						\
384 	}								\
385 }
386 
387 /* mtype init for page_get_replacement_page */
388 
389 #define	MTYPE_PGR_INIT(mtype, flags, pp, mnode, pgcnt) {		\
390 	mtype = mnoderangecnt - 1;					\
391 	if (RESTRICT16M_ALLOC(freemem, pgcnt, flags)) {			\
392 		flags |= PGI_MT_RANGE16M;				\
393 	} else {							\
394 		VM_STAT_ADD(vmm_vmstats.unrestrict16mcnt);		\
395 		flags |= PGI_MT_RANGE0;					\
396 	}								\
397 }
398 
399 #define	MNODE_PGCNT(mnode)		mnode_pgcnt(mnode)
400 
401 #define	MNODETYPE_2_PFN(mnode, mtype, pfnlo, pfnhi)			\
402 	ASSERT(mnoderanges[mtype].mnr_mnode == mnode);			\
403 	pfnlo = mnoderanges[mtype].mnr_pfnlo;				\
404 	pfnhi = mnoderanges[mtype].mnr_pfnhi;
405 
406 #define	PC_BIN_MUTEX(mnode, bin, flags) ((flags & PG_FREE_LIST) ?	\
407 	&fpc_mutex[(bin) & (NPC_MUTEX - 1)][mnode] :			\
408 	&cpc_mutex[(bin) & (NPC_MUTEX - 1)][mnode])
409 
410 #define	FPC_MUTEX(mnode, i)	(&fpc_mutex[i][mnode])
411 #define	CPC_MUTEX(mnode, i)	(&cpc_mutex[i][mnode])
412 
413 #ifdef DEBUG
414 #define	CHK_LPG(pp, szc)	chk_lpg(pp, szc)
415 extern void	chk_lpg(page_t *, uchar_t);
416 #else
417 #define	CHK_LPG(pp, szc)
418 #endif
419 
420 #define	FULL_REGION_CNT(rg_szc)	\
421 	(LEVEL_SIZE(rg_szc) >> LEVEL_SHIFT(rg_szc - 1))
422 
423 /* Return the leader for this mapping size */
424 #define	PP_GROUPLEADER(pp, szc) \
425 	(&(pp)[-(int)((pp)->p_pagenum & (SZCPAGES(szc)-1))])
426 
427 /* Return the root page for this page based on p_szc */
428 #define	PP_PAGEROOT(pp) ((pp)->p_szc == 0 ? (pp) : \
429 	PP_GROUPLEADER((pp), (pp)->p_szc))
430 
431 /*
432  * The counter base must be per page_counter element to prevent
433  * races when re-indexing, and the base page size element should
434  * be aligned on a boundary of the given region size.
435  *
436  * We also round up the number of pages spanned by the counters
437  * for a given region to PC_BASE_ALIGN in certain situations to simplify
438  * the coding for some non-performance critical routines.
439  */
440 
441 #define	PC_BASE_ALIGN		((pfn_t)1 << PAGE_BSZS_SHIFT(MMU_PAGE_SIZES-1))
442 #define	PC_BASE_ALIGN_MASK	(PC_BASE_ALIGN - 1)
443 
444 /*
445  * cpu/mmu-dependent vm variables
446  */
447 extern uint_t mmu_page_sizes;
448 extern uint_t mmu_exported_page_sizes;
449 
450 /* For x86, userszc is the same as the kernel's szc */
451 #define	USERSZC_2_SZC(userszc)	(userszc)
452 #define	SZC_2_USERSZC(szc)	(szc)
453 
454 /*
455  * for hw_page_map_t, sized to hold the ratio of large page to base
456  * pagesize (1024 max)
457  */
458 typedef	short	hpmctr_t;
459 
460 /*
461  * get the setsize of the current cpu - assume homogenous for x86
462  */
463 extern int	l2cache_sz, l2cache_linesz, l2cache_assoc;
464 
465 #define	L2CACHE_ALIGN		l2cache_linesz
466 #define	L2CACHE_ALIGN_MAX	64
467 #define	CPUSETSIZE()		\
468 	(l2cache_assoc ? (l2cache_sz / l2cache_assoc) : MMU_PAGESIZE)
469 
470 /*
471  * Return the log2(pagesize(szc) / MMU_PAGESIZE) --- or the shift count
472  * for the number of base pages in this pagesize
473  */
474 #define	PAGE_BSZS_SHIFT(szc) (LEVEL_SHIFT(szc) - MMU_PAGESHIFT)
475 
476 /*
477  * Internal PG_ flags.
478  */
479 #define	PGI_RELOCONLY	0x010000	/* opposite of PG_NORELOC */
480 #define	PGI_NOCAGE	0x020000	/* cage is disabled */
481 #define	PGI_PGCPHIPRI	0x040000	/* page_get_contig_page pri alloc */
482 #define	PGI_PGCPSZC0	0x080000	/* relocate base pagesize page */
483 
484 /*
485  * PGI range flags - should not overlap PGI flags
486  */
487 #define	PGI_MT_RANGE0	0x1000000	/* mtype range to 0 */
488 #define	PGI_MT_RANGE16M	0x2000000	/* mtype range to 16m */
489 #define	PGI_MT_RANGE4G	0x4000000	/* mtype range to 4g */
490 #define	PGI_MT_NEXT	0x8000000	/* get next mtype */
491 #define	PGI_MT_RANGE	(PGI_MT_RANGE0 | PGI_MT_RANGE16M | PGI_MT_RANGE4G)
492 
493 /*
494  * hash as and addr to get a bin.
495  */
496 
497 #define	AS_2_BIN(as, seg, vp, addr, bin)				\
498 	bin = ((((uintptr_t)(addr) >> PAGESHIFT) + ((uintptr_t)(as) >> 4)) \
499 	    & page_colors_mask)
500 
501 /*
502  * cpu private vm data - accessed thru CPU->cpu_vm_data
503  *	vc_pnum_memseg: tracks last memseg visited in page_numtopp_nolock()
504  *	vc_pnext_memseg: tracks last memseg visited in page_nextn()
505  *	vc_kmptr: orignal unaligned kmem pointer for this vm_cpu_data_t
506  *	vc_kmsize: orignal kmem size for this vm_cpu_data_t
507  */
508 
509 typedef struct {
510 	struct memseg	*vc_pnum_memseg;
511 	struct memseg	*vc_pnext_memseg;
512 	void		*vc_kmptr;
513 	size_t		vc_kmsize;
514 } vm_cpu_data_t;
515 
516 /* allocation size to ensure vm_cpu_data_t resides in its own cache line */
517 #define	VM_CPU_DATA_PADSIZE						\
518 	(P2ROUNDUP(sizeof (vm_cpu_data_t), L2CACHE_ALIGN_MAX))
519 
520 /* for boot cpu before kmem is initialized */
521 extern char	vm_cpu_data0[];
522 
523 /*
524  * When a bin is empty, and we can't satisfy a color request correctly,
525  * we scan.  If we assume that the programs have reasonable spatial
526  * behavior, then it will not be a good idea to use the adjacent color.
527  * Using the adjacent color would result in virtually adjacent addresses
528  * mapping into the same spot in the cache.  So, if we stumble across
529  * an empty bin, skip a bunch before looking.  After the first skip,
530  * then just look one bin at a time so we don't miss our cache on
531  * every look. Be sure to check every bin.  Page_create() will panic
532  * if we miss a page.
533  *
534  * This also explains the `<=' in the for loops in both page_get_freelist()
535  * and page_get_cachelist().  Since we checked the target bin, skipped
536  * a bunch, then continued one a time, we wind up checking the target bin
537  * twice to make sure we get all of them bins.
538  */
539 #define	BIN_STEP	19
540 
541 #ifdef VM_STATS
542 struct vmm_vmstats_str {
543 	ulong_t pgf_alloc[MMU_PAGE_SIZES];	/* page_get_freelist */
544 	ulong_t pgf_allocok[MMU_PAGE_SIZES];
545 	ulong_t pgf_allocokrem[MMU_PAGE_SIZES];
546 	ulong_t pgf_allocfailed[MMU_PAGE_SIZES];
547 	ulong_t	pgf_allocdeferred;
548 	ulong_t	pgf_allocretry[MMU_PAGE_SIZES];
549 	ulong_t pgc_alloc;			/* page_get_cachelist */
550 	ulong_t pgc_allocok;
551 	ulong_t pgc_allocokrem;
552 	ulong_t pgc_allocokdeferred;
553 	ulong_t pgc_allocfailed;
554 	ulong_t	pgcp_alloc[MMU_PAGE_SIZES];	/* page_get_contig_pages */
555 	ulong_t	pgcp_allocfailed[MMU_PAGE_SIZES];
556 	ulong_t	pgcp_allocempty[MMU_PAGE_SIZES];
557 	ulong_t	pgcp_allocok[MMU_PAGE_SIZES];
558 	ulong_t	ptcp[MMU_PAGE_SIZES];		/* page_trylock_contig_pages */
559 	ulong_t	ptcpfreethresh[MMU_PAGE_SIZES];
560 	ulong_t	ptcpfailexcl[MMU_PAGE_SIZES];
561 	ulong_t	ptcpfailszc[MMU_PAGE_SIZES];
562 	ulong_t	ptcpfailcage[MMU_PAGE_SIZES];
563 	ulong_t	ptcpok[MMU_PAGE_SIZES];
564 	ulong_t	pgmf_alloc[MMU_PAGE_SIZES];	/* page_get_mnode_freelist */
565 	ulong_t	pgmf_allocfailed[MMU_PAGE_SIZES];
566 	ulong_t	pgmf_allocempty[MMU_PAGE_SIZES];
567 	ulong_t	pgmf_allocok[MMU_PAGE_SIZES];
568 	ulong_t	pgmc_alloc;			/* page_get_mnode_cachelist */
569 	ulong_t	pgmc_allocfailed;
570 	ulong_t	pgmc_allocempty;
571 	ulong_t	pgmc_allocok;
572 	ulong_t	pladd_free[MMU_PAGE_SIZES];	/* page_list_add/sub */
573 	ulong_t	plsub_free[MMU_PAGE_SIZES];
574 	ulong_t	pladd_cache;
575 	ulong_t	plsub_cache;
576 	ulong_t	plsubpages_szcbig;
577 	ulong_t	plsubpages_szc0;
578 	ulong_t	pff_req[MMU_PAGE_SIZES];	/* page_freelist_fill */
579 	ulong_t	pff_demote[MMU_PAGE_SIZES];
580 	ulong_t	pff_coalok[MMU_PAGE_SIZES];
581 	ulong_t	ppr_reloc[MMU_PAGE_SIZES];	/* page_relocate */
582 	ulong_t ppr_relocnoroot[MMU_PAGE_SIZES];
583 	ulong_t ppr_reloc_replnoroot[MMU_PAGE_SIZES];
584 	ulong_t ppr_relocnolock[MMU_PAGE_SIZES];
585 	ulong_t ppr_relocnomem[MMU_PAGE_SIZES];
586 	ulong_t ppr_relocok[MMU_PAGE_SIZES];
587 	ulong_t page_ctrs_coalesce;	/* page coalesce counter */
588 	ulong_t page_ctrs_cands_skip;	/* candidates useful */
589 	ulong_t page_ctrs_changed;	/* ctrs changed after locking */
590 	ulong_t page_ctrs_failed;	/* page_freelist_coalesce failed */
591 	ulong_t page_ctrs_coalesce_all;	/* page coalesce all counter */
592 	ulong_t page_ctrs_cands_skip_all; /* candidates useful for all func */
593 	ulong_t	restrict4gcnt;
594 	ulong_t	unrestrict16mcnt;	/* non-DMA 16m allocs allowed */
595 	ulong_t	pgpanicalloc;		/* PG_PANIC allocation */
596 };
597 extern struct vmm_vmstats_str vmm_vmstats;
598 #endif	/* VM_STATS */
599 
600 extern size_t page_ctrs_sz(void);
601 extern caddr_t page_ctrs_alloc(caddr_t);
602 extern void page_ctr_sub(int, int, page_t *, int);
603 extern page_t *page_freelist_fill(uchar_t, int, int, int, pfn_t);
604 extern uint_t page_get_pagecolors(uint_t);
605 
606 #ifdef	__cplusplus
607 }
608 #endif
609 
610 #endif	/* _VM_DEP_H */
611