xref: /freebsd/sys/vm/uma.h (revision 54ebdd631db8c0bba2baab0155f603a8b5cf014a)
1 /*-
2  * Copyright (c) 2002, 2003, 2004, 2005 Jeffrey Roberson <jeff@FreeBSD.org>
3  * Copyright (c) 2004, 2005 Bosko Milekic <bmilekic@FreeBSD.org>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice unmodified, this list of conditions, and the following
11  *    disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26  *
27  * $FreeBSD$
28  *
29  */
30 
31 /*
32  * uma.h - External definitions for the Universal Memory Allocator
33  *
34 */
35 
36 #ifndef VM_UMA_H
37 #define VM_UMA_H
38 
39 #include <sys/param.h>		/* For NULL */
40 #include <sys/malloc.h>		/* For M_* */
41 
42 /* User visible parameters */
43 #define UMA_SMALLEST_UNIT       (PAGE_SIZE / 256) /* Smallest item allocated */
44 
45 /* Types and type defs */
46 
47 struct uma_zone;
48 /* Opaque type used as a handle to the zone */
49 typedef struct uma_zone * uma_zone_t;
50 
51 void zone_drain(uma_zone_t);
52 
53 /*
54  * Item constructor
55  *
56  * Arguments:
57  *	item  A pointer to the memory which has been allocated.
58  *	arg   The arg field passed to uma_zalloc_arg
59  *	size  The size of the allocated item
60  *	flags See zalloc flags
61  *
62  * Returns:
63  *	0      on success
64  *      errno  on failure
65  *
66  * Discussion:
67  *	The constructor is called just before the memory is returned
68  *	to the user. It may block if necessary.
69  */
70 typedef int (*uma_ctor)(void *mem, int size, void *arg, int flags);
71 
72 /*
73  * Item destructor
74  *
75  * Arguments:
76  *	item  A pointer to the memory which has been allocated.
77  *	size  The size of the item being destructed.
78  *	arg   Argument passed through uma_zfree_arg
79  *
80  * Returns:
81  *	Nothing
82  *
83  * Discussion:
84  *	The destructor may perform operations that differ from those performed
85  *	by the initializer, but it must leave the object in the same state.
86  *	This IS type stable storage.  This is called after EVERY zfree call.
87  */
88 typedef void (*uma_dtor)(void *mem, int size, void *arg);
89 
90 /*
91  * Item initializer
92  *
93  * Arguments:
94  *	item  A pointer to the memory which has been allocated.
95  *	size  The size of the item being initialized.
96  *	flags See zalloc flags
97  *
98  * Returns:
99  *	0      on success
100  *      errno  on failure
101  *
102  * Discussion:
103  *	The initializer is called when the memory is cached in the uma zone.
104  *	The initializer and the destructor should leave the object in the same
105  *	state.
106  */
107 typedef int (*uma_init)(void *mem, int size, int flags);
108 
109 /*
110  * Item discard function
111  *
112  * Arguments:
113  * 	item  A pointer to memory which has been 'freed' but has not left the
114  *	      zone's cache.
115  *	size  The size of the item being discarded.
116  *
117  * Returns:
118  *	Nothing
119  *
120  * Discussion:
121  *	This routine is called when memory leaves a zone and is returned to the
122  *	system for other uses.  It is the counter-part to the init function.
123  */
124 typedef void (*uma_fini)(void *mem, int size);
125 
126 /*
127  * What's the difference between initializing and constructing?
128  *
129  * The item is initialized when it is cached, and this is the state that the
130  * object should be in when returned to the allocator. The purpose of this is
131  * to remove some code which would otherwise be called on each allocation by
132  * utilizing a known, stable state.  This differs from the constructor which
133  * will be called on EVERY allocation.
134  *
135  * For example, in the initializer you may want to initialize embedded locks,
136  * NULL list pointers, set up initial states, magic numbers, etc.  This way if
137  * the object is held in the allocator and re-used it won't be necessary to
138  * re-initialize it.
139  *
140  * The constructor may be used to lock a data structure, link it on to lists,
141  * bump reference counts or total counts of outstanding structures, etc.
142  *
143  */
144 
145 
146 /* Function proto types */
147 
148 /*
149  * Create a new uma zone
150  *
151  * Arguments:
152  *	name  The text name of the zone for debugging and stats. This memory
153  *		should not be freed until the zone has been deallocated.
154  *	size  The size of the object that is being created.
155  *	ctor  The constructor that is called when the object is allocated.
156  *	dtor  The destructor that is called when the object is freed.
157  *	init  An initializer that sets up the initial state of the memory.
158  *	fini  A discard function that undoes initialization done by init.
159  *		ctor/dtor/init/fini may all be null, see notes above.
160  *	align A bitmask that corresponds to the requested alignment
161  *		eg 4 would be 0x3
162  *	flags A set of parameters that control the behavior of the zone.
163  *
164  * Returns:
165  *	A pointer to a structure which is intended to be opaque to users of
166  *	the interface.  The value may be null if the wait flag is not set.
167  */
168 uma_zone_t uma_zcreate(char *name, size_t size, uma_ctor ctor, uma_dtor dtor,
169 			uma_init uminit, uma_fini fini, int align,
170 			u_int32_t flags);
171 
172 /*
173  * Create a secondary uma zone
174  *
175  * Arguments:
176  *	name  The text name of the zone for debugging and stats. This memory
177  *		should not be freed until the zone has been deallocated.
178  *	ctor  The constructor that is called when the object is allocated.
179  *	dtor  The destructor that is called when the object is freed.
180  *	zinit  An initializer that sets up the initial state of the memory
181  *		as the object passes from the Keg's slab to the Zone's cache.
182  *	zfini  A discard function that undoes initialization done by init
183  *		as the object passes from the Zone's cache to the Keg's slab.
184  *
185  *		ctor/dtor/zinit/zfini may all be null, see notes above.
186  *		Note that the zinit and zfini specified here are NOT
187  *		exactly the same as the init/fini specified to uma_zcreate()
188  *		when creating a master zone.  These zinit/zfini are called
189  *		on the TRANSITION from keg to zone (and vice-versa). Once
190  *		these are set, the primary zone may alter its init/fini
191  *		(which are called when the object passes from VM to keg)
192  *		using uma_zone_set_init/fini()) as well as its own
193  *		zinit/zfini (unset by default for master zone) with
194  *		uma_zone_set_zinit/zfini() (note subtle 'z' prefix).
195  *
196  *	master  A reference to this zone's Master Zone (Primary Zone),
197  *		which contains the backing Keg for the Secondary Zone
198  *		being added.
199  *
200  * Returns:
201  *	A pointer to a structure which is intended to be opaque to users of
202  *	the interface.  The value may be null if the wait flag is not set.
203  */
204 uma_zone_t uma_zsecond_create(char *name, uma_ctor ctor, uma_dtor dtor,
205 		    uma_init zinit, uma_fini zfini, uma_zone_t master);
206 
207 /*
208  * Definitions for uma_zcreate flags
209  *
210  * These flags share space with UMA_ZFLAGs in uma_int.h.  Be careful not to
211  * overlap when adding new features.  0xf0000000 is in use by uma_int.h.
212  */
213 #define UMA_ZONE_PAGEABLE	0x0001	/* Return items not fully backed by
214 					   physical memory XXX Not yet */
215 #define UMA_ZONE_ZINIT		0x0002	/* Initialize with zeros */
216 #define UMA_ZONE_STATIC		0x0004	/* Statically sized zone */
217 #define UMA_ZONE_OFFPAGE	0x0008	/* Force the slab structure allocation
218 					   off of the real memory */
219 #define UMA_ZONE_MALLOC		0x0010	/* For use by malloc(9) only! */
220 #define UMA_ZONE_NOFREE		0x0020	/* Do not free slabs of this type! */
221 #define UMA_ZONE_MTXCLASS	0x0040	/* Create a new lock class */
222 #define	UMA_ZONE_VM		0x0080	/*
223 					 * Used for internal vm datastructures
224 					 * only.
225 					 */
226 #define	UMA_ZONE_HASH		0x0100	/*
227 					 * Use a hash table instead of caching
228 					 * information in the vm_page.
229 					 */
230 #define	UMA_ZONE_SECONDARY	0x0200	/* Zone is a Secondary Zone */
231 #define	UMA_ZONE_REFCNT		0x0400	/* Allocate refcnts in slabs */
232 #define	UMA_ZONE_MAXBUCKET	0x0800	/* Use largest buckets */
233 
234 /* Definitions for align */
235 #define UMA_ALIGN_PTR	(sizeof(void *) - 1)	/* Alignment fit for ptr */
236 #define UMA_ALIGN_LONG	(sizeof(long) - 1)	/* "" long */
237 #define UMA_ALIGN_INT	(sizeof(int) - 1)	/* "" int */
238 #define UMA_ALIGN_SHORT	(sizeof(short) - 1)	/* "" short */
239 #define UMA_ALIGN_CHAR	(sizeof(char) - 1)	/* "" char */
240 #define UMA_ALIGN_CACHE	(0 - 1)			/* Cache line size align */
241 
242 /*
243  * Destroys an empty uma zone.  If the zone is not empty uma complains loudly.
244  *
245  * Arguments:
246  *	zone  The zone we want to destroy.
247  *
248  */
249 void uma_zdestroy(uma_zone_t zone);
250 
251 /*
252  * Allocates an item out of a zone
253  *
254  * Arguments:
255  *	zone  The zone we are allocating from
256  *	arg   This data is passed to the ctor function
257  *	flags See sys/malloc.h for available flags.
258  *
259  * Returns:
260  *	A non-null pointer to an initialized element from the zone is
261  *	guaranteed if the wait flag is M_WAITOK.  Otherwise a null pointer
262  *	may be returned if the zone is empty or the ctor failed.
263  */
264 
265 void *uma_zalloc_arg(uma_zone_t zone, void *arg, int flags);
266 
267 /*
268  * Allocates an item out of a zone without supplying an argument
269  *
270  * This is just a wrapper for uma_zalloc_arg for convenience.
271  *
272  */
273 static __inline void *uma_zalloc(uma_zone_t zone, int flags);
274 
275 static __inline void *
276 uma_zalloc(uma_zone_t zone, int flags)
277 {
278 	return uma_zalloc_arg(zone, NULL, flags);
279 }
280 
281 /*
282  * Frees an item back into the specified zone.
283  *
284  * Arguments:
285  *	zone  The zone the item was originally allocated out of.
286  *	item  The memory to be freed.
287  *	arg   Argument passed to the destructor
288  *
289  * Returns:
290  *	Nothing.
291  */
292 
293 void uma_zfree_arg(uma_zone_t zone, void *item, void *arg);
294 
295 /*
296  * Frees an item back to a zone without supplying an argument
297  *
298  * This is just a wrapper for uma_zfree_arg for convenience.
299  *
300  */
301 static __inline void uma_zfree(uma_zone_t zone, void *item);
302 
303 static __inline void
304 uma_zfree(uma_zone_t zone, void *item)
305 {
306 	uma_zfree_arg(zone, item, NULL);
307 }
308 
309 /*
310  * XXX The rest of the prototypes in this header are h0h0 magic for the VM.
311  * If you think you need to use it for a normal zone you're probably incorrect.
312  */
313 
314 /*
315  * Backend page supplier routines
316  *
317  * Arguments:
318  *	zone  The zone that is requesting pages.
319  *	size  The number of bytes being requested.
320  *	pflag Flags for these memory pages, see below.
321  *	wait  Indicates our willingness to block.
322  *
323  * Returns:
324  *	A pointer to the allocated memory or NULL on failure.
325  */
326 
327 typedef void *(*uma_alloc)(uma_zone_t zone, int size, u_int8_t *pflag, int wait);
328 
329 /*
330  * Backend page free routines
331  *
332  * Arguments:
333  *	item  A pointer to the previously allocated pages.
334  *	size  The original size of the allocation.
335  *	pflag The flags for the slab.  See UMA_SLAB_* below.
336  *
337  * Returns:
338  *	None
339  */
340 typedef void (*uma_free)(void *item, int size, u_int8_t pflag);
341 
342 
343 
344 /*
345  * Sets up the uma allocator. (Called by vm_mem_init)
346  *
347  * Arguments:
348  *	bootmem  A pointer to memory used to bootstrap the system.
349  *
350  * Returns:
351  *	Nothing
352  *
353  * Discussion:
354  *	This memory is used for zones which allocate things before the
355  *	backend page supplier can give us pages.  It should be
356  *	UMA_SLAB_SIZE * boot_pages bytes. (see uma_int.h)
357  *
358  */
359 
360 void uma_startup(void *bootmem, int boot_pages);
361 
362 /*
363  * Finishes starting up the allocator.  This should
364  * be called when kva is ready for normal allocs.
365  *
366  * Arguments:
367  *	None
368  *
369  * Returns:
370  *	Nothing
371  *
372  * Discussion:
373  *	uma_startup2 is called by kmeminit() to enable us of uma for malloc.
374  */
375 
376 void uma_startup2(void);
377 
378 /*
379  * Reclaims unused memory for all zones
380  *
381  * Arguments:
382  *	None
383  * Returns:
384  *	None
385  *
386  * This should only be called by the page out daemon.
387  */
388 
389 void uma_reclaim(void);
390 
391 /*
392  * Sets the alignment mask to be used for all zones requesting cache
393  * alignment.  Should be called by MD boot code prior to starting VM/UMA.
394  *
395  * Arguments:
396  *	align The alignment mask
397  *
398  * Returns:
399  *	Nothing
400  */
401 void uma_set_align(int align);
402 
403 /*
404  * Switches the backing object of a zone
405  *
406  * Arguments:
407  *	zone  The zone to update.
408  *	obj   The VM object to use for future allocations.
409  *	size  The size of the object to allocate.
410  *
411  * Returns:
412  *	0  if kva space can not be allocated
413  *	1  if successful
414  *
415  * Discussion:
416  *	A NULL object can be used and uma will allocate one for you.  Setting
417  *	the size will limit the amount of memory allocated to this zone.
418  *
419  */
420 struct vm_object;
421 int uma_zone_set_obj(uma_zone_t zone, struct vm_object *obj, int size);
422 
423 /*
424  * Sets a high limit on the number of items allowed in a zone
425  *
426  * Arguments:
427  *	zone  The zone to limit
428  *
429  * Returns:
430  *	Nothing
431  */
432 void uma_zone_set_max(uma_zone_t zone, int nitems);
433 
434 /*
435  * The following two routines (uma_zone_set_init/fini)
436  * are used to set the backend init/fini pair which acts on an
437  * object as it becomes allocated and is placed in a slab within
438  * the specified zone's backing keg.  These should probably not
439  * be changed once allocations have already begun, but only be set
440  * immediately upon zone creation.
441  */
442 void uma_zone_set_init(uma_zone_t zone, uma_init uminit);
443 void uma_zone_set_fini(uma_zone_t zone, uma_fini fini);
444 
445 /*
446  * The following two routines (uma_zone_set_zinit/zfini) are
447  * used to set the zinit/zfini pair which acts on an object as
448  * it passes from the backing Keg's slab cache to the
449  * specified Zone's bucket cache.  These should probably not
450  * be changed once allocations have already begun, but only be set
451  * immediately upon zone creation.
452  */
453 void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit);
454 void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini);
455 
456 /*
457  * Replaces the standard page_alloc or obj_alloc functions for this zone
458  *
459  * Arguments:
460  *	zone   The zone whose backend allocator is being changed.
461  *	allocf A pointer to the allocation function
462  *
463  * Returns:
464  *	Nothing
465  *
466  * Discussion:
467  *	This could be used to implement pageable allocation, or perhaps
468  *	even DMA allocators if used in conjunction with the OFFPAGE
469  *	zone flag.
470  */
471 
472 void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf);
473 
474 /*
475  * Used for freeing memory provided by the allocf above
476  *
477  * Arguments:
478  *	zone  The zone that intends to use this free routine.
479  *	freef The page freeing routine.
480  *
481  * Returns:
482  *	Nothing
483  */
484 
485 void uma_zone_set_freef(uma_zone_t zone, uma_free freef);
486 
487 /*
488  * These flags are setable in the allocf and visible in the freef.
489  */
490 #define UMA_SLAB_BOOT	0x01		/* Slab alloced from boot pages */
491 #define UMA_SLAB_KMEM	0x02		/* Slab alloced from kmem_map */
492 #define UMA_SLAB_KERNEL	0x04		/* Slab alloced from kernel_map */
493 #define UMA_SLAB_PRIV	0x08		/* Slab alloced from priv allocator */
494 #define UMA_SLAB_OFFP	0x10		/* Slab is managed separately  */
495 #define UMA_SLAB_MALLOC	0x20		/* Slab is a large malloc slab */
496 /* 0x40 and 0x80 are available */
497 
498 /*
499  * Used to pre-fill a zone with some number of items
500  *
501  * Arguments:
502  *	zone    The zone to fill
503  *	itemcnt The number of items to reserve
504  *
505  * Returns:
506  *	Nothing
507  *
508  * NOTE: This is blocking and should only be done at startup
509  */
510 void uma_prealloc(uma_zone_t zone, int itemcnt);
511 
512 /*
513  * Used to lookup the reference counter allocated for an item
514  * from a UMA_ZONE_REFCNT zone.  For UMA_ZONE_REFCNT zones,
515  * reference counters are allocated for items and stored in
516  * the underlying slab header.
517  *
518  * Arguments:
519  * 	zone  The UMA_ZONE_REFCNT zone to which the item belongs.
520  *	item  The address of the item for which we want a refcnt.
521  *
522  * Returns:
523  * 	A pointer to a u_int32_t reference counter.
524  */
525 u_int32_t *uma_find_refcnt(uma_zone_t zone, void *item);
526 
527 /*
528  * Used to determine if a fixed-size zone is exhausted.
529  *
530  * Arguments:
531  *	zone    The zone to check
532  *
533  * Returns:
534  * 	Non-zero if zone is exhausted.
535  */
536 int uma_zone_exhausted(uma_zone_t zone);
537 int uma_zone_exhausted_nolock(uma_zone_t zone);
538 
539 /*
540  * Exported statistics structures to be used by user space monitoring tools.
541  * Statistics stream consists of a uma_stream_header, followed by a series of
542  * alternative uma_type_header and uma_type_stat structures.
543  */
544 #define	UMA_STREAM_VERSION	0x00000001
545 struct uma_stream_header {
546 	u_int32_t	ush_version;	/* Stream format version. */
547 	u_int32_t	ush_maxcpus;	/* Value of MAXCPU for stream. */
548 	u_int32_t	ush_count;	/* Number of records. */
549 	u_int32_t	_ush_pad;	/* Pad/reserved field. */
550 };
551 
552 #define	UTH_MAX_NAME	32
553 #define	UTH_ZONE_SECONDARY	0x00000001
554 struct uma_type_header {
555 	/*
556 	 * Static per-zone data, some extracted from the supporting keg.
557 	 */
558 	char		uth_name[UTH_MAX_NAME];
559 	u_int32_t	uth_align;	/* Keg: alignment. */
560 	u_int32_t	uth_size;	/* Keg: requested size of item. */
561 	u_int32_t	uth_rsize;	/* Keg: real size of item. */
562 	u_int32_t	uth_maxpages;	/* Keg: maximum number of pages. */
563 	u_int32_t	uth_limit;	/* Keg: max items to allocate. */
564 
565 	/*
566 	 * Current dynamic zone/keg-derived statistics.
567 	 */
568 	u_int32_t	uth_pages;	/* Keg: pages allocated. */
569 	u_int32_t	uth_keg_free;	/* Keg: items free. */
570 	u_int32_t	uth_zone_free;	/* Zone: items free. */
571 	u_int32_t	uth_bucketsize;	/* Zone: desired bucket size. */
572 	u_int32_t	uth_zone_flags;	/* Zone: flags. */
573 	u_int64_t	uth_allocs;	/* Zone: number of allocations. */
574 	u_int64_t	uth_frees;	/* Zone: number of frees. */
575 	u_int64_t	uth_fails;	/* Zone: number of alloc failures. */
576 	u_int64_t	_uth_reserved1[3];	/* Reserved. */
577 };
578 
579 struct uma_percpu_stat {
580 	u_int64_t	ups_allocs;	/* Cache: number of allocations. */
581 	u_int64_t	ups_frees;	/* Cache: number of frees. */
582 	u_int64_t	ups_cache_free;	/* Cache: free items in cache. */
583 	u_int64_t	_ups_reserved[5];	/* Reserved. */
584 };
585 
586 #endif
587