]>
Commit | Line | Data |
---|---|---|
716154c5 BB |
1 | /*****************************************************************************\ |
2 | * Copyright (C) 2007-2010 Lawrence Livermore National Security, LLC. | |
3 | * Copyright (C) 2007 The Regents of the University of California. | |
4 | * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER). | |
5 | * Written by Brian Behlendorf <behlendorf1@llnl.gov>. | |
715f6251 | 6 | * UCRL-CODE-235197 |
7 | * | |
716154c5 | 8 | * This file is part of the SPL, Solaris Porting Layer. |
3d6af2dd | 9 | * For details, see <http://zfsonlinux.org/>. |
716154c5 BB |
10 | * |
11 | * The SPL is free software; you can redistribute it and/or modify it | |
12 | * under the terms of the GNU General Public License as published by the | |
13 | * Free Software Foundation; either version 2 of the License, or (at your | |
14 | * option) any later version. | |
715f6251 | 15 | * |
716154c5 | 16 | * The SPL is distributed in the hope that it will be useful, but WITHOUT |
715f6251 | 17 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
18 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
19 | * for more details. | |
20 | * | |
21 | * You should have received a copy of the GNU General Public License along | |
716154c5 BB |
22 | * with the SPL. If not, see <http://www.gnu.org/licenses/>. |
23 | \*****************************************************************************/ | |
715f6251 | 24 | |
09b414e8 | 25 | #ifndef _SPL_KMEM_H |
26 | #define _SPL_KMEM_H | |
f1ca4da6 | 27 | |
f1b59d26 | 28 | #include <linux/module.h> |
f1ca4da6 | 29 | #include <linux/slab.h> |
79b31f36 | 30 | #include <linux/vmalloc.h> |
f1ca4da6 | 31 | #include <linux/spinlock.h> |
d6a26c6a | 32 | #include <linux/rwsem.h> |
33 | #include <linux/hash.h> | |
ed316348 | 34 | #include <linux/rbtree.h> |
d6a26c6a | 35 | #include <linux/ctype.h> |
d04c8a56 | 36 | #include <asm/atomic.h> |
57d86234 | 37 | #include <sys/types.h> |
36b313da | 38 | #include <sys/vmsystm.h> |
def465ad | 39 | #include <sys/kstat.h> |
a10287e0 | 40 | #include <sys/taskq.h> |
550f1705 | 41 | |
f1ca4da6 | 42 | /* |
43 | * Memory allocation interfaces | |
44 | */ | |
cb255ae5 BB |
45 | #define KM_SLEEP GFP_KERNEL /* Can sleep, never fails */ |
46 | #define KM_NOSLEEP GFP_ATOMIC /* Can not sleep, may fail */ | |
f90096c9 | 47 | #define KM_PUSHPAGE (GFP_NOIO | __GFP_HIGH) /* Use reserved memory */ |
cb255ae5 BB |
48 | #define KM_NODEBUG __GFP_NOWARN /* Suppress warnings */ |
49 | #define KM_FLAGS __GFP_BITS_MASK | |
50 | #define KM_VMFLAGS GFP_LEVEL_MASK | |
f1ca4da6 | 51 | |
3d061e9d | 52 | /* |
53 | * Used internally, the kernel does not need to support this flag | |
54 | */ | |
55 | #ifndef __GFP_ZERO | |
a0f6da3d | 56 | # define __GFP_ZERO 0x8000 |
3d061e9d | 57 | #endif |
58 | ||
eb0f407a BB |
59 | /* |
60 | * PF_NOFS is a per-process debug flag which is set in current->flags to | |
61 | * detect when a process is performing an unsafe allocation. All tasks | |
62 | * with PF_NOFS set must strictly use KM_PUSHPAGE for allocations because | |
63 | * if they enter direct reclaim and initiate I/O the may deadlock. | |
64 | * | |
65 | * When debugging is disabled, any incorrect usage will be detected and | |
66 | * a call stack with warning will be printed to the console. The flags | |
67 | * will then be automatically corrected to allow for safe execution. If | |
68 | * debugging is enabled this will be treated as a fatal condition. | |
69 | * | |
70 | * To avoid any risk of conflicting with the existing PF_ flags. The | |
71 | * PF_NOFS bit shadows the rarely used PF_MUTEX_TESTER bit. Only when | |
72 | * CONFIG_RT_MUTEX_TESTER is not set, and we know this bit is unused, | |
73 | * will the PF_NOFS bit be valid. Happily, most existing distributions | |
74 | * ship a kernel with CONFIG_RT_MUTEX_TESTER disabled. | |
75 | */ | |
76 | #if !defined(CONFIG_RT_MUTEX_TESTER) && defined(PF_MUTEX_TESTER) | |
77 | # define PF_NOFS PF_MUTEX_TESTER | |
78 | ||
79 | static inline void | |
80 | sanitize_flags(struct task_struct *p, gfp_t *flags) | |
81 | { | |
82 | if (unlikely((p->flags & PF_NOFS) && (*flags & (__GFP_IO|__GFP_FS)))) { | |
83 | # ifdef NDEBUG | |
84 | SDEBUG_LIMIT(SD_CONSOLE | SD_WARNING, "Fixing allocation for " | |
85 | "task %s (%d) which used GFP flags 0x%x with PF_NOFS set\n", | |
86 | p->comm, p->pid, flags); | |
87 | spl_debug_dumpstack(p); | |
88 | *flags &= ~(__GFP_IO|__GFP_FS); | |
89 | # else | |
90 | PANIC("FATAL allocation for task %s (%d) which used GFP " | |
91 | "flags 0x%x with PF_NOFS set\n", p->comm, p->pid, flags); | |
92 | # endif /* NDEBUG */ | |
93 | } | |
94 | } | |
95 | #else | |
96 | # define PF_NOFS 0x00000000 | |
97 | # define sanitize_flags(p, fl) ((void)0) | |
98 | #endif /* !defined(CONFIG_RT_MUTEX_TESTER) && defined(PF_MUTEX_TESTER) */ | |
99 | ||
c89fdee4 BB |
100 | /* |
101 | * __GFP_NOFAIL looks like it will be removed from the kernel perhaps as | |
102 | * early as 2.6.32. To avoid this issue when it occurs in upstream kernels | |
103 | * we retry the allocation here as long as it is not __GFP_WAIT (GFP_ATOMIC). | |
104 | * I would prefer the caller handle the failure case cleanly but we are | |
105 | * trying to emulate Solaris and those are not the Solaris semantics. | |
106 | */ | |
107 | static inline void * | |
108 | kmalloc_nofail(size_t size, gfp_t flags) | |
109 | { | |
110 | void *ptr; | |
111 | ||
eb0f407a BB |
112 | sanitize_flags(current, &flags); |
113 | ||
c89fdee4 BB |
114 | do { |
115 | ptr = kmalloc(size, flags); | |
116 | } while (ptr == NULL && (flags & __GFP_WAIT)); | |
117 | ||
118 | return ptr; | |
119 | } | |
120 | ||
121 | static inline void * | |
122 | kzalloc_nofail(size_t size, gfp_t flags) | |
123 | { | |
124 | void *ptr; | |
125 | ||
eb0f407a BB |
126 | sanitize_flags(current, &flags); |
127 | ||
c89fdee4 BB |
128 | do { |
129 | ptr = kzalloc(size, flags); | |
130 | } while (ptr == NULL && (flags & __GFP_WAIT)); | |
131 | ||
132 | return ptr; | |
133 | } | |
134 | ||
c89fdee4 BB |
135 | static inline void * |
136 | kmalloc_node_nofail(size_t size, gfp_t flags, int node) | |
137 | { | |
138 | void *ptr; | |
139 | ||
eb0f407a BB |
140 | sanitize_flags(current, &flags); |
141 | ||
c89fdee4 BB |
142 | do { |
143 | ptr = kmalloc_node(size, flags, node); | |
144 | } while (ptr == NULL && (flags & __GFP_WAIT)); | |
145 | ||
146 | return ptr; | |
10129680 BB |
147 | } |
148 | ||
149 | static inline void * | |
150 | vmalloc_nofail(size_t size, gfp_t flags) | |
151 | { | |
152 | void *ptr; | |
153 | ||
eb0f407a BB |
154 | sanitize_flags(current, &flags); |
155 | ||
10129680 BB |
156 | /* |
157 | * Retry failed __vmalloc() allocations once every second. The | |
158 | * rational for the delay is that the likely failure modes are: | |
159 | * | |
160 | * 1) The system has completely exhausted memory, in which case | |
161 | * delaying 1 second for the memory reclaim to run is reasonable | |
162 | * to avoid thrashing the system. | |
163 | * 2) The system has memory but has exhausted the small virtual | |
164 | * address space available on 32-bit systems. Retrying the | |
165 | * allocation immediately will only result in spinning on the | |
166 | * virtual address space lock. It is better delay a second and | |
167 | * hope that another process will free some of the address space. | |
168 | * But the bottom line is there is not much we can actually do | |
169 | * since we can never safely return a failure and honor the | |
170 | * Solaris semantics. | |
171 | */ | |
172 | while (1) { | |
173 | ptr = __vmalloc(size, flags | __GFP_HIGHMEM, PAGE_KERNEL); | |
174 | if (unlikely((ptr == NULL) && (flags & __GFP_WAIT))) { | |
175 | set_current_state(TASK_INTERRUPTIBLE); | |
176 | schedule_timeout(HZ); | |
177 | } else { | |
178 | break; | |
179 | } | |
180 | } | |
181 | ||
182 | return ptr; | |
183 | } | |
184 | ||
185 | static inline void * | |
186 | vzalloc_nofail(size_t size, gfp_t flags) | |
187 | { | |
188 | void *ptr; | |
189 | ||
190 | ptr = vmalloc_nofail(size, flags); | |
191 | if (ptr) | |
192 | memset(ptr, 0, (size)); | |
193 | ||
194 | return ptr; | |
195 | } | |
c89fdee4 | 196 | |
f1ca4da6 | 197 | #ifdef DEBUG_KMEM |
a0f6da3d | 198 | |
10129680 BB |
199 | /* |
200 | * Memory accounting functions to be used only when DEBUG_KMEM is set. | |
201 | */ | |
202 | # ifdef HAVE_ATOMIC64_T | |
a0f6da3d | 203 | |
d04c8a56 BB |
204 | # define kmem_alloc_used_add(size) atomic64_add(size, &kmem_alloc_used) |
205 | # define kmem_alloc_used_sub(size) atomic64_sub(size, &kmem_alloc_used) | |
206 | # define kmem_alloc_used_read() atomic64_read(&kmem_alloc_used) | |
207 | # define kmem_alloc_used_set(size) atomic64_set(&kmem_alloc_used, size) | |
208 | # define vmem_alloc_used_add(size) atomic64_add(size, &vmem_alloc_used) | |
209 | # define vmem_alloc_used_sub(size) atomic64_sub(size, &vmem_alloc_used) | |
210 | # define vmem_alloc_used_read() atomic64_read(&vmem_alloc_used) | |
211 | # define vmem_alloc_used_set(size) atomic64_set(&vmem_alloc_used, size) | |
212 | ||
10129680 | 213 | extern atomic64_t kmem_alloc_used; |
d04c8a56 | 214 | extern unsigned long long kmem_alloc_max; |
10129680 | 215 | extern atomic64_t vmem_alloc_used; |
d04c8a56 BB |
216 | extern unsigned long long vmem_alloc_max; |
217 | ||
10129680 BB |
218 | # else /* HAVE_ATOMIC64_T */ |
219 | ||
d04c8a56 BB |
220 | # define kmem_alloc_used_add(size) atomic_add(size, &kmem_alloc_used) |
221 | # define kmem_alloc_used_sub(size) atomic_sub(size, &kmem_alloc_used) | |
222 | # define kmem_alloc_used_read() atomic_read(&kmem_alloc_used) | |
223 | # define kmem_alloc_used_set(size) atomic_set(&kmem_alloc_used, size) | |
224 | # define vmem_alloc_used_add(size) atomic_add(size, &vmem_alloc_used) | |
225 | # define vmem_alloc_used_sub(size) atomic_sub(size, &vmem_alloc_used) | |
226 | # define vmem_alloc_used_read() atomic_read(&vmem_alloc_used) | |
227 | # define vmem_alloc_used_set(size) atomic_set(&vmem_alloc_used, size) | |
228 | ||
10129680 BB |
229 | extern atomic_t kmem_alloc_used; |
230 | extern unsigned long long kmem_alloc_max; | |
231 | extern atomic_t vmem_alloc_used; | |
232 | extern unsigned long long vmem_alloc_max; | |
a0f6da3d | 233 | |
10129680 | 234 | # endif /* HAVE_ATOMIC64_T */ |
a0f6da3d | 235 | |
236 | # ifdef DEBUG_KMEM_TRACKING | |
10129680 BB |
237 | /* |
238 | * DEBUG_KMEM && DEBUG_KMEM_TRACKING | |
239 | * | |
240 | * The maximum level of memory debugging. All memory will be accounted | |
241 | * for and each allocation will be explicitly tracked. Any allocation | |
242 | * which is leaked will be reported on module unload and the exact location | |
243 | * where that memory was allocation will be reported. This level of memory | |
244 | * tracking will have a significant impact on performance and should only | |
245 | * be enabled for debugging. This feature may be enabled by passing | |
246 | * --enable-debug-kmem-tracking to configure. | |
247 | */ | |
248 | # define kmem_alloc(sz, fl) kmem_alloc_track((sz), (fl), \ | |
249 | __FUNCTION__, __LINE__, 0, 0) | |
250 | # define kmem_zalloc(sz, fl) kmem_alloc_track((sz), (fl)|__GFP_ZERO,\ | |
251 | __FUNCTION__, __LINE__, 0, 0) | |
252 | # define kmem_alloc_node(sz, fl, nd) kmem_alloc_track((sz), (fl), \ | |
253 | __FUNCTION__, __LINE__, 1, nd) | |
254 | # define kmem_free(ptr, sz) kmem_free_track((ptr), (sz)) | |
255 | ||
256 | # define vmem_alloc(sz, fl) vmem_alloc_track((sz), (fl), \ | |
257 | __FUNCTION__, __LINE__) | |
258 | # define vmem_zalloc(sz, fl) vmem_alloc_track((sz), (fl)|__GFP_ZERO,\ | |
259 | __FUNCTION__, __LINE__) | |
260 | # define vmem_free(ptr, sz) vmem_free_track((ptr), (sz)) | |
261 | ||
262 | extern void *kmem_alloc_track(size_t, int, const char *, int, int, int); | |
973e8269 | 263 | extern void kmem_free_track(const void *, size_t); |
10129680 | 264 | extern void *vmem_alloc_track(size_t, int, const char *, int); |
973e8269 | 265 | extern void vmem_free_track(const void *, size_t); |
a0f6da3d | 266 | |
267 | # else /* DEBUG_KMEM_TRACKING */ | |
10129680 BB |
268 | /* |
269 | * DEBUG_KMEM && !DEBUG_KMEM_TRACKING | |
270 | * | |
271 | * The default build will set DEBUG_KEM. This provides basic memory | |
272 | * accounting with little to no impact on performance. When the module | |
273 | * is unloaded in any memory was leaked the total number of leaked bytes | |
274 | * will be reported on the console. To disable this basic accounting | |
275 | * pass the --disable-debug-kmem option to configure. | |
276 | */ | |
277 | # define kmem_alloc(sz, fl) kmem_alloc_debug((sz), (fl), \ | |
278 | __FUNCTION__, __LINE__, 0, 0) | |
279 | # define kmem_zalloc(sz, fl) kmem_alloc_debug((sz), (fl)|__GFP_ZERO,\ | |
280 | __FUNCTION__, __LINE__, 0, 0) | |
281 | # define kmem_alloc_node(sz, fl, nd) kmem_alloc_debug((sz), (fl), \ | |
282 | __FUNCTION__, __LINE__, 1, nd) | |
283 | # define kmem_free(ptr, sz) kmem_free_debug((ptr), (sz)) | |
284 | ||
285 | # define vmem_alloc(sz, fl) vmem_alloc_debug((sz), (fl), \ | |
286 | __FUNCTION__, __LINE__) | |
287 | # define vmem_zalloc(sz, fl) vmem_alloc_debug((sz), (fl)|__GFP_ZERO,\ | |
288 | __FUNCTION__, __LINE__) | |
289 | # define vmem_free(ptr, sz) vmem_free_debug((ptr), (sz)) | |
290 | ||
291 | extern void *kmem_alloc_debug(size_t, int, const char *, int, int, int); | |
973e8269 | 292 | extern void kmem_free_debug(const void *, size_t); |
10129680 | 293 | extern void *vmem_alloc_debug(size_t, int, const char *, int); |
973e8269 | 294 | extern void vmem_free_debug(const void *, size_t); |
a0f6da3d | 295 | |
296 | # endif /* DEBUG_KMEM_TRACKING */ | |
c6dc93d6 | 297 | #else /* DEBUG_KMEM */ |
10129680 BB |
298 | /* |
299 | * !DEBUG_KMEM && !DEBUG_KMEM_TRACKING | |
300 | * | |
301 | * All debugging is disabled. There will be no overhead even for | |
302 | * minimal memory accounting. To enable basic accounting pass the | |
303 | * --enable-debug-kmem option to configure. | |
304 | */ | |
305 | # define kmem_alloc(sz, fl) kmalloc_nofail((sz), (fl)) | |
306 | # define kmem_zalloc(sz, fl) kzalloc_nofail((sz), (fl)) | |
307 | # define kmem_alloc_node(sz, fl, nd) kmalloc_node_nofail((sz), (fl), (nd)) | |
308 | # define kmem_free(ptr, sz) ((void)(sz), kfree(ptr)) | |
f1ca4da6 | 309 | |
10129680 BB |
310 | # define vmem_alloc(sz, fl) vmalloc_nofail((sz), (fl)) |
311 | # define vmem_zalloc(sz, fl) vzalloc_nofail((sz), (fl)) | |
312 | # define vmem_free(ptr, sz) ((void)(sz), vfree(ptr)) | |
79b31f36 | 313 | |
f1ca4da6 | 314 | #endif /* DEBUG_KMEM */ |
315 | ||
10129680 BB |
316 | extern int kmem_debugging(void); |
317 | extern char *kmem_vasprintf(const char *fmt, va_list ap); | |
318 | extern char *kmem_asprintf(const char *fmt, ...); | |
319 | extern char *strdup(const char *str); | |
320 | extern void strfree(char *str); | |
321 | ||
322 | ||
f1ca4da6 | 323 | /* |
10129680 BB |
324 | * Slab allocation interfaces. The SPL slab differs from the standard |
325 | * Linux SLAB or SLUB primarily in that each cache may be backed by slabs | |
326 | * allocated from the physical or virtal memory address space. The virtual | |
327 | * slabs allow for good behavior when allocation large objects of identical | |
328 | * size. This slab implementation also supports both constructors and | |
329 | * destructions which the Linux slab does not. | |
f1ca4da6 | 330 | */ |
ea3e6ca9 BB |
331 | enum { |
332 | KMC_BIT_NOTOUCH = 0, /* Don't update ages */ | |
333 | KMC_BIT_NODEBUG = 1, /* Default behavior */ | |
334 | KMC_BIT_NOMAGAZINE = 2, /* XXX: Unsupported */ | |
335 | KMC_BIT_NOHASH = 3, /* XXX: Unsupported */ | |
336 | KMC_BIT_QCACHE = 4, /* XXX: Unsupported */ | |
337 | KMC_BIT_KMEM = 5, /* Use kmem cache */ | |
338 | KMC_BIT_VMEM = 6, /* Use vmem cache */ | |
a073aeb0 BB |
339 | KMC_BIT_SLAB = 7, /* Use Linux slab cache */ |
340 | KMC_BIT_OFFSLAB = 8, /* Objects not on slab */ | |
341 | KMC_BIT_NOEMERGENCY = 9, /* Disable emergency objects */ | |
165f13c3 | 342 | KMC_BIT_DEADLOCKED = 14, /* Deadlock detected */ |
e2dcc6e2 | 343 | KMC_BIT_GROWING = 15, /* Growing in progress */ |
ea3e6ca9 BB |
344 | KMC_BIT_REAPING = 16, /* Reaping in progress */ |
345 | KMC_BIT_DESTROY = 17, /* Destroy in progress */ | |
3336e29c BB |
346 | KMC_BIT_TOTAL = 18, /* Proc handler helper bit */ |
347 | KMC_BIT_ALLOC = 19, /* Proc handler helper bit */ | |
348 | KMC_BIT_MAX = 20, /* Proc handler helper bit */ | |
ea3e6ca9 BB |
349 | }; |
350 | ||
2b354302 BB |
351 | /* kmem move callback return values */ |
352 | typedef enum kmem_cbrc { | |
353 | KMEM_CBRC_YES = 0, /* Object moved */ | |
354 | KMEM_CBRC_NO = 1, /* Object not moved */ | |
355 | KMEM_CBRC_LATER = 2, /* Object not moved, try again later */ | |
356 | KMEM_CBRC_DONT_NEED = 3, /* Neither object is needed */ | |
357 | KMEM_CBRC_DONT_KNOW = 4, /* Object unknown */ | |
358 | } kmem_cbrc_t; | |
359 | ||
ea3e6ca9 BB |
360 | #define KMC_NOTOUCH (1 << KMC_BIT_NOTOUCH) |
361 | #define KMC_NODEBUG (1 << KMC_BIT_NODEBUG) | |
362 | #define KMC_NOMAGAZINE (1 << KMC_BIT_NOMAGAZINE) | |
363 | #define KMC_NOHASH (1 << KMC_BIT_NOHASH) | |
364 | #define KMC_QCACHE (1 << KMC_BIT_QCACHE) | |
365 | #define KMC_KMEM (1 << KMC_BIT_KMEM) | |
366 | #define KMC_VMEM (1 << KMC_BIT_VMEM) | |
a073aeb0 | 367 | #define KMC_SLAB (1 << KMC_BIT_SLAB) |
ea3e6ca9 | 368 | #define KMC_OFFSLAB (1 << KMC_BIT_OFFSLAB) |
cb5c2ace | 369 | #define KMC_NOEMERGENCY (1 << KMC_BIT_NOEMERGENCY) |
165f13c3 | 370 | #define KMC_DEADLOCKED (1 << KMC_BIT_DEADLOCKED) |
e2dcc6e2 | 371 | #define KMC_GROWING (1 << KMC_BIT_GROWING) |
ea3e6ca9 BB |
372 | #define KMC_REAPING (1 << KMC_BIT_REAPING) |
373 | #define KMC_DESTROY (1 << KMC_BIT_DESTROY) | |
3336e29c BB |
374 | #define KMC_TOTAL (1 << KMC_BIT_TOTAL) |
375 | #define KMC_ALLOC (1 << KMC_BIT_ALLOC) | |
376 | #define KMC_MAX (1 << KMC_BIT_MAX) | |
ea3e6ca9 | 377 | |
0936c344 BB |
378 | #define KMC_REAP_CHUNK INT_MAX |
379 | #define KMC_DEFAULT_SEEKS 1 | |
f1ca4da6 | 380 | |
0936c344 BB |
381 | #define KMC_EXPIRE_AGE 0x1 /* Due to age */ |
382 | #define KMC_EXPIRE_MEM 0x2 /* Due to low memory */ | |
383 | ||
376dc35e BB |
384 | #define KMC_RECLAIM_ONCE 0x1 /* Force a single shrinker pass */ |
385 | ||
0936c344 | 386 | extern unsigned int spl_kmem_cache_expire; |
ff449ac4 | 387 | extern struct list_head spl_kmem_cache_list; |
388 | extern struct rw_semaphore spl_kmem_cache_sem; | |
2fb9b26a | 389 | |
4afaaefa | 390 | #define SKM_MAGIC 0x2e2e2e2e |
2fb9b26a | 391 | #define SKO_MAGIC 0x20202020 |
392 | #define SKS_MAGIC 0x22222222 | |
393 | #define SKC_MAGIC 0x2c2c2c2c | |
394 | ||
37db7d8c BB |
395 | #define SPL_KMEM_CACHE_DELAY 15 /* Minimum slab release age */ |
396 | #define SPL_KMEM_CACHE_REAP 0 /* Default reap everything */ | |
83150861 | 397 | #define SPL_KMEM_CACHE_OBJ_PER_SLAB 16 /* Target objects per slab */ |
917fef27 | 398 | #define SPL_KMEM_CACHE_OBJ_PER_SLAB_MIN 1 /* Minimum objects per slab */ |
ea3e6ca9 | 399 | #define SPL_KMEM_CACHE_ALIGN 8 /* Default object alignment */ |
2fb9b26a | 400 | |
2b354302 BB |
401 | #define POINTER_IS_VALID(p) 0 /* Unimplemented */ |
402 | #define POINTER_INVALIDATE(pp) /* Unimplemented */ | |
403 | ||
2fb9b26a | 404 | typedef int (*spl_kmem_ctor_t)(void *, void *, int); |
405 | typedef void (*spl_kmem_dtor_t)(void *, void *); | |
406 | typedef void (*spl_kmem_reclaim_t)(void *); | |
407 | ||
4afaaefa | 408 | typedef struct spl_kmem_magazine { |
9b1b8e4c | 409 | uint32_t skm_magic; /* Sanity magic */ |
4afaaefa | 410 | uint32_t skm_avail; /* Available objects */ |
411 | uint32_t skm_size; /* Magazine size */ | |
412 | uint32_t skm_refill; /* Batch refill size */ | |
9b1b8e4c | 413 | struct spl_kmem_cache *skm_cache; /* Owned by cache */ |
4afaaefa | 414 | unsigned long skm_age; /* Last cache access */ |
08850edd | 415 | unsigned int skm_cpu; /* Owned by cpu */ |
4afaaefa | 416 | void *skm_objs[0]; /* Object pointers */ |
417 | } spl_kmem_magazine_t; | |
418 | ||
2fb9b26a | 419 | typedef struct spl_kmem_obj { |
420 | uint32_t sko_magic; /* Sanity magic */ | |
2fb9b26a | 421 | void *sko_addr; /* Buffer address */ |
422 | struct spl_kmem_slab *sko_slab; /* Owned by slab */ | |
423 | struct list_head sko_list; /* Free object list linkage */ | |
2fb9b26a | 424 | } spl_kmem_obj_t; |
425 | ||
426 | typedef struct spl_kmem_slab { | |
427 | uint32_t sks_magic; /* Sanity magic */ | |
428 | uint32_t sks_objs; /* Objects per slab */ | |
429 | struct spl_kmem_cache *sks_cache; /* Owned by cache */ | |
430 | struct list_head sks_list; /* Slab list linkage */ | |
431 | struct list_head sks_free_list; /* Free object list */ | |
432 | unsigned long sks_age; /* Last modify jiffie */ | |
4afaaefa | 433 | uint32_t sks_ref; /* Ref count used objects */ |
2fb9b26a | 434 | } spl_kmem_slab_t; |
435 | ||
e2dcc6e2 BB |
436 | typedef struct spl_kmem_alloc { |
437 | struct spl_kmem_cache *ska_cache; /* Owned by cache */ | |
438 | int ska_flags; /* Allocation flags */ | |
33e94ef1 | 439 | taskq_ent_t ska_tqe; /* Task queue entry */ |
e2dcc6e2 BB |
440 | } spl_kmem_alloc_t; |
441 | ||
442 | typedef struct spl_kmem_emergency { | |
ed316348 | 443 | struct rb_node ske_node; /* Emergency tree linkage */ |
e2dcc6e2 | 444 | void *ske_obj; /* Buffer address */ |
e2dcc6e2 BB |
445 | } spl_kmem_emergency_t; |
446 | ||
2fb9b26a | 447 | typedef struct spl_kmem_cache { |
ea3e6ca9 BB |
448 | uint32_t skc_magic; /* Sanity magic */ |
449 | uint32_t skc_name_size; /* Name length */ | |
450 | char *skc_name; /* Name string */ | |
4afaaefa | 451 | spl_kmem_magazine_t *skc_mag[NR_CPUS]; /* Per-CPU warm cache */ |
452 | uint32_t skc_mag_size; /* Magazine size */ | |
453 | uint32_t skc_mag_refill; /* Magazine refill count */ | |
ea3e6ca9 BB |
454 | spl_kmem_ctor_t skc_ctor; /* Constructor */ |
455 | spl_kmem_dtor_t skc_dtor; /* Destructor */ | |
456 | spl_kmem_reclaim_t skc_reclaim; /* Reclaimator */ | |
457 | void *skc_private; /* Private data */ | |
458 | void *skc_vmp; /* Unused */ | |
a073aeb0 | 459 | struct kmem_cache *skc_linux_cache; /* Linux slab cache if used */ |
31a033ec | 460 | unsigned long skc_flags; /* Flags */ |
2fb9b26a | 461 | uint32_t skc_obj_size; /* Object size */ |
48e0606a | 462 | uint32_t skc_obj_align; /* Object alignment */ |
a1502d76 | 463 | uint32_t skc_slab_objs; /* Objects per slab */ |
ea3e6ca9 BB |
464 | uint32_t skc_slab_size; /* Slab size */ |
465 | uint32_t skc_delay; /* Slab reclaim interval */ | |
37db7d8c | 466 | uint32_t skc_reap; /* Slab reclaim count */ |
ea3e6ca9 | 467 | atomic_t skc_ref; /* Ref count callers */ |
a10287e0 | 468 | taskqid_t skc_taskqid; /* Slab reclaim task */ |
ea3e6ca9 | 469 | struct list_head skc_list; /* List of caches linkage */ |
2fb9b26a | 470 | struct list_head skc_complete_list;/* Completely alloc'ed */ |
471 | struct list_head skc_partial_list; /* Partially alloc'ed */ | |
ed316348 | 472 | struct rb_root skc_emergency_tree; /* Min sized objects */ |
d46630e0 | 473 | spinlock_t skc_lock; /* Cache lock */ |
e2dcc6e2 | 474 | wait_queue_head_t skc_waitq; /* Allocation waiters */ |
2fb9b26a | 475 | uint64_t skc_slab_fail; /* Slab alloc failures */ |
476 | uint64_t skc_slab_create;/* Slab creates */ | |
477 | uint64_t skc_slab_destroy;/* Slab destroys */ | |
d46630e0 | 478 | uint64_t skc_slab_total; /* Slab total current */ |
ea3e6ca9 | 479 | uint64_t skc_slab_alloc; /* Slab alloc current */ |
d46630e0 | 480 | uint64_t skc_slab_max; /* Slab max historic */ |
481 | uint64_t skc_obj_total; /* Obj total current */ | |
482 | uint64_t skc_obj_alloc; /* Obj alloc current */ | |
483 | uint64_t skc_obj_max; /* Obj max historic */ | |
165f13c3 | 484 | uint64_t skc_obj_deadlock; /* Obj emergency deadlocks */ |
e2dcc6e2 BB |
485 | uint64_t skc_obj_emergency; /* Obj emergency current */ |
486 | uint64_t skc_obj_emergency_max; /* Obj emergency max */ | |
2fb9b26a | 487 | } spl_kmem_cache_t; |
7afde631 | 488 | #define kmem_cache_t spl_kmem_cache_t |
2fb9b26a | 489 | |
2b354302 BB |
490 | extern spl_kmem_cache_t *spl_kmem_cache_create(char *name, size_t size, |
491 | size_t align, spl_kmem_ctor_t ctor, spl_kmem_dtor_t dtor, | |
492 | spl_kmem_reclaim_t reclaim, void *priv, void *vmp, int flags); | |
6576a1a7 | 493 | extern void spl_kmem_cache_set_move(spl_kmem_cache_t *, |
2b354302 | 494 | kmem_cbrc_t (*)(void *, void *, size_t, void *)); |
2fb9b26a | 495 | extern void spl_kmem_cache_destroy(spl_kmem_cache_t *skc); |
496 | extern void *spl_kmem_cache_alloc(spl_kmem_cache_t *skc, int flags); | |
497 | extern void spl_kmem_cache_free(spl_kmem_cache_t *skc, void *obj); | |
cef7605c | 498 | extern void spl_kmem_cache_reap_now(spl_kmem_cache_t *skc, int count); |
2fb9b26a | 499 | extern void spl_kmem_reap(void); |
f1ca4da6 | 500 | |
2fb9b26a | 501 | int spl_kmem_init(void); |
502 | void spl_kmem_fini(void); | |
5d86345d | 503 | |
f1ca4da6 | 504 | #define kmem_cache_create(name,size,align,ctor,dtor,rclm,priv,vmp,flags) \ |
2fb9b26a | 505 | spl_kmem_cache_create(name,size,align,ctor,dtor,rclm,priv,vmp,flags) |
2b354302 | 506 | #define kmem_cache_set_move(skc, move) spl_kmem_cache_set_move(skc, move) |
2fb9b26a | 507 | #define kmem_cache_destroy(skc) spl_kmem_cache_destroy(skc) |
508 | #define kmem_cache_alloc(skc, flags) spl_kmem_cache_alloc(skc, flags) | |
509 | #define kmem_cache_free(skc, obj) spl_kmem_cache_free(skc, obj) | |
cef7605c PS |
510 | #define kmem_cache_reap_now(skc) \ |
511 | spl_kmem_cache_reap_now(skc, skc->skc_reap) | |
2fb9b26a | 512 | #define kmem_reap() spl_kmem_reap() |
a1502d76 | 513 | #define kmem_virt(ptr) (((ptr) >= (void *)VMALLOC_START) && \ |
514 | ((ptr) < (void *)VMALLOC_END)) | |
f1ca4da6 | 515 | |
a073aeb0 BB |
516 | /* |
517 | * Allow custom slab allocation flags to be set for KMC_SLAB based caches. | |
518 | * One use for this function is to ensure the __GFP_COMP flag is part of | |
519 | * the default allocation mask which ensures higher order allocations are | |
520 | * properly refcounted. This flag was added to the default ->allocflags | |
521 | * as of Linux 3.11. | |
522 | */ | |
523 | static inline void | |
524 | kmem_cache_set_allocflags(spl_kmem_cache_t *skc, gfp_t flags) | |
525 | { | |
526 | if (skc->skc_linux_cache == NULL) | |
527 | return; | |
528 | ||
529 | #if defined(HAVE_KMEM_CACHE_ALLOCFLAGS) | |
530 | skc->skc_linux_cache->allocflags |= flags; | |
531 | #elif defined(HAVE_KMEM_CACHE_GFPFLAGS) | |
532 | skc->skc_linux_cache->gfpflags |= flags; | |
533 | #endif | |
534 | } | |
535 | ||
09b414e8 | 536 | #endif /* _SPL_KMEM_H */ |