4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * libcfs/include/libcfs/libcfs_private.h
38 * Various defines for libcfs.
42 #ifndef __LIBCFS_PRIVATE_H__
43 #define __LIBCFS_PRIVATE_H__
45 /* XXX this layering violation is for nidstrings */
46 #include <linux/lnet/types.h>
48 #ifndef DEBUG_SUBSYSTEM
49 # define DEBUG_SUBSYSTEM S_UNDEFINED
55 * When this is on, LASSERT macro includes check for assignment used instead
56 * of equality check, but doesn't have unlikely(). Turn this on from time to
57 * time to make test-builds. This shouldn't be on for production release.
59 #define LASSERT_CHECKED (0)
62 #define LASSERTF(cond, fmt, ...) \
64 if (unlikely(!(cond))) { \
65 LIBCFS_DEBUG_MSG_DATA_DECL(__msg_data, D_EMERG, NULL); \
66 libcfs_debug_msg(&__msg_data, \
67 "ASSERTION( %s ) failed: " fmt, #cond, \
69 lbug_with_loc(&__msg_data); \
73 #define LASSERT(cond) LASSERTF(cond, "\n")
75 # define LINVRNT(exp) ((void)sizeof!!(exp))
77 #define KLASSERT(e) LASSERT(e)
79 void lbug_with_loc(struct libcfs_debug_msg_data *) __attribute__((noreturn));
83 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_EMERG, NULL); \
84 lbug_with_loc(&msgdata); \
87 extern atomic_t libcfs_kmemory;
92 # define libcfs_kmem_inc(ptr, size) \
94 atomic_add(size, &libcfs_kmemory); \
97 # define libcfs_kmem_dec(ptr, size) \
99 atomic_sub(size, &libcfs_kmemory); \
102 # define libcfs_kmem_read() \
103 atomic_read(&libcfs_kmemory)
106 #ifndef LIBCFS_VMALLOC_SIZE
107 #define LIBCFS_VMALLOC_SIZE (2 << PAGE_CACHE_SHIFT) /* 2 pages */
110 #define LIBCFS_ALLOC_PRE(size, mask) \
112 LASSERT(!in_interrupt() || \
113 ((size) <= LIBCFS_VMALLOC_SIZE && \
114 ((mask) & GFP_ATOMIC)) != 0); \
117 #define LIBCFS_ALLOC_POST(ptr, size) \
119 if (unlikely((ptr) == NULL)) { \
120 CERROR("LNET: out of memory at %s:%d (tried to alloc '" \
121 #ptr "' = %d)\n", __FILE__, __LINE__, (int)(size)); \
122 CERROR("LNET: %d total bytes allocated by lnet\n", \
123 libcfs_kmem_read()); \
125 memset((ptr), 0, (size)); \
126 libcfs_kmem_inc((ptr), (size)); \
127 CDEBUG(D_MALLOC, "alloc '" #ptr "': %d at %p (tot %d).\n", \
128 (int)(size), (ptr), libcfs_kmem_read()); \
133 * allocate memory with GFP flags @mask
135 #define LIBCFS_ALLOC_GFP(ptr, size, mask) \
137 LIBCFS_ALLOC_PRE((size), (mask)); \
138 (ptr) = (size) <= LIBCFS_VMALLOC_SIZE ? \
139 kmalloc((size), (mask)) : vmalloc(size); \
140 LIBCFS_ALLOC_POST((ptr), (size)); \
146 #define LIBCFS_ALLOC(ptr, size) \
147 LIBCFS_ALLOC_GFP(ptr, size, __GFP_IO)
150 * non-sleeping allocator
152 #define LIBCFS_ALLOC_ATOMIC(ptr, size) \
153 LIBCFS_ALLOC_GFP(ptr, size, GFP_ATOMIC)
156 * allocate memory for specified CPU partition
157 * \a cptab != NULL, \a cpt is CPU partition id of \a cptab
158 * \a cptab == NULL, \a cpt is HW NUMA node id
160 #define LIBCFS_CPT_ALLOC_GFP(ptr, cptab, cpt, size, mask) \
162 LIBCFS_ALLOC_PRE((size), (mask)); \
163 (ptr) = (size) <= LIBCFS_VMALLOC_SIZE ? \
164 cfs_cpt_malloc((cptab), (cpt), (size), (mask)) : \
165 cfs_cpt_vmalloc((cptab), (cpt), (size)); \
166 LIBCFS_ALLOC_POST((ptr), (size)); \
169 /** default numa allocator */
170 #define LIBCFS_CPT_ALLOC(ptr, cptab, cpt, size) \
171 LIBCFS_CPT_ALLOC_GFP(ptr, cptab, cpt, size, __GFP_IO)
173 #define LIBCFS_FREE(ptr, size) \
176 if (unlikely((ptr) == NULL)) { \
177 CERROR("LIBCFS: free NULL '" #ptr "' (%d bytes) at " \
178 "%s:%d\n", s, __FILE__, __LINE__); \
181 libcfs_kmem_dec((ptr), s); \
182 CDEBUG(D_MALLOC, "kfreed '" #ptr "': %d at %p (tot %d).\n", \
183 s, (ptr), libcfs_kmem_read()); \
184 if (unlikely(s > LIBCFS_VMALLOC_SIZE)) \
190 /******************************************************************************/
192 /* htonl hack - either this, or compile with -O2. Stupid byteorder/generic.h */
193 #if defined(__GNUC__) && (__GNUC__ >= 2) && !defined(__OPTIMIZE__)
194 #define ___htonl(x) __cpu_to_be32(x)
195 #define ___htons(x) __cpu_to_be16(x)
196 #define ___ntohl(x) __be32_to_cpu(x)
197 #define ___ntohs(x) __be16_to_cpu(x)
198 #define htonl(x) ___htonl(x)
199 #define ntohl(x) ___ntohl(x)
200 #define htons(x) ___htons(x)
201 #define ntohs(x) ___ntohs(x)
204 void libcfs_debug_dumpstack(task_t *tsk);
205 void libcfs_run_upcall(char **argv);
206 void libcfs_run_lbug_upcall(struct libcfs_debug_msg_data *);
207 void libcfs_debug_dumplog(void);
208 int libcfs_debug_init(unsigned long bufsize);
209 int libcfs_debug_cleanup(void);
210 int libcfs_debug_clear_buffer(void);
211 int libcfs_debug_mark_buffer(const char *text);
213 void libcfs_debug_set_level(unsigned int debug_level);
217 * allocate per-cpu-partition data, returned value is an array of pointers,
218 * variable can be indexed by CPU ID.
219 * cptable != NULL: size of array is number of CPU partitions
220 * cptable == NULL: size of array is number of HW cores
222 void *cfs_percpt_alloc(struct cfs_cpt_table *cptab, unsigned int size);
224 * destory per-cpu-partition variable
226 void cfs_percpt_free(void *vars);
227 int cfs_percpt_number(void *vars);
228 void *cfs_percpt_current(void *vars);
229 void *cfs_percpt_index(void *vars, int idx);
231 #define cfs_percpt_for_each(var, i, vars) \
232 for (i = 0; i < cfs_percpt_number(vars) && \
233 ((var) = (vars)[i]) != NULL; i++)
236 * allocate a variable array, returned value is an array of pointers.
237 * Caller can specify length of array by count.
239 void *cfs_array_alloc(int count, unsigned int size);
240 void cfs_array_free(void *vars);
242 #define LASSERT_ATOMIC_ENABLED (1)
244 #if LASSERT_ATOMIC_ENABLED
246 /** assert value of @a is equal to @v */
247 #define LASSERT_ATOMIC_EQ(a, v) \
249 LASSERTF(atomic_read(a) == v, \
250 "value: %d\n", atomic_read((a))); \
253 /** assert value of @a is unequal to @v */
254 #define LASSERT_ATOMIC_NE(a, v) \
256 LASSERTF(atomic_read(a) != v, \
257 "value: %d\n", atomic_read((a))); \
260 /** assert value of @a is little than @v */
261 #define LASSERT_ATOMIC_LT(a, v) \
263 LASSERTF(atomic_read(a) < v, \
264 "value: %d\n", atomic_read((a))); \
267 /** assert value of @a is little/equal to @v */
268 #define LASSERT_ATOMIC_LE(a, v) \
270 LASSERTF(atomic_read(a) <= v, \
271 "value: %d\n", atomic_read((a))); \
274 /** assert value of @a is great than @v */
275 #define LASSERT_ATOMIC_GT(a, v) \
277 LASSERTF(atomic_read(a) > v, \
278 "value: %d\n", atomic_read((a))); \
281 /** assert value of @a is great/equal to @v */
282 #define LASSERT_ATOMIC_GE(a, v) \
284 LASSERTF(atomic_read(a) >= v, \
285 "value: %d\n", atomic_read((a))); \
288 /** assert value of @a is great than @v1 and little than @v2 */
289 #define LASSERT_ATOMIC_GT_LT(a, v1, v2) \
291 int __v = atomic_read(a); \
292 LASSERTF(__v > v1 && __v < v2, "value: %d\n", __v); \
295 /** assert value of @a is great than @v1 and little/equal to @v2 */
296 #define LASSERT_ATOMIC_GT_LE(a, v1, v2) \
298 int __v = atomic_read(a); \
299 LASSERTF(__v > v1 && __v <= v2, "value: %d\n", __v); \
302 /** assert value of @a is great/equal to @v1 and little than @v2 */
303 #define LASSERT_ATOMIC_GE_LT(a, v1, v2) \
305 int __v = atomic_read(a); \
306 LASSERTF(__v >= v1 && __v < v2, "value: %d\n", __v); \
309 /** assert value of @a is great/equal to @v1 and little/equal to @v2 */
310 #define LASSERT_ATOMIC_GE_LE(a, v1, v2) \
312 int __v = atomic_read(a); \
313 LASSERTF(__v >= v1 && __v <= v2, "value: %d\n", __v); \
316 #else /* !LASSERT_ATOMIC_ENABLED */
318 #define LASSERT_ATOMIC_EQ(a, v) do {} while (0)
319 #define LASSERT_ATOMIC_NE(a, v) do {} while (0)
320 #define LASSERT_ATOMIC_LT(a, v) do {} while (0)
321 #define LASSERT_ATOMIC_LE(a, v) do {} while (0)
322 #define LASSERT_ATOMIC_GT(a, v) do {} while (0)
323 #define LASSERT_ATOMIC_GE(a, v) do {} while (0)
324 #define LASSERT_ATOMIC_GT_LT(a, v1, v2) do {} while (0)
325 #define LASSERT_ATOMIC_GT_LE(a, v1, v2) do {} while (0)
326 #define LASSERT_ATOMIC_GE_LT(a, v1, v2) do {} while (0)
327 #define LASSERT_ATOMIC_GE_LE(a, v1, v2) do {} while (0)
329 #endif /* LASSERT_ATOMIC_ENABLED */
331 #define LASSERT_ATOMIC_ZERO(a) LASSERT_ATOMIC_EQ(a, 0)
332 #define LASSERT_ATOMIC_POS(a) LASSERT_ATOMIC_GT(a, 0)
334 #define CFS_ALLOC_PTR(ptr) LIBCFS_ALLOC(ptr, sizeof (*(ptr)));
335 #define CFS_FREE_PTR(ptr) LIBCFS_FREE(ptr, sizeof (*(ptr)));
338 * percpu partition lock
340 * There are some use-cases like this in Lustre:
341 * . each CPU partition has it's own private data which is frequently changed,
342 * and mostly by the local CPU partition.
343 * . all CPU partitions share some global data, these data are rarely changed.
345 * LNet is typical example.
346 * CPU partition lock is designed for this kind of use-cases:
347 * . each CPU partition has it's own private lock
348 * . change on private data just needs to take the private lock
349 * . read on shared data just needs to take _any_ of private locks
350 * . change on shared data needs to take _all_ private locks,
351 * which is slow and should be really rare.
355 CFS_PERCPT_LOCK_EX = -1, /* negative */
359 struct cfs_percpt_lock {
360 /* cpu-partition-table for this lock */
361 struct cfs_cpt_table *pcl_cptab;
362 /* exclusively locked */
363 unsigned int pcl_locked;
364 /* private lock table */
365 spinlock_t **pcl_locks;
368 /* return number of private locks */
370 cfs_percpt_lock_num(struct cfs_percpt_lock *pcl)
372 return cfs_cpt_number(pcl->pcl_cptab);
377 * create a cpu-partition lock based on CPU partition table \a cptab,
378 * each private lock has extra \a psize bytes padding data
380 struct cfs_percpt_lock *cfs_percpt_lock_alloc(struct cfs_cpt_table *cptab);
381 /* destroy a cpu-partition lock */
382 void cfs_percpt_lock_free(struct cfs_percpt_lock *pcl);
384 /* lock private lock \a index of \a pcl */
385 void cfs_percpt_lock(struct cfs_percpt_lock *pcl, int index);
386 /* unlock private lock \a index of \a pcl */
387 void cfs_percpt_unlock(struct cfs_percpt_lock *pcl, int index);
388 /* create percpt (atomic) refcount based on @cptab */
389 atomic_t **cfs_percpt_atomic_alloc(struct cfs_cpt_table *cptab, int val);
390 /* destroy percpt refcount */
391 void cfs_percpt_atomic_free(atomic_t **refs);
392 /* return sum of all percpu refs */
393 int cfs_percpt_atomic_summary(atomic_t **refs);
396 /** Compile-time assertion.
398 * Check an invariant described by a constant expression at compile time by
399 * forcing a compiler error if it does not hold. \a cond must be a constant
400 * expression as defined by the ISO C Standard:
402 * 6.8.4.2 The switch statement
404 * [#3] The expression of each case label shall be an integer
405 * constant expression and no two of the case constant
406 * expressions in the same switch statement shall have the same
407 * value after conversion...
410 #define CLASSERT(cond) do {switch(42) {case (cond): case 0: break;}} while (0)
412 /* support decl needed both by kernel and liblustre */
413 int libcfs_isknown_lnd(int type);
414 char *libcfs_lnd2modname(int type);
415 char *libcfs_lnd2str(int type);
416 int libcfs_str2lnd(const char *str);
417 char *libcfs_net2str(__u32 net);
418 char *libcfs_nid2str(lnet_nid_t nid);
419 __u32 libcfs_str2net(const char *str);
420 lnet_nid_t libcfs_str2nid(const char *str);
421 int libcfs_str2anynid(lnet_nid_t *nid, const char *str);
422 char *libcfs_id2str(lnet_process_id_t id);
423 void cfs_free_nidlist(struct list_head *list);
424 int cfs_parse_nidlist(char *str, int len, struct list_head *list);
425 int cfs_match_nid(lnet_nid_t nid, struct list_head *list);
427 /** \addtogroup lnet_addr
429 /* how an LNET NID encodes net:address */
430 /** extract the address part of an lnet_nid_t */
431 #define LNET_NIDADDR(nid) ((__u32)((nid) & 0xffffffff))
432 /** extract the network part of an lnet_nid_t */
433 #define LNET_NIDNET(nid) ((__u32)(((nid) >> 32)) & 0xffffffff)
434 /** make an lnet_nid_t from a network part and an address part */
435 #define LNET_MKNID(net,addr) ((((__u64)(net))<<32)|((__u64)(addr)))
436 /* how net encodes type:number */
437 #define LNET_NETNUM(net) ((net) & 0xffff)
438 #define LNET_NETTYP(net) (((net) >> 16) & 0xffff)
439 #define LNET_MKNET(typ,num) ((((__u32)(typ))<<16)|((__u32)(num)))
442 /* max value for numeric network address */
443 #define MAX_NUMERIC_VALUE 0xffffffff
446 #define ergo(a, b) (!(a) || (b))
447 /* logical equivalence */
448 #define equi(a, b) (!!(a) == !!(b))
450 #ifndef CFS_CURRENT_TIME
451 # define CFS_CURRENT_TIME time(0)
454 /* --------------------------------------------------------------------
456 * Support for temporary event tracing with minimal Heisenberg effect.
457 * All stuff about lwt are put in arch/kp30.h
458 * -------------------------------------------------------------------- */
460 struct libcfs_device_userstate
462 int ldu_memhog_pages;
463 struct page *ldu_memhog_root_page;
466 /* what used to be in portals_lib.h */
468 # define MIN(a,b) (((a)<(b)) ? (a): (b))
471 # define MAX(a,b) (((a)>(b)) ? (a): (b))
474 #define MKSTR(ptr) ((ptr))? (ptr) : ""
476 static inline int cfs_size_round4 (int val)
478 return (val + 3) & (~0x3);
481 #ifndef HAVE_CFS_SIZE_ROUND
482 static inline int cfs_size_round (int val)
484 return (val + 7) & (~0x7);
486 #define HAVE_CFS_SIZE_ROUND
489 static inline int cfs_size_round16(int val)
491 return (val + 0xf) & (~0xf);
494 static inline int cfs_size_round32(int val)
496 return (val + 0x1f) & (~0x1f);
499 static inline int cfs_size_round0(int val)
503 return (val + 1 + 7) & (~0x7);
506 static inline size_t cfs_round_strlen(char *fset)
508 return (size_t)cfs_size_round((int)strlen(fset) + 1);
511 /* roundup \a val to power2 */
512 static inline unsigned int cfs_power2_roundup(unsigned int val)
514 if (val != LOWEST_BIT_SET(val)) { /* not a power of 2 already */
516 val &= ~LOWEST_BIT_SET(val);
517 } while (val != LOWEST_BIT_SET(val));
518 /* ...and round up */
524 #define LOGL(var,len,ptr) \
527 memcpy((char *)ptr, (const char *)var, len); \
528 ptr += cfs_size_round(len); \
531 #define LOGU(var,len,ptr) \
534 memcpy((char *)var, (const char *)ptr, len); \
535 ptr += cfs_size_round(len); \
538 #define LOGL0(var,len,ptr) \
542 memcpy((char *)ptr, (const char *)var, len); \
543 *((char *)(ptr) + len) = 0; \
544 ptr += cfs_size_round(len + 1); \
548 * Lustre Network Driver types.
551 /* Only add to these values (i.e. don't ever change or redefine them):
552 * network addresses depend on them... */
555 GMLND = 3, /* obsolete, keep it so that libcfs_nid2str works */