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 <lnet/types.h>
48 #ifndef DEBUG_SUBSYSTEM
49 # define DEBUG_SUBSYSTEM S_UNDEFINED
57 * When this is on, LASSERT macro includes check for assignment used instead
58 * of equality check, but doesn't have unlikely(). Turn this on from time to
59 * time to make test-builds. This shouldn't be on for production release.
61 #define LASSERT_CHECKED (0)
67 * Strange construction with empty "then" clause is used to trigger compiler
68 * warnings on the assertions of the form LASSERT(a = b);
70 * "warning: suggest parentheses around assignment used as truth value"
72 * requires -Wall. Unfortunately this rules out use of likely/unlikely.
74 #define LASSERTF(cond, fmt, ...) \
79 LIBCFS_DEBUG_MSG_DATA_DECL(__msg_data, D_EMERG, NULL); \
80 libcfs_debug_msg(&__msg_data, \
81 "ASSERTION( %s ) failed: " fmt, #cond, \
83 lbug_with_loc(&__msg_data); \
87 #define LASSERT(cond) LASSERTF(cond, "\n")
89 #else /* !LASSERT_CHECKED */
91 #define LASSERTF(cond, fmt, ...) \
93 if (unlikely(!(cond))) { \
94 LIBCFS_DEBUG_MSG_DATA_DECL(__msg_data, D_EMERG, NULL); \
95 libcfs_debug_msg(&__msg_data, \
96 "ASSERTION( %s ) failed: " fmt, #cond, \
98 lbug_with_loc(&__msg_data); \
102 #define LASSERT(cond) LASSERTF(cond, "\n")
103 #endif /* !LASSERT_CHECKED */
104 #else /* !LIBCFS_DEBUG */
105 /* sizeof is to use expression without evaluating it. */
106 # define LASSERT(e) ((void)sizeof!!(e))
107 # define LASSERTF(cond, ...) ((void)sizeof!!(cond))
108 #endif /* !LIBCFS_DEBUG */
110 #ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK
112 * This is for more expensive checks that one doesn't want to be enabled all
113 * the time. LINVRNT() has to be explicitly enabled by --enable-invariants
116 # define LINVRNT(exp) LASSERT(exp)
118 # define LINVRNT(exp) ((void)sizeof!!(exp))
121 #define KLASSERT(e) LASSERT(e)
123 void lbug_with_loc(struct libcfs_debug_msg_data *) __attribute__((noreturn));
127 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_EMERG, NULL); \
128 lbug_with_loc(&msgdata); \
131 extern atomic_t libcfs_kmemory;
137 # define libcfs_kmem_inc(ptr, size) \
139 atomic_add(size, &libcfs_kmemory); \
142 # define libcfs_kmem_dec(ptr, size) \
144 atomic_sub(size, &libcfs_kmemory); \
147 # define libcfs_kmem_read() \
148 atomic_read(&libcfs_kmemory)
151 # define libcfs_kmem_inc(ptr, size) do {} while (0)
152 # define libcfs_kmem_dec(ptr, size) do {} while (0)
153 # define libcfs_kmem_read() (0)
154 #endif /* LIBCFS_DEBUG */
156 #ifndef LIBCFS_VMALLOC_SIZE
157 #define LIBCFS_VMALLOC_SIZE (2 << PAGE_CACHE_SHIFT) /* 2 pages */
160 #define LIBCFS_ALLOC_PRE(size, mask) \
162 LASSERT(!in_interrupt() || \
163 ((size) <= LIBCFS_VMALLOC_SIZE && \
164 ((mask) & GFP_ATOMIC)) != 0); \
167 #define LIBCFS_ALLOC_POST(ptr, size) \
169 if (unlikely((ptr) == NULL)) { \
170 CERROR("LNET: out of memory at %s:%d (tried to alloc '" \
171 #ptr "' = %d)\n", __FILE__, __LINE__, (int)(size)); \
172 CERROR("LNET: %d total bytes allocated by lnet\n", \
173 libcfs_kmem_read()); \
175 libcfs_kmem_inc((ptr), (size)); \
176 CDEBUG(D_MALLOC, "alloc '" #ptr "': %d at %p (tot %d).\n", \
177 (int)(size), (ptr), libcfs_kmem_read()); \
182 * allocate memory with GFP flags @mask
183 * The allocated memory is zeroed-out.
185 #define LIBCFS_ALLOC_GFP(ptr, size, mask) \
187 LIBCFS_ALLOC_PRE((size), (mask)); \
188 (ptr) = (size) <= LIBCFS_VMALLOC_SIZE ? \
189 kzalloc((size), (mask)) : vzalloc(size); \
190 LIBCFS_ALLOC_POST((ptr), (size)); \
196 #define LIBCFS_ALLOC(ptr, size) \
197 LIBCFS_ALLOC_GFP(ptr, size, GFP_NOFS)
200 * non-sleeping allocator
202 #define LIBCFS_ALLOC_ATOMIC(ptr, size) \
203 LIBCFS_ALLOC_GFP(ptr, size, GFP_ATOMIC)
206 * allocate memory for specified CPU partition
207 * \a cptab != NULL, \a cpt is CPU partition id of \a cptab
208 * \a cptab == NULL, \a cpt is HW NUMA node id
209 * The allocated memory is zeroed-out.
211 #define LIBCFS_CPT_ALLOC_GFP(ptr, cptab, cpt, size, mask) \
213 LIBCFS_ALLOC_PRE((size), (mask)); \
214 (ptr) = (size) <= LIBCFS_VMALLOC_SIZE ? \
215 cfs_cpt_malloc((cptab), (cpt), (size), (mask) | __GFP_ZERO) : \
216 cfs_cpt_vzalloc((cptab), (cpt), (size)); \
217 LIBCFS_ALLOC_POST((ptr), (size)); \
220 /** default numa allocator */
221 #define LIBCFS_CPT_ALLOC(ptr, cptab, cpt, size) \
222 LIBCFS_CPT_ALLOC_GFP(ptr, cptab, cpt, size, GFP_NOFS)
224 #define LIBCFS_FREE(ptr, size) \
227 if (unlikely((ptr) == NULL)) { \
228 CERROR("LIBCFS: free NULL '" #ptr "' (%d bytes) at " \
229 "%s:%d\n", s, __FILE__, __LINE__); \
232 libcfs_kmem_dec((ptr), s); \
233 CDEBUG(D_MALLOC, "kfreed '" #ptr "': %d at %p (tot %d).\n", \
234 s, (ptr), libcfs_kmem_read()); \
235 if (unlikely(s > LIBCFS_VMALLOC_SIZE)) \
241 /******************************************************************************/
243 /* htonl hack - either this, or compile with -O2. Stupid byteorder/generic.h */
244 #if defined(__GNUC__) && (__GNUC__ >= 2) && !defined(__OPTIMIZE__)
245 #define ___htonl(x) __cpu_to_be32(x)
246 #define ___htons(x) __cpu_to_be16(x)
247 #define ___ntohl(x) __be32_to_cpu(x)
248 #define ___ntohs(x) __be16_to_cpu(x)
249 #define htonl(x) ___htonl(x)
250 #define ntohl(x) ___ntohl(x)
251 #define htons(x) ___htons(x)
252 #define ntohs(x) ___ntohs(x)
255 void libcfs_debug_dumpstack(struct task_struct *tsk);
256 void libcfs_run_upcall(char **argv);
257 void libcfs_run_lbug_upcall(struct libcfs_debug_msg_data *);
258 void libcfs_debug_dumplog(void);
259 int libcfs_debug_init(unsigned long bufsize);
260 int libcfs_debug_cleanup(void);
261 int libcfs_debug_clear_buffer(void);
262 int libcfs_debug_mark_buffer(const char *text);
264 void libcfs_debug_set_level(unsigned int debug_level);
266 #else /* !__KERNEL__ */
270 # define LASSERT(e) assert(e)
271 # define LASSERTF(cond, ...) \
274 CERROR(__VA_ARGS__); \
277 # define LBUG() assert(0)
278 # ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK
279 # define LINVRNT(exp) LASSERT(exp)
281 # define LINVRNT(exp) ((void)sizeof!!(exp))
284 # define LASSERT(e) ((void)sizeof!!(e))
285 # define LASSERTF(cond, ...) ((void)sizeof!!(cond))
286 # define LBUG() ((void)(0))
287 # define LINVRNT(exp) ((void)sizeof!!(exp))
288 # endif /* LIBCFS_DEBUG */
289 # define KLASSERT(e) ((void)0)
290 # define printk printf
291 #define LIBCFS_ALLOC_GFP(ptr, size, mask) \
293 (ptr) = calloc(1, size); \
295 # define LIBCFS_FREE(ptr, size) do { free(ptr); } while((size) - (size))
296 # define LIBCFS_ALLOC(ptr, size) \
297 LIBCFS_ALLOC_GFP(ptr, size, 0)
298 # define LIBCFS_CPT_ALLOC_GFP(ptr, cptab, cpt, size, mask) \
299 LIBCFS_ALLOC(ptr, size)
300 # define LIBCFS_CPT_ALLOC(ptr, cptab, cpt, size) \
301 LIBCFS_ALLOC(ptr, size)
303 void libcfs_debug_dumplog(void);
304 int libcfs_debug_init(unsigned long bufsize);
305 int libcfs_debug_cleanup(void);
307 #define libcfs_debug_dumpstack(tsk) ((void)0)
310 * Generic compiler-dependent macros required for kernel
311 * build go below this comment. Actual compiler/compiler version
312 * specific implementations come from the above header files
315 #define likely(x) __builtin_expect(!!(x), 1)
316 #define unlikely(x) __builtin_expect(!!(x), 0)
318 #define likely(x) (!!(x))
319 #define unlikely(x) (!!(x))
324 struct cfs_cpt_table;
327 * allocate per-cpu-partition data, returned value is an array of pointers,
328 * variable can be indexed by CPU ID.
329 * cptable != NULL: size of array is number of CPU partitions
330 * cptable == NULL: size of array is number of HW cores
332 void *cfs_percpt_alloc(struct cfs_cpt_table *cptab, unsigned int size);
334 * destory per-cpu-partition variable
336 void cfs_percpt_free(void *vars);
337 int cfs_percpt_number(void *vars);
338 void *cfs_percpt_current(void *vars);
339 void *cfs_percpt_index(void *vars, int idx);
341 #define cfs_percpt_for_each(var, i, vars) \
342 for (i = 0; i < cfs_percpt_number(vars) && \
343 ((var) = (vars)[i]) != NULL; i++)
346 * allocate a variable array, returned value is an array of pointers.
347 * Caller can specify length of array by count.
349 void *cfs_array_alloc(int count, unsigned int size);
350 void cfs_array_free(void *vars);
352 #define LASSERT_ATOMIC_ENABLED (1)
354 #if LASSERT_ATOMIC_ENABLED
356 /** assert value of @a is equal to @v */
357 #define LASSERT_ATOMIC_EQ(a, v) \
359 LASSERTF(atomic_read(a) == v, \
360 "value: %d\n", atomic_read((a))); \
363 /** assert value of @a is unequal to @v */
364 #define LASSERT_ATOMIC_NE(a, v) \
366 LASSERTF(atomic_read(a) != v, \
367 "value: %d\n", atomic_read((a))); \
370 /** assert value of @a is little than @v */
371 #define LASSERT_ATOMIC_LT(a, v) \
373 LASSERTF(atomic_read(a) < v, \
374 "value: %d\n", atomic_read((a))); \
377 /** assert value of @a is little/equal to @v */
378 #define LASSERT_ATOMIC_LE(a, v) \
380 LASSERTF(atomic_read(a) <= v, \
381 "value: %d\n", atomic_read((a))); \
384 /** assert value of @a is great than @v */
385 #define LASSERT_ATOMIC_GT(a, v) \
387 LASSERTF(atomic_read(a) > v, \
388 "value: %d\n", atomic_read((a))); \
391 /** assert value of @a is great/equal to @v */
392 #define LASSERT_ATOMIC_GE(a, v) \
394 LASSERTF(atomic_read(a) >= v, \
395 "value: %d\n", atomic_read((a))); \
398 /** assert value of @a is great than @v1 and little than @v2 */
399 #define LASSERT_ATOMIC_GT_LT(a, v1, v2) \
401 int __v = atomic_read(a); \
402 LASSERTF(__v > v1 && __v < v2, "value: %d\n", __v); \
405 /** assert value of @a is great than @v1 and little/equal to @v2 */
406 #define LASSERT_ATOMIC_GT_LE(a, v1, v2) \
408 int __v = atomic_read(a); \
409 LASSERTF(__v > v1 && __v <= v2, "value: %d\n", __v); \
412 /** assert value of @a is great/equal to @v1 and little than @v2 */
413 #define LASSERT_ATOMIC_GE_LT(a, v1, v2) \
415 int __v = atomic_read(a); \
416 LASSERTF(__v >= v1 && __v < v2, "value: %d\n", __v); \
419 /** assert value of @a is great/equal to @v1 and little/equal to @v2 */
420 #define LASSERT_ATOMIC_GE_LE(a, v1, v2) \
422 int __v = atomic_read(a); \
423 LASSERTF(__v >= v1 && __v <= v2, "value: %d\n", __v); \
426 #else /* !LASSERT_ATOMIC_ENABLED */
428 #define LASSERT_ATOMIC_EQ(a, v) do {} while (0)
429 #define LASSERT_ATOMIC_NE(a, v) do {} while (0)
430 #define LASSERT_ATOMIC_LT(a, v) do {} while (0)
431 #define LASSERT_ATOMIC_LE(a, v) do {} while (0)
432 #define LASSERT_ATOMIC_GT(a, v) do {} while (0)
433 #define LASSERT_ATOMIC_GE(a, v) do {} while (0)
434 #define LASSERT_ATOMIC_GT_LT(a, v1, v2) do {} while (0)
435 #define LASSERT_ATOMIC_GT_LE(a, v1, v2) do {} while (0)
436 #define LASSERT_ATOMIC_GE_LT(a, v1, v2) do {} while (0)
437 #define LASSERT_ATOMIC_GE_LE(a, v1, v2) do {} while (0)
439 #endif /* LASSERT_ATOMIC_ENABLED */
441 #define LASSERT_ATOMIC_ZERO(a) LASSERT_ATOMIC_EQ(a, 0)
442 #define LASSERT_ATOMIC_POS(a) LASSERT_ATOMIC_GT(a, 0)
444 #define CFS_ALLOC_PTR(ptr) LIBCFS_ALLOC(ptr, sizeof (*(ptr)));
445 #define CFS_FREE_PTR(ptr) LIBCFS_FREE(ptr, sizeof (*(ptr)));
448 * percpu partition lock
450 * There are some use-cases like this in Lustre:
451 * . each CPU partition has it's own private data which is frequently changed,
452 * and mostly by the local CPU partition.
453 * . all CPU partitions share some global data, these data are rarely changed.
455 * LNet is typical example.
456 * CPU partition lock is designed for this kind of use-cases:
457 * . each CPU partition has it's own private lock
458 * . change on private data just needs to take the private lock
459 * . read on shared data just needs to take _any_ of private locks
460 * . change on shared data needs to take _all_ private locks,
461 * which is slow and should be really rare.
465 CFS_PERCPT_LOCK_EX = -1, /* negative */
470 struct cfs_percpt_lock {
471 /* cpu-partition-table for this lock */
472 struct cfs_cpt_table *pcl_cptab;
473 /* exclusively locked */
474 unsigned int pcl_locked;
475 /* private lock table */
476 spinlock_t **pcl_locks;
479 /* return number of private locks */
480 #define cfs_percpt_lock_num(pcl) cfs_cpt_number(pcl->pcl_cptab)
482 #else /* !__KERNEL__ */
484 # ifdef HAVE_LIBPTHREAD
486 struct cfs_percpt_lock {
487 pthread_mutex_t pcl_mutex;
490 # else /* !HAVE_LIBPTHREAD */
492 struct cfs_percpt_lock {
496 static const struct cfs_percpt_lock CFS_PERCPT_LOCK_MAGIC;
498 # endif /* HAVE_LIBPTHREAD */
499 # define cfs_percpt_lock_num(pcl) 1
500 #endif /* __KERNEL__ */
503 * create a cpu-partition lock based on CPU partition table \a cptab,
504 * each private lock has extra \a psize bytes padding data
506 struct cfs_percpt_lock *cfs_percpt_lock_alloc(struct cfs_cpt_table *cptab);
507 /* destroy a cpu-partition lock */
508 void cfs_percpt_lock_free(struct cfs_percpt_lock *pcl);
510 /* lock private lock \a index of \a pcl */
511 void cfs_percpt_lock(struct cfs_percpt_lock *pcl, int index);
512 /* unlock private lock \a index of \a pcl */
513 void cfs_percpt_unlock(struct cfs_percpt_lock *pcl, int index);
514 /* create percpt (atomic) refcount based on @cptab */
515 atomic_t **cfs_percpt_atomic_alloc(struct cfs_cpt_table *cptab, int val);
516 /* destroy percpt refcount */
517 void cfs_percpt_atomic_free(atomic_t **refs);
518 /* return sum of all percpu refs */
519 int cfs_percpt_atomic_summary(atomic_t **refs);
522 /** Compile-time assertion.
524 * Check an invariant described by a constant expression at compile time by
525 * forcing a compiler error if it does not hold. \a cond must be a constant
526 * expression as defined by the ISO C Standard:
528 * 6.8.4.2 The switch statement
530 * [#3] The expression of each case label shall be an integer
531 * constant expression and no two of the case constant
532 * expressions in the same switch statement shall have the same
533 * value after conversion...
536 #define CLASSERT(cond) do {switch (1) {case (cond): case 0: break; } } while (0)
538 /* support decl needed both by kernel and liblustre */
539 int libcfs_isknown_lnd(int type);
540 char *libcfs_lnd2modname(int type);
541 char *libcfs_lnd2str(int type);
542 int libcfs_str2lnd(const char *str);
543 char *libcfs_net2str(__u32 net);
544 char *libcfs_nid2str(lnet_nid_t nid);
545 __u32 libcfs_str2net(const char *str);
546 lnet_nid_t libcfs_str2nid(const char *str);
547 int libcfs_str2anynid(lnet_nid_t *nid, const char *str);
548 char *libcfs_id2str(lnet_process_id_t id);
549 void cfs_free_nidlist(struct list_head *list);
550 int cfs_parse_nidlist(char *str, int len, struct list_head *list);
551 int cfs_print_nidlist(char *buffer, int count,
552 struct list_head *list);
553 int cfs_match_nid(lnet_nid_t nid, struct list_head *list);
554 bool cfs_nidrange_is_contiguous(struct list_head *nidlist);
555 void cfs_nidrange_find_min_max(struct list_head *nidlist,
556 char *min_nid, char *max_nid,
559 /** \addtogroup lnet_addr
561 /* how an LNET NID encodes net:address */
562 /** extract the address part of an lnet_nid_t */
563 #define LNET_NIDADDR(nid) ((__u32)((nid) & 0xffffffff))
564 /** extract the network part of an lnet_nid_t */
565 #define LNET_NIDNET(nid) ((__u32)(((nid) >> 32)) & 0xffffffff)
566 /** make an lnet_nid_t from a network part and an address part */
567 #define LNET_MKNID(net,addr) ((((__u64)(net))<<32)|((__u64)(addr)))
568 /* how net encodes type:number */
569 #define LNET_NETNUM(net) ((net) & 0xffff)
570 #define LNET_NETTYP(net) (((net) >> 16) & 0xffff)
571 #define LNET_MKNET(typ,num) ((((__u32)(typ))<<16)|((__u32)(num)))
574 /* max value for numeric network address */
575 #define MAX_NUMERIC_VALUE 0xffffffff
578 #define ergo(a, b) (!(a) || (b))
579 /* logical equivalence */
580 #define equi(a, b) (!!(a) == !!(b))
582 #ifndef CFS_CURRENT_TIME
583 # define CFS_CURRENT_TIME time(0)
586 struct libcfs_device_userstate
588 int ldu_memhog_pages;
589 struct page *ldu_memhog_root_page;
592 /* what used to be in portals_lib.h */
594 # define MIN(a,b) (((a)<(b)) ? (a): (b))
597 # define MAX(a,b) (((a)>(b)) ? (a): (b))
600 #define MKSTR(ptr) ((ptr))? (ptr) : ""
602 static inline int cfs_size_round4 (int val)
604 return (val + 3) & (~0x3);
607 #ifndef HAVE_CFS_SIZE_ROUND
608 static inline int cfs_size_round (int val)
610 return (val + 7) & (~0x7);
612 #define HAVE_CFS_SIZE_ROUND
615 static inline int cfs_size_round16(int val)
617 return (val + 0xf) & (~0xf);
620 static inline int cfs_size_round32(int val)
622 return (val + 0x1f) & (~0x1f);
625 static inline int cfs_size_round0(int val)
629 return (val + 1 + 7) & (~0x7);
632 static inline size_t cfs_round_strlen(char *fset)
634 return (size_t)cfs_size_round((int)strlen(fset) + 1);
637 /* roundup \a val to power2 */
638 static inline unsigned int cfs_power2_roundup(unsigned int val)
640 if (val != LOWEST_BIT_SET(val)) { /* not a power of 2 already */
642 val &= ~LOWEST_BIT_SET(val);
643 } while (val != LOWEST_BIT_SET(val));
644 /* ...and round up */
650 #define LOGL(var,len,ptr) \
653 memcpy((char *)ptr, (const char *)var, len); \
654 ptr += cfs_size_round(len); \
657 #define LOGU(var,len,ptr) \
660 memcpy((char *)var, (const char *)ptr, len); \
661 ptr += cfs_size_round(len); \
665 * Lustre Network Driver types.
668 /* Only add to these values (i.e. don't ever change or redefine them):
669 * network addresses depend on them... */
672 GMLND = 3, /* obsolete, keep it so that libcfs_nid2str works */