xref: /titanic_50/usr/src/uts/common/avs/ns/sdbc/sd_hash.c (revision fcf3ce441efd61da9bb2884968af01cb7c1452cc)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/types.h>
27 #include <sys/ksynch.h>
28 #include <sys/cmn_err.h>
29 #include <sys/kmem.h>
30 #include <sys/ddi.h>
31 #include <sys/nsc_thread.h>
32 #include <sys/nsctl/nsctl.h>
33 
34 #include <sys/sdt.h>		/* dtrace is S10 or later */
35 
36 #include "sd_bcache.h"
37 #include "sd_hash.h"
38 
39 #if defined(_SD_DEBUG)
40 int _sd_hash_max_inlist = 0;
41 #endif
42 
43 
44 #define	_SD_HB_LOCKS 32
45 static kmutex_t _sd_hb_locks[_SD_HB_LOCKS];
46 
47 
48 /*
49  * _sdbc_hash_load - allocate all the locks for buckets.
50  *
51  *
52  */
53 int
_sdbc_hash_load(void)54 _sdbc_hash_load(void)
55 {
56 	int i;
57 	for (i = 0; i < _SD_HB_LOCKS; i++) {
58 		mutex_init(&_sd_hb_locks[i], NULL, MUTEX_DRIVER, NULL);
59 	}
60 	return (0);
61 }
62 
63 /*
64  * _sdbc_hash_unload - free all the locks for buckets.
65  *
66  *
67  */
68 void
_sdbc_hash_unload(void)69 _sdbc_hash_unload(void)
70 {
71 	int i;
72 	for (i = 0; i < _SD_HB_LOCKS; i++) {
73 		mutex_destroy(&_sd_hb_locks[i]);
74 	}
75 }
76 
77 
78 /*
79  * _sdbc_hash_configure - create a hash table
80  *
81  * ARGUMENTS:
82  *	num_ents - Number of entries (or hash buckets)
83  *	htype    - Type of memory to allocate.
84  *
85  * RETURNS:
86  *	The address of the hash table just created
87  *	or zero in the event of failure.
88  *
89  * USAGE:
90  *	This routine rounds of the number of entries to the next higher
91  *	power of 2. Allocate the hash buckets and initializes the locks
92  *	and returns the hash table that is created.
93  *	It is the caller's responsibility to save the hash_table and pass
94  *	it as a key for future accesses to the hash.
95  *	It is also the caller's responsibility to destroy the hash table
96  *	when appropriate.
97  */
98 
99 
100 _sd_hash_table_t *
_sdbc_hash_configure(int num_ents)101 _sdbc_hash_configure(int num_ents)
102 {
103 	_sd_hash_table_t *hash_table;
104 	_sd_hash_bucket_t *bucket;
105 	int i;
106 	int get_high_bit(int);
107 
108 	if ((hash_table = (_sd_hash_table_t *)
109 		nsc_kmem_zalloc(sizeof (_sd_hash_table_t),
110 				KM_SLEEP, sdbc_hash_mem)) == NULL)
111 		return (NULL);
112 
113 	hash_table->ht_bits = get_high_bit(num_ents);
114 	hash_table->ht_size = (1 << hash_table->ht_bits);
115 
116 	/*
117 	 * this is where we compute the mask used in the hash function
118 	 * the ht_nmask is basically an not of ht_mask used in hash
119 	 * function.
120 	 */
121 	hash_table->ht_mask = (hash_table->ht_size - 1);
122 	hash_table->ht_nmask = (~0 & ~(hash_table->ht_mask));
123 
124 	if ((hash_table->ht_buckets = (_sd_hash_bucket_t *)
125 		nsc_kmem_zalloc(hash_table->ht_size *
126 				sizeof (_sd_hash_bucket_t), KM_SLEEP,
127 				sdbc_hash_mem)) == NULL)
128 		return (NULL);
129 
130 	for (i = 0; i < (hash_table->ht_size); i++) {
131 		bucket = (hash_table->ht_buckets + i);
132 
133 		bucket->hb_lock = &_sd_hb_locks[i % _SD_HB_LOCKS];
134 		bucket->hb_head = bucket->hb_tail = NULL;
135 		bucket->hb_inlist = 0;
136 	}
137 
138 	return (hash_table);
139 }
140 
141 
142 /*
143  * _sdbc_hash_deconfigure - deconfigure a hash table
144  *
145  * ARGUMENTS:
146  *	hash_table - hash table that was created earlier on.
147  *
148  * RETURNS:
149  *	None.
150  *
151  * USAGE:
152  *	this routine deallocates memory that was allocated during the
153  *	hash create.
154  */
155 
156 void
_sdbc_hash_deconfigure(_sd_hash_table_t * hash_table)157 _sdbc_hash_deconfigure(_sd_hash_table_t *hash_table)
158 {
159 	if (!hash_table)
160 		return;
161 
162 	nsc_kmem_free(hash_table->ht_buckets,
163 			hash_table->ht_size * sizeof (_sd_hash_bucket_t));
164 
165 	nsc_kmem_free(hash_table, sizeof (_sd_hash_table_t));
166 }
167 
168 static int _sd_forced_hash_miss;
169 static int _sd_hash_collision;
170 
171 
172 /*
173  * _sd_hash_search - search the hash table for an entry
174  *
175  * ARGUMENTS:
176  *	cd	   - device that we are interested in.
177  *	block_num  - block number we are interested in.
178  *	hash_table - hash table to search in.
179  *
180  * RETURNS:
181  *	returns a hash header if a match was found in the hash table
182  *	for the device & block_num.
183  *	Else returns 0.
184  *
185  * USAGE:
186  *	This routine is called to check if a block already exists for
187  *	the device, block_num combination. If the block does not exist,
188  *	then a new block needs to be allocated and inserted into the hash
189  *	table for future references.
190  */
191 
192 _sd_hash_hd_t *
_sd_hash_search(int cd,nsc_off_t block_num,_sd_hash_table_t * table)193 _sd_hash_search(int cd, nsc_off_t block_num, _sd_hash_table_t *table)
194 {
195 	int i;
196 	_sd_hash_bucket_t *bucket;
197 	_sd_hash_hd_t *hptr;
198 #if defined(_SD_HASH_OPTIMIZE)
199 #define	MAX_HSEARCH_RETRIES	30
200 	int tries = 0;
201 	_sd_hash_hd_t *hnext;
202 	unsigned int seq;
203 
204 	i = HASH(cd, block_num, table);
205 	bucket = (table->ht_buckets + i);
206 retry_search:
207 	seq = bucket->hb_seq;
208 	for (hptr = bucket->hb_head; hptr; hptr = hnext) {
209 		/*
210 		 * Save pointer for next before checking the seq counter.
211 		 */
212 		hnext = hptr->hh_next;
213 		/*
214 		 * enforce ordering of load of hptr->hh_next
215 		 * above and bucket->hb_seq below
216 		 */
217 		sd_serialize();
218 		if (bucket->hb_seq != seq) {
219 			/*
220 			 * To avoid looping forever, break out if a certain
221 			 * limit is reached. Its okay to return miss
222 			 * since the insert will do a proper search.
223 			 */
224 			if (++tries < MAX_HSEARCH_RETRIES) goto retry_search;
225 			else {
226 				_sd_forced_hash_miss++;
227 				DTRACE_PROBE1(_sd_hash_search_end,
228 						int, _sd_forced_hash_miss);
229 				return (NULL);
230 			}
231 		}
232 		if ((hptr->hh_cd == cd) && (hptr->hh_blk_num == block_num))
233 			break;
234 		if (hptr->hh_blk_num > block_num) {
235 			DTRACE_PROBE1(_sd_hash_search_end,
236 					_sd_hash_hd_t *, hptr);
237 			return (NULL);
238 		}
239 	}
240 
241 	DTRACE_PROBE1(_sd_hash_search_end,
242 			_sd_hash_hd_t *, hptr);
243 	return (hptr);
244 #else
245 
246 	i = HASH(cd, block_num, table);
247 	bucket = (table->ht_buckets + i);
248 
249 	mutex_enter(bucket->hb_lock);
250 
251 	for (hptr = bucket->hb_head; hptr; hptr = hptr->hh_next) {
252 		if ((hptr->hh_cd == cd) && (hptr->hh_blk_num == block_num))
253 			break;
254 		/*
255 		 * the list is ordered. If we go beyond our block, no
256 		 * point searching
257 		 */
258 		if (hptr->hh_blk_num > block_num) {
259 			hptr = NULL;
260 			break;
261 		}
262 	}
263 	mutex_exit(bucket->hb_lock);
264 
265 	return (hptr);
266 #endif
267 }
268 
269 
270 /*
271  * _sd_hash_insert - insert an entry into the hash table
272  *
273  * ARGUMENTS:
274  *	cd	   - device that we are interested in.
275  *	block_num  - block number we are interested in.
276  *      hptr       - pointer to block that we are inserting.
277  *	table	   - hash table to search in.
278  *
279  * RETURNS:
280  *	Pointer to block that was passed in, except if the cd, block_num
281  *	already exists in the hash.  Caller must check for return
282  *	not equal hptr.
283  *
284  * USAGE:
285  *	this routine inserts the hptr into the appropriate hash bucket and
286  *	sets the cd, block_num in the block for future references.
287  */
288 
289 _sd_hash_hd_t *
_sd_hash_insert(int cd,nsc_off_t block_num,_sd_hash_hd_t * hptr,_sd_hash_table_t * table)290 _sd_hash_insert(int cd,
291 		nsc_off_t block_num,
292 		_sd_hash_hd_t *hptr,
293 		_sd_hash_table_t *table)
294 {
295 	int i;
296 	_sd_hash_hd_t *p;
297 	_sd_hash_bucket_t *bucket;
298 
299 	i = HASH(cd, block_num, table);
300 	bucket = (table->ht_buckets + i);
301 
302 #if defined(_SD_DEBUG)
303 	if (hptr->hh_hashed) {
304 		cmn_err(CE_WARN, "_sd_err: hptr %p bucket %p already hashed",
305 			hptr, bucket);
306 	}
307 #endif
308 	hptr->hh_cd = (ushort_t)cd;
309 	hptr->hh_blk_num = block_num;
310 
311 	mutex_enter(bucket->hb_lock);
312 
313 	for (p = bucket->hb_head; (p && (p->hh_blk_num <= block_num));
314 							p = p->hh_next) {
315 		if ((p->hh_cd == cd) && (p->hh_blk_num == block_num)) {
316 			mutex_exit(bucket->hb_lock);
317 			_sd_hash_collision++;
318 			DTRACE_PROBE2(_sd_hash_insert_end,
319 					_sd_hash_hd_t *, p,
320 					int, _sd_hash_collision);
321 
322 			return (p);
323 		}
324 	}
325 	hptr->hh_hashed = 1;
326 	/*
327 	 * At this point, (p) points to the next higher block number or is
328 	 * NULL. If it is NULL, we are queueing to the tail of list.
329 	 * Else, insert just before p
330 	 */
331 	if (p) {
332 		hptr->hh_next = p;
333 		if ((hptr->hh_prev = p->hh_prev) != NULL)
334 			p->hh_prev->hh_next = hptr;
335 		else
336 			bucket->hb_head = hptr;
337 		p->hh_prev = hptr;
338 	} else {
339 		hptr->hh_next = NULL;
340 		hptr->hh_prev = bucket->hb_tail;
341 		if (bucket->hb_head)
342 			bucket->hb_tail->hh_next = hptr;
343 		else
344 			bucket->hb_head = hptr;
345 		bucket->hb_tail = hptr;
346 	}
347 #if defined(_SD_HASH_OPTIMIZE)
348 	bucket->hb_seq++;
349 #endif
350 #if defined(_SD_DEBUG)
351 	if (_sd_hash_max_inlist < (int)++(bucket->hb_inlist))
352 		_sd_hash_max_inlist = bucket->hb_inlist;
353 #endif
354 	mutex_exit(bucket->hb_lock);
355 
356 	return (hptr);
357 }
358 
359 
360 
361 /*
362  * _sd_hash_delete - delete an entry from the hash table
363  *
364  * ARGUMENTS:
365  *	hptr	   - pointer to delete from hash table.
366  *	hash_table - hash table that was created earlier on.
367  *
368  * RETURNS:
369  *	0 on success.  -1 on errors.
370  *
371  * USAGE:
372  *	this routine deletes a hash entry from the hash table.
373  */
374 
375 int
_sd_hash_delete(_sd_hash_hd_t * hptr,_sd_hash_table_t * table)376 _sd_hash_delete(_sd_hash_hd_t *hptr, _sd_hash_table_t *table)
377 {
378 	int i;
379 	_sd_hash_bucket_t *bucket;
380 
381 	if (hptr->hh_hashed == 0) {
382 		DTRACE_PROBE(_sd_hash_delete_end1);
383 		return (-1);
384 	}
385 
386 	i = HASH(hptr->hh_cd, hptr->hh_blk_num, table);
387 	bucket = (table->ht_buckets + i);
388 
389 	/* was FAST */
390 	mutex_enter(bucket->hb_lock);
391 	if (hptr->hh_hashed == 0) {
392 		/* was FAST */
393 		mutex_exit(bucket->hb_lock);
394 		DTRACE_PROBE(_sd_hash_delete_end2);
395 		return (-1);
396 	}
397 	hptr->hh_hashed = 0;
398 #if defined(_SD_HASH_OPTIMIZE)
399 	/*
400 	 * Increment sequence counter on bucket. This will signal a lookup
401 	 * to redo the lookup since we might have broken the link used
402 	 * during the lookup.
403 	 */
404 	bucket->hb_seq++;
405 #endif
406 
407 	if (hptr->hh_prev)
408 		hptr->hh_prev->hh_next = hptr->hh_next;
409 	else
410 		bucket->hb_head = hptr->hh_next;
411 	if (hptr->hh_next)
412 		hptr->hh_next->hh_prev = hptr->hh_prev;
413 	else
414 		bucket->hb_tail = hptr->hh_prev;
415 #if defined(_SD_DEBUG)
416 	bucket->hb_inlist--;
417 #endif
418 	/* was FAST */
419 	mutex_exit(bucket->hb_lock);
420 
421 	return (0);
422 }
423 
424 /*
425  * _sd_hash_replace - replace 'old' with 'new' entry.
426  *
427  * ARGUMENTS:
428  *      old   - pointer to block being deleted (to be anonymous)
429  *      new   - pointer to block inserting in place.
430  *	table - hash table to search in.
431  *
432  * RETURNS:
433  *	pointer to inserted block.
434  *
435  * USAGE:
436  *	expects old & new to refer to same block.
437  *	new must not be already hashed.
438  */
439 
440 _sd_hash_hd_t *
_sd_hash_replace(_sd_hash_hd_t * old,_sd_hash_hd_t * new,_sd_hash_table_t * table)441 _sd_hash_replace(_sd_hash_hd_t *old, _sd_hash_hd_t *new,
442 			_sd_hash_table_t *table)
443 {
444 	int i;
445 	_sd_hash_bucket_t *bucket;
446 
447 	if ((old->hh_cd != new->hh_cd) || (old->hh_blk_num != new->hh_blk_num))
448 		cmn_err(CE_PANIC, "_sd_hash_replace: mismatch %p %p",
449 		    (void *)old, (void *)new);
450 	if (new->hh_hashed)
451 		cmn_err(CE_PANIC, "_sd_hash_replace: new %p already hashed",
452 		    (void *)new);
453 	if (old->hh_hashed == 0) {
454 		_sd_hash_hd_t *hptr;
455 		hptr = _sd_hash_insert(new->hh_cd, new->hh_blk_num, new, table);
456 
457 		DTRACE_PROBE1(_sd_hash_replace_end,
458 				_sd_hash_hd_t *, hptr);
459 
460 		return (hptr);
461 	}
462 
463 	i = HASH(old->hh_cd, old->hh_blk_num, table);
464 	bucket = (table->ht_buckets + i);
465 
466 	/* was FAST */
467 	mutex_enter(bucket->hb_lock);
468 	if (old->hh_hashed == 0) {
469 		_sd_hash_hd_t *hptr;
470 		/* was FAST */
471 		mutex_exit(bucket->hb_lock);
472 
473 		hptr = _sd_hash_insert(new->hh_cd, new->hh_blk_num, new, table);
474 
475 		DTRACE_PROBE1(_sd_hash_replace_end,
476 				_sd_hash_hd_t *, hptr);
477 		return (hptr);
478 	}
479 	old->hh_hashed = 0;
480 	new->hh_hashed = 1;
481 	new->hh_prev = old->hh_prev;
482 	new->hh_next = old->hh_next;
483 
484 	if (new->hh_prev)
485 		new->hh_prev->hh_next = new;
486 	else
487 		bucket->hb_head = new;
488 	if (new->hh_next)
489 		new->hh_next->hh_prev = new;
490 	else
491 		bucket->hb_tail = new;
492 #if defined(_SD_HASH_OPTIMIZE)
493 	bucket->hb_seq++;
494 #endif
495 	/* was FAST */
496 	mutex_exit(bucket->hb_lock);
497 
498 	return (new);
499 }
500