rf_diskqueue.h revision 1.1 1 /* $NetBSD: rf_diskqueue.h,v 1.1 1998/11/13 04:20:29 oster Exp $ */
2 /*
3 * Copyright (c) 1995 Carnegie-Mellon University.
4 * All rights reserved.
5 *
6 * Author: Mark Holland
7 *
8 * Permission to use, copy, modify and distribute this software and
9 * its documentation is hereby granted, provided that both the copyright
10 * notice and this permission notice appear in all copies of the
11 * software, derivative works or modified versions, and any portions
12 * thereof, and that both notices appear in supporting documentation.
13 *
14 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
15 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
16 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
17 *
18 * Carnegie Mellon requests users of this software to return to
19 *
20 * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU
21 * School of Computer Science
22 * Carnegie Mellon University
23 * Pittsburgh PA 15213-3890
24 *
25 * any improvements or extensions that they make and grant Carnegie the
26 * rights to redistribute these changes.
27 */
28
29 /*****************************************************************************************
30 *
31 * rf_diskqueue.h -- header file for disk queues
32 *
33 * see comments in rf_diskqueue.c
34 *
35 ****************************************************************************************/
36 /*
37 *
38 * :
39 *
40 * Log: rf_diskqueue.h,v
41 * Revision 1.31 1996/08/07 21:08:49 jimz
42 * b_proc -> kb_proc (IRIX complained)
43 *
44 * Revision 1.30 1996/06/18 20:53:11 jimz
45 * fix up disk queueing (remove configure routine,
46 * add shutdown list arg to create routines)
47 *
48 * Revision 1.29 1996/06/13 20:38:19 jimz
49 * fix queue type in DiskQueueData
50 *
51 * Revision 1.28 1996/06/10 11:55:47 jimz
52 * Straightened out some per-array/not-per-array distinctions, fixed
53 * a couple bugs related to confusion. Added shutdown lists. Removed
54 * layout shutdown function (now subsumed by shutdown lists).
55 *
56 * Revision 1.27 1996/06/07 22:26:27 jimz
57 * type-ify which_ru (RF_ReconUnitNum_t)
58 *
59 * Revision 1.26 1996/06/07 21:33:04 jimz
60 * begin using consistent types for sector numbers,
61 * stripe numbers, row+col numbers, recon unit numbers
62 *
63 * Revision 1.25 1996/06/06 17:29:12 jimz
64 * track arm position of last I/O dequeued
65 *
66 * Revision 1.24 1996/06/05 18:06:02 jimz
67 * Major code cleanup. The Great Renaming is now done.
68 * Better modularity. Better typing. Fixed a bunch of
69 * synchronization bugs. Made a lot of global stuff
70 * per-desc or per-array. Removed dead code.
71 *
72 * Revision 1.23 1996/06/02 17:31:48 jimz
73 * Moved a lot of global stuff into array structure, where it belongs.
74 * Fixed up paritylogging, pss modules in this manner. Some general
75 * code cleanup. Removed lots of dead code, some dead files.
76 *
77 * Revision 1.22 1996/05/30 23:22:16 jimz
78 * bugfixes of serialization, timing problems
79 * more cleanup
80 *
81 * Revision 1.21 1996/05/30 11:29:41 jimz
82 * Numerous bug fixes. Stripe lock release code disagreed with the taking code
83 * about when stripes should be locked (I made it consistent: no parity, no lock)
84 * There was a lot of extra serialization of I/Os which I've removed- a lot of
85 * it was to calculate values for the cache code, which is no longer with us.
86 * More types, function, macro cleanup. Added code to properly quiesce the array
87 * on shutdown. Made a lot of stuff array-specific which was (bogusly) general
88 * before. Fixed memory allocation, freeing bugs.
89 *
90 * Revision 1.20 1996/05/24 22:17:04 jimz
91 * continue code + namespace cleanup
92 * typed a bunch of flags
93 *
94 * Revision 1.19 1996/05/24 01:59:45 jimz
95 * another checkpoint in code cleanup for release
96 * time to sync kernel tree
97 *
98 * Revision 1.18 1996/05/23 21:46:35 jimz
99 * checkpoint in code cleanup (release prep)
100 * lots of types, function names have been fixed
101 *
102 * Revision 1.17 1996/05/23 00:33:23 jimz
103 * code cleanup: move all debug decls to rf_options.c, all extern
104 * debug decls to rf_options.h, all debug vars preceded by rf_
105 *
106 * Revision 1.16 1996/05/18 19:51:34 jimz
107 * major code cleanup- fix syntax, make some types consistent,
108 * add prototypes, clean out dead code, et cetera
109 *
110 * Revision 1.15 1996/05/10 19:39:31 jimz
111 * add prev pointer to DiskQueueData
112 *
113 * Revision 1.14 1996/05/10 16:24:04 jimz
114 * mark old defines as deprecated, add RF_ defines
115 *
116 * Revision 1.13 1995/12/01 15:59:04 root
117 * added copyright info
118 *
119 * Revision 1.12 1995/11/07 16:26:44 wvcii
120 * added Peek() function to diskqueuesw
121 *
122 * Revision 1.11 1995/10/05 02:33:15 jimz
123 * made queue lens longs (less instructions to read :-)
124 *
125 * Revision 1.10 1995/10/04 07:07:07 wvcii
126 * queue->numOutstanding now valid for user & sim
127 * user tested & verified, sim untested
128 *
129 * Revision 1.9 1995/09/12 00:21:37 wvcii
130 * added support for tracing disk queue time
131 *
132 * Revision 1.8 95/04/24 13:25:51 holland
133 * rewrite to move disk queues, recon, & atomic RMW to kernel
134 *
135 * Revision 1.6.10.2 1995/04/03 20:13:56 holland
136 * added numOutstanding and maxOutstanding to support moving
137 * disk queues into kernel code
138 *
139 * Revision 1.6.10.1 1995/04/03 20:03:56 holland
140 * initial checkin on branch
141 *
142 * Revision 1.6 1995/03/03 18:34:33 rachad
143 * Simulator mechanism added
144 *
145 * Revision 1.5 1995/03/01 20:25:48 holland
146 * kernelization changes
147 *
148 * Revision 1.4 1995/02/03 22:31:36 holland
149 * many changes related to kernelization
150 *
151 * Revision 1.3 1995/02/01 14:25:19 holland
152 * began changes for kernelization:
153 * changed all instances of mutex_t and cond_t to DECLARE macros
154 * converted configuration code to use config structure
155 *
156 * Revision 1.2 1994/11/29 20:36:02 danner
157 * Added symbolic constants for io_type (e.g,IO_TYPE_READ)
158 * and support for READ_OP_WRITE
159 *
160 */
161
162
163 #ifndef _RF__RF_DISKQUEUE_H_
164 #define _RF__RF_DISKQUEUE_H_
165
166 #include "rf_threadstuff.h"
167 #include "rf_acctrace.h"
168 #include "rf_alloclist.h"
169 #include "rf_types.h"
170 #include "rf_etimer.h"
171
172
173 #if defined(__NetBSD__) && defined(_KERNEL)
174 #include "rf_netbsd.h"
175 #endif
176
177
178 #define RF_IO_NORMAL_PRIORITY 1
179 #define RF_IO_LOW_PRIORITY 0
180
181 /* the data held by a disk queue entry */
182 struct RF_DiskQueueData_s {
183 RF_SectorNum_t sectorOffset; /* sector offset into the disk */
184 RF_SectorCount_t numSector; /* number of sectors to read/write */
185 RF_IoType_t type; /* read/write/nop */
186 caddr_t buf; /* buffer pointer */
187 RF_StripeNum_t parityStripeID; /* the RAID parity stripe ID this access is for */
188 RF_ReconUnitNum_t which_ru; /* which RU within this parity stripe */
189 int priority; /* the priority of this request */
190 int (*CompleteFunc)(void *,int);/* function to be called upon completion */
191 int (*AuxFunc)(void *,...); /* function called upon completion of the first I/O of a Read_Op_Write pair*/
192 void *argument; /* argument to be passed to CompleteFunc */
193 #ifdef SIMULATE
194 RF_Owner_t owner; /* which task is responsible for this request */
195 #endif /* SIMULATE */
196 void *raidPtr; /* needed for simulation */
197 RF_AccTraceEntry_t *tracerec; /* perf mon only */
198 RF_Etimer_t qtime; /* perf mon only - time request is in queue */
199 long entryTime;
200 RF_DiskQueueData_t *next;
201 RF_DiskQueueData_t *prev;
202 caddr_t buf2; /* for read-op-write */
203 dev_t dev; /* the device number for in-kernel version */
204 RF_DiskQueue_t *queue; /* the disk queue to which this req is targeted */
205 RF_DiskQueueDataFlags_t flags; /* flags controlling operation */
206
207 #ifdef KERNEL
208 struct proc *b_proc; /* the b_proc from the original bp passed into the driver for this I/O */
209 struct buf *bp; /* a bp to use to get this I/O done */
210 #endif /* KERNEL */
211 };
212
213 #define RF_LOCK_DISK_QUEUE 0x01
214 #define RF_UNLOCK_DISK_QUEUE 0x02
215
216 /* note: "Create" returns type-specific queue header pointer cast to (void *) */
217 struct RF_DiskQueueSW_s {
218 RF_DiskQueueType_t queueType;
219 void *(*Create)(RF_SectorCount_t, RF_AllocListElem_t *, RF_ShutdownList_t **); /* creation routine -- one call per queue in system */
220 void (*Enqueue)(void *,RF_DiskQueueData_t * ,int); /* enqueue routine */
221 RF_DiskQueueData_t *(*Dequeue)(void *); /* dequeue routine */
222 RF_DiskQueueData_t *(*Peek)(void *); /* peek at head of queue */
223
224 /* the rest are optional: they improve performance, but the driver will deal with it if they don't exist */
225 int (*Promote)(void *, RF_StripeNum_t, RF_ReconUnitNum_t); /* promotes priority of tagged accesses */
226 };
227
228 struct RF_DiskQueue_s {
229 RF_DiskQueueSW_t *qPtr; /* access point to queue functions */
230 void *qHdr; /* queue header, of whatever type */
231 RF_DECLARE_MUTEX(mutex) /* mutex locking data structures */
232 RF_DECLARE_COND(cond) /* condition variable for synchronization */
233 long numOutstanding; /* number of I/Os currently outstanding on disk */
234 long maxOutstanding; /* max # of I/Os that can be outstanding on a disk (in-kernel only) */
235 int curPriority; /* the priority of accs all that are currently outstanding */
236 long queueLength; /* number of requests in queue */
237 RF_DiskQueueData_t *nextLockingOp; /* a locking op that has arrived at the head of the queue & is waiting for drainage */
238 RF_DiskQueueData_t *unlockingOp; /* used at user level to communicate unlocking op b/w user (or dag exec) & disk threads */
239 int numWaiting; /* number of threads waiting on this variable. user-level only */
240 RF_DiskQueueFlags_t flags; /* terminate, locked */
241 RF_Raid_t *raidPtr; /* associated array */
242 dev_t dev; /* device number for kernel version */
243 RF_SectorNum_t last_deq_sector; /* last sector number dequeued or dispatched */
244 int row, col; /* debug only */
245 #if defined(__NetBSD__) && defined(_KERNEL)
246 struct raidcinfo *rf_cinfo; /* disks component info.. */
247 #endif
248 };
249
250 #define RF_DQ_LOCKED 0x02 /* no new accs allowed until queue is explicitly unlocked */
251
252 /* macros setting & returning information about queues and requests */
253 #define RF_QUEUE_LOCKED(_q) ((_q)->flags & RF_DQ_LOCKED)
254 #define RF_QUEUE_EMPTY(_q) (((_q)->numOutstanding == 0) && ((_q)->nextLockingOp == NULL) && !RF_QUEUE_LOCKED(_q))
255 #define RF_QUEUE_FULL(_q) ((_q)->numOutstanding == (_q)->maxOutstanding)
256
257 #define RF_LOCK_QUEUE(_q) (_q)->flags |= RF_DQ_LOCKED
258 #define RF_UNLOCK_QUEUE(_q) (_q)->flags &= ~RF_DQ_LOCKED
259
260 #define RF_LOCK_QUEUE_MUTEX(_q_,_wh_) RF_LOCK_MUTEX((_q_)->mutex)
261 #define RF_UNLOCK_QUEUE_MUTEX(_q_,_wh_) RF_UNLOCK_MUTEX((_q_)->mutex)
262
263 #define RF_LOCKING_REQ(_r) ((_r)->flags & RF_LOCK_DISK_QUEUE)
264 #define RF_UNLOCKING_REQ(_r) ((_r)->flags & RF_UNLOCK_DISK_QUEUE)
265
266 /* whether it is ok to dispatch a regular request */
267 #define RF_OK_TO_DISPATCH(_q_,_r_) \
268 (RF_QUEUE_EMPTY(_q_) || \
269 (!RF_QUEUE_FULL(_q_) && ((_r_)->priority >= (_q_)->curPriority)))
270
271 int rf_ConfigureDiskQueueSystem(RF_ShutdownList_t **listp);
272
273 void rf_TerminateDiskQueues(RF_Raid_t *raidPtr);
274
275 int rf_ConfigureDiskQueues(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
276 RF_Config_t *cfgPtr);
277
278 void rf_DiskIOEnqueue(RF_DiskQueue_t *queue, RF_DiskQueueData_t *req, int pri);
279
280 #if !defined(KERNEL) && !defined(SIMULATE)
281 void rf_BroadcastOnQueue(RF_DiskQueue_t *queue);
282 #endif /* !KERNEL && !SIMULATE */
283
284 #ifndef KERNEL
285 RF_DiskQueueData_t *rf_DiskIODequeue(RF_DiskQueue_t *queue);
286 #else /* !KERNEL */
287 void rf_DiskIOComplete(RF_DiskQueue_t *queue, RF_DiskQueueData_t *req, int status);
288 #endif /* !KERNEL */
289
290 int rf_DiskIOPromote(RF_DiskQueue_t *queue, RF_StripeNum_t parityStripeID,
291 RF_ReconUnitNum_t which_ru);
292
293 RF_DiskQueueData_t *rf_CreateDiskQueueData(RF_IoType_t typ,
294 RF_SectorNum_t ssect, RF_SectorCount_t nsect, caddr_t buf,
295 RF_StripeNum_t parityStripeID, RF_ReconUnitNum_t which_ru,
296 int (*wakeF)(void *, int),
297 void *arg, RF_DiskQueueData_t *next, RF_AccTraceEntry_t *tracerec,
298 void *raidPtr, RF_DiskQueueDataFlags_t flags, void *kb_proc);
299
300 RF_DiskQueueData_t *rf_CreateDiskQueueDataFull(RF_IoType_t typ,
301 RF_SectorNum_t ssect, RF_SectorCount_t nsect, caddr_t buf,
302 RF_StripeNum_t parityStripeID, RF_ReconUnitNum_t which_ru,
303 int (*wakeF)(void *, int),
304 void *arg, RF_DiskQueueData_t *next, RF_AccTraceEntry_t *tracerec,
305 int priority, int (*AuxFunc)(void *,...), caddr_t buf2,
306 void *raidPtr, RF_DiskQueueDataFlags_t flags, void *kb_proc);
307
308 void rf_FreeDiskQueueData(RF_DiskQueueData_t *p);
309
310 #endif /* !_RF__RF_DISKQUEUE_H_ */
311