Home | History | Annotate | Line # | Download | only in raidframe
rf_interdecluster.c revision 1.5.2.3
      1  1.5.2.2  nathanw /*	$NetBSD: rf_interdecluster.c,v 1.5.2.3 2002/10/18 02:43:49 nathanw Exp $	*/
      2      1.1    oster /*
      3      1.1    oster  * Copyright (c) 1995 Carnegie-Mellon University.
      4      1.1    oster  * All rights reserved.
      5      1.1    oster  *
      6      1.1    oster  * Author: Khalil Amiri
      7      1.1    oster  *
      8      1.1    oster  * Permission to use, copy, modify and distribute this software and
      9      1.1    oster  * its documentation is hereby granted, provided that both the copyright
     10      1.1    oster  * notice and this permission notice appear in all copies of the
     11      1.1    oster  * software, derivative works or modified versions, and any portions
     12      1.1    oster  * thereof, and that both notices appear in supporting documentation.
     13      1.1    oster  *
     14      1.1    oster  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
     15      1.1    oster  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
     16      1.1    oster  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
     17      1.1    oster  *
     18      1.1    oster  * Carnegie Mellon requests users of this software to return to
     19      1.1    oster  *
     20      1.1    oster  *  Software Distribution Coordinator  or  Software.Distribution (at) CS.CMU.EDU
     21      1.1    oster  *  School of Computer Science
     22      1.1    oster  *  Carnegie Mellon University
     23      1.1    oster  *  Pittsburgh PA 15213-3890
     24      1.1    oster  *
     25      1.1    oster  * any improvements or extensions that they make and grant Carnegie the
     26      1.1    oster  * rights to redistribute these changes.
     27      1.1    oster  */
     28      1.1    oster 
     29      1.1    oster /************************************************************
     30      1.1    oster  *
     31      1.3    oster  * rf_interdecluster.c -- implements interleaved declustering
     32      1.1    oster  *
     33      1.1    oster  ************************************************************/
     34  1.5.2.2  nathanw 
     35  1.5.2.2  nathanw #include <sys/cdefs.h>
     36  1.5.2.2  nathanw __KERNEL_RCSID(0, "$NetBSD: rf_interdecluster.c,v 1.5.2.3 2002/10/18 02:43:49 nathanw Exp $");
     37      1.1    oster 
     38      1.5    oster #include "rf_archs.h"
     39      1.5    oster 
     40      1.5    oster #if RF_INCLUDE_INTERDECLUSTER > 0
     41      1.1    oster 
     42  1.5.2.1  nathanw #include <dev/raidframe/raidframevar.h>
     43  1.5.2.1  nathanw 
     44      1.1    oster #include "rf_raid.h"
     45      1.1    oster #include "rf_interdecluster.h"
     46      1.1    oster #include "rf_dag.h"
     47      1.1    oster #include "rf_dagutils.h"
     48      1.1    oster #include "rf_dagfuncs.h"
     49      1.1    oster #include "rf_general.h"
     50      1.1    oster #include "rf_utils.h"
     51      1.1    oster #include "rf_dagffrd.h"
     52      1.1    oster #include "rf_dagdegrd.h"
     53      1.1    oster #include "rf_dagffwr.h"
     54      1.1    oster #include "rf_dagdegwr.h"
     55      1.1    oster 
     56      1.1    oster typedef struct RF_InterdeclusterConfigInfo_s {
     57      1.3    oster 	RF_RowCol_t **stripeIdentifier;	/* filled in at config time and used
     58      1.3    oster 					 * by IdentifyStripe */
     59      1.3    oster 	RF_StripeCount_t numSparingRegions;
     60      1.3    oster 	RF_StripeCount_t stripeUnitsPerSparingRegion;
     61      1.3    oster 	RF_SectorNum_t mirrorStripeOffset;
     62      1.3    oster }       RF_InterdeclusterConfigInfo_t;
     63      1.3    oster 
     64      1.3    oster int
     65      1.3    oster rf_ConfigureInterDecluster(
     66      1.3    oster     RF_ShutdownList_t ** listp,
     67      1.3    oster     RF_Raid_t * raidPtr,
     68      1.3    oster     RF_Config_t * cfgPtr)
     69      1.1    oster {
     70      1.3    oster 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
     71      1.3    oster 	RF_StripeCount_t num_used_stripeUnitsPerDisk;
     72      1.3    oster 	RF_InterdeclusterConfigInfo_t *info;
     73      1.3    oster 	RF_RowCol_t i, tmp, SUs_per_region;
     74      1.3    oster 
     75      1.3    oster 	/* create an Interleaved Declustering configuration structure */
     76      1.3    oster 	RF_MallocAndAdd(info, sizeof(RF_InterdeclusterConfigInfo_t), (RF_InterdeclusterConfigInfo_t *),
     77      1.3    oster 	    raidPtr->cleanupList);
     78      1.3    oster 	if (info == NULL)
     79      1.3    oster 		return (ENOMEM);
     80      1.3    oster 	layoutPtr->layoutSpecificInfo = (void *) info;
     81      1.3    oster 
     82      1.3    oster 	/* fill in the config structure.  */
     83      1.3    oster 	SUs_per_region = raidPtr->numCol * (raidPtr->numCol - 1);
     84      1.3    oster 	info->stripeIdentifier = rf_make_2d_array(SUs_per_region, 2, raidPtr->cleanupList);
     85      1.3    oster 	if (info->stripeIdentifier == NULL)
     86      1.3    oster 		return (ENOMEM);
     87      1.3    oster 	for (i = 0; i < SUs_per_region; i++) {
     88      1.3    oster 		info->stripeIdentifier[i][0] = i / (raidPtr->numCol - 1);
     89      1.3    oster 		tmp = i / raidPtr->numCol;
     90      1.3    oster 		info->stripeIdentifier[i][1] = (i + 1 + tmp) % raidPtr->numCol;
     91      1.3    oster 	}
     92      1.3    oster 
     93      1.3    oster 	/* no spare tables */
     94      1.3    oster 	RF_ASSERT(raidPtr->numRow == 1);
     95      1.3    oster 
     96      1.3    oster 	/* fill in the remaining layout parameters */
     97      1.3    oster 
     98      1.3    oster 	/* total number of stripes should a multiple of 2*numCol: Each sparing
     99      1.3    oster 	 * region consists of 2*numCol stripes: n-1 primary copy, n-1
    100      1.3    oster 	 * secondary copy and 2 for spare .. */
    101      1.3    oster 	num_used_stripeUnitsPerDisk = layoutPtr->stripeUnitsPerDisk - (layoutPtr->stripeUnitsPerDisk %
    102      1.3    oster 	    (2 * raidPtr->numCol));
    103      1.3    oster 	info->numSparingRegions = num_used_stripeUnitsPerDisk / (2 * raidPtr->numCol);
    104      1.3    oster 	/* this is in fact the number of stripe units (that are primary data
    105      1.3    oster 	 * copies) in the sparing region */
    106      1.3    oster 	info->stripeUnitsPerSparingRegion = raidPtr->numCol * (raidPtr->numCol - 1);
    107      1.3    oster 	info->mirrorStripeOffset = info->numSparingRegions * (raidPtr->numCol + 1);
    108      1.3    oster 	layoutPtr->numStripe = info->numSparingRegions * info->stripeUnitsPerSparingRegion;
    109      1.3    oster 	layoutPtr->numDataCol = 1;
    110      1.3    oster 	layoutPtr->dataSectorsPerStripe = layoutPtr->numDataCol * layoutPtr->sectorsPerStripeUnit;
    111      1.3    oster 	layoutPtr->numParityCol = 1;
    112      1.3    oster 
    113      1.3    oster 	layoutPtr->dataStripeUnitsPerDisk = num_used_stripeUnitsPerDisk;
    114      1.3    oster 
    115      1.3    oster 	raidPtr->sectorsPerDisk =
    116      1.3    oster 	    num_used_stripeUnitsPerDisk * layoutPtr->sectorsPerStripeUnit;
    117      1.1    oster 
    118      1.3    oster 	raidPtr->totalSectors =
    119      1.3    oster 	    (layoutPtr->numStripe) * layoutPtr->sectorsPerStripeUnit;
    120      1.1    oster 
    121      1.3    oster 	layoutPtr->stripeUnitsPerDisk = raidPtr->sectorsPerDisk / layoutPtr->sectorsPerStripeUnit;
    122      1.1    oster 
    123      1.3    oster 	return (0);
    124      1.1    oster }
    125      1.1    oster 
    126      1.3    oster int
    127      1.3    oster rf_GetDefaultNumFloatingReconBuffersInterDecluster(RF_Raid_t * raidPtr)
    128      1.1    oster {
    129      1.3    oster 	return (30);
    130      1.1    oster }
    131      1.1    oster 
    132      1.3    oster RF_HeadSepLimit_t
    133      1.3    oster rf_GetDefaultHeadSepLimitInterDecluster(RF_Raid_t * raidPtr)
    134      1.1    oster {
    135      1.3    oster 	return (raidPtr->sectorsPerDisk);
    136      1.1    oster }
    137      1.1    oster 
    138      1.3    oster RF_ReconUnitCount_t
    139      1.3    oster rf_GetNumSpareRUsInterDecluster(
    140      1.3    oster     RF_Raid_t * raidPtr)
    141      1.1    oster {
    142      1.3    oster 	RF_InterdeclusterConfigInfo_t *info = (RF_InterdeclusterConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
    143      1.1    oster 
    144      1.3    oster 	return (2 * ((RF_ReconUnitCount_t) info->numSparingRegions));
    145      1.3    oster 	/* the layout uses two stripe units per disk as spare within each
    146      1.3    oster 	 * sparing region */
    147      1.1    oster }
    148      1.1    oster /* Maps to the primary copy of the data, i.e. the first mirror pair */
    149      1.3    oster void
    150      1.3    oster rf_MapSectorInterDecluster(
    151      1.3    oster     RF_Raid_t * raidPtr,
    152      1.3    oster     RF_RaidAddr_t raidSector,
    153      1.3    oster     RF_RowCol_t * row,
    154      1.3    oster     RF_RowCol_t * col,
    155      1.3    oster     RF_SectorNum_t * diskSector,
    156      1.3    oster     int remap)
    157      1.1    oster {
    158      1.3    oster 	RF_InterdeclusterConfigInfo_t *info = (RF_InterdeclusterConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
    159      1.3    oster 	RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
    160      1.3    oster 	RF_StripeNum_t su_offset_into_disk, mirror_su_offset_into_disk;
    161      1.3    oster 	RF_StripeNum_t sparing_region_id, index_within_region;
    162      1.3    oster 	int     col_before_remap;
    163      1.3    oster 
    164      1.3    oster 	*row = 0;
    165      1.3    oster 	sparing_region_id = SUID / info->stripeUnitsPerSparingRegion;
    166      1.3    oster 	index_within_region = SUID % info->stripeUnitsPerSparingRegion;
    167      1.3    oster 	su_offset_into_disk = index_within_region % (raidPtr->numCol - 1);
    168      1.3    oster 	mirror_su_offset_into_disk = index_within_region / raidPtr->numCol;
    169      1.3    oster 	col_before_remap = index_within_region / (raidPtr->numCol - 1);
    170      1.3    oster 
    171      1.3    oster 	if (!remap) {
    172      1.3    oster 		*col = col_before_remap;;
    173      1.3    oster 		*diskSector = (su_offset_into_disk + ((raidPtr->numCol - 1) * sparing_region_id)) *
    174      1.3    oster 		    raidPtr->Layout.sectorsPerStripeUnit;
    175      1.3    oster 		*diskSector += (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
    176      1.3    oster 	} else {
    177      1.3    oster 		/* remap sector to spare space... */
    178      1.3    oster 		*diskSector = sparing_region_id * (raidPtr->numCol + 1) * raidPtr->Layout.sectorsPerStripeUnit;
    179      1.3    oster 		*diskSector += (raidPtr->numCol - 1) * raidPtr->Layout.sectorsPerStripeUnit;
    180      1.3    oster 		*diskSector += (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
    181      1.3    oster 		*col = (index_within_region + 1 + mirror_su_offset_into_disk) % raidPtr->numCol;
    182      1.3    oster 		*col = (*col + 1) % raidPtr->numCol;
    183      1.3    oster 		if (*col == col_before_remap)
    184      1.3    oster 			*col = (*col + 1) % raidPtr->numCol;
    185      1.3    oster 	}
    186      1.1    oster }
    187      1.1    oster /* Maps to the second copy of the mirror pair. */
    188      1.3    oster void
    189      1.3    oster rf_MapParityInterDecluster(
    190      1.3    oster     RF_Raid_t * raidPtr,
    191      1.3    oster     RF_RaidAddr_t raidSector,
    192      1.3    oster     RF_RowCol_t * row,
    193      1.3    oster     RF_RowCol_t * col,
    194      1.3    oster     RF_SectorNum_t * diskSector,
    195      1.3    oster     int remap)
    196      1.1    oster {
    197      1.3    oster 	RF_InterdeclusterConfigInfo_t *info = (RF_InterdeclusterConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
    198      1.3    oster 	RF_StripeNum_t sparing_region_id, index_within_region, mirror_su_offset_into_disk;
    199      1.3    oster 	RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
    200      1.3    oster 	int     col_before_remap;
    201      1.3    oster 
    202      1.3    oster 	sparing_region_id = SUID / info->stripeUnitsPerSparingRegion;
    203      1.3    oster 	index_within_region = SUID % info->stripeUnitsPerSparingRegion;
    204      1.3    oster 	mirror_su_offset_into_disk = index_within_region / raidPtr->numCol;
    205      1.3    oster 	col_before_remap = (index_within_region + 1 + mirror_su_offset_into_disk) % raidPtr->numCol;
    206      1.3    oster 
    207      1.3    oster 	*row = 0;
    208      1.3    oster 	if (!remap) {
    209      1.3    oster 		*col = col_before_remap;
    210      1.3    oster 		*diskSector = info->mirrorStripeOffset * raidPtr->Layout.sectorsPerStripeUnit;
    211      1.3    oster 		*diskSector += sparing_region_id * (raidPtr->numCol - 1) * raidPtr->Layout.sectorsPerStripeUnit;
    212      1.3    oster 		*diskSector += mirror_su_offset_into_disk * raidPtr->Layout.sectorsPerStripeUnit;
    213      1.3    oster 		*diskSector += (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
    214      1.3    oster 	} else {
    215      1.3    oster 		/* remap parity to spare space ... */
    216      1.3    oster 		*diskSector = sparing_region_id * (raidPtr->numCol + 1) * raidPtr->Layout.sectorsPerStripeUnit;
    217      1.3    oster 		*diskSector += (raidPtr->numCol) * raidPtr->Layout.sectorsPerStripeUnit;
    218      1.3    oster 		*diskSector += (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
    219      1.3    oster 		*col = index_within_region / (raidPtr->numCol - 1);
    220      1.3    oster 		*col = (*col + 1) % raidPtr->numCol;
    221      1.3    oster 		if (*col == col_before_remap)
    222      1.3    oster 			*col = (*col + 1) % raidPtr->numCol;
    223      1.3    oster 	}
    224      1.1    oster }
    225      1.1    oster 
    226      1.3    oster void
    227      1.3    oster rf_IdentifyStripeInterDecluster(
    228      1.3    oster     RF_Raid_t * raidPtr,
    229      1.3    oster     RF_RaidAddr_t addr,
    230      1.3    oster     RF_RowCol_t ** diskids,
    231      1.3    oster     RF_RowCol_t * outRow)
    232      1.1    oster {
    233      1.3    oster 	RF_InterdeclusterConfigInfo_t *info = (RF_InterdeclusterConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
    234      1.3    oster 	RF_StripeNum_t SUID;
    235      1.1    oster 
    236      1.3    oster 	SUID = addr / raidPtr->Layout.sectorsPerStripeUnit;
    237      1.3    oster 	SUID = SUID % info->stripeUnitsPerSparingRegion;
    238      1.1    oster 
    239      1.3    oster 	*outRow = 0;
    240      1.3    oster 	*diskids = info->stripeIdentifier[SUID];
    241      1.1    oster }
    242      1.1    oster 
    243      1.3    oster void
    244      1.3    oster rf_MapSIDToPSIDInterDecluster(
    245      1.3    oster     RF_RaidLayout_t * layoutPtr,
    246      1.3    oster     RF_StripeNum_t stripeID,
    247      1.3    oster     RF_StripeNum_t * psID,
    248      1.3    oster     RF_ReconUnitNum_t * which_ru)
    249      1.1    oster {
    250      1.3    oster 	*which_ru = 0;
    251      1.3    oster 	*psID = stripeID;
    252      1.1    oster }
    253      1.1    oster /******************************************************************************
    254      1.1    oster  * select a graph to perform a single-stripe access
    255      1.1    oster  *
    256      1.1    oster  * Parameters:  raidPtr    - description of the physical array
    257      1.1    oster  *              type       - type of operation (read or write) requested
    258      1.1    oster  *              asmap      - logical & physical addresses for this access
    259      1.1    oster  *              createFunc - name of function to use to create the graph
    260      1.1    oster  *****************************************************************************/
    261      1.1    oster 
    262      1.3    oster void
    263      1.3    oster rf_RAIDIDagSelect(
    264      1.3    oster     RF_Raid_t * raidPtr,
    265      1.3    oster     RF_IoType_t type,
    266      1.3    oster     RF_AccessStripeMap_t * asmap,
    267      1.3    oster     RF_VoidFuncPtr * createFunc)
    268      1.1    oster {
    269      1.3    oster 	RF_ASSERT(RF_IO_IS_R_OR_W(type));
    270      1.1    oster 
    271      1.3    oster 	if (asmap->numDataFailed + asmap->numParityFailed > 1) {
    272      1.3    oster 		RF_ERRORMSG("Multiple disks failed in a single group!  Aborting I/O operation.\n");
    273      1.3    oster 		*createFunc = NULL;
    274      1.3    oster 		return;
    275      1.3    oster 	}
    276      1.3    oster 	*createFunc = (type == RF_IO_TYPE_READ) ? (RF_VoidFuncPtr) rf_CreateFaultFreeReadDAG : (RF_VoidFuncPtr) rf_CreateRaidOneWriteDAG;
    277      1.3    oster 	if (type == RF_IO_TYPE_READ) {
    278      1.3    oster 		if (asmap->numDataFailed == 0)
    279      1.3    oster 			*createFunc = (RF_VoidFuncPtr) rf_CreateMirrorPartitionReadDAG;
    280      1.3    oster 		else
    281      1.3    oster 			*createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneDegradedReadDAG;
    282      1.3    oster 	} else
    283      1.3    oster 		*createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneWriteDAG;
    284      1.1    oster }
    285      1.5    oster #endif /* RF_INCLUDE_INTERDECLUSTER > 0 */
    286