/*      $NetBSD: rf_dagfuncs.c,v 1.35 2021/08/07 16:19:15 thorpej Exp $ */
/*
* Copyright (c) 1995 Carnegie-Mellon University.
* All rights reserved.
*
* Author: Mark Holland, William V. Courtright II
*
* Permission to use, copy, modify and distribute this software and
* its documentation is hereby granted, provided that both the copyright
* notice and this permission notice appear in all copies of the
* software, derivative works or modified versions, and any portions
* thereof, and that both notices appear in supporting documentation.
*
* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
* CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
* FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
*
* Carnegie Mellon requests users of this software to return to
*
*  Software Distribution Coordinator  or  [email protected]
*  School of Computer Science
*  Carnegie Mellon University
*  Pittsburgh PA 15213-3890
*
* any improvements or extensions that they make and grant Carnegie the
* rights to redistribute these changes.
*/

/*
* dagfuncs.c -- DAG node execution routines
*
* Rules:
* 1. Every DAG execution function must eventually cause node->status to
*    get set to "good" or "bad", and "FinishNode" to be called. In the
*    case of nodes that complete immediately (xor, NullNodeFunc, etc),
*    the node execution function can do these two things directly. In
*    the case of nodes that have to wait for some event (a disk read to
*    complete, a lock to be released, etc) to occur before they can
*    complete, this is typically achieved by having whatever module
*    is doing the operation call GenericWakeupFunc upon completion.
* 2. DAG execution functions should check the status in the DAG header
*    and NOP out their operations if the status is not "enable". However,
*    execution functions that release resources must be sure to release
*    them even when they NOP out the function that would use them.
*    Functions that acquire resources should go ahead and acquire them
*    even when they NOP, so that a downstream release node will not have
*    to check to find out whether or not the acquire was suppressed.
*/

#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: rf_dagfuncs.c,v 1.35 2021/08/07 16:19:15 thorpej Exp $");

#include <sys/param.h>
#include <sys/ioctl.h>

#include "rf_archs.h"
#include "rf_raid.h"
#include "rf_dag.h"
#include "rf_layout.h"
#include "rf_etimer.h"
#include "rf_acctrace.h"
#include "rf_diskqueue.h"
#include "rf_dagfuncs.h"
#include "rf_general.h"
#include "rf_engine.h"
#include "rf_dagutils.h"

#include "rf_kintf.h"

#if RF_INCLUDE_PARITYLOGGING > 0
#include "rf_paritylog.h"
#endif                          /* RF_INCLUDE_PARITYLOGGING > 0 */

void     (*rf_DiskReadFunc) (RF_DagNode_t *);
void     (*rf_DiskWriteFunc) (RF_DagNode_t *);
void     (*rf_DiskReadUndoFunc) (RF_DagNode_t *);
void     (*rf_DiskWriteUndoFunc) (RF_DagNode_t *);
void     (*rf_RegularXorUndoFunc) (RF_DagNode_t *);
void     (*rf_SimpleXorUndoFunc) (RF_DagNode_t *);
void     (*rf_RecoveryXorUndoFunc) (RF_DagNode_t *);

/*****************************************************************************
* main (only) configuration routine for this module
****************************************************************************/
int
rf_ConfigureDAGFuncs(RF_ShutdownList_t **listp)
{
       RF_ASSERT(((sizeof(long) == 8) && RF_LONGSHIFT == 3) ||
                 ((sizeof(long) == 4) && RF_LONGSHIFT == 2));
       rf_DiskReadFunc = rf_DiskReadFuncForThreads;
       rf_DiskReadUndoFunc = rf_DiskUndoFunc;
       rf_DiskWriteFunc = rf_DiskWriteFuncForThreads;
       rf_DiskWriteUndoFunc = rf_DiskUndoFunc;
       rf_RegularXorUndoFunc = rf_NullNodeUndoFunc;
       rf_SimpleXorUndoFunc = rf_NullNodeUndoFunc;
       rf_RecoveryXorUndoFunc = rf_NullNodeUndoFunc;
       return (0);
}



/*****************************************************************************
* the execution function associated with a terminate node
****************************************************************************/
void
rf_TerminateFunc(RF_DagNode_t *node)
{
       RF_ASSERT(node->dagHdr->numCommits == node->dagHdr->numCommitNodes);
       node->status = rf_good;
       rf_FinishNode(node, RF_THREAD_CONTEXT);
}

void
rf_TerminateUndoFunc(RF_DagNode_t *node)
{
}


/*****************************************************************************
* execution functions associated with a mirror node
*
* parameters:
*
* 0 - physical disk address of data
* 1 - buffer for holding read data
* 2 - parity stripe ID
* 3 - flags
* 4 - physical disk address of mirror (parity)
*
****************************************************************************/

void
rf_DiskReadMirrorIdleFunc(RF_DagNode_t *node)
{
       /* select the mirror copy with the shortest queue and fill in node
        * parameters with physical disk address */

       rf_SelectMirrorDiskIdle(node);
       rf_DiskReadFunc(node);
}

#if (RF_INCLUDE_CHAINDECLUSTER > 0) || (RF_INCLUDE_INTERDECLUSTER > 0) || (RF_DEBUG_VALIDATE_DAG > 0)
void
rf_DiskReadMirrorPartitionFunc(RF_DagNode_t *node)
{
       /* select the mirror copy with the shortest queue and fill in node
        * parameters with physical disk address */

       rf_SelectMirrorDiskPartition(node);
       rf_DiskReadFunc(node);
}
#endif

void
rf_DiskReadMirrorUndoFunc(RF_DagNode_t *node)
{
}



#if RF_INCLUDE_PARITYLOGGING > 0
/*****************************************************************************
* the execution function associated with a parity log update node
****************************************************************************/
void
rf_ParityLogUpdateFunc(RF_DagNode_t *node)
{
       RF_PhysDiskAddr_t *pda = (RF_PhysDiskAddr_t *) node->params[0].p;
       void *bf = (void *) node->params[1].p;
       RF_ParityLogData_t *logData;
#if RF_ACC_TRACE > 0
       RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec;
       RF_Etimer_t timer;
#endif

       if (node->dagHdr->status == rf_enable) {
#if RF_ACC_TRACE > 0
               RF_ETIMER_START(timer);
#endif
               logData = rf_CreateParityLogData(RF_UPDATE, pda, bf,
                   (RF_Raid_t *) (node->dagHdr->raidPtr),
                   node->wakeFunc, node,
                   node->dagHdr->tracerec, timer);
               if (logData)
                       rf_ParityLogAppend(logData, RF_FALSE, NULL, RF_FALSE);
               else {
#if RF_ACC_TRACE > 0
                       RF_ETIMER_STOP(timer);
                       RF_ETIMER_EVAL(timer);
                       tracerec->plog_us += RF_ETIMER_VAL_US(timer);
#endif
                       (node->wakeFunc) (node, ENOMEM);
               }
       }
}


/*****************************************************************************
* the execution function associated with a parity log overwrite node
****************************************************************************/
void
rf_ParityLogOverwriteFunc(RF_DagNode_t *node)
{
       RF_PhysDiskAddr_t *pda = (RF_PhysDiskAddr_t *) node->params[0].p;
       void *bf = (void *) node->params[1].p;
       RF_ParityLogData_t *logData;
#if RF_ACC_TRACE > 0
       RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec;
       RF_Etimer_t timer;
#endif

       if (node->dagHdr->status == rf_enable) {
#if RF_ACC_TRACE > 0
               RF_ETIMER_START(timer);
#endif
               logData = rf_CreateParityLogData(RF_OVERWRITE, pda, bf,
(RF_Raid_t *) (node->dagHdr->raidPtr),
                   node->wakeFunc, node, node->dagHdr->tracerec, timer);
               if (logData)
                       rf_ParityLogAppend(logData, RF_FALSE, NULL, RF_FALSE);
               else {
#if RF_ACC_TRACE > 0
                       RF_ETIMER_STOP(timer);
                       RF_ETIMER_EVAL(timer);
                       tracerec->plog_us += RF_ETIMER_VAL_US(timer);
#endif
                       (node->wakeFunc) (node, ENOMEM);
               }
       }
}

void
rf_ParityLogUpdateUndoFunc(RF_DagNode_t *node)
{
}

void
rf_ParityLogOverwriteUndoFunc(RF_DagNode_t *node)
{
}
#endif                          /* RF_INCLUDE_PARITYLOGGING > 0 */

/*****************************************************************************
* the execution function associated with a NOP node
****************************************************************************/
void
rf_NullNodeFunc(RF_DagNode_t *node)
{
       node->status = rf_good;
       rf_FinishNode(node, RF_THREAD_CONTEXT);
}

void
rf_NullNodeUndoFunc(RF_DagNode_t *node)
{
       node->status = rf_undone;
       rf_FinishNode(node, RF_THREAD_CONTEXT);
}


/*****************************************************************************
* the execution function associated with a disk-read node
****************************************************************************/
void
rf_DiskReadFuncForThreads(RF_DagNode_t *node)
{
       RF_DiskQueueData_t *req;
       RF_PhysDiskAddr_t *pda = (RF_PhysDiskAddr_t *) node->params[0].p;
       void *bf = (void *) node->params[1].p;
       RF_StripeNum_t parityStripeID = (RF_StripeNum_t) node->params[2].v;
       unsigned priority = RF_EXTRACT_PRIORITY(node->params[3].v);
       unsigned which_ru = RF_EXTRACT_RU(node->params[3].v);
       RF_IoType_t iotype = (node->dagHdr->status == rf_enable) ? RF_IO_TYPE_READ : RF_IO_TYPE_NOP;
       RF_DiskQueue_t *dqs = ((RF_Raid_t *) (node->dagHdr->raidPtr))->Queues;

       req = rf_CreateDiskQueueData(iotype, pda->startSector, pda->numSector,
           bf, parityStripeID, which_ru, node->wakeFunc, node,
#if RF_ACC_TRACE > 0
            node->dagHdr->tracerec,
#else
            NULL,
#endif
           (void *) (node->dagHdr->raidPtr), 0, node->dagHdr->bp);

       node->dagFuncData = (void *) req;
       rf_DiskIOEnqueue(&(dqs[pda->col]), req, priority);
}


/*****************************************************************************
* the execution function associated with a disk-write node
****************************************************************************/
void
rf_DiskWriteFuncForThreads(RF_DagNode_t *node)
{
       RF_DiskQueueData_t *req;
       RF_PhysDiskAddr_t *pda = (RF_PhysDiskAddr_t *) node->params[0].p;
       void *bf = (void *) node->params[1].p;
       RF_StripeNum_t parityStripeID = (RF_StripeNum_t) node->params[2].v;
       unsigned priority = RF_EXTRACT_PRIORITY(node->params[3].v);
       unsigned which_ru = RF_EXTRACT_RU(node->params[3].v);
       RF_IoType_t iotype = (node->dagHdr->status == rf_enable) ? RF_IO_TYPE_WRITE : RF_IO_TYPE_NOP;
       RF_DiskQueue_t *dqs = ((RF_Raid_t *) (node->dagHdr->raidPtr))->Queues;

       /* normal processing (rollaway or forward recovery) begins here */
       req = rf_CreateDiskQueueData(iotype, pda->startSector, pda->numSector,
           bf, parityStripeID, which_ru, node->wakeFunc, node,
#if RF_ACC_TRACE > 0
           node->dagHdr->tracerec,
#else
           NULL,
#endif
           (void *) (node->dagHdr->raidPtr),
           0, node->dagHdr->bp);

       node->dagFuncData = (void *) req;
       rf_DiskIOEnqueue(&(dqs[pda->col]), req, priority);
}
/*****************************************************************************
* the undo function for disk nodes
* Note:  this is not a proper undo of a write node, only locks are released.
*        old data is not restored to disk!
****************************************************************************/
void
rf_DiskUndoFunc(RF_DagNode_t *node)
{
       RF_DiskQueueData_t *req;
       RF_PhysDiskAddr_t *pda = (RF_PhysDiskAddr_t *) node->params[0].p;
       RF_DiskQueue_t *dqs = ((RF_Raid_t *) (node->dagHdr->raidPtr))->Queues;

       req = rf_CreateDiskQueueData(RF_IO_TYPE_NOP,
           0L, 0, NULL, 0L, 0, node->wakeFunc, node,
#if RF_ACC_TRACE > 0
            node->dagHdr->tracerec,
#else
            NULL,
#endif
           (void *) (node->dagHdr->raidPtr),
           0, NULL);

       node->dagFuncData = (void *) req;
       rf_DiskIOEnqueue(&(dqs[pda->col]), req, RF_IO_NORMAL_PRIORITY);
}

/*****************************************************************************
* Callback routine for DiskRead and DiskWrite nodes.  When the disk
* op completes, the routine is called to set the node status and
* inform the execution engine that the node has fired.
****************************************************************************/
void
rf_GenericWakeupFunc(void *v, int status)
{
       RF_DagNode_t *node = v;

       switch (node->status) {
       case rf_fired:
               if (status)
                       node->status = rf_bad;
               else
                       node->status = rf_good;
               break;
       case rf_recover:
               /* probably should never reach this case */
               if (status)
                       node->status = rf_panic;
               else
                       node->status = rf_undone;
               break;
       default:
               printf("rf_GenericWakeupFunc:");
               printf("node->status is %d,", node->status);
               printf("status is %d \n", status);
               RF_PANIC();
               break;
       }
       if (node->dagFuncData)
               rf_FreeDiskQueueData((RF_DiskQueueData_t *) node->dagFuncData);
       rf_FinishNode(node, RF_INTR_CONTEXT);
}


/*****************************************************************************
* there are three distinct types of xor nodes:

* A "regular xor" is used in the fault-free case where the access
* spans a complete stripe unit.  It assumes that the result buffer is
* one full stripe unit in size, and uses the stripe-unit-offset
* values that it computes from the PDAs to determine where within the
* stripe unit to XOR each argument buffer.
*
* A "simple xor" is used in the fault-free case where the access
* touches only a portion of one (or two, in some cases) stripe
* unit(s).  It assumes that all the argument buffers are of the same
* size and have the same stripe unit offset.
*
* A "recovery xor" is used in the degraded-mode case.  It's similar
* to the regular xor function except that it takes the failed PDA as
* an additional parameter, and uses it to determine what portions of
* the argument buffers need to be xor'd into the result buffer, and
* where in the result buffer they should go.
****************************************************************************/

/* xor the params together and store the result in the result field.
* assume the result field points to a buffer that is the size of one
* SU, and use the pda params to determine where within the buffer to
* XOR the input buffers.  */
void
rf_RegularXorFunc(RF_DagNode_t *node)
{
       RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[node->numParams - 1].p;
#if RF_ACC_TRACE > 0
       RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec;
       RF_Etimer_t timer;
#endif
       int     i, retcode;

       retcode = 0;
       if (node->dagHdr->status == rf_enable) {
               /* don't do the XOR if the input is the same as the output */
#if RF_ACC_TRACE > 0
               RF_ETIMER_START(timer);
#endif
               for (i = 0; i < node->numParams - 1; i += 2)
                       if (node->params[i + 1].p != node->results[0]) {
                               retcode = rf_XorIntoBuffer(raidPtr, (RF_PhysDiskAddr_t *) node->params[i].p,
                                                          (char *) node->params[i + 1].p, (char *) node->results[0]);
                       }
#if RF_ACC_TRACE > 0
               RF_ETIMER_STOP(timer);
               RF_ETIMER_EVAL(timer);
               tracerec->xor_us += RF_ETIMER_VAL_US(timer);
#endif
       }
       rf_GenericWakeupFunc(node, retcode);    /* call wake func
                                                * explicitly since no
                                                * I/O in this node */
}
/* xor the inputs into the result buffer, ignoring placement issues */
void
rf_SimpleXorFunc(RF_DagNode_t *node)
{
       RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[node->numParams - 1].p;
       int     i, retcode = 0;
#if RF_ACC_TRACE > 0
       RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec;
       RF_Etimer_t timer;
#endif

       if (node->dagHdr->status == rf_enable) {
#if RF_ACC_TRACE > 0
               RF_ETIMER_START(timer);
#endif
               /* don't do the XOR if the input is the same as the output */
               for (i = 0; i < node->numParams - 1; i += 2)
                       if (node->params[i + 1].p != node->results[0]) {
                               retcode = rf_bxor((char *) node->params[i + 1].p, (char *) node->results[0],
                                   rf_RaidAddressToByte(raidPtr, ((RF_PhysDiskAddr_t *) node->params[i].p)->numSector));
                       }
#if RF_ACC_TRACE > 0
               RF_ETIMER_STOP(timer);
               RF_ETIMER_EVAL(timer);
               tracerec->xor_us += RF_ETIMER_VAL_US(timer);
#endif
       }
       rf_GenericWakeupFunc(node, retcode);    /* call wake func
                                                * explicitly since no
                                                * I/O in this node */
}
/* this xor is used by the degraded-mode dag functions to recover lost
* data.  the second-to-last parameter is the PDA for the failed
* portion of the access.  the code here looks at this PDA and assumes
* that the xor target buffer is equal in size to the number of
* sectors in the failed PDA.  It then uses the other PDAs in the
* parameter list to determine where within the target buffer the
* corresponding data should be xored.  */
void
rf_RecoveryXorFunc(RF_DagNode_t *node)
{
       RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[node->numParams - 1].p;
       RF_RaidLayout_t *layoutPtr = (RF_RaidLayout_t *) & raidPtr->Layout;
       RF_PhysDiskAddr_t *failedPDA = (RF_PhysDiskAddr_t *) node->params[node->numParams - 2].p;
       int     i, retcode = 0;
       RF_PhysDiskAddr_t *pda;
       int     suoffset, failedSUOffset = rf_StripeUnitOffset(layoutPtr, failedPDA->startSector);
       char   *srcbuf, *destbuf;
#if RF_ACC_TRACE > 0
       RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec;
       RF_Etimer_t timer;
#endif

       if (node->dagHdr->status == rf_enable) {
#if RF_ACC_TRACE > 0
               RF_ETIMER_START(timer);
#endif
               for (i = 0; i < node->numParams - 2; i += 2)
                       if (node->params[i + 1].p != node->results[0]) {
                               pda = (RF_PhysDiskAddr_t *) node->params[i].p;
                               srcbuf = (char *) node->params[i + 1].p;
                               suoffset = rf_StripeUnitOffset(layoutPtr, pda->startSector);
                               destbuf = ((char *) node->results[0]) + rf_RaidAddressToByte(raidPtr, suoffset - failedSUOffset);
                               retcode = rf_bxor(srcbuf, destbuf, rf_RaidAddressToByte(raidPtr, pda->numSector));
                       }
#if RF_ACC_TRACE > 0
               RF_ETIMER_STOP(timer);
               RF_ETIMER_EVAL(timer);
               tracerec->xor_us += RF_ETIMER_VAL_US(timer);
#endif
       }
       rf_GenericWakeupFunc(node, retcode);
}
/*****************************************************************************
* The next three functions are utilities used by the above
* xor-execution functions.
****************************************************************************/


/*
* this is just a glorified buffer xor.  targbuf points to a buffer
* that is one full stripe unit in size.  srcbuf points to a buffer
* that may be less than 1 SU, but never more.  When the access
* described by pda is one SU in size (which by implication means it's
* SU-aligned), all that happens is (targbuf) <- (srcbuf ^ targbuf).
* When the access is less than one SU in size the XOR occurs on only
* the portion of targbuf identified in the pda.  */

int
rf_XorIntoBuffer(RF_Raid_t *raidPtr, RF_PhysDiskAddr_t *pda,
                char *srcbuf, char *targbuf)
{
       char   *targptr;
       int     sectPerSU = raidPtr->Layout.sectorsPerStripeUnit;
       int     SUOffset = pda->startSector % sectPerSU;
       int     length, retcode = 0;

       RF_ASSERT(pda->numSector <= sectPerSU);

       targptr = targbuf + rf_RaidAddressToByte(raidPtr, SUOffset);
       length = rf_RaidAddressToByte(raidPtr, pda->numSector);
       retcode = rf_bxor(srcbuf, targptr, length);
       return (retcode);
}
/* it really should be the case that the buffer pointers (returned by
* malloc) are aligned to the natural word size of the machine, so
* this is the only case we optimize for.  The length should always be
* a multiple of the sector size, so there should be no problem with
* leftover bytes at the end.  */
int
rf_bxor(char *src, char *dest, int len)
{
       unsigned mask = sizeof(long) - 1, retcode = 0;

       if (!(((unsigned long) src) & mask) &&
           !(((unsigned long) dest) & mask) && !(len & mask)) {
               retcode = rf_longword_bxor((unsigned long *) src,
                                          (unsigned long *) dest,
                                          len >> RF_LONGSHIFT);
       } else {
               RF_ASSERT(0);
       }
       return (retcode);
}

/* When XORing in kernel mode, we need to map each user page to kernel
* space before we can access it.  We don't want to assume anything
* about which input buffers are in kernel/user space, nor about their
* alignment, so in each loop we compute the maximum number of bytes
* that we can xor without crossing any page boundaries, and do only
* this many bytes before the next remap.
*
* len - is in longwords
*/
int
rf_longword_bxor(unsigned long *src, unsigned long *dest, int len)
{
       unsigned long *end = src + len;
       unsigned long d0, d1, d2, d3, s0, s1, s2, s3;   /* temps */
       unsigned long *pg_src, *pg_dest;   /* per-page source/dest pointers */
       int     longs_this_time;/* # longwords to xor in the current iteration */

       pg_src = src;
       pg_dest = dest;
       if (!pg_src || !pg_dest)
               return (EFAULT);

       while (len >= 4) {
               longs_this_time = RF_MIN(len, RF_MIN(RF_BLIP(pg_src), RF_BLIP(pg_dest)) >> RF_LONGSHIFT);       /* note len in longwords */
               src += longs_this_time;
               dest += longs_this_time;
               len -= longs_this_time;
               while (longs_this_time >= 4) {
                       d0 = pg_dest[0];
                       d1 = pg_dest[1];
                       d2 = pg_dest[2];
                       d3 = pg_dest[3];
                       s0 = pg_src[0];
                       s1 = pg_src[1];
                       s2 = pg_src[2];
                       s3 = pg_src[3];
                       pg_dest[0] = d0 ^ s0;
                       pg_dest[1] = d1 ^ s1;
                       pg_dest[2] = d2 ^ s2;
                       pg_dest[3] = d3 ^ s3;
                       pg_src += 4;
                       pg_dest += 4;
                       longs_this_time -= 4;
               }
               while (longs_this_time > 0) {   /* cannot cross any page
                                                * boundaries here */
                       *pg_dest++ ^= *pg_src++;
                       longs_this_time--;
               }

               /* either we're done, or we've reached a page boundary on one
                * (or possibly both) of the pointers */
               if (len) {
                       if (RF_PAGE_ALIGNED(src))
                               pg_src = src;
                       if (RF_PAGE_ALIGNED(dest))
                               pg_dest = dest;
                       if (!pg_src || !pg_dest)
                               return (EFAULT);
               }
       }
       while (src < end) {
               *pg_dest++ ^= *pg_src++;
               src++;
               dest++;
               len--;
               if (RF_PAGE_ALIGNED(src))
                       pg_src = src;
               if (RF_PAGE_ALIGNED(dest))
                       pg_dest = dest;
       }
       RF_ASSERT(len == 0);
       return (0);
}

#if 0
/*
  dst = a ^ b ^ c;
  a may equal dst
  see comment above longword_bxor
  len is length in longwords
*/
int
rf_longword_bxor3(unsigned long *dst, unsigned long *a, unsigned long *b,
                 unsigned long *c, int len, void *bp)
{
       unsigned long a0, a1, a2, a3, b0, b1, b2, b3;
       unsigned long *pg_a, *pg_b, *pg_c, *pg_dst;     /* per-page source/dest
                                                                * pointers */
       int     longs_this_time;/* # longs to xor in the current iteration */
       char    dst_is_a = 0;

       pg_a = a;
       pg_b = b;
       pg_c = c;
       if (a == dst) {
               pg_dst = pg_a;
               dst_is_a = 1;
       } else {
               pg_dst = dst;
       }

       /* align dest to cache line.  Can't cross a pg boundary on dst here. */
       while ((((unsigned long) pg_dst) & 0x1f)) {
               *pg_dst++ = *pg_a++ ^ *pg_b++ ^ *pg_c++;
               dst++;
               a++;
               b++;
               c++;
               if (RF_PAGE_ALIGNED(a)) {
                       pg_a = a;
                       if (!pg_a)
                               return (EFAULT);
               }
               if (RF_PAGE_ALIGNED(b)) {
                       pg_b = a;
                       if (!pg_b)
                               return (EFAULT);
               }
               if (RF_PAGE_ALIGNED(c)) {
                       pg_c = a;
                       if (!pg_c)
                               return (EFAULT);
               }
               len--;
       }

       while (len > 4) {
               longs_this_time = RF_MIN(len, RF_MIN(RF_BLIP(a), RF_MIN(RF_BLIP(b), RF_MIN(RF_BLIP(c), RF_BLIP(dst)))) >> RF_LONGSHIFT);
               a += longs_this_time;
               b += longs_this_time;
               c += longs_this_time;
               dst += longs_this_time;
               len -= longs_this_time;
               while (longs_this_time >= 4) {
                       a0 = pg_a[0];
                       longs_this_time -= 4;

                       a1 = pg_a[1];
                       a2 = pg_a[2];

                       a3 = pg_a[3];
                       pg_a += 4;

                       b0 = pg_b[0];
                       b1 = pg_b[1];

                       b2 = pg_b[2];
                       b3 = pg_b[3];
                       /* start dual issue */
                       a0 ^= b0;
                       b0 = pg_c[0];

                       pg_b += 4;
                       a1 ^= b1;

                       a2 ^= b2;
                       a3 ^= b3;

                       b1 = pg_c[1];
                       a0 ^= b0;

                       b2 = pg_c[2];
                       a1 ^= b1;

                       b3 = pg_c[3];
                       a2 ^= b2;

                       pg_dst[0] = a0;
                       a3 ^= b3;
                       pg_dst[1] = a1;
                       pg_c += 4;
                       pg_dst[2] = a2;
                       pg_dst[3] = a3;
                       pg_dst += 4;
               }
               while (longs_this_time > 0) {   /* cannot cross any page
                                                * boundaries here */
                       *pg_dst++ = *pg_a++ ^ *pg_b++ ^ *pg_c++;
                       longs_this_time--;
               }

               if (len) {
                       if (RF_PAGE_ALIGNED(a)) {
                               pg_a = a;
                               if (!pg_a)
                                       return (EFAULT);
                               if (dst_is_a)
                                       pg_dst = pg_a;
                       }
                       if (RF_PAGE_ALIGNED(b)) {
                               pg_b = b;
                               if (!pg_b)
                                       return (EFAULT);
                       }
                       if (RF_PAGE_ALIGNED(c)) {
                               pg_c = c;
                               if (!pg_c)
                                       return (EFAULT);
                       }
                       if (!dst_is_a)
                               if (RF_PAGE_ALIGNED(dst)) {
                                       pg_dst = dst;
                                       if (!pg_dst)
                                               return (EFAULT);
                               }
               }
       }
       while (len) {
               *pg_dst++ = *pg_a++ ^ *pg_b++ ^ *pg_c++;
               dst++;
               a++;
               b++;
               c++;
               if (RF_PAGE_ALIGNED(a)) {
                       pg_a = a;
                       if (!pg_a)
                               return (EFAULT);
                       if (dst_is_a)
                               pg_dst = pg_a;
               }
               if (RF_PAGE_ALIGNED(b)) {
                       pg_b = b;
                       if (!pg_b)
                               return (EFAULT);
               }
               if (RF_PAGE_ALIGNED(c)) {
                       pg_c = c;
                       if (!pg_c)
                               return (EFAULT);
               }
               if (!dst_is_a)
                       if (RF_PAGE_ALIGNED(dst)) {
                               pg_dst = dst;
                               if (!pg_dst)
                                       return (EFAULT);
                       }
               len--;
       }
       return (0);
}

int
rf_bxor3(unsigned char *dst, unsigned char *a, unsigned char *b,
        unsigned char *c, unsigned long len, void *bp)
{
       RF_ASSERT(((RF_UL(dst) | RF_UL(a) | RF_UL(b) | RF_UL(c) | len) & 0x7) == 0);

       return (rf_longword_bxor3((unsigned long *) dst, (unsigned long *) a,
               (unsigned long *) b, (unsigned long *) c, len >> RF_LONGSHIFT, bp));
}
#endif