Aren
/
NetBSD
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

- Add a B_ORDERED flag to communicate to drivers that an I/O request should 

be issued/completed in order; that is, provide a barrier for I/O queues.
- Change the buffer driver queue links to a TAILQ, rather than using
  a home-grown equivalent.  Provide BUFQ_*() macros to manipulate buffer
  queues; these deal with the barrier provided by B_ORDERED.
- Update disksort() accordingly, and provide 3 versions:
	- disksort_cylinder(): historical disksort(), which keys on
	  b_cylinder (and b_blkno for the case when b_cylinder matches).
	- disksort_blkno(): sorts only on b_blkno.  Essentially the
	  same as disksort_cylinder(), but with fewer comparisons.
	- disksort_tail(): requests are simply inserted into the queue
	  at the tail.  This is provided as an option so that drivers
	  can simply have a pointer to the appropriate sort function.
  Note that disksort() now pays attention to B_ORDERED.
This commit is contained in:
thorpej 2000-01-21 23:20:51 +00:00
parent 9c658d828e
commit 05c6b5a9a0
2 changed files with 245 additions and 44 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

189
sys/kern/subr_disk.c
View File

@ -1,7 +1,7 @@
/* $NetBSD: subr_disk.c,v 1.25 1999/02/22 16:00:01 drochner Exp $ */
/* $NetBSD: subr_disk.c,v 1.26 2000/01/21 23:20:51 thorpej Exp $ */
/*-
* Copyright (c) 1996, 1997 The NetBSD Foundation, Inc.
* Copyright (c) 1996, 1997, 1999, 2000 The NetBSD Foundation, Inc.
* All rights reserved.
*
* This code is derived from software contributed to The NetBSD Foundation
@ -98,28 +98,40 @@ int disk_count; /* number of drives in global disklist */
* Seek sort for disks. We depend on the driver which calls us using b_resid
* as the current cylinder number.
*
* The argument ap structure holds a b_actf activity chain pointer on which we
* keep two queues, sorted in ascending cylinder order. The first queue holds
* those requests which are positioned after the current cylinder (in the first
* request); the second holds requests which came in after their cylinder number
* was passed. Thus we implement a one way scan, retracting after reaching the
* end of the drive to the first request on the second queue, at which time it
* becomes the first queue.
* The argument bufq is an I/O queue for the device, on which there are
* actually two queues, sorted in ascending cylinder order. The first
* queue holds those requests which are positioned after the current
* cylinder (in the first request); the second holds requests which came
* in after their cylinder number was passed. Thus we implement a one-way
* scan, retracting after reaching the end of the drive to the first request
* on the second queue, at which time it becomes the first queue.
*
* A one-way scan is natural because of the way UNIX read-ahead blocks are
* allocated.
*
* This is further adjusted by any `barriers' which may exist in the queue.
* The bufq points to the last such ordered request.
*/
void
disksort(ap, bp)
register struct buf *ap, *bp;
disksort_cylinder(bufq, bp)
struct buf_queue *bufq;
struct buf *bp;
{
register struct buf *bq;
struct buf *bq, *nbq;
/* If the queue is empty, then it's easy. */
if (ap->b_actf == NULL) {
bp->b_actf = NULL;
ap->b_actf = bp;
/*
* If there are ordered requests on the queue, we must start
* the elevator sort after the last of these.
*/
if ((bq = bufq->bq_barrier) == NULL)
bq = BUFQ_FIRST(bufq);
/*
* If the queue is empty, of if it's an ordered request,
* it's easy; we just go on the end.
*/
if (bq == NULL || (bp->b_flags & B_ORDERED) != 0) {
BUFQ_INSERT_TAIL(bufq, bp);
return;
}
@ -127,15 +139,14 @@ disksort(ap, bp)
* If we lie after the first (currently active) request, then we
* must locate the second request list and add ourselves to it.
*/
bq = ap->b_actf;
if (bp->b_cylinder < bq->b_cylinder) {
while (bq->b_actf) {
while ((nbq = BUFQ_NEXT(bq)) != NULL) {
/*
* Check for an ``inversion'' in the normally ascending
* cylinder numbers, indicating the start of the second
* request list.
*/
if (bq->b_actf->b_cylinder < bq->b_cylinder) {
if (nbq->b_cylinder < bq->b_cylinder) {
/*
* Search the second request list for the first
* request at a larger cylinder number. We go
@ -143,18 +154,16 @@ disksort(ap, bp)
* go at end.
*/
do {
if (bp->b_cylinder <
bq->b_actf->b_cylinder)
if (bp->b_cylinder < nbq->b_cylinder)
goto insert;
if (bp->b_cylinder ==
bq->b_actf->b_cylinder &&
bp->b_blkno < bq->b_actf->b_blkno)
if (bp->b_cylinder == nbq->b_cylinder &&
bp->b_blkno < nbq->b_blkno)
goto insert;
bq = bq->b_actf;
} while (bq->b_actf);
bq = nbq;
} while ((nbq = BUFQ_NEXT(bq)) != NULL);
goto insert; /* after last */
}
bq = bq->b_actf;
bq = BUFQ_NEXT(bq);
}
/*
* No inversions... we will go after the last, and
@ -166,26 +175,134 @@ disksort(ap, bp)
* Request is at/after the current request...
* sort in the first request list.
*/
while (bq->b_actf) {
while ((nbq = BUFQ_NEXT(bq)) != NULL) {
/*
* We want to go after the current request if there is an
* inversion after it (i.e. it is the end of the first
* request list), or if the next request is a larger cylinder
* than our request.
*/
if (bq->b_actf->b_cylinder < bq->b_cylinder ||
bp->b_cylinder < bq->b_actf->b_cylinder ||
(bp->b_cylinder == bq->b_actf->b_cylinder &&
bp->b_blkno < bq->b_actf->b_blkno))
if (nbq->b_cylinder < bq->b_cylinder ||
bp->b_cylinder < nbq->b_cylinder ||
(bp->b_cylinder == nbq->b_cylinder &&
bp->b_blkno < nbq->b_blkno))
goto insert;
bq = bq->b_actf;
bq = nbq;
}
/*
* Neither a second list nor a larger request... we go at the end of
* the first list, which is the same as the end of the whole schebang.
*/
insert: bp->b_actf = bq->b_actf;
bq->b_actf = bp;
insert: BUFQ_INSERT_AFTER(bufq, bq, bp);
}
/*
* Seek sort for disks. This version sorts based on b_blkno, which
* indicates the block number.
*
* As before, there are actually two queues, sorted in ascendening block
* order. The first queue holds those requests which are positioned after
* the current block (in the first request); the second holds requests which
* came in after their block number was passed. Thus we implement a one-way
* scan, retracting after reaching the end of the driver to the first request
* on the second queue, at which time it becomes the first queue.
*
* A one-way scan is natural because of the way UNIX read-ahead blocks are
* allocated.
*
* This is further adjusted by any `barriers' which may exist in the queue.
* The bufq points to the last such ordered request.
*/
void
disksort_blkno(bufq, bp)
struct buf_queue *bufq;
struct buf *bp;
{
struct buf *bq, *nbq;
/*
* If there are ordered requests on the queue, we must start
* the elevator sort after the last of these.
*/
if ((bq = bufq->bq_barrier) == NULL)
bq = BUFQ_FIRST(bufq);
/*
* If the queue is empty, or if it's an ordered request,
* it's easy; we just go on the end.
*/
if (bq == NULL || (bp->b_flags & B_ORDERED) != 0) {
BUFQ_INSERT_TAIL(bufq, bp);
return;
}
/*
* If we lie after the first (currently active) request, then we
* must locate the second request list and add ourselves to it.
*/
if (bp->b_blkno < bq->b_blkno) {
while ((nbq = BUFQ_NEXT(bq)) != NULL) {
/*
* Check for an ``inversion'' in the normally ascending
* block numbers, indicating the start of the second
* request list.
*/
if (nbq->b_blkno < bq->b_blkno) {
/*
* Search the second request list for the first
* request at a larger block number. We go
* after that; if there is no such request, we
* go at the end.
*/
do {
if (bp->b_blkno < nbq->b_blkno)
goto insert;
bq = nbq;
} while ((nbq = BUFQ_NEXT(bq)) != NULL);
goto insert; /* after last */
}
bq = BUFQ_NEXT(bq);
}
/*
* No inversions... we will go after the last, and
* be the first request in the second request list.
*/
goto insert;
}
/*
* Request is at/after the current request...
* sort in the first request list.
*/
while ((nbq = BUFQ_NEXT(bq)) != NULL) {
/*
* We want to go after the current request if there is an
* inversion after it (i.e. it is the end of the first
* request list), or if the next request is a larger cylinder
* than our request.
*/
if (nbq->b_blkno < bq->b_blkno ||
bp->b_blkno < nbq->b_blkno)
goto insert;
bq = nbq;
}
/*
* Neither a second list nor a larger request... we go at the end of
* the first list, which is the same as the end of the whole schebang.
*/
insert: BUFQ_INSERT_AFTER(bufq, bq, bp);
}
/*
* Seek non-sort for disks. This version simply inserts requests at
* the tail of the queue.
*/
void
disksort_tail(bufq, bp)
struct buf_queue *bufq;
struct buf *bp;
{
BUFQ_INSERT_TAIL(bufq, bp);
}
/*

100
sys/sys/buf.h
View File

@ -1,4 +1,41 @@
/* $NetBSD: buf.h,v 1.35 1999/11/15 18:49:12 fvdl Exp $ */
/* $NetBSD: buf.h,v 1.36 2000/01/21 23:20:51 thorpej Exp $ */
/*-
* Copyright (c) 1999, 2000 The NetBSD Foundation, Inc.
* All rights reserved.
*
* This code is derived from software contributed to The NetBSD Foundation
* by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
* NASA Ames Research Center.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by the NetBSD
* Foundation, Inc. and its contributors.
* 4. Neither the name of The NetBSD Foundation nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
* ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
* TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
* BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
/*
* Copyright (c) 1982, 1986, 1989, 1993
@ -51,6 +88,54 @@
* To avoid including <ufs/ffs/softdep.h>
*/
LIST_HEAD(workhead, worklist);
/*
* Device driver buffer queue.
*/
struct buf_queue {
TAILQ_HEAD(bufq_head, buf) bq_head; /* actual list of buffers */
struct buf *bq_barrier; /* last B_ORDERED request */
};
#ifdef _KERNEL
#define BUFQ_FIRST(bufq) TAILQ_FIRST(&(bufq)->bq_head)
#define BUFQ_NEXT(bp) TAILQ_NEXT((bp), b_actq)
#define BUFQ_INIT(bufq) \
do { \
TAILQ_INIT(&(bufq)->bq_head); \
(bufq)->bq_barrier = NULL; \
} while (/*CONSTCOND*/0)
#define BUFQ_INSERT_TAIL(bufq, bp) \
do { \
TAILQ_INSERT_TAIL(&(bufq)->bq_head, (bp), b_actq); \
if (((bp)->b_flags & B_ORDERED) != 0) \
(bufq)->bq_barrier = (bp); \
} while (/*CONSTCOND*/0)
#define BUFQ_INSERT_AFTER(bufq, lbp, bp) \
do { \
KASSERT((bufq)->bq_barrier == NULL); \
KASSERT(((bp)->b_flags & B_ORDERED) == 0); \
TAILQ_INSERT_AFTER(&(bufq)->bq_head, (lbp), (bp), b_actq); \
} while (/*CONSTCOND*/0)
#define BUFQ_INSERT_BEFORE(bufq, lbp, bp) \
do { \
KASSERT((bufq)->bq_barrier == NULL); \
KASSERT(((bp)->b_flags & B_ORDERED) == 0); \
TAILQ_INSERT_BEFORE((lbp), (bp), b_actq); \
} while (/*CONSTCOND*/0)
#define BUFQ_REMOVE(bufq, bp) \
do { \
if ((bufq)->bq_barrier == (bp)) \
(bufq)->bq_barrier = TAILQ_PREV((bp), bufq_head, b_actq); \
TAILQ_REMOVE(&(bufq)->bq_head, (bp), b_actq); \
} while (/*CONSTCOND*/0)
#endif /* _KERNEL */
/*
* These are currently used only by the soft dependency code, hence
* are stored once in a global variable. If other subsystems wanted
@ -75,7 +160,7 @@ struct buf {
LIST_ENTRY(buf) b_hash; /* Hash chain. */
LIST_ENTRY(buf) b_vnbufs; /* Buffer's associated vnode. */
TAILQ_ENTRY(buf) b_freelist; /* Free list position if not active. */
struct buf *b_actf, **b_actb; /* Device driver queue when active. */
TAILQ_ENTRY(buf) b_actq; /* Device driver queue when active. */
struct proc *b_proc; /* Associated proc; NULL if kernel. */
volatile long b_flags; /* B_* flags. */
int b_error; /* Errno value. */
@ -109,11 +194,7 @@ struct buf {
#define b_cylinder b_resid /* Cylinder number for disksort(). */
/* Device driver compatibility definitions. */
#define b_active b_bcount /* Driver queue head: drive active. */
#define b_data b_un.b_addr /* b_un.b_addr is not changeable. */
#define b_errcnt b_resid /* Retry count while I/O in progress. */
#define iodone biodone /* Old name for biodone. */
#define iowait biowait /* Old name for biowait. */
/*
* These flags are kept in b_flags.
@ -134,6 +215,7 @@ struct buf {
#define B_INVAL 0x00002000 /* Does not contain valid info. */
#define B_LOCKED 0x00004000 /* Locked in core (not reusable). */
#define B_NOCACHE 0x00008000 /* Do not cache block after use. */
#define B_ORDERED 0x00010000 /* ordered I/O request */
#define B_PHYS 0x00040000 /* I/O to user memory. */
#define B_RAW 0x00080000 /* Set by physio for raw transfers. */
#define B_READ 0x00100000 /* Read buffer. */
@ -161,10 +243,11 @@ struct cluster_save {
/*
* Zero out the buffer's data area.
*/
#define clrbuf(bp) { \
#define clrbuf(bp) \
do { \
memset((bp)->b_data, 0, (u_int)(bp)->b_bcount); \
(bp)->b_resid = 0; \
}
} while (0)
/* Flags to low-level allocation routines. */
#define B_CLRBUF 0x01 /* Request allocated buffer be cleared. */
@ -180,6 +263,7 @@ extern int nswbuf; /* Number of swap I/O buffer headers. */
__BEGIN_DECLS
void allocbuf __P((struct buf *, int));
void bawrite __P((struct buf *));
void bowrite __P((struct buf *));
void bdirty __P((struct buf *));
void bdwrite __P((struct buf *));
void biodone __P((struct buf *));