Give physio a makeover.

- Let physio take read/write compatible args and have it use uio->uio_rw
  to determine the direction.

- physread/physwrite are now #defines for physio

- Remove the inversly named minphys(), dev->si_iosize_max takes over.

- Physio() always uses pbufs.

- Fix the check for non page-aligned transfers, now only unaligned
  transfers larger than (MAXPHYS - PAGE_SIZE) get fragmented (only
  interesting for tapes using max blocksize).

- General wash-and-clean of code.

Constructive input from: bde

3 files changed, 60 insertions, 131 deletions

diff --git a/sys/kern/kern_physio.c b/sys/kern/kern_physio.c
index 45ecf87..6a4034e 100644
--- a/sys/kern/kern_physio.c
+++ b/sys/kern/kern_physio.c
@@ -29,175 +29,108 @@
 #include <vm/vm.h>
 #include <vm/vm_extern.h>
 
-static void	physwakeup __P((struct buf *bp));
-
-int
-physread(dev_t dev, struct uio *uio, int ioflag)
-{
-	return(physio(NULL, dev, 1, minphys, uio));
-}
-
-int
-physwrite(dev_t dev, struct uio *uio, int ioflag)
+static void
+physwakeup(struct buf *bp)
 {
-	return(physio(NULL, dev, 0, minphys, uio));
+	wakeup((caddr_t) bp);
+	bp->b_flags &= ~B_CALL;
 }
 
 int
-physio(bp, dev, rw, minp, uio)
-	struct buf *bp;
-	dev_t dev;
-	int rw;
-	u_int (*minp) __P((struct buf *bp));
-	struct uio *uio;
+physio(dev_t dev, struct uio *uio, int ioflag)
 {
 	int i;
-	int bufflags = rw?B_READ:0;
 	int error;
 	int spl;
 	caddr_t sa;
 	off_t blockno;
-	int bp_alloc = (bp == 0);
-	struct buf *bpa;
+	u_int iolen;
+	struct buf *bp;
 
-	/*
-	 * Keep the process UPAGES from being swapped. (XXX for performance?)
-	 */
+	/* Keep the process UPAGES from being swapped. XXX: why ? */
 	PHOLD(curproc);
 
-	/* create and build a buffer header for a transfer */
-	bpa = getpbuf(NULL);
-	if (!bp_alloc)
-		BUF_LOCK(bp, LK_EXCLUSIVE);
-	else
-		bp = bpa;
-
-	/*
-	 * get a copy of the kva from the physical buffer
-	 */
-	sa = bpa->b_data;
+	bp = getpbuf(NULL);
+	sa = bp->b_data;
 	error = bp->b_error = 0;
 
+	/* XXX: sanity check */
+	if(dev->si_iosize_max < PAGE_SIZE) {
+		printf("WARNING: %s si_iosize_max=%d, using DFLTPHYS.\n",
+		    devtoname(dev), dev->si_iosize_max);
+		dev->si_iosize_max = DFLTPHYS;
+	}
+
 	for (i = 0; i < uio->uio_iovcnt; i++) {
 		while (uio->uio_iov[i].iov_len) {
-
+			if (uio->uio_rw == UIO_READ)
+				bp->b_flags = B_PHYS | B_CALL | B_READ;
+			else 
+				bp->b_flags = B_PHYS | B_CALL | B_WRITE;
 			bp->b_dev = dev;
-			bp->b_bcount = uio->uio_iov[i].iov_len;
-			bp->b_flags = B_PHYS | B_CALL | bufflags;
 			bp->b_iodone = physwakeup;
 			bp->b_data = uio->uio_iov[i].iov_base;
-			bp->b_bcount = minp( bp);
-			if( minp != minphys)
-				bp->b_bcount = minphys( bp);
-			bp->b_bufsize = bp->b_bcount;
-			/*
-			 * pass in the kva from the physical buffer
-			 * for the temporary kernel mapping.
-			 */
+			bp->b_bcount = uio->uio_iov[i].iov_len;
+			bp->b_offset = uio->uio_offset;
 			bp->b_saveaddr = sa;
-			blockno = uio->uio_offset >> DEV_BSHIFT;
+
+			/* Don't exceed drivers iosize limit */
+			if (bp->b_bcount > dev->si_iosize_max)
+				bp->b_bcount = dev->si_iosize_max;
+
+			/* 
+			 * Make sure the pbuf can map the request
+			 * XXX: The pbuf has kvasize = MAXPHYS so a request
+			 * XXX: larger than MAXPHYS - PAGE_SIZE must be
+			 * XXX: page aligned or it will be fragmented.
+			 */
+			iolen = ((vm_offset_t) bp->b_data) & PAGE_MASK;
+			if ((bp->b_bcount + iolen) > bp->b_kvasize) {
+				bp->b_bcount = bp->b_kvasize;
+				if (iolen != 0)
+					bp->b_bcount -= PAGE_SIZE;
+			}
+			bp->b_bufsize = bp->b_bcount;
+
+			blockno = bp->b_offset >> DEV_BSHIFT;
 			if ((daddr_t)blockno != blockno) {
-				error = EINVAL;
+				error = EINVAL; /* blockno overflow */
 				goto doerror;
 			}
 			bp->b_blkno = blockno;
-			bp->b_offset = uio->uio_offset;
 
 			if (uio->uio_segflg == UIO_USERSPACE) {
-				if (rw && !useracc(bp->b_data, bp->b_bufsize, B_WRITE)) {
-					error = EFAULT;
-					goto doerror;
-				}
-				if (!rw && !useracc(bp->b_data, bp->b_bufsize, B_READ)) {
+				if (!useracc(bp->b_data, bp->b_bufsize,
+				    bp->b_flags & B_READ)) {
 					error = EFAULT;
 					goto doerror;
 				}
-
-				/* bring buffer into kernel space */
 				vmapbuf(bp);
 			}
 
-			/* perform transfer */
 			BUF_STRATEGY(bp, 0);
-
 			spl = splbio();
 			while ((bp->b_flags & B_DONE) == 0)
 				tsleep((caddr_t)bp, PRIBIO, "physstr", 0);
 			splx(spl);
 
-			/* release mapping into kernel space */
 			if (uio->uio_segflg == UIO_USERSPACE)
 				vunmapbuf(bp);
-
-			/*
-			 * update the uio data
-			 */
-			{
-				int iolen = bp->b_bcount - bp->b_resid;
-
-				if (iolen == 0 && !(bp->b_flags & B_ERROR))
-					goto doerror;	/* EOF */
-				uio->uio_iov[i].iov_len -= iolen;
-				uio->uio_iov[i].iov_base += iolen;
-				uio->uio_resid -= iolen;
-				uio->uio_offset += iolen;
-			}
-
-			/*
-			 * check for an error
-			 */
+			iolen = bp->b_bcount - bp->b_resid;
+			if (iolen == 0 && !(bp->b_flags & B_ERROR))
+				goto doerror;	/* EOF */
+			uio->uio_iov[i].iov_len -= iolen;
+			uio->uio_iov[i].iov_base += iolen;
+			uio->uio_resid -= iolen;
+			uio->uio_offset += iolen;
 			if( bp->b_flags & B_ERROR) {
 				error = bp->b_error;
 				goto doerror;
 			}
 		}
 	}
-
-
 doerror:
-	relpbuf(bpa, NULL);
-	if (!bp_alloc) {
-		bp->b_flags &= ~B_PHYS;
-		BUF_UNLOCK(bp);
-	}
-	/*
-	 * Allow the process UPAGES to be swapped again.
-	 */
+	relpbuf(bp, NULL);
 	PRELE(curproc);
-
 	return (error);
 }
-
-u_int
-minphys(bp)
-	struct buf *bp;
-{
-	u_int maxphys;
-
-	maxphys = bp->b_dev->si_iosize_max;
-	if(!maxphys) {
-		printf("WARNING: %s maxphys = 0 ??\n", devtoname(bp->b_dev));
-		maxphys = DFLTPHYS;
-		bp->b_dev->si_iosize_max = DFLTPHYS;
-	}
-	if (bp->b_kvasize && (bp->b_kvasize < maxphys))
-		maxphys = bp->b_kvasize;
-
-	if(((vm_offset_t) bp->b_data) & PAGE_MASK) {
-		maxphys -= PAGE_SIZE;
-	}
-
-	if( bp->b_bcount > maxphys) {
-		bp->b_bcount = maxphys;
-	}
-
-	return bp->b_bcount;
-}
-
-static void
-physwakeup(bp)
-	struct buf *bp;
-{
-	wakeup((caddr_t) bp);
-	bp->b_flags &= ~B_CALL;
-}
diff --git a/sys/sys/bio.h b/sys/sys/bio.h
index ae4dbfd..05e9863 100644
--- a/sys/sys/bio.h
+++ b/sys/sys/bio.h
@@ -480,11 +480,9 @@ int	cluster_read __P((struct vnode *, u_quad_t, daddr_t, long,
 	    struct ucred *, long, int, struct buf **));
 int	cluster_wbuild __P((struct vnode *, long, daddr_t, int));
 void	cluster_write __P((struct buf *, u_quad_t));
-int	physio __P((struct buf *, dev_t, int, u_int (*)(struct buf *),
-	    struct uio *));
-int	physread __P((dev_t dev, struct uio *uio, int ioflag));
-int	physwrite __P((dev_t dev, struct uio *uio, int ioflag));
-u_int	minphys __P((struct buf *));
+int	physio __P((dev_t dev, struct uio *uio, int ioflag));
+#define physread physio
+#define physwrite physio
 void	vfs_bio_set_validclean __P((struct buf *, int base, int size));
 void	vfs_bio_clrbuf __P((struct buf *));
 void	vfs_busy_pages __P((struct buf *, int clear_modify));
diff --git a/sys/sys/buf.h b/sys/sys/buf.h
index ae4dbfd..05e9863 100644
--- a/sys/sys/buf.h
+++ b/sys/sys/buf.h
@@ -480,11 +480,9 @@ int	cluster_read __P((struct vnode *, u_quad_t, daddr_t, long,
 	    struct ucred *, long, int, struct buf **));
 int	cluster_wbuild __P((struct vnode *, long, daddr_t, int));
 void	cluster_write __P((struct buf *, u_quad_t));
-int	physio __P((struct buf *, dev_t, int, u_int (*)(struct buf *),
-	    struct uio *));
-int	physread __P((dev_t dev, struct uio *uio, int ioflag));
-int	physwrite __P((dev_t dev, struct uio *uio, int ioflag));
-u_int	minphys __P((struct buf *));
+int	physio __P((dev_t dev, struct uio *uio, int ioflag));
+#define physread physio
+#define physwrite physio
 void	vfs_bio_set_validclean __P((struct buf *, int base, int size));
 void	vfs_bio_clrbuf __P((struct buf *));
 void	vfs_busy_pages __P((struct buf *, int clear_modify));