Kill DMA_CHUNK_SIZE and related code.

Rename request nr_segments to nr_phys_segments.

14 files changed, 37 insertions, 142 deletions

diff --git a/drivers/block/DAC960.c b/drivers/block/DAC960.c
index 55d3fbe7e..eed7b446c 100644
--- a/drivers/block/DAC960.c
+++ b/drivers/block/DAC960.c
@@ -2889,7 +2889,7 @@ static boolean DAC960_ProcessRequest(DAC960_Controller_T *Controller,
   Command->LogicalDriveNumber = DAC960_LogicalDriveNumber(Request->rq_dev);
   Command->BlockNumber = Request->sector;
   Command->BlockCount = Request->nr_sectors;
-  Command->SegmentCount = Request->nr_segments;
+  Command->SegmentCount = Request->nr_phys_segments;
   Command->BufferHeader = Request->bio;
   Command->RequestBuffer = Request->buffer;
   blkdev_dequeue_request(Request);
diff --git a/drivers/block/cciss.c b/drivers/block/cciss.c
index 371755761..208a785c6 100644
--- a/drivers/block/cciss.c
+++ b/drivers/block/cciss.c
@@ -1219,7 +1219,7 @@ queue:
 		goto startio;
 
 	creq = elv_next_request(q);
-	if (creq->nr_segments > MAXSGENTRIES)
+	if (creq->nr_phys_segments > MAXSGENTRIES)
                 BUG();
 
         if (h->ctlr != MAJOR(creq->rq_dev)-MAJOR_NR )
diff --git a/drivers/block/cpqarray.c b/drivers/block/cpqarray.c
index 4ff77277d..e34f7b47e 100644
--- a/drivers/block/cpqarray.c
+++ b/drivers/block/cpqarray.c
@@ -864,7 +864,7 @@ queue_next:
 		goto startio;
 
 	creq = elv_next_request(q);
-	if (creq->nr_segments > SG_MAX)
+	if (creq->nr_phys_segments > SG_MAX)
 		BUG();
 
 	if (h->ctlr != MAJOR(creq->rq_dev)-MAJOR_NR || h->ctlr > nr_ctlr)
diff --git a/drivers/block/ll_rw_blk.c b/drivers/block/ll_rw_blk.c
index 2eef6065a..dce810ac4 100644
--- a/drivers/block/ll_rw_blk.c
+++ b/drivers/block/ll_rw_blk.c
@@ -373,7 +373,7 @@ inline int blk_contig_segment(request_queue_t *q, struct bio *bio,
 
 /*
  * map a request to scatterlist, return number of sg entries setup. Caller
- * must make sure sg can hold rq->nr_segments entries
+ * must make sure sg can hold rq->nr_phys_segments entries
  */
 int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
 {
@@ -408,7 +408,7 @@ int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg
 			} else {
 new_segment:
 				if (nsegs >= q->max_segments) {
-					printk("map: %d >= %d, i %d, segs %d, size %ld\n", nsegs, q->max_segments, i, rq->nr_segments, rq->nr_sectors);
+					printk("map: %d >= %d, i %d, phys_segs %d, size %ld\n", nsegs, q->max_segments, i, rq->nr_phys_segments, rq->nr_sectors);
 					BUG();
 				}
 
@@ -435,8 +435,8 @@ static inline int ll_new_segment(request_queue_t *q, struct request *req,
 {
 	int nr_segs = bio_hw_segments(q, bio);
 
-	if (req->nr_segments + nr_segs <= q->max_segments) {
-		req->nr_segments += nr_segs;
+	if (req->nr_phys_segments + nr_segs <= q->max_segments) {
+		req->nr_phys_segments += nr_segs;
 		return 1;
 	}
 
@@ -469,7 +469,7 @@ static int ll_front_merge_fn(request_queue_t *q, struct request *req,
 static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
 				struct request *next)
 {
-	int total_segments = req->nr_segments + next->nr_segments;
+	int total_segments = req->nr_phys_segments + next->nr_phys_segments;
 
 	if (blk_contig_segment(q, req->biotail, next->bio))
 		total_segments--;
@@ -477,7 +477,7 @@ static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
 	if (total_segments > q->max_segments)
 		return 0;
 
-	req->nr_segments = total_segments;
+	req->nr_phys_segments = total_segments;
 	return 1;
 }
 
@@ -967,16 +967,6 @@ static int __make_request(request_queue_t *q, struct bio *bio)
 	cur_nr_sectors = bio_iovec(bio)->bv_len >> 9;
 	rw = bio_data_dir(bio);
 
-#if 1 /* XXX Makes no writes actually go to disk.
-       * XXX Use this on kernels you do not trust such
-       * XXX as current 2.5.1-preX :-) -DaveM
-       */
-	if (rw & WRITE) {
-		set_bit(BIO_UPTODATE, &bio->bi_flags);
-		bio->bi_end_io(bio, nr_sectors);
-		return 0;
-	}
-#endif
 	/*
 	 * low level driver can indicate that it wants pages above a
 	 * certain limit bounced to low memory (ie for highmem, or even
@@ -1111,7 +1101,7 @@ get_rq:
 	req->hard_sector = req->sector = sector;
 	req->hard_nr_sectors = req->nr_sectors = nr_sectors;
 	req->current_nr_sectors = req->hard_cur_sectors = cur_nr_sectors;
-	req->nr_segments = bio->bi_vcnt;
+	req->nr_phys_segments = bio->bi_vcnt;
 	req->nr_hw_segments = bio_hw_segments(q, bio);
 	req->buffer = bio_data(bio);	/* see ->buffer comment above */
 	req->waiting = NULL;
diff --git a/drivers/ide/ide-dma.c b/drivers/ide/ide-dma.c
index ecdcd85d5..fb638ca30 100644
--- a/drivers/ide/ide-dma.c
+++ b/drivers/ide/ide-dma.c
@@ -232,8 +232,8 @@ static int ide_build_sglist (ide_hwif_t *hwif, struct request *rq)
 
 	nents = blk_rq_map_sg(q, rq, hwif->sg_table);
 
-	if (rq->q && nents > rq->nr_segments)
-		printk("ide-dma: received %d segments, build %d\n", rq->nr_segments, nents);
+	if (rq->q && nents > rq->nr_phys_segments)
+		printk("ide-dma: received %d phys segments, build %d\n", rq->nr_phys_segments, nents);
 
 	if (rq_data_dir(rq) == READ)
 		hwif->sg_dma_direction = PCI_DMA_FROMDEVICE;
diff --git a/drivers/scsi/scsi.c b/drivers/scsi/scsi.c
index 3713c3284..6faaad0b7 100644
--- a/drivers/scsi/scsi.c
+++ b/drivers/scsi/scsi.c
@@ -186,11 +186,6 @@ void  scsi_initialize_queue(Scsi_Device * SDpnt, struct Scsi_Host * SHpnt)
 	blk_init_queue(q, scsi_request_fn);
 	q->queuedata = (void *) SDpnt;
 
-#ifdef DMA_CHUNK_SIZE
-	if (max_segments > 64)
-		max_segments = 64;
-#endif
-
 	blk_queue_max_segments(q, max_segments);
 	blk_queue_max_sectors(q, SHpnt->max_sectors);
 
diff --git a/drivers/scsi/scsi_dma.c b/drivers/scsi/scsi_dma.c
index 3de835130..76fbc3aaf 100644
--- a/drivers/scsi/scsi_dma.c
+++ b/drivers/scsi/scsi_dma.c
@@ -246,13 +246,8 @@ void scsi_resize_dma_pool(void)
 			 */
 			if (SDpnt->type == TYPE_WORM || SDpnt->type == TYPE_ROM ||
 			    SDpnt->type == TYPE_DISK || SDpnt->type == TYPE_MOD) {
+				/* XXX This needs fixing -DaveM */
 				int nents = host->sg_tablesize;
-#ifdef DMA_CHUNK_SIZE
-				/* If the architecture does DMA sg merging, make sure
-				   we count with at least 64 entries even for HBAs
-				   which handle very few sg entries.  */
-				if (nents < 64) nents = 64;
-#endif
 				new_dma_sectors += ((nents *
 				sizeof(struct scatterlist) + 511) >> 9) *
 				 SDpnt->queue_depth;
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index a723b3404..6e5cae494 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -82,7 +82,7 @@ static void __scsi_insert_special(request_queue_t *q, struct request *rq,
 	rq->special = data;
 	rq->q = NULL;
 	rq->bio = rq->biotail = NULL;
-	rq->nr_segments = 0;
+	rq->nr_phys_segments = 0;
 	rq->elevator_sequence = 0;
 
 	/*
diff --git a/drivers/scsi/scsi_merge.c b/drivers/scsi/scsi_merge.c
index 140933cc2..23995ab3f 100644
--- a/drivers/scsi/scsi_merge.c
+++ b/drivers/scsi/scsi_merge.c
@@ -192,66 +192,16 @@ recount_segments(Scsi_Cmnd * SCpnt)
 	struct request *req = &SCpnt->request;
 	struct Scsi_Host *SHpnt = SCpnt->host;
 
-	req->nr_segments = __count_segments(req, SHpnt->unchecked_isa_dma,NULL);
+	req->nr_phys_segments = __count_segments(req, SHpnt->unchecked_isa_dma,NULL);
 }
 
-/*
- * IOMMU hackery for sparc64
- */
-#ifdef DMA_CHUNK_SIZE
-
-#define MERGEABLE_BUFFERS(X,Y) \
-	((((bvec_to_phys(__BVEC_END((X))) + __BVEC_END((X))->bv_len) | bio_to_phys((Y))) & (DMA_CHUNK_SIZE - 1)) == 0)
-
-static inline int scsi_new_mergeable(request_queue_t * q,
-				     struct request * req,
-				     struct bio *bio)
-{
-	int nr_segs = bio_hw_segments(q, bio);
-
-	/*
-	 * pci_map_sg will be able to merge these two
-	 * into a single hardware sg entry, check if
-	 * we'll have enough memory for the sg list.
-	 * scsi.c allocates for this purpose
-	 * min(64,sg_tablesize) entries.
-	 */
-	if (req->nr_segments + nr_segs > q->max_segments)
-		return 0;
-
-	req->nr_segments += nr_segs;
-	return 1;
-}
-
-static inline int scsi_new_segment(request_queue_t * q,
-				   struct request * req,
-				   struct bio *bio)
-{
-	int nr_segs = bio_hw_segments(q, bio);
-	/*
-	 * pci_map_sg won't be able to map these two
-	 * into a single hardware sg entry, so we have to
-	 * check if things fit into sg_tablesize.
-	 */
-	if (req->nr_hw_segments + nr_segs > q->max_segments)
-		return 0;
-	else if (req->nr_segments + nr_segs > q->max_segments)
-		return 0;
-
-	req->nr_hw_segments += nr_segs;
-	req->nr_segments += nr_segs;
-	return 1;
-}
-
-#else /* DMA_CHUNK_SIZE */
-
 static inline int scsi_new_segment(request_queue_t * q,
 				   struct request * req,
 				   struct bio *bio)
 {
 	int nr_segs = bio_hw_segments(q, bio);
 
-	if (req->nr_segments + nr_segs > q->max_segments) {
+	if (req->nr_phys_segments + nr_segs > q->max_segments) {
 		req->flags |= REQ_NOMERGE;
 		return 0;
 	}
@@ -260,10 +210,9 @@ static inline int scsi_new_segment(request_queue_t * q,
 	 * This will form the start of a new segment.  Bump the 
 	 * counter.
 	 */
-	req->nr_segments += nr_segs;
+	req->nr_phys_segments += nr_segs;
 	return 1;
 }
-#endif /* DMA_CHUNK_SIZE */
 
 /*
  * Function:    __scsi_merge_fn()
@@ -306,11 +255,6 @@ __inline static int __scsi_back_merge_fn(request_queue_t * q,
 		return 0;
 	}
 
-#ifdef DMA_CHUNK_SIZE
-	if (MERGEABLE_BUFFERS(req->biotail, bio))
-		return scsi_new_mergeable(q, req, bio);
-#endif
-
 	return scsi_new_segment(q, req, bio);
 }
 
@@ -323,10 +267,6 @@ __inline static int __scsi_front_merge_fn(request_queue_t * q,
 		return 0;
 	}
 
-#ifdef DMA_CHUNK_SIZE
-	if (MERGEABLE_BUFFERS(bio, req->bio))
-		return scsi_new_mergeable(q, req, bio);
-#endif
 	return scsi_new_segment(q, req, bio);
 }
 
@@ -396,7 +336,7 @@ inline static int scsi_merge_requests_fn(request_queue_t * q,
 	if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
 		return 0;
 
-	bio_segs = req->nr_segments + next->nr_segments;
+	bio_segs = req->nr_phys_segments + next->nr_phys_segments;
 	if (blk_contig_segment(q, req->biotail, next->bio))
 		bio_segs--;
 
@@ -406,26 +346,11 @@ inline static int scsi_merge_requests_fn(request_queue_t * q,
 	if (bio_segs > q->max_segments)
 		return 0;
 
-#ifdef DMA_CHUNK_SIZE
-	bio_segs = req->nr_hw_segments + next->nr_hw_segments;
-	if (blk_contig_segment(q, req->biotail, next->bio))
-		bio_segs--;
-
-	/* If dynamic DMA mapping can merge last segment in req with
-	 * first segment in next, then the check for hw segments was
-	 * done above already, so we can always merge.
-	 */
-	if (bio_segs > q->max_segments)
-		return 0;
-
-	req->nr_hw_segments = bio_segs;
-#endif
-
 	/*
 	 * This will form the start of a new segment.  Bump the 
 	 * counter.
 	 */
-	req->nr_segments = bio_segs;
+	req->nr_phys_segments = bio_segs;
 	return 1;
 }
 
@@ -483,7 +408,7 @@ __inline static int __init_io(Scsi_Cmnd * SCpnt,
 	if (!sg_count_valid) {
 		count = __count_segments(req, dma_host, NULL);
 	} else {
-		count = req->nr_segments;
+		count = req->nr_phys_segments;
 	}
 
 	/*
@@ -530,7 +455,7 @@ __inline static int __init_io(Scsi_Cmnd * SCpnt,
 		 */
 		printk("Warning - running *really* short on DMA buffers\n");
 		this_count = req->current_nr_sectors;
-		printk("SCSI: depth is %d, # segs %d, # hw segs %d\n", SHpnt->host_busy, req->nr_segments, req->nr_hw_segments);
+		printk("SCSI: depth is %d, # phys segs %d, # hw segs %d\n", SHpnt->host_busy, req->nr_phys_segments, req->nr_hw_segments);
 		goto single_segment;
 	}
 
diff --git a/include/asm-i386/io.h b/include/asm-i386/io.h
index 0c5e61d14..975f0bf61 100644
--- a/include/asm-i386/io.h
+++ b/include/asm-i386/io.h
@@ -105,13 +105,6 @@ extern void iounmap(void *addr);
 #define page_to_bus page_to_phys
 
 /*
- * can the hardware map this into one segment or not, given no other
- * constraints.
- */
-#define BIOVEC_MERGEABLE(vec1, vec2)	\
-	((bvec_to_phys((vec1)) + (vec1)->bv_len) == bvec_to_phys((vec2)))
-
-/*
  * readX/writeX() are used to access memory mapped devices. On some
  * architectures the memory mapped IO stuff needs to be accessed
  * differently. On the x86 architecture, we just read/write the
diff --git a/include/asm-sparc64/dma.h b/include/asm-sparc64/dma.h
index aeef575cc..cd8370e13 100644
--- a/include/asm-sparc64/dma.h
+++ b/include/asm-sparc64/dma.h
@@ -1,4 +1,4 @@
-/* $Id: dma.h,v 1.20 2001-12-11 11:39:57 davem Exp $
+/* $Id: dma.h,v 1.21 2001-12-13 04:16:52 davem Exp $
  * include/asm-sparc64/dma.h
  *
  * Copyright 1996 (C) David S. Miller (davem@caip.rutgers.edu)
@@ -218,12 +218,4 @@ extern int isa_dma_bridge_buggy;
 #define isa_dma_bridge_buggy 	(0)
 #endif
 
-/* We support dynamic DMA remapping and adjacent SG entries
- * which have addresses modulo DMA_CHUNK_SIZE will be merged
- * by dma_prepare_sg().
- */
-#if 0 /* XXX Buggy in 2.5.x currently... */
-#define DMA_CHUNK_SIZE 8192
-#endif
-
 #endif /* !(_ASM_SPARC64_DMA_H) */
diff --git a/include/asm-sparc64/io.h b/include/asm-sparc64/io.h
index 7d4480719..be6c5546b 100644
--- a/include/asm-sparc64/io.h
+++ b/include/asm-sparc64/io.h
@@ -1,4 +1,4 @@
-/* $Id: io.h,v 1.45 2001-12-12 00:13:56 davem Exp $ */
+/* $Id: io.h,v 1.46 2001-12-13 04:16:52 davem Exp $ */
 #ifndef __SPARC64_IO_H
 #define __SPARC64_IO_H
 
@@ -21,14 +21,6 @@ extern unsigned long bus_to_virt_not_defined_use_pci_map(volatile void *addr);
 extern unsigned long phys_base;
 #define page_to_phys(page)	((((page) - mem_map) << PAGE_SHIFT)+phys_base)
 
-#ifdef DMA_CHUNK_SIZE
-#define BIOVEC_MERGEABLE(vec1, vec2)	\
-	((((bvec_to_phys((vec1)) + (vec1)->bv_len) | bvec_to_phys((vec2))) & (DMA_CHUNK_SIZE - 1)) == 0)
-#else
-#define BIOVEC_MERGEABLE(vec1, vec2)	\
-	((bvec_to_phys((vec1)) + (vec1)->bv_len) == bvec_to_phys((vec2)))
-#endif
-
 /* Different PCI controllers we support have their PCI MEM space
  * mapped to an either 2GB (Psycho) or 4GB (Sabre) aligned area,
  * so need to chop off the top 33 or 32 bits.
diff --git a/include/linux/bio.h b/include/linux/bio.h
index a7c0c2576..dca5debcf 100644
--- a/include/linux/bio.h
+++ b/include/linux/bio.h
@@ -130,6 +130,8 @@ struct bio {
  */
 #define __BVEC_END(bio)		bio_iovec_idx((bio), (bio)->bi_vcnt - 1)
 #define __BVEC_START(bio)	bio_iovec_idx((bio), 0)
+#define BIOVEC_MERGEABLE(vec1, vec2)	\
+	((bvec_to_phys((vec1)) + (vec1)->bv_len) == bvec_to_phys((vec2)))
 #define BIO_CONTIG(bio, nxt) \
 	BIOVEC_MERGEABLE(__BVEC_END((bio)), __BVEC_START((nxt)))
 #define __BIO_SEG_BOUNDARY(addr1, addr2, mask) \
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 204ab9765..658f94e7e 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -41,8 +41,19 @@ struct request {
 					 * touch them
 					 */
 	unsigned long hard_nr_sectors;
-	unsigned short nr_segments;
+
+	/* Number of scatter-gather DMA addr+len pairs after
+	 * physical address coalescing is performed.
+	 */
+	unsigned short nr_phys_segments;
+
+	/* Number of scatter-gather addr+len pairs after
+	 * physical and DMA remapping hardware coalescing is performed.
+	 * This is the number of scatter-gather entries the driver
+	 * will actually have to deal with after DMA mapping is done.
+	 */
 	unsigned short nr_hw_segments;
+
 	unsigned int current_nr_sectors;
 	unsigned int hard_cur_sectors;
 	void *special;