Re: 2.4.21-pre3 kernel crash

[Jens Axboe]

On Mon, Jan 27 2003, Martin MOKREJ? wrote:
> On Mon, 27 Jan 2003, Ross Biro wrote:
> 
> > This looks like the same problem I ran into with IDE and highmem not
> > getting along.  Try compiling your kernel with out highmem enabled and
> > see what happenes.
> 
> Yes, that "fixes" it. Any "better solution"? ;-)
> 
> > >Trace; c024dfc1 <ide_build_sglist+181/1a0>
> > >Trace; c024e1b4 <ide_build_dmatable+54/1a0>
> > >Trace; c024e6df <__ide_dma_read+3f/150>

Someone completely lost the highmem capable scatterlist setup, *boggle*.
This should fix it.

===== drivers/ide/ide-dma.c 1.7 vs edited =====
--- 1.7/drivers/ide/ide-dma.c	Wed Nov 20 18:46:24 2002
+++ edited/drivers/ide/ide-dma.c	Mon Jan 27 20:22:06 2003
@@ -249,6 +249,7 @@
 {
 	struct buffer_head *bh;
 	struct scatterlist *sg = hwif->sg_table;
+	unsigned long lastdataend = ~0UL;
 	int nents = 0;
 
 	if (hwif->sg_dma_active)
@@ -256,22 +257,28 @@
 
 	bh = rq->bh;
 	do {
-		unsigned char *virt_addr = bh->b_data;
-		unsigned int size = bh->b_size;
+		if (bh_phys(bh) == lastdataend) {
+			sg[nents - 1].length += bh->b_size;
+			lastdataend += bh->b_size;
+			continue;
+		}
 
 		if (nents >= PRD_ENTRIES)
 			return 0;
 
-		while ((bh = bh->b_reqnext) != NULL) {
-			if ((virt_addr + size) != (unsigned char *) bh->b_data)
-				break;
-			size += bh->b_size;
-		}
 		memset(&sg[nents], 0, sizeof(*sg));
-		sg[nents].address = virt_addr;
-		sg[nents].length = size;
-		if(size == 0)
-			BUG();
+		if (bh->b_page) {
+			sg[nents].page = bh->b_page;
+			sg[nents].offset = bh_offset(bh);
+		} else {
+			if (((unsigned long) bh->b_data) < PAGE_SIZE)
+				BUG();
+
+			sg[nents].address = bh->b_data;
+		}
+
+		sg[nents].length = bh->b_size;
+		lastdataend = bh_phys(bh) + bh->b_size;
 		nents++;
 	} while (bh != NULL);
 

-- 
Jens Axboe