From: Berkley Shands <berkley@cs.wustl.edu>
Date: Mon, 5 Jan 2004 14:00:16 -0600 (CST)
Message-Id: <200401052000.i05K0Gc0000014396@mudpuddle.cs.wustl.edu>
To: berkley@cs.wustl.edu, davem@redhat.com
Subject: Re: [BUG] x86_64 pci_map_sg modifies sg list - fails multiple map/unmaps
Cc: gibbs@scsiguy.com, linux-kernel@vger.kernel.org,
       linux-scsi@vger.kernel.org
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1055
Lines: 31

	   		BUG_ON(s->length == 0);

		size += s->length; 

		/* Handle the previous not yet processed entries */

		if (i > start) {
			struct scatterlist *ps = &sg[i-1];
			/* Can only merge when the last chunk ends on a page 
			   boundary. */
			if (1 || !force_iommu || !need || (i-1 > start && ps->offset) ||
			    (ps->offset + ps->length) % PAGE_SIZE) { 
				if (pci_map_cont(sg, start, i, sg+out, pages, 
						 need) < 0)
					goto error;

If I totally disable coalescing, in arch/x86_64/pci-gart.c by adding the if (1 || ...
then the bughalt goes away. So might some performance, but a system that runs is
MUCH better than one that hard faults on a regular basis.

The shipped scsi driver still needs to be updated to NOT "offline" the drives.
the 2.0.5 driver does this quite well.

Thank you.

berkley
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/