On Fri, Feb 11, 2000 at 05:51:26PM -0500, Jay.Estabrook@compaq.com wrote:
> pci_map_sg appears to do incorrect IOMMU PTE init during processing of
> the above sglist, as it attempts to make the list virtually contiguous
> and ends up initializing more IOMMU entries than it should.
Yep. Once davem and I figured out what was wrong with the
sym53c8xx driver, I was able to reproduce this here too.
Attached is a fixed pci_iommu.c.
> qlogic driver doesn't squirrel away the count of entries returned by
> pci_map_sg and when it comes time to free the list, often does too
> many entries and runs into some bogus ones, causing panics in pci_free_sg.
This probably isn't fully correct, but a similar change to the sym
driver appears to work under load. If it's wrong it's going to fail
on scsi errors, which I don't know how to force.
diff -upr linux/drivers/scsi/qlogicisp.c 2.3.44-3/drivers/scsi/qlogicisp.c
--- linux/drivers/scsi/qlogicisp.c Sat Feb 12 17:17:17 2000
+++ 2.3.44-3/drivers/scsi/qlogicisp.c Sat Feb 12 17:22:10 2000
@@ -811,6 +811,7 @@ int isp1020_queuecommand(Scsi_Cmnd *Cmnd
ds = cmd->dataseg;
sg_count = pci_map_sg(hostdata->pci_dev, sg, Cmnd->use_sg);
+ Cmnd->use_sg = sg_count;
cmd->segment_cnt = cpu_to_le16(sg_count);
Thoughts on a better way to fix?
r~
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sat Feb 12 2000 - 21:20:45 EST