]> err.no Git - linux-2.6/commit
[SCSI] aic94xx: Don't call pci_map_sg for already-mapped scatterlists
authorDarrick J. Wong <djwong@us.ibm.com>
Sat, 11 Nov 2006 00:59:24 +0000 (16:59 -0800)
committerJames Bottomley <jejb@mulgrave.localdomain>
Wed, 18 Jul 2007 16:14:47 +0000 (11:14 -0500)
commitba330ffebb43c37cabc765c7cb0a80df01554657
treed703cb707917d1ce5daf0224b3d10b51c8825415
parent338ec57003ff9d7bc1471677e61872455977a5de
[SCSI] aic94xx: Don't call pci_map_sg for already-mapped scatterlists

It turns out that libata has already dma_map_sg'd the scatterlist
entries that go with an ata_queued_cmd by the time it calls
sas_ata_qc_issue.  sas_ata_qc_issue passes this scatterlist to aic94xx.
Unfortunately, aic94xx assumes that any scatterlist passed to it needs
to be pci_map_sg'd... which blows away the mapping that libata created!
This causes (on a x260) Calgary IOMMU table leaks and duplicate frees
when aic94xx and libata try to {pci,dma}_unmap_sg the scatterlist.

Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Key this check off ATA_PROTOCOL_STP

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
drivers/scsi/aic94xx/aic94xx_task.c