Content-Disposition: inline
Content-Transfer-Encoding: 8bit
Content-Type: text/plain; charset="US-ASCII"
MIME-Version: 1.0
Date: Thu, 26 Oct 2023 15:13:38 +0000
From: "Martin Husemann" <martin@netbsd.org>
Subject: CVS commit: [netbsd-8] src/sys/dev/pci
To: source-changes@NetBSD.org
Approved: for-source-only
Reply-To: martin@netbsd.org
Message-Id: <20231026151338.81526FADC@cvs.NetBSD.org>

Module Name:	src
Committed By:	martin
Date:		Thu Oct 26 15:13:38 UTC 2023

Modified Files:
	src/sys/dev/pci [netbsd-8]: mpii.c

Log Message:
Pull up following revision(s) (requested by buhrow in ticket #1916):

	sys/dev/pci/mpii.c: revision 1.30

Fixes for PR kern/57133:

I can now explain why this assert is firing and have a fix for it.  It is a  regression introduced in R1.22 of mpii.c.

        If a request comes in and the IOC returns a MPII_SCSIIO_STATUS_CHECK_COND condition, after
 a successful transfer, or one that is a recovered error,
 mpii(4) correctly sets the xs->error to XS_SENSE, but incorrectly sets xs->resid to 0 before
 returning the xfer to the upper scsi layers.  Once the upper layers get it, they notice the
 XS_SENSE check condition and because it's a retryable error, they increment xs_requeuecnt, set
 ERESTART and send the xfer request down to the mpii(4) layer again for a retry. What they do
 not do is reset xs->resid equal to xs->datalen.  When the xfer comes down to mpii(4) again, the
 assert happens.  The fix is for the mpii(4) driver to leave xs->resid alone when it encounters
 a MPII_SCSIIO_STATUS_CHECK_COND condition.

This bug affects NetBSD-10, netbsd-9 and netbsd-8.


To generate a diff of this commit:
cvs rdiff -u -r1.8.10.7 -r1.8.10.8 src/sys/dev/pci/mpii.c

Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.