Note: This bug is displayed in read-only format because the product is no longer active in Red Hat Bugzilla.

Bug 64149

Summary: Bug in scsi_merge.c
Product: [Retired] Red Hat Linux Reporter: Tom "spot" Callaway <tcallawa>
Component: kernelAssignee: Doug Ledford <dledford>
Status: CLOSED ERRATA QA Contact: Brian Brock <bbrock>
Severity: medium Docs Contact:
Priority: medium    
Version: 7.2CC: conway_heather, coughlan, gary_lerhaupt, john_hull, matt_domsch
Target Milestone: ---   
Target Release: ---   
Hardware: i386   
OS: Linux   
Whiteboard:
Fixed In Version: Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2003-02-13 18:20:33 UTC Type: ---
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Attachments:
Description Flags
scsi_merge.c patch none

Description Tom "spot" Callaway 2002-04-26 16:55:57 UTC 
filing on behalf of Ed Goggin @ EMC

Description of Problem:

The bug is in __init_io() of drivers/scsi/scsi_merge.c.
This bug causes partially completed i/os (special requests)
to occur in cases where an i/o request consists of only a
single segment but multiple buffer headers.  This will only
happen with an HBA driver which supports DMA clustering.  We
get hundreds if not thousands of such partially completed i/os
to occur by issuing either raw or buffered sd device i/o.

The major effect of this bug is to slow down i/o throughput
since the linux scsi subsystem appears to handle such partially
completed i/os by re-issuing the residual portion of the i/o
as a special request.

Version-Release number of selected component (if applicable):

2.4.9-31

The patch for this is simple (1 liner) and is attached to this bugreport.
Comment 1 Tom "spot" Callaway 2002-04-26 16:56:30 UTC 
Created attachment 55484 [details]
scsi_merge.c patch
Comment 2 Matt Domsch 2002-07-02 15:13:57 UTC 
This is addressed in a separate fix in the 2.4.19-rc1 kernel at least (I don't 
know when it was added exactly):

         * Don't bother with scatter-gather if there is only one segment.
         */
        if (count == 1) {
                this_count = SCpnt->request.nr_sectors;
                goto single_segment;
        }
 
The incorrect code is present in 2.4.18-3 at least.
Comment 3 Matt Domsch 2002-07-02 16:49:43 UTC 
Dell and EMC would like to see this fixed in a Pensacola errata kernel.
Comment 4 Gary Lerhaupt 2002-07-19 20:09:04 UTC 
Kyu (ykang) informed us that the patch included in this bugzilla 
will be added to the next AS errata.  Can this be confirmed?
-Gary
Comment 5 Yil-Kyu Kang 2002-07-19 20:52:22 UTC 
Yes, this patch has been accepted and we are planning on shipping this in our
next AS errata.
Comment 6 Tom Coughlan 2002-09-20 20:10:54 UTC 
This fix is checked in to the AS CVS source tree.  We expect it to ship in an
errata as soon feasible.
Comment 7 Matt Domsch 2003-02-13 18:20:33 UTC 
Fixed in kernel 2.4.9-e.9 at least, and 2.4.20-based kernels slated for 
GinGin.  Closing.