Red Hat Bugzilla – Bug 112826
LTC5732 - MMIO alignment error when inserting the olympic TR module.
Last modified: 2007-11-30 17:06:53 EST
Hardware Environment: iSeries 64bit Regatta Software Environment: RHEL 3 QU1 Beta 1 Steps to Reproduce: 1. Boot a partition with native TR. 2. Remove (rmmod) the olympic module by hand 3. Reinsert (insmod) the olympic module by hand. 4. May require one or two repetitions. Actual Results: A kernel panic is caused: Expected Results: The kernel panic should not happen. Additional Information: This problem was previously solved. It is documented in bug 1011. I am not sure why the fix has been lost from the kernel. Here is an explanation of the change that was proposed: -------- Begin snip from bug 1011 I made a change to the ./drivers/net/tokenring/olympic.c code to handle the case where the skb does not happen to be aligned on a word boundary. This generally happens on a error path where the initial call to olympic_open failed and it is doing a second try. The change was to replace the 3 lines of code that clear out the srb command field with "memset_io(init_srb,0,SRB_COMMAND_SIZE); " This change is checked into the 2.4.19 cvs tree. -------- End snip
Created attachment 96748 [details] patch from bug 1011 ------- Additional Comment #1 From Michael Ranweiler 2004-01-02 17:55 ------- patch from bug 1011 This is Al's patch from the 2.4.19 kernel - can see if it fixes...
------ Additional Comments From jsbryant@us.ibm.com 2004-01-05 17:26 ------- Mike, I have tested this patch in a kernel and verified that the kernel no longer panics on Regatta.
------ Additional Comments From mranweil@us.ibm.com 2004-01-05 17:50 ------- Julie/Mark - I don't know what you thought of that patch (was grabbed out of an old bug we had) but it makes olympic on power4 not panic, which is probably useful. ;) Any chance on the refresh of QU1?
Mike, it is unlikely that this will be taken in the U1 refresh at this late date since it doesn't fix a data corruptor or security problem. How big an issue is this? Do you have many customers with native token ring doing this type of operation (rmmod and then insmod)? We would prefer to wait until Update 2 for this fix.
------ Additional Comments From mranweil@us.ibm.com 2004-01-06 10:53 ------- Well, it's not completely critical - it only happens after the rmmod, and on power 4. But it'd be nice... (it is a panic).
------ Additional Comments From mranweil@us.ibm.com 2004-01-09 16:05 ------- Setting target to qu2.
----- Additional Comments From mranweil@us.ibm.com 2004-01-13 10:53 ------- Added Jay Vosburgh to the cc - hopefully he'll have comments on the patch... Need to get something for this problem into mainline.
----- Additional Comments From mranweil@us.ibm.com 2004-01-19 16:31 ------- Jay - you have a chance to look at the patch here?
----- Additional Comments From khoa@us.ibm.com 2004-01-25 19:19 ------- Patch available, so move this bug into FixedAwaitingTest state. Jay - please review the patch above. Thanks.
----- Additional Comments From khoa@us.ibm.com 2004-01-25 19:20 ------- This is targeted for RHEL3 QU2.
----- Additional Comments From khoa@us.ibm.com 2004-01-25 19:21 ------- Jay (Bryant) already tested this patch and found it not panic anymore, so set the state to Tested.
Fix has been submitted to our tree and will appear in Update 2.
----- Additional Comments From mranweil@us.ibm.com 2004-02-09 15:40 ------- Jay V checked for this - it's in 2.6.2 from Jeff Garzik - thanks for pushing. Closing bug.
An errata has been issued which should help the problem described in this bug report. This report is therefore being closed with a resolution of ERRATA. For more information on the solution and/or where to find the updated files, please follow the link below. You may reopen this bug report if the solution does not work for you. http://rhn.redhat.com/errata/RHSA-2004-188.html