kern/58391: Trap 12 with heavy disk load on ide vinum mirror

[Alexander Haderer]

The following reply was made to PR kern/58391; it has been noted by GNATS.

From: Alexander Haderer <alexander.haderer at charite.de>
To: Greg 'groggy' Lehey <grog at FreeBSD.org>
Cc: FreeBSD-gnats-submit at FreeBSD.org
Subject: Re: kern/58391: Trap 12 with heavy disk load on ide vinum
  mirror
Date: Mon, 27 Oct 2003 18:26:44 +0100

 Update info:
 
 Meanwhile I did further tests, here are the results:
 
 1.
 To exclude some broken hardware issues I did the test on another machine 
 (Celeron 533/i81[05]) with a similar hw/sw setup (scsi to boot, ad0+ad2 for 
 vinum-mirror). Again the kernel crashes, here we have a double pagefault, 
 second one triggered by syncing disks as it seems to me. All dumps are save 
 for later use.
 
 2.
 Again I was unable to trigger a panic on the replacement hw without using 
 vinum. It seems to be that vinum at least helps to trigger a panic when 
 doing ide disk io.
 
 3.
 Further investigations of the trap 18 div-by-zero dump gdb shows that 
 (within a vinum function) parts of the drive->partinfo.disklab struct gets 
 overwritten with zeros and the hostname. As a result, dev/vinum/vinumio.c, 
 function driveio, computation of the block number bombs out because 
 d_secsize is zero.
 
      330:   bp->b_blkno = offset / drive->partinfo.disklab->d_secsize
 
 (All with 4.8R source from CD)
 
 In the hope this is some useful information,
 
 Alexander