Conference turris::digital_unix

9308.0. "8400, KZPSA -> MTI, pciaerror panic" by DYPSS1::SCHAFER (Kalh�un!) Wed Mar 26 1997 21:06

    This sounds suspiciously similar to 9302, but what the heck ...
    
Configuration:

   DEC Unix v4.0B, 2-27-97 patches installed.
   Alpha 8400 (EV56/440) x 8, 8GB memory (4 boards)
   13 KZPSAs; 9 connected to MTI shelves

   Each MTI shelf = 1 controller serving 1 8-member RAID-5 array

   All 9 RAID-5 arrays are used by one rootdg-based LSM volume.  Each
   individual disk is sliced (g public, h private).  Strangely, each
   disk's C partition cylinder count does not match the combination of
   g & h (g+h > c).  Don't know if this has any relationship to the 
   problem or not.

   One ADVfs domain/fset (db_domain#db) uses this volume.  

Problem:

   Customer is encountering several types of panics, including pciaerror,
   machine check, and ADVfs-related crashes.  while panics are not always
   consistent, a panic is easily reproduced by the following:

	introduce a *large* I/O load (test pgm is a perl script)
	cause one RAID-5 set to enter a reconstruct state
	cause a second RAID-5 to enter a reconstruct state

   The panic cannot be duplicated if the I/O load is introduced AFTER the
   reconstruction operations have begun.  It does not matter which RAID
   sets are forced to fail.

   I'll include the latest crash output and dia output as replies to this
   note.

    since the crash and dia output are pretty long, i thought i'd just post
    pointers:
    
    	DYPSS1::crash.abe
    	DYPSS1::dia.abe
    
    DYPSS1 is 35223 for the name-impaired.  cheers,
    
    +b