Experimental Physics and
Industrial Control System

"Jeff Hill" <[email protected]> · Fri, 11 Jan 2002 17:51:54 -0700

Brian,

This response assumes that when you say "reboot" you mean a soft reboot
initiated by ^X or the "reboot" command.

In the past when we have seen soft reboot failures here they were always
caused by device drivers that do not have reboot hooks installed to turn
off VME board initiated interrupts and VME board initiated DMA
operations. During soft reboots there is, by default, no sysreset on the
VME bus and if these hardware requests continue during the boot sequence
they eventually crash the vxWorks OS while it is booting. 

As I recall, the WRS reboot hooks were run in the past _after_ the
network services and many other parts of vxWorks services were shutdown,
but this may have been fixed in a subsequent releases. In any case,
there are not currently any reboot hooks in the CA server because during
attempts to add them in the past I became convinced that vxWorks was
already so far along in the shutdown sequence that there was not
anything useful that could be done. Of course, if the situation with
vxWorks has changed then I should probably reinvestigate.

Please send additional details about the messages you have received and
why you believe that CA is responsible and I will take a look.

Jeff

> -----Original Message-----
> From: Brian McAllister [mailto:[email protected]]
> Sent: Friday, January 11, 2002 4:59 PM
> To: [email protected]
> Cc: [email protected]
> Subject: channel access reboot behavior
> 
> 
> As currently implemented, Channel Access on the IOC tries to
communicate
> with remote clients/servers when shutting down prior to reboot (this
is
> inferred from error messages seen during the shutdown).
Unfortunately,
> network services will always have been shut down first, so these
> notification attempts always fail, and will continue to do so unless
the
> VxWorks reboot processing is modified to allow user hooks to be
handled
> first, which seems unlikely.
> 
> Occasionally, one of our IOCs will hang on reboot, requiring physical
> access to push the reset button.  I am wondering if this hang might in
> part
> be due to the system trying to do things while in a partially shut
down
> state, and if there is any reason not to eliminate the CA server
behavior
> noted above, on the chance that it may be contributing to the problem.
> 
> Or perhaps, EPICS could have a "shutdown" routine, which terminates
the
> EPICS tasks appropriately, and then calls reboot ?
> 
> ----
> Brian McAllister                    Controls Programmer/Beam Physicist
> [email protected]                        MIT-Bates Linear Accelerator
> (617) 253-9537                                           Middleton, MA

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System