Simon Falvey wrote:
>
> Hi,
>
> When a system is panicing and writting a crash dump file. Is it expected
> that the kernel keeps servicing user processes? It appears that you are able
> to switch virtual consoles, log in, run things while the vmdump process is
> doing its stuff. Surely this will severly confuse the dump image being
> created, it won't be so much of a snapshot of the system when the panic took
> place as a smudge.
That's what the patch I sent to Michael is trying to address. It
attempts to stop all scheduling when a dump is taking place, and
tries to avoid schedule() being activated inside an interrupt handler.
I/O on terminals is a good point, however.
> Also, I have noticed that inside lcrash
>
> >> ? id --- produces the help text for dis
> >> ? idcmds -- produces an error
>
> also other help texts such as bt produce help texts for commands for which
> they are possible abbreviations however as the helptext does not say this it
> could be that they too are coming up with the wrong information.
These were meant to document 'kdb' commands which area really just
aliases inside of 'lcrash'. For example, old System 5 crash used
'dump', but other analyzers used 'od', etc. The point was to try and
give people the commands they were used to.
I can fix the help commands. 'idcmds' shouldn't give an error.
> Cheers
>
> Simon
>
> Simon Falvey
> Online Product Support Specialist
> VERITAS Software UK.
> Tel: +44 118 918 8105
Thanks, Simon.
--Matt
|