tech-kern archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: extend sysmon watchdog modes?



Cliff Neighbors <cliff%netbsd.org@localhost> writes:

> Basically the hypothetical system is still alive, just bogged 
> under interrupt load, and will resume normal functioning when 
> the load relents.
>
> Ability to recover makes a significantly different case from a
> system in some pathological halt, infinite fault, deadlock or spin.
>
> Also, if the watchdog tickle were called from softnet, then the 
> stack could be still functioning, vs. from callout (softclock)
> the watchdog would time out.

So you mean a system where for tens of seconds there is no callout
processing, but there is still softnet processing.   I guess I can see
that, but in my experience it seems more likely that such  a system is
irretrievably hosed than that it will recover.

> The mode of the watchdog should be a matter of policy depending on the 
> application. As others are saying, the (proposed) ITICKLE policy
> is not appropriate for some (most) systems. But the mechanism could be 
> flexible to allow it where needed.

Sure, that seems ok.   I didn't have the impression anyone objected to
adding the capability, just doubting that it made sense to use.

Attachment: pgpRcHMOf1mWm.pgp
Description: PGP signature



Home | Main Index | Thread Index | Old Index