Vivek Goyal <[EMAIL PROTECTED]> writes:

> So basically the idea is that whatever one wants to do it should be done
> in the next kernel, even notifications. But this might require some data
> from the context of previous kernel, for example destination IP address etc.
> So the associated data either needs to be passed to new kernel or it shall
> have to be retrieved from permanent storage or something like that.

Yes.  

Except when under tight time constraints this is clearly possible.
When under time constraints things are more dicy. I expect that
with a tuned kernel you can be back in user space in about 1 second.

Detecting a failure is hard and slow.  Unless the condition that
triggers the panic is the initial source of the panic it is quite
likely a time window less than 1 second has already elapsed.

So I am certainly willing to discuss this but until I see a clear
model where a new kernel will not meet the time constraints for
some fairly fundamental reasons, and that a crash notifier
will almost always meet those same time constraints, I can't imagine
a persuasive case.

The panic notifier itself is barely used and it seems to exist
only because we didn't have a general purpose policy mechanism.

Eric
_______________________________________________
fastboot mailing list
[email protected]
https://lists.osdl.org/mailman/listinfo/fastboot

Reply via email to