Re: [net-next v44] mctp pcc: Implement MCTP over PCC Transport

[Jeremy Kerr <jk@codeconstruct.com.au>]

Hi Adam,

> > > The code itself will not, as written, work on a 32 Bit system, as there
> > > are 64 bit specific code.
> > Yeah, that was more my question - what is 64-bit specific about the
> > code?
> 
> I'd really have to dig, as this decision was mixed in with the earlier 
> endinaness conversions.  I suspect, that it was origianally triggered by 
> the ACPICA insisting on Machine architecture for these values, where 
> they are supposed to be explicitly 32, 64 bit etc.

Looking at the history, it appears like this was a in response to the
AI review comments about the stats update with interrupts enabled. The
report was that it may result in tearing of the stats values on 32-bit
machines, but if you have the stats updates right (which I think
you do now?) then this is not an issue.

> It might be perfectly safe, but I have no way to test. Treat it as a 
> general trend toward not supporting newer technology on older architectures.

This is more about not imposing a configuration restriction with no
purpose. There is no need for you to test/support those configurations
though.

> Could I safely acquire a spinlock in the rx_ callback during interupt
> context?

Yes, absolutely. That's a main use-case for spinlocks, to allow
serialisation from within an atomic context.

> I thought that had a significant impact on the system.

You don't want to be doing large amounts of work within the critical
section (and with interrupts disabled), but your scenario should be
fine.

Cheers,


Jeremy