tech-net archive

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

Re: Revamping optimised in_cksum/in4_cksum/in6_cksum support

Joerg Sonnenberger wrote:

> On Mon, Jan 28, 2008 at 01:18:03AM +1100, Simon Burge wrote:
> > At least MIPS and PowerPC will generate "pref" and "dcbt" instructions
> > respectively that point to potentially invalid addresses with the code
> > as it currently is.
> I remember some discussion of the Power4 modifications for Linux to map
> NULL for the same reason. If in doubt, we can conditionalize that
> statement for platforms that don't have a save prefetch.

The cases I'm referring to aren't against zero or low addresses, but for
any generic virtual addresses that aren't mapped.

I'm thinking that it might be safer to default to no prefetch (or at
least to only prefetching known valid addresses) given that we have two
analysed architectures so far and both have the possibility of problems.


Home | Main Index | Thread Index | Old Index