Why don't you try lifting your fingers to defend your position? We know
1. the change introduces state data to tlb with no limits on the
duration of said stale data
2. some architectures have large tlbs.
For the sake of (2), hashed tlbs count as excessively large tlbs. See the
UML, ia64 and powerpcs for this kind of setup.
Now, if you want to live in microbenchmark land, then, yes, the "optimized"
but invalid behaviour will *always* win. Back in the real world, sure, I can
construct a microbenchmark that will sit on 2 pages of memory spinning until
the pages get swapped out. What does that buy us? Nothing other than
I've already pointed out that the optimization can be done in a valid way. If
you can't go through the effort of completing the above thought experiment on
your own, then I really have no reason to care about your opinion any further.
You've already stated you don't care about mine, and I've stated that I don't
feel like going to the effort to placate you. Where does that leave us?
Nowhere, I guess.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to email@example.com
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/