On Wed, Oct 5, 2016 at 10:44 PM, Mark Rutland <[email protected]> wrote: > On Wed, Oct 05, 2016 at 08:00:38PM +0000, Fredrik Markström wrote: >> On Wed, Oct 5, 2016 at 7:48 PM Robin Murphy <[email protected]> wrote: >> As far as I understand TPIDRURW isn't anything else then an architecture >> specific piece of tls since the last patch, possibly slightly faster then a >> "__thread u32 x;" >> >> The irony is that the two different ways it was handled earlier (not context >> switched or always set to zero on swap in) would have made it useful for this >> purpose. > > The "not context switched" case was also arbitrarily corrupted, and could not > have been relied upon.
Ok, I missed that, sorry ! > > The zeroing case is similar to the restartable sequences design. So that's > probably worth looking into. Ok, I'm starting to believe my best bet is to hope that those make it into the kernel eventually, until then I'll probably just go with a local solution. /Fredrik > > Thanks, > Mark.
