----- Original Message ----- > From: "Mike Holmes" <mike.hol...@linaro.org> > To: "Nicolas Morey-Chaisemartin" <nmo...@kalray.eu> > Cc: "lng-odp" <lng-odp@lists.linaro.org> > Sent: Tuesday, 21 April, 2015 7:24:44 PM > Subject: Re: [lng-odp] [PATCH] pool: Do not flush cache of non-initialized > pools
> On 21 April 2015 at 05:06, Nicolas Morey-Chaisemartin < nmo...@kalray.eu > > wrote: > > Flush cache triggers atomic_add on some of the pool atomics which have > > never > > been initialized if the pool hasn't been. > > > Signed-off-by: Nicolas Morey-Chaisemartin < nmo...@kalray.eu > > > Subject needs to isolate the basic area for the patch by adding > "linux-generic:" and we try to avoid punctuation, in this case the "Do" = > "do" > linux-generic: pool: do not flush cache of non-initialized pools Noted for next time. Do you need me to resubmit with a fixed title? > Do you have any test case we can add to test/validation that exposes the need > to do this ? > The test suit is currently sunny day focused and I am guessing you tripped > over this in real code The reason behind this patch (and the other fixing create_pool) is that we don't have a test and set in our first generation of processor but a test and clear. So if for some reason atomic operations (using the lock) are run on an unitialized atomic, it stalls forever. The ideal way to test this would be to have a Debug/Check mode that makes sure an "init" field was initialized to a dummy non zero value. But it needs a lot of ifdefs and not-so-pretty code to build a complete test case and I'm not sure that is something you would want upstream. Nicolas
_______________________________________________ lng-odp mailing list lng-odp@lists.linaro.org https://lists.linaro.org/mailman/listinfo/lng-odp