On Fri, Oct 09, 2020 at 11:53:00AM +0200, Peter Zijlstra wrote: > On Fri, Oct 09, 2020 at 10:37:51AM +0100, Will Deacon wrote: > > On Fri, Oct 09, 2020 at 11:09:27AM +0200, Peter Zijlstra wrote: > > > > Patch 4 makes it all far worse by exposing it to pretty much everybody. > > > > > > Now, I think we can fix at least the user mappings with the below delta, > > > but if archs are using non-page-table MMU sizes we'll need arch helpers. > > > > > > ARM64 is in that last boat. > > > > > > Will, can you live with the below, if not, what would you like to do, > > > make the entire function __weak so that you can override it, or hook > > > into it somewhere? > > > > Hmm, so I don't think we currently have any PMUs that set 'data->addr' > > on arm64, in which case maybe none of this currently matters for us. > > > > However, I must admit that I couldn't figure out exactly what gets exposed > > to userspace when the backend drivers don't look at the sample_type or > > do anything with the addr field. > > Patch 4: > > https://lkml.kernel.org/r/20201001135749.2804-5-kan.li...@linux.intel.com > > is the one that exposes this to everybody with perf support. It will > then report the page-size for the code address (SAMPLE_IP).
I can see there have another potentail customer to use page-size is Arm SPE, but Arm SPE is hardware trace based sample but not interrupt based sample. For this case, I think this patch set cannot be directly applied to the AUX trace data. Thanks, Leo