15:00:40 <pwhalen> #startmeeting Fedora ARM and AArch64 Status Meeting
15:00:40 <zodbot> Meeting started Tue Aug  2 15:00:40 2016 UTC.  The chair is pwhalen. Information about MeetBot at http://wiki.debian.org/MeetBot.
15:00:40 <zodbot> Useful Commands: #action #agreed #halp #info #idea #link #topic.
15:00:40 <zodbot> The meeting name has been set to 'fedora_arm_and_aarch64_status_meeting'
15:00:40 <pwhalen> #chair pwhalen pbrobinson dgilmore hrw dmarlin yselkowitz jonmasters ahs3 msalter
15:00:40 <zodbot> Current chairs: ahs3 dgilmore dmarlin hrw jonmasters msalter pbrobinson pwhalen yselkowitz
15:00:48 <pwhalen> good morning folks, whos here today?
15:00:58 * cov says hi
15:01:02 * jlinton waves
15:01:37 * pwhalen notes, ARM State of the Union talk is underway in Krakow, we may not have much of a quorum today
15:01:58 <pwhalen> mornign gents :)
15:04:19 <pwhalen> I'll leave the meeting open for a couple more minutes to see if anyone else joins
15:07:19 <pwhalen> Ok folks, i dont think we have enough of a quorum here today. is there anything urgent to discuss?
15:08:34 <jlinton> Not, really I have a set of patches for mozjs24, that backport the upstream 48bit VA changes
15:08:51 <jlinton> and the patches to move policykitd to mozjs24
15:09:32 <pwhalen> jlinton, excellent. could you send the details to the arm-list
15:09:34 <pwhalen> ?
15:09:45 <jlinton> Yes, once I get internal review done
15:10:03 <pwhalen> jlinton, great, thanks.
15:10:06 <cov> I've been looking into a CPU topology problem
15:10:29 <cov> clusters being reported as packages
15:11:12 <cov> not much Fedora-specific other than the hwloc package doesn't seem to inlcude the graphical version of lstopo
15:11:22 <cov> I might try to look into that
15:11:41 <jlinton> cov: That is with ACPI?
15:11:55 <cov> jlinton: yes
15:11:57 <jlinton> because the cache topology doesn't work with ACPI in mainline at the moment
15:12:19 <jlinton> AFAIK..
15:12:29 <cov> well right now MPIDR is getting stuffed into /sys/devices/system/cpu/cpu0/topology/physical_package_id
15:12:55 <jlinton> yah, which messes that up too.
15:13:41 <cov> I'm trying to figure out if there's any way to know how many packages a system has. Still need to learn relevant DT and currently staring at dmidecode output.
15:15:58 <jlinton> I've had the use/don't use SMBIOS discussion a few times with regard to using it to supplement ACPI (/proc/cpuinfo for one)
15:16:22 <jlinton> and I always come back to smbios' unreliability, and its more for informative purposes..
15:18:35 <cov> hmm
15:18:51 <cov> well if you have working alternatives, I'm keen to hear them
15:19:03 <jlinton> Yah, that is the problem..
15:19:29 <jlinton> I have a "hack" for the cache topology that makes an enducated guess and gets it right for most systems...
15:19:32 <jlinton> _most_
15:19:58 <jlinton> which is basically the same problem with the CPU topology code.
15:20:03 <deadrat> jlinton : oh, interesting.  .
15:21:25 <cov> yes, many cache topologies track the CPU topology exactly: https://www.open-mpi.org/projects/hwloc/lstopo/
15:22:43 <cov> an interesting exception I've seen are a few cases where multiple NUMA nodes exist inside a single package
15:23:31 <jlinton> Yah, I don't think the sharability of any given level really has a strong association with anything
15:23:47 <jlinton> aka L2 cluster shared vs L3 socket shared.. isn't a good rule
15:25:12 <jlinton> Especially if there are DRAM controller specific caches (aka like POWER8)
15:28:58 <cov> jlinton: I'd love to see your hack if you're able to share it
15:29:00 <pwhalen> sorry guys, im distracted with something else here.
15:33:45 <pwhalen> perhaps we can continue in channel, Im going to close the meeting now
15:34:38 <pwhalen> #endmeeting