09:23 <robmur01> future log-reading chewitt: it's not about the GPU version, it's about the integration - that first patch implies that the Exynos interconnect doesn't like the GPU emitting snoop signals. If it's a non-coherent interconnect then it really shouldn't care, but I guess these things happen

09:24 <robmur01> the others are just about v4l2 being fundamentally rubbish

09:25 <HdkR> Oh, I'll have a slide on that if my XDC talk is accepted :P

12:06 <chewitt> robmur01 thanks :)

14:12 <cphealy> What is the correct name for a single core G52? Is it G52 MC1 or G52 MP1?

14:21 <robmur01> yes :P

14:24 <robmur01> (I think the "MC" nomenclature began for clarity largely because the previous generation so horribly abused "MP" to represent things other than core count)

14:26 <cphealy> robmur01: does this mean that single core G52 = G52 MC1 and dual core G52 = G52 MC2?

14:27 <robmur01> yes, unlike G51/G31 where MPn has floor(n/2) actual cores

14:30 <cphealy> Perfect, thanks!

16:08 <macc24> robmur01: so G31 MP2 == G31 MP1?

16:10 <robmur01> no, the "MP2" core is fatter with about twice as much throughput

16:27 <robmur01> see the block diagram - https://developer.arm.com/Processors/Mali-G31 - denoting the second EE as optional

16:39 <alyssa> My OpenCL CTS runs are causing a kernel oops >.>

16:43 <alyssa> looks shrinker related

16:43 <alyssa> joy

17:10 <robmur01> with or without Dmitry's patches?

18:00 <alyssa> that was without, then I tried with them and the whole world broke so we'll see what I botched

18:06 <alyssa> machine might've just been having a Moment

18:21 <alyssa> seems better now \s/

18:36 <alyssa> robmur01: with!

18:36 <alyssa> (and without)

18:36 <alyssa> maybe not shrinker related let's see

18:36 <alyssa> https://rosenzweig.io/splat.txt

18:39 <alyssa> and then the OOM repeaer kicks in

18:40 <alyssa> that the WARN is drm_gem_shmem_get_pages_sgt returning an error, apparently

19:45 <robmur01> oh dear, that doesn't look good, 'specially the kernel faults with the PC not in the kernel :/

19:50 <robmur01> -ENOMEM right before OOM kicks in seems not unreasonable, though

19:50 <alyssa> robmur01: Did I do a good? :-D

19:55 <robmur01> if you somehow got Rust to nonsensically break the kernel, I'd say yes! :D

19:55 <alyssa> don't drag Rust into this :-p

20:00 <robmur01> what's odd is that the ESR reflects a regular userspace page fault, so why does the kernel think *it's* faulted?

20:03 <alyssa> https://rosenzweig.io/splat2.txt

20:03 <alyssa> in the immortal words of Britney Spears

20:03 <alyssa> oops! I did it again

20:07 <HdkR> yea yea yea yea yea.

20:11 <alyssa> so the shrinker takes a lock, which means it can't run in a atomic context

20:11 <alyssa> ...why is that an atomic context, though?

20:12 <alyssa> maybe there was a spinlock acquired in here?

20:17 <robmur01> ooh, lovely backtraces... I'll take a closer look tomorrow

20:23 <anarsoul> why the heck it's atomic if it's coming from userspace?

20:24 <anarsoul> something disable preemption?

20:24 <anarsoul> *disabled