#asahi-dev · 2022-04-26 15:37
maz_jannau_: that's what the PCIe driver does.#asahi-dev · 2022-04-20 09:56
mazj`ey: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/include/linux/compiler.h#n52#asahi-dev · 2022-04-19 18:32
mazsven: hopefully there will be less opposition this time around. I remember looking at the Corellium hack, and it didn't seem to have a massively different programming interface.#asahi-dev · 2022-04-18 11:02
mazunder some circumstances, yes. robh fixed it recently, but there may be other occurrences. here, I doubt it did spit anything wrong, as my old email address never got any close to the M1 PMU...#asahi-dev · 2022-04-18 10:59
mazI'll dig out that from the list and reply.#asahi-dev · 2022-04-18 10:58
mazwhat is the message-id?#asahi-dev · 2022-04-18 10:57
mazif you don't cc me, there is exactly zero chances that I'll see it.#asahi-dev · 2022-04-18 10:57
mazwhich email?#asahi-dev · 2022-04-18 10:56
mazYichaoYu[m]: maybe I wan't clear enough: IMO there is nothing more to add to the kernel code. what needs doing is the the JSON file that describes the events in userspace.#asahi-dev · 2022-04-18 08:26
mazYichaoYu[m]: not sure what you expect. if you have something to upstream, post the patches to the list, cc'ing the maintainers and myself.#asahi-dev · 2022-04-13 11:22
mazI'm definitely likely to drop series if the author only posts once per cycle, and I'm sure I'm not the only one.#asahi-dev · 2022-04-13 11:21
mazthe volume doesn't really change here. it is about posting regularly new updates to the series, making sure it is always fresh in the reviewer's head.#asahi-dev · 2022-04-13 11:15
mazthe main thing is to keep the versions coming once a week. once you reach the point where you have a couple of consequtive versions without any significant comment, the maintainer is likely to pick it up.#asahi-dev · 2022-04-11 13:53
mazj`ey: FWIW, I get 3:20 with LLVM-11 on the same defconfig, v5.18-rc2.#asahi-dev · 2022-04-11 12:13
mazj`ey: landing expected first half of May...#asahi-dev · 2022-04-11 12:06
maz8GB is more than enough. the compiler used may have a significant impact though (I use GCC 10.2.1)#asahi-dev · 2022-04-11 12:04
mazj`ey: 5 minutes for 'make vmlinux'? that's odd. it takes half that time on my mini (make -j9 vmlinux).#asahi-dev · 2022-04-07 16:37
mazit could be that we need to tell the root port to generate that event on LINK_*.#asahi-dev · 2022-04-07 16:36
mazsven: triggering a PME event on the PCI side should result in a scan.#asahi-dev · 2022-04-07 13:57
mazkettenis: if the system is always coherent (which is likely since the CPUs implement FWB), then the CMOs may well be implemented as NOPs.#asahi-dev · 2022-04-05 16:06
mazarnd: the debug port is the one closer to the Ethernet port.#asahi-dev · 2022-04-05 07:38
mazah, here you go! :D#asahi-dev · 2022-04-05 07:38
mazthere is also a bunch of m1n1 fixes for that, but I can't find them anymore (must have closed the tab)#asahi-dev · 2022-04-05 07:37
mazarnd: jannau has done some initial bring-up work: https://github.com/jannau/linux/tree/asahi-t6002#asahi-dev · 2022-03-30 15:21
mazit would definitely be useful to have that information in the userspace tool. but if you are doing blind testing, please make your tests publicly available.#asahi-dev · 2022-03-30 15:19
mazyou're welcome to do it at your own risk.#asahi-dev · 2022-03-30 15:18
mazYichaoYu[m]: I'm not a lawyer, and I'm not going to give any advise on whether taking this information and dumping it in the perf tool is legal or not. personally, I'm not touching ot.#asahi-dev · 2022-03-30 14:57
mazYichaoYu[m]: no, this really is a generic question about asymmetric systems, specially about ARM big-little systems (which the M1 is an instance of).#asahi-dev · 2022-03-30 14:32
mazYichaoYu[m]: so am I. this really is the wrong forum for discussion, so please post on the list and Cc me and the other ARM perf maintainers.\#asahi-dev · 2022-03-30 13:56
mazagain: the kernel side is *done*.#asahi-dev · 2022-03-30 13:56
mazhere you go: https://paste.debian.net/1236143/#asahi-dev · 2022-03-30 13:54
mazbecause you stubbornly refuse to pass the command line argument that would make it work.#asahi-dev · 2022-03-30 13:39
mazgood luck mapping those. even better luck finding an accurate architectural description that spans x86 (which the canonical events originate from) and any arm64 CPU.#asahi-dev · 2022-03-30 13:35
mazmostly because there is no way you can map the events that current CPU implements onto what the in-kernel perf consider as "canonical" events.#asahi-dev · 2022-03-30 13:34
mazmostly, yes.#asahi-dev · 2022-03-30 13:33
mazwe can have that conversation on the list if you want.#asahi-dev · 2022-03-30 13:33
mazuserspace is the right place for this, just like any other PMU.#asahi-dev · 2022-03-30 13:32
mazbecause this is kernel bloat that we don't need.#asahi-dev · 2022-03-30 13:32
mazyou don't specify a PMU, you get a random result. end of story.#asahi-dev · 2022-03-30 13:31
mazyou don't need to use raw events if you describe things to the userspace tool. perf will convert things for you, just like with any other CPUs on this side of the universe.#asahi-dev · 2022-03-30 13:30
mazagain: if you describe the event pertaining to the P core, perf will use the P core event.#asahi-dev · 2022-03-30 13:29
mazyou don't put *ANYTHING* in the kernel driver at all. you put the information in the userspace tool.#asahi-dev · 2022-03-30 13:28
mazif you pin the thread to a single core and use the PMU that matches that core, you'll get sensible results. if you don't, well...#asahi-dev · 2022-03-30 13:27
mazperf prints things that make sense to me, just as it does on any asymmetric system.#asahi-dev · 2022-03-30 13:26
mazyou need to convert the magic (and copyrighted) apple file into something that perf can ingest *at build time*.#asahi-dev · 2022-03-30 13:23
mazYichaoYu[m]: read again. the driver supports all 256 events. m1_pmu_events only describes events that have some special affinity.#asahi-dev · 2022-03-30 13:16
mazthe driver supports all the possible events. you just need to convert that file into the JSON format that the perf tool uses, and you'll get all the events with a nice description.#asahi-dev · 2022-03-30 13:15
mazYichaoYu[m]: (catching up) counting cycles (or any other event) across CPU types makes no sense. it's like adding apples and oranges.#asahi-dev · 2022-03-30 11:06
mazno, the default should be something that reflects reality.,