#
✕ nick:Chainfire
#asahi · 2022-03-23 13:55
Chainfirebah, working hard on a dynamic pagesize commit for Chromium, seems Chromium devs just committed hard-coded 16kb patch instead
#asahi · 2022-03-23 10:15
ChainfireGlanzmann: as long as it loads and shows you the UI, it's good enough for my purposes, that means the allocator fix works. So yay! And thanks! Will submit to Chromium devs later today.
#asahi · 2022-03-23 09:59
ChainfireGlanzmann: could you test this Chromium build on a 4kb kernel please? https://drive.google.com/file/d/1dBpo96r7y0Fvu4lwN1ccnz-ix-Gt3cHX/view?usp=sharing
#asahi · 2022-03-22 15:24
Chainfirescratch that, brainfart
#asahi · 2022-03-22 15:24
Chainfire"<realmoot> Only wish sublime-text / jemalloc issue was fixed" ... sublime-text works for me?
#asahi · 2022-03-22 09:40
ChainfireI have dynamic Chromium patch for 4kb/16kb working (busy with cleanup), it'll break on 64kb though
#asahi · 2022-03-22 09:37
ChainfireDoes the M1 support 64kb and does anyone have a kernel for that?
#asahi · 2022-03-22 06:18
ChainfireNot for me
#asahi · 2022-03-22 06:15
Chainfireaem64[m]: so far, if you stick to Arch, pacman -Syu
#asahi · 2022-03-21 10:20
Chainfiremarcan: ahhh that makes sense. In my attempt yesterday I set it to 150% and that's when nothing happened. What I tried now and worked was 200%.
#asahi · 2022-03-21 10:05
Chainfiregsora[m]: doh, actually logging out and back in works. The screen said reboot, which I did, and changed nothing. Seems to work now :S
#asahi · 2022-03-21 10:03
Chainfiregsora[m]> yeah reboots and all that
#asahi · 2022-03-21 09:53
ChainfireAny way to get UI scaling to work? Using System Settings -> Display Configuration -> Global Scale doesn't seem to do anything
#asahi · 2022-03-21 06:47
Chainfireaxboe> Regardless of if that method was known or not, it's not committed so downstream isn't building it and is thus irrelevant. I'm not going to let perfect be the enemy of good. That being said, I am working on the dynamic patch as well. That will break on 64kb kernels, though, some more assistance from the Mark @ Chromium is needed there I think. (MacOS going 64kb would also break it, and I think PPC 64kb is already broken)
#asahi · 2022-03-20 14:34
ChainfireGlanzmann binary is here https://drive.google.com/file/d/10Ch8V7QGTc5NR-YNMv-4Kw_IMOR-nyaB/view?usp=sharing
#asahi · 2022-03-20 14:32
ChainfireI'll link you up later if you want. And submit to Chromium if it still works for 4kb users.
#asahi · 2022-03-20 14:31
Chainfireaxboe> I'm heading out right now, but it's just a few lines, I actually hardcoded to use 16kb instead of 4kb, as that *should* work on 4kb (still looking for that tester!), and that's what Chromium already did for PPC (4kb vs 64kb) as well. 64kb will not work due to max sizes in the allocator beyond my will and time to understand/fix. Built and works on Asahi 16kb for me.
#asahi · 2022-03-20 13:42
Chainfire(just extract the .tar.gz and run the chrome binary from a (local) terminal)
#asahi · 2022-03-20 13:41
ChainfireChromium with 16k pages working: https://drive.google.com/file/d/10Ch8V7QGTc5NR-YNMv-4Kw_IMOR-nyaB/view?usp=sharing ... Seems to work fine for me. Can I get someone with a 4kB kernel to test if it doesn't break, please?
#asahi · 2022-03-20 11:40
Chainfirej`ey> yes, the code says so, but either I'm completely missing something, or the PPC build would actually fail if run on trunk right now
#asahi · 2022-03-20 11:40
ChainfireSeeing as every build takes about an hour, has to be done on an x86_64, debugging this is hell, etc, I don't want to go too deep down the rabbithole.
#asahi · 2022-03-20 11:39
ChainfireI have no idea. They do use mprotect and SEGV handlers but I'm not sure to what purpose. What I do know is they use a construct they call partitions, which requires to be 4 times the page size, but partitions larger than 64kb aren't supported. I don't see a reason why they couldn't be, but it would require a quite a number of changes to logic I don't understand and don't want to invest the time to figure it out.
#asahi · 2022-03-20 11:20
ChainfireIt seems patching Chromium to support 16kB pagesize is relatively easy (disclaimer: building but not tested yet), but supporting 64kB pagesize would require a fairly large overhaul of their memory allocator.
#asahi · 2022-03-19 14:54
Chainfirejannau> the guy from the bugreport dropped it due to lack of interest and time
#asahi · 2022-03-19 14:51
Chainfiremarcan> you note Chromium has the page-size issue, so I assume someone tried to build it? Cross from x86 or on the m1?
#asahi · 2022-03-19 13:16
Chainfirewell that code I mentioned is absolutely pointless, I just like to play with optimizing it, so no biggy there, just interest.
#asahi · 2022-03-19 13:12
Chainfiremarcan> just out of 'academic' interest, which memory controller optimization isn't enabled yet?
#asahi · 2022-03-19 07:36
Chainfirechadmed> is there a call that you know of for any app on macOS to do this pagesize change for itself?
#asahi · 2022-03-19 07:36
Chainfiresorear> I wasn't confusing them, they were intended as two separate questions. The macOS hugepages variant doesn't work on M1, I was wondering if that was due to the chip or just an oversight on Apple's part.
#asahi · 2022-03-19 07:08
ChainfireRegarding pagesize: does macOS also use 16K ? Will 64K be supported as well at some point ? Do these chips actually lack hugepages support ? I think I recall a blog post from marcan saying this, but unsure if that's accurate and up-to-date.
#asahi · 2022-03-18 21:28
ChainfireThanks for the effort, people. Just installed Asahi Desktop and running great on my MBP Max 16" so far.