#asahi · 2022-06-17 09:47
sventhat's the reason why the default kernel only works on 16k for now#asahi · 2022-06-17 09:47
svenwell.. we want all the 3rd party software to fix their code#asahi · 2022-06-17 09:46
sven16k for now, i'm waiting for a series from robin murphy improving the iommu api to go in first before picking up that 4k patch again#asahi · 2022-06-17 09:26
sventhe timeline is "when it's done" as usual#asahi · 2022-06-17 09:25
svenmodern gpus don't really do 2d acceleration afaik. it's all 3d these days#asahi · 2022-06-15 14:53
svenvery strange, yes.#asahi · 2022-06-15 14:53
svenmaybe we're missing some setup somewhere or just need to increase some delays#asahi · 2022-06-15 14:53
svenand that link needs more bandwidth than all the others#asahi · 2022-06-15 14:52
svenyup. most people only have the 1G ethernet so that specific hw isn't tested as well#asahi · 2022-06-15 14:51
svenyou mentioned that you have the 10g ethernet option, didn't you?#asahi · 2022-06-15 14:50
svenit sounds like we're doing something slightly wrong when bringing up pcie#asahi · 2022-06-15 12:21
svenit wouldn't surprise me if wifi/bluetooth disappeared from the bus but ethernet is very strange#asahi · 2022-06-15 12:17
svenanything weird in dmesg? :/#asahi · 2022-06-15 12:17
svenyeah, second broadcom is bluetooth#asahi · 2022-06-15 11:06
sventhat's liekly copyright infringement and probably also violates the NDA you need to sign to get betas#asahi · 2022-06-15 10:27
svenregarding TSO: easiest start is to add a flag to m1n1 to enable it for the entire machine during boot and see what happens and if it helps box64/fex. if performance for non-x86 code drops too much we could then carry a downstream patch with a prctl#asahi · 2022-06-10 20:11
svenlooks like someone needs to continue working on that psci stuff ;)#asahi · 2022-06-10 19:19
svencan you check if dmesg contains anything else?#asahi · 2022-06-10 19:18
sventhat last line just says something return EBUSY which doesn’t help much #asahi · 2022-06-10 19:12
sventrue, that at least gives it a chance then#asahi · 2022-06-10 19:08
svenNote that kexec isn’t going to work if zfsbootmenu brings up the secondary cores (which it will if it’s really just a Linux kernel with a fancy userland that pretends to be a bootloader)#asahi · 2022-06-10 05:41
svenit’ll only work if no stage in between tries to start more cores, fixing that requires psci#asahi · 2022-06-09 18:46
svengrub won’t do any of that which will break many things#asahi · 2022-06-09 18:45
svensee https://github.com/AsahiLinux/m1n1/blob/main/src/kboot.c for all the things that have to be patched#asahi · 2022-06-09 18:45
svenloading the DT from grub is a bad idea in general#asahi · 2022-06-09 18:08
svenfor everything else, well, it’s called alpha for a reason #asahi · 2022-06-09 18:07
sventhe dtb ideally is backward and forward compatible *for things that are upstream*#asahi · 2022-06-09 18:07
svenwould also be annoying to implement #asahi · 2022-06-09 17:28
sveniirc for something small in pcie, but it’s been a while#asahi · 2022-06-09 17:28
svenI think it’s rather rare, I’ve only seen it once or twice #asahi · 2022-06-09 17:24
svencant squash anything after the PR is sent to Linus ofc#asahi · 2022-06-09 17:24
svenI’ve seen fixes squashed in as long as they’re only on -next since some maintainers don’t mind rebasing there#asahi · 2022-06-09 17:23
svenbut anyway, nice job on ADMAC :-)#asahi · 2022-06-09 17:22
svendepending on the maintainer they might even just squash the fix in #asahi · 2022-06-08 13:53
sven(and not required because we can just keep the stub at 12.3 for m1)#asahi · 2022-06-08 13:52
svenstill quite some details left to understand before we can do that though #asahi · 2022-06-08 13:52
svenand that update ramdisk is provided and signed by apple#asahi · 2022-06-08 13:52
sven*update ramdisk#asahi · 2022-06-08 13:52
sventhe update process is probably something like “put ram disk to $special-location and set some magic nvram arguments and then reboot”#asahi · 2022-06-08 13:51
sventhe gpu firmware like the rest is stored in that stub partition and loaded by iboot before it loads m1n1 fwiw #asahi · 2022-06-07 17:22
svena long weekend with bad weather would help ;)#asahi · 2022-06-07 15:13
sveniirc someone started writing a UI for the installer (which is written in python) at some point, maybe you can also help there#asahi · 2022-06-07 15:06
sventhere are also other ways to contribute except for writing code or reverse engineering hardware. we have some people who e.g. write excellent documentation or help others in here#asahi · 2022-06-07 15:04
svenis confused#asahi · 2022-06-07 15:03
svenuh, sure?#asahi · 2022-06-07 15:02
svenyeah#asahi · 2022-06-07 15:02
svenmost of the reverse engineering happens with python+m1n1 fwiw#asahi · 2022-06-07 15:01
svenif you only know python and haven't done kernel work before#asahi · 2022-06-07 15:01
svenm1n1 is like half-python half-C, but a lot of that python code is about poking MMIO so it's going to be hard to see what's going on there#asahi · 2022-06-07 13:06
sventhere's no broken pcie express implementation. it's just that the egpu drivers make assumptions that may no longer hold on apple silicon (at least iommu granularity and memory attributes)