#asahi-alt · 2024-01-17 12:57
arndI finally managed to catch up with running mainline on my mac studio, using a subset of 010-soc, 020-dart, 030-misc, 040-dwc3, 120-spmi, 170-atcphy and 200-dcp, so far looking good#asahi-alt · 2023-03-06 14:11
arndah, makes sense#asahi-alt · 2023-03-06 14:01
arndjannau: btw, I am using the latest bits/200-dcp branch now (with 6.2.0), and I can confirm that this no longer runs into the race that caused the HDMI port to remain disabled in an earlier version, so whatever you did to fix that issue must have worked#asahi-alt · 2023-03-06 13:58
arndI'm sure I made some minor mistake in the rebase that caused it, but it seemed easier to keep testing rc1 in a vm guest for the moment#asahi-alt · 2023-03-06 13:54
arndjannau: that's what I first tried, but I ran into WARN_ON(stream_map->dart->locked) and dcp didn't initialize#asahi-alt · 2023-03-06 13:20
arndI tried updating my kernel from 6.2 to 6.3-rc1, but couldn't figure out how to merge the bits/020-t6000-dart contents with what went into mainline. I'll stick with 6.2 for now#asahi-alt · 2023-02-27 07:12
arndthese are the last messages for reference, if you want to investigate: https://www.irccloud.com/pastebin/tVuI7s9V/#asahi-alt · 2023-02-27 07:09
arndjannau: following up: after I turned on the monitor this morning, it again came back with HDMI disabled, but switching to vt1 and back fixed it as it used to#asahi-alt · 2023-02-26 14:22
arndxfce4 is a little annoying by always showing me 'Display Settings' windows every time I turn the screen back on sincen it noticed the monitor config changing twice, bu there is probably not much that can be done about that in DCP#asahi-alt · 2023-02-26 14:18
arndlet me know if you want me to try a fix on top of the whole branch#asahi-alt · 2023-02-26 14:17
arndjannau: this survived all combinations of power off/on and unplug that previously crashed it. I'm happy enough with this, thanks for the quick help!#asahi-alt · 2023-02-26 14:15
arndtrying to crash it harder#asahi-alt · 2023-02-26 14:15
arndit survived the first monitor off/on cycle: https://www.irccloud.com/pastebin/9qROgGll/#asahi-alt · 2023-02-26 14:13
arndreboots into the new kernel up to the 125aeecbc011d commit#asahi-alt · 2023-02-26 14:09
arndthe 6.1 kernel had whatever was current on Dec 30#asahi-alt · 2023-02-26 13:56
arndthe source tree is v6.2 with bits/000-devicetree, bits/010-mailbox, bits/020-t6000-dart, bits/030-misc, bits/040-dwc3, bits/050-nvme, bits/110-smc, bits/120-spmi, bits/150-xhci-firmware, bits/170-atcphy and bits/200-dcp#asahi-alt · 2023-02-26 13:55
arndjannau: here is the entire 'journalctl --system' output from boot to shutdown for the session that just broke: https://pastebin.com/CMGnvK6J#asahi-alt · 2023-02-26 13:45
arndHmm, this time switching to vt1 worked, but then switching back to xfce got it into the broken state where it no longer came back. I'll reboot and try to find the log file#asahi-alt · 2023-02-26 13:40
arndit just happened after turning it off briefly using the moniotor's power switch. Can't see what I type any more now#asahi-alt · 2023-02-26 13:39
arndI tried unplugging the HDMI cable, using the monitor power button or the external power switch (which turns off my speaker etc as well), no differences#asahi-alt · 2023-02-26 13:39
arndjannau: not any more. I used to have this problem occasionally in the past, and then switching to vt1 and back solved it. With my current kernel, it seems to happen every time, and the vt switch no longer helps#asahi-alt · 2023-02-26 13:21
arndit's my own kernel config, and I only included the branches that I thought I needed, so it's possible something is issing, but everything else seems to work fine#asahi-alt · 2023-02-26 13:20
arndjannau: after upgrading from 6.1 to 6.2 with the current asahi/bits/200-dcp branch (a2482bcc06), I no longer get the HDMI to be turned back on after I turn off the monitor over night. The only kernel message I get about this is "Feb 26 13:40:21 studio kernel: apple-dcp 38bc00000.dcp: crtc_atomic_check: disconnected but no modeset". Any idea?#asahi-alt · 2023-02-17 09:03
arndas I said, it makes no difference on 64-bit, as the only change is the set of available syscalls for 32-bit tasks#asahi-alt · 2023-02-17 09:02
arndmps: do you mean you saw no problems on 64-bit applications or 32-bit applications?#asahi-alt · 2023-02-17 09:00
arndmusl currently relies on it being enabled, so you don't have a choice#asahi-alt · 2023-02-17 09:00
arndI know, he cares more about his idea that syscalls can never go away than to actually make systems work in the future#asahi-alt · 2023-02-17 08:58
arndmps: a lot of applications make direct system calls and bypass libc, so COMPAT_32BIT_TIME is meant to help you find stuff that would break#asahi-alt · 2023-02-17 08:55
arndmps: if you have a 32-bit userland and want to ensure it actually works beyond 2038, you should probably disable it and fix your libc to not call any of the deprecated syscalls, but most people don't need to worry because they will stop running 32-bit userspace before then, or can wait for someone else to fix it#asahi-alt · 2023-02-17 08:51
arnd(there are some remnants of OSF-1 compatibility in arch/alpha that use 32-bit time_t, but we don't care about those)#asahi-alt · 2023-02-17 08:51
arndmps: it is, but 64-bit architectures never defined a 32-bit time_t, so it has no effect unless you also enable CONFIG_COMPAT#asahi-alt · 2023-02-17 08:50
arndah I see, the dependency is just for the default: "default !64BIT || COMPAT", rather than "depends on"#asahi-alt · 2023-02-17 08:49
arndmps: there have been patch proposals in the past to allow compat syscalls to be called from native 64-bit user space such as qemu-user or fex, but those never made it in#asahi-alt · 2023-02-17 08:47
arndmps: CONFIG_COMPAT is not needed, without that I don't think you can even enable COMPAT_32BIT_TIME#asahi-alt · 2023-01-25 10:25
arndcy8aer: https://www.winehq.org/pipermail/wine-devel/2020-August/171857.html made it sound like 16KB pages would be a show-stopper for Wine on Linux as much as they are on MacOS#asahi-alt · 2023-01-24 09:01
arndthe raspberry-pi kernel almost works, but I had to give it a patched dtb and a custom qemu command line that misses some of the features#asahi-alt · 2023-01-24 09:01
arndarmhf bookworm is broken because of the binutils issue, and might never get fixed again, armel needs work because neither of the armel kernels work with qemu out of the box#asahi-alt · 2023-01-24 09:00
arnddebvm-create is mostly a wrapper around mmdebstrap that ends up with a single ext4 image file that debvm-run can then boot for any of the supported architectures except the two I'm interested in#asahi-alt · 2023-01-24 08:59
arndGlanzmann: I only learned about it when helmut mentioned 'debvm', which is even cooler. So "debvm-create -m armhf -r bullseye && debvm-run" boots into a virtual machine within a few minutes#asahi-alt · 2023-01-23 16:35
arndright, doesn't matter if it's qemu or running on the CPU#asahi-alt · 2023-01-23 16:33
arndyes, I was trying to use mmdebstrap to create an armhf debian image, which will run 32-bit processes in qemu#asahi-alt · 2023-01-23 11:34
arndopened https://sourceware.org/bugzilla/show_bug.cgi?id=30033 for the binutils regression#asahi-alt · 2023-01-23 10:18
arndb) debian bookworm on armel/armhf seems to contain a broken toolchain with a linker that again requires 4KB page alignment for shared objects. Older library packages and Debian Buster/Bullseye work fine with 16KB and 64KB page arm64 kernels#asahi-alt · 2023-01-23 10:16
arnda) qemu-user-{static,binfmt} on arm64 does not register /usr/bin/qemu-arm for running 32-bit binaries because it assumes the kernel can already do it#asahi-alt · 2023-01-23 10:16
arndGlanzmann: relaying a current discussion from #debian-arm here in case someone else runs into the same issue. I tried to run 'debvm' to create a bootable qemu instance of a 32-bit arm debian, this failed for two reasons:#asahi-alt · 2022-12-30 18:34
arndI don't think it was introduced in a recent mesa snapshot, since I saw the same symptom a month ago#asahi-alt · 2022-12-30 18:32
arndjannau: Makes sense. I know nothing about Xorg internals, but my best guess would be that the 'double free' that gets reported in the log is the result of an actual bug in the error handling, while the following backtrace is meaningless because of yet another problem with the new libunwind#asahi-alt · 2022-12-30 16:25
arndmps: it worked, thanks a lot!#asahi-alt · 2022-12-30 16:18
arndI can see that I have mesa packages versioned 23.0.0-202212211820 installed, which matches the files in https://thomas.glanzmann.de/asahi/pool/main/#asahi-alt · 2022-12-30 16:14
arndyes, I added his repo and did a dist-upgrade with everything that was newer than the normal debian-stable