IRC channel logs

<PotentialUser-7>Can someone help me diagnose wifi disconnects I am having? It seems to be related to hurd as I have no issues on Guix's livecd which uses linux-libre

<PotentialUser-7>I'm using networkmanager & dhclient to connect, but every 2-5 minutes i have to run dhclient again to restore connection. my mac address changes each time but these seems to be normal behaviour of network manager. i can't seem to find much documentation for hurd, the fixes i've found for my network card require creating a ath9k.conf file in

<PotentialUser-7>/etc/modprobe.d/ does this work the same in a hurd system?

<solid_black>damo22: I don't think it's a requirement to use Docker / a Docker image

<solid_black>they use Docker on GNU/Linux, fair enough

<solid_black>obviously they don't on macOS -- Docker itself can run on macOS, by running a GNU/Linux VM, but that'd defy the point of doing CI on macOS

<solid_black>please see https://gitlab.gnome.org/GNOME/glib/-/merge_requests/3503

<solid_black>and we need to start by talking to the glib/gtk people about this

<solid_black>as in, join their chat, or open an issue on the tracker, and tell them that we'd like to offer a CI runner, and ask them how they'd like this to happen

<solid_black>(I could talk to them about it, if you'd like me to, or you or youpi could do it, or all of us)

<solid_black>PotentialUser-7: I haven't followed the discussion, but there are no "kernel modules" in the Linux sense on the Hurd, so modprobe doesn't make sense

<solid_black>also /me needs to find & watch that fosdem talk that youpi keeps mentioning :)

<PotentialUser-7>solid_black okay that makes sense, are there log files I can look at to determine what is going on with my wifi card? I'm not really sure where to go from here but I'd rather avoid changing to linux-libre kernel if I can avoid it

<solid_black>I've never tried a Wi-Fi card on the Hurd tbh, I had no idea that it even worked

<solid_black>how are you driving it? netdde? some rump thing?

<solid_black>as always, I would expect that a million fixes has gone into the Linux driver since it has been imported :(

<PotentialUser-7>solid_black i could be misunderstanding, I read a post saying GUIX deprecated linux kernel and defaults to hurd and I am using pretty standard GUIX config so I assume I am using Hurd. My uname-r shows 6.0.10-gnu.

<solid_black>that post was an April 1st :)

<solid_black>6.0.10-gnu is linux-libre

<PotentialUser-7>*facepalm*

<PotentialUser-7>i really need to check the dates of posts more often

<PotentialUser-7>thanks for clearing up my confusion lmao

<youpi>solid_black: feel free to discuss with gtk/glib people about CI, thanks!

<youpi>the hurd does not have support for wifi

<solid_black>Wayland tests all pass 🎉️

<almuhs>i've just created this code to check which cpu execute each thread, creating 3 pthread and calling to CPUID to extract the APIC ID.

<almuhs> https://gist.github.com/AlmuHS/69fe3962cbe637c3d47f7ed7b7340139

<almuhs>after run this Debian GNU/Hurd and an smp gnumach, with -smp 4, i check that the threads are executing in many cpus

<almuhs>most times, the threads are using cpu 0, but sometimes i see other cpus. This is very noticeable reducing the time before each check

<almuhs> https://pasteboard.co/TmGDqnUWnHnb.jpg

<almuhs>with -smp 8 and 16 pthread units

<almuhs>the number indicates the APIC ID of the cpu which execute the code

<azert>Hello

<azert>almuhs: try replacing the usleep with a tight loop

<almuhs>ok. But the point is that the scheduler is able to distribute threads between all cpus, although most of the time are assigned to cpu 0

<azert>void __attribute__ ((noinline)) busy_loop(unsigned max) {

<azert> for (unsigned i = 0; i < max; i++) {

<azert> __asm__ volatile("" : "+g" (i) : :);

<azert> }

<azert>}

<azert>It’s normal since everyone sleeps

<azert>Make them do work and they’ll distribute

<almuhs>the ASM is a call to CPUID, to get the APIC ID

<azert>That’s all right

<almuhs>it's the best way to be sure that which cpu are executing this thread

<azert>i am arguing for replacing the sleep with a call to busy_loop

<almuhs>yes, i know

<azert>Buy

<azert>but of course you will have to try different max value

<almuhs>what max?

<azert>The one in busy_loop

<almuhs>i can use gettimeofaday() or similar

<almuhs>if i want to be precise

<azert>Anything that syscalls is bad since it has the potential to call in the scheduler

<azert>You could run some molecular dynamics code or similar

<almuhs>ok, then i will put a simple while loop. But the counter must be very very big

<azert>Yes the counter must be bif

<azert>And you need to check it doesn’t get optimized our

<almuhs>after replacing usleep with a loop, the concurrency increase a lot https://pasteboard.co/4IR2Sq0gbdsq.jpg

<almuhs>thanks azert

<almuhs>damo22: now we can sure that all cpus are working, and the threads are distributed between all of them

<almuhs>as a interesting data, /proc doesn't distinct the pthreads of my test process

<almuhs>and its stat file shows cpu 0

<almuhs>i've just check that "top -H" doesn't shows the threads in Hurd

<almuhs>and "ps" command lacks of most output options, like "lastcpu"

IRC channel logs

2023-10-08.log