|
Kivi posted:Considering it is a single drive single disk environment within KVM I could just bypass grub and use the Kernel boot option in Virt Manager? Maybe? I've never tried it, but it might be easier to just fix grub If you do try it, report back
|
#
?
Apr 24, 2022 17:55
|
|
|
|
| # ? May 18, 2024 02:02 |
|
|
BlankSystemDaemon posted:This seems like a bit of a mess. It was fine until systemd changed that behavior. Now it can break in slightly unexpected ways. Systemd is a net good, but it definitely still has some rough patches where lennart obviously decided some core piece of technology (in this case initrd) was going it get phased out and it hasn't been, so we are dealing with slightly flawed shims
|
|
#
?
Apr 24, 2022 17:58
|
|
|
It was a bug in SeaBIOS. Compiled latest and VM boots again.
|
|
#
?
Apr 26, 2022 18:12
|
|
|
How numa-aware is the modern Linux kernel, and is there anything that I can do to help it be more NUMA aware than forcing it with numa-ctl? I've got a whole set of software tools like https://github.com/bwa-mem2/bwa-mem2 that really would prefer to be run entirely within a single NUMA node. They bottleneck on memory bandwidth and latency, and we're scheduling them on dual-socket nodes that aren't subdivided into smaller VMs. For this specific tool, the documentation recommends limiting it to a single NUMA node like so: numactl -m 0 -C 0-27,56-83 ./bwa-mem2 index human_g1k_v37.fasta This ends up difficult to manage at scale though, because we don't know on a given machine which NUMA node is currently occupied and which is free without manual inspection. Thoughts? One blunt hammer would be to partition the nodes into two smaller VMs that are a single NUMA node each, but that'd make our scheduling less efficient because we'd have smaller nodes. Should I test more and see how much improvement we get by limiting things to a single NUMA node? Could we update the software itself to prefer to stay and allocate within one node?
|
|
#
?
Apr 26, 2022 20:38
|
|
|
Does anybody know of an alt-tab switcher for KDE that looks as much as possible like the Windows 10 one? Of all the KDE switchers I tried from the gallery, I couldn't find one that both (a) had thumbnail window previews and (b) was centered in a traditional popup (instead of taking over half/all the screen).
|
|
#
?
Apr 27, 2022 09:34
|
|
|
Does anyone have experience with BCC/BPF? I'm looking at containerizing a lot of undocumented cusyom legacy software and hoping to write a BPF program to monitor when a process starts up, what child processes it has, what files it touches, and what Unix sockets or TCP sockets it opens to talk to Middleware or service applications. Basically we have a system of code that is a rats nest of thousands of bash scripts, custom C++/java applications, and a half dozen custom middlewares all calling each other. Most of it was written 15-20 years ago with little formal documentation. We want to untangle some of the major applications and determine what they need to be containerized. I'm looking at writing a BCC tool that I could have monitor the main parent process that kicks off everything and trace downstream processes that start up without me having to run a whole lot of straces. I was wondering how disruptive the kprobes would be to applications running on the system? Would I expect to see significant degredation to applications from monitoring system calls? Ideally I see this being a background process that kicks off and starts gathering this data as we go through our usual production testing and I don't want it potentially hindering application performance. I'm also very interested in what processes are opening TCP/UDP sockets because a lot of our applications open up very short lived sockets and I haven't found a good tool that can connect a socket to pid/ppid unless you're lucky enough to catch it as the socket is currently open. I wrote a tool that uses auditd to monitor accept and connect syscalls then scrapes audit logs and hits the proc table to pull parent chains but the thrashing of the proc table is pretty rough on the system and unfortunately connect/accept calls contain protocol info buried in the struct that gets passed to them. There wasn't a good way to determine what protocols they were using. BedBuglet fucked around with this message at 11:19 on Apr 27, 2022 |
|
#
?
Apr 27, 2022 11:06
|
|
|
NihilCredo posted:Does anybody know of an alt-tab switcher for KDE that looks as much as possible like the Windows 10 one? Thumbnail Grid? Seems pretty much the same, though the thumbnails are smaller than windows. I am honestly surprised how much I like KDE Plasma. I strongly disliked KDE way back when, and for the last decade or so when I've done desktop I've used XFCE. But I gave Plasma a shot, mostly because XFCE is still far from Wayland adoption. It's really nice! OTOH the Wayland vs X11 thing still seems inconclusive. I've switched back and forth a few times and both have had problems. * X11: random loss of icons, incorrect cursors, and other small annoyances that are exactly like I remember from my last desktop linux years ago. * Wayland: graphical corruption of the WM & desktop, including a weird thing where only 1 of my two monitors could have a desktop at all. App windows drew just fine, so the monitor was still there & functional. But the desktop itself was black with no panels / widgets and right-click didn't draw a menu. Disable + enable cycle in system settings just swapped which monitor had the broken desktop. Entertaining and novel brokenness!
|
|
#
?
Apr 27, 2022 18:09
|
|
|
BedBuglet posted:Does anyone have experience with BCC/BPF? I'm looking at containerizing a lot of undocumented cusyom legacy software and hoping to write a BPF program to monitor when a process starts up, what child processes it has, what files it touches, and what Unix sockets or TCP sockets it opens to talk to Middleware or service applications. Can you not just strace the parent process with -f to grab the forks?
|
|
#
?
Apr 27, 2022 22:42
|
|
|
Mr. Crow posted:Can you not just strace the parent process with -f to grab the forks? It's more, that I need to get info on the interprocess communication, files that processes or child processes open, or sockets that get opened. Strace is fine if you know the processes you want to watch but we have a situation where service applications are spawning 80+ processes based on hardware or middleware requests from other servers and those processes are using shared memory or opening files, or communicating via Unix sockets. I'm interested in a way I can monitor system calls across all processes without knowing what processes I'm looking for and in such a way that I can have it passively gather that data because we don't always know what triggers a child process. I could strace -p on specific processes if I knew what pids to look at but there are way too many processes at a given moment and some are pretty short lived and spawned on certain conditions. I think eBPF might be better for an os wide mapping tool. BedBuglet fucked around with this message at 10:17 on Apr 28, 2022 |
|
#
?
Apr 28, 2022 03:56
|
|
|
RFC2324 posted:P sure that's default in linux? Not for years on most distros in my experience. I remember doing it in the 90s.
|
|
#
?
Apr 29, 2022 15:11
|
|
|
I have revisited my nouveau driver saga after a weeklong vacation. Yeap, still loads nouveau by default! My verified my /boot/grub/grub.cfg has the parameters to disable it in all the boot selections, but when I view the command line when I am booting, none of it is there. Outside of update-grub, is there something else I'm supposed to run?
|
|
#
?
May 2, 2022 16:43
|
|
|
Rocko Bonaparte posted:I have revisited my nouveau driver saga after a weeklong vacation. Yeap, still loads nouveau by default! If the boot params are not those that are present in grub.cfg it means it's not using it. Do you have a /boot/efi/EFI/<distro>/grub.cfg? When I update the default config params I run grub2-mkconfig -o <grub cfg file>, but it's probably different for other distros (sorry, I don't remember what distro you're using ).
|
|
#
?
May 2, 2022 18:20
|
|
|
Volguus posted:If the boot params are not those that are present in grub.cfg it means it's not using it. Do you have a /boot/efi/EFI/<distro>/grub.cfg? When I update the default config params I run grub2-mkconfig -o <grub cfg file>, but it's probably different for other distros (sorry, I don't remember what distro you're using ). I'm on kubuntu so I have an ubuntu folder there with a grub.cfg in it: code:
|
|
#
?
May 2, 2022 18:43
|
|
|
Yeah I dislike working with the mapper ids for that reason - but it does mean that you can mostly unplug and re-plug disks without caring about which disk is connected where, which is a problem I have multiple times per decade. All the relevant partitions are typically mounted, and mounted using the UUID, so you can do mount | grep fe5c3484 and see what shows up?
|
|
#
?
May 2, 2022 22:09
|
|
|
Rocko Bonaparte posted:I'm on kubuntu so I have an ubuntu folder there with a grub.cfg in it: Do you have /boot encrypted as well? Is that /dev/sda2 mounted as /boot? Seems not, but I'm just asking. That's ... a ... head-scratcher. Haha, who knows what gub.cfg from where on the drive(s) is being used . If I were you I'd hunt for it since that cfg is very precious.
|
|
#
?
May 2, 2022 22:57
|
|
|
|
|
#
?
May 2, 2022 23:42
|
|
|
Rocko Bonaparte posted:I have revisited my nouveau driver saga after a weeklong vacation. Yeap, still loads nouveau by default! Have you blacklisted nouveau? Things have been fine on my systems for quite a while, but I do remember having to go in and manually blacklist nouveau to make things work a few years ago.
|
|
#
?
May 2, 2022 23:54
|
|
|
I'm going from memory, but I think if grub-install thinks it can't read /boot then it appends grub.cfg to the end of the bootloader itself? So you could try grub-install /dev/nvme0n1, but who knows that could break your whole system. Edit: Well that wouldn't quite make sense because it still has to load the kernel/initrd from /boot. You could still try it though .... Edit edit: This shouldn't make a difference either but you can also disable Secure Boot in the BIOS and then "apt-get remove grub-efi-amd64-signed shim-signed; apt-get install grub-efi-amd64-bin". When you use Secure Boot, shimx64.efi is actually the bootloader that gets loaded by EFI which then goes and loads grubx64.efi. But I've ran into werid issues with UEFI/Shim/GRUB that's alleviated by just using GRUB so that's another shot-in-the-dark option. ExcessBLarg! fucked around with this message at 00:30 on May 3, 2022 |
|
#
?
May 3, 2022 00:23
|
|
|
Volguus posted:Do you have /boot encrypted as well? Is that /dev/sda2 mounted as /boot? Seems not, but I'm just asking. That's ... a ... head-scratcher. Haha, who knows what gub.cfg from where on the drive(s) is being used . If I were you I'd hunt for it since that cfg is very precious. Okay I feel better that I'm just that stupid. I couldn't figure it out. I haven't given up though. It's gotta be somewhere. CaptainSarcastic posted:Have you blacklisted nouveau? Things have been fine on my systems for quite a while, but I do remember having to go in and manually blacklist nouveau to make things work a few years ago.  lol see the previous page and you can wade through how I beat that dead horse! It's soooo blacklisted. It should never be able to work in this town ever again, but the turd just won't flush.It's getting to the point where it might just be easier for me to rebuild the kernel with nouveau completely stripped from it. I've already stopped automatic kernel updates because it breaks the closed-source nvidia drivers, so I can just freeze on some kernel for awhile. I can't blacklist the driver from grub worth a drat but I can build kernels all day. I mean, they do take all day...
|
|
#
?
May 3, 2022 00:50
|
|
|
Rocko Bonaparte posted:
Are module blacklists maybe not global? With the fact that you have multiple kernels, that makes me wonder if your blacklist is not hitting the kernel you're running. So, what about this idea: can you switch to tty3, kill X, and manually unload the nouveau driver module? And then successfully restart X with the nvidia driver? Like, this is not a solution the problem but when I'm at the frustration point you've reached I like to get some confirmation that the thing I want can happen in any way possible. ... On my newbie adventure, I am now thinking that Wayland actually feels quite good & pretty stable... once I hammered all the apps into running Wayland-native. And stopped using ones that are x11-only. At least, for stuff like web browsers that are open all the time. Could be apophenia, but before I did that it felt like the system had a running countdown until something terrible happened, such that it needed a reboot.
|
|
#
?
May 3, 2022 04:37
|
|
|
Rocko Bonaparte posted:Okay I feel better that I'm just that stupid. I couldn't figure it out. I haven't given up though. It's gotta be somewhere. Oh, one more thing, now I remembered: On fedora they have enabled BLSCFG some time ago, which moves the linux entries away from grub.cfg (not quite sure the reasoning behind it but whatever). I have in /etc/default/grub: code:code:code:But that does mean that I don't have menu entries in there for the kernels, just code:
|
|
#
?
May 3, 2022 17:46
|
|
|
I have: Fedora, an Nvidia card, and Flatpak Steam. Sometimes the rpmfusion nvidia driver updates before the flatpak equivalent. This causes games to Not Work until the flatpak is updated (or I rollback I guess). I started to think up ways I could maybe pin the system rpm to whatever the latest flatpak has. But: - Does this problem just go away after I'm able to get an AMD card - Is the real answer to just not mash update so often you dummy
|
|
#
?
May 3, 2022 18:51
|
|
|
When I was first investigating Flatpak and saw that they were shipping their own private copies of glibc and mesa, I laughed and resolved to never touch them because clearly the entire concept was made by idiots.
|
|
#
?
May 3, 2022 19:37
|
|
|
pseudorandom name posted:When I was first investigating Flatpak and saw that they were shipping their own private copies of glibc and mesa, I laughed and resolved to never touch them because clearly the entire concept was made by idiots.
|
|
#
?
May 3, 2022 19:56
|
|
|
Chilled Milk posted:I have: Fedora, an Nvidia card, and Flatpak Steam. Sometimes the rpmfusion nvidia driver updates before the flatpak equivalent. This causes games to Not Work until the flatpak is updated (or I rollback I guess). I started to think up ways I could maybe pin the system rpm to whatever the latest flatpak has. But: Why don't you install Steam from rpmfusion too instead of using the flatpak?
|
|
#
?
May 3, 2022 20:03
|
|
|
pseudorandom name posted:When I was first investigating Flatpak and saw that they were shipping their own private copies of glibc and mesa, I laughed and resolved to never touch them because clearly the entire concept was made by idiots. The first is that flatpak/snap make a bit more sense in the context of relatively-stable kernel/driver interfaces. In this context the X11 interface is very stable, and the DRM/DRI one is reasonably so. But for things like GPU, video encoding/decoding, where the interfaces are in flux and matched kernel/userspace blobs are common, yeah, it's a mess. Maybe snaps make a bit more sense here since those aren't wholly application-oriented, whereas Flatpaks--that's the whole point. Might as well just think of the Flatpak runtimes as yet another Linux distribution to build against. The second is that duplication is annoying. I think it was a common problem a couple of years ago that any snap that linked against libGL would include LLVM for shader compilation or whatever, and since that wasn't part of the core this was eating up significant space for each application that uses OpenGL. I think Canonical resolved this by introducing the content interface, but again that makes Ubuntu snap environments--in addition to base Ubuntu--yet another Linux distribution to build against. I guess the "benefit" to all this is that it's trivial to have Ubuntu snaps built against 18.04, 20.04, 22.04 all installed simultaneously so you don't have to worry immediately about application compatibility when upgrading your desktop, unlike with PPAs in the past. Still, where these things depend on in-flux kernel interfaces who knows! I was going to give the Dolphin snap a try at some point see if it's easier for me than having to compile it from source. I don't know if I'm looking forward to it. Steam Deck couldn't come soon enough.
|
|
#
?
May 3, 2022 20:24
|
|
|
Kassad posted:Why don't you install Steam from rpmfusion too instead of using the flatpak? At one point in time, possibly no longer true, the flatpak version had fewer issues. Especially around proton. This problem in particular is largely fake since it's never more than like half a day behind at most, when it happens at all, which is maybe a couple times a year. Just one of those little papercuts. As it is, this is the only issue I have with it, other than very slow start up. Which is something I tried switching to the rpmfusion package for but it wasn't any better. Haven't tried a different distro recently to see if it's a Fedora thing but I kind of doubt it.
|
|
#
?
May 3, 2022 20:42
|
|
|
Volguus posted:Oh, one more thing, now I remembered: On fedora they have enabled BLSCFG some time ago, which moves the linux entries away from grub.cfg (not quite sure the reasoning behind it but whatever). I have in /etc/default/grub: I'm not crazy, right? This looks a hundred times more complicated, in a way that only makes life easier for massive, automated, untouched-by-human-hands setups, at the cost of making things outright labyrinthine if you're an end user who has to touch it?
|
|
#
?
May 3, 2022 20:45
|
|
|
pseudorandom name posted:When I was first investigating Flatpak and saw that they were shipping their own private copies of glibc and mesa, I laughed and resolved to never touch them because clearly the entire concept was made by idiots. Yea Linux is well known for it's stable and consistent graphics stack across distros
|
|
#
?
May 3, 2022 21:35
|
|
|
Computer viking posted:I'm not crazy, right? It seems way more straightforward to me -- one .cfg per kernel / bootable OS is a far better system than a monolithic file that renders your system unbootable if you make a mistake. "hosed up my grub, can't boot" is not exactly a rare thing to hear.
|
|
#
?
May 3, 2022 21:52
|
|
|
Mr. Crow posted:Yea Linux is well known for it's stable and consistent graphics stack across distros Shipping an obsolete version of Mesa that doesn't have drivers for your user's hardware doesn't fix any problems.
|
|
#
?
May 3, 2022 22:08
|
|
|
Computer viking posted:I'm not crazy, right? It is more complicated for humans. It's is a lot easier for a script. Install new kernel, comes with the configuration file, put it there, done. Remove old kernel, remove conf file. Done. Klyith posted:It seems way more straightforward to me -- one .cfg per kernel / bootable OS is a far better system than a monolithic file that renders your system unbootable if you make a mistake. "hosed up my grub, can't boot" is not exactly a rare thing to hear. From a human, no, it's not a rare thing to hear. From a tested script ... it better be. But I can see why it's easier to write/test a script that writes that conf file.
|
|
#
?
May 3, 2022 23:56
|
|
|
pseudorandom name posted:When I was first investigating Flatpak and saw that they were shipping their own private copies of glibc and mesa, I laughed and resolved to never touch them because clearly the entire concept was made by idiots. The only two places I've seen it used with any seriousness are in car head units and video broadcast/network appliances. The former lean on it for managing the customization process for each model/market (and OTA update resiliency), the latter use it to deploy/license a cursed matryoshka doll of hosed up linuxes that act as a singular piece of kit. Chilled Milk posted:ways I could maybe pin the system rpm to whatever the latest flatpak has Flatpak uses ostree, which is kind of like git but specifically for linux filesystem trees. You should be able to use that to have it check out the specific version you want from history. As a bonus, if you have to do any additional configuration, you should be able to save it and easily re-apply it to future versions. If you have a link to the flatpak handy I could poke at it, if you already know specifically what version/release you want and can point to it, that would make things easier E: I guess you need to go in the opposite direction? That should be more straightforward New Zealand can eat me fucked around with this message at 05:17 on May 4, 2022 |
|
#
?
May 4, 2022 05:12
|
|
|
Chilled Milk posted:I have: Fedora, an Nvidia card, and Flatpak Steam. Sometimes the rpmfusion nvidia driver updates before the flatpak equivalent. This causes games to Not Work until the flatpak is updated (or I rollback I guess). I started to think up ways I could maybe pin the system rpm to whatever the latest flatpak has. But: I have had Silverblue, an Nvidia card, and Flatpak Steam for a few months now, and I run `flatpak upgrade -y && rpm-ostree upgrade` automatically every day. I've never had this problem happen, I don't know if it's because Silverblue integrates the layering of flatpak apps and system upgrades or simply because it's rare enough. Sometimes the dependencies aren't solvable so `rpm-ostree upgrade` refuses to run for a day or so until the upstreams have updated, I don't remember it happening with NVidia deps specifically (usually it's openssl or something) but that might point to #1? If you're already comfortable with ostree I can definitely give a glowing recommendation to Silverblue though.
|
|
#
?
May 4, 2022 09:51
|
|
|
i just run debian on my gamebox now, have been since september, haven't had an issue yet. steam works great, the games i play run way better with dxvk i am using an amd gpu, though with a distro as 
|
|
#
?
May 4, 2022 11:02
|
|
|
pseudorandom name posted:Shipping an obsolete version of Mesa that doesn't have drivers for your user's hardware doesn't fix any problems. Yea that would be a shame if they weren't a full major version ahead of the most popular distros releases wouldn't it. Linux is famously known for maintaining consistent versions of core packages across distros I can't understand why a package manager whose core purpose is to provide a consistent application environment would do that, it's just bonkers! This is some "huff huff SYSTEMD!!!  " level of argumentation.
|
|
#
?
May 4, 2022 16:18
|
|
|
Mr. Crow posted:This is some "huff huff SYSTEMD!!! Is it that you can compile against the latest runtimes and ensure users can/will/are forced to upgrade to them?
|
|
#
?
May 4, 2022 16:33
|
|
|
My KDE desktop on Fedora 35 has stopped showing background processes in the system tray (mostly I miss Zoom and Synergy being there). What did I do?
|
|
#
?
May 4, 2022 19:20
|
|
|
ExcessBLarg! posted:What's the fundamental difference between compiling an application for Flatpak and Snap distribution, vs. compiling the same application for Fedora and Ubuntu packages? I'm not 100% following your question and I'm not a maintainer, nor do i care that much about flatpak (snap is a dog turd though), so your guess is as good as mine; yes I imagine its so you can have more control of your version dependencies. Flatpak at least (i think snap does also) also maintains (or provides, depending on age) several branches of the runtimes so if your application needs older versions it'll still work, unlike what frequently happens with third party packages from a distro. I'm not sold on flatpak as the be-all-end-all but neither do I think its useless or "designed by idiots". Obviously if these were easy or solved problems, 2022 would be the year of the linux desktop.
|
|
#
?
May 4, 2022 19:23
|
|
|
|
| # ? May 18, 2024 02:02 |
|
|
Part of the argument for Flatpak is that since runtimes are available for all major distributions, you can target Flatpak for distribution and "run everywhere". The problem is that it's not just Flatpak, you also have Canonical shoving Snap in the popularity contest, and I guess there's also AppImage. So realistically, if most out-of-distribution desktop applications were targeting two Linux platforms for release (latest Fedora/RH, latest Ubuntu) then the fundamental situation is still the same--you still have to target more than one delivery platform. There is benefit to not having to provide updates packages in sync with distribution releases though, or having to support multiple versions of the same distribution at once. I mean, I'm sure Flatpak/Snap both exist to solve a real packaging problem, I just don't know how well they're both achieving their goals, and how much their co-existence hurts them too.
|
|
#
?
May 4, 2022 21:04
|
|