about summary refs log tree commit diff
path: root/machines/aszlig
Commit message (Collapse)AuthorAgeFilesLines
* machines: Remove tyree and brawndoaszlig2023-09-142-126/+0
| | | | | | | | | | | | | The former no longer exists and the latter has a new hardware and it's self-managed by the owner now (which is always the best IMHO). With both machines gone, we no longer need the "managed" profile and the BFQ module that was used for the old hardware of Brawndo was no longer used anywhere else. Same with the T100HA module, since Tyree was an ASUS T100HA and apart from the module being probably very much outdated we don't have another such hardware in Vuizvui. Signed-off-by: aszlig <aszlig@nix.build>
* dnyarri: Fix eval error for gpodder attributeaszlig2023-06-021-1/+1
| | | | | | | | Regression introduced by f72625079cc9cea8abe06d1a29b10530da538c15. I accidentally forgot about the vuizvui namespace in pkgs. Signed-off-by: aszlig <aszlig@nix.build>
* dnyarri: Switch to gpodder from Gitaszlig2023-06-021-5/+1
| | | | | | | | | I already had the latest Git version laying around in my Nix user profile, so it's battle-tested enough. The main reason why I did this is because it has much better support for yt-dlp, which I prefer to youtube-dl. Signed-off-by: aszlig <aszlig@nix.build>
* slylandro: Remove hidpi specific settings for nowaszlig2023-04-091-1/+0
| | | | | | | | | | | | | | | | | This reverts commit c88fd9eaa12c8a3c06502b09c2056d3c91421952. The hardware.video.hidpi.enable option was removed a while ago[1] because it's not clear what a single boolean option should mean, so it doesn't make sense anymore to make any of our options depend on it. Forthermore, I'm experimenting with different Wayland compositors at the moment and most of the stuff that I did here is for Xorg. I expect most of the stuff to be gone when I'm settled with a Wayland setup that works for me. [1]: https://github.com/NixOS/nixpkgs/pull/222689 Signed-off-by: aszlig <aszlig@nix.build>
* slylandro: Switch to PipeWireaszlig2022-09-261-0/+11
| | | | | | | | | I've been testing PipeWire on this machine for quite a while and it works pretty well so far. My goal is to switch all my machines to it, but since Dnyarri currently has a pretty complicated audio setup I decided to not make it the default for the workstation profile yet. Signed-off-by: aszlig <aszlig@nix.build>
* slylandro: Switch location provider to geoclue2aszlig2022-09-261-0/+2
| | | | | | | | | I do have location.latitude and location.longitude set in my workstation profile, but since Slylandro is a laptop that frequently changes location it makes sense to automate this in a way that I don't need to manually set the location every time I switch places. Signed-off-by: aszlig <aszlig@nix.build>
* slylandro: Add light commandaszlig2022-09-261-0/+2
| | | | | | | | | | | | While I do use redshift to decrease the brightness at night, I also tend to sometimes use the laptop when barely awake. Having a convenient CLI makes it easier to script, eg. "mpv somefile; light -S 0". The reason why I did not add this to my generic workstation profile is because my workstations are usually placed in a way that I need to be awake enough to approach them. Signed-off-by: aszlig <aszlig@nix.build>
* slylandro: Add libinput config for touchpadaszlig2022-09-261-0/+6
| | | | | | | | | | | | | | I never got really used to the "buttonareas" click method and while "clickfinger" does have some disadvantages (eg. dragging is much more difficult without drag lock), in most other cases I find the clickfinger method a lot more convenient, since it usually only requires one hand to operate. Since we have clickfinger enabled, I also don't see a point in enabling middle click emulation since three fingers already count for a real middle click. Signed-off-by: aszlig <aszlig@nix.build>
* machines/slylandro: Remove tuxedo-keyboard moduleaszlig2022-09-081-1/+0
| | | | | | | | | This actually consists of several modules which are of poor code quality, none of them however are essential for the TUXEDO Pulse 15 because its keyboard backlight is single color so let's drop them since they just taint the kernel for no reason. Signed-off-by: aszlig <aszlig@nix.build>
* slylandro: Add quirk for disabling D3cold on NVMeaszlig2022-08-261-0/+1
| | | | | | | | | | | | | | | | | | | | | | After resume from suspend the NVMe does not wake up again when the device was put into D3cold. This is something that is worked around by TUXEDO Tomte[1] via udev rules. However, I personally don't like this approach and it can lead to race conditions when we're going into suspend before udev is initialised. Interestingly, the device does even go into NPSS via APST, but if changing to D3cold while APST is enabled, the device does not wake up again. Right now I just added a new quirk to disable D3cold during device probe for now, but we could maybe find a better workaround eg. by disabling APST before D3cold and re-enabling it again. Not sure whether this is feasible, but since I have limited time right now I can't dig more into this. [1]: https://github.com/tuxedocomputers/tuxedo-tomte/commit/2c8d71170868a2663705fbea6ac150eecb96e6ce Signed-off-by: aszlig <aszlig@nix.build>
* machines + modules: Fix Nix option definitionsaszlig2022-08-155-5/+5
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Most of the options in nix.conf are now exposed as a submodule with a freeform type and since that change[1] got introduced, we get a bunch of warnings during machine evaluation: trace: warning: The option `nix.useSandbox' defined in `...' has been renamed to `nix.settings.sandbox'. trace: warning: The option `nix.maxJobs' defined in `...' has been renamed to `nix.settings.max-jobs'. trace: warning: The option `nix.buildCores' defined in `...' has been renamed to `nix.settings.cores'. To shut them up, I went through all machines and modules and renamed the remaining options that were not renamed back then when @devhell did some renames in a0297bf921399c3243dcca99626d8697f0735abe. This was done by looking through the output of: $ git grep -A 10 '\<nix\(\.\| *=\)' machines modules After that I tested the contents of the nix.conf of all the machines against the changes this commit introduced via the following command: $ nix-build --no-out-link -E ' with import <nixpkgs/lib>; map (m: m.eval.config.environment.etc."nix/nix.conf".source) (collect (m: m ? eval) (import ./machines)) ' I've sorted the resulting nix.conf files and diffed on that result and the only difference that showed up was the following: allowed-users = * -auto-optimise-store = false auto-optimise-store = true builders-use-substitutes = true cores = 0 This is because the previous way to generate the config was by concatenating strings whereas the new way works on an attribute set, so we get deduplication by design. [1]: https://github.com/NixOS/nixpkgs/pull/139075 Signed-off-by: aszlig <aszlig@nix.build> Cc: @devhell Cc: @Profpatsch Cc: @sternenseemann
* machines/tishtushi: Remove machine configaszlig2022-08-151-56/+0
| | | | | | | | | | | | | | | | | | | | | | | | | | | I already killed that machine on 2022-08-04: tishtushi[~]0> cryptsetup erase /dev/sda2 WARNING! ======== This operation will erase all keyslots on device /dev/sda2. Device will become unusable after this operation. Are you sure? (Type 'yes' in capital letters): YES tishtushi[~]0> cryptsetup erase /dev/sda3 WARNING! ======== This operation will erase all keyslots on device /dev/sda3. Device will become unusable after this operation. Are you sure? (Type 'yes' in capital letters): YES Since some of the hardware was already broken (touchpad, keyboard, HDD, SSD and webcam), I hardly doubt that I'll use it again so it makes sense to remove the config here. Signed-off-by: aszlig <aszlig@nix.build>
* machines/shakti: Remove machine configaszlig2022-08-151-73/+0
| | | | | | | The machine no longer exists, so it doesn't make sense to continuously build it on Hydra. Signed-off-by: aszlig <aszlig@nix.build>
* slylandro: Initialise amdgpu module early in bootaszlig2022-08-031-0/+1
| | | | | | | | The amdgpu module is initialised at roughly the same time as the X server, which sometimes leads into race conditions where X won't start because the kernel driver is not yet loaded. Signed-off-by: aszlig <aszlig@nix.build>
* slylandro: Enable keyboard backlightaszlig2022-08-031-0/+1
| | | | | | | | This is a custom kernel module for TUXEDO laptops, which allows to control the backlight. I haven't played around with its settings yet, but having the module available helps with that. Signed-off-by: aszlig <aszlig@nix.build>
* slylandro: Add linux-firmwareaszlig2022-08-031-0/+1
| | | | | | This is needed for the GPU and the WiFi adapter. Signed-off-by: aszlig <aszlig@nix.build>
* slylandro: Disable caps lock keyaszlig2022-08-031-0/+1
| | | | | | | | | | | I'm used to a TypeMatrix keyboard, where caps lock is only a small key on the side instead of being prominently placed atop switch. Slylandro's internal keyboard however has that pesky caps lock key directly atop the left shift key and I'm constantly triggering it by accident. Signed-off-by: aszlig <aszlig@nix.build>
* slylandro: Set DPI for Xorgaszlig2022-08-031-0/+2
| | | | | | | | This makes the screen readable again, but might get us into trouble when switching to an external display. Again, similar to enabling bonding in the previous commit, let's find out :-) Signed-off-by: aszlig <aszlig@nix.build>
* slylandro: Use bonding for WiFi/Ethernetaszlig2022-08-031-2/+17
| | | | | | | | | This is mainly because I want to have a seamless transition between both, at least in theory. Whether this will work out to be so smooth when trying it on various networks out there remains to be seen, but let's try it :-) Signed-off-by: aszlig <aszlig@nix.build>
* machines: Re-add slylandro with different hardwareaszlig2022-08-021-0/+56
| | | | | | | | | | | | | | | | | | I've already got a temporary laptop back then where I used the same name and I introduced it in e73fcff03faed773df2500965cb9c4a4fcfbc04d and subsequently removed it in 240378dcec205b78b32c329ff02eb9bea8af2c11. With the new permanent hardware having arrived today, I decided to reuse the name, because it sounds nicer than "tishtushi" (which is my crappy laptop) and I also like the Slylandro Probes[1] a lot in Star Control. The configuration here is pretty much bare-bones as we had before with the temporary hardware and it's essentially a remix between dnyarri and the old config, more to refine later... [1]: https://wiki.uqm.stack.nl/Probe Signed-off-by: aszlig <aszlig@nix.build>
* machines/tishtushi: Add xxhash to available modsaszlig2022-04-011-0/+1
| | | | | | | | | | | | | | | I created the BTRFS module with a few different options than I did initially, one of them is that I changed the checksum algorithm to XXHASH. Unfortunately, the xxhash_generic module providing the algorithm is not available during stage 1, so mounting the volume would fail. In the local configuration.nix I already added it to availableKernelModules, but when I moved the config over to Vuizvui in e04c7e04f2fe63cece7a2f2064212bc055cdd035, I forgot to add it. Signed-off-by: aszlig <aszlig@nix.build>
* machines/tishtushi: Switch to new SSDaszlig2022-04-011-20/+9
| | | | | | | | | | | | | | | | | | | | | | Previously, the machine had an internal SSD with a size of around 20 GB, which until it died in 2018[1] was used as the /nix/store. With dnyarri getting a hardware upgrade[2], the SSD that was used back then as a bcache became obsolete. Given that the spinning rust inside Tishtushi is slow as hell and also way smaller (320 GB) than the former bcache SSD (512 GB), I decided to just replace the spinning rust with the SSD, which should at least make I/O operations bearable while falling asleep on CPU-bound operations. Since the NixOS LUKS setup now also propagates passphrases to other devices, we also no longer need the vault device because the reason it existed was to avoid being prompted multiple times for the passphrase. [1]: 54c99271f7570c32a4215a097e577272cd4210b2 [2]: be0fb40a12b5a9301509ad45fda1eda11971fa8e Signed-off-by: aszlig <aszlig@nix.build>
* machines/dnyarri: Use async discard for root FSaszlig2022-03-241-1/+3
| | | | | | | | | | | | | | | | Another thing that was lurking around in configuration.nix, so it has been battle-tested for weeks. Given that all the layers from the hardware up to the LUKS container with the filesystem support discard, it does make sense to enable it. The disadvantage of using discard with LUKS is that attackers can now gain information about the file system in use. However, this is already public knowledge so given that discard increases performance, I decided to enable it despite some warnings[1] out there. [1]: https://asalor.blogspot.com/2011/08/trim-dm-crypt-problems.html Signed-off-by: aszlig <aszlig@nix.build>
* machines/dnyarri: Add fwupdaszlig2022-03-241-0/+8
| | | | | | | | Another thing that was hanging around in configuration.nix and with my new hardware it actually became useful since it does have capsule updates. Signed-off-by: aszlig <aszlig@nix.build>
* dnyarri: Add default printeraszlig2022-03-241-0/+10
| | | | | | | | | | | | This is one of the few things that I originally added manually, simply because usually printing is not something I do regurarily. Since I had to re-do the CUPS config a few weeks ago, I decided to instead switch to ensurePrinters, simply because it makes it easier to just remove all the CUPS state and get to a known working configuration in the event that something breaks. Signed-off-by: aszlig <aszlig@nix.build>
* dnyarri: Remove the main user from the audio groupaszlig2021-07-041-7/+1
| | | | | | | | | The user shouldn't have direct access to the /dev/snd/* but go through things such as pulseaudio or pipewire. I added the user to that group back then as a workaround to quickly get something[TM] working, but nowadays it's not needed anymore. Signed-off-by: aszlig <aszlig@nix.build>
* dnyarri: Remove hplip backendaszlig2021-06-261-10/+0
| | | | | | | | I no longer have a scanner that requires to have hplip and/or any of the proprietary parts of it, so let's keep it out of the machine configuration. Signed-off-by: aszlig <aszlig@nix.build>
* dnyarri: Tie the Psi XMPP client to workspace 1aszlig2021-06-261-1/+1
| | | | | | | | Workspace one is usually my "communications workspace" and since I've been using Psi as my go-to XMPP client, it's a good idea to add it to that workspace, even though I usually start it on that workspace anyway. Signed-off-by: aszlig <aszlig@nix.build>
* dnyarri: Add identifiers for sound cardsaszlig2021-06-261-0/+12
| | | | | | | | | | | | | I already had this in my configuration.nix for quite some time and it's part of my overly complicated[TM] audio setup where I combine multiple speaker systems into one using two sound cards. Since the mapping depends on the individual channels, it's very much important to address the right sound card. So while I'm re-doing my whole audio setup, I decided to get at least the udev part out of my very messy configuration as a first start. Signed-off-by: aszlig <aszlig@nix.build>
* machines: Remove slylandroaszlig2021-06-241-54/+0
| | | | | | | | | | | | | | | As mentioned in the initial commit (e73fcff03faed773df2500965cb9c4a4fc), the machine was only temporary as a substitute for tishtushi. Since slylandro had a pretty slow dual core CPU and its own quirks, this was never a long-term solution and for the time being my intentions are to work with dnyarri's new hardware until I have a less annoying setup when I'm on the road again. While writing this message, slylandro just died a gruesome death with "cryptsetup erase", followed by "blkdiscard" on the whole drive. Signed-off-by: aszlig <aszlig@nix.build>
* dnyarri: Switch machine config to new hardwareaszlig2021-06-151-50/+10
| | | | | | | | | | | | | | | | | | | | | | | | The first (small) part of this already started in commit dc2ef245d76f900f174e87961a3d17e6aecbd6e2, where I switched from Intel CPU microcode to AMD. However, at that time the hardware was still on its way, so the actual move didn't happen yet. On June 4th 2021 at 17:22:41 CEST, I started to migrate to the new machine by using my old spinning rust configuration and btrfs balanced to NVMe-only. This means that we now no longer need to use bcache, which was a 512 GB SATA SSD, nor do we use a RAID10 configuration in btrfs. Instead, I've switched to NVMe with RAID1 exclusively, so this removes BFQ, power management workarounds and two devices. Since my old graphics card didn't work with UEFI GOP, I also had to switch to a newer one (which I wanted to avoid, but whatever...), so instead of DVI and HDMI we're now exclusively on DisplayPort. Signed-off-by: aszlig <aszlig@nix.build>
* profiles/base: Remove GRUB from base profileaszlig2021-06-023-2/+3
| | | | | | | | | | | After all the goal was to move all hardware specific stuff to the actual machine definition, which includes the boot loader. Since GRUB is enabled by default but with a higher priority value, we now no longer need to mkForce-disable the option for machines using systemd-boot. Signed-off-by: aszlig <aszlig@nix.build>
* machines/dnyarri: Switch to AMD CPU microcodeaszlig2021-06-024-0/+8
| | | | | | | | The new hardware for dnyarri no longer contains an Intel CPU, so apart from switching the microcode updates to AMD I also removed the setting from the base profile because it clearly doesn't belong there. Signed-off-by: aszlig <aszlig@nix.build>
* machines: Add temporary laptop slylandroaszlig2021-03-121-0/+52
| | | | | | | | | | | | | | Since Tishtushi had a SSD failure and thus became a majoor nuisance to work with, I got a temporary laptop from someone (since I don't know whether they want to be mentioned, I leave out their name for now) in order to be able to be more productive than waiting for several seconds for a 1 KiB text file to be saved. Right now, I'm not sure whether any firmware is needed for the temporary laptop, so this is a hardware configuration just to get started with a proper Hydra channel. Signed-off-by: aszlig <aszlig@nix.build>
* dnyarri: Fix unit names for btrfs scrub servicesaszlig2021-01-041-1/+1
| | | | | | | | | | | | | | | | | | Injecting pre/post start scripts into services called "btrfs-scrub-.service.service" isn't going to do a whole lot if the actual service name is called "btrfs-scrub-.service". During the last scrub I was wondering why caching got slow afterwards and found out that the caching devices were filled with lots of irrelevant data from the scrub. This led me to inspect what went wrong and when checking the properties of the scrub service unit, I found out that it never worked in the first place for the reason meantioned in the first paragraph. Actually using the right unit name helps a lot here, so onwards to the next scrub in February :-) Signed-off-by: aszlig <aszlig@nix.build>
* dnyarri: Set bcache mode to writethroughaszlig2020-11-101-1/+1
| | | | | | | | | | | | | | | | | | I already have this running in writethrough for long enough so that I'm pretty confident that a disaster like last time[1] should not happen anytime soon so that hereby I'm making this permanent in the machine configuration. The reason why I set this to writearound initially instead of writethrough, was that I just wanted to go with the most conservative option first. Of course, during scrub, we still disable all caching, so we should not generate too much wear on the SSD. [1]: 15008e69542774c441e388ad4c2e28a2d27f9ba0 Signed-off-by: aszlig <aszlig@nix.build>
* Use attribute sets for boot.initrd.luks.devicesaszlig2020-09-143-24/+25
| | | | | | | | | | | | | | | | | | | | This was one of the places where types.loaOf was still in place and it got removed a while[1] ago and this in turn causes evaluation to fail for quite a few machines: The option value `boot.initrd.luks.devices' in `...' is not of type `attribute set of submodules'. I've not only changed all the machines to use attribute sets but also fixed the check in core/tests.nix, because comparing against a list when the actual type is an attribute set will result in all the LUKS tests to be part of *all* channels, no matter whether you're actually using LUKS. [1]: https://github.com/NixOS/nixpkgs/commit/20d491a317d9956ddca80913f07 Signed-off-by: aszlig <aszlig@nix.build> Cc: @Profpatsch Cc: @sternenseemann
* dnyarri: Enable Wacom tablet supportaszlig2020-09-141-0/+1
| | | | | | | | I've had this laying around in my configuration.nix since quite a while, but today is one of those days where I can't stand all the cruft piling up there anymore and decided to add it here. Signed-off-by: aszlig <aszlig@nix.build>
* dnyarri: Add youtube-dl support to gPodderaszlig2020-07-281-1/+5
| | | | | | | | | | | | | | I'm using gPodder to follow a bunch of YouTube channels and the internal extractor/downloader tends to be pretty unreliable. On the other hand, youtube-dl is regularily updated and supports a ton of different formats. To make sure we can actually use the extension, gPodder needs to have access to the youtube_dl Python module, so we need to add it to the propagatedBuildInputs. Signed-off-by: aszlig <aszlig@nix.build>
* machines/tyree: Fix evaluation of weboobaszlig2020-04-041-1/+1
| | | | | | | | | | | | One (or maybe more?) dependency of weboob requires Python 3 and hence using weboob from python2Packages isn't going to evaluate and instead result in this error: error: google-api-core-1.16.0 not supported for interpreter python2.7 Using weboob from python3Packages fixes the eval error. Signed-off-by: aszlig <aszlig@nix.build>
* machines/dnyarri: Provide luks.devices as attrsetaszlig2020-01-181-2/+3
| | | | | | | | | | | | | | | | The use of types.loaOf has been deprecated since quite a while and lately[1] there is also an appropriate warning in place if an option definition relies on types.loaOf to coerce the list to an attrset. In vuizvui we didn't rely on types.loaOf, but it turned out that dnyarri still relied on it in boot.initrd.luks.devices. Since we already use attrsets for defining the LUKS devices, it's rather easy to fix and we just need to return a nameValuePair in mkDevice. [1]: https://github.com/NixOS/nixpkgs/commit/03309899eb41e50ef65001f63d6a4f078e3d4556 Signed-off-by: aszlig <aszlig@nix.build>
* treewide: use `runCommandLocal` where applicableProfpatsch2019-12-081-1/+1
| | | | | | | `runCommandLocal` was added to nixpkgs in https://github.com/NixOS/nixpkgs/pull/74642 to speed up trivial `runCommand` derivations by always building them locally. We have a few places where that’s good to use.
* machines/brawndo: Add Visual Studio Codeaszlig2019-10-301-2/+2
| | | | | | | | | Unfortunately, this is a work requirement for the user in question, so while I'd (personally) prefer to pretty much stay on free software, the circumstances prevent this :-/ Signed-off-by: aszlig <aszlig@nix.build> Cc: @dwenola
* machines/aszlig: Explicitly set useDHCPaszlig2019-10-084-1/+4
| | | | | | | | | | | | | | | | | | The usage of DHCP is no longer global since a while[1] and we now have to explicitly enable it for the interfaces in question. This actually is a good thing and makes it far less problematic if we use tunnel interfaces and other more complicated networking configuration. I added the definitions for all machines where I actually know which interfaces are in use and disabled useNetworkd for shakti, because I don't know the interface names for that machine and the machine currently isn't in use anyway, so we can add it later if needed. [1]: https://github.com/NixOS/nixpkgs/pull/69302 Signed-off-by: aszlig <aszlig@nix.build>
* machines/brawndo: Use cnijfilter2 from nixpkgsaszlig2019-09-231-13/+1
| | | | | | | https://github.com/NixOS/nixpkgs/pull/58399 has been merged a few hours ago, so we no longer need to use our manual override. Signed-off-by: aszlig <aszlig@nix.build>
* machines/managed: Switch to libinputaszlig2019-09-073-5/+0
| | | | | | | | | | | | Since NixOS/nixpkgs@466f5e534688049be2ed6e75ae5659633016b45e, libinput is enabled by default for Plasma 5, so we no longer need to explicitly enable it anymore. Furthermore, there was brawndo as the only machine still using Synaptics and it's about time it's using libinput as well, so I subsequently removed synaptics there. Signed-off-by: aszlig <aszlig@nix.build>
* machines/dnyarri: Fix allowUnfreePredicateaszlig2019-09-061-1/+1
| | | | | | | | | | | | | Since NixOS/nixpkgs@c814d72b517bb201c8bbbfc64e386c7023352886, a lot of packages now no longer have a name attribute but instead use pname, so when checking the package name within allowUnfreePredicate we need to make sure that we fall back to a default if the name attribute is not present. This fixes evaluation of the machine, however I didn't check if it actually builds. Signed-off-by: aszlig <aszlig@nix.build>
* machines/brawndo: Add cnijfilter2 version 5.70aszlig2019-06-121-1/+13
| | | | | | | | | | | | | | | | | I don't recall the exact model which was used for that machine, but it was definitely one that required version 5.70. Now I made a pull request for nixpkgs a while ago[1], which hasn't been merged and as I didn't have the chance to test it by myself I didn't merge that pull request yet. Since the user asked about that printer driver again, I decided to give version 5.70 a shot in vuizvui only and merge it upstream, once the user reports that the driver is working. [1]: https://github.com/NixOS/nixpkgs/pull/58399 Signed-off-by: aszlig <aszlig@nix.build>
* machines: Remove kzerzaaszlig2019-03-201-69/+0
| | | | | | | | This machine was used for controlling the LED lighting bars at Rockfabrik. I no longer work there and the machine has subsequently been replaced by something else, so I don't need kzerza anymore. Signed-off-by: aszlig <aszlig@nix.build>
* machines/dnyarri: Stop bcache during sleep/scrubaszlig2019-02-091-1/+29
| | | | | | | | | | | | | | | | | | | | | | | | | | I did have a major outage this week, because I was using bcache with writeback mode on a RAID10 backing storage. Fortunately, I was able to recover 99.9% of the data (only the most recent stuff wasn't recoverable), but I certainly don't want this to happen again in the future. While I did use bcache with hibernate and writeback, the interesting part is that the caching device went bonkers after a "normal" shutdown rather than a suspend/hibernate, with "normal" being "with a bunch of kernel warnings about zswap". Also, this happened around a btrfs scrub, so the inconsistency was all over the place. So first of all, I'm now going with writaround mode rather than writeback mode for the time being. Although it's slower than writeback, the chances that I need to do such a recovery again is close to 0% with writethrough and writearound because all writes are synchronous. Second, this very change makes sure that whenever the machine goes to sleep or a scrub is started, the caching is disabled and afterwards it's re-enabled. That way we shouldn't have lots of trash on the caching device. Signed-off-by: aszlig <aszlig@nix.build>
generated by cgit-pink 1.4.1 (git 2.36.1) at 2024-07-01 12:39:14 +0000