2025-11-08.log

gordon2okay, with 6.17.7 kernel i'm getting issue with nvme, after a while (about 12 hr) this happens https://0x0.st/K9em.txt00:14
gordon2happened second time already00:14
gordon2there is ofc non-zero chance that it's just nvme decided to give up the ghost, but it is suspiciously started to happen after kernel upgrade00:16
gordon2(after the nvme's controller reset smart is happy and it looks like it works just fine)00:17
gordon2nvme is wd blue00:18
sigrid6.17.7, wd blue, no issues here (4 days uptime)00:19
gordon2rk3588 reform classic?00:20
joschgordon2: you could downgrade your kernel version and see if that helps00:20
sigridgordon2: yep00:20
gordon2oki00:20
gordon2will keep you updated00:20
gordon2josch: yeah, i could00:20
gordon2but then i'll have no more vdpu :(00:21
gordon2i think i'll start with smartctl -t long /dev/nvme000:21
+ ZetaR (~user@c-76-148-139-78.hsd1.fl.comcast.net)00:34
minutegordon2: i'd recommend to try the various nvme kernel options00:54
minutejosch: i've thought about the boot message a bit. current ideas with different levels of verbosity/specificity/credits: 1. Starting Debian GNU/Linux 2. Starting MNT Reform System. 3. Starting MNT Reform Debian System. 4. Starting MNT Reform Debian GNU/Linux System.00:58
+ paperManu_ (~paperManu@107.159.15.124)01:00
minutejosch: first try to install plymouth-theme-reform_1.81-2_arm64.deb results in > trying to overwrite '/usr/share/plymouth/themes/reform-y2k/background.png', which is also in package reform-tools (1.79-2+reform20251023T130442Z+1)01:00
minutejosch: i'm updating reform-tools01:00
gordon2minute: do you mean APST stuff? i definitely do not want to turn it off as well as ASPM stuff, especially since it worked perfectly fine for months before...01:04
gordon2are there any other options of note?01:06
minuteyeah there are a bunch01:07
minutegordon2: but also, maybe just a banal contact issue. you could try reseating.01:07
minutegordon2: the usual guides talk about APST yeah. but i've never tried another value than 0 for nvme_core.default_ps_max_latency_us01:09
gordon2extended test finished successfully, according to smart it is perfectly healthy nvme with no errors logged01:09
- pomel0 (QUIT: Read error: Connection reset by peer) (~pomel0@user/pomel0)01:15
minute> sudo dpkg -i --force-overwrite plymouth-theme-reform_1.81-2_arm64.deb01:17
minute> sudo plymouth-set-default-theme -R monobar01:18
minutei also would like to get my 120Hz hdmi modes back. but too sleepy tonight after flu+covid shots01:19
+ pomel0 (~pomel0@user/pomel0)01:26
- chomwitt (QUIT: Ping timeout: 260 seconds) (~chomwitt@2a02:85f:9a5f:900:1ac0:4dff:fedb:a3f1)01:26
- AnimaInvicta (PART: !!unknown attribute: msg!!) (~AnimaInvi@88-120-179-216.subs.proxad.net)01:32
minutegordon2: i'm now restoring a bunch of patches that were overzealously removed, lets see if there's anything pcie related 01:33
gordon2ah, interesting01:34
minutefunny how the bulk of the patchstack is hdmi related01:41
minutejosch: FYI but don't know yet if it builds https://source.mnt.re/reform/reform-debian-packages/-/merge_requests/15001:56
sigridI have two patches for 6.17.7 that may or may not be related: PCI-dw-rockchip-Add-system-PM-support.patch PCI-dw-rockchip-Disable-L1-substates.patch01:58
sigridmost likely the first one doesn't matter at all01:58
minutesigrid: the first one is back in the stack now01:59
minutesigrid: the second one i'm not aware of yet i think01:59
minute(or forgot about it)01:59
- mjw (QUIT: Ping timeout: 240 seconds) (~mjw@gnu.wildebeest.org)02:00
sigridit was this one https://lore.kernel.org/linux-rockchip/20251016175345.GA993781@bhelgaas/T/02:01
sigridpretty sure there was another one, to stop advertising l1 substates(???)02:03
sigridmy memory of these patches has faded02:03
minutehmm02:25
sigridminute: https://lore.kernel.org/linux-rockchip/20251017163252.598812-2-cassel@kernel.org/02:30
sigridsee how nvme issue is reported there as well02:30
sigridI think regardless of what gordon2's issue is, this should probably be applied02:30
minutesigrid: many thanks! iirc we don't even have clkreq wired up02:36
sigridnp02:38
- ZetaR (PART: !!unknown attribute: msg!!) (~user@c-76-148-139-78.hsd1.fl.comcast.net)02:50
- digitalrane (QUIT: Ping timeout: 250 seconds) (~digitalra@user/digitalrane)03:43
- paperManu (QUIT: Ping timeout: 264 seconds) (~paperManu@107.159.15.124)03:46
- paperManu_ (QUIT: Ping timeout: 246 seconds) (~paperManu@107.159.15.124)03:52
- aelius (QUIT: Remote host closed the connection) (~aelius@user/aelius)04:10
+ aelius (~aelius@user/aelius)04:11
joschminute: it seems your patch cleanup builds. You added a bunch of patches back (thank you!) but that makes me think that maybe it's too early to abandon the whole collabora patch stack and that we still should import everything once 6.18 comes around instead of porting the current 6.17 stack to 6.18? What are your thoughts on that?09:32
joschminute: sorry, i forgot that i took the opportunity to also move the y2k theme into the new plymouth package. You could install it with dpkg --force-overwrite to ignore this09:33
joschah you already did that09:33
joschmy zealous patch removal was done in the hopes to forward-port less patches later. Evidently, when i tested the result, I did not test enough (i have no way to test hdmi with my rk3588) but the rest worked so i merged it, sorry09:34
- colinsane1 (QUIT: Ping timeout: 244 seconds) (~colinunin@97-113-72-228.tukw.qwest.net)09:40
+ gustav25 (~gustav@c-78-82-52-240.bbcust.telenor.se)10:03
+ chomwitt (~chomwitt@2a02:85f:9a5f:900:1ac0:4dff:fedb:a3f1)10:13
- vkoskiv (QUIT: Ping timeout: 264 seconds) (~vkoskiv@109-204-186-7.cust.valoonet.fi)10:27
+ vkoskiv (~vkoskiv@109-204-186-7.cust.valoonet.fi)10:29
- chomwitt (QUIT: Ping timeout: 260 seconds) (~chomwitt@2a02:85f:9a5f:900:1ac0:4dff:fedb:a3f1)11:03
minutejosch: no problem, i just have to be more diligent in my review next time. and yeah, seems too early. i hope the hdmi situation won't end up like with imx8mq.11:44
+ mjw (~mjw@178.230.4.195)12:09
- mjw (QUIT: Ping timeout: 265 seconds) (~mjw@178.230.4.195)12:39
arjanjust ordered the new charger board and wifi daughter board for the pocket, hope that will make it run more stable and let me use linux-libre with ath9k wifi card :312:49
+ chomwitt (~chomwitt@2a02:85f:9a5f:900:1ac0:4dff:fedb:a3f1)12:56
+ paperManu (~paperManu@107.159.15.124)13:47
joschminute: there seems to be no gitlab runner for https://source.mnt.re/reform/reform-a311d-uboot/ right now -- could you assign the arm64 gitlab runner to it?13:55
minutejosch: right, will do!13:55
minutearjan: which m.2 ath9k card did you find?13:55
joschminute: could you also confirm my memory about a311d u-boot: it works like rk3588 in the sense that if emmc is zero it will fall back to reading it from sd-card, yes?13:56
josch(i want to test the u-boot builds by gitlab from my MR)13:56
arjanminute: I have a qcnfa222 lying around14:02
- chomwitt (QUIT: Ping timeout: 260 seconds) (~chomwitt@2a02:85f:9a5f:900:1ac0:4dff:fedb:a3f1)14:14
joschsince we have no wiki, I "documented" what I know in this issue: https://source.mnt.re/reform/reform-handbook/-/issues/814:24
joschminute: once you have time, could you have a quick look and confirm that the information is correct? If yes, then this will be the last time I ask because then I can just look it up in the future. :)14:25
gordon2is the guide in the handbook for updating lpc firmware is up to date?14:41
gordon2just checking before i go with it14:41
gordon2also if i'll go with precompiled fw from download.sh script will i get an 2200 mAh update?14:42
+ chomwitt (~chomwitt@2a02:85f:9a5f:900:1ac0:4dff:fedb:a3f1)14:47
mlarkinjosch: thanks for the link for the u-boot selector15:03
joschgordon2: the download-fw.sh script downloads the latest artifacts from the master branch and since !96 is merged you should be getting the 2200 mAh update15:06
joschmlarkin: maybe wait until minute confirms that my memory of this is correct :)15:06
gordon2josch: there is no way of downloading older artifacts it seems =/15:08
gordon2so it is a one-way trip15:09
joschgordon2: you can build older commits locally15:11
joschgordon2: the "problem" is that there is no versioning in the reform repository15:11
gordon2josch: well, it's more about peace of mind not for the actual rollback15:11
gordon2apparently you can back up firmware from the lpc's mass storage device, not sure if you can flash it back15:20
gordon2not planning to check15:20
mlarkinjosch: I also thought that's the rihgt switch after looking at it. I didn't yet try though.15:23
mlarkinput the machine back together. will check next time it's open15:23
gordon2updated15:25
- xktr (QUIT: Ping timeout: 250 seconds) (~xktr@user/xktr)15:40
+ xktr (~xktr@user/xktr)15:41
+ mjw (~mjw@178.230.4.195)15:53
minutejosch: that's how i remember it too @ a311d.16:18
- mjw (QUIT: Ping timeout: 240 seconds) (~mjw@178.230.4.195)16:32
bremneruh. I can see the kernel dumping call traces on the serial console of this pocket, but I can't get the lpc to respond (hyper-return does nothing). Is the hardware standby switch my only option ?16:52
minutebremner: wait, hyper+return does nothing? that's not the system controller then, but the keyboard itself16:57
minutebremner: the standby switch is one option, another is unscrewing the keyboard face plate and pressing reset button under the OLED pcb16:57
bremnerthe lights are on, but I may or may not have unloaded an important module16:58
bremnerhmm. Sounds like standby switch is easier16:58
bremnerand we're back. That was exciting. I wonder what I did...17:02
minutejosch: a311d uboot runner assigned17:13
minutemy hdmi 120hz is restored :317:13
bremnerhmm. non of my test scripts reproduces the hyper+return problem. 17:16
+ mjw (~mjw@gnu.wildebeest.org)17:28
- amospalla (QUIT: Remote host closed the connection) (~jordi@user/amospalla)17:31
+ amospalla (~jordi@user/amospalla)17:32
joschthank you for assigning the runner!17:36
joschnow i have to test the artifacts17:36
joschif i soft-brick my a311d i guess i have to switch to rk3588 sooner than planned :P17:36
f_ 13:56 <josch> minute: could you also confirm my memory about a311d u-boot: it works like rk3588 in the sense that if emmc is zero it will fall back to reading it from sd-card, yes?17:37
f_josch: yes17:37
f_well, not quite.17:37
f_If the @AML header is not present in the eMMC it will fallback to SD card17:38
f_either that, or if there's a HDMI boot dongle plugged in configured to force SD boot17:38
joschf_: perfect, thank you!17:38
joschyup, i'll need that dongle if i mess anything up :D17:38
f_the boot dongle is quite configurable, you can set it up to force boot to SD, or USB17:38
f_a few amlogic boards also have a button to force USB boot17:39
f_and some librecomputer boards have a physical switch to boot from either eMMC or SPI17:39
joschyeah, i really like how on imx8mq it's a simple DIP switch to select where u-boot is loaded from17:40
f_well, you can do that on amlogic too17:40
f_but it's not done on BPI-CM4 .. sadly.17:40
f_They do have test points you can short to force USB boot .. but from memory that only forced USB boot in their vendor u-boot, and not the USB mode that's in the bootROM17:41
joschwait what?17:41
joschfile ./amlogic-boot-fip/bananapi-cm4io/aml_encrypt_g12b17:42
josch./amlogic-boot-fip/bananapi-cm4io/aml_encrypt_g12b: ELF 64-bit LSB executable, x86-64, version 1 (GNU/Linux), statically linked, for GNU/Linux 2.6.24, BuildID[sha1]=2cc6c9c070c36fb9f5bb701d3b6d01499c32798a, with debug_info, not stripped17:42
f_yes, aml_encrypt_g12b contains debug symbols17:42
joschoh no... another binary blob...17:42
f_oh, not what you meant17:42
joschi want to build it from source17:42
f_gxlimg?17:42
joschsomehow the MNT gitlab CI is just executing the blob17:42
f_I think I already mentioned gxlimg before?17:42
f_aml_encrypt_g12b has been reversed a loong while ago so it's replacable17:43
f_https://github.com/repk/gxlimg - I also packaged it in alpine17:43
joschawesome, then that's what i'll do first before flashing this17:43
joschnice, the README of glximg contains everything i want17:44
f_for g12b you want to read README.g12b instead of README17:44
f_er, README.g12a17:45
joschnice, glad to have you here :)17:45
f_you're welcome17:45
f_nice talking to you too :)17:45
f_gxlimg was originally for amlogic gxl (S905X) SoCs. It was only later extended to amlogic g12a/b (A311D etc)17:46
joschminute: would you agree with replacing https://github.com/libreelec/amlogic-boot-fip from which MNT gitlab CI currently executes proprietary x86 binary blobs by https://github.com/repk/gxlimg ?17:46
f_Also that libreelec repo supports gxlimg17:47
f_I mean, the makefiles there, they support gxlimg too17:47
f_you just need to set an env var17:47
f_https://github.com/LibreELEC/amlogic-boot-fip/blob/master/g12a.inc#L4517:48
f_but I agree overall the situation with amlogic fw is not great :/17:48
minutejosch: sure thing17:49
minuteand thanks f_ !17:49
f_You're welcome17:49
joschf_: do you think it would be a sensible way forward to build (or even package for Debian) gxlimg and then run bananapi-cm4io/Makefile or would you say we should throw out libreelec/amlogic-boot-fip completely and run the scripts ourselves?17:51
joschjust looking for a recommended way forward because you know your way around the tools17:51
f_amlogic-boot-fip is just convenient17:51
f_personally what I'd like to do is do it all myself.. but it's quite complicated17:53
joschgreat, i have some scripts to read now :)17:53
joschat least when i zero out my emmc i can run experimentsw ith u-boot on the sdcard without bricking anything17:53
joschthat's a relief :)17:54
f_do it all myself += build bl301/acs/bl21 by hand, as they're actually GPL'd blobs. They are prebuilt in the amlogic-boot-fip repo for convenience, because you need a very old gcc and such to build those18:00
f_that's mostly the hard part of "do it all myself"18:01
f_and also tracking down all those vendor u-boot sources that have those18:01
josch"because you need a very old gcc" oh dear o018:03
f_I think I needed gcc5 for building acs18:03
f_and even then the build process would fail for other things, when it finished acs and was trying buildingsomething else18:03
joschlets go step-by-step -- my main priority right now is building from source the tools that run on my main processor at build time :)18:04
f_that part is easy18:04
joschgood :)18:05
f_just need a recent compiler mostly, not hyper old  gcc518:05
f_and then it's just a `make` away18:05
josch:)18:06
f_In e.g. Alpine Linux it's even easier, `apk add gxlimg` and you're done18:07
joschah, unfortunately gxlimg has no releases or git tags... :(18:24
f_ah, yeah no it doesn't :/18:25
+ TechnoWizard (~TechnoWiz@user/TechnoWizard)18:26
- chomwitt (QUIT: Ping timeout: 260 seconds) (~chomwitt@2a02:85f:9a5f:900:1ac0:4dff:fedb:a3f1)18:55
- TechnoWizard (QUIT: Ping timeout: 260 seconds) (~TechnoWiz@user/TechnoWizard)19:26
- chrcav (QUIT: Quit: Lost terminal) (~chrcav@user/chrcav)20:19
- gordon2 (QUIT: Quit: WeeChat 4.5.2) (~gordon1@citadel.niflheim.info)20:55
+ gordon1 (~gordon1@citadel.niflheim.info)20:55
joschf_: when building with libreelec/amlogic-boot-fip and GXLIMG_PATH=../gxlimg, the output is only a u-boot.bin and not a u-boot.bin.sd.bin -- expected? What's the difference?21:01
joschwithout gxlimg I also get u-boot.bin.usb.bl2 and u-boot.bin.usb.tpl21:01
f_u-boot.bin.sd.bin IIRC is just u-boot.bin with 512 offset21:01
f_can you xxd to confirm?21:01
joschf_: confirmed21:03
joschah this makes sense21:07
joschthe old code, produced by the proprietary x86 blob did this:21:08
josch# overwrite random bytes with static content to make the build reproducible21:08
joschdd if=/dev/zero of="${dtbname}-flash.bin" bs=512 count=1 conv=notrunc21:08
joschnow i can just shift it by 512 bytes and call it a day21:08
+ digitalrane (~digitalra@user/digitalrane)21:56
joschokay, i think we want to keep libreelec/amlogic-boot-fip around for a bit longer because it contains all the proprietary ddr training blobs22:02
f_yeah, but that's not the hardest part IMO22:11
- gustav25 (QUIT: Quit: Quit) (~gustav@c-78-82-52-240.bbcust.telenor.se)22:15
+ TechnoWizard (~TechnoWiz@user/TechnoWizard)22:33
- TechnoWizard (QUIT: Remote host closed the connection) (~TechnoWiz@user/TechnoWizard)22:35
joschf_: now that that i have sources to read i was having a deeper look where these random numbers are coming from. The thing seems to be called BL2IV, suggesting that this is for initializing some cryptographic routine? This is explicitly done here: https://github.com/repk/gxlimg/blob/master/bl2.c#L13823:50
joschi've never interacted with Remi Pommarel nor Remi Pommarel -- should i just mail them about this?23:53
joschfor the reform we made these 10 byte numbers deterministic a long while ago23:54
joschi wonder why you would want them not deterministic23:54

Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!