control: found -1 linux/6.11.2-1 Hi,
On 2024-10-20 01:53, Jing Luo wrote: > On 2024-10-19 23:08, Aurelien Jarno wrote: > > Hi, > > > > On 2024-10-19 20:48, Jing Luo wrote: > > > Package: src:linux > > > Version: 6.10.12-1 > > > Severity: critical > > > Tags: upstream > > > Justification: causes serious data loss > > > > > > Hi, > > > > > > The dwmmc_starfive module causes serious filesystem corruption when > > > debian > > > is installed to emmc with linux-image-6.10.12-riscv64. > > > > The dwmmc_starfive module is providing support for both emmc and > > sd-card. Are you also able to reproduce the issue with the sd-card? > > I couldn't reproduce it with sd card: I still see kernel log spam by > dwmmc_starfive, but the filesystem itself looks ok. The filesystem passed > e2fsck and a debsums run. Ok, thanks for checking. Then it seems that the swiotlb messages, while worrying are unrelated to the data corruption. > > > The board in question > > > is Starfive Visionfive 2. > > > > How much memory do you have on this board? > > 8GB. Thanks. > > > The system tested on uses upstream u-boot and the > > > rootfs uses ext4. Please see the attached kernel log. The emmc was > > > written > > > garbage even under light I/O. > > > > What do you mean by upstream u-boot? The version in Debian or a newer > > version? Do you pass the kernel version of the device-tree to the > > bootloader? > > Ah, that's a newer u-boot, 2024.10 I built using defconfig. I use > u-boot-menu to generate the extlinux.conf so the kernel loads the device > tree from /usr/lib/linux-image... > > I tried debian's 2024.01 version of u-boot, but I'm afraid that the result > was the same. Ok, thanks for confirming, that looks all good. > > > Also: the vendor kernel doesn't have this issue. The latest 6.11.2-1 > > > is not > > > tested yet because I could not get it to boot (another bug?). > > > > This version does work on a VisionFive 2 board, admittedly I haven't > > tested the emmc part. Could you please provide more version on the > > issue, as kernel 6.10.12-1 is basically unsupported? > > I finally and unfortunately reproduced this with 6.11.2-1. Thanks, I have updated the version in the BTS accordingly. > Another thing I noticed: the filesystem corruption has a pattern (sort of), > it looks like this with /var/lib/dpkg/status: > > PPacaage apaarmrr > Saatu:: isstall ok installed > Priority: optional > Section: admin > Inttaleed-iize 2333 > Miintiine:: Dbbia ApAArmrr Taam ppkgaapprrmo--t...@lsstsaalitth.eebinn.ogg> > rrchttecuure riccv6 > > Sorrce apaarmrr (..1.--1)VVeriion 3...7-++b1RRepaace:: fiitxddat (< > 1:..2.. > > Deeend:: dbbcoff, eebcnnf >>= ..5)|| dbbcoff-200, iibc (> 2.88) > uuggssts apaarmrr-poofiees-xxtr,, apparoor-ttil > > Braaks apaarmrr-poofiees-xxtr (< 1.11),ffcixx-dtta <<< > ::4...9.--1uuunt22),s > > Cofffiees: /ecc/apparoor.//abi/3.0 f97e410509c5def279aa227c7de12e06 > /etc/apparmo..d/bbi/eernll-544-ottofrreennetoork557b88ac44e6118f55a0ddc8004733 > > /ttc/pppammordd/aii/krrne--5.--vaiill 77447eff0b114f88bf77689998882044 > eetcaapprrmo..d/bbstaactoonsXX e88b3665c668c99b5222d11dbddf0992 Hmm I am actually surprised that you are able to boot your system with so many corruption. As the characters seems duplicated and the emmc bus is a 8-bit one, it could be a timing issue between the emmc and the controller. One thing to try is lowering the speed by removing the highspeed capabilities from the device tree. You can patch the existing dtc binary file with dtc (from the device-tree-compiler package) that way: dtc /usr/lib/linux-image-6.11.2-riscv64/starfive/jh7110-starfive-visionfive-2-v1.3b.dtb | egrep -v '(cap-mmc-highspeed|mmc-hs200-1_8v)' | dtc > outfile If that helps, you might also want to try removing only mmc-hs200-1_8v. Regards Aurelien -- Aurelien Jarno GPG: 4096R/1DDD8C9B aurel...@aurel32.net http://aurel32.net
signature.asc
Description: PGP signature
