[SOLVED] Asmedia ASM1062 quirk fix available upstream in 5.4.148 (not yet PVE 6.4) and not on 5.11.22 (PVE 7.0 branch)

[guerby]

Hi,

While running PVE 6.4 (kernel 5.4.140-1-pve) on a threadripper system based on AsrockRack TRX40D8-2N2T motherboard we recently lost two SATA SSD drives on kernel write errors, after testing various things (swapping SSD, cables) we found out the issue happens (after a while) only on the two Asmedia ASM1062 SATA ports, not on the other 4 SATA ports of the motherboard which have different controlers.

Looking for similar issue we found out a description that matches our symptoms and hopefully someone developped a kernel patch for it:

https://patchwork.ozlabs.org/project/linux-pci/patch/20210317115924.31885-1-kabel@kernel.org/

This patch backport landed in 5.4.148:

https://elixir.bootlin.com/linux/v5.4.148/source/drivers/pci/quirks.c#L3255

References :

https://bugzilla.kernel.org/show_bug.cgi?id=212695
https://lwn.net/Articles/870006/
I'm announcing the release of the 5.4.148 kernel.
Marek Behún (1): PCI: Restrict ASMedia ASM1062 SATA Max Payload Size Supported


Asmedia ASM1062 is often cited on this forum for various issues, an example similar to ours:

https://forum.proxmox.com/threads/sata-devices-missing-after-update.76204/

I don't know if an updated kernel on pve 6.4 no subscription repo is planned but I'm ready to test it

On our production system on pve-enterprise repo kernel is pve-kernel-5.4.128-1-pve.

On another of our test system we have PVE 7.0 on no subscription repo with kernel 5.11.22 which doesn't have the ASmedia quirk backported in the official kernel source tree (as 5.11 is not a "long term" kernel), I didn't check if proxmox kernel source has the quirk.

 

[guerby]

More info from kernel developpers (about which version has the fix on each branch):

https://bugzilla.kernel.org/show_bug.cgi?id=212695#c4


Pali Rohár : This issue was fixed/workarounded in following commit (in 5.15) which forces MPS to 256 bytes:

https://git.kernel.org/pub/scm/linu.../?id=b12d93e9958e028856cbcb061b6e64728ca07755

And also it was backported to stable kernel versions: 5.14.6, 5.13.19, 5.10.67, 5.4.148, 4.19.207, 4.14.247, 4.9.283 and 4.4.284

 

[guerby]

pve-no-subscription just moved from 5.4.140 to pve-kernel-5.4.143-1-pve is there by any chance the 5.4.148 asmedia fix backported?

 

[guerby]

Testing 5.4.157-1-pve, after 24 hours no issue so far.

 

[guerby]

No issue after two weeks with 5.4.157.

I see that PVE 7.1 has now 5.13.19-2 which should have the fix too.

 

[guerby]

Note: it didn't last long and even with latest 5.15.64-1-pve I had an issue of losing connection to disks with ASM1062

startech 8P6G-PCIE-SATA-CARD has 4xASM1062 inside and shows the issue too.

Looking for recommandations for a PCIe 8 port SATA or 2xminiSAS PCIe non RAId card that doesn't have an ASM1062 chip inside, anyone?