After upgrading to TrueNAS 25.10.4 (kernel 6.12.91-production) and rebooting, one of my pool drives went UNAVAIL and the RAIDZ1 pool dropped to DEGRADED. The drive is connected to an AXAGON PCES-SJ2 PCIe SATA card (JMicron JMB585, PCI ID 197b:0585). In the boot log, the affected port repeatedly fails to establish a SATA link:
ata4: softreset failed (1st FIS failed)
ata4: limiting SATA link speed to 3.0 Gbps
ata4: softreset failed
ata4: reset failed, giving up
Because the link never comes up, no /dev/sd* device is created, so ZFS reports the member as UNAVAIL with zero read/write/checksum errors.
I’ve ruled out hardware: the drive and its cable test fine on a motherboard SATA port (detected in BIOS, links at 6 Gbps), and the failure follows the card across both of its ports. lspci -vvv on the JMB585 shows the PCIe side is completely healthy - link trained at full 8GT/s x1, ASPM not supported (so power management isn’t a factor), and no correctable or uncorrectable PCIe errors logged.
The decisive test: rolling back to my previous boot environment, 25.10.3.1, with no other changes, brings the drive back immediately and the pool resilvers to healthy. So the same hardware/cable/drive works on the 25.10.3.1 kernel but fails on 6.12.91, which points to a kernel-level JMB585/AHCI regression rather than a hardware fault.
Has anyone else seen JMB585 (or other JMicron) ports fail with 1st FIS failed on the 25.10.4 kernel?
If so, the advisable course of action is to file a bug report—but the issue may well be upstream in Debian rather in TrueNAS.
It is long known that JMB controllers are not the best, so other than staying on 25.10.3 or hoping that the regression will be fixed, the best course of action may be to replace the card by a SAS HBA, or at the very least some ASM 11xx SATA controller without port multiplier.
If you have a spare PCIe slot, known-good HBAs are still pretty affordable. I’d go for a LSI 94xx or higher series to minimize power consumption.
If you only need a limited number of additional ports, art of server is also selling $67 9304-4i cards. See YouTube video here, suggesting that the 9200-4i may also be a great low cost option for spinning rust.
so… while Jmicron controllers tend to range from quirky to broken-by-design (depending on if you are daring and use SSDs or 4kn disks)… seeing FIS errors implies it is specifically this is a port multiplier problem that the driver is running into.
the verdict is the same, you need a better SATA/SAS HBA. are you using this card because it’s a x1? if you aren’t limited by pcie slots you should go with an LSI/Avago/broadcom HBA.
if you are in-fact limited to a x1, an asmedia based card may be a tolerable substitute so long as it’s free of port multipliers and you are buying a design that at least has some reviews online. (the other problem with these cards is the white-label trash factories often take liberties with their implementations)
Jmicron, Marvell, and Via based controllers are wholly unsuitable for TrueNAS (I would say in-general but their quirks tend to surface more with the parallel nature of ZFS pools)
Art of the Server commands a hefty premium on eBay.
In return, he promises knowledgable support and seems to have a very good reputation, here and on eBay for genuine stuff, updated to the latest firmware, IT mode, etc. So if playing around with sas2flash, etc. is not up your alley, he can be a good source.
