Serious problem (data corruption) with EVO860 disks + AMD SATA controllers

[Hans de Goede]

Hi Jens,

We (Fedora) just got this bug report:
https://bugzilla.redhat.com/show_bug.cgi?id=1729678

But this is not Fedora specific and as the Fedora bug report mentioned
has already been reported in bugzilla.kernel.org, but it seems that
SATA bugs reported in bugzilla.kernel.org still end up getting
assigned to Tejun...

Hence this email to bring the 2 upstream bugs to your attention:
https://bugzilla.kernel.org/show_bug.cgi?id=201693
https://bugzilla.kernel.org/show_bug.cgi?id=203475

The second bug report has a patch which disables queued TRIM
support on the 860, like it is already done for the 850.

This fixes things for the second reporter but I'm not sure if that is
enough, since in the first bug report the reporter mentions trying
"libata.force=noncqtrim" not helping and I've verified his dmesg and
the parameter is being passed and processed correctly ultimately
resulting in:

ata6.00: ATA-11: Samsung SSD 860 EVO 2TB, RVT01B6Q, max UDMA/133
ata6.00: disabling queued TRIM support

The 2 reporters do have different sata chipsets though, so maybe
for the "ASMedia ASM1062 Serial ATA Controller" from the second bug
report this is enough, but there is some other issue when combined
with the SATA controller in AMD socket AM3 mobo chipsets?

The first bug report talks about FPDMA errors, so maybe the
AM3 mobos need ATA_HORKAGE_BROKEN_FPDMA_AA?

Still it seems like a good idea to merge the patch from the 2nd
bug since it at least helps some people (there is another using
confirming the patch helps in the bug).

Regards,

Hans