Re: [PATCH v4] libata: Add ATA_HORKAGE_NO_NCQ_ON_AMD for Samsung 860 and 870 SSD.

[Hans de Goede <hdegoede@redhat.com>]

Hi Tor,

On 9/1/21 9:37 AM, torvic9@mailbox.org wrote:
> (Sorry for not doing a proper reply)
> 
> Hello,
> Noob here.
> I have a Samsung 860 Pro connected to a AMD X570 chipset mainboard and
> it just works flawlessly on 5.13 and 5.14.
> Are you sure that *every* 860/870 is concerned by this problem on
> *every* AMD controller?

I am pretty sure that every 860 / 870 EVO is affected,
I am not sure if the PRO is also affected.

As for *every* AMD controller, chances are that more recent
AMD controllers are fine.

We have been trying to resolve various issues with this combo
for a long time now, see:

https://bugzilla.kernel.org/show_bug.cgi?id=201693
https://bugzilla.kernel.org/show_bug.cgi?id=203475

> Isn't this too restrictive?
> Or am I simply missing something?

The problem is that when users are hit by this they end up with
a non functional system and even fs / data  corruption. Where
as OTOH disabling NCQ leads to a (significant) performance
degradation but affected systems will still work fine.

So I believe that it is best to err on the safe side here
and accept the performance degradation as a trade-of for
fixing the fs / data corruption.

With that said, I do believe that we should allow re-enabling
ncq on this combo through libata.force on the kernel cmdline
by adding this extra bit to the patch:

diff --git a/drivers/ata/libata-core.c b/drivers/ata/libata-core.c
--- a/drivers/ata/libata-core.c
+++ b/drivers/ata/libata-core.c
@@ -6136,6 +6136,8 @@ static int __init ata_parse_force_one(char **cur,
 		{ "ncq",	.horkage_off	= ATA_HORKAGE_NONCQ },
 		{ "noncqtrim",	.horkage_on	= ATA_HORKAGE_NO_NCQ_TRIM },
 		{ "ncqtrim",	.horkage_off	= ATA_HORKAGE_NO_NCQ_TRIM },
+		{ "noncqamd",	.horkage_on	= ATA_HORKAGE_NO_NCQ_ON_AMD },
+		{ "ncqamd",	.horkage_off	= ATA_HORKAGE_NO_NCQ_ON_AMD },
 		{ "dump_id",	.horkage_on	= ATA_HORKAGE_DUMP_ID },
 		{ "pio0",	.xfer_mask	= 1 << (ATA_SHIFT_PIO + 0) },
 		{ "pio1",	.xfer_mask	= 1 << (ATA_SHIFT_PIO + 1) },

And I will also add a comment to both linked bugs to see if we can maybe
exclude the pro models from this quirk and if we can maybe narrow it down
to a subset of the AMD SATA controllers.

But that narrowing down is probably best done as a follow up fix, while just
going with this "err on the safe side" approach for now.

Regards,

Hans