Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: [PATCH v5] libata: Add ATA_HORKAGE_NO_NCQ_ON_AMD for Samsung 860
 and 870 SSD.
To:     "Martin K. Petersen" <martin.petersen@oracle.com>
Cc:     Kate Hsuan <hpa@redhat.com>, Jens Axboe <axboe@kernel.dk>,
        Damien Le Moal <damien.lemoal@wdc.com>,
        Tor Vic <torvic9@mailbox.org>, linux-ide@vger.kernel.org,
        linux-kernel@vger.kernel.org
References: <20210901151643.13562-1-hpa@redhat.com>
 <3e26e7a5-0d99-b993-d5ce-aa517e1bf1bb@redhat.com>
 <yq1h7f24y6f.fsf@ca-mkp.ca.oracle.com>
From:   Hans de Goede <hdegoede@redhat.com>
Message-ID: <238d0841-0f03-928f-5441-89d5c9dcf9b9@redhat.com>
Date:   Thu, 2 Sep 2021 22:46:28 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.11.0
MIME-Version: 1.0
In-Reply-To: <yq1h7f24y6f.fsf@ca-mkp.ca.oracle.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

Hi,

On 9/2/21 9:01 PM, Martin K. Petersen wrote:
> 
> Hans,
> 
>> So it looks like we actually need to disable NCQ for Samsung 860/870
>> devices when the SATA controller has a vendor-id of PCI_VENDOR_ID_ATI
>> rather then AMD.
> 
> That's another great data point!
> 
> I wonder if there actually is a Samsung problem (given that these drives
> work fine on other controllers). Or if it is just the queued trim
> handling that's broken on 1002:4391 controllers from ATI.
> 
> When I originally experimented with queued trim I had systems I could
> not get to work. But queued trim worked fine when the same drives were
> connected to more modern chipsets (note that this was "did not work at
> all" as opposed to "randomly corrupting data").
>
> Do we have any evidence at all of queued trim working with non-Samsung
> drives on these controllers? Not sure how many modern SATA drives
> actually implement this feature. Maybe the reason we see Samsung drives
> in the bug reports is due to a combination of popularity and the fact
> that these drives actually implement queued trim support.

The Samsung 860 / 870 series causing issues when queued trim support
is enabled are quite wide-spread, covering many different controller
models from all well known controller vendors (Intel, Asmedia, Marvell
and AMD). So disabling queued-trim support definitely is the right
thing to do (and we should have done so a long time ago, I am to
blame for this not being done sooner).

As for your theory that it is really a problem with the controller
and not the the SSDs, I honestly do not know, but I doubt it,
there are no such reports with any other vendor's SSD or newer
Samsung models, so this seems unlikely.

Regards,

Hans