Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 6 Oct 2015 12:37:37 -0700
From:      Jim Harris <jim.harris@gmail.com>
To:        Steven Hartland <killing@multiplay.co.uk>
Cc:        Sean Kelly <smkelly@smkelly.org>,  FreeBSD-STABLE Mailing List <freebsd-stable@freebsd.org>
Subject:   Re: Dell NVMe issues
Message-ID:  <CAJP=Hc869Qmtt-r51B0A-TKPZOoZW1bKXKDnHvFcWef3eTmsEg@mail.gmail.com>
In-Reply-To: <561416FB.1050802@multiplay.co.uk>
References:  <BC5F191D-FEB2-4ADC-9D6B-240C80B2301C@smkelly.org> <5613FA02.2080205@multiplay.co.uk> <CAJP=Hc9-oQnk2r48OBXVCQbMDn0URDMDb80a0i0XvUDPuuLkrA@mail.gmail.com> <561416FB.1050802@multiplay.co.uk>

next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Oct 6, 2015 at 11:46 AM, Steven Hartland <killing@multiplay.co.uk>
wrote:

> On 06/10/2015 19:03, Jim Harris wrote:
>
>
>
> On Tue, Oct 6, 2015 at 9:42 AM, Steven Hartland <
> <killing@multiplay.co.uk>killing@multiplay.co.uk> wrote:
>
>> Also looks like nvme exposes a timeout_period sysctl you could try
>> increasing that as it could be too small for a full disk TRIM.
>>
>
>> Under CAM SCSI da support we have a delete_max which limits the max
>> single request size for a delete it may be we need something similar for
>> nvme as well to prevent this as it should still be chunking the deletes to
>> ensure this sort of thing doesn't happen.
>
>
> See attached.  Sean - can you try this patch with TRIM re-enabled in ZFS?
>
> I would be curious if TRIM passes without this patch if you increase the
> timeout_period as suggested.
>
> -Jim
>
>
> Interesting does the nvme spec not provide information from the device as
> to what its optimal / max deallocate request size should be like the ATA
> spec exposes?
>

Correct - there is no way for devices to specify a max/optimal deallocate
size in NVMe.  There is an implicit limit from the 32-bit LBA length in the
DSM Range data structure defined by the spec.  So this patch is needed
anyways to make sure we don't overflow the 32-bit LBA length.  Sean's
drives are 1.6TB which fits in an unsigned 32-bit value on a 512-byte
sector formatted controller, so I don't think that is the problem in Sean's
case.

-Jim



>     Regards
>     Steve
>



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAJP=Hc869Qmtt-r51B0A-TKPZOoZW1bKXKDnHvFcWef3eTmsEg>