sector size other than 512 bytes

sector size other than 512 bytes

[TK (Tzung-Kwang) Wang-SSI]

Hi All,

I have been using FIO on SATA drives for some times and all were fine until recently I added SAS drives to the test list. Some of the SAS drives have 520-byte sector size, with DIF, and my FIO setup was not able to run on those drives. I have not found related postings in archive for possible workaround. My had also upgrade from v2.1.14 to v2.8 with the same result.

Is there a patch available for supporting sector size other than 512-byte for either Windows 7 and/or Ubuntu 14 LTS environment?

Thanks,
TK

Re: sector size other than 512 bytes

[Sitsofe Wheeler]

On 14 April 2016 at 19:44, TK (Tzung-Kwang) Wang-SSI
<tk.wang@ssi.samsung.com> wrote:
>
> I have been using FIO on SATA drives for some times and all were fine until recently I added SAS drives to the test list. Some of the SAS drives have 520-byte sector size, with DIF, and my FIO setup was not able to run on those drives. I have not found related postings in archive for possible workaround. My had also upgrade from v2.1.14 to v2.8 with the same result.
>
> Is there a patch available for supporting sector size other than 512-byte for either Windows 7 and/or Ubuntu 14 LTS environment?

Does linux report the block size as 520 in /sys ?

-- 
Sitsofe | http://sucs.org/~sits/

RE: sector size other than 512 bytes

[TK (Tzung-Kwang) Wang-SSI]

Yes, according to sg_readcap output. 

Regards,
TK

-----Original Message-----
From: Sitsofe Wheeler [mailto:sitsofe@gmail.com] 
Sent: Thursday, April 14, 2016 12:46 PM
To: TK (Tzung-Kwang) Wang-SSI
Cc: fio@vger.kernel.org
Subject: Re: sector size other than 512 bytes

On 14 April 2016 at 19:44, TK (Tzung-Kwang) Wang-SSI <tk.wang@ssi.samsung.com> wrote:
>
> I have been using FIO on SATA drives for some times and all were fine until recently I added SAS drives to the test list. Some of the SAS drives have 520-byte sector size, with DIF, and my FIO setup was not able to run on those drives. I have not found related postings in archive for possible workaround. My had also upgrade from v2.1.14 to v2.8 with the same result.
>
> Is there a patch available for supporting sector size other than 512-byte for either Windows 7 and/or Ubuntu 14 LTS environment?

Does linux report the block size as 520 in /sys ?

--
Sitsofe | http://sucs.org/~sits/

Re: sector size other than 512 bytes

[Sitsofe Wheeler]

On 14 April 2016 at 21:06, TK (Tzung-Kwang) Wang-SSI
<tk.wang@ssi.samsung.com> wrote:
> Yes, according to sg_readcap output.

It ought to be the same but what about via

/sys/block/<device>/queue/logical_block_size

?

-- 
Sitsofe | http://sucs.org/~sits/

RE: sector size other than 512 bytes

[TK (Tzung-Kwang) Wang-SSI]

That shows 512 instead of 520,

# cat /sys/block/sdb/queue/logical_block_size
512
# sg_readcap /dev/sdb
Read Capacity results:
   Last logical block address=918552575 (0x36bfffff), Number of blocks=918552576
   Logical block length=520 bytes
Hence:
   Device size: 477647339520 bytes, 455520.0 MiB, 477.65 GB

Regards,
TK


-----Original Message-----
From: Sitsofe Wheeler [mailto:sitsofe@gmail.com] 
Sent: Thursday, April 14, 2016 8:19 PM
To: TK (Tzung-Kwang) Wang-SSI
Cc: fio@vger.kernel.org
Subject: Re: sector size other than 512 bytes

On 14 April 2016 at 21:06, TK (Tzung-Kwang) Wang-SSI <tk.wang@ssi.samsung.com> wrote:
> Yes, according to sg_readcap output.

It ought to be the same but what about via

/sys/block/<device>/queue/logical_block_size

?

--
Sitsofe | http://sucs.org/~sits/

Re: sector size other than 512 bytes

[Sitsofe Wheeler]

On 15 April 2016 at 22:19, TK (Tzung-Kwang) Wang-SSI
<tk.wang@ssi.samsung.com> wrote:
> That shows 512 instead of 520,
>
> # cat /sys/block/sdb/queue/logical_block_size
> 512
> # sg_readcap /dev/sdb
> Read Capacity results:
>    Last logical block address=918552575 (0x36bfffff), Number of blocks=918552576
>    Logical block length=520 bytes
> Hence:
>    Device size: 477647339520 bytes, 455520.0 MiB, 477.65 GB

In that case I'd say the issue you're seeing is within the kernel
rather than in fio - the kernel block device is going to refuse
userspace accesses that aren't multiples of 512. Possibly raw SCSI
commands would work?

Is there anything in dmesg about "unsupported sector size" (see
https://github.com/torvalds/linux/blob/v4.5/drivers/scsi/sd.c#L2292
for the check that provokes that message)?

-- 
Sitsofe | http://sucs.org/~sits/

RE: sector size other than 512 bytes

[TK (Tzung-Kwang) Wang-SSI]

Sitsofe,

Yes, dmesg shows "Unsupported sector size 520.". So solution for this issue is have 520-byte sector size support in linux device driver, it seems.

Regards,
TK

-----Original Message-----
From: Sitsofe Wheeler [mailto:sitsofe@gmail.com] 
Sent: Saturday, April 16, 2016 3:35 AM
To: TK (Tzung-Kwang) Wang-SSI
Cc: fio@vger.kernel.org
Subject: Re: sector size other than 512 bytes

On 15 April 2016 at 22:19, TK (Tzung-Kwang) Wang-SSI <tk.wang@ssi.samsung.com> wrote:
> That shows 512 instead of 520,
>
> # cat /sys/block/sdb/queue/logical_block_size
> 512
> # sg_readcap /dev/sdb
> Read Capacity results:
>    Last logical block address=918552575 (0x36bfffff), Number of blocks=918552576
>    Logical block length=520 bytes
> Hence:
>    Device size: 477647339520 bytes, 455520.0 MiB, 477.65 GB

In that case I'd say the issue you're seeing is within the kernel rather than in fio - the kernel block device is going to refuse userspace accesses that aren't multiples of 512. Possibly raw SCSI commands would work?

Is there anything in dmesg about "unsupported sector size" (see
https://github.com/torvalds/linux/blob/v4.5/drivers/scsi/sd.c#L2292
for the check that provokes that message)?

--
Sitsofe | http://sucs.org/~sits/

Re: sector size other than 512 bytes

[Sitsofe Wheeler]

On 18 April 2016 at 21:56, TK (Tzung-Kwang) Wang-SSI
<tk.wang@ssi.samsung.com> wrote:
>
> Yes, dmesg shows "Unsupported sector size 520.". So solution for this issue is have 520-byte sector size support in linux device driver, it seems.

You might find the sg ioengine works but if it depends on regular
block device commands to get information (e.g. the disk size) then
things might fail before I/O can be submitted.

In a previous message you mention DIF (the data integrity extensions)
but to use that it's worth noting your driver and controller have to
support it in addition to the disk. Perhaps you can use lsscsi's
--protection and --protmode to find out what's available? Here's what
Red Hat had to say about DIF support in RHEL 6:
https://access.redhat.com/solutions/41548 but I don't know if that
impacts your case.

-- 
Sitsofe | http://sucs.org/~sits/