[U-Boot] NFS and timeouts?

[U-Boot] NFS and timeouts?

[Tom Rini]

Hey all,

I've noticed now, but not dug into a problem that goes like this.  On
every platform that I have tried NFS on now (and I wasn't a user before
the test came in) I see:
# nfs 80000000 /tftpboot/1MiBtest.bin
link up on port 0, speed 1000, full duplex
#################################################################
         #################################################################
         #################################################################
         ##########T T T T
done
Bytes transferred = 1048576 (100000 hex)

for the same 1MiB file.  The link line will vary from board to board but
the end result is that I always see 4 T (for timeout) at the end of the
transfer.  On boards where I am doing this on gigabit the initial
transfer is fast enough that the timeout doesn't cause failure.  On the
boards where I'm at 100Mbit instead however, I fail.

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20161013/8d91101d/attachment.sig>

[U-Boot] NFS and timeouts?

[Simon Glass]

Hi Tom,

On 13 October 2016 at 13:11, Tom Rini <trini@konsulko.com> wrote:
> Hey all,
>
> I've noticed now, but not dug into a problem that goes like this.  On
> every platform that I have tried NFS on now (and I wasn't a user before
> the test came in) I see:
> # nfs 80000000 /tftpboot/1MiBtest.bin
> link up on port 0, speed 1000, full duplex
> #################################################################
>          #################################################################
>          #################################################################
>          ##########T T T T
> done
> Bytes transferred = 1048576 (100000 hex)
>
> for the same 1MiB file.  The link line will vary from board to board but
> the end result is that I always see 4 T (for timeout) at the end of the
> transfer.  On boards where I am doing this on gigabit the initial
> transfer is fast enough that the timeout doesn't cause failure.  On the
> boards where I'm at 100Mbit instead however, I fail.

I have seen this also - what type of interface are you using?

Regards,
Simon

[U-Boot] NFS and timeouts?

[Joe Hershberger]

On Fri, Oct 14, 2016 at 12:52 AM, Simon Glass <sjg@chromium.org> wrote:
> Hi Tom,
>
> On 13 October 2016 at 13:11, Tom Rini <trini@konsulko.com> wrote:
>> Hey all,
>>
>> I've noticed now, but not dug into a problem that goes like this.  On
>> every platform that I have tried NFS on now (and I wasn't a user before
>> the test came in) I see:
>> # nfs 80000000 /tftpboot/1MiBtest.bin
>> link up on port 0, speed 1000, full duplex
>> #################################################################
>>          #################################################################
>>          #################################################################
>>          ##########T T T T
>> done
>> Bytes transferred = 1048576 (100000 hex)
>>
>> for the same 1MiB file.  The link line will vary from board to board but
>> the end result is that I always see 4 T (for timeout) at the end of the
>> transfer.  On boards where I am doing this on gigabit the initial
>> transfer is fast enough that the timeout doesn't cause failure.  On the
>> boards where I'm at 100Mbit instead however, I fail.
>
> I have seen this also - what type of interface are you using?

Does this happen in your testing, Guillaume?

[U-Boot] NFS and timeouts?

[Tom Rini]

On Thu, Oct 13, 2016 at 04:52:14PM -0600, Simon Glass wrote:
> Hi Tom,
> 
> On 13 October 2016 at 13:11, Tom Rini <trini@konsulko.com> wrote:
> > Hey all,
> >
> > I've noticed now, but not dug into a problem that goes like this.  On
> > every platform that I have tried NFS on now (and I wasn't a user before
> > the test came in) I see:
> > # nfs 80000000 /tftpboot/1MiBtest.bin
> > link up on port 0, speed 1000, full duplex
> > #################################################################
> >          #################################################################
> >          #################################################################
> >          ##########T T T T
> > done
> > Bytes transferred = 1048576 (100000 hex)
> >
> > for the same 1MiB file.  The link line will vary from board to board but
> > the end result is that I always see 4 T (for timeout) at the end of the
> > transfer.  On boards where I am doing this on gigabit the initial
> > transfer is fast enough that the timeout doesn't cause failure.  On the
> > boards where I'm at 100Mbit instead however, I fail.
> 
> I have seen this also - what type of interface are you using?

I see it on CPSW (am335x, dra7xx) and the USB smsc95xx on rpi3 and the
USB ASIX on omap3_beagle.

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20161013/6e9e9954/attachment.sig>

[U-Boot] NFS and timeouts?

[Guillaume Gardet]

Hi,


Le 14/10/2016 ? 01:00, Joe Hershberger a ?crit :
> On Fri, Oct 14, 2016 at 12:52 AM, Simon Glass <sjg@chromium.org> wrote:
>> Hi Tom,
>>
>> On 13 October 2016 at 13:11, Tom Rini <trini@konsulko.com> wrote:
>>> Hey all,
>>>
>>> I've noticed now, but not dug into a problem that goes like this.  On
>>> every platform that I have tried NFS on now (and I wasn't a user before
>>> the test came in) I see:
>>> # nfs 80000000 /tftpboot/1MiBtest.bin
>>> link up on port 0, speed 1000, full duplex
>>> #################################################################
>>>           #################################################################
>>>           #################################################################
>>>           ##########T T T T
>>> done
>>> Bytes transferred = 1048576 (100000 hex)
>>>
>>> for the same 1MiB file.  The link line will vary from board to board but
>>> the end result is that I always see 4 T (for timeout) at the end of the
>>> transfer.  On boards where I am doing this on gigabit the initial
>>> transfer is fast enough that the timeout doesn't cause failure.  On the
>>> boards where I'm at 100Mbit instead however, I fail.
>> I have seen this also - what type of interface are you using?
> Does this happen in your testing, Guillaume?
>

No, I never saw those timeouts. I will try to reproduce here.
On which boards does it happen?

Guillaume

[U-Boot] NFS and timeouts?

[Tom Rini]

On Fri, Oct 14, 2016 at 11:13:11AM +0200, Guillaume Gardet wrote:
> Hi,
> 
> 
> Le 14/10/2016 ? 01:00, Joe Hershberger a ?crit :
> >On Fri, Oct 14, 2016 at 12:52 AM, Simon Glass <sjg@chromium.org> wrote:
> >>Hi Tom,
> >>
> >>On 13 October 2016 at 13:11, Tom Rini <trini@konsulko.com> wrote:
> >>>Hey all,
> >>>
> >>>I've noticed now, but not dug into a problem that goes like this.  On
> >>>every platform that I have tried NFS on now (and I wasn't a user before
> >>>the test came in) I see:
> >>># nfs 80000000 /tftpboot/1MiBtest.bin
> >>>link up on port 0, speed 1000, full duplex
> >>>#################################################################
> >>>          #################################################################
> >>>          #################################################################
> >>>          ##########T T T T
> >>>done
> >>>Bytes transferred = 1048576 (100000 hex)
> >>>
> >>>for the same 1MiB file.  The link line will vary from board to board but
> >>>the end result is that I always see 4 T (for timeout) at the end of the
> >>>transfer.  On boards where I am doing this on gigabit the initial
> >>>transfer is fast enough that the timeout doesn't cause failure.  On the
> >>>boards where I'm at 100Mbit instead however, I fail.
> >>I have seen this also - what type of interface are you using?
> >Does this happen in your testing, Guillaume?
> >
> 
> No, I never saw those timeouts. I will try to reproduce here.
> On which boards does it happen?

AM335x GP EVM, DRA72x EVM ("J6 Eco"), Beagleboard xM, RPi 3 (32 or 64bit
mode).  I suspect it's something either config or network related.
Everything is on the same 24 port gigabit switch.

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20161014/b3db8416/attachment.sig>

[U-Boot] NFS and timeouts?

[Guillaume Gardet]

Hi,


Le 14/10/2016 ? 15:40, Tom Rini a ?crit :
> On Fri, Oct 14, 2016 at 11:13:11AM +0200, Guillaume Gardet wrote:
>> Hi,
>>
>>
>> Le 14/10/2016 ? 01:00, Joe Hershberger a ?crit :
>>> On Fri, Oct 14, 2016 at 12:52 AM, Simon Glass <sjg@chromium.org> wrote:
>>>> Hi Tom,
>>>>
>>>> On 13 October 2016 at 13:11, Tom Rini <trini@konsulko.com> wrote:
>>>>> Hey all,
>>>>>
>>>>> I've noticed now, but not dug into a problem that goes like this.  On
>>>>> every platform that I have tried NFS on now (and I wasn't a user before
>>>>> the test came in) I see:
>>>>> # nfs 80000000 /tftpboot/1MiBtest.bin
>>>>> link up on port 0, speed 1000, full duplex
>>>>> #################################################################
>>>>>           #################################################################
>>>>>           #################################################################
>>>>>           ##########T T T T
>>>>> done
>>>>> Bytes transferred = 1048576 (100000 hex)
>>>>>
>>>>> for the same 1MiB file.  The link line will vary from board to board but
>>>>> the end result is that I always see 4 T (for timeout) at the end of the
>>>>> transfer.  On boards where I am doing this on gigabit the initial
>>>>> transfer is fast enough that the timeout doesn't cause failure.  On the
>>>>> boards where I'm at 100Mbit instead however, I fail.
>>>> I have seen this also - what type of interface are you using?
>>> Does this happen in your testing, Guillaume?
>>>
>> No, I never saw those timeouts. I will try to reproduce here.
>> On which boards does it happen?
> AM335x GP EVM, DRA72x EVM ("J6 Eco"), Beagleboard xM, RPi 3 (32 or 64bit
> mode).  I suspect it's something either config or network related.
> Everything is on the same 24 port gigabit switch.
>

I just tried 2016.11-rc2 on a Beagleboard xM with:
* nfs download from a NFSv2 server
* nfs download from a NFSv3 server
* tftpboot download from a TFTP server

And all are working fine here.


Guillaume