linuxptp-users

[Linuxptp-users] Wildly jumping offsets casued by phc2sys

[Robb <rba...@gm...>]

For whatever reason phc2sys seems to be blowing up the offsets.

My set up is as follows:

Two CentOS 6.5 machines
2.6.32-504.16.2.el6.x86_64
linuxptp-1.3-1.el6.x86_64

Machine1 eth1 [igb    5.0.5-k]    Intel I210 (rev 03)
Machine2 eth0 [e1000e 2.3.2-k] Intel I217-LM (rev 05)

Machine1 eth1 hooked directly to Machine2 eth0

Machine1 (master):
sudo ptp4l -i eth1 -m -2
sudo phc2sys -s eth1 -c CLOCK_REALTIME -w

Machine2:
sudo ptp4l -i eth0 -m -2
sudo phc2sys -s eth0 -c CLOCK_REALTIME -w

Normally the offset blows up immediately unless I sync the machines with NTP
before doing anything else.  If I do get them to sync it doesn't last very
long and the following will happen on Machine2:

16:30:03 ptp4l: [4679.655] master offset        777 s2 freq   +4403
path delay     67209
16:30:04 phc2sys: [4679.935] sys offset      -658 s2 freq   +3877 delay   1836
16:30:04 ptp4l: [4680.655] master offset       2280 s2 freq   +6139
path delay     67209
16:30:05 phc2sys: [4680.936] sys offset      1353 s2 freq   +5690 delay   1504
16:30:05 ptp4l: [4681.655] master offset      -1349 s2 freq   +3194
path delay     67123
16:30:06 phc2sys: [4681.936] sys offset    -14484 s2 freq   -9741 delay   1840
16:30:06 ptp4l: [4682.656] master offset 70368744179084 s2 freq
+599999999 path delay     67123
16:30:07 phc2sys: [4682.936] sys offset -70368575979115 s2 freq
-500000 delay   1440
16:30:07 ptp4l: [4683.656] master offset 70368127382914 s2 freq
+599999999 path delay  17621798
16:30:08 phc2sys: [4683.936] sys offset -70367975616192 s2 freq
-500000 delay   1842
16:30:08 ptp4l: [4684.657] master offset 70367522033588 s2 freq
+599999999 path delay  22920656
16:30:09 phc2sys: [4684.937] sys offset -70367375237559 s2 freq
-500000 delay   1441


I have read that there was some problems with the e1000e driver so I tried with
the 3.1.0.2-NAPI version, but still have the same problem as above.

Does anyone have any ideas what could be the problem here?

Thanks.

Re: [Linuxptp-users] Wildly jumping offsets casued by phc2sys

[Gary E. M. <ge...@re...>]

Yo Robb!

On Fri, 29 May 2015 16:59:29 -0500
Robb <rba...@gm...> wrote:

> For whatever reason phc2sys seems to be blowing up the offsets.

I have the same problem with hardware sync.  There was a discussion on
linuxptp-devel on this last May, but since the email archives seem to be
down right now I can't point you at it.

https://sourceforge.net/p/linuxptp/mailman/linuxptp-devel/

> Does anyone have any ideas what could be the problem here?

Seemingly not.

RGDS
GARY
---------------------------------------------------------------------------
Gary E. Miller Rellim 109 NW Wilmington Ave., Suite E, Bend, OR 97703
	ge...@re...  Tel:+1(541)382-8588

Re: [Linuxptp-users] Wildly jumping offsets casued by phc2sys

[Gary E. M. <ge...@re...>]

Yo Robb!

On Fri, 29 May 2015 15:23:50 -0700
"Gary E. Miller" <ge...@re...> wrote:

> There was a discussion on linuxptp-devel on this last May,

Whoops, I meant March.

RGDS
GARY
---------------------------------------------------------------------------
Gary E. Miller Rellim 109 NW Wilmington Ave., Suite E, Bend, OR 97703
	ge...@re...  Tel:+1(541)382-8588

Re: [Linuxptp-users] Wildly jumping offsets casued by phc2sys

[Richard C. <ric...@gm...>]

On Fri, May 29, 2015 at 03:23:50PM -0700, Gary E. Miller wrote:
> 
> I have the same problem with hardware sync.  There was a discussion on
> linuxptp-devel on this last May, but since the email archives seem to be
> down right now I can't point you at it.

The SF archives are unusable, IMHO, even when they are up.  Use gmane
instead.  The link to the discussion is:

   http://thread.gmane.org/gmane.comp.linux.ptp.devel/2494/

Thanks,
Richard

Re: [Linuxptp-users] Wildly jumping offsets casued by phc2sys

[Richard C. <ric...@gm...>]

On Fri, May 29, 2015 at 04:59:29PM -0500, Robb wrote:
> For whatever reason phc2sys seems to be blowing up the offsets.
> 
> My set up is as follows:
> 
> Two CentOS 6.5 machines
> 2.6.32-504.16.2.el6.x86_64
> linuxptp-1.3-1.el6.x86_64
> 
> Machine1 eth1 [igb    5.0.5-k]    Intel I210 (rev 03)
> Machine2 eth0 [e1000e 2.3.2-k] Intel I217-LM (rev 05)

The I217-LM has a known HW bug.

> I have read that there was some problems with the e1000e driver so I tried with
> the 3.1.0.2-NAPI version, but still have the same problem as above.
> 
> Does anyone have any ideas what could be the problem here?

This isn't a driver problem but rather a HW bug.  I don't have that HW,
and I have not yet heard of any SW (driver) work arounds.  You might
ask on the e1000e list if Intel has a solution to this problem.

Thanks,
Richard

Re: [Linuxptp-users] Wildly jumping offsets casued by phc2sys

[Robb <rba...@gm...>]

Yanir at:
https://sourceforge.net/p/e1000/mailman/message/34158396/
was able to help me with this problem.

He provided me with a patch to get this working.
I have tested it and it has been working for > hour with
no errors!

Here is a the patch + what I did in case it helps anyone else.

wget -O e1000e-3.1.0.2.tar.gz
"http://sourceforge.net/projects/e1000/files/e1000e%20stable/3.1.0.2/e1000e-3.1.0.2.tar.gz/download"
tar xvf e1000e-3.1.0.2.tar.gz
cd e1000e-3.1.0.2/src; patch < ../../fix_systim_issues-3.1.0.2.patch
sudo make install

$ cat fix_systim_issues-3.1.0.2.patch
--- netdev.c.orig 2015-06-01 16:49:37.802312304 +0300
+++ netdev.c 2015-06-01 16:51:19.468306452 +0300
@@ -4614,7 +4614,10 @@ static cycle_t e1000e_cyclecounter_read(
  cycle_t systim, systim_next;

  /* latch SYSTIMH on read of SYSTIML */
- systim = (cycle_t)er32(SYSTIML);
+ u32 systim_overflow_latch_fix = 0x3FFFFFFF;
+ do {
+ systim = (cycle_t)er32(SYSTIML);
+ } while (systim > systim_overflow_latch_fix);
  systim |= (cycle_t)er32(SYSTIMH) << 32;

  if ((hw->mac.type == e1000_82574) || (hw->mac.type == e1000_82583)) {



I apologize, my email composer (gmail) seems to delete the tabs in the patch.



On Sat, May 30, 2015 at 12:11 AM, Richard Cochran
<ric...@gm...> wrote:
> On Fri, May 29, 2015 at 04:59:29PM -0500, Robb wrote:
>> For whatever reason phc2sys seems to be blowing up the offsets.
>>
>> My set up is as follows:
>>
>> Two CentOS 6.5 machines
>> 2.6.32-504.16.2.el6.x86_64
>> linuxptp-1.3-1.el6.x86_64
>>
>> Machine1 eth1 [igb    5.0.5-k]    Intel I210 (rev 03)
>> Machine2 eth0 [e1000e 2.3.2-k] Intel I217-LM (rev 05)
>
> The I217-LM has a known HW bug.
>
>> I have read that there was some problems with the e1000e driver so I tried with
>> the 3.1.0.2-NAPI version, but still have the same problem as above.
>>
>> Does anyone have any ideas what could be the problem here?
>
> This isn't a driver problem but rather a HW bug.  I don't have that HW,
> and I have not yet heard of any SW (driver) work arounds.  You might
> ask on the e1000e list if Intel has a solution to this problem.
>
> Thanks,
> Richard

Re: [Linuxptp-users] Wildly jumping offsets casued by phc2sys

[Keller, J. E <jac...@in...>]

Ah. I was about to forward you to Yanir when I saw this email. I believe
the e1000e team is currently working on validating a release that can be
put on sourceforge which will include this fix.

Regards,
Jake

On Mon, 2015-06-01 at 10:42 -0500, Robb wrote:
> Yanir at:
> https://sourceforge.net/p/e1000/mailman/message/34158396/
> was able to help me with this problem.
> 
> He provided me with a patch to get this working.
> I have tested it and it has been working for > hour with
> no errors!
> 
> Here is a the patch + what I did in case it helps anyone else.
> 
> wget -O e1000e-3.1.0.2.tar.gz
> "http://sourceforge.net/projects/e1000/files/e1000e%20stable/3.1.0.2/e1000e-3.1.0.2.tar.gz/download"
> tar xvf e1000e-3.1.0.2.tar.gz
> cd e1000e-3.1.0.2/src; patch < ../../fix_systim_issues-3.1.0.2.patch
> sudo make install
> 
> $ cat fix_systim_issues-3.1.0.2.patch
> --- netdev.c.orig 2015-06-01 16:49:37.802312304 +0300
> +++ netdev.c 2015-06-01 16:51:19.468306452 +0300
> @@ -4614,7 +4614,10 @@ static cycle_t e1000e_cyclecounter_read(
>   cycle_t systim, systim_next;
> 
>   /* latch SYSTIMH on read of SYSTIML */
> - systim = (cycle_t)er32(SYSTIML);
> + u32 systim_overflow_latch_fix = 0x3FFFFFFF;
> + do {
> + systim = (cycle_t)er32(SYSTIML);
> + } while (systim > systim_overflow_latch_fix);
>   systim |= (cycle_t)er32(SYSTIMH) << 32;
> 
>   if ((hw->mac.type == e1000_82574) || (hw->mac.type == e1000_82583)) {
> 
> 
> 
> I apologize, my email composer (gmail) seems to delete the tabs in the patch.
> 
> 
> 
> On Sat, May 30, 2015 at 12:11 AM, Richard Cochran
> <ric...@gm...> wrote:
> > On Fri, May 29, 2015 at 04:59:29PM -0500, Robb wrote:
> >> For whatever reason phc2sys seems to be blowing up the offsets.
> >>
> >> My set up is as follows:
> >>
> >> Two CentOS 6.5 machines
> >> 2.6.32-504.16.2.el6.x86_64
> >> linuxptp-1.3-1.el6.x86_64
> >>
> >> Machine1 eth1 [igb    5.0.5-k]    Intel I210 (rev 03)
> >> Machine2 eth0 [e1000e 2.3.2-k] Intel I217-LM (rev 05)
> >
> > The I217-LM has a known HW bug.
> >
> >> I have read that there was some problems with the e1000e driver so I tried with
> >> the 3.1.0.2-NAPI version, but still have the same problem as above.
> >>
> >> Does anyone have any ideas what could be the problem here?
> >
> > This isn't a driver problem but rather a HW bug.  I don't have that HW,
> > and I have not yet heard of any SW (driver) work arounds.  You might
> > ask on the e1000e list if Intel has a solution to this problem.
> >
> > Thanks,
> > Richard
> 
> ------------------------------------------------------------------------------
> _______________________________________________
> Linuxptp-users mailing list
> Lin...@li...
> https://lists.sourceforge.net/lists/listinfo/linuxptp-users