Testing selectable timestamping - where are the tools for this feature?

Testing selectable timestamping - where are the tools for this feature?

[Russell King (Oracle)]

Hi Kory,

I've finally found some cycles (and time when I'm next to the platform)
to test the selectable timestamping feature. However, I'm struggling to
get it to work.

In your email
https://lore.kernel.org/20240709-feature_ptp_netnext-v17-0-b5317f50df2a@bootlin.com/
you state that "You can test it with the ethtool source on branch
feature_ptp of: https://github.com/kmaincent/ethtool". I cloned this
repository, checked out the feature_ptp branch, and while building
I get the following warnings:

netlink/desc-ethtool.c:241:37: warning: ‘__ts_desc’ defined but not used [-Wunus
ed-const-variable=]
  241 | static const struct pretty_nla_desc __ts_desc[] = {
      |                                     ^~~~~~~~~
netlink/ts.c: In function ‘ts_get_reply_cb’:
netlink/ts.c:23:14: warning: unused variable ‘str’ [-Wunused-variable]
   23 |         char str[10] = {'\0'};
      |              ^~~
ethtool.c:4865:12: warning: ‘do_set_ptp’ defined but not used [-Wunused-functio ]
 4865 | static int do_set_ptp(struct cmd_context *ctx)
      |            ^~~~~~~~~~
ethtool.c:4839:12: warning: ‘do_get_ptp’ defined but not used [-Wunused-functio ]
 4839 | static int do_get_ptp(struct cmd_context *ctx)
      |            ^~~~~~~~~~
ethtool.c:4817:12: warning: ‘do_list_ptp’ defined but not used [-Wunused-function]
 4817 | static int do_list_ptp(struct cmd_context *ctx)
      |            ^~~~~~~~~~~

My conclusion is... your ethtool sources for testing this feature are
broken, or this is no longer the place to test this feature.

Presumably there _is_ something somewhere that allows one to exercise
this new code that Jakub merged on July 15th (commit 30b356005048)?
Please could you point me in the appropriate direction ASAP - time is
very short if I'm going to give this a test on the setup where both
the MAC and PHY support PTP. Essentially, this afternoon - or its
going to be at least a month before the next opportunity.

Thanks.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!

Re: Testing selectable timestamping - where are the tools for this feature?

[Kory Maincent]

Hello Russell,

On Wed, 13 Nov 2024 14:46:25 +0000
"Russell King (Oracle)" <linux@armlinux.org.uk> wrote:

> Hi Kory,
> 
> I've finally found some cycles (and time when I'm next to the platform)
> to test the selectable timestamping feature. However, I'm struggling to
> get it to work.
> 
> In your email
> https://lore.kernel.org/20240709-feature_ptp_netnext-v17-0-b5317f50df2a@bootlin.com/
> you state that "You can test it with the ethtool source on branch
> feature_ptp of: https://github.com/kmaincent/ethtool". I cloned this
> repository, checked out the feature_ptp branch, and while building
> I get the following warnings:
> 
> My conclusion is... your ethtool sources for testing this feature are
> broken, or this is no longer the place to test this feature.

Yeah, it was for v3 of the patch series. It didn't follow up to v19, I am using
ynl tool which is the easiest way to test it.
As there were a lot of changes along the way, updating ethtool every time was
not a good idea.

Use ynl tool. Commands are described in the last patch of the series:
https://lore.kernel.org/all/20241030-feature_ptp_netnext-v19-10-94f8aadc9d5c@bootlin.com/

You simply need to install python python-yaml and maybe others python
subpackages.
Copy the tool "tools/net/ynl" and the specs "Documentation/netlink/" on the
board.

Then run the ynl commands.

> Presumably there _is_ something somewhere that allows one to exercise
> this new code that Jakub merged on July 15th (commit 30b356005048)?

You can indeed test this merge with mainline ethtool.
Specially the commit 2dd356005901 which changes the default hwtstamp to MAC for
newly supported PHY PTP as you requested.
With your marvell PTP support patch it should still report and use ts info from
the MAC.

Regards,
-- 
Köry Maincent, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com

Re: Testing selectable timestamping - where are the tools for this feature?

[Russell King (Oracle)]

On Wed, Nov 13, 2024 at 04:16:02PM +0100, Kory Maincent wrote:
> Hello Russell,
> 
> On Wed, 13 Nov 2024 14:46:25 +0000
> "Russell King (Oracle)" <linux@armlinux.org.uk> wrote:
> 
> > Hi Kory,
> > 
> > I've finally found some cycles (and time when I'm next to the platform)
> > to test the selectable timestamping feature. However, I'm struggling to
> > get it to work.
> > 
> > In your email
> > https://lore.kernel.org/20240709-feature_ptp_netnext-v17-0-b5317f50df2a@bootlin.com/
> > you state that "You can test it with the ethtool source on branch
> > feature_ptp of: https://github.com/kmaincent/ethtool". I cloned this
> > repository, checked out the feature_ptp branch, and while building
> > I get the following warnings:
> > 
> > My conclusion is... your ethtool sources for testing this feature are
> > broken, or this is no longer the place to test this feature.
> 
> Yeah, it was for v3 of the patch series. It didn't follow up to v19, I am using
> ynl tool which is the easiest way to test it.
> As there were a lot of changes along the way, updating ethtool every time was
> not a good idea.
> 
> Use ynl tool. Commands are described in the last patch of the series:
> https://lore.kernel.org/all/20241030-feature_ptp_netnext-v19-10-94f8aadc9d5c@bootlin.com/
> 
> You simply need to install python python-yaml and maybe others python
> subpackages.
> Copy the tool "tools/net/ynl" and the specs "Documentation/netlink/" on the
> board.
> 
> Then run the ynl commands.

Thanks... fairly unweildly but at least it's functional. However,
running the first, I immediately find a problem:

# ./ynl/cli.py --spec netlink/specs/ethtool.yaml --no-schema --dump tsinfo-get --json '{"header":{"dev-name":"eth0"}}'

One would expect this to only return results for eth0 ? I get:

[{'header': {'dev-index': 1, 'dev-name': 'lo'},
  'timestamping': {'bits': {'bit': [{'index': 1, 'name': 'software-transmit'},
                                    {'index': 3, 'name': 'software-receive'},
                                    {'index': 4,
                                     'name': 'software-system-clock'}]},
                   'nomask': True,
                   'size': 17}},
... one entry per network device ...

Also, I don't see more than one timestamper on any interface - there
should be two on eth2, one for the MAC and one for the PHY. I see the
timestamper for the mvpp2 MAC, but nothing for the PHY. The PTP clock
on the PHY is definitely registered (/dev/ptp0), which means
phydev->mii_ts should be pointing to the MII timestamper for the PHY.

I've also tried with --json '{"header":{"dev-name":"eth2"}}' but no
difference - it still reports all interfaces and only one timestamper
for eth2.

I'm waiting for a kernel to build with nlmon modular to investigate
further.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!

Re: Testing selectable timestamping - where are the tools for this feature?

[Kory Maincent]

On Wed, 13 Nov 2024 15:57:56 +0000
"Russell King (Oracle)" <linux@armlinux.org.uk> wrote:

> On Wed, Nov 13, 2024 at 04:16:02PM +0100, Kory Maincent wrote:
> > Hello Russell,
> > 
> > On Wed, 13 Nov 2024 14:46:25 +0000
> > "Russell King (Oracle)" <linux@armlinux.org.uk> wrote:
> >   
> > > Hi Kory,
> > > 
> > > I've finally found some cycles (and time when I'm next to the platform)
> > > to test the selectable timestamping feature. However, I'm struggling to
> > > get it to work.
> > > 
> > > In your email
> > > https://lore.kernel.org/20240709-feature_ptp_netnext-v17-0-b5317f50df2a@bootlin.com/
> > > you state that "You can test it with the ethtool source on branch
> > > feature_ptp of: https://github.com/kmaincent/ethtool". I cloned this
> > > repository, checked out the feature_ptp branch, and while building
> > > I get the following warnings:
> > > 
> > > My conclusion is... your ethtool sources for testing this feature are
> > > broken, or this is no longer the place to test this feature.  
> > 
> > Yeah, it was for v3 of the patch series. It didn't follow up to v19, I am
> > using ynl tool which is the easiest way to test it.
> > As there were a lot of changes along the way, updating ethtool every time
> > was not a good idea.
> > 
> > Use ynl tool. Commands are described in the last patch of the series:
> > https://lore.kernel.org/all/20241030-feature_ptp_netnext-v19-10-94f8aadc9d5c@bootlin.com/
> > 
> > You simply need to install python python-yaml and maybe others python
> > subpackages.
> > Copy the tool "tools/net/ynl" and the specs "Documentation/netlink/" on the
> > board.
> > 
> > Then run the ynl commands.  
> 
> Thanks... fairly unweildly but at least it's functional. However,
> running the first, I immediately find a problem:
> 
> # ./ynl/cli.py --spec netlink/specs/ethtool.yaml --no-schema --dump
> tsinfo-get --json '{"header":{"dev-name":"eth0"}}'
> 
> One would expect this to only return results for eth0 ?

Indeed it should! That's weird, I will investigate.

> Also, I don't see more than one timestamper on any interface - there
> should be two on eth2, one for the MAC and one for the PHY. I see the
> timestamper for the mvpp2 MAC, but nothing for the PHY. The PTP clock
> on the PHY is definitely registered (/dev/ptp0), which means
> phydev->mii_ts should be pointing to the MII timestamper for the PHY.
> 
> I've also tried with --json '{"header":{"dev-name":"eth2"}}' but no
> difference - it still reports all interfaces and only one timestamper
> for eth2.

Sorry forgot to explain that you need to register PTP clock with the function
phydev_ptp_clock_register() in the PHY driver.

It will be changed in v20 as request by Jakub. I will save the hwtstamp source
and phydev pointer in the netdev core instead.

Regards,
-- 
Köry Maincent, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com

Re: Testing selectable timestamping - where are the tools for this feature?

[Kory Maincent]

On Wed, 13 Nov 2024 17:14:43 +0100
Kory Maincent <kory.maincent@bootlin.com> wrote:

> On Wed, 13 Nov 2024 15:57:56 +0000
> "Russell King (Oracle)" <linux@armlinux.org.uk> wrote:
> 
> > On Wed, Nov 13, 2024 at 04:16:02PM +0100, Kory Maincent wrote:  
> > > Hello Russell,
> > > 
> > > On Wed, 13 Nov 2024 14:46:25 +0000
> > > "Russell King (Oracle)" <linux@armlinux.org.uk> wrote:
> > >     
>  [...]  
> > > 
> > > Yeah, it was for v3 of the patch series. It didn't follow up to v19, I am
> > > using ynl tool which is the easiest way to test it.
> > > As there were a lot of changes along the way, updating ethtool every time
> > > was not a good idea.
> > > 
> > > Use ynl tool. Commands are described in the last patch of the series:
> > > https://lore.kernel.org/all/20241030-feature_ptp_netnext-v19-10-94f8aadc9d5c@bootlin.com/
> > > 
> > > You simply need to install python python-yaml and maybe others python
> > > subpackages.
> > > Copy the tool "tools/net/ynl" and the specs "Documentation/netlink/" on
> > > the board.
> > > 
> > > Then run the ynl commands.    
> > 
> > Thanks... fairly unweildly but at least it's functional. However,
> > running the first, I immediately find a problem:
> > 
> > # ./ynl/cli.py --spec netlink/specs/ethtool.yaml --no-schema --dump
> > tsinfo-get --json '{"header":{"dev-name":"eth0"}}'
> > 
> > One would expect this to only return results for eth0 ?  
> 
> Indeed it should! That's weird, I will investigate.
> 
> > Also, I don't see more than one timestamper on any interface - there
> > should be two on eth2, one for the MAC and one for the PHY. I see the
> > timestamper for the mvpp2 MAC, but nothing for the PHY. The PTP clock
> > on the PHY is definitely registered (/dev/ptp0), which means
> > phydev->mii_ts should be pointing to the MII timestamper for the PHY.
> > 
> > I've also tried with --json '{"header":{"dev-name":"eth2"}}' but no
> > difference - it still reports all interfaces and only one timestamper
> > for eth2.  
> 
> Sorry forgot to explain that you need to register PTP clock with the function
> phydev_ptp_clock_register() in the PHY driver.

And netdev_ptp_clock_register() in the MAC driver.

> 
> It will be changed in v20 as request by Jakub. I will save the hwtstamp source
> and phydev pointer in the netdev core instead.
> 
> Regards,



-- 
Köry Maincent, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com

Re: Testing selectable timestamping - where are the tools for this feature?

[Russell King (Oracle)]

On Wed, Nov 13, 2024 at 05:14:43PM +0100, Kory Maincent wrote:
> On Wed, 13 Nov 2024 15:57:56 +0000
> "Russell King (Oracle)" <linux@armlinux.org.uk> wrote:
> 
> > On Wed, Nov 13, 2024 at 04:16:02PM +0100, Kory Maincent wrote:
> > > Hello Russell,
> > > 
> > > On Wed, 13 Nov 2024 14:46:25 +0000
> > > "Russell King (Oracle)" <linux@armlinux.org.uk> wrote:
> > >   
> > > > Hi Kory,
> > > > 
> > > > I've finally found some cycles (and time when I'm next to the platform)
> > > > to test the selectable timestamping feature. However, I'm struggling to
> > > > get it to work.
> > > > 
> > > > In your email
> > > > https://lore.kernel.org/20240709-feature_ptp_netnext-v17-0-b5317f50df2a@bootlin.com/
> > > > you state that "You can test it with the ethtool source on branch
> > > > feature_ptp of: https://github.com/kmaincent/ethtool". I cloned this
> > > > repository, checked out the feature_ptp branch, and while building
> > > > I get the following warnings:
> > > > 
> > > > My conclusion is... your ethtool sources for testing this feature are
> > > > broken, or this is no longer the place to test this feature.  
> > > 
> > > Yeah, it was for v3 of the patch series. It didn't follow up to v19, I am
> > > using ynl tool which is the easiest way to test it.
> > > As there were a lot of changes along the way, updating ethtool every time
> > > was not a good idea.
> > > 
> > > Use ynl tool. Commands are described in the last patch of the series:
> > > https://lore.kernel.org/all/20241030-feature_ptp_netnext-v19-10-94f8aadc9d5c@bootlin.com/
> > > 
> > > You simply need to install python python-yaml and maybe others python
> > > subpackages.
> > > Copy the tool "tools/net/ynl" and the specs "Documentation/netlink/" on the
> > > board.
> > > 
> > > Then run the ynl commands.  
> > 
> > Thanks... fairly unweildly but at least it's functional. However,
> > running the first, I immediately find a problem:
> > 
> > # ./ynl/cli.py --spec netlink/specs/ethtool.yaml --no-schema --dump
> > tsinfo-get --json '{"header":{"dev-name":"eth0"}}'
> > 
> > One would expect this to only return results for eth0 ?
> 
> Indeed it should! That's weird, I will investigate.
> 
> > Also, I don't see more than one timestamper on any interface - there
> > should be two on eth2, one for the MAC and one for the PHY. I see the
> > timestamper for the mvpp2 MAC, but nothing for the PHY. The PTP clock
> > on the PHY is definitely registered (/dev/ptp0), which means
> > phydev->mii_ts should be pointing to the MII timestamper for the PHY.
> > 
> > I've also tried with --json '{"header":{"dev-name":"eth2"}}' but no
> > difference - it still reports all interfaces and only one timestamper
> > for eth2.
> 
> Sorry forgot to explain that you need to register PTP clock with the function
> phydev_ptp_clock_register() in the PHY driver.

That is certainly inconvenient. Marvell's PHY PTP/TAI implementation is
used elsewhere, so the TAI driver is not specific to it being on a PHY.
The drivers/ptp/marvell*.c that you find in my patches is meant to
eventually replace drivers/net/dsa/mv88e6xxx/ptp.c - the same hardware
block is in Marvell's DSA switches.

> It will be changed in v20 as request by Jakub. I will save the hwtstamp source
> and phydev pointer in the netdev core instead.

I think I'll shelve this until that happens, hopefully that will mean
my existing code will work and remain a re-usable library.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!

Re: Testing selectable timestamping - where are the tools for this feature?

[Kory Maincent]

On Wed, 13 Nov 2024 16:52:20 +0000
"Russell King (Oracle)" <linux@armlinux.org.uk> wrote:

> On Wed, Nov 13, 2024 at 05:14:43PM +0100, Kory Maincent wrote:
> > On Wed, 13 Nov 2024 15:57:56 +0000
> > "Russell King (Oracle)" <linux@armlinux.org.uk> wrote:
> >   
> > > On Wed, Nov 13, 2024 at 04:16:02PM +0100, Kory Maincent wrote:  
>  [...]  
>  [...]  
>  [...]  
> > > 
> > > Thanks... fairly unweildly but at least it's functional. However,
> > > running the first, I immediately find a problem:
> > > 
> > > # ./ynl/cli.py --spec netlink/specs/ethtool.yaml --no-schema --dump
> > > tsinfo-get --json '{"header":{"dev-name":"eth0"}}'
> > > 
> > > One would expect this to only return results for eth0 ?  
> > 
> > Indeed it should! That's weird, I will investigate.
> >   
> > > Also, I don't see more than one timestamper on any interface - there
> > > should be two on eth2, one for the MAC and one for the PHY. I see the
> > > timestamper for the mvpp2 MAC, but nothing for the PHY. The PTP clock
> > > on the PHY is definitely registered (/dev/ptp0), which means
> > > phydev->mii_ts should be pointing to the MII timestamper for the PHY.
> > > 
> > > I've also tried with --json '{"header":{"dev-name":"eth2"}}' but no
> > > difference - it still reports all interfaces and only one timestamper
> > > for eth2.  
> > 
> > Sorry forgot to explain that you need to register PTP clock with the
> > function phydev_ptp_clock_register() in the PHY driver.  
> 
> That is certainly inconvenient. Marvell's PHY PTP/TAI implementation is
> used elsewhere, so the TAI driver is not specific to it being on a PHY.
> The drivers/ptp/marvell*.c that you find in my patches is meant to
> eventually replace drivers/net/dsa/mv88e6xxx/ptp.c - the same hardware
> block is in Marvell's DSA switches.
> 
> > It will be changed in v20 as request by Jakub. I will save the hwtstamp
> > source and phydev pointer in the netdev core instead.  
> 
> I think I'll shelve this until that happens, hopefully that will mean
> my existing code will work and remain a re-usable library.

Ack, Thanks anyway.

-- 
Köry Maincent, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com

Re: Testing selectable timestamping - where are the tools for this feature?

[Russell King (Oracle)]

On Wed, Nov 13, 2024 at 03:57:56PM +0000, Russell King (Oracle) wrote:
> On Wed, Nov 13, 2024 at 04:16:02PM +0100, Kory Maincent wrote:
> > You simply need to install python python-yaml and maybe others python
> > subpackages.
> > Copy the tool "tools/net/ynl" and the specs "Documentation/netlink/" on the
> > board.
> > 
> > Then run the ynl commands.
> 
> Thanks... fairly unweildly but at least it's functional. However,
> running the first, I immediately find a problem:
> 
> # ./ynl/cli.py --spec netlink/specs/ethtool.yaml --no-schema --dump tsinfo-get --json '{"header":{"dev-name":"eth0"}}'
> 
> One would expect this to only return results for eth0 ? I get:

Here's the nlmon packet capture for the ynl request:

        0x0000:  0004 0338 0000 0000 0000 0000 0000 0010  ...8............
        0x0010:  2400 0000 1500 0503 f9d3 0000 0000 0000  $...............
        0x0020:  1901 0000 1000 0180 0900 0200 6574 6832  ............eth2
        0x0030:  0000 0000                                ....

Length: 0x00000024
Family ID: 0x0015 (ethtool)
Flags: 0x0305 (Return all matching, Specify tree root, ack, request)
Sequence: 0x0000d3f9
Port ID: 0x00000000
Command: 0x19
Family Version: 0x01

Then 16 bytes of data that does contain the interface name given to
YNL. I haven't parsed that, it seems to require manual effort to do
so as wireshark is unable to do so.

I'd be guessing to draw any conclusions from this without deeper
analysis.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!

Re: Testing selectable timestamping - where are the tools for this feature?

[Kory Maincent]

On Thu, 14 Nov 2024 20:20:58 +0000
"Russell King (Oracle)" <linux@armlinux.org.uk> wrote:

> On Wed, Nov 13, 2024 at 03:57:56PM +0000, Russell King (Oracle) wrote:
> > On Wed, Nov 13, 2024 at 04:16:02PM +0100, Kory Maincent wrote:  
> > > You simply need to install python python-yaml and maybe others python
> > > subpackages.
> > > Copy the tool "tools/net/ynl" and the specs "Documentation/netlink/" on
> > > the board.
> > > 
> > > Then run the ynl commands.  
> > 
> > Thanks... fairly unweildly but at least it's functional. However,
> > running the first, I immediately find a problem:
> > 
> > # ./ynl/cli.py --spec netlink/specs/ethtool.yaml --no-schema --dump
> > tsinfo-get --json '{"header":{"dev-name":"eth0"}}'
> > 
> > One would expect this to only return results for eth0 ? I get:  
> 
> Here's the nlmon packet capture for the ynl request:
> 
>         0x0000:  0004 0338 0000 0000 0000 0000 0000 0010  ...8............
>         0x0010:  2400 0000 1500 0503 f9d3 0000 0000 0000  $...............
>         0x0020:  1901 0000 1000 0180 0900 0200 6574 6832  ............eth2
>         0x0030:  0000 0000                                ....
> 
> Length: 0x00000024
> Family ID: 0x0015 (ethtool)
> Flags: 0x0305 (Return all matching, Specify tree root, ack, request)
> Sequence: 0x0000d3f9
> Port ID: 0x00000000
> Command: 0x19
> Family Version: 0x01
> 
> Then 16 bytes of data that does contain the interface name given to
> YNL. I haven't parsed that, it seems to require manual effort to do
> so as wireshark is unable to do so.
> 
> I'd be guessing to draw any conclusions from this without deeper
> analysis.

Ok thanks, I will take a look at it

Regards,
-- 
Köry Maincent, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com