Hi Matan,

Hi Konstantin
From: Ananyev, Konstantin, Monday, January 22, 2018 10:49 PM

quoted

Hi Matan,

quoted

-----Original Message-----
From: Matan Azrad [mailto:matan@mellanox.com]
Sent: Monday, January 22, 2018 1:23 PM
To: Ananyev, Konstantin <redacted>; Gaëtan Rivet
[off-list ref]
Cc: Thomas Monjalon <redacted>; Wu, Jingjing
[off-list ref]; dev@dpdk.org; Neil Horman
[off-list ref]; Richardson, Bruce
[off-list ref]
Subject: RE: [PATCH v3 7/7] app/testpmd: adjust ethdev port ownership


Hi
From: Ananyev, Konstantin [mailto:konstantin.ananyev@intel.com]

quoted

Hi lads,

quoted

Hi Matan,

On Fri, Jan 19, 2018 at 01:35:10PM +0000, Matan Azrad wrote:

quoted

Hi Konstantin

From: Ananyev, Konstantin, Friday, January 19, 2018 3:09 PM

quoted

quoted

-----Original Message-----
From: Matan Azrad [mailto:matan@mellanox.com]
Sent: Friday, January 19, 2018 12:52 PM
To: Ananyev, Konstantin <redacted>;
Thomas Monjalon [off-list ref]; Gaetan Rivet

[off-list ref];

quoted

Wu, Jingjing [off-list ref]
Cc: dev@dpdk.org; Neil Horman <nhorman@tuxdriver.com>;
Richardson, Bruce [off-list ref]
Subject: RE: [PATCH v3 7/7] app/testpmd: adjust ethdev port
ownership

Hi Konstantin

From: Ananyev, Konstantin, Friday, January 19, 2018 2:38 PM

quoted

To: Matan Azrad <redacted>; Thomas Monjalon
[off-list ref]; Gaetan Rivet

[off-list ref];

quoted

quoted

quoted

Wu,

quoted

quoted

Jingjing [off-list ref]
Cc: dev@dpdk.org; Neil Horman <nhorman@tuxdriver.com>;
Richardson, Bruce [off-list ref]
Subject: RE: [PATCH v3 7/7] app/testpmd: adjust ethdev
port ownership

Hi Matan,

quoted

-----Original Message-----
From: Matan Azrad [mailto:matan@mellanox.com]
Sent: Thursday, January 18, 2018 4:35 PM
To: Thomas Monjalon <redacted>; Gaetan Rivet
[off-list ref]; Wu, Jingjing
[off-list ref]
Cc: dev@dpdk.org; Neil Horman <nhorman@tuxdriver.com>;

Richardson,

quoted

quoted

quoted

Bruce [off-list ref]; Ananyev, Konstantin
[off-list ref]
Subject: [PATCH v3 7/7] app/testpmd: adjust ethdev port
ownership

Testpmd should not use ethdev ports which are managed by
other DPDK entities.

Set Testpmd ownership to each port which is not used by
other entity and prevent any usage of ethdev ports which
are not owned by

Testpmd.

quoted

quoted

quoted

Signed-off-by: Matan Azrad <redacted>
---
 app/test-pmd/cmdline.c      | 89 +++++++++++++++++++------

----

quoted

quoted

-------

quoted

quoted

quoted

----

quoted

quoted

-----

quoted

 app/test-pmd/cmdline_flow.c |  2 +-
 app/test-pmd/config.c       | 37 ++++++++++---------
 app/test-pmd/parameters.c   |  4 +-
 app/test-pmd/testpmd.c      | 63 ++++++++++++++++++++----

----

quoted

quoted

----

quoted

quoted

quoted

quoted

quoted

quoted

 app/test-pmd/testpmd.h      |  3 ++
 6 files changed, 103 insertions(+), 95 deletions(-)

diff --git a/app/test-pmd/cmdline.c

b/app/test-pmd/cmdline.c index
31919ba..6199c64 100644

--- a/app/test-pmd/cmdline.c
+++ b/app/test-pmd/cmdline.c

@@ -1394,7 +1394,7 @@ struct cmd_config_speed_all {
 			&link_speed) < 0)
 		return;

-	RTE_ETH_FOREACH_DEV(pid) {
+	RTE_ETH_FOREACH_DEV_OWNED_BY(pid,

my_owner.id) {

quoted

quoted

quoted

quoted

quoted

quoted

quoted

Why do we need all these changes?
As I understand you changed definition of
RTE_ETH_FOREACH_DEV(), so no testpmd should work ok
default

(no_owner case).

quoted

quoted

quoted

quoted

quoted

Am I missing something here?

Now, After Gaetan suggestion RTE_ETH_FOREACH_DEV(pid) will
iterate

over all valid and ownerless ports.

Yes.

quoted

Here Testpmd wants to iterate over its owned ports.

Why? Why it can't just iterate over all valid and ownerless ports?
As I understand it would be enough to fix current problems and
would allow us to avoid any changes in testmpd (which I think
is a good

thing).

quoted

quoted

Yes, I understand that this big change is very daunted, But I
think the current a lot of bugs in testpmd(regarding port
ownership) even more

daunted.

quoted

Look,
Testpmd initiates some of its internal databases depends on
specific port iteration, In some time someone may take ownership
of Testpmd

ports and testpmd will continue to touch them.

But if someone will take the ownership (assign new owner_id) that
port will not appear in RTE_ETH_FOREACH_DEV() any more.

Yes, but testpmd sometimes depends on previous iteration using internal

database.

quoted

So it uses internal database that was updated by old iteration.

That sounds like just a bug in testpmd that need to be fixed, no?

If Testpmd already took ownership for these ports(like I did), it is ok.

Hmm, why not just to fix testpmd, if there is a bug?
As I said all control ops here are done by one thread, so it should be pretty easy.
Or are you talking about race conditions?

quoted

Any particular places where outdated device info is used?

For example, look for the stream management in testpmd(I think I saw it there).

Anything particular?

quoted

quoted

quoted

quoted

If I look back on the fail-safe, its sole purpose is to have
seamless hotplug with existing applications.

Port ownership is a genericization of some functions introduced by
the fail-safe, that could structure DPDK further. It should allow
applications to have a seamless integration with subsystems using
port ownership. Without this, port ownership cannot be used.

Testpmd should be fixed, but follow the most common design
patterns of DPDK applications. Going with port ownership seems
like a paradigm shift.

quoted

In addition
Using the old iterator in some places in testpmd will cause a
race for run-

time new ports(can be created by failsafe or any hotplug code):

quoted

quoted

- testpmd finds an ownerless port(just now created) by the old
iterator and start traffic there,
- failsafe takes ownership of this new port and start traffic there.
Problem!

Could you shed a bit more light here - it would be race condition
between whom and whom?

Sure.

quoted

As I remember in testpmd all control ops are done within one thread
(main lcore).

But other dpdk entity can use another thread, for example:
Failsafe uses the host thread(using alarm callback) to create a new port and

to take ownership of a port.

Hm, and you create new ports inside failsafe PMD, right and then set new
owner_id for it?

Yes.

quoted

And all this in alarm in interrupt thread?

Yes.

quoted

If so I wonder how you can guarantee that no-one else will set different
owner_id between
rte_eth_dev_allocate() and rte_eth_dev_owner_set()?

I check it (see failsafe patch to this series - V5).
Function: fs_bus_init.

You are talking about that peace of code:
+		ret = rte_eth_dev_owner_set(pid, &PRIV(dev)->my_owner);
+		if (ret) {
+			INFO("sub_device %d owner set failed (%s),"
+			     " will try again later", i, strerror(ret));
+			continue;

right?
So you just wouldn't include that device into your failsafe device.
But that probably not what user wanted, especially if he bothered to create
a special new low-level device for you.
If that' s the use case, then I think you need to set device ownership at creation time -
inside dev_allocate().
Again that would avoid such racing conditions inside testpmd.

quoted

Could you point me to that place (I am not really familiar with familiar with
failsafe code)?

quoted

The race:
Testpmd iterates over all ports by the master thread.
Failsafe takes ownership of a port by the host thread and start using it.
=> The two dpdk entities may use the device at same time!

Ok, if failsafe really assigns its owner_id(s) to ports that are already in use by
the app, then how such scheme supposed to work at all?

If the app works well (with the new rules) it already took ownership and failsafe will see it and will wait until the application release it.

Ok, and why application would need to release it?
How it would know that failsafe device wants to use it now?
Again where is a guarantee that after app released it some other entity wouldn't grab it for itself?

Every dpdk entity should know which port it wants to manage,
If 2 entities want to manage the same device -  it can be ok and port ownership can synchronize the usage.

Probably, application which will run fail-safe wants to manage only the fail-safe port and therefor to take ownership only for it.

quoted

I.E. application has a port - it assigns some owner_id != 0 to it, then PMD tries
to set its owner_id tot the same port.
Obviously failsafe's set_owner() will always fail in such case.

Yes, and will try again after some time.

Same question again - how app will know that it has to release the port ownership?

quoted

From what I hear we need to introduce a concept of 'default owner id'.
I.E. when failsafe PMD is created - user assigns some owner_id to it (default).
Then failsafe PMD generates it's own owner_id and assigns it only to the
ports whose current owner_id is equal either 0 or 'default' owner_id.

It is a suggestion and we need to think about it more (I'm talking about it with Gaetan in another thread).
Actually I think, if we want a generic solution to the generic problem the current solution is ok.

From what I heard - every app that wants to use failsafe PMD would require quite a lot of changes.

It doesn't look ok to me.

quoted

quoted

Obeying the new ownership rules can prevent all these races.

When we discussed RFC of owner_id patch, I thought we all agreed that
owner_id  API shouldn't be mandatory - i.e. existing apps not required to
change to work normally with that.

Yes, it is not mandatory if app doesn't use hotplug.

I think with hotplug it is mandatory in the most cases.

Yes in failsafe you always install this alarm handler, so even
if the app would have its own  way to handle hotplug  devices -
it would suddenly need to use this new owner API - even if it doesn't need to.
Why it has to be?

And it can ease the secondary process model too.

Again, in the generic ownership problem as discussed in RFC:
Every entity, include app, should know which ports it wants to manage and to take ownership only for them.

quoted

Though right now it seems that application changes seems necessary, at least
to work ok with failsafe PMD.

And for solving the generic problem of ownership.(will defend from future issues by sure).

quoted

Which makes we wonder was it some sort of misunderstanding or we did we
do something wrong here?

Mistakes can be done all the time, but I think we are all understand the big issue of ownership and how the current solution solves it.
fail-safe it is just a current example for the problems which are possible because of the generic ownership issue.

Honestly that seems too much changes for the app just to make failsafe PMD work correctly.
IMO - It should be some way to support it without causing changes in each DPDK application  -
otherwise something is wrong with the PMD itself.
If let say that ownership model is required to make failsafe PMD to operate -
it should be done in a transparent way to the user.
Probably something like Gaetan suggested in another mail or so.
Konstantin

Thanks,
Matan

quoted

Konstantin

quoted

quoted

The only way to attach/detach port with it - invoke testpmd CLI
"attach/detach" port.

Konstantin

quoted

Testpmd does not handle detection of new port. If it did, testing
fail-safe with it would be wrong.

At startup, RTE_ETH_FOREACH_DEV already fixed the issue of
registering DEFERRED ports. There are still remaining issues
regarding this, but I think they should be fixed. The architecture
does not need to be completely moved to port ownership.

If anything, this should serve as a test for your API with common
applications. I think you'd prefer to know and debug with testpmd
instead of firing up VPP or something like that to determine what
went wrong with using the fail-safe.

quoted

In addition
As a good example for well-done application (free from ownership
bugs) I tried here to adjust Tespmd to the new rules and BTW to
fix a

lot of bugs.

Testpmd has too much cruft, it won't ever be a good example of a
well-done application.

If you want to demonstrate ownership, I think you should start an
example application demonstrating race conditions and their mitigation.

I think that would be interesting for many DPDK users.

quoted

So actually applications which are not aware to the port
ownership still are exposed to races, but if there use the old
iterator(with the new

change) the amount of races decreases.

quoted

Thanks, Matan.

quoted

Konstantin

quoted

I added to Testpmd ability to take an ownership of ports as
the new ownership and synchronization rules suggested, Since
Tespmd is a DPDK entity which wants that no one will touch
its owned ports, It must allocate

an unique ID, set owner for its ports (see in main function)
and recognizes them by its owner ID.

quoted

quoted

Konstantin

Regards,
--
Gaëtan Rivet
6WIND