DPDK patches and discussions
 help / color / mirror / Atom feed
* [dpdk-dev]  [PATCH] ethdev: add lock-less txq capability flag
@ 2017-04-21 12:22 Jerin Jacob
  2017-04-24 12:34 ` Thomas Monjalon
  2017-07-06  6:21 ` [dpdk-dev] [PATCH v2] ethdev: introduce lock-free " Jerin Jacob
  0 siblings, 2 replies; 12+ messages in thread
From: Jerin Jacob @ 2017-04-21 12:22 UTC (permalink / raw)
  To: dev
  Cc: thomas, bruce.richardson, harry.van.haaren, hemant.agrawal,
	gage.eads, nipun.gupta, santosh.shukla, Jerin Jacob

if this flag is advertised by a PMD, Multiple threads can
invoke rte_eth_tx_burst() concurrently on the same tx queue
without SW lock. This is an HW feature found in some NICs
and useful in the following use cases if HW supports it.

1) Remove explicit spinlock in some application where cores
to tx queues are not mapped 1:1.
example: OVS has such instance
https://github.com/openvswitch/ovs/blob/master/lib/netdev-dpdk.c#L292
https://github.com/openvswitch/ovs/blob/master/lib/netdev-dpdk.c#L1821

2) In eventdev use case, Avoid dedicating a separate tx core
for transmit and thus enables more scaling as all workers
can transmit the packets.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
---
 lib/librte_ether/rte_ethdev.h | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h
index f9de9a2b6..696bc9d44 100644
--- a/lib/librte_ether/rte_ethdev.h
+++ b/lib/librte_ether/rte_ethdev.h
@@ -899,6 +899,10 @@ struct rte_eth_conf {
 #define DEV_TX_OFFLOAD_IPIP_TNL_TSO     0x00000800    /**< Used for tunneling packet. */
 #define DEV_TX_OFFLOAD_GENEVE_TNL_TSO   0x00001000    /**< Used for tunneling packet. */
 #define DEV_TX_OFFLOAD_MACSEC_INSERT    0x00002000
+#define DEV_TX_OFFLOAD_TXQ_MT_LOCKFREE	0x00004000
+/**< Multiple threads can invoke rte_eth_tx_burst() concurrently on the same
+ * tx queue without SW lock.
+ */
 
 struct rte_pci_device;
 
-- 
2.12.2

^ permalink raw reply	[flat|nested] 12+ messages in thread

* Re: [dpdk-dev] [PATCH] ethdev: add lock-less txq capability flag
  2017-04-21 12:22 [dpdk-dev] [PATCH] ethdev: add lock-less txq capability flag Jerin Jacob
@ 2017-04-24 12:34 ` Thomas Monjalon
  2017-04-27 11:00   ` Jerin Jacob
  2017-07-06  6:21 ` [dpdk-dev] [PATCH v2] ethdev: introduce lock-free " Jerin Jacob
  1 sibling, 1 reply; 12+ messages in thread
From: Thomas Monjalon @ 2017-04-24 12:34 UTC (permalink / raw)
  To: Jerin Jacob
  Cc: dev, bruce.richardson, harry.van.haaren, hemant.agrawal,
	gage.eads, nipun.gupta, santosh.shukla

21/04/2017 14:22, Jerin Jacob:
> if this flag is advertised by a PMD, Multiple threads can
> invoke rte_eth_tx_burst() concurrently on the same tx queue
> without SW lock. This is an HW feature found in some NICs
> and useful in the following use cases if HW supports it.

Which hardware supports it?

[...]
> --- a/lib/librte_ether/rte_ethdev.h
> +++ b/lib/librte_ether/rte_ethdev.h
> +#define DEV_TX_OFFLOAD_TXQ_MT_LOCKFREE	0x00004000
> +/**< Multiple threads can invoke rte_eth_tx_burst() concurrently on the
> same + * tx queue without SW lock.
> + */

Why TXQ in the name? DEV_TX_OFFLOAD_MT_LOCKFREE would be enough.
I wonder whether "lock free" wording is confusing because
the locks are probably handled in HW.
I think the good wording is "offloaded multi-thread capability",
maybe with a naming like DEV_TX_OFFLOAD_MT.

Anyway we should reference this flag in rte_eth_tx_burst()
and give more details in doc/guides/prog_guide/poll_mode_drv.rst.

Should we wait a first hardware PoC to add this flag?
Candidate for 17.08?

^ permalink raw reply	[flat|nested] 12+ messages in thread

* Re: [dpdk-dev] [PATCH] ethdev: add lock-less txq capability flag
  2017-04-24 12:34 ` Thomas Monjalon
@ 2017-04-27 11:00   ` Jerin Jacob
  2017-07-05 17:46     ` Thomas Monjalon
  0 siblings, 1 reply; 12+ messages in thread
From: Jerin Jacob @ 2017-04-27 11:00 UTC (permalink / raw)
  To: Thomas Monjalon
  Cc: dev, bruce.richardson, harry.van.haaren, hemant.agrawal,
	gage.eads, nipun.gupta, santosh.shukla

-----Original Message-----
> Date: Mon, 24 Apr 2017 14:34:59 +0200
> From: Thomas Monjalon <thomas@monjalon.net>
> To: Jerin Jacob <jerin.jacob@caviumnetworks.com>
> Cc: dev@dpdk.org, bruce.richardson@intel.com, harry.van.haaren@intel.com,
>  hemant.agrawal@nxp.com, gage.eads@intel.com, nipun.gupta@nxp.com,
>  santosh.shukla@caviumnetworks.com
> Subject: Re: [dpdk-dev]  [PATCH] ethdev: add lock-less txq capability flag
> 
> 21/04/2017 14:22, Jerin Jacob:
> > if this flag is advertised by a PMD, Multiple threads can
> > invoke rte_eth_tx_burst() concurrently on the same tx queue
> > without SW lock. This is an HW feature found in some NICs
> > and useful in the following use cases if HW supports it.
> 
> Which hardware supports it?

Cavium OCTEONTX Packet transmission HW block(PKO).

> 
> [...]
> > --- a/lib/librte_ether/rte_ethdev.h
> > +++ b/lib/librte_ether/rte_ethdev.h
> > +#define DEV_TX_OFFLOAD_TXQ_MT_LOCKFREE	0x00004000
> > +/**< Multiple threads can invoke rte_eth_tx_burst() concurrently on the
> > same + * tx queue without SW lock.
> > + */
> 
> Why TXQ in the name? DEV_TX_OFFLOAD_MT_LOCKFREE would be enough.

OK

> I wonder whether "lock free" wording is confusing because
> the locks are probably handled in HW.

Yes. Another reason why it possible because HW is not using ring like scheme(head and tail pointers)
for Tx. "lock free" wording is more from software perspective.

> I think the good wording is "offloaded multi-thread capability",
> maybe with a naming like DEV_TX_OFFLOAD_MT.

I think SW lock free is the capability here.IMO, it better to reflect the
capability in the name(DEV_TX_OFFLOAD_MT_LOCKFREE).

> 
> Anyway we should reference this flag in rte_eth_tx_burst()
> and give more details in doc/guides/prog_guide/poll_mode_drv.rst.

OK. Will address in v2.

> 
> Should we wait a first hardware PoC to add this flag?

If we are in agreement for method expose this feature through capability
flag then should we really need to wait for driver implementation to
accept the patch? Thoughts ?
This flag has impact on how we handle the even dev applications if HW support available.
We are in the process of up-streaming OCTEONTX ethdev driver, but there
are few thing like with external pool manager and eventdev integration
needs to be sorted out cleanly before that.

> Candidate for 17.08?

^ permalink raw reply	[flat|nested] 12+ messages in thread

* Re: [dpdk-dev] [PATCH] ethdev: add lock-less txq capability flag
  2017-04-27 11:00   ` Jerin Jacob
@ 2017-07-05 17:46     ` Thomas Monjalon
  2017-07-06  6:00       ` Jerin Jacob
  0 siblings, 1 reply; 12+ messages in thread
From: Thomas Monjalon @ 2017-07-05 17:46 UTC (permalink / raw)
  To: Jerin Jacob
  Cc: dev, bruce.richardson, harry.van.haaren, hemant.agrawal,
	gage.eads, nipun.gupta, santosh.shukla

27/04/2017 13:00, Jerin Jacob:
> -----Original Message-----
> > Date: Mon, 24 Apr 2017 14:34:59 +0200
> > From: Thomas Monjalon <thomas@monjalon.net>
> > To: Jerin Jacob <jerin.jacob@caviumnetworks.com>
> > Cc: dev@dpdk.org, bruce.richardson@intel.com, harry.van.haaren@intel.com,
> >  hemant.agrawal@nxp.com, gage.eads@intel.com, nipun.gupta@nxp.com,
> >  santosh.shukla@caviumnetworks.com
> > Subject: Re: [dpdk-dev]  [PATCH] ethdev: add lock-less txq capability flag
> > 
> > 21/04/2017 14:22, Jerin Jacob:
> > > if this flag is advertised by a PMD, Multiple threads can
> > > invoke rte_eth_tx_burst() concurrently on the same tx queue
> > > without SW lock. This is an HW feature found in some NICs
> > > and useful in the following use cases if HW supports it.
> > 
> > Which hardware supports it?
> 
> Cavium OCTEONTX Packet transmission HW block(PKO).
> 
> > 
> > [...]
> > > --- a/lib/librte_ether/rte_ethdev.h
> > > +++ b/lib/librte_ether/rte_ethdev.h
> > > +#define DEV_TX_OFFLOAD_TXQ_MT_LOCKFREE	0x00004000
> > > +/**< Multiple threads can invoke rte_eth_tx_burst() concurrently on the
> > > same + * tx queue without SW lock.
> > > + */
> > 
> > Why TXQ in the name? DEV_TX_OFFLOAD_MT_LOCKFREE would be enough.
> 
> OK
> 
> > I wonder whether "lock free" wording is confusing because
> > the locks are probably handled in HW.
> 
> Yes. Another reason why it possible because HW is not using ring like scheme(head and tail pointers)
> for Tx. "lock free" wording is more from software perspective.
> 
> > I think the good wording is "offloaded multi-thread capability",
> > maybe with a naming like DEV_TX_OFFLOAD_MT.
> 
> I think SW lock free is the capability here.IMO, it better to reflect the
> capability in the name(DEV_TX_OFFLOAD_MT_LOCKFREE).
> 
> > 
> > Anyway we should reference this flag in rte_eth_tx_burst()
> > and give more details in doc/guides/prog_guide/poll_mode_drv.rst.
> 
> OK. Will address in v2.
> 
> > 
> > Should we wait a first hardware PoC to add this flag?
> 
> If we are in agreement for method expose this feature through capability
> flag then should we really need to wait for driver implementation to
> accept the patch? Thoughts ?
> This flag has impact on how we handle the even dev applications if HW support available.
> We are in the process of up-streaming OCTEONTX ethdev driver, but there
> are few thing like with external pool manager and eventdev integration
> needs to be sorted out cleanly before that.
> 
> > Candidate for 17.08?

Any news of a v2?

^ permalink raw reply	[flat|nested] 12+ messages in thread

* Re: [dpdk-dev] [PATCH] ethdev: add lock-less txq capability flag
  2017-07-05 17:46     ` Thomas Monjalon
@ 2017-07-06  6:00       ` Jerin Jacob
  0 siblings, 0 replies; 12+ messages in thread
From: Jerin Jacob @ 2017-07-06  6:00 UTC (permalink / raw)
  To: Thomas Monjalon
  Cc: dev, bruce.richardson, harry.van.haaren, hemant.agrawal,
	gage.eads, nipun.gupta, santosh.shukla

-----Original Message-----
> Date: Wed, 05 Jul 2017 19:46:54 +0200
> From: Thomas Monjalon <thomas@monjalon.net>
> To: Jerin Jacob <jerin.jacob@caviumnetworks.com>
> Cc: dev@dpdk.org, bruce.richardson@intel.com, harry.van.haaren@intel.com,
>  hemant.agrawal@nxp.com, gage.eads@intel.com, nipun.gupta@nxp.com,
>  santosh.shukla@caviumnetworks.com
> Subject: Re: [dpdk-dev] [PATCH] ethdev: add lock-less txq capability flag
> 
> 27/04/2017 13:00, Jerin Jacob:
> > -----Original Message-----
> > > Which hardware supports it?
> > 
> > Cavium OCTEONTX Packet transmission HW block(PKO).
> > 
> > > 
> > > [...]
> > > > --- a/lib/librte_ether/rte_ethdev.h
> > > > +++ b/lib/librte_ether/rte_ethdev.h
> > > > +#define DEV_TX_OFFLOAD_TXQ_MT_LOCKFREE	0x00004000
> > > > +/**< Multiple threads can invoke rte_eth_tx_burst() concurrently on the
> > > > same + * tx queue without SW lock.
> > > > + */
> > > 
> > > Why TXQ in the name? DEV_TX_OFFLOAD_MT_LOCKFREE would be enough.
> > 
> > OK
> > 
> > > I wonder whether "lock free" wording is confusing because
> > > the locks are probably handled in HW.
> > 
> > Yes. Another reason why it possible because HW is not using ring like scheme(head and tail pointers)
> > for Tx. "lock free" wording is more from software perspective.
> > 
> > > I think the good wording is "offloaded multi-thread capability",
> > > maybe with a naming like DEV_TX_OFFLOAD_MT.
> > 
> > I think SW lock free is the capability here.IMO, it better to reflect the
> > capability in the name(DEV_TX_OFFLOAD_MT_LOCKFREE).
> > 
> > > 
> > > Anyway we should reference this flag in rte_eth_tx_burst()
> > > and give more details in doc/guides/prog_guide/poll_mode_drv.rst.
> > 
> > OK. Will address in v2.
> > 
> > > 
> > > Should we wait a first hardware PoC to add this flag?
> > 
> > If we are in agreement for method expose this feature through capability
> > flag then should we really need to wait for driver implementation to
> > accept the patch? Thoughts ?
> > This flag has impact on how we handle the even dev applications if HW support available.
> > We are in the process of up-streaming OCTEONTX ethdev driver, but there
> > are few thing like with external pool manager and eventdev integration
> > needs to be sorted out cleanly before that.
> > 
> > > Candidate for 17.08?
> 
> Any news of a v2?

I will send the v2.

> 

^ permalink raw reply	[flat|nested] 12+ messages in thread

* [dpdk-dev] [PATCH v2] ethdev: introduce lock-free txq capability flag
  2017-04-21 12:22 [dpdk-dev] [PATCH] ethdev: add lock-less txq capability flag Jerin Jacob
  2017-04-24 12:34 ` Thomas Monjalon
@ 2017-07-06  6:21 ` Jerin Jacob
  2017-07-08 16:08   ` Thomas Monjalon
  2017-07-10 16:59   ` [dpdk-dev] [PATCH v3 ] " Jerin Jacob
  1 sibling, 2 replies; 12+ messages in thread
From: Jerin Jacob @ 2017-07-06  6:21 UTC (permalink / raw)
  To: dev; +Cc: thomas, ferruh.yigit, john.mcnamara, Jerin Jacob

Introducing the DEV_TX_OFFLOAD_MT_LOCKFREE TX capability flag.
if a PMD advertises DEV_TX_OFFLOAD_MT_LOCKFREE capable, multiple threads
can invoke rte_eth_tx_burst() concurrently on the same tx queue without
SW lock. This PMD feature will be useful in the following use cases and
found in the OCTEON family of NPUs.

1) Remove explicit spinlock in some applications where lcores
to TX queues are not mapped 1:1.
example: OVS has such instance
https://github.com/openvswitch/ovs/blob/master/lib/netdev-dpdk.c#L299
https://github.com/openvswitch/ovs/blob/master/lib/netdev-dpdk.c#L1859
See the the usage of tx_lock spinlock.

2) In the eventdev use case, Avoid dedicating a separate TX core for
transmitting and thus enables more scaling as all workers can
send the packets.

v2:
- Changed the flag name to DEV_TX_OFFLOAD_MT_LOCKFREE(Thomas)
- Updated the documentation in doc/guides/prog_guide/poll_mode_drv.rst
and rte_eth_tx_burst() doxgen comments(Thomas)

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
---
 doc/guides/prog_guide/poll_mode_drv.rst | 14 ++++++++++++--
 lib/librte_ether/rte_ethdev.h           |  8 ++++++++
 2 files changed, 20 insertions(+), 2 deletions(-)

diff --git a/doc/guides/prog_guide/poll_mode_drv.rst b/doc/guides/prog_guide/poll_mode_drv.rst
index 4987f70a1..e6ac2978f 100644
--- a/doc/guides/prog_guide/poll_mode_drv.rst
+++ b/doc/guides/prog_guide/poll_mode_drv.rst
@@ -84,7 +84,7 @@ Whenever needed and appropriate, asynchronous communication should be introduced
 
 Avoiding lock contention is a key issue in a multi-core environment.
 To address this issue, PMDs are designed to work with per-core private resources as much as possible.
-For example, a PMD maintains a separate transmit queue per-core, per-port.
+For example, a PMD maintains a separate transmit queue per-core, per-port, if the PMD is not ``DEV_TX_OFFLOAD_MT_LOCKFREE`` capable.
 In the same way, every receive queue of a port is assigned to and polled by a single logical core (lcore).
 
 To comply with Non-Uniform Memory Access (NUMA), memory management is designed to assign to each logical core
@@ -146,6 +146,15 @@ This is also true for the pipe-line model provided all logical cores used are lo
 
 Multiple logical cores should never share receive or transmit queues for interfaces since this would require global locks and hinder performance.
 
+If the PMD is ``DEV_TX_OFFLOAD_MT_LOCKFREE`` capable, multiple threads can invoke ``rte_eth_tx_burst()``
+concurrently on the same tx queue without SW lock.This PMD feature found in some NICs and
+useful in the following use cases if PMD supports it. See `Hardware Offload`_ for details.
+
+*  Remove explicit spinlock in some applications where lcores to TX queues are not mapped 1:1.
+
+*  In the eventdev use case, Avoid dedicating a separate TX core for transmitting and thus
+   enables more scaling as all workers can send the packets.
+
 Device Identification and Configuration
 ---------------------------------------
 
@@ -290,7 +299,8 @@ Hardware Offload
 
 Depending on driver capabilities advertised by
 ``rte_eth_dev_info_get()``, the PMD may support hardware offloading
-feature like checksumming, TCP segmentation or VLAN insertion.
+feature like checksumming, TCP segmentation, VLAN insertion or
+lockfree multithreaded TX burst on the same TX queue.
 
 The support of these offload features implies the addition of dedicated
 status bit(s) and value field(s) into the rte_mbuf data structure, along
diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h
index d1076c821..8874ea504 100644
--- a/lib/librte_ether/rte_ethdev.h
+++ b/lib/librte_ether/rte_ethdev.h
@@ -901,6 +901,10 @@ struct rte_eth_conf {
 #define DEV_TX_OFFLOAD_IPIP_TNL_TSO     0x00000800    /**< Used for tunneling packet. */
 #define DEV_TX_OFFLOAD_GENEVE_TNL_TSO   0x00001000    /**< Used for tunneling packet. */
 #define DEV_TX_OFFLOAD_MACSEC_INSERT    0x00002000
+#define DEV_TX_OFFLOAD_MT_LOCKFREE      0x00004000
+/**< Multiple threads can invoke rte_eth_tx_burst() concurrently on the same
+ * tx queue without SW lock.
+ */
 
 struct rte_pci_device;
 
@@ -2933,6 +2937,10 @@ static inline int rte_eth_tx_descriptor_status(uint8_t port_id,
  * rte_eth_tx_burst() function must [attempt to] free the *rte_mbuf*  buffers
  * of those packets whose transmission was effectively completed.
  *
+ * If the PMD is DEV_TX_OFFLOAD_MT_LOCKFREE capable, multiple threads can
+ * invoke this function concurrently on the same tx queue without SW lock.
+ * @see rte_eth_dev_info_get, struct rte_eth_txconf::txq_flags
+ *
  * @param port_id
  *   The port identifier of the Ethernet device.
  * @param queue_id
-- 
2.13.2

^ permalink raw reply	[flat|nested] 12+ messages in thread

* Re: [dpdk-dev] [PATCH v2] ethdev: introduce lock-free txq capability flag
  2017-07-06  6:21 ` [dpdk-dev] [PATCH v2] ethdev: introduce lock-free " Jerin Jacob
@ 2017-07-08 16:08   ` Thomas Monjalon
  2017-07-10 16:56     ` Jerin Jacob
  2017-07-10 16:59   ` [dpdk-dev] [PATCH v3 ] " Jerin Jacob
  1 sibling, 1 reply; 12+ messages in thread
From: Thomas Monjalon @ 2017-07-08 16:08 UTC (permalink / raw)
  To: Jerin Jacob; +Cc: dev, ferruh.yigit, john.mcnamara

Hi Jerin,

Thanks for the update.
I think we can add this new flag in 17.08.
I prefer waiting John's review, especially for doc wording,
before applying it. I consider it does not hurt to add it post-rc1.

See below for my first comment on the doc.

06/07/2017 08:21, Jerin Jacob:
> Introducing the DEV_TX_OFFLOAD_MT_LOCKFREE TX capability flag.
> if a PMD advertises DEV_TX_OFFLOAD_MT_LOCKFREE capable, multiple threads
> can invoke rte_eth_tx_burst() concurrently on the same tx queue without
> SW lock. This PMD feature will be useful in the following use cases and
> found in the OCTEON family of NPUs.
> 
> 1) Remove explicit spinlock in some applications where lcores
> to TX queues are not mapped 1:1.
> example: OVS has such instance
> https://github.com/openvswitch/ovs/blob/master/lib/netdev-dpdk.c#L299
> https://github.com/openvswitch/ovs/blob/master/lib/netdev-dpdk.c#L1859
> See the the usage of tx_lock spinlock.
> 
> 2) In the eventdev use case, Avoid dedicating a separate TX core for
> transmitting and thus enables more scaling as all workers can
> send the packets.
> 
> v2:
> - Changed the flag name to DEV_TX_OFFLOAD_MT_LOCKFREE(Thomas)
> - Updated the documentation in doc/guides/prog_guide/poll_mode_drv.rst
> and rte_eth_tx_burst() doxgen comments(Thomas)
> 
> Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
> ---
[...]
> +If the PMD is ``DEV_TX_OFFLOAD_MT_LOCKFREE`` capable, multiple threads can invoke ``rte_eth_tx_burst()``
> +concurrently on the same tx queue without SW lock.This PMD feature found in some NICs and

A space is missing after the dot.
Note: my preference is to start next sentence on a new line (in RST source).

> +useful in the following use cases if PMD supports it. See `Hardware Offload`_ for details.

This sentence is confusing. I would remove "if PMD supports it".
After "following use cases", should we add a colon?
The relation with `Hardware Offload`_ is not obvious.

> +*  Remove explicit spinlock in some applications where lcores to TX queues are not mapped 1:1.

Can we reword "lcores to TX queues"?
I suggest "lcores are not mapped to Tx queues with 1:1 relation".

> +*  In the eventdev use case, Avoid dedicating a separate TX core for transmitting and thus

Uppercase in the middle of the sentence spotted.

> +   enables more scaling as all workers can send the packets.

^ permalink raw reply	[flat|nested] 12+ messages in thread

* Re: [dpdk-dev] [PATCH v2] ethdev: introduce lock-free txq capability flag
  2017-07-08 16:08   ` Thomas Monjalon
@ 2017-07-10 16:56     ` Jerin Jacob
  0 siblings, 0 replies; 12+ messages in thread
From: Jerin Jacob @ 2017-07-10 16:56 UTC (permalink / raw)
  To: Thomas Monjalon; +Cc: dev, ferruh.yigit, john.mcnamara

-----Original Message-----
> Date: Sat, 08 Jul 2017 18:08:57 +0200
> From: Thomas Monjalon <thomas@monjalon.net>
> To: Jerin Jacob <jerin.jacob@caviumnetworks.com>
> Cc: dev@dpdk.org, ferruh.yigit@intel.com, john.mcnamara@intel.com
> Subject: Re: [dpdk-dev] [PATCH v2] ethdev: introduce lock-free txq
>  capability flag
> 
> Hi Jerin,
> 

Hi Thomas. Thanks for the review.

> Thanks for the update.
> I think we can add this new flag in 17.08.
> I prefer waiting John's review, especially for doc wording,

Make sense. I will send the version.

> before applying it. I consider it does not hurt to add it post-rc1.
> 
> See below for my first comment on the doc.

Fixed in next version.

^ permalink raw reply	[flat|nested] 12+ messages in thread

* [dpdk-dev] [PATCH v3 ] ethdev: introduce lock-free txq capability flag
  2017-07-06  6:21 ` [dpdk-dev] [PATCH v2] ethdev: introduce lock-free " Jerin Jacob
  2017-07-08 16:08   ` Thomas Monjalon
@ 2017-07-10 16:59   ` Jerin Jacob
  2017-07-13 12:02     ` Hemant Agrawal
  2017-07-13 18:42     ` santosh
  1 sibling, 2 replies; 12+ messages in thread
From: Jerin Jacob @ 2017-07-10 16:59 UTC (permalink / raw)
  To: dev; +Cc: thomas, john.mcnamara, Jerin Jacob

Introducing the DEV_TX_OFFLOAD_MT_LOCKFREE TX capability flag.
if a PMD advertises DEV_TX_OFFLOAD_MT_LOCKFREE capable, multiple threads
can invoke rte_eth_tx_burst() concurrently on the same tx queue without
SW lock. This PMD feature will be useful in the following use cases and
found in the OCTEON family of NPUs.

1) Remove explicit spinlock in some applications where lcores
to TX queues are not mapped 1:1.
example: OVS has such instance
https://github.com/openvswitch/ovs/blob/master/lib/netdev-dpdk.c#L299
https://github.com/openvswitch/ovs/blob/master/lib/netdev-dpdk.c#L1859
See the the usage of tx_lock spinlock.

2) In the eventdev use case, avoid dedicating a separate TX core for
transmitting and thus enables more scaling as all workers can
send the packets.

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
---
v3:
- Addressed Thomas's documentation review comments
http://dpdk.org/ml/archives/dev/2017-July/070672.html

v2:
- Changed the flag name to DEV_TX_OFFLOAD_MT_LOCKFREE(Thomas)
- Updated the documentation in doc/guides/prog_guide/poll_mode_drv.rst
and rte_eth_tx_burst() doxgen comments(Thomas)
---
 doc/guides/prog_guide/poll_mode_drv.rst | 15 +++++++++++++--
 lib/librte_ether/rte_ethdev.h           |  8 ++++++++
 2 files changed, 21 insertions(+), 2 deletions(-)

diff --git a/doc/guides/prog_guide/poll_mode_drv.rst b/doc/guides/prog_guide/poll_mode_drv.rst
index 4987f70a1..1ac8f7ebf 100644
--- a/doc/guides/prog_guide/poll_mode_drv.rst
+++ b/doc/guides/prog_guide/poll_mode_drv.rst
@@ -84,7 +84,7 @@ Whenever needed and appropriate, asynchronous communication should be introduced
 
 Avoiding lock contention is a key issue in a multi-core environment.
 To address this issue, PMDs are designed to work with per-core private resources as much as possible.
-For example, a PMD maintains a separate transmit queue per-core, per-port.
+For example, a PMD maintains a separate transmit queue per-core, per-port, if the PMD is not ``DEV_TX_OFFLOAD_MT_LOCKFREE`` capable.
 In the same way, every receive queue of a port is assigned to and polled by a single logical core (lcore).
 
 To comply with Non-Uniform Memory Access (NUMA), memory management is designed to assign to each logical core
@@ -146,6 +146,16 @@ This is also true for the pipe-line model provided all logical cores used are lo
 
 Multiple logical cores should never share receive or transmit queues for interfaces since this would require global locks and hinder performance.
 
+If the PMD is ``DEV_TX_OFFLOAD_MT_LOCKFREE`` capable, multiple threads can invoke ``rte_eth_tx_burst()``
+concurrently on the same tx queue without SW lock. This PMD feature found in some NICs and useful in the following use cases:
+
+*  Remove explicit spinlock in some applications where lcores are not mapped to Tx queues with 1:1 relation.
+
+*  In the eventdev use case, avoid dedicating a separate TX core for transmitting and thus
+   enables more scaling as all workers can send the packets.
+
+See `Hardware Offload`_ for ``DEV_TX_OFFLOAD_MT_LOCKFREE`` capability probing details.
+
 Device Identification and Configuration
 ---------------------------------------
 
@@ -290,7 +300,8 @@ Hardware Offload
 
 Depending on driver capabilities advertised by
 ``rte_eth_dev_info_get()``, the PMD may support hardware offloading
-feature like checksumming, TCP segmentation or VLAN insertion.
+feature like checksumming, TCP segmentation, VLAN insertion or
+lockfree multithreaded TX burst on the same TX queue.
 
 The support of these offload features implies the addition of dedicated
 status bit(s) and value field(s) into the rte_mbuf data structure, along
diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h
index fd6baf37a..11fe13a07 100644
--- a/lib/librte_ether/rte_ethdev.h
+++ b/lib/librte_ether/rte_ethdev.h
@@ -927,6 +927,10 @@ struct rte_eth_conf {
 #define DEV_TX_OFFLOAD_IPIP_TNL_TSO     0x00000800    /**< Used for tunneling packet. */
 #define DEV_TX_OFFLOAD_GENEVE_TNL_TSO   0x00001000    /**< Used for tunneling packet. */
 #define DEV_TX_OFFLOAD_MACSEC_INSERT    0x00002000
+#define DEV_TX_OFFLOAD_MT_LOCKFREE      0x00004000
+/**< Multiple threads can invoke rte_eth_tx_burst() concurrently on the same
+ * tx queue without SW lock.
+ */
 
 struct rte_pci_device;
 
@@ -2961,6 +2965,10 @@ static inline int rte_eth_tx_descriptor_status(uint8_t port_id,
  * rte_eth_tx_burst() function must [attempt to] free the *rte_mbuf*  buffers
  * of those packets whose transmission was effectively completed.
  *
+ * If the PMD is DEV_TX_OFFLOAD_MT_LOCKFREE capable, multiple threads can
+ * invoke this function concurrently on the same tx queue without SW lock.
+ * @see rte_eth_dev_info_get, struct rte_eth_txconf::txq_flags
+ *
  * @param port_id
  *   The port identifier of the Ethernet device.
  * @param queue_id
-- 
2.13.2

^ permalink raw reply	[flat|nested] 12+ messages in thread

* Re: [dpdk-dev] [PATCH v3 ] ethdev: introduce lock-free txq capability flag
  2017-07-10 16:59   ` [dpdk-dev] [PATCH v3 ] " Jerin Jacob
@ 2017-07-13 12:02     ` Hemant Agrawal
  2017-07-18 13:43       ` Thomas Monjalon
  2017-07-13 18:42     ` santosh
  1 sibling, 1 reply; 12+ messages in thread
From: Hemant Agrawal @ 2017-07-13 12:02 UTC (permalink / raw)
  To: Jerin Jacob, dev; +Cc: thomas, john.mcnamara

On 7/10/2017 10:29 PM, Jerin Jacob wrote:
> Introducing the DEV_TX_OFFLOAD_MT_LOCKFREE TX capability flag.
> if a PMD advertises DEV_TX_OFFLOAD_MT_LOCKFREE capable, multiple threads
> can invoke rte_eth_tx_burst() concurrently on the same tx queue without
> SW lock. This PMD feature will be useful in the following use cases and
> found in the OCTEON family of NPUs.
>
> 1) Remove explicit spinlock in some applications where lcores
> to TX queues are not mapped 1:1.
> example: OVS has such instance
> https://github.com/openvswitch/ovs/blob/master/lib/netdev-dpdk.c#L299
> https://github.com/openvswitch/ovs/blob/master/lib/netdev-dpdk.c#L1859
> See the the usage of tx_lock spinlock.
>
> 2) In the eventdev use case, avoid dedicating a separate TX core for
> transmitting and thus enables more scaling as all workers can
> send the packets.
>
> Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
> ---
> v3:
> - Addressed Thomas's documentation review comments
> http://dpdk.org/ml/archives/dev/2017-July/070672.html
>
> v2:
> - Changed the flag name to DEV_TX_OFFLOAD_MT_LOCKFREE(Thomas)
> - Updated the documentation in doc/guides/prog_guide/poll_mode_drv.rst
> and rte_eth_tx_burst() doxgen comments(Thomas)
> ---
>  doc/guides/prog_guide/poll_mode_drv.rst | 15 +++++++++++++--
>  lib/librte_ether/rte_ethdev.h           |  8 ++++++++
>  2 files changed, 21 insertions(+), 2 deletions(-)
>
> diff --git a/doc/guides/prog_guide/poll_mode_drv.rst b/doc/guides/prog_guide/poll_mode_drv.rst
> index 4987f70a1..1ac8f7ebf 100644
> --- a/doc/guides/prog_guide/poll_mode_drv.rst
> +++ b/doc/guides/prog_guide/poll_mode_drv.rst
> @@ -84,7 +84,7 @@ Whenever needed and appropriate, asynchronous communication should be introduced
>
>  Avoiding lock contention is a key issue in a multi-core environment.
>  To address this issue, PMDs are designed to work with per-core private resources as much as possible.
> -For example, a PMD maintains a separate transmit queue per-core, per-port.
> +For example, a PMD maintains a separate transmit queue per-core, per-port, if the PMD is not ``DEV_TX_OFFLOAD_MT_LOCKFREE`` capable.
>  In the same way, every receive queue of a port is assigned to and polled by a single logical core (lcore).
>
>  To comply with Non-Uniform Memory Access (NUMA), memory management is designed to assign to each logical core
> @@ -146,6 +146,16 @@ This is also true for the pipe-line model provided all logical cores used are lo
>
>  Multiple logical cores should never share receive or transmit queues for interfaces since this would require global locks and hinder performance.
>
> +If the PMD is ``DEV_TX_OFFLOAD_MT_LOCKFREE`` capable, multiple threads can invoke ``rte_eth_tx_burst()``
> +concurrently on the same tx queue without SW lock. This PMD feature found in some NICs and useful in the following use cases:
> +
> +*  Remove explicit spinlock in some applications where lcores are not mapped to Tx queues with 1:1 relation.
> +
> +*  In the eventdev use case, avoid dedicating a separate TX core for transmitting and thus
> +   enables more scaling as all workers can send the packets.
> +
> +See `Hardware Offload`_ for ``DEV_TX_OFFLOAD_MT_LOCKFREE`` capability probing details.
> +
>  Device Identification and Configuration
>  ---------------------------------------
>
> @@ -290,7 +300,8 @@ Hardware Offload
>
>  Depending on driver capabilities advertised by
>  ``rte_eth_dev_info_get()``, the PMD may support hardware offloading
> -feature like checksumming, TCP segmentation or VLAN insertion.
> +feature like checksumming, TCP segmentation, VLAN insertion or
> +lockfree multithreaded TX burst on the same TX queue.
>
>  The support of these offload features implies the addition of dedicated
>  status bit(s) and value field(s) into the rte_mbuf data structure, along
> diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h
> index fd6baf37a..11fe13a07 100644
> --- a/lib/librte_ether/rte_ethdev.h
> +++ b/lib/librte_ether/rte_ethdev.h
> @@ -927,6 +927,10 @@ struct rte_eth_conf {
>  #define DEV_TX_OFFLOAD_IPIP_TNL_TSO     0x00000800    /**< Used for tunneling packet. */
>  #define DEV_TX_OFFLOAD_GENEVE_TNL_TSO   0x00001000    /**< Used for tunneling packet. */
>  #define DEV_TX_OFFLOAD_MACSEC_INSERT    0x00002000
> +#define DEV_TX_OFFLOAD_MT_LOCKFREE      0x00004000
> +/**< Multiple threads can invoke rte_eth_tx_burst() concurrently on the same
> + * tx queue without SW lock.
> + */
>
>  struct rte_pci_device;
>
> @@ -2961,6 +2965,10 @@ static inline int rte_eth_tx_descriptor_status(uint8_t port_id,
>   * rte_eth_tx_burst() function must [attempt to] free the *rte_mbuf*  buffers
>   * of those packets whose transmission was effectively completed.
>   *
> + * If the PMD is DEV_TX_OFFLOAD_MT_LOCKFREE capable, multiple threads can
> + * invoke this function concurrently on the same tx queue without SW lock.
> + * @see rte_eth_dev_info_get, struct rte_eth_txconf::txq_flags
> + *
>   * @param port_id
>   *   The port identifier of the Ethernet device.
>   * @param queue_id
>
You may also like to add this capability in "doc/guides/nics/features/*.ini"

Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>

^ permalink raw reply	[flat|nested] 12+ messages in thread

* Re: [dpdk-dev] [PATCH v3 ] ethdev: introduce lock-free txq capability flag
  2017-07-10 16:59   ` [dpdk-dev] [PATCH v3 ] " Jerin Jacob
  2017-07-13 12:02     ` Hemant Agrawal
@ 2017-07-13 18:42     ` santosh
  1 sibling, 0 replies; 12+ messages in thread
From: santosh @ 2017-07-13 18:42 UTC (permalink / raw)
  To: dev, Jerin Jacob

On Monday 10 July 2017 10:29 PM, Jerin Jacob wrote:

> Introducing the DEV_TX_OFFLOAD_MT_LOCKFREE TX capability flag.
> if a PMD advertises DEV_TX_OFFLOAD_MT_LOCKFREE capable, multiple threads
> can invoke rte_eth_tx_burst() concurrently on the same tx queue without
> SW lock. This PMD feature will be useful in the following use cases and
> found in the OCTEON family of NPUs.
>
> 1) Remove explicit spinlock in some applications where lcores
> to TX queues are not mapped 1:1.
> example: OVS has such instance
> https://github.com/openvswitch/ovs/blob/master/lib/netdev-dpdk.c#L299
> https://github.com/openvswitch/ovs/blob/master/lib/netdev-dpdk.c#L1859
> See the the usage of tx_lock spinlock.
>
> 2) In the eventdev use case, avoid dedicating a separate TX core for
> transmitting and thus enables more scaling as all workers can
> send the packets.
>
> Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
> ---
> v3:
> - Addressed Thomas's documentation review comments
> http://dpdk.org/ml/archives/dev/2017-July/070672.html
>
> v2:
> - Changed the flag name to DEV_TX_OFFLOAD_MT_LOCKFREE(Thomas)
> - Updated the documentation in doc/guides/prog_guide/poll_mode_drv.rst
> and rte_eth_tx_burst() doxgen comments(Thomas)
> ---
>  doc/guides/prog_guide/poll_mode_drv.rst | 15 +++++++++++++--
>  lib/librte_ether/rte_ethdev.h           |  8 ++++++++
>  2 files changed, 21 insertions(+), 2 deletions(-)
>
> diff --git a/doc/guides/prog_guide/poll_mode_drv.rst b/doc/guides/prog_guide/poll_mode_drv.rst
> index 4987f70a1..1ac8f7ebf 100644
> --- a/doc/guides/prog_guide/poll_mode_drv.rst
> +++ b/doc/guides/prog_guide/poll_mode_drv.rst
> @@ -84,7 +84,7 @@ Whenever needed and appropriate, asynchronous communication should be introduced
>  
>  Avoiding lock contention is a key issue in a multi-core environment.
>  To address this issue, PMDs are designed to work with per-core private resources as much as possible.
> -For example, a PMD maintains a separate transmit queue per-core, per-port.
> +For example, a PMD maintains a separate transmit queue per-core, per-port, if the PMD is not ``DEV_TX_OFFLOAD_MT_LOCKFREE`` capable.
>  In the same way, every receive queue of a port is assigned to and polled by a single logical core (lcore).
>  
>  To comply with Non-Uniform Memory Access (NUMA), memory management is designed to assign to each logical core
> @@ -146,6 +146,16 @@ This is also true for the pipe-line model provided all logical cores used are lo
>  
>  Multiple logical cores should never share receive or transmit queues for interfaces since this would require global locks and hinder performance.
>  
> +If the PMD is ``DEV_TX_OFFLOAD_MT_LOCKFREE`` capable, multiple threads can invoke ``rte_eth_tx_burst()``
> +concurrently on the same tx queue without SW lock. This PMD feature found in some NICs and useful in the following use cases:
> +
> +*  Remove explicit spinlock in some applications where lcores are not mapped to Tx queues with 1:1 relation.
> +
> +*  In the eventdev use case, avoid dedicating a separate TX core for transmitting and thus
> +   enables more scaling as all workers can send the packets.
> +
> +See `Hardware Offload`_ for ``DEV_TX_OFFLOAD_MT_LOCKFREE`` capability probing details.
> +
>  Device Identification and Configuration
>  ---------------------------------------
>  
> @@ -290,7 +300,8 @@ Hardware Offload
>  
>  Depending on driver capabilities advertised by
>  ``rte_eth_dev_info_get()``, the PMD may support hardware offloading
> -feature like checksumming, TCP segmentation or VLAN insertion.
> +feature like checksumming, TCP segmentation, VLAN insertion or
> +lockfree multithreaded TX burst on the same TX queue.
>  
>  The support of these offload features implies the addition of dedicated
>  status bit(s) and value field(s) into the rte_mbuf data structure, along
> diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h
> index fd6baf37a..11fe13a07 100644
> --- a/lib/librte_ether/rte_ethdev.h
> +++ b/lib/librte_ether/rte_ethdev.h
> @@ -927,6 +927,10 @@ struct rte_eth_conf {
>  #define DEV_TX_OFFLOAD_IPIP_TNL_TSO     0x00000800    /**< Used for tunneling packet. */
>  #define DEV_TX_OFFLOAD_GENEVE_TNL_TSO   0x00001000    /**< Used for tunneling packet. */
>  #define DEV_TX_OFFLOAD_MACSEC_INSERT    0x00002000
> +#define DEV_TX_OFFLOAD_MT_LOCKFREE      0x00004000
> +/**< Multiple threads can invoke rte_eth_tx_burst() concurrently on the same
> + * tx queue without SW lock.
> + */
>  
>  struct rte_pci_device;
>  
> @@ -2961,6 +2965,10 @@ static inline int rte_eth_tx_descriptor_status(uint8_t port_id,
>   * rte_eth_tx_burst() function must [attempt to] free the *rte_mbuf*  buffers
>   * of those packets whose transmission was effectively completed.
>   *
> + * If the PMD is DEV_TX_OFFLOAD_MT_LOCKFREE capable, multiple threads can
> + * invoke this function concurrently on the same tx queue without SW lock.
> + * @see rte_eth_dev_info_get, struct rte_eth_txconf::txq_flags
> + *
>   * @param port_id
>   *   The port identifier of the Ethernet device.
>   * @param queue_id

Acked-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>

^ permalink raw reply	[flat|nested] 12+ messages in thread

* Re: [dpdk-dev] [PATCH v3 ] ethdev: introduce lock-free txq capability flag
  2017-07-13 12:02     ` Hemant Agrawal
@ 2017-07-18 13:43       ` Thomas Monjalon
  0 siblings, 0 replies; 12+ messages in thread
From: Thomas Monjalon @ 2017-07-18 13:43 UTC (permalink / raw)
  To: Jerin Jacob; +Cc: dev, Hemant Agrawal, john.mcnamara

13/07/2017 15:02, Hemant Agrawal:
> On 7/10/2017 10:29 PM, Jerin Jacob wrote:
> > Introducing the DEV_TX_OFFLOAD_MT_LOCKFREE TX capability flag.
> > if a PMD advertises DEV_TX_OFFLOAD_MT_LOCKFREE capable, multiple threads
> > can invoke rte_eth_tx_burst() concurrently on the same tx queue without
> > SW lock. This PMD feature will be useful in the following use cases and
> > found in the OCTEON family of NPUs.
> >
> > 1) Remove explicit spinlock in some applications where lcores
> > to TX queues are not mapped 1:1.
> > example: OVS has such instance
> > https://github.com/openvswitch/ovs/blob/master/lib/netdev-dpdk.c#L299
> > https://github.com/openvswitch/ovs/blob/master/lib/netdev-dpdk.c#L1859
> > See the the usage of tx_lock spinlock.
> >
> > 2) In the eventdev use case, avoid dedicating a separate TX core for
> > transmitting and thus enables more scaling as all workers can
> > send the packets.
> >
> > Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
> 
> You may also like to add this capability in "doc/guides/nics/features/*.ini"

I've just added the feature "Lock-free Tx queue" in features/default.ini.

> Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>

Applied, thanks

^ permalink raw reply	[flat|nested] 12+ messages in thread

end of thread, other threads:[~2017-07-18 13:43 UTC | newest]

Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-04-21 12:22 [dpdk-dev] [PATCH] ethdev: add lock-less txq capability flag Jerin Jacob
2017-04-24 12:34 ` Thomas Monjalon
2017-04-27 11:00   ` Jerin Jacob
2017-07-05 17:46     ` Thomas Monjalon
2017-07-06  6:00       ` Jerin Jacob
2017-07-06  6:21 ` [dpdk-dev] [PATCH v2] ethdev: introduce lock-free " Jerin Jacob
2017-07-08 16:08   ` Thomas Monjalon
2017-07-10 16:56     ` Jerin Jacob
2017-07-10 16:59   ` [dpdk-dev] [PATCH v3 ] " Jerin Jacob
2017-07-13 12:02     ` Hemant Agrawal
2017-07-18 13:43       ` Thomas Monjalon
2017-07-13 18:42     ` santosh

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).