[dpdk-dev] [PATCH] eventdev: Add rte_errno return values to the enqueue and dequeue functions

[Eads, Gage]

>  -----Original Message-----
>  From: Richardson, Bruce
>  Sent: Monday, February 13, 2017 6:08 AM
>  To: Jerin Jacob <jerin.jacob at caviumnetworks.com>
>  Cc: Eads, Gage <gage.eads at intel.com>; dev at dpdk.org;
>  hemant.agrawal at nxp.com; Van Haaren, Harry <harry.van.haaren at intel.com>;
>  nipun.gupta at nxp.com
>  Subject: Re: [PATCH] eventdev: Add rte_errno return values to the enqueue and
>  dequeue functions
>  
>  On Mon, Feb 13, 2017 at 05:18:11PM +0530, Jerin Jacob wrote:
>  > On Mon, Feb 13, 2017 at 10:38:55AM +0000, Bruce Richardson wrote:
>  > > On Fri, Feb 10, 2017 at 03:02:21PM -0600, Gage Eads wrote:
>  > > > This change allows user software to differentiate between an
>  > > > invalid argument (such as an invalid queue_id or sched_type in an
>  > > > enqueued event) and backpressure from the event device.
>  > > >
>  > > > The port and device ID checks are placed in
>  > > > RTE_LIBRTE_EVENTDEV_DEBUG header guards to avoid the performance
>  hit in non-debug execution.
>  > > >
>  > > > Signed-off-by: Gage Eads <gage.eads at intel.com>
>  > > > ---
>  > >
>  > > Do we have some idea of the performance hit from these? It may be
>  > > too soon to know, given we don't have many drivers to test with, but
>  > > if there is no perf hit seen with the SW driver, I think we should
>  > > look to just always do this, rather than having it compile-time off.
>  > > If it does
>  >
>  > IMO, It is better put to under compile-time like ethdev. It is
>  > difficult predict the performance regression on wide range of cores
>  > that DPDK runs now. I think we need to add following additional checks
>  > based on Gage header file change
>  >
>  > 1) Per event queue ID is valid or not?
>  > 2) Per event's sched type doesn't match the capabilities of the destination
>  queue.
>  >
>  
>  Ok, if we are expanding the number of checks then I definitely think it needs to
>  be compile-time selected.

My thinking is that, unlike checking the dev ID or port ID, per-event checks should be in the PMD itself since (presumably) it will loop over the events anyway. I also think it makes sense for the per-event error checking not to be compile-time selectable because a PMD needs to handle the invalid queue ID or sched_type case anyway.

>  
>  /Bruce
>  >
>  > > prove to be a performance problem we can look to #ifdef it out later.
>  > >
>  > > /Bruce