> -----Original Message----- > From: Jerin Jacob [mailto:jerin.ja...@caviumnetworks.com] > Sent: Tuesday, February 7, 2017 6:58 AM > To: Van Haaren, Harry <harry.van.haa...@intel.com> > Cc: dev@dpdk.org; Richardson, Bruce <bruce.richard...@intel.com> > Subject: Re: [PATCH v2 07/15] event/sw: add support for event queues > > On Mon, Feb 06, 2017 at 10:25:18AM +0000, Van Haaren, Harry wrote: > > > -----Original Message----- > > > From: Jerin Jacob [mailto:jerin.ja...@caviumnetworks.com] > > > Sent: Monday, February 6, 2017 9:25 AM > > > To: Van Haaren, Harry <harry.van.haa...@intel.com> > > > Cc: dev@dpdk.org; Richardson, Bruce <bruce.richard...@intel.com> > > > Subject: Re: [PATCH v2 07/15] event/sw: add support for event queues > > > > > > On Tue, Jan 31, 2017 at 04:14:25PM +0000, Harry van Haaren wrote: > > > > From: Bruce Richardson <bruce.richard...@intel.com> > > > > > > > > Add in the data structures for the event queues, and the eventdev > > > > functions to create and destroy those queues. > > > > > > > > Signed-off-by: Bruce Richardson <bruce.richard...@intel.com> > > > > Signed-off-by: Harry van Haaren <harry.van.haa...@intel.com> > > > > --- > > > > drivers/event/sw/iq_ring.h | 176 > > > > ++++++++++++++++++++++++++++++++++++++++++++ > > > > drivers/event/sw/sw_evdev.c | 158 > > > > +++++++++++++++++++++++++++++++++++++++ > > > > drivers/event/sw/sw_evdev.h | 75 +++++++++++++++++++ > > > > 3 files changed, 409 insertions(+) > > > > create mode 100644 drivers/event/sw/iq_ring.h > > > > > > > > + */ > > > > + > > > > +/* > > > > + * Ring structure definitions used for the internal ring buffers of the > > > > + * SW eventdev implementation. These are designed for single-core use > > > > only. > > > > + */ > > > > > > If I understand it correctly, IQ and QE rings are single producer and > > > single consumer rings. By the specification, multiple producers through > > > multiple ports can enqueue to the event queues at a time.Does SW > > > implementation > > > support that? or am I missing something here? > > > > You're right that the IQ and QE rings are Single Producer, Single Consumer > > rings. More > specifically, the QE is a ring for sending rte_event structs between cores, > while the IQ ring > is optimized for internal use in the scheduler core - and should not be used > to send events > between cores. Note that the design of the SW scheduler includes a central > core for performing > scheduling. > > Thanks Harry. One question though, > In RTE_EVENT_DEV_CAP_DISTRIBUTED_SCHED mode, multiple SW schedulers can > be active. Right? If so, We need multi consumer. Right?
Note that the sw scheduler is a centralized scheduler, and does not support distributing the *scheduling* work itself. In the case of having multiple software schedulers, each instance is its own scheduling domain - they don't interact with each other directly. There is no need for Multi-Producer / Multi-Consumer with this design as there is never more than 1 thread accessing the producer/consumer side of a ring. > > > > > In other works, the QE rings transfer events from the worker core to the > > scheduler - and the > scheduler pushes the events into what you call the "event queues" (aka, the > atomic/ordered > queue itself). These "event queues" are IQ instances. On egress from the > scheduler, the event > passes through a QE ring to the worker. > > > > The result is that despite that only SP/SC rings are used, multiple workers > > can enqueue to > any event queue. > > Got it. >
