Today it is difficult to know if the SW Eventdev PMD is making
forward progress when it runs an iteration of its service. This
commit adds two xstats to give better visibility to the application.
The new xstats provide an application with which Eventdev ports
received work in the last iteration of scheduling, as well if
forward progress was made by the scheduler.
This patch implements an xstat for the SW PMD that exposes a
bitmask of ports that were scheduled to. In the unlikely case
that the SW PMD instance has 64 or more ports, return UINT64_MAX.
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
The sw PMD implements xstats reset by having the xstat get operations
return a value to the statistic's value at the last reset. The value at the
last reset is maintained in the per-xstat reset_value field, but the PMD
was setting reset_value = current - reset_value instead of reset_value =
current.
Fixes: c1ad03df7a ("event/sw: support xstats")
Cc: stable@dpdk.org
Signed-off-by: Gage Eads <gage.eads@intel.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
This commit introduces dynamically-sized IQs, by switching the underlying
data structure from a fixed-size ring to a linked list of queue 'chunks.'
This has a number of benefits:
- Certain corner cases were observed in which all of a pipeline's flows
could be pinned to one port for extended periods, effectively turning a
multi-core pipeline into single-core one. This was caused by an event
producer having a larger new_event_threshold than the IQ depth, and
injecting large numbers of packets that are ultimately backpressured in a
worker's rx_ring, causing those packets' flows to be scheduled to that
port.
The dynamically sized IQ does not have this problem because each IQ can
grow large enough to store all the system's events, such that
backpressure will not reach the worker_ring.
- Slight performance improvement (~1-2%) in high throughput scenarios,
tested with eventdev_pipeline_sw_pmd.
This implementation has a small increase in the queue storage memory
footprint (~70KB). This commit also removes the iq_size xstat, which no
longer applies to this implementation.
Signed-off-by: Gage Eads <gage.eads@intel.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
Replace the BSD license header with the SPDX tag for files
with only an Intel copyright on them.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
This commit allows the xstats_get() API to return just a single
value based on its ID. Previously, the "ret_n_lt_stats" value
set to 1 ensured a check to take place that the array was larger
than the available xstats.
The xstats_get() API allows retriving of individual stats - hence
this check should be removed - by setting "ret_n_lt_stats" to zero.
Fixes: c1ad03df7a ("event/sw: support xstats")
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Now that we have a standard event ring implementation for passing events
core-to-core, use that in place of the custom event rings in the software
eventdev.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
This commit adds a new statistic to the SW eventdev PMD.
The statistic shows how many packets were sent from a
queue to a port. This provides information on how traffic
from a specific queue is being load-balanced to worker cores.
Note that these numbers should be compared across all queue
stages - the load-balancing does not try to perfectly share
each queue's traffic, rather it balances the overall traffic
from all queues to the ports.
The statistic is printed from the rte_eventdev_dump() function,
as well as being made available via the xstats API.
Unit tests have been updated to expect more per-queue statistics,
and the correctness of counts and counts after reset is verified.
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Add support for xstats to report out on the state of the eventdev.
Useful for debugging and for unit tests, as well as observability
at runtime and performance tuning of apps to work well with the
scheduler.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: David Hunt <david.hunt@intel.com>
