Re: [dpdk-dev] [PATCH 2/2] event/sw: use dynamically-sized IQs

[Pavan Nikhilesh <pbhagavatula@caviumnetworks.com>]

On Wed, Nov 29, 2017 at 09:08:34PM -0600, Gage Eads wrote:
> 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>
> ---


Sw eventdev crashes when used alongside Rx adapter. The crash happens when
pumping traffic at > 1.4mpps. This commit seems responsible for this.


Apply the following Rx adapter patch http://dpdk.org/dev/patchwork/patch/31977/
Command used:
./build/eventdev_pipeline_sw_pmd -c 0xfffff8 --vdev="event_sw" -- -r0x800
-t0x100 -w F000 -e 0x10

Backtrace:

Thread 4 "lcore-slave-4" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0xffffb6c8f040 (LWP 25291)]
0x0000aaaaaadcc0d4 in iq_dequeue_burst (count=48, ev=0xffffb6c8dd38,
iq=0xffff9f764720, sw=0xffff9f332600) at
/root/clean/rebase/dpdk-next-eventdev/drivers/event/sw/iq_chunk.h:142
142 ev[total++] = current->events[index++];


(gdb) bt
#0  0x0000aaaaaadcc0d4 in iq_dequeue_burst (count=48, ev=0xffffb6c8dd38,
iq=0xffff9f764720, sw=0xffff9f332600) at
/root/clean/rebase/dpdk-next-eventdev/drivers/event/sw/iq_chunk.h:142
#1  sw_schedule_atomic_to_cq (sw=0xffff9f332600, qid=0xffff9f764700, iq_num=0,
count=48) at
/root/clean/rebase/dpdk-next-eventdev/drivers/event/sw/sw_evdev_scheduler.c:74
#2  0x0000aaaaaadcdc44 in sw_schedule_qid_to_cq (sw=0xffff9f332600) at
/root/clean/rebase/dpdk-next-eventdev/drivers/event/sw/sw_evdev_scheduler.c:262
#3  0x0000aaaaaadd069c in sw_event_schedule (dev=0xaaaaaafbd200
<rte_event_devices>) at
/root/clean/rebase/dpdk-next-eventdev/drivers/event/sw/sw_evdev_scheduler.c:564
#4  0x0000aaaaaadca008 in sw_sched_service_func (args=0xaaaaaafbd200
<rte_event_devices>) at
/root/clean/rebase/dpdk-next-eventdev/drivers/event/sw/sw_evdev.c:767
#5  0x0000aaaaaab54740 in rte_service_runner_do_callback (s=0xffff9fffdf80,
cs=0xffff9ffef900, service_idx=0) at
/root/clean/rebase/dpdk-next-eventdev/lib/librte_eal/common/rte_service.c:349
#6  0x0000aaaaaab54868 in service_run (i=0, cs=0xffff9ffef900,
service_mask=18446744073709551615) at
/root/clean/rebase/dpdk-next-eventdev/lib/librte_eal/common/rte_service.c:376
#7  0x0000aaaaaab54954 in rte_service_run_iter_on_app_lcore (id=0,
serialize_mt_unsafe=1) at
/root/clean/rebase/dpdk-next-eventdev/lib/librte_eal/common/rte_service.c:405
#8  0x0000aaaaaaaef04c in schedule_devices (lcore_id=4) at
/root/clean/rebase/dpdk-next-eventdev/examples/eventdev_pipeline_sw_pmd/main.c:223
#9  0x0000aaaaaaaef234 in worker (arg=0xffff9f331d80) at
/root/clean/rebase/dpdk-next-eventdev/examples/eventdev_pipeline_sw_pmd/main.c:274
#10 0x0000aaaaaab4382c in eal_thread_loop (arg=0x0) at
/root/clean/rebase/dpdk-next-eventdev/lib/librte_eal/linuxapp/eal/eal_thread.c:182
#11 0x0000ffffb7e46d64 in start_thread () from /usr/lib/libpthread.so.0
#12 0x0000ffffb7da8bbc in thread_start () from /usr/lib/libc.so.6

Segfault seems to happen in sw_event_schedule and only happens under high
traffic load.

Thanks,
Pavan

>  drivers/event/sw/iq_chunk.h           | 212 ++++++++++++++++++++++++++++++++++
>  drivers/event/sw/iq_ring.h            | 172 ---------------------------
>  drivers/event/sw/sw_evdev.c           |  59 ++++++----
>  drivers/event/sw/sw_evdev.h           |  16 ++-
>  drivers/event/sw/sw_evdev_scheduler.c |  40 +++----
>  drivers/event/sw/sw_evdev_xstats.c    |  12 +-
>  test/test/test_eventdev_sw.c          |  15 +--
>  7 files changed, 288 insertions(+), 238 deletions(-)
>  create mode 100644 drivers/event/sw/iq_chunk.h
>  delete mode 100644 drivers/event/sw/iq_ring.h
>

<snip>

> @@ -1743,7 +1739,6 @@ xstats_id_reset_tests(struct test *t)
>  		0, /* tx */
>  		0, /* drop */
>  		7, /* inflight */
> -		512, /* iq size */
>  		0, /* iq 0 used */
>  		0, /* iq 1 used */
>  		0, /* iq 2 used */
> --
> 2.7.4
>