Bug 88 - DPDK service core getting crashed with eventdev scheduler
Summary: DPDK service core getting crashed with eventdev scheduler
Status: CONFIRMED
Alias: None
Product: DPDK
Classification: Unclassified
Component: eventdev (show other bugs)
Version: 18.05
Hardware: x86 Linux
: Normal critical
Target Milestone: ---
Assignee: Harry van Haaren (Intel)
URL:
Depends on:
Blocks:
 
Reported: 2018-08-31 05:55 CEST by Guruprasad T S
Modified: 2018-09-21 11:53 CEST (History)
2 users (show)



Attachments

Description Guruprasad T S 2018-08-31 05:55:07 CEST
While testing performance with eventdev, I am seeing a crash in dpdk service core. Test scenario is given below
1) DPDK 1805
2) Eventdev 2 ordered Queues(Q0 and Q1) & 2 Ports
2) 2 Workers threads + 1 Service core for running scheduling.
3) 1st worker is getting packet from NIC and enqueueing to Q0(packet processing is done by both the workers)
4) Packet rate of 5.6Mpps (Pkt size 64 bytes)
 

Thread 7 "lcore-slave-6" received signal SIGSEGV, Segmentation fault.

[Switching to Thread 0x7fdde35fd700 (LWP 9067)]

0x00007fffb3ce1101 in iq_alloc_chunk (sw=0x7ff9e80c3500)

at dpdk/dpdk-18.05/drivers/event/sw/iq_chunk.h:35

35 sw->chunk_list_head = chunk->next;

(gdb) bt

#0 0x00007fffb3ce1101 in iq_alloc_chunk (sw=0x7ff9e80c3500)

at dpdk/dpdk-18.05/drivers/event/sw/iq_chunk.h:35

#1 iq_enqueue (ev=0x7ff9e80d3800, iq=0x7ff9e85119a0, sw=0x7ff9e80c3500)

at dpdk/dpdk-18.05/drivers/event/sw/iq_chunk.h:79

#2 __pull_port_lb (allow_reorder=1, port_id=0, sw=0x7ff9e80c3500)

at dpdk/dpdk-18.05/drivers/event/sw/sw_evdev_scheduler.c:435

#3 sw_schedule_pull_port_lb (sw=0x7ff9e80c3500, port_id=0)

at dpdk/dpdk-18.05/drivers/event/sw/sw_evdev_scheduler.c:452

#4 0x00007fffb3ce2824 in sw_event_schedule (dev=0x7fffb490b6c0 <rte_event_devices>)

at dpdk/dpdk-18.05/drivers/event/sw/sw_evdev_scheduler.c:524

#5 0x00007fffb3cdbb1a in sw_sched_service_func (args=0x7fffb490b6c0 <rte_event_devices>)

at dpdk/dpdk-18.05/drivers/event/sw/sw_evdev.c:795

#6 0x00007fffb35e6326 in rte_service_runner_do_callback (s=0x7ff9e93fe000, cs=0x7ff9e93dad40, service_idx=0)

at dpdk/dpdk-18.05/lib/librte_eal/common/rte_service.c:345

#7 0x00007fffb35e643f in service_run (i=0, cs=0x7ff9e93dad40, service_mask=1)

at dpdk/dpdk-18.05/lib/librte_eal/common/rte_service.c:372

#8 0x00007fffb35e6599 in rte_service_runner_func (arg=0x0)

at dpdk/dpdk-18.05/lib/librte_eal/common/rte_service.c:422

#9 0x00007fffb35b872a in eal_thread_loop (arg=0x0)

at dpdk/dpdk-18.05/lib/librte_eal/linuxapp/eal/eal_thread.c:153

#10 0x00007ffff614b6ba in start_thread (arg=0x7fdde35fd700) at pthread_create.c:333

#11 0x00007ffff5c7d41d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109

 
Is this a known issue at high load or am I doing something wrong?
Comment 1 Jerin 2018-08-31 14:08:58 CEST
The bug is specific to driver/event/sw.
Comment 2 Harry van Haaren (Intel) 2018-09-12 18:30:14 CEST
Hi,

Would you provide the exact command that you're running when starting the application? Are you using the eventdev-pipeline application? Currently I have no method to reproduce this.

There is not expected to be any crash possible in DPDK code - unless there is memory corruption from the application. That said, the IQ chunks is non-trivial, and I'd be happy to investigate if you can provide a way for me to reproduce the issue.

Thanks, -Harry
Comment 3 Harry van Haaren (Intel) 2018-09-21 11:53:25 CEST
Hi,

Ping on this bug - is there a way I can reproduce this?

Thanks, -Harry

Note You need to log in before you can comment on or make changes to this bug.