vpp buffer issue on Lx2160

cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

vpp buffer issue on Lx2160

10,305 Views
Blue_Sky1
Contributor II

Hey, Guys,

I have started VPP on my lx2160 board, and iperf tests were done.Some problems were found during testing:

1、the value of Used, it doesn't look right.

vpp# show buffers
Pool Name Index NUMA Size Data Size Total     Avail Cached Used
default-numa-0 0 0     2496 2048        100000 199488 512 42948672

2、I just tested 5m, and allocated quickly reached its maximum value, making it impossible to receive any more packets

iperf -u -c 10.8.19.178 -p 2152 -i 1 -b 5m -t 99999

vpp# show dpdk buffer
name="vpp pool 0" available = 0 allocated = 100000 total = 100000

stupid_0-1720505651128.png

3、The vpp_main thread is prone to crash when testing larger amounts of traffic,check the core with gdb and it's not the same every time.

(gdb) bt
#0 __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:50
#1 0x0000ffff8c312aac in __GI_abort () at abort.c:79
#2 0x0000aaaaacf387e4 in os_exit (code=code@entry=1)
at /root/test/flexbuild_lsdk2108_github/components/apps/networking/vpp/src/vpp/vnet/main.c:433
#3 0x0000ffff8c643f60 in unix_signal_handler (signum=11, si=<optimized out>, uc=<optimized out>)
at /root/test/flexbuild_lsdk2108_github/components/apps/networking/vpp/src/vlib/unix/main.c:187
#4 <signal handler called>
#5 0x0000ffff8c5f8b04 in dispatch_suspended_process (last_time_stamp=49550713257,
process_index=<optimized out>, vm=<optimized out>)
at /root/test/flexbuild_lsdk2108_github/components/apps/networking/vpp/src/vlib/main.c:1644
#6 vlib_main_or_worker_loop (is_main=1, vm=0xffff8c66fe40 <vlib_global_main>)
at /root/test/flexbuild_lsdk2108_github/components/apps/networking/vpp/src/vlib/main.c:1989
#7 vlib_main_loop (vm=0xffff8c66fe40 <vlib_global_main>)
at /root/test/flexbuild_lsdk2108_github/components/apps/networking/vpp/src/vlib/main.c:2032
#8 vlib_main (vm=<optimized out>, vm@entry=0xffff8c66fe40 <vlib_global_main>,
input=input@entry=0xffff4a631fc8)
at /root/test/flexbuild_lsdk2108_github/components/apps/networking/vpp/src/vlib/main.c:2278
#9 0x0000ffff8c642d8c in thread0 (arg=281473037303360)
at /root/test/flexbuild_lsdk2108_github/components/apps/networking/vpp/src/vlib/unix/main.c:670
#10 0x0000ffff8c487f7c in clib_calljmp () from /usr/lib/aarch64-linux-gnu/libvppinfra.so.21.08.0
Backtrace stopped: not enough registers or memory available to unwind further

 

The startup.conf is shown in the attachment.

Test Env:LSDK21.08

Any replies will be helpful.

Thanks

Tags (2)
0 Kudos
Reply
4 Replies

10,241 Views
yipingwang
NXP TechSupport
NXP TechSupport

Please refer to "9.3.5 Executing VPP" in LSDK 21.08 user manual.

Please execute the following command before executing VPP.

cd /usr/local/dpdk/dpaa2
. ./dynamic_dpl.sh dpmac.1 dpmac.2
mkdir /mnt/hugepages
mount -t hugetlbfs none /mnt/hugepages
echo 256 > /proc/sys/vm/nr_hugepages

0 Kudos
Reply

10,234 Views
Blue_Sky1
Contributor II

Hi,yiping:

These steps were performed before running vpp.Run Script Reference Attachment,

Before:

stupid_0-1721209546314.png

After:

stupid_1-1721211312350.png

And the  hugepage has been set.

stupid_4-1721212104205.png

Problems remain.

Thanks

 

595 Views
dunk
Contributor II

Hi,

I faced the same issue when simply injecting the *slow protocol* Ethernet packet to the VPP.

You can watch the buffer "show dpdk buffer" to see the buffer is not *freed* while dealing with every *slow protocol* packet coming throught the VPP ports.

When the buffer is all allocated, VPP will then crash.

0 Kudos
Reply

585 Views
Blue_Sky1
Contributor II
Hi,
Yes, I used show dpdk buffer later, but ran into another problem.When I use iperf for udp traffic testing, even if the test traffic is small, the dpdk buffer runs out very quickly (having modified the buffer's number to be larger) and no longer receives data.

Thanks
0 Kudos
Reply