topic Re: IPCF cost too much cpuload in S32G

IPCF cost too much cpuload

jun13_chen — Tue, 25 Jul 2023 02:29:44 GMT

I have test IPCF transfer data from A core to M core, and M core send back the same data.

After finished all the transmission, I saw too much cpuload with ksoftirqd, the stacktrace is in function of ipc_queue_pop.

Following is top result and perf result:

top - 14:30:15 up 2 min, 1 user, load average: 0.95, 0.32, 0.11
Tasks: 107 total, 2 running, 105 sleeping, 0 stopped, 0 zombie
%Cpu(s): 0.2 us, 0.5 sy, 0.0 ni, 49.4 id, 0.3 wa, 0.0 hi, 49.6 si, 0.0 st
MiB Mem : 991.6 total, 888.7 free, 62.9 used, 40.0 buff/cache
MiB Swap: 0.0 total, 0.0 free, 0.0 used. 916.1 avail Mem

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
10 root 20 0 0 0 0 R 99.3 0.0 1:05.50 ksoftirqd+

Samples: 15K of event 'cycles', 4000 Hz, Event count (approx.): 3603202572 lost:
Overhead Shared Object Symbol
79.10% [ipc_shm_dev] [k] ipc_queue_pop
5.78% [ipc_shm_dev] [k] ipc_hw_irq_enable
2.02% [kernel] [k] tasklet_action_common.constprop.0
1.65% [ipc_shm_dev] [k] $x
1.57% [kernel] [k] __softirqentry_text_start
1.43% [kernel] [k] wake_up_var
1.10% [kernel] [k] __tasklet_schedule
0.63% [ipc_shm_dev] [k] 0x00000000000000d4
0.56% [kernel] [k] rcu_softirq_qs

I am sure the problem is caused by following patch in release of IPCF bsp37.0

commit 3df0b2e261b0c6be50e70bf6e54c1b3eb0e9dd6b
Author: Hoang Nguyen Le <hoang.nguyenle@nxp.com>
Date: Fri Mar 3 17:56:35 2023 +0700

Add do while statement to process
all BDs until no data left

Issue: IPCF-2948

Signed-off-by: Hoang Nguyen Le <hoang.nguyenle@nxp.com>

How to solve the problem?

Re: IPCF cost too much cpuload

Daniel-Aguirre — Tue, 25 Jul 2023 17:25:35 GMT

Hi,

Are you seeing this behavior only under BSP37.0? Or under other BSP's?

Also, which S32G series are you using? Are you using any NXP platform?

Please, let us know.

Re: IPCF cost too much cpuload

jun13_chen — Wed, 26 Jul 2023 01:24:10 GMT

yeah, only seeing under BSP37.0，we use s32g254a soc.

Re: IPCF cost too much cpuload

Daniel-Aguirre — Wed, 26 Jul 2023 17:49:53 GMT

Hi,

Thanks for your feedback. In that case, we will send this situation to the internal team for them to verify the behavior you are experiencing.

Since this may require a patch, this may take more than the expected time and could even be pushed to further BSP releases. We do apologize.

Please, let us know.

Re: IPCF cost too much cpuload

Daniel-Aguirre — Fri, 28 Jul 2023 21:03:13 GMT

Hi,

We have received the following update from the internal team:

"I checked the customer's comments, it seems customer didn't change the ipc-shm.bb file to get the correct IPCF version. Could you please asked customer to follow the IPCF's user manual to get the correct IPCF package.

And to reproduce this issue on my side, could you please help to clarify which test tool is used to test the CPU loading? And which version of IPCF is using on the M7 side?"

Please, let us know.