Process exit with orphan rings when spawning multiple processes

[tokers]

Hello!

I am using the QAT driver (qat1.7.l.4.2.0-00012), and I am using QAT_Engine through OpenSSL, ultimately called by Nginx.

I occurred the problem " Process exit with orphan rings when spawning multiple processes", the following messages was showed by dmesg.

Jul 13 18:12:58 UPYUN_SSD kernel: dh895xcc 0000:03:00.0: Process 26515 nginx exit with orphan rings
Jul 13 18:12:58 UPYUN_SSD kernel: dh895xcc 0000:03:00.0: Process 26517 nginx exit with orphan rings
Jul 13 18:12:59 UPYUN_SSD kernel: dh895xcc 0000:03:00.0: Process 26518 nginx exit with orphan rings
Jul 13 18:12:59 UPYUN_SSD kernel: dh895xcc 0000:03:00.0: Process 26521 nginx exit with orphan rings
Jul 13 18:12:59 UPYUN_SSD kernel: dh895xcc 0000:03:00.0: Process 26520 nginx exit with orphan rings
Jul 13 18:12:59 UPYUN_SSD kernel: dh895xcc 0000:03:00.0: Process 26516 nginx exit with orphan rings
Jul 13 18:12:59 UPYUN_SSD kernel: dh895xcc 0000:03:00.0: Process 26514 nginx exit with orphan rings
Jul 13 18:12:59 UPYUN_SSD kernel: dh895xcc 0000:03:00.0: Process 26519 nginx exit with orphan rings
Jul 13 18:13:18 UPYUN_SSD kernel: dh895xcc 0000:03:00.0: Process 26568 nginx exit with orphan rings

The Nginx is configured to 8 workers, there are also some messages in error.log:

[error] SalCtrl_GetEnabledServices() - : Failed to get enabled services from ADF
[error] SalCtrl_ServiceEventHandler() - : Failed to get enabled services

I found this issue was recorded in this document: https://01.org/sites/default/files/downloads/intelr-quickassist-technology/336211qatrelnotespublicrev001.pdf . The description is:

 If multiple processes start a user space service access layer (icp_sal_userStart) and they
all exit together, the syslog may show a message "Process <PID> <NAME> exit with
orphan rings"

In order to prevent all the workers exit together, I tried to postpone the each worker's exit() calling by making each worker sleeps for a random time, but it seems useless.

Since the document is released in July 2017, Is there any new scheme that we can fix this issue?

[tokers]

It seems that the issue is caused by worker's permission. After changing the nginx worker process user to root, this problem was disappear.