Re: [dpdk-users] DPDK 18.11 ASENI_MB segfault

["Zhang, Roy Fan" <roy.fan.zhang@intel.com>]

Hi,

Can you share more information?
Which process does the enqueue and dequeue operation?
A segment of code or pseudo code would be nice.

Regards,
Fan

> -----Original Message-----
> From: users [mailto:users-bounces@dpdk.org] On Behalf Of Pathak, Pravin
> Sent: Thursday, March 7, 2019 10:31 PM
> To: users@dpdk.org
> Subject: [dpdk-users] DPDK 18.11 ASENI_MB segfault
> 
> Hi -
> I have multi-process application in which primary process initializes all crypto
> devices and sets up sessions. Secondary process uses the pre-initialized
> devices and sessions.
> It was working fine till DPDK 18.05. When switched to DPDK 18.11,  secondary
> process started giving seg fault on receive. Primary process can still send and
> receive packets to/from crypto devices.
> When switched to use HW crypto (QAT) instead of software crypto, same
> application works fine on DPDK 18.11.  Is the usage model correct? Anything
> changed in DPDK 18.11 which might explain this?
> 
> Regards
> Pravin
> 
> GDB backtrace
> 
> Program terminated with signal SIGSEGV, Segmentation fault.
> #0  post_process_mb_job (qp=0x7007df2d600, job=0x55b48daa6b00)
>     at /.../dpdk-18.11/drivers/crypto/aesni_mb/rte_aesni_mb_pmd.c:931
> ---Type <return> to continue, or q <return> to quit---
> 931             if (likely(op->status ==
> RTE_CRYPTO_OP_STATUS_NOT_PROCESSED)) {
> [Current thread is 1 (Thread 0x7f7e8b2f4c00 (LWP 39715))]
> (gdb)
> (gdb) p op
> $1 = (struct rte_crypto_op *) 0x0
> (gdb) p sess
> $2 = <optimized out>
> (gdb)