Segmentation Fault in IPsec_MB PMD When Resetting Crypto Device Stats After Upgrade to DPDK 24.11.3

[Shoaib Choudhary <shoaibsic@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 1750 bytes --]

*Description:*
After upgrading from DPDK 24.11 to 24.11.3, we are encountering a
segmentation fault in the IPsec_MB PMD (librte_crypto_ipsec_mb.so.25) when
attempting to reset crypto device statistics
using rte_cryptodev_stats_reset(). This issue was not observed in DPDK
24.11.

*Reproduction Context:*
The crash occurs during automated traffic tests that exercise GTP loopback
scenarios on a DPDK-based system. The test suite covers both fragmented and
non-fragmented packets, and uses multiple crypto algorithms (AES, SNOW3G,
ZUC) with Out-of-Place (OOP) processing enabled and disabled. Traffic is
sent through the datapath, and statistics are collected and reset as part
of the test flow. The segmentation fault is consistently triggered when
resetting crypto device stats after such traffic tests.

*Crash Details:*

   - Faulting function: rte_cryptodev_stats_reset() (called from
   application code)
   - Faulting library: librte_crypto_ipsec_mb.so.25
   - Fault address: 0x58
   - Only occurs in DPDK 24.11.3; not seen in 24.11

*Workarounds Attempted:*

   - Added device validation and error handling before stats reset
   - Implemented workaround to avoid direct stats reset (using baseline
   stats), Crash persists despite these changes
   - After implementing logic to skip rte_cryptodev_stats_reset() for
   IPsec_MB, SNOW3G, and AESNI_MB PMDs, the original segmentation fault was
   avoided. However, we then observed a separate crash during subsequent
   crypto device operations (such as session creation or device configuration).

*Request:*
Is this a known regression in DPDK 24.11.3? Are there any recommended fixes
or patches available? Any guidance on further debugging or mitigation would
be appreciated.


Regards,

Shoaib

[-- Attachment #2: Type: text/html, Size: 8809 bytes --]