TLS client handshake with ECDHE-SM2-X causes coredump under multi-core pressure testing

[wxy279]

Hi,
I’m currently developing a TLS client based on IPP-CP. When using RSA for handshake, everything works fine even under pressure testing with multiple cores (e.g., 2 cores achieve around 10,000 TPS per second without issues).
However, when I switch to SM2 for handshake, I encounter a problem: under single-core pressure testing, things look okay (about 500 TPS sustained for 150 seconds). But when I enable two cores, the program coredumps. The stack trace in the coredump appears incorrect and seems to indicate a memory corruption (possibly out-of-bounds memory write).
The CPU on the test device is Intel(R) Xeon(R) Silver 4314.
I’d like to ask: are there any special considerations or known issues in the SM2 handshake implementation when used in multi-core environments?
Thanks!

[wxy279]

Addition: ciphersuit are ECDHE-SM2-SM4-GCM-SM3 and ECDHE-SM2-SM4-CBC-SM3

[wxy279]

backtrace as following:

(gdb) bt
0x00007ff20a23b306 in k1_ippsHashUpdate_rmf () from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0xce6ebe2cd46d5595 in ?? ()
0x00007ff20a65e758 in k1_computeZa_user_id_hash_sm2.ctx ()
from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0xc2dbe553dc410846 in ?? ()
0x0000000000000020 in ?? ()
0x000000016c183340 in ?? ()
0x00007ff20a65e758 in k1_computeZa_user_id_hash_sm2.ctx ()
from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0x0000000000000001 in ?? ()
0x0000000000000030 in ?? ()
0x0000000000000002 in ?? ()
0x00007ff20a261312 in k1_KDF_sm3 () from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0x0100000000000080 in ?? ()
0x00000001ce64be20 in ?? ()
0x00000000000000e0 in ?? ()
0x936ad722c57d6800 in ?? ()
0x00007ff20a657fa0 in k1_ippsHashMethod_SM3_NI.method ()
from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0x00007ff20a65e8e8 in cpSM2KE_compute_hash_SM3.ctx ()
from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0x936ad722c57d6800 in ?? ()
0x00000000000000e0 in ?? ()
0x0000000000000020 in ?? ()
0x00000001ce64be7f in ?? ()
0x00000001ce644e70 in ?? ()
0x00000001ce634b00 in ?? ()
0x00007ff206262ba0 in ?? ()
0x00007ff20a42337b in k1_gfec_key_exchange_sm2_shared_key_avx512 ()
from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0x020592cce7af8111 in ?? ()
0x00000001ce64be20 in ?? ()
0x00000001ce64be60 in ?? ()
0x00007ff200000000 in ?? ()
0x00000001ce64bf58 in ?? ()
0x00000001ce64be40 in ?? ()
0x00000001ce644c00 in ?? ()
0x0000000000000020 in ?? ()
0x0000000000000018 in ?? ()
0x00007ff20a40aae0 in ?? () from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0x00000001ce64be80 in ?? ()
0x00000001ce649c80 in ?? ()
0x0000000000000100 in ?? ()
0x00007ff20a65c6f8 in k1_gsArithGF_p521r1_avx512.m ()
from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0x00007ff20a3f37ff in fesm2_from_mont_norder () from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0x00000001ce644eb0 in ?? ()
0x00000001ce64bd10 in ?? ()
0x00000018e4584476 in ?? ()
0x0000003000000010 in ?? ()
0x0000000000000010 in ?? ()
0x0000000000000000 in ?? ()

[wxy279]

I want to know SM2DHE functions are threaded-safe ? My programm is multi-thread , every thread can handle the TLS traffic, hope somebody can answer my confusion.

[rcao8]

Hi @wxy279 All Intel Cryptography are thread-safe design. As #97 issue , please share out a simple reproducer as well as the test steps?

[wxy279]

test_ipp_sm2dhe_keyshare.txt

[wxy279]

Hi, @rcao8 Thank you very much for your help. I’ve placed all the relevant code in the file (see the txt file above). I’ve consolidated all necessary input parameters and output results into a single function. By running this function in parallel using two threads, the issue can be reproduced. It should be noted that the memory allocation and deallocation, as well as the thread used to run the function, are based on the DPDK development environment. The threads that are launched do not perform any other operations; they simply call this function. Therefore, I believe that neither the origin of the thread nor the library used for memory allocation and deallocation should be the cause of the issue. If there’s anything missing in the provided code, please let me know and I’ll respond promptly. Once again, thank you for your assistance.

[wxy279]

Hi, @rcao8 I run into the same issue with ciphersuit ECC-SM2-SM4-GCM-SM3 when call ippsGFpECEncryptSM2_Ext, the backtrace as following, so there must be something I don't know or have misunderstood.
0x00007f184d020306 in k1_ippsHashUpdate_rmf () from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0x00007f184d43cfa0 in k1_ippsHashMethod_SM3_NI.method ()
from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0x00007f184d443758 in k1_computeZa_user_id_hash_sm2.ctx ()
from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0x00007f18490479f0 in ?? ()
0x0000000000000020 in ?? ()
0x000000017fbb41c1 in ?? ()
0x00007f184d443758 in k1_computeZa_user_id_hash_sm2.ctx ()
from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0x0000000000000002 in ?? ()
0x0000000000000010 in ?? ()
0x0000000000000002 in ?? ()
0x00007f184d046312 in k1_KDF_sm3 () from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12
0x0200000000000040 in ?? ()
0x000000018fdf1010 in ?? ()
0x531bd2478ae7b6c5 in ?? ()
0x47f4681e60798734 in ?? ()
0x1ebeeb379e6c342e in ?? ()
0x2fcd4917f55fd376 in ?? ()
0xba529969950d1300 in ?? ()
0x0000000000000020 in ?? ()
0x000000018fdf1030 in ?? ()
0x000000018fdf0f00 in ?? ()
0x0000000000000030 in ?? ()
0x000000017fbb4140 in ?? ()
0x00007f184d04323c in k1_ippsGFpECEncryptSM2_Ext ()
from /opt/intel/oneapi/ippcp/2025.1/lib/libippcp.so.12

[stanciuadrian]

Hello @wxy279 !
Thanks for providing the PoC code along with the details.
We have reproduced the issue and we're working on a patch.

[wxy279]

Hello @wxy279 ! Thanks for providing the PoC code along with the details. We have reproduced the issue and we're working on a patch.

Good, I'am waiting to verify the fix.