Fix sporadic CI failure in crypto test for nrf platform

[pan-apple]

Problem

CI is intermittently failing in crypto tests while running nrf platform tests.

Failure logs are

1: [ Test-CHIP-SecureChannel ]
1: mbedTLS error: BIGNUM - Memory allocation failed
1: 
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:53: assertion failed: "channel.Init(keypair, keypair2.Pubkey(), nullptr, 0, (const uint8_t *) info, sizeof(info)) == CHIP_NO_ERROR"
1: mbedTLS error: BIGNUM - Memory allocation failed
1: 
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:57: assertion failed: "channel.Init(keypair, keypair2.Pubkey(), nullptr, 0, (const uint8_t *) info, sizeof(info)) == CHIP_ERROR_INCORRECT_STATE"
1: mbedTLS error: BIGNUM - Memory allocation failed
1: 
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:64: assertion failed: "channel2.Init(keypair, keypair2.Pubkey(), (const uint8_t *) salt, sizeof(salt), (const uint8_t *) info, sizeof(info)) == CHIP_NO_ERROR"
1: [ Test-CHIP-SecureChannel : Init    ] : FAILED
1: mbedTLS error: BIGNUM - Memory allocation failed
1: 
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:90: assertion failed: "channel.Init(keypair, keypair2.Pubkey(), (const uint8_t *) salt, sizeof(salt), (const uint8_t *) info, sizeof(info)) == CHIP_NO_ERROR"
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:95: assertion failed: "channel.Encrypt(nullptr, 0, nullptr, packetHeader, mac) == CHIP_ERROR_INVALID_ARGUMENT"
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:96: assertion failed: "channel.Encrypt(plain_text, 0, nullptr, packetHeader, mac) == CHIP_ERROR_INVALID_ARGUMENT"
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:97: assertion failed: "channel.Encrypt(plain_text, sizeof(plain_text), nullptr, packetHeader, mac) == CHIP_ERROR_INVALID_ARGUMENT"
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:101: assertion failed: "channel.Encrypt(plain_text, sizeof(plain_text), output, packetHeader, mac) == CHIP_NO_ERROR"
1: [ Test-CHIP-SecureChannel : Encrypt ] : FAILED
1: mbedTLS error: BIGNUM - Memory allocation failed
1: 
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:121: assertion failed: "channel.Init(keypair, keypair2.Pubkey(), (const uint8_t *) salt, sizeof(salt), (const uint8_t *) info, sizeof(info)) == CHIP_NO_ERROR"
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:124: assertion failed: "channel.Encrypt(plain_text, sizeof(plain_text), encrypted, packetHeader, mac) == CHIP_NO_ERROR"
1: mbedTLS error: BIGNUM - Memory allocation failed
1: 
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:132: assertion failed: "channel2.Init(keypair2, keypair.Pubkey(), (const uint8_t *) salt, sizeof(salt), (const uint8_t *) info, sizeof(info)) == CHIP_NO_ERROR"
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:137: assertion failed: "channel2.Decrypt(nullptr, 0, nullptr, packetHeader, mac) == CHIP_ERROR_INVALID_ARGUMENT"
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:138: assertion failed: "channel2.Decrypt(encrypted, 0, nullptr, packetHeader, mac) == CHIP_ERROR_INVALID_ARGUMENT"
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:139: assertion failed: "channel2.Decrypt(encrypted, sizeof(encrypted), nullptr, packetHeader, mac) == CHIP_ERROR_INVALID_ARGUMENT"
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:143: assertion failed: "channel2.Decrypt(encrypted, sizeof(plain_text), output, packetHeader, mac) == CHIP_NO_ERROR"
1: ../../../../../../src/transport/tests/TestSecureSession.cpp:145: assertion failed: "memcmp(plain_text, output, sizeof(plain_text)) == 0"
1: [ Test-CHIP-SecureChannel : Decrypt ] : FAILED
1: Failed Tests:   3 / 3
1: Failed Asserts: 16 / 26

Summary of Changes

PASESession.Clear() was calling mSpake2p.Init(nullptr);, but it doesn't clear the state of the Spake2p object. It initialized the SHA256 context, which was eventually leaked. The test runs were calling this repeatedly that drained the buffer pool from mbedTLS library.

[bzbarsky-apple]

Ah, and the point is since we are initing mSpake2p with mCommissioningHash and we are clearing that we don't need to mess with mSpake2p? This could actually use a comment explaining why we don't need to touch that member here. Looks good to me with the comment added.

[github-actions]

Size increase report for "esp32-example-build" from 4c56ede

File	Section	File	VM
chip-all-clusters-app.elf	.flash.text	-8	-8

Full report output

BLOAT REPORT

Files found only in the build output:
    report.csv

Comparing ./master_artifact/chip-all-clusters-app.elf and ./pull_artifact/chip-all-clusters-app.elf:

sections,vmsize,filesize
.debug_loc,0,6
.debug_line,0,-4
.flash.text,-8,-8
.debug_info,0,-22

[andy31415]

It says 'we don't need to' which sounds a as if we could but it does not matter.

However from the change, it seems that calling 'init' actually caused unit test failues, so we likely we 'must not init spake2p'. Should we update the comment? I think the comment may make sense within this PR, however looking in time the comment could use a bit more context. Like:

// clearing mCommisioningHash will automatically clear mSpake2p. 
// NOTE: do NOT call mSpake2p.Init(nullptr) here since that will leak the spake hash.

... assuming the above comment is really what is going on.
Could we fix the possiblity of a memory leak when init is called?