Repository navigation

Espressif: add ESP32-C5 hardware acceleration #892

Workflow file for this run

.github/workflows/autosar.yml at 8f71cd3

	name: AUTOSAR Tests

	# START OF COMMON SECTION
	on:
	push:
	branches: [ 'release/**' ]
	pull_request:
	types: [opened, synchronize, reopened, ready_for_review]
	branches: [ '*' ]

	concurrency:
	group: ${{ github.workflow }}-${{ github.ref }}
	cancel-in-progress: true

	permissions:
	contents: read
	# END OF COMMON SECTION

	jobs:
	autosar_port_test:
	name: AUTOSAR port test (${{ matrix.name }})
	if: ${{ (github.repository_owner == 'wolfssl') && (github.event_name != 'pull_request' \|\| github.event.pull_request.draft == false) }}
	runs-on: ubuntu-24.04
	timeout-minutes: 15
	strategy:
	fail-fast: false
	matrix:
	include:
	# Default key selection: GetKey() scans the keystore and takes the
	# first slot whose key element ID and key length match the job.
	# Covers singleshot_test, update_test, random_test and key_test.
	- name: default
	config: "--enable-autosar"
	expect_redirect: false

	# Key input redirection pins each key element to one keystore slot at
	# compile time. Nothing built this half of the port before: GetKey()
	# used runtime struct members as switch case labels, so no build with
	# REDIRECTION_CONFIG defined compiled at all, and redirect_test is
	# gated behind that same macro -- the one test covering the feature
	# could never run. The slot assignment below is the port README's
	# example config: cipher key to slot 1, cipher IV to slot 4.
	- name: key-redirection
	config: >-
	--enable-autosar
	CPPFLAGS="-DREDIRECTION_CONFIG=0x03
	-DREDIRECTION_IN1_KEYID=1 -DREDIRECTION_IN1_KEYELMID=0x01
	-DREDIRECTION_IN2_KEYID=4 -DREDIRECTION_IN2_KEYELMID=0x05"
	expect_redirect: true

	# The MAC services are an add-on enable, so they need their own entry:
	# nothing else builds csm.c's Csm_MacGenerate/Csm_MacVerify or the
	# driver's CMAC path, and mac_test is compiled in only here. The entry
	# also covers the option pulling wolfCrypt's CMAC in by itself --
	# WOLFSSL_AES_DIRECT included, without which wc_InitCmac() compiles
	# fine and then fails at runtime.
	- name: cmac
	config: "--enable-autosar --enable-autosar-cmac"
	expect_redirect: false
	expect_mac: true

	# Csm_ConfigType's devId is what turns the Crypto driver into a front
	# end for an HSM, and devid_test is compiled in only with the crypto
	# callback machinery present. It registers a counting callback and
	# asserts the cipher, MAC and RNG jobs all reach it with the right
	# devId, and that Csm_Init(NULL) still runs entirely in software.
	- name: cryptocb
	config: >-
	--enable-autosar --enable-autosar-cmac --enable-cryptocb
	CPPFLAGS=-DWOLF_CRYPTO_CB_FREE
	expect_redirect: false
	expect_mac: true
	expect_devid: true

	# Key handles: a keystore slot names a key in the device instead of
	# holding it, so the key never enters RAM. keyhandle_test checks that
	# a named key produces the same ciphertext as the key itself, which
	# is what proves the identifier reached the device rather than
	# something merely looking encrypted. WOLF_PRIVATE_KEY_ID is a raw
	# -D, not a configure option.
	- name: keyhandle
	config: >-
	--enable-autosar --enable-autosar-cmac --enable-cryptocb
	CPPFLAGS="-DWOLF_PRIVATE_KEY_ID -DWOLF_CRYPTO_CB_FREE"
	expect_redirect: false
	expect_mac: true
	expect_devid: true
	expect_keyhandle: true

	# Development error reporting. With WOLFSSL_AUTOSAR_DET the port
	# forwards to Det_ReportError(), which the test program supplies --
	# hence the static build, so the undefined symbol resolves from the
	# application rather than from the shared library. det_test asserts
	# the module and error IDs are the ones the CSM specification
	# assigns; an earlier version of this port reported a local 0-based
	# enum, so every ID a DET would have seen was wrong.
	- name: det
	config: >-
	--enable-autosar --enable-autosar-cmac
	--enable-static --disable-shared
	CPPFLAGS=-DWOLFSSL_AUTOSAR_DET
	expect_redirect: false
	expect_mac: true
	expect_det: true

	# DET without the MAC services. csm.c compiles the Det_ReportError()
	# call whenever WOLFSSL_AUTOSAR_DET is defined, so a build that does
	# not define the symbol fails to link -- which this configuration did,
	# because test.c had its definition behind CMAC as well. det_test
	# needs a service that reports a development error to drive it, and
	# only the MAC services do, so it does not run here; the point of the
	# entry is that the link succeeds.
	- name: det-nocmac
	config: >-
	--enable-autosar
	--enable-static --disable-shared
	CPPFLAGS=-DWOLFSSL_AUTOSAR_DET
	expect_redirect: false
	expect_no_det_test: true

	# Key redirection together with the MAC services. That combination
	# cannot tell CRYPTO_KE_CIPHER_KEY from CRYPTO_KE_MAC_KEY -- both are
	# element 0x01 and redirection pins one slot for every service -- so
	# crypto.c refuses it at build time unless the integrator says they
	# only use one of the two services. This entry is that case, and the
	# step below asserts the refusal itself.
	#
	# Neither entry above covered redirection with CMAC, which is how the
	# ambiguity went unnoticed: keysep_test failed there and nothing ran
	# it.
	- name: key-redirection-cmac
	config: >-
	--enable-autosar --enable-autosar-cmac
	CPPFLAGS="-DREDIRECTION_CONFIG=0x03
	-DREDIRECTION_IN1_KEYID=1 -DREDIRECTION_IN1_KEYELMID=0x01
	-DREDIRECTION_IN2_KEYID=4 -DREDIRECTION_IN2_KEYELMID=0x05
	-DWOLFSSL_AUTOSAR_ALLOW_AMBIGUOUS_KEY_ELEMENT"
	expect_redirect: true
	expect_mac: true
	expect_ambiguous_opt_out: true

	# The escape hatch from the ambiguous-key-element refusal. It is a
	# security-relevant default being turned off, so it gets its own
	# entry rather than being trusted to compile: keysep_test drops the
	# refusal assertion here and must still pass on the named-slot path,
	# which is what the ECU being told to use it depends on.
	- name: ambiguous-key
	config: >-
	--enable-autosar --enable-autosar-cmac
	CPPFLAGS=-DWOLFSSL_AUTOSAR_ALLOW_AMBIGUOUS_KEY_ELEMENT
	expect_redirect: false
	expect_mac: true
	steps:
	- uses: actions/checkout@v5
	name: Checkout wolfSSL

	- name: Autogen
	run: ./autogen.sh

	# --enable-autosar-cmac is an add-on, not the default. Enabling it
	# without --enable-autosar has to be an error rather than a quiet no-op.
	#
	# In tree, and before the real configure below. AC_MSG_ERROR exits before
	# AC_OUTPUT, so this writes no Makefile, options.h or config.status for
	# the build to trip over. Doing it afterwards out of tree does not work:
	# automake's VPATH guard rejects a source directory that already has a
	# config.status, so the log would carry that error instead of the one
	# being asserted.
	- name: Assert autosar-cmac requires autosar
	if: ${{ matrix.expect_mac }}
	run: \|
	if ./configure --enable-autosar-cmac > cmac-alone.log 2>&1; then
	echo "--enable-autosar-cmac configured without --enable-autosar" >&2
	exit 1
	fi
	grep -q 'requires --enable-autosar' cmac-alone.log

	- name: Build the AUTOSAR port test
	run: \|
	./configure ${{ matrix.config }}
	make -j$(nproc) wolfcrypt/src/port/autosar/test.test

	- name: Run the AUTOSAR port test
	run: \|
	set -o pipefail
	./wolfcrypt/src/port/autosar/test.test \| tee autosar-test.log

	# A build that silently loses the redirection macros still passes every
	# test above, because redirect_test is simply not compiled in -- which is
	# how the feature stayed broken. Assert it actually ran, so this entry
	# cannot go green without having exercised redirection.
	- name: Assert redirect_test ran
	if: ${{ matrix.expect_redirect }}
	run: grep -qx 'redirect_test \.\.\. pass' autosar-test.log

	# The converse: the default entry must not be silently picking up a
	# redirection config from somewhere, or it stops covering the scan path.
	- name: Assert redirect_test did not run
	if: ${{ !matrix.expect_redirect }}
	run: \|
	if grep -q 'redirect_test' autosar-test.log; then
	echo "redirect_test ran in the default config" >&2
	exit 1
	fi

	# Same reasoning as the redirect assertion: mac_test is compiled in only
	# when the option took effect, so its absence would pass silently.
	- name: Assert mac_test ran
	if: ${{ matrix.expect_mac }}
	run: grep -qx 'mac_test \.\.\. pass' autosar-test.log

	# Same reasoning again: devid_test is absent without --enable-cryptocb,
	# so its silence would look like success.
	- name: Assert devid_test ran
	if: ${{ matrix.expect_devid }}
	run: grep -qx 'devid_test \.\.\. pass' autosar-test.log

	- name: Assert keyhandle_test ran
	if: ${{ matrix.expect_keyhandle }}
	run: grep -qx 'keyhandle_test \.\.\. pass' autosar-test.log

	# A MAC job whose key lives in the device. Compiled in only with the key
	# handle and MAC options together, so its silence would look like
	# success -- and it is the case that proves the MAC services can use a
	# key they never see.
	- name: Assert machandle_test ran
	if: ${{ matrix.expect_keyhandle }}
	run: grep -qx 'machandle_test \.\.\. pass' autosar-test.log

	# reinit_test runs everywhere, but only observes contexts actually being
	# freed where the device is told about frees.
	- name: Assert reinit_test ran
	run: grep -qx 'reinit_test \.\.\. pass' autosar-test.log

	# cancel_test is the counterpart: a re-init is refused while a job is
	# open, so an abandoned job has to be cancellable or the configuration
	# could never be changed again.
	- name: Assert cancel_test ran
	run: grep -qx 'cancel_test \.\.\. pass' autosar-test.log

	# The threaded case. Single-threaded builds compile it out, and none of
	# these entries are single-threaded, so silence here means it stopped
	# being built rather than that it had nothing to do.
	- name: Assert gate_test ran
	run: grep -qx 'gate_test \.\.\. pass' autosar-test.log

	- name: Assert det_test ran
	if: ${{ matrix.expect_det }}
	run: grep -qx 'det_test \.\.\. pass' autosar-test.log

	# The DET-without-CMAC entry exists for the link, so det_test cannot run
	# there. Assert that rather than leaving the entry asserting nothing.
	- name: Assert det_test did not run
	if: ${{ matrix.expect_no_det_test }}
	run: \|
	if grep -q 'det_test' autosar-test.log; then
	echo "det_test ran without the MAC services" >&2
	exit 1
	fi

	# keysep_test provisions a cipher key and a MAC key at the same time,
	# which is the case the shared key element ID 0x01 used to get wrong: the
	# MAC job took whichever slot came first. It needs both services, so it
	# is compiled in exactly where mac_test is.
	# The opt-out exists because the combination is otherwise refused. Assert
	# that refusal, or the entry above would silently become the only thing
	# covering redirection with CMAC while the guard rotted.
	- name: Assert redirection with CMAC is refused without the opt-out
	if: ${{ matrix.expect_ambiguous_opt_out }}
	run: \|
	# Touch the source first. The step above already built this target,
	# so make would otherwise report it up to date and exit 0 without
	# recompiling anything -- which reads as "the refusal did not fire"
	# and fails the job on a build that never happened.
	#
	# CPPFLAGS on the command line replaces what configure recorded,
	# which for this entry is these same defines plus the opt-out. So
	# this rebuilds crypto.c with the opt-out removed and nothing else
	# changed.
	touch wolfcrypt/src/port/autosar/crypto.c
	set +e
	make wolfcrypt/src/port/autosar/test.test \
	CPPFLAGS="-DREDIRECTION_CONFIG=0x03 \
	-DREDIRECTION_IN1_KEYID=1 -DREDIRECTION_IN1_KEYELMID=0x01 \
	-DREDIRECTION_IN2_KEYID=4 -DREDIRECTION_IN2_KEYELMID=0x05" \
	> ambig.log 2>&1
	rc=$?
	set -e
	if [ "$rc" -eq 0 ]; then
	echo "redirection + CMAC built without the opt-out" >&2
	exit 1
	fi
	grep -q 'cannot tell a cipher key from a MAC key' ambig.log
	# Leave the tree as the entry configured it, for any step after this.
	touch wolfcrypt/src/port/autosar/crypto.c
	make wolfcrypt/src/port/autosar/test.test > /dev/null

	- name: Assert keysep_test ran
	if: ${{ matrix.expect_mac }}
	run: grep -qx 'keysep_test \.\.\. pass' autosar-test.log

	- name: Print errors
	if: ${{ failure() }}
	run: \|
	if [ -f config.log ] ; then
	tail -n 100 config.log
	fi

	autosar_cmake:
	name: AUTOSAR CMake build and link
	if: ${{ (github.repository_owner == 'wolfssl') && (github.event_name != 'pull_request' \|\| github.event.pull_request.draft == false) }}
	runs-on: ubuntu-24.04
	timeout-minutes: 15
	steps:
	- uses: actions/checkout@v5
	name: Checkout wolfSSL

	- name: Configure and build with WOLFSSL_AUTOSAR=yes
	run: \|
	cmake -S . -B build_autosar -DWOLFSSL_AUTOSAR=yes \
	-DBUILD_SHARED_LIBS=ON -DWOLFSSL_EXAMPLES=no \
	-DWOLFSSL_CRYPT_TESTS=no
	cmake --build build_autosar -j$(nproc)

	# The MAC services have their own CMake logic -- WOLFSSL_AUTOSAR_CMAC
	# implies WOLFSSL_CMAC and WOLFSSL_AES_DIRECT, and that implication runs
	# after the options are read -- and every CMAC entry in the matrix above
	# is Autotools, so without this a regression in the CMake path reaches
	# nobody until an integrator hits it.
	- name: Configure and build with WOLFSSL_AUTOSAR_CMAC=yes
	run: \|
	cmake -S . -B build_autosar_cmac -DWOLFSSL_AUTOSAR=yes \
	-DWOLFSSL_AUTOSAR_CMAC=yes -DBUILD_SHARED_LIBS=ON \
	-DWOLFSSL_EXAMPLES=no -DWOLFSSL_CRYPT_TESTS=no
	cmake --build build_autosar_cmac -j$(nproc)

	# The implication is the point: asking for the MAC services alone has to
	# pull in the wolfCrypt pieces they call, or every MAC job fails at run
	# time in a build that looks configured correctly.
	- name: Assert the CMAC implication reached options.h
	run: \|
	for d in WOLFSSL_AUTOSAR_CMAC WOLFSSL_CMAC WOLFSSL_AES_DIRECT ; do
	grep -q "define $d" build_autosar_cmac/wolfssl/options.h \|\| {
	echo "$d missing from the CMake AUTOSAR CMAC build" >&2
	exit 1
	}
	done

	- name: Link a MAC application against the AUTOSAR API
	run: \|
	cat > cmac_link_probe.c <<'EOF'
	#include <wolfssl/options.h>
	#include <wolfssl/wolfcrypt/port/autosar/Csm.h>
	int main(void)
	{
	unsigned char mac[16];
	unsigned int macSz = (unsigned int)sizeof(mac);

	Csm_Init(NULL);
	(void)Csm_MacGenerate(1u, CRYPTO_OPERATIONMODE_SINGLECALL,
	(const unsigned char*)"msg", 3u, mac, &macSz);
	return 0;
	}
	EOF
	cc -o cmac_link_probe cmac_link_probe.c \
	-I. -Ibuild_autosar_cmac -Lbuild_autosar_cmac -lwolfssl
	LD_LIBRARY_PATH=build_autosar_cmac ./cmac_link_probe

	# And the prerequisite the option carries: the MAC services are an
	# add-on, so asking for them without the port has to fail at configure
	# time rather than produce a library whose Csm_Mac* calls do not link.
	- name: Assert WOLFSSL_AUTOSAR_CMAC requires WOLFSSL_AUTOSAR
	run: \|
	if cmake -S . -B build_cmac_alone -DWOLFSSL_AUTOSAR_CMAC=yes \
	-DWOLFSSL_EXAMPLES=no -DWOLFSSL_CRYPT_TESTS=no \
	> cmac_alone.log 2>&1 ; then
	echo "WOLFSSL_AUTOSAR_CMAC configured without WOLFSSL_AUTOSAR" >&2
	exit 1
	fi
	grep -q 'requires WOLFSSL_AUTOSAR' cmac_alone.log

	# The option used to append only -DWOLFSSL_AUTOSAR, never the port
	# sources, so WOLFSSL_AUTOSAR reached options.h while csm.c, cryif.c and
	# crypto.c went uncompiled: an application's Csm_* calls compiled against
	# Csm.h and then failed to link. Linking one here is the assertion --
	# grepping the source list would not catch a future regression in which
	# the files build but the symbols do not get exported.
	- name: Link an application against the AUTOSAR API
	run: \|
	cat > csm_link_probe.c <<'EOF'
	#include <wolfssl/options.h>
	#include <wolfssl/wolfcrypt/port/autosar/Csm.h>
	int main(void) { Csm_Init(NULL); return 0; }
	EOF
	cc -o csm_link_probe csm_link_probe.c \
	-I. -Ibuild_autosar -Lbuild_autosar -lwolfssl
	LD_LIBRARY_PATH=build_autosar ./csm_link_probe

	# The other branch of the CMake source selection: a user-settings build
	# takes WOLFSSL_AUTOSAR from user_settings.h, where CMake cannot see it,
	# so the source list keys on WOLFSSL_USER_SETTINGS as well. Both jobs
	# above pass WOLFSSL_AUTOSAR=yes on the command line and would not
	# notice that branch breaking -- which is how an ECU build came to link
	# against Csm.h with nothing behind it. The port is named ONLY in
	# user_settings.h here, on purpose.
	- name: Link an application against a user-settings CMake build
	run: \|
	mkdir -p us_autosar
	{ cat examples/configs/user_settings_all.h ; \
	echo '#define WOLFSSL_AUTOSAR' ; \
	echo '#define WOLFSSL_AUTOSAR_CMAC' ; \
	} > us_autosar/user_settings.h
	# The include path has to reach the assembler too, not just the C
	# compiler: the SP assembly sources run through the preprocessor and
	# include settings.h, which includes "user_settings.h", and
	# CMAKE_C_FLAGS does not apply to the ASM language. Without this the
	# build fails on sp_x86_64_asm.S with "user_settings.h: No such file
	# or directory" -- which is worth knowing for a real user-settings
	# CMake build, not just for this probe.
	cmake -S . -B build_us -DWOLFSSL_USER_SETTINGS=yes \
	-DWOLFSSL_EXAMPLES=no -DWOLFSSL_CRYPT_TESTS=no \
	-DBUILD_SHARED_LIBS=ON \
	-DCMAKE_C_FLAGS="-I$PWD/us_autosar" \
	-DCMAKE_ASM_FLAGS="-I$PWD/us_autosar"
	cmake --build build_us -j$(nproc)
	cat > us_probe.c <<'EOF'
	#include <wolfssl/wolfcrypt/settings.h>
	#include <wolfssl/wolfcrypt/port/autosar/Csm.h>
	int main(void)
	{
	Csm_ConfigType cfg = WOLFSSL_CSM_CONFIG_DEFAULT;
	unsigned char mac[16];
	unsigned int macSz = (unsigned int)sizeof(mac);

	Csm_Init(&cfg);
	(void)Csm_MacGenerate(1u, CRYPTO_OPERATIONMODE_SINGLECALL,
	(const unsigned char*)"msg", 3u, mac, &macSz);
	return (Csm_CancelJob(1u, CRYPTO_OPERATIONMODE_FINISH) == E_OK)
	? 1 : 0;
	}
	EOF
	# Shared, like the probes above: a static libwolfssl.a would make
	# this link line carry the library's own dependencies -- libm, for
	# DiscreteLogWorkFactor()'s pow() and log() -- which says nothing
	# about the AUTOSAR sources being in the build.
	cc -o us_probe us_probe.c -DWOLFSSL_USER_SETTINGS \
	-Ius_autosar -I. -Ibuild_us -Lbuild_us -lwolfssl
	LD_LIBRARY_PATH=build_us ./us_probe

	# The converse: a default build must not compile the port.
	- name: Assert the port stays out of a default build
	run: \|
	cmake -S . -B build_default -DWOLFSSL_EXAMPLES=no \
	-DWOLFSSL_CRYPT_TESTS=no > /dev/null
	if grep -q 'port/autosar' \
	build_default/CMakeFiles/wolfssl.dir/build.make; then
	echo "autosar port compiled without WOLFSSL_AUTOSAR" >&2
	exit 1
	fi

Navigation Menu

Repository navigation

Espressif: add ESP32-C5 hardware acceleration #892

Workflow file

Espressif: add ESP32-C5 hardware acceleration #892

Uh oh!

Workflow file for this run