Let perftools to also bench OpenSSL forks.

[Sashan]

This change adjusts existing perftools so they cne be built by 3rd party SSL libraries (a.k.a. OpenSSL forks). Currently those tools can be used to bench OpenSSL forks:
randbytes, rsasign, handshake, sslnew, x509storeissuer.
evp_setpeer, writeread

The change also adds bench_config_forks.sh which installs forks and builds perftools. And bench_run_forks.sh which runs tests and plots results.

Fixes openssl/project#1845

[Sashan]

If you want to test thsoe changes to play around with them you need to apply diff below to cloned P$:

diff --git a/bench-scripts/bench_config_forks.sh b/bench-scripts/bench_config_forks.sh
index 9d21b52..96c2799 100755
--- a/bench-scripts/bench_config_forks.sh
+++ b/bench-scripts/bench_config_forks.sh
@@ -28,14 +28,14 @@ function build_perftools {
     # you must change link to repository, so script pulls
     # modified sources
     #
-    typeset PERFTOOLS='https://github.com/openssl/perftools'
+    typeset PERFTOOLS='https://github.com/sashan/perftools'
 
     cd ${WORKSPACE_ROOT} || exit 1
     #
     # you may also need to change clone command to
     # checkout correct branch.
     #
-    git clone ${PERFTOOLS} || exit 1
+    git clone -b project.1845 ${PERFTOOLS} || exit 1
     cd perftools/source || exit 1
 
     cmake -S ${WORKSPACE_ROOT}/perftools/source \

[nhorman]

is there a reason we're using ksh here as opposed to just sh?

[Sashan]

I got used to typeset to limit the scope of variables in shell. and I think typeset is ksh specific. I don't know if there is portable equivalent of typeset.

[nhorman]

Do we want to run with a pinned version of boring here? Or should be be using the latest on master, since we're (I think) trying to get regular performance numbers updated over time?

[Sashan]

the master branch may not build. it happened to me once or twice. hence I opted for a tag.

[Sashan]

I've changed the code to use master branch and added comment how to work around in case of breakgae.

[Sashan]

The code currently depends on .ksh script which does not fit well with ansible target after poking to cmake documentation I think it's worth to try bit harder and use cmake to fetch and build 3rd-party openssl libraries.

the PR is going to land after #86 anyway. and #86 creates enough conflicts to resolve.

[Sashan]

needs review the CMakeLists.txt after rebase.

[Sashan]

close + re-open to re-run CI.

the failing tests correctly find openssl-4.0 but they fail to find SSL_set1_dnsname() symbol.

[Sashan]

just for the record, this is the change which makes CI happy

diff --git a/source/CMakeLists.txt b/source/CMakeLists.txt
index 56bd224..8a1d186 100644
--- a/source/CMakeLists.txt
+++ b/source/CMakeLists.txt
@@ -205,7 +205,6 @@ else()
     find_package(OpenSSL REQUIRED ${OPENSSL_CONFIG_MODE})
     target_link_libraries(perf PUBLIC OpenSSL::SSL OpenSSL::Crypto)
     set(CMAKE_REQUIRED_LIBRARIES OpenSSL::SSL OpenSSL::Crypto)
-    set(CMAKE_REQUIRED_LIBRARIES OpenSSL::Crypto)
     set(CMAKE_REQUIRED_INCLUDES "${OPENSSL_INCLUDE_DIR}")
 endif()

the offending line got introduced during conflict resolution when rebasing to main branch

my local test systems use cmake version 3.31, the github instance seems to be using 3.16. it's interesting the local systems with cmake3.31 don't notice the issue everything works there.

I was lucky to meet my cmake --trace-expand friend which produces very verbose log on github. this item in log file helped me to figure out what went wrong:

     1	2026-03-09T12:39:28.0714710Z Linking C executable cmTC_cd7ad
     2	2026-03-09T12:39:28.0715130Z /usr/bin/cmake -E cmake_link_script CMakeFiles/cmTC_cd7ad.dir/link.txt --verbose=1
     3	2026-03-09T12:39:28.0716560Z /usr/bin/cc -DDEBUG -Werror -Wall     CMakeFiles/cmTC_cd7ad.dir/CheckSymbolExists.c.o  -o cmTC_cd7ad  -Wl,-rpath,/__w/perftools/perftools/openssl /__w/perftools/perftools/openssl/libcrypto.so 
     4	2026-03-09T12:39:28.0717014Z /usr/bin/ld: CMakeFiles/cmTC_cd7ad.dir/CheckSymbolExists.c.o: in function `main':
     5	2026-03-09T12:39:28.0717423Z CheckSymbolExists.c:(.text+0x1f): undefined reference to `SSL_set1_dnsname'
     6	2026-03-09T12:39:28.0717611Z collect2: error: ld returned 1 exit status

line 3 shows the linker command to build a binary. note the linker uses libcrypto.so instead of libssl.so. This brought my attention to set(CMAKE_REQUIRED_LIBRARIES, ...) line. Have not investigated further if the change in behavior is caused by cmake version (3.16 vs. 3.31) or by another difference in environements.