Details
-
Bug
-
Status: Closed (View Workflow)
-
Critical
-
Resolution: Fixed
-
10.6.21, 10.11.11
Description
While shipping latest February 2025 security updates to Ubuntu it was noticed that the riscv64 builds failed:
https://bugs.launchpad.net/ubuntu/+source/mariadb/+bug/2099785
[ 20%] Building C object mysys/CMakeFiles/mysys.dir/lf_alloc-pin.c.o
|
cd /<<PKGBUILDDIR>>/builddir/mysys && /usr/bin/cc -DHAVE_CONFIG_H -D_FILE_OFFSET_BITS=64 -I/<<PKGBUILDDIR>>/wsrep-lib/include -I/<<PKGBUILDDIR>>/wsrep-lib/wsrep-API/v26 -I/<<PKGBUILDDIR>>/builddir/include -I/<<PKGBUILDDIR>>/include/providers -I/<<PKGBUILDDIR>>/include -I/<<PKGBUILDDIR>>/mysys -g -O2 -fno-omit-frame-pointer -ffile-prefix-map=/<<PKGBUILDDIR>>=. -fstack-protector-strong -Wformat -Werror=format-security -fno-stack-clash-protection -fdebug-prefix-map=/<<PKGBUILDDIR>>=/usr/src/mariadb-1:10.11.11-0ubuntu0.24.04.1 -Wdate-time -D_FORTIFY_SOURCE=3 -Wdate-time -D_FORTIFY_SOURCE=3 -pie -fPIC -fstack-protector --param=ssp-buffer-size=4 -O3 -g -DNDEBUG -g -fno-omit-frame-pointer -fno-strict-aliasing -Wno-uninitialized -fno-omit-frame-pointer -DDBUG_OFF -Wall -Wdeclaration-after-statement -Wenum-compare -Wenum-conversion -Wextra -Wformat-security -Wmissing-braces -Wno-format-truncation -Wno-init-self -Wno-nonnull-compare -Wno-unused-parameter -Wvla -Wwrite-strings -std=gnu99 -Wdate-time -D_FORTIFY_SOURCE=3 -fPIC -MD -MT mysys/CMakeFiles/mysys.dir/lf_alloc-pin.c.o -MF CMakeFiles/mysys.dir/lf_alloc-pin.c.o.d -o CMakeFiles/mysys.dir/lf_alloc-pin.c.o -c /<<PKGBUILDDIR>>/mysys/lf_alloc-pin.c
|
/tmp/cclf2UNo.s: Assembler messages:
|
/tmp/cclf2UNo.s:265: Error: unrecognized opcode `pause', extension `zihintpause' required
|
/tmp/cclf2UNo.s:1168: Error: unrecognized opcode `pause', extension `zihintpause' required
|
make[4]: *** [mysys/CMakeFiles/mysys.dir/build.make:1367: mysys/CMakeFiles/mysys.dir/lf_alloc-pin.c.o] Error 1
|
[ 12%] Building C object storage/mroonga/vendor/groonga/lib/CMakeFiles/libgroonga.dir/id.c.o
|
In file included from /<<PKGBUILDDIR>>/mysys/lf_hash.cc:30:
|
/<<PKGBUILDDIR>>/include/my_cpu.h: In function ‘void MY_RELAX_CPU()’:
|
/<<PKGBUILDDIR>>/include/my_cpu.h:100:3: error: ‘__builtin_riscv_pause’ was not declared in this scope; did you mean ‘__builtin_riscv_fsflags’?
|
100 | __builtin_riscv_pause();
|
| ^~~~~~~~~~~~~~~~~~~~~
|
| __builtin_riscv_fsflags
|
Can you marko advise how we could patch the source of 10.11.11 and 10.6.21 so that the riscv64 builders in Ubuntu 24.04 and 22.04 respectively work?
Attachments
Issue Links
- is caused by
-
-
- Closed
-
Activity
otto, while the community pull request may feel a bit hacky, it is more portable than my initial suggestion, because it should work with any reasonably recent clang or GCC version. Can you please test it? Just replace the compiler builtin function with the following:
__asm__ volatile(".long 0x0100000f" ::: "memory"); |
Actually I saw this failure while preparing https://salsa.debian.org/mariadb-team/mariadb-server/-/merge_requests/96 but I didn't think it was severe initially.
I am now testing the fix in https://salsa.debian.org/mariadb-team/mariadb-server/-/merge_requests/108 and will post results in a couple of hours.
otto, thank you. I see that also the later build failed because you tested a version of the pull request instead of how I suggested it to be revised. I believe that it is simplest to use the numeric encoding of the instruction on any version of clang or GCC. Today morning’s update of the pull request should work for your testing, since you’re using GCC 11 and the pull request is (currently) treating GCC 14 or later in a special way.
I approved the community fix https://github.com/MariaDB/server/pull/3871 and otto tested it. Without this simple patch, the 10.6 branch would fail to build on RISC-V.
Alternate clang version - https://github.com/MariaDB/server/pull/3871 that might be made use of.