On Tue, Apr 29, 2025 at 08:44:50PM +0200, Dimitry Andric wrote:
> On 29 Apr 2025, at 19:43, Shawn Webb <shawn.w...@hardenedbsd.org> wrote:
> > 
> > On Sun, Apr 27, 2025 at 07:42:44PM +0200, Dimitry Andric wrote:
> >> On 27 Apr 2025, at 17:04, Shawn Webb <shawn.w...@hardenedbsd.org> wrote:
> >>> 
> >>> On Sat, Apr 26, 2025 at 06:06:54PM +0200, Dimitry Andric wrote:
> >> ...
> >>>> Please let me know if you encounter any problems resulting due to this
> >>>> change, as I intend to MFC it. For example, I tried covering all
> >>>> incremental build scenarios, but I may have missed some corner case.
> >>> 
> >>> Hey Dimitry,
> >>> 
> >>> I suspect this may be a problem specific to HardenedBSD, but it looks
> >>> like cc occasionally crashes. It hits an assert at
> >>> /usr/src/contrib/llvm-project/clang/lib/Driver/Driver.cpp:2702.
> >>> 
> >>> I can reproduce this by running `env SHELL=/bin/sh make buildenv` at
> >>> the top of /usr/src. Though, it doesn't reproduce 100%, but perhaps
> >>> around 60%.
> >> 
> >> It's asserting on this line:
> >> 
> >>            assert(!CCGenDiagnostics && "stdin produces no crash 
> >> reproducer");
> >> 
> >> I think during make buildenv the make framework will run cc --version
> >> and ld --version to get at the compiler and linker version, but it could
> >> be that it's doing some weird combination that hasn't been thought of.
> >> Can you get the exact command line out of the debugger?
> > 
> > For some reason, I'm failing to figure out how to view the program's
> > invocation using lldb or gdb. But, running `strings
> > /usr/obj/usr/src/amd64.amd64/cc.core`, I see this, which may or may
> > not be the arguments passed to cc:
> > 
> > ==== BEGIN OUTPUT ====
> > "-cc1" "-ferror-limit" "19" "-o" "-" "-disable-free" "-E" "-x" "c" "-" 
> > "-tune-cpu" "generic" "-target-cpu" "x86-64" "-triple"
> > "x86_64-unknown-freebsd15.0" "-resource-dir" "/usr/lib/clang/19" "-isystem" 
> > "/usr/lib/clang/19/include" "-internal-externc-i
> > system" "/usr/include" "-std=gnu17" "-fskip-odr-check-in-gmf" 
> > "-ftrivial-auto-var-init=zero" "-fgnuc-version=4.2.1" "-ffp-con
> > tract=on" "-fno-experimental-relative-c++-abi-vtables" 
> > "-fno-file-reproducible" "-O0" "-fdebug-compilation-dir=/usr/obj/usr/s
> > rc/amd64.amd64" "-fcoverage-compilation-dir=/usr/obj/usr/src/amd64.amd64" 
> > "-faddrsig" "-mrelocation-model" "static" "-debugge
> > r-tuning=gdb" "-funwind-tables=2" "-mconstructor-aliases" 
> > "-clear-ast-before-backend" "-main-file-name" "-" "-mframe-pointer=
> > all" "-fdiagnostics-hotness-threshold=0" 
> > "-fdiagnostics-misexpect-tolerance=0" "-D" "__GCC_HAVE_DWARF2_CFI_ASM=1"
> > ==== END OUTPUT ====
> > 
> > I hope that helps. Otherwise, can you provide me a hint as to how to
> > get lldb/gdb to show the arguments passed to cc?
> 
> "settings show target.run-args", but I just checked buildenv, and it really 
> only runs "cc --version" twice. If you do that in a loop, does it sometimes 
> crash?

I think that `settings show...` command only pertains to actually
executing the application:

==== BEGIN OUTPUT ====
$ lldb /usr/bin/cc -c /usr/obj/usr/src/amd64.amd64/cc.core
(lldb) target create "/usr/bin/cc" --core "/usr/obj/usr/src/amd64.amd64/cc.core"
Core file '/usr/obj/usr/src/amd64.amd64/cc.core' (x86_64) was loaded.
(lldb) settings show target.run-args
target.run-args (arguments) =
==== END OUTPUT ====

Running `cc --version` does not reproduce the crash. I ran that in a
loop over 200 iterations with zero crashes. I can reproduce it
somewhat reliably with `env SHELL=/bin/sh make buildenv` at the top of
the src tree.

Thanks,

-- 
Shawn Webb
Cofounder / Security Engineer
HardenedBSD

Signal Username:  shawn_webb.74
Tor-ified Signal: +1 303-901-1600 / shawn_webb_opsec.50
https://git.hardenedbsd.org/hardenedbsd/pubkeys/-/raw/master/Shawn_Webb/03A4CBEBB82EA5A67D9F3853FF2E67A277F8E1FA.pub.asc

Attachment: signature.asc
Description: PGP signature

Reply via email to