testsuite: test/elf/mold-wrapper.sh fails since eb7433f3ac004b6b12c783bbc9ea88ca6223de96

[marxin]

Fails on x86_64-linux-gnu with:

$ ./test/elf/mold-wrapper.sh
Testing mold-wrapper ... execl: Bad address
$ 

valgrind tells:

--18102-- WARNING: unhandled amd64-linux syscall: 334
--18102-- You may be able to write your own handler.
--18102-- Read the file README_MISSING_SYSCALL_OR_IOCTL.
--18102-- Nevertheless we consider this a bug.  Please report
--18102-- it at http://valgrind.org/support/bug_reports.html.
==18102== Syscall param execve(argv) points to uninitialised byte(s)
==18102==    at 0x49803EB: execve (syscall-template.S:120)
==18102==    by 0x4858603: execl (in /home/marxin/Programming/mold/mold-wrapper.so)
==18102==    by 0x4011FE: main (in /home/marxin/Programming/mold/out/test/elf/mold-wrapper/exe)
==18102==  Address 0x1ffefff8d0 is on thread 1's stack
==18102==  in frame #1, created by execl (???:)
==18102== 
==18102== Syscall param execve(argv[i]) points to unaddressable byte(s)
==18102==    at 0x49803EB: execve (syscall-template.S:120)
==18102==    by 0x4858603: execl (in /home/marxin/Programming/mold/mold-wrapper.so)
==18102==    by 0x4011FE: main (in /home/marxin/Programming/mold/out/test/elf/mold-wrapper/exe)
==18102==  Address 0x40000 is not stack'd, malloc'd or (recently) free'd

[Alcaro]

For the record, 334 is rseq(), and is unrelated to the exec issue. Ignore it, or upgrade to Valgrind 3.19 (released three weeks ago). https://bugs.kde.org/show_bug.cgi?id=405377

[marxin]

Correct, new valgrind does not show that syscall error. Thanks.

[rui314]

It looks like it's resolved, so I'm closing this bug.

[marxin]

Sorry, I was unclear. It's still present:

$ LD_PRELOAD=/home/marxin/Programming/mold/mold-wrapper.so MOLD_PATH=out/test/elf/mold-wrapper/a.sh MOLD_PATH=out/test/elf/mold-wrapper/a.sh valgrind out/test/elf/mold-wrapper/exe execl
==3517== Memcheck, a memory error detector
==3517== Copyright (C) 2002-2022, and GNU GPL'd, by Julian Seward et al.
==3517== Using Valgrind-3.19.0 and LibVEX; rerun with -h for copyright info
==3517== Command: out/test/elf/mold-wrapper/exe execl
==3517== 
==3517== Syscall param execve(argv) points to uninitialised byte(s)
==3517==    at 0x49833EB: execve (syscall-template.S:120)
==3517==    by 0x485B603: execl (in /home/marxin/Programming/mold/mold-wrapper.so)
==3517==    by 0x4011FE: main (in /home/marxin/Programming/mold/out/test/elf/mold-wrapper/exe)
==3517==  Address 0x1ffefff870 is on thread 1's stack
==3517==  in frame #1, created by execl (???:)
==3517== 
==3517== Syscall param execve(argv[i]) points to unaddressable byte(s)
==3517==    at 0x49833EB: execve (syscall-template.S:120)
==3517==    by 0x485B603: execl (in /home/marxin/Programming/mold/mold-wrapper.so)
==3517==    by 0x4011FE: main (in /home/marxin/Programming/mold/out/test/elf/mold-wrapper/exe)
==3517==  Address 0x40000 is not stack'd, malloc'd or (recently) free'd
==3517== 
==3517== execve(0x1ffeffff49(out/test/elf/mold-wrapper/a.sh), 0x1ffefff860, 0x1ffefffb00) failed, errno 14
==3517== EXEC FAILED: I can't recover from execve() failing, so I'm dying.
==3517== Add more stringent tests in PRE(sys_execve), or work out how to recover.

[rui314]

I believe it's a false positive. We pass an environment variable to execve as a first argument, and Valgrind seems to complain that the environment variable is not initialized. From the Valgrind's view, it's no surprise that the variable looks like uninitialized because no one wrote to that memory area holding the environment variable. But it's actually initialized by the kernel before the program starts.

[marxin]

I don't care much about valgrind, but the test crashes when I run it natively! Please see the very first message execl: Bad address.

[rui314]

Ah, I missed that failure. I fixed it in the above commit. Thank you for your report!