Bug 508145

Summary: ppc64le needs ld.so hardwire for strcmp
Product: [Developer tools] valgrind Reporter: Mark Wielaard <mark>
Component: memcheckAssignee: Mark Wielaard <mark>
Status: RESOLVED FIXED    
Severity: normal CC: cel
Priority: NOR    
Version First Reported In: 3.25 GIT   
Target Milestone: ---   
Platform: Other   
OS: Linux   
Latest Commit: Version Fixed/Implemented In:
Sentry Crash Report:
Attachments: Proposed hardwire

Description Mark Wielaard 2025-08-12 00:41:18 UTC 
When dlopen is used we might end up in an assembly powerpc/strcmp.S variant that is optimized in a way memcheck cannot proof correct:

==145240== Conditional jump or move depends on uninitialised value(s)
==145240==    at 0x4042590: strcmp (strcmp.S:76)
==145240==    by 0x401BC37: _dl_name_match_p (dl-misc.c:69)
==145240==    by 0x401CA37: dl_open_worker_begin (dl-open.c:553)
==145240==    by 0x4002593: _dl_catch_exception (dl-catch.c:237)
==145240==    by 0x401C5E3: dl_open_worker (dl-open.c:782)
==145240==    by 0x4002593: _dl_catch_exception (dl-catch.c:237)
==145240==    by 0x401DFDB: _dl_open (dl-open.c:903)
==145240==    by 0x443D58F: dlopen_doit (dlopen.c:56)
==145240==    by 0x4002593: _dl_catch_exception (dl-catch.c:237)
==145240==    by 0x400276B: _dl_catch_error (dl-catch.c:256)
==145240==    by 0x443CDF3: _dlerror_run (dlerror.c:138)
==145240==    by 0x443D673: dlopen_implementation (dlopen.c:71)
==145240==    by 0x443D673: dlopen@@GLIBC_2.34 (dlopen.c:81)

We try to intercept strcmp in ld.so, but might fail when strcmp is called before our interception code is loaded.

Having an hardwire for ld.so strcmp (earlier intercept) would solve this.
Comment 1 Mark Wielaard 2025-08-12 00:42:58 UTC 
Created attachment 183979 [details]
Proposed hardwire
Comment 2 Mark Wielaard 2025-08-18 13:34:39 UTC 
Tested on Fedora 43/Rawhide ppc64le setup. Pushed as:

commit 78fe3625f6b8ed4de28527d71c4a98d70e5b3035
Author: Mark Wielaard <mark@klomp.org>
Date:   Mon Aug 18 15:30:47 2025 +0200

    Add ppc64le linux hardwire for ld64.so.2 strcmp
    
    When dlopen is used we might end up in an assembly powerpc/strcmp.S
    variant that is optimized in a way memcheck cannot proof correct. We
    try to intercept strcmp in ld.so, but might fail when strcmp is called
    before our interception code is loaded. Having an hardwire for ld.so
    strcmp (earlier intercept) would solve this.
    
    https://bugs.kde.org/show_bug.cgi?id=508145