Bug 17644

Summary:

libSM: valgrind warning on _SmcProcessMessage

Product:

xorg

Reporter:

Caolán McNamara <caolanm>

Component:

Lib/other

Assignee:

Xorg Project Team <xorg-team>

Status:

RESOLVED FIXED

QA Contact:

Xorg Project Team <xorg-team>

Severity:

normal

Priority:

medium

CC:

dbaron

Version:

unspecified

Hardware:

All

OS:

Linux (All)

Whiteboard:

i915 platform:

i915 features:

Attachments:

Description	Flags
little patch	none
alternative patch?	none
the valgrind warning in Firefox	none

Description Caolán McNamara 2008-09-18 09:05:41 UTC

Created attachment 18983 [details] [review]
little patch

libSM callocs some memory, at some later stage some malloced memory is splatted on the calloc memory, which makes valgrind worried that some expected-to-be-zero stuff is now random. Its turns out to be a bogus error as the used bytes are correctly set, and only the unused holes are left with uninitialized foo. But its an ugly warning, and the only one left that appears when valgrinding openoffice.org. Maybe there's a possible argument in favour of the patch that memsetting it to 0 might allow some tiny compression advantage over LBX or something

==6986== Syscall param write(buf) points to uninitialised byte(s)
==6986==    at 0x353960DECB: write (in /lib64/libpthread-2.8.90.so)
==6986==    by 0x354560996E: (within /usr/lib64/libICE.so.6.3.0)
==6986==    by 0x354560DBCF: _IceWrite (in /usr/lib64/libICE.so.6.3.0)
==6986==    by 0x354560DCB3: IceFlush (in /usr/lib64/libICE.so.6.3.0)
==6986==    by 0xFE3398B: SessionManagerClient::saveDone() (sm.cxx:393)
==6986==    by 0xFE33BC4: SessionManagerClient::SaveYourselfProc(_SmcConn*, void*, int, int, int, int) (sm.cxx:332)
==6986==    by 0x3545206496: _SmcProcessMessage (in /usr/lib64/libSM.so.6.0.0)
==6986==    by 0x3545612213: IceProcessMessages (in /usr/lib64/libICE.so.6.3.0)
==6986==    by 0xFE334CC: ICEConnectionWorker (sm.cxx:642)
==6986==    by 0x4C3A1BF: (within /usr/lib64/openoffice.org/ure/lib/libuno_sal.so.3)
==6986==    by 0x35396073D9: (within /lib64/libpthread-2.8.90.so)
==6986==    by 0x3538AE9B8C: clone (in /lib64/libc-2.8.90.so)
==6986==  Address 0xa2d7354 is 12 bytes inside a block of size 1,024 alloc'd
==6986==    at 0x4A05174: calloc (vg_replace_malloc.c:397)
==6986==    by 0x3545605E59: IceOpenConnection (in /usr/lib64/libICE.so.6.3.0)
==6986==    by 0x35452027A0: SmcOpenConnection (in /usr/lib64/libSM.so.6.0.0)
==6986==    by 0xFE3423A: SessionManagerClient::open() (sm.cxx:437)
==6986==    by 0xFB9455C: GtkSalFrame::Show(unsigned char, unsigned char) (gtkframe.cxx:1264)
==6986==    by 0x7B89C17: Window::Show(unsigned char, unsigned short) (window.cxx:6515)
==6986==    by 0x7B8993F: Window::Show(unsigned char, unsigned short) (window.cxx:6502)
==6986==    by 0x15591288: (within /usr/lib64/openoffice.org/basis3.0/program/libspllx.so)
==6986==    by 0x50268B0: (within /usr/lib64/openoffice.org/basis3.0/program/libsofficeapp.so)
==6986==    by 0x502A01E: (within /usr/lib64/openoffice.org/basis3.0/program/libsofficeapp.so)
==6986==    by 0x79B6673: ImplSVMain() (svmain.cxx:195)
==6986==    by 0x79B6714: SVMain() (svmain.cxx:236)

Comment 1 Adam Jackson 2008-10-17 14:22:27 UTC

Fixed, thanks!

Comment 2 David Baron 2009-11-27 12:40:22 UTC

Created attachment 31514 [details] [review]
alternative patch?

I was investigating this same problem (in terms of valgrind warnings while starting Firefox), and came up with this (somewhat different) patch that you may or may not consider better; instead of calling memset over the whole buffer, it calls memset on the holes as it creates them.

This approach has the advantage of fixing not just the one caller fixed above, but all callers of STORE_ARRAY8 and STORE_LISTOF_PROPERTY.

Comment 3 David Baron 2009-11-27 12:41:22 UTC

Created attachment 31515 [details]
the valgrind warning in Firefox

Here's the valgrind warning in Firefox; Ubuntu 9.10 still hasn't picked up the previous patch in this bug, so I'm still seeing it.

Comment 4 David Baron 2009-12-23 17:56:21 UTC

I stuck the above patch in bug 25785.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.