[OE-core] [PATCH] glibc: add -fno-builtin-strlen when not using -O2

[Andre McCurdy]

On Sun, Dec 11, 2016 at 11:41 PM, Huang, Jie (Jackie)
<Jackie.Huang at windriver.com> wrote:
>> -----Original Message-----
>> From: Khem Raj [mailto:raj.khem at gmail.com]
>> Sent: Monday, December 12, 2016 2:37 PM
>> To: Huang, Jie (Jackie)
>> Cc: Patches and discussions about the oe-core layer
>> Subject: Re: [OE-core] [PATCH] glibc: add -fno-builtin-strlen when not using -O2
>>
>> On Sun, Dec 11, 2016 at 9:42 PM,  <jackie.huang at windriver.com> wrote:
>> > From: Jackie Huang <jackie.huang at windriver.com>
>> >
>> > The strlen will be inlined when compile with -O, -O1 or -Os,
>> > so there is no symbol for strlen in ld-linux-x86-64.so.2,
>> > causing a fatal error in valgrind:
>> >
>> > valgrind: Fatal error at startup: a function redirection
>> > valgrind: which is mandatory for this platform-tool combination
>> > valgrind: cannot be set up. Details of the redirection are:
>> > valgrind:
>> > valgrind: A must-be-redirected function
>> > valgrind: whose name matches the pattern: strlen
>> > valgrind: in an object with soname matching: ld-linux-x86-64.so.2
>> >
>> > so add -fno-builtin-strlen when compile with -O, -O1 or -Os.
>>
>> This is a bug in valgrind, I read the same on forums. KDE has proposed a fix
>> https://bugsfiles.kde.org/attachment.cgi?id=82043
>> can you check if this fixes the issue
>
> Thanks, I will check that and change to patch for valgrind if it works.

For reference, here's the patch I've been using. It's a slightly more
generic fix than the one in the KDE bug report.

diff --git a/coregrind/m_redir.c b/coregrind/m_redir.c
index 7e4df8d..640a346 100644
--- a/coregrind/m_redir.c
+++ b/coregrind/m_redir.c
@@ -1220,7 +1220,18 @@ static void add_hardwired_spec (const  HChar*
sopatt, const HChar* fnpatt,
    spec->from_fnpatt = CONST_CAST(HChar *,fnpatt);
    spec->to_addr     = to_addr;
    spec->isWrap      = False;
-   spec->mandatory   = mandatory;
+
+   /* Hack: Depending on how glibc was compiled (e.g. optimised for size or
+      built with _FORTIFY_SOURCE enabled) the strlen symbol might not be found.
+      Therefore although we should still try to intercept it, don't make it
+      mandatory to do so. We over-ride "mandatory" here to avoid the need to
+      patch the many different architecture specific callers to
+      add_hardwired_spec(). */
+   if (0==VG_(strcmp)("strlen", fnpatt))
+      spec->mandatory = NULL;
+   else
+      spec->mandatory = mandatory;
+
    /* VARIABLE PARTS */
    spec->mark        = False; /* not significant */
    spec->done        = False; /* not significant */