408071 - GCC miscompiles NSS when optimizing for size on Linux/ARM

Status: RESOLVED WONTFIX

(NSS :: Libraries, defect)

Description

[Ilpo Ruotsalainen]

User-Agent:       Opera/9.24 (X11; Linux i686; U; en)
Build Identifier: NSS_3_12_ALPHA_2B

The change to use optimize-for-size (bug 389781) in optimized builds causes crash during xulrunner startup on Linux/ARM, likely due to GCC miscompiling something. Only happens when optimizing for size (-Os), changing it back to optimize for speed (-O2) or disabling optimization makes this crash go away.

gcc --version says:
sbox-arm-none-linux-gnueabi-gcc (CodeSourcery Sourcery G++ Lite 2007q3-51) 4.2.1

Reproducible: Always

Comment 1

[Ilpo Ruotsalainen]

Attachment: Backtrace from crash

Comment 2

[Julien Pierre]

gcc bugs should be filed against gcc, not NSS . You should contact the gcc developers to get this fixed. You can already build optimized for speed as before by not setting OPT_CODE_SIZE = 1  in the NSS build , so no NSS change needs to be made to workaround this gcc bug .

Comment 3

[Doug Turner (:dougt)]

Attachment: WORKAROUND

Comment 4

[Nelson Bolyard (seldom reads bugmail)]

Comment on attachment 310804 [details] [diff] [review]
WORKAROUND

Changing the makefile to stop optimizing for code size may be OK for one single platform, but not for all platforms.

Comment 5

[Wan-Teh Chang]

Comment on attachment 310804 [details] [diff] [review]
WORKAROUND

Please try something like this:

DEFAULT_GMAKE_FLAGS += BUILD_OPT=1
ifeq ($(OS_ARCH)_$(TARGET_CPU),Linux_arm)
# Work around a GCC bug (see Bugzilla bug 408071).
DEFAULT_GMAKE_FLAGS += OPT_CODE_SIZE=0
else
DEFAULT_GMAKE_FLAGS += OPT_CODE_SIZE=1
endif

Comment 6

[Doug Turner (:dougt)]

brad, you're using 4.2.x, right?  Is this still a problem for you.

Comment 7

[Nelson Bolyard (seldom reads bugmail)]

If someone wants to submit a patch to change NSS (not PSM) to build without
space optimizations on Linux/Arm platforms ONLY, that would be considered.
Otherwise, this bug is a WONTFIX.  

The best solution, IMO, is to get a working compiler.

Comment 8

[Doug Turner (:dougt)]

wontfix.

Comment 9

[Wan-Teh Chang]

Doug, what is your workaround for this GCC optimization bug?

Comment 10

[Doug Turner (:dougt)]

I am not seeing it.  I asked blassey, and he has not.  we are using CS2007q3 gcc 4.2.