Re: [PATCH] fix miscompiling with GCC 4.5 -finline-functions

From: Denis Kaganovich
Date: Fri Jan 07 2011 - 20:05:57 EST


On Thu, 6 Jan 2011 13:07:01 -0500, Arnaud Lacombe wrote:
On Thu, Jan 6, 2011 at 12:42 PM, Dzianis Kahanovich
<mahatma@xxxxxxxxxxxxxx> wrote:
Andrew Morton wrote:

Fixing broken automatic inlining for GCC 4.5+ Â(breaks build with
-finline-functions or -O3 on x86_*).


Please always quote the compiler output when addressing build errors
and warnings.

Sorry. linux-next, yesterday, gcc version 4.5.1 (Gentoo 4.5.1-r1 p1.4, pie-0.4.5):

make -j5 -s HOSTCC=x86_64-pc-linux-gnu-gcc ARCH=i386
CROSS_COMPILE=i586-pc-linux-gnu- all -i
arch/x86/kvm/vmx.c: Assembler messages:
arch/x86/kvm/vmx.c:4036: Error: symbol `.Llaunched' is already defined
arch/x86/kvm/vmx.c:4048: Error: symbol `.Lkvm_vmx_return' is already defined
i586-pc-linux-gnu-ld: cannot find arch/x86/kvm/vmx.o: No such file or directory
...
ERROR: "__bad_udelay" [drivers/media/radio/radio-aimslab.ko] undefined!

did you report this to gcc folks ?

Btw, you said it showed with "-O3" or "-finline-functions", but I do
not see any specific option passed to make. Could you remove the "-j5"
switch and add "V=1" to see the full compiler command line ?

Yes, big sorry. Distilling output again I found - there are problem of conjunction of "-finline-function -ftracer". Time ago I report similar problem about -ftracer in gcc bugzilla ("make defconfig && make all" with -ftracer, in various places in various versions).

BUT: First time (weeks ago) "-ftracer" must be first candidate for me to strip. Or I got this error first time without -ftracer or it is some kind of my own bug. Just fyi.

But output anymore ;) for first bug only:

make HOSTCC=x86_64-pc-linux-gnu-gcc ARCH=i386 CROSS_COMPILE=i586-pc-linux-gnu- all
...
i586-pc-linux-gnu-gcc -Wp,-MD,arch/x86/kvm/.vmx.o.d -nostdinc -isystem /usr/lib/gcc/i586-pc-linux-gnu/4.5.1/include -I/var/tmp/portage/sys-kernel/next-sources-9999/work/linux-9999/arch/x86/include -Iinclude -include include/generated/autoconf.h -D__KERNEL__ -Wall -Wundef -Wstrict-prototypes -Wno-trigraphs -fno-strict-aliasing -fno-common -Werror-implicit-function-declaration -Wno-format-security -fno-delete-null-pointer-checks -O2 -finline-functions -ftracer -m32 -msoft-float -mregparm=3 -freg-struct-return -mpreferred-stack-boundary=2 -Wa, -ffreestanding -DCONFIG_AS_CFI=1 -DCONFIG_AS_CFI_SIGNAL_FRAME=1 -DCONFIG_AS_CFI_SECTIONS=1 -pipe -Wno-sign-compare -fno-asynchronous-unwind-tables -mno-sse -mno-mmx -mno-sse2 -mno-3dnow -Wframe-larger-than=2048 -fno-stack-protector -fomit-frame-pointer -Wdeclaration-after-statement -Wno-pointer-sign -fno-strict-overflow -fconserve-stack -DCC_HAVE_ASM_GOTO -Ivirt/kvm -Iarch/x86/kvm -I. -DMODULE -D"KBUILD_STR(s)=#s" -D"KBUILD_BASENAME=KBUILD_STR(vmx)" -D"KBUILD_MODNAME=KBUILD_STR(kvm_intel)" -c -o arch/x86/kvm/vmx.o arch/x86/kvm/vmx.c
arch/x86/kvm/vmx.c: Assembler messages:
arch/x86/kvm/vmx.c:4036: Error: symbol `.Llaunched' is already defined
arch/x86/kvm/vmx.c:4048: Error: symbol `.Lkvm_vmx_return' is already defined



Generally speaking, neither "-O3" or "-finline-functions" seems to be
passed by default by kbuild, so I am not sure they are meant to be
supported at all.

I know, -O3 is "uncommon" subject. Just I see "noinline" wide used, then sometimes sources adopted to auto-inlining. But -ftracer even more ncommon subject? yes & sorry...


- Arnaud

I unsure in precise version boundaries, but first found on release tree weeks
ago and first error only sometimes happened on x86_64 (version related?). On my
own builds I used -fno-inline-functions in local Makefile's, but this is more
local fix.


--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -569,7 +569,7 @@ static inline void ept_sync_individual_a
  Â}
Â}

-static unsigned long vmcs_readl(unsigned long field)
+static noinline unsigned long vmcs_readl(unsigned long field)
Â{
  Âunsigned long value = 0;

--- a/drivers/media/radio/radio-aimslab.c
+++ b/drivers/media/radio/radio-aimslab.c
@@ -71,7 +71,7 @@ static struct rtrack rtrack_card;

Â/* local things */

-static void sleep_delay(long n)
+static noinline void sleep_delay(long n)
Â{
  Â/* Sleep nicely for 'n' uS */
  Âint d = n / msecs_to_jiffies(1000);

A golden rule is that when a programmer reads some code, he should be
able to understand why it's there. ÂThere is no way on this little
earth that a programmer will be able to look at this code and say
"ah-hah, that must be a workaround for gcc-4.5 -finline-functions!".

We fix that problem this way:

--- a/arch/x86/kvm/vmx.c~fix-miscompiling-with-gcc-45-finline-functions-fix
+++ a/arch/x86/kvm/vmx.c
@@ -569,6 +569,7 @@ static inline void ept_sync_individual_a
   }
Â}

+/* noinline works around gcc-4.5+ build error with -finline-functions */
Âstatic noinline unsigned long vmcs_readl(unsigned long field)
Â{
   unsigned long value = 0;
--- a/drivers/media/radio/radio-aimslab.c~fix-miscompiling-with-gcc-45-finline-functions-fix
+++ a/drivers/media/radio/radio-aimslab.c
@@ -71,6 +71,7 @@ static struct rtrack rtrack_card;

Â/* local things */

+/* noinline works around gcc-4.5+ build error with -finline-functions */
Âstatic noinline void sleep_delay(long n)
Â{
   /* Sleep nicely for 'n' uS */
_





--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/