Hello Heiko,
On 30-11-15 11:03, Heiko Schocher wrote:
Am 30.11.2015 um 10:20 schrieb Jeroen Hofstee:
Hello Heiko,
On 30-11-15 08:47, Heiko Schocher wrote:
compiling U-Boot for openrd_base_defconfig with gcc 5.x shows the following warning:
CC fs/ubifs/super.o In file included from fs/ubifs/ubifs.h:35:0, from fs/ubifs/super.c:37: fs/ubifs/super.c: In function 'atomic_inc': ./arch/arm/include/asm/atomic.h:55:2: warning: 'flags' is used uninitialized in this function [-Wuninitialized] local_irq_save(flags);
Since flags is an "out" argument, something else must be wrong. There should be no need to initialize it, since local_irq_save should do that afaik.
yes, you are right, it should be, but gcc 5.x seems to have problems with it ... compiled code size for the openrd_base config is same with my patch ...
Hmm... for the openrd_base compile local_irq_save() is used from: arch/arm/thumb1/include/asm/proc-armv/system.h
with: static inline void local_irq_save( unsigned long flags __attribute__((unused))) { __asm__ __volatile__ ("" : : : "memory"); }
flasg marked as unused ... seems correct to me, but I have no idea, why gcc 5.x prints a warning ... any ideas?
Well, guessing, order of the passes inside the compiler matter, if the inline is done first, and thereafter the sanity check is done, the latter can't detect an uninitialized variable is passed, since it has already been removed. If you do it the other way around, it will warn first since it won't know the latter pass will remove it. If I have to guess, I think that is what is happening. [ of course it can be made a bit smarter, by e.g. looking at the __attribute__((unused)), but more guessing, I think gcc5 just ignores that one ].
Even more funny is the fact that these functions are not macro's to _prevent_ compiler warnings ;) So cc-ing Albert since he wrote it.
" /* * Redefine all original macros with static inline functions containing * a simple memory barrier, so that they produce the same instruction * ordering constraints as their original counterparts. * We use static inline functions rather than macros so that we can tell * the compiler to not complain about unused arguments. */ "
If we can find a version which doesn't pose other requirements then the linux versions (like flags must be initialized) _and_ they don't warn as well, I guess we can agree that would be preferred.
I am not sure to what extend though, some options below to modify local_irq_save itself to prevent the warning.... 4 and 5 are warning free, not sure it is a brilliant thing to do though .... Albert?
Regards, Jeroen
-------------------------------------------------------------------------------------------- // triggers: "is used uninitialized in this function" with gcc5 static inline void my_argument_maybe_unused1(int flags __attribute__((unused))) { }
// triggers -Wunused-variable #define my_argument_maybe_unused2(flags) do {} while(0)
// triggers set but not used #define my_argument_maybe_unused3(flags) do { flags = 0; } while(0)
// so, lets use flags instead, but clearly unused so it gets optimized out later ;) // no warnings with gcc 5, clang 3.4.... #define my_argument_maybe_unused4(flags) do { int dummy __attribute__((unused)) = sizeof(flags); } while(0)
// combined preprocessor / compiler foodoo static inline void _my_argument_maybe_unused5(int *flags __attribute__((unused))) { } #define my_argument_maybe_unused5(flags) do { _my_argument_maybe_unused5(&flags); } while(0)
int main() { int i_am_used_in_linux_not_in_uboot1; int i_am_used_in_linux_not_in_uboot2; int i_am_used_in_linux_not_in_uboot3; int i_am_used_in_linux_not_in_uboot4; int i_am_used_in_linux_not_in_uboot5;
my_argument_maybe_unused1(i_am_used_in_linux_not_in_uboot1); my_argument_maybe_unused2(i_am_used_in_linux_not_in_uboot2); my_argument_maybe_unused3(i_am_used_in_linux_not_in_uboot3); my_argument_maybe_unused4(i_am_used_in_linux_not_in_uboot4); my_argument_maybe_unused5(i_am_used_in_linux_not_in_uboot5);
return 0; }
/* arm-linux-gnueabihf-gcc (Linaro GCC 5.1-2015.08) 5.1.1 20150608 arm-linux-gnueabihf-gcc -Wall -O2 warnings.c
good, no code generated..
000102c0 <main>: 102c0: 2000 movs r0, #0 102c2: 4770 bx lr
host, clang... guess this fine too, no idea what the nops are though... 00000000004004f0 <main>: 4004f0: 31 c0 xor %eax,%eax 4004f2: c3 retq 4004f3: 66 2e 0f 1f 84 00 00 nopw %cs:0x0(%rax,%rax,1) 4004fa: 00 00 00 4004fd: 0f 1f 00 nopl (%rax)
*/