(Sorry forgot to the patch itself.)
diff -cibwr system.h.saved system.h
*** system.h.saved Wed Jun 25 01:24:32 2003
--- system.h Wed Jun 25 01:26:11 2003
***************
*** 187,193 ****
static inline void __set_64bit_var (unsigned long long *ptr,
unsigned long long value)
{
! __set_64bit(ptr,ll_low(value), ll_high(value));
}
#define set_64bit(ptr,value) \
--- 187,198 ----
static inline void __set_64bit_var (unsigned long long *ptr,
unsigned long long value)
{
! union {
! unsigned long long ull;
! unsigned int ui2[2];
! } t;
! t.ull = value;
! __set_64bit(ptr,t.ui2[0], t.ui2[1]);
}
#define set_64bit(ptr,value) \
Patch to (x86) asm/system.h to suppress warning by GCC 3.3.
I would like to suggest a patch to x86 asm/system.h of the kernel
2.4.21. Patch is attached at the end.
GCC version.
gcc -v
Reading specs from /usr/lib/gcc-lib/i386-linux/3.3/specs
Configured with: ../src/configure -v --enable-languages=c,c++,java,f77,pascal,objc,ada,treelang --prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info --with-gxx-include-dir=/usr/include/c++/3.3 --enable-shared --with-system-zlib --enable-nls --without-included-gettext --enable-__cxa_atexit --enable-clocale=gnu --enable-debug --enable-java-gc=boehm --enable-java-awt=xlib --enable-objc-gc i386-linux
Thread model: posix
gcc version 3.3 (Debian)
Background:
linux kernel 2.4.21 asm/system.h header produces warnings when
compiled using GCC 3.3 (Debian).
For example, when I compile lm_sensors package
( See the following URL for lm_sensors
http://www2.lm-sensors.nu/~lm78/index.html )
I get many warnings similar to the following.
--- quote from compilation ---
gcc -D__KERNEL__ -DMODULE -DEXPORT_SYMTAB -fomit-frame-pointer -I. -Ikernel/include -I/usr/local/include -I/lib/modules/2.4.21/build/include -nostdinc -I /usr/lib/gcc-lib/i386-linux/3.3/include -Wall -O2 -c kernel/busses/i2c-nforce2.c -o kernel/busses/i2c-nforce2.o
In file included from /lib/modules/2.4.21/build/include/asm/semaphore.h:39,
from /lib/modules/2.4.21/build/include/linux/fs.h:200,
from /lib/modules/2.4.21/build/include/linux/capability.h:17,
from /lib/modules/2.4.21/build/include/linux/binfmts.h:5,
from /lib/modules/2.4.21/build/include/linux/sched.h:9,
from /lib/modules/2.4.21/build/include/linux/mm.h:4,
from /lib/modules/2.4.21/build/include/linux/slab.h:14,
from /lib/modules/2.4.21/build/include/asm/pci.h:32,
from /lib/modules/2.4.21/build/include/linux/pci.h:654,
from kernel/busses/i2c-nforce2.c:35:
/lib/modules/2.4.21/build/include/asm/system.h: In function `__set_64bit_var':
/lib/modules/2.4.21/build/include/asm/system.h:190: warning: dereferencing type-punned pointer will break strict-aliasing rules
/lib/modules/2.4.21/build/include/asm/system.h:190: warning: dereferencing type-punned pointer will break strict-aliasing rules
kernel/busses/i2c-nforce2.c: In function `nforce2_access':
kernel/busses/i2c-nforce2.c:148: warning: `len' might be used uninitialized in this function
--- end quote.
Note the warning message "dereferencing type-punned pointer will break
strict-aliasing rules".
>From (x86) asm/system.h, we find the following code on line 190
#define ll_low(x) *(((unsigned int*)&(x))+0)
#define ll_high(x) *(((unsigned int*)&(x))+1)
static inline void __set_64bit_var (unsigned long long *ptr,
unsigned long long value)
{
---> __set_64bit(ptr,ll_low(value), ll_high(value));
}
We see that the "unsigned long long" value is handled using ll_low()
and ll_high() macros, which indeed play loose with pointers.
My suggested patch is basically as follows.
static inline void __set_64bit_var (unsigned long long *ptr,
unsigned long long value)
{
union {
unsigned long long ull;
unsigned int ui2[2];
} t;
t.ull = value;
__set_64bit(ptr,t.ui2[0], t.ui2[1]);
}
Verification:
I checked that the generated assembly language code
for my modification is not that bad.
Indeed the modified code generates a little cleaner output
than the current C code does IMHO.
(Or am I missing something? Maybe )
The modified code also works correctly as far as I can see.
So I suggest that we modify the asm/system.h (for x86)
(Maybe someone can offer better fix which may not
break hidden assumptions/requirements missed by me.
In any case, suppressing warnings from a system-supplied
header ought to be a priority.)
Testing was done by the following C code.
---
#include <stdio.h>
#include <stdlib.h>
/* The following inlines were taken straight from
asm/system.h
*/
static inline void __set_64bit (unsigned long long * ptr,
unsigned int low, unsigned int high)
{
__asm__ __volatile__ (
"\n1:\t"
"movl (%0), %%eax\n\t"
"movl 4(%0), %%edx\n\t"
"lock cmpxchg8b (%0)\n\t"
"jnz 1b"
: /* no outputs */
: "D"(ptr),
"b"(low),
"c"(high)
: "ax","dx","memory");
}
static inline void __set_64bit_constant (unsigned long long *ptr,
unsigned long long value)
{
__set_64bit(ptr,(unsigned int)(value), (unsigned int)((value)>>32ULL));
}
#define ll_low(x) *(((unsigned int*)&(x))+0)
#define ll_high(x) *(((unsigned int*)&(x))+1)
static inline void __set_64bit_var (unsigned long long *ptr,
unsigned long long value)
{
__set_64bit(ptr,ll_low(value), ll_high(value));
}
/* Here is my modification.
Does not produce warning from the compiler GCC 3.3,
and seem to generate cleaner code.
*/
static inline void __set_64bit_var_2 (unsigned long long *ptr,
unsigned long long value)
{
union {
unsigned long long ull;
unsigned int ui2[2];
} t;
t.ull = value;
__set_64bit(ptr,t.ui2[0], t.ui2[1]);
}
unsigned long long g_ull;
unsigned long long *g_ullptr;
int
test2()
{
__set_64bit_var(g_ullptr, g_ull);
return 0;
}
int test3()
{
__set_64bit_var_2(g_ullptr, g_ull);
return 0;
}
int
main()
{
unsigned long long t;
g_ullptr = &t;
g_ull = 0x1100220033004400ull;
test2();
if(t != g_ull)
abort();
g_ull = 0x0011002200330044ull;
test3();
if( t != g_ull )
abort();
exit(0);
}
---
gcc -Wall -O2 -S test-macro.c
test-macro.c: In function `__set_64bit_var':
test-macro.c:35: warning: dereferencing type-punned pointer will break strict-aliasing rules
test-macro.c:35: warning: dereferencing type-punned pointer will break strict-aliasing rules
cat test-macro.s
.file "test-macro.c"
.text
.p2align 2,,3
.globl test2
.type test2, @function
test2:
pushl %ebp
movl %esp, %ebp
pushl %edi
pushl %ebx
subl $8, %esp
movl g_ull, %eax
movl g_ull+4, %edx
movl %eax, -16(%ebp)
movl g_ullptr, %edi
movl %edx, -12(%ebp)
movl -16(%ebp), %ebx
movl -12(%ebp), %ecx
#APP
1: movl (%edi), %eax
movl 4(%edi), %edx
lock cmpxchg8b (%edi)
jnz 1b
#NO_APP
addl $8, %esp
popl %ebx
xorl %eax, %eax
popl %edi
leave
ret
.size test2, .-test2
.p2align 2,,3
.globl test3
.type test3, @function
test3:
pushl %ebp
movl %esp, %ebp
pushl %edi
pushl %ebx
movl g_ullptr, %edi
movl g_ull, %ebx
movl g_ull+4, %ecx
#APP
1: movl (%edi), %eax
movl 4(%edi), %edx
lock cmpxchg8b (%edi)
jnz 1b
#NO_APP
popl %ebx
xorl %eax, %eax
popl %edi
leave
ret
.size test3, .-test3
.p2align 2,,3
.globl main
.type main, @function
main:
pushl %ebp
movl %esp, %ebp
pushl %esi
pushl %ebx
subl $16, %esp
andl $-16, %esp
leal -16(%ebp), %eax
movl %eax, g_ullptr
movl $855655424, g_ull
movl $285221376, g_ull+4
call test2
movl -12(%ebp), %ebx
movl -16(%ebp), %ecx
movl g_ull, %eax
movl g_ull+4, %edx
movl %ebx, %esi
xorl %edx, %esi
xorl %ecx, %eax
orl %eax, %esi
jne .L10
movl $3342404, g_ull
movl $1114146, g_ull+4
call test3
movl -12(%ebp), %ebx
movl -16(%ebp), %ecx
movl g_ull, %eax
movl g_ull+4, %edx
movl %ebx, %esi
xorl %edx, %esi
xorl %ecx, %eax
orl %eax, %esi
je .L9
.L10:
call abort
.L9:
subl $12, %esp
pushl $0
call exit
.size main, .-main
.comm g_ull,8,8
.comm g_ullptr,4,4
.ident "GCC: (GNU) 3.3 (Debian)"
---
Run the program.
gcc -Wall -O2 test-macro.c
test-macro.c: In function `__set_64bit_var':
test-macro.c:35: warning: dereferencing type-punned pointer will break strict-aliasing rules
test-macro.c:35: warning: dereferencing type-punned pointer will break strict-aliasing rules
./a.out
(no error.)
PS: i am not on linux-kernel and I would appreciate it if
follow-up is cc: to me as well.
PPS: I wondered why linux kernel compilatio itsefl didn't produce
the warning and found the use of -fno-strict-aliasing on gcc command
line. Hmm...
PPPS: In any case, the code produced seem to be cleaner (and possibly
faster. Will someone count the cycles?) and so the patch seems
preferable.
[end of memo]