Skip to contentSkip to conversation

Cleanup possibility in asm-i386/string.h

DISCUSSION Andi Kleen 4 participants 12 messages Feb 7, 2006 – Feb 10, 2006

lore
Share & tools
lore

Andi Kleen

author
#0 lore raw

Adrian,

If you feel the need to remove some more code: Now that gcc 2.95 isn't supported anymore there isn't really a need to keep the handwritten inline string functions in asm-i386/string.h around. Just declaring them as normal externs will cause gcc to use its builtin expansions, which are typically better than these old inline functions with inline assembly.

For out of line the C versions in lib/string.c can be used (by not setting __ARCH_*) x86-64 did it like this forever and I guess it would be valuable cleanup for i386 too.

-Andi

#1 lore raw

Hi,

Quoted 8 lines · Andi Kleen

If you feel the need to remove some more code: Now that gcc 2.95 isn't supported anymore there isn't really a need to keep the handwritten inline string functions in asm-i386/string.h around. Just declaring them as normal externs will cause gcc to use its builtin expansions, which are typically better than these old inline functions with inline assembly.

For out of line the C versions in lib/string.c can be used (by not setting __ARCH_*) x86-64 did it like this forever and I guess it would be valuable cleanup for i386 too.

The only problem is that we compile with -ffreestanding which implies -fno-builtin, so just declaring them as normal externs is not enough and you have to something like this:

#define __HAVE_ARCH_MEMSET
extern void *memset(void *, int, __kernel_size_t); #define memset(d, c, n) __builtin_memset(d, c, n)

(BTW you do this already in x86-64.)

Another problem here is because of -fno-builtin it's not easy to use the generic functions as fallback. x86-64 basically does this:

#define strlen __builtin_strlen
size_t strlen(const char * s);

#ifndef __HAVE_ARCH_STRLEN
extern __kernel_size_t strlen(const char *);
#endif

This means you define a prototype for the builtin function and not for the normal function. I'm not sure this is really intended.

bye, Roman

author
#2 lore raw
Quoted 17 lines · Roman Zippel

Hi,

Quoted 8 lines · Andi Kleen

If you feel the need to remove some more code: Now that gcc 2.95 isn't supported anymore there isn't really a need to keep the handwritten inline string functions in asm-i386/string.h around. Just declaring them as normal externs will cause gcc to use its builtin expansions, which are typically better than these old inline functions with inline assembly.

For out of line the C versions in lib/string.c can be used (by not setting __ARCH_*) x86-64 did it like this forever and I guess it would be valuable cleanup for i386 too.

The only problem is that we compile with -ffreestanding which implies -fno-builtin, so just declaring them as normal externs is not enough and you have to something like this:

Quoted 5 lines

#define __HAVE_ARCH_MEMSET
extern void *memset(void *, int, __kernel_size_t); #define memset(d, c, n) __builtin_memset(d, c, n)

(BTW you do this already in x86-64.)

Yes thinking about it the x86-64 string.h could just be copied over to i386. It should already do everything correctly and I don't think there is anything really 64bit specific in there.

Quoted 6 lines

Another problem here is because of -fno-builtin it's not easy to use the generic functions as fallback. x86-64 basically does this:

#define strlen __builtin_strlen
size_t strlen(const char * s);

Quoted 6 lines

#ifndef __HAVE_ARCH_STRLEN
extern __kernel_size_t strlen(const char *);
#endif

This means you define a prototype for the builtin function and not for the normal function. I'm not sure this is really intended.

What good would be a prototype for a symbol that is defined to a different symbol?

-Andi

#3 lore raw

Hi,

Quoted 4 lines · Andi Kleen
Quoted 2 lines

This means you define a prototype for the builtin function and not for the normal function. I'm not sure this is really intended.

What good would be a prototype for a symbol that is defined to a different symbol?

The point is you define a prototype for a builtin function, I'm not sure that's a good thing to do.
Actually I'd prefer to remove -ffreestanding again, especially because it disables builtin functions, which we have to painfully enable all again one by one, instead of leaving it just to gcc.

bye, Roman

#4 lore raw
Quoted 7 lines · Roman Zippel

Hi,

Quoted 3 lines · Andi Kleen
Quoted 2 lines

This means you define a prototype for the builtin function and not for the normal function. I'm not sure this is really intended.

Quoted 7 lines
Quoted 1 line

What good would be a prototype for a symbol that is defined to a different symbol?

The point is you define a prototype for a builtin function, I'm not sure that's a good thing to do.
Actually I'd prefer to remove -ffreestanding again, especially because it disables builtin functions, which we have to painfully enable all again one by one, instead of leaving it just to gcc.

I remember playing with using more gcc builtins in the kernel some time ago, and some gcc builtin used a different library function, which was a function the kernel did not supply.

I don't remember the exact details, but this was the reason why I preferred using builtins only when explicitely enabled.

Quoted 1 line

bye, Roman

cu
Adrian

Signature · 4 lines
       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed
author
#5 lore raw
Quoted 16 lines · Adrian Bunk
Quoted 7 lines · Roman Zippel

Hi,

Quoted 3 lines · Andi Kleen
Quoted 2 lines

This means you define a prototype for the builtin function and not for the normal function. I'm not sure this is really intended.

Quoted 7 lines
Quoted 1 line

What good would be a prototype for a symbol that is defined to a different symbol?

The point is you define a prototype for a builtin function, I'm not sure that's a good thing to do.
Actually I'd prefer to remove -ffreestanding again, especially because it disables builtin functions, which we have to painfully enable all again one by one, instead of leaving it just to gcc.

Quoted 3 lines

I remember playing with using more gcc builtins in the kernel some time ago, and some gcc builtin used a different library function, which was a function the kernel did not supply.

It works fine on x86-64. If something is missing it can be also supplied.

-Andi

#6 lore raw
Quoted 10 lines · Andi Kleen
Quoted 9 lines · Adrian Bunk
Quoted 8 lines · Roman Zippel

Hi,

Quoted 3 lines · Andi Kleen
Quoted 1 line

This means you define a prototype for the builtin function and not for the normal function. I'm not sure this is really intended.

What good would be a prototype for a symbol that is defined to a different symbol?

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

The point is you define a prototype for a builtin function, I'm not sure that's a good thing to do. Actually I'd prefer to remove -ffreestanding again, especially because it disables builtin
functions, which we have to painfully enable all again one by one, instead of leaving it just to gcc.

I remember playing with using more gcc builtins in the kernel some time ago, and some gcc builtin used a different library function, which was a function the kernel did not supply.

It works fine on x86-64. If something is missing it can be also supplied.

I don't remember exactly, but I think the problem was something like this (even if not this exact case, it was similarly obscure): If -ffreestanding was not specified, then the following code would generate an implicit call to memcpy() or some other library function.

struct a {
[... large struct with lots of fields ...]
}

struct a first = { ..... };
[... more code ...];
struct a second = first; /* <== This line would generate implicit memcpy */

Cheers,
Kyle Moffett

--
Diplomacy involves walking softly and _carrying_ a big stick. Actually using the big stick means the diplomacy part failed.
-- Rob Landley

#7 lore raw
In reply to Andi Kleen

Hi,

Quoted 5 lines · Andi Kleen
Quoted 3 lines

I remember playing with using more gcc builtins in the kernel some time ago, and some gcc builtin used a different library function, which was a function the kernel did not supply.

It works fine on x86-64. If something is missing it can be also supplied.

I think I now see what the real problem was, x86-64 does:

#define strcpy __builtin_strcpy

which also renames the version in lib/string.c, so x86-64 never had a fallback copy for __builtin_sprintf.
Can we please get rid of -freestanding and fix x86-64 instead?

bye, Roman

author
#8 lore raw
Quoted 10 lines · Roman Zippel

Hi,

Quoted 5 lines · Andi Kleen
Quoted 3 lines

I remember playing with using more gcc builtins in the kernel some time ago, and some gcc builtin used a different library function, which was a function the kernel did not supply.

It works fine on x86-64. If something is missing it can be also supplied.

Quoted 7 lines

I think I now see what the real problem was, x86-64 does:

#define strcpy __builtin_strcpy

which also renames the version in lib/string.c, so x86-64 never had a fallback copy for __builtin_sprintf.
Can we please get rid of -freestanding and fix x86-64 instead?

Ok I can fix that. Just removing the defines should be ok i guess (afaik gcc detects them automatically as the builtin)

I don't know if the freestanding in the main Makefile isn't needed for other architectures so I won't touch it right now.

-Andi

#9 lore raw

Hi,

Quoted 8 lines · Andi Kleen
Quoted 5 lines

#define strcpy __builtin_strcpy

which also renames the version in lib/string.c, so x86-64 never had a fallback copy for __builtin_sprintf.
Can we please get rid of -freestanding and fix x86-64 instead?

Ok I can fix that. Just removing the defines should be ok i guess (afaik gcc detects them automatically as the builtin)

Not with -freestanding.

Quoted 2 lines

I don't know if the freestanding in the main Makefile isn't needed for other architectures so I won't touch it right now.

Well, it was added for x86-64...
It shouldn't break anything, that isn't already broken.

bye, Roman

author
#10 lore raw
Quoted 6 lines · Roman Zippel

Hi,

Quoted 2 lines · Andi Kleen
Quoted 2 lines

#define strcpy __builtin_strcpy

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

which also renames the version in lib/string.c, so x86-64 never had a fallback copy for __builtin_sprintf.
Can we please get rid of -freestanding and fix x86-64 instead?

Ok I can fix that. Just removing the defines should be ok i guess (afaik gcc detects them automatically as the builtin)

Not with -freestanding.

True.

Quoted 6 lines
Quoted 3 lines

I don't know if the freestanding in the main Makefile isn't needed for other architectures so I won't touch it right now.

Well, it was added for x86-64...
It shouldn't break anything, that isn't already broken.

Ok i will remove it then.

Thanks,
-Andi

#11 lore raw
In reply to Adrian Bunk

Hi,

Quoted 6 lines · Adrian Bunk

I remember playing with using more gcc builtins in the kernel some time ago, and some gcc builtin used a different library function, which was a function the kernel did not supply.

I don't remember the exact details, but this was the reason why I preferred using builtins only when explicitely enabled.

gcc can turn a sprintf() into strcpy(), which is a valid thing to do even in a kernel environment. It can be turned off selectively, if it should be a problem, so using -free-standing is IMO complete overkill.

bye, Roman