Re: Userspace compiler support of "long long"

Re: Userspace compiler support of "long long"

[Kyle Moffett]

On Jun 27, 2007, at 13:32:40, Adrian Bunk wrote:
> AFAIR the Intel compiler claims to be gcc.
>
> But these are by far not the only C compilers under Linux, and the  
> more important points are:
>
> Is there any userspace Linux compiler that does not support "long  
> long"?

Don't know, but I'd guess not.


> If yes, is there any other way to tell that something is a 64bit  
> int on 32bit architectures?

Not that I know of.  Probably the straight #else conditional is OK.   
We should also merge up the types since *EVERY* linux architecture  
has these same types:

typedef   signed char      __s8;
typedef unsigned char      __u8;
typedef   signed short     __s16;
typedef unsigned short     __u16;
typedef   signed int       __s32;
typedef unsigned int       __u32;

Then all 64-bit archs have:
typedef   signed long      __s64;
typedef unsigned long      __u64;

While all 32-bit archs have:
typedef   signed long long __s64;
typedef unsigned long long __u64;

The only trick is if you care about building 32-bit compat code using  
64-bit linux kernel headers.  In that case we should probably just  
make all archs use "long long" for their 64-bit integers, unless  
there's some platform I'm not remembering where "long long" is 128- 
bits or bigger.  The other benefit is that people could then just use  
the printf format "%llu" for 64-bit integers instead of having to  
conditionalize it all over the place.

I'm working on a patch now.

Cheers,
Kyle Moffett

Re: Userspace compiler support of "long long"

[Randy Dunlap]

On Wed, 27 Jun 2007 18:30:52 -0400 Kyle Moffett wrote:

> On Jun 27, 2007, at 13:32:40, Adrian Bunk wrote:
> > AFAIR the Intel compiler claims to be gcc.
> >
> > But these are by far not the only C compilers under Linux, and the  
> > more important points are:
> >
> > Is there any userspace Linux compiler that does not support "long  
> > long"?
> 
> Don't know, but I'd guess not.
> 
> 
> > If yes, is there any other way to tell that something is a 64bit  
> > int on 32bit architectures?
> 
> Not that I know of.  Probably the straight #else conditional is OK.   
> We should also merge up the types since *EVERY* linux architecture  
> has these same types:
> 
> typedef   signed char      __s8;
> typedef unsigned char      __u8;
> typedef   signed short     __s16;
> typedef unsigned short     __u16;
> typedef   signed int       __s32;
> typedef unsigned int       __u32;
> 
> Then all 64-bit archs have:
> typedef   signed long      __s64;
> typedef unsigned long      __u64;
> 
> While all 32-bit archs have:
> typedef   signed long long __s64;
> typedef unsigned long long __u64;
> 
> The only trick is if you care about building 32-bit compat code using  
> 64-bit linux kernel headers.  In that case we should probably just  
> make all archs use "long long" for their 64-bit integers, unless  
> there's some platform I'm not remembering where "long long" is 128- 
> bits or bigger.  The other benefit is that people could then just use  
> the printf format "%llu" for 64-bit integers instead of having to  
> conditionalize it all over the place.
> 
> I'm working on a patch now.

LDD3 ch. 11 says that long on Sparc64 is 32 bits.
Same for "ppc" (don't know which power* arch. they mean by that).

---
~Randy
*** Remember to use Documentation/SubmitChecklist when testing your code ***

Re: Userspace compiler support of "long long"

[Randy Dunlap]

On Wed, 27 Jun 2007 15:57:15 -0700 Randy Dunlap wrote:

> On Wed, 27 Jun 2007 18:30:52 -0400 Kyle Moffett wrote:
> 
> > On Jun 27, 2007, at 13:32:40, Adrian Bunk wrote:
> > > AFAIR the Intel compiler claims to be gcc.
> > >
> > > But these are by far not the only C compilers under Linux, and the  
> > > more important points are:
> > >
> > > Is there any userspace Linux compiler that does not support "long  
> > > long"?
> > 
> > Don't know, but I'd guess not.
> > 
> > 
> > > If yes, is there any other way to tell that something is a 64bit  
> > > int on 32bit architectures?
> > 
> > Not that I know of.  Probably the straight #else conditional is OK.   
> > We should also merge up the types since *EVERY* linux architecture  
> > has these same types:
> > 
> > typedef   signed char      __s8;
> > typedef unsigned char      __u8;
> > typedef   signed short     __s16;
> > typedef unsigned short     __u16;
> > typedef   signed int       __s32;
> > typedef unsigned int       __u32;
> > 
> > Then all 64-bit archs have:
> > typedef   signed long      __s64;
> > typedef unsigned long      __u64;
> > 
> > While all 32-bit archs have:
> > typedef   signed long long __s64;
> > typedef unsigned long long __u64;
> > 
> > The only trick is if you care about building 32-bit compat code using  
> > 64-bit linux kernel headers.  In that case we should probably just  
> > make all archs use "long long" for their 64-bit integers, unless  
> > there's some platform I'm not remembering where "long long" is 128- 
> > bits or bigger.  The other benefit is that people could then just use  
> > the printf format "%llu" for 64-bit integers instead of having to  
> > conditionalize it all over the place.
> > 
> > I'm working on a patch now.
> 
> LDD3 ch. 11 says that long on Sparc64 is 32 bits.
> Same for "ppc" (don't know which power* arch. they mean by that).

Hm, I suppose that table only applies to userspace, not kernel...

---
~Randy
*** Remember to use Documentation/SubmitChecklist when testing your code ***

Re: Userspace compiler support of "long long"

[Andi Kleen]

On Thursday 28 June 2007 00:30:52 Kyle Moffett wrote:
> On Jun 27, 2007, at 13:32:40, Adrian Bunk wrote:
> > AFAIR the Intel compiler claims to be gcc.
> >
> > But these are by far not the only C compilers under Linux, and the  
> > more important points are:
> >
> > Is there any userspace Linux compiler that does not support "long  
> > long"?
> 
> Don't know, but I'd guess not.

Tendra C and probably lcc. I would guess tinycc too.

> The only trick is if you care about building 32-bit compat code using  
> 64-bit linux kernel headers.  In that case we should probably just  
> make all archs use "long long" for their 64-bit integers, unless  
> there's some platform I'm not remembering where "long long" is 128- 
> bits or bigger.  The other benefit is that people could then just use  
> the printf format "%llu" for 64-bit integers instead of having to  
> conditionalize it all over the place.
> 
> I'm working on a patch now.

Changing this will give you a zillion warnings for printk formats.

-Andi
 

Re: Userspace compiler support of "long long"

[Geert Uytterhoeven]

On Wed, 27 Jun 2007, Randy Dunlap wrote:
> On Wed, 27 Jun 2007 15:57:15 -0700 Randy Dunlap wrote:
> > LDD3 ch. 11 says that long on Sparc64 is 32 bits.
> > Same for "ppc" (don't know which power* arch. they mean by that).
> 
> Hm, I suppose that table only applies to userspace, not kernel...

32-bit userspace?

On 64-bit, `long' is 64-bit on all platforms supported by Linux.

Gr{oetje,eeting}s,

						Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
							    -- Linus Torvalds

Re: Userspace compiler support of "long long"

[Kyle McMartin]

On Wed, Jun 27, 2007 at 04:16:48PM -0700, Randy Dunlap wrote:
> > LDD3 ch. 11 says that long on Sparc64 is 32 bits.
> > Same for "ppc" (don't know which power* arch. they mean by that).
> 
> Hm, I suppose that table only applies to userspace, not kernel...
> 

Doing 64-bit Linux non-LP64 would be an interesting exercise in masochism...

Re: Userspace compiler support of "long long"

[Matthew Wilcox]

On Wed, Jun 27, 2007 at 06:30:52PM -0400, Kyle Moffett wrote:
> Then all 64-bit archs have:
> typedef   signed long      __s64;
> typedef unsigned long      __u64;
> 
> While all 32-bit archs have:
> typedef   signed long long __s64;
> typedef unsigned long long __u64;

include/asm-parisc/types.h:typedef unsigned long long __u64;

For both 32 and 64-bit.

include/asm-sh64/types.h:typedef unsigned long long __u64;
include/asm-x86_64/types.h:typedef unsigned long long  __u64;

So that's three architectures that violate your first assertion.

Re: Userspace compiler support of "long long"

[H. Peter Anvin]

Kyle Moffett wrote:
>> 
> The only trick is if you care about building 32-bit compat code using 
> 64-bit linux kernel headers.  In that case we should probably just make 
> all archs use "long long" for their 64-bit integers, unless there's some 
> platform I'm not remembering where "long long" is 128-bits or bigger.  
> The other benefit is that people could then just use the printf format 
> "%llu" for 64-bit integers instead of having to conditionalize it all 
> over the place.
> 

No, you really don't want to do that, because then u64 != uint64_t on 
those platforms.

	-hpa

Re: Userspace compiler support of "long long"

[Jan Engelhardt]

On Jun 28 2007 04:12, Geert Uytterhoeven wrote:
>On Wed, 27 Jun 2007, Randy Dunlap wrote:
>> On Wed, 27 Jun 2007 15:57:15 -0700 Randy Dunlap wrote:
>> > LDD3 ch. 11 says that long on Sparc64 is 32 bits.
>> > Same for "ppc" (don't know which power* arch. they mean by that).
>> 
>> Hm, I suppose that table only applies to userspace, not kernel...
>
>32-bit userspace?
>
>On 64-bit, `long' is 64-bit on all platforms supported by Linux.

All types are as wide as the compiler makes them.

Compiler  short  int long llong
Turbo C      16   16   32     -
GCC -m32     16   32   32    64
GCC -m64     16   32   64    64



	Jan
-- 

Re: Userspace compiler support of "long long"

[Geert Uytterhoeven]

On Thu, 28 Jun 2007, Jan Engelhardt wrote:
> On Jun 28 2007 04:12, Geert Uytterhoeven wrote:
> >On Wed, 27 Jun 2007, Randy Dunlap wrote:
> >> On Wed, 27 Jun 2007 15:57:15 -0700 Randy Dunlap wrote:
> >> > LDD3 ch. 11 says that long on Sparc64 is 32 bits.
> >> > Same for "ppc" (don't know which power* arch. they mean by that).
> >> 
> >> Hm, I suppose that table only applies to userspace, not kernel...
> >
> >32-bit userspace?
> >
> >On 64-bit, `long' is 64-bit on all platforms supported by Linux.
> 
> All types are as wide as the compiler makes them.
> 
> Compiler  short  int long llong
> Turbo C      16   16   32     -
> GCC -m32     16   32   32    64
> GCC -m64     16   32   64    64

We do not support building Linux with Turbo C (or MS Visual C for Win64
P64).

Gr{oetje,eeting}s,

						Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
							    -- Linus Torvalds

Re: Userspace compiler support of "long long"

[David Woodhouse]

On Thu, 2007-06-28 at 13:34 +0200, Geert Uytterhoeven wrote:
> We do not support building Linux with Turbo C (or MS Visual C for
> Win64 P64). 

We're talking about types which are exposed to userspace.

-- 
dwmw2

Re: Userspace compiler support of "long long"

[Kyle Moffett]

On Jun 27, 2007, at 20:30:42, Andi Kleen wrote:
> On Thursday 28 June 2007 00:30:52 Kyle Moffett wrote:
>> The only trick is if you care about building 32-bit compat code  
>> using 64-bit linux kernel headers.  In that case we should  
>> probably just make all archs use "long long" for their 64-bit  
>> integers, unless there's some platform I'm not remembering where  
>> "long long" is 128-bits or bigger.  The other benefit is that  
>> people could then just use the printf format "%llu" for 64-bit  
>> integers instead of having to conditionalize it all over the place.
>>
>> I'm working on a patch now.
>
> Changing this will give you a zillion warnings for printk formats.

Why?  If this were a problem then we'd be getting a zillion warnings  
*now* from all the 32-bit archs (which already use "long long" for 64- 
bit.  This would actually make it _easier_ to get the printk formats  
right, as you could always use %ull for 64-bit types without having  
to cast for 64-bit platforms.

This is another way to get around the "build 32-bit-compat userspace  
on 64-bit kernel headers" problem:  It tells GCC to use the  
"smallest" available type of the given size, which ends up being  
exactly the types we use now.  On the other hand, it only works for  
GCC which sort of ruins most of the reason for changing the types in  
the first place.

typedef   signed __s8  __attribute__((__mode__(__QI__)));
typedef unsigned __u8  __attribute__((__mode__(__QI__)));
typedef   signed __s16 __attribute__((__mode__(__HI__)));
typedef unsigned __u16 __attribute__((__mode__(__HI__)));
typedef   signed __s32 __attribute__((__mode__(__SI__)));
typedef unsigned __u32 __attribute__((__mode__(__SI__)));
typedef   signed __s64 __attribute__((__mode__(__DI__)));
typedef unsigned __u64 __attribute__((__mode__(__DI__)));

Cheers,
Kyle Moffett

Re: Userspace compiler support of "long long"

[Kyle Moffett]

On Jun 27, 2007, at 23:57:54, Matthew Wilcox wrote:
> On Wed, Jun 27, 2007 at 06:30:52PM -0400, Kyle Moffett wrote:
>> Then all 64-bit archs have:
>> typedef   signed long      __s64;
>> typedef unsigned long      __u64;
>>
>> While all 32-bit archs have:
>> typedef   signed long long __s64;
>> typedef unsigned long long __u64;
>
> include/asm-parisc/types.h:typedef unsigned long long __u64;
>
> For both 32 and 64-bit.
>
> include/asm-sh64/types.h:typedef unsigned long long __u64;
> include/asm-x86_64/types.h:typedef unsigned long long  __u64;
>
> So that's three architectures that violate your first assertion.

Oh, ok, that makes it even easier to say this with certainty:   
Changing the other 64-bit archs to use "long long" for their 64-bit  
numbers will not cause additional warnings.  I'm also almost certain  
there are no architectures which use "long long" for 128-bit  
integers. (Moreover, I can't find hardly anything which does 128-bit  
integers at all).

Cheers,
Kyle Moffett

Re: Userspace compiler support of "long long"

[Jakub Jelinek]

On Thu, Jun 28, 2007 at 07:53:51AM -0400, Kyle Moffett wrote:
> On Jun 27, 2007, at 23:57:54, Matthew Wilcox wrote:
> >On Wed, Jun 27, 2007 at 06:30:52PM -0400, Kyle Moffett wrote:
> >>Then all 64-bit archs have:
> >>typedef   signed long      __s64;
> >>typedef unsigned long      __u64;
> >>
> >>While all 32-bit archs have:
> >>typedef   signed long long __s64;
> >>typedef unsigned long long __u64;
> >
> >include/asm-parisc/types.h:typedef unsigned long long __u64;
> >
> >For both 32 and 64-bit.
> >
> >include/asm-sh64/types.h:typedef unsigned long long __u64;
> >include/asm-x86_64/types.h:typedef unsigned long long  __u64;
> >
> >So that's three architectures that violate your first assertion.
> 
> Oh, ok, that makes it even easier to say this with certainty:   
> Changing the other 64-bit archs to use "long long" for their 64-bit  
> numbers will not cause additional warnings.  I'm also almost certain  
> there are no architectures which use "long long" for 128-bit  
> integers. (Moreover, I can't find hardly anything which does 128-bit  
> integers at all).

unsigned long and unsigned long long have the same size, precision
and alignment on all LP64 arches, that's true.  But they have
different ranks and more importantly they mangle differently in C++.
So, whether some user exposed type uses unsigned long or unsigned long long
is part of the ABI, whether that's size_t, uintptr_t, uint64_t, u_int64_t
or any other type, you can't change it without breaking the ABI.

	Jakub

Re: Userspace compiler support of "long long"

[Kyle Moffett]

On Jun 28, 2007, at 08:08:03, Jakub Jelinek wrote:
> On Thu, Jun 28, 2007 at 07:53:51AM -0400, Kyle Moffett wrote:
>> Oh, ok, that makes it even easier to say this with certainty:  
>> Changing the other 64-bit archs to use "long long" for their 64- 
>> bit numbers will not cause additional warnings.  I'm also almost  
>> certain there are no architectures which use "long long" for 128- 
>> bit integers. (Moreover, I can't find hardly anything which does  
>> 128-bit integers at all).
>
> unsigned long and unsigned long long have the same size, precision  
> and alignment on all LP64 arches, that's true.  But they have  
> different ranks and more importantly they mangle differently in C+ 
> +.  So, whether some user exposed type uses unsigned long or  
> unsigned long long is part of the ABI, whether that's size_t,  
> uintptr_t, uint64_t, u_int64_t or any other type, you can't change  
> it without breaking the ABI.

That sounds *extraordinarily* broken.  Hopefully this would *not*  
affect the type of a function which is passed a C "struct" containing  
the "long long", right?

Hmm, I guess the question is:  Do we support people directly passing  
__u64 to C++ functions in userspace?  I could understand, perhaps,  
passing around structures defined in the kernel headers, but  
certainly not the kernel-internal types.  The only reason we even  
export those is so we can have a private set of bit-size-defined  
types with which to define kernel ABI structures.

Cheers,
Kyle Moffett

Re: Userspace compiler support of "long long"

[Kyle Moffett]

On Jun 28, 2007, at 07:36:14, David Woodhouse wrote:
> On Thu, 2007-06-28 at 13:34 +0200, Geert Uytterhoeven wrote:
>> We do not support building Linux with Turbo C (or MS Visual C for  
>> Win64 P64).
>
> We're talking about types which are exposed to userspace.

Yes, and all 64-bit software built using kernel headers must be built  
in LP64 mode, anything else is pure insanity.  On LP64 (IE: how the  
kernel itself is compiled on EVERY 64-bit arch):

char == 8 bits
short == 16 bits
int == 32 bits
long == 64 bits
pointer == 64 bits
long long == 64 bits

On LP32 (IE: how the kernel itself is compiled on EVERY 32-bit arch):

char == 8 bits
short == 16 bits
int == 32 bits
long == 32 bits
pointer == 32 bits
long long == 64 bits

Ergo we can simply require that if you want to use kernel headers you  
must be using the same mode as the kernel is compiled in (LP32 or LP64).

The simplest guaranteed-not-to-break way to do this on _every_  
supported platform is:
typedef   signed char  __s8;
typedef unsigned char  __s8;
typedef   signed short __s16;
typedef unsigned short __s16;
typedef   signed int   __s32;
typedef unsigned int   __s32;
# if __STDC_VERSION__ >= 19901L
typedef   signed long long __s64;
typedef unsigned long long __s64;
# elif defined(__GNUC__)
__extension__ typedef   signed long long __s64;
__extension__ typedef unsigned long long __s64;
# endif

If you have some other compiler that works under linux *AND* supports  
a 64-bit type in non-C99-mode (whether "long long" or something  
else), then they are welcome to submit patches to fix it.

Cheers,
Kyle Moffett