On Tue, Jan 4, 2011 at 1:07 PM, Rajesh S R <srrajesh1989@gmail.com> wrote:

On Tue, Jan 4, 2011 at 11:28 PM, Mulyadi Santosa <mulyadi.santosa@gmail.com> wrote:

...

On Wed, Jan 5, 2011 at 00:40, mohit verma <mohit89mlnc@gmail.com> wrote:

...

hi all, i have seen many places in  kernel where the variables specially the structures should be of  fixed size independent of the architecture. i went through the  definitions of them  but dint  get  clearly (or frankly say ...dint get them even a bit) .

Your question isn't specific enough, so I'll just guess. Let's say "int". In 32 bit, AFAIK it's  4 byte, but in 64 bit (like IA 64, not sure if it's x64) it's 8 byte. So, if you just say "int", you will likely getting screwed up.

By using types like u_int or something like that, you pretty much say "I mean 4 byte kind of integer" etc

Still there can be padding issues due to byte alignment, which may vary across architecture. Am not sure if that is controllable (probably some pragma to gcc?). Probably, OP is asking about it?

I believe you are referring to __attribute__( ( packed ) )

Hi.. On Wed, Jan 5, 2011 at 01:22, mohit verma <mohit89mlnc@gmail.com> wrote:

that is it mulyadi. but how the compiler or kernel forces the things to get initialized in that fixed size only independent of the architecture??

well, AFAIK by mapping that new type into native one...for example, let's say I have "u_int", which in turn when this code is compiled for x86 32 bit, it is a typedef of "int". -- regards, Mulyadi Santosa Freelance Linux trainer and consultant blog: the-hydra.blogspot.com training: mulyaditraining.blogspot.com

Linux (compiler actually) supports C99 fixed-width types such as 8, 16, 32, 64 bits. Just look through the include/linux/types.h and small example here: http://lxr.linux.no/#linux+v2.6.36/arch/powerpc/boot/types.h#L8

...

while writing portable applications always remember that "unsigned long" is the size of pointer and not necessarily unsigned int. For example on x86_64 sizeof(unsigned int) != sizeof(void *). -syed

A nice kernel document regarding unaligned memory access. It may not be directly answering all the questions asked, but once gone through and understood completely, it will become easy to figure out how to write portable kernel code. http://lxr.linux.no/#linux+v2.6.36/Documentation/unaligned-memory-access.txt Rajat On Wed, Jan 5, 2011 at 4:29 AM, julie Sullivan <kernelmail.jms@gmail.com> wrote:

...

i have seen many places in  kernel where the variables specially the structures should be of  fixed size independent of the architecture. i went through the  definitions of them  but dint  get  clearly (or frankly  say ...dint get them even a bit) .

Hi Mohit

I'm not sure whether we are interpreting your question correctly. Do you mean

1. you've seen some code in the kernel which you think means the size of a structure/ variable (and its resulting binary footprint) is set to be the same (in bytes), regardless of the architecture, and you are confused about it?

2. you think that there should be a way of fixing the structure/variable (binary footprint) size to be the same (in bytes) regardless of the architecture and you are wondering if this is possible?

In my (uninformed) opinion (2) is not be possible with the kernel due to portability issues - not only do natural word types differ (as others here are explaining) but you have no control over what optimization settings the kernel's user might set in gcc, for example. This is one of the problems with trying to maintain closed-source drivers and other binary code for the kernel, as I understand.

Thanks Julie

_______________________________________________ Kernelnewbies mailing list Kernelnewbies@kernelnewbies.org http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies