64-bit integeer data types and the Linux kernel

64-bit integeer data types and the Linux kernel

[Quentin Barnes]

I've been going over some of my ARM port work of the Systemtap
runtime environment in anticipation of sending it out to this list.
The port work brought up a Linux kernel question I had a few years
ago and never got a satisfactory answer.

I noticed in runtime/arith.c, Systemtap has to supply 64-bit support
arithmetic for some architectures.  (I've added ARM support now.)

Why does Systemtap have to do this?  (Rhetorical question!)  Because
the kernel doesn't always have these missing 64-bit integer routines
for all architectures.  Okay, why doesn't the kernel fully support
64-bit integer types for all architectures?

I never got an answer to that last question when I asked it three
years ago when helping someone port a driver who needed to do a
divide of a 64-bit integer.  Can anyone these days give me the
reason or logic as to why?

If it is because of space concerns, then why can't it be made a
configurable kernel option that Systemtap or anyone else can enable
as need be?  (That's a real question.  Can anyone answer that
question for me?)  Has anyone tried this angle with regards to
Systemtap's needs?

I tried googling to see if I could find an old discussion on this
but got way too many hits and none seemed pertinent.  If someone
could point me to old postings on this issue, I'd appreciate it.

Quentin

Re: 64-bit integeer data types and the Linux kernel

[Martin Hunt]

On Wed, 2007-05-02 at 11:33 -0500, Quentin Barnes wrote:

> Why does Systemtap have to do this?  (Rhetorical question!)  Because
> the kernel doesn't always have these missing 64-bit integer routines
> for all architectures.  Okay, why doesn't the kernel fully support
> 64-bit integer types for all architectures?

There was some discussion about fixing this a month ago or so. 
http://lkml.org/lkml/2007/2/23/321