Hi,
I have tabulated lists of static and global arrays in the
2.4.16 kernel. I am posting the information here in case it may be
of interest to some of you. I would recommend starting with the
linux/kernel and linux/fs tables, as these are the most complete.
http://lse.sourceforge.net/resource/#staticarray
Background:
-----------
These tables were created as part of a larger effort to identify
static resource limits which could perhaps be made dynamic or
configurable. The notes column should eventually contain the
following information.
(1) A _very_ brief description of the array
(2) Any static limits associated with resource
allocation should be noted.
See the linux/kernel and linux/fs tables for examples of these notes.
Moving Forward:
---------------
Many of the arrays in these tables have not yet been annotated.
For me this has been very educational - lots of unfamiliar source
code to read. For many of you, it would be simply a matter of noting
things you already understand. If you find this information useful
and feel inclined to contribute, I would welcome contributions and/or
modifications to the notes in these tables.
- Russ
--
Russ Weight ([email protected])
Linux Technology Center
On Fri, 14 Dec 2001 16:31:10 -0800,
Russ Weight <[email protected]> wrote:
> I have tabulated lists of static and global arrays in the
>2.4.16 kernel. I am posting the information here in case it may be
>of interest to some of you. I would recommend starting with the
>linux/kernel and linux/fs tables, as these are the most complete.
>
> http://lse.sourceforge.net/resource/#staticarray
Stating the obvious: For any array with dimension [32] or [64], check
if the source code uses [NR_CPUS]. IMHO there is no point in trying to
make those arrays dynamic in size, you penalize the 99% of small
machines on the off chance that somebody is going to run single system
image Linux on a box with > 32/64 processors. People using such large
machines can recompile the kernel with a new value of NR_CPUS.
As a separate problem, there are still places in the kernel which
assume the number of cpus can be represented in a bitmap of type long.
That code would be worth tracking down and changing to remove the
assumption that NR_CPUS <= 8*sizeof(long). Until that is done, you
cannot run SSI Linux on machines with > 32/64 processors.
Note to self: after the bitmap code limit has been removed, make
NR_CPUS a config option to get ready for huge machines, make
CONFIG_NR_CPUS a critical config option.