Re: Rusty's module talk at the Kernel Summit

[Rogier Wolff]

Adam J. Richter wrote:
> >The total saving over all 2.5.24 modules is 4% of the total module
> >sizes, rounded to page boundaries.
> 
>       As individual space optimizations go, 4% is respectable,
> especially for something that has no cost, helps detect bugs and
> simplifies the kernel.  It is hard to think of many potential
> other space improvements that would as effective, especially as
> function of implementation effort.  In comparison, my vmlinux is
> 5% init sections.  So, if init sections are worth it for the
> core kernel, they should be worth it for modules.

Ehmmm. You normally load one big 1Mb kernel, freeing about 40 or 50k
at init time.

You normally load a couple of modules, totalling much less. 

Hmm. Just checked on a system with sound as modules, I see half a
megabyte of modules. So maybe that 20k is worth it. On the other hand,
you only load half a megabyte of shit if you have the RAM to spare.
20k is not worth the time I spend typing this....

> >Most of that saving comes from a few modules.
> 
> 	This makes me wonder if __init procedures are not being
> aggressively identified.  I wonder if people would use __init a little
> more if they knew they would get the benefit of it in the module case.
> Perhaps someday someone will write a tool to identify procedures that
> are only called from init sections.

Sometimes the "error path" will try to reset/reinit the chip. You will
not see that happening during a normal usage cycle, but you will get
bitten if you remove the init based on an actual call-trace.... 

> 	Kernel modules have been a way of life for me for years, and I
> don't think I've ever caught a kernel bug by the mechanism that you

This happens often enough "during development" that the bugs get fixed
before you get to see them....

> describe.  However, I see no harm in having a debugging option that
> always vmalloc'ed kernel modules.  This faciilty could be entirely
> configuarable from user level by having insmod allocate a module of
> *exactly* one page for modules that were less than a page (since you
> would only want to kmalloc modules that were *less* than one page).

As far as I know, kmallocing more than half a page will actually
allocate the whole page. 

			Roger. 

-- 
** [e-mail] ** http://www.BitWizard.nl/ ** [phone] **
*-- BitWizard writes Linux device drivers for any device you may have! --*
* There are old pilots, and there are bold pilots. 
* There are also old, bald pilots.