John Levon wrote:
> On Fri, Mar 14, 2008 at 08:49:43AM -0700, Bart Smaalders wrote:
> 
>>> Has someone thought through the implications of the IPS approach on
>>> the CTF parts of the kernel build? (See
>>> http://blogs.sun.com/levon/entry/reducing_ctf_overhead if you need
>>> context)
>>> For the sake of kmdb and crash dumps, the CTF data for each kernel
>>> binary is present in the memory image of a booted kernel. 
>> Since the ctf data is present in the in memory portion of the elf files,
>> it is part of the elf hash and thus any file w/ varying cft data should be
>> part of the new package(s).
> 
> This is not the problem I'm raising.
> 
>> There are distinct implications of CTF on the number of binaries changed
>> in each build; this will need to be explored.
> 
> Well, it's a little more fundamental than that - without *some* labelling,
> every kernel change (that is, every new pkg revision of genunix) means
> every kernel module is rebuilt and re-issued too.
> 

There are a lot of kernel changes that don't affect genunix.

> The current system has been a significant pain point for the current S10
> leads - and it's not even being used correctly. There are a number of
> ways we can rethink things, but this is one of those rare things where
> the packaging system impacts directly on the build, so we do need to
> deal with this wrt IPS.
> 

The packaging system doesn't affect the build here; it's the other way 
around.
Are you saying that even though two builds produce identical versions of a
kernel module, we must always take the later one?

>> If necessary, we may need to work on making the CTF construction
>> process more deterministic.
> 
> This is extremely difficult for the needs we have in this case, if not
> impossible.

So what you're implying is that we have to replace all kernel binaries
every time we relink?  Or that if I build the same ON software 10 times,
I'll end up with 10 different versions of all the kernel modules which
are not interchangable?

- Bart


-- 
Bart Smaalders                  Solaris Kernel Performance
[EMAIL PROTECTED]               http://blogs.sun.com/barts
"You will contribute more with mercurial than with thunderbird."
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to