Re: TODO file

[John L. <le...@mo...>]

On Sun, Dec 30, 2001 at 02:40:56AM +0100, Philippe Elie wrote:

> TODO file:
> 
> >> o prepare for rtc:
> >>  - move apic init to module init
> 
> what you means? Apic init are already marked as __init function.

we need to make the init part of the CPU type detection, i.e. so we 
can return CPU_NOAPIC and export that in a sysctl for the GUI etc.
on module load.

> I've not looked implementation of init/exit section but
> lvpt_apic_restore() is marked __exit and is called from
> oprof_init() which is marked as __init. Is it safe ?

Theoretically no ... please remove the __exit (and any others
like that)

> >>  - remove user-space proc type code
> 
> - so op_start can check parameters only after probing the module
>  because op_events need to know the proc type. It is not probably
> a problem but
> 
> - op_help is usable as a standalone programm to see all the events so we
> need to keep the user space proc type code

ok. The point is user-space cannot determine if we have a good local APIC
without help from the module. The user-space detect code can do a "best-guess"
attempt if it can't find /proc/sys/dev/oprofile/cpu_type or whatever, but this
is for op_help only. op_start and oprof_start should load the module first.

> >> o determine where we need ctx_sw_on/off to prevent pre-emption + test
> 
> Hummm, I'm probably unable to make that even if I understand what it
> means.

There might be some points in the code that would break a kernel with the pre-emption
patch applied (look for Robert Love under http://kernelnewbies.org/patches/).

Not sure though, I've been too lazy to even have a cursory look at our code for
this.

> >> o move .oprofile to name=value pairs to avoid conflicts on change.
> 
> There is a "thing" for that in util/persistent.h and an example test in
> oprofile-tests/understanting/persistent-test.cpp, but I don't like the
> output format I use. I hate persistency of C++ object...

I have to admit I've still not looked properly at any of the things you've
been doing recently. I really should. Unfortunately I *have* totally goosed
my distribution it seems - oprof_start still core dumps for me :/

regards
john

-- 
"We're standing there pounding a dead parrot on the counter, and the management
response is to frantically swap in new counters to see if that fixes the
problem."
	- Peter Gutmann