This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH 02/10] Add the low level infrastructure for pthreads lockelision with TSX
On 01/13/2013 02:41 PM, Andi Kleen wrote:
>> We need a framework and policy for tunables.
>>
>> I'll try to work something up and send it out to the list for discussion.
>
> AFAIK there are two choices: environment variable(s) or config file.
>
> config file is nicer, but harder to use (it's very convenient to set per
> app for testing). However what would work is also having an environment
> variable to override the config file.
>
> config file would require one more syscall per program startup to check
> if the file is there (or 3+ if it's actually there), so it's slower.
For tunables I want to avoid configuration files at all costs.
The process *should* start with everything it needs to start running.
I'm going to be suggesting we should expose a generic API and
environment variables.
>> Again, this is totally not your fault Andi, you've just touched a sore
>> spot where the *community* needs to do some work.
>
> We already have a lot of scripts that use my existing format, so if
> there's not a strong reason I would prefer to keep at least the basic
> format. This is mainly for the basics (enabling/disabling), not the individual
> tunables. It's also useful to have a simple way to specify at least
> enabling/disabling.
Sure, but the fact that you have a lot of scripts shouldn't play into
the design of what is right for our users :-)
Honestly I think PTHREAD_MUTEX_ELISION=on/off is a better environment
variable since it explains almost intuitively what it does, but I'll
follow up on that separately.
Cheers,
Carlos.