On 2015-12-14 at 11:06 "'Davide Libenzi' via Akaros"
<[email protected]> wrote:
> > We've been a little lax with adding code and dealing with licenses
> > in the past.  I'd like it to be clear from the commit message where
> > code came from.  I take it then that you added libpfm4-4.6.0 or
> > something from http://perfmon2.sourceforge.net/, possibly from
> > git://git.code.sf.net/p/perfmon2/libpfm4.
> >
> 
> Should I amend the commit, or add a README into the library (or both)?

thanks for amending it.  i was taking care of some other things and
didn't get to this email in time.  =)

> > Still, I'm fine with the lock (per cpu or global).  I (and
> > perhaps whoever reads the code in the future) just want to know if
> > there's something to be worried about.
> >
> 
> Well, multiple call from different tasks can come in, and the
> operations against the per CPU counters are definitely not atomic.

They should be atomic on Akaros, since the kernel messages (which
smp_do uses) aren't interruptible.  Let's leave the locks in for now -
they aren't hurting.

> > which is before perfmon_install_session_alloc().
> >
> > That's the sort of thing that could catch someone in the future,
> > even if it is unlikely.  I wasn't asking for the code to change; I
> > was just saying that a comment would have made it more clear both
> > for me and for future developers.  We ought to be explicit about
> > any of the invariants about locking or concurrent access.
> >
> 
> Should I add?

I'd appreciate a one-liner in there that says something like "pa must
be read-only once we publish it"

> I will revise that with Makefrag-user-app idea, until we decide for
> the future layout.

Sounds great.

Thanks for the fixes btw.  =)

Barret


-- 
You received this message because you are subscribed to the Google Groups 
"Akaros" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To post to this group, send email to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to