Re: [Valgrind-developers] multi threaded valgrind: a first (hacked) prototype for discussions

[Dave G. <go...@mc...>]

On Feb 16, 2012, at 1:21 AM CST, Bart Van Assche wrote:

> On Wed, Feb 15, 2012 at 10:28 AM, Julian Seward <js...@ac...> wrote:
>> And if a thread leaves generated code in order to remove
>> a translation for whatever reason .. how do we do that without
>> crashing the other running threads?
> 
> Maybe we should report this to the user as a (data race) error instead
> of trying to address this now.

What about using Userspace RCU to protect access to the translation table?  Wouldn't this allow you to ensure that the remover won't clobber any threads currently using that translation while keeping the fast-path fast?

Or is this too complex for the current stage of design?

-Dave