Subj : Re: Question for mr lockfree :)
To   : comp.programming.threads
From : Chris Thomasson
Date : Wed May 18 2005 12:28 am

>> For the time being, it is. I have yet to find a "proper" way to reliably
>> grow the number of per-thread hazard pointers without stopping the 
>> world..
>
> It could be done but I don't think it should be done if it adds an extra 
> level
> of indirection.

I did an implementation that allows for any number of hazard pointers 
per-thread, but it needed an extra memory barrier for the bit of extra state 
that needed to be scanned by the polling thread. It used per-thread 
hash-tables instead of hazard pointers. It hashed the pointer that needed to 
be protected into an index of a per-thread table and incremented the 
corresponding counter. This allows for a single hazard to protect multiple 
objects. I guess I should call the algorithm hazard proxies instead of 
hazard pointers.




> I don't think you need more than one hazard pointer per
> thread unless you do some kind of lock-free recursive traversal of a tree 
> or
> something.

This is exactly why I implemented it.




> You can always handle it as an error equivalent to stack overflow.

Yes, that would make it a lot easier.

:)

.