Re: New filesystem for Linux

[Mikulas Patocka <mikulas@artax.karlin.mff.cuni.cz>]

> > I have these questions:
> >
> > * There is a rw semaphore that is locked for read for nearly all operations 
> > and locked for write only rarely. However locking for read causes cache 
> > line pingpong on SMP systems. Do you have an idea how to make it better?
> >
> > It could be improved by making a semaphore for each CPU and locking for 
> > read only the CPU's semaphore and for write all semaphores. Or is there a 
> > better method?
>
> If you believe you need a semaphore for protecting a mostly read structure, 
> then RCU is certainly a good candidate. (ie no locked operation at all)

RCU is for non-blocking operation only.

Maybe it could be use for maintaining an unlocked variable that prevents 
readers from entering a critical section --- but waiting in a write for a 
quiscent state would hurt.

> The problem with a per_cpu biglock is that you may consume a lot of RAM for 
> big NR_CPUS. Count 32 KB per 'biglock' if NR_CPUS=1024
>
> > * This leads to another observation --- on i386 locking a semaphore is 2 
> > instructions, on x86_64 it is a call to two nested functions. Has it some 
> > reason or was it just implementator's laziness? Given the fact that locked 
> > instruction takes 16 ticks on Opteron (and can overlap about 2 ticks with 
> > other instructions), it would make sense to have optimized semaphores too.
>
> Hum, please dont use *lazy*, this could make Andi unhappy :)
>
> What are you calling semaphore exactly ?
> Did you read Documentation/mutex-design.txt ?

I see --- I could replace that semaphore with mutex. But rw-semaphores 
can't be replaces with them.

Mikulas
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/