This is the mail archive of the mailing list for the glibc project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Native POSIX Thread Library(NPTL) ARM Supporting Patches (1/3)

On Thu, May 29, 2003 at 06:26:52PM +0100, Philip Blundell wrote:
> On Thu, 2003-05-29 at 16:54, Daniel Jacobowitz wrote:
> > On Thu, May 29, 2003 at 03:28:35PM +0100, Philip Blundell wrote:
> > > On Thu, 2003-05-29 at 15:21, Jakub Jelinek wrote:
> > > > If ARMs are never going to be SMP, this seems like good idea
> > > > if there are no spare registers.
> > > 
> > > Yep, there are indeed no spare registers.  
> > > 
> > > Right now, there are no SMP ARM systems.  We might see some in the
> > > future, but there is always the option of locking down a line in the
> > > d-cache for use as CPU-local scratchpad memory.  I guess if that's a
> > > possibility, it might be better for the kernel to pick the address of
> > > the thread ID variable.
> > 
> > I suspect that having the system call wouldn't be a total performance
> > killer; how many cycles is e.g. getuid() on ARM?
> Couple of hundred, maybe?  I think it comes out at about 400ns on a
> netwinder.

My instinct suggests that's good enough.  For instance, GCC can be told
that the syscall is a const function so it can eliminate duplicate
calls, et cetera.

That's just my instinct though; feel free to disagree :)  This should
certainly be decided before we go much further, since it will be part
of the ABI.

Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]