On Thu, Feb 28, 2002 at 01:00:59AM +0100, Oliver Neukum wrote:
> On Thursday 28 February 2002 00:29, Greg KH wrote:
> > On Thu, Feb 28, 2002 at 12:18:41AM +0100, Oliver Neukum wrote:
> > > Why ?
> > > If the pool is formed by allocating the memory for urbs in chunks
> > > the positive effect on TLB and cache is there whether the allocation
> > > in the drivers is dynamic or not.
> > > In fact, could we use the slab cache as is ?
> >
> > We could.  But the point I was trying to make is that USB drivers only
> > create urbs very infrequently (on device insertion), so any speed
> > benefits are quite limited (we don't really care about speed when
> > initializing a device.)
> 
> I must disagree partially.
> The effect on TLB usage is the same whether the allocation
> is dynamic or not. Furthermore we should plan for usb2.0
> devices where it might be become noticeable.

I agree.

> AFAIK a Pentium class CPU has 16 entries for large pages.
> This means that we can cache TLB entries for 64MB.
> For my machine this makes a miss probability of 3/4
> with randomly distributed urbs.
> If we can't use large pages things become nasty.
> USB on a 486 ?

Think USB on uCLinux (it's true) :)

greg k-h

_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel

Reply via email to