Re: vm / physmem info
Re: vm / physmem info
- Subject: Re: vm / physmem info
- From: Andrew Gallatin <email@hidden>
- Date: Mon, 20 Jun 2005 09:31:46 -0400 (EDT)
Mike Smith writes:
>
> On Jun 17, 2005, at 11:24 AM, Glen Beane wrote:
>
> > I just really need (actually, would like) to get a *rough* idea of
> > how memory is fragmented - if I have a HPC application that
> > allocates a large amount of memory, how fragmented is that in
> > physical ram. It's not the end of the world if I can't, but it
> > would definitely be helpful (for a number of reasons that are
> > outside the scope of my question)...
> >
> > I'm really only concerned with one or two processes
>
> If you can make some minor changes to the application and write an I/
> O kit driver, you could easily have the app issue an I/O against the
> driver and then have the driver use the IOKit functionality to do the
> virtual->physical mapping. That avoids trying to go backwards, which
> is what you're describing there.
I haven't had my coffee yet, so excuse me if I'm missing something
simple... But won't using an IOKit driver to do virt->phys
translations require using IOMemoryDescriptors and "preparing" the
memory? This would require page-ins if the memory was not currently
resident, so you might see 2 different virtual address mapped to the
same physical address..
If it is somehow possible to get a hold of the thread's pmap, it seems
like it *might* be better to walk the address space using the
(unsupported) pmap_find_phys.
Drew
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Darwin-kernel mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden