2002-11-10 01:54:56

by James Bottomley

[permalink] [raw]
Subject: Re: [parisc-linux] Untested port of parisc_device to generic device interface

[email protected] said:
>
> I'd like to know more about what these machines look like in the real
> world. Specifically, I am interested in the trade-off of having a
> parameter to wback_fake_consistent so that it could be enabled or
> disabled on an individual basis.

Actually, so would I. I can suspect why there might exist machines like this
(say the consistent attribute is settable at the pgd level)

> I suspect that the parameter is not worth the clutter because these
> "partially consistent" machines either have a large amount of
> consistent memory, so the case of the allocation failing in the is not
> worth supporting, or it is easy to check for consistent memory on them
> with something like "if ((unsigned long) vaddr < 0xwhatever)", but I'm
> just guessing.

Well, if it has to be done, it can be done by making alloc_consistent return a
handle rather than an address and making wback/invalidate take the handle (but
it's certainly not ideal).

James



2002-11-10 02:08:46

by Matthew Wilcox

[permalink] [raw]
Subject: Re: [parisc-linux] Untested port of parisc_device to generic device interface

On Sat, Nov 09, 2002 at 09:01:29PM -0500, J.E.J. Bottomley wrote:
> Actually, so would I. I can suspect why there might exist machines like this
> (say the consistent attribute is settable at the pgd level)

well, there's a limited amount of space available for consistent mappings
on some machines. it's basically the same as the vmalloc space. i think
the best way to handle this is simply to fail to initialise if you can't
get the consistent memory you need.

--
Revolutions do not require corporate support.