[lkml]   [1998]   [Aug]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: 64 bit PCI and Linux

On Sun, 23 Aug 1998, David S. Miller wrote:
> Keep in mind that Linux will probably never support a >2GB physical
> address space, let alone a 36-bit one... The memory management issues
> are just too hairy and hard to make work when the physical address
> space is larger than the virtual one.

Note that Linux probably _will_ support access to the high physical memory
as a buffer cache or similar. It's the only reasonable use for it, and
it's the same thing NT does, for example. For once MS did a reasonable
thing - they told intel to forget about their original extremely hairy
scheme, and the new PAE36 things in the Xeon etc CPU's make it easy to
access the high memory without doing the horrible cludgy 3-level page
table thing.

(Not that three level page tables are bad - they aren't - but as done by
intel they just sucked ;)

So there might actually be interest in using 64-bit PCI on intel, but
probably only in a very limited sense: using 64-bit DMA in order to fill
and write from the upper memory region. That should be fairly transparent
from anything else - it doesn't involve remapping shared areas or anything
like that.

[ Side issue: I don't actually know how shared PCI devices show up in a
machine with 4GB of memory, so that may be a problem. It's not a problem
on non-intel machines, because on the alpha and others the PCI space is
completely separated from normal memory, so it should be ok to have
memory and PCI mapped at the same address even though it probably means
that you lose device-to-device direct transfers ]


To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
Please read the FAQ at

 \ /
  Last update: 2005-03-22 13:44    [W:0.066 / U:9.872 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site