All of lore.kernel.org
 help / color / mirror / Atom feed
* hypersparc dvma aliasing
@ 2004-11-17  2:28 Bob Breuer
  2004-11-17  3:35 ` Bob Breuer
                   ` (8 more replies)
  0 siblings, 9 replies; 10+ messages in thread
From: Bob Breuer @ 2004-11-17  2:28 UTC (permalink / raw)
  To: sparclinux

After further testing, it certainly seems that hypersparcs
have aliasing issues between dvma and physical addresses.

Here are some results that I get using a minimum of cache
flushing with 512K of cache, and all dvma areas are cacheable:

iommu_get_one -- cache(va) = cache(pa):
va 0xf0777000, pa 0x00777000, dvma 0xf0010000 -- fails
va 0xf0777000, pa 0x00777000, dvma 0xf0777000 -- works

iommu_map_dma_area -- cache(va) != cache(addr):
va 0xfbd66000, addr 0xfff0e000, pba 0xf0066000 -- fails
va 0xfbd80000, addr 0xfff00000, pba 0xf0000000 -- works
va 0xfbd66000, addr 0xfff0e000, pba 0xf010e000 -- works


I've got some additional changes that allocate dvma addresses
with the right cache color and allow for less cache flushing.
I will post a patch shortly.

Bob


^ permalink raw reply	[flat|nested] 10+ messages in thread

end of thread, other threads:[~2004-11-24 16:54 UTC | newest]

Thread overview: 10+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2004-11-17  2:28 hypersparc dvma aliasing Bob Breuer
2004-11-17  3:35 ` Bob Breuer
2004-11-17  3:47 ` William Lee Irwin III
2004-11-17  6:11 ` David S. Miller
2004-11-23  6:18 ` Bob Breuer
2004-11-23  7:37 ` David S. Miller
2004-11-23 14:28 ` Stan Benoit
2004-11-23 15:35 ` Bob Breuer
2004-11-23 23:21 ` Stan Benoit
2004-11-24 16:54 ` William Lee Irwin III

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.