Closed xiay-lcw closed 8 years ago
Cherry-picked that patch in my branch, thanks! Sorry for the delay in handling this.
Mmm, actually this is causing issues with GM206:
nouveau 0000:01:00.0: fifo: write fault at 00002e7000 engine 05 [BAR3] client 08 [HOST_CPU_NB] reason 0d [REGION_VIOLATION] on channel -1
GM20B does not use BAR3, which is probably why you have not noticed this?
In addition I suppose this should be the first patch of a series that enables 128K pages. As a standalone patch it is harder to justify upstream. Could you check that issue with BAR3 and resend it along with the full 128K-enablement series?
Urrk, seems like Github arbitrarily added a lot of people who are not related to this because I rebased my tree... this pull request system is really crap. Sorry everyone.
Page directory size depends on address space length and large page size. For future support of large page size other than 128K and/or variable large page size, we need to need to replace hard-coded page directory size with proper calculation.
Signed-off-by: Xia Yang xiay@nvidia.com