From: Keith P. <ke...@ke...> - 2008-08-22 06:00:33
|
On Thu, 2008-08-21 at 20:21 -0700, Jesse Barnes wrote: > Ok, as discussed on IRC this also needs either full pin/map/unmap/unpin on the > client side, or real fault support like TTM, or we could fragment and/or run > out of GTT space, which would be bad. This needs faulting support; that didn't look hard to me though, there are fairly straightforward functions that invalidate VMA ranges; future access hits a fault function that you can provide. -- kei...@in... |