I worked around problems by uploading only the part of the image that is actually displayed, so that the source point actually is always 0,0, and the source image is no larger than the user's screen.
This fixed problems on my computer but I am waiting for more testing reports befor e closing this bug as fixed.

Update : many bw problems have been fixed by my integration of #4686 (?) nv patch in nouveau, and extending of it to all cards from NV10 to NV30.
There still are some issues that nvidia does not seem to be willing to fix in nv. They're difficult to trigger, so not very important.

status update via irc:

< curro__> heh, nv1x bandwidth problems, that code is a bit broken
< curro__> it seems it did the calculations as if there was always just a single CRTC scanning out
< curro__> i workarounded it on 5603fe7f92d6b06c87f48007989ec1d66d3258bc by making it assume a worst case scenario
< curro__> it probably fixed that bug too
< pq> heh, ahuillet, can you re-test? :-)
< ahuillet> pq : I cannot test any longer
< ahuillet> I don't have a working NV1x setup
< curro__> pq: he would have to backport the fix to UMS
< curro__> pq: anyways i don't think my fix is the right thing to do
< curro__> but i haven't found the time to rework that bandwidth code

Reassigning to bug defaults.

This bug will probably go away if the overlay is ported to KMS, forcing from userspace the CRTC arbitration parameters the kernel sets after 2d809ca83bb makes the corruption on the right end of the screen disappear when playing a 1920x1080 YV12 video (at least on my nv11).

UMS is removed and anything should be ported to KMS. Is this still an issue or could we close this bug?

Closing this bug since it references a no longer existent part of nouveau and we got no new info about this.