[Libdlo] [PATCH 2.6.32-rc7 1/1] udlfb: add dynamic modeset support

[Bernie Thompson]

Hi Roberto,

On the IOCTLs, do you think we could get rid of them entirely?

The IOCTL operations I see in displaylink-mod are:
 A) blit and copy
 B) set mode
 C) return driver names

C is information queries that we could better expose through sysfs.  A
and B mirror standard fbdev operations.

What kept the X server from using the starndard fbdev ops?  Is there
anything that keeps us on the IOCTLs, if we update the X server to
match?

Other notes:
* I need to update the patch (make a v3) that does reserve the pages
(as your code did originally).  Reasoning: Since the fbcon cursor uses
blit, and may come in on interrupt context (we know that from
Alexander's report of a kernel oops), we can't sleep on that path.
Since mmap or defio might not be in use (?), we need to explicitly
reserve pages to not allow vm to swap any out.
* I left check_var and set_par (from displaylink-mod) out of the
patch, but they really should go in as a part of modeset changes
* I realized that the new IOCTLs in the displaylink-mod codeline
weren't about rendering operations (although the copy/flip blit was
removed).  So my note that udlfb or udlfb+patch might have different
performance from displaylink-mod wasn't right - theoretically, they
should be the same.

Thanks,
Bernie