This slipped in as a debugging aid, and never got turned off. The driver
appears to work fine without it on an i915 system, and for the non-default EXA
option, we'd rather see issues found than continue running with debugging aids
and hiding them behind bad performance.
It was basing off of the clock rate, but we have an override to use the
existing dual channel state when we can detect it, so the two settings were
conflicting.
Remove TV format from mode name, instead use an explicit output property and
split the input resolution from the tv format. Add properties to set the
blank area on all four sides of the image.
Originally we smashed vertex header to store texture
coordinates, this is working as we only use sf/wm kernel
and disable all other stages on pipeline. But better to
not do this. This also cleans up vertex elements state
and makes vertex buffer order looks "normal".
- Use the existing single/dual-channel state when available, as changing it
doesn't appear to work out.
- Set the power state of the CLKB and B0-B3 pairs according to whether
choose to go dual-channel or not.
- Restore the LVDS register at the appropriate point (before DPLLs are
re-programmed.
For some reason, certain chips don't correctly enable the SDVO hardware when
this register is written only once. We're following what the BIOS code does
and writing it twice now, but with extra posting reads to boot. Yes, this is
cult-and-paste, but it fixes problems found on deployed hardware.
Linux cannot allocate a large fixed buffer for the HW cursors as needed for
FreeBSD; instead, allocate four separate buffers. The code now prefers to
allocate one buffer (less overhead) and falls back to separate buffers only
when necessary.
Changing this value slows the entire I2C bus down, making it far more
reliable on older monitors. Note the same change has been made in the core X
server code; this change is included here to ensure that older X servers
work reliably with this driver.
Now, we allocate one single block of memory for cursors, and either succeed or
fail once, rather than trying to support partial fallback modes that generally
resulted in pain due to being untested. In particular, this fixes cursors on
FreeBSD, which only allowed one large physically-contiguous allocation.
Eric Anholt
Dave Airlie
Keith Packard
Wang Zhenyu
Nian Wu
Keith Packard
Michel Dänzer