XAA PixmapOps: Sync before accessing unwrapped callbacks.
When using any XAAPixmapOps, we call into unknown but freshly
unwrapped callbacks (like fb ones). Unlike the XAA*Fallback calls,
we did so without syncing first, exposing us to all kinds of
synchronisation issues.
I believe that the rendering errors appeared now because *PaintWindow
vanished (e4d11e58), and we just use miPaintWindow instead. This
takes a less direct route to the hw and ends up at
PolyFillRectPixmap, which very often left drawing artifacts.
We now sync accordingly, and no longer get the rendering artifacts i
was methodically reproducing on radeonhd, radeon, unichrome...
Also, in order to allow driver authors to remove extensive syncing
or flushing to hide this issue, create XAA_VERSION_ defines, put
them in xaa.h and bump the patchlevel.
tested by naddy@ and Edd Barrett. ok oga@.
PCI address space is completely seperate, so we don't have to claim any
resources to avoid clashes with main memory and firmware. Makes accelerated
graphics work on the Tadpole SPARCLE.
ok matthieu@, oga@
privsep ( O_NONBLOCK | O_RDWR | O_EXCL) by turning the list of allowed
devices into a struct, with the flags we're supposed to use, then using
these values with open(). Add /dev/dri/card0 there too (more'll be needed to be
added when it matters).
This gives privsep with dri a chance to work.
ok matthieu
CPUs in basic VESA mode:
X86EMU: handle CPUID instruction
Starting X on an AMD Geode LX system such as an Alix board with VGA
connector, the Xorg driver attempts to issue INT 10/0 to go to mode
3 (VGA). The emulator, running the BIOS code, would then spit out:
c000:0282: A2 ILLEGAL EXTENDED X86 OPCODE!
The opcode was 0F A2, or CPUID; it was not implemented in the
emulator. With this patch it handles the CPUID instruction in one of
two ways:
1) if ran on __i386__ or __x86_64__ then it calls the CPUID instruction
directly.
2) if ran elsewhere it returns a canned 486dx4 set of values for function 1.
Tested with & ok matthieu
overwritten with the enqueueInputProc (from upstream git)
cf. https://bugs.freedesktop.org/show_bug.cgi?id=13511
This should fix hangs people were seeing with icewm, cwm and most
probably other WM under some circumstances.
ok matthieu@
each cmsg_len (ie. msg_controllen = sum of CMSG_ALIGN(cmsg_len). This
works now that kernel fd passing has been fixed to accept a bit of
sloppiness because of this ABI repair.
lots of discussion with kettenis. From deraadt@
