| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
|
|
- set dac_cntl on non-primary crtcs
- set XCRT_CNT_EN in CRTC_EXT_CNTL just because
- fix warnings in calls to xf86PrintModeline (wants screen index not
pScrnInfo)
|
|
- quick fix for the cursor handling to update to the latest
server bits
- make sure connector type is CONNECTOR_PROPRIETARY for LVDS.
|
|
This attempts to keep it inside the 32-bit limit when multiplying things out
later in the randr tree.
Let me know if I screwed this up..
|
|
I noticed fglrx has DDC for the panel in the rs480 laptop, however radeon
didn't pick it up, so I valgrinded fglrx and spotted 0x1a0/0x1a4 accesses
I actually noticed this before from the BIOS but never figured it out.
So now I get DDC from the LCD on this laptop.
|
|
|
|
|
|
Due to the hardware layout RMX ddc_mode has to be set.
If ddc_mode is set, RADEONValdiateFPModes() shouldn't be called.
Bugzilla #10620 (3).
|
|
According to Lisa Wu, this is correct regarding the VESA EEDID standard.
Bugzilla #10620 (2), original patch by Lisa Wu @ATI
|
|
Some scans used to only check every second entry, some stopped at the entry
before the last entry.
Bugzilla #10620 (1), original patch by Lisa Wu @ATI
|
|
If the secondary head isn't found (Monitor unplugged etc.) but MergedFB
is configured, the driver segfaults because it tries to access the mode
list private structures, which are not filled in.
|
|
|
|
Thanks to Matthew Garrett and Ubuntu for the hw loan to get this working.
Still no 3D driver support but at least you should get CP acceleration for
2D now.
|
|
This makes sure all modes of both CRTCs will be available by default with
MergedFB.
|
|
This makes sure mode lists will always be sorted from larger to smaller.
|
|
|
|
|
|
|
|
Fixes https://bugs.freedesktop.org/show_bug.cgi?id=10442 .
|
|
|
|
|
|
|
|
|
|
when IsSecondary is true, crtc1 is NULL
Noticed by Sverre Froyen.
|
|
|
|
atimisc is PCI-only now, we can get the chip name with xf86TokenToString().
|
|
There is no need to override pATI->ChipRevision for GX/CX, as it is only
reported with a printf.
|
|
atimisc is PCI-only now, so we only need to scan the PCI space.
|
|
|
|
|
|
Subdrivers are now loaded from the wrapper Probe function rather than at screen
creation time.
The wrapper Identify callback only prints chip families now, chip lists are
printed when a subdriver is loaded. This also avoids duplication of subdriver
Identify callbacks.
Unknown radeons should still get a list of known radeon and then probe fails...
Probe for atimisc last to avoid needless loading in most cases (r128, radeon).
|
|
For an MMIO-only driver, the extra MMIO address probing is not needed since we
already compute the MMIO address during probe and do not make any BIOS calls.
For a PIO driver, this extra probing could even result in a wrong MMIO address
since the checking is done by reading through PIO, not MMIO. The MMIO address
is later corrected but having the extra probe there is useless.
|
|
Partially revert "Minor refactoring of pATI->Block0Base computation."
This parially reverts commit 2c8b33b761e4411451aea1eea3a89d629aeea40d.
|
|
|
|
|
|
I can't see how these tests could fail on 32-bit. OTOH they are potentially
problematic with 64-bit PCI addresses.
|
|
- We reset write-combining of the FB when MMIO registers fall in the linear
aperture.
- We set MMIOInLinear in all three aperture mappings since the mapping code
depends on it.
- We map the linear aperture in its entirety for MMIOInLinear also.
This simplifies mapping of HW cursor and the LE aperture on BE archs, as
they always fall in the linear aperture.
- I assumed that BAR[2] is page-aligned (we know it's 4K-aligned for sure).
|
|
We will not be able to handle this when we do mappings for PCI BARs only and
it should be handled with a CPIO driver anyway.
|
|
We don't automatically require bumped minor versions.
|
|
In particular, don't sync again after accelerated DownloadFromScreen, which
syncs implicitly. This avoids calling into the kernel when it's not necessary,
which can be relevant in some situations.
|
|
|
|
Don't flush indirect buffer in BlockHandler; it's done in LeaveServer.
Also set the EXA engine mode to unknown only at the end.
|
|
There have been several reports of stability issues with things like the
xscreensaver hack hypercube.
|
|
Walk the SAREA texList and bump the age of every active object, so their owners
will consider them kicked out when they grab the HW lock next time.
|
|
|
|
|
|
|
|
|
