drivers/gpu/drm/savage/savage_state.c: add missing kfree

Most of the error handling code in this function frees the buffers
kcmd_addr, kvb_addr, and kbox_addr allocated at the beginning of this
function.  These two branches are changed to do the same.

Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drivers/gpu/drm/radeon/radeon_cs.c: eliminate possible double free

The function radeon_cs_parser_init is only called from two places, in
drivers/gpu/drm/radeon/radeon_cs.c and drivers/gpu/drm/radeon/r600_cs.c.
In each case, if the call fails another function is called that frees all
of the kdata and dpage information in the chunks array.  So this
information should not be freed in radeon_cs_parser_init as well.

Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon: dp only enable enhanced framing if sink supports.

DisplayPort spec v1.1a, Table 2-52.

Same commit to Intel by ajax in 9962c9252e46eda7058067cbe73bdf1ed74b0d37

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: skip db/cb/streamout checking when possible on r600

Signed-off-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: compute GPU addresses correctly on r600

Signed-off-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: remove some unused variables in evergreen_cs_track

and document the other unused ones.

Signed-off-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: skip db/cb/streamout checking when possible on evergreen

Signed-off-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: optimize streamout checking for evergreen

Signed-off-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: compute GPU addresses correctly on evergreen

There are also two fixes:
- In DRAW_INDEX_2, we read idx_value, but should have read idx+1.
- When correcting SQ_VTX_CONSTANT_WORD1_0.SIZE, we should subtract
  the offset.

Signed-off-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: make some DRM errors more informative

Signed-off-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: skip cb/db checking if SX_MISC is 1 on r600+

Signed-off-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/usb: move usb support into a separate module

In order to satisfy all the various Kconfig options between
USB and DRM, we need to split the USB code out into a separate module
and export symbols to it.

This fixes build problems in -next reported by sfr.

Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: fix build with UDL if USB is a module

reported by sfr on -next merge.

Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/udl: initial UDL driver (v4)

This is an initial drm/kms driver for the displaylink devices.

Supports fb_defio,
supports KMS dumb interface
supports 24bpp via conversion to 16bpp, hw can do this better.
supports hot unplug using new drm core features.

On an unplug, it disables connector polling, unplugs connectors
from sysfs, unplugs fbdev layer (using Kay's API), drops all the
USB device URBs, and call the drm core to unplug the device.

This driver is based in large parts on udlfb.c so I've licensed
it under GPLv2.

Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: add core support for unplugging a device (v2)

Two parts to this, one is simple unplug from sysfs for the device node.

The second adds an unplugged state, if we have device opens, we
just set the unplugged state and return, if we have no device
opens we drop the drm device.

If after a lastclose we discover we are unplugged we then
drop the drm device.

v2: use an atomic for unplugged and wrap it for users,
add checks on open + mmap + ioctl entry points.

Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/modeset: add helper to unplug all connectors from sysfs

In order to get correct ordering at hot-unplug for userspace,
we need to tear down all the sysfs bits at the correct time.

This adds a helper to allow drivers to remove the sysfs nodes
for all connectors.

Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/sysfs: protect sysfs removal code against being run twice.

a step towards correct hot unplug for USB devices, we need to
remove the userspace facing bits at the unplug time for correct
udev operation.

Signed-off-by: Dave Airlie <airlied@redhat.com>

udlfb: remove sysfs framebuffer device with USB .disconnect()

The USB graphics card driver delays the unregistering of the framebuffer
device to a workqueue, which breaks the userspace visible remove uevent
sequence. Recent userspace tools started to support USB graphics card
hotplug out-of-the-box and rely on proper events sent by the kernel.

The framebuffer device is a direct child of the USB interface which is
removed immediately after the USB .disconnect() callback. But the fb device
in /sys stays around until its final cleanup, at a time where all the parent
devices have been removed already.

To work around that, we remove the sysfs fb device directly in the USB
.disconnect() callback and leave only the cleanup of the internal fb
data to the delayed work.

Before:
 add      /devices/pci0000:00/0000:00:1d.0/usb2/2-1/2-1.2 (usb)
 add      /devices/pci0000:00/0000:00:1d.0/usb2/2-1/2-1.2/2-1.2:1.0 (usb)
 add      /devices/pci0000:00/0000:00:1d.0/usb2/2-1/2-1.2/2-1.2:1.0/graphics/fb0 (graphics)
 remove   /devices/pci0000:00/0000:00:1d.0/usb2/2-1/2-1.2/2-1.2:1.0 (usb)
 remove   /devices/pci0000:00/0000:00:1d.0/usb2/2-1/2-1.2 (usb)
 remove   /2-1.2:1.0/graphics/fb0 (graphics)

After:
 add      /devices/pci0000:00/0000:00:1d.0/usb2/2-1/2-1.2 (usb)
 add      /devices/pci0000:00/0000:00:1d.0/usb2/2-1/2-1.2/2-1.2:1.0 (usb)
 add      /devices/pci0000:00/0000:00:1d.0/usb2/2-1/2-1.2/2-1.2:1.0/graphics/fb1 (graphics)
 remove   /devices/pci0000:00/0000:00:1d.0/usb2/2-1/2-1.2/2-1.2:1.0/graphics/fb1 (graphics)
 remove   /devices/pci0000:00/0000:00:1d.0/usb2/2-1/2-1.2/2-1.2:1.0 (usb)
 remove   /devices/pci0000:00/0000:00:1d.0/usb2/2-1/2-1.2 (usb)

Cc: stable@vger.kernel.org
Tested-by: Bernie Thompson <bernie@plugable.com>
Acked-by: Bernie Thompson <bernie@plugable.com>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Florian Tobias Schandinat <FlorianSchandinat@gmx.de>

drm: Merge tag 'v3.3-rc7' into drm-core-next

Merge the fixes so far into core-next, needed to test
intel driver.

Conflicts:
drivers/gpu/drm/i915/intel_ringbuffer.c

drm: Add drm_mode_copy()

Add a helper function to copy a display mode. Use it in
drm_mode_duplicate() and nouveau mode_fixup hooks.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Use a flexible array member for blob property data

The blob property data is always allocated immediately after the object
header. No need for the extra indirection when accessing it, just use
a flexible array member.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Handle drm_object_get() failures

Check drm_mode_object_get() return value everywhere.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Make drm_crtc_convert_{umode, to_umode} static and constify their params

drm_crtc_convert_umode() and drm_crtc_convert_to_umode() are never
used outside drm_crtc.c, so make them static. Also make the input
mode structure const for both functions.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Fix drm_mode_attachmode_crtc()

Change drm_mode_attachmode_crtc() to take an "all or nothing" approach.
If an error is returned, there are no side effects visible.

Also change the function to always duplicate the mode passed in.

Also change the function to not give up when it finds the first
connector without and encoder.

A simpler approach would be to just remove the function completely as
it's unused currently.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Check CRTC viewport against framebuffer size

Make sure the requested CRTC viewport fits inside the
framebuffer.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Check user mode against overflows

The internal mode representation drm_display_mode uses signed data
types. When converting the user mode to internal representation,
check that the unsigned values don't overflow the signed datatypes.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Fix memory leak in drm_mode_setcrtc()

The mode passed to the .set_config() hook was never freed. The drivers
will make a copy of the mode, so simply free it when done.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Make drm_mode_attachmode() void

drm_mode_attachmode() always returns 0. Change the return type to void.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Check crtc x and y coordinates

The crtc x/y panning coordinates are stored as signed integers
internally. The user provides them as unsigned, so we should check
that the user provided values actually fit in the internal datatypes.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Warn if mode to umode conversion overflows the destination types

When converting from a drm_display_mode to drm_mode_modeinfo, print a
warning if the the timings values don't fit into the __u16 datatype.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Change drm_display_mode::type to unsigned

The drm_display_mode type is a bitmask so it should be unsigned.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Reject mode set with current fb if no current fb is bound

When doing a mode set with the special fb id -1, reject the mode set if
no fb is currently bound to the crtc.

Also remove the pointless list traversal to find the current crtc based
on the current crtc :)

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: suspend/resume support for Cedartrail

Update our tree to match the current driver head.

Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: Fix resume paths

We fall apart somewhat on resume because we don't invoke all the resume
methods as we should. Fix the silly error in the logic.

Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/nouveau: add userspace fallback hints.

This lets the modesetting driver work better.

Signed-off-by: Dave Airlie <airlied@redhat.com>

Merge branch 'drm-nouveau-next' of git://anongit.freedesktop.org/git/nouveau/linux-2.6 into drm-core-next

* 'drm-nouveau-next' of git://anongit.freedesktop.org/git/nouveau/linux-2.6: (57 commits)
  drm/nouveau: map first page of mmio early and determine chipset earlier
  drm/nvd0/disp: disconnect encoders before reprogramming them
  drm/nvd0/disp: move syncs/magic setup to or mode_set
  drm/nouveau/dp: account for channel coding overhead in link training
  drm/nvd0/disp: fix dcb sor link matching in supervisor handler
  drm/nvd0/disp: initial implementation of displayport
  drm/nouveau/dp: make dp dpms function common, call from sor code instead
  drm/nv50/hwsq: some nv92 fixes
  drm/nouveau/dp: move all nv50/sor-specific code out of nouveau_dp.c
  drm/nouveau/dp: make functions for executing various bios tables
  drm/nouveau/pm: fix oops if chipset has no pm support at all
  drm/nouveau/bios: rework vbios shadowing
  drm/nouveau/bios: attempt acpi rom fetch before pcirom
  drm/nvd0/disp: attempt to handle more than 2 crtcs if possible
  drm/nvc0/vram: get part count from PUNITS
  drm/nv40/pm: fix fanspeed regression
  drm/nouveau/pm: several fixes for nvc0 memory timings
  drm/nvc0/pm: restrict pll mode to clocks that can actually use it
  drm/nouveau/dp: fix bad comparison in dp_link_train_commit()
  drm/nouveau/mxm: call mxmi to determine revision before calling mxms
  ...

drm/nouveau: map first page of mmio early and determine chipset earlier

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nvd0/disp: disconnect encoders before reprogramming them

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nvd0/disp: move syncs/magic setup to or mode_set

NVIDIA appear to do these around the same place they do the MODE_CTRL
methods, and for DP at least we need to bash some extra bits in "syncs"
to keep EVO happy.

It's a bit of a guess as to the 6/8bpc, but i have no better idea yet.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/dp: account for channel coding overhead in link training

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nvd0/disp: fix dcb sor link matching in supervisor handler

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nvd0/disp: initial implementation of displayport

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/dp: make dp dpms function common, call from sor code instead

GF119 will use this too.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv50/hwsq: some nv92 fixes

The shift from hwsq_data = 0x1400 to 0x080000 actually happened in nv94, not nv92
This fixes some reclocking issues on my newly acquired nv92

Signed-off-by: Martin Peres <martin.peres@labri.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/dp: move all nv50/sor-specific code out of nouveau_dp.c

Off-chip encoders (which we don't support yet anyway), and newer chipsets
(such as NVD9...), will need their own code for this.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/dp: make functions for executing various bios tables

More code to do the same thing, but will make it easier to handle various
changes that could possibly happen the the VBIOS tables.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: fix oops if chipset has no pm support at all

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/bios: rework vbios shadowing

Refactored to allow shadowing of VBIOS images longer than 64KiB, which
allows us to pass the VBIOS checksum test on certain boards.

There's also a workaround for reading the PROM VBIOS on some chipsets.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/bios: attempt acpi rom fetch before pcirom

There's cards out there with completely messed up PCIROM images that have
a perfectly valid signature.. Sigh!

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nvd0/disp: attempt to handle more than 2 crtcs if possible

Theoretically handles CRTC2/CRTC3, should any GF119 out there actually
have them enabled.  The room is there for the regs etc, so why not :)

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nvc0/vram: get part count from PUNITS

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv40/pm: fix fanspeed regression

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: several fixes for nvc0 memory timings

This patch fixes two small issues in timing generation as spotted on
several NVCx cards.

In addition, the header of the file is updated to also contain (some of)
the current developers of this code.

Signed-off-by: Roy Spliet <r.spliet@student.tudelft.nl>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nvc0/pm: restrict pll mode to clocks that can actually use it

Fixes reclocking failure on some chips where we attempted to set PDAEMON
to PLL mode.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/dp: fix bad comparison in dp_link_train_commit()

The comparison (lpre == DP_TRAIN_PRE_EMPHASIS_9_5) is always false:
lpre is initialized as (lane & 0x0c) >> 2, which is at most 3, while
DP_TRAIN_PRE_EMPHASIS_9_5 is defined as (3 << 3).

Signed-off-by: Xi Wang <xi.wang@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/mxm: call mxmi to determine revision before calling mxms

There's a HP laptop out there where the MXM version in the VBIOS doesn't
match what the ACPI implementation is expecting.  These tables will accept
0x00 to MXMS to return latest version, but *only* if MXMI has been called
first..

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: init only after display subsystem has been created

This patch fixes an oops cause by pm_trigger accessing the (uninitialised)
crtc list.

Reported-by: Roy Spliet <r.spliet@student.tudelft.nl>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nvc0/fb: detect presense of second rank

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv50/display: expose color vibrance control

Signed-off-by: Christoph Bumiller <e0425955@student.tuwien.ac.at>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv50-nvc0/vm: support unsnooped system memory

v2 (Emil Velikov <emil.l.velikov@gmail.com>):
- Fixed a regression on certain nv50 IGP due to not passing the correct
  target type to nv50_vm_addr()

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Tested-by: Johannes Obermayr <johannesobermayr@gmx.de>

drm/nouveau: recognise DCB connector type for DP+DVI+VGA DMS-59

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/mem: handle dll_off for ddr2/ddr3

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nouveau/pm: extend profile interface for destroy/init/fini

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nouveau/pm: rework to allow selecting separate profiles for ac/battery

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nouveau/pm: fix dll off -> dll on transitions

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nouveau/pm: detect when we need dll disabled for gddr3

Fixes minor flickering on NVS295 when at perflvl 0.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nv50: fix detection of second vram rank

Goes a long way to correcting NVS295 memory reclocking issues.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nouveau/pm: track mr2 for gddr3

There's some "extended" GDDR3 chipsets out there with EMRS2 settings that
change the layout of MRS/EMRS1 bitmaps.. Sigh.. Still need to track down
how exactly we're supposed to handle this.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nv50/pm: wait for all fifo-connected engines to idle before reclocking

Signed-off-by: Martin Peres <martin.peres@labri.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv50/pm: use hwsq for engine reclocking too

Idea from Martin Peres, different implementation by me.

v2: Martin Peres:
- fix mast calculation

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nv50/disp: more accurate function to determine active crtcs

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nv50/pm: initial work towards proper memory reclocking, with timings

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nouveau/pm: introduce ram reclocking helper

This will probably result in more lines of code, however, we're going to
have at least 3 slightly different implementations of this very soon and
I'd rather keep the ram reclocking logic separate from the hw specifics.

DDR2/DDR3/GDDR3 implemented thus far, others will be added as necessary.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nouveau/pm: embed timings into perflvl structs

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nouveau/pm: calculate memory timings at perflvl creation time

Statically generating the PFB register and MR values for each timing set
turns out to be insufficient.  There's at least one (so far) known piece
of information which effects MR values which is stored in the perflvl
entry on some chipsets (and in another table on later ones), which is
disconnected from the timing table entries.

After this change we will generate a timing set based on an input clock
frequency instead, and have this data stored in the performance level
data.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nouveau/pm: readback boot perflvl *before* parsing vbios

We might want/need the boot data to generate the other perflevels.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nouveau/pm: implement DDR2/DDR3/GDDR3/GDDR5 MR generation and validation

Roy Spliet:
- Implement according to specs
- Simplify
- Make array for mc latency registers

Martin Peres:
- squash and split all the commits from Roy
- rework following Ben Skeggs comments
- add a form of timings validation
- store the initial timings for later use

Ben Skeggs
- merge slightly modified tidy-up patch with this one
- remove perflvl-dropping logic for the moment

Signed-off-by: Roy Spliet <r.spliet@student.tudelft.nl>
Signed-off-by: Martin Peres <martin.peres@labri.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: restructure bios table parsing

It turns out we need access to some additional information in various VBIOS
tables to handle PFB memory timings correctly.

Rather than hack in parsing of the new stuff in some kludgy way, I've
restructured the VBIOS parsing to be more primitive, so we can use them in
more flexible ways in the future.

The perflvl->timing association code is disabled for the moment until it can
be reworked.  We don't use this stuff yet anyway, so no harm done.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nouveau/pm: avoid potential divide-by-zero

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Martin Peres <martin.peres@labri.fr>

drm/nouveau: Fix module parameter description formats

Module parameter descriptions don't take a trailing \n, otherwise it
breaks formatting of modinfo's output. Also remove trailing space.

Signed-off-by: Jean Delvare <jdelvare@suse.de>
Cc: David Airlie <airlied@linux.ie>
Cc: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: improve memory timing generation

- Rename several VBIOS entries to closer match the real world
- Add the missing 0x100238 and 0x100240 register values
- Parse bit 14 of the VBIOS timing table
- "Magic value" -> tCWL, fixing some minor bugs in the process
- Also name a few more by their name rather than their number.
- Some values seem to be dependent on the memory type. Fix

Edits by Martin Peres <martin.peres@labri.fr>:
- this is a squash commit
- reworked for fixing some style issues

Signed-off-by: Roy Spliet <r.spliet@student.tudelft.nl>
Signed-off-by: Martin Peres <martin.peres@labri.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: improve the reclocking logs' readability

Signed-off-by: Martin Peres <martin.peres@labri.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: move pwm_divisor to the nouveau_pm_fan struct

Signed-off-by: Martin Peres <martin.peres@labri.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: restore fan speed after suspend

Signed-off-by: Martin Peres <martin.peres@labri.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: style fixes

Signed-off-by: Martin Peres <martin.peres@labri.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: rework the init/takedown ordering

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nvc0: add initial memory type detection

Uses only the VBIOS tables, from what I can tell this is what NVIDIA do
too, I was able to change the detected memory type by modifying this table
on a NVC1 chipset.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv50: hopefully handle the DDR2/DDR3 memtype detection somewhat better

M version 2 appears to have a table with some form of memory type info
available.

NVIDIA appear to ignore the table information except for this DDR2/DDR3
case (which has the same value in 0x100714).  My guess is this is due to
some of the supported memory types not being represented in the table.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv50: add memory type detection

DDR1/DDR[23] confirmed on NVA8 (see note about DDR3 in source) by changing
the value and watching the binary driver's behaviour.

GDDR3/4 values confirmed on a NV96 via the same method above.  That GDDR4
is present is interesting, as far as I can see no boards using it were ever
released.

GDDR5 value is based on VBIOS images of known GDDR5 boards.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv20-nv40: add memory type detection

NV20/NV30 is partially educated guesswork at this point, based on any
information around about available memory types and a horribly unspeakable
amount of vbios image scouring.  I'm not entirely certain the GDDR3 define
is correct, I have not spotted a single vbios with that value yet (though
it is mentioned in some 1218-using nv4x vbios), but there are reports that
some nv3x did use it..

NV40(100914) confirmed by switching an NV49 to DDR1/DDR2 values and making
sure that the binary driver behaviour showed it had detected DDR1/DDR2
instead of GDDR3 before dying horribly.

NV40(100474) confirmed by doing much the same task as above on an NV44,
except this was *much* easier as changing the values didn't seem to have
any noticable effect on the memory controller aside from changing the
binary driver's behaviour.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv20: split PFB code out of nv10_fb.c

Most functions were quite different between NV10/NV20 already, and they're
about to get even more so.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: memory type detection for the really old chipsets

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: move vram detection funcs to chipset-specific fb code

Also, display detected memory type in logs - though, we don't even try to
detect this yet.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Linux 3.3-rc7

Merge branch 'drm-gma500-alanc' into drm-core-next

* drm-gma500-alanc: (47 commits)
  gma500: psb_irq: mark mid_{enable, disable}_pipe_event() as static
  gma500: mark psb_intel_sdvo_hdmi_sink_detect() as static
  gma500: psb_intel_display: drop unused variables
  gma500: mark psb_intel_pipe_set_base() as static
  gma500: drop unused psb_intel_modeset_cleanup()
  gma500: oaktrail_lvds_init() drop unused variable
  gma500: oaktrail_hdmi_i2c_access() drop unused variable
  gma500: mark oaktrail_backlight_init() as static
  gma500: oaktrail_hdmi: fix -Wmissing-field-initializers warning
  gma500: oaktrail_hdmi: drop dead code
  gma500: oaktrail_crtc: drop unused variables
  gma500: oaktrail_crtc: mark few functions as static
  gma500: mdfld_intel_display: drop unused variables
  gma500: mdfld_dsi_pkg_sender: fix -Wtype-limits warning
  gma500: mdfld_dsi_dpi: drop unused variables
  gma500: mdfld_device: mark few functions as static
  gma500: cdv_intel_lvds: #if 0 currently unused functions
  gma500: cdv_intel_lvds: mark few functions as static
  gma500: cdv_intel_hdmi: add missing include
  gma500: cdv_intel_crt: mark few functions as static
  ...

gma500: psb_irq: mark mid_{enable, disable}_pipe_event() as static

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: mark psb_intel_sdvo_hdmi_sink_detect() as static

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: psb_intel_display: drop unused variables

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: mark psb_intel_pipe_set_base() as static

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: drop unused psb_intel_modeset_cleanup()

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>