drm/nouveau: Take fence spinlock before reading the last sequence.

It fixes a race between the TTM delayed work queue and the GEM IOCTLs
(fdo bug 29583) uncovered by the BKL removal.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: drop drm_global_mutex before sleeping in submission path

If we keep hold of the mutex here, the process which currently holds the
buffer object will never be able to release it, causing a deadlock.

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

drm: export drm_global_mutex for drivers to use

Nouveau needs to be able to drop the mutex before sleeping to prevent a
deadlock from occuring.

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

drm/nv20: Don't use pushbuf calls on the original nv20.

The "return" command is buggy on the original nv20, it jumps back to
the caller address as expected, but it doesn't clear the subroutine
active bit making the subsequent pushbuf calls fail with a "stack"
overflow.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: Fix TMDS on some DCB1.5 boards.

The TMDS output of an nv11 was being detected as LVDS, because it uses
DCB type 2 for TMDS instead of type 4.

Reported-by: Bertrand VIEILLE <Vieille.Bertrand@free.fr>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: Fix backlight control on PPC machines with an internal TMDS panel.

This commit fixes fdo bug 29685.

Reported-by: Vlado Plaga <rechner@vlado-do.de>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv30: Apply modesetting to the correct slave encoder

Signed-off-by: Patrice Mandin <patmandin@gmail.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: Use a helper function to match PCI device/subsystem IDs.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv50: add dcb type 14 to enum to prevent compiler complaint

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

drm/nouveau: fix earlier mistake when fixing merge conflict

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

drm/nvc0: fix thinko in instmem suspend/resume

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

drm/nouveau: Workaround missing GPIO tables on an Apple iMac G4 NV18.

This should fix the reported TV-out load detection false positives
(fdo bug 29455).

Reported-by: Vlado Plaga <rechner@vlado-do.de>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: Add TV-out quirk for an MSI nForce2 IGP.

The blob also thinks there's a TV connected, so hardware bug...

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv50-nvc0: ramht_size is meant to be in bytes, not entries

Fixes an infinite loop that can happen in RAMHT lookup.

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

drm/nouveau: punt some more log messages to debug level

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

drm/nouveau: remove warning about unknown tmds table revisions

This message is apparently confusing people, and is being blamed for some
modesetting issues.  Lets remove the message, and instead replace it
with an unconditional printout of the table revision.

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

drm/nouveau: check for error when allocating/mapping dummy page

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

drm/nouveau: fix race condition when under memory pressure

When VRAM is running out it's possible that the client's push buffers get
evicted to main memory.  When they're validated back in, the GPU may
be used for the copy back to VRAM, but the existing synchronisation code
only deals with inter-channel sync, not sync between PFIFO and PGRAPH on
the same channel.  This leads to PFIFO fetching from command buffers that
haven't quite been copied by PGRAPH yet.

This patch marks push buffers as so, and forces any GPU-assisted buffer
moves to be done on a different channel, which triggers the correct
synchronisation to happen before we submit them.

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

drm/nv50: fix minor thinko from nvc0 changes

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

drm/nouveau: Don't try DDC on the dummy I2C channel.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nvc0: fix typo in PRAMIN flush

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

drm/nouveau: Fix DCB TMDS config parsing.

Thinko caused by 43bda05428a3d2021f3c12220073e0251c65df8b.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>

drm/nv30: Fix PFB init for nv31.

Fixes a regression introduced by 58bbb63720c8997e0136fe1884101e7ca40d68fd
(fdo bug 29324).

Reported-by: Johannes Obermayr <johannesobermayr@gmx.de>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv04: Fix up SGRAM density detection.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/i2c/ch7006: Don't use POWER_LEVEL_FULL_POWER_OFF on early chip versions.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: Init dcb->or on cards that have no usable DCB table.

We need a valid OR value because there're a few nv17 cards with DCB v1.4.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: reduce severity of some "error" messages

There's some known configurations where the lack of these tables/scripts
is perfectly normal, reduce visibilty of complaint messages to debug.

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

drm/nvc0: backup bar3 channel on suspend

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

drm/nouveau: implement init table opcodex 0x5e and 0x9a

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Marcin Kościelnicki <koriakin@0x04.net>

drm/nouveau: implement init table op 0x57, INIT_LTIME

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Marcin Kościelnicki <koriakin@0x04.net>

drm/nvc0: implement crtc pll setting

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

drm/nvc0: fix evo dma object so we display something

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

drm/nvc0: rudimentary instmem support

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

drm/nvc0: implement memory detection

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

drm/nvc0: allow INIT_GPIO

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

drm/nvc0: starting point for GF100 support, everything stubbed

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

drm/nv30: Workaround dual TMDS brain damage.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: No need to set slave TV encoder configs explicitly.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv17-nv4x: Attempt to init some external TMDS transmitters.

sil164 and friends are the most common, usually they just need to be
poked once because a fixed configuration is enough for any modes and
clocks, so they worked without this patch if the BIOS had done a good
job on POST. Display couldn't survive a suspend/resume cycle though.
Unfortunately, BIOS scripts are useless here.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv10: Fix up switching of NV10TCL_DMA_VTXBUF.

Not very nice, but I don't think there's a simpler workaround.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: Ack the context switch interrupt before switching contexts.

Leaving the IRQ unack'ed while switching contexts makes the switch
fail randomly on some nv1x.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: Fix the INIT_CONFIGURE_PREINIT BIOS opcode.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: Don't pass misaligned offsets to io_mapping_map_atomic_wc().

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: unwind on load errors

nouveau_load() just returned directly if there was an error instead of
releasing resources.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Reviewed-by: Marcin Slusarz <marcin.slusarz@gmail.com>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: Fix AGP reset when AGP FW is already enabled on init.

Previously nouveau_mem_reset_agp() was only disabling AGP fast writes
when coming back from suspend. However, the "locked out of the card
because of FW" problem can also be reproduced on init if you
unload/reload nouveau.ko several times. This patch makes the AGP code
reset FW on init.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: remove unused ttm bo list

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

drm/nv50: use custom i2c algo for dp auxch

This makes it easier to see how this is working, and lets us transfer the
EDID in blocks of 16 bytes.

The primary reason for this change is because debug logs are rather hard
to read with the hundreds of single-byte auxch transactions that occur.

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

agp: intel-agp: do not use PCI resources before pci_enable_device()

IRQ and resource[] may not have correct values until
after PCI hotplug setup occurs at pci_enable_device() time.

The semantic match that finds this problem is as follows:

// <smpl>
@@
identifier x;
identifier request ~= "pci_request.*|pci_resource.*";
@@

(
* x->irq
|
* x->resource
|
* request(x, ...)
)
 ...
*pci_enable_device(x)
// </smpl>

Signed-off-by: Kulikov Vasiliy <segooon@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

agp: efficeon-agp: do not use PCI resources before pci_enable_device()

IRQ and resource[] may not have correct values until
after PCI hotplug setup occurs at pci_enable_device() time.

The semantic match that finds this problem is as follows:

// <smpl>
@@
identifier x;
identifier request ~= "pci_request.*|pci_resource.*";
@@

(
* x->irq
|
* x->resource
|
* request(x, ...)
)
 ...
*pci_enable_device(x)
// </smpl>

Signed-off-by: Kulikov Vasiliy <segooon@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: kill BKL from common code

This restricts the use of the big kernel lock to the i830 and i810
device drivers. The three remaining users in common code (open, ioctl
and release) get converted to a new mutex, the drm_global_mutex,
making the locking stricter than the big kernel lock.

This may have a performance impact, but only in those cases that
currently don't use DRM_UNLOCKED flag in the ioctl list and would
benefit from that anyway.

The reason why i810 and i830 cannot use drm_global_mutex in their
mmap functions is a lock-order inversion problem between the current
use of the BKL and mmap_sem in these drivers. Since the BKL has
release-on-sleep semantics, it's harmless but it would cause trouble
if we replace the BKL with a mutex.

Instead, these drivers get their own ioctl wrappers that take the
BKL around every ioctl call and then set their own handlers as
DRM_UNLOCKED.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: David Airlie <airlied@linux.ie>
Cc: dri-devel@lists.freedesktop.org
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/kms: Simplify setup of the initial I2C encoder config.

In most use cases the driver will be using the same static config all
the time: interpreting i2c_board_info::platform_data as the default
config we can can save the GPU driver a redundant set_config() call.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm,io-mapping: Specify slot to use for atomic mappings

This is required should we ever attempt to use an io-mapping where
KM_USER0 is verboten, such as inside an IRQ context.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: only expose underscan on avivo chips

R4xx also uses the atom add connector function, but underscan is only
supported on avivo chips.

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon: add new pci ids

New evergreen and r7xx ids.

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Cleanup after failing to create master->unique and dev->name

v2: Userspace (notably xf86-video-{intel,ati}) became confused when
drmSetInterfaceVersion() started returning -EBUSY as they used a second
call (the first done in drmOpen()) to check their master credentials.
Since userspace wants to be able to repeatedly call
drmSetInterfaceVersion() allow them to do so.

v3: Rebase to drm-core-next.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon: tone down overchatty acpi debug messages.

On non laptop systems we'll see these the whole time, so make them
less important.

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

drm/radeon/kms: enable underscan option for digital connectors

This connector attribute allows you to enable or disable underscan
on a digital output to compensate for panels that automatically
overscan (e.g., many HDMI TVs).  Valid values for the attribute are:

off - forces underscan off
on - forces underscan on
auto - enables underscan if an HDMI TV is connected, off otherwise

default value is auto.

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: fix calculation of h/v scaling factors

Prior to this patch the code was dividing the src_v by the dst_h
and vice versa, rather than src_v/dst_v and src_h/dst_h.

This could lead to problems in the calculation of the display
watermarks.

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

Merge remote branch 'intel/drm-intel-next' of /ssd/git/drm-next into drm-core-next

* 'intel/drm-intel-next' of /ssd/git/drm-next: (230 commits)
  drm/i915: Clear the Ironlake dithering flags when the pipe doesn't want it.
  drm/agp/i915: trim stolen space to 32M
  drm/i915: Unset cursor if out-of-bounds upon mode change (v4)
  drm/i915: Unreference object not handle on creation
  drm/i915: Attempt to uncouple object after catastrophic failure in unbind
  drm/i915: Repeat unbinding during free if interrupted (v6)
  drm/i915: Refactor i915_gem_retire_requests()
  drm/i915: Warn if we run out of FIFO space for a mode
  drm/i915: Round up the watermark entries (v3)
  drm/i915: Typo in (unused) register mask for overlay.
  drm/i915: Check overlay stride errata for i830 and i845
  drm/i915: Validate the mode for eDP by using fixed panel size
  drm/i915: Always use the fixed panel timing for eDP
  drm/i915: Enable panel fitting for eDP
  drm/i915: Add fixed panel mode parsed from EDID for eDP without fixed mode in VBT
  drm/i915/sdvo: Set sync polarity based on actual mode
  drm/i915/hdmi: Set sync polarity based on actual mode
  drm/i915/pch: Set transcoder sync polarity for DP based on actual mode
  drm/i915: Initialize LVDS and eDP outputs before anything else
  drm/i915/dp: Correctly report eDP in the core connector type
  ...

drm/radeon/kms/igp: sideport is AMD only

Intel variants don't support it.

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: handle the case of no active displays properly in the bandwidth code

Logic was:
if (mode0 && mode1)
else if (mode0)
else

Should be:
if (mode0 && mode1)
else if (mode0)
else if (mode1)

Otherwise we may end up calculating the priority regs with
unitialized values.

Fixes:
https://bugzilla.kernel.org/show_bug.cgi?id=16492

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: move ttm global code to core drm

I wrote this for the prime sharing work, but I also noticed other external
non-upstream drivers from a large company carrying a similiar patch, so I
may as well ship it in master.

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

drm/i915: Clear the Ironlake dithering flags when the pipe doesn't want it.

My fine DisplayPort output was getting ST dithering forever after
having had the LVDS enabled at one point.

Signed-off-by: Eric Anholt <eric@anholt.net>

drm/radeon/kms: make sure HPD is set to NONE on analog-only connectors

HPD is digital only.

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: make sure rio_mem is valid before unmapping it

If we were not able to map the io bar in device init, don't attempt
to unmap it in device fini.  All radeons should have a io bar, so
I doubt this would ever trigger, but just to be on the safe side...

Pointed out by: Alberto Milone <alberto.milone@canonical.com>
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/agp/i915: trim stolen space to 32M

Some BIOSes will claim a large chunk of stolen space.  Unless we
reclaim it, our aperture for remapping buffer objects will be
constrained.  So clamp the stolen space to 32M and ignore the rest.

Fixes https://bugzilla.kernel.org/show_bug.cgi?id=15469 among others.

Adding the ignored stolen memory back into the general pool using the
memory hotplug code is left as an exercise for the reader.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Simon Farnsworth <simon.farnsworth@onelan.com>
Tested-by: Artem S. Tashkinov <t.artem@mailcity.com>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Unset cursor if out-of-bounds upon mode change (v4)

The docs warn that to position the cursor such that no part of it is
visible on the pipe is an undefined operation. Avoid such circumstances
upon changing the mode, or at any other time, by unsetting the cursor if
it moves out of bounds.

"For normal high resolution display modes, the cursor must have at least a
single pixel positioned over the active screen.” (p143, p148 of the hardware
registers docs).

Fixes:

  Bug 24748 - [965G] Graphics crashes when resolution is changed with KMS
              enabled
  https://bugs.freedesktop.org/show_bug.cgi?id=24748

v2: Only update the cursor registers if they change.
v3: Fix the unsigned comparision of x,y against width,height.
v4: Always set CUR.BASE or else the cursor may become corrupt.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reported-by: Christian Eggers <ceggers@gmx.de>
Cc: Christopher James Halse Rogers <chalserogers@gmail.com>
Cc: stable@kernel.org
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Unreference object not handle on creation

When creating an object, we create the handle by which it is known to
the process and which own the reference to the object. That reference to
the new handle is what we want to transfer to the process, not the lost
reference to the object; so free the local object reference *not* the
process's handle reference.

This brings i915_gem_object_create_ioctl() into line with
drm_gem_open_ioctl()

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Attempt to uncouple object after catastrophic failure in unbind

If we fail to flush outstanding GPU writes but return the memory to the
system, we risk corrupting memory should the GPU recovery and complete
those writes. On the other hand, if we bail early and free the object
then we have a definite use-after-free and real memory corruption.
Choose the lesser of two evils, since in order to recover from the hung
GPU we need to completely reset it, those pending writes should
never happen.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Repeat unbinding during free if interrupted (v6)

If during the freeing of an object the unbind is interrupted by a system
call, which is quite possible if we have outstanding GPU writes that
must be flushed, the unbind is silently aborted. This still leaves the
AGP region and backing pages allocated, and perhaps more importantly,
the object remains upon the various lists exposing us to memory
corruption.

I think this is the cause behind the use-after-free, such as

  Bug 15664 - Graphics hang and kernel backtrace when starting Azureus
              with Compiz enabled
  https://bugzilla.kernel.org/show_bug.cgi?id=15664

v2: Daniel Vetter reminded me that kernel space programming is never easy.
We cannot simply spin to clear the pending signal and so must deferred
the freeing of the object until later.
v3: Run from the top level retire requests.
v4: Tested with P(return -ERESTARTSYS)=.5 from i915_gem_do_wait_request()
v5: Rebase against Eric's for-linus tree.
v6: Refactor, split and add a comment about avoiding unbounded recursion.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel@ffwll.ch>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Refactor i915_gem_retire_requests()

Combine the iteration over active render rings into a common function.
This is in preparation for reusing the idle function to also retire
deferred free requests.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Warn if we run out of FIFO space for a mode

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Round up the watermark entries (v3)

Even though "we have enough padding that it should be ok", round up the
watermark entries to the next unit to be on the safe side...

v2: Use the DIV_ROUND_UP macro
v3: Spotted a few more missing round-ups.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Typo in (unused) register mask for overlay.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Check overlay stride errata for i830 and i845

Apparently i830 and i845 cannot handle any stride that is not a multiple
of 256, unlike their brethren which do support 64 byte aligned strides.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Validate the mode for eDP by using fixed panel size

Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Always use the fixed panel timing for eDP

Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Enable panel fitting for eDP

When trying to set other display mode besides the fixed panel mode, the
panel fitting should be enabled. This is similar to LVDS.

Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Add fixed panel mode parsed from EDID for eDP without fixed mode in VBT

Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915/sdvo: Set sync polarity based on actual mode

Signed-off-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915/hdmi: Set sync polarity based on actual mode

Signed-off-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915/pch: Set transcoder sync polarity for DP based on actual mode

Signed-off-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Initialize LVDS and eDP outputs before anything else

This makes them sort to the front in X, which makes them likely to be
the primary outputs if you haven't specified a preference in your DE,
which is likely to be what you want.

Signed-off-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915/dp: Correctly report eDP in the core connector type

Do this for both real eDP and for PCH_DP_D when used as the eDP
connection.

Signed-off-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915/dp: Rename has_edp to is_pch_edp to reflect its real meaning

Signed-off-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Refactor panel fitting on the LVDS. (v2)

Move the common routines into separate functions to not only increase
readability, but also throwaway surplus code.

In doing so, we review the calculation of the aspect preserving scaling
and avoid the use of fixed-point until we need to calculate the accurate
scale factor.

v2: Improve comments as suggested by Jesse.

1 files changed, 105 insertions(+), 194 deletions(-)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Remove the redundant check for a fixed_panel_mode

We already checked just a couple of lines above that we have found a
fixed_panel_mode for the LVDS, so remove the surplus check.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: apply DP bandwidth workaround for PCH eDP as well

Fixes https://bugs.freedesktop.org/show_bug.cgi?id=29141 though the
workaround itself is still a bit of a mystery.

Tested-by: Adam Hill <sidepipeuk@yahoo.co.uk>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>

agp/intel: Destroy the scatterlist on allocation failure

A side-effect of being able to use custom page allocations with the
sg_table is that it cannot reap the partially constructed scatterlist if
fails to allocate a page. So we need to call sg_free_table() ourselves
if sg_alloc_table() fails.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc Dave Airlie <airlied@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>

Merge remote branch 'origin/master' into drm-intel-next

This resolves the conflict in the EDP code, which has been rather
popular to hack on recently.

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

agp/intel: Add actual definitions of the Sandybridge PTE caching bits.

agp/intel: Support the extended physical addressing bits on Sandybridge.

Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
[anholt: Split this patch out of a larger patch for Sandybridge fixes]
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Include any alternate names by which the device is known.

When trying to keep track of features between the kernel, the 2D driver,
mesa and the specs, it helps to list any other name by which the device
is referred to.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Use 128k alignment for untiled display surface on i965 (v2)

The original i965, including the revised G35 and Q35, requires an
alignment of 128K for the display surface with linear memory, so
increase the requirement from 64k for these chipsets. For the later
chipsets in the i965 family, only a 4k alignment is required. (So
long as we do not start performing asynchronous flips.)

Note the impact of this should be slight as on i965 we should be using a
tiled frontbuffer for anything up to a 4096x4096 display.

v2: compilation fixes and note that the docs do not exclude the G35 from
the extra alignment.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915/pch: Cosmetic fix to FDI link training

Unmask the bits for link training reporting before starting link
training.  If stage 1 training finished before we unmask them, then we'd
spin around in a loop a few times until smashing on through.  Which is
harmless, since training _did_ succeed, it just looks ugly in dmesg.

Signed-off-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: cleanup: use ARRAY_SIZE()

NUM_TV_MODES is the same as ARRAY_SIZE(tv_modes).  In the end, I
decided it was cleaner to remove NUM_TV_MODES and just use
ARRAY_SIZE(tv_modes) through out.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Include instdone[1] in hangcheck

References:

  Bug 26691 - Spurious hangcheck whilst executing a long shader over a
              large vertex buffer
  https://bugs.freedesktop.org/show_bug.cgi?id=26691

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Remove the WARN when failing to set tiling.

We generally issue an error message at the point of failure, and so this
warning with a fairly pointless stacktrace is superfluous and ugly.
Needless to say, the common trigger for this WARN happens to be EIO
where this is pure noise.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: fix FDI frequency check

Since mode->clock is in kHz we should be checking against 2700000
instead of just 27000.  This patch gets my x201s working again (well
working as well as it ever was anyway).

When looking for this I also noticed we set link_bw to 270000, but the
calculation is different.  Does it also need to use kHz or we using
10kHz internally?

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>

drm/i915: Sparse warning about invalid value for burst_ena in tv_modes

drivers/gpu/drm/i915/intel_tv.c|479 col 16| warning: cast truncates bits
from constant value (8 becomes 0)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>