Add infrastructure to wait for all planes updates to finish by using
an atomic_t variable to track how many pending updates we are waiting
plus a wait_queue for the wait part.

It also changes vblank behaviour and keeps it enabled for all types
of updates
Signed-off-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: NInki Dae <inki.dae@samsung.com>

The atomic modesetting interfaces supports async commits that should be
implemented by the drivers. If drm core requests an async commit
exynos_atomic_commit() will now schedule a work task to run the update later.

It also serializes commits that needs to run on the same crtc, putting the
following commit to wait until the current one is finished.
Signed-off-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: NInki Dae <inki.dae@samsung.com>

fimd_update_plane() programs BUF_START[win] and during the update
BUF_START[win] is copied to BUF_START_S[win] (its shadow register)
and starts scanning out, then it raises a irq.

The fimd_irq_handler, in the case we have a pending_fb, will check
the fb value was copied to START_S register and finish the update
in case of success.

Based on patch from Daniel Kurtz <djkurtz@chromium.org>
Signed-off-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: NInki Dae <inki.dae@samsung.com>

This macro is need to get the value of the START shadow register, that
will tell if an framebuffer is currently displayed on the screen or not.
Signed-off-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: NInki Dae <inki.dae@samsung.com>

The current code was ignoring the end of update for all overlay planes,
caring only for the primary plane update in case of pageflip.

This change adds a change to start to check for pending updates for all
planes through exynos_plane->pending_fb. At the start of plane update the
pending_fb is set with the fb to be shown on the screen. Then only when to
fb is already presented in the screen we set pending_fb to NULL to
signal that the update was finished.
Signed-off-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: NInki Dae <inki.dae@samsung.com>

fixup! drm/exynos: check for pending fb before finish update

Only set/clear the update bit in the CRTC's .atomic_begin()/flush()
so all planes are really committed at the same time.
Signed-off-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: NInki Dae <inki.dae@samsung.com>

From: Gustavo Padovan <gustavo.padovan@collabora.co.uk>

.prepare_plane() and .cleanup_plane() allows to perform extra operations
before and after the update of planes. For FIMD for example this will
be used to enable disable the shadow protection bit.
Signed-off-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: NInki Dae <inki.dae@samsung.com>

Unify handling of finished plane update to prepare for a following patch
that will check for the START and START_S regs to really make sure that
the plane was updated.
Signed-off-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: NInki Dae <inki.dae@samsung.com>

struct drm_crtc already stores the enabled state of the crtc
thus we don't need to replicate enabled in exynos_drm_crtc.
Signed-off-by: NGustavo Padovan <gustavo.padovan@collabora.co.uk>
Signed-off-by: NInki Dae <inki.dae@samsung.com>

align with mask code in overlay.c, Ben can clean the naming
up later if necessary.
Signed-off-by: NDave Airlie <airlied@redhat.com>

Reported-by: Ilia Mirkin on irc.
Signed-off-by: NDave Airlie <airlied@redhat.com>

Rather large pull request this time around, due to the long-pending
cleanup of the kernel driver being here.  There's a stupidly large
number of commits for that, as I wanted to have the series be
bisectable at a fairly fine-grained level.  That said, a very large
portion of the churn in the rework was automated, and a very large
number of boards from right across the whole range we support have
been tested.  I'm fairly confident there shouldn't be (too many)
issues from this.

Beyond correcting some not-so-great design decisions and making the
code a lot easier to work with, there's not much exciting (lower
memory usage, GPU VM should be a lot faster, etc) to be gained by the
end-user as a result of the cleanup, it mostly lays the groundwork for
future improvements.

A big thanks goes to Alexandre Courbot for testing/debugging the GK20A
codepaths for me :)

Highlights:
- A heap of perfmon work, providing a more useful userspace interface
and specifying counters for a bunch of boards
- Support for GT200 reclocking + other misc pm improvements
- Initial patches towards supporting GM20B (Tegra X1)
- Maxwell DisplayPort fixes
- Cleanup of the kernel driver
- The usual collection of random fixes

* 'linux-4.3' of git://anongit.freedesktop.org/git/nouveau/linux-2.6: (312 commits)
  drm/nouveau: bump driver version for release
  drm/nouveau/tegra: merge platform setup from nouveau drm
  drm/nouveau/pci: merge agp handling from nouveau drm
  drm/nouveau/device: remove pci/platform_device from common struct
  drm/nouveau/device: import pciid list and integrate quirks with it
  drm/nouveau/device: cleaner abstraction for device resource functions
  drm/nouveau/mc: move device irq handling to platform-specific code
  drm/nouveau/mc/gf100-: handle second interrupt tree
  drm/nouveau/mc: abstract interface to master intr registers
  drm/nouveau/pci: new subdev
  drm/nouveau/object: merge with handle
  drm/nouveau/core: remove the remainder of the previous style
  drm/nouveau/mpeg: convert to new-style nvkm_engine
  drm/nouveau/sw: convert to new-style nvkm_engine
  drm/nouveau/pm: convert to new-style nvkm_engine
  drm/nouveau/gr: convert to new-style nvkm_engine
  drm/nouveau/fifo: convert to new-style nvkm_engine
  drm/nouveau/disp: convert to new-style nvkm_engine
  drm/nouveau/dma: convert to new-style nvkm_engine
  drm/nouveau/cipher: convert to new-style nvkm_engine
  ...

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

The copyright header in nvkm/engine/device/platform.c has been replaced
with the NVIDIA one from drm/nouveau_platform.c, as most of the actual
code is now theirs.
Signed-off-by: NBen Skeggs <bskeggs@redhat.com>

This commit reinstates the pre-DEVINIT AGP fiddling that was broken in
an earlier commit.
Signed-off-by: NBen Skeggs <bskeggs@redhat.com>

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

PCI IDs taken from the NVIDIA binary driver, with permission.
Signed-off-by: NBen Skeggs <bskeggs@redhat.com>

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

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

Doesn't fix any known issue, but best be safe in case control is handed
to us from firmware with these left enabled.
Signed-off-by: NBen Skeggs <bskeggs@redhat.com>

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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