drm/nouveau/pwr: fix missing mutex unlock in a failure path

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

drm/nv40/therm: fix slowing down fan when pstate undefined

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

drm/nv11-: synchronise flips to vblank, unless async flip requested

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

drm/nvc0-: remove nasty fifo swmthd hack for flip completion method

Not required anymore as flips are always done on the kernel's channel,
which means we can use a proper software object class instead.

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

drm/nv10-: we no longer need to create nvsw object on user channels

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

drm/nouveau: always queue flips relative to kernel channel activity

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

drm/nouveau: there is no need to reserve/fence the new fb when flipping

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

drm/nouveau: when bailing out of a pushbuf ioctl, do not remove previous fence

Cc: stable@vger.kernel.org
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: allow nouveau_fence_ref() to be a noop

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

drm/nvc8/mc: msi rearm is via the nvc0 method

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

drm/nouveau/fb: implement various bits of work towards memory reclocking

Not even remotely ready for the vast majority of the world.

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

drm/nouveau: implement a simple sysfs interface to new pm code

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

drm/nouveau/device: initial control object class, with pstate control methods

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

drm/nouveau/clk: implement power state and engine clock control in core

User control of this has been hard-coded as disabled for now.

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

drm/nouveau/volt: implement voltage control in core

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

drm/nouveau/bios: parsing for various tables required for power management

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

drm/nouveau/perfmon: initial infrastructure to expose performance counters

Internal use only at this point.  Userspace later.

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

drm/nouveau/bus: add interfaces/helpers for sequencer

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

drm/nouveau/bus: make external class definitions pointers

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

drm/nouveau/pwr: initial implementation

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

drm/nouveau/therm: update target fanspeed outside of therm lock

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

drm/nouveau/therm: automatic mode by default

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

drm/nouveau/therm: no toggle fan control either if we can't guarantee no pwm connected

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

drm/nvc0/therm: allow fan control if we've killed the vbios ppwr ucode

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

drm/nouveau/therm: if no bios trip/linear info, default to perf-suggested speed

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

drm/nouveau/therm: add hook for clk to suggest fanspeed to therm

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

drm/nouveau/pwr: assign a subdev id for upcoming implementation

THERM will use this before it exists.

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

drm/nouveau/gpio: return different error code for not found vs invalid

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

drm/nouveau/drm/pm: remove everything except the hwmon interfaces to THERM

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

drm/nouveau/core: make all info-level messages silent for runtime pm

Removes the need for special handling of messages in init paths.

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

drm/nouveau/fifo: make external class definitions into pointers

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

drm/nouveau/device: recognise GK208

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

drm/nvc0-/gr: fix a number of missing explicit array terminators...

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

drm/nouveau/mc: fix (incorrect) reporting of unknown pending intr bits

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

drm/nouveau/disp: semi-complete link training sequence even if display disappears

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

drm/nouveau/bios: some older boards have shorter displayport tables

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

drm/nouveau/fbcon: bracket entrypoints with a per-device enabled check

We don't necessarily have fbcon on all devices these days.

Fixes suspend on (at least) Quadro NVS 450.

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

drm/nvd0-/disp: reorder writes to lane current control regs

Matches the nv94-nvc0 commit with the same title.  Doesn't fix a reported issue,
but NVIDIA use this ordering here too, so let's do it just in case.

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

drm/nv94-nvc0/disp: reorder writes to lane current control regs

Fixes link training issues on some boards.

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

drm/nouveau/disp: log if DP link training fails

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

drm/nouveau/i2c: use a custom bitbanging delay for the adt7473

This patch adds a way to define a custom delay when scanning for i2c devices
because the adt7473 sometimes doesn't like the default bitbanging udelay.

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

drm/nouveau/agp: add a quirk list to limit agp modes

Certain combinations of hardware can't actually support the maximum
detected speed. Add a quirk list that lists pairs of hostbridge/chip pci
ids and the mode that they should work with.

See https://bugs.freedesktop.org/show_bug.cgi?id=20341

Reported-by: Jason Detring <detringj@gmail.com>
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nvd9-/disp: disable display underflow reporting at init

Reported-by: Andy Ritger <aritger@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nva0/clk: fix accidental limiting of pll coefficients

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

drm/nv50-nvaf/fb: split fbram oclass in preparation for reclocking

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

drm/nouveau/fb: merge more bits and pieces into oclass definitions

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

drm/nouveau/fb: remove ram oclass argument from base fb constructor

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

drm/nv50-nvaf/fb: split the class definitions up a bit

These will diverge further in the future.

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

drm/nouveau/fb: make external class definitions pointers

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

drm/nvaa/mc: blacklist msi to off by default

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

drm/nv50-nv86,nv92/mc: rearm msi via pci config space, rather than mmio mirror

This is what NVIDIA do on these chipsets, let's hope it works around
the reported MSI failures for us on NV86.

v2: updated to include G92, as per information provided by NVIDIA.

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

drm/nvc0,nvc4/mc: handle 0xc0's "special" msi rearm

v2. updated to cover GF104, as per information provided by NVIDIA.

Reported-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/mc: store static data in nouveau_mc class definition

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

drm/nouveau/mc: fetch NV_PMC_INTR again after re-arming MSI

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

drm/nouveau/mc: bracket interrupt handler with NV_PMC_INTR_EN disable/re-enable

This looks to be what NVIDIA do pretty much everywhere, since forever.

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

drm/nouveau/mc: have single entry and exit points to the interrupt handler

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

drm/nouveau/mc: msi rearm write via subdev, not device

This way we can catch it with debugging on for PMC subdev.

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

drm/nouveau/device: use an additional bit from NV_PMC_BOOT_0 to identify chipset

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

drm/nouveau/bios/init: return failure condition on invalid opcodes

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

drm/nv31/mpeg: remove need for separate refcnt on engine use

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

drm/nouveau/therm: ack any pending IRQ at init

This is safe because ptherm hasn't been configured yet and will be a
little further down the initialization path. Ptherm should be safe
regarding to runtime reconfiguration.

v2:
  - do not limit this patch to nv84-a3 and make it nv84+

v3:
  - move the ack to fini()
  - disable IRQs on fini()
  - silently ignore un-requested IRQs

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

drm/nv31/mpeg: split the nv31 and nv40 dma setting implementations

NV31 has different config bits than NV40+ do. Also fix the DMA_IMAGE
VRAM-only setting to check the right bits.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv31/mpeg: store chan singleton in engine, use it for dispatch

This makes nv31+ able to actually perform the nv_call, since previously
the inst was not available.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv40/mpeg: use the nv31-provided classes

Since nv40 only covers pre-nv44 now, it can use the nv31-provided
functions.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv44/mpeg: create a copy of the nv31/nv40 impls

The nv31/nv40 impls are actually fairly nv44-specific, since they assume
the presence of the instance register/context switching. Create a copy
before nv31/nv40 get fixed.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv31/mpeg: no need to set compat mode differently for nv44 gr

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv10/kms: add plane support for nv10-nv40

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv10: fix chipset checks, mostly for the benefit of nv1a

NV1A is numerically higher than NV17 but generationally lower. Use the
new card type to help disambiguate.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv10: introduce a new NV_11 card type

NV11/17/1F/18 come after NV10/15/16/1A. In order to facilitate using
numerical comparisons, split up the two sets into different card types.

This change should be a no-op except that the relevant cards will see
NV11 printed instead of NV10 for the family.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: fix backlight mask on ppc powerbook

This code was originally moved to using nv_mask by d31e078d84. This
should not have any actual effect since the mask isn't applied to the
value.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: remove prototype for non-existent nouveau_connector_bpp

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/vic: rename PUNK1C1 to PVIC

NVIDIA's name for what rnndb calls PVCOMP.

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

drm/nouveau/therm: kill some over-zealous debugging

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

drm/nouveau/core: split lock into list+exec and enable refcount locks

This fixes a reported locking inversion when interacting with the DRM
core's vblank routines.

Reviewed-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/core: convert event handler apis to split create/enable semantics

This is a necessary step towards being able to work with the insane locking
requirements of the DRM core's vblank routines, and a nice cleanup as a
side-effect.

This is similar in spirit to the interfaces that Peter Hurley arrived at
with his nouveau_event rcu conversion series.

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

drm/nv50-/sw: share engine/channel constructor between implementations

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

drm/nouveau/sw: prepare for the sharing of constructors between implementations

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

drm/nv50-/sw: make vblank tracking data private to the implementations

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

drm/nv50-/sw: share engine/channel struct definitions between implementations

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

drm/nouveau/core: Allow asymmetric nouveau_event_get/_put

Most nouveau event handlers have storage in 'static' containers
(structures with lifetimes nearly equivalent to the drm_device),
but are dangerously reused via nouveau_event_get/_put. For
example, if nouveau_event_get is called more than once for a
given handler, the event handler list will be corrupted.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/core: Move event index check from critical section

The index_nr field is constant for the lifetime of the event, so
serialized access is unnecessary.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/core: Add priv field for event handlers

Provide private field for event handlers exclusive use.
Convert nouveau_fence_wait_uevent() and
nouveau_fence_wait_uevent_handler(); drop struct nouveau_fence_uevent.

Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: off by one in nouveau_drm_vblank_enable()

The test here should be ">= ARRAY_SIZE()" instead of "> ARRAY_SIZE()".

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: remove pointless assignment

self-assignment of a variable doesn't make a lot of sense.

Signed-off-by: Dave Jones <davej@fedoraproject.org>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

Merge branch 'ttm-next-3.13' of git://people.freedesktop.org/~thomash/linux into drm-next

- A couple of fixes that never made it into fixes-3.12
- Make NO_EVICT bo's available for shrinkers when on delayed-delete list
- Allow retrying page-faults that need to wait for GPU.

* 'ttm-next-3.13' of git://people.freedesktop.org/~thomash/linux:
  drm/ttm: Fix memory type compatibility check
  drm/ttm: Fix ttm_bo_move_memcpy
  drm/ttm: Handle in-memory region copies
  drm/ttm: Make NO_EVICT bos available to shrinkers pending destruction
  drm/ttm: Allow vm fault retries

Merge branch 'vmwgfx-next-3.13' of git://people.freedesktop.org/~thomash/linux into drm-next

Pull request for vmwgfx. Currently just the DMA address stuff.

* 'vmwgfx-next-3.13' of git://people.freedesktop.org/~thomash/linux:
  drm/vmwgfx: Use the linux DMA api to get valid device addresses of pages
  drm/ttm: Enable the dma page pool also for intel IOMMUs

drm/ttm: Fix memory type compatibility check

Also check the busy placements before deciding to move a buffer object.
Failing to do this may result in a completely unneccessary move within a
single memory type.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Cc: stable@vger.kernel.org

drm/ttm: Fix ttm_bo_move_memcpy

All error paths will want to keep the mm node, so handle this at the
function exit. This fixes an ioremap failure error path.
Also add some comments to make the function a bit easier to understand.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Cc: stable@vger.kernel.org

drm/ttm: Handle in-memory region copies

Fix the case where the ttm pointer may be NULL causing
a NULL pointer dereference.

Signed-off-by: Jakob Bornecrantz <jakob@vmware.com>
Signed-off-by: Thomas Hellström <thellstrom@vmware.com>
Cc: stable@vger.kernel.org

drm/ttm: Make NO_EVICT bos available to shrinkers pending destruction

NO_EVICT bos that are not idle when all references are dropped are put on
the delayed destroy list. However, since they are not on LRU lists, they
are not available to shrinkers at that point, and buffers on the delayed
destroy list are not checked very often for idle.

So when these buffers are put on the delayed destroy list, clear the
NO_EVICT flag and put them on the right LRU list. This way they are
immediately available for eviction or shrinkers and will not cause false
OOMS.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>

drm/ttm: Allow vm fault retries

Make use of the FAULT_FLAG_ALLOW_RETRY flag to allow dropping the
mmap_sem while waiting for bo idle.

FAULT_FLAG_ALLOW_RETRY appears to be primarily designed for disk waits
but should work just as fine for GPU waits..

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>

drm/vmwgfx: Use the linux DMA api to get valid device addresses of pages

The code handles three different cases:
1) physical page addresses. The ttm page array is used.
2) DMA subsystem addresses. A scatter-gather list is used.
3) Coherent pages. The ttm dma pool is used, together with the dma_ttm
array os dma_addr_t

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>

drm/ttm: Enable the dma page pool also for intel IOMMUs

Used by the vmwgfx driver

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Jakob Bornecrantz <jakob@vmware.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>

qxl: add a connector property to denote hotplug should rescan modes.

So GNOME userspace has an issue with when it rescans for modes on hotplug
events, if the monitor has no EDID it assumes that nothing has changed on
EDID as with real hw we'd never have new modes without a new EDID, and they
kind off rely on the behaviour now, however with virtual GPUs we would
like to rescan the modes and get a new preferred mode on hotplug events
to handle dynamic guest resizing (where you resize the host window and the
guest resizes with it).

This is a simple property we can make userspace watch for to trigger new
behaviour based on it, and can be used to replaced EDID hacks in virtual
drivers.

Acked-by: Marc-André Lureau <marcandre.lureau@gmail.com> (on irc)
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: delay minor destruction to drm_dev_free()

Instead of freeing minors in drm_dev_unregister(), we only unplug them and
delay the free to drm_dev_free(). Note that if drm_dev_register() has
never been called, minors are NULL and this has no effect.

This change is needed to allow early device unregistration. If we want to
call drm_dev_unregister() on live devices, we need to guarantee that
minors are still valid (but unplugged). This way, any open file can still
access file_priv->minor->dev to get the DRM device. However, the minor is
unplugged so no new users can occur.

Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: remove minor-id during unplug

Don't delay minor removal to drm_put_minor(). Otherwise, user-space can
still open the minor and cause the kernel to oops. Instead, remove the
minor during unplug so any new open() will fail to access this minor.

Note that open() and drm_unplug_minor() are both protected by the global
DRM mutex so we're fine.

Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: cleanup debugfs in drm_unplug_minor()

There is no reason to delay debugfs-cleanup to drm_put_minor(). We should
forbid any access to debugfs files once the device is dead. Chances they
oops once a card was unplugged are very high, anyway.

Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: make drm_get_minor() static

drm_get_minor() is only used in one file. Make it static and add a
kernel-doc comment which documents the current semantics.

Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: simplify drm_put_minor()

Allow passing NULL as minor to simplify DRM destruction paths. Also remove
the double-pointer reset as it is no longer needed. drm_put_minor() is
only called when the underlying object is destroyed. Hence, resetting
minors to NULL is not necessary.

As drm_put_minor() is no longer used by other DRM files, we can make it
static, too.

Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: call drm_unplug_minor() from drm_put_minor()

This protects drm_unplug_minor() against repeated calls so we can use it
in drm_put_minor(). This allows us to further simplify it in follow-ups as
we no longer do minor-destruction in both functions but only in
drm_unplug_minor().
Also add kernel-doc comments about what these calls do.

[airlied: fixup for changes to kdev stuff]
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>