drm/nv04-nv40: Fix NULL dereference when we fail to find an LVDS native mode.

Reported-by: Alex Buell <alex.buell@munted.org.uk>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv10: Fix crash when allocating a BO larger than half the available VRAM.

Reported-by: Alex Buell <alex.buell@munted.org.uk>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv50: fix display on 0x50

Accidently busted a while back.  We'll be creating objects that aren't
necessary here, but, they're never used so no harm..

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

drm/nouveau: correctly pair hwmon_init and hwmon_fini

I broke this with my commit
07cfe0e7a820ecad078c04e9c2a102521709145d

This fixes fdo #33434

Signed-off-by: Lucas Stach <dev@lynxeye.de>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv50: fix regression on IGPs

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

drm/nvc0/grctx: correct an off-by-one

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

drm/nv50: Fix race with PFIFO during PGRAPH context destruction.

Reported-by: Xavier Chantry <chantry.xavier@gmail.com>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: Workaround incorrect DCB entry on a GeForce3 Ti 200.

Fixes the DVI-D output on that board (fdo bug 32645).

Reported-by: Bryan Quigley <BryanQuigley@Ubuntu.com>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nvc0: implement irq handler for whatever's at 0x14xxxx

This is just barely enough to stop a never-ending IRQ storm that can
be triggered by our 3D driver.  We have no idea what this engine is..

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

drm/nvc0: fix incorrect TPC register setup

Was hitting TPC+1's regs by accident, oops.

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

drm/nouveau: probe for adt7473 before f75375

There's a reported case where probing for f75375 causes the system
to hang completely, in this case there's an adt7473 at the same
i2c address.

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

drm/nouveau: remove dead function definition

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

drm/nouveau: fix gpu page faults triggered by plymouthd

The switch to separate BAR and channel address spaces made the fbcon memory
address calculation incorrect on NV50+ boards, this commit fixes that.

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

drm/nouveau: greatly simplify mm, killing some bugs in the process

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

drm/nvc0: enable protection of system-use-only structures in vm

Somehow missed this in the original merge of the nvc0 code.

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

drm/nv40: initialise 0x17xx on all chipsets that have it

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

drm/nv40: make detection of 0x4097-ful chipsets available everywhere

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

drm/nouveau: fix hwmon device binding

Bind the hwmon structs to nouveau device kobj. This makes sure
the hwmon files are created in the device subdir in line with
all other hwmon drivers.

Signed-off-by: Lucas Stach <dev@lynxeye.de>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: create grctx on the fly on all chipsets

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

drm/nvc0: fix init without firmware present

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

drm/nvc0/pgraph: fix 0x406028/0x405870 init

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

drm/nvc0/pgraph: more unit names

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

drm/nvc0/pfifo: support for chipsets with only one PSUBFIFO (0xc1)

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

drm/nvc0: reserve only subc 0 for kernel use

Current 3D driver expects this behaviour.  While this could be changed,
there's no compelling reason to reserve more than one subchannel for the
DRM.  If we ever need to use an object other then M2MF, we can just
re-bind subchannel 0 as required.

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

drm/nv50: sync up gr data error names with rnn, use for nvc0 also

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

drm/nvc0: parse a couple more PGRAPH_INTR

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

drm/nvc0: nuke left-over debug messages

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

drm/nvc0: kill off a couple more magics

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

drm/nouveau: Validate channel indices passed from userspace.

When hacking the libdrm for improvements, I triggered a kernel crash
related to the fact that the NOUVEAU_NOTIFIEROBJ_ALLOC ioctl calls
nouveau_channel_get with an unchecked channel index.
The patch ensures that the channel index is an unsigned and validates
its value in nouveau_channel_get.

Signed-off-by: Michel Hermier <hermier@frugalware.org>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: Only select ACPI_VIDEO if its dependencies are met

CONFIG_ACPI_VIDEO depends on more than just CONFIG_ACPI, so add those
dependencies to the Kconfig select condition.  The case where some
dependencies fail to be satisfied should be handled correctly, because
in that case the ACPI_VIDEO symbols we use are converted into
static-inline stubs.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nvc0: accelerate ttm buffer moves

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

drm/nvc0: initial support for tiled buffer objects

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

drm/nvc0: implement fbcon acceleration

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

drm/nvc0: implement pgraph engine hooks

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

drm/nvc0: implement pfifo engine hooks

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

drm/nvc0: implement fencing

Just simple REF_CNT fencing for the moment.

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

drm/nvc0: fix channel dma init paths

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

drm/nvc0: skip dma object creation for drm channel

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

drm/nvc0: implement channel structure initialisation

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

drm/nvc0: gpuobj_new need only check validity and init the relevant engine

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

drm/nvc0: reject the notifier_alloc ioctl

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

drm/nvc0: create shared channel vm

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

drm/nvc0: initial vm implementation, use for bar1/bar3 management

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

drm/nvc0: import initial vm backend

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

drm/nouveau: modify vm to accomodate dual page tables for nvc0

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

drm/nv50: add missing license header to nv50_fbcon.c

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

drm/nv50: fix smatch warning in nv50_vram.c

Reported-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: sizeof() vs ARRAY_SIZE()

ARRAY_SIZE() was intended here, sizeof() is too large.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/radeon/kms: properly print ontario chip id

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

drm/radeon/kms: fix vram start calculation on ontario (v2)

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

Merge remote branch 'nouveau/drm-nouveau-next' of ../drm-nouveau-next into drm-core-next

* 'nouveau/drm-nouveau-next' of ../drm-nouveau-next: (93 commits)
  drm/nv50: fix a couple of vm init issues
  drm/nv04-nv40: Fix up PCI(E) GART DMA object bus address calculation.
  drm/nouveau: kick vram functions out into an "engine"
  drm/nouveau: allow gpuobj vinst to be a virtual address when necessary
  drm/nv50: tidy up PCIEGART implementation
  drm/nv50: enable non-contig vram allocations where requested
  drm/nv50: enable 4KiB pages for small vram allocations
  drm/nv50: implement global channel address space on new VM code
  drm/nv50: implement BAR1/BAR3 management on top of new VM code
  drm/nv50: import new vm code
  drm/nv50: implement custom vram mm
  drm/nouveau: Avoid potential race between nouveau_fence_update() and context takedown.
  drm/nouveau: fix use of drm_mm_node in semaphore object
  drm/nouveau: wrap calls to ttm_bo_validate()
  drm/nouveau: no need to zero dma objects, we fill them completely anyway
  drm/nouveau: introduce a util function to wait on reg != val
  drm/nouveau: implicitly insert non-DMA objects into RAMHT
  drm/nouveau: make fifo.create_context() responsible for mapping control regs
  drm/nouveau: Spin for a bit in nouveau_fence_wait() before yielding the CPU.
  drm/nouveau: Use WC memory on the AGP GART.
  ...

drm/ttm: delay freeing of old node during move_memcpy until after iounmap

Drivers using their own implementation of io_mem_reserve/io_mem_free are
likely to store the tracking information for the map in mem.mm_node, so
it can't be freed while still mapped.

Signed-off-by: Ben Skeggs<bskeggs@redhat.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon: add initial tracepoint support.

this adds a bo create, and fence seq tracking tracepoints.

This is just an initial set to play around with, we should investigate
what others we need would be useful.

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

drm/nv50: fix a couple of vm init issues

Fixes overwriting the first page table entry when testing that the PRAMIN
BAR can be correctly read/written, and adds an additional bar flush after
poking the BAR3 control regs.

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

drm/nv04-nv40: Fix up PCI(E) GART DMA object bus address calculation.

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

drm/nouveau: kick vram functions out into an "engine"

NVC0 will be able to share some of nv50's paths this way.  This also makes
it the card-specific vram code responsible for deciding if a given set
of tile_flags is valid, rather than duplicating the allowed types in
nv50_vram.c and nouveau_gem.c

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

drm/nouveau: allow gpuobj vinst to be a virtual address when necessary

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

drm/nv50: tidy up PCIEGART implementation

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

drm/nv50: enable non-contig vram allocations where requested

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

drm/nv50: enable 4KiB pages for small vram allocations

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

drm/nv50: implement global channel address space on new VM code

As of this commit, it's guaranteed that if an object is in VRAM that its
GPU virtual address will be constant.

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

drm/nv50: implement BAR1/BAR3 management on top of new VM code

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

drm/nv50: import new vm code

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

drm/nv50: implement custom vram mm

This is required on nv50 as we need to be able to have more precise control
over physical VRAM allocations to avoid buffer corruption when using
buffers of mixed memory types.

This removes some nasty overallocation/alignment that we were previously
using to "control" this problem.

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

drm/nouveau: Avoid potential race between nouveau_fence_update() and context takedown.

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

drm/nouveau: fix use of drm_mm_node in semaphore object

At some point in the future, this bo won't necessarily be backed by
a drm_mm_node, so use the start/size fields of the ttm_mem_reg instead.

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

drm/nouveau: wrap calls to ttm_bo_validate()

This will be used later to fixup bo.offset with a buffer's fixed GPU
virtual address.

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

drm/nouveau: no need to zero dma objects, we fill them completely anyway

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

drm/nouveau: introduce a util function to wait on reg != val

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

drm/nouveau: implicitly insert non-DMA objects into RAMHT

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

drm/nouveau: make fifo.create_context() responsible for mapping control regs

The regs belong to PFIFO, they're different for pretty much the same
generations we need different PFIFO control for, and NVC0 is going
to be even more different than the rest.

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

drm/nouveau: Spin for a bit in nouveau_fence_wait() before yielding the CPU.

Sleeping doesn't pay off for very short delays in comparison with the
minimum granularity of schedule_timeout().

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

drm/nouveau: Use WC memory on the AGP GART.

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

drm/nouveau: Synchronize with the user channel before GPU object destruction.

There have been reports of PFIFO cache errors during context take down
(fdo bug 31637). They are caused by some GPU objects being taken out
while the channel is still potentially processing commands. Make sure
that all the previous rendering has landed before releasing a GPU
object.

Reported-by: Grzesiek Sójka <pld@pfu.pl>
Reported-by: Patrice Mandin <patmandin@gmail.com>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Acked-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv04-nv10: Don't re-enable FIFO access multiple times after IRQ dispatch.

nvxx_graph_isr is already taking care of it. In some cases this
could've made you miss PGRAPH interrupts (e.g. when you were supposed
to get several IRQs of the same kind in a row).

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

drm/nv04-nv40: Give "gpuobj->cinst" the same meaning as on nv50.

No functional changes, just simplify some code paths a bit.

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

drm/nouveau: fabricate DCB encoder table for iMac G4

In typical Apple fashion there's no standard information about what
encoders are present on this machine, this patch adds a quirk to
provide it.

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

drm/nouveau: tidy up and extend dma object creation interfaces

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

drm/nouveau: remove some useless GETPARAMs

These have been unused since UMS support was ripped out, so lets remove
them completely.

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

drm/nv84: fix minor issues in PCRYPT implementation

Fix running of destroy_context() when create_context() has never been
called for the channel, and fill in engine's tlb_flush() function pointer.

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

drm/nouveau: remove dummy page use from PCI(E)GART, use PTE present instead

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

drm/nv50: fix 0x100c90 init for NVAF

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

drm/nv50: allocate page for unknown PFB object in nv50_fb.c

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

drm/nouveau: rework gpu-specific instmem interfaces

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

drm/nouveau: simplify gpuobj suspend/resume

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

drm/nv50: rework PGPIO IRQ handling and hotplug detection

Allows callers to install their own handlers for when a GPIO line
changes state (such as for hotplug detect).

This also fixes a bug where we weren't acknowledging the GPIO IRQ
until after the bottom half had run, causing a severe IRQ storm
in some cases.

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

drm/nv04-nv40: unregister irq handler on destroy

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

drm/nouveau: tidy+move PGRAPH ISRs to their respective *_graph.c files

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

drm/nouveau: move PFIFO ISR into nv04_fifo.c

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

drm/nv04-nv40: register vblank isr

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

drm/nouveau: move bitfield/enum helpers to nouveau_util.c

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

drm/nv50: use register/unregister functionality for PDISPLAY ISR

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

drm/nv50: move GPIO ISR to nv50_gpio.c

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

drm/nv84: move PCRYPT ISR out of nouveau_irq.c

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

drm/nouveau: allow irq handlers to be installed by engine-specific code

Lets start to clean up this mess!

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

drm/nv50: 0x50c0 apparently works on NVA3+ too, so lets allow it

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

drm/nv50: fix compute object class

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

drm/nouveau: Fix sleep while atomic in nouveau_bo_fence().

Reported-by: Pekka Paalanen <pq@iki.fi>
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nv20: Add Z compression support.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Tested-by: Xavier Chantry <chantry.xavier@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau: Rework tile region handling.

The point is to share more code between the PFB/PGRAPH tile region
hooks, and give the hardware specific functions a chance to allocate
per-region resources.

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