drm/nouveau/clk: cosmetic changes

This is purely preparation for upcoming commits, there should be no
code changes here.

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

drm/nouveau/bus: cosmetic changes

This is purely preparation for upcoming commits, there should be no
code changes here.

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

drm/nouveau/bios: cosmetic changes

This is purely preparation for upcoming commits, there should be no
code changes here.

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

drm/nouveau/bar: cosmetic changes

This is purely preparation for upcoming commits, there should be no
code changes here.

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

drm/nouveau/device: include core/device.h automatically for subdevs/engines

Pretty much every subdev/engine is going to need access to nvkm_device
shortly to touch registers and/or output messages.

The odd placement of the includes is necessary to work around some
inter-dependencies that currently exist.  This will be fixed later.

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

drm/nouveau/device: add direct pointer to struct device

A future commit will hide the platform/pci specifics from nvkm_device,
but it's still very useful in a lot of places to have access to the
Linux device struct.

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

drm/nouveau/device: add direct pointers to subdevs from nvkm_device

Will be used in upcoming commits to remove the need for lookup/runtime
type-checking functions when accessing foreign subdevs.

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

drm/nouveau/subdev: add direct pointer to nvkm_device

Will be utilised in upcoming commits to remove the need for heuristics
to lookup the device a subdev belongs to.

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

drm/nouveau/lib: various tweaks

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

drm/nouveau/pmu/gk104: implement a hackish workaround for a hw bug

Only a handful of machines have this enabled by default, where it's been
proven to work.  The workaround can be explicitly enabled with a module
option also.

Still waiting on feedback from NVIDIA for a proper idea of exactly what
this fix is doing, and how to implement it properly.

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

drm/nouveau/disp/dp: gm1xx appears to have same dp lane ordering as gm2xx

Fixes 2-lane DP on Quadro K620.

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

drm/nouveau/disp/dp: fix some tx_pu mishandling

We only need to mask 0x0f on GM2xx, and want to keep the higher bits on
earlier cards.

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

drm/nouveau/bios/dp: use alternate set of drvctl values where necessary

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

drm/nouveau/bios/dcb: accept "maxwell" lane count values for dcb 4.0

We previously assumed that the values "2" and "4" were new in DCB 4.1,
however, there's at least one GM107 DCB 4.0 board (Quadro K620) that
uses the newer values.

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

drm/nouveau/fb/sddr3: add WR/CWL values seen on a GK208

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

drm/nouveau/nv46: Change mc subdev oclass from nv44 to nv4c

MSI interrupts appear to not work for nv46 based cards. Change the mc
subdev oclass for these cards from nv44 to nv4c, the nv4c mc code is
identical to the nv44 mc code except that it does not use msi
(it does not define a msi_rearm callback).

BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=90435
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm/gf100: only use PBFB_BROADCAST.PM_UNK100 for PBFB signals

High level hardware events related to PBFB will monitor all partitions.
While we are at it, fix bitfield for this mux.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm/gf100: remove multiple definitions of GPC_DOM signal 0x0e

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm/gf100: remove undefined TEX.PM_UNKC8 mux

This mux only exists on GF108+ (except for GF110 one), but since it is
not used by the userspace we can drop it for now.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: allow zeroed signals to enable sources

Hardware signals index 0x00 are defined for some domains and they have
to be allowed to enable sources like the others.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm/nv50: TPC[0x3] must be used for PGRAPH muxs on G80

I thought that using TPC[0x0] like for G84:GT215 was sufficient on G80,
but it's actually not the case. According to NVIDIA PerfKit on Windows,
we have to configure PGRAPH related muxs on TPC[0x3] for this chipset.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm/nv50: fix wrong addr for ZCULL source on G80:GT215

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/bios: add opcodes 0x73 and 0x77

No known VBIOSes use these, but they are present in the actual VBIOS
table parsing logic. No harm in adding these too.

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

drm/nouveau/platform: recognize GM20B

Allow the platform driver to recognize GM20B.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/device: recognize GM20B

Recognize GM20B and assign the right engines and subdevs.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/gr: add GM20B support

Add support for GM20B's graphics engine, based on GK20A. Note that this
code alone will not allow the engine to initialize on released devices
which require PMU-assisted secure boot.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/fifo: add GM20B fifo

GM20B has a 512-channels FIFO similar to GK104.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/gr/gk20a: use same initialization sequence as nvgpu

GK20A's initialization was based on GK104, but differences exist in the
way the initial context is built and the initialization process itself.

This patch follows the same initialization sequence as nvgpu performs
to avoid bad surprises. Since the register bundles initialization also
differ considerably from GK104, the register packs are now loaded from
firmware files, again similarly to what is done with nvgpu.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/gr: use NVIDIA-provided external firmwares

NVIDIA will officially start providing GR firmwares through
linux-firmware for GPUs that require it. Change the GR firmware lookup
function to use these files.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm/gk104: add compute signals/sources

These signals and sources have been reverse engineered from CUPTI
(Linux). Graphics signals exposed by PerfKit (Windows only) will be
added later. I need to reverse engineer them and it's a bit painful.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm/gk104: re-use gf100_pm_ctor()

gk104_pm_ctor() is equal to gf100_pm_ctor().

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm/nv40: rename pcounter domains to 'pc' instead of 'pm'

This trivial patch makes thing more consistent since hardware signals
names are prefixed by 'pcXX'.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: expose name of domains

This is going to be very useful for GF100+ because each GPC can
have its own domain of counters.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/drm/nouveau/clk: fix tstate to pstate calculation

According to the tstate calculation in nvkm_clk_tstate(),
the range of tstate is from -(clk->state_nr - 1) to 0,
it mean the tstate is negative value. But in nvkm_pstate_work(),
it use (clk->state_nr - 1 - clk->tstate) to limit pstate,
it's not correct.
This patch fix it to use (clk->state_nr - 1 + clk->tstate) to
limit pstate.

Signed-off-by: Wei Ni <wni@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: some fixes related to sources

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: fix signals/sources for GT200+

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm/gf100: add compute signals/sources

These signals and sources have been reverse engineered from CUPTI
(Linux). Graphics signals exposed by PerfKit (Windows only) will be
added later. I need to reverse engineer them and it's a bit painful.

This commit also adds a new class for GF108 and GF117.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm/gf100: allow to share GPC, HUB and PART domains

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: stack perfdom class under perfmon

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

drm/nouveau/pm: swap perfmon/perfdom code to avoid forward decl in next commit

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

drm/nouveau/pm/nv50: add compute and graphics signals/sources

These signals and sources have been reverse engineered from NVIDIA
PerfKit (Windows) and CUPTI (Linux), they will be used to build complex
hardware events from the userspace.

This commit also adds a new class for GT200.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: allow the userspace to configure sources

Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: allow to configure domains instead of simple counters

Configuring counters from the userspace require the kernel to handle some
logic related to performance counters. Basically, it has to find a free
slot to assign a counter, to handle extra counting modes like B4/B6 and it
must return and error when it can't configure a counter.

In my opinion, the kernel should not handle all of that logic but it
should only write the configuration sent by the userspace without
checking anything. In other words, it should overwrite the configuration
even if it's already counting and do not return any errors.

This patch allows the userspace to configure a domain instead of
separate counters. This has the advantage to move all of the logic to
the userspace.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: allow the userspace to schedule hardware counters

This adds a new method NVIF_PERFCTR_V0_INIT which starts a batch of
hardware counters for sampling. This will allow the userspace to start
a monitoring session using the INIT method and to stop it with SAMPLE,
for example before and after a frame is rendered.

This commit temporarily breaks nv_perfmon but this is going to be fixed
with the upcoming patch.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: implement NVIF_PERFMON_V0_QUERY_SOURCE method

This allows to query the ID, the mask and the user-readable name of
sources for each signal.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: allow to query the number of sources for a signal

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: add concept of sources

A source (or multiplexer) is a tuple addr+mask+shift which allows to
control a block of signals. The maximum number of sources that a signal
can define is arbitrary limited to 8 and this should be large enough.
This patch allows to define multi-level of sources for a signal.

Each different sources are stored to a global list and will be exposed
to the userspace through the nvif interface in order to avoid conflicts.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: allow to monitor hardware signal index 0x00

This signal index must be always allowed even if it's not clearly
defined in a domain in order to monitor a counter like 0x03020100
because it's the default value of signals.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: use hardware signals indexes instead of user-readable names

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: change signal iter to u16

16 bits is large enough to store the maximum number of signals available
for one domain (i.e. 256).

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: allow to query signals by domain

This will allow to configure performance counters with hardware signal
indexes instead of user-readable names in an upcoming patch.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: implement NVIF_PERFMON_V0_QUERY_DOMAIN method

This allows to query the number of available domains, including the
number of hardware counter and the number of signals per domain.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: prevent creating a perfctr object when signals are not found

Since a new class has been introduced to query signals, we can now
return an error when the userspace wants to monitor unknown signals.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: reorganize the nvif interface

This commit introduces the NVIF_IOCTL_NEW_V0_PERFMON class which will be
used in order to query domains, signals and sources. This separates the
querying and the counting interface.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: remove unused nvkm_perfsig_wrap() function

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Martin Peres <martin.peres@free.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/pm: remove pmu signals

PDAEMON signals don't have to be exposed by the perfmon engine.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Martin Peres <martin.peres@free.fr>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/clk/nv50: Enable user reclocking for NVA0

Tested on a few cards. Probably works quite well for most, given they should
all be GDDR3.

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/fb/gddr3: Add a few CL and WR entries observed on GTX260

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/fb/ramnv50: GDDR3 script for NVA0

This looks surprisingly similar to scripts on earlier cards as well
but they don't seem to work just yet. That... and I don't have any, which
makes it a tough job to reverse engineer.

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/bios/ramcfg: Separate out RON pull value

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/bios/rammap: Parse perf mode as if it's a rammap entry

Some of the bits in there are similar to the bits in the gt215 rammap.

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/fb/ramnv50: Ressurect timing code, use proper timing/rammap handlers

Might need some generalisation to < GT200. For those: use at your own risk!

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/fb/ramgt215: No need to cuss like that

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/fb/ramnv50: Make 0x100da0 per-partition

Like on GT215

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drm/nouveau/bios/rammap: Pull DLLoff bit out of version 0x10 struct

In preparation of NV50 reclocking, where there is no version

Signed-off-by: Roy Spliet <rspliet@eclipso.eu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

remove unnecessary include

This was merged with core/device.h in an earlier commit, but somehow
never got removed.

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

Merge branch 'drm-dwhdmi-devel' of git://ftp.arm.linux.org.uk/~rmk/linux-arm into drm-next

Here are some development updates for the Synopsis Designware HDMI driver,
which clean up some of the code, and start preparing to add audio support
to the driver.  This series of patches are based on a couple of dependent
commits from the ALSA tree.

Briefly, the updates are:
- move comments which should have moved with the phy values to the IMX
  part of the driver.
- clean up the phy configuration: to all lookups before starting to
  program the phy.
- clean up the HDMI clock regenerator code
- use the drm_hdmi_avi_infoframe_from_display_mode() helper which allows
  the code to be subsequently simplified
- remove the unused 'regmap' pointer in struct dw_hdmi
- use the bridge drm device rather than the connector (we're the bridge
  code)
- remove private hsync/vsync/interlaced flags, getting them from the
  DRM mode structure instead.
- implement interface functions to support audio - setting the audio
  sample rate, and enabling the audio clocks.
- removal of broken pixel repetition support
- cleanup DVI vs HDMI sink handling
- enable audio only if connected device supports audio
- avoid double-enabling bridge in the sink path (once in mode_set, and
  again in commit)
- rename mis-named dw_hdmi_phy_enable_power()
- fix bridge enable/disable handing, so a plug-in event doesn't
  reconfigure the bridge if DRM has disabled the output
- fix from Vladimir Zapolskiy for the I2CM_ADDRESS macro name

These are primerily preparitory patches for the AHB audio driver and
the I2S audio driver (from Rockchip) for this IP.

* 'drm-dwhdmi-devel' of git://ftp.arm.linux.org.uk/~rmk/linux-arm:
  drm: bridge/dw_hdmi: fix register I2CM_ADDRESS register name
  drm: bridge/dw_hdmi: fix phy enable/disable handling
  drm: bridge/dw_hdmi: rename dw_hdmi_phy_enable_power()
  drm: bridge/dw_hdmi: avoid enabling interface in mode_set
  drm: bridge/dw_hdmi: enable audio only if sink supports audio
  drm: bridge/dw_hdmi: clean up HDMI vs DVI mode handling
  drm: bridge/dw_hdmi: don't support any pixel doubled modes
  drm: bridge/dw_hdmi: remove pixel repetition setting for all VICs
  drm: bridge/dw_hdmi: introduce interfaces to enable and disable audio
  drm: bridge/dw_hdmi: introduce interface to setting sample rate
  drm: bridge/dw_hdmi: remove mhsyncpolarity/mvsyncpolarity/minterlaced
  drm: bridge/dw_hdmi: use our own drm_device
  drm: bridge/dw_hdmi: remove unused 'regmap' struct member
  drm: bridge/dw_hdmi: simplify hdmi_config_AVI() a little
  drm: bridge/dw_hdmi: use drm_hdmi_avi_infoframe_from_display_mode()
  drm: bridge/dw_hdmi: clean up hdmi_set_clk_regenerator()
  drm: bridge/dw_hdmi: clean up phy configuration
  drm: imx/dw_hdmi: move phy comments
  drm/edid: add function to help find SADs

Merge tag 'vmwgfx-next-15-08-21' of git://people.freedesktop.org/~thomash/linux into drm-next

Pull request of 15-08-21

The third pull request for 4.3. Contains two fixes for regressions introduced
with previous pull requests.

* tag 'vmwgfx-next-15-08-21' of git://people.freedesktop.org/~thomash/linux:
  drm/vmwgfx: Remove duplicate ttm_bo_device_release
  drm/vmwgfx: Fix a circular locking dependency in the fbdev code

Merge branch 'drm-next-4.3' of git://people.freedesktop.org/~agd5f/linux into drm-next

- DP fixes for radeon and amdgpu
- IH ring fix for tonga and fiji
- Lots of GPU scheduler fixes
- Misc additional fixes

* 'drm-next-4.3' of git://people.freedesktop.org/~agd5f/linux: (42 commits)
  drm/amdgpu: fix wait queue handling in the scheduler
  drm/amdgpu: remove extra parameters from scheduler callbacks
  drm/amdgpu: wake up scheduler only when neccessary
  drm/amdgpu: remove entity idle timeout v2
  drm/amdgpu: fix postclose order
  drm/amdgpu: use IB for copy buffer of eviction
  drm/amdgpu: adjust the judgement of removing fence callback
  drm/amdgpu: fix no sync_wait in copy_buffer
  drm/amdgpu: fix last_vm_update fence is not effetive for sched fence
  drm/amdgpu: add priv data to sched
  drm/amdgpu: add owner for sched fence
  drm/amdgpu: remove entity reference from sched fence
  drm/amdgpu: fix and cleanup amd_sched_entity_push_job
  drm/amdgpu: remove amdgpu_bo_list_clone
  drm/amdgpu: remove the context from amdgpu_job
  drm/amdgpu: remove unused parameters to amd_sched_create
  drm/amdgpu: remove sched_lock
  drm/amdgpu: remove prepare_job callback
  drm/amdgpu: cleanup a scheduler function name
  drm/amdgpu: reorder scheduler functions
  ...

drm/amdgpu: fix wait queue handling in the scheduler

Freeing up a queue after signalling it isn't race free.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: remove extra parameters from scheduler callbacks

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: wake up scheduler only when neccessary

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: remove entity idle timeout v2

Removing the entity from scheduling can deadlock the whole system.
Wait forever till the remaining IBs are scheduled.

v2: fix comment as well

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com> (v1)

drm/amdgpu: fix postclose order

The context needs to finish before everything else.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: use IB for copy buffer of eviction

This aids handling buffers moves with the scheduler.

Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>

drm/amdgpu: adjust the judgement of removing fence callback

The fence in the array may be skipped if wait_all is false,
thus the related callback is not initialized with list head.
So removing this kind callback will cause NULL pointer reference.

Signed-off-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>

drm/amdgpu: fix no sync_wait in copy_buffer

when eviction is happening, if don't handle
dependency, then the fence could be dead off.

Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>

drm/amdgpu: fix last_vm_update fence is not effetive for sched fence

Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>

drm/amdgpu: add priv data to sched

Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>

drm/amdgpu: add owner for sched fence

Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>

drm/amdgpu: remove entity reference from sched fence

Entity don't live as long as scheduler fences.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: fix and cleanup amd_sched_entity_push_job

Calling schedule() is probably the worse things we can do.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: remove amdgpu_bo_list_clone

Not used any more.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: remove the context from amdgpu_job

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: remove unused parameters to amd_sched_create

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: remove sched_lock

It isn't protecting anything.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: remove prepare_job callback

Not used any more.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: cleanup a scheduler function name

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: reorder scheduler functions

Keep run queue, entity and scheduler handling together.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: fix scheduler thread creation error checking

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: fix entity wakeup race condition

That actually didn't worked at all.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: cleanup entity picking

Cleanup function name, stop checking scheduler ready twice, but
check if kernel thread should stop instead.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: remove some more unused entity members v2

None of them are used any more.

v2: fix type in error message

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: fix user fences when scheduler is enabled

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drm/amdgpu: fix fence wait in sync_fence, instead should be in sync_rings

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>

drm/amdgpu: rework scheduler submission handling.

Remove active_hw_rq and it's protecting queue_lock, they are unused.

User 32bit atomic for hw_rq_count, 64bits for counting to three is a bit
overkill.

Cleanup the function name and remove incorrect comments.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>

drm/amdgpu: remove v_seq handling from the scheduler v2

Simply not used any more. Only keep 32bit atomic for fence sequence numbering.

v2: trivial rebase

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v1)
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com> (v1)
Reviewed-by: Chunming Zhou <david1.zhou@amd.com> (v1)

drm/amdgpu: improve sa_bo->fence by kernel fence

Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>

drm/amdgpu: modify amdgpu_fence_wait_any() to amdgpu_fence_wait_multiple()

Rename the function and update the related code with this modified function.
Add the new parameter of bool wait_all.

If wait_all is true, it will return when all fences are signaled or timeout.
If wait_all is false, it will return when any fence is signaled or timeout.

Signed-off-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Monk Liu <monk.liu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

drm/amdgpu: fix IH ring allocation for bus addresses (v2)

Use pci_alloc_consistent rather than kzalloc since we
need 256 byte aligned memory for the ring buffer.

v2: fix copy paste typo in free function noticed
by Jammy.

bug:
https://bugs.freedesktop.org/show_bug.cgi?id=91749

Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org