Since I wrote the below patch if you run a debug kernel you can a
dma debug warning like:
nouveau 0000:1f:00.0: DMA-API: device driver tries to sync DMA memory it has not allocated [device address=0x000000016e012000] [size=4096 bytes]
The old nouveau code wasn't consolidate the pages like the ttm code,
but the dma-debug expects the sync code to give it the same base/range
pairs as the allocator.
Fix the nouveau sync code to consolidate pages before calling the
sync code.
Fixes: bd549d35b4 ("nouveau: use ttm populate mapping functions. (v2)")
Reported-by: Lyude Paul <lyude@redhat.com>
Reviewed-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/417588/
Noticed that I wasn't paying close enough attention the last time I looked
at our audio callbacks, as I completely missed the fact that we were
figuring out which audio-enabled connector goes to each encoder by checking
it's state, but without grabbing any of the appropriate modesetting locks
to do so.
That being said however: trying to grab modesetting locks in our audio
callbacks would be very painful due to the potential for locking inversion
between HDA and DRM. So, let's instead just copy what i915 does again - add
our own audio lock to protect audio related state, and store each audio
enabled connector in each nouveau_encoder struct so that we don't need to
check any atomic states.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
drm_encoder->crtc is deprecated for atomic drivers, but
nouveau_encoder->crtc is safe.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Despite being an atomic driver, nouveau has a lot of leftover code that
relies on retrieving information regarding the new atomic state from
members of drm_encoder and drm_crtc. The first field being used,
drm_encoder.crtc, is deprecated for atomic drivers. The second field being
used is drm_crtc.state, which is only really sensible to use outside of an
atomic modeset.
So, add some helpers to lookup the current crtc for a given outp from the
atomic state. Then, convert most of the code in dispnv50/disp.c to use said
new helper, along with the relevant DRM atomic helpers for retrieving the
new encoder/crtc combinations for a new atomic state.
Note that we don't get rid of the nouveau_encoder.crtc field entirely for
three reasons:
- Legacy modesetting for pre-nv50 still uses it
- It doesn't cause any locking issues
- We need it for the HDA callbacks, as grabbing atomic modesetting locks in
those would be a mess.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Just to be more consistent with the order of args that DRM helpers like
drm_atomic_get_new_crtc_state() use.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
I have a strange dejavu feeling that I tried to submit a patch for this in
the past, but that it was rejected. I can't remember though, but I'm
further convinced this patch is the right thing to do anyway.
We label the to-be-committed head state in nv50_msto_atomic_enable() as
armh. Normally armh implies a state which is currently armed in hardware.
nv50_msto_atomic_enable() is called _after_ drm_atomic_swap_state()
however, but before the commit tail ends, which means that said state is
not actually armed on hardware.
As well - take note that this is the same convention followed in all of the
other atomic_enable() callbacks.
So, let's correct this to asyh.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
No functional changes, just change the function names to match the
callbacks they provide.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Noticed these in both the disable (which we'll be getting rid of in a
moment) and the atomic disable callbacks: both callback types check whether
or not there's actually a CRTC assigned to the given encoder. However, as
->atomic_disable and ->disable will never be called without a CRTC assigned
to the given encoder there's no point in this check. So just remove it.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Previous hardware allowed a MMU fault to be generated by software to
trigger a context switch for engine recovery. Turing has the capability
to preempt all work from a specific runlist processor and removed the
registers currently used for triggering MMU faults. Attempting to access
these non-existent registers results in further errors, so use the
runlist preemption register instead.
Signed-off-by: Alistair Popple <apopple@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Some of the low level FIFO interrupt status bits have changed for
Turing. Update the handling of these to match the hardware.
Signed-off-by: Alistair Popple <apopple@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Turing requires some changes to FIFO interrupt handling due to changes
in HW register layout. It also requires some changes to implement robust
channel (RC) recovery. This preparatory patch moves the functions
requiring changes into nvkm/engine/fifo/tu102.c so they can be altered
without affecting gk104 and other users. It should not introduce any
functional changes.
Signed-off-by: Alistair Popple <apopple@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
This is no longer needed now that tu102_mc_intr_stat has been updated to
look at the correct top-level interrupt bits.
Signed-off-by: Alistair Popple <apopple@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Turing reports MMU fault interrupts via new top level interrupt
registers. The old PMC MMU interrupt vector is not used by the HW. This
means we can remap the new top-level MMU interrupt to the exisiting PMC
MMU bit which simplifies the implementation until all interrupts are
moved over to using the new top level registers.
Signed-off-by: Alistair Popple <apopple@nvidia.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Since I'm almost certain I didn't get capability checking right for
pre-volta chipsets, let's start logging any caps we find to make things
like this obvious in the future.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
This was a mistake that was present before, but never got noticed until
we converted over to using nvidia's class headers for display
programming. Luckily though it never caused any problems, since we
always end up calling crc907d_set_src() after head907d_mode().
So, let's get rid of this.
Signed-off-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
While we do handle the additional cursor sizes introduced in NVE4, it looks
like we accidentally broke this when converting over to use Nvidia's
display headers. Since we now use NVVAL in dispnv50/head907d.c in order to
format the value for the cursor layout and NVD9 only had one byte reserved
vs. the 2 bytes reserved in later generations, we end up accidentally
stripping the second bit in the cursor layout format parameter - causing us
to set the wrong cursor size.
This fixes that by adding our own curs_set hook for 917d which uses the
NV917D headers.
Cc: Martin Peres <martin.peres@free.fr>
Cc: Jeremy Cline <jcline@redhat.com>
Cc: Simon Ser <contact@emersion.fr>
Cc: <stable@vger.kernel.org> # v5.9+
Signed-off-by: Lyude Paul <lyude@redhat.com>
Fixes: ed0b86a90b ("drm/nouveau/kms/nv50-: use NVIDIA's headers for core head_curs_set()")
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Fixes a crash when trying to create a channel on e.g. Turing GPUs when
NOUVEAU_SVM_INIT was called before.
Fixes: eeaf06ac1a ("drm/nouveau/svm: initial support for shared virtual memory")
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Commit f844eb485e introduced a regression for
NV50, which lead to visual artifacts, tearing and eventual crashes.
In the changes of f844eb485e only the first line
was correctly translated to the new NVIDIA header macros:
- PUSH_NVSQ(push, NV827C, 0x0110, 0,
- 0x0114, 0);
+ PUSH_MTHD(push, NV827C, SET_PROCESSING,
+ NVDEF(NV827C, SET_PROCESSING, USE_GAIN_OFS, DISABLE));
The lower part ("0x0114, 0") was probably omitted by accident.
This patch restores the push of the missing data and fixes the regression.
Signed-off-by: Bastian Beranek <bastian.beischer@rwth-aachen.de>
Fixes: f844eb485e ("drm/nouveau/kms/nv50-: use NVIDIA's headers for wndw image_set()")
Link: https://gitlab.freedesktop.org/drm/nouveau/-/issues/14
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
UEFI/RM no longer use IED scripts from the VBIOS, though they appear to
have been updated for use by the x86 VBIOS code, so we should be able to
continue using them for the moment.
Unfortunately, we require some hacks to do so, as the BeforeLinkTraining
IED script became a pointer to an array of scripts instead, without a
revbump of the relevant tables.
There's also some changes to SOR clock divider fiddling, which are
hopefully correct enough that things work as they should.
AFAIK, GA100 shouldn't have display, so it hasn't been added.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Fortunately, all the interrupts we need to bring up basic display support
are contained in a single leaf register, allowing this basic (but hackish)
implementation.
There's a bunch more invasive patches to come implementing all this in a
better/more complete way, but trying to get a minimal series out first.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
