- replaces the hacked-up version that existed solely to support TTM
v2. remove earlier hack preventing use of non-stall intr for fences
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
- replaces the hacked-up version that existed solely to support TTM
- noop until the next commit, adding proper support for ampere host
v2. fixup for ga103 early merge
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Exposes a bunch of the new features that became possible as a result
of the earlier commits. DRM will build on this in the future to add
support for features such as SCG ("async compute") and multi-device
rendering, as part of the work necessary to be able to write a half-
decent vulkan driver - finally.
For the moment, this just crudely ports DRM to the API changes.
- channel class interfaces now the same for all HW classes
- channel group class exposed (SCG)
- channel runqueue selector exposed (SCG)
- channel sub-device id control exposed (multi-device rendering)
- channel names in logging will reflect creating process, not fd owner
- explicit USERD allocation required by VOLTA_CHANNEL_GPFIFO_A and newer
- drm is smarter about determining the appropriate channel class to use
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
DRM uses this to setup fence-related items.
- nouveau_chan.runlist will always be "0" for the moment, not an issue
as GPUs prior to ampere have system-wide channel IDs,
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Adds the basic skeleton for common channel (group) interfaces.
- common behaviour between <gk104 and >=gk104 impl's
- separates priv/user channel objects
- passthrough to existing object for now, kludges removed later
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
This removes some now-unnecessary nesting of workqueues.
v2:
- use ?: (lyude)
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Also fixes vblank interrupts being left enabled when they're not meant
to be as a result of races/bugs in previous event handling code.
v2:
- use ?: (lyude)
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
This replaces the twisty, confusing, relationship between nvkm_event and
nvkm_notify with something much simpler, and less racey. It also places
events in the object tree hierarchy, which will allow a heap of the code
tracking events across allocation/teardown/suspend to be removed.
This commit just adds the new interfaces, and passes the owning subdev to
the event constructor to enable debug-tracing in the new code.
v2:
- use ?: (lyude)
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
This moves control of link retraining in response to HPD IRQ to the
KMS driver's HPD IRQ handler.
NVKM still handles checking link status for the moment, this can be
moved to the KMS driver when it takes explicit control of link rate
selection.
v2:
- skip source config on retrain (fixes some retrain failures)
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
This removes the need for NVKM to track DP HPD events, as the KMS
driver follows them already, and has better information available.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Aside from fixing MST->SST switching (KMS never turned off MST link config),
this should preserve existing behaviour for the moment, but provide a path
for the KMS driver to have more explicit control of the DP link, which has
been requested by Lyude.
More research into modeset/supervisor interactions is needed before we can
have fully explicit control from the KMS driver.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
v2:
- fix typo in sorhdmi/g84 struct initialiser (kbuild test robot)
v3:
- less convoluted flow control in nvkm_uoutp_mthd_acquire_tmds() (lyude)
v4:
- we don't support hdmi on original nv50, don't try
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
There are various pieces of information we pass to NVKM about the next
modeset, which are generally used while handling supervisor interrupts.
We had to start passing in some information about audio requirements a
while back to allocate an appropriate SOR in ACQUIRE, so we may as well
move all this type of information here for other protocols too.
Certain methods will be blocked on non-acquired outputs now, preventing
NULL pointer derefs from KMS driver bugs.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Will be used to more cleanly implement existing method interfaces that
take some confusing (IEDTkey, inherited from VBIOS, which RM no longer
uses on Ampere) match values to determine which display path to operate
on.
Methods will be protected from racing with supervisor, and from being
called where they shouldn't be (ie. without an OR assigned).
v2:
- use ?: (lyude)
v3:
- fix return code if noacquire() method fails
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
And use it to bail early in DP detection and avoid futile AUX transactions.
This could be used on other connector types too in theory, but it's not
something we've ever done before and I'd rather not risk breaking working
systems without looking into it more closely.
It's safe for DP though. We already do this by checking an AUX register
that contains HPD status and aborting the transaction. However, this is
much deeper in the stack - after taking various mutexes, poking HW for no
good reason, and making a mess in debug logs.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Will be used to provide more solid driver interfaces in general, but
the immediate motivation is work towards fixing issues with handling
hotplug/DP IRQ events.
Its use is currently limited to where we support non-polled hotplug
already (ie. any GPU since NV40ish era, where our DCB handling works
well enough), until that gets cleaned up someday.
v2:
- use ?: (lyude)
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Replaces a bunch of unnecessarily duplicated boilerplate in per-chipset
code with a simpler, common, implementation.
Channel "awaken" notify code is completely gone for now. KMS has never
made use of it so far, and event notify handling is about to be changed
in general anyway.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
About to expose head/output path/connector objects everywhere, so we will
need support for child classes prior to nv50 now.
Somewhat cleaner than the code >=nv50 used previously.
v2:
- use ?: (lyude)
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We don't currently have any kind of real acceleration on Ampere GPUs,
but the TTM memcpy() fallback paths aren't really designed to handle
copies between different devices, such as on Optimus systems, and
result in a kernel OOPS.
A few options were investigated to try and fix this, but didn't work
out, and likely would have resulted in a very unpleasant experience
for users anyway.
This commit adds just enough support for setting up a single channel
connected to a copy engine, which the kernel can use to accelerate
the buffer copies between devices. Userspace has no access to this
incomplete channel support, but it's suitable for TTM's needs.
A more complete implementation of host(fifo) for Ampere GPUs is in
the works, but the required changes are far too invasive that they
would be unsuitable to backport to fix this issue on current kernels.
v2: fix GPFIFO length in RAMFC (reported by Karol)
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Cc: Lyude Paul <lyude@redhat.com>
Cc: Karol Herbst <kherbst@redhat.com>
Cc: <stable@vger.kernel.org> # v5.12+
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Tested-by: Karol Herbst <kherbst@redhat.com>
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20210916220406.666454-1-skeggsb@gmail.com
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
No longer required now that userspace can't touch anything that might
need it, and should fix DRM MM operations racing with each other, and
the random hangs/crashes that come with that.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Long ago, there had been plans for making use of a bunch of these APIs
from userspace and there's various checks in place to stop misbehaving.
Countless other projects have occurred in the meantime, and the pieces
didn't finish falling into place for that to happen.
They will (hopefully) in the not-too-distant future, but it won't look
quite as insane. The super checks are causing problems right now, and
are going to be removed.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
I honestly don't even know why... These have never been used.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Some NVIDIA GPUs do not support direct atomic access to system memory via
PCIe. Instead this must be emulated by granting the GPU exclusive access
to the memory. This is achieved by replacing CPU page table entries with
special swap entries that fault on userspace access.
The driver then grants the GPU permission to update the page undergoing
atomic access via the GPU page tables. When CPU access to the page is
required a CPU fault is raised which calls into the device driver via MMU
notifiers to revoke the atomic access. The original page table entries
are then restored allowing CPU access to proceed.
Link: https://lkml.kernel.org/r/20210616105937.23201-11-apopple@nvidia.com
Signed-off-by: Alistair Popple <apopple@nvidia.com>
Reviewed-by: Ben Skeggs <bskeggs@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Peter Xu <peterx@redhat.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Shakeel Butt <shakeelb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
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>
GA100 hidden behind a module option, as it's not been as well verified
since initial bring-up and may need additional changes.
There's no display anyway, so this can wait for a bit.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
We had a, what was supposed to be temporary, hack in the KMS code where we'd
completely drain an EVO/NVD channel's push buffer when wrapping to the start
again, instead of treating it as a ring buffer.
Let's fix that, finally.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
This commit pulls in a bunch of new push buffer macros which are able to
support NVIDIA's class headers, and provide more useful debug output and
error checking (compile-time, where possible) than we had previously.
Will incrementally transition each function over to the unified interfaces.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
