We have drm_crtc_vblank_crtc() to get the struct drm_vblank_crtc pointer
for a crtc. Use it instead of poking at dev->vblank[] directly.
However, we also need to get the crtc to start with. We could use
drm_crtc_from_index(), but refactor to use drm_for_each_crtc() instead.
This is all a bit tedious, and perhaps the driver shouldn't be poking at
vblank->enabled directly in the first place. But at least hide away the
dev->vblank[] access in drm_vblank.c where it belongs.
Cc: Patrik Jakobsson <patrik.r.jakobsson@gmail.com>
Acked-by: Patrik Jakobsson <patrik.r.jakobsson@gmail.com>
Link: https://patch.msgid.link/27b2c6772c68120d0d5ec28477db0d993743e955.1762513240.git.jani.nikula@intel.com
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Pull Kbuild fixes from Nathan Chancellor:
- Strip trailing padding bytes from modules.builtin.modinfo to fix
error during modules_install with certain versions of kmod
- Drop unused static inline function warning in .c files with clang
from W=1 to W=2
- Ensure kernel-doc.py invocations use the PYTHON3 make variable to
ensure user's choice of Python interpreter is always respected
* tag 'kbuild-fixes-6.18-3' of git://git.kernel.org/pub/scm/linux/kernel/git/kbuild/linux:
kbuild: Let kernel-doc.py use PYTHON3 override
compiler_types: Move unused static inline functions warning to W=2
kbuild: Strip trailing padding bytes from modules.builtin.modinfo
It is possible to force a specific version of python to be used when
building the kernel by passing PYTHON3= on the make command line.
However kernel-doc.py is currently called with python3 hard-coded and
thus ignores this setting.
Use $(PYTHON3) to run $(KERNELDOC) so that the desired version of
python is used.
Signed-off-by: Jean Delvare <jdelvare@suse.de>
Reviewed-by: Nicolas Schier <nsc@kernel.org>
Reviewed-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Link: https://patch.msgid.link/20251107192933.2bfe9e57@endymion
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
This reverts commit ebe7556050.
Tested the latest kernel on my GB203 and this seems to break it somehow.
Nov 09 04:16:14 bighp kernel: nouveau 0000:02:00.0: gsp: GSP-FMC boot failed (mbox: 0x0000000b)
Nov 09 04:16:14 bighp kernel: nouveau 0000:02:00.0: gsp: init failed, -5
Nov 09 04:16:14 bighp kernel: nouveau 0000:02:00.0: init failed with -5
Nov 09 04:16:14 bighp kernel: nouveau: drm:00000000:00000080: init failed with -5
Nov 09 04:16:14 bighp kernel: nouveau 0000:02:00.0: drm: Device allocation failed: -5
Nov 09 04:16:14 bighp kernel: nouveau 0000:02:00.0: probe with driver nouveau failed with error -5
Not sure why, I went over the patch and thought it should have worked, but there must be some
32-bit problem maybe in the FMC boot path.
Signed-off-by: Dave Airlie <airlied@redhat.com>
There are a few remaining cases where we *do* want to use `as`,
because we specifically want to strip the data that does not fit into
the destination type. Comment these uses to clear confusion about the
intent.
Acked-by: Danilo Krummrich <dakr@kernel.org>
[acourbot@nvidia.com: fix merge conflicts after rebase.]
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Message-ID: <20251029-nova-as-v3-6-6a30c7333ad9@nvidia.com>
Use the newly-introduced `num` module to replace the use of `as`
wherever it is safe to do. This ensures that a given conversion cannot
lose data if its source or destination type ever changes.
Acked-by: Danilo Krummrich <dakr@kernel.org>
[acourbot@nvidia.com: fix merge conflicts after rebase.]
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Message-ID: <20251029-nova-as-v3-5-6a30c7333ad9@nvidia.com>
The core library's `From` implementations do not cover conversions
that are not portable or future-proof. For instance, even though it is
safe today, `From<usize>` is not implemented for `u64` because of the
possibility to support larger-than-64bit architectures in the future.
However, the kernel supports a narrower set of architectures, and in the
case of Nova we only support 64-bit. This makes it helpful and desirable
to provide more infallible conversions, lest we need to rely on the `as`
keyword and carry the risk of silently losing data.
Thus, introduce a new module `num` that provides safe const functions
performing more conversions allowed by the build target, as well as
`FromSafeCast` and `IntoSafeCast` traits that are just extensions of
`From` and `Into` to conversions that are known to be lossless.
Suggested-by: Danilo Krummrich <dakr@kernel.org>
Link: https://lore.kernel.org/rust-for-linux/DDK4KADWJHMG.1FUPL3SDR26XF@kernel.org/
Acked-by: Danilo Krummrich <dakr@kernel.org>
[acourbot@nvidia.com: fix merge conflicts after rebase.]
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Message-ID: <20251029-nova-as-v3-4-6a30c7333ad9@nvidia.com>
Instead of trying very hard to find the largest fair number of GuC
doorbell IDs that could be allocated for VFs on the current GT, pick
some smaller rounded down to power-of-two value that is more likely
to be provisioned in the same manner by the other PF instance:
num VFs | num doorbells
--------+--------------
63..32 | 4
31..16 | 8
15..8 | 16
7..4 | 32
3..2 | 64
1 | 128 (regular PF)
1 | 240 (admin only PF)
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Piotr Piórkowski <piotr.piorkowski@intel.com>
Link: https://patch.msgid.link/20251105183253.863-3-michal.wajdeczko@intel.com
Instead of trying very hard to find the largest fair number of GuC
context IDs that could be allocated for VFs on the current GT, pick
some smaller rounded down to power-of-two value that is more likely
to be provisioned in the same manner by the other PF instance:
num VFs | num contexts
--------+-------------
63..32 | 1024
31..16 | 2048
15..8 | 4096
7..4 | 8192
3..2 | 16384
1 | 32768 (regular PF)
1 | 64512 (admin only PF)
Add also helper function to determine if the PF is admin-only,
and for now use .probe_display flag for that.
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Piotr Piórkowski <piotr.piorkowski@intel.com>
Link: https://patch.msgid.link/20251105183253.863-2-michal.wajdeczko@intel.com
The LT PHY code is abusing intel_de_wait_custom() in all kinds of weird
ways. Get rid of the weird fast timeouts, and just use the slow ones.
For consistency with intel_wait_for_register() we'll stick to the
default 2 usec fast timeout for all cases.
Someone really needs to properly document where all these magic numbers
came from...
This will let us eventually nuke intel_de_wait_custom() and convert
over to poll_timeout_us().
v2: Go for the longer (ms) timeout in case it actually matters
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251106152049.21115-9-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
There was a completely unjustified change to the cx0 powerdown
timeout, and the way it was done now prevents future conversion
to poll_timeout_us().
Assuming there was some reason the bigger timeout let's nuke
the old short timeout (XELPDP_PORT_POWERDOWN_UPDATE_TIMEOUT_US)
nd replace it with the bigger timeout
(XELPDP_PORT_POWERDOWN_UPDATE_TIMEOUT_MS).
For consistency with intel_wait_for_register() we'll stick to the
default 2 usec for the fast timeout.
v2: Go for the longer (ms) timeout in case it actually matters
v3: Note the defaullt 2 usec fast timeout (Jani)
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251106152049.21115-5-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
The current attempted split between xe/i915 vs. display
for intel_frontbuffer is a mess:
- the i915 rcu leaks through the interface to the display side
- the obj->frontbuffer write-side is now protected by a display
specific spinlock even though the actual obj->framebuffer
pointer lives in a i915 specific structure
- the kref is getting poked directly from both sides
- i915_active is still on the display side
Clean up the mess by moving everything about the frontbuffer
lifetime management to the i915/xe side:
- the rcu usage is now completely contained in i915
- frontbuffer_lock is moved into i915
- kref is on the i915/xe side (xe needs the refcount as well
due to intel_frontbuffer_queue_flush()->intel_frontbuffer_ref())
- the bo (and its refcounting) is no longer on the display side
- i915_active is contained in i915
I was pondering whether we could do this in some kind of smaller
steps, and perhaps we could, but it would probably have to start
with a bunch of reverts (which for sure won't go cleanly anymore).
So not convinced it's worth the hassle.
Acked-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251016185408.22735-10-ville.syrjala@linux.intel.com
intel_bo_frontbuffer_flush_for_display() is a bit too low level
to be directly in the high level dirtyfb code. Move the calls
into intel_frontbuffer_flush().
There is a slight behavioural change here in that we now skip
the flush if the bo is not a current scanout buffer (front->bits
== 0). But that is fine as the flush will eventually happen via
the fb pinning code if/when the bo becomes a scanout buffer again.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251016185408.22735-6-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Get rid of intel_frontbuffer_flip_{prepare,complete}() (and
the accompanying flip_bits) since they are unused.
I suppose these could technically provide a minor optimization
over intel_frontbuffer_flip() in that the flush would get
deferred further if new rendering were to sneak in between the
prepare() and complete() calls. But for correctness it should
not make any difference since another flush will anyway follow
once the new rendering finishes.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251016185408.22735-4-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Get rid of intel_frontbuffer_flip_{prepare,complete}() from
the overlay code and just use intel_frontbuffer_flip() instead.
The only difference between these are the light interactions
with the ORIGIN_CS busyness tracking, but since the only user
of this is the overlay/xf86-video-intel/Xv the buffer will
always be filled by the CPU and thus we'll never see any
ORIGIN_CS frontbuffer activity there anyway. Also I don't
think we actually have anything covered by the frontbuffer
tracking that affects the overlay (FBC is on the primary
plane, DRRS isn't currently enabled on the platforms with
overlay, and PSR doesn't exist in the hardware).
Acked-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251016185408.22735-3-ville.syrjala@linux.intel.com
This patch solves one of the existing mentions of COHA, a task
in the Nova task list about improving the `CoherentAllocation` API.
It uses the `write` method from `CoherentAllocation`.
Signed-off-by: Daniel del Castillo <delcastillodelarosadaniel@gmail.com>
[acourbot@nvidia.com: set prefix to "gpu: nova-core:".]
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Message-ID: <20251104193756.57726-3-delcastillodelarosadaniel@gmail.com>
Some comments that already existed didn't start with a capital
letter, this patch fixes that.
Signed-off-by: Daniel del Castillo <delcastillodelarosadaniel@gmail.com>
[acourbot@nvidia.com: set prefix to "gpu: nova-core:".]
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Message-ID: <20251104193756.57726-2-delcastillodelarosadaniel@gmail.com>
This patch solves one of the existing mentions of COHA, a task
in the Nova task list about improving the `CoherentAllocation` API.
It uses the new `from_bytes` method from the `FromBytes` trait as
well as the `as_slice` and `as_slice_mut` methods from
`CoherentAllocation`.
Signed-off-by: Daniel del Castillo <delcastillodelarosadaniel@gmail.com>
[acourbot@nvidia.com: set prefix to "gpu: nova-core:".]
[acourbot@nvidia.com: fix merge conflict after imports refactor.]
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Message-ID: <20251104193756.57726-1-delcastillodelarosadaniel@gmail.com>
As per [1], we need one "use" item per line, in order to reduce merge
conflicts. Furthermore, we need a trailing ", //" in order to tell
rustfmt(1) to leave it alone.
This does that for the entire nova-core driver.
[1] https://docs.kernel.org/rust/coding-guidelines.html#imports
Acked-by: Danilo Krummrich <dakr@kernel.org>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
[acourbot@nvidia.com: remove imports already in prelude as pointed out
by Danilo.]
[acourbot@nvidia.com: remove a few unneeded trailing `//`.]
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Message-ID: <20251107021006.434109-1-jhubbard@nvidia.com>
First, we can't assume pipe == crtc index. If a pipe is fused off in
between, it no longer holds. intel_crtc_for_pipe() is the only proper
way to get from a pipe to the corresponding crtc.
Second, drivers aren't supposed to access or index drm->vblank[]
directly. There's drm_crtc_vblank_crtc() for this.
Use both functions to fix the pipe to vblank conversion.
Fixes: f02658c46c ("drm/i915/psr: Add mechanism to notify PSR of pipe enable/disable")
Cc: Jouni Högander <jouni.hogander@intel.com>
Cc: stable@vger.kernel.org # v6.16+
Reviewed-by: Jouni Högander <jouni.hogander@intel.com>
Link: https://patch.msgid.link/20251106200000.1455164-1-jani.nikula@intel.com
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Prevent application hangs caused by out-of-order fence signaling when
user fences are attached. Use drm_syncobj (via dma-fence-chain) to
guarantee that each user fence signals in order, regardless of the
signaling order of the attached fences. Ensure user fence writebacks to
user space occur in the correct sequence.
v7:
- Skip drm_syncbj create of error (CI)
Fixes: dd08ebf6c3 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Link: https://patch.msgid.link/20251031234050.3043507-2-matthew.brost@intel.com
(cherry picked from commit adda4e855a)
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Currently, wait_event_timeout() in drm_wait_one_vblank() uses a 100ms
timeout. Under heavy scheduling pressure or rare delayed vblank
handling, this can trigger WARNs unnecessarily.
Increase the timeout to 1000ms to reduce spurious WARNs, while still
catching genuine issues.
Reported-by: syzbot+147ba789658184f0ce04@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=147ba789658184f0ce04
Tested-by: syzbot+147ba789658184f0ce04@syzkaller.appspotmail.com
Signed-off-by: Chintan Patel <chintanlike@gmail.com>
v2:
- Dropped unnecessary in-code comment (suggested by Thomas Zimmermann)
- Removed else branch, only log timeout case
v3:
- Replaced drm_dbg_kms()/manual logging with drm_err() (suggested by Ville Syrjälä)
- Removed unnecessary curr = drm_vblank_count() (suggested by Thomas Zimmermann)
- Fixed commit message wording ("invalid userspace calls" → "delayed vblank handling")
v4:
- Keep the original drm_WARN() to catch genuine kernel issues
- Increased timeout from 100ms → 1000ms to reduce spurious WARNs (suggested by Thomas Zimmermann)
Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://patch.msgid.link/20251028034337.6341-1-chintanlike@gmail.com
In the general workqueue implementation, if a user enqueues a work item
using schedule_delayed_work() the used wq is "system_wq" (per-cpu wq)
while queue_delayed_work() use WORK_CPU_UNBOUND (used when a cpu is not
specified). The same applies to schedule_work() that is using system_wq
and queue_work(), that makes use again of WORK_CPU_UNBOUND.
This lack of consistency cannot be addressed without refactoring the API.
For more details see the Link tag below.
This continues the effort to refactor worqueue APIs, which has begun
with the change introducing new workqueues and a new alloc_workqueue flag:
commit 128ea9f6cc ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566 ("workqueue: Add new WQ_PERCPU flag")
Use the successor of system_wq, system_percpu_wq, for the scheduler's
default timeout_wq. system_wq will be removed in a few release cycles.
Link: https://lore.kernel.org/all/20250221112003.1dSuoGyc@linutronix.de/
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Signed-off-by: Philipp Stanner <phasta@kernel.org>
Link: https://patch.msgid.link/20251106150121.256367-1-marco.crivellari@suse.com
