If we were not specified a correct DTB, we will not be able to print
anything useful with panic() which requires a working console of some
sort. Utilize cfe_die() to callback into the CFE default UART and print
some useful diagnostics.
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com> Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Add a cfe_die() implementation which is useful when the kernel does an
early panic and no console is registered. This allows us to print
useful diagnostics such as an invalid DTB having been
configured/selected.
Since the BMIPS_GENERIC kernel can be built with support for multiple
processors, we need to do a runtime determination of the type of CPU
that we are executing on to perform the appropriate XKS01 disabling.
Since cfe_init() + cfe_die() could be conceivably called at very early
stages of the kernel boot, before cpu_probe(), we do not rely on the
structure(s) populated by cpu_probe().
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com> Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
The lp50xx LEDs expects to have single-color LED children with unit
addresses. This is required by the driver and provided by existing
DTSes. Fix the binding to match actual usage.
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org> Reviewed-by: Rob Herring <robh@kernel.org>
[This depends on previous schema-fixing patch, so I deleted Fixes; we
don't want to deal with this in stable] Signed-off-by: Pavel Machek <pavel@ucw.cz>
dt-bindings: leds: class-multicolor: reference class directly in multi-led node
The leds/common.yaml is referenced directly in each LED node, which
leads to people doing the same with leds/leds-class-multicolor.yaml.
This is not correct because leds-class-multicolor.yaml defined multi-led
property and its children. Some schemas implemented this incorrect.
Rework this to match same behavior common.yaml, so expect the multi-led
node to reference the leds-class-multicolor.yaml. Fixing allows to add
unevaluatedProperties:false.
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org> Reviewed-by: Rob Herring <robh@kernel.org> Signed-off-by: Pavel Machek <pavel@ucw.cz>
Broadcom used 2 LEDs hardware blocks for their BCM63xx SoCs:
1. Older one (BCM6318, BCM6328, BCM6362, BCM63268, BCM6838)
2. Newer one (BCM6848, BCM6858, BCM63138, BCM63148, BCM63381, BCM68360)
The newer one was also later also used on BCM4908 SoC.
Old block is already documented in the leds-bcm6328.yaml. This binding
documents the new one which uses different registers & programming. It's
first used in BCM63138 thus the binding name.
Signed-off-by: Rafał Miłecki <rafal@milecki.pl> Reviewed-by: Rob Herring <robh@kernel.org> Reviewed-by: Florian Fainelli <f.fainelli@gmail.com> Signed-off-by: Pavel Machek <pavel@ucw.cz>
Paul Elder [Tue, 14 Jun 2022 19:10:33 +0000 (20:10 +0100)]
media: rkisp1: debug: Add dump file in debugfs for MI main path registers
Add a register dump file in debugfs for some of the buffer-related
registers for the main path in MI (for the base address, the size, and
the offset). Also dump the corresponding shadow registers.
Laurent Pinchart [Tue, 14 Jun 2022 19:11:17 +0000 (20:11 +0100)]
media: rkisp1: Add infrastructure to support ISP features
Different ISP versions implement different sets of features. The driver
already takes the version into account in several places, but this
approach won't scale well for features that are found in different
versions. Introduce a new mechanism using a features bitmask in the
rkisp1_info structure to indicate which features the ISP support.
The first feature bit tells if the ISP has an internal CSI-2 receiver,
which is not available in all ISP versions.
Paul Elder [Tue, 14 Jun 2022 19:11:16 +0000 (20:11 +0100)]
media: rkisp1: Support the ISP parallel input
The ISP has a parallel input, exposed through port 1 in the device tree
node. While the driver supports configuring the ISP for the parallel and
BT.656 input modes, the DT parsing code, the subdev bound handler and
the ISP stream start handler only support the CSI input. Extend them to
support the parallel input.
Paul Elder [Tue, 14 Jun 2022 19:11:15 +0000 (20:11 +0100)]
media: dt-bindings: media: rkisp1: Add port for parallel interface
The rkisp1 can take an input on the parallel interface. Add a port for
it, and update the required field. At least one port is required, and
both may be specified.
Signed-off-by: Paul Elder <paul.elder@ideasonboard.com> Reviewed-by: Rob Herring <robh@kernel.org> Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com> Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Laurent Pinchart [Tue, 14 Jun 2022 19:11:13 +0000 (20:11 +0100)]
media: rkisp1: csi: Plumb the CSI RX subdev
Connect the CSI receiver subdevice between the sensors and the ISP. This
includes:
- Calling the subdevice via the v4l2 subdev API
- Moving the async notifier for the sensor from the ISP to the CSI
receiver
- In the ISP, create a media link to the CSI receiver, and remove the
media link creation to the sensor
- In the CSI receiver, create a media link to the sensor
Paul Elder [Tue, 14 Jun 2022 19:11:12 +0000 (20:11 +0100)]
media: rkisp1: csi: Implement a V4L2 subdev for the CSI receiver
The CSI receiver is a component separate from the ISP or the resizers.
It is actually optional, not all device model include a CSI receiver. On
some SoCs CSI-2 support can be provided through an external CSI-2
receiver, connected to the ISP's parallel input.
To support those use cases, create a V4L2 subdev to model the CSI
receiver. It will allow the driver to handle both internal and external
CSI receivers the same way.
The next commit will plumb the CSI subdev to the rest of the driver,
replacing direct function calls.
Laurent Pinchart [Tue, 14 Jun 2022 19:11:11 +0000 (20:11 +0100)]
media: rkisp1: isp: Disallow multiple active sources
The ISP supports multiple source subdevs, but can only capture from a
single one at a time. The source is selected through link setup. The
driver finds the active source in its .s_stream() handler using the
media_entity_remote_pad() function. This fails to reject invalid
configurations with multiple active sources. Fix it by using the
media_pad_remote_pad_unique() helper instead, and inline
rkisp1_isp_get_source() in rkisp1_isp_s_stream() as the function is
small and has a single caller.
Rename the rkisp1_get_remote_source() function to
rkisp1_isp_get_source() to use a consistent rkisp1_isp_* prefix for all
ISP functions, and drop the "remote" as the source can't be local.
Laurent Pinchart [Tue, 14 Jun 2022 19:11:04 +0000 (20:11 +0100)]
media: rkisp1: isp: Merge ISP_ACQ_PROP configuration in single variable
The rkisp1_config_isp() function stores the value of the input selection
and polarity configuration in two different local variables, OR'ed
together when writing the register. Merge them into a single acq_prop
variable.
Laurent Pinchart [Tue, 14 Jun 2022 19:11:03 +0000 (20:11 +0100)]
media: rkisp1: isp: Move input configuration to rkisp1_config_isp()
The ISP_ACQ_PROP register is set twice, once in rkisp1_config_isp() for
most of its fields, and once in rkisp1_config_dvp() (called from
rkisp1_config_path()) to configure the input selection field. Move the
latter to rkisp1_config_isp() to write the register once only, and drop
the now empty rkisp1_config_dvp() function.
Laurent Pinchart [Tue, 14 Jun 2022 19:11:02 +0000 (20:11 +0100)]
media: rkisp1: isp: Pass rkisp1_isp pointer to internal ISP functions
Replace the rkisp1_device pointer argument to the internal functions of
the ISP implementation with a rkisp1_isp object. This makes the code
flow more logical, as the functions operate on the ISP object.
Laurent Pinchart [Tue, 14 Jun 2022 19:11:01 +0000 (20:11 +0100)]
media: rkisp1: isp: Add rkisp1_device backpointer to rkisp1_isp
The rkisp1_isp structure documentation mentions a backpointer field to
rkisp1_device, but the field is missing. Add it, and use it to replace
more complicated constructs using container_of() on the v4l2_device.
Laurent Pinchart [Tue, 14 Jun 2022 19:10:59 +0000 (20:10 +0100)]
media: rkisp1: isp: Rename rkisp1_device.active_sensor to source
The active_sensor field of the rkisp1_device structure points to the ASD
data for the active source. The source may however not be a sensor, so
the naming is a bit confusing. Furthermore, the driver doesn't need to
access the full ASD from the active_sensor field, only the subdev
pointer is needed, when stopping streaming.
Rename the field to source, and turn it into a v4l2_subdev pointer.
Laurent Pinchart [Tue, 14 Jun 2022 19:10:58 +0000 (20:10 +0100)]
media: rkisp1: isp: Pass mbus type and flags to rkisp1_config_cif()
To prepare for the removal of the active_sensor field from the
rkisp1_device structure, pass the media bus type of flag to the
rkisp1_config_cif() function instead of accessing them through
active_sensor.
Laurent Pinchart [Tue, 14 Jun 2022 19:10:55 +0000 (20:10 +0100)]
media: rkisp1: csi: Pass sensor pointer to rkisp1_csi_config()
To prepare for the removal of the active_sensor field from the
rkisp1_device structure, pass the sensor pointer to the
rkisp1_csi_config() function instead of accessing it through
active_sensor.
Laurent Pinchart [Tue, 14 Jun 2022 19:10:52 +0000 (20:10 +0100)]
media: rkisp1: csi: Handle CSI-2 RX configuration fully in rkisp1-csi.c
The ISP layer now calls multiple functions of the CSI-2 RX layer to
configure, start and stop it, with the steps for the last two
operations. Move those calls to rkisp1_mipi_csi2_start() and
rkisp1_mipi_csi2_stop() to simplify the ISP code and the API exposed by
the CSI-2 receiver component.
Laurent Pinchart [Tue, 14 Jun 2022 19:10:51 +0000 (20:10 +0100)]
media: rkisp1: isp: Start CSI-2 receiver before ISP
Make sure the ISP is ready to receive data before starting the CSI-2
receiver by starting it first. Similarly, stop the CSI-2 receiver before
the ISP when stopping streaming.
Laurent Pinchart [Tue, 14 Jun 2022 19:10:49 +0000 (20:10 +0100)]
media: rkisp1: Fix sensor source pad retrieval at bound time
When a sensor is bound, its source pad is retrieved in the .bound()
operation with a call to media_entity_get_fwnode_pad(). The function
should be called with the source endpoint fwnode of the sensor, but is
instead called with the sensor's device fwnode.
Fix this, which involves storing a reference to the source endpoint
fwnode in the rkisp1_sensor_async structure, and thus implementing the
subdev notifier .destroy() operation to release the reference.
Laurent Pinchart [Tue, 14 Jun 2022 19:10:45 +0000 (20:10 +0100)]
media: rkisp1: Create link from sensor to ISP at notifier bound time
Links from sensors to the ISP can be created as sensors are bound. Move
the link creation from rkisp1_create_links() to the bound notifier, and
clean up the rkisp1_create_links() function while at it.
Laurent Pinchart [Tue, 14 Jun 2022 19:10:44 +0000 (20:10 +0100)]
media: rkisp1: Reject sensors without pixel rate control at bound time
The rkisp1 driver requires the sensor to implement the pixel rate
control. Trying to operate without it will cause an error when starting
streaming. Catch the issue earlier, at bound time.
Laurent Pinchart [Tue, 14 Jun 2022 19:10:43 +0000 (20:10 +0100)]
media: rkisp1: Move sensor .s_stream() call to ISP
Move the calls to the active sensor's .s_stream() operation to the ISP
subdev's .s_stream(). This groups all handling of the active sensor in
one place, preparing for a rework of that code.
Paul Elder [Tue, 14 Jun 2022 19:10:41 +0000 (20:10 +0100)]
media: rkisp1: Make rkisp1_isp_mbus_info common
The upcoming CSI receiver split from the ISP to a separate source file
will need to be able to access the list of formats supported by the
driver. Move it out of the ISP's header and into the common header, and
add helper functions for accessing it so that the format list doesn't
need to be stored in the header.
leds: pwm-multicolor: Don't show -EPROBE_DEFER as errors
When requesting a PWM it might return -EPROBE_DEFER if it hasn't probed
yet. This is not an error, so just propagate the -EPROBE_DEFER without
logging anything. There is already dev_err_probe for exactly this
situation.
Fixes: 9fa2762110dd ("leds: Add PWM multicolor driver") Signed-off-by: Emil Renner Berthing <emil.renner.berthing@canonical.com> Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be> Signed-off-by: Pavel Machek <pavel@ucw.cz>
Laurent Pinchart [Tue, 14 Jun 2022 19:10:40 +0000 (20:10 +0100)]
media: rkisp1: Access ISP version from info pointer
The ISP version is stored in the media_device hw_revision field and
access from there in the driver. Now that we store a pointer to the
match data in the rkisp1_device structure, access the ISP version from
there to make the code clearer and avoid depending on the media_device.
Paul Elder [Tue, 14 Jun 2022 19:10:39 +0000 (20:10 +0100)]
media: rkisp1: Save info pointer in rkisp1_device
To make it possible to use the rkisp1_info after probe time (for
instance to make code conditional on the ISP version), save it in the
main rkisp1_device structure. To achieve this, also move the info
structure into the common header, and document it.
While at it, drop a NULL check in rkisp1_probe() for the match data as
it can't happen.
Laurent Pinchart [Tue, 14 Jun 2022 19:10:38 +0000 (20:10 +0100)]
media: rkisp1: Rename rkisp1_match_data to rkisp1_info
The rkisp1_match_data structure contains device model-specific
information. It it referenced from OF match data, but that's an
implementation detail. Rename it to rkisp1_info to reflect its main
purpose.
Marc Zyngier [Sun, 17 Jul 2022 10:55:58 +0000 (11:55 +0100)]
Merge branch kvm-arm64/sysreg-cleanup-5.20 into kvmarm-master/next
* kvm-arm64/sysreg-cleanup-5.20:
: .
: Long overdue cleanup of the sysreg userspace access,
: with extra scrubbing on the vgic side of things.
: From the cover letter:
:
: "Schspa Shi recently reported[1] that some of the vgic code interacting
: with userspace was reading uninitialised stack memory, and although
: that read wasn't used any further, it prompted me to revisit this part
: of the code.
:
: Needless to say, this area of the kernel is pretty crufty, and shows a
: bunch of issues in other parts of the KVM/arm64 infrastructure. This
: series tries to remedy a bunch of them:
:
: - Sanitise the way we deal with sysregs from userspace: at the moment,
: each and every .set_user/.get_user callback has to implement its own
: userspace accesses (directly or indirectly). It'd be much better if
: that was centralised so that we can reason about it.
:
: - Enforce that all AArch64 sysregs are 64bit. Always. This was sort of
: implied by the code, but it took some effort to convince myself that
: this was actually the case.
:
: - Move the vgic-v3 sysreg userspace accessors to the userspace
: callbacks instead of hijacking the vcpu trap callback. This allows
: us to reuse the sysreg infrastructure.
:
: - Consolidate userspace accesses for both GICv2, GICv3 and common code
: as much as possible.
:
: - Cleanup a bunch of not-very-useful helpers, tidy up some of the code
: as we touch it.
:
: [1] https://lore.kernel.org/r/m2h740zz1i.fsf@gmail.com"
: .
KVM: arm64: Get rid or outdated comments
KVM: arm64: Descope kvm_arm_sys_reg_{get,set}_reg()
KVM: arm64: Get rid of find_reg_by_id()
KVM: arm64: vgic: Tidy-up calls to vgic_{get,set}_common_attr()
KVM: arm64: vgic: Consolidate userspace access for base address setting
KVM: arm64: vgic-v2: Add helper for legacy dist/cpuif base address setting
KVM: arm64: vgic: Use {get,put}_user() instead of copy_{from.to}_user
KVM: arm64: vgic-v2: Consolidate userspace access for MMIO registers
KVM: arm64: vgic-v3: Consolidate userspace access for MMIO registers
KVM: arm64: vgic-v3: Use u32 to manage the line level from userspace
KVM: arm64: vgic-v3: Convert userspace accessors over to FIELD_GET/FIELD_PREP
KVM: arm64: vgic-v3: Make the userspace accessors use sysreg API
KVM: arm64: vgic-v3: Push user access into vgic_v3_cpu_sysregs_uaccess()
KVM: arm64: vgic-v3: Simplify vgic_v3_has_cpu_sysregs_attr()
KVM: arm64: Get rid of reg_from/to_user()
KVM: arm64: Consolidate sysreg userspace accesses
KVM: arm64: Rely on index_to_param() for size checks on userspace access
KVM: arm64: Introduce generic get_user/set_user helpers for system registers
KVM: arm64: Reorder handling of invariant sysregs from userspace
KVM: arm64: Add get_reg_by_id() as a sys_reg_desc retrieving helper
Marc Zyngier [Mon, 4 Jul 2022 17:11:04 +0000 (18:11 +0100)]
KVM: arm64: Get rid or outdated comments
Once apon a time, the 32bit KVM/arm port was the reference, while
the arm64 version was the new kid on the block, without a clear
future... This was a long time ago.
Having kvm_arm_sys_reg_get_reg and co in kvm_host.h gives the
impression that these functions are free to be called from
anywhere.
Not quite. They really are tied to out internal sysreg handling,
and they would be better off in the sys_regs.h header, which is
private. kvm_host.h could also get a bit of a diet, so let's
just do that.
Marc Zyngier [Thu, 14 Jul 2022 07:10:09 +0000 (08:10 +0100)]
KVM: arm64: vgic: Tidy-up calls to vgic_{get,set}_common_attr()
The userspace accessors have an early call to vgic_{get,set}_common_attr()
that makes the code hard to follow. Move it to the default: clause of
the decoding switch statement, which results in a nice cleanup.
This requires us to move the handling of the pending table into the
common handling, even if it is strictly a GICv3 feature (it has the
benefit of keeping the whole control group handling in the same
function).
Also cleanup vgic_v3_{get,set}_attr() while we're at it, deduplicating
the calls to vgic_v3_attr_regs_access().
Suggested-by: Reiji Watanabe <reijiw@google.com> Signed-off-by: Marc Zyngier <maz@kernel.org>
Marc Zyngier [Tue, 5 Jul 2022 13:34:33 +0000 (14:34 +0100)]
KVM: arm64: vgic-v2: Add helper for legacy dist/cpuif base address setting
We carry a legacy interface to set the base addresses for GICv2.
As this is currently plumbed into the same handling code as
the modern interface, it limits the evolution we can make there.
Add a helper dedicated to this handling, with a view of maybe
removing this in the future.
Marc Zyngier [Tue, 5 Jul 2022 10:27:37 +0000 (11:27 +0100)]
KVM: arm64: vgic: Use {get,put}_user() instead of copy_{from.to}_user
Tidy-up vgic_get_common_attr() and vgic_set_common_attr() to use
{get,put}_user() instead of the more complex (and less type-safe)
copy_{from,to}_user().
Reviewed-by: Reiji Watanabe <reijiw@google.com> Signed-off-by: Marc Zyngier <maz@kernel.org>
Marc Zyngier [Tue, 5 Jul 2022 09:26:07 +0000 (10:26 +0100)]
KVM: arm64: vgic-v3: Consolidate userspace access for MMIO registers
For userspace accesses to GICv3 MMIO registers (and related data),
vgic_v3_{get,set}_attr are littered with {get,put}_user() calls,
making it hard to audit and reason about.
Consolidate all userspace accesses in vgic_v3_attr_regs_access(),
making the code far simpler to audit.
Reviewed-by: Reiji Watanabe <reijiw@google.com> Signed-off-by: Marc Zyngier <maz@kernel.org>
Marc Zyngier [Tue, 5 Jul 2022 07:11:54 +0000 (08:11 +0100)]
KVM: arm64: vgic-v3: Convert userspace accessors over to FIELD_GET/FIELD_PREP
The GICv3 userspace accessors are all about dealing with conversion
between fields from architectural registers and internal representations.
However, and owing to the age of this code, the accessors use
a combination of shift/mask that is hard to read. It is nonetheless
easy to make it better by using the FIELD_{GET,PREP} macros that solely
rely on a mask.
This results in somewhat nicer looking code, and is probably easier
to maintain.
Reviewed-by: Reiji Watanabe <reijiw@google.com> Signed-off-by: Marc Zyngier <maz@kernel.org>
Marc Zyngier [Mon, 4 Jul 2022 08:57:38 +0000 (09:57 +0100)]
KVM: arm64: vgic-v3: Make the userspace accessors use sysreg API
The vgic-v3 sysreg accessors have been ignored as the rest of the
sysreg internal API was evolving, and are stuck with the .access
method (which is normally reserved to the guest's own access)
for the userspace accesses (which should use the .set/.get_user()
methods).
Catch up with the program and repaint all the accessors so that
they fit into the normal userspace model, and plug the result into
the helpers that have been introduced earlier.
Reviewed-by: Reiji Watanabe <reijiw@google.com> Signed-off-by: Marc Zyngier <maz@kernel.org>
Marc Zyngier [Mon, 4 Jul 2022 07:07:44 +0000 (08:07 +0100)]
KVM: arm64: vgic-v3: Push user access into vgic_v3_cpu_sysregs_uaccess()
In order to start making the vgic sysreg access from userspace
similar to all the other sysregs, push the userspace memory
access one level down into vgic_v3_cpu_sysregs_uaccess().
The next step will be to rely on the sysreg infrastructure
to perform this task.
Reviewed-by: Reiji Watanabe <reijiw@google.com> Signed-off-by: Marc Zyngier <maz@kernel.org>
Finding out whether a sysreg exists has little to do with that
register being accessed, so drop the is_write parameter.
Also, the reg pointer is completely unused, and we're better off
just passing the attr pointer to the function.
This result in a small cleanup of the calling site, with a new
helper converting the vGIC view of a sysreg into the canonical
one (this is purely cosmetic, as the encoding is the same).
Reviewed-by: Reiji Watanabe <reijiw@google.com> Signed-off-by: Marc Zyngier <maz@kernel.org>
Marc Zyngier [Mon, 4 Jul 2022 16:55:43 +0000 (17:55 +0100)]
KVM: arm64: Get rid of reg_from/to_user()
These helpers are only used by the invariant stuff now, and while
they pretend to support non-64bit registers, this only serves as
a way to scare the casual reviewer...
Replace these helpers with our good friends get/put_user(), and
don't look back.
Reviewed-by: Reiji Watanabe <reijiw@google.com> Signed-off-by: Marc Zyngier <maz@kernel.org>
Marc Zyngier [Mon, 4 Jul 2022 16:27:00 +0000 (17:27 +0100)]
KVM: arm64: Consolidate sysreg userspace accesses
Until now, the .set_user and .get_user callbacks have to implement
(directly or not) the userspace memory accesses. Although this gives
us maximem flexibility, this is also a maintenance burden, making it
hard to audit, and I'd feel much better if it was all located in
a single place.
So let's do just that, simplifying most of the function signatures
in the process (the callbacks are now only concerned with the
data itself, and not with userspace).
Reviewed-by: Reiji Watanabe <reijiw@google.com> Signed-off-by: Marc Zyngier <maz@kernel.org>
Marc Zyngier [Sun, 3 Jul 2022 15:06:51 +0000 (16:06 +0100)]
KVM: arm64: Introduce generic get_user/set_user helpers for system registers
The userspace access to the system registers is done using helpers
that hardcode the table that is looked up. extract some generic
helpers from this, moving the handling of hidden sysregs into
the core code.
Reviewed-by: Reiji Watanabe <reijiw@google.com> Signed-off-by: Marc Zyngier <maz@kernel.org>
With CONFIG_RANDOMIZE_BASE=y vmlinux addresses will resolve incorrectly
from kallsyms. Fix this by adding the KASLR offset before printing the
symbols.
Laurent Pinchart [Tue, 14 Jun 2022 19:10:34 +0000 (20:10 +0100)]
media: rkisp1: Enable compilation on ARCH_MXC
The ISP used by the Rockchip RK3399 is also found in the NXP i.MX8MP.
Enable compilation of the driver for the MXC architecture in addition to
ARCH_ROCKCHIP.
Laurent Pinchart [Tue, 14 Jun 2022 19:11:10 +0000 (20:11 +0100)]
media: mc-entity: Add a new helper function to get a remote pad for a pad
The newly added media_entity_remote_source_pad_unique() helper function
handles use cases where the entity has a link enabled uniqueness
constraint covering all pads. There are use cases where the constraint
covers a specific pad only. Add a new media_pad_remote_pad_unique()
function to handle this. It operates as
media_entity_remote_source_pad_unique(), but on a given pad instead of
on the entity.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com> Acked-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com> Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Laurent Pinchart [Tue, 14 Jun 2022 19:11:09 +0000 (20:11 +0100)]
media: mc-entity: Add a new helper function to get a remote pad
The media_entity_remote_pad_first() helper function returns the first
remote pad it finds connected to a given pad. Beside being possibly
non-deterministic (as it stops at the first enabled link), the fact that
it returns the first match makes it unsuitable for drivers that need to
guarantee that a single link is enabled, for instance when an entity can
process data from one of multiple sources at a time.
For those use cases, add a new helper function,
media_entity_remote_pad_unique(), that operates on an entity and returns
a remote pad, with a guarantee that only one link is enabled. To ease
its use in drivers, also add an inline wrapper that locates source pads
specifically. A wrapper that locates sink pads can easily be added when
needed.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com> Acked-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com> Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Laurent Pinchart [Sat, 25 Jun 2022 17:02:24 +0000 (18:02 +0100)]
media: mc-entity: Rename media_entity_remote_pad() to media_pad_remote_pad_first()
The media_entity_remote_pad() is misnamed, as it operates on a pad and
not an entity. Rename it to media_pad_remote_pad_first() to clarify its
behaviour.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com> Reviewed-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com> Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Laurent Pinchart [Tue, 14 Jun 2022 19:10:48 +0000 (20:10 +0100)]
media: v4l2-async: Add notifier operation to destroy asd instances
Drivers typically extend the v4l2_async_subdev structure by embedding it
in a driver-specific structure, to store per-subdev custom data. The
v4l2_async_subdev instances are freed by the v4l2-async framework, which
makes this mechanism cumbersome to use safely when custom data needs
special treatment to be destroyed (such as freeing additional memory, or
releasing references to kernel objects).
To ease this, add a .destroy() operation to the
v4l2_async_notifier_operations structure. The operation is called right
before the v4l2_async_subdev is freed, giving drivers a chance to
destroy data if needed.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com> Reviewed-by: Hans Verkuil <hverkuil-cisco@xs4all.nl> Acked-by: Sakari Ailus <sakari.ailus@linux.intel.com> Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Jernej Skrabec [Mon, 20 Jun 2022 17:55:14 +0000 (18:55 +0100)]
media: cedrus: Add error handling for failed setup
During decoding setup stage for complex codecs like HEVC driver can
detect inconsistent values in controls or some other task, like
allocating memory, can fail.
Currently setup stage has no way of signalling error. Change return type
of setup callback to int and if returned value is not zero, skip
decoding and finish job immediately with error flag.
While currently there is only one place when setup can fail, it's
expected that there will be more such cases in the future, when HEVC
decoding is improved.
Jernej Skrabec [Mon, 20 Jun 2022 17:55:13 +0000 (18:55 +0100)]
media: cedrus: Improve error messages for controls
Currently error messages when control creation fails are very sparse.
Granted, user should never observe them. However, developer working on
codecs can. In such cases additional information like which control
creation failed and error number are very useful.
Jernej Skrabec [Mon, 20 Jun 2022 17:55:12 +0000 (18:55 +0100)]
media: cedrus: h265: Fix logic for not low delay flag
Now that we know real purpose of "not low delay" flag, logic for
applying this flag should be fixed too. According to vendor and
reference implementation, low delay is signaled when POC of current
frame is lower than POC of at least one reference of a slice.
Implement mentioned logic and invert it to conform to flag meaning. Also
don't apply flag for I frames. They don't have any reference.
Jernej Skrabec [Mon, 20 Jun 2022 17:55:11 +0000 (18:55 +0100)]
media: cedrus: h265: Fix flag name
Bit 21 in register 0x24 (slice header info 1) actually represents
negated version of low delay flag. This can be seen in vendor Cedar
library source code. While this flag is not part of the standard, it can
be found in reference HEVC implementation.
When allocating aux buffers for postprocessing, it's assumed that base
buffer size is the same as that of output. Coincidentally, that's true
most of the time, but not always. 10-bit source also needs aux buffer
size which is appropriate for 10-bit native format, even if the output
format is 8-bit. Similarly, mv sizes and other extra buffer size also
depends on source width/height, not destination.
In preparation for supporting 10-bit formats, add mechanism which will
filter formats based on pixel depth.
Hantro G2 supports only one decoding format natively and that is based
on bit depth of current video frame. Additionally, it makes no sense to
upconvert bitness, so filter those out too.
projects / people / ms / linux.git / log
