platform/kernel/linux-rpi.git
8 months agodrm/panel: jdi-lt070me05000: Use gpiod_set_value_cansleep
Dave Stevenson [Thu, 15 Apr 2021 16:30:35 +0000 (17:30 +0100)]
drm/panel: jdi-lt070me05000: Use gpiod_set_value_cansleep

There is no reason why the control GPIOs for the panel can not
be connected to I2C or similar GPIO interfaces that may need to
sleep, therefore switch from gpiod_set_value to
gpiod_set_value_cansleep calls to configure them.
Without that you get complaints from gpiolib every time the state
is changed.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agostaging: fbtft: Add minipitft13 variant
Phil Elwell [Fri, 19 Feb 2021 10:25:01 +0000 (10:25 +0000)]
staging: fbtft: Add minipitft13 variant

The Adafruit Mini-PiTFT13 display needs offsets applying when rotated,
so use the "variant" mechanism to select a custom set_addr_win method
using a dedicated compatible string of "fbtft,minipitft13".

See: https://github.com/raspberrypi/firmware/issues/1524

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agostaging/bcm2835-camera: Add support for DMABUFs
Dave Stevenson [Wed, 17 Mar 2021 12:34:57 +0000 (12:34 +0000)]
staging/bcm2835-camera: Add support for DMABUFs

DMABUFs are all handled by videobuf2, so there is no reason not
to enable support for them.

Note that this driver is still using the vmalloc allocator, so
the buffers it allocates will not be compatible with the codec
or ISP driver that require contiguous buffers. However this
driver should be able to import the buffers allocated by them.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agomedia: i2c: ov5647: Parse and register properties
Laurent Pinchart [Fri, 3 Jul 2020 22:45:08 +0000 (01:45 +0300)]
media: i2c: ov5647: Parse and register properties

Parse device properties and register controls for them using the V4L2
fwnode properties helpers.

Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
8 months agomedia: ov5647: Fix return codes from ov5647_write/ov5647_read functions.
David Plowman [Wed, 15 Jan 2020 13:40:38 +0000 (13:40 +0000)]
media: ov5647: Fix return codes from ov5647_write/ov5647_read functions.

Previously they were returning positive non-zero codes for success,
which were getting passed up the call stack. Since release 4.19,
do_dentry_open (fs/open.c) has been catching these and flagging an
error. (So this driver has been broken since that date.)

Fixes: 3c2472a [media] media: i2c: Add support for OV5647 sensor
Signed-off-by: David Plowman <david.plowman@raspberrypi.org>
Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agomedia: i2c: add ov9281 driver.
Zefa Chen [Fri, 17 May 2019 10:23:03 +0000 (18:23 +0800)]
media: i2c: add ov9281 driver.

Change-Id: I7b77250bbc56d2f861450cf77271ad15f9b88ab1
Signed-off-by: Zefa Chen <zefa.chen@rock-chips.com>
media: i2c: ov9281: fix mclk issue when probe multiple camera.

Takes the ov9281 part only from the Rockchip's patch.

Change-Id: I30e833baf2c1bb07d6d87ddb3b00759ab45a90e4
Signed-off-by: Zefa Chen <zefa.chen@rock-chips.com>
media: i2c: ov9281: add enum_frame_interval function for iq tool 2.2 and hal3

Adds the ov9281 parts of the Rockchip patch adding enum_frame_interval to
a large number of drivers.

Change-Id: I03344cd6cf278dd7c18fce8e97479089ef185a5c
Signed-off-by: Zefa Chen <zefa.chen@rock-chips.com>
media: i2c: ov9281: Fixup for recent kernel releases, and remove custom code

The Rockchip driver was based on a 4.4 kernel, and had several custom
Rockchip parts.

Update to 5.4 kernel APIs, with the relevant controls required by
libcamera, and remove custom Rockchip parts.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
media: i2c: ov9281: Read chip ID via 2 reads

Vision Components have made an OV9281 module which blocks reading
back the majority of registers to comply with NDAs, and in doing
so doesn't allow auto-increment register reading as used when
reading the chip ID.

Use two reads and manually combine the results.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
media: i2c: ov9281: Add support for 8 bit readout

The sensor supports 8 bit mode as well as 10bit, so add the
relevant code to allow selection of this.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
media: ov9281: Add 1280x720 and 640x480 modes

Breaks out common register set and adds the different registers
for 1280x720 (cropped) and 640x480 (skipped) modes

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
Fixed picture line bug in all ov9281 modes

Signed-off-by: Mathias Anhalt <mathiasanhalt@web.de>
Added hflip and vflip controls to ov9281

Signed-off-by: Mathias Anhalt <mathiasanhalt@web.de>
media: i2c: ov9281: Remove override of subdev name

From the original Rockchip driver, the subdev was renamed
from the default to being "mov9281 <dev_name>" whereas the
default would have been "ov9281 <dev_name>".

Remove the override to drop back to the default rather than
a vendor custom string.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
media: v4l2-subdev: add subdev-wide state struct

Signed-off-by: Dom Cobley <popcornmix@gmail.com>
media: i2c: ov9281: Add fwnode properties controls

Add call to v4l2_ctrl_new_fwnode_properties to read and
create the fwnode based controls.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
media: i2c: ov9281: Sensor should report RAW color space

Tested on Raspberry Pi running libcamera.

Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
Partial revert "media: i2c: add ov9281 driver."

This partially reverts commit 84e98e3a4f3eecb168ceb80231c3e8252929892e.

The commit had merged some changes to other drivers with adding the ov9281
driver. Only the ov9281 parts have been reverted.

8 months agomedia/v4l2_m2m: In buffered mode run jobs if either port is streaming
Dave Stevenson [Mon, 1 Feb 2021 18:48:47 +0000 (18:48 +0000)]
media/v4l2_m2m: In buffered mode run jobs if either port is streaming

In order to get the intended behaviour of the stateful video
decoder API where only the OUTPUT queue needs to be enabled and fed
buffers in order to get the SOURCE_CHANGED event that configures the
CAPTURE queue, we want the device to run should either queue be
streaming.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agoAssign crypto aliases to different AES implementation modules
Ben Avison [Mon, 8 Mar 2021 15:32:25 +0000 (15:32 +0000)]
Assign crypto aliases to different AES implementation modules

The kernel modules aes-neon-blk and aes-neon-bs perform poorly, at least on
Cortex-A72 without crypto extensions. In fact, aes-arm64 outperforms them
on benchmarks, despite it being a simpler implementation (only accelerating
the single-block AES cipher).

For modes of operation where multiple cipher blocks can be processed in
parallel, aes-neon-bs outperforms aes-neon-blk by around 60-70% and aes-arm64
is another 10-20% faster still. But the difference is even more marked with
modes of operation with dependencies between neighbouring blocks, such as
CBC encryption, which defeat parallelism: in these cases, aes-arm64 is
typically around 250% faster than either aes-neon-blk or aes-neon-bs.

The key trade-off with aes-arm64 is that the look-up tables are situated in
RAM. This leaves them potentially open to cache timing attacks. The two other
modules, by contrast, load the look-up tables into NEON registers and so are
able to perform in constant time.

This patch aims to load aes-arm64 more often.

If none of the currently-loaded crypto modules implement a given algorithm,
a new one is typically selected for loading using a platform-neutral alias
describing the required algorithm. To enable users to still
load aes-neon-blk or aes-neon-bs if they really want them, while still
ensuring that aes-arm64 is usually selected, remove the aliases from
aes-neonbs-glue.c and aes-glue.c and apply them to aes-cipher-glue.c, but
still build the two NEON modules.

Since aes-glue.c can also be used to build aes-ce-blk, leave them enabled
if USE_V8_CRYPTO_EXTENSIONS is defined, to ensure they are selected if we
in future use a CPU which has the crypto extensions enabled.

Note that the algorithm priority specifiers are unchanged, so if
aes-neon-bs is loaded at the same time as aes-arm64, the former will be
used in preference. However, aes-neon-blk and aes-arm64 have tied priority,
so whichever module was loaded first will be used (assuming aes-neon-bs is
not loaded).

Signed-off-by: Ben Avison <bavison@riscosopen.org>
8 months agomedia: bcm2835-unicam: Fix bug in buffer swapping logic
Naushir Patuck [Fri, 5 Mar 2021 15:40:45 +0000 (15:40 +0000)]
media: bcm2835-unicam: Fix bug in buffer swapping logic

If multiple sets of interrupts occur simultaneously, it may be unsafe
to swap buffers, as the hardware may already be re-using the current
buffers. In such cases, avoid swapping buffers, and wait for the next
opportunity at the Frame End interrupt to signal completion.

Additionally, check the packet compare status when watching for frame
end for buffers swaps, as this could also signify a frame end event.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agobcm2835-isp: Allow formats with different colour spaces.
David Plowman [Tue, 12 Jan 2021 13:55:39 +0000 (13:55 +0000)]
bcm2835-isp: Allow formats with different colour spaces.

Each supported format now includes a mask showing the allowed colour
spaces, as well as a default colour space for when one was not
specified.

Additionally we translate the colour space to mmal format and pass it
over to the VideoCore.

Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
8 months agokbuild: Silence unavoidable dtc overlay warnings
Phil Elwell [Fri, 29 Jan 2021 10:34:11 +0000 (10:34 +0000)]
kbuild: Silence unavoidable dtc overlay warnings

Much effort has been put into finding ways to avoid warnings from dtc
about overlays, usually to do with the presence of #address-cells and
size-cells, but not exclusively so. Since the issues being warned about
are harmless, suppress the warnings to declutter the build output and
to avoid alarming users.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agospi: bcm2835: Workaround/fix for zero-length transfers
Phil Elwell [Thu, 28 Jan 2021 11:30:04 +0000 (11:30 +0000)]
spi: bcm2835: Workaround/fix for zero-length transfers

A relatively recent commit ([1]) contained optimisation for the PIO
SPI FIFO-filling functions. The commit message includes the phrase
"[t]he blind and counted loops are always called with nonzero count".
This is technically true, but it is still possible for count to become
zero before the loop is entered - if tfr->len is zero. Moving the loop
exit condition to the end of the loop saves a few cycles, but results
in a near-infinite loop should the revised count be zero on entry.

Strangely, zero-lengthed transfers aren't filtered by the SPI framework
and, even more strangely, the Python3 spidev library is triggering them
for no obvious reason.

Avoid the problem completely by bailing out of the main transfer
function early if trf->len is zero, although there may be a case for
moving the mitigation into the framework.

See: https://github.com/raspberrypi/linux/issues/4100

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
[1] 26751de25d25 ("spi: bcm2835: Micro-optimise FIFO loops")

8 months agostaging: vc04_services: ISP: Add colour denoise control
Naushir Patuck [Thu, 14 Jan 2021 09:20:52 +0000 (09:20 +0000)]
staging: vc04_services: ISP: Add colour denoise control

Add colour denoise control to the bcm2835 driver through a new v4l2
control: V4L2_CID_USER_BCM2835_ISP_CDN.

Add the accompanying MMAL configuration structure definitions as well.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agouapi: bcm2835-isp: Add colour denoise configuration
Naushir Patuck [Thu, 14 Jan 2021 09:18:42 +0000 (09:18 +0000)]
uapi: bcm2835-isp: Add colour denoise configuration

Add a configuration structure for colour denoise to the bcm2835_isp
driver.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agostaging/bcm2835-isp: Log the number of excess supported formats
Dave Stevenson [Thu, 7 Jan 2021 11:43:22 +0000 (11:43 +0000)]
staging/bcm2835-isp: Log the number of excess supported formats

When logging that the firmware has provided more supported formats
than we had allocated storage for, log the number allocated and
returned.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agostaging/bcm2835-isp: Add the unpacked (16bpp) raw formats
Dave Stevenson [Thu, 7 Jan 2021 11:37:10 +0000 (11:37 +0000)]
staging/bcm2835-isp: Add the unpacked (16bpp) raw formats

Now that the firmware supports the unpacked (16bpp) variants
of the MIPI raw formats, add the mappings.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agostaging/vc04_services: Add additional unpacked raw formats
Dave Stevenson [Thu, 7 Jan 2021 10:43:20 +0000 (10:43 +0000)]
staging/vc04_services: Add additional unpacked raw formats

Support has been added for the unpacked (16bpp) versions of
the MIPI raw 10, 12, and 14 formats, so add the 4CCs for them.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agonet: lan78xx: Ack pending PHY ints when resetting
Phil Elwell [Tue, 15 Dec 2020 16:38:37 +0000 (16:38 +0000)]
net: lan78xx: Ack pending PHY ints when resetting

lan78xx_link_reset explicitly clears the MAC's view of the PHY's IRQ
status. In doing so it potentially leaves the PHY with a pending
interrupt that will never be acknowledged, at which point no further
interrupts will be generated.

Avoid the problem by acknowledging any pending PHY interrupt after
clearing the MAC's status bit.

See: https://github.com/raspberrypi/linux/issues/2937

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agoPCI: brcmstb: Advertise MSI-X support
Phil Elwell [Thu, 3 Dec 2020 13:44:42 +0000 (13:44 +0000)]
PCI: brcmstb: Advertise MSI-X support

Although the BRCMSTB PCIe interface doesn't technically support the
MSI-X spec, in practise it seems to work provided no more than 32
MSI-Xs are required. Add the required flag to the driver to allow
experimentation with devices that demand MSI-X support.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Clear clock state when stopping streaming
Naushir Patuck [Wed, 2 Dec 2020 16:48:41 +0000 (16:48 +0000)]
media: bcm2835-unicam: Clear clock state when stopping streaming

Commit 65e08c465020d4c5b51afb452efc2246d80fd66f failed to clear the
clock state when the device stopped streaming. Fix this, as it might
again cause the same problems when doing an unprepare.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Return early from stop_streaming() if stopped
Naushir Patuck [Wed, 2 Dec 2020 15:26:09 +0000 (15:26 +0000)]
media: bcm2835-unicam: Return early from stop_streaming() if stopped

clk_disable_unprepare() is called unconditionally in stop_streaming().
This is incorrect in the cases where start_streaming() fails, and
unprepares all clocks as part of the failure cleanup. To avoid this,
ensure that clk_disable_unprepare() is only called in stop_streaming()
if the clocks are in a prepared state.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Correctly handle error propagation for stream on
Naushir Patuck [Wed, 2 Dec 2020 15:22:23 +0000 (15:22 +0000)]
media: bcm2835-unicam: Correctly handle error propagation for stream on

On a failure in start_streaming(), the error code would not propagate to
the calling function on all conditions. This would cause the userland
caller to not know of the failure.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agodt-bindings: Add compatible for BCM2711 DSI1
Dave Stevenson [Thu, 12 Nov 2020 17:01:52 +0000 (17:01 +0000)]
dt-bindings: Add compatible for BCM2711 DSI1

DSI1 on BCM2711 doesn't require the DMA workaround that is used
on BCM2835/6/7, therefore it needs a new compatible string.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agodrm/panel/raspberrypi-ts: Insert delay before polling for startup state
Dave Stevenson [Tue, 10 Nov 2020 11:21:56 +0000 (11:21 +0000)]
drm/panel/raspberrypi-ts: Insert delay before polling for startup state

In switching to the hardware I2C controller there is an issue
where we seem to not get back the correct state from the Pi
touchscreen.
Insert a delay before polling to avoid this condition.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agodrm/panel/raspberrypi-touchscreen: Use independent I2C actions with delay.
Dave Stevenson [Thu, 23 Apr 2020 09:17:18 +0000 (10:17 +0100)]
drm/panel/raspberrypi-touchscreen: Use independent I2C actions with delay.

We now have the hardware I2C controller pinmuxed to the drive the
display I2C, but this controller does not support clock stretching.
The Atmel micro-controller in the panel requires clock stretching
to allow it to prepare any data to be read.

Split the rpi_touchscreen_i2c_read into two independent transactions with
a delay between them for the Atmel to prepare the data.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agoInput: edt-ft5x06: Poll the device if no interrupt is configured.
Dave Stevenson [Fri, 6 Nov 2020 18:45:10 +0000 (18:45 +0000)]
Input: edt-ft5x06: Poll the device if no interrupt is configured.

Not all systems have the interrupt line wired up, so switch to
polling the touchscreen off a timer if no interrupt line is
configured.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
input: edt-ft5x06: Handle unreliable TOUCH_UP events

The ft5x06 is unreliable in sending touch up events, so some
touch IDs can become stuck in the detected state.

Ensure that IDs that are unreported by the controller are
released.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
input: edt-ft5x06: Only look at the number of points reported

Register 0x02 in the FT5x06 is TD_STATUS containing the number
of valid touch points being reported.

Iterate over that number of points rather than all that are
supported on the device.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
input: edt-ft5x06: Only read data for number of points reported

Rather than always reading the maximum number of points supported
by the chip (which may be as high as 10), read the number of
active points first, and read data for just those.
In most cases this will result in less data on the I2C bus,
with only the maximum touch points taking more due to a second
read that has to configure the start address.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
input: edt-ft5x06: Fix patch reading only the number of points reported

Fix bad conflict resolution from upstream updates. Need to read
from tsdata->tdata_offset bytes, not from tsdata->offset.
Also fix logging of i2c read errors to cover both transactions.

Fixes: 7216fcfe2e5f ("input: edt-ft5x06: Only read data for number of points reported")
Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agophy: broadcom: Add bcm54213pe configuration
Phil Elwell [Thu, 29 Oct 2020 14:10:56 +0000 (14:10 +0000)]
phy: broadcom: Add bcm54213pe configuration

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agophy: broadcom: split out the BCM54213PE from the BCM54210E IDs
Jonathan Bell [Tue, 14 May 2019 16:00:41 +0000 (17:00 +0100)]
phy: broadcom: split out the BCM54213PE from the BCM54210E IDs

The last nibble is a revision ID, and the 54213pe is a later rev
than the 54210e. Running the 54210e setup code on a 54213pe results
in a broken RGMII interface.

Signed-off-by: Jonathan Bell <jonathan@raspberrypi.org>
8 months agofirmware: raspberrypi: Add support for tryonce reboot flag
Tim Gover [Tue, 20 Oct 2020 10:55:37 +0000 (11:55 +0100)]
firmware: raspberrypi: Add support for tryonce reboot flag

Define a new mailbox (SET_REBOOT_FLAGS) which may be used to
pass optional flags to the Raspberry Pi firmware that changes
the behaviour of the bootloader and firmware during a reboot.

Currently this just defines the 'tryboot' flag which causes
the firmware to load tryboot.txt instead config.txt. This
alternate configuration file can be used to specify the
path of an alternate firmware and kernels allowing a fallback
mechanism to be implemented for OS upgrades.

8 months agowatchdog: bcm2835: Ignore params after the partition number
Tim Gover [Thu, 22 Oct 2020 14:30:55 +0000 (15:30 +0100)]
watchdog: bcm2835: Ignore params after the partition number

Use sscanf to extract the partition number and ignore extra parameters
which are only relevant to other reboot notifiers.

8 months agobcm2708_fb: Fix a build warning
Phil Elwell [Tue, 27 Oct 2020 12:12:22 +0000 (12:12 +0000)]
bcm2708_fb: Fix a build warning

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agorpisense-fb: Set pseudo_pallete to prevent crash on fbcon takeover
Serge Schneider [Mon, 26 Oct 2020 16:38:21 +0000 (16:38 +0000)]
rpisense-fb: Set pseudo_pallete to prevent crash on fbcon takeover

Signed-off-by: Serge Schneider <serge@raspberrypi.com>
8 months agogpio: Add gpio-fsm driver
Phil Elwell [Wed, 30 Sep 2020 11:00:54 +0000 (12:00 +0100)]
gpio: Add gpio-fsm driver

The gpio-fsm driver implements simple state machines that allow GPIOs
to be controlled in response to inputs from other GPIOs - real and
soft/virtual - and time delays. It can:
+ create dummy GPIOs for drivers that demand them,
+ drive multiple GPIOs from a single input, with optional delays,
+ add a debounce circuit to an input,
+ drive pattern sequences onto LEDs
etc.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
gpio-fsm: Fix a build warning

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
gpio-fsm: Rename 'num-soft-gpios' to avoid warning

As of 5.10, the Device Tree parser warns about properties that look
like references to "suppliers" of various services. "num-soft-gpios"
resembles a declaration of a GPIO called "num-soft", causing the value
to be interpreted as a phandle, the owner of which is checked for a
"#gpio-cells" property.

To avoid this warning, rename the gpio-fsm property to "num-swgpios".

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
gpio-fsm: Show state info in /sys/class/gpio-fsm

Add gpio-fsm sysfs entries under /sys/class/gpio-fsm. For each state
machine show the current state, which state (if any) will be entered
after a delay, and the current value of that delay.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
gpio-fsm: Fix shutdown timeout handling

The driver is intended to jump directly to a shutdown state in the
event of a timeout during shutdown, but the sense of the test was
inverted.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
gpio-fsm: Clamp the delay time to zero

The sysfs delay_ms value is calculated live, and it is possible for
the time left to appear to be negative briefly if the timer handling
hasn't completed. Ensure the displayed value never goes below zero,
for the sake of appearances.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
gpio-fsm: Sort functions into a more logical order

Move some functions into a more logical ordering. This change causes
no functional change and is essentially cosmetic.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
gpio_fsm: Rework the atomic-vs-non-atomic split

Partition the code to separate atomic and non-atomic methods so that
none of them have to handle both cases. The result avoids using deferred
work unless necessary, and should be easier to understand.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agostaging: vc04_services: ISP: Add a more complex ISP processing component
Naushir Patuck [Thu, 23 Apr 2020 09:17:37 +0000 (10:17 +0100)]
staging: vc04_services: ISP: Add a more complex ISP processing component

Driver for the BCM2835 ISP hardware block.  This driver uses the MMAL
component to program the ISP hardware through the VC firmware.

The ISP component can produce two video stream outputs, and Bayer
image statistics. This can't be encompassed in a simple V4L2
M2M device, so create a new device that registers 4 video nodes.

This patch squashes all the development patches from the earlier
rpi-5.4.y branch into one

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agouapi: bcm2835-isp: Add bcm2835-isp uapi header file
Dave Stevenson [Mon, 12 Oct 2020 16:03:14 +0000 (17:03 +0100)]
uapi: bcm2835-isp: Add bcm2835-isp uapi header file

This file defines the userland interface to the bcm2835-isp driver
that will follow in a separate commit.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agostaging: vc04_services: Add a V4L2 M2M codec driver
Dave Stevenson [Thu, 8 Oct 2020 19:24:12 +0000 (20:24 +0100)]
staging: vc04_services: Add a V4L2 M2M codec driver

This adds a V4L2 memory to memory device that wraps the MMAL
video decode and video_encode components for H264 and MJPEG encode
and decode, MPEG4, H263, and VP8 decode (and MPEG2 decode
if the appropriate licence has been purchased).

This patch squashes all the work done in developing the driver
on the Raspberry Pi rpi-5.4.y kernel branch.
Thanks to Kieran Bingham, Aman Gupta, Chen-Yu Tsai, and
Marek Behún for their contributions. Please refer to the
rpi-5.4.y branch for the full history.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835-codec: Ensure OUTPUT timestamps are always forwarded

The firmware by default tries to ensure that decoded frame
timestamps always increment. This is counter to the V4L2 API
which wants exactly the OUTPUT queue timestamps passed to the
CAPTURE queue buffers.

Disable the firmware option.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/vc04_services/codec: Add support for CID MPEG_HEADER_MODE

Control V4L2_CID_MPEG_VIDEO_HEADER_MODE controls whether the encoder
is meant to emit the header bytes as a separate packet or with the
first encoded frame.
Add support for it.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/vc04_services/codec: Clear last buf dequeued flag on START

It appears that the V4L2 M2M framework requires the driver to manually
call vb2_clear_last_buffer_dequeued on the CAPTURE queue during a
V4L2_DEC_CMD_START.
Add such a call.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/vc04-services/codec: Fix logical precedence issue

Two issues identified with operator precedence in logical
expressions. Fix them.

https://github.com/raspberrypi/linux/issues/4040

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
vc04_services: bcm2835-codec: Switch to s32fract

staging/bcm2835-codec: Add the unpacked (16bpp) raw formats

Now that the firmware supports the unpacked (16bpp) variants
of the MIPI raw formats, add the mappings.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835-codec: Log the number of excess supported formats

When logging that the firmware has provided more supported formats
than we had allocated storage for, log the number allocated and
returned.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835-codec: Add support for pixel aspect ratio

If the format is detected by the driver and a V4L2_EVENT_SOURCE_CHANGE
event is generated, then pass on the pixel aspect ratio as well.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835-codec: Implement additional g_selection calls for decode

v4l_cropcap calls our vidioc_g_pixelaspect function to get the pixel
aspect ratio, but also calls g_selection for V4L2_SEL_TGT_CROP_BOUNDS
and V4L2_SEL_TGT_CROP_DEFAULT. Whilst it allows for vidioc_g_pixelaspect
not to be implemented, it doesn't allow for either of the other two.

Add in support for the additional selection targets.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835-codec: Add VC-1 support.

Providing the relevant licence has been purchased, then Pi0-3
can decode VC-1.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835-codec: Fix support for levels 4.1 and 4.2

The driver said it supported H264 levels 4.1 and 4.2, but
was missing the V4L2 to MMAL mappings.

Add in those mappings.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835-codec: Set the colourspace appropriately for RGB formats

Video decode supports YUV and RGB formats. YUV needs to report SMPTE170M
or REC709 appropriately, whilst RGB should report SRGB.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835-codec: Pass corrupt frame flag.

MMAL has the flag MMAL_BUFFER_HEADER_FLAG_CORRUPTED but that
wasn't being passed through, so add it.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835-codec: Do not update crop from S_FMT after res change

During decode, setting the CAPTURE queue format was setting the crop
rectangle to the requested height before aligning up the format to
cater for simple clients that weren't expecting to deal with cropping
and the SELECTION API.
This caused problems on some resolution change events if the client
didn't also then use the selection API.

Disable the crop update after a resolution change.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
bcm2835: Allow compressed frames to set sizeimage (#4386)

Allow the user to set sizeimage in TRY_FMT and S_FMT if the format
flags have V4L2_FMT_FLAG_COMPRESSED set

Signed-off-by: John Cox <jc@kynesim.co.uk>
staging/bcm2835-codec: Change the default codec res to 32x32

In order to effectively guarantee that a V4L2_EVENT_SOURCE_CHANGE
event occurs, adopt a default resolution of 32x32 so that it
is incredibly unlikely to be decoding a stream of that resolution
and therefore failing to note a "change" requiring the event.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835-codec: Add support for decoding interlaced streams

The video decoder can support decoding interlaced streams, so add
the required plumbing to signal this correctly.

The encoder and ISP do NOT support interlaced data, so trying to
configure an interlaced format on those nodes will be rejected.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835-codec: Correct ENUM_FRAMESIZES stepsize to 2

Being YUV420 formats, the step size is always 2 to avoid part
chroma subsampling.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835-codec: Return buffers to QUEUED not ERROR state

Should start_streaming fail, or buffers be queued during
stop_streaming, they should be returned to the core as QUEUED
and not (as currently) as ERROR.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835_codec: Log MMAL flags in hex

The flags is a bitmask, so it's far easier to interpret as hex
data instead of decimal.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging: bcm2835-codec: Allow custom specified strides/bytesperline.

If the client provides a bytesperline value in try_fmt/s_fmt then
validate it and correct if necessary.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835_codec: Add support for image_fx to deinterlace

Adds another /dev/video node wrapping image_fx doing deinterlace.

Co-developed-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
Signed-off-by: Dom Cobley <popcornmix@gmail.com>
staging/bcm2835-v4l2_codec: Fix for encode selection API

Matches correct behaviour from DECODE and DEINTERLACE

Signed-off-by: Dom Cobley <popcornmix@gmail.com>
staging: bcm2835-codec: Allow decode res changed before STREAMON(CAPTURE)

The V4L2 stateful video decoder API requires that you can STREAMON
on only the OUTPUT queue, feed in buffers, and wait for the
SOURCE_CHANGE event.
This requires that we enable the MMAL output port at the same time
as the input port, because the output port is the one that creates
the SOURCE_CHANGED event.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835-codec: Do not send buffers to the VPU unless streaming

With video decode we now enable both input and output ports on
the component. This means that buffers will get passed to the VPU
earlier than desired if they are queued befoer STREAMON.

Check that the queue is streaming before sending buffers to the VPU.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging: bcm2835-codec: Format changed should trigger drain

When a format changed event occurs, the spec says that it
triggers an implicit drain, and that needs to be signalled
via -EPIPE.

For BCM2835, the format changed event happens at the point
the format change occurs, so no further buffers exist from
before the resolution changed point. We therefore signal the
last buffer immediately.
We don't have a V4L2 available to us at this point, so set
the videobuf2 queue last_buffer_dequeued flag directly.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging: bcm2835-codec: Signal the firmware to stop on all changes

The firmware defaults to not stopping video decode if only the
pixel aspect ratio or colourspace change. V4L2 requires us
to stop decoding on any change, therefore tell the firmware
of the desire for this alternate behaviour.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging: bcm2835-codec: Queue flushed buffers instead of completing

When a buffer is returned on a port that is disabled, return it
to the videobuf2 QUEUED state instead of DONE which returns it
to the client.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging: bcm2835_codec: Correct flushing code for refcounting

Completions don't reference count, so setting the completion
on the first buffer returned and then not reinitialising it
means that the flush function doesn't behave as intended.

Signal the completion when the last buffer is returned.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging: bcm2835-codec: Ensure all ctrls are set on streamon

Currently the code was only setting some controls from
bcm2835_codec_set_ctrls, but it's simpler to use
v4l2_ctrl_handler_setup to avoid forgetting to adding new
controls to the list.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging: bcm2835-codec: Add support for H&V Flips to ISP

The ISP can do H & V flips whilst resizing or converting
the image, so expose that via V4L2_CID_[H|V]FLIP.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
bcm2835-v4l2-codec: Remove advertised support of VP8

The support for this format by firmware is very limited
and won't be faster than the arm.

Signed-off-by: Dom Cobley <popcornmix@gmail.com>
Pass V4L2_CID_MPEG_VIDEO_H264_MIN_QP/MAX_QP to bcm2835-v4l2-codec

Following raspberrypi/linux#4704. This is necessary to set up
quantization for variable bitrate to avoid video flickering.

staging/bcm2835-codec: bytesperline for YUV420/YVU420 needs to be 64

Matching https://github.com/raspberrypi/linux/pull/4419, the ISP
block (which is also used on the input of the encoder, and output
of the decoder) needs the base address of all planes to be aligned
to multiples of 32. This includes the chroma planes of YUV420 and
YVU420.
If the height is only a multiple of 2 (not 4), then you get an odd
number of lines in the second plane, which means the 3rd plane
starts at a multiple of bytesperline/2.

Set the minimum bytesperline alignment to 64 for those formats
so that the plane alignment is always right.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/bcm2835-codec: Allow a different stride alignment per role

Deinterlace and decode aren't affected in the same way as encode
and ISP by the alignment requirement on 3 plane YUV420.
Decode would be affected, but it always aligns the height up to
a macroblock, and uses the selection API to reflect that.

Add in the facility to set the bytesperline alignment per role.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging: vc04_services: codec: Add support for V4L2_PIX_FMT_RGBA32 format

We already support V4L2_PIX_FMT_BGR32 which is the same thing with red
and blue swapped, so it makes sense to include this variant as well.

Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
bcm2835-codec: Return empty buffers to the VPU instead of queueing to vbuf2

The encoder can skip frames totally should rate control overshoot
the target bitrate too far. In this situation it generates an
output buffer of length 0.
V4L2 treats a buffer of length 0 as an end of stream flag, which is
not appropriate in this case, therefore we can not return that buffer
to the client.

The driver was returning the buffer to videobuf2 in the QUEUED state,
however that buffer was then not dequeued again, so the number of
buffers was reduced each time this happened. In the pathological
case of using GStreamer's videotestsrc in mode 1 for noise, this happens
sufficiently frequently to totally stall the pipeline.

If the port is still enabled then return the buffer straight back to
the VPU rather than to videobuf2.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
vc04_services: bcm2835-codec: Add support for V4L2_PIX_FMT_NV12_COL128

V4L2_PIX_FMT_NV12_COL128 is supported by the ISP and the input of
video_encode, output of video_decode, and both input and output
of the ISP.

Add in the plumbing to support the format on those ports.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
vc04_services: bcm2835-codec: Set crop_height for compressed formats

In particular for the encoder where the CAPTURE format dictates
the parameters given to the codec we need to be able to set the
value passed as the crop_height for the compressed format.
There's no crop available for cropped modes, so always set
crop_height to the requested height.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
vc04_services: bcm2835-codec: Set port format from s_selection

s_selection allows the crop region of an uncompressed pixel
format to be specified, but it wasn't passing the setting on to
the firmware. Depending on call order this would potentially
mean that the crop wasn't actioned.

Set the port format on s_selection if we have a component created.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
bcm2835-codec: /dev/video31 as interface to image_encode JPEG encoder

Signed-off-by: Maxim Devaev <mdevaev@gmail.com>
bcm2835-v4l2-codec: support H.264 5.0 and 5.1 levels

vc04_services: bcm2835-codec:  Remove redundant role check

vidioc_try_encoder_cmd checks the role, but the ioctl is disabled
for any roles in which it is invalid.

Remove the redundant check.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
vc04_services: bcm2835-codec: Allow encoder_cmd on ISP and deinterlace

ISP and deinterlace also need a mechanism for passing effectively
an EOS through the pipeline to signal when all buffers have been
processed.

VIDIOC_ENCODER_CMD does exactly this for encoders, so reuse the same
function for ISP and deinterlace.
(VIDIOC_DECODER_CMD is slightly different in that it also passes
details of when and how to stop, so is not as relevant).

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
vc04_services: bcm2835_codec: Allow larger images through the ISP

Whilst the codecs are restricted to 1920x1080 / 1080x1920, the ISP
isn't, but the limits advertised via V4L2 was 1920x1920 for all
roles.

Increase the limit to 16k x 16k for the ISP.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
media: bcm2835-v4l2-codec: Enable selection ioctl for ISP

The ISP cases do nothing. Remove the break that separates them from the
deinterlace case so they now do the same as deinterlace. This enables
simple width & height setting, but does not enable setting left and
top coordinates.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: bcm2835-v4l2-codec: Add profile & level ctrls to decode

In order to support discovery of what profile & levels are supported by
stateful decoders implement the profile and level controls where they
are defined by V4L2.

Signed-off-by: John Cox <jc@kynesim.co.uk>
vc04_services: bcm2835_codec: Ignore READ_ONLY ctrls in s_ctrl

In adding the MPEG2/MPEG4/H264 level and profile controls to
the decoder, they weren't declared as read-only, nor handlers
added to bcm2835_codec_s_ctrl. That resulted in an error message
"Invalid control" being logged every time v4l2_ctrl_handler_setup
was called from bcm2835_codec_create_component.

Define those controls as read only, and exit early from s_ctrl
on read only controls.

Fixes: "media: bcm2835-v4l2-codec: Add profile & level ctrls to decode"
Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
vc04_services: bcm2835_codec: Set MPEG2_LEVEL control to READ_ONLY

V4L2_CID_MPEG_VIDEO_MPEG2_LEVEL  was missed from
"vc04_services: bcm2835_codec: Ignore READ_ONLY ctrls in s_ctrl"

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging: bcm2835-codec: Add V4L2_CID_MPEG_VIDEO_B_FRAMES control

FFmpeg insists on trying to set V4L2_CID_MPEG_VIDEO_B_FRAMES to
0, and generates an error should it fail.
As our encoder doesn't support B frames, add a stub handler for
it to silence FFmpeg.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging: bcm2835-codec: Add support for V4L2_CID_MPEG_VIDEO_GOP_SIZE

For H264, V4L2_CID_MPEG_VIDEO_H264_I_PERIOD is meant to be the intra
I-frame period, whilst V4L2_CID_MPEG_VIDEO_GOP_SIZE is the intra IDR
frame period.
The firmware encoder doesn't produce I-frames that aren't IDR as well,
therefore V4L2_CID_MPEG_VIDEO_GOP_SIZE is technically the correct
control, however users may have adopted V4L2_CID_MPEG_VIDEO_H264_I_PERIOD.

Add support for V4L2_CID_MPEG_VIDEO_GOP_SIZE controlling the encoder,
and have VIDIOC_S_CTRL for V4L2_CID_MPEG_VIDEO_H264_I_PERIOD update
the value for V4L2_CID_MPEG_VIDEO_GOP_SIZE (the reverse is not
implemented).

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging: bcm2835-codec: Add missing alignment for V4L2_PIX_FMT_RGBA32

The patch adding image encode (JPEG) to the driver missed adding
the alignment constraint for V4L2_PIX_FMT_RGBA32, which meant
it ended up giving a stride and size of 0.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging: bcm2835-codec: Downgrade the level for a debug message

The debug message from bcm2835_codec_buf_prepare when the buffer
size is incorrect can be a little spammy if the application isn't
careful on how it drives it, therefore drop the priority of the
message.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agostaging: mmal-vchiq: Use vc-sm-cma to support zero copy
Dave Stevenson [Tue, 25 Sep 2018 15:07:55 +0000 (16:07 +0100)]
staging: mmal-vchiq: Use vc-sm-cma to support zero copy

With the vc-sm-cma driver we can support zero copy of buffers between
the kernel and VPU. Add this support to mmal-vchiq.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
vc-sm-cma: fixed kbuild problem

error logs:
  drivers/staging/vc04_services/vc-sm-cma/Kconfig:1:error: recursive dependency detected!
  drivers/staging/vc04_services/vc-sm-cma/Kconfig:1:      symbol BCM_VC_SM_CMA is selected by BCM2835_VCHIQ_MMAL
  drivers/staging/vc04_services/vchiq-mmal/Kconfig:1:     symbol BCM2835_VCHIQ_MMAL depends on BCM2835_VCHIQ
  drivers/staging/vc04_services/Kconfig:14:       symbol BCM2835_VCHIQ is selected by BCM_VC_SM_CMA
  For a resolution refer to Documentation/kbuild/kconfig-language.rst
  subsection "Kconfig recursive dependency limitations"

Tested-by: make ARCH=arm64 bcm2711_defconfig
Test platform: fedora 33
Branch: rpi-5.10.y

8 months agostaging: mmal-vchiq: Add monochrome image formats
Dave Stevenson [Wed, 6 May 2020 17:11:14 +0000 (18:11 +0100)]
staging: mmal-vchiq: Add monochrome image formats

Adds support for monochrome image formats in the various
MIPI packings.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agostaging: vchiq-mmal: Add support for 14bit Bayer
Dave Stevenson [Wed, 6 May 2020 17:09:04 +0000 (18:09 +0100)]
staging: vchiq-mmal: Add support for 14bit Bayer

Add in the missing defines.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agostaging: vc04_services: Add new vc-sm-cma driver
Dave Stevenson [Thu, 8 Oct 2020 17:49:52 +0000 (18:49 +0100)]
staging: vc04_services: Add new vc-sm-cma driver

Add Broadcom VideoCore Shared Memory support.

This new driver allows contiguous memory blocks to be imported
into the VideoCore VPU memory map, and manages the lifetime of
those objects, only releasing the source dmabuf once the VPU has
confirmed it has finished with it.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging: vcsm-cma: Fix memory leak from not detaching dmabuf

When importing there was a missing call to detach the buffer,
so each import leaked the sg table entry.

Actually the release process for both locally allocated and
imported buffers is identical, so fix them to both use the same
function.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
staging/vc-sm-cma: Avoid log spamming on Pi0/1 over cache alias.

Pi 0/1 use the 0x80000000 cache alias as the ARM also sees the world
through the VPU L2 cache.
vc-sm-cma was trying to ensure it was in an uncached alias (0xc), and
complaining on every allocation if it weren't. Reduce this logging.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
vc-sm-cma: Restore correct cache maintainance operations

We have been using the more expensive flush operations rather than
invalidate and clean since kernel rpi-5.9.y

These are exposed with:
52f1453513ba95084ab811a030032fe605b0cbe2 Re-expose some dmi APIs for use in VCSM

But I believe that commit was dropped when (non-cma) vc-sm was dropped,
and didn't get updated when the commit was restored

Signed-off-by: Dom Cobley <popcornmix@gmail.com>
staging: vc04_services: Fix clang14 warning

Insert a break to fix a fallthrough warning from clang14. Since the
fallthrough was to another break, this is a cosmetic change.

See: https://github.com/raspberrypi/linux/issues/5078

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
vc04_services/vc-sm-cma: Handle upstream require vchiq_instance to be passed around

vc04_services/vc-sm-cma: Switch one-bit bitfields to bool

Clang 16 warns:

../drivers/staging/vc04_services/vc-sm-cma/vc_sm.c:816:19: warning: implicit truncation from 'int' to a one-bit wide bit-field changes value from 1 to -1 [-Wsingle-bit-bitfield-constant-conversion]
        buffer->imported = 1;
                         ^ ~
../drivers/staging/vc04_services/vc-sm-cma/vc_sm.c:822:17: warning: implicit truncation from 'int' to a one-bit wide bit-field changes value from 1 to -1 [-Wsingle-bit-bitfield-constant-conversion]
        buffer->in_use = 1;
                       ^ ~
2 warnings generated.

Signed-off-by: Alexander Winkowski <dereference23@outlook.com>
vc04_services: vcsm-cma: Detach from the correct dmabuf

Commit d3292daee319 ("dma-buf: Make locking consistent in dma_buf_detach()")
added checking that the same dmabuf for which dma_buf_attach
was called is passed into dma_buf_detach, which flagged up
that vcsm-cma was passing in the wrong dmabuf.

Correct this so that we don't get the WARN on every dma_buf
release.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agostaging: bcm2835-camera: Replace deprecated V4L2_PIX_FMT_BGR32
Dave Stevenson [Thu, 8 Oct 2020 14:35:14 +0000 (15:35 +0100)]
staging: bcm2835-camera: Replace deprecated V4L2_PIX_FMT_BGR32

V4L2_PIX_FMT_BGR32 is deprecated as it is ambiguous over where
the alpha byte is. Cheese/GStreamer appear to get it wrong for
one, and qv4l2 gets red and blue swapped.

Swap to the newer V4L2_PIX_FMT_BGRX32 format.

https://www.raspberrypi.org/forums/viewtopic.php?f=38&t=267736&p=1738912

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agochar: Add broadcom char drivers back to build files
popcornmix [Mon, 28 Sep 2020 19:23:30 +0000 (20:23 +0100)]
char: Add broadcom char drivers back to build files

See: https://github.com/raspberrypi/linux/issues/3875
Signed-off-by: popcornmix <popcornmix@gmail.com>
8 months agonet: bcmgenet: Reset RBUF on first open
Phil Elwell [Fri, 25 Sep 2020 14:07:23 +0000 (15:07 +0100)]
net: bcmgenet: Reset RBUF on first open

If the RBUF logic is not reset when the kernel starts then there
may be some data left over from any network boot loader. If the
64-byte packet headers are enabled then this can be fatal.

Extend bcmgenet_dma_disable to do perform the reset, but not when
called from bcmgenet_resume in order to preserve a wake packet.

N.B. This different handling of resume is just based on a hunch -
why else wouldn't one reset the RBUF as well as the TBUF? If this
isn't the case then it's easy to change the patch to make the RBUF
reset unconditional.

See: https://github.com/raspberrypi/linux/issues/3850

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agobrcmfmac: Increase power saving delay to 2s
Phil Elwell [Mon, 3 Feb 2020 09:32:22 +0000 (09:32 +0000)]
brcmfmac: Increase power saving delay to 2s

Increase the delay before entering the lower power state to 2 seconds
(the maximum allowed) in order to reduce the packet latencies,
particularly for inbound packets.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agostaging/fbtft: Add support for display variants
Phil Elwell [Tue, 1 Sep 2020 17:15:27 +0000 (18:15 +0100)]
staging/fbtft: Add support for display variants

Display variants are intended as a replacement for the now-deleted
fbtft_device drivers. Drivers can register additional compatible
strings with a custom callback that can make the required changes
to the fbtft_display structure.

Start the ball rolling by adding adafruit18, adafruit18_green and
sainsmart18 displays.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agomedia: bcm2835-unicam: change minimum number of vb2_queue buffers to 1
David Plowman [Thu, 28 May 2020 10:09:48 +0000 (11:09 +0100)]
media: bcm2835-unicam: change minimum number of vb2_queue buffers to 1

Since the unicam driver was modified to write to a dummy buffer when no
user-supplied buffer is available, it can now write to and return a
buffer even when there's only a single one. Enable this by changing the
min_buffers_needed in the vb2_queue; it will be useful for enabling
still captures without allocating more memory than absolutely necessary.

Signed-off-by: David Plowman <david.plowman@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Retain packing information on G_FMT
Dave Stevenson [Tue, 19 May 2020 10:46:47 +0000 (11:46 +0100)]
media: bcm2835-unicam: Retain packing information on G_FMT

The change to retrieve the pixel format always on g_fmt didn't
check whether the native or unpacked version of the format
had been requested, and always returned the packed one.
Correct this so that the packing setting is retained whereever
possible.

Fixes "9d59e89 media: bcm2835-unicam: Re-fetch mbus code from subdev
on a g_fmt call"

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Fixup review comments from Hans.
Dave Stevenson [Tue, 23 Jun 2020 14:14:05 +0000 (15:14 +0100)]
media: bcm2835-unicam: Fixup review comments from Hans.

Updates the driver based on the upstream review comments from
Hans Verkuil at https://patchwork.linuxtv.org/patch/63531/

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agomedia: bcm2835: unicam: Fix uninitialized warning
Jacko Dirks [Tue, 5 May 2020 12:33:31 +0000 (14:33 +0200)]
media: bcm2835: unicam: Fix uninitialized warning

Signed-off-by: Jacko Dirks <jdirks.linuxdev@gmail.com>
8 months agomedia: bcm2835-unicam: Always service interrupts
Dave Stevenson [Wed, 13 May 2020 17:28:27 +0000 (18:28 +0100)]
media: bcm2835-unicam: Always service interrupts

From when bringing up the driver, there was a check in the isr
to ignore interrupts (claiming them handled) should the driver
not be streaming.

The VPU now will not register a camera driver if it finds a
CSI2 node enabled in device tree, therefore this flawed check is
redundant.

https://github.com/raspberrypi/linux/issues/3602

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agomedia: i2c: tc358743: Only allow supported pixel fmts in set_fmt
Dave Stevenson [Fri, 10 Jul 2020 11:40:50 +0000 (12:40 +0100)]
media: i2c: tc358743: Only allow supported pixel fmts in set_fmt

Fix commit "media: tc358743: Return an appropriate colorspace from
tc358743_set_fmt" to ensure that the format passed in to set_fmt
is checked to be valid, and reset to the current format if not.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Drop WARN on uing direct cache alias
Dave Stevenson [Thu, 27 Aug 2020 15:30:26 +0000 (16:30 +0100)]
media: bcm2835-unicam: Drop WARN on uing direct cache alias

Pi 0&1 pass all ARM accesses through the VPU L2 cache, therefore
the dma-ranges property sets the cache alias bits to other
than the direct alias, hence this WARN was firing.

It was overprotective coding, so assume that everything is OK
with the dma-ranges, and remove the WARN.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agoleds: Add the actpwr trigger
Phil Elwell [Mon, 13 Jul 2020 09:33:19 +0000 (10:33 +0100)]
leds: Add the actpwr trigger

The actpwr trigger is a meta trigger that cycles between an inverted
mmc0 and default-on. It is written in a way that could fairly easily
be generalised to support alternative sets of source triggers.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agodt-bindings: bcm2835-unicam: Update documentation with new clock params
Naushir Patuck [Mon, 11 May 2020 12:06:27 +0000 (13:06 +0100)]
dt-bindings: bcm2835-unicam: Update documentation with new clock params

Update the documentation to reflect the new "VPU" clock needed
by the bcm2835-unicam driver.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agomedia: bcm2835: unicam: Set VPU min clock freq to 250Mhz.
Naushir Patuck [Mon, 11 May 2020 12:02:22 +0000 (13:02 +0100)]
media: bcm2835: unicam: Set VPU min clock freq to 250Mhz.

When streaming with Unicam, the VPU must have a clock frequency of at
least 250Mhz.  Otherwise, the input fifos could overrun, causing
image corruption.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Ensure type is VIDEO_CAPTURE in [g|s]_selection
Dave Stevenson [Tue, 7 Jul 2020 13:52:43 +0000 (14:52 +0100)]
media: bcm2835-unicam: Ensure type is VIDEO_CAPTURE in [g|s]_selection

[g|s]_selection pass in a buffer type that needs to be validated
before passing on to the sensor subdev.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Reinstate V4L2_CAP_READWRITE in the caps
Dave Stevenson [Tue, 7 Jul 2020 13:23:40 +0000 (14:23 +0100)]
media: bcm2835-unicam: Reinstate V4L2_CAP_READWRITE in the caps

v4l2-compliance throws a failure if the device doesn't advertise
V4L2_CAP_READWRITE but allows read or write operations.
We do support read, so reinstate the flag.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Add support for unpacked 14bit Bayer formats
Dave Stevenson [Wed, 1 Jul 2020 09:57:57 +0000 (10:57 +0100)]
media: bcm2835-unicam: Add support for unpacked 14bit Bayer formats

Now that the 14bit non-packed Bayer formats are defined, add them
into the supported formats lookup table.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Add support for 14bit mono sources
Dave Stevenson [Thu, 25 Jun 2020 17:03:47 +0000 (18:03 +0100)]
media: bcm2835-unicam: Add support for 14bit mono sources

Now that V4L2_PIX_FMT_Y14 and V4L2_PIX_FMT_Y14P are defined,
allow passing 14bit mono data through the peripheral.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Add support for 12bit mono packed format
Dave Stevenson [Thu, 25 Jun 2020 16:53:32 +0000 (17:53 +0100)]
media: bcm2835-unicam: Add support for 12bit mono packed format

Now that V4L2_PIX_FMT_Y12P is defined, allow passing raw 12bit
mono packed data through the peripheral.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agomedia: Add a pixel format for MIPI packed 14bit luma only.
Dave Stevenson [Thu, 25 Jun 2020 16:51:03 +0000 (17:51 +0100)]
media: Add a pixel format for MIPI packed 14bit luma only.

This is the format used by monochrome 14bit image sensors.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agomedia: Add a pixel format for MIPI packed 12bit luma only.
Dave Stevenson [Thu, 7 May 2020 15:59:03 +0000 (16:59 +0100)]
media: Add a pixel format for MIPI packed 12bit luma only.

This is the format used by monochrome 12bit image sensors.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agoserial: 8250: bcm2835aux - defer if clock is zero
Phil Elwell [Thu, 2 Jul 2020 12:53:20 +0000 (13:53 +0100)]
serial: 8250: bcm2835aux - defer if clock is zero

See: https://github.com/raspberrypi/linux/issues/3700

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Avoid gcc warning over {0} on endpoint
Dave Stevenson [Fri, 26 Jun 2020 14:53:44 +0000 (15:53 +0100)]
media: bcm2835-unicam: Avoid gcc warning over {0} on endpoint

Older gcc versions object to = { 0 } initialisation if the first
elemtn in the structure is a substructure.

Use = { } to avoid this compiler warning.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Add support for get_mbus_config to set num lanes
Dave Stevenson [Tue, 23 Jun 2020 13:32:51 +0000 (14:32 +0100)]
media: bcm2835-unicam: Add support for get_mbus_config to set num lanes

Use the get_mbus_config pad subdev call to allow a source to use
fewer than the number of CSI2 lanes defined in device tree.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Kconfig/Makefile for CCP2/CSI2 camera interface
Naushir Patuck [Wed, 1 Apr 2020 07:39:49 +0000 (08:39 +0100)]
media: bcm2835-unicam: Kconfig/Makefile for CCP2/CSI2 camera interface

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agomedia: bcm2835-unicam: Driver for CCP2/CSI2 camera interface
Naushir Patuck [Mon, 4 May 2020 09:25:41 +0000 (12:25 +0300)]
media: bcm2835-unicam: Driver for CCP2/CSI2 camera interface

Add a driver for the Unicam camera receiver block on BCM283x processors.
Compared to the bcm2835-camera driver present in staging, this driver
handles the Unicam block only (CSI-2 receiver), and doesn't depend on
the VC4 firmware running on the VPU.

The commit is made up of a series of changes cherry-picked from the
rpi-5.4.y branch of https://github.com/raspberrypi/linux/ with
additional enhancements, forward-ported to the mainline kernel.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reported-by: kbuild test robot <lkp@intel.com>
8 months agobrcmfmac: Prefer a ccode from OTP over nvram file
Phil Elwell [Fri, 26 Jun 2020 10:51:05 +0000 (11:51 +0100)]
brcmfmac: Prefer a ccode from OTP over nvram file

Allow the nvram file to set a default ccode (regulatory domain) without
overriding one set in OTP.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agomedia: irs1125: Keep HW in sync after imager reset
Markus Proeller [Tue, 16 Jun 2020 11:33:56 +0000 (13:33 +0200)]
media: irs1125: Keep HW in sync after imager reset

When closing the video device, the irs1125 is put in power down state.
To keep V4L2 ctrls and the HW in sync, v4l2_ctrl_handler_setup is
called after power up.

The compound ctrl IRS1125_CID_MOD_PLL however has a default value
of all zeros, which puts the imager into a non responding state.
Thus, this ctrl is not written by the driver into HW after power up.
The userspace has to take care to write senseful data.

Signed-off-by: Markus Proeller <markus.proeller@pieye.org>
8 months agomedia: irs1125: Atomic access to imager reconfiguration
Markus Proeller [Tue, 16 Jun 2020 11:31:36 +0000 (13:31 +0200)]
media: irs1125: Atomic access to imager reconfiguration

Instead of changing the exposure and framerate settings for all sequences,
they can be changed for every sequence individually now. Therefore the
IRS1125_CID_SAFE_RECONFIG ctrl has been removed and replaced by
IRS1125_CID_SAFE_RECONFIG_S<seq_num>_EXPO and *_FRAME ctrls.

The consistency check in the sequence ctrl IRS1125_CID_SEQ_CONFIG
is removed.

Signed-off-by: Markus Proeller <markus.proeller@pieye.org>
8 months agomedia: irs1125: Refactoring and debug messages
Markus Proeller [Tue, 16 Jun 2020 11:27:42 +0000 (13:27 +0200)]
media: irs1125: Refactoring and debug messages

Changed some variable names to comply with checkpatch --strict mode.
Debug messages added.

Signed-off-by: Markus Proeller <markus.proeller@pieye.org>
8 months agomedia: irs1125: Using i2c_transfer for ic2 reads
Markus Proeller [Tue, 16 Jun 2020 11:24:31 +0000 (13:24 +0200)]
media: irs1125: Using i2c_transfer for ic2 reads

Reading data over i2c is done by using i2c_transfer to ensure that this
operation can't be interrupted.

Signed-off-by: Markus Proeller <markus.proeller@pieye.org>
8 months agoPCI: brcmstb: Add DT property to control L1SS
Phil Elwell [Thu, 11 Jun 2020 08:57:03 +0000 (09:57 +0100)]
PCI: brcmstb: Add DT property to control L1SS

The BRCM PCIe block has controls to enable control of the CLKREQ#
signal by the L1SS, and to gate the refclk with the CLKREQ# input.
These controls are mutually exclusive - the upstream code sets the
latter, but some use cases require the former.

Add a Device Tree property - brcm,enable-l1ss - to switch to the
L1SS configuration.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agogpiolib: Don't prevent IRQ usage of output GPIOs
Phil Elwell [Tue, 24 Apr 2018 13:42:27 +0000 (14:42 +0100)]
gpiolib: Don't prevent IRQ usage of output GPIOs

Upstream Linux deems using output GPIOs to generate IRQs as a bogus
use case, even though the BCM2835 GPIO controller is capable of doing
so. A number of users would like to make use of this facility, so
disable the checks.

See: https://github.com/raspberrypi/linux/issues/2527

Signed-off-by: Phil Elwell <phil@raspberrypi.org>
8 months agostaging:vc04_services: bcm2835-camera: Request headers with I-frame
Dave Stevenson [Fri, 15 May 2020 12:48:59 +0000 (13:48 +0100)]
staging:vc04_services: bcm2835-camera: Request headers with I-frame

V4L2 wishes to have the codec header bytes in the same buffer as the
first encoded frame, so it does become 1-in 1-out for encoding.
The firmware now has an option to do this, so enable it.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agostaging: vc04_services: mmal-vchiq: Update parameters list
Dave Stevenson [Fri, 15 May 2020 12:42:10 +0000 (13:42 +0100)]
staging: vc04_services: mmal-vchiq: Update parameters list

Adds in a couple of new MMAL parameter defines.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agovideo: bcm2708_fb: Disable FB if no displays found
Phil Elwell [Tue, 5 May 2020 18:45:41 +0000 (19:45 +0100)]
video: bcm2708_fb: Disable FB if no displays found

If the firmware hasn't detected a display, the driver would assume
one display was available, but because it had failed to retrieve the
display size it would try to allocate a zero-sized buffer.

Avoid the allocation failure by bailing out early if no display is
found.

See: https://github.com/raspberrypi/linux/issues/3598

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agomedia: i2c: tc358743: Fix fallthrough warning
Jacko Dirks [Tue, 5 May 2020 12:28:14 +0000 (14:28 +0200)]
media: i2c: tc358743: Fix fallthrough warning

Signed-off-by: Jacko Dirks <jdirks.linuxdev@gmail.com>
8 months agostaging: vchiq: Load bcm2835_isp driver from vchiq
Naushir Patuck [Wed, 22 Apr 2020 07:32:32 +0000 (08:32 +0100)]
staging: vchiq: Load bcm2835_isp driver from vchiq

bcmn2835_isp is a platform driver dependent on vchiq,
therefore add the load/unload functions for it to vchiq.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agomedia: uapi: v4l-ctrls: Add CID base for the bcm2835-isp driver
Naushir Patuck [Tue, 21 Apr 2020 14:06:19 +0000 (15:06 +0100)]
media: uapi: v4l-ctrls: Add CID base for the bcm2835-isp driver

We are reserving controls for the new bcm2835-isp driver.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agomedia: uapi: v4l2-core: Add ISP statistics output V4L2 fourcc type
Naushir Patuck [Thu, 23 Apr 2020 09:20:26 +0000 (10:20 +0100)]
media: uapi: v4l2-core: Add ISP statistics output V4L2 fourcc type

Add V4L2_META_FMT_BCM2835_ISP_STATS V4L2 format type.

This new format will be used by the BCM2835 ISP device to return
out ISP statistics for 3A.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agomedia: uapi: Add MEDIA_BUS_FMT_SENSOR_DATA media bus format
Naushir Patuck [Tue, 21 Jan 2020 14:06:47 +0000 (14:06 +0000)]
media: uapi: Add MEDIA_BUS_FMT_SENSOR_DATA media bus format

This patch adds MEDIA_BUS_FMT_SENSOR_DATA used by the bcm2835-unicam
driver to support CSI-2 embedded data streams from camera sensors.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agomedia: uapi: v4l2-core: Add sensor ancillary data V4L2 foucc type.
Naushir Patuck [Wed, 1 Apr 2020 07:46:29 +0000 (08:46 +0100)]
media: uapi: v4l2-core: Add sensor ancillary data V4L2 foucc type.

Add V4L2_META_FMT_SENSOR_DATA format 4CC.

This new format will be used by the BCM2835 Unicam device to return
out camera sensor embedded data.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
8 months agospi: Force CS_HIGH if GPIO descriptors are used
Phil Elwell [Fri, 17 Apr 2020 09:46:19 +0000 (10:46 +0100)]
spi: Force CS_HIGH if GPIO descriptors are used

Commit f3186dd87669 ("spi: Optionally use GPIO descriptors for CS GPIOs")
amended of_spi_parse_dt() to always set SPI_CS_HIGH for SPI slaves whose
Chip Select is defined by a "cs-gpios" devicetree property.

This change breaks drivers whose probe functions set the mode field of
the spi_device because in doing so they clear the SPI_CS_HIGH flag.

Fix by setting SPI_CS_HIGH in spi_setup (under the same conditions as
in of_spi_parse_dt()).

See also: 83b2a8fe43bd ("spi: spidev: Fix CS polarity if GPIO descriptors are used")

Fixes: f3186dd87669 ("spi: Optionally use GPIO descriptors for CS GPIOs")
Signed-off-by: Phil Elwell <phil@raspberrypi.com>
SQUASH: spi: Demote SPI_CS_HIGH warning to KERN_DEBUG

This warning is unavoidable from a client's perspective and
doesn't indicate anything wrong (just surprising).

SQUASH with "spi: use_gpio_descriptor fixup moved to spi_setup"

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agostaging: media: rpivid: Add Raspberry Pi V4L2 H265 decoder
John Cox [Thu, 5 Mar 2020 18:30:41 +0000 (18:30 +0000)]
staging: media: rpivid: Add Raspberry Pi V4L2 H265 decoder

This driver is for the HEVC/H265 decoder block on the Raspberry
Pi 4, and conforms to the V4L2 stateless decoder API.

Signed-off-by: John Cox <jc@kynesim.co.uk>
staging: media: rpivid: Select MEDIA_CONTROLLER and MEDIA_CONTROLLER_REQUEST_API

MEDIA_CONTROLLER_REQUEST_API is a hidden option. If rpivid depends on it,
the user would need to first enable another driver that selects
MEDIA_CONTROLLER_REQUEST_API, and only then rpivid would become available.

By selecting it instead of depending on it, it becomes possible to enable
rpivid without having to enable other potentially unnecessary drivers.

Signed-off-by: Hristo Venev <hristo@venev.name>
rpivid_h265: Fix width/height typo

Signed-off-by: popcornmix <popcornmix@gmail.com>
rpivid_h625: Fix build warnings

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
staging: rpivid: Fix crash when CMA alloc fails

If realloc to increase coeff size fails then attempt to re-allocate
the original size.  If that also fails then flag a fatal error to abort
all further decode.

Signed-off-by: John Cox <jc@kynesim.co.uk>
rpivid: Request maximum hevc clock

Query maximum and minimum clock from driver
and use those

Signed-off-by: Dom Cobley <popcornmix@gmail.com>
rpivid: Switch to new clock api

Signed-off-by: Dom Cobley <popcornmix@gmail.com>
rpivid: Only clk_request_done once

Fixes: 25486f49bfe2e3ae13b90478d1eebd91413136ad
Signed-off-by: Dom Cobley <popcornmix@gmail.com>
media: rpivid: Remove the need to have num_entry_points set

VAAPI H265 has num entry points but never sets it. Allow a VAAPI
shim to work without requiring rewriting the VAAPI driver.
num_entry_points can be calculated from the slice_segment_addr
of the next slice so delay processing until we have that.

Also includes some minor cosmetics.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Convert to MPLANE

Use multi-planar interface rather than single plane interface. This
allows dmabufs holding compressed data to be resized.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Add an enable count to irq claim Qs

Add an enable count to the irq Q structures to allow the irq logic to
block further callbacks if resources associated with the irq are not
yet available.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Add a Pass0 to accumulate slices and rework job finish

Due to overheads in assembling controls and requests it is worth having
the slice assembly phase separate from the h/w pass1 processing. Create
a queue to service pass1 rather than have the pass1 finished callback
trigger the next slice job.

This requires a rework of the logic that splits up the buffer and
request done events. This code contains two ways of doing that, we use
Ezequiel Garcias <ezequiel@collabora.com> solution, but expect that
in the future this will be handled by the framework in a cleaner manner.

Fix up the handling of some of the memory exhaustion crashes uncovered
in the process of writing this code.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Map cmd buffer directly

It is unnecessary to have a separate dmabuf to hold the cmd buffer.
Map it directly from the kmalloc.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Improve values returned when setting output format

Guess a better value for the compressed bitstream buffer size

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Improve stream_on/off conformance & clock setup

Fix stream on & off such that failures leave the driver in the correct
state.  Ensure that the clock is on when we are streaming and off when
all contexts attached to this device have stopped streaming.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Improve SPS/PPS error handling/validation

Move size and width checking from bitstream processing to control
validation

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Fix H265 aux ent reuse of the same slot

It is legitimate, though unusual, for an aux ent associated with a slot
to be selected in phase 0 before a previous selection has been used and
released in phase 2. Fix such that if the slot is found to be in use
that the aux ent associated with it is reused rather than an new aux
ent being created. This fixes a problem where when the first aux ent
was released the second was lost track of.

This bug spotted in Nick's testing. It may explain some other occasional,
unreliable decode error reports where dmesg included "Missing DPB AUX
ent" logging.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Update to compile with new hevc decode params

DPB entries have moved from slice params to the new decode params
attribute - update to deal with this.  Also fixes fallthrough
warnings which seem to be new in 5.14.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Make slice ctrl dynamic

Allows the user to submit a whole frames worth of slice headers in
one lump along with a single bitstream dmabuf for the whole lot.
This saves potentially a lot of bitstream copying.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Only create aux entries for H265 if needed

Only create aux entries of mv info for frames where that info might
be used by a later frame.  This saves some memory bandwidth and
potentially some memory.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Avoid returning EINVAL to a G_FMT ioctl

V4L2 spec says that G/S/TRY_FMT IOCTLs should never return errors for
anything other than wrong buffer types. Improve the capture format
function such that this is so and unsupported values get converted
to supported ones properly.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Remove unused ctx state variable and defines

Remove unused ctx state tracking variable and associated defines.
Their presence implies they might be used, but they aren't.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Ensure IRQs have completed before uniniting context

Before uniniting the decode context sync with the IRQ queues to ensure
that decode no longer has any buffers in use.  This fixes a problem that
manifested as ffmpeg leaking CMA buffers when it did a stream off on
OUTPUT before CAPTURE, though in reality it was probably much more
dangerous than that.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: remove min_buffers_needed from src queue

Remove min_buffers_needed=1 from src queue init. Src buffers are bound
to media requests therefore this setting is not needed and generates
a WARN in kernel 5.16.

Signed-off-by: John Cox <jc@kynesim.co.uk>
rpivid: Use clk_get_max_rate()

The driver was using clk_round_rate() to figure out the maximum clock
rate that was allowed for the HEVC clock.

Since we have a function to return it directly now, let's use it.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
media: rpivid: Apply V4L2 stateless API changes

media: rpivid: Fix fallthrough warning

Replace old-style /* FALLTHRU */ with fallthrough;

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Set min value as well as max for HEVC_DECODE_MODE

As only one value can be accepted set both min and max to that value.

Signed-off-by: John Cox <jc@kynesim.co.uk>
media: rpivid: Accept ANNEX_B start codes

Allow the START_CODE control to take ANNEX_B as a value. This makes no
difference to any part of the decode process as the added bytes are in
data that we ignore. This helps my testing and may help userland code
that expects to send those bytes.

Signed-off-by: John Cox <jc@kynesim.co.uk>
rpivid: Convert to new clock rate API

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
8 months agoRFC: media: Add media_request_{pin,unpin} API
Ezequiel Garcia [Sun, 21 Mar 2021 19:38:54 +0000 (16:38 -0300)]
RFC: media: Add media_request_{pin,unpin} API

This is probably not the API we will want to add, but it
should show what semantics are needed by drivers.

The goal is to allow the OUTPUT (aka source) buffer and the
controls associated to a request to be released from the request,
and in particular return the OUTPUT buffer back to userspace,
without signalling the media request fd.

This is useful for devices that are able to pre-process
the OUTPUT buffer, therefore able to release it before
the decoding is finished. These drivers should signal
the media request fd only after the CAPTURE buffer is done.

Tested-by: John Cox <jc@kynesim.co.uk>
Signed-off-by: Ezequiel Garcia <ezequiel@collabora.com>
8 months agomedia: dt-bindings: media: Add binding for the Raspberry PI HEVC decoder
Dave Stevenson [Tue, 17 Mar 2020 10:53:16 +0000 (10:53 +0000)]
media: dt-bindings: media: Add binding for the Raspberry PI HEVC decoder

Adds a binding for the HEVC decoder found on the BCM2711 / Raspberry Pi 4.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agomedia: videodev2.h: Add a format for column YUV4:2:0 modes
Dave Stevenson [Fri, 24 Jan 2020 14:28:21 +0000 (14:28 +0000)]
media: videodev2.h: Add a format for column YUV4:2:0 modes

Some of the Broadcom codec blocks use a column based YUV4:2:0 image
format, so add the documentation and defines for both 8 and 10 bit
versions.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
8 months agobcmgenet: Disable skip_umac_reset by default
popcornmix [Wed, 1 Apr 2020 10:22:44 +0000 (11:22 +0100)]
bcmgenet: Disable skip_umac_reset by default

Possible fixed upstream by 'net: bcmgenet: keep MAC in reset until PHY is up'

Signed-off-by: popcornmix <popcornmix@gmail.com>
8 months agodt-bindings: pci: Add DT docs for Brcmstb PCIe device
Jim Quinlan [Mon, 15 Jan 2018 23:28:39 +0000 (18:28 -0500)]
dt-bindings: pci: Add DT docs for Brcmstb PCIe device

The DT bindings description of the Brcmstb PCIe device is described.  This
node can be used by almost all Broadcom settop box chips, using
ARM, ARM64, or MIPS CPU architectures.

Signed-off-by: Jim Quinlan <jim2101024@gmail.com>
8 months agoof: overlay: Correct symbol path fixups
Phil Elwell [Thu, 6 Feb 2020 12:23:15 +0000 (12:23 +0000)]
of: overlay: Correct symbol path fixups

When symbols from overlays are added to the live tree their paths must
be rebased. The translated symbol is normally the result of joining
the fragment-relative path (with a leading "/") to the target path
(either copied directly from the "target-path" property or resolved
from the phandle). This translation fails when the target is the root
node (a common case for Raspberry Pi overlays) because the resulting
path starts with a double slash. For example, if target-path is "/" and
the fragment adds a node called "newnode", the label associated with
that node will be assigned the path "//newnode", which can't be found
in the tree.

Fix the failure case by explicitly replacing a target path of "/" with
an empty string.

Fixes: d1651b03c2df ("of: overlay: add overlay symbols to live device tree")

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agovideo: fbdev: bcm2708_fb: Use common compat header
Phil Elwell [Mon, 2 Mar 2020 14:42:23 +0000 (14:42 +0000)]
video: fbdev: bcm2708_fb: Use common compat header

The definition of compat_ptr is now common for most platforms, but
requires the inclusion of <linux/compat.h>.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agopinctrl: bcm2835: Remove gpiochip on error
Phil Elwell [Mon, 6 Jan 2020 16:04:30 +0000 (16:04 +0000)]
pinctrl: bcm2835: Remove gpiochip on error

A failure in gpiochip_irqchip_add leads to a leak of a gpiochip. Fix
the leak with the use of devm_gpiochip_add_data.

Fixes: 85ae9e512f43 ("pinctrl: bcm2835: switch to GPIOLIB_IRQCHIP")
Signed-off-by: Phil Elwell <phil@raspberrypi.org>
8 months agostaging: vchiq_arm: Give vchiq children DT nodes
Phil Elwell [Mon, 3 Feb 2020 17:30:46 +0000 (17:30 +0000)]
staging: vchiq_arm: Give vchiq children DT nodes

vchiq kernel clients are now instantiated as platform drivers rather
than using DT, but the children of the vchiq interface may still
benefit from access to DT properties. Give them the option of a
a sub-node of the vchiq parent for configuration and to allow
them to be disabled.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
8 months agoInitialise rpi-firmware before clk-bcm2835
Luke Hinds [Wed, 22 Jan 2020 16:03:00 +0000 (16:03 +0000)]
Initialise rpi-firmware before clk-bcm2835

The IMA (Integrity Measurement Architecture) looks for a TPM (Trusted
Platform Module) having been registered when it initialises; otherwise
it assumes there is no TPM. It has been observed on BCM2835 that IMA
is initialised before TPM, and that initialising the BCM2835 clock
driver before the firmware driver has the effect of reversing this
order.

Change the firmware driver to initialise at core_initcall, delaying the
BCM2835 clock driver to postcore_initcall.

See: https://github.com/raspberrypi/linux/issues/3291
     https://github.com/raspberrypi/linux/pull/3297

Signed-off-by: Luke Hinds <lhinds@redhat.com>
Co-authored-by: Phil Elwell <phil@raspberrypi.org>
8 months agoARM: bcm: Backport BCM2711 support from upstream
Stefan Wahren [Fri, 27 Dec 2019 10:40:56 +0000 (11:40 +0100)]
ARM: bcm: Backport BCM2711 support from upstream

Make the BCM2711 a different machine, but keep it in board_bcm2835.

Signed-off-by: Stefan Wahren <wahrenst@gmx.net>
8 months agostaging: vchiq: Use the old dma controller for OF config on platform devices
Dave Stevenson [Tue, 18 Jun 2019 11:15:50 +0000 (12:15 +0100)]
staging: vchiq: Use the old dma controller for OF config on platform devices

vchiq on Pi4 is no longer under the soc node, therefore it
doesn't get the dma-ranges for the VPU.

Switch to using the configuration of the old dma controller as
that will set the dma-ranges correctly.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.org>
8 months agostaging: vchiq_arm: Set up dma ranges on child devices
Dave Stevenson [Tue, 29 Jan 2019 16:13:25 +0000 (16:13 +0000)]
staging: vchiq_arm: Set up dma ranges on child devices

The VCHIQ driver now loads the audio, camera, codec, and vc-sm
drivers as platform drivers. However they were not being given
the correct DMA configuration.

Call of_dma_configure with the parent (VCHIQ) parameters to be
inherited by the child.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.org>
8 months agoclk-raspberrypi: Allow cpufreq driver to also adjust gpu clocks
popcornmix [Mon, 9 Sep 2019 14:49:56 +0000 (15:49 +0100)]
clk-raspberrypi: Allow cpufreq driver to also adjust gpu clocks

For performance/power it is beneficial to adjust gpu clocks with arm clock.
This is how the downstream cpufreq driver works

Signed-off-by: popcornmix <popcornmix@gmail.com>
8 months agov3d_gem: Kick the clock so firmware knows we are using firmware clock interface
popcornmix [Thu, 5 Sep 2019 16:59:14 +0000 (17:59 +0100)]
v3d_gem: Kick the clock so firmware knows we are using firmware clock interface

Setting the v3d clock to low value allows firmware to handle dvfs in case
where v3d hardware is not being actively used (e.g. console use).

Signed-off-by: popcornmix <popcornmix@gmail.com>