The SerDes mode setting at the end of rtpcs_931x_setup_serdes is
currently broken although it is mostly similar to what the SDK does [1].
It prevents several modes from being set, especially fiber modes. This
seems to be one (if not the only) issue for currently missing SFP
support.
Add a small helper function which takes cares of setting the mode, to
keep the still valid different procedure when using XSGMII mode. Only
this helper is called in rtpcs_931x_setup_serdes to keep it clean there.
As a functional change, call mode application in every case, not just
for SGMII, QSGMII and USXGMII. We can assume the SDK is misleading in
this case, either accidentially or on purpose. This makes SFP modules
work in theory. In practice, there still seem to be device-specific
issues which need to be fixed later. These issues may include no link
detection or link flapping.
[1] f7f85ffc14/sources/rtk-dms1250/src/dal/mango/dal_mango_construct.c (L2266)
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/20736
Signed-off-by: Robert Marko <robimarko@gmail.com>
The SDK and our code for finally applying the hardware mode are quite
confusing. There are two different "places" where a mode can be set,
in a SerDes register and in a global SerDes mode register. Neither the
SDK nor any of the datasheet/documentation serve any explanation for
that. The functions are just named "fiber_mode_set" and "mii_mode_set"
which is basically as useless as it can be to understand it.
Try to get rid off this confusion by naming the functions
'sds_set_mac_mode' and 'sds_set_ip_mode' to make clear where the mode
is set. While at it, also clarify the naming of 'config_mode' by
renaming it to 'config_hw_mode'.
The naming is based on the following assumption:
> Realtek uses an SerDes IP core design (probably from another vendor)
> in their switch. This supports a variety of modes and must be
> configured properly for each mode. Usually, changing the mode in the
> MAC's registers triggers a proper configuration of the SerDes IP block
> in the background.
> However, for some modes this seems to be incomplete, at least missing
> important parts so it doesn't work on its own in the end. In this
> case, the SerDes IP block needs to be configured manually with the
> missing bits to make it work.
There are several places in the SDK that support this assumption, both
for RTL931X and RTL930X (as they are somewhat similar), e.g. [1].
[1] f7f85ffc14/sources/rtk-dms1250/src/dal/longan/dal_longan_sds.c (L1746)
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/20736
Signed-off-by: Robert Marko <robimarko@gmail.com>
Improve the current mode setting functions to address a few issues:
- add missing mode from SDK reference (instead of using the
corresponding value as a default value)
- use and return error values
- give internal variable a more meaningful name
- consistently use lowercase letters in hex values
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/20736
Signed-off-by: Robert Marko <robimarko@gmail.com>
Merge the unused helper 'rtpcs_931x_sds_fiber_disable' into Fiber mode
setting, and drop the helper itself. As with the MII helper in a
previous commit, functionality is basically the same just with a value
for OFF mode. If functionality is required later, Fiber mode setting can
be used with the OFF mode instead of carrying the unused helper until
it's used.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/20736
Signed-off-by: Robert Marko <robimarko@gmail.com>
Use the hardware mode instead of PHY_INTERFACE_MODE_* mode for mode
setting. Adjust all affected functions and switch cases accordingly.
In MII mode setting, drop the case for 2500Base-X as this is a mistake.
In the SDK reference code, this doesn't exist [1].
[1] 69d2890a2e/sources/rtk-dms1250/src/hal/phy/phy_rtl9310.c (L817)
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/20736
Signed-off-by: Robert Marko <robimarko@gmail.com>
Merge the 'rtpcs_931x_sds_disable' helper into the MII mode setting and
use that instead. The helper was essentially doing same just with a
value for OFF state.
Moreover, the name of the helper was confusing. It implied it disables
the whole SerDes. However, this is used in Fiber mode setting and thus,
cannot completely disable the SerDes.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/20736
Signed-off-by: Robert Marko <robimarko@gmail.com>
The function 'rtpcs_931x_sds_mii_mode_set' does not correctly write the
register. It just write a plain value at the determined register
address. While this works for SerDes with (id mod 4 == 0), it doesn't
for the other SerDes.
Fix that by using a corresponding shift and writing only some bits
instead of the whole register.
While at it, drop an unneeded blank line, add comment to explain a bit
that is set and use the BIT(..) helper for that instead of manual shift.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/20736
Signed-off-by: Robert Marko <robimarko@gmail.com>
Bring the RTL931X functions into a proper order for two purposes:
- be able to reuse code
- have blocks of logically connected functions
This just moves code, no functional changes.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/20736
Signed-off-by: Robert Marko <robimarko@gmail.com>
Fix the implementation to clear symbol errors. Up to now, it was barely
functional because important modes weren't handled. Add another function
(which is needed in the future anyway) and pick to missing bits from
that function into the symbol clearing function to handle the modes
1000Base-X and 10GBaseR too. All that is based on [1].
While at it, rename the function to adhere to a common naming scheme and
fix some minor style issues.
[1] 69d2890a2e/sources/rtk-dms1250/src/hal/phy/phy_rtl9310.c (L703)
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/20736
Signed-off-by: Robert Marko <robimarko@gmail.com>
Now that the bus is completely configured by the dts
compatible, drop the hard coded family detection.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21274
Signed-off-by: Robert Marko <robimarko@gmail.com>
Place the reset function into the config structure.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21274
Signed-off-by: Robert Marko <robimarko@gmail.com>
There is no need to give the mdio bus a family dependent name.
Name the bus similar to the SerDes mdio bus.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21274
Signed-off-by: Robert Marko <robimarko@gmail.com>
Move the read/write functions to where they belong.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21274
Signed-off-by: Robert Marko <robimarko@gmail.com>
For all families the driver uses the same read/write functions.
Drop the duplicate per-family initialization.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21274
Signed-off-by: Robert Marko <robimarko@gmail.com>
Relocate the cpu port definition over to the new configuration
structure. There are several family specific places where the
driver checks port ranges. These are a wild mix of arbitrary
values (64) or constant defines. Remove them as the the central
read/write functions already have a proper check with -ENODEV
bail out in place.
With the cleanup drop the port defines as there is only one
consumer (config structure) left.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21274
Signed-off-by: Robert Marko <robimarko@gmail.com>
The mdio driver already has a family specific compatible that
is used in the dts to denote the device type. Nevertheless it
is not evaluated. The driver determines the family on its own
in rtmdio_get_family().
Break up this dependency by providing a configuration structure
that can be automatically assigned by device_get_match_data().
For now only move the rawpage variable over there.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21274
Signed-off-by: Robert Marko <robimarko@gmail.com>
The ethernet and dsa drivers still rely on several defines from
the central mach include. Move important defines over to the
driver specific includes as a first decoupling step.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21263
Signed-off-by: Robert Marko <robimarko@gmail.com>
RTL930x devices have highmem starting address at 0x20000000.
The Linksys LGS328C highmem definition is wrongly shared with
the larger LGS352C RTL931x model and starts at 0x90000000.
Fix it by splitting the definition.
Fixes: 853d73f ("realtek: add support for Linksys LGS328C")
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21262
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Add a dedicated helper 'rtpcs_931x_sds_power' to power down/up a SerDes.
Up to now, plain register writes to the corresponding register were used
in several places and with slightly different syntax. Use the helper to
replace the existing boilerplate code for powering on/off SerDes.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21184
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Move the configuration of 1000Base-X mode into its own function. This
way, redundant code can be replaced by a single function call.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21184
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Reorder the modes within the switch statement in
rtpcs_931x_sds_config_mode to have a proper and non-confusing order.
Have fiber modes first and MII modes after, and sort all modes in
ascending speed/capabilities order.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21184
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Pass the SerDes hardware mode to rtpcs_931x_sds_config_mode instead of
the PHY_INTERFACE_MODE_* mode and use it. Replace matched modes
accordingly.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21184
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
The PCS driver is intended to use a dedicated hardware mode in the
future because the hardware capabilities and modes of the SerDes do not
align with the PHY_INTERFACE_MODE_* modes.
For the beginning, use the previously introduced generic mode mapping
in the SerDes setup and assign the determined hardware mode to the
corresponding field of a SerDes instance. This allows parallel usage of
both mode representations for the time being, until the phy_interface_t
modes can be dropped from internal functions completely.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21184
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Slim the rtpcs_931x_setup_serdes entrypoint by moving the SerDes
configuration sequences for the different SerDes modes into a dedicated
function called rtpcs_931x_sds_config_mode. They form a logical block
similar to what the SDK does and can be taken out to follow the
'divide & conquer' principle.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21184
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Configuring USXGMII on RTL931X was disable because it is not setup
properly right now. Affected devices need to rely on U-boot setup for
now. However, it is not the proper way to return an error in case we
just want to skip USXGMII configuration. Thus, just return 0 in that
case indicating it is properly configured.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21184
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Add a generic mapper for mapping the PHY_INTERFACE_MODE_* modes to the
SerDes hardware modes RTPCS_SDS_MODE_*. This is to be used by all
subtargets step-by-step and avoids having subtarget-specific code for
that simple mapping.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21184
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Rename the mode field in rtpcs_serdes from 'mode' to 'hw_mode' to make
clear what it actually is, to avoid confusion with the phy_interface_t
'mode' usually passed to functions.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21184
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
The ethernet driver gets all the data from the DTS now. No need
to access the shared soc_info structure any longer.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21183
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
family_id is now part of the register set and automatically assigned
during initialization. Make use of it. This is basically a conversion
from priv->family_id to priv->r->family_id.
While we are here convert some hard coded family ids to their proper
defines.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21183
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Now that the register configuration contains the family data derive it
from the DTS via device_get_match_data().
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21183
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
The ethernet driver currently determines the family from the soc_info
structure. To avoid the dependency add the family to the local register
structure.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21183
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
The ethernet driver uses registers in the switchcore range.
Rearrange the DTS nodes accordingly. This allows to make use
of regmap with syscon_node_to_regmap(np->parent) later.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21183
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
The ethernet driver currently "requests" the memory region of the
ethernet device according to the DTS settings. This is wrong in
two ways:
- The DTS uses arbitrary regions out of the switchcore memory
- The variable is afterwards overwritten by DMA allocation
Drop it.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21183
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
The Realtek Otto ethernet driver currently uses a single compatible
for all different models. Split this into the the four well known
subtargets. This allows to get rid of the central mach/soc include
later.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21183
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
With kernel 6.18 the package shared structure will be hidden from
the phy drivers. Workaround that in advance by providing a shared
private structure that covers all information that is needed for
normal operation.
For further simplification provide a new join() function that takes
care of a consistent initialization.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/21204
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Add a new recipe 'rt-loader-no-uimage' that passes the kernel load
address to rt-loader, causing it to use that instead of it's initial run
address.
The usual behavior is fine for uImages where the load address is
predefined in the header, U-boot loads the image to that address and
then runs it, rt-loader just takes over that address. For non-uImage
instead, the address is tightly coupled to where the image has been
transferred during serial or TFTP upload. This may not be possible on
several devices. Passing a separate kernel load address to rt-loader
decouples that and avoids taking the pain to change the load address of
the kernel itself.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21248
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
rt-loader currently has two operation modes, piggy-backed and
standalone. In standalone mode, the kernel load address is read from the
uImage in flash. In piggy-backed mode, rt-loader instead uses its
initial run address (aka run address during first run) as the kernel
load address. This is safe and works fine for all devices either using
U-boot or having no issue uploading an image to the default kernel load
address 0x80100000.
To extend usecases, allow to specify a kernel load address when
building rt-loader. In this case, rt-loader uses this address instead of
the address inferred at runtime.
On certain Zyxel devices, this allows to upload and boot an rt-loader
piggy-backed image to an alternate address but keep the default kernel
load address of 0x80100000. BootExt on these devices occupies memory
above and will crash during transfer when this address is used as upload
location. Using this extension, the image can be uploaded to e.g.
0x80300000 and rt-loader will use 0x80100000 as the final load address.
This avoid taking the pain the adjust the load address of the kernel
itself.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21248
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Use the correct identifier 'rtsds_of_match' instead of
'rtsds_mdio_of_match' because the latter doesn't exist.
This doesn't cause an error for 6.12. However, with 6.18 the
implementation of MODULE_DEVICE_TABLE has changed to use 'static' and
'used' [1] instead of 'extern' and 'unused' [2].
[1] 7d0a66e4bb/include/linux/module.h (L260)
[2] adc218676e/include/linux/module.h (L249)
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21182
Signed-off-by: Robert Marko <robimarko@gmail.com>
As a first real usage of the new SerDes struct, move the polarity
configuration there. It was previously located in the global rtpcs_ctrl
struct as an array, indexed by SerDes id. Because this is per-SerDes
information, the new SerDes struct is the correct place to live in.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21146
Signed-off-by: Robert Marko <robimarko@gmail.com>
By using references to pre-initiated SerDes instances instead of plain
SerDes number, there is no need to check for the range anymore in
various places. During driver/pcs init it is ensured that only valid
SerDes will reach the configuration functions.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21146
Signed-off-by: Robert Marko <robimarko@gmail.com>
Also switch set_autoneg (and related helper rtpcs_sds_modify) to the
SerDes struct instead of the plain SerDes id by using just the reference
to the SerDes instance instead of (ctrl, sds_id) tuple. This completes
the transition.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21146
Signed-off-by: Robert Marko <robimarko@gmail.com>
Make use of the previously added SerDes struct in SerDes setup and all
functions in its call path by removing (ctrl, sds_num) being passed to
every function call and instead just pass the reference to the
corresponding SerDes instance.
Various SerDes calculations for even, odd and neighbor are unified by
switching to previously introduced helpers.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21146
Signed-off-by: Robert Marko <robimarko@gmail.com>
Drop usage of the to-be-phased-out SerDes id stored in rtpcs_link and
use the reference to the SerDes instance to use the embedded id in
rtpcs_serdes instead.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21146
Signed-off-by: Robert Marko <robimarko@gmail.com>
Upon creation of a phylink_pcs instance by calling rtpcs_create, assign
a reference to the corresponding SerDes to the link structure. In the
next step, this should be used everywhere instead of the plain SerDes
number.
Rename the field used to hold the SerDes number from 'sds' to 'sds_num'
and name the new field 'sds' to make clear what is what.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21146
Signed-off-by: Robert Marko <robimarko@gmail.com>
Add dedicated helpers to get references to even, odd and neigbor SerDes
if needed. This should replace the various calculations scattered
throughout the code, providing a unified way to work with adjacent
SerDes.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21146
Signed-off-by: Robert Marko <robimarko@gmail.com>
Add a separate structure for a SerDes. This is needed to appropriately
store per-SerDes information, which in turn is needed for future work.
Additionally, it's intended to reduce boilerplate and several
inconsistencies.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21146
Signed-off-by: Robert Marko <robimarko@gmail.com>
Use a separate configuration field for the number of SerDes for each
variant of the Realtek Otto family. Add this field to the config
structure, assign it and use it during driver probe. This narrows
possible error cases and is needed for upcoming extensions.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21146
Signed-off-by: Robert Marko <robimarko@gmail.com>
The Realtek SerDes mode capabilities do not map 1:1 to the
PHY_INTERFACE_MODE_* modes used in the kernel and passed to the PCS.
For example, some PHY chips use the proprietary XSGMII mode for which
there isn't an equivalent in the kernel, or HSGMII.
In the past, this led to problems and confusion using kernel's XGMII to
handle the XSGMII mode, and needed a downstream patch for HSGMII. They
have been solved/worked around for now, but XSGMII is currently not
implemented at all. And who knows what might come in the future.
To make our life easier, introduce a dedicated internal representation
of SerDes modes which differs from kernel's PHY_INTERFACE_MODE_*. This
allows us to map "external" modes to different internal modes as needed
instead of carrying the PHY_INTERFACE_MODE_* through the whole SerDes
configuration code. The PCS driver needs to map PHY_INTERFACE_MODE_* to
RTPCS_SDS_MODE_* in pcs_config, and the latter should be used as the
only one.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21146
Signed-off-by: Robert Marko <robimarko@gmail.com>
Drop the unused and broken function rtpcs_930x_sds_clock_wait from the
PCS driver. The proper working variant is already some lines above and
called rtpcs_930x_sds_wait_clock_ready.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21146
Signed-off-by: Robert Marko <robimarko@gmail.com>
Since the beginning, the PCS driver had the ability to call its
rtpcs_create without a reference to a valid PCS node. A comment in the
code mentions that this is done for RTL838X and its built-in octa-PHY
which is connected directly instead of via a SerDes. Further
explanations are not provided.
Drop this ability and make the rtpcs_create call in the dsa driver
conditional. As the built-in PHY of RTL838X isn't attached to a SerDes,
there is no obvious point of having the PCS driver in that chain. The
ports are marked as internal and have no pcs-handle, thus no phylink_pcs
instance should be created.
Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Link: https://github.com/openwrt/openwrt/pull/21146
Signed-off-by: Robert Marko <robimarko@gmail.com>
