Linux uses the prefix "ethernet" whereas u-boot uses "eth". This is from
the linux tree:
$ grep "eth[0-9].*=.*&" arch/**/*dts{,i}|wc -l
0
$ grep "ethernet[0-9].*=.*&" arch/**/*dts{,i}|wc -l
633
In u-boot device trees both prefixes are used. Until recently the only
user of the ethernet alias was the sandbox test device tree. This
changed with commit fc054d563b ("net: Introduce DSA class for Ethernet
switches"). There, the MAC addresses are inherited based on the devices
sequence IDs which is in turn given by the device tree.
Before there are more users in u-boot and both worlds will differ even
more, rename the alias prefix to "ethernet" to match the linux ones.
Also adapt the test cases and rename any old aliases in the u-boot
device trees.
Cc: David Wu <david.wu@rock-chips.com>
Signed-off-by: Michael Walle <michael@walle.cc>
Reviewed-by: Fabio Estevam <festevam@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
If the environment variable "ethact" is not set, the first device in the
uclass is returned. This depends on the probing order of the ethernet
devices. Moreover it is not not configurable at all.
Try to return the ethernet device with sequence id 0 first which then
can be configured by the aliases in a device tree. Fall back to the old
mechanism in case of an error.
Signed-off-by: Michael Walle <michael@walle.cc>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
On systems that use CONFIG_OF_LIVE, the "ofnode" type is defined
as const struct device_node *np, while on the flat DT systems it
is defined as a long of_offset into gd->fdt_blob.
It is desirable that the fixed PHY driver uses the higher-level
ofnode abstraction instead of parsing gd->fdt_blob directly,
because that enables it to work on live OF systems.
The fixed PHY driver has used a nasty hack since its introduction in
commit db40c1aa1c ("drivers/net/phy: add fixed-phy /
fixed-link support"),
which is to pass the long gd->fdt_blob offset inside int phydev->addr
(a value that normally holds the MDIO bus address at which the PHY
responds). Even ignoring the fact that the types were already
mismatched leading to a potential truncation (flat OF offset was
supposed to be a long and not an int), we really cannot extend this
hack any longer, because there's no way an int will hold the other
representation of ofnode, the struct device_node *np.
So we unfortunately need to do the right thing, which is to use the
framework introduced by Grygorii Strashko in
commit eef0b8a930 ("net: phy: add ofnode node to struct phy_device").
This will populate phydev->node for the fixed PHY.
Note that phydev->node will not be valid in the probe function, since
that is called synchronously from phy_device_create and we really have
no way of passing the ofnode directly through the phy_device_create API.
So we do what other drivers do too: we move the OF parsing logic from
the .probe to the .config method of the PHY driver. The new function
will be called at phy_config() time.
I do believe I've converted all the possible call paths for creating
a PHY with PHY_FIXED_ID, so there is really no reason to maintain
compatibility with the old logic of retrieving a flat OF tree offset
from phydev->addr. We just pass 0 to phydev->addr now.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Message-Id: <20210216224804.3355044-2-olteanv@gmail.com>
[bmeng: keep fixedphy_probe(); update mdio-uclass.c to handle fixed phy]
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Priyanka Jain <priyanka.jain@nxp.com>
Switch to use the ofnode_phy_is_fixed_link() API which can support
both the new and old DT bindings.
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Priyanka Jain <priyanka.jain@nxp.com>
Because we probe the master ourselves (and fail if there is no master),
it is not possible that we don't have a master device.
There is one catch though: device removal. We don't support that. It
wasn't supported neither before this patch. Because the master device
was only set in .pre_probe(), if a device was removed master_dev was a
dangling pointer and transmitting a frame cause a panic. I don't see a
good solution without having some sort of notify machanism when a
udevice is removed.
Signed-off-by: Michael Walle <michael@walle.cc>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Tested-by: Michael Walle <michael@walle.cc> [DSA unit tests]
Reviewed-by: Priyanka Jain <priyanka.jain@nxp.com>
Because the uclass has the "*_auto" properties set, the driver model
will take care of allocating the private structures for us and they
can't be NULL. Drop the checks.
Signed-off-by: Michael Walle <michael@walle.cc>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
Reviewed-by: Priyanka Jain <priyanka.jain@nxp.com>
DSA needs to have the master device probed first for MAC inheritance.
Until now, it only works by chance because the only user (LS1028A SoC)
will probe the master device first. The probe order is given by the PCI
device ordering, thus it works because the master device has a "smaller"
BDF then the switch device.
Explicitly probe the master device in dsa_port_probe().
Fixes: fc054d563b ("net: Introduce DSA class for Ethernet switches")
Signed-off-by: Michael Walle <michael@walle.cc>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
Reviewed-by: Priyanka Jain <priyanka.jain@nxp.com>
It doesn't make sense to have DSA without a master port. Error out early
if there is no master port.
Fixes: fc054d563b ("net: Introduce DSA class for Ethernet switches")
Signed-off-by: Michael Walle <michael@walle.cc>
Reviewed-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Ramon Fried <rfried.dev@gmail.com>
Reviewed-by: Priyanka Jain <priyanka.jain@nxp.com>
When trying to port our mpc8309-based board to DM_ETH, on top of
Heiko's patches, I found that nothing in mdio-uclass.c seems to
support the use of a fixed-link subnode of the ethernet DT node. That
is, the ethernet node looks like
enet0: ethernet@2000 {
device_type = "network";
compatible = "ucc_geth";
...
fixed-link {
reg = <0xffffffff>;
speed = <100>;
full-duplex;
};
but the current code expects there to be phy-handle property. Adding
that, i.e.
phy-handle = <&enet0phy>;
enet0phy: fixed-link {
just makes the code break a few lines later since a fixed-link node
doesn't have a reg property. Ignoring the dtc complaint and adding a
dummy reg property, we of course hit "can't find MDIO bus for node
ethernet@2000" since indeed, the parent node of the phy node does not
represent an MDIO bus. So that's obviously the wrong path.
Now, in linux, it seems that the fixed link case is treated specially;
in the of_phy_get_and_connect() which roughly corresponds to
dm_eth_connect_phy_handle() we have
if (of_phy_is_fixed_link(np)) {
ret = of_phy_register_fixed_link(np);
...
} else {
phy_np = of_parse_phandle(np, "phy-handle", 0);
...
}
phy = of_phy_connect(dev, phy_np, hndlr, 0, iface);
And U-Boot's phy_connect() does have support for fixed-link
subnodes. Calling phy_connect() directly with NULL bus and a dummy
address does seem to make the ethernet work.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Heiko Schocher <hs@denx.de>
in tftpboot, if ack was already sent previously for this
packet, don't send again.
Fixes: cc6b87ecaa ("net: tftp: Add client support for RFC 7440")
Reported-by: Suneel Garapati <suneelglinux@gmail.com>
Signed-off-by: Ramon Fried <rfried.dev@gmail.com>
Tested-by: Suneel Garapati <suneelglinux@gmail.com>
Tested-by: Oliver Graute <oliver.graute@kococonnector.com>
DSA stands for Distributed Switch Architecture and it covers switches that
are connected to the CPU through an Ethernet link and generally use frame
tags to pass information about the source/destination ports to/from CPU.
Front panel ports are presented as regular ethernet devices in U-Boot and
they are expected to support the typical networking commands.
DSA switches may be cascaded, DSA class code does not currently support
this.
Signed-off-by: Alex Marginean <alexandru.marginean@nxp.com>
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Move this out of the common header and include it only where needed. In
a number of cases this requires adding "struct udevice;" to avoid adding
another large header or in other cases replacing / adding missing header
files that had been pulled in, very indirectly. Finally, we have a few
cases where we did not need to include <asm/global_data.h> at all, so
remove that include.
Signed-off-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Tom Rini <trini@konsulko.com>
This function may be used in SPL where devicetree is not available.
Use the correct macro so that the function does not try to read it.
Signed-off-by: Simon Glass <sjg@chromium.org>
This reverts commit 1231184caa.
While the change is fine in theory, a number of tests need to be updated
to match.
Signed-off-by: Tom Rini <trini@konsulko.com>
dev_read_alias_seq() used uc_drv->name compared to alias
stem string, Ethernet's alias stem uses "ethernet", which
does not match the eth-uclass driver name "eth", can not
get the correct index of ethernet alias namer. So it seems
change uclass driver name to match the alias stem is a more
reasonable way.
Signed-off-by: David Wu <david.wu@rock-chips.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
When calling srand_mac we use a weak seed dependent on the
mac address. If present, use a RNG device instead to incerase entropy.
Signed-off-by: Matthias Brugger <mbrugger@suse.com>
Reviewed-by: Torsten Duwe <duwe@suse.de>
Fix typo which would cause a build error.
Fixes: 3eaac6307d ("net: introduce packet capture support")
Signed-off-by: Jorge Ramirez-Ortiz <jorge@foundries.io>
When the tftp server did not send any OACK, the tftp_next_ack variable
was not set to the correct value . As the server was transmitting
blocks we generated a lot of 'Received unexpected block: $n, expected
$n+1' error messages. Depending on the timeout setting the transfer
could still complete though.
Signed-off-by: Harm Berntsen <harm.berntsen@nedap.com>
CC: Ramon Fried <rfried.dev@gmail.com>
Reviewed-By: Ramon Fried <rfried.dev@gmail.com>
While doing DHCP the interface IP is set to 0.0.0.0. This causes the
check in net.c on dst_ip to be effectively skipped, and all IP datagrams
are accepted up the IP stack. In the case of an ICMP_ECHO_REQUEST for the
matching MAC address (regardless of destination IP), the result is that
an ICMP_ECHO_REPLY is sent. The source address of the ICMP_ECHO_REPLY is
0.0.0.0, which is an illegal source address.
This can happen in common practice with the following sequence:
DHCP (U-Boot or OS) acquires IP address 10.0.0.1
System reboots
U-Boot starts DHCP and send DHCP DISCOVER
DHCP server decides to OFFER 10.0.0.1 again
(perhaps because of existing lease or manual configuration)
DHCP server tries to PING 10.0.0.1 to see if anyone is squatting on it
DHCP server still has our MAC address in its ARP table for 10.0.0.1
U-Boot receives PING, and responds with an illegal source address
This may further result in a the DHCP server seeing the response as
confirmation that someone is squatting on 10.0.0.1, and picking a
new IP address from the pool to try again
Signed-off-by: David Rivshin <drivshin@allworx.com>
Running the start() handler twice without a stop() inbetween completely
breaks communication for some ethernet drivers like fec_mxc.
eth_halt() is called before each eth_init(). Due to the switch to
eth_is_active() in commit 68acb51f44 ("net: Only call halt on a driver
that has been init'ed"), this is not sufficient anymore when netconsole
is active: eth_init_state_only()/eth_halt_state_only() manipulate the
state check that is performed by eth_is_active() without actually
calling into the driver.
The issue can be triggered by starting a network operation (e.g. ping or
tftp) while netconsole is active.
Add an additional "running" flag that reflects the actual state of the
driver and use it to ensure that eth_halt() actually stops the device as
it is supposed to.
Fixes: 68acb51f44 ("net: Only call halt on a driver that has been init'ed")
Signed-off-by: Matthias Schiffer <matthias.schiffer@ew.tq-group.com>
Up to now the bootefi command used the last file loaded to determine the
boot partition. This has led to errors when the fdt had been loaded from
another partition after the EFI binary.
Before setting the boot device from a loaded file check if it is a PE-COFF
image or a FIT image.
For a PE-COFF image remember address and size, boot device and path.
For a FIT image remember boot device and path.
If the PE-COFF image is overwritten by loading another file, forget it.
Do not allow to start an image via bootefi which is not the last loaded
PE-COFF image.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
At present ofnode is present in the device even if it is never used. With
of-platdata this field is not used, so can be removed. In preparation for
this, change the access to go through inline functions.
Signed-off-by: Simon Glass <sjg@chromium.org>
Most drivers use these access methods but a few do not. Update them.
In some cases the access is not permitted, so mark those with a FIXME tag
for the maintainer to check.
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Acked-by: Pratyush Yadav <p.yadav@ti.com>
Now that there is only one sequence number (rather than both requested and
assigned ones) we can simplify this function. Also update its caller to
simplify the logic.
Signed-off-by: Simon Glass <sjg@chromium.org>
Checking for seq == -1 is effectively checking that the device is
activated. The new sequence numbers are never -1 for a bound device, so
update the check.
Also drop the note about valid sequence numbers so it is accurate with the
new approach.
Signed-off-by: Simon Glass <sjg@chromium.org>
At present various drivers etc. access the device's 'seq' member directly.
This makes it harder to change the meaning of that member. Change access
to go through a function instead.
The drivers/i2c/lpc32xx_i2c.c file is left unchanged for now.
Signed-off-by: Simon Glass <sjg@chromium.org>
We use 'priv' for private data but often use 'platdata' for platform data.
We can't really use 'pdata' since that is ambiguous (it could mean private
or platform data).
Rename some of the latter variables to end with 'plat' for consistency.
Signed-off-by: Simon Glass <sjg@chromium.org>
This construct is quite long-winded. In earlier days it made some sense
since auto-allocation was a strange concept. But with driver model now
used pretty universally, we can shorten this to 'auto'. This reduces
verbosity and makes it easier to read.
Coincidentally it also ensures that every declaration is on one line,
thus making dtoc's job easier.
Signed-off-by: Simon Glass <sjg@chromium.org>
CONFIG_TIMESTAMP is not related to the RTC drivers. It does not make any
sense to let the updating of the RTC by the sntp command depend on it.
Drop the CONFIG_TIMESTAMP checks.
Furthermore function dm_rtc_set() is enabled by CONFIG_DM_RTC. There is no
reason to require CONFIG_CMD_DATE when using a driver model RTC. The UEFI
sub-system can consume the RTC functions even if there is not date command.
Only check CONFIG_CMD_DATE when using a non-driver model RTC.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
- ProxyDHCP allows a second DHCP server to exist alongside your main
DHCP server and supply additional BOOTP related options
- When u-boot sends out a DHCP request, the real DHCP server will
respond with a normal response containing the new client IP address
while simultaneously the ProxyDHCP server will respond with a blank
client IP address and a `bootfile` option
- This patch adds CONFIG_SERVERIP_FROM_PROXYDHCP (default false) to
enable this behavior and CONFIG_SERVERIP_FROM_PROXYDHCP_DELAY_MS
(default 100) which tells u-boot to wait additional time after
receiving the main DHCP response to give the ProxyDHCP response time
to arrive
- The PXE spec for ProxyDHCP is more complicated than the solution
added here as diagramed on page 16:
http://www.pix.net/software/pxeboot/archive/pxespec.pdf:
```
DHCP Discover will be retried four times. The four timeouts are 4, 8, 16
and 32 seconds respectively. If a DHCPOFFER is received without an Option
timeouts in an attempt to receive a PXE response.
```
- Adding a simple delay worked for my purposes but let me know if a
more robust solution is required
Signed-off-by: Lyle Franklin <lylejfranklin@gmail.com>
With a define for the magic number of packets received as batch
we can make sure that the EFI network stack caches the same amount
of packets.
Signed-off-by: Patrick Wildt <patrick@blueri.se>
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
net_init does not always succeed, and there is no existing mechanism to
discover errors. This patch allows callers of net_init (such as net_init)
to handle errors. The root issue is that eth_get_dev can fail, but
net_init_loop doesn't expose that. The ideal way to fix eth_get_dev would
be to return an error with ERR_PTR, but there are a lot of callers, and all
of them just check if it's NULL. Another approach would be to change the
signature to something like
int eth_get_dev(struct udevice **pdev)
but that would require rewriting all of the many callers.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This commits update the support of sntp to use
the framework udp. This change allows to remove
all the reference to sntp in the main network
file net/net.c.
Signed-off-by: Philippe Reynes <philippe.reynes@softathome.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
This commit adds a generic udp protocol framework in the
network loop. So protocol based on udp may be implemented
without modifying the network loop (for example custom
wait magic packet).
Signed-off-by: Philippe Reynes <philippe.reynes@softathome.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Write the 'No ethernet found' message via the log drivers. This allows
suppressing it during output via the syslog driver.
This fixes the problem reported in:
[PATCH 0/4] log: Fix the syslog spam when running tests
https://lists.denx.de/pipermail/u-boot/2020-September/426343.html
Reported-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
When load the last block, the "len" might not be a block size. This cause
loading the incorrect last block data.
The fix change "len" to tftp_block_size and minus one tftp_block_size
for offset calculation.
Use same offset calculation formula as in store_block().
Signed-off-by: Ley Foon Tan <ley.foon.tan@intel.com>
Reviewed-By: Ramon Fried <rfried.dev@gmail.com>
tftp_cur_block start with 1 for first block, but tftp_cur_block counter is
start with zero when block number is rollover. The existing code
"tftp_cur_block - 1" will cause the block number become -1 in store_block()
when tftp_cur_block is 0 when tftp_cur_block is rollover.
The fix pass in tftp_cur_block to store_block() and minus the
tftp_block_size when do the offset calculation.
Signed-off-by: Ley Foon Tan <ley.foon.tan@intel.com>
Reviewed-By: Ramon Fried <rfried.dev@gmail.com>
Fixes missing update to tftp_prev_block counter before increase
tftp_cur_block counter when do the tftpput operation.
tftp_prev_block counter is used in update_block_number() function to
check whether block number (sequence number) is rollover. This bug
cause the tftpput command fail to upload a large file when block
number is greater than 16-bit (0xFFFF).
Signed-off-by: Ley Foon Tan <ley.foon.tan@intel.com>
Reviewed-By: Ramon Fried <rfried.dev@gmail.com>
The name of the device we are working on is `ethdev` and not just `dev`.
Signed-off-by: Sean Anderson <seanga2@gmail.com>
Tested-by: Patrick Delaunay <patrick.delaunay@st.com>
Android 10 adds support for dynamic partitions and in order to support
this userspace fastboot must be used[1]. New tool fastbootd is
included into recovery.
Userspace fastboot works from recovery and is launched if:
1) - Dynamic partitioning is enabled
2) - Boot control block has 'boot-fastboot' value into command field
The bootloader is expected to load and boot into the recovery image
upon seeing boot-fastboot in the BCB command. Recovery then parses the
BCB message and switches to fastbootd mode[2].
Please note that boot script is expected to handle 'boot-fastboot'
command in BCB and load into recovery mode.
Bootloader must support 'reboot fastboot' command which should reboot
device into userspace fastboot to accomodate those changes[3].
Another command that bootloader must support[3] is 'reboot recovery'. This
command should simply reboot device into recovery mode.
[1] - https://source.android.com/devices/bootloader/fastbootd
[2] - https://source.android.com/devices/bootloader/fastbootd#unified_fastboot_and_recovery
[3] - https://source.android.com/devices/bootloader/fastbootd#modifications_to_the_bootloader
Signed-off-by: Roman Kovalivskyi <roman.kovalivskyi@globallogic.com>
Signed-off-by: Roman Stratiienko <r.stratiienko@gmail.com>
Change-Id: I9d2bdc9a6f6f31ea98572fe155e1cc8341e9af76
Reset the stored ping IP address before entering a netloop with different
protocol to ensure that it won't be interrupted by the received
correct ICMP_ECHO_REPLY packet.
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Add support for RFC 7440: "TFTP Windowsize Option".
This optional feature allows the client and server
to negotiate a window size of consecutive blocks to send as an
alternative for replacing the single-block lockstep schema.
windowsize can be defined statically during compilation by
setting CONFIG_TFTP_WINDOWSIZE, or defined in runtime by
setting an environment variable: "tftpwindowsize"
If not defined, the windowsize is set to 1, meaning that it
behaves as it was never defined.
Choosing the appropriate windowsize depends on the specific
network topology, underlying NIC.
You should test various windowsize scenarios and see which
best work for you.
Setting a windowsize too big can actually decreases performance.
Signed-off-by: Ramon Fried <rfried.dev@gmail.com>
Reviewed-by: Marek Vasut <marex@denx.de>
This header file should not be included in other header files. Remove it
and use other headers and C inclusions instead.
Signed-off-by: Simon Glass <sjg@chromium.org>
The Linux coding style guide (Documentation/process/coding-style.rst)
clearly says:
It's a **mistake** to use typedef for structures and pointers.
Besides, using typedef for structures is annoying when you try to make
headers self-contained.
Let's say you have the following function declaration in a header:
void foo(bd_t *bd);
This is not self-contained since bd_t is not defined.
To tell the compiler what 'bd_t' is, you need to include <asm/u-boot.h>
#include <asm/u-boot.h>
void foo(bd_t *bd);
Then, the include direcective pulls in more bloat needlessly.
If you use 'struct bd_info' instead, it is enough to put a forward
declaration as follows:
struct bd_info;
void foo(struct bd_info *bd);
Right, typedef'ing bd_t is a mistake.
I used coccinelle to generate this commit.
The semantic patch that makes this change is as follows:
<smpl>
@@
typedef bd_t;
@@
-bd_t
+struct bd_info
</smpl>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Similar to patch 821fec0ceb ("net: remove scary warning about EEPROM
provided MAC address") this removes the somewhat awkward "warning" on
boards using DM_ETH:
In many parts of the computing world having a unique MAC address
sitting in some on-NIC storage is considered the normal case.
If there is a properly provided MAC address (either from ROM or from DT),
remove the warning to not scare the user unnecessarily.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-By: Ramon Fried <rfried.dev@gmail.com>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
The random_port() routine is not used anywhere else. Make it local to
dns.c to reduce code clutter, and shrink generated code a little.
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
RFC2348, RFC2349:
- Option string is case in-sensitive.
- Client must generate ERR pkt in case option value mismatch in server OACK
- Fix debug print for options
Signed-off-by: Ravik Hasija <rahasij@linux.microsoft.com>
Reviewed-By: Ramon Fried <rfried.dev@gmail.com>
During packet sequence number wraparound the show_block_marker() API was
not called, as a result the progress marker doesn't stay within column
boundary. Use position in file instead of sequence number to align the
marker.
Signed-off-by: Ravik Hasija <rahasij@linux.microsoft.com>
Reviewed-By: Ramon Fried <rfried.dev@gmail.com>
