Adds the ability for I2C drivers to report synchronous transfer stats
using a I2C specific macro to define the device instance.
The macro creates a container for device_state which allows for per
instance device class common data structure to be used in the device
class api (ex: i2c.h). This is used to maintain per driver instance
stats for all i2c drivers. This is a reusable idea across other device
classes as desired.
Using Kconfig device class stats may be turned on/off individually
this way as well, in this case I2C_STATS.
Signed-off-by: Tom Burdick <thomas.burdick@intel.com>
- Rename to "action" to make its purpose more clear
- Use the _cb suffix to align with naming used for callbacks in other
areas.
Signed-off-by: Gerard Marull-Paretas <gerard.marull@nordicsemi.no>
The NET_DEVICE_DT_DEFINE_INSTANCE() macro was passing an invalid
dev_name to Z_NET_DEVICE_INIT_INSTANCE(). Change this parameter to get
device name based on the node_id, matching other macros.
Signed-off-by: Keith Short <keithshort@google.com>
This migrates all the current iterable section usages to the external
API, dropping the "Z_" prefix:
Z_ITERABLE_SECTION_ROM
Z_ITERABLE_SECTION_ROM_GC_ALLOWED
Z_ITERABLE_SECTION_RAM
Z_ITERABLE_SECTION_RAM_GC_ALLOWED
Z_STRUCT_SECTION_ITERABLE
Z_STRUCT_SECTION_ITERABLE_ALTERNATE
Z_STRUCT_SECTION_FOREACH
Signed-off-by: Fabio Baltieri <fabiobaltieri@google.com>
The k_work handler cannot manipulate the used k_work. This means
that it is not easy to cleanup the net_pkt because it contains
k_work in it. Because of this, use k_fifo instead between
RX thread and network driver, and between application and TX
thread.
A echo-server/client run with IPv4 and UDP gave following
results:
Using k_work
------------
TX traffic class statistics:
TC Priority Sent pkts bytes time
[0] BK (1) 21922 5543071 103 us [0->41->26->34=101 us]
[1] BE (0) 0 0 -
RX traffic class statistics:
TC Priority Recv pkts bytes time
[0] BK (0) 0 0 -
[1] BE (0) 21925 6039151 97 us [0->21->16->37->20=94 us]
Using k_fifo
------------
TX traffic class statistics:
TC Priority Sent pkts bytes time
[0] BK (1) 15079 3811118 94 us [0->36->23->32=91 us]
[1] BE (0) 0 0 -
RX traffic class statistics:
TC Priority Recv pkts bytes time
[0] BK (1) 0 0 -
[1] BE (0) 15073 4150947 79 us [0->17->12->32->14=75 us]
So using k_fifo gives about 10% better performance with same workload.
Fixes#34690
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Do not try to set or get the interface MTU if the interface
pointer is NULL.
Coverity-CID: 220541
Fixes#34000
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Before this change all struct net_if objects accessed by for example
net_if_get_by_index() are placed in one linker area (i.e. net_if_area)
with the same "name" - '_net_if.static.net_if'. This may cause problems
when the order of struct net_if elements is important.
With the same names for all elements there is no guarantee of placement
order. After this change the unique device name is appended, so
SORT_BY_NAME() linker command places objects in reproductible manner.
Signed-off-by: Lukasz Majewski <lukma@denx.de>
This can be used to implement tunneling, VPN etc. The virtual
interfaces can be chained together to support multilayer
network interfaces.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Application can disable IPv4 or IPv6 later if those are not
needed nor used for a given network interface.
Fixes#14581
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
As we move to making the `label` property optional, we need to account
for widespread use of `dev->name` in contexts where the pointer is not
validated before it is dereferenced. Use an empty string instead of a
null pointer when no label is provided.
Signed-off-by: Peter Bigot <peter.bigot@nordicsemi.no>
Generated device and driver names have changed with commits [1] and [2],
but net offload drivers missed the conversion. Update net offload DT
device macros now.
[1] commit 8c1bef535b ("device: support generating defines from
devicetree nodes with no label")
[2] commit f91e9fba51 ("device: fix potential truncation of DT-derived
device names")
Signed-off-by: Marcin Niestroj <m.niestroj@grinn-global.com>
While using the encoded path to a device tree node guarantees a unique
identifier for the corresponding device there is a limit on the number
of characters of that name that can be captured when looking up a
device by name from user mode, and the path can exceed that limit.
Synthesize a unique name from the node dependency ordinal instead, and
update the gen_defines script to record the name associated with the
full path in the extern declaration.
Add a build-time check that no device is created with a name that
violates the user mode requirement.
Also update the network device DTS helper functions to use the same
inference for dev_name and label that the real one does, since they
bypass the real one.
Signed-off-by: Peter Bigot <peter.bigot@nordicsemi.no>
The proper usage of @param in callbacks is currently unsupported by
Doxygen so not warnings are generated. The issues fixed with this commit
where found while adding support to Doxygen for validating @param in
callbacks like it currently does for functions.
Signed-off-by: Fabio Utzig <fabio.utzig@nordicsemi.no>
Provide versions of NET_DEVICE_INIT, NET_DEVICE_INIT_INSTANCE, and
NET_DEVICE_OFFLOAD_INIT that are both devicetree node and instance
aware. We use the _DEFINE suffix for the DT versions to match the
naming convention in device.h
Signed-off-by: Kumar Gala <kumar.gala@linaro.org>
The "tc" field which was holding the traffic class thread
priority is not used nor needed so remove it from the struct
to save some space.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Now that device_api attribute is unmodified at runtime, as well as all
the other attributes, it is possible to switch all device driver
instance to be constant.
A coccinelle rule is used for this:
@r_const_dev_1
disable optional_qualifier
@
@@
-struct device *
+const struct device *
@r_const_dev_2
disable optional_qualifier
@
@@
-struct device * const
+const struct device *
Fixes#27399
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
The indentation went wrong when the integer types was changed
from the type "u8_t" to type "uint8_t". This changed the length
of the type and caused the code to look bad in places.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
net: route: Add prefix-based ipv6 multicast forwarding
This adds/reenables the feature of multicast routing/forwarding.
The forwarding decision is based on the added multicast routes
and the new network interface flag:
NET_IF_FORWARD_MULTICASTS.
Signed-off-by: Jan Georgi <jan.georgi@lemonbeat.com>
There use to be a config attribute of type device_config, on which
config_info could be found. config_info was thus named that way to
differentiate easily from config attribute in struct device.
Now that there is no such device_config structure, config_info in
structure device now, can be renamed to config.
Semantically, it makes for sense. We have an attribute pointing to the
configuration of the device driver instance. Configuration information
is correct but has a redundant meaning.
Fixes#27397
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Make sure that only those threads that have been granted access
to net_if objects, can call the functions that modify net_if data.
The CONFIG_NET_IF_USERSPACE_ACCESS config option is also removed
as it is no longer needed after this change.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
The NET_IF_OFFLOAD_INIT() macro was not converted to use
Z_STRUCT_SECTION_ITERABLE() macros in commit 62580748e5
("net: Use section foreach macros"). This meant that for those boards
that use offloaded network stack, the network interface was not created
and network connectivity was lost.
Fixes#27037
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
The Z_STRUCT_SECTION_ITERABLE() provides an alignment so remove
the 32 byte alignment for net_if and ppp_protocol_handler structs.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Use system provided Z_STRUCT_SECTION_FOREACH() and
Z_STRUCT_SECTION_ITERABLE() macros instead of manually coding
everything for network sections.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
This parameter points to the driver instance private data, and not to
its configuration data, which one is set via cfg_info parameter.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Such state needs to be set _from_ the PM API functions and not the other
way round. So if a network device driver does not support such API, it
will not be able to set the core net_if on PM state, obviously.
Currently, these functions only set/unset NET_IF_SUSPENDED flag.
More logic will be added later, to decide whether the net_if can be
actually set to suspend mode or not and also to take care of all timers
related to the interface.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
This function can be used for example by network power management
to check if the network interface can be suspended or not.
If there are network packets in transmit queue, then the network
interface cannot be suspended yet.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
By changing the various *NET_DEVICE* macros. It is up to the device
drivers to either set a proper PM function or, if not supported or PM
disabled, to use device_pm_control_nop relevantly.
All existing macro calls are updated. Since no PM support was added so
far, device_pm_control_nop is used as the default everywhere.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
commit 971ae59913 ("net: pkt: Make sure iface is not null when
accessing L2") fixed net_if_l2 where iface was NULL, however if
iface->if_dev is NULL, the check breaks and returns an offset of
NULL (0x82 or so). This is incorrect.
Let's add a check for iface->if_dev as well.
Signed-off-by: Michael Scott <mike@foundries.io>
It is possible that the network interface is not set when we
check the interface in net_pkt.c:pkt_buffer_length(). For example
in icmpv6 unit test the interface is left as NULL as the test does
not care about what network interface is used. For real hw like
mimxrt1050_evk, which supports Ethernet, we need to add additional
checks for the interface being non-null.
Fixes#20088
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
A socket-offloaded interface should bypass interface initialization in
the same way as net-offloaded interface does.
Signed-off-by: Robert Lubos <robert.lubos@nordicsemi.no>
Allow user to disable native IP stack and use offloaded IP
stack instead. It is also possible to enable both at the same
time if needed.
Fixes#18105
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Removing an IPv4 router was missing, as well as finding the default
router for an IPv4 address.
Note howevere that IPv4 router features are not used anywhere yet. But
at least the API is there and is a 1:1 to IPv6, if that matters.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
- router lifetime is always a u16_t so fixing
net_if_ipv6_router_update_lifetime() signature.
- Coalescing router timers into one: this reduces the net_if_router
structure by 22 bytes
- refactor IPv6 and IPv4 router code so it's handled in generic
functions, to avoid duplicating 90% of the code for each family. This
also fixes the lifetime support for IPv4 which was missing.
Note however that IPv4 routing support seems to be missing as none of
the relevant functions are used anywhere yet.
Fixes#8728
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
This reduces the size of struct net_if_ipv6 by 24 bytes by moving
the k_delayed_work attribute into net_if core code.
Then each net_if_ipv6 can be added to the timer handler via a slist.
This does not make much gain if the system has only 1 network interface
It starts to be interesting if it has 2+ network interfaces then.
Fixes#8728
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
This reduces the size of struct net_if_addr by 24 bytes by moving
the k_delayed_work attribute into net_if core code.
Then each net_if_addr can be added to the timer handler via a slist.
This does not make much gain if the system has only 1 unicast IPv6
address. It's a nice memory improvment once it has 2+ unicast IPv6
address. Note that having IPv4 enabled along with IPv6 will also see
memory improvements since both IPv6 and IPv4 use the same struct
net_if_addr.
Fixes#8728
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
DAD use dad_count attribute in struct net_if_addr, since DAD is ran on
each and every ipv6 unicast address.
Fixes#8728
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
