Constifying global data allows to save lots of RAM. This improvement
allows to save 224 bytes of RAM when compiled on ARM, with default
configuration.
Signed-off-by: Marcin Niestroj <m.niestroj@grinn-global.com>
Constifying global data allows to save lots of RAM. Defining data inside
function as 'static const' allows on the other hand to save stack usage
and reduced code size (because data doesn't have to be copied on stack
in runtime).
This improvement allows to save 448 bytes of RAM and 88 bytes of ROM
when compiled on ARM, with default configuration.
Signed-off-by: Marcin Niestroj <m.niestroj@grinn-global.com>
modem_cmd and setup_cmd structures are used only to store static
information about commands to be sent and replies to be
received. Reference it as const, so that it is possible to define const
instances and save some RAM space by placing those definitions entirely
in ROM.
Signed-off-by: Marcin Niestroj <m.niestroj@grinn-global.com>
The commit 5e97d779bb ("device: convert DEVICE_INIT to DEVICE_DEFINE
or SYS_DEVICE_DEFINE") did not convert the gsm_ppp.c properly. The
power management parameter was not set to NULL but was pointing to
the data pointer.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
Convert handful of users of DEVICE_INIT to DEVICE_DEFINE or
SYS_DEVICE_DEFINE to allow deprecation of DEVICE_INIT.
Signed-off-by: Kumar Gala <kumar.gala@linaro.org>
Replaces all existing variants of value clamping with the MIN and MAX
macros with the CLAMP macro.
Signed-off-by: Trond Einar Snekvik <Trond.Einar.Snekvik@nordicsemi.no>
All net_bufs allocated to modem_cmd_handler's data->rx_buf are consumed
synchronously in ublox-sara-r4.c in the same thread. This means that
allocating them with timeout makes no sense, because timeout
will *always* be hit when there are no more buffers in net_buf_pool.
Get rid of the unnecessary timeout, as it doesn't help and just slows
down processing of incoming data, increasing possibility of data
overrun.
Signed-off-by: Marcin Niestroj <m.niestroj@grinn-global.com>
All net_bufs allocated to modem_cmd_handler's data->rx_buf are consumed
synchronously in gsm_ppp.c in the same thread. This means that
allocating them with timeout makes no sense, because timeout
will *always* be hit when there are no more buffers in net_buf_pool.
Get rid of the unnecessary timeout, as it doesn't help and just slows
down processing of incoming data, increasing possibility of data
overrun.
Signed-off-by: Marcin Niestroj <m.niestroj@grinn-global.com>
So far a dedicated buffer was used for data read from modem
interface. New net_bufs were allocated and filled later, which means
that data was lost when no more net_bufs were available in the pool.
Prevent data loss by allocating net_buf before attempting any read on
modem interface. Process incoming data in a loop as long as reading from
interface results in new data. Also remove dedicated buffer
(data->read_buf) and directly fill net_buf content instead. As a side
effect there are less memory copy operations and RAM usage is reduced.
Pre-allocated net_buf is now always appended to data->rx_buf. When there
was no (more) data read from interface to such net_buf, then this empty
net_buf will be on the end of data->rx_buf fragment list. Update
skipcrlf() and findcrlf() implementations to explicitly check for each
net_buf length, instead of blindly assuming them to have at least single
byte.
Signed-off-by: Marcin Niestroj <m.niestroj@grinn-global.com>
cmd_handler_process() does two major things:
- reads data from modem interface and fills data->rx_buf,
- processes data in data->rx_buf.
Split implementation accordingly to two separate functions, which
improves readability (less automatic variables to follow at once) and
simplifies refactoring of each action.
No functional change was intended in this commit.
Signed-off-by: Marcin Niestroj <m.niestroj@grinn-global.com>
If CONFIG_MODEM_SIM_NUMBERS is 'n' than the data_imei field of struct
modem_context doesn't exist, so add ifdef protection for that case.
Signed-off-by: Kumar Gala <kumar.gala@linaro.org>
If other threads are accessing the modem concurrently during gsm_ppp
initialization, they can easily corrupt UART comms. Normally this is not
noticable since those services are usually started only after modem
setup has been completed. But with the new gsm_ppp stop/start
functionality, the need of synchronizing concurrent access in a
structured way has arisen.
This commit ensures modem_cmd_handler is locked when concurrent access
to the modem UART would interfere with the driver, or otherwise cause
problems. Since the semaphore is not available during this period, all
essential calls to modem_cmd_send has been replaced with the non-locking
equivalents.
Signed-off-by: Benjamin Lindqvist <benjamin.lindqvist@endian.se>
A semaphore is used by modem_cmd_handler to protect against concurrent
UART access. This is only used for the duration of sending a command,
but there are cases when one wants to prevent UART access for other
reasons, such as when powering off the modem.
This commit exposes functionality for hogging this semaphore without
having to send a command. Furthermore, a non-locking equivalent for
modem_cmd_handler_setup_cmds is added which was previously missing.
Signed-off-by: Benjamin Lindqvist <benjamin.lindqvist@endian.se>
Application might need to know the GSM modem device name so
provide it in the header file.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
When PPP is muxed, using uart_poll_out resulted in each byte getting
wrapped in a muxing header. This led to UART bombardment which
can quickly cause some modems to hang and panic. This was observed
regularly using a SIMCOM7600E modem.
A perfect fix would involve rewriting ppp.c, uart_mux.c and
modem_iface_uart.c to all use another UART API, but that would be more
invasive by several orders of magnitude than this one, which utilizes
the fact that the uart_mux implementation of uart_fifo_fill does NOT
require ISR context. Since the Zephyr UART API states that the behavior
of uart_fifo_fill outside of ISR context is implementation defined, this
should be kosher.
Signed-off-by: Benjamin Lindqvist <benjamin.lindqvist@endian.se>
These changes enable applications to restart the networking stack which
was previously not possible without rebooting the device. This was a
major show-stopper because it made power management impossible, and
furthermore made it impossible to recover from a bad modem state without
rebooting.
This has been verified to work on a SIMCOM7600E modem, both with and
without CONFIG_GSM_MUX enabled.
Signed-off-by: Benjamin Lindqvist <benjamin.lindqvist@endian.se>
The char pointer that is logged could get scoped out and so should be
strduped before logging.
Signed-off-by: Benjamin Lindqvist <benjamin.lindqvist@endian.se>
Attempting to initialize PPP without first ensuring we're attached to
the network packet service will sometimes lead to "NO CARRIER" errors
when we initialize PPP with the modem. This has been observed
reproducibly on some SIMCOM7600E modems.
This commit holds off on PPP initialization until modem has reported
that it is indeed attached by using the "AT+CGATT?" command (see 3GPP TS
27.007)
Signed-off-by: Benjamin Lindqvist <benjamin.lindqvist@endian.se>
There are cases where the socket would not be closed.
offload_put function will now always close the socket
properly.
Remove error log messages as they are redundant.
on_cmd_sock_error_code handles printing errors.
Signed-off-by: Ryan Erickson <ryan.erickson@lairdconnect.com>
If HW flow control is enabled, then modem_context framework won't drain
UART FIFO blindly, but will stop when there is no more space in RX
ring_buf. This prevents data loss by "pausing" incoming data on hardware
level.
Signed-off-by: Marcin Niestroj <m.niestroj@grinn-global.com>
So far all received bytes over UART where blindly drained and pushed to
ring_buf. This approach is okay for UART devices without configured HW
flow control, as it basically decouples data processing from ISR handler
and gives more time before data overrun. However when HW flow control
is enabled, such behavior somehow suppresses UART flow control
advantage, because data can overrun when pushing to ring_buf.
Allow drivers utilizing modem_context framework to pass information
about whether HW flow control is enabled or not. If it is enabled, then
read data from UART FIFO to the point when RX ring_buf is totally filled
and follow such situation by disabling RX interrupt. Incoming data will
be paused on HW level, so there is lots of time for RX thread to process
ring_buf content. Reenable RX interrupts after all data in ring_buf is
processed, so that number of context switches is kept at minimum level.
Signed-off-by: Marcin Niestroj <m.niestroj@grinn-global.com>
This API allows to drop use of preallocated isr_buf. Most importantly as
a result RAM usage is reduced for each driver utilizing modem_context
framework. Additionally there is less copying done in ISR context, as
data is direcly read from UART FIFO to ring_buf.
Signed-off-by: Marcin Niestroj <m.niestroj@grinn-global.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 driver utilizes the CONFIG_NET_OFFLOAD setting to avoid the
normal handling of IP packets, and instead uses a socket-like
UART interface to handle incoming and outgoing data via AT commands.
Signed-off-by: Ryan Erickson <ryan.erickson@lairdconnect.com>
PR #27485 introduced new implementations for close() that do not build
correctly. This commit fixes the problem by re-implementing them.
Fixes#27587
Signed-off-by: Vincent Wan <vwan@ti.com>
This adds an option to query the modem for the SIM's IMSI and ICCID
numbers, just like the modem's IMEI is queried today. This requires
the SIM to be present, which might not be the case for all
applications, so it can be disabled.
Signed-off-by: Göran Weinholt <goran.weinholt@endian.se>
Signed-off-by: Benjamin Lindqvist <benjamin.lindqvist@endian.se>
Do not route close() calls via ioctl() as that is error prone
and quite pointless. Instead create a callback for close() in
fdtable and use it directly.
Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
It was already using uart_irq_callback_user_data_set below, now it also
uses uart_irq_callback_user_data_t as callback type, so let's normalize
the callbacks.
Fixes#26923
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Unit tests were failing to build because random header was included by
kernel_includes.h. The problem is that rand32.h includes a generated
file that is either not generated or not included when building unit
tests. Also, it is better to limit the scope of this file to where it is
used.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
By using the Zephyr-native zsock_ family of types and functions, these
drivers will be decoupled from NET_SOCKETS_POSIX_NAMES.
Signed-off-by: Adam Porter <porter.adam@gmail.com>
Used for permission validation when accessing the associated file
descriptors from user mode.
There often get defined in implementation code, expand the search
to look in drivers/ and subsys/net/.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
The implementation of offload_getaddrinfo in this driver failed
when the node it was called with was an IP address. This condition
was never detected, and as a consequence a DNS query was done on
the IP address instead of returning it directly.
Also, the port was set first after running the DNS query.
As a consequence, if the IP address would have been returned directly,
this would have been done without a port been set.
Both errors are fixed in this patch.
Signed-off-by: Hans Wilmers <hans@wilmers.no>
