The PinePhone has circuitry that makes this pin active-high, but the
PinePhone Pro doesn't appear to, which means we would need read active
high / active low from the config file. Since the modem works fine
without setting this pin, remove the logic to handle the W_DISBALE pin.
If the reset sequence fails, we don't want to transition to the powered
state, instead we should stay in the resetting state so we eventually
timeout and try a gpio reset. If the reset command succeeds the modem
will eventually send RDY and transition the state to STARTED.
Since this might take longer to change the modem state, add 15 seconds
(for a total of 45) to the at reset timeout just in case. It looks like
the modem typically takes about 20 between sending "AT+CFUN=1,1" and
receiving "RDY", so this should be plenty.
If the modem doesn't respond to the at reboot sequence, this will
attempt reboot the modem using both the power pin and then the reset pin
in case that doesn't work.
This splits the logic from gpio_check_poweroff into two separate
functions, allowing the caller more control over when the reset line is
asserted to force the modem to turn off.
Previously if the modem didn't respond to the power key sequence in
quit_app, gpio_check_poweroff wouldn't set the reset pin, and
eg25-manager could leave the modem running after exiting.
It seems `off_t` is now 64-bits wide on 32-bit architectures, so using
`glong` causes a build failure in this case. Switch to using the correct
type from the `sendfile()` prototype to fix this.
BH PinePhone don't have the modem's `STATUS` pin connected to the SoC,
and as such require using `libusb` for checking the modem power state.
We didn't handle this case previously due to lack of on-device testing,
causing BH phones to fail with newer versions.
This patch ignores the `status` GPIO for devices relying on `libusb`,
which are only pre-CE PinePhones.
`modem_reset()` can be called from multiple places, both tied to a
specific timer. In order to prevent calling this function multiple times
in a row, disable all related timers as soon as we enter this function.
When going into firmware upgrade mode, the modem comes back with
different IDs than in normal use. We should make sure we cancel the
reset sequence when that happens, and not start a new one.
Cancel reset if modem is back before reset sequence is started.
When upgrading the modem through fwupd, it will enter fastboot.
Therefore, it disappears from the USB bus for a few ms.
However, the eg25-manager considers this as an issue and
resets the modem a bit later during the upgrade process.
To avoid this, cancel the reset sequence if the modem is already
back before the reset sequence is started. This will take less than 3
seconds.
The problem addressed by monitoring the modem status through udev seems
tied to the specific USB controller used by the A64. On RK3399 devices,
this quirk is apparently unneeded, and actually harmful as it resets the
modem USB connection while ModemManager is already configuring it.
This commit adds an optional config parameter for disabling this module
(enabled by default).
The timeout was meant to give more time to find the modem, yet the
callback didn't actually check if it came back but blindly issued a
reset.
Add a check to the callback to see if the modem was found and only
reset if not.
Fixes https://gitlab.com/mobian1/devices/eg25-manager/-/issues/28
The PinePhone Pro uses the same EG25 modem as the OG PinePhone, but with
a different SoC. It also uses an ALC5616 audio codec directly hooked up
to the modem (which is I2S master in this case).
The config is therefore almost identical to the PinePhone rev1.2 except
for the gpios, UART port used and `AT+QDAI` config (to account for the
different audio setup).
Instead of assuming we're running on the PinePhone (and therefore
hardcoding gpiochip labels and line numbers), make all of those
configurable. Legacy config files will still be parsed as long as they
lack the `chips` key.
Existing config files are also updated to match the new config format.
Previously this code checked if the buffer was full after writing to it,
which meant that the buffer could overflow.
This checks for an overflow before copying into the buffer and only
copies the data that will fit.
Executing a reset each time the modem is unbound is a bit too extreme:
the modem sometimes recovers by itself and only needs a "soft" reset
sequence (unbind/bind).
This commit introduces a short timer (2s) so we the modem can settle in.
If reset fails after this time, the modem is probably completely broken,
or already rebooting, so we can safely issue a reset AT command.
`modem_reset()` could previously either fail silently, or fall back to
using AT commands without indicating what happened. This commit adds
informative messages and makes sure we fall back to resetting using AT
commands whenever we encounter an error.
By instructing sendfile() to send the whole content of the assistance
data at once results in a big data transfer (> 4kb) first followed by a
huge amount of small transfers (64 or 128b). This "overloads" the modem
which needs to process smaller chunks of data and have more time to do
so (experimentation shows that 1kb need 100ms for processing, or the
upload will subsequently fail).
This commit sets the transfer size to 256b and increase the timeout
between each transfer to 25ms.
If we clear QFUPL right after we finished the file transfer, we will
execute the following commands right away, leading to interleaved
replies from the modem (i.e. the reply to QGPSXTRATIME being received
with the notification of upload completion).
Keeping the command in the queue allows us to use the callback a second
time once the upload is complete and resume assistance data processing
accordingly.
After the modem send "RDY", other messages are received. With the
current implementation, this causes eg25-manager to mark the modem as
configured before it is even picked up by ModemManager. Adding an
additional status check helps preventing this issue.
If a user created a custom config file and we added new fields in a
newer version, `eg25-manager` will crash by assuming all config files
are complete and up-to-date.
Moreover, creating a custom config, even to change only one option,
required copying a complete default config file and then editing the
relevant field(s). This could lead to issues when upgrading, as default
values of fields unrelated to the user change could be modified and not
applied to the user config.
This patch addresses both these issues by:
* making sure at least one config file exists
* requiring only the "default" config file (the one under `/usr/share`)
to include all the required fields
* trying to use user config for each field, falling back to the default
config file if the field isn't present in the user config
* exiting with a meaningful error message in case the default config
file is missing a required field or section
That way, it will be possible to have a minimal user config file
containing only the field(s) needing a different value than the default
one, falling back to the values in the default config file.
Fixes#23
In some cases, there might be an issue with `eg25-manager` trying to
reset the modem while it's being configured. This leads to infinite boot
loops where the modem is constantly reset soon after it booted.
Fixes#24
