A perpetual WIP PR #22
Draft
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
d3zd3z
added a commit
to tangybbq/keyboard-firmware
that referenced
this pull request
Oct 18, 2024
Feedback on zephyr-lang-rust has resulted in numerous API changes. The good news is that I only have a small number of commits on top of zephyr-lang-rust to support this keyboard firmware, and all currently live in: zephyrproject-rtos/zephyr-lang-rust#22
d3zd3z
force-pushed
the
dt
branch
2 times, most recently
from
601ced8 to
d4754e1
Compare
d3zd3z
force-pushed
the
hack
branch
2 times, most recently
from
69930a4 to
4985edf
Compare
d3zd3z
force-pushed
the
dt
branch
2 times, most recently
from
e640970 to
7e8a847
Compare
This code parses the DTS file generated by the Zephyr build, along with a few entries from the generated header file, to build a representation of the device tree. There is a notion of "augments" that add various methods. This is currently just hard-coded. Signed-off-by: David Brown <[email protected]>
Blinky is a rust port of the samples/blinky application from the main zephyr repo. It performs the same function, but using the DT and GPIO abstractions provided in the zephyr::sys module. Signed-off-by: David Brown <[email protected]>
Create two modules with wrappers for Zephyr gpio and flash devices. These have no methods, but allow the DT generated code to compile, with `get_instance` methods that return these values. Signed-off-by: David Brown <[email protected]>
The is_ready method on both `Gpio` and `GpioPin` will ultimately call the underlying `device_is_ready` entry. Signed-off-by: David Brown <[email protected]>
Create wrappers for config and pin toggle on the gpios. This is enough to allow the blink app to work. Signed-off-by: David Brown <[email protected]>
Move the module declaration for the device tree up into `lib.rs` to allow insertion of allow directives to eliminate documentation warnings on the generated devicetree. Signed-off-by: David Brown <[email protected]>
Instead of hardcoding all of the augments in the code, put these augments into a data structure. Load the actual rules from a yaml file (with a future ability to extend this with per-app or per-module defined augments. Convert all of the existing hardcoded augment rules into the initial augment file.
Add constructors to the individual device types. These are unsafe, and are all referenced from the generated devicetree code.
Move all of the device implementations into a `zephyr::device` module. In this module, add a `Unique` type that supports the constructors on the device requiring a unique instance. The device tree augmentation code adds a declaration for these uniqueness markers for each instance, passing it into the constructor. Signed-off-by: David Brown <[email protected]>
Gpios in Zephyr are inherently unsafe. There is a shared controller that is not just used by the pins defined here, but extensively across various drivers. As such, make all of the gpio pin operations themselves unsafe. We can help, a little bit, to at least enforce uniqueness with the Rust drivers that use gpios by requiring them to take a mutable instance of `GpioToken`, which has a singleton getter. Signed-off-by: David Brown <[email protected]>
Prevents a warning on boards where there are no gpios or flash controllers are defined. Signed-off-by: David Brown <[email protected]>
This allows a build on the nrf51, preventing an error when the partitions are detected, but the controller wasn't. Signed-off-by: David Brown <[email protected]>
This fixes a weird issue with bindgen missing the `__device_dts_ord_nn` declarations in some circumstances. It is unclear when this was occuring, and hopefully it doesn't return at some point. Signed-off-by: David Brown <[email protected]>
Instead of trying to hand off data through a file between the build of different crates, which was causing build failures in applications that need to use cfgs based on the presence of DT nodes, instead, just parse the DT again, and recalculate the node tree from it. This should fix build issues with the all nodes txt file missing. Signed-off-by: David Brown <[email protected]>
Several things have become unsafe, so use those in unsafe blocks. The GPIO driver now has a token that must be passed to each action, to enforce single threadded use. Signed-off-by: David Brown <[email protected]>
Add a simplistic uart wrapper, and drivers around it. All methods are unsafe. Signed-off-by: David Brown <[email protected]>
Add a simple wrapper for the led strip driver. Signed-off-by: David Brown <[email protected]>
Add methods for get and set of pin logical values. Signed-off-by: David Brown <[email protected]>
This isn't really quite right, as the "unique" values here are per pin, and if the pin were to be used across the system, nothing would enforce these to be unique. Signed-off-by: David Brown <[email protected]>
This is a special keyboard matrix type used by the bbq-keyboard firmware. It doesn't belong here, and can be moved out, once we support additional augment files. Signed-off-by: David Brown <[email protected]>
Add support for Leds, and a simple augment for the pwm drivers.
Fix the DT node names in the generated `dt_cfgs` entries to match the names of the nodes. Need to apply `fix_id` in a few places to make sure non-identifier characters are converted to underscores. Signed-off-by: David Brown <[email protected]>
Enable use of the thread analyzer, if it is enabled. Signed-off-by: David Brown <[email protected]>
Add these bindings so that app code can directly use these. Signed-off-by: David Brown <[email protected]>
Add an `.into_irq()` to the Uart device that turns it into an interrupt driven interface. Currently, read is implemented, with a `.try_read()` method that will try, with a timeout, to read data from the interface. The methods are all marked as 'unsafe' currently, until a more thorough safety analysis can be made. Signed-off-by: David Brown <[email protected]>
A blocking write allows us to do the write without needing an intermediate copy of the data. The thread's buffer is directly copied into the uart's fifo. To make this safe, the write call must block until all of the data has been placed in the fifo. The uart.h irq interface is a little quirky. Namely, the uart fifo can only be accessed from the irq handler. To make this possible, the drivers will implement usually a soft interrupt to the irq handler, for the initial transmission. There is also an async interface in uart.h, but numerous devices don't support it. Signed-off-by: David Brown <[email protected]>
Signed-off-by: David Brown <[email protected]>
Add a timeout parameter to the 'write' method on the UartIrq interface. The method now returns the number of bytes that were actually written. Signed-off-by: David Brown <[email protected]>
This method had a lifetime error, and allowed the inner uart to be leaked beyond its lifetime. This should prevent the returned device from outliving the UartIrq, or even being used after mutable methods on the UartIrq have been used. For now, don't even return a mutable method until we evaluate what should be permissible. This at least allows reading the line status. In addition, since the return Uart is not owned by the caller, it is not permissible to call any of the `into_*` methods. Signed-off-by: David Brown <[email protected]>
With the 'Uart' encapsulated within the `UartIrq`, having a separate field for device is redundant with the one inside of `Uart`. Remove this redundant field, and change the references to it to use the one inside the Uart. Signed-off-by: David Brown <[email protected]>
Move this specific type into its own crate. Signed-off-by: David Brown <[email protected]>
Add ring size parameters to the UartIrq type. These will later be used to migrate the interface to an async interface (not rust asyc). Signed-off-by: David Brown <[email protected]>
Update the writing to be an async interface. Instead of a blocking write, there is a `write_enqueue` to add the message, and a `write_wait` which is a blocking wait for a write to handle. For this to work, the UartIrq needs to own the buffer for the duration of the write. At this point, the buffers are just `Vec<u8>`, but a future change will make this more general, so that, for example, Pooled buffers can be used. The buffer should never be dropped within this code. Signed-off-by: David Brown <[email protected]>
Because this interface has registered IRQs, prevent drops. They are probably safe due to the use of the leaked Arc, but the data is in fact leaked. For now, just cause Drop to panic, as normal use won't ever drop this. Signed-off-by: David Brown <[email protected]>
Change the read interface to also be async. The user must enqueue one or more buffers for the driver to place data into. Signed-off-by: David Brown <[email protected]>
The result from the async_write holds the buffer, but the fields are private. Provide two accessors to make this result useful: - as_slice: returns a slice of the useful part of the buffer - into_inner: Transforms into the buffer, so it can be reused. Signed-off-by: David Brown <[email protected]>
Currently, the IRQ handling code requires a SpinMutex which is only available with Arc, which requires allocation. Disable this use of the driver in this situation. Signed-off-by: David Brown <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR holds onto changes that would be considered "work in progress", but can be reviewed. Ideally, this should be based on top of other pending PRs so that the actual change here is kept minimal.
Built upon #15