| PORTING LIBUSB TO OTHER PLATFORMS |
| |
| Introduction |
| ============ |
| |
| This document is aimed at developers wishing to port libusb to unsupported |
| platforms. I believe the libusb API is OS-independent, so by supporting |
| multiple operating systems we pave the way for cross-platform USB device |
| drivers. |
| |
| Implementation-wise, the basic idea is that you provide an interface to |
| libusb's internal "backend" API, which performs the appropriate operations on |
| your target platform. |
| |
| In terms of USB I/O, your backend provides functionality to submit |
| asynchronous transfers (synchronous transfers are implemented in the higher |
| layers, based on the async interface). Your backend must also provide |
| functionality to cancel those transfers. |
| |
| Your backend must also provide an event handling function to "reap" ongoing |
| transfers and process their results. |
| |
| The backend must also provide standard functions for other USB operations, |
| e.g. setting configuration, obtaining descriptors, etc. |
| |
| |
| File descriptors for I/O polling |
| ================================ |
| |
| For libusb to work, your event handling function obviously needs to be called |
| at various points in time. Your backend must provide a set of file descriptors |
| which libusb and its users can pass to poll() or select() to determine when |
| it is time to call the event handling function. |
| |
| On Linux, this is easy: the usbfs kernel interface exposes a file descriptor |
| which can be passed to poll(). If something similar is not true for your |
| platform, you can emulate this using an internal library thread to reap I/O as |
| necessary, and a pipe() with the main library to raise events. The file |
| descriptor of the pipe can then be provided to libusb as an event source. |
| |
| |
| Interface semantics and documentation |
| ===================================== |
| |
| Documentation of the backend interface can be found in libusbi.h inside the |
| usbi_os_backend structure definition. |
| |
| Your implementations of these functions will need to call various internal |
| libusb functions, prefixed with "usbi_". Documentation for these functions |
| can be found in the .c files where they are implemented. |
| |
| You probably want to skim over *all* the documentation before starting your |
| implementation. For example, you probably need to allocate and store private |
| OS-specific data for device handles, but the documentation for the mechanism |
| for doing so is probably not the first thing you will see. |
| |
| The Linux backend acts as a good example - view it as a reference |
| implementation which you should try to match the behaviour of. |
| |
| |
| Getting started |
| =============== |
| |
| 1. Modify configure.ac to detect your platform appropriately (see the OS_LINUX |
| stuff for an example). |
| |
| 2. Implement your backend in the libusb/os/ directory, modifying |
| libusb/os/Makefile.am appropriately. |
| |
| 3. Add preprocessor logic to the top of libusb/core.c to statically assign the |
| right usbi_backend for your platform. |
| |
| 4. Produce and test your implementation. |
| |
| 5. Send your implementation to libusb-devel mailing list. |
| |
| |
| Implementation difficulties? Questions? |
| ======================================= |
| |
| If you encounter difficulties porting libusb to your platform, please raise |
| these issues on the libusb-devel mailing list. Where possible and sensible, I |
| am interested in solving problems preventing libusb from operating on other |
| platforms. |
| |
| The libusb-devel mailing list is also a good place to ask questions and |
| make suggestions about the internal API. Hopefully we can produce some |
| better documentation based on your questions and other input. |
| |
| You are encouraged to get involved in the process; if the library needs |
| some infrastructure additions/modifications to better support your platform, |
| you are encouraged to make such changes (in cleanly distinct patch |
| submissions). Even if you do not make such changes yourself, please do raise |
| the issues on the mailing list at the very minimum. |
| |