USBDevice: Avoid forcing end device to be derived from USBDevice class

[facchinm]

Description

By replacing the existing callbacks with proper mbed::callback we can use the USBDevice infrastructure without deriving directly from USBDevice.

The patch could bring a (tiny, big ?) performance hit due to the indirection, but would make the USBDevice class usable as a singleton.

A usage example can be found here https://github.com/arduino/ArduinoCore-nRF528x-mbedos/tree/master/cores/arduino/USB and https://github.com/bcmi-labs/ArduinoCore-nRF528x-mbedos/tree/master/libraries/USBHID

PluggableUSBDevice implements USBDevice and calls phy init() just once; all device drivers (derived from mbed ones) plug on it to create a composite device but they don't derive from USBDevice.

Pull request type

[ ] Fix
[x] Refactor
[ ] Target update
[ ] Functionality change
[ ] Docs update
[ ] Test update
[ ] Breaking change

Reviewers

Release Notes

[ciarmcom]

@facchinm, thank you for your changes.
@ARMmbed/mbed-os-maintainers please review.

[0xc0170]

I can see there are lot of conflicts, please review

[facchinm]

Rebased upon master, thanks.

[0xc0170]

@maciejbocianski please review

[maciejbocianski]

Generally looks good,
tests-usb_device-basic needs adaptation to your changes, you can use my code maciejbocianski@092b26b

@fkjagodzinski @c1728p9 what do you think

[kjbracey]

If you're changing to Callback, could you please avoid giving it NULL - change to nullptr. Will avoid issues with some optimisation work I'm doing on Callback. Will save me touching this again later.

(That does make it a 5.14 change - we're not putting any C++11-isms into 5.13 patches. But this doesn't look like a patch anyway).

[fkjagodzinski]

Yeah, I think this patch OK but can't really comment on performance aspect.

The patch could bring a (tiny, big ?) performance hit due to the indirection, but would make the USBDevice class usable as a singleton.

@c1728p9 could you comment on that, please?

[c1728p9]

This PR seems reasonable to me. Does the use of callback increase or decrease RAM used? Also, you'll need a few more changes if you want to make USBDevice usable without inheriting from it:

• Protected functions need to be made public so they can be called
  • Callback completion handlers such as complete_request
  • Synchronization functions

Those changes could be made in a separate PR though as this change look good on its own merits.

[0xc0170]

CI started

[0xc0170]

@facchinm Is this breaking change or functionality change? If yes, Release Notes should be added to your first comment above.

[facchinm]

@0xc0170 the only breaking change is the removal of template<typename T> bool endpoint_add(usb_ep_t endpoint, uint32_t max_packet, usb_ep_type_t type, void (T::*callback)() overload that could affect 3rd party, out of tree USB classes (I didn't find any in the wild but they could exist of course).
I'm amending the first commit and add this information

[mbed-ci]

Test run: FAILED

Summary: 3 of 4 test jobs failed
Build number : 1
Build artifacts

Failed test jobs:

• jenkins-ci/mbed-os-ci_build-IAR
• jenkins-ci/mbed-os-ci_build-GCC_ARM
• jenkins-ci/mbed-os-ci_build-ARM

[0xc0170]

Please review build logs, there are failures

[0xc0170]

CI restarted

[facchinm]

By fixing the tests I came up with another implementation, much cleaner and, more important, without breaking changes.
Should I close this PR and open another or forcepushing is ok?

[0xc0170]

You can force push

[0xc0170]

Once you do force push , please update PR with the current status to notify us - what has changed.

[facchinm]

Sorry, I updated the description on the top but probably this didn't trigger any notification.
Anyway, the new implementation leave everything in its place and adds the possibility to use a regular Callback as argument.

[0xc0170]

Ci restarted

@maciejbocianski Please review

[mbed-ci]

Test run: SUCCESS

Summary: 11 of 11 test jobs passed
Build number : 3
Build artifacts

[maciejbocianski]

LGTM

[fkjagodzinski]

Does the use of callback increase or decrease RAM used?

FYI, this patch increased the size of the USBDevice class by 240 B. It is now 824 B.

[fkjagodzinski]

@donatieng, @ARMmbed/mbed-os-hal, this information ^^^ might be useful for future reference. I think there's room for improvement if needed.