On Sun, Sep 17, 2017 at 1:58 PM, Simon Glass sjg@chromium.org wrote:
Hi Heinrich,
On 15 September 2017 at 02:06, Heinrich Schuchardt xypron.glpk@gmx.de wrote:
This patch series provides:
- corrections for the EFI event services
- a test framework to check the EFI API implementation
- unit tests covering the event services
The EFI selftest is written such that it could be easily turned into a standalone EFI application. But this would require modifying the build procedures for EFI. Objcopy cannot generate the necessary relocations.
The unit tests are identified by entries in a linker generated array to make them as self sufficient as possible.
A Python test case is supplied to call run the EFI tests.
Tested with Travis CI https://travis-ci.org/xypron2/u-boot/jobs/275733784
Of all my efi_loader patches these are the first I would like to see merged.
Simon has commented on some other patches that he misses comments for all EFI API functions. I will add these with a separate patch.
Heinrich Schuchardt (10): efi_loader: allow return value in EFI_CALL efi_selftest: provide an EFI selftest application test/py: add a test calling the EFI selftest efi_loader: implement queueing of the notification function efi_loader: efi_set_timer: reset signaled state efi_selftest: provide unit test for event services efi_loader: implement task priority level (TPL) efi_selftest: test task priority levels efi_loader: notify when ExitBootServices is invoked efi_selftest: check notification of ExitBootServices
This progress makes significant progress on EFI testing at last. I'm very pleased to see it. Thank you for all the work you have put into this.
In addition to this (not instead of) I would like to see EFI code running under sandbox. I don't at present see a good reason why this cannot be done. I am going to try to enable EFI loader support in sandbox to a basic level and then we can see how hard it is to get some of your tests running directly in sandbox. If that works out then we can add that into the mix.
fwiw, I started on trying to get EFI_LOADER working in sandbox earlier today.. but ran into issues w/ setjmp. I probably should have kept my WIP but it was nothing too hard to reproduce (kconfig adds an "|| SANDBOX" to depends, and one or two "#elif defined(CONFIG_SANDBOX)".. nothing that would take too long to get back to the point I was at stuck on setjmp/longjmp (and lack of standard system hdrs).. I just switched to qemu
Other than booting a real OS, seems theoretically possible to get EFI_LOADER working in sandbox. It should be enough to (w/ suitable 'host bind') load/run Shell.efi and eventually SCT.efi.
BR, -R
I think this would make for an easier development environment for new EFI features. For some years I have developed all new features in sandbox and find it painful and unproductive when I need to test every change manually on a board. It should also allow us to run your tests (perhaps with some adaptation) with 'make tests' on a local machine using sandbox. Ultimately it should be possible to expand test coverage to cover all significant EFI logic.
[..]
Regards, Simon