[U-Boot] [PATCH 0/2] efi_loader: Implement reset on RPi

[Andreas Färber]

Am 14.07.2016 um 08:18 schrieb Alexander Graf:
>> Am 14.07.2016 um 06:48 schrieb Andreas Färber <afaerber at suse.de>:
>>
>> Hi Alex,
>>
>>> Am 05.06.2016 um 23:17 schrieb Alexander Graf:
>>> If Linux finds an EFI implementation it always uses the EFI reset handler to
>>> reboot or power down the system.
>>
>> Hm, I thought my powerdown issues on the Jetson TK1 were for lack of
>> CONFIG_AS3277_RESET - sounds like it could be due to EFI instead?
> 
> It depends. IIRC the TK1 is 32bit, where you're probably using grub2 which is not efi Linux aware, but instead boots over the zImage protocol. In that case Linux doesn't know about efi runtime services.

We've confirmed in the meantime that the Jetson TK1 issues were
unrelated to EFI and could be worked around by enabling some as3722
kernel option.

>>> Unfortunately we haven't implemented that one yet. In fact, while we prepared
>>> for RTS handling, we never actually implemented a single user.
>>>
>>> This is going to change today. This simple patch set enables RTS reset support
>>> for the RPi systems, allowing you to reboot and shut down the rpi if booted
>>> via EFI.
>>>
>>> It also lays the groundwork to show how to implement that functionality at all,
>>> so I expect more boards to follow.
>>>
>>> Alexander Graf (2):
>>>  efi_loader: Allow boards to implement get_time and reset_system
>>>  ARM: bcm283x: Implement EFI RTS reset_system
>>>
>>> arch/arm/mach-bcm283x/include/mach/wdog.h |   2 +-
>>> arch/arm/mach-bcm283x/reset.c             |  59 +++++++++++++++--
>>> cmd/bootefi.c                             |   4 ++
>>> include/efi_loader.h                      |  18 ++++++
>>> lib/efi_loader/efi_runtime.c              | 101 ++++++++++++++++++++++++++----
>>> 5 files changed, 166 insertions(+), 18 deletions(-)
>>
>> This all looks very non-generic and would mean that every board will
>> need to be patched individually, which is unrealistic to be tested by
>> just the two of us.
>>
>> Can't you patch the reset_cpu() declaration (common.h/sysreset.h)
>> instead of all its implementations? We might still need to patch
>> individual implementations but I don't see why any reset_cpu()
>> implementation should be in a different section than others.
> 
> Hmm. There are 2 minor problems:
> 
>   1) Efi also supports power off on top of reset
>   2) We would have to convert all boards at once, rather than one by one, as we couldn't distinguish between efi aware and unaware ones

I don't see why we would need to convert everything at once either way.

> 
> And one major issue:
> 
> All device memory pointers used by the reset functions need to be marked as such in the efi memory map and live relocated when entering runtime mode. So we need to manually touch every function either way.
> 
> That mesns we could either make a generic, broken version. Or we just convert one by one for systems that we can verify it on :). I hope that I designed the APIs easily enough that people who are not us enable RTS support on other platforms too :)


Ping! Anyone any comments on the two open questions of uppercase vs.
lowercase and placement of attribute?

1/2 should not be affected by those discussion points and already
featured in a second series (LS2080ARDB).

Regards,
Andreas

-- 
SUSE Linux GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Felix Imendörffer, Jane Smithard, Graham Norton
HRB 21284 (AG Nürnberg)