[U-Boot] [PATCH 1/5] efi_loader: bootmgr: support BootNext and BootCurrent variable behavior

Heinrich Schuchardt xypron.glpk at gmx.de
Wed Dec 26 21:33:23 UTC 2018


On 12/26/18 10:23 PM, Alexander Graf wrote:
> 
> 
> On 25.12.18 10:36, AKASHI Takahiro wrote:
>> On Sun, Dec 23, 2018 at 03:03:51AM +0100, Alexander Graf wrote:
>>>
>>>
>>> On 18.12.18 06:02, AKASHI Takahiro wrote:
>>>> See UEFI v2.7, section 3.1.2 for details of the specification.
>>>>
>>>> With my efishell command[1], you can try as the following:
>>>>   => efi boot add 1 SHELL ...
>>>>   => efi boot add 2 HELLO ...
>>>>   => efi boot order 1 2
>>>>   => efi bootmgr
>>>>      (starting SHELL ...)
>>>>   => efi setvar BootNext =H0200
>>>>   => efi bootmgr
>>>>      (starting HELLO ...)
>>>>   => efi dumpvar
>>>>   <snip ...>
>>>>   BootCurrent: {boot,run}(blob)
>>>>   00000000:  02 00                    ..
>>>>   BootOrder: {boot,run}(blob)
>>>>   00000000:  01 00 02 00              ....
>>>>
>>>> Using "run -e" would be more human-friendly, though.
>>>>
>>>> [1] https://lists.denx.de/pipermail/u-boot/2018-November/346450.html
>>>>
>>>> Signed-off-by: AKASHI Takahiro <takahiro.akashi at linaro.org>
>>>> ---
>>>>  lib/efi_loader/efi_bootmgr.c | 37 +++++++++++++++++++++++++++++++++++-
>>>>  1 file changed, 36 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/lib/efi_loader/efi_bootmgr.c b/lib/efi_loader/efi_bootmgr.c
>>>> index a095df3f540b..a54ae28343ce 100644
>>>> --- a/lib/efi_loader/efi_bootmgr.c
>>>> +++ b/lib/efi_loader/efi_bootmgr.c
>>>> @@ -145,11 +145,21 @@ static void *try_load_entry(uint16_t n, struct efi_device_path **device_path,
>>>>  	efi_deserialize_load_option(&lo, load_option);
>>>>  
>>>>  	if (lo.attributes & LOAD_OPTION_ACTIVE) {
>>>> +		u32 attributes;
>>>>  		efi_status_t ret;
>>>>  
>>>>  		debug("%s: trying to load \"%ls\" from %pD\n",
>>>>  		      __func__, lo.label, lo.file_path);
>>>>  
>>>> +		attributes = EFI_VARIABLE_BOOTSERVICE_ACCESS |
>>>> +			     EFI_VARIABLE_RUNTIME_ACCESS;
>>>> +		size = sizeof(n);
>>>> +		ret = rs->set_variable(L"BootCurrent",
>>>> +				       (efi_guid_t *)&efi_global_variable_guid,
>>>> +				       attributes, size, &n);
>>>
>>> Every call into UEFI land (rs->foo(), bs->foo(), etc) has to go via the
>>> EFI_CALL() macro. Otherwise we may destroy the "gd" pointer on ARM.
>>
>> OK, but let me make sure one thing:
>> My efishell calls efi_* functions directly in some places as
>> Not all the features can be implemented only with boot/runtime services.
>> In those cases, we don't have to use EFI_CALL(), right?
> 
> If your "efishell" is a UEFI binary, you can directly call boot/runtime
> services. If it runs as part of the U-Boot code base, every call to
> boot/runtime service callbacks *must* go through EFI_CALL().

No, that is not how the call counting has been set up. You will get an
assert error if debugging is enabled. We use EFI_CALL when we want to
call an API function from inside an API function.

Just have a look at all the selftest code. It never uses EFI_CALL.

In this respect the bootmgr code is broken.

Best regards

Heinrich

> 
> 
> Alex
> 



More information about the U-Boot mailing list