[U-Boot] [PATCH v4 1/1] efi_loader: write protocol GUID in OpenProtocol

[Rob Clark]

On Wed, Aug 23, 2017 at 5:44 PM, Heinrich Schuchardt <xypron.glpk at gmx.de> wrote:
> On 08/18/2017 05:45 PM, Heinrich Schuchardt wrote:
>> To understand what is happening in OpenProtocol or LocateProtocol
>> it is necessary to know the protocol interface GUID.
>> Let's write a debug message.
>>
>> Cc: Rob Clark <robdclark at gmail.com>
>> Signed-off-by: Heinrich Schuchardt <xypron.glpk at gmx.de>
>> ---
>> v4:
>>       use %pUl format string
>>       correct indention of output
>> v3:
>>       commit message changed
>> v2:
>>       fix typo in commit message
>> ---
>>  include/efi_loader.h          |  8 ++++++++
>>  lib/efi_loader/efi_boottime.c | 11 +++++++++++
>>  2 files changed, 19 insertions(+)
>>
>> diff --git a/include/efi_loader.h b/include/efi_loader.h
>> index 870854fb61..5b20af808b 100644
>> --- a/include/efi_loader.h
>> +++ b/include/efi_loader.h
>> @@ -17,6 +17,7 @@
>>
>>  int __efi_entry_check(void);
>>  int __efi_exit_check(void);
>> +const char *__efi_nesting(void);
>>  const char *__efi_nesting_inc(void);
>>  const char *__efi_nesting_dec(void);
>>
>> @@ -64,6 +65,13 @@ const char *__efi_nesting_dec(void);
>>       debug("%sEFI: Return From: %s\n", __efi_nesting_dec(), #exp); \
>>       } while(0)
>>
>> +/*
>> + * Write GUID
>> + */
>> +#define EFI_PRINT_GUID(txt, guid) ({ \
>> +     debug("%sEFI: %s %pUl\n", __efi_nesting(), txt, guid); \
>> +     })
>> +
>>  extern struct efi_runtime_services efi_runtime_services;
>>  extern struct efi_system_table systab;
>>
>> diff --git a/lib/efi_loader/efi_boottime.c b/lib/efi_loader/efi_boottime.c
>> index 5aeaeeca75..6af033458b 100644
>> --- a/lib/efi_loader/efi_boottime.c
>> +++ b/lib/efi_loader/efi_boottime.c
>> @@ -112,6 +112,11 @@ static const char *indent_string(int level)
>>       return &indent[max - level];
>>  }
>>
>> +const char *__efi_nesting(void)
>> +{
>> +     return indent_string(nesting_level);
>> +}
>> +
>>  const char *__efi_nesting_inc(void)
>>  {
>>       return indent_string(nesting_level++);
>> @@ -1163,6 +1170,8 @@ static efi_status_t EFIAPI efi_locate_protocol(efi_guid_t *protocol,
>>       if (!protocol || !protocol_interface)
>>               return EFI_EXIT(EFI_INVALID_PARAMETER);
>>
>> +     EFI_PRINT_GUID("protocol", protocol);
>> +
>>       list_for_each(lhandle, &efi_obj_list) {
>>               struct efi_object *efiobj;
>>
>> @@ -1366,6 +1375,8 @@ static efi_status_t EFIAPI efi_open_protocol(
>>               goto out;
>>       }
>>
>> +     EFI_PRINT_GUID("protocol", protocol);
>> +
>>       switch (attributes) {
>>       case EFI_OPEN_PROTOCOL_BY_HANDLE_PROTOCOL:
>>       case EFI_OPEN_PROTOCOL_GET_PROTOCOL:
>>
>
> Hello Rob,
>
> with the patch installed I sometimes got failures on Travis when grub is
> loaded.
> https://travis-ci.org/xypron2/u-boot/builds/267753128?utm_source=github_status&utm_medium=notification
>
> I added some code for debugging and found that after efi_exit is called
> it opens the EFI_LOADED_IMAGE_PROTOCOL.
>
> This leads to an error in __efi_entry_check():
>
> do_bootefi_exec:270 Jumping to 0x9ee22400
>
> previous call
> /home/travis/build/xypron2/u-boot/lib/efi_loader/efi_boottime.c(869)
> efi_exit, ret = 0
>
>   EFI: protocol 5b1b31a1-9562-11d2-8e3f-00a0c969723b
>
> previous call
> /home/travis/build/xypron2/u-boot/lib/efi_loader/efi_boottime.c(1567)
> efi_open_protocol, ret = 0
>
> previous call
> /home/travis/build/xypron2/u-boot/lib/efi_loader/efi_boottime.c(1365)
> efi_locate_protocol, ret = 0
>
> previous call
> /home/travis/build/xypron2/u-boot/lib/efi_loader/efi_console.c(54)
> efi_cin_get_mode, ret = 0
>
> previous call
> /home/travis/build/xypron2/u-boot/lib/efi_loader/efi_boottime.c(1365)
> efi_locate_protocol, ret = 0
>
> previous call
> /home/travis/build/xypron2/u-boot/lib/efi_loader/efi_console.c(54)
> efi_cin_get_mode, ret = 0
>
> /home/travis/build/xypron2/u-boot/lib/efi_loader/efi_boottime.c:227:
> efi_allocate_pages_ext: Assertion `__efi_entry_check()' failed.
>
> resetting ...
>
> I will work on a patch to fix it.
>

hmm, odd that this patch would trigger it..

maybe adding 'ccflags-y += -DDEBUG=1' in efi_loader/Makefile (and then
possibly an #undef DEBUG at top of efi_console.c to make the debug
spam not *too* much) would help track it down?

I probably need to rebase my stack of patches again.  I've been busy
on other things and haven't had time for u-boot for last couple
weeks..

BR,
-R