Re: [PATCH v5 3/3] efi: Add tee-based EFI variable driver

[Jan Kiszka <jan.kiszka@siemens.com>]

On 09.06.23 08:34, Ilias Apalodimas wrote:
> Hi Jan,
> 
> [...]
> 
>>> --- a/drivers/firmware/efi/vars.c
>>> +++ b/drivers/firmware/efi/vars.c
>>> @@ -123,7 +123,7 @@ EXPORT_SYMBOL_GPL(efivars_unregister);
>>>
>>>  bool efivar_supports_writes(void)
>>>  {
>>> -       return __efivars && __efivars->ops->set_variable;
>>> +       return __efivars && __efivars->ops->set_variable != set_variable_int;
>>>  }
>>>  EXPORT_SYMBOL_GPL(efivar_supports_writes);
>>>
>>> diff --git a/fs/efivarfs/super.c b/fs/efivarfs/super.c
>>> index e028fafa04f3..e40b5c4c5106 100644
>>> --- a/fs/efivarfs/super.c
>>> +++ b/fs/efivarfs/super.c
>>> @@ -242,9 +242,6 @@ static int efivarfs_fill_super(struct super_block
>>> *sb, struct fs_context *fc)
>>>         sb->s_d_op              = &efivarfs_d_ops;
>>>         sb->s_time_gran         = 1;
>>>
>>> -       if (!efivar_supports_writes())
>>> -               sb->s_flags |= SB_RDONLY;
>>> -
>>>         inode = efivarfs_get_inode(sb, NULL, S_IFDIR | 0755, 0, true);
>>>         if (!inode)
>>>                 return -ENOMEM;
>>> diff --git a/include/linux/efi.h b/include/linux/efi.h
>>> index 58d1c271d3b0..ec0ac6ef50a3 100644
>>> --- a/include/linux/efi.h
>>> +++ b/include/linux/efi.h
>>> @@ -1084,6 +1084,10 @@ int efivars_register(struct efivars *efivars,
>>>                      const struct efivar_operations *ops);
>>>  int efivars_unregister(struct efivars *efivars);
>>>
>>> +efi_status_t set_variable_int(efi_char16_t *name, efi_guid_t *vendor,
>>> +                             u32 attributes, unsigned long data_size,
>>> +                             void *data);
>>> +
>>>  void efivars_generic_ops_register(void);
>>>  void efivars_generic_ops_unregister(void);
>>>
>>> Thanks
>>> /Ilias
>>
>> As just written in my other reply: The root cause is the dependency on
>> tee-supplicant daemon. That needs to be resolved, and then also r/w
>> mounting will just work.
> 
> That's partially true.  If we solve the dependency your problem will
> go away only if everything gets compiled as built in.  But if you have
> them as modules there's still a chance you mount the efivarfs before
> installing all the modules.  In that case, you'll end up with the same
> problem no?

Obviously, this will need proper probing of the TA services in the
proper order so that the STMM driver is pulled in before efivarfs gets used.

> 
> That's why I think this patch (or a variation of it) is useful.  It
> solves the kernel panic you are seeing if you remount the efivarfs as
> RW and It unifies the way the kernel responds to userspace no matter
> what the firmware does with its setvariableRT service.

I'm not against fixes crashes, but the r/w issue is a different thing IMHO.

Jan

-- 
Siemens AG, Technology
Competence Center Embedded Linux