[PATCH] riscv: Only enable OF_BOARD_FIXUP for S-Mode

[Sean Anderson]

On 9/14/20 2:38 AM, Bin Meng wrote:
> On Sat, Sep 12, 2020 at 2:25 AM Sean Anderson <seanga2 at gmail.com> wrote:
>>
>> On 9/11/20 10:43 AM, Bin Meng wrote:
>>> On Fri, Sep 11, 2020 at 6:20 PM Sean Anderson <seanga2 at gmail.com> wrote:
>>>>
>>>> On 9/11/20 3:29 AM, Bin Meng wrote:
>>>>> Hi Sean,
>>>>>
>>>>> On Sat, Sep 5, 2020 at 9:22 PM Sean Anderson <seanga2 at gmail.com> wrote:
>>>>>>
>>>>>> It is unsafe to enable OF_BOARD_FIXUP only based on OF_SEPARATE.
>>>>>> OF_SEPARATE may indicate that the user wishes U-Boot to use a different
>>>>>> device tree than one obtained via OF_PRIOR_STAGE. However, OF_SEPARATE may
>>>>>> also indicate that the device tree which would be obtained via
>>>>>> OF_PRIOR_STAGE is invalid, nonexistant, or otherwise unusable. In this
>>>>>
>>>>> typo: nonexistent
>>>>>
>>>>>> latter case, enabling OF_BOARD_FIXUP will result in corruption of the
>>>>>> device tree. To remedy this, only enable OF_BOARD_FIXUP if U-Boot is
>>>>>> configured for S-Mode.
>>>>>>
>>>>>> Fixes: 1c17e55594a394ced7de88d91be294eaf8c564c1
>>>>>
>>>>> nits: the format should be: commit_id ("description")>
>>>>>> Signed-off-by: Sean Anderson <seanga2 at gmail.com>
>>>>>> ---
>>>>>>
>>>>>>  arch/riscv/Kconfig | 2 +-
>>>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
>>>>>> index 009a545fcf..13fac51483 100644
>>>>>> --- a/arch/riscv/Kconfig
>>>>>> +++ b/arch/riscv/Kconfig
>>>>>> @@ -288,6 +288,6 @@ config STACK_SIZE_SHIFT
>>>>>>         default 14
>>>>>>
>>>>>>  config OF_BOARD_FIXUP
>>>>>> -       default y if OF_SEPARATE
>>>>>> +       default y if OF_SEPARATE && RISCV_SMODE
>>>>>
>>>>> Is that your board is running U-Boot M-mode with OF_SEPARATE that does not work?
>>>>
>>>> Yes, because the reason we use OF_SEPARATE is because no device tree is
>>>> passed to U-Boot. Trying to use the device tree passed to U-Boot even
>>>
>>> I don't get it. If no device tree is passed to U-Boot, why using
>>> OF_SEPARATE in the first place?
>>
>> Because it has to come from somewhere. Where else would U-Boot get the
>> device tree?
> 
> Sounds like there was some misunderstanding on "passed to U-Boot" ..
> But I got it now.
> 
>>
>>>> though OF_SEPARATE is enabled results in garbage being written to the
>>>
>>> What garbage data is written?
>>
>> It might not be garbage written. I didn't document the exact failure
>> mode at the time I discovered this bug, so I went back to try and
>> reproduce it for a more thorough analysis. However, I was unable to
>> reproduce this bug, even on the branch where I originally triggered it.
>> I documented my reasoning behind this patch at [1]. In my testing, I
>> could only trigger a "periodic-32" bug.
>>
>> In any case, this behavior could still cause problems in the future.
>> From my testing, on the k210, a1 usually holds some address on the ROM's
>> stack. However, if it (for instance) instead held an address which
> 
> So U-Boot on K210 boots with M-mode from the K210 ROM, and the ROM
> code does not hold DTB address in a1 before jumping to U-Boot, right?
> 
>> raised a load access fault, or was misaligned, then booting would fail.
>> In the general case, I was very surpised that U-Boot was using the value
>> of a1 on entry even with OF_SEPARATE specified. I would expect it only
>> to use that value if configured with OF_PRIOR_STAGE.
> 
> Because U-Boot S-mode needs to fix up the DT when OF_SEPERATE is used.

Right. It's just unexpected because OF_SEPARATE appears to imply to both
use a separate device tree and to not use the passed-in device tree.
This is because it is mutually exclusive with OF_PRIOR_STAGE. However,
with OF_BOARD_FIXUP, it's as if one has selected both OF_SEPARATE and
OF_PRIOR_STAGE at once. I think defaulting OF_BOARD_FIXUP to y only
S-Mode is more likely to result in unsurprising behavior on new boards.

--Sean