[PATCH v1] ARM: tegra: Fix misplaced tegra_uart_config in decompressor

[Florian Fainelli]

On 12/15/2020 10:56 AM, Dmitry Osipenko wrote:
> 15.12.2020 21:22, Florian Fainelli пишет:
>>
>>
>> On 12/15/2020 8:53 AM, Dmitry Osipenko wrote:
>>> 15.12.2020 19:40, Florian Fainelli пишет:
>>>>
>>>>
>>>> On 12/15/2020 8:17 AM, Dmitry Osipenko wrote:
>>>>> 15.12.2020 19:04, Florian Fainelli пишет:
>>>>>>
>>>>>>
>>>>>> On 12/15/2020 5:52 AM, Dmitry Osipenko wrote:
>>>>>>> The tegra_uart_config of the DEBUG_LL code is now placed right at the
>>>>>>> start of the .text section after commit which enabled debug output in the
>>>>>>> decompressor. Tegra devices are not booting anymore if DEBUG_LL is enabled
>>>>>>> since tegra_uart_config data is executes as a code. Fix the misplaced
>>>>>>> tegra_uart_config storage by embedding it into the code.
>>>>>>>
>>>>>>> Cc: stable at vger.kernel.org
>>>>>>> Fixes: 2596a72d3384 ("ARM: 9009/1: uncompress: Enable debug in head.S")
>>>>>>> Signed-off-by: Dmitry Osipenko <digetx at gmail.com>
>>>>>>> ---
>>>>>>>  arch/arm/include/debug/tegra.S | 54 +++++++++++++++++-----------------
>>>>>>>  1 file changed, 27 insertions(+), 27 deletions(-)
>>>>>>
>>>>>> Looks like arch/arm/include/debug/brcmstb.S would need the same
>>>>>> treatment since the implementation was copied from tegra.S.
>>>>>>
>>>>>
>>>>> Good catch, will you be able to test the brcm and make a patch?
>>>>
>>>> Yes, absolutely, building a kernel to test right now.
>>>>
>>>
>>> Thank you.
>>>
>>> BTW, I noticed that the problem is more visible on a thumb2 kernel
>>> build, i.e. you should get a more reliable hang on boot. On a non-thumb2
>>> kernel the hanging behaviour seems depends on a device / bootloader. I
>>> haven't tried to figure out what exactly makes the difference, perhaps
>>> it should be a memory layout / state.
>>
>> To build with a CONFIG_THUMB2_KERNEL I had to fetch:
>>
>> https://www.armlinux.org.uk/developer/patches/viewpatch.php?id=9018/2
>>
>> to avoid a build error, too bad this missed v5.10 final but hopefully it
>> can make it soon.
> 
> The VFP fix was applied to the -next very recently, it should propagate
> to v5.10 eventually.
> 
>> With CONFIG_THUMB2_KERNEL=y, I am not getting the head.S output where it
>> prints the start/end of the compressed kernel:
>>
>> C:0x420800C0-0x4321B0E0->0x4212AB00-0x432C5B20
>> Uncompressing Linux... done, booting the kernel.
>> [    0.000000] Booting Linux on physical CPU 0x0
>> [    0.000000] Linux version 5.10.0-g148842c98a24
>> (fainelli at fainelli-desktop) (arm-linux-gcc (GCC) 8.3.0, GNU ld (GNU
>> Binutils) 2.32) #71 SMP Tue Dec 15 09:53:09 PST 2020
>>
>> I am only getting:
>>
>> Uncompressing Linux... done, booting the kernel.
>>
>> Is that the same for you?
> 
> No, start/end are printed for both THUMB2 and ARM kernels here.
> 
>> Looking at the disassembly of head.o it definitively has
>> brcmstb_uart_config in the .text section as the beginning just like you
>> mentioned in your commit message.
>>
>> Disassembly of section .text:
>>
>> 00000000 <brcmstb_uart_config>:
>>    0:   00000001        andeq   r0, r0, r1
>>         ...
>>    c:   467c            mov     r4, pc
>>    e:   f004 4478       and.w   r4, r4, #4160749568     ; 0xf8000000
>>   12:   f504 4400       add.w   r4, r4, #32768  ; 0x8000
>>   16:   4678            mov     r0, pc
>>   18:   42a0            cmp     r0, r4
>>   1a:   bf3f            itttt   cc
>>   1c:   48d4            ldrcc   r0, [pc, #848]  ; (370 <LC1+0x8>)
>>   1e:   4478            addcc   r0, pc
>>   20:   4284            cmpcc   r4, r0
>>   22:   f044 0401       orrcc.w r4, r4, #1
>>   26:   bf28            it      cs
>>   28:   f000 f9aa       blcs    380 <cache_on>
>>
>> however after applying a fix similar to yours, we do end-up with the
>> expected data embedded within the code and given brcmstb.S would be
>> subject to the same issue as tegra.S, it would not hurt.
>>
> 
> Have you checked whether start/end printed after applying the fix?

Yes I did, it is not printed when CONFIG_THUMB2_KERNEL=y, but it is when
CONFIG_THUMB2_KERNEL=n. I don't have a JTAG adapter on this board right
now to single step.
-- 
Florian