[PATCH] imx: Revert "imx: mx6ull: fix REFTOP_VBGADJ setting" and fix comments

[Emanuele Ghidoli]

On 31/10/2025 02:48, Fabio Estevam wrote:
> On Mon, Oct 27, 2025 at 8:56 AM Fabio Estevam <festevam at gmail.com> wrote:
>>
>> From: Ye Li <ye.li at nxp.com>
>>
>> The change in commit c1af358cf51a ("imx: mx6ull: fix REFTOP_VBGADJ
>> setting") is not correct. According to latest i.MX6ULL fuse table,
>> the REFTOP_TRIM[2:0] fuse is defined as below. So the previous codes
>> are correct, only need to fix the comments.
>>
>> Set the REFTOP_VBGADJ field in PMU_MISC0 register based on the
>> REFTOP_TRIM fuse:
>> '000" - set REFTOP_VBGADJ[2:0] to 3'b000
>> '001" - set REFTOP_VBGADJ[2:0] to 3'b001
>> '010" - set REFTOP_VBGADJ[2:0] to 3'b010
>> '011" - set REFTOP_VBGADJ[2:0] to 3'b011
>> '100" - set REFTOP_VBGADJ[2:0] to 3'b100
>> '101" - set REFTOP_VBGADJ[2:0] to 3'b101
>> '110" - set REFTOP_VBGADJ[2:0] to 3'b110
>> '111" - set REFTOP_VBGADJ[2:0] to 3'b111
>>
>> Sven Schwermer reported:
>>
>> "These are the symptoms: Very few of the thousands of devices that we
>> manufacture (< 1%) fail to decompress a large XZ archive once in Linux
>> userspace. We have not found any form of memory testing or other load
>> that reproduces this issue. Reverting the offending patch fixes this
>> behavior on the affected devices."
>>
>> Revert the offending commit to fix the regression.
>>
>> Reported-by: Sven Schwermer <u-boot at schwermer.no>
>> Signed-off-by: Ye Li <ye.li at nxp.com>
>> [fabio: Fixed checkpatch warning and added Sven's report]
>> Signed-off-by: Fabio Estevam <festevam at gmail.com>
> 
> Applied, thanks.
Hi Fabio, Ye,

At Toradex we have been struggling since quite some time (back to 2018) with
the temperature readings and the thermal shutdown behavior on i.MX6ULL. The
issue we observed is that the thermal shutdown seems to trigger too early.

I have investigated the history around this topic and verified some internal
notes in our bug tracker. With this revert applied, and on the devices we have
(all trimmed with value 0), the reported temperature increases by about 10 °C
compared to before the revert. The absolute accuracy of the readout is
questionable: for example, on a device kept at around 20 °C, the temperature
readout was ~19 °C before the revert, and ~31 °C after the revert.

Consider that for devices with fuse value of 6, there is a -10 degrees offset
after this revert (so, the other way around).

I understand that for Sven this revert improves stability during large XZ
decompression in userspace for few devices, but for us it means that the
thermal shutdown happens roughly 10 °C earlier, which is a serious issue in
our case.

I would kindly ask NXP to double-check how these devices were trimmed. My gut
feeling is that there might be an inconsistency, either the trimming scheme is
not backward-compatible (e.g. old and new devices behave differently, and the
code should cope, I don't know how, with this), or this patch must be reverted
(because introduces a regression).

Thank you and kind regards,
Emanuele