[PATCH v22 5/7] soc: mediatek: SVS: add debug commands

AngeloGioacchino Del Regno angelogioacchino.delregno at collabora.com
Mon Jan 31 03:11:51 PST 2022


Il 27/01/22 04:39, Roger Lu ha scritto:
> The purpose of SVS is to help find the suitable voltages
> for DVFS. Therefore, if SVS bank voltages are concerned
> to be wrong, we can adjust SVS bank voltages by this patch.
> 
> Signed-off-by: Roger Lu <roger.lu at mediatek.com>


Hello Roger,
I was thinking about what this patch is adding... and I have a few considerations.

It's nice to have a debugging mechanism to read the status and dump registers, as
that's very helpful when doing heavy debugging of the IP... but adding the
possibility to write a voltage offset may be very dangerous: think about the case
in which, either for misconfiguration, or for any other reason, the debugfs entry
that allows writing voffset becomes user-writable, or a user writes an impossibly
high voffset.
In case a very low (negative) voffset is entered, the platform would crash (denial
of service); if a very high voffset is entered, hardware damage may occur.

For this reason, there are two proposals:
1. If you want to keep the debugfs voffset write, please constrain the permissible
    voffset to an acceptable range that at least makes it unlikely to damage the HW;
    Moreover, since voffset write is a feature that would be used in very limited
    debugging cases, I think that this should be implemented over a build-time
    configuration barrier... something like CONFIG_MTK_SVS_DEBUG_ALLOW_WRITE, or
    similar;
2. Since it's very unlikely for someone to really play that much with a voltage
    offset during runtime, and since this looks like something very machine specific
    (perhaps addressing board-specific quirks?), I would suggest to add this as a
    device-tree parameter instead, such as "mediatek,svs-voffset", as it is indeed
    possible to specify both positive or negative values in DT.

I would prefer proposal 2, as it looks generally cleaner and way less risky.

Regards,
Angelo



More information about the Linux-mediatek mailing list