Re: [PATCH v3 3/6] thermal: tegra: soctherm-fuse: prepare calibration for Tegra114 support

[Svyatoslav Ryhel]

чт, 21 серп. 2025 р. о 10:42 Mikko Perttunen <mperttunen@xxxxxxxxxx> пише:
>
> On Wednesday, August 20, 2025 8:42 PM Svyatoslav Ryhel wrote:
> > The Tegra114 has a different fuse calibration register layout and address
> > compared to other Tegra SoCs, requiring SOCTHERM shift, mask, register
> > address, and nominal calibration values to be configurable.
> >
> > Additionally, a use_lower_precision option was implemented to account for
> > the Tegra114's 0.5C thermal data output, which differs from the 1C
> > precision of newer SoCs.
> >
> > Signed-off-by: Svyatoslav Ryhel <clamor95@xxxxxxxxx>
> > ---
> >  drivers/thermal/tegra/soctherm-fuse.c     | 31 ++++++++++++++++-------
> >  drivers/thermal/tegra/soctherm.h          |  8 +++++-
> >  drivers/thermal/tegra/tegra124-soctherm.c |  6 +++++
> >  drivers/thermal/tegra/tegra132-soctherm.c |  6 +++++
> >  drivers/thermal/tegra/tegra210-soctherm.c |  6 +++++
> >  5 files changed, 47 insertions(+), 10 deletions(-)
> >
> > diff --git a/drivers/thermal/tegra/soctherm-fuse.c
> > b/drivers/thermal/tegra/soctherm-fuse.c index 190f95280e0b..d27876dd9b2a
> > 100644
> > --- a/drivers/thermal/tegra/soctherm-fuse.c
> > +++ b/drivers/thermal/tegra/soctherm-fuse.c
> > @@ -9,15 +9,10 @@
> >
> >  #include "soctherm.h"
> >
> > -#define NOMINAL_CALIB_FT                     105
> > -#define NOMINAL_CALIB_CP                     25
> > -
> >  #define FUSE_TSENSOR_CALIB_CP_TS_BASE_MASK   0x1fff
> >  #define FUSE_TSENSOR_CALIB_FT_TS_BASE_MASK   (0x1fff << 13)
> >  #define FUSE_TSENSOR_CALIB_FT_TS_BASE_SHIFT  13
> >
> > -#define FUSE_TSENSOR_COMMON                  0x180
> > -
> >  /*
> >   * Tegra210: Layout of bits in FUSE_TSENSOR_COMMON:
> >   *    3                   2                   1                   0
> > @@ -26,7 +21,7 @@
> >   * |       BASE_FT       |      BASE_CP      | SHFT_FT | SHIFT_CP  |
> >   * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
> >   *
> > - * Tegra12x, etc:
> > + * Tegra124:
> >   * In chips prior to Tegra210, this fuse was incorrectly sized as 26 bits,
> >   * and didn't hold SHIFT_CP in [31:26]. Therefore these missing six bits
> >   * were obtained via the FUSE_SPARE_REALIGNMENT_REG register [5:0].
> > @@ -44,6 +39,13 @@
> >   * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
> >   * |---------------------------------------------------| SHIFT_CP  |
> >   * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
> > + *
> > + * Tegra114: Layout of bits in FUSE_TSENSOR_COMMON aka FUSE_VSENSOR_CALIB:
> > + *    3                   2                   1                   0
> > + *  1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0
> > + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
> > + * | SHFT_FT |       BASE_FT       | SHIFT_CP  |      BASE_CP      |
> > + * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
> >   */
> >
> >  #define CALIB_COEFFICIENT 1000000LL
> > @@ -77,7 +79,7 @@ int tegra_calc_shared_calib(const struct
> > tegra_soctherm_fuse *tfuse, s32 shifted_cp, shifted_ft;
> >       int err;
> >
> > -     err = tegra_fuse_readl(FUSE_TSENSOR_COMMON, &val);
> > +     err = tegra_fuse_readl(tfuse->fuse_common_reg, &val);
> >       if (err)
> >               return err;
> >
> > @@ -96,10 +98,21 @@ int tegra_calc_shared_calib(const struct
> > tegra_soctherm_fuse *tfuse, return err;
> >       }
> >
> > +     shifted_cp = (val & tfuse->fuse_shift_cp_mask) >>
> > +                  tfuse->fuse_shift_cp_shift;
> >       shifted_cp = sign_extend32(val, 5);
> >
> > -     shared->actual_temp_cp = 2 * NOMINAL_CALIB_CP + shifted_cp;
> > -     shared->actual_temp_ft = 2 * NOMINAL_CALIB_FT + shifted_ft;
> > +     shared->actual_temp_cp = 2 * tfuse->nominal_calib_cp + shifted_cp;
> > +     shared->actual_temp_ft = 2 * tfuse->nominal_calib_ft + shifted_ft;
> > +
> > +     /*
> > +      * Tegra114 provides fuse thermal corrections in 0.5C while newer
> > +      * SoCs provide data in 1C
> > +      */
>
> I've been looking a bit into these fuses, and from what I can tell the
> precision for these fuses should be in 0.5C units for all of Tegra114, 124,
> and 210. The documented nominal CP (cold) and FT (hot) temperatures for
> Tegra114 should be 25C and 90C respectively.
>
> The reason for the code '2 * NOMINAL_CALIB_XX + shifted_xx' then is that the
> value of 'actual_temp_xx' is in 0.5C units -- NOMINAL_CALIB_XX being in 1C
> units and being multiplied by 2 to match the units of the shifted_xx values
> coming from fuses.
>
> If you're getting correct values with your code, clearly there's more hijinks
> going on.
>

I have based this code on downstream kernel which sometimes can be
quite challenging to understand correctly. If you assume that Tegra114
fits into existing driver even more, that is great, I will test it and
remove unnecessary parts.