Re: [RFC v1 2/3] drivers: nvmem: Add driver for QTI qfprom-efuse support

[Doug Anderson]

Hi,

On Wed, May 20, 2020 at 7:35 AM Srinivas Kandagatla
<srinivas.kandagatla@xxxxxxxxxx> wrote:
>
> On 18/05/2020 19:31, Doug Anderson wrote:
> > Hi,
> >
> > On Mon, May 18, 2020 at 3:45 AM Srinivas Kandagatla
> > <srinivas.kandagatla@xxxxxxxxxx> wrote:
> >>
> >> On 18/05/2020 11:39, Ravi Kumar Bokka (Temp) wrote:
> >>>
> >>> Based on the compatible, do i need to separate probe function for
> >>> qfprom-efuse and maintain separate nvmem object to register nvmem
> >>> framework. Is this what you are suggesting to implementing this in to
> >>> one existing driver?
> >>
> >> Yes for same driver we should add new compatible string and add support
> >> to this in existing qfprom driver.
> >> Ideally we should allocate nvmem_config object at probe with different
> >> parameters based on compatible string.
> >
> > I wish I had better documentation for exactly what was in the SoC
> > instead of the heavily redacted stuff Qualcomm provides.  Really the
> > answer here is: how do you best describe the hardware?  OK, so I just
> > spent the past hour or so trying to patch together all the bits and
> > fragments that Qualcomm provided me.  Just like a scavenger hunt!
> > Fun!  The best I can patch together is that there is a single QFPROM
> > with these ranges:
> >
> > 0x00780000 - 0x007800FF
> > QFPROM HW registers, range 1/2
> >
> > 0x00780120 - 0x007808FF
> > QFPROM "raw" space
> >
>
> so this is the only region is the QFPROM fuses can be programmed!
>
> > 0x00782000 - 0x007820FF
> > QFPROM HW registers, range 2/2
> >
> > 0x00784120 - 0x007848FF
> > QFPROM "corrected" space
>
> Is this some kind of FEC corrected regions?

Yes.


> > 0x00786000 - 0x00786FFF
> > QFPROM memory range that I don't really understand and maybe we don't
> > worry about right now?
>
> >
> > Did I get that right?  If so, is there a prize for winning the scavenger hunt?
> >
> > ---
> >
> > If so then, IMO, it wouldn't be insane to actually keep it as two
> > drivers and two device tree nodes, as you've done.  I'd defer to
> > Srinivas and Rob Herring, though.  The existing driver would be a
> > read-only driver and provide access to the "corrected" versions of all
> > the registers.  Its node would have "#address-cells = <1>" and
> > "#size-cells = <1>" because it's expected that other drivers might
> > need to refer to data stored here.
> >
> > Your new driver would be read-write and provide access to the "raw"
> > values.  A read from your new driver would not necessarily equal a
> > read from the old driver if the FEC (forward error correction) kicked
>
> Is this only applicable for corrected address space?

I guess I was proposing a two dts-node / two drive approach here.

dts node #1:just covers the memory range for accessing the FEC-corrected data
driver #1: read-only and reads the FEC-corrected data

dts node #2: covers the memory range that's _not_ the FEC-corrected
memory range.
driver #2: read-write.  reading reads uncorrected data

Does that seem sane?


> > in.  Other drivers should never refer to the non-corrected values so
> > you wouldn't have "#address-cells" and "#size-cells".  The only way to
> > really read or write it would be through sysfs.
> >
> > It would be super important to document what's happening, of course.
> > ...and ideally name them to make it clearer too.
> >
> > ---
> >
> > Another alternative (if Srinivas and/or Rob H prefer it) would be to
> > deprecate the old driver and/or bindings and say that there really
> > should just be one node and one driver.  In that case you'd replace
> > the old node with:
> >
> > qfprom@780000 {
> >    compatible = "qcom,sc7180-qfprom-efuse";
>
> May be "qcom,sc7180-qfprom"
>
>
> >    reg = <0 0x00780000 0 0x6fff>;
> >    #address-cells = <1>;
> >    #size-cells = <1>;
> >
> >    clocks = <&gcc GCC_SEC_CTRL_CLK_SRC>;
> >    clock-names = "sec";
> >
> >    qusb2p_hstx_trim: hstx-trim-primary@25b {
> >      reg = <0x25b 0x1>;
> >      bits = <1 3>;
> >    };
> > };
> >
> > You'd use the of_match_table solution to figure out the relevant
> > offsets (0x120, 0x2000, 0x4120, 0x6000) for sc7180 and this new driver
> > would be responsible for being able to read the corrected values and
>
>
> Encompassing these offsets in driver as part of the register defines
> itself should be a good start!
>
> It will also be nice to understand how similar this thing is with w.rt
> other Qcom SoCs?

At least sdm845 is about the same.  I cross-referenced docs I had with
sc7180 and sdm845 and that's how I came up with my model for how this
works.


> > also for programming.  In this case I'm not sure how (assuming it's
> > valuable) you'd provide read access to the uncorrected data.
> I will leave this question to the author of the driver.
>
> --srini
>
> >
> >
> > -Doug
> >