Re: [PATCH v3 2/4] dt-bindings: nvmem: Add properties needed for blowing fuses

[Doug Anderson]

Hi,

On Thu, Jun 18, 2020 at 7:01 AM Srinivas Kandagatla
<srinivas.kandagatla@xxxxxxxxxx> wrote:
>
>
>
> On 18/06/2020 14:48, Doug Anderson wrote:
> > Hi,
> >
> > On Thu, Jun 18, 2020 at 3:10 AM Srinivas Kandagatla
> > <srinivas.kandagatla@xxxxxxxxxx> wrote:
> >>
> >> +Adding SBoyd.
> >>
> >> On 17/06/2020 18:22, Doug Anderson wrote:
> >>> Hi,
> >>>
> >>> On Wed, Jun 17, 2020 at 8:19 AM Srinivas Kandagatla
> >>> <srinivas.kandagatla@xxxxxxxxxx> wrote:
> >>>>
> >>>>
> >>>>
> >>>> On 17/06/2020 15:51, Douglas Anderson wrote:
> >>>>> From: Ravi Kumar Bokka <rbokka@xxxxxxxxxxxxxx>
> >>>>>
> >>>>> On some systems it's possible to actually blow the fuses in the qfprom
> >>>>> from the kernel.  Add properties to support that.
> >>>>>
> >>>>> NOTE: Whether this is possible depends on the BIOS settings and
> >>>>> whether the kernel has permissions here, so not all boards will be
> >>>>> able to blow fuses in the kernel.
> >>>>>
> >>>>> Signed-off-by: Ravi Kumar Bokka <rbokka@xxxxxxxxxxxxxx>
> >>>>> Signed-off-by: Douglas Anderson <dianders@xxxxxxxxxxxx>
> >>>>> ---
> >>>>>
> >>>>> Changes in v3:
> >>>>> - Add an extra reg range (at 0x6000 offset for SoCs checked)
> >>>>> - Define two options for reg: 1 item or 4 items.
> >>>>> - No reg-names.
> >>>>> - Add "clocks" and "clock-names" to list of properties.
> >>>>> - Clock is now "sec", not "secclk".
> >>>>> - Add "vcc-supply" to list of properties.
> >>>>> - Fixed up example.
> >>>>>
> >>>>>     .../bindings/nvmem/qcom,qfprom.yaml           | 45 ++++++++++++++++++-
> >>>>>     1 file changed, 43 insertions(+), 2 deletions(-)
> >>>>>
> >>>>> diff --git a/Documentation/devicetree/bindings/nvmem/qcom,qfprom.yaml b/Documentation/devicetree/bindings/nvmem/qcom,qfprom.yaml
> >>>>> index 5efa5e7c4d81..b195212c6193 100644
> >>>>> --- a/Documentation/devicetree/bindings/nvmem/qcom,qfprom.yaml
> >>>>> +++ b/Documentation/devicetree/bindings/nvmem/qcom,qfprom.yaml
> >>>>> @@ -17,8 +17,27 @@ properties:
> >>>>>         const: qcom,qfprom
> >>>>>
> >>>>>       reg:
> >>>>> -    items:
> >>>>> -      - description: The corrected region.
> >>>>> +    # If the QFPROM is read-only OS image then only the corrected region
> >>>>> +    # needs to be provided.  If the QFPROM is writable then all 4 regions
> >>>>> +    # must be provided.
> >>>>> +    oneOf:
> >>>>> +      - items:
> >>>>> +          - description: The corrected region.
> >>>>> +      - items:
> >>>>> +          - description: The corrected region.
> >>>>> +          - description: The raw region.
> >>>>> +          - description: The config region.
> >>>>> +          - description: The security control region.
> >>>>> +
> >>>>> +  # Clock must be provided if QFPROM is writable from the OS image.
> >>>>> +  clocks:
> >>>>> +    maxItems: 1
> >>>>
> >>>>
> >>>>> +  clock-names:
> >>>>> +    const: sec
> >>>>
> >>>> Do we need clock-names for just one clock here?
> >>>
> >>> I think technically you can get by without, but convention is that
> >>> clock-names are always provided for clocks.  It's talked about in the
> >>> same link I sent that talked about reg-names:
> >>>
> >>> https://lore.kernel.org/r/CAL_Jsq+MMunmVWqeW9v2RyzsMKP+=kMzeTHNMG4JDHM7Fy0HBg@xxxxxxxxxxxxxx/
> >>>
> >>
> >> TBH, This is total confusion!!!
> >>
> >> when to use "*-names" Device tree bindings is totally depended on Linux
> >> Subsystem interfaces!
> >>
> >> And what is the starting point to draw this line?
> >
> > Definitely confusing and mostly because the dts stuff grew organically
> > for a while there.  It does feel like Rob is pretty clear on the
> > current state of things and the policy in the link I provided, though.
> >
> >
> >>> Specifically, Rob said:
> >>>
> >>>> That probably is because the clock binding has had clock-names from
> >>>> the start (it may have been the first one). That was probably partly
> >>>> due to the clock API also was mainly by name already if we want to
> >>>> admit Linux influence on bindings
> >>>
> >>> Basically the standard way for getting clocks in Linux is
> >>> clk_get(name).  With just one clock you can call clk_get(NULL) and I
> >>> believe that works, but when you add the 2nd clock then you have to
> >>> switch APIs to one of the less-commonly-used variants.
> >>
> >> In previous NON-DT life clk_get api name argument comes from the clk
> >> names that clk provider registered the clocks with.
> >>
> >> If I remember this correctly, the name that is refereed here for
> >> clk_get() is old clkdev api based on clk_lookups and is not the same as
> >> clk-names that we have in Device tree. Atleast in this case!
> >>
> >> clk-names has two objectives in DT:
> >> 1> To find the index of the clock in the clocks DT property.
> >>
> >> 2> If actual clk name is specified then if "1" fails then name could
> >> potentially fallback to use old clkdev based clk_lookups.
> >>
> >> In this specific case we have "sec" as clock-names which is totally used
> >> for indexing into clocks property and it can not be used for (2) as
> >> there is no clk named "sec" registered by any of the clk providers.
> >>
> >> So this does not justify the reasoning why "clock-names" should be used
> >> while "reg-names" should not be used!. Both of them are going to be
> >> finally used for indexing into there respective properties.
> >
> > Right, you just have to accept the fact that logic doesn't come into
> > play here.  For clocks, always use "clk-names" but also always use a
> > consistent order (which is now more enforced by the schema checker).
> > For "reg" almost never use "reg-names".
> >
>
> On the other note:
>
> clock-names are not mandatory according to
> Documentation/devicetree/bindings/clock/clock-bindings.txt
>
> For this particular case where clock-names = "sec" is totally used for
> indexing and nothing else!

So I guess in the one-clock case it's more optional and if you feel
strongly I'll get rid of clk-names here.  ...but if we ever need
another clock we probably will want to add it back and (I could be
corrected) I believe it's convention to specify clk-names even with
one clock.

I won't say it's impossible to get by without clock-names when you
have more than one clock, but (almost) nobody does it.  It's hard to
quickly come up with a good way to search for this, but skimming
through:

git grep -C5 'clocks.*,' -- arch/arm64/boot/dts

...you don't find too many examples of no clock-names and you find
_lots_ of examples where clock-names are specified.  One example that
_does_ have multiple clocks and doesn't specify clock-names is
"simple-framebuffer".  Looking at the Linux driver you can see they
have to use the special "of_clk_get()" variant to handle it.


-Doug