Re: [PATCH 1/2] firmware: arm_scmi: Fix frequency truncation by promoting multiplier to u64

[Sudeep Holla]

On Fri, Dec 01, 2023 at 04:17:56PM +0000, Cristian Marussi wrote:
> On Fri, Dec 01, 2023 at 02:39:35PM +0000, Sudeep Holla wrote:
> > On Thu, Nov 30, 2023 at 08:43:42PM +0000, Sudeep Holla wrote:
> > > Fix the frequency truncation for all values equal to or greater 4GHz by
> > > updating the multiplier 'mult_factor' to u64 type. It is also possible
> > > that the multiplier itself can be greater than or equal to 2^32. So we need
> > > to also fix the equation computing the value of the multiplier.
> > > 
> > > Fixes: a9e3fbfaa0ff ("firmware: arm_scmi: add initial support for performance protocol")
> > > Reported-by: Sibi Sankar <quic_sibis@xxxxxxxxxxx>
> > > Closes: https://lore.kernel.org/all/20231129065748.19871-3-quic_sibis@xxxxxxxxxxx/
> > > Cc: Cristian Marussi <cristian.marussi@xxxxxxx>
> > > Signed-off-by: Sudeep Holla <sudeep.holla@xxxxxxx>
> > > ---
> > >  drivers/firmware/arm_scmi/perf.c | 6 +++---
> > >  1 file changed, 3 insertions(+), 3 deletions(-)
> > > 
> > > diff --git a/drivers/firmware/arm_scmi/perf.c b/drivers/firmware/arm_scmi/perf.c
> > > index 81dd5c5e5533..8ce449922e55 100644
> > > --- a/drivers/firmware/arm_scmi/perf.c
> > > +++ b/drivers/firmware/arm_scmi/perf.c
> > > @@ -152,7 +152,7 @@ struct perf_dom_info {
> > >  	u32 opp_count;
> > >  	u32 sustained_freq_khz;
> > >  	u32 sustained_perf_level;
> > > -	u32 mult_factor;
> > > +	u64 mult_factor;
> > 
> > I have now changed this to unsigned long instead of u64 to fix the 32-bit
> > build failure[1].
> 
> Right, I was caught a few times too by this kind of failures on v7 :D
>

😄

> ... but this 32bit issue makes me wonder what to do in such a case...
>

Same here, but the frequency calculations are also unsigned long in higher
layers, so I don't see any point in making it u64(also 32-bit doesn't
support 32bit value to be divided by a 64bit value which adds unnecessary
complications here).

> ...I mean, on 32bit if the calculated freq oveflows, there is just
> nothing we can do on v7 without overcomplicating the code...but I suppose
> it is unplausible to have such high freq on a v7...

Yes this is exactly the argument I made myself and got convinced to keep
it unsigned long(KISS approach) unless we need it on v7.

> as a palliative I can only think of some sort of overflow check (only on v7)
> that could trigger a warning ... but it is hardly worth the effort
> probably..
>

Not sure myself.

--
Regards,
Sudeep