On 27/02/2025 23:22:36+0900, William Breathitt Gray wrote:
> On Thu, Feb 27, 2025 at 02:53:30PM +0100, Kamel Bouhara wrote:
> > On Thu, Feb 27, 2025 at 01:59:57PM +0900, William Breathitt Gray wrote:
> > > Let me make sure I understand the situation correctly. This SoC has two
> > > Timer Counter Blocks (TCB) and each TCB has three Timer Counter Channels
> > > (TCC); each TCC has a Counter Value (CV) and three general registers
> > > (RA, RB, RC); RA and RB can store Captures, and RC can be used for
> > > Compare operations.
> > >
> > > If that is true, then the correct way for this hardware to be exposed is
> > > to have each TCB be a Counter device where each TCC is exposed as a
> > > Count. So for this SoC: two Counter devices as counter0 and counter1;
> > > count0, count1, and count2 as the three TCC; i.e. counter0/count{0,1,2}
> > > and counter1/count{0,1,2}.
>
> [...]
>
> > > Kamel, what would it take for us to rectify this situation so that the
> > > TCC are organized together by TCB under the same Counter devices?
> >
> > Hello,
> >
> > Indeed, each TCC operates independently except when quadrature mode is
> > enabled. I assume this approach was taken to provide more flexibility by
> > exposing them separately.
> >
> > Currently only one channel is configured this would need to rework the
> > driver to make the 3 TCCs exposed.
> >
> > Greetings,
> > Kamel
>
> Skimming through the driver, it looks like what we'll need is for
> mchp_tc_counts[] to have all three TCCs defined, then have
> mchp_tc_probe() match on a TCB node and configure each TCC. Once that's
> setup, then whenever we need to identify which TCC a callback is
> exposing, we can get it from count->id.
>
> So for example, the TC_CV register offset is calculated as 0x00 +
> channel * 0x40 + 0x10. In the count_read() callback we can leverage
> count->id to identify the TCC and thus get the respective TC_CV register
> at offset + count->id * 0x40 + 0x10.
>
We can't do that because the TCC of a single TCB can have a mix of
different features. I struggled with the breakage to move away from the
one TCB, one feature state we had.
Be fore this, it was not possible to mix features on a single TCB, now,
we can have e.g. the clocksource on TCC 0 and 1 of TCB0 and a PWM on
TCC 2. mchp_tc_probe must not match on a TCB node...
--
Alexandre Belloni, co-owner and COO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
