On Tue, 25 Aug 2020, Rob Herring wrote:
> On Tue, Aug 25, 2020 at 4:02 PM Nicolas Pitre <nico@xxxxxxxxxxx> wrote:
>
> > Currently there are very few implementations. One of them lives in an
> > FPGA and the example below is actually the DT entry I use for it. I'm
> > guessing specific vendor implementations will have their own tweaks
> > eventually, such as clock sources and whatnot.
>
> Yes, exactly. And bugs too.
Obviously. ;-)
> > But that is outside of
> > the spec (actually the spec defines a register area for eventual vendor
> > specific usage). But I have no visibility into that and of course the
> > code has no provision for that yet either.
> >
> > So I imagine there will be something like this in dts files eventually:
> >
> > compatibvle = "intel,foobar_soc_i3c_hci", "mipi-i3c-hci";
> >
> > Is that what you mean?
>
> Yes. Even your FPGA is tied to some implementation...
It is, but so far it's self-discoverable.
> > > Also, which version of the spec does this compatible correspond to?
> >
> > All of them.
> >
> > > Or are there not HCI differences in the spec versions you mention in
> > > the cover letter?
> >
> > The hardware is self advertising per the spec. So there is no need to
> > carry such distinction in the DT compatible. Even vendor extensions are
> > tagged with MIPI vendor IDs in the hardware directly.
>
> Oh good, folks are learning. :)
>
> Is the vendor ID (and revision) discoverable even if no vendor
> extensions?
Yes. It's even in the very first 32-bit word from the register space.
Here's the relevant code:
#define HCI_VERSION 0x00 /* HCI Version (in BCD) */
[...]
/* Validate HCI hardware version */
regval = reg_read(HCI_VERSION);
hci->version_major = (regval >> 8) & 0xf;
hci->version_minor = (regval >> 4) & 0xf;
hci->revision = regval & 0xf;
NOTE("HCI v%u.%u r%02u",
hci->version_major, hci->version_minor, hci->revision);
/* known versions */
switch (regval & ~0xf) {
case 0x100: /* version 1.0 */
case 0x110: /* version 1.1 */
case 0x200: /* version 2.0 */
break;
default:
ERR("unsupported HCI version");
return -EPROTONOSUPPORT;
}
Then there is a register that provides the relative offset to another
register area where "extended attributes" are to be found. Those
attributes are also spec defined. One of the standard attributes
contains the MIPI vendor ID, the vendor version ID and vendor product
ID. And then there is a range of attributes which are vendor defined.
That's where specific stuff like fancy clock controls would be located
and vendor specific tweaks to be applied. But so far
everything can be predicated on hardware-provided data.
> If so, then I'm more comfortable with "mipi-i3c-hci" on it's own. The
> exception will be if there's setup needed to discover the h/w which
> seems likely. In that case, we should probably do compatible strings
> based on VID/PID like PCI, USB, etc. No need to define that now I
> guess, but please add some sort of summary of the above about the
> discoverability of the HCI implementer and features.
OK.
Nicolas
