On Wed, Aug 15, 2018 at 8:51 AM Michal Simek <michal.simek@xxxxxxxxxx> wrote:
>
> Hi Rob,
>
> On 15.8.2018 16:34, Tom Rini wrote:
> > On Wed, Aug 15, 2018 at 04:30:15PM +0200, Michal Simek wrote:
> >> On 15.8.2018 16:17, Tom Rini wrote:
> >>> On Mon, Aug 13, 2018 at 05:53:38PM +0200, Jens Wiklander wrote:
> >>>
> >>>> Just as /chosen may contain devices /firmware may contain devices, scan
> >>>> for devices under /firmware too.
> >>>>
> >>>> Signed-off-by: Jens Wiklander <jens.wiklander@xxxxxxxxxx>
> >>>> ---
> >>>> drivers/core/root.c | 15 ++++++++++-----
> >>>> 1 file changed, 10 insertions(+), 5 deletions(-)
> >>>>
> >>>> diff --git a/drivers/core/root.c b/drivers/core/root.c
> >>>> index 72bcc7d7f2a3..0dca507e1187 100644
> >>>> --- a/drivers/core/root.c
> >>>> +++ b/drivers/core/root.c
> >>>> @@ -265,9 +265,15 @@ static int dm_scan_fdt_node(struct udevice *parent, const void *blob,
> >>>> for (offset = fdt_first_subnode(blob, offset);
> >>>> offset > 0;
> >>>> offset = fdt_next_subnode(blob, offset)) {
> >>>> - /* "chosen" node isn't a device itself but may contain some: */
> >>>> - if (!strcmp(fdt_get_name(blob, offset, NULL), "chosen")) {
> >>>> - pr_debug("parsing subnodes of \"chosen\"\n");
> >>>> + const char *node_name = fdt_get_name(blob, offset, NULL);
> >>>> +
> >>>> + /*
> >>>> + * The "chosen" and "firmware" nodes aren't devices
> >>>> + * themselves but may contain some:
> >>>> + */
> >>>> + if (!strcmp(node_name, "chosen") ||
> >>>> + !strcmp(node_name, "firmware")) {
> >>>> + pr_debug("parsing subnodes of \"%s\"\n", node_name);
> >>>>
> >>>> err = dm_scan_fdt_node(parent, blob, offset,
> >>>> pre_reloc_only);
> >>>
> >>> So, the /firmware node seems special. Have you talked with the
> >>> devicetree folks to get it listed in the spec? That would seem rather
> >>> valuable for something used by many parties. Thanks!
> >>>
> >>
> >> some days ago we have sent a patch for this too.
> >> https://lists.denx.de/pipermail/u-boot/2018-August/338049.html
> >>
> >> It is using different way but it should do the same thing.
> >
> > OK, so it sounds like in terms of code clean-ups, we need something like
> > what you reference and then some further clean-ups on top of that
> > perhaps for other places to call dm_scan_fdt_ofnode_path() for special
> > cases. And in terms of formalized specification bits, I do think
> > /firmware should perhaps get kicked up to the spec itself so that it's
> > very clear to all consumers.
>
> I was also checking latest devicetree spec and there is no record about
> /firmware node and how it is supposed to be used.
Patches welcome. :)
It's really only a container to define non-discoverable firmware
interfaces. Typically accessed thru a secure call (for ARM) or a
mailbox. It is primarily just convention rather than a strict
requirement. We have to support firmware nodes at the top-level too
anyways.
Rob
