On 10/13/21 5:12 PM, Linus Walleij wrote:

On Tue, Oct 12, 2021 at 2:37 PM Alvin Šipraga [off-list ref] wrote:

quoted

This patch adds a realtek-smi subdriver for the RTL8365MB-VC 4+1 port
10/100/1000M switch controller. The driver has been developed based on a
GPL-licensed OS-agnostic Realtek vendor driver known as rtl8367c found
in the OpenWrt source tree.

(...)

quoted

Co-developed-by: Michael Rasmussen <redacted>
Signed-off-by: Michael Rasmussen <redacted>
Signed-off-by: Alvin Šipraga <alsi@bang-olufsen.dk>

Overall this driver looks very good :)

Some minor nits below:

quoted

+static irqreturn_t rtl8365mb_irq(int irq, void *data)
+{

(...)

quoted

+       if (!line_changes)
+               goto out_none;
+
+       while (line_changes) {
+               int line = __ffs(line_changes);
+               int child_irq;
+
+               line_changes &= ~BIT(line);
+
+               child_irq = irq_find_mapping(smi->irqdomain, line);
+               handle_nested_irq(child_irq);
+       }

What about just:

for_each_set_bit(offset, &line_changes, 32) {
   child_irq = irq_find_mapping(smi->irqdomain, line);
   handle_nested_irq(child_irq);
}

?

I don't know how many or which bits are valid IRQs, 16 maybe rather
than 32.

quoted

+static struct irq_chip rtl8365mb_irq_chip = {
+       .name = "rtl8365mb",
+       /* The hardware doesn't support masking IRQs on a per-port basis */
+};

I would rathe make this a dynamically allocated struct inside
struct rtl8365mb, so the irqchip lives with the instance of the
chip. (Which is nice if there would happen to be two of these
chips in a system.)

Forgive my ignorance, but is it actually necessary? Can't multiple 
instances of the switch still use the same irq_chip structure? That 
seems to be OK for the dummy chip, for example (dummychip.c). I also 
failed to find a driver which does it the way you suggest.

I will incorporate the rest of your feedback into v3. :)

quoted

+static int _rtl8365mb_irq_enable(struct realtek_smi *smi, bool enable)

I'm personally a bit allergic to _rand_underscore_naming, as sometimes
that means "inner function" and sometimes it means "compiler intrinsic"
I would just name it rtl8365mb_irq_config_commit()

(no strong opinion)

quoted

+       /* Configure chip interrupt signal polarity */
+       irq_trig = irqd_get_trigger_type(irq_get_irq_data(irq));

Nice that you preserve this edge trigger config from the machine
description (DT)!

With this fixed or not (your preference)
Reviewed-by: Linus Walleij <redacted>

Yours,
Linus Walleij