Re: [PATCH] net: dsa: mt7530: disable LEDs before reset

Arınç

On 2024-03-12 04:41, Arınç ÜNAL wrote:

quoted

Wouldn't it be a better idea to increase the delay between
reset_control_assert() and reset_control_deassert(), and
gpiod_set_value_cansleep(priv->reset, 0) and
gpiod_set_value_cansleep(priv->reset, 1) instead of disabling the LED
controller and delaying before reset?

I've done some additional tests to see what the difference would be
between increasing the reset hold period vs. disabling the LEDs then
sleeping before reset.


I wanted to know:

When an active link is present on Port 3 at boot, what are the
minimum, maximum and average periods between ESW_P3_LED_0
going low and the signal inversion that seems to co-incide with
reset_control_deassert() for each approach?


I created two patches:

WITH INCREASED RESET DELAY

As I submitted a patch that added an intended 1000 - 1100 usec delay
before reset, I thought it would be fair to increase the reset hold
period by the same value.

---%---

--- a/drivers/net/dsa/mt7530.c
+++ b/drivers/net/dsa/mt7530.c

@@ -2243,11 +2243,11 @@ mt7530_setup(struct dsa_switch *ds)
          */
         if (priv->mcm) {
                 reset_control_assert(priv->rstc);
-               usleep_range(1000, 1100);
+               usleep_range(2000, 2200);
                 reset_control_deassert(priv->rstc);
         } else {
                 gpiod_set_value_cansleep(priv->reset, 0);
-               usleep_range(1000, 1100);
+               usleep_range(2000, 2200);
                 gpiod_set_value_cansleep(priv->reset, 1);
         }
---%---

DISABLE LEDS BEFORE RESET

Reset hold period unchanged from the intended 1000 - 1100 usec.

---%---

--- a/drivers/net/dsa/mt7530.c
+++ b/drivers/net/dsa/mt7530.c

@@ -2238,6 +2238,12 @@ mt7530_setup(struct dsa_switch *ds)
                 }
         }

+       /* Disable LEDs before reset to prevent the MT7530 sampling a
+        * potentially incorrect HT_XTAL_FSEL value.
+        */
+       mt7530_write(priv, MT7530_LED_EN, 0);
+       usleep_range(1000, 1100);
+
         /* Reset whole chip through gpio pin or memory-mapped registers for
          * different type of hardware
          */
---%---

I ran 20 tests per patch, applied exclusively. 40 tests in total.

      <-- ESW_P3_LED_0 Low Period before Reset Deassertion -->

   TEST    WITH INCREASED RESET DELAY    DISABLE LEDS BEFORE RESET
      #                        (usec)                       (usec)
-------------------------------------------------------------------
      1                           182                         4790
      2                           370                         3470
      3                           240                         4635
      4                          1475                         4850
      5                            70                         4775
      6                          2730                         4575
      7                          3180                         4565
      8                           265                         5650
      9                           270                         4690
     10                          1525                         4450
     11                          3210                         4735
     12                           120                         4690
     13                           185                         4625
     14                           305                         7020
     15                          2975                         4720
     16                           245                         4675
     17                           350                         4740
     18                            80                        17920
     19                           150                        17665
     20                           100                         4620

    Min                            70                         3470
    Max                          3210                        17920

   Mean                           270                         4720
    Avg                       923.421                     6161.579


Every test resulted in a 40MHz HT_XTAL_FSEL, but after seeing 70 usec
and 80 usec periods I wondered how many more tests it may take before
an 25MHz HT_XTAL_FSEL appears.

I was also surprised by the 17920 usec and 17665 usec periods listed
under the DISABLED LEDS BEFORE RESET column. Nothing unusual seemed
to be happening, at least as far as the kernel message output was
concerned.

What do you make of these results?