[RFC PATCH net-next] net: dsa: mt7530: support setting ageing time

[RFC PATCH net-next] net: dsa: mt7530: support setting ageing time

[DENG Qingfang]

MT7530 has a global address age control register, so use it to set
ageing time.

The applied timer is (AGE_CNT + 1) * (AGE_UNIT + 1) seconds

Signed-off-by: DENG Qingfang <dqfext@gmail.com>
---
 drivers/net/dsa/mt7530.c | 41 ++++++++++++++++++++++++++++++++++++++++
 drivers/net/dsa/mt7530.h | 13 +++++++++++++
 2 files changed, 54 insertions(+)

RFC:
1. What is the expected behaviour if the timer is too big or too small?
   - return -ERANGE or -EINVAL;
     or
   - if it is too big, apply the maximum value; if it is too small,
     disable learning;

2. Is there a better algorithm to find the closest pair?

diff --git a/drivers/net/dsa/mt7530.c b/drivers/net/dsa/mt7530.c
index 6408402a44f5..99bf8fed6536 100644
--- a/drivers/net/dsa/mt7530.c
+++ b/drivers/net/dsa/mt7530.c
@@ -870,6 +870,46 @@ mt7530_get_sset_count(struct dsa_switch *ds, int port, int sset)
 	return ARRAY_SIZE(mt7530_mib);
 }
 
+static int
+mt7530_set_ageing_time(struct dsa_switch *ds, unsigned int msecs)
+{
+	struct mt7530_priv *priv = ds->priv;
+	unsigned int secs = msecs / 1000;
+	unsigned int tmp_age_count;
+	unsigned int error = -1;
+	unsigned int age_count;
+	unsigned int age_unit;
+
+	/* Applied timer is (AGE_CNT + 1) * (AGE_UNIT + 1) seconds */
+	if (secs < 1 || secs > (AGE_CNT_MAX + 1) * (AGE_UNIT_MAX + 1))
+		return -ERANGE;
+
+	/* iterate through all possible age_count to find the closest pair */
+	for (tmp_age_count = 0; tmp_age_count <= AGE_CNT_MAX; ++tmp_age_count) {
+		unsigned int tmp_age_unit = secs / (tmp_age_count + 1) - 1;
+
+		if (tmp_age_unit <= AGE_UNIT_MAX) {
+			unsigned int tmp_error = secs -
+				(tmp_age_count + 1) * (tmp_age_unit + 1);
+
+			/* found a closer pair */
+			if (error > tmp_error) {
+				error = tmp_error;
+				age_count = tmp_age_count;
+				age_unit = tmp_age_unit;
+			}
+
+			/* found the exact match, so break the loop */
+			if (!error)
+				break;
+		}
+	}
+
+	mt7530_write(priv, MT7530_AAC, AGE_CNT(age_count) | AGE_UNIT(age_unit));
+
+	return 0;
+}
+
 static void mt7530_setup_port5(struct dsa_switch *ds, phy_interface_t interface)
 {
 	struct mt7530_priv *priv = ds->priv;
@@ -2564,6 +2604,7 @@ static const struct dsa_switch_ops mt7530_switch_ops = {
 	.phy_write		= mt753x_phy_write,
 	.get_ethtool_stats	= mt7530_get_ethtool_stats,
 	.get_sset_count		= mt7530_get_sset_count,
+	.set_ageing_time	= mt7530_set_ageing_time,
 	.port_enable		= mt7530_port_enable,
 	.port_disable		= mt7530_port_disable,
 	.port_change_mtu	= mt7530_port_change_mtu,
diff --git a/drivers/net/dsa/mt7530.h b/drivers/net/dsa/mt7530.h
index ee3523a7537e..32d8969b3ace 100644
--- a/drivers/net/dsa/mt7530.h
+++ b/drivers/net/dsa/mt7530.h
@@ -161,6 +161,19 @@ enum mt7530_vlan_egress_attr {
 	MT7530_VLAN_EGRESS_STACK = 3,
 };
 
+/* Register for address age control */
+#define MT7530_AAC			0xa0
+/* Disable ageing */
+#define  AGE_DIS			BIT(20)
+/* Age count */
+#define  AGE_CNT_MASK			GENMASK(19, 12)
+#define  AGE_CNT_MAX			0xff
+#define  AGE_CNT(x)			(AGE_CNT_MASK & ((x) << 12))
+/* Age unit */
+#define  AGE_UNIT_MASK			GENMASK(11, 0)
+#define  AGE_UNIT_MAX			0xfff
+#define  AGE_UNIT(x)			(AGE_UNIT_MASK & (x))
+
 /* Register for port STP state control */
 #define MT7530_SSP_P(x)			(0x2000 + ((x) * 0x100))
 #define  FID_PST(x)			((x) & 0x3)
-- 
2.25.1

Re: [RFC PATCH net-next] net: dsa: mt7530: support setting ageing time

[Andrew Lunn]

On Thu, Nov 19, 2020 at 02:40:20PM +0800, DENG Qingfang wrote:
> MT7530 has a global address age control register, so use it to set
> ageing time.
> 
> The applied timer is (AGE_CNT + 1) * (AGE_UNIT + 1) seconds
> 
> Signed-off-by: DENG Qingfang <dqfext@gmail.com>
> ---
>  drivers/net/dsa/mt7530.c | 41 ++++++++++++++++++++++++++++++++++++++++
>  drivers/net/dsa/mt7530.h | 13 +++++++++++++
>  2 files changed, 54 insertions(+)
> 
> RFC:
> 1. What is the expected behaviour if the timer is too big or too small?
>    - return -ERANGE or -EINVAL;

ERANGE is good. 

>      or
>    - if it is too big, apply the maximum value; if it is too small,
>      disable learning;
> 
> 2. Is there a better algorithm to find the closest pair?

The bridge code will default to 300 seconds. And after a topology
change, it sets it to 2 * the forwarding delay, which defaults to 15
seconds. So maybe you can look for these two values, and use
pre-computed values?

You still need to handle other values, the user can configure these.

	     Andrew

Re: [RFC PATCH net-next] net: dsa: mt7530: support setting ageing time

[DENG Qingfang]

On Fri, Nov 20, 2020 at 10:26 AM Andrew Lunn <andrew@lunn.ch> wrote:
> The bridge code will default to 300 seconds. And after a topology
> change, it sets it to 2 * the forwarding delay, which defaults to 15
> seconds. So maybe you can look for these two values, and use
> pre-computed values?

15 and 300 are not larger than 4096 (AGE_UNIT_MAX + 1) so the exact
match can always be found in the first iteration:

age_count = 0, age_unit = 14, secs = (0 + 1) * (14 + 1) = 15
age_count = 0, age_unit = 299, secs = (0 + 1) * (299 + 1) = 300

>
> You still need to handle other values, the user can configure these.
>
>              Andrew