diff mbox series

[v2,2/8] platform/mellanox: mlxreg-hotplug: Improve mechanism of ASIC health discovery

Message ID 1532644864-164459-3-git-send-email-vadimp@mellanox.com (mailing list archive)
State Accepted, archived
Delegated to: Darren Hart
Headers show
Series platform/x86: Mellanox: add new features | expand

Commit Message

Vadim Pasternak July 26, 2018, 10:40 p.m. UTC
Simplify the logic of ASIC health discovery.
ASIC device can indicate its health state as a good, booting or dormant.
During ASIC reset the device is dropped to dormant state and should get to
the stable good health state through the intermediate booting state.
The sequence for getting to the steady state health after reset is:
(dormant -> booting -> good)+.
Initial implementation assumes that ?good? within this sequence is always
repeated twice and device is getting steady state only after the second
?good?. This patch removes this dependency, since the second ?good? is
received because of the noise on line and can be ignored. Device reaches
steady state after the first ?good? is received.

Signed-off-by: Vadim Pasternak <vadimp@mellanox.com>
---
v1-v2:
 Comments pointed out by Darren:
 - Edit comment for ASIC steady state flow.
 - Make logic more clear.
 Fixes added by Vadim:
 - Ignore transition from good to good state. Just acknowledge the
   received event in such case.
 - Remove MLXREG_HOTPLUG_RST_CNTR, since it?s not necessary after the
   above fix.
---
 drivers/platform/mellanox/mlxreg-hotplug.c | 34 +++++++++++++++++++++++-------
 1 file changed, 26 insertions(+), 8 deletions(-)
diff mbox series

Patch

diff --git a/drivers/platform/mellanox/mlxreg-hotplug.c b/drivers/platform/mellanox/mlxreg-hotplug.c
index ac97aa0..f363e078 100644
--- a/drivers/platform/mellanox/mlxreg-hotplug.c
+++ b/drivers/platform/mellanox/mlxreg-hotplug.c
@@ -50,9 +50,8 @@ 
 #define MLXREG_HOTPLUG_MASK_OFF		2
 #define MLXREG_HOTPLUG_AGGR_MASK_OFF	1
 
-/* ASIC health parameters. */
-#define MLXREG_HOTPLUG_HEALTH_MASK	0x02
-#define MLXREG_HOTPLUG_RST_CNTR		3
+/* ASIC good health mask. */
+#define MLXREG_HOTPLUG_GOOD_HEALTH_MASK	0x02
 
 #define MLXREG_HOTPLUG_ATTRS_MAX	24
 #define MLXREG_HOTPLUG_NOT_ASSERT	3
@@ -325,21 +324,40 @@  mlxreg_hotplug_health_work_helper(struct mlxreg_hotplug_priv_data *priv,
 			goto out;
 
 		regval &= data->mask;
-		item->cache = regval;
-		if (regval == MLXREG_HOTPLUG_HEALTH_MASK) {
-			if ((data->health_cntr++ == MLXREG_HOTPLUG_RST_CNTR) ||
-			    !priv->after_probe) {
+
+		if (item->cache == regval)
+			goto ack_event;
+
+		/*
+		 * ASIC health indication is provided through two bits. Bits
+		 * value 0x2 indicates that ASIC reached the good health, value
+		 * 0x0 indicates ASIC the bad health or dormant state and value
+		 * 0x3 indicates the booting state. During ASIC reset it should
+		 * pass the following states: dormant -> booting -> good.
+		 */
+		if (regval == MLXREG_HOTPLUG_GOOD_HEALTH_MASK) {
+			if (!data->attached) {
+				/*
+				 * ASIC is in steady state. Connect associated
+				 * device, if configured.
+				 */
 				mlxreg_hotplug_device_create(priv, data);
 				data->attached = true;
 			}
 		} else {
 			if (data->attached) {
+				/*
+				 * ASIC health is failed after ASIC has been
+				 * in steady state. Disconnect associated
+				 * device, if it has been connected.
+				 */
 				mlxreg_hotplug_device_destroy(data);
 				data->attached = false;
 				data->health_cntr = 0;
 			}
 		}
-
+		item->cache = regval;
+ack_event:
 		/* Acknowledge event. */
 		ret = regmap_write(priv->regmap, data->reg +
 				   MLXREG_HOTPLUG_EVENT_OFF, 0);