Skip to contentSkip to conversation

ethdev: avoid blocking telemetry callback for link status

PATCH Applied Bruce Richardson 6 participants 8 messages Jan 14, 2021 – Jan 19, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ethdev: avoid blocking telemetry callback for link status

When querying the link status via telemetry interface, we don't want the client to have to wait for multiple seconds for a reply. Therefore use "rte_eth_link_get_nowait()" rather than "rte_eth_link_get()" in the telemetry callback.

Fixes c190daedb9b1 ("ethdev: add telemetry callbacks")
---
 lib/librte_ethdev/rte_ethdev.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_ethdev/rte_ethdev.c +1 -1
index 17ddacc78..1f4545fe0 100644
@@ -5692,7 +5692,7 @@ eth_dev_handle_port_link_status(const char *cmd __rte_unused,
 	if (!rte_eth_dev_is_valid_port(port_id))
 		return -1;
 
-	ret = rte_eth_link_get(port_id, &link);
+	ret = rte_eth_link_get_nowait(port_id, &link);
 	if (ret < 0)
 		return -1;
 
Signature · 1 line
2.27.0
Acked-by on v1 1/1
#1 lore raw

Hi Bruce,

Quoted 39 lines

-----Original Message-----
From: Richardson, Bruce <[email protected]>
Sent: Thursday 14 January 2021 12:18
To: [email protected]
Cc: Richardson, Bruce <[email protected]>; [email protected]; Thomas Monjalon <[email protected]>; Yigit, Ferruh <[email protected]>; Andrew Rybchenko <[email protected]>; Power, Ciara <[email protected]>; Wiles, Keith <[email protected]>
Subject: [PATCH] ethdev: avoid blocking telemetry callback for link status

When querying the link status via telemetry interface, we don't want the client to have to wait for multiple seconds for a reply. Therefore use "rte_eth_link_get_nowait()" rather than "rte_eth_link_get()" in the telemetry callback.

Fixes c190daedb9b1 ("ethdev: add telemetry callbacks")

---
lib/librte_ethdev/rte_ethdev.c | 2 +-

1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_ethdev/rte_ethdev.c +1 -2
index 17ddacc78..1f4545fe0 100644
@@ -5692,7 +5692,7 @@ eth_dev_handle_port_link_status(const char *cmd
 __rte_unused,
 	if (!rte_eth_dev_is_valid_port(port_id))
 		return -1;
 
-	ret = rte_eth_link_get(port_id, &link);
+	ret = rte_eth_link_get_nowait(port_id, &link);
 	if (ret < 0)
 		return -1;
 
--
Quoted 1 line

2.27.0

This change looks good to me.

#2 lore raw
Quoted 18 lines
Quoted 13 lines

When querying the link status via telemetry interface, we don't want the client to have to wait for multiple seconds for a reply. Therefore use "rte_eth_link_get_nowait()" rather than "rte_eth_link_get()" in the telemetry callback.

Fixes c190daedb9b1 ("ethdev: add telemetry callbacks")

---

lib/librte_ethdev/rte_ethdev.c eth_dev_handle_port_link_status(const char *cmd · line 5692

Quoted 2 lines
lib/librte_ethdev/rte_ethdev.c +1 -1
@@ -5692,7 +5692,7 @@ eth_dev_handle_port_link_status(const char *cmd
-	ret = rte_eth_link_get(port_id, &link);
+	ret = rte_eth_link_get_nowait(port_id, &link);

This change looks good to me.

Quoted 1 line
Applied
#3 lore raw
Quoted 18 lines · Thomas Monjalon
Quoted 18 lines
Quoted 13 lines

When querying the link status via telemetry interface, we don't want the client to have to wait for multiple seconds for a reply. Therefore use "rte_eth_link_get_nowait()" rather than "rte_eth_link_get()" in the telemetry callback.

Fixes c190daedb9b1 ("ethdev: add telemetry callbacks")

---

lib/librte_ethdev/rte_ethdev.c eth_dev_handle_port_link_status(const char *cmd · line 5692

Quoted 2 lines
lib/librte_ethdev/rte_ethdev.c +1 -1
@@ -5692,7 +5692,7 @@ eth_dev_handle_port_link_status(const char *cmd
-	ret = rte_eth_link_get(port_id, &link);
+	ret = rte_eth_link_get_nowait(port_id, &link);

This change looks good to me.

Quoted 4 lines
Quoted 1 line

Applied to dpdk-next-net/main, thanks.

#4 lore raw
In reply to Bruce Richardson

Hi, Bruce and all,
Do you know the difference between "rte_eth_link_get" and
"rte_eth_link_get_nowait"? I know they call funciton "link_update" with differenct parameter "wait_to_complete"(set 1 means wait, set 0
means not wait). But how to define the "wait" time, and why it shoud wait?
On the further, What are the application scenarios of the two
APIs?

Look forward to your reply, thanks.

Quoted 27 lines · Bruce Richardson

When querying the link status via telemetry interface, we don't want the client to have to wait for multiple seconds for a reply. Therefore use "rte_eth_link_get_nowait()" rather than "rte_eth_link_get()" in the telemetry callback.

Fixes c190daedb9b1 ("ethdev: add telemetry callbacks")
---
  lib/librte_ethdev/rte_ethdev.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
lib/librte_ethdev/rte_ethdev.c +1 -1
index 17ddacc78..1f4545fe0 100644
@@ -5692,7 +5692,7 @@ eth_dev_handle_port_link_status(const char *cmd __rte_unused,
  	if (!rte_eth_dev_is_valid_port(port_id))
  		return -1;
 
-	ret = rte_eth_link_get(port_id, &link);
+	ret = rte_eth_link_get_nowait(port_id, &link);
  	if (ret < 0)
  		return -1;
 
#5 lore raw
Quoted 8 lines · On Tue, 19 Jan 2021 09:06:48 +0800 "Min Hu (Connor)" <[email protected]> wrote:

Hi, Bruce and all,
Do you know the difference between "rte_eth_link_get" and
"rte_eth_link_get_nowait"? I know they call funciton "link_update" with differenct parameter "wait_to_complete"(set 1 means wait, set 0
means not wait). But how to define the "wait" time, and why it shoud wait?
On the further, What are the application scenarios of the two
APIs?

The default behavior of rte_eth_link_get (in some drivers) is to wait for link to come up. Many drivers don't do this. It seems mostly the Intel ones that do.

#6 lore raw

Thank Stephen,
but in which the scenarios, it should wait link to up, in which scenarios, it should not ?
By the way, how to define the "wait" time value ?

Quoted 17 lines · Stephen Hemminger
Quoted 8 lines · On Tue, 19 Jan 2021 09:06:48 +0800 "Min Hu (Connor)" <[email protected]> wrote:

Hi, Bruce and all,
Do you know the difference between "rte_eth_link_get" and
"rte_eth_link_get_nowait"? I know they call funciton "link_update" with differenct parameter "wait_to_complete"(set 1 means wait, set 0
means not wait). But how to define the "wait" time, and why it shoud wait?
On the further, What are the application scenarios of the two
APIs?

The default behavior of rte_eth_link_get (in some drivers) is to wait for link to come up. Many drivers don't do this. It seems mostly the Intel ones that do.
.

#7 lore raw
Quoted 5 lines · Min Hu (Connor)

Thank Stephen,
but in which the scenarios, it should wait link to up, in which scenarios, it should not ?
By the way, how to define the "wait" time value ?

I believe the documentation somewhere refers to a wait time of up to 9 seconds, but I don't think it's terribly well defined. Some Intel NICs will stall for a few seconds if the link is not yet stable or if the link is down. With the link up, I don't think any NIC fails to return immediately.
However, if delays in the app are something you want to avoid, the _nowait varient is the one to call to be sure.

/Bruce

Quoted 20 lines · Stephen Hemminger
Quoted 17 lines
Quoted 8 lines · On Tue, 19 Jan 2021 09:06:48 +0800 "Min Hu (Connor)" <[email protected]> wrote:

Hi, Bruce and all,
Do you know the difference between "rte_eth_link_get" and
"rte_eth_link_get_nowait"? I know they call funciton "link_update" with differenct parameter "wait_to_complete"(set 1 means wait, set 0
means not wait). But how to define the "wait" time, and why it shoud wait?
On the further, What are the application scenarios of the two
APIs?

The default behavior of rte_eth_link_get (in some drivers) is to wait for link to come up. Many drivers don't do this. It seems mostly the Intel ones that do.
.