Skip to contentSkip to conversation

net: xsk: update tx queue consumer immediately after transmission

PATCH Superseded Jason Xing 2 participants 3 messages Jun 23, 2025 – Jun 24, 2025

Subsystems (4)

v2 of 6 Superseded Latest is v6 →

lore
Share & tools
lore

Jason Xing

author
#0 lore raw

PATCH v2 1/1 net: xsk: update tx queue consumer immediately after transmission

From: Jason Xing <[email protected]>

For afxdp, the return value of sendto() syscall doesn't reflect how many descs handled in the kernel. One of use cases is that when user-space application tries to know the number of transmitted skbs and then decides if it continues to send, say, is it stopped due to max tx budget?

The following formular can be used after sending to learn how many skbs/descs the kernel takes care of:

tx_queue.consumers_before - tx_queue.consumers_after

Prior to the current patch, in non-zc mode, the consumer of tx queue is not immediately updated at the end of each sendto syscall when error occurs, which leads to the consumer value out-of-dated from the perspective of user space. So this patch requires store operation to pass the cached value to the shared value to handle the problem.

More than those explicit errors appearing in the while() loop in __xsk_generic_xmit(), there are a few possible error cases that might be neglected in the following call trace:
__xsk_generic_xmit()
xskq_cons_peek_desc()
xskq_cons_read_desc()
xskq_cons_is_valid_desc()
It will also cause the premature exit in the while() loop even if not all the descs are consumed.

Based on the above analysis, using 'cached_prod != cached_cons' could cover all the possible cases because it represents there are remaining descs that are not handled and cached_cons are not updated to the global state of consumer at this time.

---
V2

1. filter out those good cases because only those that return error need updates.
Side note:
1. in non-batched zero copy mode, at the end of every caller of xsk_tx_peek_desc(), there is always a xsk_tx_release() function that used to update the local consumer to the global state of consumer. So for the zero copy mode, no need to change at all.
2. Actually I have no strong preference between v1 (see the above link) and v2 because smp_store_release() shouldn't cause side effect. Considering the exactitude of writing code, v2 is a more preferable one.

---
 net/xdp/xsk.c | 3 +++
 1 file changed, 3 insertions(+)
net/xdp/xsk.c +3 -0
index 5542675dffa9..b9223a2a6ada 100644
@@ -856,6 +856,9 @@ static int __xsk_generic_xmit(struct sock *sk)
 	}
 
 out:
+	if (xs->tx->cached_prod != xs->tx->cached_cons)
+		__xskq_cons_release(xs->tx);
+
 	if (sent_frame)
 		if (xsk_tx_writeable(xs))
 			sk->sk_write_space(sk);
Signature · 1 line
2.43.5
#1 lore raw
Quoted 57 lines · Jason Xing

From: Jason Xing <[email protected]>

For afxdp, the return value of sendto() syscall doesn't reflect how many descs handled in the kernel. One of use cases is that when user-space application tries to know the number of transmitted skbs and then decides if it continues to send, say, is it stopped due to max tx budget?

The following formular can be used after sending to learn how many skbs/descs the kernel takes care of:

tx_queue.consumers_before - tx_queue.consumers_after

Prior to the current patch, in non-zc mode, the consumer of tx queue is not immediately updated at the end of each sendto syscall when error occurs, which leads to the consumer value out-of-dated from the perspective of user space. So this patch requires store operation to pass the cached value to the shared value to handle the problem.

More than those explicit errors appearing in the while() loop in __xsk_generic_xmit(), there are a few possible error cases that might be neglected in the following call trace:
__xsk_generic_xmit()
xskq_cons_peek_desc()
xskq_cons_read_desc()
xskq_cons_is_valid_desc()
It will also cause the premature exit in the while() loop even if not all the descs are consumed.

Based on the above analysis, using 'cached_prod != cached_cons' could cover all the possible cases because it represents there are remaining descs that are not handled and cached_cons are not updated to the global state of consumer at this time.

---
V2

1. filter out those good cases because only those that return error need updates.
Side note:
1. in non-batched zero copy mode, at the end of every caller of xsk_tx_peek_desc(), there is always a xsk_tx_release() function that used to update the local consumer to the global state of consumer. So for the zero copy mode, no need to change at all.
2. Actually I have no strong preference between v1 (see the above link) and v2 because smp_store_release() shouldn't cause side effect. Considering the exactitude of writing code, v2 is a more preferable one.

---
 net/xdp/xsk.c | 3 +++
 1 file changed, 3 insertions(+)
net/xdp/xsk.c +0 -0
index 5542675dffa9..b9223a2a6ada 100644

net/xdp/xsk.c static int __xsk_generic_xmit(struct sock *sk) · line 859

Quoted 4 lines
net/xdp/xsk.c +1 -0
@@ -856,6 +856,9 @@ static int __xsk_generic_xmit(struct sock *sk)
 	}
 
 out:
+	if (xs->tx->cached_prod != xs->tx->cached_cons)

Can we use xskq_has_descs() here instead?

And still would be nice to verify this with a test. Should not be much work, most of the things (setup/etc) are already there, so you won't have to write everything from scratch...

author
#2 lore raw
Quoted 61 lines · Stanislav Fomichev
Quoted 59 lines · Jason Xing

From: Jason Xing <[email protected]>

For afxdp, the return value of sendto() syscall doesn't reflect how many descs handled in the kernel. One of use cases is that when user-space application tries to know the number of transmitted skbs and then decides if it continues to send, say, is it stopped due to max tx budget?

The following formular can be used after sending to learn how many skbs/descs the kernel takes care of:

tx_queue.consumers_before - tx_queue.consumers_after

Prior to the current patch, in non-zc mode, the consumer of tx queue is not immediately updated at the end of each sendto syscall when error occurs, which leads to the consumer value out-of-dated from the perspective of user space. So this patch requires store operation to pass the cached value to the shared value to handle the problem.

More than those explicit errors appearing in the while() loop in __xsk_generic_xmit(), there are a few possible error cases that might be neglected in the following call trace:
__xsk_generic_xmit()
xskq_cons_peek_desc()
xskq_cons_read_desc()
xskq_cons_is_valid_desc()
It will also cause the premature exit in the while() loop even if not all the descs are consumed.

Based on the above analysis, using 'cached_prod != cached_cons' could cover all the possible cases because it represents there are remaining descs that are not handled and cached_cons are not updated to the global state of consumer at this time.

---
V2

1. filter out those good cases because only those that return error need updates.
Side note:
1. in non-batched zero copy mode, at the end of every caller of xsk_tx_peek_desc(), there is always a xsk_tx_release() function that used to update the local consumer to the global state of consumer. So for the zero copy mode, no need to change at all.
2. Actually I have no strong preference between v1 (see the above link) and v2 because smp_store_release() shouldn't cause side effect. Considering the exactitude of writing code, v2 is a more preferable one.

---
 net/xdp/xsk.c | 3 +++
 1 file changed, 3 insertions(+)
net/xdp/xsk.c +0 -0
index 5542675dffa9..b9223a2a6ada 100644
@@ -856,6 +856,9 @@ static int __xsk_generic_xmit(struct sock *sk)
      }
 
Quoted 4 lines

net/xdp/xsk.c static int __xsk_generic_xmit(struct sock *sk) · line 859

Quoted 2 lines
net/xdp/xsk.c +1 -0
 out:
+     if (xs->tx->cached_prod != xs->tx->cached_cons)

Can we use xskq_has_descs() here instead?

Sure, thanks.

Quoted 4 lines

And still would be nice to verify this with a test. Should not be much work, most of the things (setup/etc) are already there, so you won't have to write everything from scratch...

Ah, finally I managed to make it work but a bit ugly... Let me try if I can find a
better way to write the test patch.

Thanks,
Jason