Skip to contentSkip to conversation

bpf: cpumap: report Rx queue index to xdp_rxq_info

PATCH In review Ujwal Kundur 4 participants 6 messages Jun 9, 2025 – Jun 10, 2025

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 bpf: cpumap: report Rx queue index to xdp_rxq_info

Refer to the Rx queue using a XDP frame's attached netdev and ascertain the queue index from it.

---
 kernel/bpf/cpumap.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
kernel/bpf/cpumap.c +2 -1
index 67e8a2fc1a99..8230292deac1 100644
@@ -34,6 +34,7 @@
 #include <linux/btf_ids.h>
 
 #include <linux/netdevice.h>
+#include <net/netdev_rx_queue.h>
 #include <net/gro.h>
 
 /* General idea: XDP packets getting XDP redirected to another CPU,
@@ -196,7 +197,7 @@ static int cpu_map_bpf_prog_run_xdp(struct bpf_cpu_map_entry *rcpu,
 
 		rxq.dev = xdpf->dev_rx;
 		rxq.mem.type = xdpf->mem_type;
-		/* TODO: report queue_index to xdp_rxq_info */
+		rxq.queue_index = get_netdev_rx_queue_index(xdpf->dev_rx->_rx);
 
 		xdp_convert_frame_to_buff(xdpf, &xdp);
 
Signature · 1 line
2.20.1
#1 lore raw
Quoted 21 lines · Ujwal Kundur

Refer to the Rx queue using a XDP frame's attached netdev and ascertain the queue index from it.

---
  kernel/bpf/cpumap.c | 3 ++-
  1 file changed, 2 insertions(+), 1 deletion(-)
kernel/bpf/cpumap.c +1 -0
index 67e8a2fc1a99..8230292deac1 100644
@@ -34,6 +34,7 @@
  #include <linux/btf_ids.h>
 
  #include <linux/netdevice.h>
+#include <net/netdev_rx_queue.h>
  #include <net/gro.h>
 
  /* General idea: XDP packets getting XDP redirected to another CPU,

kernel/bpf/cpumap.c static int cpu_map_bpf_prog_run_xdp(struct bpf_cpu_map_entry *rcpu, · line 200

Quoted 5 lines
kernel/bpf/cpumap.c +1 -1
@@ -196,7 +197,7 @@ static int cpu_map_bpf_prog_run_xdp(struct bpf_cpu_map_entry *rcpu,
 
  		rxq.dev = xdpf->dev_rx;
  		rxq.mem.type = xdpf->mem_type;
-		/* TODO: report queue_index to xdp_rxq_info */
+		rxq.queue_index = get_netdev_rx_queue_index(xdpf->dev_rx->_rx);

This looks wrong...
I think this will always return index 0

--Jesper

#2 lore raw
In reply to Ujwal Kundur
Quoted 21 lines · Ujwal Kundur

Refer to the Rx queue using a XDP frame's attached netdev and ascertain the queue index from it.

---
 kernel/bpf/cpumap.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
kernel/bpf/cpumap.c +1 -0
index 67e8a2fc1a99..8230292deac1 100644
@@ -34,6 +34,7 @@
 #include <linux/btf_ids.h>
 
 #include <linux/netdevice.h>
+#include <net/netdev_rx_queue.h>
 #include <net/gro.h>
 
 /* General idea: XDP packets getting XDP redirected to another CPU,

kernel/bpf/cpumap.c static int cpu_map_bpf_prog_run_xdp(struct bpf_cpu_map_entry *rcpu, · line 200

Quoted 5 lines
kernel/bpf/cpumap.c +1 -1
@@ -196,7 +197,7 @@ static int cpu_map_bpf_prog_run_xdp(struct bpf_cpu_map_entry *rcpu,
 
 		rxq.dev = xdpf->dev_rx;
 		rxq.mem.type = xdpf->mem_type;
-		/* TODO: report queue_index to xdp_rxq_info */
+		rxq.queue_index = get_netdev_rx_queue_index(xdpf->dev_rx->_rx);

This is pretty nonsensical; the definition of the function you're calling is this:

static inline unsigned int get_netdev_rx_queue_index(struct netdev_rx_queue *queue)

{
	struct net_device *dev = queue->dev;
	int index = queue - dev->_rx;

	BUG_ON(index >= dev->num_rx_queues);
	return index;
}

So passing dev->_rx to that function will always return 0; which is what the field is already initialised to...

-Toke

#3 lore raw
In reply to Ujwal Kundur

From: Ujwal Kundur <[email protected]>
Date: Mon, 9 Jun 2025 23:08:52 +0530

Quoted 21 lines

Refer to the Rx queue using a XDP frame's attached netdev and ascertain the queue index from it.

---
 kernel/bpf/cpumap.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
kernel/bpf/cpumap.c +1 -0
index 67e8a2fc1a99..8230292deac1 100644
@@ -34,6 +34,7 @@
 #include <linux/btf_ids.h>
 
 #include <linux/netdevice.h>
+#include <net/netdev_rx_queue.h>
 #include <net/gro.h>
 
 /* General idea: XDP packets getting XDP redirected to another CPU,

kernel/bpf/cpumap.c static int cpu_map_bpf_prog_run_xdp(struct bpf_cpu_map_entry *rcpu, · line 200

Quoted 5 lines
kernel/bpf/cpumap.c +1 -1
@@ -196,7 +197,7 @@ static int cpu_map_bpf_prog_run_xdp(struct bpf_cpu_map_entry *rcpu,
 
 		rxq.dev = xdpf->dev_rx;
 		rxq.mem.type = xdpf->mem_type;
-		/* TODO: report queue_index to xdp_rxq_info */
+		rxq.queue_index = get_netdev_rx_queue_index(xdpf->dev_rx->_rx);

I won't repeat what the folks above already told you.

I'll just add that you may want to take a look at Lorenzo's series[0]. Rx queue index is sorta HW hint, so it shouldn't be a problem to add the corresponding field to xdp_rx_meta.
Then, you can expand cpumap's code to try reading that HW meta if present.

kernel/bpf/cpumap.c static int cpu_map_bpf_prog_run_xdp(struct bpf_cpu_map_entry *rcpu, · line 202

Quoted 2 lines
kernel/bpf/cpumap.c +0 -0
 		xdp_convert_frame_to_buff(xdpf, &xdp);

[0]
https://lore.kernel.org/bpf/174897275458.1677018.15827867832940584671.stgit@firesoul

Thanks,
Olek

#4 lore raw
Quoted 5 lines

This looks wrong...
I think this will always return index 0

So passing dev->_rx to that function will always return 0; which is what the field is already initialised to...

I didn't realize that would always return 0, sorry I should've tried to understand that statement better.

Quoted 4 lines

I'll just add that you may want to take a look at Lorenzo's series[0]. Rx queue index is sorta HW hint, so it shouldn't be a problem to add the corresponding field to xdp_rx_meta.
Then, you can expand cpumap's code to try reading that HW meta if present.

Thank you! I also tried to work backwards to figure out how the queue_index would be used if present in xdp_rxq_info but that wasn't immediately apparent to me.
I'm keen on learning/contributing to the BPF part of the network stack and this seemed like a good first patch to take up -- I'll understand this better and try again.

Thanks for your time,
Ujwal

#5 lore raw
Quoted 14 lines · Ujwal Kundur
Quoted 5 lines

This looks wrong...
I think this will always return index 0

So passing dev->_rx to that function will always return 0; which is what the field is already initialised to...

I didn't realize that would always return 0, sorry I should've tried to understand that statement better.

Quoted 4 lines

I'll just add that you may want to take a look at Lorenzo's series[0]. Rx queue index is sorta HW hint, so it shouldn't be a problem to add the corresponding field to xdp_rx_meta.
Then, you can expand cpumap's code to try reading that HW meta if present.

Quoted 6 lines

Thank you! I also tried to work backwards to figure out how the queue_index would be used if present in xdp_rxq_info but that wasn't immediately apparent to me.
I'm keen on learning/contributing to the BPF part of the network stack and this seemed like a good first patch to take up -- I'll understand this better and try again.

Sounds good! Don't be discouraged by having to try again, that's perfectly normal :)

-Toke