[v2] bpf/cpumap.c: Remove unnecessary TODO comment

[PATCH bpf-next v2] bpf/cpumap.c: Remove unnecessary TODO comment

Posted by Mehdi Ben Hadj Khelifa 3 months, 2 weeks ago

After discussion with bpf maintainers[1], queue_index could
be propagated to the remote XDP program by the xdp_md struct[2]
which makes this todo a misguide for future effort.

[1]:https://lore.kernel.org/all/87y0q23j2w.fsf@cloudflare.com/
[2]:https://docs.ebpf.io/linux/helper-function/bpf_xdp_adjust_meta/

Signed-off-by: Mehdi Ben Hadj Khelifa <mehdi.benhadjkhelifa@gmail.com>
---
Changelog:

Changes from v1:

-Added a comment to clarify that RX queue_index is lost after the frame
redirection.

Link:https://lore.kernel.org/bpf/d9819687-5b0d-4bfa-9aec-aef71b847383@gmail.com/T/#mcb6a0315f174d02db3c9bc4fa556cc939c87a706
 kernel/bpf/cpumap.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/kernel/bpf/cpumap.c b/kernel/bpf/cpumap.c
index 703e5df1f4ef..6856a4a67840 100644
--- a/kernel/bpf/cpumap.c
+++ b/kernel/bpf/cpumap.c
@@ -195,7 +195,10 @@ 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 */
+		/* The NIC RX queue_index is lost after the frame redirection
+		 * but in case of need, it can be passed as a custom XDP
+		 * metadata via xdp_md struct to the remote XDP program
+		 */
 
 		xdp_convert_frame_to_buff(xdpf, &xdp);
 
-- 
2.51.1.dirty

Re: [PATCH bpf-next v2] bpf/cpumap.c: Remove unnecessary TODO comment

Posted by Jesper Dangaard Brouer 3 months, 2 weeks ago


On 20/10/2025 19.02, Mehdi Ben Hadj Khelifa wrote:
> After discussion with bpf maintainers[1], queue_index could
> be propagated to the remote XDP program by the xdp_md struct[2]
> which makes this todo a misguide for future effort.
> 
> [1]:https://lore.kernel.org/all/87y0q23j2w.fsf@cloudflare.com/
> [2]:https://docs.ebpf.io/linux/helper-function/bpf_xdp_adjust_meta/
> 
> Signed-off-by: Mehdi Ben Hadj Khelifa <mehdi.benhadjkhelifa@gmail.com>
> ---
> Changelog:
> 
> Changes from v1:
> 
> -Added a comment to clarify that RX queue_index is lost after the frame
> redirection.
> 
> Link:https://lore.kernel.org/bpf/d9819687-5b0d-4bfa-9aec-aef71b847383@gmail.com/T/#mcb6a0315f174d02db3c9bc4fa556cc939c87a706
>   kernel/bpf/cpumap.c | 5 ++++-
>   1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/bpf/cpumap.c b/kernel/bpf/cpumap.c
> index 703e5df1f4ef..6856a4a67840 100644
> --- a/kernel/bpf/cpumap.c
> +++ b/kernel/bpf/cpumap.c
> @@ -195,7 +195,10 @@ 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 */
> +		/* The NIC RX queue_index is lost after the frame redirection
> +		 * but in case of need, it can be passed as a custom XDP
> +		 * metadata via xdp_md struct to the remote XDP program

Argh, saying XDP metadata is accessed via the xdp_md struct is just wrong.

Nacked-by: Jesper Dangaard Brouer <hawk@kernel.org>

> +		 */
>   
>   		xdp_convert_frame_to_buff(xdpf, &xdp);
>

Re: [PATCH bpf-next v2] bpf/cpumap.c: Remove unnecessary TODO comment

Posted by Mehdi Ben Hadj Khelifa 3 months, 2 weeks ago

On 10/21/25 10:37 AM, Jesper Dangaard Brouer wrote:
> 
> 
> On 20/10/2025 19.02, Mehdi Ben Hadj Khelifa wrote:
>> After discussion with bpf maintainers[1], queue_index could
>> be propagated to the remote XDP program by the xdp_md struct[2]
>> which makes this todo a misguide for future effort.
>>
>> [1]:https://lore.kernel.org/all/87y0q23j2w.fsf@cloudflare.com/
>> [2]:https://docs.ebpf.io/linux/helper-function/bpf_xdp_adjust_meta/
>>
>> Signed-off-by: Mehdi Ben Hadj Khelifa <mehdi.benhadjkhelifa@gmail.com>
>> ---
>> Changelog:
>>
>> Changes from v1:
>>
>> -Added a comment to clarify that RX queue_index is lost after the frame
>> redirection.
>>
>> Link:https://lore.kernel.org/bpf/d9819687-5b0d-4bfa-9aec- 
>> aef71b847383@gmail.com/T/#mcb6a0315f174d02db3c9bc4fa556cc939c87a706
>>   kernel/bpf/cpumap.c | 5 ++++-
>>   1 file changed, 4 insertions(+), 1 deletion(-)
>>
>> diff --git a/kernel/bpf/cpumap.c b/kernel/bpf/cpumap.c
>> index 703e5df1f4ef..6856a4a67840 100644
>> --- a/kernel/bpf/cpumap.c
>> +++ b/kernel/bpf/cpumap.c
>> @@ -195,7 +195,10 @@ 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 */
>> +        /* The NIC RX queue_index is lost after the frame redirection
>> +         * but in case of need, it can be passed as a custom XDP
>> +         * metadata via xdp_md struct to the remote XDP program
> 
> Argh, saying XDP metadata is accessed via the xdp_md struct is just wrong.
> 
Ack, I didn't clarify that XDP metadata should be propagated via the 
bpf_xdp_adjust_meta like mentionned in the link[2]... Maybe I was 
thinking more in the technical side that xdp_md->data_meta would hold 
the value internally... I will send a v3 with appropriate changes.
Thanks for the review.

Best Regards,
Mehdi Ben Hadj Khelifa
> Nacked-by: Jesper Dangaard Brouer <hawk@kernel.org>
> 
>> +         */
>>           xdp_convert_frame_to_buff(xdpf, &xdp);
>