Re: [PATCH net-next 14/15] net: Reference bpf_redirect_info via task_struct on PREEMPT_RT.

[Jesper Dangaard Brouer <hawk@kernel.org>]

On 07/05/2024 14.36, Sebastian Andrzej Siewior wrote:
> On 2024-05-06 16:09:47 [-0700], Alexei Starovoitov wrote:
>>>> On PREEMPT_RT the pointer to bpf_net_context is saved task's
>>>> task_struct. On non-PREEMPT_RT builds the pointer saved in a per-CPU
>>>> variable (which is always NODE-local memory). Using always the
>>>> bpf_net_context approach has the advantage that there is almost zero
>>>> differences between PREEMPT_RT and non-PREEMPT_RT builds.
>>>
>>> Did you ever manage to get any performance data to see if this has an
>>> impact?
>>>
>>> [...]
>>>
>>>> +static inline struct bpf_net_context *bpf_net_ctx_get(void)
>>>> +{
>>>> +     struct bpf_net_context *bpf_net_ctx = this_cpu_read(bpf_net_context);
>>>> +
>>>> +     WARN_ON_ONCE(!bpf_net_ctx);
>>>
>>> If we have this WARN...
>>>

When asking for change anyhow...

XDP redirect is an extreme fast-path.

Adding an WARN macro cause adding an 'ud2' instruction that cause CPU
instruction-cache to stop pre-fetching.

For this reason we in include/net/xdp.h have #define XDP_WARN and
function xdp_warn() that lives in net/core/xdp.c.
See how it is used in xdp_update_frame_from_buff().

Described in https://git.kernel.org/torvalds/c/34cc0b338a61
  - 34cc0b338a61 ("xdp: Xdp_frame add member frame_sz and handle in 
convert_to_xdp_frame")



>>>> +static inline struct bpf_redirect_info *bpf_net_ctx_get_ri(void)
>>>> +{
>>>> +     struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
>>>> +
>>>> +     if (!bpf_net_ctx)
>>>> +             return NULL;
>>>
>>> ... do we really need all the NULL checks?
>>
>> Indeed.
>> Let's drop all NULL checks, since they definitely add overhead.
>> I'd also remove ifdef CONFIG_PREEMPT_RT and converge on single implementation:
>> static inline struct bpf_net_context * bpf_net_ctx_get(void)
>> {
>>   return current->bpf_net_context;
>> }
> 
> Okay, let me do that then.

I need/want to echo Toke's request to benchmark these changes.

--Jesper