On Wed, May 15, 2024 at 6:43 AM Sebastian Andrzej Siewior
[off-list ref] wrote:

On 2024-05-14 13:54:43 [+0200], Toke Høiland-Jørgensen wrote:

quoted

quoted

--- a/include/linux/sched.h
+++ b/include/linux/sched.h

@@ -1504,6 +1505,8 @@ struct task_struct {
    /* Used for BPF run context */
    struct bpf_run_ctx              *bpf_ctx;
 #endif
+   /* Used by BPF for per-TASK xdp storage */
+   struct bpf_net_context          *bpf_net_context;

Okay, so if we are going the route of always putting this in 'current',
why not just embed the whole struct bpf_net_context inside task_struct,
instead of mucking about with the stack-allocated structures and
setting/clearing of pointers?

The whole struct bpf_net_context has 112 bytes. task_struct has 12352
bytes in my debug-config or 7296 bytes with defconfig on x86-64. Adding
it unconditionally would grow task_struct by ~1% but it would make
things way easier: The NULL case goes away, the assignment and cleanup
goes away, the INIT_LIST_HEAD can be moved to fork(). If the size
increase is not an issue then why not. Let me prepare…

I think 112 bytes or whatever the size of bpf_net_context is a bit
too much to consume in task_struct.
Yes, it's big, but there are systems with 1m threads. 112Mbyte is not
that small.
bpf_net_ctx_set/get are not in critical path and get_ri will be
inlined without any conditionals, so performance should be the same.