On Mon, Aug 24, 2026 at 10:17 PM Leon Hwang <[email protected]> wrote: > > On 25/8/26 04:04, Andrii Nakryiko wrote: > > On Sun, Aug 23, 2026 at 10:02 PM Leon Hwang <[email protected]> wrote: > [...] > >>> > >> It is to verify the mmap change in libbpf's bpf_map_mmap_resize() for > >> percpu data map, like those subtests in global_map_resize.c. > >> > > > > this didn't explain much, but I think what you are trying to say is > > that this won't work for PERCPU arrays with PAGE_SIZE=64KB, because > > PERCPU value is limited to 32KB today, is that right? > > > > Right. Limiting PAGE_SIZE=4KB is to avoid the > test_global_percpu_data__load() failure due to PERCPU value 32KB limit > when PAGE_SIZE is 64KB. > > > > > so maybe actually model that in the test itself, expect -E2BIG if > > PAGE_SIZE > 32KB or something, and otherwise check the resize? > > > How about skipping test_percpu_data_on_cpus() when > test_global_percpu_data__load() returns -E2BIG? Because > bpf_map__set_value_size() has already been verified before calling > test_global_percpu_data__load(). Afterward, it's unnecessary to check > PAGE_SIZE value. >
let's just add a comment about per-cpu limit of 32KB and why PAGE_SIZE of 64KB or more is not going to work, and skip the test ignoring load failures seems worse > Thanks, > Leon >
