On Wed, Dec 18, 2019 at 11:11 AM Yonghong Song <yhs@xxxxxx> wrote:
>
>
>
> On 12/18/19 6:23 AM, Paul Chaignon wrote:
> > Currently, userspace programs have to update the values of all CPUs at
> > once when updating per-cpu maps. This limitation prevents the update of
> > a single CPU's value without the risk of missing concurrent updates on
> > other CPU's values.
> >
> > This patch allows userspace to update the value of a specific CPU in
> > per-cpu maps. The CPU whose value should be updated is encoded in the
> > 32 upper-bits of the flags argument, as follows. The new BPF_CPU flag
> > can be combined with existing flags.
> >
> > bpf_map_update_elem(..., cpuid << 32 | BPF_CPU)
>
> Some additional comments beyond Alexei's one.
>
> >
> > Signed-off-by: Paul Chaignon <paul.chaignon@xxxxxxxxxx>
> > ---
> > include/uapi/linux/bpf.h | 4 +++
> > kernel/bpf/arraymap.c | 19 ++++++++-----
> > kernel/bpf/hashtab.c | 49 ++++++++++++++++++++--------------
> > kernel/bpf/local_storage.c | 16 +++++++----
> > kernel/bpf/syscall.c | 17 +++++++++---
> > tools/include/uapi/linux/bpf.h | 4 +++
> > 6 files changed, 74 insertions(+), 35 deletions(-)
> >
> > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > index dbbcf0b02970..2efb17d2c77a 100644
> > --- a/include/uapi/linux/bpf.h
> > +++ b/include/uapi/linux/bpf.h
> > @@ -316,6 +316,10 @@ enum bpf_attach_type {
> > #define BPF_NOEXIST 1 /* create new element if it didn't exist */
> > #define BPF_EXIST 2 /* update existing element */
> > #define BPF_F_LOCK 4 /* spin_lock-ed map_lookup/map_update */
> > +#define BPF_CPU 8 /* single-cpu update for per-cpu maps */
> > +
> > +/* CPU mask for single-cpu updates */
> > +#define BPF_CPU_MASK 0xFFFFFFFF00000000ULL
> BPF_F_CPU_MASK?
Maybe even define this as a function-like macro:
#define BPF_F_CPU_NR(cpu) ((cpu) << 32)
so that it can be easily combined with other flags in code like so:
bpf_map_update_element(...., BPF_F_LOCK | BPF_F_CPU_NR(10))
BPF_F_CPU_NR can automatically set BPF_F_CPU flag as well, btw.
[...]
