netdev - Re: [RFC PATCH bpf-next 04/13] xsk: introduce AF

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Wed, 6 May 2020 09:22:57 +0200
From:   Björn Töpel <bjorn.topel@...il.com>
To:     Maxim Mikityanskiy <maximmi@...lanox.com>
Cc:     "Karlsson, Magnus" <magnus.karlsson@...el.com>,
        Björn Töpel <bjorn.topel@...el.com>,
        Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        David Miller <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Jesper Dangaard Brouer <hawk@...nel.org>,
        John Fastabend <john.fastabend@...il.com>,
        Netdev <netdev@...r.kernel.org>, bpf <bpf@...r.kernel.org>,
        Jonathan Lemon <jonathan.lemon@...il.com>,
        "Fijalkowski, Maciej" <maciej.fijalkowski@...el.com>
Subject: Re: [RFC PATCH bpf-next 04/13] xsk: introduce AF_XDP buffer
 allocation API

On Tue, 5 May 2020 at 16:28, Maxim Mikityanskiy <maximmi@...lanox.com> wrote:
>
[...]
> >
> > -     if (((d->addr + d->len) & q->chunk_mask) != (d->addr & q->chunk_mask) ||
> > -         d->options) {
> > +static inline bool xskq_cons_is_valid_desc(struct xsk_queue *q,
> > +                                        struct xdp_desc *d,
> > +                                        struct xdp_umem *umem)
> > +{
> > +     if (!xp_validate_desc(umem->pool, d)) {
>
> I did some performance debugging and came to conclusion that this
> function call is the culprit of the TX speed degradation that I
> experience. I still don't know if it's the only reason or not, but I
> clearly see a degradation when xskq_cons_is_valid_desc is not fully
> inlined, but calls a function. E.g., I've put the code that handles the
> aligned mode into a separate function in a different file, and it caused
> the similar speed decrease.
>

Thanks for looking in to, and finding this!

I'll make sure the xp_validate_desc() call is inlined for the next
revision.

A note to myself: I need to check the performance for an LLVM build
with LTO enabled.


Cheers,
Björn