linux-kernel - Re: [PATCH] af_packet: TPACKET

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 for Android: free password hash cracker in your pocket
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CA+FuTSe1WXLGKd2zNLmQiKTZeNN64R-vGJTNMuVD_4VA8AN5Fg@mail.gmail.com>
Date:   Wed, 15 Jul 2020 18:35:00 -0400
From:   Willem de Bruijn <willemdebruijn.kernel@...il.com>
To:     Jakub Kicinski <kuba@...nel.org>
Cc:     John Ogness <john.ogness@...utronix.de>,
        "David S . Miller" <davem@...emloft.net>,
        Sebastian Andrzej Siewior <bigeasy@...utronix.de>,
        Thomas Gleixner <tglx@...utronix.de>,
        Network Development <netdev@...r.kernel.org>,
        linux-kernel <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] af_packet: TPACKET_V3: replace busy-wait loop

On Wed, Jul 15, 2020 at 4:21 PM Jakub Kicinski <kuba@...nel.org> wrote:
>
> On Tue,  7 Jul 2020 17:28:04 +0206 John Ogness wrote:
> > A busy-wait loop is used to implement waiting for bits to be copied
> > from the skb to the kernel buffer before retiring a block. This is
> > a problem on PREEMPT_RT because the copying task could be preempted
> > by the busy-waiting task and thus live lock in the busy-wait loop.
> >
> > Replace the busy-wait logic with an rwlock_t. This provides lockdep
> > coverage and makes the code RT ready.
> >
> > Signed-off-by: John Ogness <john.ogness@...utronix.de>
>
> Is taking a lock and immediately releasing it better than a completion?
> Seems like the lock is guaranteed to dirty a cache line, which would
> otherwise be avoided here.
>
> Willem, would you be able to take a look as well? Is this path
> performance sensitive in real life?

No objections from me.

I guess this resolves the issue on preempt_rt, because the spinlocks act as
mutexes. It will still spin on write_lock otherwise, no huge difference from
existing logic.





>
> > diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
> > index 29bd405adbbd..dd1eec2dd6ef 100644
> > --- a/net/packet/af_packet.c
> > +++ b/net/packet/af_packet.c
> > @@ -593,6 +593,7 @@ static void init_prb_bdqc(struct packet_sock *po,
> >                                               req_u->req3.tp_block_size);
> >       p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
> >       p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
> > +     rwlock_init(&p1->blk_fill_in_prog_lock);
> >
> >       p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
> >       prb_init_ft_ops(p1, req_u);
> > @@ -659,10 +660,9 @@ static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
> >        *
> >        */
> >       if (BLOCK_NUM_PKTS(pbd)) {
> > -             while (atomic_read(&pkc->blk_fill_in_prog)) {
> > -                     /* Waiting for skb_copy_bits to finish... */
> > -                     cpu_relax();
> > -             }
> > +             /* Waiting for skb_copy_bits to finish... */
> > +             write_lock(&pkc->blk_fill_in_prog_lock);
> > +             write_unlock(&pkc->blk_fill_in_prog_lock);
> >       }
> >
> >       if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
> > @@ -921,10 +921,9 @@ static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
> >                * the timer-handler already handled this case.
> >                */
> >               if (!(status & TP_STATUS_BLK_TMO)) {
> > -                     while (atomic_read(&pkc->blk_fill_in_prog)) {
> > -                             /* Waiting for skb_copy_bits to finish... */
> > -                             cpu_relax();
> > -                     }
> > +                     /* Waiting for skb_copy_bits to finish... */
> > +                     write_lock(&pkc->blk_fill_in_prog_lock);
> > +                     write_unlock(&pkc->blk_fill_in_prog_lock);
> >               }
> >               prb_close_block(pkc, pbd, po, status);
> >               return;
> > @@ -944,7 +943,8 @@ static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
> >  static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
> >  {
> >       struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
> > -     atomic_dec(&pkc->blk_fill_in_prog);
> > +
> > +     read_unlock(&pkc->blk_fill_in_prog_lock);
> >  }
> >
> >  static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
> > @@ -998,7 +998,7 @@ static void prb_fill_curr_block(char *curr,
> >       pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
> >       BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
> >       BLOCK_NUM_PKTS(pbd) += 1;
> > -     atomic_inc(&pkc->blk_fill_in_prog);
> > +     read_lock(&pkc->blk_fill_in_prog_lock);
> >       prb_run_all_ft_ops(pkc, ppd);
> >  }
> >
> > diff --git a/net/packet/internal.h b/net/packet/internal.h
> > index 907f4cd2a718..fd41ecb7f605 100644
> > --- a/net/packet/internal.h
> > +++ b/net/packet/internal.h
> > @@ -39,7 +39,7 @@ struct tpacket_kbdq_core {
> >       char            *nxt_offset;
> >       struct sk_buff  *skb;
> >
> > -     atomic_t        blk_fill_in_prog;
> > +     rwlock_t        blk_fill_in_prog_lock;
> >
> >       /* Default is set to 8ms */
> >  #define DEFAULT_PRB_RETIRE_TOV       (8)
>