> -----Original Message-----
> From: Elior, Ariel
> Sent: Wednesday, March 21, 2018 7:10 PM
> To: da...@davemloft.net
> Cc: netdev@vger.kernel.org; Kalderon, Michal <michal.kalde...@cavium.com>;
> Chopra, Manish <manish.cho...@cavium.com>
> Subject: RE: [PATCH net 1/1] qede: Fix barrier usage after tx doorbell write.
>
> > Subject: [PATCH net 1/1] qede: Fix barrier usage after tx doorbell write.
> >
> > Since commit c5ad119fb6c09b0297446be05bd66602fa564758
> > ("net: sched: pfifo_fast use skb_array") driver is exposed to an issue
> > where it is hitting NULL skbs while handling TX completions. Driver
> > uses mmiowb() to flush the writes to the doorbell bar which is a
> > write-combined bar, however on x86
> > mmiowb() does not flush the write combined buffer.
> >
> > This patch fixes this problem by replacing mmiowb() with wmb() after
> > the write combined doorbell write so that writes are flushed and
> > synchronized from more than one processor.
> >
> > Signed-off-by: Ariel Elior <ariel.el...@cavium.com>
> > Signed-off-by: Manish Chopra <manish.cho...@cavium.com>
> > ---
> > drivers/net/ethernet/qlogic/qede/qede_fp.c | 10 ++++------
> > 1 files changed, 4 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/net/ethernet/qlogic/qede/qede_fp.c
> > b/drivers/net/ethernet/qlogic/qede/qede_fp.c
> > index dafc079..2e921ca 100644
> > --- a/drivers/net/ethernet/qlogic/qede/qede_fp.c
> > +++ b/drivers/net/ethernet/qlogic/qede/qede_fp.c
> > @@ -320,13 +320,11 @@ static inline void
> > qede_update_tx_producer(struct qede_tx_queue *txq)
> > barrier();
> > writel(txq->tx_db.raw, txq->doorbell_addr);
> >
> > - /* mmiowb is needed to synchronize doorbell writes from more than
> one
> > - * processor. It guarantees that the write arrives to the device before
> > - * the queue lock is released and another start_xmit is called (possibly
> > - * on another CPU). Without this barrier, the next doorbell can bypass
> > - * this doorbell. This is applicable to IA64/Altix systems.
> > + /* Fence required to flush the write combined buffer, since another
> > + * CPU may write to the same doorbell address and data may be lost
> > + * due to relaxed order nature of write combined bar.
> > */
> > - mmiowb();
> > + wmb();
> > }
> >
> > static int qede_xdp_xmit(struct qede_dev *edev, struct qede_fastpath
> > *fp,
> > --
> > 1.7.1
>
> Hi Dave,
> This patch appears as "superseded" in patchwork. I am not really sure why
> that is
> - I noticed some other barrier work is going on, but none of it will solve
> this
> issue. This patch solves an important bug in the driver - please consider
> applying
> it.
> Thanks,
> Ariel
Hi Dave,
The other patchwork which is going on to use writel_relaxed doesn't solve this
issue.
I think the writel_relaxed patchwork might have caused this fix as superseded
since those changes are in same area/function.
This is an important fix which is after the write combined doorbell write.
Please let me know if I should re-submit this fix or not ?
Thanks,
Manish
