Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20190512012508.10608-1-elder@linaro.org> <20190512012508.10608-10-elder@linaro.org>
 <CAK8P3a0eYWN6mMwft5OSu8wQQo=kWh5safGFFNkDCELZJyiMmQ@mail.gmail.com> <58b03138-d99f-b22f-5f8a-969612163135@linaro.org>
In-Reply-To: <58b03138-d99f-b22f-5f8a-969612163135@linaro.org>
From:   Arnd Bergmann <arnd@arndb.de>
Date:   Wed, 15 May 2019 22:50:17 +0200
Message-ID: <CAK8P3a1CD2CjLzvWAeZcpEo5U2-fX1wgjKQ2-1oyigwE_KwBqg@mail.gmail.com>
Subject: Re: [PATCH 09/18] soc: qcom: ipa: GSI transactions
To:     Alex Elder <elder@linaro.org>
Cc:     David Miller <davem@davemloft.net>,
        Bjorn Andersson <bjorn.andersson@linaro.org>,
        Ilias Apalodimas <ilias.apalodimas@linaro.org>,
        syadagir@codeaurora.org, mjavid@codeaurora.org,
        evgreen@chromium.org, Ben Chan <benchan@google.com>,
        Eric Caruso <ejcaruso@google.com>, abhishek.esse@gmail.com,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, May 15, 2019 at 2:26 PM Alex Elder <elder@linaro.org> wrote:
> On 5/15/19 2:34 AM, Arnd Bergmann wrote:
> >> +/* Cancel a channel's pending transactions */
> >> +void gsi_channel_trans_cancel_pending(struct gsi_channel *channel)
> >> +{
> >> +       struct gsi_trans_info *trans_info = &channel->trans_info;
> >> +       u32 evt_ring_id = channel->evt_ring_id;
> >> +       struct gsi *gsi = channel->gsi;
> >> +       struct gsi_evt_ring *evt_ring;
> >> +       struct gsi_trans *trans;
> >> +       unsigned long flags;
> >> +
> >> +       evt_ring = &gsi->evt_ring[evt_ring_id];
> >> +
> >> +       spin_lock_irqsave(&evt_ring->ring.spinlock, flags);
> >> +
> >> +       list_for_each_entry(trans, &trans_info->pending, links)
> >> +               trans->result = -ECANCELED;
> >> +
> >> +       list_splice_tail_init(&trans_info->pending, &trans_info->complete);
> >> +
> >> +       spin_unlock_irqrestore(&evt_ring->ring.spinlock, flags);
> >> +
> >> +       spin_lock_irqsave(&gsi->spinlock, flags);
> >> +
> >> +       if (gsi->event_enable_bitmap & BIT(evt_ring_id))
> >> +               gsi_event_handle(gsi, evt_ring_id);
> >> +
> >> +       spin_unlock_irqrestore(&gsi->spinlock, flags);
> >> +}
> >
> > That is a lot of irqsave()/irqrestore() operations. Do you actually call
> > all of these functions from hardirq context?
>
> The transaction list is definitely updated in IRQ context,
> but I think it is no longer updated in hardirq context (the
> softirq was a recent change).  This particular function is
> definitely not called in a hardirq context, so I can remove
> the irqsave/irqrestore.

If you want to protect against concurrent softirqs, you still
need spin_lock_bh(), which is cheaper than spin_lock_irqsave()
but still requires writing to the shared cache line for the
atomic update of the lock.

> I'll survey my spinlock use throughout the driver and will
> remove any irqsave/irqrestore used in non-hardirq contexts.

Ok. I actually hope that most of the spinlocks can be
removed from the data path entirely. I just replied on the
ring.spinlock, which I think can go away and be replaced
either with two atomic_t values (rp_local and wp_local
only; 'wp' appears to be unused), or even just an smp_rmb()/
smp_wmb() pair for each access. The gsi register spinlock
can probably be avoided as well if we stop disabling and
renabling the interrupts as I suggested.

gsi_trans_info->spinlock is harder to get rid of unfortunately,
as that would require changing the way you do the doubly linked
lists.

     Arnd