[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CANMq1KC0nDNRUGm8Y5Lw1nH1uecYx7FnG9sWY=Gg9RhisroafA@mail.gmail.com>
Date: Sat, 31 Oct 2020 09:49:11 +0800
From: Nicolas Boichat <drinkcat@...omium.org>
To: Houlong Wei <houlong.wei@...iatek.com>
Cc: Matthias Brugger <matthias.bgg@...il.com>,
Jassi Brar <jassisinghbrar@...il.com>,
Daniel Kurtz <djkurtz@...omium.org>,
lkml <linux-kernel@...r.kernel.org>,
linux-arm Mailing List <linux-arm-kernel@...ts.infradead.org>,
"moderated list:ARM/Mediatek SoC support"
<linux-mediatek@...ts.infradead.org>,
srv_heupstream <srv_heupstream@...iatek.com>,
CK HU <ck.hu@...iatek.com>,
Bibby Hsieh <bibby.hsieh@...iatek.com>,
Daoyuan Huang <daoyuan.huang@...iatek.com>,
Dennis-YC Hsieh <dennis-yc.hsieh@...iatek.com>,
ginny.chen@...iatek.com,
Yongqiang Niu <yongqiang.niu@...iatek.com>,
Nicolas Boichat <drinkcat@...omium.org>
Subject: Re: [PATCH] soc: mediatek: cmdq: fixup possible timeout issue
Thanks for the patch.
On Thu, Oct 22, 2020 at 5:44 PM Houlong Wei <houlong.wei@...iatek.com> wrote:
>
> Fixes: 576f1b4bc802 ("soc: mediatek: Add Mediatek CMDQ helper")
nit: This belongs right next to the Sob line, but I guess/hope
Matthias can help you fix that ,-)
>
> There may be possible timeout issue when lots of cmdq packets are
> flushed to the same cmdq client. The necessary modifications are as
> below.
> 1.Adjust the timer timeout period as client->timeout_ms * client->pkt_cnt.
> 2.Optimize the time to start the timer.
>
Reviewed-by: Nicolas Boichat <drinkcat@...omium.org>
> Signed-off-by: Houlong Wei <houlong.wei@...iatek.com>
> ---
> drivers/soc/mediatek/mtk-cmdq-helper.c | 22 ++++++++++++++++++----
> 1 file changed, 18 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/soc/mediatek/mtk-cmdq-helper.c b/drivers/soc/mediatek/mtk-cmdq-helper.c
> index dc644cfb6419..31142c193527 100644
> --- a/drivers/soc/mediatek/mtk-cmdq-helper.c
> +++ b/drivers/soc/mediatek/mtk-cmdq-helper.c
> @@ -350,7 +350,8 @@ static void cmdq_pkt_flush_async_cb(struct cmdq_cb_data data)
> del_timer(&client->timer);
> else
> mod_timer(&client->timer, jiffies +
> - msecs_to_jiffies(client->timeout_ms));
> + msecs_to_jiffies(client->timeout_ms *
> + client->pkt_cnt));
> spin_unlock_irqrestore(&client->lock, flags);
> }
>
> @@ -379,9 +380,7 @@ int cmdq_pkt_flush_async(struct cmdq_pkt *pkt, cmdq_async_flush_cb cb,
>
> if (client->timeout_ms != CMDQ_NO_TIMEOUT) {
> spin_lock_irqsave(&client->lock, flags);
> - if (client->pkt_cnt++ == 0)
> - mod_timer(&client->timer, jiffies +
> - msecs_to_jiffies(client->timeout_ms));
> + client->pkt_cnt++;
> spin_unlock_irqrestore(&client->lock, flags);
> }
>
> @@ -391,6 +390,21 @@ int cmdq_pkt_flush_async(struct cmdq_pkt *pkt, cmdq_async_flush_cb cb,
> /* We can send next packet immediately, so just call txdone. */
> mbox_client_txdone(client->chan, 0);
>
> + if (client->timeout_ms != CMDQ_NO_TIMEOUT) {
> + spin_lock_irqsave(&client->lock, flags);
> + /*
> + * GCE HW maybe execute too quickly and the callback function
> + * may be invoked earlier. If this happens, pkt_cnt is reduced
> + * by 1 in cmdq_pkt_flush_async_cb(). The timer is set only if
> + * pkt_cnt is greater than 0.
> + */
> + if (client->pkt_cnt > 0)
> + mod_timer(&client->timer, jiffies +
> + msecs_to_jiffies(client->timeout_ms *
> + client->pkt_cnt));
> + spin_unlock_irqrestore(&client->lock, flags);
> + }
> +
> return 0;
> }
> EXPORT_SYMBOL(cmdq_pkt_flush_async);
> --
> 2.18.0
Powered by blists - more mailing lists