Re: flexcan missing error state transitions

From: Andri Yngvason <andri.yngvason@marel.com>
To: "ZHU Yi (ST-FIR/ENG1-Zhu)" <Yi.Zhu5@cn.bosch.com>,
	"linux-can@vger.kernel.org" <linux-can@vger.kernel.org>,
	"mkl@pengutronix.de" <mkl@pengutronix.de>,
	"wg@grandegger.com" <wg@grandegger.com>
Cc: "hs@denx.de" <hs@denx.de>,
	"Jonas Mark (ST-FIR/ENG1)" <Mark.Jonas@de.bosch.com>,
	"RUAN Tingquan (ST-FIR/ENG1-Zhu)" <Tingquan.Ruan@cn.bosch.com>
Subject: Re: flexcan missing error state transitions
Date: Mon, 21 Aug 2017 16:18:23 +0000	[thread overview]
Message-ID: <150333230375.13890.2773067532209799515@maxwell> (raw)
In-Reply-To: <5213bc7ebd454acea568aba682d696ca@SGPMBX1017.APAC.bosch.com>

Hello Yi ZHU,

Quoting ZHU Yi (ST-FIR/ENG1-Zhu) (2017-08-18 07:47:41)
> Sadly, the flexcan core does not generate error state interrupts for all state transitions.
> This is required for the current flexcan driver to operate smoothly. We propose the following
> workaround to mitigate this problem:
> 1. update state upon _any_ interrupt
> 2. update state when user explicitly inquires (via do_get_state callback)

For now, I think that it would be good enough to implement option 1 as this is
how things were before the regression. Unless there's a performance penalty?

I am not sure that I like the userspace polling idea. I think it would be better
to have the polling implemented within the driver because it keeps the interface
"simple" and there will be fewer differences between different platforms. I.e.
the same userpspace code should work for flexcan, sja1000, mscan, etc..

Regards,
Andri