Message-ID: <5167F972.2000707@atmel.com>
Date:	Fri, 12 Apr 2013 14:09:22 +0200
From:	Nicolas Ferre <nicolas.ferre@...el.com>
To:	Johan Hovold <jhovold@...il.com>,
	Jean-Christophe PLAGNIOL-VILLARD <plagnioj@...osoft.com>
CC:	<rtc-linux@...glegroups.com>, Andrew Victor <linux@...im.org.za>,
	Alessandro Zummo <a.zummo@...ertech.it>,
	<linux-arm-kernel@...ts.infradead.org>,
	<linux-kernel@...r.kernel.org>
Subject: Re: [rtc-linux] Re: [PATCH v2 0/5] ARM: at91: fix hanged boot

On 04/12/2013 11:33 AM, Johan Hovold :
> On Thu, Apr 11, 2013 at 06:54:14PM +0200, Jean-Christophe PLAGNIOL-VILLARD wrote:
>> On 17:55 Thu 11 Apr     , Johan Hovold wrote:
>>> On Mon, Mar 11, 2013 at 07:07:54PM +0100, Johan Hovold wrote:
>>>> These patches fix a few severe issues affecting most AT91 SOCs where
>>>> boot can hang after a non-general reset, and where the only way to get
>>>> the system booting again is to do a general reset -- something which
>>>> could require physically removing any backup battery.
>>>
>>> Have you had time to look at these patches yet, Nicolas?
>>>
>>> I don't think not having decided on the path forward for DT-support for
>>> rtc-at91sam9 needs to be a blocker. The rtt-nodes will be needed in any
>>> case.
>>>
>>> I could respin the series on top of the DT-patch for rtc-at91rm9200, and
>>> add interrupt and status-disabled properties to the DT-nodes as well.
>> for this this is still a no go
>>
>> this way too much ugly
> 
> I understand that you prefer fixing every bootloader. I was just making
> sure everyone agrees that that is the best solution.
> 
> The two interrupt masks has to be cleared before the kernel enables the
> system interrupt; either it needs to be done by the bootloader or by the
> at91 arch code.
> 
> The various bootloaders may not know anything about RTT or RTC, but
> have all made sure interrupts are disabled before executing the kernel.
> That is, they have fulfilled the requirement that interrupts must be
> disabled.
> 
> So the trade-off seems to be: Either we fix this once and for all using
> the infrastructure already in place in the kernel (DT), or risk further
> (apparently) bricked systems as there are bound to be bootloaders that
> won't get updated.

Note that I didn't read your patch series yet, so I am not commenting on
the implementation.

BUT, from my experience with customers facing this issue, I do thing
that we must provide a solution (even in Linux kernel itself).


> [...]
> 
>>>> The problems stem from the fact that the RTC and RTT-peripherals are
>>>> powered by backup power (VDDBU) and are not reset on wake-up, user,
>>>> watchdog or software reset. Consequently, RTC and RTT-alarms and their
>>>> interrupts may be enabled at boot, leading to a system lock-up when an
>>>> interrupt arrives on the shared system-interrupt line before the
>>>> appropriate handler (e.g. RTC-driver) has been installed.
>>>>
>>>> The easiest way to trigger this is to simply wake up from an RTC-alarm
>>>> on at91sam9g45. The RTC-driver currently does not disable interrupts at
>>>> shutdown so even after a clean shut-down the system will always hang
>>>> after waking up.
>>>>
>>>> The first patch fixes this very general case of RTC-wake up after a
>>>> clean shutdown in the RTC-driver and is marked for stable as it is
>>>> perfectly straight-forward. [ Note that the other, RTT-based, AT91
>>>> RTC-driver already disables its interrupts at shutdown. ]
> 
> And what about this patch? If it's decided that every bootloader needs
> to be updated, then perhaps it's better to risk bricked systems also
> after a clean shutdown to enforce those updates? Should we then remove
> the corresponding disable of interrupts at shutdown from the rtc-at91sam9
> driver by the same logic?
> 
>>>> The more general problem can be triggered, for example, by doing a
>>>> user-reset while updating the RTC-time or if an RTC or RTT-alarm goes
>>>> off after a non-clean shutdown.
>>>>
>>>> To fix this I propose that arch-code should mask the relevant interrupts
>>>> before enabling the system interrupt at early boot, and this is what
>>>> the fifth patch does. To access the RTC-registers I choose to revert a
>>>> recent patch that moved the register definitions to drivers/rtc.
>>>>
>>>> Arguably, the relevant interrupts could also be disabled in bootloaders,
>>>> but I suggest fixing it in the kernel once and for all.
> 
> Thanks,
> Johan
> 
> 


-- 
Nicolas Ferre
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists