From mboxrd@z Thu Jan 1 00:00:00 1970 From: Lee Jones Subject: Re: [PATCH] mfd: mt6397: irq domain should initialize before mfd_add_devices() Date: Thu, 7 Apr 2016 08:52:14 +0100 Message-ID: <20160407075214.GX3323@x1> References: <1459322708-21363-1-git-send-email-henryc.chen@mediatek.com> <56FB99DB.6010901@openwrt.org> <1459388452.18284.2.camel@mtksdaap41> <1459391533.21870.15.camel@mtksdaap41> <56FCE910.8030100@openwrt.org> <1459431711.395.6.camel@mtksdaap41> <56FD2F18.4050508@openwrt.org> <1459910866.18284.27.camel@mtksdaap41> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: Content-Disposition: inline In-Reply-To: <1459910866.18284.27.camel@mtksdaap41> Sender: linux-kernel-owner@vger.kernel.org To: Henry Chen Cc: John Crispin , Yingjoe Chen , Sascha Hauer , linux-kernel@vger.kernel.org, Flora Fu , linux-mediatek@lists.infradead.org, linux@roeck-us.net, Matthias Brugger , linux-arm-kernel@lists.infradead.org List-Id: linux-mediatek@lists.infradead.org On Wed, 06 Apr 2016, Henry Chen wrote: > On Thu, 2016-03-31 at 16:07 +0200, John Crispin wrote: > >=20 > > On 31/03/2016 15:41, Yingjoe Chen wrote: > > > On Thu, 2016-03-31 at 11:08 +0200, John Crispin wrote: > > >> > > >> On 31/03/2016 04:32, Yingjoe Chen wrote: > > >>> On Thu, 2016-03-31 at 09:40 +0800, Henry Chen wrote: > > >>>> On Wed, 2016-03-30 at 11:18 +0200, John Crispin wrote: > > >>>>> Hi, > > >>>>> > > >>>>> small nitpick inline > > >>>>> > > >>>>> On 30/03/2016 09:25, Henry Chen wrote: > > >>>>>> Some sub driver like RTC module need irq domain from parent = to create > > >>>>>> irq mapping when driver initialize. so move mt6397_irq_init(= ) before > > >>>>>> mfd_add_devices(). > > >>>>>> > > >>>>>> Signed-off-by: Henry Chen > > >>>>>> --- > > >>>>>> This patch fixed the below warning based on "Linux kernel v4= =2E6-rc1" > > >>>>>> WARNING: CPU: 1 PID: 132 at kernel/mediatek/kernel/irq/irqdo= main.c:471 > > >>>>>> irq_create_mapping+0xc4/0xd0 > > >>>>>> --- > > >>>>>> drivers/mfd/mt6397-core.c | 37 ++++++++++++++++++++++------= --------- > > >>>>>> 1 file changed, 22 insertions(+), 15 deletions(-) > > >>>>>> > > >>>>>> diff --git a/drivers/mfd/mt6397-core.c b/drivers/mfd/mt6397-= core.c > > >>>>>> index 8e8d932..a879223 100644 > > >>>>>> --- a/drivers/mfd/mt6397-core.c > > >>>>>> +++ b/drivers/mfd/mt6397-core.c > > >>>>>> @@ -270,22 +270,36 @@ static int mt6397_probe(struct platfor= m_device *pdev) > > >>>>>> goto fail_irq; > > >>>>>> } > > >>>>>> =20 > > >>>>>> + pmic->irq =3D platform_get_irq(pdev, 0); > > >>>>>> + > > >>>>>> switch (id & 0xff) { > > >>>>>> case MT6323_CID_CODE: > > >>>>>> - pmic->int_con[0] =3D MT6323_INT_CON0; > > >>>>>> - pmic->int_con[1] =3D MT6323_INT_CON1; > > >>>>>> - pmic->int_status[0] =3D MT6323_INT_STATUS0; > > >>>>>> - pmic->int_status[1] =3D MT6323_INT_STATUS1; > > >>>>>> + if (pmic->irq > 0) { > > >>>>> > > >>>>> should this not be > > >>>>> > > >>>>> if (pmic->irq >=3D 0) { > > >>>>> > > >>>>> i think the code before your patch was wrong as linux irqs st= art with 0. > > >>>>> > > >>>>> John > > >>>> Hi John, > > >>>> > > >>>> Thanks, I will modify this. > > >>> > > >>> Linux irq start from 1, 0 is invalid. I can't find the document= saying > > >>> this now, but you could see this from irq_create_mapping() in > > >>> kernel/irq/irqdomain.c > > >>> > > >>> I think the code should have check return from platform_get_irq= and > > >>> handle -EPROBE_DEFER, but maybe it should be another patch? > > >>> > > >>> BTW, in this function, it is possible that pmic->irq_domain wil= l be NULL > > >>> in fail_irq error handling. We should check before calling > > >>> irq_domain_remove. > > >>> > > >>> Joe.C > > >>> > > >> > > >> Hi, > > >> > > >> looking at > > >> http://lxr.free-electrons.com/source/drivers/base/platform.c#L87= there > > >> is a check in line #100 ret >=3D 0 > > >> > > >> checking the return value of pmic->irq =3D platform_get_irq(pdev= , 0); > > >> should follow the same pattern i think .. unless i have a thinko= and am > > >> reading the code wrong. > > >=20 > > >=20 > > > I'm not sure why platform_get_irq() check for 0, but I think the = code > > > logic is differnet. > > >=20 > > > When platform_get_irq() return 0 to our code, it means we don't h= ave > > > valid irq to use. In this case it doesn't make any sense to conti= nue > > > init irq. > > >=20 > > >=20 > > > Joe.C > > >=20 > >=20 > >=20 > > --> http://lwn.net/Articles/470820/ > >=20 > > indeed ARM has changed this is seems. was not aware of this change, > > sorry for the noise > >=20 > > John >=20 > Hi Lee/Joe/John, >=20 > Thanks for comment, that means we can keep this patch for 4.6-rc1 > regression (RTC get NULL irq issue), right? Just re-send everything with any Acks that you collected. > I will send another patch to fixed above error handling problem. --=20 Lee Jones Linaro STMicroelectronics Landing Team Lead Linaro.org =E2=94=82 Open source software for ARM SoCs =46ollow Linaro: Facebook | Twitter | Blog