| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <1508210661.21840.86.camel@mtkswgap22>
Date: Tue, 17 Oct 2017 11:24:21 +0800
From: Sean Wang <sean.wang@...iatek.com>
To: Alexandre Belloni <alexandre.belloni@...e-electrons.com>
CC: <a.zummo@...ertech.it>, <robh+dt@...nel.org>,
<mark.rutland@....com>, <linux-rtc@...r.kernel.org>,
<devicetree@...r.kernel.org>, <linux-mediatek@...ts.infradead.org>,
<linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 2/4] rtc: mediatek: add driver for RTC on MT7622 SoC
On Mon, 2017-10-16 at 16:17 +0800, Sean Wang wrote:
> Hi Alexandre,
>
> Thanks for your valuable suggestions on the driver.
>
> I added comments inline and will have following-ups in the next version
>
> Sean
>
> On Thu, 2017-10-12 at 23:20 +0200, Alexandre Belloni wrote:
> > Hi,
> >
> > On 22/09/2017 at 11:33:15 +0800, sean.wang@...iatek.com wrote:
> > > diff --git a/drivers/rtc/rtc-mediatek.c b/drivers/rtc/rtc-mediatek.c
> >
> > I'm pretty sure this should be named rtc-mt7622.c instead of
> > rtc-mediatek.c, exactly for the same reason you have patch 3/4.
> >
>
> It's okay for naming with rtc-mt7622.c at this moment. But if more SoCs
> support gets into the driver, I will consider again to give a more
> generic name for the driver.
>
> > > +static void mtk_w32(struct mtk_rtc *rtc, u32 reg, u32 val)
> > > +{
> > > + __raw_writel(val, rtc->base + reg);
> >
> > Do you really need the __raw accessors? What about running your SoC in
> > BE mode? I guess the _relaxed version are fast enough.
> >
>
> SoC runs on LE mode. I also think it's fine and enough to use _relaxed
> version instead of __raw version.
>
> > > +}
> > > +
> > > +static u32 mtk_r32(struct mtk_rtc *rtc, u32 reg)
> > > +{
> > > + return __raw_readl(rtc->base + reg);
> > > +}
> > > +
> >
> >
> > > +static void mtk_rtc_hw_init(struct mtk_rtc *hw)
> > > +{
> > > + /* The setup of the init sequence is for allowing RTC got to work */
> > > + mtk_w32(hw, MTK_RTC_PWRCHK1, RTC_PWRCHK1_MAGIC);
> > > + mtk_w32(hw, MTK_RTC_PWRCHK2, RTC_PWRCHK2_MAGIC);
> > > + mtk_w32(hw, MTK_RTC_KEY, RTC_KEY_MAGIC);
> > > + mtk_w32(hw, MTK_RTC_PROT1, RTC_PROT1_MAGIC);
> > > + mtk_w32(hw, MTK_RTC_PROT2, RTC_PROT2_MAGIC);
> > > + mtk_w32(hw, MTK_RTC_PROT3, RTC_PROT3_MAGIC);
> > > + mtk_w32(hw, MTK_RTC_PROT4, RTC_PROT4_MAGIC);
> > > + mtk_rmw(hw, MTK_RTC_DEBNCE, RTC_DEBNCE_MASK, 0);
> > > + mtk_clr(hw, MTK_RTC_CTL, RTC_RC_STOP);
> > > +}
> > > +
> > > +static void mtk_rtc_get_alarm_or_time(struct mtk_rtc *hw, struct rtc_time *tm,
> > > + int time_alarm)
> > > +{
> > > + u32 year, mon, mday, wday, hour, min, sec;
> > > +
> > > + /*
> > > + * Read again until all fields are not changed for all fields in the
> > > + * consistent state.
> > > + */
> > > + do {
> > > + year = mtk_r32(hw, MTK_RTC_TREG(time_alarm, MTK_YEA));
> > > + mon = mtk_r32(hw, MTK_RTC_TREG(time_alarm, MTK_MON));
> > > + wday = mtk_r32(hw, MTK_RTC_TREG(time_alarm, MTK_DOW));
> > > + mday = mtk_r32(hw, MTK_RTC_TREG(time_alarm, MTK_DOM));
> > > + hour = mtk_r32(hw, MTK_RTC_TREG(time_alarm, MTK_HOU));
> > > + min = mtk_r32(hw, MTK_RTC_TREG(time_alarm, MTK_MIN));
> > > + sec = mtk_r32(hw, MTK_RTC_TREG(time_alarm, MTK_SEC));
> > > + } while (year != mtk_r32(hw, MTK_RTC_TREG(time_alarm, MTK_YEA)) ||
> > > + mon != mtk_r32(hw, MTK_RTC_TREG(time_alarm, MTK_MON)) ||
> > > + mday != mtk_r32(hw, MTK_RTC_TREG(time_alarm, MTK_DOM)) ||
> > > + wday != mtk_r32(hw, MTK_RTC_TREG(time_alarm, MTK_DOW)) ||
> > > + hour != mtk_r32(hw, MTK_RTC_TREG(time_alarm, MTK_HOU)) ||
> > > + min != mtk_r32(hw, MTK_RTC_TREG(time_alarm, MTK_MIN)) ||
> > > + sec != mtk_r32(hw, MTK_RTC_TREG(time_alarm, MTK_SEC))
> > > + );
> >
> > I'm pretty sure only checking sec is enough because it is highly
> > unlikely that 7 reads take a minute.
> >
>
> You're right. I made something stupid here. Only checking on sec is
> enough and will give simpler and better code.
>
> > > +static irqreturn_t mtk_rtc_alarmirq(int irq, void *id)
> > > +{
> > > + struct mtk_rtc *hw = (struct mtk_rtc *)id;
> > > + u32 irq_sta;
> > > +
> > > + /* Stop alarm also implicitly disable the alarm interrupt */
> > > + mtk_w32(hw, MTK_RTC_AL_CTL, 0);
> >
> > You stop the alarm here, before testing whether the alarm really
> > happened.
> >
>
> Okay. I will exchange the order for alarm stopping and the examination
> whether the alarm is really expired.
>
> > > + irq_sta = mtk_r32(hw, MTK_RTC_INT);
> > > + if (irq_sta & RTC_INT_AL_STA) {
> > > + rtc_update_irq(hw->rtc, 1, RTC_IRQF | RTC_AF);
> > > +
> > > + /* Ack alarm interrupt status */
> > > + mtk_w32(hw, MTK_RTredundantC_INT, RTC_INT_AL_STA);
> > > + return IRQ_HANDLED;
> > > + }
> > > +
> > > + return IRQ_NONE;
> > > +}
> > > +
> > > +static int mtk_rtc_gettime(struct device *dev, struct rtc_time *tm)
> > > +{
> > > + struct mtk_rtc *hw = dev_get_drvdata(dev);
> > > +
> > > + mtk_rtc_get_alarm_or_time(hw, tm, MTK_TC);
> > > +
> > > + return rtc_valid_tm(tm);
> > > +}
> > > +
> > > +static int mtk_rtc_settime(struct device *dev, struct rtc_time *tm)
> > > +{
> > > + struct mtk_rtc *hw = dev_get_drvdata(dev);
> > > +
> > > + /* Stop time counter before setting a new one*/
> > > + mtk_set(hw, MTK_RTC_CTL, RTC_RC_STOP);
> > > +
> > > + /* Epoch == 1900 */
> > > + if (tm->tm_year < 100 || tm->tm_year > 199)
> > > + return -EINVAL;
> >
> > Year is a 32 bits register, what makes the RTC fail in 2100? Is it
> > because of the leap year handling?
> >
>
> I made something stupid again here: rtc hardware doesn't have such the
> limitation. I just felt alarm set up prior to 2100 is enough in my
> initial thought, but it seemed I shouldn't do this. I will remove the
> sanity condition.
>
Sorry for that I gave incorrect information for the RTC in the previous
reply: After check again the usage of the register, the maximum number
of the year the RTC can hold is 99 and then wraparound to 0 when
overflow occurs although the year register is a 32 bits register.
Therefore, the sanity for tm->tm_year is still required for the both
setup handler on alarm and rtc to ensure the user input data is valid,
where the current driver assume it's valid when tm->tm_year is between
2000 and 2099. I'll add more comments for the hardware limitation.
Sean
>
> > > +static int mtk_rtc_setalarm(struct device *dev, struct rtc_wkalrm *wkalrm)
> > > +{
> > > + struct mtk_rtc *hw = dev_get_drvdata(dev);
> > > + struct rtc_time *alrm_tm = &wkalrm->time;
> > > +
> > > + /* Epoch == 1900 */
> > > + if (alrm_tm->tm_year < 100 || alrm_tm->tm_year > 199)
> > > + return -EINVAL;
> > > +
> >
> > Ditto.
> >
> Ditto. those condition will be removed.
>
> > > +
> > > + dev_info(&pdev->dev, "MediaTek SoC based RTC enabled\n");
> > > +
> >
> > I think the rtc core is verbose enough that this message is not needed.
> >
>
> Okay. the redundant and specific log prompt would be removed as well.
>
>
> >
>
Powered by blists - more mailing lists