Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1280499 > unrolled thread

Re: [PATCH 2/6] clk: qcom: gdsc: Add support for gdscs with gds hw controller

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH 2/6] clk: qcom: gdsc: Add support for gdscs with gds hw  controller Stephen Boyd <sboyd@codeaurora.org> - 2015-12-01 03:30 +0100
    Re: [PATCH 2/6] clk: qcom: gdsc: Add support for gdscs with gds hw  controller Rajendra Nayak <rnayak@codeaurora.org> - 2015-12-01 04:10 +0100
      Re: [PATCH 2/6] clk: qcom: gdsc: Add support for gdscs with gds hw  controller Stephen Boyd <sboyd@codeaurora.org> - 2015-12-01 08:30 +0100
        Re: [PATCH 2/6] clk: qcom: gdsc: Add support for gdscs with gds hw  controller "Rajendra Nayak" <rnayak@codeaurora.org> - 2015-12-01 17:10 +0100

#1280499 — Re: [PATCH 2/6] clk: qcom: gdsc: Add support for gdscs with gds hw controller

On 11/26, Rajendra Nayak wrote:
> @@ -58,30 +58,34 @@ static int gdsc_toggle_logic(struct gdsc *sc, bool en)
>  {
>  	int ret;
>  	u32 val = en ? 0 : SW_COLLAPSE_MASK;
> -	u32 check = en ? PWR_ON_MASK : 0;
>  	unsigned long timeout;
> +	unsigned int status_reg = sc->gdscr;
>  
>  	ret = regmap_update_bits(sc->regmap, sc->gdscr, SW_COLLAPSE_MASK, val);
>  	if (ret)
>  		return ret;
>  
>  	timeout = jiffies + usecs_to_jiffies(TIMEOUT_US);
> -	do {
> -		ret = regmap_read(sc->regmap, sc->gdscr, &val);
> -		if (ret)
> -			return ret;
>  
> -		if ((val & PWR_ON_MASK) == check)
> +	if (sc->gds_hw_ctrl) {
> +		status_reg = sc->gds_hw_ctrl;
> +	/*
> +	 * The gds hw controller asserts/de-asserts the status bit soon
> +	 * after it receives a power on/off request from a master.
> +	 * The controller then takes around 8 xo cycles to start its internal
> +	 * state machine and update the status bit. During this time, the
> +	 * status bit does not reflect the true status of the core.
> +	 * Add a delay of 1 us between writing to the SW_COLLAPSE bit and
> +	 * polling the status bit
> +	 */

Please indent this correctly to the udelay indent level.

> +		udelay(1);
> +	}
> +
> +	do {
> +		if (gdsc_is_enabled(sc, status_reg) == en)
>  			return 0;
>  	} while (time_before(jiffies, timeout));
>  
> -	ret = regmap_read(sc->regmap, sc->gdscr, &val);
> -	if (ret)
> -		return ret;
> -
> -	if ((val & PWR_ON_MASK) == check)
> -		return 0;
> -

This opens a bug where we timeout and then the status bit changes
after the timeout. One more check is good and should stay. We
could also change this to ktime instead of jiffies. That would be
a good improvement.

>  	return -ETIMEDOUT;
>  }
>  
> @@ -165,6 +169,7 @@ static int gdsc_init(struct gdsc *sc)
>  {
>  	u32 mask, val;
>  	int on, ret;
> +	unsigned int reg;
>  
>  	/*
>  	 * Disable HW trigger: collapse/restore occur based on registers writes.
> @@ -185,7 +190,8 @@ static int gdsc_init(struct gdsc *sc)
>  			return ret;
>  	}
>  
> -	on = gdsc_is_enabled(sc);
> +	reg = sc->gds_hw_ctrl ? sc->gds_hw_ctrl : sc->gdscr;
> +	on = gdsc_is_enabled(sc, reg);

If the gdsc is voteable, then we need to make sure that the vote
is from us when we boot up. Otherwise the kernel may think that
the gdsc is enabled, but it gets turned off by some other master
later on. I don't know if this causes some sort of problem for
the power domain framework, but we can't rely on the status bit
unless we're sure that we've actually set the register to enable
it. In the normal enable/disable path we'll always know we set
the register, so this really only matters once when we boot up.

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[toc] | [next] | [standalone]

#1280527

On 12/01/2015 07:52 AM, Stephen Boyd wrote:
> On 11/26, Rajendra Nayak wrote:
>> @@ -58,30 +58,34 @@ static int gdsc_toggle_logic(struct gdsc *sc, bool en)
>>  {
>>  	int ret;
>>  	u32 val = en ? 0 : SW_COLLAPSE_MASK;
>> -	u32 check = en ? PWR_ON_MASK : 0;
>>  	unsigned long timeout;
>> +	unsigned int status_reg = sc->gdscr;
>>  
>>  	ret = regmap_update_bits(sc->regmap, sc->gdscr, SW_COLLAPSE_MASK, val);
>>  	if (ret)
>>  		return ret;
>>  
>>  	timeout = jiffies + usecs_to_jiffies(TIMEOUT_US);
>> -	do {
>> -		ret = regmap_read(sc->regmap, sc->gdscr, &val);
>> -		if (ret)
>> -			return ret;
>>  
>> -		if ((val & PWR_ON_MASK) == check)
>> +	if (sc->gds_hw_ctrl) {
>> +		status_reg = sc->gds_hw_ctrl;
>> +	/*
>> +	 * The gds hw controller asserts/de-asserts the status bit soon
>> +	 * after it receives a power on/off request from a master.
>> +	 * The controller then takes around 8 xo cycles to start its internal
>> +	 * state machine and update the status bit. During this time, the
>> +	 * status bit does not reflect the true status of the core.
>> +	 * Add a delay of 1 us between writing to the SW_COLLAPSE bit and
>> +	 * polling the status bit
>> +	 */
> 
> Please indent this correctly to the udelay indent level.

will do.

> 
>> +		udelay(1);
>> +	}
>> +
>> +	do {
>> +		if (gdsc_is_enabled(sc, status_reg) == en)
>>  			return 0;
>>  	} while (time_before(jiffies, timeout));
>>  
>> -	ret = regmap_read(sc->regmap, sc->gdscr, &val);
>> -	if (ret)
>> -		return ret;
>> -
>> -	if ((val & PWR_ON_MASK) == check)
>> -		return 0;
>> -
> 
> This opens a bug where we timeout and then the status bit changes
> after the timeout. One more check is good and should stay. We
> could also change this to ktime instead of jiffies. That would be
> a good improvement.

If the status bit does change after timeout maybe the timeout isn't
really enough and needs to be increased?

> 
>>  	return -ETIMEDOUT;
>>  }
>>  
>> @@ -165,6 +169,7 @@ static int gdsc_init(struct gdsc *sc)
>>  {
>>  	u32 mask, val;
>>  	int on, ret;
>> +	unsigned int reg;
>>  
>>  	/*
>>  	 * Disable HW trigger: collapse/restore occur based on registers writes.
>> @@ -185,7 +190,8 @@ static int gdsc_init(struct gdsc *sc)
>>  			return ret;
>>  	}
>>  
>> -	on = gdsc_is_enabled(sc);
>> +	reg = sc->gds_hw_ctrl ? sc->gds_hw_ctrl : sc->gdscr;
>> +	on = gdsc_is_enabled(sc, reg);
> 
> If the gdsc is voteable, then we need to make sure that the vote
> is from us when we boot up. Otherwise the kernel may think that
> the gdsc is enabled, but it gets turned off by some other master
> later on. I don't know if this causes some sort of problem for
> the power domain framework, but we can't rely on the status bit
> unless we're sure that we've actually set the register to enable
> it. In the normal enable/disable path we'll always know we set
> the register, so this really only matters once when we boot up.

right, thanks for catching this. However if we vote for a votable
GDSC just because its ON at boot (due to someone else having voted)
we won't ever remove the vote keeping it always enabled.

I think a safe way would be to consider all votable gdscs for which
*we* haven't voted explicitly to be disabled at boot?

> 

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[toc] | [prev] | [next] | [standalone]

#1280612

On 12/01, Rajendra Nayak wrote:
> 
> On 12/01/2015 07:52 AM, Stephen Boyd wrote:
> > On 11/26, Rajendra Nayak wrote:
> > 
> >> +		udelay(1);
> >> +	}
> >> +
> >> +	do {
> >> +		if (gdsc_is_enabled(sc, status_reg) == en)
> >>  			return 0;
> >>  	} while (time_before(jiffies, timeout));
> >>  
> >> -	ret = regmap_read(sc->regmap, sc->gdscr, &val);
> >> -	if (ret)
> >> -		return ret;
> >> -
> >> -	if ((val & PWR_ON_MASK) == check)
> >> -		return 0;
> >> -
> > 
> > This opens a bug where we timeout and then the status bit changes
> > after the timeout. One more check is good and should stay. We
> > could also change this to ktime instead of jiffies. That would be
> > a good improvement.
> 
> If the status bit does change after timeout maybe the timeout isn't
> really enough and needs to be increased?

The problem is more that this isn't a tight loop with interrupts
disabled, so we may schedule the task away for quite some time
before we come back here and test the timeout against the jiffies
value. So it's best to always check the register one more time
after we bail out in case this occurs.

> 
> > 
> >>  	return -ETIMEDOUT;
> >>  }
> >>  
> >> @@ -165,6 +169,7 @@ static int gdsc_init(struct gdsc *sc)
> >>  {
> >>  	u32 mask, val;
> >>  	int on, ret;
> >> +	unsigned int reg;
> >>  
> >>  	/*
> >>  	 * Disable HW trigger: collapse/restore occur based on registers writes.
> >> @@ -185,7 +190,8 @@ static int gdsc_init(struct gdsc *sc)
> >>  			return ret;
> >>  	}
> >>  
> >> -	on = gdsc_is_enabled(sc);
> >> +	reg = sc->gds_hw_ctrl ? sc->gds_hw_ctrl : sc->gdscr;
> >> +	on = gdsc_is_enabled(sc, reg);
> > 
> > If the gdsc is voteable, then we need to make sure that the vote
> > is from us when we boot up. Otherwise the kernel may think that
> > the gdsc is enabled, but it gets turned off by some other master
> > later on. I don't know if this causes some sort of problem for
> > the power domain framework, but we can't rely on the status bit
> > unless we're sure that we've actually set the register to enable
> > it. In the normal enable/disable path we'll always know we set
> > the register, so this really only matters once when we boot up.
> 
> right, thanks for catching this. However if we vote for a votable
> GDSC just because its ON at boot (due to someone else having voted)
> we won't ever remove the vote keeping it always enabled.
> 
> I think a safe way would be to consider all votable gdscs for which
> *we* haven't voted explicitly to be disabled at boot?
> 

Agreed, when we boot we should consider GDSCs that are indicating
they're enabled via the bit 31 status bit but without the sw
enable mask set as "disabled" even though they're actually
enabled by some other master in the SoC.

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[toc] | [prev] | [next] | [standalone]

#1280991

[]..

>> >>  	return -ETIMEDOUT;
>> >>  }
>> >>
>> >> @@ -165,6 +169,7 @@ static int gdsc_init(struct gdsc *sc)
>> >>  {
>> >>  	u32 mask, val;
>> >>  	int on, ret;
>> >> +	unsigned int reg;
>> >>
>> >>  	/*
>> >>  	 * Disable HW trigger: collapse/restore occur based on registers
>> writes.
>> >> @@ -185,7 +190,8 @@ static int gdsc_init(struct gdsc *sc)
>> >>  			return ret;
>> >>  	}
>> >>
>> >> -	on = gdsc_is_enabled(sc);
>> >> +	reg = sc->gds_hw_ctrl ? sc->gds_hw_ctrl : sc->gdscr;
>> >> +	on = gdsc_is_enabled(sc, reg);
>> >
>> > If the gdsc is voteable, then we need to make sure that the vote
>> > is from us when we boot up. Otherwise the kernel may think that
>> > the gdsc is enabled, but it gets turned off by some other master
>> > later on. I don't know if this causes some sort of problem for
>> > the power domain framework, but we can't rely on the status bit
>> > unless we're sure that we've actually set the register to enable
>> > it. In the normal enable/disable path we'll always know we set
>> > the register, so this really only matters once when we boot up.
>>
>> right, thanks for catching this. However if we vote for a votable
>> GDSC just because its ON at boot (due to someone else having voted)
>> we won't ever remove the vote keeping it always enabled.
>>
>> I think a safe way would be to consider all votable gdscs for which
>> *we* haven't voted explicitly to be disabled at boot?
>>
>
> Agreed, when we boot we should consider GDSCs that are indicating
> they're enabled via the bit 31 status bit but without the sw
> enable mask set as "disabled" even though they're actually
> enabled by some other master in the SoC.

Thinking about this a bit more, your earlier suggestion of voting
for the GDSC explicitly seemed to work too, and also seemed cleaner.
genpd ends up removing the vote if there aren't any users as part
of genpd_poweroff_unused()

-- 
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation

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

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web