The Linux Kernel Mailing List
 help / color / mirror / Atom feed
* Re: [PATCH] mfd: axp20x: Change volatile ranges on axp803
       [not found]   ` <voxgbzeunk7doe64ej3eq22kgxw7y4x7wwqsm72t546qreonmi@ypnuw5esnddo>
@ 2026-05-09 13:00     ` Jernej Škrabec
  0 siblings, 0 replies; only message in thread
From: Jernej Škrabec @ 2026-05-09 13:00 UTC (permalink / raw)
  To: Ondřej Jirman <megi@xff.cz>, Jernej Škrabec,
	linux-sunxi, linux-kernel, Lee Jones, Chen-Yu Tsai

Dne sobota, 21. marec 2026 ob 19:26:43 Srednjeevropski poletni čas je Ondřej Jirman napisal(a):
> On Sat, Mar 21, 2026 at 07:25:49AM +0100, Jernej Škrabec wrote:
> > Dne nedelja, 15. marec 2026 ob 13:49:31 Srednjeevropski standardni čas je Ondřej Jirman napisal(a):
> > > From: Ondrej Jirman <megi@xff.cz>
> > > 
> > > Regulator control registers can be cached. They don't update by
> > > themselves. Enable cache on them, to speed up voltage changes,
> > > particularly DCDC2 which needs to be handled quickly due to being used
> > > for CPUX cores and is changed very frequently by some cpufreq schedulers.
> > 
> > That's not really the case, DCDC2-DCDC6 registers have DVM finished status
> > bit, which is volatile and it is right to have them in volatile range.
> > However, it is true that other registers are non-volatile, so that range can
> > be considerably shrinked. It seems that AXP288 have same register layout
> > but I can't say for sure (I only found datasheet for AXP288C), so it's best
> > to leave it alone and apply changes only to AXP803. 
> 
> While true, it's not used by any kernel drivers, and having the whole register
> marked volatile for a single unused bit is contributing 10s of % of permanent
> CPU load just from cpufreq DVFS operations alone when using schedutil governor,
> which is doing DVFS hundred or more times per second.
> 
> If someone will want to use this bit in the regulator driver, they can use eg.
> regmap_read_bypassed() to access the bit, while keepig the other parts of
> regulator implementation performant.

This sounds like a good compromise.

> 
> But I guess it may deserve a comment in the code.

For sure.

BR Jernej

> 
> Kind regards,
> 	o.
> 
> > Best regards,
> > Jernej
> > 
> > > 
> > > This shrinks register access over RSB bus to one write per voltage
> > > change (down from 4 or so).
> > > 
> > > Signed-off-by: Ondrej Jirman <megi@xff.cz>
> > > ---
> > >  drivers/mfd/axp20x.c | 28 +++++++++++++++++++++++++++-
> > >  1 file changed, 27 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/mfd/axp20x.c b/drivers/mfd/axp20x.c
> > > index 0d0d40972eb8..2ea409029d2f 100644
> > > --- a/drivers/mfd/axp20x.c
> > > +++ b/drivers/mfd/axp20x.c
> > > @@ -172,6 +172,18 @@ static const struct regmap_range axp288_volatile_ranges[] = {
> > >  	regmap_reg_range(AXP20X_FG_RES, AXP288_FG_CC_CAP_REG),
> > >  };
> > >  
> > > +static const struct regmap_range axp803_volatile_ranges[] = {
> > > +	regmap_reg_range(AXP20X_PWR_INPUT_STATUS, AXP288_POWER_REASON),
> > > +	regmap_reg_range(AXP288_BC_GLOBAL, AXP288_BC_GLOBAL),
> > > +	regmap_reg_range(AXP288_BC_DET_STAT, AXP20X_VBUS_IPSOUT_MGMT),
> > > +	regmap_reg_range(AXP20X_CHRG_BAK_CTRL, AXP20X_CHRG_BAK_CTRL),
> > > +	regmap_reg_range(AXP20X_IRQ1_EN, AXP20X_IPSOUT_V_HIGH_L),
> > > +	regmap_reg_range(AXP20X_TIMER_CTRL, AXP20X_TIMER_CTRL),
> > > +	regmap_reg_range(AXP20X_GPIO1_CTRL, AXP22X_GPIO_STATE),
> > > +	regmap_reg_range(AXP288_RT_BATT_V_H, AXP288_RT_BATT_V_L),
> > > +	regmap_reg_range(AXP20X_FG_RES, AXP288_FG_CC_CAP_REG),
> > > +};
> > > +
> > >  static const struct regmap_access_table axp288_writeable_table = {
> > >  	.yes_ranges	= axp288_writeable_ranges,
> > >  	.n_yes_ranges	= ARRAY_SIZE(axp288_writeable_ranges),
> > > @@ -182,6 +194,11 @@ static const struct regmap_access_table axp288_volatile_table = {
> > >  	.n_yes_ranges	= ARRAY_SIZE(axp288_volatile_ranges),
> > >  };
> > >  
> > > +static const struct regmap_access_table axp803_volatile_table = {
> > > +	.yes_ranges	= axp803_volatile_ranges,
> > > +	.n_yes_ranges	= ARRAY_SIZE(axp803_volatile_ranges),
> > > +};
> > > +
> > >  static const struct regmap_range axp806_writeable_ranges[] = {
> > >  	regmap_reg_range(AXP20X_DATACACHE(0), AXP20X_DATACACHE(3)),
> > >  	regmap_reg_range(AXP806_PWR_OUT_CTRL1, AXP806_CLDO3_V_CTRL),
> > > @@ -456,6 +473,15 @@ static const struct regmap_config axp22x_regmap_config = {
> > >  	.cache_type	= REGCACHE_MAPLE,
> > >  };
> > >  
> > > +static const struct regmap_config axp803_regmap_config = {
> > > +	.reg_bits	= 8,
> > > +	.val_bits	= 8,
> > > +	.wr_table	= &axp288_writeable_table,
> > > +	.volatile_table	= &axp803_volatile_table,
> > > +	.max_register	= AXP288_FG_TUNE5,
> > > +	.cache_type	= REGCACHE_MAPLE,
> > > +};
> > > +
> > >  static const struct regmap_config axp288_regmap_config = {
> > >  	.reg_bits	= 8,
> > >  	.val_bits	= 8,
> > > @@ -1368,7 +1394,7 @@ int axp20x_match_device(struct axp20x_dev *axp20x)
> > >  	case AXP803_ID:
> > >  		axp20x->nr_cells = ARRAY_SIZE(axp803_cells);
> > >  		axp20x->cells = axp803_cells;
> > > -		axp20x->regmap_cfg = &axp288_regmap_config;
> > > +		axp20x->regmap_cfg = &axp803_regmap_config;
> > >  		axp20x->regmap_irq_chip = &axp803_regmap_irq_chip;
> > >  		break;
> > >  	case AXP806_ID:
> > > 
> > 
> > 
> > 
> > 
> 





^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2026-05-09 13:00 UTC | newest]

Thread overview: (only message) (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
     [not found] <20260315124932.3669260-1-megi@xff.cz>
     [not found] ` <5084920.31r3eYUQgx@jernej-laptop>
     [not found]   ` <voxgbzeunk7doe64ej3eq22kgxw7y4x7wwqsm72t546qreonmi@ypnuw5esnddo>
2026-05-09 13:00     ` [PATCH] mfd: axp20x: Change volatile ranges on axp803 Jernej Škrabec

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox