Re: [PATCH 1/3] drm/bridge: analogix_dp: deassert DP_PHY_PD on power-on for exynos

From: Alexandre Marquet

Date: Sun Nov 15 2026 - 17:20:45 EST


Hi,

Thank you for this feedback.

Le Mon, Nov 09, 2026 at 02:15:25PM +0900, Inki Dae a écrit :
> Hi,
>
> 2026년 10월 27일 (화) 오후 9:17, Alexandre Marquet via B4 Relay
> <devnull+tb.a-marquet.fr@xxxxxxxxxx>님이 작성:
> >
> > From: Alexandre Marquet <tb@xxxxxxxxxxxx>
> >
> > On exynos platforms at least, calling
> > analogix_dp_set_analog_power_down() to power on (with enable=false) CH0
> > to CH3 or the AUX bus currently produces no effect, unless the function
> > is first called to power on everything (by setting block to POWER_ALL or
> > ANALOG_TOTAL). The reason for this is that DP_PHY_PD being asserted
> > (which is its default state) renders every other register bit
> > ineffective.
> >
> > This is particularly an issue for AUX bus initialization, that tends to
> > happen before anything else (e.g., to request EDID), and thus, before
> > anything has any chance to request a power on of the whole analog
> > section of the chip.
> >
> > This commit allows analogix_dp_set_analog_power_down() to power on any
> > block, by deasserting DP_PHY_PD if needed.
> >
> > Signed-off-by: Alexandre Marquet <tb@xxxxxxxxxxxx>
> > ---
> > drivers/gpu/drm/bridge/analogix/analogix_dp_reg.c | 19 +++++++++++++------
> > 1 file changed, 13 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/bridge/analogix/analogix_dp_reg.c b/drivers/gpu/drm/bridge/analogix/analogix_dp_reg.c
> > index c1344a3f013a..9f634d4e4793 100644
> > --- a/drivers/gpu/drm/bridge/analogix/analogix_dp_reg.c
> > +++ b/drivers/gpu/drm/bridge/analogix/analogix_dp_reg.c
> > @@ -256,6 +256,19 @@ void analogix_dp_set_analog_power_down(struct analogix_dp_device *dp,
> > if (dp->plat_data && analogix_dp_is_rockchip(dp->plat_data->dev_type))
> > phy_pd_addr = ANALOGIX_DP_PD;
> >
> > + reg = readl(dp->reg_base + phy_pd_addr);
> > +
> > + if (dp->plat_data && (dp->plat_data->dev_type == EXYNOS_DP)) {
> > + /* On exynos, register AUX_PD and CH[0-3]_PD have no effect if
> > + * DP_PHY_PD is asserted. So, when powering any of them up, we
> > + * need to deassert DP_PHY_PD first.
> > + */
> > + if (!enable && (reg & DP_PHY_PD)) {
> > + reg &= ~DP_PHY_PD;
> > + reg |= AUX_PD | CH0_PD | CH1_PD | CH2_PD | CH3_PD;
> > + }
>
> When DP_PHY_PD is set, this path clears it but leaves the bandgap
> power-down bits unchanged. If the register retains 0xff from
> POWER_ALL(true), an AUX_BLOCK, false call produces 0xcf; both
> DP_INC_BG and DP_EXP_BG remain set. Could you check whether AUX can
> operate in that state?

It can. But what happens is that after calling POWER_ALL(true) (thus
setting the register to 0xff), when reading the register again, then I
get 0x7f. Hence, after calling AUX_BLOCK(false), we get 0x4f and not
0xcf. And although DP_INC_BG remains set as a result, AUX still works.
That being said, it sounds wrong not to deassert DP_INC_BG. This was
likely added to POWER_ALL for a reason... DP_EXP_BG, on the other hand,
does not seem to have any effect (and is not used anywhere, as far as I
can tell).

> I suggest limiting the new handling to AUX_BLOCK and CH[0-3]_BLOCK.
> When the bandgap bits are set, keep the other blocks powered down,
> release DP_INC_BG, wait as the existing POWER_ALL(false) path does,
> and then release the remaining required bits and the requested block.
> This also avoids changing ANALOG_TOTAL, false, which currently only
> clears DP_PHY_PD.

Right, in this case I'm leaning towards writing a small helper function
for AUX_BLOCK and CH[0-3]. Something like (not tested):

void analogix_dp_write_analog_power_down(struct analogix_dp_device *dp,
u32 phy_pd_addr,
u32 mask,
bool enable)
{
u32 reg = readl(dp->reg_base + phy_pd_addr);

if (enable)
reg |= mask;
else {
if (reg & DP_INC_BG)) {
reg |= AUX_PD | CH0_PD | CH1_PD | CH2_PD | CH3_PD;
writel(reg, dp->reg_base + phy_pd_addr);
usleep_range(10, 15);
reg &= ~DP_INC_BG;
writel(reg, dp->reg_base + phy_pd_addr);
usleep_range(10, 15);
}

if (reg & DP_PHY_PD) {
reg |= AUX_PD | CH0_PD | CH1_PD | CH2_PD | CH3_PD;
reg &= ~DP_PHY_PD;
}

reg &= ~mask;
}

writel(reg, dp->reg_base + phy_pd_addr);
}

Also, since DP_INC_BG looks functionally equivalent to DP_PHY_PD on
rockchip, I'm tented to make the behaviour consistent accross exynos and
rockchip. But I have no rockchip board to test.

>
> Thanks,
> Inki Dae
>

Best regards,
Alex.

> > + }
> > +
> > switch (block) {
> > case AUX_BLOCK:
> > if (dp->plat_data && analogix_dp_is_rockchip(dp->plat_data->dev_type))
> > @@ -263,7 +276,6 @@ void analogix_dp_set_analog_power_down(struct analogix_dp_device *dp,
> > else
> > mask = AUX_PD;
> >
> > - reg = readl(dp->reg_base + phy_pd_addr);
> > if (enable)
> > reg |= mask;
> > else
> > @@ -272,7 +284,6 @@ void analogix_dp_set_analog_power_down(struct analogix_dp_device *dp,
> > break;
> > case CH0_BLOCK:
> > mask = CH0_PD;
> > - reg = readl(dp->reg_base + phy_pd_addr);
> >
> > if (enable)
> > reg |= mask;
> > @@ -282,7 +293,6 @@ void analogix_dp_set_analog_power_down(struct analogix_dp_device *dp,
> > break;
> > case CH1_BLOCK:
> > mask = CH1_PD;
> > - reg = readl(dp->reg_base + phy_pd_addr);
> >
> > if (enable)
> > reg |= mask;
> > @@ -292,7 +302,6 @@ void analogix_dp_set_analog_power_down(struct analogix_dp_device *dp,
> > break;
> > case CH2_BLOCK:
> > mask = CH2_PD;
> > - reg = readl(dp->reg_base + phy_pd_addr);
> >
> > if (enable)
> > reg |= mask;
> > @@ -302,7 +311,6 @@ void analogix_dp_set_analog_power_down(struct analogix_dp_device *dp,
> > break;
> > case CH3_BLOCK:
> > mask = CH3_PD;
> > - reg = readl(dp->reg_base + phy_pd_addr);
> >
> > if (enable)
> > reg |= mask;
> > @@ -321,7 +329,6 @@ void analogix_dp_set_analog_power_down(struct analogix_dp_device *dp,
> > else
> > mask = DP_PHY_PD;
> >
> > - reg = readl(dp->reg_base + phy_pd_addr);
> > if (enable)
> > reg |= mask;
> > else
> >
> > --
> > 2.43.0
> >
> >
> >