Re: [PATCH v2 1/2] arm64: dts: ti: k3-am65: disable optional peripherals by default

Re: [PATCH v2 1/2] arm64: dts: ti: k3-am65: disable optional peripherals by default

[Nishanth Menon]

Rob, Tony, Arnd, SoC maintainers,

On 15:02-20220203, Matthias Schiffer wrote:
> All peripherals that require pinmuxing or other configuration to work
> should be disabled by default. Dependent DTS are adjusted accordingly.

https://lore.kernel.org/linux-arm-kernel/20201112183538.6805-1-nm@ti.com/
reversal all over again.

Is there a specific pattern we are intending to use here? Because, if we
are going down this path (which would be a major churn across multiple
downstream trees as well) - I'd rather have this as a documented
standard and not just a TI approach and will need to be done across all
K3 devices.

Are you aware of such a documented guideline, rather than "word of
mouth"? Maybe I have'nt looked deep enough, but checking..

> 
> The following nodes are now "disabled" according to dtx_diff and were not
> overridden to "okay", as they define no pinctrl:


> k3-am654-base-board:
> - mcu_i2c0
> - mcu_spi0..2
> - mcu_uart0
> - cal
> - main_i2c3
> - ehrpwm0..5
> - main_uart1..2
> - main_spi1..4
> 
> k3-am65-iot2050*:
> - mci_spi1..2
> - cal
> - ehrpwm0..5
> - main_spi0..4
> 
> Signed-off-by: Matthias Schiffer <matthias.schiffer@ew.tq-group.com>
> ---
[...]

> diff --git a/arch/arm64/boot/dts/ti/k3-am65-main.dtsi b/arch/arm64/boot/dts/ti/k3-am65-main.dtsi
> index ce8bb4a61011e..5aa425d1ba802 100644
> --- a/arch/arm64/boot/dts/ti/k3-am65-main.dtsi
> +++ b/arch/arm64/boot/dts/ti/k3-am65-main.dtsi
> @@ -64,6 +64,7 @@ serdes0: serdes@900000 {
>  		ti,serdes-clk = <&serdes0_clk>;

[...]

> @@ -1337,6 +1388,7 @@ icssg2_mdio: mdio@32400 {
>  			#address-cells = <1>;
>  			#size-cells = <0>;
>  			bus_freq = <1000000>;
> +			status = "disabled";
>  		};
>  	};
>  };
> diff --git a/arch/arm64/boot/dts/ti/k3-am65-mcu.dtsi b/arch/arm64/boot/dts/ti/k3-am65-mcu.dtsi
> index 8d592bf41d6f1..57ac3a493adbe 100644
> --- a/arch/arm64/boot/dts/ti/k3-am65-mcu.dtsi
> +++ b/arch/arm64/boot/dts/ti/k3-am65-mcu.dtsi
> @@ -22,11 +22,12 @@ phy_gmii_sel: phy@4040 {
>  
>  	mcu_uart0: serial@40a00000 {
>  		compatible = "ti,am654-uart";
> -			reg = <0x00 0x40a00000 0x00 0x100>;
> -			interrupts = <GIC_SPI 565 IRQ_TYPE_LEVEL_HIGH>;
> -			clock-frequency = <96000000>;
> -			current-speed = <115200>;
> -			power-domains = <&k3_pds 149 TI_SCI_PD_EXCLUSIVE>;
> +		reg = <0x00 0x40a00000 0x00 0x100>;
> +		interrupts = <GIC_SPI 565 IRQ_TYPE_LEVEL_HIGH>;
> +		clock-frequency = <96000000>;
> +		current-speed = <115200>;
> +		power-domains = <&k3_pds 149 TI_SCI_PD_EXCLUSIVE>;

When doing these kind of changes, do not include ancillary tab cleanups.
keep such cleanups separate patch.

> +		status = "disabled";

>  	};
>  
>  	mcu_ram: sram@41c00000 {
> @@ -46,6 +47,7 @@ mcu_i2c0: i2c@40b00000 {
>  		clock-names = "fck";
>  		clocks = <&k3_clks 114 1>;
>  		power-domains = <&k3_pds 114 TI_SCI_PD_EXCLUSIVE>;
> +		status = "disabled";
>  	};
>  
>  	mcu_spi0: spi@40300000 {
> @@ -56,6 +58,7 @@ mcu_spi0: spi@40300000 {
>  		power-domains = <&k3_pds 142 TI_SCI_PD_EXCLUSIVE>;
>  		#address-cells = <1>;
>  		#size-cells = <0>;
> +		status = "disabled";
>  	};
>  
-- 
Regards,
Nishanth Menon
Key (0xDDB5849D1736249D)/Fingerprint: F8A2 8693 54EB 8232 17A3  1A34 DDB5 849D 1736 249D

Re: [PATCH v2 1/2] arm64: dts: ti: k3-am65: disable optional peripherals by default

[Tony Lindgren]

Hi,

* Nishanth Menon <nm@ti.com> [220204 14:30]:
> Rob, Tony, Arnd, SoC maintainers,
> 
> On 15:02-20220203, Matthias Schiffer wrote:
> > All peripherals that require pinmuxing or other configuration to work
> > should be disabled by default. Dependent DTS are adjusted accordingly.

Disabling SoC internal devices by default is not a good policy. The
devices are available even if not pinned out. Disabling device by default
causes runtime PM to not work as the kernel will completely ignore the
disabled devices. And this means you add a dependency to some certain
version of a bootloader for PM to work.

Additionally tagging devices as disabled by default (and then again
re-enabling them in the board specific dts files) is just pointless
churn and bloat. See for example commit 12afc0cf8121 ("ARM: dts: Drop
pointless status changing for am3 musb") :)

If you really want to disable some devices for memory usage or other
reasons, do it in the board specific dts files.

> https://lore.kernel.org/linux-arm-kernel/20201112183538.6805-1-nm@ti.com/
> reversal all over again.
> 
> Is there a specific pattern we are intending to use here? Because, if we
> are going down this path (which would be a major churn across multiple
> downstream trees as well) - I'd rather have this as a documented
> standard and not just a TI approach and will need to be done across all
> K3 devices.
> 
> Are you aware of such a documented guideline, rather than "word of
> mouth"? Maybe I have'nt looked deep enough, but checking..

For SoCs that don't implement runtime PM the policy can be different
without causing any harm. But for any SoCs implementing runtime PM, an
unknown state from the bootloader is not going to work.

Regards,

Tony

Re: [PATCH v2 1/2] arm64: dts: ti: k3-am65: disable optional peripherals by default

[Matthias Schiffer]

On Mon, 2022-02-07 at 08:54 +0200, Tony Lindgren wrote:
> Hi,
> 
> * Nishanth Menon <nm@ti.com> [220204 14:30]:
> > Rob, Tony, Arnd, SoC maintainers,
> > 
> > On 15:02-20220203, Matthias Schiffer wrote:
> > > All peripherals that require pinmuxing or other configuration to
> > > work
> > > should be disabled by default. Dependent DTS are adjusted
> > > accordingly.
> 
> Disabling SoC internal devices by default is not a good policy. The
> devices are available even if not pinned out. Disabling device by
> default
> causes runtime PM to not work as the kernel will completely ignore
> the
> disabled devices. And this means you add a dependency to some certain
> version of a bootloader for PM to work.
> 
> Additionally tagging devices as disabled by default (and then again
> re-enabling them in the board specific dts files) is just pointless
> churn and bloat. See for example commit 12afc0cf8121 ("ARM: dts: Drop
> pointless status changing for am3 musb") :)
> 
> If you really want to disable some devices for memory usage or other
> reasons, do it in the board specific dts files.

Attempting to use unmuxed peripherals often leads to ugly errors - I2C
without pullups appearing busy, UARTs in endless break condition, ...
Such errors are often seen as defects in hardware or software by people
who aren't familiar with the internals.

I can see the issue with bootloaders leaving peripherals in an unknown
state, but I'm not happy with keeping such devices enabled in the
kernel either.

Generally I think that it's a bootloader's responsiblity to disable
unneeded devices - the kernel may not even have a driver for some
peripherals, leading to the same behaviour as a "disabled" status. For
this reason I believe that it should always be okay to set unneeded
devices to "disabled", and it should be considered a safe default.

I'm not sure what the consensus on these issues is. I'm more familiar
with NXP's i.MX and Layerscape SoCs, where it's common to have all
muxable peripherals set to "disabled" in the base DTSI, and a quick
grep through a few dts directories gives me the impression that this is
the case for most other vendors as well.

Regards,
Matthias


> 
> > https://lore.kernel.org/linux-arm-kernel/20201112183538.6805-1-nm@ti.com/
> > reversal all over again.
> > 
> > Is there a specific pattern we are intending to use here? Because,
> > if we
> > are going down this path (which would be a major churn across
> > multiple
> > downstream trees as well) - I'd rather have this as a documented
> > standard and not just a TI approach and will need to be done across
> > all
> > K3 devices.
> > 
> > Are you aware of such a documented guideline, rather than "word of
> > mouth"? Maybe I have'nt looked deep enough, but checking..
> 
> For SoCs that don't implement runtime PM the policy can be different
> without causing any harm. But for any SoCs implementing runtime PM,
> an
> unknown state from the bootloader is not going to work.
> 
> Regards,
> 
> Tony

Re: [PATCH v2 1/2] arm64: dts: ti: k3-am65: disable optional peripherals by default

[Tony Lindgren]

* Matthias Schiffer <matthias.schiffer@ew.tq-group.com> [220207 08:45]:
> Generally I think that it's a bootloader's responsiblity to disable
> unneeded devices - the kernel may not even have a driver for some
> peripherals, leading to the same behaviour as a "disabled" status. For
> this reason I believe that it should always be okay to set unneeded
> devices to "disabled", and it should be considered a safe default.

Not possible, think kexec for example :) How would the previous kernel
even know what to disable if Linux has no idea about the devices?

If there are issues you're seeing, it's likely a bug in some of the
device drivers for not checking for the necessary resources like
pinctrl for i2c lines.

> I'm not sure what the consensus on these issues is. I'm more familiar
> with NXP's i.MX and Layerscape SoCs, where it's common to have all
> muxable peripherals set to "disabled" in the base DTSI, and a quick
> grep through a few dts directories gives me the impression that this is
> the case for most other vendors as well.

This approach only works for SoCs that don't need the kernel to idle
devices for runtime PM.

Regards,

Tony

Re: [PATCH v2 1/2] arm64: dts: ti: k3-am65: disable optional peripherals by default

[Matthias Schiffer]

On Mon, 2022-02-07 at 13:25 +0200, Tony Lindgren wrote:
> * Matthias Schiffer <matthias.schiffer@ew.tq-group.com> [220207
> 08:45]:
> > Generally I think that it's a bootloader's responsiblity to disable
> > unneeded devices - the kernel may not even have a driver for some
> > peripherals, leading to the same behaviour as a "disabled" status.
> > For
> > this reason I believe that it should always be okay to set unneeded
> > devices to "disabled", and it should be considered a safe default.
> 
> Not possible, think kexec for example :) How would the previous
> kernel
> even know what to disable if Linux has no idea about the devices?

Well, optimally, bootloader and all kernels would agree on the devices
that are actually available, but I get your point.

> 
> If there are issues you're seeing, it's likely a bug in some of the
> device drivers for not checking for the necessary resources like
> pinctrl for i2c lines.

I don't think it's common for individual drivers to care about pinctrl
unless switching between different pin settings is required at runtime.
Many drivers can be used on different hardware, some of which may
require pinmuxing, while others don't.

Also, what is the expected behavior of a driver that is probed for an
unusable device? Wouldn't this require some as-of-yet nonexisting
status between "okay" and "disabled" that conveys something like "probe
this device, initialize (and disable) PM, but don't register anything",
so no unusable devices become visible to userspace (and possibly other
kernel drivers)?

> 
> > I'm not sure what the consensus on these issues is. I'm more
> > familiar
> > with NXP's i.MX and Layerscape SoCs, where it's common to have all
> > muxable peripherals set to "disabled" in the base DTSI, and a quick
> > grep through a few dts directories gives me the impression that
> > this is
> > the case for most other vendors as well.
> 
> This approach only works for SoCs that don't need the kernel to idle
> devices for runtime PM.

I'm pretty sure that most modern SoCs I looked at have runtime PM, and
it is simply expected that unusable devices are never enabled in the
first place, so there is no need for the kernel to know about them.

Regards,
Matthias




> Regards,
> 
> Tony

Re: [PATCH v2 1/2] arm64: dts: ti: k3-am65: disable optional peripherals by default

[Tony Lindgren]

* Matthias Schiffer <matthias.schiffer@ew.tq-group.com> [220208 10:53]:
> On Mon, 2022-02-07 at 13:25 +0200, Tony Lindgren wrote:
> > * Matthias Schiffer <matthias.schiffer@ew.tq-group.com> [220207
> > 08:45]:
> > > Generally I think that it's a bootloader's responsiblity to disable
> > > unneeded devices - the kernel may not even have a driver for some
> > > peripherals, leading to the same behaviour as a "disabled" status.
> > > For
> > > this reason I believe that it should always be okay to set unneeded
> > > devices to "disabled", and it should be considered a safe default.
> > 
> > Not possible, think kexec for example :) How would the previous
> > kernel
> > even know what to disable if Linux has no idea about the devices?
> 
> Well, optimally, bootloader and all kernels would agree on the devices
> that are actually available, but I get your point.
> 
> > 
> > If there are issues you're seeing, it's likely a bug in some of the
> > device drivers for not checking for the necessary resources like
> > pinctrl for i2c lines.
> 
> I don't think it's common for individual drivers to care about pinctrl
> unless switching between different pin settings is required at runtime.
> Many drivers can be used on different hardware, some of which may
> require pinmuxing, while others don't.

Yeah that's true, some configurations only do pin muxing in the
bootloader. So pins are not a good criteria for devicetree status enabled
for when the device is operational.

Probably a better criteria for devicetree "operational" status is the
device can be clocked and configured or idled. Some devices like GPUs
can render to memory with no external pin configuration for example.

Following Linux running on a PC analogy.. If ACPI has some device that
causes driver warnings on Linux boot, do we patch the ACPI table and
pretend the device does not exist? Or do we patch the device driver to
deal with the random buggy bootloader state for the device? :)

> Also, what is the expected behavior of a driver that is probed for an
> unusable device? Wouldn't this require some as-of-yet nonexisting
> status between "okay" and "disabled" that conveys something like "probe
> this device, initialize (and disable) PM, but don't register anything",
> so no unusable devices become visible to userspace (and possibly other
> kernel drivers)?

I did some experimental patches several years ago to add devicetree
status for incomplete, but eventually came to the conclusion that it
was not really needed. Feel free to revisit that if you have the
spare cycles :)

Having the drivers check for the resources like clocks and then just
idle the device after probe solved the issues I was seeing for warnings
and kexec. In some cases the device may need to be reset or at least
properly reconfigured in the probe as the state can be unknown from the
bootloader. That's about all there is to it. Sure you could save some
memory with less instances for some devices, so maybe the status =
"incomplete" could be used to do the trick for that.

> > > I'm not sure what the consensus on these issues is. I'm more
> > > familiar
> > > with NXP's i.MX and Layerscape SoCs, where it's common to have all
> > > muxable peripherals set to "disabled" in the base DTSI, and a quick
> > > grep through a few dts directories gives me the impression that
> > > this is
> > > the case for most other vendors as well.
> > 
> > This approach only works for SoCs that don't need the kernel to idle
> > devices for runtime PM.
> 
> I'm pretty sure that most modern SoCs I looked at have runtime PM, and
> it is simply expected that unusable devices are never enabled in the
> first place, so there is no need for the kernel to know about them.

Yeah well that assumption is the difference in getting runtime PM to
work in a sane way across multiple SoCs and devices :)

Devices tagged with status = "disabled" are completely ignored by the
kernel. Interconnect and bus related code may not know the details on
how to reset and idle the child devices. Relying on firmware to do the
reset and idle of unused devices may be too generic, can be buggy, and
probably depends on the firmware revision.

Regards,

Tony

Re: [PATCH v2 1/2] arm64: dts: ti: k3-am65: disable optional peripherals by default

[Matthias Schiffer]

On Tue, 2022-02-08 at 13:52 +0200, Tony Lindgren wrote:
> * Matthias Schiffer <matthias.schiffer@ew.tq-group.com> [220208
> 10:53]:
> > On Mon, 2022-02-07 at 13:25 +0200, Tony Lindgren wrote:
> > > * Matthias Schiffer <matthias.schiffer@ew.tq-group.com> [220207
> > > 08:45]:
> > > > Generally I think that it's a bootloader's responsiblity to
> > > > disable
> > > > unneeded devices - the kernel may not even have a driver for
> > > > some
> > > > peripherals, leading to the same behaviour as a "disabled"
> > > > status.
> > > > For
> > > > this reason I believe that it should always be okay to set
> > > > unneeded
> > > > devices to "disabled", and it should be considered a safe
> > > > default.
> > > 
> > > Not possible, think kexec for example :) How would the previous
> > > kernel
> > > even know what to disable if Linux has no idea about the devices?
> > 
> > Well, optimally, bootloader and all kernels would agree on the
> > devices
> > that are actually available, but I get your point.
> > 
> > > If there are issues you're seeing, it's likely a bug in some of
> > > the
> > > device drivers for not checking for the necessary resources like
> > > pinctrl for i2c lines.
> > 
> > I don't think it's common for individual drivers to care about
> > pinctrl
> > unless switching between different pin settings is required at
> > runtime.
> > Many drivers can be used on different hardware, some of which may
> > require pinmuxing, while others don't.
> 
> Yeah that's true, some configurations only do pin muxing in the
> bootloader. So pins are not a good criteria for devicetree status
> enabled
> for when the device is operational.
> 
> Probably a better criteria for devicetree "operational" status is the
> device can be clocked and configured or idled. Some devices like GPUs
> can render to memory with no external pin configuration for example.
> 

I don't think any properties currently exist that could or should be
used to decide whether a device is operational. Clocks etc. are usually
internal to the SoC and thus already set in the SoC DTSI. Pins and
power supplies may be specific to a mainboard, but can also be
optional. Whether an I2C bus can be operational may solely depend on
whether external pullups are connected to the pins or not.

The idea of an "incomplete" status like you mention below sounds better
to me. I also thought about adding something like a "probe_disabled()"
that is called instead of probe() for "disabled" devices, but I assume
that would cause a boot time penalty on systems that have many
"disabled" devices and don't actually need this...


> Following Linux running on a PC analogy.. If ACPI has some device
> that
> causes driver warnings on Linux boot, do we patch the ACPI table and
> pretend the device does not exist? Or do we patch the device driver
> to
> deal with the random buggy bootloader state for the device? :)
> 
> > Also, what is the expected behavior of a driver that is probed for
> > an
> > unusable device? Wouldn't this require some as-of-yet nonexisting
> > status between "okay" and "disabled" that conveys something like
> > "probe
> > this device, initialize (and disable) PM, but don't register
> > anything",
> > so no unusable devices become visible to userspace (and possibly
> > other
> > kernel drivers)?
> 
> I did some experimental patches several years ago to add devicetree
> status for incomplete, but eventually came to the conclusion that it
> was not really needed. Feel free to revisit that if you have the
> spare cycles :)
> 
> Having the drivers check for the resources like clocks and then just
> idle the device after probe solved the issues I was seeing for
> warnings
> and kexec. In some cases the device may need to be reset or at least
> properly reconfigured in the probe as the state can be unknown from
> the
> bootloader. That's about all there is to it. Sure you could save some
> memory with less instances for some devices, so maybe the status =
> "incomplete" could be used to do the trick for that.

I don't really care about memory usage. What I do care about is that
incorrect userspace usage doesn't cause ugly kernel warnings (for
example timeouts for i2cdetect on unmuxed bus) when we can avoid it,
because such issues always lead to support requests.

Not being able to hide non-operational devices from userspace feels
like a regression from older hardware.

> 
> > > > I'm not sure what the consensus on these issues is. I'm more
> > > > familiar
> > > > with NXP's i.MX and Layerscape SoCs, where it's common to have
> > > > all
> > > > muxable peripherals set to "disabled" in the base DTSI, and a
> > > > quick
> > > > grep through a few dts directories gives me the impression that
> > > > this is
> > > > the case for most other vendors as well.
> > > 
> > > This approach only works for SoCs that don't need the kernel to
> > > idle
> > > devices for runtime PM.
> > 
> > I'm pretty sure that most modern SoCs I looked at have runtime PM,
> > and
> > it is simply expected that unusable devices are never enabled in
> > the
> > first place, so there is no need for the kernel to know about them.
> 
> Yeah well that assumption is the difference in getting runtime PM to
> work in a sane way across multiple SoCs and devices :)
> 
> Devices tagged with status = "disabled" are completely ignored by the
> kernel. Interconnect and bus related code may not know the details on
> how to reset and idle the child devices. Relying on firmware to do
> the
> reset and idle of unused devices may be too generic, can be buggy,
> and
> probably depends on the firmware revision.

Well, so far it seems like the `status = "disabled"` is just being
pushed from the SoC DTSIs to the board DTSs on TI hardware. For the
AM64 platform (which is fairly similar to the AM65), both mainboards
that currently exist disable unused UARTs, I2C/SPI busses, PWMs, ...
(Some of these might be disabled to make them usable from the R5/M4
cores, but I don't think that the case for all of them - "reserved"
would be more appropriate than "disabled" in these cases anyways)

AFAICT, disabling non-operatational devices in the board DTS instead of
the SoC DTSI is worse than the alternatives in every way:

- Verbose board DTS: You have to think about all the devices that exist
in the SoC, not just the ones you want to use
- Adding new nodes without `status = "disabled" to SoC DTSI can
potentially cause issues on dependent boards
- It doesn't solve the issues that not having `status = "disabled"` in
the DTSI is supposed to solve

Regards,
Matthias


> 
> Regards,
> 
> Tony

Re: [PATCH v2 1/2] arm64: dts: ti: k3-am65: disable optional peripherals by default

[Tony Lindgren]

Hi,

* Matthias Schiffer <matthias.schiffer@ew.tq-group.com> [220228 10:29]:
> AFAICT, disabling non-operatational devices in the board DTS instead of
> the SoC DTSI is worse than the alternatives in every way:
> 
> - Verbose board DTS: You have to think about all the devices that exist
> in the SoC, not just the ones you want to use
> - Adding new nodes without `status = "disabled" to SoC DTSI can
> potentially cause issues on dependent boards
> - It doesn't solve the issues that not having `status = "disabled"` in
> the DTSI is supposed to solve

My preference is the least amount of tinkering in the dts files
naturally :) It really does not matter if the extra dts churn is to
enable or disable devices, it should not be needed at all.

To summarize, my main point really is the following:

There should not be any need to tag the SoC internal devices with anything
in the dts files. The device drivers should be able to just deal with the
situation. IMO devices should be tagged with disabled or reserved when
they are not accessible for example because of being used by secure mode
for example. If the the status needs to be set to anything, it really is
a symptom of incomplete handling somewhere.

Regards,

Tony

Re: [PATCH v2 1/2] arm64: dts: ti: k3-am65: disable optional peripherals by default

[Matthias Schiffer]

On Wed, 2022-03-09 at 11:11 +0200, Tony Lindgren wrote:
> Hi,
> 
> * Matthias Schiffer <matthias.schiffer@ew.tq-group.com> [220228
> 10:29]:
> > AFAICT, disabling non-operatational devices in the board DTS
> > instead of
> > the SoC DTSI is worse than the alternatives in every way:
> > 
> > - Verbose board DTS: You have to think about all the devices that
> > exist
> > in the SoC, not just the ones you want to use
> > - Adding new nodes without `status = "disabled" to SoC DTSI can
> > potentially cause issues on dependent boards
> > - It doesn't solve the issues that not having `status = "disabled"`
> > in
> > the DTSI is supposed to solve
> 
> My preference is the least amount of tinkering in the dts files
> naturally :) It really does not matter if the extra dts churn is to
> enable or disable devices, it should not be needed at all.
> 
> To summarize, my main point really is the following:
> 
> There should not be any need to tag the SoC internal devices with
> anything
> in the dts files. The device drivers should be able to just deal with
> the
> situation. IMO devices should be tagged with disabled or reserved
> when
> they are not accessible for example because of being used by secure
> mode
> for example. If the the status needs to be set to anything, it really
> is
> a symptom of incomplete handling somewhere.
> 
> Regards,
> 
> Tony


Hi Tony,

while I agree that it would be great if drivers could just detect when
hardware is not available, this is simply not how most drivers work -
when you instantiate the driver via a non-disabled(/reserved/...) DT
node, the driver expects a usable device.

Especially for busses like I2C, there is no way for a driver to
reliably detect whether the bus is usable or not. (There are several
states that can't really be distinguished: Is pinmuxing missing, or
does the device not need any muxing? Is a line low because it is not
actually connected to anything, or is there another master currently
using the bus, or is the bus stuck due to a faulty device?)

Which is why it is the convention for SoC DTSI files disable nodes for
devices that may be unusable. Taking UARTs as an example, a quick grep
for "serial@" nodes in arch/arm/boot/dts/*.dtsi and
arch/arm64/boot/dts/*/*.dtsi shows that the vast majority of these
nodes is disabled by default.

Regards,
Matthias

Re: [PATCH v2 1/2] arm64: dts: ti: k3-am65: disable optional peripherals by default

[Tony Lindgren]

* Matthias Schiffer <matthias.schiffer@ew.tq-group.com> [220309 11:09]:
> while I agree that it would be great if drivers could just detect when
> hardware is not available, this is simply not how most drivers work -
> when you instantiate the driver via a non-disabled(/reserved/...) DT
> node, the driver expects a usable device.
> 
> Especially for busses like I2C, there is no way for a driver to
> reliably detect whether the bus is usable or not. (There are several
> states that can't really be distinguished: Is pinmuxing missing, or
> does the device not need any muxing? Is a line low because it is not
> actually connected to anything, or is there another master currently
> using the bus, or is the bus stuck due to a faulty device?)

Well how about set only the problem devices with status = "disabled"
with a proper comment in the SoC dtsi file?

See for example what has been done in arch/arm64/boot/dts/apple that
has been pretty widely reviewed and done with a good taste :)

Not sure what can be done to idle the unused devices in the disabled
case though, maybe some firmware call to disable all unclaimed devices
could be done if it does not exist already. The firmware may not have
the capability to idle devices that need firmware loaded to idle them
for example though.

Regards,

Tony