[1/2] xhci: Introduce "disable-usb3" DT property/quirk

  Some systems may have xHCI controllers that enumerate USB 3.0 ports, but
these ports nevertheless cannot be used. Perhaps enabling them triggers a
hardware bug, or perhaps they simply aren't connected and it would be
confusing to the user to see an unusable USB 3.0 rhub show up -- whatever
the case may be, it's reasonable to want to disable these ports.

Add a DT property (and associated quirk) to the xHCI driver that skips
over (i.e. ignores and doesn't initialize) any USB 3.0 ports discovered
during driver initialization.

Signed-off-by: Sam Edwards <CFSworks@gmail.com>
---
 Documentation/devicetree/bindings/usb/usb-xhci.yaml | 4 ++++
 drivers/usb/host/xhci-mem.c                         | 4 ++++
 drivers/usb/host/xhci-plat.c                        | 3 +++
 drivers/usb/host/xhci.h                             | 1 +
 4 files changed, 12 insertions(+)
  

On 08/12/2023 22:04, Sam Edwards wrote:
> Some systems may have xHCI controllers that enumerate USB 3.0 ports, but
> these ports nevertheless cannot be used. Perhaps enabling them triggers a
> hardware bug, or perhaps they simply aren't connected and it would be
> confusing to the user to see an unusable USB 3.0 rhub show up -- whatever
> the case may be, it's reasonable to want to disable these ports.
> 
> Add a DT property (and associated quirk) to the xHCI driver that skips
> over (i.e. ignores and doesn't initialize) any USB 3.0 ports discovered
> during driver initialization.
> 
> Signed-off-by: Sam Edwards <CFSworks@gmail.com>
> ---
>  Documentation/devicetree/bindings/usb/usb-xhci.yaml | 4 ++++

Bindings are always separate patches.

Please do not sneak in properties without DT review.

Please use scripts/get_maintainers.pl to get a list of necessary people
and lists to CC. It might happen, that command when run on an older
kernel, gives you outdated entries. Therefore please be sure you base
your patches on recent Linux kernel.

You missed at least devicetree list (maybe more), so this won't be
tested by automated tooling. Performing review on untested code might be
a waste of time, thus I will skip this patch entirely till you follow
the process allowing the patch to be tested.

Please kindly resend and include all necessary To/Cc entries.

Best regards,
Krzysztof
  

On 12/9/23 06:53, Krzysztof Kozlowski wrote:
> On 08/12/2023 22:04, Sam Edwards wrote:
>> Some systems may have xHCI controllers that enumerate USB 3.0 ports, but
>> these ports nevertheless cannot be used. Perhaps enabling them triggers a
>> hardware bug, or perhaps they simply aren't connected and it would be
>> confusing to the user to see an unusable USB 3.0 rhub show up -- whatever
>> the case may be, it's reasonable to want to disable these ports.
>>
>> Add a DT property (and associated quirk) to the xHCI driver that skips
>> over (i.e. ignores and doesn't initialize) any USB 3.0 ports discovered
>> during driver initialization.
>>
>> Signed-off-by: Sam Edwards <CFSworks@gmail.com>
>> ---
>>   Documentation/devicetree/bindings/usb/usb-xhci.yaml | 4 ++++

Hi Krzysztof,

> 
> Bindings are always separate patches.
> 
> Please do not sneak in properties without DT review.
> 

It makes sense that the new property should be introduced in a separate 
patch. I'll ensure that is the case in v2. (If there is one -- see below.)

> Please use scripts/get_maintainers.pl to get a list of necessary people
> and lists to CC. It might happen, that command when run on an older
> kernel, gives you outdated entries. Therefore please be sure you base
> your patches on recent Linux kernel.
> 
> You missed at least devicetree list (maybe more), so this won't be
> tested by automated tooling.

I have tried (and failed) to find the documentation for this 
linux-devicetree bot. Do you have the link? In particular, I'd like to 
ensure that patch 2/2 (the one that actually changes established 
behavior) is tested sufficiently thoroughly.

> Performing review on untested code might be
> a waste of time, thus I will skip this patch entirely till you follow
> the process allowing the patch to be tested.

That's fine; this patch has just failed review anyway (due to the new 
property not being introduced in a separate patch), and I'll need to 
prepare and send a v2 to proceed. However as I mentioned in the cover, 
this is a semi-RFC. I haven't discussed the overall idea with anyone 
yet, so to avoid wasting my own time, I need to give the USB folks ample 
opportunity to object to the proposed changes or suggest improvements 
before investing more effort in refining the patchset.

As of now, I'm only seeking commentary, not formal review. I'd 
appreciate any insights on the approach I've taken and whether there are 
any potential challenges or alternatives that haven't been explored yet. 
Therefore, I'll hold off on CC-ing linux-devicetree at this stage to 
keep the focus on the broader concept, and will loop them in (with any 
other recipients as appropriate) for v2 when (and if!) there's consensus 
here on linux-usb that the general direction is worth pursuing.

> 
> Please kindly resend and include all necessary To/Cc entries.
> 
> Best regards,
> Krzysztof
> 

Happy Saturday,
Sam
  

On 09/12/2023 20:26, Sam Edwards wrote:
>> Performing review on untested code might be
>> a waste of time, thus I will skip this patch entirely till you follow
>> the process allowing the patch to be tested.
> 
> That's fine; this patch has just failed review anyway (due to the new 
> property not being introduced in a separate patch), and I'll need to 
> prepare and send a v2 to proceed. However as I mentioned in the cover, 
> this is a semi-RFC. I haven't discussed the overall idea with anyone 
> yet, so to avoid wasting my own time, I need to give the USB folks ample 

It does not really explain why you did not Cc some of the maintainers.
If this is a RFC, even though not marked as such in subject prefix, then
I guess all maintainers should be involved for comments.

Best regards,
Krzysztof
  

On 12/10/23 04:10, Krzysztof Kozlowski wrote:
> It does not really explain why you did not Cc some of the maintainers.
> If this is a RFC, even though not marked as such in subject prefix, then
> I guess all maintainers should be involved for comments.

Hi Krzysztof,

More simply put: I don't want to waste anyone's time by seeking 
review/commentary from linux-devicetree on a property enabling a feature 
that someone here on linux-usb might soon deem (summarily) untenable.

I agree fully that I need (and want) the DT reviewers' sign-off on this 
new property, and I'll get more serious about that (and other things) in 
v2. I just need to know if there's going to be a v2 first.

Warm regards,
Sam
  

Am Freitag, 8. Dezember 2023, 22:04:57 CET schrieb Sam Edwards:
> Some systems may have xHCI controllers that enumerate USB 3.0 ports, but
> these ports nevertheless cannot be used. Perhaps enabling them triggers a
> hardware bug, or perhaps they simply aren't connected and it would be
> confusing to the user to see an unusable USB 3.0 rhub show up -- whatever
> the case may be, it's reasonable to want to disable these ports.
> 
> Add a DT property (and associated quirk) to the xHCI driver that skips
> over (i.e. ignores and doesn't initialize) any USB 3.0 ports discovered
> during driver initialization.
> 
> Signed-off-by: Sam Edwards <CFSworks@gmail.com>

I'm very much unsure, where the line goes between hw-quirk and
dt-is-not-a-configuration-space - in this specific instance.

DT is meant to describe the actual hardware present and not how
any operating system supports it.

So having that usb3phy present in the kernel - even if only in
a more limited form as you describe would be my preference.


But for a short-term thing, the usb3-phy in the binding is optional, so
so you could "just" deduce the no-usb3 state in your code from its
absence from the dt-node?


Heiko



>  Documentation/devicetree/bindings/usb/usb-xhci.yaml | 4 ++++
>  drivers/usb/host/xhci-mem.c                         | 4 ++++
>  drivers/usb/host/xhci-plat.c                        | 3 +++
>  drivers/usb/host/xhci.h                             | 1 +
>  4 files changed, 12 insertions(+)
> 
> diff --git a/Documentation/devicetree/bindings/usb/usb-xhci.yaml b/Documentation/devicetree/bindings/usb/usb-xhci.yaml
> index 180a261c3e8f..8a64e747260a 100644
> --- a/Documentation/devicetree/bindings/usb/usb-xhci.yaml
> +++ b/Documentation/devicetree/bindings/usb/usb-xhci.yaml
> @@ -25,6 +25,10 @@ properties:
>      description: Set if the controller has broken port disable mechanism
>      type: boolean
>  
> +  disable-usb3:
> +    description: Ignore (don't initialize, don't use) USB3 ports
> +    type: boolean
> +
>    imod-interval-ns:
>      description: Interrupt moderation interval
>      default: 5000
> diff --git a/drivers/usb/host/xhci-mem.c b/drivers/usb/host/xhci-mem.c
> index 0a37f0d511cf..bf8fcab626e4 100644
> --- a/drivers/usb/host/xhci-mem.c
> +++ b/drivers/usb/host/xhci-mem.c
> @@ -1968,6 +1968,10 @@ static void xhci_add_in_port(struct xhci_hcd *xhci, unsigned int num_ports,
>  	minor_revision = XHCI_EXT_PORT_MINOR(temp);
>  
>  	if (major_revision == 0x03) {
> +		/* Ignore USB3 ports entirely if USB3 support is disabled. */
> +		if (xhci->quirks & XHCI_DISABLE_USB3)
> +			return;
> +
>  		rhub = &xhci->usb3_rhub;
>  		/*
>  		 * Some hosts incorrectly use sub-minor version for minor
> diff --git a/drivers/usb/host/xhci-plat.c b/drivers/usb/host/xhci-plat.c
> index b93161374293..75285fb5bbbc 100644
> --- a/drivers/usb/host/xhci-plat.c
> +++ b/drivers/usb/host/xhci-plat.c
> @@ -249,6 +249,9 @@ int xhci_plat_probe(struct platform_device *pdev, struct device *sysdev, const s
>  		if (device_property_read_bool(tmpdev, "quirk-broken-port-ped"))
>  			xhci->quirks |= XHCI_BROKEN_PORT_PED;
>  
> +		if (device_property_read_bool(tmpdev, "disable-usb3"))
> +			xhci->quirks |= XHCI_DISABLE_USB3;
> +
>  		device_property_read_u32(tmpdev, "imod-interval-ns",
>  					 &xhci->imod_interval);
>  	}
> diff --git a/drivers/usb/host/xhci.h b/drivers/usb/host/xhci.h
> index 5df370482521..c53fbeea478f 100644
> --- a/drivers/usb/host/xhci.h
> +++ b/drivers/usb/host/xhci.h
> @@ -1906,6 +1906,7 @@ struct xhci_hcd {
>  #define XHCI_RESET_TO_DEFAULT	BIT_ULL(44)
>  #define XHCI_ZHAOXIN_TRB_FETCH	BIT_ULL(45)
>  #define XHCI_ZHAOXIN_HOST	BIT_ULL(46)
> +#define XHCI_DISABLE_USB3	BIT_ULL(47)
>  
>  	unsigned int		num_active_eps;
>  	unsigned int		limit_active_eps;
>
  

On 12/12/23 12:31, Heiko Stuebner wrote:
> Am Freitag, 8. Dezember 2023, 22:04:57 CET schrieb Sam Edwards:
>> Some systems may have xHCI controllers that enumerate USB 3.0 ports, but
>> these ports nevertheless cannot be used. Perhaps enabling them triggers a
>> hardware bug, or perhaps they simply aren't connected and it would be
>> confusing to the user to see an unusable USB 3.0 rhub show up -- whatever
>> the case may be, it's reasonable to want to disable these ports.
>>
>> Add a DT property (and associated quirk) to the xHCI driver that skips
>> over (i.e. ignores and doesn't initialize) any USB 3.0 ports discovered
>> during driver initialization.
>>
>> Signed-off-by: Sam Edwards <CFSworks@gmail.com>

Ahoy Heiko,

Glad to have you in the thread!

> I'm very much unsure, where the line goes between hw-quirk and
> dt-is-not-a-configuration-space - in this specific instance.
> 
> DT is meant to describe the actual hardware present and not how
> any operating system supports it.

I can plumb this quirk through something other than DT if there's an 
objection to it; however, the intended meaning of this property is 
indeed hardware-description and not OS-configuration: "This xHC does NOT 
support USB3. Do not let it tell you otherwise by enumerating USB3 
ports; it is mistaken. Attempting to enable the USB3 'ports' will at 
worst provoke a hardware bug and at best register a usb3 rhub for 
something that doesn't physically exist."

For v2, I should probably rename the property from something 
"imperative" ("disable-usb3;") to something "declarative," like 
"no-usb3;" or "usb3-unconnected;" in order to reflect this intended 
meaning better. The linux-devicetree reviewers (which I will be looping 
in with v2) will probably prefer a name like that anyway.

> So having that usb3phy present in the kernel - even if only in
> a more limited form as you describe would be my preference.
> 
> 
> But for a short-term thing, the usb3-phy in the binding is optional, so
> so you could "just" deduce the no-usb3 state in your code from its
> absence from the dt-node?

Ah, so while there is a short-term benefit, this is actually meant to be 
long-term: on the Turing RK1 PCB itself, USB-DWC3 #0 is only wired for 
USB 2.0, not USB 3.0. The corresponding USBDP PHY pins are connected to 
(dedicated) DisplayPort pins.

This means there are 4 options:
1. Proceed without enabling that USB-DWC3 at all, giving up USB 2.0.
2. Enable the USBDP in USB3 mode, gaining USB 2.0 but giving up DP.
3. Enable the USBDP in DP mode, having the USB3-PIPE backend
    initialized (but disconnected from the frontend), as a "pacifier"
    for the DWC3 so it doesn't start misbehaving.
4. Disable the USB3 ports on the xHC side, by...
    a. Telling the xHCI driver that it shouldn't enable USB3 ports.
    b. Configuring the DWC3 not to enable USB3
       (see the RK3588 TRM for usb3otg0_host_u3_port_disable)
    c. Configuring the DWC3's number of USB3 ports
       (usb3otg0_host_num_u3_port)

Users won't like #1 or #2, and may be confused by #3 due to the dummy 
port that shows up. Options #2 and #3 aren't possible until a USBDP 
driver lands (though you're right that this is only a short-term 
concern). Option 4a is what I'm currently trying to do because I believe 
having physically-inaccessible USB3 ports show up in Linux is 
undesirable in any case, but I can switch to 4b/4c if we want something 
more SoC-specific.

So (as far as RK3588 is concerned) with option(s) 4(a/b/c), while this 
does let USB2.0 work before the USBDP driver lands, the "real" benefit 
is that it requires neither a dummy port nor powering up a block of the 
chip (the PHY's USB3-specific block) that isn't actually usable/useful 
because it doesn't connect to a USB 3.0 socket/device.

Though if that still doesn't sound appealing, I'm happy to change up the 
approach -- I guess the question then is whether it's because you *like* 
option 3, or *dislike* option 4.

Cheers,
Sam

> 
> 
> Heiko
> 
> 
> 
>>   Documentation/devicetree/bindings/usb/usb-xhci.yaml | 4 ++++
>>   drivers/usb/host/xhci-mem.c                         | 4 ++++
>>   drivers/usb/host/xhci-plat.c                        | 3 +++
>>   drivers/usb/host/xhci.h                             | 1 +
>>   4 files changed, 12 insertions(+)
>>
>> diff --git a/Documentation/devicetree/bindings/usb/usb-xhci.yaml b/Documentation/devicetree/bindings/usb/usb-xhci.yaml
>> index 180a261c3e8f..8a64e747260a 100644
>> --- a/Documentation/devicetree/bindings/usb/usb-xhci.yaml
>> +++ b/Documentation/devicetree/bindings/usb/usb-xhci.yaml
>> @@ -25,6 +25,10 @@ properties:
>>       description: Set if the controller has broken port disable mechanism
>>       type: boolean
>>   
>> +  disable-usb3:
>> +    description: Ignore (don't initialize, don't use) USB3 ports
>> +    type: boolean
>> +
>>     imod-interval-ns:
>>       description: Interrupt moderation interval
>>       default: 5000
>> diff --git a/drivers/usb/host/xhci-mem.c b/drivers/usb/host/xhci-mem.c
>> index 0a37f0d511cf..bf8fcab626e4 100644
>> --- a/drivers/usb/host/xhci-mem.c
>> +++ b/drivers/usb/host/xhci-mem.c
>> @@ -1968,6 +1968,10 @@ static void xhci_add_in_port(struct xhci_hcd *xhci, unsigned int num_ports,
>>   	minor_revision = XHCI_EXT_PORT_MINOR(temp);
>>   
>>   	if (major_revision == 0x03) {
>> +		/* Ignore USB3 ports entirely if USB3 support is disabled. */
>> +		if (xhci->quirks & XHCI_DISABLE_USB3)
>> +			return;
>> +
>>   		rhub = &xhci->usb3_rhub;
>>   		/*
>>   		 * Some hosts incorrectly use sub-minor version for minor
>> diff --git a/drivers/usb/host/xhci-plat.c b/drivers/usb/host/xhci-plat.c
>> index b93161374293..75285fb5bbbc 100644
>> --- a/drivers/usb/host/xhci-plat.c
>> +++ b/drivers/usb/host/xhci-plat.c
>> @@ -249,6 +249,9 @@ int xhci_plat_probe(struct platform_device *pdev, struct device *sysdev, const s
>>   		if (device_property_read_bool(tmpdev, "quirk-broken-port-ped"))
>>   			xhci->quirks |= XHCI_BROKEN_PORT_PED;
>>   
>> +		if (device_property_read_bool(tmpdev, "disable-usb3"))
>> +			xhci->quirks |= XHCI_DISABLE_USB3;
>> +
>>   		device_property_read_u32(tmpdev, "imod-interval-ns",
>>   					 &xhci->imod_interval);
>>   	}
>> diff --git a/drivers/usb/host/xhci.h b/drivers/usb/host/xhci.h
>> index 5df370482521..c53fbeea478f 100644
>> --- a/drivers/usb/host/xhci.h
>> +++ b/drivers/usb/host/xhci.h
>> @@ -1906,6 +1906,7 @@ struct xhci_hcd {
>>   #define XHCI_RESET_TO_DEFAULT	BIT_ULL(44)
>>   #define XHCI_ZHAOXIN_TRB_FETCH	BIT_ULL(45)
>>   #define XHCI_ZHAOXIN_HOST	BIT_ULL(46)
>> +#define XHCI_DISABLE_USB3	BIT_ULL(47)
>>   
>>   	unsigned int		num_active_eps;
>>   	unsigned int		limit_active_eps;
>>
> 
> 
> 
>