2021-06-03 07:05:01

by Raviteja Narayanam

[permalink] [raw]
Subject: Need suggestion for 'access_type' of AMBA pl011 serial driver

Hi,

The uart peripheral on Xilinx Versal platform is ARM primecell.
Our environment is 32-bit access type but the ARM primecell uart in pl011 driver has default 16 bit access type.
(https://github.com/torvalds/linux/blob/master/drivers/tty/serial/amba-pl011.c#L2665 access_32b is false for 'vendor_arm')
This is causing asynchronous abort on our platform when any UART register is written from the pl011 driver.

Need suggestion on how we can address this issue and if the below approach is fine.

As this is platform specific issue, we can have a new device tree property (memory_access_type), specifying the 32 bit type.
In the probe function, override the behavior (uap->port.iotype) if this property is present in DT.
In this way, we can have support for our SOC, without breaking any legacy ones.

Regards,
Raviteja N


2021-06-03 07:16:14

by Greg Kroah-Hartman

[permalink] [raw]
Subject: Re: Need suggestion for 'access_type' of AMBA pl011 serial driver

On Thu, Jun 03, 2021 at 07:03:25AM +0000, Raviteja Narayanam wrote:
> Hi,
>
> The uart peripheral on Xilinx Versal platform is ARM primecell.
> Our environment is 32-bit access type but the ARM primecell uart in pl011 driver has default 16 bit access type.
> (https://github.com/torvalds/linux/blob/master/drivers/tty/serial/amba-pl011.c#L2665 access_32b is false for 'vendor_arm')
> This is causing asynchronous abort on our platform when any UART register is written from the pl011 driver.
>
> Need suggestion on how we can address this issue and if the below approach is fine.
>
> As this is platform specific issue, we can have a new device tree property (memory_access_type), specifying the 32 bit type.
> In the probe function, override the behavior (uap->port.iotype) if this property is present in DT.
> In this way, we can have support for our SOC, without breaking any legacy ones.

There is no other way to determine what this device is other than
platform data? Why not just set the vendor id in your device to a new
one and provide a correct setting in that new vendor data structure,
like all other devices for this chip have done before?

If it is the same id, and works differently, then that is a hardware
issue that you need to take up with your hardware designers to get fixed
as that is obviously a problem.

thanks,

greg k-h

2021-06-03 09:02:12

by Vladimir Zapolskiy

[permalink] [raw]
Subject: Re: Need suggestion for 'access_type' of AMBA pl011 serial driver

Hello Raviteja,

On 6/3/21 10:03 AM, Raviteja Narayanam wrote:
> Hi,
>
> The uart peripheral on Xilinx Versal platform is ARM primecell. Our
> environment is 32-bit access type but the ARM primecell uart in pl011
> driver has default 16 bit access type.
> (https://github.com/torvalds/linux/blob/master/drivers/tty/serial/amba-pl011.c#L2665
> access_32b is false for 'vendor_arm') This is causing asynchronous
> abort on our platform when any UART register is written from the
> pl011 driver.
>
> Need suggestion on how we can address this issue and if the below
> approach is fine.

Check drivers/tty/serial/amba-pl011.c code, there are quite many
controllers with 32-bit access, for instance ZTE UART and SBSA UART.

In other words this case is already well covered in the code, nothing
extraordinary is required or have to be invented here.

> As this is platform specific issue, we can have a new device tree
> property (memory_access_type), specifying the 32 bit type. In the
> probe function, override the behavior (uap->port.iotype) if this
> property is present in DT. In this way, we can have support for our
> SOC, without breaking any legacy ones.

No, there is totally no need for this kind of a device tree property.

Instead please introduce a new compatible of your controller, just
like it's been done for the SBSA UART controller:

compatible = "arm,sbsa-uart", "arm,pl011", "arm,primecell";

Then based on a match against your new compatible select a proper
configuration of the device driver, as I've said above similar cases
are already found in the code.

--
Best wishes,
Vladimir