Re: [PATCH] iio: acpi: Accept comma separated values in ROTM strings

[Andy Shevchenko <andriy.shevchenko@intel.com>]

On Mon, Oct 12, 2026 at 03:41:44PM +0800, Yu-Hsiang Tseng wrote:
> The ROTM method of the MXC6655 accelerometer on the Higole F9B, an
> Alder Lake-N handheld (AMI BIOS F9D_16G_P800217, DMI strings are all
> "Default string"), separates the values with a comma and a space:
> 
> 	Method (ROTM, 0, NotSerialized)
> 	{
> 	    Name (SBUF, Package (0x03)
> 	    {
> 	        "-1, 0, 0",
> 	        "0, 1, 0",
> 	        "0, 0, 1"
> 	    })
> 	    Return (SBUF)
> 	}
> 
> iio_read_acpi_mount_matrix() only accepts the space separated form used
> in Microsoft's example ("0 1 0"). It rejects this matrix and mxc4005
> falls back to the identity matrix:
> 
> 	mxc4005 i2c-MXC6655:00: Incorrect ACPI mount matrix string format
> 	mxc4005 i2c-MXC6655:00: mounting matrix not found: using identity...
> 
> With the X axis not inverted, iio-sensor-proxy reports left-up when the
> device is held right-up and vice versa, so the desktop turns the
> picture upside down whenever the handheld is held in landscape.
> 
> It is not the only BIOS written this way. The BOSC0200 on the AYANEO
> Air Plus uses commas in its ROMS method as well, with a space before
> one of them, and its ROMK method has "1, 0,0":
> 
> 	Method (ROMS, 0, NotSerialized)
> 	{
> 	    Name (SBUF, Package (0x03)
> 	    {
> 	        "-1, 0, 0",
> 	        "0, -1 , 0",
> 	        "0, 0, -1"
> 	    })
> 	    Return (SBUF)
> 	}
> 
> The kernel's sscanf() does not support '%*[', so there is no clean way
> to take both separators in a single format string. That leaves either
> a second sscanf() call with a comma format, or normalizing the
> separator first. Do the latter: turn the commas into spaces before
> parsing, so the existing "%d %d %d" format, which already takes any
> amount of white space around the values, keeps working unchanged. The
> strings live in the buffer acpi_evaluate_object() allocated for us, so
> they can be modified in place.
> 
> With this change in_accel_mount_matrix on the Higole F9B reads
> "-1, 0, 0; 0, 1, 0; 0, 0, 1" and, with no udev hwdb override, the
> desktop rotates correctly in all four orientations. The AYANEO Air
> Plus was not tested; systemd's hwdb has an entry for it, which
> iio-sensor-proxy prefers over the kernel's matrix.

Makes sense, and fix is neat.
Reviewed-by: Andy Shevchenko <andriy.shevchenko@intel.com>

-- 
With Best Regards,
Andy Shevchenko