On 8/15/25 9:12 AM, Luca Weiss wrote:
> Hi Wolfram,
>
> On 2025-08-11 14:13, Wolfram Sang wrote:
>> On Sun, Aug 10, 2025 at 05:37:53PM +0200, Luca Weiss wrote:
>>> Add a config for the v1.2.5 CCI found on msm8953 which has different
>>
>> Given the above version number...
>>
>>> static const struct of_device_id cci_dt_match[] = {
>>> { .compatible = "qcom,msm8226-cci", .data = &cci_v1_data},
>>> + { .compatible = "qcom,msm8953-cci", .data = &cci_msm8953_data},
>>
>> ... why don't we use it here to stay consistent? cci_v1_2_5_data?
>
> I don't think the existing 'v2' or 'v1' configs have much to do with the
> actual
> HW_VERSION of the IP block. For example on of the newer Qualcomm SoCs has HW
> version 1.7.0 and is many years newer than the msm8996 which was called 'v2'.
Most of the i2c-speed-dependent configs are electrical and therefore may
be configured as you wish.. I recall Sony Xperia kernels made changes to
that rather often, whether really necessary - I don't know.
The programming guide suggests a couple sets of values, picked in order
to meet the I2C spec.
The queue depths and max r/w lengths are per-instance (SoC specific) and
change very rarely.
>
> I'm also not sure what these parameters depend on, if it's CCI HW version, or
> something else. So naming it after the SoC should be a safer bet. Also the
> msm8974-cci was only named 'v1.5' because it's an inbetween mix of the v1 and
> v2 that were already upstream so arguably that one shouldn't have been called
> v1.5 in the first place either.
AFAICS there is no "v2" CCI in existence, msm8996 is v1.4.0 and newer
platforms are v1.x.y.
JFYI there's a revision ID register at (base + 0x0) which the driver's probe
function conveniently reads and prints with dev_dbg.
Konrad
