> ... > > > +static int aw9610x_read_chipid(struct aw9610x *aw9610x) > > +{ > > + unsigned char cnt = 0; > > + u32 reg_val; > > + int ret; > > + > > + while (cnt < AW_READ_CHIPID_RETRIES) { Why retries? > > + ret = aw9610x_i2c_read(aw9610x, REG_CHIPID, ®_val); > > + if (ret < 0) { > > + cnt++; > > + usleep_range(2000, 3000); > > + } else { > > + reg_val = FIELD_GET(AW9610X_CHIPID_MASK, reg_val); > > + break; > > + } > > + } > > + > > + if (reg_val == AW9610X_CHIP_ID) > > + return 0; > > So devices are detectable? Encode this in the bindings (oneOf and a > fallback compatible) and drop unneeded entry from ID tables. Hi Krzysztof, I think this is not a good idea. Even though these two are detectable, this breaks if along comes a 3rd device in the future which is truly compatible with one of these two parts but that we don't yet know about (so can't discover). For that part we will want to provide a meaningful fallback compatible. It needs to fallback to either the 3 channel or the 5 channel chip and handle it as appropriate. (Note that this difference is non obvious as right now the code pretends there are always 5 channels and that needs fixing). If the chips provided a register that told all the chip specific data like how many channels, then sure making one fallback to the other would be fine as future devices could use those standard registers. With just an Id register, we can't discover enough. Hence these two parts should not be listed as compatible with each other. Jonathan > > Best regards, > Krzysztof >