Hi Andy, On Thu, 25 Aug 2022 23:12:43 +0300, Andy Shevchenko wrote: > On Thu, Aug 25, 2022 at 3:54 PM Jean Delvare <jdelvare@xxxxxxx> wrote: > > > > When CONFIG_OF isn't set, we can optimize out dmard06_of_match as it > > will never be used. > > NACK. There is a magic PRP0001 for ACPI case. OK, I wasn't aware of this. As of_match_device() is a stub when CONFIG_OF isn't set, I thought the table could never be used. But now I see that the acpi subsystem accesses the table directly, so you are correct, the optimization I suggested is bogus. Now I'm curious, is there a well-defined subset of device names that can be found in an ACPI table? If not, does that mean that any driver with an OF entry could match, therefore of_match_ptr() should be removed from the kernel entirely? Another possibility would be to stub out of_match_ptr() only if neither CONFIG_OF nor CONFIG_ACPI is set. But I'm not sure that would be worth, as I expected either to be set on almost all kernels in practice (except on s390x, but you wouldn't build any of these drivers there anyway). -- Jean Delvare SUSE L3 Support
