Message ID | 5561935F.9060704@web.de |
---|---|
State | Rejected |
Headers | show |
On 2015-05-24 11:01, Dirk Neukirchen wrote: > > unify config across platforms > RESOURCE_COUNTERS is deactive on all platforms > > Signed-off-by: Dirk Neukirchen <dirkneukirchen@web.de> > --- Removing target/subtarget kernel config lines individually does not make any sense. These files are generated by make kernel_menuconfig or make kernel_oldconfig. - Felix
On 24.05.2015 11:15, Felix Fietkau wrote: > On 2015-05-24 11:01, Dirk Neukirchen wrote: >> >> unify config across platforms >> RESOURCE_COUNTERS is deactive on all platforms >> >> Signed-off-by: Dirk Neukirchen <dirkneukirchen@web.de> >> --- > Removing target/subtarget kernel config lines individually does not make > any sense. These files are generated by make kernel_menuconfig or make > kernel_oldconfig. > > - Felix > Thanks for clarifying - _but_ if I rerun "make kernel_menuconfig" with current trunk I get several big changesets that might break some targets if I look at the symbol changes on some of them: Example: x86 with selected KPROBES in menuconfig needs # CONFIG_X86_DECODER_SELFTEST is not set but X86_DECODER_SELFTEST is removed from target/x86/config-3.18 with "make kernel_menuconfig" but other symbols from arch/x86/Kconfig.debug are kept in that file make target/linux/compile : does not work because the missing symbol make kernel_oldconfig : does not add or question this symbol a patch that has "make kernel_menuconfig" changes for all targets (might be missing some subtargets) is available at: https://github.com/plntyk/openwrt-patches/blob/master/kernel_refresh/0016-refreshes-according-to-make-kernel_menuconfig.patch (from a tree where some changes regarding kernel version/symbols were already applied) I did run "make dirclean" before arch switching and a target/linux/clean run before running make kernel_menuconfig. I really dont get it where the menuconfig changes are coming from Would patching the arch/<targetarch>/configs/*defconfig help/be the correct fix? The patch that does this sort of thing for brcm2708 looks really bad (patch #023) "Malta"(-LE) maybe needs a defconfig since kernel_menuconfig wants to switch to the big endian variant Those "instable" symbols are a bit annoying when they will get "accidentally" reverted
On 2015-05-27 00:02, Dirk Neukirchen wrote: > On 24.05.2015 11:15, Felix Fietkau wrote: >> On 2015-05-24 11:01, Dirk Neukirchen wrote: >>> >>> unify config across platforms >>> RESOURCE_COUNTERS is deactive on all platforms >>> >>> Signed-off-by: Dirk Neukirchen <dirkneukirchen@web.de> >>> --- >> Removing target/subtarget kernel config lines individually does not make >> any sense. These files are generated by make kernel_menuconfig or make >> kernel_oldconfig. >> >> - Felix >> > > Thanks for clarifying - _but_ if I rerun "make kernel_menuconfig" > with current trunk I get several big changesets that might break > some targets if I look at the symbol changes on some of them: > > Example: > x86 with selected KPROBES in menuconfig needs > # CONFIG_X86_DECODER_SELFTEST is not set > > but > X86_DECODER_SELFTEST is removed from target/x86/config-3.18 with "make kernel_menuconfig" > but other symbols from arch/x86/Kconfig.debug are kept in that file > > make target/linux/compile : does not work because the missing symbol > make kernel_oldconfig : does not add or question this symbol Adding it to generic/config-* will fix this. > a patch that has "make kernel_menuconfig" changes > for all targets (might be missing some subtargets) is available at: > https://github.com/plntyk/openwrt-patches/blob/master/kernel_refresh/0016-refreshes-according-to-make-kernel_menuconfig.patch > (from a tree where some changes regarding kernel version/symbols were already applied) > > I did run "make dirclean" before arch switching and a target/linux/clean run before running make kernel_menuconfig. > > I really dont get it where the menuconfig changes are coming from > Would patching the arch/<targetarch>/configs/*defconfig help/be the correct fix? That should not be necessary. > The patch that does this sort of thing for brcm2708 looks really bad > (patch #023) Not sure where this delta is coming from or how this config was generated. This needs to be looked at more closely. Either way, all target config files should be kept in a way that make kernel_menuconfig works properly and does not break stuff. > "Malta"(-LE) maybe needs a defconfig since kernel_menuconfig wants to switch to the big endian variant Here's one way to resolve this: copy the generated .config from the kernel tree for each subtarget somewhere. Then refresh the target config file. Afterwards, go through each subtarget, replace the config-* file in there with the saved .config, refresh using kernel_oldconfig afterwards. This should ensure that the change/refresh will not result in any effective kernel configuration changes. - Felix
diff --git a/target/linux/sunxi/config-3.18 b/target/linux/sunxi/config-3.18 index ddc3f7a..fde6a78 100644 --- a/target/linux/sunxi/config-3.18 +++ b/target/linux/sunxi/config-3.18 @@ -386,7 +386,6 @@ CONFIG_REGULATOR_FIXED_VOLTAGE=y # CONFIG_REGULATOR_USERSPACE_CONSUMER is not set CONFIG_RELAY=y CONFIG_RESET_CONTROLLER=y -CONFIG_RESOURCE_COUNTERS=y # CONFIG_RFKILL_REGULATOR is not set CONFIG_RFS_ACCEL=y CONFIG_RPS=y
unify config across platforms RESOURCE_COUNTERS is deactive on all platforms Signed-off-by: Dirk Neukirchen <dirkneukirchen@web.de> --- target/linux/sunxi/config-3.18 | 1 - 1 file changed, 1 deletion(-)