[OpenWrt-Devel] [PATCH v3] kernel: revert bad module stripping patch

Philip Prindeville philipp_subx at redfish-solutions.com
Mon Oct 29 19:50:19 EDT 2018


Some hardware (it’s rare but not unheard of) can only be reset by unloading and reloading the module that controls it.  Otherwise, you have to reboot the box.  If you build all of your drivers in, then rebooting is all you have.

My firewall has been up for 3 years.  I update it with opkg, including modules.

It would have been up longer, but I have to swap out the UPS that it’s on tomorrow.

Well, at least now I can sysupgrade to a newer kernel…


> On Oct 29, 2018, at 5:01 PM, Daniel Santos <daniel.santos at pobox.com> wrote:
> 
> I would like to take this opportunity to mention that the best way to
> save space here is to not build modules -- make them built-ins.  I did a
> quick experiment of this and instead of saving 4k, my *image* is a full
> 256k smaller.  I haven't analysed the specifics, but also this means
> less RAM consumed because squashfs uses the page cache for uncompressed
> files.  Further, modules inherently have greater overhead, even after
> __init sections have been discarded.
> 
> sed -i 's/=m/=y/g;'
> build_dir/target-mipsel_24kc_musl/linux-ramips_mt7620/linux-4.14.63/.config
> make
> 
> The only downside is that built-ins cannot be unloaded and will always
> occupy a portion of RAM.  But having them built into the kernel is far
> more efficient.
> 
> Daniel
> 
> On 10/29/2018 05:54 PM, Daniel Santos wrote:
>> Never EVER do this!  Decades of computer science has taught us "best
>> practices" -- ways of doing things that provide an optimal outcome for
>> all parties involved.  While this type of hack is fine if you're working
>> alone, it is not a "best practice" and is completely unacceptable for a
>> large FOSS project.  This is especially insidious because it goes behind
>> the back of the programmer / user / hacker and makes a change without
>> their consent or direction -- this is the type of behaviour we've come
>> to expect from Google, m$, Apple and the like, not from FOSS.
>> 
>> Doing --strip-all is not appropriate default behaviour as this includes
>> relocation data and breaks things like kmemleak.  The presence of debug
>> information in modules is already controlled by CONFIG_DEBUG_INFO,
>> CONFIG_DEBUG_INFO_REDUCED, etc.  Force-stripping just makes life crappy
>> for anybody who needs to debug the a kernel built in OpenWRT.  (Some of
>> us do run gdb on the kernel.)
>> 
>> If --strip-all really is desired, it should probably be added to the
>> CONFIG_MODULE_STRIPPED patch, and should be done at install time
>> (probably via objcopy), not at link time.  This leaves the in-tree
>> modules useful for debugging, decompiling, diagnostics, profiling, etc.
>> 
>> Signed-off-by: Daniel Santos <daniel.santos at pobox.com>
>> ---
>> .../generic/hack-4.14/202-reduce_module_size.patch | 24 ----------------------
>> .../generic/hack-4.9/202-reduce_module_size.patch  | 24 ----------------------
>> 2 files changed, 48 deletions(-)
>> delete mode 100644 target/linux/generic/hack-4.14/202-reduce_module_size.patch
>> delete mode 100644 target/linux/generic/hack-4.9/202-reduce_module_size.patch
>> 
>> diff --git a/target/linux/generic/hack-4.14/202-reduce_module_size.patch b/target/linux/generic/hack-4.14/202-reduce_module_size.patch
>> deleted file mode 100644
>> index 2cbb6add9a..0000000000
>> --- a/target/linux/generic/hack-4.14/202-reduce_module_size.patch
>> +++ /dev/null
>> @@ -1,24 +0,0 @@
>> -From fd66884da2f96d2a7ea73f58b1b86251b959a913 Mon Sep 17 00:00:00 2001
>> -From: Felix Fietkau <nbd at nbd.name>
>> -Date: Fri, 7 Jul 2017 16:56:19 +0200
>> -Subject: kernel: strip unnecessary symbol table information from kernel modules
>> -
>> -reduces default squashfs size on ar71xx by about 4k
>> -
>> -lede-commit: 058d331a39077f159ca8922f1f422a1346d6aa67
>> -Signed-off-by: Felix Fietkau <nbd at nbd.name>
>> ----
>> - Makefile | 2 +-
>> - 1 file changed, 1 insertion(+), 1 deletion(-)
>> -
>> ---- a/Makefile
>> -+++ b/Makefile
>> -@@ -425,7 +425,7 @@ KBUILD_AFLAGS_KERNEL :=
>> - KBUILD_CFLAGS_KERNEL :=
>> - KBUILD_AFLAGS_MODULE  := -DMODULE
>> - KBUILD_CFLAGS_MODULE  := -DMODULE
>> --KBUILD_LDFLAGS_MODULE := -T $(srctree)/scripts/module-common.lds
>> -+KBUILD_LDFLAGS_MODULE = -T $(srctree)/scripts/module-common.lds $(if $(CONFIG_PROFILING),,-s)
>> - GCC_PLUGINS_CFLAGS :=
>> - 
>> - export ARCH SRCARCH SUBARCH CONFIG_SHELL HOSTCC HOSTCFLAGS CROSS_COMPILE AS LD
>> diff --git a/target/linux/generic/hack-4.9/202-reduce_module_size.patch b/target/linux/generic/hack-4.9/202-reduce_module_size.patch
>> deleted file mode 100644
>> index f744b945fe..0000000000
>> --- a/target/linux/generic/hack-4.9/202-reduce_module_size.patch
>> +++ /dev/null
>> @@ -1,24 +0,0 @@
>> -From fd66884da2f96d2a7ea73f58b1b86251b959a913 Mon Sep 17 00:00:00 2001
>> -From: Felix Fietkau <nbd at nbd.name>
>> -Date: Fri, 7 Jul 2017 16:56:19 +0200
>> -Subject: kernel: strip unnecessary symbol table information from kernel modules
>> -
>> -reduces default squashfs size on ar71xx by about 4k
>> -
>> -lede-commit: 058d331a39077f159ca8922f1f422a1346d6aa67
>> -Signed-off-by: Felix Fietkau <nbd at nbd.name>
>> ----
>> - Makefile | 2 +-
>> - 1 file changed, 1 insertion(+), 1 deletion(-)
>> -
>> ---- a/Makefile
>> -+++ b/Makefile
>> -@@ -403,7 +403,7 @@ KBUILD_AFLAGS_KERNEL :=
>> - KBUILD_CFLAGS_KERNEL :=
>> - KBUILD_AFLAGS_MODULE  := -DMODULE
>> - KBUILD_CFLAGS_MODULE  := -DMODULE
>> --KBUILD_LDFLAGS_MODULE := -T $(srctree)/scripts/module-common.lds
>> -+KBUILD_LDFLAGS_MODULE = -T $(srctree)/scripts/module-common.lds $(if $(CONFIG_PROFILING),,-s)
>> - GCC_PLUGINS_CFLAGS :=
>> - 
>> - # Read KERNELRELEASE from include/config/kernel.release (if it exists)
> 
> 
> _______________________________________________
> openwrt-devel mailing list
> openwrt-devel at lists.openwrt.org
> https://lists.openwrt.org/mailman/listinfo/openwrt-devel


_______________________________________________
openwrt-devel mailing list
openwrt-devel at lists.openwrt.org
https://lists.openwrt.org/mailman/listinfo/openwrt-devel


More information about the openwrt-devel mailing list