Hi Joe,
c4e789572557 ("s390/nospec: build expoline.o for modules_prepare
target") is now in linux.git.
Note: arch/s390/lib/expoline.o is moved to
arch/s390/lib/expoline/expoline.o. This means kernel-devel package in
fedora should also include this updated file path.
Thanks
On 7/1/22 23:39, Joe Lawrence wrote:
On 7/1/22 5:18 PM, C. Erastus Toe wrote:
On Wed, Jun 29, 2022 at 11:16 AM Joe Lawrence <joe.lawrence@xxxxxxxxxx
<mailto:joe.lawrence@xxxxxxxxxx>> wrote:
On 6/27/22 8:50 AM, Vasily Gorbik wrote:
> Hi Joe,
>
> sorry for late reply.
>
>> I couldn't find the upstream patch post for 1d2ad084800e
("s390/nospec:
>> add an option to use thunk-extern"), so replying off-list here. Feel
>> free to cc the appropriate list.
>>
>> Regarding this change, as I understand it, when
CONFIG_EXPOLINE_EXTERN=y
>> out-of-tree kernel modules will need to link against
>> arch/s390x/lib/expoline.o, right?
>>
>> And if so, shouldn't the top level 'prepare_modules' target create
>> expoline.o for this purpose?
>
> Thanks for bringing this up. I definitely missed out-of-tree
kernel modules
> build case without a prebuilt kernel. On the other hand this
post-linking
> trick is a rip off from powerpc:
>
> KBUILD_LDFLAGS_MODULE += arch/powerpc/lib/crtsavres.o
>
> So, now I wonder why powerpc doesn't have crtsavres.o in
'prepare_modules'.
>
> Anyhow, below is couple of patches to consider. The first one is
> meant to be backportable, as the second one requires 4efd417f298b.
>
> I had to move expoline.S to a separate directory to be able to
call into
> its Makefile for 'prepare_modules' and avoid warnings for other
targets
> defined in the same Makefile. Not sure if there are better kbuild
tricks
> I could use. Another option I thought about is to keep expoline.S
where
> it is and add a condition into that Makefile:
> expoline_prepare: prepare0
> $(Q)$(MAKE) $(build)=arch/s390/lib expoline_prepare=1
arch/s390/lib/expoline.o
>
> arch/s390/lib/Makefile:
> # first target defined
> obj-$(CONFIG_EXPOLINE_EXTERN) += expoline.o
> ifndef expoline_prepare
> # ...other targets...
>
> Vasily Gorbik (2):
> s390/nospec: build expoline.o for modules_prepare target
> s390/nospec: remove unneeded header includes
>
> arch/s390/Makefile | 8 +++++++-
> arch/s390/include/asm/nospec-insn.h | 2 --
> arch/s390/lib/Makefile | 3 ++-
> arch/s390/lib/expoline/Makefile | 3 +++
> arch/s390/lib/{ => expoline}/expoline.S | 0
> 5 files changed, 12 insertions(+), 4 deletions(-)
> create mode 100644 arch/s390/lib/expoline/Makefile
> rename arch/s390/lib/{ => expoline}/expoline.S (100%)
>
Thanks, Vasily. We'll test these with OOT and the original gitlab
pipeline where we spotted potential issue with packaging and report
back.
Hi,
Successfully tested the first patch in a rhel-9 backport. (had to skip
the second as it has dependencies on other patches like [1] that
deprecated symbols like __LC_BR_R1. Without those, the build resulted in
a flood of: depmod: WARNING: <module>.ko needs unknown symbol __LC_BR_R1.)
For ("s390/nospec: build expoline.o for modules_prepare target"),
Tested-by: C. Erastus Toe <ctoe@xxxxxxxxxx <mailto:ctoe@xxxxxxxxxx>>
[1] 4efd417f298b ("s390: raise minimum supported machine generation to z10")
And then for the entire series (tested on top of v5.19-rc4),
Tested-by: Joe Lawrence <joe.lawrence@xxxxxxxxxx>
--
Sumanth
