On Thu, Feb 28, 2019 at 1:42 PM Vincenzo Frascino
[off-list ref] wrote:

quoted

On 28/02/2019 12:38, Arnd Bergmann wrote:

quoted

On Thu, Feb 28, 2019 at 1:09 PM Vincenzo Frascino
[off-list ref] wrote:

quoted

thank you for testing my code.

On 28/02/2019 11:40, Arnd Bergmann wrote:

quoted

On Fri, Feb 22, 2019 at 1:25 PM Vincenzo Frascino
[off-list ref] wrote:
  ^

I think this is all harmless, but we need to limit the set of header files
that can be included indirectly when compiling the vdso in arm32 mode.

I agree with you on the principle, but I am unable to replicate the warnings.
Did you make any customization to the default defconfig? What is the top
commit-id you have on your tree?

I observed this on top of v5.0-rc7 earlier, and have since rebased to
today's linux-next, where it still happens with the regular 'make defconfig',
on all compiler versions I have (4.8 through 8.1).

Thanks, I will try to rebase my patches on linux-next and see if I can reproduce it.

I just found a related issue: when CONFIG_CPU_BIG_ENDIAN is
set, everything breaks in the compat vdso because of the included
arm64 headers in arm32 code. That should be easier to reproduce.
For the moment, I'm using this patch:

--- a/arch/arm64/Kconfig
+++ b/arch/arm64/Kconfig

@@ -103,7 +103,7 @@ config ARM64
        select GENERIC_STRNLEN_USER
        select GENERIC_TIME_VSYSCALL
        select GENERIC_GETTIMEOFDAY
-       select GENERIC_COMPAT_VDSO
+       select GENERIC_COMPAT_VDSO if !CPU_BIG_ENDIAN
        select HANDLE_DOMAIN_IRQ
        select HARDIRQS_SW_RESEND
        select HAVE_PCI