Re: linux-next: build failure in Linus' tree

Hi all,

On Tue, 20 Jul 2021 14:11:01 +0200 Jakub Kicinski [off-list ref] wrote:

On Tue, 20 Jul 2021 16:45:31 +1000, Stephen Rothwell wrote:

quoted

On Thu, 15 Jul 2021 09:50:32 +1000 Stephen Rothwell [off-list ref] wrote:  

quoted

While compiling Linus' tree, a powerpc-allmodconfig build (and others)
with gcc 4.9 failed like this:

drivers/net/ethernet/microchip/sparx5/sparx5_netdev.c: In function 'ifh_encode_bitfield':
include/linux/compiler_types.h:328:38: error: call to '__compiletime_assert_431' declared with attribute error: Unsupported width, must be <= 40
  _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
                                      ^
include/linux/compiler_types.h:309:4: note: in definition of macro '__compiletime_assert'
    prefix ## suffix();    \
    ^
include/linux/compiler_types.h:328:2: note: in expansion of macro '_compiletime_assert'
  _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
  ^
drivers/net/ethernet/microchip/sparx5/sparx5_netdev.c:28:2: note: in expansion of macro 'compiletime_assert'
  compiletime_assert(width <= 40, "Unsupported width, must be <= 40");
  ^

Caused by commit

  f3cad2611a77 ("net: sparx5: add hostmode with phylink support")

I guess this is caused by the call to ifh_encode_bitfield() not being
inlined.    

I am still getting these failures.  

Bjarni, Steen, could you address this build failure ASAP?

We can't have a compile time asserts in static functions, if the code
is optimized for size chances are the function won't get inlined. clang
is pretty bad at propagating constants to compile time asserts, too.
Please remove this check, or refactor it to be done in a macro, or ..

I am still getting these failures.

-- 
Cheers,
Stephen Rothwell