powerpc/Makefile: Don't pass -mcpu=powerpc64 when building 32-bit
authorMichael Ellerman <mpe@ellerman.id.au>
Tue, 15 Feb 2022 11:28:58 +0000 (22:28 +1100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 8 Apr 2022 12:23:36 +0000 (14:23 +0200)
commit29322cd03678561dddb2475db51b1dc7124120b6
treebada6190652bf1195d2bf53fed1b67e6fcf441f9
parent451472dc178faf1dd3dff96e1ce67f9bdbf6f611
powerpc/Makefile: Don't pass -mcpu=powerpc64 when building 32-bit

[ Upstream commit 2863dd2db23e0407f6c50b8ba5c0e55abef894f1 ]

When CONFIG_GENERIC_CPU=y (true for all our defconfigs) we pass
-mcpu=powerpc64 to the compiler, even when we're building a 32-bit
kernel.

This happens because we have an ifdef CONFIG_PPC_BOOK3S_64/else block in
the Makefile that was written before 32-bit supported GENERIC_CPU. Prior
to that the else block only applied to 64-bit Book3E.

The GCC man page says -mcpu=powerpc64 "[specifies] a pure ... 64-bit big
endian PowerPC ... architecture machine [type], with an appropriate,
generic processor model assumed for scheduling purposes."

It's unclear how that interacts with -m32, which we are also passing,
although obviously -m32 is taking precedence in some sense, as the
32-bit kernel only contains 32-bit instructions.

This was noticed by inspection, not via any bug reports, but it does
affect code generation. Comparing before/after code generation, there
are some changes to instruction scheduling, and the after case (with
-mcpu=powerpc64 removed) the compiler seems more keen to use r8.

Fix it by making the else case only apply to Book3E 64, which excludes
32-bit.

Fixes: 0e00a8c9fd92 ("powerpc: Allow CPU selection also on PPC32")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220215112858.304779-1-mpe@ellerman.id.au
Signed-off-by: Sasha Levin <sashal@kernel.org>
arch/powerpc/Makefile