Re: powerpc: build failures in Linus' tree

[Michael Ellerman <mpe@ellerman.id.au>]

Willy Tarreau <w@1wt.eu> writes:
> On Sun, Aug 02, 2020 at 07:20:19PM +0200, Willy Tarreau wrote:
>> On Sun, Aug 02, 2020 at 08:48:42PM +1000, Stephen Rothwell wrote:
>> > Hi all,
>> > 
>> > We are getting build failures in some PowerPC configs for Linus' tree.
>> > See e.g. http://kisskb.ellerman.id.au/kisskb/buildresult/14306515/
>> > 
>> > In file included from /kisskb/src/arch/powerpc/include/asm/paca.h:18,
>> >                  from /kisskb/src/arch/powerpc/include/asm/percpu.h:13,
>> >                  from /kisskb/src/include/linux/random.h:14,
>> >                  from /kisskb/src/include/linux/net.h:18,
>> >                  from /kisskb/src/net/ipv6/ip6_fib.c:20:
>> > /kisskb/src/arch/powerpc/include/asm/mmu.h:139:22: error: unknown type name 'next_tlbcam_idx'
>> >   139 | DECLARE_PER_CPU(int, next_tlbcam_idx);
>> > 
>> > I assume this is caused by commit
>> > 
>> >   1c9df907da83 ("random: fix circular include dependency on arm64 after addition of percpu.h")
>> > 
>> > But I can't see how, sorry.
>> 
>> So there, asm/mmu.h includes asm/percpu.h, which includes asm/paca.h, which
>> includes asm/mmu.h.
>> 
>> I suspect that we can remove asm/paca.h from asm/percpu.h as it *seems*
>> to be only used by the #define __my_cpu_offset but I don't know if anything
>> will break further, especially if this __my_cpu_offset is used anywhere
>> without this paca definition.
>
> I tried this and it fixed 5.8 for me with your config above. I'm appending
> a patch that does just this. I didn't test other configs as I don't know
> which ones to test though. If it fixes the problem for you, maybe it can
> be picked by the PPC maintainers.
>
> Willy
> From bcd64a7d0f3445c9a75d3b4dc4837d2ce61660c9 Mon Sep 17 00:00:00 2001
> From: Willy Tarreau <w@1wt.eu>
> Date: Mon, 3 Aug 2020 05:27:57 +0200
> Subject: powerpc: fix circular dependency in percpu.h
>
> After random.h started to include percpu.h (commit f227e3e), several
> archs broke in circular dependencies around percpu.h.
>
> In https://lore.kernel.org/lkml/20200802204842.36bca162@canb.auug.org.au/
> Stephen Rothwell reported breakage for powerpc with CONFIG_PPC_FSL_BOOK3E.
>
> It turns out that asm/percpu.h includes asm/paca.h, which itself
> includes mmu.h, which includes percpu.h when CONFIG_PPC_FSL_BOOK3E=y.
>
> Percpu seems to include asm/paca.h only for local_paca which is used in
> the __my_cpu_offset macro. Removing this include solves the issue for
> this config.
>
> Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
> Fixes: f227e3e ("random32: update the net random state on interrupt and activity")
> Link: https://lore.kernel.org/lkml/20200802204842.36bca162@canb.auug.org.au/
> Cc: Linus Torvalds <torvalds@linux-foundation.org>
> Cc: Michael Ellerman <mpe@ellerman.id.au>
> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
> Cc: Paul Mackerras <paulus@samba.org>
> Signed-off-by: Willy Tarreau <w@1wt.eu>
> ---
>  arch/powerpc/include/asm/percpu.h | 2 --
>  1 file changed, 2 deletions(-)
>
> diff --git a/arch/powerpc/include/asm/percpu.h b/arch/powerpc/include/asm/percpu.h
> index dce863a..cd3f6e5 100644
> --- a/arch/powerpc/include/asm/percpu.h
> +++ b/arch/powerpc/include/asm/percpu.h
> @@ -10,8 +10,6 @@
>  
>  #ifdef CONFIG_SMP
>  
> -#include <asm/paca.h>
> -
>  #define __my_cpu_offset local_paca->data_offset
>  
>  #endif /* CONFIG_SMP */

If we just move the include of asm/paca.h below asm-generic/percpu.h
then it avoids the bad circular dependency and we still have paca.h
included from percpu.h as before.

eg:

diff --git a/arch/powerpc/include/asm/percpu.h b/arch/powerpc/include/asm/percpu.h
index dce863a7635c..8e5b7d0b851c 100644
--- a/arch/powerpc/include/asm/percpu.h
+++ b/arch/powerpc/include/asm/percpu.h
@@ -10,8 +10,6 @@
 
 #ifdef CONFIG_SMP
 
-#include <asm/paca.h>
-
 #define __my_cpu_offset local_paca->data_offset
 
 #endif /* CONFIG_SMP */
@@ -19,4 +17,6 @@
 
 #include <asm-generic/percpu.h>
 
+#include <asm/paca.h>
+
 #endif /* _ASM_POWERPC_PERCPU_H_ */


So I think I'm inclined to merge that as a minimal fix that's easy to
backport.

cheers