On Tue 2020-06-23 22:19:37, Herbert Xu wrote:

On Tue, Jun 23, 2020 at 02:16:38PM +0200, Petr Mladek wrote:

quoted

I have removed the problematic commit for now. It tried to remove
some cyclic dependencies from heavily used include files. It clearly
needs more love.

Hmm, the cyclic dependencies are there because you didn't pull in
the lockdep_types patch.  The printk patch must go on top of the
lockdep_types patch.  How about just putting this into the x86 tree
alongside the lockdep_types patch?

I see the problem with both patches.

In file included from include/linux/printk.h:10,
                 from include/linux/kernel.h:15,
                 from include/linux/list.h:9,
                 from include/linux/lockdep.h:43,
                 from include/linux/spinlock_types.h:18,
                 from include/linux/genalloc.h:32,
                 from drivers/soc/fsl/qe/qe_common.c:16:
include/linux/ratelimit_types.h:16:2: error: unknown type name 'raw_spinlock_t'
   16 |  raw_spinlock_t lock;  /* protect the state */

It is similar cycle:

spinlock_types.h -> lockdep.h -> printk.h -> ratelimit.h -> spinlock_types.h

But this time it happens via list.h -> kernel.h ->printk.h.
Where list.h needs READ_ONCE() stuff from compiler.h.


My "allmodconfig" build has successfully finished with the following extra
 fix on top of the two patches:

diff --git a/include/linux/list.h b/include/linux/list.h
index aff44d34f4e4..6d606c4036ce 100644
--- a/include/linux/list.h
+++ b/include/linux/list.h

@@ -6,7 +6,7 @@
 #include <linux/stddef.h>
 #include <linux/poison.h>
 #include <linux/const.h>
-#include <linux/kernel.h>
+#include <linux/compiler.h>
 
 /*
  * Simple doubly linked list implementation.

I suggest to bundle this into the 2nd patch that makes linux/printk.h
self-contained.

Best Regards,
Petr

PS: And yes, it makes sense to push both patches via a single tree to
make sure that the lockdep.h split is done first.