]> git.kernelconcepts.de Git - karo-tx-linux.git/commit
Fix build failure due to hwirq.h needing smp_lock.h
authorLinus Torvalds <torvalds@linux-foundation.org>
Wed, 17 Nov 2010 22:58:36 +0000 (14:58 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 17 Nov 2010 22:58:36 +0000 (14:58 -0800)
commit7957f0a857754c555e07f58a3fb83ac29501478c
tree120976183d3f871b2023a745e888d71f96fbcfb3
parent460781b54253e3ed10a0a2a433bdc548ec952269
Fix build failure due to hwirq.h needing smp_lock.h

Arnd Bergmann did an automated scripting run to find left-over instances
of <linux/smp_lock.h>, and had made it trigger it on the normal BKL use
of lock_kernel and unlock_lernel (and apparently release_kernel_lock and
reacquire_kernel_lock too, used by the scheduler).

That resulted in commit 451a3c24b013 ("BKL: remove extraneous #include
<smp_lock.h>").

However, hardirq.h was the only remaining user of the old
'kernel_locked()' interface, and Arnd's script hadn't checked for that.
So depending on your configuration and what header files had been
included, you would get errors like "implicit declaration of function
'kernel_locked'" during the build.

The right fix is not to just re-instate the smp_lock.h include - it is
to just remove 'kernel_locked()' entirely, since the only use was this
one special low-level detail.  Just make hardirq.h do it directly.

In fact this simplifies and clarifies the code, because some trivial
analysis makes it clear that hardirq.h only ever used _one_ of the two
definitions of kernel_locked(), so we can remove the other one entirely.

Reported-by: Zimny Lech <napohybelskurwysynom2010@gmail.com>
Reported-and-acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/hardirq.h
include/linux/smp_lock.h