From: Heiko Carstens <heiko.carstens@de.ibm.com>
To: Ingo Molnar <mingo@elte.hu>
Cc: Andrew Morton <akpm@linux-foundation.org>,
Linus Torvalds <torvalds@linux-foundation.org>,
David Miller <davem@davemloft.net>,
Benjamin Herrenschmidt <benh@kernel.crashing.org>,
Paul Mackerras <paulus@samba.org>,
Geert Uytterhoeven <geert@linux-m68k.org>,
Roman Zippel <zippel@linux-m68k.org>,
linux-arch@vger.kernel.org,
Peter Zijlstra <a.p.zijlstra@chello.nl>,
Arnd Bergmann <arnd@arndb.de>,
Nick Piggin <nickpiggin@yahoo.com.au>,
Martin Schwidefsky <schwidefsky@de.ibm.com>,
Horst Hartmann <horsth@linux.vnet.ibm.com>,
Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com>
Subject: Re: [patch 0/8] Allow inlined spinlocks again V5
Date: Sat, 29 Aug 2009 15:59:07 +0200 [thread overview]
Message-ID: <20090829135906.GA11215@osiris.boeblingen.de.ibm.com> (raw)
In-Reply-To: <20090829111642.GA17951@elte.hu>
On Sat, Aug 29, 2009 at 01:16:42PM +0200, Ingo Molnar wrote:
> * Heiko Carstens <heiko.carstens@de.ibm.com> wrote:
> > This patch set allows to have inlined spinlocks again.
> >
> > V5: simplify ifdefs as pointed out by Linus. Fix architecture compile
> > breakages caused by this change.
> >
> > Linus, Ingo, do you still have objections?
>
> Ok, this looks pretty clean.
>
> Two small comments. One (small) worry is that the configuration
> space of this construct:
>
> +#define __spin_lock_is_small
> [...]
>
> Is 2^28.
>
> Could we perhaps shape this in a form that makes it 'formally' a
> manually tuned inlining decision - where we already accept such kind
> of per function decisions and know how to handle them?
>
> I.e. rename it to something like:
>
> #define __always_inline__write_unlock_irqrestore
>
> That way it fits into our existing forced-inlining attributes
> (visually and name-space wise), which means the addition of 'inline'
> or '__forced_inline', or the removal of such attributes.
Ok, I'll wait for more comments and post an updated version.
> The other comment i have: you dont seem to have preserved the
> current auto-inlining of spin_lock() we did before, have you?
> Without that this patchset causes a (small) performance regression
> on every architectures but s390.
I assume you're talking of spin_unlock() and not spin_lock()?
It still gets inlined (spinlock.h):
/*
* We inline the unlock functions in the nondebug case:
*/
#if defined(CONFIG_DEBUG_SPINLOCK) || defined(CONFIG_PREEMPT) || \
!defined(CONFIG_SMP)
# define spin_unlock(lock) _spin_unlock(lock)
[...]
#else
# define spin_unlock(lock) \
do {__raw_spin_unlock(&(lock)->raw_lock); __release(lock); } while (0)
[...]
Or in other words, specifying __spin_lock_is_small causes only that the
out-of-line variant doesn't get generated anymore.
It gets inlined regardless (if config options allow).
Now this could be simplified/cleanup by doing something like this:
/*
* We inline the unlock functions in the nondebug case:
*/
#if defined(CONFIG_DEBUG_SPINLOCK) || defined(CONFIG_PREEMPT) || \
!defined(CONFIG_SMP)
# define spin_unlock(lock) _spin_unlock(lock)
[...]
#else
# define spin_unlock(lock) \
do {__raw_spin_unlock(&(lock)->raw_lock); __release(lock); } while (0)
[...]
However it sucks a bit that we have two different ways to force inlining.
So we could and probably should simplify this. The patch below (untested)
would do that:
---
include/linux/spinlock.h | 46 +++++----------------------------------
include/linux/spinlock_api_smp.h | 12 ++++++++++
2 files changed, 18 insertions(+), 40 deletions(-)
Index: linux-2.6/include/linux/spinlock.h
===================================================================
--- linux-2.6.orig/include/linux/spinlock.h
+++ linux-2.6/include/linux/spinlock.h
@@ -259,50 +259,16 @@ static inline void smp_mb__after_lock(vo
#define spin_lock_irq(lock) _spin_lock_irq(lock)
#define spin_lock_bh(lock) _spin_lock_bh(lock)
-
#define read_lock_irq(lock) _read_lock_irq(lock)
#define read_lock_bh(lock) _read_lock_bh(lock)
-
#define write_lock_irq(lock) _write_lock_irq(lock)
#define write_lock_bh(lock) _write_lock_bh(lock)
-
-/*
- * We inline the unlock functions in the nondebug case:
- */
-#if defined(CONFIG_DEBUG_SPINLOCK) || defined(CONFIG_PREEMPT) || \
- !defined(CONFIG_SMP)
-# define spin_unlock(lock) _spin_unlock(lock)
-# define read_unlock(lock) _read_unlock(lock)
-# define write_unlock(lock) _write_unlock(lock)
-# define spin_unlock_irq(lock) _spin_unlock_irq(lock)
-# define read_unlock_irq(lock) _read_unlock_irq(lock)
-# define write_unlock_irq(lock) _write_unlock_irq(lock)
-#else
-# define spin_unlock(lock) \
- do {__raw_spin_unlock(&(lock)->raw_lock); __release(lock); } while (0)
-# define read_unlock(lock) \
- do {__raw_read_unlock(&(lock)->raw_lock); __release(lock); } while (0)
-# define write_unlock(lock) \
- do {__raw_write_unlock(&(lock)->raw_lock); __release(lock); } while (0)
-# define spin_unlock_irq(lock) \
-do { \
- __raw_spin_unlock(&(lock)->raw_lock); \
- __release(lock); \
- local_irq_enable(); \
-} while (0)
-# define read_unlock_irq(lock) \
-do { \
- __raw_read_unlock(&(lock)->raw_lock); \
- __release(lock); \
- local_irq_enable(); \
-} while (0)
-# define write_unlock_irq(lock) \
-do { \
- __raw_write_unlock(&(lock)->raw_lock); \
- __release(lock); \
- local_irq_enable(); \
-} while (0)
-#endif
+#define spin_unlock(lock) _spin_unlock(lock)
+#define read_unlock(lock) _read_unlock(lock)
+#define write_unlock(lock) _write_unlock(lock)
+#define spin_unlock_irq(lock) _spin_unlock_irq(lock)
+#define read_unlock_irq(lock) _read_unlock_irq(lock)
+#define write_unlock_irq(lock) _write_unlock_irq(lock)
#define spin_unlock_irqrestore(lock, flags) \
do { \
Index: linux-2.6/include/linux/spinlock_api_smp.h
===================================================================
--- linux-2.6.orig/include/linux/spinlock_api_smp.h
+++ linux-2.6/include/linux/spinlock_api_smp.h
@@ -60,6 +60,18 @@ void __lockfunc _read_unlock_irqrestore(
void __lockfunc _write_unlock_irqrestore(rwlock_t *lock, unsigned long flags)
__releases(lock);
+/*
+ * We inline the unlock functions in the nondebug case:
+ */
+#if !defined(CONFIG_DEBUG_SPINLOCK) && !defined(CONFIG_PREEMPT)
+#define __spin_unlock_is_small
+#define __read_unlock_is_small
+#define __write_unlock_is_small
+#define __spin_unlock_irq_is_small
+#define __read_unlock_irq_is_small
+#define __write_unlock_irq_is_small
+#endif
+
#ifndef CONFIG_DEBUG_SPINLOCK
#ifndef CONFIG_GENERIC_LOCKBREAK
next prev parent reply other threads:[~2009-08-29 13:59 UTC|newest]
Thread overview: 24+ messages / expand[flat|nested] mbox.gz Atom feed top
2009-08-29 10:21 [patch 0/8] Allow inlined spinlocks again V5 Heiko Carstens
2009-08-29 10:21 ` Heiko Carstens
2009-08-29 10:21 ` [patch 1/8] powerpc: rename __spin_try_lock() and friends Heiko Carstens
2009-08-29 10:21 ` Heiko Carstens
2009-08-29 10:21 ` [patch 2/8] sparc: " Heiko Carstens
2009-08-29 10:21 ` Heiko Carstens
2009-08-29 22:27 ` David Miller
2009-08-29 10:21 ` [patch 3/8] m68k/asm-offsets: rename pt_regs offset defines Heiko Carstens
2009-08-29 10:21 ` Heiko Carstens
2009-08-29 10:21 ` [patch 4/8] m68k/asm-offsets: rename signal defines Heiko Carstens
2009-08-29 10:21 ` Heiko Carstens
2009-08-29 10:21 ` [patch 5/8] m68k: calculate thread_info offset with asm offset Heiko Carstens
2009-08-29 10:21 ` Heiko Carstens
2009-08-29 10:21 ` [patch 6/8] spinlock: move spinlock function bodies to header file Heiko Carstens
2009-08-29 10:21 ` Heiko Carstens
2009-08-29 10:21 ` [patch 7/8] spinlock: allow inlined spinlocks Heiko Carstens
2009-08-29 10:21 ` Heiko Carstens
2009-08-29 10:21 ` [patch 8/8] spinlock: inline code for all locking variants on s390 Heiko Carstens
2009-08-29 10:21 ` Heiko Carstens
2009-08-29 11:16 ` [patch 0/8] Allow inlined spinlocks again V5 Ingo Molnar
2009-08-29 13:59 ` Heiko Carstens [this message]
2009-08-29 14:13 ` Ingo Molnar
2009-08-31 8:59 ` Heiko Carstens
2009-09-01 13:19 ` Ingo Molnar
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20090829135906.GA11215@osiris.boeblingen.de.ibm.com \
--to=heiko.carstens@de.ibm.com \
--cc=a.p.zijlstra@chello.nl \
--cc=akpm@linux-foundation.org \
--cc=arnd@arndb.de \
--cc=benh@kernel.crashing.org \
--cc=davem@davemloft.net \
--cc=ehrhardt@linux.vnet.ibm.com \
--cc=geert@linux-m68k.org \
--cc=horsth@linux.vnet.ibm.com \
--cc=linux-arch@vger.kernel.org \
--cc=mingo@elte.hu \
--cc=nickpiggin@yahoo.com.au \
--cc=paulus@samba.org \
--cc=schwidefsky@de.ibm.com \
--cc=torvalds@linux-foundation.org \
--cc=zippel@linux-m68k.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).