From: Andi Kleen <ak@suse.de>
To: linux-kernel@vger.kernel.org
Subject: [PATCH x86] [12/16] Optimize lock prefix switching to run less frequently
Date: Thu, 3 Jan 2008 16:42:26 +0100 (CET) [thread overview]
Message-ID: <20080103154226.67D6414DDB@wotan.suse.de> (raw)
In-Reply-To: <20080103442.621670000@suse.de>
On VMs implemented using JITs that cache translated code changing the lock
prefixes is a quite costly operation that forces the JIT to throw away and
retranslate a lot of code.
Previously a SMP kernel would rewrite the locks once for each CPU which
is quite unnecessary. This patch changes the code to never switch at boot in
the normal case (SMP kernel booting with >1 CPU) or only once for SMP kernel
on UP.
This makes a significant difference in boot up performance on AMD SimNow!
Also I expect it to be a little faster on native systems too because a smp
switch does a lot of text_poke()s which each synchronize the pipeline.
Signed-off-by: Andi Kleen <ak@suse.de>
---
arch/x86/kernel/alternative.c | 16 ++++++++++++++--
include/linux/smp.h | 2 ++
init/main.c | 2 +-
3 files changed, 17 insertions(+), 3 deletions(-)
Index: linux/arch/x86/kernel/alternative.c
===================================================================
--- linux.orig/arch/x86/kernel/alternative.c
+++ linux/arch/x86/kernel/alternative.c
@@ -273,6 +273,7 @@ struct smp_alt_module {
};
static LIST_HEAD(smp_alt_modules);
static DEFINE_SPINLOCK(smp_alt);
+static int smp_mode = 1; /* protected by smp_alt */
void alternatives_smp_module_add(struct module *mod, char *name,
void *locks, void *locks_end,
@@ -354,7 +355,14 @@ void alternatives_smp_switch(int smp)
BUG_ON(!smp && (num_online_cpus() > 1));
spin_lock_irqsave(&smp_alt, flags);
- if (smp) {
+
+ /*
+ * Avoid unnecessary switches because it forces JIT based VMs to
+ * throw away all cached translations, which can be quite costly.
+ */
+ if (smp == smp_mode) {
+ /* nothing */
+ } else if (smp) {
printk(KERN_INFO "SMP alternatives: switching to SMP code\n");
clear_cpu_cap(&boot_cpu_data, X86_FEATURE_UP);
clear_cpu_cap(&cpu_data(0), X86_FEATURE_UP);
@@ -369,6 +377,7 @@ void alternatives_smp_switch(int smp)
alternatives_smp_unlock(mod->locks, mod->locks_end,
mod->text, mod->text_end);
}
+ smp_mode = smp;
spin_unlock_irqrestore(&smp_alt, flags);
}
@@ -441,7 +450,10 @@ void __init alternative_instructions(voi
alternatives_smp_module_add(NULL, "core kernel",
__smp_locks, __smp_locks_end,
_text, _etext);
- alternatives_smp_switch(0);
+
+ /* Only switch to UP mode if we don't immediately boot others */
+ if (num_possible_cpus() == 1 || max_cpus == 0)
+ alternatives_smp_switch(0);
}
#endif
apply_paravirt(__parainstructions, __parainstructions_end);
Index: linux/include/linux/smp.h
===================================================================
--- linux.orig/include/linux/smp.h
+++ linux/include/linux/smp.h
@@ -78,6 +78,8 @@ int on_each_cpu(void (*func) (void *info
*/
void smp_prepare_boot_cpu(void);
+extern unsigned int max_cpus;
+
#else /* !SMP */
/*
Index: linux/init/main.c
===================================================================
--- linux.orig/init/main.c
+++ linux/init/main.c
@@ -128,7 +128,7 @@ static char *ramdisk_execute_command;
#ifdef CONFIG_SMP
/* Setup configured maximum number of CPUs to activate */
-static unsigned int __initdata max_cpus = NR_CPUS;
+unsigned int __initdata max_cpus = NR_CPUS;
/*
* Setup routine for controlling SMP activation
next prev parent reply other threads:[~2008-01-03 15:46 UTC|newest]
Thread overview: 103+ messages / expand[flat|nested] mbox.gz Atom feed top
2008-01-03 15:42 [PATCH x86] [0/16] Various i386/x86-64 changes Andi Kleen
2008-01-03 15:42 ` [PATCH x86] [1/16] Make clocksource watchdog cycle through online CPUs Andi Kleen
2008-01-04 8:51 ` Ingo Molnar
2008-01-04 12:59 ` Andi Kleen
2008-01-03 15:42 ` [PATCH x86] [2/16] Add a counter for per cpu clocksource watchdog checks and report them in /proc/interrupts Andi Kleen
2008-01-04 8:53 ` Ingo Molnar
2008-01-03 15:42 ` [PATCH x86] [3/16] Turn irq debugging options into module params Andi Kleen
2008-01-04 8:55 ` Ingo Molnar
2008-01-03 15:42 ` [PATCH x86] [4/16] Add /proc/irq/*/spurious to dump the spurious irq debugging state Andi Kleen
2008-01-04 8:58 ` Ingo Molnar
2008-01-04 12:22 ` Andi Kleen
2008-01-03 15:42 ` [PATCH x86] [5/16] Replace hard coded reservations in x86-64 early boot code with dynamic table Andi Kleen
2008-01-04 9:00 ` Ingo Molnar
2008-01-04 12:23 ` Andi Kleen
2008-01-04 12:38 ` [PATCH x86] [5/16] Replace hard coded reservations in x86-64 early boot code with dynamic table II Andi Kleen
2008-01-04 14:41 ` Ingo Molnar
2008-01-04 15:03 ` Andi Kleen
2008-01-04 17:51 ` Sam Ravnborg
2008-01-04 18:06 ` Andi Kleen
2008-01-04 18:47 ` Joe Perches
2008-01-04 22:13 ` Andi Kleen
2008-01-04 22:46 ` J. Bruce Fields
2008-01-04 20:56 ` Sam Ravnborg
2008-01-04 12:41 ` [PATCH x86] [5/16] Replace hard coded reservations in x86-64 early boot code with dynamic table Cyrill Gorcunov
2008-01-03 15:42 ` [PATCH x86] [6/16] Add a new arch_early_alloc() interface for x86-64 Andi Kleen
2008-01-03 16:22 ` Eric Dumazet
2008-01-03 17:17 ` Andi Kleen
2008-01-03 15:42 ` [PATCH x86] [7/16] Convert lockdep to use arch_early_alloc() if available for its large arrays Andi Kleen
2008-01-04 9:03 ` Ingo Molnar
2008-01-03 15:42 ` [PATCH x86] [8/16] Make lockdep_init __init Andi Kleen
2008-01-04 9:06 ` Ingo Molnar
2008-01-03 15:42 ` [PATCH x86] [9/16] Remove CPU capabitilites printks on i386 Andi Kleen
2008-01-04 9:11 ` Ingo Molnar
2008-01-03 15:42 ` [PATCH x86] [10/16] Document fdimage/isoimage completely in make help Andi Kleen
2008-01-04 9:00 ` Sam Ravnborg
2008-01-04 9:15 ` Ingo Molnar
2008-01-03 15:42 ` [PATCH x86] [11/16] Compile apm and voyager module only when selected in Kconfig Andi Kleen
2008-01-04 4:15 ` Stephen Rothwell
2008-01-03 15:42 ` Andi Kleen [this message]
2008-01-04 9:42 ` [PATCH x86] [12/16] Optimize lock prefix switching to run less frequently Thomas Gleixner
2008-01-04 12:17 ` Andi Kleen
2008-01-04 14:19 ` Thomas Gleixner
2008-01-04 14:39 ` Andi Kleen
2008-01-04 15:02 ` Pekka Enberg
2008-01-04 15:04 ` Andi Kleen
2008-01-04 20:34 ` Thomas Gleixner
2008-01-04 22:11 ` Andi Kleen
2008-01-05 1:19 ` Ingo Molnar
2008-01-05 14:37 ` Andi Kleen
2008-01-03 15:42 ` [PATCH x86] [13/16] i386: Set CFQ as default in i386 defconfig Andi Kleen
2008-01-04 9:19 ` Ingo Molnar
2008-01-03 15:42 ` [PATCH x86] [14/16] Enable RDC321X subarch only on 32bit Andi Kleen
2008-01-04 9:22 ` Ingo Molnar
2008-01-03 15:42 ` [PATCH x86] [15/16] Force __cpuinit on for CONFIG_PM without HOTPLUG_CPU Andi Kleen
2008-01-03 17:22 ` Rafael J. Wysocki
2008-01-03 17:42 ` Andi Kleen
2008-01-03 21:41 ` Rafael J. Wysocki
2008-01-04 9:23 ` Ingo Molnar
2008-01-10 17:14 ` Rafael J. Wysocki
2008-01-03 18:14 ` Adrian Bunk
2008-01-03 18:43 ` Andi Kleen
2008-01-10 9:54 ` Adrian Bunk
2008-01-10 9:58 ` Andi Kleen
2008-01-10 10:19 ` Adrian Bunk
2008-01-10 11:15 ` Andi Kleen
2008-01-10 11:26 ` Adrian Bunk
2008-01-10 11:42 ` Andi Kleen
2008-01-10 12:47 ` Adrian Bunk
2008-01-10 13:12 ` Andi Kleen
2008-01-10 15:09 ` Adrian Bunk
2008-01-14 13:52 ` Ingo Molnar
2008-01-14 14:09 ` Sam Ravnborg
2008-01-14 14:58 ` Ingo Molnar
2008-01-14 15:01 ` Ingo Molnar
2008-01-14 18:20 ` Adrian Bunk
2008-01-14 19:10 ` Ingo Molnar
2008-01-14 19:52 ` Adrian Bunk
2008-01-14 20:01 ` Sam Ravnborg
2008-01-14 20:18 ` Adrian Bunk
2008-01-14 20:27 ` Sam Ravnborg
2008-01-14 20:34 ` Adrian Bunk
2008-01-15 22:14 ` Ingo Molnar
2008-01-15 22:51 ` Adrian Bunk
2008-01-14 19:59 ` Sam Ravnborg
2008-01-15 22:12 ` [patch for v2.6.24] fix section mismatch warning in page_alloc.c Ingo Molnar
2008-01-14 15:05 ` [PATCH x86] [15/16] Force __cpuinit on for CONFIG_PM without HOTPLUG_CPU Ingo Molnar
2008-01-14 15:24 ` Ingo Molnar
2008-01-14 20:12 ` Sam Ravnborg
2008-01-15 15:17 ` Ingo Molnar
2008-01-15 16:25 ` Sam Ravnborg
2008-01-15 17:11 ` Andi Kleen
2008-01-15 17:27 ` Adrian Bunk
2008-01-15 18:21 ` Sam Ravnborg
2008-01-15 18:29 ` Andi Kleen
2008-01-15 18:31 ` Sam Ravnborg
2008-01-15 18:47 ` Adrian Bunk
2008-01-14 18:17 ` Adrian Bunk
2008-01-14 15:25 ` Adrian Bunk
2008-01-10 17:17 ` Rafael J. Wysocki
2008-01-11 23:06 ` [PATCH] x86: Change unnecessary dependencies on CONFIG_PM Rafael J. Wysocki
2008-01-03 15:42 ` [PATCH x86] [16/16] Mark memory_setup __init Andi Kleen
2008-01-04 9:25 ` Ingo Molnar
2008-01-04 9:53 ` 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=20080103154226.67D6414DDB@wotan.suse.de \
--to=ak@suse.de \
--cc=linux-kernel@vger.kernel.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
all inboxes | Powered by JetHome®