From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ingo Molnar Subject: Re: [PATCH] x86 idle: repair large-server 50-watt idle-power regression Date: Thu, 19 Dec 2013 13:22:57 +0100 Message-ID: <20131219122257.GC11279@gmail.com> References: Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from mail-ea0-f170.google.com ([209.85.215.170]:38364 "EHLO mail-ea0-f170.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756555Ab3LSMXC (ORCPT ); Thu, 19 Dec 2013 07:23:02 -0500 Content-Disposition: inline In-Reply-To: Sender: linux-pm-owner@vger.kernel.org List-Id: linux-pm@vger.kernel.org To: Len Brown Cc: x86@kernel.org, linux-pm@vger.kernel.org, linux-kernel@vger.kernel.org, Len Brown , stable@vger.kernel.org, Linus Torvalds , "H. Peter Anvin" , Thomas Gleixner , Peter Zijlstra , Mike Galbraith , Borislav Petkov * Len Brown wrote: > From: Len Brown > > Linux 3.10 changed the timing of how thread_info->flags is touched: > > x86: Use generic idle loop > (7d1a941731fabf27e5fb6edbebb79fe856edb4e5) > > This caused Intel NHM-EX and WSM-EX servers to experience a large number > of immediate MONITOR/MWAIT break wakeups, which caused cpuidle to demote > from deep C-states to shallow C-states, which caused these platforms > to experience a significant increase in idle power. > > Note that this issue was already present before the commit above, > however, it wasn't seen often enough to be noticed in power measurements. > > Here we extend an errata workaround from the Core2 EX "Dunnington" > to extend to NHM-EX and WSM-EX, to prevent these immediate > returns from MWAIT, reducing idle power on these platforms. > > While only acpi_idle ran on Dunnington, intel_idle > may also run on these two newer systems. > As of today, there are no other models that are known > to need this tweak. > > ref: https://lkml.org/lkml/2013/12/7/22 > Signed-off-by: Len Brown > Cc: # 3.12.x, 3.11.x, 3.10.x > --- > arch/x86/kernel/cpu/intel.c | 3 ++- > drivers/idle/intel_idle.c | 3 +++ > 2 files changed, 5 insertions(+), 1 deletion(-) > > diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c > index dc1ec0d..ea04b34 100644 > --- a/arch/x86/kernel/cpu/intel.c > +++ b/arch/x86/kernel/cpu/intel.c > @@ -387,7 +387,8 @@ static void init_intel(struct cpuinfo_x86 *c) > set_cpu_cap(c, X86_FEATURE_PEBS); > } > > - if (c->x86 == 6 && c->x86_model == 29 && cpu_has_clflush) > + if (c->x86 == 6 && cpu_has_clflush && > + (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47)) > set_cpu_cap(c, X86_FEATURE_CLFLUSH_MONITOR); > > #ifdef CONFIG_X86_64 > diff --git a/drivers/idle/intel_idle.c b/drivers/idle/intel_idle.c > index 92d1206..f80b700 100644 > --- a/drivers/idle/intel_idle.c > +++ b/drivers/idle/intel_idle.c > @@ -377,6 +377,9 @@ static int intel_idle(struct cpuidle_device *dev, > > if (!current_set_polling_and_test()) { > > + if (this_cpu_has(X86_FEATURE_CLFLUSH_MONITOR)) > + clflush((void *)¤t_thread_info()->flags); > + > __monitor((void *)¤t_thread_info()->flags, 0, 0); I don't think either of these casts to '(void *)' is needed, both the clflush() and __monitor() will take pointers. Looks good to me otherwise - except that maybe the best way to represent this quirk would be for the CLFLUSH+MONITOR sequence to be a single 'instruction' which is patched in dynamically during bootup, using our usual alternatives framework. On non-affected CPUs a NOP would remain in place of the CLFLUSH, eliminating the branch above. So the whole thing could be thought of as a slightly more complex 'monitor' instruction - not exposing the quirk details to actual usage sites. Thanks, Ingo