From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1756221Ab0I3Sub (ORCPT ); Thu, 30 Sep 2010 14:50:31 -0400 Received: from mga02.intel.com ([134.134.136.20]:12244 "EHLO mga02.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752018Ab0I3Sua (ORCPT ); Thu, 30 Sep 2010 14:50:30 -0400 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="4.57,260,1283756400"; d="scan'208";a="559479444" Subject: [patch] intel_idle: Voluntary leave_mm before entering deeper c-states From: Suresh Siddha Reply-To: Suresh Siddha To: Len Brown , LKML Cc: Arjan van de Ven Content-Type: text/plain Organization: Intel Corp Date: Thu, 30 Sep 2010 11:50:13 -0700 Message-Id: <1285872613.2676.4.camel@sbsiddha-MOBL3.sc.intel.com> Mime-Version: 1.0 X-Mailer: Evolution 2.26.3 (2.26.3-1.fc11) Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Avoid TLB flush IPIs for the cores in deeper c-states by voluntary leave_mm() before entering into that state. CPUs tend to flush TLB in those c-states anyways. ACPI processor idle driver does this and was left out when intel idle got introduced. Signed-off-by: Suresh Siddha --- diff --git a/drivers/idle/intel_idle.c b/drivers/idle/intel_idle.c index a10152b..dff4663 100755 --- a/drivers/idle/intel_idle.c +++ b/drivers/idle/intel_idle.c @@ -108,7 +108,7 @@ static struct cpuidle_state nehalem_cstates[MWAIT_MAX_NUM_CSTATES] = { .name = "NHM-C3", .desc = "MWAIT 0x10", .driver_data = (void *) 0x10, - .flags = CPUIDLE_FLAG_TIME_VALID, + .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, .exit_latency = 20, .power_usage = 500, .target_residency = 80, @@ -117,7 +117,7 @@ static struct cpuidle_state nehalem_cstates[MWAIT_MAX_NUM_CSTATES] = { .name = "NHM-C6", .desc = "MWAIT 0x20", .driver_data = (void *) 0x20, - .flags = CPUIDLE_FLAG_TIME_VALID, + .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, .exit_latency = 200, .power_usage = 350, .target_residency = 800, @@ -149,7 +149,7 @@ static struct cpuidle_state atom_cstates[MWAIT_MAX_NUM_CSTATES] = { .name = "ATM-C4", .desc = "MWAIT 0x30", .driver_data = (void *) 0x30, - .flags = CPUIDLE_FLAG_TIME_VALID, + .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, .exit_latency = 100, .power_usage = 250, .target_residency = 400, @@ -159,7 +159,7 @@ static struct cpuidle_state atom_cstates[MWAIT_MAX_NUM_CSTATES] = { .name = "ATM-C6", .desc = "MWAIT 0x40", .driver_data = (void *) 0x40, - .flags = CPUIDLE_FLAG_TIME_VALID, + .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED, .exit_latency = 200, .power_usage = 150, .target_residency = 800, @@ -185,6 +185,16 @@ static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state) local_irq_disable(); + /* + * If the state flag indicates that the TLB will be flushed or if this + * is the deepest c-state supported, do a voluntary leave mm to avoid + * costly and mostly unnecessary wakeups for flushing the user TLB's + * associated with the active mm. + */ + if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED || + (&dev->states[dev->state_count - 1] == state)) + leave_mm(cpu); + if (!(lapic_timer_reliable_states & (1 << (cstate)))) clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu); diff --git a/include/linux/cpuidle.h b/include/linux/cpuidle.h index 36ca972..1be416b 100644 --- a/include/linux/cpuidle.h +++ b/include/linux/cpuidle.h @@ -53,6 +53,7 @@ struct cpuidle_state { #define CPUIDLE_FLAG_BALANCED (0x40) /* medium latency, moderate savings */ #define CPUIDLE_FLAG_DEEP (0x80) /* high latency, large savings */ #define CPUIDLE_FLAG_IGNORE (0x100) /* ignore during this idle period */ +#define CPUIDLE_FLAG_TLB_FLUSHED (0x200) /* tlb will be flushed */ #define CPUIDLE_DRIVER_FLAGS_MASK (0xFFFF0000)