Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S964829AbVKVBfZ (ORCPT ); Mon, 21 Nov 2005 20:35:25 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S964828AbVKVBfZ (ORCPT ); Mon, 21 Nov 2005 20:35:25 -0500 Received: from e35.co.us.ibm.com ([32.97.110.153]:59864 "EHLO e35.co.us.ibm.com") by vger.kernel.org with ESMTP id S964829AbVKVBfY (ORCPT ); Mon, 21 Nov 2005 20:35:24 -0500 Date: Mon, 21 Nov 2005 18:35:22 -0700 From: john stultz To: lkml Cc: Ingo Molnar , Darren Hart , Nishanth Aravamudan , Frank Sorenson , George Anzinger , Roman Zippel , Ulrich Windl , Thomas Gleixner , john stultz , john stultz Message-Id: <20051122013522.18537.97944.sendpatchset@cog.beaverton.ibm.com> In-Reply-To: <20051122013515.18537.76463.sendpatchset@cog.beaverton.ibm.com> References: <20051122013515.18537.76463.sendpatchset@cog.beaverton.ibm.com> Subject: [PATCH 1/13] Time: Reduced NTP rework (part 1) Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 4876 Lines: 159 All, With Roman's suggestions, I've been working on reducing the footprint of my timeofday patches. This is the first of two patches that reworks some of the interrupt time NTP adjustments so that it could be re-used with the timeofday patches. The motivation of the change is to logically separate the code which adjusts xtime and the code that decides, based on the NTP state variables, how much per tick to adjust xtime. Thus this patch should not affect the existing behavior, but just separate the logical functionality so it can be re-used. This patch applies on top of 2.6.15-rc1-mm2. thanks -john Signed-off-by: John Stultz linux-2.6.15-rc1-mm2_timeofday-ntp-part1_B11.patch ============================================ diff -ruN tod-mm_1/kernel/timer.c tod-mm_2/kernel/timer.c --- tod-mm_1/kernel/timer.c 2005-11-21 16:39:15.000000000 -0800 +++ tod-mm_2/kernel/timer.c 2005-11-21 16:43:03.000000000 -0800 @@ -595,6 +595,7 @@ long time_reftime; /* time at last adjustment (s) */ long time_adjust; long time_next_adjust; +long time_adjust_step; /* per tick time_adjust step */ /* * this routine handles the overflow of the microsecond field @@ -722,45 +723,52 @@ #endif } -/* in the NTP reference this is called "hardclock()" */ -static void update_wall_time_one_tick(void) +/** + * ntp_advance() - increments the NTP state machine + * + * Must be holding the xtime writelock when calling. + * + */ +static void ntp_advance(unsigned long interval_ns) { - long time_adjust_step, delta_nsec; + static unsigned long interval_sum; - if ((time_adjust_step = time_adjust) != 0 ) { - /* - * We are doing an adjtime thing. Prepare time_adjust_step to - * be within bounds. Note that a positive time_adjust means we - * want the clock to run faster. - * - * Limit the amount of the step to be in the range - * -tickadj .. +tickadj - */ - time_adjust_step = min(time_adjust_step, (long)tickadj); - time_adjust_step = max(time_adjust_step, (long)-tickadj); + /* increment the interval sum */ + interval_sum += interval_ns; + + /* calculate the per tick singleshot adjtime adjustment step */ + while (interval_ns >= tick_nsec) { + time_adjust_step = time_adjust; + if (time_adjust_step) { + /* We are doing an adjtime thing. + * + * Prepare time_adjust_step to be within bounds. + * Note that a positive time_adjust means we want + * the clock to run faster. + * + * Limit the amount of the step to be in the range + * -tickadj .. +tickadj + */ + time_adjust_step = min(time_adjust_step, (long)tickadj); + time_adjust_step = max(time_adjust_step, + (long)-tickadj); - /* Reduce by this step the amount of time left */ - time_adjust -= time_adjust_step; - } - delta_nsec = tick_nsec + time_adjust_step * 1000; - /* - * Advance the phase, once it gets to one microsecond, then - * advance the tick more. - */ - time_phase += time_adj; - if ((time_phase >= FINENSEC) || (time_phase <= -FINENSEC)) { - long ltemp = shift_right(time_phase, (SHIFT_SCALE - 10)); - time_phase -= ltemp << (SHIFT_SCALE - 10); - delta_nsec += ltemp; + /* Reduce by this step the amount of time left */ + time_adjust -= time_adjust_step; + } + interval_ns -= tick_nsec; } - xtime.tv_nsec += delta_nsec; - time_interpolator_update(delta_nsec); /* Changes by adjtime() do not take effect till next tick. */ if (time_next_adjust != 0) { time_adjust = time_next_adjust; time_next_adjust = 0; } + + while (interval_sum >= NSEC_PER_SEC) { + interval_sum -= NSEC_PER_SEC; + second_overflow(); + } } /* @@ -772,14 +780,36 @@ */ static void update_wall_time(unsigned long ticks) { + long delta_nsec; + do { ticks--; - update_wall_time_one_tick(); - if (xtime.tv_nsec >= 1000000000) { - xtime.tv_nsec -= 1000000000; + + /* Calculate the nsec delta using the + * precomputed NTP adjustments: + * tick_nsec, time_adjust_step, time_adj + */ + delta_nsec = tick_nsec + time_adjust_step * 1000; + /* + * Advance the phase, once it gets to one microsecond, then + * advance the tick more. + */ + time_phase += time_adj; + if ((time_phase >= FINENSEC) || (time_phase <= -FINENSEC)) { + long ltemp = shift_right(time_phase, + (SHIFT_SCALE - 10)); + time_phase -= ltemp << (SHIFT_SCALE - 10); + delta_nsec += ltemp; + } + + xtime.tv_nsec += delta_nsec; + if (xtime.tv_nsec >= NSEC_PER_SEC) { + xtime.tv_nsec -= NSEC_PER_SEC; xtime.tv_sec++; - second_overflow(); } + ntp_advance(tick_nsec); + time_interpolator_update(delta_nsec); + } while (ticks); } - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/