From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1757633AbdKOMxM (ORCPT ); Wed, 15 Nov 2017 07:53:12 -0500 Received: from us-smtp-delivery-107.mimecast.com ([216.205.24.107]:47505 "EHLO us-smtp-delivery-107.mimecast.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754719AbdKOMxG (ORCPT ); Wed, 15 Nov 2017 07:53:06 -0500 X-MC-Unique: HG4ow0_bMOyqzWccDrfvXw-1 Subject: Re: [RFC] Improving udelay/ndelay on platforms where that is possible From: Marc Gonzalez To: Linus Torvalds CC: Alan Cox , LKML , Linux ARM , Steven Rostedt , Ingo Molnar , Thomas Gleixner , Peter Zijlstra , John Stultz , Douglas Anderson , Nicolas Pitre , Mark Rutland , Will Deacon , Jonathan Austin , Arnd Bergmann , Kevin Hilman , Russell King , Michael Turquette , Stephen Boyd , Boris Brezillon , Thibaud Cornic , Mason References: <20171101175325.2557ce85@alans-desktop> <4b707ce0-6067-ab36-e167-1acf348d26bf@free.fr> Message-ID: <11393e07-b042-180c-3bcd-484bf51eada6@sigmadesigns.com> Date: Wed, 15 Nov 2017 13:51:54 +0100 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Firefox/52.0 SeaMonkey/2.49.1 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit X-Originating-IP: [172.27.0.114] Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 01/11/2017 20:38, Marc Gonzalez wrote: > OK, I'll just send my patch, and then crawl back under my rock. Linus, As promised, the patch is provided below. And as promised, I will no longer bring this up on LKML. FWIW, I have checked that the computed value matches the expected value for all HZ and delay_us, and for a few clock frequencies, using the following program: $ cat delays.c #include #define MEGA 1000000u typedef unsigned int uint; typedef unsigned long long u64; #define DIV_ROUND_UP(n,d) (((n) + (d) - 1) / (d)) static const uint HZ_tab[] = { 100, 250, 300, 1000 }; static void check_cycle_count(uint freq, uint HZ, uint delay_us) { uint UDELAY_MULT = (2147 * HZ) + (483648 * HZ / MEGA); uint lpj = DIV_ROUND_UP(freq, HZ); uint computed = ((u64)lpj * delay_us * UDELAY_MULT >> 31) + 1; uint expected = DIV_ROUND_UP((u64)delay_us * freq, MEGA); if (computed != expected) printf("freq=%u HZ=%u delay_us=%u comp=%u exp=%u\n", freq, HZ, delay_us, computed, expected); } int main(void) { uint idx, delay_us, freq; for (freq = 3*MEGA; freq <= 100*MEGA; freq += 3*MEGA) for (idx = 0; idx < sizeof HZ_tab / sizeof *HZ_tab; ++idx) for (delay_us = 1; delay_us <= 2000; ++delay_us) check_cycle_count(freq, HZ_tab[idx], delay_us); return 0; } -- >8 -- Subject: [PATCH] ARM: Tweak clock-based udelay implementation In 9f8197980d87a ("delay: Add explanation of udelay() inaccuracy") Russell pointed out that loop-based delays may return early. On the arm platform, delays may be either loop-based or clock-based. This patch tweaks the clock-based implementation so that udelay(N) is guaranteed to spin at least N microseconds. Signed-off-by: Marc Gonzalez --- arch/arm/lib/delay.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/arch/arm/lib/delay.c b/arch/arm/lib/delay.c index 2cef11884857..0a25712077ec 100644 --- a/arch/arm/lib/delay.c +++ b/arch/arm/lib/delay.c @@ -58,15 +58,15 @@ static void __timer_delay(unsigned long cycles) { cycles_t start = get_cycles(); - while ((get_cycles() - start) < cycles) + while ((get_cycles() - start) <= cycles) cpu_relax(); } static void __timer_const_udelay(unsigned long xloops) { - unsigned long long loops = xloops; - loops *= arm_delay_ops.ticks_per_jiffy; - __timer_delay(loops >> UDELAY_SHIFT); + u64 tmp = (u64)xloops * arm_delay_ops.ticks_per_jiffy; + unsigned long cycles = tmp >> UDELAY_SHIFT; + __timer_delay(cycles + 1); /* Round up in 1 instruction */ } static void __timer_udelay(unsigned long usecs) @@ -92,7 +92,7 @@ void __init register_current_timer_delay(const struct delay_timer *timer) if (!delay_calibrated && (!delay_res || (res < delay_res))) { pr_info("Switching to timer-based delay loop, resolution %lluns\n", res); delay_timer = timer; - lpj_fine = timer->freq / HZ; + lpj_fine = DIV_ROUND_UP(timer->freq, HZ); delay_res = res; /* cpufreq may scale loops_per_jiffy, so keep a private copy */ -- 2.15.0