From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751943AbcB2V3K (ORCPT ); Mon, 29 Feb 2016 16:29:10 -0500 Received: from mga11.intel.com ([192.55.52.93]:49418 "EHLO mga11.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751397AbcB2V3I (ORCPT ); Mon, 29 Feb 2016 16:29:08 -0500 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.22,521,1449561600"; d="scan'208";a="923751308" From: "Christopher S. Hall" To: tglx@linutronix.de, luto@amacapital.net, john.stultz@linaro.org Cc: hpa@zytor.com, x86@kernel.org, linux-kernel@vger.kernel.org, peterz@infradead.org, "Christopher S. Hall" Subject: Re: [PATCH v7 6/8] x86: tsc: Always Running Timer (ART) correlated clocksource Date: Mon, 29 Feb 2016 06:22:38 -0800 Message-Id: <1456755758-2808-1-git-send-email-christopher.s.hall@intel.com> X-Mailer: git-send-email 2.1.4 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Mon, 22 Feb 2016 18:49:10 -0800, Andy Lutomirski wrote: >> an >> earlier related thread Peter Zijlstra asserts that TSC adjust "had >> better" be >> 0.(http://lkml.iu.edu/hypermail/linux/kernel/1507.3/03734.html). The comment more or less reinforces your description below of such a BIOS as "crappy" > There are three interesting cases that I can think of: > > 1. Crappy BIOS that sets TSC_ADJUST. As the not-so-proud owner of a > piece of crap motherboard that actively messes with the TSC, I don't > trust BIOS. I have added code that read this offset MSR at initialization time. This isn't bulletproof, but it covers the most common case where BIOS boots with TSC adjust set to something non-zero. > 2. Hypervisors. What if we're running as a guest with an ART-using > NIC passed through? I don't see a lot of use cases where virtualization would be used. Currently this feature (ART) isn't available on server platforms. Without explicit support, there are a lot of ways this can go wrong. For now, I've added an extra check that disables ART on a VM. > 3. Hypothetical improved future kernel that politely uses TSC_ADJUST > to keep the TSC from going backwards across suspend/resume. This seems like a non-trivial change. Without know how this code is going to work, it would be difficult to test the ART solution. *** Commit message below: On modern Intel systems TSC is derived from the new Always Running Timer (ART). ART can be captured simultaneous to the capture of audio and network device clocks, allowing a correlation between timebases to be constructed. Upon capture, the driver converts the captured ART value to the appropriate system clock using the correlated clocksource mechanism. On systems that support ART a new CPUID leaf (0x15) returns parameters “m” and “n” such that: TSC_value = (ART_value * m) / n + k [n >= 1] [k is an offset that can adjusted by a privileged agent. The IA32_TSC_ADJUST MSR is an example of an interface to adjust k. See 17.14.4 of the Intel SDM for more details] Reviewed-by: Thomas Gleixner Signed-off-by: Christopher S. Hall [jstultz: Tweaked to fix build issue, also reworked math for 64bit division on 32bit systems] Signed-off-by: John Stultz --- arch/x86/include/asm/cpufeature.h | 2 +- arch/x86/include/asm/tsc.h | 2 ++ arch/x86/kernel/tsc.c | 59 +++++++++++++++++++++++++++++++++++++++ 3 files changed, 62 insertions(+), 1 deletion(-) diff --git a/arch/x86/include/asm/cpufeature.h b/arch/x86/include/asm/cpufeature.h index 7ad8c94..ff557b4 100644 --- a/arch/x86/include/asm/cpufeature.h +++ b/arch/x86/include/asm/cpufeature.h @@ -85,7 +85,7 @@ #define X86_FEATURE_P4 ( 3*32+ 7) /* "" P4 */ #define X86_FEATURE_CONSTANT_TSC ( 3*32+ 8) /* TSC ticks at a constant rate */ #define X86_FEATURE_UP ( 3*32+ 9) /* smp kernel running on up */ -/* free, was #define X86_FEATURE_FXSAVE_LEAK ( 3*32+10) * "" FXSAVE leaks FOP/FIP/FOP */ +#define X86_FEATURE_ART (3*32+10) /* Platform has always running timer (ART) */ #define X86_FEATURE_ARCH_PERFMON ( 3*32+11) /* Intel Architectural PerfMon */ #define X86_FEATURE_PEBS ( 3*32+12) /* Precise-Event Based Sampling */ #define X86_FEATURE_BTS ( 3*32+13) /* Branch Trace Store */ diff --git a/arch/x86/include/asm/tsc.h b/arch/x86/include/asm/tsc.h index 6d7c547..174c421 100644 --- a/arch/x86/include/asm/tsc.h +++ b/arch/x86/include/asm/tsc.h @@ -29,6 +29,8 @@ static inline cycles_t get_cycles(void) return rdtsc(); } +extern struct system_counterval_t convert_art_to_tsc(cycle_t art); + extern void tsc_init(void); extern void mark_tsc_unstable(char *reason); extern int unsynchronized_tsc(void); diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c index 3d743da..8644b47 100644 --- a/arch/x86/kernel/tsc.c +++ b/arch/x86/kernel/tsc.c @@ -43,6 +43,11 @@ static DEFINE_STATIC_KEY_FALSE(__use_tsc); int tsc_clocksource_reliable; +static u32 art_to_tsc_numerator; +static u32 art_to_tsc_denominator; +static u64 art_to_tsc_offset; +struct clocksource *art_related_clocksource; + /* * Use a ring-buffer like data structure, where a writer advances the head by * writing a new data entry and a reader advances the tail when it observes a @@ -949,10 +954,41 @@ static struct notifier_block time_cpufreq_notifier_block = { .notifier_call = time_cpufreq_notifier }; +#define ART_CPUID_LEAF (0x15) +#define ART_MIN_DENOMINATOR (1) + + +/* + * If ART is present detect the numerator:denominator to convert to TSC + */ +static void detect_art(void) +{ + unsigned int unused[2]; + + if (boot_cpu_data.cpuid_level < ART_CPUID_LEAF) + return; + + cpuid(ART_CPUID_LEAF, &art_to_tsc_denominator, + &art_to_tsc_numerator, unused, unused+1); + + /* Don't enable ART in a VM, non-stop TSC required */ + if (boot_cpu_has(X86_FEATURE_HYPERVISOR) || + !boot_cpu_has(X86_FEATURE_NONSTOP_TSC) || + art_to_tsc_denominator < ART_MIN_DENOMINATOR) + return; + + if (rdmsrl_safe(MSR_IA32_TSC_ADJUST, &art_to_tsc_offset)) + return; + + /* Make this sticky over multiple CPU init calls */ + setup_force_cpu_cap(X86_FEATURE_ART); +} + static int __init cpufreq_tsc(void) { if (!cpu_has_tsc) return 0; + if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) return 0; cpufreq_register_notifier(&time_cpufreq_notifier_block, @@ -1071,6 +1107,25 @@ int unsynchronized_tsc(void) return 0; } +/* + * Convert ART to TSC given numerator/denominator found in detect_art() + */ +struct system_counterval_t convert_art_to_tsc(cycle_t art) +{ + u64 tmp, res, rem; + + rem = do_div(art, art_to_tsc_denominator); + + res = art * art_to_tsc_numerator; + tmp = rem * art_to_tsc_numerator; + + do_div(tmp, art_to_tsc_denominator); + res += tmp + art_to_tsc_offset; + + return (struct system_counterval_t) {.cs = art_related_clocksource, + .cycles = res}; +} +EXPORT_SYMBOL(convert_art_to_tsc); static void tsc_refine_calibration_work(struct work_struct *work); static DECLARE_DELAYED_WORK(tsc_irqwork, tsc_refine_calibration_work); @@ -1142,6 +1197,8 @@ static void tsc_refine_calibration_work(struct work_struct *work) (unsigned long)tsc_khz % 1000); out: + if (boot_cpu_has(X86_FEATURE_ART)) + art_related_clocksource = &clocksource_tsc; clocksource_register_khz(&clocksource_tsc, tsc_khz); } @@ -1235,6 +1292,8 @@ void __init tsc_init(void) mark_tsc_unstable("TSCs unsynchronized"); check_system_tsc_reliable(); + + detect_art(); } #ifdef CONFIG_SMP -- 2.1.4