Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1758921AbcJ1HzD (ORCPT ); Fri, 28 Oct 2016 03:55:03 -0400 Received: from ex13-edg-ou-001.vmware.com ([208.91.0.189]:33174 "EHLO EX13-EDG-OU-001.vmware.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757718AbcJ1Hyw (ORCPT ); Fri, 28 Oct 2016 03:54:52 -0400 From: Alexey Makhalov To: , , , , , CC: , , , , Alexey Makhalov Subject: [PATCH v3 3/3] x86/vmware: Add paravirt sched clock Date: Fri, 28 Oct 2016 00:54:32 -0700 Message-ID: <20161028075432.90579-4-amakhalov@vmware.com> X-Mailer: git-send-email 2.10.1 In-Reply-To: <20161028075432.90579-1-amakhalov@vmware.com> References: <20161028075432.90579-1-amakhalov@vmware.com> MIME-Version: 1.0 Content-Type: text/plain Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 3428 Lines: 106 The default sched_clock() implementation is native_sched_clock(). It contains code to handle non constant frequency TSCs, which creates overhead for systems with constant frequency TSCs. The vmware hypervisor guarantees a constant frequency TSC, so native_sched_clock() is not required and slower than a dedicated function which operates with one time calculated conversion factors. Calculate the conversion factors at boot time from the tsc frequency and install an optimized sched_clock() function via paravirt ops. The paravirtualized clock can be disabled on the kernel command line with the new 'no-vmw-sched-clock' option. Signed-off-by: Alexey Makhalov Acked-by: Alok N Kataria --- Documentation/kernel-parameters.txt | 4 ++++ arch/x86/kernel/cpu/vmware.c | 42 +++++++++++++++++++++++++++++++++++++ 2 files changed, 46 insertions(+) diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt index 37babf9..b3b2ec0 100644 --- a/Documentation/kernel-parameters.txt +++ b/Documentation/kernel-parameters.txt @@ -2754,6 +2754,10 @@ bytes respectively. Such letter suffixes can also be entirely omitted. no-kvmapf [X86,KVM] Disable paravirtualized asynchronous page fault handling. + no-vmw-sched-clock + [X86,PV_OPS] Disable paravirtualized VMware scheduler + clock and use the default one. + no-steal-acc [X86,KVM] Disable paravirtualized steal time accounting. steal time is computed, but won't influence scheduler behaviour diff --git a/arch/x86/kernel/cpu/vmware.c b/arch/x86/kernel/cpu/vmware.c index 098a524..cdbe38b 100644 --- a/arch/x86/kernel/cpu/vmware.c +++ b/arch/x86/kernel/cpu/vmware.c @@ -24,10 +24,15 @@ #include #include #include +#include #include #include #include #include +#include + +#undef pr_fmt +#define pr_fmt(fmt) "vmware: " fmt #define CPUID_VMWARE_INFO_LEAF 0x40000000 #define VMWARE_HYPERVISOR_MAGIC 0x564D5868 @@ -62,10 +67,47 @@ static unsigned long vmware_get_tsc_khz(void) } #ifdef CONFIG_PARAVIRT +static struct cyc2ns_data vmware_cyc2ns __ro_after_init; +static int vmw_sched_clock __initdata = 1; + +static __init int setup_vmw_sched_clock(char *s) +{ + vmw_sched_clock = 0; + return 0; +} +early_param("no-vmw-sched-clock", setup_vmw_sched_clock); + +static unsigned long long vmware_sched_clock(void) +{ + unsigned long long ns; + + ns = mul_u64_u32_shr(rdtsc(), vmware_cyc2ns.cyc2ns_mul, + vmware_cyc2ns.cyc2ns_shift); + ns -= vmware_cyc2ns.cyc2ns_offset; + return ns; +} + +static void __init vmware_sched_clock_setup(void) +{ + struct cyc2ns_data *d = &vmware_cyc2ns; + unsigned long long tsc_now = rdtsc(); + + clocks_calc_mult_shift(&d->cyc2ns_mul, &d->cyc2ns_shift, + vmware_tsc_khz, NSEC_PER_MSEC, 0); + d->cyc2ns_offset = mul_u64_u32_shr(tsc_now, d->cyc2ns_mul, + d->cyc2ns_shift); + + pv_time_ops.sched_clock = vmware_sched_clock; + pr_info("using sched offset of %llu ns\n", d->cyc2ns_offset); +} + static void __init vmware_paravirt_ops_setup(void) { pv_info.name = "VMware hypervisor"; pv_cpu_ops.io_delay = paravirt_nop; + + if (vmware_tsc_khz && vmw_sched_clock) + vmware_sched_clock_setup(); } #else #define vmware_paravirt_ops_setup() do {} while (0) -- 2.10.1