x86/vmware: Add paravirt sched clock

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 <amakhalov@vmware.com>
Acked-by: Alok N Kataria <akataria@vmware.com>
Cc: linux-doc@vger.kernel.org
Cc: pv-drivers@vmware.com
Cc: corbet@lwn.net
Cc: virtualization@lists.linux-foundation.org
Link: http://lkml.kernel.org/r/20161028075432.90579-4-amakhalov@vmware.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
index 37babf9..b3b2ec0 100644 (file)
--- 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 (file)
--- a/arch/x86/kernel/cpu/vmware.c
+++ b/arch/x86/kernel/cpu/vmware.c
@@ -24,10 +24,15 @@
 #include <linux/dmi.h>
 #include <linux/init.h>
 #include <linux/export.h>
+#include <linux/clocksource.h>
 #include <asm/div64.h>
 #include <asm/x86_init.h>
 #include <asm/hypervisor.h>
 #include <asm/apic.h>
+#include <asm/timer.h>
+
+#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)