Attachment 317536 Details for Bug 463573 – x86: Code to detect if running on VMware.

[patch] x86: Code to detect if running on VMware.

vm_tsc_calibration.patch (text/plain), 14.24 KB, created by Alok Kataria on 2008-09-23 21:35:33 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Alok Kataria

Created: 2008-09-23 21:35:33 UTC

Size: 14.24 KB

patch

obsolete

>x86: Code to detect if running on VMware.
>
>From: Alok N Kataria <akataria@vmware.com>
>
>Use backdoor calls to detect whether running on VMware.
>If so get the tsc frequency from the hypervisor instead of calibrating it
>since calibration can be error prone in virtualized environment.
>
>Finally, if a VMware hypervisor is detected, skip the code that detects
>broken PIT wiring.  In a VM, this code can result in false positives,
>due to difference in timing.  We know that virtual hardware always has
>the correct wiring anyway.
>
>Signed-off-by: Alok N Kataria <akataria@vmware.com>
>Signed-off-by: Dan Hecht <dhecht@vmware.com>
>---
>
> arch/i386/kernel/cpu/common.c             |    9 +++
> arch/i386/kernel/io_apic.c                |   10 ++++
> arch/i386/kernel/tsc.c                    |   34 ++++++++++++
> arch/x86_64/kernel/io_apic.c              |   10 ++++
> arch/x86_64/kernel/setup.c                |   11 ++++
> arch/x86_64/kernel/time.c                 |   16 ++++++
> include/asm-i386/mach-xen/asm/processor.h |    4 +
> include/asm-i386/processor.h              |    4 +
> include/asm-x86_64/processor.h            |    4 +
> include/linux/vmware_backdoor.h           |   79 +++++++++++++++++++++++++++++
> 10 files changed, 181 insertions(+), 0 deletions(-)
> create mode 100644 include/linux/vmware_backdoor.h
>
>
>diff --git a/arch/i386/kernel/cpu/common.c b/arch/i386/kernel/cpu/common.c
>index f279008..bf2ba97 100644
>--- a/arch/i386/kernel/cpu/common.c
>+++ b/arch/i386/kernel/cpu/common.c
>@@ -5,6 +5,8 @@
> #include <linux/module.h>
> #include <linux/percpu.h>
> #include <linux/bootmem.h>
>+#include <linux/vmware_backdoor.h>
>+
> #include <asm/semaphore.h>
> #include <asm/processor.h>
> #include <asm/i387.h>
>@@ -181,6 +183,11 @@ static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c, int early)
> 	this_cpu = &default_cpu;
> }
> 
>+static void __cpuinit get_hypervisor_vendor(struct cpuinfo_x86 *c)
>+{
>+	if (vmware_platform())
>+		c->x86_hyper_vendor = X86_HYPER_VENDOR_VMWARE;
>+}
> 
> static int __init x86_fxsr_setup(char * s)
> {
>@@ -237,6 +244,7 @@ static void __init early_cpu_detect(void)
> 	struct cpuinfo_x86 *c = &boot_cpu_data;
> 
> 	c->x86_cache_alignment = 32;
>+	c->x86_hyper_vendor = X86_HYPER_VENDOR_NONE;
> 
> 	if (!have_cpuid_p())
> 		return;
>@@ -263,6 +271,7 @@ static void __init early_cpu_detect(void)
> 		if (cap0 & (1<<19))
> 			c->x86_cache_alignment = ((misc >> 8) & 0xff) * 8;
> 	}
>+	get_hypervisor_vendor(c);
> }
> 
> void __cpuinit generic_identify(struct cpuinfo_x86 * c)
>diff --git a/arch/i386/kernel/io_apic.c b/arch/i386/kernel/io_apic.c
>index 4fb32c5..79f3d9b 100644
>--- a/arch/i386/kernel/io_apic.c
>+++ b/arch/i386/kernel/io_apic.c
>@@ -18,6 +18,8 @@
>  *					and Rolf G. Tews
>  *					for testing these extensively
>  *	Paul Diefenbaugh	:	Added full ACPI support
>+ *
>+ *	Copyright (c) 2008, VMware, Inc. Skip timer_irq_works on hypervisors.
>  */
> 
> #include <linux/mm.h>
>@@ -1883,6 +1885,14 @@ static int __init timer_irq_works(void)
> 	unsigned long t1 = jiffies;
> 
> 	local_irq_enable();
>+
>+	/*
>+	 * Hypervisors don't emulate old broken boards, and this time
>+	 * sensitive test is prone to failure on a hypervisor, so skip it.
>+	 */
>+	if (boot_cpu_data.x86_hyper_vendor != X86_HYPER_VENDOR_NONE)
>+		return 1;
>+
> 	/* Let ten ticks pass... */
> 	mdelay((10 * 1000) / HZ);
> 
>diff --git a/arch/i386/kernel/tsc.c b/arch/i386/kernel/tsc.c
>index 16dbae7..eff9958 100644
>--- a/arch/i386/kernel/tsc.c
>+++ b/arch/i386/kernel/tsc.c
>@@ -2,6 +2,7 @@
>  * This code largely moved from arch/i386/kernel/timer/timer_tsc.c
>  * which was originally moved from arch/i386/kernel/time.c.
>  * See comments there for proper credits.
>+ * Copyright (c) 2008  VMware, Inc. Get TSC frequency from the hypervisor
>  */
> 
> #include <linux/clocksource.h>
>@@ -10,6 +11,7 @@
> #include <linux/jiffies.h>
> #include <linux/init.h>
> #include <linux/dmi.h>
>+#include <linux/vmware_backdoor.h>
> 
> #include <asm/delay.h>
> #include <asm/tsc.h>
>@@ -24,6 +26,7 @@
>  * an extra value to store the TSC freq
>  */
> unsigned int tsc_khz;
>+extern unsigned long preset_lpj;
> 
> int tsc_disable __cpuinitdata = 0;
> 
>@@ -202,6 +205,29 @@ void tsc_init(void)
> 	cpu_khz = calculate_cpu_khz();
> 	tsc_khz = cpu_khz;
> 
>+	if (boot_cpu_data.x86_hyper_vendor == X86_HYPER_VENDOR_VMWARE) {
>+		/* Ask the hypervisor for the TSC frequency since
>+		 * calibrating it inside a VM is prone to error. */
>+		unsigned long vm_tsc_khz = vmware_get_tsc_khz();
>+		if (vm_tsc_khz) {
>+			tsc_khz = vm_tsc_khz;
>+			cpu_khz = tsc_khz;
>+			preset_lpj = (vm_tsc_khz * 1000) / HZ;
>+			/*
>+			 * Enable "lazy" timer emulation.  Rather than holding
>+			 * back virtual time when timer interrupt delivery falls
>+			 * behind and attempting to "catch up", in lazy mode,
>+			 * missed periodic interrupts are skipped and virtual
>+			 * time always reflects real time.
>+			 */
>+			if (!vmware_enable_lazy_timer_emulation())
>+				printk(KERN_WARNING "time.c: failed to enable "
>+						"lazy timer emulation.\n");
>+		} else
>+			printk(KERN_WARNING "time.c: failed to get tsc "
>+					"frequency from hypervisor.\n");
>+	}
>+
> 	if (!cpu_khz)
> 		return;
> 
>@@ -414,6 +440,8 @@ static void verify_tsc_freq(unsigned long unused)
> 	u64 now_tsc, interval_tsc;
> 	unsigned long now_jiffies, interval_jiffies;
> 
>+	if (boot_cpu_data.x86_hyper_vendor == X86_HYPER_VENDOR_VMWARE)
>+		return;
> 
> 	if (check_tsc_unstable())
> 		return;
>@@ -458,6 +486,12 @@ static __init int unsynchronized_tsc(void)
> 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
>  		return 0;
> 
>+	/*
>+	 * TSC's are normally synchronized on a hypervisor.
>+	 */
>+	if (boot_cpu_data.x86_hyper_vendor == X86_HYPER_VENDOR_VMWARE)
>+		return 0;
>+
> 	/* assume multi socket systems are not synchronized: */
>  	return num_possible_cpus() > 1;
> }
>diff --git a/arch/x86_64/kernel/io_apic.c b/arch/x86_64/kernel/io_apic.c
>index 610d316..e6b1492 100644
>--- a/arch/x86_64/kernel/io_apic.c
>+++ b/arch/x86_64/kernel/io_apic.c
>@@ -18,6 +18,8 @@
>  *					and Rolf G. Tews
>  *					for testing these extensively
>  *	Paul Diefenbaugh	:	Added full ACPI support
>+ *
>+ *	Copyright (c) 2008, VMware, Inc. Skip timer_irq_works on hypervisors.
>  */
> 
> #include <linux/mm.h>
>@@ -1460,6 +1462,14 @@ static int __init timer_irq_works(void)
> 	unsigned long t1 = jiffies;
> 
> 	local_irq_enable();
>+
>+	/*
>+	 * Hypervisors don't emulate old broken boards, and this time
>+	 * sensitive test is prone to failure on a hypervisor, so skip it.
>+	 */
>+	if (boot_cpu_data.x86_hyper_vendor != X86_HYPER_VENDOR_NONE)
>+		return 1;
>+
> 	/* Let ten ticks pass... */
> 	mdelay((10 * 1000) / HZ);
> 
>diff --git a/arch/x86_64/kernel/setup.c b/arch/x86_64/kernel/setup.c
>index 19ca938..4365057 100644
>--- a/arch/x86_64/kernel/setup.c
>+++ b/arch/x86_64/kernel/setup.c
>@@ -5,6 +5,8 @@
>  *
>  *  Nov 2001 Dave Jones <davej@suse.de>
>  *  Forked from i386 setup code.
>+ *
>+ *  Copyright (c) 2008  VMware, Inc.  Detect if running on a VMware Hypervisor.
>  */
> 
> /*
>@@ -45,6 +47,7 @@
> #include <linux/dmi.h>
> #include <linux/dma-mapping.h>
> #include <linux/ctype.h>
>+#include <linux/vmware_backdoor.h>
> 
> #include <asm/mtrr.h>
> #include <asm/uaccess.h>
>@@ -1043,6 +1046,12 @@ static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
> 		c->x86_vendor = X86_VENDOR_UNKNOWN;
> }
> 
>+static void __cpuinit get_hypervisor_vendor(struct cpuinfo_x86 *c)
>+{
>+	if (vmware_platform())
>+		c->x86_hyper_vendor = X86_HYPER_VENDOR_VMWARE;
>+}
>+
> struct cpu_model_info {
> 	int vendor;
> 	int family;
>@@ -1066,6 +1075,7 @@ void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
> 	c->x86_cache_alignment = c->x86_clflush_size;
> 	c->x86_max_cores = 1;
> 	c->extended_cpuid_level = 0;
>+	c->x86_hyper_vendor = X86_HYPER_VENDOR_NONE;
> 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
> 
> 	/* Get vendor name */
>@@ -1075,6 +1085,7 @@ void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
> 	      (unsigned int *)&c->x86_vendor_id[4]);
> 		
> 	get_cpu_vendor(c);
>+	get_hypervisor_vendor(c);
> 
> 	/* Initialize the standard set of capabilities */
> 	/* Note that the vendor-specific code below might override */
>diff --git a/arch/x86_64/kernel/time.c b/arch/x86_64/kernel/time.c
>index 867760d..fab019d 100644
>--- a/arch/x86_64/kernel/time.c
>+++ b/arch/x86_64/kernel/time.c
>@@ -10,6 +10,7 @@
>  *  Copyright (c) 1998  Andrea Arcangeli
>  *  Copyright (c) 2002,2006  Vojtech Pavlik
>  *  Copyright (c) 2003  Andi Kleen
>+ *  Copyright (c) 2008  VMware, Inc. get TSC frequency from the hypervisor.
>  *  RTC support code taken from arch/i386/kernel/timers/time_hpet.c
>  */
> 
>@@ -28,6 +29,7 @@
> #include <linux/cpu.h>
> #include <linux/kallsyms.h>
> #include <linux/acpi.h>
>+#include <linux/vmware_backdoor.h>
> #ifdef CONFIG_ACPI
> #include <acpi/achware.h>	/* for PM timer frequency */
> #include <acpi/acpi_bus.h>
>@@ -89,6 +91,7 @@ unsigned long __wall_jiffies __section_wall_jiffies = INITIAL_JIFFIES;
> struct timespec __xtime __section_xtime;
> struct timezone __sys_tz __section_sys_tz;
> 
>+extern unsigned long preset_lpj;
> /*
>  * do_gettimeoffset() returns microseconds since last timer interrupt was
>  * triggered by hardware. A memory read of HPET is slower than a register read
>@@ -996,6 +999,19 @@ void __init time_init(void)
> 		boot_cpu_data.x86 == 16)
> 		cpu_khz = tsc_calibrate_cpu_khz();
> 
>+	if (boot_cpu_data.x86_hyper_vendor == X86_HYPER_VENDOR_VMWARE) {
>+		/* Ask the hypervisor for the TSC frequency since
>+		 * calibrating it inside a VM is prone to error. */
>+		unsigned long vm_tsc_khz = vmware_get_tsc_khz();
>+		if (vm_tsc_khz) {
>+			tsc_khz = vm_tsc_khz;
>+			cpu_khz = tsc_khz;
>+			preset_lpj = (vm_tsc_khz * 1000) / HZ;
>+		} else
>+			printk(KERN_WARNING "time.c: failed to get tsc "
>+					"frequency from hypervisor.\n");
>+	}
>+
> 	vxtime.mode = VXTIME_TSC;
> 	vxtime.quot = (USEC_PER_SEC << US_SCALE) / vxtime_hz;
> 	vxtime.tsc_quot = (USEC_PER_MSEC << US_SCALE) / cpu_khz;
>diff --git a/include/asm-i386/mach-xen/asm/processor.h b/include/asm-i386/mach-xen/asm/processor.h
>index d8a4b7b..e5e3402 100644
>--- a/include/asm-i386/mach-xen/asm/processor.h
>+++ b/include/asm-i386/mach-xen/asm/processor.h
>@@ -78,6 +78,7 @@ struct cpuinfo_x86 {
> 	__u8 phys_proc_id; 		/* Physical processor id. */
> 	__u8 cpu_core_id;  		/* Core id */
> #endif
>+	__u8    x86_hyper_vendor;
> } __attribute__((__aligned__(SMP_CACHE_BYTES)));
> 
> #define X86_VENDOR_INTEL 0
>@@ -92,6 +93,9 @@ struct cpuinfo_x86 {
> #define X86_VENDOR_NUM 9
> #define X86_VENDOR_UNKNOWN 0xff
> 
>+#define X86_HYPER_VENDOR_NONE 0
>+#define X86_HYPER_VENDOR_VMWARE 1
>+
> /*
>  * capabilities of CPUs
>  */
>diff --git a/include/asm-i386/processor.h b/include/asm-i386/processor.h
>index c108225..6977965 100644
>--- a/include/asm-i386/processor.h
>+++ b/include/asm-i386/processor.h
>@@ -77,6 +77,7 @@ struct cpuinfo_x86 {
> 	__u8 phys_proc_id; 		/* Physical processor id. */
> 	__u8 cpu_core_id;  		/* Core id */
> #endif
>+	__u8    x86_hyper_vendor;
> } __attribute__((__aligned__(SMP_CACHE_BYTES)));
> 
> #define X86_VENDOR_INTEL 0
>@@ -91,6 +92,9 @@ struct cpuinfo_x86 {
> #define X86_VENDOR_NUM 9
> #define X86_VENDOR_UNKNOWN 0xff
> 
>+#define X86_HYPER_VENDOR_NONE 0
>+#define X86_HYPER_VENDOR_VMWARE 1
>+
> /*
>  * capabilities of CPUs
>  */
>diff --git a/include/asm-x86_64/processor.h b/include/asm-x86_64/processor.h
>index 2dd19a7..d7331f5 100644
>--- a/include/asm-x86_64/processor.h
>+++ b/include/asm-x86_64/processor.h
>@@ -74,6 +74,7 @@ struct cpuinfo_x86 {
> 	__u8	phys_proc_id;	/* Physical Processor id. */
> 	__u8	cpu_core_id;	/* Core id. */
> #endif
>+	__u8    x86_hyper_vendor;
> } ____cacheline_aligned;
> 
> #define X86_VENDOR_INTEL 0
>@@ -87,6 +88,9 @@ struct cpuinfo_x86 {
> #define X86_VENDOR_NUM 8
> #define X86_VENDOR_UNKNOWN 0xff
> 
>+#define X86_HYPER_VENDOR_NONE 0
>+#define X86_HYPER_VENDOR_VMWARE 1
>+
> #ifdef CONFIG_SMP
> extern struct cpuinfo_x86 cpu_data[];
> #define current_cpu_data cpu_data[smp_processor_id()]
>diff --git a/include/linux/vmware_backdoor.h b/include/linux/vmware_backdoor.h
>new file mode 100644
>index 0000000..49a31fc
>--- /dev/null
>+++ b/include/linux/vmware_backdoor.h
>@@ -0,0 +1,79 @@
>+/*
>+ * Copyright (C) 2007-2008, VMware, Inc.
>+ *
>+ * This program is free software; you can redistribute it and/or modify
>+ * it under the terms of the GNU General Public License as published by
>+ * the Free Software Foundation, version 2 and no later version.
>+ *
>+ * This program is distributed in the hope that it will be useful, but
>+ * WITHOUT ANY WARRANTY; without even the implied warranty of
>+ * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
>+ * NON INFRINGEMENT.  See the GNU General Public License for more
>+ * details.
>+ *
>+ * You should have received a copy of the GNU General Public License
>+ * along with this program; if not, write to the Free Software
>+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
>+ *
>+ * Send any feedback to akataria@vmware.com, dhecht@vmware.com
>+ */
>+
>+#ifndef _VMWARE_BACKDOOR_H
>+#define _VMWARE_BACKDOOR_H
>+
>+#include <linux/types.h>
>+
>+#define VMWARE_BDOOR_MAGIC     0x564D5868
>+#define VMWARE_BDOOR_PORT      0x5658
>+
>+#define VMWARE_BDOOR_CMD_GETVERSION         10
>+#define VMWARE_BDOOR_CMD_GETHZ              45
>+#define VMWARE_BDOOR_CMD_LAZYTIMEREMULATION 49
>+
>+#define VMWARE_BDOOR(cmd, eax, ebx, ecx, edx)                         \
>+	__asm__("inl (%%dx)" :                                        \
>+		"=a"(eax), "=c"(ecx), "=d"(edx), "=b"(ebx) :          \
>+		"0"(VMWARE_BDOOR_MAGIC), "1"(VMWARE_BDOOR_CMD_##cmd), \
>+		"2"(VMWARE_BDOOR_PORT), "3"(0) :                      \
>+		"memory");
>+
>+
>+static inline int vmware_platform(void)
>+{
>+	uint32_t eax, ebx, ecx, edx;
>+	VMWARE_BDOOR(GETVERSION, eax, ebx, ecx, edx);
>+	return eax != (uint32_t)-1 && ebx == VMWARE_BDOOR_MAGIC;
>+}
>+
>+static inline unsigned int vmware_get_tsc_khz(void)
>+{
>+	uint64_t tsc_hz;
>+	uint32_t eax, ebx, ecx, edx;
>+
>+	VMWARE_BDOOR(GETHZ, eax, ebx, ecx, edx);
>+
>+	if (eax == (uint32_t)-1)
>+		return 0;
>+	tsc_hz = eax | (((uint64_t)ebx) << 32);
>+	do_div(tsc_hz, 1000);
>+	BUG_ON(tsc_hz >> 32);
>+	return tsc_hz;
>+}
>+
>+/*
>+ * Usually, all timer devices in the VM are kept consistent, but this
>+ * can cause time in the VM to fall behind (e.g., if the VM is
>+ * descheduled for long periods of time, the hypervisor may not have had
>+ * an opportunity to deliver timer interrupts, and so the time in the VM
>+ * will be behind until it can catch up.  In lazy timer emulation mode,
>+ * the TSC tracks real time regardless of whether e.g. timer interrupts
>+ * are delivered on time.
>+ */
>+static inline int vmware_enable_lazy_timer_emulation(void)
>+{
>+	uint32_t eax, ebx, ecx, edx;
>+	VMWARE_BDOOR(LAZYTIMEREMULATION, eax, ebx, ecx, edx);
>+	return ebx == VMWARE_BDOOR_MAGIC;
>+}
>+
>+#endif

Actions: View | Diff

Attachments on bug 463573: 317536 | 317537 | 317538 | 317727 | 317728 | 317734 | 317735 | 319716 | 319717 | 319718 | 319719 | 319720 | 319799 | 319817 | 320098 | 320133 | 320485 | 321043 | 325724 | 325725 | 325726 | 325727 | 325728 | 326672 | 329135 | 329213 | 332979