[PATCH v4 0/6] xen/PMU: PMU support for Xen PV guests

[PATCH v4 0/6] xen/PMU: PMU support for Xen PV guests

[Boris Ostrovsky]

Changes in v4:

* Don't write to APIC while processing PMU interrupt, defer until PMU flush
* pmu_mode_store() will verify whole string, not just first strlen(mode)
  characters
* Use get_zeroed_page() instead of vzalloc() in xen_pmu_init()
* Replace per_cpu() with this_cpu_ptr() where possible

Changes in v3:

* Handle cases when hypervisor symbols strings don't fit into the buffer
provided by kernel. (Buffer size is no longer a constant shared between
Xen and kernel)
* Reworked data structures in xensym.c to handle better the fact that
  buffer is a Xen handle.
* Changed VPMU modes' names to "self", "all", "off"
* Moved a few (fairly minor) things around to make things bisectable
* Added Documentation/ABI/testing/sysfs-hypervisor-pmu
* Added a blurb to describe XEN_SYMS in Kconfig
* Code style changes


This is the Linux side of Xen PMU support for PV(H) guests, including
dom0. Only kernel changes are here, toolstack patch will be provided
separately.

Here is description from the hypervisor patch submission that applies
to this series as well:

This version has following limitations:
* For accurate profiling of dom0/Xen dom0 VCPUs should be pinned.
* Hypervisor code is only profiled on processors that have running dom0 VCPUs
on them.
* No backtrace support.
* Will fail to load under XSM: we ran out of bits in permissions vector and
this needs to be fixed separately

A few notes that may help reviewing:

* A shared data structure (xenpmu_data_t) between each PV VPCU and hypervisor
CPU is used for passing registers' values as well as PMU state at the time of
PMU interrupt.
* PMU interrupts are taken by hypervisor either as NMIs or regular vector
interrupts for both HVM and PV(H). The interrupts are sent as NMIs to HVM guests
and as virtual interrupts to PV(H) guests
* PV guest's interrupt handler does not read/write PMU MSRs directly. Instead, it
accesses xenpmu_data_t and flushes it to HW it before returning.
* PMU mode is controlled at runtime via /sys/hypervisor/pmu/pmu/{pmu_mode,pmu_flags}
in addition to 'vpmu' boot option (which is preserved for back compatibility).
The following modes are provided:
  * disable: VPMU is off
  * enable: VPMU is on. Guests can profile themselves, dom0 profiles itself and Xen
  * priv_enable: dom0 only profiling. dom0 collects samples for everyone. Sampling
    in guests is suspended.
* /proc/xen/xensyms file exports hypervisor's symbols to dom0 (similar to
/proc/kallsyms)
* VPMU infrastructure is now used for HVM, PV and PVH and therefore has been moved
up from hvm subtree


Boris Ostrovsky (6):
  xen: xensyms support
  xen/PMU: Sysfs interface for setting Xen PMU mode
  xen/PMU: Initialization code for Xen PMU
  xen/PMU: Describe vendor-specific PMU registers
  xen/PMU: Intercept PMU-related MSR and APIC accesses
  xen/PMU: PMU emulation code

 Documentation/ABI/testing/sysfs-hypervisor-pmu |  22 +
 arch/x86/include/asm/xen/hypercall.h           |   6 +
 arch/x86/include/asm/xen/interface.h           |  41 ++
 arch/x86/xen/Makefile                          |   2 +-
 arch/x86/xen/enlighten.c                       |  24 +-
 arch/x86/xen/pmu.c                             | 546 +++++++++++++++++++++++++
 arch/x86/xen/pmu.h                             |  15 +
 arch/x86/xen/smp.c                             |  31 +-
 arch/x86/xen/xen-head.S                        |   5 +-
 drivers/xen/Kconfig                            |   8 +
 drivers/xen/sys-hypervisor.c                   | 118 ++++++
 drivers/xen/xenfs/Makefile                     |   1 +
 drivers/xen/xenfs/super.c                      |   3 +
 drivers/xen/xenfs/xenfs.h                      |   1 +
 drivers/xen/xenfs/xensyms.c                    | 152 +++++++
 include/xen/interface/platform.h               |  19 +
 include/xen/interface/xen.h                    |   2 +
 include/xen/interface/xenpmu.h                 |  67 +++
 18 files changed, 1057 insertions(+), 6 deletions(-)
 create mode 100644 Documentation/ABI/testing/sysfs-hypervisor-pmu
 create mode 100644 arch/x86/xen/pmu.c
 create mode 100644 arch/x86/xen/pmu.h
 create mode 100644 drivers/xen/xenfs/xensyms.c
 create mode 100644 include/xen/interface/xenpmu.h

-- 
1.8.1.4

[PATCH v4 1/6] xen: xensyms support

[Boris Ostrovsky]

Export Xen symbols to dom0 via /proc/xen/xensyms (similar to /proc/kallsyms).

Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Reviewed-by: David Vrabel <david.vrabel@citrix.com>
---
 drivers/xen/Kconfig              |   8 +++
 drivers/xen/xenfs/Makefile       |   1 +
 drivers/xen/xenfs/super.c        |   3 +
 drivers/xen/xenfs/xenfs.h        |   1 +
 drivers/xen/xenfs/xensyms.c      | 152 +++++++++++++++++++++++++++++++++++++++
 include/xen/interface/platform.h |  19 +++++
 6 files changed, 184 insertions(+)
 create mode 100644 drivers/xen/xenfs/xensyms.c

diff --git a/drivers/xen/Kconfig b/drivers/xen/Kconfig
index 38fb36e..01c3d4e 100644
--- a/drivers/xen/Kconfig
+++ b/drivers/xen/Kconfig
@@ -240,4 +240,12 @@ config XEN_MCE_LOG
 config XEN_HAVE_PVMMU
        bool
 
+config XEN_SYMS
+       bool "Xen symbols"
+       depends on XEN_DOM0 && XENFS
+       default y if KALLSYMS
+       help
+          Exports hypervisor symbols (along with their types and addresses) via
+          /proc/xen/xensyms file, similar to /proc/kallsyms
+
 endmenu
diff --git a/drivers/xen/xenfs/Makefile b/drivers/xen/xenfs/Makefile
index b019865..1a83010 100644
--- a/drivers/xen/xenfs/Makefile
+++ b/drivers/xen/xenfs/Makefile
@@ -2,3 +2,4 @@ obj-$(CONFIG_XENFS) += xenfs.o
 
 xenfs-y			  = super.o
 xenfs-$(CONFIG_XEN_DOM0) += xenstored.o
+xenfs-$(CONFIG_XEN_SYMS) += xensyms.o
diff --git a/drivers/xen/xenfs/super.c b/drivers/xen/xenfs/super.c
index 06092e0..8559a71 100644
--- a/drivers/xen/xenfs/super.c
+++ b/drivers/xen/xenfs/super.c
@@ -57,6 +57,9 @@ static int xenfs_fill_super(struct super_block *sb, void *data, int silent)
 		{ "privcmd", &xen_privcmd_fops, S_IRUSR|S_IWUSR },
 		{ "xsd_kva", &xsd_kva_file_ops, S_IRUSR|S_IWUSR},
 		{ "xsd_port", &xsd_port_file_ops, S_IRUSR|S_IWUSR},
+#ifdef CONFIG_XEN_SYMS
+		{ "xensyms", &xensyms_ops, S_IRUSR},
+#endif
 		{""},
 	};
 
diff --git a/drivers/xen/xenfs/xenfs.h b/drivers/xen/xenfs/xenfs.h
index 6b80c77..2c5934e 100644
--- a/drivers/xen/xenfs/xenfs.h
+++ b/drivers/xen/xenfs/xenfs.h
@@ -3,5 +3,6 @@
 
 extern const struct file_operations xsd_kva_file_ops;
 extern const struct file_operations xsd_port_file_ops;
+extern const struct file_operations xensyms_ops;
 
 #endif	/* _XENFS_XENBUS_H */
diff --git a/drivers/xen/xenfs/xensyms.c b/drivers/xen/xenfs/xensyms.c
new file mode 100644
index 0000000..ed16f8d
--- /dev/null
+++ b/drivers/xen/xenfs/xensyms.c
@@ -0,0 +1,152 @@
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/seq_file.h>
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/proc_fs.h>
+#include <linux/slab.h>
+#include <xen/interface/platform.h>
+#include <asm/xen/hypercall.h>
+#include <xen/xen-ops.h>
+#include "xenfs.h"
+
+
+#define XEN_KSYM_NAME_LEN 127 /* Hypervisor may have different name length */
+
+struct xensyms {
+	struct xen_platform_op op;
+	char *name;
+	uint32_t namelen;
+};
+
+/* Grab next output page from the hypervisor */
+static int xensyms_next_sym(struct xensyms *xs)
+{
+	int ret;
+	struct xenpf_symdata *symdata = &xs->op.u.symdata;
+	uint64_t symnum;
+
+	memset(xs->name, 0, xs->namelen);
+	symdata->namelen = xs->namelen;
+
+	symnum = symdata->symnum;
+
+	ret = HYPERVISOR_dom0_op(&xs->op);
+	if (ret < 0)
+		return ret;
+
+	/*
+	 * If hypervisor's symbol didn't fit into the buffer then allocate
+	 * a larger buffer and try again
+	 */
+	if (unlikely(symdata->namelen > xs->namelen)) {
+		kfree(xs->name);
+
+		xs->namelen = symdata->namelen;
+		xs->name = kzalloc(xs->namelen, GFP_KERNEL);
+		if (!xs->name)
+			return 1;
+
+		set_xen_guest_handle(symdata->name, xs->name);
+		symdata->symnum--; /* Rewind */
+
+		ret = HYPERVISOR_dom0_op(&xs->op);
+		if (ret < 0)
+			return ret;
+	}
+
+	if (symdata->symnum == symnum)
+		/* End of symbols */
+		return 1;
+
+	return 0;
+}
+
+static void *xensyms_start(struct seq_file *m, loff_t *pos)
+{
+	struct xensyms *xs = (struct xensyms *)m->private;
+
+	xs->op.u.symdata.symnum = *pos;
+
+	if (xensyms_next_sym(xs))
+		return NULL;
+
+	return m->private;
+}
+
+static void *xensyms_next(struct seq_file *m, void *p, loff_t *pos)
+{
+	struct xensyms *xs = (struct xensyms *)m->private;
+
+	xs->op.u.symdata.symnum = ++(*pos);
+
+	if (xensyms_next_sym(xs))
+		return NULL;
+
+	return p;
+}
+
+static int xensyms_show(struct seq_file *m, void *p)
+{
+	struct xensyms *xs = (struct xensyms *)m->private;
+	struct xenpf_symdata *symdata = &xs->op.u.symdata;
+
+	seq_printf(m, "%016llx %c %s\n", symdata->address,
+		   symdata->type, xs->name);
+
+	return 0;
+}
+
+static void xensyms_stop(struct seq_file *m, void *p)
+{
+}
+
+static const struct seq_operations xensyms_seq_ops = {
+	.start = xensyms_start,
+	.next = xensyms_next,
+	.show = xensyms_show,
+	.stop = xensyms_stop,
+};
+
+static int xensyms_open(struct inode *inode, struct file *file)
+{
+	struct seq_file *m;
+	struct xensyms *xs;
+	int ret;
+
+	ret = seq_open_private(file, &xensyms_seq_ops,
+			       sizeof(struct xensyms));
+	if (ret)
+		return ret;
+
+	m = file->private_data;
+	xs = (struct xensyms *)m->private;
+
+	xs->namelen = XEN_KSYM_NAME_LEN + 1;
+	xs->name = kzalloc(xs->namelen, GFP_KERNEL);
+	if (!xs->name) {
+		seq_release_private(inode, file);
+		return -ENOMEM;
+	}
+	set_xen_guest_handle(xs->op.u.symdata.name, xs->name);
+	xs->op.cmd = XENPF_get_symbol;
+	xs->op.u.symdata.namelen = xs->namelen;
+
+	return 0;
+}
+
+static int xensyms_release(struct inode *inode, struct file *file)
+{
+	struct seq_file *m = file->private_data;
+	struct xensyms *xs = (struct xensyms *)m->private;
+
+	kfree(xs->name);
+	return seq_release_private(inode, file);
+}
+
+const struct file_operations xensyms_ops = {
+	.open = xensyms_open,
+	.read = seq_read,
+	.llseek = seq_lseek,
+	.release = xensyms_release
+};
diff --git a/include/xen/interface/platform.h b/include/xen/interface/platform.h
index f1331e3..a3499f8 100644
--- a/include/xen/interface/platform.h
+++ b/include/xen/interface/platform.h
@@ -352,6 +352,24 @@ struct xenpf_core_parking {
 };
 DEFINE_GUEST_HANDLE_STRUCT(xenpf_core_parking);
 
+#define XENPF_get_symbol      61
+struct xenpf_symdata {
+	/* IN/OUT variables */
+	uint32_t	namelen; /* size of 'name' buffer */
+
+	/* IN/OUT variables */
+	uint32_t	symnum; /* IN:  Symbol to read                       */
+				/* OUT: Next available symbol. If same as IN */
+				/* then  we reached the end                  */
+
+	/* OUT variables */
+	char		type;
+	uint64_t	address;
+
+	GUEST_HANDLE(char) name;
+};
+DEFINE_GUEST_HANDLE_STRUCT(xenpf_symdata);
+
 struct xen_platform_op {
 	uint32_t cmd;
 	uint32_t interface_version; /* XENPF_INTERFACE_VERSION */
@@ -372,6 +390,7 @@ struct xen_platform_op {
 		struct xenpf_cpu_hotadd        cpu_add;
 		struct xenpf_mem_hotadd        mem_add;
 		struct xenpf_core_parking      core_parking;
+		struct xenpf_symdata           symdata;
 		uint8_t                        pad[128];
 	} u;
 };
-- 
1.8.1.4

[PATCH v4 2/6] xen/PMU: Sysfs interface for setting Xen PMU mode

[Boris Ostrovsky]

Set Xen's PMU mode via /sys/hypervisor/pmu/pmu_mode. Add XENPMU hypercall.

Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
---
 Documentation/ABI/testing/sysfs-hypervisor-pmu |  22 +++++
 arch/x86/include/asm/xen/hypercall.h           |   6 ++
 arch/x86/xen/xen-head.S                        |   5 +-
 drivers/xen/sys-hypervisor.c                   | 118 +++++++++++++++++++++++++
 include/xen/interface/xen.h                    |   1 +
 include/xen/interface/xenpmu.h                 |  48 ++++++++++
 6 files changed, 199 insertions(+), 1 deletion(-)
 create mode 100644 Documentation/ABI/testing/sysfs-hypervisor-pmu
 create mode 100644 include/xen/interface/xenpmu.h

diff --git a/Documentation/ABI/testing/sysfs-hypervisor-pmu b/Documentation/ABI/testing/sysfs-hypervisor-pmu
new file mode 100644
index 0000000..3123a5d
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-hypervisor-pmu
@@ -0,0 +1,22 @@
+What:		/sys/hypervisor/pmu/pmu_mode
+Date:		July 2014
+KernelVersion:	3.17
+Contact:	Boris Ostrovsky <boris.ostrovsky@oracle.com>
+Description:
+		Describes mode that Xen's performance-monitoring unit (PMU)
+		uses. Accepted values are
+			"off"  -- PMU is disabled
+			"self" -- The guest can profile itself and, if it is
+				  privileged (e.g. dom0), the hypervisor
+			"all" --  The guest profiles itself, the hypervisor and
+				  all other guests. Only available to privileged
+				  guests.
+
+What:           /sys/hypervisor/pmu/pmu_features
+Date:           July 2014
+KernelVersion:  3.17
+Contact:        Boris Ostrovsky <boris.ostrovsky@oracle.com>
+Description:
+		Describes Xen PMU features (as an integer). A set bit indicates
+		that the corresponding feature is enabled. See
+		include/xen/interface/xenpmu.h for available features
diff --git a/arch/x86/include/asm/xen/hypercall.h b/arch/x86/include/asm/xen/hypercall.h
index ca08a27..83aea80 100644
--- a/arch/x86/include/asm/xen/hypercall.h
+++ b/arch/x86/include/asm/xen/hypercall.h
@@ -465,6 +465,12 @@ HYPERVISOR_tmem_op(
 	return _hypercall1(int, tmem_op, op);
 }
 
+static inline int
+HYPERVISOR_xenpmu_op(unsigned int op, void *arg)
+{
+	return _hypercall2(int, xenpmu_op, op, arg);
+}
+
 static inline void
 MULTI_fpu_taskswitch(struct multicall_entry *mcl, int set)
 {
diff --git a/arch/x86/xen/xen-head.S b/arch/x86/xen/xen-head.S
index 485b695..628852f 100644
--- a/arch/x86/xen/xen-head.S
+++ b/arch/x86/xen/xen-head.S
@@ -93,8 +93,11 @@ NEXT_HYPERCALL(sysctl)
 NEXT_HYPERCALL(domctl)
 NEXT_HYPERCALL(kexec_op)
 NEXT_HYPERCALL(tmem_op) /* 38 */
+ENTRY(xenclient_rsvd)
+	.skip 32
+NEXT_HYPERCALL(xenpmu_op) /* 40 */
 ENTRY(xen_hypercall_rsvr)
-	.skip 320
+	.skip 256
 NEXT_HYPERCALL(mca) /* 48 */
 NEXT_HYPERCALL(arch_1)
 NEXT_HYPERCALL(arch_2)
diff --git a/drivers/xen/sys-hypervisor.c b/drivers/xen/sys-hypervisor.c
index 96453f8..d8d9bee 100644
--- a/drivers/xen/sys-hypervisor.c
+++ b/drivers/xen/sys-hypervisor.c
@@ -20,6 +20,7 @@
 #include <xen/xenbus.h>
 #include <xen/interface/xen.h>
 #include <xen/interface/version.h>
+#include <xen/interface/xenpmu.h>
 
 #define HYPERVISOR_ATTR_RO(_name) \
 static struct hyp_sysfs_attr  _name##_attr = __ATTR_RO(_name)
@@ -368,6 +369,115 @@ static void xen_properties_destroy(void)
 	sysfs_remove_group(hypervisor_kobj, &xen_properties_group);
 }
 
+struct pmu_mode {
+	const char *name;
+	uint32_t mode;
+};
+
+struct pmu_mode pmu_modes[] = {
+	{"self", XENPMU_MODE_SELF},
+	{"all", XENPMU_MODE_ALL},
+	{"off", XENPMU_MODE_OFF}
+};
+
+static ssize_t pmu_mode_store(struct hyp_sysfs_attr *attr,
+			      const char *buffer, size_t len)
+{
+	int ret;
+	struct xen_pmu_params xp;
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(pmu_modes); i++) {
+		if (strncmp(buffer, pmu_modes[i].name, len - 1) == 0) {
+			xp.val = pmu_modes[i].mode;
+			break;
+		}
+	}
+
+	if (i == ARRAY_SIZE(pmu_modes))
+		return -EINVAL;
+
+	ret = HYPERVISOR_xenpmu_op(XENPMU_mode_set, &xp);
+	if (ret)
+		return ret;
+
+	return len;
+}
+
+static ssize_t pmu_mode_show(struct hyp_sysfs_attr *attr, char *buffer)
+{
+	int ret;
+	struct xen_pmu_params xp;
+	int i;
+	uint32_t mode;
+
+	ret = HYPERVISOR_xenpmu_op(XENPMU_mode_get, &xp);
+	if (ret)
+		return ret;
+
+	mode = (uint32_t)xp.val;
+	for (i = 0; i < ARRAY_SIZE(pmu_modes); i++) {
+		if (mode == pmu_modes[i].mode)
+			return sprintf(buffer, "%s\n", pmu_modes[i].name);
+	}
+
+	return -EINVAL;
+}
+HYPERVISOR_ATTR_RW(pmu_mode);
+
+static ssize_t pmu_features_store(struct hyp_sysfs_attr *attr,
+				  const char *buffer, size_t len)
+{
+	int ret;
+	uint32_t features;
+	struct xen_pmu_params xp;
+
+	ret = kstrtou32(buffer, 0, &features);
+	if (ret)
+		return ret;
+
+	xp.val = features;
+	ret = HYPERVISOR_xenpmu_op(XENPMU_feature_set, &xp);
+	if (ret)
+		return ret;
+
+	return len;
+}
+
+static ssize_t pmu_features_show(struct hyp_sysfs_attr *attr, char *buffer)
+{
+	int ret;
+	struct xen_pmu_params xp;
+
+	ret = HYPERVISOR_xenpmu_op(XENPMU_feature_get, &xp);
+	if (ret)
+		return ret;
+
+	return sprintf(buffer, "0x%x\n", (uint32_t)xp.val);
+}
+HYPERVISOR_ATTR_RW(pmu_features);
+
+static struct attribute *xen_pmu_attrs[] = {
+	&pmu_mode_attr.attr,
+	&pmu_features_attr.attr,
+	NULL
+};
+
+static const struct attribute_group xen_pmu_group = {
+	.name = "pmu",
+	.attrs = xen_pmu_attrs,
+};
+
+static int __init xen_pmu_init(void)
+{
+	return sysfs_create_group(hypervisor_kobj, &xen_pmu_group);
+}
+
+static void xen_pmu_destroy(void)
+{
+	sysfs_remove_group(hypervisor_kobj, &xen_pmu_group);
+}
+
 static int __init hyper_sysfs_init(void)
 {
 	int ret;
@@ -390,9 +500,16 @@ static int __init hyper_sysfs_init(void)
 	ret = xen_properties_init();
 	if (ret)
 		goto prop_out;
+	if (xen_initial_domain()) {
+		ret = xen_pmu_init();
+		if (ret)
+			goto pmu_out;
+	}
 
 	goto out;
 
+pmu_out:
+	xen_properties_destroy();
 prop_out:
 	xen_sysfs_uuid_destroy();
 uuid_out:
@@ -407,6 +524,7 @@ out:
 
 static void __exit hyper_sysfs_exit(void)
 {
+	xen_pmu_destroy();
 	xen_properties_destroy();
 	xen_compilation_destroy();
 	xen_sysfs_uuid_destroy();
diff --git a/include/xen/interface/xen.h b/include/xen/interface/xen.h
index de08213..9ba0370 100644
--- a/include/xen/interface/xen.h
+++ b/include/xen/interface/xen.h
@@ -58,6 +58,7 @@
 #define __HYPERVISOR_physdev_op           33
 #define __HYPERVISOR_hvm_op               34
 #define __HYPERVISOR_tmem_op              38
+#define __HYPERVISOR_xenpmu_op            40
 
 /* Architecture-specific hypercall definitions. */
 #define __HYPERVISOR_arch_0               48
diff --git a/include/xen/interface/xenpmu.h b/include/xen/interface/xenpmu.h
new file mode 100644
index 0000000..be5ea9d
--- /dev/null
+++ b/include/xen/interface/xenpmu.h
@@ -0,0 +1,48 @@
+#ifndef __XEN_PUBLIC_XENPMU_H__
+#define __XEN_PUBLIC_XENPMU_H__
+
+#include "xen.h"
+
+#define XENPMU_VER_MAJ    0
+#define XENPMU_VER_MIN    1
+
+/* HYPERVISOR_xenpmu_op commands */
+#define XENPMU_mode_get        0
+#define XENPMU_mode_set        1
+#define XENPMU_feature_get     2
+#define XENPMU_feature_set     3
+
+/* Parameter structure for HYPERVISOR_xenpmu_op call */
+struct xen_pmu_params {
+	/* IN/OUT parameters */
+	struct {
+		uint32_t maj;
+		uint32_t min;
+	} version;
+	uint64_t val;
+
+	/* IN parameters */
+	uint64_t vcpu;
+};
+
+
+/* PMU modes:
+ * - XENPMU_MODE_OFF:   No PMU virtualization
+ * - XENPMU_MODE_SELF:  Guests can profile themselves, dom0 profiles
+ *                      itself and Xen
+ * - XENPMU_MODE_ALL:   Only dom0 has access to VPMU and it profiles
+ *                      everyone: itself, the hypervisor and the guests.
+ */
+#define XENPMU_MODE_OFF           0
+#define XENPMU_MODE_SELF          (1<<0)
+#define XENPMU_MODE_ALL           (1<<1)
+
+
+/*
+ * PMU features:
+ * - XENPMU_FEATURE_INTEL_BTS: Intel BTS support (ignored on AMD)
+ */
+#define XENPMU_FEATURE_INTEL_BTS  1
+
+
+#endif /* __XEN_PUBLIC_XENPMU_H__ */
-- 
1.8.1.4

[PATCH v4 3/6] xen/PMU: Initialization code for Xen PMU

[Boris Ostrovsky]

Map shared data structure that will hold CPU registers, VPMU context, V/PCPU IDs
of the CPU interrupted by PMU interrupt. Hypervisor fills this information in
its handler and passes it to the guest for further processing.

Set up PMU VIRQ.

Now that perf infrastructure will assume that PMU is availble on a PV guest we
need to be careful and make sure that accesses via RDPMC instruction don't
cause fatal traps by the hypervisor. Provide a nop RDPMC handler.

For the same reason avoid issuing a warning on a write to APIC's LVTPC.

Both of these will be made functional in later patches.


Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
---
 arch/x86/include/asm/xen/interface.h |  41 ++++++++++
 arch/x86/xen/Makefile                |   2 +-
 arch/x86/xen/enlighten.c             |  10 ++-
 arch/x86/xen/pmu.c                   | 145 +++++++++++++++++++++++++++++++++++
 arch/x86/xen/pmu.h                   |  11 +++
 arch/x86/xen/smp.c                   |  31 +++++++-
 include/xen/interface/xen.h          |   1 +
 include/xen/interface/xenpmu.h       |  17 ++++
 8 files changed, 255 insertions(+), 3 deletions(-)
 create mode 100644 arch/x86/xen/pmu.c
 create mode 100644 arch/x86/xen/pmu.h

diff --git a/arch/x86/include/asm/xen/interface.h b/arch/x86/include/asm/xen/interface.h
index 3400dba..fa64648 100644
--- a/arch/x86/include/asm/xen/interface.h
+++ b/arch/x86/include/asm/xen/interface.h
@@ -172,6 +172,47 @@ struct vcpu_guest_context {
 #endif
 };
 DEFINE_GUEST_HANDLE_STRUCT(vcpu_guest_context);
+
+/* AMD PMU registers and structures */
+struct xen_pmu_amd_ctxt {
+	uint32_t counters;       /* Offset to counter MSRs */
+	uint32_t ctrls;          /* Offset to control MSRs */
+};
+
+/* Intel PMU registers and structures */
+struct xen_pmu_cntr_pair {
+	uint64_t counter;
+	uint64_t control;
+};
+
+struct xen_pmu_intel_ctxt {
+	uint64_t global_ctrl;
+	uint64_t global_ovf_ctrl;
+	uint64_t global_status;
+	uint64_t fixed_ctrl;
+	uint64_t ds_area;
+	uint64_t pebs_enable;
+	uint64_t debugctl;
+	uint32_t fixed_counters;  /* Offset to fixed counter MSRs */
+	uint32_t arch_counters;   /* Offset to architectural counter MSRs */
+};
+
+struct xen_arch_pmu {
+	union {
+		struct cpu_user_regs regs;
+		uint8_t pad1[256];
+	};
+	union {
+		uint32_t lapic_lvtpc;
+		uint64_t pad2;
+	};
+	union {
+		struct xen_pmu_amd_ctxt amd;
+		struct xen_pmu_intel_ctxt intel;
+#define XENPMU_CTXT_PAD_SZ  128
+		uint8_t pad3[XENPMU_CTXT_PAD_SZ];
+	};
+};
 #endif	/* !__ASSEMBLY__ */
 
 /*
diff --git a/arch/x86/xen/Makefile b/arch/x86/xen/Makefile
index 96ab2c0..b187df5 100644
--- a/arch/x86/xen/Makefile
+++ b/arch/x86/xen/Makefile
@@ -13,7 +13,7 @@ CFLAGS_mmu.o			:= $(nostackp)
 obj-y		:= enlighten.o setup.o multicalls.o mmu.o irq.o \
 			time.o xen-asm.o xen-asm_$(BITS).o \
 			grant-table.o suspend.o platform-pci-unplug.o \
-			p2m.o
+			p2m.o pmu.o
 
 obj-$(CONFIG_EVENT_TRACING) += trace.o
 
diff --git a/arch/x86/xen/enlighten.c b/arch/x86/xen/enlighten.c
index ffb101e..57764ce 100644
--- a/arch/x86/xen/enlighten.c
+++ b/arch/x86/xen/enlighten.c
@@ -962,8 +962,16 @@ static u32 xen_apic_read(u32 reg)
 	return op.u.pcpu_info.apic_id << 24;
 }
 
+unsigned long long xen_read_pmc(int counter)
+{
+	return 0;
+}
+
 static void xen_apic_write(u32 reg, u32 val)
 {
+	if (reg == APIC_LVTPC)
+		return;
+
 	/* Warn to see if there's any stray references */
 	WARN_ON(1);
 }
@@ -1248,7 +1256,7 @@ static const struct pv_cpu_ops xen_cpu_ops __initconst = {
 	.write_msr = xen_write_msr_safe,
 
 	.read_tsc = native_read_tsc,
-	.read_pmc = native_read_pmc,
+	.read_pmc = xen_read_pmc,
 
 	.read_tscp = native_read_tscp,
 
diff --git a/arch/x86/xen/pmu.c b/arch/x86/xen/pmu.c
new file mode 100644
index 0000000..799e229
--- /dev/null
+++ b/arch/x86/xen/pmu.c
@@ -0,0 +1,145 @@
+#include <linux/types.h>
+#include <linux/interrupt.h>
+
+#include <asm/xen/hypercall.h>
+#include <xen/page.h>
+#include <xen/interface/xen.h>
+#include <xen/interface/vcpu.h>
+#include <xen/interface/xenpmu.h>
+
+#include "xen-ops.h"
+#include "pmu.h"
+
+/* x86_pmu.handle_irq definition */
+#include "../kernel/cpu/perf_event.h"
+
+
+/* Shared page between hypervisor and domain */
+DEFINE_PER_CPU(struct xen_pmu_data *, xenpmu_shared);
+#define get_xenpmu_data()    per_cpu(xenpmu_shared, smp_processor_id())
+
+/* perf callbacks*/
+int xen_is_in_guest(void)
+{
+	const struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+
+	if (!xenpmu_data) {
+		WARN_ONCE(1, "%s: pmudata not initialized\n", __func__);
+		return 0;
+	}
+
+	if (!xen_initial_domain() || (xenpmu_data->domain_id >= DOMID_SELF))
+		return 0;
+
+	return 1;
+}
+
+static int xen_is_user_mode(void)
+{
+	const struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+
+	if (!xenpmu_data) {
+		WARN_ONCE(1, "%s: pmudata not initialized\n", __func__);
+		return 0;
+	}
+
+	return ((xenpmu_data->pmu.regs.cs & 3) == 3);
+}
+
+static unsigned long xen_get_guest_ip(void)
+{
+	const struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+
+	if (!xenpmu_data) {
+		WARN_ONCE(1, "%s: pmudata not initialized\n", __func__);
+		return 0;
+	}
+
+	return xenpmu_data->pmu.regs.eip;
+}
+
+static struct perf_guest_info_callbacks xen_guest_cbs = {
+	.is_in_guest            = xen_is_in_guest,
+	.is_user_mode           = xen_is_user_mode,
+	.get_guest_ip           = xen_get_guest_ip,
+};
+
+/* Convert registers from Xen's format to Linux' */
+static void xen_convert_regs(const struct cpu_user_regs *xen_regs,
+			     struct pt_regs *regs)
+{
+	regs->ip = xen_regs->eip;
+	regs->cs = xen_regs->cs;
+}
+
+irqreturn_t xen_pmu_irq_handler(int irq, void *dev_id)
+{
+	int ret = IRQ_NONE;
+	struct pt_regs regs;
+	const struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+
+	if (!xenpmu_data) {
+		WARN_ONCE(1, "%s: pmudata not initialized\n", __func__);
+		return ret;
+	}
+
+	xen_convert_regs(&xenpmu_data->pmu.regs, &regs);
+	if (x86_pmu.handle_irq(&regs))
+		ret = IRQ_HANDLED;
+
+	return ret;
+}
+
+bool is_xen_pmu(int cpu)
+{
+	return (per_cpu(xenpmu_shared, cpu) != NULL);
+}
+
+int xen_pmu_init(int cpu)
+{
+	int ret = 0;
+	struct xen_pmu_params xp;
+	unsigned long pfn;
+	struct xen_pmu_data *xenpmu_data;
+
+	BUILD_BUG_ON(sizeof(struct xen_pmu_data) > PAGE_SIZE);
+	xenpmu_data = (struct xen_pmu_data *)get_zeroed_page(GFP_KERNEL);
+	if (!xenpmu_data) {
+		ret = -ENOMEM;
+		goto fail;
+	}
+	pfn = virt_to_pfn(xenpmu_data);
+
+	xp.val = pfn_to_mfn(pfn);
+	xp.vcpu = cpu;
+	xp.version.maj = XENPMU_VER_MAJ;
+	xp.version.min = XENPMU_VER_MIN;
+	ret = HYPERVISOR_xenpmu_op(XENPMU_init, &xp);
+	if (ret)
+		goto fail;
+
+	per_cpu(xenpmu_shared, cpu) = xenpmu_data;
+
+	if (cpu == 0)
+		perf_register_guest_info_callbacks(&xen_guest_cbs);
+
+	return ret;
+
+fail:
+	free_pages((unsigned long)xenpmu_data, 0);
+	return ret;
+}
+
+void xen_pmu_finish(int cpu)
+{
+	struct xen_pmu_params xp;
+
+	xp.vcpu = cpu;
+	xp.version.maj = XENPMU_VER_MAJ;
+	xp.version.min = XENPMU_VER_MIN;
+
+	(void)HYPERVISOR_xenpmu_op(XENPMU_finish, &xp);
+
+	free_pages((unsigned long)per_cpu(xenpmu_shared, cpu), 0);
+	per_cpu(xenpmu_shared, cpu) = NULL;
+}
diff --git a/arch/x86/xen/pmu.h b/arch/x86/xen/pmu.h
new file mode 100644
index 0000000..d52e8db
--- /dev/null
+++ b/arch/x86/xen/pmu.h
@@ -0,0 +1,11 @@
+#ifndef __XEN_PMU_H
+#define __XEN_PMU_H
+
+#include <xen/interface/xenpmu.h>
+
+irqreturn_t xen_pmu_irq_handler(int irq, void *dev_id);
+int xen_pmu_init(int cpu);
+void xen_pmu_finish(int cpu);
+bool is_xen_pmu(int cpu);
+
+#endif /* __XEN_PMU_H */
diff --git a/arch/x86/xen/smp.c b/arch/x86/xen/smp.c
index 7005974..7ea6296 100644
--- a/arch/x86/xen/smp.c
+++ b/arch/x86/xen/smp.c
@@ -26,6 +26,7 @@
 
 #include <xen/interface/xen.h>
 #include <xen/interface/vcpu.h>
+#include <xen/interface/xenpmu.h>
 
 #include <asm/xen/interface.h>
 #include <asm/xen/hypercall.h>
@@ -37,6 +38,7 @@
 #include <xen/hvc-console.h>
 #include "xen-ops.h"
 #include "mmu.h"
+#include "pmu.h"
 
 cpumask_var_t xen_cpu_initialized_map;
 
@@ -49,6 +51,7 @@ static DEFINE_PER_CPU(struct xen_common_irq, xen_callfunc_irq) = { .irq = -1 };
 static DEFINE_PER_CPU(struct xen_common_irq, xen_callfuncsingle_irq) = { .irq = -1 };
 static DEFINE_PER_CPU(struct xen_common_irq, xen_irq_work) = { .irq = -1 };
 static DEFINE_PER_CPU(struct xen_common_irq, xen_debug_irq) = { .irq = -1 };
+static DEFINE_PER_CPU(struct xen_common_irq, xen_pmu_irq) = { .irq = -1 };
 
 static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id);
 static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id);
@@ -147,11 +150,18 @@ static void xen_smp_intr_free(unsigned int cpu)
 		kfree(per_cpu(xen_irq_work, cpu).name);
 		per_cpu(xen_irq_work, cpu).name = NULL;
 	}
+
+	if (per_cpu(xen_pmu_irq, cpu).irq >= 0) {
+		unbind_from_irqhandler(per_cpu(xen_pmu_irq, cpu).irq, NULL);
+		per_cpu(xen_pmu_irq, cpu).irq = -1;
+		kfree(per_cpu(xen_pmu_irq, cpu).name);
+		per_cpu(xen_pmu_irq, cpu).name = NULL;
+	}
 };
 static int xen_smp_intr_init(unsigned int cpu)
 {
 	int rc;
-	char *resched_name, *callfunc_name, *debug_name;
+	char *resched_name, *callfunc_name, *debug_name, *pmu_name;
 
 	resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu);
 	rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR,
@@ -217,6 +227,18 @@ static int xen_smp_intr_init(unsigned int cpu)
 	per_cpu(xen_irq_work, cpu).irq = rc;
 	per_cpu(xen_irq_work, cpu).name = callfunc_name;
 
+	if (is_xen_pmu(cpu)) {
+		pmu_name = kasprintf(GFP_KERNEL, "pmu%d", cpu);
+		rc = bind_virq_to_irqhandler(VIRQ_XENPMU, cpu,
+					     xen_pmu_irq_handler,
+					     IRQF_PERCPU|IRQF_NOBALANCING,
+					     pmu_name, NULL);
+		if (rc < 0)
+			goto fail;
+		per_cpu(xen_pmu_irq, cpu).irq = rc;
+		per_cpu(xen_pmu_irq, cpu).name = pmu_name;
+	}
+
 	return 0;
 
  fail:
@@ -334,6 +356,9 @@ static void __init xen_smp_prepare_cpus(unsigned int max_cpus)
 	}
 	set_cpu_sibling_map(0);
 
+	if (xen_pmu_init(0))
+		pr_err("Could not initialize VPMU for VCPU 0\n");
+
 	if (xen_smp_intr_init(0))
 		BUG();
 
@@ -463,6 +488,9 @@ static int xen_cpu_up(unsigned int cpu, struct task_struct *idle)
 		/* Just in case we booted with a single CPU. */
 		alternatives_enable_smp();
 
+	if (xen_pmu_init(cpu))
+		pr_err("Could not initialize VPMU for VCPU %u\n", cpu);
+
 	rc = xen_smp_intr_init(cpu);
 	if (rc)
 		return rc;
@@ -504,6 +532,7 @@ static void xen_cpu_die(unsigned int cpu)
 	xen_smp_intr_free(cpu);
 	xen_uninit_lock_cpu(cpu);
 	xen_teardown_timer(cpu);
+	xen_pmu_finish(cpu);
 }
 
 static void xen_play_dead(void) /* used only with HOTPLUG_CPU */
diff --git a/include/xen/interface/xen.h b/include/xen/interface/xen.h
index 9ba0370..25df701 100644
--- a/include/xen/interface/xen.h
+++ b/include/xen/interface/xen.h
@@ -81,6 +81,7 @@
 #define VIRQ_DOM_EXC    3  /* (DOM0) Exceptional event for some domain.   */
 #define VIRQ_DEBUGGER   6  /* (DOM0) A domain has paused for debugging.   */
 #define VIRQ_PCPU_STATE 9  /* (DOM0) PCPU state changed                   */
+#define VIRQ_XENPMU    13  /* PMC interrupt                               */
 
 /* Architecture-specific VIRQ definitions. */
 #define VIRQ_ARCH_0    16
diff --git a/include/xen/interface/xenpmu.h b/include/xen/interface/xenpmu.h
index be5ea9d..78073dc 100644
--- a/include/xen/interface/xenpmu.h
+++ b/include/xen/interface/xenpmu.h
@@ -11,6 +11,8 @@
 #define XENPMU_mode_set        1
 #define XENPMU_feature_get     2
 #define XENPMU_feature_set     3
+#define XENPMU_init            4
+#define XENPMU_finish          5
 
 /* Parameter structure for HYPERVISOR_xenpmu_op call */
 struct xen_pmu_params {
@@ -44,5 +46,20 @@ struct xen_pmu_params {
  */
 #define XENPMU_FEATURE_INTEL_BTS  1
 
+/*
+ * PMU MSRs are cached in the context so the PV guest doesn't need to trap to
+ * the hypervisor
+ */
+#define PMU_CACHED 1
+
+/* Shared between hypervisor and PV domain */
+struct xen_pmu_data {
+	uint32_t domain_id;
+	uint32_t vcpu_id;
+	uint32_t pcpu_id;
+	uint32_t pmu_flags;
+
+	struct xen_arch_pmu pmu;
+};
 
 #endif /* __XEN_PUBLIC_XENPMU_H__ */
-- 
1.8.1.4

[PATCH v4 4/6] xen/PMU: Describe vendor-specific PMU registers

[Boris Ostrovsky]

AMD and Intel PMU register initialization and helpers that determine whether a
register belongs to PMU.

This and some of subsequent PMU emulation code is somewhat similar to Xen's PMU
implementation.

Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Reviewed-by: David Vrabel <david.vrabel@citrix.com>
---
 arch/x86/xen/pmu.c | 153 ++++++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 152 insertions(+), 1 deletion(-)

diff --git a/arch/x86/xen/pmu.c b/arch/x86/xen/pmu.c
index 799e229..f893631 100644
--- a/arch/x86/xen/pmu.c
+++ b/arch/x86/xen/pmu.c
@@ -18,6 +18,155 @@
 DEFINE_PER_CPU(struct xen_pmu_data *, xenpmu_shared);
 #define get_xenpmu_data()    per_cpu(xenpmu_shared, smp_processor_id())
 
+
+/* AMD PMU */
+#define F15H_NUM_COUNTERS   6
+#define F10H_NUM_COUNTERS   4
+
+static __read_mostly uint32_t amd_counters_base;
+static __read_mostly uint32_t amd_ctrls_base;
+static __read_mostly int amd_msr_step;
+static __read_mostly int k7_counters_mirrored;
+static __read_mostly int amd_num_counters;
+
+/* Intel PMU */
+#define MSR_TYPE_COUNTER            0
+#define MSR_TYPE_CTRL               1
+#define MSR_TYPE_GLOBAL             2
+#define MSR_TYPE_ARCH_COUNTER       3
+#define MSR_TYPE_ARCH_CTRL          4
+
+/* Number of general pmu registers (CPUID.EAX[0xa].EAX[8..15]) */
+#define PMU_GENERAL_NR_SHIFT        8
+#define PMU_GENERAL_NR_BITS         8
+#define PMU_GENERAL_NR_MASK         (((1 << PMU_GENERAL_NR_BITS) - 1) \
+				     << PMU_GENERAL_NR_SHIFT)
+
+/* Number of fixed pmu registers (CPUID.EDX[0xa].EDX[0..4]) */
+#define PMU_FIXED_NR_SHIFT          0
+#define PMU_FIXED_NR_BITS           5
+#define PMU_FIXED_NR_MASK           (((1 << PMU_FIXED_NR_BITS) - 1) \
+				     << PMU_FIXED_NR_SHIFT)
+
+/* Alias registers (0x4c1) for full-width writes to PMCs */
+#define MSR_PMC_ALIAS_MASK          (~(MSR_IA32_PERFCTR0 ^ MSR_IA32_PMC0))
+
+static __read_mostly int intel_num_arch_counters, intel_num_fixed_counters;
+
+
+static void xen_pmu_arch_init(void)
+{
+	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
+
+		switch (boot_cpu_data.x86) {
+		case 0x15:
+			amd_num_counters = F15H_NUM_COUNTERS;
+			amd_counters_base = MSR_F15H_PERF_CTR;
+			amd_ctrls_base = MSR_F15H_PERF_CTL;
+			amd_msr_step = 2;
+			k7_counters_mirrored = 1;
+			break;
+		case 0x10:
+		case 0x12:
+		case 0x14:
+		case 0x16:
+		default:
+			amd_num_counters = F10H_NUM_COUNTERS;
+			amd_counters_base = MSR_K7_PERFCTR0;
+			amd_ctrls_base = MSR_K7_EVNTSEL0;
+			amd_msr_step = 1;
+			k7_counters_mirrored = 0;
+			break;
+		}
+	} else {
+		uint32_t eax, ebx, ecx, edx;
+
+		cpuid(0xa, &eax, &ebx, &ecx, &edx);
+
+		intel_num_arch_counters = (eax & PMU_GENERAL_NR_MASK) >>
+			PMU_GENERAL_NR_SHIFT;
+		intel_num_fixed_counters = (edx & PMU_FIXED_NR_MASK) >>
+			PMU_FIXED_NR_SHIFT;
+	}
+}
+
+static inline uint32_t get_fam15h_addr(u32 addr)
+{
+	switch (addr) {
+	case MSR_K7_PERFCTR0:
+	case MSR_K7_PERFCTR1:
+	case MSR_K7_PERFCTR2:
+	case MSR_K7_PERFCTR3:
+		return MSR_F15H_PERF_CTR + (addr - MSR_K7_PERFCTR0);
+	case MSR_K7_EVNTSEL0:
+	case MSR_K7_EVNTSEL1:
+	case MSR_K7_EVNTSEL2:
+	case MSR_K7_EVNTSEL3:
+		return MSR_F15H_PERF_CTL + (addr - MSR_K7_EVNTSEL0);
+	default:
+		break;
+	}
+
+	return addr;
+}
+
+static inline bool is_amd_pmu_msr(unsigned int msr)
+{
+	if ((msr >= MSR_F15H_PERF_CTL &&
+	     msr < MSR_F15H_PERF_CTR + (amd_num_counters * 2)) ||
+	    (msr >= MSR_K7_EVNTSEL0 &&
+	     msr < MSR_K7_PERFCTR0 + amd_num_counters))
+		return true;
+
+	return false;
+}
+
+static int is_intel_pmu_msr(u32 msr_index, int *type, int *index)
+{
+	u32 msr_index_pmc;
+
+	switch (msr_index) {
+	case MSR_CORE_PERF_FIXED_CTR_CTRL:
+	case MSR_IA32_DS_AREA:
+	case MSR_IA32_PEBS_ENABLE:
+		*type = MSR_TYPE_CTRL;
+		return true;
+
+	case MSR_CORE_PERF_GLOBAL_CTRL:
+	case MSR_CORE_PERF_GLOBAL_STATUS:
+	case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
+		*type = MSR_TYPE_GLOBAL;
+		return true;
+
+	default:
+
+		if ((msr_index >= MSR_CORE_PERF_FIXED_CTR0) &&
+		    (msr_index < MSR_CORE_PERF_FIXED_CTR0 +
+				 intel_num_fixed_counters)) {
+			*index = msr_index - MSR_CORE_PERF_FIXED_CTR0;
+			*type = MSR_TYPE_COUNTER;
+			return true;
+		}
+
+		if ((msr_index >= MSR_P6_EVNTSEL0) &&
+		    (msr_index < MSR_P6_EVNTSEL0 +  intel_num_arch_counters)) {
+			*index = msr_index - MSR_P6_EVNTSEL0;
+			*type = MSR_TYPE_ARCH_CTRL;
+			return true;
+		}
+
+		msr_index_pmc = msr_index & MSR_PMC_ALIAS_MASK;
+		if ((msr_index_pmc >= MSR_IA32_PERFCTR0) &&
+		    (msr_index_pmc < MSR_IA32_PERFCTR0 +
+				     intel_num_arch_counters)) {
+			*type = MSR_TYPE_ARCH_COUNTER;
+			*index = msr_index_pmc - MSR_IA32_PERFCTR0;
+			return true;
+		}
+		return false;
+	}
+}
+
 /* perf callbacks*/
 int xen_is_in_guest(void)
 {
@@ -120,8 +269,10 @@ int xen_pmu_init(int cpu)
 
 	per_cpu(xenpmu_shared, cpu) = xenpmu_data;
 
-	if (cpu == 0)
+	if (cpu == 0) {
 		perf_register_guest_info_callbacks(&xen_guest_cbs);
+		xen_pmu_arch_init();
+	}
 
 	return ret;
 
-- 
1.8.1.4

[PATCH v4 5/6] xen/PMU: Intercept PMU-related MSR and APIC accesses

[Boris Ostrovsky]

Provide interfaces for recognizing accesses to PMU-related MSRs and LVTPC APIC
and process these accesses in Xen PMU code.

(The interrupt handler performs XENPMU_flush right away in the beginning since
no PMU emulation is available. It will be added with a later patch).

Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Reviewed-by: David Vrabel <david.vrabel@citrix.com>
---
 arch/x86/xen/enlighten.c       | 26 ++++++++----
 arch/x86/xen/pmu.c             | 95 +++++++++++++++++++++++++++++++++++++++++-
 arch/x86/xen/pmu.h             |  4 ++
 include/xen/interface/xenpmu.h |  2 +
 4 files changed, 118 insertions(+), 9 deletions(-)

diff --git a/arch/x86/xen/enlighten.c b/arch/x86/xen/enlighten.c
index 57764ce..9ab5f76 100644
--- a/arch/x86/xen/enlighten.c
+++ b/arch/x86/xen/enlighten.c
@@ -82,6 +82,7 @@
 #include "mmu.h"
 #include "smp.h"
 #include "multicalls.h"
+#include "pmu.h"
 
 EXPORT_SYMBOL_GPL(hypercall_page);
 
@@ -962,15 +963,12 @@ static u32 xen_apic_read(u32 reg)
 	return op.u.pcpu_info.apic_id << 24;
 }
 
-unsigned long long xen_read_pmc(int counter)
-{
-	return 0;
-}
-
 static void xen_apic_write(u32 reg, u32 val)
 {
-	if (reg == APIC_LVTPC)
+	if (reg == APIC_LVTPC) {
+		(void)pmu_apic_update(reg);
 		return;
+	}
 
 	/* Warn to see if there's any stray references */
 	WARN_ON(1);
@@ -1076,6 +1074,17 @@ static inline void xen_write_cr8(unsigned long val)
 	BUG_ON(val);
 }
 #endif
+
+static u64 xen_read_msr_safe(unsigned int msr, int *err)
+{
+	u64 val;
+
+	if (pmu_msr_read(msr, &val, err))
+		return val;
+
+	return native_read_msr_safe(msr, err);
+}
+
 static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
 {
 	int ret;
@@ -1116,7 +1125,8 @@ static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
 		break;
 
 	default:
-		ret = native_write_msr_safe(msr, low, high);
+		if (!pmu_msr_write(msr, low, high, &ret))
+			ret = native_write_msr_safe(msr, low, high);
 	}
 
 	return ret;
@@ -1252,7 +1262,7 @@ static const struct pv_cpu_ops xen_cpu_ops __initconst = {
 
 	.wbinvd = native_wbinvd,
 
-	.read_msr = native_read_msr_safe,
+	.read_msr = xen_read_msr_safe,
 	.write_msr = xen_write_msr_safe,
 
 	.read_tsc = native_read_tsc,
diff --git a/arch/x86/xen/pmu.c b/arch/x86/xen/pmu.c
index f893631..959172e 100644
--- a/arch/x86/xen/pmu.c
+++ b/arch/x86/xen/pmu.c
@@ -51,6 +51,8 @@ static __read_mostly int amd_num_counters;
 /* Alias registers (0x4c1) for full-width writes to PMCs */
 #define MSR_PMC_ALIAS_MASK          (~(MSR_IA32_PERFCTR0 ^ MSR_IA32_PMC0))
 
+#define INTEL_PMC_TYPE_SHIFT        30
+
 static __read_mostly int intel_num_arch_counters, intel_num_fixed_counters;
 
 
@@ -167,6 +169,91 @@ static int is_intel_pmu_msr(u32 msr_index, int *type, int *index)
 	}
 }
 
+bool pmu_msr_read(unsigned int msr, uint64_t *val, int *err)
+{
+
+	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
+		if (is_amd_pmu_msr(msr)) {
+			*val = native_read_msr_safe(msr, err);
+			return true;
+		}
+	} else {
+		int type, index;
+
+		if (is_intel_pmu_msr(msr, &type, &index)) {
+			*val = native_read_msr_safe(msr, err);
+			return true;
+		}
+	}
+
+	return false;
+}
+
+bool pmu_msr_write(unsigned int msr, uint32_t low, uint32_t high, int *err)
+{
+	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
+		if (is_amd_pmu_msr(msr)) {
+			*err = native_write_msr_safe(msr, low, high);
+			return true;
+		}
+	} else {
+		int type, index;
+
+		if (is_intel_pmu_msr(msr, &type, &index)) {
+			*err = native_write_msr_safe(msr, low, high);
+			return true;
+		}
+	}
+
+	return false;
+}
+
+unsigned long long xen_amd_read_pmc(int counter)
+{
+	uint32_t msr;
+	int err;
+
+	msr = amd_counters_base + (counter * amd_msr_step);
+	return native_read_msr_safe(msr, &err);
+}
+
+unsigned long long xen_intel_read_pmc(int counter)
+{
+	int err;
+	uint32_t msr;
+
+	if (counter & (1<<INTEL_PMC_TYPE_SHIFT))
+		msr = MSR_CORE_PERF_FIXED_CTR0 + (counter & 0xffff);
+	else
+		msr = MSR_IA32_PERFCTR0 + counter;
+
+	return native_read_msr_safe(msr, &err);
+}
+
+unsigned long long xen_read_pmc(int counter)
+{
+	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
+		return xen_amd_read_pmc(counter);
+	else
+		return xen_intel_read_pmc(counter);
+}
+
+int pmu_apic_update(uint32_t val)
+{
+	int ret;
+	struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+
+	if (!xenpmu_data) {
+		WARN_ONCE(1, "%s: pmudata not initialized\n", __func__);
+		return -EINVAL;
+	}
+
+	xenpmu_data->pmu.lapic_lvtpc = val;
+	ret = HYPERVISOR_xenpmu_op(XENPMU_lvtpc_set, NULL);
+
+	return ret;
+}
+
 /* perf callbacks*/
 int xen_is_in_guest(void)
 {
@@ -223,7 +310,7 @@ static void xen_convert_regs(const struct cpu_user_regs *xen_regs,
 
 irqreturn_t xen_pmu_irq_handler(int irq, void *dev_id)
 {
-	int ret = IRQ_NONE;
+	int err, ret = IRQ_NONE;
 	struct pt_regs regs;
 	const struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
 
@@ -232,6 +319,12 @@ irqreturn_t xen_pmu_irq_handler(int irq, void *dev_id)
 		return ret;
 	}
 
+	err = HYPERVISOR_xenpmu_op(XENPMU_flush, NULL);
+	if (err) {
+		WARN_ONCE(1, "%s failed hypercall, err: %d\n", __func__, err);
+		return ret;
+	}
+
 	xen_convert_regs(&xenpmu_data->pmu.regs, &regs);
 	if (x86_pmu.handle_irq(&regs))
 		ret = IRQ_HANDLED;
diff --git a/arch/x86/xen/pmu.h b/arch/x86/xen/pmu.h
index d52e8db..30bfbcf 100644
--- a/arch/x86/xen/pmu.h
+++ b/arch/x86/xen/pmu.h
@@ -7,5 +7,9 @@ irqreturn_t xen_pmu_irq_handler(int irq, void *dev_id);
 int xen_pmu_init(int cpu);
 void xen_pmu_finish(int cpu);
 bool is_xen_pmu(int cpu);
+bool pmu_msr_read(unsigned int msr, uint64_t *val, int *err);
+bool pmu_msr_write(unsigned int msr, uint32_t low, uint32_t high, int *err);
+int pmu_apic_update(uint32_t reg);
+unsigned long long xen_read_pmc(int counter);
 
 #endif /* __XEN_PMU_H */
diff --git a/include/xen/interface/xenpmu.h b/include/xen/interface/xenpmu.h
index 78073dc..59aaedd 100644
--- a/include/xen/interface/xenpmu.h
+++ b/include/xen/interface/xenpmu.h
@@ -13,6 +13,8 @@
 #define XENPMU_feature_set     3
 #define XENPMU_init            4
 #define XENPMU_finish          5
+#define XENPMU_lvtpc_set       6
+#define XENPMU_flush           7
 
 /* Parameter structure for HYPERVISOR_xenpmu_op call */
 struct xen_pmu_params {
-- 
1.8.1.4

[PATCH v4 6/6] xen/PMU: PMU emulation code

[Boris Ostrovsky]

Add PMU emulation code that runs when we are processing a PMU interrupt. This code
will allow us not to trap to hypervisor on each MSR/LVTPC access (of which there
may be quite a few in the handler).

Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
---
 arch/x86/xen/pmu.c | 213 ++++++++++++++++++++++++++++++++++++++++++++++-------
 1 file changed, 185 insertions(+), 28 deletions(-)

diff --git a/arch/x86/xen/pmu.c b/arch/x86/xen/pmu.c
index 959172e..9acec38 100644
--- a/arch/x86/xen/pmu.c
+++ b/arch/x86/xen/pmu.c
@@ -13,11 +13,20 @@
 /* x86_pmu.handle_irq definition */
 #include "../kernel/cpu/perf_event.h"
 
+#define XENPMU_IRQ_PROCESSING    1
+struct xenpmu {
+	/* Shared page between hypervisor and domain */
+	struct xen_pmu_data *xenpmu_data;
 
-/* Shared page between hypervisor and domain */
-DEFINE_PER_CPU(struct xen_pmu_data *, xenpmu_shared);
-#define get_xenpmu_data()    per_cpu(xenpmu_shared, smp_processor_id())
+	uint8_t flags;
+};
+DEFINE_PER_CPU(struct xenpmu, xenpmu_shared);
+#define get_xenpmu_data()    (this_cpu_ptr(&xenpmu_shared)->xenpmu_data)
+#define get_xenpmu_flags()   (this_cpu_ptr(&xenpmu_shared)->flags)
 
+/* Macro for computing address of a PMU MSR bank */
+#define field_offset(ctxt, field) ((void *)((uintptr_t)ctxt + \
+					    (uintptr_t)ctxt->field))
 
 /* AMD PMU */
 #define F15H_NUM_COUNTERS   6
@@ -169,19 +178,124 @@ static int is_intel_pmu_msr(u32 msr_index, int *type, int *index)
 	}
 }
 
-bool pmu_msr_read(unsigned int msr, uint64_t *val, int *err)
+static bool xen_intel_pmu_emulate(unsigned int msr, u64 *val, int type,
+				  int index, bool is_read)
+{
+	uint64_t *reg = NULL;
+	struct xen_pmu_intel_ctxt *ctxt;
+	uint64_t *fix_counters;
+	struct xen_pmu_cntr_pair *arch_cntr_pair;
+	struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+	uint8_t xenpmu_flags = get_xenpmu_flags();
+
+
+	if (!xenpmu_data || !(xenpmu_flags & XENPMU_IRQ_PROCESSING))
+		return false;
+
+	ctxt = &xenpmu_data->pmu.intel;
+
+	switch (msr) {
+	case MSR_CORE_PERF_GLOBAL_OVF_CTRL:
+		reg = &ctxt->global_ovf_ctrl;
+		break;
+	case MSR_CORE_PERF_GLOBAL_STATUS:
+		reg = &ctxt->global_status;
+		break;
+	case MSR_CORE_PERF_GLOBAL_CTRL:
+		reg = &ctxt->global_ctrl;
+		break;
+	case MSR_CORE_PERF_FIXED_CTR_CTRL:
+		reg = &ctxt->fixed_ctrl;
+		break;
+	default:
+		switch (type) {
+		case MSR_TYPE_COUNTER:
+			fix_counters = field_offset(ctxt, fixed_counters);
+			reg = &fix_counters[index];
+			break;
+		case MSR_TYPE_ARCH_COUNTER:
+			arch_cntr_pair = field_offset(ctxt, arch_counters);
+			reg = &arch_cntr_pair[index].counter;
+			break;
+		case MSR_TYPE_ARCH_CTRL:
+			arch_cntr_pair = field_offset(ctxt, arch_counters);
+			reg = &arch_cntr_pair[index].control;
+			break;
+		default:
+			return false;
+		}
+	}
+
+	if (reg) {
+		if (is_read)
+			*val = *reg;
+		else {
+			*reg = *val;
+
+			if (msr == MSR_CORE_PERF_GLOBAL_OVF_CTRL)
+				ctxt->global_status &= (~(*val));
+		}
+		return true;
+	}
+
+	return false;
+}
+
+static bool xen_amd_pmu_emulate(unsigned int msr, u64 *val, bool is_read)
 {
+	uint64_t *reg = NULL;
+	int i, off = 0;
+	struct xen_pmu_amd_ctxt *ctxt;
+	uint64_t *counter_regs, *ctrl_regs;
+	struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+	uint8_t xenpmu_flags = get_xenpmu_flags();
+
+	if (!xenpmu_data || !(xenpmu_flags & XENPMU_IRQ_PROCESSING))
+		return false;
+
+	if (k7_counters_mirrored &&
+	    ((msr >= MSR_K7_EVNTSEL0) && (msr <= MSR_K7_PERFCTR3)))
+		msr = get_fam15h_addr(msr);
 
+	ctxt = &xenpmu_data->pmu.amd;
+	for (i = 0; i < amd_num_counters; i++) {
+		if (msr == amd_ctrls_base + off) {
+			ctrl_regs = field_offset(ctxt, ctrls);
+			reg = &ctrl_regs[i];
+			break;
+		} else if (msr == amd_counters_base + off) {
+			counter_regs = field_offset(ctxt, counters);
+			reg = &counter_regs[i];
+			break;
+		}
+		off += amd_msr_step;
+	}
+
+	if (reg) {
+		if (is_read)
+			*val = *reg;
+		else
+			*reg = *val;
+
+		return true;
+	}
+	return false;
+}
+
+bool pmu_msr_read(unsigned int msr, uint64_t *val, int *err)
+{
 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
 		if (is_amd_pmu_msr(msr)) {
-			*val = native_read_msr_safe(msr, err);
+			if (!xen_amd_pmu_emulate(msr, val, 1))
+				*val = native_read_msr_safe(msr, err);
 			return true;
 		}
 	} else {
 		int type, index;
 
 		if (is_intel_pmu_msr(msr, &type, &index)) {
-			*val = native_read_msr_safe(msr, err);
+			if (!xen_intel_pmu_emulate(msr, val, type, index, 1))
+				*val = native_read_msr_safe(msr, err);
 			return true;
 		}
 	}
@@ -191,16 +305,20 @@ bool pmu_msr_read(unsigned int msr, uint64_t *val, int *err)
 
 bool pmu_msr_write(unsigned int msr, uint32_t low, uint32_t high, int *err)
 {
+	uint64_t val = ((uint64_t)high << 32) | low;
+
 	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
 		if (is_amd_pmu_msr(msr)) {
-			*err = native_write_msr_safe(msr, low, high);
+			if (!xen_amd_pmu_emulate(msr, &val, 0))
+				*err = native_write_msr_safe(msr, low, high);
 			return true;
 		}
 	} else {
 		int type, index;
 
 		if (is_intel_pmu_msr(msr, &type, &index)) {
-			*err = native_write_msr_safe(msr, low, high);
+			if (!xen_intel_pmu_emulate(msr, &val, type, index, 0))
+				*err = native_write_msr_safe(msr, low, high);
 			return true;
 		}
 	}
@@ -210,24 +328,52 @@ bool pmu_msr_write(unsigned int msr, uint32_t low, uint32_t high, int *err)
 
 unsigned long long xen_amd_read_pmc(int counter)
 {
-	uint32_t msr;
-	int err;
+	struct xen_pmu_amd_ctxt *ctxt;
+	uint64_t *counter_regs;
+	struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+	uint8_t xenpmu_flags = get_xenpmu_flags();
 
-	msr = amd_counters_base + (counter * amd_msr_step);
-	return native_read_msr_safe(msr, &err);
+	if (!xenpmu_data || !(xenpmu_flags & XENPMU_IRQ_PROCESSING)) {
+		uint32_t msr;
+		int err;
+
+		msr = amd_counters_base + (counter * amd_msr_step);
+		return native_read_msr_safe(msr, &err);
+	}
+
+	ctxt = &xenpmu_data->pmu.amd;
+	counter_regs = field_offset(ctxt, counters);
+	return counter_regs[counter];
 }
 
 unsigned long long xen_intel_read_pmc(int counter)
 {
-	int err;
-	uint32_t msr;
+	struct xen_pmu_intel_ctxt *ctxt;
+	uint64_t *fixed_counters;
+	struct xen_pmu_cntr_pair *arch_cntr_pair;
+	struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+	uint8_t xenpmu_flags = get_xenpmu_flags();
 
-	if (counter & (1<<INTEL_PMC_TYPE_SHIFT))
-		msr = MSR_CORE_PERF_FIXED_CTR0 + (counter & 0xffff);
-	else
-		msr = MSR_IA32_PERFCTR0 + counter;
+	if (!xenpmu_data || !(xenpmu_flags & XENPMU_IRQ_PROCESSING)) {
+		uint32_t msr;
+		int err;
+
+		if (counter & (1 << INTEL_PMC_TYPE_SHIFT))
+			msr = MSR_CORE_PERF_FIXED_CTR0 + (counter & 0xffff);
+		else
+			msr = MSR_IA32_PERFCTR0 + counter;
 
-	return native_read_msr_safe(msr, &err);
+		return native_read_msr_safe(msr, &err);
+	}
+
+	ctxt = &xenpmu_data->pmu.intel;
+	if (counter & (1 << INTEL_PMC_TYPE_SHIFT)) {
+		fixed_counters = field_offset(ctxt, fixed_counters);
+		return fixed_counters[counter & 0xffff];
+	} else {
+		arch_cntr_pair = field_offset(ctxt, arch_counters);
+		return arch_cntr_pair[counter].counter;
+	}
 }
 
 unsigned long long xen_read_pmc(int counter)
@@ -249,6 +395,10 @@ int pmu_apic_update(uint32_t val)
 	}
 
 	xenpmu_data->pmu.lapic_lvtpc = val;
+
+	if (get_xenpmu_flags() & XENPMU_IRQ_PROCESSING)
+		return 0;
+
 	ret = HYPERVISOR_xenpmu_op(XENPMU_lvtpc_set, NULL);
 
 	return ret;
@@ -313,28 +463,34 @@ irqreturn_t xen_pmu_irq_handler(int irq, void *dev_id)
 	int err, ret = IRQ_NONE;
 	struct pt_regs regs;
 	const struct xen_pmu_data *xenpmu_data = get_xenpmu_data();
+	uint8_t xenpmu_flags = get_xenpmu_flags();
 
 	if (!xenpmu_data) {
 		WARN_ONCE(1, "%s: pmudata not initialized\n", __func__);
 		return ret;
 	}
 
-	err = HYPERVISOR_xenpmu_op(XENPMU_flush, NULL);
-	if (err) {
-		WARN_ONCE(1, "%s failed hypercall, err: %d\n", __func__, err);
-		return ret;
-	}
+	this_cpu_ptr(&xenpmu_shared)->flags = xenpmu_flags |
+					      XENPMU_IRQ_PROCESSING;
 
 	xen_convert_regs(&xenpmu_data->pmu.regs, &regs);
 	if (x86_pmu.handle_irq(&regs))
 		ret = IRQ_HANDLED;
 
+	/* Write out cached context to HW */
+	err = HYPERVISOR_xenpmu_op(XENPMU_flush, NULL);
+	this_cpu_ptr(&xenpmu_shared)->flags = xenpmu_flags;
+	if (err) {
+		WARN_ONCE(1, "%s failed hypercall, err: %d\n", __func__, err);
+		return IRQ_NONE;
+	}
+
 	return ret;
 }
 
 bool is_xen_pmu(int cpu)
 {
-	return (per_cpu(xenpmu_shared, cpu) != NULL);
+	return (get_xenpmu_data() != NULL);
 }
 
 int xen_pmu_init(int cpu)
@@ -360,7 +516,8 @@ int xen_pmu_init(int cpu)
 	if (ret)
 		goto fail;
 
-	per_cpu(xenpmu_shared, cpu) = xenpmu_data;
+	per_cpu(xenpmu_shared, cpu).xenpmu_data = xenpmu_data;
+	per_cpu(xenpmu_shared, cpu).flags = 0;
 
 	if (cpu == 0) {
 		perf_register_guest_info_callbacks(&xen_guest_cbs);
@@ -384,6 +541,6 @@ void xen_pmu_finish(int cpu)
 
 	(void)HYPERVISOR_xenpmu_op(XENPMU_finish, &xp);
 
-	free_pages((unsigned long)per_cpu(xenpmu_shared, cpu), 0);
-	per_cpu(xenpmu_shared, cpu) = NULL;
+	free_pages((unsigned long)per_cpu(xenpmu_shared, cpu).xenpmu_data, 0);
+	per_cpu(xenpmu_shared, cpu).xenpmu_data = NULL;
 }
-- 
1.8.1.4