powerpc/perf: Add support for the hv gpci (get performance counter info) interface
authorCody P Schafer <cody@linux.vnet.ibm.com>
Fri, 14 Mar 2014 05:00:41 +0000 (16:00 +1100)
committerBenjamin Herrenschmidt <benh@kernel.crashing.org>
Sun, 23 Mar 2014 22:48:31 +0000 (09:48 +1100)
This provides a basic link between perf and hv_gpci. Notably, it does
not yet support transactions and does not list any events (they can
still be manually composed).

Example usage via perf tool:

perf stat -e 'hv_gpci/counter_info_version=3,offset=0,length=8,secondary_index=0,starting_index=0xffffffff,request=0x10/' -r 0 -C 0 -x ' ' sleep 0.1

Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
arch/powerpc/perf/hv-gpci.c [new file with mode: 0644]

diff --git a/arch/powerpc/perf/hv-gpci.c b/arch/powerpc/perf/hv-gpci.c
new file mode 100644 (file)
index 0000000..278ba7b
--- /dev/null
@@ -0,0 +1,294 @@
+/*
+ * Hypervisor supplied "gpci" ("get performance counter info") performance
+ * counter support
+ *
+ * Author: Cody P Schafer <cody@linux.vnet.ibm.com>
+ * Copyright 2014 IBM Corporation.
+ *
+ * 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; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#define pr_fmt(fmt) "hv-gpci: " fmt
+
+#include <linux/init.h>
+#include <linux/perf_event.h>
+#include <asm/firmware.h>
+#include <asm/hvcall.h>
+#include <asm/io.h>
+
+#include "hv-gpci.h"
+#include "hv-common.h"
+
+/*
+ * Example usage:
+ *  perf stat -e 'hv_gpci/counter_info_version=3,offset=0,length=8,
+ *               secondary_index=0,starting_index=0xffffffff,request=0x10/' ...
+ */
+
+/* u32 */
+EVENT_DEFINE_RANGE_FORMAT(request, config, 0, 31);
+/* u32 */
+EVENT_DEFINE_RANGE_FORMAT(starting_index, config, 32, 63);
+/* u16 */
+EVENT_DEFINE_RANGE_FORMAT(secondary_index, config1, 0, 15);
+/* u8 */
+EVENT_DEFINE_RANGE_FORMAT(counter_info_version, config1, 16, 23);
+/* u8, bytes of data (1-8) */
+EVENT_DEFINE_RANGE_FORMAT(length, config1, 24, 31);
+/* u32, byte offset */
+EVENT_DEFINE_RANGE_FORMAT(offset, config1, 32, 63);
+
+static struct attribute *format_attrs[] = {
+       &format_attr_request.attr,
+       &format_attr_starting_index.attr,
+       &format_attr_secondary_index.attr,
+       &format_attr_counter_info_version.attr,
+
+       &format_attr_offset.attr,
+       &format_attr_length.attr,
+       NULL,
+};
+
+static struct attribute_group format_group = {
+       .name = "format",
+       .attrs = format_attrs,
+};
+
+#define HV_CAPS_ATTR(_name, _format)                           \
+static ssize_t _name##_show(struct device *dev,                        \
+                           struct device_attribute *attr,      \
+                           char *page)                         \
+{                                                              \
+       struct hv_perf_caps caps;                               \
+       unsigned long hret = hv_perf_caps_get(&caps);           \
+       if (hret)                                               \
+               return -EIO;                                    \
+                                                               \
+       return sprintf(page, _format, caps._name);              \
+}                                                              \
+static struct device_attribute hv_caps_attr_##_name = __ATTR_RO(_name)
+
+static ssize_t kernel_version_show(struct device *dev,
+                                  struct device_attribute *attr,
+                                  char *page)
+{
+       return sprintf(page, "0x%x\n", COUNTER_INFO_VERSION_CURRENT);
+}
+
+DEVICE_ATTR_RO(kernel_version);
+HV_CAPS_ATTR(version, "0x%x\n");
+HV_CAPS_ATTR(ga, "%d\n");
+HV_CAPS_ATTR(expanded, "%d\n");
+HV_CAPS_ATTR(lab, "%d\n");
+HV_CAPS_ATTR(collect_privileged, "%d\n");
+
+static struct attribute *interface_attrs[] = {
+       &dev_attr_kernel_version.attr,
+       &hv_caps_attr_version.attr,
+       &hv_caps_attr_ga.attr,
+       &hv_caps_attr_expanded.attr,
+       &hv_caps_attr_lab.attr,
+       &hv_caps_attr_collect_privileged.attr,
+       NULL,
+};
+
+static struct attribute_group interface_group = {
+       .name = "interface",
+       .attrs = interface_attrs,
+};
+
+static const struct attribute_group *attr_groups[] = {
+       &format_group,
+       &interface_group,
+       NULL,
+};
+
+#define GPCI_MAX_DATA_BYTES \
+       (1024 - sizeof(struct hv_get_perf_counter_info_params))
+
+static unsigned long single_gpci_request(u32 req, u32 starting_index,
+               u16 secondary_index, u8 version_in, u32 offset, u8 length,
+               u64 *value)
+{
+       unsigned long ret;
+       size_t i;
+       u64 count;
+
+       struct {
+               struct hv_get_perf_counter_info_params params;
+               uint8_t bytes[GPCI_MAX_DATA_BYTES];
+       } __packed __aligned(sizeof(uint64_t)) arg = {
+               .params = {
+                       .counter_request = cpu_to_be32(req),
+                       .starting_index = cpu_to_be32(starting_index),
+                       .secondary_index = cpu_to_be16(secondary_index),
+                       .counter_info_version_in = version_in,
+               }
+       };
+
+       ret = plpar_hcall_norets(H_GET_PERF_COUNTER_INFO,
+                       virt_to_phys(&arg), sizeof(arg));
+       if (ret) {
+               pr_devel("hcall failed: 0x%lx\n", ret);
+               return ret;
+       }
+
+       /*
+        * we verify offset and length are within the zeroed buffer at event
+        * init.
+        */
+       count = 0;
+       for (i = offset; i < offset + length; i++)
+               count |= arg.bytes[i] << (i - offset);
+
+       *value = count;
+       return ret;
+}
+
+static u64 h_gpci_get_value(struct perf_event *event)
+{
+       u64 count;
+       unsigned long ret = single_gpci_request(event_get_request(event),
+                                       event_get_starting_index(event),
+                                       event_get_secondary_index(event),
+                                       event_get_counter_info_version(event),
+                                       event_get_offset(event),
+                                       event_get_length(event),
+                                       &count);
+       if (ret)
+               return 0;
+       return count;
+}
+
+static void h_gpci_event_update(struct perf_event *event)
+{
+       s64 prev;
+       u64 now = h_gpci_get_value(event);
+       prev = local64_xchg(&event->hw.prev_count, now);
+       local64_add(now - prev, &event->count);
+}
+
+static void h_gpci_event_start(struct perf_event *event, int flags)
+{
+       local64_set(&event->hw.prev_count, h_gpci_get_value(event));
+}
+
+static void h_gpci_event_stop(struct perf_event *event, int flags)
+{
+       h_gpci_event_update(event);
+}
+
+static int h_gpci_event_add(struct perf_event *event, int flags)
+{
+       if (flags & PERF_EF_START)
+               h_gpci_event_start(event, flags);
+
+       return 0;
+}
+
+static int h_gpci_event_init(struct perf_event *event)
+{
+       u64 count;
+       u8 length;
+
+       /* Not our event */
+       if (event->attr.type != event->pmu->type)
+               return -ENOENT;
+
+       /* config2 is unused */
+       if (event->attr.config2) {
+               pr_devel("config2 set when reserved\n");
+               return -EINVAL;
+       }
+
+       /* unsupported modes and filters */
+       if (event->attr.exclude_user   ||
+           event->attr.exclude_kernel ||
+           event->attr.exclude_hv     ||
+           event->attr.exclude_idle   ||
+           event->attr.exclude_host   ||
+           event->attr.exclude_guest  ||
+           is_sampling_event(event)) /* no sampling */
+               return -EINVAL;
+
+       /* no branch sampling */
+       if (has_branch_stack(event))
+               return -EOPNOTSUPP;
+
+       length = event_get_length(event);
+       if (length < 1 || length > 8) {
+               pr_devel("length invalid\n");
+               return -EINVAL;
+       }
+
+       /* last byte within the buffer? */
+       if ((event_get_offset(event) + length) > GPCI_MAX_DATA_BYTES) {
+               pr_devel("request outside of buffer: %zu > %zu\n",
+                               (size_t)event_get_offset(event) + length,
+                               GPCI_MAX_DATA_BYTES);
+               return -EINVAL;
+       }
+
+       /* check if the request works... */
+       if (single_gpci_request(event_get_request(event),
+                               event_get_starting_index(event),
+                               event_get_secondary_index(event),
+                               event_get_counter_info_version(event),
+                               event_get_offset(event),
+                               length,
+                               &count)) {
+               pr_devel("gpci hcall failed\n");
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
+static int h_gpci_event_idx(struct perf_event *event)
+{
+       return 0;
+}
+
+static struct pmu h_gpci_pmu = {
+       .task_ctx_nr = perf_invalid_context,
+
+       .name = "hv_gpci",
+       .attr_groups = attr_groups,
+       .event_init  = h_gpci_event_init,
+       .add         = h_gpci_event_add,
+       .del         = h_gpci_event_stop,
+       .start       = h_gpci_event_start,
+       .stop        = h_gpci_event_stop,
+       .read        = h_gpci_event_update,
+       .event_idx   = h_gpci_event_idx,
+};
+
+static int hv_gpci_init(void)
+{
+       int r;
+       unsigned long hret;
+       struct hv_perf_caps caps;
+
+       if (!firmware_has_feature(FW_FEATURE_LPAR)) {
+               pr_info("not a virtualized system, not enabling\n");
+               return -ENODEV;
+       }
+
+       hret = hv_perf_caps_get(&caps);
+       if (hret) {
+               pr_info("could not obtain capabilities, error 0x%80lx, not enabling\n",
+                               hret);
+               return -ENODEV;
+       }
+
+       r = perf_pmu_register(&h_gpci_pmu, h_gpci_pmu.name, -1);
+       if (r)
+               return r;
+
+       return 0;
+}
+
+device_initcall(hv_gpci_init);