From: Thiago Jung Bauermann <bauerman@linux.vnet.ibm.com>
To: linuxppc-dev@lists.ozlabs.org
Cc: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>,
Thiago Jung Bauermann <bauerman@linux.vnet.ibm.com>
Subject: [PATCH v2 8/8] powerpc/perf/hv-24x7: Aggregate result elements on POWER9 SMT8
Date: Wed, 28 Jun 2017 20:10:26 -0300 [thread overview]
Message-ID: <20170628231026.23190-9-bauerman@linux.vnet.ibm.com> (raw)
In-Reply-To: <20170628231026.23190-1-bauerman@linux.vnet.ibm.com>
On POWER9 SMT8 the 24x7 API returns two result elements for physical core
and virtual CPU events and we need to add their counts to get the final
result.
Signed-off-by: Thiago Jung Bauermann <bauerman@linux.vnet.ibm.com>
---
arch/powerpc/perf/hv-24x7.c | 54 +++++++++++++++++++++++++++++++++++----------
1 file changed, 42 insertions(+), 12 deletions(-)
diff --git a/arch/powerpc/perf/hv-24x7.c b/arch/powerpc/perf/hv-24x7.c
index 38f6a81683a8..8a61b2415186 100644
--- a/arch/powerpc/perf/hv-24x7.c
+++ b/arch/powerpc/perf/hv-24x7.c
@@ -31,6 +31,9 @@
/* Version of the 24x7 hypervisor API that we should use in this machine. */
static int interface_version;
+/* Whether we have to aggregate result data for some domains. */
+static bool aggregate_result_elements;
+
static bool domain_is_valid(unsigned domain)
{
switch (domain) {
@@ -58,6 +61,15 @@ static bool is_physical_domain(unsigned domain)
}
}
+/* Domains for which more than one result element are returned for each event. */
+static bool domain_needs_aggregation(unsigned int domain)
+{
+ return aggregate_result_elements &&
+ (domain == HV_PERF_DOMAIN_PHYS_CORE ||
+ (domain >= HV_PERF_DOMAIN_VCPU_HOME_CORE &&
+ domain <= HV_PERF_DOMAIN_VCPU_REMOTE_NODE));
+}
+
static const char *domain_name(unsigned domain)
{
if (!domain_is_valid(domain))
@@ -1145,17 +1157,23 @@ static int add_event_to_24x7_request(struct perf_event *event,
req->starting_ix = cpu_to_be16(idx);
req->max_ix = cpu_to_be16(1);
- if (request_buffer->interface_version > 1 &&
- req->performance_domain != HV_PERF_DOMAIN_PHYS_CHIP) {
- req->starting_thread_group_ix = idx % 2;
- req->max_num_thread_groups = 1;
+ if (request_buffer->interface_version > 1) {
+ if (domain_needs_aggregation(req->performance_domain))
+ req->max_num_thread_groups = -1;
+ else if (req->performance_domain != HV_PERF_DOMAIN_PHYS_CHIP) {
+ req->starting_thread_group_ix = idx % 2;
+ req->max_num_thread_groups = 1;
+ }
}
return 0;
}
/**
- * get_count_from_result - get event count from the given result
+ * get_count_from_result - get event count from all result elements in result
+ *
+ * If the event corresponding to this result needs aggregation of the result
+ * element values, then this function does that.
*
* @event: Event associated with @res.
* @resb: Result buffer containing @res.
@@ -1172,6 +1190,8 @@ static int get_count_from_result(struct perf_event *event,
u16 data_size = be16_to_cpu(res->result_element_data_size);
unsigned int data_offset;
void *element_data;
+ int i;
+ u64 count;
/*
* We can bail out early if the result is empty.
@@ -1189,8 +1209,10 @@ static int get_count_from_result(struct perf_event *event,
/*
* Since we always specify 1 as the maximum for the smallest resource
* we're requesting, there should to be only one element per result.
+ * Except when an event needs aggregation, in which case there are more.
*/
- if (num_elements != 1) {
+ if (num_elements != 1 &&
+ !domain_needs_aggregation(event_get_domain(event))) {
pr_err("Error: result of request %hhu has %hu elements\n",
res->result_ix, num_elements);
@@ -1211,13 +1233,17 @@ static int get_count_from_result(struct perf_event *event,
data_offset = offsetof(struct hv_24x7_result_element_v2,
element_data);
- element_data = res->elements + data_offset;
+ /* Go through the result elements in the result. */
+ for (i = count = 0, element_data = res->elements + data_offset;
+ i < num_elements;
+ i++, element_data += data_size + data_offset)
+ count += be64_to_cpu(*((u64 *) element_data));
- *countp = be64_to_cpu(*((u64 *) element_data));
+ *countp = count;
- /* The next result is after the result element. */
+ /* The next result is after the last result element. */
if (next)
- *next = element_data + data_size;
+ *next = element_data - data_offset;
return 0;
}
@@ -1565,9 +1591,13 @@ static int hv_24x7_init(void)
}
/* POWER8 only supports v1, while POWER9 only supports v2. */
- if (cpu_has_feature(CPU_FTR_ARCH_300))
+ if (cpu_has_feature(CPU_FTR_ARCH_300)) {
interface_version = 2;
- else
+
+ /* SMT8 in POWER9 needs to aggregate result elements. */
+ if (threads_per_core == 8)
+ aggregate_result_elements = true;
+ } else
interface_version = 1;
hret = hv_perf_caps_get(&caps);
--
2.13.0
prev parent reply other threads:[~2017-06-28 23:11 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-06-28 23:10 [PATCH v2 0/8] Support for 24x7 hcall interface version 2 Thiago Jung Bauermann
2017-06-28 23:10 ` [PATCH v2 1/8] powerpc/perf/hv-24x7: Fix passing of catalog version number Thiago Jung Bauermann
2017-06-28 23:10 ` [PATCH v2 2/8] powerpc/perf/hv-24x7: Fix off-by-one error in request_buffer check Thiago Jung Bauermann
2017-06-28 23:10 ` [PATCH v2 3/8] powerpc/perf/hv-24x7: Properly iterate through results Thiago Jung Bauermann
2017-06-28 23:10 ` [PATCH v2 4/8] powerpc-perf/hx-24x7: Don't log failed hcall twice Thiago Jung Bauermann
2017-06-28 23:10 ` [PATCH v2 5/8] powerpc/perf/hv-24x7: Fix return value of hcalls Thiago Jung Bauermann
2017-06-28 23:10 ` [PATCH v2 6/8] powerpc/perf/hv-24x7: Minor improvements Thiago Jung Bauermann
2017-06-28 23:10 ` [PATCH v2 7/8] powerpc/perf/hv-24x7: Support v2 of the hypervisor API Thiago Jung Bauermann
2017-06-28 23:10 ` Thiago Jung Bauermann [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20170628231026.23190-9-bauerman@linux.vnet.ibm.com \
--to=bauerman@linux.vnet.ibm.com \
--cc=linuxppc-dev@lists.ozlabs.org \
--cc=sukadev@linux.vnet.ibm.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).