From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mx0a-001b2d01.pphosted.com (mx0b-001b2d01.pphosted.com [148.163.158.5]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by lists.ozlabs.org (Postfix) with ESMTPS id 3wf0F54K0QzDqMs for ; Fri, 2 Jun 2017 07:03:37 +1000 (AEST) Received: from pps.filterd (m0098417.ppops.net [127.0.0.1]) by mx0a-001b2d01.pphosted.com (8.16.0.20/8.16.0.20) with SMTP id v51Kwpr1058492 for ; Thu, 1 Jun 2017 17:03:26 -0400 Received: from e24smtp05.br.ibm.com (e24smtp05.br.ibm.com [32.104.18.26]) by mx0a-001b2d01.pphosted.com with ESMTP id 2att680tgf-1 (version=TLSv1.2 cipher=AES256-SHA bits=256 verify=NOT) for ; Thu, 01 Jun 2017 17:03:26 -0400 Received: from localhost by e24smtp05.br.ibm.com with IBM ESMTP SMTP Gateway: Authorized Use Only! Violators will be prosecuted for from ; Thu, 1 Jun 2017 18:03:24 -0300 Received: from d24av01.br.ibm.com (d24av01.br.ibm.com [9.8.31.91]) by d24relay04.br.ibm.com (8.14.9/8.14.9/NCO v10.0) with ESMTP id v51L3MUe62783510 for ; Thu, 1 Jun 2017 18:03:22 -0300 Received: from d24av01.br.ibm.com (localhost [127.0.0.1]) by d24av01.br.ibm.com (8.14.4/8.14.4/NCO v10.0 AVout) with ESMTP id v51L3NlT003131 for ; Thu, 1 Jun 2017 18:03:23 -0300 From: Thiago Jung Bauermann To: linuxppc-dev@lists.ozlabs.org Cc: Sukadev Bhattiprolu , Thiago Jung Bauermann Subject: [PATCH 8/8] powerpc/perf/hv-24x7: Aggregate result elements on POWER9 SMT8 Date: Thu, 1 Jun 2017 18:02:27 -0300 In-Reply-To: <1496350947-30951-1-git-send-email-bauerman@linux.vnet.ibm.com> References: <1496350947-30951-1-git-send-email-bauerman@linux.vnet.ibm.com> Message-Id: <1496350947-30951-9-git-send-email-bauerman@linux.vnet.ibm.com> List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , 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 --- arch/powerpc/perf/hv-24x7.c | 58 ++++++++++++++++++++++++++++++++++----------- 1 file changed, 44 insertions(+), 14 deletions(-) diff --git a/arch/powerpc/perf/hv-24x7.c b/arch/powerpc/perf/hv-24x7.c index 95c44f1d2fd2..641f385e7eb0 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)) @@ -1149,17 +1161,23 @@ 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. @@ -1176,7 +1194,8 @@ 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 ret = 0; + int i, ret = 0; + u64 count; /* * We can bail out early if the result is empty. @@ -1192,9 +1211,11 @@ static int get_count_from_result(struct perf_event *event, } /* - * This code assumes that a result has only one element. + * This code assumes that a result has only one element, except + * when an event needs aggregation. */ - if (num_elements != 1) { + if (num_elements != 1 && + !domain_needs_aggregation(event_get_domain(event))) { pr_debug("Error: result of request %hhu has %hu elements\n", res->result_ix, num_elements); @@ -1225,14 +1246,19 @@ 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) + if (!ret) + count += be64_to_cpu(*((u64 *) element_data)); if (!ret) - *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 ret; } @@ -1580,9 +1606,13 @@ 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); -- 2.7.4