Message ID | 20240227181632.659133-3-quic_sibis@quicinc.com |
---|---|
State | New |
Headers |
Return-Path: <linux-kernel+bounces-83827-ouuuleilei=gmail.com@vger.kernel.org> Delivered-To: ouuuleilei@gmail.com Received: by 2002:a05:7300:a81b:b0:108:e6aa:91d0 with SMTP id bq27csp2883389dyb; Tue, 27 Feb 2024 10:23:00 -0800 (PST) X-Forwarded-Encrypted: i=3; AJvYcCXyTZvFT5nHMUCQDkZD5QCPWD4HBROo7N8J2bGjDjqP9pOO5hh79S2snQb6RWdcK0jBoN91VpLKkcTrODi1nUxIPa1zOA== X-Google-Smtp-Source: AGHT+IEmjd+h/GamJnH0Lm7Ay1xk7PPL257OkMOLvnzYqx4pyLQ+JMbStJ5inbeVJVLhupD1QIqq X-Received: by 2002:a05:6a20:8993:b0:1a0:df79:656f with SMTP id h19-20020a056a20899300b001a0df79656fmr2601365pzg.31.1709058180334; Tue, 27 Feb 2024 10:23:00 -0800 (PST) ARC-Seal: i=2; a=rsa-sha256; t=1709058180; cv=pass; d=google.com; s=arc-20160816; b=NIz1H1pm5HzoF1NPY4C/dDGdmDrriwdjViO796diQdWfOoBT1kjrYkKe1gWb95zt5q aBiqnSPHNZ1azH90fPscu5Nu32BGDxENXVvL2rV/5vYTidl4Y2QasCqaKKY2OYjHoJQY MGpkvzSR5wgCNuUyD33CojEPK+gNroiokBrD3XfUgeNkegkVY9x5RF2C61RkRbVHKfc2 DyS+VELh1NdEJLvADBMOmUgZS+lEb0hsSJHifxgy9DsJ4bYxsvX0OsvBhEarBz2aKLcU 31hbjYfGpMjhgyp/U9FCrEnBYyIOkaLSaGG8cWbHWGYi09/+IARlXWMB1cGBDPUz1ndi CjMQ== ARC-Message-Signature: i=2; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=content-transfer-encoding:mime-version:list-unsubscribe :list-subscribe:list-id:precedence:references:in-reply-to:message-id :date:subject:cc:to:from:dkim-signature; bh=ornixnyEGdWMbFwNeDMGNQb9hYIkJYlnDtUaJ6GrfQ8=; fh=S1FWNMyHQafB8jSVFMYxufENq6nTcLKph3ifanCJdss=; b=Tg7e0d8N1CVDvBNazbkTRFsOTmx8VBzOpvg1TI7C6urTpTs8fqDjDzTD9skaPULGKD Mvs96z9uEhzKyc3oyQHkUdW8iep4mqgpz0xWC3169u7n3Fl0c6Qf6Eh/wIsic+h0YQAo wwa/vGc9S23RtiZhamsRFQ6ipPCfjOZJBOUwyFArQq7OT5X+IXbn+HYeI72ybMoUFTfB N9lYZd3V4UtrNwk4rxvip+csVu/qLXCkXPt0AYyILV4VOn9OF+DB7TzK7ITPYI4sdhiO 3UV4uOHu/5h+90dmBWgsmAZa2jfrSAFENI3JZTBJRO4fKxzrzRQuhDyfQaqFfYu1XZHp GPHQ==; dara=google.com ARC-Authentication-Results: i=2; mx.google.com; dkim=pass header.i=@quicinc.com header.s=qcppdkim1 header.b=phSOurph; arc=pass (i=1 spf=pass spfdomain=quicinc.com dkim=pass dkdomain=quicinc.com dmarc=pass fromdomain=quicinc.com); spf=pass (google.com: domain of linux-kernel+bounces-83827-ouuuleilei=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) smtp.mailfrom="linux-kernel+bounces-83827-ouuuleilei=gmail.com@vger.kernel.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=quicinc.com Received: from sv.mirrors.kernel.org (sv.mirrors.kernel.org. [139.178.88.99]) by mx.google.com with ESMTPS id ch5-20020a056a00288500b006e530ea9c43si3638691pfb.302.2024.02.27.10.23.00 for <ouuuleilei@gmail.com> (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 27 Feb 2024 10:23:00 -0800 (PST) Received-SPF: pass (google.com: domain of linux-kernel+bounces-83827-ouuuleilei=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99; Authentication-Results: mx.google.com; dkim=pass header.i=@quicinc.com header.s=qcppdkim1 header.b=phSOurph; arc=pass (i=1 spf=pass spfdomain=quicinc.com dkim=pass dkdomain=quicinc.com dmarc=pass fromdomain=quicinc.com); spf=pass (google.com: domain of linux-kernel+bounces-83827-ouuuleilei=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) smtp.mailfrom="linux-kernel+bounces-83827-ouuuleilei=gmail.com@vger.kernel.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=quicinc.com Received: from smtp.subspace.kernel.org (wormhole.subspace.kernel.org [52.25.139.140]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by sv.mirrors.kernel.org (Postfix) with ESMTPS id 31C0329050C for <ouuuleilei@gmail.com>; Tue, 27 Feb 2024 18:21:23 +0000 (UTC) Received: from localhost.localdomain (localhost.localdomain [127.0.0.1]) by smtp.subspace.kernel.org (Postfix) with ESMTP id 38AFB15958C; Tue, 27 Feb 2024 18:17:31 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=quicinc.com header.i=@quicinc.com header.b="phSOurph" Received: from mx0a-0031df01.pphosted.com (mx0a-0031df01.pphosted.com [205.220.168.131]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id D78111487D8; Tue, 27 Feb 2024 18:17:26 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=205.220.168.131 ARC-Seal: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1709057848; cv=none; b=iYbLcmIK7ElVQkEmpAVpa181efoU0T0jtH2ec0JBWH9ci0/0xRlWqnO4cvXuw16SAPOjy4ddqx495bgJ/Bhi5ArnZNeeR5BXgIzFoBE1/LrgTP6hK77tGguS9DJz5hIguOeKh0e+pGssiT8uLIIOQQjYqfxE268FUEA19hTcrvM= ARC-Message-Signature: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1709057848; c=relaxed/simple; bh=zAsfk+v4hJQHezpar3y9bfxnsJGJOGSKx2JbFHeCxXM=; h=From:To:CC:Subject:Date:Message-ID:In-Reply-To:References: MIME-Version:Content-Type; b=GLlHk+I5c6QFAQrkyV+v/Cp/M+XNGmvpnAErszNo+cSzaetdM0ECd+396V8pdcIXZ5pvDx/5j/EjHJLyl4QIy+XY1io06shPop9rZysh10CPXRe2Tjsd7QFgk5g9++z83oqw/Ahh5p8IUjNQZqXEZYgLMqNk1YN+Lk7WR4i8yoU= ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=quicinc.com; spf=pass smtp.mailfrom=quicinc.com; dkim=pass (2048-bit key) header.d=quicinc.com header.i=@quicinc.com header.b=phSOurph; arc=none smtp.client-ip=205.220.168.131 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=quicinc.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=quicinc.com Received: from pps.filterd (m0279867.ppops.net [127.0.0.1]) by mx0a-0031df01.pphosted.com (8.17.1.24/8.17.1.24) with ESMTP id 41R9Y43L004484; Tue, 27 Feb 2024 18:17:16 GMT DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=quicinc.com; h= from:to:cc:subject:date:message-id:in-reply-to:references :mime-version:content-transfer-encoding:content-type; s= qcppdkim1; bh=ornixnyEGdWMbFwNeDMGNQb9hYIkJYlnDtUaJ6GrfQ8=; b=ph SOurphNJeuGw3Lsf1A0AT6CsAzIuey4z21gIpqJ4hErPFuknPAGR5pPbhjqemBnR 8IEPKqjhxlxbiIfXDoQ+vAZ4bHBrIQjYh9F5vIJ0a01XN0icou2fIzm3o3VrdQAV DGWwL4veU0Gaq6RDF0/4pHcUIJXrqvtU3wB82laHU868aYPvJEzNu7r9Ycu9Ahz9 /EWAmLLe7jjfgacAPbTkDKuf7HEpwCuZNtH+HhCOQOH0nruIfkBes/qV8mGvV2aZ Deu65RpXf1kKd67vYMFPi0ONr4eNf+rwbpvl55cd52NpNKLl6Dz2r83FUs2H+U0D EH4FJMYhUYimpMHIrf5w== Received: from nalasppmta05.qualcomm.com (Global_NAT1.qualcomm.com [129.46.96.20]) by mx0a-0031df01.pphosted.com (PPS) with ESMTPS id 3whd7b18t2-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NOT); Tue, 27 Feb 2024 18:17:15 +0000 (GMT) Received: from nalasex01a.na.qualcomm.com (nalasex01a.na.qualcomm.com [10.47.209.196]) by NALASPPMTA05.qualcomm.com (8.17.1.5/8.17.1.5) with ESMTPS id 41RIH3Ae010219 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NOT); Tue, 27 Feb 2024 18:17:03 GMT Received: from hu-sibis-blr.qualcomm.com (10.80.80.8) by nalasex01a.na.qualcomm.com (10.47.209.196) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.2.1118.40; Tue, 27 Feb 2024 10:16:59 -0800 From: Sibi Sankar <quic_sibis@quicinc.com> To: <sudeep.holla@arm.com>, <cristian.marussi@arm.com>, <rafael@kernel.org>, <viresh.kumar@linaro.org>, <morten.rasmussen@arm.com>, <dietmar.eggemann@arm.com>, <lukasz.luba@arm.com>, <pierre.gondois@arm.com> CC: <linux-arm-kernel@lists.infradead.org>, <linux-pm@vger.kernel.org>, <linux-kernel@vger.kernel.org>, <quic_mdtipton@quicinc.com>, <linux-arm-msm@vger.kernel.org>, Sibi Sankar <quic_sibis@quicinc.com> Subject: [PATCH V3 2/2] cpufreq: scmi: Register for limit change notifications Date: Tue, 27 Feb 2024 23:46:32 +0530 Message-ID: <20240227181632.659133-3-quic_sibis@quicinc.com> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20240227181632.659133-1-quic_sibis@quicinc.com> References: <20240227181632.659133-1-quic_sibis@quicinc.com> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: <linux-kernel.vger.kernel.org> List-Subscribe: <mailto:linux-kernel+subscribe@vger.kernel.org> List-Unsubscribe: <mailto:linux-kernel+unsubscribe@vger.kernel.org> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Content-Type: text/plain X-ClientProxiedBy: nasanex01a.na.qualcomm.com (10.52.223.231) To nalasex01a.na.qualcomm.com (10.47.209.196) X-QCInternal: smtphost X-Proofpoint-Virus-Version: vendor=nai engine=6200 definitions=5800 signatures=585085 X-Proofpoint-GUID: ZEb7q_z4kIURjHSXm1GTUsykMg-r9OjA X-Proofpoint-ORIG-GUID: ZEb7q_z4kIURjHSXm1GTUsykMg-r9OjA X-Proofpoint-Virus-Version: vendor=baseguard engine=ICAP:2.0.272,Aquarius:18.0.1011,Hydra:6.0.619,FMLib:17.11.176.26 definitions=2024-02-27_05,2024-02-27_01,2023-05-22_02 X-Proofpoint-Spam-Details: rule=outbound_notspam policy=outbound score=0 clxscore=1015 impostorscore=0 mlxlogscore=999 priorityscore=1501 spamscore=0 adultscore=0 malwarescore=0 mlxscore=0 lowpriorityscore=0 phishscore=0 bulkscore=0 suspectscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.19.0-2402120000 definitions=main-2402270142 X-getmail-retrieved-from-mailbox: INBOX X-GMAIL-THRID: 1792077390284684748 X-GMAIL-MSGID: 1792077390284684748 |
Series |
firmware: arm_scmi: Register and handle limits change notification
|
|
Commit Message
Sibi Sankar
Feb. 27, 2024, 6:16 p.m. UTC
Register for limit change notifications if supported and use the throttled
frequency from the notification to apply HW pressure.
Signed-off-by: Sibi Sankar <quic_sibis@quicinc.com>
---
v3:
* Sanitize range_max received from the notifier. [Pierre]
* Update commit message.
drivers/cpufreq/scmi-cpufreq.c | 29 ++++++++++++++++++++++++++++-
1 file changed, 28 insertions(+), 1 deletion(-)
Comments
On 2/27/24 18:16, Sibi Sankar wrote: > Register for limit change notifications if supported and use the throttled > frequency from the notification to apply HW pressure. > > Signed-off-by: Sibi Sankar <quic_sibis@quicinc.com> > --- > > v3: > * Sanitize range_max received from the notifier. [Pierre] > * Update commit message. > > drivers/cpufreq/scmi-cpufreq.c | 29 ++++++++++++++++++++++++++++- > 1 file changed, 28 insertions(+), 1 deletion(-) > > diff --git a/drivers/cpufreq/scmi-cpufreq.c b/drivers/cpufreq/scmi-cpufreq.c > index 76a0ddbd9d24..78b87b72962d 100644 > --- a/drivers/cpufreq/scmi-cpufreq.c > +++ b/drivers/cpufreq/scmi-cpufreq.c > @@ -25,9 +25,13 @@ struct scmi_data { > int domain_id; > int nr_opp; > struct device *cpu_dev; > + struct cpufreq_policy *policy; > cpumask_var_t opp_shared_cpus; > + struct notifier_block limit_notify_nb; > }; > > +const struct scmi_handle *handle; > +static struct scmi_device *scmi_dev; > static struct scmi_protocol_handle *ph; > static const struct scmi_perf_proto_ops *perf_ops; > static struct cpufreq_driver scmi_cpufreq_driver; > @@ -151,6 +155,20 @@ static struct freq_attr *scmi_cpufreq_hw_attr[] = { > NULL, > }; > > +static int scmi_limit_notify_cb(struct notifier_block *nb, unsigned long event, void *data) > +{ > + struct scmi_data *priv = container_of(nb, struct scmi_data, limit_notify_nb); > + struct scmi_perf_limits_report *limit_notify = data; > + struct cpufreq_policy *policy = priv->policy; > + > + policy->max = clamp(limit_notify->range_max_freq/HZ_PER_KHZ, policy->cpuinfo.min_freq, > + policy->cpuinfo.max_freq); Please take the division operation out of this clamp() call, somewhere above. Currently it 'blurs' these stuff, while it's important convertion to khz. You can call it e.g.: limit_freq_khz = limit_notify->range_max_freq / HZ_PER_KHZ; then use in clamp(limit_freq_khz, ...) > + > + cpufreq_update_pressure(policy); > + > + return NOTIFY_OK; > +} > + > static int scmi_cpufreq_init(struct cpufreq_policy *policy) > { > int ret, nr_opp, domain; > @@ -269,6 +287,15 @@ static int scmi_cpufreq_init(struct cpufreq_policy *policy) > } > } > > + priv->limit_notify_nb.notifier_call = scmi_limit_notify_cb; > + ret = handle->notify_ops->devm_event_notifier_register(scmi_dev, SCMI_PROTOCOL_PERF, > + SCMI_EVENT_PERFORMANCE_LIMITS_CHANGED, > + &domain, > + &priv->limit_notify_nb); > + if (ret) > + dev_warn(cpu_dev, > + "failed to register for limits change notifier for domain %d\n", domain); > + > priv->policy = policy; > > return 0; > @@ -342,8 +369,8 @@ static int scmi_cpufreq_probe(struct scmi_device *sdev) > { > int ret; > struct device *dev = &sdev->dev; > - const struct scmi_handle *handle; It should be a compilation error... > > + scmi_dev = sdev; > handle = sdev->handle; due to usage here, wasn't it? > > if (!handle)
On 2/28/24 18:54, Lukasz Luba wrote: > > > On 2/27/24 18:16, Sibi Sankar wrote: >> Register for limit change notifications if supported and use the >> throttled >> frequency from the notification to apply HW pressure. Lukasz, Thanks for taking time to review the series! >> >> Signed-off-by: Sibi Sankar <quic_sibis@quicinc.com> >> --- >> >> v3: >> * Sanitize range_max received from the notifier. [Pierre] >> * Update commit message. >> >> drivers/cpufreq/scmi-cpufreq.c | 29 ++++++++++++++++++++++++++++- >> 1 file changed, 28 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/cpufreq/scmi-cpufreq.c >> b/drivers/cpufreq/scmi-cpufreq.c >> index 76a0ddbd9d24..78b87b72962d 100644 >> --- a/drivers/cpufreq/scmi-cpufreq.c >> +++ b/drivers/cpufreq/scmi-cpufreq.c >> @@ -25,9 +25,13 @@ struct scmi_data { >> int domain_id; >> int nr_opp; >> struct device *cpu_dev; >> + struct cpufreq_policy *policy; >> cpumask_var_t opp_shared_cpus; >> + struct notifier_block limit_notify_nb; >> }; >> +const struct scmi_handle *handle; >> +static struct scmi_device *scmi_dev; >> static struct scmi_protocol_handle *ph; >> static const struct scmi_perf_proto_ops *perf_ops; >> static struct cpufreq_driver scmi_cpufreq_driver; >> @@ -151,6 +155,20 @@ static struct freq_attr *scmi_cpufreq_hw_attr[] = { >> NULL, >> }; >> +static int scmi_limit_notify_cb(struct notifier_block *nb, unsigned >> long event, void *data) >> +{ >> + struct scmi_data *priv = container_of(nb, struct scmi_data, >> limit_notify_nb); >> + struct scmi_perf_limits_report *limit_notify = data; >> + struct cpufreq_policy *policy = priv->policy; >> + >> + policy->max = clamp(limit_notify->range_max_freq/HZ_PER_KHZ, >> policy->cpuinfo.min_freq, >> + policy->cpuinfo.max_freq); > > Please take the division operation out of this clamp() call, somewhere > above. Currently it 'blurs' these stuff, while it's important convertion > to khz. You can call it e.g.: > > limit_freq_khz = limit_notify->range_max_freq / HZ_PER_KHZ; > > then use in clamp(limit_freq_khz, ...) ack > >> + >> + cpufreq_update_pressure(policy); >> + >> + return NOTIFY_OK; >> +} >> + >> static int scmi_cpufreq_init(struct cpufreq_policy *policy) >> { >> int ret, nr_opp, domain; >> @@ -269,6 +287,15 @@ static int scmi_cpufreq_init(struct >> cpufreq_policy *policy) >> } >> } >> + priv->limit_notify_nb.notifier_call = scmi_limit_notify_cb; >> + ret = handle->notify_ops->devm_event_notifier_register(scmi_dev, >> SCMI_PROTOCOL_PERF, >> + SCMI_EVENT_PERFORMANCE_LIMITS_CHANGED, >> + &domain, >> + &priv->limit_notify_nb); >> + if (ret) >> + dev_warn(cpu_dev, >> + "failed to register for limits change notifier for >> domain %d\n", domain); >> + >> priv->policy = policy; >> return 0; >> @@ -342,8 +369,8 @@ static int scmi_cpufreq_probe(struct scmi_device >> *sdev) >> { >> int ret; >> struct device *dev = &sdev->dev; >> - const struct scmi_handle *handle; > > It should be a compilation error... > >> + scmi_dev = sdev; >> handle = sdev->handle; > > due to usage here, wasn't it? Not really, isn't it getting the first initialization here? Are there any compiler options that I need to turn on to catch these? -Sibi > >> if (!handle)
On 2/28/24 17:00, Sibi Sankar wrote: > > > On 2/28/24 18:54, Lukasz Luba wrote: >> >> >> On 2/27/24 18:16, Sibi Sankar wrote: >>> Register for limit change notifications if supported and use the >>> throttled >>> frequency from the notification to apply HW pressure. > > Lukasz, > > Thanks for taking time to review the series! > >>> >>> Signed-off-by: Sibi Sankar <quic_sibis@quicinc.com> >>> --- >>> >>> v3: >>> * Sanitize range_max received from the notifier. [Pierre] >>> * Update commit message. >>> >>> drivers/cpufreq/scmi-cpufreq.c | 29 ++++++++++++++++++++++++++++- >>> 1 file changed, 28 insertions(+), 1 deletion(-) >>> >>> diff --git a/drivers/cpufreq/scmi-cpufreq.c >>> b/drivers/cpufreq/scmi-cpufreq.c >>> index 76a0ddbd9d24..78b87b72962d 100644 >>> --- a/drivers/cpufreq/scmi-cpufreq.c >>> +++ b/drivers/cpufreq/scmi-cpufreq.c >>> @@ -25,9 +25,13 @@ struct scmi_data { >>> int domain_id; >>> int nr_opp; >>> struct device *cpu_dev; >>> + struct cpufreq_policy *policy; >>> cpumask_var_t opp_shared_cpus; >>> + struct notifier_block limit_notify_nb; >>> }; >>> +const struct scmi_handle *handle; I've missed this bit here. >>> +static struct scmi_device *scmi_dev; >>> static struct scmi_protocol_handle *ph; >>> static const struct scmi_perf_proto_ops *perf_ops; >>> static struct cpufreq_driver scmi_cpufreq_driver; >>> @@ -151,6 +155,20 @@ static struct freq_attr *scmi_cpufreq_hw_attr[] = { >>> NULL, >>> }; >>> +static int scmi_limit_notify_cb(struct notifier_block *nb, unsigned >>> long event, void *data) >>> +{ >>> + struct scmi_data *priv = container_of(nb, struct scmi_data, >>> limit_notify_nb); >>> + struct scmi_perf_limits_report *limit_notify = data; >>> + struct cpufreq_policy *policy = priv->policy; >>> + >>> + policy->max = clamp(limit_notify->range_max_freq/HZ_PER_KHZ, >>> policy->cpuinfo.min_freq, >>> + policy->cpuinfo.max_freq); >> >> Please take the division operation out of this clamp() call, somewhere >> above. Currently it 'blurs' these stuff, while it's important convertion >> to khz. You can call it e.g.: >> >> limit_freq_khz = limit_notify->range_max_freq / HZ_PER_KHZ; >> >> then use in clamp(limit_freq_khz, ...) > > ack > >> >>> + >>> + cpufreq_update_pressure(policy); >>> + >>> + return NOTIFY_OK; >>> +} >>> + >>> static int scmi_cpufreq_init(struct cpufreq_policy *policy) >>> { >>> int ret, nr_opp, domain; >>> @@ -269,6 +287,15 @@ static int scmi_cpufreq_init(struct >>> cpufreq_policy *policy) >>> } >>> } >>> + priv->limit_notify_nb.notifier_call = scmi_limit_notify_cb; >>> + ret = handle->notify_ops->devm_event_notifier_register(scmi_dev, >>> SCMI_PROTOCOL_PERF, >>> + SCMI_EVENT_PERFORMANCE_LIMITS_CHANGED, >>> + &domain, >>> + &priv->limit_notify_nb); >>> + if (ret) >>> + dev_warn(cpu_dev, >>> + "failed to register for limits change notifier for >>> domain %d\n", domain); >>> + >>> priv->policy = policy; >>> return 0; >>> @@ -342,8 +369,8 @@ static int scmi_cpufreq_probe(struct scmi_device >>> *sdev) >>> { >>> int ret; >>> struct device *dev = &sdev->dev; >>> - const struct scmi_handle *handle; >> >> It should be a compilation error... >> >>> + scmi_dev = sdev; >>> handle = sdev->handle; >> >> due to usage here, wasn't it? > > Not really, isn't it getting the first initialization here? > Are there any compiler options that I need to turn on to > catch these? Yes, you're right, my apologies for confusion. I couldn't apply the series due issues in two patch sets in your dependency list. Now when I have been manually applying the changes I spotted it.
On 2/29/24 09:59, Lukasz Luba wrote: > > > On 2/28/24 17:00, Sibi Sankar wrote: >> >> >> On 2/28/24 18:54, Lukasz Luba wrote: >>> >>> >>> On 2/27/24 18:16, Sibi Sankar wrote: >>>> Register for limit change notifications if supported and use the >>>> throttled >>>> frequency from the notification to apply HW pressure. >> >> Lukasz, >> >> Thanks for taking time to review the series! >> >>>> >>>> Signed-off-by: Sibi Sankar <quic_sibis@quicinc.com> >>>> --- >>>> >>>> v3: >>>> * Sanitize range_max received from the notifier. [Pierre] >>>> * Update commit message. >>>> >>>> drivers/cpufreq/scmi-cpufreq.c | 29 ++++++++++++++++++++++++++++- >>>> 1 file changed, 28 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/drivers/cpufreq/scmi-cpufreq.c >>>> b/drivers/cpufreq/scmi-cpufreq.c >>>> index 76a0ddbd9d24..78b87b72962d 100644 >>>> --- a/drivers/cpufreq/scmi-cpufreq.c >>>> +++ b/drivers/cpufreq/scmi-cpufreq.c >>>> @@ -25,9 +25,13 @@ struct scmi_data { >>>> int domain_id; >>>> int nr_opp; >>>> struct device *cpu_dev; >>>> + struct cpufreq_policy *policy; >>>> cpumask_var_t opp_shared_cpus; >>>> + struct notifier_block limit_notify_nb; >>>> }; >>>> +const struct scmi_handle *handle; > > I've missed this bit here. So for this change we actually have to ask Cristian or Sudeep because I'm not sure if we have only one 'handle' instance for all cpufreq devices. If we have different 'handle' we cannot move it to the global single pointer. Sudeep, Cristian what do you think?
On Thu, Feb 29, 2024 at 10:22:39AM +0000, Lukasz Luba wrote: > > > On 2/29/24 09:59, Lukasz Luba wrote: > > > > > > On 2/28/24 17:00, Sibi Sankar wrote: > > > > > > > > > On 2/28/24 18:54, Lukasz Luba wrote: > > > > > > > > > > > > On 2/27/24 18:16, Sibi Sankar wrote: > > > > > Register for limit change notifications if supported and use > > > > > the throttled > > > > > frequency from the notification to apply HW pressure. > > > > > > Lukasz, > > > > > > Thanks for taking time to review the series! > > > > > > > > > > > > > Signed-off-by: Sibi Sankar <quic_sibis@quicinc.com> > > > > > --- > > > > > > > > > > v3: > > > > > * Sanitize range_max received from the notifier. [Pierre] > > > > > * Update commit message. > > > > > > > > > > drivers/cpufreq/scmi-cpufreq.c | 29 ++++++++++++++++++++++++++++- > > > > > 1 file changed, 28 insertions(+), 1 deletion(-) > > > > > > > > > > diff --git a/drivers/cpufreq/scmi-cpufreq.c > > > > > b/drivers/cpufreq/scmi-cpufreq.c > > > > > index 76a0ddbd9d24..78b87b72962d 100644 > > > > > --- a/drivers/cpufreq/scmi-cpufreq.c > > > > > +++ b/drivers/cpufreq/scmi-cpufreq.c > > > > > @@ -25,9 +25,13 @@ struct scmi_data { > > > > > int domain_id; > > > > > int nr_opp; > > > > > struct device *cpu_dev; > > > > > + struct cpufreq_policy *policy; > > > > > cpumask_var_t opp_shared_cpus; > > > > > + struct notifier_block limit_notify_nb; > > > > > }; > > > > > +const struct scmi_handle *handle; > > > > I've missed this bit here. > > So for this change we actually have to ask Cristian or Sudeep > because I'm not sure if we have only one 'handle' instance > for all cpufreq devices. > > If we have different 'handle' we cannot move it to the > global single pointer. > > Sudeep, Cristian what do you think? I was just replying noticing this :D .... since SCMI drivers can be probed multiple times IF you defined multiple scmi top nodes in your DT containing the same protocol nodes, they receive a distinct sdev/handle/ph for each probe...so any attempt to globalize these wont work...BUT... ..this is a bit of a weird setup BUT it is not against the spec and it can be used to parallelize more the SCMI accesses to disjont set of resources within the same protocol (a long story here...) AND this type of setup is something that it is already used by some other colleagues of Sibi working on a different line of products (AFAIK)... So, for these reasons, usually, all the other SCMI drivers have per-instance non-global references to handle/sdev/ph.... ..having said that, thought, looking at the structure of CPUFReq drivers, I am not sure that they can stand such a similar setup where multiple instances of this same driver are probed ... indeed the existent *ph refs above is already global....so it wont already work anyway in case of multiple instances now... ..and if I look at how CPUFreq expects the signature of scmi_cpufreq_get_rate() to be annd how it is implemented now using the global *ph reference, it is clearly already not working cleanly on a multi-instance setup... ..now...I can imagine how to (maybe) fix the above removing the globals and fixing this, BUT the question, more generally, is CPUFreq supposed to work at all in this multi-probed mode of operation ? Does it even make sense to be able to support this in CPUFREQ ? (as an example in cpufreq,c there is static global cpufreq_driver pointing to the arch-specific configured driver BUT that also holds some .driver_data AND that cleraly wont be instance specific if you probe multiple times and register with CPUFreq multiple times...) More questions than answers here :D Thanks, Cristian
On 2/29/24 11:28, Cristian Marussi wrote: > On Thu, Feb 29, 2024 at 10:22:39AM +0000, Lukasz Luba wrote: >> >> >> On 2/29/24 09:59, Lukasz Luba wrote: >>> >>> >>> On 2/28/24 17:00, Sibi Sankar wrote: >>>> >>>> >>>> On 2/28/24 18:54, Lukasz Luba wrote: >>>>> >>>>> >>>>> On 2/27/24 18:16, Sibi Sankar wrote: >>>>>> Register for limit change notifications if supported and use >>>>>> the throttled >>>>>> frequency from the notification to apply HW pressure. >>>> >>>> Lukasz, >>>> >>>> Thanks for taking time to review the series! >>>> >>>>>> >>>>>> Signed-off-by: Sibi Sankar <quic_sibis@quicinc.com> >>>>>> --- >>>>>> >>>>>> v3: >>>>>> * Sanitize range_max received from the notifier. [Pierre] >>>>>> * Update commit message. >>>>>> >>>>>> � drivers/cpufreq/scmi-cpufreq.c | 29 ++++++++++++++++++++++++++++- >>>>>> � 1 file changed, 28 insertions(+), 1 deletion(-) >>>>>> >>>>>> diff --git a/drivers/cpufreq/scmi-cpufreq.c >>>>>> b/drivers/cpufreq/scmi-cpufreq.c >>>>>> index 76a0ddbd9d24..78b87b72962d 100644 >>>>>> --- a/drivers/cpufreq/scmi-cpufreq.c >>>>>> +++ b/drivers/cpufreq/scmi-cpufreq.c >>>>>> @@ -25,9 +25,13 @@ struct scmi_data { >>>>>> ����� int domain_id; >>>>>> ����� int nr_opp; >>>>>> ����� struct device *cpu_dev; >>>>>> +��� struct cpufreq_policy *policy; >>>>>> ����� cpumask_var_t opp_shared_cpus; >>>>>> +��� struct notifier_block limit_notify_nb; >>>>>> � }; >>>>>> +const struct scmi_handle *handle; >>> >>> I've missed this bit here. >> >> So for this change we actually have to ask Cristian or Sudeep >> because I'm not sure if we have only one 'handle' instance >> for all cpufreq devices. >> >> If we have different 'handle' we cannot move it to the >> global single pointer. >> >> Sudeep, Cristian what do you think? > > I was just replying noticing this :D .... since SCMI drivers can be > probed multiple times IF you defined multiple scmi top nodes in your DT > containing the same protocol nodes, they receive a distinct sdev/handle/ph > for each probe...so any attempt to globalize these wont work...BUT... > > ...this is a bit of a weird setup BUT it is not against the spec and it can > be used to parallelize more the SCMI accesses to disjont set of resources > within the same protocol (a long story here...) AND this type of setup is > something that it is already used by some other colleagues of Sibi working > on a different line of products (AFAIK)... > > So, for these reasons, usually, all the other SCMI drivers have per-instance > non-global references to handle/sdev/ph.... > > ...having said that, thought, looking at the structure of CPUFReq > drivers, I am not sure that they can stand such a similar setup > where multiple instances of this same driver are probed > > .... indeed the existent *ph refs above is already global....so it wont already > work anyway in case of multiple instances now... > > ...and if I look at how CPUFreq expects the signature of scmi_cpufreq_get_rate() > to be annd how it is implemented now using the global *ph reference, it is > clearly already not working cleanly on a multi-instance setup... > > ...now...I can imagine how to (maybe) fix the above removing the globals and > fixing this, BUT the question, more generally, is CPUFreq supposed to work at all in > this multi-probed mode of operation ? > Does it even make sense to be able to support this in CPUFREQ ? > > (as an example in cpufreq,c there is static global cpufreq_driver > pointing to the arch-specific configured driver BUT that also holds > some .driver_data AND that cleraly wont be instance specific if you > probe multiple times and register with CPUFreq multiple times...) > > More questions than answers here :D > Thanks Cristian for instant response. Yes, indeed now we have more questions :) (which is good). But that's good description of the situation. So lets consider a few option what we could do now: 1. Let Sibi add another global state the 'handle' but add a BUG_ON() or WARN_ON() in the probe path if the next 'handle' instance is different than already set in global. This would simply mean that we don't support (yet) such configuration in a platform. As you said, we already have the *ph global, so maybe such platforms with multiple instances for this particular cpufreq and performance protocol don't exist yet. 2. Ask Sibi to wait with this change, till we refactor the exiting driver such that it could support easily those multiple instances. Then pick up this patch set. Although, we would also like to have those notifications from our Juno SCP reference FW, so the feature is useful. 3. Ask Sibi to refactor his patch to somehow get the 'handle' in different way, using exiting code and not introduce this global. IHMO we could do this in steps: 1. and then 2. When we create some mock platform to test this refactoring we can start cleaning it. What do you think? Regards, Lukasz
On Thu, Feb 29, 2024 at 11:45:41AM +0000, Lukasz Luba wrote: > > > On 2/29/24 11:28, Cristian Marussi wrote: > > On Thu, Feb 29, 2024 at 10:22:39AM +0000, Lukasz Luba wrote: > > > > > > > > > On 2/29/24 09:59, Lukasz Luba wrote: > > > > > > > > > > > > On 2/28/24 17:00, Sibi Sankar wrote: > > > > > > > > > > > > > > > On 2/28/24 18:54, Lukasz Luba wrote: > > > > > > > > > > > > > > > > > > On 2/27/24 18:16, Sibi Sankar wrote: > > > > > > > Register for limit change notifications if supported and use > > > > > > > the throttled > > > > > > > frequency from the notification to apply HW pressure. > > > > > > > > > > Lukasz, > > > > > > > > > > Thanks for taking time to review the series! > > > > > > > > > > > > > > > > > > > Signed-off-by: Sibi Sankar <quic_sibis@quicinc.com> > > > > > > > --- > > > > > > > > > > > > > > v3: > > > > > > > * Sanitize range_max received from the notifier. [Pierre] > > > > > > > * Update commit message. > > > > > > > > > > > > > > � drivers/cpufreq/scmi-cpufreq.c | 29 ++++++++++++++++++++++++++++- > > > > > > > � 1 file changed, 28 insertions(+), 1 deletion(-) > > > > > > > > > > > > > > diff --git a/drivers/cpufreq/scmi-cpufreq.c > > > > > > > b/drivers/cpufreq/scmi-cpufreq.c > > > > > > > index 76a0ddbd9d24..78b87b72962d 100644 > > > > > > > --- a/drivers/cpufreq/scmi-cpufreq.c > > > > > > > +++ b/drivers/cpufreq/scmi-cpufreq.c > > > > > > > @@ -25,9 +25,13 @@ struct scmi_data { > > > > > > > ����� int domain_id; > > > > > > > ����� int nr_opp; > > > > > > > ����� struct device *cpu_dev; > > > > > > > +��� struct cpufreq_policy *policy; > > > > > > > ����� cpumask_var_t opp_shared_cpus; > > > > > > > +��� struct notifier_block limit_notify_nb; > > > > > > > � }; > > > > > > > +const struct scmi_handle *handle; > > > > > > > > I've missed this bit here. > > > > > > So for this change we actually have to ask Cristian or Sudeep > > > because I'm not sure if we have only one 'handle' instance > > > for all cpufreq devices. > > > > > > If we have different 'handle' we cannot move it to the > > > global single pointer. > > > > > > Sudeep, Cristian what do you think? > > > > I was just replying noticing this :D .... since SCMI drivers can be > > probed multiple times IF you defined multiple scmi top nodes in your DT > > containing the same protocol nodes, they receive a distinct sdev/handle/ph > > for each probe...so any attempt to globalize these wont work...BUT... > > > > ...this is a bit of a weird setup BUT it is not against the spec and it can > > be used to parallelize more the SCMI accesses to disjont set of resources > > within the same protocol (a long story here...) AND this type of setup is > > something that it is already used by some other colleagues of Sibi working > > on a different line of products (AFAIK)... > > > > So, for these reasons, usually, all the other SCMI drivers have per-instance > > non-global references to handle/sdev/ph.... > > > > ...having said that, thought, looking at the structure of CPUFReq > > drivers, I am not sure that they can stand such a similar setup > > where multiple instances of this same driver are probed > > > > .... indeed the existent *ph refs above is already global....so it wont already > > work anyway in case of multiple instances now... > > > > ...and if I look at how CPUFreq expects the signature of scmi_cpufreq_get_rate() > > to be annd how it is implemented now using the global *ph reference, it is > > clearly already not working cleanly on a multi-instance setup... > > > > ...now...I can imagine how to (maybe) fix the above removing the globals and > > fixing this, BUT the question, more generally, is CPUFreq supposed to work at all in > > this multi-probed mode of operation ? > > Does it even make sense to be able to support this in CPUFREQ ? > > > > (as an example in cpufreq,c there is static global cpufreq_driver > > pointing to the arch-specific configured driver BUT that also holds > > some .driver_data AND that cleraly wont be instance specific if you > > probe multiple times and register with CPUFreq multiple times...) > > > > More questions than answers here :D > > > > Thanks Cristian for instant response. Yes, indeed now we have more > questions :) (which is good). But that's good description of the > situation. > > So lets consider a few option what we could do now: > 1. Let Sibi add another global state the 'handle' but add > a BUG_ON() or WARN_ON() in the probe path if the next > 'handle' instance is different than already set in global. > This would simply mean that we don't support (yet) > such configuration in a platform. As you said, we > already have the *ph global, so maybe such platforms > with multiple instances for this particular cpufreq and > performance protocol don't exist yet. Yes this is the quickst way (and a WARN_ON() is better I'd say) but there are similar issues of "unicity" currently already with another vendor SCMI drivers and custom protocol currently under review, so I was thinking to add a new common mechanism in SCMI to handle this ... not thought about this really in depth and I want to chat with Sudeep about this... > 2. Ask Sibi to wait with this change, till we refactor the > exiting driver such that it could support easily those > multiple instances. Then pick up this patch set. > Although, we would also like to have those notifications from our > Juno SCP reference FW, so the feature is useful. > 3. Ask Sibi to refactor his patch to somehow get the 'handle' > in different way, using exiting code and not introduce this global. > > IHMO we could do this in steps: 1. and then 2. When > we create some mock platform to test this refactoring we can > start cleaning it. > Both of these options really beg an answer to my original previous q question...if we somehow enable this multi-probe support in the scmi-cpufreq.c driver by avoiding glbals refs, does this work at all in the context of CPUFreq ? ..or it is just that CPUFreq cannot handle such a configuration (and maybe dont want to) and so the only solution here is just 1. at first and then a common refined mechanism (as mentioned above) to ensure this "unicity" of the probes for some drivers ? I'm not familiar enough to grasp if this "multi-probed" mode of operation is allowed/supported by CPUFreq and, more important, if it makes any sense at all to be a supported mode... Thanks, Cristian
On 2/29/24 12:11, Cristian Marussi wrote: > On Thu, Feb 29, 2024 at 11:45:41AM +0000, Lukasz Luba wrote: >> >> >> On 2/29/24 11:28, Cristian Marussi wrote: >>> On Thu, Feb 29, 2024 at 10:22:39AM +0000, Lukasz Luba wrote: >>>> >>>> >>>> On 2/29/24 09:59, Lukasz Luba wrote: >>>>> >>>>> >>>>> On 2/28/24 17:00, Sibi Sankar wrote: >>>>>> >>>>>> >>>>>> On 2/28/24 18:54, Lukasz Luba wrote: >>>>>>> >>>>>>> >>>>>>> On 2/27/24 18:16, Sibi Sankar wrote: >>>>>>>> Register for limit change notifications if supported and use >>>>>>>> the throttled >>>>>>>> frequency from the notification to apply HW pressure. >>>>>> >>>>>> Lukasz, >>>>>> >>>>>> Thanks for taking time to review the series! >>>>>> >>>>>>>> >>>>>>>> Signed-off-by: Sibi Sankar <quic_sibis@quicinc.com> >>>>>>>> --- >>>>>>>> >>>>>>>> v3: >>>>>>>> * Sanitize range_max received from the notifier. [Pierre] >>>>>>>> * Update commit message. >>>>>>>> >>>>>>>> � drivers/cpufreq/scmi-cpufreq.c | 29 ++++++++++++++++++++++++++++- >>>>>>>> � 1 file changed, 28 insertions(+), 1 deletion(-) >>>>>>>> >>>>>>>> diff --git a/drivers/cpufreq/scmi-cpufreq.c >>>>>>>> b/drivers/cpufreq/scmi-cpufreq.c >>>>>>>> index 76a0ddbd9d24..78b87b72962d 100644 >>>>>>>> --- a/drivers/cpufreq/scmi-cpufreq.c >>>>>>>> +++ b/drivers/cpufreq/scmi-cpufreq.c >>>>>>>> @@ -25,9 +25,13 @@ struct scmi_data { >>>>>>>> ����� int domain_id; >>>>>>>> ����� int nr_opp; >>>>>>>> ����� struct device *cpu_dev; >>>>>>>> +��� struct cpufreq_policy *policy; >>>>>>>> ����� cpumask_var_t opp_shared_cpus; >>>>>>>> +��� struct notifier_block limit_notify_nb; >>>>>>>> � }; >>>>>>>> +const struct scmi_handle *handle; >>>>> >>>>> I've missed this bit here. >>>> >>>> So for this change we actually have to ask Cristian or Sudeep >>>> because I'm not sure if we have only one 'handle' instance >>>> for all cpufreq devices. >>>> >>>> If we have different 'handle' we cannot move it to the >>>> global single pointer. >>>> >>>> Sudeep, Cristian what do you think? >>> >>> I was just replying noticing this :D .... since SCMI drivers can be >>> probed multiple times IF you defined multiple scmi top nodes in your DT >>> containing the same protocol nodes, they receive a distinct sdev/handle/ph >>> for each probe...so any attempt to globalize these wont work...BUT... >>> >>> ...this is a bit of a weird setup BUT it is not against the spec and it can >>> be used to parallelize more the SCMI accesses to disjont set of resources >>> within the same protocol (a long story here...) AND this type of setup is >>> something that it is already used by some other colleagues of Sibi working >>> on a different line of products (AFAIK)... >>> >>> So, for these reasons, usually, all the other SCMI drivers have per-instance >>> non-global references to handle/sdev/ph.... >>> >>> ...having said that, thought, looking at the structure of CPUFReq >>> drivers, I am not sure that they can stand such a similar setup >>> where multiple instances of this same driver are probed >>> >>> .... indeed the existent *ph refs above is already global....so it wont already >>> work anyway in case of multiple instances now... >>> >>> ...and if I look at how CPUFreq expects the signature of scmi_cpufreq_get_rate() >>> to be annd how it is implemented now using the global *ph reference, it is >>> clearly already not working cleanly on a multi-instance setup... >>> >>> ...now...I can imagine how to (maybe) fix the above removing the globals and >>> fixing this, BUT the question, more generally, is CPUFreq supposed to work at all in >>> this multi-probed mode of operation ? >>> Does it even make sense to be able to support this in CPUFREQ ? >>> >>> (as an example in cpufreq,c there is static global cpufreq_driver >>> pointing to the arch-specific configured driver BUT that also holds >>> some .driver_data AND that cleraly wont be instance specific if you >>> probe multiple times and register with CPUFreq multiple times...) >>> >>> More questions than answers here :D >>> >> >> Thanks Cristian for instant response. Yes, indeed now we have more >> questions :) (which is good). But that's good description of the >> situation. >> >> So lets consider a few option what we could do now: >> 1. Let Sibi add another global state the 'handle' but add >> a BUG_ON() or WARN_ON() in the probe path if the next >> 'handle' instance is different than already set in global. >> This would simply mean that we don't support (yet) >> such configuration in a platform. As you said, we >> already have the *ph global, so maybe such platforms >> with multiple instances for this particular cpufreq and >> performance protocol don't exist yet. > > Yes this is the quickst way (and a WARN_ON() is better I'd say) but there > are similar issues of "unicity" currently already with another vendor SCMI > drivers and custom protocol currently under review, so I was thinking to > add a new common mechanism in SCMI to handle this ... not thought about > this really in depth and I want to chat with Sudeep about this... > >> 2. Ask Sibi to wait with this change, till we refactor the >> exiting driver such that it could support easily those >> multiple instances. Then pick up this patch set. >> Although, we would also like to have those notifications from our >> Juno SCP reference FW, so the feature is useful. >> 3. Ask Sibi to refactor his patch to somehow get the 'handle' >> in different way, using exiting code and not introduce this global. >> > >> IHMO we could do this in steps: 1. and then 2. When >> we create some mock platform to test this refactoring we can >> start cleaning it. >> > > Both of these options really beg an answer to my original previous q > question...if we somehow enable this multi-probe support in the > scmi-cpufreq.c driver by avoiding glbals refs, does this work at all in > the context of CPUFreq ? I don't know yet. > > ...or it is just that CPUFreq cannot handle such a configuration (and > maybe dont want to) and so the only solution here is just 1. at first and > then a common refined mechanism (as mentioned above) to ensure this "unicity" > of the probes for some drivers ? This sounds reasonable. > > I'm not familiar enough to grasp if this "multi-probed" mode of operation is > allowed/supported by CPUFreq and, more important, if it makes any sense > at all to be a supported mode... > OK, let me check some stuff in the code and think for a while on that. Thanks Cristian! Sibi, please give me a few days. In the meantime you can continue on the 'boost' patch set probably.
On 2/29/24 19:45, Lukasz Luba wrote: > > > On 2/29/24 12:11, Cristian Marussi wrote: >> On Thu, Feb 29, 2024 at 11:45:41AM +0000, Lukasz Luba wrote: >>> >>> >>> On 2/29/24 11:28, Cristian Marussi wrote: >>>> On Thu, Feb 29, 2024 at 10:22:39AM +0000, Lukasz Luba wrote: >>>>> >>>>> >>>>> On 2/29/24 09:59, Lukasz Luba wrote: >>>>>> >>>>>> >>>>>> On 2/28/24 17:00, Sibi Sankar wrote: >>>>>>> >>>>>>> >>>>>>> On 2/28/24 18:54, Lukasz Luba wrote: >>>>>>>> >>>>>>>> >>>>>>>> On 2/27/24 18:16, Sibi Sankar wrote: >>>>>>>>> Register for limit change notifications if supported and use >>>>>>>>> the throttled >>>>>>>>> frequency from the notification to apply HW pressure. >>>>>>> >>>>>>> Lukasz, >>>>>>> >>>>>>> Thanks for taking time to review the series! >>>>>>> >>>>>>>>> >>>>>>>>> Signed-off-by: Sibi Sankar <quic_sibis@quicinc.com> >>>>>>>>> --- >>>>>>>>> >>>>>>>>> v3: >>>>>>>>> * Sanitize range_max received from the notifier. [Pierre] >>>>>>>>> * Update commit message. >>>>>>>>> >>>>>>>>> � drivers/cpufreq/scmi-cpufreq.c | 29 >>>>>>>>> ++++++++++++++++++++++++++++- >>>>>>>>> � 1 file changed, 28 insertions(+), 1 deletion(-) >>>>>>>>> >>>>>>>>> diff --git a/drivers/cpufreq/scmi-cpufreq.c >>>>>>>>> b/drivers/cpufreq/scmi-cpufreq.c >>>>>>>>> index 76a0ddbd9d24..78b87b72962d 100644 >>>>>>>>> --- a/drivers/cpufreq/scmi-cpufreq.c >>>>>>>>> +++ b/drivers/cpufreq/scmi-cpufreq.c >>>>>>>>> @@ -25,9 +25,13 @@ struct scmi_data { >>>>>>>>> ����� int domain_id; >>>>>>>>> ����� int nr_opp; >>>>>>>>> ����� struct device *cpu_dev; >>>>>>>>> +��� struct cpufreq_policy *policy; >>>>>>>>> ����� cpumask_var_t opp_shared_cpus; >>>>>>>>> +��� struct notifier_block limit_notify_nb; >>>>>>>>> � }; >>>>>>>>> +const struct scmi_handle *handle; >>>>>> >>>>>> I've missed this bit here. >>>>> >>>>> So for this change we actually have to ask Cristian or Sudeep >>>>> because I'm not sure if we have only one 'handle' instance >>>>> for all cpufreq devices. >>>>> >>>>> If we have different 'handle' we cannot move it to the >>>>> global single pointer. >>>>> >>>>> Sudeep, Cristian what do you think? >>>> >>>> I was just replying noticing this :D .... since SCMI drivers can be >>>> probed multiple times IF you defined multiple scmi top nodes in your DT >>>> containing the same protocol nodes, they receive a distinct >>>> sdev/handle/ph >>>> for each probe...so any attempt to globalize these wont work...BUT... >>>> >>>> ...this is a bit of a weird setup BUT it is not against the spec and >>>> it can >>>> be used to parallelize more the SCMI accesses to disjont set of >>>> resources >>>> within the same protocol (a long story here...) AND this type of >>>> setup is >>>> something that it is already used by some other colleagues of Sibi >>>> working >>>> on a different line of products (AFAIK)... >>>> >>>> So, for these reasons, usually, all the other SCMI drivers have >>>> per-instance >>>> non-global references to handle/sdev/ph.... >>>> >>>> ...having said that, thought, looking at the structure of CPUFReq >>>> drivers, I am not sure that they can stand such a similar setup >>>> where multiple instances of this same driver are probed >>>> >>>> .... indeed the existent *ph refs above is already global....so it >>>> wont already >>>> work anyway in case of multiple instances now... >>>> >>>> ...and if I look at how CPUFreq expects the signature of >>>> scmi_cpufreq_get_rate() >>>> to be annd how it is implemented now using the global *ph reference, >>>> it is >>>> clearly already not working cleanly on a multi-instance setup... >>>> >>>> ...now...I can imagine how to (maybe) fix the above removing the >>>> globals and >>>> fixing this, BUT the question, more generally, is CPUFreq supposed >>>> to work at all in >>>> this multi-probed mode of operation ? >>>> Does it even make sense to be able to support this in CPUFREQ ? >>>> >>>> (as an example in cpufreq,c there is static global cpufreq_driver >>>> pointing to the arch-specific configured driver BUT that also holds >>>> some .driver_data AND that cleraly wont be instance specific if you >>>> probe multiple times and register with CPUFreq multiple times...) >>>> >>>> More questions than answers here :D >>>> >>> >>> Thanks Cristian for instant response. Yes, indeed now we have more >>> questions :) (which is good). But that's good description of the >>> situation. >>> >>> So lets consider a few option what we could do now: >>> 1. Let Sibi add another global state the 'handle' but add >>> a BUG_ON() or WARN_ON() in the probe path if the next >>> 'handle' instance is different than already set in global. >>> This would simply mean that we don't support (yet) >>> such configuration in a platform. As you said, we >>> already have the *ph global, so maybe such platforms >>> with multiple instances for this particular cpufreq and >>> performance protocol don't exist yet. >> >> Yes this is the quickst way (and a WARN_ON() is better I'd say) but there >> are similar issues of "unicity" currently already with another vendor >> SCMI >> drivers and custom protocol currently under review, so I was thinking to >> add a new common mechanism in SCMI to handle this ... not thought about >> this really in depth and I want to chat with Sudeep about this... >> >>> 2. Ask Sibi to wait with this change, till we refactor the >>> exiting driver such that it could support easily those >>> multiple instances. Then pick up this patch set. >>> Although, we would also like to have those notifications from our >>> Juno SCP reference FW, so the feature is useful. >>> 3. Ask Sibi to refactor his patch to somehow get the 'handle' >>> in different way, using exiting code and not introduce this global. >>> >> >>> IHMO we could do this in steps: 1. and then 2. When >>> we create some mock platform to test this refactoring we can >>> start cleaning it. I should be able to volunteer a platform to test against when we have things ready. >>> >> >> Both of these options really beg an answer to my original previous q >> question...if we somehow enable this multi-probe support in the >> scmi-cpufreq.c driver by avoiding glbals refs, does this work at all in >> the context of CPUFreq ? > > I don't know yet. > >> >> ...or it is just that CPUFreq cannot handle such a configuration (and >> maybe dont want to) and so the only solution here is just 1. at first and >> then a common refined mechanism (as mentioned above) to ensure this >> "unicity" >> of the probes for some drivers ? > > This sounds reasonable. > >> >> I'm not familiar enough to grasp if this "multi-probed" mode of >> operation is >> allowed/supported by CPUFreq and, more important, if it makes any sense >> at all to be a supported mode... >> > > OK, let me check some stuff in the code and think for a while on that. > Thanks Cristian! > > Sibi, please give me a few days. In the meantime you can continue > on the 'boost' patch set probably. sure, thanks. I've plenty things to send out so no hurry ;) -Sibi
diff --git a/drivers/cpufreq/scmi-cpufreq.c b/drivers/cpufreq/scmi-cpufreq.c index 76a0ddbd9d24..78b87b72962d 100644 --- a/drivers/cpufreq/scmi-cpufreq.c +++ b/drivers/cpufreq/scmi-cpufreq.c @@ -25,9 +25,13 @@ struct scmi_data { int domain_id; int nr_opp; struct device *cpu_dev; + struct cpufreq_policy *policy; cpumask_var_t opp_shared_cpus; + struct notifier_block limit_notify_nb; }; +const struct scmi_handle *handle; +static struct scmi_device *scmi_dev; static struct scmi_protocol_handle *ph; static const struct scmi_perf_proto_ops *perf_ops; static struct cpufreq_driver scmi_cpufreq_driver; @@ -151,6 +155,20 @@ static struct freq_attr *scmi_cpufreq_hw_attr[] = { NULL, }; +static int scmi_limit_notify_cb(struct notifier_block *nb, unsigned long event, void *data) +{ + struct scmi_data *priv = container_of(nb, struct scmi_data, limit_notify_nb); + struct scmi_perf_limits_report *limit_notify = data; + struct cpufreq_policy *policy = priv->policy; + + policy->max = clamp(limit_notify->range_max_freq/HZ_PER_KHZ, policy->cpuinfo.min_freq, + policy->cpuinfo.max_freq); + + cpufreq_update_pressure(policy); + + return NOTIFY_OK; +} + static int scmi_cpufreq_init(struct cpufreq_policy *policy) { int ret, nr_opp, domain; @@ -269,6 +287,15 @@ static int scmi_cpufreq_init(struct cpufreq_policy *policy) } } + priv->limit_notify_nb.notifier_call = scmi_limit_notify_cb; + ret = handle->notify_ops->devm_event_notifier_register(scmi_dev, SCMI_PROTOCOL_PERF, + SCMI_EVENT_PERFORMANCE_LIMITS_CHANGED, + &domain, + &priv->limit_notify_nb); + if (ret) + dev_warn(cpu_dev, + "failed to register for limits change notifier for domain %d\n", domain); + priv->policy = policy; return 0; @@ -342,8 +369,8 @@ static int scmi_cpufreq_probe(struct scmi_device *sdev) { int ret; struct device *dev = &sdev->dev; - const struct scmi_handle *handle; + scmi_dev = sdev; handle = sdev->handle; if (!handle)