[V4,4/5] soc: qcom: Introduce SCMI based Memlat (Memory Latency) governor

Introduce a client driver that uses the memlat algorithm string
hosted on QCOM SCMI Generic Extension Protocol to detect memory
latency workloads and control frequency/level of the various
memory buses (DDR/LLCC/DDR_QOS).

Co-developed-by: Shivnandan Kumar <quic_kshivnan@quicinc.com>
Signed-off-by: Shivnandan Kumar <quic_kshivnan@quicinc.com>
Co-developed-by: Ramakrishna Gottimukkula <quic_rgottimu@quicinc.com>
Signed-off-by: Ramakrishna Gottimukkula <quic_rgottimu@quicinc.com>
Co-developed-by: Amir Vajid <avajid@quicinc.com>
Signed-off-by: Amir Vajid <avajid@quicinc.com>
Signed-off-by: Sibi Sankar <quic_sibis@quicinc.com>
---

v3:
* Add missing enum in the scmi memlat driver and fix documentation [Konrad]
* Add checks for max memory and monitor [Shivnandan]
* Fix typo from START_TIMER -> STOP_TIMER [Shivnandan]
* Make populate_physical_mask func to void [Shivnandan]
* Remove unecessary zero set [Shivnandan]
* Use __free(device node) in init_cpufreq-memfreqmap [Christian/Konrad]
* Use sdev->dev.of_node directly [Christian]
* use return dev_err_probe in multiple places [Christian]

 drivers/soc/qcom/Kconfig                   |  12 +
 drivers/soc/qcom/Makefile                  |   1 +
 drivers/soc/qcom/qcom_scmi_memlat_client.c | 569 +++++++++++++++++++++
 3 files changed, 582 insertions(+)
 create mode 100644 drivers/soc/qcom/qcom_scmi_memlat_client.c

On Mon, Oct 07, 2024 at 11:40:22AM GMT, Sibi Sankar wrote:
> Introduce a client driver that uses the memlat algorithm string
> hosted on QCOM SCMI Generic Extension Protocol to detect memory
> latency workloads and control frequency/level of the various
> memory buses (DDR/LLCC/DDR_QOS).

This sounds like a devfreq implementation. Please provide a reason why
it doesn't use existing API (even if to export the information to the
userspace).

> 
> Co-developed-by: Shivnandan Kumar <quic_kshivnan@quicinc.com>
> Signed-off-by: Shivnandan Kumar <quic_kshivnan@quicinc.com>
> Co-developed-by: Ramakrishna Gottimukkula <quic_rgottimu@quicinc.com>
> Signed-off-by: Ramakrishna Gottimukkula <quic_rgottimu@quicinc.com>
> Co-developed-by: Amir Vajid <avajid@quicinc.com>
> Signed-off-by: Amir Vajid <avajid@quicinc.com>
> Signed-off-by: Sibi Sankar <quic_sibis@quicinc.com>
> ---
> 
> v3:
> * Add missing enum in the scmi memlat driver and fix documentation [Konrad]
> * Add checks for max memory and monitor [Shivnandan]
> * Fix typo from START_TIMER -> STOP_TIMER [Shivnandan]
> * Make populate_physical_mask func to void [Shivnandan]
> * Remove unecessary zero set [Shivnandan]
> * Use __free(device node) in init_cpufreq-memfreqmap [Christian/Konrad]
> * Use sdev->dev.of_node directly [Christian]
> * use return dev_err_probe in multiple places [Christian]
> 
>  drivers/soc/qcom/Kconfig                   |  12 +
>  drivers/soc/qcom/Makefile                  |   1 +
>  drivers/soc/qcom/qcom_scmi_memlat_client.c | 569 +++++++++++++++++++++
>  3 files changed, 582 insertions(+)
>  create mode 100644 drivers/soc/qcom/qcom_scmi_memlat_client.c
> 
> diff --git a/drivers/soc/qcom/Kconfig b/drivers/soc/qcom/Kconfig
> index 74b9121240f8..1b6dd40d69ea 100644
> --- a/drivers/soc/qcom/Kconfig
> +++ b/drivers/soc/qcom/Kconfig
> @@ -295,4 +295,16 @@ config QCOM_PBS
>  	  This module provides the APIs to the client drivers that wants to send the
>  	  PBS trigger event to the PBS RAM.
>  
> +config QCOM_SCMI_MEMLAT_CLIENT
> +	tristate "Qualcomm Technologies Inc. SCMI client driver"
> +	depends on QCOM_SCMI_GENERIC_EXT || COMPILE_TEST
> +	help
> +	  This driver uses the MEMLAT (memory latency) algorithm string
> +	  hosted on QCOM SCMI Vendor Protocol to detect memory latency

How can it use the string to detect workloads? Most likely you mean something like "uses memlat extensions".
Also s/QCOM/Qualcomm/ in the help text.

> +	  workloads and control frequency/level of the various memory
> +	  buses (DDR/LLCC/DDR_QOS).
> +
> +	  This driver defines/documents the parameter IDs used while configuring
> +	  the memory buses.
> +
>  endmenu
> diff --git a/drivers/soc/qcom/Makefile b/drivers/soc/qcom/Makefile
> index acbca2ab5cc2..28549bb141bc 100644
> --- a/drivers/soc/qcom/Makefile
> +++ b/drivers/soc/qcom/Makefile
> @@ -36,6 +36,7 @@ obj-$(CONFIG_QCOM_APR) += apr.o
>  obj-$(CONFIG_QCOM_LLCC) += llcc-qcom.o
>  obj-$(CONFIG_QCOM_KRYO_L2_ACCESSORS) +=	kryo-l2-accessors.o
>  obj-$(CONFIG_QCOM_ICC_BWMON)	+= icc-bwmon.o
> +obj-$(CONFIG_QCOM_SCMI_MEMLAT_CLIENT)	+= qcom_scmi_memlat_client.o
>  qcom_ice-objs			+= ice.o
>  obj-$(CONFIG_QCOM_INLINE_CRYPTO_ENGINE)	+= qcom_ice.o
>  obj-$(CONFIG_QCOM_PBS) +=	qcom-pbs.o
> diff --git a/drivers/soc/qcom/qcom_scmi_memlat_client.c b/drivers/soc/qcom/qcom_scmi_memlat_client.c
> new file mode 100644
> index 000000000000..05198bf1f7ec
> --- /dev/null
> +++ b/drivers/soc/qcom/qcom_scmi_memlat_client.c
> @@ -0,0 +1,569 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * Copyright (c) 2024, Qualcomm Innovation Center, Inc. All rights reserved.
> + */
> +
> +#include <linux/cpu.h>
> +#include <linux/err.h>
> +#include <linux/errno.h>
> +#include <linux/init.h>
> +#include <linux/kernel.h>
> +#include <linux/module.h>
> +#include <linux/of.h>
> +#include <linux/platform_device.h>
> +#include <linux/scmi_protocol.h>
> +#include <linux/scmi_qcom_protocol.h>
> +#include <linux/units.h>
> +#include <dt-bindings/firmware/qcom,scmi-memlat.h>
> +
> +#define MEMLAT_ALGO_STR				0x4D454D4C4154 /* MEMLAT */
> +#define INVALID_IDX				0xff
> +#define MAX_MEMORY_TYPES			3
> +#define MAX_MONITOR_CNT				4
> +#define MAX_NAME_LEN				20
> +#define MAX_MAP_ENTRIES				7
> +#define CPUCP_DEFAULT_SAMPLING_PERIOD_MS	4
> +#define CPUCP_DEFAULT_FREQ_METHOD		1
> +
> +/**
> + * enum scmi_memlat_protocol_cmd - parameter_ids supported by the "MEMLAT" algo_str hosted
> + *                                 by the Qualcomm Generic Vendor Protocol on the SCMI controller.
> + *
> + * MEMLAT (Memory Latency) monitors the counters to detect memory latency bound workloads
> + * and scales the frequency/levels of the memory buses accordingly.
> + *
> + * @MEMLAT_SET_MEM_GROUP: initializes the frequency/level scaling functions for the memory bus.
> + * @MEMLAT_SET_MONITOR: configures the monitor to work on a specific memory bus.
> + * @MEMLAT_SET_COMMON_EV_MAP: set up common counters used to monitor the cpu frequency.
> + * @MEMLAT_SET_GRP_EV_MAP: set up any specific counters used to monitor the memory bus.
> + * @MEMLAT_IPM_CEIL: set the IPM (Instruction Per Misses) ceiling per monitor.
> + * @MEMLAT_SAMPLE_MS: set the sampling period for all the monitors.
> + * @MEMLAT_MON_FREQ_MAP: setup the cpufreq to memfreq map.
> + * @MEMLAT_SET_MIN_FREQ: set the max frequency of the memory bus.
> + * @MEMLAT_SET_MAX_FREQ: set the min frequency of the memory bus.
> + * @MEMLAT_START_TIMER: start all the monitors with the requested sampling period.
> + * @MEMLAT_STOP_TIMER: stop all the running monitors.
> + * @MEMLAT_SET_EFFECTIVE_FREQ_METHOD: set the method used to determine cpu frequency.
> + */
> +enum scmi_memlat_protocol_cmd {
> +	MEMLAT_SET_MEM_GROUP = 16,
> +	MEMLAT_SET_MONITOR,
> +	MEMLAT_SET_COMMON_EV_MAP,
> +	MEMLAT_SET_GRP_EV_MAP,
> +	MEMLAT_IPM_CEIL = 23,
> +	MEMLAT_SAMPLE_MS = 31,
> +	MEMLAT_MON_FREQ_MAP,
> +	MEMLAT_SET_MIN_FREQ,
> +	MEMLAT_SET_MAX_FREQ,
> +	MEMLAT_START_TIMER = 36,
> +	MEMLAT_STOP_TIMER,
> +	MEMLAT_SET_EFFECTIVE_FREQ_METHOD = 39,
> +};
> +
> +struct map_table {
> +	u16 v1;
> +	u16 v2;

Huh? Why can't it be cpufreq and memfreq with some suffix?

> +};
> +
> +struct map_param_msg {
> +	u32 hw_type;
> +	u32 mon_idx;
> +	u32 nr_rows;
> +	struct map_table tbl[MAX_MAP_ENTRIES];
> +} __packed;
> +
> +struct node_msg {
> +	u32 cpumask;
> +	u32 hw_type;
> +	u32 mon_type;
> +	u32 mon_idx;
> +	char mon_name[MAX_NAME_LEN];
> +};
> +
> +struct scalar_param_msg {
> +	u32 hw_type;
> +	u32 mon_idx;
> +	u32 val;
> +};
> +
> +enum common_ev_idx {
> +	INST_IDX,
> +	CYC_IDX,
> +	CONST_CYC_IDX,
> +	FE_STALL_IDX,
> +	BE_STALL_IDX,
> +	NUM_COMMON_EVS
> +};
> +
> +enum grp_ev_idx {
> +	MISS_IDX,
> +	WB_IDX,
> +	ACC_IDX,
> +	NUM_GRP_EVS
> +};
> +
> +#define EV_CPU_CYCLES		0
> +#define EV_INST_RETIRED		2
> +#define EV_L2_D_RFILL		5
> +
> +struct ev_map_msg {
> +	u32 num_evs;
> +	u32 hw_type;
> +	u32 cid[NUM_COMMON_EVS];
> +};
> +
> +struct cpufreq_memfreq_map {
> +	unsigned int cpufreq_mhz;
> +	unsigned int memfreq_khz;
> +};
> +
> +struct scmi_monitor_info {
> +	struct cpufreq_memfreq_map *freq_map;
> +	char mon_name[MAX_NAME_LEN];
> +	u32 mon_idx;
> +	u32 mon_type;
> +	u32 ipm_ceil;
> +	u32 mask;
> +	u32 freq_map_len;
> +};
> +
> +struct scmi_memory_info {
> +	struct scmi_monitor_info *monitor[MAX_MONITOR_CNT];
> +	u32 hw_type;
> +	int monitor_cnt;
> +	u32 min_freq;
> +	u32 max_freq;
> +};
> +
> +struct scmi_memlat_info {
> +	struct scmi_protocol_handle *ph;
> +	const struct qcom_generic_ext_ops *ops;
> +	struct scmi_memory_info *memory[MAX_MEMORY_TYPES];
> +	u32 cluster_info[NR_CPUS];
> +	int memory_cnt;
> +};
> +
> +static int populate_cluster_info(u32 *cluster_info)
> +{
> +	char name[MAX_NAME_LEN];
> +	int i = 0;
> +
> +	struct device_node *cn __free(device_node) = of_find_node_by_path("/cpus");
> +	if (!cn)
> +		return -ENODEV;
> +
> +	struct device_node *map __free(device_node) = of_get_child_by_name(cn, "cpu-map");
> +	if (!map)
> +		return -ENODEV;
> +
> +	do {
> +		snprintf(name, sizeof(name), "cluster%d", i);
> +		struct device_node *c __free(device_node) = of_get_child_by_name(map, name);
> +		if (!c)
> +			break;
> +
> +		*(cluster_info + i) = of_get_child_count(c);
> +		i++;
> +	} while (1);

Can you use existing API from drivers/base/arch_topology.c? If not, can
it be extended to support your usecase?

> +
> +	return 0;
> +}
> +
> +static void populate_physical_mask(struct device_node *np, u32 *mask, u32 *cluster_info)
> +{
> +	struct device_node *dev_phandle __free(device_node);
> +	int cpu, i = 0, physical_id;
> +
> +	do {
> +		dev_phandle = of_parse_phandle(np, "cpus", i++);
> +		cpu = of_cpu_node_to_id(dev_phandle);
> +		if (cpu != -ENODEV) {
> +			physical_id = topology_core_id(cpu);
> +			for (int j = 0; j < topology_cluster_id(cpu); j++)
> +				physical_id += *(cluster_info + j);
> +			*mask |= BIT(physical_id);
> +		}
> +	} while (dev_phandle);
> +}
> +
> +static struct cpufreq_memfreq_map *init_cpufreq_memfreq_map(struct device *dev,
> +							    struct scmi_memory_info *memory,
> +							    struct device_node *of_node,
> +							    u32 *cnt)
> +{
> +	struct device_node *tbl_np __free(device_node), *opp_np __free(device_node);
> +	struct cpufreq_memfreq_map *tbl;
> +	int ret, i = 0;
> +	u32 level, len;
> +	u64 rate;
> +
> +	tbl_np = of_parse_phandle(of_node, "operating-points-v2", 0);

Please use existing API to parse OPP tables or document a reason why it
can't be used.

> +	if (!tbl_np)
> +		return ERR_PTR(-ENODEV);
> +
> +	len = min(of_get_available_child_count(tbl_np), MAX_MAP_ENTRIES);
> +	if (len == 0)
> +		return ERR_PTR(-ENODEV);
> +
> +	tbl = devm_kzalloc(dev, (len + 1) * sizeof(struct cpufreq_memfreq_map),
> +			   GFP_KERNEL);
> +	if (!tbl)
> +		return ERR_PTR(-ENOMEM);
> +
> +	for_each_available_child_of_node(tbl_np, opp_np) {
> +		ret = of_property_read_u64_index(opp_np, "opp-hz", 0, &rate);
> +		if (ret < 0)
> +			return ERR_PTR(ret);
> +
> +		tbl[i].cpufreq_mhz = rate / HZ_PER_MHZ;
> +
> +		if (memory->hw_type != QCOM_MEM_TYPE_DDR_QOS) {
> +			ret = of_property_read_u64_index(opp_np, "opp-hz", 1, &rate);
> +			if (ret < 0)
> +				return ERR_PTR(ret);
> +
> +			tbl[i].memfreq_khz = rate / HZ_PER_KHZ;
> +		} else {
> +			ret = of_property_read_u32(opp_np, "opp-level", &level);
> +			if (ret < 0)
> +				return ERR_PTR(ret);
> +
> +			tbl[i].memfreq_khz = level;
> +		}
> +
> +		dev_dbg(dev, "Entry%d CPU:%u, Mem:%u\n", i, tbl[i].cpufreq_mhz, tbl[i].memfreq_khz);
> +		i++;
> +	}
> +	*cnt = len;
> +
> +	return tbl;
> +}
> +
> +static int process_scmi_memlat_of_node(struct scmi_device *sdev, struct scmi_memlat_info *info)
> +{
> +	struct scmi_monitor_info *monitor;
> +	struct scmi_memory_info *memory;
> +	char name[MAX_NAME_LEN];
> +	u64 memfreq[2];
> +	int ret;
> +
> +	ret = populate_cluster_info(info->cluster_info);
> +	if (ret < 0) {
> +		dev_err_probe(&sdev->dev, ret, "failed to populate cluster info\n");
> +		goto err;
> +	}
> +
> +	of_node_get(sdev->dev.of_node);
> +	do {
> +		snprintf(name, sizeof(name), "memory-%d", info->memory_cnt);
> +		struct device_node *memory_np __free(device_node) =
> +			of_find_node_by_name(sdev->dev.of_node, name);
> +
> +		if (!memory_np)
> +			break;
> +
> +		if (info->memory_cnt >= MAX_MEMORY_TYPES)
> +			return dev_err_probe(&sdev->dev, -EINVAL,
> +					     "failed to parse unsupported memory type\n");
> +
> +		memory = devm_kzalloc(&sdev->dev, sizeof(*memory), GFP_KERNEL);
> +		if (!memory) {
> +			ret = -ENOMEM;
> +			goto err;
> +		}
> +
> +		ret = of_property_read_u32(memory_np, "qcom,memory-type", &memory->hw_type);
> +		if (ret) {
> +			dev_err_probe(&sdev->dev, ret, "failed to read memory type\n");
> +			goto err;
> +		}
> +
> +		ret = of_property_read_u64_array(memory_np, "freq-table-hz", memfreq, 2);
> +		if (ret && (ret != -EINVAL)) {
> +			dev_err_probe(&sdev->dev, ret, "failed to read min/max freq\n");
> +			goto err;
> +		}

Can we get this information from the OPP table instead?

> +
> +		if (memory->hw_type != QCOM_MEM_TYPE_DDR_QOS) {
> +			memory->min_freq = memfreq[0] / HZ_PER_KHZ;
> +			memory->max_freq = memfreq[1] / HZ_PER_KHZ;
> +		} else {
> +			memory->min_freq = memfreq[0];
> +			memory->max_freq = memfreq[1];

Why? At least invert the logic here, please. The DDR_QOS is a special
case, not all other kinds of memory.

> +		}
> +		info->memory[info->memory_cnt++] = memory;
> +
> +		do {
> +			snprintf(name, sizeof(name), "monitor-%d", memory->monitor_cnt);
> +			struct device_node *monitor_np __free(device_node) =
> +				of_get_child_by_name(memory_np, name);
> +
> +			if (!monitor_np)
> +				break;
> +
> +			if (memory->monitor_cnt >= MAX_MONITOR_CNT)

Why do you need to limit it? Is it a protocol limitation or an
artificial driver limitation? Can monitors be allocated dynamically?

> +				return dev_err_probe(&sdev->dev, -EINVAL,
> +						     "failed to parse unsupported monitor\n");
> +
> +			monitor = devm_kzalloc(&sdev->dev, sizeof(*monitor), GFP_KERNEL);
> +			if (!monitor) {
> +				ret = -ENOMEM;
> +				goto err;
> +			}
> +
> +			monitor->mon_type = of_property_read_bool(monitor_np, "qcom,compute-type");
> +			if (!monitor->mon_type) {
> +				ret = of_property_read_u32(monitor_np, "qcom,ipm-ceil",
> +							   &monitor->ipm_ceil);
> +				if (ret) {
> +					dev_err_probe(&sdev->dev, ret,
> +						      "failed to read IPM ceiling\n");
> +					goto err;
> +				}
> +			}
> +
> +			/*
> +			 * Variants of the SoC having reduced number of cpus operate
> +			 * with the same number of logical cpus but the physical
> +			 * cpu disabled will differ between parts. Calculate the
> +			 * physical cpu number using cluster information instead.
> +			 */
> +			populate_physical_mask(monitor_np, &monitor->mask, info->cluster_info);
> +
> +			monitor->freq_map = init_cpufreq_memfreq_map(&sdev->dev, memory, monitor_np,
> +								     &monitor->freq_map_len);
> +			if (IS_ERR(monitor->freq_map)) {
> +				dev_err_probe(&sdev->dev, PTR_ERR(monitor->freq_map),
> +					      "failed to populate cpufreq-memfreq map\n");
> +				goto err;
> +			}
> +
> +			strscpy(monitor->mon_name, name, sizeof(monitor->mon_name));
> +			monitor->mon_idx = memory->monitor_cnt;
> +
> +			memory->monitor[memory->monitor_cnt++] = monitor;
> +		} while (1);
> +
> +		if (!memory->monitor_cnt) {
> +			ret = -EINVAL;
> +			dev_err_probe(&sdev->dev, ret, "failed to find monitor nodes\n");
> +			goto err;
> +		}
> +	} while (1);
> +
> +	if (!info->memory_cnt) {
> +		ret = -EINVAL;
> +		dev_err_probe(&sdev->dev, ret, "failed to find memory nodes\n");
> +	}
> +
> +err:
> +	of_node_put(sdev->dev.of_node);
> +
> +	return ret;
> +}
> +
> +static int configure_cpucp_common_events(struct scmi_memlat_info *info)
> +{
> +	const struct qcom_generic_ext_ops *ops = info->ops;
> +	u8 ev_map[NUM_COMMON_EVS];
> +	struct ev_map_msg msg;
> +
> +	memset(ev_map, 0xFF, NUM_COMMON_EVS);
> +
> +	msg.num_evs = NUM_COMMON_EVS;
> +	msg.hw_type = INVALID_IDX;
> +	msg.cid[INST_IDX] = EV_INST_RETIRED;
> +	msg.cid[CYC_IDX] = EV_CPU_CYCLES;
> +	msg.cid[CONST_CYC_IDX] = INVALID_IDX;
> +	msg.cid[FE_STALL_IDX] = INVALID_IDX;
> +	msg.cid[BE_STALL_IDX] = INVALID_IDX;
> +
> +	return ops->set_param(info->ph, &msg, sizeof(msg), MEMLAT_ALGO_STR,
> +			      MEMLAT_SET_COMMON_EV_MAP);
> +}
> +
> +static int configure_cpucp_grp(struct device *dev, struct scmi_memlat_info *info, int memory_index)
> +{
> +	const struct qcom_generic_ext_ops *ops = info->ops;
> +	struct scmi_memory_info *memory = info->memory[memory_index];
> +	struct ev_map_msg ev_msg;
> +	u8 ev_map[NUM_GRP_EVS];
> +	struct node_msg msg;
> +	int ret;
> +
> +	msg.cpumask = 0;
> +	msg.hw_type = memory->hw_type;
> +	msg.mon_type = 0;
> +	msg.mon_idx = 0;
> +	ret = ops->set_param(info->ph, &msg, sizeof(msg), MEMLAT_ALGO_STR, MEMLAT_SET_MEM_GROUP);
> +	if (ret < 0)
> +		return dev_err_probe(dev, ret, "failed to configure mem type %d\n",
> +				     memory->hw_type);
> +
> +	memset(ev_map, 0xFF, NUM_GRP_EVS);
> +	ev_msg.num_evs = NUM_GRP_EVS;
> +	ev_msg.hw_type = memory->hw_type;
> +	ev_msg.cid[MISS_IDX] = EV_L2_D_RFILL;
> +	ev_msg.cid[WB_IDX] = INVALID_IDX;
> +	ev_msg.cid[ACC_IDX] = INVALID_IDX;
> +	ret = ops->set_param(info->ph, &ev_msg, sizeof(ev_msg), MEMLAT_ALGO_STR,
> +			     MEMLAT_SET_GRP_EV_MAP);
> +	if (ret < 0)
> +		return dev_err_probe(dev, ret, "failed to configure event map for mem type %d\n",
> +				     memory->hw_type);
> +
> +	return ret;
> +}
> +
> +static int configure_cpucp_mon(struct device *dev, struct scmi_memlat_info *info,
> +			       int memory_index, int monitor_index)
> +{
> +	const struct qcom_generic_ext_ops *ops = info->ops;
> +	struct scmi_memory_info *memory = info->memory[memory_index];
> +	struct scmi_monitor_info *monitor = memory->monitor[monitor_index];
> +	struct scalar_param_msg scalar_msg;
> +	struct map_param_msg map_msg;
> +	struct node_msg msg;
> +	int ret;
> +	int i;
> +
> +	msg.cpumask = monitor->mask;
> +	msg.hw_type = memory->hw_type;
> +	msg.mon_type = monitor->mon_type;
> +	msg.mon_idx = monitor->mon_idx;
> +	strscpy(msg.mon_name, monitor->mon_name, sizeof(msg.mon_name));
> +	ret = ops->set_param(info->ph, &msg, sizeof(msg), MEMLAT_ALGO_STR, MEMLAT_SET_MONITOR);
> +	if (ret < 0)
> +		return dev_err_probe(dev, ret, "failed to configure monitor %s\n",
> +				     monitor->mon_name);
> +
> +	scalar_msg.hw_type = memory->hw_type;
> +	scalar_msg.mon_idx = monitor->mon_idx;
> +	scalar_msg.val = monitor->ipm_ceil;
> +	ret = ops->set_param(info->ph, &scalar_msg, sizeof(scalar_msg), MEMLAT_ALGO_STR,
> +			     MEMLAT_IPM_CEIL);
> +	if (ret < 0)
> +		return dev_err_probe(dev, ret, "failed to set ipm ceil for %s\n",
> +				     monitor->mon_name);
> +
> +	map_msg.hw_type = memory->hw_type;
> +	map_msg.mon_idx = monitor->mon_idx;
> +	map_msg.nr_rows = monitor->freq_map_len;
> +	for (i = 0; i < monitor->freq_map_len; i++) {
> +		map_msg.tbl[i].v1 = monitor->freq_map[i].cpufreq_mhz;
> +		map_msg.tbl[i].v2 = monitor->freq_map[i].memfreq_khz;
> +	}
> +	ret = ops->set_param(info->ph, &map_msg, sizeof(map_msg), MEMLAT_ALGO_STR,
> +			     MEMLAT_MON_FREQ_MAP);
> +	if (ret < 0)
> +		return dev_err_probe(dev, ret, "failed to configure freq_map for %s\n",
> +				     monitor->mon_name);
> +
> +	scalar_msg.hw_type = memory->hw_type;
> +	scalar_msg.mon_idx = monitor->mon_idx;
> +	scalar_msg.val = memory->min_freq;
> +	ret = ops->set_param(info->ph, &scalar_msg, sizeof(scalar_msg), MEMLAT_ALGO_STR,
> +			     MEMLAT_SET_MIN_FREQ);
> +	if (ret < 0)
> +		return dev_err_probe(dev, ret, "failed to set min_freq for %s\n",
> +				     monitor->mon_name);
> +
> +	scalar_msg.hw_type = memory->hw_type;
> +	scalar_msg.mon_idx = monitor->mon_idx;
> +	scalar_msg.val = memory->max_freq;
> +	ret = ops->set_param(info->ph, &scalar_msg, sizeof(scalar_msg), MEMLAT_ALGO_STR,
> +			     MEMLAT_SET_MAX_FREQ);
> +	if (ret < 0)
> +		dev_err_probe(dev, ret, "failed to set max_freq for %s\n", monitor->mon_name);
> +
> +	return ret;
> +}
> +
> +static int cpucp_memlat_init(struct scmi_device *sdev)
> +{
> +	const struct scmi_handle *handle = sdev->handle;
> +	const struct qcom_generic_ext_ops *ops;
> +	struct scmi_protocol_handle *ph;
> +	struct scmi_memlat_info *info;
> +	u32 cpucp_freq_method = CPUCP_DEFAULT_FREQ_METHOD;
> +	u32 cpucp_sample_ms = CPUCP_DEFAULT_SAMPLING_PERIOD_MS;
> +	int ret, i, j;
> +
> +	if (!handle)
> +		return -ENODEV;
> +
> +	ops = handle->devm_protocol_get(sdev, SCMI_PROTOCOL_QCOM_GENERIC, &ph);
> +	if (IS_ERR(ops))
> +		return PTR_ERR(ops);
> +
> +	info = devm_kzalloc(&sdev->dev, sizeof(*info), GFP_KERNEL);
> +	if (!info)
> +		return -ENOMEM;
> +
> +	ret = process_scmi_memlat_of_node(sdev, info);
> +	if (ret)
> +		return ret;
> +
> +	info->ph = ph;
> +	info->ops = ops;
> +
> +	/* Configure common events ids */
> +	ret = configure_cpucp_common_events(info);
> +	if (ret < 0)
> +		return dev_err_probe(&sdev->dev, ret, "failed to configure common events\n");
> +
> +	for (i = 0; i < info->memory_cnt; i++) {
> +		/* Configure per group parameters */
> +		ret = configure_cpucp_grp(&sdev->dev, info, i);
> +		if (ret < 0)
> +			return ret;
> +
> +		for (j = 0; j < info->memory[i]->monitor_cnt; j++) {
> +			/* Configure per monitor parameters */
> +			ret = configure_cpucp_mon(&sdev->dev, info, i, j);
> +			if (ret < 0)
> +				return ret;
> +		}
> +	}
> +
> +	/* Set loop sampling time */
> +	ret = ops->set_param(ph, &cpucp_sample_ms, sizeof(cpucp_sample_ms), MEMLAT_ALGO_STR,
> +			     MEMLAT_SAMPLE_MS);
> +	if (ret < 0)
> +		return dev_err_probe(&sdev->dev, ret, "failed to set sample_ms\n");
> +
> +	/* Set the effective cpu frequency calculation method */
> +	ret = ops->set_param(ph, &cpucp_freq_method, sizeof(cpucp_freq_method), MEMLAT_ALGO_STR,
> +			     MEMLAT_SET_EFFECTIVE_FREQ_METHOD);
> +	if (ret < 0)
> +		return dev_err_probe(&sdev->dev, ret,
> +				     "failed to set effective frequency calc method\n");
> +
> +	/* Start sampling and voting timer */
> +	ret = ops->start_activity(ph, NULL, 0, MEMLAT_ALGO_STR, MEMLAT_START_TIMER);
> +	if (ret < 0)
> +		dev_err_probe(&sdev->dev, ret, "failed to start memory group timer\n");
> +
> +	return ret;
> +}
> +
> +static int scmi_client_probe(struct scmi_device *sdev)
> +{
> +	return cpucp_memlat_init(sdev);

Inline it here, please.

> +}
> +
> +static const struct scmi_device_id scmi_id_table[] = {
> +	{ SCMI_PROTOCOL_QCOM_GENERIC, "qcom-generic-ext" },
> +	{ },
> +};
> +MODULE_DEVICE_TABLE(scmi, scmi_id_table);
> +
> +static struct scmi_driver qcom_scmi_client_drv = {
> +	.name		= "scmi-qcom-generic-ext-memlat",
> +	.probe		= scmi_client_probe,
> +	.id_table	= scmi_id_table,
> +};
> +module_scmi_driver(qcom_scmi_client_drv);
> +
> +MODULE_DESCRIPTION("QTI SCMI client driver");
> +MODULE_LICENSE("GPL");
> -- 
> 2.34.1
>

On 10/7/24 23:27, Dmitry Baryshkov wrote:
> On Mon, Oct 07, 2024 at 11:40:22AM GMT, Sibi Sankar wrote:
>> Introduce a client driver that uses the memlat algorithm string
>> hosted on QCOM SCMI Generic Extension Protocol to detect memory
>> latency workloads and control frequency/level of the various
>> memory buses (DDR/LLCC/DDR_QOS).
> 
> This sounds like a devfreq implementation. Please provide a reason why
> it doesn't use existing API (even if to export the information to the
> userspace).

IIRC, you asked the same question when the RFC version of it
was posted and I replied to it back then.

https://lore.kernel.org/lkml/0adaa065-3883-ebfe-8259-05ebdbd821eb@quicinc.com/

The series does not yet export any information to userspace yet
and when it does get added in the future, it would have no way
of populating governor node with the current way devfreq framework
is structured. Since this series is all about just enabling basic
support, I guess what you ask can be accomodated when we do start
exporting this info to userspace.

> 
>>
>> Co-developed-by: Shivnandan Kumar <quic_kshivnan@quicinc.com>
>> Signed-off-by: Shivnandan Kumar <quic_kshivnan@quicinc.com>
>> Co-developed-by: Ramakrishna Gottimukkula <quic_rgottimu@quicinc.com>
>> Signed-off-by: Ramakrishna Gottimukkula <quic_rgottimu@quicinc.com>
>> Co-developed-by: Amir Vajid <avajid@quicinc.com>
>> Signed-off-by: Amir Vajid <avajid@quicinc.com>
>> Signed-off-by: Sibi Sankar <quic_sibis@quicinc.com>
>> ---
>>
>> v3:
>> * Add missing enum in the scmi memlat driver and fix documentation [Konrad]
>> * Add checks for max memory and monitor [Shivnandan]
>> * Fix typo from START_TIMER -> STOP_TIMER [Shivnandan]
>> * Make populate_physical_mask func to void [Shivnandan]
>> * Remove unecessary zero set [Shivnandan]
>> * Use __free(device node) in init_cpufreq-memfreqmap [Christian/Konrad]
>> * Use sdev->dev.of_node directly [Christian]
>> * use return dev_err_probe in multiple places [Christian]
>>
>>   drivers/soc/qcom/Kconfig                   |  12 +
>>   drivers/soc/qcom/Makefile                  |   1 +
>>   drivers/soc/qcom/qcom_scmi_memlat_client.c | 569 +++++++++++++++++++++
>>   3 files changed, 582 insertions(+)
>>   create mode 100644 drivers/soc/qcom/qcom_scmi_memlat_client.c
>>
>> diff --git a/drivers/soc/qcom/Kconfig b/drivers/soc/qcom/Kconfig
>> index 74b9121240f8..1b6dd40d69ea 100644
>> --- a/drivers/soc/qcom/Kconfig
>> +++ b/drivers/soc/qcom/Kconfig
>> @@ -295,4 +295,16 @@ config QCOM_PBS
>>   	  This module provides the APIs to the client drivers that wants to send the
>>   	  PBS trigger event to the PBS RAM.
>>   
>> +config QCOM_SCMI_MEMLAT_CLIENT
>> +	tristate "Qualcomm Technologies Inc. SCMI client driver"
>> +	depends on QCOM_SCMI_GENERIC_EXT || COMPILE_TEST
>> +	help
>> +	  This driver uses the MEMLAT (memory latency) algorithm string
>> +	  hosted on QCOM SCMI Vendor Protocol to detect memory latency
> 
> How can it use the string to detect workloads? Most likely you mean something like "uses memlat extensions".
> Also s/QCOM/Qualcomm/ in the help text.

The generic vendor protocol extension works by associating algorithms to
strings. But like you said it can be re-worded to avoid confusion.

> 
>> +	  workloads and control frequency/level of the various memory
>> +	  buses (DDR/LLCC/DDR_QOS).
>> +
>> +	  This driver defines/documents the parameter IDs used while configuring
>> +	  the memory buses.
>> +
>>   endmenu
>> diff --git a/drivers/soc/qcom/Makefile b/drivers/soc/qcom/Makefile
>> index acbca2ab5cc2..28549bb141bc 100644
>> --- a/drivers/soc/qcom/Makefile
>> +++ b/drivers/soc/qcom/Makefile
>> @@ -36,6 +36,7 @@ obj-$(CONFIG_QCOM_APR) += apr.o
>>   obj-$(CONFIG_QCOM_LLCC) += llcc-qcom.o
>>   obj-$(CONFIG_QCOM_KRYO_L2_ACCESSORS) +=	kryo-l2-accessors.o
>>   obj-$(CONFIG_QCOM_ICC_BWMON)	+= icc-bwmon.o
>> +obj-$(CONFIG_QCOM_SCMI_MEMLAT_CLIENT)	+= qcom_scmi_memlat_client.o
>>   qcom_ice-objs			+= ice.o
>>   obj-$(CONFIG_QCOM_INLINE_CRYPTO_ENGINE)	+= qcom_ice.o
>>   obj-$(CONFIG_QCOM_PBS) +=	qcom-pbs.o
>> diff --git a/drivers/soc/qcom/qcom_scmi_memlat_client.c b/drivers/soc/qcom/qcom_scmi_memlat_client.c
>> new file mode 100644
>> index 000000000000..05198bf1f7ec
>> --- /dev/null
>> +++ b/drivers/soc/qcom/qcom_scmi_memlat_client.c
>> @@ -0,0 +1,569 @@
>> +// SPDX-License-Identifier: GPL-2.0-only
>> +/*
>> + * Copyright (c) 2024, Qualcomm Innovation Center, Inc. All rights reserved.
>> + */
>> +
>> +#include <linux/cpu.h>
>> +#include <linux/err.h>
>> +#include <linux/errno.h>
>> +#include <linux/init.h>
>> +#include <linux/kernel.h>
>> +#include <linux/module.h>
>> +#include <linux/of.h>
>> +#include <linux/platform_device.h>
>> +#include <linux/scmi_protocol.h>
>> +#include <linux/scmi_qcom_protocol.h>
>> +#include <linux/units.h>
>> +#include <dt-bindings/firmware/qcom,scmi-memlat.h>
>> +
>> +#define MEMLAT_ALGO_STR				0x4D454D4C4154 /* MEMLAT */
>> +#define INVALID_IDX				0xff
>> +#define MAX_MEMORY_TYPES			3
>> +#define MAX_MONITOR_CNT				4
>> +#define MAX_NAME_LEN				20
>> +#define MAX_MAP_ENTRIES				7
>> +#define CPUCP_DEFAULT_SAMPLING_PERIOD_MS	4
>> +#define CPUCP_DEFAULT_FREQ_METHOD		1
>> +
>> +/**
>> + * enum scmi_memlat_protocol_cmd - parameter_ids supported by the "MEMLAT" algo_str hosted
>> + *                                 by the Qualcomm Generic Vendor Protocol on the SCMI controller.
>> + *
>> + * MEMLAT (Memory Latency) monitors the counters to detect memory latency bound workloads
>> + * and scales the frequency/levels of the memory buses accordingly.
>> + *
>> + * @MEMLAT_SET_MEM_GROUP: initializes the frequency/level scaling functions for the memory bus.
>> + * @MEMLAT_SET_MONITOR: configures the monitor to work on a specific memory bus.
>> + * @MEMLAT_SET_COMMON_EV_MAP: set up common counters used to monitor the cpu frequency.
>> + * @MEMLAT_SET_GRP_EV_MAP: set up any specific counters used to monitor the memory bus.
>> + * @MEMLAT_IPM_CEIL: set the IPM (Instruction Per Misses) ceiling per monitor.
>> + * @MEMLAT_SAMPLE_MS: set the sampling period for all the monitors.
>> + * @MEMLAT_MON_FREQ_MAP: setup the cpufreq to memfreq map.
>> + * @MEMLAT_SET_MIN_FREQ: set the max frequency of the memory bus.
>> + * @MEMLAT_SET_MAX_FREQ: set the min frequency of the memory bus.
>> + * @MEMLAT_START_TIMER: start all the monitors with the requested sampling period.
>> + * @MEMLAT_STOP_TIMER: stop all the running monitors.
>> + * @MEMLAT_SET_EFFECTIVE_FREQ_METHOD: set the method used to determine cpu frequency.
>> + */
>> +enum scmi_memlat_protocol_cmd {
>> +	MEMLAT_SET_MEM_GROUP = 16,
>> +	MEMLAT_SET_MONITOR,
>> +	MEMLAT_SET_COMMON_EV_MAP,
>> +	MEMLAT_SET_GRP_EV_MAP,
>> +	MEMLAT_IPM_CEIL = 23,
>> +	MEMLAT_SAMPLE_MS = 31,
>> +	MEMLAT_MON_FREQ_MAP,
>> +	MEMLAT_SET_MIN_FREQ,
>> +	MEMLAT_SET_MAX_FREQ,
>> +	MEMLAT_START_TIMER = 36,
>> +	MEMLAT_STOP_TIMER,
>> +	MEMLAT_SET_EFFECTIVE_FREQ_METHOD = 39,
>> +};
>> +
>> +struct map_table {
>> +	u16 v1;
>> +	u16 v2;
> 
> Huh? Why can't it be cpufreq and memfreq with some suffix?

ack

> 
>> +};
>> +
>> +struct map_param_msg {
>> +	u32 hw_type;
>> +	u32 mon_idx;
>> +	u32 nr_rows;
>> +	struct map_table tbl[MAX_MAP_ENTRIES];
>> +} __packed;
>> +
>> +struct node_msg {
>> +	u32 cpumask;
>> +	u32 hw_type;
>> +	u32 mon_type;
>> +	u32 mon_idx;
>> +	char mon_name[MAX_NAME_LEN];
>> +};
>> +
>> +struct scalar_param_msg {
>> +	u32 hw_type;
>> +	u32 mon_idx;
>> +	u32 val;
>> +};
>> +
>> +enum common_ev_idx {
>> +	INST_IDX,
>> +	CYC_IDX,
>> +	CONST_CYC_IDX,
>> +	FE_STALL_IDX,
>> +	BE_STALL_IDX,
>> +	NUM_COMMON_EVS
>> +};
>> +
>> +enum grp_ev_idx {
>> +	MISS_IDX,
>> +	WB_IDX,
>> +	ACC_IDX,
>> +	NUM_GRP_EVS
>> +};
>> +
>> +#define EV_CPU_CYCLES		0
>> +#define EV_INST_RETIRED		2
>> +#define EV_L2_D_RFILL		5
>> +
>> +struct ev_map_msg {
>> +	u32 num_evs;
>> +	u32 hw_type;
>> +	u32 cid[NUM_COMMON_EVS];
>> +};
>> +
>> +struct cpufreq_memfreq_map {
>> +	unsigned int cpufreq_mhz;
>> +	unsigned int memfreq_khz;
>> +};
>> +
>> +struct scmi_monitor_info {
>> +	struct cpufreq_memfreq_map *freq_map;
>> +	char mon_name[MAX_NAME_LEN];
>> +	u32 mon_idx;
>> +	u32 mon_type;
>> +	u32 ipm_ceil;
>> +	u32 mask;
>> +	u32 freq_map_len;
>> +};
>> +
>> +struct scmi_memory_info {
>> +	struct scmi_monitor_info *monitor[MAX_MONITOR_CNT];
>> +	u32 hw_type;
>> +	int monitor_cnt;
>> +	u32 min_freq;
>> +	u32 max_freq;
>> +};
>> +
>> +struct scmi_memlat_info {
>> +	struct scmi_protocol_handle *ph;
>> +	const struct qcom_generic_ext_ops *ops;
>> +	struct scmi_memory_info *memory[MAX_MEMORY_TYPES];
>> +	u32 cluster_info[NR_CPUS];
>> +	int memory_cnt;
>> +};
>> +
>> +static int populate_cluster_info(u32 *cluster_info)
>> +{
>> +	char name[MAX_NAME_LEN];
>> +	int i = 0;
>> +
>> +	struct device_node *cn __free(device_node) = of_find_node_by_path("/cpus");
>> +	if (!cn)
>> +		return -ENODEV;
>> +
>> +	struct device_node *map __free(device_node) = of_get_child_by_name(cn, "cpu-map");
>> +	if (!map)
>> +		return -ENODEV;
>> +
>> +	do {
>> +		snprintf(name, sizeof(name), "cluster%d", i);
>> +		struct device_node *c __free(device_node) = of_get_child_by_name(map, name);
>> +		if (!c)
>> +			break;
>> +
>> +		*(cluster_info + i) = of_get_child_count(c);
>> +		i++;
>> +	} while (1);
> 
> Can you use existing API from drivers/base/arch_topology.c? If not, can
> it be extended to support your usecase?

ack. But I'm pretty sure it's going to take a while for reaching such
an agreement so I'll drop this feature during the next re-spin.

> 
>> +
>> +	return 0;
>> +}
>> +
>> +static void populate_physical_mask(struct device_node *np, u32 *mask, u32 *cluster_info)
>> +{
>> +	struct device_node *dev_phandle __free(device_node);
>> +	int cpu, i = 0, physical_id;
>> +
>> +	do {
>> +		dev_phandle = of_parse_phandle(np, "cpus", i++);
>> +		cpu = of_cpu_node_to_id(dev_phandle);
>> +		if (cpu != -ENODEV) {
>> +			physical_id = topology_core_id(cpu);
>> +			for (int j = 0; j < topology_cluster_id(cpu); j++)
>> +				physical_id += *(cluster_info + j);
>> +			*mask |= BIT(physical_id);
>> +		}
>> +	} while (dev_phandle);
>> +}
>> +
>> +static struct cpufreq_memfreq_map *init_cpufreq_memfreq_map(struct device *dev,
>> +							    struct scmi_memory_info *memory,
>> +							    struct device_node *of_node,
>> +							    u32 *cnt)
>> +{
>> +	struct device_node *tbl_np __free(device_node), *opp_np __free(device_node);
>> +	struct cpufreq_memfreq_map *tbl;
>> +	int ret, i = 0;
>> +	u32 level, len;
>> +	u64 rate;
>> +
>> +	tbl_np = of_parse_phandle(of_node, "operating-points-v2", 0);
> 
> Please use existing API to parse OPP tables or document a reason why it
> can't be used.

Thanks, I had them documented as opens in the coverletter. Dropped them
since no one had any comments on it during V3. Will add them as comments
to this driver instead.

https://lore.kernel.org/lkml/20240702191440.2161623-1-quic_sibis@quicinc.com/

re-copying things again:
opp-tables are used but they don't get to be added to the scmi device
(thus we rely on a lot of manual parsing) because the memlat client 
driver doesn't vote on these resources clocks/interconnects/power-domain
from the kernel and some of the resources aren't modeled in the first
place like DDR_QOS.


> 
>> +	if (!tbl_np)
>> +		return ERR_PTR(-ENODEV);
>> +
>> +	len = min(of_get_available_child_count(tbl_np), MAX_MAP_ENTRIES);
>> +	if (len == 0)
>> +		return ERR_PTR(-ENODEV);
>> +
>> +	tbl = devm_kzalloc(dev, (len + 1) * sizeof(struct cpufreq_memfreq_map),
>> +			   GFP_KERNEL);
>> +	if (!tbl)
>> +		return ERR_PTR(-ENOMEM);
>> +
>> +	for_each_available_child_of_node(tbl_np, opp_np) {
>> +		ret = of_property_read_u64_index(opp_np, "opp-hz", 0, &rate);
>> +		if (ret < 0)
>> +			return ERR_PTR(ret);
>> +
>> +		tbl[i].cpufreq_mhz = rate / HZ_PER_MHZ;
>> +
>> +		if (memory->hw_type != QCOM_MEM_TYPE_DDR_QOS) {
>> +			ret = of_property_read_u64_index(opp_np, "opp-hz", 1, &rate);
>> +			if (ret < 0)
>> +				return ERR_PTR(ret);
>> +
>> +			tbl[i].memfreq_khz = rate / HZ_PER_KHZ;
>> +		} else {
>> +			ret = of_property_read_u32(opp_np, "opp-level", &level);
>> +			if (ret < 0)
>> +				return ERR_PTR(ret);
>> +
>> +			tbl[i].memfreq_khz = level;
>> +		}
>> +
>> +		dev_dbg(dev, "Entry%d CPU:%u, Mem:%u\n", i, tbl[i].cpufreq_mhz, tbl[i].memfreq_khz);
>> +		i++;
>> +	}
>> +	*cnt = len;
>> +
>> +	return tbl;
>> +}
>> +
>> +static int process_scmi_memlat_of_node(struct scmi_device *sdev, struct scmi_memlat_info *info)
>> +{
>> +	struct scmi_monitor_info *monitor;
>> +	struct scmi_memory_info *memory;
>> +	char name[MAX_NAME_LEN];
>> +	u64 memfreq[2];
>> +	int ret;
>> +
>> +	ret = populate_cluster_info(info->cluster_info);
>> +	if (ret < 0) {
>> +		dev_err_probe(&sdev->dev, ret, "failed to populate cluster info\n");
>> +		goto err;
>> +	}
>> +
>> +	of_node_get(sdev->dev.of_node);
>> +	do {
>> +		snprintf(name, sizeof(name), "memory-%d", info->memory_cnt);
>> +		struct device_node *memory_np __free(device_node) =
>> +			of_find_node_by_name(sdev->dev.of_node, name);
>> +
>> +		if (!memory_np)
>> +			break;
>> +
>> +		if (info->memory_cnt >= MAX_MEMORY_TYPES)
>> +			return dev_err_probe(&sdev->dev, -EINVAL,
>> +					     "failed to parse unsupported memory type\n");
>> +
>> +		memory = devm_kzalloc(&sdev->dev, sizeof(*memory), GFP_KERNEL);
>> +		if (!memory) {
>> +			ret = -ENOMEM;
>> +			goto err;
>> +		}
>> +
>> +		ret = of_property_read_u32(memory_np, "qcom,memory-type", &memory->hw_type);
>> +		if (ret) {
>> +			dev_err_probe(&sdev->dev, ret, "failed to read memory type\n");
>> +			goto err;
>> +		}
>> +
>> +		ret = of_property_read_u64_array(memory_np, "freq-table-hz", memfreq, 2);
>> +		if (ret && (ret != -EINVAL)) {
>> +			dev_err_probe(&sdev->dev, ret, "failed to read min/max freq\n");
>> +			goto err;
>> +		}
> 
> Can we get this information from the OPP table instead?

we don't list all the available ddr/llcc freqs in the opp-table
so that we can keep the table constant across platforms.

> 
>> +
>> +		if (memory->hw_type != QCOM_MEM_TYPE_DDR_QOS) {
>> +			memory->min_freq = memfreq[0] / HZ_PER_KHZ;
>> +			memory->max_freq = memfreq[1] / HZ_PER_KHZ;
>> +		} else {
>> +			memory->min_freq = memfreq[0];
>> +			memory->max_freq = memfreq[1];
> 
> Why? At least invert the logic here, please. The DDR_QOS is a special
> case, not all other kinds of memory.

ack
> 
>> +		}
>> +		info->memory[info->memory_cnt++] = memory;
>> +
>> +		do {
>> +			snprintf(name, sizeof(name), "monitor-%d", memory->monitor_cnt);
>> +			struct device_node *monitor_np __free(device_node) =
>> +				of_get_child_by_name(memory_np, name);
>> +
>> +			if (!monitor_np)
>> +				break;
>> +
>> +			if (memory->monitor_cnt >= MAX_MONITOR_CNT)
> 
> Why do you need to limit it? Is it a protocol limitation or an
> artificial driver limitation? Can monitors be allocated dynamically?

Yeah, they are limited to a max of 5 in firmware.

> 
>> +				return dev_err_probe(&sdev->dev, -EINVAL,
>> +						     "failed to parse unsupported monitor\n");
>> +
>> +			monitor = devm_kzalloc(&sdev->dev, sizeof(*monitor), GFP_KERNEL);
>> +			if (!monitor) {
>> +				ret = -ENOMEM;
>> +				goto err;
>> +			}
>> +
>> +			monitor->mon_type = of_property_read_bool(monitor_np, "qcom,compute-type");
>> +			if (!monitor->mon_type) {
>> +				ret = of_property_read_u32(monitor_np, "qcom,ipm-ceil",
>> +							   &monitor->ipm_ceil);
>> +				if (ret) {
>> +					dev_err_probe(&sdev->dev, ret,
>> +						      "failed to read IPM ceiling\n");
>> +					goto err;
>> +				}
>> +			}
>> +
>> +			/*
>> +			 * Variants of the SoC having reduced number of cpus operate
>> +			 * with the same number of logical cpus but the physical
>> +			 * cpu disabled will differ between parts. Calculate the
>> +			 * physical cpu number using cluster information instead.
>> +			 */
>> +			populate_physical_mask(monitor_np, &monitor->mask, info->cluster_info);
>> +
>> +			monitor->freq_map = init_cpufreq_memfreq_map(&sdev->dev, memory, monitor_np,
>> +								     &monitor->freq_map_len);
>> +			if (IS_ERR(monitor->freq_map)) {
>> +				dev_err_probe(&sdev->dev, PTR_ERR(monitor->freq_map),
>> +					      "failed to populate cpufreq-memfreq map\n");
>> +				goto err;
>> +			}
>> +
>> +			strscpy(monitor->mon_name, name, sizeof(monitor->mon_name));
>> +			monitor->mon_idx = memory->monitor_cnt;
>> +
>> +			memory->monitor[memory->monitor_cnt++] = monitor;
>> +		} while (1);
>> +
>> +		if (!memory->monitor_cnt) {
>> +			ret = -EINVAL;
>> +			dev_err_probe(&sdev->dev, ret, "failed to find monitor nodes\n");
>> +			goto err;
>> +		}
>> +	} while (1);
>> +
>> +	if (!info->memory_cnt) {
>> +		ret = -EINVAL;
>> +		dev_err_probe(&sdev->dev, ret, "failed to find memory nodes\n");
>> +	}
>> +
>> +err:
>> +	of_node_put(sdev->dev.of_node);
>> +
>> +	return ret;
>> +}
>> +
>> +static int configure_cpucp_common_events(struct scmi_memlat_info *info)
>> +{
>> +	const struct qcom_generic_ext_ops *ops = info->ops;
>> +	u8 ev_map[NUM_COMMON_EVS];
>> +	struct ev_map_msg msg;
>> +
>> +	memset(ev_map, 0xFF, NUM_COMMON_EVS);
>> +
>> +	msg.num_evs = NUM_COMMON_EVS;
>> +	msg.hw_type = INVALID_IDX;
>> +	msg.cid[INST_IDX] = EV_INST_RETIRED;
>> +	msg.cid[CYC_IDX] = EV_CPU_CYCLES;
>> +	msg.cid[CONST_CYC_IDX] = INVALID_IDX;
>> +	msg.cid[FE_STALL_IDX] = INVALID_IDX;
>> +	msg.cid[BE_STALL_IDX] = INVALID_IDX;
>> +
>> +	return ops->set_param(info->ph, &msg, sizeof(msg), MEMLAT_ALGO_STR,
>> +			      MEMLAT_SET_COMMON_EV_MAP);
>> +}
>> +
>> +static int configure_cpucp_grp(struct device *dev, struct scmi_memlat_info *info, int memory_index)
>> +{
>> +	const struct qcom_generic_ext_ops *ops = info->ops;
>> +	struct scmi_memory_info *memory = info->memory[memory_index];
>> +	struct ev_map_msg ev_msg;
>> +	u8 ev_map[NUM_GRP_EVS];
>> +	struct node_msg msg;
>> +	int ret;
>> +
>> +	msg.cpumask = 0;
>> +	msg.hw_type = memory->hw_type;
>> +	msg.mon_type = 0;
>> +	msg.mon_idx = 0;
>> +	ret = ops->set_param(info->ph, &msg, sizeof(msg), MEMLAT_ALGO_STR, MEMLAT_SET_MEM_GROUP);
>> +	if (ret < 0)
>> +		return dev_err_probe(dev, ret, "failed to configure mem type %d\n",
>> +				     memory->hw_type);
>> +
>> +	memset(ev_map, 0xFF, NUM_GRP_EVS);
>> +	ev_msg.num_evs = NUM_GRP_EVS;
>> +	ev_msg.hw_type = memory->hw_type;
>> +	ev_msg.cid[MISS_IDX] = EV_L2_D_RFILL;
>> +	ev_msg.cid[WB_IDX] = INVALID_IDX;
>> +	ev_msg.cid[ACC_IDX] = INVALID_IDX;
>> +	ret = ops->set_param(info->ph, &ev_msg, sizeof(ev_msg), MEMLAT_ALGO_STR,
>> +			     MEMLAT_SET_GRP_EV_MAP);
>> +	if (ret < 0)
>> +		return dev_err_probe(dev, ret, "failed to configure event map for mem type %d\n",
>> +				     memory->hw_type);
>> +
>> +	return ret;
>> +}
>> +
>> +static int configure_cpucp_mon(struct device *dev, struct scmi_memlat_info *info,
>> +			       int memory_index, int monitor_index)
>> +{
>> +	const struct qcom_generic_ext_ops *ops = info->ops;
>> +	struct scmi_memory_info *memory = info->memory[memory_index];
>> +	struct scmi_monitor_info *monitor = memory->monitor[monitor_index];
>> +	struct scalar_param_msg scalar_msg;
>> +	struct map_param_msg map_msg;
>> +	struct node_msg msg;
>> +	int ret;
>> +	int i;
>> +
>> +	msg.cpumask = monitor->mask;
>> +	msg.hw_type = memory->hw_type;
>> +	msg.mon_type = monitor->mon_type;
>> +	msg.mon_idx = monitor->mon_idx;
>> +	strscpy(msg.mon_name, monitor->mon_name, sizeof(msg.mon_name));
>> +	ret = ops->set_param(info->ph, &msg, sizeof(msg), MEMLAT_ALGO_STR, MEMLAT_SET_MONITOR);
>> +	if (ret < 0)
>> +		return dev_err_probe(dev, ret, "failed to configure monitor %s\n",
>> +				     monitor->mon_name);
>> +
>> +	scalar_msg.hw_type = memory->hw_type;
>> +	scalar_msg.mon_idx = monitor->mon_idx;
>> +	scalar_msg.val = monitor->ipm_ceil;
>> +	ret = ops->set_param(info->ph, &scalar_msg, sizeof(scalar_msg), MEMLAT_ALGO_STR,
>> +			     MEMLAT_IPM_CEIL);
>> +	if (ret < 0)
>> +		return dev_err_probe(dev, ret, "failed to set ipm ceil for %s\n",
>> +				     monitor->mon_name);
>> +
>> +	map_msg.hw_type = memory->hw_type;
>> +	map_msg.mon_idx = monitor->mon_idx;
>> +	map_msg.nr_rows = monitor->freq_map_len;
>> +	for (i = 0; i < monitor->freq_map_len; i++) {
>> +		map_msg.tbl[i].v1 = monitor->freq_map[i].cpufreq_mhz;
>> +		map_msg.tbl[i].v2 = monitor->freq_map[i].memfreq_khz;
>> +	}
>> +	ret = ops->set_param(info->ph, &map_msg, sizeof(map_msg), MEMLAT_ALGO_STR,
>> +			     MEMLAT_MON_FREQ_MAP);
>> +	if (ret < 0)
>> +		return dev_err_probe(dev, ret, "failed to configure freq_map for %s\n",
>> +				     monitor->mon_name);
>> +
>> +	scalar_msg.hw_type = memory->hw_type;
>> +	scalar_msg.mon_idx = monitor->mon_idx;
>> +	scalar_msg.val = memory->min_freq;
>> +	ret = ops->set_param(info->ph, &scalar_msg, sizeof(scalar_msg), MEMLAT_ALGO_STR,
>> +			     MEMLAT_SET_MIN_FREQ);
>> +	if (ret < 0)
>> +		return dev_err_probe(dev, ret, "failed to set min_freq for %s\n",
>> +				     monitor->mon_name);
>> +
>> +	scalar_msg.hw_type = memory->hw_type;
>> +	scalar_msg.mon_idx = monitor->mon_idx;
>> +	scalar_msg.val = memory->max_freq;
>> +	ret = ops->set_param(info->ph, &scalar_msg, sizeof(scalar_msg), MEMLAT_ALGO_STR,
>> +			     MEMLAT_SET_MAX_FREQ);
>> +	if (ret < 0)
>> +		dev_err_probe(dev, ret, "failed to set max_freq for %s\n", monitor->mon_name);
>> +
>> +	return ret;
>> +}
>> +
>> +static int cpucp_memlat_init(struct scmi_device *sdev)
>> +{
>> +	const struct scmi_handle *handle = sdev->handle;
>> +	const struct qcom_generic_ext_ops *ops;
>> +	struct scmi_protocol_handle *ph;
>> +	struct scmi_memlat_info *info;
>> +	u32 cpucp_freq_method = CPUCP_DEFAULT_FREQ_METHOD;
>> +	u32 cpucp_sample_ms = CPUCP_DEFAULT_SAMPLING_PERIOD_MS;
>> +	int ret, i, j;
>> +
>> +	if (!handle)
>> +		return -ENODEV;
>> +
>> +	ops = handle->devm_protocol_get(sdev, SCMI_PROTOCOL_QCOM_GENERIC, &ph);
>> +	if (IS_ERR(ops))
>> +		return PTR_ERR(ops);
>> +
>> +	info = devm_kzalloc(&sdev->dev, sizeof(*info), GFP_KERNEL);
>> +	if (!info)
>> +		return -ENOMEM;
>> +
>> +	ret = process_scmi_memlat_of_node(sdev, info);
>> +	if (ret)
>> +		return ret;
>> +
>> +	info->ph = ph;
>> +	info->ops = ops;
>> +
>> +	/* Configure common events ids */
>> +	ret = configure_cpucp_common_events(info);
>> +	if (ret < 0)
>> +		return dev_err_probe(&sdev->dev, ret, "failed to configure common events\n");
>> +
>> +	for (i = 0; i < info->memory_cnt; i++) {
>> +		/* Configure per group parameters */
>> +		ret = configure_cpucp_grp(&sdev->dev, info, i);
>> +		if (ret < 0)
>> +			return ret;
>> +
>> +		for (j = 0; j < info->memory[i]->monitor_cnt; j++) {
>> +			/* Configure per monitor parameters */
>> +			ret = configure_cpucp_mon(&sdev->dev, info, i, j);
>> +			if (ret < 0)
>> +				return ret;
>> +		}
>> +	}
>> +
>> +	/* Set loop sampling time */
>> +	ret = ops->set_param(ph, &cpucp_sample_ms, sizeof(cpucp_sample_ms), MEMLAT_ALGO_STR,
>> +			     MEMLAT_SAMPLE_MS);
>> +	if (ret < 0)
>> +		return dev_err_probe(&sdev->dev, ret, "failed to set sample_ms\n");
>> +
>> +	/* Set the effective cpu frequency calculation method */
>> +	ret = ops->set_param(ph, &cpucp_freq_method, sizeof(cpucp_freq_method), MEMLAT_ALGO_STR,
>> +			     MEMLAT_SET_EFFECTIVE_FREQ_METHOD);
>> +	if (ret < 0)
>> +		return dev_err_probe(&sdev->dev, ret,
>> +				     "failed to set effective frequency calc method\n");
>> +
>> +	/* Start sampling and voting timer */
>> +	ret = ops->start_activity(ph, NULL, 0, MEMLAT_ALGO_STR, MEMLAT_START_TIMER);
>> +	if (ret < 0)
>> +		dev_err_probe(&sdev->dev, ret, "failed to start memory group timer\n");
>> +
>> +	return ret;
>> +}
>> +
>> +static int scmi_client_probe(struct scmi_device *sdev)
>> +{
>> +	return cpucp_memlat_init(sdev);
> 
> Inline it here, please.

ack.

-Sibi

> 
>> +}
>> +
>> +static const struct scmi_device_id scmi_id_table[] = {
>> +	{ SCMI_PROTOCOL_QCOM_GENERIC, "qcom-generic-ext" },
>> +	{ },
>> +};
>> +MODULE_DEVICE_TABLE(scmi, scmi_id_table);
>> +
>> +static struct scmi_driver qcom_scmi_client_drv = {
>> +	.name		= "scmi-qcom-generic-ext-memlat",
>> +	.probe		= scmi_client_probe,
>> +	.id_table	= scmi_id_table,
>> +};
>> +module_scmi_driver(qcom_scmi_client_drv);
>> +
>> +MODULE_DESCRIPTION("QTI SCMI client driver");
>> +MODULE_LICENSE("GPL");
>> -- 
>> 2.34.1
>>
>

On Mon, Oct 07, 2024 at 11:40:22AM +0530, Sibi Sankar wrote:
> Introduce a client driver that uses the memlat algorithm string
> hosted on QCOM SCMI Generic Extension Protocol to detect memory
> latency workloads and control frequency/level of the various
> memory buses (DDR/LLCC/DDR_QOS).
> 

Hi,

a few small remarks, down below.

> Co-developed-by: Shivnandan Kumar <quic_kshivnan@quicinc.com>
> Signed-off-by: Shivnandan Kumar <quic_kshivnan@quicinc.com>
> Co-developed-by: Ramakrishna Gottimukkula <quic_rgottimu@quicinc.com>
> Signed-off-by: Ramakrishna Gottimukkula <quic_rgottimu@quicinc.com>
> Co-developed-by: Amir Vajid <avajid@quicinc.com>
> Signed-off-by: Amir Vajid <avajid@quicinc.com>
> Signed-off-by: Sibi Sankar <quic_sibis@quicinc.com>
> ---

[snip]

> +static int populate_cluster_info(u32 *cluster_info)
> +{
> +	char name[MAX_NAME_LEN];
> +	int i = 0;
> +
> +	struct device_node *cn __free(device_node) = of_find_node_by_path("/cpus");
> +	if (!cn)
> +		return -ENODEV;

Not sure if this is some new coding style accepted for the new cleanup.h
fancy stuff (sincere question/doubt...so please take this with a grain of salt),
BUT, if not, you should consider grouping this definition/initialization to
the start of the block whose scope they are in...like:


	struct device_node *cn __free(device_node) = of_find_node_by_path("/cpus");
	struct device_node *map __free(device_node) = NULL;
	char name[MAX_NAME_LEN];
	int i = 0;

	if (!cn)
		return -ENODEV;

	map = of_get_child_by_name(cn, "cpu-map");
	if (!map)
		return -ENODEV;

> +
> +	struct device_node *map __free(device_node) = of_get_child_by_name(cn, "cpu-map");
> +	if (!map)
> +		return -ENODEV;
> +

As said...

> +	do {
> +		snprintf(name, sizeof(name), "cluster%d", i);
> +		struct device_node *c __free(device_node) = of_get_child_by_name(map, name);
> +		if (!c)
> +			break;
> +
> +		*(cluster_info + i) = of_get_child_count(c);
> +		i++;
> +	} while (1);
> +
> +	return 0;
> +}
> +
> +static void populate_physical_mask(struct device_node *np, u32 *mask, u32 *cluster_info)
> +{
> +	struct device_node *dev_phandle __free(device_node);

...so this cleanups on return....

> +	int cpu, i = 0, physical_id;
> +
> +	do {
> +		dev_phandle = of_parse_phandle(np, "cpus", i++);

BUT wont this be needed to be of_put, between calls to of_parse_phandle
inside this loop ? ... so cannot this be done like

	int cpu, i = 0, physical_id;

	while (1) {
		struct device_node *dev_phandle __free(device_node) = of_parse_phandle(np, "cpus", i++);
	
		if (!dev_phandle)
			break;

		cpu = of_cpu_node_to_id(dev_phandle);
		if (cpu != -ENODEV) {
			....
	}

...not even build tested ... ah... :P


> +		cpu = of_cpu_node_to_id(dev_phandle);
> +		if (cpu != -ENODEV) {
> +			physical_id = topology_core_id(cpu);
> +			for (int j = 0; j < topology_cluster_id(cpu); j++)
> +				physical_id += *(cluster_info + j);
> +			*mask |= BIT(physical_id);
> +		}
> +	} while (dev_phandle);
> +}
> +
> +static struct cpufreq_memfreq_map *init_cpufreq_memfreq_map(struct device *dev,
> +							    struct scmi_memory_info *memory,
> +							    struct device_node *of_node,
> +							    u32 *cnt)
> +{
> +	struct device_node *tbl_np __free(device_node), *opp_np __free(device_node);
> +	struct cpufreq_memfreq_map *tbl;
> +	int ret, i = 0;
> +	u32 level, len;
> +	u64 rate;
> +
> +	tbl_np = of_parse_phandle(of_node, "operating-points-v2", 0);
> +	if (!tbl_np)
> +		return ERR_PTR(-ENODEV);
> +
> +	len = min(of_get_available_child_count(tbl_np), MAX_MAP_ENTRIES);
> +	if (len == 0)
> +		return ERR_PTR(-ENODEV);
> +
> +	tbl = devm_kzalloc(dev, (len + 1) * sizeof(struct cpufreq_memfreq_map),
> +			   GFP_KERNEL);
> +	if (!tbl)
> +		return ERR_PTR(-ENOMEM);
> +
> +	for_each_available_child_of_node(tbl_np, opp_np) {

This seems to lack a of+node_put at the end but possibly the scoped
version  for_each_available_child_of_node_scoped() will do it for you...

> +		ret = of_property_read_u64_index(opp_np, "opp-hz", 0, &rate);
> +		if (ret < 0)
> +			return ERR_PTR(ret);
> +
> +		tbl[i].cpufreq_mhz = rate / HZ_PER_MHZ;
> +
> +		if (memory->hw_type != QCOM_MEM_TYPE_DDR_QOS) {
> +			ret = of_property_read_u64_index(opp_np, "opp-hz", 1, &rate);
> +			if (ret < 0)
> +				return ERR_PTR(ret);
> +
> +			tbl[i].memfreq_khz = rate / HZ_PER_KHZ;
> +		} else {
> +			ret = of_property_read_u32(opp_np, "opp-level", &level);
> +			if (ret < 0)
> +				return ERR_PTR(ret);
> +
> +			tbl[i].memfreq_khz = level;
> +		}
> +
> +		dev_dbg(dev, "Entry%d CPU:%u, Mem:%u\n", i, tbl[i].cpufreq_mhz, tbl[i].memfreq_khz);
> +		i++;
> +	}
> +	*cnt = len;
> +
> +	return tbl;
> +}
> +
> +static int process_scmi_memlat_of_node(struct scmi_device *sdev, struct scmi_memlat_info *info)
> +{
> +	struct scmi_monitor_info *monitor;
> +	struct scmi_memory_info *memory;
> +	char name[MAX_NAME_LEN];
> +	u64 memfreq[2];
> +	int ret;
> +
> +	ret = populate_cluster_info(info->cluster_info);
> +	if (ret < 0) {
> +		dev_err_probe(&sdev->dev, ret, "failed to populate cluster info\n");
> +		goto err;
> +	}
> +
> +	of_node_get(sdev->dev.of_node);

cant you use cleanup.h magic also for this and get rid of a few gotos down below ?
...this function seems the ideal case fot that...

> +	do {
> +		snprintf(name, sizeof(name), "memory-%d", info->memory_cnt);
> +		struct device_node *memory_np __free(device_node) =
> +			of_find_node_by_name(sdev->dev.of_node, name);
> +
> +		if (!memory_np)
> +			break;
> +
> +		if (info->memory_cnt >= MAX_MEMORY_TYPES)

Shouldn't the MAX_MEMORY_TYPES something discoverable at runtime through
some command of your vendor protocol ? for better future scalability I
mean...maybe I am overthinking... 

> +			return dev_err_probe(&sdev->dev, -EINVAL,
> +					     "failed to parse unsupported memory type\n");
> +
> +		memory = devm_kzalloc(&sdev->dev, sizeof(*memory), GFP_KERNEL);
> +		if (!memory) {
> +			ret = -ENOMEM;
> +			goto err;
> +		}
> +

Thanks,
Cristian

On Tue, Oct 22, 2024 at 01:48:25PM +0530, Sibi Sankar wrote:
> 
> 

Hi Sibi,

one more thing down below...

> On 10/7/24 23:27, Dmitry Baryshkov wrote:
> > On Mon, Oct 07, 2024 at 11:40:22AM GMT, Sibi Sankar wrote:
> > > Introduce a client driver that uses the memlat algorithm string
> > > hosted on QCOM SCMI Generic Extension Protocol to detect memory
> > > latency workloads and control frequency/level of the various
> > > memory buses (DDR/LLCC/DDR_QOS).

[snip]

> > > +		}
> > > +		info->memory[info->memory_cnt++] = memory;
> > > +
> > > +		do {
> > > +			snprintf(name, sizeof(name), "monitor-%d", memory->monitor_cnt);
> > > +			struct device_node *monitor_np __free(device_node) =
> > > +				of_get_child_by_name(memory_np, name);
> > > +
> > > +			if (!monitor_np)
> > > +				break;
> > > +
> > > +			if (memory->monitor_cnt >= MAX_MONITOR_CNT)
> > 
> > Why do you need to limit it? Is it a protocol limitation or an
> > artificial driver limitation? Can monitors be allocated dynamically?
> 
> Yeah, they are limited to a max of 5 in firmware.

Similarly as I already commented elsewhere, in an SCMI-based platform
this is something you should be able to discover at run-time with a
query command of some kind (or using some bits of the existing cmds)
....so that you dont have this dependency on constants defined and
built into the platform fw....what if your fw evolves to allow more
monitors ?

Thanks,
Cristian