Re: [PATCH 3/6] perf/marvell : Odyssey LLC-TAD performance monitor support

[Mark Rutland]

On Fri, Jun 30, 2023 at 05:33:48PM +0530, Gowthami Thiagarajan wrote:
> Each TAD provides eight 64-bit counters for monitoring
> cache behavior.The driver always configures the same counter for
> all the TADs. The user would end up effectively reserving one of
> eight counters in every TAD to look across all TADs.
> The occurrences of events are aggregated and presented to the user
> at the end of running the workload. The driver does not provide a
> way for the user to partition TADs so that different TADs are used for
> different applications.
> 
> The performance events reflect various internal or interface activities.
> By combining the values from multiple performance counters, cache
> performance can be measured in terms such as: cache miss rate, cache
> allocations, interface retry rate, internal resource occupancy, etc.
> 
> Each supported counter's event and formatting information is exposed
> to sysfs at /sys/devices/tad/. Use perf tool stat command to measure
> the pmu events. For instance:
> 
> perf stat -e tad_hit_ltg,tad_hit_dtg <workload>
> 
> Signed-off-by: Gowthami Thiagarajan <gthiagarajan@xxxxxxxxxxx>

This generally looks ok; I have a few comments below.

[...]

> +static void tad_pmu_event_counter_stop(struct perf_event *event, int flags)
> +{
> +	struct tad_pmu *tad_pmu = to_tad_pmu(event->pmu);
> +	struct hw_perf_event *hwc = &event->hw;
> +	u32 counter_idx = hwc->idx;
> +	int tad_region;
> +
> +	/* TAD()_PFC() stop counting on the write
> +	 * which sets TAD()_PRF()[CNTSEL] == 0
> +	 */

Please fix the comment style.

Likewise for all other instances within this file.

[...]

> +static int tad_pmu_event_counter_add(struct perf_event *event, int flags)
> +{
> +	struct tad_pmu *tad_pmu = to_tad_pmu(event->pmu);
> +	struct hw_perf_event *hwc = &event->hw;
> +	int idx;
> +
> +	/* Get a free counter for this event */
> +	idx = find_first_zero_bit(tad_pmu->counters_map, TAD_MAX_COUNTERS);
> +	if (idx == TAD_MAX_COUNTERS)
> +		return -EAGAIN;
> +
> +	set_bit(idx, tad_pmu->counters_map);
> +
> +	hwc->idx = idx;
> +	hwc->state = PERF_HES_STOPPED;
> +	tad_pmu->events[idx] = event;
> +
> +	if (flags & PERF_EF_START)
> +		tad_pmu_event_counter_start(event, flags);
> +
> +	return 0;
> +}
> +
> +static int tad_pmu_event_init(struct perf_event *event)
> +{
> +	struct tad_pmu *tad_pmu = to_tad_pmu(event->pmu);
> +
> +	if (event->attr.type != event->pmu->type)
> +		return -ENOENT;

Why is this not rejecting smapling events, as patch 1 does?

> +
> +	if (!event->attr.disabled)
> +		return -EINVAL;

Why?

> +
> +	if (event->state != PERF_EVENT_STATE_OFF)
> +		return -EINVAL;

Event groups need to be verified here too.

[...]

> +static int tad_pmu_probe(struct platform_device *pdev)
> +{
> +	struct device *dev = &pdev->dev;
> +	struct tad_region *regions;
> +	struct tad_pmu *tad_pmu;
> +	struct resource *res;
> +	u32 tad_pmu_page_size;
> +	u32 tad_page_size;
> +	u32 tad_cnt;
> +	int i, ret;
> +	char *name;
> +
> +	tad_pmu = devm_kzalloc(&pdev->dev, sizeof(*tad_pmu), GFP_KERNEL);
> +	if (!tad_pmu)
> +		return -ENOMEM;
> +
> +	platform_set_drvdata(pdev, tad_pmu);
> +
> +	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> +	if (!res) {
> +		dev_err(&pdev->dev, "Mem resource not found\n");
> +		return -ENODEV;
> +	}
> +
> +	ret = device_property_read_u32(dev, "marvell,tad-page-size", &tad_page_size);
> +	if (ret) {
> +		dev_err(&pdev->dev, "Can't find tad-page-size property\n");
> +		return ret;
> +	}
> +
> +	ret = device_property_read_u32(dev, "marvell,tad-pmu-page-size",
> +				       &tad_pmu_page_size);
> +	if (ret) {
> +		dev_err(&pdev->dev, "Can't find tad-pmu-page-size property\n");
> +		return ret;
> +	}

Why do you think these properties are necessary?

These should almost certainly be provided by IO resources, and shouldn't need a
custom property.

Thanks,
Mark.