[PATCH v3] perf: arm_dsu: Support DSU ACPI devices

[Tuan Phan]

> On Sep 7, 2020, at 4:02 AM, Will Deacon <will at kernel.org> wrote:
> 
> [+ Suzuki as I'd like his Ack on this]
> 
> On Fri, Aug 14, 2020 at 05:39:40PM -0700, Tuan Phan wrote:
>> Add support for probing device from ACPI node.
>> Each DSU ACPI node and its associated cpus are inside a cluster node.
>> 
>> Signed-off-by: Tuan Phan <tuanphan at os.amperecomputing.com>
>> ---
>> Changes in v3:
>> - Based on the latest ARM ACPI binding at: https://developer.arm.com/documentation/den0093/c/
>> 
>> Changes in v2:
>> - Removed IRQF_SHARED.
>> - Fixed ACPI runtime detection.
>> 
>> drivers/perf/arm_dsu_pmu.c | 68 ++++++++++++++++++++++++++++++++++++++++------
>> 1 file changed, 60 insertions(+), 8 deletions(-)
>> 
>> diff --git a/drivers/perf/arm_dsu_pmu.c b/drivers/perf/arm_dsu_pmu.c
>> index 96ed93c..4be355d 100644
>> --- a/drivers/perf/arm_dsu_pmu.c
>> +++ b/drivers/perf/arm_dsu_pmu.c
>> @@ -11,6 +11,7 @@
>> #define DRVNAME		PMUNAME "_pmu"
>> #define pr_fmt(fmt)	DRVNAME ": " fmt
>> 
>> +#include <linux/acpi.h>
>> #include <linux/bitmap.h>
>> #include <linux/bitops.h>
>> #include <linux/bug.h>
>> @@ -603,18 +604,21 @@ static struct dsu_pmu *dsu_pmu_alloc(struct platform_device *pdev)
>> }
>> 
>> /**
>> - * dsu_pmu_dt_get_cpus: Get the list of CPUs in the cluster.
>> + * dsu_pmu_dt_get_cpus: Get the list of CPUs in the cluster
>> + * from device tree.
>>  */
>> -static int dsu_pmu_dt_get_cpus(struct device_node *dev, cpumask_t *mask)
>> +static int dsu_pmu_dt_get_cpus(struct platform_device *pdev)
>> {
>> 	int i = 0, n, cpu;
>> 	struct device_node *cpu_node;
>> +	struct dsu_pmu *dsu_pmu =
>> +		(struct dsu_pmu *) platform_get_drvdata(pdev);
>> 
>> -	n = of_count_phandle_with_args(dev, "cpus", NULL);
>> +	n = of_count_phandle_with_args(pdev->dev.of_node, "cpus", NULL);
>> 	if (n <= 0)
>> 		return -ENODEV;
>> 	for (; i < n; i++) {
>> -		cpu_node = of_parse_phandle(dev, "cpus", i);
>> +		cpu_node = of_parse_phandle(pdev->dev.of_node, "cpus", i);
>> 		if (!cpu_node)
>> 			break;
>> 		cpu = of_cpu_node_to_id(cpu_node);
>> @@ -626,11 +630,51 @@ static int dsu_pmu_dt_get_cpus(struct device_node *dev, cpumask_t *mask)
>> 		 */
>> 		if (cpu < 0)
>> 			continue;
>> -		cpumask_set_cpu(cpu, mask);
>> +		cpumask_set_cpu(cpu, &dsu_pmu->associated_cpus);
>> 	}
>> 	return 0;
>> }
>> 
>> +/**
>> + * dsu_pmu_acpi_get_cpus: Get the list of CPUs in the cluster
>> + * from ACPI.
>> + */
>> +static int dsu_pmu_acpi_get_cpus(struct platform_device *pdev)
>> +{
>> +	int cpu;
>> +	struct dsu_pmu *dsu_pmu = (struct dsu_pmu *) platform_get_drvdata(pdev);
>> +
>> +	/*
>> +	 * A dsu pmu node is inside a cluster parent node along with cpu nodes.
>> +	 * We need to find out all cpus that have the same parent with this pmu.
>> +	 */
>> +	for_each_possible_cpu(cpu) {
>> +		struct acpi_device *acpi_dev = ACPI_COMPANION(get_cpu_device(cpu));
>> +
>> +		if (acpi_dev &&
>> +			acpi_dev->parent == ACPI_COMPANION(&pdev->dev)->parent)
>> +			cpumask_set_cpu(cpu, &dsu_pmu->associated_cpus);
>> +	}
>> +
>> +	return 0;
>> +}
>> +
>> +static int dsu_pmu_platform_get_cpus(struct platform_device *pdev)
>> +{
>> +	int ret = -ENOENT;
>> +	struct fwnode_handle *fwnode = dev_fwnode(&pdev->dev);
>> +
>> +	if (IS_ERR_OR_NULL(fwnode))
>> +		return ret;
>> +
>> +	if (is_of_node(fwnode))
>> +		ret = dsu_pmu_dt_get_cpus(pdev);
>> +	else if (is_acpi_device_node(fwnode))
>> +		ret = dsu_pmu_acpi_get_cpus(pdev);
>> +
>> +	return ret;
>> +}
>> +
>> /*
>>  * dsu_pmu_probe_pmu: Probe the PMU details on a CPU in the cluster.
>>  */
>> @@ -683,7 +727,9 @@ static int dsu_pmu_device_probe(struct platform_device *pdev)
>> 	if (IS_ERR(dsu_pmu))
>> 		return PTR_ERR(dsu_pmu);
>> 
>> -	rc = dsu_pmu_dt_get_cpus(pdev->dev.of_node, &dsu_pmu->associated_cpus);
>> +	platform_set_drvdata(pdev, dsu_pmu);
> 
> Hmm, this is a bit nasty because we haven't finished initialising the
> dsu_pmu yet. I think it would actually be cleaner if you kept:
> 
> 	static int dsu_pmu_dt_get_cpus(struct device_node *dev, cpumask_t *mask)
> 
> for the DT case and added:
> 
> 	static int dsu_pmu_acpi_get_cpus(struct device *dev, cpumask_t *mask)
> 
> for the ACPI case. I suppose the DT case could take the struct device * too
> if you wanted the prototypes to be the same.

Sure, I can do that.
> 
>> +	rc = dsu_pmu_platform_get_cpus(pdev);
>> 	if (rc) {
>> 		dev_warn(&pdev->dev, "Failed to parse the CPUs\n");
>> 		return rc;
>> @@ -705,7 +751,6 @@ static int dsu_pmu_device_probe(struct platform_device *pdev)
>> 	}
>> 
>> 	dsu_pmu->irq = irq;
>> -	platform_set_drvdata(pdev, dsu_pmu);
>> 	rc = cpuhp_state_add_instance(dsu_pmu_cpuhp_state,
>> 						&dsu_pmu->cpuhp_node);
>> 	if (rc)
>> @@ -752,11 +797,19 @@ static const struct of_device_id dsu_pmu_of_match[] = {
>> 	{ .compatible = "arm,dsu-pmu", },
>> 	{},
>> };
>> +MODULE_DEVICE_TABLE(of, dsu_pmu_of_match);
>> +
>> +static const struct acpi_device_id dsu_pmu_acpi_match[] = {
>> +	{ "ARMHD500", 0},
>> +	{},
>> +};
>> +MODULE_DEVICE_TABLE(acpi, dsu_pmu_acpi_match);
>> 
>> static struct platform_driver dsu_pmu_driver = {
>> 	.driver = {
>> 		.name	= DRVNAME,
>> 		.of_match_table = of_match_ptr(dsu_pmu_of_match),
>> +		.acpi_match_table = ACPI_PTR(dsu_pmu_acpi_match),
>> 		.suppress_bind_attrs = true,
>> 	},
>> 	.probe = dsu_pmu_device_probe,
>> @@ -826,7 +879,6 @@ static void __exit dsu_pmu_exit(void)
>> module_init(dsu_pmu_init);
>> module_exit(dsu_pmu_exit);
>> 
>> -MODULE_DEVICE_TABLE(of, dsu_pmu_of_match);
> 
> Why have you moved this line?

I moved it to just behind the dsu_pmu_of_match definition as we now also have dsu_pmu_acpi_match.

Tuan
> 
> Will