Re: [PATCH v3 1/2] perf x86: Infrastructure for exposing an Uncore unit to PMON mapping

[Greg KH]

On Mon, Jan 13, 2020 at 04:54:43PM +0300, roman.sudarikov@xxxxxxxxxxxxxxx wrote:
> From: Roman Sudarikov <roman.sudarikov@xxxxxxxxxxxxxxx>
> 
> Intel® Xeon® Scalable processor family (code name Skylake-SP) makes
> significant changes in the integrated I/O (IIO) architecture. The new
> solution introduces IIO stacks which are responsible for managing traffic
> between the PCIe domain and the Mesh domain. Each IIO stack has its own
> PMON block and can handle either DMI port, x16 PCIe root port, MCP-Link
> or various built-in accelerators. IIO PMON blocks allow concurrent
> monitoring of I/O flows up to 4 x4 bifurcation within each IIO stack.
> 
> Software is supposed to program required perf counters within each IIO
> stack and gather performance data. The tricky thing here is that IIO PMON
> reports data per IIO stack but users have no idea what IIO stacks are -
> they only know devices which are connected to the platform.
> 
> Understanding IIO stack concept to find which IIO stack that particular
> IO device is connected to, or to identify an IIO PMON block to program
> for monitoring specific IIO stack assumes a lot of implicit knowledge
> about given Intel server platform architecture.
> 
> Usage example:
>     /sys/devices/uncore_<type>_<pmu_idx>/platform_mapping
> 
> Each Uncore unit type, by its nature, can be mapped to its own context,
> for example:
> 1. CHA - each uncore_cha_<pmu_idx> is assigned to manage a distinct slice
>    of LLC capacity;
> 2. UPI - each uncore_upi_<pmu_idx> is assigned to manage one link of Intel
>    UPI Subsystem;
> 3. IIO - each uncore_iio_<pmu_idx> is assigned to manage one stack of the
>    IIO module;
> 4. IMC - each uncore_imc_<pmu_idx> is assigned to manage one channel of
>    Memory Controller.
> 
> Implementation details:
> Two callbacks added to struct intel_uncore_type to discover and map Uncore
> units to PMONs:
>     int (*get_topology)(void)
>     int (*set_mapping)(struct intel_uncore_pmu *pmu)
> 
> Details of IIO Uncore unit mapping to IIO PMON:
> Each IIO stack is either DMI port, x16 PCIe root port, MCP-Link or various
> built-in accelerators. For Uncore IIO Unit type, the platform_mapping file
> holds bus numbers of devices, which can be monitored by that IIO PMON block
> on each die.
> 
> Co-developed-by: Alexander Antonov <alexander.antonov@xxxxxxxxx>
> Reviewed-by: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>
> Signed-off-by: Alexander Antonov <alexander.antonov@xxxxxxxxx>
> Signed-off-by: Roman Sudarikov <roman.sudarikov@xxxxxxxxxxxxxxx>
> ---
>  arch/x86/events/intel/uncore.c | 37 +++++++++++++++++++++++++++++++++-
>  arch/x86/events/intel/uncore.h |  9 ++++++++-
>  2 files changed, 44 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/x86/events/intel/uncore.c b/arch/x86/events/intel/uncore.c
> index 86467f85c383..2c53ad44b51f 100644
> --- a/arch/x86/events/intel/uncore.c
> +++ b/arch/x86/events/intel/uncore.c
> @@ -905,6 +905,32 @@ static void uncore_types_exit(struct intel_uncore_type **types)
>  		uncore_type_exit(*types);
>  }
>  
> +static struct attribute *empty_attrs[] = {
> +	NULL,
> +};
> +
> +static const struct attribute_group empty_group = {
> +	.attrs = empty_attrs,
> +};

What is this for?  Why is it needed?  It doesn't do anything?

> +
> +static ssize_t platform_mapping_show(struct device *dev,
> +				struct device_attribute *attr, char *buf)
> +{
> +	struct intel_uncore_pmu *pmu = dev_get_drvdata(dev);
> +
> +	return snprintf(buf, PAGE_SIZE - 1, "%s\n", pmu->mapping);
> +}
> +static DEVICE_ATTR_RO(platform_mapping);

You are creating new sysfs attributes without any Documentation/ABI
updates, which is not ok.  Please fix this up for your next round of
patches.

> +static struct attribute *mapping_attrs[] = {
> +	&dev_attr_platform_mapping.attr,
> +	NULL,
> +};
> +
> +static const struct attribute_group uncore_mapping_group = {
> +	.attrs = mapping_attrs,
> +};

ATTRIBUTE_GROUPS()?

Messing around with single attribute_group lists is usually a sign that
something is really wrong as the driver core should handle arrays of
attribute group lists instead.


> +
>  static int __init uncore_type_init(struct intel_uncore_type *type, bool setid)
>  {
>  	struct intel_uncore_pmu *pmus;
> @@ -950,10 +976,19 @@ static int __init uncore_type_init(struct intel_uncore_type *type, bool setid)
>  			attr_group->attrs[j] = &type->event_descs[j].attr.attr;
>  
>  		type->events_group = &attr_group->group;
> -	}
> +	} else
> +		type->events_group = &empty_group;

Why???

Didn't we fix up the x86 attributes to work properly and not mess around
with trying to merge groups and the like?  Please don't perpetuate that
more...

thanks,

greg k-h