Re: [PATCH v4 1/5] arm64: perf: Basic uncore counter support for Cavium ThunderX SOC

[Jan Glauber]

Hi Will,

thanks for the review!

On Tue, Nov 08, 2016 at 11:50:10PM +0000, Will Deacon wrote:
> Hi Jan,
> 
> Thanks for posting an updated series. I have a few minor comments, which
> we can hopefully address in time for 4.10.
> 
> Also, have you run the perf fuzzer with this series applied?

No, that's new to me. Will try it.

> https://github.com/deater/perf_event_tests
> 
> (build the tests and look under the fuzzer/ directory for the tool)
> 
> On Sat, Oct 29, 2016 at 01:55:29PM +0200, Jan Glauber wrote:
> > Provide "uncore" facilities for different non-CPU performance
> > counter units.
> > 
> > The uncore PMUs can be found under /sys/bus/event_source/devices.
> > All counters are exported via sysfs in the corresponding events
> > files under the PMU directory so the perf tool can list the event names.
> > 
> > There are some points that are special in this implementation:
> > 
> > 1) The PMU detection relies on PCI device detection. If a
> >    matching PCI device is found the PMU is created. The code can deal
> >    with multiple units of the same type, e.g. more than one memory
> >    controller.
> > 
> > 2) Counters are summarized across different units of the same type
> >    on one NUMA node but not across NUMA nodes.
> >    For instance L2C TAD 0..7 are presented as a single counter
> >    (adding the values from TAD 0 to 7). Although losing the ability
> >    to read a single value the merged values are easier to use.
> > 
> > 3) The counters are not CPU related. A random CPU is picked regardless
> >    of the NUMA node. There is a small performance penalty for accessing
> >    counters on a remote note but reading a performance counter is a
> >    slow operation anyway.
> > 
> > Signed-off-by: Jan Glauber <jglauber@xxxxxxxxxx>
> > ---
> >  drivers/perf/Kconfig                |  13 ++
> >  drivers/perf/Makefile               |   1 +
> >  drivers/perf/uncore/Makefile        |   1 +
> >  drivers/perf/uncore/uncore_cavium.c | 351 ++++++++++++++++++++++++++++++++++++
> >  drivers/perf/uncore/uncore_cavium.h |  71 ++++++++
> 
> We already have "uncore" PMUs under drivers/perf, so I'd prefer that we
> renamed this a bit to reflect better what's going on. How about:
> 
>   drivers/perf/cavium/
> 
> and then
> 
>   drivers/perf/cavium/uncore_thunder.[ch]
> 
> ?

OK, agreed.

> >  include/linux/cpuhotplug.h          |   1 +
> >  6 files changed, 438 insertions(+)
> >  create mode 100644 drivers/perf/uncore/Makefile
> >  create mode 100644 drivers/perf/uncore/uncore_cavium.c
> >  create mode 100644 drivers/perf/uncore/uncore_cavium.h
> > 
> > diff --git a/drivers/perf/Kconfig b/drivers/perf/Kconfig
> > index 4d5c5f9..3266c87 100644
> > --- a/drivers/perf/Kconfig
> > +++ b/drivers/perf/Kconfig
> > @@ -19,4 +19,17 @@ config XGENE_PMU
> >          help
> >            Say y if you want to use APM X-Gene SoC performance monitors.
> >  
> > +config UNCORE_PMU
> > +	bool
> 
> This isn't needed.

I thought about a symbol for all uncore pmus. But when drivers/perf is
already that it can be removed.

> > +
> > +config UNCORE_PMU_CAVIUM
> > +	depends on PERF_EVENTS && NUMA && ARM64
> > +	bool "Cavium uncore PMU support"
> 
> Please mentioned Thunder somewhere, since that's the SoC being supported.

OK.

> > +	select UNCORE_PMU
> > +	default y
> > +	help
> > +	  Say y if you want to access performance counters of subsystems
> > +	  on a Cavium SOC like cache controller, memory controller or
> > +	  processor interconnect.
> > +
> >  endmenu
> > diff --git a/drivers/perf/Makefile b/drivers/perf/Makefile
> > index b116e98..d6c02c9 100644
> > --- a/drivers/perf/Makefile
> > +++ b/drivers/perf/Makefile
> > @@ -1,2 +1,3 @@
> >  obj-$(CONFIG_ARM_PMU) += arm_pmu.o
> >  obj-$(CONFIG_XGENE_PMU) += xgene_pmu.o
> > +obj-y += uncore/
> > diff --git a/drivers/perf/uncore/Makefile b/drivers/perf/uncore/Makefile
> > new file mode 100644
> > index 0000000..6130e18
> > --- /dev/null
> > +++ b/drivers/perf/uncore/Makefile
> > @@ -0,0 +1 @@
> > +obj-$(CONFIG_UNCORE_PMU_CAVIUM) += uncore_cavium.o
> > diff --git a/drivers/perf/uncore/uncore_cavium.c b/drivers/perf/uncore/uncore_cavium.c
> > new file mode 100644
> > index 0000000..a7b4277
> > --- /dev/null
> > +++ b/drivers/perf/uncore/uncore_cavium.c
> > @@ -0,0 +1,351 @@
> > +/*
> > + * Cavium Thunder uncore PMU support.
> > + *
> > + * Copyright (C) 2015,2016 Cavium Inc.
> > + * Author: Jan Glauber <jan.glauber@xxxxxxxxxx>
> > + */
> > +
> > +#include <linux/cpufeature.h>
> > +#include <linux/numa.h>
> > +#include <linux/slab.h>
> > +
> > +#include "uncore_cavium.h"
> > +
> > +/*
> > + * Some notes about the various counters supported by this "uncore" PMU
> > + * and the design:
> > + *
> > + * All counters are 64 bit long.
> > + * There are no overflow interrupts.
> > + * Counters are summarized per node/socket.
> > + * Most devices appear as separate PCI devices per socket with the exception
> > + * of OCX TLK which appears as one PCI device per socket and contains several
> > + * units with counters that are merged.
> > + * Some counters are selected via a control register (L2C TAD) and read by
> > + * a number of counter registers, others (L2C CBC, LMC & OCX TLK) have
> > + * one dedicated counter per event.
> > + * Some counters are not stoppable (L2C CBC & LMC).
> > + * Some counters are read-only (LMC).
> > + * All counters belong to PCI devices, the devices may have additional
> > + * drivers but we assume we are the only user of the counter registers.
> > + * We map the whole PCI BAR so we must be careful to forbid access to
> > + * addresses that contain neither counters nor counter control registers.
> > + */
> > +
> > +void thunder_uncore_read(struct perf_event *event)
> > +{
> 
> Rather than have a bunch of global symbols that are called from the
> individual drivers, why don't you pass a struct of function pointers to
> their respective init functions and keep the internals private?

Most of these functions are already in struct pmu. What I can do is
set the shared functions in thunder_uncore_setup as default, and
only override them as needed (like thunder_uncore_read_ocx_tlk)
or the other way around (use default if not set already).
Then I can get rid of them.

> > +	struct thunder_uncore *uncore = to_uncore(event->pmu);
> > +	struct hw_perf_event *hwc = &event->hw;
> > +	struct thunder_uncore_node *node;
> > +	struct thunder_uncore_unit *unit;
> > +	u64 prev, delta, new = 0;
> > +
> > +	node = get_node(hwc->config, uncore);
> > +
> > +	/* read counter values from all units on the node */
> > +	list_for_each_entry(unit, &node->unit_list, entry)
> > +		new += readq(hwc->event_base + unit->map);
> > +
> > +	prev = local64_read(&hwc->prev_count);
> > +	local64_set(&hwc->prev_count, new);
> > +	delta = new - prev;
> > +	local64_add(delta, &event->count);
> > +}
> > +
> > +int thunder_uncore_add(struct perf_event *event, int flags, u64 config_base,
> > +		       u64 event_base)
> > +{
> > +	struct thunder_uncore *uncore = to_uncore(event->pmu);
> > +	struct hw_perf_event *hwc = &event->hw;
> > +	struct thunder_uncore_node *node;
> > +	int id;
> > +
> > +	node = get_node(hwc->config, uncore);
> > +	id = get_id(hwc->config);
> > +
> > +	if (!cmpxchg(&node->events[id], NULL, event))
> > +		hwc->idx = id;
> 
> Does this need to be a full-fat cmpxchg? Who are you racing with?

Just copy'n'paste from the existing drivers. I guess it can be relaxed.

> > +
> > +	if (hwc->idx == -1)
> > +		return -EBUSY;
> 
> This would be much clearer as an else statement after the cmpxchg.

Agreed.

> > +
> > +	hwc->config_base = config_base;
> > +	hwc->event_base = event_base;
> > +	hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
> > +
> > +	if (flags & PERF_EF_START)
> > +		uncore->pmu.start(event, PERF_EF_RELOAD);
> > +
> > +	return 0;
> > +}
> > +
> > +void thunder_uncore_del(struct perf_event *event, int flags)
> > +{
> > +	struct thunder_uncore *uncore = to_uncore(event->pmu);
> > +	struct hw_perf_event *hwc = &event->hw;
> > +	struct thunder_uncore_node *node;
> > +	int i;
> > +
> > +	event->pmu->stop(event, PERF_EF_UPDATE);
> > +
> > +	/*
> > +	 * For programmable counters we need to check where we installed it.
> > +	 * To keep this function generic always test the more complicated
> > +	 * case (free running counters won't need the loop).
> > +	 */
> > +	node = get_node(hwc->config, uncore);
> > +	for (i = 0; i < node->num_counters; i++) {
> > +		if (cmpxchg(&node->events[i], event, NULL) == event)
> > +			break;
> > +	}
> > +	hwc->idx = -1;
> > +}
> > +
> > +void thunder_uncore_start(struct perf_event *event, int flags)
> > +{
> > +	struct thunder_uncore *uncore = to_uncore(event->pmu);
> > +	struct hw_perf_event *hwc = &event->hw;
> > +	struct thunder_uncore_node *node;
> > +	struct thunder_uncore_unit *unit;
> > +	u64 new = 0;
> > +
> > +	/* read counter values from all units on the node */
> > +	node = get_node(hwc->config, uncore);
> > +	list_for_each_entry(unit, &node->unit_list, entry)
> > +		new += readq(hwc->event_base + unit->map);
> > +	local64_set(&hwc->prev_count, new);
> > +
> > +	hwc->state = 0;
> > +	perf_event_update_userpage(event);
> > +}
> > +
> > +void thunder_uncore_stop(struct perf_event *event, int flags)
> > +{
> > +	struct hw_perf_event *hwc = &event->hw;
> > +
> > +	hwc->state |= PERF_HES_STOPPED;
> > +
> > +	if ((flags & PERF_EF_UPDATE) && !(hwc->state & PERF_HES_UPTODATE)) {
> > +		thunder_uncore_read(event);
> > +		hwc->state |= PERF_HES_UPTODATE;
> > +	}
> > +}
> > +
> > +int thunder_uncore_event_init(struct perf_event *event)
> > +{
> > +	struct hw_perf_event *hwc = &event->hw;
> > +	struct thunder_uncore_node *node;
> > +	struct thunder_uncore *uncore;
> > +
> > +	if (event->attr.type != event->pmu->type)
> > +		return -ENOENT;
> > +
> > +	/* we do not support sampling */
> > +	if (is_sampling_event(event))
> > +		return -EINVAL;
> > +
> > +	/* counters do not have these bits */
> > +	if (event->attr.exclude_user	||
> > +	    event->attr.exclude_kernel	||
> > +	    event->attr.exclude_host	||
> > +	    event->attr.exclude_guest	||
> > +	    event->attr.exclude_hv	||
> > +	    event->attr.exclude_idle)
> > +		return -EINVAL;
> > +
> > +	uncore = to_uncore(event->pmu);
> > +	if (!uncore)
> > +		return -ENODEV;
> > +	if (!uncore->event_valid(event->attr.config & UNCORE_EVENT_ID_MASK))
> > +		return -EINVAL;
> > +
> > +	/* check NUMA node */
> > +	node = get_node(event->attr.config, uncore);
> > +	if (!node) {
> > +		pr_debug("Invalid NUMA node selected\n");
> > +		return -EINVAL;
> > +	}
> > +
> > +	hwc->config = event->attr.config;
> > +	hwc->idx = -1;
> > +	return 0;
> > +}
> > +
> > +static ssize_t thunder_uncore_attr_show_cpumask(struct device *dev,
> > +						struct device_attribute *attr,
> > +						char *buf)
> > +{
> > +	struct pmu *pmu = dev_get_drvdata(dev);
> > +	struct thunder_uncore *uncore =
> > +		container_of(pmu, struct thunder_uncore, pmu);
> > +
> > +	return cpumap_print_to_pagebuf(true, buf, &uncore->active_mask);
> > +}
> > +static DEVICE_ATTR(cpumask, S_IRUGO, thunder_uncore_attr_show_cpumask, NULL);
> > +
> > +static struct attribute *thunder_uncore_attrs[] = {
> > +	&dev_attr_cpumask.attr,
> > +	NULL,
> > +};
> > +
> > +struct attribute_group thunder_uncore_attr_group = {
> > +	.attrs = thunder_uncore_attrs,
> > +};
> > +
> > +ssize_t thunder_events_sysfs_show(struct device *dev,
> > +				  struct device_attribute *attr,
> > +				  char *page)
> > +{
> > +	struct perf_pmu_events_attr *pmu_attr =
> > +		container_of(attr, struct perf_pmu_events_attr, attr);
> > +
> > +	if (pmu_attr->event_str)
> > +		return sprintf(page, "%s", pmu_attr->event_str);
> > +
> > +	return 0;
> > +}
> > +
> > +/* node attribute depending on number of NUMA nodes */
> > +static ssize_t node_show(struct device *dev, struct device_attribute *attr,
> > +			 char *page)
> > +{
> > +	if (NODES_SHIFT)
> > +		return sprintf(page, "config:16-%d\n", 16 + NODES_SHIFT - 1);
> 
> If NODES_SHIFT is 1, you'll end up with "config:16-16", which might confuse
> userspace.

So should I use "config:16" in that case? Is it OK to use this also for
NODES_SHIFT=0 ?

> > +	else
> > +		return sprintf(page, "config:16\n");
> > +}
> > +
> > +struct device_attribute format_attr_node = __ATTR_RO(node);
> > +
> > +/*
> > + * Thunder uncore events are independent from CPUs. Provide a cpumask
> > + * nevertheless to prevent perf from adding the event per-cpu and just
> > + * set the mask to one online CPU. Use the same cpumask for all uncore
> > + * devices.
> > + *
> > + * There is a performance penalty for accessing a device from a CPU on
> > + * another socket, but we do not care (yet).
> > + */
> > +static int thunder_uncore_offline_cpu(unsigned int old_cpu, struct hlist_node *node)
> > +{
> > +	struct thunder_uncore *uncore = hlist_entry_safe(node, struct thunder_uncore, node);
> 
> Why _safe?

Not required, will remove.

> > +	int new_cpu;
> > +
> > +	if (!cpumask_test_and_clear_cpu(old_cpu, &uncore->active_mask))
> > +		return 0;
> > +	new_cpu = cpumask_any_but(cpu_online_mask, old_cpu);
> > +	if (new_cpu >= nr_cpu_ids)
> > +		return 0;
> > +	perf_pmu_migrate_context(&uncore->pmu, old_cpu, new_cpu);
> > +	cpumask_set_cpu(new_cpu, &uncore->active_mask);
> > +	return 0;
> > +}
> > +
> > +static struct thunder_uncore_node * __init alloc_node(struct thunder_uncore *uncore,
> > +						      int node_id, int counters)
> > +{
> > +	struct thunder_uncore_node *node;
> > +
> > +	node = kzalloc(sizeof(*node), GFP_KERNEL);
> > +	if (!node)
> > +		return NULL;
> > +	node->num_counters = counters;
> > +	INIT_LIST_HEAD(&node->unit_list);
> > +	return node;
> > +}
> > +
> > +int __init thunder_uncore_setup(struct thunder_uncore *uncore, int device_id,
> > +				struct pmu *pmu, int counters)
> > +{
> > +	unsigned int vendor_id = PCI_VENDOR_ID_CAVIUM;
> > +	struct thunder_uncore_unit  *unit, *tmp;
> > +	struct thunder_uncore_node *node;
> > +	struct pci_dev *pdev = NULL;
> > +	int ret, node_id, found = 0;
> > +
> > +	/* detect PCI devices */
> > +	while ((pdev = pci_get_device(vendor_id, device_id, pdev))) {
> 
> Redundant brackets?

OK

> > +		if (!pdev)
> > +			break;
> 
> Redundant check?

OK

> > +		node_id = dev_to_node(&pdev->dev);
> > +
> > +		/* allocate node if necessary */
> > +		if (!uncore->nodes[node_id])
> > +			uncore->nodes[node_id] = alloc_node(uncore, node_id, counters);
> > +
> > +		node = uncore->nodes[node_id];
> > +		if (!node) {
> > +			ret = -ENOMEM;
> > +			goto fail;
> > +		}
> > +
> > +		unit = kzalloc(sizeof(*unit), GFP_KERNEL);
> > +		if (!unit) {
> > +			ret = -ENOMEM;
> > +			goto fail;
> > +		}
> > +
> > +		unit->pdev = pdev;
> > +		unit->map = ioremap(pci_resource_start(pdev, 0),
> > +				    pci_resource_len(pdev, 0));
> > +		list_add(&unit->entry, &node->unit_list);
> > +		node->nr_units++;
> > +		found++;
> > +	}
> > +
> > +	if (!found)
> > +		return -ENODEV;
> > +
> > +	cpuhp_state_add_instance_nocalls(CPUHP_AP_UNCORE_CAVIUM_ONLINE,
> > +                                         &uncore->node);
> > +
> > +	/*
> > +	 * perf PMU is CPU dependent in difference to our uncore devices.
> > +	 * Just pick a CPU and migrate away if it goes offline.
> > +	 */
> > +	cpumask_set_cpu(smp_processor_id(), &uncore->active_mask);
> > +
> > +	uncore->pmu = *pmu;
> > +	ret = perf_pmu_register(&uncore->pmu, uncore->pmu.name, -1);
> > +	if (ret)
> > +		goto fail;
> > +
> > +	return 0;
> > +
> > +fail:
> > +	node_id = 0;
> > +	while (uncore->nodes[node_id]) {
> > +		node = uncore->nodes[node_id];
> > +
> > +		list_for_each_entry_safe(unit, tmp, &node->unit_list, entry) {
> 
> Why do you need the _safe variant?

OK, not needed

> Will