Re: [PATCH v8 2/2] ThunderX2, perf : Add Cavium ThunderX2 SoC UNCORE PMU driver

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Mon, Dec 3, 2018 at 5:41 PM Will Deacon <will.deacon@xxxxxxx> wrote:
>
> On Thu, Nov 22, 2018 at 03:04:35AM +0000, Kulkarni, Ganapatrao wrote:
> > This patch adds a perf driver for the PMU UNCORE devices DDR4 Memory
> > Controller(DMC) and Level 3 Cache(L3C). Each PMU supports up to 4
> > counters. All counters lack overflow interrupt and are
> > sampled periodically.
> >
> > Signed-off-by: Ganapatrao Kulkarni <ganapatrao.kulkarni@xxxxxxxxxx>
> > ---
> >  drivers/perf/Kconfig         |   9 +
> >  drivers/perf/Makefile        |   1 +
> >  drivers/perf/thunderx2_pmu.c | 869 +++++++++++++++++++++++++++++++++++
> >  include/linux/cpuhotplug.h   |   1 +
> >  4 files changed, 880 insertions(+)
> >  create mode 100644 drivers/perf/thunderx2_pmu.c
> >
> > diff --git a/drivers/perf/Kconfig b/drivers/perf/Kconfig
> > index 08ebaf7cca8b..af9bc178495d 100644
> > --- a/drivers/perf/Kconfig
> > +++ b/drivers/perf/Kconfig
> > @@ -87,6 +87,15 @@ config QCOM_L3_PMU
> >          Adds the L3 cache PMU into the perf events subsystem for
> >          monitoring L3 cache events.
> >
> > +config THUNDERX2_PMU
> > +     tristate "Cavium ThunderX2 SoC PMU UNCORE"
> > +     depends on ARCH_THUNDER2 && ARM64 && ACPI && NUMA
> > +     default m
> > +     help
> > +        Provides support for ThunderX2 UNCORE events.
> > +        The SoC has PMU support in its L3 cache controller (L3C) and
> > +        in the DDR4 Memory Controller (DMC).
> > +
> >  config XGENE_PMU
> >          depends on ARCH_XGENE
> >          bool "APM X-Gene SoC PMU"
> > diff --git a/drivers/perf/Makefile b/drivers/perf/Makefile
> > index b3902bd37d53..909f27fd9db3 100644
> > --- a/drivers/perf/Makefile
> > +++ b/drivers/perf/Makefile
> > @@ -7,5 +7,6 @@ obj-$(CONFIG_ARM_PMU_ACPI) += arm_pmu_acpi.o
> >  obj-$(CONFIG_HISI_PMU) += hisilicon/
> >  obj-$(CONFIG_QCOM_L2_PMU)    += qcom_l2_pmu.o
> >  obj-$(CONFIG_QCOM_L3_PMU) += qcom_l3_pmu.o
> > +obj-$(CONFIG_THUNDERX2_PMU) += thunderx2_pmu.o
> >  obj-$(CONFIG_XGENE_PMU) += xgene_pmu.o
> >  obj-$(CONFIG_ARM_SPE_PMU) += arm_spe_pmu.o
> > diff --git a/drivers/perf/thunderx2_pmu.c b/drivers/perf/thunderx2_pmu.c
> > new file mode 100644
> > index 000000000000..e6509ba868ab
> > --- /dev/null
> > +++ b/drivers/perf/thunderx2_pmu.c
> > @@ -0,0 +1,869 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +/*
> > + * CAVIUM THUNDERX2 SoC PMU UNCORE
> > + * Copyright (C) 2018 Cavium Inc.
> > + * Author: Ganapatrao Kulkarni <gkulkarni@xxxxxxxxxx>
> > + */
> > +
> > +#include <linux/acpi.h>
> > +#include <linux/cpuhotplug.h>
> > +#include <linux/perf_event.h>
> > +#include <linux/platform_device.h>
> > +
> > +/* Each ThunderX2(TX2) Socket has a L3C and DMC UNCORE PMU device.
> > + * Each UNCORE PMU device consists of 4 independent programmable counters.
> > + * Counters are 32 bit and do not support overflow interrupt,
> > + * they need to be sampled before overflow(i.e, at every 2 seconds).
> > + */
> > +
> > +#define TX2_PMU_MAX_COUNTERS         4
> > +#define TX2_PMU_DMC_CHANNELS         8
> > +#define TX2_PMU_L3_TILES             16
> > +
> > +#define TX2_PMU_HRTIMER_INTERVAL     (2 * NSEC_PER_SEC)
> > +#define GET_EVENTID(ev)                      ((ev->hw.config) & 0x1ff)
>
> I think this should be 0x1f.

yes it should be, i will update it.
>
> > +#define GET_COUNTERID(ev)            ((ev->hw.idx) & 0x3)
> > + /* 1 byte per counter(4 counters).
> > +  * Event id is encoded in bits [5:1] of a byte,
> > +  */
> > +#define DMC_EVENT_CFG(idx, val)              ((val) << (((idx) * 8) + 1))
> > +
> > +#define L3C_COUNTER_CTL                      0xA8
> > +#define L3C_COUNTER_DATA             0xAC
> > +#define DMC_COUNTER_CTL                      0x234
> > +#define DMC_COUNTER_DATA             0x240
> > +
> > +/* L3C event IDs */
> > +#define L3_EVENT_READ_REQ            0xD
> > +#define L3_EVENT_WRITEBACK_REQ               0xE
> > +#define L3_EVENT_INV_N_WRITE_REQ     0xF
> > +#define L3_EVENT_INV_REQ             0x10
> > +#define L3_EVENT_EVICT_REQ           0x13
> > +#define L3_EVENT_INV_N_WRITE_HIT     0x14
> > +#define L3_EVENT_INV_HIT             0x15
> > +#define L3_EVENT_READ_HIT            0x17
> > +#define L3_EVENT_MAX                 0x18
> > +
> > +/* DMC event IDs */
> > +#define DMC_EVENT_COUNT_CYCLES               0x1
> > +#define DMC_EVENT_WRITE_TXNS         0xB
> > +#define DMC_EVENT_DATA_TRANSFERS     0xD
> > +#define DMC_EVENT_READ_TXNS          0xF
> > +#define DMC_EVENT_MAX                        0x10
> > +
> > +enum tx2_uncore_type {
> > +     PMU_TYPE_L3C,
> > +     PMU_TYPE_DMC,
> > +     PMU_TYPE_INVALID,
> > +};
> > +
> > +/*
> > + * pmu on each socket has 2 uncore devices(dmc and l3c),
> > + * each device has 4 counters.
> > + */
> > +struct tx2_uncore_pmu {
> > +     struct hlist_node hpnode;
> > +     struct list_head  entry;
> > +     struct pmu pmu;
> > +     char *name;
> > +     int node;
> > +     int cpu;
> > +     u32    max_counters;
> > +     u32    prorate_factor;
> > +     u32    max_events;
> > +     u64 hrtimer_interval;
> > +     void __iomem *base;
> > +     DECLARE_BITMAP(active_counters, TX2_PMU_MAX_COUNTERS);
> > +     struct perf_event *events[TX2_PMU_MAX_COUNTERS];
> > +     struct device *dev;
> > +     struct hrtimer hrtimer;
> > +     const struct attribute_group **attr_groups;
> > +     enum tx2_uncore_type type;
> > +     void    (*init_cntr_base)(struct perf_event *event,
> > +                     struct tx2_uncore_pmu *tx2_pmu);
> > +     void    (*stop_event)(struct perf_event *event);
> > +     void    (*start_event)(struct perf_event *event, int flags);
> > +};
> > +
> > +static LIST_HEAD(tx2_pmus);
> > +
> > +static inline struct tx2_uncore_pmu *pmu_to_tx2_pmu(struct pmu *pmu)
> > +{
> > +     return container_of(pmu, struct tx2_uncore_pmu, pmu);
> > +}
> > +
> > +/*
> > + * sysfs format attributes
> > + */
> > +static ssize_t tx2_pmu_format_show(struct device *dev,
> > +                                  struct device_attribute *attr, char *buf)
> > +{
> > +     struct dev_ext_attribute *eattr;
> > +
> > +     eattr = container_of(attr, struct dev_ext_attribute, attr);
> > +     return sprintf(buf, "%s\n", (char *) eattr->var);
> > +}
> > +
> > +#define FORMAT_ATTR(_name, _config) \
> > +     (&((struct dev_ext_attribute[]) { \
> > +        { \
> > +        .attr = __ATTR(_name, 0444, tx2_pmu_format_show, NULL), \
> > +        .var = (void *) _config, \
> > +        } \
> > +     })[0].attr.attr)
> > +
> > +static struct attribute *l3c_pmu_format_attrs[] = {
> > +     FORMAT_ATTR(event,      "config:0-4"),
> > +     NULL,
> > +};
> > +
> > +static struct attribute *dmc_pmu_format_attrs[] = {
> > +     FORMAT_ATTR(event,      "config:0-4"),
> > +     NULL,
> > +};
>
> We have PMU_FORMAT_ATTR, PMU_EVENT_ATTR etc in the core code to help here.
> Please try to use them.

ok, i will try to use it.
>
> > +static const struct attribute_group l3c_pmu_format_attr_group = {
> > +     .name = "format",
> > +     .attrs = l3c_pmu_format_attrs,
> > +};
> > +
> > +static const struct attribute_group dmc_pmu_format_attr_group = {
> > +     .name = "format",
> > +     .attrs = dmc_pmu_format_attrs,
> > +};
> > +
> > +/*
> > + * sysfs event attributes
> > + */
> > +static ssize_t tx2_pmu_event_show(struct device *dev,
> > +                                 struct device_attribute *attr, char *buf)
> > +{
> > +     struct dev_ext_attribute *eattr;
> > +
> > +     eattr = container_of(attr, struct dev_ext_attribute, attr);
> > +     return sprintf(buf, "config=0x%lx\n", (unsigned long) eattr->var);
> > +}
>
> Shouldn't this be "event=" instead of "config="?

yep, thanks
>
> Will

Thanks,
Ganapat



[Index of Archives]     [Kernel Newbies]     [Security]     [Netfilter]     [Bugtraq]     [Linux FS]     [Yosemite Forum]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Samba]     [Video 4 Linux]     [Device Mapper]     [Linux Resources]

  Powered by Linux