Re: [PATCH v7 16/26] PM / devfreq: rockchip-dfi: Add perf support

[Sascha Hauer]

On Mon, Oct 09, 2023 at 06:48:43AM +0900, Chanwoo Choi wrote:
> On 23. 7. 4. 18:32, Sascha Hauer wrote:
> > The DFI is a unit which is suitable for measuring DDR utilization, but
> > so far it could only be used as an event driver for the DDR frequency
> > scaling driver. This adds perf support to the DFI driver.
> > 
> > Usage with the 'perf' tool can look like:
> > 
> > perf stat -a -e rockchip_ddr/cycles/,\
> > 		rockchip_ddr/read-bytes/,\
> > 		rockchip_ddr/write-bytes/,\
> > 		rockchip_ddr/bytes/ sleep 1
> > 
> >  Performance counter stats for 'system wide':
> > 
> >         1582524826      rockchip_ddr/cycles/
> >            1802.25 MB   rockchip_ddr/read-bytes/
> >            1793.72 MB   rockchip_ddr/write-bytes/
> >            3595.90 MB   rockchip_ddr/bytes/
> > 
> >        1.014369709 seconds time elapsed
> > 
> > perf support has been tested on a RK3568 and a RK3399, the latter with
> > dual channel DDR.
> > 
> > Signed-off-by: Sascha Hauer <s.hauer@xxxxxxxxxxxxxx>
> > Reviewed-by: Sebastian Reichel <sebastian.reichel@xxxxxxxxxxxxx>
> > ---
> > 
> > Notes:
> >     Changes since v5:
> >     - Add missing initialization of &dfi->last_perf_count
> >     
> >     Changes since v4:
> >     
> >     - use __stringify to ensure event type definitions and event numbers in sysfs are consistent
> >     - only use 64bit values in structs holding counters
> >     - support monitoring individual DDR channels
> >     - fix return value in rockchip_ddr_perf_event_init(): -EOPNOTSUPP -> -EINVAL
> >     - check for invalid event->attr.config values
> >     - start hrtimer to trigger in one second, not immediately
> >     - use devm_add_action_or_reset()
> >     - add suppress_bind_attrs
> >     - enable DDRMON during probe when perf is enabled
> >     - use a seqlock to protect perf reading the counters from the hrtimer callback modifying them
> > 
> >  drivers/devfreq/event/rockchip-dfi.c | 442 ++++++++++++++++++++++++++-
> >  include/soc/rockchip/rk3399_grf.h    |   2 +
> >  include/soc/rockchip/rk3568_grf.h    |   1 +
> >  3 files changed, 440 insertions(+), 5 deletions(-)
> > 
> > diff --git a/drivers/devfreq/event/rockchip-dfi.c b/drivers/devfreq/event/rockchip-dfi.c
> > index 50e497455dc69..969b62f071b83 100644
> > --- a/drivers/devfreq/event/rockchip-dfi.c
> > +++ b/drivers/devfreq/event/rockchip-dfi.c
> > @@ -16,10 +16,12 @@
> >  #include <linux/regmap.h>
> >  #include <linux/slab.h>
> >  #include <linux/list.h>
> > +#include <linux/seqlock.h>
> >  #include <linux/of.h>
> >  #include <linux/of_device.h>
> >  #include <linux/bitfield.h>
> >  #include <linux/bits.h>
> > +#include <linux/perf_event.h>
> >  
> >  #include <soc/rockchip/rockchip_grf.h>
> >  #include <soc/rockchip/rk3399_grf.h>
> > @@ -41,19 +43,39 @@
> >  					 DDRMON_CTRL_LPDDR4 | \
> >  					 DDRMON_CTRL_LPDDR23)
> >  
> > +#define DDRMON_CH0_WR_NUM		0x20
> > +#define DDRMON_CH0_RD_NUM		0x24
> >  #define DDRMON_CH0_COUNT_NUM		0x28
> >  #define DDRMON_CH0_DFI_ACCESS_NUM	0x2c
> >  #define DDRMON_CH1_COUNT_NUM		0x3c
> >  #define DDRMON_CH1_DFI_ACCESS_NUM	0x40
> >  
> > +#define PERF_EVENT_CYCLES		0x0
> > +#define PERF_EVENT_READ_BYTES		0x1
> > +#define PERF_EVENT_WRITE_BYTES		0x2
> > +#define PERF_EVENT_READ_BYTES0		0x3
> > +#define PERF_EVENT_WRITE_BYTES0		0x4
> > +#define PERF_EVENT_READ_BYTES1		0x5
> > +#define PERF_EVENT_WRITE_BYTES1		0x6
> > +#define PERF_EVENT_READ_BYTES2		0x7
> > +#define PERF_EVENT_WRITE_BYTES2		0x8
> > +#define PERF_EVENT_READ_BYTES3		0x9
> > +#define PERF_EVENT_WRITE_BYTES3		0xa
> > +#define PERF_EVENT_BYTES		0xb
> > +#define PERF_ACCESS_TYPE_MAX		0xc
> > +
> >  /**
> >   * struct dmc_count_channel - structure to hold counter values from the DDR controller
> >   * @access:       Number of read and write accesses
> >   * @clock_cycles: DDR clock cycles
> > + * @read_access:  number of read accesses
> > + * @write_acccess: number of write accesses
> >   */
> >  struct dmc_count_channel {
> > -	u32 access;
> > -	u32 clock_cycles;
> > +	u64 access;
> > +	u64 clock_cycles;
> > +	u64 read_access;
> > +	u64 write_access;
> >  };
> >  
> >  struct dmc_count {
> > @@ -69,6 +91,11 @@ struct rockchip_dfi {
> >  	struct devfreq_event_dev *edev;
> >  	struct devfreq_event_desc desc;
> >  	struct dmc_count last_event_count;
> > +
> > +	struct dmc_count last_perf_count;
> > +	struct dmc_count total_count;
> > +	seqlock_t count_seqlock; /* protects last_perf_count and total_count */
> > +
> >  	struct device *dev;
> >  	void __iomem *regs;
> >  	struct regmap *regmap_pmu;
> > @@ -77,6 +104,14 @@ struct rockchip_dfi {
> >  	struct mutex mutex;
> >  	u32 ddr_type;
> >  	unsigned int channel_mask;
> > +	enum cpuhp_state cpuhp_state;
> > +	struct hlist_node node;
> > +	struct pmu pmu;
> > +	struct hrtimer timer;
> > +	unsigned int cpu;
> > +	int active_events;
> > +	int burst_len;
> > +	int buswidth[DMC_MAX_CHANNELS];
> >  };
> >  
> >  static int rockchip_dfi_enable(struct rockchip_dfi *dfi)
> > @@ -145,7 +180,7 @@ static void rockchip_dfi_disable(struct rockchip_dfi *dfi)
> >  	mutex_unlock(&dfi->mutex);
> >  }
> >  
> > -static void rockchip_dfi_read_counters(struct rockchip_dfi *dfi, struct dmc_count *count)
> > +static void rockchip_dfi_read_counters(struct rockchip_dfi *dfi, struct dmc_count *c)
> 
> Actually, this change is not related to the patch's role which supports perf.
> Also, it is better to use 'res' argument name because rockchip_ddr_perf_counters_add()
> used the 'struct dmc_count *res' argument name.

Indeed the variable rename is not required here and for consistency
with rockchip_ddr_perf_counters_add() 'res' would be a better name.

Are you fine with renaming 'c' to 'res' in this patch or do you want me
to make a separate patch from the renaming?

> 
> 
> >  {
> >  	u32 i;
> >  	void __iomem *dfi_regs = dfi->regs;
> > @@ -153,13 +188,36 @@ static void rockchip_dfi_read_counters(struct rockchip_dfi *dfi, struct dmc_coun
> >  	for (i = 0; i < DMC_MAX_CHANNELS; i++) {
> >  		if (!(dfi->channel_mask & BIT(i)))
> >  			continue;
> > -		count->c[i].access = readl_relaxed(dfi_regs +
> > +		c->c[i].read_access = readl_relaxed(dfi_regs +
> > +				DDRMON_CH0_RD_NUM + i * 20);
> > +		c->c[i].write_access = readl_relaxed(dfi_regs +
> > +				DDRMON_CH0_WR_NUM + i * 20);
> > +		c->c[i].access = readl_relaxed(dfi_regs +
> >  				DDRMON_CH0_DFI_ACCESS_NUM + i * 20);
> > -		count->c[i].clock_cycles = readl_relaxed(dfi_regs +
> > +		c->c[i].clock_cycles = readl_relaxed(dfi_regs +
> >  				DDRMON_CH0_COUNT_NUM + i * 20);
> >  	}
> >  }
> >  
> > +static void rockchip_ddr_perf_counters_add(struct rockchip_dfi *dfi,
> > +					   const struct dmc_count *now,
> > +					   struct dmc_count *res)
> > +{
> > +	const struct dmc_count *last = &dfi->last_perf_count;
> > +	int i;
> > +
> > +	for (i = 0; i < DMC_MAX_CHANNELS; i++) {
> > +		res->c[i].read_access = dfi->total_count.c[i].read_access +
> > +			(u32)(now->c[i].read_access - last->c[i].read_access);
> > +		res->c[i].write_access = dfi->total_count.c[i].write_access +
> > +			(u32)(now->c[i].write_access - last->c[i].write_access);
> > +		res->c[i].access = dfi->total_count.c[i].access +
> > +			(u32)(now->c[i].access - last->c[i].access);
> > +		res->c[i].clock_cycles = dfi->total_count.c[i].clock_cycles +
> > +			(u32)(now->c[i].clock_cycles - last->c[i].clock_cycles);
> > +	}
> > +}
> > +
> >  static int rockchip_dfi_event_disable(struct devfreq_event_dev *edev)
> >  {
> >  	struct rockchip_dfi *dfi = devfreq_event_get_drvdata(edev);
> > @@ -223,6 +281,370 @@ static const struct devfreq_event_ops rockchip_dfi_ops = {
> >  	.set_event = rockchip_dfi_set_event,
> >  };
> >  

[...]

> > +static u64 rockchip_ddr_perf_event_get_count(struct perf_event *event)
> > +{
> > +	struct rockchip_dfi *dfi = container_of(event->pmu, struct rockchip_dfi, pmu);
> > +	int blen = dfi->burst_len;
> > +	struct dmc_count total, now;
> > +	unsigned int seq;
> > +	u64 c = 0;
> 
> Actually, it is difficult to understand the meaning of 'c' local variable name.
> Need to use the more clear vairable name instead of 'c'.

'c' is short for 'count' as in the function name xxx_get_count(). It is
initialized to 0, filled with values throughout the function and
returned at the end. Which other name do you suggest?

Sascha

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |