spmi-pmic-arb: add irq tracepoints to the pmic-arb driver

The spmi-pmic-arb is also an interrupt controller. It gets a single aggregate irq and disseminates it to individual pmic-peripheral drivers. Each pmic-peripheral has a unique apid number, and can have multiple interrupt capable functions. The registered apid range shows the lowest and highest apid numbers of pmic-peripheral drivers which request irqs. Pid is the base address of that peripheral. For performance measurement, tracepoints are added at the beginning of the aggregate irq and at the end of the individual pmic-peripheral irqs.

Following is a list showing the new tracepoint events:

spmi_pmic_arb_aggregate_irq_start: aggregate irq number and registered
apid range.

spmi_pmic_arb_apid_irq_end: apid, irq, func_num, sid and pid.

SPMI Interrupts tracepoints can be enabled like:

echo 1 >/sys/kernel/debug/tracing/events/spmi-pmic-arb/enable

and will dump messages that can be viewed in /sys/kernel/debug/tracing/trace that look like:
... spmi_pmic_arb_aggregate_irq_start: irq=150 registered apid range=(3,189) ... spmi_pmic_arb_apid_irq_end: apid=3 irq=1 func_num=0 sid=0 pid=0x8

Suggested-by Sagar Dharia <[email protected]>

Signed-off-by Gilad Avidov <[email protected]>

Signed-off-by Ankit Gupta <[email protected]>

---

Quoted 27 lines

The spmi-pmic-arb is also an interrupt controller. It gets a single aggregate irq and disseminates it to individual pmic-peripheral drivers. Each pmic-peripheral has a unique apid number, and can have multiple interrupt capable functions. The registered apid range shows the lowest and highest apid numbers of pmic-peripheral drivers which request irqs. Pid is the base address of that peripheral. For performance measurement, tracepoints are added at the beginning of the aggregate irq and at the end of the individual pmic-peripheral irqs.

Following is a list showing the new tracepoint events:

spmi_pmic_arb_aggregate_irq_start: aggregate irq number and registered
apid range.

spmi_pmic_arb_apid_irq_end: apid, irq, func_num, sid and pid.

SPMI Interrupts tracepoints can be enabled like:

echo 1 >/sys/kernel/debug/tracing/events/spmi-pmic-arb/enable

and will dump messages that can be viewed in /sys/kernel/debug/tracing/trace that look like:
... spmi_pmic_arb_aggregate_irq_start: irq=150 registered apid range=(3,189)
... spmi_pmic_arb_apid_irq_end: apid=3 irq=1 func_num=0 sid=0 pid=0x8

Quoted 4 lines

Suggested-by Sagar Dharia <[email protected]>

Signed-off-by Gilad Avidov <[email protected]>

Signed-off-by Ankit Gupta <[email protected]>

---

How is this any better than irq tracepoints that we already have for generic irqs?

--
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project

Quoted 4 lines

How is this any better than irq tracepoints that we already have for generic irqs?

It is better than generic irq tracepoints because it provides bus specific information (sid and address(pid) of slave write), driver specific information (apid (pmic-peripheral) and func_num) and statistics (apid range).
Recall that *slave* read/write cannot be traced by the spmi framework ftrace.

Quoted 11 lines · Ankit Gupta

Quoted 4 lines

How is this any better than irq tracepoints that we already have for generic irqs?

It is better than generic irq tracepoints because it provides bus specific information (sid and address(pid) of slave write), driver specific information (apid (pmic-peripheral) and func_num) and statistics (apid range).
Recall that *slave* read/write cannot be traced by the spmi framework ftrace.

Don't we already get all this information based on how we map interrupts to devices in DT? It feels to me that the same argument here could be applied to all the random gpio expanders and chained interrupt controllers that we support in the kernel.

Quoted 11 lines · On Wed, 27 May 2015 13:06:29 -0700 Stephen Boyd <[email protected]> wrote:

Quoted 10 lines · Ankit Gupta

Quoted 3 lines

How is this any better than irq tracepoints that we already have for generic irqs?

It is better than generic irq tracepoints because it provides bus specific information (sid and address(pid) of slave write), driver specific information (apid (pmic-peripheral) and func_num) and statistics (apid range).
Recall that *slave* read/write cannot be traced by the spmi framework ftrace.

Quoted 5 lines

Don't we already get all this information based on how we map interrupts to devices in DT? It feels to me that the same argument here could be applied to all the random gpio expanders and chained interrupt controllers that we support in the kernel.

We don't.
We could get the same information if we had the irq-domain and hw-irq 32bit value. While these values are available from /proc/interrupt, they are not available from the irq event tracing.

This patch traces spmi slave-originated events. If we had one such slave we could cross information from both sources to filter the trace events. However, we have hundreds of transaction-capable spmi slaves.

Quoted 19 lines · Gilad Avidov

Quoted 11 lines · On Wed, 27 May 2015 13:06:29 -0700 Stephen Boyd <[email protected]> wrote:

Quoted 10 lines · Ankit Gupta

Quoted 3 lines

How is this any better than irq tracepoints that we already have for generic irqs?

It is better than generic irq tracepoints because it provides bus specific information (sid and address(pid) of slave write), driver specific information (apid (pmic-peripheral) and func_num) and statistics (apid range).
Recall that *slave* read/write cannot be traced by the spmi framework ftrace.

Quoted 5 lines

Don't we already get all this information based on how we map interrupts to devices in DT? It feels to me that the same argument here could be applied to all the random gpio expanders and chained interrupt controllers that we support in the kernel.

Quoted 5 lines

We don't.
We could get the same information if we had the irq-domain and hw-irq 32bit value. While these values are available from /proc/interrupt, they are not available from the irq event tracing.

Hm.. maybe we should add those trace events then into the irq domain layer? Or even extend the irq tracepoints to include the hw-irq and irq-domain name?

Quoted 6 lines

This patch traces spmi slave-originated events. If we had one such slave we could cross information from both sources to filter the trace events. However, we have hundreds of transaction-capable spmi slaves.

Sorry I'm not following this point. If it was about tracing slave-originated events wouldn't it be more than irqs then?

The spmi-pmic-arb is also an interrupt controller. It gets a single aggregate irq and disseminates it to individual pmic-peripheral drivers. Each pmic-peripheral has a unique apid number, and can have multiple interrupt capable functions. The registered apid range shows the lowest and highest apid numbers of pmic-peripheral drivers which request irqs. Pid is the base address of that peripheral. For performance measurement, tracepoints are added at the beginning of the aggregate irq and at the end of the individual pmic-peripheral irqs.

Following is a list showing the new tracepoint events:

spmi_pmic_arb_aggregate_irq_start: aggregate irq number and registered
apid range.

spmi_pmic_arb_apid_irq_end: apid, irq, func_num, sid and pid.

SPMI Interrupts tracepoints can be enabled like:

echo 1 >/sys/kernel/debug/tracing/events/spmi-pmic-arb/enable

and will dump messages that can be viewed in /sys/kernel/debug/tracing/trace that look like:
... spmi_pmic_arb_aggregate_irq_start: irq=150 registered apid range=(3,189) ... spmi_pmic_arb_apid_irq_end: apid=3 irq=1 func_num=0 sid=0 pid=0x8

Suggested-by Sagar Dharia <[email protected]>

Signed-off-by Gilad Avidov <[email protected]>

Signed-off-by Ankit Gupta <[email protected]>

---
 drivers/spmi/spmi-pmic-arb.c         | 15 ++++++---
 include/trace/events/spmi-pmic-arb.h | 62 ++++++++++++++++++++++++++++++++++++
 2 files changed, 72 insertions(+), 5 deletions(-)
 create mode 100644 include/trace/events/spmi-pmic-arb.h

drivers/spmi/spmi-pmic-arb.c +6 -0

index 20559ab..342a71d 100644
@@ -23,6 +23,9 @@
 #include <linux/slab.h>
 #include <linux/spmi.h>
 
+#define CREATE_TRACE_POINTS
+#include <trace/events/spmi-pmic-arb.h>
+
 /* PMIC Arbiter configuration registers */
 #define PMIC_ARB_VERSION		0x0000
 #define PMIC_ARB_INT_EN			0x0004

@@ -375,16 +378,17 @@ static void periph_interrupt(struct spmi_pmic_arb_dev *pa, u8 apid)
 	unsigned int irq;
 	u32 status;
 	int id;
+	u16 ppid = pa->apid_to_ppid[apid];
+	u8 sid = (ppid >> 8) & 0x0F;
+	u8 pid = ppid & 0xFF;
 
 	status = readl_relaxed(pa->intr + SPMI_PIC_IRQ_STATUS(apid));

drivers/spmi/spmi-pmic-arb.c +2 -1

 	}
 }
 

@@ -399,7 +403,8 @@ static void pmic_arb_chained_irq(unsigned int irq, struct irq_desc *desc)
 	int i, id;
 
 	chained_irq_enter(chip, desc);
-
+	trace_spmi_pmic_arb_aggregate_irq_start(irq, pa->min_apid,
+						pa->max_apid);
 	for (i = first; i <= last; ++i) {
 		status = readl_relaxed(intr +
 				       SPMI_PIC_OWNER_ACC_STATUS(pa->ee, i));

include/trace/events/spmi-pmic-arb.h new file +62 -0

new file mode 100644
index 0000000..6c4dbca
@@ -0,0 +1,62 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM spmi-pmic-arb
+
+#if !defined(_TRACE_SPMI_PMIC_ARB_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_SPMI_PMIC_ARB_H
+
+#include <linux/spmi.h>
+#include <linux/tracepoint.h>
+
+/*
+ * drivers/spmi/spmi-pmic-arb.c
+ */
+
+TRACE_EVENT(spmi_pmic_arb_aggregate_irq_start,
+	TP_PROTO(unsigned int irq, int first, int last),
+	TP_ARGS(irq, first, last),
+
+	TP_STRUCT__entry(
+		__field		( unsigned int,   irq   )
+		__field		( int,            first )
+		__field		( int,            last  )
+	),
+
+	TP_fast_assign(
+		__entry->irq    = irq;
+		__entry->first  = first;
+		__entry->last   = last;
+	),
+
+	TP_printk("irq=%d registered apid range=(%d,%d)",
+		  (int)__entry->irq, __entry->first, __entry->last)
+);
+
+TRACE_EVENT(spmi_pmic_arb_apid_irq_end,
+	TP_PROTO(u8 apid, unsigned int irq, int func_num, u8 sid, u8 pid),
+	TP_ARGS(apid, irq, func_num, sid, pid),
+
+	TP_STRUCT__entry(
+		__field		( u8,           apid     )
+		__field		( unsigned int, irq      )
+		__field		( int,          func_num )
+		__field		( u8,           sid      )
+		__field		( u8,           pid      )
+	),
+
+	TP_fast_assign(
+		__entry->apid     = apid;
+		__entry->irq      = irq;
+		__entry->func_num = func_num;
+		__entry->sid      = sid;
+		__entry->pid      = pid;
+	),
+
+	TP_printk("apid=%d irq=%d func_num=%d sid=%d pid=0x%d",
+		  (int)__entry->apid, (int)__entry->irq, (int)__entry->func_num,
+		  (int)__entry->sid, (int)__entry->pid)
+);
+
+#endif /* _TRACE_SPMI_PMIC_ARB_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>

Quoted 61 lines · On Tue, 26 May 2015 17:39:15 -0600 Ankit Gupta <[email protected]> wrote:

The spmi-pmic-arb is also an interrupt controller. It gets a single aggregate irq and disseminates it to individual pmic-peripheral drivers. Each pmic-peripheral has a unique apid number, and can have multiple interrupt capable functions. The registered apid range shows the lowest and highest apid numbers of pmic-peripheral drivers which request irqs. Pid is the base address of that peripheral. For performance measurement, tracepoints are added at the beginning of the aggregate irq and at the end of the individual pmic-peripheral irqs.

Following is a list showing the new tracepoint events:

spmi_pmic_arb_aggregate_irq_start: aggregate irq number and registered
apid range.

spmi_pmic_arb_apid_irq_end: apid, irq, func_num, sid and pid.

SPMI Interrupts tracepoints can be enabled like:

echo 1 >/sys/kernel/debug/tracing/events/spmi-pmic-arb/enable

and will dump messages that can be viewed in /sys/kernel/debug/tracing/trace that look like:
... spmi_pmic_arb_aggregate_irq_start: irq=150 registered apid range=(3,189)
... spmi_pmic_arb_apid_irq_end: apid=3 irq=1 func_num=0 sid=0 pid=0x8

Suggested-by Sagar Dharia <[email protected]>

Signed-off-by Gilad Avidov <[email protected]>

Signed-off-by Ankit Gupta <[email protected]>

---
 drivers/spmi/spmi-pmic-arb.c         | 15 ++++++---
 include/trace/events/spmi-pmic-arb.h | 62
++++++++++++++++++++++++++++++++++++
 2 files changed, 72 insertions(+), 5 deletions(-)
 create mode 100644 include/trace/events/spmi-pmic-arb.h

drivers/spmi/spmi-pmic-arb.c +6 -0

index 20559ab..342a71d 100644
@@ -23,6 +23,9 @@
 #include <linux/slab.h>
 #include <linux/spmi.h>
 
+#define CREATE_TRACE_POINTS
+#include <trace/events/spmi-pmic-arb.h>
+
 /* PMIC Arbiter configuration registers */
 #define PMIC_ARB_VERSION		0x0000
 #define PMIC_ARB_INT_EN			0x0004

@@ -375,16 +378,17 @@ static void periph_interrupt(struct
 spmi_pmic_arb_dev *pa, u8 apid)
 	unsigned int irq;
 	u32 status;
 	int id;
+	u16 ppid = pa->apid_to_ppid[apid];
+	u8 sid = (ppid >> 8) & 0x0F;
+	u8 pid = ppid & 0xFF;
 
 	status = readl_relaxed(pa->intr + SPMI_PIC_IRQ_STATUS(apid));

drivers/spmi/spmi-pmic-arb.c static void periph_interrupt(struct · line 392

Quoted 10 lines

drivers/spmi/spmi-pmic-arb.c +2 -4

 	while (status) {
 		id = ffs(status) - 1;
 		status &= ~(1 << id);
-		irq = irq_find_mapping(pa->domain,
-				       pa->apid_to_ppid[apid] << 16
-				     | id << 8
-				     | apid);
+		irq = irq_find_mapping(pa->domain, ppid << 16 | id << 8 | apid);
 		generic_handle_irq(irq);
+		trace_spmi_pmic_arb_apid_irq_end(apid, irq, id, sid, pid);

It looks like sid is only used for the tracepoint processing. Instead of doing the work up above "(ppid >> 8) & 0x0F" that would only be used in the unlikely event that you happen to be tracing, what about moving the work below into the tracepoint, by passing in ppid, and having:

	__entry->sid = (ppid >> 8) & 0x0F;

That way you would save some CPU cycles when not tracing.

-- Steve

Quoted 75 lines

drivers/spmi/spmi-pmic-arb.c +2 -1

 	}
 }

@@ -399,7 +403,8 @@ static void pmic_arb_chained_irq(unsigned int irq,
 struct irq_desc *desc)
 	int i, id;
 
 	chained_irq_enter(chip, desc);
-
+	trace_spmi_pmic_arb_aggregate_irq_start(irq, pa->min_apid,
+						pa->max_apid);
 	for (i = first; i <= last; ++i) {
 		status = readl_relaxed(intr +

drivers/spmi/spmi-pmic-arb.c +0 -0

 				       SPMI_PIC_OWNER_ACC_STATUS(pa->ee, i));

include/trace/events/spmi-pmic-arb.h new file +4 -0

new file mode 100644
index 0000000..6c4dbca
@@ -0,0 +1,62 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM spmi-pmic-arb
+
+#if !defined(_TRACE_SPMI_PMIC_ARB_H) ||

defined(TRACE_HEADER_MULTI_READ)

include/trace/events/spmi-pmic-arb.h new file +48 -0

+#define _TRACE_SPMI_PMIC_ARB_H
+
+#include <linux/spmi.h>
+#include <linux/tracepoint.h>
+
+/*
+ * drivers/spmi/spmi-pmic-arb.c
+ */
+
+TRACE_EVENT(spmi_pmic_arb_aggregate_irq_start,
+	TP_PROTO(unsigned int irq, int first, int last),
+	TP_ARGS(irq, first, last),
+
+	TP_STRUCT__entry(
+		__field		( unsigned int,   irq   )
+		__field		( int,            first )
+		__field		( int,            last  )
+	),
+
+	TP_fast_assign(
+		__entry->irq    = irq;
+		__entry->first  = first;
+		__entry->last   = last;
+	),
+
+	TP_printk("irq=%d registered apid range=(%d,%d)",
+		  (int)__entry->irq, __entry->first, __entry->last)
+);
+
+TRACE_EVENT(spmi_pmic_arb_apid_irq_end,
+	TP_PROTO(u8 apid, unsigned int irq, int func_num, u8 sid, u8 pid),
+	TP_ARGS(apid, irq, func_num, sid, pid),
+
+	TP_STRUCT__entry(
+		__field		( u8,           apid     )
+		__field		( unsigned int, irq      )
+		__field		( int,          func_num )
+		__field		( u8,           sid      )
+		__field		( u8,           pid      )
+	),
+
+	TP_fast_assign(
+		__entry->apid     = apid;
+		__entry->irq      = irq;
+		__entry->func_num = func_num;
+		__entry->sid      = sid;
+		__entry->pid      = pid;
+	),

include/trace/events/spmi-pmic-arb.h line 62

Quoted 10 lines

include/trace/events/spmi-pmic-arb.h new file +10 -0

+
+	TP_printk("apid=%d irq=%d func_num=%d sid=%d pid=0x%d",
+		  (int)__entry->apid, (int)__entry->irq, (int)__entry->func_num,
+		  (int)__entry->sid, (int)__entry->pid)
+);
+
+#endif /* _TRACE_SPMI_PMIC_ARB_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>

--
To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html