[PATCH] Add minimal support for software performance counters for ia64

[PATCH] Add minimal support for software performance counters for ia64

[William Cohen]

[-- Attachment #1: Type: text/plain, Size: 777 bytes --]

The following patch adds minimal support for software performance counters for ia64.

For the kernel the patch enables configuration of the perf counter
option, adds the perf_counter_open syscall, and includes a minimal machine
specific asm/perf.h header file.

For the perf tool the header perf.h includes an ia64 specific section
and the options used in the Makefile were adjusted to allow
compilation on ia64.  The -Wcast-align gives "cast increases required
alignment of target type" warning for the list_for_each_entry()
macro. The "-fno-strict-aliasing" was used to avoid warnings for
"dereferencing type-punned pointer will break strict-aliasing rules"
warnings for sscanf() functions in util/trace-event-parse.c

-Will

Signed-off-by: William Cohen <wcohen@redhat.com>

[-- Attachment #2: perf_ia64a.patch --]
[-- Type: text/plain, Size: 3998 bytes --]

diff --git a/arch/ia64/Kconfig b/arch/ia64/Kconfig
index 170042b..8471526 100644
--- a/arch/ia64/Kconfig
+++ b/arch/ia64/Kconfig
@@ -29,6 +29,7 @@ config IA64
 	select HAVE_KVM
 	select HAVE_ARCH_TRACEHOOK
 	select HAVE_DMA_API_DEBUG
+	select HAVE_PERF_COUNTERS
 	default y
 	help
 	  The Itanium Processor Family is Intel's 64-bit successor to
diff --git a/arch/ia64/include/asm/unistd.h b/arch/ia64/include/asm/unistd.h
index 5a5347f..f69d1c9 100644
--- a/arch/ia64/include/asm/unistd.h
+++ b/arch/ia64/include/asm/unistd.h
@@ -311,11 +311,12 @@
 #define __NR_preadv			1319
 #define __NR_pwritev			1320
 #define __NR_rt_tgsigqueueinfo		1321
+#define __NR_perf_counter_open		1322
 
 #ifdef __KERNEL__
 
 
-#define NR_syscalls			298 /* length of syscall table */
+#define NR_syscalls			299 /* length of syscall table */
 
 /*
  * The following defines stop scripts/checksyscalls.sh from complaining about
diff --git a/arch/ia64/kernel/entry.S b/arch/ia64/kernel/entry.S
index d0e7d37..4cf88b0 100644
--- a/arch/ia64/kernel/entry.S
+++ b/arch/ia64/kernel/entry.S
@@ -1806,6 +1806,7 @@ sys_call_table:
 	data8 sys_preadv
 	data8 sys_pwritev			// 1320
 	data8 sys_rt_tgsigqueueinfo
+	data8 sys_perf_counter_open
 
 	.org sys_call_table + 8*NR_syscalls	// guard against failures to increase NR_syscalls
 #endif /* __IA64_ASM_PARAVIRTUALIZED_NATIVE */
diff --git a/include/asm-ia64/perf_counter.h b/include/asm-ia64/perf_counter.h
new file mode 100644
index 0000000..cab4352
--- /dev/null
+++ b/include/asm-ia64/perf_counter.h
@@ -0,0 +1,9 @@
+#ifndef __ASM_IA64_PERF_COUNTER_H
+#define __ASM_IA64_PERF_COUNTER_H
+
+/* IA64 only supports software counters through this interface. */
+static inline void set_perf_counter_pending(void) {}
+
+#define PERF_COUNTER_INDEX_OFFSET	0
+
+#endif /* __ASM_IA64_PERF_COUNTER_H */
diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index c481a51..69db3de 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -159,8 +159,10 @@ uname_V := $(shell sh -c 'uname -v 2>/dev/null || echo not')
 
 # If we're on a 64-bit kernel, use -m64
 ifndef NO_64BIT
-	ifneq ($(patsubst %64,%,$(uname_M)),$(uname_M))
-	  M64 := -m64
+	ifeq ($(patsubst %ia64,%,$(uname_M)),$(uname_M))
+	  ifneq ($(patsubst %64,%,$(uname_M)),$(uname_M))
+	    M64 := -m64
+	  endif
 	endif
 endif
 
@@ -169,9 +171,9 @@ endif
 #
 # Include saner warnings here, which can catch bugs:
 #
-EXTRA_WARNINGS = -Wcast-align -Wformat -Wformat-security -Wformat-y2k -Wshadow -Winit-self -Wpacked -Wredundant-decls -Wstack-protector -Wstrict-aliasing=3 -Wswitch-default -Wswitch-enum -Wno-system-headers -Wundef -Wvolatile-register-var -Wwrite-strings -Wbad-function-cast -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Wold-style-definition -Wstrict-prototypes -Wdeclaration-after-statement
+EXTRA_WARNINGS = -Wformat -Wformat-security -Wformat-y2k -Wshadow -Winit-self -Wpacked -Wredundant-decls -Wstack-protector -Wstrict-aliasing=3 -Wswitch-default -Wswitch-enum -Wno-system-headers -Wundef -Wvolatile-register-var -Wwrite-strings -Wbad-function-cast -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Wold-style-definition -Wstrict-prototypes -Wdeclaration-after-statement
 
-CFLAGS = $(M64) -ggdb3 -Wall -Wextra -std=gnu99 -Werror -O6 -fstack-protector-all -D_FORTIFY_SOURCE=2 $(EXTRA_WARNINGS)
+CFLAGS = $(M64) -ggdb3 -Wall -Wextra -std=gnu99 -Werror -O6 -fstack-protector-all -D_FORTIFY_SOURCE=2 -fno-strict-aliasing $(EXTRA_WARNINGS)
 LDFLAGS = -lpthread -lrt -lelf -lm
 ALL_CFLAGS = $(CFLAGS)
 ALL_LDFLAGS = $(LDFLAGS)
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index e5148e2..90179c9 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -41,6 +41,12 @@
 #define cpu_relax()	asm volatile("" ::: "memory");
 #endif
 
+#ifdef __ia64__
+#include "../../arch/ia64/include/asm/unistd.h"
+#define rmb()		asm volatile("" ::: "memory")
+#define cpu_relax()	asm volatile("" ::: "memory");
+#endif
+
 #include <time.h>
 #include <unistd.h>
 #include <sys/types.h>

Re: [PATCH] Add minimal support for software performance counters for ia64

[Peter Zijlstra]

On Wed, 2009-08-19 at 12:04 -0400, William Cohen wrote:
> +#ifdef __ia64__
> +#include "../../arch/ia64/include/asm/unistd.h"
> +#define rmb()          asm volatile("" ::: "memory")
> +#define cpu_relax()    asm volatile("" ::: "memory");
> +#endif

Surely the itanic has a real memory barrier? The kernel seems to use
something along the lines of mf.

Also, cpu_relax() seems to end up being asm volatile("hint @pause" :::
"memory"), although we don't seem to have an actual user of cpu_relax()
in perf atm.

Re: [PATCH] Add minimal support for software performance counters for ia64

[William Cohen]

[-- Attachment #1: Type: text/plain, Size: 713 bytes --]

Peter Zijlstra wrote:
> On Wed, 2009-08-19 at 12:04 -0400, William Cohen wrote:
>> +#ifdef __ia64__
>> +#include "../../arch/ia64/include/asm/unistd.h"
>> +#define rmb()          asm volatile("" ::: "memory")
>> +#define cpu_relax()    asm volatile("" ::: "memory");
>> +#endif
> 
> Surely the itanic has a real memory barrier? The kernel seems to use
> something along the lines of mf.
> 
> Also, cpu_relax() seems to end up being asm volatile("hint @pause" :::
> "memory"), although we don't seem to have an actual user of cpu_relax()
> in perf atm.
> 
> 
> 

Peter, thanks for the comment. I took a look at the rmb() and cpu_relax() code
in the kernel and revised the patch with the appropriate defines

-Will

[-- Attachment #2: perf_ia64b.patch --]
[-- Type: text/plain, Size: 4010 bytes --]

diff --git a/arch/ia64/Kconfig b/arch/ia64/Kconfig
index 170042b..8471526 100644
--- a/arch/ia64/Kconfig
+++ b/arch/ia64/Kconfig
@@ -29,6 +29,7 @@ config IA64
 	select HAVE_KVM
 	select HAVE_ARCH_TRACEHOOK
 	select HAVE_DMA_API_DEBUG
+	select HAVE_PERF_COUNTERS
 	default y
 	help
 	  The Itanium Processor Family is Intel's 64-bit successor to
diff --git a/arch/ia64/include/asm/unistd.h b/arch/ia64/include/asm/unistd.h
index 5a5347f..f69d1c9 100644
--- a/arch/ia64/include/asm/unistd.h
+++ b/arch/ia64/include/asm/unistd.h
@@ -311,11 +311,12 @@
 #define __NR_preadv			1319
 #define __NR_pwritev			1320
 #define __NR_rt_tgsigqueueinfo		1321
+#define __NR_perf_counter_open		1322
 
 #ifdef __KERNEL__
 
 
-#define NR_syscalls			298 /* length of syscall table */
+#define NR_syscalls			299 /* length of syscall table */
 
 /*
  * The following defines stop scripts/checksyscalls.sh from complaining about
diff --git a/arch/ia64/kernel/entry.S b/arch/ia64/kernel/entry.S
index d0e7d37..4cf88b0 100644
--- a/arch/ia64/kernel/entry.S
+++ b/arch/ia64/kernel/entry.S
@@ -1806,6 +1806,7 @@ sys_call_table:
 	data8 sys_preadv
 	data8 sys_pwritev			// 1320
 	data8 sys_rt_tgsigqueueinfo
+	data8 sys_perf_counter_open
 
 	.org sys_call_table + 8*NR_syscalls	// guard against failures to increase NR_syscalls
 #endif /* __IA64_ASM_PARAVIRTUALIZED_NATIVE */
diff --git a/include/asm-ia64/perf_counter.h b/include/asm-ia64/perf_counter.h
new file mode 100644
index 0000000..cab4352
--- /dev/null
+++ b/include/asm-ia64/perf_counter.h
@@ -0,0 +1,9 @@
+#ifndef __ASM_IA64_PERF_COUNTER_H
+#define __ASM_IA64_PERF_COUNTER_H
+
+/* IA64 only supports software counters through this interface. */
+static inline void set_perf_counter_pending(void) {}
+
+#define PERF_COUNTER_INDEX_OFFSET	0
+
+#endif /* __ASM_IA64_PERF_COUNTER_H */
diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index c481a51..69db3de 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -159,8 +159,10 @@ uname_V := $(shell sh -c 'uname -v 2>/dev/null || echo not')
 
 # If we're on a 64-bit kernel, use -m64
 ifndef NO_64BIT
-	ifneq ($(patsubst %64,%,$(uname_M)),$(uname_M))
-	  M64 := -m64
+	ifeq ($(patsubst %ia64,%,$(uname_M)),$(uname_M))
+	  ifneq ($(patsubst %64,%,$(uname_M)),$(uname_M))
+	    M64 := -m64
+	  endif
 	endif
 endif
 
@@ -169,9 +171,9 @@ endif
 #
 # Include saner warnings here, which can catch bugs:
 #
-EXTRA_WARNINGS = -Wcast-align -Wformat -Wformat-security -Wformat-y2k -Wshadow -Winit-self -Wpacked -Wredundant-decls -Wstack-protector -Wstrict-aliasing=3 -Wswitch-default -Wswitch-enum -Wno-system-headers -Wundef -Wvolatile-register-var -Wwrite-strings -Wbad-function-cast -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Wold-style-definition -Wstrict-prototypes -Wdeclaration-after-statement
+EXTRA_WARNINGS = -Wformat -Wformat-security -Wformat-y2k -Wshadow -Winit-self -Wpacked -Wredundant-decls -Wstack-protector -Wstrict-aliasing=3 -Wswitch-default -Wswitch-enum -Wno-system-headers -Wundef -Wvolatile-register-var -Wwrite-strings -Wbad-function-cast -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Wold-style-definition -Wstrict-prototypes -Wdeclaration-after-statement
 
-CFLAGS = $(M64) -ggdb3 -Wall -Wextra -std=gnu99 -Werror -O6 -fstack-protector-all -D_FORTIFY_SOURCE=2 $(EXTRA_WARNINGS)
+CFLAGS = $(M64) -ggdb3 -Wall -Wextra -std=gnu99 -Werror -O6 -fstack-protector-all -D_FORTIFY_SOURCE=2 -fno-strict-aliasing $(EXTRA_WARNINGS)
 LDFLAGS = -lpthread -lrt -lelf -lm
 ALL_CFLAGS = $(CFLAGS)
 ALL_LDFLAGS = $(LDFLAGS)
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index e5148e2..c849337 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -41,6 +41,12 @@
 #define cpu_relax()	asm volatile("" ::: "memory");
 #endif
 
+#ifdef __ia64__
+#include "../../arch/ia64/include/asm/unistd.h"
+#define rmb()		asm volatile("mf" ::: "memory")
+#define cpu_relax()	asm volatile("hint @pause" ::: "memory")
+#endif
+
 #include <time.h>
 #include <unistd.h>
 #include <sys/types.h>

Re: [PATCH] Add minimal support for software performance counters for ia64

[Peter Zijlstra]

On Wed, 2009-08-19 at 15:04 -0400, William Cohen wrote:
> Peter Zijlstra wrote:
> > On Wed, 2009-08-19 at 12:04 -0400, William Cohen wrote:
> >> +#ifdef __ia64__
> >> +#include "../../arch/ia64/include/asm/unistd.h"
> >> +#define rmb()          asm volatile("" ::: "memory")
> >> +#define cpu_relax()    asm volatile("" ::: "memory");
> >> +#endif
> > 
> > Surely the itanic has a real memory barrier? The kernel seems to use
> > something along the lines of mf.
> > 
> > Also, cpu_relax() seems to end up being asm volatile("hint @pause" :::
> > "memory"), although we don't seem to have an actual user of cpu_relax()
> > in perf atm.
> > 
> > 
> > 
> 
> Peter, thanks for the comment. I took a look at the rmb() and cpu_relax() code
> in the kernel and revised the patch with the appropriate defines

Looks good now, thanks!

Re: [PATCH] Add minimal support for software performance counters for ia64

[William Cohen]

Peter Zijlstra wrote:
> On Wed, 2009-08-19 at 15:04 -0400, William Cohen wrote:
>> Peter Zijlstra wrote:
>>> On Wed, 2009-08-19 at 12:04 -0400, William Cohen wrote:
>>>> +#ifdef __ia64__
>>>> +#include "../../arch/ia64/include/asm/unistd.h"
>>>> +#define rmb()          asm volatile("" ::: "memory")
>>>> +#define cpu_relax()    asm volatile("" ::: "memory");
>>>> +#endif
>>> Surely the itanic has a real memory barrier? The kernel seems to use
>>> something along the lines of mf.
>>>
>>> Also, cpu_relax() seems to end up being asm volatile("hint @pause" :::
>>> "memory"), although we don't seem to have an actual user of cpu_relax()
>>> in perf atm.
>>>
>>>
>>>
>> Peter, thanks for the comment. I took a look at the rmb() and cpu_relax() code
>> in the kernel and revised the patch with the appropriate defines
> 
> Looks good now, thanks!
> 

When looking through the patch today I realized the new perf_counter.h was
placed in include/asm-ia64. It should be placed in
arch/ia64/include/asm/perf_counter.h.

-Will

Re: [PATCH] Add minimal support for software performance counters for ia64

[Ingo Molnar]

* William Cohen <wcohen@redhat.com> wrote:

> Peter Zijlstra wrote:
> > On Wed, 2009-08-19 at 15:04 -0400, William Cohen wrote:
> >> Peter Zijlstra wrote:
> >>> On Wed, 2009-08-19 at 12:04 -0400, William Cohen wrote:
> >>>> +#ifdef __ia64__
> >>>> +#include "../../arch/ia64/include/asm/unistd.h"
> >>>> +#define rmb()          asm volatile("" ::: "memory")
> >>>> +#define cpu_relax()    asm volatile("" ::: "memory");
> >>>> +#endif
> >>> Surely the itanic has a real memory barrier? The kernel seems to use
> >>> something along the lines of mf.
> >>>
> >>> Also, cpu_relax() seems to end up being asm volatile("hint @pause" :::
> >>> "memory"), although we don't seem to have an actual user of cpu_relax()
> >>> in perf atm.
> >>>
> >>>
> >>>
> >> Peter, thanks for the comment. I took a look at the rmb() and cpu_relax() code
> >> in the kernel and revised the patch with the appropriate defines
> > 
> > Looks good now, thanks!
> > 
> 
> When looking through the patch today I realized the new 
> perf_counter.h was placed in include/asm-ia64. It should be placed 
> in arch/ia64/include/asm/perf_counter.h.

mind sending out a v2 patch with all review feedback folded in, for 
IA64 maintainers to ack if they like it?

Thanks,

	Ingo

Re: [PATCH] Add minimal support for software performance counters for ia64

[William Cohen]

[-- Attachment #1: Type: text/plain, Size: 1445 bytes --]

Ingo Molnar wrote:
> * William Cohen <wcohen@redhat.com> wrote:
> 
>> Peter Zijlstra wrote:
>>> On Wed, 2009-08-19 at 15:04 -0400, William Cohen wrote:
>>>> Peter Zijlstra wrote:
>>>>> On Wed, 2009-08-19 at 12:04 -0400, William Cohen wrote:
>>>>>> +#ifdef __ia64__
>>>>>> +#include "../../arch/ia64/include/asm/unistd.h"
>>>>>> +#define rmb()          asm volatile("" ::: "memory")
>>>>>> +#define cpu_relax()    asm volatile("" ::: "memory");
>>>>>> +#endif
>>>>> Surely the itanic has a real memory barrier? The kernel seems to use
>>>>> something along the lines of mf.
>>>>>
>>>>> Also, cpu_relax() seems to end up being asm volatile("hint @pause" :::
>>>>> "memory"), although we don't seem to have an actual user of cpu_relax()
>>>>> in perf atm.
>>>>>
>>>>>
>>>>>
>>>> Peter, thanks for the comment. I took a look at the rmb() and cpu_relax() code
>>>> in the kernel and revised the patch with the appropriate defines
>>> Looks good now, thanks!
>>>
>> When looking through the patch today I realized the new 
>> perf_counter.h was placed in include/asm-ia64. It should be placed 
>> in arch/ia64/include/asm/perf_counter.h.
> 
> mind sending out a v2 patch with all review feedback folded in, for 
> IA64 maintainers to ack if they like it?
> 
> Thanks,
> 
> 	Ingo

Hi Ingo,

The following is the revised patch with the corrected macros and the include
file in the correct place.

Signed-off-by: William Cohen <wcohen@redhat.com>

-Will

[-- Attachment #2: perf_ia64c.patch --]
[-- Type: text/plain, Size: 4025 bytes --]

diff --git a/arch/ia64/Kconfig b/arch/ia64/Kconfig
index 170042b..8471526 100644
--- a/arch/ia64/Kconfig
+++ b/arch/ia64/Kconfig
@@ -29,6 +29,7 @@ config IA64
 	select HAVE_KVM
 	select HAVE_ARCH_TRACEHOOK
 	select HAVE_DMA_API_DEBUG
+	select HAVE_PERF_COUNTERS
 	default y
 	help
 	  The Itanium Processor Family is Intel's 64-bit successor to
diff --git a/arch/ia64/include/asm/unistd.h b/arch/ia64/include/asm/unistd.h
index 5a5347f..f69d1c9 100644
--- a/arch/ia64/include/asm/unistd.h
+++ b/arch/ia64/include/asm/unistd.h
@@ -311,11 +311,12 @@
 #define __NR_preadv			1319
 #define __NR_pwritev			1320
 #define __NR_rt_tgsigqueueinfo		1321
+#define __NR_perf_counter_open		1322
 
 #ifdef __KERNEL__
 
 
-#define NR_syscalls			298 /* length of syscall table */
+#define NR_syscalls			299 /* length of syscall table */
 
 /*
  * The following defines stop scripts/checksyscalls.sh from complaining about
diff --git a/arch/ia64/kernel/entry.S b/arch/ia64/kernel/entry.S
index d0e7d37..4cf88b0 100644
--- a/arch/ia64/kernel/entry.S
+++ b/arch/ia64/kernel/entry.S
@@ -1806,6 +1806,7 @@ sys_call_table:
 	data8 sys_preadv
 	data8 sys_pwritev			// 1320
 	data8 sys_rt_tgsigqueueinfo
+	data8 sys_perf_counter_open
 
 	.org sys_call_table + 8*NR_syscalls	// guard against failures to increase NR_syscalls
 #endif /* __IA64_ASM_PARAVIRTUALIZED_NATIVE */
diff --git a/arch/ia64/include/asm/perf_counter.h b/arch/ia64/include/asm/perf_counter.h
new file mode 100644
index 0000000..cab4352
--- /dev/null
+++ b/arch/ia64/include/asm/perf_counter.h
@@ -0,0 +1,9 @@
+#ifndef __ASM_IA64_PERF_COUNTER_H
+#define __ASM_IA64_PERF_COUNTER_H
+
+/* IA64 only supports software counters through this interface. */
+static inline void set_perf_counter_pending(void) {}
+
+#define PERF_COUNTER_INDEX_OFFSET	0
+
+#endif /* __ASM_IA64_PERF_COUNTER_H */
diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index c481a51..69db3de 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -159,8 +159,10 @@ uname_V := $(shell sh -c 'uname -v 2>/dev/null || echo not')
 
 # If we're on a 64-bit kernel, use -m64
 ifndef NO_64BIT
-	ifneq ($(patsubst %64,%,$(uname_M)),$(uname_M))
-	  M64 := -m64
+	ifeq ($(patsubst %ia64,%,$(uname_M)),$(uname_M))
+	  ifneq ($(patsubst %64,%,$(uname_M)),$(uname_M))
+	    M64 := -m64
+	  endif
 	endif
 endif
 
@@ -169,9 +171,9 @@ endif
 #
 # Include saner warnings here, which can catch bugs:
 #
-EXTRA_WARNINGS = -Wcast-align -Wformat -Wformat-security -Wformat-y2k -Wshadow -Winit-self -Wpacked -Wredundant-decls -Wstack-protector -Wstrict-aliasing=3 -Wswitch-default -Wswitch-enum -Wno-system-headers -Wundef -Wvolatile-register-var -Wwrite-strings -Wbad-function-cast -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Wold-style-definition -Wstrict-prototypes -Wdeclaration-after-statement
+EXTRA_WARNINGS = -Wformat -Wformat-security -Wformat-y2k -Wshadow -Winit-self -Wpacked -Wredundant-decls -Wstack-protector -Wstrict-aliasing=3 -Wswitch-default -Wswitch-enum -Wno-system-headers -Wundef -Wvolatile-register-var -Wwrite-strings -Wbad-function-cast -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Wold-style-definition -Wstrict-prototypes -Wdeclaration-after-statement
 
-CFLAGS = $(M64) -ggdb3 -Wall -Wextra -std=gnu99 -Werror -O6 -fstack-protector-all -D_FORTIFY_SOURCE=2 $(EXTRA_WARNINGS)
+CFLAGS = $(M64) -ggdb3 -Wall -Wextra -std=gnu99 -Werror -O6 -fstack-protector-all -D_FORTIFY_SOURCE=2 -fno-strict-aliasing $(EXTRA_WARNINGS)
 LDFLAGS = -lpthread -lrt -lelf -lm
 ALL_CFLAGS = $(CFLAGS)
 ALL_LDFLAGS = $(LDFLAGS)
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index e5148e2..c849337 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -41,6 +41,12 @@
 #define cpu_relax()	asm volatile("" ::: "memory");
 #endif
 
+#ifdef __ia64__
+#include "../../arch/ia64/include/asm/unistd.h"
+#define rmb()		asm volatile("mf" ::: "memory")
+#define cpu_relax()	asm volatile("hint @pause" ::: "memory")
+#endif
+
 #include <time.h>
 #include <unistd.h>
 #include <sys/types.h>

[PATCH, V2] perf_counter, IA64: Add support for software performance counters

[Ingo Molnar]

* William Cohen <wcohen@redhat.com> wrote:

> The following is the revised patch with the corrected macros and 
> the include file in the correct place.

Note, this is not how v2 patches are generally expected to be 
submitted upstream - all info is supposed to be re-submitted with a 
new v2 email title to make it easy and context-less for maintainers.

I've done that here: i've tidied up the changelog (to match IA64 and 
perf_counter patch title standards), added a diffstat, added the 
acks/signoffs the patch got in the discussion and included the whole 
thing below, for IA64 folks to comment on.

Thanks,

	Ingo

---------------------->
Subject: perf_counter, IA64: Add support for software performance counters
From: William Cohen <wcohen@redhat.com>
Date: Fri, 21 Aug 2009 11:09:29 -0400

The patch enables configuration of the perf counter option on
IA64, adds the perf_counter_open syscall, and includes a
minimal machine specific asm/perf.h header file.

For the perf tool the header perf.h includes an ia64 specific
section and the options used in the Makefile were adjusted to
allow compilation on ia64.

[ The -Wcast-align gives "cast increases required alignment of
  target type" warning for the list_for_each_entry() macro. The
  "-fno-strict-aliasing" was used to avoid warnings for
  "dereferencing type-punned pointer will break strict-aliasing
  rules" warnings for sscanf() functions in
  util/trace-event-parse.c ]

Signed-off-by: William Cohen <wcohen@redhat.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <4A8EB8A9.9000900@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 arch/ia64/Kconfig                    |    1 +
 arch/ia64/include/asm/perf_counter.h |    9 +++++++++
 arch/ia64/include/asm/unistd.h       |    3 ++-
 arch/ia64/kernel/entry.S             |    1 +
 tools/perf/Makefile                  |   10 ++++++----
 tools/perf/perf.h                    |    6 ++++++
 6 files changed, 25 insertions(+), 5 deletions(-)

Index: tip/arch/ia64/Kconfig
===================================================================
--- tip.orig/arch/ia64/Kconfig
+++ tip/arch/ia64/Kconfig
@@ -29,6 +29,7 @@ config IA64
 	select HAVE_KVM
 	select HAVE_ARCH_TRACEHOOK
 	select HAVE_DMA_API_DEBUG
+	select HAVE_PERF_COUNTERS
 	default y
 	help
 	  The Itanium Processor Family is Intel's 64-bit successor to
Index: tip/arch/ia64/include/asm/perf_counter.h
===================================================================
--- /dev/null
+++ tip/arch/ia64/include/asm/perf_counter.h
@@ -0,0 +1,9 @@
+#ifndef __ASM_IA64_PERF_COUNTER_H
+#define __ASM_IA64_PERF_COUNTER_H
+
+/* IA64 only supports software counters through this interface. */
+static inline void set_perf_counter_pending(void) {}
+
+#define PERF_COUNTER_INDEX_OFFSET	0
+
+#endif /* __ASM_IA64_PERF_COUNTER_H */
Index: tip/arch/ia64/include/asm/unistd.h
===================================================================
--- tip.orig/arch/ia64/include/asm/unistd.h
+++ tip/arch/ia64/include/asm/unistd.h
@@ -311,11 +311,12 @@
 #define __NR_preadv			1319
 #define __NR_pwritev			1320
 #define __NR_rt_tgsigqueueinfo		1321
+#define __NR_perf_counter_open		1322
 
 #ifdef __KERNEL__
 
 
-#define NR_syscalls			298 /* length of syscall table */
+#define NR_syscalls			299 /* length of syscall table */
 
 /*
  * The following defines stop scripts/checksyscalls.sh from complaining about
Index: tip/arch/ia64/kernel/entry.S
===================================================================
--- tip.orig/arch/ia64/kernel/entry.S
+++ tip/arch/ia64/kernel/entry.S
@@ -1806,6 +1806,7 @@ sys_call_table:
 	data8 sys_preadv
 	data8 sys_pwritev			// 1320
 	data8 sys_rt_tgsigqueueinfo
+	data8 sys_perf_counter_open
 
 	.org sys_call_table + 8*NR_syscalls	// guard against failures to increase NR_syscalls
 #endif /* __IA64_ASM_PARAVIRTUALIZED_NATIVE */
Index: tip/tools/perf/Makefile
===================================================================
--- tip.orig/tools/perf/Makefile
+++ tip/tools/perf/Makefile
@@ -159,8 +159,10 @@ uname_V := $(shell sh -c 'uname -v 2>/de
 
 # If we're on a 64-bit kernel, use -m64
 ifndef NO_64BIT
-	ifneq ($(patsubst %64,%,$(uname_M)),$(uname_M))
-	  M64 := -m64
+	ifeq ($(patsubst %ia64,%,$(uname_M)),$(uname_M))
+	  ifneq ($(patsubst %64,%,$(uname_M)),$(uname_M))
+	    M64 := -m64
+	  endif
 	endif
 endif
 
@@ -169,9 +171,9 @@ endif
 #
 # Include saner warnings here, which can catch bugs:
 #
-EXTRA_WARNINGS = -Wcast-align -Wformat -Wformat-security -Wformat-y2k -Wshadow -Winit-self -Wpacked -Wredundant-decls -Wstack-protector -Wstrict-aliasing=3 -Wswitch-default -Wswitch-enum -Wno-system-headers -Wundef -Wvolatile-register-var -Wwrite-strings -Wbad-function-cast -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Wold-style-definition -Wstrict-prototypes -Wdeclaration-after-statement
+EXTRA_WARNINGS = -Wformat -Wformat-security -Wformat-y2k -Wshadow -Winit-self -Wpacked -Wredundant-decls -Wstack-protector -Wstrict-aliasing=3 -Wswitch-default -Wswitch-enum -Wno-system-headers -Wundef -Wvolatile-register-var -Wwrite-strings -Wbad-function-cast -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Wold-style-definition -Wstrict-prototypes -Wdeclaration-after-statement
 
-CFLAGS = $(M64) -ggdb3 -Wall -Wextra -std=gnu99 -Werror -O6 -fstack-protector-all -D_FORTIFY_SOURCE=2 $(EXTRA_WARNINGS)
+CFLAGS = $(M64) -ggdb3 -Wall -Wextra -std=gnu99 -Werror -O6 -fstack-protector-all -D_FORTIFY_SOURCE=2 -fno-strict-aliasing $(EXTRA_WARNINGS)
 LDFLAGS = -lpthread -lrt -lelf -lm
 ALL_CFLAGS = $(CFLAGS)
 ALL_LDFLAGS = $(LDFLAGS)
Index: tip/tools/perf/perf.h
===================================================================
--- tip.orig/tools/perf/perf.h
+++ tip/tools/perf/perf.h
@@ -41,6 +41,12 @@
 #define cpu_relax()	asm volatile("" ::: "memory");
 #endif
 
+#ifdef __ia64__
+#include "../../arch/ia64/include/asm/unistd.h"
+#define rmb()		asm volatile("mf" ::: "memory")
+#define cpu_relax()	asm volatile("hint @pause" ::: "memory")
+#endif
+
 #include <time.h>
 #include <unistd.h>
 #include <sys/types.h>

Re: [PATCH] Add minimal support for software performance counters

[Ingo Molnar]

* William Cohen <wcohen@redhat.com> wrote:

> The following patch adds minimal support for software 
> performance counters for ia64.

Nice!

> For the kernel the patch enables configuration of the 
> perf counter option, adds the perf_counter_open 
> syscall, and includes a minimal machine specific 
> asm/perf.h header file.
> 
> For the perf tool the header perf.h includes an ia64 
> specific section and the options used in the Makefile 
> were adjusted to allow compilation on ia64.  The 
> -Wcast-align gives "cast increases required alignment 
> of target type" warning for the list_for_each_entry() 
> macro. The "-fno-strict-aliasing" was used to avoid 
> warnings for "dereferencing type-punned pointer will 
> break strict-aliasing rules" warnings for sscanf() 
> functions in util/trace-event-parse.c

Tony, Fenghua Yu, this patch already depends on .32 
changes in the perfcounters tree - would be OK to you if 
we merged this patch into the perfcounters tree? Plus 
perhaps merge a syscall placeholder patch into the ia64 
tree?

Thanks,

	Ingo

Re: [PATCH] Add minimal support for software performance counters for ia64

[Ingo Molnar]

* William Cohen <wcohen@redhat.com> wrote:

> The following patch adds minimal support for software 
> performance counters for ia64.

Nice!

> For the kernel the patch enables configuration of the 
> perf counter option, adds the perf_counter_open 
> syscall, and includes a minimal machine specific 
> asm/perf.h header file.
> 
> For the perf tool the header perf.h includes an ia64 
> specific section and the options used in the Makefile 
> were adjusted to allow compilation on ia64.  The 
> -Wcast-align gives "cast increases required alignment 
> of target type" warning for the list_for_each_entry() 
> macro. The "-fno-strict-aliasing" was used to avoid 
> warnings for "dereferencing type-punned pointer will 
> break strict-aliasing rules" warnings for sscanf() 
> functions in util/trace-event-parse.c

Tony, Fenghua Yu, this patch already depends on .32 
changes in the perfcounters tree - would be OK to you if 
we merged this patch into the perfcounters tree? Plus 
perhaps merge a syscall placeholder patch into the ia64 
tree?

Thanks,

	Ingo