[PATCH] basic perf support for sparc

[PATCH] basic perf support for sparc

[Jens Axboe]

Hi,

This adds the little necessary bits to get perf going at
least. Callgraph doesn't work, but basic profiling does.

Building the perf tool is somewhat involved on sparc64
though, since 64-bit versions of zlib/libelf/bfd aren't
directly available (at least on debian 5.x). But once you
get there, it runs :-). Would it be easier/functional
to build 32-bit userland perf instead?

$  perf record -f ls > /dev/null
[ perf record: Captured and wrote 0.004 MB perf.data (~194 samples) ]
$ perf report --sort comm,dso,symbol | head -n10
# Samples: 98
#
# Overhead  Command  Shared Object  Symbol
# ........  .......  .............  ......
#
    25.51%       ls       f7f62e00  [.] 0x000000f7f62e00
     6.12%     perf  [kernel]       [.] 0xfffff801007e4b38
     6.12%       ls  [kernel]       [k] sparc64_realfault_common
     3.06%       ls  [kernel]       [k] tg_shares_up
     3.06%       ls  [kernel]       [k] find_next_bit

$ perf record -f -g ls > /dev/null
[ perf record: Captured and wrote 0.005 MB perf.data (~223 samples) ]
$ perf report -g | head -n10
Warning: empty node in callchain tree
Warning: empty node in callchain tree
Warning: empty node in callchain tree
[...]

Patch is against latest -git.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

diff --git a/arch/sparc/Kconfig b/arch/sparc/Kconfig
index 3f8b6a9..2f951a9 100644
--- a/arch/sparc/Kconfig
+++ b/arch/sparc/Kconfig
@@ -44,6 +44,7 @@ config SPARC64
 	select RTC_DRV_BQ4802
 	select RTC_DRV_SUN4V
 	select RTC_DRV_STARFIRE
+	select HAVE_PERF_COUNTERS
 
 config ARCH_DEFCONFIG
 	string
diff --git a/arch/sparc/include/asm/unistd.h b/arch/sparc/include/asm/unistd.h
index b2c406d..706df66 100644
--- a/arch/sparc/include/asm/unistd.h
+++ b/arch/sparc/include/asm/unistd.h
@@ -395,8 +395,9 @@
 #define __NR_preadv		324
 #define __NR_pwritev		325
 #define __NR_rt_tgsigqueueinfo	326
+#define __NR_perf_counter_open	327
 
-#define NR_SYSCALLS		327
+#define NR_SYSCALLS		328
 
 #ifdef __32bit_syscall_numbers__
 /* Sparc 32-bit only has the "setresuid32", "getresuid32" variants,
diff --git a/arch/sparc/kernel/systbls_64.S b/arch/sparc/kernel/systbls_64.S
index 6b3ee88..80ebf20 100644
--- a/arch/sparc/kernel/systbls_64.S
+++ b/arch/sparc/kernel/systbls_64.S
@@ -158,4 +158,4 @@ sys_call_table:
 /*310*/	.word sys_utimensat, sys_signalfd, sys_timerfd_create, sys_eventfd, sys_fallocate
 	.word sys_timerfd_settime, sys_timerfd_gettime, sys_signalfd4, sys_eventfd2, sys_epoll_create1
 /*320*/	.word sys_dup3, sys_pipe2, sys_inotify_init1, sys_accept4, sys_preadv
-	.word sys_pwritev, sys_rt_tgsigqueueinfo
+	.word sys_pwritev, sys_rt_tgsigqueueinfo, sys_perf_counter_open

diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index e5148e2..2abeb20 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -41,6 +41,12 @@
 #define cpu_relax()	asm volatile("" ::: "memory");
 #endif
 
+#ifdef __sparc__
+#include "../../arch/sparc/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>

-- 
Jens Axboe

Re: [PATCH] basic perf support for sparc

[Jens Axboe]

On Wed, Jul 29 2009, Jens Axboe wrote:
> Building the perf tool is somewhat involved on sparc64
> though, since 64-bit versions of zlib/libelf/bfd aren't
> directly available (at least on debian 5.x). But once you
> get there, it runs :-). Would it be easier/functional
> to build 32-bit userland perf instead?

Same is true on ppc64, btw. How are others handling this?

-- 
Jens Axboe

Re: [PATCH] basic perf support for sparc

[Anton Blanchard]

Hi,

> > Building the perf tool is somewhat involved on sparc64
> > though, since 64-bit versions of zlib/libelf/bfd aren't
> > directly available (at least on debian 5.x). But once you
> > get there, it runs :-). Would it be easier/functional
> > to build 32-bit userland perf instead?
> 
> Same is true on ppc64, btw. How are others handling this?

The requirement for libz was removed, so up until recently we only needed
a 64bit version of elfutils which is easy to build.

It looks like we now have a requirement on binutils which is considerably
more painful to build. One option is to make the bfd requirement optional, all
you lose would be the ability to see c++ demangled names I think.

Anton

Re: [PATCH] basic perf support for sparc

[Jens Axboe]

On Sat, Aug 01 2009, Anton Blanchard wrote:
> 
> Hi,
> 
> > > Building the perf tool is somewhat involved on sparc64
> > > though, since 64-bit versions of zlib/libelf/bfd aren't
> > > directly available (at least on debian 5.x). But once you
> > > get there, it runs :-). Would it be easier/functional
> > > to build 32-bit userland perf instead?
> > 
> > Same is true on ppc64, btw. How are others handling this?
> 
> The requirement for libz was removed, so up until recently we only needed
> a 64bit version of elfutils which is easy to build.
> 
> It looks like we now have a requirement on binutils which is considerably
> more painful to build. One option is to make the bfd requirement optional, all
> you lose would be the ability to see c++ demangled names I think.

Right, binutils is the ugly one. I got a libbfd.so built for both ppc
and sparc, but it wasn't just a make && make install job. Personally I
could not care less about losing c++ demangled name support, so that
approach sounds fine to me :-)

-- 
Jens Axboe

Re: [PATCH] basic perf support for sparc

[Arnaldo Carvalho de Melo]

Em Sat, Aug 01, 2009 at 10:20:48AM +0200, Jens Axboe escreveu:
> On Sat, Aug 01 2009, Anton Blanchard wrote:
> > 
> > Hi,
> > 
> > > > Building the perf tool is somewhat involved on sparc64
> > > > though, since 64-bit versions of zlib/libelf/bfd aren't
> > > > directly available (at least on debian 5.x). But once you
> > > > get there, it runs :-). Would it be easier/functional
> > > > to build 32-bit userland perf instead?
> > > 
> > > Same is true on ppc64, btw. How are others handling this?
> > 
> > The requirement for libz was removed, so up until recently we only needed
> > a 64bit version of elfutils which is easy to build.
> > 
> > It looks like we now have a requirement on binutils which is considerably
> > more painful to build. One option is to make the bfd requirement optional, all
> > you lose would be the ability to see c++ demangled names I think.
> 
> Right, binutils is the ugly one. I got a libbfd.so built for both ppc
> and sparc, but it wasn't just a make && make install job. Personally I
> could not care less about losing c++ demangled name support, so that
> approach sounds fine to me :-)

Exactly, for a huge number of developers not being able to see demangled
C++ is okay, so I agree on adding smarts to not demangle when
binutils-devel is not available.

I thought about extracting the demangling bits out of binutils, ran away
screaming. I also hoped elfutils would have that by now, but it doesn't.

- Arnaldo

Re: [PATCH] basic perf support for sparc

[Ingo Molnar]

* Arnaldo Carvalho de Melo <acme@redhat.com> wrote:

> Em Sat, Aug 01, 2009 at 10:20:48AM +0200, Jens Axboe escreveu:
> > On Sat, Aug 01 2009, Anton Blanchard wrote:
> > > 
> > > Hi,
> > > 
> > > > > Building the perf tool is somewhat involved on sparc64
> > > > > though, since 64-bit versions of zlib/libelf/bfd aren't
> > > > > directly available (at least on debian 5.x). But once you
> > > > > get there, it runs :-). Would it be easier/functional
> > > > > to build 32-bit userland perf instead?
> > > > 
> > > > Same is true on ppc64, btw. How are others handling this?
> > > 
> > > The requirement for libz was removed, so up until recently we only needed
> > > a 64bit version of elfutils which is easy to build.
> > > 
> > > It looks like we now have a requirement on binutils which is considerably
> > > more painful to build. One option is to make the bfd requirement optional, all
> > > you lose would be the ability to see c++ demangled names I think.
> > 
> > Right, binutils is the ugly one. I got a libbfd.so built for 
> > both ppc and sparc, but it wasn't just a make && make install 
> > job. Personally I could not care less about losing c++ demangled 
> > name support, so that approach sounds fine to me :-)
> 
> Exactly, for a huge number of developers not being able to see 
> demangled C++ is okay, so I agree on adding smarts to not demangle 
> when binutils-devel is not available.
> 
> I thought about extracting the demangling bits out of binutils, 
> ran away screaming. I also hoped elfutils would have that by now, 
> but it doesn't.

Could we somehow define a weak symbol for those library functions 
ourselves and thus just fall back to that (which does nothing) 
instead of failing the link?

	Ingo

Re: [PATCH] basic perf support for sparc

[Kyle McMartin]

On Sun, Aug 02, 2009 at 08:41:48PM +0200, Ingo Molnar wrote:
> Could we somehow define a weak symbol for those library functions 
> ourselves and thus just fall back to that (which does nothing) 
> instead of failing the link?
> 

Well, you could just dlopen the object, but that's kind of gross...

Why not abuse the perf Makefile, which already has this kind of
portability gunk in it?

$ make NO_BFD_DEMANGLE=1
succeeds with no libbfd.

$ make
succeeds when libbfd is around.

diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index a5e9b87..67ab097 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -345,7 +345,6 @@ BUILTIN_OBJS += builtin-stat.o
 BUILTIN_OBJS += builtin-top.o
 
 PERFLIBS = $(LIB_FILE)
-EXTLIBS = -lbfd
 
 #
 # Platform specific tweaks
@@ -541,6 +540,11 @@ endif
 ifdef NO_EXTERNAL_GREP
 	BASIC_CFLAGS += -DNO_EXTERNAL_GREP
 endif
+ifdef NO_BFD_DEMANGLE
+	BASIC_CFLAGS += -DNO_BFD_DEMANGLE
+else
+	EXTLIBS += -lbfd
+endif
 
 ifeq ($(PERL_PATH),)
 NO_PERL=NoThanks
diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
index 2810605..736b8db 100644
--- a/tools/perf/util/symbol.c
+++ b/tools/perf/util/symbol.c
@@ -6,10 +6,20 @@
 #include <libelf.h>
 #include <gelf.h>
 #include <elf.h>
+
+#ifndef NO_BFD_DEMANGLE
 #include <bfd.h>
+#endif /* NO_BFD_DEMANGLE */
 
 const char *sym_hist_filter;
 
+#ifdef NO_BFD_DEMANGLE
+static inline char *bfd_demangle(void __used *v, const char __used *c,
+				int __used i) {
+	return NULL;
+}
+#endif
+
 #ifndef DMGL_PARAMS
 #define DMGL_PARAMS      (1 << 0)       /* Include function args */
 #define DMGL_ANSI        (1 << 1)       /* Include const, volatile, etc */

Re: [PATCH] basic perf support for sparc

[Ingo Molnar]

* Kyle McMartin <kyle@mcmartin.ca> wrote:

> On Sun, Aug 02, 2009 at 08:41:48PM +0200, Ingo Molnar wrote:
> > Could we somehow define a weak symbol for those library functions 
> > ourselves and thus just fall back to that (which does nothing) 
> > instead of failing the link?
> > 
> 
> Well, you could just dlopen the object, but that's kind of 
> gross...

hm, why is it gross? I think it's a nicely plug-and-play tooling 
solution - instead of some obscure make flag.

Since we define the prototype anyway:

> +static inline char *bfd_demangle(void __used *v, const char __used *c,
> +				int __used i) {
> +	return NULL;
> +}

We are sticking to a given version of the API. We could turn that 
into a function pointer and fill it in during startup via dlopen(). 
If it's NULL then we dont call it and assume a value of NULL.

Mind submitting such a version of your fix? It would nicely decrease 
the build requirements cross section surface of perf.

	Ingo

Re: [PATCH] basic perf support for sparc

[Kyle McMartin]

On Sun, Aug 02, 2009 at 09:50:43PM +0200, Ingo Molnar wrote:
> We are sticking to a given version of the API. We could turn that 
> into a function pointer and fill it in during startup via dlopen(). 
> If it's NULL then we dont call it and assume a value of NULL.
> 
> Mind submitting such a version of your fix? It would nicely decrease 
> the build requirements cross section surface of perf.
> 

Assuming I remember how to use dlopen/dlsym, this might work.
Unfortunately I can't easily test anything this weekend, so
I don't actually know if it works... but it links ok.

cheers, Kyle

diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index a5e9b87..8fb2e12 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -345,7 +345,7 @@ BUILTIN_OBJS += builtin-stat.o
 BUILTIN_OBJS += builtin-top.o
 
 PERFLIBS = $(LIB_FILE)
-EXTLIBS = -lbfd
+EXTLIBS = -ldl
 
 #
 # Platform specific tweaks
diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
index 2810605..e41cb71 100644
--- a/tools/perf/util/symbol.c
+++ b/tools/perf/util/symbol.c
@@ -4,9 +4,9 @@
 #include "symbol.h"
 
 #include <libelf.h>
+#include <dlfcn.h>
 #include <gelf.h>
 #include <elf.h>
-#include <bfd.h>
 
 const char *sym_hist_filter;
 
@@ -512,6 +512,11 @@ out:
 	return 0;
 }
 
+static char *perf_bfd_demangle(void __used *v,
+	const char __used *name, int __used opt) {
+	return NULL;
+}
+
 static int dso__load_sym(struct dso *self, int fd, const char *name,
 			 symbol_filter_t filter, int verbose, struct module *mod)
 {
@@ -526,6 +531,8 @@ static int dso__load_sym(struct dso *self, int fd, const char *name,
 	Elf_Scn *sec, *sec_strndx;
 	Elf *elf;
 	int nr = 0, kernel = !strcmp("[kernel]", self->name);
+	char *(*demangle)(void *v, const char *name, int opt) = NULL;
+	void *handle;
 
 	elf = elf_begin(fd, ELF_C_READ_MMAP, NULL);
 	if (elf == NULL) {
@@ -578,6 +585,13 @@ static int dso__load_sym(struct dso *self, int fd, const char *name,
 						     NULL) != NULL);
 	} else self->adjust_symbols = 0;
 
+	handle = dlopen("libbfd.so", RTLD_NOW);
+	if (!handle)
+		demangle = perf_bfd_demangle;
+	demangle = dlsym(handle, "bfd_demangle");
+	if (!demangle)
+		demangle = perf_bfd_demangle;
+
 	elf_symtab__for_each_symbol(syms, nr_syms, index, sym) {
 		struct symbol *f;
 		const char *name;
@@ -592,7 +606,7 @@ static int dso__load_sym(struct dso *self, int fd, const char *name,
 
 		sec = elf_getscn(elf, sym.st_shndx);
 		if (!sec)
-			goto out_elf_end;
+			goto out_dlclose;
 
 		gelf_getshdr(sec, &shdr);
 
@@ -617,7 +631,7 @@ static int dso__load_sym(struct dso *self, int fd, const char *name,
 			else {
 				fprintf(stderr, "dso__load_sym() module %s lookup of %s failed\n",
 					mod->name, section_name);
-				goto out_elf_end;
+				goto out_dlclose;
 			}
 		}
 		/*
@@ -626,7 +640,7 @@ static int dso__load_sym(struct dso *self, int fd, const char *name,
 		 * to it...
 		 */
 		name = elf_sym__name(&sym, symstrs);
-		demangled = bfd_demangle(NULL, name, DMGL_PARAMS | DMGL_ANSI);
+		demangled = demangle(NULL, name, DMGL_PARAMS | DMGL_ANSI);
 		if (demangled != NULL)
 			name = demangled;
 
@@ -634,7 +648,7 @@ static int dso__load_sym(struct dso *self, int fd, const char *name,
 				self->sym_priv_size, obj_start, verbose);
 		free(demangled);
 		if (!f)
-			goto out_elf_end;
+			goto out_dlclose;
 
 		if (filter && filter(self, f))
 			symbol__delete(f, self->sym_priv_size);
@@ -646,6 +660,9 @@ static int dso__load_sym(struct dso *self, int fd, const char *name,
 	}
 
 	err = nr;
+out_dlclose:
+	if (handle)
+		dlclose(handle);
 out_elf_end:
 	elf_end(elf);
 out_close:

Re: [PATCH] basic perf support for sparc

[Ingo Molnar]

* Kyle McMartin <kyle@mcmartin.ca> wrote:

> On Sun, Aug 02, 2009 at 09:50:43PM +0200, Ingo Molnar wrote:
> > We are sticking to a given version of the API. We could turn that 
> > into a function pointer and fill it in during startup via dlopen(). 
> > If it's NULL then we dont call it and assume a value of NULL.
> > 
> > Mind submitting such a version of your fix? It would nicely decrease 
> > the build requirements cross section surface of perf.
> > 
> 
> Assuming I remember how to use dlopen/dlsym, this might work. 
> Unfortunately I can't easily test anything this weekend, so I 
> don't actually know if it works... but it links ok.

looks good to me. Arnaldo, mind picking this up?

	Ingo

Re: [PATCH] basic perf support for sparc

[Arnaldo Carvalho de Melo]

Em Sun, Aug 02, 2009 at 04:11:32PM -0400, Kyle McMartin escreveu:
> On Sun, Aug 02, 2009 at 09:50:43PM +0200, Ingo Molnar wrote:
> > We are sticking to a given version of the API. We could turn that 
> > into a function pointer and fill it in during startup via dlopen(). 
> > If it's NULL then we dont call it and assume a value of NULL.
> > 
> > Mind submitting such a version of your fix? It would nicely decrease 
> > the build requirements cross section surface of perf.
> > 
> 
> Assuming I remember how to use dlopen/dlsym, this might work.
> Unfortunately I can't easily test anything this weekend, so
> I don't actually know if it works... but it links ok.
> 
> cheers, Kyle

Doesn't seem to work, tried with firefox, with matching
xulrunner-debuginfo package installed and symbols didn't got properly
demangled, gotta be afk now, but will try to figure this out later.

[]s.

- Arnaldo

Re: [PATCH] basic perf support for sparc

[Arnaldo Carvalho de Melo]

Em Sun, Aug 02, 2009 at 05:47:01PM -0300, Arnaldo Carvalho de Melo escreveu:
> Em Sun, Aug 02, 2009 at 04:11:32PM -0400, Kyle McMartin escreveu:
> > On Sun, Aug 02, 2009 at 09:50:43PM +0200, Ingo Molnar wrote:
> > > We are sticking to a given version of the API. We could turn that 
> > > into a function pointer and fill it in during startup via dlopen(). 
> > > If it's NULL then we dont call it and assume a value of NULL.
> > > 
> > > Mind submitting such a version of your fix? It would nicely decrease 
> > > the build requirements cross section surface of perf.
> > > 
> > 
> > Assuming I remember how to use dlopen/dlsym, this might work.
> > Unfortunately I can't easily test anything this weekend, so
> > I don't actually know if it works... but it links ok.
> > 
> > cheers, Kyle
> 
> Doesn't seem to work, tried with firefox, with matching
> xulrunner-debuginfo package installed and symbols didn't got properly
> demangled, gotta be afk now, but will try to figure this out later.

The problem is that at least on f11, libbfd.so is a... linker script,
not an ELF shared library as the name seemed to imply.

But if I cheat and do a dlopen on the full name, that includes the
version, it works. Argh, I don't know why this has to be so contorted,
does anybody here understands this?

Anyway, the reworked patch below works and additionally doesn't dlopens
at each DSO, but just once and only if it finds a symbol starting with
an underscore, an heuristic to avoid loading bfd if we don't need it.

Looking at each DW_AT_language attribute in the DW_TAG_compile_unit
tags, trying to figure out if there is a C++ object in the mix seems too
much, we would have to traverse all of the compile units, touching most
of the debuginfo file :-\

- Arnaldo

diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index 4b20fa4..8fb2e12 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -345,7 +345,7 @@ BUILTIN_OBJS += builtin-stat.o
 BUILTIN_OBJS += builtin-top.o
 
 PERFLIBS = $(LIB_FILE)
-EXTLIBS = -lbfd -liberty
+EXTLIBS = -ldl
 
 #
 # Platform specific tweaks
diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
index b4fe057..625f975 100644
--- a/tools/perf/util/symbol.c
+++ b/tools/perf/util/symbol.c
@@ -4,9 +4,9 @@
 #include "symbol.h"
 
 #include <libelf.h>
+#include <dlfcn.h>
 #include <gelf.h>
 #include <elf.h>
-#include <bfd.h>
 
 const char *sym_hist_filter;
 
@@ -512,6 +512,30 @@ out:
 	return 0;
 }
 
+static char *def_perf_bfd_demangle(void __used *v, const char __used *name,
+				   int __used opt)
+{
+	return NULL;
+}
+
+static char *(*perf_bfd_demangle)(void *v, const char *name, int opt);
+
+static void load_perf_bfd_demangle(void)
+{
+	void *handle = dlopen("libbfd-2.19.51.0.2-17.fc11.so", RTLD_NOW);
+
+	if (handle) {
+		perf_bfd_demangle = dlsym(handle, "bfd_demangle");
+		if (perf_bfd_demangle)
+			return;
+		dlclose(handle);
+	} else
+		fprintf(stderr, "%s: problems finding bfd_demangle: %s.\n",
+			__func__, dlerror());
+
+	perf_bfd_demangle = def_perf_bfd_demangle;
+}
+
 static int dso__load_sym(struct dso *self, int fd, const char *name,
 			 symbol_filter_t filter, int verbose, struct module *mod)
 {
@@ -581,7 +605,7 @@ static int dso__load_sym(struct dso *self, int fd, const char *name,
 	elf_symtab__for_each_symbol(syms, nr_syms, index, sym) {
 		struct symbol *f;
 		const char *name;
-		char *demangled;
+		char *demangled = NULL;
 		u64 obj_start;
 		struct section *section = NULL;
 		int is_label = elf_sym__is_label(&sym);
@@ -626,9 +650,16 @@ static int dso__load_sym(struct dso *self, int fd, const char *name,
 		 * to it...
 		 */
 		name = elf_sym__name(&sym, symstrs);
-		demangled = bfd_demangle(NULL, name, DMGL_PARAMS | DMGL_ANSI);
-		if (demangled != NULL)
-			name = demangled;
+		if (name && name[0] == '_') {
+			if (perf_bfd_demangle == NULL)
+				load_perf_bfd_demangle();
+
+			if (perf_bfd_demangle != NULL) {
+				demangled = perf_bfd_demangle(NULL, name, DMGL_PARAMS | DMGL_ANSI);
+				if (demangled != NULL)
+					name = demangled;
+			}
+		}
 
 		f = symbol__new(sym.st_value, sym.st_size, name,
 				self->sym_priv_size, obj_start, verbose);

Re: [PATCH] basic perf support for sparc

[Kyle McMartin]

On Sun, Aug 02, 2009 at 10:54:30PM -0300, Arnaldo Carvalho de Melo wrote:
> The problem is that at least on f11, libbfd.so is a... linker script,
> not an ELF shared library as the name seemed to imply.
> 

Ugh, that's weird. My fedora boxes are the same, but it's correctly a
symlink on my Debian box.
lrwxrwxrwx 1 root root     26 Aug  3 23:30 /usr/lib/libbfd.so ->
libbfd-2.19.51.20090723.so

Kind of strange it's a reference to bfd.a too. :/

> But if I cheat and do a dlopen on the full name, that includes the
> version, it works. Argh, I don't know why this has to be so contorted,
> does anybody here understands this?
> 
> Anyway, the reworked patch below works and additionally doesn't dlopens
> at each DSO, but just once and only if it finds a symbol starting with
> an underscore, an heuristic to avoid loading bfd if we don't need it.
> 

Looks better than mine. :)

cheers, Kyle

Re: [PATCH] basic perf support for sparc

[Ingo Molnar]

* Arnaldo Carvalho de Melo <acme@redhat.com> wrote:

> +static void load_perf_bfd_demangle(void)
> +{
> +	void *handle = dlopen("libbfd-2.19.51.0.2-17.fc11.so", RTLD_NOW);

Hm, this does not look like a very generic solution. Is there some 
way to do a library search to figure out the name? I guess a glob 
match on /usr/lib/libbfd*.so?

	Ingo

Re: [PATCH] basic perf support for sparc

[Peter Zijlstra]

On Tue, 2009-08-04 at 11:25 +0200, Ingo Molnar wrote:
> * Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
> 
> > +static void load_perf_bfd_demangle(void)
> > +{
> > +	void *handle = dlopen("libbfd-2.19.51.0.2-17.fc11.so", RTLD_NOW);
> 
> Hm, this does not look like a very generic solution. Is there some 
> way to do a library search to figure out the name? I guess a glob 
> match on /usr/lib/libbfd*.so?

I'd simply go for libbfd.so and raise a bug report against fedora. A
link archive named .so is just stupid.

Re: [PATCH] basic perf support for sparc

[David Miller]

From: Peter Zijlstra <a.p.zijlstra@chello.nl>
Date: Tue, 04 Aug 2009 11:29:00 +0200

> On Tue, 2009-08-04 at 11:25 +0200, Ingo Molnar wrote:
>> * Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
>> 
>> > +static void load_perf_bfd_demangle(void)
>> > +{
>> > +	void *handle = dlopen("libbfd-2.19.51.0.2-17.fc11.so", RTLD_NOW);
>> 
>> Hm, this does not look like a very generic solution. Is there some 
>> way to do a library search to figure out the name? I guess a glob 
>> match on /usr/lib/libbfd*.so?
> 
> I'd simply go for libbfd.so and raise a bug report against fedora. A
> link archive named .so is just stupid.

It really isn't, I find it surprising why people are so up in
arms about this.  It's a quite common technique.

Re: [PATCH] basic perf support for sparc

[Frederic Riss]

2009/8/4 Ingo Molnar <mingo@elte.hu>:
>
> * Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
>
>> +static void load_perf_bfd_demangle(void)
>> +{
>> +     void *handle = dlopen("libbfd-2.19.51.0.2-17.fc11.so", RTLD_NOW);
>
> Hm, this does not look like a very generic solution. Is there some
> way to do a library search to figure out the name? I guess a glob
> match on /usr/lib/libbfd*.so?

If all you want is C++ symbols demangling, why not just look for the
c++filt binary on the system? It accepts mangled names on its stdin
and outputs the demangled name on stdout.  You can keep it running in
the background, and communicate through pipes. Of course it's not as
effective as a library call, but I guess that it'll be much more
generic and also much more likely to be found on a developer's system
than a libbfd shared library.

Fred.

Re: [PATCH] basic perf support for sparc

[Peter Zijlstra]

On Tue, 2009-08-04 at 12:32 +0200, Frederic Riss wrote:
> 2009/8/4 Ingo Molnar <mingo@elte.hu>:
> >
> > * Arnaldo Carvalho de Melo <acme@redhat.com> wrote:
> >
> >> +static void load_perf_bfd_demangle(void)
> >> +{
> >> +     void *handle = dlopen("libbfd-2.19.51.0.2-17.fc11.so", RTLD_NOW);
> >
> > Hm, this does not look like a very generic solution. Is there some
> > way to do a library search to figure out the name? I guess a glob
> > match on /usr/lib/libbfd*.so?
> 
> If all you want is C++ symbols demangling, why not just look for the
> c++filt binary on the system? It accepts mangled names on its stdin
> and outputs the demangled name on stdout.  You can keep it running in
> the background, and communicate through pipes. Of course it's not as
> effective as a library call, but I guess that it'll be much more
> generic and also much more likely to be found on a developer's system
> than a libbfd shared library.

Oddly enough c++filt uses libbfd :-)

$ ldd `which c++filt`
        linux-gate.so.1 =>  (0xffffe000)
        libbfd-2.19.1.so => /usr/lib/libbfd-2.19.1.so (0xf7de9000)
        libz.so.1 => /lib/libz.so.1 (0xf7dd3000)
        libc.so.6 => /lib/tls/i686/cmov/libc.so.6 (0xf7c6f000)
        /lib/ld-linux.so.2 (0xf7ef9000)

Re: [PATCH] basic perf support for sparc

[Frederic Riss]

2009/8/4 Peter Zijlstra <a.p.zijlstra@chello.nl>:
> On Tue, 2009-08-04 at 12:32 +0200, Frederic Riss wrote:
>> If all you want is C++ symbols demangling, why not just look for the
>> c++filt binary on the system? It accepts mangled names on its stdin
>> and outputs the demangled name on stdout.  You can keep it running in
>> the background, and communicate through pipes. Of course it's not as
>> effective as a library call, but I guess that it'll be much more
>> generic and also much more likely to be found on a developer's system
>> than a libbfd shared library.
>
> Oddly enough c++filt uses libbfd :-)
>
> $ ldd `which c++filt`
>        linux-gate.so.1 =>  (0xffffe000)
>        libbfd-2.19.1.so => /usr/lib/libbfd-2.19.1.so (0xf7de9000)
>        libz.so.1 => /lib/libz.so.1 (0xf7dd3000)
>        libc.so.6 => /lib/tls/i686/cmov/libc.so.6 (0xf7c6f000)
>        /lib/ld-linux.so.2 (0xf7ef9000)

Hehe... It depends how distro build it. IIRC, the default binutils
configure links statically, but I might be wrong. Anyway, this
approach still has the advantage that the binary is called 'c++filt'
and not c++filt-2.19.1-fc11 :-)

Fred

Re: [PATCH] basic perf support for sparc

[Ingo Molnar]

Btw., how about the solution below? Will the libbfd system library 
version override this weak alias properly?

	Ingo

-------------->
>From 238762507bb41689df191160ada1986b012b57df Mon Sep 17 00:00:00 2001
From: Ingo Molnar <mingo@elte.hu>
Date: Tue, 4 Aug 2009 13:27:41 +0200
Subject: [PATCH] perf_counter tools: Provide default bfd_demangle() function in case it's not around

Provide weak alias (which does nothing) in case the system does not
offer one.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 tools/perf/util/symbol.c |   11 +++++++++++
 1 files changed, 11 insertions(+), 0 deletions(-)

diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
index b4fe057..7bd5128 100644
--- a/tools/perf/util/symbol.c
+++ b/tools/perf/util/symbol.c
@@ -8,6 +8,17 @@
 #include <elf.h>
 #include <bfd.h>
 
+/*
+ * Weak wrapper in case a library version is not available:
+ */
+extern char * __attribute__((weak)) bfd_demangle (bfd *, const char *, int);
+
+char * __attribute__((weak))
+ bfd_demangle (bfd *bfd __used, const char *name __used, int flags __used)
+{
+	return NULL;
+}
+
 const char *sym_hist_filter;
 
 #ifndef DMGL_PARAMS

Re: [PATCH] basic perf support for sparc

[Peter Zijlstra]

On Sun, 2009-08-02 at 15:44 -0400, Kyle McMartin wrote:

> Why not abuse the perf Makefile, which already has this kind of
> portability gunk in it?

Does this work for people?

---
Subject: perf: autodetect libbfd
From: Peter Zijlstra <a.p.zijlstra@chello.nl>
Date: Wed Aug 05 14:05:16 CEST 2009

Since the C++ demangling isn't needed for everybody and bfd/iberty
aren't widely/easily available on all machines, make it optional.

It allows you to forcefully disable demangling by using NO_DEMANGLE=1
and otherwise tries to detect libbfd/libiberty combinations that
result in a compiling demangler.

Cc: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 tools/perf/Makefile      |   17 ++++++++++++++++-
 tools/perf/util/symbol.c |    9 +++++++++
 2 files changed, 25 insertions(+), 1 deletion(-)

Index: linux-2.6/tools/perf/Makefile
===================================================================
--- linux-2.6.orig/tools/perf/Makefile
+++ linux-2.6/tools/perf/Makefile
@@ -345,7 +345,6 @@ BUILTIN_OBJS += builtin-stat.o
 BUILTIN_OBJS += builtin-top.o
 
 PERFLIBS = $(LIB_FILE)
-EXTLIBS = -lbfd -liberty
 
 #
 # Platform specific tweaks
@@ -374,6 +373,22 @@ ifeq ($(uname_S),Darwin)
 	PTHREAD_LIBS =
 endif
 
+ifdef NO_DEMANGLE
+	BASIC_CFLAGS += -DNO_DEMANGLE
+else
+	has_bfd := $(shell sh -c "(echo '\#include <stdio.h>'; echo '\#include <bfd.h>'; echo '\#ifndef DMGL_PARAMS'; echo '\#define DMGL_PARAMS (1 << 0)'; echo '\#define DMGL_ANSI (1 << 1)'; echo '\#endif'; echo 'int main(int argc, char **argv) { bfd_demangle(NULL, argv[0], DMGL_PARAMS | DMGL_ANSI); return 0; }') | gcc -x c - -lbfd > /dev/null 2>&1 && echo y")
+
+	has_bfd_iberty := $(shell sh -c "(echo '\#include <stdio.h>'; echo '\#include <bfd.h>'; echo '\#ifndef DMGL_PARAMS'; echo '\#define DMGL_PARAMS (1 << 0)'; echo '\#define DMGL_ANSI (1 << 1)'; echo '\#endif'; echo 'int main(int argc, char **argv) { bfd_demangle(NULL, argv[0], DMGL_PARAMS | DMGL_ANSI); return 0; }') | gcc -x c - -lbfd -liberty > /dev/null 2>&1 && echo y")
+
+	ifeq ($(has_bfd),y)
+		EXTLIBS += -lbfd
+	else ifeq ($(has_bfd_iberty),y)
+		EXTLIBS += -lbfd -liberty
+	else
+		BASIC_CFLAGS += -DNO_DEMANGLE
+	endif
+endif
+
 ifndef CC_LD_DYNPATH
 	ifdef NO_R_TO_GCC_LINKER
 		# Some gcc does not accept and pass -R to the linker to specify
Index: linux-2.6/tools/perf/util/symbol.c
===================================================================
--- linux-2.6.orig/tools/perf/util/symbol.c
+++ linux-2.6/tools/perf/util/symbol.c
@@ -6,7 +6,16 @@
 #include <libelf.h>
 #include <gelf.h>
 #include <elf.h>
+
+#ifndef NO_DEMANGLE
 #include <bfd.h>
+#else
+static inline
+char *bfd_demangle(void __used *v, const char __used *c, int __used i)
+{
+	return NULL;
+}
+#endif
 
 const char *sym_hist_filter;
 

Re: [PATCH] basic perf support for sparc

[Jens Axboe]

On Wed, Aug 05 2009, Peter Zijlstra wrote:
> On Sun, 2009-08-02 at 15:44 -0400, Kyle McMartin wrote:
> 
> > Why not abuse the perf Makefile, which already has this kind of
> > portability gunk in it?
> 
> Does this work for people?

Works for me, just tested it.

-- 
Jens Axboe

Re: [PATCH] basic perf support for sparc

[Ingo Molnar]

* Jens Axboe <jens.axboe@oracle.com> wrote:

> On Wed, Aug 05 2009, Peter Zijlstra wrote:
> > On Sun, 2009-08-02 at 15:44 -0400, Kyle McMartin wrote:
> > 
> > > Why not abuse the perf Makefile, which already has this kind of
> > > portability gunk in it?
> > 
> > Does this work for people?
> 
> Works for me, just tested it.

Thanks Jens - i've committed it and will push it Linus-wards.

	Ingo

[tip:perfcounters/urgent] perf: Auto-detect libbfd

[tip-bot for Peter Zijlstra]

Commit-ID:  2cdbc46d7b2cb0acb68c3ecad93b000552121fa6
Gitweb:     http://git.kernel.org/tip/2cdbc46d7b2cb0acb68c3ecad93b000552121fa6
Author:     Peter Zijlstra <a.p.zijlstra@chello.nl>
AuthorDate: Wed, 5 Aug 2009 14:05:16 +0200
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Wed, 5 Aug 2009 14:12:08 +0200

perf: Auto-detect libbfd

Since the C++ demangling isn't needed for everybody and
bfd/iberty aren't widely/easily available on all machines, make
it optional.

It also allows you to forcefully disable demangling by using
NO_DEMANGLE=1 and otherwise tries to detect libbfd/libiberty
combinations that result in a compiling demangler.

Reported-by: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Kyle McMartin <kyle@mcmartin.ca>
LKML-Reference: <20090801082048.GX12579@kernel.dk>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


---
 tools/perf/Makefile      |   18 +++++++++++++++++-
 tools/perf/util/symbol.c |    9 +++++++++
 2 files changed, 26 insertions(+), 1 deletions(-)

diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index 4b20fa4..ff905ac 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -345,7 +345,6 @@ BUILTIN_OBJS += builtin-stat.o
 BUILTIN_OBJS += builtin-top.o
 
 PERFLIBS = $(LIB_FILE)
-EXTLIBS = -lbfd -liberty
 
 #
 # Platform specific tweaks
@@ -374,6 +373,23 @@ ifeq ($(uname_S),Darwin)
 	PTHREAD_LIBS =
 endif
 
+ifdef NO_DEMANGLE
+	BASIC_CFLAGS += -DNO_DEMANGLE
+else
+
+	has_bfd := $(shell sh -c "(echo '\#include <stdio.h>'; echo '\#include <bfd.h>'; echo '\#ifndef DMGL_PARAMS'; echo '\#define DMGL_PARAMS (1 << 0)'; echo '\#define DMGL_ANSI (1 << 1)'; echo '\#endif'; echo 'int main(int argc, char **argv) { bfd_demangle(NULL, argv[0], DMGL_PARAMS | DMGL_ANSI); return 0; }') | gcc -x c - -lbfd > /dev/null 2>&1 && echo y")
+
+	has_bfd_iberty := $(shell sh -c "(echo '\#include <stdio.h>'; echo '\#include <bfd.h>'; echo '\#ifndef DMGL_PARAMS'; echo '\#define DMGL_PARAMS (1 << 0)'; echo '\#define DMGL_ANSI (1 << 1)'; echo '\#endif'; echo 'int main(int argc, char **argv) { bfd_demangle(NULL, argv[0], DMGL_PARAMS | DMGL_ANSI); return 0; }') | gcc -x c - -lbfd -liberty > /dev/null 2>&1 && echo y")
+
+	ifeq ($(has_bfd),y)
+		EXTLIBS += -lbfd
+	else ifeq ($(has_bfd_iberty),y)
+		EXTLIBS += -lbfd -liberty
+	else
+		BASIC_CFLAGS += -DNO_DEMANGLE
+	endif
+endif
+
 ifndef CC_LD_DYNPATH
 	ifdef NO_R_TO_GCC_LINKER
 		# Some gcc does not accept and pass -R to the linker to specify
diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c
index b4fe057..0580b94 100644
--- a/tools/perf/util/symbol.c
+++ b/tools/perf/util/symbol.c
@@ -6,7 +6,16 @@
 #include <libelf.h>
 #include <gelf.h>
 #include <elf.h>
+
+#ifndef NO_DEMANGLE
 #include <bfd.h>
+#else
+static inline
+char *bfd_demangle(void __used *v, const char __used *c, int __used i)
+{
+	return NULL;
+}
+#endif
 
 const char *sym_hist_filter;
 

Re: [tip:perfcounters/urgent] perf: Auto-detect libbfd

[Peter Zijlstra]

On Wed, 2009-08-05 at 12:16 +0000, tip-bot for Peter Zijlstra wrote:
> Commit-ID:  2cdbc46d7b2cb0acb68c3ecad93b000552121fa6
> Gitweb:     http://git.kernel.org/tip/2cdbc46d7b2cb0acb68c3ecad93b000552121fa6
> Author:     Peter Zijlstra <a.p.zijlstra@chello.nl>
> AuthorDate: Wed, 5 Aug 2009 14:05:16 +0200
> Committer:  Ingo Molnar <mingo@elte.hu>
> CommitDate: Wed, 5 Aug 2009 14:12:08 +0200
> 
> perf: Auto-detect libbfd
> 
> Since the C++ demangling isn't needed for everybody and
> bfd/iberty aren't widely/easily available on all machines, make
> it optional.
> 
> It also allows you to forcefully disable demangling by using
> NO_DEMANGLE=1 and otherwise tries to detect libbfd/libiberty
> combinations that result in a compiling demangler.


The below simpifies the C proglet and extends it to cover auto-detection
of libelf as well, it also generates an error in case of libelf being
MIA and an warning for libbfd.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 tools/perf/Makefile |   15 +++++++++++----
 1 files changed, 11 insertions(+), 4 deletions(-)

diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index ff905ac..e9db128 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -158,8 +158,10 @@ uname_P := $(shell sh -c 'uname -p 2>/dev/null || echo not')
 uname_V := $(shell sh -c 'uname -v 2>/dev/null || echo not')
 
 # If we're on a 64-bit kernel, use -m64
-ifneq ($(patsubst %64,%,$(uname_M)),$(uname_M))
-  M64 := -m64
+ifndef NO_64BIT
+	ifneq ($(patsubst %64,%,$(uname_M)),$(uname_M))
+	  M64 := -m64
+	endif
 endif
 
 # CFLAGS and LDFLAGS are for the users to override from the command line.
@@ -373,19 +375,24 @@ ifeq ($(uname_S),Darwin)
 	PTHREAD_LIBS =
 endif
 
+ifneq ($(shell sh -c "(echo '\#include <libelf.h>'; echo 'int main(void) { Elf * elf = elf_begin(0, ELF_C_READ_MMAP, 0); return (int)elf; }') | $(CC) -x c - $(ALL_CFLAGS) -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -o /dev/null $(ALL_LDFLAGS) > /dev/null 2>&1 && echo y"), y)
+	msg := $(error No libelf.h/libelf found, please install libelf-dev[el]);
+endif
+
 ifdef NO_DEMANGLE
 	BASIC_CFLAGS += -DNO_DEMANGLE
 else
 
-	has_bfd := $(shell sh -c "(echo '\#include <stdio.h>'; echo '\#include <bfd.h>'; echo '\#ifndef DMGL_PARAMS'; echo '\#define DMGL_PARAMS (1 << 0)'; echo '\#define DMGL_ANSI (1 << 1)'; echo '\#endif'; echo 'int main(int argc, char **argv) { bfd_demangle(NULL, argv[0], DMGL_PARAMS | DMGL_ANSI); return 0; }') | gcc -x c - -lbfd > /dev/null 2>&1 && echo y")
+	has_bfd := $(shell sh -c "(echo '\#include <bfd.h>'; echo 'int main(void) { bfd_demangle(0, 0, 0); return 0; }') | $(CC) -x c - $(ALL_CFLAGS) -o /dev/null $(ALL_LDFLAGS) -lbfd > /dev/null 2>&1 && echo y")
 
-	has_bfd_iberty := $(shell sh -c "(echo '\#include <stdio.h>'; echo '\#include <bfd.h>'; echo '\#ifndef DMGL_PARAMS'; echo '\#define DMGL_PARAMS (1 << 0)'; echo '\#define DMGL_ANSI (1 << 1)'; echo '\#endif'; echo 'int main(int argc, char **argv) { bfd_demangle(NULL, argv[0], DMGL_PARAMS | DMGL_ANSI); return 0; }') | gcc -x c - -lbfd -liberty > /dev/null 2>&1 && echo y")
+	has_bfd_iberty := $(shell sh -c "(echo '\#include <bfd.h>'; echo 'int main(void) { bfd_demangle(0, 0, 0); return 0; }') | $(CC) -x c - $(ALL_CFLAGS) -o /dev/null $(ALL_LDFLAGS) -lbfd -liberty > /dev/null 2>&1 && echo y")
 
 	ifeq ($(has_bfd),y)
 		EXTLIBS += -lbfd
 	else ifeq ($(has_bfd_iberty),y)
 		EXTLIBS += -lbfd -liberty
 	else
+		msg := $(warning No bfd.h/libbfd found, install binutils-dev[el] to gain symbol demangling)
 		BASIC_CFLAGS += -DNO_DEMANGLE
 	endif
 endif

Re: [tip:perfcounters/urgent] perf: Auto-detect libbfd

[Peter Zijlstra]

> +ifneq ($(shell sh -c "(echo '\#include <libelf.h>'; echo 'int
> main(void) { Elf * elf = elf_begin(0, ELF_C_READ_MMAP, 0); return
> (int)elf; }') | $(CC) -x c - $(ALL_CFLAGS) -D_LARGEFILE64_SOURCE
> -D_FILE_OFFSET_BITS=64 -o /dev/null $(ALL_LDFLAGS) > /dev/null 2>&1 &&
> echo y"), y)

s/(int)/(long)/

---
Index: linux-2.6/tools/perf/Makefile
===================================================================
--- linux-2.6.orig/tools/perf/Makefile
+++ linux-2.6/tools/perf/Makefile
@@ -158,8 +158,10 @@ uname_P := $(shell sh -c 'uname -p 2>/de
 uname_V := $(shell sh -c 'uname -v 2>/dev/null || echo not')
 
 # If we're on a 64-bit kernel, use -m64
-ifneq ($(patsubst %64,%,$(uname_M)),$(uname_M))
-  M64 := -m64
+ifndef NO_64BIT
+	ifneq ($(patsubst %64,%,$(uname_M)),$(uname_M))
+	  M64 := -m64
+	endif
 endif
 
 # CFLAGS and LDFLAGS are for the users to override from the command line.
@@ -373,18 +375,24 @@ ifeq ($(uname_S),Darwin)
 	PTHREAD_LIBS =
 endif
 
+ifneq ($(shell sh -c "(echo '\#include <libelf.h>'; echo 'int main(void) { Elf * elf = elf_begin(0, ELF_C_READ_MMAP, 0); return (long)elf; }') | $(CC) -x c - $(ALL_CFLAGS) -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -o /dev/null $(ALL_LDFLAGS) > /dev/null 2>&1 && echo y"), y)
+	msg := $(error No libelf.h/libelf found, please install libelf-dev[el]);
+endif
+
 ifdef NO_DEMANGLE
 	BASIC_CFLAGS += -DNO_DEMANGLE
 else
-	has_bfd := $(shell sh -c "(echo '\#include <stdio.h>'; echo '\#include <bfd.h>'; echo '\#ifndef DMGL_PARAMS'; echo '\#define DMGL_PARAMS (1 << 0)'; echo '\#define DMGL_ANSI (1 << 1)'; echo '\#endif'; echo 'int main(int argc, char **argv) { bfd_demangle(NULL, argv[0], DMGL_PARAMS | DMGL_ANSI); return 0; }') | gcc -x c - -lbfd > /dev/null 2>&1 && echo y")
 
-	has_bfd_iberty := $(shell sh -c "(echo '\#include <stdio.h>'; echo '\#include <bfd.h>'; echo '\#ifndef DMGL_PARAMS'; echo '\#define DMGL_PARAMS (1 << 0)'; echo '\#define DMGL_ANSI (1 << 1)'; echo '\#endif'; echo 'int main(int argc, char **argv) { bfd_demangle(NULL, argv[0], DMGL_PARAMS | DMGL_ANSI); return 0; }') | gcc -x c - -lbfd -liberty > /dev/null 2>&1 && echo y")
+	has_bfd := $(shell sh -c "(echo '\#include <bfd.h>'; echo 'int main(void) { bfd_demangle(0, 0, 0); return 0; }') | $(CC) -x c - $(ALL_CFLAGS) -o /dev/null $(ALL_LDFLAGS) -lbfd > /dev/null 2>&1 && echo y")
+
+	has_bfd_iberty := $(shell sh -c "(echo '\#include <bfd.h>'; echo 'int main(void) { bfd_demangle(0, 0, 0); return 0; }') | $(CC) -x c - $(ALL_CFLAGS) -o /dev/null $(ALL_LDFLAGS) -lbfd -liberty > /dev/null 2>&1 && echo y")
 
 	ifeq ($(has_bfd),y)
 		EXTLIBS += -lbfd
 	else ifeq ($(has_bfd_iberty),y)
 		EXTLIBS += -lbfd -liberty
 	else
+		msg := $(warning No bfd.h/libbfd found, install binutils-dev[el] to gain symbol demangling)
 		BASIC_CFLAGS += -DNO_DEMANGLE
 	endif
 endif

Re: [PATCH] basic perf support for sparc

[David Miller]

From: Jens Axboe <jens.axboe@oracle.com>
Date: Wed, 29 Jul 2009 13:25:10 +0200

> -#define NR_SYSCALLS		327
> +#define NR_SYSCALLS		328

When you increase this value, you have to add entries to all of the
syscall tables.  The syscall dispatch checks against this as a limit,
so if you don't explicitly add an entry to all the tables, it's
possible to deref garbage past the end of the table and try to jump to
it as a syscall.

And if you somehow arrange for adding a compat syscall entry here for
this, and build the perf tools 32-bit, you can forego all of these
rediculious issues with trying to get a 64-bit BFD library.  If the
perf tools are written portably and use types like u64 etc. for
holding addresses and similar things, this should not be an issue.

The 32-bit sparc BFD library has full support for all the 64-bit
binary formats and whatnot.

Re: [PATCH] basic perf support for sparc

[Ingo Molnar]

* David Miller <davem@davemloft.net> wrote:

> From: Jens Axboe <jens.axboe@oracle.com>
> Date: Wed, 29 Jul 2009 13:25:10 +0200
> 
> > -#define NR_SYSCALLS		327
> > +#define NR_SYSCALLS		328
> 
> When you increase this value, you have to add entries to all of 
> the syscall tables.  The syscall dispatch checks against this as a 
> limit, so if you don't explicitly add an entry to all the tables, 
> it's possible to deref garbage past the end of the table and try 
> to jump to it as a syscall.
> 
> And if you somehow arrange for adding a compat syscall entry here 
> for this, and build the perf tools 32-bit, you can forego all of 
> these rediculious issues with trying to get a 64-bit BFD library.  
> If the perf tools are written portably and use types like u64 etc. 
> for holding addresses and similar things, this should not be an 
> issue.
> 
> The 32-bit sparc BFD library has full support for all the 64-bit 
> binary formats and whatnot.

That would work too. On x86 perf works all across the compatibility 
spectrum, and we do use strict u32/u64 typing and ABIs.

Note that we'll also solve (remove) the binutils-devel dependency, 
it creates a way too large set of external build constraints for 
perf. But in any case both 32-bit and 64-bit perf should work just 
fine.

	Ingo

Re: [PATCH] basic perf support for sparc

[Jens Axboe]

On Sun, Aug 02 2009, David Miller wrote:
> From: Jens Axboe <jens.axboe@oracle.com>
> Date: Wed, 29 Jul 2009 13:25:10 +0200
> 
> > -#define NR_SYSCALLS		327
> > +#define NR_SYSCALLS		328
> 
> When you increase this value, you have to add entries to all of the
> syscall tables.  The syscall dispatch checks against this as a limit,
> so if you don't explicitly add an entry to all the tables, it's
> possible to deref garbage past the end of the table and try to jump to
> it as a syscall.

Oops, missed the 32-bit table. Does the below look better?

> And if you somehow arrange for adding a compat syscall entry here for
> this, and build the perf tools 32-bit, you can forego all of these
> rediculious issues with trying to get a 64-bit BFD library.  If the
> perf tools are written portably and use types like u64 etc. for
> holding addresses and similar things, this should not be an issue.
> 
> The 32-bit sparc BFD library has full support for all the 64-bit
> binary formats and whatnot.

I'm assuming that this will happen automatically at some point, instead
of perf defaulting to building with the same 32/64-bit of the kernel.

diff --git a/arch/sparc/Kconfig b/arch/sparc/Kconfig
index 3f8b6a9..2f951a9 100644
--- a/arch/sparc/Kconfig
+++ b/arch/sparc/Kconfig
@@ -44,6 +44,7 @@ config SPARC64
 	select RTC_DRV_BQ4802
 	select RTC_DRV_SUN4V
 	select RTC_DRV_STARFIRE
+	select HAVE_PERF_COUNTERS
 
 config ARCH_DEFCONFIG
 	string
diff --git a/arch/sparc/include/asm/unistd.h b/arch/sparc/include/asm/unistd.h
index b2c406d..706df66 100644
--- a/arch/sparc/include/asm/unistd.h
+++ b/arch/sparc/include/asm/unistd.h
@@ -395,8 +395,9 @@
 #define __NR_preadv		324
 #define __NR_pwritev		325
 #define __NR_rt_tgsigqueueinfo	326
+#define __NR_perf_counter_open	327
 
-#define NR_SYSCALLS		327
+#define NR_SYSCALLS		328
 
 #ifdef __32bit_syscall_numbers__
 /* Sparc 32-bit only has the "setresuid32", "getresuid32" variants,
diff --git a/arch/sparc/kernel/systbls_32.S b/arch/sparc/kernel/systbls_32.S
index 6909016..0418157 100644
--- a/arch/sparc/kernel/systbls_32.S
+++ b/arch/sparc/kernel/systbls_32.S
@@ -82,5 +82,5 @@ sys_call_table:
 /*310*/	.long sys_utimensat, sys_signalfd, sys_timerfd_create, sys_eventfd, sys_fallocate
 /*315*/	.long sys_timerfd_settime, sys_timerfd_gettime, sys_signalfd4, sys_eventfd2, sys_epoll_create1
 /*320*/	.long sys_dup3, sys_pipe2, sys_inotify_init1, sys_accept4, sys_preadv
-/*325*/	.long sys_pwritev, sys_rt_tgsigqueueinfo
+/*325*/	.long sys_pwritev, sys_rt_tgsigqueueinfo, sys_perf_counter_open
 
diff --git a/arch/sparc/kernel/systbls_64.S b/arch/sparc/kernel/systbls_64.S
index 6b3ee88..80ebf20 100644
--- a/arch/sparc/kernel/systbls_64.S
+++ b/arch/sparc/kernel/systbls_64.S
@@ -158,4 +158,4 @@ sys_call_table:
 /*310*/	.word sys_utimensat, sys_signalfd, sys_timerfd_create, sys_eventfd, sys_fallocate
 	.word sys_timerfd_settime, sys_timerfd_gettime, sys_signalfd4, sys_eventfd2, sys_epoll_create1
 /*320*/	.word sys_dup3, sys_pipe2, sys_inotify_init1, sys_accept4, sys_preadv
-	.word sys_pwritev, sys_rt_tgsigqueueinfo
+	.word sys_pwritev, sys_rt_tgsigqueueinfo, sys_perf_counter_open
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index e5148e2..2abeb20 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -41,6 +41,12 @@
 #define cpu_relax()	asm volatile("" ::: "memory");
 #endif
 
+#ifdef __sparc__
+#include "../../arch/sparc/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>

-- 
Jens Axboe

Re: [PATCH] basic perf support for sparc

[David Miller]

From: Jens Axboe <jens.axboe@oracle.com>
Date: Thu, 6 Aug 2009 09:02:42 +0200

> On Sun, Aug 02 2009, David Miller wrote:
>> From: Jens Axboe <jens.axboe@oracle.com>
>> Date: Wed, 29 Jul 2009 13:25:10 +0200
>> 
>> > -#define NR_SYSCALLS		327
>> > +#define NR_SYSCALLS		328
>> 
>> When you increase this value, you have to add entries to all of the
>> syscall tables.  The syscall dispatch checks against this as a limit,
>> so if you don't explicitly add an entry to all the tables, it's
>> possible to deref garbage past the end of the table and try to jump to
>> it as a syscall.
> 
> Oops, missed the 32-bit table. Does the below look better?

I'll take a look at this as soon as I can Jens, sorry for taking
so long.

Re: [PATCH] basic perf support for sparc

[Jens Axboe]

On Wed, Aug 12 2009, David Miller wrote:
> From: Jens Axboe <jens.axboe@oracle.com>
> Date: Thu, 6 Aug 2009 09:02:42 +0200
> 
> > On Sun, Aug 02 2009, David Miller wrote:
> >> From: Jens Axboe <jens.axboe@oracle.com>
> >> Date: Wed, 29 Jul 2009 13:25:10 +0200
> >> 
> >> > -#define NR_SYSCALLS		327
> >> > +#define NR_SYSCALLS		328
> >> 
> >> When you increase this value, you have to add entries to all of the
> >> syscall tables.  The syscall dispatch checks against this as a limit,
> >> so if you don't explicitly add an entry to all the tables, it's
> >> possible to deref garbage past the end of the table and try to jump to
> >> it as a syscall.
> > 
> > Oops, missed the 32-bit table. Does the below look better?
> 
> I'll take a look at this as soon as I can Jens, sorry for taking
> so long.

No problem, I figured you'd get to it eventually and in the mean time
I'll just keep patching it in.

-- 
Jens Axboe

Re: [PATCH] basic perf support for sparc

[David Miller]

From: Jens Axboe <jens.axboe@oracle.com>
Date: Thu, 6 Aug 2009 09:02:42 +0200

> On Sun, Aug 02 2009, David Miller wrote:
>> From: Jens Axboe <jens.axboe@oracle.com>
>> Date: Wed, 29 Jul 2009 13:25:10 +0200
>> 
>> > -#define NR_SYSCALLS		327
>> > +#define NR_SYSCALLS		328
>> 
>> When you increase this value, you have to add entries to all of the
>> syscall tables.  The syscall dispatch checks against this as a limit,
>> so if you don't explicitly add an entry to all the tables, it's
>> possible to deref garbage past the end of the table and try to jump to
>> it as a syscall.
> 
> Oops, missed the 32-bit table. Does the below look better?

You got the native 32-bit table, and the native 64-bit table,
but you missed sparc64's 32-bit compat syscall table :-)

There are two syscall tables in arch/sparc/systbls.S, you only
got the native 64-bit one.

Please fix this up, and the patch is fine with that cured,
so make it a formal submission with commit message and signoffs
thanks!

> I'm assuming that this will happen automatically at some point, instead
> of perf defaulting to building with the same 32/64-bit of the kernel.

Right, it looks like Ingo and co. are going to take care of that.

Re: [PATCH] basic perf support for sparc

[Jens Axboe]

On Sun, Aug 16 2009, David Miller wrote:
> From: Jens Axboe <jens.axboe@oracle.com>
> Date: Thu, 6 Aug 2009 09:02:42 +0200
> 
> > On Sun, Aug 02 2009, David Miller wrote:
> >> From: Jens Axboe <jens.axboe@oracle.com>
> >> Date: Wed, 29 Jul 2009 13:25:10 +0200
> >> 
> >> > -#define NR_SYSCALLS		327
> >> > +#define NR_SYSCALLS		328
> >> 
> >> When you increase this value, you have to add entries to all of the
> >> syscall tables.  The syscall dispatch checks against this as a limit,
> >> so if you don't explicitly add an entry to all the tables, it's
> >> possible to deref garbage past the end of the table and try to jump to
> >> it as a syscall.
> > 
> > Oops, missed the 32-bit table. Does the below look better?
> 
> You got the native 32-bit table, and the native 64-bit table,
> but you missed sparc64's 32-bit compat syscall table :-)
> 
> There are two syscall tables in arch/sparc/systbls.S, you only
> got the native 64-bit one.
> 
> Please fix this up, and the patch is fine with that cured,
> so make it a formal submission with commit message and signoffs
> thanks!

Gotcha, I missed the compat table. So below is a full patch, signed off
and everything. It also adds the select HAVE_PERF_COUNTERS for sparc32,
but it's only tested on sparc64.

> > I'm assuming that this will happen automatically at some point, instead
> > of perf defaulting to building with the same 32/64-bit of the kernel.
> 
> Right, it looks like Ingo and co. are going to take care of that.

It's already much better, I can just type make and it'll work. Still
builds 64-bit, but at least it only requires zlib and elf which I
already got fixed.

commit 4c66ea6871ad059b35e35e915fd7910724dc0bd6
Author: Jens Axboe <jens.axboe@oracle.com>
Date:   Mon Aug 17 08:40:50 2009 +0200

    sparc: add basic support for 'perf'
    
    This wires up the perf_counter_open() syscall so that basic
    software support for perf is working.
    
    Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

diff --git a/arch/sparc/Kconfig b/arch/sparc/Kconfig
index 3f8b6a9..71a9a11 100644
--- a/arch/sparc/Kconfig
+++ b/arch/sparc/Kconfig
@@ -25,6 +25,7 @@ config SPARC
 	select ARCH_WANT_OPTIONAL_GPIOLIB
 	select RTC_CLASS
 	select RTC_DRV_M48T59
+	select HAVE_PERF_COUNTERS
 
 config SPARC32
 	def_bool !64BIT
@@ -44,6 +45,7 @@ config SPARC64
 	select RTC_DRV_BQ4802
 	select RTC_DRV_SUN4V
 	select RTC_DRV_STARFIRE
+	select HAVE_PERF_COUNTERS
 
 config ARCH_DEFCONFIG
 	string
diff --git a/arch/sparc/include/asm/unistd.h b/arch/sparc/include/asm/unistd.h
index b2c406d..706df66 100644
--- a/arch/sparc/include/asm/unistd.h
+++ b/arch/sparc/include/asm/unistd.h
@@ -395,8 +395,9 @@
 #define __NR_preadv		324
 #define __NR_pwritev		325
 #define __NR_rt_tgsigqueueinfo	326
+#define __NR_perf_counter_open	327
 
-#define NR_SYSCALLS		327
+#define NR_SYSCALLS		328
 
 #ifdef __32bit_syscall_numbers__
 /* Sparc 32-bit only has the "setresuid32", "getresuid32" variants,
diff --git a/arch/sparc/kernel/systbls_32.S b/arch/sparc/kernel/systbls_32.S
index 6909016..0418157 100644
--- a/arch/sparc/kernel/systbls_32.S
+++ b/arch/sparc/kernel/systbls_32.S
@@ -82,5 +82,5 @@ sys_call_table:
 /*310*/	.long sys_utimensat, sys_signalfd, sys_timerfd_create, sys_eventfd, sys_fallocate
 /*315*/	.long sys_timerfd_settime, sys_timerfd_gettime, sys_signalfd4, sys_eventfd2, sys_epoll_create1
 /*320*/	.long sys_dup3, sys_pipe2, sys_inotify_init1, sys_accept4, sys_preadv
-/*325*/	.long sys_pwritev, sys_rt_tgsigqueueinfo
+/*325*/	.long sys_pwritev, sys_rt_tgsigqueueinfo, sys_perf_counter_open
 
diff --git a/arch/sparc/kernel/systbls_64.S b/arch/sparc/kernel/systbls_64.S
index 6b3ee88..01e4849 100644
--- a/arch/sparc/kernel/systbls_64.S
+++ b/arch/sparc/kernel/systbls_64.S
@@ -83,7 +83,7 @@ sys_call_table32:
 /*310*/	.word compat_sys_utimensat, compat_sys_signalfd, sys_timerfd_create, sys_eventfd, compat_sys_fallocate
 	.word compat_sys_timerfd_settime, compat_sys_timerfd_gettime, compat_sys_signalfd4, sys_eventfd2, sys_epoll_create1
 /*320*/	.word sys_dup3, sys_pipe2, sys_inotify_init1, sys_accept4, compat_sys_preadv
-	.word compat_sys_pwritev, compat_sys_rt_tgsigqueueinfo
+	.word compat_sys_pwritev, compat_sys_rt_tgsigqueueinfo, sys_perf_counter_open
 
 #endif /* CONFIG_COMPAT */
 
@@ -158,4 +158,4 @@ sys_call_table:
 /*310*/	.word sys_utimensat, sys_signalfd, sys_timerfd_create, sys_eventfd, sys_fallocate
 	.word sys_timerfd_settime, sys_timerfd_gettime, sys_signalfd4, sys_eventfd2, sys_epoll_create1
 /*320*/	.word sys_dup3, sys_pipe2, sys_inotify_init1, sys_accept4, sys_preadv
-	.word sys_pwritev, sys_rt_tgsigqueueinfo
+	.word sys_pwritev, sys_rt_tgsigqueueinfo, sys_perf_counter_open
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index e5148e2..2abeb20 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -41,6 +41,12 @@
 #define cpu_relax()	asm volatile("" ::: "memory");
 #endif
 
+#ifdef __sparc__
+#include "../../arch/sparc/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>

-- 
Jens Axboe

Re: [PATCH] basic perf support for sparc

[Ingo Molnar]

* Jens Axboe <jens.axboe@oracle.com> wrote:

> On Sun, Aug 16 2009, David Miller wrote:
>
> > > I'm assuming that this will happen automatically at some point, instead
> > > of perf defaulting to building with the same 32/64-bit of the kernel.
> > 
> > Right, it looks like Ingo and co. are going to take care of that.
> 
> It's already much better, I can just type make and it'll work. 
> Still builds 64-bit, but at least it only requires zlib and elf 
> which I already got fixed.

Yes, those fixes/enhancements are already in -rc6 so if you base it 
on that or later kernels it should be fine out of box.

	Ingo

Re: [PATCH] basic perf support for sparc

[David Miller]

From: Jens Axboe <jens.axboe@oracle.com>
Date: Mon, 17 Aug 2009 08:48:51 +0200

>     sparc: add basic support for 'perf'
>     
>     This wires up the perf_counter_open() syscall so that basic
>     software support for perf is working.
>     
>     Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

Does this build for you without adding an
arch/sparc/include/asm/perf_counter.h that looks
something like the following?

#ifndef _ASM_SPARC_PERF_COUNTER_H
#define _ASM_SPARC_PERF_COUNTER_H

#define PERF_COUNTER_INDEX_OFFSET	0

#endif

Or is this somehow now required in the -tip trees?

Re: [PATCH] basic perf support for sparc

[Ingo Molnar]

* David Miller <davem@davemloft.net> wrote:

> From: Jens Axboe <jens.axboe@oracle.com>
> Date: Mon, 17 Aug 2009 08:48:51 +0200
> 
> >     sparc: add basic support for 'perf'
> >     
> >     This wires up the perf_counter_open() syscall so that basic
> >     software support for perf is working.
> >     
> >     Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
> 
> Does this build for you without adding an
> arch/sparc/include/asm/perf_counter.h that looks
> something like the following?
> 
> #ifndef _ASM_SPARC_PERF_COUNTER_H
> #define _ASM_SPARC_PERF_COUNTER_H
> 
> #define PERF_COUNTER_INDEX_OFFSET	0
> 
> #endif
> 
> Or is this somehow now required in the -tip trees?

This used to be required but i recently fixed this (and that fix is 
upstream as well) via:

  f738eb1: perf_counter: Fix the PARISC build

there's now a default define of 0 so you dont have to define it and 
can leave out this chunk.

( That index is only interesting if the architecture has a way to 
  allow unprivileged user-space to access counter registers 
  directly. In that case the index reflects the offset from the 
  (constantly changing) dynamix index which we put into the mmap 
  header. With Sparc not having a hw-PMU implementation this index 
  is entirely uninteresting at this stage. )

	Ingo

Re: [PATCH] basic perf support for sparc

[David Miller]

From: Ingo Molnar <mingo@elte.hu>
Date: Fri, 4 Sep 2009 07:02:56 +0200

> 
> * David Miller <davem@davemloft.net> wrote:
> 
>> Does this build for you without adding an
>> arch/sparc/include/asm/perf_counter.h that looks
>> something like the following?
>> 
>> #ifndef _ASM_SPARC_PERF_COUNTER_H
>> #define _ASM_SPARC_PERF_COUNTER_H
>> 
>> #define PERF_COUNTER_INDEX_OFFSET	0
>> 
>> #endif
>> 
>> Or is this somehow now required in the -tip trees?
> 
> This used to be required but i recently fixed this (and that fix is 
> upstream as well) via:
> 
>   f738eb1: perf_counter: Fix the PARISC build
> 
> there's now a default define of 0 so you dont have to define it and 
> can leave out this chunk.
> 
> ( That index is only interesting if the architecture has a way to 
>   allow unprivileged user-space to access counter registers 
>   directly. In that case the index reflects the offset from the 
>   (constantly changing) dynamix index which we put into the mmap 
>   header. With Sparc not having a hw-PMU implementation this index 
>   is entirely uninteresting at this stage. )

But you still do need at least an empty perf_counter.h file
right?  Jens must have left the file out of his submission
by accident, and that's what I'm trying to get to the bottom
of here :-)

I assume there was a similar change to deal with references to
set_perf_counter_pending() too or is at least a NOP definition
still needed?

Re: [PATCH] basic perf support for sparc

[Jens Axboe]

On Thu, Sep 03 2009, David Miller wrote:
> From: Ingo Molnar <mingo@elte.hu>
> Date: Fri, 4 Sep 2009 07:02:56 +0200
> 
> > 
> > * David Miller <davem@davemloft.net> wrote:
> > 
> >> Does this build for you without adding an
> >> arch/sparc/include/asm/perf_counter.h that looks
> >> something like the following?
> >> 
> >> #ifndef _ASM_SPARC_PERF_COUNTER_H
> >> #define _ASM_SPARC_PERF_COUNTER_H
> >> 
> >> #define PERF_COUNTER_INDEX_OFFSET	0
> >> 
> >> #endif
> >> 
> >> Or is this somehow now required in the -tip trees?
> > 
> > This used to be required but i recently fixed this (and that fix is 
> > upstream as well) via:
> > 
> >   f738eb1: perf_counter: Fix the PARISC build
> > 
> > there's now a default define of 0 so you dont have to define it and 
> > can leave out this chunk.
> > 
> > ( That index is only interesting if the architecture has a way to 
> >   allow unprivileged user-space to access counter registers 
> >   directly. In that case the index reflects the offset from the 
> >   (constantly changing) dynamix index which we put into the mmap 
> >   header. With Sparc not having a hw-PMU implementation this index 
> >   is entirely uninteresting at this stage. )
> 
> But you still do need at least an empty perf_counter.h file
> right?  Jens must have left the file out of his submission
> by accident, and that's what I'm trying to get to the bottom
> of here :-)
> 
> I assume there was a similar change to deal with references to
> set_perf_counter_pending() too or is at least a NOP definition
> still needed?

It wasn't required when I built and used it (and sent the patch), I used
the posted patch as-is. It's been a few weeks since I last updated and
ran that box, let me double check after morning coffee and send you a
fresh patch (if needed) :-)

-- 
Jens Axboe

Re: [PATCH] basic perf support for sparc

[Jens Axboe]

On Fri, Sep 04 2009, Jens Axboe wrote:
> On Thu, Sep 03 2009, David Miller wrote:
> > From: Ingo Molnar <mingo@elte.hu>
> > Date: Fri, 4 Sep 2009 07:02:56 +0200
> > 
> > > 
> > > * David Miller <davem@davemloft.net> wrote:
> > > 
> > >> Does this build for you without adding an
> > >> arch/sparc/include/asm/perf_counter.h that looks
> > >> something like the following?
> > >> 
> > >> #ifndef _ASM_SPARC_PERF_COUNTER_H
> > >> #define _ASM_SPARC_PERF_COUNTER_H
> > >> 
> > >> #define PERF_COUNTER_INDEX_OFFSET	0
> > >> 
> > >> #endif
> > >> 
> > >> Or is this somehow now required in the -tip trees?
> > > 
> > > This used to be required but i recently fixed this (and that fix is 
> > > upstream as well) via:
> > > 
> > >   f738eb1: perf_counter: Fix the PARISC build
> > > 
> > > there's now a default define of 0 so you dont have to define it and 
> > > can leave out this chunk.
> > > 
> > > ( That index is only interesting if the architecture has a way to 
> > >   allow unprivileged user-space to access counter registers 
> > >   directly. In that case the index reflects the offset from the 
> > >   (constantly changing) dynamix index which we put into the mmap 
> > >   header. With Sparc not having a hw-PMU implementation this index 
> > >   is entirely uninteresting at this stage. )
> > 
> > But you still do need at least an empty perf_counter.h file
> > right?  Jens must have left the file out of his submission
> > by accident, and that's what I'm trying to get to the bottom
> > of here :-)
> > 
> > I assume there was a similar change to deal with references to
> > set_perf_counter_pending() too or is at least a NOP definition
> > still needed?
> 
> It wasn't required when I built and used it (and sent the patch), I used
> the posted patch as-is. It's been a few weeks since I last updated and
> ran that box, let me double check after morning coffee and send you a
> fresh patch (if needed) :-)

It seems I had a left-over arch/sparc/include/asm/perf_counter.h from
earlier experiments that was never checked in, so that is why it worked
for me. include/linux/sched.h does:

#ifdef CONFIG_PERF_COUNTERS
# include <asm/perf_counter.h>
#endif

so that's no way around an empty stub file. Updated patch below, this is
what I have been using.

commit 99d9c7f3326a73e2b9952c99d6c7a15bf761858f
Author: Jens Axboe <jens.axboe@oracle.com>
Date:   Fri Sep 4 08:31:22 2009 +0200

    sparc: add basic support for 'perf'
    
    This wires up the perf_counter_open() syscall so that basic
    software support for perf is working.
    
    Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

diff --git a/arch/sparc/Kconfig b/arch/sparc/Kconfig
index 3f8b6a9..71a9a11 100644
--- a/arch/sparc/Kconfig
+++ b/arch/sparc/Kconfig
@@ -25,6 +25,7 @@ config SPARC
 	select ARCH_WANT_OPTIONAL_GPIOLIB
 	select RTC_CLASS
 	select RTC_DRV_M48T59
+	select HAVE_PERF_COUNTERS
 
 config SPARC32
 	def_bool !64BIT
@@ -44,6 +45,7 @@ config SPARC64
 	select RTC_DRV_BQ4802
 	select RTC_DRV_SUN4V
 	select RTC_DRV_STARFIRE
+	select HAVE_PERF_COUNTERS
 
 config ARCH_DEFCONFIG
 	string
diff --git a/arch/sparc/include/asm/perf_counter.h b/arch/sparc/include/asm/perf_counter.h
new file mode 100644
index 0000000..664b624
--- /dev/null
+++ b/arch/sparc/include/asm/perf_counter.h
@@ -0,0 +1,8 @@
+#ifndef __ASM_SPARC_PERF_COUNTER_H
+#define __ASM_SPARC_PERF_COUNTER_H
+
+static inline void set_perf_counter_pending(void) { }
+
+#define	PERF_COUNTER_INDEX_OFFSET	0
+
+#endif 
diff --git a/arch/sparc/include/asm/unistd.h b/arch/sparc/include/asm/unistd.h
index b2c406d..706df66 100644
--- a/arch/sparc/include/asm/unistd.h
+++ b/arch/sparc/include/asm/unistd.h
@@ -395,8 +395,9 @@
 #define __NR_preadv		324
 #define __NR_pwritev		325
 #define __NR_rt_tgsigqueueinfo	326
+#define __NR_perf_counter_open	327
 
-#define NR_SYSCALLS		327
+#define NR_SYSCALLS		328
 
 #ifdef __32bit_syscall_numbers__
 /* Sparc 32-bit only has the "setresuid32", "getresuid32" variants,
diff --git a/arch/sparc/kernel/systbls_32.S b/arch/sparc/kernel/systbls_32.S
index 6909016..0418157 100644
--- a/arch/sparc/kernel/systbls_32.S
+++ b/arch/sparc/kernel/systbls_32.S
@@ -82,5 +82,5 @@ sys_call_table:
 /*310*/	.long sys_utimensat, sys_signalfd, sys_timerfd_create, sys_eventfd, sys_fallocate
 /*315*/	.long sys_timerfd_settime, sys_timerfd_gettime, sys_signalfd4, sys_eventfd2, sys_epoll_create1
 /*320*/	.long sys_dup3, sys_pipe2, sys_inotify_init1, sys_accept4, sys_preadv
-/*325*/	.long sys_pwritev, sys_rt_tgsigqueueinfo
+/*325*/	.long sys_pwritev, sys_rt_tgsigqueueinfo, sys_perf_counter_open
 
diff --git a/arch/sparc/kernel/systbls_64.S b/arch/sparc/kernel/systbls_64.S
index 2ee7250..91b06b7 100644
--- a/arch/sparc/kernel/systbls_64.S
+++ b/arch/sparc/kernel/systbls_64.S
@@ -83,7 +83,7 @@ sys_call_table32:
 /*310*/	.word compat_sys_utimensat, compat_sys_signalfd, sys_timerfd_create, sys_eventfd, compat_sys_fallocate
 	.word compat_sys_timerfd_settime, compat_sys_timerfd_gettime, compat_sys_signalfd4, sys_eventfd2, sys_epoll_create1
 /*320*/	.word sys_dup3, sys_pipe2, sys_inotify_init1, sys_accept4, compat_sys_preadv
-	.word compat_sys_pwritev, compat_sys_rt_tgsigqueueinfo
+	.word compat_sys_pwritev, compat_sys_rt_tgsigqueueinfo, sys_perf_counter_open
 
 #endif /* CONFIG_COMPAT */
 
@@ -158,4 +158,4 @@ sys_call_table:
 /*310*/	.word sys_utimensat, sys_signalfd, sys_timerfd_create, sys_eventfd, sys_fallocate
 	.word sys_timerfd_settime, sys_timerfd_gettime, sys_signalfd4, sys_eventfd2, sys_epoll_create1
 /*320*/	.word sys_dup3, sys_pipe2, sys_inotify_init1, sys_accept4, sys_preadv
-	.word sys_pwritev, sys_rt_tgsigqueueinfo
+	.word sys_pwritev, sys_rt_tgsigqueueinfo, sys_perf_counter_open
diff --git a/tools/perf/perf.h b/tools/perf/perf.h
index e5148e2..2abeb20 100644
--- a/tools/perf/perf.h
+++ b/tools/perf/perf.h
@@ -41,6 +41,12 @@
 #define cpu_relax()	asm volatile("" ::: "memory");
 #endif
 
+#ifdef __sparc__
+#include "../../arch/sparc/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>

-- 
Jens Axboe

Re: [PATCH] basic perf support for sparc

[Ingo Molnar]

* Jens Axboe <jens.axboe@oracle.com> wrote:

> On Fri, Sep 04 2009, Jens Axboe wrote:
> > On Thu, Sep 03 2009, David Miller wrote:
> > > From: Ingo Molnar <mingo@elte.hu>
> > > Date: Fri, 4 Sep 2009 07:02:56 +0200
> > > 
> > > > 
> > > > * David Miller <davem@davemloft.net> wrote:
> > > > 
> > > >> Does this build for you without adding an
> > > >> arch/sparc/include/asm/perf_counter.h that looks
> > > >> something like the following?
> > > >> 
> > > >> #ifndef _ASM_SPARC_PERF_COUNTER_H
> > > >> #define _ASM_SPARC_PERF_COUNTER_H
> > > >> 
> > > >> #define PERF_COUNTER_INDEX_OFFSET	0
> > > >> 
> > > >> #endif
> > > >> 
> > > >> Or is this somehow now required in the -tip trees?
> > > > 
> > > > This used to be required but i recently fixed this (and that fix is 
> > > > upstream as well) via:
> > > > 
> > > >   f738eb1: perf_counter: Fix the PARISC build
> > > > 
> > > > there's now a default define of 0 so you dont have to define it and 
> > > > can leave out this chunk.
> > > > 
> > > > ( That index is only interesting if the architecture has a way to 
> > > >   allow unprivileged user-space to access counter registers 
> > > >   directly. In that case the index reflects the offset from the 
> > > >   (constantly changing) dynamix index which we put into the mmap 
> > > >   header. With Sparc not having a hw-PMU implementation this index 
> > > >   is entirely uninteresting at this stage. )
> > > 
> > > But you still do need at least an empty perf_counter.h file
> > > right?  Jens must have left the file out of his submission
> > > by accident, and that's what I'm trying to get to the bottom
> > > of here :-)
> > > 
> > > I assume there was a similar change to deal with references to
> > > set_perf_counter_pending() too or is at least a NOP definition
> > > still needed?
> > 
> > It wasn't required when I built and used it (and sent the 
> > patch), I used the posted patch as-is. It's been a few weeks 
> > since I last updated and ran that box, let me double check after 
> > morning coffee and send you a fresh patch (if needed) :-)
> 
> It seems I had a left-over arch/sparc/include/asm/perf_counter.h 
> from earlier experiments that was never checked in, so that is why 
> it worked for me. include/linux/sched.h does:
> 
> #ifdef CONFIG_PERF_COUNTERS
> # include <asm/perf_counter.h>
> #endif
> 
> so that's no way around an empty stub file. Updated patch below, 
> this is what I have been using.

That stub is not wasted: it will be filled in with real arch hw-PMU 
details, once that's implemented too.

hw-PMU support has its advantages: it can provide NMI sampling that 
allows 'perf report' to pierce irqs-off critical sections, and it 
also gives access to non-time based metrics such as instructions, 
cache-misses, etc. - depending on what the CPU can do.

That way you can tell at a glance what a workload does:

 aldebaran:~> perf stat ./hackbench 10
 Time: 0.109

 Performance counter stats for './hackbench 10':

    1191.574039  task-clock-msecs         #      7.768 CPUs 
          50363  context-switches         #      0.042 M/sec
           4249  CPU-migrations           #      0.004 M/sec
          17710  page-faults              #      0.015 M/sec
     3600730931  cycles                   #   3021.827 M/sec
     1573681316  instructions             #      0.437 IPC  
       15394883  cache-references         #     12.920 M/sec
        5005241  cache-misses             #      4.201 M/sec

    0.153389368  seconds time elapsed

without hw-PMU support it looks like this:

 venus:~> perf stat ./hackbench 1
 Time: 7.600

 Performance counter stats for './hackbench 1':

   28218.316944  task-clock-ticks     #       3.590 CPU utilization factor
          62700  context-switches     #       0.002 M/sec
          11112  CPU-migrations       #       0.000 M/sec
           1919  page-faults          #       0.000 M/sec
  <not counted>  cycles              
  <not counted>  instructions        
  <not counted>  cache-references    
  <not counted>  cache-misses        

 Wall-clock time elapsed:  7860.000004 msecs

The soft stats and time measurements work just fine - the hardware 
metrics are not counted.

	Ingo

Re: [PATCH] basic perf support for sparc

[David Miller]

From: Jens Axboe <jens.axboe@oracle.com>
Date: Fri, 4 Sep 2009 08:34:19 +0200

> It seems I had a left-over arch/sparc/include/asm/perf_counter.h from
> earlier experiments that was never checked in, so that is why it worked
> for me. include/linux/sched.h does:

That explains everything :-)

>     sparc: add basic support for 'perf'
>     
>     This wires up the perf_counter_open() syscall so that basic
>     software support for perf is working.
>     
>     Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

Applied to sparc-next-2.6, thanks!

As time allows I'll work on hw counter support.