Skip to contentSkip to conversation

basic perf support for sparc

PATCH Applied Jens Axboe 9 participants 41 messages Jul 29, 2009 – Sep 4, 2009

Subsystems (2)
lore
Share & tools
lore

Jens Axboe

author
#0 lore raw

PATCH 1/1 basic perf support for sparc

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.

arch/sparc/Kconfig +1 -0
index 3f8b6a9..2f951a9 100644
@@ -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
arch/sparc/include/asm/unistd.h +2 -1
index b2c406d..706df66 100644
@@ -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,
arch/sparc/kernel/systbls_64.S +1 -1
index 6b3ee88..80ebf20 100644
@@ -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
tools/perf/perf.h +6 -0
index e5148e2..2abeb20 100644
@@ -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>
Signature · 1 line
Jens Axboe
author
#1 lore raw
Quoted 5 lines · Jens Axboe

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?

Signature · 1 line
Jens Axboe
#2 lore raw

Hi,

Quoted 7 lines
Quoted 5 lines

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

author
#3 lore raw
Quoted 9 lines · Anton Blanchard

Hi,

Quoted 6 lines
Quoted 5 lines

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?

Quoted 8 lines
Quoted 1 line

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 :-)

Signature · 1 line
Jens Axboe
#4 lore raw

Em Sat, Aug 01, 2009 at 10:20:48AM +0200, Jens Axboe escreveu:

Quoted 15 lines · Anton Blanchard
Quoted 14 lines

Hi,

Quoted 7 lines
Quoted 5 lines

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.

Quoted 8 lines
Quoted 3 lines

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

#5 lore raw
Quoted 25 lines · Arnaldo Carvalho de Melo

Em Sat, Aug 01, 2009 at 10:20:48AM +0200, Jens Axboe escreveu:

Quoted 15 lines · Anton Blanchard
Quoted 14 lines

Hi,

Quoted 7 lines
Quoted 5 lines

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.

Quoted 8 lines
Quoted 3 lines

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 :-)

Quoted 7 lines

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

#6 lore raw
Quoted 4 lines · Ingo Molnar

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.
tools/perf/Makefile +5 -1
index a5e9b87..67ab097 100644
@@ -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
tools/perf/util/symbol.c +10 -0
index 2810605..736b8db 100644
@@ -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 */
#7 lore raw
Quoted 8 lines · Kyle McMartin
Quoted 4 lines · Ingo Molnar

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:

tools/perf/util/symbol.c line 20

Quoted 4 lines
tools/perf/util/symbol.c +4 -0
+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

#8 lore raw
Quoted 7 lines · Ingo Molnar

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

tools/perf/Makefile +1 -1
index a5e9b87..8fb2e12 100644
@@ -345,7 +345,7 @@ BUILTIN_OBJS += builtin-stat.o
 BUILTIN_OBJS += builtin-top.o
 
 PERFLIBS = $(LIB_FILE)
-EXTLIBS = -lbfd
+EXTLIBS = -ldl
 
 #
 # Platform specific tweaks
tools/perf/util/symbol.c +22 -5
index 2810605..e41cb71 100644
@@ -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:
#9 lore raw
Quoted 5 lines · Kyle McMartin
Quoted 4 lines · Ingo Molnar

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.

Quoted 7 lines
Quoted 3 lines

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

#10 lore raw
In reply to Kyle McMartin

Em Sun, Aug 02, 2009 at 04:11:32PM -0400, Kyle McMartin escreveu:

Quoted 9 lines · Ingo Molnar
Quoted 7 lines

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.

Quoted 5 lines

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

#11 lore raw

Em Sun, Aug 02, 2009 at 05:47:01PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 14 lines

Em Sun, Aug 02, 2009 at 04:11:32PM -0400, Kyle McMartin escreveu:

Quoted 13 lines · Ingo Molnar
Quoted 7 lines

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.

Quoted 5 lines
Quoted 1 line

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

tools/perf/Makefile +1 -1
index 4b20fa4..8fb2e12 100644
@@ -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
tools/perf/util/symbol.c +36 -5

67 lines not on this page · raw message

#12 lore raw
Quoted 3 lines · Arnaldo Carvalho de Melo

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. :/

Quoted 8 lines

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

#13 lore raw
In reply to Arnaldo Carvalho de Melo

tools/perf/util/symbol.c out: · line 525

Quoted 3 lines · Arnaldo Carvalho de Melo
tools/perf/util/symbol.c +3 -0
+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

#14 lore raw
Quoted 9 lines · Ingo Molnar

tools/perf/util/symbol.c out: · line 525

Quoted 3 lines · Arnaldo Carvalho de Melo
tools/perf/util/symbol.c +3 -0
+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.

maintainer
#15 lore raw

From: Peter Zijlstra <[email protected]>
Date: Tue, 04 Aug 2009 11:29:00 +0200

Quoted 11 lines · Ingo Molnar
Quoted 9 lines · Arnaldo Carvalho de Melo

tools/perf/util/symbol.c out: · line 525

Quoted 4 lines
tools/perf/util/symbol.c +3 -0
+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?

Quoted 2 lines

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.

#16 lore raw
In reply to Ingo Molnar

2009/8/4 Ingo Molnar <[email protected]>:

Quoted 10 lines · Arnaldo Carvalho de Melo

tools/perf/util/symbol.c out: · line 525

Quoted 4 lines
tools/perf/util/symbol.c +3 -0
+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.

#17 lore raw
Quoted 12 lines · Frederic Riss

2009/8/4 Ingo Molnar <[email protected]>:

Quoted 10 lines · Arnaldo Carvalho de Melo

tools/perf/util/symbol.c out: · line 525

Quoted 4 lines
tools/perf/util/symbol.c +3 -0
+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?

Quoted 7 lines

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)
#18 lore raw

2009/8/4 Peter Zijlstra <[email protected]>:

Quoted 9 lines · Frederic Riss
Quoted 7 lines

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.

Quoted 8 lines

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

#19 lore raw
In reply to Ingo Molnar

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

Ingo

-------------->

Quoted 1 line

From 238762507bb41689df191160ada1986b012b57df Mon Sep 17 00:00:00 2001

From: Ingo Molnar <[email protected]>
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 <[email protected]>
Cc: Mike Galbraith <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
LKML-Reference: <new-submission>

---
 tools/perf/util/symbol.c |   11 +++++++++++
 1 files changed, 11 insertions(+), 0 deletions(-)
tools/perf/util/symbol.c +11 -0
index b4fe057..7bd5128 100644
@@ -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
#20 lore raw
In reply to Kyle McMartin
Quoted 2 lines · Kyle McMartin

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 <[email protected]>
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.

---
 tools/perf/Makefile      |   17 ++++++++++++++++-
 tools/perf/util/symbol.c |    9 +++++++++
 2 files changed, 25 insertions(+), 1 deletion(-)
tools/perf/Makefile +16 -1
Index: 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
tools/perf/util/symbol.c +9 -0
Index: 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;
 
author
#21 lore raw
Quoted 6 lines · Peter Zijlstra
Quoted 2 lines · Kyle McMartin

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.

Signature · 1 line
Jens Axboe
#22 lore raw
Quoted 9 lines · Jens Axboe
Quoted 6 lines · Peter Zijlstra
Quoted 2 lines · Kyle McMartin

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

#23 lore raw

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

In reply to Jens Axboe
Commit-ID:  2cdbc46d7b2cb0acb68c3ecad93b000552121fa6
Gitweb:     http://git.kernel.org/tip/2cdbc46d7b2cb0acb68c3ecad93b000552121fa6
Author:     Peter Zijlstra <[email protected]>
AuthorDate: Wed, 5 Aug 2009 14:05:16 +0200
Committer:  Ingo Molnar <[email protected]>
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.

Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Kyle McMartin <[email protected]>

LKML-Reference: <[email protected]>

---
 tools/perf/Makefile      |   18 +++++++++++++++++-
 tools/perf/util/symbol.c |    9 +++++++++
 2 files changed, 26 insertions(+), 1 deletions(-)
tools/perf/Makefile +17 -1
index 4b20fa4..ff905ac 100644
@@ -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
tools/perf/util/symbol.c +9 -0
index b4fe057..0580b94 100644
@@ -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;
 
#24 lore raw

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

Quoted 9 lines · tip-bot for Peter Zijlstra
Commit-ID:  2cdbc46d7b2cb0acb68c3ecad93b000552121fa6
Gitweb:     http://git.kernel.org/tip/2cdbc46d7b2cb0acb68c3ecad93b000552121fa6
Author:     Peter Zijlstra <[email protected]>
AuthorDate: Wed, 5 Aug 2009 14:05:16 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 5 Aug 2009 14:12:08 +0200

perf: Auto-detect libbfd

Quoted 7 lines

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.

---
 tools/perf/Makefile |   15 +++++++++++----
 1 files changed, 11 insertions(+), 4 deletions(-)
tools/perf/Makefile +11 -4
index ff905ac..e9db128 100644
@@ -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
maintainer
#25 lore raw

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

Quoted 6 lines

+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)/

---

tools/perf/Makefile +12 -4
Index: 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
maintainer
#26 lore raw
In reply to Jens Axboe

From: Jens Axboe <[email protected]>
Date: Wed, 29 Jul 2009 13:25:10 +0200

arch/sparc/include/asm/unistd.h line 400

Quoted 2 lines
arch/sparc/include/asm/unistd.h +1 -1
-#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.

#27 lore raw
Quoted 19 lines · David Miller

From: Jens Axboe <[email protected]>
Date: Wed, 29 Jul 2009 13:25:10 +0200

arch/sparc/include/asm/unistd.h line 400

Quoted 2 lines
arch/sparc/include/asm/unistd.h +1 -1
-#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.

Quoted 2 lines

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

author
#28 lore raw
In reply to David Miller
Quoted 11 lines · David Miller

From: Jens Axboe <[email protected]>
Date: Wed, 29 Jul 2009 13:25:10 +0200

Review

Quoted 2 lines
(quoted diff) +1 -1
-#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?

Quoted 8 lines

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.

arch/sparc/Kconfig +1 -0
index 3f8b6a9..2f951a9 100644
@@ -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
arch/sparc/include/asm/unistd.h +2 -1
index b2c406d..706df66 100644
@@ -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,
arch/sparc/kernel/systbls_32.S +1 -1
index 6909016..0418157 100644
@@ -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
 
arch/sparc/kernel/systbls_64.S +1 -1
index 6b3ee88..80ebf20 100644
@@ -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
tools/perf/perf.h +6 -0
index e5148e2..2abeb20 100644
@@ -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>
Signature · 1 line
Jens Axboe
maintainer
#29 lore raw

From: Jens Axboe <[email protected]>
Date: Thu, 6 Aug 2009 09:02:42 +0200

Quoted 13 lines · David Miller
Quoted 11 lines

From: Jens Axboe <[email protected]>
Date: Wed, 29 Jul 2009 13:25:10 +0200

arch/sparc/include/asm/unistd.h line 400

Quoted 2 lines
arch/sparc/include/asm/unistd.h +1 -1
-#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.

Quoted 1 line

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.

author
#30 lore raw
Quoted 16 lines · David Miller

From: Jens Axboe <[email protected]>
Date: Thu, 6 Aug 2009 09:02:42 +0200

Quoted 13 lines · David Miller
Quoted 11 lines

From: Jens Axboe <[email protected]>
Date: Wed, 29 Jul 2009 13:25:10 +0200

arch/sparc/include/asm/unistd.h line 400

Quoted 2 lines
arch/sparc/include/asm/unistd.h +1 -1
-#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.

Quoted 4 lines
Quoted 1 line

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.

Signature · 1 line
Jens Axboe
maintainer
#31 lore raw
In reply to Jens Axboe

From: Jens Axboe <[email protected]>
Date: Thu, 6 Aug 2009 09:02:42 +0200

Quoted 13 lines · David Miller
Quoted 11 lines

From: Jens Axboe <[email protected]>
Date: Wed, 29 Jul 2009 13:25:10 +0200

arch/sparc/include/asm/unistd.h line 400

Quoted 2 lines
arch/sparc/include/asm/unistd.h +1 -1
-#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.

Quoted 1 line

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!

Quoted 2 lines

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.

author
#32 lore raw
Quoted 21 lines · David Miller

From: Jens Axboe <[email protected]>
Date: Thu, 6 Aug 2009 09:02:42 +0200

Quoted 13 lines · David Miller
Quoted 11 lines

From: Jens Axboe <[email protected]>
Date: Wed, 29 Jul 2009 13:25:10 +0200

Review

Quoted 2 lines
(quoted diff) +1 -1
-#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.

Quoted 1 line

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 :-)

Quoted 6 lines

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.

Quoted 4 lines
Quoted 2 lines

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 <[email protected]>
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.

arch/sparc/Kconfig +2 -0
index 3f8b6a9..71a9a11 100644
@@ -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
arch/sparc/include/asm/unistd.h +2 -1
index b2c406d..706df66 100644
@@ -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,
arch/sparc/kernel/systbls_32.S +1 -1
index 6909016..0418157 100644
@@ -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
 
arch/sparc/kernel/systbls_64.S +2 -2
index 6b3ee88..01e4849 100644
@@ -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
tools/perf/perf.h +6 -0
index e5148e2..2abeb20 100644
@@ -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>
Signature · 1 line
Jens Axboe
#33 lore raw
Quoted 10 lines · Jens Axboe
Quoted 4 lines · David Miller
Quoted 2 lines

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

maintainer
#34 lore raw
In reply to Jens Axboe

From: Jens Axboe <[email protected]>
Date: Mon, 17 Aug 2009 08:48:51 +0200

Quoted 6 lines

sparc: add basic support for 'perf'

This wires up the perf_counter_open() syscall so that basic software support for perf is working.

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?

#35 lore raw
Quoted 14 lines · David Miller

From: Jens Axboe <[email protected]>
Date: Mon, 17 Aug 2009 08:48:51 +0200

Quoted 6 lines

sparc: add basic support for 'perf'

This wires up the perf_counter_open() syscall so that basic software support for perf is working.

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

Quoted 8 lines

#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

maintainer
#36 lore raw

From: Ingo Molnar <[email protected]>
Date: Fri, 4 Sep 2009 07:02:56 +0200

Quoted 24 lines · David Miller
Quoted 5 lines

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

Quoted 8 lines

#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.

Quoted 6 lines

( 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?

author
#37 lore raw
Quoted 34 lines · David Miller

From: Ingo Molnar <[email protected]>
Date: Fri, 4 Sep 2009 07:02:56 +0200

Quoted 24 lines · David Miller
Quoted 5 lines

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

Quoted 8 lines

#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.

Quoted 6 lines

( 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. )

Quoted 8 lines

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) :-)

Signature · 1 line
Jens Axboe
author
#38 lore raw
Quoted 40 lines · Jens Axboe
Quoted 39 lines · David Miller

From: Ingo Molnar <[email protected]>
Date: Fri, 4 Sep 2009 07:02:56 +0200

Quoted 24 lines · David Miller
Quoted 5 lines

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

Quoted 8 lines

#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.

Quoted 6 lines

( 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 :-)

Quoted 8 lines
Quoted 3 lines

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 <[email protected]>
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.

arch/sparc/Kconfig +2 -0
index 3f8b6a9..71a9a11 100644
@@ -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
arch/sparc/include/asm/perf_counter.h new file +8 -0
new file mode 100644
index 0000000..664b624
@@ -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 
arch/sparc/include/asm/unistd.h +2 -1
index b2c406d..706df66 100644
@@ -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,
arch/sparc/kernel/systbls_32.S +1 -1
index 6909016..0418157 100644
@@ -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
 
arch/sparc/kernel/systbls_64.S +2 -2
index 2ee7250..91b06b7 100644
@@ -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
tools/perf/perf.h +6 -0
index e5148e2..2abeb20 100644
@@ -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>
Signature · 1 line
Jens Axboe
#39 lore raw
Quoted 54 lines · Jens Axboe
Quoted 40 lines · Jens Axboe
Quoted 39 lines · David Miller

From: Ingo Molnar <[email protected]>
Date: Fri, 4 Sep 2009 07:02:56 +0200

Quoted 24 lines · David Miller
Quoted 5 lines

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

Quoted 8 lines

#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.

Quoted 6 lines

( 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 :-)

Quoted 8 lines
Quoted 3 lines

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:

Quoted 6 lines

#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

maintainer Applied
#40 lore raw
In reply to Jens Axboe

From: Jens Axboe <[email protected]>
Date: Fri, 4 Sep 2009 08:34:19 +0200

Quoted 3 lines

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 :-)

Quoted 6 lines

sparc: add basic support for 'perf'

This wires up the perf_counter_open() syscall so that basic software support for perf is working.

Applied to sparc-next-2.6, thanks!

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