Port work-in-progress oprofile with Cortex-A9 support.

Change-Id: I26ecf3697c651dae31050081d6918ee5fc87915d

12 files changed, 255 insertions, 205 deletions

diff --git a/daemon/opd_events.c b/daemon/opd_events.c
index fa017e4..b544fb3 100644
--- a/daemon/opd_events.c
+++ b/daemon/opd_events.c
@@ -142,19 +142,8 @@ struct opd_event * find_counter_event(unsigned long counter)
 	}
 
 	fprintf(stderr, "Unknown event for counter %lu\n", counter);
-        /*
-         * ANDROID FIXME - from time to time there seems to be 1 phantom event
-         * reported from counter 3 when only counter 0 is enabled. Instead of
-         * crashing the daemon and losing tons of useful samples, we just
-         * charge the erroneous single count to the first event.
-         */
-#ifdef ANDROID
-        return &opd_events[0];
-#else
 	abort();
 	return NULL;
-
-#endif
 }
 
 
diff --git a/daemon/opd_perfmon.c b/daemon/opd_perfmon.c
index a1b158a..97f307c 100644
--- a/daemon/opd_perfmon.c
+++ b/daemon/opd_perfmon.c
@@ -30,6 +30,8 @@
 #include <stdlib.h>
 #include <string.h>
 #include <errno.h>
+#include <sys/types.h>
+#include <sys/stat.h>
 #ifdef HAVE_SCHED_SETAFFINITY
 #include <sched.h>
 #endif
@@ -98,7 +100,6 @@ static struct child * children;
 static void perfmon_start_child(int ctx_fd)
 {
 	if (perfmonctl(ctx_fd, PFM_START, 0, 0) == -1) {
-		perror("Couldn't start perfmon: ");
 		exit(EXIT_FAILURE);
 	}
 }
@@ -107,7 +108,6 @@ static void perfmon_start_child(int ctx_fd)
 static void perfmon_stop_child(int ctx_fd)
 {
 	if (perfmonctl(ctx_fd, PFM_STOP, 0, 0) == -1) {
-		perror("Couldn't stop perfmon: ");
 		exit(EXIT_FAILURE);
 	}
 }
@@ -141,7 +141,6 @@ static void child_sigusr2(int val __attribute__((unused)))
 
 static void child_sigterm(int val __attribute__((unused)))
 {
-	printf("Child received SIGTERM, killing parent.\n");
 	kill(getppid(), SIGTERM);
 }
 
@@ -149,15 +148,15 @@ static void child_sigterm(int val __attribute__((unused)))
 static void set_affinity(size_t cpu)
 {
 	cpu_set_t set;
+	int err;
 
 	CPU_ZERO(&set);
 	CPU_SET(cpu, &set);
 
-	int err = sched_setaffinity(getpid(), sizeof(set), &set);
+	err = sched_setaffinity(getpid(), sizeof(set), &set);
 
 	if (err == -1) {
-		fprintf(stderr, "Failed to set affinity: %s\n",
-			    strerror(errno));
+		perror("Failed to set affinity");
 		exit(EXIT_FAILURE);
 	}
 }
@@ -178,7 +177,7 @@ static void setup_signals(void)
 	sigemptyset(&act.sa_mask);
 
 	if (sigaction(SIGUSR1, &act, NULL)) {
-		perror("oprofiled: install of SIGUSR1 handler failed: ");
+		perror("oprofiled: install of SIGUSR1 handler failed");
 		exit(EXIT_FAILURE);
 	}
 
@@ -187,7 +186,7 @@ static void setup_signals(void)
 	sigemptyset(&act.sa_mask);
 
 	if (sigaction(SIGUSR2, &act, NULL)) {
-		perror("oprofiled: install of SIGUSR2 handler failed: ");
+		perror("oprofiled: install of SIGUSR2 handler failed");
 		exit(EXIT_FAILURE);
 	}
 
@@ -196,7 +195,7 @@ static void setup_signals(void)
 	sigemptyset(&act.sa_mask);
 
 	if (sigaction(SIGTERM, &act, NULL)) {
-		perror("oprofiled: install of SIGTERM handler failed: ");
+		perror("oprofiled: install of SIGTERM handler failed");
 		exit(EXIT_FAILURE);
 	}
 }
@@ -214,8 +213,7 @@ static void create_context(struct child * self)
 
 	err = perfmonctl(0, PFM_CREATE_CONTEXT, &ctx, 1);
 	if (err == -1) {
-		fprintf(stderr, "CREATE_CONTEXT failed: %s\n",
-		        strerror(errno));
+		perror("CREATE_CONTEXT failed");
 		exit(EXIT_FAILURE);
 	}
 
@@ -268,13 +266,13 @@ static void write_pmu(struct child * self)
 
 	err = perfmonctl(self->ctx_fd, PFM_WRITE_PMCS, pc, i);
 	if (err == -1) {
-		perror("Couldn't write PMCs: ");
+		perror("Couldn't write PMCs");
 		exit(EXIT_FAILURE);
 	}
 
 	err = perfmonctl(self->ctx_fd, PFM_WRITE_PMDS, pd, i);
 	if (err == -1) {
-		perror("Couldn't write PMDs: ");
+		perror("Couldn't write PMDs");
 		exit(EXIT_FAILURE);
 	}
 }
@@ -290,7 +288,7 @@ static void load_context(struct child * self)
 
 	err = perfmonctl(self->ctx_fd, PFM_LOAD_CONTEXT, &load_args, 1);
 	if (err == -1) {
-		perror("Couldn't load context: ");
+		perror("Couldn't load context");
 		exit(EXIT_FAILURE);
 	}
 }
@@ -304,13 +302,17 @@ static void notify_parent(struct child * self, size_t cpu)
 		if (ret == sizeof(size_t))
 			break;
 		if (ret < 0 && errno != EINTR) {
-			fprintf(stderr, "Failed to write child pipe with %s\n",
-			        strerror(errno));
+			perror("Failed to write child pipe:");
 			exit(EXIT_FAILURE);
 		}
 	}
 }
 
+static struct child * inner_child;
+void close_pipe(void)
+{
+	close(inner_child->up_pipe[1]);
+}
 
 static void run_child(size_t cpu)
 {
@@ -321,6 +323,19 @@ static void run_child(size_t cpu)
 	self->sigusr2 = 0;
 	self->sigterm = 0;
 
+	inner_child = self;
+	if (atexit(close_pipe)){
+		close_pipe();
+		exit(EXIT_FAILURE);
+	}
+
+	umask(0);
+	/* Change directory to allow directory to be removed */
+	if (chdir("/") < 0) {
+		perror("Unable to chdir to \"/\"");
+		exit(EXIT_FAILURE);
+	}
+
 	setup_signals();
 
 	set_affinity(cpu);
@@ -333,6 +348,11 @@ static void run_child(size_t cpu)
 
 	notify_parent(self, cpu);
 
+	/* Redirect standard files to /dev/null */
+	freopen( "/dev/null", "r", stdin);
+	freopen( "/dev/null", "w", stdout);
+	freopen( "/dev/null", "w", stderr);
+
 	for (;;) {
 		sigset_t sigmask;
 		sigfillset(&sigmask);
@@ -341,15 +361,11 @@ static void run_child(size_t cpu)
 		sigdelset(&sigmask, SIGTERM);
 
 		if (self->sigusr1) {
-			printf("PFM_START on CPU%d\n", (int)cpu);
-			fflush(stdout);
 			perfmon_start_child(self->ctx_fd);
 			self->sigusr1 = 0;
 		}
 
 		if (self->sigusr2) {
-			printf("PFM_STOP on CPU%d\n", (int)cpu);
-			fflush(stdout);
 			perfmon_stop_child(self->ctx_fd);
 			self->sigusr2 = 0;
 		}
@@ -367,9 +383,8 @@ static void wait_for_child(struct child * child)
 		ret = read(child->up_pipe[0], &tmp, sizeof(size_t));
 		if (ret == sizeof(size_t))
 			break;
-		if (ret < 0 && errno != EINTR) {
-			fprintf(stderr, "Failed to read child pipe with %s\n",
-			        strerror(errno));
+		if ((ret < 0 && errno != EINTR) || ret == 0 ) {
+			perror("Failed to read child pipe");
 			exit(EXIT_FAILURE);
 		}
 	}
@@ -377,7 +392,6 @@ static void wait_for_child(struct child * child)
 	fflush(stdout);
 
 	close(child->up_pipe[0]);
-	close(child->up_pipe[1]);
 }
 
 static struct child* xen_ctx;
@@ -417,25 +431,26 @@ void perfmon_init(void)
 	nr_cpus = nr;
 
 	children = xmalloc(sizeof(struct child) * nr_cpus);
+	bzero(children, sizeof(struct child) * nr_cpus);
 
 	for (i = 0; i < nr_cpus; ++i) {
 		int ret;
 
 		if (pipe(children[i].up_pipe)) {
-			perror("Couldn't create child pipe.\n");
+			perror("Couldn't create child pipe");
 			exit(EXIT_FAILURE);
 		}
 
 		ret = fork();
 		if (ret == -1) {
-			fprintf(stderr, "Couldn't fork perfmon child.\n");
+			perror("Couldn't fork perfmon child");
 			exit(EXIT_FAILURE);
 		} else if (ret == 0) {
-			printf("Running perfmon child on CPU%d.\n", (int)i);
-			fflush(stdout);
+			close(children[i].up_pipe[0]);
 			run_child(i);
 		} else {
 			children[i].pid = ret;
+			close(children[i].up_pipe[1]);
 			printf("Waiting on CPU%d\n", (int)i);
 			wait_for_child(&children[i]);
 		}
@@ -454,8 +469,12 @@ void perfmon_exit(void)
 		return;
 
 	for (i = 0; i < nr_cpus; ++i) {
-		kill(children[i].pid, SIGKILL);
-		waitpid(children[i].pid, NULL, 0);
+		if (children[i].pid) {
+			int c_pid = children[i].pid;
+			children[i].pid = 0;
+			if (kill(c_pid, SIGKILL)==0)
+				waitpid(c_pid, NULL, 0);
+		}
 	}
 }
 
@@ -472,8 +491,12 @@ void perfmon_start(void)
 		return;
 	}
 
-	for (i = 0; i < nr_cpus; ++i)
-		kill(children[i].pid, SIGUSR1);
+	for (i = 0; i < nr_cpus; ++i) {
+		if (kill(children[i].pid, SIGUSR1)) {
+			perror("Unable to start perfmon");
+			exit(EXIT_FAILURE);
+		}
+	}
 }
 
 
@@ -490,7 +513,10 @@ void perfmon_stop(void)
 	}
 	
 	for (i = 0; i < nr_cpus; ++i)
-		kill(children[i].pid, SIGUSR2);
+		if (kill(children[i].pid, SIGUSR2)) {
+			perror("Unable to stop perfmon");
+			exit(EXIT_FAILURE);
+		}
 }
 
 #endif /* __ia64__ */
diff --git a/daemon/opd_pipe.c b/daemon/opd_pipe.c
index a5c334a..3c81979 100644
--- a/daemon/opd_pipe.c
+++ b/daemon/opd_pipe.c
@@ -22,6 +22,7 @@
 #include <sys/stat.h>
 
 static int fifo;
+static FILE * fifo_fd = NULL;
 
 void opd_create_pipe(void)
 {
@@ -48,6 +49,8 @@ void opd_open_pipe(void)
 
 void opd_close_pipe(void)
 {
+	if (fifo_fd)
+		fclose(fifo_fd);
 	close(fifo);
 }
 
@@ -58,14 +61,14 @@ int is_jitconv_requested(void)
 	static long nr_drops = 0;
 	/* modulus to output only a few warnings to avoid flooding oprofiled.log */
 	static int mod_cnt_drops = 1;
-	FILE * fd;
 	char line[256];
 	int i, ret = 0;
 
 	/* get a file descriptor to the pipe */
-	fd = fdopen(fifo, "r");
+	if (!fifo_fd)
+		fifo_fd = fdopen(fifo, "r");
 
-	if (fd == NULL) {
+	if (fifo_fd == NULL) {
 		perror("oprofiled: couldn't create file descriptor: ");
 		exit(EXIT_FAILURE);
 	}
@@ -73,7 +76,7 @@ int is_jitconv_requested(void)
 	/* read up to 99 lines to check for 'do_jitconv' */
 	for (i = 0; i < 99; i++) {
 		/* just break if no new line is found */
-		if (fgets(line, 256, fd) == NULL)
+		if (fgets(line, 256, fifo_fd) == NULL)
 			break;
 		line[strlen(line) - 1] = '\0';
 
diff --git a/libop/op_cpu_type.c b/libop/op_cpu_type.c
index e168b43..4633fd1 100644
--- a/libop/op_cpu_type.c
+++ b/libop/op_cpu_type.c
@@ -51,7 +51,7 @@ static struct cpu_descr const cpu_descrs[MAX_CPU_TYPE] = {
 	{ "MIPS 20K", "mips/20K", CPU_MIPS_20K, 1},
 	{ "MIPS 24K", "mips/24K", CPU_MIPS_24K, 2},
 	{ "MIPS 25K", "mips/25K", CPU_MIPS_25K, 2},
-	{ "MIPS 34K", "mips/34K", CPU_MIPS_34K, 4},
+	{ "MIPS 34K", "mips/34K", CPU_MIPS_34K, 2},
 	{ "MIPS 5K", "mips/5K", CPU_MIPS_5K, 2},
 	{ "MIPS R10000", "mips/r10000", CPU_MIPS_R10000, 2 },
 	{ "MIPS R12000", "mips/r12000", CPU_MIPS_R12000, 4 },
@@ -70,18 +70,23 @@ static struct cpu_descr const cpu_descrs[MAX_CPU_TYPE] = {
 	{ "ppc64 Cell Broadband Engine", "ppc64/cell-be", CPU_PPC64_CELL, 8 },
 	{ "AMD64 family10", "x86-64/family10", CPU_FAMILY10, 4 },
 	{ "ppc64 PA6T", "ppc64/pa6t", CPU_PPC64_PA6T, 6 },
-	{ "ARM MPCore", "arm/mpcore", CPU_ARM_MPCORE, 2 },
+	{ "ARM 11MPCore", "arm/mpcore", CPU_ARM_MPCORE, 2 },
 	{ "ARM V6 PMU", "arm/armv6", CPU_ARM_V6, 3 },
 	{ "ppc64 POWER5++", "ppc64/power5++", CPU_PPC64_POWER5pp, 6 },
 	{ "e300", "ppc/e300", CPU_PPC_E300, 4 },
 	{ "AVR32", "avr32", CPU_AVR32, 3 },
-	{ "ARM V7 PMNC", "arm/armv7", CPU_ARM_V7, 5 },
+	{ "ARM Cortex-A8", "arm/armv7", CPU_ARM_V7, 5 },
  	{ "Intel Architectural Perfmon", "i386/arch_perfmon", CPU_ARCH_PERFMON, 0},
 	{ "AMD64 family11h", "x86-64/family11h", CPU_FAMILY11H, 4 },
 	{ "ppc64 POWER7", "ppc64/power7", CPU_PPC64_POWER7, 6 },
 	{ "ppc64 compat version 1", "ppc64/ibm-compat-v1", CPU_PPC64_IBM_COMPAT_V1, 4 },
    	{ "Intel Core/i7", "i386/core_i7", CPU_CORE_I7, 4 },
    	{ "Intel Atom", "i386/atom", CPU_ATOM, 2 },
+	{ "Loongson2", "mips/loongson2", CPU_MIPS_LOONGSON2, 2 },
+	{ "Intel Nehalem microarchitecture", "i386/nehalem", CPU_NEHALEM, 4 },
+	{ "ARM Cortex-A9", "arm/armv7-ca9", CPU_ARM_V7_CA9, 7 },
+	{ "MIPS 74K", "mips/74K", CPU_MIPS_74K, 4},
+	{ "MIPS 1004K", "mips/1004K", CPU_MIPS_1004K, 2},
 };
  
 static size_t const nr_cpu_descrs = sizeof(cpu_descrs) / sizeof(struct cpu_descr);
diff --git a/libop/op_cpu_type.h b/libop/op_cpu_type.h
index 133a4f8..990e213 100644
--- a/libop/op_cpu_type.h
+++ b/libop/op_cpu_type.h
@@ -72,13 +72,18 @@ typedef enum {
 	CPU_PPC64_POWER5pp,  /**< ppc64 Power5++ family */
 	CPU_PPC_E300, /**< e300 */
 	CPU_AVR32, /**< AVR32 */
-	CPU_ARM_V7, /**< ARM V7 */
+	CPU_ARM_V7, /**< ARM Cortex-A8 */
  	CPU_ARCH_PERFMON, /**< Intel architectural perfmon */
 	CPU_FAMILY11H, /**< AMD family 11h */
 	CPU_PPC64_POWER7, /**< ppc64 POWER7 family */
 	CPU_PPC64_IBM_COMPAT_V1, /**< IBM PPC64 processor compat mode version 1 */
    	CPU_CORE_I7, /* Intel Core i7, Nehalem */
    	CPU_ATOM, /* First generation Intel Atom */
+	CPU_MIPS_LOONGSON2, /* < loongson2 family */
+	CPU_NEHALEM, /* Intel Nehalem microarchitecture */
+	CPU_ARM_V7_CA9, /**< ARM Cortex-A9 */
+	CPU_MIPS_74K, /**< MIPS 74K */
+	CPU_MIPS_1004K, /**< MIPS 1004K */
 	MAX_CPU_TYPE
 } op_cpu;
 
diff --git a/libop/op_events.c b/libop/op_events.c
index ad95d86..c3fcf57 100644
--- a/libop/op_events.c
+++ b/libop/op_events.c
@@ -970,6 +970,8 @@ void op_default_event(op_cpu cpu_type, struct op_default_event_descr * descr)
 		case CPU_FAMILY11H:
  		case CPU_ATOM:
  		case CPU_CORE_I7:
+		case CPU_NEHALEM:
+		case CPU_MIPS_LOONGSON2:
 			descr->name = "CPU_CLK_UNHALTED";
 			break;
 
@@ -1005,6 +1007,7 @@ void op_default_event(op_cpu cpu_type, struct op_default_event_descr * descr)
 		case CPU_ARM_MPCORE:
 		case CPU_ARM_V6:
 		case CPU_ARM_V7:
+		case CPU_ARM_V7_CA9:
 		case CPU_AVR32:
 			descr->name = "CPU_CYCLES";
 			break;
@@ -1029,10 +1032,9 @@ void op_default_event(op_cpu cpu_type, struct op_default_event_descr * descr)
 			break;
 
 		case CPU_MIPS_24K:
-			descr->name = "INSTRUCTIONS";
-			break;
-
 		case CPU_MIPS_34K:
+		case CPU_MIPS_74K:
+		case CPU_MIPS_1004K:
 			descr->name = "INSTRUCTIONS";
 			break;
 
diff --git a/libop/op_hw_specific.h b/libop/op_hw_specific.h
index b1e3b4c..35080ad 100644
--- a/libop/op_hw_specific.h
+++ b/libop/op_hw_specific.h
@@ -20,8 +20,7 @@ static inline int cpuid_vendor(char *vnd)
 		char v[12];
 	} v;
 	unsigned eax;
-	asm volatile(	"pushl %%ebx; cpuid; movl %%ebx, %1; popl %%ebx"
-			: "=a" (eax), "=S" (v.b), "=c" (v.c), "=d" (v.d) : "0" (0));
+	asm("cpuid" : "=a" (eax), "=b" (v.b), "=c" (v.c), "=d" (v.d) : "0" (0));
 	return !strncmp(v.v, vnd, 12);
 }
 
@@ -47,8 +46,7 @@ static inline void workaround_nehalem_aaj79(unsigned *ebx)
 
 	if (!cpuid_vendor("GenuineIntel"))
 		return;
-	asm volatile(	"pushl %%ebx; cpuid; movl %%ebx, %1; popl %%ebx"
-			: "=a" (v.eax) : "0" (1) : "ecx","edx");
+	asm("cpuid" : "=a" (v.eax) : "0" (1) : "ecx","ebx","edx");
 	model = (v.ext_model << 4) + v.model;
 	if (v.family != 6 || model != 26 || v.stepping > 4)
 		return;
@@ -59,8 +57,7 @@ static inline unsigned arch_get_filter(op_cpu cpu_type)
 {
 	if (cpu_type == CPU_ARCH_PERFMON) { 
 		unsigned ebx, eax;
-		asm volatile(	"pushl %%ebx; cpuid; movl %%ebx, %1; popl %%ebx"
-				: "=a" (eax), "=S" (ebx) : "0" (0xa) : "ecx","edx");
+		asm("cpuid" : "=a" (eax), "=b" (ebx) : "0" (0xa) : "ecx","edx");
 		workaround_nehalem_aaj79(&ebx);
 		return ebx & num_to_mask(eax >> 24);
 	}
@@ -71,8 +68,7 @@ static inline int arch_num_counters(op_cpu cpu_type)
 {
 	if (cpu_type == CPU_ARCH_PERFMON) {
 		unsigned v;
-		asm volatile(	"pushl %%ebx; cpuid; movl %%eax, %1; popl %%ebx"
-				: "=a" (v) : "0" (0xa) : "ecx","edx");
+		asm("cpuid" : "=a" (v) : "0" (0xa) : "ebx","ecx","edx");
 		return (v >> 8) & 0xff;
 	} 
 	return -1;
@@ -81,8 +77,7 @@ static inline int arch_num_counters(op_cpu cpu_type)
 static inline unsigned arch_get_counter_mask(void)
 {
 	unsigned v;
-	asm volatile(	"pushl %%ebx; cpuid; movl %%ebx, %1; popl %%ebx"
-			: "=a" (v) : "0" (0xa) : "ecx","edx");
+	asm("cpuid" : "=a" (v) : "0" (0xa) : "ebx","ecx","edx");
 	return num_to_mask((v >> 8) & 0xff);	
 }
 
diff --git a/libop/op_xml_events.c b/libop/op_xml_events.c
index 5b9ac7d..67fc0a5 100644
--- a/libop/op_xml_events.c
+++ b/libop/op_xml_events.c
@@ -16,38 +16,38 @@
 #include "op_xml_out.h"
 
 static op_cpu cpu_type;
-#define MAX_BUFFER 4096
+#define MAX_BUFFER 16384
+static char buffer[MAX_BUFFER];
+
 void open_xml_events(char const * title, char const * doc, op_cpu the_cpu_type)
 {
 	char const * schema_version = "1.0";
-	char buffer[MAX_BUFFER];
 
 	buffer[0] = '\0';
 	cpu_type = the_cpu_type;
-	open_xml_element(HELP_EVENTS, 0, buffer);
-	open_xml_element(HELP_HEADER, 1, buffer);
-	init_xml_str_attr(HELP_TITLE, title, buffer);
-	init_xml_str_attr(SCHEMA_VERSION, schema_version, buffer);
-	init_xml_str_attr(HELP_DOC, doc, buffer);
-	close_xml_element(NONE, 0, buffer);
+	open_xml_element(HELP_EVENTS, 0, buffer, MAX_BUFFER);
+	open_xml_element(HELP_HEADER, 1, buffer, MAX_BUFFER);
+	init_xml_str_attr(HELP_TITLE, title, buffer, MAX_BUFFER);
+	init_xml_str_attr(SCHEMA_VERSION, schema_version, buffer, MAX_BUFFER);
+	init_xml_str_attr(HELP_DOC, doc, buffer, MAX_BUFFER);
+	close_xml_element(NONE, 0, buffer, MAX_BUFFER);
 	printf("%s", buffer);
 }
 
 void close_xml_events(void)
 {
-	char buffer[MAX_BUFFER];
-
 	buffer[0] = '\0';
-	close_xml_element(HELP_EVENTS, 0, buffer);
+	close_xml_element(HELP_EVENTS, 0, buffer, MAX_BUFFER);
 	printf("%s", buffer);
 }
 
-static void xml_do_arch_specific_event_help(struct op_event const * event,
-					    char * buffer)
+static void xml_do_arch_specific_event_help(struct op_event const *event,
+					    char *buffer, size_t size)
 {
 	switch (cpu_type) {
 	case CPU_PPC64_CELL:
-		init_xml_int_attr(HELP_EVENT_GROUP, event->val / 100, buffer);
+		init_xml_int_attr(HELP_EVENT_GROUP, event->val / 100, buffer,
+				  size);
 		break;
 	default:
 		break;
@@ -60,34 +60,41 @@ void xml_help_for_event(struct op_event const * event)
 	uint i;
 	int nr_counters;
 	int has_nested = strcmp(event->unit->name, "zero");
-	char buffer[MAX_BUFFER];
 
 	buffer[0] = '\0';
-	open_xml_element(HELP_EVENT, 1, buffer);
-	init_xml_str_attr(HELP_EVENT_NAME, event->name, buffer);
-	xml_do_arch_specific_event_help(event, buffer);
-	init_xml_str_attr(HELP_EVENT_DESC, event->desc, buffer);
+	open_xml_element(HELP_EVENT, 1, buffer, MAX_BUFFER);
+	init_xml_str_attr(HELP_EVENT_NAME, event->name, buffer, MAX_BUFFER);
+	xml_do_arch_specific_event_help(event, buffer, MAX_BUFFER);
+	init_xml_str_attr(HELP_EVENT_DESC, event->desc, buffer, MAX_BUFFER);
 
 	nr_counters = op_get_nr_counters(cpu_type);
-	init_xml_int_attr(HELP_COUNTER_MASK, event->counter_mask, buffer);
-	init_xml_int_attr(HELP_MIN_COUNT, event->min_count, buffer);
+	init_xml_int_attr(HELP_COUNTER_MASK, event->counter_mask, buffer,
+			  MAX_BUFFER);
+	if (event->ext)
+		init_xml_str_attr(HELP_EXT, event->ext, buffer, MAX_BUFFER);
+	init_xml_int_attr(HELP_MIN_COUNT, event->min_count,
+			  buffer, MAX_BUFFER);
 
 	if (has_nested) {
-		close_xml_element(NONE, 1, buffer);
-		open_xml_element(HELP_UNIT_MASKS, 1, buffer);
-		init_xml_int_attr(HELP_DEFAULT_MASK, event->unit->default_mask, buffer);
-		close_xml_element(NONE, 1, buffer);
+		close_xml_element(NONE, 1, buffer, MAX_BUFFER);
+		open_xml_element(HELP_UNIT_MASKS, 1, buffer, MAX_BUFFER);
+		init_xml_int_attr(HELP_DEFAULT_MASK, event->unit->default_mask,
+				  buffer, MAX_BUFFER);
+		close_xml_element(NONE, 1, buffer, MAX_BUFFER);
 		for (i = 0; i < event->unit->num; i++) {
-			open_xml_element(HELP_UNIT_MASK, 1, buffer);
+			open_xml_element(HELP_UNIT_MASK, 1, buffer, MAX_BUFFER);
 			init_xml_int_attr(HELP_UNIT_MASK_VALUE,
-					  event->unit->um[i].value, buffer);
+					  event->unit->um[i].value,
+					  buffer, MAX_BUFFER);
 			init_xml_str_attr(HELP_UNIT_MASK_DESC,
-					  event->unit->um[i].desc, buffer);
-			close_xml_element(NONE, 0, buffer);
+					  event->unit->um[i].desc,
+					  buffer, MAX_BUFFER);
+			close_xml_element(NONE, 0, buffer, MAX_BUFFER);
 		}
-		close_xml_element(HELP_UNIT_MASKS, 0, buffer);
+		close_xml_element(HELP_UNIT_MASKS, 0, buffer, MAX_BUFFER);
 	}
-	close_xml_element(has_nested ? HELP_EVENT : NONE, has_nested, buffer);
+	close_xml_element(has_nested ? HELP_EVENT : NONE, has_nested,
+			  buffer, MAX_BUFFER);
 	printf("%s", buffer);
 }
 
diff --git a/libop/op_xml_out.c b/libop/op_xml_out.c
index d779c45..b12919e 100644
--- a/libop/op_xml_out.c
+++ b/libop/op_xml_out.c
@@ -77,6 +77,7 @@ char const * xml_tag_map[] = {
 		"desc",
 		"counter_mask",
 		"min_count",
+		"ext",
 	"unit_masks",
 		"default",
 	"unit_mask",
@@ -91,143 +92,176 @@ char const * xml_tag_name(tag_t tag)
 }
 
 
-void open_xml_element(tag_t tag, int with_attrs, char * buffer)
+void open_xml_element(tag_t tag, int with_attrs, char *buffer, size_t max)
 {
-	char const * tag_name = xml_tag_name(tag);
-	unsigned int const max_len = strlen(tag_name) + 3;
-	char tmp_buf[MAX_BUF_LEN];
+	char *buf;
+	int size, ret;
 
-	if (max_len >= sizeof(tmp_buf))
-		fprintf(stderr,"Warning: open_xml_element: buffer overflow %d\n", max_len);
+	buffer[max - 1] = '\0';
+	size = strlen(buffer);
+	buf = &buffer[size];
+	size = max - 1 - size;
 
-	if (snprintf(tmp_buf, sizeof(tmp_buf), "<%s%s", tag_name,
-		(with_attrs ? " " : ">\n")) < 0) {
+	ret = snprintf(buf, size, "<%s%s", xml_tag_name(tag),
+		       (with_attrs ? " " : ">\n"));
+
+	if (ret < 0 || ret >= size) {
 		fprintf(stderr,"open_xml_element: snprintf failed\n");
 		exit(EXIT_FAILURE);
 	}
-	strncat(buffer, tmp_buf, sizeof(tmp_buf));
 }
 
 
-void close_xml_element(tag_t tag, int has_nested, char * buffer)
+void close_xml_element(tag_t tag, int has_nested, char *buffer, size_t max)
 {
-	char const * tag_name = xml_tag_name(tag);
-	unsigned int const max_len = strlen(tag_name) + 3;
-	char tmp_buf[MAX_BUF_LEN];
+	char *buf;
+	int size, ret;
 
-	if (max_len >= sizeof(tmp_buf))
-		fprintf(stderr,"Warning: close_xml_element: buffer overflow %d\n", max_len);
+	buffer[max - 1] = '\0';
+	size = strlen(buffer);
+	buf = &buffer[size];
+	size = max - 1 - size;
 
-	if (tag == NONE) {
-		if (snprintf(tmp_buf, sizeof(tmp_buf), "%s\n", (has_nested ? ">" : "/>")) < 0) {
-			fprintf(stderr, "close_xml_element: snprintf failed\n");
-			exit(EXIT_FAILURE);
-		}
-	} else {
-		if (snprintf(tmp_buf, sizeof(tmp_buf), "</%s>\n", tag_name) < 0) {
-			fprintf(stderr, "close_xml_element: snprintf failed\n");
-			exit(EXIT_FAILURE);
-		}
+	if (tag == NONE)
+		ret = snprintf(buf, size, "%s\n", (has_nested ? ">" : "/>"));
+	else
+		ret = snprintf(buf, size, "</%s>\n", xml_tag_name(tag));
+
+	if (ret < 0 || ret >= size) {
+		fprintf(stderr, "close_xml_element: snprintf failed\n");
+		exit(EXIT_FAILURE);
 	}
-	strncat(buffer, tmp_buf, sizeof(tmp_buf));
 }
 
 
-void init_xml_int_attr(tag_t attr, int value, char * buffer)
+void init_xml_int_attr(tag_t attr, int value, char *buffer, size_t max)
 {
-	char const * attr_name = xml_tag_name(attr);
-	char tmp_buf[MAX_BUF_LEN];
-	unsigned int const max_len = strlen(attr_name) + 50;
+	char *buf;
+	int size, ret;
 
-	if (max_len >= sizeof(tmp_buf)) {
-		fprintf(stderr,
-			"Warning: init_xml_int_attr: buffer overflow %d\n", max_len);
-	}
+	buffer[max - 1] = '\0';
+	size = strlen(buffer);
+	buf = &buffer[size];
+	size = max - 1 - size;
 
+	ret = snprintf(buf, size, " %s=\"%d\"", xml_tag_name(attr), value);
 
-	if (snprintf(tmp_buf, sizeof(tmp_buf), " %s=\"%d\"", attr_name, value) < 0) {
+	if (ret < 0 || ret >= size) {
 		fprintf(stderr,"init_xml_int_attr: snprintf failed\n");
 		exit(EXIT_FAILURE);
 	}
-	strncat(buffer, tmp_buf, sizeof(tmp_buf));
 }
 
 
-void init_xml_dbl_attr(tag_t attr, double value, char * buffer)
+void init_xml_dbl_attr(tag_t attr, double value, char *buffer, size_t max)
 {
-	char const * attr_name = xml_tag_name(attr);
-	unsigned int const max_len = strlen(attr_name) + 50;
-	char tmp_buf[MAX_BUF_LEN];
+	char *buf;
+	int size, ret;
+
+	buffer[max - 1] = '\0';
+	size = strlen(buffer);
+	buf = &buffer[size];
+	size = max - 1 - size;
 
-	if (max_len >= sizeof(tmp_buf))
-		fprintf(stderr, "Warning: init_xml_dbl_attr: buffer overflow %d\n", max_len);
+	ret = snprintf(buf, size, " %s=\"%.2f\"", xml_tag_name(attr), value);
 
-	if (snprintf(tmp_buf, sizeof(tmp_buf), " %s=\"%.2f\"", attr_name, value) < 0) {
+	if (ret < 0 || ret >= size) {
 		fprintf(stderr, "init_xml_dbl_attr: snprintf failed\n");
 		exit(EXIT_FAILURE);
 	}
-	strncat(buffer, tmp_buf, sizeof(tmp_buf));
 }
 
 
-static char * xml_quote(char const * str, char * quote_buf)
+static void xml_quote(char const *str, char *buffer, size_t max)
 {
-	int i;
-	int pos = 0;
-	int len = strlen(str);
+	char *buf;
+	char *quote;
+	char *pos = (char*)str;
+	size_t size;
+	int ret;
 
-	
-	quote_buf[pos++] = '"';
+	buffer[max - 1] = '\0';
+	size = strlen(buffer);
+	buf = &buffer[size];
+	size = max - 1 - size;
 
-	for (i = 0; i < len; i++) {
-		if (pos >= MAX_BUF_LEN - 10) {
-			fprintf(stderr,"quote_str: buffer overflow %d\n", pos);
-			exit(EXIT_FAILURE);
-		}
+	if (size < strlen(pos) + 2)
+		goto Error;
 
-		switch(str[i]) {
+	*buf = '"';
+	buf++;
+	size--;
+
+	while (*pos) {
+		switch(*pos) {
 		case '&':
-			strncpy(quote_buf + pos, "&amp;", 5);
-			pos += 5;
+			quote = "&amp;";
 			break;
 		case '<':
-			strncpy(quote_buf + pos, "&lt;", 4);
-			pos += 4;
+			quote = "&lt;";
 			break;
 		case '>':
-			strncpy(quote_buf + pos, "&gt;", 4);
-			pos += 4;
+                        quote = "&gt;";
 			break;
 		case '"':
-			strncpy(quote_buf + pos, "&quot;", 6);
-			pos += 6;
+			quote = "&quot;";
 			break;
 		default:
-			quote_buf[pos++] = str[i];
-			break;
+			*buf = *pos;
+			pos++;
+			buf++;
+			size--;
+			continue;
 		}
+
+		pos++;
+		ret = snprintf(buf, size, "%s", quote);
+		if (ret < 0 || ret >= (int)size)
+			goto Error;
+		buf += ret;
+		size -= ret;
+		if (size < strlen(pos))
+			goto Error;
 	}
 
-	quote_buf[pos++] = '"';
-	quote_buf[pos++] = '\0';
-	return quote_buf;
+	if (!size)
+		goto Error;
+
+	*buf = '"';
+	buf++;
+	*buf = '\0';
+
+	return;
+
+Error:
+	fprintf(stderr,"quote_str: buffer overflow\n");
+	exit(EXIT_FAILURE);
 }
 
 
-void init_xml_str_attr(tag_t attr, char const * str, char * buffer)
+void init_xml_str_attr(tag_t attr, char const *str, char *buffer, size_t max)
 {
-	char tmp_buf[MAX_BUF_LEN];
-	char quote_buf[MAX_BUF_LEN];
-	char const * attr_name = xml_tag_name(attr);
-	char const * quote_str = xml_quote(str, quote_buf);
-	const unsigned int max_len = strlen(attr_name) + strlen(quote_str) + 10;
+	char *buf;
+	int size, ret;
 
-	if (max_len >= sizeof(tmp_buf))
-		fprintf(stderr, "Warning: init_xml_str_attr: buffer overflow %d\n", max_len);
+	buffer[max - 1] = '\0';
+	size = strlen(buffer);
+	buf = &buffer[size];
+	size = max - 1 - size;
 
-	if (snprintf(tmp_buf, sizeof(tmp_buf), " %s=""%s""", attr_name, quote_str) < 0) {
-		fprintf(stderr,"init_xml_str_attr: snprintf failed\n");
-		exit(EXIT_FAILURE);
-	}
-	strncat(buffer, tmp_buf, sizeof(tmp_buf));
+	ret = snprintf(buf, size, " %s=", xml_tag_name(attr));
+	if (ret < 0 || ret >= size)
+		goto Error;
+
+	buf += ret;
+	size -= ret;
+
+	if (!size)
+		goto Error;
+
+	xml_quote(str, buf, size);
+	return;
+Error:
+	fprintf(stderr,"init_xml_str_attr: snprintf failed\n");
+	exit(EXIT_FAILURE);
 }
diff --git a/libop/op_xml_out.h b/libop/op_xml_out.h
index 52e8d8f..b0188b2 100644
--- a/libop/op_xml_out.h
+++ b/libop/op_xml_out.h
@@ -51,6 +51,7 @@ typedef enum {
 	HELP_EVENT_DESC,
 	HELP_COUNTER_MASK,
 	HELP_MIN_COUNT,
+	HELP_EXT,
 	HELP_UNIT_MASKS,
 	HELP_DEFAULT_MASK,
 	HELP_UNIT_MASK,
@@ -59,11 +60,11 @@ typedef enum {
 	} tag_t;
 
 char const * xml_tag_name(tag_t tag);
-void open_xml_element(tag_t tag, int with_attrs, char * result);
-void close_xml_element(tag_t tag, int has_nested, char * result);
-void init_xml_int_attr(tag_t attr, int value, char * result);
-void init_xml_dbl_attr(tag_t attr, double value, char * result);
-void init_xml_str_attr(tag_t attr, char const * str, char * result);
+void open_xml_element(tag_t tag, int with_attrs, char *buffer, size_t size);
+void close_xml_element(tag_t tag, int has_nested, char *buffer, size_t size);
+void init_xml_int_attr(tag_t attr, int value, char *buffer, size_t size);
+void init_xml_dbl_attr(tag_t attr, double value, char *buffer, size_t size);
+void init_xml_str_attr(tag_t attr, char const *str, char *buffer, size_t size);
 
 #ifdef __cplusplus
 }
diff --git a/libutil/fscanf.c b/libutil/fscanf.c
deleted file mode 100644
index d567324..0000000
--- a/libutil/fscanf.c
+++ /dev/null
@@ -1,22 +0,0 @@
-#include <stdio.h> 
-
-// ugly hack because we don't have fscanf
-
-int fscanf(FILE* stream, const char* format, int* value)
-{
-    int c;
-    int r = 0;
-    do {
-        c = fgetc(stream);
-        if (c>='0' && c<='9') {
-            r = r*10 + (c-'0');
-            continue;
-        }
-        break;
-    } while (1);
-
-    *value = r;
-
-    // gahhhh
-    return 1;
-}
diff --git a/libutil/op_cpufreq.c b/libutil/op_cpufreq.c
index 78a6333..c95cb58 100644
--- a/libutil/op_cpufreq.c
+++ b/libutil/op_cpufreq.c
@@ -51,6 +51,11 @@ double op_cpu_frequency(void)
 			fval = uval / 1E6;
 			break;
 		}
+		/* mips including loongson2 */
+		if (sscanf(line, "BogoMIPS		: %lu", &uval) == 1) {
+			fval = uval * 3 / 2;
+			break;
+		}
 		/* s390 doesn't provide cpu freq, checked up to 2.6-test4 */
 
 		free(line);