X-Git-Url: http://git.liburcu.org/?a=blobdiff_plain;f=liblttng-ust%2Flttng-context-perf-counters.c;h=7e87be94cfa5a5f7c19d63fe449b8c3ae70d5a76;hb=77d7fa989f79f3c2cb14d26576cbbb54b0637594;hp=c98e2ebdceeb2786862fcb95d8f331ae115aab0d;hpb=d286ad50d958f988399fadc0a53bf4770b7a10ae;p=lttng-ust.git

diff --git a/liblttng-ust/lttng-context-perf-counters.c b/liblttng-ust/lttng-context-perf-counters.c
index c98e2ebd..7e87be94 100644
--- a/liblttng-ust/lttng-context-perf-counters.c
+++ b/liblttng-ust/lttng-context-perf-counters.c
@@ -25,9 +25,9 @@
 #include <string.h>
 #include <stdlib.h>
 #include <stdio.h>
+#include <stdbool.h>
 #include <sys/mman.h>
 #include <sys/syscall.h>
-#include <linux/perf_event.h>
 #include <lttng/ust-events.h>
 #include <lttng/ust-tracer.h>
 #include <lttng/ringbuffer-config.h>
@@ -38,6 +38,7 @@
 #include <urcu/ref.h>
 #include <usterr-signal-safe.h>
 #include <signal.h>
+#include "perf_event.h"
 #include "lttng-tracer-core.h"
 
 /*
@@ -79,6 +80,22 @@ size_t perf_counter_get_size(struct lttng_ctx_field *field, size_t offset)
 	return size;
 }
 
+static
+uint64_t read_perf_counter_syscall(
+		struct lttng_perf_counter_thread_field *thread_field)
+{
+	uint64_t count;
+
+	if (caa_unlikely(thread_field->fd < 0))
+		return 0;
+
+	if (caa_unlikely(read(thread_field->fd, &count, sizeof(count))
+				< sizeof(count)))
+		return 0;
+
+	return count;
+}
+
 #if defined(__x86_64__) || defined(__i386__)
 
 static
@@ -91,13 +108,17 @@ uint64_t rdpmc(unsigned int counter)
 	return low | ((uint64_t) high) << 32;
 }
 
-static bool arch_perf_use_read(void)
+static
+bool has_rdpmc(struct perf_event_mmap_page *pc)
 {
-	return false;
+	if (caa_unlikely(!pc->cap_bit0_is_deprecated))
+		return false;
+	/* Since Linux kernel 3.12. */
+	return pc->cap_user_rdpmc;
 }
 
 static
-uint64_t read_perf_counter(
+uint64_t arch_read_perf_counter(
 		struct lttng_perf_counter_thread_field *thread_field)
 {
 	uint32_t seq, idx;
@@ -112,45 +133,51 @@ uint64_t read_perf_counter(
 		cmm_barrier();
 
 		idx = pc->index;
-		if (idx)
-			count = pc->offset + rdpmc(idx - 1);
-		else
-			count = 0;
-
+		if (caa_likely(has_rdpmc(pc) && idx)) {
+			int64_t pmcval;
+
+			pmcval = rdpmc(idx - 1);
+			/* Sign-extend the pmc register result. */
+			pmcval <<= 64 - pc->pmc_width;
+			pmcval >>= 64 - pc->pmc_width;
+			count = pc->offset + pmcval;
+		} else {
+			/* Fall-back on system call if rdpmc cannot be used. */
+			return read_perf_counter_syscall(thread_field);
+		}
 		cmm_barrier();
 	} while (CMM_LOAD_SHARED(pc->lock) != seq);
 
 	return count;
 }
 
-#elif defined (__ARM_ARCH_7A__)
-
-static bool arch_perf_use_read(void)
-{
-	return true;
-}
-
 static
-uint64_t read_perf_counter(
-		struct lttng_perf_counter_thread_field *thread_field)
+int arch_perf_keep_fd(struct lttng_perf_counter_thread_field *thread_field)
 {
-	uint64_t count;
+	struct perf_event_mmap_page *pc = thread_field->pc;
 
-	if (caa_unlikely(thread_field->fd < 0))
+	if (!pc)
 		return 0;
+	return !has_rdpmc(pc);
+}
 
-	if (caa_unlikely(read(thread_field->fd, &count, sizeof(count))
-				< sizeof(count)))
-		return 0;
+#else
 
-	return count;
+/* Generic (slow) implementation using a read system call. */
+static
+uint64_t arch_read_perf_counter(
+		struct lttng_perf_counter_thread_field *thread_field)
+{
+	return read_perf_counter_syscall(thread_field);
 }
 
-#else /* defined(__x86_64__) || defined(__i386__) || defined(__ARM_ARCH_7A__) */
-
-#error "Perf event counters are only supported on x86 and ARMv7 so far."
+static
+int arch_perf_keep_fd(struct lttng_perf_counter_thread_field *thread_field)
+{
+	return 1;
+}
 
-#endif /* #else defined(__x86_64__) || defined(__i386__) || defined(__ARM_ARCH_7A__) */
+#endif
 
 static
 int sys_perf_event_open(struct perf_event_attr *attr,
@@ -187,9 +214,7 @@ void close_perf_fd(int fd)
 	}
 }
 
-static
-struct perf_event_mmap_page *setup_perf(
-		struct lttng_perf_counter_thread_field *thread_field)
+static void setup_perf(struct lttng_perf_counter_thread_field *thread_field)
 {
 	void *perf_addr;
 
@@ -197,13 +222,12 @@ struct perf_event_mmap_page *setup_perf(
 			PROT_READ, MAP_SHARED, thread_field->fd, 0);
 	if (perf_addr == MAP_FAILED)
 		perf_addr = NULL;
+	thread_field->pc = perf_addr;
 
-	if (!arch_perf_use_read()) {
+	if (!arch_perf_keep_fd(thread_field)) {
 		close_perf_fd(thread_field->fd);
 		thread_field->fd = -1;
 	}
-
-	return perf_addr;
 }
 
 static
@@ -278,7 +302,7 @@ struct lttng_perf_counter_thread_field *
 	thread_field->field = perf_field;
 	thread_field->fd = open_perf_fd(&perf_field->attr);
 	if (thread_field->fd >= 0)
-		thread_field->pc = setup_perf(thread_field);
+		setup_perf(thread_field);
 	/*
 	 * Note: thread_field->pc can be NULL if setup_perf() fails.
 	 * Also, thread_field->fd can be -1 if open_perf_fd() fails.
@@ -323,7 +347,7 @@ uint64_t wrapper_perf_counter_read(struct lttng_ctx_field *field)
 
 	perf_field = field->u.perf_counter;
 	perf_thread_field = get_thread_field(perf_field);
-	return read_perf_counter(perf_thread_field);
+	return arch_read_perf_counter(perf_thread_field);
 }
 
 static