[PATCH i-g-t] tests/perf_pmu: Restore runtime PM at subtest exit

[PATCH i-g-t] tests/perf_pmu: Restore runtime PM at subtest exit

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Restore runtime PM state (via a newly added library function) when the
test which sets it up exit. This was we avoid running all subsequent sub-
tests in the aggressive runtime PM mode.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 lib/igt_pm.c     | 13 ++++++++++++-
 lib/igt_pm.h     |  1 +
 tests/perf_pmu.c |  3 +++
 3 files changed, 16 insertions(+), 1 deletion(-)

diff --git a/lib/igt_pm.c b/lib/igt_pm.c
index 8ac132269d79..2e2eea5d3b2d 100644
--- a/lib/igt_pm.c
+++ b/lib/igt_pm.c
@@ -297,7 +297,13 @@ int pm_status_fd = -1;
 static char __igt_pm_runtime_autosuspend[64];
 static char __igt_pm_runtime_control[64];
 
-static void __igt_pm_runtime_exit_handler(int sig)
+/**
+ * igt_restore_runtime_pm:
+ *
+ * Restores the runtime PM configuration as it was before the call to
+ * igt_setup_runtime_pm.
+ */
+void igt_restore_runtime_pm(void)
 {
 	int fd;
 
@@ -326,6 +332,11 @@ static void __igt_pm_runtime_exit_handler(int sig)
 	close(fd);
 }
 
+static void __igt_pm_runtime_exit_handler(int sig)
+{
+	igt_restore_runtime_pm();
+}
+
 /**
  * igt_setup_runtime_pm:
  *
diff --git a/lib/igt_pm.h b/lib/igt_pm.h
index eced39f8801a..10cc6794e4e7 100644
--- a/lib/igt_pm.h
+++ b/lib/igt_pm.h
@@ -47,6 +47,7 @@ enum igt_runtime_pm_status {
 };
 
 bool igt_setup_runtime_pm(void);
+void igt_restore_runtime_pm(void);
 enum igt_runtime_pm_status igt_get_runtime_pm_status(void);
 bool igt_wait_for_pm_status(enum igt_runtime_pm_status status);
 
diff --git a/tests/perf_pmu.c b/tests/perf_pmu.c
index a1d36ac4fa9d..9a20abb6b95c 100644
--- a/tests/perf_pmu.c
+++ b/tests/perf_pmu.c
@@ -1441,6 +1441,9 @@ test_rc6(int gem_fd, unsigned int flags)
 	close(fw);
 	close(fd);
 
+	if (flags & TEST_RUNTIME_PM)
+		igt_restore_runtime_pm();
+
 	assert_within_epsilon(busy - prev, 0.0, tolerance);
 }
 
-- 
2.17.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [igt-dev] [PATCH i-g-t] tests/perf_pmu: Restore runtime PM at subtest exit

[Chris Wilson]

Quoting Tvrtko Ursulin (2018-07-19 17:37:56)
> From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> 
> Restore runtime PM state (via a newly added library function) when the
> test which sets it up exit. This was we avoid running all subsequent sub-
> tests in the aggressive runtime PM mode.
> 
> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> ---
>  lib/igt_pm.c     | 13 ++++++++++++-
>  lib/igt_pm.h     |  1 +
>  tests/perf_pmu.c |  3 +++
>  3 files changed, 16 insertions(+), 1 deletion(-)
> 
> diff --git a/lib/igt_pm.c b/lib/igt_pm.c
> index 8ac132269d79..2e2eea5d3b2d 100644
> --- a/lib/igt_pm.c
> +++ b/lib/igt_pm.c
> @@ -297,7 +297,13 @@ int pm_status_fd = -1;
>  static char __igt_pm_runtime_autosuspend[64];
>  static char __igt_pm_runtime_control[64];
>  
> -static void __igt_pm_runtime_exit_handler(int sig)
> +/**
> + * igt_restore_runtime_pm:
> + *
> + * Restores the runtime PM configuration as it was before the call to
> + * igt_setup_runtime_pm.
> + */
> +void igt_restore_runtime_pm(void)
>  {
>         int fd;

Should we eat __igt_pm_runtime_autosuspend so we only apply it once for
each igt_setup_runtime_pm?
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH i-g-t v2] tests/perf_pmu: Restore runtime PM at subtest exit

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Restore runtime PM state (via a newly added library function) when the
test which sets it up exit. This was we avoid running all subsequent sub-
tests in the aggressive runtime PM mode.

v2:
 * Skip double restore. (Chris Wilson)
 * Close previously leaked fd.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 lib/igt_pm.c     | 20 ++++++++++++++++++--
 lib/igt_pm.h     |  1 +
 tests/perf_pmu.c |  3 +++
 3 files changed, 22 insertions(+), 2 deletions(-)

diff --git a/lib/igt_pm.c b/lib/igt_pm.c
index 8ac132269d79..512ad9b31951 100644
--- a/lib/igt_pm.c
+++ b/lib/igt_pm.c
@@ -291,16 +291,24 @@ void igt_pm_restore_sata_link_power_management(int8_t *pm_data)
 	free(file_name);
 }
 #define POWER_DIR "/sys/devices/pci0000:00/0000:00:02.0/power"
-/* We just leak this on exit ... */
 int pm_status_fd = -1;
 
 static char __igt_pm_runtime_autosuspend[64];
 static char __igt_pm_runtime_control[64];
 
-static void __igt_pm_runtime_exit_handler(int sig)
+/**
+ * igt_restore_runtime_pm:
+ *
+ * Restores the runtime PM configuration as it was before the call to
+ * igt_setup_runtime_pm.
+ */
+void igt_restore_runtime_pm(void)
 {
 	int fd;
 
+	if (pm_status_fd < 0)
+		return;
+
 	igt_debug("Restoring runtime management to '%s' and '%s'\n",
 		  __igt_pm_runtime_autosuspend,
 		  __igt_pm_runtime_control);
@@ -324,6 +332,14 @@ static void __igt_pm_runtime_exit_handler(int sig)
 		igt_warn("Failed to restore runtime pm control to '%s'\n",
 			 __igt_pm_runtime_control);
 	close(fd);
+
+	close(pm_status_fd);
+	pm_status_fd = -1;
+}
+
+static void __igt_pm_runtime_exit_handler(int sig)
+{
+	igt_restore_runtime_pm();
 }
 
 /**
diff --git a/lib/igt_pm.h b/lib/igt_pm.h
index eced39f8801a..10cc6794e4e7 100644
--- a/lib/igt_pm.h
+++ b/lib/igt_pm.h
@@ -47,6 +47,7 @@ enum igt_runtime_pm_status {
 };
 
 bool igt_setup_runtime_pm(void);
+void igt_restore_runtime_pm(void);
 enum igt_runtime_pm_status igt_get_runtime_pm_status(void);
 bool igt_wait_for_pm_status(enum igt_runtime_pm_status status);
 
diff --git a/tests/perf_pmu.c b/tests/perf_pmu.c
index a1d36ac4fa9d..9a20abb6b95c 100644
--- a/tests/perf_pmu.c
+++ b/tests/perf_pmu.c
@@ -1441,6 +1441,9 @@ test_rc6(int gem_fd, unsigned int flags)
 	close(fw);
 	close(fd);
 
+	if (flags & TEST_RUNTIME_PM)
+		igt_restore_runtime_pm();
+
 	assert_within_epsilon(busy - prev, 0.0, tolerance);
 }
 
-- 
2.17.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [igt-dev] [PATCH i-g-t v2] tests/perf_pmu: Restore runtime PM at subtest exit

[Chris Wilson]

Quoting Tvrtko Ursulin (2018-07-20 10:42:55)
> From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> 
> Restore runtime PM state (via a newly added library function) when the
> test which sets it up exit. This was we avoid running all subsequent sub-
> tests in the aggressive runtime PM mode.
> 
> v2:
>  * Skip double restore. (Chris Wilson)
>  * Close previously leaked fd.
> 
> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> ---
>  lib/igt_pm.c     | 20 ++++++++++++++++++--
>  lib/igt_pm.h     |  1 +
>  tests/perf_pmu.c |  3 +++
>  3 files changed, 22 insertions(+), 2 deletions(-)
> 
> diff --git a/lib/igt_pm.c b/lib/igt_pm.c
> index 8ac132269d79..512ad9b31951 100644
> --- a/lib/igt_pm.c
> +++ b/lib/igt_pm.c
> @@ -291,16 +291,24 @@ void igt_pm_restore_sata_link_power_management(int8_t *pm_data)
>         free(file_name);
>  }
>  #define POWER_DIR "/sys/devices/pci0000:00/0000:00:02.0/power"
> -/* We just leak this on exit ... */
>  int pm_status_fd = -1;
>  
>  static char __igt_pm_runtime_autosuspend[64];
>  static char __igt_pm_runtime_control[64];
>  
> -static void __igt_pm_runtime_exit_handler(int sig)
> +/**
> + * igt_restore_runtime_pm:
> + *
> + * Restores the runtime PM configuration as it was before the call to
> + * igt_setup_runtime_pm.
> + */
> +void igt_restore_runtime_pm(void)
>  {
>         int fd;
>  
> +       if (pm_status_fd < 0)
> +               return;
> +
>         igt_debug("Restoring runtime management to '%s' and '%s'\n",
>                   __igt_pm_runtime_autosuspend,
>                   __igt_pm_runtime_control);

Hmm. Did we make our igt_debug() signal safe? The open/read/write/close
are ok.  igt_vlog() is definetly not signal-safe. Not the fault of this
patch, but I strongly advise that the debug/warn are removed from this
function and handled in the caller for !exit_handler.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

[PATCH i-g-t v3] tests/perf_pmu: Restore runtime PM at subtest exit

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

Restore runtime PM state (via a newly added library function) when the
test which sets it up exit. This was we avoid running all subsequent sub-
tests in the aggressive runtime PM mode.

v2:
 * Skip double restore. (Chris Wilson)
 * Close previously leaked fd.

v3:
 * Refactor atexit handlers to be signal safe. (Chris Wilson)
 * Restore audio runtime PM status.
 * Add a different PCI path to audio device.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 lib/igt_pm.c     | 152 ++++++++++++++++++++++++++++++++++++-----------
 lib/igt_pm.h     |   1 +
 tests/perf_pmu.c |   3 +
 3 files changed, 122 insertions(+), 34 deletions(-)

diff --git a/lib/igt_pm.c b/lib/igt_pm.c
index 8ac132269d79..1fe8f5f46399 100644
--- a/lib/igt_pm.c
+++ b/lib/igt_pm.c
@@ -63,36 +63,73 @@ enum {
 /* Remember to fix this if adding longer strings */
 #define MAX_POLICY_STRLEN	strlen(MAX_PERFORMANCE_STR)
 
+static const char *const __igt_pm_audio_runtime_control_paths[] = {
+	"/sys/bus/pci/devices/0000:00:03.0/power/control",
+	"/sys/bus/pci/devices/0000:00:1f.3/power/control",
+};
+
 static char __igt_pm_audio_runtime_power_save[64];
+static const char * __igt_pm_audio_runtime_control_path;
 static char __igt_pm_audio_runtime_control[64];
 
-static void __igt_pm_audio_runtime_exit_handler(int sig)
+static int __igt_pm_audio_restore_runtime_pm(void)
 {
 	int fd;
 
-	igt_debug("Restoring audio power management to '%s' and '%s'\n",
-		  __igt_pm_audio_runtime_power_save,
-		  __igt_pm_audio_runtime_control);
+	if (!__igt_pm_audio_runtime_power_save[0])
+		return 0;
 
 	fd = open("/sys/module/snd_hda_intel/parameters/power_save", O_WRONLY);
 	if (fd < 0)
-		return;
+		return errno;
+
 	if (write(fd, __igt_pm_audio_runtime_power_save,
 		  strlen(__igt_pm_audio_runtime_power_save)) !=
-	    strlen(__igt_pm_audio_runtime_power_save))
-		igt_warn("Failed to restore audio power_save to '%s'\n",
-			 __igt_pm_audio_runtime_power_save);
+	    strlen(__igt_pm_audio_runtime_power_save)) {
+		close(fd);
+		return errno;
+	}
+
 	close(fd);
 
-	fd = open("/sys/bus/pci/devices/0000:00:03.0/power/control", O_WRONLY);
+	fd = open(__igt_pm_audio_runtime_control_path, O_WRONLY);
 	if (fd < 0)
-		return;
+		return errno;
+
 	if (write(fd, __igt_pm_audio_runtime_control,
 		  strlen(__igt_pm_audio_runtime_control)) !=
-	    strlen(__igt_pm_audio_runtime_control))
-		igt_warn("Failed to restore audio control to '%s'\n",
-			 __igt_pm_audio_runtime_control);
+	    strlen(__igt_pm_audio_runtime_control)) {
+		close(fd);
+		return errno;
+	}
+
 	close(fd);
+
+	__igt_pm_audio_runtime_power_save[0] = 0;
+
+	return 0;
+}
+
+static void igt_pm_audio_restore_runtime_pm(void)
+{
+	int ret;
+
+	if (!__igt_pm_audio_runtime_power_save[0])
+		return;
+
+	igt_debug("Restoring audio power management to '%s' and '%s'\n",
+		  __igt_pm_audio_runtime_power_save,
+		  __igt_pm_audio_runtime_control);
+
+	ret = __igt_pm_audio_restore_runtime_pm();
+	if (ret)
+		igt_warn("Failed to restore runtime audio PM! (errno=%d)\n",
+			 ret);
+}
+
+static void __igt_pm_audio_runtime_exit_handler(int sig)
+{
+	__igt_pm_audio_restore_runtime_pm();
 }
 
 static void strchomp(char *str)
@@ -116,7 +153,7 @@ static void strchomp(char *str)
  */
 void igt_pm_enable_audio_runtime_pm(void)
 {
-	int fd;
+	int fd, i;
 
 	/* Check if already enabled. */
 	if (__igt_pm_audio_runtime_power_save[0])
@@ -131,14 +168,23 @@ void igt_pm_enable_audio_runtime_pm(void)
 		igt_assert_eq(write(fd, "1\n", 2), 2);
 		close(fd);
 	}
-	fd = open("/sys/bus/pci/devices/0000:00:03.0/power/control", O_RDWR);
-	if (fd >= 0) {
-		igt_assert(read(fd, __igt_pm_audio_runtime_control,
-				sizeof(__igt_pm_audio_runtime_control)) > 0);
-		strchomp(__igt_pm_audio_runtime_control);
-		igt_assert_eq(write(fd, "auto\n", 5), 5);
-		close(fd);
+
+	for (i = 0; i < ARRAY_SIZE(__igt_pm_audio_runtime_control_paths); i++) {
+		const char *path = __igt_pm_audio_runtime_control_paths[i];
+
+		fd = open(path, O_RDWR);
+		if (fd >= 0) {
+			__igt_pm_audio_runtime_control_path = path;
+			igt_assert(read(fd, __igt_pm_audio_runtime_control,
+					sizeof(__igt_pm_audio_runtime_control)) > 0);
+			strchomp(__igt_pm_audio_runtime_control);
+			igt_assert_eq(write(fd, "auto\n", 5), 5);
+			close(fd);
+			break;
+		}
 	}
+	if (!__igt_pm_audio_runtime_control_path)
+		igt_warn("Failed to find audio PCI device!\n");
 
 	igt_debug("Saved audio power management as '%s' and '%s'\n",
 		  __igt_pm_audio_runtime_power_save,
@@ -291,39 +337,77 @@ void igt_pm_restore_sata_link_power_management(int8_t *pm_data)
 	free(file_name);
 }
 #define POWER_DIR "/sys/devices/pci0000:00/0000:00:02.0/power"
-/* We just leak this on exit ... */
 int pm_status_fd = -1;
 
 static char __igt_pm_runtime_autosuspend[64];
 static char __igt_pm_runtime_control[64];
 
-static void __igt_pm_runtime_exit_handler(int sig)
+static int __igt_restore_runtime_pm(void)
 {
 	int fd;
 
-	igt_debug("Restoring runtime management to '%s' and '%s'\n",
-		  __igt_pm_runtime_autosuspend,
-		  __igt_pm_runtime_control);
+	if (pm_status_fd < 0)
+		return 0;
 
 	fd = open(POWER_DIR "/autosuspend_delay_ms", O_WRONLY);
 	if (fd < 0)
-		return;
+		return errno;
+
 	if (write(fd, __igt_pm_runtime_autosuspend,
 		  strlen(__igt_pm_runtime_autosuspend)) !=
-	    strlen(__igt_pm_runtime_autosuspend))
-		igt_warn("Failed to restore runtime pm autosuspend delay to '%s'\n",
-			 __igt_pm_runtime_autosuspend);
+	    strlen(__igt_pm_runtime_autosuspend)) {
+		close(fd);
+		return errno;
+	}
+
 	close(fd);
 
 	fd = open(POWER_DIR "/control", O_WRONLY);
 	if (fd < 0)
-		return;
+		return errno;
+
 	if (write(fd, __igt_pm_runtime_control,
 		  strlen(__igt_pm_runtime_control)) !=
-	    strlen(__igt_pm_runtime_control))
-		igt_warn("Failed to restore runtime pm control to '%s'\n",
-			 __igt_pm_runtime_control);
+	    strlen(__igt_pm_runtime_control)) {
+		close(fd);
+		return errno;
+	}
+
 	close(fd);
+
+	close(pm_status_fd);
+	pm_status_fd = -1;
+
+	return 0;
+}
+
+/**
+ * igt_restore_runtime_pm:
+ *
+ * Restores the runtime PM configuration as it was before the call to
+ * igt_setup_runtime_pm.
+ */
+void igt_restore_runtime_pm(void)
+{
+	int ret;
+
+	if (pm_status_fd < 0)
+		return;
+
+	igt_debug("Restoring runtime PM management to '%s' and '%s'\n",
+		  __igt_pm_runtime_autosuspend,
+		  __igt_pm_runtime_control);
+
+	ret = __igt_restore_runtime_pm();
+	if (ret)
+		igt_warn("Failed to restore runtime PM! (errno=%d)\n", ret);
+
+	igt_pm_audio_restore_runtime_pm();
+}
+
+static void __igt_pm_runtime_exit_handler(int sig)
+{
+	__igt_restore_runtime_pm();
 }
 
 /**
diff --git a/lib/igt_pm.h b/lib/igt_pm.h
index eced39f8801a..10cc6794e4e7 100644
--- a/lib/igt_pm.h
+++ b/lib/igt_pm.h
@@ -47,6 +47,7 @@ enum igt_runtime_pm_status {
 };
 
 bool igt_setup_runtime_pm(void);
+void igt_restore_runtime_pm(void);
 enum igt_runtime_pm_status igt_get_runtime_pm_status(void);
 bool igt_wait_for_pm_status(enum igt_runtime_pm_status status);
 
diff --git a/tests/perf_pmu.c b/tests/perf_pmu.c
index a1d36ac4fa9d..9a20abb6b95c 100644
--- a/tests/perf_pmu.c
+++ b/tests/perf_pmu.c
@@ -1441,6 +1441,9 @@ test_rc6(int gem_fd, unsigned int flags)
 	close(fw);
 	close(fd);
 
+	if (flags & TEST_RUNTIME_PM)
+		igt_restore_runtime_pm();
+
 	assert_within_epsilon(busy - prev, 0.0, tolerance);
 }
 
-- 
2.17.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [igt-dev] [PATCH i-g-t v3] tests/perf_pmu: Restore runtime PM at subtest exit

[Chris Wilson]

Quoting Tvrtko Ursulin (2018-07-20 13:12:22)
> From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
> 
> Restore runtime PM state (via a newly added library function) when the
> test which sets it up exit. This was we avoid running all subsequent sub-
> tests in the aggressive runtime PM mode.
> 
> v2:
>  * Skip double restore. (Chris Wilson)
>  * Close previously leaked fd.
> 
> v3:
>  * Refactor atexit handlers to be signal safe. (Chris Wilson)
>  * Restore audio runtime PM status.
>  * Add a different PCI path to audio device.
> 
> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

(Just ship it already! But I do wonder if there's a better way to find
the control device (search by PCI vendor, class?) and if we shouldn't
make the control/value be uniform in their handling, then just iterate
over the area of (path, value) for restoring.)
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx