[PATCH 0/4] Introduce userspace-driven ALSA timers

[PATCH 0/4] Introduce userspace-driven ALSA timers

[Ivan Orlov]

There are multiple possible timer sources which could be useful for
the sound stream synchronization: hrtimers, hardware clocks (e.g. PTP),
timer wheels (jiffies). Currently, using one of them to synchronize
the audio stream of snd-aloop module would require writing a
kernel-space driver which exports an ALSA timer through the
snd_timer interface.

However, it is not really convenient for application developers, who may
want to define their custom timer sources for audio synchronization.

For instance, we could have a network application which receives frames
and sends them to snd-aloop pcm device, and another application
listening on the other end of snd-aloop. It makes sense to transfer a
new period of data only when certain amount of frames is received
through the network, but definitely not when a certain amount of jiffies
on a local system elapses. Since all of the devices are purely virtual
it won't introduce any glitches and will help the application developers
to avoid using sample-rate conversion.

This patch series introduces userspace-driven ALSA timers: virtual
timers which are created and controlled from userspace. The timer can
be created from the userspace using the new ioctl SNDRV_TIMER_IOCTL_CREATE.
After creating a timer, it becomes available for use system-wide, so it
can be passed to snd-aloop as a timer source (timer_source parameter
would be "-1.SNDRV_TIMER_GLOBAL_UDRIVEN.{timer_id}"). When the userspace
app decides to trigger a timer, it calls another ioctl
SNDRV_TIMER_IOCTL_TRIGGER on the file descriptor of a timer. It
initiates a transfer of a new period of data.

Userspace-driven timers are associated with file descriptors. If the
application wishes to destroy the timer, it can simply release the file
descriptor of a virtual timer.

I believe introducing new ioctl calls is quite inconvenient (as we have
a limited amount of them), but other possible ways of app <-> kernel
communication (like virtual FS) seem completely inappropriate for this
task (but I'd love to discuss alternative solutions).

This patch series also updates the snd-aloop module so the global timers
can be used as a timer_source for it (it allows using userspace-driven
timers as timer source).

Ivan Orlov (4):
  ALSA: aloop: Allow using global timers
  Docs/sound: Add documentation for userspace-driven ALSA timers
  ALSA: timer: Introduce virtual userspace-driven timers
  selftests: ALSA: Cover userspace-driven timers with test

 Documentation/sound/index.rst               |   1 +
 Documentation/sound/utimers.rst             | 120 +++++++++++
 include/uapi/sound/asound.h                 |  17 ++
 sound/core/Kconfig                          |  11 +
 sound/core/timer.c                          | 226 ++++++++++++++++++++
 sound/drivers/aloop.c                       |   2 +
 tools/testing/selftests/alsa/Makefile       |   2 +-
 tools/testing/selftests/alsa/global-timer.c |  87 ++++++++
 tools/testing/selftests/alsa/utimer-test.c  | 133 ++++++++++++
 9 files changed, 598 insertions(+), 1 deletion(-)
 create mode 100644 Documentation/sound/utimers.rst
 create mode 100644 tools/testing/selftests/alsa/global-timer.c
 create mode 100644 tools/testing/selftests/alsa/utimer-test.c

-- 
2.34.1

[PATCH 1/4] ALSA: aloop: Allow using global timers

[Ivan Orlov]

Allow using global timers as a timer source when card id is equal to -1
in the timer_source parameter.

Signed-off-by: Ivan Orlov <ivan.orlov0322@gmail.com>
---
 sound/drivers/aloop.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/sound/drivers/aloop.c b/sound/drivers/aloop.c
index d6dd4b8c750a..a1e5e7fe9c72 100644
--- a/sound/drivers/aloop.c
+++ b/sound/drivers/aloop.c
@@ -1130,6 +1130,8 @@ static int loopback_parse_timer_id(const char *str,
 			}
 		}
 	}
+	if (card_idx == -1)
+		tid->dev_class = SNDRV_TIMER_CLASS_GLOBAL;
 	if (!err && tid) {
 		tid->card = card_idx;
 		tid->device = dev;
-- 
2.34.1

[PATCH 2/4] Docs/sound: Add documentation for userspace-driven ALSA timers

[Ivan Orlov]

Add the documentation which describes the new userspace-driven timers
API introduced in this patch series. The documentation contains:

- Description of userspace-driven ALSA timers, what they are for
- Description of the timers API
- Example of how the timers can be created and triggered
- How the timers can be used as a timer sources for snd-aloop module

Suggested-by: Axel Holzinger <aholzinger@gmx.de>
Signed-off-by: Ivan Orlov <ivan.orlov0322@gmail.com>
---
 Documentation/sound/index.rst   |   1 +
 Documentation/sound/utimers.rst | 120 ++++++++++++++++++++++++++++++++
 2 files changed, 121 insertions(+)
 create mode 100644 Documentation/sound/utimers.rst

diff --git a/Documentation/sound/index.rst b/Documentation/sound/index.rst
index 7e67e12730d3..c437f2a4bc85 100644
--- a/Documentation/sound/index.rst
+++ b/Documentation/sound/index.rst
@@ -13,6 +13,7 @@ Sound Subsystem Documentation
    alsa-configuration
    hd-audio/index
    cards/index
+   utimers
 
 .. only::  subproject and html
 
diff --git a/Documentation/sound/utimers.rst b/Documentation/sound/utimers.rst
new file mode 100644
index 000000000000..0bd875ea9986
--- /dev/null
+++ b/Documentation/sound/utimers.rst
@@ -0,0 +1,120 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+=======================
+Userspace-driven timers
+=======================
+
+:Author: Ivan Orlov <ivan.orlov0322@gmail.com>
+
+Preface
+=======
+
+This document describes the userspace-driven timers: virtual ALSA timers
+which could be created and controlled by userspace applications using
+IOCTL calls. Such timers could be useful when synchronizing audio
+stream with timer sources which we don't have ALSA timers exported for
+(e.g. PTP clocks), and when synchronizing the audio stream going through
+two virtual sound devices using ``snd-aloop`` (for instance, when
+we have a network application sending frames to one snd-aloop device,
+and another sound application listening on the other end of snd-aloop).
+
+Enabling userspace-driven timers
+================================
+
+The userspace-driven timers could be enabled in the kernel using the
+``CONFIG_SND_UTIMER`` configuration option. It depends on the
+``CONFIG_SND_TIMER`` option, so it also should be enabled.
+
+Userspace-driven timers API
+===========================
+
+Userspace application can create a userspace-driven ALSA timer by
+executing the ``SNDRV_TIMER_IOCTL_CREATE`` ioctl call on the
+``/dev/snd/timer`` device file descriptor. The ``snd_utimer_info``
+structure should be passed as an ioctl argument:
+
+::
+
+    struct snd_utimer_info {
+        snd_pcm_uframes_t frame_rate;
+        snd_pcm_uframes_t period_size;
+        unsigned int id;
+    }
+
+``frame_rate`` and ``period_size`` set the desired frame rate and period
+size emulated by the virtual timer respectively. ``id`` field gets
+overwritten by the ioctl, and the identifier you get in this field after
+the call can be used as a timer subdevice number when passing the timer
+to ``snd-aloop`` kernel module or other userspace applications. There
+could be up to 128 userspace-driven timers in the system at one moment
+of time, thus the id value ranges from 0 to 127.
+
+Besides from overwriting the ``snd_utimer_info`` struct, ioctl returns
+a timer file descriptor, which can be used to trigger the timer. This
+guarantees that the timer can only be triggered by the process which
+created it. The timer then can be triggered with
+``SNDRV_TIMER_IOCTL_TRIGGER`` ioctl call on the timer file descriptor.
+
+So, the example code for creating and triggering the timer would be:
+
+::
+
+    static const struct snd_utimer_info utimer_info = {
+        .frame_rate = 8000,
+        .period_size = 4410,
+        .id = -1,
+    };
+
+    int timer_device_fd = open("/dev/snd/timer",  O_RDWR | O_CLOEXEC);
+    int utimer_fd = ioctl(timer_device_fd, SNDRV_TIMER_IOCTL_CREATE, &utimer_info);
+
+    if (utimer_fd < 0) {
+        perror("Failed to create the timer");
+        return -1;
+    }
+
+    ...
+
+    /*
+     * Now we want to trigger the timer. Callbacks of all of the
+     * timer instances binded to this timer will be executed after
+     * this call.
+     */
+    ioctl(utimer_fd, SNDRV_TIMER_IOCTL_TRIGGER, NULL);
+
+    ...
+
+    /* Now, destroy the timer */
+    close(utimer_fd);
+
+
+More detailed example of creating and ticking the timer could be found
+in the utimer ALSA selftest.
+
+Userspace-driven timers and snd-aloop
+-------------------------------------
+
+Userspace-driven timers could be easily used with ``snd-aloop`` module
+when synchronizing two sound applications on both ends of the virtual
+sound loopback. For instance, if one of the applications receives sound
+frames from network and sends them to snd-aloop pcm device, and another
+application listens for frames on the other snd-aloop pcm device, it
+makes sense that the ALSA middle layer should initiate a data
+transaction when the new period of data is received through network, but
+not when the certain amount of jiffies elapses. Userspace-driven ALSA
+timers could be used to achieve this.
+
+To use userspace-driven ALSA timer as a timer source of snd-aloop, pass
+the following string as the snd-aloop ``timer_source`` parameter:
+
+::
+
+  # modprobe snd-aloop timer_source="-1.4.<utimer_id>"
+
+Where ``utimer_id`` is the id of the timer you created with
+``SNDRV_TIMER_IOCTL_CREATE``, and ``4`` is the number of
+userspace-driven timers device (``SNDRV_TIMER_GLOBAL_UDRIVEN``).
+
+After that, each time you trigger the timer with
+``SNDRV_TIMER_IOCTL_TRIGGER`` the new period of data will be transferred
+from one snd-aloop device to another.
-- 
2.34.1

[PATCH 3/4] ALSA: timer: Introduce virtual userspace-driven timers

[Ivan Orlov]

Implement two ioctl calls in order to support virtual userspace-driven
ALSA timers.

The first ioctl is SNDRV_TIMER_IOCTL_CREATE, which gets the
snd_utimer_info struct as a parameter and returns a file descriptor of
a virtual timer. It also updates the `id` field of the snd_utimer_info
struct, which provides a unique identifier for the timer (basically,
the subdevice number which can be used when creating timer instances).

This patch also introduces a tiny id allocator for the userspace-driven
timers, which guarantees that we don't have more than 128 of them in the
system.

Another ioctl is SNDRV_TIMER_IOCTL_TRIGGER, which allows us to trigger
the virtual timer (and calls snd_timer_interrupt for the timer under
the hood), causing all of the timer instances binded to this timer to
execute their callbacks.

The maximum amount of ticks available for the timer is 1 for the sake of
simplification of the userspace API. 'start', 'stop', 'open' and 'close'
callbacks for the userspace-driven timers are empty since we don't
really do any hardware initialization here.

Suggested-by: Axel Holzinger <aholzinger@gmx.de>
Signed-off-by: Ivan Orlov <ivan.orlov0322@gmail.com>
---
 include/uapi/sound/asound.h |  17 +++
 sound/core/Kconfig          |  11 ++
 sound/core/timer.c          | 226 ++++++++++++++++++++++++++++++++++++
 3 files changed, 254 insertions(+)

diff --git a/include/uapi/sound/asound.h b/include/uapi/sound/asound.h
index 8bf7e8a0eb6f..ade952a54edd 100644
--- a/include/uapi/sound/asound.h
+++ b/include/uapi/sound/asound.h
@@ -894,6 +894,7 @@ enum {
 #define SNDRV_TIMER_GLOBAL_RTC		1	/* unused */
 #define SNDRV_TIMER_GLOBAL_HPET		2
 #define SNDRV_TIMER_GLOBAL_HRTIMER	3
+#define SNDRV_TIMER_GLOBAL_UDRIVEN	4
 
 /* info flags */
 #define SNDRV_TIMER_FLG_SLAVE		(1<<0)	/* cannot be controlled */
@@ -974,6 +975,20 @@ struct snd_timer_status {
 };
 #endif
 
+/*
+ * This structure describes the userspace-driven timer. Such timers are purely virtual,
+ * and can only be triggered from software (for instance, by userspace application).
+ */
+struct snd_utimer_info {
+	/*
+	 * To pretend being a normal timer, we need to know the frame rate and
+	 * the period size in frames.
+	 */
+	snd_pcm_uframes_t frame_rate;
+	snd_pcm_uframes_t period_size;
+	unsigned int id;
+};
+
 #define SNDRV_TIMER_IOCTL_PVERSION	_IOR('T', 0x00, int)
 #define SNDRV_TIMER_IOCTL_NEXT_DEVICE	_IOWR('T', 0x01, struct snd_timer_id)
 #define SNDRV_TIMER_IOCTL_TREAD_OLD	_IOW('T', 0x02, int)
@@ -990,6 +1005,8 @@ struct snd_timer_status {
 #define SNDRV_TIMER_IOCTL_CONTINUE	_IO('T', 0xa2)
 #define SNDRV_TIMER_IOCTL_PAUSE		_IO('T', 0xa3)
 #define SNDRV_TIMER_IOCTL_TREAD64	_IOW('T', 0xa4, int)
+#define SNDRV_TIMER_IOCTL_CREATE	_IOWR('T', 0xa5, struct snd_utimer_info)
+#define SNDRV_TIMER_IOCTL_TRIGGER	_IO('T', 0xa6)
 
 #if __BITS_PER_LONG == 64
 #define SNDRV_TIMER_IOCTL_TREAD SNDRV_TIMER_IOCTL_TREAD_OLD
diff --git a/sound/core/Kconfig b/sound/core/Kconfig
index b970a1734647..3cf82641fc67 100644
--- a/sound/core/Kconfig
+++ b/sound/core/Kconfig
@@ -251,6 +251,17 @@ config SND_JACK_INJECTION_DEBUG
 	  Say Y if you are debugging via jack injection interface.
 	  If unsure select "N".
 
+config SND_UTIMER
+	bool "Enable support for userspace-controlled virtual timers"
+	depends on SND_TIMER
+	help
+	  Say Y to enable the support of userspace-controlled timers. These
+	  timers are purely virtual, and they are supposed to be triggered
+	  from userspace. They could be quite useful when synchronizing the
+	  sound timing with userspace applications (for instance, when sending
+	  data through snd-aloop).
+
+
 config SND_VMASTER
 	bool
 
diff --git a/sound/core/timer.c b/sound/core/timer.c
index d104adc75a8b..177bd06f7b60 100644
--- a/sound/core/timer.c
+++ b/sound/core/timer.c
@@ -13,6 +13,8 @@
 #include <linux/module.h>
 #include <linux/string.h>
 #include <linux/sched/signal.h>
+#include <linux/anon_inodes.h>
+#include <linux/units.h>
 #include <sound/core.h>
 #include <sound/timer.h>
 #include <sound/control.h>
@@ -109,6 +111,17 @@ struct snd_timer_status64 {
 	unsigned char reserved[64];	/* reserved */
 };
 
+#ifdef CONFIG_SND_UTIMER
+#define SNDRV_UTIMERS_MAX_COUNT 128
+#define SNDRV_UTIMER_NAME_LEN 20
+/* Internal data structure for keeping the state of the userspace-driven timer */
+struct snd_utimer {
+	char *name;
+	struct snd_timer *timer;
+	unsigned int id;
+};
+#endif
+
 #define SNDRV_TIMER_IOCTL_STATUS64	_IOR('T', 0x14, struct snd_timer_status64)
 
 /* list of timers */
@@ -2009,6 +2022,217 @@ enum {
 	SNDRV_TIMER_IOCTL_PAUSE_OLD = _IO('T', 0x23),
 };
 
+#ifdef CONFIG_SND_UTIMER
+/*
+ * Since userspace-driven timers are passed to userspace, we need to have an identifier
+ * which will allow us to use them (basically, the subdevice number of udriven timer).
+ *
+ * We have a pool of SNDRV_UTIMERS_MAX_COUNT ids from 0 to (SNDRV_UTIMERS_MAX_COUNT - 1).
+ * When we take one of them, the corresponding entry in snd_utimer_ids becomes true.
+ */
+static bool snd_utimer_ids[SNDRV_UTIMERS_MAX_COUNT];
+
+static void snd_utimer_put_id(struct snd_utimer *utimer)
+{
+	int timer_id = utimer->id;
+
+	snd_BUG_ON(timer_id < 0 || timer_id >= SNDRV_UTIMERS_MAX_COUNT);
+	snd_utimer_ids[timer_id] = false;
+}
+
+static int snd_utimer_take_id(void)
+{
+	size_t i;
+
+	for (i = 0; i < SNDRV_UTIMERS_MAX_COUNT; i++) {
+		if (!snd_utimer_ids[i]) {
+			snd_utimer_ids[i] = true;
+			return i;
+		}
+	}
+
+	return -EBUSY;
+}
+
+static void snd_utimer_free(struct snd_utimer *utimer)
+{
+	snd_timer_free(utimer->timer);
+	snd_utimer_put_id(utimer);
+	kfree(utimer);
+}
+
+static int snd_utimer_release(struct inode *inode, struct file *file)
+{
+	struct snd_utimer *utimer = (struct snd_utimer *)file->private_data;
+
+	snd_utimer_free(utimer);
+	return 0;
+}
+
+static int snd_utimer_trigger(struct file *file)
+{
+	struct snd_utimer *utimer = (struct snd_utimer *)file->private_data;
+
+	snd_timer_interrupt(utimer->timer, utimer->timer->sticks);
+	return 0;
+}
+
+static long snd_utimer_ioctl(struct file *file, unsigned int ioctl, unsigned long arg)
+{
+	switch (ioctl) {
+	case SNDRV_TIMER_IOCTL_TRIGGER:
+		return snd_utimer_trigger(file);
+	}
+
+	return -ENOTTY;
+}
+
+static const struct file_operations snd_utimer_fops = {
+	.llseek = noop_llseek,
+	.release = snd_utimer_release,
+	.unlocked_ioctl = snd_utimer_ioctl,
+};
+
+static int snd_utimer_start(struct snd_timer *t)
+{
+	return 0;
+}
+
+static int snd_utimer_stop(struct snd_timer *t)
+{
+	return 0;
+}
+
+static int snd_utimer_open(struct snd_timer *t)
+{
+	return 0;
+}
+
+static int snd_utimer_close(struct snd_timer *t)
+{
+	return 0;
+}
+
+static const struct snd_timer_hardware timer_hw = {
+	.flags = SNDRV_TIMER_HW_AUTO | SNDRV_TIMER_HW_WORK,
+	.open = snd_utimer_open,
+	.close = snd_utimer_close,
+	.start = snd_utimer_start,
+	.stop = snd_utimer_stop,
+};
+
+static int snd_utimer_create(struct snd_utimer_info *utimer_info,
+			     struct snd_utimer **r_utimer)
+{
+	struct snd_utimer *utimer;
+	struct snd_timer *timer;
+	struct snd_timer_id tid;
+	int utimer_id;
+	int err = 0;
+	char *timer_name;
+
+	utimer = kzalloc(sizeof(*utimer), GFP_KERNEL);
+	if (!utimer)
+		return -ENOMEM;
+
+	timer_name = kzalloc(SNDRV_UTIMER_NAME_LEN, GFP_KERNEL);
+	if (!timer_name) {
+		kfree(utimer);
+		return -ENOMEM;
+	}
+
+	/* We hold the ioctl lock here so we won't get a race condition when allocating id */
+	utimer_id = snd_utimer_take_id();
+	if (utimer_id < 0) {
+		err = utimer_id;
+		goto err_take_id;
+	}
+
+	sprintf(timer_name, "snd-utimer%d", utimer_id);
+	utimer->name = timer_name;
+	utimer->id = utimer_id;
+
+	tid.dev_sclass = SNDRV_TIMER_SCLASS_APPLICATION;
+	tid.dev_class = SNDRV_TIMER_CLASS_GLOBAL;
+	tid.card = -1;
+	tid.device = SNDRV_TIMER_GLOBAL_UDRIVEN;
+	tid.subdevice = utimer_id;
+
+	err = snd_timer_new(NULL, timer_name, &tid, &timer);
+	if (err < 0) {
+		pr_err("Can't create userspace-driven timer\n");
+		goto err_timer_new;
+	}
+
+	timer->module = THIS_MODULE;
+	timer->hw = timer_hw;
+	timer->hw.resolution = NANO / utimer_info->frame_rate * utimer_info->period_size;
+	timer->hw.ticks = 1;
+	timer->max_instances = MAX_SLAVE_INSTANCES;
+
+	utimer->timer = timer;
+
+	err = snd_timer_global_register(timer);
+	if (err < 0) {
+		pr_err("Can't register a userspace-driven timer\n");
+		goto err_timer_reg;
+	}
+
+	*r_utimer = utimer;
+	return 0;
+
+err_timer_reg:
+	snd_timer_free(timer);
+err_timer_new:
+	snd_utimer_put_id(utimer);
+err_take_id:
+	kfree(timer_name);
+	kfree(utimer);
+
+	return err;
+}
+
+static int snd_utimer_ioctl_create(struct file *file,
+				   struct snd_utimer_info __user *_utimer_info)
+{
+	struct snd_utimer *utimer;
+	struct snd_utimer_info *utimer_info;
+	int err;
+
+	utimer_info = memdup_user(_utimer_info, sizeof(*utimer_info));
+	if (IS_ERR(utimer_info))
+		return PTR_ERR(no_free_ptr(utimer_info));
+
+	err = snd_utimer_create(utimer_info, &utimer);
+	if (err < 0) {
+		kfree(utimer_info);
+		return err;
+	}
+
+	utimer_info->id = utimer->id;
+
+	err = copy_to_user(_utimer_info, utimer_info, sizeof(*utimer_info));
+	if (err) {
+		snd_utimer_free(utimer);
+		kfree(utimer_info);
+		return -EFAULT;
+	}
+
+	kfree(utimer_info);
+
+	return anon_inode_getfd(utimer->name, &snd_utimer_fops, utimer, O_RDWR | O_CLOEXEC);
+}
+
+#else
+
+static int snd_utimer_ioctl_create(struct file *file,
+				   struct snd_utimer_info __user *_utimer_info)
+{
+	return -EINVAL;
+}
+
+#endif
+
 static long __snd_timer_user_ioctl(struct file *file, unsigned int cmd,
 				 unsigned long arg, bool compat)
 {
@@ -2053,6 +2277,8 @@ static long __snd_timer_user_ioctl(struct file *file, unsigned int cmd,
 	case SNDRV_TIMER_IOCTL_PAUSE:
 	case SNDRV_TIMER_IOCTL_PAUSE_OLD:
 		return snd_timer_user_pause(file);
+	case SNDRV_TIMER_IOCTL_CREATE:
+		return snd_utimer_ioctl_create(file, argp);
 	}
 	return -ENOTTY;
 }
-- 
2.34.1

[PATCH 4/4] selftests: ALSA: Cover userspace-driven timers with test

[Ivan Orlov]

Add a test for the new functionality of userspace-driven timers and the
tool which allows us to count timer ticks in a certain time period. The
test:

1. Creates a userspace-driven timer with ioctl to /dev/snd/timer
2. Starts the `global-timer` application to count the ticks of the timer
from step 1.
3. Asynchronously triggers the timer multiple times with some interval
4. Compares the amount of caught ticks with the amount of trigger calls.

Since we can't include <alsa/asoundlib.h> and <sound/asound.h> in one
file due to overlapping declarations, I have to split the test into two
applications: one of them counts the amount of timer ticks in the
defined time period, and another one is the actual test which creates
the timer, triggers it periodically and starts the first app to count
the amount of ticks in a separate thread.

Besides from testing the functionality itself, the test represents a
sample application showing userspace-driven ALSA timers API.

Signed-off-by: Ivan Orlov <ivan.orlov0322@gmail.com>
---
 tools/testing/selftests/alsa/Makefile       |   2 +-
 tools/testing/selftests/alsa/global-timer.c |  87 +++++++++++++
 tools/testing/selftests/alsa/utimer-test.c  | 133 ++++++++++++++++++++
 3 files changed, 221 insertions(+), 1 deletion(-)
 create mode 100644 tools/testing/selftests/alsa/global-timer.c
 create mode 100644 tools/testing/selftests/alsa/utimer-test.c

diff --git a/tools/testing/selftests/alsa/Makefile b/tools/testing/selftests/alsa/Makefile
index c1ce39874e2b..0d5bd8ea900b 100644
--- a/tools/testing/selftests/alsa/Makefile
+++ b/tools/testing/selftests/alsa/Makefile
@@ -12,7 +12,7 @@ LDLIBS+=-lpthread
 
 OVERRIDE_TARGETS = 1
 
-TEST_GEN_PROGS := mixer-test pcm-test test-pcmtest-driver
+TEST_GEN_PROGS := mixer-test pcm-test utimer-test test-pcmtest-driver global-timer
 
 TEST_GEN_PROGS_EXTENDED := libatest.so
 
diff --git a/tools/testing/selftests/alsa/global-timer.c b/tools/testing/selftests/alsa/global-timer.c
new file mode 100644
index 000000000000..c15ec0ba851a
--- /dev/null
+++ b/tools/testing/selftests/alsa/global-timer.c
@@ -0,0 +1,87 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * This tool is used by the utimer test, and it allows us to
+ * count the ticks of a global timer in a certain time frame
+ * (which is set by `timeout` parameter).
+ *
+ * Author: Ivan Orlov <ivan.orlov0322@gmail.com>
+ */
+#include <stdio.h>
+#include <stdlib.h>
+#include <alsa/asoundlib.h>
+#include <time.h>
+
+static int ticked;
+static void async_callback(snd_async_handler_t *ahandler)
+{
+	ticked++;
+}
+
+static char timer_name[64];
+static void bind_to_timer(int device, int subdevice, int timeout)
+{
+	snd_timer_t *handle;
+	snd_timer_params_t *params;
+	snd_async_handler_t *ahandler;
+
+	time_t end;
+
+	sprintf(timer_name, "hw:CLASS=%d,SCLASS=%d,DEV=%d,SUBDEV=%d",
+		SND_TIMER_CLASS_GLOBAL, SND_TIMER_SCLASS_NONE,
+		device, subdevice);
+
+	snd_timer_params_alloca(&params);
+
+	if (snd_timer_open(&handle, timer_name, SND_TIMER_OPEN_NONBLOCK) < 0) {
+		perror("Can't open the timer");
+		exit(EXIT_FAILURE);
+	}
+
+	snd_timer_params_set_auto_start(params, 1);
+	snd_timer_params_set_ticks(params, 1);
+	if (snd_timer_params(handle, params) < 0) {
+		perror("Can't set timer params");
+		exit(EXIT_FAILURE);
+	}
+
+	if (snd_async_add_timer_handler(&ahandler, handle, async_callback, NULL) < 0) {
+		perror("Can't create a handler");
+		exit(EXIT_FAILURE);
+	}
+	end = time(NULL) + timeout;
+	if (snd_timer_start(handle) < 0) {
+		perror("Failed to start the timer");
+		exit(EXIT_FAILURE);
+	}
+	printf("Timer has started\n");
+	while (time(NULL) <= end) {
+		/*
+		 * Waiting for the timeout to elapse. Can't use sleep here, as it gets
+		 * constantly interrupted by the signal from the timer (SIGIO)
+		 */
+	}
+	snd_timer_stop(handle);
+	snd_timer_close(handle);
+}
+
+int main(int argc, char *argv[])
+{
+	int device, subdevice, timeout;
+
+	if (argc < 4) {
+		perror("Usage: %s <device> <subdevice> <timeout>");
+		return EXIT_FAILURE;
+	}
+
+	setlinebuf(stdout);
+
+	device = atoi(argv[1]);
+	subdevice = atoi(argv[2]);
+	timeout = atoi(argv[3]);
+
+	bind_to_timer(device, subdevice, timeout);
+
+	printf("Total ticks count: %d\n", ticked);
+
+	return EXIT_SUCCESS;
+}
diff --git a/tools/testing/selftests/alsa/utimer-test.c b/tools/testing/selftests/alsa/utimer-test.c
new file mode 100644
index 000000000000..38de90e507fe
--- /dev/null
+++ b/tools/testing/selftests/alsa/utimer-test.c
@@ -0,0 +1,133 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * This test covers the functionality of userspace-driven ALSA timers. Such timers
+ * are purely virtual (so they don't directly depend on the hardware), and they could be
+ * created and triggered by userspace applications.
+ *
+ * Author: Ivan Orlov <ivan.orlov0322@gmail.com>
+ */
+#include "../kselftest_harness.h"
+#include <sound/asound.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <sys/ioctl.h>
+#include <stdlib.h>
+#include <pthread.h>
+#include <string.h>
+
+#define FRAME_RATE 8000
+#define PERIOD_SIZE 4410
+#define TICKS_COUNT 10
+#define TICKS_RECORDING_DELTA 5
+#define TIMER_OUTPUT_BUF_LEN 1024
+#define TIMER_FREQ_SEC 1
+#define RESULT_PREFIX_LEN strlen("Total ticks count: ")
+
+enum timer_app_event {
+	TIMER_APP_STARTED,
+	TIMER_APP_RESULT,
+	TIMER_NO_EVENT,
+};
+
+FIXTURE(timer_f) {
+	int utimer_fd;
+	struct snd_utimer_info *utimer_info;
+};
+
+FIXTURE_SETUP(timer_f) {
+	int timer_dev_fd;
+
+	if (geteuid())
+		SKIP(return, "This test needs root to run!");
+
+	self->utimer_info = calloc(1, sizeof(*self->utimer_info));
+	ASSERT_NE(NULL, self->utimer_info);
+
+	self->utimer_info->frame_rate = FRAME_RATE;
+	self->utimer_info->period_size = PERIOD_SIZE;
+
+	timer_dev_fd = open("/dev/snd/timer", O_RDONLY);
+	ASSERT_GE(timer_dev_fd, 0);
+
+	self->utimer_fd = ioctl(timer_dev_fd, SNDRV_TIMER_IOCTL_CREATE, self->utimer_info);
+	ASSERT_GE(self->utimer_fd, 0);
+
+	close(timer_dev_fd);
+}
+
+FIXTURE_TEARDOWN(timer_f) {
+	close(self->utimer_fd);
+	free(self->utimer_info);
+}
+
+static void *ticking_func(void *data)
+{
+	int i;
+	int *fd = (int *)data;
+
+	for (i = 0; i < TICKS_COUNT; i++) {
+		/* Well, trigger the timer! */
+		ioctl(*fd, SNDRV_TIMER_IOCTL_TRIGGER, NULL);
+		sleep(TIMER_FREQ_SEC);
+	}
+}
+
+static enum timer_app_event parse_timer_output(const char *s)
+{
+	if (strstr(s, "Timer has started"))
+		return TIMER_APP_STARTED;
+	if (strstr(s, "Total ticks count"))
+		return TIMER_APP_RESULT;
+
+	return TIMER_NO_EVENT;
+}
+
+static int parse_timer_result(const char *s)
+{
+	char *end;
+	long d;
+
+	d = strtol(s + RESULT_PREFIX_LEN, &end, 10);
+	if (end == s + RESULT_PREFIX_LEN)
+		return -1;
+
+	return d;
+}
+
+/*
+ * This test triggers the timer and counts ticks at the same time. The amount
+ * of the timer trigger calls should be equal to the amount of ticks received.
+ */
+TEST_F(timer_f, utimer) {
+	char command[64];
+	pthread_t ticking_thread;
+	int total_ticks = 0;
+	FILE *rfp;
+	char *buf = malloc(TIMER_OUTPUT_BUF_LEN);
+
+	ASSERT_NE(buf, NULL);
+
+	/* The timeout should be the ticks interval * count of ticks + some delta */
+	sprintf(command, "./global-timer %d %d %d", SNDRV_TIMER_GLOBAL_UDRIVEN,
+		self->utimer_info->id, TICKS_COUNT * TIMER_FREQ_SEC + TICKS_RECORDING_DELTA);
+
+	rfp = popen(command, "r");
+	while (fgets(buf, TIMER_OUTPUT_BUF_LEN, rfp)) {
+		buf[TIMER_OUTPUT_BUF_LEN - 1] = 0;
+		switch (parse_timer_output(buf)) {
+		case TIMER_APP_STARTED:
+			/* global-timer waits for timer to trigger, so start the ticking thread */
+			pthread_create(&ticking_thread, NULL, ticking_func,
+				       &self->utimer_fd);
+			break;
+		case TIMER_APP_RESULT:
+			total_ticks = parse_timer_result(buf);
+			break;
+		}
+	}
+	pthread_join(ticking_thread, NULL);
+	ASSERT_EQ(total_ticks, TICKS_COUNT);
+	pclose(rfp);
+}
+
+TEST_HARNESS_MAIN
-- 
2.34.1

Re: [PATCH 4/4] selftests: ALSA: Cover userspace-driven timers with test

[kernel test robot]

Hi Ivan,

kernel test robot noticed the following build warnings:

[auto build test WARNING on tiwai-sound/for-next]
[also build test WARNING on tiwai-sound/for-linus linus/master v6.10 next-20240726]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Ivan-Orlov/ALSA-aloop-Allow-using-global-timers/20240726-155158
base:   https://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound.git for-next
patch link:    https://lore.kernel.org/r/20240726074750.626671-5-ivan.orlov0322%40gmail.com
patch subject: [PATCH 4/4] selftests: ALSA: Cover userspace-driven timers with test
:::::: branch date: 2 days ago
:::::: commit date: 2 days ago
compiler: clang version 18.1.5 (https://github.com/llvm/llvm-project 617a15a9eac96088ae5e9134248d8236e34b91b1)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20240728/202407281359.vMYdESrT-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/r/202407281359.vMYdESrT-lkp@intel.com/

All warnings (new ones prefixed by >>):

>> utimer-test.c:73:1: warning: non-void function does not return a value [-Wreturn-type]
      73 | }
         | ^
>> utimer-test.c:117:11: warning: enumeration value 'TIMER_NO_EVENT' not handled in switch [-Wswitch]
     117 |                 switch (parse_timer_output(buf)) {
         |                         ^~~~~~~~~~~~~~~~~~~~~~~
   2 warnings generated.


vim +73 tools/testing/selftests/alsa/utimer-test.c

5cf58530fa628b Ivan Orlov 2024-07-26  62  
5cf58530fa628b Ivan Orlov 2024-07-26  63  static void *ticking_func(void *data)
5cf58530fa628b Ivan Orlov 2024-07-26  64  {
5cf58530fa628b Ivan Orlov 2024-07-26  65  	int i;
5cf58530fa628b Ivan Orlov 2024-07-26  66  	int *fd = (int *)data;
5cf58530fa628b Ivan Orlov 2024-07-26  67  
5cf58530fa628b Ivan Orlov 2024-07-26  68  	for (i = 0; i < TICKS_COUNT; i++) {
5cf58530fa628b Ivan Orlov 2024-07-26  69  		/* Well, trigger the timer! */
5cf58530fa628b Ivan Orlov 2024-07-26  70  		ioctl(*fd, SNDRV_TIMER_IOCTL_TRIGGER, NULL);
5cf58530fa628b Ivan Orlov 2024-07-26  71  		sleep(TIMER_FREQ_SEC);
5cf58530fa628b Ivan Orlov 2024-07-26  72  	}
5cf58530fa628b Ivan Orlov 2024-07-26 @73  }
5cf58530fa628b Ivan Orlov 2024-07-26  74  

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH 3/4] ALSA: timer: Introduce virtual userspace-driven timers

[Christophe JAILLET]

Le 26/07/2024 à 09:47, Ivan Orlov a écrit :
> Implement two ioctl calls in order to support virtual userspace-driven
> ALSA timers.
> 
> The first ioctl is SNDRV_TIMER_IOCTL_CREATE, which gets the
> snd_utimer_info struct as a parameter and returns a file descriptor of
> a virtual timer. It also updates the `id` field of the snd_utimer_info
> struct, which provides a unique identifier for the timer (basically,
> the subdevice number which can be used when creating timer instances).
> 
> This patch also introduces a tiny id allocator for the userspace-driven
> timers, which guarantees that we don't have more than 128 of them in the
> system.
> 
> Another ioctl is SNDRV_TIMER_IOCTL_TRIGGER, which allows us to trigger
> the virtual timer (and calls snd_timer_interrupt for the timer under
> the hood), causing all of the timer instances binded to this timer to
> execute their callbacks.
> 
> The maximum amount of ticks available for the timer is 1 for the sake of
> simplification of the userspace API. 'start', 'stop', 'open' and 'close'
> callbacks for the userspace-driven timers are empty since we don't
> really do any hardware initialization here.
> 
> Suggested-by: Axel Holzinger <aholzinger@gmx.de>
> Signed-off-by: Ivan Orlov <ivan.orlov0322@gmail.com>
> ---

Hi,

...

> diff --git a/sound/core/Kconfig b/sound/core/Kconfig
> index b970a1734647..3cf82641fc67 100644
> --- a/sound/core/Kconfig
> +++ b/sound/core/Kconfig
> @@ -251,6 +251,17 @@ config SND_JACK_INJECTION_DEBUG
>   	  Say Y if you are debugging via jack injection interface.
>   	  If unsure select "N".
>   
> +config SND_UTIMER
> +	bool "Enable support for userspace-controlled virtual timers"
> +	depends on SND_TIMER
> +	help
> +	  Say Y to enable the support of userspace-controlled timers. These
> +	  timers are purely virtual, and they are supposed to be triggered
> +	  from userspace. They could be quite useful when synchronizing the
> +	  sound timing with userspace applications (for instance, when sending
> +	  data through snd-aloop).
> +

Unneeded extra new line.

> +
>   config SND_VMASTER
>   	bool
>   

...

> +static void snd_utimer_free(struct snd_utimer *utimer)
> +{
> +	snd_timer_free(utimer->timer);
> +	snd_utimer_put_id(utimer);

Missing kfree(utimer->name); ?

> +	kfree(utimer);
> +}

...

> +static int snd_utimer_create(struct snd_utimer_info *utimer_info,
> +			     struct snd_utimer **r_utimer)
> +{
> +	struct snd_utimer *utimer;
> +	struct snd_timer *timer;
> +	struct snd_timer_id tid;
> +	int utimer_id;
> +	int err = 0;
> +	char *timer_name;
> +
> +	utimer = kzalloc(sizeof(*utimer), GFP_KERNEL);
> +	if (!utimer)
> +		return -ENOMEM;
> +
> +	timer_name = kzalloc(SNDRV_UTIMER_NAME_LEN, GFP_KERNEL);

kasprintf(GFP_KERNEL, "snd-utimer%d", utimer_id); ?
and SNDRV_UTIMER_NAME_LEN becomes useless too.

In snd_timer_new() it is copied in a char[64] anyway, and if utimer_id 
is small, we could even save a few bytes of memory.

CJ

> +	if (!timer_name) {
> +		kfree(utimer);
> +		return -ENOMEM;
> +	}
> +
> +	/* We hold the ioctl lock here so we won't get a race condition when allocating id */
> +	utimer_id = snd_utimer_take_id();
> +	if (utimer_id < 0) {
> +		err = utimer_id;
> +		goto err_take_id;
> +	}
> +
> +	sprintf(timer_name, "snd-utimer%d", utimer_id);
> +	utimer->name = timer_name;
> +	utimer->id = utimer_id;

...

Re: [PATCH 3/4] ALSA: timer: Introduce virtual userspace-driven timers

[Christophe JAILLET]

Le 26/07/2024 à 09:47, Ivan Orlov a écrit :
> Implement two ioctl calls in order to support virtual userspace-driven
> ALSA timers.
> 
> The first ioctl is SNDRV_TIMER_IOCTL_CREATE, which gets the
> snd_utimer_info struct as a parameter and returns a file descriptor of
> a virtual timer. It also updates the `id` field of the snd_utimer_info
> struct, which provides a unique identifier for the timer (basically,
> the subdevice number which can be used when creating timer instances).
> 
> This patch also introduces a tiny id allocator for the userspace-driven
> timers, which guarantees that we don't have more than 128 of them in the
> system.
> 
> Another ioctl is SNDRV_TIMER_IOCTL_TRIGGER, which allows us to trigger
> the virtual timer (and calls snd_timer_interrupt for the timer under
> the hood), causing all of the timer instances binded to this timer to
> execute their callbacks.
> 
> The maximum amount of ticks available for the timer is 1 for the sake of
> simplification of the userspace API. 'start', 'stop', 'open' and 'close'
> callbacks for the userspace-driven timers are empty since we don't
> really do any hardware initialization here.
> 
> Suggested-by: Axel Holzinger <aholzinger@gmx.de>
> Signed-off-by: Ivan Orlov <ivan.orlov0322@gmail.com>
> ---

...

> +#ifdef CONFIG_SND_UTIMER
> +/*
> + * Since userspace-driven timers are passed to userspace, we need to have an identifier
> + * which will allow us to use them (basically, the subdevice number of udriven timer).
> + *
> + * We have a pool of SNDRV_UTIMERS_MAX_COUNT ids from 0 to (SNDRV_UTIMERS_MAX_COUNT - 1).
> + * When we take one of them, the corresponding entry in snd_utimer_ids becomes true.
> + */
> +static bool snd_utimer_ids[SNDRV_UTIMERS_MAX_COUNT];
> +
> +static void snd_utimer_put_id(struct snd_utimer *utimer)
> +{
> +	int timer_id = utimer->id;
> +
> +	snd_BUG_ON(timer_id < 0 || timer_id >= SNDRV_UTIMERS_MAX_COUNT);
> +	snd_utimer_ids[timer_id] = false;
> +}
> +
> +static int snd_utimer_take_id(void)
> +{
> +	size_t i;
> +
> +	for (i = 0; i < SNDRV_UTIMERS_MAX_COUNT; i++) {
> +		if (!snd_utimer_ids[i]) {
> +			snd_utimer_ids[i] = true;
> +			return i;
> +		}
> +	}
> +
> +	return -EBUSY;
> +}

Also the bitmap API could be useful here.

CJ

Re: [PATCH 3/4] ALSA: timer: Introduce virtual userspace-driven timers

[Ivan Orlov]

On 7/28/24 07:52, Christophe JAILLET wrote:
> Hi,
> 

Hi Christophe,

> ...
> 
>> diff --git a/sound/core/Kconfig b/sound/core/Kconfig
>> index b970a1734647..3cf82641fc67 100644
>> --- a/sound/core/Kconfig
>> +++ b/sound/core/Kconfig
>> @@ -251,6 +251,17 @@ config SND_JACK_INJECTION_DEBUG
>>         Say Y if you are debugging via jack injection interface.
>>         If unsure select "N".
>> +config SND_UTIMER
>> +    bool "Enable support for userspace-controlled virtual timers"
>> +    depends on SND_TIMER
>> +    help
>> +      Say Y to enable the support of userspace-controlled timers. These
>> +      timers are purely virtual, and they are supposed to be triggered
>> +      from userspace. They could be quite useful when synchronizing the
>> +      sound timing with userspace applications (for instance, when 
>> sending
>> +      data through snd-aloop).
>> +
> 
> Unneeded extra new line.
> 
>> +
>>   config SND_VMASTER
>>       bool
> 
> ...
> 
>> +static void snd_utimer_free(struct snd_utimer *utimer)
>> +{
>> +    snd_timer_free(utimer->timer);
>> +    snd_utimer_put_id(utimer);
> 
> Missing kfree(utimer->name); ?
> 

Yeah, it definitely should be here... Thank you for finding this!

>> +    kfree(utimer);
>> +}
> 
> ...
> 
>> +static int snd_utimer_create(struct snd_utimer_info *utimer_info,
>> +                 struct snd_utimer **r_utimer)
>> +{
>> +    struct snd_utimer *utimer;
>> +    struct snd_timer *timer;
>> +    struct snd_timer_id tid;
>> +    int utimer_id;
>> +    int err = 0;
>> +    char *timer_name;
>> +
>> +    utimer = kzalloc(sizeof(*utimer), GFP_KERNEL);
>> +    if (!utimer)
>> +        return -ENOMEM;
>> +
>> +    timer_name = kzalloc(SNDRV_UTIMER_NAME_LEN, GFP_KERNEL);
> 
> kasprintf(GFP_KERNEL, "snd-utimer%d", utimer_id); ?
> and SNDRV_UTIMER_NAME_LEN becomes useless too.
> 
> In snd_timer_new() it is copied in a char[64] anyway, and if utimer_id 
> is small, we could even save a few bytes of memory.
> 

Wow, cool, I haven't heard of kasprintf but now I'll use it here in V2. 
Thanks!

-- 
Kind regards,
Ivan Orlov

Re: [PATCH 3/4] ALSA: timer: Introduce virtual userspace-driven timers

[Ivan Orlov]

On 7/28/24 07:59, Christophe JAILLET wrote:
> Le 26/07/2024 à 09:47, Ivan Orlov a écrit :
>> Implement two ioctl calls in order to support virtual userspace-driven
>> ALSA timers.
>>
>> The first ioctl is SNDRV_TIMER_IOCTL_CREATE, which gets the
>> snd_utimer_info struct as a parameter and returns a file descriptor of
>> a virtual timer. It also updates the `id` field of the snd_utimer_info
>> struct, which provides a unique identifier for the timer (basically,
>> the subdevice number which can be used when creating timer instances).
>>
>> This patch also introduces a tiny id allocator for the userspace-driven
>> timers, which guarantees that we don't have more than 128 of them in the
>> system.
>>
>> Another ioctl is SNDRV_TIMER_IOCTL_TRIGGER, which allows us to trigger
>> the virtual timer (and calls snd_timer_interrupt for the timer under
>> the hood), causing all of the timer instances binded to this timer to
>> execute their callbacks.
>>
>> The maximum amount of ticks available for the timer is 1 for the sake of
>> simplification of the userspace API. 'start', 'stop', 'open' and 'close'
>> callbacks for the userspace-driven timers are empty since we don't
>> really do any hardware initialization here.
>>
>> Suggested-by: Axel Holzinger <aholzinger@gmx.de>
>> Signed-off-by: Ivan Orlov <ivan.orlov0322@gmail.com>
>> ---
> 
> ...
> 
>> +#ifdef CONFIG_SND_UTIMER
>> +/*
>> + * Since userspace-driven timers are passed to userspace, we need to 
>> have an identifier
>> + * which will allow us to use them (basically, the subdevice number 
>> of udriven timer).
>> + *
>> + * We have a pool of SNDRV_UTIMERS_MAX_COUNT ids from 0 to 
>> (SNDRV_UTIMERS_MAX_COUNT - 1).
>> + * When we take one of them, the corresponding entry in 
>> snd_utimer_ids becomes true.
>> + */
>> +static bool snd_utimer_ids[SNDRV_UTIMERS_MAX_COUNT];
>> +
>> +static void snd_utimer_put_id(struct snd_utimer *utimer)
>> +{
>> +    int timer_id = utimer->id;
>> +
>> +    snd_BUG_ON(timer_id < 0 || timer_id >= SNDRV_UTIMERS_MAX_COUNT);
>> +    snd_utimer_ids[timer_id] = false;
>> +}
>> +
>> +static int snd_utimer_take_id(void)
>> +{
>> +    size_t i;
>> +
>> +    for (i = 0; i < SNDRV_UTIMERS_MAX_COUNT; i++) {
>> +        if (!snd_utimer_ids[i]) {
>> +            snd_utimer_ids[i] = true;
>> +            return i;
>> +        }
>> +    }
>> +
>> +    return -EBUSY;
>> +}
> 
> Also the bitmap API could be useful here.
> 

Awesome, will use it in V2.

Thank you!


-- 
Kind regards,
Ivan Orlov

Re: [PATCH 3/4] ALSA: timer: Introduce virtual userspace-driven timers

[Christophe JAILLET]

Le 28/07/2024 à 10:51, Ivan Orlov a écrit :
> On 7/28/24 07:59, Christophe JAILLET wrote:
>> Le 26/07/2024 à 09:47, Ivan Orlov a écrit :
>>> Implement two ioctl calls in order to support virtual userspace-driven
>>> ALSA timers.
>>>
>>> The first ioctl is SNDRV_TIMER_IOCTL_CREATE, which gets the
>>> snd_utimer_info struct as a parameter and returns a file descriptor of
>>> a virtual timer. It also updates the `id` field of the snd_utimer_info
>>> struct, which provides a unique identifier for the timer (basically,
>>> the subdevice number which can be used when creating timer instances).
>>>
>>> This patch also introduces a tiny id allocator for the userspace-driven
>>> timers, which guarantees that we don't have more than 128 of them in the
>>> system.
>>>
>>> Another ioctl is SNDRV_TIMER_IOCTL_TRIGGER, which allows us to trigger
>>> the virtual timer (and calls snd_timer_interrupt for the timer under
>>> the hood), causing all of the timer instances binded to this timer to
>>> execute their callbacks.
>>>
>>> The maximum amount of ticks available for the timer is 1 for the sake of
>>> simplification of the userspace API. 'start', 'stop', 'open' and 'close'
>>> callbacks for the userspace-driven timers are empty since we don't
>>> really do any hardware initialization here.
>>>
>>> Suggested-by: Axel Holzinger <aholzinger@gmx.de>
>>> Signed-off-by: Ivan Orlov <ivan.orlov0322@gmail.com>
>>> ---
>>
>> ...
>>
>>> +#ifdef CONFIG_SND_UTIMER
>>> +/*
>>> + * Since userspace-driven timers are passed to userspace, we need to 
>>> have an identifier
>>> + * which will allow us to use them (basically, the subdevice number 
>>> of udriven timer).
>>> + *
>>> + * We have a pool of SNDRV_UTIMERS_MAX_COUNT ids from 0 to 
>>> (SNDRV_UTIMERS_MAX_COUNT - 1).
>>> + * When we take one of them, the corresponding entry in 
>>> snd_utimer_ids becomes true.
>>> + */
>>> +static bool snd_utimer_ids[SNDRV_UTIMERS_MAX_COUNT];
>>> +
>>> +static void snd_utimer_put_id(struct snd_utimer *utimer)
>>> +{
>>> +    int timer_id = utimer->id;
>>> +
>>> +    snd_BUG_ON(timer_id < 0 || timer_id >= SNDRV_UTIMERS_MAX_COUNT);
>>> +    snd_utimer_ids[timer_id] = false;
>>> +}
>>> +
>>> +static int snd_utimer_take_id(void)
>>> +{
>>> +    size_t i;
>>> +
>>> +    for (i = 0; i < SNDRV_UTIMERS_MAX_COUNT; i++) {
>>> +        if (!snd_utimer_ids[i]) {
>>> +            snd_utimer_ids[i] = true;
>>> +            return i;
>>> +        }
>>> +    }
>>> +
>>> +    return -EBUSY;
>>> +}
>>
>> Also the bitmap API could be useful here.
>>
> 
> Awesome, will use it in V2.

Hmm, maybe DEFINE_IDA(), ida_alloc_max() and ida_free() would be even 
better.

CJ

> 
> Thank you!
> 
> 

Re: [PATCH 3/4] ALSA: timer: Introduce virtual userspace-driven timers

[Ivan Orlov]

On 7/28/24 10:30, Christophe JAILLET wrote:
> Le 28/07/2024 à 10:51, Ivan Orlov a écrit :
>> On 7/28/24 07:59, Christophe JAILLET wrote:
>>> Le 26/07/2024 à 09:47, Ivan Orlov a écrit :
>>>> Implement two ioctl calls in order to support virtual userspace-driven
>>>> ALSA timers.
>>>>
>>>> The first ioctl is SNDRV_TIMER_IOCTL_CREATE, which gets the
>>>> snd_utimer_info struct as a parameter and returns a file descriptor of
>>>> a virtual timer. It also updates the `id` field of the snd_utimer_info
>>>> struct, which provides a unique identifier for the timer (basically,
>>>> the subdevice number which can be used when creating timer instances).
>>>>
>>>> This patch also introduces a tiny id allocator for the userspace-driven
>>>> timers, which guarantees that we don't have more than 128 of them in 
>>>> the
>>>> system.
>>>>
>>>> Another ioctl is SNDRV_TIMER_IOCTL_TRIGGER, which allows us to trigger
>>>> the virtual timer (and calls snd_timer_interrupt for the timer under
>>>> the hood), causing all of the timer instances binded to this timer to
>>>> execute their callbacks.
>>>>
>>>> The maximum amount of ticks available for the timer is 1 for the 
>>>> sake of
>>>> simplification of the userspace API. 'start', 'stop', 'open' and 
>>>> 'close'
>>>> callbacks for the userspace-driven timers are empty since we don't
>>>> really do any hardware initialization here.
>>>>
>>>> Suggested-by: Axel Holzinger <aholzinger@gmx.de>
>>>> Signed-off-by: Ivan Orlov <ivan.orlov0322@gmail.com>
>>>> ---
>>>
>>> ...
>>>
>>>> +#ifdef CONFIG_SND_UTIMER
>>>> +/*
>>>> + * Since userspace-driven timers are passed to userspace, we need 
>>>> to have an identifier
>>>> + * which will allow us to use them (basically, the subdevice number 
>>>> of udriven timer).
>>>> + *
>>>> + * We have a pool of SNDRV_UTIMERS_MAX_COUNT ids from 0 to 
>>>> (SNDRV_UTIMERS_MAX_COUNT - 1).
>>>> + * When we take one of them, the corresponding entry in 
>>>> snd_utimer_ids becomes true.
>>>> + */
>>>> +static bool snd_utimer_ids[SNDRV_UTIMERS_MAX_COUNT];
>>>> +
>>>> +static void snd_utimer_put_id(struct snd_utimer *utimer)
>>>> +{
>>>> +    int timer_id = utimer->id;
>>>> +
>>>> +    snd_BUG_ON(timer_id < 0 || timer_id >= SNDRV_UTIMERS_MAX_COUNT);
>>>> +    snd_utimer_ids[timer_id] = false;
>>>> +}
>>>> +
>>>> +static int snd_utimer_take_id(void)
>>>> +{
>>>> +    size_t i;
>>>> +
>>>> +    for (i = 0; i < SNDRV_UTIMERS_MAX_COUNT; i++) {
>>>> +        if (!snd_utimer_ids[i]) {
>>>> +            snd_utimer_ids[i] = true;
>>>> +            return i;
>>>> +        }
>>>> +    }
>>>> +
>>>> +    return -EBUSY;
>>>> +}
>>>
>>> Also the bitmap API could be useful here.
>>>
>>
>> Awesome, will use it in V2.
> 
> Hmm, maybe DEFINE_IDA(), ida_alloc_max() and ida_free() would be even 
> better.
> 

It looks like IDA allocator uses XArrays under the hood to allocate ids 
between 0 and INT_MAX... Considering the fact, that we currently could 
have up to 128 userspace-driven timers in the system, using XArrays 
seems a bit redundant, and I believe bitmap approach would be more 
efficient. What do you think?

-- 
Kind regards,
Ivan Orlov

Re: [PATCH 3/4] ALSA: timer: Introduce virtual userspace-driven timers

[Christophe JAILLET]

Le 28/07/2024 à 11:42, Ivan Orlov a écrit :
> On 7/28/24 10:30, Christophe JAILLET wrote:
>> Le 28/07/2024 à 10:51, Ivan Orlov a écrit :
>>> On 7/28/24 07:59, Christophe JAILLET wrote:
>>>> Le 26/07/2024 à 09:47, Ivan Orlov a écrit :
>>>>> Implement two ioctl calls in order to support virtual userspace-driven
>>>>> ALSA timers.
>>>>>
>>>>> The first ioctl is SNDRV_TIMER_IOCTL_CREATE, which gets the
>>>>> snd_utimer_info struct as a parameter and returns a file descriptor of
>>>>> a virtual timer. It also updates the `id` field of the snd_utimer_info
>>>>> struct, which provides a unique identifier for the timer (basically,
>>>>> the subdevice number which can be used when creating timer instances).
>>>>>
>>>>> This patch also introduces a tiny id allocator for the 
>>>>> userspace-driven
>>>>> timers, which guarantees that we don't have more than 128 of them 
>>>>> in the
>>>>> system.
>>>>>
>>>>> Another ioctl is SNDRV_TIMER_IOCTL_TRIGGER, which allows us to trigger
>>>>> the virtual timer (and calls snd_timer_interrupt for the timer under
>>>>> the hood), causing all of the timer instances binded to this timer to
>>>>> execute their callbacks.
>>>>>
>>>>> The maximum amount of ticks available for the timer is 1 for the 
>>>>> sake of
>>>>> simplification of the userspace API. 'start', 'stop', 'open' and 
>>>>> 'close'
>>>>> callbacks for the userspace-driven timers are empty since we don't
>>>>> really do any hardware initialization here.
>>>>>
>>>>> Suggested-by: Axel Holzinger <aholzinger@gmx.de>
>>>>> Signed-off-by: Ivan Orlov <ivan.orlov0322@gmail.com>
>>>>> ---
>>>>
>>>> ...
>>>>
>>>>> +#ifdef CONFIG_SND_UTIMER
>>>>> +/*
>>>>> + * Since userspace-driven timers are passed to userspace, we need 
>>>>> to have an identifier
>>>>> + * which will allow us to use them (basically, the subdevice 
>>>>> number of udriven timer).
>>>>> + *
>>>>> + * We have a pool of SNDRV_UTIMERS_MAX_COUNT ids from 0 to 
>>>>> (SNDRV_UTIMERS_MAX_COUNT - 1).
>>>>> + * When we take one of them, the corresponding entry in 
>>>>> snd_utimer_ids becomes true.
>>>>> + */
>>>>> +static bool snd_utimer_ids[SNDRV_UTIMERS_MAX_COUNT];
>>>>> +
>>>>> +static void snd_utimer_put_id(struct snd_utimer *utimer)
>>>>> +{
>>>>> +    int timer_id = utimer->id;
>>>>> +
>>>>> +    snd_BUG_ON(timer_id < 0 || timer_id >= SNDRV_UTIMERS_MAX_COUNT);
>>>>> +    snd_utimer_ids[timer_id] = false;
>>>>> +}
>>>>> +
>>>>> +static int snd_utimer_take_id(void)
>>>>> +{
>>>>> +    size_t i;
>>>>> +
>>>>> +    for (i = 0; i < SNDRV_UTIMERS_MAX_COUNT; i++) {
>>>>> +        if (!snd_utimer_ids[i]) {
>>>>> +            snd_utimer_ids[i] = true;
>>>>> +            return i;
>>>>> +        }
>>>>> +    }
>>>>> +
>>>>> +    return -EBUSY;
>>>>> +}
>>>>
>>>> Also the bitmap API could be useful here.
>>>>
>>>
>>> Awesome, will use it in V2.
>>
>> Hmm, maybe DEFINE_IDA(), ida_alloc_max() and ida_free() would be even 
>> better.
>>
> 
> It looks like IDA allocator uses XArrays under the hood to allocate ids 
> between 0 and INT_MAX... Considering the fact, that we currently could 
> have up to 128 userspace-driven timers in the system, using XArrays 
> seems a bit redundant, and I believe bitmap approach would be more 
> efficient. What do you think?
> 

I may be wrong but I think that ida allocates hunks for 1024 bits (128 
bytes * 8) at a time. (see [1])

So with this extra sape and the sapce for the xarray, it would waste a 
few bytes of memory, yes.

With ida, there is also some locking that may be unnecessary (but harmless)


Hoping, I got it right, here are a few numbers:

On a x86_64, with allmodconfig:

Your initial patch:
    text	   data	    bss	    dec	    hex	filename
   55020	   1783	    268	  57071	   deef	sound/core/timer.o

With ida:
   54763	   1631	    116	  56510	   dcbe	sound/core/timer.o
+ 128 bytes of runtime memory allocation

With bitmap:
   54805	   1535	    132	  56472	   dc98	sound/core/timer.o


I think that the code would be slightly more elegant with ida, but 
implementing it with a bitmap does not add that much complexity.

CJ


[1]: 
https://elixir.bootlin.com/linux/v6.10.2/source/include/linux/idr.h#L238

Re: [PATCH 3/4] ALSA: timer: Introduce virtual userspace-driven timers

[Ivan Orlov]

On 7/28/24 11:29, Christophe JAILLET wrote:
> 
> I may be wrong but I think that ida allocates hunks for 1024 bits (128 
> bytes * 8) at a time. (see [1])
> 
> So with this extra sape and the sapce for the xarray, it would waste a 
> few bytes of memory, yes.
> 
> With ida, there is also some locking that may be unnecessary (but harmless)
> 
> 
> Hoping, I got it right, here are a few numbers:
> 
> On a x86_64, with allmodconfig:
> 
> Your initial patch:
>     text       data        bss        dec        hex    filename
>    55020       1783        268      57071       deef    sound/core/timer.o
> 
> With ida:
>    54763       1631        116      56510       dcbe    sound/core/timer.o
> + 128 bytes of runtime memory allocation
> 
> With bitmap:
>    54805       1535        132      56472       dc98    sound/core/timer.o
> 
> 
> I think that the code would be slightly more elegant with ida, but 
> implementing it with a bitmap does not add that much complexity.
> 

Ah, alright, I agree that the code would be cleaner when using IDA, and 
such a small memory overhead won't be significant/noticeable. I'm going 
to use IDA in the V2 instead of bitmap API, thank you so much for 
pointing me to it (I was wondering if the Kernel has a generic ID 
allocator and now I finally know it does :) ).

Thank you!

-- 
Kind regards,
Ivan Orlov