[PATCH 0/3] selftests: ALSA: pcmtest: Test error injection and pause

[PATCH 0/3] selftests: ALSA: pcmtest: Test error injection and pause

[HyeongJun An]

The pcmtest driver can return errors from its PCM callbacks and
supports pause, but the selftest covers neither. This series adds tests
for both.

Patch 1 moves the capture data check into a helper, so patch 3 can
reuse it. Patch 2 tests each inject_*_err parameter. Patch 3 tests
pausing a capture stream.

Tested on x86_64 with the sound/for-next driver, built for a 6.17 host.
All 10 tests pass in three runs. To check that the pause test catches
real bugs, I loaded two broken drivers. One keeps running on PAUSE_PUSH.
The other restarts the data on PAUSE_RELEASE. Only capture_pause fails
on each.

HyeongJun An (3):
  selftests: ALSA: pcmtest: Move the capture data check into a helper
  selftests: ALSA: pcmtest: Test errors from the PCM callbacks
  selftests: ALSA: pcmtest: Test pausing a capture stream

 .../selftests/alsa/test-pcmtest-driver.c      | 167 +++++++++++++++++-
 1 file changed, 158 insertions(+), 9 deletions(-)

-- 
2.43.0

[PATCH 1/3] selftests: ALSA: pcmtest: Move the capture data check into a helper

[HyeongJun An]

The capture test checks the captured bytes against the patterns in an
inline loop. A pause test added later needs the same check, so move the
loop into find_capture_mismatch(). It returns the offset of the first
wrong byte, or -1 when all bytes match.

No functional change.

Assisted-by: Claude:claude-opus-5-5
Signed-off-by: HyeongJun An <sammiee5311@gmail.com>
---
 .../selftests/alsa/test-pcmtest-driver.c      | 29 +++++++++++++------
 1 file changed, 20 insertions(+), 9 deletions(-)

diff --git a/tools/testing/selftests/alsa/test-pcmtest-driver.c b/tools/testing/selftests/alsa/test-pcmtest-driver.c
index 18ee72d5d4c9..fc4d12542808 100644
--- a/tools/testing/selftests/alsa/test-pcmtest-driver.c
+++ b/tools/testing/selftests/alsa/test-pcmtest-driver.c
@@ -82,6 +82,25 @@ static size_t get_sec_buf_len(unsigned int rate, unsigned long channels, snd_pcm
 	return rate * channels * snd_pcm_format_physical_width(format) / 8;
 }
 
+/*
+ * Check interleaved capture data against the patterns. Return the offset of the first wrong
+ * byte, or -1 if all bytes match.
+ */
+static long find_capture_mismatch(const unsigned char *it, struct pcmtest_test_params *params)
+{
+	size_t i, cur_ch, pos_in_ch;
+
+	for (i = 0; i < params->sec_buf_len * params->time; i++) {
+		cur_ch = (i / params->sample_size) % CH_NUM;
+		pos_in_ch = i / params->sample_size / CH_NUM * params->sample_size
+			    + (i % params->sample_size);
+		if (it[i] != patterns[cur_ch].buf[pos_in_ch % patterns[cur_ch].len])
+			return i;
+	}
+
+	return -1;
+}
+
 static int setup_handle(snd_pcm_t **handle, snd_pcm_sw_params_t *swparams,
 			snd_pcm_hw_params_t *hwparams, struct pcmtest_test_params *params,
 			int card, snd_pcm_stream_t stream)
@@ -207,9 +226,7 @@ TEST_F(pcmtest, playback) {
  */
 TEST_F(pcmtest, capture) {
 	snd_pcm_t *handle;
-	unsigned char *it;
 	snd_pcm_sframes_t read_res;
-	int i, cur_ch, pos_in_ch;
 	void *samples;
 	struct pcmtest_test_params *params = &self->params;
 
@@ -226,13 +243,7 @@ TEST_F(pcmtest, capture) {
 	read_res = snd_pcm_readi(handle, samples, params->rate * params->time);
 	ASSERT_EQ(read_res, params->rate * params->time);
 	snd_pcm_close(handle);
-	it = (unsigned char *)samples;
-	for (i = 0; i < self->params.sec_buf_len * self->params.time; i++) {
-		cur_ch = (i / params->sample_size) % CH_NUM;
-		pos_in_ch = i / params->sample_size / CH_NUM * params->sample_size
-			    + (i % params->sample_size);
-		ASSERT_EQ(it[i], patterns[cur_ch].buf[pos_in_ch % patterns[cur_ch].len]);
-	}
+	ASSERT_EQ(find_capture_mismatch(samples, params), -1);
 	free(samples);
 }
 
-- 
2.43.0

[PATCH 2/3] selftests: ALSA: pcmtest: Test errors from the PCM callbacks

[HyeongJun An]

The driver can return an error from its open, hw_params, prepare and
trigger callbacks, controlled by the inject_*_err module parameters.
No test uses them, so nothing checks that these errors reach user space.

Add a test for each callback. It sets the parameter, checks that the
matching call fails with the driver's error, then clears the parameter
and checks that the call succeeds. The trigger test also checks that a
failed start leaves the stream in the PREPARED state. The fixture
teardown clears all parameters, so a failed test does not affect the
next one.

Assisted-by: Claude:claude-opus-5-5
Signed-off-by: HyeongJun An <sammiee5311@gmail.com>
---
 .../selftests/alsa/test-pcmtest-driver.c      | 89 +++++++++++++++++++
 1 file changed, 89 insertions(+)

diff --git a/tools/testing/selftests/alsa/test-pcmtest-driver.c b/tools/testing/selftests/alsa/test-pcmtest-driver.c
index fc4d12542808..380364ee7924 100644
--- a/tools/testing/selftests/alsa/test-pcmtest-driver.c
+++ b/tools/testing/selftests/alsa/test-pcmtest-driver.c
@@ -5,6 +5,7 @@
  *
  * Copyright 2023 Ivan Orlov <ivan.orlov0322@gmail.com>
  */
+#include <errno.h>
 #include <string.h>
 #include <alsa/asoundlib.h>
 #include "kselftest_harness.h"
@@ -58,6 +59,20 @@ static int read_patterns(void)
 	return 0;
 }
 
+static int set_param(const char *name, int value)
+{
+	char path[128];
+	FILE *f;
+
+	sprintf(path, "/sys/module/snd_pcmtest/parameters/%s", name);
+	f = fopen(path, "w");
+	if (!f)
+		return -1;
+	fprintf(f, "%d", value);
+
+	return fclose(f) ? -1 : 0;
+}
+
 static int get_test_results(char *debug_name)
 {
 	int result;
@@ -141,6 +156,10 @@ FIXTURE(pcmtest) {
 };
 
 FIXTURE_TEARDOWN(pcmtest) {
+	set_param("inject_open_err", 0);
+	set_param("inject_hwpars_err", 0);
+	set_param("inject_prepare_err", 0);
+	set_param("inject_trigger_err", 0);
 }
 
 FIXTURE_SETUP(pcmtest) {
@@ -343,4 +362,74 @@ TEST_F(pcmtest, reset_ioctl) {
 	snd_pcm_close(handle);
 }
 
+/*
+ * Here we check that an error from each PCM callback reaches user space. The driver returns
+ * the error while the matching 'inject_*_err' parameter is set, and works again once it is
+ * cleared.
+ */
+TEST_F(pcmtest, open_err) {
+	snd_pcm_t *handle;
+	char pcm_name[32];
+
+	sprintf(pcm_name, "hw:%d,0,0", self->card);
+
+	ASSERT_EQ(set_param("inject_open_err", 1), 0);
+	ASSERT_EQ(snd_pcm_open(&handle, pcm_name, SND_PCM_STREAM_PLAYBACK, 0), -EBUSY);
+
+	ASSERT_EQ(set_param("inject_open_err", 0), 0);
+	ASSERT_EQ(snd_pcm_open(&handle, pcm_name, SND_PCM_STREAM_PLAYBACK, 0), 0);
+	snd_pcm_close(handle);
+}
+
+TEST_F(pcmtest, hw_params_err) {
+	snd_pcm_t *handle;
+	char pcm_name[32];
+
+	snd_pcm_hw_params_alloca(&self->hwparams);
+	sprintf(pcm_name, "hw:%d,0,0", self->card);
+	ASSERT_EQ(snd_pcm_open(&handle, pcm_name, SND_PCM_STREAM_PLAYBACK, 0), 0);
+	ASSERT_EQ(snd_pcm_hw_params_any(handle, self->hwparams), 0);
+
+	ASSERT_EQ(set_param("inject_hwpars_err", 1), 0);
+	ASSERT_EQ(snd_pcm_hw_params(handle, self->hwparams), -EBUSY);
+
+	ASSERT_EQ(set_param("inject_hwpars_err", 0), 0);
+	ASSERT_EQ(snd_pcm_hw_params(handle, self->hwparams), 0);
+	snd_pcm_close(handle);
+}
+
+TEST_F(pcmtest, prepare_err) {
+	snd_pcm_t *handle;
+
+	snd_pcm_sw_params_alloca(&self->swparams);
+	snd_pcm_hw_params_alloca(&self->hwparams);
+	ASSERT_EQ(setup_handle(&handle, self->swparams, self->hwparams, &self->params,
+			       self->card, SND_PCM_STREAM_PLAYBACK), 0);
+
+	ASSERT_EQ(set_param("inject_prepare_err", 1), 0);
+	ASSERT_EQ(snd_pcm_prepare(handle), -EINVAL);
+
+	ASSERT_EQ(set_param("inject_prepare_err", 0), 0);
+	ASSERT_EQ(snd_pcm_prepare(handle), 0);
+	snd_pcm_close(handle);
+}
+
+TEST_F(pcmtest, trigger_err) {
+	snd_pcm_t *handle;
+
+	snd_pcm_sw_params_alloca(&self->swparams);
+	snd_pcm_hw_params_alloca(&self->hwparams);
+	ASSERT_EQ(setup_handle(&handle, self->swparams, self->hwparams, &self->params,
+			       self->card, SND_PCM_STREAM_CAPTURE), 0);
+
+	ASSERT_EQ(set_param("inject_trigger_err", 1), 0);
+	ASSERT_EQ(snd_pcm_start(handle), -EINVAL);
+	ASSERT_EQ(snd_pcm_state(handle), SND_PCM_STATE_PREPARED);
+
+	ASSERT_EQ(set_param("inject_trigger_err", 0), 0);
+	ASSERT_EQ(snd_pcm_start(handle), 0);
+	ASSERT_EQ(snd_pcm_state(handle), SND_PCM_STATE_RUNNING);
+	snd_pcm_close(handle);
+}
+
 TEST_HARNESS_MAIN
-- 
2.43.0

[PATCH 3/3] selftests: ALSA: pcmtest: Test pausing a capture stream

[HyeongJun An]

The driver supports pause, but no test uses it.

Add a capture test that pauses the stream halfway for half a second.
The core does not read the hardware pointer while the stream is
paused, so a driver that keeps running shows up only at the release,
as a jump of the pointer. Check that the jump is under a tenth of a
second, and that the data carries on from where it stopped.

Assisted-by: Claude:claude-opus-5-5
Signed-off-by: HyeongJun An <sammiee5311@gmail.com>
---
 .../selftests/alsa/test-pcmtest-driver.c      | 49 +++++++++++++++++++
 1 file changed, 49 insertions(+)

diff --git a/tools/testing/selftests/alsa/test-pcmtest-driver.c b/tools/testing/selftests/alsa/test-pcmtest-driver.c
index 380364ee7924..b796a5613c6f 100644
--- a/tools/testing/selftests/alsa/test-pcmtest-driver.c
+++ b/tools/testing/selftests/alsa/test-pcmtest-driver.c
@@ -7,6 +7,7 @@
  */
 #include <errno.h>
 #include <string.h>
+#include <unistd.h>
 #include <alsa/asoundlib.h>
 #include "kselftest_harness.h"
 
@@ -362,6 +363,54 @@ TEST_F(pcmtest, reset_ioctl) {
 	snd_pcm_close(handle);
 }
 
+/*
+ * Here we pause a capture stream halfway. The core does not read the hardware pointer while
+ * the stream is paused, so a driver that keeps running only shows up after the release, as a
+ * jump of the pointer. The data must also continue where it stopped, so the whole buffer still
+ * matches the patterns.
+ */
+TEST_F(pcmtest, capture_pause) {
+	snd_pcm_t *handle;
+	snd_pcm_sframes_t read_res, paused_avail, avail, frames, half;
+	unsigned char *samples;
+	struct pcmtest_test_params *params = &self->params;
+
+	samples = calloc(params->sec_buf_len * params->time, 1);
+	ASSERT_NE(samples, NULL);
+
+	snd_pcm_sw_params_alloca(&self->swparams);
+	snd_pcm_hw_params_alloca(&self->hwparams);
+	ASSERT_EQ(setup_handle(&handle, self->swparams, self->hwparams, params,
+			       self->card, SND_PCM_STREAM_CAPTURE), 0);
+
+	frames = params->rate * params->time;
+	half = frames / 2;
+	read_res = snd_pcm_readi(handle, samples, half);
+	ASSERT_EQ(read_res, half);
+
+	ASSERT_EQ(snd_pcm_pause(handle, 1), 0);
+	ASSERT_EQ(snd_pcm_state(handle), SND_PCM_STATE_PAUSED);
+	paused_avail = snd_pcm_avail(handle);
+	ASSERT_GE(paused_avail, 0);
+	usleep(500000);
+
+	/*
+	 * At 8000 Hz the pause is 4000 frames long. Allow a tenth of a second (800 frames) for
+	 * the release.
+	 */
+	ASSERT_EQ(snd_pcm_pause(handle, 0), 0);
+	avail = snd_pcm_avail(handle);
+	ASSERT_GE(avail, 0);
+	ASSERT_LT(avail - paused_avail, (snd_pcm_sframes_t)params->rate / 10);
+	read_res = snd_pcm_readi(handle, samples + snd_pcm_frames_to_bytes(handle, half),
+				 frames - half);
+	ASSERT_EQ(read_res, frames - half);
+	snd_pcm_close(handle);
+
+	ASSERT_EQ(find_capture_mismatch(samples, params), -1);
+	free(samples);
+}
+
 /*
  * Here we check that an error from each PCM callback reaches user space. The driver returns
  * the error while the matching 'inject_*_err' parameter is set, and works again once it is
-- 
2.43.0

Re: [PATCH 0/3] selftests: ALSA: pcmtest: Test error injection and pause

[Takashi Iwai]

On Thu, 22 Oct 2026 13:12:41 +0200,
HyeongJun An wrote:
> 
> The pcmtest driver can return errors from its PCM callbacks and
> supports pause, but the selftest covers neither. This series adds tests
> for both.
> 
> Patch 1 moves the capture data check into a helper, so patch 3 can
> reuse it. Patch 2 tests each inject_*_err parameter. Patch 3 tests
> pausing a capture stream.
> 
> Tested on x86_64 with the sound/for-next driver, built for a 6.17 host.
> All 10 tests pass in three runs. To check that the pause test catches
> real bugs, I loaded two broken drivers. One keeps running on PAUSE_PUSH.
> The other restarts the data on PAUSE_RELEASE. Only capture_pause fails
> on each.
> 
> HyeongJun An (3):
>   selftests: ALSA: pcmtest: Move the capture data check into a helper
>   selftests: ALSA: pcmtest: Test errors from the PCM callbacks
>   selftests: ALSA: pcmtest: Test pausing a capture stream

JFYI, I postpone merging this for 7.5, as it doesn't look like a
regression fix.


thanks,

Takashi