Mikmod Sound System / Bugs / #30 wasapi driver segfault upon MikMod

Stephane Denis - 2026-05-07

assigned_to: Stephane Denis
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Stephane Denis - 2026-05-07

Problem is in WASAPI_Exit
VC1_Exit clear the voices but doesn't re-create them. needs to be removed

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Stephane Denis - 2026-05-07

Fixed in commit 099b986137b115f16821ce341048a5de1e331ce1
Removed VC1_Exit on DirectSound and Wasapi.
Tested with mikmod app

open mikmod with a module as parameter

change configuration (Loops)

use configuration

-> No more crash (vinf is not deleted in virtch code).
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Stephane Denis - 2026-05-07

status: open --> closed
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Ozkan Sezer - 2026-05-07

Every driver calls VC_Exit() in their Exit() procedure, none does segfault. drv_ds didn't used to crash either. Why is this a problem with drv_wasapi?

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Stephane Denis - 2026-05-07

the issue is MikMod_Reset, reset (with WASAPI_PlayStart) but the voice are not recreated.
on better fix is to do
md_driver->Exit();
md_driver->SetNumVoices();

because if the music is played and mikmod is reset, then you need to recreate the voices else it will continue to write to vinf.

tell me if it better and will revert and use that better fix instead

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Ozkan Sezer - 2026-05-07

Sounds logical. But what I do not understand is why is it a problem only with drv_wasapi? And drv_ds did not crash before either.. Thinking about it, drv_ds leaves first update to MiMod_Update() via DS_Update(): see do_update in there.

tell me if it better and will revert and use that better fix instead

Maybe?. I'm still trying to understand things, though..

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Stephane Denis - 2026-05-08

the gap is mmreset exit the driver and immediately call mddriver->PlayStart() and wasapi audiodevicestart immediately start to produce audiodevicerender. there is no latency, that's why the voice need to be reset before voice start.

that's the difference from other driver.

The real fix is that voice needs to be recreated because calling playstart. ( SetNumVoices)

it didn't crash for other driver, but this is definitely something was omitted.

Note that is happens only the the music is playing, technically with the fix you don't need to have mikmod to reload the music, you can keep the current music position

Here the callstack

audio_device_render(AudioDevice , unsigned char , unsigned int) << Call VC1_Volume and crash because vinf not yet recreated (VC1_SetNumVoices not called until music is reloaded by mikmod)
audio_device_start(AudioDevice )
WASAPI_PlayStart() << Audio is restarting
_mm_reset(const char )
MikMod_Reset(const char ) << Reset config
Player_SetConfig(CONFIG )
handle_menu(MMENU )
menu_handle_key(MWINDOW , int)
win_handle_key(int)
win_main_iteration()
win_run()
main(int, char * *)

technically you can remove also the

framesAvail = dev->bufferFrames - padding;
if (framesAvail > 0) {
hr = IAudioRenderClient_GetBuffer(dev->renderClient, framesAvail, &pData);
if (FAILED(hr)) return FALSE;
if (pData) {
BOOL produced = audio_device_render(dev, pData, framesAvail);
if (!produced)
memset(pData, 0, (size_t)framesAvail * bytes_per_frame(dev->mixFormat));
}
IAudioRenderClient_ReleaseBuffer(dev->renderClient, framesAvail, 0);
}

this the part that 'calls player tick to honor pending frames' at start of the driver start

Last edit: Stephane Denis 2026-05-08

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

I tested the following patch which (after reverting the previous fix),
out-sources device-start to MikMod_Update(), like it is done in drv_ds
and dr_xaudio. Tested on a windows-11 setup.

What do you think?

(And I can't think of any app that uses these drivers and also not call
MikMod_Update: Can I be mistaken?)

diff --git a/libmikmod/drivers/drv_ds.c b/libmikmod/drivers/drv_ds.c
index f5977dd..9ec638f 100644
--- a/libmikmod/drivers/drv_ds.c
+++ b/libmikmod/drivers/drv_ds.c
@@ -328,6 +328,8 @@ static void DS_Exit(void)
        IDirectSound_Release(pSoundCard);
        pSoundCard = NULL;
    }
+

+   VC_Exit();
 }

 static BOOL do_update = 0;
diff --git a/libmikmod/drivers/drv_wasapi.c b/libmikmod/drivers/drv_wasapi.c
index 642eaba..160f7e0 100644
--- a/libmikmod/drivers/drv_wasapi.c
+++ b/libmikmod/drivers/drv_wasapi.c
@@ -745,24 +745,22 @@ static void WASAPI_Exit(void)
         g_wasapi = NULL;
     }
     g_started = FALSE;

+    VC_Exit();
 }

+static BOOL do_update = 0;
+
 static int WASAPI_PlayStart(void)
 {
     if (VC_PlayStart()) return 1;


-    if (g_wasapi && !audio_device_start(g_wasapi)) {
-        VC_PlayStop();
-        _mm_errno = MMERR_OPENING_AUDIO;
-        return 1;
-    }
-
-    g_started = TRUE;
+    do_update = 1;
     return 0;
 }

 static void WASAPI_PlayStop(void)
 {

+    do_update = 0;
     if (g_wasapi && g_started)
         audio_device_stop(g_wasapi);
     g_started = FALSE;
@@ -770,7 +768,21 @@ static void WASAPI_PlayStop(void)
 }

 /* Event-driven backend: no polling needed */
-static void WASAPI_Update(void) {}
+static void WASAPI_Update(void)
+{

+    if (!do_update)
+        return;
+    do_update = 0;
+
+    if (g_wasapi && !audio_device_start(g_wasapi)) {
+        VC_PlayStop();
+        _mm_errno = MMERR_OPENING_AUDIO;
+        WASAPI_Exit();
+        return;
+    }
+
+    g_started = TRUE;
+}

 MIKMODAPI MDRIVER drv_wasapi = {
     NULL,
diff --git a/libmikmod/drivers/drv_sdl.c b/libmikmod/drivers/drv_sdl.c
index e5a1ba2..4fc6a0d 100644
--- a/libmikmod/drivers/drv_sdl.c
+++ b/libmikmod/drivers/drv_sdl.c
@@ -47,19 +47,15 @@ static int enabled = 0;

 static void SDLSoundCallback(void *userdata, Uint8 *stream, int len)
 {

-    if (!enabled) return;
-    if (enabled < 0) {
-        if (++enabled == 0)
-            enabled = 1;
+    if (enabled <= 0) {
+        VC_SilenceBytes((SBYTE *) stream, (ULONG)len);
         return;
     }
-
     MUTEX_LOCK(vars);
     if (Player_Paused_internal()) {
         VC_SilenceBytes((SBYTE *) stream, (ULONG)len);
     }
-    else
-    {
+    else {
         int got = (int) VC_WriteBytes((SBYTE *) stream, (ULONG)len);
         if (got < len) {   /* fill the rest with silence, then */
             VC_SilenceBytes((SBYTE *) &stream[got], (ULONG)(len-got));
@@ -124,7 +120,9 @@ static int SDLDrv_Init(void)
     if (VC_Init())
         return 1;


-    enabled = -2; /* delay the callback 2 frames */
+    enabled = -1;
+    SDL_PauseAudio(0);
+
     return 0;
 }

@@ -145,7 +143,9 @@ static int SDLDrv_Reset(void)

 static void SDLDrv_Update( void )
 {
-/* do nothing */

+    if (enabled < 0) {
+        enabled = 1;
+    }
 }

 static void SDLDrv_PlayStop(void)

Ozkan Sezer - 2026-05-10

Tested the patch on windows-10 too, and applied as b8e3709ebf58ac4d28479bfaec33b9499dcc598c

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Stephane Denis - 2026-05-10

Yes that's a good approach. i approve.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

wasapi driver segfault upon MikMod_Reset()

Group

Searches

Help

#30 wasapi driver segfault upon MikMod_Reset()

Discussion