diff -Naur SDL2-2.0.10/src/haptic/linux/SDL_syshaptic.c SDL2-2.0.10.patched/src/haptic/linux/SDL_syshaptic.c
--- SDL2-2.0.10/src/haptic/linux/SDL_syshaptic.c	2019-07-24 21:32:36.000000000 -0700
+++ SDL2-2.0.10.patched/src/haptic/linux/SDL_syshaptic.c	2019-09-04 10:52:36.784736281 -0700
@@ -974,6 +974,7 @@
                            SDL_HapticEffect * data)
 {
     struct ff_effect linux_effect;
+    int retry_count = 0;

     /* Create the new effect */
     if (SDL_SYS_ToFFEffect(&linux_effect, data) != 0) {
@@ -981,10 +982,29 @@
     }
     linux_effect.id = effect->hweffect->effect.id;

+retry:
     /* See if it can be uploaded. */
     if (ioctl(haptic->hwdata->fd, EVIOCSFF, &linux_effect) < 0) {
-        return SDL_SetError("Haptic: Error updating the effect: %s",
-                            strerror(errno));
+        int err = errno;
+        char const * err_str = strerror(err);
+
+        /*
+         * For some strange and inexplicable reason, the kernel driver appears
+         * to "forget" about our effect for... "reasons."
+         *
+         * The only difference between creating a "new" effect and "updating" an
+         * effect, in the kernel API, is the use of the effect id of "-1" to
+         * request a new id.
+         *
+         * If we get EINVAL, try to reinitialize the effect by setting the index
+         * back to -1, but don't retry more than once. EINVAL is pretty vague
+         * and there could be other causes.
+         */
+        if ((err == EINVAL) && (retry_count++ == 0)) {
+            linux_effect.id = -1;
+            goto retry;
+        }
+        return SDL_SetError("Haptic: Error updating the effect: %s", err_str);
     }

     /* Copy the new effect into memory. */