diff --git a/sys/dev/pcm/channel.c b/sys/dev/pcm/channel.c
index bb46e03f8e88..38e1796bee97 100644
--- a/sys/dev/pcm/channel.c
+++ b/sys/dev/pcm/channel.c
@@ -268,7 +268,7 @@ chn_write(pcm_channel *c, struct uio *buf)
 	}
 	a = (1 << c->align) - 1;
 	c->flags |= CHN_F_WRITING;
-	while (buf->uio_resid > 0) {
+	while ((c->smegcnt + buf->uio_resid) > a) {
 		s = spltty();
 		chn_dmaupdate(c);
 		splx(s);
@@ -282,10 +282,9 @@ chn_write(pcm_channel *c, struct uio *buf)
 			continue;
 		}
 		/* ensure we always have a whole number of samples */
-		l = min(b->fl, b->bufsize - b->fp);
-		if (l & a) panic("unaligned write %d, %d", l, a + 1);
-		l &= ~a;
-		w = c->feeder->feed(c->feeder, b->buf + b->fp, l, buf);
+		l = min(b->fl, b->bufsize - b->fp) & ~a;
+		if (l == 0) break;
+		w = c->feeder->feed(c->feeder, c, b->buf + b->fp, l, buf);
 		if (w == 0) panic("no feed");
 		s = spltty();
 		b->rl += w;
@@ -294,6 +293,12 @@ chn_write(pcm_channel *c, struct uio *buf)
 	      	splx(s);
 		if (b->rl && !b->dl) chn_stintr(c);
 	}
+	if ((ret == 0) && (buf->uio_resid > 0)) {
+		l = buf->uio_resid;
+		if ((c->smegcnt + l) >= SMEGBUFSZ) panic("resid overflow %d", l);
+		uiomove(c->smegbuf + c->smegcnt, l, buf);
+		c->smegcnt += l;
+	}
 	c->flags &= ~CHN_F_WRITING;
 	return ret;
 }
@@ -418,7 +423,7 @@ chn_read(pcm_channel *c, struct uio *buf)
 		}
 		/* ensure we always have a whole number of samples */
 		l = min(b->rl, b->bufsize - b->rp) & DMA_ALIGN_MASK;
-		w = c->feeder->feed(c->feeder, b->buf + b->rp, l, buf);
+		w = c->feeder->feed(c->feeder, c, b->buf + b->rp, l, buf);
 		s = spltty();
 		b->rl -= w;
 		b->fl += w;
@@ -474,6 +479,7 @@ chn_resetbuf(pcm_channel *c)
 	u_int16_t data, *p;
 	u_int32_t i;
 
+	c->smegcnt = 0;
 	c->buffer.sample_size = 1;
 	c->buffer.sample_size <<= (c->hwfmt & AFMT_STEREO)? 1 : 0;
 	c->buffer.sample_size <<= (c->hwfmt & AFMT_16BIT)? 1 : 0;
diff --git a/sys/dev/pcm/datatypes.h b/sys/dev/pcm/datatypes.h
index 4a38679e45c5..a4af384a1c01 100644
--- a/sys/dev/pcm/datatypes.h
+++ b/sys/dev/pcm/datatypes.h
@@ -75,7 +75,8 @@ struct _snd_dbuf {
 
 typedef int (pcmfeed_init_t)(pcm_feeder *feeder);
 typedef int (pcmfeed_free_t)(pcm_feeder *feeder);
-typedef int (pcmfeed_feed_t)(pcm_feeder *feeder, u_int8_t *buffer, u_int32_t count, struct uio *stream);
+typedef int (pcmfeed_feed_t)(pcm_feeder *feeder, pcm_channel *c, u_int8_t *buffer,
+			     u_int32_t count, struct uio *stream);
 
 struct _pcm_feeder {
 	char name[16];
@@ -121,6 +122,9 @@ struct _pcm_channel {
 
 	int direction;
 	snd_dbuf buffer;
+#define	SMEGBUFSZ 4
+	u_int8_t smegbuf[SMEGBUFSZ];
+	u_int32_t smegcnt;
 	void *devinfo;
 };
 
diff --git a/sys/dev/pcm/feeder.c b/sys/dev/pcm/feeder.c
index 4a59806803c7..055dc98d65df 100644
--- a/sys/dev/pcm/feeder.c
+++ b/sys/dev/pcm/feeder.c
@@ -106,10 +106,17 @@ static unsigned char u8_to_ulaw[] = {
 /*****************************************************************************/
 
 static int
-feed_root(pcm_feeder *feeder, u_int8_t *buffer, u_int32_t count, struct uio *stream)
+feed_root(pcm_feeder *feeder, pcm_channel *ch, u_int8_t *buffer, u_int32_t count, struct uio *stream)
 {
 	int ret, tmp = 0, c = 0;
 	if (!count) panic("feed_root: count == 0");
+	count &= ~((1 << ch->align) - 1);
+	if (!count) panic("feed_root: aligned count == 0");
+	if (ch->smegcnt > 0) {
+		c = min(ch->smegcnt, count);
+		bcopy(ch->smegbuf, buffer, c);
+		ch->smegcnt -= c;
+	}
 	while ((stream->uio_resid > 0) && (c < count)) {
 		tmp = stream->uio_resid;
 		ret = uiomove(buffer + c, count - c, stream);
@@ -125,10 +132,10 @@ pcm_feeder feeder_root = { "root", 0, NULL, NULL, feed_root };
 /*****************************************************************************/
 
 static int
-feed_8to16(pcm_feeder *f, u_int8_t *b, u_int32_t count, struct uio *stream)
+feed_8to16(pcm_feeder *f, pcm_channel *c, u_int8_t *b, u_int32_t count, struct uio *stream)
 {
 	int i, j, k;
-	k = f->source->feed(f->source, b, count / 2, stream);
+	k = f->source->feed(f->source, c, b, count / 2, stream);
 	j = k - 1;
 	i = j * 2 + 1;
 	while (i > 0 && j >= 0) {
@@ -157,11 +164,11 @@ feed_16to8_free(pcm_feeder *f)
 }
 
 static int
-feed_16to8le(pcm_feeder *f, u_int8_t *b, u_int32_t count, struct uio *stream)
+feed_16to8le(pcm_feeder *f, pcm_channel *c, u_int8_t *b, u_int32_t count, struct uio *stream)
 {
 	u_int32_t i = 0, toget = count * 2;
 	int j = 1, k;
-	k = f->source->feed(f->source, f->data, min(toget, FEEDBUFSZ), stream);
+	k = f->source->feed(f->source, c, f->data, min(toget, FEEDBUFSZ), stream);
 	while (j < k) {
 		b[i++] = ((u_int8_t *)f->data)[j];
 		j += 2;
@@ -174,9 +181,9 @@ static pcm_feeder feeder_16to8le =
 /*****************************************************************************/
 
 static int
-feed_monotostereo8(pcm_feeder *f, u_int8_t *b, u_int32_t count, struct uio *stream)
+feed_monotostereo8(pcm_feeder *f, pcm_channel *c, u_int8_t *b, u_int32_t count, struct uio *stream)
 {
-	int i, j, k = f->source->feed(f->source, b, count / 2, stream);
+	int i, j, k = f->source->feed(f->source, c, b, count / 2, stream);
 	j = k - 1;
 	i = j * 2 + 1;
 	while (i > 0 && j >= 0) {
@@ -207,11 +214,11 @@ feed_stereotomono8_free(pcm_feeder *f)
 }
 
 static int
-feed_stereotomono8(pcm_feeder *f, u_int8_t *b, u_int32_t count, struct uio *stream)
+feed_stereotomono8(pcm_feeder *f, pcm_channel *c, u_int8_t *b, u_int32_t count, struct uio *stream)
 {
 	u_int32_t i = 0, toget = count * 2;
 	int j = 0, k;
-	k = f->source->feed(f->source, f->data, min(toget, FEEDBUFSZ), stream);
+	k = f->source->feed(f->source, c, f->data, min(toget, FEEDBUFSZ), stream);
 	while (j < k) {
 		b[i++] = ((u_int8_t *)f->data)[j];
 		j += 2;
@@ -225,10 +232,10 @@ static pcm_feeder feeder_stereotomono8 =
 /*****************************************************************************/
 
 static int
-feed_endian(pcm_feeder *f, u_int8_t *b, u_int32_t count, struct uio *stream)
+feed_endian(pcm_feeder *f, pcm_channel *c, u_int8_t *b, u_int32_t count, struct uio *stream)
 {
 	u_int8_t t;
-	int i = 0, j = f->source->feed(f->source, b, count, stream);
+	int i = 0, j = f->source->feed(f->source, c, b, count, stream);
 	while (i < j) {
 		t = b[i];
 		b[i] = b[i + 1];
@@ -242,9 +249,9 @@ static pcm_feeder feeder_endian = { "endian", -1, NULL, NULL, feed_endian };
 /*****************************************************************************/
 
 static int
-feed_sign(pcm_feeder *f, u_int8_t *b, u_int32_t count, struct uio *stream)
+feed_sign(pcm_feeder *f, pcm_channel *c, u_int8_t *b, u_int32_t count, struct uio *stream)
 {
-	int i = 0, j = f->source->feed(f->source, b, count, stream);
+	int i = 0, j = f->source->feed(f->source, c, b, count, stream);
 	int ssz = (int)f->data, ofs = ssz - 1;
 	while (i < j) {
 		b[i + ofs] ^= 0x80;
@@ -260,9 +267,9 @@ static pcm_feeder feeder_sign16 =
 /*****************************************************************************/
 
 static int
-feed_table(pcm_feeder *f, u_int8_t *b, u_int32_t count, struct uio *stream)
+feed_table(pcm_feeder *f, pcm_channel *c, u_int8_t *b, u_int32_t count, struct uio *stream)
 {
-	int i = 0, j = f->source->feed(f->source, b, count, stream);
+	int i = 0, j = f->source->feed(f->source, c, b, count, stream);
 	while (i < j) {
 		b[i] = ((u_int8_t *)f->data)[b[i]];
 		i++;
diff --git a/sys/dev/sound/pcm/channel.c b/sys/dev/sound/pcm/channel.c
index bb46e03f8e88..38e1796bee97 100644
--- a/sys/dev/sound/pcm/channel.c
+++ b/sys/dev/sound/pcm/channel.c
@@ -268,7 +268,7 @@ chn_write(pcm_channel *c, struct uio *buf)
 	}
 	a = (1 << c->align) - 1;
 	c->flags |= CHN_F_WRITING;
-	while (buf->uio_resid > 0) {
+	while ((c->smegcnt + buf->uio_resid) > a) {
 		s = spltty();
 		chn_dmaupdate(c);
 		splx(s);
@@ -282,10 +282,9 @@ chn_write(pcm_channel *c, struct uio *buf)
 			continue;
 		}
 		/* ensure we always have a whole number of samples */
-		l = min(b->fl, b->bufsize - b->fp);
-		if (l & a) panic("unaligned write %d, %d", l, a + 1);
-		l &= ~a;
-		w = c->feeder->feed(c->feeder, b->buf + b->fp, l, buf);
+		l = min(b->fl, b->bufsize - b->fp) & ~a;
+		if (l == 0) break;
+		w = c->feeder->feed(c->feeder, c, b->buf + b->fp, l, buf);
 		if (w == 0) panic("no feed");
 		s = spltty();
 		b->rl += w;
@@ -294,6 +293,12 @@ chn_write(pcm_channel *c, struct uio *buf)
 	      	splx(s);
 		if (b->rl && !b->dl) chn_stintr(c);
 	}
+	if ((ret == 0) && (buf->uio_resid > 0)) {
+		l = buf->uio_resid;
+		if ((c->smegcnt + l) >= SMEGBUFSZ) panic("resid overflow %d", l);
+		uiomove(c->smegbuf + c->smegcnt, l, buf);
+		c->smegcnt += l;
+	}
 	c->flags &= ~CHN_F_WRITING;
 	return ret;
 }
@@ -418,7 +423,7 @@ chn_read(pcm_channel *c, struct uio *buf)
 		}
 		/* ensure we always have a whole number of samples */
 		l = min(b->rl, b->bufsize - b->rp) & DMA_ALIGN_MASK;
-		w = c->feeder->feed(c->feeder, b->buf + b->rp, l, buf);
+		w = c->feeder->feed(c->feeder, c, b->buf + b->rp, l, buf);
 		s = spltty();
 		b->rl -= w;
 		b->fl += w;
@@ -474,6 +479,7 @@ chn_resetbuf(pcm_channel *c)
 	u_int16_t data, *p;
 	u_int32_t i;
 
+	c->smegcnt = 0;
 	c->buffer.sample_size = 1;
 	c->buffer.sample_size <<= (c->hwfmt & AFMT_STEREO)? 1 : 0;
 	c->buffer.sample_size <<= (c->hwfmt & AFMT_16BIT)? 1 : 0;
diff --git a/sys/dev/sound/pcm/datatypes.h b/sys/dev/sound/pcm/datatypes.h
index 4a38679e45c5..a4af384a1c01 100644
--- a/sys/dev/sound/pcm/datatypes.h
+++ b/sys/dev/sound/pcm/datatypes.h
@@ -75,7 +75,8 @@ struct _snd_dbuf {
 
 typedef int (pcmfeed_init_t)(pcm_feeder *feeder);
 typedef int (pcmfeed_free_t)(pcm_feeder *feeder);
-typedef int (pcmfeed_feed_t)(pcm_feeder *feeder, u_int8_t *buffer, u_int32_t count, struct uio *stream);
+typedef int (pcmfeed_feed_t)(pcm_feeder *feeder, pcm_channel *c, u_int8_t *buffer,
+			     u_int32_t count, struct uio *stream);
 
 struct _pcm_feeder {
 	char name[16];
@@ -121,6 +122,9 @@ struct _pcm_channel {
 
 	int direction;
 	snd_dbuf buffer;
+#define	SMEGBUFSZ 4
+	u_int8_t smegbuf[SMEGBUFSZ];
+	u_int32_t smegcnt;
 	void *devinfo;
 };
 
diff --git a/sys/dev/sound/pcm/feeder.c b/sys/dev/sound/pcm/feeder.c
index 4a59806803c7..055dc98d65df 100644
--- a/sys/dev/sound/pcm/feeder.c
+++ b/sys/dev/sound/pcm/feeder.c
@@ -106,10 +106,17 @@ static unsigned char u8_to_ulaw[] = {
 /*****************************************************************************/
 
 static int
-feed_root(pcm_feeder *feeder, u_int8_t *buffer, u_int32_t count, struct uio *stream)
+feed_root(pcm_feeder *feeder, pcm_channel *ch, u_int8_t *buffer, u_int32_t count, struct uio *stream)
 {
 	int ret, tmp = 0, c = 0;
 	if (!count) panic("feed_root: count == 0");
+	count &= ~((1 << ch->align) - 1);
+	if (!count) panic("feed_root: aligned count == 0");
+	if (ch->smegcnt > 0) {
+		c = min(ch->smegcnt, count);
+		bcopy(ch->smegbuf, buffer, c);
+		ch->smegcnt -= c;
+	}
 	while ((stream->uio_resid > 0) && (c < count)) {
 		tmp = stream->uio_resid;
 		ret = uiomove(buffer + c, count - c, stream);
@@ -125,10 +132,10 @@ pcm_feeder feeder_root = { "root", 0, NULL, NULL, feed_root };
 /*****************************************************************************/
 
 static int
-feed_8to16(pcm_feeder *f, u_int8_t *b, u_int32_t count, struct uio *stream)
+feed_8to16(pcm_feeder *f, pcm_channel *c, u_int8_t *b, u_int32_t count, struct uio *stream)
 {
 	int i, j, k;
-	k = f->source->feed(f->source, b, count / 2, stream);
+	k = f->source->feed(f->source, c, b, count / 2, stream);
 	j = k - 1;
 	i = j * 2 + 1;
 	while (i > 0 && j >= 0) {
@@ -157,11 +164,11 @@ feed_16to8_free(pcm_feeder *f)
 }
 
 static int
-feed_16to8le(pcm_feeder *f, u_int8_t *b, u_int32_t count, struct uio *stream)
+feed_16to8le(pcm_feeder *f, pcm_channel *c, u_int8_t *b, u_int32_t count, struct uio *stream)
 {
 	u_int32_t i = 0, toget = count * 2;
 	int j = 1, k;
-	k = f->source->feed(f->source, f->data, min(toget, FEEDBUFSZ), stream);
+	k = f->source->feed(f->source, c, f->data, min(toget, FEEDBUFSZ), stream);
 	while (j < k) {
 		b[i++] = ((u_int8_t *)f->data)[j];
 		j += 2;
@@ -174,9 +181,9 @@ static pcm_feeder feeder_16to8le =
 /*****************************************************************************/
 
 static int
-feed_monotostereo8(pcm_feeder *f, u_int8_t *b, u_int32_t count, struct uio *stream)
+feed_monotostereo8(pcm_feeder *f, pcm_channel *c, u_int8_t *b, u_int32_t count, struct uio *stream)
 {
-	int i, j, k = f->source->feed(f->source, b, count / 2, stream);
+	int i, j, k = f->source->feed(f->source, c, b, count / 2, stream);
 	j = k - 1;
 	i = j * 2 + 1;
 	while (i > 0 && j >= 0) {
@@ -207,11 +214,11 @@ feed_stereotomono8_free(pcm_feeder *f)
 }
 
 static int
-feed_stereotomono8(pcm_feeder *f, u_int8_t *b, u_int32_t count, struct uio *stream)
+feed_stereotomono8(pcm_feeder *f, pcm_channel *c, u_int8_t *b, u_int32_t count, struct uio *stream)
 {
 	u_int32_t i = 0, toget = count * 2;
 	int j = 0, k;
-	k = f->source->feed(f->source, f->data, min(toget, FEEDBUFSZ), stream);
+	k = f->source->feed(f->source, c, f->data, min(toget, FEEDBUFSZ), stream);
 	while (j < k) {
 		b[i++] = ((u_int8_t *)f->data)[j];
 		j += 2;
@@ -225,10 +232,10 @@ static pcm_feeder feeder_stereotomono8 =
 /*****************************************************************************/
 
 static int
-feed_endian(pcm_feeder *f, u_int8_t *b, u_int32_t count, struct uio *stream)
+feed_endian(pcm_feeder *f, pcm_channel *c, u_int8_t *b, u_int32_t count, struct uio *stream)
 {
 	u_int8_t t;
-	int i = 0, j = f->source->feed(f->source, b, count, stream);
+	int i = 0, j = f->source->feed(f->source, c, b, count, stream);
 	while (i < j) {
 		t = b[i];
 		b[i] = b[i + 1];
@@ -242,9 +249,9 @@ static pcm_feeder feeder_endian = { "endian", -1, NULL, NULL, feed_endian };
 /*****************************************************************************/
 
 static int
-feed_sign(pcm_feeder *f, u_int8_t *b, u_int32_t count, struct uio *stream)
+feed_sign(pcm_feeder *f, pcm_channel *c, u_int8_t *b, u_int32_t count, struct uio *stream)
 {
-	int i = 0, j = f->source->feed(f->source, b, count, stream);
+	int i = 0, j = f->source->feed(f->source, c, b, count, stream);
 	int ssz = (int)f->data, ofs = ssz - 1;
 	while (i < j) {
 		b[i + ofs] ^= 0x80;
@@ -260,9 +267,9 @@ static pcm_feeder feeder_sign16 =
 /*****************************************************************************/
 
 static int
-feed_table(pcm_feeder *f, u_int8_t *b, u_int32_t count, struct uio *stream)
+feed_table(pcm_feeder *f, pcm_channel *c, u_int8_t *b, u_int32_t count, struct uio *stream)
 {
-	int i = 0, j = f->source->feed(f->source, b, count, stream);
+	int i = 0, j = f->source->feed(f->source, c, b, count, stream);
 	while (i < j) {
 		b[i] = ((u_int8_t *)f->data)[b[i]];
 		i++;