ALSA: emu10k1: fix sample signedness issues in wavetable loader

The hardware supports S16LE and U8 samples, while U16LE and S8 (which
the driver implicitly claims to support) require sign flipping.

Note that this matters only for the GUS patch loader, as the implemented
SoundFont v2.01 spec is limited to S16LE.
Signed-off-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
Message-ID: <20240406064830.1029573-10-oswald.buddenhagen@gmx.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>

include/sound/emu10k1.h

@@ -1882,8 +1882,8 @@ int snd_emu10k1_alloc_pages_maybe_wider(struct snd_emu10k1 *emu, size_t size,
 					struct snd_dma_buffer *dmab);
 struct snd_util_memblk *snd_emu10k1_synth_alloc(struct snd_emu10k1 *emu, unsigned int size);
 int snd_emu10k1_synth_free(struct snd_emu10k1 *emu, struct snd_util_memblk *blk);
-int snd_emu10k1_synth_bzero(struct snd_emu10k1 *emu, struct snd_util_memblk *blk, int offset, int size);
-int snd_emu10k1_synth_copy_from_user(struct snd_emu10k1 *emu, struct snd_util_memblk *blk, int offset, const char __user *data, int size);
+int snd_emu10k1_synth_memset(struct snd_emu10k1 *emu, struct snd_util_memblk *blk, int offset, int size, u8 value);
+int snd_emu10k1_synth_copy_from_user(struct snd_emu10k1 *emu, struct snd_util_memblk *blk, int offset, const char __user *data, int size, u32 xor);
 int snd_emu10k1_memblk_map(struct snd_emu10k1 *emu, struct snd_emu10k1_memblk *blk);
 
 /* voice allocation */

sound/pci/emu10k1/emu10k1_patch.c

@@ -26,6 +26,8 @@ snd_emu10k1_sample_new(struct snd_emux *rec, struct snd_sf_sample *sp,
 		       struct snd_util_memhdr *hdr,
 		       const void __user *data, long count)
 {
+	u8 fill;
+	u32 xor;
 	int offset;
 	int truesize, size, blocksize;
 	unsigned int start_addr;
@@ -41,6 +43,14 @@ snd_emu10k1_sample_new(struct snd_emux *rec, struct snd_sf_sample *sp,
 		       sp->v.sample);
 	}
 
+	if (sp->v.mode_flags & SNDRV_SFNT_SAMPLE_8BITS) {
+		fill = 0x80;
+		xor = (sp->v.mode_flags & SNDRV_SFNT_SAMPLE_UNSIGNED) ? 0 : 0x80808080;
+	} else {
+		fill = 0;
+		xor = (sp->v.mode_flags & SNDRV_SFNT_SAMPLE_UNSIGNED) ? 0x80008000 : 0;
+	}
+
 	/* compute true data size to be loaded */
 	truesize = sp->v.size + BLANK_HEAD_SIZE;
 	if (sp->v.mode_flags & SNDRV_SFNT_SAMPLE_NO_BLANK)
@@ -65,14 +75,14 @@ snd_emu10k1_sample_new(struct snd_emux *rec, struct snd_sf_sample *sp,
 	size = BLANK_HEAD_SIZE;
 	if (! (sp->v.mode_flags & SNDRV_SFNT_SAMPLE_8BITS))
 		size *= 2;
-	snd_emu10k1_synth_bzero(emu, sp->block, offset, size);
+	snd_emu10k1_synth_memset(emu, sp->block, offset, size, fill);
 	offset += size;
 
 	/* copy provided samples */
 	size = sp->v.size;
 	if (! (sp->v.mode_flags & SNDRV_SFNT_SAMPLE_8BITS))
 		size *= 2;
-	if (snd_emu10k1_synth_copy_from_user(emu, sp->block, offset, data, size)) {
+	if (snd_emu10k1_synth_copy_from_user(emu, sp->block, offset, data, size, xor)) {
 		snd_emu10k1_synth_free(emu, sp->block);
 		sp->block = NULL;
 		return -EFAULT;
@@ -81,7 +91,7 @@ snd_emu10k1_sample_new(struct snd_emux *rec, struct snd_sf_sample *sp,
 
 	/* clear rest of samples (if any) */
 	if (offset < blocksize)
-		snd_emu10k1_synth_bzero(emu, sp->block, offset, blocksize - offset);
+		snd_emu10k1_synth_memset(emu, sp->block, offset, blocksize - offset, fill);
 
 	if (sp->v.mode_flags & SNDRV_SFNT_SAMPLE_NO_BLANK) {
 		/* if no blank loop is attached in the sample, add it */
@@ -91,20 +101,6 @@ snd_emu10k1_sample_new(struct snd_emux *rec, struct snd_sf_sample *sp,
 		}
 	}
 
-#if 0 /* not supported yet */
-	if (sp->v.mode_flags & SNDRV_SFNT_SAMPLE_UNSIGNED) {
-		/* unsigned -> signed */
-		if (! (sp->v.mode_flags & SNDRV_SFNT_SAMPLE_8BITS)) {
-			unsigned short *wblock = (unsigned short*)block;
-			for (i = 0; i < truesize; i++)
-				wblock[i] ^= 0x8000;
-		} else {
-			for (i = 0; i < truesize; i++)
-				block[i] ^= 0x80;
-		}
-	}
-#endif
-
 	/* recalculate offset */
 	start_addr = BLANK_HEAD_SIZE * 2;
 	if (! (sp->v.mode_flags & SNDRV_SFNT_SAMPLE_8BITS))

sound/pci/emu10k1/memory.c

@@ -565,10 +565,10 @@ static inline void *offset_ptr(struct snd_emu10k1 *emu, int page, int offset)
 }
 
 /*
- * bzero(blk + offset, size)
+ * memset(blk + offset, value, size)
  */
-int snd_emu10k1_synth_bzero(struct snd_emu10k1 *emu, struct snd_util_memblk *blk,
-			    int offset, int size)
+int snd_emu10k1_synth_memset(struct snd_emu10k1 *emu, struct snd_util_memblk *blk,
+			     int offset, int size, u8 value)
 {
 	int page, nextofs, end_offset, temp, temp1;
 	void *ptr;
@@ -588,20 +588,47 @@ int snd_emu10k1_synth_bzero(struct snd_emu10k1 *emu, struct snd_util_memblk *blk
 			temp = temp1;
 		ptr = offset_ptr(emu, page + p->first_page, offset);
 		if (ptr)
-			memset(ptr, 0, temp);
+			memset(ptr, value, temp);
 		offset = nextofs;
 		page++;
 	} while (offset < end_offset);
 	return 0;
 }
 
-EXPORT_SYMBOL(snd_emu10k1_synth_bzero);
+EXPORT_SYMBOL(snd_emu10k1_synth_memset);
+
+// Note that the value is assumed to be suitably repetitive.
+static void xor_range(void *ptr, int size, u32 value)
+{
+	if ((long)ptr & 1) {
+		*(u8 *)ptr ^= (u8)value;
+		ptr++;
+		size--;
+	}
+	if (size > 1 && ((long)ptr & 2)) {
+		*(u16 *)ptr ^= (u16)value;
+		ptr += 2;
+		size -= 2;
+	}
+	while (size > 3) {
+		*(u32 *)ptr ^= value;
+		ptr += 4;
+		size -= 4;
+	}
+	if (size > 1) {
+		*(u16 *)ptr ^= (u16)value;
+		ptr += 2;
+		size -= 2;
+	}
+	if (size > 0)
+		*(u8 *)ptr ^= (u8)value;
+}
 
 /*
- * copy_from_user(blk + offset, data, size)
+ * copy_from_user(blk + offset, data, size) ^ xor
  */
 int snd_emu10k1_synth_copy_from_user(struct snd_emu10k1 *emu, struct snd_util_memblk *blk,
-				     int offset, const char __user *data, int size)
+				     int offset, const char __user *data, int size, u32 xor)
 {
 	int page, nextofs, end_offset, temp, temp1;
 	void *ptr;
@@ -620,8 +647,12 @@ int snd_emu10k1_synth_copy_from_user(struct snd_emu10k1 *emu, struct snd_util_me
 		if (temp1 < temp)
 			temp = temp1;
 		ptr = offset_ptr(emu, page + p->first_page, offset);
-		if (ptr && copy_from_user(ptr, data, temp))
-			return -EFAULT;
+		if (ptr) {
+			if (copy_from_user(ptr, data, temp))
+				return -EFAULT;
+			if (xor)
+				xor_range(ptr, temp, xor);
+		}
 		offset = nextofs;
 		data += temp;
 		page++;