ALSA: hda - Fix playback noise with 24/32 bit sample size on BXT

In BXT-P A0, HD-Audio DMA requests is later than expected,
and makes an audio stream sensitive to system latencies when
24/32 bits are playing.
Adjusting threshold of DMA fifo to force the DMA request
sooner to improve latency tolerance at the expense of power.

v2: move Intel specific code to hda_intel.c

Signed-off-by: Lu, Han <han.lu@intel.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>

include/sound/hda_register.h

@@ -93,6 +93,9 @@ enum { SDI0, SDI1, SDI2, SDI3, SDO0, SDO1, SDO2, SDO3 };
 #define AZX_REG_HSW_EM4			0x100c
 #define AZX_REG_HSW_EM5			0x1010
 
+/* Skylake/Broxton display HD-A controller Extended Mode registers */
+#define AZX_REG_SKL_EM4L		0x1040
+
 /* PCI space */
 #define AZX_PCIREG_TCSEL		0x44
 

sound/pci/hda/hda_intel.c

@@ -355,6 +355,8 @@ enum {
 					((pci)->device == 0x0d0c) || \
 					((pci)->device == 0x160c))
 
+#define IS_BROXTON(pci)	((pci)->device == 0x5a98)
+
 static char *driver_short_names[] = {
 	[AZX_DRIVER_ICH] = "HDA Intel",
 	[AZX_DRIVER_PCH] = "HDA Intel PCH",
@@ -506,15 +508,36 @@ static void azx_init_pci(struct azx *chip)
         }
 }
 
+/*
+ * In BXT-P A0, HD-Audio DMA requests is later than expected,
+ * and makes an audio stream sensitive to system latencies when
+ * 24/32 bits are playing.
+ * Adjusting threshold of DMA fifo to force the DMA request
+ * sooner to improve latency tolerance at the expense of power.
+ */
+static void bxt_reduce_dma_latency(struct azx *chip)
+{
+	u32 val;
+
+	val = azx_readl(chip, SKL_EM4L);
+	val &= (0x3 << 20);
+	azx_writel(chip, SKL_EM4L, val);
+}
+
 static void hda_intel_init_chip(struct azx *chip, bool full_reset)
 {
 	struct hdac_bus *bus = azx_bus(chip);
+	struct pci_dev *pci = chip->pci;
 
 	if (chip->driver_caps & AZX_DCAPS_I915_POWERWELL)
 		snd_hdac_set_codec_wakeup(bus, true);
 	azx_init_chip(chip, full_reset);
 	if (chip->driver_caps & AZX_DCAPS_I915_POWERWELL)
 		snd_hdac_set_codec_wakeup(bus, false);
+
+	/* reduce dma latency to avoid noise */
+	if (IS_BROXTON(pci))
+		bxt_reduce_dma_latency(chip);
 }
 
 /* calculate runtime delay from LPIB */