diff --git a/sound/soc/intel/sst-dsp.h b/sound/soc/intel/sst-dsp.h
index 2753b85ac863..f291e32f0077 100644
--- a/sound/soc/intel/sst-dsp.h
+++ b/sound/soc/intel/sst-dsp.h
@@ -159,12 +159,18 @@
 #define SST_VDRTCTL3		0xaC
 
 /* VDRTCTL0 */
-#define SST_VDRTCL0_APLLSE_MASK		1
-#define SST_VDRTCL0_DSRAMPGE_SHIFT	16
-#define SST_VDRTCL0_DSRAMPGE_MASK	(0xffff << SST_VDRTCL0_DSRAMPGE_SHIFT)
-#define SST_VDRTCL0_ISRAMPGE_SHIFT	6
+#define SST_VDRTCL0_D3PGD		(1 << 0)
+#define SST_VDRTCL0_D3SRAMPGD		(1 << 1)
+#define SST_VDRTCL0_DSRAMPGE_SHIFT	12
+#define SST_VDRTCL0_DSRAMPGE_MASK	(0xfffff << SST_VDRTCL0_DSRAMPGE_SHIFT)
+#define SST_VDRTCL0_ISRAMPGE_SHIFT	2
 #define SST_VDRTCL0_ISRAMPGE_MASK	(0x3ff << SST_VDRTCL0_ISRAMPGE_SHIFT)
 
+/* VDRTCTL2 */
+#define SST_VDRTCL2_DCLCGE		(1 << 1)
+#define SST_VDRTCL2_DTCGE		(1 << 10)
+#define SST_VDRTCL2_APLLSE_MASK		(1 << 31)
+
 /* PMCS */
 #define SST_PMCS		0x84
 #define SST_PMCS_PS_MASK	0x3
diff --git a/sound/soc/intel/sst-haswell-dsp.c b/sound/soc/intel/sst-haswell-dsp.c
index 86aea343bfec..57039b00efc2 100644
--- a/sound/soc/intel/sst-haswell-dsp.c
+++ b/sound/soc/intel/sst-haswell-dsp.c
@@ -250,17 +250,42 @@ static irqreturn_t hsw_irq(int irq, void *context)
 static void hsw_set_dsp_D3(struct sst_dsp *sst)
 {
 	u32 val;
+	u32 reg;
 
-	/* switch off audio PLL, DRAM & IRAM blocks */
+	/* Disable core clock gating (VDRTCTL2.DCLCGE = 0) */
+	reg = readl(sst->addr.pci_cfg + SST_VDRTCTL2);
+	reg &= ~(SST_VDRTCL2_DCLCGE | SST_VDRTCL2_DTCGE);
+	writel(reg, sst->addr.pci_cfg + SST_VDRTCTL2);
+
+	/* enable power gating and switch off DRAM & IRAM blocks */
 	val = readl(sst->addr.pci_cfg + SST_VDRTCTL0);
-	val |= SST_VDRTCL0_APLLSE_MASK | SST_VDRTCL0_DSRAMPGE_MASK |
+	val |= SST_VDRTCL0_DSRAMPGE_MASK |
 		SST_VDRTCL0_ISRAMPGE_MASK;
+	val &= ~(SST_VDRTCL0_D3PGD | SST_VDRTCL0_D3SRAMPGD);
 	writel(val, sst->addr.pci_cfg + SST_VDRTCTL0);
 
-	/* Set D3 state */
+	/* switch off audio PLL */
+	val = readl(sst->addr.pci_cfg + SST_VDRTCTL2);
+	val |= SST_VDRTCL2_APLLSE_MASK;
+	writel(val, sst->addr.pci_cfg + SST_VDRTCTL2);
+
+	/* disable MCLK(clkctl.smos = 0) */
+	sst_dsp_shim_update_bits_unlocked(sst, SST_CLKCTL,
+		SST_CLKCTL_MASK, 0);
+
+	/* Set D3 state, delay 50 us */
 	val = readl(sst->addr.pci_cfg + SST_PMCS);
 	val |= SST_PMCS_PS_MASK;
 	writel(val, sst->addr.pci_cfg + SST_PMCS);
+	udelay(50);
+
+	/* Enable core clock gating (VDRTCTL2.DCLCGE = 1), delay 50 us */
+	reg = readl(sst->addr.pci_cfg + SST_VDRTCTL2);
+	reg |= SST_VDRTCL2_DCLCGE | SST_VDRTCL2_DTCGE;
+	writel(reg, sst->addr.pci_cfg + SST_VDRTCTL2);
+
+	udelay(50);
+
 }
 
 static void hsw_reset(struct sst_dsp *sst)
@@ -283,6 +308,16 @@ static int hsw_set_dsp_D0(struct sst_dsp *sst)
 	int tries = 10;
 	u32 reg;
 
+	/* Disable core clock gating (VDRTCTL2.DCLCGE = 0) */
+	reg = readl(sst->addr.pci_cfg + SST_VDRTCTL2);
+	reg &= ~(SST_VDRTCL2_DCLCGE | SST_VDRTCL2_DTCGE);
+	writel(reg, sst->addr.pci_cfg + SST_VDRTCTL2);
+
+	/* Disable D3PG (VDRTCTL0.D3PGD = 1) */
+	reg = readl(sst->addr.pci_cfg + SST_VDRTCTL0);
+	reg |= SST_VDRTCL0_D3PGD;
+	writel(reg, sst->addr.pci_cfg + SST_VDRTCTL0);
+
 	/* Set D0 state */
 	reg = readl(sst->addr.pci_cfg + SST_PMCS);
 	reg &= ~SST_PMCS_PS_MASK;
@@ -300,14 +335,6 @@ static int hsw_set_dsp_D0(struct sst_dsp *sst)
 	return -ENODEV;
 
 finish:
-	hsw_reset(sst);
-
-	/* switch on audio PLL, DRAM & IRAM blocks */
-	reg = readl(sst->addr.pci_cfg + SST_VDRTCTL0);
-	reg &= ~(SST_VDRTCL0_APLLSE_MASK | SST_VDRTCL0_DSRAMPGE_MASK |
-		SST_VDRTCL0_ISRAMPGE_MASK);
-	writel(reg, sst->addr.pci_cfg + SST_VDRTCTL0);
-
 	/* select SSP1 19.2MHz base clock, SSP clock 0, turn off Low Power Clock */
 	sst_dsp_shim_update_bits_unlocked(sst, SST_CSR,
 		SST_CSR_S1IOCS | SST_CSR_SBCS1 | SST_CSR_LPCS, 0x0);
@@ -322,6 +349,28 @@ finish:
 		SST_CLKCTL_MASK | SST_CLKCTL_DCPLCG | SST_CLKCTL_SCOE0,
 		SST_CLKCTL_MASK | SST_CLKCTL_DCPLCG | SST_CLKCTL_SCOE0);
 
+	/* Stall and reset core, set CSR */
+	hsw_reset(sst);
+
+	/* Enable core clock gating (VDRTCTL2.DCLCGE = 1), delay 50 us */
+	reg = readl(sst->addr.pci_cfg + SST_VDRTCTL2);
+	reg |= SST_VDRTCL2_DCLCGE | SST_VDRTCL2_DTCGE;
+	writel(reg, sst->addr.pci_cfg + SST_VDRTCTL2);
+
+	udelay(50);
+
+	/* switch on audio PLL */
+	reg = readl(sst->addr.pci_cfg + SST_VDRTCTL2);
+	reg &= ~SST_VDRTCL2_APLLSE_MASK;
+	writel(reg, sst->addr.pci_cfg + SST_VDRTCTL2);
+
+	/* set default power gating control, enable power gating control for all blocks. that is,
+	can't be accessed, please enable each block before accessing. */
+	reg = readl(sst->addr.pci_cfg + SST_VDRTCTL0);
+	reg |= SST_VDRTCL0_DSRAMPGE_MASK | SST_VDRTCL0_ISRAMPGE_MASK;
+	writel(reg, sst->addr.pci_cfg + SST_VDRTCTL0);
+
+
 	/* disable DMA finish function for SSP0 & SSP1 */
 	sst_dsp_shim_update_bits_unlocked(sst, SST_CSR2, SST_CSR2_SDFD_SSP1,
 		SST_CSR2_SDFD_SSP1);
@@ -343,9 +392,6 @@ finish:
 	sst_dsp_shim_write(sst, 0x80, 0x6);
 	sst_dsp_shim_write(sst, 0xe0, 0x300a);
 
-	/* disable all clock gating */
-	writel(0x0, sst->addr.pci_cfg + SST_VDRTCTL2);
-
 	return 0;
 }
 
@@ -497,6 +543,11 @@ static int hsw_block_enable(struct sst_mem_block *block)
 	dev_dbg(block->dsp->dev, " enabled block %d:%d at offset 0x%x\n",
 		block->type, block->index, block->offset);
 
+	/* Disable core clock gating (VDRTCTL2.DCLCGE = 0) */
+	val = readl(sst->addr.pci_cfg + SST_VDRTCTL2);
+	val &= ~SST_VDRTCL2_DCLCGE;
+	writel(val, sst->addr.pci_cfg + SST_VDRTCTL2);
+
 	val = readl(sst->addr.pci_cfg + SST_VDRTCTL0);
 	bit = hsw_block_get_bit(block);
 	writel(val & ~bit, sst->addr.pci_cfg + SST_VDRTCTL0);
@@ -504,6 +555,13 @@ static int hsw_block_enable(struct sst_mem_block *block)
 	/* wait 18 DSP clock ticks */
 	udelay(10);
 
+	/* Enable core clock gating (VDRTCTL2.DCLCGE = 1), delay 50 us */
+	val = readl(sst->addr.pci_cfg + SST_VDRTCTL2);
+	val |= SST_VDRTCL2_DCLCGE;
+	writel(val, sst->addr.pci_cfg + SST_VDRTCTL2);
+
+	udelay(50);
+
 	/*add a dummy read before the SRAM block is written, otherwise the writing may miss bytes sometimes.*/
 	sst_mem_block_dummy_read(block);
 	return 0;
@@ -521,10 +579,26 @@ static int hsw_block_disable(struct sst_mem_block *block)
 	dev_dbg(block->dsp->dev, " disabled block %d:%d at offset 0x%x\n",
 		block->type, block->index, block->offset);
 
+	/* Disable core clock gating (VDRTCTL2.DCLCGE = 0) */
+	val = readl(sst->addr.pci_cfg + SST_VDRTCTL2);
+	val &= ~SST_VDRTCL2_DCLCGE;
+	writel(val, sst->addr.pci_cfg + SST_VDRTCTL2);
+
+
 	val = readl(sst->addr.pci_cfg + SST_VDRTCTL0);
 	bit = hsw_block_get_bit(block);
 	writel(val | bit, sst->addr.pci_cfg + SST_VDRTCTL0);
 
+	/* wait 18 DSP clock ticks */
+	udelay(10);
+
+	/* Enable core clock gating (VDRTCTL2.DCLCGE = 1), delay 50 us */
+	val = readl(sst->addr.pci_cfg + SST_VDRTCTL2);
+	val |= SST_VDRTCL2_DCLCGE;
+	writel(val, sst->addr.pci_cfg + SST_VDRTCTL2);
+
+	udelay(50);
+
 	return 0;
 }