drm/amdgpu: convert nbio to use callbacks (v2)

Cleans up and consolidates all of the per-asic logic.

v2: squash in "drm/amdgpu: fix NULL err for sriov detect" (Chunming)
Acked-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -1428,16 +1428,52 @@ typedef void (*amdgpu_block_wreg_t)(struct amdgpu_device*, uint32_t, uint32_t, u
 /*
  * amdgpu nbio functions
  *
- * Fix me :
- * 	Put more NBIO specifc func wraper here , for now just try to minimize the
- *	change to avoid use SOC15_REG_OFFSET in the constant array
  */
+struct nbio_hdp_flush_reg {
+	u32 ref_and_mask_cp0;
+	u32 ref_and_mask_cp1;
+	u32 ref_and_mask_cp2;
+	u32 ref_and_mask_cp3;
+	u32 ref_and_mask_cp4;
+	u32 ref_and_mask_cp5;
+	u32 ref_and_mask_cp6;
+	u32 ref_and_mask_cp7;
+	u32 ref_and_mask_cp8;
+	u32 ref_and_mask_cp9;
+	u32 ref_and_mask_sdma0;
+	u32 ref_and_mask_sdma1;
+};
 
 struct amdgpu_nbio_funcs {
-	u32 (*get_hdp_flush_req_offset)(struct amdgpu_device*);
-	u32 (*get_hdp_flush_done_offset)(struct amdgpu_device*);
-	u32 (*get_pcie_index_offset)(struct amdgpu_device*);
-	u32 (*get_pcie_data_offset)(struct amdgpu_device*);
+	const struct nbio_hdp_flush_reg *hdp_flush_reg;
+	u32 (*get_hdp_flush_req_offset)(struct amdgpu_device *adev);
+	u32 (*get_hdp_flush_done_offset)(struct amdgpu_device *adev);
+	u32 (*get_pcie_index_offset)(struct amdgpu_device *adev);
+	u32 (*get_pcie_data_offset)(struct amdgpu_device *adev);
+	u32 (*get_rev_id)(struct amdgpu_device *adev);
+	u32 (*get_atombios_scratch_regs)(struct amdgpu_device *adev, uint32_t idx);
+	void (*set_atombios_scratch_regs)(struct amdgpu_device *adev,
+					  uint32_t idx, uint32_t val);
+	void (*mc_access_enable)(struct amdgpu_device *adev, bool enable);
+	void (*hdp_flush)(struct amdgpu_device *adev);
+	u32 (*get_memsize)(struct amdgpu_device *adev);
+	void (*sdma_doorbell_range)(struct amdgpu_device *adev, int instance,
+				    bool use_doorbell, int doorbell_index);
+	void (*enable_doorbell_aperture)(struct amdgpu_device *adev,
+					 bool enable);
+	void (*enable_doorbell_selfring_aperture)(struct amdgpu_device *adev,
+						  bool enable);
+	void (*ih_doorbell_range)(struct amdgpu_device *adev,
+				  bool use_doorbell, int doorbell_index);
+	void (*update_medium_grain_clock_gating)(struct amdgpu_device *adev,
+						 bool enable);
+	void (*update_medium_grain_light_sleep)(struct amdgpu_device *adev,
+						bool enable);
+	void (*get_clockgating_state)(struct amdgpu_device *adev,
+				      u32 *flags);
+	void (*ih_control)(struct amdgpu_device *adev);
+	void (*init_registers)(struct amdgpu_device *adev);
+	void (*detect_hw_virt)(struct amdgpu_device *adev);
 };
 
 

drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c

@@ -3552,12 +3552,7 @@ static void gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 {
 	struct amdgpu_device *adev = ring->adev;
 	u32 ref_and_mask, reg_mem_engine;
-	const struct nbio_hdp_flush_reg *nbio_hf_reg;
-
-	if (ring->adev->flags & AMD_IS_APU)
-		nbio_hf_reg = &nbio_v7_0_hdp_flush_reg;
-	else
-		nbio_hf_reg = &nbio_v6_1_hdp_flush_reg;
+	const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio_funcs->hdp_flush_reg;
 
 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
 		switch (ring->me) {

drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c

@@ -38,8 +38,6 @@
 #include "soc15_common.h"
 #include "umc/umc_6_0_sh_mask.h"
 
-#include "nbio_v6_1.h"
-#include "nbio_v7_0.h"
 #include "gfxhub_v1_0.h"
 #include "mmhub_v1_0.h"
 
@@ -332,10 +330,7 @@ static void gmc_v9_0_gart_flush_gpu_tlb(struct amdgpu_device *adev,
 	unsigned i, j;
 
 	/* flush hdp cache */
-	if (adev->flags & AMD_IS_APU)
-		nbio_v7_0_hdp_flush(adev);
-	else
-		nbio_v6_1_hdp_flush(adev);
+	adev->nbio_funcs->hdp_flush(adev);
 
 	spin_lock(&adev->mc.invalidate_lock);
 
@@ -702,8 +697,7 @@ static int gmc_v9_0_mc_init(struct amdgpu_device *adev)
 
 	/* size in MB on si */
 	adev->mc.mc_vram_size =
-		((adev->flags & AMD_IS_APU) ? nbio_v7_0_get_memsize(adev) :
-		 nbio_v6_1_get_memsize(adev)) * 1024ULL * 1024ULL;
+		adev->nbio_funcs->get_memsize(adev) * 1024ULL * 1024ULL;
 	adev->mc.real_vram_size = adev->mc.mc_vram_size;
 
 	if (!(adev->flags & AMD_IS_APU)) {
@@ -951,10 +945,7 @@ static int gmc_v9_0_gart_enable(struct amdgpu_device *adev)
 	WREG32_SOC15(HDP, 0, mmHDP_HOST_PATH_CNTL, tmp);
 
 	/* After HDP is initialized, flush HDP.*/
-	if (adev->flags & AMD_IS_APU)
-		nbio_v7_0_hdp_flush(adev);
-	else
-		nbio_v6_1_hdp_flush(adev);
+	adev->nbio_funcs->hdp_flush(adev);
 
 	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
 		value = false;

drivers/gpu/drm/amd/amdgpu/nbio_v6_1.c

@@ -33,7 +33,7 @@
 #define smnPCIE_CNTL2                                                                                   0x11180070
 #define smnPCIE_CONFIG_CNTL                                                                             0x11180044
 
-u32 nbio_v6_1_get_rev_id(struct amdgpu_device *adev)
+static u32 nbio_v6_1_get_rev_id(struct amdgpu_device *adev)
 {
         u32 tmp = RREG32_SOC15(NBIO, 0, mmRCC_DEV0_EPF0_STRAP0);
 
@@ -43,19 +43,19 @@ u32 nbio_v6_1_get_rev_id(struct amdgpu_device *adev)
 	return tmp;
 }
 
-u32 nbio_v6_1_get_atombios_scratch_regs(struct amdgpu_device *adev,
-					uint32_t idx)
+static u32 nbio_v6_1_get_atombios_scratch_regs(struct amdgpu_device *adev,
+					       uint32_t idx)
 {
 	return RREG32_SOC15_OFFSET(NBIO, 0, mmBIOS_SCRATCH_0, idx);
 }
 
-void nbio_v6_1_set_atombios_scratch_regs(struct amdgpu_device *adev,
-					 uint32_t idx, uint32_t val)
+static void nbio_v6_1_set_atombios_scratch_regs(struct amdgpu_device *adev,
+						uint32_t idx, uint32_t val)
 {
 	WREG32_SOC15_OFFSET(NBIO, 0, mmBIOS_SCRATCH_0, idx, val);
 }
 
-void nbio_v6_1_mc_access_enable(struct amdgpu_device *adev, bool enable)
+static void nbio_v6_1_mc_access_enable(struct amdgpu_device *adev, bool enable)
 {
 	if (enable)
 		WREG32_SOC15(NBIO, 0, mmBIF_FB_EN,
@@ -65,17 +65,17 @@ void nbio_v6_1_mc_access_enable(struct amdgpu_device *adev, bool enable)
 		WREG32_SOC15(NBIO, 0, mmBIF_FB_EN, 0);
 }
 
-void nbio_v6_1_hdp_flush(struct amdgpu_device *adev)
+static void nbio_v6_1_hdp_flush(struct amdgpu_device *adev)
 {
 	WREG32_SOC15_NO_KIQ(NBIO, 0, mmBIF_BX_PF0_HDP_MEM_COHERENCY_FLUSH_CNTL, 0);
 }
 
-u32 nbio_v6_1_get_memsize(struct amdgpu_device *adev)
+static u32 nbio_v6_1_get_memsize(struct amdgpu_device *adev)
 {
 	return RREG32_SOC15(NBIO, 0, mmRCC_PF_0_0_RCC_CONFIG_MEMSIZE);
 }
 
-void nbio_v6_1_sdma_doorbell_range(struct amdgpu_device *adev, int instance,
+static void nbio_v6_1_sdma_doorbell_range(struct amdgpu_device *adev, int instance,
 				  bool use_doorbell, int doorbell_index)
 {
 	u32 reg = instance == 0 ? SOC15_REG_OFFSET(NBIO, 0, mmBIF_SDMA0_DOORBELL_RANGE) :
@@ -93,14 +93,14 @@ void nbio_v6_1_sdma_doorbell_range(struct amdgpu_device *adev, int instance,
 
 }
 
-void nbio_v6_1_enable_doorbell_aperture(struct amdgpu_device *adev,
-					bool enable)
+static void nbio_v6_1_enable_doorbell_aperture(struct amdgpu_device *adev,
+					       bool enable)
 {
 	WREG32_FIELD15(NBIO, 0, RCC_PF_0_0_RCC_DOORBELL_APER_EN, BIF_DOORBELL_APER_EN, enable ? 1 : 0);
 }
 
-void nbio_v6_1_enable_doorbell_selfring_aperture(struct amdgpu_device *adev,
-					bool enable)
+static void nbio_v6_1_enable_doorbell_selfring_aperture(struct amdgpu_device *adev,
+							bool enable)
 {
 	u32 tmp = 0;
 
@@ -119,8 +119,8 @@ void nbio_v6_1_enable_doorbell_selfring_aperture(struct amdgpu_device *adev,
 }
 
 
-void nbio_v6_1_ih_doorbell_range(struct amdgpu_device *adev,
-				bool use_doorbell, int doorbell_index)
+static void nbio_v6_1_ih_doorbell_range(struct amdgpu_device *adev,
+					bool use_doorbell, int doorbell_index)
 {
 	u32 ih_doorbell_range = RREG32_SOC15(NBIO, 0 , mmBIF_IH_DOORBELL_RANGE);
 
@@ -133,7 +133,7 @@ void nbio_v6_1_ih_doorbell_range(struct amdgpu_device *adev,
 	WREG32_SOC15(NBIO, 0, mmBIF_IH_DOORBELL_RANGE, ih_doorbell_range);
 }
 
-void nbio_v6_1_ih_control(struct amdgpu_device *adev)
+static void nbio_v6_1_ih_control(struct amdgpu_device *adev)
 {
 	u32 interrupt_cntl;
 
@@ -149,8 +149,8 @@ void nbio_v6_1_ih_control(struct amdgpu_device *adev)
 	WREG32_SOC15(NBIO, 0, mmINTERRUPT_CNTL, interrupt_cntl);
 }
 
-void nbio_v6_1_update_medium_grain_clock_gating(struct amdgpu_device *adev,
-						bool enable)
+static void nbio_v6_1_update_medium_grain_clock_gating(struct amdgpu_device *adev,
+						       bool enable)
 {
 	uint32_t def, data;
 
@@ -177,8 +177,8 @@ void nbio_v6_1_update_medium_grain_clock_gating(struct amdgpu_device *adev,
 		WREG32_PCIE(smnCPM_CONTROL, data);
 }
 
-void nbio_v6_1_update_medium_grain_light_sleep(struct amdgpu_device *adev,
-					       bool enable)
+static void nbio_v6_1_update_medium_grain_light_sleep(struct amdgpu_device *adev,
+						      bool enable)
 {
 	uint32_t def, data;
 
@@ -197,7 +197,8 @@ void nbio_v6_1_update_medium_grain_light_sleep(struct amdgpu_device *adev,
 		WREG32_PCIE(smnPCIE_CNTL2, data);
 }
 
-void nbio_v6_1_get_clockgating_state(struct amdgpu_device *adev, u32 *flags)
+static void nbio_v6_1_get_clockgating_state(struct amdgpu_device *adev,
+					    u32 *flags)
 {
 	int data;
 
@@ -232,7 +233,7 @@ static u32 nbio_v6_1_get_pcie_data_offset(struct amdgpu_device *adev)
 	return SOC15_REG_OFFSET(NBIO, 0, mmPCIE_DATA);
 }
 
-const struct nbio_hdp_flush_reg nbio_v6_1_hdp_flush_reg = {
+static const struct nbio_hdp_flush_reg nbio_v6_1_hdp_flush_reg = {
 	.ref_and_mask_cp0 = BIF_BX_PF0_GPU_HDP_FLUSH_DONE__CP0_MASK,
 	.ref_and_mask_cp1 = BIF_BX_PF0_GPU_HDP_FLUSH_DONE__CP1_MASK,
 	.ref_and_mask_cp2 = BIF_BX_PF0_GPU_HDP_FLUSH_DONE__CP2_MASK,
@@ -247,15 +248,7 @@ const struct nbio_hdp_flush_reg nbio_v6_1_hdp_flush_reg = {
 	.ref_and_mask_sdma1 = BIF_BX_PF0_GPU_HDP_FLUSH_DONE__SDMA1_MASK
 };
 
-const struct amdgpu_nbio_funcs nbio_v6_1_funcs = {
-	.get_hdp_flush_req_offset = nbio_v6_1_get_hdp_flush_req_offset,
-	.get_hdp_flush_done_offset = nbio_v6_1_get_hdp_flush_done_offset,
-	.get_pcie_index_offset = nbio_v6_1_get_pcie_index_offset,
-	.get_pcie_data_offset = nbio_v6_1_get_pcie_data_offset,
-};
-
-
-void nbio_v6_1_detect_hw_virt(struct amdgpu_device *adev)
+static void nbio_v6_1_detect_hw_virt(struct amdgpu_device *adev)
 {
 	uint32_t reg;
 
@@ -272,7 +265,7 @@ void nbio_v6_1_detect_hw_virt(struct amdgpu_device *adev)
 	}
 }
 
-void nbio_v6_1_init_registers(struct amdgpu_device *adev)
+static void nbio_v6_1_init_registers(struct amdgpu_device *adev)
 {
 	uint32_t def, data;
 
@@ -283,3 +276,27 @@ void nbio_v6_1_init_registers(struct amdgpu_device *adev)
 	if (def != data)
 		WREG32_PCIE(smnPCIE_CONFIG_CNTL, data);
 }
+
+const struct amdgpu_nbio_funcs nbio_v6_1_funcs = {
+	.hdp_flush_reg = &nbio_v6_1_hdp_flush_reg,
+	.get_hdp_flush_req_offset = nbio_v6_1_get_hdp_flush_req_offset,
+	.get_hdp_flush_done_offset = nbio_v6_1_get_hdp_flush_done_offset,
+	.get_pcie_index_offset = nbio_v6_1_get_pcie_index_offset,
+	.get_pcie_data_offset = nbio_v6_1_get_pcie_data_offset,
+	.get_rev_id = nbio_v6_1_get_rev_id,
+	.get_atombios_scratch_regs = nbio_v6_1_get_atombios_scratch_regs,
+	.set_atombios_scratch_regs = nbio_v6_1_set_atombios_scratch_regs,
+	.mc_access_enable = nbio_v6_1_mc_access_enable,
+	.hdp_flush = nbio_v6_1_hdp_flush,
+	.get_memsize = nbio_v6_1_get_memsize,
+	.sdma_doorbell_range = nbio_v6_1_sdma_doorbell_range,
+	.enable_doorbell_aperture = nbio_v6_1_enable_doorbell_aperture,
+	.enable_doorbell_selfring_aperture = nbio_v6_1_enable_doorbell_selfring_aperture,
+	.ih_doorbell_range = nbio_v6_1_ih_doorbell_range,
+	.update_medium_grain_clock_gating = nbio_v6_1_update_medium_grain_clock_gating,
+	.update_medium_grain_light_sleep = nbio_v6_1_update_medium_grain_light_sleep,
+	.get_clockgating_state = nbio_v6_1_get_clockgating_state,
+	.ih_control = nbio_v6_1_ih_control,
+	.init_registers = nbio_v6_1_init_registers,
+	.detect_hw_virt = nbio_v6_1_detect_hw_virt,
+};

drivers/gpu/drm/amd/amdgpu/nbio_v6_1.h

@@ -26,31 +26,6 @@
 
 #include "soc15_common.h"
 
-extern const struct nbio_hdp_flush_reg nbio_v6_1_hdp_flush_reg;
 extern const struct amdgpu_nbio_funcs nbio_v6_1_funcs;
 
-int nbio_v6_1_init(struct amdgpu_device *adev);
-u32 nbio_v6_1_get_atombios_scratch_regs(struct amdgpu_device *adev,
-                                        uint32_t idx);
-void nbio_v6_1_set_atombios_scratch_regs(struct amdgpu_device *adev,
-                                         uint32_t idx, uint32_t val);
-void nbio_v6_1_mc_access_enable(struct amdgpu_device *adev, bool enable);
-void nbio_v6_1_hdp_flush(struct amdgpu_device *adev);
-u32 nbio_v6_1_get_memsize(struct amdgpu_device *adev);
-void nbio_v6_1_sdma_doorbell_range(struct amdgpu_device *adev, int instance,
-				  bool use_doorbell, int doorbell_index);
-void nbio_v6_1_enable_doorbell_aperture(struct amdgpu_device *adev,
-					bool enable);
-void nbio_v6_1_enable_doorbell_selfring_aperture(struct amdgpu_device *adev,
-					bool enable);
-void nbio_v6_1_ih_doorbell_range(struct amdgpu_device *adev,
-				bool use_doorbell, int doorbell_index);
-void nbio_v6_1_ih_control(struct amdgpu_device *adev);
-u32 nbio_v6_1_get_rev_id(struct amdgpu_device *adev);
-void nbio_v6_1_update_medium_grain_clock_gating(struct amdgpu_device *adev, bool enable);
-void nbio_v6_1_update_medium_grain_light_sleep(struct amdgpu_device *adev, bool enable);
-void nbio_v6_1_get_clockgating_state(struct amdgpu_device *adev, u32 *flags);
-void nbio_v6_1_detect_hw_virt(struct amdgpu_device *adev);
-void nbio_v6_1_init_registers(struct amdgpu_device *adev);
-
 #endif

drivers/gpu/drm/amd/amdgpu/nbio_v7_0.c

@@ -31,8 +31,10 @@
 
 #define smnNBIF_MGCG_CTRL_LCLK	0x1013a05c
 
+#define smnCPM_CONTROL                                                                                  0x11180460
+#define smnPCIE_CNTL2                                                                                   0x11180070
 
-u32 nbio_v7_0_get_rev_id(struct amdgpu_device *adev)
+static u32 nbio_v7_0_get_rev_id(struct amdgpu_device *adev)
 {
         u32 tmp = RREG32_SOC15(NBIO, 0, mmRCC_DEV0_EPF0_STRAP0);
 
@@ -42,19 +44,19 @@ u32 nbio_v7_0_get_rev_id(struct amdgpu_device *adev)
 	return tmp;
 }
 
-u32 nbio_v7_0_get_atombios_scratch_regs(struct amdgpu_device *adev,
+static u32 nbio_v7_0_get_atombios_scratch_regs(struct amdgpu_device *adev,
 					uint32_t idx)
 {
 	return RREG32_SOC15_OFFSET(NBIO, 0, mmBIOS_SCRATCH_0, idx);
 }
 
-void nbio_v7_0_set_atombios_scratch_regs(struct amdgpu_device *adev,
-					 uint32_t idx, uint32_t val)
+static void nbio_v7_0_set_atombios_scratch_regs(struct amdgpu_device *adev,
+						uint32_t idx, uint32_t val)
 {
 	WREG32_SOC15_OFFSET(NBIO, 0, mmBIOS_SCRATCH_0, idx, val);
 }
 
-void nbio_v7_0_mc_access_enable(struct amdgpu_device *adev, bool enable)
+static void nbio_v7_0_mc_access_enable(struct amdgpu_device *adev, bool enable)
 {
 	if (enable)
 		WREG32_SOC15(NBIO, 0, mmBIF_FB_EN,
@@ -63,18 +65,18 @@ void nbio_v7_0_mc_access_enable(struct amdgpu_device *adev, bool enable)
 		WREG32_SOC15(NBIO, 0, mmBIF_FB_EN, 0);
 }
 
-void nbio_v7_0_hdp_flush(struct amdgpu_device *adev)
+static void nbio_v7_0_hdp_flush(struct amdgpu_device *adev)
 {
 	WREG32_SOC15_NO_KIQ(NBIO, 0, mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0);
 }
 
-u32 nbio_v7_0_get_memsize(struct amdgpu_device *adev)
+static u32 nbio_v7_0_get_memsize(struct amdgpu_device *adev)
 {
 	return RREG32_SOC15(NBIO, 0, mmRCC_CONFIG_MEMSIZE);
 }
 
-void nbio_v7_0_sdma_doorbell_range(struct amdgpu_device *adev, int instance,
-				  bool use_doorbell, int doorbell_index)
+static void nbio_v7_0_sdma_doorbell_range(struct amdgpu_device *adev, int instance,
+					  bool use_doorbell, int doorbell_index)
 {
 	u32 reg = instance == 0 ? SOC15_REG_OFFSET(NBIO, 0, mmBIF_SDMA0_DOORBELL_RANGE) :
 			SOC15_REG_OFFSET(NBIO, 0, mmBIF_SDMA1_DOORBELL_RANGE);
@@ -90,14 +92,20 @@ void nbio_v7_0_sdma_doorbell_range(struct amdgpu_device *adev, int instance,
 	WREG32(reg, doorbell_range);
 }
 
-void nbio_v7_0_enable_doorbell_aperture(struct amdgpu_device *adev,
-					bool enable)
+static void nbio_v7_0_enable_doorbell_aperture(struct amdgpu_device *adev,
+					       bool enable)
 {
 	WREG32_FIELD15(NBIO, 0, RCC_DOORBELL_APER_EN, BIF_DOORBELL_APER_EN, enable ? 1 : 0);
 }
 
-void nbio_v7_0_ih_doorbell_range(struct amdgpu_device *adev,
-				bool use_doorbell, int doorbell_index)
+static void nbio_v7_0_enable_doorbell_selfring_aperture(struct amdgpu_device *adev,
+							bool enable)
+{
+
+}
+
+static void nbio_v7_0_ih_doorbell_range(struct amdgpu_device *adev,
+					bool use_doorbell, int doorbell_index)
 {
 	u32 ih_doorbell_range = RREG32_SOC15(NBIO, 0 , mmBIF_IH_DOORBELL_RANGE);
 
@@ -127,8 +135,8 @@ static void nbio_7_0_write_syshub_ind_mmr(struct amdgpu_device *adev, uint32_t o
 	WREG32_SOC15(NBIO, 0, mmSYSHUB_DATA, data);
 }
 
-void nbio_v7_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
-						bool enable)
+static void nbio_v7_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
+						       bool enable)
 {
 	uint32_t def, data;
 
@@ -166,7 +174,43 @@ void nbio_v7_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
 		nbio_7_0_write_syshub_ind_mmr(adev, ixSYSHUB_MMREG_IND_SYSHUB_MGCG_CTRL_SHUBCLK, data);
 }
 
-void nbio_v7_0_ih_control(struct amdgpu_device *adev)
+static void nbio_v7_0_update_medium_grain_light_sleep(struct amdgpu_device *adev,
+						      bool enable)
+{
+	uint32_t def, data;
+
+	def = data = RREG32_PCIE(smnPCIE_CNTL2);
+	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_BIF_LS)) {
+		data |= (PCIE_CNTL2__SLV_MEM_LS_EN_MASK |
+			 PCIE_CNTL2__MST_MEM_LS_EN_MASK |
+			 PCIE_CNTL2__REPLAY_MEM_LS_EN_MASK);
+	} else {
+		data &= ~(PCIE_CNTL2__SLV_MEM_LS_EN_MASK |
+			  PCIE_CNTL2__MST_MEM_LS_EN_MASK |
+			  PCIE_CNTL2__REPLAY_MEM_LS_EN_MASK);
+	}
+
+	if (def != data)
+		WREG32_PCIE(smnPCIE_CNTL2, data);
+}
+
+static void nbio_v7_0_get_clockgating_state(struct amdgpu_device *adev,
+					    u32 *flags)
+{
+	int data;
+
+	/* AMD_CG_SUPPORT_BIF_MGCG */
+	data = RREG32_PCIE(smnCPM_CONTROL);
+	if (data & CPM_CONTROL__LCLK_DYN_GATE_ENABLE_MASK)
+		*flags |= AMD_CG_SUPPORT_BIF_MGCG;
+
+	/* AMD_CG_SUPPORT_BIF_LS */
+	data = RREG32_PCIE(smnPCIE_CNTL2);
+	if (data & PCIE_CNTL2__SLV_MEM_LS_EN_MASK)
+		*flags |= AMD_CG_SUPPORT_BIF_LS;
+}
+
+static void nbio_v7_0_ih_control(struct amdgpu_device *adev)
 {
 	u32 interrupt_cntl;
 
@@ -217,10 +261,37 @@ const struct nbio_hdp_flush_reg nbio_v7_0_hdp_flush_reg = {
 	.ref_and_mask_sdma1 = GPU_HDP_FLUSH_DONE__SDMA1_MASK,
 };
 
+static void nbio_v7_0_detect_hw_virt(struct amdgpu_device *adev)
+{
+	if (is_virtual_machine())	/* passthrough mode exclus sriov mod */
+		adev->virt.caps |= AMDGPU_PASSTHROUGH_MODE;
+}
+
+static void nbio_v7_0_init_registers(struct amdgpu_device *adev)
+{
+
+}
+
 const struct amdgpu_nbio_funcs nbio_v7_0_funcs = {
+	.hdp_flush_reg = &nbio_v7_0_hdp_flush_reg,
 	.get_hdp_flush_req_offset = nbio_v7_0_get_hdp_flush_req_offset,
 	.get_hdp_flush_done_offset = nbio_v7_0_get_hdp_flush_done_offset,
 	.get_pcie_index_offset = nbio_v7_0_get_pcie_index_offset,
 	.get_pcie_data_offset = nbio_v7_0_get_pcie_data_offset,
+	.get_rev_id = nbio_v7_0_get_rev_id,
+	.get_atombios_scratch_regs = nbio_v7_0_get_atombios_scratch_regs,
+	.set_atombios_scratch_regs = nbio_v7_0_set_atombios_scratch_regs,
+	.mc_access_enable = nbio_v7_0_mc_access_enable,
+	.hdp_flush = nbio_v7_0_hdp_flush,
+	.get_memsize = nbio_v7_0_get_memsize,
+	.sdma_doorbell_range = nbio_v7_0_sdma_doorbell_range,
+	.enable_doorbell_aperture = nbio_v7_0_enable_doorbell_aperture,
+	.enable_doorbell_selfring_aperture = nbio_v7_0_enable_doorbell_selfring_aperture,
+	.ih_doorbell_range = nbio_v7_0_ih_doorbell_range,
+	.update_medium_grain_clock_gating = nbio_v7_0_update_medium_grain_clock_gating,
+	.update_medium_grain_light_sleep = nbio_v7_0_update_medium_grain_light_sleep,
+	.get_clockgating_state = nbio_v7_0_get_clockgating_state,
+	.ih_control = nbio_v7_0_ih_control,
+	.init_registers = nbio_v7_0_init_registers,
+	.detect_hw_virt = nbio_v7_0_detect_hw_virt,
 };
-

drivers/gpu/drm/amd/amdgpu/nbio_v7_0.h

@@ -26,25 +26,6 @@
 
 #include "soc15_common.h"
 
-extern const struct nbio_hdp_flush_reg nbio_v7_0_hdp_flush_reg;
 extern const struct amdgpu_nbio_funcs nbio_v7_0_funcs;
 
-int nbio_v7_0_init(struct amdgpu_device *adev);
-u32 nbio_v7_0_get_atombios_scratch_regs(struct amdgpu_device *adev,
-                                        uint32_t idx);
-void nbio_v7_0_set_atombios_scratch_regs(struct amdgpu_device *adev,
-                                         uint32_t idx, uint32_t val);
-void nbio_v7_0_mc_access_enable(struct amdgpu_device *adev, bool enable);
-void nbio_v7_0_hdp_flush(struct amdgpu_device *adev);
-u32 nbio_v7_0_get_memsize(struct amdgpu_device *adev);
-void nbio_v7_0_sdma_doorbell_range(struct amdgpu_device *adev, int instance,
-				  bool use_doorbell, int doorbell_index);
-void nbio_v7_0_enable_doorbell_aperture(struct amdgpu_device *adev,
-					bool enable);
-void nbio_v7_0_ih_doorbell_range(struct amdgpu_device *adev,
-				bool use_doorbell, int doorbell_index);
-void nbio_v7_0_ih_control(struct amdgpu_device *adev);
-u32 nbio_v7_0_get_rev_id(struct amdgpu_device *adev);
-void nbio_v7_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
-						bool enable);
 #endif

drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c

@@ -359,12 +359,7 @@ static void sdma_v4_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 {
 	struct amdgpu_device *adev = ring->adev;
 	u32 ref_and_mask = 0;
-	const struct nbio_hdp_flush_reg *nbio_hf_reg;
-
-	if (ring->adev->flags & AMD_IS_APU)
-		nbio_hf_reg = &nbio_v7_0_hdp_flush_reg;
-	else
-		nbio_hf_reg = &nbio_v6_1_hdp_flush_reg;
+	const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio_funcs->hdp_flush_reg;
 
 	if (ring == &ring->adev->sdma.instance[0].ring)
 		ref_and_mask = nbio_hf_reg->ref_and_mask_sdma0;
@@ -629,10 +624,8 @@ static int sdma_v4_0_gfx_resume(struct amdgpu_device *adev)
 		}
 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_DOORBELL), doorbell);
 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_DOORBELL_OFFSET), doorbell_offset);
-		if (adev->flags & AMD_IS_APU)
-			nbio_v7_0_sdma_doorbell_range(adev, i, ring->use_doorbell, ring->doorbell_index);
-		else
-			nbio_v6_1_sdma_doorbell_range(adev, i, ring->use_doorbell, ring->doorbell_index);
+		adev->nbio_funcs->sdma_doorbell_range(adev, i, ring->use_doorbell,
+						      ring->doorbell_index);
 
 		if (amdgpu_sriov_vf(adev))
 			sdma_v4_0_ring_set_wptr(ring);

drivers/gpu/drm/amd/amdgpu/soc15.c

@@ -228,10 +228,7 @@ static void soc15_se_cac_wreg(struct amdgpu_device *adev, u32 reg, u32 v)
 
 static u32 soc15_get_config_memsize(struct amdgpu_device *adev)
 {
-	if (adev->flags & AMD_IS_APU)
-		return nbio_v7_0_get_memsize(adev);
-	else
-		return nbio_v6_1_get_memsize(adev);
+	return adev->nbio_funcs->get_memsize(adev);
 }
 
 static const u32 vega10_golden_init[] =
@@ -460,9 +457,8 @@ static int soc15_asic_reset(struct amdgpu_device *adev)
 
 	/* wait for asic to come out of reset */
 	for (i = 0; i < adev->usec_timeout; i++) {
-		u32 memsize = (adev->flags & AMD_IS_APU) ?
-			nbio_v7_0_get_memsize(adev) :
-			nbio_v6_1_get_memsize(adev);
+		u32 memsize = adev->nbio_funcs->get_memsize(adev);
+
 		if (memsize != 0xffffffff)
 			break;
 		udelay(1);
@@ -527,14 +523,10 @@ static void soc15_program_aspm(struct amdgpu_device *adev)
 }
 
 static void soc15_enable_doorbell_aperture(struct amdgpu_device *adev,
-					bool enable)
+					   bool enable)
 {
-	if (adev->flags & AMD_IS_APU) {
-		nbio_v7_0_enable_doorbell_aperture(adev, enable);
-	} else {
-		nbio_v6_1_enable_doorbell_aperture(adev, enable);
-		nbio_v6_1_enable_doorbell_selfring_aperture(adev, enable);
-	}
+	adev->nbio_funcs->enable_doorbell_aperture(adev, enable);
+	adev->nbio_funcs->enable_doorbell_selfring_aperture(adev, enable);
 }
 
 static const struct amdgpu_ip_block_version vega10_common_ip_block =
@@ -558,7 +550,12 @@ int soc15_set_ip_blocks(struct amdgpu_device *adev)
 		return -EINVAL;
 	}
 
-	nbio_v6_1_detect_hw_virt(adev);
+	if (adev->flags & AMD_IS_APU)
+		adev->nbio_funcs = &nbio_v7_0_funcs;
+	else
+		adev->nbio_funcs = &nbio_v6_1_funcs;
+
+	adev->nbio_funcs->detect_hw_virt(adev);
 
 	if (amdgpu_sriov_vf(adev))
 		adev->virt.ops = &xgpu_ai_virt_ops;
@@ -612,10 +609,7 @@ int soc15_set_ip_blocks(struct amdgpu_device *adev)
 
 static uint32_t soc15_get_rev_id(struct amdgpu_device *adev)
 {
-	if (adev->flags & AMD_IS_APU)
-		return nbio_v7_0_get_rev_id(adev);
-	else
-		return nbio_v6_1_get_rev_id(adev);
+	return adev->nbio_funcs->get_rev_id(adev);
 }
 
 static const struct amdgpu_asic_funcs soc15_asic_funcs =
@@ -651,11 +645,6 @@ static int soc15_common_early_init(void *handle)
 
 	adev->asic_funcs = &soc15_asic_funcs;
 
-	if (adev->flags & AMD_IS_APU)
-		adev->nbio_funcs = &nbio_v7_0_funcs;
-	else
-		adev->nbio_funcs = &nbio_v6_1_funcs;
-
 	if (amdgpu_get_ip_block(adev, AMD_IP_BLOCK_TYPE_PSP) &&
 		(amdgpu_ip_block_mask & (1 << AMD_IP_BLOCK_TYPE_PSP)))
 		psp_enabled = true;
@@ -763,8 +752,7 @@ static int soc15_common_hw_init(void *handle)
 	/* enable aspm */
 	soc15_program_aspm(adev);
 	/* setup nbio registers */
-	if (!(adev->flags & AMD_IS_APU))
-		nbio_v6_1_init_registers(adev);
+	adev->nbio_funcs->init_registers(adev);
 	/* enable the doorbell aperture */
 	soc15_enable_doorbell_aperture(adev, true);
 
@@ -925,9 +913,9 @@ static int soc15_common_set_clockgating_state(void *handle,
 
 	switch (adev->asic_type) {
 	case CHIP_VEGA10:
-		nbio_v6_1_update_medium_grain_clock_gating(adev,
+		adev->nbio_funcs->update_medium_grain_clock_gating(adev,
 				state == AMD_CG_STATE_GATE ? true : false);
-		nbio_v6_1_update_medium_grain_light_sleep(adev,
+		adev->nbio_funcs->update_medium_grain_light_sleep(adev,
 				state == AMD_CG_STATE_GATE ? true : false);
 		soc15_update_hdp_light_sleep(adev,
 				state == AMD_CG_STATE_GATE ? true : false);
@@ -941,9 +929,9 @@ static int soc15_common_set_clockgating_state(void *handle,
 				state == AMD_CG_STATE_GATE ? true : false);
 		break;
 	case CHIP_RAVEN:
-		nbio_v7_0_update_medium_grain_clock_gating(adev,
+		adev->nbio_funcs->update_medium_grain_clock_gating(adev,
 				state == AMD_CG_STATE_GATE ? true : false);
-		nbio_v6_1_update_medium_grain_light_sleep(adev,
+		adev->nbio_funcs->update_medium_grain_light_sleep(adev,
 				state == AMD_CG_STATE_GATE ? true : false);
 		soc15_update_hdp_light_sleep(adev,
 				state == AMD_CG_STATE_GATE ? true : false);
@@ -968,7 +956,7 @@ static void soc15_common_get_clockgating_state(void *handle, u32 *flags)
 	if (amdgpu_sriov_vf(adev))
 		*flags = 0;
 
-	nbio_v6_1_get_clockgating_state(adev, flags);
+	adev->nbio_funcs->get_clockgating_state(adev, flags);
 
 	/* AMD_CG_SUPPORT_HDP_LS */
 	data = RREG32(SOC15_REG_OFFSET(HDP, 0, mmHDP_MEM_POWER_LS));

drivers/gpu/drm/amd/amdgpu/soc15_common.h

@@ -24,22 +24,6 @@
 #ifndef __SOC15_COMMON_H__
 #define __SOC15_COMMON_H__
 
-struct nbio_hdp_flush_reg {
-	u32 ref_and_mask_cp0;
-	u32 ref_and_mask_cp1;
-	u32 ref_and_mask_cp2;
-	u32 ref_and_mask_cp3;
-	u32 ref_and_mask_cp4;
-	u32 ref_and_mask_cp5;
-	u32 ref_and_mask_cp6;
-	u32 ref_and_mask_cp7;
-	u32 ref_and_mask_cp8;
-	u32 ref_and_mask_cp9;
-	u32 ref_and_mask_sdma0;
-	u32 ref_and_mask_sdma1;
-};
-
-
 /* Register Access Macros */
 #define SOC15_REG_OFFSET(ip, inst, reg)	(adev->reg_offset[ip##_HWIP][inst][reg##_BASE_IDX] + reg)
 

drivers/gpu/drm/amd/amdgpu/vega10_ih.c

@@ -95,10 +95,7 @@ static int vega10_ih_irq_init(struct amdgpu_device *adev)
 	/* disable irqs */
 	vega10_ih_disable_interrupts(adev);
 
-	if (adev->flags & AMD_IS_APU)
-		nbio_v7_0_ih_control(adev);
-	else
-		nbio_v6_1_ih_control(adev);
+	adev->nbio_funcs->ih_control(adev);
 
 	ih_rb_cntl = RREG32_SOC15(OSSSYS, 0, mmIH_RB_CNTL);
 	/* Ring Buffer base. [39:8] of 40-bit address of the beginning of the ring buffer*/
@@ -149,10 +146,8 @@ static int vega10_ih_irq_init(struct amdgpu_device *adev)
 						 ENABLE, 0);
 	}
 	WREG32_SOC15(OSSSYS, 0, mmIH_DOORBELL_RPTR, ih_doorbell_rtpr);
-	if (adev->flags & AMD_IS_APU)
-		nbio_v7_0_ih_doorbell_range(adev, adev->irq.ih.use_doorbell, adev->irq.ih.doorbell_index);
-	else
-		nbio_v6_1_ih_doorbell_range(adev, adev->irq.ih.use_doorbell, adev->irq.ih.doorbell_index);
+	adev->nbio_funcs->ih_doorbell_range(adev, adev->irq.ih.use_doorbell,
+					    adev->irq.ih.doorbell_index);
 
 	tmp = RREG32_SOC15(OSSSYS, 0, mmIH_STORM_CLIENT_LIST_CNTL);
 	tmp = REG_SET_FIELD(tmp, IH_STORM_CLIENT_LIST_CNTL,