1 files changed, 175 insertions, 84 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index c76ac0dfe572..282197f4ffb1 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -411,6 +411,11 @@ static const uint32_t ecc_umc_mcumc_ctrl_mask_addrs[] = {
 	(0x001d43e0 + 0x00001800),
 };
 
+static inline bool gmc_v9_0_is_multi_chiplet(struct amdgpu_device *adev)
+{
+	return !!adev->aid_mask;
+}
+
 static int gmc_v9_0_ecc_interrupt_state(struct amdgpu_device *adev,
 		struct amdgpu_irq_src *src,
 		unsigned int type,
@@ -623,6 +628,9 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev,
 		}
 	}
 
+	if (kgd2kfd_vmfault_fast_path(adev, entry, retry_fault))
+		return 1;
+
 	if (!printk_ratelimit())
 		return 0;
 
@@ -644,8 +652,7 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev,
 		addr, entry->client_id,
 		soc15_ih_clientid_name[entry->client_id]);
 
-	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
-	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4))
+	if (gmc_v9_0_is_multi_chiplet(adev))
 		dev_err(adev->dev, "  cookie node_id %d fault from die %s%d%s\n",
 			node_id, node_id % 4 == 3 ? "RSV" : "AID", node_id / 4,
 			node_id % 4 == 1 ? ".XCD0" : node_id % 4 == 2 ? ".XCD1" : "");
@@ -672,6 +679,12 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev,
 	    (amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(9, 4, 2)))
 		return 0;
 
+	/* Only print L2 fault status if the status register could be read and
+	 * contains useful information
+	 */
+	if (!status)
+		return 0;
+
 	if (!amdgpu_sriov_vf(adev))
 		WREG32_P(hub->vm_l2_pro_fault_cntl, 1, ~1);
 
@@ -788,8 +801,7 @@ static bool gmc_v9_0_use_invalidate_semaphore(struct amdgpu_device *adev,
 				       uint32_t vmhub)
 {
 	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 2) ||
-	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
-	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4))
+	    gmc_v9_0_is_multi_chiplet(adev))
 		return false;
 
 	return ((vmhub == AMDGPU_MMHUB0(0) ||
@@ -1124,16 +1136,21 @@ static void gmc_v9_0_get_coherence_flags(struct amdgpu_device *adev,
 					 uint64_t *flags)
 {
 	struct amdgpu_device *bo_adev = amdgpu_ttm_adev(bo->tbo.bdev);
-	bool is_vram = bo->tbo.resource->mem_type == TTM_PL_VRAM;
-	bool coherent = bo->flags & (AMDGPU_GEM_CREATE_COHERENT | AMDGPU_GEM_CREATE_EXT_COHERENT);
+	bool is_vram = bo->tbo.resource &&
+		bo->tbo.resource->mem_type == TTM_PL_VRAM;
+	bool coherent = bo->flags & (AMDGPU_GEM_CREATE_COHERENT |
+				     AMDGPU_GEM_CREATE_EXT_COHERENT);
 	bool ext_coherent = bo->flags & AMDGPU_GEM_CREATE_EXT_COHERENT;
 	bool uncached = bo->flags & AMDGPU_GEM_CREATE_UNCACHED;
 	struct amdgpu_vm *vm = mapping->bo_va->base.vm;
 	unsigned int mtype_local, mtype;
+	uint32_t gc_ip_version = amdgpu_ip_version(adev, GC_HWIP, 0);
 	bool snoop = false;
 	bool is_local;
 
-	switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
+	dma_resv_assert_held(bo->tbo.base.resv);
+
+	switch (gc_ip_version) {
 	case IP_VERSION(9, 4, 1):
 	case IP_VERSION(9, 4, 2):
 		if (is_vram) {
@@ -1147,10 +1164,7 @@ static void gmc_v9_0_get_coherence_flags(struct amdgpu_device *adev,
 				/* FIXME: is this still needed? Or does
 				 * amdgpu_ttm_tt_pde_flags already handle this?
 				 */
-				if ((amdgpu_ip_version(adev, GC_HWIP, 0) ==
-					     IP_VERSION(9, 4, 2) ||
-				     amdgpu_ip_version(adev, GC_HWIP, 0) ==
-					     IP_VERSION(9, 4, 3)) &&
+				if (gc_ip_version == IP_VERSION(9, 4, 2) &&
 				    adev->gmc.xgmi.connected_to_cpu)
 					snoop = true;
 			} else {
@@ -1174,6 +1188,7 @@ static void gmc_v9_0_get_coherence_flags(struct amdgpu_device *adev,
 		break;
 	case IP_VERSION(9, 4, 3):
 	case IP_VERSION(9, 4, 4):
+	case IP_VERSION(9, 5, 0):
 		/* Only local VRAM BOs or system memory on non-NUMA APUs
 		 * can be assumed to be local in their entirety. Choose
 		 * MTYPE_NC as safe fallback for all system memory BOs on
@@ -1198,20 +1213,17 @@ static void gmc_v9_0_get_coherence_flags(struct amdgpu_device *adev,
 		if (uncached) {
 			mtype = MTYPE_UC;
 		} else if (ext_coherent) {
-			if (adev->rev_id)
-				mtype = is_local ? MTYPE_CC : MTYPE_UC;
-			else
-				mtype = MTYPE_UC;
+			mtype = is_local ? MTYPE_CC : MTYPE_UC;
 		} else if (adev->flags & AMD_IS_APU) {
 			mtype = is_local ? mtype_local : MTYPE_NC;
 		} else {
 			/* dGPU */
 			if (is_local)
 				mtype = mtype_local;
-			else if (is_vram)
-				mtype = MTYPE_NC;
-			else
+			else if (gc_ip_version < IP_VERSION(9, 5, 0) && !is_vram)
 				mtype = MTYPE_UC;
+			else
+				mtype = MTYPE_NC;
 		}
 
 		break;
@@ -1251,9 +1263,8 @@ static void gmc_v9_0_get_vm_pte(struct amdgpu_device *adev,
 		*flags &= ~AMDGPU_PTE_VALID;
 	}
 
-	if (bo && bo->tbo.resource)
-		gmc_v9_0_get_coherence_flags(adev, mapping->bo_va->base.bo,
-					     mapping, flags);
+	if ((*flags & AMDGPU_PTE_VALID) && bo)
+		gmc_v9_0_get_coherence_flags(adev, bo, mapping, flags);
 }
 
 static void gmc_v9_0_override_vm_pte_flags(struct amdgpu_device *adev,
@@ -1265,8 +1276,8 @@ static void gmc_v9_0_override_vm_pte_flags(struct amdgpu_device *adev,
 	/* Only GFX 9.4.3 APUs associate GPUs with NUMA nodes. Local system
 	 * memory can use more efficient MTYPEs.
 	 */
-	if (amdgpu_ip_version(adev, GC_HWIP, 0) != IP_VERSION(9, 4, 3) &&
-	    amdgpu_ip_version(adev, GC_HWIP, 0) != IP_VERSION(9, 4, 4))
+	if (!(adev->flags & AMD_IS_APU) ||
+	    amdgpu_ip_version(adev, GC_HWIP, 0) != IP_VERSION(9, 4, 3))
 		return;
 
 	/* Only direct-mapped memory allows us to determine the NUMA node from
@@ -1322,7 +1333,7 @@ static void gmc_v9_0_override_vm_pte_flags(struct amdgpu_device *adev,
 				mtype_local = MTYPE_CC;
 
 			*flags = AMDGPU_PTE_MTYPE_VG10(*flags, mtype_local);
-		} else if (adev->rev_id) {
+		} else {
 			/* MTYPE_UC case */
 			*flags = AMDGPU_PTE_MTYPE_VG10(*flags, MTYPE_CC);
 		}
@@ -1387,14 +1398,44 @@ gmc_v9_0_get_memory_partition(struct amdgpu_device *adev, u32 *supp_modes)
 }
 
 static enum amdgpu_memory_partition
+gmc_v9_0_query_vf_memory_partition(struct amdgpu_device *adev)
+{
+	switch (adev->gmc.num_mem_partitions) {
+	case 0:
+		return UNKNOWN_MEMORY_PARTITION_MODE;
+	case 1:
+		return AMDGPU_NPS1_PARTITION_MODE;
+	case 2:
+		return AMDGPU_NPS2_PARTITION_MODE;
+	case 4:
+		return AMDGPU_NPS4_PARTITION_MODE;
+	default:
+		return AMDGPU_NPS1_PARTITION_MODE;
+	}
+
+	return AMDGPU_NPS1_PARTITION_MODE;
+}
+
+static enum amdgpu_memory_partition
 gmc_v9_0_query_memory_partition(struct amdgpu_device *adev)
 {
 	if (amdgpu_sriov_vf(adev))
-		return AMDGPU_NPS1_PARTITION_MODE;
+		return gmc_v9_0_query_vf_memory_partition(adev);
 
 	return gmc_v9_0_get_memory_partition(adev, NULL);
 }
 
+static bool gmc_v9_0_need_reset_on_init(struct amdgpu_device *adev)
+{
+	if (adev->nbio.funcs && adev->nbio.funcs->is_nps_switch_requested &&
+	    adev->nbio.funcs->is_nps_switch_requested(adev)) {
+		adev->gmc.reset_flags |= AMDGPU_GMC_INIT_RESET_NPS;
+		return true;
+	}
+
+	return false;
+}
+
 static const struct amdgpu_gmc_funcs gmc_v9_0_gmc_funcs = {
 	.flush_gpu_tlb = gmc_v9_0_flush_gpu_tlb,
 	.flush_gpu_tlb_pasid = gmc_v9_0_flush_gpu_tlb_pasid,
@@ -1406,6 +1447,8 @@ static const struct amdgpu_gmc_funcs gmc_v9_0_gmc_funcs = {
 	.override_vm_pte_flags = gmc_v9_0_override_vm_pte_flags,
 	.get_vbios_fb_size = gmc_v9_0_get_vbios_fb_size,
 	.query_mem_partition_mode = &gmc_v9_0_query_memory_partition,
+	.request_mem_partition_mode = &amdgpu_gmc_request_memory_partition,
+	.need_reset_on_init = &gmc_v9_0_need_reset_on_init,
 };
 
 static void gmc_v9_0_set_gmc_funcs(struct amdgpu_device *adev)
@@ -1452,14 +1495,13 @@ static void gmc_v9_0_set_umc_funcs(struct amdgpu_device *adev)
 			adev->umc.channel_idx_tbl = &umc_v6_7_channel_idx_tbl_second[0][0];
 		break;
 	case IP_VERSION(12, 0, 0):
+	case IP_VERSION(12, 5, 0):
 		adev->umc.max_ras_err_cnt_per_query =
 			UMC_V12_0_TOTAL_CHANNEL_NUM(adev) * UMC_V12_0_BAD_PAGE_NUM_PER_CHANNEL;
 		adev->umc.channel_inst_num = UMC_V12_0_CHANNEL_INSTANCE_NUM;
 		adev->umc.umc_inst_num = UMC_V12_0_UMC_INSTANCE_NUM;
 		adev->umc.node_inst_num /= UMC_V12_0_UMC_INSTANCE_NUM;
 		adev->umc.channel_offs = UMC_V12_0_PER_CHANNEL_OFFSET;
-		adev->umc.active_mask = adev->aid_mask;
-		adev->umc.retire_unit = UMC_V12_0_BAD_PAGE_NUM_PER_CHANNEL;
 		if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu)
 			adev->umc.ras = &umc_v12_0_ras;
 		break;
@@ -1478,6 +1520,7 @@ static void gmc_v9_0_set_mmhub_funcs(struct amdgpu_device *adev)
 		adev->mmhub.funcs = &mmhub_v1_7_funcs;
 		break;
 	case IP_VERSION(1, 8, 0):
+	case IP_VERSION(1, 8, 1):
 		adev->mmhub.funcs = &mmhub_v1_8_funcs;
 		break;
 	default:
@@ -1499,6 +1542,7 @@ static void gmc_v9_0_set_mmhub_ras_funcs(struct amdgpu_device *adev)
 		adev->mmhub.ras = &mmhub_v1_7_ras;
 		break;
 	case IP_VERSION(1, 8, 0):
+	case IP_VERSION(1, 8, 1):
 		adev->mmhub.ras = &mmhub_v1_8_ras;
 		break;
 	default:
@@ -1509,8 +1553,7 @@ static void gmc_v9_0_set_mmhub_ras_funcs(struct amdgpu_device *adev)
 
 static void gmc_v9_0_set_gfxhub_funcs(struct amdgpu_device *adev)
 {
-	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
-	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4))
+	if (gmc_v9_0_is_multi_chiplet(adev))
 		adev->gfxhub.funcs = &gfxhub_v1_2_funcs;
 	else
 		adev->gfxhub.funcs = &gfxhub_v1_0_funcs;
@@ -1545,9 +1588,46 @@ static void gmc_v9_0_set_xgmi_ras_funcs(struct amdgpu_device *adev)
 		adev->gmc.xgmi.ras = &xgmi_ras;
 }
 
-static int gmc_v9_0_early_init(void *handle)
+static void gmc_v9_0_init_nps_details(struct amdgpu_device *adev)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	enum amdgpu_memory_partition mode;
+	uint32_t supp_modes;
+	int i;
+
+	adev->gmc.supported_nps_modes = 0;
+
+	if (amdgpu_sriov_vf(adev) || (adev->flags & AMD_IS_APU))
+		return;
+
+	mode = gmc_v9_0_get_memory_partition(adev, &supp_modes);
+
+	/* Mode detected by hardware and supported modes available */
+	if ((mode != UNKNOWN_MEMORY_PARTITION_MODE) && supp_modes) {
+		while ((i = ffs(supp_modes))) {
+			if (AMDGPU_ALL_NPS_MASK & BIT(i))
+				adev->gmc.supported_nps_modes |= BIT(i);
+			supp_modes &= supp_modes - 1;
+		}
+	} else {
+		/*TODO: Check PSP version also which supports NPS switch. Otherwise keep
+	 * supported modes as 0.
+	 */
+		switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
+		case IP_VERSION(9, 4, 3):
+		case IP_VERSION(9, 4, 4):
+			adev->gmc.supported_nps_modes =
+				BIT(AMDGPU_NPS1_PARTITION_MODE) |
+				BIT(AMDGPU_NPS4_PARTITION_MODE);
+			break;
+		default:
+			break;
+		}
+	}
+}
+
+static int gmc_v9_0_early_init(struct amdgpu_ip_block *ip_block)
+{
+	struct amdgpu_device *adev = ip_block->adev;
 
 	/*
 	 * 9.4.0, 9.4.1 and 9.4.3 don't have XGMI defined
@@ -1555,8 +1635,7 @@ static int gmc_v9_0_early_init(void *handle)
 	 */
 	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 0) ||
 	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 1) ||
-	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
-	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4))
+	    gmc_v9_0_is_multi_chiplet(adev))
 		adev->gmc.xgmi.supported = true;
 
 	if (amdgpu_ip_version(adev, XGMI_HWIP, 0) == IP_VERSION(6, 1, 0)) {
@@ -1565,8 +1644,7 @@ static int gmc_v9_0_early_init(void *handle)
 			adev->smuio.funcs->is_host_gpu_xgmi_supported(adev);
 	}
 
-	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
-	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4)) {
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3)) {
 		enum amdgpu_pkg_type pkg_type =
 			adev->smuio.funcs->get_pkg_type(adev);
 		/* On GFXIP 9.4.3. APU, there is no physical VRAM domain present
@@ -1601,9 +1679,9 @@ static int gmc_v9_0_early_init(void *handle)
 	return 0;
 }
 
-static int gmc_v9_0_late_init(void *handle)
+static int gmc_v9_0_late_init(struct amdgpu_ip_block *ip_block)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct amdgpu_device *adev = ip_block->adev;
 	int r;
 
 	r = amdgpu_gmc_allocate_vm_inv_eng(adev);
@@ -1729,6 +1807,7 @@ static int gmc_v9_0_mc_init(struct amdgpu_device *adev)
 		case IP_VERSION(9, 4, 2):
 		case IP_VERSION(9, 4, 3):
 		case IP_VERSION(9, 4, 4):
+		case IP_VERSION(9, 5, 0):
 		default:
 			adev->gmc.gart_size = 512ULL << 20;
 			break;
@@ -1900,6 +1979,8 @@ gmc_v9_0_init_sw_mem_ranges(struct amdgpu_device *adev,
 
 	switch (mode) {
 	case UNKNOWN_MEMORY_PARTITION_MODE:
+		adev->gmc.num_mem_partitions = 0;
+		break;
 	case AMDGPU_NPS1_PARTITION_MODE:
 		adev->gmc.num_mem_partitions = 1;
 		break;
@@ -1919,7 +2000,7 @@ gmc_v9_0_init_sw_mem_ranges(struct amdgpu_device *adev,
 
 	/* Use NPS range info, if populated */
 	r = amdgpu_gmc_get_nps_memranges(adev, mem_ranges,
-					 adev->gmc.num_mem_partitions);
+					 &adev->gmc.num_mem_partitions);
 	if (!r) {
 		l = 0;
 		for (i = 1; i < adev->gmc.num_mem_partitions; ++i) {
@@ -1929,6 +2010,11 @@ gmc_v9_0_init_sw_mem_ranges(struct amdgpu_device *adev,
 		}
 
 	} else {
+		if (!adev->gmc.num_mem_partitions) {
+			dev_err(adev->dev,
+				"Not able to detect NPS mode, fall back to NPS1");
+			adev->gmc.num_mem_partitions = 1;
+		}
 		/* Fallback to sw based calculation */
 		size = (adev->gmc.real_vram_size + SZ_16M) >> AMDGPU_GPU_PAGE_SHIFT;
 		size /= adev->gmc.num_mem_partitions;
@@ -1985,12 +2071,15 @@ static void gmc_v9_4_3_init_vram_info(struct amdgpu_device *adev)
 {
 	adev->gmc.vram_type = AMDGPU_VRAM_TYPE_HBM;
 	adev->gmc.vram_width = 128 * 64;
+
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 5, 0))
+		adev->gmc.vram_type = AMDGPU_VRAM_TYPE_HBM3E;
 }
 
-static int gmc_v9_0_sw_init(void *handle)
+static int gmc_v9_0_sw_init(struct amdgpu_ip_block *ip_block)
 {
 	int r, vram_width = 0, vram_type = 0, vram_vendor = 0, dma_addr_bits;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct amdgpu_device *adev = ip_block->adev;
 	unsigned long inst_mask = adev->aid_mask;
 
 	adev->gfxhub.funcs->init(adev);
@@ -1999,8 +2088,7 @@ static int gmc_v9_0_sw_init(void *handle)
 
 	spin_lock_init(&adev->gmc.invalidate_lock);
 
-	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
-	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4)) {
+	if (gmc_v9_0_is_multi_chiplet(adev)) {
 		gmc_v9_4_3_init_vram_info(adev);
 	} else if (!adev->bios) {
 		if (adev->flags & AMD_IS_APU) {
@@ -2084,6 +2172,7 @@ static int gmc_v9_0_sw_init(void *handle)
 		break;
 	case IP_VERSION(9, 4, 3):
 	case IP_VERSION(9, 4, 4):
+	case IP_VERSION(9, 5, 0):
 		bitmap_set(adev->vmhubs_mask, AMDGPU_GFXHUB(0),
 				  NUM_XCC(adev->gfx.xcc_mask));
 
@@ -2149,8 +2238,7 @@ static int gmc_v9_0_sw_init(void *handle)
 
 	amdgpu_gmc_get_vbios_allocations(adev);
 
-	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
-	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4)) {
+	if (gmc_v9_0_is_multi_chiplet(adev)) {
 		r = gmc_v9_0_init_mem_ranges(adev);
 		if (r)
 			return r;
@@ -2165,6 +2253,7 @@ static int gmc_v9_0_sw_init(void *handle)
 	if (r)
 		return r;
 
+	gmc_v9_0_init_nps_details(adev);
 	/*
 	 * number of VMs
 	 * VMID 0 is reserved for System
@@ -2178,8 +2267,7 @@ static int gmc_v9_0_sw_init(void *handle)
 	adev->vm_manager.first_kfd_vmid =
 		(amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 1) ||
 		 amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 2) ||
-		 amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
-		 amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4)) ?
+		 gmc_v9_0_is_multi_chiplet(adev)) ?
 			3 :
 			8;
 
@@ -2191,19 +2279,17 @@ static int gmc_v9_0_sw_init(void *handle)
 	if (r)
 		return r;
 
-	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
-	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4))
+	if (gmc_v9_0_is_multi_chiplet(adev))
 		amdgpu_gmc_sysfs_init(adev);
 
 	return 0;
 }
 
-static int gmc_v9_0_sw_fini(void *handle)
+static int gmc_v9_0_sw_fini(struct amdgpu_ip_block *ip_block)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct amdgpu_device *adev = ip_block->adev;
 
-	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
-	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4))
+	if (gmc_v9_0_is_multi_chiplet(adev))
 		amdgpu_gmc_sysfs_fini(adev);
 
 	amdgpu_gmc_ras_fini(adev);
@@ -2308,9 +2394,9 @@ static int gmc_v9_0_gart_enable(struct amdgpu_device *adev)
 	return 0;
 }
 
-static int gmc_v9_0_hw_init(void *handle)
+static int gmc_v9_0_hw_init(struct amdgpu_ip_block *ip_block)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct amdgpu_device *adev = ip_block->adev;
 	bool value;
 	int i, r;
 
@@ -2324,13 +2410,6 @@ static int gmc_v9_0_hw_init(void *handle)
 	adev->gmc.flush_tlb_needs_extra_type_2 =
 		amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 0) &&
 		adev->gmc.xgmi.num_physical_nodes;
-	/*
-	 * TODO: This workaround is badly documented and had a buggy
-	 * implementation. We should probably verify what we do here.
-	 */
-	adev->gmc.flush_tlb_needs_extra_type_0 =
-		amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) &&
-		adev->rev_id == 0;
 
 	/* The sequence of these two function calls matters.*/
 	gmc_v9_0_init_golden_registers(adev);
@@ -2348,7 +2427,7 @@ static int gmc_v9_0_hw_init(void *handle)
 	adev->hdp.funcs->init_registers(adev);
 
 	/* After HDP is initialized, flush HDP.*/
-	adev->hdp.funcs->flush_hdp(adev, NULL);
+	amdgpu_device_flush_hdp(adev, NULL);
 
 	if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
 		value = false;
@@ -2393,9 +2472,9 @@ static void gmc_v9_0_gart_disable(struct amdgpu_device *adev)
 	adev->mmhub.funcs->gart_disable(adev);
 }
 
-static int gmc_v9_0_hw_fini(void *handle)
+static int gmc_v9_0_hw_fini(struct amdgpu_ip_block *ip_block)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct amdgpu_device *adev = ip_block->adev;
 
 	gmc_v9_0_gart_disable(adev);
 
@@ -2413,58 +2492,70 @@ static int gmc_v9_0_hw_fini(void *handle)
 	if (adev->mmhub.funcs->update_power_gating)
 		adev->mmhub.funcs->update_power_gating(adev, false);
 
-	amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
+	/*
+	 * For minimal init, late_init is not called, hence VM fault/RAS irqs
+	 * are not enabled.
+	 */
+	if (adev->init_lvl->level != AMDGPU_INIT_LEVEL_MINIMAL_XGMI) {
+		amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
 
-	if (adev->gmc.ecc_irq.funcs &&
-		amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__UMC))
-		amdgpu_irq_put(adev, &adev->gmc.ecc_irq, 0);
+		if (adev->gmc.ecc_irq.funcs &&
+		    amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__UMC))
+			amdgpu_irq_put(adev, &adev->gmc.ecc_irq, 0);
+	}
 
 	return 0;
 }
 
-static int gmc_v9_0_suspend(void *handle)
+static int gmc_v9_0_suspend(struct amdgpu_ip_block *ip_block)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-
-	return gmc_v9_0_hw_fini(adev);
+	return gmc_v9_0_hw_fini(ip_block);
 }
 
-static int gmc_v9_0_resume(void *handle)
+static int gmc_v9_0_resume(struct amdgpu_ip_block *ip_block)
 {
+	struct amdgpu_device *adev = ip_block->adev;
 	int r;
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
-	r = gmc_v9_0_hw_init(adev);
+	/* If a reset is done for NPS mode switch, read the memory range
+	 * information again.
+	 */
+	if (adev->gmc.reset_flags & AMDGPU_GMC_INIT_RESET_NPS) {
+		gmc_v9_0_init_sw_mem_ranges(adev, adev->gmc.mem_partitions);
+		adev->gmc.reset_flags &= ~AMDGPU_GMC_INIT_RESET_NPS;
+	}
+
+	r = gmc_v9_0_hw_init(ip_block);
 	if (r)
 		return r;
 
-	amdgpu_vmid_reset_all(adev);
+	amdgpu_vmid_reset_all(ip_block->adev);
 
 	return 0;
 }
 
-static bool gmc_v9_0_is_idle(void *handle)
+static bool gmc_v9_0_is_idle(struct amdgpu_ip_block *ip_block)
 {
 	/* MC is always ready in GMC v9.*/
 	return true;
 }
 
-static int gmc_v9_0_wait_for_idle(void *handle)
+static int gmc_v9_0_wait_for_idle(struct amdgpu_ip_block *ip_block)
 {
 	/* There is no need to wait for MC idle in GMC v9.*/
 	return 0;
 }
 
-static int gmc_v9_0_soft_reset(void *handle)
+static int gmc_v9_0_soft_reset(struct amdgpu_ip_block *ip_block)
 {
 	/* XXX for emulation.*/
 	return 0;
 }
 
-static int gmc_v9_0_set_clockgating_state(void *handle,
+static int gmc_v9_0_set_clockgating_state(struct amdgpu_ip_block *ip_block,
 					enum amd_clockgating_state state)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct amdgpu_device *adev = ip_block->adev;
 
 	adev->mmhub.funcs->set_clockgating(adev, state);
 
@@ -2473,16 +2564,16 @@ static int gmc_v9_0_set_clockgating_state(void *handle,
 	return 0;
 }
 
-static void gmc_v9_0_get_clockgating_state(void *handle, u64 *flags)
+static void gmc_v9_0_get_clockgating_state(struct amdgpu_ip_block *ip_block, u64 *flags)
 {
-	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+	struct amdgpu_device *adev = ip_block->adev;
 
 	adev->mmhub.funcs->get_clockgating(adev, flags);
 
 	athub_v1_0_get_clockgating(adev, flags);
 }
 
-static int gmc_v9_0_set_powergating_state(void *handle,
+static int gmc_v9_0_set_powergating_state(struct amdgpu_ip_block *ip_block,
 					enum amd_powergating_state state)
 {
 	return 0;