Merge tag 'amd-drm-next-5.6-2020-01-09' of git://people.freedesktop.org/~agd5f/linux into drm-next

amd-drm-next-5.6-2020-01-09:

amdgpu:
- Enable DCN support on POWER
- Enable GFXOFF for Raven1 refresh
- Clean up MM engine idle handlers
- HDMI 2.0 audio fixes
- Fixes for some 10 bpc EDP panels
- Watermark fixes for renoir
- SR-IOV fixes
- Runtime pm robustness fixes
- Arcturus VCN fixes
- RAS fixes
- BACO fixes for Arcturus
- Stable pstate fixes for swSMU
- HDCP fixes
- PSP cleanup
- HDMI fixes
- Misc cleanups

amdkfd:
- Spread interrupt work across cores to reduce latency
- Topology fixes for APUs
- GPU reset improvements

UAPI:
- Enable DRIVER_SYNCOBJ_TIMELINE for vulkan
- Return better error values for kfd process ioctl

Signed-off-by: Dave Airlie <[email protected]>
From: Alex Deucher <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]

Author: airlied

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

@@ -636,9 +636,8 @@ struct amdgpu_fw_vram_usage {
 	struct amdgpu_bo *reserved_bo;
 	void *va;
 
-	/* Offset on the top of VRAM, used as c2p write buffer.
+	/* GDDR6 training support flag.
 	*/
-	u64 mem_train_fb_loc;
 	bool mem_train_support;
 };
 
@@ -994,8 +993,6 @@ struct amdgpu_device {
 
 	bool                            pm_sysfs_en;
 	bool                            ucode_sysfs_en;
-
-	bool				in_baco;
 };
 
 static inline struct amdgpu_device *amdgpu_ttm_adev(struct ttm_bo_device *bdev)

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

@@ -46,6 +46,8 @@
 #include "soc15.h"
 #include "soc15d.h"
 #include "amdgpu_amdkfd_gfx_v9.h"
+#include "gfxhub_v1_0.h"
+#include "mmhub_v9_4.h"
 
 #define HQD_N_REGS 56
 #define DUMP_REG(addr) do {				\
@@ -258,6 +260,22 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
 	return 0;
 }
 
+static void kgd_set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
+		uint64_t page_table_base)
+{
+	struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
+		pr_err("trying to set page table base for wrong VMID %u\n",
+		       vmid);
+		return;
+	}
+
+	mmhub_v9_4_setup_vm_pt_regs(adev, vmid, page_table_base);
+
+	gfxhub_v1_0_setup_vm_pt_regs(adev, vmid, page_table_base);
+}
+
 const struct kfd2kgd_calls arcturus_kfd2kgd = {
 	.program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings,
 	.set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping,
@@ -277,7 +295,7 @@ const struct kfd2kgd_calls arcturus_kfd2kgd = {
 	.get_atc_vmid_pasid_mapping_info =
 			kgd_gfx_v9_get_atc_vmid_pasid_mapping_info,
 	.get_tile_config = kgd_gfx_v9_get_tile_config,
-	.set_vm_context_page_table_base = kgd_gfx_v9_set_vm_context_page_table_base,
+	.set_vm_context_page_table_base = kgd_set_vm_context_page_table_base,
 	.invalidate_tlbs = kgd_gfx_v9_invalidate_tlbs,
 	.invalidate_tlbs_vmid = kgd_gfx_v9_invalidate_tlbs_vmid,
 	.get_hive_id = amdgpu_amdkfd_get_hive_id,

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c

@@ -40,7 +40,6 @@
 #include "soc15d.h"
 #include "mmhub_v1_0.h"
 #include "gfxhub_v1_0.h"
-#include "mmhub_v9_4.h"
 
 
 enum hqd_dequeue_request_type {
@@ -758,8 +757,8 @@ uint32_t kgd_gfx_v9_address_watch_get_offset(struct kgd_dev *kgd,
 	return 0;
 }
 
-void kgd_gfx_v9_set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
-		uint64_t page_table_base)
+static void kgd_gfx_v9_set_vm_context_page_table_base(struct kgd_dev *kgd,
+			uint32_t vmid, uint64_t page_table_base)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 
@@ -769,14 +768,7 @@ void kgd_gfx_v9_set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmi
 		return;
 	}
 
-	/* TODO: take advantage of per-process address space size. For
-	 * now, all processes share the same address space size, like
-	 * on GFX8 and older.
-	 */
-	if (adev->asic_type == CHIP_ARCTURUS) {
-		mmhub_v9_4_setup_vm_pt_regs(adev, vmid, page_table_base);
-	} else
-		mmhub_v1_0_setup_vm_pt_regs(adev, vmid, page_table_base);
+	mmhub_v1_0_setup_vm_pt_regs(adev, vmid, page_table_base);
 
 	gfxhub_v1_0_setup_vm_pt_regs(adev, vmid, page_table_base);
 }

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h

@@ -57,8 +57,6 @@ uint32_t kgd_gfx_v9_address_watch_get_offset(struct kgd_dev *kgd,
 
 bool kgd_gfx_v9_get_atc_vmid_pasid_mapping_info(struct kgd_dev *kgd,
 					uint8_t vmid, uint16_t *p_pasid);
-void kgd_gfx_v9_set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
-		uint64_t page_table_base);
 int kgd_gfx_v9_invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid);
 int kgd_gfx_v9_invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid);
 int kgd_gfx_v9_get_tile_config(struct kgd_dev *kgd,

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

@@ -2022,7 +2022,7 @@ int amdgpu_atombios_init(struct amdgpu_device *adev)
 	if (adev->is_atom_fw) {
 		amdgpu_atomfirmware_scratch_regs_init(adev);
 		amdgpu_atomfirmware_allocate_fb_scratch(adev);
-		ret = amdgpu_atomfirmware_get_mem_train_fb_loc(adev);
+		ret = amdgpu_atomfirmware_get_mem_train_info(adev);
 		if (ret) {
 			DRM_ERROR("Failed to get mem train fb location.\n");
 			return ret;

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

@@ -525,16 +525,12 @@ static int gddr6_mem_train_support(struct amdgpu_device *adev)
 	return ret;
 }
 
-int amdgpu_atomfirmware_get_mem_train_fb_loc(struct amdgpu_device *adev)
+int amdgpu_atomfirmware_get_mem_train_info(struct amdgpu_device *adev)
 {
 	struct atom_context *ctx = adev->mode_info.atom_context;
-	unsigned char *bios = ctx->bios;
-	struct vram_reserve_block *reserved_block;
-	int index, block_number;
+	int index;
 	uint8_t frev, crev;
 	uint16_t data_offset, size;
-	uint32_t start_address_in_kb;
-	uint64_t offset;
 	int ret;
 
 	adev->fw_vram_usage.mem_train_support = false;
@@ -569,32 +565,6 @@ int amdgpu_atomfirmware_get_mem_train_fb_loc(struct amdgpu_device *adev)
 		return -EINVAL;
 	}
 
-	reserved_block = (struct vram_reserve_block *)
-		(bios + data_offset + sizeof(struct atom_common_table_header));
-	block_number = ((unsigned int)size - sizeof(struct atom_common_table_header))
-		/ sizeof(struct vram_reserve_block);
-	reserved_block += (block_number > 0) ? block_number-1 : 0;
-	DRM_DEBUG("block_number:0x%04x, last block: 0x%08xkb sz, %dkb fw, %dkb drv.\n",
-		  block_number,
-		  le32_to_cpu(reserved_block->start_address_in_kb),
-		  le16_to_cpu(reserved_block->used_by_firmware_in_kb),
-		  le16_to_cpu(reserved_block->used_by_driver_in_kb));
-	if (reserved_block->used_by_firmware_in_kb > 0) {
-		start_address_in_kb = le32_to_cpu(reserved_block->start_address_in_kb);
-		offset = (uint64_t)start_address_in_kb * ONE_KiB;
-		if ((offset & (ONE_MiB - 1)) < (4 * ONE_KiB + 1) ) {
-			offset -= ONE_MiB;
-		}
-
-		offset &= ~(ONE_MiB - 1);
-		adev->fw_vram_usage.mem_train_fb_loc = offset;
-		adev->fw_vram_usage.mem_train_support = true;
-		DRM_DEBUG("mem_train_fb_loc:0x%09llx.\n", offset);
-		ret = 0;
-	} else {
-		DRM_ERROR("used_by_firmware_in_kb is 0!\n");
-		ret = -EINVAL;
-	}
-
-	return ret;
+	adev->fw_vram_usage.mem_train_support = true;
+	return 0;
 }

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

@@ -31,7 +31,7 @@ void amdgpu_atomfirmware_scratch_regs_init(struct amdgpu_device *adev);
 int amdgpu_atomfirmware_allocate_fb_scratch(struct amdgpu_device *adev);
 int amdgpu_atomfirmware_get_vram_info(struct amdgpu_device *adev,
 	int *vram_width, int *vram_type, int *vram_vendor);
-int amdgpu_atomfirmware_get_mem_train_fb_loc(struct amdgpu_device *adev);
+int amdgpu_atomfirmware_get_mem_train_info(struct amdgpu_device *adev);
 int amdgpu_atomfirmware_get_clock_info(struct amdgpu_device *adev);
 int amdgpu_atomfirmware_get_gfx_info(struct amdgpu_device *adev);
 bool amdgpu_atomfirmware_mem_ecc_supported(struct amdgpu_device *adev);

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

@@ -613,7 +613,17 @@ static bool amdgpu_atpx_detect(void)
 	bool d3_supported = false;
 	struct pci_dev *parent_pdev;
 
-	while ((pdev = pci_get_class(PCI_BASE_CLASS_DISPLAY << 16, pdev)) != NULL) {
+	while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, pdev)) != NULL) {
+		vga_count++;
+
+		has_atpx |= (amdgpu_atpx_pci_probe_handle(pdev) == true);
+
+		parent_pdev = pci_upstream_bridge(pdev);
+		d3_supported |= parent_pdev && parent_pdev->bridge_d3;
+		amdgpu_atpx_get_quirks(pdev);
+	}
+
+	while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_OTHER << 8, pdev)) != NULL) {
 		vga_count++;
 
 		has_atpx |= (amdgpu_atpx_pci_probe_handle(pdev) == true);

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

@@ -74,7 +74,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 			   struct amdgpu_ctx *ctx)
 {
 	unsigned num_entities = amdgpu_ctx_total_num_entities();
-	unsigned i, j, k;
+	unsigned i, j;
 	int r;
 
 	if (priority < 0 || priority >= DRM_SCHED_PRIORITY_MAX)
@@ -121,72 +121,57 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 	ctx->override_priority = DRM_SCHED_PRIORITY_UNSET;
 
 	for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
-		struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
-		struct drm_sched_rq *rqs[AMDGPU_MAX_RINGS];
-		unsigned num_rings = 0;
-		unsigned num_rqs = 0;
+		struct drm_gpu_scheduler **scheds;
+		struct drm_gpu_scheduler *sched;
+		unsigned num_scheds = 0;
 
 		switch (i) {
 		case AMDGPU_HW_IP_GFX:
-			rings[0] = &adev->gfx.gfx_ring[0];
-			num_rings = 1;
+			sched = &adev->gfx.gfx_ring[0].sched;
+			scheds = &sched;
+			num_scheds = 1;
 			break;
 		case AMDGPU_HW_IP_COMPUTE:
-			for (j = 0; j < adev->gfx.num_compute_rings; ++j)
-				rings[j] = &adev->gfx.compute_ring[j];
-			num_rings = adev->gfx.num_compute_rings;
+			scheds = adev->gfx.compute_sched;
+			num_scheds = adev->gfx.num_compute_sched;
 			break;
 		case AMDGPU_HW_IP_DMA:
-			for (j = 0; j < adev->sdma.num_instances; ++j)
-				rings[j] = &adev->sdma.instance[j].ring;
-			num_rings = adev->sdma.num_instances;
+			scheds = adev->sdma.sdma_sched;
+			num_scheds = adev->sdma.num_sdma_sched;
 			break;
 		case AMDGPU_HW_IP_UVD:
-			rings[0] = &adev->uvd.inst[0].ring;
-			num_rings = 1;
+			sched = &adev->uvd.inst[0].ring.sched;
+			scheds = &sched;
+			num_scheds = 1;
 			break;
 		case AMDGPU_HW_IP_VCE:
-			rings[0] = &adev->vce.ring[0];
-			num_rings = 1;
+			sched = &adev->vce.ring[0].sched;
+			scheds = &sched;
+			num_scheds = 1;
 			break;
 		case AMDGPU_HW_IP_UVD_ENC:
-			rings[0] = &adev->uvd.inst[0].ring_enc[0];
-			num_rings = 1;
+			sched = &adev->uvd.inst[0].ring_enc[0].sched;
+			scheds = &sched;
+			num_scheds = 1;
 			break;
 		case AMDGPU_HW_IP_VCN_DEC:
-			for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
-				if (adev->vcn.harvest_config & (1 << j))
-					continue;
-				rings[num_rings++] = &adev->vcn.inst[j].ring_dec;
-			}
+			scheds = adev->vcn.vcn_dec_sched;
+			num_scheds =  adev->vcn.num_vcn_dec_sched;
 			break;
 		case AMDGPU_HW_IP_VCN_ENC:
-			for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
-				if (adev->vcn.harvest_config & (1 << j))
-					continue;
-				for (k = 0; k < adev->vcn.num_enc_rings; ++k)
-					rings[num_rings++] = &adev->vcn.inst[j].ring_enc[k];
-			}
+			scheds = adev->vcn.vcn_enc_sched;
+			num_scheds =  adev->vcn.num_vcn_enc_sched;
 			break;
 		case AMDGPU_HW_IP_VCN_JPEG:
-			for (j = 0; j < adev->jpeg.num_jpeg_inst; ++j) {
-				if (adev->jpeg.harvest_config & (1 << j))
-					continue;
-				rings[num_rings++] = &adev->jpeg.inst[j].ring_dec;
-			}
+			scheds = adev->jpeg.jpeg_sched;
+			num_scheds =  adev->jpeg.num_jpeg_sched;
 			break;
 		}
 
-		for (j = 0; j < num_rings; ++j) {
-			if (!rings[j]->adev)
-				continue;
-
-			rqs[num_rqs++] = &rings[j]->sched.sched_rq[priority];
-		}
-
 		for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j)
 			r = drm_sched_entity_init(&ctx->entities[i][j].entity,
-						  rqs, num_rqs, &ctx->guilty);
+						  priority, scheds,
+						  num_scheds, &ctx->guilty);
 		if (r)
 			goto error_cleanup_entities;
 	}
@@ -627,3 +612,45 @@ void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr)
 	idr_destroy(&mgr->ctx_handles);
 	mutex_destroy(&mgr->lock);
 }
+
+void amdgpu_ctx_init_sched(struct amdgpu_device *adev)
+{
+	int i, j;
+
+	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
+		adev->gfx.gfx_sched[i] = &adev->gfx.gfx_ring[i].sched;
+		adev->gfx.num_gfx_sched++;
+	}
+
+	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
+		adev->gfx.compute_sched[i] = &adev->gfx.compute_ring[i].sched;
+		adev->gfx.num_compute_sched++;
+	}
+
+	for (i = 0; i < adev->sdma.num_instances; i++) {
+		adev->sdma.sdma_sched[i] = &adev->sdma.instance[i].ring.sched;
+		adev->sdma.num_sdma_sched++;
+	}
+
+	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+		if (adev->vcn.harvest_config & (1 << i))
+			continue;
+		adev->vcn.vcn_dec_sched[adev->vcn.num_vcn_dec_sched++] =
+			&adev->vcn.inst[i].ring_dec.sched;
+	}
+
+	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+		if (adev->vcn.harvest_config & (1 << i))
+			continue;
+		for (j = 0; j < adev->vcn.num_enc_rings; ++j)
+			adev->vcn.vcn_enc_sched[adev->vcn.num_vcn_enc_sched++] =
+				&adev->vcn.inst[i].ring_enc[j].sched;
+	}
+
+	for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
+		if (adev->jpeg.harvest_config & (1 << i))
+			continue;
+		adev->jpeg.jpeg_sched[adev->jpeg.num_jpeg_sched++] =
+			&adev->jpeg.inst[i].ring_dec.sched;
+	}
+}

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

@@ -87,4 +87,7 @@ void amdgpu_ctx_mgr_entity_fini(struct amdgpu_ctx_mgr *mgr);
 long amdgpu_ctx_mgr_entity_flush(struct amdgpu_ctx_mgr *mgr, long timeout);
 void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr);
 
+void amdgpu_ctx_init_sched(struct amdgpu_device *adev);
+
+
 #endif