Merge tag 'amd-drm-next-6.7-2023-11-03' of https://gitlab.freedesktop…

….org/agd5f/linux into drm-next

amd-drm-next-6.7-2023-11-03:

amdgpu:
- Fix RAS support check
- RAS fixes
- MES fixes
- SMU13 fixes
- Contiguous memory allocation fix
- BACO fixes
- GPU reset fixes
- Min power limit fixes
- GFX11 fixes
- USB4/TB hotplug fixes
- ARM regression fix
- GFX9.4.3 fixes
- KASAN/KCSAN stack size check fixes
- SR-IOV fixes
- SMU14 fixes
- PSP13 fixes
- Display blend fixes
- Flexible array size fixes

amdkfd:
- GPUVM fix

radeon:
- Flexible array size fixes

Signed-off-by: Dave Airlie <airlied@redhat.com>

From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231103173203.4912-1-alexander.deucher@amd.com

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

@@ -363,9 +363,6 @@ struct amdgpu_ip_block_version {
 	const struct amd_ip_funcs *funcs;
 };
 
-#define HW_REV(_Major, _Minor, _Rev) \
-	((((uint32_t) (_Major)) << 16) | ((uint32_t) (_Minor) << 8) | ((uint32_t) (_Rev)))
-
 struct amdgpu_ip_block {
 	struct amdgpu_ip_block_status status;
 	const struct amdgpu_ip_block_version *version;

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

@@ -425,6 +425,32 @@ static int amdgpu_amdkfd_bo_validate(struct amdgpu_bo *bo, uint32_t domain,
 	return ret;
 }
 
+static int amdgpu_amdkfd_bo_validate_and_fence(struct amdgpu_bo *bo,
+					       uint32_t domain,
+					       struct dma_fence *fence)
+{
+	int ret = amdgpu_bo_reserve(bo, false);
+
+	if (ret)
+		return ret;
+
+	ret = amdgpu_amdkfd_bo_validate(bo, domain, true);
+	if (ret)
+		goto unreserve_out;
+
+	ret = dma_resv_reserve_fences(bo->tbo.base.resv, 1);
+	if (ret)
+		goto unreserve_out;
+
+	dma_resv_add_fence(bo->tbo.base.resv, fence,
+			   DMA_RESV_USAGE_BOOKKEEP);
+
+unreserve_out:
+	amdgpu_bo_unreserve(bo);
+
+	return ret;
+}
+
 static int amdgpu_amdkfd_validate_vm_bo(void *_unused, struct amdgpu_bo *bo)
 {
 	return amdgpu_amdkfd_bo_validate(bo, bo->allowed_domains, false);
@@ -1784,6 +1810,15 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 		}
 		bo->allowed_domains = AMDGPU_GEM_DOMAIN_GTT;
 		bo->preferred_domains = AMDGPU_GEM_DOMAIN_GTT;
+	} else {
+		mutex_lock(&avm->process_info->lock);
+		if (avm->process_info->eviction_fence &&
+		    !dma_fence_is_signaled(&avm->process_info->eviction_fence->base))
+			ret = amdgpu_amdkfd_bo_validate_and_fence(bo, domain,
+				&avm->process_info->eviction_fence->base);
+		mutex_unlock(&avm->process_info->lock);
+		if (ret)
+			goto err_validate_bo;
 	}
 
 	if (offset)
@@ -1793,6 +1828,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 
 allocate_init_user_pages_failed:
 err_pin_bo:
+err_validate_bo:
 	remove_kgd_mem_from_kfd_bo_list(*mem, avm->process_info);
 	drm_vma_node_revoke(&gobj->vma_node, drm_priv);
 err_node_allow:
@@ -1866,10 +1902,6 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
 	if (unlikely(ret))
 		return ret;
 
-	/* The eviction fence should be removed by the last unmap.
-	 * TODO: Log an error condition if the bo still has the eviction fence
-	 * attached
-	 */
 	amdgpu_amdkfd_remove_eviction_fence(mem->bo,
 					process_info->eviction_fence);
 	pr_debug("Release VA 0x%llx - 0x%llx\n", mem->va,
@@ -1998,19 +2030,6 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 	if (unlikely(ret))
 		goto out_unreserve;
 
-	if (mem->mapped_to_gpu_memory == 0 &&
-	    !amdgpu_ttm_tt_get_usermm(bo->tbo.ttm)) {
-		/* Validate BO only once. The eviction fence gets added to BO
-		 * the first time it is mapped. Validate will wait for all
-		 * background evictions to complete.
-		 */
-		ret = amdgpu_amdkfd_bo_validate(bo, domain, true);
-		if (ret) {
-			pr_debug("Validate failed\n");
-			goto out_unreserve;
-		}
-	}
-
 	list_for_each_entry(entry, &mem->attachments, list) {
 		if (entry->bo_va->base.vm != avm || entry->is_mapped)
 			continue;
@@ -2037,10 +2056,6 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 			 mem->mapped_to_gpu_memory);
 	}
 
-	if (!amdgpu_ttm_tt_get_usermm(bo->tbo.ttm) && !bo->tbo.pin_count)
-		dma_resv_add_fence(bo->tbo.base.resv,
-				   &avm->process_info->eviction_fence->base,
-				   DMA_RESV_USAGE_BOOKKEEP);
 	ret = unreserve_bo_and_vms(&ctx, false, false);
 
 	goto out;
@@ -2074,7 +2089,6 @@ int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
 		struct amdgpu_device *adev, struct kgd_mem *mem, void *drm_priv)
 {
 	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
-	struct amdkfd_process_info *process_info = avm->process_info;
 	unsigned long bo_size = mem->bo->tbo.base.size;
 	struct kfd_mem_attachment *entry;
 	struct bo_vm_reservation_context ctx;
@@ -2115,15 +2129,6 @@ int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
 			 mem->mapped_to_gpu_memory);
 	}
 
-	/* If BO is unmapped from all VMs, unfence it. It can be evicted if
-	 * required.
-	 */
-	if (mem->mapped_to_gpu_memory == 0 &&
-	    !amdgpu_ttm_tt_get_usermm(mem->bo->tbo.ttm) &&
-	    !mem->bo->tbo.pin_count)
-		amdgpu_amdkfd_remove_eviction_fence(mem->bo,
-						process_info->eviction_fence);
-
 unreserve_out:
 	unreserve_bo_and_vms(&ctx, false, false);
 out:
@@ -2351,8 +2356,20 @@ int amdgpu_amdkfd_gpuvm_import_dmabuf(struct amdgpu_device *adev,
 	amdgpu_sync_create(&(*mem)->sync);
 	(*mem)->is_imported = true;
 
+	mutex_lock(&avm->process_info->lock);
+	if (avm->process_info->eviction_fence &&
+	    !dma_fence_is_signaled(&avm->process_info->eviction_fence->base))
+		ret = amdgpu_amdkfd_bo_validate_and_fence(bo, (*mem)->domain,
+				&avm->process_info->eviction_fence->base);
+	mutex_unlock(&avm->process_info->lock);
+	if (ret)
+		goto err_remove_mem;
+
 	return 0;
 
+err_remove_mem:
+	remove_kgd_mem_from_kfd_bo_list(*mem, avm->process_info);
+	drm_vma_node_revoke(&obj->vma_node, drm_priv);
 err_free_mem:
 	kfree(*mem);
 err_put_obj:

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

@@ -29,6 +29,7 @@
 #include "amdgpu.h"
 #include "atom.h"
 
+#include <linux/device.h>
 #include <linux/pci.h>
 #include <linux/slab.h>
 #include <linux/acpi.h>
@@ -287,6 +288,10 @@ static bool amdgpu_atrm_get_bios(struct amdgpu_device *adev)
 	if (adev->flags & AMD_IS_APU)
 		return false;
 
+	/* ATRM is for on-platform devices only */
+	if (dev_is_removable(&adev->pdev->dev))
+		return false;
+
 	while ((pdev = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, pdev)) != NULL) {
 		dhandle = ACPI_HANDLE(&pdev->dev);
 		if (!dhandle)

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

@@ -1116,6 +1116,11 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 			return r;
 	}
 
+	/* FIXME: In theory this loop shouldn't be needed any more when
+	 * amdgpu_vm_handle_moved handles all moved BOs that are reserved
+	 * with p->ticket. But removing it caused test regressions, so I'm
+	 * leaving it here for now.
+	 */
 	amdgpu_bo_list_for_each_entry(e, p->bo_list) {
 		bo_va = e->bo_va;
 		if (bo_va == NULL)
@@ -1130,7 +1135,7 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 			return r;
 	}
 
-	r = amdgpu_vm_handle_moved(adev, vm);
+	r = amdgpu_vm_handle_moved(adev, vm, &p->exec.ticket);
 	if (r)
 		return r;
 

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

@@ -41,6 +41,7 @@
 #include <drm/drm_fb_helper.h>
 #include <drm/drm_probe_helper.h>
 #include <drm/amdgpu_drm.h>
+#include <linux/device.h>
 #include <linux/vgaarb.h>
 #include <linux/vga_switcheroo.h>
 #include <linux/efi.h>
@@ -1073,6 +1074,8 @@ static int amdgpu_device_asic_init(struct amdgpu_device *adev)
 	    amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(11, 0, 0)) {
 		amdgpu_psp_wait_for_bootloader(adev);
 		ret = amdgpu_atomfirmware_asic_init(adev, true);
+		/* TODO: check the return val and stop device initialization if boot fails */
+		amdgpu_psp_query_boot_status(adev);
 		return ret;
 	} else {
 		return amdgpu_atom_asic_init(adev->mode_info.atom_context);
@@ -2223,7 +2226,6 @@ static int amdgpu_device_parse_gpu_info_fw(struct amdgpu_device *adev)
  */
 static int amdgpu_device_ip_early_init(struct amdgpu_device *adev)
 {
-	struct drm_device *dev = adev_to_drm(adev);
 	struct pci_dev *parent;
 	int i, r;
 	bool total;
@@ -2294,7 +2296,7 @@ static int amdgpu_device_ip_early_init(struct amdgpu_device *adev)
 	    (amdgpu_is_atpx_hybrid() ||
 	     amdgpu_has_atpx_dgpu_power_cntl()) &&
 	    ((adev->flags & AMD_IS_APU) == 0) &&
-	    !pci_is_thunderbolt_attached(to_pci_dev(dev->dev)))
+	    !dev_is_removable(&adev->pdev->dev))
 		adev->flags |= AMD_IS_PX;
 
 	if (!(adev->flags & AMD_IS_APU)) {
@@ -3962,13 +3964,23 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 				}
 			}
 		} else {
-			tmp = amdgpu_reset_method;
-			/* It should do a default reset when loading or reloading the driver,
-			 * regardless of the module parameter reset_method.
-			 */
-			amdgpu_reset_method = AMD_RESET_METHOD_NONE;
-			r = amdgpu_asic_reset(adev);
-			amdgpu_reset_method = tmp;
+			switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) {
+			case IP_VERSION(13, 0, 0):
+			case IP_VERSION(13, 0, 7):
+			case IP_VERSION(13, 0, 10):
+				r = psp_gpu_reset(adev);
+				break;
+			default:
+				tmp = amdgpu_reset_method;
+				/* It should do a default reset when loading or reloading the driver,
+				 * regardless of the module parameter reset_method.
+				 */
+				amdgpu_reset_method = AMD_RESET_METHOD_NONE;
+				r = amdgpu_asic_reset(adev);
+				amdgpu_reset_method = tmp;
+				break;
+			}
+
 			if (r) {
 				dev_err(adev->dev, "asic reset on init failed\n");
 				goto failed;
@@ -4132,7 +4144,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 
 	px = amdgpu_device_supports_px(ddev);
 
-	if (px || (!pci_is_thunderbolt_attached(adev->pdev) &&
+	if (px || (!dev_is_removable(&adev->pdev->dev) &&
 				apple_gmux_detect(NULL, NULL)))
 		vga_switcheroo_register_client(adev->pdev,
 					       &amdgpu_switcheroo_ops, px);
@@ -4282,7 +4294,7 @@ void amdgpu_device_fini_sw(struct amdgpu_device *adev)
 
 	px = amdgpu_device_supports_px(adev_to_drm(adev));
 
-	if (px || (!pci_is_thunderbolt_attached(adev->pdev) &&
+	if (px || (!dev_is_removable(&adev->pdev->dev) &&
 				apple_gmux_detect(NULL, NULL)))
 		vga_switcheroo_unregister_client(adev->pdev);
 
@@ -5566,10 +5578,6 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
 			drm_sched_start(&ring->sched, true);
 		}
 
-		if (adev->enable_mes &&
-		    amdgpu_ip_version(adev, GC_HWIP, 0) != IP_VERSION(11, 0, 3))
-			amdgpu_mes_self_test(tmp_adev);
-
 		if (!drm_drv_uses_atomic_modeset(adev_to_drm(tmp_adev)) && !job_signaled)
 			drm_helper_resume_force_mode(adev_to_drm(tmp_adev));
 

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

@@ -99,6 +99,7 @@
 MODULE_FIRMWARE(FIRMWARE_IP_DISCOVERY);
 
 #define mmRCC_CONFIG_MEMSIZE	0xde3
+#define mmMP0_SMN_C2PMSG_33	0x16061
 #define mmMM_INDEX		0x0
 #define mmMM_INDEX_HI		0x6
 #define mmMM_DATA		0x1
@@ -239,8 +240,26 @@ static int amdgpu_discovery_read_binary_from_sysmem(struct amdgpu_device *adev,
 static int amdgpu_discovery_read_binary_from_mem(struct amdgpu_device *adev,
 						 uint8_t *binary)
 {
-	uint64_t vram_size = (uint64_t)RREG32(mmRCC_CONFIG_MEMSIZE) << 20;
-	int ret = 0;
+	uint64_t vram_size;
+	u32 msg;
+	int i, ret = 0;
+
+	/* It can take up to a second for IFWI init to complete on some dGPUs,
+	 * but generally it should be in the 60-100ms range.  Normally this starts
+	 * as soon as the device gets power so by the time the OS loads this has long
+	 * completed.  However, when a card is hotplugged via e.g., USB4, we need to
+	 * wait for this to complete.  Once the C2PMSG is updated, we can
+	 * continue.
+	 */
+	if (dev_is_removable(&adev->pdev->dev)) {
+		for (i = 0; i < 1000; i++) {
+			msg = RREG32(mmMP0_SMN_C2PMSG_33);
+			if (msg & 0x80000000)
+				break;
+			msleep(1);
+		}
+	}
+	vram_size = (uint64_t)RREG32(mmRCC_CONFIG_MEMSIZE) << 20;
 
 	if (vram_size) {
 		uint64_t pos = vram_size - DISCOVERY_TMR_OFFSET;
@@ -2449,6 +2468,9 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev)
 	if (amdgpu_ip_version(adev, XGMI_HWIP, 0) == IP_VERSION(4, 8, 0))
 		adev->gmc.xgmi.supported = true;
 
+	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3))
+		adev->ip_versions[XGMI_HWIP][0] = IP_VERSION(6, 4, 0);
+
 	/* set NBIO version */
 	switch (amdgpu_ip_version(adev, NBIO_HWIP, 0)) {
 	case IP_VERSION(6, 1, 0):

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

@@ -409,7 +409,7 @@ amdgpu_dma_buf_move_notify(struct dma_buf_attachment *attach)
 		if (!r)
 			r = amdgpu_vm_clear_freed(adev, vm, NULL);
 		if (!r)
-			r = amdgpu_vm_handle_moved(adev, vm);
+			r = amdgpu_vm_handle_moved(adev, vm, ticket);
 
 		if (r && r != -EBUSY)
 			DRM_ERROR("Failed to invalidate VM page tables (%d))\n",

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

@@ -2041,6 +2041,14 @@ static const struct pci_device_id pciidlist[] = {
 
 MODULE_DEVICE_TABLE(pci, pciidlist);
 
+static const struct amdgpu_asic_type_quirk asic_type_quirks[] = {
+	/* differentiate between P10 and P11 asics with the same DID */
+	{0x67FF, 0xE3, CHIP_POLARIS10},
+	{0x67FF, 0xE7, CHIP_POLARIS10},
+	{0x67FF, 0xF3, CHIP_POLARIS10},
+	{0x67FF, 0xF7, CHIP_POLARIS10},
+};
+
 static const struct drm_driver amdgpu_kms_driver;
 
 static void amdgpu_get_secondary_funcs(struct amdgpu_device *adev)
@@ -2083,6 +2091,22 @@ static void amdgpu_init_debug_options(struct amdgpu_device *adev)
 	}
 }
 
+static unsigned long amdgpu_fix_asic_type(struct pci_dev *pdev, unsigned long flags)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(asic_type_quirks); i++) {
+		if (pdev->device == asic_type_quirks[i].device &&
+			pdev->revision == asic_type_quirks[i].revision) {
+				flags &= ~AMD_ASIC_MASK;
+				flags |= asic_type_quirks[i].type;
+				break;
+			}
+	}
+
+	return flags;
+}
+
 static int amdgpu_pci_probe(struct pci_dev *pdev,
 			    const struct pci_device_id *ent)
 {
@@ -2110,15 +2134,8 @@ static int amdgpu_pci_probe(struct pci_dev *pdev,
 			 "See modparam exp_hw_support\n");
 		return -ENODEV;
 	}
-	/* differentiate between P10 and P11 asics with the same DID */
-	if (pdev->device == 0x67FF &&
-	    (pdev->revision == 0xE3 ||
-	     pdev->revision == 0xE7 ||
-	     pdev->revision == 0xF3 ||
-	     pdev->revision == 0xF7)) {
-		flags &= ~AMD_ASIC_MASK;
-		flags |= CHIP_POLARIS10;
-	}
+
+	flags = amdgpu_fix_asic_type(pdev, flags);
 
 	/* Due to hardware bugs, S/G Display on raven requires a 1:1 IOMMU mapping,
 	 * however, SME requires an indirect IOMMU mapping because the encryption