From f544afac3f34124088b981c63843a3cc48f4ee3e Mon Sep 17 00:00:00 2001
From: Amber Lin <Amber.Lin@amd.com>
Date: Tue, 17 May 2022 23:41:01 +0800
Subject: drm/amdgpu: Add kgd2kfd for GC 9.4.3

New GC (v9.4.3) and ATHUB (v1.8.0) versions
are used. Add kgd_gfx_v9_4_3_*
functions if registers in use of kgd_gfx_v9_*
functions are changed or have different offset.

Signed-off-by: Amber Lin <Amber.Lin@amd.com>
Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Mukul Joshi <mukul.joshi@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 00f528eb9812..1510041a6ee1 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -51,6 +51,7 @@ extern const struct kfd2kgd_calls gfx_v8_kfd2kgd;
 extern const struct kfd2kgd_calls gfx_v9_kfd2kgd;
 extern const struct kfd2kgd_calls arcturus_kfd2kgd;
 extern const struct kfd2kgd_calls aldebaran_kfd2kgd;
+extern const struct kfd2kgd_calls gc_9_4_3_kfd2kgd;
 extern const struct kfd2kgd_calls gfx_v10_kfd2kgd;
 extern const struct kfd2kgd_calls gfx_v10_3_kfd2kgd;
 extern const struct kfd2kgd_calls gfx_v11_kfd2kgd;
@@ -328,7 +329,7 @@ struct kfd_dev *kgd2kfd_probe(struct amdgpu_device *adev, bool vf)
 			break;
 		case IP_VERSION(9, 4, 3):
 			gfx_target_version = 90400;
-			f2g = &aldebaran_kfd2kgd;
+			f2g = &gc_9_4_3_kfd2kgd;
 			break;
 		/* Navi10 */
 		case IP_VERSION(10, 1, 10):
-- 
cgit 


From 8dc1db3172ae2f17ae71e33b608a33411ce8a1aa Mon Sep 17 00:00:00 2001
From: Mukul Joshi <mukul.joshi@amd.com>
Date: Wed, 14 Sep 2022 16:39:48 +0800
Subject: drm/amdkfd: Introduce kfd_node struct (v5)

Introduce a new structure, kfd_node, which will now represent
a compute node. kfd_node is carved out of kfd_dev structure.
kfd_dev struct now will become the parent of kfd_node, and will
store common resources such as doorbells, GTT sub-alloctor etc.
kfd_node struct will store all resources specific to a compute
node, such as device queue manager, interrupt handling etc.

This is the first step in adding compute partition support in KFD.

v2: introduce kfd_node struct to gc v11 (Hawking)
v3: make reference to kfd_dev struct through kfd_node (Morris)
v4: use kfd_node instead for kfd isr/mqd functions (Morris)
v5: rebase (Alex)

Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Tested-by: Amber Lin <Amber.Lin@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Morris Zhang <Shiwu.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 258 +++++++++++++++++++-------------
 1 file changed, 157 insertions(+), 101 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 1510041a6ee1..23d9a7f77055 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -61,7 +61,7 @@ static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size,
 static void kfd_gtt_sa_fini(struct kfd_dev *kfd);
 
 static int kfd_resume_iommu(struct kfd_dev *kfd);
-static int kfd_resume(struct kfd_dev *kfd);
+static int kfd_resume(struct kfd_node *kfd);
 
 static void kfd_device_info_set_sdma_info(struct kfd_dev *kfd)
 {
@@ -441,8 +441,6 @@ struct kfd_dev *kgd2kfd_probe(struct amdgpu_device *adev, bool vf)
 	memset(&kfd->doorbell_available_index, 0,
 		sizeof(kfd->doorbell_available_index));
 
-	atomic_set(&kfd->sram_ecc_flag, 0);
-
 	ida_init(&kfd->doorbell_ida);
 
 	return kfd;
@@ -489,41 +487,106 @@ static void kfd_cwsr_init(struct kfd_dev *kfd)
 	}
 }
 
-static int kfd_gws_init(struct kfd_dev *kfd)
+static int kfd_gws_init(struct kfd_node *node)
 {
 	int ret = 0;
+	struct kfd_dev *kfd = node->kfd;
 
-	if (kfd->dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS)
+	if (node->dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS)
 		return 0;
 
-	if (hws_gws_support || (KFD_IS_SOC15(kfd) &&
-		((KFD_GC_VERSION(kfd) == IP_VERSION(9, 0, 1)
+	if (hws_gws_support || (KFD_IS_SOC15(node) &&
+		((KFD_GC_VERSION(node) == IP_VERSION(9, 0, 1)
 			&& kfd->mec2_fw_version >= 0x81b3) ||
-		(KFD_GC_VERSION(kfd) <= IP_VERSION(9, 4, 0)
+		(KFD_GC_VERSION(node) <= IP_VERSION(9, 4, 0)
 			&& kfd->mec2_fw_version >= 0x1b3)  ||
-		(KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 1)
+		(KFD_GC_VERSION(node) == IP_VERSION(9, 4, 1)
 			&& kfd->mec2_fw_version >= 0x30)   ||
-		(KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 2)
+		(KFD_GC_VERSION(node) == IP_VERSION(9, 4, 2)
 			&& kfd->mec2_fw_version >= 0x28) ||
-		(KFD_GC_VERSION(kfd) >= IP_VERSION(10, 3, 0)
-			&& KFD_GC_VERSION(kfd) < IP_VERSION(11, 0, 0)
+		(KFD_GC_VERSION(node) >= IP_VERSION(10, 3, 0)
+			&& KFD_GC_VERSION(node) < IP_VERSION(11, 0, 0)
 			&& kfd->mec2_fw_version >= 0x6b))))
-		ret = amdgpu_amdkfd_alloc_gws(kfd->adev,
-				kfd->adev->gds.gws_size, &kfd->gws);
+		ret = amdgpu_amdkfd_alloc_gws(node->adev,
+				node->adev->gds.gws_size, &node->gws);
 
 	return ret;
 }
 
-static void kfd_smi_init(struct kfd_dev *dev)
+static void kfd_smi_init(struct kfd_node *dev)
 {
 	INIT_LIST_HEAD(&dev->smi_clients);
 	spin_lock_init(&dev->smi_lock);
 }
 
+static int kfd_init_node(struct kfd_node *node)
+{
+	int err = -1;
+
+	if (kfd_interrupt_init(node)) {
+		dev_err(kfd_device, "Error initializing interrupts\n");
+		goto kfd_interrupt_error;
+	}
+
+	node->dqm = device_queue_manager_init(node);
+	if (!node->dqm) {
+		dev_err(kfd_device, "Error initializing queue manager\n");
+		goto device_queue_manager_error;
+	}
+
+	if (kfd_gws_init(node)) {
+		dev_err(kfd_device, "Could not allocate %d gws\n",
+			node->adev->gds.gws_size);
+		goto gws_error;
+	}
+
+	if (kfd_resume(node))
+		goto kfd_resume_error;
+
+	if (kfd_topology_add_device(node)) {
+		dev_err(kfd_device, "Error adding device to topology\n");
+		goto kfd_topology_add_device_error;
+	}
+
+	kfd_smi_init(node);
+
+	return 0;
+
+kfd_topology_add_device_error:
+kfd_resume_error:
+gws_error:
+	device_queue_manager_uninit(node->dqm);
+device_queue_manager_error:
+	kfd_interrupt_exit(node);
+kfd_interrupt_error:
+	if (node->gws)
+		amdgpu_amdkfd_free_gws(node->adev, node->gws);
+
+	/* Cleanup the node memory here */
+	kfree(node);
+	return err;
+}
+
+static void kfd_cleanup_node(struct kfd_dev *kfd)
+{
+	struct kfd_node *knode = kfd->node;
+
+	device_queue_manager_uninit(knode->dqm);
+	kfd_interrupt_exit(knode);
+	kfd_topology_remove_device(knode);
+	if (knode->gws)
+		amdgpu_amdkfd_free_gws(knode->adev, knode->gws);
+	kfree(knode);
+	kfd->node = NULL;
+}
+
 bool kgd2kfd_device_init(struct kfd_dev *kfd,
 			 const struct kgd2kfd_shared_resources *gpu_resources)
 {
 	unsigned int size, map_process_packet_size;
+	struct kfd_node *node;
+	uint32_t first_vmid_kfd, last_vmid_kfd, vmid_num_kfd;
+	unsigned int max_proc_per_quantum;
 
 	kfd->mec_fw_version = amdgpu_amdkfd_get_fw_version(kfd->adev,
 			KGD_ENGINE_MEC1);
@@ -533,10 +596,9 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 			KGD_ENGINE_SDMA1);
 	kfd->shared_resources = *gpu_resources;
 
-	kfd->vm_info.first_vmid_kfd = ffs(gpu_resources->compute_vmid_bitmap)-1;
-	kfd->vm_info.last_vmid_kfd = fls(gpu_resources->compute_vmid_bitmap)-1;
-	kfd->vm_info.vmid_num_kfd = kfd->vm_info.last_vmid_kfd
-			- kfd->vm_info.first_vmid_kfd + 1;
+	first_vmid_kfd = ffs(gpu_resources->compute_vmid_bitmap)-1;
+	last_vmid_kfd = fls(gpu_resources->compute_vmid_bitmap)-1;
+	vmid_num_kfd = last_vmid_kfd - first_vmid_kfd + 1;
 
 	/* Allow BIF to recode atomics to PCIe 3.0 AtomicOps.
 	 * 32 and 64-bit requests are possible and must be
@@ -557,9 +619,9 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 
 	/* Verify module parameters regarding mapped process number*/
 	if (hws_max_conc_proc >= 0)
-		kfd->max_proc_per_quantum = min((u32)hws_max_conc_proc, kfd->vm_info.vmid_num_kfd);
+		max_proc_per_quantum = min((u32)hws_max_conc_proc, vmid_num_kfd);
 	else
-		kfd->max_proc_per_quantum = kfd->vm_info.vmid_num_kfd;
+		max_proc_per_quantum = vmid_num_kfd;
 
 	/* calculate max size of mqds needed for queues */
 	size = max_num_of_queues_per_device *
@@ -609,26 +671,6 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 
 	kfd->noretry = kfd->adev->gmc.noretry;
 
-	if (kfd_interrupt_init(kfd)) {
-		dev_err(kfd_device, "Error initializing interrupts\n");
-		goto kfd_interrupt_error;
-	}
-
-	kfd->dqm = device_queue_manager_init(kfd);
-	if (!kfd->dqm) {
-		dev_err(kfd_device, "Error initializing queue manager\n");
-		goto device_queue_manager_error;
-	}
-
-	/* If supported on this device, allocate global GWS that is shared
-	 * by all KFD processes
-	 */
-	if (kfd_gws_init(kfd)) {
-		dev_err(kfd_device, "Could not allocate %d gws\n",
-			kfd->adev->gds.gws_size);
-		goto gws_error;
-	}
-
 	/* If CRAT is broken, won't set iommu enabled */
 	kfd_double_confirm_iommu_support(kfd);
 
@@ -642,46 +684,54 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 
 	svm_migrate_init(kfd->adev);
 
-	if (kfd_resume_iommu(kfd))
-		goto device_iommu_error;
-
-	if (kfd_resume(kfd))
-		goto kfd_resume_error;
-
-	amdgpu_amdkfd_get_local_mem_info(kfd->adev, &kfd->local_mem_info);
+	/* Allocate the KFD node */
+	node = kzalloc(sizeof(struct kfd_node), GFP_KERNEL);
+	if (!node) {
+		dev_err(kfd_device, "Error allocating KFD node\n");
+		goto node_alloc_error;
+	}
 
-	if (kfd_topology_add_device(kfd)) {
-		dev_err(kfd_device, "Error adding device to topology\n");
-		goto kfd_topology_add_device_error;
+	node->adev = kfd->adev;
+	node->kfd = kfd;
+	node->kfd2kgd = kfd->kfd2kgd;
+	node->vm_info.vmid_num_kfd = vmid_num_kfd;
+	node->vm_info.first_vmid_kfd = first_vmid_kfd;
+	node->vm_info.last_vmid_kfd = last_vmid_kfd;
+	node->max_proc_per_quantum = max_proc_per_quantum;
+	atomic_set(&node->sram_ecc_flag, 0);
+
+	/* Initialize the KFD node */
+	if (kfd_init_node(node)) {
+		dev_err(kfd_device, "Error initializing KFD node\n");
+		goto node_init_error;
 	}
+	kfd->node = node;
 
-	kfd_smi_init(kfd);
+	if (kfd_resume_iommu(kfd))
+		goto kfd_resume_iommu_error;
+
+	amdgpu_amdkfd_get_local_mem_info(kfd->adev, &kfd->local_mem_info);
 
 	kfd->init_complete = true;
 	dev_info(kfd_device, "added device %x:%x\n", kfd->adev->pdev->vendor,
 		 kfd->adev->pdev->device);
 
 	pr_debug("Starting kfd with the following scheduling policy %d\n",
-		kfd->dqm->sched_policy);
+		node->dqm->sched_policy);
 
 	goto out;
 
-kfd_topology_add_device_error:
-kfd_resume_error:
+kfd_resume_iommu_error:
+	kfd_cleanup_node(kfd);
+node_init_error:
+node_alloc_error:
 device_iommu_error:
-gws_error:
-	device_queue_manager_uninit(kfd->dqm);
-device_queue_manager_error:
-	kfd_interrupt_exit(kfd);
-kfd_interrupt_error:
 	kfd_doorbell_fini(kfd);
 kfd_doorbell_error:
 	kfd_gtt_sa_fini(kfd);
 kfd_gtt_sa_init_error:
 	amdgpu_amdkfd_free_gtt_mem(kfd->adev, kfd->gtt_mem);
 alloc_gtt_mem_failure:
-	if (kfd->gws)
-		amdgpu_amdkfd_free_gws(kfd->adev, kfd->gws);
 	dev_err(kfd_device,
 		"device %x:%x NOT added due to errors\n",
 		kfd->adev->pdev->vendor, kfd->adev->pdev->device);
@@ -692,15 +742,11 @@ out:
 void kgd2kfd_device_exit(struct kfd_dev *kfd)
 {
 	if (kfd->init_complete) {
-		device_queue_manager_uninit(kfd->dqm);
-		kfd_interrupt_exit(kfd);
-		kfd_topology_remove_device(kfd);
+		kfd_cleanup_node(kfd);
 		kfd_doorbell_fini(kfd);
 		ida_destroy(&kfd->doorbell_ida);
 		kfd_gtt_sa_fini(kfd);
 		amdgpu_amdkfd_free_gtt_mem(kfd->adev, kfd->gtt_mem);
-		if (kfd->gws)
-			amdgpu_amdkfd_free_gws(kfd->adev, kfd->gws);
 	}
 
 	kfree(kfd);
@@ -708,16 +754,18 @@ void kgd2kfd_device_exit(struct kfd_dev *kfd)
 
 int kgd2kfd_pre_reset(struct kfd_dev *kfd)
 {
+	struct kfd_node *node = kfd->node;
+
 	if (!kfd->init_complete)
 		return 0;
 
-	kfd_smi_event_update_gpu_reset(kfd, false);
+	kfd_smi_event_update_gpu_reset(node, false);
 
-	kfd->dqm->ops.pre_reset(kfd->dqm);
+	node->dqm->ops.pre_reset(node->dqm);
 
 	kgd2kfd_suspend(kfd, false);
 
-	kfd_signal_reset_event(kfd);
+	kfd_signal_reset_event(node);
 	return 0;
 }
 
@@ -730,18 +778,19 @@ int kgd2kfd_pre_reset(struct kfd_dev *kfd)
 int kgd2kfd_post_reset(struct kfd_dev *kfd)
 {
 	int ret;
+	struct kfd_node *node = kfd->node;
 
 	if (!kfd->init_complete)
 		return 0;
 
-	ret = kfd_resume(kfd);
+	ret = kfd_resume(node);
 	if (ret)
 		return ret;
 	atomic_dec(&kfd_locked);
 
-	atomic_set(&kfd->sram_ecc_flag, 0);
+	atomic_set(&node->sram_ecc_flag, 0);
 
-	kfd_smi_event_update_gpu_reset(kfd, true);
+	kfd_smi_event_update_gpu_reset(node, true);
 
 	return 0;
 }
@@ -753,6 +802,8 @@ bool kfd_is_locked(void)
 
 void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm)
 {
+	struct kfd_node *node = kfd->node;
+
 	if (!kfd->init_complete)
 		return;
 
@@ -763,18 +814,19 @@ void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm)
 			kfd_suspend_all_processes();
 	}
 
-	kfd->dqm->ops.stop(kfd->dqm);
+	node->dqm->ops.stop(node->dqm);
 	kfd_iommu_suspend(kfd);
 }
 
 int kgd2kfd_resume(struct kfd_dev *kfd, bool run_pm)
 {
 	int ret, count;
+	struct kfd_node *node = kfd->node;
 
 	if (!kfd->init_complete)
 		return 0;
 
-	ret = kfd_resume(kfd);
+	ret = kfd_resume(node);
 	if (ret)
 		return ret;
 
@@ -809,15 +861,15 @@ static int kfd_resume_iommu(struct kfd_dev *kfd)
 	return err;
 }
 
-static int kfd_resume(struct kfd_dev *kfd)
+static int kfd_resume(struct kfd_node *node)
 {
 	int err = 0;
 
-	err = kfd->dqm->ops.start(kfd->dqm);
+	err = node->dqm->ops.start(node->dqm);
 	if (err)
 		dev_err(kfd_device,
 			"Error starting queue manager for device %x:%x\n",
-			kfd->adev->pdev->vendor, kfd->adev->pdev->device);
+			node->adev->pdev->vendor, node->adev->pdev->device);
 
 	return err;
 }
@@ -843,6 +895,7 @@ void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry)
 	uint32_t patched_ihre[KFD_MAX_RING_ENTRY_SIZE];
 	bool is_patched = false;
 	unsigned long flags;
+	struct kfd_node *node = kfd->node;
 
 	if (!kfd->init_complete)
 		return;
@@ -852,16 +905,16 @@ void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry)
 		return;
 	}
 
-	spin_lock_irqsave(&kfd->interrupt_lock, flags);
+	spin_lock_irqsave(&node->interrupt_lock, flags);
 
-	if (kfd->interrupts_active
-	    && interrupt_is_wanted(kfd, ih_ring_entry,
+	if (node->interrupts_active
+	    && interrupt_is_wanted(node, ih_ring_entry,
 				   patched_ihre, &is_patched)
-	    && enqueue_ih_ring_entry(kfd,
+	    && enqueue_ih_ring_entry(node,
 				     is_patched ? patched_ihre : ih_ring_entry))
-		kfd_queue_work(kfd->ih_wq, &kfd->interrupt_work);
+		kfd_queue_work(node->ih_wq, &node->interrupt_work);
 
-	spin_unlock_irqrestore(&kfd->interrupt_lock, flags);
+	spin_unlock_irqrestore(&node->interrupt_lock, flags);
 }
 
 int kgd2kfd_quiesce_mm(struct mm_struct *mm, uint32_t trigger)
@@ -999,10 +1052,11 @@ static inline uint32_t *kfd_gtt_sa_calc_cpu_addr(void *start_addr,
 	return (uint32_t *) ((uint64_t) start_addr + bit_num * chunk_size);
 }
 
-int kfd_gtt_sa_allocate(struct kfd_dev *kfd, unsigned int size,
+int kfd_gtt_sa_allocate(struct kfd_node *node, unsigned int size,
 			struct kfd_mem_obj **mem_obj)
 {
 	unsigned int found, start_search, cur_size;
+	struct kfd_dev *kfd = node->kfd;
 
 	if (size == 0)
 		return -EINVAL;
@@ -1102,8 +1156,10 @@ kfd_gtt_no_free_chunk:
 	return -ENOMEM;
 }
 
-int kfd_gtt_sa_free(struct kfd_dev *kfd, struct kfd_mem_obj *mem_obj)
+int kfd_gtt_sa_free(struct kfd_node *node, struct kfd_mem_obj *mem_obj)
 {
+	struct kfd_dev *kfd = node->kfd;
+
 	/* Act like kfree when trying to free a NULL object */
 	if (!mem_obj)
 		return 0;
@@ -1126,28 +1182,28 @@ int kfd_gtt_sa_free(struct kfd_dev *kfd, struct kfd_mem_obj *mem_obj)
 void kgd2kfd_set_sram_ecc_flag(struct kfd_dev *kfd)
 {
 	if (kfd)
-		atomic_inc(&kfd->sram_ecc_flag);
+		atomic_inc(&kfd->node->sram_ecc_flag);
 }
 
-void kfd_inc_compute_active(struct kfd_dev *kfd)
+void kfd_inc_compute_active(struct kfd_node *node)
 {
-	if (atomic_inc_return(&kfd->compute_profile) == 1)
-		amdgpu_amdkfd_set_compute_idle(kfd->adev, false);
+	if (atomic_inc_return(&node->kfd->compute_profile) == 1)
+		amdgpu_amdkfd_set_compute_idle(node->adev, false);
 }
 
-void kfd_dec_compute_active(struct kfd_dev *kfd)
+void kfd_dec_compute_active(struct kfd_node *node)
 {
-	int count = atomic_dec_return(&kfd->compute_profile);
+	int count = atomic_dec_return(&node->kfd->compute_profile);
 
 	if (count == 0)
-		amdgpu_amdkfd_set_compute_idle(kfd->adev, true);
+		amdgpu_amdkfd_set_compute_idle(node->adev, true);
 	WARN_ONCE(count < 0, "Compute profile ref. count error");
 }
 
 void kgd2kfd_smi_event_throttle(struct kfd_dev *kfd, uint64_t throttle_bitmask)
 {
 	if (kfd && kfd->init_complete)
-		kfd_smi_event_update_thermal_throttling(kfd, throttle_bitmask);
+		kfd_smi_event_update_thermal_throttling(kfd->node, throttle_bitmask);
 }
 
 /* kfd_get_num_sdma_engines returns the number of PCIe optimized SDMA and
@@ -1155,19 +1211,19 @@ void kgd2kfd_smi_event_throttle(struct kfd_dev *kfd, uint64_t throttle_bitmask)
  * When the device has more than two engines, we reserve two for PCIe to enable
  * full-duplex and the rest are used as XGMI.
  */
-unsigned int kfd_get_num_sdma_engines(struct kfd_dev *kdev)
+unsigned int kfd_get_num_sdma_engines(struct kfd_node *node)
 {
 	/* If XGMI is not supported, all SDMA engines are PCIe */
-	if (!kdev->adev->gmc.xgmi.supported)
-		return kdev->adev->sdma.num_instances;
+	if (!node->adev->gmc.xgmi.supported)
+		return node->adev->sdma.num_instances;
 
-	return min(kdev->adev->sdma.num_instances, 2);
+	return min(node->adev->sdma.num_instances, 2);
 }
 
-unsigned int kfd_get_num_xgmi_sdma_engines(struct kfd_dev *kdev)
+unsigned int kfd_get_num_xgmi_sdma_engines(struct kfd_node *node)
 {
 	/* After reserved for PCIe, the rest of engines are XGMI */
-	return kdev->adev->sdma.num_instances - kfd_get_num_sdma_engines(kdev);
+	return node->adev->sdma.num_instances - kfd_get_num_sdma_engines(node);
 }
 
 #if defined(CONFIG_DEBUG_FS)
@@ -1175,7 +1231,7 @@ unsigned int kfd_get_num_xgmi_sdma_engines(struct kfd_dev *kdev)
 /* This function will send a package to HIQ to hang the HWS
  * which will trigger a GPU reset and bring the HWS back to normal state
  */
-int kfd_debugfs_hang_hws(struct kfd_dev *dev)
+int kfd_debugfs_hang_hws(struct kfd_node *dev)
 {
 	if (dev->dqm->sched_policy != KFD_SCHED_POLICY_HWS) {
 		pr_err("HWS is not enabled");
-- 
cgit 


From 74c5b85da75475c73a8f040397610fbfcc2c3e78 Mon Sep 17 00:00:00 2001
From: Mukul Joshi <mukul.joshi@amd.com>
Date: Mon, 9 May 2022 16:33:38 -0400
Subject: drm/amdkfd: Add spatial partitioning support in KFD

This patch introduces multi-partition support in KFD.
This patch includes:
- Support for maximum 8 spatial partitions in KFD.
- Initialize one HIQ per partition.
- Management of VMID range depending on partition mode.
- Management of doorbell aperture space between all
  partitions.
- Each partition does its own queue management, interrupt
  handling, SMI event reporting.
- IOMMU, if enabled with multiple partitions, will only work
  on first partition.
- SPM is only supported on the first partition.
- Currently, there is no support for resetting individual
  partitions. All partitions will reset together.

Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Tested-by: Amber Lin <Amber.Lin@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 214 ++++++++++++++++++++++++--------
 1 file changed, 161 insertions(+), 53 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 23d9a7f77055..37c6dc5c37bf 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -567,23 +567,27 @@ kfd_interrupt_error:
 	return err;
 }
 
-static void kfd_cleanup_node(struct kfd_dev *kfd)
+static void kfd_cleanup_nodes(struct kfd_dev *kfd, unsigned int num_nodes)
 {
-	struct kfd_node *knode = kfd->node;
-
-	device_queue_manager_uninit(knode->dqm);
-	kfd_interrupt_exit(knode);
-	kfd_topology_remove_device(knode);
-	if (knode->gws)
-		amdgpu_amdkfd_free_gws(knode->adev, knode->gws);
-	kfree(knode);
-	kfd->node = NULL;
+	struct kfd_node *knode;
+	unsigned int i;
+
+	for (i = 0; i < num_nodes; i++) {
+		knode = kfd->nodes[i];
+		device_queue_manager_uninit(knode->dqm);
+		kfd_interrupt_exit(knode);
+		kfd_topology_remove_device(knode);
+		if (knode->gws)
+			amdgpu_amdkfd_free_gws(knode->adev, knode->gws);
+		kfree(knode);
+		kfd->nodes[i] = NULL;
+	}
 }
 
 bool kgd2kfd_device_init(struct kfd_dev *kfd,
 			 const struct kgd2kfd_shared_resources *gpu_resources)
 {
-	unsigned int size, map_process_packet_size;
+	unsigned int size, map_process_packet_size, i;
 	struct kfd_node *node;
 	uint32_t first_vmid_kfd, last_vmid_kfd, vmid_num_kfd;
 	unsigned int max_proc_per_quantum;
@@ -596,9 +600,18 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 			KGD_ENGINE_SDMA1);
 	kfd->shared_resources = *gpu_resources;
 
-	first_vmid_kfd = ffs(gpu_resources->compute_vmid_bitmap)-1;
-	last_vmid_kfd = fls(gpu_resources->compute_vmid_bitmap)-1;
-	vmid_num_kfd = last_vmid_kfd - first_vmid_kfd + 1;
+	if (kfd->adev->gfx.num_xcd == 0 || kfd->adev->gfx.num_xcd == 1 ||
+	    kfd->adev->gfx.num_xcc_per_xcp == 0)
+		kfd->num_nodes = 1;
+	else
+		kfd->num_nodes =
+			kfd->adev->gfx.num_xcd/kfd->adev->gfx.num_xcc_per_xcp;
+	if (kfd->num_nodes == 0) {
+		dev_err(kfd_device,
+			"KFD num nodes cannot be 0, GC inst: %d, num_xcc_in_node: %d\n",
+			kfd->adev->gfx.num_xcd, kfd->adev->gfx.num_xcc_per_xcp);
+		goto out;
+	}
 
 	/* Allow BIF to recode atomics to PCIe 3.0 AtomicOps.
 	 * 32 and 64-bit requests are possible and must be
@@ -617,6 +630,26 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		return false;
 	}
 
+	first_vmid_kfd = ffs(gpu_resources->compute_vmid_bitmap)-1;
+	last_vmid_kfd = fls(gpu_resources->compute_vmid_bitmap)-1;
+	vmid_num_kfd = last_vmid_kfd - first_vmid_kfd + 1;
+
+	/* For GFX9.4.3, we need special handling for VMIDs depending on
+	 * partition mode.
+	 * In CPX mode, the VMID range needs to be shared between XCDs.
+	 * Additionally, there are 13 VMIDs (3-15) available for KFD. To
+	 * divide them equally, we change starting VMID to 4 and not use
+	 * VMID 3.
+	 * If the VMID range changes for GFX9.4.3, then this code MUST be
+	 * revisited.
+	 */
+	if (KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 3) &&
+	    kfd->adev->gfx.partition_mode == AMDGPU_CPX_PARTITION_MODE &&
+	    kfd->num_nodes != 1) {
+		vmid_num_kfd /= 2;
+		first_vmid_kfd = last_vmid_kfd + 1 - vmid_num_kfd*2;
+	}
+
 	/* Verify module parameters regarding mapped process number*/
 	if (hws_max_conc_proc >= 0)
 		max_proc_per_quantum = min((u32)hws_max_conc_proc, vmid_num_kfd);
@@ -682,6 +715,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 
 	kfd_cwsr_init(kfd);
 
+	/* TODO: Needs to be updated for memory partitioning */
 	svm_migrate_init(kfd->adev);
 
 	/* Allocate the KFD node */
@@ -700,12 +734,51 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	node->max_proc_per_quantum = max_proc_per_quantum;
 	atomic_set(&node->sram_ecc_flag, 0);
 
-	/* Initialize the KFD node */
-	if (kfd_init_node(node)) {
-		dev_err(kfd_device, "Error initializing KFD node\n");
-		goto node_init_error;
+	dev_info(kfd_device, "Total number of KFD nodes to be created: %d\n",
+				kfd->num_nodes);
+	for (i = 0; i < kfd->num_nodes; i++) {
+		node = kzalloc(sizeof(struct kfd_node), GFP_KERNEL);
+		if (!node)
+			goto node_alloc_error;
+
+		node->adev = kfd->adev;
+		node->kfd = kfd;
+		node->kfd2kgd = kfd->kfd2kgd;
+		node->vm_info.vmid_num_kfd = vmid_num_kfd;
+		node->num_xcc_per_node = max(1U, kfd->adev->gfx.num_xcc_per_xcp);
+		node->start_xcc_id = node->num_xcc_per_node * i;
+
+		if (KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 3) &&
+		    kfd->adev->gfx.partition_mode == AMDGPU_CPX_PARTITION_MODE &&
+		    kfd->num_nodes != 1) {
+			/* For GFX9.4.3 and CPX mode, first XCD gets VMID range
+			 * 4-9 and second XCD gets VMID range 10-15.
+			 */
+
+			node->vm_info.first_vmid_kfd = (i%2 == 0) ?
+						first_vmid_kfd :
+						first_vmid_kfd+vmid_num_kfd;
+			node->vm_info.last_vmid_kfd = (i%2 == 0) ?
+						last_vmid_kfd-vmid_num_kfd :
+						last_vmid_kfd;
+			node->compute_vmid_bitmap =
+				((0x1 << (node->vm_info.last_vmid_kfd + 1)) - 1) -
+				((0x1 << (node->vm_info.first_vmid_kfd)) - 1);
+		} else {
+			node->vm_info.first_vmid_kfd = first_vmid_kfd;
+			node->vm_info.last_vmid_kfd = last_vmid_kfd;
+			node->compute_vmid_bitmap =
+				gpu_resources->compute_vmid_bitmap;
+		}
+		node->max_proc_per_quantum = max_proc_per_quantum;
+		atomic_set(&node->sram_ecc_flag, 0);
+		/* Initialize the KFD node */
+		if (kfd_init_node(node)) {
+			dev_err(kfd_device, "Error initializing KFD node\n");
+			goto node_init_error;
+		}
+		kfd->nodes[i] = node;
 	}
-	kfd->node = node;
 
 	if (kfd_resume_iommu(kfd))
 		goto kfd_resume_iommu_error;
@@ -722,9 +795,9 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	goto out;
 
 kfd_resume_iommu_error:
-	kfd_cleanup_node(kfd);
 node_init_error:
 node_alloc_error:
+	kfd_cleanup_nodes(kfd, i);
 device_iommu_error:
 	kfd_doorbell_fini(kfd);
 kfd_doorbell_error:
@@ -742,7 +815,9 @@ out:
 void kgd2kfd_device_exit(struct kfd_dev *kfd)
 {
 	if (kfd->init_complete) {
-		kfd_cleanup_node(kfd);
+		/* Cleanup KFD nodes */
+		kfd_cleanup_nodes(kfd, kfd->num_nodes);
+		/* Cleanup common/shared resources */
 		kfd_doorbell_fini(kfd);
 		ida_destroy(&kfd->doorbell_ida);
 		kfd_gtt_sa_fini(kfd);
@@ -754,18 +829,23 @@ void kgd2kfd_device_exit(struct kfd_dev *kfd)
 
 int kgd2kfd_pre_reset(struct kfd_dev *kfd)
 {
-	struct kfd_node *node = kfd->node;
+	struct kfd_node *node;
+	int i;
 
 	if (!kfd->init_complete)
 		return 0;
 
-	kfd_smi_event_update_gpu_reset(node, false);
-
-	node->dqm->ops.pre_reset(node->dqm);
+	for (i = 0; i < kfd->num_nodes; i++) {
+		node = kfd->nodes[i];
+		kfd_smi_event_update_gpu_reset(node, false);
+		node->dqm->ops.pre_reset(node->dqm);
+	}
 
 	kgd2kfd_suspend(kfd, false);
 
-	kfd_signal_reset_event(node);
+	for (i = 0; i < kfd->num_nodes; i++)
+		kfd_signal_reset_event(kfd->nodes[i]);
+
 	return 0;
 }
 
@@ -778,19 +858,25 @@ int kgd2kfd_pre_reset(struct kfd_dev *kfd)
 int kgd2kfd_post_reset(struct kfd_dev *kfd)
 {
 	int ret;
-	struct kfd_node *node = kfd->node;
+	struct kfd_node *node;
+	int i;
 
 	if (!kfd->init_complete)
 		return 0;
 
-	ret = kfd_resume(node);
-	if (ret)
-		return ret;
-	atomic_dec(&kfd_locked);
+	for (i = 0; i < kfd->num_nodes; i++) {
+		ret = kfd_resume(kfd->nodes[i]);
+		if (ret)
+			return ret;
+	}
 
-	atomic_set(&node->sram_ecc_flag, 0);
+	atomic_dec(&kfd_locked);
 
-	kfd_smi_event_update_gpu_reset(node, true);
+	for (i = 0; i < kfd->num_nodes; i++) {
+		node = kfd->nodes[i];
+		atomic_set(&node->sram_ecc_flag, 0);
+		kfd_smi_event_update_gpu_reset(node, true);
+	}
 
 	return 0;
 }
@@ -802,7 +888,8 @@ bool kfd_is_locked(void)
 
 void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm)
 {
-	struct kfd_node *node = kfd->node;
+	struct kfd_node *node;
+	int i;
 
 	if (!kfd->init_complete)
 		return;
@@ -814,21 +901,25 @@ void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm)
 			kfd_suspend_all_processes();
 	}
 
-	node->dqm->ops.stop(node->dqm);
+	for (i = 0; i < kfd->num_nodes; i++) {
+		node = kfd->nodes[i];
+		node->dqm->ops.stop(node->dqm);
+	}
 	kfd_iommu_suspend(kfd);
 }
 
 int kgd2kfd_resume(struct kfd_dev *kfd, bool run_pm)
 {
-	int ret, count;
-	struct kfd_node *node = kfd->node;
+	int ret, count, i;
 
 	if (!kfd->init_complete)
 		return 0;
 
-	ret = kfd_resume(node);
-	if (ret)
-		return ret;
+	for (i = 0; i < kfd->num_nodes; i++) {
+		ret = kfd_resume(kfd->nodes[i]);
+		if (ret)
+			return ret;
+	}
 
 	/* for runtime resume, skip unlocking kfd */
 	if (!run_pm) {
@@ -892,10 +983,10 @@ static inline void kfd_queue_work(struct workqueue_struct *wq,
 /* This is called directly from KGD at ISR. */
 void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry)
 {
-	uint32_t patched_ihre[KFD_MAX_RING_ENTRY_SIZE];
+	uint32_t patched_ihre[KFD_MAX_RING_ENTRY_SIZE], i;
 	bool is_patched = false;
 	unsigned long flags;
-	struct kfd_node *node = kfd->node;
+	struct kfd_node *node;
 
 	if (!kfd->init_complete)
 		return;
@@ -905,16 +996,22 @@ void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry)
 		return;
 	}
 
-	spin_lock_irqsave(&node->interrupt_lock, flags);
-
-	if (node->interrupts_active
-	    && interrupt_is_wanted(node, ih_ring_entry,
-				   patched_ihre, &is_patched)
-	    && enqueue_ih_ring_entry(node,
-				     is_patched ? patched_ihre : ih_ring_entry))
-		kfd_queue_work(node->ih_wq, &node->interrupt_work);
+	for (i = 0; i < kfd->num_nodes; i++) {
+		node = kfd->nodes[i];
+		spin_lock_irqsave(&node->interrupt_lock, flags);
+
+		if (node->interrupts_active
+		    && interrupt_is_wanted(node, ih_ring_entry,
+			    	patched_ihre, &is_patched)
+		    && enqueue_ih_ring_entry(node,
+			    	is_patched ? patched_ihre : ih_ring_entry)) {
+			kfd_queue_work(node->ih_wq, &node->interrupt_work);
+			spin_unlock_irqrestore(&node->interrupt_lock, flags);
+				return;
+		}
+		spin_unlock_irqrestore(&node->interrupt_lock, flags);
+	}
 
-	spin_unlock_irqrestore(&node->interrupt_lock, flags);
 }
 
 int kgd2kfd_quiesce_mm(struct mm_struct *mm, uint32_t trigger)
@@ -1181,8 +1278,13 @@ int kfd_gtt_sa_free(struct kfd_node *node, struct kfd_mem_obj *mem_obj)
 
 void kgd2kfd_set_sram_ecc_flag(struct kfd_dev *kfd)
 {
+	/*
+	 * TODO: Currently update SRAM ECC flag for first node.
+	 * This needs to be updated later when we can
+	 * identify SRAM ECC error on other nodes also.
+	 */
 	if (kfd)
-		atomic_inc(&kfd->node->sram_ecc_flag);
+		atomic_inc(&kfd->nodes[0]->sram_ecc_flag);
 }
 
 void kfd_inc_compute_active(struct kfd_node *node)
@@ -1202,8 +1304,14 @@ void kfd_dec_compute_active(struct kfd_node *node)
 
 void kgd2kfd_smi_event_throttle(struct kfd_dev *kfd, uint64_t throttle_bitmask)
 {
+	/*
+	 * TODO: For now, raise the throttling event only on first node.
+	 * This will need to change after we are able to determine
+	 * which node raised the throttling event.
+	 */
 	if (kfd && kfd->init_complete)
-		kfd_smi_event_update_thermal_throttling(kfd->node, throttle_bitmask);
+		kfd_smi_event_update_thermal_throttling(kfd->nodes[0],
+							throttle_bitmask);
 }
 
 /* kfd_get_num_sdma_engines returns the number of PCIe optimized SDMA and
-- 
cgit 


From a805889a15315f7fa78c1c4bb2f1875c7c43f919 Mon Sep 17 00:00:00 2001
From: Mukul Joshi <mukul.joshi@amd.com>
Date: Mon, 9 May 2022 22:52:39 -0400
Subject: drm/amdkfd: Update SDMA queue management for GFX9.4.3

This patch updates SDMA queue management for multi XCC in GFX9.4.3.
- Allocate/deallocate SDMA queues from the correct SDMA engines
  based on the partition mode.
- Updates the kgd2kfd interface to fetch the correct SDMA register
  addresses.
- It also fixes dumping correct SDMA queue info in debugfs.

v2: squash in fix "drm/amdkfd: Fix XGMI SDMA user-mode queue allocation"

Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 37c6dc5c37bf..ec5f85ff34e5 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -741,6 +741,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		if (!node)
 			goto node_alloc_error;
 
+		node->node_id = i;
 		node->adev = kfd->adev;
 		node->kfd = kfd;
 		node->kfd2kgd = kfd->kfd2kgd;
@@ -1323,15 +1324,16 @@ unsigned int kfd_get_num_sdma_engines(struct kfd_node *node)
 {
 	/* If XGMI is not supported, all SDMA engines are PCIe */
 	if (!node->adev->gmc.xgmi.supported)
-		return node->adev->sdma.num_instances;
+		return node->adev->sdma.num_instances/(int)node->kfd->num_nodes;
 
-	return min(node->adev->sdma.num_instances, 2);
+	return min(node->adev->sdma.num_instances/(int)node->kfd->num_nodes, 2);
 }
 
 unsigned int kfd_get_num_xgmi_sdma_engines(struct kfd_node *node)
 {
 	/* After reserved for PCIe, the rest of engines are XGMI */
-	return node->adev->sdma.num_instances - kfd_get_num_sdma_engines(node);
+	return node->adev->sdma.num_instances/(int)node->kfd->num_nodes -
+		kfd_get_num_sdma_engines(node);
 }
 
 #if defined(CONFIG_DEBUG_FS)
-- 
cgit 


From 92085240ef9c0ec60c27a60b3cc0d4f5266fa511 Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Tue, 3 May 2022 10:16:46 -0400
Subject: drm/amdkfd: add gpu compute cores io links for gfx9.4.3

The PSP TA will only provide xGMI topology info for links between GPU
sockets so links between partitions from different sockets will be
hardcoded as 3 xGMI hops with 1 hops weighted as xGMI and 2 hops
weighted with a new intra-socket weight to indicate the longest
possible distance.

If the link between a partition and the CPU is non-PCIe, then assume
the CPU (CCDs) is located within the same socket as the partition
and represent the link as an intra-socket weighted single hop XGMI link
with memory bandwidth.

Links between partitions within a single socket will be abstracted as
single hop xGMI links weighted with the new intra-socket weight and
will have memory bandwidth.

Finally, use the unused function bits in the location ID to represent the
coordinates of the compute partition within its socket.

A follow on patch will resolve the requirement for GPU socket xGMI
link representation sometime later.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 8 ++++++++
 1 file changed, 8 insertions(+)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index ec5f85ff34e5..971a3aa3294a 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -702,6 +702,14 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	if (amdgpu_use_xgmi_p2p)
 		kfd->hive_id = kfd->adev->gmc.xgmi.hive_id;
 
+	/*
+	 * For GFX9.4.3, the KFD abstracts all partitions within a socket as
+	 * xGMI connected in the topology so assign a unique hive id per
+	 * device based on the pci device location if device is in PCIe mode.
+	 */
+	if (!kfd->hive_id && (KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 3)) && kfd->num_nodes > 1)
+		kfd->hive_id = pci_dev_id(kfd->adev->pdev);
+
 	kfd->noretry = kfd->adev->gmc.noretry;
 
 	/* If CRAT is broken, won't set iommu enabled */
-- 
cgit 


From 1bd6dd21fcd53ac78a9018b96699ef1aa99a3e59 Mon Sep 17 00:00:00 2001
From: Mukul Joshi <mukul.joshi@amd.com>
Date: Tue, 31 May 2022 16:31:28 -0400
Subject: drm/amdkfd: Add SDMA info for SDMA 4.4.2

Update SDMA queue information for SDMA 4.4.2.

Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 1 +
 1 file changed, 1 insertion(+)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 971a3aa3294a..8e5d785b8824 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -82,6 +82,7 @@ static void kfd_device_info_set_sdma_info(struct kfd_dev *kfd)
 	case IP_VERSION(4, 2, 0):/* VEGA20 */
 	case IP_VERSION(4, 2, 2):/* ARCTURUS */
 	case IP_VERSION(4, 4, 0):/* ALDEBARAN */
+	case IP_VERSION(4, 4, 2):
 	case IP_VERSION(5, 0, 0):/* NAVI10 */
 	case IP_VERSION(5, 0, 1):/* CYAN_SKILLFISH */
 	case IP_VERSION(5, 0, 2):/* NAVI14 */
-- 
cgit 


From d1d22df174ae512c57374f517e346b608f61555c Mon Sep 17 00:00:00 2001
From: Mukul Joshi <mukul.joshi@amd.com>
Date: Tue, 7 Jun 2022 14:46:18 -0400
Subject: drm/amdkfd: Populate memory info before adding GPU node to topology

The local memory info needs to be fetched before the GPU node is added
to topology. Without this, the sysfs is incorrectly populated and the
size is reported as 0. This was causing rocr tests to fail. This issue
was caused because of a bad merge.

Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Reviewed-by: Amber Lin <Amber.Lin@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 8e5d785b8824..829e32433faf 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -743,6 +743,8 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	node->max_proc_per_quantum = max_proc_per_quantum;
 	atomic_set(&node->sram_ecc_flag, 0);
 
+	amdgpu_amdkfd_get_local_mem_info(kfd->adev, &kfd->local_mem_info);
+
 	dev_info(kfd_device, "Total number of KFD nodes to be created: %d\n",
 				kfd->num_nodes);
 	for (i = 0; i < kfd->num_nodes; i++) {
@@ -793,8 +795,6 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	if (kfd_resume_iommu(kfd))
 		goto kfd_resume_iommu_error;
 
-	amdgpu_amdkfd_get_local_mem_info(kfd->adev, &kfd->local_mem_info);
-
 	kfd->init_complete = true;
 	dev_info(kfd_device, "added device %x:%x\n", kfd->adev->pdev->vendor,
 		 kfd->adev->pdev->device);
-- 
cgit 


From 8078f1c610fdcdd8003e2c538fb04af41fa5c269 Mon Sep 17 00:00:00 2001
From: Lijo Lazar <lijo.lazar@amd.com>
Date: Wed, 29 Jun 2022 11:41:53 +0530
Subject: drm/amdgpu: Change num_xcd to xcc_mask

Instead of number of XCCs, keep a mask of XCCs for the exact XCCs
available on the ASIC. XCC configuration could differ based on
different ASIC configs.

v2:
	Rename num_xcd to num_xcc (Hawking)
	Use smaller xcc_mask size, changed to u16 (Le)

Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Le Ma <Le.Ma@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 829e32433faf..df96c4c508a0 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -592,6 +592,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	struct kfd_node *node;
 	uint32_t first_vmid_kfd, last_vmid_kfd, vmid_num_kfd;
 	unsigned int max_proc_per_quantum;
+	int num_xcd;
 
 	kfd->mec_fw_version = amdgpu_amdkfd_get_fw_version(kfd->adev,
 			KGD_ENGINE_MEC1);
@@ -601,16 +602,15 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 			KGD_ENGINE_SDMA1);
 	kfd->shared_resources = *gpu_resources;
 
-	if (kfd->adev->gfx.num_xcd == 0 || kfd->adev->gfx.num_xcd == 1 ||
-	    kfd->adev->gfx.num_xcc_per_xcp == 0)
+	num_xcd = NUM_XCC(kfd->adev->gfx.xcc_mask);
+	if (num_xcd == 0 || num_xcd == 1 || kfd->adev->gfx.num_xcc_per_xcp == 0)
 		kfd->num_nodes = 1;
 	else
-		kfd->num_nodes =
-			kfd->adev->gfx.num_xcd/kfd->adev->gfx.num_xcc_per_xcp;
+		kfd->num_nodes = num_xcd / kfd->adev->gfx.num_xcc_per_xcp;
 	if (kfd->num_nodes == 0) {
 		dev_err(kfd_device,
 			"KFD num nodes cannot be 0, GC inst: %d, num_xcc_in_node: %d\n",
-			kfd->adev->gfx.num_xcd, kfd->adev->gfx.num_xcc_per_xcp);
+			num_xcd, kfd->adev->gfx.num_xcc_per_xcp);
 		goto out;
 	}
 
-- 
cgit 


From fe1f05df5919c67c3add49efb55e251a8d78ee4e Mon Sep 17 00:00:00 2001
From: Mukul Joshi <mukul.joshi@amd.com>
Date: Tue, 31 May 2022 14:39:36 -0400
Subject: drm/amdkfd: Rework kfd_locked handling

Currently, even if kfd_locked is set, a process is first
created and then removed to work around a race condition
in updating kfd_locked flag. Rework kfd_locked handling to
ensure no processes is created if kfd_locked is set. This
is achieved by updating kfd_locked under kfd_processes_mutex.
With this there is no need for kfd_locked to be an atomic
counter. Instead, it can be a regular integer.

Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 21 ++++++++++++++++-----
 1 file changed, 16 insertions(+), 5 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index df96c4c508a0..eb2b44fddf74 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -42,7 +42,7 @@
  * once locked, kfd driver will stop any further GPU execution.
  * create process (open) will return -EAGAIN.
  */
-static atomic_t kfd_locked = ATOMIC_INIT(0);
+static int kfd_locked;
 
 #ifdef CONFIG_DRM_AMDGPU_CIK
 extern const struct kfd2kgd_calls gfx_v7_kfd2kgd;
@@ -880,7 +880,9 @@ int kgd2kfd_post_reset(struct kfd_dev *kfd)
 			return ret;
 	}
 
-	atomic_dec(&kfd_locked);
+	mutex_lock(&kfd_processes_mutex);
+	--kfd_locked;
+	mutex_unlock(&kfd_processes_mutex);
 
 	for (i = 0; i < kfd->num_nodes; i++) {
 		node = kfd->nodes[i];
@@ -893,21 +895,27 @@ int kgd2kfd_post_reset(struct kfd_dev *kfd)
 
 bool kfd_is_locked(void)
 {
-	return  (atomic_read(&kfd_locked) > 0);
+	lockdep_assert_held(&kfd_processes_mutex);
+	return  (kfd_locked > 0);
 }
 
 void kgd2kfd_suspend(struct kfd_dev *kfd, bool run_pm)
 {
 	struct kfd_node *node;
 	int i;
+	int count;
 
 	if (!kfd->init_complete)
 		return;
 
 	/* for runtime suspend, skip locking kfd */
 	if (!run_pm) {
+		mutex_lock(&kfd_processes_mutex);
+		count = ++kfd_locked;
+		mutex_unlock(&kfd_processes_mutex);
+
 		/* For first KFD device suspend all the KFD processes */
-		if (atomic_inc_return(&kfd_locked) == 1)
+		if (count == 1)
 			kfd_suspend_all_processes();
 	}
 
@@ -933,7 +941,10 @@ int kgd2kfd_resume(struct kfd_dev *kfd, bool run_pm)
 
 	/* for runtime resume, skip unlocking kfd */
 	if (!run_pm) {
-		count = atomic_dec_return(&kfd_locked);
+		mutex_lock(&kfd_processes_mutex);
+		count = --kfd_locked;
+		mutex_unlock(&kfd_processes_mutex);
+
 		WARN_ONCE(count < 0, "KFD suspend / resume ref. error");
 		if (count == 0)
 			ret = kfd_resume_all_processes();
-- 
cgit 


From 0c7315e7d5ef9b36ca4db32ffeb34a187cbaf231 Mon Sep 17 00:00:00 2001
From: Mukul Joshi <mukul.joshi@amd.com>
Date: Fri, 10 Jun 2022 09:41:29 -0400
Subject: drm/amdkfd: Add device repartition support

GFX9.4.3 will support dynamic repartitioning of the GPU through sysfs.
Add device repartitioning support in KFD to repartition GPU from one
mode to other.

v2: squash in fix ("drm/amdkfd: Fix warning kgd2kfd_unlock_kfd defined but not used")

Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index eb2b44fddf74..293787290e36 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -1356,6 +1356,27 @@ unsigned int kfd_get_num_xgmi_sdma_engines(struct kfd_node *node)
 		kfd_get_num_sdma_engines(node);
 }
 
+int kgd2kfd_check_and_lock_kfd(void)
+{
+	mutex_lock(&kfd_processes_mutex);
+	if (!hash_empty(kfd_processes_table) || kfd_is_locked()) {
+		mutex_unlock(&kfd_processes_mutex);
+		return -EBUSY;
+	}
+
+	++kfd_locked;
+	mutex_unlock(&kfd_processes_mutex);
+
+	return 0;
+}
+
+void kgd2kfd_unlock_kfd(void)
+{
+	mutex_lock(&kfd_processes_mutex);
+	--kfd_locked;
+	mutex_unlock(&kfd_processes_mutex);
+}
+
 #if defined(CONFIG_DEBUG_FS)
 
 /* This function will send a package to HIQ to hang the HWS
-- 
cgit 


From 8e7fd19380f9187dae3ad18a61793b1752dfa097 Mon Sep 17 00:00:00 2001
From: Lijo Lazar <lijo.lazar@amd.com>
Date: Wed, 16 Nov 2022 17:15:47 +0530
Subject: drm/amdgpu: Switch to SOC partition funcs

For GFXv9.4.3, use SOC level partition switch implementation rather than
keeping them at GFX IP level. Change the exisiting implementation in
GFX IP for keeping partition mode and restrict it to only GFX related
switch.

Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 293787290e36..7a963d0a34e2 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -34,6 +34,7 @@
 #include "kfd_smi_events.h"
 #include "kfd_migrate.h"
 #include "amdgpu.h"
+#include "amdgpu_xcp.h"
 
 #define MQD_SIZE_ALIGNED 768
 
@@ -592,7 +593,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	struct kfd_node *node;
 	uint32_t first_vmid_kfd, last_vmid_kfd, vmid_num_kfd;
 	unsigned int max_proc_per_quantum;
-	int num_xcd;
+	int num_xcd, partition_mode;
 
 	kfd->mec_fw_version = amdgpu_amdkfd_get_fw_version(kfd->adev,
 			KGD_ENGINE_MEC1);
@@ -644,8 +645,9 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	 * If the VMID range changes for GFX9.4.3, then this code MUST be
 	 * revisited.
 	 */
+	partition_mode = amdgpu_xcp_query_partition_mode(kfd->adev->xcp_mgr);
 	if (KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 3) &&
-	    kfd->adev->gfx.partition_mode == AMDGPU_CPX_PARTITION_MODE &&
+	    partition_mode == AMDGPU_CPX_PARTITION_MODE &&
 	    kfd->num_nodes != 1) {
 		vmid_num_kfd /= 2;
 		first_vmid_kfd = last_vmid_kfd + 1 - vmid_num_kfd*2;
@@ -761,7 +763,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		node->start_xcc_id = node->num_xcc_per_node * i;
 
 		if (KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 3) &&
-		    kfd->adev->gfx.partition_mode == AMDGPU_CPX_PARTITION_MODE &&
+		    partition_mode == AMDGPU_CPX_PARTITION_MODE &&
 		    kfd->num_nodes != 1) {
 			/* For GFX9.4.3 and CPX mode, first XCD gets VMID range
 			 * 4-9 and second XCD gets VMID range 10-15.
-- 
cgit 


From 8c45a8340dd097ea0d6be6f718c4882283d9645d Mon Sep 17 00:00:00 2001
From: Philip Yang <Philip.Yang@amd.com>
Date: Tue, 24 Jan 2023 10:10:14 -0500
Subject: drm/amdkfd: Cleanup KFD nodes creation

kfd node allocation outside kfd->num_nodes loop is not needed and causes
memory leak because kfd->num_nodes is at least equal to 1.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 18 ++----------------
 1 file changed, 2 insertions(+), 16 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 7a963d0a34e2..d7cffd91f1d7 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -729,26 +729,12 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	/* TODO: Needs to be updated for memory partitioning */
 	svm_migrate_init(kfd->adev);
 
-	/* Allocate the KFD node */
-	node = kzalloc(sizeof(struct kfd_node), GFP_KERNEL);
-	if (!node) {
-		dev_err(kfd_device, "Error allocating KFD node\n");
-		goto node_alloc_error;
-	}
-
-	node->adev = kfd->adev;
-	node->kfd = kfd;
-	node->kfd2kgd = kfd->kfd2kgd;
-	node->vm_info.vmid_num_kfd = vmid_num_kfd;
-	node->vm_info.first_vmid_kfd = first_vmid_kfd;
-	node->vm_info.last_vmid_kfd = last_vmid_kfd;
-	node->max_proc_per_quantum = max_proc_per_quantum;
-	atomic_set(&node->sram_ecc_flag, 0);
-
 	amdgpu_amdkfd_get_local_mem_info(kfd->adev, &kfd->local_mem_info);
 
 	dev_info(kfd_device, "Total number of KFD nodes to be created: %d\n",
 				kfd->num_nodes);
+
+	/* Allocate the KFD nodes */
 	for (i = 0; i < kfd->num_nodes; i++) {
 		node = kzalloc(sizeof(struct kfd_node), GFP_KERNEL);
 		if (!node)
-- 
cgit 


From ded7d99eb5b78931cec30dd49cd4097d0ac770e1 Mon Sep 17 00:00:00 2001
From: Lijo Lazar <lijo.lazar@amd.com>
Date: Mon, 16 Jan 2023 10:55:38 +0530
Subject: drm/amdgpu: Add flags for partition mode query

It's not required to take lock on all cases while querying partition
mode. Querying partition mode during KFD init process doesn't need to
take a lock. Init process after a switch will already be happening under
lock. Control the behaviour by adding flags to xcp_query_partition_mode.

Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index d7cffd91f1d7..4293cbf9ceb0 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -645,7 +645,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	 * If the VMID range changes for GFX9.4.3, then this code MUST be
 	 * revisited.
 	 */
-	partition_mode = amdgpu_xcp_query_partition_mode(kfd->adev->xcp_mgr);
+	partition_mode = amdgpu_xcp_query_partition_mode(kfd->adev->xcp_mgr, AMDGPU_XCP_FL_LOCKED);
 	if (KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 3) &&
 	    partition_mode == AMDGPU_CPX_PARTITION_MODE &&
 	    kfd->num_nodes != 1) {
-- 
cgit 


From a75f2271a4936265c8a189ab06f9eb89e343b441 Mon Sep 17 00:00:00 2001
From: Lijo Lazar <lijo.lazar@amd.com>
Date: Thu, 9 Feb 2023 14:44:13 +0530
Subject: drm/amdkfd: Add xcp reference to kfd node

Fetch xcp information from xcp_mgr and also add xcc_mask to kfd node.

Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Le Ma <le.ma@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 4293cbf9ceb0..647c3313c27e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -594,6 +594,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	uint32_t first_vmid_kfd, last_vmid_kfd, vmid_num_kfd;
 	unsigned int max_proc_per_quantum;
 	int num_xcd, partition_mode;
+	int xcp_idx;
 
 	kfd->mec_fw_version = amdgpu_amdkfd_get_fw_version(kfd->adev,
 			KGD_ENGINE_MEC1);
@@ -603,11 +604,8 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 			KGD_ENGINE_SDMA1);
 	kfd->shared_resources = *gpu_resources;
 
-	num_xcd = NUM_XCC(kfd->adev->gfx.xcc_mask);
-	if (num_xcd == 0 || num_xcd == 1 || kfd->adev->gfx.num_xcc_per_xcp == 0)
-		kfd->num_nodes = 1;
-	else
-		kfd->num_nodes = num_xcd / kfd->adev->gfx.num_xcc_per_xcp;
+	kfd->num_nodes = amdgpu_xcp_get_num_xcp(kfd->adev->xcp_mgr);
+
 	if (kfd->num_nodes == 0) {
 		dev_err(kfd_device,
 			"KFD num nodes cannot be 0, GC inst: %d, num_xcc_in_node: %d\n",
@@ -735,7 +733,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 				kfd->num_nodes);
 
 	/* Allocate the KFD nodes */
-	for (i = 0; i < kfd->num_nodes; i++) {
+	for (i = 0, xcp_idx = 0; i < kfd->num_nodes; i++) {
 		node = kzalloc(sizeof(struct kfd_node), GFP_KERNEL);
 		if (!node)
 			goto node_alloc_error;
@@ -745,6 +743,15 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		node->kfd = kfd;
 		node->kfd2kgd = kfd->kfd2kgd;
 		node->vm_info.vmid_num_kfd = vmid_num_kfd;
+		node->xcp = amdgpu_get_next_xcp(kfd->adev->xcp_mgr, &xcp_idx);
+		/* TODO : Check if error handling is needed */
+		if (node->xcp)
+			amdgpu_xcp_get_inst_details(node->xcp, AMDGPU_XCP_GFX,
+						    &node->xcc_mask);
+		else
+			node->xcc_mask =
+				(1U << NUM_XCC(kfd->adev->gfx.xcc_mask)) - 1;
+
 		node->num_xcc_per_node = max(1U, kfd->adev->gfx.num_xcc_per_xcp);
 		node->start_xcc_id = node->num_xcc_per_node * i;
 
-- 
cgit 


From c4050ff1a43eec08498b1ed876efc6213592dba0 Mon Sep 17 00:00:00 2001
From: Lijo Lazar <lijo.lazar@amd.com>
Date: Thu, 9 Feb 2023 16:30:53 +0530
Subject: drm/amdkfd: Use xcc mask for identifying xcc

Instead of start xcc id and number of xcc per node, use the xcc mask
which is the mask of logical ids of xccs belonging to a parition.

Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Le Ma <le.ma@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 647c3313c27e..b5497d2ee984 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -745,15 +745,14 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		node->vm_info.vmid_num_kfd = vmid_num_kfd;
 		node->xcp = amdgpu_get_next_xcp(kfd->adev->xcp_mgr, &xcp_idx);
 		/* TODO : Check if error handling is needed */
-		if (node->xcp)
+		if (node->xcp) {
 			amdgpu_xcp_get_inst_details(node->xcp, AMDGPU_XCP_GFX,
 						    &node->xcc_mask);
-		else
+			++xcp_idx;
+		} else {
 			node->xcc_mask =
 				(1U << NUM_XCC(kfd->adev->gfx.xcc_mask)) - 1;
-
-		node->num_xcc_per_node = max(1U, kfd->adev->gfx.num_xcc_per_xcp);
-		node->start_xcc_id = node->num_xcc_per_node * i;
+		}
 
 		if (KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 3) &&
 		    partition_mode == AMDGPU_CPX_PARTITION_MODE &&
-- 
cgit 


From 4c6ce75fdd628c43aea11448ed41b52119dae42b Mon Sep 17 00:00:00 2001
From: Philip Yang <Philip.Yang@amd.com>
Date: Thu, 26 Jan 2023 18:11:29 -0500
Subject: drm/amdkfd: Show KFD node memory partition info

Show KFD node memory partition id and size, add helper function
KFD_XCP_MEMORY_SIZE to get kfd node memory size, will be used
later to support memory accounting per partition.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index b5497d2ee984..db5b53fcdf11 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -724,7 +724,6 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 
 	kfd_cwsr_init(kfd);
 
-	/* TODO: Needs to be updated for memory partitioning */
 	svm_migrate_init(kfd->adev);
 
 	amdgpu_amdkfd_get_local_mem_info(kfd->adev, &kfd->local_mem_info);
@@ -754,6 +753,12 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 				(1U << NUM_XCC(kfd->adev->gfx.xcc_mask)) - 1;
 		}
 
+		if (node->xcp) {
+			dev_info(kfd_device, "KFD node %d partition %d size %lldM\n",
+				node->node_id, node->xcp->mem_id,
+				KFD_XCP_MEMORY_SIZE(node) >> 20);
+		}
+
 		if (KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 3) &&
 		    partition_mode == AMDGPU_CPX_PARTITION_MODE &&
 		    kfd->num_nodes != 1) {
-- 
cgit 


From 315e29eca57f85107cc6f687c2d510aa532fb3f0 Mon Sep 17 00:00:00 2001
From: Mukul Joshi <mukul.joshi@amd.com>
Date: Mon, 20 Mar 2023 11:21:38 -0400
Subject: drm/amdkfd: Move local_mem_info to kfd_node

We need to track memory usage on a per partition basis. To do
that, store the local memory information in KFD node instead
of kfd device.

v2: squash in fix ("amdkfd: Use mem_id to access mem_partition info")

Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index db5b53fcdf11..d41da964d2f5 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -726,7 +726,6 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 
 	svm_migrate_init(kfd->adev);
 
-	amdgpu_amdkfd_get_local_mem_info(kfd->adev, &kfd->local_mem_info);
 
 	dev_info(kfd_device, "Total number of KFD nodes to be created: %d\n",
 				kfd->num_nodes);
@@ -756,7 +755,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		if (node->xcp) {
 			dev_info(kfd_device, "KFD node %d partition %d size %lldM\n",
 				node->node_id, node->xcp->mem_id,
-				KFD_XCP_MEMORY_SIZE(node) >> 20);
+				KFD_XCP_MEMORY_SIZE(node->adev, node->node_id) >> 20);
 		}
 
 		if (KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 3) &&
@@ -783,6 +782,10 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		}
 		node->max_proc_per_quantum = max_proc_per_quantum;
 		atomic_set(&node->sram_ecc_flag, 0);
+
+		amdgpu_amdkfd_get_local_mem_info(kfd->adev,
+					&node->local_mem_info, node->xcp->id);
+
 		/* Initialize the KFD node */
 		if (kfd_init_node(node)) {
 			dev_err(kfd_device, "Error initializing KFD node\n");
-- 
cgit 


From 25f50704343de1bea70100ad41621b5737a6a96b Mon Sep 17 00:00:00 2001
From: Philip Yang <Philip.Yang@amd.com>
Date: Thu, 23 Mar 2023 08:45:56 -0400
Subject: drm/amdkfd: APU mode set max svm range pages

svm_migrate_init set the max svm range pages based on the KFD nodes
partition size. APU mode don't init pgmap because there is no migration.

kgd2kfd_device_init calls svm_migrate_init after KFD nodes allocation
and initialization.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index d41da964d2f5..882ff86bba08 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -724,9 +724,6 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 
 	kfd_cwsr_init(kfd);
 
-	svm_migrate_init(kfd->adev);
-
-
 	dev_info(kfd_device, "Total number of KFD nodes to be created: %d\n",
 				kfd->num_nodes);
 
@@ -794,6 +791,8 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		kfd->nodes[i] = node;
 	}
 
+	svm_migrate_init(kfd->adev);
+
 	if (kfd_resume_iommu(kfd))
 		goto kfd_resume_iommu_error;
 
-- 
cgit 


From 84b4dd3f84de424a68e1fda0d483530ddaa92b45 Mon Sep 17 00:00:00 2001
From: Philip Yang <Philip.Yang@amd.com>
Date: Fri, 31 Mar 2023 11:18:12 -0400
Subject: drm/amdkfd: Refactor migrate init to support partition switch

Rename smv_migrate_init to a better name kgd2kfd_init_zone_device
because it setup zone devive pgmap for page migration and keep it in
kfd_migrate.c to access static functions svm_migrate_pgmap_ops. Call it
only once in amdgpu_device_ip_init after adev ip blocks are initialized,
but before amdgpu_amdkfd_device_init initialize kfd nodes which enable
SVM support based on pgmap.

svm_range_set_max_pages is called by kgd2kfd_device_init everytime after
switching compute partition mode.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 882ff86bba08..bf32e547182c 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -32,6 +32,7 @@
 #include "kfd_iommu.h"
 #include "amdgpu_amdkfd.h"
 #include "kfd_smi_events.h"
+#include "kfd_svm.h"
 #include "kfd_migrate.h"
 #include "amdgpu.h"
 #include "amdgpu_xcp.h"
@@ -791,7 +792,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		kfd->nodes[i] = node;
 	}
 
-	svm_migrate_init(kfd->adev);
+	svm_range_set_max_pages(kfd->adev);
 
 	if (kfd_resume_iommu(kfd))
 		goto kfd_resume_iommu_error;
-- 
cgit 


From 0409022c15623d91c112e51f38cb62633becd432 Mon Sep 17 00:00:00 2001
From: Hawking Zhang <Hawking.Zhang@amd.com>
Date: Thu, 11 May 2023 17:01:03 +0800
Subject: drm/amdkfd: Fix null ptr access

Avoid access null xcp_mgr pointer.

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index bf32e547182c..2cfef3f9456f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -644,12 +644,14 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	 * If the VMID range changes for GFX9.4.3, then this code MUST be
 	 * revisited.
 	 */
-	partition_mode = amdgpu_xcp_query_partition_mode(kfd->adev->xcp_mgr, AMDGPU_XCP_FL_LOCKED);
-	if (KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 3) &&
-	    partition_mode == AMDGPU_CPX_PARTITION_MODE &&
-	    kfd->num_nodes != 1) {
-		vmid_num_kfd /= 2;
-		first_vmid_kfd = last_vmid_kfd + 1 - vmid_num_kfd*2;
+	if (kfd->adev->xcp_mgr) {
+		partition_mode = amdgpu_xcp_query_partition_mode(kfd->adev->xcp_mgr,
+								 AMDGPU_XCP_FL_LOCKED);
+		if (partition_mode == AMDGPU_CPX_PARTITION_MODE &&
+		    kfd->num_nodes != 1) {
+			vmid_num_kfd /= 2;
+			first_vmid_kfd = last_vmid_kfd + 1 - vmid_num_kfd*2;
+		}
 	}
 
 	/* Verify module parameters regarding mapped process number*/
-- 
cgit 


From 9a3ce1a7a9e5372d8c275bf3fbef4456c8407145 Mon Sep 17 00:00:00 2001
From: Hawking Zhang <Hawking.Zhang@amd.com>
Date: Fri, 12 May 2023 13:22:57 +0800
Subject: drm/amdgpu: Do not access members of xcp w/o check (v2)

Not all the asic needs xcp. ensure check xcp availabity
before accessing its member.

v2: add missing change in kfd_topology.c

Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Le Ma <le.ma@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 2cfef3f9456f..986543a000bf 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -784,7 +784,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 		atomic_set(&node->sram_ecc_flag, 0);
 
 		amdgpu_amdkfd_get_local_mem_info(kfd->adev,
-					&node->local_mem_info, node->xcp->id);
+					&node->local_mem_info, node->xcp);
 
 		/* Initialize the KFD node */
 		if (kfd_init_node(node)) {
-- 
cgit 


From 2ad00e753ae13b0c523a579fb04372787f77cce9 Mon Sep 17 00:00:00 2001
From: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Date: Fri, 19 May 2023 10:44:41 +0530
Subject: drm/amdgpu: Fix uninitalized variable in kgd2kfd_device_init
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

drivers/gpu/drm/amd/amdgpu/../amdkfd/kfd_device.c:613:4: error: variable 'num_xcd' is uninitialized when used here [-Werror,-Wuninitialized]
                        num_xcd, kfd->adev->gfx.num_xcc_per_xcp);
                        ^~~~~~~
include/linux/dev_printk.h:144:65: note: expanded from macro 'dev_err'
        dev_printk_index_wrap(_dev_err, KERN_ERR, dev, dev_fmt(fmt), ##__VA_ARGS__)
                                                                       ^~~~~~~~~~~
include/linux/dev_printk.h:110:23: note: expanded from macro 'dev_printk_index_wrap'
                _p_func(dev, fmt, ##__VA_ARGS__);                       \
                                    ^~~~~~~~~~~
drivers/gpu/drm/amd/amdgpu/../amdkfd/kfd_device.c:597:13: note: initialize the variable 'num_xcd' to silence this warning
        int num_xcd, partition_mode;
                   ^
                    = 0
1 error generated.

Cc: Luben Tuikov <luben.tuikov@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: Christian König <christian.koenig@amd.com>
Cc: Felix Kuehling <Felix.Kuehling@amd.com>
Cc: Mukul Joshi <mukul.joshi@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 986543a000bf..e92b93b2c14c 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -594,7 +594,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	struct kfd_node *node;
 	uint32_t first_vmid_kfd, last_vmid_kfd, vmid_num_kfd;
 	unsigned int max_proc_per_quantum;
-	int num_xcd, partition_mode;
+	int partition_mode;
 	int xcp_idx;
 
 	kfd->mec_fw_version = amdgpu_amdkfd_get_fw_version(kfd->adev,
@@ -609,8 +609,8 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 
 	if (kfd->num_nodes == 0) {
 		dev_err(kfd_device,
-			"KFD num nodes cannot be 0, GC inst: %d, num_xcc_in_node: %d\n",
-			num_xcd, kfd->adev->gfx.num_xcc_per_xcp);
+			"KFD num nodes cannot be 0, num_xcc_in_node: %d\n",
+			kfd->adev->gfx.num_xcc_per_xcp);
 		goto out;
 	}
 
-- 
cgit 


From 55a6dc60b47c817c644af2b505d46815d8b9219e Mon Sep 17 00:00:00 2001
From: Mukul Joshi <mukul.joshi@amd.com>
Date: Tue, 23 May 2023 11:59:53 -0400
Subject: drm/amdkfd: Set event interrupt class for GFX 9.4.3

Fix the warning during driver load because the event
interrupt class is not set for GFX9.4.3.

Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 1 +
 1 file changed, 1 insertion(+)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index e92b93b2c14c..862a50f7b490 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -139,6 +139,7 @@ static void kfd_device_info_set_event_interrupt_class(struct kfd_dev *kfd)
 	case IP_VERSION(9, 4, 0): /* VEGA20 */
 	case IP_VERSION(9, 4, 1): /* ARCTURUS */
 	case IP_VERSION(9, 4, 2): /* ALDEBARAN */
+	case IP_VERSION(9, 4, 3): /* GC 9.4.3 */
 	case IP_VERSION(10, 3, 1): /* VANGOGH */
 	case IP_VERSION(10, 3, 3): /* YELLOW_CARP */
 	case IP_VERSION(10, 3, 6): /* GC 10.3.6 */
-- 
cgit 


From 28ebbb4981cb1fad12e0b1227dbecc88810b1ee8 Mon Sep 17 00:00:00 2001
From: Alex Deucher <alexander.deucher@amd.com>
Date: Wed, 24 May 2023 14:30:12 -0400
Subject: drm/amdkfd: fix gfx_target_version for certain 11.0.3 devices

Certain boards with GC IP 11.0.3 need slightly different handling
in the shader compiler due to board specific bounding box
optimizations.

Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 862a50f7b490..ebc3c3f965f9 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -411,8 +411,15 @@ struct kfd_dev *kgd2kfd_probe(struct amdgpu_device *adev, bool vf)
 			f2g = &gfx_v11_kfd2kgd;
 			break;
 		case IP_VERSION(11, 0, 3):
-			/* Note: Compiler version is 11.0.1 while HW version is 11.0.3 */
-			gfx_target_version = 110001;
+			if ((adev->pdev->device == 0x7460 &&
+			     adev->pdev->revision == 0x00) ||
+			    (adev->pdev->device == 0x7461 &&
+			     adev->pdev->revision == 0x00))
+				/* Note: Compiler version is 11.0.5 while HW version is 11.0.3 */
+				gfx_target_version = 110005;
+			else
+				/* Note: Compiler version is 11.0.1 while HW version is 11.0.3 */
+				gfx_target_version = 110001;
 			f2g = &gfx_v11_kfd2kgd;
 			break;
 		default:
-- 
cgit 


From 07a1475279244cd8eea81bec44fd5f0a9d6871f8 Mon Sep 17 00:00:00 2001
From: Graham Sider <Graham.Sider@amd.com>
Date: Mon, 3 Apr 2023 15:31:53 -0400
Subject: drm/amdkfd: Add new gfx_target_versions for GC 9.4.3

For GC 9.4.3, set gfx_target_version to 90402 for rev 1 and later (APU
or dGPU), 90401 for rev 0 dGPU, and 90400 for rev 0 APU.

Signed-off-by: Graham Sider <Graham.Sider@amd.com>
Acked-by: Alex Deucher <Alexander.Deucher@amd.com>
Reviewed-by: Amber Lin <Amber.Lin@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index ebc3c3f965f9..e84ad1c5ef44 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -332,7 +332,9 @@ struct kfd_dev *kgd2kfd_probe(struct amdgpu_device *adev, bool vf)
 			f2g = &aldebaran_kfd2kgd;
 			break;
 		case IP_VERSION(9, 4, 3):
-			gfx_target_version = 90400;
+			gfx_target_version = adev->rev_id >= 1 ? 90402
+					   : adev->flags & AMD_IS_APU ? 90400
+					   : 90401;
 			f2g = &gc_9_4_3_kfd2kgd;
 			break;
 		/* Navi10 */
-- 
cgit 


From d3116d9f27b89d363dd528e42fcf4895a15e0c3c Mon Sep 17 00:00:00 2001
From: Yang Li <yang.lee@linux.alibaba.com>
Date: Wed, 31 May 2023 10:08:11 +0800
Subject: drm/amdkfd: clean up one inconsistent indenting

drivers/gpu/drm/amd/amdgpu/../amdkfd/kfd_device.c:1036 kgd2kfd_interrupt() warn: inconsistent indenting

Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index e84ad1c5ef44..f0ed6e6416c3 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -1042,7 +1042,7 @@ void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry)
 			    	is_patched ? patched_ihre : ih_ring_entry)) {
 			kfd_queue_work(node->ih_wq, &node->interrupt_work);
 			spin_unlock_irqrestore(&node->interrupt_lock, flags);
-				return;
+			return;
 		}
 		spin_unlock_irqrestore(&node->interrupt_lock, flags);
 	}
-- 
cgit 


From 12fb1ad70d65edc3405884792d044fa79df7244f Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Fri, 22 Apr 2022 12:26:18 -0400
Subject: drm/amdkfd: update process interrupt handling for debug events

The debugger must be notified by any debugger subscribed exception
that comes from hardware interrupts.

If a debugger session exits, any exceptions it subscribed to may still
have interrupts in the interrupt ring buffer or KGD/KFD pipeline.
To prevent a new session from inheriting stale interrupts, when a new
queue is created, open an interrupt drain and allow the IH ring to drain
from a timestamped checkpoint.  Then inject a custom IV so that once
the custom IV is picked up by the KFD, it's safe to close the drain
and proceed with queue creation.

The drain must also be on debug disable as SW interrupts may still
be processed.  Drain at this time and clear all the exception status.

The debugger may also not be attached nor subscibed to certain
exceptions so forward them directly to the runtime.

GFX10 also requires its own IV processing, hence the creation of
kfd_int_process_v10.c.  This is because the IV from SQ interrupts are
packed into a new continguous format unlike GFX9. To make this clear,
a separate interrupting handling code file was created.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index f0ed6e6416c3..2c36bb578633 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -140,6 +140,8 @@ static void kfd_device_info_set_event_interrupt_class(struct kfd_dev *kfd)
 	case IP_VERSION(9, 4, 1): /* ARCTURUS */
 	case IP_VERSION(9, 4, 2): /* ALDEBARAN */
 	case IP_VERSION(9, 4, 3): /* GC 9.4.3 */
+		kfd->device_info.event_interrupt_class = &event_interrupt_class_v9;
+		break;
 	case IP_VERSION(10, 3, 1): /* VANGOGH */
 	case IP_VERSION(10, 3, 3): /* YELLOW_CARP */
 	case IP_VERSION(10, 3, 6): /* GC 10.3.6 */
@@ -153,7 +155,7 @@ static void kfd_device_info_set_event_interrupt_class(struct kfd_dev *kfd)
 	case IP_VERSION(10, 3, 2): /* NAVY_FLOUNDER */
 	case IP_VERSION(10, 3, 4): /* DIMGREY_CAVEFISH */
 	case IP_VERSION(10, 3, 5): /* BEIGE_GOBY */
-		kfd->device_info.event_interrupt_class = &event_interrupt_class_v9;
+		kfd->device_info.event_interrupt_class = &event_interrupt_class_v10;
 		break;
 	case IP_VERSION(11, 0, 0):
 	case IP_VERSION(11, 0, 1):
-- 
cgit 


From e0f85f4690d089cc1a60337decafb1acf7eec45e Mon Sep 17 00:00:00 2001
From: Jonathan Kim <jonathan.kim@amd.com>
Date: Fri, 6 May 2022 14:58:55 -0400
Subject: drm/amdkfd: add debug set and clear address watch points operation

Shader read, write and atomic memory operations can be alerted to the
debugger as an address watch exception.

Allow the debugger to pass in a watch point to a particular memory
address per device.

Note that there exists only 4 watch points per devices to date, so have
the KFD keep track of what watch points are allocated or not.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 2 ++
 1 file changed, 2 insertions(+)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 2c36bb578633..9fc9d32cb579 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -811,6 +811,8 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd,
 	if (kfd_resume_iommu(kfd))
 		goto kfd_resume_iommu_error;
 
+	spin_lock_init(&kfd->watch_points_lock);
+
 	kfd->init_complete = true;
 	dev_info(kfd_device, "added device %x:%x\n", kfd->adev->pdev->vendor,
 		 kfd->adev->pdev->device);
-- 
cgit 


From 597364adc0fcf71617b3adbe647b6eec76e27554 Mon Sep 17 00:00:00 2001
From: Mukul Joshi <mukul.joshi@amd.com>
Date: Wed, 31 May 2023 11:22:03 -0400
Subject: drm/amdkfd: Fix reserved SDMA queues handling

This patch fixes a regression caused by a bad merge where
the handling of reserved SDMA queues was accidentally removed.
With the fix, the reserved SDMA queues are again correctly
marked as unavailable for allocation.

Fixes: a805889a1531 ("drm/amdkfd: Update SDMA queue management for GFX9.4.3")
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_device.c | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

(limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device.c')

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 9fc9d32cb579..9d4abfd8b55e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -106,20 +106,19 @@ static void kfd_device_info_set_sdma_info(struct kfd_dev *kfd)
 		kfd->device_info.num_sdma_queues_per_engine = 8;
 	}
 
+	bitmap_zero(kfd->device_info.reserved_sdma_queues_bitmap, KFD_MAX_SDMA_QUEUES);
+
 	switch (sdma_version) {
 	case IP_VERSION(6, 0, 0):
+	case IP_VERSION(6, 0, 1):
 	case IP_VERSION(6, 0, 2):
 	case IP_VERSION(6, 0, 3):
 		/* Reserve 1 for paging and 1 for gfx */
 		kfd->device_info.num_reserved_sdma_queues_per_engine = 2;
 		/* BIT(0)=engine-0 queue-0; BIT(1)=engine-1 queue-0; BIT(2)=engine-0 queue-1; ... */
-		kfd->device_info.reserved_sdma_queues_bitmap = 0xFULL;
-		break;
-	case IP_VERSION(6, 0, 1):
-		/* Reserve 1 for paging and 1 for gfx */
-		kfd->device_info.num_reserved_sdma_queues_per_engine = 2;
-		/* BIT(0)=engine-0 queue-0; BIT(1)=engine-0 queue-1; ... */
-		kfd->device_info.reserved_sdma_queues_bitmap = 0x3ULL;
+		bitmap_set(kfd->device_info.reserved_sdma_queues_bitmap, 0,
+			   kfd->adev->sdma.num_instances *
+			   kfd->device_info.num_reserved_sdma_queues_per_engine);
 		break;
 	default:
 		break;
-- 
cgit