Commit b41a3829 authored by Jonathan Kim's avatar Jonathan Kim Committed by Alex Deucher
Browse files

drm/amdkfd: fix debug watchpoints for logical devices 



The number of watchpoints should be set and constrained per logical
partition device, not by the socket device.

Signed-off-by: default avatarJonathan Kim <jonathan.kim@amd.com>
Reviewed-by: default avatarHarish Kasiviswanathan <harish.kasiviswanathan@amd.com>
Signed-off-by: default avatarAlex Deucher <alexander.deucher@amd.com>
parent 67c4ca9f

drivers/gpu/drm/amd/amdkfd/kfd_debug.c

+10 −10
Original line number Diff line number Diff line
@@ -365,47 +365,47 @@ static int kfd_dbg_get_dev_watch_id(struct kfd_process_device *pdd, int *watch_i 


	*watch_id = KFD_DEBUGGER_INVALID_WATCH_POINT_ID;



	spin_lock(&pdd->dev->kfd->watch_points_lock);

	spin_lock(&pdd->dev->watch_points_lock);



	for (i = 0; i < MAX_WATCH_ADDRESSES; i++) {

		/* device watchpoint in use so skip */

		if ((pdd->dev->kfd->alloc_watch_ids >> i) & 0x1)

		if ((pdd->dev->alloc_watch_ids >> i) & 0x1)

			continue;



		pdd->alloc_watch_ids |= 0x1 << i;

		pdd->dev->kfd->alloc_watch_ids |= 0x1 << i;

		pdd->dev->alloc_watch_ids |= 0x1 << i;

		*watch_id = i;

		spin_unlock(&pdd->dev->kfd->watch_points_lock);

		spin_unlock(&pdd->dev->watch_points_lock);

		return 0;

	}



	spin_unlock(&pdd->dev->kfd->watch_points_lock);

	spin_unlock(&pdd->dev->watch_points_lock);



	return -ENOMEM;

}



static void kfd_dbg_clear_dev_watch_id(struct kfd_process_device *pdd, int watch_id)

{

	spin_lock(&pdd->dev->kfd->watch_points_lock);

	spin_lock(&pdd->dev->watch_points_lock);



	/* process owns device watch point so safe to clear */

	if ((pdd->alloc_watch_ids >> watch_id) & 0x1) {

		pdd->alloc_watch_ids &= ~(0x1 << watch_id);

		pdd->dev->kfd->alloc_watch_ids &= ~(0x1 << watch_id);

		pdd->dev->alloc_watch_ids &= ~(0x1 << watch_id);

	}



	spin_unlock(&pdd->dev->kfd->watch_points_lock);

	spin_unlock(&pdd->dev->watch_points_lock);

}



static bool kfd_dbg_owns_dev_watch_id(struct kfd_process_device *pdd, int watch_id)

{

	bool owns_watch_id = false;



	spin_lock(&pdd->dev->kfd->watch_points_lock);

	spin_lock(&pdd->dev->watch_points_lock);

	owns_watch_id = watch_id < MAX_WATCH_ADDRESSES &&

			((pdd->alloc_watch_ids >> watch_id) & 0x1);



	spin_unlock(&pdd->dev->kfd->watch_points_lock);

	spin_unlock(&pdd->dev->watch_points_lock);



	return owns_watch_id;

}

drivers/gpu/drm/amd/amdkfd/kfd_device.c

+3 −2
Original line number Diff line number Diff line
@@ -884,13 +884,14 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, 
			dev_err(kfd_device, "Error initializing KFD node\n");

			goto node_init_error;

		}



		spin_lock_init(&node->watch_points_lock);



		kfd->nodes[i] = node;

	}



	svm_range_set_max_pages(kfd->adev);



	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);

drivers/gpu/drm/amd/amdkfd/kfd_priv.h

+4 −4
Original line number Diff line number Diff line
@@ -310,6 +310,10 @@ struct kfd_node { 
	struct kfd_local_mem_info local_mem_info;



	struct kfd_dev *kfd;



	/* Track per device allocated watch points */

	uint32_t alloc_watch_ids;

	spinlock_t watch_points_lock;

};



struct kfd_dev {
@@ -362,10 +366,6 @@ struct kfd_dev { 
	struct kfd_node *nodes[MAX_KFD_NODES];

	unsigned int num_nodes;



	/* Track per device allocated watch points */

	uint32_t alloc_watch_ids;

	spinlock_t watch_points_lock;



	/* Kernel doorbells for KFD device */

	struct amdgpu_bo *doorbells;