[PATCH 12/17] drm/amdkfd: allow split HQD on per-queue granularity v5

Andres Rodriguez andresx7 at gmail.com
Fri Apr 21 20:02:37 UTC 2017


Update the KGD to KFD interface to allow sharing pipes with queue
granularity instead of pipe granularity.

This allows for more interesting pipe/queue splits.

v2: fix overflow check for res.queue_mask
v3: fix shift overflow when setting res.queue_mask
v4: fix comment in is_pipeline_enabled()
v5: clamp res.queue_mask to the first MEC only

Reviewed-by: Edward O'Callaghan <funfunctor at folklore1984.net>
Reviewed-by: Felix Kuehling <Felix.Kuehling at amd.com>
Acked-by: Christian König <christian.koenig at amd.com>
Signed-off-by: Andres Rodriguez <andresx7 at gmail.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c         |  22 ++++-
 drivers/gpu/drm/amd/amdkfd/kfd_device.c            |   4 +
 .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c  | 104 ++++++++++++++-------
 .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.h  |  10 +-
 .../drm/amd/amdkfd/kfd_device_queue_manager_cik.c  |   2 +-
 drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c    |   3 +-
 .../gpu/drm/amd/amdkfd/kfd_process_queue_manager.c |   2 +-
 drivers/gpu/drm/amd/include/kgd_kfd_interface.h    |  17 +++-
 drivers/gpu/drm/radeon/radeon_kfd.c                |  21 ++++-
 9 files changed, 130 insertions(+), 55 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
index 3200ff9..8fc5aa3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
@@ -78,48 +78,64 @@ bool amdgpu_amdkfd_load_interface(struct amdgpu_device *adev)
 	return true;
 }
 
 void amdgpu_amdkfd_fini(void)
 {
 	if (kgd2kfd) {
 		kgd2kfd->exit();
 		symbol_put(kgd2kfd_init);
 	}
 }
 
 void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev)
 {
 	if (kgd2kfd)
 		adev->kfd = kgd2kfd->probe((struct kgd_dev *)adev,
 					adev->pdev, kfd2kgd);
 }
 
 void amdgpu_amdkfd_device_init(struct amdgpu_device *adev)
 {
+	int i;
+	int last_valid_bit;
 	if (adev->kfd) {
 		struct kgd2kfd_shared_resources gpu_resources = {
 			.compute_vmid_bitmap = 0xFF00,
-
-			.first_compute_pipe = 1,
-			.compute_pipe_count = 4 - 1,
+			.num_mec = adev->gfx.mec.num_mec,
+			.num_pipe_per_mec = adev->gfx.mec.num_pipe_per_mec,
+			.num_queue_per_pipe = adev->gfx.mec.num_queue_per_pipe
 		};
 
+		/* this is going to have a few of the MSBs set that we need to
+		 * clear */
+		bitmap_complement(gpu_resources.queue_bitmap,
+				  adev->gfx.mec.queue_bitmap,
+				  KGD_MAX_QUEUES);
+
+		/* According to linux/bitmap.h we shouldn't use bitmap_clear if
+		 * nbits is not compile time constant */
+		last_valid_bit = adev->gfx.mec.num_mec
+				* adev->gfx.mec.num_pipe_per_mec
+				* adev->gfx.mec.num_queue_per_pipe;
+		for (i = last_valid_bit; i < KGD_MAX_QUEUES; ++i)
+			clear_bit(i, gpu_resources.queue_bitmap);
+
 		amdgpu_doorbell_get_kfd_info(adev,
 				&gpu_resources.doorbell_physical_address,
 				&gpu_resources.doorbell_aperture_size,
 				&gpu_resources.doorbell_start_offset);
 
 		kgd2kfd->device_init(adev->kfd, &gpu_resources);
 	}
 }
 
 void amdgpu_amdkfd_device_fini(struct amdgpu_device *adev)
 {
 	if (adev->kfd) {
 		kgd2kfd->device_exit(adev->kfd);
 		adev->kfd = NULL;
 	}
 }
 
 void amdgpu_amdkfd_interrupt(struct amdgpu_device *adev,
 		const void *ih_ring_entry)
 {
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index 3f95f7c..88187bf 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -209,40 +209,44 @@ static int iommu_invalid_ppr_cb(struct pci_dev *pdev, int pasid,
 			pasid,
 			address,
 			flags);
 
 	dev = kfd_device_by_pci_dev(pdev);
 	BUG_ON(dev == NULL);
 
 	kfd_signal_iommu_event(dev, pasid, address,
 			flags & PPR_FAULT_WRITE, flags & PPR_FAULT_EXEC);
 
 	return AMD_IOMMU_INV_PRI_RSP_INVALID;
 }
 
 bool kgd2kfd_device_init(struct kfd_dev *kfd,
 			 const struct kgd2kfd_shared_resources *gpu_resources)
 {
 	unsigned int size;
 
 	kfd->shared_resources = *gpu_resources;
 
+	/* We only use the first MEC */
+	if (kfd->shared_resources.num_mec > 1)
+		kfd->shared_resources.num_mec = 1;
+
 	/* calculate max size of mqds needed for queues */
 	size = max_num_of_queues_per_device *
 			kfd->device_info->mqd_size_aligned;
 
 	/*
 	 * calculate max size of runlist packet.
 	 * There can be only 2 packets at once
 	 */
 	size += (KFD_MAX_NUM_OF_PROCESSES * sizeof(struct pm4_map_process) +
 		max_num_of_queues_per_device *
 		sizeof(struct pm4_map_queues) + sizeof(struct pm4_runlist)) * 2;
 
 	/* Add size of HIQ & DIQ */
 	size += KFD_KERNEL_QUEUE_SIZE * 2;
 
 	/* add another 512KB for all other allocations on gart (HPD, fences) */
 	size += 512 * 1024;
 
 	if (kfd->kfd2kgd->init_gtt_mem_allocation(
 			kfd->kgd, size, &kfd->gtt_mem,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index c064dea..955aa30 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -46,55 +46,78 @@ static int create_compute_queue_nocpsch(struct device_queue_manager *dqm,
 
 static int execute_queues_cpsch(struct device_queue_manager *dqm, bool lock);
 static int destroy_queues_cpsch(struct device_queue_manager *dqm,
 				bool preempt_static_queues, bool lock);
 
 static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm,
 					struct queue *q,
 					struct qcm_process_device *qpd);
 
 static void deallocate_sdma_queue(struct device_queue_manager *dqm,
 				unsigned int sdma_queue_id);
 
 static inline
 enum KFD_MQD_TYPE get_mqd_type_from_queue_type(enum kfd_queue_type type)
 {
 	if (type == KFD_QUEUE_TYPE_SDMA)
 		return KFD_MQD_TYPE_SDMA;
 	return KFD_MQD_TYPE_CP;
 }
 
-unsigned int get_first_pipe(struct device_queue_manager *dqm)
+static bool is_pipe_enabled(struct device_queue_manager *dqm, int mec, int pipe)
+{
+	int i;
+	int pipe_offset = mec * dqm->dev->shared_resources.num_pipe_per_mec
+		+ pipe * dqm->dev->shared_resources.num_queue_per_pipe;
+
+	/* queue is available for KFD usage if bit is 1 */
+	for (i = 0; i <  dqm->dev->shared_resources.num_queue_per_pipe; ++i)
+		if (test_bit(pipe_offset + i,
+			      dqm->dev->shared_resources.queue_bitmap))
+			return true;
+	return false;
+}
+
+unsigned int get_mec_num(struct device_queue_manager *dqm)
 {
 	BUG_ON(!dqm || !dqm->dev);
-	return dqm->dev->shared_resources.first_compute_pipe;
+
+	return dqm->dev->shared_resources.num_mec;
 }
 
-unsigned int get_pipes_num(struct device_queue_manager *dqm)
+unsigned int get_queues_num(struct device_queue_manager *dqm)
 {
 	BUG_ON(!dqm || !dqm->dev);
-	return dqm->dev->shared_resources.compute_pipe_count;
+	return bitmap_weight(dqm->dev->shared_resources.queue_bitmap,
+				KGD_MAX_QUEUES);
 }
 
-static inline unsigned int get_pipes_num_cpsch(void)
+unsigned int get_queues_per_pipe(struct device_queue_manager *dqm)
 {
-	return PIPE_PER_ME_CP_SCHEDULING;
+	BUG_ON(!dqm || !dqm->dev);
+	return dqm->dev->shared_resources.num_queue_per_pipe;
+}
+
+unsigned int get_pipes_per_mec(struct device_queue_manager *dqm)
+{
+	BUG_ON(!dqm || !dqm->dev);
+	return dqm->dev->shared_resources.num_pipe_per_mec;
 }
 
 void program_sh_mem_settings(struct device_queue_manager *dqm,
 					struct qcm_process_device *qpd)
 {
 	return dqm->dev->kfd2kgd->program_sh_mem_settings(
 						dqm->dev->kgd, qpd->vmid,
 						qpd->sh_mem_config,
 						qpd->sh_mem_ape1_base,
 						qpd->sh_mem_ape1_limit,
 						qpd->sh_mem_bases);
 }
 
 static int allocate_vmid(struct device_queue_manager *dqm,
 			struct qcm_process_device *qpd,
 			struct queue *q)
 {
 	int bit, allocated_vmid;
 
 	if (dqm->vmid_bitmap == 0)
@@ -183,63 +206,67 @@ static int create_queue_nocpsch(struct device_queue_manager *dqm,
 
 	/*
 	 * Unconditionally increment this counter, regardless of the queue's
 	 * type or whether the queue is active.
 	 */
 	dqm->total_queue_count++;
 	pr_debug("Total of %d queues are accountable so far\n",
 			dqm->total_queue_count);
 
 	mutex_unlock(&dqm->lock);
 	return 0;
 }
 
 static int allocate_hqd(struct device_queue_manager *dqm, struct queue *q)
 {
 	bool set;
 	int pipe, bit, i;
 
 	set = false;
 
-	for (pipe = dqm->next_pipe_to_allocate, i = 0; i < get_pipes_num(dqm);
-			pipe = ((pipe + 1) % get_pipes_num(dqm)), ++i) {
+	for (pipe = dqm->next_pipe_to_allocate, i = 0; i < get_pipes_per_mec(dqm);
+			pipe = ((pipe + 1) % get_pipes_per_mec(dqm)), ++i) {
+
+		if (!is_pipe_enabled(dqm, 0, pipe))
+			continue;
+
 		if (dqm->allocated_queues[pipe] != 0) {
 			bit = find_first_bit(
 				(unsigned long *)&dqm->allocated_queues[pipe],
-				QUEUES_PER_PIPE);
+				get_queues_per_pipe(dqm));
 
 			clear_bit(bit,
 				(unsigned long *)&dqm->allocated_queues[pipe]);
 			q->pipe = pipe;
 			q->queue = bit;
 			set = true;
 			break;
 		}
 	}
 
 	if (!set)
 		return -EBUSY;
 
 	pr_debug("kfd: DQM %s hqd slot - pipe (%d) queue(%d)\n",
 				__func__, q->pipe, q->queue);
 	/* horizontal hqd allocation */
-	dqm->next_pipe_to_allocate = (pipe + 1) % get_pipes_num(dqm);
+	dqm->next_pipe_to_allocate = (pipe + 1) % get_pipes_per_mec(dqm);
 
 	return 0;
 }
 
 static inline void deallocate_hqd(struct device_queue_manager *dqm,
 				struct queue *q)
 {
 	set_bit(q->queue, (unsigned long *)&dqm->allocated_queues[q->pipe]);
 }
 
 static int create_compute_queue_nocpsch(struct device_queue_manager *dqm,
 					struct queue *q,
 					struct qcm_process_device *qpd)
 {
 	int retval;
 	struct mqd_manager *mqd;
 
 	BUG_ON(!dqm || !q || !qpd);
 
 	mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_COMPUTE);
@@ -452,95 +479,84 @@ static int unregister_process_nocpsch(struct device_queue_manager *dqm,
 out:
 	mutex_unlock(&dqm->lock);
 	return retval;
 }
 
 static int
 set_pasid_vmid_mapping(struct device_queue_manager *dqm, unsigned int pasid,
 			unsigned int vmid)
 {
 	uint32_t pasid_mapping;
 
 	pasid_mapping = (pasid == 0) ? 0 :
 		(uint32_t)pasid |
 		ATC_VMID_PASID_MAPPING_VALID;
 
 	return dqm->dev->kfd2kgd->set_pasid_vmid_mapping(
 						dqm->dev->kgd, pasid_mapping,
 						vmid);
 }
 
-int init_pipelines(struct device_queue_manager *dqm,
-			unsigned int pipes_num, unsigned int first_pipe)
-{
-	BUG_ON(!dqm || !dqm->dev);
-
-	pr_debug("kfd: In func %s\n", __func__);
-
-	return 0;
-}
-
 static void init_interrupts(struct device_queue_manager *dqm)
 {
 	unsigned int i;
 
 	BUG_ON(dqm == NULL);
 
-	for (i = 0 ; i < get_pipes_num(dqm) ; i++)
-		dqm->dev->kfd2kgd->init_interrupts(dqm->dev->kgd,
-				i + get_first_pipe(dqm));
+	for (i = 0 ; i < get_pipes_per_mec(dqm) ; i++)
+		if (is_pipe_enabled(dqm, 0, i))
+			dqm->dev->kfd2kgd->init_interrupts(dqm->dev->kgd, i);
 }
 
 static int init_scheduler(struct device_queue_manager *dqm)
 {
-	int retval;
+	int retval = 0;
 
 	BUG_ON(!dqm);
 
 	pr_debug("kfd: In %s\n", __func__);
 
-	retval = init_pipelines(dqm, get_pipes_num(dqm), get_first_pipe(dqm));
 	return retval;
 }
 
 static int initialize_nocpsch(struct device_queue_manager *dqm)
 {
 	int i;
 
 	BUG_ON(!dqm);
 
 	pr_debug("kfd: In func %s num of pipes: %d\n",
-			__func__, get_pipes_num(dqm));
+			__func__, get_pipes_per_mec(dqm));
 
 	mutex_init(&dqm->lock);
 	INIT_LIST_HEAD(&dqm->queues);
 	dqm->queue_count = dqm->next_pipe_to_allocate = 0;
 	dqm->sdma_queue_count = 0;
-	dqm->allocated_queues = kcalloc(get_pipes_num(dqm),
+	dqm->allocated_queues = kcalloc(get_pipes_per_mec(dqm),
 					sizeof(unsigned int), GFP_KERNEL);
 	if (!dqm->allocated_queues) {
 		mutex_destroy(&dqm->lock);
 		return -ENOMEM;
 	}
 
-	for (i = 0; i < get_pipes_num(dqm); i++)
-		dqm->allocated_queues[i] = (1 << QUEUES_PER_PIPE) - 1;
+	for (i = 0; i < get_pipes_per_mec(dqm); i++)
+		dqm->allocated_queues[i] = (1 << get_queues_per_pipe(dqm)) - 1;
 
 	dqm->vmid_bitmap = (1 << VMID_PER_DEVICE) - 1;
 	dqm->sdma_bitmap = (1 << CIK_SDMA_QUEUES) - 1;
 
 	init_scheduler(dqm);
 	return 0;
 }
 
 static void uninitialize_nocpsch(struct device_queue_manager *dqm)
 {
 	int i;
 
 	BUG_ON(!dqm);
 
 	BUG_ON(dqm->queue_count > 0 || dqm->processes_count > 0);
 
 	kfree(dqm->allocated_queues);
 	for (i = 0 ; i < KFD_MQD_TYPE_MAX ; i++)
 		kfree(dqm->mqds[i]);
 	mutex_destroy(&dqm->lock);
@@ -613,71 +629,91 @@ static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm,
 		return retval;
 	}
 
 	retval = mqd->load_mqd(mqd, q->mqd, 0,
 				0, NULL);
 	if (retval != 0) {
 		deallocate_sdma_queue(dqm, q->sdma_id);
 		mqd->uninit_mqd(mqd, q->mqd, q->mqd_mem_obj);
 		return retval;
 	}
 
 	return 0;
 }
 
 /*
  * Device Queue Manager implementation for cp scheduler
  */
 
 static int set_sched_resources(struct device_queue_manager *dqm)
 {
+	int i, mec;
 	struct scheduling_resources res;
-	unsigned int queue_num, queue_mask;
 
 	BUG_ON(!dqm);
 
 	pr_debug("kfd: In func %s\n", __func__);
 
-	queue_num = get_pipes_num_cpsch() * QUEUES_PER_PIPE;
-	queue_mask = (1 << queue_num) - 1;
 	res.vmid_mask = (1 << VMID_PER_DEVICE) - 1;
 	res.vmid_mask <<= KFD_VMID_START_OFFSET;
-	res.queue_mask = queue_mask << (get_first_pipe(dqm) * QUEUES_PER_PIPE);
+
+	res.queue_mask = 0;
+	for (i = 0; i < KGD_MAX_QUEUES; ++i) {
+		mec = (i / dqm->dev->shared_resources.num_queue_per_pipe)
+			/ dqm->dev->shared_resources.num_pipe_per_mec;
+
+		if (!test_bit(i, dqm->dev->shared_resources.queue_bitmap))
+			continue;
+
+		/* only acquire queues from the first MEC */
+		if (mec > 0)
+			continue;
+
+		/* This situation may be hit in the future if a new HW
+		 * generation exposes more than 64 queues. If so, the
+		 * definition of res.queue_mask needs updating */
+		if (WARN_ON(i > (sizeof(res.queue_mask)*8))) {
+			pr_err("Invalid queue enabled by amdgpu: %d\n", i);
+			break;
+		}
+
+		res.queue_mask |= (1ull << i);
+	}
 	res.gws_mask = res.oac_mask = res.gds_heap_base =
 						res.gds_heap_size = 0;
 
 	pr_debug("kfd: scheduling resources:\n"
 			"      vmid mask: 0x%8X\n"
 			"      queue mask: 0x%8llX\n",
 			res.vmid_mask, res.queue_mask);
 
 	return pm_send_set_resources(&dqm->packets, &res);
 }
 
 static int initialize_cpsch(struct device_queue_manager *dqm)
 {
 	int retval;
 
 	BUG_ON(!dqm);
 
 	pr_debug("kfd: In func %s num of pipes: %d\n",
-			__func__, get_pipes_num_cpsch());
+			__func__, get_pipes_per_mec(dqm));
 
 	mutex_init(&dqm->lock);
 	INIT_LIST_HEAD(&dqm->queues);
 	dqm->queue_count = dqm->processes_count = 0;
 	dqm->sdma_queue_count = 0;
 	dqm->active_runlist = false;
 	retval = dqm->ops_asic_specific.initialize(dqm);
 	if (retval != 0)
 		goto fail_init_pipelines;
 
 	return 0;
 
 fail_init_pipelines:
 	mutex_destroy(&dqm->lock);
 	return retval;
 }
 
 static int start_cpsch(struct device_queue_manager *dqm)
 {
 	struct device_process_node *node;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
index a625b91..66b9615 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
@@ -13,42 +13,40 @@
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  * OTHER DEALINGS IN THE SOFTWARE.
  *
  */
 
 #ifndef KFD_DEVICE_QUEUE_MANAGER_H_
 #define KFD_DEVICE_QUEUE_MANAGER_H_
 
 #include <linux/rwsem.h>
 #include <linux/list.h>
 #include "kfd_priv.h"
 #include "kfd_mqd_manager.h"
 
 #define QUEUE_PREEMPT_DEFAULT_TIMEOUT_MS	(500)
-#define QUEUES_PER_PIPE				(8)
-#define PIPE_PER_ME_CP_SCHEDULING		(3)
 #define CIK_VMID_NUM				(8)
 #define KFD_VMID_START_OFFSET			(8)
 #define VMID_PER_DEVICE				CIK_VMID_NUM
 #define KFD_DQM_FIRST_PIPE			(0)
 #define CIK_SDMA_QUEUES				(4)
 #define CIK_SDMA_QUEUES_PER_ENGINE		(2)
 #define CIK_SDMA_ENGINE_NUM			(2)
 
 struct device_process_node {
 	struct qcm_process_device *qpd;
 	struct list_head list;
 };
 
 /**
  * struct device_queue_manager_ops
  *
  * @create_queue: Queue creation routine.
  *
  * @destroy_queue: Queue destruction routine.
  *
@@ -165,37 +163,37 @@ struct device_queue_manager {
 	unsigned int		processes_count;
 	unsigned int		queue_count;
 	unsigned int		sdma_queue_count;
 	unsigned int		total_queue_count;
 	unsigned int		next_pipe_to_allocate;
 	unsigned int		*allocated_queues;
 	unsigned int		sdma_bitmap;
 	unsigned int		vmid_bitmap;
 	uint64_t		pipelines_addr;
 	struct kfd_mem_obj	*pipeline_mem;
 	uint64_t		fence_gpu_addr;
 	unsigned int		*fence_addr;
 	struct kfd_mem_obj	*fence_mem;
 	bool			active_runlist;
 };
 
 void device_queue_manager_init_cik(struct device_queue_manager_asic_ops *ops);
 void device_queue_manager_init_vi(struct device_queue_manager_asic_ops *ops);
 void program_sh_mem_settings(struct device_queue_manager *dqm,
 					struct qcm_process_device *qpd);
-int init_pipelines(struct device_queue_manager *dqm,
-		unsigned int pipes_num, unsigned int first_pipe);
-unsigned int get_first_pipe(struct device_queue_manager *dqm);
-unsigned int get_pipes_num(struct device_queue_manager *dqm);
+unsigned int get_mec_num(struct device_queue_manager *dqm);
+unsigned int get_queues_num(struct device_queue_manager *dqm);
+unsigned int get_queues_per_pipe(struct device_queue_manager *dqm);
+unsigned int get_pipes_per_mec(struct device_queue_manager *dqm);
 
 static inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd)
 {
 	return (pdd->lds_base >> 16) & 0xFF;
 }
 
 static inline unsigned int
 get_sh_mem_bases_nybble_64(struct kfd_process_device *pdd)
 {
 	return (pdd->lds_base >> 60) & 0x0E;
 }
 
 #endif /* KFD_DEVICE_QUEUE_MANAGER_H_ */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c
index c6f435a..48dc056 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c
@@ -134,22 +134,22 @@ static int register_process_cik(struct device_queue_manager *dqm,
 }
 
 static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q,
 				struct qcm_process_device *qpd)
 {
 	uint32_t value = (1 << SDMA0_RLC0_VIRTUAL_ADDR__ATC__SHIFT);
 
 	if (q->process->is_32bit_user_mode)
 		value |= (1 << SDMA0_RLC0_VIRTUAL_ADDR__PTR32__SHIFT) |
 				get_sh_mem_bases_32(qpd_to_pdd(qpd));
 	else
 		value |= ((get_sh_mem_bases_nybble_64(qpd_to_pdd(qpd))) <<
 				SDMA0_RLC0_VIRTUAL_ADDR__SHARED_BASE__SHIFT) &
 				SDMA0_RLC0_VIRTUAL_ADDR__SHARED_BASE_MASK;
 
 	q->properties.sdma_vm_addr = value;
 }
 
 static int initialize_cpsch_cik(struct device_queue_manager *dqm)
 {
-	return init_pipelines(dqm, get_pipes_num(dqm), get_first_pipe(dqm));
+	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
index ca8c093..7131998 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
@@ -48,42 +48,41 @@ static unsigned int build_pm4_header(unsigned int opcode, size_t packet_size)
 	header.count = packet_size/sizeof(uint32_t) - 2;
 	header.type = PM4_TYPE_3;
 
 	return header.u32all;
 }
 
 static void pm_calc_rlib_size(struct packet_manager *pm,
 				unsigned int *rlib_size,
 				bool *over_subscription)
 {
 	unsigned int process_count, queue_count;
 	unsigned int map_queue_size;
 
 	BUG_ON(!pm || !rlib_size || !over_subscription);
 
 	process_count = pm->dqm->processes_count;
 	queue_count = pm->dqm->queue_count;
 
 	/* check if there is over subscription*/
 	*over_subscription = false;
-	if ((process_count > 1) ||
-		queue_count > PIPE_PER_ME_CP_SCHEDULING * QUEUES_PER_PIPE) {
+	if ((process_count > 1) || queue_count > get_queues_num(pm->dqm)) {
 		*over_subscription = true;
 		pr_debug("kfd: over subscribed runlist\n");
 	}
 
 	map_queue_size =
 		(pm->dqm->dev->device_info->asic_family == CHIP_CARRIZO) ?
 		sizeof(struct pm4_mes_map_queues) :
 		sizeof(struct pm4_map_queues);
 	/* calculate run list ib allocation size */
 	*rlib_size = process_count * sizeof(struct pm4_map_process) +
 		     queue_count * map_queue_size;
 
 	/*
 	 * Increase the allocation size in case we need a chained run list
 	 * when over subscription
 	 */
 	if (*over_subscription)
 		*rlib_size += sizeof(struct pm4_runlist);
 
 	pr_debug("kfd: runlist ib size %d\n", *rlib_size);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
index e1fb40b..32cdf2b 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
@@ -192,41 +192,41 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 	if (retval != 0)
 		return retval;
 
 	if (list_empty(&pqm->queues)) {
 		pdd->qpd.pqm = pqm;
 		dev->dqm->ops.register_process(dev->dqm, &pdd->qpd);
 	}
 
 	pqn = kzalloc(sizeof(struct process_queue_node), GFP_KERNEL);
 	if (!pqn) {
 		retval = -ENOMEM;
 		goto err_allocate_pqn;
 	}
 
 	switch (type) {
 	case KFD_QUEUE_TYPE_SDMA:
 	case KFD_QUEUE_TYPE_COMPUTE:
 		/* check if there is over subscription */
 		if ((sched_policy == KFD_SCHED_POLICY_HWS_NO_OVERSUBSCRIPTION) &&
 		((dev->dqm->processes_count >= VMID_PER_DEVICE) ||
-		(dev->dqm->queue_count >= PIPE_PER_ME_CP_SCHEDULING * QUEUES_PER_PIPE))) {
+		(dev->dqm->queue_count >= get_queues_num(dev->dqm)))) {
 			pr_err("kfd: over-subscription is not allowed in radeon_kfd.sched_policy == 1\n");
 			retval = -EPERM;
 			goto err_create_queue;
 		}
 
 		retval = create_cp_queue(pqm, dev, &q, &q_properties, f, *qid);
 		if (retval != 0)
 			goto err_create_queue;
 		pqn->q = q;
 		pqn->kq = NULL;
 		retval = dev->dqm->ops.create_queue(dev->dqm, q, &pdd->qpd,
 						&q->properties.vmid);
 		pr_debug("DQM returned %d for create_queue\n", retval);
 		print_queue(q);
 		break;
 	case KFD_QUEUE_TYPE_DIQ:
 		kq = kernel_queue_init(dev, KFD_QUEUE_TYPE_DIQ);
 		if (kq == NULL) {
 			retval = -ENOMEM;
 			goto err_create_queue;
diff --git a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
index 67f6d19..91ef148 100644
--- a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
+++ b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
@@ -12,78 +12,85 @@
  * all copies or substantial portions of the Software.
  *
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  * OTHER DEALINGS IN THE SOFTWARE.
  */
 
 /*
  * This file defines the private interface between the
  * AMD kernel graphics drivers and the AMD KFD.
  */
 
 #ifndef KGD_KFD_INTERFACE_H_INCLUDED
 #define KGD_KFD_INTERFACE_H_INCLUDED
 
 #include <linux/types.h>
+#include <linux/bitmap.h>
 
 struct pci_dev;
 
-#define KFD_INTERFACE_VERSION 1
+#define KFD_INTERFACE_VERSION 2
 #define KGD_MAX_QUEUES 128
 
 struct kfd_dev;
 struct kgd_dev;
 
 struct kgd_mem;
 
 enum kgd_memory_pool {
 	KGD_POOL_SYSTEM_CACHEABLE = 1,
 	KGD_POOL_SYSTEM_WRITECOMBINE = 2,
 	KGD_POOL_FRAMEBUFFER = 3,
 };
 
 enum kgd_engine_type {
 	KGD_ENGINE_PFP = 1,
 	KGD_ENGINE_ME,
 	KGD_ENGINE_CE,
 	KGD_ENGINE_MEC1,
 	KGD_ENGINE_MEC2,
 	KGD_ENGINE_RLC,
 	KGD_ENGINE_SDMA1,
 	KGD_ENGINE_SDMA2,
 	KGD_ENGINE_MAX
 };
 
 struct kgd2kfd_shared_resources {
 	/* Bit n == 1 means VMID n is available for KFD. */
 	unsigned int compute_vmid_bitmap;
 
-	/* Compute pipes are counted starting from MEC0/pipe0 as 0. */
-	unsigned int first_compute_pipe;
+	/* number of mec available from the hardware */
+	uint32_t num_mec;
 
-	/* Number of MEC pipes available for KFD. */
-	unsigned int compute_pipe_count;
+	/* number of pipes per mec */
+	uint32_t num_pipe_per_mec;
+
+	/* number of queues per pipe */
+	uint32_t num_queue_per_pipe;
+
+	/* Bit n == 1 means Queue n is available for KFD */
+	DECLARE_BITMAP(queue_bitmap, KGD_MAX_QUEUES);
 
 	/* Base address of doorbell aperture. */
 	phys_addr_t doorbell_physical_address;
 
 	/* Size in bytes of doorbell aperture. */
 	size_t doorbell_aperture_size;
 
 	/* Number of bytes at start of aperture reserved for KGD. */
 	size_t doorbell_start_offset;
 };
 
 /**
  * struct kfd2kgd_calls
  *
  * @init_gtt_mem_allocation: Allocate a buffer on the gart aperture.
  * The buffer can be used for mqds, hpds, kernel queue, fence and runlists
  *
  * @free_gtt_mem: Frees a buffer that was allocated on the gart aperture
  *
  * @get_vmem_size: Retrieves (physical) size of VRAM
diff --git a/drivers/gpu/drm/radeon/radeon_kfd.c b/drivers/gpu/drm/radeon/radeon_kfd.c
index a06e3b1..699fe7f 100644
--- a/drivers/gpu/drm/radeon/radeon_kfd.c
+++ b/drivers/gpu/drm/radeon/radeon_kfd.c
@@ -162,48 +162,63 @@ int radeon_kfd_init(void)
 	return ret;
 }
 
 void radeon_kfd_fini(void)
 {
 	if (kgd2kfd) {
 		kgd2kfd->exit();
 		symbol_put(kgd2kfd_init);
 	}
 }
 
 void radeon_kfd_device_probe(struct radeon_device *rdev)
 {
 	if (kgd2kfd)
 		rdev->kfd = kgd2kfd->probe((struct kgd_dev *)rdev,
 			rdev->pdev, &kfd2kgd);
 }
 
 void radeon_kfd_device_init(struct radeon_device *rdev)
 {
+	int i, queue, pipe, mec;
+
 	if (rdev->kfd) {
 		struct kgd2kfd_shared_resources gpu_resources = {
 			.compute_vmid_bitmap = 0xFF00,
-
-			.first_compute_pipe = 1,
-			.compute_pipe_count = 4 - 1,
+			.num_mec = 1,
+			.num_pipe_per_mec = 4,
+			.num_queue_per_pipe = 8
 		};
 
+		bitmap_zero(gpu_resources.queue_bitmap, KGD_MAX_QUEUES);
+
+		for (i = 0; i < KGD_MAX_QUEUES; ++i) {
+			queue = i % gpu_resources.num_queue_per_pipe;
+			pipe = (i / gpu_resources.num_queue_per_pipe)
+				% gpu_resources.num_pipe_per_mec;
+			mec = (i / gpu_resources.num_queue_per_pipe)
+				/ gpu_resources.num_pipe_per_mec;
+
+			if (mec == 0 && pipe > 0)
+				set_bit(i, gpu_resources.queue_bitmap);
+		}
+
 		radeon_doorbell_get_kfd_info(rdev,
 				&gpu_resources.doorbell_physical_address,
 				&gpu_resources.doorbell_aperture_size,
 				&gpu_resources.doorbell_start_offset);
 
 		kgd2kfd->device_init(rdev->kfd, &gpu_resources);
 	}
 }
 
 void radeon_kfd_device_fini(struct radeon_device *rdev)
 {
 	if (rdev->kfd) {
 		kgd2kfd->device_exit(rdev->kfd);
 		rdev->kfd = NULL;
 	}
 }
 
 void radeon_kfd_interrupt(struct radeon_device *rdev, const void *ih_ring_entry)
 {
 	if (rdev->kfd)
-- 
2.9.3



More information about the amd-gfx mailing list