Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit b1f389ec authored by Dave Airlie's avatar Dave Airlie
Browse files

Merge tag 'drm-amdkfd-next-2015-03-25' of git://people.freedesktop.org/~gabbayo/linux into drm-next 

- Allow amdkfd to work with multiple kgd instances. This is in preparation for
  AMD's new open source kernel graphic driver (amdgpu), and for the new
  AMD APU, Carrizo.

- Convert timestamping to use 64bit time accessors

- Three other minor changes.

* tag 'drm-amdkfd-next-2015-03-25' of git://people.freedesktop.org/~gabbayo/linux:
  drm/amdkfd: Add multiple kgd support
  drm/amdkfd: Convert timestamping to use 64bit time accessors
  drm/amdkfd: add debug prints for process teardown
  drm/amdkfd: Remove unused field from struct qcm_process_device
  drm/amdkfd: rename fence_wait_timeout
parents 517bc045 cea405b1

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

+7 −6
Original line number Original line Diff line number Diff line
@@ -435,21 +435,22 @@ static int kfd_ioctl_get_clock_counters(struct file *filep, 
{

{

	struct kfd_ioctl_get_clock_counters_args *args = data;

	struct kfd_ioctl_get_clock_counters_args *args = data;

	struct kfd_dev *dev;

	struct kfd_dev *dev;

	struct timespec time;

	struct timespec64 time;





	dev = kfd_device_by_id(args->gpu_id);

	dev = kfd_device_by_id(args->gpu_id);

	if (dev == NULL)

	if (dev == NULL)

		return -EINVAL;

		return -EINVAL;





	/* Reading GPU clock counter from KGD */

	/* Reading GPU clock counter from KGD */

	args->gpu_clock_counter = kfd2kgd->get_gpu_clock_counter(dev->kgd);

	args->gpu_clock_counter =

		dev->kfd2kgd->get_gpu_clock_counter(dev->kgd);





	/* No access to rdtsc. Using raw monotonic time */

	/* No access to rdtsc. Using raw monotonic time */

	getrawmonotonic(&time);

	getrawmonotonic64(&time);

	args->cpu_clock_counter = (uint64_t)timespec_to_ns(&time);

	args->cpu_clock_counter = (uint64_t)timespec64_to_ns(&time);





	get_monotonic_boottime(&time);

	get_monotonic_boottime64(&time);

	args->system_clock_counter = (uint64_t)timespec_to_ns(&time);

	args->system_clock_counter = (uint64_t)timespec64_to_ns(&time);





	/* Since the counter is in nano-seconds we use 1GHz frequency */

	/* Since the counter is in nano-seconds we use 1GHz frequency */

	args->system_clock_freq = 1000000000;

	args->system_clock_freq = 1000000000;

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

+12 −5
Original line number Original line Diff line number Diff line
@@ -94,7 +94,8 @@ static const struct kfd_device_info *lookup_device_info(unsigned short did) 
	return NULL;

	return NULL;

}

}





struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd, struct pci_dev *pdev)

struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd,

	struct pci_dev *pdev, const struct kfd2kgd_calls *f2g)

{

{

	struct kfd_dev *kfd;

	struct kfd_dev *kfd;
@@ -112,6 +113,11 @@ struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd, struct pci_dev *pdev) 
	kfd->device_info = device_info;

	kfd->device_info = device_info;

	kfd->pdev = pdev;

	kfd->pdev = pdev;

	kfd->init_complete = false;

	kfd->init_complete = false;

	kfd->kfd2kgd = f2g;



	mutex_init(&kfd->doorbell_mutex);

	memset(&kfd->doorbell_available_index, 0,

		sizeof(kfd->doorbell_available_index));





	return kfd;

	return kfd;

}

}
@@ -200,7 +206,8 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, 
	/* add another 512KB for all other allocations on gart (HPD, fences) */

	/* add another 512KB for all other allocations on gart (HPD, fences) */

	size += 512 * 1024;

	size += 512 * 1024;





	if (kfd2kgd->init_gtt_mem_allocation(kfd->kgd, size, &kfd->gtt_mem,

	if (kfd->kfd2kgd->init_gtt_mem_allocation(

			kfd->kgd, size, &kfd->gtt_mem,

			&kfd->gtt_start_gpu_addr, &kfd->gtt_start_cpu_ptr)){

			&kfd->gtt_start_gpu_addr, &kfd->gtt_start_cpu_ptr)){

		dev_err(kfd_device,

		dev_err(kfd_device,

			"Could not allocate %d bytes for device (%x:%x)\n",

			"Could not allocate %d bytes for device (%x:%x)\n",
@@ -270,7 +277,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, 
kfd_topology_add_device_error:

kfd_topology_add_device_error:

	kfd_gtt_sa_fini(kfd);

	kfd_gtt_sa_fini(kfd);

kfd_gtt_sa_init_error:

kfd_gtt_sa_init_error:

	kfd2kgd->free_gtt_mem(kfd->kgd, kfd->gtt_mem);

	kfd->kfd2kgd->free_gtt_mem(kfd->kgd, kfd->gtt_mem);

	dev_err(kfd_device,

	dev_err(kfd_device,

		"device (%x:%x) NOT added due to errors\n",

		"device (%x:%x) NOT added due to errors\n",

		kfd->pdev->vendor, kfd->pdev->device);

		kfd->pdev->vendor, kfd->pdev->device);
@@ -285,7 +292,7 @@ void kgd2kfd_device_exit(struct kfd_dev *kfd) 
		amd_iommu_free_device(kfd->pdev);

		amd_iommu_free_device(kfd->pdev);

		kfd_topology_remove_device(kfd);

		kfd_topology_remove_device(kfd);

		kfd_gtt_sa_fini(kfd);

		kfd_gtt_sa_fini(kfd);

		kfd2kgd->free_gtt_mem(kfd->kgd, kfd->gtt_mem);

		kfd->kfd2kgd->free_gtt_mem(kfd->kgd, kfd->gtt_mem);

	}

	}





	kfree(kfd);

	kfree(kfd);

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

+11 −7
Original line number Original line Diff line number Diff line
@@ -82,7 +82,8 @@ static inline unsigned int get_pipes_num_cpsch(void) 
void program_sh_mem_settings(struct device_queue_manager *dqm,

void program_sh_mem_settings(struct device_queue_manager *dqm,

					struct qcm_process_device *qpd)

					struct qcm_process_device *qpd)

{

{

	return kfd2kgd->program_sh_mem_settings(dqm->dev->kgd, qpd->vmid,

	return dqm->dev->kfd2kgd->program_sh_mem_settings(

						dqm->dev->kgd, qpd->vmid,

						qpd->sh_mem_config,

						qpd->sh_mem_config,

						qpd->sh_mem_ape1_base,

						qpd->sh_mem_ape1_base,

						qpd->sh_mem_ape1_limit,

						qpd->sh_mem_ape1_limit,
@@ -457,9 +458,12 @@ set_pasid_vmid_mapping(struct device_queue_manager *dqm, unsigned int pasid, 
{

{

	uint32_t pasid_mapping;

	uint32_t pasid_mapping;





	pasid_mapping = (pasid == 0) ? 0 : (uint32_t)pasid |

	pasid_mapping = (pasid == 0) ? 0 :

		(uint32_t)pasid |

		ATC_VMID_PASID_MAPPING_VALID;

		ATC_VMID_PASID_MAPPING_VALID;

	return kfd2kgd->set_pasid_vmid_mapping(dqm->dev->kgd, pasid_mapping,



	return dqm->dev->kfd2kgd->set_pasid_vmid_mapping(

						dqm->dev->kgd, pasid_mapping,

						vmid);

						vmid);

}

}
@@ -511,7 +515,7 @@ int init_pipelines(struct device_queue_manager *dqm, 
		pipe_hpd_addr = dqm->pipelines_addr + i * CIK_HPD_EOP_BYTES;

		pipe_hpd_addr = dqm->pipelines_addr + i * CIK_HPD_EOP_BYTES;

		pr_debug("kfd: pipeline address %llX\n", pipe_hpd_addr);

		pr_debug("kfd: pipeline address %llX\n", pipe_hpd_addr);

		/* = log2(bytes/4)-1 */

		/* = log2(bytes/4)-1 */

		kfd2kgd->init_pipeline(dqm->dev->kgd, inx,

		dqm->dev->kfd2kgd->init_pipeline(dqm->dev->kgd, inx,

				CIK_HPD_EOP_BYTES_LOG2 - 3, pipe_hpd_addr);

				CIK_HPD_EOP_BYTES_LOG2 - 3, pipe_hpd_addr);

	}

	}
@@ -897,7 +901,7 @@ static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q, 
	return retval;

	return retval;

}

}





static int fence_wait_timeout(unsigned int *fence_addr,

static int amdkfd_fence_wait_timeout(unsigned int *fence_addr,

				unsigned int fence_value,

				unsigned int fence_value,

				unsigned long timeout)

				unsigned long timeout)

{

{
@@ -953,7 +957,7 @@ static int destroy_queues_cpsch(struct device_queue_manager *dqm, bool lock) 
	pm_send_query_status(&dqm->packets, dqm->fence_gpu_addr,

	pm_send_query_status(&dqm->packets, dqm->fence_gpu_addr,

				KFD_FENCE_COMPLETED);

				KFD_FENCE_COMPLETED);

	/* should be timed out */

	/* should be timed out */

	fence_wait_timeout(dqm->fence_addr, KFD_FENCE_COMPLETED,

	amdkfd_fence_wait_timeout(dqm->fence_addr, KFD_FENCE_COMPLETED,

				QUEUE_PREEMPT_DEFAULT_TIMEOUT_MS);

				QUEUE_PREEMPT_DEFAULT_TIMEOUT_MS);

	pm_release_ib(&dqm->packets);

	pm_release_ib(&dqm->packets);

	dqm->active_runlist = false;

	dqm->active_runlist = false;

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

+7 −10
Original line number Original line Diff line number Diff line
@@ -32,9 +32,6 @@ 
 * and that's assures that any user process won't get access to the

 * and that's assures that any user process won't get access to the

 * kernel doorbells page

 * kernel doorbells page

 */

 */

static DEFINE_MUTEX(doorbell_mutex);

static unsigned long doorbell_available_index[

	DIV_ROUND_UP(KFD_MAX_NUM_OF_QUEUES_PER_PROCESS, BITS_PER_LONG)] = { 0 };





#define KERNEL_DOORBELL_PASID 1

#define KERNEL_DOORBELL_PASID 1

#define KFD_SIZE_OF_DOORBELL_IN_BYTES 4

#define KFD_SIZE_OF_DOORBELL_IN_BYTES 4
@@ -170,12 +167,12 @@ u32 __iomem *kfd_get_kernel_doorbell(struct kfd_dev *kfd, 




	BUG_ON(!kfd || !doorbell_off);

	BUG_ON(!kfd || !doorbell_off);





	mutex_lock(&doorbell_mutex);

	mutex_lock(&kfd->doorbell_mutex);

	inx = find_first_zero_bit(doorbell_available_index,

	inx = find_first_zero_bit(kfd->doorbell_available_index,

					KFD_MAX_NUM_OF_QUEUES_PER_PROCESS);

					KFD_MAX_NUM_OF_QUEUES_PER_PROCESS);





	__set_bit(inx, doorbell_available_index);

	__set_bit(inx, kfd->doorbell_available_index);

	mutex_unlock(&doorbell_mutex);

	mutex_unlock(&kfd->doorbell_mutex);





	if (inx >= KFD_MAX_NUM_OF_QUEUES_PER_PROCESS)

	if (inx >= KFD_MAX_NUM_OF_QUEUES_PER_PROCESS)

		return NULL;

		return NULL;
@@ -203,9 +200,9 @@ void kfd_release_kernel_doorbell(struct kfd_dev *kfd, u32 __iomem *db_addr) 




	inx = (unsigned int)(db_addr - kfd->doorbell_kernel_ptr);

	inx = (unsigned int)(db_addr - kfd->doorbell_kernel_ptr);





	mutex_lock(&doorbell_mutex);

	mutex_lock(&kfd->doorbell_mutex);

	__clear_bit(inx, doorbell_available_index);

	__clear_bit(inx, kfd->doorbell_available_index);

	mutex_unlock(&doorbell_mutex);

	mutex_unlock(&kfd->doorbell_mutex);

}

}





inline void write_kernel_doorbell(u32 __iomem *db, u32 value)

inline void write_kernel_doorbell(u32 __iomem *db, u32 value)

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

+1 −11
Original line number Original line Diff line number Diff line
@@ -34,7 +34,6 @@ 
#define KFD_DRIVER_MINOR	7

#define KFD_DRIVER_MINOR	7

#define KFD_DRIVER_PATCHLEVEL	1

#define KFD_DRIVER_PATCHLEVEL	1





const struct kfd2kgd_calls *kfd2kgd;

static const struct kgd2kfd_calls kgd2kfd = {

static const struct kgd2kfd_calls kgd2kfd = {

	.exit		= kgd2kfd_exit,

	.exit		= kgd2kfd_exit,

	.probe		= kgd2kfd_probe,

	.probe		= kgd2kfd_probe,
@@ -55,9 +54,7 @@ module_param(max_num_of_queues_per_device, int, 0444); 
MODULE_PARM_DESC(max_num_of_queues_per_device,

MODULE_PARM_DESC(max_num_of_queues_per_device,

	"Maximum number of supported queues per device (1 = Minimum, 4096 = default)");

	"Maximum number of supported queues per device (1 = Minimum, 4096 = default)");





bool kgd2kfd_init(unsigned interface_version,

bool kgd2kfd_init(unsigned interface_version, const struct kgd2kfd_calls **g2f)

		  const struct kfd2kgd_calls *f2g,

		  const struct kgd2kfd_calls **g2f)

{

{

	/*

	/*

	 * Only one interface version is supported,

	 * Only one interface version is supported,
@@ -66,11 +63,6 @@ bool kgd2kfd_init(unsigned interface_version, 
	if (interface_version != KFD_INTERFACE_VERSION)

	if (interface_version != KFD_INTERFACE_VERSION)

		return false;

		return false;





	/* Protection against multiple amd kgd loads */

	if (kfd2kgd)

		return true;



	kfd2kgd = f2g;

	*g2f = &kgd2kfd;

	*g2f = &kgd2kfd;





	return true;

	return true;
@@ -85,8 +77,6 @@ static int __init kfd_module_init(void) 
{

{

	int err;

	int err;





	kfd2kgd = NULL;



	/* Verify module parameters */

	/* Verify module parameters */

	if ((sched_policy < KFD_SCHED_POLICY_HWS) ||

	if ((sched_policy < KFD_SCHED_POLICY_HWS) ||

		(sched_policy > KFD_SCHED_POLICY_NO_HWS)) {

		(sched_policy > KFD_SCHED_POLICY_NO_HWS)) {