@@ -7,6 +7,7 @@ ccflags-y := -Iinclude/drm
amdkfd-y := kfd_module.o kfd_device.o kfd_chardev.o kfd_topology.o \
kfd_pasid.o kfd_doorbell.o kfd_vidmem.o kfd_aperture.o \
kfd_process.o kfd_queue.o kfd_mqd_manager.o \
- kfd_kernel_queue.o kfd_packet_manager.o
+ kfd_kernel_queue.o kfd_packet_manager.o \
+ kfd_process_queue_manager.o
obj-$(CONFIG_HSA_RADEON) += amdkfd.o
@@ -277,6 +277,9 @@ struct kfd_process_device {
/* The user-mode address of the doorbell mapping for this device. */
doorbell_t __user *doorbell_mapping;
+ /* per-process-per device QCM data structure */
+ struct qcm_process_device qpd;
+
/* Is this process/pasid bound to this device? (amd_iommu_bind_pasid) */
bool bound;
@@ -312,6 +315,8 @@ struct kfd_process {
*/
struct list_head per_device_data;
+ struct process_queue_manager pqm;
+
/* The process's queues. */
size_t queue_array_size;
@@ -382,11 +387,23 @@ inline uint32_t upper_32(uint64_t x);
int init_queue(struct queue **q, struct queue_properties properties);
void uninit_queue(struct queue *q);
+void print_queue_properties(struct queue_properties *q);
void print_queue(struct queue *q);
struct kernel_queue *kernel_queue_init(struct kfd_dev *dev, enum kfd_queue_type type);
void kernel_queue_uninit(struct kernel_queue *kq);
+/* Process Queue Manager */
+struct process_queue_node {
+ struct queue *q;
+ struct kernel_queue *kq;
+ struct list_head process_queue_list;
+};
+
+int pqm_init(struct process_queue_manager *pqm, struct kfd_process *p);
+void pqm_uninit(struct process_queue_manager *pqm);
+int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid);
+
/* Packet Manager */
#define KFD_HIQ_TIMEOUT (500)
@@ -163,8 +163,16 @@ static struct kfd_process *create_process(const struct task_struct *thread)
INIT_LIST_HEAD(&process->per_device_data);
+ err = pqm_init(&process->pqm, process);
+ if (err != 0)
+ goto err_process_pqm_init;
+
return process;
+err_process_pqm_init:
+ kfd_pasid_free(process->pasid);
+ list_del(&process->processes_list);
+ thread->mm->kfd_process = NULL;
err_alloc_pasid:
kfree(process->queues);
err_alloc_queues:
@@ -185,6 +193,9 @@ struct kfd_process_device *kfd_get_process_device_data(struct kfd_dev *dev,
pdd = kzalloc(sizeof(*pdd), GFP_KERNEL);
if (pdd != NULL) {
pdd->dev = dev;
+ INIT_LIST_HEAD(&pdd->qpd.queues_list);
+ INIT_LIST_HEAD(&pdd->qpd.priv_queue_list);
+ pdd->qpd.dqm = dev->dqm;
list_add(&pdd->per_device_list, &p->per_device_data);
}
@@ -246,6 +257,8 @@ void kfd_unbind_process_from_device(struct kfd_dev *dev, pasid_t pasid)
mutex_lock(&p->mutex);
+ pqm_uninit(&p->pqm);
+
/*
* Just mark pdd as unbound, because we still need it to call
* amd_iommu_unbind_pasid() in when the process exits.
new file mode 100644
@@ -0,0 +1,343 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * 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.
+ *
+ */
+
+#include <linux/slab.h>
+#include <linux/list.h>
+#include "kfd_device_queue_manager.h"
+#include "kfd_priv.h"
+#include "kfd_kernel_queue.h"
+
+static inline struct process_queue_node *get_queue_by_qid(struct process_queue_manager *pqm, unsigned int qid)
+{
+ struct process_queue_node *pqn;
+
+ BUG_ON(!pqm);
+
+ list_for_each_entry(pqn, &pqm->queues, process_queue_list) {
+ if (pqn->q && pqn->q->properties.queue_id == qid)
+ return pqn;
+ if (pqn->kq && pqn->kq->queue->properties.queue_id == qid)
+ return pqn;
+ }
+
+ return NULL;
+}
+
+static int find_available_queue_slot(struct process_queue_manager *pqm, unsigned int *qid)
+{
+ unsigned long found;
+
+ BUG_ON(!pqm || !qid);
+
+ pr_debug("kfd: in %s\n", __func__);
+
+ found = find_first_zero_bit(pqm->queue_slot_bitmap, MAX_PROCESS_QUEUES);
+
+ pr_debug("kfd: the new slot id %lu\n", found);
+
+ if (found >= MAX_PROCESS_QUEUES)
+ return -ENOMEM;
+
+ set_bit(found, pqm->queue_slot_bitmap);
+ *qid = found;
+
+ return 0;
+}
+
+int pqm_init(struct process_queue_manager *pqm, struct kfd_process *p)
+{
+ BUG_ON(!pqm);
+
+ INIT_LIST_HEAD(&pqm->queues);
+ pqm->queue_slot_bitmap = kzalloc(DIV_ROUND_UP(MAX_PROCESS_QUEUES, BITS_PER_BYTE), GFP_KERNEL);
+ if (pqm->queue_slot_bitmap == NULL)
+ return -ENOMEM;
+ pqm->process = p;
+
+ return 0;
+}
+
+void pqm_uninit(struct process_queue_manager *pqm)
+{
+ int retval;
+ struct process_queue_node *pqn, *next;
+
+ BUG_ON(!pqm);
+
+ pr_debug("In func %s\n", __func__);
+
+ list_for_each_entry_safe(pqn, next, &pqm->queues, process_queue_list) {
+ retval = pqm_destroy_queue(
+ pqm,
+ (pqn->q != NULL) ?
+ pqn->q->properties.queue_id :
+ pqn->kq->queue->properties.queue_id);
+
+ if (retval != 0) {
+ pr_err("kfd: failed to destroy queue\n");
+ return;
+ }
+ }
+ kfree(pqm->queue_slot_bitmap);
+}
+
+static int create_cp_queue(struct process_queue_manager *pqm, struct kfd_dev *dev, struct queue **q,
+ struct queue_properties *q_properties, struct file *f, unsigned int qid)
+{
+ int retval;
+
+ retval = 0;
+
+ q_properties->doorbell_ptr = kfd_get_doorbell(f, pqm->process, dev, qid);
+ if (!q_properties->doorbell_ptr)
+ return -ENOMEM;
+
+ q_properties->doorbell_off = kfd_queue_id_to_doorbell(dev, pqm->process, qid);
+
+ /* let DQM handle it*/
+ q_properties->vmid = 0;
+ q_properties->queue_id = qid;
+ q_properties->type = KFD_QUEUE_TYPE_COMPUTE;
+
+ retval = init_queue(q, *q_properties);
+ if (retval != 0)
+ goto err_init_queue;
+
+ (*q)->device = dev;
+ (*q)->process = pqm->process;
+
+ pr_debug("kfd: PQM After init queue");
+
+ return retval;
+
+err_init_queue:
+ return retval;
+}
+
+int pqm_create_queue(struct process_queue_manager *pqm,
+ struct kfd_dev *dev,
+ struct file *f,
+ struct queue_properties *properties,
+ unsigned int flags,
+ enum kfd_queue_type type,
+ unsigned int *qid)
+{
+ int retval;
+ struct kfd_process_device *pdd;
+ struct queue_properties q_properties;
+ struct queue *q;
+ struct process_queue_node *pqn;
+ struct kernel_queue *kq;
+
+ BUG_ON(!pqm || !dev || !properties || !qid);
+
+ memset(&q_properties, 0, sizeof(struct queue_properties));
+ memcpy(&q_properties, properties, sizeof(struct queue_properties));
+
+ pdd = kfd_get_process_device_data(dev, pqm->process);
+ BUG_ON(!pdd);
+
+ retval = find_available_queue_slot(pqm, qid);
+ if (retval != 0)
+ return retval;
+
+ if (list_empty(&pqm->queues)) {
+ pdd->qpd.pqm = pqm;
+ dev->dqm->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_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))) {
+ 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->create_queue(dev->dqm, q, &pdd->qpd, &q->properties.vmid);
+ print_queue(q);
+ break;
+ case KFD_QUEUE_TYPE_DIQ:
+ kq = kernel_queue_init(dev, KFD_QUEUE_TYPE_DIQ);
+ if (kq == NULL) {
+ kernel_queue_uninit(kq);
+ goto err_create_queue;
+ }
+ kq->queue->properties.queue_id = *qid;
+ pqn->kq = kq;
+ pqn->q = NULL;
+ retval = dev->dqm->create_kernel_queue(dev->dqm, kq, &pdd->qpd);
+ break;
+ default:
+ BUG();
+ break;
+ }
+
+ if (retval != 0) {
+ pr_err("kfd: error dqm create queue\n");
+ goto err_create_queue;
+ }
+
+ pr_debug("kfd: PQM After DQM create queue\n");
+
+ list_add(&pqn->process_queue_list, &pqm->queues);
+
+ retval = dev->dqm->execute_queues(dev->dqm);
+ if (retval != 0) {
+ if (pqn->kq)
+ dev->dqm->destroy_kernel_queue(dev->dqm, pqn->kq, &pdd->qpd);
+ if (pqn->q)
+ dev->dqm->destroy_queue(dev->dqm, &pdd->qpd, pqn->q);
+
+ goto err_execute_runlist;
+ }
+
+ *properties = q->properties;
+ pr_debug("kfd: PQM done creating queue\n");
+ print_queue_properties(properties);
+
+ return retval;
+
+err_execute_runlist:
+ list_del(&pqn->process_queue_list);
+err_create_queue:
+ kfree(pqn);
+err_allocate_pqn:
+ clear_bit(*qid, pqm->queue_slot_bitmap);
+ return retval;
+}
+
+int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid)
+{
+ struct process_queue_node *pqn;
+ struct kfd_process_device *pdd;
+ struct device_queue_manager *dqm;
+ struct kfd_dev *dev;
+ int retval;
+
+ dqm = NULL;
+
+ BUG_ON(!pqm);
+ retval = 0;
+
+ pr_debug("kfd: In Func %s\n", __func__);
+
+ pqn = get_queue_by_qid(pqm, qid);
+ BUG_ON(!pqn);
+
+ dev = NULL;
+ if (pqn->kq)
+ dev = pqn->kq->dev;
+ if (pqn->q)
+ dev = pqn->q->device;
+ BUG_ON(!dev);
+
+ pdd = kfd_get_process_device_data(dev, pqm->process);
+ BUG_ON(!pdd);
+
+ if (pqn->kq) {
+ /* destroy kernel queue (DIQ) */
+ dqm = pqn->kq->dev->dqm;
+ dqm->destroy_kernel_queue(dqm, pqn->kq, &pdd->qpd);
+ kernel_queue_uninit(pqn->kq);
+ }
+
+ if (pqn->q) {
+ dqm = pqn->q->device->dqm;
+ retval = dqm->destroy_queue(dqm, &pdd->qpd, pqn->q);
+ if (retval != 0)
+ return retval;
+
+ uninit_queue(pqn->q);
+ }
+
+ list_del(&pqn->process_queue_list);
+ kfree(pqn);
+ clear_bit(qid, pqm->queue_slot_bitmap);
+
+ if (list_empty(&pqm->queues))
+ dqm->unregister_process(dqm, &pdd->qpd);
+
+ retval = dqm->execute_queues(dqm);
+
+ return retval;
+}
+
+int pqm_update_queue(struct process_queue_manager *pqm, unsigned int qid, struct queue_properties *p)
+{
+ int retval;
+ struct process_queue_node *pqn;
+
+ BUG_ON(!pqm);
+
+ pqn = get_queue_by_qid(pqm, qid);
+ BUG_ON(!pqn);
+
+ pqn->q->properties.queue_address = p->queue_address;
+ pqn->q->properties.queue_size = p->queue_size;
+ pqn->q->properties.queue_percent = p->queue_percent;
+ pqn->q->properties.priority = p->priority;
+
+ retval = pqn->q->device->dqm->destroy_queues(pqn->q->device->dqm);
+ if (retval != 0)
+ return retval;
+
+ retval = pqn->q->device->dqm->update_queue(pqn->q->device->dqm, pqn->q);
+ if (retval != 0)
+ return retval;
+
+ retval = pqn->q->device->dqm->execute_queues(pqn->q->device->dqm);
+ if (retval != 0)
+ return retval;
+
+ return 0;
+}
+
+struct kernel_queue *pqm_get_kernel_queue(struct process_queue_manager *pqm, unsigned int qid)
+{
+ struct process_queue_node *pqn;
+
+ BUG_ON(!pqm);
+
+ pqn = get_queue_by_qid(pqm, qid);
+ if (pqn && pqn->kq)
+ return pqn->kq;
+
+ return NULL;
+}
+
+