gpu: nvgpu: Replace kref for refcounting in nvgpu

- added wrapper struct nvgpu_ref over nvgpu_atomic_t
- added nvgpu_ref_* APIs to access the above struct

JIRA NVGPU-140

Change-Id: Id47f897995dd4721751f7610b6d4d4fbfe4d6b9a
Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1540899
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svccoveritychecker <svccoveritychecker@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Konsta Holtta <kholtta@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>

11 files changed, 45 insertions, 43 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/fence_gk20a.c b/drivers/gpu/nvgpu/gk20a/fence_gk20a.c
index a7250b17..fdfef3da 100644
--- a/drivers/gpu/nvgpu/gk20a/fence_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/fence_gk20a.c
@@ -33,10 +33,10 @@
 struct gk20a_fence_ops {
         int (*wait)(struct gk20a_fence *, long timeout);
         bool (*is_expired)(struct gk20a_fence *);
-        void *(*free)(struct kref *);
+        void *(*free)(struct nvgpu_ref *);
 };
 
-static void gk20a_fence_free(struct kref *ref)
+static void gk20a_fence_free(struct nvgpu_ref *ref)
 {
         struct gk20a_fence *f =
                 container_of(ref, struct gk20a_fence, ref);
@@ -59,13 +59,13 @@ static void gk20a_fence_free(struct kref *ref)
 void gk20a_fence_put(struct gk20a_fence *f)
 {
         if (f)
-                kref_put(&f->ref, gk20a_fence_free);
+                nvgpu_ref_put(&f->ref, gk20a_fence_free);
 }
 
 struct gk20a_fence *gk20a_fence_get(struct gk20a_fence *f)
 {
         if (f)
-                kref_get(&f->ref);
+                nvgpu_ref_get(&f->ref);
         return f;
 }
 
@@ -175,7 +175,7 @@ struct gk20a_fence *gk20a_alloc_fence(struct channel_gk20a *c)
                 fence = nvgpu_kzalloc(c->g, sizeof(struct gk20a_fence));
 
         if (fence) {
-                kref_init(&fence->ref);
+                nvgpu_ref_init(&fence->ref);
                 fence->g = c->g;
         }
 
diff --git a/drivers/gpu/nvgpu/gk20a/fence_gk20a.h b/drivers/gpu/nvgpu/gk20a/fence_gk20a.h
index 140f5488..e0eb09b6 100644
--- a/drivers/gpu/nvgpu/gk20a/fence_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/fence_gk20a.h
@@ -33,7 +33,7 @@ struct gk20a_fence {
 
         /* Valid for all fence types: */
         bool valid;
-        struct kref ref;
+        struct nvgpu_ref ref;
         bool wfi;
         struct sync_fence *sync_fence;
         const struct gk20a_fence_ops *ops;
diff --git a/drivers/gpu/nvgpu/gk20a/fifo_gk20a.h b/drivers/gpu/nvgpu/gk20a/fifo_gk20a.h
index a6eae8ca..fb4932c8 100644
--- a/drivers/gpu/nvgpu/gk20a/fifo_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/fifo_gk20a.h
@@ -159,7 +159,7 @@ struct fifo_gk20a {
                 nvgpu_atomic_t get;
                 bool enabled;
                 u64 *sorted;
-                struct kref ref;
+                struct nvgpu_ref ref;
                 struct nvgpu_mutex lock;
         } profile;
 #endif
diff --git a/drivers/gpu/nvgpu/gk20a/gk20a.c b/drivers/gpu/nvgpu/gk20a/gk20a.c
index 550b22c0..639ec4b5 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/gk20a.c
@@ -511,7 +511,7 @@ int gk20a_init_gpu_characteristics(struct gk20a *g)
 /*
  * Free the gk20a struct.
  */
-static void gk20a_free_cb(struct kref *refcount)
+static void gk20a_free_cb(struct nvgpu_ref *refcount)
 {
         struct gk20a *g = container_of(refcount,
                 struct gk20a, refcount);
@@ -544,10 +544,11 @@ struct gk20a * __must_check gk20a_get(struct gk20a *g)
          * the code will never be in such a situation that this race is
          * possible.
          */
-        success = kref_get_unless_zero(&g->refcount);
+        success = nvgpu_ref_get_unless_zero(&g->refcount);
 
         gk20a_dbg(gpu_dbg_shutdown, "GET: refs currently %d %s",
-                atomic_read(&g->refcount.refcount), success ? "" : "(FAILED)");
+                nvgpu_atomic_read(&g->refcount.refcount),
+                        success ? "" : "(FAILED)");
 
         return success ? g : NULL;
 }
@@ -571,7 +572,7 @@ void gk20a_put(struct gk20a *g)
          *  ... Freeing GK20A struct!
          */
         gk20a_dbg(gpu_dbg_shutdown, "PUT: refs currently %d",
-                atomic_read(&g->refcount.refcount));
+                nvgpu_atomic_read(&g->refcount.refcount));
 
-        kref_put(&g->refcount, gk20a_free_cb);
+        nvgpu_ref_put(&g->refcount, gk20a_free_cb);
 }
diff --git a/drivers/gpu/nvgpu/gk20a/gk20a.h b/drivers/gpu/nvgpu/gk20a/gk20a.h
index 5efa846d..15e81291 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/gk20a.h
@@ -1036,7 +1036,7 @@ struct gk20a {
 
         nvgpu_atomic_t usage_count;
 
-        struct kref refcount;
+        struct nvgpu_ref refcount;
 
         struct resource *reg_mem;
         void __iomem *regs;
diff --git a/drivers/gpu/nvgpu/gk20a/mm_gk20a.c b/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
index 2ce78cef..3030c170 100644
--- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
@@ -811,7 +811,7 @@ int nvgpu_vm_get_buffers(struct vm_gk20a *vm,
                 mapped_buffer = mapped_buffer_from_rbtree_node(node);
                 if (mapped_buffer->user_mapped) {
                         buffer_list[i] = mapped_buffer;
-                        kref_get(&mapped_buffer->ref);
+                        nvgpu_ref_get(&mapped_buffer->ref);
                         i++;
                 }
                 nvgpu_rbtree_enum_next(&node, node);
@@ -827,7 +827,7 @@ int nvgpu_vm_get_buffers(struct vm_gk20a *vm,
         return 0;
 }
 
-void gk20a_vm_unmap_locked_kref(struct kref *ref)
+void gk20a_vm_unmap_locked_ref(struct nvgpu_ref *ref)
 {
         struct nvgpu_mapped_buf *mapped_buffer =
                 container_of(ref, struct nvgpu_mapped_buf, ref);
@@ -849,8 +849,8 @@ void nvgpu_vm_put_buffers(struct vm_gk20a *vm,
         vm->kref_put_batch = &batch;
 
         for (i = 0; i < num_buffers; ++i)
-                kref_put(&mapped_buffers[i]->ref,
-                         gk20a_vm_unmap_locked_kref);
+                nvgpu_ref_put(&mapped_buffers[i]->ref,
+                         gk20a_vm_unmap_locked_ref);
 
         vm->kref_put_batch = NULL;
         nvgpu_vm_mapping_batch_finish_locked(vm, &batch);
@@ -882,8 +882,9 @@ static void nvgpu_vm_unmap_user(struct vm_gk20a *vm, u64 offset,
                 nvgpu_timeout_init(vm->mm->g, &timeout, 10000,
                                    NVGPU_TIMER_RETRY_TIMER);
                 do {
-                        if (atomic_read(&mapped_buffer->ref.refcount) == 1)
-                                break;
+                        if (nvgpu_atomic_read(
+                                &mapped_buffer->ref.refcount) == 1)
+                                        break;
                         nvgpu_udelay(5);
                 } while (!nvgpu_timeout_expired_msg(&timeout,
                                             "sync-unmap failed on 0x%llx"));
@@ -902,7 +903,7 @@ static void nvgpu_vm_unmap_user(struct vm_gk20a *vm, u64 offset,
                 vm->num_user_mapped_buffers--;
 
         vm->kref_put_batch = batch;
-        kref_put(&mapped_buffer->ref, gk20a_vm_unmap_locked_kref);
+        nvgpu_ref_put(&mapped_buffer->ref, gk20a_vm_unmap_locked_ref);
         vm->kref_put_batch = NULL;
 
         nvgpu_mutex_release(&vm->update_gmmu_lock);
diff --git a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
index e8b90c8f..82a4ee85 100644
--- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
@@ -434,6 +434,6 @@ extern const struct gk20a_mmu_level gk20a_mm_levels_128k[];
 
 int gk20a_mm_get_buffer_info(struct device *dev, int dmabuf_fd,
                              u64 *buffer_id, u64 *buffer_len);
-void gk20a_vm_unmap_locked_kref(struct kref *ref);
+void gk20a_vm_unmap_locked_ref(struct nvgpu_ref *ref);
 
 #endif /* MM_GK20A_H */
diff --git a/drivers/gpu/nvgpu/gk20a/sched_gk20a.c b/drivers/gpu/nvgpu/gk20a/sched_gk20a.c
index 014848ba..ac54addd 100644
--- a/drivers/gpu/nvgpu/gk20a/sched_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/sched_gk20a.c
@@ -189,7 +189,7 @@ static int gk20a_sched_dev_ioctl_get_params(struct gk20a_sched_ctrl *sched,
                 return -EINVAL;
 
         tsg = &f->tsg[tsgid];
-        if (!kref_get_unless_zero(&tsg->refcount))
+        if (!nvgpu_ref_get_unless_zero(&tsg->refcount))
                 return -ENXIO;
 
         arg->pid = tsg->tgid;   /* kernel tgid corresponds to user pid */
@@ -206,7 +206,7 @@ static int gk20a_sched_dev_ioctl_get_params(struct gk20a_sched_ctrl *sched,
                 arg->compute_preempt_mode = 0;
         }
 
-        kref_put(&tsg->refcount, gk20a_tsg_release);
+        nvgpu_ref_put(&tsg->refcount, gk20a_tsg_release);
 
         return 0;
 }
@@ -227,7 +227,7 @@ static int gk20a_sched_dev_ioctl_tsg_set_timeslice(
                 return -EINVAL;
 
         tsg = &f->tsg[tsgid];
-        if (!kref_get_unless_zero(&tsg->refcount))
+        if (!nvgpu_ref_get_unless_zero(&tsg->refcount))
                 return -ENXIO;
 
         err = gk20a_busy(g);
@@ -239,7 +239,7 @@ static int gk20a_sched_dev_ioctl_tsg_set_timeslice(
         gk20a_idle(g);
 
 done:
-        kref_put(&tsg->refcount, gk20a_tsg_release);
+        nvgpu_ref_put(&tsg->refcount, gk20a_tsg_release);
 
         return err;
 }
@@ -260,7 +260,7 @@ static int gk20a_sched_dev_ioctl_tsg_set_runlist_interleave(
                 return -EINVAL;
 
         tsg = &f->tsg[tsgid];
-        if (!kref_get_unless_zero(&tsg->refcount))
+        if (!nvgpu_ref_get_unless_zero(&tsg->refcount))
                 return -ENXIO;
 
         err = gk20a_busy(g);
@@ -272,7 +272,7 @@ static int gk20a_sched_dev_ioctl_tsg_set_runlist_interleave(
         gk20a_idle(g);
 
 done:
-        kref_put(&tsg->refcount, gk20a_tsg_release);
+        nvgpu_ref_put(&tsg->refcount, gk20a_tsg_release);
 
         return err;
 }
@@ -320,7 +320,7 @@ static int gk20a_sched_dev_ioctl_get_tsg(struct gk20a_sched_ctrl *sched,
                 return -EINVAL;
 
         tsg = &f->tsg[tsgid];
-        if (!kref_get_unless_zero(&tsg->refcount))
+        if (!nvgpu_ref_get_unless_zero(&tsg->refcount))
                 return -ENXIO;
 
         nvgpu_mutex_acquire(&sched->status_lock);
@@ -328,7 +328,7 @@ static int gk20a_sched_dev_ioctl_get_tsg(struct gk20a_sched_ctrl *sched,
                 nvgpu_warn(g, "tsgid=%d already referenced", tsgid);
                 /* unlock status_lock as gk20a_tsg_release locks it */
                 nvgpu_mutex_release(&sched->status_lock);
-                kref_put(&tsg->refcount, gk20a_tsg_release);
+                nvgpu_ref_put(&tsg->refcount, gk20a_tsg_release);
                 return -ENXIO;
         }
 
@@ -364,7 +364,7 @@ static int gk20a_sched_dev_ioctl_put_tsg(struct gk20a_sched_ctrl *sched,
         nvgpu_mutex_release(&sched->status_lock);
 
         tsg = &f->tsg[tsgid];
-        kref_put(&tsg->refcount, gk20a_tsg_release);
+        nvgpu_ref_put(&tsg->refcount, gk20a_tsg_release);
 
         return 0;
 }
@@ -507,7 +507,7 @@ int gk20a_sched_dev_release(struct inode *inode, struct file *filp)
         for (tsgid = 0; tsgid < f->num_channels; tsgid++) {
                 if (NVGPU_SCHED_ISSET(tsgid, sched->ref_tsg_bitmap)) {
                         tsg = &f->tsg[tsgid];
-                        kref_put(&tsg->refcount, gk20a_tsg_release);
+                        nvgpu_ref_put(&tsg->refcount, gk20a_tsg_release);
                 }
         }
 
diff --git a/drivers/gpu/nvgpu/gk20a/sync_gk20a.c b/drivers/gpu/nvgpu/gk20a/sync_gk20a.c
index deaf19a1..a8e824b6 100644
--- a/drivers/gpu/nvgpu/gk20a/sync_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/sync_gk20a.c
@@ -43,7 +43,7 @@ struct gk20a_sync_timeline {
  */
 struct gk20a_sync_pt {
         struct gk20a                    *g;
-        struct kref                     refcount;
+        struct nvgpu_ref                        refcount;
         u32                             thresh;
         struct nvgpu_semaphore          *sema;
         struct gk20a_sync_timeline      *obj;
@@ -170,7 +170,7 @@ static struct gk20a_sync_timeline *to_gk20a_timeline(struct sync_timeline *obj)
         return (struct gk20a_sync_timeline *)obj;
 }
 
-static void gk20a_sync_pt_free_shared(struct kref *ref)
+static void gk20a_sync_pt_free_shared(struct nvgpu_ref *ref)
 {
         struct gk20a_sync_pt *pt =
                 container_of(ref, struct gk20a_sync_pt, refcount);
@@ -192,7 +192,7 @@ static struct gk20a_sync_pt *gk20a_sync_pt_create_shared(
         if (!shared)
                 return NULL;
 
-        kref_init(&shared->refcount);
+        nvgpu_ref_init(&shared->refcount);
         shared->g = g;
         shared->obj = obj;
         shared->sema = sema;
@@ -229,7 +229,7 @@ static void gk20a_sync_pt_free_inst(struct sync_pt *sync_pt)
 {
         struct gk20a_sync_pt *pt = to_gk20a_sync_pt(sync_pt);
         if (pt)
-                kref_put(&pt->refcount, gk20a_sync_pt_free_shared);
+                nvgpu_ref_put(&pt->refcount, gk20a_sync_pt_free_shared);
 }
 
 static struct sync_pt *gk20a_sync_pt_dup_inst(struct sync_pt *sync_pt)
@@ -242,7 +242,7 @@ static struct sync_pt *gk20a_sync_pt_dup_inst(struct sync_pt *sync_pt)
         if (!pti)
                 return NULL;
         pti->shared = pt;
-        kref_get(&pt->refcount);
+        nvgpu_ref_get(&pt->refcount);
         return &pti->pt;
 }
 
diff --git a/drivers/gpu/nvgpu/gk20a/tsg_gk20a.c b/drivers/gpu/nvgpu/gk20a/tsg_gk20a.c
index 99d72292..f3e87a13 100644
--- a/drivers/gpu/nvgpu/gk20a/tsg_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/tsg_gk20a.c
@@ -104,7 +104,7 @@ int gk20a_tsg_bind_channel(struct tsg_gk20a *tsg,
         nvgpu_list_add_tail(&ch->ch_entry, &tsg->ch_list);
         up_write(&tsg->ch_list_lock);
 
-        kref_get(&tsg->refcount);
+        nvgpu_ref_get(&tsg->refcount);
 
         gk20a_dbg(gpu_dbg_fn, "BIND tsg:%d channel:%d\n",
                                         tsg->tsgid, ch->chid);
@@ -122,7 +122,7 @@ int gk20a_tsg_unbind_channel(struct channel_gk20a *ch)
         nvgpu_list_del(&ch->ch_entry);
         up_write(&tsg->ch_list_lock);
 
-        kref_put(&tsg->refcount, gk20a_tsg_release);
+        nvgpu_ref_put(&tsg->refcount, gk20a_tsg_release);
 
         ch->tsgid = NVGPU_INVALID_TSG_ID;
 
@@ -257,7 +257,7 @@ struct tsg_gk20a *gk20a_tsg_open(struct gk20a *g)
 
         tsg->g = g;
         tsg->num_active_channels = 0;
-        kref_init(&tsg->refcount);
+        nvgpu_ref_init(&tsg->refcount);
 
         tsg->tsg_gr_ctx = NULL;
         tsg->vm = NULL;
@@ -287,11 +287,11 @@ struct tsg_gk20a *gk20a_tsg_open(struct gk20a *g)
         return tsg;
 
 clean_up:
-        kref_put(&tsg->refcount, gk20a_tsg_release);
+        nvgpu_ref_put(&tsg->refcount, gk20a_tsg_release);
         return NULL;
 }
 
-void gk20a_tsg_release(struct kref *ref)
+void gk20a_tsg_release(struct nvgpu_ref *ref)
 {
         struct tsg_gk20a *tsg = container_of(ref, struct tsg_gk20a, refcount);
         struct gk20a *g = tsg->g;
diff --git a/drivers/gpu/nvgpu/gk20a/tsg_gk20a.h b/drivers/gpu/nvgpu/gk20a/tsg_gk20a.h
index 40e12105..9195d3d3 100644
--- a/drivers/gpu/nvgpu/gk20a/tsg_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/tsg_gk20a.h
@@ -25,7 +25,7 @@ struct channel_gk20a;
 
 bool gk20a_is_channel_marked_as_tsg(struct channel_gk20a *ch);
 struct tsg_gk20a *gk20a_tsg_open(struct gk20a *g);
-void gk20a_tsg_release(struct kref *ref);
+void gk20a_tsg_release(struct nvgpu_ref *ref);
 
 int gk20a_init_tsg_support(struct gk20a *g, u32 tsgid);
 struct tsg_gk20a *tsg_gk20a_from_ch(struct channel_gk20a *ch);
@@ -36,7 +36,7 @@ struct tsg_gk20a {
         bool in_use;
         int tsgid;
 
-        struct kref refcount;
+        struct nvgpu_ref refcount;
 
         struct nvgpu_list_node ch_list;
         int num_active_channels;