11 files changed, 78 insertions, 134 deletions

diff --git a/drivers/gpu/nvgpu/common/linux/ioctl_tsg.c b/drivers/gpu/nvgpu/common/linux/ioctl_tsg.c
index 25805ea0..d530fff6 100644
--- a/drivers/gpu/nvgpu/common/linux/ioctl_tsg.c
+++ b/drivers/gpu/nvgpu/common/linux/ioctl_tsg.c
@@ -18,7 +18,6 @@
 #include <linux/file.h>
 #include <linux/cdev.h>
 #include <linux/uaccess.h>
-#include <linux/nvhost.h>
 #include <uapi/linux/nvgpu.h>
 #include <linux/anon_inodes.h>
 
diff --git a/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c b/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c
index 07157dc9..3d313ce8 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c
@@ -23,6 +23,7 @@
 #include <nvgpu/atomic.h>
 #include <nvgpu/bug.h>
 #include <nvgpu/list.h>
+#include <nvgpu/nvhost.h>
 
 #include "channel_sync_gk20a.h"
 #include "gk20a.h"
@@ -35,12 +36,11 @@
 #endif
 
 #ifdef CONFIG_TEGRA_GK20A_NVHOST
-#include <linux/nvhost.h>
 
 struct gk20a_channel_syncpt {
         struct gk20a_channel_sync ops;
         struct channel_gk20a *c;
-        struct platform_device *host1x_pdev;
+        struct nvgpu_nvhost_dev *nvhost_dev;
         u32 id;
         struct nvgpu_mem syncpt_buf;
 };
@@ -54,12 +54,12 @@ static int gk20a_channel_syncpt_wait_syncpt(struct gk20a_channel_sync *s,
         struct channel_gk20a *c = sp->c;
         int err = 0;
 
-        if (!nvhost_syncpt_is_valid_pt_ext(sp->host1x_pdev, id)) {
+        if (!nvgpu_nvhost_syncpt_is_valid_pt_ext(sp->nvhost_dev, id)) {
                 nvgpu_warn(c->g, "invalid wait id in gpfifo submit, elided");
                 return 0;
         }
 
-        if (nvhost_syncpt_is_expired_ext(sp->host1x_pdev, id, thresh))
+        if (nvgpu_nvhost_syncpt_is_expired_ext(sp->nvhost_dev, id, thresh))
                 return 0;
 
         err = gk20a_channel_alloc_priv_cmdbuf(c,
@@ -94,7 +94,7 @@ static int gk20a_channel_syncpt_wait_fd(struct gk20a_channel_sync *s, int fd,
         int err = 0;
         u32 wait_cmd_size = 0;
 
-        sync_fence = nvhost_sync_fdget(fd);
+        sync_fence = nvgpu_nvhost_sync_fdget(fd);
         if (!sync_fence)
                 return -EINVAL;
 
@@ -105,9 +105,9 @@ static int gk20a_channel_syncpt_wait_fd(struct gk20a_channel_sync *s, int fd,
         for (i = 0; i < sync_fence->num_fences; i++) {
                 pt = sync_pt_from_fence(sync_fence->cbs[i].sync_pt);
 #endif
-                wait_id = nvhost_sync_pt_id(pt);
-                if (!wait_id || !nvhost_syncpt_is_valid_pt_ext(sp->host1x_pdev,
-                                        wait_id)) {
+                wait_id = nvgpu_nvhost_sync_pt_id(pt);
+                if (!wait_id || !nvgpu_nvhost_syncpt_is_valid_pt_ext(
+                                        sp->nvhost_dev, wait_id)) {
                         sync_fence_put(sync_fence);
                         return -EINVAL;
                 }
@@ -117,7 +117,7 @@ static int gk20a_channel_syncpt_wait_fd(struct gk20a_channel_sync *s, int fd,
         }
 #endif
 
-        num_wait_cmds = nvhost_sync_num_pts(sync_fence);
+        num_wait_cmds = nvgpu_nvhost_sync_num_pts(sync_fence);
         if (num_wait_cmds == 0) {
                 sync_fence_put(sync_fence);
                 return 0;
@@ -141,10 +141,10 @@ static int gk20a_channel_syncpt_wait_fd(struct gk20a_channel_sync *s, int fd,
                 struct fence *f = sync_fence->cbs[i].sync_pt;
                 struct sync_pt *pt = sync_pt_from_fence(f);
 #endif
-                u32 wait_id = nvhost_sync_pt_id(pt);
-                u32 wait_value = nvhost_sync_pt_thresh(pt);
+                u32 wait_id = nvgpu_nvhost_sync_pt_id(pt);
+                u32 wait_value = nvgpu_nvhost_sync_pt_thresh(pt);
 
-                if (nvhost_syncpt_is_expired_ext(sp->host1x_pdev,
+                if (nvgpu_nvhost_syncpt_is_expired_ext(sp->nvhost_dev,
                                 wait_id, wait_value)) {
                         nvgpu_memset(c->g, wait_cmd->mem,
                         (wait_cmd->off + i * wait_cmd_size) * sizeof(u32),
@@ -206,7 +206,7 @@ static int __gk20a_channel_syncpt_incr(struct gk20a_channel_sync *s,
         c->g->ops.fifo.add_syncpt_incr_cmd(c->g, wfi_cmd,
                         incr_cmd, sp->id, sp->syncpt_buf.gpu_va);
 
-        thresh = nvhost_syncpt_incr_max_ext(sp->host1x_pdev, sp->id, 2);
+        thresh = nvgpu_nvhost_syncpt_incr_max_ext(sp->nvhost_dev, sp->id, 2);
 
         if (register_irq) {
                 struct channel_gk20a *referenced = gk20a_channel_get(c);
@@ -217,8 +217,8 @@ static int __gk20a_channel_syncpt_incr(struct gk20a_channel_sync *s,
                         /* note: channel_put() is in
                          * gk20a_channel_syncpt_update() */
 
-                        err = nvhost_intr_register_notifier(
-                                sp->host1x_pdev,
+                        err = nvgpu_nvhost_intr_register_notifier(
+                                sp->nvhost_dev,
                                 sp->id, thresh,
                                 gk20a_channel_syncpt_update, c);
                         if (err)
@@ -234,7 +234,7 @@ static int __gk20a_channel_syncpt_incr(struct gk20a_channel_sync *s,
                 }
         }
 
-        err = gk20a_fence_from_syncpt(fence, sp->host1x_pdev, sp->id, thresh,
+        err = gk20a_fence_from_syncpt(fence, sp->nvhost_dev, sp->id, thresh,
                                          wfi_cmd, need_sync_fence);
         if (err)
                 goto clean_up_priv_cmd;
@@ -290,7 +290,7 @@ static void gk20a_channel_syncpt_set_min_eq_max(struct gk20a_channel_sync *s)
 {
         struct gk20a_channel_syncpt *sp =
                 container_of(s, struct gk20a_channel_syncpt, ops);
-        nvhost_syncpt_set_min_eq_max_ext(sp->host1x_pdev, sp->id);
+        nvgpu_nvhost_syncpt_set_min_eq_max_ext(sp->nvhost_dev, sp->id);
 }
 
 static void gk20a_channel_syncpt_signal_timeline(
@@ -314,8 +314,8 @@ static void gk20a_channel_syncpt_destroy(struct gk20a_channel_sync *s)
 
         sp->c->g->ops.fifo.free_syncpt_buf(sp->c, &sp->syncpt_buf);
 
-        nvhost_syncpt_set_min_eq_max_ext(sp->host1x_pdev, sp->id);
-        nvhost_syncpt_put_ref_ext(sp->host1x_pdev, sp->id);
+        nvgpu_nvhost_syncpt_set_min_eq_max_ext(sp->nvhost_dev, sp->id);
+        nvgpu_nvhost_syncpt_put_ref_ext(sp->nvhost_dev, sp->id);
         nvgpu_kfree(sp->c->g, sp);
 }
 
@@ -330,12 +330,12 @@ gk20a_channel_syncpt_create(struct channel_gk20a *c)
                 return NULL;
 
         sp->c = c;
-        sp->host1x_pdev = c->g->host1x_dev;
+        sp->nvhost_dev = c->g->nvhost_dev;
 
         snprintf(syncpt_name, sizeof(syncpt_name),
                 "%s_%d", c->g->name, c->hw_chid);
 
-        sp->id = nvhost_get_syncpt_host_managed(sp->host1x_pdev,
+        sp->id = nvgpu_nvhost_get_syncpt_host_managed(sp->nvhost_dev,
                                                 c->hw_chid, syncpt_name);
         if (!sp->id) {
                 nvgpu_kfree(c->g, sp);
@@ -346,7 +346,7 @@ gk20a_channel_syncpt_create(struct channel_gk20a *c)
         sp->c->g->ops.fifo.alloc_syncpt_buf(sp->c, sp->id,
                                 &sp->syncpt_buf);
 
-        nvhost_syncpt_set_min_eq_max_ext(sp->host1x_pdev, sp->id);
+        nvgpu_nvhost_syncpt_set_min_eq_max_ext(sp->nvhost_dev, sp->id);
 
         atomic_set(&sp->ops.refcount, 0);
         sp->ops.wait_syncpt             = gk20a_channel_syncpt_wait_syncpt;
diff --git a/drivers/gpu/nvgpu/gk20a/dbg_gpu_gk20a.c b/drivers/gpu/nvgpu/gk20a/dbg_gpu_gk20a.c
index 4bfa041e..f018ef89 100644
--- a/drivers/gpu/nvgpu/gk20a/dbg_gpu_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/dbg_gpu_gk20a.c
@@ -20,7 +20,6 @@
 #include <linux/file.h>
 #include <linux/cdev.h>
 #include <linux/uaccess.h>
-#include <linux/nvhost.h>
 #include <linux/dma-buf.h>
 #include <uapi/linux/nvgpu.h>
 
diff --git a/drivers/gpu/nvgpu/gk20a/fence_gk20a.c b/drivers/gpu/nvgpu/gk20a/fence_gk20a.c
index 7c5d33c7..5392e77c 100644
--- a/drivers/gpu/nvgpu/gk20a/fence_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/fence_gk20a.c
@@ -21,6 +21,7 @@
 #include <nvgpu/semaphore.h>
 #include <nvgpu/kmem.h>
 #include <nvgpu/soc.h>
+#include <nvgpu/nvhost.h>
 
 #include "gk20a.h"
 #include "channel_gk20a.h"
@@ -30,11 +31,6 @@
 #include "../drivers/staging/android/sync.h"
 #endif
 
-#ifdef CONFIG_TEGRA_GK20A_NVHOST
-#include <linux/nvhost.h>
-#include <linux/nvhost_ioctl.h>
-#endif
-
 struct gk20a_fence_ops {
         int (*wait)(struct gk20a_fence *, long timeout);
         bool (*is_expired)(struct gk20a_fence *);
@@ -268,8 +264,8 @@ int gk20a_fence_from_semaphore(
 
 static int gk20a_syncpt_fence_wait(struct gk20a_fence *f, long timeout)
 {
-        return nvhost_syncpt_wait_timeout_ext(
-                        f->host1x_pdev, f->syncpt_id, f->syncpt_value,
+        return nvgpu_nvhost_syncpt_wait_timeout_ext(
+                        f->nvhost_dev, f->syncpt_id, f->syncpt_value,
                         (u32)timeout, NULL, NULL);
 }
 
@@ -281,13 +277,14 @@ static bool gk20a_syncpt_fence_is_expired(struct gk20a_fence *f)
          * syncpt value to be updated. For this case, we force a read
          * of the value from HW, and then check for expiration.
          */
-        if (!nvhost_syncpt_is_expired_ext(f->host1x_pdev, f->syncpt_id,
+        if (!nvgpu_nvhost_syncpt_is_expired_ext(f->nvhost_dev, f->syncpt_id,
                                 f->syncpt_value)) {
                 u32 val;
 
-                if (!nvhost_syncpt_read_ext_check(f->host1x_pdev,
+                if (!nvgpu_nvhost_syncpt_read_ext_check(f->nvhost_dev,
                                 f->syncpt_id, &val)) {
-                        return nvhost_syncpt_is_expired_ext(f->host1x_pdev,
+                        return nvgpu_nvhost_syncpt_is_expired_ext(
+                                        f->nvhost_dev,
                                         f->syncpt_id, f->syncpt_value);
                 }
         }
@@ -302,7 +299,7 @@ static const struct gk20a_fence_ops gk20a_syncpt_fence_ops = {
 
 int gk20a_fence_from_syncpt(
                 struct gk20a_fence *fence_out,
-                struct platform_device *host1x_pdev,
+                struct nvgpu_nvhost_dev *nvhost_dev,
                 u32 id, u32 value, bool wfi,
                 bool need_sync_fence)
 {
@@ -310,14 +307,9 @@ int gk20a_fence_from_syncpt(
         struct sync_fence *sync_fence = NULL;
 
 #ifdef CONFIG_SYNC
-        struct nvhost_ctrl_sync_fence_info pt = {
-                .id = id,
-                .thresh = value
-        };
-
         if (need_sync_fence) {
-                sync_fence = nvhost_sync_create_fence(host1x_pdev, &pt, 1,
-                                                      "fence");
+                sync_fence = nvgpu_nvhost_sync_create_fence(nvhost_dev,
+                                        id, value, 1, "fence");
                 if (IS_ERR(sync_fence))
                         return -1;
         }
@@ -331,7 +323,7 @@ int gk20a_fence_from_syncpt(
 #endif
                 return -EINVAL;
         }
-        f->host1x_pdev = host1x_pdev;
+        f->nvhost_dev = nvhost_dev;
         f->syncpt_id = id;
         f->syncpt_value = value;
 
@@ -344,7 +336,7 @@ int gk20a_fence_from_syncpt(
 #else
 int gk20a_fence_from_syncpt(
                 struct gk20a_fence *fence_out,
-                struct platform_device *host1x_pdev,
+                struct nvgpu_nvhost_dev *nvhost_dev,
                 u32 id, u32 value, bool wfi,
                 bool need_sync_fence)
 {
diff --git a/drivers/gpu/nvgpu/gk20a/fence_gk20a.h b/drivers/gpu/nvgpu/gk20a/fence_gk20a.h
index c479f359..140f5488 100644
--- a/drivers/gpu/nvgpu/gk20a/fence_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/fence_gk20a.h
@@ -43,7 +43,7 @@ struct gk20a_fence {
         struct nvgpu_cond *semaphore_wq;
 
         /* Valid for fences created from syncpoints: */
-        struct platform_device *host1x_pdev;
+        struct nvgpu_nvhost_dev *nvhost_dev;
         u32 syncpt_id;
         u32 syncpt_value;
 
@@ -62,7 +62,7 @@ int gk20a_fence_from_semaphore(
 
 int gk20a_fence_from_syncpt(
                 struct gk20a_fence *fence_out,
-                struct platform_device *host1x_pdev,
+                struct nvgpu_nvhost_dev *nvhost_dev,
                 u32 id, u32 value, bool wfi,
                 bool need_sync_fence);
 
diff --git a/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c b/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c
index 7eb28f2f..4b0667c5 100644
--- a/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c
@@ -18,9 +18,6 @@
  */
 
 #include <trace/events/gk20a.h>
-#ifdef CONFIG_TEGRA_GK20A_NVHOST
-#include <linux/nvhost.h>
-#endif
 
 #include <nvgpu/dma.h>
 #include <nvgpu/timers.h>
@@ -32,6 +29,7 @@
 #include <nvgpu/bug.h>
 #include <nvgpu/log2.h>
 #include <nvgpu/debug.h>
+#include <nvgpu/nvhost.h>
 
 #include "gk20a.h"
 #include "ctxsw_trace_gk20a.h"
@@ -3537,7 +3535,7 @@ void gk20a_dump_channel_status_ramfc(struct gk20a *g,
                 gk20a_debug_output(o, "%s on syncpt %u (%s) val %u\n",
                         (status == 3 || status == 8) ? "Waiting" : "Waited",
                         pbdma_syncpointb_syncpt_index_v(syncpointb),
-                        nvhost_syncpt_get_name(g->host1x_dev,
+                        nvgpu_nvhost_syncpt_get_name(g->nvhost_dev,
                                 pbdma_syncpointb_syncpt_index_v(syncpointb)),
                         pbdma_syncpointa_payload_v(syncpointa));
 #endif
diff --git a/drivers/gpu/nvgpu/gk20a/gk20a.h b/drivers/gpu/nvgpu/gk20a/gk20a.h
index 8691c73d..b4884af1 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/gk20a.h
@@ -981,7 +981,6 @@ struct nvgpu_bios {
 
 struct gk20a {
         struct device *dev;
-        struct platform_device *host1x_dev;
         struct nvgpu_nvhost_dev *nvhost_dev;
 
         /*
diff --git a/drivers/gpu/nvgpu/gk20a/gk20a_sysfs.c b/drivers/gpu/nvgpu/gk20a/gk20a_sysfs.c
index 1065968b..4a79a142 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a_sysfs.c
+++ b/drivers/gpu/nvgpu/gk20a/gk20a_sysfs.c
@@ -28,6 +28,7 @@
 #endif
 
 #include <nvgpu/kmem.h>
+#include <nvgpu/nvhost.h>
 
 #include "gk20a.h"
 #include "gk20a/platform_gk20a.h"
@@ -889,8 +890,6 @@ static DEVICE_ATTR(max_timeslice_us, ROOTRW, max_timeslice_us_read,
 
 void gk20a_remove_sysfs(struct device *dev)
 {
-        struct gk20a *g = get_gk20a(dev);
-
         device_remove_file(dev, &dev_attr_elcg_enable);
         device_remove_file(dev, &dev_attr_blcg_enable);
         device_remove_file(dev, &dev_attr_slcg_enable);
@@ -917,9 +916,9 @@ void gk20a_remove_sysfs(struct device *dev)
         device_remove_file(dev, &dev_attr_min_timeslice_us);
         device_remove_file(dev, &dev_attr_max_timeslice_us);
 
-        if (g->host1x_dev && (dev->parent != &g->host1x_dev->dev)) {
-                sysfs_remove_link(&g->host1x_dev->dev.kobj, dev_name(dev));
-        }
+#ifdef CONFIG_TEGRA_GK20A_NVHOST
+        nvgpu_nvhost_remove_symlink(get_gk20a(dev));
+#endif
 
         if (strcmp(dev_name(dev), "gpu.0")) {
                 struct kobject *kobj = &dev->kobj;
@@ -931,7 +930,6 @@ void gk20a_remove_sysfs(struct device *dev)
 
 void gk20a_create_sysfs(struct device *dev)
 {
-        struct gk20a *g = gk20a_from_dev(dev);
         int error = 0;
 
         error |= device_create_file(dev, &dev_attr_elcg_enable);
@@ -960,11 +958,9 @@ void gk20a_create_sysfs(struct device *dev)
         error |= device_create_file(dev, &dev_attr_min_timeslice_us);
         error |= device_create_file(dev, &dev_attr_max_timeslice_us);
 
-        if (g->host1x_dev && (dev->parent != &g->host1x_dev->dev)) {
-                error |= sysfs_create_link(&g->host1x_dev->dev.kobj,
-                                           &dev->kobj,
-                                           dev_name(dev));
-        }
+#ifdef CONFIG_TEGRA_GK20A_NVHOST
+        error |= nvgpu_nvhost_create_symlink(get_gk20a(dev));
+#endif
 
         if (strcmp(dev_name(dev), "gpu.0")) {
                 struct kobject *kobj = &dev->kobj;
diff --git a/drivers/gpu/nvgpu/gk20a/platform_vgpu_tegra.c b/drivers/gpu/nvgpu/gk20a/platform_vgpu_tegra.c
index 049d394e..e33fc29f 100644
--- a/drivers/gpu/nvgpu/gk20a/platform_vgpu_tegra.c
+++ b/drivers/gpu/nvgpu/gk20a/platform_vgpu_tegra.c
@@ -13,38 +13,20 @@
  * more details.
  */
 
-#include <linux/of_platform.h>
-
 #include "gk20a.h"
 #include "hal_gk20a.h"
 #include "platform_gk20a.h"
 
+#include <nvgpu/nvhost.h>
+
 static int gk20a_tegra_probe(struct device *dev)
 {
+#ifdef CONFIG_TEGRA_GK20A_NVHOST
         struct gk20a_platform *platform = dev_get_drvdata(dev);
-        struct device_node *np = dev->of_node;
-        const __be32 *host1x_ptr;
-        struct platform_device *host1x_pdev = NULL;
-
-        host1x_ptr = of_get_property(np, "nvidia,host1x", NULL);
-        if (host1x_ptr) {
-                struct device_node *host1x_node =
-                        of_find_node_by_phandle(be32_to_cpup(host1x_ptr));
-
-                host1x_pdev = of_find_device_by_node(host1x_node);
-                if (!host1x_pdev) {
-                        dev_warn(dev, "host1x device not available");
-                        return -EPROBE_DEFER;
-                }
-
-        } else {
-                host1x_pdev = to_platform_device(dev->parent);
-                dev_warn(dev, "host1x reference not found. assuming host1x to be parent");
-        }
-
-        platform->g->host1x_dev = host1x_pdev;
-
+        return nvgpu_get_nvhost_dev(platform->g);
+#else
         return 0;
+#endif
 }
 
 struct gk20a_platform vgpu_tegra_platform = {
diff --git a/drivers/gpu/nvgpu/tegra/linux/platform_gk20a_tegra.c b/drivers/gpu/nvgpu/tegra/linux/platform_gk20a_tegra.c
index 8a83f7d0..f55ea6d2 100644
--- a/drivers/gpu/nvgpu/tegra/linux/platform_gk20a_tegra.c
+++ b/drivers/gpu/nvgpu/tegra/linux/platform_gk20a_tegra.c
@@ -15,9 +15,6 @@
 
 #include <linux/version.h>
 #include <linux/of_platform.h>
-#ifdef CONFIG_TEGRA_GK20A_NVHOST
-#include <linux/nvhost.h>
-#endif
 #include <linux/debugfs.h>
 #include <linux/platform_data/tegra_edp.h>
 #include <linux/delay.h>
@@ -48,6 +45,7 @@
 #include <nvgpu/kmem.h>
 #include <nvgpu/bug.h>
 #include <nvgpu/enabled.h>
+#include <nvgpu/nvhost.h>
 
 #include <nvgpu/linux/dma.h>
 
@@ -721,8 +719,8 @@ void gk20a_tegra_debug_dump(struct device *dev)
         struct gk20a_platform *platform = gk20a_get_platform(dev);
         struct gk20a *g = platform->g;
 
-        if (g->host1x_dev)
-                nvhost_debug_dump_device(g->host1x_dev);
+        if (g->nvhost_dev)
+                nvgpu_nvhost_debug_dump_device(g->nvhost_dev);
 #endif
 }
 
@@ -732,8 +730,8 @@ int gk20a_tegra_busy(struct device *dev)
         struct gk20a_platform *platform = gk20a_get_platform(dev);
         struct gk20a *g = platform->g;
 
-        if (g->host1x_dev)
-                return nvhost_module_busy_ext(g->host1x_dev);
+        if (g->nvhost_dev)
+                return nvgpu_nvhost_module_busy_ext(g->nvhost_dev);
 #endif
         return 0;
 }
@@ -744,8 +742,8 @@ void gk20a_tegra_idle(struct device *dev)
         struct gk20a_platform *platform = gk20a_get_platform(dev);
         struct gk20a *g = platform->g;
 
-        if (g->host1x_dev)
-                nvhost_module_idle_ext(g->host1x_dev);
+        if (g->nvhost_dev)
+                nvgpu_nvhost_module_idle_ext(g->nvhost_dev);
 #endif
 }
 
@@ -864,8 +862,6 @@ static int gk20a_tegra_probe(struct device *dev)
 {
         struct gk20a_platform *platform = dev_get_drvdata(dev);
         struct device_node *np = dev->of_node;
-        const __be32 *host1x_ptr;
-        struct platform_device *host1x_pdev = NULL;
         bool joint_xpu_rail = false;
         int ret;
 
@@ -887,23 +883,11 @@ static int gk20a_tegra_probe(struct device *dev)
         }
 #endif
 
-        host1x_ptr = of_get_property(np, "nvidia,host1x", NULL);
-        if (host1x_ptr) {
-                struct device_node *host1x_node =
-                        of_find_node_by_phandle(be32_to_cpup(host1x_ptr));
-
-                host1x_pdev = of_find_device_by_node(host1x_node);
-                if (!host1x_pdev) {
-                        dev_warn(dev, "host1x device not available");
-                        return -EPROBE_DEFER;
-                }
-
-        } else {
-                host1x_pdev = to_platform_device(dev->parent);
-                dev_warn(dev, "host1x reference not found. assuming host1x to be parent");
-        }
-
-        platform->g->host1x_dev = host1x_pdev;
+#ifdef CONFIG_TEGRA_GK20A_NVHOST
+        ret = nvgpu_get_nvhost_dev(platform->g);
+        if (ret)
+                return ret;
+#endif
 
 #ifdef CONFIG_OF
         joint_xpu_rail = of_property_read_bool(of_chosen,
@@ -967,6 +951,10 @@ static int gk20a_tegra_remove(struct device *dev)
         /* deinitialise tegra specific scaling quirks */
         gk20a_tegra_scale_exit(dev);
 
+#ifdef CONFIG_TEGRA_GK20A_NVHOST
+        nvgpu_free_nvhost_dev(get_gk20a(dev));
+#endif
+
         return 0;
 }
 
diff --git a/drivers/gpu/nvgpu/tegra/linux/platform_gp10b_tegra.c b/drivers/gpu/nvgpu/tegra/linux/platform_gp10b_tegra.c
index 8b530fac..235473d3 100644
--- a/drivers/gpu/nvgpu/tegra/linux/platform_gp10b_tegra.c
+++ b/drivers/gpu/nvgpu/tegra/linux/platform_gp10b_tegra.c
@@ -14,7 +14,6 @@
  */
 
 #include <linux/of_platform.h>
-#include <linux/nvhost.h>
 #include <linux/debugfs.h>
 #include <linux/dma-buf.h>
 #include <linux/nvmap.h>
@@ -30,6 +29,7 @@
 #include <nvgpu/bug.h>
 #include <nvgpu/enabled.h>
 #include <nvgpu/hashtable.h>
+#include <nvgpu/nvhost.h>
 
 #include "clk.h"
 
@@ -130,26 +130,14 @@ static void gp10b_tegra_scale_exit(struct device *dev)
 static int gp10b_tegra_probe(struct device *dev)
 {
         struct gk20a_platform *platform = dev_get_drvdata(dev);
-        struct gk20a *g = platform->g;
-        struct device_node *np = dev->of_node;
-        struct device_node *host1x_node;
-        struct platform_device *host1x_pdev;
-        const __be32 *host1x_ptr;
-
-        host1x_ptr = of_get_property(np, "nvidia,host1x", NULL);
-        if (!host1x_ptr) {
-                nvgpu_err(g, "host1x device not available");
-                return -ENOSYS;
-        }
+#ifdef CONFIG_TEGRA_GK20A_NVHOST
+        int ret;
 
-        host1x_node = of_find_node_by_phandle(be32_to_cpup(host1x_ptr));
-        host1x_pdev = of_find_device_by_node(host1x_node);
-        if (!host1x_pdev) {
-                nvgpu_err(g, "host1x device not available");
-                return -ENOSYS;
-        }
+        ret = nvgpu_get_nvhost_dev(platform->g);
+        if (ret)
+                return ret;
+#endif
 
-        platform->g->host1x_dev = host1x_pdev;
         platform->bypass_smmu = !device_is_iommuable(dev);
         platform->disable_bigpage = platform->bypass_smmu;
 
@@ -190,8 +178,11 @@ int gp10b_tegra_remove(struct device *dev)
         /* deinitialise tegra specific scaling quirks */
         gp10b_tegra_scale_exit(dev);
 
-        return 0;
+#ifdef CONFIG_TEGRA_GK20A_NVHOST
+        nvgpu_free_nvhost_dev(get_gk20a(dev));
+#endif
 
+        return 0;
 }
 
 static bool gp10b_tegra_is_railgated(struct device *dev)