gpu: nvgpu: support kernel-3.10 version

Make necessary changes to support nvgpu on kernel-3.10

This includes below changes
- PROBE_PREFER_ASYNCHRONOUS is defined only for K3.10
- Fence handling and struct sync_fence is different between
  K3.10 and K3.18
- variable status in struct sync_fence is atomic on K3.18
  whereas it is int on K3.10
- if SOC == T132, set soc_name = "tegra13x"
- ioremap_cache() is not defined on K3.10 ARM versions,
  hence use ioremap_cached()

Bug 200188753

Change-Id: I18d77eb1404e15054e8510d67c9a61c0f1883e2b
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/1121092
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>

drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c

@@ -16,6 +16,7 @@
  */
 
 #include <linux/gk20a.h>
+#include <linux/version.h>
 
 #include "channel_sync_gk20a.h"
 #include "gk20a.h"
@@ -93,10 +94,12 @@ static int gk20a_channel_syncpt_wait_fd(struct gk20a_channel_sync *s, int fd,
 	int i;
 	int num_wait_cmds;
 	struct sync_fence *sync_fence;
+	struct sync_pt *pt;
 	struct priv_cmd_entry *wait_cmd = NULL;
 	struct gk20a_channel_syncpt *sp =
 		container_of(s, struct gk20a_channel_syncpt, ops);
 	struct channel_gk20a *c = sp->c;
+	u32 wait_id;
 	int err = 0;
 
 	sync_fence = nvhost_sync_fdget(fd);
@@ -104,9 +107,13 @@ static int gk20a_channel_syncpt_wait_fd(struct gk20a_channel_sync *s, int fd,
 		return -EINVAL;
 
 	/* validate syncpt ids */
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3,18,0)
+	list_for_each_entry(pt, &sync_fence->pt_list_head, pt_list) {
+#else
 	for (i = 0; i < sync_fence->num_fences; i++) {
-		struct sync_pt *pt = sync_pt_from_fence(sync_fence->cbs[i].sync_pt);
-		u32 wait_id = nvhost_sync_pt_id(pt);
+		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)) {
 			sync_fence_put(sync_fence);
@@ -129,9 +136,13 @@ static int gk20a_channel_syncpt_wait_fd(struct gk20a_channel_sync *s, int fd,
 	}
 
 	i = 0;
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3,18,0)
+	list_for_each_entry(pt, &sync_fence->pt_list_head, pt_list) {
+#else
 	for (i = 0; i < sync_fence->num_fences; i++) {
 		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);
 
@@ -144,6 +155,9 @@ static int gk20a_channel_syncpt_wait_fd(struct gk20a_channel_sync *s, int fd,
 		} else
 			add_wait_cmd(&wait_cmd->ptr[i * 4], wait_id,
 					wait_value);
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3,18,0)
+		i++;
+#endif
 	}
 	WARN_ON(i != num_wait_cmds);
 	sync_fence_put(sync_fence);

drivers/gpu/nvgpu/gk20a/fence_gk20a.c

@@ -15,6 +15,7 @@
 
 #include <linux/gk20a.h>
 #include <linux/file.h>
+#include <linux/version.h>
 
 #include "gk20a.h"
 #include "semaphore_gk20a.h"
@@ -153,10 +154,17 @@ struct gk20a_fence *gk20a_fence_from_semaphore(
 	struct sync_fence *sync_fence = NULL;
 
 #ifdef CONFIG_SYNC
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3,18,0)
+	sync_fence = gk20a_sync_fence_create(timeline, semaphore,
+					     dependency, "f-gk20a-0x%04llx",
+					     ((uintptr_t)(void *)semaphore->value) &
+					     0xffff);
+#else
 	sync_fence = gk20a_sync_fence_create(timeline, semaphore,
 					     dependency, "f-gk20a-0x%04llx",
 					     ((u64)(void *)semaphore->value) &
 					     0xffff);
+#endif
 	if (!sync_fence)
 		return NULL;
 #endif

drivers/gpu/nvgpu/gk20a/gk20a.c

@@ -43,6 +43,7 @@
 #include <linux/reset.h>
 
 #include <linux/sched.h>
+#include <linux/version.h>
 
 #include "gk20a.h"
 #include "debug_gk20a.h"
@@ -1667,7 +1668,9 @@ static struct platform_driver gk20a_driver = {
 	.driver = {
 		.owner = THIS_MODULE,
 		.name = "gk20a",
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(3,18,0)
 		.probe_type = PROBE_PREFER_ASYNCHRONOUS,
+#endif
 #ifdef CONFIG_OF
 		.of_match_table = tegra_gk20a_of_match,
 #endif

drivers/gpu/nvgpu/gk20a/platform_gk20a_tegra.c

@@ -770,6 +770,9 @@ static int gk20a_tegra_probe(struct device *dev)
 		}
 	}
 
+	if (tegra_get_chipid() == TEGRA_CHIPID_TEGRA13)
+		platform->soc_name = "tegra13x";
+
 	gk20a_tegra_get_clocks(dev);
 
 	return 0;

drivers/gpu/nvgpu/gk20a/sync_gk20a.c

@@ -15,6 +15,7 @@
 
 #include "sync_gk20a.h"
 
+#include <linux/version.h>
 #include <linux/kernel.h>
 #include <linux/file.h>
 #include <linux/fs.h>
@@ -149,7 +150,11 @@ static struct gk20a_sync_pt *gk20a_sync_pt_create_shared(
 
 	/* Store the dependency fence for this pt. */
 	if (dependency) {
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3,18,0)
+		if (dependency->status == 0) {
+#else
 		if (!atomic_read(&dependency->status)) {
+#endif
 			shared->dep = dependency;
 		} else {
 			shared->dep_timestamp = ktime_get();
@@ -214,6 +219,9 @@ static int gk20a_sync_pt_has_signaled(struct sync_pt *sync_pt)
 {
 	struct gk20a_sync_pt *pt = to_gk20a_sync_pt(sync_pt);
 	struct gk20a_sync_timeline *obj = pt->obj;
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3,18,0)
+	struct sync_pt *pos;
+#endif
 	bool signaled = true;
 
 	spin_lock(&pt->lock);
@@ -233,6 +241,12 @@ static int gk20a_sync_pt_has_signaled(struct sync_pt *sync_pt)
 		 * first.*/
 		if (pt->dep) {
 			s64 ns = 0;
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3,18,0)
+			struct list_head *dep_pts = &pt->dep->pt_list_head;
+			list_for_each_entry(pos, dep_pts, pt_list) {
+				ns = max(ns, ktime_to_ns(pos->timestamp));
+			}
+#else
 			struct fence *fence;
 			int i;
 
@@ -240,6 +254,7 @@ static int gk20a_sync_pt_has_signaled(struct sync_pt *sync_pt)
 				fence = pt->dep->cbs[i].sync_pt;
 				ns = max(ns, ktime_to_ns(fence->timestamp));
 			}
+#endif
 			pt->dep_timestamp = ns_to_ktime(ns);
 			sync_fence_put(pt->dep);
 			pt->dep = NULL;
@@ -260,7 +275,11 @@ static inline ktime_t gk20a_sync_pt_duration(struct sync_pt *sync_pt)
 	struct gk20a_sync_pt *pt = to_gk20a_sync_pt(sync_pt);
 	if (!gk20a_sync_pt_has_signaled(sync_pt) || !pt->dep_timestamp.tv64)
 		return ns_to_ktime(0);
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3,18,0)
+	return ktime_sub(sync_pt->timestamp, pt->dep_timestamp);
+#else
 	return ktime_sub(sync_pt->base.timestamp, pt->dep_timestamp);
+#endif
 }
 
 static int gk20a_sync_pt_compare(struct sync_pt *a, struct sync_pt *b)

drivers/gpu/nvgpu/vgpu/fecs_trace_vgpu.c

@@ -14,6 +14,7 @@
 #include <linux/string.h>
 #include <linux/tegra-ivc.h>
 #include <linux/tegra_vgpu.h>
+#include <linux/version.h>
 
 #include "gk20a/gk20a.h"
 #include "gk20a/ctxsw_trace_gk20a.h"
@@ -61,7 +62,11 @@ static int vgpu_fecs_trace_init(struct gk20a *g)
 		goto fail;
 	}
 
+#if LINUX_VERSION_CODE < KERNEL_VERSION(3,18,0)
+	vcst->buf = ioremap_cached(vcst->cookie->ipa, vcst->cookie->size);
+#else
 	vcst->buf = ioremap_cache(vcst->cookie->ipa, vcst->cookie->size);
+#endif
 	if (!vcst->buf) {
 		dev_info(dev_from_gk20a(g), "ioremap_cache failed\n");
 		err = -EINVAL;