gpu: nvgpu: allocate separate client managed syncpoint for User

We right now allocate a nvgpu managed syncpoint in c->sync and share
that with user space

But to avoid conflicts between user space and kernel space increments
allocate a separate "client managed" syncpoint for User space in c->user_sync

Add new API nvgpu_nvhost_get_syncpt_client_managed() to request a client managed
syncpoint from nvhost.
Note that nvhost/nvgpu do not keep track of MAX/threshold value of this syncpoint

Update gk20a_channel_syncpt_create() to receive a flag to indicate whether a
User space syncpoint is required or not

Unset NVGPU_SUPPORT_USER_SYNCPOINT for gp10b since we don't want to allocate
double syncpoints per channel on that platform

For gv11b, once we move to use user space submits, support for c->sync will be
dropped so we keep using only one syncpoint per channel

Bug 200326065
Jira NVGPU-179

Change-Id: I78d94de4276db1c897ea2a4fe4c2db8b2a179722
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1665828
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

drivers/gpu/nvgpu/common/linux/channel.c

@@ -444,7 +444,7 @@ static int gk20a_submit_prepare_syncs(struct channel_gk20a *c,
 	if (g->aggressive_sync_destroy_thresh) {
 		nvgpu_mutex_acquire(&c->sync_lock);
 		if (!c->sync) {
-			c->sync = gk20a_channel_sync_create(c);
+			c->sync = gk20a_channel_sync_create(c, false);
 			if (!c->sync) {
 				err = -ENOMEM;
 				nvgpu_mutex_release(&c->sync_lock);

drivers/gpu/nvgpu/common/linux/ioctl_channel.c

@@ -972,11 +972,11 @@ static int nvgpu_ioctl_channel_get_user_syncpoint(struct channel_gk20a *ch,
 	}
 
 	nvgpu_mutex_acquire(&ch->sync_lock);
-	if (ch->sync) {
+	if (ch->user_sync) {
 		nvgpu_mutex_release(&ch->sync_lock);
 	} else {
-		ch->sync = gk20a_channel_sync_create(ch);
+		ch->user_sync = gk20a_channel_sync_create(ch, true);
-		if (!ch->sync) {
+		if (!ch->user_sync) {
 			nvgpu_mutex_release(&ch->sync_lock);
 			return -ENOMEM;
 		}
@@ -989,11 +989,11 @@ static int nvgpu_ioctl_channel_get_user_syncpoint(struct channel_gk20a *ch,
 		}
 	}
 
-	args->syncpoint_id = ch->sync->syncpt_id(ch->sync);
+	args->syncpoint_id = ch->user_sync->syncpt_id(ch->user_sync);
 	args->syncpoint_max = nvgpu_nvhost_syncpt_read_maxval(g->nvhost_dev,
 						args->syncpoint_id);
 	if (nvgpu_is_enabled(g, NVGPU_SUPPORT_SYNCPOINT_ADDRESS))
-		args->gpu_va = ch->sync->syncpt_address(ch->sync);
+		args->gpu_va = ch->user_sync->syncpt_address(ch->user_sync);
 	else
 		args->gpu_va = 0;
 

drivers/gpu/nvgpu/common/linux/nvhost.c

@@ -138,6 +138,14 @@ u32 nvgpu_nvhost_get_syncpt_host_managed(
 			param, syncpt_name);
 }
 
+u32 nvgpu_nvhost_get_syncpt_client_managed(
+	struct nvgpu_nvhost_dev *nvhost_dev,
+	const char *syncpt_name)
+{
+	return nvhost_get_syncpt_client_managed(nvhost_dev->host1x_pdev,
+			syncpt_name);
+}
+
 int nvgpu_nvhost_syncpt_wait_timeout_ext(
 	struct nvgpu_nvhost_dev *nvhost_dev, u32 id,
 	u32 thresh, u32 timeout, u32 *value, struct timespec *ts)

drivers/gpu/nvgpu/gk20a/channel_gk20a.c

@@ -438,6 +438,10 @@ static void gk20a_free_channel(struct channel_gk20a *ch, bool force)
 		gk20a_channel_sync_destroy(ch->sync);
 		ch->sync = NULL;
 	}
+	if (ch->user_sync) {
+		gk20a_channel_sync_destroy(ch->user_sync);
+		ch->user_sync = NULL;
+	}
 	nvgpu_mutex_release(&ch->sync_lock);
 
 	/*
@@ -1147,7 +1151,7 @@ int gk20a_channel_alloc_gpfifo(struct channel_gk20a *c,
 
 	if (!g->aggressive_sync_destroy_thresh) {
 		nvgpu_mutex_acquire(&c->sync_lock);
-		c->sync = gk20a_channel_sync_create(c);
+		c->sync = gk20a_channel_sync_create(c, false);
 		if (!c->sync) {
 			err = -ENOMEM;
 			nvgpu_mutex_release(&c->sync_lock);

drivers/gpu/nvgpu/gk20a/channel_gk20a.h

@@ -226,6 +226,7 @@ struct channel_gk20a {
 
 	struct nvgpu_mutex sync_lock;
 	struct gk20a_channel_sync *sync;
+	struct gk20a_channel_sync *user_sync;
 
 #ifdef CONFIG_TEGRA_GR_VIRTUALIZATION
 	u64 virt_ctx;

drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c

@@ -315,7 +315,7 @@ static void gk20a_channel_syncpt_destroy(struct gk20a_channel_sync *s)
 }
 
 static struct gk20a_channel_sync *
-gk20a_channel_syncpt_create(struct channel_gk20a *c)
+gk20a_channel_syncpt_create(struct channel_gk20a *c, bool user_managed)
 {
 	struct gk20a_channel_syncpt *sp;
 	char syncpt_name[32];
@@ -327,11 +327,19 @@ gk20a_channel_syncpt_create(struct channel_gk20a *c)
 	sp->c = c;
 	sp->nvhost_dev = c->g->nvhost_dev;
 
+	if (user_managed) {
+		snprintf(syncpt_name, sizeof(syncpt_name),
+			"%s_%d_user", c->g->name, c->chid);
+
+		sp->id = nvgpu_nvhost_get_syncpt_client_managed(sp->nvhost_dev,
+						syncpt_name);
+	} else {
 		snprintf(syncpt_name, sizeof(syncpt_name),
 			"%s_%d", c->g->name, c->chid);
 
 		sp->id = nvgpu_nvhost_get_syncpt_host_managed(sp->nvhost_dev,
 						c->chid, syncpt_name);
+	}
 	if (!sp->id) {
 		nvgpu_kfree(c->g, sp);
 		nvgpu_err(c->g, "failed to get free syncpt");
@@ -892,7 +900,7 @@ static void gk20a_channel_semaphore_destroy(struct gk20a_channel_sync *s)
 }
 
 static struct gk20a_channel_sync *
-gk20a_channel_semaphore_create(struct channel_gk20a *c)
+gk20a_channel_semaphore_create(struct channel_gk20a *c, bool user_managed)
 {
 	int asid = -1;
 	struct gk20a_channel_semaphore *sema;
@@ -940,13 +948,14 @@ void gk20a_channel_sync_destroy(struct gk20a_channel_sync *sync)
 	sync->destroy(sync);
 }
 
-struct gk20a_channel_sync *gk20a_channel_sync_create(struct channel_gk20a *c)
+struct gk20a_channel_sync *gk20a_channel_sync_create(struct channel_gk20a *c,
+	bool user_managed)
 {
 #ifdef CONFIG_TEGRA_GK20A_NVHOST
 	if (gk20a_platform_has_syncpoints(c->g))
-		return gk20a_channel_syncpt_create(c);
+		return gk20a_channel_syncpt_create(c, user_managed);
 #endif
-	return gk20a_channel_semaphore_create(c);
+	return gk20a_channel_semaphore_create(c, user_managed);
 }
 
 bool gk20a_channel_sync_needs_sync_framework(struct gk20a *g)

drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.h

@@ -110,7 +110,8 @@ struct gk20a_channel_sync {
 };
 
 void gk20a_channel_sync_destroy(struct gk20a_channel_sync *sync);
-struct gk20a_channel_sync *gk20a_channel_sync_create(struct channel_gk20a *c);
+struct gk20a_channel_sync *gk20a_channel_sync_create(struct channel_gk20a *c,
+	bool user_managed);
 bool gk20a_channel_sync_needs_sync_framework(struct gk20a *g);
 
 #ifdef CONFIG_SYNC

drivers/gpu/nvgpu/gp10b/gp10b.c

@@ -116,6 +116,5 @@ int gp10b_init_gpu_characteristics(struct gk20a *g)
 	gk20a_init_gpu_characteristics(g);
 	gp10b_detect_ecc_enabled_units(g);
 	__nvgpu_set_enabled(g, NVGPU_SUPPORT_RESCHEDULE_RUNLIST, true);
-	__nvgpu_set_enabled(g, NVGPU_SUPPORT_USER_SYNCPOINT, true);
 	return 0;
 }

drivers/gpu/nvgpu/include/nvgpu/nvhost.h

@@ -65,6 +65,8 @@ void nvgpu_nvhost_syncpt_put_ref_ext(struct nvgpu_nvhost_dev *nvhost_dev,
 u32 nvgpu_nvhost_get_syncpt_host_managed(struct nvgpu_nvhost_dev *nvhost_dev,
 	u32 param,
 	const char *syncpt_name);
+u32 nvgpu_nvhost_get_syncpt_client_managed(struct nvgpu_nvhost_dev *nvhost_dev,
+	const char *syncpt_name);
 
 int nvgpu_nvhost_create_symlink(struct gk20a *g);
 void nvgpu_nvhost_remove_symlink(struct gk20a *g);