gpu: nvgpu: Use gv11b_css_hw_set_handled_snapshots for GV11B

The value of NV_PERF_PMASYS_MEM_BUMP is different for Volta
and NVGPU_IOCTL_CHANNEL_CYCLE_STATS_SNAPSHOT_CMD_FLUSH did not
have correct behavior on GV11B due to that.
The patch adds an instance of css_hw_set_handled_snapshots
for Volta to fix that.
The patch also renames css_hw_set_handled_snapshots
to gk20a_css_hw_set_handled_snapshots to make it more clear
that the function is arch dependent.

Bug 1960846

Change-Id: I92c35a862ecd7f918dd1458c086fc7ae42ca8fc5
Signed-off-by: Martin Radev <mradev@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1662427
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

drivers/gpu/nvgpu/gk20a/css_gr_gk20a.c

@@ -75,7 +75,7 @@ static inline u32 css_hw_get_pending_snapshots(struct gk20a *g)
 }
 
 /* informs hw how many snapshots have been processed (frees up fifo space) */
-void css_hw_set_handled_snapshots(struct gk20a *g, u32 done)
+void gk20a_css_hw_set_handled_snapshots(struct gk20a *g, u32 done)
 {
 	if (done > 0) {
 		gk20a_writel(g, perf_pmasys_mem_bump_r(),
@@ -102,7 +102,7 @@ static void css_hw_reset_streaming(struct gk20a *g)
 			perf_pmasys_control_membuf_clear_status_doit_f());
 
 	/* pointing all pending snapshots as handled */
-	css_hw_set_handled_snapshots(g, css_hw_get_pending_snapshots(g));
+	gk20a_css_hw_set_handled_snapshots(g, css_hw_get_pending_snapshots(g));
 }
 
 /*

drivers/gpu/nvgpu/gk20a/css_gr_gk20a.h

@@ -129,7 +129,7 @@ struct gk20a_cs_snapshot {
 	struct gk20a_cs_snapshot_fifo_entry	*hw_get;
 };
 
-void css_hw_set_handled_snapshots(struct gk20a *g, u32 done);
+void gk20a_css_hw_set_handled_snapshots(struct gk20a *g, u32 done);
 int css_hw_enable_snapshot(struct channel_gk20a *ch,
 				struct gk20a_cs_snapshot_client *cs_client);
 void css_hw_disable_snapshot(struct gr_gk20a *gr);

drivers/gpu/nvgpu/gm20b/hal_gm20b.c

@@ -577,7 +577,7 @@ static const struct gpu_ops gm20b_ops = {
 		.enable_snapshot = css_hw_enable_snapshot,
 		.disable_snapshot = css_hw_disable_snapshot,
 		.check_data_available = css_hw_check_data_available,
-		.set_handled_snapshots = css_hw_set_handled_snapshots,
+		.set_handled_snapshots = gk20a_css_hw_set_handled_snapshots,
 		.allocate_perfmon_ids = css_gr_allocate_perfmon_ids,
 		.release_perfmon_ids = css_gr_release_perfmon_ids,
 	},

drivers/gpu/nvgpu/gp106/hal_gp106.c

@@ -689,7 +689,7 @@ static const struct gpu_ops gp106_ops = {
 		.enable_snapshot = css_hw_enable_snapshot,
 		.disable_snapshot = css_hw_disable_snapshot,
 		.check_data_available = css_hw_check_data_available,
-		.set_handled_snapshots = css_hw_set_handled_snapshots,
+		.set_handled_snapshots = gk20a_css_hw_set_handled_snapshots,
 		.allocate_perfmon_ids = css_gr_allocate_perfmon_ids,
 		.release_perfmon_ids = css_gr_release_perfmon_ids,
 	},

drivers/gpu/nvgpu/gp10b/hal_gp10b.c

@@ -623,7 +623,7 @@ static const struct gpu_ops gp10b_ops = {
 		.enable_snapshot = css_hw_enable_snapshot,
 		.disable_snapshot = css_hw_disable_snapshot,
 		.check_data_available = css_hw_check_data_available,
-		.set_handled_snapshots = css_hw_set_handled_snapshots,
+		.set_handled_snapshots = gk20a_css_hw_set_handled_snapshots,
 		.allocate_perfmon_ids = css_gr_allocate_perfmon_ids,
 		.release_perfmon_ids = css_gr_release_perfmon_ids,
 	},

drivers/gpu/nvgpu/gv100/hal_gv100.c

@@ -699,7 +699,7 @@ static const struct gpu_ops gv100_ops = {
 		.enable_snapshot = gv11b_css_hw_enable_snapshot,
 		.disable_snapshot = gv11b_css_hw_disable_snapshot,
 		.check_data_available = gv11b_css_hw_check_data_available,
-		.set_handled_snapshots = css_hw_set_handled_snapshots,
+		.set_handled_snapshots = gv11b_css_hw_set_handled_snapshots,
 		.allocate_perfmon_ids = css_gr_allocate_perfmon_ids,
 		.release_perfmon_ids = css_gr_release_perfmon_ids,
 	},

drivers/gpu/nvgpu/gv11b/css_gr_gv11b.c

@@ -57,15 +57,6 @@ static inline u32 css_hw_get_pending_snapshots(struct gk20a *g)
 			sizeof(struct gk20a_cs_snapshot_fifo_entry);
 }
 
-/* informs hw how many snapshots have been processed (frees up fifo space) */
-static inline void gv11b_css_hw_set_handled_snapshots(struct gk20a *g, u32 done)
-{
-	if (done > 0) {
-		gk20a_writel(g, perf_pmasys_mem_bump_r(),
-		     done * sizeof(struct gk20a_cs_snapshot_fifo_entry));
-	}
-}
-
 /* disable streaming to memory */
 static void gv11b_css_hw_reset_streaming(struct gk20a *g)
 {
@@ -86,6 +77,17 @@ static void gv11b_css_hw_reset_streaming(struct gk20a *g)
 	gv11b_css_hw_set_handled_snapshots(g, css_hw_get_pending_snapshots(g));
 }
 
+/* informs hw how many snapshots have been processed (frees up fifo space) */
+void gv11b_css_hw_set_handled_snapshots(struct gk20a *g, u32 done)
+{
+	if (done == 0u) {
+		return;
+	}
+
+	gk20a_writel(g, perf_pmasys_mem_bump_r(),
+		done * sizeof(struct gk20a_cs_snapshot_fifo_entry));
+}
+
 int gv11b_css_hw_enable_snapshot(struct channel_gk20a *ch,
 				struct gk20a_cs_snapshot_client *cs_client)
 {

drivers/gpu/nvgpu/gv11b/css_gr_gv11b.h

@@ -30,5 +30,6 @@ int gv11b_css_hw_enable_snapshot(struct channel_gk20a *ch,
 void gv11b_css_hw_disable_snapshot(struct gr_gk20a *gr);
 int gv11b_css_hw_check_data_available(struct channel_gk20a *ch, u32 *pending,
 			bool *hw_overflow);
+void gv11b_css_hw_set_handled_snapshots(struct gk20a *g, u32 done);
 
 #endif /* CSS_GR_GV11B_H */

drivers/gpu/nvgpu/gv11b/hal_gv11b.c

@@ -688,7 +688,7 @@ static const struct gpu_ops gv11b_ops = {
 		.enable_snapshot = gv11b_css_hw_enable_snapshot,
 		.disable_snapshot = gv11b_css_hw_disable_snapshot,
 		.check_data_available = gv11b_css_hw_check_data_available,
-		.set_handled_snapshots = css_hw_set_handled_snapshots,
+		.set_handled_snapshots = gv11b_css_hw_set_handled_snapshots,
 		.allocate_perfmon_ids = css_gr_allocate_perfmon_ids,
 		.release_perfmon_ids = css_gr_release_perfmon_ids,
 	},