gpu: nvgpu: add cycle stats to debugger node

Add NVGPU_DBG_GPU_IOCTL_CYCLE_STATS to debugger node, to
install/uninstall a buffer for cycle stats.

Add NVGPU_DBG_GPU_IOCTL_CYCLE_STATS_SNAPSHOT to debugger
node, to attach/flush/detach a buffer for Mode-E streamout.

Those ioctls will apply to the first channel in the debug session.

Bug 200464613
Jira NVGPU-1442

Change-Id: I0b96d9a07c016690140292fa5886fda545697ee6
Signed-off-by: Thomas Fleury <tfleury@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/2002060
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

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

@@ -130,8 +130,7 @@ void gk20a_channel_free_cycle_stats_buffer(struct channel_gk20a *ch)
 	nvgpu_mutex_release(&ch->cyclestate.cyclestate_buffer_mutex);
 }
 
-static int gk20a_channel_cycle_stats(struct channel_gk20a *ch,
+int gk20a_channel_cycle_stats(struct channel_gk20a *ch, int dmabuf_fd)
-		       struct nvgpu_cycle_stats_args *args)
 {
 	struct dma_buf *dmabuf;
 	void *virtual_address;
@@ -141,10 +140,10 @@ static int gk20a_channel_cycle_stats(struct channel_gk20a *ch,
 	if (!nvgpu_is_enabled(ch->g, NVGPU_SUPPORT_CYCLE_STATS))
 		return -ENOSYS;
 
-	if (args->dmabuf_fd && !priv->cyclestate_buffer_handler) {
+	if (dmabuf_fd && !priv->cyclestate_buffer_handler) {
 
 		/* set up new cyclestats buffer */
-		dmabuf = dma_buf_get(args->dmabuf_fd);
+		dmabuf = dma_buf_get(dmabuf_fd);
 		if (IS_ERR(dmabuf))
 			return PTR_ERR(dmabuf);
 		virtual_address = dma_buf_vmap(dmabuf);
@@ -156,12 +155,12 @@ static int gk20a_channel_cycle_stats(struct channel_gk20a *ch,
 		ch->cyclestate.cyclestate_buffer_size = dmabuf->size;
 		return 0;
 
-	} else if (!args->dmabuf_fd && priv->cyclestate_buffer_handler) {
+	} else if (!dmabuf_fd && priv->cyclestate_buffer_handler) {
 		gk20a_channel_free_cycle_stats_buffer(ch);
 		return 0;
 
-	} else if (!args->dmabuf_fd && !priv->cyclestate_buffer_handler) {
+	} else if (!dmabuf_fd && !priv->cyclestate_buffer_handler) {
-		/* no requst from GL */
+		/* no request from GL */
 		return 0;
 
 	} else {
@@ -170,7 +169,7 @@ static int gk20a_channel_cycle_stats(struct channel_gk20a *ch,
 	}
 }
 
-static int gk20a_flush_cycle_stats_snapshot(struct channel_gk20a *ch)
+int gk20a_flush_cycle_stats_snapshot(struct channel_gk20a *ch)
 {
 	int ret;
 
@@ -184,7 +183,7 @@ static int gk20a_flush_cycle_stats_snapshot(struct channel_gk20a *ch)
 	return ret;
 }
 
-static int gk20a_attach_cycle_stats_snapshot(struct channel_gk20a *ch,
+int gk20a_attach_cycle_stats_snapshot(struct channel_gk20a *ch,
 				u32 dmabuf_fd,
 				u32 perfmon_id_count,
 				u32 *perfmon_id_start)
@@ -1289,7 +1288,7 @@ long gk20a_channel_ioctl(struct file *filp,
 			break;
 		}
 		err = gk20a_channel_cycle_stats(ch,
-				(struct nvgpu_cycle_stats_args *)buf);
+				((struct nvgpu_cycle_stats_args *)buf)->dmabuf_fd);
 		gk20a_idle(ch->g);
 		break;
 #endif

drivers/gpu/nvgpu/os/linux/ioctl_channel.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017, NVIDIA CORPORATION.  All rights reserved.
+ * Copyright (c) 2017-2019, NVIDIA CORPORATION.  All rights reserved.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms and conditions of the GNU General Public License,
@@ -36,9 +36,16 @@ long gk20a_channel_ioctl(struct file *filp,
 int gk20a_channel_open_ioctl(struct gk20a *g,
 		struct nvgpu_channel_open_args *args);
 
-int gk20a_channel_free_cycle_stats_snapshot(struct channel_gk20a *ch);
+int gk20a_channel_cycle_stats(struct channel_gk20a *ch, int dmabuf_fd);
 void gk20a_channel_free_cycle_stats_buffer(struct channel_gk20a *ch);
 
+int gk20a_attach_cycle_stats_snapshot(struct channel_gk20a *ch,
+				u32 dmabuf_fd,
+				u32 perfmon_id_count,
+				u32 *perfmon_id_start);
+int gk20a_flush_cycle_stats_snapshot(struct channel_gk20a *ch);
+int gk20a_channel_free_cycle_stats_snapshot(struct channel_gk20a *ch);
+
 extern const struct file_operations gk20a_channel_ops;
 
 u32 nvgpu_get_common_runlist_level(u32 level);

drivers/gpu/nvgpu/os/linux/ioctl_dbg.c

@@ -42,6 +42,7 @@
 #include "os_linux.h"
 #include "platform_gk20a.h"
 #include "ioctl_dbg.h"
+#include "ioctl_channel.h"
 #include "dmabuf_vidmem.h"
 
 struct dbg_session_gk20a_linux {
@@ -1855,6 +1856,87 @@ static int nvgpu_dbg_gpu_set_sm_exception_type_mask(struct dbg_session_gk20a *db
 	return err;
 }
 
+#if defined(CONFIG_GK20A_CYCLE_STATS)
+static int nvgpu_dbg_gpu_cycle_stats(struct dbg_session_gk20a *dbg_s,
+			struct nvgpu_dbg_gpu_cycle_stats_args *args)
+{
+	struct channel_gk20a *ch = NULL;
+	int err;
+
+	ch = nvgpu_dbg_gpu_get_session_channel(dbg_s);
+	if (ch == NULL) {
+		return -EINVAL;
+	}
+
+	err = gk20a_busy(ch->g);
+	if (err != 0) {
+		return err;
+	}
+
+	err = gk20a_channel_cycle_stats(ch, args->dmabuf_fd);
+
+	gk20a_idle(ch->g);
+	return err;
+}
+
+static int nvgpu_dbg_gpu_cycle_stats_snapshot(struct dbg_session_gk20a *dbg_s,
+		struct nvgpu_dbg_gpu_cycle_stats_snapshot_args *args)
+{
+	struct channel_gk20a *ch = NULL;
+	int err;
+
+	if (!args->dmabuf_fd) {
+		return -EINVAL;
+	}
+
+	nvgpu_speculation_barrier();
+
+	ch = nvgpu_dbg_gpu_get_session_channel(dbg_s);
+	if (ch == NULL) {
+		return -EINVAL;
+	}
+
+	/* is it allowed to handle calls for current GPU? */
+	if (!nvgpu_is_enabled(ch->g, NVGPU_SUPPORT_CYCLE_STATS_SNAPSHOT)) {
+		return -ENOSYS;
+	}
+
+	err = gk20a_busy(ch->g);
+	if (err != 0) {
+		return err;
+	}
+
+	/* handle the command (most frequent cases first) */
+	switch (args->cmd) {
+	case NVGPU_DBG_GPU_IOCTL_CYCLE_STATS_SNAPSHOT_CMD_FLUSH:
+		err = gk20a_flush_cycle_stats_snapshot(ch);
+		args->extra = 0;
+		break;
+
+	case NVGPU_DBG_GPU_IOCTL_CYCLE_STATS_SNAPSHOT_CMD_ATTACH:
+		err = gk20a_attach_cycle_stats_snapshot(ch,
+						args->dmabuf_fd,
+						args->extra,
+						&args->extra);
+		break;
+
+	case NVGPU_DBG_GPU_IOCTL_CYCLE_STATS_SNAPSHOT_CMD_DETACH:
+		err = gk20a_channel_free_cycle_stats_snapshot(ch);
+		args->extra = 0;
+		break;
+
+	default:
+		pr_err("cyclestats: unknown command %u\n", args->cmd);
+		err = -EINVAL;
+		break;
+	}
+
+	gk20a_idle(ch->g);
+	return err;
+}
+
+#endif
+
 int gk20a_dbg_gpu_dev_open(struct inode *inode, struct file *filp)
 {
 	struct nvgpu_os_linux *l = container_of(inode->i_cdev,
@@ -2011,6 +2093,18 @@ long gk20a_dbg_gpu_dev_ioctl(struct file *filp, unsigned int cmd,
 		   (struct nvgpu_dbg_gpu_set_sm_exception_type_mask_args *)buf);
 		break;
 
+#ifdef CONFIG_GK20A_CYCLE_STATS
+	case NVGPU_DBG_GPU_IOCTL_CYCLE_STATS:
+		err = nvgpu_dbg_gpu_cycle_stats(dbg_s,
+				(struct nvgpu_dbg_gpu_cycle_stats_args *)buf);
+		break;
+
+	case NVGPU_DBG_GPU_IOCTL_CYCLE_STATS_SNAPSHOT:
+		err = nvgpu_dbg_gpu_cycle_stats_snapshot(dbg_s,
+				(struct nvgpu_dbg_gpu_cycle_stats_snapshot_args *)buf);
+		break;
+#endif
+
 	default:
 		nvgpu_err(g,
 			   "unrecognized dbg gpu ioctl cmd: 0x%x",

include/uapi/linux/nvgpu.h

@@ -1,7 +1,7 @@
 /*
  * NVGPU Public Interface Header
  *
- * Copyright (c) 2011-2018, NVIDIA CORPORATION.  All rights reserved.
+ * Copyright (c) 2011-2019, NVIDIA CORPORATION.  All rights reserved.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms and conditions of the GNU General Public License,
@@ -1421,8 +1421,33 @@ struct nvgpu_dbg_gpu_set_sm_exception_type_mask_args {
 	_IOW(NVGPU_DBG_GPU_IOCTL_MAGIC, 23, \
 			struct nvgpu_dbg_gpu_set_sm_exception_type_mask_args)
 
+struct nvgpu_dbg_gpu_cycle_stats_args {
+	__u32 dmabuf_fd;
+	__u32 reserved;
+};
+
+#define NVGPU_DBG_GPU_IOCTL_CYCLE_STATS	\
+	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 24, struct nvgpu_dbg_gpu_cycle_stats_args)
+
+/* cycle stats snapshot buffer support for mode E */
+struct nvgpu_dbg_gpu_cycle_stats_snapshot_args {
+	__u32 cmd;		/* in: command to handle     */
+	__u32 dmabuf_fd;	/* in: dma buffer handler    */
+	__u32 extra;		/* in/out: extra payload e.g.*/
+				/*    counter/start perfmon  */
+	__u32 reserved;
+};
+
+/* valid commands to control cycle stats shared buffer */
+#define NVGPU_DBG_GPU_IOCTL_CYCLE_STATS_SNAPSHOT_CMD_FLUSH   0
+#define NVGPU_DBG_GPU_IOCTL_CYCLE_STATS_SNAPSHOT_CMD_ATTACH  1
+#define NVGPU_DBG_GPU_IOCTL_CYCLE_STATS_SNAPSHOT_CMD_DETACH  2
+
+#define NVGPU_DBG_GPU_IOCTL_CYCLE_STATS_SNAPSHOT	\
+	_IOWR(NVGPU_DBG_GPU_IOCTL_MAGIC, 25, struct nvgpu_dbg_gpu_cycle_stats_snapshot_args)
+
 #define NVGPU_DBG_GPU_IOCTL_LAST		\
-	_IOC_NR(NVGPU_DBG_GPU_IOCTL_SET_SM_EXCEPTION_TYPE_MASK)
+	_IOC_NR(NVGPU_DBG_GPU_IOCTL_CYCLE_STATS_SNAPSHOT)
 
 #define NVGPU_DBG_GPU_IOCTL_MAX_ARG_SIZE		\
 	sizeof(struct nvgpu_dbg_gpu_access_fb_memory_args)