video: tegra: host: use virtual address instead of mem_handle

add below functions which support using virtual addresses
instead of mem_handle
_nvhost_cdma_push_gather()
_push_buffer_push_to()

Bug 1380122
Bug 1380147

Change-Id: I6d33eed4d4d254be7f15f839095ad685351db74e
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/301731
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>

5 files changed, 67 insertions, 1 deletions

diff --git a/drivers/video/tegra/host/chip_support.h b/drivers/video/tegra/host/chip_support.h
index ba26401f6bbb..3f015bac8558 100644
--- a/drivers/video/tegra/host/chip_support.h
+++ b/drivers/video/tegra/host/chip_support.h
@@ -67,6 +67,9 @@ struct nvhost_pushbuffer_ops {
 	void (*push_to)(struct push_buffer *,
 			struct mem_mgr *, struct mem_handle *,
 			u32 op1, u32 op2);
+	void (*_push_to)(struct push_buffer *,
+			dma_addr_t iova,
+			u32 op1, u32 op2);
 	void (*pop_from)(struct push_buffer *,
 			 unsigned int slots);
 	u32 (*space)(struct push_buffer *);
diff --git a/drivers/video/tegra/host/host1x/host1x_cdma.c b/drivers/video/tegra/host/host1x/host1x_cdma.c
index 36d393285b09..4dc6a06f2906 100644
--- a/drivers/video/tegra/host/host1x/host1x_cdma.c
+++ b/drivers/video/tegra/host/host1x/host1x_cdma.c
@@ -141,6 +141,20 @@ static void push_buffer_push_to(struct push_buffer *pb,
 	pb->cur = (cur + 8) & (PUSH_BUFFER_SIZE - 1);
 }
 
+static void _push_buffer_push_to(struct push_buffer *pb,
+				dma_addr_t iova,
+				u32 op1, u32 op2)
+{
+	u32 cur = pb->cur;
+	u32 *p = (u32 *)((uintptr_t)pb->mapped + cur);
+	u32 cur_nvmap = (cur/8) & (NVHOST_GATHER_QUEUE_SIZE - 1);
+	WARN_ON(cur == pb->fence);
+	*(p++) = op1;
+	*(p++) = op2;
+	pb->client_handle[cur_nvmap].iova = iova;
+	pb->cur = (cur + 8) & (PUSH_BUFFER_SIZE - 1);
+}
+
 /**
  * Pop a number of two word slots from the push buffer
  * Caller must ensure push buffer is not empty
@@ -157,6 +171,7 @@ static void push_buffer_pop_from(struct push_buffer *pb,
 		struct mem_mgr_handle *h = &pb->client_handle[cur_fence_nvmap];
 		h->client = NULL;
 		h->handle = NULL;
+		h->iova = 0;
 	}
 	/* Advance the next write position */
 	pb->fence = (pb->fence + slots * 8) & (PUSH_BUFFER_SIZE - 1);
@@ -563,6 +578,7 @@ static const struct nvhost_pushbuffer_ops host1x_pushbuffer_ops = {
 	.init = push_buffer_init,
 	.destroy = push_buffer_destroy,
 	.push_to = push_buffer_push_to,
+	._push_to = _push_buffer_push_to,
 	.pop_from = push_buffer_pop_from,
 	.space = push_buffer_space,
 	.putptr = push_buffer_putptr,
diff --git a/drivers/video/tegra/host/host1x/host1x_hwctx.h b/drivers/video/tegra/host/host1x/host1x_hwctx.h
index 0a32026a1b11..8b379902254a 100644
--- a/drivers/video/tegra/host/host1x/host1x_hwctx.h
+++ b/drivers/video/tegra/host/host1x/host1x_hwctx.h
@@ -3,7 +3,7 @@
  *
  * Tegra Graphics Host HOST1X Hardware Context Interface
  *
- * Copyright (c) 2012-2013, NVIDIA Corporation.
+ * Copyright (c) 2012-2013, NVIDIA Corporation. All rights reserved.
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -43,6 +43,8 @@ struct host1x_hwctx {
 	struct sg_table *restore_sgt;
 	dma_addr_t restore_phys;
 	u32 restore_size;
+	u32 *cpuva;
+	dma_addr_t iova;
 };
 
 struct host1x_hwctx_handler {
diff --git a/drivers/video/tegra/host/nvhost_cdma.c b/drivers/video/tegra/host/nvhost_cdma.c
index 19e8bf13865f..d2adc0e7a83d 100644
--- a/drivers/video/tegra/host/nvhost_cdma.c
+++ b/drivers/video/tegra/host/nvhost_cdma.c
@@ -455,6 +455,27 @@ static void trace_write_gather(struct nvhost_cdma *cdma,
 	}
 }
 
+static void _trace_write_gather(struct nvhost_cdma *cdma,
+		u32 *cpuva, dma_addr_t iova,
+		u32 offset, u32 words)
+{
+	if (iova) {
+		u32 i;
+		/*
+		 * Write in batches of 128 as there seems to be a limit
+		 * of how much you can output to ftrace at once.
+		 */
+		for (i = 0; i < words; i += TRACE_MAX_LENGTH) {
+			trace_nvhost_cdma_push_gather(
+				cdma_to_channel(cdma)->dev->name,
+				(u32)((uintptr_t)iova),
+				min(words - i, TRACE_MAX_LENGTH),
+				offset + i * sizeof(u32),
+				cpuva);
+		}
+	}
+}
+
 /**
  * Push two words into a push buffer slot
  * Blocks as necessary if the push buffer is full.
@@ -492,6 +513,26 @@ void nvhost_cdma_push_gather(struct nvhost_cdma *cdma,
 	cdma_pb_op().push_to(pb, client, handle, op1, op2);
 }
 
+void _nvhost_cdma_push_gather(struct nvhost_cdma *cdma,
+			u32 *cpuva, dma_addr_t iova,
+			u32 offset, u32 op1, u32 op2)
+{
+	u32 slots_free = cdma->slots_free;
+	struct push_buffer *pb = &cdma->push_buffer;
+
+	if (iova)
+		_trace_write_gather(cdma, cpuva, iova, offset, op1 & 0x1fff);
+
+	if (slots_free == 0) {
+		cdma_op().kick(cdma);
+		slots_free = nvhost_cdma_wait_locked(cdma,
+				CDMA_EVENT_PUSH_BUFFER_SPACE);
+	}
+	cdma->slots_free = slots_free - 1;
+	cdma->slots_used++;
+	cdma_pb_op()._push_to(pb, iova, op1, op2);
+}
+
 /**
  * End a cdma submit
  * Kick off DMA, add job to the sync queue, and a number of slots to be freed
diff --git a/drivers/video/tegra/host/nvhost_cdma.h b/drivers/video/tegra/host/nvhost_cdma.h
index 6bf2524d2dc8..5f5951dca933 100644
--- a/drivers/video/tegra/host/nvhost_cdma.h
+++ b/drivers/video/tegra/host/nvhost_cdma.h
@@ -50,6 +50,7 @@ struct mem_handle;
 struct mem_mgr_handle {
 	struct mem_mgr *client;
 	struct mem_handle *handle;
+	dma_addr_t iova;
 };
 
 struct push_buffer {
@@ -111,6 +112,9 @@ void	nvhost_cdma_push(struct nvhost_cdma *cdma, u32 op1, u32 op2);
 void	nvhost_cdma_push_gather(struct nvhost_cdma *cdma,
 		struct mem_mgr *client,
 		struct mem_handle *handle, u32 offset, u32 op1, u32 op2);
+void	_nvhost_cdma_push_gather(struct nvhost_cdma *cdma,
+		u32 *cpuva, dma_addr_t iova,
+		u32 offset, u32 op1, u32 op2);
 void	nvhost_cdma_end(struct nvhost_cdma *cdma,
 		struct nvhost_job *job);
 void	nvhost_cdma_update(struct nvhost_cdma *cdma);