[char-misc-next 00/12] mei: Add DMA ring

[char-misc-next 00/12] mei: Add DMA ring

[Tomas Winkler]

This series adds an alternative method for
transferring data between the mei driver and the device
via a DMA ring. The DMA ring allows transferring
data in bigger chunks, up to 128K, than the HW ring 512B.
The actual sizes depend on particular MEI generations.
The HW ring is faster for packets that fits
into the HW ring while a packet that would require
fragmentation is faster to send via the DMA ring.

Alexander Usyskin (3):
  mei: define dma ring buffer sizes for PCH12 HW and newer
  mei: bump hbm version to 2.1
  mei: me: mark CNP devices as having dma support

Tomas Winkler (9):
  mei: add support for variable length mei headers.
  mei: hbm: define dma ring setup protocol
  mei: hbm: introduce dma bit in the message header
  mei: restrict dma ring support to hbm version 2.1
  mei: dma ring buffers allocation
  mei: hbm: setup dma ring
  mei: hw: add dma ring control block
  mei: dma ring: implement rx circular buffer logic
  mei: dma ring: implement transmit flow

 drivers/misc/mei/Makefile    |   1 +
 drivers/misc/mei/client.c    |  84 +++++++++----
 drivers/misc/mei/debugfs.c   |   2 +
 drivers/misc/mei/dma-ring.c  | 278 +++++++++++++++++++++++++++++++++++++++++++
 drivers/misc/mei/hbm.c       | 126 ++++++++++++++++++--
 drivers/misc/mei/hbm.h       |   2 +
 drivers/misc/mei/hw-me.c     |  53 ++++++---
 drivers/misc/mei/hw-me.h     |   4 +
 drivers/misc/mei/hw-txe.c    |  42 +++----
 drivers/misc/mei/hw.h        |  98 ++++++++++++++-
 drivers/misc/mei/init.c      |   2 +-
 drivers/misc/mei/interrupt.c |  45 ++++---
 drivers/misc/mei/mei_dev.h   |  57 +++++++--
 drivers/misc/mei/pci-me.c    |   4 +-
 14 files changed, 700 insertions(+), 98 deletions(-)
 create mode 100644 drivers/misc/mei/dma-ring.c

-- 
2.14.4

[char-misc-next 01/12] mei: add support for variable length mei headers.

[Tomas Winkler]

Remove header size knowledge from me and txe hw layers,
this requires to change the write handler to accept
header and its length as well as data and its length.

HBM messages are fixed to use basic header, hence we add mei_hbm2slots()
that converts HBM message length and mei message header,
while mei_data2slots() converts data length directly to the slots.

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
---
 drivers/misc/mei/client.c    | 31 ++++++++++++++++++-------------
 drivers/misc/mei/hbm.c       | 30 ++++++++++++++++++++++--------
 drivers/misc/mei/hw-me.c     | 34 ++++++++++++++++++++--------------
 drivers/misc/mei/hw-txe.c    | 42 +++++++++++++++++++++---------------------
 drivers/misc/mei/interrupt.c |  4 ++--
 drivers/misc/mei/mei_dev.h   | 25 +++++++++++++++++++------
 6 files changed, 102 insertions(+), 64 deletions(-)

diff --git a/drivers/misc/mei/client.c b/drivers/misc/mei/client.c
index 207d2f5d5702..0a9173827461 100644
--- a/drivers/misc/mei/client.c
+++ b/drivers/misc/mei/client.c
@@ -863,7 +863,7 @@ int mei_cl_irq_disconnect(struct mei_cl *cl, struct mei_cl_cb *cb,
 	int slots;
 	int ret;
 
-	msg_slots = mei_data2slots(sizeof(struct hbm_client_connect_request));
+	msg_slots = mei_hbm2slots(sizeof(struct hbm_client_connect_request));
 	slots = mei_hbuf_empty_slots(dev);
 	if (slots < 0)
 		return -EOVERFLOW;
@@ -1055,11 +1055,10 @@ int mei_cl_irq_connect(struct mei_cl *cl, struct mei_cl_cb *cb,
 	int slots;
 	int rets;
 
-	msg_slots = mei_data2slots(sizeof(struct hbm_client_connect_request));
-
 	if (mei_cl_is_other_connecting(cl))
 		return 0;
 
+	msg_slots = mei_hbm2slots(sizeof(struct hbm_client_connect_request));
 	slots = mei_hbuf_empty_slots(dev);
 	if (slots < 0)
 		return -EOVERFLOW;
@@ -1299,7 +1298,7 @@ int mei_cl_irq_notify(struct mei_cl *cl, struct mei_cl_cb *cb,
 	int ret;
 	bool request;
 
-	msg_slots = mei_data2slots(sizeof(struct hbm_client_connect_request));
+	msg_slots = mei_hbm2slots(sizeof(struct hbm_client_connect_request));
 	slots = mei_hbuf_empty_slots(dev);
 	if (slots < 0)
 		return -EOVERFLOW;
@@ -1571,6 +1570,7 @@ int mei_cl_irq_write(struct mei_cl *cl, struct mei_cl_cb *cb,
 	struct mei_device *dev;
 	struct mei_msg_data *buf;
 	struct mei_msg_hdr mei_hdr;
+	size_t hdr_len = sizeof(mei_hdr);
 	size_t len;
 	size_t hbuf_len;
 	int hbuf_slots;
@@ -1601,7 +1601,8 @@ int mei_cl_irq_write(struct mei_cl *cl, struct mei_cl_cb *cb,
 		rets = -EOVERFLOW;
 		goto err;
 	}
-	hbuf_len = mei_slots2data(hbuf_slots) - sizeof(struct mei_msg_hdr);
+
+	hbuf_len = mei_slots2data(hbuf_slots);
 
 	mei_msg_hdr_init(&mei_hdr, cb);
 
@@ -1609,11 +1610,11 @@ int mei_cl_irq_write(struct mei_cl *cl, struct mei_cl_cb *cb,
 	 * Split the message only if we can write the whole host buffer
 	 * otherwise wait for next time the host buffer is empty.
 	 */
-	if (hbuf_len >= len) {
+	if (len + hdr_len <= hbuf_len) {
 		mei_hdr.length = len;
 		mei_hdr.msg_complete = 1;
 	} else if ((u32)hbuf_slots == mei_hbuf_depth(dev)) {
-		mei_hdr.length = hbuf_len;
+		mei_hdr.length = hbuf_len - hdr_len;
 	} else {
 		return 0;
 	}
@@ -1621,7 +1622,8 @@ int mei_cl_irq_write(struct mei_cl *cl, struct mei_cl_cb *cb,
 	cl_dbg(dev, cl, "buf: size = %zu idx = %zu\n",
 			cb->buf.size, cb->buf_idx);
 
-	rets = mei_write_message(dev, &mei_hdr, buf->data + cb->buf_idx);
+	rets = mei_write_message(dev, &mei_hdr, hdr_len,
+				 buf->data + cb->buf_idx, mei_hdr.length);
 	if (rets)
 		goto err;
 
@@ -1661,6 +1663,7 @@ ssize_t mei_cl_write(struct mei_cl *cl, struct mei_cl_cb *cb)
 	struct mei_device *dev;
 	struct mei_msg_data *buf;
 	struct mei_msg_hdr mei_hdr;
+	size_t hdr_len = sizeof(mei_hdr);
 	size_t len;
 	size_t hbuf_len;
 	int hbuf_slots;
@@ -1716,15 +1719,17 @@ ssize_t mei_cl_write(struct mei_cl *cl, struct mei_cl_cb *cb)
 		goto out;
 	}
 
-	hbuf_len = mei_slots2data(hbuf_slots) - sizeof(struct mei_msg_hdr);
-	if (hbuf_len >= len) {
+	hbuf_len = mei_slots2data(hbuf_slots);
+
+	if (len + hdr_len <= hbuf_len) {
 		mei_hdr.length = len;
 		mei_hdr.msg_complete = 1;
 	} else {
-		mei_hdr.length = hbuf_len;
+		mei_hdr.length = hbuf_len - hdr_len;
 	}
 
-	rets = mei_write_message(dev, &mei_hdr, buf->data);
+	rets = mei_write_message(dev, &mei_hdr, hdr_len,
+				 buf->data, mei_hdr.length);
 	if (rets)
 		goto err;
 
@@ -1761,7 +1766,7 @@ ssize_t mei_cl_write(struct mei_cl *cl, struct mei_cl_cb *cb)
 		}
 	}
 
-	rets = len;
+	rets = buf->size;
 err:
 	cl_dbg(dev, cl, "rpm: autosuspend\n");
 	pm_runtime_mark_last_busy(dev->dev);
diff --git a/drivers/misc/mei/hbm.c b/drivers/misc/mei/hbm.c
index fe6595fe94f1..8b3fd9ff6566 100644
--- a/drivers/misc/mei/hbm.c
+++ b/drivers/misc/mei/hbm.c
@@ -95,6 +95,20 @@ static int mei_cl_conn_status_to_errno(enum mei_cl_connect_status status)
 	}
 }
 
+/**
+ * mei_hbm_write_message - wrapper for sending hbm messages.
+ *
+ * @dev: mei device
+ * @hdr: mei header
+ * @data: payload
+ */
+static inline int mei_hbm_write_message(struct mei_device *dev,
+					struct mei_msg_hdr *hdr,
+					const void *data)
+{
+	return mei_write_message(dev, hdr, sizeof(*hdr), data, hdr->length);
+}
+
 /**
  * mei_hbm_idle - set hbm to idle state
  *
@@ -174,7 +188,7 @@ static inline int mei_hbm_cl_write(struct mei_device *dev, struct mei_cl *cl,
 	mei_hbm_hdr(&mei_hdr, len);
 	mei_hbm_cl_hdr(cl, hbm_cmd, buf, len);
 
-	return mei_write_message(dev, &mei_hdr, buf);
+	return mei_hbm_write_message(dev, &mei_hdr, buf);
 }
 
 /**
@@ -267,7 +281,7 @@ int mei_hbm_start_req(struct mei_device *dev)
 	start_req.host_version.minor_version = HBM_MINOR_VERSION;
 
 	dev->hbm_state = MEI_HBM_IDLE;
-	ret = mei_write_message(dev, &mei_hdr, &start_req);
+	ret = mei_hbm_write_message(dev, &mei_hdr, &start_req);
 	if (ret) {
 		dev_err(dev->dev, "version message write failed: ret = %d\n",
 			ret);
@@ -304,7 +318,7 @@ static int mei_hbm_enum_clients_req(struct mei_device *dev)
 	enum_req.flags |= dev->hbm_f_ie_supported ?
 			  MEI_HBM_ENUM_F_IMMEDIATE_ENUM : 0;
 
-	ret = mei_write_message(dev, &mei_hdr, &enum_req);
+	ret = mei_hbm_write_message(dev, &mei_hdr, &enum_req);
 	if (ret) {
 		dev_err(dev->dev, "enumeration request write failed: ret = %d.\n",
 			ret);
@@ -373,7 +387,7 @@ static int mei_hbm_add_cl_resp(struct mei_device *dev, u8 addr, u8 status)
 	resp.me_addr = addr;
 	resp.status  = status;
 
-	ret = mei_write_message(dev, &mei_hdr, &resp);
+	ret = mei_hbm_write_message(dev, &mei_hdr, &resp);
 	if (ret)
 		dev_err(dev->dev, "add client response write failed: ret = %d\n",
 			ret);
@@ -430,7 +444,7 @@ int mei_hbm_cl_notify_req(struct mei_device *dev,
 
 	req.start = start;
 
-	ret = mei_write_message(dev, &mei_hdr, &req);
+	ret = mei_hbm_write_message(dev, &mei_hdr, &req);
 	if (ret)
 		dev_err(dev->dev, "notify request failed: ret = %d\n", ret);
 
@@ -555,7 +569,7 @@ static int mei_hbm_prop_req(struct mei_device *dev, unsigned long start_idx)
 	prop_req.hbm_cmd = HOST_CLIENT_PROPERTIES_REQ_CMD;
 	prop_req.me_addr = addr;
 
-	ret = mei_write_message(dev, &mei_hdr, &prop_req);
+	ret = mei_hbm_write_message(dev, &mei_hdr, &prop_req);
 	if (ret) {
 		dev_err(dev->dev, "properties request write failed: ret = %d\n",
 			ret);
@@ -592,7 +606,7 @@ int mei_hbm_pg(struct mei_device *dev, u8 pg_cmd)
 	memset(&req, 0, len);
 	req.hbm_cmd = pg_cmd;
 
-	ret = mei_write_message(dev, &mei_hdr, &req);
+	ret = mei_hbm_write_message(dev, &mei_hdr, &req);
 	if (ret)
 		dev_err(dev->dev, "power gate command write failed.\n");
 	return ret;
@@ -618,7 +632,7 @@ static int mei_hbm_stop_req(struct mei_device *dev)
 	req.hbm_cmd = HOST_STOP_REQ_CMD;
 	req.reason = DRIVER_STOP_REQUEST;
 
-	return mei_write_message(dev, &mei_hdr, &req);
+	return mei_hbm_write_message(dev, &mei_hdr, &req);
 }
 
 /**
diff --git a/drivers/misc/mei/hw-me.c b/drivers/misc/mei/hw-me.c
index c50671cf47eb..0e3c31595dda 100644
--- a/drivers/misc/mei/hw-me.c
+++ b/drivers/misc/mei/hw-me.c
@@ -517,28 +517,31 @@ static u32 mei_me_hbuf_depth(const struct mei_device *dev)
 	return hw->hbuf_depth;
 }
 
-
 /**
  * mei_me_hbuf_write - writes a message to host hw buffer.
  *
  * @dev: the device structure
- * @header: mei HECI header of message
- * @buf: message payload will be written
+ * @hdr: header of message
+ * @hdr_len: header length in bytes: must be multiplication of a slot (4bytes)
+ * @data: payload
+ * @data_len: payload length in bytes
  *
- * Return: -EIO if write has failed
+ * Return: 0 if success, < 0 - otherwise.
  */
 static int mei_me_hbuf_write(struct mei_device *dev,
-			     struct mei_msg_hdr *header,
-			     const unsigned char *buf)
+			     const void *hdr, size_t hdr_len,
+			     const void *data, size_t data_len)
 {
 	unsigned long rem;
-	unsigned long length = header->length;
 	unsigned long i;
-	u32 *reg_buf = (u32 *)buf;
+	const u32 *reg_buf;
 	u32 dw_cnt;
 	int empty_slots;
 
-	dev_dbg(dev->dev, MEI_HDR_FMT, MEI_HDR_PRM(header));
+	if (WARN_ON(!hdr || !data || hdr_len & 0x3))
+		return -EINVAL;
+
+	dev_dbg(dev->dev, MEI_HDR_FMT, MEI_HDR_PRM((struct mei_msg_hdr *)hdr));
 
 	empty_slots = mei_hbuf_empty_slots(dev);
 	dev_dbg(dev->dev, "empty slots = %hu.\n", empty_slots);
@@ -546,20 +549,23 @@ static int mei_me_hbuf_write(struct mei_device *dev,
 	if (empty_slots < 0)
 		return -EOVERFLOW;
 
-	dw_cnt = mei_data2slots(length);
+	dw_cnt = mei_data2slots(hdr_len + data_len);
 	if (dw_cnt > (u32)empty_slots)
 		return -EMSGSIZE;
 
-	mei_me_hcbww_write(dev, *((u32 *) header));
+	reg_buf = hdr;
+	for (i = 0; i < hdr_len / MEI_SLOT_SIZE; i++)
+		mei_me_hcbww_write(dev, reg_buf[i]);
 
-	for (i = 0; i < length / MEI_SLOT_SIZE; i++)
+	reg_buf = data;
+	for (i = 0; i < data_len / MEI_SLOT_SIZE; i++)
 		mei_me_hcbww_write(dev, reg_buf[i]);
 
-	rem = length & 0x3;
+	rem = data_len & 0x3;
 	if (rem > 0) {
 		u32 reg = 0;
 
-		memcpy(&reg, &buf[length - rem], rem);
+		memcpy(&reg, (const u8 *)data + data_len - rem, rem);
 		mei_me_hcbww_write(dev, reg);
 	}
 
diff --git a/drivers/misc/mei/hw-txe.c b/drivers/misc/mei/hw-txe.c
index 7e2026894e9f..8449fe0367ff 100644
--- a/drivers/misc/mei/hw-txe.c
+++ b/drivers/misc/mei/hw-txe.c
@@ -689,37 +689,34 @@ static void mei_txe_hw_config(struct mei_device *dev)
 		hw->aliveness, hw->readiness);
 }
 
-
 /**
  * mei_txe_write - writes a message to device.
  *
  * @dev: the device structure
- * @header: header of message
- * @buf: message buffer will be written
+ * @hdr: header of message
+ * @hdr_len: header length in bytes - must multiplication of a slot (4bytes)
+ * @data: payload
+ * @data_len: paylead length in bytes
  *
- * Return: 0 if success, <0 - otherwise.
+ * Return: 0 if success, < 0 - otherwise.
  */
-
 static int mei_txe_write(struct mei_device *dev,
-			 struct mei_msg_hdr *header,
-			 const unsigned char *buf)
+			 const void *hdr, size_t hdr_len,
+			 const void *data, size_t data_len)
 {
 	struct mei_txe_hw *hw = to_txe_hw(dev);
 	unsigned long rem;
-	unsigned long length;
-	unsigned long i;
+	const u32 *reg_buf;
 	u32 slots = TXE_HBUF_DEPTH;
-	u32 *reg_buf = (u32 *)buf;
 	u32 dw_cnt;
+	unsigned long i, j;
 
-	if (WARN_ON(!header || !buf))
+	if (WARN_ON(!hdr || !data || hdr_len & 0x3))
 		return -EINVAL;
 
-	length = header->length;
-
-	dev_dbg(dev->dev, MEI_HDR_FMT, MEI_HDR_PRM(header));
+	dev_dbg(dev->dev, MEI_HDR_FMT, MEI_HDR_PRM((struct mei_msg_hdr *)hdr));
 
-	dw_cnt = mei_data2slots(length);
+	dw_cnt = mei_data2slots(hdr_len + data_len);
 	if (dw_cnt > slots)
 		return -EMSGSIZE;
 
@@ -737,17 +734,20 @@ static int mei_txe_write(struct mei_device *dev,
 		return -EAGAIN;
 	}
 
-	mei_txe_input_payload_write(dev, 0, *((u32 *)header));
+	reg_buf = hdr;
+	for (i = 0; i < hdr_len / MEI_SLOT_SIZE; i++)
+		mei_txe_input_payload_write(dev, i, reg_buf[i]);
 
-	for (i = 0; i < length / 4; i++)
-		mei_txe_input_payload_write(dev, i + 1, reg_buf[i]);
+	reg_buf = data;
+	for (j = 0; j < data_len / MEI_SLOT_SIZE; j++)
+		mei_txe_input_payload_write(dev, i + j, reg_buf[j]);
 
-	rem = length & 0x3;
+	rem = data_len & 0x3;
 	if (rem > 0) {
 		u32 reg = 0;
 
-		memcpy(&reg, &buf[length - rem], rem);
-		mei_txe_input_payload_write(dev, i + 1, reg);
+		memcpy(&reg, (const u8 *)data + data_len - rem, rem);
+		mei_txe_input_payload_write(dev, i + j, reg);
 	}
 
 	/* after each write the whole buffer is consumed */
diff --git a/drivers/misc/mei/interrupt.c b/drivers/misc/mei/interrupt.c
index 6217cebcad3d..5a661cbdf2ae 100644
--- a/drivers/misc/mei/interrupt.c
+++ b/drivers/misc/mei/interrupt.c
@@ -173,7 +173,7 @@ static int mei_cl_irq_disconnect_rsp(struct mei_cl *cl, struct mei_cl_cb *cb,
 	int slots;
 	int ret;
 
-	msg_slots = mei_data2slots(sizeof(struct hbm_client_connect_response));
+	msg_slots = mei_hbm2slots(sizeof(struct hbm_client_connect_response));
 	slots = mei_hbuf_empty_slots(dev);
 	if (slots < 0)
 		return -EOVERFLOW;
@@ -208,7 +208,7 @@ static int mei_cl_irq_read(struct mei_cl *cl, struct mei_cl_cb *cb,
 	if (!list_empty(&cl->rd_pending))
 		return 0;
 
-	msg_slots = mei_data2slots(sizeof(struct hbm_flow_control));
+	msg_slots = mei_hbm2slots(sizeof(struct hbm_flow_control));
 	slots = mei_hbuf_empty_slots(dev);
 	if (slots < 0)
 		return -EOVERFLOW;
diff --git a/drivers/misc/mei/mei_dev.h b/drivers/misc/mei/mei_dev.h
index e9f37085a628..06fb5fc67fe9 100644
--- a/drivers/misc/mei/mei_dev.h
+++ b/drivers/misc/mei/mei_dev.h
@@ -300,8 +300,8 @@ struct mei_hw_ops {
 	bool (*hbuf_is_ready)(struct mei_device *dev);
 	u32 (*hbuf_depth)(const struct mei_device *dev);
 	int (*write)(struct mei_device *dev,
-		     struct mei_msg_hdr *hdr,
-		     const unsigned char *buf);
+		     const void *hdr, size_t hdr_len,
+		     const void *data, size_t data_len);
 
 	int (*rdbuf_full_slots)(struct mei_device *dev);
 
@@ -528,14 +528,26 @@ static inline unsigned long mei_secs_to_jiffies(unsigned long sec)
 }
 
 /**
- * mei_data2slots - get slots - number of (dwords) from a message length
- *	+ size of the mei header
+ * mei_data2slots - get slots number from a message length
  *
  * @length: size of the messages in bytes
  *
  * Return: number of slots
  */
 static inline u32 mei_data2slots(size_t length)
+{
+	return DIV_ROUND_UP(length, MEI_SLOT_SIZE);
+}
+
+/**
+ * mei_hbm2slots - get slots number from a hbm message length
+ *                 length + size of the mei message header
+ *
+ * @length: size of the messages in bytes
+ *
+ * Return: number of slots
+ */
+static inline u32 mei_hbm2slots(size_t length)
 {
 	return DIV_ROUND_UP(sizeof(struct mei_msg_hdr) + length, MEI_SLOT_SIZE);
 }
@@ -656,9 +668,10 @@ static inline u32 mei_hbuf_depth(const struct mei_device *dev)
 }
 
 static inline int mei_write_message(struct mei_device *dev,
-				    struct mei_msg_hdr *hdr, const void *buf)
+				    const void *hdr, size_t hdr_len,
+				    const void *data, size_t data_len)
 {
-	return dev->ops->write(dev, hdr, buf);
+	return dev->ops->write(dev, hdr, hdr_len, data, data_len);
 }
 
 static inline u32 mei_read_hdr(const struct mei_device *dev)
-- 
2.14.4

[char-misc-next 02/12] mei: hbm: define dma ring setup protocol

[Tomas Winkler]

The protocol defines how to setup an I/O ring on top of host
memory to utilize the device DMA engine for faster transport.

Three memory buffers are allocated.
A Host circular buffer for from the Host to Device communication.
A Device circular buffer for from Device to the Host communication.
And finally a Control block where the pointers for the both
circular buffers are managed.

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
---
 drivers/misc/mei/hw.h | 49 +++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 49 insertions(+)

diff --git a/drivers/misc/mei/hw.h b/drivers/misc/mei/hw.h
index 3b325d955fbe..76fb502e6f71 100644
--- a/drivers/misc/mei/hw.h
+++ b/drivers/misc/mei/hw.h
@@ -122,6 +122,9 @@
 #define MEI_HBM_NOTIFY_RES_CMD              0x90
 #define MEI_HBM_NOTIFICATION_CMD            0x11
 
+#define MEI_HBM_DMA_SETUP_REQ_CMD           0x12
+#define MEI_HBM_DMA_SETUP_RES_CMD           0x92
+
 /*
  * MEI Stop Reason
  * used by hbm_host_stop_request.reason
@@ -449,4 +452,50 @@ struct hbm_notification {
 	u8 reserved[1];
 } __packed;
 
+/**
+ * struct hbm_dma_mem_dscr - dma ring
+ *
+ * @addr_hi: the high 32bits of 64 bit address
+ * @addr_lo: the low  32bits of 64 bit address
+ * @size   : size in bytes (must be power of 2)
+ */
+struct hbm_dma_mem_dscr {
+	u32 addr_hi;
+	u32 addr_lo;
+	u32 size;
+} __packed;
+
+enum {
+	DMA_DSCR_HOST = 0,
+	DMA_DSCR_DEVICE = 1,
+	DMA_DSCR_CTRL = 2,
+	DMA_DSCR_NUM,
+};
+
+/**
+ * struct hbm_dma_setup_request - dma setup request
+ *
+ * @hbm_cmd: bus message command header
+ * @reserved: reserved for alignment
+ * @dma_dscr: dma descriptor for HOST, DEVICE, and CTRL
+ */
+struct hbm_dma_setup_request {
+	u8 hbm_cmd;
+	u8 reserved[3];
+	struct hbm_dma_mem_dscr dma_dscr[DMA_DSCR_NUM];
+} __packed;
+
+/**
+ * struct hbm_dma_setup_response - dma setup response
+ *
+ * @hbm_cmd: bus message command header
+ * @status: 0 on success; otherwise DMA setup failed.
+ * @reserved: reserved for alignment
+ */
+struct hbm_dma_setup_response {
+	u8 hbm_cmd;
+	u8 status;
+	u8 reserved[2];
+} __packed;
+
 #endif
-- 
2.14.4

[char-misc-next 03/12] mei: hbm: introduce dma bit in the message header

[Tomas Winkler]

Add dma_ring bit in the mei message header for conveying
that the message data itself are on the dma ring.

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
---
 drivers/misc/mei/client.c  |  1 +
 drivers/misc/mei/hbm.c     |  1 +
 drivers/misc/mei/hw.h      | 16 ++++++++++++----
 drivers/misc/mei/mei_dev.h |  4 ++--
 4 files changed, 16 insertions(+), 6 deletions(-)

diff --git a/drivers/misc/mei/client.c b/drivers/misc/mei/client.c
index 0a9173827461..4ab6251d418e 100644
--- a/drivers/misc/mei/client.c
+++ b/drivers/misc/mei/client.c
@@ -1551,6 +1551,7 @@ static void mei_msg_hdr_init(struct mei_msg_hdr *mei_hdr, struct mei_cl_cb *cb)
 	mei_hdr->length = 0;
 	mei_hdr->reserved = 0;
 	mei_hdr->msg_complete = 0;
+	mei_hdr->dma_ring = 0;
 	mei_hdr->internal = cb->internal;
 }
 
diff --git a/drivers/misc/mei/hbm.c b/drivers/misc/mei/hbm.c
index 8b3fd9ff6566..b0086c4aefe4 100644
--- a/drivers/misc/mei/hbm.c
+++ b/drivers/misc/mei/hbm.c
@@ -145,6 +145,7 @@ static inline void mei_hbm_hdr(struct mei_msg_hdr *hdr, size_t length)
 	hdr->me_addr = 0;
 	hdr->length = length;
 	hdr->msg_complete = 1;
+	hdr->dma_ring = 0;
 	hdr->reserved = 0;
 	hdr->internal = 0;
 }
diff --git a/drivers/misc/mei/hw.h b/drivers/misc/mei/hw.h
index 76fb502e6f71..5fb8844a6db5 100644
--- a/drivers/misc/mei/hw.h
+++ b/drivers/misc/mei/hw.h
@@ -190,19 +190,27 @@ enum  mei_cl_disconnect_status {
 	MEI_CL_DISCONN_SUCCESS = MEI_HBMS_SUCCESS
 };
 
-/*
- *  MEI BUS Interface Section
+/**
+ * struct mei_msg_hdr - MEI BUS Interface Section
+ *
+ * @me_addr: device address
+ * @host_addr: host address
+ * @length: message length
+ * @reserved: reserved
+ * @dma_ring: message is on dma ring
+ * @internal: message is internal
+ * @msg_complete: last packet of the message
  */
 struct mei_msg_hdr {
 	u32 me_addr:8;
 	u32 host_addr:8;
 	u32 length:9;
-	u32 reserved:5;
+	u32 reserved:4;
+	u32 dma_ring:1;
 	u32 internal:1;
 	u32 msg_complete:1;
 } __packed;
 
-
 struct mei_bus_message {
 	u8 hbm_cmd;
 	u8 data[0];
diff --git a/drivers/misc/mei/mei_dev.h b/drivers/misc/mei/mei_dev.h
index 06fb5fc67fe9..975afaa38a66 100644
--- a/drivers/misc/mei/mei_dev.h
+++ b/drivers/misc/mei/mei_dev.h
@@ -714,10 +714,10 @@ static inline void mei_dbgfs_deregister(struct mei_device *dev) {}
 int mei_register(struct mei_device *dev, struct device *parent);
 void mei_deregister(struct mei_device *dev);
 
-#define MEI_HDR_FMT "hdr:host=%02d me=%02d len=%d internal=%1d comp=%1d"
+#define MEI_HDR_FMT "hdr:host=%02d me=%02d len=%d dma=%1d internal=%1d comp=%1d"
 #define MEI_HDR_PRM(hdr)                  \
 	(hdr)->host_addr, (hdr)->me_addr, \
-	(hdr)->length, (hdr)->internal, (hdr)->msg_complete
+	(hdr)->length, (hdr)->dma_ring, (hdr)->internal, (hdr)->msg_complete
 
 ssize_t mei_fw_status2str(struct mei_fw_status *fw_sts, char *buf, size_t len);
 /**
-- 
2.14.4

[char-misc-next 04/12] mei: restrict dma ring support to hbm version 2.1

[Tomas Winkler]

Only a firmware with version 2.1 and above supports dma ring feature.

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
---
 drivers/misc/mei/debugfs.c | 2 ++
 drivers/misc/mei/hbm.c     | 6 ++++++
 drivers/misc/mei/hw.h      | 6 ++++++
 drivers/misc/mei/mei_dev.h | 2 ++
 4 files changed, 16 insertions(+)

diff --git a/drivers/misc/mei/debugfs.c b/drivers/misc/mei/debugfs.c
index c815da91089c..7b5df8fd6c5a 100644
--- a/drivers/misc/mei/debugfs.c
+++ b/drivers/misc/mei/debugfs.c
@@ -183,6 +183,8 @@ static ssize_t mei_dbgfs_read_devstate(struct file *fp, char __user *ubuf,
 				 dev->hbm_f_fa_supported);
 		pos += scnprintf(buf + pos, bufsz - pos, "\tOS: %01d\n",
 				 dev->hbm_f_os_supported);
+		pos += scnprintf(buf + pos, bufsz - pos, "\tDR: %01d\n",
+				 dev->hbm_f_dr_supported);
 	}
 
 	pos += scnprintf(buf + pos, bufsz - pos, "pg:  %s, %s\n",
diff --git a/drivers/misc/mei/hbm.c b/drivers/misc/mei/hbm.c
index b0086c4aefe4..09e233d4c0de 100644
--- a/drivers/misc/mei/hbm.c
+++ b/drivers/misc/mei/hbm.c
@@ -1007,6 +1007,12 @@ static void mei_hbm_config_features(struct mei_device *dev)
 	/* OS ver message Support */
 	if (dev->version.major_version >= HBM_MAJOR_VERSION_OS)
 		dev->hbm_f_os_supported = 1;
+
+	/* DMA Ring Support */
+	if (dev->version.major_version > HBM_MAJOR_VERSION_DR ||
+	    (dev->version.major_version == HBM_MAJOR_VERSION_DR &&
+	     dev->version.minor_version >= HBM_MINOR_VERSION_DR))
+		dev->hbm_f_dr_supported = 1;
 }
 
 /**
diff --git a/drivers/misc/mei/hw.h b/drivers/misc/mei/hw.h
index 5fb8844a6db5..65655925791a 100644
--- a/drivers/misc/mei/hw.h
+++ b/drivers/misc/mei/hw.h
@@ -80,6 +80,12 @@
 #define HBM_MINOR_VERSION_OS               0
 #define HBM_MAJOR_VERSION_OS               2
 
+/*
+ * MEI version with dma ring support
+ */
+#define HBM_MINOR_VERSION_DR               1
+#define HBM_MAJOR_VERSION_DR               2
+
 /* Host bus message command opcode */
 #define MEI_HBM_CMD_OP_MSK                  0x7f
 /* Host bus message command RESPONSE */
diff --git a/drivers/misc/mei/mei_dev.h b/drivers/misc/mei/mei_dev.h
index 975afaa38a66..377397e1b5a5 100644
--- a/drivers/misc/mei/mei_dev.h
+++ b/drivers/misc/mei/mei_dev.h
@@ -418,6 +418,7 @@ struct mei_fw_version {
  * @hbm_f_fa_supported  : hbm feature fixed address client
  * @hbm_f_ie_supported  : hbm feature immediate reply to enum request
  * @hbm_f_os_supported  : hbm feature support OS ver message
+ * @hbm_f_dr_supported  : hbm feature dma ring supported
  *
  * @fw_ver : FW versions
  *
@@ -495,6 +496,7 @@ struct mei_device {
 	unsigned int hbm_f_fa_supported:1;
 	unsigned int hbm_f_ie_supported:1;
 	unsigned int hbm_f_os_supported:1;
+	unsigned int hbm_f_dr_supported:1;
 
 	struct mei_fw_version fw_ver[MEI_MAX_FW_VER_BLOCKS];
 
-- 
2.14.4

[char-misc-next 05/12] mei: define dma ring buffer sizes for PCH12 HW and newer

[Tomas Winkler]

From: Alexander Usyskin <alexander.usyskin@intel.com>

Define dma ring buffer sizes for PCH12 (CLN HW and newer)

Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
---
 drivers/misc/mei/hw-me.c | 13 +++++++++++++
 drivers/misc/mei/hw-me.h |  4 ++++
 2 files changed, 17 insertions(+)

diff --git a/drivers/misc/mei/hw-me.c b/drivers/misc/mei/hw-me.c
index 0e3c31595dda..0759c3a668de 100644
--- a/drivers/misc/mei/hw-me.c
+++ b/drivers/misc/mei/hw-me.c
@@ -19,6 +19,7 @@
 #include <linux/kthread.h>
 #include <linux/interrupt.h>
 #include <linux/pm_runtime.h>
+#include <linux/sizes.h>
 
 #include "mei_dev.h"
 #include "hbm.h"
@@ -1389,6 +1390,11 @@ static bool mei_me_fw_type_sps(struct pci_dev *pdev)
 	.fw_status.status[4] = PCI_CFG_HFS_5,   \
 	.fw_status.status[5] = PCI_CFG_HFS_6
 
+#define MEI_CFG_DMA_128 \
+	.dma_size[DMA_DSCR_HOST] = SZ_128K, \
+	.dma_size[DMA_DSCR_DEVICE] = SZ_128K, \
+	.dma_size[DMA_DSCR_CTRL] = PAGE_SIZE
+
 /* ICH Legacy devices */
 static const struct mei_cfg mei_me_ich_cfg = {
 	MEI_CFG_ICH_HFS,
@@ -1421,6 +1427,12 @@ static const struct mei_cfg mei_me_pch8_sps_cfg = {
 	MEI_CFG_FW_SPS,
 };
 
+/* Cannon Lake and newer devices */
+static const struct mei_cfg mei_me_pch12_cfg = {
+	MEI_CFG_PCH8_HFS,
+	MEI_CFG_DMA_128,
+};
+
 /*
  * mei_cfg_list - A list of platform platform specific configurations.
  * Note: has to be synchronized with  enum mei_cfg_idx.
@@ -1433,6 +1445,7 @@ static const struct mei_cfg *const mei_cfg_list[] = {
 	[MEI_ME_PCH_CPT_PBG_CFG] = &mei_me_pch_cpt_pbg_cfg,
 	[MEI_ME_PCH8_CFG] = &mei_me_pch8_cfg,
 	[MEI_ME_PCH8_SPS_CFG] = &mei_me_pch8_sps_cfg,
+	[MEI_ME_PCH12_CFG] = &mei_me_pch12_cfg,
 };
 
 const struct mei_cfg *mei_me_get_cfg(kernel_ulong_t idx)
diff --git a/drivers/misc/mei/hw-me.h b/drivers/misc/mei/hw-me.h
index 0c6fe71d1212..bbcc5fc106cd 100644
--- a/drivers/misc/mei/hw-me.h
+++ b/drivers/misc/mei/hw-me.h
@@ -31,10 +31,12 @@
  *
  * @fw_status: FW status
  * @quirk_probe: device exclusion quirk
+ * @dma_size: device DMA buffers size
  */
 struct mei_cfg {
 	const struct mei_fw_status fw_status;
 	bool (*quirk_probe)(struct pci_dev *pdev);
+	size_t dma_size[DMA_DSCR_NUM];
 };
 
 
@@ -80,6 +82,7 @@ struct mei_me_hw {
  * @MEI_ME_PCH8_SPS_CFG:   Platform Controller Hub Gen8 and newer
  *                         servers platforms with quirk for
  *                         SPS firmware exclusion.
+ * @MEI_ME_PCH12_CFG:      Platform Controller Hub Gen12 and newer
  * @MEI_ME_NUM_CFG:        Upper Sentinel.
  */
 enum mei_cfg_idx {
@@ -90,6 +93,7 @@ enum mei_cfg_idx {
 	MEI_ME_PCH_CPT_PBG_CFG,
 	MEI_ME_PCH8_CFG,
 	MEI_ME_PCH8_SPS_CFG,
+	MEI_ME_PCH12_CFG,
 	MEI_ME_NUM_CFG,
 };
 
-- 
2.14.4

[char-misc-next 06/12] mei: dma ring buffers allocation

[Tomas Winkler]

Allocate DMA ring buffers from managed coherent memory.

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
---
 drivers/misc/mei/Makefile   |   1 +
 drivers/misc/mei/dma-ring.c | 103 ++++++++++++++++++++++++++++++++++++++++++++
 drivers/misc/mei/hw-me.c    |   6 +++
 drivers/misc/mei/mei_dev.h  |  20 +++++++++
 4 files changed, 130 insertions(+)
 create mode 100644 drivers/misc/mei/dma-ring.c

diff --git a/drivers/misc/mei/Makefile b/drivers/misc/mei/Makefile
index cd6825afa8e1..d9215fc4e499 100644
--- a/drivers/misc/mei/Makefile
+++ b/drivers/misc/mei/Makefile
@@ -9,6 +9,7 @@ mei-objs += hbm.o
 mei-objs += interrupt.o
 mei-objs += client.o
 mei-objs += main.o
+mei-objs += dma-ring.o
 mei-objs += bus.o
 mei-objs += bus-fixup.o
 mei-$(CONFIG_DEBUG_FS) += debugfs.o
diff --git a/drivers/misc/mei/dma-ring.c b/drivers/misc/mei/dma-ring.c
new file mode 100644
index 000000000000..368012116196
--- /dev/null
+++ b/drivers/misc/mei/dma-ring.c
@@ -0,0 +1,103 @@
+// SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
+/*
+ * Copyright(c) 2016 - 2018 Intel Corporation. All rights reserved.
+ */
+#include <linux/dma-mapping.h>
+#include <linux/mei.h>
+
+#include "mei_dev.h"
+
+/**
+ * mei_dmam_dscr_alloc - allocate a managed coherent buffer
+ *     for the dma descriptor
+ *
+ * @dev: mei_device
+ * @dscr: dma descriptor
+ *
+ * Return: 0 on success or zero allocation request
+ *         -EINVAL if size is not power of 2
+ *         -ENOMEM of allocation has failed
+ */
+static int mei_dmam_dscr_alloc(struct mei_device *dev,
+			       struct mei_dma_dscr *dscr)
+{
+	if (!dscr->size)
+		return 0;
+
+	if (WARN_ON(!is_power_of_2(dscr->size)))
+		return -EINVAL;
+
+	if (dscr->vaddr)
+		return 0;
+
+	dscr->vaddr = dmam_alloc_coherent(dev->dev, dscr->size, &dscr->daddr,
+					  GFP_KERNEL);
+	if (!dscr->vaddr)
+		return -ENOMEM;
+
+	return 0;
+}
+
+/**
+ * mei_dmam_dscr_free - free a managed coherent buffer
+ *     from the dma descriptor
+ *
+ * @dev: mei_device
+ * @dscr: dma descriptor
+ */
+static void mei_dmam_dscr_free(struct mei_device *dev,
+			       struct mei_dma_dscr *dscr)
+{
+	if (!dscr->vaddr)
+		return;
+
+	dmam_free_coherent(dev->dev, dscr->size, dscr->vaddr, dscr->daddr);
+	dscr->vaddr = NULL;
+}
+
+/**
+ * mei_dmam_ring_free - free dma ring buffers
+ *
+ * @dev: mei device
+ */
+void mei_dmam_ring_free(struct mei_device *dev)
+{
+	int i;
+
+	for (i = 0; i < DMA_DSCR_NUM; i++)
+		mei_dmam_dscr_free(dev, &dev->dr_dscr[i]);
+}
+
+/**
+ * mei_dmam_ring_alloc - allocate dma ring buffers
+ *
+ * @dev: mei device
+ *
+ * Return: -ENOMEM on allocation failure 0 otherwise
+ */
+int mei_dmam_ring_alloc(struct mei_device *dev)
+{
+	int i;
+
+	for (i = 0; i < DMA_DSCR_NUM; i++)
+		if (mei_dmam_dscr_alloc(dev, &dev->dr_dscr[i]))
+			goto err;
+
+	return 0;
+
+err:
+	mei_dmam_ring_free(dev);
+	return -ENOMEM;
+}
+
+/**
+ * mei_dma_ring_is_allocated - check if dma ring is allocated
+ *
+ * @dev: mei device
+ *
+ * Return: true if dma ring is allocated
+ */
+bool mei_dma_ring_is_allocated(struct mei_device *dev)
+{
+	return !!dev->dr_dscr[DMA_DSCR_HOST].vaddr;
+}
diff --git a/drivers/misc/mei/hw-me.c b/drivers/misc/mei/hw-me.c
index 0759c3a668de..3fbbadfa2ae1 100644
--- a/drivers/misc/mei/hw-me.c
+++ b/drivers/misc/mei/hw-me.c
@@ -1471,15 +1471,21 @@ struct mei_device *mei_me_dev_init(struct pci_dev *pdev,
 {
 	struct mei_device *dev;
 	struct mei_me_hw *hw;
+	int i;
 
 	dev = devm_kzalloc(&pdev->dev, sizeof(struct mei_device) +
 			   sizeof(struct mei_me_hw), GFP_KERNEL);
 	if (!dev)
 		return NULL;
+
 	hw = to_me_hw(dev);
 
+	for (i = 0; i < DMA_DSCR_NUM; i++)
+		dev->dr_dscr[i].size = cfg->dma_size[i];
+
 	mei_device_init(dev, &pdev->dev, &mei_me_hw_ops);
 	hw->cfg = cfg;
+
 	return dev;
 }
 
diff --git a/drivers/misc/mei/mei_dev.h b/drivers/misc/mei/mei_dev.h
index 377397e1b5a5..a6796e3f712b 100644
--- a/drivers/misc/mei/mei_dev.h
+++ b/drivers/misc/mei/mei_dev.h
@@ -122,6 +122,19 @@ struct mei_msg_data {
 	unsigned char *data;
 };
 
+/**
+ * struct mei_dma_dscr - dma address descriptor
+ *
+ * @vaddr: dma buffer virtual address
+ * @daddr: dma buffer physical address
+ * @size : dma buffer size
+ */
+struct mei_dma_dscr {
+	void *vaddr;
+	dma_addr_t daddr;
+	size_t size;
+};
+
 /* Maximum number of processed FW status registers */
 #define MEI_FW_STATUS_MAX 6
 /* Minimal  buffer for FW status string (8 bytes in dw + space or '\0') */
@@ -409,6 +422,7 @@ struct mei_fw_version {
  * @rd_msg_hdr  : read message header storage
  *
  * @hbuf_is_ready : query if the host host/write buffer is ready
+ * @dr_dscr: DMA ring descriptors: TX, RX, and CTRL
  *
  * @version     : HBM protocol version in use
  * @hbm_f_pg_supported  : hbm feature pgi protocol
@@ -488,6 +502,8 @@ struct mei_device {
 	/* write buffer */
 	bool hbuf_is_ready;
 
+	struct mei_dma_dscr dr_dscr[DMA_DSCR_NUM];
+
 	struct hbm_version version;
 	unsigned int hbm_f_pg_supported:1;
 	unsigned int hbm_f_dc_supported:1;
@@ -578,6 +594,10 @@ int mei_restart(struct mei_device *dev);
 void mei_stop(struct mei_device *dev);
 void mei_cancel_work(struct mei_device *dev);
 
+int mei_dmam_ring_alloc(struct mei_device *dev);
+void mei_dmam_ring_free(struct mei_device *dev);
+bool mei_dma_ring_is_allocated(struct mei_device *dev);
+
 /*
  *  MEI interrupt functions prototype
  */
-- 
2.14.4

[char-misc-next 07/12] mei: hbm: setup dma ring

[Tomas Winkler]

DMA ring is allocated upon HBM handshake and the ring parameters are set
via dedicated HBM_DMA_SETUP request command. The firmware will perform
its setup and respond with a status. On failure the DMA buffers are
released.

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
---
 drivers/misc/mei/hbm.c | 87 ++++++++++++++++++++++++++++++++++++++++++++++++--
 drivers/misc/mei/hbm.h |  2 ++
 2 files changed, 86 insertions(+), 3 deletions(-)

diff --git a/drivers/misc/mei/hbm.c b/drivers/misc/mei/hbm.c
index 09e233d4c0de..a3cdffb9ef62 100644
--- a/drivers/misc/mei/hbm.c
+++ b/drivers/misc/mei/hbm.c
@@ -65,6 +65,7 @@ const char *mei_hbm_state_str(enum mei_hbm_state state)
 	MEI_HBM_STATE(IDLE);
 	MEI_HBM_STATE(STARTING);
 	MEI_HBM_STATE(STARTED);
+	MEI_HBM_STATE(DR_SETUP);
 	MEI_HBM_STATE(ENUM_CLIENTS);
 	MEI_HBM_STATE(CLIENT_PROPERTIES);
 	MEI_HBM_STATE(STOPPED);
@@ -295,6 +296,47 @@ int mei_hbm_start_req(struct mei_device *dev)
 	return 0;
 }
 
+/**
+ * mei_hbm_dma_setup_req - setup DMA request
+ *
+ * @dev: the device structure
+ *
+ * Return: 0 on success and < 0 on failure
+ */
+static int mei_hbm_dma_setup_req(struct mei_device *dev)
+{
+	struct mei_msg_hdr mei_hdr;
+	struct hbm_dma_setup_request req;
+	const size_t len = sizeof(struct hbm_dma_setup_request);
+	unsigned int i;
+	int ret;
+
+	mei_hbm_hdr(&mei_hdr, len);
+
+	memset(&req, 0, len);
+	req.hbm_cmd = MEI_HBM_DMA_SETUP_REQ_CMD;
+	for (i = 0; i < DMA_DSCR_NUM; i++) {
+		phys_addr_t paddr;
+
+		paddr = dev->dr_dscr[i].daddr;
+		req.dma_dscr[i].addr_hi = upper_32_bits(paddr);
+		req.dma_dscr[i].addr_lo = lower_32_bits(paddr);
+		req.dma_dscr[i].size = dev->dr_dscr[i].size;
+	}
+
+	ret = mei_hbm_write_message(dev, &mei_hdr, &req);
+	if (ret) {
+		dev_err(dev->dev, "dma setup request write failed: ret = %d.\n",
+			ret);
+		return ret;
+	}
+
+	dev->hbm_state = MEI_HBM_DR_SETUP;
+	dev->init_clients_timer = MEI_CLIENTS_INIT_TIMEOUT;
+	mei_schedule_stall_timer(dev);
+	return 0;
+}
+
 /**
  * mei_hbm_enum_clients_req - sends enumeration client request message.
  *
@@ -1044,6 +1086,7 @@ int mei_hbm_dispatch(struct mei_device *dev, struct mei_msg_hdr *hdr)
 	struct hbm_host_version_response *version_res;
 	struct hbm_props_response *props_res;
 	struct hbm_host_enum_response *enum_res;
+	struct hbm_dma_setup_response *dma_setup_res;
 	struct hbm_add_client_request *add_cl_req;
 	int ret;
 
@@ -1108,14 +1151,52 @@ int mei_hbm_dispatch(struct mei_device *dev, struct mei_msg_hdr *hdr)
 			return -EPROTO;
 		}
 
-		if (mei_hbm_enum_clients_req(dev)) {
-			dev_err(dev->dev, "hbm: start: failed to send enumeration request\n");
-			return -EIO;
+		if (dev->hbm_f_dr_supported) {
+			if (mei_dmam_ring_alloc(dev))
+				dev_info(dev->dev, "running w/o dma ring\n");
+			if (mei_dma_ring_is_allocated(dev)) {
+				if (mei_hbm_dma_setup_req(dev))
+					return -EIO;
+
+				wake_up(&dev->wait_hbm_start);
+				break;
+			}
 		}
 
+		dev->hbm_f_dr_supported = 0;
+		mei_dmam_ring_free(dev);
+
+		if (mei_hbm_enum_clients_req(dev))
+			return -EIO;
+
 		wake_up(&dev->wait_hbm_start);
 		break;
 
+	case MEI_HBM_DMA_SETUP_RES_CMD:
+		dev_dbg(dev->dev, "hbm: dma setup response: message received.\n");
+
+		dev->init_clients_timer = 0;
+
+		if (dev->hbm_state != MEI_HBM_DR_SETUP) {
+			dev_err(dev->dev, "hbm: dma setup response: state mismatch, [%d, %d]\n",
+				dev->dev_state, dev->hbm_state);
+			return -EPROTO;
+		}
+
+		dma_setup_res = (struct hbm_dma_setup_response *)mei_msg;
+
+		if (dma_setup_res->status) {
+			dev_info(dev->dev, "hbm: dma setup response: failure = %d %s\n",
+				 dma_setup_res->status,
+				 mei_hbm_status_str(dma_setup_res->status));
+			dev->hbm_f_dr_supported = 0;
+			mei_dmam_ring_free(dev);
+		}
+
+		if (mei_hbm_enum_clients_req(dev))
+			return -EIO;
+		break;
+
 	case CLIENT_CONNECT_RES_CMD:
 		dev_dbg(dev->dev, "hbm: client connect response: message received.\n");
 		mei_hbm_cl_res(dev, cl_cmd, MEI_FOP_CONNECT);
diff --git a/drivers/misc/mei/hbm.h b/drivers/misc/mei/hbm.h
index a2025a5083a3..0171a7e79bab 100644
--- a/drivers/misc/mei/hbm.h
+++ b/drivers/misc/mei/hbm.h
@@ -26,6 +26,7 @@ struct mei_cl;
  *
  * @MEI_HBM_IDLE : protocol not started
  * @MEI_HBM_STARTING : start request message was sent
+ * @MEI_HBM_DR_SETUP : dma ring setup request message was sent
  * @MEI_HBM_ENUM_CLIENTS : enumeration request was sent
  * @MEI_HBM_CLIENT_PROPERTIES : acquiring clients properties
  * @MEI_HBM_STARTED : enumeration was completed
@@ -34,6 +35,7 @@ struct mei_cl;
 enum mei_hbm_state {
 	MEI_HBM_IDLE = 0,
 	MEI_HBM_STARTING,
+	MEI_HBM_DR_SETUP,
 	MEI_HBM_ENUM_CLIENTS,
 	MEI_HBM_CLIENT_PROPERTIES,
 	MEI_HBM_STARTED,
-- 
2.14.4

[char-misc-next 08/12] mei: hw: add dma ring control block

[Tomas Winkler]

The DMA ring control block contains write and read
indices for host and device circular buffers.

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
---
 drivers/misc/mei/dma-ring.c | 21 +++++++++++++++++++++
 drivers/misc/mei/hbm.c      |  2 ++
 drivers/misc/mei/hw.h       | 23 +++++++++++++++++++++++
 drivers/misc/mei/mei_dev.h  |  1 +
 4 files changed, 47 insertions(+)

diff --git a/drivers/misc/mei/dma-ring.c b/drivers/misc/mei/dma-ring.c
index 368012116196..70901319534e 100644
--- a/drivers/misc/mei/dma-ring.c
+++ b/drivers/misc/mei/dma-ring.c
@@ -101,3 +101,24 @@ bool mei_dma_ring_is_allocated(struct mei_device *dev)
 {
 	return !!dev->dr_dscr[DMA_DSCR_HOST].vaddr;
 }
+
+static inline
+struct hbm_dma_ring_ctrl *mei_dma_ring_ctrl(struct mei_device *dev)
+{
+	return (struct hbm_dma_ring_ctrl *)dev->dr_dscr[DMA_DSCR_CTRL].vaddr;
+}
+
+/**
+ * mei_dma_ring_reset - reset the dma control block
+ *
+ * @dev: mei device
+ */
+void mei_dma_ring_reset(struct mei_device *dev)
+{
+	struct hbm_dma_ring_ctrl *ctrl = mei_dma_ring_ctrl(dev);
+
+	if (!ctrl)
+		return;
+
+	memset(ctrl, 0, sizeof(*ctrl));
+}
diff --git a/drivers/misc/mei/hbm.c b/drivers/misc/mei/hbm.c
index a3cdffb9ef62..f15d44bda28e 100644
--- a/drivers/misc/mei/hbm.c
+++ b/drivers/misc/mei/hbm.c
@@ -324,6 +324,8 @@ static int mei_hbm_dma_setup_req(struct mei_device *dev)
 		req.dma_dscr[i].size = dev->dr_dscr[i].size;
 	}
 
+	mei_dma_ring_reset(dev);
+
 	ret = mei_hbm_write_message(dev, &mei_hdr, &req);
 	if (ret) {
 		dev_err(dev->dev, "dma setup request write failed: ret = %d.\n",
diff --git a/drivers/misc/mei/hw.h b/drivers/misc/mei/hw.h
index 65655925791a..4f09bbcdbc2a 100644
--- a/drivers/misc/mei/hw.h
+++ b/drivers/misc/mei/hw.h
@@ -512,4 +512,27 @@ struct hbm_dma_setup_response {
 	u8 reserved[2];
 } __packed;
 
+/**
+ * struct mei_dma_ring_ctrl - dma ring control block
+ *
+ * @hbuf_wr_idx: host circular buffer write index in slots
+ * @reserved1: reserved for alignment
+ * @hbuf_rd_idx: host circular buffer read index in slots
+ * @reserved2: reserved for alignment
+ * @dbuf_wr_idx: device circular buffer write index in slots
+ * @reserved3: reserved for alignment
+ * @dbuf_rd_idx: device circular buffer read index in slots
+ * @reserved4: reserved for alignment
+ */
+struct hbm_dma_ring_ctrl {
+	u32 hbuf_wr_idx;
+	u32 reserved1;
+	u32 hbuf_rd_idx;
+	u32 reserved2;
+	u32 dbuf_wr_idx;
+	u32 reserved3;
+	u32 dbuf_rd_idx;
+	u32 reserved4;
+} __packed;
+
 #endif
diff --git a/drivers/misc/mei/mei_dev.h b/drivers/misc/mei/mei_dev.h
index a6796e3f712b..033b5eff8e59 100644
--- a/drivers/misc/mei/mei_dev.h
+++ b/drivers/misc/mei/mei_dev.h
@@ -597,6 +597,7 @@ void mei_cancel_work(struct mei_device *dev);
 int mei_dmam_ring_alloc(struct mei_device *dev);
 void mei_dmam_ring_free(struct mei_device *dev);
 bool mei_dma_ring_is_allocated(struct mei_device *dev);
+void mei_dma_ring_reset(struct mei_device *dev);
 
 /*
  *  MEI interrupt functions prototype
-- 
2.14.4

[char-misc-next 09/12] mei: dma ring: implement rx circular buffer logic

[Tomas Winkler]

Implement circular buffer protocol over receive dma
buffer. Add extension to the mei message header that holds
length of the buffer on the dma buffer.

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
---
 drivers/misc/mei/client.c    |  2 +-
 drivers/misc/mei/dma-ring.c  | 63 ++++++++++++++++++++++++++++++++++++++++++++
 drivers/misc/mei/hw.h        |  4 +++
 drivers/misc/mei/init.c      |  2 +-
 drivers/misc/mei/interrupt.c | 41 +++++++++++++++++++---------
 drivers/misc/mei/mei_dev.h   |  3 ++-
 6 files changed, 100 insertions(+), 15 deletions(-)

diff --git a/drivers/misc/mei/client.c b/drivers/misc/mei/client.c
index 4ab6251d418e..859a5e1469c9 100644
--- a/drivers/misc/mei/client.c
+++ b/drivers/misc/mei/client.c
@@ -478,7 +478,7 @@ struct mei_cl_cb *mei_cl_alloc_cb(struct mei_cl *cl, size_t length,
 	if (length == 0)
 		return cb;
 
-	cb->buf.data = kmalloc(length, GFP_KERNEL);
+	cb->buf.data = kmalloc(roundup(length, MEI_SLOT_SIZE), GFP_KERNEL);
 	if (!cb->buf.data) {
 		mei_io_cb_free(cb);
 		return NULL;
diff --git a/drivers/misc/mei/dma-ring.c b/drivers/misc/mei/dma-ring.c
index 70901319534e..2d2ce6f4ef3b 100644
--- a/drivers/misc/mei/dma-ring.c
+++ b/drivers/misc/mei/dma-ring.c
@@ -122,3 +122,66 @@ void mei_dma_ring_reset(struct mei_device *dev)
 
 	memset(ctrl, 0, sizeof(*ctrl));
 }
+
+/**
+ * mei_dma_ring_reset - copy from dma ring into buffer
+ *
+ * @dev: mei device
+ * @buf: data buffer
+ * @offset: offset in slots.
+ * @n: number of slots to copy.
+ */
+static size_t mei_dma_copy_from(struct mei_device *dev, unsigned char *buf,
+				u32 offset, u32 n)
+{
+	unsigned char *dbuf = dev->dr_dscr[DMA_DSCR_DEVICE].vaddr;
+
+	size_t b_offset = offset << 2;
+	size_t b_n = n << 2;
+
+	memcpy(buf, dbuf + b_offset, b_n);
+
+	return b_n;
+}
+
+/**
+ * mei_dma_ring_read - read data from the ring
+ *
+ * @dev: mei device
+ * @buf: buffer to read into: may be NULL in case of droping the data.
+ * @len: length to read.
+ */
+void mei_dma_ring_read(struct mei_device *dev, unsigned char *buf, u32 len)
+{
+	struct hbm_dma_ring_ctrl *ctrl = mei_dma_ring_ctrl(dev);
+	u32 dbuf_depth;
+	u32 rd_idx, rem, slots;
+
+	if (WARN_ON(!ctrl))
+		return;
+
+	dev_dbg(dev->dev, "reading from dma %u bytes\n", len);
+
+	if (!len)
+		return;
+
+	dbuf_depth = dev->dr_dscr[DMA_DSCR_DEVICE].size >> 2;
+	rd_idx = READ_ONCE(ctrl->dbuf_rd_idx) & (dbuf_depth - 1);
+	slots = mei_data2slots(len);
+
+	/* if buf is NULL we drop the packet by advancing the pointer.*/
+	if (!buf)
+		goto out;
+
+	if (rd_idx + slots > dbuf_depth) {
+		buf += mei_dma_copy_from(dev, buf, rd_idx, dbuf_depth - rd_idx);
+		rem = slots - (dbuf_depth - rd_idx);
+		rd_idx = 0;
+	} else {
+		rem = slots;
+	}
+
+	mei_dma_copy_from(dev, buf, rd_idx, rem);
+out:
+	WRITE_ONCE(ctrl->dbuf_rd_idx, ctrl->dbuf_rd_idx + slots);
+}
diff --git a/drivers/misc/mei/hw.h b/drivers/misc/mei/hw.h
index 4f09bbcdbc2a..acbccb8dba34 100644
--- a/drivers/misc/mei/hw.h
+++ b/drivers/misc/mei/hw.h
@@ -206,6 +206,7 @@ enum  mei_cl_disconnect_status {
  * @dma_ring: message is on dma ring
  * @internal: message is internal
  * @msg_complete: last packet of the message
+ * @extension: extension of the header
  */
 struct mei_msg_hdr {
 	u32 me_addr:8;
@@ -215,8 +216,11 @@ struct mei_msg_hdr {
 	u32 dma_ring:1;
 	u32 internal:1;
 	u32 msg_complete:1;
+	u32 extension[0];
 } __packed;
 
+#define MEI_MSG_HDR_MAX 2
+
 struct mei_bus_message {
 	u8 hbm_cmd;
 	u8 data[0];
diff --git a/drivers/misc/mei/init.c b/drivers/misc/mei/init.c
index 4888ebc076b7..eb026e2a0537 100644
--- a/drivers/misc/mei/init.c
+++ b/drivers/misc/mei/init.c
@@ -151,7 +151,7 @@ int mei_reset(struct mei_device *dev)
 
 	mei_hbm_reset(dev);
 
-	dev->rd_msg_hdr = 0;
+	memset(dev->rd_msg_hdr, 0, sizeof(dev->rd_msg_hdr));
 
 	if (ret) {
 		dev_err(dev->dev, "hw_reset failed ret = %d\n", ret);
diff --git a/drivers/misc/mei/interrupt.c b/drivers/misc/mei/interrupt.c
index 5a661cbdf2ae..055c2d89b310 100644
--- a/drivers/misc/mei/interrupt.c
+++ b/drivers/misc/mei/interrupt.c
@@ -75,6 +75,8 @@ static inline int mei_cl_hbm_equal(struct mei_cl *cl,
  */
 static void mei_irq_discard_msg(struct mei_device *dev, struct mei_msg_hdr *hdr)
 {
+	if (hdr->dma_ring)
+		mei_dma_ring_read(dev, NULL, hdr->extension[0]);
 	/*
 	 * no need to check for size as it is guarantied
 	 * that length fits into rd_msg_buf
@@ -100,6 +102,7 @@ static int mei_cl_irq_read_msg(struct mei_cl *cl,
 	struct mei_device *dev = cl->dev;
 	struct mei_cl_cb *cb;
 	size_t buf_sz;
+	u32 length;
 
 	cb = list_first_entry_or_null(&cl->rd_pending, struct mei_cl_cb, list);
 	if (!cb) {
@@ -119,25 +122,31 @@ static int mei_cl_irq_read_msg(struct mei_cl *cl,
 		goto discard;
 	}
 
-	buf_sz = mei_hdr->length + cb->buf_idx;
+	length = mei_hdr->dma_ring ? mei_hdr->extension[0] : mei_hdr->length;
+
+	buf_sz = length + cb->buf_idx;
 	/* catch for integer overflow */
 	if (buf_sz < cb->buf_idx) {
 		cl_err(dev, cl, "message is too big len %d idx %zu\n",
-		       mei_hdr->length, cb->buf_idx);
+		       length, cb->buf_idx);
 		cb->status = -EMSGSIZE;
 		goto discard;
 	}
 
 	if (cb->buf.size < buf_sz) {
 		cl_dbg(dev, cl, "message overflow. size %zu len %d idx %zu\n",
-			cb->buf.size, mei_hdr->length, cb->buf_idx);
+			cb->buf.size, length, cb->buf_idx);
 		cb->status = -EMSGSIZE;
 		goto discard;
 	}
 
+	if (mei_hdr->dma_ring)
+		mei_dma_ring_read(dev, cb->buf.data + cb->buf_idx, length);
+
+	/*  for DMA read 0 length to generate an interrupt to the device */
 	mei_read_slots(dev, cb->buf.data + cb->buf_idx, mei_hdr->length);
 
-	cb->buf_idx += mei_hdr->length;
+	cb->buf_idx += length;
 
 	if (mei_hdr->msg_complete) {
 		cl_dbg(dev, cl, "completed read length = %zu\n", cb->buf_idx);
@@ -247,6 +256,9 @@ static inline int hdr_is_valid(u32 msg_hdr)
 	if (!msg_hdr || mei_hdr->reserved)
 		return -EBADMSG;
 
+	if (mei_hdr->dma_ring && mei_hdr->length != MEI_SLOT_SIZE)
+		return -EBADMSG;
+
 	return 0;
 }
 
@@ -267,20 +279,20 @@ int mei_irq_read_handler(struct mei_device *dev,
 	struct mei_cl *cl;
 	int ret;
 
-	if (!dev->rd_msg_hdr) {
-		dev->rd_msg_hdr = mei_read_hdr(dev);
+	if (!dev->rd_msg_hdr[0]) {
+		dev->rd_msg_hdr[0] = mei_read_hdr(dev);
 		(*slots)--;
 		dev_dbg(dev->dev, "slots =%08x.\n", *slots);
 
-		ret = hdr_is_valid(dev->rd_msg_hdr);
+		ret = hdr_is_valid(dev->rd_msg_hdr[0]);
 		if (ret) {
 			dev_err(dev->dev, "corrupted message header 0x%08X\n",
-				dev->rd_msg_hdr);
+				dev->rd_msg_hdr[0]);
 			goto end;
 		}
 	}
 
-	mei_hdr = (struct mei_msg_hdr *)&dev->rd_msg_hdr;
+	mei_hdr = (struct mei_msg_hdr *)dev->rd_msg_hdr;
 	dev_dbg(dev->dev, MEI_HDR_FMT, MEI_HDR_PRM(mei_hdr));
 
 	if (mei_slots2data(*slots) < mei_hdr->length) {
@@ -291,6 +303,12 @@ int mei_irq_read_handler(struct mei_device *dev,
 		goto end;
 	}
 
+	if (mei_hdr->dma_ring) {
+		dev->rd_msg_hdr[1] = mei_read_hdr(dev);
+		(*slots)--;
+		mei_hdr->length = 0;
+	}
+
 	/*  HBM message */
 	if (hdr_is_hbm(mei_hdr)) {
 		ret = mei_hbm_dispatch(dev, mei_hdr);
@@ -324,7 +342,7 @@ int mei_irq_read_handler(struct mei_device *dev,
 			goto reset_slots;
 		}
 		dev_err(dev->dev, "no destination client found 0x%08X\n",
-				dev->rd_msg_hdr);
+				dev->rd_msg_hdr[0]);
 		ret = -EBADMSG;
 		goto end;
 	}
@@ -334,9 +352,8 @@ int mei_irq_read_handler(struct mei_device *dev,
 
 reset_slots:
 	/* reset the number of slots and header */
+	memset(dev->rd_msg_hdr, 0, sizeof(dev->rd_msg_hdr));
 	*slots = mei_count_full_read_slots(dev);
-	dev->rd_msg_hdr = 0;
-
 	if (*slots == -EOVERFLOW) {
 		/* overflow - reset */
 		dev_err(dev->dev, "resetting due to slots overflow.\n");
diff --git a/drivers/misc/mei/mei_dev.h b/drivers/misc/mei/mei_dev.h
index 033b5eff8e59..bfd181fbd90c 100644
--- a/drivers/misc/mei/mei_dev.h
+++ b/drivers/misc/mei/mei_dev.h
@@ -497,7 +497,7 @@ struct mei_device {
 #endif /* CONFIG_PM */
 
 	unsigned char rd_msg_buf[MEI_RD_MSG_BUF_SIZE];
-	u32 rd_msg_hdr;
+	u32 rd_msg_hdr[MEI_MSG_HDR_MAX];
 
 	/* write buffer */
 	bool hbuf_is_ready;
@@ -598,6 +598,7 @@ int mei_dmam_ring_alloc(struct mei_device *dev);
 void mei_dmam_ring_free(struct mei_device *dev);
 bool mei_dma_ring_is_allocated(struct mei_device *dev);
 void mei_dma_ring_reset(struct mei_device *dev);
+void mei_dma_ring_read(struct mei_device *dev, unsigned char *buf, u32 len);
 
 /*
  *  MEI interrupt functions prototype
-- 
2.14.4

[char-misc-next 10/12] mei: dma ring: implement transmit flow

[Tomas Winkler]

Implement a circular buffer on allocated system memory. Read and write
indices are stored on the control block which is also shared between the
device and the host.
Two new functions are exported from the DMA module: mei_dma_ring_write,
and mei_dma_ring_empty_slots. The former simply copy a packet on the TX
DMA circular buffer and later, returns the number of empty slots on the
TX DMA circular buffer.

Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
---
 drivers/misc/mei/client.c   | 56 ++++++++++++++++++++++------
 drivers/misc/mei/dma-ring.c | 91 +++++++++++++++++++++++++++++++++++++++++++++
 drivers/misc/mei/mei_dev.h  |  2 +
 3 files changed, 138 insertions(+), 11 deletions(-)

diff --git a/drivers/misc/mei/client.c b/drivers/misc/mei/client.c
index 859a5e1469c9..9dbf2ac1ca4b 100644
--- a/drivers/misc/mei/client.c
+++ b/drivers/misc/mei/client.c
@@ -1573,10 +1573,13 @@ int mei_cl_irq_write(struct mei_cl *cl, struct mei_cl_cb *cb,
 	struct mei_msg_hdr mei_hdr;
 	size_t hdr_len = sizeof(mei_hdr);
 	size_t len;
-	size_t hbuf_len;
+	size_t hbuf_len, dr_len;
 	int hbuf_slots;
+	u32 dr_slots;
+	u32 dma_len;
 	int rets;
 	bool first_chunk;
+	const void *data;
 
 	if (WARN_ON(!cl || !cl->dev))
 		return -ENODEV;
@@ -1597,6 +1600,7 @@ int mei_cl_irq_write(struct mei_cl *cl, struct mei_cl_cb *cb,
 	}
 
 	len = buf->size - cb->buf_idx;
+	data = buf->data + cb->buf_idx;
 	hbuf_slots = mei_hbuf_empty_slots(dev);
 	if (hbuf_slots < 0) {
 		rets = -EOVERFLOW;
@@ -1604,6 +1608,8 @@ int mei_cl_irq_write(struct mei_cl *cl, struct mei_cl_cb *cb,
 	}
 
 	hbuf_len = mei_slots2data(hbuf_slots);
+	dr_slots = mei_dma_ring_empty_slots(dev);
+	dr_len = mei_slots2data(dr_slots);
 
 	mei_msg_hdr_init(&mei_hdr, cb);
 
@@ -1614,23 +1620,32 @@ int mei_cl_irq_write(struct mei_cl *cl, struct mei_cl_cb *cb,
 	if (len + hdr_len <= hbuf_len) {
 		mei_hdr.length = len;
 		mei_hdr.msg_complete = 1;
+	} else if (dr_slots && hbuf_len >= hdr_len + sizeof(dma_len)) {
+		mei_hdr.dma_ring = 1;
+		if (len > dr_len)
+			len = dr_len;
+		else
+			mei_hdr.msg_complete = 1;
+
+		mei_hdr.length = sizeof(dma_len);
+		dma_len = len;
+		data = &dma_len;
 	} else if ((u32)hbuf_slots == mei_hbuf_depth(dev)) {
 		mei_hdr.length = hbuf_len - hdr_len;
 	} else {
 		return 0;
 	}
 
-	cl_dbg(dev, cl, "buf: size = %zu idx = %zu\n",
-			cb->buf.size, cb->buf_idx);
+	if (mei_hdr.dma_ring)
+		mei_dma_ring_write(dev, buf->data + cb->buf_idx, len);
 
-	rets = mei_write_message(dev, &mei_hdr, hdr_len,
-				 buf->data + cb->buf_idx, mei_hdr.length);
+	rets = mei_write_message(dev, &mei_hdr, hdr_len, data, mei_hdr.length);
 	if (rets)
 		goto err;
 
 	cl->status = 0;
 	cl->writing_state = MEI_WRITING;
-	cb->buf_idx += mei_hdr.length;
+	cb->buf_idx += len;
 
 	if (first_chunk) {
 		if (mei_cl_tx_flow_ctrl_creds_reduce(cl)) {
@@ -1665,11 +1680,13 @@ ssize_t mei_cl_write(struct mei_cl *cl, struct mei_cl_cb *cb)
 	struct mei_msg_data *buf;
 	struct mei_msg_hdr mei_hdr;
 	size_t hdr_len = sizeof(mei_hdr);
-	size_t len;
-	size_t hbuf_len;
+	size_t len, hbuf_len, dr_len;
 	int hbuf_slots;
+	u32 dr_slots;
+	u32 dma_len;
 	ssize_t rets;
 	bool blocking;
+	const void *data;
 
 	if (WARN_ON(!cl || !cl->dev))
 		return -ENODEV;
@@ -1681,10 +1698,12 @@ ssize_t mei_cl_write(struct mei_cl *cl, struct mei_cl_cb *cb)
 
 	buf = &cb->buf;
 	len = buf->size;
-	blocking = cb->blocking;
 
 	cl_dbg(dev, cl, "len=%zd\n", len);
 
+	blocking = cb->blocking;
+	data = buf->data;
+
 	rets = pm_runtime_get(dev->dev);
 	if (rets < 0 && rets != -EINPROGRESS) {
 		pm_runtime_put_noidle(dev->dev);
@@ -1721,16 +1740,31 @@ ssize_t mei_cl_write(struct mei_cl *cl, struct mei_cl_cb *cb)
 	}
 
 	hbuf_len = mei_slots2data(hbuf_slots);
+	dr_slots = mei_dma_ring_empty_slots(dev);
+	dr_len =  mei_slots2data(dr_slots);
 
 	if (len + hdr_len <= hbuf_len) {
 		mei_hdr.length = len;
 		mei_hdr.msg_complete = 1;
+	} else if (dr_slots && hbuf_len >= hdr_len + sizeof(dma_len)) {
+		mei_hdr.dma_ring = 1;
+		if (len > dr_len)
+			len = dr_len;
+		else
+			mei_hdr.msg_complete = 1;
+
+		mei_hdr.length = sizeof(dma_len);
+		dma_len = len;
+		data = &dma_len;
 	} else {
 		mei_hdr.length = hbuf_len - hdr_len;
 	}
 
+	if (mei_hdr.dma_ring)
+		mei_dma_ring_write(dev, buf->data, len);
+
 	rets = mei_write_message(dev, &mei_hdr, hdr_len,
-				 buf->data, mei_hdr.length);
+				 data, mei_hdr.length);
 	if (rets)
 		goto err;
 
@@ -1739,7 +1773,7 @@ ssize_t mei_cl_write(struct mei_cl *cl, struct mei_cl_cb *cb)
 		goto err;
 
 	cl->writing_state = MEI_WRITING;
-	cb->buf_idx = mei_hdr.length;
+	cb->buf_idx = len;
 
 out:
 	if (mei_hdr.msg_complete)
diff --git a/drivers/misc/mei/dma-ring.c b/drivers/misc/mei/dma-ring.c
index 2d2ce6f4ef3b..f809db5c0129 100644
--- a/drivers/misc/mei/dma-ring.c
+++ b/drivers/misc/mei/dma-ring.c
@@ -144,6 +144,27 @@ static size_t mei_dma_copy_from(struct mei_device *dev, unsigned char *buf,
 	return b_n;
 }
 
+/**
+ * mei_dma_copy_to - copy to a buffer to the dma ring
+ *
+ * @dev: mei device
+ * @buf: data buffer
+ * @offset: offset in slots.
+ * @n: number of slots to copy.
+ */
+static size_t mei_dma_copy_to(struct mei_device *dev, unsigned char *buf,
+			      u32 offset, u32 n)
+{
+	unsigned char *hbuf = dev->dr_dscr[DMA_DSCR_HOST].vaddr;
+
+	size_t b_offset = offset << 2;
+	size_t b_n = n << 2;
+
+	memcpy(hbuf + b_offset, buf, b_n);
+
+	return b_n;
+}
+
 /**
  * mei_dma_ring_read - read data from the ring
  *
@@ -185,3 +206,73 @@ void mei_dma_ring_read(struct mei_device *dev, unsigned char *buf, u32 len)
 out:
 	WRITE_ONCE(ctrl->dbuf_rd_idx, ctrl->dbuf_rd_idx + slots);
 }
+
+static inline u32 mei_dma_ring_hbuf_depth(struct mei_device *dev)
+{
+	return dev->dr_dscr[DMA_DSCR_HOST].size >> 2;
+}
+
+/**
+ * mei_dma_ring_empty_slots - calaculate number of empty slots in dma ring
+ *
+ * @dev: mei_device
+ *
+ * Return: number of empty slots
+ */
+u32 mei_dma_ring_empty_slots(struct mei_device *dev)
+{
+	struct hbm_dma_ring_ctrl *ctrl = mei_dma_ring_ctrl(dev);
+	u32 wr_idx, rd_idx, hbuf_depth, empty;
+
+	if (!mei_dma_ring_is_allocated(dev))
+		return 0;
+
+	if (WARN_ON(!ctrl))
+		return 0;
+
+	/* easier to work in slots */
+	hbuf_depth = mei_dma_ring_hbuf_depth(dev);
+	rd_idx = READ_ONCE(ctrl->hbuf_rd_idx);
+	wr_idx = READ_ONCE(ctrl->hbuf_wr_idx);
+
+	if (rd_idx > wr_idx)
+		empty = rd_idx - wr_idx;
+	else
+		empty = hbuf_depth - (wr_idx - rd_idx);
+
+	return empty;
+}
+
+/**
+ * mei_dma_ring_write - write data to dma ring host buffer
+ *
+ * @dev: mei_device
+ * @buf: data will be written
+ * @len: data length
+ */
+void mei_dma_ring_write(struct mei_device *dev, unsigned char *buf, u32 len)
+{
+	struct hbm_dma_ring_ctrl *ctrl = mei_dma_ring_ctrl(dev);
+	u32 hbuf_depth;
+	u32 wr_idx, rem, slots;
+
+	if (WARN_ON(!ctrl))
+		return;
+
+	dev_dbg(dev->dev, "writing to dma %u bytes\n", len);
+	hbuf_depth = mei_dma_ring_hbuf_depth(dev);
+	wr_idx = READ_ONCE(ctrl->hbuf_wr_idx) & (hbuf_depth - 1);
+	slots = mei_data2slots(len);
+
+	if (wr_idx + slots > hbuf_depth) {
+		buf += mei_dma_copy_to(dev, buf, wr_idx, hbuf_depth - wr_idx);
+		rem = slots - (hbuf_depth - wr_idx);
+		wr_idx = 0;
+	} else {
+		rem = slots;
+	}
+
+	mei_dma_copy_to(dev, buf, wr_idx, rem);
+
+	WRITE_ONCE(ctrl->hbuf_wr_idx, ctrl->hbuf_wr_idx + slots);
+}
diff --git a/drivers/misc/mei/mei_dev.h b/drivers/misc/mei/mei_dev.h
index bfd181fbd90c..685b78ce30a5 100644
--- a/drivers/misc/mei/mei_dev.h
+++ b/drivers/misc/mei/mei_dev.h
@@ -599,6 +599,8 @@ void mei_dmam_ring_free(struct mei_device *dev);
 bool mei_dma_ring_is_allocated(struct mei_device *dev);
 void mei_dma_ring_reset(struct mei_device *dev);
 void mei_dma_ring_read(struct mei_device *dev, unsigned char *buf, u32 len);
+void mei_dma_ring_write(struct mei_device *dev, unsigned char *buf, u32 len);
+u32 mei_dma_ring_empty_slots(struct mei_device *dev);
 
 /*
  *  MEI interrupt functions prototype
-- 
2.14.4

[char-misc-next 11/12] mei: bump hbm version to 2.1

[Tomas Winkler]

From: Alexander Usyskin <alexander.usyskin@intel.com>

Bump HBM version to 2.1 to indicate DMA transfer support.

Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
---
 drivers/misc/mei/hw.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/misc/mei/hw.h b/drivers/misc/mei/hw.h
index acbccb8dba34..2b7f7677f8cc 100644
--- a/drivers/misc/mei/hw.h
+++ b/drivers/misc/mei/hw.h
@@ -35,7 +35,7 @@
 /*
  * MEI Version
  */
-#define HBM_MINOR_VERSION                   0
+#define HBM_MINOR_VERSION                   1
 #define HBM_MAJOR_VERSION                   2
 
 /*
-- 
2.14.4

[char-misc-next 12/12] mei: me: mark CNP devices as having dma support

[Tomas Winkler]

From: Alexander Usyskin <alexander.usyskin@intel.com>

Signed-off-by: Alexander Usyskin <alexander.usyskin@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
---
 drivers/misc/mei/pci-me.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/misc/mei/pci-me.c b/drivers/misc/mei/pci-me.c
index ea4e152270a3..73ace2d59dea 100644
--- a/drivers/misc/mei/pci-me.c
+++ b/drivers/misc/mei/pci-me.c
@@ -98,9 +98,9 @@ static const struct pci_device_id mei_me_pci_tbl[] = {
 	{MEI_PCI_DEVICE(MEI_DEV_ID_KBP, MEI_ME_PCH8_CFG)},
 	{MEI_PCI_DEVICE(MEI_DEV_ID_KBP_2, MEI_ME_PCH8_CFG)},
 
-	{MEI_PCI_DEVICE(MEI_DEV_ID_CNP_LP, MEI_ME_PCH8_CFG)},
+	{MEI_PCI_DEVICE(MEI_DEV_ID_CNP_LP, MEI_ME_PCH12_CFG)},
 	{MEI_PCI_DEVICE(MEI_DEV_ID_CNP_LP_4, MEI_ME_PCH8_CFG)},
-	{MEI_PCI_DEVICE(MEI_DEV_ID_CNP_H, MEI_ME_PCH8_CFG)},
+	{MEI_PCI_DEVICE(MEI_DEV_ID_CNP_H, MEI_ME_PCH12_CFG)},
 	{MEI_PCI_DEVICE(MEI_DEV_ID_CNP_H_4, MEI_ME_PCH8_CFG)},
 
 	/* required last entry */
-- 
2.14.4

Re: [char-misc-next 06/12] mei: dma ring buffers allocation

[Greg Kroah-Hartman]

On Tue, Jul 31, 2018 at 09:35:38AM +0300, Tomas Winkler wrote:
> --- /dev/null
> +++ b/drivers/misc/mei/dma-ring.c
> @@ -0,0 +1,103 @@
> +// SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0

I thought Intel was not doing this type of crazy nonsense anymore and
just properly creating new files that were GPL-2.0.

Are you _sure_ you want to do this?

> +/*
> + * Copyright(c) 2016 - 2018 Intel Corporation. All rights reserved.

How is a new file copyright 2016?  You have had this file laying around
for 2 years?

thanks,

greg k-h

Re: [char-misc-next 00/12] mei: Add DMA ring

[Greg Kroah-Hartman]

On Tue, Jul 31, 2018 at 09:35:32AM +0300, Tomas Winkler wrote:
> This series adds an alternative method for
> transferring data between the mei driver and the device
> via a DMA ring. The DMA ring allows transferring
> data in bigger chunks, up to 128K, than the HW ring 512B.
> The actual sizes depend on particular MEI generations.
> The HW ring is faster for packets that fits
> into the HW ring while a packet that would require
> fragmentation is faster to send via the DMA ring.

I've applied the first 5 patches now.  Please fix up and resend the
rest.

thanks,

greg k-h

RE: [char-misc-next 06/12] mei: dma ring buffers allocation

[Winkler, Tomas]

> On Tue, Jul 31, 2018 at 09:35:38AM +0300, Tomas Winkler wrote:
> > --- /dev/null
> > +++ b/drivers/misc/mei/dma-ring.c
> > @@ -0,0 +1,103 @@
> > +// SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
> 
> I thought Intel was not doing this type of crazy nonsense anymore and just
> properly creating new files that were GPL-2.0.
> 
> Are you _sure_ you want to do this?
> 
> > +/*
> > + * Copyright(c) 2016 - 2018 Intel Corporation. All rights reserved.
> 
> How is a new file copyright 2016?  You have had this file laying around for 2
> years?
Yes, it takes time till a HW is released to public..
Tomas

RE: [char-misc-next 06/12] mei: dma ring buffers allocation

[Winkler, Tomas]

> Subject: RE: [char-misc-next 06/12] mei: dma ring buffers allocation
> 
> 
> > On Tue, Jul 31, 2018 at 09:35:38AM +0300, Tomas Winkler wrote:
> > > --- /dev/null
> > > +++ b/drivers/misc/mei/dma-ring.c
> > > @@ -0,0 +1,103 @@
> > > +// SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
> >
> > I thought Intel was not doing this type of crazy nonsense anymore and
> > just properly creating new files that were GPL-2.0.
> >
> > Are you _sure_ you want to do this?
I would prefer to keep it that way actually,  hoping eventually relicense most of the files in the driver under the dual license, 
currently it's really kind of mix. 
I'm not sure where this was discussed before, so you've surprised me with this comment.
There is no hidden agenda behind this move, just let the BSD and virtualization world look at the code more freely we have quite a gap there.

Thanks
Tomas

RE: [char-misc-next 00/12] mei: Add DMA ring

[Winkler, Tomas]

> On Tue, Jul 31, 2018 at 09:35:32AM +0300, Tomas Winkler wrote:
> > This series adds an alternative method for transferring data between
> > the mei driver and the device via a DMA ring. The DMA ring allows
> > transferring data in bigger chunks, up to 128K, than the HW ring 512B.
> > The actual sizes depend on particular MEI generations.
> > The HW ring is faster for packets that fits into the HW ring while a
> > packet that would require fragmentation is faster to send via the DMA
> > ring.
> 
> I've applied the first 5 patches now.  Please fix up and resend the rest.

Greg, as I wrote, we wish to stay with the dual license in that file,  please can you apply the patches as they are?

Thanks
Tomas

RE: [char-misc-next 00/12] mei: Add DMA ring

[Winkler, Tomas]

> 
> > On Tue, Jul 31, 2018 at 09:35:32AM +0300, Tomas Winkler wrote:
> > > This series adds an alternative method for transferring data between
> > > the mei driver and the device via a DMA ring. The DMA ring allows
> > > transferring data in bigger chunks, up to 128K, than the HW ring 512B.
> > > The actual sizes depend on particular MEI generations.
> > > The HW ring is faster for packets that fits into the HW ring while a
> > > packet that would require fragmentation is faster to send via the
> > > DMA ring.
> >
> > I've applied the first 5 patches now.  Please fix up and resend the rest.
> 
> Greg, as I wrote, we wish to stay with the dual license in that file,  please can
> you apply the patches as they are?

Hi Greg, can you please respond, we have half of the series merged only, it should be completed.
Thanks
Tomas

Re: [char-misc-next 00/12] mei: Add DMA ring

['Greg Kroah-Hartman']

On Sun, Aug 19, 2018 at 05:34:35AM +0000, Winkler, Tomas wrote:
> 
> > 
> > > On Tue, Jul 31, 2018 at 09:35:32AM +0300, Tomas Winkler wrote:
> > > > This series adds an alternative method for transferring data between
> > > > the mei driver and the device via a DMA ring. The DMA ring allows
> > > > transferring data in bigger chunks, up to 128K, than the HW ring 512B.
> > > > The actual sizes depend on particular MEI generations.
> > > > The HW ring is faster for packets that fits into the HW ring while a
> > > > packet that would require fragmentation is faster to send via the
> > > > DMA ring.
> > >
> > > I've applied the first 5 patches now.  Please fix up and resend the rest.
> > 
> > Greg, as I wrote, we wish to stay with the dual license in that file,  please can
> > you apply the patches as they are?
> 
> Hi Greg, can you please respond, we have half of the series merged only, it should be completed.

It's the middle of the merge window, with major security things
happening at the same time, my ability to be persistant and annoyed at
minor things like this is very low...

Anyway, the patch series is long gone from my queue, you will have to
resend them please.  But first, I want to see an Intel lawyer's
signed-off-by on them when you do resend them as the last time I talked
to them, they said they were not going to be doing dual-licensed kernel
code anymore for a whole bunch of reasons I don't want to get into on a
public mailing list.

So go verify this please, and get their approval before resending.

thanks,

greg k-h

Re: [char-misc-next 00/12] mei: Add DMA ring

[Greg Kroah-Hartman]

On Wed, Aug 08, 2018 at 04:11:56PM +0000, Winkler, Tomas wrote:
> > On Tue, Jul 31, 2018 at 09:35:32AM +0300, Tomas Winkler wrote:
> > > This series adds an alternative method for transferring data between
> > > the mei driver and the device via a DMA ring. The DMA ring allows
> > > transferring data in bigger chunks, up to 128K, than the HW ring 512B.
> > > The actual sizes depend on particular MEI generations.
> > > The HW ring is faster for packets that fits into the HW ring while a
> > > packet that would require fragmentation is faster to send via the DMA
> > > ring.
> > 
> > I've applied the first 5 patches now.  Please fix up and resend the rest.
> 
> Greg, as I wrote, we wish to stay with the dual license in that file,
> please can you apply the patches as they are?

I don't have these anymore in my queue, can you rebase and resend?

thanks,

greg k-h

Re: [char-misc-next 06/12] mei: dma ring buffers allocation

[Greg Kroah-Hartman]

On Sun, Aug 05, 2018 at 08:38:30PM +0000, Winkler, Tomas wrote:
> 
> > Subject: RE: [char-misc-next 06/12] mei: dma ring buffers allocation
> > 
> > 
> > > On Tue, Jul 31, 2018 at 09:35:38AM +0300, Tomas Winkler wrote:
> > > > --- /dev/null
> > > > +++ b/drivers/misc/mei/dma-ring.c
> > > > @@ -0,0 +1,103 @@
> > > > +// SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
> > >
> > > I thought Intel was not doing this type of crazy nonsense anymore and
> > > just properly creating new files that were GPL-2.0.
> > >
> > > Are you _sure_ you want to do this?
> I would prefer to keep it that way actually,  hoping eventually
> relicense most of the files in the driver under the dual license, 
> currently it's really kind of mix. 
> I'm not sure where this was discussed before, so you've surprised me with this comment.

Please go talk to the open source legal team at Intel, this has been
discussed with them a number of times now.  Look at the IB drivers for
one example of how this has changed.

thanks,

greg k-h