[PATCH 5/7] af9015 usb bus repeater.

[PATCH 5/7] af9015 usb bus repeater.

[Malcolm Priestley]

This a bus repeater for af9015 devices. Commands usually fail because of other
activity on the usb bus. 

Afatech drivers can repeat up to ten times on the usb bus.

bulk failures that report -ETIMEDOUT or -EBUSY are repeated. If the device fails
it usually return 0x55 in the first byte.

I am working on a patch to move parts of this to the dvb-usb common area to
be used by other drivers.


Signed-off-by: Malcolm Priestley<tvboxspy@gmail.com>

---
 drivers/media/dvb/dvb-usb/af9015.c |   61 ++++++++++++++++++++++--------------
 1 files changed, 37 insertions(+), 24 deletions(-)

diff --git a/drivers/media/dvb/dvb-usb/af9015.c b/drivers/media/dvb/dvb-usb/af9015.c
index 9077ac4..ac134b6 100644
--- a/drivers/media/dvb/dvb-usb/af9015.c
+++ b/drivers/media/dvb/dvb-usb/af9015.c
@@ -68,7 +68,7 @@ static struct af9013_config af9015_af9013_config[] = {
 	}
 };
 
-static int af9015_rw_udev(struct usb_device *udev, struct req_t *req)
+static int af9015_io_udev(struct usb_device *udev, struct req_t *req)
 {
 #define BUF_LEN 63
 #define REQ_HDR_LEN 8 /* send header size */
@@ -79,9 +79,6 @@ static int af9015_rw_udev(struct usb_device *udev, struct req_t *req)
 	u8 msg_len = REQ_HDR_LEN;
 	static u8 seq; /* packet sequence number */
 
-	if (mutex_lock_interruptible(&af9015_usb_mutex) < 0)
-		return -EAGAIN;
-
 	buf[0] = req->cmd;
 	buf[1] = seq++;
 	buf[2] = req->i2c_addr;
@@ -114,16 +111,14 @@ static int af9015_rw_udev(struct usb_device *udev, struct req_t *req)
 		break;
 	default:
 		err("unknown command:%d", req->cmd);
-		ret = -1;
-		goto error_unlock;
+		return -EINVAL;
 	}
 
 	/* buffer overflow check */
 	if ((write && (req->data_len > BUF_LEN - REQ_HDR_LEN)) ||
 		(!write && (req->data_len > BUF_LEN - ACK_HDR_LEN))) {
 		err("too much data; cmd:%d len:%d", req->cmd, req->data_len);
-		ret = -EINVAL;
-		goto error_unlock;
+		return -EINVAL;
 	}
 
 	/* write requested */
@@ -142,13 +137,13 @@ static int af9015_rw_udev(struct usb_device *udev, struct req_t *req)
 		err("bulk message failed:%d (%d/%d)", ret, msg_len, act_len);
 	else
 		if (act_len != msg_len)
-			ret = -1; /* all data is not send */
+			return -EAGAIN; /* all data is not send */
 	if (ret)
-		goto error_unlock;
+		goto error;
 
 	/* no ack for those packets */
 	if (req->cmd == DOWNLOAD_FIRMWARE || req->cmd == RECONNECT_USB)
-		goto exit_unlock;
+		return 0;
 
 	/* write receives seq + status = 2 bytes
 	   read receives seq + status + data = 2 + N bytes */
@@ -158,30 +153,48 @@ static int af9015_rw_udev(struct usb_device *udev, struct req_t *req)
 
 	ret = usb_bulk_msg(udev, usb_rcvbulkpipe(udev, 0x81), buf, msg_len,
 		&act_len, AF9015_USB_TIMEOUT);
-	if (ret) {
-		err("recv bulk message failed:%d", ret);
-		ret = -1;
-		goto error_unlock;
-	}
+	if (ret)
+		goto error;
 
 	deb_xfer("<<< ");
 	debug_dump(buf, act_len, deb_xfer);
 
 	/* check status */
-	if (buf[1]) {
-		err("command failed:%d", buf[1]);
-		ret = -1;
-		goto error_unlock;
-	}
+	if (buf[1])
+		return -EAGAIN;
 
 	/* read request, copy returned data to return buf */
 	if (!write)
 		memcpy(req->data, &buf[ACK_HDR_LEN], req->data_len);
 
-error_unlock:
-exit_unlock:
-	mutex_unlock(&af9015_usb_mutex);
+error:
+	if (ret == -ETIMEDOUT || ret == -EBUSY)
+		return -EAGAIN;
+
+	return ret;
+}
+
+#define AF9015_RETRY 5
+static int af9015_rw_udev(struct usb_device *udev, struct req_t *req)
+{
+	int ret = 0, i;
+
+	if (mutex_lock_interruptible(&af9015_usb_mutex) < 0)
+		return -EAGAIN;
+
+	for (i = 0; i < AF9015_RETRY; i++) {
+		if (req == NULL)
+			break;
+		ret = af9015_io_udev(udev, req);
+		if (ret != -EAGAIN)
+			break;
+		udelay(250);
+	}
+	if (ret && req)
+		err("Command failed:%x i2c addr:%x error:%d",
+			req->cmd, req->i2c_addr, ret);
 
+	mutex_unlock(&af9015_usb_mutex);
 	return ret;
 }
 
-- 
1.7.5.4

Re: [PATCH 5/7] af9015 usb bus repeater.

[Antti Palosaari]

On 11/12/2011 05:56 PM, Malcolm Priestley wrote:
> This a bus repeater for af9015 devices. Commands usually fail because of other
> activity on the usb bus.
>
> Afatech drivers can repeat up to ten times on the usb bus.
>
> bulk failures that report -ETIMEDOUT or -EBUSY are repeated. If the device fails
> it usually return 0x55 in the first byte.
>
> I am working on a patch to move parts of this to the dvb-usb common area to
> be used by other drivers.

Repeating does not help for those I2C errors. I have already tested it. 
IIRC 01 and 02 was the error codes returned in case of I2C read / write.

Which command gives 0x55 for failing?

And generally only very first command will fail. There is already repeat 
for that. Generally I think it is good idea to add some repeating, since 
it is needed. But it does not make much sense to repeat for those common 
I2C errors since it does not matter - you can repeat forever it still 
fails. So first need to fix main error source.

And it is indeed good idea to generalize it, but I think it is possible 
only for error situations when error is returned by platform call.
1. general repeat for platform calls
2. repeat for firmware command fail (is there need?)

regards
Antti

-- 
http://palosaari.fi/