drivers/media/pci/saa7164/saa7164-cmd.c - linux - Git at Google

 // SPDX-License-Identifier: GPL-2.0-or-later
 /*
  *  Driver for the NXP SAA7164 PCIe bridge
  *
  *  Copyright (c) 2010-2015 Steven Toth <stoth@kernellabs.com>
  */

 #include <linux/wait.h>

 #include "saa7164.h"

 static int saa7164_cmd_alloc_seqno(struct saa7164_dev *dev)
 {
 	int i, ret = -1;

 	mutex_lock(&dev->lock);
 	for (i = 0; i < SAA_CMD_MAX_MSG_UNITS; i++) {
 		if (dev->cmds[i].inuse == 0) {
 			dev->cmds[i].inuse = 1;
 			dev->cmds[i].signalled = 0;
 			dev->cmds[i].timeout = 0;
 			ret = dev->cmds[i].seqno;
 			break;
 		}
 	}
 	mutex_unlock(&dev->lock);

 	return ret;
 }

 static void saa7164_cmd_free_seqno(struct saa7164_dev *dev, u8 seqno)
 {
 	mutex_lock(&dev->lock);
 	if ((dev->cmds[seqno].inuse == 1) &&
 		(dev->cmds[seqno].seqno == seqno)) {
 		dev->cmds[seqno].inuse = 0;
 		dev->cmds[seqno].signalled = 0;
 		dev->cmds[seqno].timeout = 0;
 	}
 	mutex_unlock(&dev->lock);
 }

 static void saa7164_cmd_timeout_seqno(struct saa7164_dev *dev, u8 seqno)
 {
 	mutex_lock(&dev->lock);
 	if ((dev->cmds[seqno].inuse == 1) &&
 		(dev->cmds[seqno].seqno == seqno)) {
 		dev->cmds[seqno].timeout = 1;
 	}
 	mutex_unlock(&dev->lock);
 }

 static u32 saa7164_cmd_timeout_get(struct saa7164_dev *dev, u8 seqno)
 {
 	int ret = 0;

 	mutex_lock(&dev->lock);
 	if ((dev->cmds[seqno].inuse == 1) &&
 		(dev->cmds[seqno].seqno == seqno)) {
 		ret = dev->cmds[seqno].timeout;
 	}
 	mutex_unlock(&dev->lock);

 	return ret;
 }

 /* Commands to the f/w get marshelled to/from this code then onto the PCI
  * -bus/c running buffer. */
 int saa7164_irq_dequeue(struct saa7164_dev *dev)
 {
 	int ret = SAA_OK, i = 0;
 	u32 timeout;
 	wait_queue_head_t *q = NULL;
 	u8 tmp[512];
 	dprintk(DBGLVL_CMD, "%s()\n", __func__);

 	/* While any outstand message on the bus exists... */
 	do {

 		/* Peek the msg bus */
 		struct tmComResInfo tRsp = { 0, 0, 0, 0, 0, 0 };
 		ret = saa7164_bus_get(dev, &tRsp, NULL, 1);
 		if (ret != SAA_OK)
 			break;

 		q = &dev->cmds[tRsp.seqno].wait;
 		timeout = saa7164_cmd_timeout_get(dev, tRsp.seqno);
 		dprintk(DBGLVL_CMD, "%s() timeout = %d\n", __func__, timeout);
 		if (!timeout) {
 			dprintk(DBGLVL_CMD,
 				"%s() signalled seqno(%d) (for dequeue)\n",
 				__func__, tRsp.seqno);
 			dev->cmds[tRsp.seqno].signalled = 1;
 			wake_up(q);
 		} else {
 			printk(KERN_ERR
 				"%s() found timed out command on the bus\n",
 					__func__);

 			/* Clean the bus */
 			ret = saa7164_bus_get(dev, &tRsp, &tmp, 0);
 			printk(KERN_ERR "%s() ret = %x\n", __func__, ret);
 			if (ret == SAA_ERR_EMPTY)
 				/* Someone else already fetched the response */
 				return SAA_OK;

 			if (ret != SAA_OK)
 				return ret;
 		}

 		/* It's unlikely to have more than 4 or 5 pending messages,
 		 * ensure we exit at some point regardless.
 		 */
 	} while (i++ < 32);

 	return ret;
 }

 /* Commands to the f/w get marshelled to/from this code then onto the PCI
  * -bus/c running buffer. */
 static int saa7164_cmd_dequeue(struct saa7164_dev *dev)
 {
 	int ret;
 	u32 timeout;
 	wait_queue_head_t *q = NULL;
 	u8 tmp[512];
 	dprintk(DBGLVL_CMD, "%s()\n", __func__);

 	while (true) {

 		struct tmComResInfo tRsp = { 0, 0, 0, 0, 0, 0 };
 		ret = saa7164_bus_get(dev, &tRsp, NULL, 1);
 		if (ret == SAA_ERR_EMPTY)
 			return SAA_OK;

 		if (ret != SAA_OK)
 			return ret;

 		q = &dev->cmds[tRsp.seqno].wait;
 		timeout = saa7164_cmd_timeout_get(dev, tRsp.seqno);
 		dprintk(DBGLVL_CMD, "%s() timeout = %d\n", __func__, timeout);
 		if (timeout) {
 			printk(KERN_ERR "found timed out command on the bus\n");

 			/* Clean the bus */
 			ret = saa7164_bus_get(dev, &tRsp, &tmp, 0);
 			printk(KERN_ERR "ret = %x\n", ret);
 			if (ret == SAA_ERR_EMPTY)
 				/* Someone else already fetched the response */
 				return SAA_OK;

 			if (ret != SAA_OK)
 				return ret;

 			if (tRsp.flags & PVC_CMDFLAG_CONTINUE)
 				printk(KERN_ERR "split response\n");
 			else
 				saa7164_cmd_free_seqno(dev, tRsp.seqno);

 			printk(KERN_ERR " timeout continue\n");
 			continue;
 		}

 		dprintk(DBGLVL_CMD, "%s() signalled seqno(%d) (for dequeue)\n",
 			__func__, tRsp.seqno);
 		dev->cmds[tRsp.seqno].signalled = 1;
 		wake_up(q);
 		return SAA_OK;
 	}
 }

 static int saa7164_cmd_set(struct saa7164_dev *dev, struct tmComResInfo *msg,
 			   void *buf)
 {
 	struct tmComResBusInfo *bus = &dev->bus;
 	u8 cmd_sent;
 	u16 size, idx;
 	u32 cmds;
 	void *tmp;
 	int ret = -1;

 	if (!msg) {
 		printk(KERN_ERR "%s() !msg\n", __func__);
 		return SAA_ERR_BAD_PARAMETER;
 	}

 	mutex_lock(&dev->cmds[msg->id].lock);

 	size = msg->size;
 	cmds = size / bus->m_wMaxReqSize;
 	if (size % bus->m_wMaxReqSize == 0)
 		cmds -= 1;

 	cmd_sent = 0;

 	/* Split the request into smaller chunks */
 	for (idx = 0; idx < cmds; idx++) {

 		msg->flags |= SAA_CMDFLAG_CONTINUE;
 		msg->size = bus->m_wMaxReqSize;
 		tmp = buf + idx * bus->m_wMaxReqSize;

 		ret = saa7164_bus_set(dev, msg, tmp);
 		if (ret != SAA_OK) {
 			printk(KERN_ERR "%s() set failed %d\n", __func__, ret);

 			if (cmd_sent) {
 				ret = SAA_ERR_BUSY;
 				goto out;
 			}
 			ret = SAA_ERR_OVERFLOW;
 			goto out;
 		}
 		cmd_sent = 1;
 	}

 	/* If not the last command... */
 	if (idx != 0)
 		msg->flags &= ~SAA_CMDFLAG_CONTINUE;

 	msg->size = size - idx * bus->m_wMaxReqSize;

 	ret = saa7164_bus_set(dev, msg, buf + idx * bus->m_wMaxReqSize);
 	if (ret != SAA_OK) {
 		printk(KERN_ERR "%s() set last failed %d\n", __func__, ret);

 		if (cmd_sent) {
 			ret = SAA_ERR_BUSY;
 			goto out;
 		}
 		ret = SAA_ERR_OVERFLOW;
 		goto out;
 	}
 	ret = SAA_OK;

 out:
 	mutex_unlock(&dev->cmds[msg->id].lock);
 	return ret;
 }

 /* Wait for a signal event, without holding a mutex. Either return TIMEOUT if
  * the event never occurred, or SAA_OK if it was signaled during the wait.
  */
 static int saa7164_cmd_wait(struct saa7164_dev *dev, u8 seqno)
 {
 	wait_queue_head_t *q = NULL;
 	int ret = SAA_BUS_TIMEOUT;
 	unsigned long stamp;
 	int r;

 	if (saa_debug >= 4)
 		saa7164_bus_dump(dev);

 	dprintk(DBGLVL_CMD, "%s(seqno=%d)\n", __func__, seqno);

 	mutex_lock(&dev->lock);
 	if ((dev->cmds[seqno].inuse == 1) &&
 		(dev->cmds[seqno].seqno == seqno)) {
 		q = &dev->cmds[seqno].wait;
 	}
 	mutex_unlock(&dev->lock);

 	if (q) {
 		/* If we haven't been signalled we need to wait */
 		if (dev->cmds[seqno].signalled == 0) {
 			stamp = jiffies;
 			dprintk(DBGLVL_CMD,
 				"%s(seqno=%d) Waiting (signalled=%d)\n",
 				__func__, seqno, dev->cmds[seqno].signalled);

 			/* Wait for signalled to be flagged or timeout */
 			/* In a highly stressed system this can easily extend
 			 * into multiple seconds before the deferred worker
 			 * is scheduled, and we're woken up via signal.
 			 * We typically are signalled in < 50ms but it can
 			 * take MUCH longer.
 			 */
 			wait_event_timeout(*q, dev->cmds[seqno].signalled,
 				(HZ * waitsecs));
 			r = time_before(jiffies, stamp + (HZ * waitsecs));
 			if (r)
 				ret = SAA_OK;
 			else
 				saa7164_cmd_timeout_seqno(dev, seqno);

 			dprintk(DBGLVL_CMD, "%s(seqno=%d) Waiting res = %d (signalled=%d)\n",
 				__func__, seqno, r,
 				dev->cmds[seqno].signalled);
 		} else
 			ret = SAA_OK;
 	} else
 		printk(KERN_ERR "%s(seqno=%d) seqno is invalid\n",
 			__func__, seqno);

 	return ret;
 }

 void saa7164_cmd_signal(struct saa7164_dev *dev, u8 seqno)
 {
 	int i;
 	dprintk(DBGLVL_CMD, "%s()\n", __func__);

 	mutex_lock(&dev->lock);
 	for (i = 0; i < SAA_CMD_MAX_MSG_UNITS; i++) {
 		if (dev->cmds[i].inuse == 1) {
 			dprintk(DBGLVL_CMD,
 				"seqno %d inuse, sig = %d, t/out = %d\n",
 				dev->cmds[i].seqno,
 				dev->cmds[i].signalled,
 				dev->cmds[i].timeout);
 		}
 	}

 	for (i = 0; i < SAA_CMD_MAX_MSG_UNITS; i++) {
 		if ((dev->cmds[i].inuse == 1) && ((i == 0) ||
 			(dev->cmds[i].signalled) || (dev->cmds[i].timeout))) {
 			dprintk(DBGLVL_CMD, "%s(seqno=%d) calling wake_up\n",
 				__func__, i);
 			dev->cmds[i].signalled = 1;
 			wake_up(&dev->cmds[i].wait);
 		}
 	}
 	mutex_unlock(&dev->lock);
 }

 int saa7164_cmd_send(struct saa7164_dev *dev, u8 id, enum tmComResCmd command,
 	u16 controlselector, u16 size, void *buf)
 {
 	struct tmComResInfo command_t, *pcommand_t;
 	struct tmComResInfo response_t, *presponse_t;
 	u8 errdata[256];
 	u16 resp_dsize;
 	u16 data_recd;
 	u32 loop;
 	int ret;
 	int safety = 0;

 	dprintk(DBGLVL_CMD, "%s(unitid = %s (%d) , command = 0x%x, sel = 0x%x)\n",
 		__func__, saa7164_unitid_name(dev, id), id,
 		command, controlselector);

 	if ((size == 0) || (buf == NULL)) {
 		printk(KERN_ERR "%s() Invalid param\n", __func__);
 		return SAA_ERR_BAD_PARAMETER;
 	}

 	/* Prepare some basic command/response structures */
 	memset(&command_t, 0, sizeof(command_t));
 	memset(&response_t, 0, sizeof(response_t));
 	pcommand_t = &command_t;
 	presponse_t = &response_t;
 	command_t.id = id;
 	command_t.command = command;
 	command_t.controlselector = controlselector;
 	command_t.size = size;

 	/* Allocate a unique sequence number */
 	ret = saa7164_cmd_alloc_seqno(dev);
 	if (ret < 0) {
 		printk(KERN_ERR "%s() No free sequences\n", __func__);
 		ret = SAA_ERR_NO_RESOURCES;
 		goto out;
 	}

 	command_t.seqno = (u8)ret;

 	/* Send Command */
 	resp_dsize = size;
 	pcommand_t->size = size;

 	dprintk(DBGLVL_CMD, "%s() pcommand_t.seqno = %d\n",
 		__func__, pcommand_t->seqno);

 	dprintk(DBGLVL_CMD, "%s() pcommand_t.size = %d\n",
 		__func__, pcommand_t->size);

 	ret = saa7164_cmd_set(dev, pcommand_t, buf);
 	if (ret != SAA_OK) {
 		printk(KERN_ERR "%s() set command failed %d\n", __func__, ret);

 		if (ret != SAA_ERR_BUSY)
 			saa7164_cmd_free_seqno(dev, pcommand_t->seqno);
 		else
 			/* Flag a timeout, because at least one
 			 * command was sent */
 			saa7164_cmd_timeout_seqno(dev, pcommand_t->seqno);

 		goto out;
 	}

 	/* With split responses we have to collect the msgs piece by piece */
 	data_recd = 0;
 	loop = 1;
 	while (loop) {
 		dprintk(DBGLVL_CMD, "%s() loop\n", __func__);

 		ret = saa7164_cmd_wait(dev, pcommand_t->seqno);
 		dprintk(DBGLVL_CMD, "%s() loop ret = %d\n", __func__, ret);

 		/* if power is down and this is not a power command ... */

 		if (ret == SAA_BUS_TIMEOUT) {
 			printk(KERN_ERR "Event timed out\n");
 			saa7164_cmd_timeout_seqno(dev, pcommand_t->seqno);
 			return ret;
 		}

 		if (ret != SAA_OK) {
 			printk(KERN_ERR "spurious error\n");
 			return ret;
 		}

 		/* Peek response */
 		ret = saa7164_bus_get(dev, presponse_t, NULL, 1);
 		if (ret == SAA_ERR_EMPTY) {
 			dprintk(4, "%s() SAA_ERR_EMPTY\n", __func__);
 			continue;
 		}
 		if (ret != SAA_OK) {
 			printk(KERN_ERR "peek failed\n");
 			return ret;
 		}

 		dprintk(DBGLVL_CMD, "%s() presponse_t->seqno = %d\n",
 			__func__, presponse_t->seqno);

 		dprintk(DBGLVL_CMD, "%s() presponse_t->flags = 0x%x\n",
 			__func__, presponse_t->flags);

 		dprintk(DBGLVL_CMD, "%s() presponse_t->size = %d\n",
 			__func__, presponse_t->size);

 		/* Check if the response was for our command */
 		if (presponse_t->seqno != pcommand_t->seqno) {

 			dprintk(DBGLVL_CMD,
 				"wrong event: seqno = %d, expected seqno = %d, will dequeue regardless\n",
 				presponse_t->seqno, pcommand_t->seqno);

 			ret = saa7164_cmd_dequeue(dev);
 			if (ret != SAA_OK) {
 				printk(KERN_ERR "dequeue failed, ret = %d\n",
 					ret);
 				if (safety++ > 16) {
 					printk(KERN_ERR
 					"dequeue exceeded, safety exit\n");
 					return SAA_ERR_BUSY;
 				}
 			}

 			continue;
 		}

 		if ((presponse_t->flags & PVC_RESPONSEFLAG_ERROR) != 0) {

 			memset(&errdata[0], 0, sizeof(errdata));

 			ret = saa7164_bus_get(dev, presponse_t, &errdata[0], 0);
 			if (ret != SAA_OK) {
 				printk(KERN_ERR "get error(2)\n");
 				return ret;
 			}

 			saa7164_cmd_free_seqno(dev, pcommand_t->seqno);

 			dprintk(DBGLVL_CMD, "%s() errdata %02x%02x%02x%02x\n",
 				__func__, errdata[0], errdata[1], errdata[2],
 				errdata[3]);

 			/* Map error codes */
 			dprintk(DBGLVL_CMD, "%s() cmd, error code  = 0x%x\n",
 				__func__, errdata[0]);

 			switch (errdata[0]) {
 			case PVC_ERRORCODE_INVALID_COMMAND:
 				dprintk(DBGLVL_CMD, "%s() INVALID_COMMAND\n",
 					__func__);
 				ret = SAA_ERR_INVALID_COMMAND;
 				break;
 			case PVC_ERRORCODE_INVALID_DATA:
 				dprintk(DBGLVL_CMD, "%s() INVALID_DATA\n",
 					__func__);
 				ret = SAA_ERR_BAD_PARAMETER;
 				break;
 			case PVC_ERRORCODE_TIMEOUT:
 				dprintk(DBGLVL_CMD, "%s() TIMEOUT\n", __func__);
 				ret = SAA_ERR_TIMEOUT;
 				break;
 			case PVC_ERRORCODE_NAK:
 				dprintk(DBGLVL_CMD, "%s() NAK\n", __func__);
 				ret = SAA_ERR_NULL_PACKET;
 				break;
 			case PVC_ERRORCODE_UNKNOWN:
 			case PVC_ERRORCODE_INVALID_CONTROL:
 				dprintk(DBGLVL_CMD,
 					"%s() UNKNOWN OR INVALID CONTROL\n",
 					__func__);
 				ret = SAA_ERR_NOT_SUPPORTED;
 				break;
 			default:
 				dprintk(DBGLVL_CMD, "%s() UNKNOWN\n", __func__);
 				ret = SAA_ERR_NOT_SUPPORTED;
 			}

 			/* See of other commands are on the bus */
 			if (saa7164_cmd_dequeue(dev) != SAA_OK)
 				printk(KERN_ERR "dequeue(2) failed\n");

 			return ret;
 		}

 		/* If response is invalid */
 		if ((presponse_t->id != pcommand_t->id) ||
 			(presponse_t->command != pcommand_t->command) ||
 			(presponse_t->controlselector !=
 				pcommand_t->controlselector) ||
 			(((resp_dsize - data_recd) != presponse_t->size) &&
 				!(presponse_t->flags & PVC_CMDFLAG_CONTINUE)) ||
 			((resp_dsize - data_recd) < presponse_t->size)) {

 			/* Invalid */
 			dprintk(DBGLVL_CMD, "%s() Invalid\n", __func__);
 			ret = saa7164_bus_get(dev, presponse_t, NULL, 0);
 			if (ret != SAA_OK) {
 				printk(KERN_ERR "get failed\n");
 				return ret;
 			}

 			/* See of other commands are on the bus */
 			if (saa7164_cmd_dequeue(dev) != SAA_OK)
 				printk(KERN_ERR "dequeue(3) failed\n");
 			continue;
 		}

 		/* OK, now we're actually getting out correct response */
 		ret = saa7164_bus_get(dev, presponse_t, buf + data_recd, 0);
 		if (ret != SAA_OK) {
 			printk(KERN_ERR "get failed\n");
 			return ret;
 		}

 		data_recd = presponse_t->size + data_recd;
 		if (resp_dsize == data_recd) {
 			dprintk(DBGLVL_CMD, "%s() Resp recd\n", __func__);
 			break;
 		}

 		/* See of other commands are on the bus */
 		if (saa7164_cmd_dequeue(dev) != SAA_OK)
 			printk(KERN_ERR "dequeue(3) failed\n");
 	} /* (loop) */

 	/* Release the sequence number allocation */
 	saa7164_cmd_free_seqno(dev, pcommand_t->seqno);

 	/* if powerdown signal all pending commands */

 	dprintk(DBGLVL_CMD, "%s() Calling dequeue then exit\n", __func__);

 	/* See of other commands are on the bus */
 	if (saa7164_cmd_dequeue(dev) != SAA_OK)
 		printk(KERN_ERR "dequeue(4) failed\n");

 	ret = SAA_OK;
 out:
 	return ret;
 }
	// SPDX-License-Identifier: GPL-2.0-or-later
	/*
	* Driver for the NXP SAA7164 PCIe bridge
	*
	* Copyright (c) 2010-2015 Steven Toth <stoth@kernellabs.com>
	*/

	#include <linux/wait.h>

	#include "saa7164.h"

	static int saa7164_cmd_alloc_seqno(struct saa7164_dev *dev)
	{
	int i, ret = -1;

	mutex_lock(&dev->lock);
	for (i = 0; i < SAA_CMD_MAX_MSG_UNITS; i++) {
	if (dev->cmds[i].inuse == 0) {
	dev->cmds[i].inuse = 1;
	dev->cmds[i].signalled = 0;
	dev->cmds[i].timeout = 0;
	ret = dev->cmds[i].seqno;
	break;
	}
	}
	mutex_unlock(&dev->lock);

	return ret;
	}

	static void saa7164_cmd_free_seqno(struct saa7164_dev *dev, u8 seqno)
	{
	mutex_lock(&dev->lock);
	if ((dev->cmds[seqno].inuse == 1) &&
	(dev->cmds[seqno].seqno == seqno)) {
	dev->cmds[seqno].inuse = 0;
	dev->cmds[seqno].signalled = 0;
	dev->cmds[seqno].timeout = 0;
	}
	mutex_unlock(&dev->lock);
	}

	static void saa7164_cmd_timeout_seqno(struct saa7164_dev *dev, u8 seqno)
	{
	mutex_lock(&dev->lock);
	if ((dev->cmds[seqno].inuse == 1) &&
	(dev->cmds[seqno].seqno == seqno)) {
	dev->cmds[seqno].timeout = 1;
	}
	mutex_unlock(&dev->lock);
	}

	static u32 saa7164_cmd_timeout_get(struct saa7164_dev *dev, u8 seqno)
	{
	int ret = 0;

	mutex_lock(&dev->lock);
	if ((dev->cmds[seqno].inuse == 1) &&
	(dev->cmds[seqno].seqno == seqno)) {
	ret = dev->cmds[seqno].timeout;
	}
	mutex_unlock(&dev->lock);

	return ret;
	}

	/* Commands to the f/w get marshelled to/from this code then onto the PCI
	* -bus/c running buffer. */
	int saa7164_irq_dequeue(struct saa7164_dev *dev)
	{
	int ret = SAA_OK, i = 0;
	u32 timeout;
	wait_queue_head_t *q = NULL;
	u8 tmp[512];
	dprintk(DBGLVL_CMD, "%s()\n", __func__);

	/* While any outstand message on the bus exists... */
	do {

	/* Peek the msg bus */
	struct tmComResInfo tRsp = { 0, 0, 0, 0, 0, 0 };
	ret = saa7164_bus_get(dev, &tRsp, NULL, 1);
	if (ret != SAA_OK)
	break;

	q = &dev->cmds[tRsp.seqno].wait;
	timeout = saa7164_cmd_timeout_get(dev, tRsp.seqno);
	dprintk(DBGLVL_CMD, "%s() timeout = %d\n", __func__, timeout);
	if (!timeout) {
	dprintk(DBGLVL_CMD,
	"%s() signalled seqno(%d) (for dequeue)\n",
	__func__, tRsp.seqno);
	dev->cmds[tRsp.seqno].signalled = 1;
	wake_up(q);
	} else {
	printk(KERN_ERR
	"%s() found timed out command on the bus\n",
	__func__);

	/* Clean the bus */
	ret = saa7164_bus_get(dev, &tRsp, &tmp, 0);
	printk(KERN_ERR "%s() ret = %x\n", __func__, ret);
	if (ret == SAA_ERR_EMPTY)
	/* Someone else already fetched the response */
	return SAA_OK;

	if (ret != SAA_OK)
	return ret;
	}

	/* It's unlikely to have more than 4 or 5 pending messages,
	* ensure we exit at some point regardless.
	*/
	} while (i++ < 32);

	return ret;
	}

	/* Commands to the f/w get marshelled to/from this code then onto the PCI
	* -bus/c running buffer. */
	static int saa7164_cmd_dequeue(struct saa7164_dev *dev)
	{
	int ret;
	u32 timeout;
	wait_queue_head_t *q = NULL;
	u8 tmp[512];
	dprintk(DBGLVL_CMD, "%s()\n", __func__);

	while (true) {

	struct tmComResInfo tRsp = { 0, 0, 0, 0, 0, 0 };
	ret = saa7164_bus_get(dev, &tRsp, NULL, 1);
	if (ret == SAA_ERR_EMPTY)
	return SAA_OK;

	if (ret != SAA_OK)
	return ret;

	q = &dev->cmds[tRsp.seqno].wait;
	timeout = saa7164_cmd_timeout_get(dev, tRsp.seqno);
	dprintk(DBGLVL_CMD, "%s() timeout = %d\n", __func__, timeout);
	if (timeout) {
	printk(KERN_ERR "found timed out command on the bus\n");

	/* Clean the bus */
	ret = saa7164_bus_get(dev, &tRsp, &tmp, 0);
	printk(KERN_ERR "ret = %x\n", ret);
	if (ret == SAA_ERR_EMPTY)
	/* Someone else already fetched the response */
	return SAA_OK;

	if (ret != SAA_OK)
	return ret;

	if (tRsp.flags & PVC_CMDFLAG_CONTINUE)
	printk(KERN_ERR "split response\n");
	else
	saa7164_cmd_free_seqno(dev, tRsp.seqno);

	printk(KERN_ERR " timeout continue\n");
	continue;
	}

	dprintk(DBGLVL_CMD, "%s() signalled seqno(%d) (for dequeue)\n",
	__func__, tRsp.seqno);
	dev->cmds[tRsp.seqno].signalled = 1;
	wake_up(q);
	return SAA_OK;
	}
	}

	static int saa7164_cmd_set(struct saa7164_dev dev, struct tmComResInfo msg,
	void *buf)
	{
	struct tmComResBusInfo *bus = &dev->bus;
	u8 cmd_sent;
	u16 size, idx;
	u32 cmds;
	void *tmp;
	int ret = -1;

	if (!msg) {
	printk(KERN_ERR "%s() !msg\n", __func__);
	return SAA_ERR_BAD_PARAMETER;
	}

	mutex_lock(&dev->cmds[msg->id].lock);

	size = msg->size;
	cmds = size / bus->m_wMaxReqSize;
	if (size % bus->m_wMaxReqSize == 0)
	cmds -= 1;

	cmd_sent = 0;

	/* Split the request into smaller chunks */
	for (idx = 0; idx < cmds; idx++) {

	msg->flags \|= SAA_CMDFLAG_CONTINUE;
	msg->size = bus->m_wMaxReqSize;
	tmp = buf + idx * bus->m_wMaxReqSize;

	ret = saa7164_bus_set(dev, msg, tmp);
	if (ret != SAA_OK) {
	printk(KERN_ERR "%s() set failed %d\n", __func__, ret);

	if (cmd_sent) {
	ret = SAA_ERR_BUSY;
	goto out;
	}
	ret = SAA_ERR_OVERFLOW;
	goto out;
	}
	cmd_sent = 1;
	}

	/* If not the last command... */
	if (idx != 0)
	msg->flags &= ~SAA_CMDFLAG_CONTINUE;

	msg->size = size - idx * bus->m_wMaxReqSize;

	ret = saa7164_bus_set(dev, msg, buf + idx * bus->m_wMaxReqSize);
	if (ret != SAA_OK) {
	printk(KERN_ERR "%s() set last failed %d\n", __func__, ret);

	if (cmd_sent) {
	ret = SAA_ERR_BUSY;
	goto out;
	}
	ret = SAA_ERR_OVERFLOW;
	goto out;
	}
	ret = SAA_OK;

	out:
	mutex_unlock(&dev->cmds[msg->id].lock);
	return ret;
	}

	/* Wait for a signal event, without holding a mutex. Either return TIMEOUT if
	* the event never occurred, or SAA_OK if it was signaled during the wait.
	*/
	static int saa7164_cmd_wait(struct saa7164_dev *dev, u8 seqno)
	{
	wait_queue_head_t *q = NULL;
	int ret = SAA_BUS_TIMEOUT;
	unsigned long stamp;
	int r;

	if (saa_debug >= 4)
	saa7164_bus_dump(dev);

	dprintk(DBGLVL_CMD, "%s(seqno=%d)\n", __func__, seqno);

	mutex_lock(&dev->lock);
	if ((dev->cmds[seqno].inuse == 1) &&
	(dev->cmds[seqno].seqno == seqno)) {
	q = &dev->cmds[seqno].wait;
	}
	mutex_unlock(&dev->lock);

	if (q) {
	/* If we haven't been signalled we need to wait */
	if (dev->cmds[seqno].signalled == 0) {
	stamp = jiffies;
	dprintk(DBGLVL_CMD,
	"%s(seqno=%d) Waiting (signalled=%d)\n",
	__func__, seqno, dev->cmds[seqno].signalled);

	/* Wait for signalled to be flagged or timeout */
	/* In a highly stressed system this can easily extend
	* into multiple seconds before the deferred worker
	* is scheduled, and we're woken up via signal.
	* We typically are signalled in < 50ms but it can
	* take MUCH longer.
	*/
	wait_event_timeout(*q, dev->cmds[seqno].signalled,
	(HZ * waitsecs));
	r = time_before(jiffies, stamp + (HZ * waitsecs));
	if (r)
	ret = SAA_OK;
	else
	saa7164_cmd_timeout_seqno(dev, seqno);

	dprintk(DBGLVL_CMD, "%s(seqno=%d) Waiting res = %d (signalled=%d)\n",
	__func__, seqno, r,
	dev->cmds[seqno].signalled);
	} else
	ret = SAA_OK;
	} else
	printk(KERN_ERR "%s(seqno=%d) seqno is invalid\n",
	__func__, seqno);

	return ret;
	}

	void saa7164_cmd_signal(struct saa7164_dev *dev, u8 seqno)
	{
	int i;
	dprintk(DBGLVL_CMD, "%s()\n", __func__);

	mutex_lock(&dev->lock);
	for (i = 0; i < SAA_CMD_MAX_MSG_UNITS; i++) {
	if (dev->cmds[i].inuse == 1) {
	dprintk(DBGLVL_CMD,
	"seqno %d inuse, sig = %d, t/out = %d\n",
	dev->cmds[i].seqno,
	dev->cmds[i].signalled,
	dev->cmds[i].timeout);
	}
	}

	for (i = 0; i < SAA_CMD_MAX_MSG_UNITS; i++) {
	if ((dev->cmds[i].inuse == 1) && ((i == 0) \|\|
	(dev->cmds[i].signalled) \|\| (dev->cmds[i].timeout))) {
	dprintk(DBGLVL_CMD, "%s(seqno=%d) calling wake_up\n",
	__func__, i);
	dev->cmds[i].signalled = 1;
	wake_up(&dev->cmds[i].wait);
	}
	}
	mutex_unlock(&dev->lock);
	}

	int saa7164_cmd_send(struct saa7164_dev *dev, u8 id, enum tmComResCmd command,
	u16 controlselector, u16 size, void *buf)
	{
	struct tmComResInfo command_t, *pcommand_t;
	struct tmComResInfo response_t, *presponse_t;
	u8 errdata[256];
	u16 resp_dsize;
	u16 data_recd;
	u32 loop;
	int ret;
	int safety = 0;

	dprintk(DBGLVL_CMD, "%s(unitid = %s (%d) , command = 0x%x, sel = 0x%x)\n",
	__func__, saa7164_unitid_name(dev, id), id,
	command, controlselector);

	if ((size == 0) \|\| (buf == NULL)) {
	printk(KERN_ERR "%s() Invalid param\n", __func__);
	return SAA_ERR_BAD_PARAMETER;
	}

	/* Prepare some basic command/response structures */
	memset(&command_t, 0, sizeof(command_t));
	memset(&response_t, 0, sizeof(response_t));
	pcommand_t = &command_t;
	presponse_t = &response_t;
	command_t.id = id;
	command_t.command = command;
	command_t.controlselector = controlselector;
	command_t.size = size;

	/* Allocate a unique sequence number */
	ret = saa7164_cmd_alloc_seqno(dev);
	if (ret < 0) {
	printk(KERN_ERR "%s() No free sequences\n", __func__);
	ret = SAA_ERR_NO_RESOURCES;
	goto out;
	}

	command_t.seqno = (u8)ret;

	/* Send Command */
	resp_dsize = size;
	pcommand_t->size = size;

	dprintk(DBGLVL_CMD, "%s() pcommand_t.seqno = %d\n",
	__func__, pcommand_t->seqno);

	dprintk(DBGLVL_CMD, "%s() pcommand_t.size = %d\n",
	__func__, pcommand_t->size);

	ret = saa7164_cmd_set(dev, pcommand_t, buf);
	if (ret != SAA_OK) {
	printk(KERN_ERR "%s() set command failed %d\n", __func__, ret);

	if (ret != SAA_ERR_BUSY)
	saa7164_cmd_free_seqno(dev, pcommand_t->seqno);
	else
	/* Flag a timeout, because at least one
	* command was sent */
	saa7164_cmd_timeout_seqno(dev, pcommand_t->seqno);

	goto out;
	}

	/* With split responses we have to collect the msgs piece by piece */
	data_recd = 0;
	loop = 1;
	while (loop) {
	dprintk(DBGLVL_CMD, "%s() loop\n", __func__);

	ret = saa7164_cmd_wait(dev, pcommand_t->seqno);
	dprintk(DBGLVL_CMD, "%s() loop ret = %d\n", __func__, ret);

	/* if power is down and this is not a power command ... */

	if (ret == SAA_BUS_TIMEOUT) {
	printk(KERN_ERR "Event timed out\n");
	saa7164_cmd_timeout_seqno(dev, pcommand_t->seqno);
	return ret;
	}

	if (ret != SAA_OK) {
	printk(KERN_ERR "spurious error\n");
	return ret;
	}

	/* Peek response */
	ret = saa7164_bus_get(dev, presponse_t, NULL, 1);
	if (ret == SAA_ERR_EMPTY) {
	dprintk(4, "%s() SAA_ERR_EMPTY\n", __func__);
	continue;
	}
	if (ret != SAA_OK) {
	printk(KERN_ERR "peek failed\n");
	return ret;
	}

	dprintk(DBGLVL_CMD, "%s() presponse_t->seqno = %d\n",
	__func__, presponse_t->seqno);

	dprintk(DBGLVL_CMD, "%s() presponse_t->flags = 0x%x\n",
	__func__, presponse_t->flags);

	dprintk(DBGLVL_CMD, "%s() presponse_t->size = %d\n",
	__func__, presponse_t->size);

	/* Check if the response was for our command */
	if (presponse_t->seqno != pcommand_t->seqno) {

	dprintk(DBGLVL_CMD,
	"wrong event: seqno = %d, expected seqno = %d, will dequeue regardless\n",
	presponse_t->seqno, pcommand_t->seqno);

	ret = saa7164_cmd_dequeue(dev);
	if (ret != SAA_OK) {
	printk(KERN_ERR "dequeue failed, ret = %d\n",
	ret);
	if (safety++ > 16) {
	printk(KERN_ERR
	"dequeue exceeded, safety exit\n");
	return SAA_ERR_BUSY;
	}
	}

	continue;
	}

	if ((presponse_t->flags & PVC_RESPONSEFLAG_ERROR) != 0) {

	memset(&errdata[0], 0, sizeof(errdata));

	ret = saa7164_bus_get(dev, presponse_t, &errdata[0], 0);
	if (ret != SAA_OK) {
	printk(KERN_ERR "get error(2)\n");
	return ret;
	}

	saa7164_cmd_free_seqno(dev, pcommand_t->seqno);

	dprintk(DBGLVL_CMD, "%s() errdata %02x%02x%02x%02x\n",
	__func__, errdata[0], errdata[1], errdata[2],
	errdata[3]);

	/* Map error codes */
	dprintk(DBGLVL_CMD, "%s() cmd, error code = 0x%x\n",
	__func__, errdata[0]);

	switch (errdata[0]) {
	case PVC_ERRORCODE_INVALID_COMMAND:
	dprintk(DBGLVL_CMD, "%s() INVALID_COMMAND\n",
	__func__);
	ret = SAA_ERR_INVALID_COMMAND;
	break;
	case PVC_ERRORCODE_INVALID_DATA:
	dprintk(DBGLVL_CMD, "%s() INVALID_DATA\n",
	__func__);
	ret = SAA_ERR_BAD_PARAMETER;
	break;
	case PVC_ERRORCODE_TIMEOUT:
	dprintk(DBGLVL_CMD, "%s() TIMEOUT\n", __func__);
	ret = SAA_ERR_TIMEOUT;
	break;
	case PVC_ERRORCODE_NAK:
	dprintk(DBGLVL_CMD, "%s() NAK\n", __func__);
	ret = SAA_ERR_NULL_PACKET;
	break;
	case PVC_ERRORCODE_UNKNOWN:
	case PVC_ERRORCODE_INVALID_CONTROL:
	dprintk(DBGLVL_CMD,
	"%s() UNKNOWN OR INVALID CONTROL\n",
	__func__);
	ret = SAA_ERR_NOT_SUPPORTED;
	break;
	default:
	dprintk(DBGLVL_CMD, "%s() UNKNOWN\n", __func__);
	ret = SAA_ERR_NOT_SUPPORTED;
	}

	/* See of other commands are on the bus */
	if (saa7164_cmd_dequeue(dev) != SAA_OK)
	printk(KERN_ERR "dequeue(2) failed\n");

	return ret;
	}

	/* If response is invalid */
	if ((presponse_t->id != pcommand_t->id) \|\|
	(presponse_t->command != pcommand_t->command) \|\|
	(presponse_t->controlselector !=
	pcommand_t->controlselector) \|\|
	(((resp_dsize - data_recd) != presponse_t->size) &&
	!(presponse_t->flags & PVC_CMDFLAG_CONTINUE)) \|\|
	((resp_dsize - data_recd) < presponse_t->size)) {

	/* Invalid */
	dprintk(DBGLVL_CMD, "%s() Invalid\n", __func__);
	ret = saa7164_bus_get(dev, presponse_t, NULL, 0);
	if (ret != SAA_OK) {
	printk(KERN_ERR "get failed\n");
	return ret;
	}

	/* See of other commands are on the bus */
	if (saa7164_cmd_dequeue(dev) != SAA_OK)
	printk(KERN_ERR "dequeue(3) failed\n");
	continue;
	}

	/* OK, now we're actually getting out correct response */
	ret = saa7164_bus_get(dev, presponse_t, buf + data_recd, 0);
	if (ret != SAA_OK) {
	printk(KERN_ERR "get failed\n");
	return ret;
	}

	data_recd = presponse_t->size + data_recd;
	if (resp_dsize == data_recd) {
	dprintk(DBGLVL_CMD, "%s() Resp recd\n", __func__);
	break;
	}

	/* See of other commands are on the bus */
	if (saa7164_cmd_dequeue(dev) != SAA_OK)
	printk(KERN_ERR "dequeue(3) failed\n");
	} /* (loop) */

	/* Release the sequence number allocation */
	saa7164_cmd_free_seqno(dev, pcommand_t->seqno);

	/* if powerdown signal all pending commands */

	dprintk(DBGLVL_CMD, "%s() Calling dequeue then exit\n", __func__);

	/* See of other commands are on the bus */
	if (saa7164_cmd_dequeue(dev) != SAA_OK)
	printk(KERN_ERR "dequeue(4) failed\n");

	ret = SAA_OK;
	out:
	return ret;
	}