Add the rt linux 4.1.3-rt3 as base

Import the rt linux 4.1.3-rt3 as OPNFV kvm base.

It's from git://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-rt-devel.git linux-4.1.y-rt and
the base is:

commit 0917f823c59692d751951bf5ea699a2d1e2f26a2
Author: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Date:   Sat Jul 25 12:13:34 2015 +0200

    Prepare v4.1.3-rt3

    Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

We lose all the git history this way and it's not good. We
should apply another opnfv project repo in future.

Change-Id: I87543d81c9df70d99c5001fbdf646b202c19f423
Signed-off-by: Yunhong Jiang <yunhong.jiang@intel.com>

1 files changed, 1063 insertions, 0 deletions

diff --git a/kernel/drivers/media/platform/omap3isp/ispstat.c b/kernel/drivers/media/platform/omap3isp/ispstat.c
new file mode 100644
index 000000000..20434e83e
--- /dev/null
+++ b/kernel/drivers/media/platform/omap3isp/ispstat.c
@@ -0,0 +1,1063 @@
+/*
+ * ispstat.c
+ *
+ * TI OMAP3 ISP - Statistics core
+ *
+ * Copyright (C) 2010 Nokia Corporation
+ * Copyright (C) 2009 Texas Instruments, Inc
+ *
+ * Contacts: David Cohen <dacohen@gmail.com>
+ *	     Laurent Pinchart <laurent.pinchart@ideasonboard.com>
+ *	     Sakari Ailus <sakari.ailus@iki.fi>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/dma-mapping.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+
+#include "isp.h"
+
+#define ISP_STAT_USES_DMAENGINE(stat)	((stat)->dma_ch != NULL)
+
+/*
+ * MAGIC_SIZE must always be the greatest common divisor of
+ * AEWB_PACKET_SIZE and AF_PAXEL_SIZE.
+ */
+#define MAGIC_SIZE		16
+#define MAGIC_NUM		0x55
+
+/* HACK: AF module seems to be writing one more paxel data than it should. */
+#define AF_EXTRA_DATA		OMAP3ISP_AF_PAXEL_SIZE
+
+/*
+ * HACK: H3A modules go to an invalid state after have a SBL overflow. It makes
+ * the next buffer to start to be written in the same point where the overflow
+ * occurred instead of the configured address. The only known way to make it to
+ * go back to a valid state is having a valid buffer processing. Of course it
+ * requires at least a doubled buffer size to avoid an access to invalid memory
+ * region. But it does not fix everything. It may happen more than one
+ * consecutive SBL overflows. In that case, it might be unpredictable how many
+ * buffers the allocated memory should fit. For that case, a recover
+ * configuration was created. It produces the minimum buffer size for each H3A
+ * module and decrease the change for more SBL overflows. This recover state
+ * will be enabled every time a SBL overflow occur. As the output buffer size
+ * isn't big, it's possible to have an extra size able to fit many recover
+ * buffers making it extreamily unlikely to have an access to invalid memory
+ * region.
+ */
+#define NUM_H3A_RECOVER_BUFS	10
+
+/*
+ * HACK: Because of HW issues the generic layer sometimes need to have
+ * different behaviour for different statistic modules.
+ */
+#define IS_H3A_AF(stat)		((stat) == &(stat)->isp->isp_af)
+#define IS_H3A_AEWB(stat)	((stat) == &(stat)->isp->isp_aewb)
+#define IS_H3A(stat)		(IS_H3A_AF(stat) || IS_H3A_AEWB(stat))
+
+static void __isp_stat_buf_sync_magic(struct ispstat *stat,
+				      struct ispstat_buffer *buf,
+				      u32 buf_size, enum dma_data_direction dir,
+				      void (*dma_sync)(struct device *,
+					dma_addr_t, unsigned long, size_t,
+					enum dma_data_direction))
+{
+	/* Sync the initial and final magic words. */
+	dma_sync(stat->isp->dev, buf->dma_addr, 0, MAGIC_SIZE, dir);
+	dma_sync(stat->isp->dev, buf->dma_addr + (buf_size & PAGE_MASK),
+		 buf_size & ~PAGE_MASK, MAGIC_SIZE, dir);
+}
+
+static void isp_stat_buf_sync_magic_for_device(struct ispstat *stat,
+					       struct ispstat_buffer *buf,
+					       u32 buf_size,
+					       enum dma_data_direction dir)
+{
+	if (ISP_STAT_USES_DMAENGINE(stat))
+		return;
+
+	__isp_stat_buf_sync_magic(stat, buf, buf_size, dir,
+				  dma_sync_single_range_for_device);
+}
+
+static void isp_stat_buf_sync_magic_for_cpu(struct ispstat *stat,
+					    struct ispstat_buffer *buf,
+					    u32 buf_size,
+					    enum dma_data_direction dir)
+{
+	if (ISP_STAT_USES_DMAENGINE(stat))
+		return;
+
+	__isp_stat_buf_sync_magic(stat, buf, buf_size, dir,
+				  dma_sync_single_range_for_cpu);
+}
+
+static int isp_stat_buf_check_magic(struct ispstat *stat,
+				    struct ispstat_buffer *buf)
+{
+	const u32 buf_size = IS_H3A_AF(stat) ?
+			     buf->buf_size + AF_EXTRA_DATA : buf->buf_size;
+	u8 *w;
+	u8 *end;
+	int ret = -EINVAL;
+
+	isp_stat_buf_sync_magic_for_cpu(stat, buf, buf_size, DMA_FROM_DEVICE);
+
+	/* Checking initial magic numbers. They shouldn't be here anymore. */
+	for (w = buf->virt_addr, end = w + MAGIC_SIZE; w < end; w++)
+		if (likely(*w != MAGIC_NUM))
+			ret = 0;
+
+	if (ret) {
+		dev_dbg(stat->isp->dev, "%s: beginning magic check does not "
+					"match.\n", stat->subdev.name);
+		return ret;
+	}
+
+	/* Checking magic numbers at the end. They must be still here. */
+	for (w = buf->virt_addr + buf_size, end = w + MAGIC_SIZE;
+	     w < end; w++) {
+		if (unlikely(*w != MAGIC_NUM)) {
+			dev_dbg(stat->isp->dev, "%s: ending magic check does "
+				"not match.\n", stat->subdev.name);
+			return -EINVAL;
+		}
+	}
+
+	isp_stat_buf_sync_magic_for_device(stat, buf, buf_size,
+					   DMA_FROM_DEVICE);
+
+	return 0;
+}
+
+static void isp_stat_buf_insert_magic(struct ispstat *stat,
+				      struct ispstat_buffer *buf)
+{
+	const u32 buf_size = IS_H3A_AF(stat) ?
+			     stat->buf_size + AF_EXTRA_DATA : stat->buf_size;
+
+	isp_stat_buf_sync_magic_for_cpu(stat, buf, buf_size, DMA_FROM_DEVICE);
+
+	/*
+	 * Inserting MAGIC_NUM at the beginning and end of the buffer.
+	 * buf->buf_size is set only after the buffer is queued. For now the
+	 * right buf_size for the current configuration is pointed by
+	 * stat->buf_size.
+	 */
+	memset(buf->virt_addr, MAGIC_NUM, MAGIC_SIZE);
+	memset(buf->virt_addr + buf_size, MAGIC_NUM, MAGIC_SIZE);
+
+	isp_stat_buf_sync_magic_for_device(stat, buf, buf_size,
+					   DMA_BIDIRECTIONAL);
+}
+
+static void isp_stat_buf_sync_for_device(struct ispstat *stat,
+					 struct ispstat_buffer *buf)
+{
+	if (ISP_STAT_USES_DMAENGINE(stat))
+		return;
+
+	dma_sync_sg_for_device(stat->isp->dev, buf->sgt.sgl,
+			       buf->sgt.nents, DMA_FROM_DEVICE);
+}
+
+static void isp_stat_buf_sync_for_cpu(struct ispstat *stat,
+				      struct ispstat_buffer *buf)
+{
+	if (ISP_STAT_USES_DMAENGINE(stat))
+		return;
+
+	dma_sync_sg_for_cpu(stat->isp->dev, buf->sgt.sgl,
+			    buf->sgt.nents, DMA_FROM_DEVICE);
+}
+
+static void isp_stat_buf_clear(struct ispstat *stat)
+{
+	int i;
+
+	for (i = 0; i < STAT_MAX_BUFS; i++)
+		stat->buf[i].empty = 1;
+}
+
+static struct ispstat_buffer *
+__isp_stat_buf_find(struct ispstat *stat, int look_empty)
+{
+	struct ispstat_buffer *found = NULL;
+	int i;
+
+	for (i = 0; i < STAT_MAX_BUFS; i++) {
+		struct ispstat_buffer *curr = &stat->buf[i];
+
+		/*
+		 * Don't select the buffer which is being copied to
+		 * userspace or used by the module.
+		 */
+		if (curr == stat->locked_buf || curr == stat->active_buf)
+			continue;
+
+		/* Don't select uninitialised buffers if it's not required */
+		if (!look_empty && curr->empty)
+			continue;
+
+		/* Pick uninitialised buffer over anything else if look_empty */
+		if (curr->empty) {
+			found = curr;
+			break;
+		}
+
+		/* Choose the oldest buffer */
+		if (!found ||
+		    (s32)curr->frame_number - (s32)found->frame_number < 0)
+			found = curr;
+	}
+
+	return found;
+}
+
+static inline struct ispstat_buffer *
+isp_stat_buf_find_oldest(struct ispstat *stat)
+{
+	return __isp_stat_buf_find(stat, 0);
+}
+
+static inline struct ispstat_buffer *
+isp_stat_buf_find_oldest_or_empty(struct ispstat *stat)
+{
+	return __isp_stat_buf_find(stat, 1);
+}
+
+static int isp_stat_buf_queue(struct ispstat *stat)
+{
+	if (!stat->active_buf)
+		return STAT_NO_BUF;
+
+	ktime_get_ts(&stat->active_buf->ts);
+
+	stat->active_buf->buf_size = stat->buf_size;
+	if (isp_stat_buf_check_magic(stat, stat->active_buf)) {
+		dev_dbg(stat->isp->dev, "%s: data wasn't properly written.\n",
+			stat->subdev.name);
+		return STAT_NO_BUF;
+	}
+	stat->active_buf->config_counter = stat->config_counter;
+	stat->active_buf->frame_number = stat->frame_number;
+	stat->active_buf->empty = 0;
+	stat->active_buf = NULL;
+
+	return STAT_BUF_DONE;
+}
+
+/* Get next free buffer to write the statistics to and mark it active. */
+static void isp_stat_buf_next(struct ispstat *stat)
+{
+	if (unlikely(stat->active_buf))
+		/* Overwriting unused active buffer */
+		dev_dbg(stat->isp->dev, "%s: new buffer requested without "
+					"queuing active one.\n",
+					stat->subdev.name);
+	else
+		stat->active_buf = isp_stat_buf_find_oldest_or_empty(stat);
+}
+
+static void isp_stat_buf_release(struct ispstat *stat)
+{
+	unsigned long flags;
+
+	isp_stat_buf_sync_for_device(stat, stat->locked_buf);
+	spin_lock_irqsave(&stat->isp->stat_lock, flags);
+	stat->locked_buf = NULL;
+	spin_unlock_irqrestore(&stat->isp->stat_lock, flags);
+}
+
+/* Get buffer to userspace. */
+static struct ispstat_buffer *isp_stat_buf_get(struct ispstat *stat,
+					       struct omap3isp_stat_data *data)
+{
+	int rval = 0;
+	unsigned long flags;
+	struct ispstat_buffer *buf;
+
+	spin_lock_irqsave(&stat->isp->stat_lock, flags);
+
+	while (1) {
+		buf = isp_stat_buf_find_oldest(stat);
+		if (!buf) {
+			spin_unlock_irqrestore(&stat->isp->stat_lock, flags);
+			dev_dbg(stat->isp->dev, "%s: cannot find a buffer.\n",
+				stat->subdev.name);
+			return ERR_PTR(-EBUSY);
+		}
+		if (isp_stat_buf_check_magic(stat, buf)) {
+			dev_dbg(stat->isp->dev, "%s: current buffer has "
+				"corrupted data\n.", stat->subdev.name);
+			/* Mark empty because it doesn't have valid data. */
+			buf->empty = 1;
+		} else {
+			/* Buffer isn't corrupted. */
+			break;
+		}
+	}
+
+	stat->locked_buf = buf;
+
+	spin_unlock_irqrestore(&stat->isp->stat_lock, flags);
+
+	if (buf->buf_size > data->buf_size) {
+		dev_warn(stat->isp->dev, "%s: userspace's buffer size is "
+					 "not enough.\n", stat->subdev.name);
+		isp_stat_buf_release(stat);
+		return ERR_PTR(-EINVAL);
+	}
+
+	isp_stat_buf_sync_for_cpu(stat, buf);
+
+	rval = copy_to_user(data->buf,
+			    buf->virt_addr,
+			    buf->buf_size);
+
+	if (rval) {
+		dev_info(stat->isp->dev,
+			 "%s: failed copying %d bytes of stat data\n",
+			 stat->subdev.name, rval);
+		buf = ERR_PTR(-EFAULT);
+		isp_stat_buf_release(stat);
+	}
+
+	return buf;
+}
+
+static void isp_stat_bufs_free(struct ispstat *stat)
+{
+	struct device *dev = ISP_STAT_USES_DMAENGINE(stat)
+			   ? NULL : stat->isp->dev;
+	unsigned int i;
+
+	for (i = 0; i < STAT_MAX_BUFS; i++) {
+		struct ispstat_buffer *buf = &stat->buf[i];
+
+		if (!buf->virt_addr)
+			continue;
+
+		sg_free_table(&buf->sgt);
+
+		dma_free_coherent(dev, stat->buf_alloc_size, buf->virt_addr,
+				  buf->dma_addr);
+
+		buf->dma_addr = 0;
+		buf->virt_addr = NULL;
+		buf->empty = 1;
+	}
+
+	dev_dbg(stat->isp->dev, "%s: all buffers were freed.\n",
+		stat->subdev.name);
+
+	stat->buf_alloc_size = 0;
+	stat->active_buf = NULL;
+}
+
+static int isp_stat_bufs_alloc_one(struct device *dev,
+				   struct ispstat_buffer *buf,
+				   unsigned int size)
+{
+	int ret;
+
+	buf->virt_addr = dma_alloc_coherent(dev, size, &buf->dma_addr,
+					    GFP_KERNEL | GFP_DMA);
+	if (!buf->virt_addr)
+		return -ENOMEM;
+
+	ret = dma_get_sgtable(dev, &buf->sgt, buf->virt_addr, buf->dma_addr,
+			      size);
+	if (ret < 0) {
+		dma_free_coherent(dev, size, buf->virt_addr, buf->dma_addr);
+		buf->virt_addr = NULL;
+		buf->dma_addr = 0;
+		return ret;
+	}
+
+	return 0;
+}
+
+/*
+ * The device passed to the DMA API depends on whether the statistics block uses
+ * ISP DMA, external DMA or PIO to transfer data.
+ *
+ * The first case (for the AEWB and AF engines) passes the ISP device, resulting
+ * in the DMA buffers being mapped through the ISP IOMMU.
+ *
+ * The second case (for the histogram engine) should pass the DMA engine device.
+ * As that device isn't accessible through the OMAP DMA engine API the driver
+ * passes NULL instead, resulting in the buffers being mapped directly as
+ * physical pages.
+ *
+ * The third case (for the histogram engine) doesn't require any mapping. The
+ * buffers could be allocated with kmalloc/vmalloc, but we still use
+ * dma_alloc_coherent() for consistency purpose.
+ */
+static int isp_stat_bufs_alloc(struct ispstat *stat, u32 size)
+{
+	struct device *dev = ISP_STAT_USES_DMAENGINE(stat)
+			   ? NULL : stat->isp->dev;
+	unsigned long flags;
+	unsigned int i;
+
+	spin_lock_irqsave(&stat->isp->stat_lock, flags);
+
+	BUG_ON(stat->locked_buf != NULL);
+
+	/* Are the old buffers big enough? */
+	if (stat->buf_alloc_size >= size) {
+		spin_unlock_irqrestore(&stat->isp->stat_lock, flags);
+		return 0;
+	}
+
+	if (stat->state != ISPSTAT_DISABLED || stat->buf_processing) {
+		dev_info(stat->isp->dev,
+			 "%s: trying to allocate memory when busy\n",
+			 stat->subdev.name);
+		spin_unlock_irqrestore(&stat->isp->stat_lock, flags);
+		return -EBUSY;
+	}
+
+	spin_unlock_irqrestore(&stat->isp->stat_lock, flags);
+
+	isp_stat_bufs_free(stat);
+
+	stat->buf_alloc_size = size;
+
+	for (i = 0; i < STAT_MAX_BUFS; i++) {
+		struct ispstat_buffer *buf = &stat->buf[i];
+		int ret;
+
+		ret = isp_stat_bufs_alloc_one(dev, buf, size);
+		if (ret < 0) {
+			dev_err(stat->isp->dev,
+				"%s: Failed to allocate DMA buffer %u\n",
+				stat->subdev.name, i);
+			isp_stat_bufs_free(stat);
+			return ret;
+		}
+
+		buf->empty = 1;
+
+		dev_dbg(stat->isp->dev,
+			"%s: buffer[%u] allocated. dma=0x%08lx virt=0x%08lx",
+			stat->subdev.name, i,
+			(unsigned long)buf->dma_addr,
+			(unsigned long)buf->virt_addr);
+	}
+
+	return 0;
+}
+
+static void isp_stat_queue_event(struct ispstat *stat, int err)
+{
+	struct video_device *vdev = stat->subdev.devnode;
+	struct v4l2_event event;
+	struct omap3isp_stat_event_status *status = (void *)event.u.data;
+
+	memset(&event, 0, sizeof(event));
+	if (!err) {
+		status->frame_number = stat->frame_number;
+		status->config_counter = stat->config_counter;
+	} else {
+		status->buf_err = 1;
+	}
+	event.type = stat->event_type;
+	v4l2_event_queue(vdev, &event);
+}
+
+
+/*
+ * omap3isp_stat_request_statistics - Request statistics.
+ * @data: Pointer to return statistics data.
+ *
+ * Returns 0 if successful.
+ */
+int omap3isp_stat_request_statistics(struct ispstat *stat,
+				     struct omap3isp_stat_data *data)
+{
+	struct ispstat_buffer *buf;
+
+	if (stat->state != ISPSTAT_ENABLED) {
+		dev_dbg(stat->isp->dev, "%s: engine not enabled.\n",
+			stat->subdev.name);
+		return -EINVAL;
+	}
+
+	mutex_lock(&stat->ioctl_lock);
+	buf = isp_stat_buf_get(stat, data);
+	if (IS_ERR(buf)) {
+		mutex_unlock(&stat->ioctl_lock);
+		return PTR_ERR(buf);
+	}
+
+	data->ts.tv_sec = buf->ts.tv_sec;
+	data->ts.tv_usec = buf->ts.tv_nsec / NSEC_PER_USEC;
+	data->config_counter = buf->config_counter;
+	data->frame_number = buf->frame_number;
+	data->buf_size = buf->buf_size;
+
+	buf->empty = 1;
+	isp_stat_buf_release(stat);
+	mutex_unlock(&stat->ioctl_lock);
+
+	return 0;
+}
+
+/*
+ * omap3isp_stat_config - Receives new statistic engine configuration.
+ * @new_conf: Pointer to config structure.
+ *
+ * Returns 0 if successful, -EINVAL if new_conf pointer is NULL, -ENOMEM if
+ * was unable to allocate memory for the buffer, or other errors if parameters
+ * are invalid.
+ */
+int omap3isp_stat_config(struct ispstat *stat, void *new_conf)
+{
+	int ret;
+	unsigned long irqflags;
+	struct ispstat_generic_config *user_cfg = new_conf;
+	u32 buf_size = user_cfg->buf_size;
+
+	if (!new_conf) {
+		dev_dbg(stat->isp->dev, "%s: configuration is NULL\n",
+			stat->subdev.name);
+		return -EINVAL;
+	}
+
+	mutex_lock(&stat->ioctl_lock);
+
+	dev_dbg(stat->isp->dev, "%s: configuring module with buffer "
+		"size=0x%08lx\n", stat->subdev.name, (unsigned long)buf_size);
+
+	ret = stat->ops->validate_params(stat, new_conf);
+	if (ret) {
+		mutex_unlock(&stat->ioctl_lock);
+		dev_dbg(stat->isp->dev, "%s: configuration values are "
+					"invalid.\n", stat->subdev.name);
+		return ret;
+	}
+
+	if (buf_size != user_cfg->buf_size)
+		dev_dbg(stat->isp->dev, "%s: driver has corrected buffer size "
+			"request to 0x%08lx\n", stat->subdev.name,
+			(unsigned long)user_cfg->buf_size);
+
+	/*
+	 * Hack: H3A modules may need a doubled buffer size to avoid access
+	 * to a invalid memory address after a SBL overflow.
+	 * The buffer size is always PAGE_ALIGNED.
+	 * Hack 2: MAGIC_SIZE is added to buf_size so a magic word can be
+	 * inserted at the end to data integrity check purpose.
+	 * Hack 3: AF module writes one paxel data more than it should, so
+	 * the buffer allocation must consider it to avoid invalid memory
+	 * access.
+	 * Hack 4: H3A need to allocate extra space for the recover state.
+	 */
+	if (IS_H3A(stat)) {
+		buf_size = user_cfg->buf_size * 2 + MAGIC_SIZE;
+		if (IS_H3A_AF(stat))
+			/*
+			 * Adding one extra paxel data size for each recover
+			 * buffer + 2 regular ones.
+			 */
+			buf_size += AF_EXTRA_DATA * (NUM_H3A_RECOVER_BUFS + 2);
+		if (stat->recover_priv) {
+			struct ispstat_generic_config *recover_cfg =
+				stat->recover_priv;
+			buf_size += recover_cfg->buf_size *
+				    NUM_H3A_RECOVER_BUFS;
+		}
+		buf_size = PAGE_ALIGN(buf_size);
+	} else { /* Histogram */
+		buf_size = PAGE_ALIGN(user_cfg->buf_size + MAGIC_SIZE);
+	}
+
+	ret = isp_stat_bufs_alloc(stat, buf_size);
+	if (ret) {
+		mutex_unlock(&stat->ioctl_lock);
+		return ret;
+	}
+
+	spin_lock_irqsave(&stat->isp->stat_lock, irqflags);
+	stat->ops->set_params(stat, new_conf);
+	spin_unlock_irqrestore(&stat->isp->stat_lock, irqflags);
+
+	/*
+	 * Returning the right future config_counter for this setup, so
+	 * userspace can *know* when it has been applied.
+	 */
+	user_cfg->config_counter = stat->config_counter + stat->inc_config;
+
+	/* Module has a valid configuration. */
+	stat->configured = 1;
+	dev_dbg(stat->isp->dev, "%s: module has been successfully "
+		"configured.\n", stat->subdev.name);
+
+	mutex_unlock(&stat->ioctl_lock);
+
+	return 0;
+}
+
+/*
+ * isp_stat_buf_process - Process statistic buffers.
+ * @buf_state: points out if buffer is ready to be processed. It's necessary
+ *	       because histogram needs to copy the data from internal memory
+ *	       before be able to process the buffer.
+ */
+static int isp_stat_buf_process(struct ispstat *stat, int buf_state)
+{
+	int ret = STAT_NO_BUF;
+
+	if (!atomic_add_unless(&stat->buf_err, -1, 0) &&
+	    buf_state == STAT_BUF_DONE && stat->state == ISPSTAT_ENABLED) {
+		ret = isp_stat_buf_queue(stat);
+		isp_stat_buf_next(stat);
+	}
+
+	return ret;
+}
+
+int omap3isp_stat_pcr_busy(struct ispstat *stat)
+{
+	return stat->ops->busy(stat);
+}
+
+int omap3isp_stat_busy(struct ispstat *stat)
+{
+	return omap3isp_stat_pcr_busy(stat) | stat->buf_processing |
+		(stat->state != ISPSTAT_DISABLED);
+}
+
+/*
+ * isp_stat_pcr_enable - Disables/Enables statistic engines.
+ * @pcr_enable: 0/1 - Disables/Enables the engine.
+ *
+ * Must be called from ISP driver when the module is idle and synchronized
+ * with CCDC.
+ */
+static void isp_stat_pcr_enable(struct ispstat *stat, u8 pcr_enable)
+{
+	if ((stat->state != ISPSTAT_ENABLING &&
+	     stat->state != ISPSTAT_ENABLED) && pcr_enable)
+		/* Userspace has disabled the module. Aborting. */
+		return;
+
+	stat->ops->enable(stat, pcr_enable);
+	if (stat->state == ISPSTAT_DISABLING && !pcr_enable)
+		stat->state = ISPSTAT_DISABLED;
+	else if (stat->state == ISPSTAT_ENABLING && pcr_enable)
+		stat->state = ISPSTAT_ENABLED;
+}
+
+void omap3isp_stat_suspend(struct ispstat *stat)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&stat->isp->stat_lock, flags);
+
+	if (stat->state != ISPSTAT_DISABLED)
+		stat->ops->enable(stat, 0);
+	if (stat->state == ISPSTAT_ENABLED)
+		stat->state = ISPSTAT_SUSPENDED;
+
+	spin_unlock_irqrestore(&stat->isp->stat_lock, flags);
+}
+
+void omap3isp_stat_resume(struct ispstat *stat)
+{
+	/* Module will be re-enabled with its pipeline */
+	if (stat->state == ISPSTAT_SUSPENDED)
+		stat->state = ISPSTAT_ENABLING;
+}
+
+static void isp_stat_try_enable(struct ispstat *stat)
+{
+	unsigned long irqflags;
+
+	if (stat->priv == NULL)
+		/* driver wasn't initialised */
+		return;
+
+	spin_lock_irqsave(&stat->isp->stat_lock, irqflags);
+	if (stat->state == ISPSTAT_ENABLING && !stat->buf_processing &&
+	    stat->buf_alloc_size) {
+		/*
+		 * Userspace's requested to enable the engine but it wasn't yet.
+		 * Let's do that now.
+		 */
+		stat->update = 1;
+		isp_stat_buf_next(stat);
+		stat->ops->setup_regs(stat, stat->priv);
+		isp_stat_buf_insert_magic(stat, stat->active_buf);
+
+		/*
+		 * H3A module has some hw issues which forces the driver to
+		 * ignore next buffers even if it was disabled in the meantime.
+		 * On the other hand, Histogram shouldn't ignore buffers anymore
+		 * if it's being enabled.
+		 */
+		if (!IS_H3A(stat))
+			atomic_set(&stat->buf_err, 0);
+
+		isp_stat_pcr_enable(stat, 1);
+		spin_unlock_irqrestore(&stat->isp->stat_lock, irqflags);
+		dev_dbg(stat->isp->dev, "%s: module is enabled.\n",
+			stat->subdev.name);
+	} else {
+		spin_unlock_irqrestore(&stat->isp->stat_lock, irqflags);
+	}
+}
+
+void omap3isp_stat_isr_frame_sync(struct ispstat *stat)
+{
+	isp_stat_try_enable(stat);
+}
+
+void omap3isp_stat_sbl_overflow(struct ispstat *stat)
+{
+	unsigned long irqflags;
+
+	spin_lock_irqsave(&stat->isp->stat_lock, irqflags);
+	/*
+	 * Due to a H3A hw issue which prevents the next buffer to start from
+	 * the correct memory address, 2 buffers must be ignored.
+	 */
+	atomic_set(&stat->buf_err, 2);
+
+	/*
+	 * If more than one SBL overflow happen in a row, H3A module may access
+	 * invalid memory region.
+	 * stat->sbl_ovl_recover is set to tell to the driver to temporarily use
+	 * a soft configuration which helps to avoid consecutive overflows.
+	 */
+	if (stat->recover_priv)
+		stat->sbl_ovl_recover = 1;
+	spin_unlock_irqrestore(&stat->isp->stat_lock, irqflags);
+}
+
+/*
+ * omap3isp_stat_enable - Disable/Enable statistic engine as soon as possible
+ * @enable: 0/1 - Disables/Enables the engine.
+ *
+ * Client should configure all the module registers before this.
+ * This function can be called from a userspace request.
+ */
+int omap3isp_stat_enable(struct ispstat *stat, u8 enable)
+{
+	unsigned long irqflags;
+
+	dev_dbg(stat->isp->dev, "%s: user wants to %s module.\n",
+		stat->subdev.name, enable ? "enable" : "disable");
+
+	/* Prevent enabling while configuring */
+	mutex_lock(&stat->ioctl_lock);
+
+	spin_lock_irqsave(&stat->isp->stat_lock, irqflags);
+
+	if (!stat->configured && enable) {
+		spin_unlock_irqrestore(&stat->isp->stat_lock, irqflags);
+		mutex_unlock(&stat->ioctl_lock);
+		dev_dbg(stat->isp->dev, "%s: cannot enable module as it's "
+			"never been successfully configured so far.\n",
+			stat->subdev.name);
+		return -EINVAL;
+	}
+
+	if (enable) {
+		if (stat->state == ISPSTAT_DISABLING)
+			/* Previous disabling request wasn't done yet */
+			stat->state = ISPSTAT_ENABLED;
+		else if (stat->state == ISPSTAT_DISABLED)
+			/* Module is now being enabled */
+			stat->state = ISPSTAT_ENABLING;
+	} else {
+		if (stat->state == ISPSTAT_ENABLING) {
+			/* Previous enabling request wasn't done yet */
+			stat->state = ISPSTAT_DISABLED;
+		} else if (stat->state == ISPSTAT_ENABLED) {
+			/* Module is now being disabled */
+			stat->state = ISPSTAT_DISABLING;
+			isp_stat_buf_clear(stat);
+		}
+	}
+
+	spin_unlock_irqrestore(&stat->isp->stat_lock, irqflags);
+	mutex_unlock(&stat->ioctl_lock);
+
+	return 0;
+}
+
+int omap3isp_stat_s_stream(struct v4l2_subdev *subdev, int enable)
+{
+	struct ispstat *stat = v4l2_get_subdevdata(subdev);
+
+	if (enable) {
+		/*
+		 * Only set enable PCR bit if the module was previously
+		 * enabled through ioctl.
+		 */
+		isp_stat_try_enable(stat);
+	} else {
+		unsigned long flags;
+		/* Disable PCR bit and config enable field */
+		omap3isp_stat_enable(stat, 0);
+		spin_lock_irqsave(&stat->isp->stat_lock, flags);
+		stat->ops->enable(stat, 0);
+		spin_unlock_irqrestore(&stat->isp->stat_lock, flags);
+
+		/*
+		 * If module isn't busy, a new interrupt may come or not to
+		 * set the state to DISABLED. As Histogram needs to read its
+		 * internal memory to clear it, let interrupt handler
+		 * responsible of changing state to DISABLED. If the last
+		 * interrupt is coming, it's still safe as the handler will
+		 * ignore the second time when state is already set to DISABLED.
+		 * It's necessary to synchronize Histogram with streamoff, once
+		 * the module may be considered idle before last SDMA transfer
+		 * starts if we return here.
+		 */
+		if (!omap3isp_stat_pcr_busy(stat))
+			omap3isp_stat_isr(stat);
+
+		dev_dbg(stat->isp->dev, "%s: module is being disabled\n",
+			stat->subdev.name);
+	}
+
+	return 0;
+}
+
+/*
+ * __stat_isr - Interrupt handler for statistic drivers
+ */
+static void __stat_isr(struct ispstat *stat, int from_dma)
+{
+	int ret = STAT_BUF_DONE;
+	int buf_processing;
+	unsigned long irqflags;
+	struct isp_pipeline *pipe;
+
+	/*
+	 * stat->buf_processing must be set before disable module. It's
+	 * necessary to not inform too early the buffers aren't busy in case
+	 * of SDMA is going to be used.
+	 */
+	spin_lock_irqsave(&stat->isp->stat_lock, irqflags);
+	if (stat->state == ISPSTAT_DISABLED) {
+		spin_unlock_irqrestore(&stat->isp->stat_lock, irqflags);
+		return;
+	}
+	buf_processing = stat->buf_processing;
+	stat->buf_processing = 1;
+	stat->ops->enable(stat, 0);
+
+	if (buf_processing && !from_dma) {
+		if (stat->state == ISPSTAT_ENABLED) {
+			spin_unlock_irqrestore(&stat->isp->stat_lock, irqflags);
+			dev_err(stat->isp->dev,
+				"%s: interrupt occurred when module was still "
+				"processing a buffer.\n", stat->subdev.name);
+			ret = STAT_NO_BUF;
+			goto out;
+		} else {
+			/*
+			 * Interrupt handler was called from streamoff when
+			 * the module wasn't busy anymore to ensure it is being
+			 * disabled after process last buffer. If such buffer
+			 * processing has already started, no need to do
+			 * anything else.
+			 */
+			spin_unlock_irqrestore(&stat->isp->stat_lock, irqflags);
+			return;
+		}
+	}
+	spin_unlock_irqrestore(&stat->isp->stat_lock, irqflags);
+
+	/* If it's busy we can't process this buffer anymore */
+	if (!omap3isp_stat_pcr_busy(stat)) {
+		if (!from_dma && stat->ops->buf_process)
+			/* Module still need to copy data to buffer. */
+			ret = stat->ops->buf_process(stat);
+		if (ret == STAT_BUF_WAITING_DMA)
+			/* Buffer is not ready yet */
+			return;
+
+		spin_lock_irqsave(&stat->isp->stat_lock, irqflags);
+
+		/*
+		 * Histogram needs to read its internal memory to clear it
+		 * before be disabled. For that reason, common statistic layer
+		 * can return only after call stat's buf_process() operator.
+		 */
+		if (stat->state == ISPSTAT_DISABLING) {
+			stat->state = ISPSTAT_DISABLED;
+			spin_unlock_irqrestore(&stat->isp->stat_lock, irqflags);
+			stat->buf_processing = 0;
+			return;
+		}
+		pipe = to_isp_pipeline(&stat->subdev.entity);
+		stat->frame_number = atomic_read(&pipe->frame_number);
+
+		/*
+		 * Before this point, 'ret' stores the buffer's status if it's
+		 * ready to be processed. Afterwards, it holds the status if
+		 * it was processed successfully.
+		 */
+		ret = isp_stat_buf_process(stat, ret);
+
+		if (likely(!stat->sbl_ovl_recover)) {
+			stat->ops->setup_regs(stat, stat->priv);
+		} else {
+			/*
+			 * Using recover config to increase the chance to have
+			 * a good buffer processing and make the H3A module to
+			 * go back to a valid state.
+			 */
+			stat->update = 1;
+			stat->ops->setup_regs(stat, stat->recover_priv);
+			stat->sbl_ovl_recover = 0;
+
+			/*
+			 * Set 'update' in case of the module needs to use
+			 * regular configuration after next buffer.
+			 */
+			stat->update = 1;
+		}
+
+		isp_stat_buf_insert_magic(stat, stat->active_buf);
+
+		/*
+		 * Hack: H3A modules may access invalid memory address or send
+		 * corrupted data to userspace if more than 1 SBL overflow
+		 * happens in a row without re-writing its buffer's start memory
+		 * address in the meantime. Such situation is avoided if the
+		 * module is not immediately re-enabled when the ISR misses the
+		 * timing to process the buffer and to setup the registers.
+		 * Because of that, pcr_enable(1) was moved to inside this 'if'
+		 * block. But the next interruption will still happen as during
+		 * pcr_enable(0) the module was busy.
+		 */
+		isp_stat_pcr_enable(stat, 1);
+		spin_unlock_irqrestore(&stat->isp->stat_lock, irqflags);
+	} else {
+		/*
+		 * If a SBL overflow occurs and the H3A driver misses the timing
+		 * to process the buffer, stat->buf_err is set and won't be
+		 * cleared now. So the next buffer will be correctly ignored.
+		 * It's necessary due to a hw issue which makes the next H3A
+		 * buffer to start from the memory address where the previous
+		 * one stopped, instead of start where it was configured to.
+		 * Do not "stat->buf_err = 0" here.
+		 */
+
+		if (stat->ops->buf_process)
+			/*
+			 * Driver may need to erase current data prior to
+			 * process a new buffer. If it misses the timing, the
+			 * next buffer might be wrong. So should be ignored.
+			 * It happens only for Histogram.
+			 */
+			atomic_set(&stat->buf_err, 1);
+
+		ret = STAT_NO_BUF;
+		dev_dbg(stat->isp->dev, "%s: cannot process buffer, "
+					"device is busy.\n", stat->subdev.name);
+	}
+
+out:
+	stat->buf_processing = 0;
+	isp_stat_queue_event(stat, ret != STAT_BUF_DONE);
+}
+
+void omap3isp_stat_isr(struct ispstat *stat)
+{
+	__stat_isr(stat, 0);
+}
+
+void omap3isp_stat_dma_isr(struct ispstat *stat)
+{
+	__stat_isr(stat, 1);
+}
+
+int omap3isp_stat_subscribe_event(struct v4l2_subdev *subdev,
+				  struct v4l2_fh *fh,
+				  struct v4l2_event_subscription *sub)
+{
+	struct ispstat *stat = v4l2_get_subdevdata(subdev);
+
+	if (sub->type != stat->event_type)
+		return -EINVAL;
+
+	return v4l2_event_subscribe(fh, sub, STAT_NEVENTS, NULL);
+}
+
+int omap3isp_stat_unsubscribe_event(struct v4l2_subdev *subdev,
+				    struct v4l2_fh *fh,
+				    struct v4l2_event_subscription *sub)
+{
+	return v4l2_event_unsubscribe(fh, sub);
+}
+
+void omap3isp_stat_unregister_entities(struct ispstat *stat)
+{
+	v4l2_device_unregister_subdev(&stat->subdev);
+}
+
+int omap3isp_stat_register_entities(struct ispstat *stat,
+				    struct v4l2_device *vdev)
+{
+	return v4l2_device_register_subdev(vdev, &stat->subdev);
+}
+
+static int isp_stat_init_entities(struct ispstat *stat, const char *name,
+				  const struct v4l2_subdev_ops *sd_ops)
+{
+	struct v4l2_subdev *subdev = &stat->subdev;
+	struct media_entity *me = &subdev->entity;
+
+	v4l2_subdev_init(subdev, sd_ops);
+	snprintf(subdev->name, V4L2_SUBDEV_NAME_SIZE, "OMAP3 ISP %s", name);
+	subdev->grp_id = 1 << 16;	/* group ID for isp subdevs */
+	subdev->flags |= V4L2_SUBDEV_FL_HAS_EVENTS | V4L2_SUBDEV_FL_HAS_DEVNODE;
+	v4l2_set_subdevdata(subdev, stat);
+
+	stat->pad.flags = MEDIA_PAD_FL_SINK | MEDIA_PAD_FL_MUST_CONNECT;
+	me->ops = NULL;
+
+	return media_entity_init(me, 1, &stat->pad, 0);
+}
+
+int omap3isp_stat_init(struct ispstat *stat, const char *name,
+		       const struct v4l2_subdev_ops *sd_ops)
+{
+	int ret;
+
+	stat->buf = kcalloc(STAT_MAX_BUFS, sizeof(*stat->buf), GFP_KERNEL);
+	if (!stat->buf)
+		return -ENOMEM;
+
+	isp_stat_buf_clear(stat);
+	mutex_init(&stat->ioctl_lock);
+	atomic_set(&stat->buf_err, 0);
+
+	ret = isp_stat_init_entities(stat, name, sd_ops);
+	if (ret < 0) {
+		mutex_destroy(&stat->ioctl_lock);
+		kfree(stat->buf);
+	}
+
+	return ret;
+}
+
+void omap3isp_stat_cleanup(struct ispstat *stat)
+{
+	media_entity_cleanup(&stat->subdev.entity);
+	mutex_destroy(&stat->ioctl_lock);
+	isp_stat_bufs_free(stat);
+	kfree(stat->buf);
+}