openwrt/target/linux/brcm2708/patches-4.1/0010-cma-Add-vc_cma-driver-to-enable-use-of-CMA.patch

From 2bb9c01de4c90a8b7a9b52be95f504177abeaf45 Mon Sep 17 00:00:00 2001
From: popcornmix <popcornmix@gmail.com>
Date: Wed, 3 Jul 2013 00:31:47 +0100
Subject: [PATCH 010/121] cma: Add vc_cma driver to enable use of CMA

Signed-off-by: popcornmix <popcornmix@gmail.com>

vc_cma: Make the vc_cma area the default contiguous DMA area
---
 drivers/char/Kconfig                  |    2 +
 drivers/char/Makefile                 |    1 +
 drivers/char/broadcom/Kconfig         |   15 +
 drivers/char/broadcom/Makefile        |    1 +
 drivers/char/broadcom/vc_cma/Makefile |   14 +
 drivers/char/broadcom/vc_cma/vc_cma.c | 1193 +++++++++++++++++++++++++++++++++
 include/linux/broadcom/vc_cma.h       |   29 +
 7 files changed, 1255 insertions(+)
 create mode 100644 drivers/char/broadcom/Kconfig
 create mode 100644 drivers/char/broadcom/Makefile
 create mode 100644 drivers/char/broadcom/vc_cma/Makefile
 create mode 100644 drivers/char/broadcom/vc_cma/vc_cma.c
 create mode 100644 include/linux/broadcom/vc_cma.h

--- a/drivers/char/Kconfig
+++ b/drivers/char/Kconfig
@@ -590,6 +590,8 @@ config DEVPORT
 
 source "drivers/s390/char/Kconfig"
 
+source "drivers/char/broadcom/Kconfig"
+
 config MSM_SMD_PKT
 	bool "Enable device interface for some SMD packet ports"
 	default n
--- a/drivers/char/Makefile
+++ b/drivers/char/Makefile
@@ -62,3 +62,4 @@ js-rtc-y = rtc.o
 
 obj-$(CONFIG_TILE_SROM)		+= tile-srom.o
 obj-$(CONFIG_XILLYBUS)		+= xillybus/
+obj-$(CONFIG_BRCM_CHAR_DRIVERS) += broadcom/
--- /dev/null
+++ b/drivers/char/broadcom/Kconfig
@@ -0,0 +1,15 @@
+#
+# Broadcom char driver config
+#
+
+menuconfig BRCM_CHAR_DRIVERS
+	bool "Broadcom Char Drivers"
+	help
+	  Broadcom's char drivers
+
+config BCM_VC_CMA
+	bool "Videocore CMA"
+	depends on CMA && BRCM_CHAR_DRIVERS && BCM2708_VCHIQ
+	default n
+        help
+          Helper for videocore CMA access.
--- /dev/null
+++ b/drivers/char/broadcom/Makefile
@@ -0,0 +1 @@
+obj-$(CONFIG_BCM_VC_CMA)	+= vc_cma/
--- /dev/null
+++ b/drivers/char/broadcom/vc_cma/Makefile
@@ -0,0 +1,14 @@
+ccflags-y  += -Wall -Wstrict-prototypes -Wno-trigraphs
+ccflags-y  += -Werror
+ccflags-y  += -Iinclude/linux/broadcom
+ccflags-y  += -Idrivers/misc/vc04_services
+ccflags-y  += -Idrivers/misc/vc04_services/interface/vchi
+ccflags-y  += -Idrivers/misc/vc04_services/interface/vchiq_arm
+
+ccflags-y  += -D__KERNEL__
+ccflags-y  += -D__linux__
+ccflags-y  += -Werror
+
+obj-$(CONFIG_BCM_VC_CMA) += vc-cma.o
+
+vc-cma-objs := vc_cma.o
--- /dev/null
+++ b/drivers/char/broadcom/vc_cma/vc_cma.c
@@ -0,0 +1,1193 @@
+/**
+ * Copyright (c) 2010-2012 Broadcom. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions, and the following disclaimer,
+ *    without modification.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. The names of the above-listed copyright holders may not be used
+ *    to endorse or promote products derived from this software without
+ *    specific prior written permission.
+ *
+ * ALTERNATIVELY, this software may be distributed under the terms of the
+ * GNU General Public License ("GPL") version 2, as published by the Free
+ * Software Foundation.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
+ * IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
+ * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
+ * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+ * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/kthread.h>
+#include <linux/fs.h>
+#include <linux/device.h>
+#include <linux/cdev.h>
+#include <linux/mm.h>
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
+#include <linux/dma-mapping.h>
+#include <linux/dma-contiguous.h>
+#include <linux/platform_device.h>
+#include <linux/uaccess.h>
+#include <asm/cacheflush.h>
+
+#include "vc_cma.h"
+
+#include "vchiq_util.h"
+#include "vchiq_connected.h"
+//#include "debug_sym.h"
+//#include "vc_mem.h"
+
+#define DRIVER_NAME  "vc-cma"
+
+#define LOG_DBG(fmt, ...) \
+	if (vc_cma_debug) \
+		printk(KERN_INFO fmt "\n", ##__VA_ARGS__)
+#define LOG_INFO(fmt, ...) \
+	printk(KERN_INFO fmt "\n", ##__VA_ARGS__)
+#define LOG_ERR(fmt, ...) \
+	printk(KERN_ERR fmt "\n", ##__VA_ARGS__)
+
+#define VC_CMA_FOURCC VCHIQ_MAKE_FOURCC('C', 'M', 'A', ' ')
+#define VC_CMA_VERSION 2
+
+#define VC_CMA_CHUNK_ORDER 6	/* 256K */
+#define VC_CMA_CHUNK_SIZE (4096 << VC_CMA_CHUNK_ORDER)
+#define VC_CMA_MAX_PARAMS_PER_MSG \
+	((VCHIQ_MAX_MSG_SIZE - sizeof(unsigned short))/sizeof(unsigned short))
+#define VC_CMA_RESERVE_COUNT_MAX 16
+
+#define PAGES_PER_CHUNK (VC_CMA_CHUNK_SIZE / PAGE_SIZE)
+
+#define VCADDR_TO_PHYSADDR(vcaddr) (mm_vc_mem_phys_addr + vcaddr)
+
+#define loud_error(...) \
+	LOG_ERR("===== " __VA_ARGS__)
+
+enum {
+	VC_CMA_MSG_QUIT,
+	VC_CMA_MSG_OPEN,
+	VC_CMA_MSG_TICK,
+	VC_CMA_MSG_ALLOC,	/* chunk count */
+	VC_CMA_MSG_FREE,	/* chunk, chunk, ... */
+	VC_CMA_MSG_ALLOCATED,	/* chunk, chunk, ... */
+	VC_CMA_MSG_REQUEST_ALLOC,	/* chunk count */
+	VC_CMA_MSG_REQUEST_FREE,	/* chunk count */
+	VC_CMA_MSG_RESERVE,	/* bytes lo, bytes hi */
+	VC_CMA_MSG_UPDATE_RESERVE,
+	VC_CMA_MSG_MAX
+};
+
+struct cma_msg {
+	unsigned short type;
+	unsigned short params[VC_CMA_MAX_PARAMS_PER_MSG];
+};
+
+struct vc_cma_reserve_user {
+	unsigned int pid;
+	unsigned int reserve;
+};
+
+/* Device (/dev) related variables */
+static dev_t vc_cma_devnum;
+static struct class *vc_cma_class;
+static struct cdev vc_cma_cdev;
+static int vc_cma_inited;
+static int vc_cma_debug;
+
+/* Proc entry */
+static struct proc_dir_entry *vc_cma_proc_entry;
+
+phys_addr_t vc_cma_base;
+struct page *vc_cma_base_page;
+unsigned int vc_cma_size;
+EXPORT_SYMBOL(vc_cma_size);
+unsigned int vc_cma_initial;
+unsigned int vc_cma_chunks;
+unsigned int vc_cma_chunks_used;
+unsigned int vc_cma_chunks_reserved;
+
+
+void *vc_cma_dma_alloc;
+unsigned int vc_cma_dma_size;
+
+static int in_loud_error;
+
+unsigned int vc_cma_reserve_total;
+unsigned int vc_cma_reserve_count;
+struct vc_cma_reserve_user vc_cma_reserve_users[VC_CMA_RESERVE_COUNT_MAX];
+static DEFINE_SEMAPHORE(vc_cma_reserve_mutex);
+static DEFINE_SEMAPHORE(vc_cma_worker_queue_push_mutex);
+
+static u64 vc_cma_dma_mask = DMA_BIT_MASK(32);
+static struct platform_device vc_cma_device = {
+	.name = "vc-cma",
+	.id = 0,
+	.dev = {
+		.dma_mask = &vc_cma_dma_mask,
+		.coherent_dma_mask = DMA_BIT_MASK(32),
+		},
+};
+
+static VCHIQ_INSTANCE_T cma_instance;
+static VCHIQ_SERVICE_HANDLE_T cma_service;
+static VCHIU_QUEUE_T cma_msg_queue;
+static struct task_struct *cma_worker;
+
+static int vc_cma_set_reserve(unsigned int reserve, unsigned int pid);
+static int vc_cma_alloc_chunks(int num_chunks, struct cma_msg *reply);
+static VCHIQ_STATUS_T cma_service_callback(VCHIQ_REASON_T reason,
+					   VCHIQ_HEADER_T * header,
+					   VCHIQ_SERVICE_HANDLE_T service,
+					   void *bulk_userdata);
+static void send_vc_msg(unsigned short type,
+			unsigned short param1, unsigned short param2);
+static bool send_worker_msg(VCHIQ_HEADER_T * msg);
+
+static int early_vc_cma_mem(char *p)
+{
+	unsigned int new_size;
+	printk(KERN_NOTICE "early_vc_cma_mem(%s)", p);
+	vc_cma_size = memparse(p, &p);
+	vc_cma_initial = vc_cma_size;
+	if (*p == '/')
+		vc_cma_size = memparse(p + 1, &p);
+	if (*p == '@')
+		vc_cma_base = memparse(p + 1, &p);
+
+	new_size = (vc_cma_size - ((-vc_cma_base) & (VC_CMA_CHUNK_SIZE - 1)))
+	    & ~(VC_CMA_CHUNK_SIZE - 1);
+	if (new_size > vc_cma_size)
+		vc_cma_size = 0;
+	vc_cma_initial = (vc_cma_initial + VC_CMA_CHUNK_SIZE - 1)
+	    & ~(VC_CMA_CHUNK_SIZE - 1);
+	if (vc_cma_initial > vc_cma_size)
+		vc_cma_initial = vc_cma_size;
+	vc_cma_base = (vc_cma_base + VC_CMA_CHUNK_SIZE - 1)
+	    & ~(VC_CMA_CHUNK_SIZE - 1);
+
+	printk(KERN_NOTICE " -> initial %x, size %x, base %x", vc_cma_initial,
+	       vc_cma_size, (unsigned int)vc_cma_base);
+
+	return 0;
+}
+
+early_param("vc-cma-mem", early_vc_cma_mem);
+
+void vc_cma_early_init(void)
+{
+	LOG_DBG("vc_cma_early_init - vc_cma_chunks = %d", vc_cma_chunks);
+	if (vc_cma_size) {
+		int rc = platform_device_register(&vc_cma_device);
+		LOG_DBG("platform_device_register -> %d", rc);
+	}
+}
+
+void vc_cma_reserve(void)
+{
+	/* if vc_cma_size is set, then declare vc CMA area of the same
+	 * size from the end of memory
+	 */
+	if (vc_cma_size) {
+		if (dma_declare_contiguous(&vc_cma_device.dev, vc_cma_size,
+					   vc_cma_base, 0) == 0) {
+			if (!dev_get_cma_area(NULL)) {
+				/* There is no default CMA area - make this
+				   the default */
+				struct cma *vc_cma_area = dev_get_cma_area(
+					&vc_cma_device.dev);
+				dma_contiguous_set_default(vc_cma_area);
+				LOG_INFO("vc_cma_reserve - using vc_cma as "
+					 "the default contiguous DMA area");
+			}
+		} else {
+			LOG_ERR("vc_cma: dma_declare_contiguous(%x,%x) failed",
+				vc_cma_size, (unsigned int)vc_cma_base);
+			vc_cma_size = 0;
+		}
+	}
+	vc_cma_chunks = vc_cma_size / VC_CMA_CHUNK_SIZE;
+}
+
+/****************************************************************************
+*
+*   vc_cma_open
+*
+***************************************************************************/
+
+static int vc_cma_open(struct inode *inode, struct file *file)
+{
+	(void)inode;
+	(void)file;
+
+	return 0;
+}
+
+/****************************************************************************
+*
+*   vc_cma_release
+*
+***************************************************************************/
+
+static int vc_cma_release(struct inode *inode, struct file *file)
+{
+	(void)inode;
+	(void)file;
+
+	vc_cma_set_reserve(0, current->tgid);
+
+	return 0;
+}
+
+/****************************************************************************
+*
+*   vc_cma_ioctl
+*
+***************************************************************************/
+
+static long vc_cma_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+	int rc = 0;
+
+	(void)cmd;
+	(void)arg;
+
+	switch (cmd) {
+	case VC_CMA_IOC_RESERVE:
+		rc = vc_cma_set_reserve((unsigned int)arg, current->tgid);
+		if (rc >= 0)
+			rc = 0;
+		break;
+	default:
+		LOG_ERR("vc-cma: Unknown ioctl %x", cmd);
+		return -ENOTTY;
+	}
+
+	return rc;
+}
+
+/****************************************************************************
+*
+*   File Operations for the driver.
+*
+***************************************************************************/
+
+static const struct file_operations vc_cma_fops = {
+	.owner = THIS_MODULE,
+	.open = vc_cma_open,
+	.release = vc_cma_release,
+	.unlocked_ioctl = vc_cma_ioctl,
+};
+
+/****************************************************************************
+*
+*   vc_cma_proc_open
+*
+***************************************************************************/
+
+static int vc_cma_show_info(struct seq_file *m, void *v)
+{
+	int i;
+
+	seq_printf(m, "Videocore CMA:\n");
+	seq_printf(m, "   Base       : %08x\n", (unsigned int)vc_cma_base);
+	seq_printf(m, "   Length     : %08x\n", vc_cma_size);
+	seq_printf(m, "   Initial    : %08x\n", vc_cma_initial);
+	seq_printf(m, "   Chunk size : %08x\n", VC_CMA_CHUNK_SIZE);
+	seq_printf(m, "   Chunks     : %4d (%d bytes)\n",
+		   (int)vc_cma_chunks,
+		   (int)(vc_cma_chunks * VC_CMA_CHUNK_SIZE));
+	seq_printf(m, "   Used       : %4d (%d bytes)\n",
+		   (int)vc_cma_chunks_used,
+		   (int)(vc_cma_chunks_used * VC_CMA_CHUNK_SIZE));
+	seq_printf(m, "   Reserved   : %4d (%d bytes)\n",
+		   (unsigned int)vc_cma_chunks_reserved,
+		   (int)(vc_cma_chunks_reserved * VC_CMA_CHUNK_SIZE));
+
+	for (i = 0; i < vc_cma_reserve_count; i++) {
+		struct vc_cma_reserve_user *user = &vc_cma_reserve_users[i];
+		seq_printf(m, "     PID %5d: %d bytes\n", user->pid,
+			   user->reserve);
+	}
+	seq_printf(m, "   dma_alloc  : %p (%d pages)\n",
+		   vc_cma_dma_alloc ? page_address(vc_cma_dma_alloc) : 0,
+		   vc_cma_dma_size);
+
+	seq_printf(m, "\n");
+
+	return 0;
+}
+
+static int vc_cma_proc_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, vc_cma_show_info, NULL);
+}
+
+/****************************************************************************
+*
+*   vc_cma_proc_write
+*
+***************************************************************************/
+
+static int vc_cma_proc_write(struct file *file,
+			     const char __user *buffer,
+			     size_t size, loff_t *ppos)
+{
+	int rc = -EFAULT;
+	char input_str[20];
+
+	memset(input_str, 0, sizeof(input_str));
+
+	if (size > sizeof(input_str)) {
+		LOG_ERR("%s: input string length too long", __func__);
+		goto out;
+	}
+
+	if (copy_from_user(input_str, buffer, size - 1)) {
+		LOG_ERR("%s: failed to get input string", __func__);
+		goto out;
+	}
+#define ALLOC_STR "alloc"
+#define FREE_STR "free"
+#define DEBUG_STR "debug"
+#define RESERVE_STR "reserve"
+#define DMA_ALLOC_STR "dma_alloc"
+#define DMA_FREE_STR "dma_free"
+	if (strncmp(input_str, ALLOC_STR, strlen(ALLOC_STR)) == 0) {
+		int alloc_size;
+		char *p = input_str + strlen(ALLOC_STR);
+
+		while (*p == ' ')
+			p++;
+		alloc_size = memparse(p, NULL);
+		LOG_INFO("/proc/vc-cma: alloc %d", alloc_size);
+		if (alloc_size)
+			send_vc_msg(VC_CMA_MSG_REQUEST_FREE,
+				    alloc_size / VC_CMA_CHUNK_SIZE, 0);
+		else
+			LOG_ERR("invalid size '%s'", p);
+		rc = size;
+	} else if (strncmp(input_str, FREE_STR, strlen(FREE_STR)) == 0) {
+		int alloc_size;
+		char *p = input_str + strlen(FREE_STR);
+
+		while (*p == ' ')
+			p++;
+		alloc_size = memparse(p, NULL);
+		LOG_INFO("/proc/vc-cma: free %d", alloc_size);
+		if (alloc_size)
+			send_vc_msg(VC_CMA_MSG_REQUEST_ALLOC,
+				    alloc_size / VC_CMA_CHUNK_SIZE, 0);
+		else
+			LOG_ERR("invalid size '%s'", p);
+		rc = size;
+	} else if (strncmp(input_str, DEBUG_STR, strlen(DEBUG_STR)) == 0) {
+		char *p = input_str + strlen(DEBUG_STR);
+		while (*p == ' ')
+			p++;
+		if ((strcmp(p, "on") == 0) || (strcmp(p, "1") == 0))
+			vc_cma_debug = 1;
+		else if ((strcmp(p, "off") == 0) || (strcmp(p, "0") == 0))
+			vc_cma_debug = 0;
+		LOG_INFO("/proc/vc-cma: debug %s", vc_cma_debug ? "on" : "off");
+		rc = size;
+	} else if (strncmp(input_str, RESERVE_STR, strlen(RESERVE_STR)) == 0) {
+		int alloc_size;
+		int reserved;
+		char *p = input_str + strlen(RESERVE_STR);
+		while (*p == ' ')
+			p++;
+		alloc_size = memparse(p, NULL);
+
+		reserved = vc_cma_set_reserve(alloc_size, current->tgid);
+		rc = (reserved >= 0) ? size : reserved;
+	} else if (strncmp(input_str, DMA_ALLOC_STR, strlen(DMA_ALLOC_STR)) == 0) {
+		int alloc_size;
+		char *p = input_str + strlen(DMA_ALLOC_STR);
+		while (*p == ' ')
+			p++;
+		alloc_size = memparse(p, NULL);
+
+		if (vc_cma_dma_alloc) {
+		    dma_release_from_contiguous(NULL, vc_cma_dma_alloc,
+						vc_cma_dma_size);
+		    vc_cma_dma_alloc = NULL;
+		    vc_cma_dma_size = 0;
+		}
+		vc_cma_dma_alloc = dma_alloc_from_contiguous(NULL, alloc_size, 0);
+		vc_cma_dma_size = (vc_cma_dma_alloc ? alloc_size : 0);
+		if (vc_cma_dma_alloc)
+			LOG_INFO("dma_alloc(%d pages) -> %p", alloc_size, page_address(vc_cma_dma_alloc));
+		else
+			LOG_ERR("dma_alloc(%d pages) failed", alloc_size);
+		rc = size;
+	} else if (strncmp(input_str, DMA_FREE_STR, strlen(DMA_FREE_STR)) == 0) {
+		if (vc_cma_dma_alloc) {
+		    dma_release_from_contiguous(NULL, vc_cma_dma_alloc,
+						vc_cma_dma_size);
+		    vc_cma_dma_alloc = NULL;
+		    vc_cma_dma_size = 0;
+		}
+		rc = size;
+	}
+
+out:
+	return rc;
+}
+
+/****************************************************************************
+*
+*   File Operations for /proc interface.
+*
+***************************************************************************/
+
+static const struct file_operations vc_cma_proc_fops = {
+	.open = vc_cma_proc_open,
+	.read = seq_read,
+	.write = vc_cma_proc_write,
+	.llseek = seq_lseek,
+	.release = single_release
+};
+
+static int vc_cma_set_reserve(unsigned int reserve, unsigned int pid)
+{
+	struct vc_cma_reserve_user *user = NULL;
+	int delta = 0;
+	int i;
+
+	if (down_interruptible(&vc_cma_reserve_mutex))
+		return -ERESTARTSYS;
+
+	for (i = 0; i < vc_cma_reserve_count; i++) {
+		if (pid == vc_cma_reserve_users[i].pid) {
+			user = &vc_cma_reserve_users[i];
+			delta = reserve - user->reserve;
+			if (reserve)
+				user->reserve = reserve;
+			else {
+				/* Remove this entry by copying downwards */
+				while ((i + 1) < vc_cma_reserve_count) {
+					user[0].pid = user[1].pid;
+					user[0].reserve = user[1].reserve;
+					user++;
+					i++;
+				}
+				vc_cma_reserve_count--;
+				user = NULL;
+			}
+			break;
+		}
+	}
+
+	if (reserve && !user) {
+		if (vc_cma_reserve_count == VC_CMA_RESERVE_COUNT_MAX) {
+			LOG_ERR("vc-cma: Too many reservations - "
+				"increase CMA_RESERVE_COUNT_MAX");
+			up(&vc_cma_reserve_mutex);
+			return -EBUSY;
+		}
+		user = &vc_cma_reserve_users[vc_cma_reserve_count];
+		user->pid = pid;
+		user->reserve = reserve;
+		delta = reserve;
+		vc_cma_reserve_count++;
+	}
+
+	vc_cma_reserve_total += delta;
+
+	send_vc_msg(VC_CMA_MSG_RESERVE,
+		    vc_cma_reserve_total & 0xffff, vc_cma_reserve_total >> 16);
+
+	send_worker_msg((VCHIQ_HEADER_T *) VC_CMA_MSG_UPDATE_RESERVE);
+
+	LOG_DBG("/proc/vc-cma: reserve %d (PID %d) - total %u",
+		reserve, pid, vc_cma_reserve_total);
+
+	up(&vc_cma_reserve_mutex);
+
+	return vc_cma_reserve_total;
+}
+
+static VCHIQ_STATUS_T cma_service_callback(VCHIQ_REASON_T reason,
+					   VCHIQ_HEADER_T * header,
+					   VCHIQ_SERVICE_HANDLE_T service,
+					   void *bulk_userdata)
+{
+	switch (reason) {
+	case VCHIQ_MESSAGE_AVAILABLE:
+		if (!send_worker_msg(header))
+			return VCHIQ_RETRY;
+		break;
+	case VCHIQ_SERVICE_CLOSED:
+		LOG_DBG("CMA service closed");
+		break;
+	default:
+		LOG_ERR("Unexpected CMA callback reason %d", reason);
+		break;
+	}
+	return VCHIQ_SUCCESS;
+}
+
+static void send_vc_msg(unsigned short type,
+			unsigned short param1, unsigned short param2)
+{
+	unsigned short msg[] = { type, param1, param2 };
+	VCHIQ_ELEMENT_T elem = { &msg, sizeof(msg) };
+	VCHIQ_STATUS_T ret;
+	vchiq_use_service(cma_service);
+	ret = vchiq_queue_message(cma_service, &elem, 1);
+	vchiq_release_service(cma_service);
+	if (ret != VCHIQ_SUCCESS)
+		LOG_ERR("vchiq_queue_message returned %x", ret);
+}
+
+static bool send_worker_msg(VCHIQ_HEADER_T * msg)
+{
+	if (down_interruptible(&vc_cma_worker_queue_push_mutex))
+		return false;
+	vchiu_queue_push(&cma_msg_queue, msg);
+	up(&vc_cma_worker_queue_push_mutex);
+	return true;
+}
+
+static int vc_cma_alloc_chunks(int num_chunks, struct cma_msg *reply)
+{
+	int i;
+	for (i = 0; i < num_chunks; i++) {
+		struct page *chunk;
+		unsigned int chunk_num;
+		uint8_t *chunk_addr;
+		size_t chunk_size = PAGES_PER_CHUNK << PAGE_SHIFT;
+
+		chunk = dma_alloc_from_contiguous(&vc_cma_device.dev,
+						  PAGES_PER_CHUNK,
+						  VC_CMA_CHUNK_ORDER);
+		if (!chunk)
+			break;
+
+		chunk_addr = page_address(chunk);
+		dmac_flush_range(chunk_addr, chunk_addr + chunk_size);
+		outer_inv_range(__pa(chunk_addr), __pa(chunk_addr) +
+			chunk_size);
+
+		chunk_num =
+		    (page_to_phys(chunk) - vc_cma_base) / VC_CMA_CHUNK_SIZE;
+		BUG_ON(((page_to_phys(chunk) - vc_cma_base) %
+			VC_CMA_CHUNK_SIZE) != 0);
+		if (chunk_num >= vc_cma_chunks) {
+			phys_addr_t _pa = vc_cma_base + vc_cma_size - 1;
+			LOG_ERR("%s: ===============================",
+				__func__);
+			LOG_ERR("%s: chunk phys %x, vc_cma %pa-%pa - "
+				"bad SPARSEMEM configuration?",
+				__func__, (unsigned int)page_to_phys(chunk),
+				&vc_cma_base, &_pa);
+			LOG_ERR("%s: dev->cma_area = %p", __func__,
+				(void*)0/*vc_cma_device.dev.cma_area*/);
+			LOG_ERR("%s: ===============================",
+				__func__);
+			break;
+		}
+		reply->params[i] = chunk_num;
+		vc_cma_chunks_used++;
+	}
+
+	if (i < num_chunks) {
+		LOG_ERR("%s: dma_alloc_from_contiguous failed "
+			"for %x bytes (alloc %d of %d, %d free)",
+			__func__, VC_CMA_CHUNK_SIZE, i,
+			num_chunks, vc_cma_chunks - vc_cma_chunks_used);
+		num_chunks = i;
+	}
+
+	LOG_DBG("CMA allocated %d chunks -> %d used",
+		num_chunks, vc_cma_chunks_used);
+	reply->type = VC_CMA_MSG_ALLOCATED;
+
+	{
+		VCHIQ_ELEMENT_T elem = {
+			reply,
+			offsetof(struct cma_msg, params[0]) +
+			    num_chunks * sizeof(reply->params[0])
+		};
+		VCHIQ_STATUS_T ret;
+		vchiq_use_service(cma_service);
+		ret = vchiq_queue_message(cma_service, &elem, 1);
+		vchiq_release_service(cma_service);
+		if (ret != VCHIQ_SUCCESS)
+			LOG_ERR("vchiq_queue_message return " "%x", ret);
+	}
+
+	return num_chunks;
+}
+
+static int cma_worker_proc(void *param)
+{
+	static struct cma_msg reply;
+	(void)param;
+
+	while (1) {
+		VCHIQ_HEADER_T *msg;
+		static struct cma_msg msg_copy;
+		struct cma_msg *cma_msg = &msg_copy;
+		int type, msg_size;
+
+		msg = vchiu_queue_pop(&cma_msg_queue);
+		if ((unsigned int)msg >= VC_CMA_MSG_MAX) {
+			msg_size = msg->size;
+			memcpy(&msg_copy, msg->data, msg_size);
+			type = cma_msg->type;
+			vchiq_release_message(cma_service, msg);
+		} else {
+			msg_size = 0;
+			type = (int)msg;
+			if (type == VC_CMA_MSG_QUIT)
+				break;
+			else if (type == VC_CMA_MSG_UPDATE_RESERVE) {
+				msg = NULL;
+				cma_msg = NULL;
+			} else {
+				BUG();
+				continue;
+			}
+		}
+
+		switch (type) {
+		case VC_CMA_MSG_ALLOC:{
+				int num_chunks, free_chunks;
+				num_chunks = cma_msg->params[0];
+				free_chunks =
+				    vc_cma_chunks - vc_cma_chunks_used;
+				LOG_DBG("CMA_MSG_ALLOC(%d chunks)", num_chunks);
+				if (num_chunks > VC_CMA_MAX_PARAMS_PER_MSG) {
+					LOG_ERR
+					    ("CMA_MSG_ALLOC - chunk count (%d) "
+					     "exceeds VC_CMA_MAX_PARAMS_PER_MSG (%d)",
+					     num_chunks,
+					     VC_CMA_MAX_PARAMS_PER_MSG);
+					num_chunks = VC_CMA_MAX_PARAMS_PER_MSG;
+				}
+
+				if (num_chunks > free_chunks) {
+					LOG_ERR
+					    ("CMA_MSG_ALLOC - chunk count (%d) "
+					     "exceeds free chunks (%d)",
+					     num_chunks, free_chunks);
+					num_chunks = free_chunks;
+				}
+
+				vc_cma_alloc_chunks(num_chunks, &reply);
+			}
+			break;
+
+		case VC_CMA_MSG_FREE:{
+				int chunk_count =
+				    (msg_size -
+				     offsetof(struct cma_msg,
+					      params)) /
+				    sizeof(cma_msg->params[0]);
+				int i;
+				BUG_ON(chunk_count <= 0);
+
+				LOG_DBG("CMA_MSG_FREE(%d chunks - %x, ...)",
+					chunk_count, cma_msg->params[0]);
+				for (i = 0; i < chunk_count; i++) {
+					int chunk_num = cma_msg->params[i];
+					struct page *page = vc_cma_base_page +
+					    chunk_num * PAGES_PER_CHUNK;
+					if (chunk_num >= vc_cma_chunks) {
+						LOG_ERR
+						    ("CMA_MSG_FREE - chunk %d of %d"
+						     " (value %x) exceeds maximum "
+						     "(%x)", i, chunk_count,
+						     chunk_num,
+						     vc_cma_chunks - 1);
+						break;
+					}
+
+					if (!dma_release_from_contiguous
+					    (&vc_cma_device.dev, page,
+					     PAGES_PER_CHUNK)) {
+						phys_addr_t _pa = page_to_phys(page);
+						LOG_ERR
+						    ("CMA_MSG_FREE - failed to "
+						     "release chunk %d (phys %pa, "
+						     "page %x)", chunk_num,
+						     &_pa,
+						     (unsigned int)page);
+					}
+					vc_cma_chunks_used--;
+				}
+				LOG_DBG("CMA released %d chunks -> %d used",
+					i, vc_cma_chunks_used);
+			}
+			break;
+
+		case VC_CMA_MSG_UPDATE_RESERVE:{
+				int chunks_needed =
+				    ((vc_cma_reserve_total + VC_CMA_CHUNK_SIZE -
+				      1)
+				     / VC_CMA_CHUNK_SIZE) -
+				    vc_cma_chunks_reserved;
+
+				LOG_DBG
+				    ("CMA_MSG_UPDATE_RESERVE(%d chunks needed)",
+				     chunks_needed);
+
+				/* Cap the reservations to what is available */
+				if (chunks_needed > 0) {
+					if (chunks_needed >
+					    (vc_cma_chunks -
+					     vc_cma_chunks_used))
+						chunks_needed =
+						    (vc_cma_chunks -
+						     vc_cma_chunks_used);
+
+					chunks_needed =
+					    vc_cma_alloc_chunks(chunks_needed,
+								&reply);
+				}
+
+				LOG_DBG
+				    ("CMA_MSG_UPDATE_RESERVE(%d chunks allocated)",
+				     chunks_needed);
+				vc_cma_chunks_reserved += chunks_needed;
+			}
+			break;
+
+		default:
+			LOG_ERR("unexpected msg type %d", type);
+			break;
+		}
+	}
+
+	LOG_DBG("quitting...");
+	return 0;
+}
+
+/****************************************************************************
+*
+*   vc_cma_connected_init
+*
+*   This function is called once the videocore has been connected.
+*
+***************************************************************************/
+
+static void vc_cma_connected_init(void)
+{
+	VCHIQ_SERVICE_PARAMS_T service_params;
+
+	LOG_DBG("vc_cma_connected_init");
+
+	if (!vchiu_queue_init(&cma_msg_queue, 16)) {
+		LOG_ERR("could not create CMA msg queue");
+		goto fail_queue;
+	}
+
+	if (vchiq_initialise(&cma_instance) != VCHIQ_SUCCESS)
+		goto fail_vchiq_init;
+
+	vchiq_connect(cma_instance);
+
+	service_params.fourcc = VC_CMA_FOURCC;
+	service_params.callback = cma_service_callback;
+	service_params.userdata = NULL;
+	service_params.version = VC_CMA_VERSION;
+	service_params.version_min = VC_CMA_VERSION;
+
+	if (vchiq_open_service(cma_instance, &service_params,
+			       &cma_service) != VCHIQ_SUCCESS) {
+		LOG_ERR("failed to open service - already in use?");
+		goto fail_vchiq_open;
+	}
+
+	vchiq_release_service(cma_service);
+
+	cma_worker = kthread_create(cma_worker_proc, NULL, "cma_worker");
+	if (!cma_worker) {
+		LOG_ERR("could not create CMA worker thread");
+		goto fail_worker;
+	}
+	set_user_nice(cma_worker, -20);
+	wake_up_process(cma_worker);
+
+	return;
+
+fail_worker:
+	vchiq_close_service(cma_service);
+fail_vchiq_open:
+	vchiq_shutdown(cma_instance);
+fail_vchiq_init:
+	vchiu_queue_delete(&cma_msg_queue);
+fail_queue:
+	return;
+}
+
+void
+loud_error_header(void)
+{
+	if (in_loud_error)
+		return;
+
+	LOG_ERR("============================================================"
+		"================");
+	LOG_ERR("============================================================"
+		"================");
+	LOG_ERR("=====");
+
+	in_loud_error = 1;
+}
+
+void
+loud_error_footer(void)
+{
+	if (!in_loud_error)
+		return;
+
+	LOG_ERR("=====");
+	LOG_ERR("============================================================"
+		"================");
+	LOG_ERR("============================================================"
+		"================");
+
+	in_loud_error = 0;
+}
+
+#if 1
+static int check_cma_config(void) { return 1; }
+#else
+static int
+read_vc_debug_var(VC_MEM_ACCESS_HANDLE_T handle,
+	const char *symbol,
+	void *buf, size_t bufsize)
+{
+	VC_MEM_ADDR_T vcMemAddr;
+	size_t vcMemSize;
+	uint8_t *mapAddr;
+	off_t  vcMapAddr;
+
+	if (!LookupVideoCoreSymbol(handle, symbol,
+		&vcMemAddr,
+		&vcMemSize)) {
+		loud_error_header();
+		loud_error(
+			"failed to find VC symbol \"%s\".",
+			symbol);
+		loud_error_footer();
+		return 0;
+	}
+
+	if (vcMemSize != bufsize) {
+		loud_error_header();
+		loud_error(
+			"VC symbol \"%s\" is the wrong size.",
+			symbol);
+		loud_error_footer();
+		return 0;
+	}
+
+	vcMapAddr = (off_t)vcMemAddr & VC_MEM_TO_ARM_ADDR_MASK;
+	vcMapAddr += mm_vc_mem_phys_addr;
+	mapAddr = ioremap_nocache(vcMapAddr, vcMemSize);
+	if (mapAddr == 0) {
+		loud_error_header();
+		loud_error(
+			"failed to ioremap \"%s\" @ 0x%x "
+			"(phys: 0x%x, size: %u).",
+			symbol,
+			(unsigned int)vcMapAddr,
+			(unsigned int)vcMemAddr,
+			(unsigned int)vcMemSize);
+		loud_error_footer();
+		return 0;
+	}
+
+	memcpy(buf, mapAddr, bufsize);
+	iounmap(mapAddr);
+
+	return 1;
+}
+
+
+static int
+check_cma_config(void)
+{
+	VC_MEM_ACCESS_HANDLE_T mem_hndl;
+	VC_MEM_ADDR_T mempool_start;
+	VC_MEM_ADDR_T mempool_end;
+	VC_MEM_ADDR_T mempool_offline_start;
+	VC_MEM_ADDR_T mempool_offline_end;
+	VC_MEM_ADDR_T cam_alloc_base;
+	VC_MEM_ADDR_T cam_alloc_size;
+	VC_MEM_ADDR_T cam_alloc_end;
+	int success = 0;
+
+	if (OpenVideoCoreMemory(&mem_hndl) != 0)
+		goto out;
+
+	/* Read the relevant VideoCore variables */
+	if (!read_vc_debug_var(mem_hndl, "__MEMPOOL_START",
+		&mempool_start,
+		sizeof(mempool_start)))
+		goto close;
+
+	if (!read_vc_debug_var(mem_hndl, "__MEMPOOL_END",
+		&mempool_end,
+		sizeof(mempool_end)))
+		goto close;
+
+	if (!read_vc_debug_var(mem_hndl, "__MEMPOOL_OFFLINE_START",
+		&mempool_offline_start,
+		sizeof(mempool_offline_start)))
+		goto close;
+
+	if (!read_vc_debug_var(mem_hndl, "__MEMPOOL_OFFLINE_END",
+		&mempool_offline_end,
+		sizeof(mempool_offline_end)))
+		goto close;
+
+	if (!read_vc_debug_var(mem_hndl, "cam_alloc_base",
+		&cam_alloc_base,
+		sizeof(cam_alloc_base)))
+		goto close;
+
+	if (!read_vc_debug_var(mem_hndl, "cam_alloc_size",
+		&cam_alloc_size,
+		sizeof(cam_alloc_size)))
+		goto close;
+
+	cam_alloc_end = cam_alloc_base + cam_alloc_size;
+
+	success = 1;
+
+	/* Now the sanity checks */
+	if (!mempool_offline_start)
+		mempool_offline_start = mempool_start;
+	if (!mempool_offline_end)
+		mempool_offline_end = mempool_end;
+
+	if (VCADDR_TO_PHYSADDR(mempool_offline_start) != vc_cma_base) {
+		loud_error_header();
+		loud_error(
+			"__MEMPOOL_OFFLINE_START(%x -> %lx) doesn't match "
+			"vc_cma_base(%x)",
+			mempool_offline_start,
+			VCADDR_TO_PHYSADDR(mempool_offline_start),
+			vc_cma_base);
+		success = 0;
+	}
+
+	if (VCADDR_TO_PHYSADDR(mempool_offline_end) !=
+		(vc_cma_base + vc_cma_size)) {
+		loud_error_header();
+		loud_error(
+			"__MEMPOOL_OFFLINE_END(%x -> %lx) doesn't match "
+			"vc_cma_base(%x) + vc_cma_size(%x) = %x",
+			mempool_offline_start,
+			VCADDR_TO_PHYSADDR(mempool_offline_end),
+			vc_cma_base, vc_cma_size, vc_cma_base + vc_cma_size);
+		success = 0;
+	}
+
+	if (mempool_end < mempool_start) {
+		loud_error_header();
+		loud_error(
+			"__MEMPOOL_END(%x) must not be before "
+			"__MEMPOOL_START(%x)",
+			mempool_end,
+			mempool_start);
+		success = 0;
+	}
+
+	if (mempool_offline_end < mempool_offline_start) {
+		loud_error_header();
+		loud_error(
+			"__MEMPOOL_OFFLINE_END(%x) must not be before "
+			"__MEMPOOL_OFFLINE_START(%x)",
+			mempool_offline_end,
+			mempool_offline_start);
+		success = 0;
+	}
+
+	if (mempool_offline_start < mempool_start) {
+		loud_error_header();
+		loud_error(
+			"__MEMPOOL_OFFLINE_START(%x) must not be before "
+			"__MEMPOOL_START(%x)",
+			mempool_offline_start,
+			mempool_start);
+		success = 0;
+	}
+
+	if (mempool_offline_end > mempool_end) {
+		loud_error_header();
+		loud_error(
+			"__MEMPOOL_OFFLINE_END(%x) must not be after "
+			"__MEMPOOL_END(%x)",
+			mempool_offline_end,
+			mempool_end);
+		success = 0;
+	}
+
+	if ((cam_alloc_base < mempool_end) &&
+		(cam_alloc_end > mempool_start)) {
+		loud_error_header();
+		loud_error(
+			"cam_alloc pool(%x-%x) overlaps "
+			"mempool(%x-%x)",
+			cam_alloc_base, cam_alloc_end,
+			mempool_start, mempool_end);
+		success = 0;
+	}
+
+	loud_error_footer();
+
+close:
+	CloseVideoCoreMemory(mem_hndl);
+
+out:
+	return success;
+}
+#endif
+
+static int vc_cma_init(void)
+{
+	int rc = -EFAULT;
+	struct device *dev;
+
+	if (!check_cma_config())
+		goto out_release;
+
+	LOG_INFO("vc-cma: Videocore CMA driver");
+	LOG_INFO("vc-cma: vc_cma_base      = %pa", &vc_cma_base);
+	LOG_INFO("vc-cma: vc_cma_size      = 0x%08x (%u MiB)",
+		 vc_cma_size, vc_cma_size / (1024 * 1024));
+	LOG_INFO("vc-cma: vc_cma_initial   = 0x%08x (%u MiB)",
+		 vc_cma_initial, vc_cma_initial / (1024 * 1024));
+
+	vc_cma_base_page = phys_to_page(vc_cma_base);
+
+	if (vc_cma_chunks) {
+		int chunks_needed = vc_cma_initial / VC_CMA_CHUNK_SIZE;
+
+		for (vc_cma_chunks_used = 0;
+		     vc_cma_chunks_used < chunks_needed; vc_cma_chunks_used++) {
+			struct page *chunk;
+			chunk = dma_alloc_from_contiguous(&vc_cma_device.dev,
+							  PAGES_PER_CHUNK,
+							  VC_CMA_CHUNK_ORDER);
+			if (!chunk)
+				break;
+			BUG_ON(((page_to_phys(chunk) - vc_cma_base) %
+				VC_CMA_CHUNK_SIZE) != 0);
+		}
+		if (vc_cma_chunks_used != chunks_needed) {
+			LOG_ERR("%s: dma_alloc_from_contiguous failed (%d "
+				"bytes, allocation %d of %d)",
+				__func__, VC_CMA_CHUNK_SIZE,
+				vc_cma_chunks_used, chunks_needed);
+			goto out_release;
+		}
+
+		vchiq_add_connected_callback(vc_cma_connected_init);
+	}
+
+	rc = alloc_chrdev_region(&vc_cma_devnum, 0, 1, DRIVER_NAME);
+	if (rc < 0) {
+		LOG_ERR("%s: alloc_chrdev_region failed (rc=%d)", __func__, rc);
+		goto out_release;
+	}
+
+	cdev_init(&vc_cma_cdev, &vc_cma_fops);
+	rc = cdev_add(&vc_cma_cdev, vc_cma_devnum, 1);
+	if (rc != 0) {
+		LOG_ERR("%s: cdev_add failed (rc=%d)", __func__, rc);
+		goto out_unregister;
+	}
+
+	vc_cma_class = class_create(THIS_MODULE, DRIVER_NAME);
+	if (IS_ERR(vc_cma_class)) {
+		rc = PTR_ERR(vc_cma_class);
+		LOG_ERR("%s: class_create failed (rc=%d)", __func__, rc);
+		goto out_cdev_del;
+	}
+
+	dev = device_create(vc_cma_class, NULL, vc_cma_devnum, NULL,
+			    DRIVER_NAME);
+	if (IS_ERR(dev)) {
+		rc = PTR_ERR(dev);
+		LOG_ERR("%s: device_create failed (rc=%d)", __func__, rc);
+		goto out_class_destroy;
+	}
+
+	vc_cma_proc_entry = proc_create(DRIVER_NAME, 0444, NULL, &vc_cma_proc_fops);
+	if (vc_cma_proc_entry == NULL) {
+		rc = -EFAULT;
+		LOG_ERR("%s: proc_create failed", __func__);
+		goto out_device_destroy;
+	}
+
+	vc_cma_inited = 1;
+	return 0;
+
+out_device_destroy:
+	device_destroy(vc_cma_class, vc_cma_devnum);
+
+out_class_destroy:
+	class_destroy(vc_cma_class);
+	vc_cma_class = NULL;
+
+out_cdev_del:
+	cdev_del(&vc_cma_cdev);
+
+out_unregister:
+	unregister_chrdev_region(vc_cma_devnum, 1);
+
+out_release:
+	/* It is tempting to try to clean up by calling
+	   dma_release_from_contiguous for all allocated chunks, but it isn't
+	   a very safe thing to do. If vc_cma_initial is non-zero it is because
+	   VideoCore is already using that memory, so giving it back to Linux
+	   is likely to be fatal.
+	 */
+	return -1;
+}
+
+/****************************************************************************
+*
+*   vc_cma_exit
+*
+***************************************************************************/
+
+static void __exit vc_cma_exit(void)
+{
+	LOG_DBG("%s: called", __func__);
+
+	if (vc_cma_inited) {
+		remove_proc_entry(DRIVER_NAME, NULL);
+		device_destroy(vc_cma_class, vc_cma_devnum);
+		class_destroy(vc_cma_class);
+		cdev_del(&vc_cma_cdev);
+		unregister_chrdev_region(vc_cma_devnum, 1);
+	}
+}
+
+module_init(vc_cma_init);
+module_exit(vc_cma_exit);
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Broadcom Corporation");
--- /dev/null
+++ b/include/linux/broadcom/vc_cma.h
@@ -0,0 +1,29 @@
+/*****************************************************************************
+* Copyright 2012 Broadcom Corporation.  All rights reserved.
+*
+* Unless you and Broadcom execute a separate written software license
+* agreement governing use of this software, this software is licensed to you
+* under the terms of the GNU General Public License version 2, available at
+* http://www.broadcom.com/licenses/GPLv2.php (the "GPL").
+*
+* Notwithstanding the above, under no circumstances may you combine this
+* software in any way with any other Broadcom software provided under a
+* license other than the GPL, without Broadcom's express prior written
+* consent.
+*****************************************************************************/
+
+#if !defined( VC_CMA_H )
+#define VC_CMA_H
+
+#include <linux/ioctl.h>
+
+#define VC_CMA_IOC_MAGIC 0xc5
+
+#define VC_CMA_IOC_RESERVE _IO(VC_CMA_IOC_MAGIC, 0)
+
+#ifdef __KERNEL__
+extern void __init vc_cma_early_init(void);
+extern void __init vc_cma_reserve(void);
+#endif
+
+#endif /* VC_CMA_H */