]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
HSI: cmt_speech: Add cmt-speech driver
authorKai Vehmanen <kai.vehmanen@nokia.com>
Wed, 2 Jun 2010 19:23:34 +0000 (22:23 +0300)
committerSebastian Reichel <sre@kernel.org>
Tue, 31 Mar 2015 20:14:04 +0000 (22:14 +0200)
Introduces the cmt-speech driver, which implements
a character device interface for transferring speech
data frames over HSI/SSI.

The driver is used to exchange voice/speech data between
the Nokia N900/N950/N9's modem and its cpu.

Signed-off-by: Kai Vehmanen <kai.vehmanen@nokia.com>
Signed-off-by: Carlos Chinea <carlos.chinea@nokia.com>
Signed-off-by: Joni Lapilainen <joni.lapilainen@gmail.com>
Since the original driver has been written for 2.6.28 some
build fixes and general cleanups have been added by me:

 * fix build for 4.0 kernel
 * replace GFP_ATOMIC with GFP_KERNEL in cs_alloc_cmds()
 * add sanity check for CS_SET_WAKELINE ioctl
 * cleanup driver initialisation
 * rename driver to cmt-speech to be consistent with
   ssi-protocol driver
 * move cs-protocol.h to include/uapi/linux/hsi, since
   it describes a userspace API
 * replace hardcoded channels numbers with values provided
   via the HSI framework (e.g. coming from DT)

Acked-by: Aaro Koskinen <aaro.koskinen@iki.fi>
Tested-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Sebastian Reichel <sre@kernel.org>
drivers/hsi/clients/Kconfig
drivers/hsi/clients/Makefile
drivers/hsi/clients/cmt_speech.c [new file with mode: 0644]
include/uapi/linux/hsi/Kbuild
include/uapi/linux/hsi/cs-protocol.h [new file with mode: 0644]

index bc60dec3f586fec38c995af8300506a5f7caf5b5..86c849506f3433f76b7bd7ca5d98ed9cb10b1e43 100644 (file)
@@ -13,6 +13,16 @@ config NOKIA_MODEM
 
        If unsure, say N.
 
+config CMT_SPEECH
+       tristate "CMT speech"
+       depends on HSI && SSI_PROTOCOL
+       help
+       If you say Y here, you will enable the CMT speech protocol used
+       by Nokia modems. If you say M the protocol will be available as
+       module named cmt_speech.
+
+       If unsure, say N.
+
 config SSI_PROTOCOL
        tristate "SSI protocol"
        depends on HSI && PHONET && OMAP_SSI
index 4d5bc0e0b27b814b53dd076715d0cdfdb04f7d96..26072326640702152d6c383cc4aaf068cb6080f1 100644 (file)
@@ -4,4 +4,5 @@
 
 obj-$(CONFIG_NOKIA_MODEM)      += nokia-modem.o
 obj-$(CONFIG_SSI_PROTOCOL)     += ssi_protocol.o
+obj-$(CONFIG_CMT_SPEECH)       += cmt_speech.o
 obj-$(CONFIG_HSI_CHAR)         += hsi_char.o
diff --git a/drivers/hsi/clients/cmt_speech.c b/drivers/hsi/clients/cmt_speech.c
new file mode 100644 (file)
index 0000000..e9560ef
--- /dev/null
@@ -0,0 +1,1456 @@
+/*
+ * cmt_speech.c - HSI CMT speech driver
+ *
+ * Copyright (C) 2008,2009,2010 Nokia Corporation. All rights reserved.
+ *
+ * Contact: Kai Vehmanen <kai.vehmanen@nokia.com>
+ * Original author: Peter Ujfalusi <peter.ujfalusi@nokia.com>
+ *
+ * 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.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA
+ * 02110-1301 USA
+ */
+
+#include <linux/errno.h>
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/init.h>
+#include <linux/device.h>
+#include <linux/miscdevice.h>
+#include <linux/mm.h>
+#include <linux/slab.h>
+#include <linux/fs.h>
+#include <linux/poll.h>
+#include <linux/sched.h>
+#include <linux/ioctl.h>
+#include <linux/uaccess.h>
+#include <linux/pm_qos.h>
+#include <linux/hsi/hsi.h>
+#include <linux/hsi/ssi_protocol.h>
+#include <linux/hsi/cs-protocol.h>
+
+#define CS_MMAP_SIZE   PAGE_SIZE
+
+struct char_queue {
+       struct list_head        list;
+       u32                     msg;
+};
+
+struct cs_char {
+       unsigned int            opened;
+       struct hsi_client       *cl;
+       struct cs_hsi_iface     *hi;
+       struct list_head        chardev_queue;
+       struct list_head        dataind_queue;
+       int                     dataind_pending;
+       /* mmap things */
+       unsigned long           mmap_base;
+       unsigned long           mmap_size;
+       spinlock_t              lock;
+       struct fasync_struct    *async_queue;
+       wait_queue_head_t       wait;
+       /* hsi channel ids */
+       int                     channel_id_cmd;
+       int                     channel_id_data;
+};
+
+#define SSI_CHANNEL_STATE_READING      1
+#define SSI_CHANNEL_STATE_WRITING      (1 << 1)
+#define SSI_CHANNEL_STATE_POLL         (1 << 2)
+#define SSI_CHANNEL_STATE_ERROR                (1 << 3)
+
+#define TARGET_MASK                    0xf000000
+#define TARGET_REMOTE                  (1 << CS_DOMAIN_SHIFT)
+#define TARGET_LOCAL                   0
+
+/* Number of pre-allocated commands buffers */
+#define CS_MAX_CMDS                    4
+
+/*
+ * During data transfers, transactions must be handled
+ * within 20ms (fixed value in cmtspeech HSI protocol)
+ */
+#define CS_QOS_LATENCY_FOR_DATA_USEC   20000
+
+/* Timeout to wait for pending HSI transfers to complete */
+#define CS_HSI_TRANSFER_TIMEOUT_MS      500
+
+
+#define RX_PTR_BOUNDARY_SHIFT          8
+#define RX_PTR_MAX_SHIFT               (RX_PTR_BOUNDARY_SHIFT + \
+                                               CS_MAX_BUFFERS_SHIFT)
+struct cs_hsi_iface {
+       struct hsi_client               *cl;
+       struct hsi_client               *master;
+
+       unsigned int                    iface_state;
+       unsigned int                    wakeline_state;
+       unsigned int                    control_state;
+       unsigned int                    data_state;
+
+       /* state exposed to application */
+       struct cs_mmap_config_block     *mmap_cfg;
+
+       unsigned long                   mmap_base;
+       unsigned long                   mmap_size;
+
+       unsigned int                    rx_slot;
+       unsigned int                    tx_slot;
+
+       /* note: for security reasons, we do not trust the contents of
+        * mmap_cfg, but instead duplicate the variables here */
+       unsigned int                    buf_size;
+       unsigned int                    rx_bufs;
+       unsigned int                    tx_bufs;
+       unsigned int                    rx_ptr_boundary;
+       unsigned int                    rx_offsets[CS_MAX_BUFFERS];
+       unsigned int                    tx_offsets[CS_MAX_BUFFERS];
+
+       /* size of aligned memory blocks */
+       unsigned int                    slot_size;
+       unsigned int                    flags;
+
+       struct list_head                cmdqueue;
+
+       struct hsi_msg                  *data_rx_msg;
+       struct hsi_msg                  *data_tx_msg;
+       wait_queue_head_t               datawait;
+
+       struct pm_qos_request           pm_qos_req;
+
+       spinlock_t                      lock;
+};
+
+static struct cs_char cs_char_data;
+
+static void cs_hsi_read_on_control(struct cs_hsi_iface *hi);
+static void cs_hsi_read_on_data(struct cs_hsi_iface *hi);
+
+static inline void rx_ptr_shift_too_big(void)
+{
+       BUILD_BUG_ON((1LLU << RX_PTR_MAX_SHIFT) > UINT_MAX);
+}
+
+static void cs_notify(u32 message, struct list_head *head)
+{
+       struct char_queue *entry;
+
+       spin_lock(&cs_char_data.lock);
+
+       if (!cs_char_data.opened) {
+               spin_unlock(&cs_char_data.lock);
+               goto out;
+       }
+
+       entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
+       if (!entry) {
+               dev_err(&cs_char_data.cl->device,
+                       "Can't allocate new entry for the queue.\n");
+               spin_unlock(&cs_char_data.lock);
+               goto out;
+       }
+
+       entry->msg = message;
+       list_add_tail(&entry->list, head);
+
+       spin_unlock(&cs_char_data.lock);
+
+       wake_up_interruptible(&cs_char_data.wait);
+       kill_fasync(&cs_char_data.async_queue, SIGIO, POLL_IN);
+
+out:
+       return;
+}
+
+static u32 cs_pop_entry(struct list_head *head)
+{
+       struct char_queue *entry;
+       u32 data;
+
+       entry = list_entry(head->next, struct char_queue, list);
+       data = entry->msg;
+       list_del(&entry->list);
+       kfree(entry);
+
+       return data;
+}
+
+static void cs_notify_control(u32 message)
+{
+       cs_notify(message, &cs_char_data.chardev_queue);
+}
+
+static void cs_notify_data(u32 message, int maxlength)
+{
+       cs_notify(message, &cs_char_data.dataind_queue);
+
+       spin_lock(&cs_char_data.lock);
+       cs_char_data.dataind_pending++;
+       while (cs_char_data.dataind_pending > maxlength &&
+                               !list_empty(&cs_char_data.dataind_queue)) {
+               dev_dbg(&cs_char_data.cl->device, "data notification "
+               "queue overrun (%u entries)\n", cs_char_data.dataind_pending);
+
+               cs_pop_entry(&cs_char_data.dataind_queue);
+               cs_char_data.dataind_pending--;
+       }
+       spin_unlock(&cs_char_data.lock);
+}
+
+static inline void cs_set_cmd(struct hsi_msg *msg, u32 cmd)
+{
+       u32 *data = sg_virt(msg->sgt.sgl);
+       *data = cmd;
+}
+
+static inline u32 cs_get_cmd(struct hsi_msg *msg)
+{
+       u32 *data = sg_virt(msg->sgt.sgl);
+       return *data;
+}
+
+static void cs_release_cmd(struct hsi_msg *msg)
+{
+       struct cs_hsi_iface *hi = msg->context;
+
+       list_add_tail(&msg->link, &hi->cmdqueue);
+}
+
+static void cs_cmd_destructor(struct hsi_msg *msg)
+{
+       struct cs_hsi_iface *hi = msg->context;
+
+       spin_lock(&hi->lock);
+
+       dev_dbg(&cs_char_data.cl->device, "control cmd destructor\n");
+
+       if (hi->iface_state != CS_STATE_CLOSED)
+               dev_err(&hi->cl->device, "Cmd flushed while driver active\n");
+
+       if (msg->ttype == HSI_MSG_READ)
+               hi->control_state &=
+                       ~(SSI_CHANNEL_STATE_POLL | SSI_CHANNEL_STATE_READING);
+       else if (msg->ttype == HSI_MSG_WRITE &&
+                       hi->control_state & SSI_CHANNEL_STATE_WRITING)
+               hi->control_state &= ~SSI_CHANNEL_STATE_WRITING;
+
+       cs_release_cmd(msg);
+
+       spin_unlock(&hi->lock);
+}
+
+static struct hsi_msg *cs_claim_cmd(struct cs_hsi_iface* ssi)
+{
+       struct hsi_msg *msg;
+
+       BUG_ON(list_empty(&ssi->cmdqueue));
+
+       msg = list_first_entry(&ssi->cmdqueue, struct hsi_msg, link);
+       list_del(&msg->link);
+       msg->destructor = cs_cmd_destructor;
+
+       return msg;
+}
+
+static void cs_free_cmds(struct cs_hsi_iface *ssi)
+{
+       struct hsi_msg *msg, *tmp;
+
+       list_for_each_entry_safe(msg, tmp, &ssi->cmdqueue, link) {
+               list_del(&msg->link);
+               msg->destructor = NULL;
+               kfree(sg_virt(msg->sgt.sgl));
+               hsi_free_msg(msg);
+       }
+}
+
+static int cs_alloc_cmds(struct cs_hsi_iface *hi)
+{
+       struct hsi_msg *msg;
+       u32 *buf;
+       unsigned int i;
+
+       INIT_LIST_HEAD(&hi->cmdqueue);
+
+       for (i = 0; i < CS_MAX_CMDS; i++) {
+               msg = hsi_alloc_msg(1, GFP_KERNEL);
+               if (!msg)
+                       goto out;
+               buf = kmalloc(sizeof(*buf), GFP_KERNEL);
+               if (!buf) {
+                       hsi_free_msg(msg);
+                       goto out;
+               }
+               sg_init_one(msg->sgt.sgl, buf, sizeof(*buf));
+               msg->channel = cs_char_data.channel_id_cmd;
+               msg->context = hi;
+               list_add_tail(&msg->link, &hi->cmdqueue);
+       }
+
+       return 0;
+
+out:
+       cs_free_cmds(hi);
+       return -ENOMEM;
+}
+
+static void cs_hsi_data_destructor(struct hsi_msg *msg)
+{
+       struct cs_hsi_iface *hi = msg->context;
+       const char *dir = (msg->ttype == HSI_MSG_READ) ? "TX" : "RX";
+
+       dev_dbg(&cs_char_data.cl->device, "Freeing data %s message\n", dir);
+
+       spin_lock(&hi->lock);
+       if (hi->iface_state != CS_STATE_CLOSED)
+               dev_err(&cs_char_data.cl->device,
+                               "Data %s flush while device active\n", dir);
+       if (msg->ttype == HSI_MSG_READ)
+               hi->data_state &=
+                       ~(SSI_CHANNEL_STATE_POLL | SSI_CHANNEL_STATE_READING);
+       else
+               hi->data_state &= ~SSI_CHANNEL_STATE_WRITING;
+
+       msg->status = HSI_STATUS_COMPLETED;
+       if (unlikely(waitqueue_active(&hi->datawait)))
+               wake_up_interruptible(&hi->datawait);
+
+       spin_unlock(&hi->lock);
+}
+
+static int cs_hsi_alloc_data(struct cs_hsi_iface *hi)
+{
+       struct hsi_msg *txmsg, *rxmsg;
+       int res = 0;
+
+       rxmsg = hsi_alloc_msg(1, GFP_KERNEL);
+       if (!rxmsg) {
+               res = -ENOMEM;
+               goto out1;
+       }
+       rxmsg->channel = cs_char_data.channel_id_data;
+       rxmsg->destructor = cs_hsi_data_destructor;
+       rxmsg->context = hi;
+
+       txmsg = hsi_alloc_msg(1, GFP_KERNEL);
+       if (!txmsg) {
+               res = -ENOMEM;
+               goto out2;
+       }
+       txmsg->channel = cs_char_data.channel_id_data;
+       txmsg->destructor = cs_hsi_data_destructor;
+       txmsg->context = hi;
+
+       hi->data_rx_msg = rxmsg;
+       hi->data_tx_msg = txmsg;
+
+       return 0;
+
+out2:
+       hsi_free_msg(rxmsg);
+out1:
+       return res;
+}
+
+static void cs_hsi_free_data_msg(struct hsi_msg *msg)
+{
+       WARN_ON(msg->status != HSI_STATUS_COMPLETED &&
+                                       msg->status != HSI_STATUS_ERROR);
+       hsi_free_msg(msg);
+}
+
+static void cs_hsi_free_data(struct cs_hsi_iface *hi)
+{
+       cs_hsi_free_data_msg(hi->data_rx_msg);
+       cs_hsi_free_data_msg(hi->data_tx_msg);
+}
+
+static inline void __cs_hsi_error_pre(struct cs_hsi_iface *hi,
+                                       struct hsi_msg *msg, const char *info,
+                                       unsigned int *state)
+{
+       spin_lock(&hi->lock);
+       dev_err(&hi->cl->device, "HSI %s error, msg %d, state %u\n",
+               info, msg->status, *state);
+}
+
+static inline void __cs_hsi_error_post(struct cs_hsi_iface *hi)
+{
+       spin_unlock(&hi->lock);
+}
+
+static inline void __cs_hsi_error_read_bits(unsigned int *state)
+{
+       *state |= SSI_CHANNEL_STATE_ERROR;
+       *state &= ~(SSI_CHANNEL_STATE_READING | SSI_CHANNEL_STATE_POLL);
+}
+
+static inline void __cs_hsi_error_write_bits(unsigned int *state)
+{
+       *state |= SSI_CHANNEL_STATE_ERROR;
+       *state &= ~SSI_CHANNEL_STATE_WRITING;
+}
+
+static void cs_hsi_control_read_error(struct cs_hsi_iface *hi,
+                                                       struct hsi_msg *msg)
+{
+       __cs_hsi_error_pre(hi, msg, "control read", &hi->control_state);
+       cs_release_cmd(msg);
+       __cs_hsi_error_read_bits(&hi->control_state);
+       __cs_hsi_error_post(hi);
+}
+
+static void cs_hsi_control_write_error(struct cs_hsi_iface *hi,
+                                                       struct hsi_msg *msg)
+{
+       __cs_hsi_error_pre(hi, msg, "control write", &hi->control_state);
+       cs_release_cmd(msg);
+       __cs_hsi_error_write_bits(&hi->control_state);
+       __cs_hsi_error_post(hi);
+
+}
+
+static void cs_hsi_data_read_error(struct cs_hsi_iface *hi, struct hsi_msg *msg)
+{
+       __cs_hsi_error_pre(hi, msg, "data read", &hi->data_state);
+       __cs_hsi_error_read_bits(&hi->data_state);
+       __cs_hsi_error_post(hi);
+}
+
+static void cs_hsi_data_write_error(struct cs_hsi_iface *hi,
+                                                       struct hsi_msg *msg)
+{
+       __cs_hsi_error_pre(hi, msg, "data write", &hi->data_state);
+       __cs_hsi_error_write_bits(&hi->data_state);
+       __cs_hsi_error_post(hi);
+}
+
+static void cs_hsi_read_on_control_complete(struct hsi_msg *msg)
+{
+       u32 cmd = cs_get_cmd(msg);
+       struct cs_hsi_iface *hi = msg->context;
+
+       spin_lock(&hi->lock);
+       hi->control_state &= ~SSI_CHANNEL_STATE_READING;
+       if (msg->status == HSI_STATUS_ERROR) {
+               dev_err(&hi->cl->device, "Control RX error detected\n");
+               cs_hsi_control_read_error(hi, msg);
+               spin_unlock(&hi->lock);
+               goto out;
+       }
+       dev_dbg(&hi->cl->device, "Read on control: %08X\n", cmd);
+       cs_release_cmd(msg);
+       if (hi->flags & CS_FEAT_TSTAMP_RX_CTRL) {
+               struct timespec *tstamp =
+                       &hi->mmap_cfg->tstamp_rx_ctrl;
+               do_posix_clock_monotonic_gettime(tstamp);
+       }
+       spin_unlock(&hi->lock);
+
+       cs_notify_control(cmd);
+
+out:
+       cs_hsi_read_on_control(hi);
+}
+
+static void cs_hsi_peek_on_control_complete(struct hsi_msg *msg)
+{
+       struct cs_hsi_iface *hi = msg->context;
+       int ret;
+
+       if (msg->status == HSI_STATUS_ERROR) {
+               dev_err(&hi->cl->device, "Control peek RX error detected\n");
+               cs_hsi_control_read_error(hi, msg);
+               return;
+       }
+
+       WARN_ON(!(hi->control_state & SSI_CHANNEL_STATE_READING));
+
+       dev_dbg(&hi->cl->device, "Peek on control complete, reading\n");
+       msg->sgt.nents = 1;
+       msg->complete = cs_hsi_read_on_control_complete;
+       ret = hsi_async_read(hi->cl, msg);
+       if (ret)
+               cs_hsi_control_read_error(hi, msg);
+}
+
+static void cs_hsi_read_on_control(struct cs_hsi_iface *hi)
+{
+       struct hsi_msg *msg;
+       int ret;
+
+       spin_lock(&hi->lock);
+       if (hi->control_state & SSI_CHANNEL_STATE_READING) {
+               dev_err(&hi->cl->device, "Control read already pending (%d)\n",
+                       hi->control_state);
+               spin_unlock(&hi->lock);
+               return;
+       }
+       if (hi->control_state & SSI_CHANNEL_STATE_ERROR) {
+               dev_err(&hi->cl->device, "Control read error (%d)\n",
+                       hi->control_state);
+               spin_unlock(&hi->lock);
+               return;
+       }
+       hi->control_state |= SSI_CHANNEL_STATE_READING;
+       dev_dbg(&hi->cl->device, "Issuing RX on control\n");
+       msg = cs_claim_cmd(hi);
+       spin_unlock(&hi->lock);
+
+       msg->sgt.nents = 0;
+       msg->complete = cs_hsi_peek_on_control_complete;
+       ret = hsi_async_read(hi->cl, msg);
+       if (ret)
+               cs_hsi_control_read_error(hi, msg);
+}
+
+static void cs_hsi_write_on_control_complete(struct hsi_msg *msg)
+{
+       struct cs_hsi_iface *hi = msg->context;
+       if (msg->status == HSI_STATUS_COMPLETED) {
+               spin_lock(&hi->lock);
+               hi->control_state &= ~SSI_CHANNEL_STATE_WRITING;
+               cs_release_cmd(msg);
+               spin_unlock(&hi->lock);
+       } else if (msg->status == HSI_STATUS_ERROR) {
+               cs_hsi_control_write_error(hi, msg);
+       } else {
+               dev_err(&hi->cl->device,
+                       "unexpected status in control write callback %d\n",
+                       msg->status);
+       }
+}
+
+static int cs_hsi_write_on_control(struct cs_hsi_iface *hi, u32 message)
+{
+       struct hsi_msg *msg;
+       int ret;
+
+       spin_lock(&hi->lock);
+       if (hi->control_state & SSI_CHANNEL_STATE_ERROR) {
+               spin_unlock(&hi->lock);
+               return -EIO;
+       }
+       if (hi->control_state & SSI_CHANNEL_STATE_WRITING) {
+               dev_err(&hi->cl->device,
+                       "Write still pending on control channel.\n");
+               spin_unlock(&hi->lock);
+               return -EBUSY;
+       }
+       hi->control_state |= SSI_CHANNEL_STATE_WRITING;
+       msg = cs_claim_cmd(hi);
+       spin_unlock(&hi->lock);
+
+       cs_set_cmd(msg, message);
+       msg->sgt.nents = 1;
+       msg->complete = cs_hsi_write_on_control_complete;
+       dev_dbg(&hi->cl->device,
+               "Sending control message %08X\n", message);
+       ret = hsi_async_write(hi->cl, msg);
+       if (ret) {
+               dev_err(&hi->cl->device,
+                       "async_write failed with %d\n", ret);
+               cs_hsi_control_write_error(hi, msg);
+       }
+
+       /*
+        * Make sure control read is always pending when issuing
+        * new control writes. This is needed as the controller
+        * may flush our messages if e.g. the peer device reboots
+        * unexpectedly (and we cannot directly resubmit a new read from
+        * the message destructor; see cs_cmd_destructor()).
+        */
+       if (!(hi->control_state & SSI_CHANNEL_STATE_READING)) {
+               dev_err(&hi->cl->device, "Restarting control reads\n");
+               cs_hsi_read_on_control(hi);
+       }
+
+       return 0;
+}
+
+static void cs_hsi_read_on_data_complete(struct hsi_msg *msg)
+{
+       struct cs_hsi_iface *hi = msg->context;
+       u32 payload;
+
+       if (unlikely(msg->status == HSI_STATUS_ERROR)) {
+               cs_hsi_data_read_error(hi, msg);
+               return;
+       }
+
+       spin_lock(&hi->lock);
+       WARN_ON(!(hi->data_state & SSI_CHANNEL_STATE_READING));
+       hi->data_state &= ~SSI_CHANNEL_STATE_READING;
+       payload = CS_RX_DATA_RECEIVED;
+       payload |= hi->rx_slot;
+       hi->rx_slot++;
+       hi->rx_slot %= hi->rx_ptr_boundary;
+       /* expose current rx ptr in mmap area */
+       hi->mmap_cfg->rx_ptr = hi->rx_slot;
+       if (unlikely(waitqueue_active(&hi->datawait)))
+               wake_up_interruptible(&hi->datawait);
+       spin_unlock(&hi->lock);
+
+       cs_notify_data(payload, hi->rx_bufs);
+       cs_hsi_read_on_data(hi);
+}
+
+static void cs_hsi_peek_on_data_complete(struct hsi_msg *msg)
+{
+       struct cs_hsi_iface *hi = msg->context;
+       u32 *address;
+       int ret;
+
+       if (unlikely(msg->status == HSI_STATUS_ERROR)) {
+               cs_hsi_data_read_error(hi, msg);
+               return;
+       }
+       if (unlikely(hi->iface_state != CS_STATE_CONFIGURED)) {
+               dev_err(&hi->cl->device, "Data received in invalid state\n");
+               cs_hsi_data_read_error(hi, msg);
+               return;
+       }
+
+       spin_lock(&hi->lock);
+       WARN_ON(!(hi->data_state & SSI_CHANNEL_STATE_POLL));
+       hi->data_state &= ~SSI_CHANNEL_STATE_POLL;
+       hi->data_state |= SSI_CHANNEL_STATE_READING;
+       spin_unlock(&hi->lock);
+
+       address = (u32 *)(hi->mmap_base +
+                               hi->rx_offsets[hi->rx_slot % hi->rx_bufs]);
+       sg_init_one(msg->sgt.sgl, address, hi->buf_size);
+       msg->sgt.nents = 1;
+       msg->complete = cs_hsi_read_on_data_complete;
+       ret = hsi_async_read(hi->cl, msg);
+       if (ret)
+               cs_hsi_data_read_error(hi, msg);
+}
+
+/*
+ * Read/write transaction is ongoing. Returns false if in
+ * SSI_CHANNEL_STATE_POLL state.
+ */
+static inline int cs_state_xfer_active(unsigned int state)
+{
+       return (state & SSI_CHANNEL_STATE_WRITING) ||
+               (state & SSI_CHANNEL_STATE_READING);
+}
+
+/*
+ * No pending read/writes
+ */
+static inline int cs_state_idle(unsigned int state)
+{
+       return !(state & ~SSI_CHANNEL_STATE_ERROR);
+}
+
+static void cs_hsi_read_on_data(struct cs_hsi_iface *hi)
+{
+       struct hsi_msg *rxmsg;
+       int ret;
+
+       spin_lock(&hi->lock);
+       if (hi->data_state &
+               (SSI_CHANNEL_STATE_READING | SSI_CHANNEL_STATE_POLL)) {
+               dev_dbg(&hi->cl->device, "Data read already pending (%u)\n",
+                       hi->data_state);
+               spin_unlock(&hi->lock);
+               return;
+       }
+       hi->data_state |= SSI_CHANNEL_STATE_POLL;
+       spin_unlock(&hi->lock);
+
+       rxmsg = hi->data_rx_msg;
+       sg_init_one(rxmsg->sgt.sgl, (void *)hi->mmap_base, 0);
+       rxmsg->sgt.nents = 0;
+       rxmsg->complete = cs_hsi_peek_on_data_complete;
+
+       ret = hsi_async_read(hi->cl, rxmsg);
+       if (ret)
+               cs_hsi_data_read_error(hi, rxmsg);
+}
+
+static void cs_hsi_write_on_data_complete(struct hsi_msg *msg)
+{
+       struct cs_hsi_iface *hi = msg->context;
+
+       if (msg->status == HSI_STATUS_COMPLETED) {
+               spin_lock(&hi->lock);
+               hi->data_state &= ~SSI_CHANNEL_STATE_WRITING;
+               if (unlikely(waitqueue_active(&hi->datawait)))
+                       wake_up_interruptible(&hi->datawait);
+               spin_unlock(&hi->lock);
+       } else {
+               cs_hsi_data_write_error(hi, msg);
+       }
+}
+
+static int cs_hsi_write_on_data(struct cs_hsi_iface *hi, unsigned int slot)
+{
+       u32 *address;
+       struct hsi_msg *txmsg;
+       int ret;
+
+       spin_lock(&hi->lock);
+       if (hi->iface_state != CS_STATE_CONFIGURED) {
+               dev_err(&hi->cl->device, "Not configured, aborting\n");
+               ret = -EINVAL;
+               goto error;
+       }
+       if (hi->data_state & SSI_CHANNEL_STATE_ERROR) {
+               dev_err(&hi->cl->device, "HSI error, aborting\n");
+               ret = -EIO;
+               goto error;
+       }
+       if (hi->data_state & SSI_CHANNEL_STATE_WRITING) {
+               dev_err(&hi->cl->device, "Write pending on data channel.\n");
+               ret = -EBUSY;
+               goto error;
+       }
+       hi->data_state |= SSI_CHANNEL_STATE_WRITING;
+       spin_unlock(&hi->lock);
+
+       hi->tx_slot = slot;
+       address = (u32 *)(hi->mmap_base + hi->tx_offsets[hi->tx_slot]);
+       txmsg = hi->data_tx_msg;
+       sg_init_one(txmsg->sgt.sgl, address, hi->buf_size);
+       txmsg->complete = cs_hsi_write_on_data_complete;
+       ret = hsi_async_write(hi->cl, txmsg);
+       if (ret)
+               cs_hsi_data_write_error(hi, txmsg);
+
+       return ret;
+
+error:
+       spin_unlock(&hi->lock);
+       if (ret == -EIO)
+               cs_hsi_data_write_error(hi, hi->data_tx_msg);
+
+       return ret;
+}
+
+static unsigned int cs_hsi_get_state(struct cs_hsi_iface *hi)
+{
+       return hi->iface_state;
+}
+
+static int cs_hsi_command(struct cs_hsi_iface *hi, u32 cmd)
+{
+       int ret = 0;
+
+       local_bh_disable();
+       switch (cmd & TARGET_MASK) {
+       case TARGET_REMOTE:
+               ret = cs_hsi_write_on_control(hi, cmd);
+               break;
+       case TARGET_LOCAL:
+               if ((cmd & CS_CMD_MASK) == CS_TX_DATA_READY)
+                       ret = cs_hsi_write_on_data(hi, cmd & CS_PARAM_MASK);
+               else
+                       ret = -EINVAL;
+               break;
+       default:
+               ret = -EINVAL;
+               break;
+       }
+       local_bh_enable();
+
+       return ret;
+}
+
+static void cs_hsi_set_wakeline(struct cs_hsi_iface *hi, bool new_state)
+{
+       int change = 0;
+
+       spin_lock_bh(&hi->lock);
+       if (hi->wakeline_state != new_state) {
+               hi->wakeline_state = new_state;
+               change = 1;
+               dev_dbg(&hi->cl->device, "setting wake line to %d (%p)\n",
+                       new_state, hi->cl);
+       }
+       spin_unlock_bh(&hi->lock);
+
+       if (change) {
+               if (new_state)
+                       ssip_slave_start_tx(hi->master);
+               else
+                       ssip_slave_stop_tx(hi->master);
+       }
+
+       dev_dbg(&hi->cl->device, "wake line set to %d (%p)\n",
+               new_state, hi->cl);
+}
+
+static void set_buffer_sizes(struct cs_hsi_iface *hi, int rx_bufs, int tx_bufs)
+{
+       hi->rx_bufs = rx_bufs;
+       hi->tx_bufs = tx_bufs;
+       hi->mmap_cfg->rx_bufs = rx_bufs;
+       hi->mmap_cfg->tx_bufs = tx_bufs;
+
+       if (hi->flags & CS_FEAT_ROLLING_RX_COUNTER) {
+               /*
+                * For more robust overrun detection, let the rx
+                * pointer run in range 0..'boundary-1'. Boundary
+                * is a multiple of rx_bufs, and limited in max size
+                * by RX_PTR_MAX_SHIFT to allow for fast ptr-diff
+                * calculation.
+                */
+               hi->rx_ptr_boundary = (rx_bufs << RX_PTR_BOUNDARY_SHIFT);
+               hi->mmap_cfg->rx_ptr_boundary = hi->rx_ptr_boundary;
+       } else {
+               hi->rx_ptr_boundary = hi->rx_bufs;
+       }
+}
+
+static int check_buf_params(struct cs_hsi_iface *hi,
+                                       const struct cs_buffer_config *buf_cfg)
+{
+       size_t buf_size_aligned = L1_CACHE_ALIGN(buf_cfg->buf_size) *
+                                       (buf_cfg->rx_bufs + buf_cfg->tx_bufs);
+       size_t ctrl_size_aligned = L1_CACHE_ALIGN(sizeof(*hi->mmap_cfg));
+       int r = 0;
+
+       if (buf_cfg->rx_bufs > CS_MAX_BUFFERS ||
+                                       buf_cfg->tx_bufs > CS_MAX_BUFFERS) {
+               r = -EINVAL;
+       } else if ((buf_size_aligned + ctrl_size_aligned) >= hi->mmap_size) {
+               dev_err(&hi->cl->device, "No space for the requested buffer "
+                       "configuration\n");
+               r = -ENOBUFS;
+       }
+
+       return r;
+}
+
+/**
+ * Block until pending data transfers have completed.
+ */
+static int cs_hsi_data_sync(struct cs_hsi_iface *hi)
+{
+       int r = 0;
+
+       spin_lock_bh(&hi->lock);
+
+       if (!cs_state_xfer_active(hi->data_state)) {
+               dev_dbg(&hi->cl->device, "hsi_data_sync break, idle\n");
+               goto out;
+       }
+
+       for (;;) {
+               int s;
+               DEFINE_WAIT(wait);
+               if (!cs_state_xfer_active(hi->data_state))
+                       goto out;
+               if (signal_pending(current)) {
+                       r = -ERESTARTSYS;
+                       goto out;
+               }
+               /**
+                * prepare_to_wait must be called with hi->lock held
+                * so that callbacks can check for waitqueue_active()
+                */
+               prepare_to_wait(&hi->datawait, &wait, TASK_INTERRUPTIBLE);
+               spin_unlock_bh(&hi->lock);
+               s = schedule_timeout(
+                       msecs_to_jiffies(CS_HSI_TRANSFER_TIMEOUT_MS));
+               spin_lock_bh(&hi->lock);
+               finish_wait(&hi->datawait, &wait);
+               if (!s) {
+                       dev_dbg(&hi->cl->device,
+                               "hsi_data_sync timeout after %d ms\n",
+                               CS_HSI_TRANSFER_TIMEOUT_MS);
+                       r = -EIO;
+                       goto out;
+               }
+       }
+
+out:
+       spin_unlock_bh(&hi->lock);
+       dev_dbg(&hi->cl->device, "hsi_data_sync done with res %d\n", r);
+
+       return r;
+}
+
+static void cs_hsi_data_enable(struct cs_hsi_iface *hi,
+                                       struct cs_buffer_config *buf_cfg)
+{
+       unsigned int data_start, i;
+
+       BUG_ON(hi->buf_size == 0);
+
+       set_buffer_sizes(hi, buf_cfg->rx_bufs, buf_cfg->tx_bufs);
+
+       hi->slot_size = L1_CACHE_ALIGN(hi->buf_size);
+       dev_dbg(&hi->cl->device,
+                       "setting slot size to %u, buf size %u, align %u\n",
+                       hi->slot_size, hi->buf_size, L1_CACHE_BYTES);
+
+       data_start = L1_CACHE_ALIGN(sizeof(*hi->mmap_cfg));
+       dev_dbg(&hi->cl->device,
+                       "setting data start at %u, cfg block %u, align %u\n",
+                       data_start, sizeof(*hi->mmap_cfg), L1_CACHE_BYTES);
+
+       for (i = 0; i < hi->mmap_cfg->rx_bufs; i++) {
+               hi->rx_offsets[i] = data_start + i * hi->slot_size;
+               hi->mmap_cfg->rx_offsets[i] = hi->rx_offsets[i];
+               dev_dbg(&hi->cl->device, "DL buf #%u at %u\n",
+                                       i, hi->rx_offsets[i]);
+       }
+       for (i = 0; i < hi->mmap_cfg->tx_bufs; i++) {
+               hi->tx_offsets[i] = data_start +
+                       (i + hi->mmap_cfg->rx_bufs) * hi->slot_size;
+               hi->mmap_cfg->tx_offsets[i] = hi->tx_offsets[i];
+               dev_dbg(&hi->cl->device, "UL buf #%u at %u\n",
+                                       i, hi->rx_offsets[i]);
+       }
+
+       hi->iface_state = CS_STATE_CONFIGURED;
+}
+
+static void cs_hsi_data_disable(struct cs_hsi_iface *hi, int old_state)
+{
+       if (old_state == CS_STATE_CONFIGURED) {
+               dev_dbg(&hi->cl->device,
+                       "closing data channel with slot size 0\n");
+               hi->iface_state = CS_STATE_OPENED;
+       }
+}
+
+static int cs_hsi_buf_config(struct cs_hsi_iface *hi,
+                                       struct cs_buffer_config *buf_cfg)
+{
+       int r = 0;
+       unsigned int old_state = hi->iface_state;
+
+       spin_lock_bh(&hi->lock);
+       /* Prevent new transactions during buffer reconfig */
+       if (old_state == CS_STATE_CONFIGURED)
+               hi->iface_state = CS_STATE_OPENED;
+       spin_unlock_bh(&hi->lock);
+
+       /*
+        * make sure that no non-zero data reads are ongoing before
+        * proceeding to change the buffer layout
+        */
+       r = cs_hsi_data_sync(hi);
+       if (r < 0)
+               return r;
+
+       WARN_ON(cs_state_xfer_active(hi->data_state));
+
+       spin_lock_bh(&hi->lock);
+       r = check_buf_params(hi, buf_cfg);
+       if (r < 0)
+               goto error;
+
+       hi->buf_size = buf_cfg->buf_size;
+       hi->mmap_cfg->buf_size = hi->buf_size;
+       hi->flags = buf_cfg->flags;
+
+       hi->rx_slot = 0;
+       hi->tx_slot = 0;
+       hi->slot_size = 0;
+
+       if (hi->buf_size)
+               cs_hsi_data_enable(hi, buf_cfg);
+       else
+               cs_hsi_data_disable(hi, old_state);
+
+       spin_unlock_bh(&hi->lock);
+
+       if (old_state != hi->iface_state) {
+               if (hi->iface_state == CS_STATE_CONFIGURED) {
+                       pm_qos_add_request(&hi->pm_qos_req,
+                               PM_QOS_CPU_DMA_LATENCY,
+                               CS_QOS_LATENCY_FOR_DATA_USEC);
+                       local_bh_disable();
+                       cs_hsi_read_on_data(hi);
+                       local_bh_enable();
+               } else if (old_state == CS_STATE_CONFIGURED) {
+                       pm_qos_remove_request(&hi->pm_qos_req);
+               }
+       }
+       return r;
+
+error:
+       spin_unlock_bh(&hi->lock);
+       return r;
+}
+
+static int cs_hsi_start(struct cs_hsi_iface **hi, struct hsi_client *cl,
+                       unsigned long mmap_base, unsigned long mmap_size)
+{
+       int err = 0;
+       struct cs_hsi_iface *hsi_if = kzalloc(sizeof(*hsi_if), GFP_KERNEL);
+
+       dev_dbg(&cl->device, "cs_hsi_start\n");
+
+       if (!hsi_if) {
+               err = -ENOMEM;
+               goto leave0;
+       }
+       spin_lock_init(&hsi_if->lock);
+       hsi_if->cl = cl;
+       hsi_if->iface_state = CS_STATE_CLOSED;
+       hsi_if->mmap_cfg = (struct cs_mmap_config_block *)mmap_base;
+       hsi_if->mmap_base = mmap_base;
+       hsi_if->mmap_size = mmap_size;
+       memset(hsi_if->mmap_cfg, 0, sizeof(*hsi_if->mmap_cfg));
+       init_waitqueue_head(&hsi_if->datawait);
+       err = cs_alloc_cmds(hsi_if);
+       if (err < 0) {
+               dev_err(&cl->device, "Unable to alloc HSI messages\n");
+               goto leave1;
+       }
+       err = cs_hsi_alloc_data(hsi_if);
+       if (err < 0) {
+               dev_err(&cl->device, "Unable to alloc HSI messages for data\n");
+               goto leave2;
+       }
+       err = hsi_claim_port(cl, 1);
+       if (err < 0) {
+               dev_err(&cl->device,
+                               "Could not open, HSI port already claimed\n");
+               goto leave3;
+       }
+       hsi_if->master = ssip_slave_get_master(cl);
+       if (IS_ERR(hsi_if->master)) {
+               dev_err(&cl->device, "Could not get HSI master client\n");
+               goto leave4;
+       }
+       if (!ssip_slave_running(hsi_if->master)) {
+               err = -ENODEV;
+               dev_err(&cl->device,
+                               "HSI port not initialized\n");
+               goto leave4;
+       }
+
+       hsi_if->iface_state = CS_STATE_OPENED;
+       local_bh_disable();
+       cs_hsi_read_on_control(hsi_if);
+       local_bh_enable();
+
+       dev_dbg(&cl->device, "cs_hsi_start...done\n");
+
+       BUG_ON(!hi);
+       *hi = hsi_if;
+
+       return 0;
+
+leave4:
+       hsi_release_port(cl);
+leave3:
+       cs_hsi_free_data(hsi_if);
+leave2:
+       cs_free_cmds(hsi_if);
+leave1:
+       kfree(hsi_if);
+leave0:
+       dev_dbg(&cl->device, "cs_hsi_start...done/error\n\n");
+
+       return err;
+}
+
+static void cs_hsi_stop(struct cs_hsi_iface *hi)
+{
+       dev_dbg(&hi->cl->device, "cs_hsi_stop\n");
+       cs_hsi_set_wakeline(hi, 0);
+       ssip_slave_put_master(hi->master);
+
+       /* hsi_release_port() needs to be called with CS_STATE_CLOSED */
+       hi->iface_state = CS_STATE_CLOSED;
+       hsi_release_port(hi->cl);
+
+       /*
+        * hsi_release_port() should flush out all the pending
+        * messages, so cs_state_idle() should be true for both
+        * control and data channels.
+        */
+       WARN_ON(!cs_state_idle(hi->control_state));
+       WARN_ON(!cs_state_idle(hi->data_state));
+
+       if (pm_qos_request_active(&hi->pm_qos_req))
+               pm_qos_remove_request(&hi->pm_qos_req);
+
+       spin_lock_bh(&hi->lock);
+       cs_hsi_free_data(hi);
+       cs_free_cmds(hi);
+       spin_unlock_bh(&hi->lock);
+       kfree(hi);
+}
+
+static int cs_char_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
+{
+       struct cs_char *csdata = vma->vm_private_data;
+       struct page *page;
+
+       page = virt_to_page(csdata->mmap_base);
+       get_page(page);
+       vmf->page = page;
+
+       return 0;
+}
+
+static struct vm_operations_struct cs_char_vm_ops = {
+       .fault  = cs_char_vma_fault,
+};
+
+static int cs_char_fasync(int fd, struct file *file, int on)
+{
+       struct cs_char *csdata = file->private_data;
+
+       if (fasync_helper(fd, file, on, &csdata->async_queue) < 0)
+               return -EIO;
+
+       return 0;
+}
+
+static unsigned int cs_char_poll(struct file *file, poll_table *wait)
+{
+       struct cs_char *csdata = file->private_data;
+       unsigned int ret = 0;
+
+       poll_wait(file, &cs_char_data.wait, wait);
+       spin_lock_bh(&csdata->lock);
+       if (!list_empty(&csdata->chardev_queue))
+               ret = POLLIN | POLLRDNORM;
+       else if (!list_empty(&csdata->dataind_queue))
+               ret = POLLIN | POLLRDNORM;
+       spin_unlock_bh(&csdata->lock);
+
+       return ret;
+}
+
+static ssize_t cs_char_read(struct file *file, char __user *buf, size_t count,
+                                                               loff_t *unused)
+{
+       struct cs_char *csdata = file->private_data;
+       u32 data;
+       ssize_t retval;
+
+       if (count < sizeof(data))
+               return -EINVAL;
+
+       for (;;) {
+               DEFINE_WAIT(wait);
+
+               spin_lock_bh(&csdata->lock);
+               if (!list_empty(&csdata->chardev_queue)) {
+                       data = cs_pop_entry(&csdata->chardev_queue);
+               } else if (!list_empty(&csdata->dataind_queue)) {
+                       data = cs_pop_entry(&csdata->dataind_queue);
+                       csdata->dataind_pending--;
+               } else {
+                       data = 0;
+               }
+               spin_unlock_bh(&csdata->lock);
+
+               if (data)
+                       break;
+               if (file->f_flags & O_NONBLOCK) {
+                       retval = -EAGAIN;
+                       goto out;
+               } else if (signal_pending(current)) {
+                       retval = -ERESTARTSYS;
+                       goto out;
+               }
+               prepare_to_wait_exclusive(&csdata->wait, &wait,
+                                               TASK_INTERRUPTIBLE);
+               schedule();
+               finish_wait(&csdata->wait, &wait);
+       }
+
+       retval = put_user(data, (u32 __user *)buf);
+       if (!retval)
+               retval = sizeof(data);
+
+out:
+       return retval;
+}
+
+static ssize_t cs_char_write(struct file *file, const char __user *buf,
+                                               size_t count, loff_t *unused)
+{
+       struct cs_char *csdata = file->private_data;
+       u32 data;
+       int err;
+       ssize_t retval;
+
+       if (count < sizeof(data))
+               return -EINVAL;
+
+       if (get_user(data, (u32 __user *)buf))
+               retval = -EFAULT;
+       else
+               retval = count;
+
+       err = cs_hsi_command(csdata->hi, data);
+       if (err < 0)
+               retval = err;
+
+       return retval;
+}
+
+static long cs_char_ioctl(struct file *file, unsigned int cmd,
+                               unsigned long arg)
+{
+       struct cs_char *csdata = file->private_data;
+       int r = 0;
+
+       switch (cmd) {
+       case CS_GET_STATE: {
+               unsigned int state;
+
+               state = cs_hsi_get_state(csdata->hi);
+               if (copy_to_user((void __user *)arg, &state, sizeof(state)))
+                       r = -EFAULT;
+
+               break;
+       }
+       case CS_SET_WAKELINE: {
+               unsigned int state;
+
+               if (copy_from_user(&state, (void __user *)arg, sizeof(state))) {
+                       r = -EFAULT;
+                       break;
+               }
+
+               if (state > 1) {
+                       r = -EINVAL;
+                       break;
+               }
+
+               cs_hsi_set_wakeline(csdata->hi, !!state);
+
+               break;
+       }
+       case CS_GET_IF_VERSION: {
+               unsigned int ifver = CS_IF_VERSION;
+
+               if (copy_to_user((void __user *)arg, &ifver, sizeof(ifver)))
+                       r = -EFAULT;
+
+               break;
+       }
+       case CS_CONFIG_BUFS: {
+               struct cs_buffer_config buf_cfg;
+
+               if (copy_from_user(&buf_cfg, (void __user *)arg,
+                                                       sizeof(buf_cfg)))
+                       r = -EFAULT;
+               else
+                       r = cs_hsi_buf_config(csdata->hi, &buf_cfg);
+
+               break;
+       }
+       default:
+               r = -ENOTTY;
+               break;
+       }
+
+       return r;
+}
+
+static int cs_char_mmap(struct file *file, struct vm_area_struct *vma)
+{
+       if (vma->vm_end < vma->vm_start)
+               return -EINVAL;
+
+       if (((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) != 1)
+               return -EINVAL;
+
+       vma->vm_flags |= VM_IO | VM_DONTDUMP | VM_DONTEXPAND;
+       vma->vm_ops = &cs_char_vm_ops;
+       vma->vm_private_data = file->private_data;
+
+       return 0;
+}
+
+static int cs_char_open(struct inode *unused, struct file *file)
+{
+       int ret = 0;
+       unsigned long p;
+
+       spin_lock_bh(&cs_char_data.lock);
+       if (cs_char_data.opened) {
+               ret = -EBUSY;
+               spin_unlock_bh(&cs_char_data.lock);
+               goto out1;
+       }
+       cs_char_data.opened = 1;
+       cs_char_data.dataind_pending = 0;
+       spin_unlock_bh(&cs_char_data.lock);
+
+       p = get_zeroed_page(GFP_KERNEL);
+       if (!p) {
+               ret = -ENOMEM;
+               goto out2;
+       }
+
+       ret = cs_hsi_start(&cs_char_data.hi, cs_char_data.cl, p, CS_MMAP_SIZE);
+       if (ret) {
+               dev_err(&cs_char_data.cl->device, "Unable to initialize HSI\n");
+               goto out3;
+       }
+
+       /* these are only used in release so lock not needed */
+       cs_char_data.mmap_base = p;
+       cs_char_data.mmap_size = CS_MMAP_SIZE;
+
+       file->private_data = &cs_char_data;
+
+       return 0;
+
+out3:
+       free_page(p);
+out2:
+       spin_lock_bh(&cs_char_data.lock);
+       cs_char_data.opened = 0;
+       spin_unlock_bh(&cs_char_data.lock);
+out1:
+       return ret;
+}
+
+static void cs_free_char_queue(struct list_head *head)
+{
+       struct char_queue *entry;
+       struct list_head *cursor, *next;
+
+       if (!list_empty(head)) {
+               list_for_each_safe(cursor, next, head) {
+                       entry = list_entry(cursor, struct char_queue, list);
+                       list_del(&entry->list);
+                       kfree(entry);
+               }
+       }
+
+}
+
+static int cs_char_release(struct inode *unused, struct file *file)
+{
+       struct cs_char *csdata = file->private_data;
+
+       cs_hsi_stop(csdata->hi);
+       spin_lock_bh(&csdata->lock);
+       csdata->hi = NULL;
+       free_page(csdata->mmap_base);
+       cs_free_char_queue(&csdata->chardev_queue);
+       cs_free_char_queue(&csdata->dataind_queue);
+       csdata->opened = 0;
+       spin_unlock_bh(&csdata->lock);
+
+       return 0;
+}
+
+static const struct file_operations cs_char_fops = {
+       .owner          = THIS_MODULE,
+       .read           = cs_char_read,
+       .write          = cs_char_write,
+       .poll           = cs_char_poll,
+       .unlocked_ioctl = cs_char_ioctl,
+       .mmap           = cs_char_mmap,
+       .open           = cs_char_open,
+       .release        = cs_char_release,
+       .fasync         = cs_char_fasync,
+};
+
+static struct miscdevice cs_char_miscdev = {
+       .minor  = MISC_DYNAMIC_MINOR,
+       .name   = "cmt_speech",
+       .fops   = &cs_char_fops
+};
+
+static int cs_hsi_client_probe(struct device *dev)
+{
+       int err = 0;
+       struct hsi_client *cl = to_hsi_client(dev);
+
+       dev_dbg(dev, "hsi_client_probe\n");
+       init_waitqueue_head(&cs_char_data.wait);
+       spin_lock_init(&cs_char_data.lock);
+       cs_char_data.opened = 0;
+       cs_char_data.cl = cl;
+       cs_char_data.hi = NULL;
+       INIT_LIST_HEAD(&cs_char_data.chardev_queue);
+       INIT_LIST_HEAD(&cs_char_data.dataind_queue);
+
+       cs_char_data.channel_id_cmd = hsi_get_channel_id_by_name(cl,
+               "speech-control");
+       if (cs_char_data.channel_id_cmd < 0) {
+               err = cs_char_data.channel_id_cmd;
+               dev_err(dev, "Could not get cmd channel (%d)\n", err);
+               return err;
+       }
+
+       cs_char_data.channel_id_data = hsi_get_channel_id_by_name(cl,
+               "speech-data");
+       if (cs_char_data.channel_id_data < 0) {
+               err = cs_char_data.channel_id_data;
+               dev_err(dev, "Could not get data channel (%d)\n", err);
+               return err;
+       }
+
+       err = misc_register(&cs_char_miscdev);
+       if (err)
+               dev_err(dev, "Failed to register: %d\n", err);
+
+       return err;
+}
+
+static int cs_hsi_client_remove(struct device *dev)
+{
+       struct cs_hsi_iface *hi;
+
+       dev_dbg(dev, "hsi_client_remove\n");
+       misc_deregister(&cs_char_miscdev);
+       spin_lock_bh(&cs_char_data.lock);
+       hi = cs_char_data.hi;
+       cs_char_data.hi = NULL;
+       spin_unlock_bh(&cs_char_data.lock);
+       if (hi)
+               cs_hsi_stop(hi);
+
+       return 0;
+}
+
+static struct hsi_client_driver cs_hsi_driver = {
+       .driver = {
+               .name   = "cmt-speech",
+               .owner  = THIS_MODULE,
+               .probe  = cs_hsi_client_probe,
+               .remove = cs_hsi_client_remove,
+       },
+};
+
+static int __init cs_char_init(void)
+{
+       pr_info("CMT speech driver added\n");
+       return hsi_register_client_driver(&cs_hsi_driver);
+}
+module_init(cs_char_init);
+
+static void __exit cs_char_exit(void)
+{
+       hsi_unregister_client_driver(&cs_hsi_driver);
+       pr_info("CMT speech driver removed\n");
+}
+module_exit(cs_char_exit);
+
+MODULE_ALIAS("hsi:cmt-speech");
+MODULE_AUTHOR("Kai Vehmanen <kai.vehmanen@nokia.com>");
+MODULE_AUTHOR("Peter Ujfalusi <peter.ujfalusi@nokia.com>");
+MODULE_DESCRIPTION("CMT speech driver");
+MODULE_LICENSE("GPL v2");
index 30ab3cd3b8a55b7f347eda392cae4578645af9e0..a16a00544258c2e99e46fd1632548f79d875075f 100644 (file)
@@ -1,2 +1,2 @@
 # UAPI Header export list
-header-y += hsi_char.h
+header-y += hsi_char.h cs-protocol.h
diff --git a/include/uapi/linux/hsi/cs-protocol.h b/include/uapi/linux/hsi/cs-protocol.h
new file mode 100644 (file)
index 0000000..4957bba
--- /dev/null
@@ -0,0 +1,113 @@
+/*
+ * cmt-speech interface definitions
+ *
+ * Copyright (C) 2008,2009,2010 Nokia Corporation. All rights reserved.
+ *
+ * Contact: Kai Vehmanen <kai.vehmanen@nokia.com>
+ * Original author: Peter Ujfalusi <peter.ujfalusi@nokia.com>
+ *
+ * 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.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA
+ * 02110-1301 USA
+ */
+
+#ifndef _CS_PROTOCOL_H
+#define _CS_PROTOCOL_H
+
+#include <linux/types.h>
+#include <linux/ioctl.h>
+
+/* chardev parameters */
+#define CS_DEV_FILE_NAME               "/dev/cmt_speech"
+
+/* user-space API versioning */
+#define CS_IF_VERSION                  2
+
+/* APE kernel <-> user space messages */
+#define CS_CMD_SHIFT                   28
+#define CS_DOMAIN_SHIFT                        24
+
+#define CS_CMD_MASK                    0xff000000
+#define CS_PARAM_MASK                  0xffffff
+
+#define CS_CMD(id, dom) \
+       (((id) << CS_CMD_SHIFT) | ((dom) << CS_DOMAIN_SHIFT))
+
+#define CS_ERROR                       CS_CMD(1, 0)
+#define CS_RX_DATA_RECEIVED            CS_CMD(2, 0)
+#define CS_TX_DATA_READY               CS_CMD(3, 0)
+#define CS_TX_DATA_SENT                        CS_CMD(4, 0)
+
+/* params to CS_ERROR indication */
+#define CS_ERR_PEER_RESET              0
+
+/* ioctl interface */
+
+/* parameters to CS_CONFIG_BUFS ioctl */
+#define CS_FEAT_TSTAMP_RX_CTRL         (1 << 0)
+#define CS_FEAT_ROLLING_RX_COUNTER     (2 << 0)
+
+/* parameters to CS_GET_STATE ioctl */
+#define CS_STATE_CLOSED                        0
+#define CS_STATE_OPENED                        1 /* resource allocated */
+#define CS_STATE_CONFIGURED            2 /* data path active */
+
+/* maximum number of TX/RX buffers */
+#define CS_MAX_BUFFERS_SHIFT           4
+#define CS_MAX_BUFFERS                 (1 << CS_MAX_BUFFERS_SHIFT)
+
+/* Parameters for setting up the data buffers */
+struct cs_buffer_config {
+       __u32 rx_bufs;  /* number of RX buffer slots */
+       __u32 tx_bufs;  /* number of TX buffer slots */
+       __u32 buf_size; /* bytes */
+       __u32 flags;    /* see CS_FEAT_* */
+       __u32 reserved[4];
+};
+
+/*
+ * Struct describing the layout and contents of the driver mmap area.
+ * This information is meant as read-only information for the application.
+ */
+struct cs_mmap_config_block {
+       __u32 reserved1;
+       __u32 buf_size;         /* 0=disabled, otherwise the transfer size */
+       __u32 rx_bufs;          /* # of RX buffers */
+       __u32 tx_bufs;          /* # of TX buffers */
+       __u32 reserved2;
+       /* array of offsets within the mmap area for each RX and TX buffer */
+       __u32 rx_offsets[CS_MAX_BUFFERS];
+       __u32 tx_offsets[CS_MAX_BUFFERS];
+       __u32 rx_ptr;
+       __u32 rx_ptr_boundary;
+       __u32 reserved3[2];
+       /*
+        * if enabled with CS_FEAT_TSTAMP_RX_CTRL, monotonic
+        * timestamp taken when the last control command was received
+        */
+       struct timespec tstamp_rx_ctrl;
+};
+
+#define CS_IO_MAGIC            'C'
+
+#define CS_IOW(num, dtype)     _IOW(CS_IO_MAGIC, num, dtype)
+#define CS_IOR(num, dtype)     _IOR(CS_IO_MAGIC, num, dtype)
+#define CS_IOWR(num, dtype)    _IOWR(CS_IO_MAGIC, num, dtype)
+#define CS_IO(num)             _IO(CS_IO_MAGIC, num)
+
+#define CS_GET_STATE           CS_IOR(21, unsigned int)
+#define CS_SET_WAKELINE                CS_IOW(23, unsigned int)
+#define CS_GET_IF_VERSION      CS_IOR(30, unsigned int)
+#define CS_CONFIG_BUFS         CS_IOW(31, struct cs_buffer_config)
+
+#endif /* _CS_PROTOCOL_H */