]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
[media] cec: add HDMI CEC framework (adapter)
authorHans Verkuil <hans.verkuil@cisco.com>
Sat, 25 Jun 2016 12:44:16 +0000 (09:44 -0300)
committerMauro Carvalho Chehab <mchehab@s-opensource.com>
Tue, 28 Jun 2016 12:58:43 +0000 (09:58 -0300)
The added HDMI CEC framework provides a generic kernel interface for
HDMI CEC devices.

Note that the CEC framework is added to staging/media and that the
cec.h and cec-funcs.h headers are not exported yet. While the kABI
is mature, I would prefer to allow the uABI some more time before
it is mainlined in case it needs more tweaks.

This adds the cec-adap.c source that deals with the low-level CEC
messaging and logical address handling.

Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
[k.debski@samsung.com: change kthread handling when setting logical address]
[k.debski@samsung.com: code cleanup and fixes]
[k.debski@samsung.com: add possibility to clear assigned logical addresses]
Signed-off-by: Kamil Debski <kamil@wypas.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab@s-opensource.com>
drivers/staging/media/cec/cec-adap.c [new file with mode: 0644]

diff --git a/drivers/staging/media/cec/cec-adap.c b/drivers/staging/media/cec/cec-adap.c
new file mode 100644 (file)
index 0000000..5ffa839
--- /dev/null
@@ -0,0 +1,1627 @@
+/*
+ * cec-adap.c - HDMI Consumer Electronics Control framework - CEC adapter
+ *
+ * Copyright 2016 Cisco Systems, Inc. and/or its affiliates. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/errno.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/kmod.h>
+#include <linux/ktime.h>
+#include <linux/slab.h>
+#include <linux/mm.h>
+#include <linux/string.h>
+#include <linux/types.h>
+
+#include "cec-priv.h"
+
+static int cec_report_features(struct cec_adapter *adap, unsigned int la_idx);
+static int cec_report_phys_addr(struct cec_adapter *adap, unsigned int la_idx);
+
+/*
+ * 400 ms is the time it takes for one 16 byte message to be
+ * transferred and 5 is the maximum number of retries. Add
+ * another 100 ms as a margin. So if the transmit doesn't
+ * finish before that time something is really wrong and we
+ * have to time out.
+ *
+ * This is a sign that something it really wrong and a warning
+ * will be issued.
+ */
+#define CEC_XFER_TIMEOUT_MS (5 * 400 + 100)
+
+#define call_op(adap, op, arg...) \
+       (adap->ops->op ? adap->ops->op(adap, ## arg) : 0)
+
+#define call_void_op(adap, op, arg...)                 \
+       do {                                            \
+               if (adap->ops->op)                      \
+                       adap->ops->op(adap, ## arg);    \
+       } while (0)
+
+static int cec_log_addr2idx(const struct cec_adapter *adap, u8 log_addr)
+{
+       int i;
+
+       for (i = 0; i < adap->log_addrs.num_log_addrs; i++)
+               if (adap->log_addrs.log_addr[i] == log_addr)
+                       return i;
+       return -1;
+}
+
+static unsigned int cec_log_addr2dev(const struct cec_adapter *adap, u8 log_addr)
+{
+       int i = cec_log_addr2idx(adap, log_addr);
+
+       return adap->log_addrs.primary_device_type[i < 0 ? 0 : i];
+}
+
+/*
+ * Queue a new event for this filehandle. If ts == 0, then set it
+ * to the current time.
+ *
+ * The two events that are currently defined do not need to keep track
+ * of intermediate events, so no actual queue of events is needed,
+ * instead just store the latest state and the total number of lost
+ * messages.
+ *
+ * Should new events be added in the future that require intermediate
+ * results to be queued as well, then a proper queue data structure is
+ * required. But until then, just keep it simple.
+ */
+void cec_queue_event_fh(struct cec_fh *fh,
+                       const struct cec_event *new_ev, u64 ts)
+{
+       struct cec_event *ev = &fh->events[new_ev->event - 1];
+
+       if (ts == 0)
+               ts = ktime_get_ns();
+
+       mutex_lock(&fh->lock);
+       if (new_ev->event == CEC_EVENT_LOST_MSGS &&
+           fh->pending_events & (1 << new_ev->event)) {
+               /*
+                * If there is already a lost_msgs event, then just
+                * update the lost_msgs count. This effectively
+                * merges the old and new events into one.
+                */
+               ev->lost_msgs.lost_msgs += new_ev->lost_msgs.lost_msgs;
+               goto unlock;
+       }
+
+       /*
+        * Intermediate states are not interesting, so just
+        * overwrite any older event.
+        */
+       *ev = *new_ev;
+       ev->ts = ts;
+       fh->pending_events |= 1 << new_ev->event;
+
+unlock:
+       mutex_unlock(&fh->lock);
+       wake_up_interruptible(&fh->wait);
+}
+
+/* Queue a new event for all open filehandles. */
+static void cec_queue_event(struct cec_adapter *adap,
+                           const struct cec_event *ev)
+{
+       u64 ts = ktime_get_ns();
+       struct cec_fh *fh;
+
+       mutex_lock(&adap->devnode.fhs_lock);
+       list_for_each_entry(fh, &adap->devnode.fhs, list)
+               cec_queue_event_fh(fh, ev, ts);
+       mutex_unlock(&adap->devnode.fhs_lock);
+}
+
+/*
+ * Queue a new message for this filehandle. If there is no more room
+ * in the queue, then send the LOST_MSGS event instead.
+ */
+static void cec_queue_msg_fh(struct cec_fh *fh, const struct cec_msg *msg)
+{
+       static const struct cec_event ev_lost_msg = {
+               .event = CEC_EVENT_LOST_MSGS,
+               .lost_msgs.lost_msgs = 1,
+       };
+       struct cec_msg_entry *entry;
+
+       mutex_lock(&fh->lock);
+       entry = kmalloc(sizeof(*entry), GFP_KERNEL);
+       if (!entry)
+               goto lost_msgs;
+
+       entry->msg = *msg;
+       /* Add new msg at the end of the queue */
+       list_add_tail(&entry->list, &fh->msgs);
+
+       /*
+        * if the queue now has more than CEC_MAX_MSG_QUEUE_SZ
+        * messages, drop the oldest one and send a lost message event.
+        */
+       if (fh->queued_msgs == CEC_MAX_MSG_QUEUE_SZ) {
+               list_del(&entry->list);
+               goto lost_msgs;
+       }
+       fh->queued_msgs++;
+       mutex_unlock(&fh->lock);
+       wake_up_interruptible(&fh->wait);
+       return;
+
+lost_msgs:
+       mutex_unlock(&fh->lock);
+       cec_queue_event_fh(fh, &ev_lost_msg, 0);
+}
+
+/*
+ * Queue the message for those filehandles that are in monitor mode.
+ * If valid_la is true (this message is for us or was sent by us),
+ * then pass it on to any monitoring filehandle. If this message
+ * isn't for us or from us, then only give it to filehandles that
+ * are in MONITOR_ALL mode.
+ *
+ * This can only happen if the CEC_CAP_MONITOR_ALL capability is
+ * set and the CEC adapter was placed in 'monitor all' mode.
+ */
+static void cec_queue_msg_monitor(struct cec_adapter *adap,
+                                 const struct cec_msg *msg,
+                                 bool valid_la)
+{
+       struct cec_fh *fh;
+       u32 monitor_mode = valid_la ? CEC_MODE_MONITOR :
+                                     CEC_MODE_MONITOR_ALL;
+
+       mutex_lock(&adap->devnode.fhs_lock);
+       list_for_each_entry(fh, &adap->devnode.fhs, list) {
+               if (fh->mode_follower >= monitor_mode)
+                       cec_queue_msg_fh(fh, msg);
+       }
+       mutex_unlock(&adap->devnode.fhs_lock);
+}
+
+/*
+ * Queue the message for follower filehandles.
+ */
+static void cec_queue_msg_followers(struct cec_adapter *adap,
+                                   const struct cec_msg *msg)
+{
+       struct cec_fh *fh;
+
+       mutex_lock(&adap->devnode.fhs_lock);
+       list_for_each_entry(fh, &adap->devnode.fhs, list) {
+               if (fh->mode_follower == CEC_MODE_FOLLOWER)
+                       cec_queue_msg_fh(fh, msg);
+       }
+       mutex_unlock(&adap->devnode.fhs_lock);
+}
+
+/* Notify userspace of an adapter state change. */
+static void cec_post_state_event(struct cec_adapter *adap)
+{
+       struct cec_event ev = {
+               .event = CEC_EVENT_STATE_CHANGE,
+       };
+
+       ev.state_change.phys_addr = adap->phys_addr;
+       ev.state_change.log_addr_mask = adap->log_addrs.log_addr_mask;
+       cec_queue_event(adap, &ev);
+}
+
+/*
+ * A CEC transmit (and a possible wait for reply) completed.
+ * If this was in blocking mode, then complete it, otherwise
+ * queue the message for userspace to dequeue later.
+ *
+ * This function is called with adap->lock held.
+ */
+static void cec_data_completed(struct cec_data *data)
+{
+       /*
+        * Delete this transmit from the filehandle's xfer_list since
+        * we're done with it.
+        *
+        * Note that if the filehandle is closed before this transmit
+        * finished, then the release() function will set data->fh to NULL.
+        * Without that we would be referring to a closed filehandle.
+        */
+       if (data->fh)
+               list_del(&data->xfer_list);
+
+       if (data->blocking) {
+               /*
+                * Someone is blocking so mark the message as completed
+                * and call complete.
+                */
+               data->completed = true;
+               complete(&data->c);
+       } else {
+               /*
+                * No blocking, so just queue the message if needed and
+                * free the memory.
+                */
+               if (data->fh)
+                       cec_queue_msg_fh(data->fh, &data->msg);
+               kfree(data);
+       }
+}
+
+/*
+ * A pending CEC transmit needs to be cancelled, either because the CEC
+ * adapter is disabled or the transmit takes an impossibly long time to
+ * finish.
+ *
+ * This function is called with adap->lock held.
+ */
+static void cec_data_cancel(struct cec_data *data)
+{
+       /*
+        * It's either the current transmit, or it is a pending
+        * transmit. Take the appropriate action to clear it.
+        */
+       if (data->adap->transmitting == data)
+               data->adap->transmitting = NULL;
+       else
+               list_del_init(&data->list);
+
+       /* Mark it as an error */
+       data->msg.ts = ktime_get_ns();
+       data->msg.tx_status = CEC_TX_STATUS_ERROR |
+                             CEC_TX_STATUS_MAX_RETRIES;
+       data->attempts = 0;
+       data->msg.tx_error_cnt = 1;
+       data->msg.reply = 0;
+       /* Queue transmitted message for monitoring purposes */
+       cec_queue_msg_monitor(data->adap, &data->msg, 1);
+
+       cec_data_completed(data);
+}
+
+/*
+ * Main CEC state machine
+ *
+ * Wait until the thread should be stopped, or we are not transmitting and
+ * a new transmit message is queued up, in which case we start transmitting
+ * that message. When the adapter finished transmitting the message it will
+ * call cec_transmit_done().
+ *
+ * If the adapter is disabled, then remove all queued messages instead.
+ *
+ * If the current transmit times out, then cancel that transmit.
+ */
+int cec_thread_func(void *_adap)
+{
+       struct cec_adapter *adap = _adap;
+
+       for (;;) {
+               unsigned int signal_free_time;
+               struct cec_data *data;
+               bool timeout = false;
+               u8 attempts;
+
+               if (adap->transmitting) {
+                       int err;
+
+                       /*
+                        * We are transmitting a message, so add a timeout
+                        * to prevent the state machine to get stuck waiting
+                        * for this message to finalize and add a check to
+                        * see if the adapter is disabled in which case the
+                        * transmit should be canceled.
+                        */
+                       err = wait_event_interruptible_timeout(adap->kthread_waitq,
+                               kthread_should_stop() ||
+                               adap->phys_addr == CEC_PHYS_ADDR_INVALID ||
+                               (!adap->transmitting &&
+                                !list_empty(&adap->transmit_queue)),
+                               msecs_to_jiffies(CEC_XFER_TIMEOUT_MS));
+                       timeout = err == 0;
+               } else {
+                       /* Otherwise we just wait for something to happen. */
+                       wait_event_interruptible(adap->kthread_waitq,
+                               kthread_should_stop() ||
+                               (!adap->transmitting &&
+                                !list_empty(&adap->transmit_queue)));
+               }
+
+               mutex_lock(&adap->lock);
+
+               if (adap->phys_addr == CEC_PHYS_ADDR_INVALID ||
+                   kthread_should_stop()) {
+                       /*
+                        * If the adapter is disabled, or we're asked to stop,
+                        * then cancel any pending transmits.
+                        */
+                       while (!list_empty(&adap->transmit_queue)) {
+                               data = list_first_entry(&adap->transmit_queue,
+                                                       struct cec_data, list);
+                               cec_data_cancel(data);
+                       }
+                       if (adap->transmitting)
+                               cec_data_cancel(adap->transmitting);
+
+                       /*
+                        * Cancel the pending timeout work. We have to unlock
+                        * the mutex when flushing the work since
+                        * cec_wait_timeout() will take it. This is OK since
+                        * no new entries can be added to wait_queue as long
+                        * as adap->transmitting is NULL, which it is due to
+                        * the cec_data_cancel() above.
+                        */
+                       while (!list_empty(&adap->wait_queue)) {
+                               data = list_first_entry(&adap->wait_queue,
+                                                       struct cec_data, list);
+
+                               if (!cancel_delayed_work(&data->work)) {
+                                       mutex_unlock(&adap->lock);
+                                       flush_scheduled_work();
+                                       mutex_lock(&adap->lock);
+                               }
+                               cec_data_cancel(data);
+                       }
+                       goto unlock;
+               }
+
+               if (adap->transmitting && timeout) {
+                       /*
+                        * If we timeout, then log that. This really shouldn't
+                        * happen and is an indication of a faulty CEC adapter
+                        * driver, or the CEC bus is in some weird state.
+                        */
+                       dprintk(0, "message %*ph timed out!\n",
+                               adap->transmitting->msg.len,
+                               adap->transmitting->msg.msg);
+                       /* Just give up on this. */
+                       cec_data_cancel(adap->transmitting);
+                       goto unlock;
+               }
+
+               /*
+                * If we are still transmitting, or there is nothing new to
+                * transmit, then just continue waiting.
+                */
+               if (adap->transmitting || list_empty(&adap->transmit_queue))
+                       goto unlock;
+
+               /* Get a new message to transmit */
+               data = list_first_entry(&adap->transmit_queue,
+                                       struct cec_data, list);
+               list_del_init(&data->list);
+               /* Make this the current transmitting message */
+               adap->transmitting = data;
+
+               /*
+                * Suggested number of attempts as per the CEC 2.0 spec:
+                * 4 attempts is the default, except for 'secondary poll
+                * messages', i.e. poll messages not sent during the adapter
+                * configuration phase when it allocates logical addresses.
+                */
+               if (data->msg.len == 1 && adap->is_configured)
+                       attempts = 2;
+               else
+                       attempts = 4;
+
+               /* Set the suggested signal free time */
+               if (data->attempts) {
+                       /* should be >= 3 data bit periods for a retry */
+                       signal_free_time = CEC_SIGNAL_FREE_TIME_RETRY;
+               } else if (data->new_initiator) {
+                       /* should be >= 5 data bit periods for new initiator */
+                       signal_free_time = CEC_SIGNAL_FREE_TIME_NEW_INITIATOR;
+               } else {
+                       /*
+                        * should be >= 7 data bit periods for sending another
+                        * frame immediately after another.
+                        */
+                       signal_free_time = CEC_SIGNAL_FREE_TIME_NEXT_XFER;
+               }
+               if (data->attempts == 0)
+                       data->attempts = attempts;
+
+               /* Tell the adapter to transmit, cancel on error */
+               if (adap->ops->adap_transmit(adap, data->attempts,
+                                            signal_free_time, &data->msg))
+                       cec_data_cancel(data);
+
+unlock:
+               mutex_unlock(&adap->lock);
+
+               if (kthread_should_stop())
+                       break;
+       }
+       return 0;
+}
+
+/*
+ * Called by the CEC adapter if a transmit finished.
+ */
+void cec_transmit_done(struct cec_adapter *adap, u8 status, u8 arb_lost_cnt,
+                      u8 nack_cnt, u8 low_drive_cnt, u8 error_cnt)
+{
+       struct cec_data *data;
+       struct cec_msg *msg;
+
+       dprintk(2, "cec_transmit_done %02x\n", status);
+       mutex_lock(&adap->lock);
+       data = adap->transmitting;
+       if (!data) {
+               /*
+                * This can happen if a transmit was issued and the cable is
+                * unplugged while the transmit is ongoing. Ignore this
+                * transmit in that case.
+                */
+               dprintk(1, "cec_transmit_done without an ongoing transmit!\n");
+               goto unlock;
+       }
+
+       msg = &data->msg;
+
+       /* Drivers must fill in the status! */
+       WARN_ON(status == 0);
+       msg->ts = ktime_get_ns();
+       msg->tx_status |= status;
+       msg->tx_arb_lost_cnt += arb_lost_cnt;
+       msg->tx_nack_cnt += nack_cnt;
+       msg->tx_low_drive_cnt += low_drive_cnt;
+       msg->tx_error_cnt += error_cnt;
+
+       /* Mark that we're done with this transmit */
+       adap->transmitting = NULL;
+
+       /*
+        * If there are still retry attempts left and there was an error and
+        * the hardware didn't signal that it retried itself (by setting
+        * CEC_TX_STATUS_MAX_RETRIES), then we will retry ourselves.
+        */
+       if (data->attempts > 1 &&
+           !(status & (CEC_TX_STATUS_MAX_RETRIES | CEC_TX_STATUS_OK))) {
+               /* Retry this message */
+               data->attempts--;
+               /* Add the message in front of the transmit queue */
+               list_add(&data->list, &adap->transmit_queue);
+               goto wake_thread;
+       }
+
+       data->attempts = 0;
+
+       /* Always set CEC_TX_STATUS_MAX_RETRIES on error */
+       if (!(status & CEC_TX_STATUS_OK))
+               msg->tx_status |= CEC_TX_STATUS_MAX_RETRIES;
+
+       /* Queue transmitted message for monitoring purposes */
+       cec_queue_msg_monitor(adap, msg, 1);
+
+       /*
+        * Clear reply and timeout on error or if the adapter is no longer
+        * configured. It makes no sense to wait for a reply in that case.
+        */
+       if (!(status & CEC_TX_STATUS_OK) || !adap->is_configured) {
+               msg->reply = 0;
+               msg->timeout = 0;
+       }
+
+       if (msg->timeout) {
+               /*
+                * Queue the message into the wait queue if we want to wait
+                * for a reply.
+                */
+               list_add_tail(&data->list, &adap->wait_queue);
+               schedule_delayed_work(&data->work,
+                                     msecs_to_jiffies(msg->timeout));
+       } else {
+               /* Otherwise we're done */
+               cec_data_completed(data);
+       }
+
+wake_thread:
+       /*
+        * Wake up the main thread to see if another message is ready
+        * for transmitting or to retry the current message.
+        */
+       wake_up_interruptible(&adap->kthread_waitq);
+unlock:
+       mutex_unlock(&adap->lock);
+}
+EXPORT_SYMBOL_GPL(cec_transmit_done);
+
+/*
+ * Called when waiting for a reply times out.
+ */
+static void cec_wait_timeout(struct work_struct *work)
+{
+       struct cec_data *data = container_of(work, struct cec_data, work.work);
+       struct cec_adapter *adap = data->adap;
+
+       mutex_lock(&adap->lock);
+       /*
+        * Sanity check in case the timeout and the arrival of the message
+        * happened at the same time.
+        */
+       if (list_empty(&data->list))
+               goto unlock;
+
+       /* Mark the message as timed out */
+       list_del_init(&data->list);
+       data->msg.ts = ktime_get_ns();
+       data->msg.rx_status = CEC_RX_STATUS_TIMEOUT;
+       cec_data_completed(data);
+unlock:
+       mutex_unlock(&adap->lock);
+}
+
+/*
+ * Transmit a message. The fh argument may be NULL if the transmit is not
+ * associated with a specific filehandle.
+ *
+ * This function is called with adap->lock held.
+ */
+int cec_transmit_msg_fh(struct cec_adapter *adap, struct cec_msg *msg,
+                       struct cec_fh *fh, bool block)
+{
+       struct cec_data *data;
+       u8 last_initiator = 0xff;
+       unsigned int timeout;
+       int res = 0;
+
+       if (msg->reply && msg->timeout == 0) {
+               /* Make sure the timeout isn't 0. */
+               msg->timeout = 1000;
+       }
+
+       /* Sanity checks */
+       if (msg->len == 0 || msg->len > CEC_MAX_MSG_SIZE) {
+               dprintk(1, "cec_transmit_msg: invalid length %d\n", msg->len);
+               return -EINVAL;
+       }
+       if (msg->timeout && msg->len == 1) {
+               dprintk(1, "cec_transmit_msg: can't reply for poll msg\n");
+               return -EINVAL;
+       }
+       if (msg->len == 1) {
+               if (cec_msg_initiator(msg) != 0xf ||
+                   cec_msg_destination(msg) == 0xf) {
+                       dprintk(1, "cec_transmit_msg: invalid poll message\n");
+                       return -EINVAL;
+               }
+               if (cec_has_log_addr(adap, cec_msg_destination(msg))) {
+                       /*
+                        * If the destination is a logical address our adapter
+                        * has already claimed, then just NACK this.
+                        * It depends on the hardware what it will do with a
+                        * POLL to itself (some OK this), so it is just as
+                        * easy to handle it here so the behavior will be
+                        * consistent.
+                        */
+                       msg->tx_status = CEC_TX_STATUS_NACK |
+                                        CEC_TX_STATUS_MAX_RETRIES;
+                       msg->tx_nack_cnt = 1;
+                       return 0;
+               }
+       }
+       if (msg->len > 1 && !cec_msg_is_broadcast(msg) &&
+           cec_has_log_addr(adap, cec_msg_destination(msg))) {
+               dprintk(1, "cec_transmit_msg: destination is the adapter itself\n");
+               return -EINVAL;
+       }
+       if (cec_msg_initiator(msg) != 0xf &&
+           !cec_has_log_addr(adap, cec_msg_initiator(msg))) {
+               dprintk(1, "cec_transmit_msg: initiator has unknown logical address %d\n",
+                       cec_msg_initiator(msg));
+               return -EINVAL;
+       }
+       if (!adap->is_configured && !adap->is_configuring)
+               return -ENONET;
+
+       data = kzalloc(sizeof(*data), GFP_KERNEL);
+       if (!data)
+               return -ENOMEM;
+
+       if (msg->len > 1 && msg->msg[1] == CEC_MSG_CDC_MESSAGE) {
+               msg->msg[2] = adap->phys_addr >> 8;
+               msg->msg[3] = adap->phys_addr & 0xff;
+       }
+
+       if (msg->timeout)
+               dprintk(2, "cec_transmit_msg: %*ph (wait for 0x%02x%s)\n",
+                       msg->len, msg->msg, msg->reply, !block ? ", nb" : "");
+       else
+               dprintk(2, "cec_transmit_msg: %*ph%s\n",
+                       msg->len, msg->msg, !block ? " (nb)" : "");
+
+       msg->rx_status = 0;
+       msg->tx_status = 0;
+       msg->tx_arb_lost_cnt = 0;
+       msg->tx_nack_cnt = 0;
+       msg->tx_low_drive_cnt = 0;
+       msg->tx_error_cnt = 0;
+       data->msg = *msg;
+       data->fh = fh;
+       data->adap = adap;
+       data->blocking = block;
+
+       /*
+        * Determine if this message follows a message from the same
+        * initiator. Needed to determine the free signal time later on.
+        */
+       if (msg->len > 1) {
+               if (!(list_empty(&adap->transmit_queue))) {
+                       const struct cec_data *last;
+
+                       last = list_last_entry(&adap->transmit_queue,
+                                              const struct cec_data, list);
+                       last_initiator = cec_msg_initiator(&last->msg);
+               } else if (adap->transmitting) {
+                       last_initiator =
+                               cec_msg_initiator(&adap->transmitting->msg);
+               }
+       }
+       data->new_initiator = last_initiator != cec_msg_initiator(msg);
+       init_completion(&data->c);
+       INIT_DELAYED_WORK(&data->work, cec_wait_timeout);
+
+       data->msg.sequence = adap->sequence++;
+       if (fh)
+               list_add_tail(&data->xfer_list, &fh->xfer_list);
+       list_add_tail(&data->list, &adap->transmit_queue);
+       if (!adap->transmitting)
+               wake_up_interruptible(&adap->kthread_waitq);
+
+       /* All done if we don't need to block waiting for completion */
+       if (!block)
+               return 0;
+
+       /*
+        * If we don't get a completion before this time something is really
+        * wrong and we time out.
+        */
+       timeout = CEC_XFER_TIMEOUT_MS;
+       /* Add the requested timeout if we have to wait for a reply as well */
+       if (msg->timeout)
+               timeout += msg->timeout;
+
+       /*
+        * Release the lock and wait, retake the lock afterwards.
+        */
+       mutex_unlock(&adap->lock);
+       res = wait_for_completion_killable_timeout(&data->c,
+                                                  msecs_to_jiffies(timeout));
+       mutex_lock(&adap->lock);
+
+       if (data->completed) {
+               /* The transmit completed (possibly with an error) */
+               *msg = data->msg;
+               kfree(data);
+               return 0;
+       }
+       /*
+        * The wait for completion timed out or was interrupted, so mark this
+        * as non-blocking and disconnect from the filehandle since it is
+        * still 'in flight'. When it finally completes it will just drop the
+        * result silently.
+        */
+       data->blocking = false;
+       if (data->fh)
+               list_del(&data->xfer_list);
+       data->fh = NULL;
+
+       if (res == 0) { /* timed out */
+               /* Check if the reply or the transmit failed */
+               if (msg->timeout && (msg->tx_status & CEC_TX_STATUS_OK))
+                       msg->rx_status = CEC_RX_STATUS_TIMEOUT;
+               else
+                       msg->tx_status = CEC_TX_STATUS_MAX_RETRIES;
+       }
+       return res > 0 ? 0 : res;
+}
+
+/* Helper function to be used by drivers and this framework. */
+int cec_transmit_msg(struct cec_adapter *adap, struct cec_msg *msg,
+                    bool block)
+{
+       int ret;
+
+       mutex_lock(&adap->lock);
+       ret = cec_transmit_msg_fh(adap, msg, NULL, block);
+       mutex_unlock(&adap->lock);
+       return ret;
+}
+EXPORT_SYMBOL_GPL(cec_transmit_msg);
+
+/*
+ * I don't like forward references but without this the low-level
+ * cec_received_msg() function would come after a bunch of high-level
+ * CEC protocol handling functions. That was very confusing.
+ */
+static int cec_receive_notify(struct cec_adapter *adap, struct cec_msg *msg,
+                             bool is_reply);
+
+/* Called by the CEC adapter if a message is received */
+void cec_received_msg(struct cec_adapter *adap, struct cec_msg *msg)
+{
+       struct cec_data *data;
+       u8 msg_init = cec_msg_initiator(msg);
+       u8 msg_dest = cec_msg_destination(msg);
+       bool is_reply = false;
+       bool valid_la = true;
+
+       mutex_lock(&adap->lock);
+       msg->ts = ktime_get_ns();
+       msg->rx_status = CEC_RX_STATUS_OK;
+       msg->tx_status = 0;
+       msg->sequence = msg->reply = msg->timeout = 0;
+       msg->flags = 0;
+
+       dprintk(2, "cec_received_msg: %*ph\n", msg->len, msg->msg);
+
+       /* Check if this message was for us (directed or broadcast). */
+       if (!cec_msg_is_broadcast(msg))
+               valid_la = cec_has_log_addr(adap, msg_dest);
+
+       /* It's a valid message and not a poll or CDC message */
+       if (valid_la && msg->len > 1 && msg->msg[1] != CEC_MSG_CDC_MESSAGE) {
+               u8 cmd = msg->msg[1];
+               bool abort = cmd == CEC_MSG_FEATURE_ABORT;
+
+               /* The aborted command is in msg[2] */
+               if (abort)
+                       cmd = msg->msg[2];
+
+               /*
+                * Walk over all transmitted messages that are waiting for a
+                * reply.
+                */
+               list_for_each_entry(data, &adap->wait_queue, list) {
+                       struct cec_msg *dst = &data->msg;
+                       u8 dst_reply;
+
+                       /* Does the command match? */
+                       if ((abort && cmd != dst->msg[1]) ||
+                           (!abort && cmd != dst->reply))
+                               continue;
+
+                       /* Does the addressing match? */
+                       if (msg_init != cec_msg_destination(dst) &&
+                           !cec_msg_is_broadcast(dst))
+                               continue;
+
+                       /* We got a reply */
+                       msg->sequence = dst->sequence;
+                       dst_reply = dst->reply;
+                       *dst = *msg;
+                       dst->reply = dst_reply;
+                       if (abort) {
+                               dst->reply = 0;
+                               dst->rx_status |= CEC_RX_STATUS_FEATURE_ABORT;
+                       }
+                       /* Remove it from the wait_queue */
+                       list_del_init(&data->list);
+
+                       /* Cancel the pending timeout work */
+                       if (!cancel_delayed_work(&data->work)) {
+                               mutex_unlock(&adap->lock);
+                               flush_scheduled_work();
+                               mutex_lock(&adap->lock);
+                       }
+                       /*
+                        * Mark this as a reply, provided someone is still
+                        * waiting for the answer.
+                        */
+                       if (data->fh)
+                               is_reply = true;
+                       cec_data_completed(data);
+                       break;
+               }
+       }
+       mutex_unlock(&adap->lock);
+
+       /* Pass the message on to any monitoring filehandles */
+       cec_queue_msg_monitor(adap, msg, valid_la);
+
+       /* We're done if it is not for us or a poll message */
+       if (!valid_la || msg->len <= 1)
+               return;
+
+       /*
+        * Process the message on the protocol level. If is_reply is true,
+        * then cec_receive_notify() won't pass on the reply to the listener(s)
+        * since that was already done by cec_data_completed() above.
+        */
+       cec_receive_notify(adap, msg, is_reply);
+}
+EXPORT_SYMBOL_GPL(cec_received_msg);
+
+/* Logical Address Handling */
+
+/*
+ * Attempt to claim a specific logical address.
+ *
+ * This function is called with adap->lock held.
+ */
+static int cec_config_log_addr(struct cec_adapter *adap,
+                              unsigned int idx,
+                              unsigned int log_addr)
+{
+       struct cec_log_addrs *las = &adap->log_addrs;
+       struct cec_msg msg = { };
+       int err;
+
+       if (cec_has_log_addr(adap, log_addr))
+               return 0;
+
+       /* Send poll message */
+       msg.len = 1;
+       msg.msg[0] = 0xf0 | log_addr;
+       err = cec_transmit_msg_fh(adap, &msg, NULL, true);
+
+       /*
+        * While trying to poll the physical address was reset
+        * and the adapter was unconfigured, so bail out.
+        */
+       if (!adap->is_configuring)
+               return -EINTR;
+
+       if (err)
+               return err;
+
+       if (msg.tx_status & CEC_TX_STATUS_OK)
+               return 0;
+
+       /*
+        * Message not acknowledged, so this logical
+        * address is free to use.
+        */
+       err = adap->ops->adap_log_addr(adap, log_addr);
+       if (err)
+               return err;
+
+       las->log_addr[idx] = log_addr;
+       las->log_addr_mask |= 1 << log_addr;
+       adap->phys_addrs[log_addr] = adap->phys_addr;
+
+       dprintk(2, "claimed addr %d (%d)\n", log_addr,
+               las->primary_device_type[idx]);
+       return 1;
+}
+
+/*
+ * Unconfigure the adapter: clear all logical addresses and send
+ * the state changed event.
+ *
+ * This function is called with adap->lock held.
+ */
+static void cec_adap_unconfigure(struct cec_adapter *adap)
+{
+       WARN_ON(adap->ops->adap_log_addr(adap, CEC_LOG_ADDR_INVALID));
+       adap->log_addrs.log_addr_mask = 0;
+       adap->is_configuring = false;
+       adap->is_configured = false;
+       memset(adap->phys_addrs, 0xff, sizeof(adap->phys_addrs));
+       wake_up_interruptible(&adap->kthread_waitq);
+       cec_post_state_event(adap);
+}
+
+/*
+ * Attempt to claim the required logical addresses.
+ */
+static int cec_config_thread_func(void *arg)
+{
+       /* The various LAs for each type of device */
+       static const u8 tv_log_addrs[] = {
+               CEC_LOG_ADDR_TV, CEC_LOG_ADDR_SPECIFIC,
+               CEC_LOG_ADDR_INVALID
+       };
+       static const u8 record_log_addrs[] = {
+               CEC_LOG_ADDR_RECORD_1, CEC_LOG_ADDR_RECORD_2,
+               CEC_LOG_ADDR_RECORD_3,
+               CEC_LOG_ADDR_BACKUP_1, CEC_LOG_ADDR_BACKUP_2,
+               CEC_LOG_ADDR_INVALID
+       };
+       static const u8 tuner_log_addrs[] = {
+               CEC_LOG_ADDR_TUNER_1, CEC_LOG_ADDR_TUNER_2,
+               CEC_LOG_ADDR_TUNER_3, CEC_LOG_ADDR_TUNER_4,
+               CEC_LOG_ADDR_BACKUP_1, CEC_LOG_ADDR_BACKUP_2,
+               CEC_LOG_ADDR_INVALID
+       };
+       static const u8 playback_log_addrs[] = {
+               CEC_LOG_ADDR_PLAYBACK_1, CEC_LOG_ADDR_PLAYBACK_2,
+               CEC_LOG_ADDR_PLAYBACK_3,
+               CEC_LOG_ADDR_BACKUP_1, CEC_LOG_ADDR_BACKUP_2,
+               CEC_LOG_ADDR_INVALID
+       };
+       static const u8 audiosystem_log_addrs[] = {
+               CEC_LOG_ADDR_AUDIOSYSTEM,
+               CEC_LOG_ADDR_INVALID
+       };
+       static const u8 specific_use_log_addrs[] = {
+               CEC_LOG_ADDR_SPECIFIC,
+               CEC_LOG_ADDR_BACKUP_1, CEC_LOG_ADDR_BACKUP_2,
+               CEC_LOG_ADDR_INVALID
+       };
+       static const u8 *type2addrs[6] = {
+               [CEC_LOG_ADDR_TYPE_TV] = tv_log_addrs,
+               [CEC_LOG_ADDR_TYPE_RECORD] = record_log_addrs,
+               [CEC_LOG_ADDR_TYPE_TUNER] = tuner_log_addrs,
+               [CEC_LOG_ADDR_TYPE_PLAYBACK] = playback_log_addrs,
+               [CEC_LOG_ADDR_TYPE_AUDIOSYSTEM] = audiosystem_log_addrs,
+               [CEC_LOG_ADDR_TYPE_SPECIFIC] = specific_use_log_addrs,
+       };
+       static const u16 type2mask[] = {
+               [CEC_LOG_ADDR_TYPE_TV] = CEC_LOG_ADDR_MASK_TV,
+               [CEC_LOG_ADDR_TYPE_RECORD] = CEC_LOG_ADDR_MASK_RECORD,
+               [CEC_LOG_ADDR_TYPE_TUNER] = CEC_LOG_ADDR_MASK_TUNER,
+               [CEC_LOG_ADDR_TYPE_PLAYBACK] = CEC_LOG_ADDR_MASK_PLAYBACK,
+               [CEC_LOG_ADDR_TYPE_AUDIOSYSTEM] = CEC_LOG_ADDR_MASK_AUDIOSYSTEM,
+               [CEC_LOG_ADDR_TYPE_SPECIFIC] = CEC_LOG_ADDR_MASK_SPECIFIC,
+       };
+       struct cec_adapter *adap = arg;
+       struct cec_log_addrs *las = &adap->log_addrs;
+       int err;
+       int i, j;
+
+       mutex_lock(&adap->lock);
+       dprintk(1, "physical address: %x.%x.%x.%x, claim %d logical addresses\n",
+               cec_phys_addr_exp(adap->phys_addr), las->num_log_addrs);
+       las->log_addr_mask = 0;
+
+       if (las->log_addr_type[0] == CEC_LOG_ADDR_TYPE_UNREGISTERED)
+               goto configured;
+
+       for (i = 0; i < las->num_log_addrs; i++) {
+               unsigned int type = las->log_addr_type[i];
+               const u8 *la_list;
+               u8 last_la;
+
+               /*
+                * The TV functionality can only map to physical address 0.
+                * For any other address, try the Specific functionality
+                * instead as per the spec.
+                */
+               if (adap->phys_addr && type == CEC_LOG_ADDR_TYPE_TV)
+                       type = CEC_LOG_ADDR_TYPE_SPECIFIC;
+
+               la_list = type2addrs[type];
+               last_la = las->log_addr[i];
+               las->log_addr[i] = CEC_LOG_ADDR_INVALID;
+               if (last_la == CEC_LOG_ADDR_INVALID ||
+                   last_la == CEC_LOG_ADDR_UNREGISTERED ||
+                   !(last_la & type2mask[type]))
+                       last_la = la_list[0];
+
+               err = cec_config_log_addr(adap, i, last_la);
+               if (err > 0) /* Reused last LA */
+                       continue;
+
+               if (err < 0)
+                       goto unconfigure;
+
+               for (j = 0; la_list[j] != CEC_LOG_ADDR_INVALID; j++) {
+                       /* Tried this one already, skip it */
+                       if (la_list[j] == last_la)
+                               continue;
+                       /* The backup addresses are CEC 2.0 specific */
+                       if ((la_list[j] == CEC_LOG_ADDR_BACKUP_1 ||
+                            la_list[j] == CEC_LOG_ADDR_BACKUP_2) &&
+                           las->cec_version < CEC_OP_CEC_VERSION_2_0)
+                               continue;
+
+                       err = cec_config_log_addr(adap, i, la_list[j]);
+                       if (err == 0) /* LA is in use */
+                               continue;
+                       if (err < 0)
+                               goto unconfigure;
+                       /* Done, claimed an LA */
+                       break;
+               }
+
+               if (la_list[j] == CEC_LOG_ADDR_INVALID)
+                       dprintk(1, "could not claim LA %d\n", i);
+       }
+
+configured:
+       if (adap->log_addrs.log_addr_mask == 0) {
+               /* Fall back to unregistered */
+               las->log_addr[0] = CEC_LOG_ADDR_UNREGISTERED;
+               las->log_addr_mask = 1 << las->log_addr[0];
+       }
+       adap->is_configured = true;
+       adap->is_configuring = false;
+       cec_post_state_event(adap);
+       mutex_unlock(&adap->lock);
+
+       for (i = 0; i < las->num_log_addrs; i++) {
+               if (las->log_addr[i] == CEC_LOG_ADDR_INVALID)
+                       continue;
+
+               /*
+                * Report Features must come first according
+                * to CEC 2.0
+                */
+               if (las->log_addr[i] != CEC_LOG_ADDR_UNREGISTERED)
+                       cec_report_features(adap, i);
+               cec_report_phys_addr(adap, i);
+       }
+       mutex_lock(&adap->lock);
+       adap->kthread_config = NULL;
+       mutex_unlock(&adap->lock);
+       complete(&adap->config_completion);
+       return 0;
+
+unconfigure:
+       for (i = 0; i < las->num_log_addrs; i++)
+               las->log_addr[i] = CEC_LOG_ADDR_INVALID;
+       cec_adap_unconfigure(adap);
+       adap->kthread_config = NULL;
+       mutex_unlock(&adap->lock);
+       complete(&adap->config_completion);
+       return 0;
+}
+
+/*
+ * Called from either __cec_s_phys_addr or __cec_s_log_addrs to claim the
+ * logical addresses.
+ *
+ * This function is called with adap->lock held.
+ */
+static void cec_claim_log_addrs(struct cec_adapter *adap, bool block)
+{
+       if (WARN_ON(adap->is_configuring || adap->is_configured))
+               return;
+
+       init_completion(&adap->config_completion);
+
+       /* Ready to kick off the thread */
+       adap->is_configuring = true;
+       adap->kthread_config = kthread_run(cec_config_thread_func, adap,
+                                          "ceccfg-%s", adap->name);
+       if (IS_ERR(adap->kthread_config)) {
+               adap->kthread_config = NULL;
+       } else if (block) {
+               mutex_unlock(&adap->lock);
+               wait_for_completion(&adap->config_completion);
+               mutex_lock(&adap->lock);
+       }
+}
+
+/* Set a new physical address and send an event notifying userspace of this.
+ *
+ * This function is called with adap->lock held.
+ */
+void __cec_s_phys_addr(struct cec_adapter *adap, u16 phys_addr, bool block)
+{
+       if (phys_addr == adap->phys_addr)
+               return;
+
+       if (phys_addr == CEC_PHYS_ADDR_INVALID ||
+           adap->phys_addr != CEC_PHYS_ADDR_INVALID) {
+               adap->phys_addr = CEC_PHYS_ADDR_INVALID;
+               cec_post_state_event(adap);
+               cec_adap_unconfigure(adap);
+               /* Disabling monitor all mode should always succeed */
+               if (adap->monitor_all_cnt)
+                       WARN_ON(call_op(adap, adap_monitor_all_enable, false));
+               WARN_ON(adap->ops->adap_enable(adap, false));
+               if (phys_addr == CEC_PHYS_ADDR_INVALID)
+                       return;
+       }
+
+       if (adap->ops->adap_enable(adap, true))
+               return;
+
+       if (adap->monitor_all_cnt &&
+           call_op(adap, adap_monitor_all_enable, true)) {
+               WARN_ON(adap->ops->adap_enable(adap, false));
+               return;
+       }
+       adap->phys_addr = phys_addr;
+       cec_post_state_event(adap);
+       if (adap->log_addrs.num_log_addrs)
+               cec_claim_log_addrs(adap, block);
+}
+
+void cec_s_phys_addr(struct cec_adapter *adap, u16 phys_addr, bool block)
+{
+       if (IS_ERR_OR_NULL(adap))
+               return;
+
+       if (WARN_ON(adap->capabilities & CEC_CAP_PHYS_ADDR))
+               return;
+       mutex_lock(&adap->lock);
+       __cec_s_phys_addr(adap, phys_addr, block);
+       mutex_unlock(&adap->lock);
+}
+EXPORT_SYMBOL_GPL(cec_s_phys_addr);
+
+/*
+ * Called from either the ioctl or a driver to set the logical addresses.
+ *
+ * This function is called with adap->lock held.
+ */
+int __cec_s_log_addrs(struct cec_adapter *adap,
+                     struct cec_log_addrs *log_addrs, bool block)
+{
+       u16 type_mask = 0;
+       int i;
+
+       if (!log_addrs || log_addrs->num_log_addrs == 0) {
+               adap->log_addrs.num_log_addrs = 0;
+               cec_adap_unconfigure(adap);
+               return 0;
+       }
+
+       /* Ensure the osd name is 0-terminated */
+       log_addrs->osd_name[sizeof(log_addrs->osd_name) - 1] = '\0';
+
+       /* Sanity checks */
+       if (log_addrs->num_log_addrs > adap->available_log_addrs) {
+               dprintk(1, "num_log_addrs > %d\n", adap->available_log_addrs);
+               return -EINVAL;
+       }
+
+       /*
+        * Vendor ID is a 24 bit number, so check if the value is
+        * within the correct range.
+        */
+       if (log_addrs->vendor_id != CEC_VENDOR_ID_NONE &&
+           (log_addrs->vendor_id & 0xff000000) != 0)
+               return -EINVAL;
+
+       if (log_addrs->cec_version != CEC_OP_CEC_VERSION_1_4 &&
+           log_addrs->cec_version != CEC_OP_CEC_VERSION_2_0)
+               return -EINVAL;
+
+       if (log_addrs->num_log_addrs > 1)
+               for (i = 0; i < log_addrs->num_log_addrs; i++)
+                       if (log_addrs->log_addr_type[i] ==
+                                       CEC_LOG_ADDR_TYPE_UNREGISTERED) {
+                               dprintk(1, "num_log_addrs > 1 can't be combined with unregistered LA\n");
+                               return -EINVAL;
+                       }
+
+       if (log_addrs->cec_version < CEC_OP_CEC_VERSION_2_0) {
+               memset(log_addrs->all_device_types, 0,
+                      sizeof(log_addrs->all_device_types));
+               memset(log_addrs->features, 0, sizeof(log_addrs->features));
+       }
+
+       for (i = 0; i < log_addrs->num_log_addrs; i++) {
+               u8 *features = log_addrs->features[i];
+               bool op_is_dev_features = false;
+
+               log_addrs->log_addr[i] = CEC_LOG_ADDR_INVALID;
+               if (type_mask & (1 << log_addrs->log_addr_type[i])) {
+                       dprintk(1, "duplicate logical address type\n");
+                       return -EINVAL;
+               }
+               type_mask |= 1 << log_addrs->log_addr_type[i];
+               if ((type_mask & (1 << CEC_LOG_ADDR_TYPE_RECORD)) &&
+                   (type_mask & (1 << CEC_LOG_ADDR_TYPE_PLAYBACK))) {
+                       /* Record already contains the playback functionality */
+                       dprintk(1, "invalid record + playback combination\n");
+                       return -EINVAL;
+               }
+               if (log_addrs->primary_device_type[i] >
+                                       CEC_OP_PRIM_DEVTYPE_PROCESSOR) {
+                       dprintk(1, "unknown primary device type\n");
+                       return -EINVAL;
+               }
+               if (log_addrs->primary_device_type[i] == 2) {
+                       dprintk(1, "invalid primary device type\n");
+                       return -EINVAL;
+               }
+               if (log_addrs->log_addr_type[i] > CEC_LOG_ADDR_TYPE_UNREGISTERED) {
+                       dprintk(1, "unknown logical address type\n");
+                       return -EINVAL;
+               }
+               if (log_addrs->cec_version < CEC_OP_CEC_VERSION_2_0)
+                       continue;
+
+               for (i = 0; i < ARRAY_SIZE(log_addrs->features[0]); i++) {
+                       if ((features[i] & 0x80) == 0) {
+                               if (op_is_dev_features)
+                                       break;
+                               op_is_dev_features = true;
+                       }
+               }
+               if (!op_is_dev_features ||
+                   i == ARRAY_SIZE(log_addrs->features[0])) {
+                       dprintk(1, "malformed features\n");
+                       return -EINVAL;
+               }
+       }
+
+       if (log_addrs->cec_version >= CEC_OP_CEC_VERSION_2_0) {
+               if (log_addrs->num_log_addrs > 2) {
+                       dprintk(1, "CEC 2.0 allows no more than 2 logical addresses\n");
+                       return -EINVAL;
+               }
+               if (log_addrs->num_log_addrs == 2) {
+                       if (!(type_mask & ((1 << CEC_LOG_ADDR_TYPE_AUDIOSYSTEM) |
+                                          (1 << CEC_LOG_ADDR_TYPE_TV)))) {
+                               dprintk(1, "Two LAs is only allowed for audiosystem and TV\n");
+                               return -EINVAL;
+                       }
+                       if (!(type_mask & ((1 << CEC_LOG_ADDR_TYPE_PLAYBACK) |
+                                          (1 << CEC_LOG_ADDR_TYPE_RECORD)))) {
+                               dprintk(1, "An audiosystem/TV can only be combined with record or playback\n");
+                               return -EINVAL;
+                       }
+               }
+       }
+
+       log_addrs->log_addr_mask = adap->log_addrs.log_addr_mask;
+       adap->log_addrs = *log_addrs;
+       if (adap->phys_addr != CEC_PHYS_ADDR_INVALID)
+               cec_claim_log_addrs(adap, block);
+       return 0;
+}
+
+int cec_s_log_addrs(struct cec_adapter *adap,
+                   struct cec_log_addrs *log_addrs, bool block)
+{
+       int err;
+
+       if (WARN_ON(adap->capabilities & CEC_CAP_LOG_ADDRS))
+               return -EINVAL;
+       mutex_lock(&adap->lock);
+       err = __cec_s_log_addrs(adap, log_addrs, block);
+       mutex_unlock(&adap->lock);
+       return err;
+}
+EXPORT_SYMBOL_GPL(cec_s_log_addrs);
+
+/* High-level core CEC message handling */
+
+/* Transmit the Report Features message */
+static int cec_report_features(struct cec_adapter *adap, unsigned int la_idx)
+{
+       struct cec_msg msg = { };
+       const struct cec_log_addrs *las = &adap->log_addrs;
+       const u8 *features = las->features[la_idx];
+       bool op_is_dev_features = false;
+       unsigned int idx;
+
+       /* This is 2.0 and up only */
+       if (adap->log_addrs.cec_version < CEC_OP_CEC_VERSION_2_0)
+               return 0;
+
+       /* Report Features */
+       msg.msg[0] = (las->log_addr[la_idx] << 4) | 0x0f;
+       msg.len = 4;
+       msg.msg[1] = CEC_MSG_REPORT_FEATURES;
+       msg.msg[2] = adap->log_addrs.cec_version;
+       msg.msg[3] = las->all_device_types[la_idx];
+
+       /* Write RC Profiles first, then Device Features */
+       for (idx = 0; idx < ARRAY_SIZE(las->features[0]); idx++) {
+               msg.msg[msg.len++] = features[idx];
+               if ((features[idx] & CEC_OP_FEAT_EXT) == 0) {
+                       if (op_is_dev_features)
+                               break;
+                       op_is_dev_features = true;
+               }
+       }
+       return cec_transmit_msg(adap, &msg, false);
+}
+
+/* Transmit the Report Physical Address message */
+static int cec_report_phys_addr(struct cec_adapter *adap, unsigned int la_idx)
+{
+       const struct cec_log_addrs *las = &adap->log_addrs;
+       struct cec_msg msg = { };
+
+       /* Report Physical Address */
+       msg.msg[0] = (las->log_addr[la_idx] << 4) | 0x0f;
+       cec_msg_report_physical_addr(&msg, adap->phys_addr,
+                                    las->primary_device_type[la_idx]);
+       dprintk(2, "config: la %d pa %x.%x.%x.%x\n",
+               las->log_addr[la_idx],
+                       cec_phys_addr_exp(adap->phys_addr));
+       return cec_transmit_msg(adap, &msg, false);
+}
+
+/* Transmit the Feature Abort message */
+static int cec_feature_abort_reason(struct cec_adapter *adap,
+                                   struct cec_msg *msg, u8 reason)
+{
+       struct cec_msg tx_msg = { };
+
+       /*
+        * Don't reply with CEC_MSG_FEATURE_ABORT to a CEC_MSG_FEATURE_ABORT
+        * message!
+        */
+       if (msg->msg[1] == CEC_MSG_FEATURE_ABORT)
+               return 0;
+       cec_msg_set_reply_to(&tx_msg, msg);
+       cec_msg_feature_abort(&tx_msg, msg->msg[1], reason);
+       return cec_transmit_msg(adap, &tx_msg, false);
+}
+
+static int cec_feature_abort(struct cec_adapter *adap, struct cec_msg *msg)
+{
+       return cec_feature_abort_reason(adap, msg,
+                                       CEC_OP_ABORT_UNRECOGNIZED_OP);
+}
+
+static int cec_feature_refused(struct cec_adapter *adap, struct cec_msg *msg)
+{
+       return cec_feature_abort_reason(adap, msg,
+                                       CEC_OP_ABORT_REFUSED);
+}
+
+/*
+ * Called when a CEC message is received. This function will do any
+ * necessary core processing. The is_reply bool is true if this message
+ * is a reply to an earlier transmit.
+ *
+ * The message is either a broadcast message or a valid directed message.
+ */
+static int cec_receive_notify(struct cec_adapter *adap, struct cec_msg *msg,
+                             bool is_reply)
+{
+       bool is_broadcast = cec_msg_is_broadcast(msg);
+       u8 dest_laddr = cec_msg_destination(msg);
+       u8 init_laddr = cec_msg_initiator(msg);
+       u8 devtype = cec_log_addr2dev(adap, dest_laddr);
+       int la_idx = cec_log_addr2idx(adap, dest_laddr);
+       bool is_directed = la_idx >= 0;
+       bool from_unregistered = init_laddr == 0xf;
+       struct cec_msg tx_cec_msg = { };
+
+       dprintk(1, "cec_receive_notify: %*ph\n", msg->len, msg->msg);
+
+       if (adap->ops->received) {
+               /* Allow drivers to process the message first */
+               if (adap->ops->received(adap, msg) != -ENOMSG)
+                       return 0;
+       }
+
+       /*
+        * REPORT_PHYSICAL_ADDR, CEC_MSG_USER_CONTROL_PRESSED and
+        * CEC_MSG_USER_CONTROL_RELEASED messages always have to be
+        * handled by the CEC core, even if the passthrough mode is on.
+        * The others are just ignored if passthrough mode is on.
+        */
+       switch (msg->msg[1]) {
+       case CEC_MSG_GET_CEC_VERSION:
+       case CEC_MSG_GIVE_DEVICE_VENDOR_ID:
+       case CEC_MSG_ABORT:
+       case CEC_MSG_GIVE_DEVICE_POWER_STATUS:
+       case CEC_MSG_GIVE_PHYSICAL_ADDR:
+       case CEC_MSG_GIVE_OSD_NAME:
+       case CEC_MSG_GIVE_FEATURES:
+               /*
+                * Skip processing these messages if the passthrough mode
+                * is on.
+                */
+               if (adap->passthrough)
+                       goto skip_processing;
+               /* Ignore if addressing is wrong */
+               if (is_broadcast || from_unregistered)
+                       return 0;
+               break;
+
+       case CEC_MSG_USER_CONTROL_PRESSED:
+       case CEC_MSG_USER_CONTROL_RELEASED:
+               /* Wrong addressing mode: don't process */
+               if (is_broadcast || from_unregistered)
+                       goto skip_processing;
+               break;
+
+       case CEC_MSG_REPORT_PHYSICAL_ADDR:
+               /*
+                * This message is always processed, regardless of the
+                * passthrough setting.
+                *
+                * Exception: don't process if wrong addressing mode.
+                */
+               if (!is_broadcast)
+                       goto skip_processing;
+               break;
+
+       default:
+               break;
+       }
+
+       cec_msg_set_reply_to(&tx_cec_msg, msg);
+
+       switch (msg->msg[1]) {
+       /* The following messages are processed but still passed through */
+       case CEC_MSG_REPORT_PHYSICAL_ADDR:
+               adap->phys_addrs[init_laddr] =
+                       (msg->msg[2] << 8) | msg->msg[3];
+               dprintk(1, "Reported physical address %04x for logical address %d\n",
+                       adap->phys_addrs[init_laddr], init_laddr);
+               break;
+
+       case CEC_MSG_USER_CONTROL_PRESSED:
+               if (!(adap->capabilities & CEC_CAP_RC))
+                       break;
+
+#if IS_ENABLED(CONFIG_RC_CORE)
+               switch (msg->msg[2]) {
+               /*
+                * Play function, this message can have variable length
+                * depending on the specific play function that is used.
+                */
+               case 0x60:
+                       if (msg->len == 2)
+                               rc_keydown(adap->rc, RC_TYPE_CEC,
+                                          msg->msg[2], 0);
+                       else
+                               rc_keydown(adap->rc, RC_TYPE_CEC,
+                                          msg->msg[2] << 8 | msg->msg[3], 0);
+                       break;
+               /*
+                * Other function messages that are not handled.
+                * Currently the RC framework does not allow to supply an
+                * additional parameter to a keypress. These "keys" contain
+                * other information such as channel number, an input number
+                * etc.
+                * For the time being these messages are not processed by the
+                * framework and are simply forwarded to the user space.
+                */
+               case 0x56: case 0x57:
+               case 0x67: case 0x68: case 0x69: case 0x6a:
+                       break;
+               default:
+                       rc_keydown(adap->rc, RC_TYPE_CEC, msg->msg[2], 0);
+                       break;
+               }
+#endif
+               break;
+
+       case CEC_MSG_USER_CONTROL_RELEASED:
+               if (!(adap->capabilities & CEC_CAP_RC))
+                       break;
+#if IS_ENABLED(CONFIG_RC_CORE)
+               rc_keyup(adap->rc);
+#endif
+               break;
+
+       /*
+        * The remaining messages are only processed if the passthrough mode
+        * is off.
+        */
+       case CEC_MSG_GET_CEC_VERSION:
+               cec_msg_cec_version(&tx_cec_msg, adap->log_addrs.cec_version);
+               return cec_transmit_msg(adap, &tx_cec_msg, false);
+
+       case CEC_MSG_GIVE_PHYSICAL_ADDR:
+               /* Do nothing for CEC switches using addr 15 */
+               if (devtype == CEC_OP_PRIM_DEVTYPE_SWITCH && dest_laddr == 15)
+                       return 0;
+               cec_msg_report_physical_addr(&tx_cec_msg, adap->phys_addr, devtype);
+               return cec_transmit_msg(adap, &tx_cec_msg, false);
+
+       case CEC_MSG_GIVE_DEVICE_VENDOR_ID:
+               if (adap->log_addrs.vendor_id == CEC_VENDOR_ID_NONE)
+                       return cec_feature_abort(adap, msg);
+               cec_msg_device_vendor_id(&tx_cec_msg, adap->log_addrs.vendor_id);
+               return cec_transmit_msg(adap, &tx_cec_msg, false);
+
+       case CEC_MSG_ABORT:
+               /* Do nothing for CEC switches */
+               if (devtype == CEC_OP_PRIM_DEVTYPE_SWITCH)
+                       return 0;
+               return cec_feature_refused(adap, msg);
+
+       case CEC_MSG_GIVE_OSD_NAME: {
+               if (adap->log_addrs.osd_name[0] == 0)
+                       return cec_feature_abort(adap, msg);
+               cec_msg_set_osd_name(&tx_cec_msg, adap->log_addrs.osd_name);
+               return cec_transmit_msg(adap, &tx_cec_msg, false);
+       }
+
+       case CEC_MSG_GIVE_FEATURES:
+               if (adap->log_addrs.cec_version >= CEC_OP_CEC_VERSION_2_0)
+                       return cec_report_features(adap, la_idx);
+               return 0;
+
+       default:
+               /*
+                * Unprocessed messages are aborted if userspace isn't doing
+                * any processing either.
+                */
+               if (is_directed && !is_reply && !adap->follower_cnt &&
+                   !adap->cec_follower && msg->msg[1] != CEC_MSG_FEATURE_ABORT)
+                       return cec_feature_abort(adap, msg);
+               break;
+       }
+
+skip_processing:
+       /* If this was a reply, then we're done */
+       if (is_reply)
+               return 0;
+
+       /*
+        * Send to the exclusive follower if there is one, otherwise send
+        * to all followers.
+        */
+       if (adap->cec_follower)
+               cec_queue_msg_fh(adap->cec_follower, msg);
+       else
+               cec_queue_msg_followers(adap, msg);
+       return 0;
+}
+
+/*
+ * Helper functions to keep track of the 'monitor all' use count.
+ *
+ * These functions are called with adap->lock held.
+ */
+int cec_monitor_all_cnt_inc(struct cec_adapter *adap)
+{
+       int ret = 0;
+
+       if (adap->monitor_all_cnt == 0)
+               ret = call_op(adap, adap_monitor_all_enable, 1);
+       if (ret == 0)
+               adap->monitor_all_cnt++;
+       return ret;
+}
+
+void cec_monitor_all_cnt_dec(struct cec_adapter *adap)
+{
+       adap->monitor_all_cnt--;
+       if (adap->monitor_all_cnt == 0)
+               WARN_ON(call_op(adap, adap_monitor_all_enable, 0));
+}
+
+#ifdef CONFIG_MEDIA_CEC_DEBUG
+/*
+ * Log the current state of the CEC adapter.
+ * Very useful for debugging.
+ */
+int cec_adap_status(struct seq_file *file, void *priv)
+{
+       struct cec_adapter *adap = dev_get_drvdata(file->private);
+       struct cec_data *data;
+
+       mutex_lock(&adap->lock);
+       seq_printf(file, "configured: %d\n", adap->is_configured);
+       seq_printf(file, "configuring: %d\n", adap->is_configuring);
+       seq_printf(file, "phys_addr: %x.%x.%x.%x\n",
+                  cec_phys_addr_exp(adap->phys_addr));
+       seq_printf(file, "number of LAs: %d\n", adap->log_addrs.num_log_addrs);
+       seq_printf(file, "LA mask: 0x%04x\n", adap->log_addrs.log_addr_mask);
+       if (adap->cec_follower)
+               seq_printf(file, "has CEC follower%s\n",
+                          adap->passthrough ? " (in passthrough mode)" : "");
+       if (adap->cec_initiator)
+               seq_puts(file, "has CEC initiator\n");
+       if (adap->monitor_all_cnt)
+               seq_printf(file, "file handles in Monitor All mode: %u\n",
+                          adap->monitor_all_cnt);
+       data = adap->transmitting;
+       if (data)
+               seq_printf(file, "transmitting message: %*ph (reply: %02x)\n",
+                          data->msg.len, data->msg.msg, data->msg.reply);
+       list_for_each_entry(data, &adap->transmit_queue, list) {
+               seq_printf(file, "queued tx message: %*ph (reply: %02x)\n",
+                          data->msg.len, data->msg.msg, data->msg.reply);
+       }
+       list_for_each_entry(data, &adap->wait_queue, list) {
+               seq_printf(file, "message waiting for reply: %*ph (reply: %02x)\n",
+                          data->msg.len, data->msg.msg, data->msg.reply);
+       }
+
+       call_void_op(adap, adap_status, file);
+       mutex_unlock(&adap->lock);
+       return 0;
+}
+#endif