mirrors/gstreamer
mirrors/gstreamer
1
1
Fork 0
mirror of https://gitlab.freedesktop.org/gstreamer/gstreamer.git synced 2024-11-18 15:51:11 +00:00
Code Issues 1 Projects Releases Wiki Activity
gstreamer/libs/gst/net/gstptpclock.c

2568 lines
79 KiB
C
Raw Normal View History

ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
/* GStreamer
* Copyright (C) 2015 Sebastian Dröge <sebastian@centricular.com>
*
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Library General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This library 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
* Library General Public License for more details.
*
* You should have received a copy of the GNU Library General Public
* License along with this library; if not, write to the
* Free Software Foundation, Inc., 51 Franklin St, Fifth Floor,
* Boston, MA 02110-1301, USA.
*/
/**
* SECTION:gstptpclock
* @short_description: Special clock that synchronizes to a remote time
* provider via PTP (IEEE1588:2008).
* @see_also: #GstClock, #GstNetClientClock, #GstPipeline
*
* GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only
* mode, that allows a GStreamer pipeline to synchronize to a PTP network
* clock in some specific domain.
*
* The PTP subsystem can be initialized with gst_ptp_init(), which then starts
* a helper process to do the actual communication via the PTP ports. This is
* required as PTP listens on ports < 1024 and thus requires special
* privileges. Once this helper process is started, the main process will
* synchronize to all PTP domains that are detected on the selected
* interfaces.
*
* gst_ptp_clock_new() then allows to create a GstClock that provides the PTP
* time from a master clock inside a specific PTP domain. This clock will only
* return valid timestamps once the timestamps in the PTP domain are known. To
ptpclock: Fix documentation a bit
2015-06-06 10:35:58 +00:00
* check this, you can use gst_clock_wait_for_sync(), the GstClock::synced
* signal and gst_clock_is_synced().
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
*
*
* To gather statistics about the PTP clock synchronization,
* gst_ptp_statistics_callback_add() can be used. This gives the application
* the possibility to collect all kinds of statistics from the clock
* synchronization.
*
* Since: 1.6
*
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include "gstptpclock.h"
#include "gstptp_private.h"
ptp: Fix build on Windows, and in general the GI build when PTP support was not available It's not going to work on Windows still, the helper process needs to be ported.
2015-06-08 08:46:24 +00:00
#ifdef HAVE_SYS_WAIT_H
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
#include <sys/wait.h>
ptp: Fix build on Windows, and in general the GI build when PTP support was not available It's not going to work on Windows still, the helper process needs to be ported.
2015-06-08 08:46:24 +00:00
#endif
#ifdef G_OS_WIN32
#include <windows.h>
#endif
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
#include <sys/types.h>
#include <unistd.h>
#include <gst/base/base.h>
GST_DEBUG_CATEGORY_STATIC (ptp_debug);
#define GST_CAT_DEFAULT (ptp_debug)
/* IEEE 1588 7.7.3.1 */
#define PTP_ANNOUNCE_RECEIPT_TIMEOUT 4
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
/* Use a running average for calculating the mean path delay instead
* of just using the last measurement. Enabling this helps in unreliable
* networks, like wifi, with often changing delays
*
* Undef for following IEEE1588-2008 by the letter
*/
#define USE_RUNNING_AVERAGE_DELAY 1
/* Filter out any measurements that are above a certain threshold compared to
* previous measurements. Enabling this helps filtering out outliers that
* happen fairly often in unreliable networks, like wifi.
*
* Undef for following IEEE1588-2008 by the letter
*/
#define USE_MEASUREMENT_FILTERING 1
/* Select the first clock from which we capture a SYNC message as the master
* clock of the domain until we are ready to run the best master clock
* algorithm. This allows faster syncing but might mean a change of the master
* clock in the beginning. As all clocks in a domain are supposed to use the
* same time, this shouldn't be much of a problem.
*
* Undef for following IEEE1588-2008 by the letter
*/
#define USE_OPPORTUNISTIC_CLOCK_SELECTION 1
ptp: Add #define to only use SYNC messages for which we can send DELAY_REQ https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-06-02 13:24:06 +00:00
/* Only consider SYNC messages for which we are allowed to send a DELAY_REQ
* afterwards. This allows better synchronization in networks with varying
* delays, as for every other SYNC message we would have to assume that it's
* the average of what we saw before. But that might be completely off
*/
#define USE_ONLY_SYNC_WITH_DELAY 1
ptp: Add median based pre-filtering of delays If the delay measurement is too far away from the median of the window of last delay measurements, we discard it. This increases accuracy on wifi a lot. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-06-03 09:04:48 +00:00
/* Filter out delay measurements that are too far away from the median of the
* last delay measurements, currently those that are more than 2 times as big.
* This increases accuracy a lot on wifi.
*/
#define USE_MEDIAN_PRE_FILTERING 1
#define MEDIAN_PRE_FILTERING_WINDOW 9
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
/* How many updates should be skipped at maximum when using USE_MEASUREMENT_FILTERING */
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
#define MAX_SKIPPED_UPDATES 5
typedef enum
{
PTP_MESSAGE_TYPE_SYNC = 0x0,
PTP_MESSAGE_TYPE_DELAY_REQ = 0x1,
PTP_MESSAGE_TYPE_PDELAY_REQ = 0x2,
PTP_MESSAGE_TYPE_PDELAY_RESP = 0x3,
PTP_MESSAGE_TYPE_FOLLOW_UP = 0x8,
PTP_MESSAGE_TYPE_DELAY_RESP = 0x9,
PTP_MESSAGE_TYPE_PDELAY_RESP_FOLLOW_UP = 0xA,
PTP_MESSAGE_TYPE_ANNOUNCE = 0xB,
PTP_MESSAGE_TYPE_SIGNALING = 0xC,
PTP_MESSAGE_TYPE_MANAGEMENT = 0xD
} PtpMessageType;
typedef struct
{
guint64 seconds_field; /* 48 bits valid */
guint32 nanoseconds_field;
} PtpTimestamp;
#define PTP_TIMESTAMP_TO_GST_CLOCK_TIME(ptp) (ptp.seconds_field * GST_SECOND + ptp.nanoseconds_field)
#define GST_CLOCK_TIME_TO_PTP_TIMESTAMP_SECONDS(gst) (((GstClockTime) gst) / GST_SECOND)
#define GST_CLOCK_TIME_TO_PTP_TIMESTAMP_NANOSECONDS(gst) (((GstClockTime) gst) % GST_SECOND)
typedef struct
{
guint64 clock_identity;
guint16 port_number;
} PtpClockIdentity;
static gint
compare_clock_identity (const PtpClockIdentity * a, const PtpClockIdentity * b)
{
if (a->clock_identity < b->clock_identity)
return -1;
else if (a->clock_identity > b->clock_identity)
return 1;
if (a->port_number < b->port_number)
return -1;
else if (a->port_number > b->port_number)
return 1;
return 0;
}
typedef struct
{
guint8 clock_class;
guint8 clock_accuracy;
guint16 offset_scaled_log_variance;
} PtpClockQuality;
typedef struct
{
guint8 transport_specific;
PtpMessageType message_type;
/* guint8 reserved; */
guint8 version_ptp;
guint16 message_length;
guint8 domain_number;
/* guint8 reserved; */
guint16 flag_field;
gint64 correction_field; /* 48.16 fixed point nanoseconds */
/* guint32 reserved; */
PtpClockIdentity source_port_identity;
guint16 sequence_id;
guint8 control_field;
gint8 log_message_interval;
union
{
struct
{
PtpTimestamp origin_timestamp;
gint16 current_utc_offset;
/* guint8 reserved; */
guint8 grandmaster_priority_1;
PtpClockQuality grandmaster_clock_quality;
guint8 grandmaster_priority_2;
guint64 grandmaster_identity;
guint16 steps_removed;
guint8 time_source;
} announce;
struct
{
PtpTimestamp origin_timestamp;
} sync;
struct
{
PtpTimestamp precise_origin_timestamp;
} follow_up;
struct
{
PtpTimestamp origin_timestamp;
} delay_req;
struct
{
PtpTimestamp receive_timestamp;
PtpClockIdentity requesting_port_identity;
} delay_resp;
} message_specific;
} PtpMessage;
static GMutex ptp_lock;
static GCond ptp_cond;
static gboolean initted = FALSE;
ptp: Fix build on Windows, and in general the GI build when PTP support was not available It's not going to work on Windows still, the helper process needs to be ported.
2015-06-08 08:46:24 +00:00
#ifdef HAVE_PTP
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
static gboolean supported = TRUE;
ptp: Fix build on Windows, and in general the GI build when PTP support was not available It's not going to work on Windows still, the helper process needs to be ported.
2015-06-08 08:46:24 +00:00
#else
static gboolean supported = FALSE;
#endif
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
static GPid ptp_helper_pid;
static GThread *ptp_helper_thread;
static GMainContext *main_context;
static GMainLoop *main_loop;
static GIOChannel *stdin_channel, *stdout_channel;
static GRand *delay_req_rand;
ptpclock: Use a system clock for the time observations instead of gst_util_get_timestamp() They don't necessarily use the same underlying clocks (e.g. on Windows), or might be configured to a different clock type (monotonic vs. real time clock). We need the values a clean system clock returns, as those are the values used by the internal clocks.
2015-06-06 12:19:21 +00:00
static GstClock *observation_system_clock;
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
static PtpClockIdentity ptp_clock_id = { GST_PTP_CLOCK_ID_NONE, 0 };
typedef struct
{
GstClockTime receive_time;
PtpClockIdentity master_clock_identity;
guint8 grandmaster_priority_1;
PtpClockQuality grandmaster_clock_quality;
guint8 grandmaster_priority_2;
guint64 grandmaster_identity;
guint16 steps_removed;
guint8 time_source;
guint16 sequence_id;
} PtpAnnounceMessage;
typedef struct
{
PtpClockIdentity master_clock_identity;
GstClockTime announce_interval; /* last interval we received */
GQueue announce_messages;
} PtpAnnounceSender;
typedef struct
{
guint domain;
PtpClockIdentity master_clock_identity;
guint16 sync_seqnum;
GstClockTime sync_recv_time_local; /* t2 */
GstClockTime sync_send_time_remote; /* t1, might be -1 if FOLLOW_UP pending */
GstClockTime follow_up_recv_time_local;
GSource *timeout_source;
guint16 delay_req_seqnum;
GstClockTime delay_req_send_time_local; /* t3, -1 if we wait for FOLLOW_UP */
GstClockTime delay_req_recv_time_remote; /* t4, -1 if we wait */
GstClockTime delay_resp_recv_time_local;
gint64 correction_field_sync; /* sum of the correction fields of SYNC/FOLLOW_UP */
gint64 correction_field_delay; /* sum of the correction fields of DELAY_RESP */
} PtpPendingSync;
static void
ptp_pending_sync_free (PtpPendingSync * sync)
{
if (sync->timeout_source)
g_source_destroy (sync->timeout_source);
g_free (sync);
}
typedef struct
{
guint domain;
GstClockTime last_ptp_time;
GstClockTime last_local_time;
gint skipped_updates;
/* Used for selecting the master/grandmaster */
GList *announce_senders;
/* Last selected master clock */
gboolean have_master_clock;
PtpClockIdentity master_clock_identity;
guint64 grandmaster_identity;
/* Last SYNC or FOLLOW_UP timestamp we received */
GstClockTime last_ptp_sync_time;
GstClockTime sync_interval;
GstClockTime mean_path_delay;
GstClockTime last_delay_req, min_delay_req_interval;
guint16 last_delay_req_seqnum;
ptp: Add median based pre-filtering of delays If the delay measurement is too far away from the median of the window of last delay measurements, we discard it. This increases accuracy on wifi a lot. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-06-03 09:04:48 +00:00
GstClockTime last_path_delays[MEDIAN_PRE_FILTERING_WINDOW];
gint last_path_delays_missing;
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
GQueue pending_syncs;
GstClock *domain_clock;
} PtpDomainData;
static GList *domain_data;
static GMutex domain_clocks_lock;
static GList *domain_clocks;
/* Protected by PTP lock */
static void emit_ptp_statistics (guint8 domain, const GstStructure * stats);
static GHookList domain_stats_hooks;
static gint domain_stats_n_hooks;
static gboolean domain_stats_hooks_initted = FALSE;
/* Converts log2 seconds to GstClockTime */
static GstClockTime
log2_to_clock_time (gint l)
{
if (l < 0)
return GST_SECOND >> (-l);
else
return GST_SECOND << l;
}
static void
dump_ptp_message (PtpMessage * msg)
{
GST_TRACE ("PTP message:");
GST_TRACE ("\ttransport_specific: %u", msg->transport_specific);
GST_TRACE ("\tmessage_type: 0x%01x", msg->message_type);
GST_TRACE ("\tversion_ptp: %u", msg->version_ptp);
GST_TRACE ("\tmessage_length: %u", msg->message_length);
GST_TRACE ("\tdomain_number: %u", msg->domain_number);
GST_TRACE ("\tflag_field: 0x%04x", msg->flag_field);
GST_TRACE ("\tcorrection_field: %" G_GINT64_FORMAT ".%03u",
(msg->correction_field / 65536),
(guint) ((msg->correction_field & 0xffff) * 1000) / 65536);
GST_TRACE ("\tsource_port_identity: 0x%016" G_GINT64_MODIFIER "x %u",
msg->source_port_identity.clock_identity,
msg->source_port_identity.port_number);
GST_TRACE ("\tsequence_id: %u", msg->sequence_id);
GST_TRACE ("\tcontrol_field: 0x%02x", msg->control_field);
GST_TRACE ("\tmessage_interval: %" GST_TIME_FORMAT,
GST_TIME_ARGS (log2_to_clock_time (msg->log_message_interval)));
switch (msg->message_type) {
case PTP_MESSAGE_TYPE_ANNOUNCE:
GST_TRACE ("\tANNOUNCE:");
GST_TRACE ("\t\torigin_timestamp: %" G_GUINT64_FORMAT ".%09u",
msg->message_specific.announce.origin_timestamp.seconds_field,
msg->message_specific.announce.origin_timestamp.nanoseconds_field);
GST_TRACE ("\t\tcurrent_utc_offset: %d",
msg->message_specific.announce.current_utc_offset);
GST_TRACE ("\t\tgrandmaster_priority_1: %u",
msg->message_specific.announce.grandmaster_priority_1);
GST_TRACE ("\t\tgrandmaster_clock_quality: 0x%02x 0x%02x %u",
msg->message_specific.announce.grandmaster_clock_quality.clock_class,
msg->message_specific.announce.
grandmaster_clock_quality.clock_accuracy,
msg->message_specific.announce.
grandmaster_clock_quality.offset_scaled_log_variance);
GST_TRACE ("\t\tgrandmaster_priority_2: %u",
msg->message_specific.announce.grandmaster_priority_2);
GST_TRACE ("\t\tgrandmaster_identity: 0x%016" G_GINT64_MODIFIER "x",
msg->message_specific.announce.grandmaster_identity);
GST_TRACE ("\t\tsteps_removed: %u",
msg->message_specific.announce.steps_removed);
GST_TRACE ("\t\ttime_source: 0x%02x",
msg->message_specific.announce.time_source);
break;
case PTP_MESSAGE_TYPE_SYNC:
GST_TRACE ("\tSYNC:");
GST_TRACE ("\t\torigin_timestamp: %" G_GUINT64_FORMAT ".%09u",
msg->message_specific.sync.origin_timestamp.seconds_field,
msg->message_specific.sync.origin_timestamp.nanoseconds_field);
break;
case PTP_MESSAGE_TYPE_FOLLOW_UP:
GST_TRACE ("\tFOLLOW_UP:");
GST_TRACE ("\t\tprecise_origin_timestamp: %" G_GUINT64_FORMAT ".%09u",
msg->message_specific.follow_up.
precise_origin_timestamp.seconds_field,
msg->message_specific.follow_up.
precise_origin_timestamp.nanoseconds_field);
break;
case PTP_MESSAGE_TYPE_DELAY_REQ:
GST_TRACE ("\tDELAY_REQ:");
GST_TRACE ("\t\torigin_timestamp: %" G_GUINT64_FORMAT ".%09u",
msg->message_specific.delay_req.origin_timestamp.seconds_field,
msg->message_specific.delay_req.origin_timestamp.nanoseconds_field);
break;
case PTP_MESSAGE_TYPE_DELAY_RESP:
GST_TRACE ("\tDELAY_RESP:");
GST_TRACE ("\t\treceive_timestamp: %" G_GUINT64_FORMAT ".%09u",
msg->message_specific.delay_resp.receive_timestamp.seconds_field,
msg->message_specific.delay_resp.receive_timestamp.nanoseconds_field);
GST_TRACE ("\t\trequesting_port_identity: 0x%016" G_GINT64_MODIFIER
"x %u",
msg->message_specific.delay_resp.
requesting_port_identity.clock_identity,
msg->message_specific.delay_resp.
requesting_port_identity.port_number);
break;
default:
break;
}
GST_TRACE (" ");
}
/* IEEE 1588-2008 5.3.3 */
static gboolean
parse_ptp_timestamp (PtpTimestamp * timestamp, GstByteReader * reader)
{
g_return_val_if_fail (gst_byte_reader_get_remaining (reader) >= 10, FALSE);
timestamp->seconds_field =
(((guint64) gst_byte_reader_get_uint32_be_unchecked (reader)) << 16) |
gst_byte_reader_get_uint16_be_unchecked (reader);
timestamp->nanoseconds_field =
gst_byte_reader_get_uint32_be_unchecked (reader);
if (timestamp->nanoseconds_field >= 1000000000)
return FALSE;
return TRUE;
}
/* IEEE 1588-2008 13.3 */
static gboolean
parse_ptp_message_header (PtpMessage * msg, GstByteReader * reader)
{
guint8 b;
g_return_val_if_fail (gst_byte_reader_get_remaining (reader) >= 34, FALSE);
b = gst_byte_reader_get_uint8_unchecked (reader);
msg->transport_specific = b >> 4;
msg->message_type = b & 0x0f;
b = gst_byte_reader_get_uint8_unchecked (reader);
msg->version_ptp = b & 0x0f;
if (msg->version_ptp != 2) {
GST_WARNING ("Unsupported PTP message version (%u != 2)", msg->version_ptp);
return FALSE;
}
msg->message_length = gst_byte_reader_get_uint16_be_unchecked (reader);
if (gst_byte_reader_get_remaining (reader) + 4 < msg->message_length) {
GST_WARNING ("Not enough data (%u < %u)",
gst_byte_reader_get_remaining (reader) + 4, msg->message_length);
return FALSE;
}
msg->domain_number = gst_byte_reader_get_uint8_unchecked (reader);
gst_byte_reader_skip_unchecked (reader, 1);
msg->flag_field = gst_byte_reader_get_uint16_be_unchecked (reader);
msg->correction_field = gst_byte_reader_get_uint64_be_unchecked (reader);
gst_byte_reader_skip_unchecked (reader, 4);
msg->source_port_identity.clock_identity =
gst_byte_reader_get_uint64_be_unchecked (reader);
msg->source_port_identity.port_number =
gst_byte_reader_get_uint16_be_unchecked (reader);
msg->sequence_id = gst_byte_reader_get_uint16_be_unchecked (reader);
msg->control_field = gst_byte_reader_get_uint8_unchecked (reader);
msg->log_message_interval = gst_byte_reader_get_uint8_unchecked (reader);
return TRUE;
}
/* IEEE 1588-2008 13.5 */
static gboolean
parse_ptp_message_announce (PtpMessage * msg, GstByteReader * reader)
{
g_return_val_if_fail (msg->message_type == PTP_MESSAGE_TYPE_ANNOUNCE, FALSE);
if (gst_byte_reader_get_remaining (reader) < 20)
return FALSE;
if (!parse_ptp_timestamp (&msg->message_specific.announce.origin_timestamp,
reader))
return FALSE;
msg->message_specific.announce.current_utc_offset =
gst_byte_reader_get_uint16_be_unchecked (reader);
gst_byte_reader_skip_unchecked (reader, 1);
msg->message_specific.announce.grandmaster_priority_1 =
gst_byte_reader_get_uint8_unchecked (reader);
msg->message_specific.announce.grandmaster_clock_quality.clock_class =
gst_byte_reader_get_uint8_unchecked (reader);
msg->message_specific.announce.grandmaster_clock_quality.clock_accuracy =
gst_byte_reader_get_uint8_unchecked (reader);
msg->message_specific.announce.
grandmaster_clock_quality.offset_scaled_log_variance =
gst_byte_reader_get_uint16_be_unchecked (reader);
msg->message_specific.announce.grandmaster_priority_2 =
gst_byte_reader_get_uint8_unchecked (reader);
msg->message_specific.announce.grandmaster_identity =
gst_byte_reader_get_uint64_be_unchecked (reader);
msg->message_specific.announce.steps_removed =
gst_byte_reader_get_uint16_be_unchecked (reader);
msg->message_specific.announce.time_source =
gst_byte_reader_get_uint8_unchecked (reader);
return TRUE;
}
/* IEEE 1588-2008 13.6 */
static gboolean
parse_ptp_message_sync (PtpMessage * msg, GstByteReader * reader)
{
g_return_val_if_fail (msg->message_type == PTP_MESSAGE_TYPE_SYNC, FALSE);
if (gst_byte_reader_get_remaining (reader) < 10)
return FALSE;
if (!parse_ptp_timestamp (&msg->message_specific.sync.origin_timestamp,
reader))
return FALSE;
return TRUE;
}
/* IEEE 1588-2008 13.6 */
static gboolean
parse_ptp_message_delay_req (PtpMessage * msg, GstByteReader * reader)
{
g_return_val_if_fail (msg->message_type == PTP_MESSAGE_TYPE_DELAY_REQ, FALSE);
if (gst_byte_reader_get_remaining (reader) < 10)
return FALSE;
if (!parse_ptp_timestamp (&msg->message_specific.delay_req.origin_timestamp,
reader))
return FALSE;
return TRUE;
}
/* IEEE 1588-2008 13.7 */
static gboolean
parse_ptp_message_follow_up (PtpMessage * msg, GstByteReader * reader)
{
g_return_val_if_fail (msg->message_type == PTP_MESSAGE_TYPE_FOLLOW_UP, FALSE);
if (gst_byte_reader_get_remaining (reader) < 10)
return FALSE;
if (!parse_ptp_timestamp (&msg->message_specific.
follow_up.precise_origin_timestamp, reader))
return FALSE;
return TRUE;
}
/* IEEE 1588-2008 13.8 */
static gboolean
parse_ptp_message_delay_resp (PtpMessage * msg, GstByteReader * reader)
{
g_return_val_if_fail (msg->message_type == PTP_MESSAGE_TYPE_DELAY_RESP,
FALSE);
if (gst_byte_reader_get_remaining (reader) < 20)
return FALSE;
if (!parse_ptp_timestamp (&msg->message_specific.delay_resp.receive_timestamp,
reader))
return FALSE;
msg->message_specific.delay_resp.requesting_port_identity.clock_identity =
gst_byte_reader_get_uint64_be_unchecked (reader);
msg->message_specific.delay_resp.requesting_port_identity.port_number =
gst_byte_reader_get_uint16_be_unchecked (reader);
return TRUE;
}
static gboolean
parse_ptp_message (PtpMessage * msg, const guint8 * data, gsize size)
{
GstByteReader reader;
gboolean ret = FALSE;
gst_byte_reader_init (&reader, data, size);
if (!parse_ptp_message_header (msg, &reader)) {
GST_WARNING ("Failed to parse PTP message header");
return FALSE;
}
switch (msg->message_type) {
case PTP_MESSAGE_TYPE_SYNC:
ret = parse_ptp_message_sync (msg, &reader);
break;
case PTP_MESSAGE_TYPE_FOLLOW_UP:
ret = parse_ptp_message_follow_up (msg, &reader);
break;
case PTP_MESSAGE_TYPE_DELAY_REQ:
ret = parse_ptp_message_delay_req (msg, &reader);
break;
case PTP_MESSAGE_TYPE_DELAY_RESP:
ret = parse_ptp_message_delay_resp (msg, &reader);
break;
case PTP_MESSAGE_TYPE_ANNOUNCE:
ret = parse_ptp_message_announce (msg, &reader);
break;
default:
/* ignore for now */
break;
}
return ret;
}
static gint
compare_announce_message (const PtpAnnounceMessage * a,
const PtpAnnounceMessage * b)
{
/* IEEE 1588 Figure 27 */
if (a->grandmaster_identity == b->grandmaster_identity) {
if (a->steps_removed + 1 < b->steps_removed)
return -1;
else if (a->steps_removed > b->steps_removed + 1)
return 1;
/* Error cases are filtered out earlier */
if (a->steps_removed < b->steps_removed)
return -1;
else if (a->steps_removed > b->steps_removed)
return 1;
/* Error cases are filtered out earlier */
if (a->master_clock_identity.clock_identity <
b->master_clock_identity.clock_identity)
return -1;
else if (a->master_clock_identity.clock_identity >
b->master_clock_identity.clock_identity)
return 1;
/* Error cases are filtered out earlier */
if (a->master_clock_identity.port_number <
b->master_clock_identity.port_number)
return -1;
else if (a->master_clock_identity.port_number >
b->master_clock_identity.port_number)
return 1;
else
g_assert_not_reached ();
return 0;
}
if (a->grandmaster_priority_1 < b->grandmaster_priority_1)
return -1;
else if (a->grandmaster_priority_1 > b->grandmaster_priority_1)
return 1;
if (a->grandmaster_clock_quality.clock_class <
b->grandmaster_clock_quality.clock_class)
return -1;
else if (a->grandmaster_clock_quality.clock_class >
b->grandmaster_clock_quality.clock_class)
return 1;
if (a->grandmaster_clock_quality.clock_accuracy <
b->grandmaster_clock_quality.clock_accuracy)
return -1;
else if (a->grandmaster_clock_quality.clock_accuracy >
b->grandmaster_clock_quality.clock_accuracy)
return 1;
if (a->grandmaster_clock_quality.offset_scaled_log_variance <
b->grandmaster_clock_quality.offset_scaled_log_variance)
return -1;
else if (a->grandmaster_clock_quality.offset_scaled_log_variance >
b->grandmaster_clock_quality.offset_scaled_log_variance)
return 1;
if (a->grandmaster_priority_2 < b->grandmaster_priority_2)
return -1;
else if (a->grandmaster_priority_2 > b->grandmaster_priority_2)
return 1;
if (a->grandmaster_identity < b->grandmaster_identity)
return -1;
else if (a->grandmaster_identity > b->grandmaster_identity)
return 1;
else
g_assert_not_reached ();
return 0;
}
static void
select_best_master_clock (PtpDomainData * domain, GstClockTime now)
{
GList *qualified_messages = NULL;
GList *l, *m;
PtpAnnounceMessage *best = NULL;
/* IEEE 1588 9.3.2.5 */
for (l = domain->announce_senders; l; l = l->next) {
PtpAnnounceSender *sender = l->data;
GstClockTime window = 4 * sender->announce_interval;
gint count = 0;
for (m = sender->announce_messages.head; m; m = m->next) {
PtpAnnounceMessage *msg = m->data;
if (now - msg->receive_time <= window)
count++;
}
/* Only include the newest message of announce senders that had at least 2
* announce messages in the last 4 announce intervals. Which also means
* that we wait at least 4 announce intervals before we select a master
* clock. Until then we just report based on the newest SYNC we received
*/
if (count >= 2) {
qualified_messages =
g_list_prepend (qualified_messages,
g_queue_peek_tail (&sender->announce_messages));
}
}
if (!qualified_messages) {
GST_DEBUG
("No qualified announce messages for domain %u, can't select a master clock",
domain->domain);
domain->have_master_clock = FALSE;
return;
}
for (l = qualified_messages; l; l = l->next) {
PtpAnnounceMessage *msg = l->data;
if (!best || compare_announce_message (msg, best) < 0)
best = msg;
}
if (domain->have_master_clock
&& compare_clock_identity (&domain->master_clock_identity,
&best->master_clock_identity) == 0) {
GST_DEBUG ("Master clock in domain %u did not change", domain->domain);
} else {
GST_DEBUG ("Selected master clock for domain %u: 0x%016" G_GINT64_MODIFIER
"x %u with grandmaster clock 0x%016" G_GINT64_MODIFIER "x",
domain->domain, best->master_clock_identity.clock_identity,
best->master_clock_identity.port_number, best->grandmaster_identity);
domain->have_master_clock = TRUE;
domain->grandmaster_identity = best->grandmaster_identity;
/* Opportunistic master clock selection likely gave us the same master
* clock before, no need to reset all statistics */
if (compare_clock_identity (&domain->master_clock_identity,
&best->master_clock_identity) != 0) {
memcpy (&domain->master_clock_identity, &best->master_clock_identity,
sizeof (PtpClockIdentity));
domain->mean_path_delay = 0;
domain->last_delay_req = 0;
ptp: Add median based pre-filtering of delays If the delay measurement is too far away from the median of the window of last delay measurements, we discard it. This increases accuracy on wifi a lot. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-06-03 09:04:48 +00:00
domain->last_path_delays_missing = 9;
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
domain->min_delay_req_interval = 0;
domain->sync_interval = 0;
domain->last_ptp_sync_time = 0;
domain->skipped_updates = 0;
g_queue_foreach (&domain->pending_syncs, (GFunc) ptp_pending_sync_free,
NULL);
g_queue_clear (&domain->pending_syncs);
}
if (g_atomic_int_get (&domain_stats_n_hooks)) {
GstStructure *stats =
gst_structure_new (GST_PTP_STATISTICS_BEST_MASTER_CLOCK_SELECTED,
"domain", G_TYPE_UINT, domain->domain,
"master-clock-id", G_TYPE_UINT64,
domain->master_clock_identity.clock_identity,
"master-clock-port", G_TYPE_UINT,
domain->master_clock_identity.port_number,
"grandmaster-clock-id", G_TYPE_UINT64, domain->grandmaster_identity,
NULL);
emit_ptp_statistics (domain->domain, stats);
gst_structure_free (stats);
}
}
}
static void
handle_announce_message (PtpMessage * msg, GstClockTime receive_time)
{
GList *l;
PtpDomainData *domain = NULL;
PtpAnnounceSender *sender = NULL;
PtpAnnounceMessage *announce;
/* IEEE1588 9.3.2.2 e)
* Don't consider messages with the alternate master flag set
*/
if ((msg->flag_field & 0x0100))
return;
/* IEEE 1588 9.3.2.5 d)
* Don't consider announce messages with steps_removed>=255
*/
if (msg->message_specific.announce.steps_removed >= 255)
return;
for (l = domain_data; l; l = l->next) {
PtpDomainData *tmp = l->data;
if (tmp->domain == msg->domain_number) {
domain = tmp;
break;
}
}
if (!domain) {
gchar *clock_name;
domain = g_new0 (PtpDomainData, 1);
domain->domain = msg->domain_number;
clock_name = g_strdup_printf ("ptp-clock-%u", domain->domain);
domain->domain_clock =
g_object_new (GST_TYPE_SYSTEM_CLOCK, "name", clock_name, NULL);
g_free (clock_name);
g_queue_init (&domain->pending_syncs);
ptp: Add median based pre-filtering of delays If the delay measurement is too far away from the median of the window of last delay measurements, we discard it. This increases accuracy on wifi a lot. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-06-03 09:04:48 +00:00
domain->last_path_delays_missing = 9;
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
domain_data = g_list_prepend (domain_data, domain);
g_mutex_lock (&domain_clocks_lock);
domain_clocks = g_list_prepend (domain_clocks, domain);
g_mutex_unlock (&domain_clocks_lock);
if (g_atomic_int_get (&domain_stats_n_hooks)) {
GstStructure *stats =
gst_structure_new (GST_PTP_STATISTICS_NEW_DOMAIN_FOUND, "domain",
G_TYPE_UINT, domain->domain, "clock", GST_TYPE_CLOCK,
domain->domain_clock, NULL);
emit_ptp_statistics (domain->domain, stats);
gst_structure_free (stats);
}
}
for (l = domain->announce_senders; l; l = l->next) {
PtpAnnounceSender *tmp = l->data;
if (compare_clock_identity (&tmp->master_clock_identity,
&msg->source_port_identity) == 0) {
sender = tmp;
break;
}
}
if (!sender) {
sender = g_new0 (PtpAnnounceSender, 1);
memcpy (&sender->master_clock_identity, &msg->source_port_identity,
sizeof (PtpClockIdentity));
g_queue_init (&sender->announce_messages);
domain->announce_senders =
g_list_prepend (domain->announce_senders, sender);
}
for (l = sender->announce_messages.head; l; l = l->next) {
PtpAnnounceMessage *tmp = l->data;
/* IEEE 1588 9.3.2.5 c)
* Don't consider identical messages, i.e. duplicates
*/
if (tmp->sequence_id == msg->sequence_id)
return;
}
sender->announce_interval = log2_to_clock_time (msg->log_message_interval);
announce = g_new0 (PtpAnnounceMessage, 1);
announce->receive_time = receive_time;
announce->sequence_id = msg->sequence_id;
memcpy (&announce->master_clock_identity, &msg->source_port_identity,
sizeof (PtpClockIdentity));
announce->grandmaster_identity =
msg->message_specific.announce.grandmaster_identity;
announce->grandmaster_priority_1 =
msg->message_specific.announce.grandmaster_priority_1;
announce->grandmaster_clock_quality.clock_class =
msg->message_specific.announce.grandmaster_clock_quality.clock_class;
announce->grandmaster_clock_quality.clock_accuracy =
msg->message_specific.announce.grandmaster_clock_quality.clock_accuracy;
announce->grandmaster_clock_quality.offset_scaled_log_variance =
msg->message_specific.announce.
grandmaster_clock_quality.offset_scaled_log_variance;
announce->grandmaster_priority_2 =
msg->message_specific.announce.grandmaster_priority_2;
announce->steps_removed = msg->message_specific.announce.steps_removed;
announce->time_source = msg->message_specific.announce.time_source;
g_queue_push_tail (&sender->announce_messages, announce);
select_best_master_clock (domain, receive_time);
}
static gboolean
send_delay_req_timeout (PtpPendingSync * sync)
{
StdIOHeader header = { 0, };
guint8 delay_req[44];
GstByteWriter writer;
GIOStatus status;
gsize written;
GError *err = NULL;
header.type = TYPE_EVENT;
header.size = 44;
gst_byte_writer_init_with_data (&writer, delay_req, 44, FALSE);
gst_byte_writer_put_uint8_unchecked (&writer, PTP_MESSAGE_TYPE_DELAY_REQ);
gst_byte_writer_put_uint8_unchecked (&writer, 2);
gst_byte_writer_put_uint16_be_unchecked (&writer, 44);
gst_byte_writer_put_uint8_unchecked (&writer, sync->domain);
gst_byte_writer_put_uint8_unchecked (&writer, 0);
gst_byte_writer_put_uint16_be_unchecked (&writer, 0);
gst_byte_writer_put_uint64_be_unchecked (&writer, 0);
gst_byte_writer_put_uint32_be_unchecked (&writer, 0);
gst_byte_writer_put_uint64_be_unchecked (&writer,
ptp_clock_id.clock_identity);
gst_byte_writer_put_uint16_be_unchecked (&writer, ptp_clock_id.port_number);
gst_byte_writer_put_uint16_be_unchecked (&writer, sync->delay_req_seqnum);
gst_byte_writer_put_uint8_unchecked (&writer, 0x01);
gst_byte_writer_put_uint8_unchecked (&writer, 0x7f);
gst_byte_writer_put_uint64_be_unchecked (&writer, 0);
gst_byte_writer_put_uint16_be_unchecked (&writer, 0);
status =
g_io_channel_write_chars (stdout_channel, (gchar *) & header,
sizeof (header), &written, &err);
if (status == G_IO_STATUS_ERROR) {
g_warning ("Failed to write to stdout: %s", err->message);
return G_SOURCE_REMOVE;
} else if (status == G_IO_STATUS_EOF) {
g_message ("EOF on stdout");
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
} else if (status != G_IO_STATUS_NORMAL) {
g_warning ("Unexpected stdout write status: %d", status);
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
} else if (written != sizeof (header)) {
g_warning ("Unexpected write size: %" G_GSIZE_FORMAT, written);
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
}
ptpclock: Use a system clock for the time observations instead of gst_util_get_timestamp() They don't necessarily use the same underlying clocks (e.g. on Windows), or might be configured to a different clock type (monotonic vs. real time clock). We need the values a clean system clock returns, as those are the values used by the internal clocks.
2015-06-06 12:19:21 +00:00
sync->delay_req_send_time_local =
gst_clock_get_time (observation_system_clock);
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
status =
g_io_channel_write_chars (stdout_channel,
(const gchar *) delay_req, 44, &written, &err);
if (status == G_IO_STATUS_ERROR) {
g_warning ("Failed to write to stdout: %s", err->message);
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
} else if (status == G_IO_STATUS_EOF) {
g_message ("EOF on stdout");
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
} else if (status != G_IO_STATUS_NORMAL) {
g_warning ("Unexpected stdout write status: %d", status);
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
} else if (written != 44) {
g_warning ("Unexpected write size: %" G_GSIZE_FORMAT, written);
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
}
return G_SOURCE_REMOVE;
}
static gboolean
send_delay_req (PtpDomainData * domain, PtpPendingSync * sync)
{
ptpclock: Use a system clock for the time observations instead of gst_util_get_timestamp() They don't necessarily use the same underlying clocks (e.g. on Windows), or might be configured to a different clock type (monotonic vs. real time clock). We need the values a clean system clock returns, as those are the values used by the internal clocks.
2015-06-06 12:19:21 +00:00
GstClockTime now = gst_clock_get_time (observation_system_clock);
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
guint timeout;
GSource *timeout_source;
if (domain->last_delay_req != 0
&& domain->last_delay_req + domain->min_delay_req_interval > now)
return FALSE;
domain->last_delay_req = now;
sync->delay_req_seqnum = domain->last_delay_req_seqnum++;
/* IEEE 1588 9.5.11.2 */
if (domain->last_delay_req == 0 || domain->min_delay_req_interval == 0)
timeout = 0;
else
timeout =
g_rand_int_range (delay_req_rand, 0,
(domain->min_delay_req_interval * 2) / GST_MSECOND);
sync->timeout_source = timeout_source = g_timeout_source_new (timeout);
g_source_set_priority (timeout_source, G_PRIORITY_DEFAULT);
g_source_set_callback (timeout_source, (GSourceFunc) send_delay_req_timeout,
sync, NULL);
g_source_attach (timeout_source, main_context);
return TRUE;
}
/* Filtering of outliers for RTT and time calculations inspired
* by the code from gstnetclientclock.c
*/
static void
update_ptp_time (PtpDomainData * domain, PtpPendingSync * sync)
{
GstClockTime internal_time, external_time, rate_num, rate_den;
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
GstClockTime corrected_ptp_time, corrected_local_time;
gdouble r_squared = 0.0;
gboolean synced;
GstClockTimeDiff discont = 0;
GstClockTime estimated_ptp_time = GST_CLOCK_TIME_NONE;
#ifdef USE_MEASUREMENT_FILTERING
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
GstClockTime orig_internal_time, orig_external_time, orig_rate_num,
orig_rate_den;
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
GstClockTime new_estimated_ptp_time;
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
GstClockTime max_discont, estimated_ptp_time_min, estimated_ptp_time_max;
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
gboolean now_synced;
ptpclock: fix compilation Don't put code between declarations. Fix use of uninitialized variables
2015-06-03 13:22:31 +00:00
#endif
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
ptpclock: fix compilation Don't put code between declarations. Fix use of uninitialized variables
2015-06-03 13:22:31 +00:00
#ifdef USE_ONLY_SYNC_WITH_DELAY
ptpclock: Use the current path delay for calculation the local/remote clock times The mean might currently be changing, and the current path delay is the closest we can get to the actual delay around the current SYNC message.
2015-06-07 06:59:23 +00:00
GstClockTime mean_path_delay;
ptpclock: fix compilation Don't put code between declarations. Fix use of uninitialized variables
2015-06-03 13:22:31 +00:00
if (sync->delay_req_send_time_local == GST_CLOCK_TIME_NONE)
return;
ptpclock: Use the current path delay for calculation the local/remote clock times The mean might currently be changing, and the current path delay is the closest we can get to the actual delay around the current SYNC message.
2015-06-07 06:59:23 +00:00
/* IEEE 1588 11.3 */
mean_path_delay =
(sync->delay_req_recv_time_remote - sync->sync_send_time_remote +
sync->sync_recv_time_local - sync->delay_req_send_time_local -
(sync->correction_field_sync + sync->correction_field_delay +
32768) / 65536) / 2;
ptpclock: fix compilation Don't put code between declarations. Fix use of uninitialized variables
2015-06-03 13:22:31 +00:00
#endif
/* IEEE 1588 11.2 */
corrected_ptp_time =
sync->sync_send_time_remote +
(sync->correction_field_sync + 32768) / 65536;
ptpclock: Use the current path delay for calculation the local/remote clock times The mean might currently be changing, and the current path delay is the closest we can get to the actual delay around the current SYNC message.
2015-06-07 06:59:23 +00:00
#ifdef USE_ONLY_SYNC_WITH_DELAY
corrected_local_time = sync->sync_recv_time_local - mean_path_delay;
#else
ptpclock: fix compilation Don't put code between declarations. Fix use of uninitialized variables
2015-06-03 13:22:31 +00:00
corrected_local_time = sync->sync_recv_time_local - domain->mean_path_delay;
ptpclock: Use the current path delay for calculation the local/remote clock times The mean might currently be changing, and the current path delay is the closest we can get to the actual delay around the current SYNC message.
2015-06-07 06:59:23 +00:00
#endif
ptpclock: fix compilation Don't put code between declarations. Fix use of uninitialized variables
2015-06-03 13:22:31 +00:00
#ifdef USE_MEASUREMENT_FILTERING
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
/* We check this here and when updating the mean path delay, because
* we can get here without a delay response too */
if (sync->follow_up_recv_time_local != GST_CLOCK_TIME_NONE
&& sync->follow_up_recv_time_local >
sync->sync_recv_time_local + 2 * domain->mean_path_delay) {
GST_WARNING ("Sync-follow-up delay for domain %u too big: %" GST_TIME_FORMAT
" > 2 * %" GST_TIME_FORMAT, domain->domain,
GST_TIME_ARGS (sync->follow_up_recv_time_local),
GST_TIME_ARGS (domain->mean_path_delay));
ptpclock: fix compilation Don't put code between declarations. Fix use of uninitialized variables
2015-06-03 13:22:31 +00:00
synced = FALSE;
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
goto out;
}
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
#endif
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
/* Set an initial local-remote relation */
if (domain->last_ptp_time == 0)
gst_clock_set_calibration (domain->domain_clock, corrected_local_time,
corrected_ptp_time, 1, 1);
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
#ifdef USE_MEASUREMENT_FILTERING
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
/* Check if the corrected PTP time is +/- 3/4 RTT around what we would
* estimate with our present knowledge about the clock
*/
/* Store what the clock produced as 'now' before this update */
gst_clock_get_calibration (GST_CLOCK_CAST (domain->domain_clock),
&orig_internal_time, &orig_external_time, &orig_rate_num, &orig_rate_den);
internal_time = orig_internal_time;
external_time = orig_external_time;
rate_num = orig_rate_num;
rate_den = orig_rate_den;
/* 3/4 RTT window around the estimation */
max_discont = domain->mean_path_delay * 3 / 2;
/* Check if the estimated sync time is inside our window */
estimated_ptp_time_min = corrected_local_time - max_discont;
estimated_ptp_time_min =
gst_clock_adjust_with_calibration (GST_CLOCK_CAST (domain->domain_clock),
estimated_ptp_time_min, internal_time, external_time, rate_num, rate_den);
estimated_ptp_time_max = corrected_local_time + max_discont;
estimated_ptp_time_max =
gst_clock_adjust_with_calibration (GST_CLOCK_CAST (domain->domain_clock),
estimated_ptp_time_max, internal_time, external_time, rate_num, rate_den);
synced = (estimated_ptp_time_min < corrected_ptp_time
&& corrected_ptp_time < estimated_ptp_time_max);
GST_DEBUG ("Adding observation for domain %u: %" GST_TIME_FORMAT " - %"
GST_TIME_FORMAT, domain->domain,
GST_TIME_ARGS (corrected_ptp_time), GST_TIME_ARGS (corrected_local_time));
GST_DEBUG ("Synced %d: %" GST_TIME_FORMAT " < %" GST_TIME_FORMAT " < %"
GST_TIME_FORMAT, synced, GST_TIME_ARGS (estimated_ptp_time_min),
GST_TIME_ARGS (corrected_ptp_time),
GST_TIME_ARGS (estimated_ptp_time_max));
if (gst_clock_add_observation_unapplied (domain->domain_clock,
corrected_local_time, corrected_ptp_time, &r_squared,
&internal_time, &external_time, &rate_num, &rate_den)) {
GST_DEBUG ("Regression gave r_squared: %f", r_squared);
/* Old estimated PTP time based on receive time and path delay */
estimated_ptp_time = corrected_local_time;
estimated_ptp_time =
gst_clock_adjust_with_calibration (GST_CLOCK_CAST
(domain->domain_clock), estimated_ptp_time, orig_internal_time,
orig_external_time, orig_rate_num, orig_rate_den);
/* New estimated PTP time based on receive time and path delay */
new_estimated_ptp_time = corrected_local_time;
new_estimated_ptp_time =
gst_clock_adjust_with_calibration (GST_CLOCK_CAST
(domain->domain_clock), new_estimated_ptp_time, internal_time,
external_time, rate_num, rate_den);
discont = GST_CLOCK_DIFF (estimated_ptp_time, new_estimated_ptp_time);
if (synced && ABS (discont) > max_discont) {
GstClockTimeDiff offset;
GST_DEBUG ("Too large a discont %s%" GST_TIME_FORMAT
", clamping to 1/4 average RTT = %" GST_TIME_FORMAT,
(discont < 0 ? "-" : ""), GST_TIME_ARGS (ABS (discont)),
GST_TIME_ARGS (max_discont));
if (discont > 0) { /* Too large a forward step - add a -ve offset */
offset = max_discont - discont;
if (-offset > external_time)
external_time = 0;
else
external_time += offset;
} else { /* Too large a backward step - add a +ve offset */
offset = -(max_discont + discont);
external_time += offset;
}
discont += offset;
} else {
GST_DEBUG ("Discont %s%" GST_TIME_FORMAT " (max: %" GST_TIME_FORMAT ")",
(discont < 0 ? "-" : ""), GST_TIME_ARGS (ABS (discont)),
GST_TIME_ARGS (max_discont));
}
/* Check if the estimated sync time is now (still) inside our window */
estimated_ptp_time_min = corrected_local_time - max_discont;
estimated_ptp_time_min =
gst_clock_adjust_with_calibration (GST_CLOCK_CAST
(domain->domain_clock), estimated_ptp_time_min, internal_time,
external_time, rate_num, rate_den);
estimated_ptp_time_max = corrected_local_time + max_discont;
estimated_ptp_time_max =
gst_clock_adjust_with_calibration (GST_CLOCK_CAST
(domain->domain_clock), estimated_ptp_time_max, internal_time,
external_time, rate_num, rate_den);
now_synced = (estimated_ptp_time_min < corrected_ptp_time
&& corrected_ptp_time < estimated_ptp_time_max);
GST_DEBUG ("Now synced %d: %" GST_TIME_FORMAT " < %" GST_TIME_FORMAT " < %"
GST_TIME_FORMAT, now_synced, GST_TIME_ARGS (estimated_ptp_time_min),
GST_TIME_ARGS (corrected_ptp_time),
GST_TIME_ARGS (estimated_ptp_time_max));
if (synced || now_synced || domain->skipped_updates > MAX_SKIPPED_UPDATES) {
gst_clock_set_calibration (GST_CLOCK_CAST (domain->domain_clock),
internal_time, external_time, rate_num, rate_den);
domain->skipped_updates = 0;
domain->last_ptp_time = corrected_ptp_time;
domain->last_local_time = corrected_local_time;
} else {
domain->skipped_updates++;
}
} else {
domain->last_ptp_time = corrected_ptp_time;
domain->last_local_time = corrected_local_time;
}
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
#else
GST_DEBUG ("Adding observation for domain %u: %" GST_TIME_FORMAT " - %"
GST_TIME_FORMAT, domain->domain,
GST_TIME_ARGS (corrected_ptp_time), GST_TIME_ARGS (corrected_local_time));
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
gst_clock_get_calibration (GST_CLOCK_CAST (domain->domain_clock),
&internal_time, &external_time, &rate_num, &rate_den);
estimated_ptp_time = corrected_local_time;
estimated_ptp_time =
gst_clock_adjust_with_calibration (GST_CLOCK_CAST
(domain->domain_clock), estimated_ptp_time, internal_time,
external_time, rate_num, rate_den);
gst_clock_add_observation (domain->domain_clock,
corrected_local_time, corrected_ptp_time, &r_squared);
gst_clock_get_calibration (GST_CLOCK_CAST (domain->domain_clock),
&internal_time, &external_time, &rate_num, &rate_den);
synced = TRUE;
domain->last_ptp_time = corrected_ptp_time;
domain->last_local_time = corrected_local_time;
#endif
#ifdef USE_MEASUREMENT_FILTERING
out:
#endif
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
if (g_atomic_int_get (&domain_stats_n_hooks)) {
GstStructure *stats = gst_structure_new (GST_PTP_STATISTICS_TIME_UPDATED,
"domain", G_TYPE_UINT, domain->domain,
"mean-path-delay-avg", GST_TYPE_CLOCK_TIME, domain->mean_path_delay,
"local-time", GST_TYPE_CLOCK_TIME, corrected_local_time,
"ptp-time", GST_TYPE_CLOCK_TIME, corrected_ptp_time,
"estimated-ptp-time", GST_TYPE_CLOCK_TIME, estimated_ptp_time,
"discontinuity", G_TYPE_INT64, discont,
"synced", G_TYPE_BOOLEAN, synced,
"r-squared", G_TYPE_DOUBLE, r_squared,
"internal-time", GST_TYPE_CLOCK_TIME, internal_time,
"external-time", GST_TYPE_CLOCK_TIME, external_time,
"rate-num", G_TYPE_UINT64, rate_num,
"rate-den", G_TYPE_UINT64, rate_den,
"rate", G_TYPE_DOUBLE, (gdouble) (rate_num) / rate_den,
NULL);
emit_ptp_statistics (domain->domain, stats);
gst_structure_free (stats);
}
}
ptp: Add median based pre-filtering of delays If the delay measurement is too far away from the median of the window of last delay measurements, we discard it. This increases accuracy on wifi a lot. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-06-03 09:04:48 +00:00
#ifdef USE_MEDIAN_PRE_FILTERING
static gint
compare_clock_time (const GstClockTime * a, const GstClockTime * b)
{
if (*a < *b)
return -1;
else if (*a > *b)
return 1;
return 0;
}
#endif
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
static gboolean
update_mean_path_delay (PtpDomainData * domain, PtpPendingSync * sync)
{
ptp: Add median based pre-filtering of delays If the delay measurement is too far away from the median of the window of last delay measurements, we discard it. This increases accuracy on wifi a lot. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-06-03 09:04:48 +00:00
#ifdef USE_MEDIAN_PRE_FILTERING
ptpclock: Use #define everywhere instead of G_N_ELEMENTS()
2015-06-06 12:34:39 +00:00
GstClockTime last_path_delays[MEDIAN_PRE_FILTERING_WINDOW];
ptp: Add median based pre-filtering of delays If the delay measurement is too far away from the median of the window of last delay measurements, we discard it. This increases accuracy on wifi a lot. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-06-03 09:04:48 +00:00
GstClockTime median;
gint i;
#endif
ptpclock: fix compilation Don't put code between declarations. Fix use of uninitialized variables
2015-06-03 13:22:31 +00:00
GstClockTime mean_path_delay, delay_req_delay = 0;
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
gboolean ret;
/* IEEE 1588 11.3 */
mean_path_delay =
(sync->delay_req_recv_time_remote - sync->sync_send_time_remote +
sync->sync_recv_time_local - sync->delay_req_send_time_local -
(sync->correction_field_sync + sync->correction_field_delay +
32768) / 65536) / 2;
ptp: Add median based pre-filtering of delays If the delay measurement is too far away from the median of the window of last delay measurements, we discard it. This increases accuracy on wifi a lot. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-06-03 09:04:48 +00:00
#ifdef USE_MEDIAN_PRE_FILTERING
ptpclock: Use #define everywhere instead of G_N_ELEMENTS()
2015-06-06 12:34:39 +00:00
for (i = 1; i < MEDIAN_PRE_FILTERING_WINDOW; i++)
ptp: Add median based pre-filtering of delays If the delay measurement is too far away from the median of the window of last delay measurements, we discard it. This increases accuracy on wifi a lot. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-06-03 09:04:48 +00:00
domain->last_path_delays[i - 1] = domain->last_path_delays[i];
domain->last_path_delays[i - 1] = mean_path_delay;
if (domain->last_path_delays_missing) {
domain->last_path_delays_missing--;
} else {
memcpy (&last_path_delays, &domain->last_path_delays,
sizeof (last_path_delays));
g_qsort_with_data (&last_path_delays,
ptpclock: Use #define everywhere instead of G_N_ELEMENTS()
2015-06-06 12:34:39 +00:00
MEDIAN_PRE_FILTERING_WINDOW, sizeof (GstClockTime),
ptp: Add median based pre-filtering of delays If the delay measurement is too far away from the median of the window of last delay measurements, we discard it. This increases accuracy on wifi a lot. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-06-03 09:04:48 +00:00
(GCompareDataFunc) compare_clock_time, NULL);
ptpclock: Use #define everywhere instead of G_N_ELEMENTS()
2015-06-06 12:34:39 +00:00
median = last_path_delays[MEDIAN_PRE_FILTERING_WINDOW / 2];
ptp: Add median based pre-filtering of delays If the delay measurement is too far away from the median of the window of last delay measurements, we discard it. This increases accuracy on wifi a lot. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-06-03 09:04:48 +00:00
/* FIXME: We might want to use something else here, like only allowing
* things in the interquartile range, or also filtering away delays that
* are too small compared to the median. This here worked well enough
* in tests so far.
*/
if (mean_path_delay > 2 * median) {
GST_WARNING ("Path delay for domain %u too big compared to median: %"
GST_TIME_FORMAT " > 2 * %" GST_TIME_FORMAT, domain->domain,
GST_TIME_ARGS (mean_path_delay), GST_TIME_ARGS (median));
ret = FALSE;
goto out;
}
}
#endif
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
#ifdef USE_RUNNING_AVERAGE_DELAY
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
/* Track an average round trip time, for a bit of smoothing */
/* Always update before discarding a sample, so genuine changes in
* the network get picked up, eventually */
if (domain->mean_path_delay == 0)
domain->mean_path_delay = mean_path_delay;
else if (mean_path_delay < domain->mean_path_delay) /* Shorter RTTs carry more weight than longer */
domain->mean_path_delay =
(3 * domain->mean_path_delay + mean_path_delay) / 4;
else
domain->mean_path_delay =
(15 * domain->mean_path_delay + mean_path_delay) / 16;
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
#else
domain->mean_path_delay = mean_path_delay;
#endif
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
#ifdef USE_MEASUREMENT_FILTERING
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
if (sync->follow_up_recv_time_local != GST_CLOCK_TIME_NONE &&
domain->mean_path_delay != 0
&& sync->follow_up_recv_time_local >
sync->sync_recv_time_local + 2 * domain->mean_path_delay) {
GST_WARNING ("Sync-follow-up delay for domain %u too big: %" GST_TIME_FORMAT
" > 2 * %" GST_TIME_FORMAT, domain->domain,
ptp: Fix debug output to print the difference instead of absolute values
2015-06-03 14:08:43 +00:00
GST_TIME_ARGS (sync->follow_up_recv_time_local -
sync->sync_recv_time_local),
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
GST_TIME_ARGS (domain->mean_path_delay));
ret = FALSE;
goto out;
}
if (mean_path_delay > 2 * domain->mean_path_delay) {
GST_WARNING ("Mean path delay for domain %u too big: %" GST_TIME_FORMAT
" > 2 * %" GST_TIME_FORMAT, domain->domain,
GST_TIME_ARGS (mean_path_delay),
GST_TIME_ARGS (domain->mean_path_delay));
ret = FALSE;
goto out;
}
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
#endif
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
delay_req_delay =
sync->delay_resp_recv_time_local - sync->delay_req_send_time_local;
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
#ifdef USE_MEASUREMENT_FILTERING
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
/* delay_req_delay is a RTT, so 2 times the path delay */
if (delay_req_delay > 4 * domain->mean_path_delay) {
GST_WARNING ("Delay-request-response delay for domain %u too big: %"
GST_TIME_FORMAT " > 4 * %" GST_TIME_FORMAT, domain->domain,
GST_TIME_ARGS (delay_req_delay),
GST_TIME_ARGS (domain->mean_path_delay));
ret = FALSE;
goto out;
}
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
#endif
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
ret = TRUE;
GST_DEBUG ("Got mean path delay for domain %u: %" GST_TIME_FORMAT " (new: %"
GST_TIME_FORMAT ")", domain->domain,
GST_TIME_ARGS (domain->mean_path_delay), GST_TIME_ARGS (mean_path_delay));
GST_DEBUG ("Delay request delay for domain %u: %" GST_TIME_FORMAT,
domain->domain, GST_TIME_ARGS (delay_req_delay));
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
#ifdef USE_MEASUREMENT_FILTERING
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
out:
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
#endif
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
if (g_atomic_int_get (&domain_stats_n_hooks)) {
GstStructure *stats =
gst_structure_new (GST_PTP_STATISTICS_PATH_DELAY_MEASURED,
"domain", G_TYPE_UINT, domain->domain,
"mean-path-delay-avg", GST_TYPE_CLOCK_TIME, domain->mean_path_delay,
"mean-path-delay", GST_TYPE_CLOCK_TIME, mean_path_delay,
"delay-request-delay", GST_TYPE_CLOCK_TIME, delay_req_delay, NULL);
emit_ptp_statistics (domain->domain, stats);
gst_structure_free (stats);
}
return ret;
}
static void
handle_sync_message (PtpMessage * msg, GstClockTime receive_time)
{
GList *l;
PtpDomainData *domain = NULL;
PtpPendingSync *sync = NULL;
/* Don't consider messages with the alternate master flag set */
if ((msg->flag_field & 0x0100))
return;
for (l = domain_data; l; l = l->next) {
PtpDomainData *tmp = l->data;
if (msg->domain_number == tmp->domain) {
domain = tmp;
break;
}
}
if (!domain) {
gchar *clock_name;
ptp: Add median based pre-filtering of delays If the delay measurement is too far away from the median of the window of last delay measurements, we discard it. This increases accuracy on wifi a lot. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-06-03 09:04:48 +00:00
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
domain = g_new0 (PtpDomainData, 1);
domain->domain = msg->domain_number;
clock_name = g_strdup_printf ("ptp-clock-%u", domain->domain);
domain->domain_clock =
g_object_new (GST_TYPE_SYSTEM_CLOCK, "name", clock_name, NULL);
g_free (clock_name);
g_queue_init (&domain->pending_syncs);
ptp: Add median based pre-filtering of delays If the delay measurement is too far away from the median of the window of last delay measurements, we discard it. This increases accuracy on wifi a lot. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-06-03 09:04:48 +00:00
domain->last_path_delays_missing = 9;
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
domain_data = g_list_prepend (domain_data, domain);
g_mutex_lock (&domain_clocks_lock);
domain_clocks = g_list_prepend (domain_clocks, domain);
g_mutex_unlock (&domain_clocks_lock);
}
/* If we have a master clock, ignore this message if it's not coming from there */
if (domain->have_master_clock
&& compare_clock_identity (&domain->master_clock_identity,
&msg->source_port_identity) != 0)
return;
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
#ifdef USE_OPPORTUNISTIC_CLOCK_SELECTION
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
/* Opportunistic selection of master clock */
if (!domain->have_master_clock)
memcpy (&domain->master_clock_identity, &msg->source_port_identity,
sizeof (PtpClockIdentity));
ptp: Add #defines to enable/disable improvements for unreliable networks We should do some more measurements with all these and check how much sense they make for PTP. Also enabling them means not following IEEE1588-2008 by the letter anymore. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-15 13:58:51 +00:00
#else
if (!domain->have_master_clock)
return;
#endif
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
domain->sync_interval = log2_to_clock_time (msg->log_message_interval);
/* Check if duplicated */
for (l = domain->pending_syncs.head; l; l = l->next) {
PtpPendingSync *tmp = l->data;
if (tmp->sync_seqnum == msg->sequence_id)
return;
}
if (msg->message_specific.sync.origin_timestamp.seconds_field >
GST_CLOCK_TIME_NONE / GST_SECOND) {
GST_FIXME ("Unsupported sync message seconds field value: %"
G_GUINT64_FORMAT " > %" G_GUINT64_FORMAT,
msg->message_specific.sync.origin_timestamp.seconds_field,
GST_CLOCK_TIME_NONE / GST_SECOND);
return;
}
sync = g_new0 (PtpPendingSync, 1);
sync->domain = domain->domain;
sync->sync_seqnum = msg->sequence_id;
sync->sync_recv_time_local = receive_time;
sync->sync_send_time_remote = GST_CLOCK_TIME_NONE;
sync->follow_up_recv_time_local = GST_CLOCK_TIME_NONE;
sync->delay_req_send_time_local = GST_CLOCK_TIME_NONE;
sync->delay_req_recv_time_remote = GST_CLOCK_TIME_NONE;
sync->delay_resp_recv_time_local = GST_CLOCK_TIME_NONE;
/* 0.5 correction factor for division later */
sync->correction_field_sync = msg->correction_field;
if ((msg->flag_field & 0x0200)) {
/* Wait for FOLLOW_UP */
} else {
sync->sync_send_time_remote =
PTP_TIMESTAMP_TO_GST_CLOCK_TIME (msg->message_specific.
sync.origin_timestamp);
if (domain->last_ptp_sync_time != 0
&& domain->last_ptp_sync_time >= sync->sync_send_time_remote) {
GST_WARNING ("Backwards PTP times in domain %u: %" GST_TIME_FORMAT " >= %"
GST_TIME_FORMAT, domain->domain,
GST_TIME_ARGS (domain->last_ptp_sync_time),
GST_TIME_ARGS (sync->sync_send_time_remote));
ptp_pending_sync_free (sync);
sync = NULL;
return;
}
domain->last_ptp_sync_time = sync->sync_send_time_remote;
if (send_delay_req (domain, sync)) {
/* Sent delay request */
} else {
update_ptp_time (domain, sync);
ptp_pending_sync_free (sync);
sync = NULL;
}
}
if (sync)
g_queue_push_tail (&domain->pending_syncs, sync);
}
static void
handle_follow_up_message (PtpMessage * msg, GstClockTime receive_time)
{
GList *l;
PtpDomainData *domain = NULL;
PtpPendingSync *sync = NULL;
/* Don't consider messages with the alternate master flag set */
if ((msg->flag_field & 0x0100))
return;
for (l = domain_data; l; l = l->next) {
PtpDomainData *tmp = l->data;
if (msg->domain_number == tmp->domain) {
domain = tmp;
break;
}
}
if (!domain)
return;
/* If we have a master clock, ignore this message if it's not coming from there */
if (domain->have_master_clock
&& compare_clock_identity (&domain->master_clock_identity,
&msg->source_port_identity) != 0)
return;
/* Check if we know about this one */
for (l = domain->pending_syncs.head; l; l = l->next) {
PtpPendingSync *tmp = l->data;
if (tmp->sync_seqnum == msg->sequence_id) {
sync = tmp;
break;
}
}
if (!sync)
return;
/* Got a FOLLOW_UP for this already */
if (sync->sync_send_time_remote != GST_CLOCK_TIME_NONE)
return;
if (sync->sync_recv_time_local >= receive_time) {
GST_ERROR ("Got bogus follow up in domain %u: %" GST_TIME_FORMAT " > %"
GST_TIME_FORMAT, domain->domain,
GST_TIME_ARGS (sync->sync_recv_time_local),
GST_TIME_ARGS (receive_time));
g_queue_remove (&domain->pending_syncs, sync);
ptp_pending_sync_free (sync);
return;
}
sync->correction_field_sync += msg->correction_field;
sync->sync_send_time_remote =
PTP_TIMESTAMP_TO_GST_CLOCK_TIME (msg->message_specific.
follow_up.precise_origin_timestamp);
sync->follow_up_recv_time_local = receive_time;
if (domain->last_ptp_sync_time >= sync->sync_send_time_remote) {
GST_WARNING ("Backwards PTP times in domain %u: %" GST_TIME_FORMAT " >= %"
GST_TIME_FORMAT, domain->domain,
GST_TIME_ARGS (domain->last_ptp_sync_time),
GST_TIME_ARGS (sync->sync_send_time_remote));
g_queue_remove (&domain->pending_syncs, sync);
ptp_pending_sync_free (sync);
sync = NULL;
return;
}
domain->last_ptp_sync_time = sync->sync_send_time_remote;
if (send_delay_req (domain, sync)) {
/* Sent delay request */
} else {
update_ptp_time (domain, sync);
g_queue_remove (&domain->pending_syncs, sync);
ptp_pending_sync_free (sync);
sync = NULL;
}
}
static void
handle_delay_resp_message (PtpMessage * msg, GstClockTime receive_time)
{
GList *l;
PtpDomainData *domain = NULL;
PtpPendingSync *sync = NULL;
/* Don't consider messages with the alternate master flag set */
if ((msg->flag_field & 0x0100))
return;
for (l = domain_data; l; l = l->next) {
PtpDomainData *tmp = l->data;
if (msg->domain_number == tmp->domain) {
domain = tmp;
break;
}
}
if (!domain)
return;
/* If we have a master clock, ignore this message if it's not coming from there */
if (domain->have_master_clock
&& compare_clock_identity (&domain->master_clock_identity,
&msg->source_port_identity) != 0)
return;
/* Not for us */
if (msg->message_specific.delay_resp.
requesting_port_identity.clock_identity != ptp_clock_id.clock_identity
|| msg->message_specific.delay_resp.
requesting_port_identity.port_number != ptp_clock_id.port_number)
return;
domain->min_delay_req_interval =
log2_to_clock_time (msg->log_message_interval);
/* Check if we know about this one */
for (l = domain->pending_syncs.head; l; l = l->next) {
PtpPendingSync *tmp = l->data;
if (tmp->delay_req_seqnum == msg->sequence_id) {
sync = tmp;
break;
}
}
if (!sync)
return;
/* Got a DELAY_RESP for this already */
if (sync->delay_req_recv_time_remote != GST_CLOCK_TIME_NONE)
return;
if (sync->delay_req_send_time_local > receive_time) {
GST_ERROR ("Got bogus delay response in domain %u: %" GST_TIME_FORMAT " > %"
GST_TIME_FORMAT, domain->domain,
GST_TIME_ARGS (sync->delay_req_send_time_local),
GST_TIME_ARGS (receive_time));
g_queue_remove (&domain->pending_syncs, sync);
ptp_pending_sync_free (sync);
return;
}
sync->correction_field_delay = msg->correction_field;
sync->delay_req_recv_time_remote =
PTP_TIMESTAMP_TO_GST_CLOCK_TIME (msg->message_specific.
delay_resp.receive_timestamp);
sync->delay_resp_recv_time_local = receive_time;
if (domain->mean_path_delay != 0
&& sync->sync_send_time_remote > sync->delay_req_recv_time_remote) {
GST_WARNING ("Sync send time after delay req receive time for domain %u: %"
GST_TIME_FORMAT " > %" GST_TIME_FORMAT, domain->domain,
GST_TIME_ARGS (sync->sync_send_time_remote),
GST_TIME_ARGS (sync->delay_req_recv_time_remote));
g_queue_remove (&domain->pending_syncs, sync);
ptp_pending_sync_free (sync);
return;
}
if (update_mean_path_delay (domain, sync))
update_ptp_time (domain, sync);
g_queue_remove (&domain->pending_syncs, sync);
ptp_pending_sync_free (sync);
}
static void
handle_ptp_message (PtpMessage * msg, GstClockTime receive_time)
{
/* Ignore our own messages */
if (msg->source_port_identity.clock_identity == ptp_clock_id.clock_identity &&
msg->source_port_identity.port_number == ptp_clock_id.port_number)
return;
switch (msg->message_type) {
case PTP_MESSAGE_TYPE_ANNOUNCE:
handle_announce_message (msg, receive_time);
break;
case PTP_MESSAGE_TYPE_SYNC:
handle_sync_message (msg, receive_time);
break;
case PTP_MESSAGE_TYPE_FOLLOW_UP:
handle_follow_up_message (msg, receive_time);
break;
case PTP_MESSAGE_TYPE_DELAY_RESP:
handle_delay_resp_message (msg, receive_time);
break;
default:
break;
}
}
static gboolean
have_stdin_data_cb (GIOChannel * channel, GIOCondition condition,
gpointer user_data)
{
GIOStatus status;
StdIOHeader header;
gchar buffer[8192];
GError *err = NULL;
gsize read;
if ((condition & G_IO_STATUS_EOF)) {
GST_ERROR ("Got EOF on stdin");
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
}
status =
g_io_channel_read_chars (channel, (gchar *) & header, sizeof (header),
&read, &err);
if (status == G_IO_STATUS_ERROR) {
GST_ERROR ("Failed to read from stdin: %s", err->message);
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
} else if (status == G_IO_STATUS_EOF) {
GST_ERROR ("Got EOF on stdin");
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
} else if (status != G_IO_STATUS_NORMAL) {
GST_ERROR ("Unexpected stdin read status: %d", status);
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
} else if (read != sizeof (header)) {
GST_ERROR ("Unexpected read size: %" G_GSIZE_FORMAT, read);
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
} else if (header.size > 8192) {
GST_ERROR ("Unexpected size: %u", header.size);
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
}
status = g_io_channel_read_chars (channel, buffer, header.size, &read, &err);
if (status == G_IO_STATUS_ERROR) {
GST_ERROR ("Failed to read from stdin: %s", err->message);
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
} else if (status == G_IO_STATUS_EOF) {
GST_ERROR ("EOF on stdin");
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
} else if (status != G_IO_STATUS_NORMAL) {
GST_ERROR ("Unexpected stdin read status: %d", status);
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
} else if (read != header.size) {
GST_ERROR ("Unexpected read size: %" G_GSIZE_FORMAT, read);
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
}
switch (header.type) {
case TYPE_EVENT:
case TYPE_GENERAL:{
ptpclock: Use a system clock for the time observations instead of gst_util_get_timestamp() They don't necessarily use the same underlying clocks (e.g. on Windows), or might be configured to a different clock type (monotonic vs. real time clock). We need the values a clean system clock returns, as those are the values used by the internal clocks.
2015-06-06 12:19:21 +00:00
GstClockTime receive_time = gst_clock_get_time (observation_system_clock);
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
PtpMessage msg;
if (parse_ptp_message (&msg, (const guint8 *) buffer, header.size)) {
dump_ptp_message (&msg);
handle_ptp_message (&msg, receive_time);
}
break;
}
default:
case TYPE_CLOCK_ID:{
if (header.size != 8) {
GST_ERROR ("Unexpected clock id size (%u != 8)", header.size);
g_main_loop_quit (main_loop);
return G_SOURCE_REMOVE;
}
g_mutex_lock (&ptp_lock);
ptp_clock_id.clock_identity = GST_READ_UINT64_BE (buffer);
ptp_clock_id.port_number = getpid ();
GST_DEBUG ("Got clock id 0x%016" G_GINT64_MODIFIER "x %u",
ptp_clock_id.clock_identity, ptp_clock_id.port_number);
g_cond_signal (&ptp_cond);
g_mutex_unlock (&ptp_lock);
break;
}
}
return G_SOURCE_CONTINUE;
}
/* Cleanup all announce messages and announce message senders
* that are timed out by now, and clean up all pending syncs
* that are missing their FOLLOW_UP or DELAY_RESP */
static gboolean
cleanup_cb (gpointer data)
{
ptpclock: Use a system clock for the time observations instead of gst_util_get_timestamp() They don't necessarily use the same underlying clocks (e.g. on Windows), or might be configured to a different clock type (monotonic vs. real time clock). We need the values a clean system clock returns, as those are the values used by the internal clocks.
2015-06-06 12:19:21 +00:00
GstClockTime now = gst_clock_get_time (observation_system_clock);
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
GList *l, *m, *n;
for (l = domain_data; l; l = l->next) {
PtpDomainData *domain = l->data;
for (n = domain->announce_senders; n;) {
PtpAnnounceSender *sender = n->data;
gboolean timed_out = TRUE;
/* Keep only 5 messages per sender around */
while (g_queue_get_length (&sender->announce_messages) > 5) {
PtpAnnounceMessage *msg = g_queue_pop_head (&sender->announce_messages);
g_free (msg);
}
for (m = sender->announce_messages.head; m; m = m->next) {
PtpAnnounceMessage *msg = m->data;
if (msg->receive_time +
sender->announce_interval * PTP_ANNOUNCE_RECEIPT_TIMEOUT > now) {
timed_out = FALSE;
break;
}
}
if (timed_out) {
GST_DEBUG ("Announce sender 0x%016" G_GINT64_MODIFIER "x %u timed out",
sender->master_clock_identity.clock_identity,
sender->master_clock_identity.port_number);
g_queue_foreach (&sender->announce_messages, (GFunc) g_free, NULL);
g_queue_clear (&sender->announce_messages);
}
if (g_queue_get_length (&sender->announce_messages) == 0) {
GList *tmp = n->next;
if (compare_clock_identity (&sender->master_clock_identity,
&domain->master_clock_identity) == 0)
GST_WARNING ("currently selected master clock timed out");
g_free (sender);
domain->announce_senders =
g_list_delete_link (domain->announce_senders, n);
n = tmp;
} else {
n = n->next;
}
}
select_best_master_clock (domain, now);
/* Clean up any pending syncs */
for (n = domain->pending_syncs.head; n;) {
PtpPendingSync *sync = n->data;
gboolean timed_out = FALSE;
/* Time out pending syncs after 4 sync intervals or 10 seconds,
* and pending delay reqs after 4 delay req intervals or 10 seconds
*/
if (sync->delay_req_send_time_local != GST_CLOCK_TIME_NONE &&
((domain->min_delay_req_interval != 0
&& sync->delay_req_send_time_local +
4 * domain->min_delay_req_interval < now)
|| (sync->delay_req_send_time_local + 10 * GST_SECOND < now))) {
timed_out = TRUE;
} else if ((domain->sync_interval != 0
&& sync->sync_recv_time_local + 4 * domain->sync_interval < now)
|| (sync->sync_recv_time_local + 10 * GST_SECOND < now)) {
timed_out = TRUE;
}
if (timed_out) {
GList *tmp = n->next;
ptp_pending_sync_free (sync);
g_queue_delete_link (&domain->pending_syncs, n);
n = tmp;
} else {
n = n->next;
}
}
}
return G_SOURCE_CONTINUE;
}
static gpointer
ptp_helper_main (gpointer data)
{
GSource *cleanup_source;
GST_DEBUG ("Starting PTP helper loop");
/* Check all 5 seconds, if we have to cleanup ANNOUNCE or pending syncs message */
cleanup_source = g_timeout_source_new_seconds (5);
g_source_set_priority (cleanup_source, G_PRIORITY_DEFAULT);
g_source_set_callback (cleanup_source, (GSourceFunc) cleanup_cb, NULL, NULL);
g_source_attach (cleanup_source, main_context);
g_source_unref (cleanup_source);
g_main_loop_run (main_loop);
GST_DEBUG ("Stopped PTP helper loop");
g_mutex_lock (&ptp_lock);
ptp_clock_id.clock_identity = GST_PTP_CLOCK_ID_NONE;
ptp_clock_id.port_number = 0;
initted = FALSE;
g_cond_signal (&ptp_cond);
g_mutex_unlock (&ptp_lock);
return NULL;
}
/**
* gst_ptp_is_supported:
*
* Check if PTP clocks are generally supported on this system, and if previous
* initializations did not fail.
*
* Returns: %TRUE if PTP clocks are generally supported on this system, and
* previous initializations did not fail.
*
* Since: 1.6
*/
gboolean
gst_ptp_is_supported (void)
{
return supported;
}
/**
* gst_ptp_is_initialized:
*
* Check if the GStreamer PTP clock subsystem is initialized.
*
* Returns: %TRUE if the GStreamer PTP clock subsystem is intialized.
*
* Since: 1.6
*/
gboolean
gst_ptp_is_initialized (void)
{
return initted;
}
/**
* gst_ptp_init:
* @clock_id: PTP clock id of this process' clock or %GST_PTP_CLOCK_ID_NONE
* @interfaces: (transfer none) (array zero-terminated=1): network interfaces to run the clock on
*
* Initialize the GStreamer PTP subsystem and create a PTP ordinary clock in
* slave-only mode for all domains on the given @interfaces with the
* given @clock_id.
*
* If @clock_id is %GST_PTP_CLOCK_ID_NONE, a clock id is automatically
* generated from the MAC address of the first network interface.
*
*
* This function is automatically called by gst_ptp_clock_new() with default
* parameters if it wasn't called before.
*
* Returns: %TRUE if the GStreamer PTP clock subsystem could be initialized.
*
* Since: 1.6
*/
gboolean
gst_ptp_init (guint64 clock_id, gchar ** interfaces)
{
gboolean ret;
const gchar *env;
gchar **argv = NULL;
gint argc, argc_c;
gint fd_r, fd_w;
GError *err = NULL;
GSource *stdin_source;
GST_DEBUG_CATEGORY_INIT (ptp_debug, "ptp", 0, "PTP clock");
g_mutex_lock (&ptp_lock);
if (!supported) {
GST_ERROR ("PTP not supported");
ret = FALSE;
goto done;
}
if (initted) {
GST_DEBUG ("PTP already initialized");
ret = TRUE;
goto done;
}
if (ptp_helper_pid) {
GST_DEBUG ("PTP currently initializing");
goto wait;
}
if (!domain_stats_hooks_initted) {
g_hook_list_init (&domain_stats_hooks, sizeof (GHook));
domain_stats_hooks_initted = TRUE;
}
argc = 1;
if (clock_id != GST_PTP_CLOCK_ID_NONE)
argc += 2;
if (interfaces != NULL)
argc += 2 * g_strv_length (interfaces);
argv = g_new0 (gchar *, argc + 2);
argc_c = 0;
env = g_getenv ("GST_PTP_HELPER_1_0");
if (env == NULL)
env = g_getenv ("GST_PTP_HELPER");
if (env != NULL && *env != '\0') {
GST_LOG ("Trying GST_PTP_HELPER env var: %s", env);
argv[argc_c++] = g_strdup (env);
} else {
argv[argc_c++] = g_strdup (GST_PTP_HELPER_INSTALLED);
}
if (clock_id != GST_PTP_CLOCK_ID_NONE) {
argv[argc_c++] = g_strdup ("-c");
argv[argc_c++] = g_strdup_printf ("0x%016" G_GINT64_MODIFIER "x", clock_id);
}
if (interfaces != NULL) {
gchar **ptr = interfaces;
while (*ptr) {
argv[argc_c++] = g_strdup ("-i");
argv[argc_c++] = g_strdup (*ptr);
ptr++;
}
}
main_context = g_main_context_new ();
main_loop = g_main_loop_new (main_context, FALSE);
ptp_helper_thread =
g_thread_try_new ("ptp-helper-thread", ptp_helper_main, NULL, &err);
if (!ptp_helper_thread) {
GST_ERROR ("Failed to start PTP helper thread: %s", err->message);
g_clear_error (&err);
ret = FALSE;
goto done;
}
if (!g_spawn_async_with_pipes (NULL, argv, NULL, 0, NULL, NULL,
&ptp_helper_pid, &fd_w, &fd_r, NULL, &err)) {
GST_ERROR ("Failed to start ptp helper process: %s", err->message);
g_clear_error (&err);
ret = FALSE;
supported = FALSE;
goto done;
}
stdin_channel = g_io_channel_unix_new (fd_r);
g_io_channel_set_encoding (stdin_channel, NULL, NULL);
g_io_channel_set_buffered (stdin_channel, FALSE);
g_io_channel_set_close_on_unref (stdin_channel, TRUE);
stdin_source =
g_io_create_watch (stdin_channel, G_IO_IN | G_IO_PRI | G_IO_HUP);
g_source_set_priority (stdin_source, G_PRIORITY_DEFAULT);
g_source_set_callback (stdin_source, (GSourceFunc) have_stdin_data_cb, NULL,
NULL);
g_source_attach (stdin_source, main_context);
g_source_unref (stdin_source);
/* Create stdout channel */
stdout_channel = g_io_channel_unix_new (fd_w);
g_io_channel_set_encoding (stdout_channel, NULL, NULL);
g_io_channel_set_close_on_unref (stdout_channel, TRUE);
g_io_channel_set_buffered (stdout_channel, FALSE);
delay_req_rand = g_rand_new ();
ptpclock: Use a system clock for the time observations instead of gst_util_get_timestamp() They don't necessarily use the same underlying clocks (e.g. on Windows), or might be configured to a different clock type (monotonic vs. real time clock). We need the values a clean system clock returns, as those are the values used by the internal clocks.
2015-06-06 12:19:21 +00:00
observation_system_clock =
g_object_new (GST_TYPE_SYSTEM_CLOCK, "name", "ptp-observation-clock",
NULL);
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
initted = TRUE;
wait:
GST_DEBUG ("Waiting for PTP to be initialized");
while (ptp_clock_id.clock_identity == GST_PTP_CLOCK_ID_NONE && initted)
g_cond_wait (&ptp_cond, &ptp_lock);
ret = initted;
if (ret) {
GST_DEBUG ("Initialized and got clock id 0x%016" G_GINT64_MODIFIER "x %u",
ptp_clock_id.clock_identity, ptp_clock_id.port_number);
} else {
GST_ERROR ("Failed to initialize");
supported = FALSE;
}
done:
g_strfreev (argv);
if (!ret) {
if (ptp_helper_pid) {
ptp: Fix build on Windows, and in general the GI build when PTP support was not available It's not going to work on Windows still, the helper process needs to be ported.
2015-06-08 08:46:24 +00:00
#ifndef G_OS_WIN32
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
kill (ptp_helper_pid, SIGKILL);
waitpid (ptp_helper_pid, NULL, 0);
ptp: Fix build on Windows, and in general the GI build when PTP support was not available It's not going to work on Windows still, the helper process needs to be ported.
2015-06-08 08:46:24 +00:00
#else
TerminateProcess (ptp_helper_pid, 1);
WaitForSingleObject (ptp_helper_pid, INFINITE);
#endif
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
g_spawn_close_pid (ptp_helper_pid);
}
ptp_helper_pid = 0;
if (stdin_channel)
g_io_channel_unref (stdin_channel);
stdin_channel = NULL;
if (stdout_channel)
g_io_channel_unref (stdout_channel);
stdout_channel = NULL;
if (main_loop && ptp_helper_thread) {
g_main_loop_quit (main_loop);
g_thread_join (ptp_helper_thread);
}
ptp_helper_thread = NULL;
if (main_loop)
g_main_loop_unref (main_loop);
main_loop = NULL;
if (main_context)
g_main_context_unref (main_context);
main_context = NULL;
if (delay_req_rand)
g_rand_free (delay_req_rand);
delay_req_rand = NULL;
ptpclock: Use a system clock for the time observations instead of gst_util_get_timestamp() They don't necessarily use the same underlying clocks (e.g. on Windows), or might be configured to a different clock type (monotonic vs. real time clock). We need the values a clean system clock returns, as those are the values used by the internal clocks.
2015-06-06 12:19:21 +00:00
if (observation_system_clock)
gst_object_unref (observation_system_clock);
observation_system_clock = NULL;
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
}
g_mutex_unlock (&ptp_lock);
return ret;
}
/**
* gst_ptp_deinit:
*
* Deinitialize the GStreamer PTP subsystem and stop the PTP clock. If there
* are any remaining GstPtpClock instances, they won't be further synchronized
* to the PTP network clock.
*
* Since: 1.6
*/
void
gst_ptp_deinit (void)
{
GList *l, *m;
g_mutex_lock (&ptp_lock);
if (ptp_helper_pid) {
ptp: Fix build on Windows, and in general the GI build when PTP support was not available It's not going to work on Windows still, the helper process needs to be ported.
2015-06-08 08:46:24 +00:00
#ifndef G_OS_WIN32
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
kill (ptp_helper_pid, SIGKILL);
waitpid (ptp_helper_pid, NULL, 0);
ptp: Fix build on Windows, and in general the GI build when PTP support was not available It's not going to work on Windows still, the helper process needs to be ported.
2015-06-08 08:46:24 +00:00
#else
TerminateProcess (ptp_helper_pid, 1);
WaitForSingleObject (ptp_helper_pid, INFINITE);
#endif
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
g_spawn_close_pid (ptp_helper_pid);
}
ptp_helper_pid = 0;
if (stdin_channel)
g_io_channel_unref (stdin_channel);
stdin_channel = NULL;
if (stdout_channel)
g_io_channel_unref (stdout_channel);
stdout_channel = NULL;
if (main_loop && ptp_helper_thread) {
GThread *tmp = ptp_helper_thread;
ptp_helper_thread = NULL;
g_mutex_unlock (&ptp_lock);
g_main_loop_quit (main_loop);
g_thread_join (tmp);
g_mutex_lock (&ptp_lock);
}
if (main_loop)
g_main_loop_unref (main_loop);
main_loop = NULL;
if (main_context)
g_main_context_unref (main_context);
main_context = NULL;
if (delay_req_rand)
g_rand_free (delay_req_rand);
delay_req_rand = NULL;
ptpclock: Use a system clock for the time observations instead of gst_util_get_timestamp() They don't necessarily use the same underlying clocks (e.g. on Windows), or might be configured to a different clock type (monotonic vs. real time clock). We need the values a clean system clock returns, as those are the values used by the internal clocks.
2015-06-06 12:19:21 +00:00
if (observation_system_clock)
gst_object_unref (observation_system_clock);
observation_system_clock = NULL;
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
for (l = domain_data; l; l = l->next) {
PtpDomainData *domain = l->data;
for (m = domain->announce_senders; m; m = m->next) {
PtpAnnounceSender *sender = m->data;
g_queue_foreach (&sender->announce_messages, (GFunc) g_free, NULL);
g_queue_clear (&sender->announce_messages);
g_free (sender);
}
g_list_free (domain->announce_senders);
g_queue_foreach (&domain->pending_syncs, (GFunc) ptp_pending_sync_free,
NULL);
g_queue_clear (&domain->pending_syncs);
gst_object_unref (domain->domain_clock);
g_free (domain);
}
g_list_free (domain_data);
domain_data = NULL;
g_list_foreach (domain_clocks, (GFunc) g_free, NULL);
g_list_free (domain_clocks);
domain_clocks = NULL;
ptp_clock_id.clock_identity = GST_PTP_CLOCK_ID_NONE;
ptp_clock_id.port_number = 0;
initted = FALSE;
g_mutex_unlock (&ptp_lock);
}
#define DEFAULT_DOMAIN 0
enum
{
PROP_0,
PROP_DOMAIN,
PROP_INTERNAL_CLOCK
};
#define GST_PTP_CLOCK_GET_PRIVATE(obj) \
(G_TYPE_INSTANCE_GET_PRIVATE ((obj), GST_TYPE_PTP_CLOCK, GstPtpClockPrivate))
struct _GstPtpClockPrivate
{
guint domain;
GstClock *domain_clock;
gulong domain_stats_id;
};
#define gst_ptp_clock_parent_class parent_class
G_DEFINE_TYPE (GstPtpClock, gst_ptp_clock, GST_TYPE_SYSTEM_CLOCK);
static void gst_ptp_clock_set_property (GObject * object, guint prop_id,
const GValue * value, GParamSpec * pspec);
static void gst_ptp_clock_get_property (GObject * object, guint prop_id,
GValue * value, GParamSpec * pspec);
static void gst_ptp_clock_finalize (GObject * object);
static GstClockTime gst_ptp_clock_get_internal_time (GstClock * clock);
static void
gst_ptp_clock_class_init (GstPtpClockClass * klass)
{
GObjectClass *gobject_class;
GstClockClass *clock_class;
gobject_class = G_OBJECT_CLASS (klass);
clock_class = GST_CLOCK_CLASS (klass);
g_type_class_add_private (klass, sizeof (GstPtpClockPrivate));
gobject_class->finalize = gst_ptp_clock_finalize;
gobject_class->get_property = gst_ptp_clock_get_property;
gobject_class->set_property = gst_ptp_clock_set_property;
g_object_class_install_property (gobject_class, PROP_DOMAIN,
g_param_spec_uint ("domain", "Domain",
"The PTP domain", 0, G_MAXUINT8,
DEFAULT_DOMAIN,
G_PARAM_READWRITE | G_PARAM_CONSTRUCT_ONLY | G_PARAM_STATIC_STRINGS));
g_object_class_install_property (gobject_class, PROP_INTERNAL_CLOCK,
g_param_spec_object ("internal-clock", "Internal Clock",
"Internal clock", GST_TYPE_CLOCK,
G_PARAM_READABLE | G_PARAM_STATIC_STRINGS));
clock_class->get_internal_time = gst_ptp_clock_get_internal_time;
}
static void
gst_ptp_clock_init (GstPtpClock * self)
{
GstPtpClockPrivate *priv;
self->priv = priv = GST_PTP_CLOCK_GET_PRIVATE (self);
GST_OBJECT_FLAG_SET (self, GST_CLOCK_FLAG_CAN_SET_MASTER);
GST_OBJECT_FLAG_SET (self, GST_CLOCK_FLAG_NEEDS_STARTUP_SYNC);
priv->domain = DEFAULT_DOMAIN;
}
static gboolean
gst_ptp_clock_ensure_domain_clock (GstPtpClock * self)
{
gboolean got_clock = TRUE;
if (G_UNLIKELY (!self->priv->domain_clock)) {
g_mutex_lock (&domain_clocks_lock);
if (!self->priv->domain_clock) {
GList *l;
got_clock = FALSE;
for (l = domain_clocks; l; l = l->next) {
PtpDomainData *clock_data = l->data;
if (clock_data->domain == self->priv->domain
&& clock_data->last_ptp_time != 0) {
self->priv->domain_clock = clock_data->domain_clock;
got_clock = TRUE;
break;
}
}
}
g_mutex_unlock (&domain_clocks_lock);
if (got_clock) {
g_object_notify (G_OBJECT (self), "internal-clock");
gst_clock_set_synced (GST_CLOCK (self), TRUE);
}
}
return got_clock;
}
static gboolean
gst_ptp_clock_stats_callback (guint8 domain, const GstStructure * stats,
gpointer user_data)
{
GstPtpClock *self = user_data;
if (domain != self->priv->domain
|| !gst_structure_has_name (stats, GST_PTP_STATISTICS_TIME_UPDATED))
return TRUE;
/* Let's set our internal clock */
if (!gst_ptp_clock_ensure_domain_clock (self))
return TRUE;
self->priv->domain_stats_id = 0;
return FALSE;
}
static void
gst_ptp_clock_set_property (GObject * object, guint prop_id,
const GValue * value, GParamSpec * pspec)
{
GstPtpClock *self = GST_PTP_CLOCK (object);
switch (prop_id) {
case PROP_DOMAIN:
self->priv->domain = g_value_get_uint (value);
gst_ptp_clock_ensure_domain_clock (self);
if (!self->priv->domain_clock)
self->priv->domain_stats_id =
gst_ptp_statistics_callback_add (gst_ptp_clock_stats_callback, self,
NULL);
break;
default:
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
break;
}
}
static void
gst_ptp_clock_get_property (GObject * object, guint prop_id,
GValue * value, GParamSpec * pspec)
{
GstPtpClock *self = GST_PTP_CLOCK (object);
switch (prop_id) {
case PROP_DOMAIN:
g_value_set_uint (value, self->priv->domain);
break;
case PROP_INTERNAL_CLOCK:
gst_ptp_clock_ensure_domain_clock (self);
g_value_set_object (value, self->priv->domain_clock);
break;
default:
G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec);
break;
}
}
static void
gst_ptp_clock_finalize (GObject * object)
{
GstPtpClock *self = GST_PTP_CLOCK (object);
if (self->priv->domain_stats_id)
gst_ptp_statistics_callback_remove (self->priv->domain_stats_id);
G_OBJECT_CLASS (gst_ptp_clock_parent_class)->finalize (object);
}
static GstClockTime
gst_ptp_clock_get_internal_time (GstClock * clock)
{
GstPtpClock *self = GST_PTP_CLOCK (clock);
gst_ptp_clock_ensure_domain_clock (self);
if (!self->priv->domain_clock) {
GST_ERROR_OBJECT (self, "Domain %u has no clock yet and is not synced",
self->priv->domain);
return GST_CLOCK_TIME_NONE;
}
return gst_clock_get_time (self->priv->domain_clock);
}
/**
* gst_ptp_clock_new:
* @name: Name of the clock
* @domain: PTP domain
*
* Creates a new PTP clock instance that exports the PTP time of the master
* clock in @domain. This clock can be slaved to other clocks as needed.
*
* If gst_ptp_init() was not called before, this will call gst_ptp_init() with
* default parameters.
*
*
* This clock only returns valid timestamps after it received the first
* times from the PTP master clock on the network. Once this happens the
ptpclock: Fix documentation a bit
2015-06-06 10:35:58 +00:00
* GstPtpClock::internal-clock property will become non-NULL. You can
* check this with gst_clock_wait_for_sync(), the GstClock::synced signal and
* gst_clock_is_synced().
ptp: Initial implementation of a PTP clock GstPtpClock implements a PTP (IEEE1588:2008) ordinary clock in slave-only mode, that allows a GStreamer pipeline to synchronize to a PTP network clock in some specific domain. The PTP subsystem can be initialized with gst_ptp_init(), which then starts a helper process to do the actual communication via the PTP ports. This is required as PTP listens on ports < 1024 and thus requires special privileges. Once this helper process is started, the main process will synchronize to all PTP domains that are detected on the selected interfaces. gst_ptp_clock_new() then allows to create a GstClock that provides the PTP time from a master clock inside a specific PTP domain. This clock will only return valid timestamps once the timestamps in the PTP domain are known. To check this, the GstPtpClock::internal-clock property and the related notify::clock signal can be used. Once the internal clock is not NULL, the PTP domain's time is known. Alternatively you can wait for this with gst_ptp_clock_wait_ready(). To gather statistics about the PTP clock synchronization, gst_ptp_statistics_callback_add() can be used. This gives the application the possibility to collect all kinds of statistics from the clock synchronization. https://bugzilla.gnome.org/show_bug.cgi?id=749391
2015-05-14 10:18:25 +00:00
*
* Since: 1.6
*/
GstClock *
gst_ptp_clock_new (const gchar * name, guint domain)
{
g_return_val_if_fail (name != NULL, NULL);
g_return_val_if_fail (domain <= G_MAXUINT8, NULL);
if (!initted && !gst_ptp_init (GST_PTP_CLOCK_ID_NONE, NULL)) {
GST_ERROR ("Failed to initialize PTP");
return NULL;
}
return g_object_new (GST_TYPE_PTP_CLOCK, "name", name, "domain", domain,
NULL);
}
typedef struct
{
guint8 domain;
const GstStructure *stats;
} DomainStatsMarshalData;
static void
domain_stats_marshaller (GHook * hook, DomainStatsMarshalData * data)
{
GstPtpStatisticsCallback callback = (GstPtpStatisticsCallback) hook->func;
if (!callback (data->domain, data->stats, hook->data))
g_hook_destroy (&domain_stats_hooks, hook->hook_id);
}
static void
emit_ptp_statistics (guint8 domain, const GstStructure * stats)
{
DomainStatsMarshalData data = { domain, stats };
g_mutex_lock (&ptp_lock);
g_hook_list_marshal (&domain_stats_hooks, TRUE,
(GHookMarshaller) domain_stats_marshaller, &data);
g_mutex_unlock (&ptp_lock);
}
/**
* gst_ptp_statistics_callback_add:
* @callback: GstPtpStatisticsCallback to call
* @user_data: Data to pass to the callback
* @destroy_data: GDestroyNotify to destroy the data
*
* Installs a new statistics callback for gathering PTP statistics. See
* GstPtpStatisticsCallback for a list of statistics that are provided.
*
* Returns: Id for the callback that can be passed to
* gst_ptp_statistics_callback_remove()
*
* Since: 1.6
*/
gulong
gst_ptp_statistics_callback_add (GstPtpStatisticsCallback callback,
gpointer user_data, GDestroyNotify destroy_data)
{
GHook *hook;
g_mutex_lock (&ptp_lock);
if (!domain_stats_hooks_initted) {
g_hook_list_init (&domain_stats_hooks, sizeof (GHook));
domain_stats_hooks_initted = TRUE;
}
hook = g_hook_alloc (&domain_stats_hooks);
hook->func = callback;
hook->data = user_data;
hook->destroy = destroy_data;
g_hook_prepend (&domain_stats_hooks, hook);
g_atomic_int_add (&domain_stats_n_hooks, 1);
g_mutex_unlock (&ptp_lock);
return hook->hook_id;
}
/**
* gst_ptp_statistics_callback_remove:
* @id: Callback id to remove
*
* Removes a PTP statistics callback that was previously added with
* gst_ptp_statistics_callback_add().
*
* Since: 1.6
*/
void
gst_ptp_statistics_callback_remove (gulong id)
{
g_mutex_lock (&ptp_lock);
if (g_hook_destroy (&domain_stats_hooks, id))
g_atomic_int_add (&domain_stats_n_hooks, -1);
g_mutex_unlock (&ptp_lock);
}
Reference in a new issue Copy permalink