rpms/linuxptp
5
0
Fork 0
Code Issues Pull Requests Releases Activity

import linuxptp-3.1.1-6.el9_2.1

Browse Source
This commit is contained in:
CentOS Sources 2023-05-09 05:21:49 +00:00 committed by Stepan Oksanichenko
parent a6200bae87
commit 1cc1c4d830
4 changed files with 707 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
SOURCES/linuxptp-eintr.patch Normal file
View File

@ -0,0 +1,30 @@
commit c8de9f33e437fb15cff32c0028f651d7cb916a37
Author: Miroslav Lichvar <mlichvar@redhat.com>
Date: Tue Sep 13 13:44:10 2022 +0200
sk: Handle EINTR when waiting for transmit timestamp.
If ptp4l received a signal in poll() waiting for a transmit timestamp,
multiple (possibly confusing) error messages are logged before exit.
If poll() returns with EINTR, call it once again to get the timestamp
and avoid logging the errors. Don't call it in a loop to avoid getting
stuck in case the timestamp is lost and the signal is repeated before
the poll timeout can be reached.
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
diff --git a/sk.c b/sk.c
index 80075be..d27abff 100644
--- a/sk.c
+++ b/sk.c
@@ -354,6 +354,9 @@ int sk_receive(int fd, void *buf, int buflen,
if (flags == MSG_ERRQUEUE) {
struct pollfd pfd = { fd, sk_events, 0 };
res = poll(&pfd, 1, sk_tx_timeout);
+ /* Retry once on EINTR to avoid logging errors before exit */
+ if (res < 0 && errno == EINTR)
+ res = poll(&pfd, 1, sk_tx_timeout);
if (res < 1) {
pr_err(res ? "poll for tx timestamp failed: %m" :
"timed out while polling for tx timestamp");

208
SOURCES/linuxptp-freqcheck.patch Normal file
View File

@ -0,0 +1,208 @@
commit 513c7457865dba262a43e03fbe9178f4b08ba319
Author: Miroslav Lichvar <mlichvar@redhat.com>
Date: Mon Oct 24 15:25:30 2022 +0200
Drop support for old kernels returning zero frequency.
Kernels before 3.10 had a bug in reading of the system clock frequency,
which was worked around by commit da347d7a36f2 ("ptp4l: Set clock
frequency on start").
Drop this workaround and support for the old kernels to make
clockadj_get_freq() useful.
(Rebased to 3.1.1)
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
diff --git a/clock.c b/clock.c
index f3df220..9079428 100644
--- a/clock.c
+++ b/clock.c
@@ -1147,11 +1147,6 @@ struct clock *clock_create(enum clock_type type, struct config *config,
if (c->clkid != CLOCK_INVALID) {
fadj = (int) clockadj_get_freq(c->clkid);
- /* Due to a bug in older kernels, the reading may silently fail
- and return 0. Set the frequency back to make sure fadj is
- the actual frequency of the clock. */
- clockadj_set_freq(c->clkid, fadj);
-
/* Disable write phase mode if not implemented by driver */
if (c->write_phase_mode && !phc_has_writephase(c->clkid)) {
pr_err("clock does not support write phase mode");
@@ -1746,7 +1741,6 @@ int clock_switch_phc(struct clock *c, int phc_index)
return -1;
}
fadj = (int) clockadj_get_freq(clkid);
- clockadj_set_freq(clkid, fadj);
servo = servo_create(c->config, c->servo_type, -fadj, max_adj, 0);
if (!servo) {
pr_err("Switching PHC, failed to create clock servo");
diff --git a/phc2sys.c b/phc2sys.c
index 5f6fbaa..af948eb 100644
--- a/phc2sys.c
+++ b/phc2sys.c
@@ -147,9 +147,6 @@ static struct servo *servo_add(struct phc2sys_private *priv,
clockadj_init(clock->clkid);
ppb = clockadj_get_freq(clock->clkid);
- /* The reading may silently fail and return 0, reset the frequency to
- make sure ppb is the actual frequency of the clock. */
- clockadj_set_freq(clock->clkid, ppb);
if (clock->clkid == CLOCK_REALTIME) {
sysclk_set_leap(0);
max_ppb = sysclk_max_freq();
diff --git a/ts2phc_slave.c b/ts2phc_slave.c
index 749efe5..5541d91 100644
--- a/ts2phc_slave.c
+++ b/ts2phc_slave.c
@@ -183,10 +183,6 @@ static struct ts2phc_slave *ts2phc_slave_create(struct config *cfg, const char *
pr_debug("PHC slave %s has ptp index %d", device, junk);
fadj = (int) clockadj_get_freq(slave->clk);
- /* Due to a bug in older kernels, the reading may silently fail
- and return 0. Set the frequency back to make sure fadj is
- the actual frequency of the clock. */
- clockadj_set_freq(slave->clk, fadj);
max_adj = phc_max_adj(slave->clk);
commit 80d6875b0072c7ea636a5631e590fa72d169c351
Author: Miroslav Lichvar <mlichvar@redhat.com>
Date: Mon Oct 24 15:25:31 2022 +0200
Don't accept errors in clockadj_get_freq().
Exit if an error is returned from the clock_adjtime() call in
clockadj_get_freq(). No recoverable errors are expected.
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
diff --git a/clockadj.c b/clockadj.c
index 957dc57..4c920b9 100644
--- a/clockadj.c
+++ b/clockadj.c
@@ -19,6 +19,7 @@
#include <errno.h>
#include <math.h>
+#include <stdlib.h>
#include <string.h>
#include <unistd.h>
@@ -72,6 +73,7 @@ double clockadj_get_freq(clockid_t clkid)
memset(&tx, 0, sizeof(tx));
if (clock_adjtime(clkid, &tx) < 0) {
pr_err("failed to read out the clock frequency adjustment: %m");
+ exit(1);
} else {
f = tx.freq / 65.536;
if (clkid == CLOCK_REALTIME && realtime_nominal_tick && tx.tick)
commit 211cbfe810ef4082d1af8e9b50f65d4f6fd7246a
Author: Miroslav Lichvar <mlichvar@redhat.com>
Date: Mon Oct 24 15:25:32 2022 +0200
Extend clockcheck to check for changes in frequency.
Before setting the new frequency offset on a clock update, compare the
current frequency returned by the kernel with the value saved from the
previous update. Print a warning message if the difference is larger
than 1 ppb, allowing for rounding errors in conversion to and from
double. The kernel caches the value set by clock_adjtime() in shifted
ppm, it doesn't request it from the driver, which can have a lower
resulution.
This should detect misconfigurations where multiple processes are trying
to control the clock (e.g. another ptp4l/phc2sys instance or an NTP
client), even when they don't step the clock.
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
diff --git a/clock.c b/clock.c
index 9079428..eea7983 100644
--- a/clock.c
+++ b/clock.c
@@ -1760,6 +1760,9 @@ int clock_switch_phc(struct clock *c, int phc_index)
static void clock_synchronize_locked(struct clock *c, double adj)
{
+ if (c->sanity_check) {
+ clockcheck_freq(c->sanity_check, clockadj_get_freq(c->clkid));
+ }
clockadj_set_freq(c->clkid, -adj);
if (c->clkid == CLOCK_REALTIME) {
sysclk_set_sync();
diff --git a/clockcheck.c b/clockcheck.c
index f0141be..b5a69cc 100644
--- a/clockcheck.c
+++ b/clockcheck.c
@@ -123,6 +123,16 @@ void clockcheck_set_freq(struct clockcheck *cc, int freq)
cc->freq_known = 1;
}
+int clockcheck_freq(struct clockcheck *cc, int freq)
+{
+ /* Allow difference of 1 ppb due to conversion to/from double */
+ if (cc->freq_known && abs(cc->current_freq - freq) > 1) {
+ pr_warning("clockcheck: clock frequency changed unexpectedly!");
+ return 1;
+ }
+ return 0;
+}
+
void clockcheck_step(struct clockcheck *cc, int64_t step)
{
if (cc->last_ts)
diff --git a/clockcheck.h b/clockcheck.h
index 1ff86eb..4b09b98 100644
--- a/clockcheck.h
+++ b/clockcheck.h
@@ -54,6 +54,14 @@ int clockcheck_sample(struct clockcheck *cc, uint64_t ts);
*/
void clockcheck_set_freq(struct clockcheck *cc, int freq);
+/**
+ * Check whether the frequency correction did not change unexpectedly.
+ * @param cc Pointer to a clock check obtained via @ref clockcheck_create().
+ * @param freq Current reading of the frequency correction in ppb.
+ * @return Zero if the frequency did not change, non-zero otherwise.
+ */
+int clockcheck_freq(struct clockcheck *cc, int freq);
+
/**
* Inform clock check that the clock was stepped.
* @param cc Pointer to a clock check obtained via @ref clockcheck_create().
diff --git a/phc2sys.c b/phc2sys.c
index af948eb..4daa1bf 100644
--- a/phc2sys.c
+++ b/phc2sys.c
@@ -566,6 +566,9 @@ static void update_clock(struct phc2sys_private *priv, struct clock *clock,
/* Fall through. */
case SERVO_LOCKED:
case SERVO_LOCKED_STABLE:
+ if (clock->sanity_check)
+ clockcheck_freq(clock->sanity_check,
+ clockadj_get_freq(clock->clkid));
clockadj_set_freq(clock->clkid, -ppb);
if (clock->clkid == CLOCK_REALTIME)
sysclk_set_sync();
diff --git a/ptp4l.8 b/ptp4l.8
index 4917240..f760e2b 100644
--- a/ptp4l.8
+++ b/ptp4l.8
@@ -605,8 +605,10 @@ This option used to be called
The maximum allowed frequency offset between uncorrected clock and the system
monotonic clock in parts per billion (ppb). This is used as a sanity check of
the synchronized clock. When a larger offset is measured, a warning message
-will be printed and the servo will be reset. When set to 0, the sanity check is
-disabled. The default is 200000000 (20%).
+will be printed and the servo will be reset. If the frequency correction set by
+ptp4l changes unexpectedly between updates of the clock (e.g. due to another
+process trying to control the clock), a warning message will be printed. When
+set to 0, the sanity check is disabled. The default is 200000000 (20%).
.TP
.B initial_delay
The initial path delay of the clock in nanoseconds used for synchronization of

451
SOURCES/linuxptp-vlanbond.patch Normal file
View File

@ -0,0 +1,451 @@
commit afd709924edcea26fec24a6253d1b85fc94c88e6
Author: Hangbin Liu <liuhangbin@gmail.com>
Date: Wed May 25 14:46:14 2022 +0800
ptp4l: init iface->ts_label when interface created
It's redundant to ensure tslabel if we could get ts_info directly from
the interface. By init iface->ts_label when interface created, there is
no need to take care of ts_label on other places.
It will also make the later vlan over bonding support patch more clear
after removing function interface_ensure_tslabel().
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
diff --git a/clock.c b/clock.c
index 7ffbd6a..d4f273f 100644
--- a/clock.c
+++ b/clock.c
@@ -1006,9 +1006,8 @@ struct clock *clock_create(enum clock_type type, struct config *config,
required_modes = clock_required_modes(c);
STAILQ_FOREACH(iface, &config->interfaces, list) {
memset(ts_label, 0, sizeof(ts_label));
- rtnl_get_ts_device(interface_name(iface), ts_label);
- interface_set_label(iface, ts_label);
- interface_ensure_tslabel(iface);
+ if (!rtnl_get_ts_device(interface_name(iface), ts_label))
+ interface_set_label(iface, ts_label);
interface_get_tsinfo(iface);
if (interface_tsinfo_valid(iface) &&
!interface_tsmodes_supported(iface, required_modes)) {
diff --git a/interface.c b/interface.c
index 445a270..6c2630c 100644
--- a/interface.c
+++ b/interface.c
@@ -24,6 +24,7 @@ struct interface *interface_create(const char *name)
return NULL;
}
strncpy(iface->name, name, MAX_IFNAME_SIZE);
+ strncpy(iface->ts_label, name, MAX_IFNAME_SIZE);
iface->vclock = -1;
return iface;
@@ -34,13 +35,6 @@ void interface_destroy(struct interface *iface)
free(iface);
}
-void interface_ensure_tslabel(struct interface *iface)
-{
- if (!iface->ts_label[0]) {
- memcpy(iface->ts_label, iface->name, MAX_IFNAME_SIZE);
- }
-}
-
int interface_get_tsinfo(struct interface *iface)
{
return sk_get_ts_info(iface->ts_label, &iface->ts_info);
diff --git a/interface.h b/interface.h
index 752f4f1..5fc7836 100644
--- a/interface.h
+++ b/interface.h
@@ -33,12 +33,6 @@ struct interface *interface_create(const char *name);
*/
void interface_destroy(struct interface *iface);
-/**
- * Ensures that an interface has a proper time stamping label.
- * @param iface The interface of interest.
- */
-void interface_ensure_tslabel(struct interface *iface);
-
/**
* Populate the time stamping information of a given interface.
* @param iface The interface of interest.
diff --git a/nsm.c b/nsm.c
index 5aa925b..8d4a362 100644
--- a/nsm.c
+++ b/nsm.c
@@ -274,9 +274,8 @@ static int nsm_open(struct nsm *nsm, struct config *cfg)
STAILQ_FOREACH(iface, &cfg->interfaces, list) {
ifname = interface_name(iface);
memset(ts_label, 0, sizeof(ts_label));
- rtnl_get_ts_device(ifname, ts_label);
- interface_set_label(iface, ts_label);
- interface_ensure_tslabel(iface);
+ if (!rtnl_get_ts_device(ifname, ts_label))
+ interface_set_label(iface, ts_label);
count++;
}
if (count != 1) {
diff --git a/pmc_common.c b/pmc_common.c
index 756edf5..ee2ece6 100644
--- a/pmc_common.c
+++ b/pmc_common.c
@@ -405,7 +405,6 @@ struct pmc *pmc_create(struct config *cfg, enum transport_type transport_type,
pr_err("failed to create interface");
goto failed;
}
- interface_ensure_tslabel(pmc->iface);
if (transport_open(pmc->transport, pmc->iface,
&pmc->fdarray, TS_SOFTWARE)) {
commit 6aab466eb0563c1025c78ab496b22a8833b02d20
Author: Vincent Cheng <vincent.cheng.xh@renesas.com>
Date: Fri Jan 8 10:21:14 2021 -0500
port: Fix link down/up to continue using phc_index set from command line -p option.
In the scenario where a port link goes down and up, current code checks
the port's phc_index against the interface's phc_index and if they are
different will set the port phc_index to the interface phc_index.
If the phc_index was initially set by the command line -p option, then we
end up using the wrong phc_index.
Fix is to skip updating the port phc_index with the interface phc_index
when port link is back up if it was initialy set from the command line.
Signed-off-by: Vincent Cheng <vincent.cheng.xh@renesas.com>
diff --git a/port.c b/port.c
index 70b6e60..d22f30f 100644
--- a/port.c
+++ b/port.c
@@ -2601,6 +2601,11 @@ void port_link_status(void *ctx, int linkup, int ts_index)
"timestamping mode, set link status down by force.",
interface_label(p->iface));
p->link_status = LINK_DOWN | LINK_STATE_CHANGED;
+ } else if (p->phc_from_cmdline) {
+ pr_warning("port %d: taking /dev/ptp%d from the "
+ "command line, not the attached ptp%d",
+ portnum(p), p->phc_index,
+ interface_phc_index(p->iface));
} else if (p->phc_index != interface_phc_index(p->iface)) {
p->phc_index = interface_phc_index(p->iface);
@@ -3107,6 +3112,7 @@ struct port *port_open(const char *phc_device,
"not the attached ptp%d", number, phc_device,
interface_phc_index(interface));
p->phc_index = phc_index;
+ p->phc_from_cmdline = 1;
} else {
pr_err("port %d: PHC device mismatch", number);
pr_err("port %d: /dev/ptp%d requested, ptp%d attached",
diff --git a/port_private.h b/port_private.h
index fcabaa6..6e40e15 100644
--- a/port_private.h
+++ b/port_private.h
@@ -69,6 +69,7 @@ struct port {
struct fdarray fda;
int fault_fd;
int phc_index;
+ int phc_from_cmdline;
void (*dispatch)(struct port *p, enum fsm_event event, int mdiff);
enum fsm_event (*event)(struct port *p, int fd_index);
commit 45f20e824fdfb8748d73495ccc7578dc70866282
Author: Hangbin Liu <liuhangbin@gmail.com>
Date: Wed May 25 14:46:15 2022 +0800
port: refactor port_link_status
Split port_change_phc() logic form port_link_status() to improve the
readability of the code
(Rebased to 3.1.1)
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
diff --git a/port.c b/port.c
index d22f30f..031b212 100644
--- a/port.c
+++ b/port.c
@@ -2563,10 +2563,43 @@ static void bc_dispatch(struct port *p, enum fsm_event event, int mdiff)
}
}
+static void port_change_phc(struct port *p)
+{
+ int required_modes;
+
+ /* Only switch a non-vclock PHC with HW time stamping. */
+ if (!interface_tsinfo_valid(p->iface) ||
+ interface_get_vclock(p->iface) >= 0 ||
+ interface_phc_index(p->iface) < 0)
+ return;
+
+ required_modes = clock_required_modes(p->clock);
+ if (!interface_tsmodes_supported(p->iface, required_modes)) {
+ pr_err("interface '%s' does not support requested "
+ "timestamping mode, set link status down by force.",
+ interface_label(p->iface));
+ p->link_status = LINK_DOWN | LINK_STATE_CHANGED;
+ } else if (p->phc_from_cmdline) {
+ pr_warning("port %d: taking /dev/ptp%d from the "
+ "command line, not the attached ptp%d",
+ portnum(p), p->phc_index,
+ interface_phc_index(p->iface));
+ } else if (p->phc_index != interface_phc_index(p->iface)) {
+ p->phc_index = interface_phc_index(p->iface);
+
+ if (clock_switch_phc(p->clock, p->phc_index)) {
+ p->last_fault_type = FT_SWITCH_PHC;
+ port_dispatch(p, EV_FAULT_DETECTED, 0);
+ return;
+ }
+ clock_sync_interval(p->clock, p->log_sync_interval);
+ }
+}
+
void port_link_status(void *ctx, int linkup, int ts_index)
{
char ts_label[MAX_IFNAME_SIZE + 1] = {0};
- int link_state, required_modes;
+ int link_state;
const char *old_ts_label;
struct port *p = ctx;
@@ -2590,33 +2623,7 @@ void port_link_status(void *ctx, int linkup, int ts_index)
if (p->link_status & LINK_UP &&
(p->link_status & LINK_STATE_CHANGED || p->link_status & TS_LABEL_CHANGED)) {
interface_get_tsinfo(p->iface);
-
- /* Only switch a non-vclock PHC with HW time stamping. */
- if (interface_tsinfo_valid(p->iface) &&
- interface_get_vclock(p->iface) < 0 &&
- interface_phc_index(p->iface) >= 0) {
- required_modes = clock_required_modes(p->clock);
- if (!interface_tsmodes_supported(p->iface, required_modes)) {
- pr_err("interface '%s' does not support requested "
- "timestamping mode, set link status down by force.",
- interface_label(p->iface));
- p->link_status = LINK_DOWN | LINK_STATE_CHANGED;
- } else if (p->phc_from_cmdline) {
- pr_warning("port %d: taking /dev/ptp%d from the "
- "command line, not the attached ptp%d",
- portnum(p), p->phc_index,
- interface_phc_index(p->iface));
- } else if (p->phc_index != interface_phc_index(p->iface)) {
- p->phc_index = interface_phc_index(p->iface);
-
- if (clock_switch_phc(p->clock, p->phc_index)) {
- p->last_fault_type = FT_SWITCH_PHC;
- port_dispatch(p, EV_FAULT_DETECTED, 0);
- return;
- }
- clock_sync_interval(p->clock, p->log_sync_interval);
- }
- }
+ port_change_phc(p);
}
/*
commit 878c2eac77cc7abd1d86620a82ae9046545439ea
Author: Hangbin Liu <liuhangbin@gmail.com>
Date: Wed May 25 14:46:16 2022 +0800
ptp4l: add VLAN over bond support
Latest Linux kernel has supported getting active port's PHC from bond
directly. This would help topology like VLAN over bond to get PHC natively,
as VLAN also albe to get downlink's PHC.
To achieve this, the new hwtstamp flag is needed when passing hwtstamp_config
to kernel. Let's supply the flag first, and fall back without flag if user
run on old kernel.
We also need to update the PHC index in port_link_status() when there is
a RTNL event. The PHC info will be update in later port_change_phc().
Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
diff --git a/clock.c b/clock.c
index d4f273f..f3df220 100644
--- a/clock.c
+++ b/clock.c
@@ -1758,6 +1758,9 @@ int clock_switch_phc(struct clock *c, int phc_index)
c->clkid = clkid;
c->servo = servo;
c->servo_state = SERVO_UNLOCKED;
+
+ pr_info("Switched to /dev/ptp%d as PTP clock", phc_index);
+
return 0;
}
diff --git a/missing.h b/missing.h
index c5194f4..5f684dd 100644
--- a/missing.h
+++ b/missing.h
@@ -73,6 +73,12 @@ struct so_timestamping {
};
#endif
+#ifndef HWTSTAMP_FLAG_BONDED_PHC_INDEX
+enum {
+ HWTSTAMP_FLAG_BONDED_PHC_INDEX = (1<<0),
+};
+#endif
+
#ifdef PTP_EXTTS_REQUEST2
#define PTP_EXTTS_REQUEST_FAILED "PTP_EXTTS_REQUEST2 failed: %m"
#else
diff --git a/port.c b/port.c
index 031b212..ae9c4d3 100644
--- a/port.c
+++ b/port.c
@@ -2619,12 +2619,18 @@ void port_link_status(void *ctx, int linkup, int ts_index)
pr_notice("port %hu: ts label changed to %s", portnum(p), ts_label);
}
+ /* phc index may changed while ts_label keeps the same after failover.
+ * e.g. vlan over bond. Since the lower link changed, we still set
+ * the TS_LABEL_CHANGED flag.
+ */
+ interface_get_tsinfo(p->iface);
+ if (p->phc_index != interface_phc_index(p->iface))
+ p->link_status |= TS_LABEL_CHANGED;
+
/* Both link down/up and change ts_label may change phc index. */
if (p->link_status & LINK_UP &&
- (p->link_status & LINK_STATE_CHANGED || p->link_status & TS_LABEL_CHANGED)) {
- interface_get_tsinfo(p->iface);
+ (p->link_status & LINK_STATE_CHANGED || p->link_status & TS_LABEL_CHANGED))
port_change_phc(p);
- }
/*
* A port going down can affect the BMCA result.
diff --git a/sk.c b/sk.c
index b55d6b5..80075be 100644
--- a/sk.c
+++ b/sk.c
@@ -66,6 +66,11 @@ static int hwts_init(int fd, const char *device, int rx_filter,
init_ifreq(&ifreq, &cfg, device);
+ cfg.flags = HWTSTAMP_FLAG_BONDED_PHC_INDEX;
+ /* Fall back without flag if user run new build on old kernel */
+ if (ioctl(fd, SIOCGHWTSTAMP, &ifreq) == -EINVAL)
+ init_ifreq(&ifreq, &cfg, device);
+
switch (sk_hwts_filter_mode) {
case HWTS_FILTER_CHECK:
err = ioctl(fd, SIOCGHWTSTAMP, &ifreq);
commit 98edb4d72b3d1d4e988f810b9a9cb1078212e5dc
Author: Miroslav Lichvar <mlichvar@redhat.com>
Date: Tue Nov 22 17:07:16 2022 +0100
port: Avoid faults with vclocks and PHC from command line.
After commit afeabf3c90ed ("ptp4l: add VLAN over bond support") the
TS_LABEL_CHANGED flag was set on link status changes when the used
PHC index was different from the PHC index of the interface.
This caused the port to be constantly switching to the faulty state when
using vclocks, or a different PHC device was forced with the -p option,
where it is expected the used PHC doesn't match the interface's PHC.
Rework port_link_status() and port_change_phc() to avoid setting the
flag and switch the clock only in the cases where it is expected.
Fixes: afeabf3c90ed ("ptp4l: add VLAN over bond support")
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
diff --git a/port.c b/port.c
index ae9c4d3..96e097a 100644
--- a/port.c
+++ b/port.c
@@ -2567,10 +2567,13 @@ static void port_change_phc(struct port *p)
{
int required_modes;
- /* Only switch a non-vclock PHC with HW time stamping. */
- if (!interface_tsinfo_valid(p->iface) ||
+ /* Try to switch only if the interface is up, it has HW time stamping
+ using a non-vclock PHC, and the PHC actually changed. */
+ if (!(p->link_status & LINK_UP) ||
+ !interface_tsinfo_valid(p->iface) ||
interface_get_vclock(p->iface) >= 0 ||
- interface_phc_index(p->iface) < 0)
+ interface_phc_index(p->iface) < 0 ||
+ p->phc_index == interface_phc_index(p->iface))
return;
required_modes = clock_required_modes(p->clock);
@@ -2584,7 +2587,7 @@ static void port_change_phc(struct port *p)
"command line, not the attached ptp%d",
portnum(p), p->phc_index,
interface_phc_index(p->iface));
- } else if (p->phc_index != interface_phc_index(p->iface)) {
+ } else {
p->phc_index = interface_phc_index(p->iface);
if (clock_switch_phc(p->clock, p->phc_index)) {
@@ -2619,18 +2622,12 @@ void port_link_status(void *ctx, int linkup, int ts_index)
pr_notice("port %hu: ts label changed to %s", portnum(p), ts_label);
}
- /* phc index may changed while ts_label keeps the same after failover.
- * e.g. vlan over bond. Since the lower link changed, we still set
- * the TS_LABEL_CHANGED flag.
- */
+ /* The PHC index may change even with the same ts_label, e.g. after
+ failover with VLAN over bond. */
interface_get_tsinfo(p->iface);
- if (p->phc_index != interface_phc_index(p->iface))
- p->link_status |= TS_LABEL_CHANGED;
- /* Both link down/up and change ts_label may change phc index. */
- if (p->link_status & LINK_UP &&
- (p->link_status & LINK_STATE_CHANGED || p->link_status & TS_LABEL_CHANGED))
- port_change_phc(p);
+ /* Switch the clock if needed */
+ port_change_phc(p);
/*
* A port going down can affect the BMCA result.
commit 730a775fec159f012f3e0f65bb059465e9f7b49b
Author: Miroslav Lichvar <mlichvar@redhat.com>
Date: Wed Mar 15 11:24:16 2023 +0100
port: Don't switch to PHC with SW timestamping.
When ptp4l was configured to use SW timestamping, but the interface
supported also HW timestamping, ptp4l detected a change of the PHC on
start, switched to it from the system clock, and tried to control the
PHC using SW timestamps.
Don't switch the PHC if the current PHC index is -1, which is expected
with SW timestamping and in the free-running mode.
Fixes: afeabf3c90ed ("ptp4l: add VLAN over bond support")
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
diff --git a/port.c b/port.c
index 96e097a..32ee83b 100644
--- a/port.c
+++ b/port.c
@@ -2570,6 +2570,7 @@ static void port_change_phc(struct port *p)
/* Try to switch only if the interface is up, it has HW time stamping
using a non-vclock PHC, and the PHC actually changed. */
if (!(p->link_status & LINK_UP) ||
+ p->phc_index < 0 ||
!interface_tsinfo_valid(p->iface) ||
interface_get_vclock(p->iface) >= 0 ||
interface_phc_index(p->iface) < 0 ||

19
SPECS/linuxptp.spec
View File

@ -4,7 +4,7 @@
Name: linuxptp Name: linuxptp
Version: 3.1.1 Version: 3.1.1
Release: 5%{?dist} Release: 6%{?dist}.1
Summary: PTP implementation for Linux Summary: PTP implementation for Linux
License: GPLv2+ License: GPLv2+
@ -45,6 +45,12 @@ Patch11: linuxptp-clockcheck.patch
Patch12: linuxptp-vclock.patch Patch12: linuxptp-vclock.patch
# handle PHC read failing with EBUSY in phc2sys # handle PHC read failing with EBUSY in phc2sys
Patch13: linuxptp-phcerr.patch Patch13: linuxptp-phcerr.patch
# add support for VLAN over bond
Patch14: linuxptp-vlanbond.patch
# handle EINTR when waiting for transmit timestamp
Patch15: linuxptp-eintr.patch
# check for unexpected changes in frequency offset
Patch16: linuxptp-freqcheck.patch
BuildRequires: gcc gcc-c++ make systemd BuildRequires: gcc gcc-c++ make systemd
@ -71,6 +77,9 @@ Supporting legacy APIs and other platforms is not a goal.
%patch11 -p1 -b .clockcheck %patch11 -p1 -b .clockcheck
%patch12 -p1 -b .vclock %patch12 -p1 -b .vclock
%patch13 -p1 -b .phcerr %patch13 -p1 -b .phcerr
%patch14 -p1 -b .vlanbond
%patch15 -p1 -b .eintr
%patch16 -p1 -b .freqcheck
mv linuxptp-testsuite-%{testsuite_ver}* testsuite mv linuxptp-testsuite-%{testsuite_ver}* testsuite
mv clknetsim-%{clknetsim_ver}* testsuite/clknetsim mv clknetsim-%{clknetsim_ver}* testsuite/clknetsim
@ -133,6 +142,14 @@ PATH=..:$PATH ./run
%{_mandir}/man8/*.8* %{_mandir}/man8/*.8*
%changelog %changelog
* Mon Mar 20 2023 Miroslav Lichvar <mlichvar@redhat.com> 3.1.1-6.el9_2.1
- don't switch from system clock to PHC with SW timestamping (#2178233)
* Thu Jan 05 2023 Miroslav Lichvar <mlichvar@redhat.com> 3.1.1-6
- add support for VLAN over bond (#2120521)
- handle EINTR when waiting for transmit timestamp (#2128786)
- check for unexpected changes in frequency offset (#2150815)
* Thu Jul 28 2022 Miroslav Lichvar <mlichvar@redhat.com> 3.1.1-5 * Thu Jul 28 2022 Miroslav Lichvar <mlichvar@redhat.com> 3.1.1-5
- disable PHC switch with vclocks (#2066452) - disable PHC switch with vclocks (#2066452)