| Commit message (Collapse) | Author | Age |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
"FSL_stressPTP" tool is supposed to test the capability characteristics of
a ptp stack porting. The stresstest shall measure the overall load of the
stack under test by sending delay request and signalling messages.The higher
the overall load on the stack is, the higher your packet loss will be.
The patch fix the timestamp error in high overall load stress test like:
./FSL_stressPTP -u -b 10.192.242.10 -d 10.192.242.6 -e eth0 -m 00:01:02:04:04:19
-i 100 -t 10 -f -6 -s -6 -c 10
And enet interrupt coalescing feature introduce ptp packet latency, which may
cause that the timestamp don't store the related timestamp buffer when stack
use ioctl interface to access the message. So add some 4ms delay after getting
timestamp fail to let the next looking up success. Suggest to disable interrupt
coalescing feature by ethtool when run 1588 cases.
Signed-off-by: Fugang Duan <B38611@freescale.com>
(cherry picked from commit 11107938cc5a4a7cb4e5fc5e00d248183ed43d93)
|
| |
|
|
|
|
|
|
|
|
|
| |
1588 convergence process (setup time about 15s) is not ideal:
Applied a time jump on the reference master which causes both slaves
to apply a time jump as well and then synchronize back to the nanoseconds.
Optimize the 1588 adjust algorithm to get better convergent action.
Signed-off-by: Fugang Duan <B38611@freescale.com>
(cherry picked from commit 46a727f7ef469ae7c77c3458efd8a0de2396c360)
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
LinuxPTP stack
IEEE 1588 module has one hw issue in capturing the ATVR register. According
to the user manual it is:
ENET0->ATCR |= ENET_ATCR_CAPTURE_MASK;
while(ENET0->ATCR & ENET_ATCR_CAPTURE_MASK);
ts_counter_ns = ENET0->ATVR;
Incorrect behavior for ENET_ATCR[Capture and Restart Bits]. These bits will always
read a value zero. According to SPEC, when these bits are set to 1'b1, these should
hold value 1'b1 until the counter value is capture in the register clock domain.
Unfortunately there is a bug with the way the bit "ENET_ATCR_CAPTURE" clears.
So need something like:
ENET0->ATCR |= ENET_ATCR_CAPTURE_MASK;
wait();
ts_counter_ns = ENET0->ATVR;
The wait-time to be at least 6 clock cycle of the slower clock between the register
clock and the 1588 clock. The 1588 ts_clk is 25Mhz, register clock is 66Mhz, so the
wait-time must be greater than 240ns (40ns * 6). The workaround is that adding 1us
delay before read ATVR.
Signed-off-by: Fugang Duan <B38611@freescale.com>
|
| |
|
|
|
|
|
|
|
|
| |
Avoid fec register access when clock is disabled.
After fec probed or "ifconfig ethx down", it disable fec clock
to save power. PTP use period timer softirq to read ptp timer
count register regardless any ethernet status, which cause
system hang. The patch fix the issue.
Signed-off-by: Fugang Duan <B38611@freescale.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The timestamps generated in the i.MX drivers are generated by the
nanoseconds part coming from the 1588 clock. But the number of seconds
are maintained in a private structure of the interface. Those are
updated in a 1588 clock rollover interrupt.
The timestamp is generated right before a rollover of a second and the
timestamp value is constructed afterwards. Therefore the bigger part of
the timestamp is wrong (the second).
commit:54181c1d83e04b18e63c7723ac80f974b760e019
Suggested solution (pseudo-code):
If( actual-time.nsec < timestamp.nsec )
Timestamp.sec = fpp->prtc -1;
Else
Timestamp.sec = fpp->prtc;
But it is not perfect and there still exist potenitial second sync issue.
So, the patch Suggested solution (pseudo-code):
If( actual-time.nsec < timestamp.nsec &&
!FEC_IEVENT[TS_TIMER] )
Timestamp.sec = fpp->prtc -1;
Else
Timestamp.sec = fpp->prtc;
Signed-off-by: Fugang Duan <B38611@freescale.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The timestamps generated in the i.MX drivers are generated by the
nanoseconds part coming from the 1588 clock. But the number of seconds
are maintained in a private structure of the interface. Those are
updated in a 1588 clock rollover interrupt.
The timestamp is generated right before a rollover of a second and the
timestamp value is constructed afterwards. Therefore the bigger part of
the timestamp is wrong (the second).
Suggested solution (pseudo-code):
If( actual-time.nsec < timestamp.nsec )
Timestamp.sec = fpp->prtc -1;
Else
Timestamp.sec = fpp->prtc;
Signed-off-by: Fugang Duan <B38611@freescale.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
interrupt context
Kernel warning dump by enable kernel config "CONFIG_DEBUG_MUTEXES":
------------[ cut here ]------------
WARNING: at kernel/mutex.c:577 mutex_trylock+0x180/0x1d0()
DEBUG_LOCKS_WARN_ON(in_interrupt())
Modules linked in:
CPU: 0 PID: 68 Comm: kworker/0:2 Tainted: G W 3.10.17-16855-ga44de14 #1325
Workqueue: events phy_state_machine
[<80014cbc>] (unwind_backtrace+0x0/0x138) from [<8001251c>]
(show_stack+0x10/0x14)
[<8001251c>] (show_stack+0x10/0x14) from [<80026754>]
(warn_slowpath_common+0x4c/0x68)
[<80026754>] (warn_slowpath_common+0x4c/0x68) from [<80026804>]
(warn_slowpath_fmt+0x30/0x40)
[<80026804>] (warn_slowpath_fmt+0x30/0x40) from [<8069f6b0>]
(mutex_trylock+0x180/0x1d0)
[<8069f6b0>] (mutex_trylock+0x180/0x1d0) from [<804dce7c>]
(clk_prepare_lock+0xc/0xd8)
[<804dce7c>] (clk_prepare_lock+0xc/0xd8) from [<804ddbcc>]
(clk_get_rate+0xc/0x5c)
[<804ddbcc>] (clk_get_rate+0xc/0x5c) from [<803b7528>]
(fec_ptp_start_cyclecounter+0x1c/0x198)
[<803b7528>] (fec_ptp_start_cyclecounter+0x1c/0x198) from [<803b5928>]
(fec_restart+0x6e8/0x870)
[<803b5928>] (fec_restart+0x6e8/0x870) from [<803b5d50>]
(fec_enet_adjust_link+0x7c/0xb4)
[<803b5d50>] (fec_enet_adjust_link+0x7c/0xb4) from [<803b07b8>]
(phy_state_machine+0xfc/0x394)
[<803b07b8>] (phy_state_machine+0xfc/0x394) from [<8003f03c>]
(process_one_work+0x198/0x428)
[<8003f03c>] (process_one_work+0x198/0x428) from [<8003fd24>]
(worker_thread+0x144/0x3a4)
[<8003fd24>] (worker_thread+0x144/0x3a4) from [<800458d8>]
(kthread+0xa4/0xb0)
[<800458d8>] (kthread+0xa4/0xb0) from [<8000ebd8>] (ret_from_fork+0x14/0x3c)
---[ end trace d1930b3e1c195329 ]---
Root cause:
Worker thread call netif_tx_lock_bh() to diable the softirq preempt, and
then call clk_get_rate() to get ptp clock rate.
In fact, netif_tx_lock_bh()->local_bh_disable(), which make in_interrupt()
to be ture. clk_get_rate()->clk_prepare_lock()->mutex_trylock(), and
mutex_trylock() cannot use at interrupt context, otherwise there have
kernel dump.
So, remove the clk_get_rate() in there.
Signed-off-by: Fugang Duan <B38611@freescale.com>
|
| |
|
|
|
|
|
| |
These patch add ptp support for IXXAT stack.
Cherry picked from commit 1c8839574ac87826f53b96f987975a9bb1d72915.
Signed-off-by: Luwei Zhou <B45643@freescale.com>
|
| |
|
|
|
|
|
| |
Passing pdev in fec_ptp_init() is enough, since we can get ndev locally.
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
| |
|
|
|
|
|
|
|
|
|
| |
Use a more current logging message style.
Convert the printks where a struct net_device is available to
netdev_<level>. Convert the other printks to pr_<level> and
add pr_fmt where appropriate.
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently when booting a mx6 device we get the following on boot:
registered PHC device on eth%d
Fix it by printing the network device name only after it gets registered, so
that the following can be read now:
fec 2188000.ethernet eth0: registered PHC device 0
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
| |
|
|
|
|
|
|
|
|
| |
fec_ptp.ko can't run individually
rename fec.c to fec_main.c
Build fec.o and fec_ptp.o into one fec.ko
Remove unnessary EXPORT_SYMBOL in fec_ptp
Signed-off-by: Frank Li <Frank.Li@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Since commit ff43da86c69 (NET: FEC: dynamtic check DMA desc buff type) the
following build error happens when CONFIG_FEC=m
ERROR: "fec_ptp_init" [drivers/net/ethernet/freescale/fec.ko] undefined!
ERROR: "fec_ptp_ioctl" [drivers/net/ethernet/freescale/fec.ko] undefined!
ERROR: "fec_ptp_start_cyclecounter" [drivers/net/ethernet/freescale/fec.ko] undefined!
Fix it by exporting the required fec_ptp symbols.
Reported-by: Uwe Kleine-Koenig <u.kleine-koenig@pengutronix.de>
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
=========================================================
[ INFO: possible irq lock inversion dependency detected ]
3.8.0-rc5+ #82 Not tainted
---------------------------------------------------------
swapper/0/0 just changed the state of lock:
(&(&fep->hw_lock)->rlock){..-...}, at: [<8034e2f8>] fec_enet_start_xmit+0x48/0x 2cc
but this lock took another, SOFTIRQ-unsafe lock in the past:
(prepare_lock){+.+.+.}
and interrupts could create inverse lock ordering between them.
other info that might help us debug this:
Possible interrupt unsafe locking scenario:
CPU0 CPU1
---- ----
lock(prepare_lock);
local_irq_disable()
lock(&(&fep->hw_lock)->rlock);
lock(prepare_lock);
<Interrupt>
lock(&(&fep->hw_lock)->rlock);
*** DEADLOCK ***
Signed-off-by: Frank Li <Frank.Li@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
move below calculate out of spin lock section
diff = fep->cc.mult;
diff *= ppb;
diff = div_u64(diff, 1000000000ULL);
diff is local variable and not neccesary in spin lock
Signed-off-by: Frank Li <Frank.Li@freescale.com>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
|
This patch adds a driver for the FEC(MX6) that offers time
stamping and a PTP haderware clock. Because FEC\ENET(MX6)
hardware frequency adjustment is complex, we have implemented
this in software by changing the multiplication factor of the
timecounter.
Signed-off-by: Frank Li <Frank.Li@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
