Hi Richard, All,
This is re-spin of patches to add CPSW IRQ and HW_TS_PUSH events support I've
sent long time ago [1]. In this series, I've tried to restructure and split changes,
and also add few additional optimizations comparing to initial RFC submission [1].
The HW_TS_PUSH events intended to serve for different timesync purposes on of
which is to add PPS generation function, which can be implemented as below:
+-----------------+
| Control |
| application |
+------->+ +----------+
| | | |
| | | |
| +-----------------+ |
| |
| |
| PTP_EXTTS_REQUEST |
| |
| |
+----------------------------------------------------------------+
| | Kernel
+-------+----------+ +-------v--------+
| \dev\ptpX | | /sys/class/pwm/|
| | | |
+-------^----------+ +-------+--------+
| |
| |
| +-------v-------------------+
+-------+----------+ | |
| CPTS driver | |pwm/pwm-omap-dmtimer.c |
| | +---------------------------+
+-------^----------+ |clocksource/timer_ti_dm.c |
| +-------+-------------------+
|HWx_TS_PUSH evt |
+----------------------------------------------------------------+
| | HW
+-------+----------+ +-------v--------+
| CPTS | | DMTimer |
| | | |
| HWx_TS_PUSH X<-----------------+ |
| + | |
+------------------+ +-------+--------+
|
X timer4
As per my knowledge there is at least one public implemented above PPS generation
schema from Tusori Tibor [2] based on initial HW_TS_PUSH enable submission[1].
And now there is work done by Lokesh Vutla <[email protected]> published to
enable PWM enable/improve PWM adjustment from user space [3][4][5].
Main changes comparing to initial submission:
- TX timestamp processing deferred to ptp worker only
- both CPTS IRQ and polling events processing supported to make it work for
Keystone 2 also
- switch to use new .gettimex64() interface
- no DT updates as number of HWx_TS_PUSH inputs is static per HW
Testing on am571x-idk/omap2plus_defconfig/+CONFIG_PREEMPT=y:
1) testing HW_TS_PUSH
- enable pwm in DT
pwm16: dmtimer-pwm {
compatible = "ti,omap-dmtimer-pwm";
ti,timers = <&timer16>;
#pwm-cells = <3>;
};
- configure and start pwm
echo 0 > /sys/class/pwm/pwmchip0/export
echo 1000000000 > /sys/class/pwm/pwmchip0/pwm0/period
echo 500000000 > /sys/class/pwm/pwmchip0/pwm0/duty_cycle
echo 1 > /sys/class/pwm/pwmchip0/pwm0/enable
- test HWx_TS_PUSH using Kernel selftest testptp application
./tools/testing/selftests/ptp/testptp -d /dev/ptp0 -e 1000 -i 3
2) testing phc2sys
# ./linuxptp/phc2sys -s CLOCK_REALTIME -c eth0 -m -O 0 -u30
phc2sys[1616.791]: eth0 rms 408190379792180864 max 1580914543017209856 freq +864 +/- 4635 delay 645 +/- 29
phc2sys[1646.795]: eth0 rms 41 max 108 freq +0 +/- 36 delay 656 +/- 29
phc2sys[1676.800]: eth0 rms 43 max 83 freq +2 +/- 38 delay 650 +/- 0
phc2sys[1706.804]: eth0 rms 39 max 87 freq +4 +/- 34 delay 672 +/- 55
phc2sys[1736.808]: eth0 rms 35 max 66 freq +1 +/- 30 delay 667 +/- 49
phc2sys[1766.813]: eth0 rms 38 max 79 freq +2 +/- 33 delay 656 +/- 29
phc2sys[1796.817]: eth0 rms 45 max 98 freq +1 +/- 39 delay 656 +/- 29
phc2sys[1826.821]: eth0 rms 40 max 87 freq +5 +/- 35 delay 650 +/- 0
phc2sys[1856.826]: eth0 rms 29 max 76 freq -0 +/- 25 delay 656 +/- 29
phc2sys[1886.830]: eth0 rms 40 max 97 freq +4 +/- 35 delay 667 +/- 49
phc2sys[1916.834]: eth0 rms 42 max 94 freq +2 +/- 36 delay 661 +/- 41
phc2sys[1946.839]: eth0 rms 40 max 91 freq +2 +/- 35 delay 661 +/- 41
phc2sys[1976.843]: eth0 rms 46 max 88 freq -0 +/- 40 delay 667 +/- 49
phc2sys[2006.847]: eth0 rms 49 max 97 freq +2 +/- 43 delay 650 +/- 0
3) testing ptp4l
- 1G connection
# ./linuxptp/ptp4l -P -2 -H -i eth0 -l 6 -m -q -p /dev/ptp0 -f ptp.cfg -s
ptp4l[862.891]: port 1: UNCALIBRATED to SLAVE on MASTER_CLOCK_SELECTED
ptp4l[923.894]: rms 1019697354682 max 5768279314068 freq +26053 +/- 72 delay 488 +/- 1
ptp4l[987.896]: rms 13 max 26 freq +26005 +/- 29 delay 488 +/- 1
ptp4l[1051.899]: rms 14 max 50 freq +25895 +/- 21 delay 488 +/- 1
ptp4l[1115.901]: rms 11 max 27 freq +25878 +/- 17 delay 488 +/- 1
ptp4l[1179.904]: rms 10 max 27 freq +25857 +/- 12 delay 488 +/- 1
ptp4l[1243.906]: rms 14 max 37 freq +25851 +/- 15 delay 488 +/- 1
ptp4l[1307.909]: rms 12 max 33 freq +25835 +/- 15 delay 488 +/- 1
ptp4l[1371.911]: rms 11 max 27 freq +25832 +/- 14 delay 488 +/- 1
ptp4l[1435.914]: rms 11 max 26 freq +25823 +/- 11 delay 488 +/- 1
ptp4l[1499.916]: rms 10 max 29 freq +25829 +/- 11 delay 489 +/- 1
ptp4l[1563.919]: rms 11 max 27 freq +25827 +/- 12 delay 488 +/- 1
- 10M connection
# ./linuxptp/ptp4l -P -2 -H -i eth0 -l 6 -m -q -p /dev/ptp0 -f ptp.cfg -s
ptp4l[51.955]: port 1: UNCALIBRATED to SLAVE on MASTER_CLOCK_SELECTED
ptp4l[112.957]: rms 279468848453933920 max 1580914542977391360 freq +25390 +/- 3207 delay 8222 +/- 36
ptp4l[176.960]: rms 254 max 522 freq +25809 +/- 219 delay 8271 +/- 30
ptp4l[240.962]: rms 271 max 684 freq +25868 +/- 234 delay 8249 +/- 22
ptp4l[304.965]: rms 263 max 556 freq +25894 +/- 227 delay 8225 +/- 47
ptp4l[368.967]: rms 238 max 648 freq +25908 +/- 204 delay 8234 +/- 40
ptp4l[432.970]: rms 274 max 658 freq +25932 +/- 237 delay 8241 +/- 22
ptp4l[496.972]: rms 247 max 557 freq +25943 +/- 213 delay 8223 +/- 26
ptp4l[560.974]: rms 291 max 756 freq +25968 +/- 251 delay 8244 +/- 41
ptp4l[624.977]: rms 249 max 697 freq +25975 +/- 216 delay 8258 +/- 22
Changes in v5:
- fixed build issue
Changes in v4:
- fixed comments from Richard Cochran
- dropped patch "net: ethernet: ti: cpts: move rx timestamp processing to ptp
worker only"
- added "Acked-by" from Richard Cochran <[email protected]>
- dependencies resolved, patch merged
Changes in v3:
- fixed rebase mess
- fixed build issues
Changes in v2 (broken):
- fixed (formatting) comments from David Miller <[email protected]>
v4: https://patchwork.ozlabs.org/project/netdev/cover/[email protected]/
v3: https://patchwork.ozlabs.org/project/netdev/cover/[email protected]/
v2: https://patchwork.ozlabs.org/cover/1258339/
v1: https://patchwork.ozlabs.org/cover/1254708/
[1] https://lore.kernel.org/patchwork/cover/799251/
[2] https://usermanual.wiki/Document/SetupGuide.632280828.pdf
https://github.com/t-tibor/msc_thesis
[3] https://patchwork.kernel.org/cover/11421329/
[4] https://patchwork.kernel.org/cover/11433197/
[5] https://sourceforge.net/p/linuxptp/mailman/message/36943248/
Grygorii Strashko (10):
net: ethernet: ti: cpts: use dev_yy() api for logs
net: ethernet: ti: cpts: separate hw counter read from timecounter
net: ethernet: ti: cpts: move tc mult update in cpts_fifo_read()
net: ethernet: ti: cpts: switch to use new .gettimex64() interface
net: ethernet: ti: cpts: optimize packet to event matching
net: ethernet: ti: cpts: move tx timestamp processing to ptp worker only
net: ethernet: ti: cpts: rework locking
net: ethernet: ti: cpts: add irq support
net: ethernet: ti: cpts: add support for HW_TS_PUSH events
net: ethernet: ti: cpsw: enable cpts irq
drivers/net/ethernet/ti/cpsw.c | 21 ++
drivers/net/ethernet/ti/cpsw_new.c | 20 ++
drivers/net/ethernet/ti/cpsw_priv.c | 17 +-
drivers/net/ethernet/ti/cpsw_priv.h | 2 +
drivers/net/ethernet/ti/cpts.c | 422 +++++++++++++++++---------
drivers/net/ethernet/ti/cpts.h | 27 +-
drivers/net/ethernet/ti/netcp_ethss.c | 3 +-
7 files changed, 363 insertions(+), 149 deletions(-)
--
2.17.1
The CPTS HW latches and saves CPTS counter value in CPTS fifo immediately
after writing to CPSW_CPTS_PUSH.TS_PUSH (bit 0), so the total time that the
driver needs to read the CPTS timestamp is the time required CPSW_CPTS_PUSH
write to actually reach HW.
Hence switch CPTS driver to implement new .gettimex64() callback for more
precise measurement of the offset between a PHC and the system clock which
is measured as time between
write(CPSW_CPTS_PUSH)
read(CPSW_CPTS_PUSH)
Signed-off-by: Grygorii Strashko <[email protected]>
Acked-by: Richard Cochran <[email protected]>
---
drivers/net/ethernet/ti/cpts.c | 22 ++++++++++++++--------
1 file changed, 14 insertions(+), 8 deletions(-)
diff --git a/drivers/net/ethernet/ti/cpts.c b/drivers/net/ethernet/ti/cpts.c
index a2974b542bed..1f738bb3df74 100644
--- a/drivers/net/ethernet/ti/cpts.c
+++ b/drivers/net/ethernet/ti/cpts.c
@@ -203,9 +203,13 @@ static u64 cpts_systim_read(const struct cyclecounter *cc)
return READ_ONCE(cpts->cur_timestamp);
}
-static void cpts_update_cur_time(struct cpts *cpts, int match)
+static void cpts_update_cur_time(struct cpts *cpts, int match,
+ struct ptp_system_timestamp *sts)
{
+ ptp_read_system_prets(sts);
cpts_write32(cpts, TS_PUSH, ts_push);
+ cpts_read32(cpts, ts_push);
+ ptp_read_system_postts(sts);
if (cpts_fifo_read(cpts, match) && match != -1)
dev_err(cpts->dev, "cpts: unable to obtain a time stamp\n");
@@ -234,7 +238,7 @@ static int cpts_ptp_adjfreq(struct ptp_clock_info *ptp, s32 ppb)
cpts->mult_new = neg_adj ? mult - diff : mult + diff;
- cpts_update_cur_time(cpts, CPTS_EV_PUSH);
+ cpts_update_cur_time(cpts, CPTS_EV_PUSH, NULL);
spin_unlock_irqrestore(&cpts->lock, flags);
@@ -253,15 +257,17 @@ static int cpts_ptp_adjtime(struct ptp_clock_info *ptp, s64 delta)
return 0;
}
-static int cpts_ptp_gettime(struct ptp_clock_info *ptp, struct timespec64 *ts)
+static int cpts_ptp_gettimeex(struct ptp_clock_info *ptp,
+ struct timespec64 *ts,
+ struct ptp_system_timestamp *sts)
{
- u64 ns;
- unsigned long flags;
struct cpts *cpts = container_of(ptp, struct cpts, info);
+ unsigned long flags;
+ u64 ns;
spin_lock_irqsave(&cpts->lock, flags);
- cpts_update_cur_time(cpts, CPTS_EV_PUSH);
+ cpts_update_cur_time(cpts, CPTS_EV_PUSH, sts);
ns = timecounter_read(&cpts->tc);
spin_unlock_irqrestore(&cpts->lock, flags);
@@ -302,7 +308,7 @@ static long cpts_overflow_check(struct ptp_clock_info *ptp)
spin_lock_irqsave(&cpts->lock, flags);
- cpts_update_cur_time(cpts, -1);
+ cpts_update_cur_time(cpts, -1, NULL);
ns = timecounter_read(&cpts->tc);
@@ -326,7 +332,7 @@ static const struct ptp_clock_info cpts_info = {
.pps = 0,
.adjfreq = cpts_ptp_adjfreq,
.adjtime = cpts_ptp_adjtime,
- .gettime64 = cpts_ptp_gettime,
+ .gettimex64 = cpts_ptp_gettimeex,
.settime64 = cpts_ptp_settime,
.enable = cpts_ptp_enable,
.do_aux_work = cpts_overflow_check,
--
2.17.1
Now CPTS driver .adjfreq() generates request to read CPTS current time
(CPTS_EV_PUSH) with intention to process all pending event using previous
frequency adjustment values before switching to the new ones. So
CPTS_EV_PUSH works as a marker to switch to the new frequency adjustment
values. Current code assumes that all job is done in .adjfreq(), but after
enabling IRQ this will not be true any more.
Hence save new frequency adjustment values (mult) and perform actual freq
adjustment in cpts_fifo_read() immediately after CPTS_EV_PUSH is received.
Signed-off-by: Grygorii Strashko <[email protected]>
Acked-by: Richard Cochran <[email protected]>
---
drivers/net/ethernet/ti/cpts.c | 8 ++++++--
drivers/net/ethernet/ti/cpts.h | 1 +
2 files changed, 7 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/ti/cpts.c b/drivers/net/ethernet/ti/cpts.c
index f40a864d8c36..a2974b542bed 100644
--- a/drivers/net/ethernet/ti/cpts.c
+++ b/drivers/net/ethernet/ti/cpts.c
@@ -165,6 +165,10 @@ static int cpts_fifo_read(struct cpts *cpts, int match)
case CPTS_EV_PUSH:
WRITE_ONCE(cpts->cur_timestamp, lo);
timecounter_read(&cpts->tc);
+ if (cpts->mult_new) {
+ cpts->cc.mult = cpts->mult_new;
+ cpts->mult_new = 0;
+ }
break;
case CPTS_EV_TX:
if (cpts_match_tx_ts(cpts, event)) {
@@ -228,9 +232,9 @@ static int cpts_ptp_adjfreq(struct ptp_clock_info *ptp, s32 ppb)
spin_lock_irqsave(&cpts->lock, flags);
- cpts_update_cur_time(cpts, CPTS_EV_PUSH);
+ cpts->mult_new = neg_adj ? mult - diff : mult + diff;
- cpts->cc.mult = neg_adj ? mult - diff : mult + diff;
+ cpts_update_cur_time(cpts, CPTS_EV_PUSH);
spin_unlock_irqrestore(&cpts->lock, flags);
diff --git a/drivers/net/ethernet/ti/cpts.h b/drivers/net/ethernet/ti/cpts.h
index 32ecd1ce4d3b..421630049ee7 100644
--- a/drivers/net/ethernet/ti/cpts.h
+++ b/drivers/net/ethernet/ti/cpts.h
@@ -116,6 +116,7 @@ struct cpts {
unsigned long ov_check_period;
struct sk_buff_head txq;
u64 cur_timestamp;
+ u32 mult_new;
};
void cpts_rx_timestamp(struct cpts *cpts, struct sk_buff *skb);
--
2.17.1
The CPSW misc IRQ need be enabled for CPTS event_pend IRQs processing. This
patch adds corresponding support to CPSW driver.
Signed-off-by: Grygorii Strashko <[email protected]>
Acked-by: Richard Cochran <[email protected]>
---
drivers/net/ethernet/ti/cpsw.c | 21 +++++++++++++++++++++
drivers/net/ethernet/ti/cpsw_new.c | 20 ++++++++++++++++++++
drivers/net/ethernet/ti/cpsw_priv.c | 12 ++++++++++++
drivers/net/ethernet/ti/cpsw_priv.h | 2 ++
4 files changed, 55 insertions(+)
diff --git a/drivers/net/ethernet/ti/cpsw.c b/drivers/net/ethernet/ti/cpsw.c
index c2c5bf87da01..09f98fa2fb4e 100644
--- a/drivers/net/ethernet/ti/cpsw.c
+++ b/drivers/net/ethernet/ti/cpsw.c
@@ -1569,6 +1569,12 @@ static int cpsw_probe(struct platform_device *pdev)
return irq;
cpsw->irqs_table[1] = irq;
+ /* get misc irq*/
+ irq = platform_get_irq(pdev, 3);
+ if (irq <= 0)
+ return irq;
+ cpsw->misc_irq = irq;
+
/*
* This may be required here for child devices.
*/
@@ -1703,6 +1709,21 @@ static int cpsw_probe(struct platform_device *pdev)
goto clean_unregister_netdev_ret;
}
+ if (!cpsw->cpts)
+ goto skip_cpts;
+
+ ret = devm_request_irq(&pdev->dev, cpsw->misc_irq, cpsw_misc_interrupt,
+ 0, dev_name(&pdev->dev), cpsw);
+ if (ret < 0) {
+ dev_err(dev, "error attaching misc irq (%d)\n", ret);
+ goto clean_unregister_netdev_ret;
+ }
+
+ /* Enable misc CPTS evnt_pend IRQ */
+ cpts_set_irqpoll(cpsw->cpts, false);
+ writel(0x10, &cpsw->wr_regs->misc_en);
+
+skip_cpts:
cpsw_notice(priv, probe,
"initialized device (regs %pa, irq %d, pool size %d)\n",
&ss_res->start, cpsw->irqs_table[0], descs_pool_size);
diff --git a/drivers/net/ethernet/ti/cpsw_new.c b/drivers/net/ethernet/ti/cpsw_new.c
index 9209e613257d..33c8dd686206 100644
--- a/drivers/net/ethernet/ti/cpsw_new.c
+++ b/drivers/net/ethernet/ti/cpsw_new.c
@@ -1896,6 +1896,11 @@ static int cpsw_probe(struct platform_device *pdev)
return irq;
cpsw->irqs_table[1] = irq;
+ irq = platform_get_irq_byname(pdev, "misc");
+ if (irq <= 0)
+ return irq;
+ cpsw->misc_irq = irq;
+
platform_set_drvdata(pdev, cpsw);
/* This may be required here for child devices. */
pm_runtime_enable(dev);
@@ -1975,6 +1980,21 @@ static int cpsw_probe(struct platform_device *pdev)
goto clean_unregister_netdev;
}
+ if (!cpsw->cpts)
+ goto skip_cpts;
+
+ ret = devm_request_irq(dev, cpsw->misc_irq, cpsw_misc_interrupt,
+ 0, dev_name(&pdev->dev), cpsw);
+ if (ret < 0) {
+ dev_err(dev, "error attaching misc irq (%d)\n", ret);
+ goto clean_unregister_netdev;
+ }
+
+ /* Enable misc CPTS evnt_pend IRQ */
+ cpts_set_irqpoll(cpsw->cpts, false);
+ writel(0x10, &cpsw->wr_regs->misc_en);
+
+skip_cpts:
ret = cpsw_register_notifiers(cpsw);
if (ret)
goto clean_unregister_netdev;
diff --git a/drivers/net/ethernet/ti/cpsw_priv.c b/drivers/net/ethernet/ti/cpsw_priv.c
index 099208927400..9d098c802c6d 100644
--- a/drivers/net/ethernet/ti/cpsw_priv.c
+++ b/drivers/net/ethernet/ti/cpsw_priv.c
@@ -114,6 +114,18 @@ irqreturn_t cpsw_rx_interrupt(int irq, void *dev_id)
return IRQ_HANDLED;
}
+irqreturn_t cpsw_misc_interrupt(int irq, void *dev_id)
+{
+ struct cpsw_common *cpsw = dev_id;
+
+ writel(0, &cpsw->wr_regs->misc_en);
+ cpdma_ctlr_eoi(cpsw->dma, CPDMA_EOI_MISC);
+ cpts_misc_interrupt(cpsw->cpts);
+ writel(0x10, &cpsw->wr_regs->misc_en);
+
+ return IRQ_HANDLED;
+}
+
int cpsw_tx_mq_poll(struct napi_struct *napi_tx, int budget)
{
struct cpsw_common *cpsw = napi_to_cpsw(napi_tx);
diff --git a/drivers/net/ethernet/ti/cpsw_priv.h b/drivers/net/ethernet/ti/cpsw_priv.h
index b8d7b924ee3d..bf4e179b4ca4 100644
--- a/drivers/net/ethernet/ti/cpsw_priv.h
+++ b/drivers/net/ethernet/ti/cpsw_priv.h
@@ -350,6 +350,7 @@ struct cpsw_common {
bool rx_irq_disabled;
bool tx_irq_disabled;
u32 irqs_table[IRQ_NUM];
+ int misc_irq;
struct cpts *cpts;
struct devlink *devlink;
int rx_ch_num, tx_ch_num;
@@ -442,6 +443,7 @@ int cpsw_run_xdp(struct cpsw_priv *priv, int ch, struct xdp_buff *xdp,
struct page *page, int port);
irqreturn_t cpsw_tx_interrupt(int irq, void *dev_id);
irqreturn_t cpsw_rx_interrupt(int irq, void *dev_id);
+irqreturn_t cpsw_misc_interrupt(int irq, void *dev_id);
int cpsw_tx_mq_poll(struct napi_struct *napi_tx, int budget);
int cpsw_tx_poll(struct napi_struct *napi_tx, int budget);
int cpsw_rx_mq_poll(struct napi_struct *napi_rx, int budget);
--
2.17.1
Now CPTS HW time reading code is implemented in timecounter->cyclecounter
.read() callback and performs following operations:
timecounter_read() ->cc.read() -> cpts_systim_read()
- request current CPTS HW time CPTS_TS_PUSH.TS_PUSH = 1
- poll CPTS FIFO for CPTS_EV_PUSH event with current HW timestamp
This approach need to be changed for the future switch to PTP PHC
.gettimex64() callback, which require to separate requesting current CPTS
HW time and processing CPTS FIFO. And for the follow up patch, which
improves .adjfreq() implementation.
This patch moves code accessing CPTS HW out of timecounter code as
following:
- convert HW timestamp of every CPTS event to PTP time (us) and store it as
part struct cpts_event;
- add CPTS context field to store current CPTS HW time (counter) value and
update it on CPTS_EV_PUSH reception;
- move code accessing CPTS HW out of timecounter code and use current CPTS
HW time (counter) from CPTS context instead;
- ensure timecounter->cycle_last is updated on CPTS_EV_PUSH reception.
After this change CPTS timecounter will only perform timekeeper role
without actually accessing CPTS HW.
Signed-off-by: Grygorii Strashko <[email protected]>
Acked-by: Richard Cochran <[email protected]>
---
drivers/net/ethernet/ti/cpts.c | 53 +++++++++++++++++-----------------
drivers/net/ethernet/ti/cpts.h | 2 ++
2 files changed, 29 insertions(+), 26 deletions(-)
diff --git a/drivers/net/ethernet/ti/cpts.c b/drivers/net/ethernet/ti/cpts.c
index 445f445185df..f40a864d8c36 100644
--- a/drivers/net/ethernet/ti/cpts.c
+++ b/drivers/net/ethernet/ti/cpts.c
@@ -112,10 +112,8 @@ static bool cpts_match_tx_ts(struct cpts *cpts, struct cpts_event *event)
(struct cpts_skb_cb_data *)skb->cb;
if (cpts_match(skb, class, seqid, mtype)) {
- u64 ns = timecounter_cyc2time(&cpts->tc, event->low);
-
memset(&ssh, 0, sizeof(ssh));
- ssh.hwtstamp = ns_to_ktime(ns);
+ ssh.hwtstamp = ns_to_ktime(event->timestamp);
skb_tstamp_tx(skb, &ssh);
found = true;
__skb_unlink(skb, &cpts->txq);
@@ -158,8 +156,16 @@ static int cpts_fifo_read(struct cpts *cpts, int match)
event->tmo = jiffies + 2;
event->high = hi;
event->low = lo;
+ event->timestamp = timecounter_cyc2time(&cpts->tc, event->low);
type = event_type(event);
+
+ dev_dbg(cpts->dev, "CPTS_EV: %d high:%08X low:%08x\n",
+ type, event->high, event->low);
switch (type) {
+ case CPTS_EV_PUSH:
+ WRITE_ONCE(cpts->cur_timestamp, lo);
+ timecounter_read(&cpts->tc);
+ break;
case CPTS_EV_TX:
if (cpts_match_tx_ts(cpts, event)) {
/* if the new event matches an existing skb,
@@ -168,7 +174,6 @@ static int cpts_fifo_read(struct cpts *cpts, int match)
break;
}
/* fall through */
- case CPTS_EV_PUSH:
case CPTS_EV_RX:
list_del_init(&event->list);
list_add_tail(&event->list, &cpts->events);
@@ -189,26 +194,17 @@ static int cpts_fifo_read(struct cpts *cpts, int match)
static u64 cpts_systim_read(const struct cyclecounter *cc)
{
- u64 val = 0;
- struct cpts_event *event;
- struct list_head *this, *next;
struct cpts *cpts = container_of(cc, struct cpts, cc);
- cpts_write32(cpts, TS_PUSH, ts_push);
- if (cpts_fifo_read(cpts, CPTS_EV_PUSH))
- dev_err(cpts->dev, "cpts: unable to obtain a time stamp\n");
+ return READ_ONCE(cpts->cur_timestamp);
+}
- list_for_each_safe(this, next, &cpts->events) {
- event = list_entry(this, struct cpts_event, list);
- if (event_type(event) == CPTS_EV_PUSH) {
- list_del_init(&event->list);
- list_add(&event->list, &cpts->pool);
- val = event->low;
- break;
- }
- }
+static void cpts_update_cur_time(struct cpts *cpts, int match)
+{
+ cpts_write32(cpts, TS_PUSH, ts_push);
- return val;
+ if (cpts_fifo_read(cpts, match) && match != -1)
+ dev_err(cpts->dev, "cpts: unable to obtain a time stamp\n");
}
/* PTP clock operations */
@@ -232,7 +228,7 @@ static int cpts_ptp_adjfreq(struct ptp_clock_info *ptp, s32 ppb)
spin_lock_irqsave(&cpts->lock, flags);
- timecounter_read(&cpts->tc);
+ cpts_update_cur_time(cpts, CPTS_EV_PUSH);
cpts->cc.mult = neg_adj ? mult - diff : mult + diff;
@@ -260,6 +256,9 @@ static int cpts_ptp_gettime(struct ptp_clock_info *ptp, struct timespec64 *ts)
struct cpts *cpts = container_of(ptp, struct cpts, info);
spin_lock_irqsave(&cpts->lock, flags);
+
+ cpts_update_cur_time(cpts, CPTS_EV_PUSH);
+
ns = timecounter_read(&cpts->tc);
spin_unlock_irqrestore(&cpts->lock, flags);
@@ -294,11 +293,14 @@ static long cpts_overflow_check(struct ptp_clock_info *ptp)
{
struct cpts *cpts = container_of(ptp, struct cpts, info);
unsigned long delay = cpts->ov_check_period;
- struct timespec64 ts;
unsigned long flags;
+ u64 ns;
spin_lock_irqsave(&cpts->lock, flags);
- ts = ns_to_timespec64(timecounter_read(&cpts->tc));
+
+ cpts_update_cur_time(cpts, -1);
+
+ ns = timecounter_read(&cpts->tc);
if (!skb_queue_empty(&cpts->txq)) {
cpts_purge_txq(cpts);
@@ -307,8 +309,7 @@ static long cpts_overflow_check(struct ptp_clock_info *ptp)
}
spin_unlock_irqrestore(&cpts->lock, flags);
- dev_dbg(cpts->dev, "cpts overflow check at %lld.%09ld\n",
- (long long)ts.tv_sec, ts.tv_nsec);
+ dev_dbg(cpts->dev, "cpts overflow check at %lld\n", ns);
return (long)delay;
}
@@ -390,7 +391,7 @@ static u64 cpts_find_ts(struct cpts *cpts, struct sk_buff *skb, int ev_type)
seqid = (event->high >> SEQUENCE_ID_SHIFT) & SEQUENCE_ID_MASK;
if (ev_type == event_type(event) &&
cpts_match(skb, class, seqid, mtype)) {
- ns = timecounter_cyc2time(&cpts->tc, event->low);
+ ns = event->timestamp;
list_del_init(&event->list);
list_add(&event->list, &cpts->pool);
break;
diff --git a/drivers/net/ethernet/ti/cpts.h b/drivers/net/ethernet/ti/cpts.h
index bb997c11ee15..32ecd1ce4d3b 100644
--- a/drivers/net/ethernet/ti/cpts.h
+++ b/drivers/net/ethernet/ti/cpts.h
@@ -94,6 +94,7 @@ struct cpts_event {
unsigned long tmo;
u32 high;
u32 low;
+ u64 timestamp;
};
struct cpts {
@@ -114,6 +115,7 @@ struct cpts {
struct cpts_event pool_data[CPTS_MAX_EVENTS];
unsigned long ov_check_period;
struct sk_buff_head txq;
+ u64 cur_timestamp;
};
void cpts_rx_timestamp(struct cpts *cpts, struct sk_buff *skb);
--
2.17.1
Add CPTS IRQ support, but do not enable it. By default, the CPTS driver
will continue working using polling mode which is required for CPTS to
continue working on platforms other than CPSW, like Keystone 2.
The CPTS IRQ support is required to enable support for HW_TS_PUSH events.
The CPSW CPTS IRQ and HW_TS_PUSH events support will be enabled in follow
up patches.
Signed-off-by: Grygorii Strashko <[email protected]>
Acked-by: Richard Cochran <[email protected]>
---
drivers/net/ethernet/ti/cpts.c | 23 ++++++++++++++++++++++-
drivers/net/ethernet/ti/cpts.h | 16 ++++++++++++++++
2 files changed, 38 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/ti/cpts.c b/drivers/net/ethernet/ti/cpts.c
index 8db9efdf1708..339796c87bf6 100644
--- a/drivers/net/ethernet/ti/cpts.c
+++ b/drivers/net/ethernet/ti/cpts.c
@@ -99,6 +99,7 @@ static void cpts_purge_txq(struct cpts *cpts)
*/
static int cpts_fifo_read(struct cpts *cpts, int match)
{
+ bool need_schedule = false;
struct cpts_event *event;
unsigned long flags;
int i, type = -1;
@@ -131,6 +132,8 @@ static int cpts_fifo_read(struct cpts *cpts, int match)
cpts->cc.mult = cpts->mult_new;
cpts->mult_new = 0;
}
+ if (!cpts->irq_poll)
+ complete(&cpts->ts_push_complete);
break;
case CPTS_EV_TX:
case CPTS_EV_RX:
@@ -139,6 +142,7 @@ static int cpts_fifo_read(struct cpts *cpts, int match)
list_del_init(&event->list);
list_add_tail(&event->list, &cpts->events);
+ need_schedule = true;
break;
case CPTS_EV_ROLL:
case CPTS_EV_HALF:
@@ -154,9 +158,18 @@ static int cpts_fifo_read(struct cpts *cpts, int match)
spin_unlock_irqrestore(&cpts->lock, flags);
+ if (!cpts->irq_poll && need_schedule)
+ ptp_schedule_worker(cpts->clock, 0);
+
return type == match ? 0 : -1;
}
+void cpts_misc_interrupt(struct cpts *cpts)
+{
+ cpts_fifo_read(cpts, -1);
+}
+EXPORT_SYMBOL_GPL(cpts_misc_interrupt);
+
static u64 cpts_systim_read(const struct cyclecounter *cc)
{
struct cpts *cpts = container_of(cc, struct cpts, cc);
@@ -169,6 +182,8 @@ static void cpts_update_cur_time(struct cpts *cpts, int match,
{
unsigned long flags;
+ reinit_completion(&cpts->ts_push_complete);
+
/* use spin_lock_irqsave() here as it has to run very fast */
spin_lock_irqsave(&cpts->lock, flags);
ptp_read_system_prets(sts);
@@ -177,8 +192,12 @@ static void cpts_update_cur_time(struct cpts *cpts, int match,
ptp_read_system_postts(sts);
spin_unlock_irqrestore(&cpts->lock, flags);
- if (cpts_fifo_read(cpts, match) && match != -1)
+ if (cpts->irq_poll && cpts_fifo_read(cpts, match) && match != -1)
dev_err(cpts->dev, "cpts: unable to obtain a time stamp\n");
+
+ if (!cpts->irq_poll &&
+ !wait_for_completion_timeout(&cpts->ts_push_complete, HZ))
+ dev_err(cpts->dev, "cpts: obtain a time stamp timeout\n");
}
/* PTP clock operations */
@@ -708,8 +727,10 @@ struct cpts *cpts_create(struct device *dev, void __iomem *regs,
cpts->dev = dev;
cpts->reg = (struct cpsw_cpts __iomem *)regs;
+ cpts->irq_poll = true;
spin_lock_init(&cpts->lock);
mutex_init(&cpts->ptp_clk_mutex);
+ init_completion(&cpts->ts_push_complete);
ret = cpts_of_parse(cpts, node);
if (ret)
diff --git a/drivers/net/ethernet/ti/cpts.h b/drivers/net/ethernet/ti/cpts.h
index f16e14d67f5f..473d0622e861 100644
--- a/drivers/net/ethernet/ti/cpts.h
+++ b/drivers/net/ethernet/ti/cpts.h
@@ -118,6 +118,8 @@ struct cpts {
u64 cur_timestamp;
u32 mult_new;
struct mutex ptp_clk_mutex; /* sync PTP interface and worker */
+ bool irq_poll;
+ struct completion ts_push_complete;
};
void cpts_rx_timestamp(struct cpts *cpts, struct sk_buff *skb);
@@ -127,6 +129,7 @@ void cpts_unregister(struct cpts *cpts);
struct cpts *cpts_create(struct device *dev, void __iomem *regs,
struct device_node *node);
void cpts_release(struct cpts *cpts);
+void cpts_misc_interrupt(struct cpts *cpts);
static inline bool cpts_can_timestamp(struct cpts *cpts, struct sk_buff *skb)
{
@@ -138,6 +141,11 @@ static inline bool cpts_can_timestamp(struct cpts *cpts, struct sk_buff *skb)
return true;
}
+static inline void cpts_set_irqpoll(struct cpts *cpts, bool en)
+{
+ cpts->irq_poll = en;
+}
+
#else
struct cpts;
@@ -173,6 +181,14 @@ static inline bool cpts_can_timestamp(struct cpts *cpts, struct sk_buff *skb)
{
return false;
}
+
+static inline void cpts_misc_interrupt(struct cpts *cpts)
+{
+}
+
+static inline void cpts_set_irqpoll(struct cpts *cpts, bool en)
+{
+}
#endif
--
2.17.1
Now the tx timestamp processing happens from different contexts - softirq
and thread/PTP worker. Enabling IRQ will add one more hard_irq context.
This makes over all defered TX timestamp processing and locking
overcomplicated. Move tx timestamp processing to PTP worker always instead.
napi_rx->cpts_tx_timestamp
if ptp_packet then
push to txq
ptp_schedule_worker()
do_aux_work->cpts_overflow_check
cpts_process_events()
Signed-off-by: Grygorii Strashko <[email protected]>
Acked-by: Richard Cochran <[email protected]>
---
drivers/net/ethernet/ti/cpts.c | 165 +++++++++++++++++++--------------
1 file changed, 94 insertions(+), 71 deletions(-)
diff --git a/drivers/net/ethernet/ti/cpts.c b/drivers/net/ethernet/ti/cpts.c
index 6efb809d58ed..55ba6b425fb5 100644
--- a/drivers/net/ethernet/ti/cpts.c
+++ b/drivers/net/ethernet/ti/cpts.c
@@ -21,6 +21,8 @@
#include "cpts.h"
#define CPTS_SKB_TX_WORK_TIMEOUT 1 /* jiffies */
+#define CPTS_SKB_RX_TX_TMO 100 /*ms */
+#define CPTS_EVENT_RX_TX_TIMEOUT (100) /* ms */
struct cpts_skb_cb_data {
u32 skb_mtype_seqid;
@@ -92,46 +94,6 @@ static void cpts_purge_txq(struct cpts *cpts)
dev_dbg(cpts->dev, "txq cleaned up %d\n", removed);
}
-static bool cpts_match_tx_ts(struct cpts *cpts, struct cpts_event *event)
-{
- struct sk_buff *skb, *tmp;
- bool found = false;
- u32 mtype_seqid;
-
- mtype_seqid = event->high &
- ((MESSAGE_TYPE_MASK << MESSAGE_TYPE_SHIFT) |
- (SEQUENCE_ID_MASK << SEQUENCE_ID_SHIFT) |
- (EVENT_TYPE_MASK << EVENT_TYPE_SHIFT));
-
- /* no need to grab txq.lock as access is always done under cpts->lock */
- skb_queue_walk_safe(&cpts->txq, skb, tmp) {
- struct skb_shared_hwtstamps ssh;
- struct cpts_skb_cb_data *skb_cb =
- (struct cpts_skb_cb_data *)skb->cb;
-
- if (mtype_seqid == skb_cb->skb_mtype_seqid) {
- memset(&ssh, 0, sizeof(ssh));
- ssh.hwtstamp = ns_to_ktime(event->timestamp);
- skb_tstamp_tx(skb, &ssh);
- found = true;
- __skb_unlink(skb, &cpts->txq);
- dev_consume_skb_any(skb);
- dev_dbg(cpts->dev, "match tx timestamp mtype_seqid %08x\n",
- mtype_seqid);
- break;
- }
-
- if (time_after(jiffies, skb_cb->tmo)) {
- /* timeout any expired skbs over 1s */
- dev_dbg(cpts->dev, "expiring tx timestamp from txq\n");
- __skb_unlink(skb, &cpts->txq);
- dev_consume_skb_any(skb);
- }
- }
-
- return found;
-}
-
/*
* Returns zero if matching event type was found.
*/
@@ -151,7 +113,6 @@ static int cpts_fifo_read(struct cpts *cpts, int match)
}
event = list_first_entry(&cpts->pool, struct cpts_event, list);
- event->tmo = jiffies + 2;
event->high = hi;
event->low = lo;
event->timestamp = timecounter_cyc2time(&cpts->tc, event->low);
@@ -169,14 +130,10 @@ static int cpts_fifo_read(struct cpts *cpts, int match)
}
break;
case CPTS_EV_TX:
- if (cpts_match_tx_ts(cpts, event)) {
- /* if the new event matches an existing skb,
- * then don't queue it
- */
- break;
- }
- /* fall through */
case CPTS_EV_RX:
+ event->tmo = jiffies +
+ msecs_to_jiffies(CPTS_EVENT_RX_TX_TIMEOUT);
+
list_del_init(&event->list);
list_add_tail(&event->list, &cpts->events);
break;
@@ -297,6 +254,84 @@ static int cpts_ptp_enable(struct ptp_clock_info *ptp,
return -EOPNOTSUPP;
}
+static bool cpts_match_tx_ts(struct cpts *cpts, struct cpts_event *event)
+{
+ struct sk_buff_head txq_list;
+ struct sk_buff *skb, *tmp;
+ unsigned long flags;
+ bool found = false;
+ u32 mtype_seqid;
+
+ mtype_seqid = event->high &
+ ((MESSAGE_TYPE_MASK << MESSAGE_TYPE_SHIFT) |
+ (SEQUENCE_ID_MASK << SEQUENCE_ID_SHIFT) |
+ (EVENT_TYPE_MASK << EVENT_TYPE_SHIFT));
+
+ __skb_queue_head_init(&txq_list);
+
+ spin_lock_irqsave(&cpts->txq.lock, flags);
+ skb_queue_splice_init(&cpts->txq, &txq_list);
+ spin_unlock_irqrestore(&cpts->txq.lock, flags);
+
+ skb_queue_walk_safe(&txq_list, skb, tmp) {
+ struct skb_shared_hwtstamps ssh;
+ struct cpts_skb_cb_data *skb_cb =
+ (struct cpts_skb_cb_data *)skb->cb;
+
+ if (mtype_seqid == skb_cb->skb_mtype_seqid) {
+ memset(&ssh, 0, sizeof(ssh));
+ ssh.hwtstamp = ns_to_ktime(event->timestamp);
+ skb_tstamp_tx(skb, &ssh);
+ found = true;
+ __skb_unlink(skb, &txq_list);
+ dev_consume_skb_any(skb);
+ dev_dbg(cpts->dev, "match tx timestamp mtype_seqid %08x\n",
+ mtype_seqid);
+ break;
+ }
+
+ if (time_after(jiffies, skb_cb->tmo)) {
+ /* timeout any expired skbs over 1s */
+ dev_dbg(cpts->dev, "expiring tx timestamp from txq\n");
+ __skb_unlink(skb, &txq_list);
+ dev_consume_skb_any(skb);
+ }
+ }
+
+ spin_lock_irqsave(&cpts->txq.lock, flags);
+ skb_queue_splice(&txq_list, &cpts->txq);
+ spin_unlock_irqrestore(&cpts->txq.lock, flags);
+
+ return found;
+}
+
+static void cpts_process_events(struct cpts *cpts)
+{
+ struct list_head *this, *next;
+ struct cpts_event *event;
+ LIST_HEAD(events_free);
+ unsigned long flags;
+ LIST_HEAD(events);
+
+ spin_lock_irqsave(&cpts->lock, flags);
+ list_splice_init(&cpts->events, &events);
+ spin_unlock_irqrestore(&cpts->lock, flags);
+
+ list_for_each_safe(this, next, &events) {
+ event = list_entry(this, struct cpts_event, list);
+ if (cpts_match_tx_ts(cpts, event) ||
+ time_after(jiffies, event->tmo)) {
+ list_del_init(&event->list);
+ list_add(&event->list, &events_free);
+ }
+ }
+
+ spin_lock_irqsave(&cpts->lock, flags);
+ list_splice_tail(&events, &cpts->events);
+ list_splice_tail(&events_free, &cpts->pool);
+ spin_unlock_irqrestore(&cpts->lock, flags);
+}
+
static long cpts_overflow_check(struct ptp_clock_info *ptp)
{
struct cpts *cpts = container_of(ptp, struct cpts, info);
@@ -305,17 +340,20 @@ static long cpts_overflow_check(struct ptp_clock_info *ptp)
u64 ns;
spin_lock_irqsave(&cpts->lock, flags);
-
cpts_update_cur_time(cpts, -1, NULL);
+ spin_unlock_irqrestore(&cpts->lock, flags);
ns = timecounter_read(&cpts->tc);
+ cpts_process_events(cpts);
+
+ spin_lock_irqsave(&cpts->txq.lock, flags);
if (!skb_queue_empty(&cpts->txq)) {
cpts_purge_txq(cpts);
if (!skb_queue_empty(&cpts->txq))
delay = CPTS_SKB_TX_WORK_TIMEOUT;
}
- spin_unlock_irqrestore(&cpts->lock, flags);
+ spin_unlock_irqrestore(&cpts->txq.lock, flags);
dev_dbg(cpts->dev, "cpts overflow check at %lld\n", ns);
return (long)delay;
@@ -409,19 +447,6 @@ static u64 cpts_find_ts(struct cpts *cpts, struct sk_buff *skb,
break;
}
}
-
- if (ev_type == CPTS_EV_TX && !ns) {
- struct cpts_skb_cb_data *skb_cb =
- (struct cpts_skb_cb_data *)skb->cb;
- /* Not found, add frame to queue for processing later.
- * The periodic FIFO check will handle this.
- */
- skb_get(skb);
- /* get the timestamp for timeouts */
- skb_cb->tmo = jiffies + msecs_to_jiffies(100);
- __skb_queue_tail(&cpts->txq, skb);
- ptp_schedule_worker(cpts->clock, 0);
- }
spin_unlock_irqrestore(&cpts->lock, flags);
return ns;
@@ -455,9 +480,7 @@ EXPORT_SYMBOL_GPL(cpts_rx_timestamp);
void cpts_tx_timestamp(struct cpts *cpts, struct sk_buff *skb)
{
struct cpts_skb_cb_data *skb_cb = (struct cpts_skb_cb_data *)skb->cb;
- struct skb_shared_hwtstamps ssh;
int ret;
- u64 ns;
if (!(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
return;
@@ -471,12 +494,12 @@ void cpts_tx_timestamp(struct cpts *cpts, struct sk_buff *skb)
dev_dbg(cpts->dev, "%s mtype seqid %08x\n",
__func__, skb_cb->skb_mtype_seqid);
- ns = cpts_find_ts(cpts, skb, CPTS_EV_TX, skb_cb->skb_mtype_seqid);
- if (!ns)
- return;
- memset(&ssh, 0, sizeof(ssh));
- ssh.hwtstamp = ns_to_ktime(ns);
- skb_tstamp_tx(skb, &ssh);
+ /* Always defer TX TS processing to PTP worker */
+ skb_get(skb);
+ /* get the timestamp for timeouts */
+ skb_cb->tmo = jiffies + msecs_to_jiffies(CPTS_SKB_RX_TX_TMO);
+ skb_queue_tail(&cpts->txq, skb);
+ ptp_schedule_worker(cpts->clock, 0);
}
EXPORT_SYMBOL_GPL(cpts_tx_timestamp);
--
2.17.1
From: Grygorii Strashko <[email protected]>
Date: Thu, 23 Apr 2020 17:20:12 +0300
> This is re-spin of patches to add CPSW IRQ and HW_TS_PUSH events support I've
> sent long time ago [1]. In this series, I've tried to restructure and split changes,
> and also add few additional optimizations comparing to initial RFC submission [1].
...
> Changes in v5:
> - fixed build issue
...
Series applied, thanks.