[1/2] net/pcap: support software Tx nanosecond timestamp
diff mbox series

Message ID 20200523172130.2285380-1-vivien.didelot@gmail.com
State Changes Requested
Delegated to: Ferruh Yigit
Headers show
Series
  • [1/2] net/pcap: support software Tx nanosecond timestamp
Related show

Checks

Context Check Description
ci/iol-testing fail Testing issues
ci/iol-mellanox-Performance success Performance Testing PASS
ci/Intel-compilation success Compilation OK
ci/iol-nxp-Performance success Performance Testing PASS
ci/iol-intel-Performance success Performance Testing PASS
ci/checkpatch success coding style OK

Commit Message

Vivien Didelot May 23, 2020, 5:21 p.m. UTC
When capturing packets into a PCAP file, DPDK currently uses
microseconds for the timestamp. But libpcap supports interpreting
tv_usec as nanoseconds depending on the file timestamp precision.

To support this, use PCAP_TSTAMP_PRECISION_NANO when creating the
empty PCAP file as specified by PCAP_OPEN_DEAD(3PCAP) and implement
nanosecond timeval addition. This also ensures that the precision
reported by capinfos is nanoseconds (9).

Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
---
 drivers/net/pcap/rte_eth_pcap.c | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)

Comments

Stephen Hemminger May 24, 2020, 1:39 a.m. UTC | #1
On Sat, 23 May 2020 13:21:29 -0400
Vivien Didelot <vivien.didelot@gmail.com> wrote:

> When capturing packets into a PCAP file, DPDK currently uses
> microseconds for the timestamp. But libpcap supports interpreting
> tv_usec as nanoseconds depending on the file timestamp precision.
> 
> To support this, use PCAP_TSTAMP_PRECISION_NANO when creating the
> empty PCAP file as specified by PCAP_OPEN_DEAD(3PCAP) and implement
> nanosecond timeval addition. This also ensures that the precision
> reported by capinfos is nanoseconds (9).
> 
> Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
> ---
>  drivers/net/pcap/rte_eth_pcap.c | 15 ++++++++++++---
>  1 file changed, 12 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/net/pcap/rte_eth_pcap.c b/drivers/net/pcap/rte_eth_pcap.c
> index b4c79d174..68588c3d7 100644
> --- a/drivers/net/pcap/rte_eth_pcap.c
> +++ b/drivers/net/pcap/rte_eth_pcap.c
> @@ -287,6 +287,8 @@ eth_null_rx(void *queue __rte_unused,
>  	return 0;
>  }
>  
> +#define NSEC_PER_SEC	1e9
> +
>  static inline void
>  calculate_timestamp(struct timeval *ts) {
>  	uint64_t cycles;
> @@ -294,8 +296,14 @@ calculate_timestamp(struct timeval *ts) {
>  
>  	cycles = rte_get_timer_cycles() - start_cycles;
>  	cur_time.tv_sec = cycles / hz;
> -	cur_time.tv_usec = (cycles % hz) * 1e6 / hz;
> -	timeradd(&start_time, &cur_time, ts);
> +	cur_time.tv_usec = (cycles % hz) * NSEC_PER_SEC / hz;
> +
> +	ts->tv_sec = start_time.tv_sec + cur_time.tv_sec;
> +	ts->tv_usec = start_time.tv_usec + cur_time.tv_usec;
> +	if (ts->tv_usec > NSEC_PER_SEC) {
> +		ts->tv_usec -= NSEC_PER_SEC;
> +		ts->tv_sec += 1;
> +	}

Please no floating point math in the fast path of capturing packets!
Ferruh Yigit June 3, 2020, 5:50 p.m. UTC | #2
On 5/23/2020 6:21 PM, Vivien Didelot wrote:
> When capturing packets into a PCAP file, DPDK currently uses
> microseconds for the timestamp. But libpcap supports interpreting
> tv_usec as nanoseconds depending on the file timestamp precision.
> 
> To support this, use PCAP_TSTAMP_PRECISION_NANO when creating the
> empty PCAP file as specified by PCAP_OPEN_DEAD(3PCAP) and implement
> nanosecond timeval addition. This also ensures that the precision
> reported by capinfos is nanoseconds (9).

Overall good idea and patch looks good.

Only concern is change in the libpcap dependency. Do you know which libpcap
version supports 'PCAP_TSTAMP_PRECISION_NANO'?
If a user of pcap PMD updates to latest DPDK and the environment doesn't have
new version of the libpcap, this change will require an environment update and
this may or may not be easy to do. That is why not sure if the updates require
dependency change should wait for the LTS or not.

Another things is the backward compatibility, as far as I understand the pcap
file recorded with nanosecond precision can be read and parsed without problem
by old application that doesn't know the nanosecond precision, but can you
please confirm this?

Thanks,
ferruh

> 
> Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
> ---
>  drivers/net/pcap/rte_eth_pcap.c | 15 ++++++++++++---
>  1 file changed, 12 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/net/pcap/rte_eth_pcap.c b/drivers/net/pcap/rte_eth_pcap.c
> index b4c79d174..68588c3d7 100644
> --- a/drivers/net/pcap/rte_eth_pcap.c
> +++ b/drivers/net/pcap/rte_eth_pcap.c
> @@ -287,6 +287,8 @@ eth_null_rx(void *queue __rte_unused,
>  	return 0;
>  }
>  
> +#define NSEC_PER_SEC	1e9
> +
>  static inline void
>  calculate_timestamp(struct timeval *ts) {
>  	uint64_t cycles;
> @@ -294,8 +296,14 @@ calculate_timestamp(struct timeval *ts) {
>  
>  	cycles = rte_get_timer_cycles() - start_cycles;
>  	cur_time.tv_sec = cycles / hz;
> -	cur_time.tv_usec = (cycles % hz) * 1e6 / hz;
> -	timeradd(&start_time, &cur_time, ts);
> +	cur_time.tv_usec = (cycles % hz) * NSEC_PER_SEC / hz;
> +
> +	ts->tv_sec = start_time.tv_sec + cur_time.tv_sec;
> +	ts->tv_usec = start_time.tv_usec + cur_time.tv_usec;
> +	if (ts->tv_usec > NSEC_PER_SEC) {
> +		ts->tv_usec -= NSEC_PER_SEC;
> +		ts->tv_sec += 1;
> +	}
>  }
>  
>  /*
> @@ -475,7 +483,8 @@ open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
>  	 * with pcap_dump_open(). We create big enough an Ethernet
>  	 * pcap holder.
>  	 */
> -	tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN);
> +	tx_pcap = pcap_open_dead_with_tstamp_precision(DLT_EN10MB,
> +			RTE_ETH_PCAP_SNAPSHOT_LEN, PCAP_TSTAMP_PRECISION_NANO);
>  	if (tx_pcap == NULL) {
>  		PMD_LOG(ERR, "Couldn't create dead pcap");
>  		return -1;
>
Stephen Hemminger June 3, 2020, 8:11 p.m. UTC | #3
On Wed, 3 Jun 2020 18:50:51 +0100
Ferruh Yigit <ferruh.yigit@intel.com> wrote:

> On 5/23/2020 6:21 PM, Vivien Didelot wrote:
> > When capturing packets into a PCAP file, DPDK currently uses
> > microseconds for the timestamp. But libpcap supports interpreting
> > tv_usec as nanoseconds depending on the file timestamp precision.
> > 
> > To support this, use PCAP_TSTAMP_PRECISION_NANO when creating the
> > empty PCAP file as specified by PCAP_OPEN_DEAD(3PCAP) and implement
> > nanosecond timeval addition. This also ensures that the precision
> > reported by capinfos is nanoseconds (9).  
> 
> Overall good idea and patch looks good.
> 
> Only concern is change in the libpcap dependency. Do you know which libpcap
> version supports 'PCAP_TSTAMP_PRECISION_NANO'?
> If a user of pcap PMD updates to latest DPDK and the environment doesn't have
> new version of the libpcap, this change will require an environment update and
> this may or may not be easy to do. That is why not sure if the updates require
> dependency change should wait for the LTS or not.
> 
> Another things is the backward compatibility, as far as I understand the pcap
> file recorded with nanosecond precision can be read and parsed without problem
> by old application that doesn't know the nanosecond precision, but can you
> please confirm this?

We should do pcapng instead of doing the pcap with nano timestamp.
My impression is that libpcap is a dormant project at this point, and the
finer grain timestamp is a hack that is only in some verisions.
That was one of the reasons pcapng started.
Stephen Hemminger June 3, 2020, 9:59 p.m. UTC | #4
On Wed, 3 Jun 2020 16:26:52 -0400
Vivien Didelot <vivien.didelot@gmail.com> wrote:

> On Wed, 3 Jun 2020 13:11:39 -0700, Stephen Hemminger <stephen@networkplumber.org> wrote:
> > On Wed, 3 Jun 2020 18:50:51 +0100
> > Ferruh Yigit <ferruh.yigit@intel.com> wrote:
> >   
> > > On 5/23/2020 6:21 PM, Vivien Didelot wrote:  
> > > > When capturing packets into a PCAP file, DPDK currently uses
> > > > microseconds for the timestamp. But libpcap supports interpreting
> > > > tv_usec as nanoseconds depending on the file timestamp precision.
> > > > 
> > > > To support this, use PCAP_TSTAMP_PRECISION_NANO when creating the
> > > > empty PCAP file as specified by PCAP_OPEN_DEAD(3PCAP) and implement
> > > > nanosecond timeval addition. This also ensures that the precision
> > > > reported by capinfos is nanoseconds (9).    
> > > 
> > > Overall good idea and patch looks good.
> > > 
> > > Only concern is change in the libpcap dependency. Do you know which libpcap
> > > version supports 'PCAP_TSTAMP_PRECISION_NANO'?
> > > If a user of pcap PMD updates to latest DPDK and the environment doesn't have
> > > new version of the libpcap, this change will require an environment update and
> > > this may or may not be easy to do. That is why not sure if the updates require
> > > dependency change should wait for the LTS or not.
> > > 
> > > Another things is the backward compatibility, as far as I understand the pcap
> > > file recorded with nanosecond precision can be read and parsed without problem
> > > by old application that doesn't know the nanosecond precision, but can you
> > > please confirm this?  
> > 
> > We should do pcapng instead of doing the pcap with nano timestamp.
> > My impression is that libpcap is a dormant project at this point, and the
> > finer grain timestamp is a hack that is only in some verisions.
> > That was one of the reasons pcapng started.  
> 
> Reading tv_usec as nanoseconds has been supported for years in libpcap.
> 
> For instance PCAP_TSTAMP_PRECISION_NANO was introduced in ba89e4a18e8b
> ("Make timestamps precision configurable") on May 17, 2013.
> 
> BTW, I have no clue what you mean by "floating point math".

It was a concern that 1e9 constant in C is interpreted as floating point.
But probably a non-issue. Have gotten burned in the past by floating point
creeping into DPDK (in Qos code).
Ferruh Yigit June 4, 2020, 10:52 a.m. UTC | #5
On 6/3/2020 9:26 PM, Vivien Didelot wrote:
> On Wed, 3 Jun 2020 13:11:39 -0700, Stephen Hemminger <stephen@networkplumber.org> wrote:
>> On Wed, 3 Jun 2020 18:50:51 +0100
>> Ferruh Yigit <ferruh.yigit@intel.com> wrote:
>>
>>> On 5/23/2020 6:21 PM, Vivien Didelot wrote:
>>>> When capturing packets into a PCAP file, DPDK currently uses
>>>> microseconds for the timestamp. But libpcap supports interpreting
>>>> tv_usec as nanoseconds depending on the file timestamp precision.
>>>>
>>>> To support this, use PCAP_TSTAMP_PRECISION_NANO when creating the
>>>> empty PCAP file as specified by PCAP_OPEN_DEAD(3PCAP) and implement
>>>> nanosecond timeval addition. This also ensures that the precision
>>>> reported by capinfos is nanoseconds (9).  
>>>
>>> Overall good idea and patch looks good.
>>>
>>> Only concern is change in the libpcap dependency. Do you know which libpcap
>>> version supports 'PCAP_TSTAMP_PRECISION_NANO'?
>>> If a user of pcap PMD updates to latest DPDK and the environment doesn't have
>>> new version of the libpcap, this change will require an environment update and
>>> this may or may not be easy to do. That is why not sure if the updates require
>>> dependency change should wait for the LTS or not.
>>>
>>> Another things is the backward compatibility, as far as I understand the pcap
>>> file recorded with nanosecond precision can be read and parsed without problem
>>> by old application that doesn't know the nanosecond precision, but can you
>>> please confirm this?
>>
>> We should do pcapng instead of doing the pcap with nano timestamp.
>> My impression is that libpcap is a dormant project at this point, and the
>> finer grain timestamp is a hack that is only in some verisions.
>> That was one of the reasons pcapng started.
> 
> Reading tv_usec as nanoseconds has been supported for years in libpcap.
> 
> For instance PCAP_TSTAMP_PRECISION_NANO was introduced in ba89e4a18e8b
> ("Make timestamps precision configurable") on May 17, 2013.

Thanks for the commit info, it looks like this support exists since
libpcap-1.5.0, and even CentOS 7 has libpcap-1.5.3-xx so looks safe to proceed.
Ferruh Yigit June 4, 2020, 10:55 a.m. UTC | #6
On 6/3/2020 9:11 PM, Stephen Hemminger wrote:
> On Wed, 3 Jun 2020 18:50:51 +0100
> Ferruh Yigit <ferruh.yigit@intel.com> wrote:
> 
>> On 5/23/2020 6:21 PM, Vivien Didelot wrote:
>>> When capturing packets into a PCAP file, DPDK currently uses
>>> microseconds for the timestamp. But libpcap supports interpreting
>>> tv_usec as nanoseconds depending on the file timestamp precision.
>>>
>>> To support this, use PCAP_TSTAMP_PRECISION_NANO when creating the
>>> empty PCAP file as specified by PCAP_OPEN_DEAD(3PCAP) and implement
>>> nanosecond timeval addition. This also ensures that the precision
>>> reported by capinfos is nanoseconds (9).  
>>
>> Overall good idea and patch looks good.
>>
>> Only concern is change in the libpcap dependency. Do you know which libpcap
>> version supports 'PCAP_TSTAMP_PRECISION_NANO'?
>> If a user of pcap PMD updates to latest DPDK and the environment doesn't have
>> new version of the libpcap, this change will require an environment update and
>> this may or may not be easy to do. That is why not sure if the updates require
>> dependency change should wait for the LTS or not.
>>
>> Another things is the backward compatibility, as far as I understand the pcap
>> file recorded with nanosecond precision can be read and parsed without problem
>> by old application that doesn't know the nanosecond precision, but can you
>> please confirm this?
> 
> We should do pcapng instead of doing the pcap with nano timestamp.
> My impression is that libpcap is a dormant project at this point, and the
> finer grain timestamp is a hack that is only in some verisions.
> That was one of the reasons pcapng started.
> 

When there is a way to get better precision timestamp from pcap, why not benefit
from it. We can discuss switching to pcapng separately I think.
Ferruh Yigit June 4, 2020, 11:16 a.m. UTC | #7
On 6/4/2020 11:52 AM, Ferruh Yigit wrote:
> On 6/3/2020 9:26 PM, Vivien Didelot wrote:
>> On Wed, 3 Jun 2020 13:11:39 -0700, Stephen Hemminger <stephen@networkplumber.org> wrote:
>>> On Wed, 3 Jun 2020 18:50:51 +0100
>>> Ferruh Yigit <ferruh.yigit@intel.com> wrote:
>>>
>>>> On 5/23/2020 6:21 PM, Vivien Didelot wrote:
>>>>> When capturing packets into a PCAP file, DPDK currently uses
>>>>> microseconds for the timestamp. But libpcap supports interpreting
>>>>> tv_usec as nanoseconds depending on the file timestamp precision.
>>>>>
>>>>> To support this, use PCAP_TSTAMP_PRECISION_NANO when creating the
>>>>> empty PCAP file as specified by PCAP_OPEN_DEAD(3PCAP) and implement
>>>>> nanosecond timeval addition. This also ensures that the precision
>>>>> reported by capinfos is nanoseconds (9).  
>>>>
>>>> Overall good idea and patch looks good.
>>>>
>>>> Only concern is change in the libpcap dependency. Do you know which libpcap
>>>> version supports 'PCAP_TSTAMP_PRECISION_NANO'?
>>>> If a user of pcap PMD updates to latest DPDK and the environment doesn't have
>>>> new version of the libpcap, this change will require an environment update and
>>>> this may or may not be easy to do. That is why not sure if the updates require
>>>> dependency change should wait for the LTS or not.
>>>>
>>>> Another things is the backward compatibility, as far as I understand the pcap
>>>> file recorded with nanosecond precision can be read and parsed without problem
>>>> by old application that doesn't know the nanosecond precision, but can you
>>>> please confirm this?
>>>
>>> We should do pcapng instead of doing the pcap with nano timestamp.
>>> My impression is that libpcap is a dormant project at this point, and the
>>> finer grain timestamp is a hack that is only in some verisions.
>>> That was one of the reasons pcapng started.
>>
>> Reading tv_usec as nanoseconds has been supported for years in libpcap.
>>
>> For instance PCAP_TSTAMP_PRECISION_NANO was introduced in ba89e4a18e8b
>> ("Make timestamps precision configurable") on May 17, 2013.
> 
> Thanks for the commit info, it looks like this support exists since
> libpcap-1.5.0, and even CentOS 7 has libpcap-1.5.3-xx so looks safe to proceed.
> 

But it may be good to mention from the change in the release notes, can you
please update the release notes (doc/guides/rel_notes/release_20_08.rst) in next
version of the patch.

Patch
diff mbox series

diff --git a/drivers/net/pcap/rte_eth_pcap.c b/drivers/net/pcap/rte_eth_pcap.c
index b4c79d174..68588c3d7 100644
--- a/drivers/net/pcap/rte_eth_pcap.c
+++ b/drivers/net/pcap/rte_eth_pcap.c
@@ -287,6 +287,8 @@  eth_null_rx(void *queue __rte_unused,
 	return 0;
 }
 
+#define NSEC_PER_SEC	1e9
+
 static inline void
 calculate_timestamp(struct timeval *ts) {
 	uint64_t cycles;
@@ -294,8 +296,14 @@  calculate_timestamp(struct timeval *ts) {
 
 	cycles = rte_get_timer_cycles() - start_cycles;
 	cur_time.tv_sec = cycles / hz;
-	cur_time.tv_usec = (cycles % hz) * 1e6 / hz;
-	timeradd(&start_time, &cur_time, ts);
+	cur_time.tv_usec = (cycles % hz) * NSEC_PER_SEC / hz;
+
+	ts->tv_sec = start_time.tv_sec + cur_time.tv_sec;
+	ts->tv_usec = start_time.tv_usec + cur_time.tv_usec;
+	if (ts->tv_usec > NSEC_PER_SEC) {
+		ts->tv_usec -= NSEC_PER_SEC;
+		ts->tv_sec += 1;
+	}
 }
 
 /*
@@ -475,7 +483,8 @@  open_single_tx_pcap(const char *pcap_filename, pcap_dumper_t **dumper)
 	 * with pcap_dump_open(). We create big enough an Ethernet
 	 * pcap holder.
 	 */
-	tx_pcap = pcap_open_dead(DLT_EN10MB, RTE_ETH_PCAP_SNAPSHOT_LEN);
+	tx_pcap = pcap_open_dead_with_tstamp_precision(DLT_EN10MB,
+			RTE_ETH_PCAP_SNAPSHOT_LEN, PCAP_TSTAMP_PRECISION_NANO);
 	if (tx_pcap == NULL) {
 		PMD_LOG(ERR, "Couldn't create dead pcap");
 		return -1;