[net-next,0/5] net: add and use dev_get_tstats64

Message

Heiner Kallweit Nov. 1, 2020, 12:33 p.m. UTC

It's a frequent pattern to use netdev->stats for the less frequently
accessed counters and per-cpu counters for the frequently accessed
counters (rx/tx bytes/packets). Add a default ndo_get_stats64()
implementation for this use case. Subsequently switch more drivers
to use this pattern.

Heiner Kallweit (5):
  net: core: add dev_get_tstats64 as a ndo_get_stats64 implementation
  net: make ip_tunnel_get_stats64 an alias for dev_get_tstats64
  ip6_tunnel: use ip_tunnel_get_stats64 as ndo_get_stats64 callback
  net: dsa: use net core stats64 handling
  tun: switch to net core provided statistics counters

 drivers/net/tun.c         | 127 ++++++++++----------------------------
 include/linux/netdevice.h |   1 +
 include/net/ip_tunnels.h  |   4 +-
 net/core/dev.c            |  15 +++++
 net/dsa/dsa.c             |   7 +--
 net/dsa/dsa_priv.h        |   2 -
 net/dsa/slave.c           |  29 +++------
 net/ipv4/ip_tunnel_core.c |   9 ---
 net/ipv6/ip6_tunnel.c     |  32 +---------
 9 files changed, 58 insertions(+), 168 deletions(-)

Comments

Saeed Mahameed Nov. 2, 2020, 10:36 p.m. UTC | #1

On Sun, 2020-11-01 at 13:33 +0100, Heiner Kallweit wrote:
> It's a frequent pattern to use netdev->stats for the less frequently
> accessed counters and per-cpu counters for the frequently accessed
> counters (rx/tx bytes/packets). Add a default ndo_get_stats64()
> implementation for this use case. Subsequently switch more drivers
> to use this pattern.
> 
> Heiner Kallweit (5):
>   net: core: add dev_get_tstats64 as a ndo_get_stats64 implementation
>   net: make ip_tunnel_get_stats64 an alias for dev_get_tstats64
>   ip6_tunnel: use ip_tunnel_get_stats64 as ndo_get_stats64 callback
>   net: dsa: use net core stats64 handling
>   tun: switch to net core provided statistics counters
> 

not many left,

$ git grep dev_fetch_sw_netstats drivers/

drivers/infiniband/hw/hfi1/ipoib_main.c:        dev_fetch_sw_netstats(s
torage, priv->netstats);
drivers/net/macsec.c:   dev_fetch_sw_netstats(s, dev->tstats);
drivers/net/usb/qmi_wwan.c:     dev_fetch_sw_netstats(stats, priv-
>stats64);
drivers/net/usb/usbnet.c:       dev_fetch_sw_netstats(stats, dev-
>stats64);
drivers/net/wireless/quantenna/qtnfmac/core.c:  dev_fetch_sw_netstats(s
tats, vif->stats64);

Why not convert them as well ?
macsec has a different implementation, but all others can be converted.

Heiner Kallweit Nov. 3, 2020, 2:46 p.m. UTC | #2

On 02.11.2020 23:36, Saeed Mahameed wrote:
> On Sun, 2020-11-01 at 13:33 +0100, Heiner Kallweit wrote:
>> It's a frequent pattern to use netdev->stats for the less frequently
>> accessed counters and per-cpu counters for the frequently accessed
>> counters (rx/tx bytes/packets). Add a default ndo_get_stats64()
>> implementation for this use case. Subsequently switch more drivers
>> to use this pattern.
>>
>> Heiner Kallweit (5):
>>   net: core: add dev_get_tstats64 as a ndo_get_stats64 implementation
>>   net: make ip_tunnel_get_stats64 an alias for dev_get_tstats64
>>   ip6_tunnel: use ip_tunnel_get_stats64 as ndo_get_stats64 callback
>>   net: dsa: use net core stats64 handling
>>   tun: switch to net core provided statistics counters
>>
> 
> not many left,
> 
> $ git grep dev_fetch_sw_netstats drivers/
> 
> drivers/infiniband/hw/hfi1/ipoib_main.c:        dev_fetch_sw_netstats(s
> torage, priv->netstats);
> drivers/net/macsec.c:   dev_fetch_sw_netstats(s, dev->tstats);
> drivers/net/usb/qmi_wwan.c:     dev_fetch_sw_netstats(stats, priv-
>> stats64);
> drivers/net/usb/usbnet.c:       dev_fetch_sw_netstats(stats, dev-
>> stats64);
> drivers/net/wireless/quantenna/qtnfmac/core.c:  dev_fetch_sw_netstats(s
> tats, vif->stats64);
> 
> Why not convert them as well ?
> macsec has a different implementation, but all others can be converted.
> 
OK, I can do this. Then the series becomes somewhat bigger.
@Jakub: Would it be ok to apply the current series and I provide the
additionally requested migrations as follow-up series?

Jakub Kicinski Nov. 3, 2020, 5:05 p.m. UTC | #3

On Tue, 3 Nov 2020 15:46:27 +0100 Heiner Kallweit wrote:
> On 02.11.2020 23:36, Saeed Mahameed wrote:
> > On Sun, 2020-11-01 at 13:33 +0100, Heiner Kallweit wrote:  
> >> It's a frequent pattern to use netdev->stats for the less frequently
> >> accessed counters and per-cpu counters for the frequently accessed
> >> counters (rx/tx bytes/packets). Add a default ndo_get_stats64()
> >> implementation for this use case. Subsequently switch more drivers
> >> to use this pattern.
> >>
> >> Heiner Kallweit (5):
> >>   net: core: add dev_get_tstats64 as a ndo_get_stats64 implementation
> >>   net: make ip_tunnel_get_stats64 an alias for dev_get_tstats64
> >>   ip6_tunnel: use ip_tunnel_get_stats64 as ndo_get_stats64 callback
> >>   net: dsa: use net core stats64 handling
> >>   tun: switch to net core provided statistics counters
> >>  
> > 
> > not many left,
> > 
> > $ git grep dev_fetch_sw_netstats drivers/
> > 
> > drivers/infiniband/hw/hfi1/ipoib_main.c:        dev_fetch_sw_netstats(s
> > torage, priv->netstats);
> > drivers/net/macsec.c:   dev_fetch_sw_netstats(s, dev->tstats);
> > drivers/net/usb/qmi_wwan.c:     dev_fetch_sw_netstats(stats, priv-  
> >> stats64);  
> > drivers/net/usb/usbnet.c:       dev_fetch_sw_netstats(stats, dev-  
> >> stats64);  
> > drivers/net/wireless/quantenna/qtnfmac/core.c:  dev_fetch_sw_netstats(s
> > tats, vif->stats64);
> > 
> > Why not convert them as well ?
> > macsec has a different implementation, but all others can be converted.
> >   
> OK, I can do this. Then the series becomes somewhat bigger.
> @Jakub: Would it be ok to apply the current series and I provide the
> additionally requested migrations as follow-up series?

Fine by me.