| Message ID | 20240930152304.472767-2-edumazet@google.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | net: prepare pacing offload support | expand |
Eric Dumazet wrote: > Some network devices have the ability to offload EDT (Earliest > Departure Time) which is the model used for TCP pacing and FQ > packet scheduler. > > Some of them implement the timing wheel mechanism described in > https://saeed.github.io/files/carousel-sigcomm17.pdf > with an associated 'timing wheel horizon'. > > This patch adds dev->max_pacing_offload_horizon expressing > this timing wheel horizon in nsec units. > > This is a read-only attribute. > > Unless a driver sets it, dev->max_pacing_offload_horizon > is zero. > > Signed-off-by: Eric Dumazet <edumazet@google.com> Reviewed-by: Willem de Bruijn <willemb@google.com> > @@ -2030,6 +2034,7 @@ static const struct nla_policy ifla_policy[IFLA_MAX+1] = { > [IFLA_ALLMULTI] = { .type = NLA_REJECT }, > [IFLA_GSO_IPV4_MAX_SIZE] = { .type = NLA_U32 }, > [IFLA_GRO_IPV4_MAX_SIZE] = { .type = NLA_U32 }, > + [IFLA_MAX_PACING_OFFLOAD_HORIZON]= { .type = NLA_REJECT }, nit: checkpatch does not like the lack of whitespace before assignment in such C99 designated initializers. Probably just stylistic.
On Mon, 30 Sep 2024 15:23:03 +0000 Eric Dumazet wrote: > @@ -1867,6 +1868,9 @@ static int rtnl_fill_ifinfo(struct sk_buff *skb, > READ_ONCE(dev->tso_max_size)) || > nla_put_u32(skb, IFLA_TSO_MAX_SEGS, > READ_ONCE(dev->tso_max_segs)) || > + nla_put_u64_64bit(skb, IFLA_MAX_PACING_OFFLOAD_HORIZON, > + READ_ONCE(dev->max_pacing_offload_horizon), > + IFLA_PAD) || nla_put_uint() ? > #ifdef CONFIG_RPS > nla_put_u32(skb, IFLA_NUM_RX_QUEUES, > READ_ONCE(dev->num_rx_queues)) || > @@ -2030,6 +2034,7 @@ static const struct nla_policy ifla_policy[IFLA_MAX+1] = { > [IFLA_ALLMULTI] = { .type = NLA_REJECT }, > [IFLA_GSO_IPV4_MAX_SIZE] = { .type = NLA_U32 }, > [IFLA_GRO_IPV4_MAX_SIZE] = { .type = NLA_U32 }, > + [IFLA_MAX_PACING_OFFLOAD_HORIZON]= { .type = NLA_REJECT }, Let's do this instead ? diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c index f0a520987085..a68de5c15b46 100644 --- a/net/core/rtnetlink.c +++ b/net/core/rtnetlink.c @@ -1975,6 +1975,7 @@ static int rtnl_fill_ifinfo(struct sk_buff *skb, } static const struct nla_policy ifla_policy[IFLA_MAX+1] = { + [IFLA_UNSPEC] = { .strict_start_type = IFLA_DPLL_PIN }, [IFLA_IFNAME] = { .type = NLA_STRING, .len = IFNAMSIZ-1 }, [IFLA_ADDRESS] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN }, [IFLA_BROADCAST] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN },
On Wed, Oct 2, 2024 at 3:47 PM Jakub Kicinski <kuba@kernel.org> wrote: > > On Mon, 30 Sep 2024 15:23:03 +0000 Eric Dumazet wrote: > > @@ -1867,6 +1868,9 @@ static int rtnl_fill_ifinfo(struct sk_buff *skb, > > READ_ONCE(dev->tso_max_size)) || > > nla_put_u32(skb, IFLA_TSO_MAX_SEGS, > > READ_ONCE(dev->tso_max_segs)) || > > + nla_put_u64_64bit(skb, IFLA_MAX_PACING_OFFLOAD_HORIZON, > > + READ_ONCE(dev->max_pacing_offload_horizon), > > + IFLA_PAD) || > > nla_put_uint() ? Yes, I can do this. Some backports hassles for us with older kernels. > > > #ifdef CONFIG_RPS > > nla_put_u32(skb, IFLA_NUM_RX_QUEUES, > > READ_ONCE(dev->num_rx_queues)) || > > @@ -2030,6 +2034,7 @@ static const struct nla_policy ifla_policy[IFLA_MAX+1] = { > > [IFLA_ALLMULTI] = { .type = NLA_REJECT }, > > [IFLA_GSO_IPV4_MAX_SIZE] = { .type = NLA_U32 }, > > [IFLA_GRO_IPV4_MAX_SIZE] = { .type = NLA_U32 }, > > + [IFLA_MAX_PACING_OFFLOAD_HORIZON]= { .type = NLA_REJECT }, > > Let's do this instead ? > > diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c > index f0a520987085..a68de5c15b46 100644 > --- a/net/core/rtnetlink.c > +++ b/net/core/rtnetlink.c > @@ -1975,6 +1975,7 @@ static int rtnl_fill_ifinfo(struct sk_buff *skb, > } > > static const struct nla_policy ifla_policy[IFLA_MAX+1] = { > + [IFLA_UNSPEC] = { .strict_start_type = IFLA_DPLL_PIN }, > [IFLA_IFNAME] = { .type = NLA_STRING, .len = IFNAMSIZ-1 }, > [IFLA_ADDRESS] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN }, > [IFLA_BROADCAST] = { .type = NLA_BINARY, .len = MAX_ADDR_LEN }, +2 Thanks.
diff --git a/Documentation/networking/net_cachelines/net_device.rst b/Documentation/networking/net_cachelines/net_device.rst index 22b07c814f4a4575d255fdf472d07c549536e543..49f03cb78c6e25109af969654c86ebeb19d38e12 100644 --- a/Documentation/networking/net_cachelines/net_device.rst +++ b/Documentation/networking/net_cachelines/net_device.rst @@ -183,3 +183,4 @@ struct_devlink_port* devlink_port struct_dpll_pin* dpll_pin struct hlist_head page_pools struct dim_irq_moder* irq_moder +u64 max_pacing_offload_horizon diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h index e87b5e4883259a0723278ae3f1bee87e940af895..9eb5d9c63630e9a29a8ce2f8bc8042a520ed8398 100644 --- a/include/linux/netdevice.h +++ b/include/linux/netdevice.h @@ -2009,6 +2009,8 @@ enum netdev_reg_state { * @dpll_pin: Pointer to the SyncE source pin of a DPLL subsystem, * where the clock is recovered. * + * @max_pacing_offload_horizon: max EDT offload horizon in nsec. + * * FIXME: cleanup struct net_device such that network protocol info * moves out. */ @@ -2399,6 +2401,8 @@ struct net_device { /** @irq_moder: dim parameters used if IS_ENABLED(CONFIG_DIMLIB). */ struct dim_irq_moder *irq_moder; + u64 max_pacing_offload_horizon; + u8 priv[] ____cacheline_aligned __counted_by(priv_len); } ____cacheline_aligned; diff --git a/include/uapi/linux/if_link.h b/include/uapi/linux/if_link.h index 6dc258993b177093a77317ee5f2deab97fb04674..506ba9c80e83a5039f003c9def8b4fce41f43847 100644 --- a/include/uapi/linux/if_link.h +++ b/include/uapi/linux/if_link.h @@ -377,6 +377,7 @@ enum { IFLA_GSO_IPV4_MAX_SIZE, IFLA_GRO_IPV4_MAX_SIZE, IFLA_DPLL_PIN, + IFLA_MAX_PACING_OFFLOAD_HORIZON, __IFLA_MAX }; diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c index f0a52098708584aa27461b7ee941fa324adcaf20..898a9e0061dc9dd7b8f8691b778873ec0fe0059e 100644 --- a/net/core/rtnetlink.c +++ b/net/core/rtnetlink.c @@ -1118,6 +1118,7 @@ static noinline size_t if_nlmsg_size(const struct net_device *dev, + nla_total_size(MAX_ADDR_LEN) /* IFLA_PERM_ADDRESS */ + rtnl_devlink_port_size(dev) + rtnl_dpll_pin_size(dev) + + nla_total_size_64bit(sizeof(u64)) /* IFLA_MAX_PACING_OFFLOAD_HORIZON */ + 0; } @@ -1867,6 +1868,9 @@ static int rtnl_fill_ifinfo(struct sk_buff *skb, READ_ONCE(dev->tso_max_size)) || nla_put_u32(skb, IFLA_TSO_MAX_SEGS, READ_ONCE(dev->tso_max_segs)) || + nla_put_u64_64bit(skb, IFLA_MAX_PACING_OFFLOAD_HORIZON, + READ_ONCE(dev->max_pacing_offload_horizon), + IFLA_PAD) || #ifdef CONFIG_RPS nla_put_u32(skb, IFLA_NUM_RX_QUEUES, READ_ONCE(dev->num_rx_queues)) || @@ -2030,6 +2034,7 @@ static const struct nla_policy ifla_policy[IFLA_MAX+1] = { [IFLA_ALLMULTI] = { .type = NLA_REJECT }, [IFLA_GSO_IPV4_MAX_SIZE] = { .type = NLA_U32 }, [IFLA_GRO_IPV4_MAX_SIZE] = { .type = NLA_U32 }, + [IFLA_MAX_PACING_OFFLOAD_HORIZON]= { .type = NLA_REJECT }, }; static const struct nla_policy ifla_info_policy[IFLA_INFO_MAX+1] = { diff --git a/tools/include/uapi/linux/if_link.h b/tools/include/uapi/linux/if_link.h index f0d71b2a3f1e1a3d0945bc3a0efe31cd95940f72..96ec2b01e725b304874816af171d2455bc7b495c 100644 --- a/tools/include/uapi/linux/if_link.h +++ b/tools/include/uapi/linux/if_link.h @@ -377,6 +377,7 @@ enum { IFLA_GSO_IPV4_MAX_SIZE, IFLA_GRO_IPV4_MAX_SIZE, IFLA_DPLL_PIN, + IFLA_MAX_PACING_OFFLOAD_HORIZON, __IFLA_MAX };
Some network devices have the ability to offload EDT (Earliest Departure Time) which is the model used for TCP pacing and FQ packet scheduler. Some of them implement the timing wheel mechanism described in https://saeed.github.io/files/carousel-sigcomm17.pdf with an associated 'timing wheel horizon'. This patch adds dev->max_pacing_offload_horizon expressing this timing wheel horizon in nsec units. This is a read-only attribute. Unless a driver sets it, dev->max_pacing_offload_horizon is zero. Signed-off-by: Eric Dumazet <edumazet@google.com> --- Documentation/networking/net_cachelines/net_device.rst | 1 + include/linux/netdevice.h | 4 ++++ include/uapi/linux/if_link.h | 1 + net/core/rtnetlink.c | 5 +++++ tools/include/uapi/linux/if_link.h | 1 + 5 files changed, 12 insertions(+)