ipv6 tcp and gro stacks will soon be able to build big TCP packets,
with an added temporary Hop By Hop header.
If GSO is involved for these large packets, we need to remove
the temporary HBH header before segmentation happens.
v2: perform HBH removal from ipv6_gso_segment() instead of
skb_segment() (Alexander feedback)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Following patches will need to add and remove local IPv6 jumbogram
options to enable BIG TCP.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
hystart_ack_delay() had the assumption that a TSO packet
would not be bigger than GSO_MAX_SIZE.
This will no longer be true.
We should use sk->sk_gso_max_size instead.
This reduces chances of spurious Hystart ACK train detections.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Make sure we will not overflow shinfo->gso_segs
Minimal TCP MSS size is 8 bytes, and shinfo->gso_segs
is a 16bit field.
TCP_MIN_GSO_SIZE is currently defined in include/net/tcp.h,
it seems cleaner to not bring tcp details into include/linux/netdevice.h
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The code for gso_max_size was added originally to allow for debugging and
workaround of buggy devices that couldn't support TSO with blocks 64K in
size. The original reason for limiting it to 64K was because that was the
existing limits of IPv4 and non-jumbogram IPv6 length fields.
With the addition of Big TCP we can remove this limit and allow the value
to potentially go up to UINT_MAX and instead be limited by the tso_max_size
value.
So in order to support this we need to go through and clean up the
remaining users of the gso_max_size value so that the values will cap at
64K for non-TCPv6 flows. In addition we can clean up the GSO_MAX_SIZE value
so that 64K becomes GSO_LEGACY_MAX_SIZE and UINT_MAX will now be the upper
limit for GSO_MAX_SIZE.
v6: (edumazet) fixed a compile error if CONFIG_IPV6=n,
in a new sk_trim_gso_size() helper.
netif_set_tso_max_size() caps the requested TSO size
with GSO_MAX_SIZE.
Signed-off-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
New netlink attributes IFLA_TSO_MAX_SIZE and IFLA_TSO_MAX_SEGS
are used to report to user-space the device TSO limits.
ip -d link sh dev eth1
...
tso_max_size 65536 tso_max_segs 65535
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Phil Edworthy says:
====================
Add Renesas RZ/V2M Ethernet support
The RZ/V2M Ethernet is very similar to R-Car Gen3 Ethernet-AVB, though
some small parts are the same as R-Car Gen2.
Other differences are:
* It has separate data (DI), error (Line 1) and management (Line 2) irqs
rather than one irq for all three.
* Instead of using the High-speed peripheral bus clock for gPTP, it has
a separate gPTP reference clock.
v4:
* Add clk_disable_unprepare() for gptp ref clk
v3:
* Really renamed irq_en_dis_regs to irq_en_dis this time
* Modified ravb_ptp_extts() to use irq_en_dis
* Added Reviewed-by tags
v2:
* Just net patches in this series
* Instead of reusing ch22 and ch24 interrupt names, use the proper names
* Renamed irq_en_dis_regs to irq_en_dis
* Squashed use of GIC reg versus GIE/GID and got rid of separate gptp_ptm_gic feature.
* Move err_mgmt_irqs code under multi_irqs
* Minor editing of the commit msgs
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
RZ/V2M Ethernet is very similar to R-Car Gen3 Ethernet-AVB, though
some small parts are the same as R-Car Gen2.
Other differences to R-Car Gen3 and Gen2 are:
* It has separate data (DI), error (Line 1) and management (Line 2) irqs
rather than one irq for all three.
* Instead of using the High-speed peripheral bus clock for gPTP, it has a
separate gPTP reference clock.
Signed-off-by: Phil Edworthy <phil.edworthy@renesas.com>
Reviewed-by: Biju Das <biju.das.jz@bp.renesas.com>
Reviewed-by: Sergey Shtylyov <s.shtylyov@omp.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
RZ/V2M has a separate gPTP reference clock that is used when the
AVB-DMAC Mode Register (CCC) gPTP Clock Select (CSEL) bits are
set to "01: High-speed peripheral bus clock".
Therefore, add a feature that allows this clock to be used for
gPTP.
Signed-off-by: Phil Edworthy <phil.edworthy@renesas.com>
Reviewed-by: Biju Das <biju.das.jz@bp.renesas.com>
Reviewed-by: Sergey Shtylyov <s.shtylyov@omp.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
R-Car has a combined interrupt line, ch22 = Line0_DiA | Line1_A | Line2_A.
RZ/V2M has separate interrupt lines for each of these, so add a feature
that allows the driver to get these interrupts and call the common handler.
Signed-off-by: Phil Edworthy <phil.edworthy@renesas.com>
Reviewed-by: Biju Das <biju.das.jz@bp.renesas.com>
Reviewed-by: Sergey Shtylyov <s.shtylyov@omp.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, when the HW has a single interrupt, the driver uses the
GIC, TIC, RIC0 registers to enable and disable interrupts.
When the HW has multiple interrupts, it uses the GIE, GID, TIE, TID,
RIE0, RID0 registers.
However, other devices, e.g. RZ/V2M, have multiple irqs and only have
the GIC, TIC, RIC0 registers.
Therefore, split this into a separate feature.
Signed-off-by: Phil Edworthy <phil.edworthy@renesas.com>
Reviewed-by: Biju Das <biju.das.jz@bp.renesas.com>
Reviewed-by: Sergey Shtylyov <s.shtylyov@omp.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
Document the Ethernet AVB IP found on RZ/V2M SoC.
It includes the Ethernet controller (E-MAC) and Dedicated Direct memory
access controller (DMAC) for transferring transmitted Ethernet frames
to and received Ethernet frames from respective storage areas in the
RAM at high speed.
The AVB-DMAC is compliant with IEEE 802.1BA, IEEE 802.1AS timing and
synchronization protocol, IEEE 802.1Qav real-time transfer, and the
IEEE 802.1Qat stream reservation protocol.
R-Car has a pair of combined interrupt lines:
ch22 = Line0_DiA | Line1_A | Line2_A
ch23 = Line0_DiB | Line1_B | Line2_B
Line0 for descriptor interrupts (which we call dia and dib).
Line1 for error related interrupts (which we call err_a and err_b).
Line2 for management and gPTP related interrupts (mgmt_a and mgmt_b).
RZ/V2M hardware has separate interrupt lines for each of these.
It has 3 clocks; the main AXI clock, the AMBA CHI (Coherent Hub
Interface) clock and a gPTP reference clock.
Signed-off-by: Phil Edworthy <phil.edworthy@renesas.com>
Reviewed-by: Biju Das <biju.das.jz@bp.renesas.com>
Reviewed-by: Sergey Shtylyov <s.shtylyov@omp.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pablo Neira Ayuso says:
====================
Netfilter updates for net-next
This is v2 including deadlock fix in conntrack ecache rework
reported by Jakub Kicinski.
The following patchset contains Netfilter updates for net-next,
mostly updates to conntrack from Florian Westphal.
1) Add a dedicated list for conntrack event redelivery.
2) Include event redelivery list in conntrack dumps of dying type.
3) Remove per-cpu dying list for event redelivery, not used anymore.
4) Add netns .pre_exit to cttimeout to zap timeout objects before
synchronize_rcu() call.
5) Remove nf_ct_unconfirmed_destroy.
6) Add generation id for conntrack extensions for conntrack
timeout and helpers.
7) Detach timeout policy from conntrack on cttimeout module removal.
8) Remove __nf_ct_unconfirmed_destroy.
9) Remove unconfirmed list.
10) Remove unconditional local_bh_disable in init_conntrack().
11) Consolidate conntrack iterator nf_ct_iterate_cleanup().
12) Detect if ctnetlink listeners exist to short-circuit event
path early.
13) Un-inline nf_ct_ecache_ext_add().
14) Add nf_conntrack_events autodetect ctnetlink listener mode
and make it default.
15) Add nf_ct_ecache_exist() to check for event cache extension.
16) Extend flowtable reverse route lookup to include source, iif,
tos and mark, from Sven Auhagen.
17) Do not verify zero checksum UDP packets in nf_reject,
from Kevin Mitchell.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
When ADQ queue groups (TCs) are created via tc mqprio command,
RSS contexts and associated RSS indirection tables are configured
automatically per TC based on the queue ranges specified for
each traffic class.
For ex:
tc qdisc add dev enp175s0f0 root mqprio num_tc 3 map 0 1 2 \
queues 2@0 8@2 4@10 hw 1 mode channel
will create 3 queue groups (TC 0-2) with queue ranges 2, 8 and 4
in 3 queue groups. Each queue group is associated with its
own RSS context and RSS indirection table.
Add support to expose RSS indirection tables for all ADQ queue
groups using ethtool RSS contexts interface.
ethtool -x enp175s0f0 context <tc-num>
Signed-off-by: Sridhar Samudrala <sridhar.samudrala@intel.com>
Signed-off-by: Sudheer Mogilappagari <sudheer.mogilappagari@intel.com>
Tested-by: Bharathi Sreenivas <bharathi.sreenivas@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Link: https://lore.kernel.org/r/20220512213249.3747424-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add the capability to map non-linear xdp frames in XDP_TX and ndo_xdp_xmit
callback.
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Tested-by: Sandeep Penigalapati <sandeep.penigalapati@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Link: https://lore.kernel.org/r/20220512212621.3746140-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Remove napi_weight statics which are set to 64 and never modified,
remnants of the out-of-tree napi_weight module param.
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Link: https://lore.kernel.org/r/20220512205603.1536771-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Steffen Klassert says:
====================
pull request (net-next): ipsec-next 2022-05-13
1) Cleanups for the code behind the XFRM offload API. This is a
preparation for the extension of the API for policy offload.
From Leon Romanovsky.
* 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/klassert/ipsec-next:
xfrm: drop not needed flags variable in XFRM offload struct
net/mlx5e: Use XFRM state direction instead of flags
netdevsim: rely on XFRM state direction instead of flags
ixgbe: propagate XFRM offload state direction instead of flags
xfrm: store and rely on direction to construct offload flags
xfrm: rename xfrm_state_offload struct to allow reuse
xfrm: delete not used number of external headers
xfrm: free not used XFRM_ESP_NO_TRAILER flag
====================
Link: https://lore.kernel.org/r/20220513151218.4010119-1-steffen.klassert@secunet.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
If CONFIG_PTP_1588_CLOCK=m and CONFIG_SFC_SIENA=y, the siena driver will fail to link:
drivers/net/ethernet/sfc/siena/ptp.o: In function `efx_ptp_remove_channel':
ptp.c:(.text+0xa28): undefined reference to `ptp_clock_unregister'
drivers/net/ethernet/sfc/siena/ptp.o: In function `efx_ptp_probe_channel':
ptp.c:(.text+0x13a0): undefined reference to `ptp_clock_register'
ptp.c:(.text+0x1470): undefined reference to `ptp_clock_unregister'
drivers/net/ethernet/sfc/siena/ptp.o: In function `efx_ptp_pps_worker':
ptp.c:(.text+0x1d29): undefined reference to `ptp_clock_event'
drivers/net/ethernet/sfc/siena/ptp.o: In function `efx_siena_ptp_get_ts_info':
ptp.c:(.text+0x301b): undefined reference to `ptp_clock_index'
To fix this build error, make SFC_SIENA depends on PTP_1588_CLOCK.
Reported-by: Hulk Robot <hulkci@huawei.com>
Fixes: d48523cb88 ("sfc: Copy shared files needed for Siena (part 2)")
Signed-off-by: Ren Zhijie <renzhijie2@huawei.com>
Acked-by: Martin Habets <habetsm.xilinx@gmail.com>
Link: https://lore.kernel.org/r/20220513012721.140871-1-renzhijie2@huawei.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The checksum is optional for UDP packets. However nf_reject would
previously require a valid checksum to elicit a response such as
ICMP_DEST_UNREACH.
Add some logic to nf_reject_verify_csum to determine if a UDP packet has
a zero checksum and should therefore not be verified.
Signed-off-by: Kevin Mitchell <kevmitch@arista.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
When creating a flow table entry, the reverse route is looked
up based on the current packet.
There can be scenarios where the user creates a custom ip rule
to route the traffic differently.
In order to support those scenarios, the lookup needs to add
more information based on the current packet.
The patch adds multiple new information to the route lookup.
Signed-off-by: Sven Auhagen <sven.auhagen@voleatech.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The pointer check usually results in a 'false positive': its likely
that the ctnetlink module is loaded but no event monitoring is enabled.
After recent change to autodetect ctnetlink usage and only allocate
the ecache extension if a listener is active, check if the extension
is present on a given conntrack.
If its not there, there is nothing to report and calls to the
notification framework can be elided.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This adds the new nf_conntrack_events=2 mode and makes it the
default.
This leverages the earlier flag in struct net to allow to avoid
the event extension as long as no event listener is active in
the namespace.
This avoids, for most cases, allocation of ct->ext area.
A followup patch will take further advantage of this by avoiding
calls down into the event framework if the extension isn't present.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Only called when new ct is allocated or the extension isn't present.
This function will be extended, place this in the conntrack module
instead of inlining.
The callers already depend on nf_conntrack module.
Return value is changed to bool, noone used the returned pointer.
Make sure that the core drops the newly allocated conntrack
if the extension is requested but can't be added.
This makes it necessary to ifdef the section, as the stub
always returns false we'd drop every new conntrack if the
the ecache extension is disabled in kconfig.
Add from data path (xt_CT, nft_ct) is unchanged.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
At this time, every new conntrack gets the 'event cache extension'
enabled for it.
This is because the 'net.netfilter.nf_conntrack_events' sysctl defaults
to 1.
Changing the default to 0 means that commands that rely on the event
notification extension, e.g. 'conntrack -E' or conntrackd, stop working.
We COULD detect if there is a listener by means of
'nfnetlink_has_listeners()' and only add the extension if this is true.
The downside is a dependency from conntrack module to nfnetlink module.
This adds a different way: inc/dec a counter whenever a ctnetlink group
is being (un)subscribed and toggle a flag in struct net.
Next patches will take advantage of this and will only add the event
extension if the flag is set.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This patch adds a structure to collect all the context data that is
passed to the cleanup iterator.
struct nf_ct_iter_data {
struct net *net;
void *data;
u32 portid;
int report;
};
There is a netns field that allows to clean up conntrack entries
specifically owned by the specified netns.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Now that the conntrack entry isn't placed on the pcpu list anymore the
bh only needs to be disabled in the 'expectation present' case.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Its not needed anymore:
A. If entry is totally new, then the rcu-protected resource
must already have been removed from global visibility before call
to nf_ct_iterate_destroy.
B. If entry was allocated before, but is not yet in the hash table
(uncofirmed case), genid gets incremented and synchronize_rcu() call
makes sure access has completed.
C. Next attempt to peek at extension area will fail for unconfirmed
conntracks, because ext->genid != genid.
D. Conntracks in the hash are iterated as before.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Increment the extid on module removal; this makes sure that even
in extreme cases any old uncofirmed entry that happened to be kept
e.g. on nfnetlink_queue list will not trip over a stale timeout
reference.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Multiple netfilter extensions store pointers to external data
in their extension area struct.
Examples:
1. Timeout policies
2. Connection tracking helpers.
No references are taken for these.
When a helper or timeout policy is removed, the conntrack table gets
traversed and affected extensions are cleared.
Conntrack entries not yet in the hashtable are referenced via a special
list, the unconfirmed list.
On removal of a policy or connection tracking helper, the unconfirmed
list gets traversed an all entries are marked as dying, this prevents
them from getting committed to the table at insertion time: core checks
for dying bit, if set, the conntrack entry gets destroyed at confirm
time.
The disadvantage is that each new conntrack has to be added to the percpu
unconfirmed list, and each insertion needs to remove it from this list.
The list is only ever needed when a policy or helper is removed -- a rare
occurrence.
Add a generation ID count: Instead of adding to the list and then
traversing that list on policy/helper removal, increment a counter
that is stored in the extension area.
For unconfirmed conntracks, the extension has the genid valid at ct
allocation time.
Removal of a helper/policy etc. increments the counter.
At confirmation time, validate that ext->genid == global_id.
If the stored number is not the same, do not allow the conntrack
insertion, just like as if a confirmed-list traversal would have flagged
the entry as dying.
After insertion, the genid is no longer relevant (conntrack entries
are now reachable via the conntrack table iterators and is set to 0.
This allows removal of the percpu unconfirmed list.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This helper tags connections not yet in the conntrack table as
dying. These nf_conn entries will be dropped instead when the
core attempts to insert them from the input or postrouting
'confirm' hook.
After the previous change, the entries get unlinked from the
list earlier, so that by the time the actual exit hook runs,
new connections no longer have a timeout policy assigned.
Its enough to walk the hashtable instead.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Make it so netns pre_exit unlinks the objects from the pernet list, so
they cannot be found anymore.
netns core issues a synchronize_rcu() before calling the exit hooks so
any the time the exit hooks run unconfirmed nf_conn entries have been
free'd or they have been committed to the hashtable.
The exit hook still tags unconfirmed entries as dying, this can
now be removed in a followup change.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Its no longer needed. Entries that need event redelivery are placed
on the new pernet dying list.
The advantage is that there is no need to take additional spinlock on
conntrack removal unless event redelivery failed or the conntrack entry
was never added to the table in the first place (confirmed bit not set).
The IPS_CONFIRMED bit now needs to be set as soon as the entry has been
unlinked from the unconfirmed list, else the destroy function may
attempt to unlink it a second time.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The new pernet dying list includes conntrack entries that await
delivery of the 'destroy' event via ctnetlink.
The old percpu dying list will be removed soon.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This disentangles event redelivery and the percpu dying list.
Because entries are now stored on a dedicated list, all
entries are in NFCT_ECACHE_DESTROY_FAIL state and all entries
still have confirmed bit set -- the reference count is at least 1.
The 'struct net' back-pointer can be removed as well.
The pcpu dying list will be removed eventually, it has no functionality.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Michael Chan says:
====================
bnxt_en: Updates for net-next
This small patchset updates the firmware interface, adds timestamping
support for all receive packets, and adds revised NVRAM package error
messages for ethtool and devlink.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of always returning -ENOPKG, decode the firmware error
code further when the HWRM_NVM_INSTALL_UPDATE firmware call fails.
Return a more suitable error code to userspace and log an error
in dmesg.
This is version 2 of the earlier patch that was reverted:
02acd39953 ("bnxt_en: parse result field when NVRAM package install fails")
In this new version, if the call is made through devlink instead of
ethtool, we'll also set the error message in extack.
Link: https://lore.kernel.org/netdev/20220307141358.4d52462e@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com/
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add driver support to enable timestamping on all RX packets
that are received by the NIC. This capability can be requested
by the applications using SIOCSHWTSTAMP ioctl with filter type
HWTSTAMP_FILTER_ALL.
Cc: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For correctness, we need to configure the packet filters for timestamping
during bnxt_open. This way they are always configured after firmware
reset or chip reset. We should not assume that the filters will always
be retained across resets.
This patch modifies the ioctl handler and always configures the PTP
filters in the bnxt_open() path.
Cc: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The main changes are timestamp support for all RX packets and new PCIe
statistics.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This driver was using the TX IRQ handler to perform all TX completion
tasks. Under heavy TX network load, this can cause significant irqs-off
latencies (found to be in the hundreds of microseconds using ftrace).
This can cause other issues, such as overrunning serial UART FIFOs when
using high baud rates with limited UART FIFO sizes.
Switch to using a NAPI poll handler to perform the TX completion work
to get this out of hard IRQ context and avoid the IRQ latency impact.
A separate poll handler is used for TX and RX since they have separate
IRQs on this controller, so that the completion work for each of them
stays on the same CPU as the interrupt.
Testing on a Xilinx MPSoC ZU9EG platform using iperf3 from a Linux PC
through a switch at 1G link speed showed no significant change in TX or
RX throughput, with approximately 941 Mbps before and after. Hard IRQ
time in the TX throughput test was significantly reduced from 12% to
below 1% on the CPU handling TX interrupts, with total hard+soft IRQ CPU
usage dropping from about 56% down to 48%.
Signed-off-by: Robert Hancock <robert.hancock@calian.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The axienet_start_xmit function was updating the tx_bd_tail variable
multiple times, with potential rollbacks on error or invalid
intermediate positions, even though this variable is also used in the
TX completion path. Use READ_ONCE where this variable is read and
WRITE_ONCE where it is written to make this update more atomic, and
move the write before the MMIO write to start the transfer, so it is
protected by that implicit write barrier.
Signed-off-by: Robert Hancock <robert.hancock@calian.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
INET_MATCH() runs without holding a lock on the socket.
We probably need to annotate most reads.
This patch makes INET_MATCH() an inline function
to ease our changes.
v2:
We remove the 32bit version of it, as modern compilers
should generate the same code really, no need to
try to be smarter.
Also make 'struct net *net' the first argument.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Rarely some of the test cases fail. Make the test more robust by increasing
the timeout of ping commands to 5 seconds.
Signed-off-by: Amit Cohen <amcohen@nvidia.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Lukas Wunner says:
====================
Polling be gone on LAN95xx
Do away with link status polling on LAN95xx USB Ethernet
and rely on interrupts instead, thereby reducing bus traffic,
CPU overhead and improving interface bringup latency.
Link to v2:
https://lore.kernel.org/netdev/cover.1651574194.git.lukas@wunner.de/
Only change since v2:
* Patch [5/7]:
* Drop call to __irq_enter_raw() which worked around a warning in
generic_handle_domain_irq(). That warning is gone since
792ea6a074 (queued on tip.git/irq/urgent).
(Marc Zyngier, Thomas Gleixner)
====================
If reading the Interrupt Source Flag register fails with -ENODEV, then
the PHY has been hot-removed and the correct response is to bail out
instead of throwing a WARN splat and attempting to suspend the PHY.
The PHY should be stopped in due course anyway as the kernel
asynchronously tears down the device.
Tested-by: Oleksij Rempel <o.rempel@pengutronix.de> # LAN9514/9512/9500
Tested-by: Ferry Toth <fntoth@gmail.com> # LAN9514
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Cache the interrupt mask to avoid re-reading it from the PHY upon every
interrupt.
This will simplify a subsequent commit which detects hot-removal in the
interrupt handler and bails out.
Analyzing and debugging PHY transactions also becomes simpler if such
redundant reads are avoided.
Last not least, interrupt overhead and latency is slightly improved.
Tested-by: Oleksij Rempel <o.rempel@pengutronix.de> # LAN9514/9512/9500
Tested-by: Ferry Toth <fntoth@gmail.com> # LAN9514
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Link status of SMSC LAN95xx chips is polled once per second, even though
they're capable of signaling PHY interrupts through the MAC layer.
Forward those interrupts to the PHY driver to avoid polling. Benefits
are reduced bus traffic, reduced CPU overhead and quicker interface
bringup.
Polling was introduced in 2016 by commit d69d169493 ("usbnet:
smsc95xx: fix link detection for disabled autonegotiation").
Back then, the LAN95xx driver neglected to enable the ENERGYON interrupt,
hence couldn't detect link-up events when auto-negotiation was disabled.
The proper solution would have been to enable the ENERGYON interrupt
instead of polling.
Since then, PHY handling was moved from the LAN95xx driver to the SMSC
PHY driver with commit 05b35e7eb9 ("smsc95xx: add phylib support").
That PHY driver is capable of link detection with auto-negotiation
disabled because it enables the ENERGYON interrupt.
Note that signaling interrupts through the MAC layer not only works with
the integrated PHY, but also with an external PHY, provided its
interrupt pin is attached to LAN95xx's nPHY_INT pin.
In the unlikely event that the interrupt pin of an external PHY is
attached to a GPIO of the SoC (or not connected at all), the driver can
be amended to retrieve the irq from the PHY's of_node.
To forward PHY interrupts to phylib, it is not sufficient to call
phy_mac_interrupt(). Instead, the PHY's interrupt handler needs to run
so that PHY interrupts are cleared. That's because according to page
119 of the LAN950x datasheet, "The source of this interrupt is a level.
The interrupt persists until it is cleared in the PHY."
https://www.microchip.com/content/dam/mchp/documents/UNG/ProductDocuments/DataSheets/LAN950x-Data-Sheet-DS00001875D.pdf
Therefore, create an IRQ domain with a single IRQ for the PHY. In the
future, the IRQ domain may be extended to support the 11 GPIOs on the
LAN95xx.
Normally the PHY interrupt should be masked until the PHY driver has
cleared it. However masking requires a (sleeping) USB transaction and
interrupts are received in (non-sleepable) softirq context. I decided
not to mask the interrupt at all (by using the dummy_irq_chip's noop
->irq_mask() callback): The USB interrupt endpoint is polled in 1 msec
intervals and normally that's sufficient to wake the PHY driver's IRQ
thread and have it clear the interrupt. If it does take longer, worst
thing that can happen is the IRQ thread is woken again. No big deal.
Because PHY interrupts are now perpetually enabled, there's no need to
selectively enable them on suspend. So remove all invocations of
smsc95xx_enable_phy_wakeup_interrupts().
In smsc95xx_resume(), move the call of phy_init_hw() before
usbnet_resume() (which restarts the status URB) to ensure that the PHY
is fully initialized when an interrupt is handled.
Tested-by: Oleksij Rempel <o.rempel@pengutronix.de> # LAN9514/9512/9500
Tested-by: Ferry Toth <fntoth@gmail.com> # LAN9514
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch> # from a PHY perspective
Cc: Andre Edich <andre.edich@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When a PHY interrupt is signaled, the SMSC LAN95xx driver updates the
MAC full duplex mode and PHY flow control registers based on cached data
in struct phy_device:
smsc95xx_status() # raises EVENT_LINK_RESET
usbnet_deferred_kevent()
smsc95xx_link_reset() # uses cached data in phydev
Simultaneously, phylib polls link status once per second and updates
that cached data:
phy_state_machine()
phy_check_link_status()
phy_read_status()
lan87xx_read_status()
genphy_read_status() # updates cached data in phydev
If smsc95xx_link_reset() wins the race against genphy_read_status(),
the registers may be updated based on stale data.
E.g. if the link was previously down, phydev->duplex is set to
DUPLEX_UNKNOWN and that's what smsc95xx_link_reset() will use, even
though genphy_read_status() may update it to DUPLEX_FULL afterwards.
PHY interrupts are currently only enabled on suspend to trigger wakeup,
so the impact of the race is limited, but we're about to enable them
perpetually.
Avoid the race by delaying execution of smsc95xx_link_reset() until
phy_state_machine() has done its job and calls back via
smsc95xx_handle_link_change().
Signaling EVENT_LINK_RESET on wakeup is not necessary because phylib
picks up link status changes through polling. So drop the declaration
of a ->link_reset() callback.
Note that the semicolon on a line by itself added in smsc95xx_status()
is a placeholder for a function call which will be added in a subsequent
commit. That function call will actually handle the INT_ENP_PHY_INT_
interrupt.
Tested-by: Oleksij Rempel <o.rempel@pengutronix.de> # LAN9514/9512/9500
Tested-by: Ferry Toth <fntoth@gmail.com> # LAN9514
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
smsc95xx_reset() resets the PHY behind the PHY driver's back, which
seems like a bad idea generally. Remove that portion of the function.
We're about to use PHY interrupts instead of polling to detect link
changes on SMSC LAN95xx chips. Because smsc95xx_reset() is called from
usbnet_open(), PHY interrupt settings are lost whenever the net_device
is brought up.
There are two other callers of smsc95xx_reset(), namely smsc95xx_bind()
and smsc95xx_reset_resume(), and both may indeed benefit from a PHY
reset. However they already perform one through their calls to
phy_connect_direct() and phy_init_hw().
Tested-by: Oleksij Rempel <o.rempel@pengutronix.de> # LAN9514/9512/9500
Tested-by: Ferry Toth <fntoth@gmail.com> # LAN9514
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Cc: Martyn Welch <martyn.welch@collabora.com>
Cc: Gabriel Hojda <ghojda@yo2urs.ro>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
