GCC complains:
net/l2tp/l2tp_ppp.c: In function ‘pppol2tp_ioctl’:
net/l2tp/l2tp_ppp.c:1073:6: warning: variable ‘val’ set but not used [-Wunused-but-set-variable]
int val;
^~~
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit da70314917 ("bpf: Document BPF_PROG_TYPE_CGROUP_SYSCTL")
Andrey proposes to put per-prog type docs under Documentation/bpf/
Let's move flow dissector documentation there as well.
Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
To make ICMPv6 closer to ICMPv4, add ratemask parameter. Since the ICMP
message types use larger numeric values, a simple bitmask doesn't fit.
I use large bitmap. The input and output are the in form of list of
ranges. Set the default to rate limit all error messages but Packet Too
Big. For Packet Too Big, use ratemask instead of hard-coded.
There are functions where icmpv6_xrlim_allow() and icmpv6_global_allow()
aren't called. This patch only adds them to icmpv6_echo_reply().
Rate limiting error messages is mandated by RFC 4443 but RFC 4890 says
that it is also acceptable to rate limit informational messages. Thus,
I removed the current hard-coded behavior of icmpv6_mask_allow() that
doesn't rate limit informational messages.
v2: Add dummy function proc_do_large_bitmap() if CONFIG_PROC_SYSCTL
isn't defined, expand the description in ip-sysctl.txt and remove
unnecessary conditional before kfree().
v3: Inline the bitmap instead of dynamically allocated. Still is a
pointer to it is needed because of the way proc_do_large_bitmap work.
Signed-off-by: Stephen Suryaputra <ssuryaextr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
phy_sanitize_settings() is called from phy_start_aneg() only, and only
if phydev->autoneg isn't set. Therefore the removed code does nothing.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
map_fds[16] is the last one index-ed by fixup_map_array_small.
Hence, the MAX_NR_MAPS should be 17 instead.
Fixes: fb2abb73e5 ("bpf, selftest: test {rd, wr}only flags and direct value access")
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
In phy_device_create() we set phydev->autoneg = 1. This isn't changed
even if the PHY doesn't support autoneg. This seems to affect very
few PHY's, and they disable phydev->autoneg in their config_init
callback. So it's more of an improvement, therefore net-next.
The patch also wouldn't apply to older kernel versions because the
link mode bitmaps have been introduced recently.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
I meet below compile errors:
"
In file included from test_tcpnotify_kern.c:12:
/usr/include/netinet/in.h:101:5: error: expected identifier
IPPROTO_HOPOPTS = 0, /* IPv6 Hop-by-Hop options. */
^
/usr/include/linux/in6.h:131:26: note: expanded from macro 'IPPROTO_HOPOPTS'
^
In file included from test_tcpnotify_kern.c:12:
/usr/include/netinet/in.h:103:5: error: expected identifier
IPPROTO_ROUTING = 43, /* IPv6 routing header. */
^
/usr/include/linux/in6.h:132:26: note: expanded from macro 'IPPROTO_ROUTING'
^
In file included from test_tcpnotify_kern.c:12:
/usr/include/netinet/in.h:105:5: error: expected identifier
IPPROTO_FRAGMENT = 44, /* IPv6 fragmentation header. */
^
/usr/include/linux/in6.h:133:26: note: expanded from macro 'IPPROTO_FRAGMENT'
"
The same compile errors are reported for test_tcpbpf_kern.c too.
My environment:
lsb_release -a:
No LSB modules are available.
Distributor ID: Ubuntu
Description: Ubuntu 16.04.6 LTS
Release: 16.04
Codename: xenial
dpkg -l | grep libc-dev:
ii libc-dev-bin 2.23-0ubuntu11 amd64 GNU C Library: Development binaries
ii linux-libc-dev:amd64 4.4.0-145.171 amd64 Linux Kernel Headers for development.
The reason is linux/in6.h and netinet/in.h aren't synchronous about how to
handle the same definitions, IPPROTO_HOPOPTS, etc.
This patch fixes the compile errors by moving <netinet/in.h> to before the
<linux/*.h>.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Having a helpful compile time warning in libbpf_util.h is not a good
idea since all warnings are treated as errors. Change this into a
comment in the code instead.
Fixes: b7e3a28019 ("libbpf: remove dependency on barrier.h in xsk.h")
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Add documentation for BPF_PROG_TYPE_CGROUP_SYSCTL, including general
info, attach type, context, return code, helpers, example and usage
considerations.
A separate file prog_cgroup_sysctl.rst is added to Documentation/bpf/.
In the future more program types can be documented in their own
prog_<name>.rst files.
Another way to place program type specific documentation would be to
group program types somehow (e.g. cgroup.rst for all cgroup-bpf
programs), but it may not scale well since some program types may belong
to different groups, e.g. BPF_PROG_TYPE_CGROUP_SKB can be documented
together with either cgroup-bpf programs or programs that access skb.
The new file is added to the index and verified by `make htmldocs` /
sanity-check by lynx.
Signed-off-by: Andrey Ignatov <rdna@fb.com>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Jeff Kirsher says:
====================
100GbE Intel Wired LAN Driver Updates 2019-04-18
This series contains updates to the ice driver only.
Anirudh fixes up code comments which had typos. Added support for DCB
into the ice driver, which required a bit of refactoring of the existing
code. Also fixed a potential race condition between closing and opening
the VSI for a MIB change event, so resolved this by grabbing the
rtnl_lock prior to closing. Added support to process LLDP MIB change
notifications. Added support for reporting DCB stats via ethtool.
Brett updates the calculation to increment ITR to use a direct
calculation instead of using estimations. This provides a more accurate
value.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Nothing really special standing out this time, iwlwifi being the most
active driver.
Major changes:
iwlwifi
* send NO_DATA events so they can be captured in radiotap
* support for multiple BSSID
* support for some new FW API versions
* support new hardware
* debugfs cleanups by Greg-KH
qtnfmac
* allow each MAC to specify its own regulatory rules
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQEcBAABAgAGBQJcuHgsAAoJEG4XJFUm622bfo8H/3uRRxsQBHGg6e3NpELaxpNV
IfrPDtvxyfILzIepBBhnZYUY0OrlTHKfMmzFBD9FFMojsxBYddnLZ/0iKUNKfwLm
KzToW/64YJ784dc+tw85gjh8I3MB+RRoD0l01M1HuOkzQ4hDNEGK3IsMHsBs/oTZ
huiqTYsTxStOj53vOpQiBFZ1pYBtvGLMxBdSepDcR27bgT1gwriynCSkSNglDH8z
/t3m6hDGtZa6uVkoIVH+BAMu6+vt+vIkU/TOdmiW/zqBL2JYq6cDE0uIb3bLAzN6
uvS1Rj42P3OwHUwFavlUBdr5Rdcj6P24S5ZhtVaGGWCBjMZI5/nO7IjzwyQnQuQ=
=/6q9
-----END PGP SIGNATURE-----
Merge tag 'wireless-drivers-next-for-davem-2019-04-18' of git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers-next
Kalle Valo says:
====================
wireless-drivers-next patches for 5.2
Nothing really special standing out this time, iwlwifi being the most
active driver.
Major changes:
iwlwifi
* send NO_DATA events so they can be captured in radiotap
* support for multiple BSSID
* support for some new FW API versions
* support new hardware
* debugfs cleanups by Greg-KH
qtnfmac
* allow each MAC to specify its own regulatory rules
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently when calculating how much to increment ITR by inside of
ice_update_itr() we do some estimations and intermediate
calculations. Instead of doing estimations, just do the
calculation directly. This allows for a more accurate value and it
makes it easier for the next person to understand and update.
Also, remove the dividing the ITR value by 2 when latency
driven because the ITR values are already so low for 100Gbps
speed. This should help get to the desired ITR value faster.
Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Update driver version to 0.7.4
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch adds code to start or stop LLDP and DCBX in firmware through
use of ethtool private flags.
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch introduces a new function ice_dcb_rebuild which reinitializes
DCB after a reset.
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch adds a new function ice_update_dcb_stats to get DCB stats
from the hardware and ethtool support for displaying these stats.
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch introduces a new function ice_tx_prepare_vlan_flags_dcb to
insert 802.1p priority information into the VLAN header
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch adds a new function ice_vsi_cfg_dcb_rings which updates a
VSI's rings based on DCB traffic class information.
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch adds support to process LLDP MIB change notifications sent
by the firmware.
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
When the firmware doesn't support LLDP or DCBX, the driver should switch
to "software LLDP mode". This patch adds support for the same.
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch adds a new function ice_pf_dcb_cfg (and related helpers)
which applies the DCB configuration obtained from the firmware. As
part of this, VSIs/netdevs are updated with traffic class information.
This patch requires a bit of a refactor of existing code.
1. For a MIB change event, the associated VSI is closed and brought up
again. The gap between closing and opening the VSI can cause a race
condition. Fix this by grabbing the rtnl_lock prior to closing the
VSI and then only free it after re-opening the VSI during a MIB
change event.
2. ice_sched_query_elem is used in ice_sched.c and with this patch, in
ice_dcb.c as well. However, ice_dcb.c is not built when CONFIG_DCB is
unset. This results in namespace warnings (ice_sched.o: Externally
defined symbols with no external references) when CONFIG_DCB is unset.
To avoid this move ice_sched_query_elem from ice_sched.c to
ice_common.c.
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch introduces a new top level function ice_init_dcb (and
related lower level helper functions) which continues the DCB init
flow.
This function uses ice_get_dcb_cfg to get, parse and store the DCB
configuration. Once this is done, it sets itself up to be notified
by the firmware on LLDP MIB change events.
Reviewed-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch introduces a skeleton for ice_init_pf_dcb, the top level
function for DCB initialization. Subsequent patches will add to this
DCB init flow.
In this patch, ice_init_pf_dcb checks if DCB is a supported capability.
If so, an admin queue call to start the LLDP and DCBx in firmware is
issued. If not, an error is reported. Note that we don't fail the driver
init if DCB init fails.
Reviewed-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Bump driver version to 0.7.3
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Capitalize abbreviations and spell out some that aren't obvious.
Reviewed-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch fixes typos in code comments.
Reviewed-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Somehow file drivers/net/wireless/realtek/rtlwifi/rtl8188ee/trx.c.rej was
incorporated into the sources. Obviously, it can be removed.
Signed-off-by: Larry Finger <Larry.Finger@lwfinger.net>
Reported-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Disabling IPv6 on an interface removes existing entries but nothing prevents
new entries from being manually added. To that end, add a new neigh_table
operation, allow_add, that is called on RTM_NEWNEIGH to see if neighbor
entries are allowed on a given device. If IPv6 is disabled on the device,
allow_add returns false and passes a message back to the user via extack.
$ echo 1 > /proc/sys/net/ipv6/conf/eth1/disable_ipv6
$ ip -6 neigh add fe80::4c88:bff:fe21:2704 dev eth1 lladdr de:ad:be:ef:01:01
Error: IPv6 is disabled on this device.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
David Ahern says:
====================
ipv6: Use fib6_result for fib_lookups
Add fib6_result as a single data structure to hold results from a fib
lookup. IPv6 currently has everything in 1 data structure - a fib6_info,
but with nexthop objects the fib6_nh can be in a nexthop or a nexthop
can be a blackhole which affects the fib6_type and flags (REJECT).
v2
- fixed 2 bugs in patch12:
i. checking return from fib6_table_lookup in fib6_lookup
ii. call to fib6_rule_saddr in fib6_rule_action_alt should use res->nh
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Add the fib6_flags and fib6_type to fib6_result. Update the lookup helpers
to set them and update post fib lookup users to use the version from the
result.
This allows nexthop objects to have blackhole nexthop.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change fib6_lookup and fib6_table_lookup to take a fib6_result and set
f6i and nh rather than returning a fib6_info. For now both always
return 0.
A later patch set can make these more like the IPv4 counterparts and
return EINVAL, EACCESS, etc based on fib6_type.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change fib6_table_lookup tracepoint to take the fib6_result and use
the fib6_info and fib6_nh from it.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pass fib6_result to rt6_select. Instead of returning the fib entry, it
will set f6i and nh based on the lookup.
find_rr_leaf is changed to remove the match option in favor of taking
fib6_result and having __find_rr_leaf set f6i in the result.
In the process, update fib6_info references in __find_rr_leaf to f6i names.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pass fib6_result to rt6_device_match with f6i set. rt6_device_match
updates f6i in the result if it finds a better match and sets nh.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change ip6_mtu_from_fib6 and fib6_mtu to take a fib6_result over a
fib6_info. Update both to use the fib6_nh from fib6_result.
Since the signature of ip6_mtu_from_fib6 is already changing, add const
to daddr and saddr.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Update rt6_insert_exception to take a fib6_result over a fib6_info.
Change ort to f6i from the fib6_result and rename to better reflect
what it references (a fib6_info).
Since this function is already getting changed, update the comments
to reference fib6_info variables rather than the older rt6_info.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that all callers are update to have a fib6_result, pass it down
to ip6_rt_get_dev_rcu, ip6_rt_copy_init, and ip6_rt_init_dst.
In the process, change ort to f6i in ip6_rt_copy_init to make it
clear it is a reference to a fib6_info.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Update ip6_rt_pcpu_alloc, rt6_get_pcpu_route and rt6_make_pcpu_route
to a fib6_result over a fib6_info.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change ip6_create_rt_rcu to take fib6_result over a fib6_info.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change ip6_rt_cache_alloc to take a fib6_result over a fib6_info.
Since ip6_rt_cache_alloc is only the caller, update the
rt6_is_gw_or_nonexthop helper to take fib6_result.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Simplify rt6_find_cached_rt for the fast path cases and pass fib6_result
to rt6_find_cached_rt. Rename the local return variable to ret to maintain
consisting with fib6_result name.
Update the comment in rt6_find_cached_rt to reference the new names in
a fib6_info vs the old name when fib entries were an rt6_info.
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add 'struct fib6_result' to hold the fib entry and fib6_nh from a fib
lookup as separate entries, similar to what IPv4 now has with fib_result.
Rename fib6_multipath_select to fib6_select_path, pass fib6_result to
it, and set f6i and nh in the result once a path selection is done.
Call fib6_select_path unconditionally for path selection which means
moving the sibling and oif check to fib6_select_path. To handle the two
different call paths (2 only call multipath_select if flowi6_oif == 0 and
the other always calls it), add a new have_oif_match that controls the
sibling walk if relevant.
Update callers of fib6_multipath_select accordingly and have them use the
fib6_info and fib6_nh from the result.
This is needed for multipath nexthop objects where a single f6i can
point to multiple fib6_nh (similar to IPv4).
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
I hit the following compilation error with gcc 4.8.5.
prog_tests/flow_dissector.c: In function ‘test_flow_dissector’:
prog_tests/flow_dissector.c:155:2: error: ‘for’ loop initial declarations are only allowed in C99 mode
for (int i = 0; i < ARRAY_SIZE(tests); i++) {
^
prog_tests/flow_dissector.c:155:2: note: use option -std=c99 or -std=gnu99 to compile your code
Let us fix the issue by avoiding this particular c99 feature.
Fixes: a5cb33464e ("selftests/bpf: make flow dissector tests more extensible")
Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Jesper Dangaard Brouer says:
====================
This patchset utilize a number of different kernel bulk APIs for optimizing
the performance for the XDP cpumap redirect feature.
Benchmark details are available here:
https://github.com/xdp-project/xdp-project/blob/master/areas/cpumap/cpumap03-optimizations.org
Performance measurements can be considered micro benchmarks, as they measure
dropping packets at different stages in the network stack.
Summary based on above:
Baseline benchmarks
- baseline-redirect: UdpNoPorts: 3,180,074
- baseline-redirect: iptables-raw drop: 6,193,534
Patch1: bpf: cpumap use ptr_ring_consume_batched
- redirect: UdpNoPorts: 3,327,729
- redirect: iptables-raw drop: 6,321,540
Patch2: net: core: introduce build_skb_around
- redirect: UdpNoPorts: 3,221,303
- redirect: iptables-raw drop: 6,320,066
Patch3: bpf: cpumap do bulk allocation of SKBs
- redirect: UdpNoPorts: 3,290,563
- redirect: iptables-raw drop: 6,650,112
Patch4: bpf: cpumap memory prefetchw optimizations for struct page
- redirect: UdpNoPorts: 3,520,250
- redirect: iptables-raw drop: 7,649,604
In this V2 submission I have chosen drop the SKB-list patch using
netif_receive_skb_list() as it was not showing a performance improvement for
these micro benchmarks.
====================
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
A lot of the performance gain comes from this patch.
While analysing performance overhead it was found that the largest CPU
stalls were caused when touching the struct page area. It is first read with
a READ_ONCE from build_skb_around via page_is_pfmemalloc(), and when freed
written by page_frag_free() call.
Measurements show that the prefetchw (W) variant operation is needed to
achieve the performance gain. We believe this optimization it two fold,
first the W-variant saves one step in the cache-coherency protocol, and
second it helps us to avoid the non-temporal prefetch HW optimizations and
bring this into all cache-levels. It might be worth investigating if
prefetch into L2 will have the same benefit.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
As cpumap now batch consume xdp_frame's from the ptr_ring, it knows how many
SKBs it need to allocate. Thus, lets bulk allocate these SKBs via
kmem_cache_alloc_bulk() API, and use the previously introduced function
build_skb_around().
Notice that the flag __GFP_ZERO asks the slab/slub allocator to clear the
memory for us. This does clear a larger area than needed, but my micro
benchmarks on Intel CPUs show that this is slightly faster due to being a
cacheline aligned area is cleared for the SKBs. (For SLUB allocator, there
is a future optimization potential, because SKBs will with high probability
originate from same page. If we can find/identify continuous memory areas
then the Intel CPU memset rep stos will have a real performance gain.)
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The function build_skb() also have the responsibility to allocate and clear
the SKB structure. Introduce a new function build_skb_around(), that moves
the responsibility of allocation and clearing to the caller. This allows
caller to use kmem_cache (slab/slub) bulk allocation API.
Next patch use this function combined with kmem_cache_alloc_bulk.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Song Liu <songliubraving@fb.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Move ptr_ring dequeue outside loop, that allocate SKBs and calls network
stack, as these operations that can take some time. The ptr_ring is a
communication channel between CPUs, where we want to reduce/limit any
cacheline bouncing.
Do a concentrated bulk dequeue via ptr_ring_consume_batched, to shorten the
period and times the remote cacheline in ptr_ring is read
Batch size 8 is both to (1) limit BH-disable period, and (2) consume one
cacheline on 64-bit archs. After reducing the BH-disable section further
then we can consider changing this, while still thinking about L1 cacheline
size being active.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Julian Wiedmann says:
====================
s390/qeth: updates 2019-04-17
please apply some additional qeth patches to net-next. This patchset
converts the driver to use the kernel's multiqueue model.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
