| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
This patch reverts commit
80b45261a0b263536b043c5ccfc4ba4fc27c2acc
which was implementing a 'cancelled' functionality to notify that
a cancelled request will not be replied.
This implementation was not used anywhere and therefore removed.
Signed-off-by: Andi Shyti <andi@etezian.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The driver uses dev_err when enabling of AER fails (e.g. PCIe AER is not
supported). The dev_info is more appropriate to avoid console pollution.
Cc: sathya.perla@emulex.com
Cc: subbu.seetharaman@emulex.com
Cc: ajit.khaparde@emulex.com
Signed-off-by: Ivan Vecera <ivecera@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Nithin Nayak Sujir <nsujir@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The 5762 devices sometimes incorrectly treat udp fragments as RMCP
packets and route to the APE. This patch sets the RX_MODE_IPV4_FRAG_FIX
bit for these devices which enables the proper behaviour.
Signed-off-by: Nithin Nayak Sujir <nsujir@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The PTP_CAPABLE tg3 devices have a gpio output that is toggled when the
free running counter matches a watchdog value. This patch adds support
to set the watchdog and enable this feature.
Since the output is controlled via bits in the EAV_REF_CLCK_CTL
register, we have to read-modify-write it when we stop/resume.
Cc: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: Nithin Nayak Sujir <nsujir@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Also remove the call to tg3_power_down_prepare() in tg3_power_down()
since tg3_close() calls it.
Signed-off-by: Nithin Nayak Sujir <nsujir@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Previously, when the interface was brought down, the driver would set
the power state to D3hot. In D3hot, we don't have access to the NVRAM.
This patch removes the call to set the power state to PCI_D3hot in
close. A following patch will implement the shutdown handler to properly
set the D3hot state when the system is going down.
Doing the above means that the TG3_PHYFLG_IS_LOW_POWER should not be
checked to validate access to the NVRAM.
Signed-off-by: Nithin Nayak Sujir <nsujir@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
During probe, the driver is incorrectly switching the power to Vaux on
the 5717 and later devices. At this point, we are in D0 state and
drawing maximum power. We also definitely have Vmain available. It
doesn't make sense to switch to Vaux since it has a lesser maximum power
draw and we might go over the limit. On a new system, we observe that
not all ports are recognized in some of the slots with this call in
place.
Signed-off-by: Nithin Nayak Sujir <nsujir@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Completion status field should also be checked for non-zero error
condition.
Signed-off-by: Eddie Wai <eddie.wai@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Update TCP delayed ACK and timestamp options setup to match latest bnx2x
firmware.
Signed-off-by: Eddie Wai <eddie.wai@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Without resetting it, the bnx2i driver cannot use different options for
different iSCSI connections.
Signed-off-by: Eddie Wai <eddie.wai@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Since unregister_netdevice_notifier() will replay the NETDEV_DOWN and
NETDEV_UNREGISTER_EVENTS, the cnic_dev_list will be cleaned up automatically.
The loop to cleanup the cnic_dev_list can be removed in cnic_release().
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
After this earlier commit to simplify probing:
commit 4bd9b0fffb193d2e288f67f81821af32df8d4349
cnic, bnx2x, bnx2: Simplify cnic probing.
we can now reliably receive netdev events and we can simplify the handling
of these events. We now remove the logic that tries to handle missed
NETDEV_REGISTER events.
This change will allow cleanup to be simplified in the next patch. We can
now rely on the play back of netdev events during
unregister_netdevice_notifier() to cleanup the structures.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This commit adds new firmware command and new firmware event. The firmware
raises the MLX4_EVENT_TYPE_OP_REQUIRED event in order to signal the driver it
needs to perform an administrative operation throughout the MLX4_CMD_GET_OP_REQ
command. At the moment the supported operation is adding/removing multicast
entries which are used by the firmware for handling NCSI traffic in B0
steering mode.
Also, had to swap the order of mlx4_init_mcg_table() and
mlx4_init_eq_table() to make sure that driver will get events only after
resources are initialized to handle it.
Signed-off-by: Yevgeny Petrilin <yevgenyp@mellanox.com>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.com>
Signed-off-by: Eugenia Emantayev <eugenia@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Fix a race between BlueFlame flow and stamping in post send flow.
Example:
SW: Build WQE 0 on the TX buffer, except the ownership bit
SW: Set ownership for WQE 0 on the TX buffer
SW: Ring doorbell for WQE 0
SW: Build WQE 1 on the TX buffer, except the ownership bit
SW: Set ownership for WQE 1 on the TX buffer
HW: Read WQE 0 and then WQE 1, before doorbell was rung/BF was done for WQE 1
HW: Produce CQEs for WQE 0 and WQE 1
SW: Process the CQEs, and stamp WQE 0 and WQE 1 accordingly (on the TX buffer)
SW: Copy WQE 1 from the TX buffer to the BF register - ALREADY STAMPED!
HW: CQE error with index 0xFFFF - the BF WQE's control segment is STAMPED,
so the BF index is 0xFFFF. Error: Invalid Opcode.
As a result QP enters the error state and no traffic can be sent.
Solution:
When stamping - do not stamp last completed wqe.
Signed-off-by: Eugenia Emantayev <eugenia@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Fixes this on PowerPC (at least):
net/core/pktgen.c: In function 'fill_packet_ipv6':
net/core/pktgen.c:2906:3: error: implicit declaration of function 'csum_ipv6_magic' [-Werror=implicit-function-declaration]
udph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, udplen, IPPROTO_UDP, 0);
^
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net-next
Jeff Kirsher says:
====================
This series contains updates to e100 and e1000e.
The e100 patch from Andy simply updates the netif_printk() to use
%*ph to dump small buffers.
The changes to e1000e include a fix from Dean Nelson to resolve a
issue where a pci_clear_master() was accidentally dropped during a
conflict resolution. Wei Young provides 2 patches, one removes an
assignment of the default ring size because it was a duplicate. The
second changes the packet split receive structure to use
PS_PAGE_BUFFERS macro for the length so that problems won't occur
when the length is changed.
The remaining patches for e1000e are from Bruce Allan, where he
provides a number of fixes and updates for I218. In addition, a
fix for 82583 which can disappear off the PCIe bus, to resolve this,
disable ASPM L1. Bruce also provides a fix to a previous commit
(commit e60b22c5b7 e1000e: fix accessing to suspended device) so that
devices are only taken out of runtime power management for those
ethtool operations that must access device registers.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
v2:
a) Also send ipv4 igmp messages with TC_PRIO_CONTROL
Cc: William Manley <william.manley@youview.com>
Cc: Lukas Tribus <luky-37@hotmail.com>
Acked-by: Benjamin LaHaise <bcrl@kvack.org>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The initialization of the PHY on I217/I218, while similar to 82579, must
also check to see if the MAC and PHY are in the same mode (PCIe vs. SMBus)
otherwise the PHY will be inaccessible by the MAC.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Jeff Pieper <jeffrey.e.pieper@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
|
|
When the device is runtime suspended (e.g. when there is no link), do not
wake it from D3 to read the PHY status; just set the values to typical
power-on defaults as is done when runtime PM is not enabled and there is no
link.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
|
|
The device IDs 0x15a0 and 0x15a1 are new SKUs that contain the same MAC as
I217 and same PHY as I218.
The device IDs 0x15a2 and 0x15a3 are the same as existing I218 SKUs.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
|
|
A previous patch (commit e60b22c5b7 e1000e: fix accessing to suspended
device) added .begin and .complete ethtool driver callbacks so that the
device was resumed from Runtime Power Management (RPM) suspend state for
all ethtool operations. This is overkill for operations which do not need
to access any registers in the device. This patch makes it so that the
device is taken out of RPM suspend only for those ethtool operations that
must access device registers.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
|
|
Tx hang is an unintended consequence of another workaround that is in the
EEPROM for an issue with the firmware at 10Mbps when K1 (a power mode of
the MAC-PHY interconnect) is enabled. The issue is resolved by setting
appropriate Tx re-transmission timeouts in the PHY and associated K1 entry
times in the MAC to allow enough transmissions to occur without triggering
a Tx hang. A similar change is needed when linked at 10Mbps to improve
latency.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Jeff Pieper <jeffrey.e.pieper@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
|
|
Alter the packet buffer allocation accordingly.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Jeff Pieper <jeffrey.e.pieper@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
|
|
The jumbo frame configuration in the MAC/PHY should be reverted on 82579
and newer parts when the interface is brought down (not just when the MTU
is changed back to standard frame size) otherwise iAMT connections (e.g.
SoL, IDE-R) will be dropped and cannot be re-acquired until the MTU is
changed again.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
|
|
The 82583 can disappear off the PCIe bus. This device is a modified 82574
which had the same problem which was fixed by disabling ASPM L1; disabling
it on 82583 fixes the issue on this device.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
|
|
In structure e1000_rx_desc_packet_split, the size of wb.upper.length is
defined by a digit. This may introduce some problem when the length is
changed.
This patch use the macro PS_PAGE_BUFFERS for the definition. And move the
definition to hw.h.
Signed-off-by: Wei Yang <weiyang@linux.vnet.ibm.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
|
|
tx_ring/rx_ring size is assigned in function e1000_alloc_queues(), which is
called by e1000_sw_init() in the early stage of e1000_probe().
This patch just remove the duplicate assignment of this default ring size
value.
Signed-off-by: Wei Yang <weiyang@linux.vnet.ibm.com>
Reviewed-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Reviewed-by: Da Yu Qiu <qiudayu@cn.ibm.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Acked-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
|
|
In attempting to resolve a minor merge conflict, commit e5f2ef7ab4690d2e8faa
(Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net) accidentally
dropped a call to pci_clear_master() that was intended to remain in place.
Commit 4e0855dff094b0d56d6b (e1000e: fix pci-device enable-counter balance)
replaced a call to pci_disable_device() by one to pci_clear_master(). And then
commit 66148babe728f3e00e13 (e1000e: fix runtime power management transitions)
deleted a number of lines starting two lines following that call.
This patch restores the call to pci_clear_master() in __e1000_shutdown().
v2: added summary lines (enclosed in parens) following commit IDs
Signed-off-by: Dean Nelson <dnelson@redhat.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Acked-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
|
|
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
|
|
After commit 4aa5dee4d9 ("net: convert resend IGMP to notifier event") we
have 1 read_unlock in bond_resend_igmp_join_requests which isn't paired
with a read_lock because it's removed by that commit.
Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Reviewed-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
UDP checksums are optional, hence pktgen has been omitting them in
favour of performance. The optional flag UDPCSUM enables UDP
checksumming. If the output device supports hardware checksumming
the skb is prepared and marked CHECKSUM_PARTIAL, otherwise the
checksum is generated in software.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Ben Greear <greearb@candelatech.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This is useful for other VSOCK transport implemented outside the
net/vmw_vsock/ directory to use these headers.
Signed-off-by: Asias He <asias@redhat.com>
Acked-by: Andy King <acking@vmware.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
git://git.infradead.org/users/dvhart/linux-2.6 into minnow
Darren Hart says:
====================
Add support for the MinnowBoard in the pch_gbe driver. This was
originally sent to LKML as part of the MinnowBoard support series. That
is now partially merged and this version of the patch has been isolated
from those changes and is now completely self-contained.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The default RX_QLEN()/TX_QLEN() didn't consider super speed
USB device, so only max 4 URBs are scheduled at the same time
for tx/rx, then USB3 NIC can't perform very well.
With this patch, both rx and tx thoughput are increased more than
100Mbps when doing iperf test on ax88179_178a USB 3.0 NIC.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch centralizes computing of max rx/tx qlen, because:
- RX_QLEN()/TX_QLEN() is called in hot path
- computing depends on device's usb speed, now we have ls/fs, hs, ss,
so more checks need to be involved
- in fact, max rx/tx qlen should not only depend on device USB
speed, but also depend on ethernet link speed, so we need to
consider that in future.
- if SG support is done, max tx qlen may need change too
Generally, hard_mtu and rx_urb_size are changed in bind(), reset()
and link_reset() callback, and change mtu network operation, this
patches introduces the API of usbnet_update_max_qlen(), and calls
it in above path.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Inspired by commit f09e2249c4f5c7c13261ec73f5a7807076af0c8e (macvtap: restore
vlan header on user read). This patch adds hardware vlan tx support for
tuntap. This is done by copying vlan header directly into userspace in
tun_put_user() instead of doing it through __vlan_put_tag() in
dev_hard_start_xmit(). This eliminates one unnecessary memmove() in
vlan_insert_tag() for 802.1ad and 802.1q traffic.
pktgen test shows about 20% improvement for 802.1q traffic:
Before:
662149pps 317Mb/sec (317831520bps) errors: 0
After:
801033pps 384Mb/sec (384495840bps) errors: 0
Cc: Basil Gor <basil.gor@gmail.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This patch consolidates the SCTP checksum calculation code from various
places to a single new function, sctp_compute_cksum(skb, offset).
Signed-off-by: Joe Stringer <joe@wand.net.nz>
Reviewed-by: Julian Anastasov <ja@ssi.bg>
Acked-by: Simon Horman <horms@verge.net.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
For small packets we can simplify xmit processing
by linearizing buffers with the header:
most packets seem to have enough head room
we can use for this purpose.
Since existing hypervisors require that header
is the first s/g element, we need a feature bit
for this.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This started out with fixing a sparse warning, then I realized that
the wrapper function bond_netpoll_info could just be removed
by rolling it into the enable code.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Reviewed-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This started out with fixing a sparse warning, then I realized that
the wrapper function team_netpoll_info could just be collapsed away
by rolling it into the enable code.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This started out with fixing a sparse warning, then I realized that
the wrapper function br_netpoll_info could just be collapsed away
by rolling it into the enable code.
Also, eliminate unnecessary goto's
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Reviewed-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
We have BOND_MODE_ROUNDROBIN pre-defined as 0, and it's the lowest
mode number.
Use it to check the arg lower bound instead of magic number 0 in
bond_mode_name.
Signed-off-by: Wang Sheng-Hui <shhuiw@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The MinnowBoard uses an AR803x PHY with the PCH GBE which requires
special handling. Use the MinnowBoard PCI Subsystem ID to detect this
and add a pci_device_id.driver_data structure and functions to handle
platform setup.
The AR803x does not implement the RGMII 2ns TX clock delay in the trace
routing nor via strapping. Add a detection method for the board and the
PHY and enable the TX clock delay via the registers.
This PHY will hibernate without link for 10 seconds. Ensure the PHY is
awake for probe and then disable hibernation. A future improvement would
be to convert pch_gbe to using PHYLIB and making sure we can wake the
PHY at the necessary times rather than permanently disabling it.
Signed-off-by: Darren Hart <dvhart@linux.intel.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Peter Waskiewicz <peter.p.waskiewicz.jr@intel.com>
Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Joe Perches <joe@perches.com>
Cc: netdev@vger.kernel.org
|
|
devm_ioremap_resource
devm_ioremap_resource does sanity checks on the given resource. No need to
duplicate this in the driver.
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Avoid using magic numbers when we have perfectly good defines just lying
around.
Signed-off-by: Darren Hart <dvhart@linux.intel.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Peter Waskiewicz <peter.p.waskiewicz.jr@intel.com>
Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: netdev@vger.kernel.org
|
|
No users outside net/core/dev.c.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Idea of this patch is to add optional limitation of number of
unsent bytes in TCP sockets, to reduce usage of kernel memory.
TCP receiver might announce a big window, and TCP sender autotuning
might allow a large amount of bytes in write queue, but this has little
performance impact if a large part of this buffering is wasted :
Write queue needs to be large only to deal with large BDP, not
necessarily to cope with scheduling delays (incoming ACKS make room
for the application to queue more bytes)
For most workloads, using a value of 128 KB or less is OK to give
applications enough time to react to POLLOUT events in time
(or being awaken in a blocking sendmsg())
This patch adds two ways to set the limit :
1) Per socket option TCP_NOTSENT_LOWAT
2) A sysctl (/proc/sys/net/ipv4/tcp_notsent_lowat) for sockets
not using TCP_NOTSENT_LOWAT socket option (or setting a zero value)
Default value being UINT_MAX (0xFFFFFFFF), meaning this has no effect.
This changes poll()/select()/epoll() to report POLLOUT
only if number of unsent bytes is below tp->nosent_lowat
Note this might increase number of sendmsg()/sendfile() calls
when using non blocking sockets,
and increase number of context switches for blocking sockets.
Note this is not related to SO_SNDLOWAT (as SO_SNDLOWAT is
defined as :
Specify the minimum number of bytes in the buffer until
the socket layer will pass the data to the protocol)
Tested:
netperf sessions, and watching /proc/net/protocols "memory" column for TCP
With 200 concurrent netperf -t TCP_STREAM sessions, amount of kernel memory
used by TCP buffers shrinks by ~55 % (20567 pages instead of 45458)
lpq83:~# echo -1 >/proc/sys/net/ipv4/tcp_notsent_lowat
lpq83:~# (super_netperf 200 -t TCP_STREAM -H remote -l 90 &); sleep 60 ; grep TCP /proc/net/protocols
TCPv6 1880 2 45458 no 208 yes ipv6 y y y y y y y y y y y y y n y y y y y
TCP 1696 508 45458 no 208 yes kernel y y y y y y y y y y y y y n y y y y y
lpq83:~# echo 131072 >/proc/sys/net/ipv4/tcp_notsent_lowat
lpq83:~# (super_netperf 200 -t TCP_STREAM -H remote -l 90 &); sleep 60 ; grep TCP /proc/net/protocols
TCPv6 1880 2 20567 no 208 yes ipv6 y y y y y y y y y y y y y n y y y y y
TCP 1696 508 20567 no 208 yes kernel y y y y y y y y y y y y y n y y y y y
Using 128KB has no bad effect on the throughput or cpu usage
of a single flow, although there is an increase of context switches.
A bonus is that we hold socket lock for a shorter amount
of time and should improve latencies of ACK processing.
lpq83:~# echo -1 >/proc/sys/net/ipv4/tcp_notsent_lowat
lpq83:~# perf stat -e context-switches ./netperf -H 7.7.7.84 -t omni -l 20 -c -i10,3
OMNI Send TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : +/-2.500% @ 99% conf.
Local Remote Local Elapsed Throughput Throughput Local Local Remote Remote Local Remote Service
Send Socket Recv Socket Send Time Units CPU CPU CPU CPU Service Service Demand
Size Size Size (sec) Util Util Util Util Demand Demand Units
Final Final % Method % Method
1651584 6291456 16384 20.00 17447.90 10^6bits/s 3.13 S -1.00 U 0.353 -1.000 usec/KB
Performance counter stats for './netperf -H 7.7.7.84 -t omni -l 20 -c -i10,3':
412,514 context-switches
200.034645535 seconds time elapsed
lpq83:~# echo 131072 >/proc/sys/net/ipv4/tcp_notsent_lowat
lpq83:~# perf stat -e context-switches ./netperf -H 7.7.7.84 -t omni -l 20 -c -i10,3
OMNI Send TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : +/-2.500% @ 99% conf.
Local Remote Local Elapsed Throughput Throughput Local Local Remote Remote Local Remote Service
Send Socket Recv Socket Send Time Units CPU CPU CPU CPU Service Service Demand
Size Size Size (sec) Util Util Util Util Demand Demand Units
Final Final % Method % Method
1593240 6291456 16384 20.00 17321.16 10^6bits/s 3.35 S -1.00 U 0.381 -1.000 usec/KB
Performance counter stats for './netperf -H 7.7.7.84 -t omni -l 20 -c -i10,3':
2,675,818 context-switches
200.029651391 seconds time elapsed
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-By: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
