wdev is used even after del_virtual_interface handler in cfg80211
in nl80211_post_doit. Since we have freed wdev in handling of
del_virtual_intf, this can result into crash while deleting
interface.
Avoid this be not declaring wdev which part of
mwifiex_private structure but struct wireless_dev type.
Signed-off-by: Avinash Patil <patila@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Cathy Luo <cluo@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
RTS threshold, fragmentation threshold are per device properties.
Setting them on any interface would be reflected for all other
interfaces as well. This patch removes unnesessary command download
per interface.
Signed-off-by: Avinash Patil <patila@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Cathy Luo <cluo@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
This is not used anywhere execpt initialization.
Signed-off-by: Avinash Patil <patila@marvell.com>
Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Signed-off-by: Cathy Luo <cluo@marvell.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
The dasd driver has a lot of duplicated code to handle
dasd_global_profile. With this patch we use the same code for the
global and the per device profiling data. Note that dasd_stats_write
had to change slightly to maintain some odd differences between
A) per device and global profile and B) proc and sysfs interface
usage.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Reviewed-by: Stefan Haberland <stefan.haberland@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Access to DASDs global statistics is done without locking which
can lead to inconsistent data. Add locking to fix this. Also move
the relevant structs in a global dasd_profile struct.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Reviewed-by: Stefan Haberland <stefan.haberland@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Make use of gcc's hotpatch support to generate better code for ftrace
function tracing.
The generated code now contains only a six byte nop in each function
prologue instead of a 24 byte code block which will be runtime patched to
support function tracing.
With the new code generation the runtime overhead for supporting function
tracing is close to zero, while the original code did show a significant
performance impact.
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
gcc supports an s390 specific function attribute called "hotpatch".
It can be used to specify the number of halfwords that shall be added before
and after a function and which shall be filled with nops for runtime patching.
s390 will use the hotpatch attribute for function tracing, therefore make
sure that the notrace function attribute either disables the mcount call
or in case of hotpatch nop generation.
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
If the kernel is compiled with function tracer support the -pg compile option
is passed to gcc to generate extra code into the prologue of each function.
This patch replaces the "open-coded" -pg compile flag with a CC_FLAGS_FTRACE
makefile variable which architectures can override if a different option
should be used for code generation.
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Christian Borntraeger reported that the now missing diag 44 calls (voluntary
time slice end) does cause a performance regression for stop_machine() calls
if a machine has more virtual cpus than the host has physical cpus.
This patch mainly reverts 57f2ffe14f ("s390: remove diag 44 calls from
cpu_relax()") with the exception that we still do not issue diag 44 calls if
running with smt enabled. Due to group scheduling algorithms when running in
LPAR this would lead to significant latencies.
However, when running in LPAR we do not have more virtual than physical cpus.
Reported-and-tested-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
This change fixes below sparse error:
drivers/ssb/main.c:94:16: warning: symbol 'ssb_sdio_func_to_bus'
was not declared. Should it be static?
Acked-by: Michael Buesch <m@bues.ch>
Signed-off-by: Pramod Gurav <pramod.gurav@smartplayin.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
The nvram file can hold a key=value combination in which the value
may have spaces, ie. 'RAW1=80 02 fe ff'. The parsing functionality
did not deal with this so it gives an error message:
[621746.311635] brcmfmac: brcmf_nvram_handle_key
warning: ln=90:col=11: '=' expected, skip invalid key entry
because RAW1=80 is being considerd as key=value pair and it expects
'=' sign after '02' for next key=value pair. This entry can be
completely ignored as firmware does not need it.
Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Daniel (Deognyoun) Kim <dekim@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
When revinfo retrieval fails we can not show the firmware version
in ethtool driver info. Store the result to be used when handling
ethtool driver info callback.
Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Obtain device revision information and store it.
Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
In some scenarios the chip number and revision may not be provided
by the bus layer. If the chip number is not filled, the common
layer will ask the firmware for this information.
Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Daniel (Deognyoun) Kim <dekim@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Introducing a define that the caller of brcmu_boardrev_str() can
use to allocate enough room for buffer passed to the function.
Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
In brcmf_usb_probe_cb() the device is checked to determine whether
it is already running firmware. However, when no firmware download
is needed it still continues to request the firmware files. This
is fixed by returning after successful setup.
Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Daniel (Deognyoun) Kim <dekim@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
During suspend the bus state is put in the down state. When data
is being transmitted during this state then the netdev queue will
be close. This patch will wake the queue on state data if the
queue was closed.
Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Reviewed-by: Daniel (Deognyoun) Kim <dekim@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Each bus driver is maintaing an exported bus state indicating
if upper layers can or cannot send data. SDIO is using this state
also for more private states. This makes handling the states and
state changes complex. This patch minimises the exposed states
and makes SDIO keep track of an internal state where necessary.
Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Reviewed-by: Daniel (Deognyoun) Kim <dekim@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Before going in suspend state the watchdog thread needs to put the
device in bus sleep state, which assures it can go in deep-sleep
state during D3 state.
Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Daniel (Deognyoun) Kim <dekim@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
The driver always called cfg80211_disconnected() with reason parameter
set to zero, ie. unknown. However, firmware does provide a valid
802.11 reason code in DEAUTH and DISASSOC event message to the driver.
This patch passes the reason code to cfg80211_disconnected().
Reviewed-by: Daniel (Deognyoun) Kim <dekim@broadcom.com>
Reviewed-by: Hante Meuleman <meuleman@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Control frames are normally handled outside DPC, but sometimes
within DPC. To simplify code always handle control within DPC.
Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Reviewed-by: Daniel (Deognyoun) Kim <dekim@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
When the system is resumed a deadlock can occur when DPC gets
entered before resume is complete. This patch fixes this by
properly checking the suspend state outside the claim_host code
block.
Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Reviewed-by: Daniel (Deognyoun) Kim <dekim@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
On every tx the flow worker is triggered. When running high
throughput data this causes an excessive amount of times the worker
gets activated. This patch starts scheduling the worker more relaxed
once outstanding tx has reached a certain depth.
Reviewed-by: Arend Van Spriel <arend@broadcom.com>
Reviewed-by: Franky (Zhenhui) Lin <frankyl@broadcom.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Signed-off-by: Hante Meuleman <meuleman@broadcom.com>
Signed-off-by: Arend van Spriel <arend@broadcom.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Access point require this API to check peer alive status.
Assume peer is alive when it is connected, because
firmware implements keep alive checks and will disconnect
peer if it is not alive.
Signed-off-by: Vladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Implement Tx status reporting using skb_complete_wifi_ack().
Signed-off-by: Vladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Interrupt moderation parameters will never be passed as module
parameters. For product, they will be hard-coded after lab testing,
and for debugging, they can be altered via debugfs.
Signed-off-by: Vladimir Shulman <qca_shulmanv@qca.qualcomm.com>
Signed-off-by: Vladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Due to HW limitation, inter-packet gap timeout max value is 13 usec.
Update of current thresholds from 15 to 13 usec.
Signed-off-by: Vladimir Shulman <qca_shulmanv@qca.qualcomm.com>
Signed-off-by: Vladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Tuning rings size for performance optimization.
Increasing Tx ring size, allows buffering more packets for HW, thus
eliminating idle periods which were observed with smaller ring at
high throughput, because HW was fetching packets faster than driver was
filling them into the TX ring. Rx ring was similarly increased to
avoid same problems in Rx.
Signed-off-by: Vladimir Shulman <qca_shulmanv@qca.qualcomm.com>
Signed-off-by: Vladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
Tx queue was hard-coded to 1000 in ether_setup. Add wil_dev_setup
function which configures tx queue len to chosen default value
after calling ether_setup.
Signed-off-by: Vladimir Shulman <qca_shulmanv@qca.qualcomm.com>
Signed-off-by: Vladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
upon cfg80211_stop_ap, a graceful AP shutdown is requested from firmware
followed by firmware reset.
In case graceful request failed, error was returned to cfg80211.
The change is to return success in this scenario, because firmware
reset will anyhow shutdown the AP.
Signed-off-by: Dedy Lansky <qca_dlansky@qca.qualcomm.com>
Signed-off-by: Vladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
netif_carrier_on indication was too late. In case Rx packet received
before netif_carrier_on indication, upper layers could not send
Tx packet back.
The fix is to indicate netif_carrier_on earlier:
for STA, indicate netif_carrier_on when association starts.
for AP/PCP, indicate netif_carrier_on upon starting AP/PCP.
Signed-off-by: Dedy Lansky <qca_dlansky@qca.qualcomm.com>
Signed-off-by: Vladimir Kondratiev <qca_vkondrat@qca.qualcomm.com>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
The Bluetooth HCI transport specification for USB device defines on how
a standard AMP controller is identified and operated. This patch adds
the needed handling to hook it up to the Bluetooth stack.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
When I added sk_pacing_rate field, I forgot to initialize its value
in the per cpu unicast_sock used in ip_send_unicast_reply()
This means that for sch_fq users, RST packets, or ACK packets sent
on behalf of TIME_WAIT sockets might be sent to slowly or even dropped
once we reach the per flow limit.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Fixes: 95bd09eb27 ("tcp: TSO packets automatic sizing")
Signed-off-by: David S. Miller <davem@davemloft.net>
TIME_WAIT sockets are not owning any skb.
ip_send_unicast_reply() and tcp_v6_send_response() both use
regular sockets.
We can safely remove a test in sch_fq and save one cache line miss,
as sk_state is far away from sk_pacing_rate.
Tested at Google for about one year.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
NET_ACT_CONNMARK fails to build if NF_CONNTRACK_MARK is disabled,
and d7924450e1 ("act_connmark: Add missing dependency on
NF_CONNTRACK_MARK") fixed that case, but missed the cased where
NF_CONNTRACK is a loadable module.
This adds the second dependency to ensure that NET_ACT_CONNMARK
can only be built-in if NF_CONNTRACK is also part of the kernel
rather than a loadable module.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The sock_iocb structure is allocate on stack for each read/write-like
operation on sockets, and contains various fields of which only the
embedded msghdr and sometimes a pointer to the scm_cookie is ever used.
Get rid of the sock_iocb and put a msghdr directly on the stack and pass
the scm_cookie explicitly to netlink_mmap_sendmsg.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The if block was supposed to have curly braces. In the current code we
complain about dropped rx packets when we shouldn't.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, it isn't possible to request checksums on the outer UDP
header of tunnels - the TUNNEL_CSUM flag is ignored. This adds
support for requesting that UDP checksums be computed on transmit
and properly reported if they are present on receive.
Signed-off-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is the first NFC pull request for 3.20.
With this one we have:
- Secure element support for the ST Micro st21nfca driver. This depends
on a few HCI internal changes in order for example to support more
than one secure element per controller.
- ACPI support for NXP's pn544 HCI driver. This controller is found on
many x86 SoCs and is typically enumerated on the ACPI bus there.
- A few st21nfca and st21nfcb fixes.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1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=SNWp
-----END PGP SIGNATURE-----
Merge tag 'nfc-next-3.20-1' of git://git.kernel.org/pub/scm/linux/kernel/git/sameo/nfc-next
NFC: 3.20 first pull request
This is the first NFC pull request for 3.20.
With this one we have:
- Secure element support for the ST Micro st21nfca driver. This depends
on a few HCI internal changes in order for example to support more
than one secure element per controller.
- ACPI support for NXP's pn544 HCI driver. This controller is found on
many x86 SoCs and is typically enumerated on the ACPI bus there.
- A few st21nfca and st21nfcb fixes.
Signed-off-by: David S. Miller <davem@davemloft.net>
The carry from the 64->32bits folding was dropped, e.g with:
saddr=0xFFFFFFFF daddr=0xFF0000FF len=0xFFFF proto=0 sum=1,
csum_tcpudp_nofold returned 0 instead of 1.
Signed-off-by: Karl Beldan <karl.beldan@rivierawaves.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Mike Frysinger <vapier@gentoo.org>
Cc: netdev@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: stable@vger.kernel.org
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The Intel Bluetooth devices use the generic USB device/interface class
descriptors that are assigned to Bluetooth H:2 conforming transports.
T: Bus=01 Lev=01 Prnt=01 Port=01 Cnt=02 Dev#= 3 Spd=12 MxCh= 0
D: Ver= 2.01 Cls=e0(wlcon) Sub=01 Prot=01 MxPS=64 #Cfgs= 1
However newer chips have a bootloader stage and require firmware to
be loaded before they are functional. To avoid any confusion for the
users, just ignore unknown Intel Bluetooth devices.
All the released Intel Bluetooth devices have an entry in the device
table identifying their setup and support requirements. The advantage
here is that older kernel can be booted with newer devices without
causing any disturbance.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
New entries to the USB blacklist/quirk device table should be sorted
by USB vendor id. Fix the recent entry fro Marvell devices.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Reported in: https://bugzilla.kernel.org/show_bug.cgi?id=92081
This patch avoids calling rtnl_notify if the device ndo_bridge_getlink
handler does not return any bytes in the skb.
Alternately, the skb->len check can be moved inside rtnl_notify.
For the bridge vlan case described in 92081, there is also a fix needed
in bridge driver to generate a proper notification. Will fix that in
subsequent patch.
v2: rebase patch on net tree
Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Neal Cardwell says:
====================
fix stretch ACK bugs in TCP CUBIC and Reno
This patch series fixes the TCP CUBIC and Reno congestion control
modules to properly handle stretch ACKs in their respective additive
increase modes, and in the transitions from slow start to additive
increase.
This finishes the project started by commit 9f9843a751 ("tcp:
properly handle stretch acks in slow start"), which fixed behavior for
TCP congestion control when handling stretch ACKs in slow start mode.
Motivation: In the Jan 2015 netdev thread 'BW regression after "tcp:
refine TSO autosizing"', Eyal Perry documented a regression that Eric
Dumazet determined was caused by improper handling of TCP stretch
ACKs.
Background: LRO, GRO, delayed ACKs, and middleboxes can cause "stretch
ACKs" that cover more than the RFC-specified maximum of 2
packets. These stretch ACKs can cause serious performance shortfalls
in common congestion control algorithms, like Reno and CUBIC, which
were designed and tuned years ago with receiver hosts that were not
using LRO or GRO, and were instead ACKing every other packet.
Testing: at Google we have been using this approach for handling
stretch ACKs for CUBIC datacenter and Internet traffic for several
years, with good results.
v2:
* fixed return type of tcp_slow_start() to be u32 instead of int
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes a bug in CUBIC that causes cwnd to increase slightly
too slowly when multiple ACKs arrive in the same jiffy.
If cwnd is supposed to increase at a rate of more than once per jiffy,
then CUBIC was sometimes too slow. Because the bic_target is
calculated for a future point in time, calculated with time in
jiffies, the cwnd can increase over the course of the jiffy while the
bic_target calculated as the proper CUBIC cwnd at time
t=tcp_time_stamp+rtt does not increase, because tcp_time_stamp only
increases on jiffy tick boundaries.
So since the cnt is set to:
ca->cnt = cwnd / (bic_target - cwnd);
as cwnd increases but bic_target does not increase due to jiffy
granularity, the cnt becomes too large, causing cwnd to increase
too slowly.
For example:
- suppose at the beginning of a jiffy, cwnd=40, bic_target=44
- so CUBIC sets:
ca->cnt = cwnd / (bic_target - cwnd) = 40 / (44 - 40) = 40/4 = 10
- suppose we get 10 acks, each for 1 segment, so tcp_cong_avoid_ai()
increases cwnd to 41
- so CUBIC sets:
ca->cnt = cwnd / (bic_target - cwnd) = 41 / (44 - 41) = 41 / 3 = 13
So now CUBIC will wait for 13 packets to be ACKed before increasing
cwnd to 42, insted of 10 as it should.
The fix is to avoid adjusting the slope (determined by ca->cnt)
multiple times within a jiffy, and instead skip to compute the Reno
cwnd, the "TCP friendliness" code path.
Reported-by: Eyal Perry <eyalpe@mellanox.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change CUBIC to properly handle stretch ACKs in additive increase mode
by passing in the count of ACKed packets to tcp_cong_avoid_ai().
In addition, because we are now precisely accounting for stretch ACKs,
including delayed ACKs, we can now remove the delayed ACK tracking and
estimation code that tracked recent delayed ACK behavior in
ca->delayed_ack.
Reported-by: Eyal Perry <eyalpe@mellanox.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
