igb: Fix possible panic caused by Rx traffic arrival while interface is down
This patch reorders disabling napi and irqs during igb_down.
This is done to avoid possible panic's found in other Intel drivers
when Rx traffic arrives while interface is going down.
igb: Fix set_ethtool function to call update nvm for entire image
This patch fixes a problem where we were only checking to update checksum
on first part of nvm image. Newer parts have multiple checksum fields and
checksum function will accommodate that as long as we call it in the first
place for any changes made.
This path allows users to get appropriate flow control setting on SerDes
devices, based on original implementation for Copper devices.
Also, since 100baseFX does not support setting flow control, so exclude
it from the setting mechanism.
This patch adds support for SFP modules media type discovery for
SGMII, which will enable driver to detect supported external PHYs,
including 100baseFXSFP module.
Matthew Vick [Wed, 24 Apr 2013 07:42:06 +0000 (07:42 +0000)]
igb: Add update to last_rx_timestamp in Rx rings
In order to support a more accurate check for a PTP Rx hang where the
device can no longer timestamp received packets, we need to update, per
ring, when the last Rx timestamp was. Because of how the PTP Rx hang logic
works, the current logic is valid, but properly updating the ring variable
increases the accuracy of the check.
igb: Changed LEDs blink mechanism to include designs using cathode
This patch addresses the changes needed to make LEDs work properly with
negative logic. This implementation uses LED Invert bit to reverse the
logic issue that occurred when LEDs are driven by cathode. Keep LEDs
blinking for SerDes devices. Also made changes to magic number and the
for loop to reduce number of shifts.
e1000e: Release mutex lock only if it has been initially acquired
This patch fixes the issue of unlocking swflag_mutex for 82574 and 82583
devices regardless of if the hw semaphore has been successfully acquired via
e1000_get_hw_semaphore_82574(). With this patch, unlocking mutex now depends
on if the hw semaphore was successfully acquired before. And 82574/82583
devices are reset regardless of whether e1000_get_hw_semaphore_82574()
returns success or failure.
Dave Airlie [Mon, 20 May 2013 23:42:55 +0000 (09:42 +1000)]
Merge branch 'drm-fixes-3.10-sun' of git://people.freedesktop.org/~agd5f/linux into drm-next
This is the pull request for AMD Sun/Hainan support. I've
split it out separately from my regular fixes stream. Hainan
is a new SI asic with no UVD or DCE hardware. The patches are
minimally invasive; basically just pci ids and skipping UVD and
DCE init for this family. Most of the changes to si.c are just
the golden register tables for the family.
* 'drm-fixes-3.10-sun' of git://people.freedesktop.org/~agd5f/linux:
drm/radeon: add Hainan pci ids
drm/radeon: add golden register settings for Hainan (v2)
drm/radeon: sun/hainan chips do not have UVD (v2)
drm/radeon: track which asics have UVD
drm/radeon: radeon-asic updates for Hainan
drm/radeon: fill in ucode loading support for Hainan
drm/radeon: don't touch DCE or VGA regs on Hainan (v3)
drm/radeon: fill in GPU init for Hainan (v2)
drm/radeon: add chip family for Hainan
Dave Airlie [Mon, 20 May 2013 23:41:33 +0000 (09:41 +1000)]
Merge branch 'drm-fixes-3.10' of git://people.freedesktop.org/~agd5f/linux into drm-next
Minor bug fixes.
* 'drm-fixes-3.10' of git://people.freedesktop.org/~agd5f/linux:
drm/radeon/dce2: use 10khz units for audio dto calculation
drm/radeon: Fix VRAM size calculation for VRAM >= 4GB
drm/radeon: Remove superfluous variable
Linus Torvalds [Mon, 20 May 2013 21:25:19 +0000 (14:25 -0700)]
Merge tag 'stable/for-linus-3.10-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen
Pull Xen fixes from Konrad Rzeszutek Wilk:
- Regression fix in xen privcmd fixing a memory leak.
- Add Documentation for tmem driver.
- Simplify and remove code in the tmem driver.
- Cleanups.
* tag 'stable/for-linus-3.10-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
xen: Fixed assignment error in if statement
xen/xenbus: Fixed over 80 character limit issue
xen/xenbus: Fixed indentation error in switch case
xen/tmem: Don't use self[ballooning|shrinking] if frontswap is off.
xen/tmem: Remove the usage of '[no|]selfballoon' and use 'tmem.selfballooning' bool instead.
xen/tmem: Remove the usage of 'noselfshrink' and use 'tmem.selfshrink' bool instead.
xen/tmem: Remove the boot options and fold them in the tmem.X parameters.
xen/tmem: s/disable_// and change the logic.
xen/tmem: Fix compile warning.
xen/tmem: Split out the different module/boot options.
xen/tmem: Move all of the boot and module parameters to the top of the file.
xen/tmem: Cleanup. Remove the parts that say temporary.
xen/privcmd: fix condition in privcmd_close()
Florian Fainelli [Sun, 19 May 2013 22:53:43 +0000 (22:53 +0000)]
phy: add phy_mac_interrupt() to use with PHY_IGNORE_INTERRUPT
There is currently no way for an Ethernet MAC driver servicing PHY link
interrupts to notify this to the PHY state machine without defining its
own state machine. Since most drivers are not so special, introduce a
helper: phy_mac_interrupt() which can be called from a link up/down
interrupt routine to update the PHY state machine. To avoid code
duplication some refactoring has been done to expose the workqueue and
its corresponding callback internally.
Florian Fainelli [Sun, 19 May 2013 22:53:42 +0000 (22:53 +0000)]
phy: fix the use of PHY_IGNORE_INTERRUPT
When a PHY device is registered with the special IRQ value
PHY_IGNORE_INTERRUPT (-2) it will not properly be handled by the PHY
library:
- it continues to poll its register, while we do not want this
because such PHY link events or register changes are serviced by an
Ethernet MAC
- it will still try to configure PHY interrupts at the PHY level, such
interrupts do not exist at the PHY but at the MAC level
- the state machine only handles PHY_POLL, but should also handle
PHY_IGNORE_INTERRUPT similarly
This patch updates the PHY state machine and initialization paths to
account for the specific PHY_IGNORE_INTERRUPT. Based on an earlier patch
by Thomas Petazzoni, and reworked to add the missing bits. Add a helper
phy_interrupt_is_valid() which specifically tests for a PHY interrupt
not to be PHY_POLL or PHY_IGNORE_INTERRUPT and use it throughout the
code.
Rasesh Mody [Mon, 20 May 2013 10:08:03 +0000 (10:08 +0000)]
bna: Enahncement to Identify Default IOC Function
User should not be allowed to delete base function of eth port. Add a new field
to the bfa ioc attributes structure to indicate if the given ioc is default
function on the port or not.
Rasesh Mody [Mon, 20 May 2013 10:08:02 +0000 (10:08 +0000)]
bna: Fix Ucast Failure Handling
Failure of the UCAST set for base mac address fails when user configures a
duplicate mac address that matches that of another vNIC on the same port.
The bna does not handle the ucast failure and keeps this address in cache.
On disable of the vNIC, bna tries to delete the failed base mac address and the
fw asserts.
On failure of ucast address, mark ucast address set to false.
Rasesh Mody [Mon, 20 May 2013 10:08:01 +0000 (10:08 +0000)]
bna: Clear Driver Config Flags When HW Resets
Driver configuration flags are retained across open/stop operations preventing
configurations to be set in next open/stop. Setting MTU on a 1020 causes
network to fail until a reboot is performed on the host.
Clear the flags when configuration resets in hardware.
David S. Miller [Mon, 20 May 2013 21:05:22 +0000 (14:05 -0700)]
Merge branch 'for-davem' of git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless
John W. Linville says:
====================
This pull request is intended for the 3.10 series. It contains a
variety of fixes for problems discovered during the merge window and
after 3.10-rc1.
For the mac80211 bits, Johannes says the following:
"This is what I have:
* a patch from Felix to fix RCU usage in his rate table code
* a patch from Ilan to add the wdev id to some notifications so they can
actually be used by userspace
* Sasha Levin found an issue in how hwsim handles devices
* a fix for a bug in the wiphy_register() error path that's been there forever
* three fixes for WoWLAN
* AP mode frame matching was erroneously giving frames to all virtual AP
interfaces (reported by Jouni)
* a fix for HT handling in my CSA changes, found by Sujith
* a fix for some locking simplifications gone wrong
* Ben Greear found more cfg80211/mac80211 state confusion
* and a fix for another bug found by Jouni: local state changes need to be
reported by mac80211 to cfg80211 so it disconnects properly."
And for the iwlwifi bits, he says:
"I have fixes for a firmware crash during resume, multicast RX,
aggregation and a workaround for a firmware scanning bug."
Along with those...
Albert Pool adds a USB ID to the rtl8192cu driver.
Arend van Spriel restores a driver option support flag that had been
removed from 3.9 due to a bug in that version of the driver.
Felix Fietkau fixes a trio of ath9k issues with a series of small
patches.
Geert Uytterhoeven provides a Kconfig fix for ath9k (which you also
merged, so it isn't in the diff here).
Larry Finger gives us a fix for a build warning on big-endian systems
for rtlwifi.
Rafał Miłecki adds some core IDs to the bcma driver.
Sujith Manoharan fixes a module unloading crash in ath9k, and corrects
some calibration settings for AR9485.
====================
This patch cures transmit timeout's with DHCP observed
while running under KVM. When the transmit ring is cleaned out,
the Byte Queue Limit values need to be reset.
Eric Dumazet [Mon, 20 May 2013 06:52:26 +0000 (06:52 +0000)]
tcp: md5: remove spinlock usage in fast path
TCP md5 code uses per cpu variables but protects access to them with
a shared spinlock, which is a contention point.
[ tcp_md5sig_pool_lock is locked twice per incoming packet ]
Makes things much simpler, by allocating crypto structures once, first
time a socket needs md5 keys, and not deallocating them as they are
really small.
Next step would be to allow crypto allocations being done in a NUMA
aware way.
Willem de Bruijn [Mon, 20 May 2013 04:02:32 +0000 (04:02 +0000)]
rps: selective flow shedding during softnet overflow
A cpu executing the network receive path sheds packets when its input
queue grows to netdev_max_backlog. A single high rate flow (such as a
spoofed source DoS) can exceed a single cpu processing rate and will
degrade throughput of other flows hashed onto the same cpu.
This patch adds a more fine grained hashtable. If the netdev backlog
is above a threshold, IRQ cpus track the ratio of total traffic of
each flow (using 4096 buckets, configurable). The ratio is measured
by counting the number of packets per flow over the last 256 packets
from the source cpu. Any flow that occupies a large fraction of this
(set at 50%) will see packet drop while above the threshold.
Tested:
Setup is a muli-threaded UDP echo server with network rx IRQ on cpu0,
kernel receive (RPS) on cpu0 and application threads on cpus 2--7
each handling 20k req/s. Throughput halves when hit with a 400 kpps
antagonist storm. With this patch applied, antagonist overload is
dropped and the server processes its complete load.
The patch is effective when kernel receive processing is the
bottleneck. The above RPS scenario is a extreme, but the same is
reached with RFS and sufficient kernel processing (iptables, packet
socket tap, ..).
Fabio Estevam [Mon, 20 May 2013 03:06:17 +0000 (03:06 +0000)]
fec: Let device core handle pinctrl
Since commit ab78029 (drivers/pinctrl: grab default handles from device core)
we can rely on device core for handling pinctrl, so remove
devm_pinctrl_get_select_default() from the driver.
Petko Manolov [Sun, 19 May 2013 23:08:47 +0000 (23:08 +0000)]
drivers: net: usb: rtl8150: concurrent URB bugfix
This patch fixes a potential race with concurrently running asynchronous
write requests. The values for device's RX control register are now
stored in dynamically allocated buffers so each URB submission has it's
own copy. Doing it the old way is data clobbering prone.
Tony Prisk [Sat, 18 May 2013 09:39:06 +0000 (09:39 +0000)]
net: velocity: Convert to generic dma functions
Remove the pci_* dma functions and replace with the more generic
versions.
In preparation of adding platform support, a new struct device *dev
is added to struct velocity_info which can be used by both the pci
and platform code.
Dan Carpenter [Sun, 19 May 2013 08:36:36 +0000 (08:36 +0000)]
isdn/kcapi: fix a small underflow
In get_capi_ctr_by_nr() and get_capi_appl_by_nr() the parameter comes
from skb->data. The current code can underflow to one space before the
start of the array.
The sanity check isn't needed in __get_capi_appl_by_nr() but I changed
it to match the others.
Wang YanQing [Thu, 9 May 2013 06:16:47 +0000 (14:16 +0800)]
TTY: Fix tty miss restart after we turn off flow-control
I meet emacs hang in start if I do the operation below:
1: echo 3 > /proc/sys/vm/drop_caches
2: emacs BigFile
3: Press CTRL-S follow 2 immediately
Then emacs hang on, CTRL-Q can't resume, the terminal
hang on, you can do nothing with this terminal except
close it.
The reason is before emacs takeover control the tty,
we use CTRL-S to XOFF it. Then when emacs takeover the
control, it may don't use the flow-control, so emacs hang.
This patch fix it.
This patch will fix a kind of strange tty relation hang problem,
I believe I meet it with vim in ssh, and also see below bug report:
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=465823
Peter Hurley [Fri, 17 May 2013 16:41:03 +0000 (12:41 -0400)]
tty/vt: Fix vc_deallocate() lock order
Now that the tty port owns the flip buffers and i/o is allowed
from the driver even when no tty is attached, the destruction
of the tty port (and the flip buffers) must ensure that no
outstanding work is pending.
Unfortunately, this creates a lock order problem with the
console_lock (see attached lockdep report [1] below).
For single console deallocation, drop the console_lock prior
to port destruction. When multiple console deallocation,
defer port destruction until the consoles have been
deallocated.
tty_port_destroy() is not required if the port has not
been used; remove from vc_allocate() failure path.
Commit "TTY: rocket, fix compilation warning" fixed a compilation
warning, but there was still a problem with !CONFIG_PCI configs. So
fix them for good by coupling the PCI functions together and moving
them inside a common #ifdef.
Lee Jones [Thu, 9 May 2013 12:50:55 +0000 (13:50 +0100)]
serial: pl011: protect attribute read from NULL platform data struct
It's completely feasible that platform data will be empty i.e. when
booting with Device Tree with no device AUXDATA. So we must protect
it's use in these use-cases, or risk a kernel Oops.
Ezequiel Garcia [Tue, 7 May 2013 11:27:16 +0000 (08:27 -0300)]
serial: 8250_dw: Add valid clk pointer check
Commit ffc3ae6dd "serial: 8250_dw: Enable runtime PM" introduced runtime
PM management, which enables/disables the clk without checking if the clk
is valid. However, this driver allows to be probed without a defined clk,
using clock-frequency, as a fallback.
Therefore, on platforms that are device tree probed using clock-frequency
instead of clk, we get an ugly NULL pointer dereference.
This patch fixes it by simply adding a check before accessing the clk api.
Linus Torvalds [Mon, 20 May 2013 18:36:52 +0000 (11:36 -0700)]
Merge tag 'hwmon-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging
Pull hwmon fixes from Guenter Roeck:
- Error path fixes for abituguru and iio_hwmon drivers.
- Drop erroneously created attributes from nct6775 driver.
- Drop redundant safety on cache lifetime for tmp401 driver.
- Add explicit maintainer for LM95234 and TMP401 drivers.
* tag 'hwmon-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging:
MAINTAINERS: Add myself as maintainer for LM95234 and TMP401 drivers
hwmon: (tmp401) Drop redundant safety on cache lifetime
hwmon: fix error return code in abituguru_probe()
hwmon: (iio_hwmon) Fix null pointer dereference
hwmon: (nct6775) Do not create non-existing attributes
hwmon: (iio_hwmon) Fix missing iio_channel_release_all call if devm_kzalloc fail
Linus Torvalds [Mon, 20 May 2013 18:36:03 +0000 (11:36 -0700)]
x86: Fix bit corruption at CPU resume time
In commit 78d77df71510 ("x86-64, init: Do not set NX bits on non-NX
capable hardware") we added the early_pmd_flags that gets the NX bit set
when a CPU supports NX. However, the new variable was marked __initdata,
because the main _use_ of this is in an __init routine.
However, the bit setting happens from secondary_startup_64(), which is
called not only at bootup, but on every secondary CPU start. Including
resuming from STR and at CPU hotplug time. So the value cannot be
__initdata.
Dan Carpenter [Sun, 19 May 2013 18:52:20 +0000 (21:52 +0300)]
USB: cxacru: potential underflow in cxacru_cm_get_array()
The value of "offd" comes off the instance->rcv_buf[] and we used it as
the offset into an array. The problem is that we check the upper bound
but not for negative values.
Maxime Ripard [Mon, 29 Apr 2013 14:07:18 +0000 (16:07 +0200)]
gpio: mxs: Use set and clear capabilities of the gpio controller
The current driver doesn't use the set and clear registers found on the
mxs gpio controller.
This leads the generic gpio controller to be using some internal value
to avoid looking up the value stored in the registers, making it behave
pretty much like a cache.
This raises some coherency problem when a gpio is not modified by the
gpio controller, while it can easily be fixed by using the set and clear
registers.
David Cohen [Mon, 6 May 2013 23:11:03 +0000 (16:11 -0700)]
gpio-langwell: fix irq conflicts when DT is not used
When DT is not used IOAPIC does not register irq domain. As result
IOAPIC won't care about gpio-langwell's virq and may cause conflict if
an irq number is equal to gpio-langwell's virq mapped beforehand.
This patch tells gpio_langwell to not ignore irq_base if != 0 and
preferably use it to avoid the conflict.
Axel Lin [Sun, 19 May 2013 11:11:54 +0000 (19:11 +0800)]
gpio: mcp23s08: Fix build error when CONFIG_SPI_MASTER=y && CONFIG_I2C=m
This patch fixes below build error when CONFIG_SPI_MASTER=y && CONFIG_I2C=m &&
CONFIG_GPIO_MCP23S08=y.
LD init/built-in.o
drivers/built-in.o: In function `mcp23017_write':
clkdev.c:(.text+0x1e14): undefined reference to `i2c_smbus_write_word_data'
drivers/built-in.o: In function `mcp23017_read':
clkdev.c:(.text+0x1e24): undefined reference to `i2c_smbus_read_word_data'
drivers/built-in.o: In function `mcp23008_write':
clkdev.c:(.text+0x1e8c): undefined reference to `i2c_smbus_write_byte_data'
drivers/built-in.o: In function `mcp23008_read':
clkdev.c:(.text+0x1e98): undefined reference to `i2c_smbus_read_byte_data'
drivers/built-in.o: In function `mcp23008_read_regs':
clkdev.c:(.text+0x1ed0): undefined reference to `i2c_smbus_read_byte_data'
drivers/built-in.o: In function `mcp23s08_init':
clkdev.c:(.init.text+0x30): undefined reference to `i2c_register_driver'
drivers/built-in.o: In function `mcp23s08_exit':
clkdev.c:(.exit.text+0x30): undefined reference to `i2c_del_driver'
make: *** [vmlinux] Error 1
When CONFIG_I2C=m, meaning we can't build the drivers in with I2C support.
Thus don't allow the drivers to be compiled as built-in when CONFIG_I2C=m.
The real fix though is to break the driver apart into a SPI part, an I2C part
and a common part. But that's something for 3.11 while this is something for
3.10/stable.
Linus Torvalds [Mon, 20 May 2013 14:59:46 +0000 (07:59 -0700)]
Merge tag 'pinctrl-fixes-v3.10-2' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl
Pull pinctrl fixes from Linus Walleij:
- Three fixes to make the boot path for device tree work properly on
the Nomadik pin controller.
- Compile warning fix for the vt8500 driver.
- Fix error path in pinctrl-single.
- Free mappings in error path of the Lantiq controller.
- Documentation fixes.
* tag 'pinctrl-fixes-v3.10-2' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl:
pinctrl/lantiq: Free mapping configs for both pin and groups
pinctrl: single: fix error return code in pcs_parse_one_pinctrl_entry()
pinctrl: generic: Fix typos and clarify comments
pinctrl: vt8500: Fix incorrect data in WM8750 pinctrl table
pinctrl: abx500: Rejiggle platform data and DT initialisation
pinctrl: abx500: Specify failed sub-driver by ID instead of driver_data
Guenter Roeck [Mon, 20 May 2013 03:44:27 +0000 (20:44 -0700)]
MAINTAINERS: Add myself as maintainer for LM95234 and TMP401 drivers
I wrote the LM95234 driver and extended the TMP401 driver substantially,
and I have hardware to test both, so it makes sense to explicitly
maintain them.
Yuval Mintz [Sun, 19 May 2013 04:41:01 +0000 (04:41 +0000)]
bnx2x: Fix GSO for 57710/57711 chips
Starting with commit 91226790bbe2dbfbba48dd79d49f2b38ef10eb97
`bnx2x: use FW 7.8.17', the bnx2x driver no longer requests the FW to perform
IP checksums for IPv4 packets.
This behaviour needs to be revised for 57710/57711 chips -
when using GSO, if the driver will not set the IP checksum flag then packets
will be transmitted by the chip without a valid IP checksum, resulting in
a drop of all such packets on the receiver-side.
David S. Miller [Mon, 20 May 2013 07:13:54 +0000 (00:13 -0700)]
Merge branch 'tg3_eee'
Nithin Nayak Sujir says:
====================
This series adds support for modifying EEE settings via ethtool. Since this can
impact Link Flap Avoidance, the driver pulls the current hardware settings if
LFA is enabled. This is similar to how we do the link settings to avoid a flap.
v2: Fixes pointed out by Ben Hutchings.
- Use MDIO_AN_EEE_LPABLE to set the lp_advertised field.
- Check that tx_lpi_timer is within valid range.
====================
Nithin Sujir [Sat, 18 May 2013 06:26:54 +0000 (06:26 +0000)]
tg3: Simplify tg3_phy_eee_config_ok() by reusing tg3_eee_pull_config()
eee_config_ok() was checking only for mismatch in advertised settings.
This patch expands the scope of eee_config_ok() to check for mismatch in
the other eee settings. On mismatch we will require a call to
tg3_setup_eee() to push the configured settings to the hardware.
Nithin Sujir [Sat, 18 May 2013 06:26:53 +0000 (06:26 +0000)]
tg3: Add tg3_eee_pull_config() function
Add tg3_eee_pull_config() to pull the settings from the hardware and
populate the eee structure.
If Link Flap Avoidance is enabled, we pull the eee settings from the hw
so as not to cause a phy reset on eee config mismatch later. This
requires moving down tg3_setup_eee() below the tg3_pull_config() to not
trample existing settings.
Nithin Sujir [Sat, 18 May 2013 06:26:52 +0000 (06:26 +0000)]
tg3: Add ethtool_eee struct and tg3_setup_eee()
Add an eee structure and update it with eee settings. This will be used
for set/get_eee operations. Add common function tg3_setup_eee() that
will be used in the subsequent patches.
françois romieu [Sat, 18 May 2013 01:24:46 +0000 (01:24 +0000)]
r8169: fix offloaded tx checksum for small packets.
8168evl offloaded checksums are wrong since commit e5195c1f31f399289347e043d6abf3ffa80f0005 ("r8169: fix 8168evl frame padding.")
pads small packets to 60 bytes (without ethernet checksum). Typical symptoms
appear as UDP checksums which are wrong by the count of added bytes.
It isn't worth compensating. Let the driver checksum.
Due to the skb length changes, TSO code is moved before the Tx descriptor gets
written.
Eric Dumazet [Fri, 17 May 2013 16:37:03 +0000 (16:37 +0000)]
x86: bpf_jit_comp: secure bpf jit against spraying attacks
hpa bringed into my attention some security related issues
with BPF JIT on x86.
This patch makes sure the bpf generated code is marked read only,
as other kernel text sections.
It also splits the unused space (we vmalloc() and only use a fraction of
the page) in two parts, so that the generated bpf code not starts at a
known offset in the page, but a pseudo random one.
Yuchung Cheng [Fri, 17 May 2013 13:45:05 +0000 (13:45 +0000)]
tcp: remove bad timeout logic in fast recovery
tcp_timeout_skb() was intended to trigger fast recovery on timeout,
unfortunately in reality it often causes spurious retransmission
storms during fast recovery. The particular sign is a fast retransmit
over the highest sacked sequence (SND.FACK).
Currently the RTO timer re-arming (as in RFC6298) offers a nice cushion
to avoid spurious timeout: when SND.UNA advances the sender re-arms
RTO and extends the timeout by icsk_rto. The sender does not offset
the time elapsed since the packet at SND.UNA was sent.
But if the next (DUP)ACK arrives later than ~RTTVAR and triggers
tcp_fastretrans_alert(), then tcp_timeout_skb() will mark any packet
sent before the icsk_rto interval lost, including one that's above the
highest sacked sequence. Most likely a large part of scorebard will be
marked.
If most packets are not lost then the subsequent DUPACKs with new SACK
blocks will cause the sender to continue to retransmit packets beyond
SND.FACK spuriously. Even if only one packet is lost the sender may
falsely retransmit almost the entire window.
The situation becomes common in the world of bufferbloat: the RTT
continues to grow as the queue builds up but RTTVAR remains small and
close to the minimum 200ms. If a data packet is lost and the DUPACK
triggered by the next data packet is slightly delayed, then a spurious
retransmission storm forms.
As the original comment on tcp_timeout_skb() suggests: the usefulness
of this feature is questionable. It also wastes cycles walking the
sack scoreboard and is actually harmful because of false recovery.
When bond_3ad_get_active_agg_info() is used in all show_ad_ functions
it is not protected against slave manipulation and since it walks over
the slaves and uses them, this can easily result in NULL pointer
dereference or use of freed memory. Both the new wrapper and the
internal function are exported to the bonding as they're needed in
different places.
bonding: arp_ip_count and arp_targets can be wrong
When getting arp_ip_targets if we encounter a bad IP, arp_ip_count still
gets increased and all the targets after the wrong one will not be probed
if arp_interval is enabled after that (unless a new IP target is added
through sysfs) because of the zero entry, in this case reading
arp_ip_target through sysfs will show valid targets even if there's a
zero entry.
Example: 1.2.3.4,4.5.6.7,blah,5.6.7.8
When retrieving the list from arp_ip_target the output would be:
1.2.3.4,4.5.6.7,5.6.7.8
but there will be a 0 entry between 4.5.6.7 and 5.6.7.8. If arp_interval
is enabled after that 5.6.7.8 will never be checked because of that.
Changing the mode without any locking can result in multiple races (e.g.
upping a bond, enslaving/releasing). Depending on which race is hit the
impact can vary from incosistent bond state to kernel crash.
Use RTNL to synchronize the mode setting with the dangerous races.
Dave Airlie [Mon, 20 May 2013 03:31:36 +0000 (13:31 +1000)]
Merge branch 'drm-nouveau-fixes-3.10' of git://anongit.freedesktop.org/git/nouveau/linux-2.6 into drm-next
Fixes some s/r problem with copy engines and ZCULL issues and playlist issues
* 'drm-nouveau-fixes-3.10' of git://anongit.freedesktop.org/git/nouveau/linux-2.6:
drm/nouveau: ensure channels are stopped before saving fences for suspend
drm/nv50/fifo: prevent races between clients updating playlists
drm/nvc0/fifo: prevent CHAN_TABLE_ERROR:CHANNEL_PENDING on fifo fini
drm/nvc0/fifo: prevent races between clients updating playlists
drm/nve0/fifo: prevent races between clients updating playlists
drm/nve0/ltcg: poke the partition count into yet another register
drm/nvc0/ltcg: fix handling of disabled partitions
drm/nvc0/ce: disable ce1 on a number of chipsets
drm/nouveau/bios: fix thinko in ZM_MASK_ADD opcode
drm/nouveau: fix build with nv50->nvc0
Ben Skeggs [Mon, 13 May 2013 06:11:12 +0000 (16:11 +1000)]
drm/nvc0/ce: disable ce1 on a number of chipsets
The falcon is present, but the rest of the copy engine doesn't appear to
be... PUNITS doesn't report disabled (maybe the bits for the copy engines
got added later?), so we end up trying to use a non-functional CE1, and
bust all sorts of things.. Most notably, suspend/resume..
projects / linux.git / log