| Age | Commit message (Collapse) | Author |
|---|
|
Jesper Dangaard Brouer says:
====================
net: mitigating kmem_cache free slowpath
This patchset is the first real use-case for kmem_cache bulk _free_.
The use of bulk _alloc_ is NOT included in this patchset. The full use
have previously been posted here [1].
The bulk free side have the largest benefit for the network stack
use-case, because network stack is hitting the kmem_cache/SLUB
slowpath when freeing SKBs, due to the amount of outstanding SKBs.
This is solved by using the new API kmem_cache_free_bulk().
Introduce new API napi_consume_skb(), that hides/handles bulk freeing
for the caller. The drivers simply need to use this call when freeing
SKBs in NAPI context, e.g. replacing their calles to dev_kfree_skb() /
dev_consume_skb_any().
Driver ixgbe is the first user of this new API.
[1] http://thread.gmane.org/gmane.linux.network/384302/focus=397373
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
There is an opportunity to bulk free SKBs during reclaiming of
resources after DMA transmit completes in ixgbe_clean_tx_irq. Thus,
bulk freeing at this point does not introduce any added latency.
Simply use napi_consume_skb() which were recently introduced. The
napi_budget parameter is needed by napi_consume_skb() to detect if it
is called from netpoll.
Benchmarking IPv4-forwarding, on CPU i7-4790K @4.2GHz (no turbo boost)
Single CPU/flow numbers: before: 1982144 pps -> after : 2064446 pps
Improvement: +82302 pps, -20 nanosec, +4.1%
(SLUB and GCC version 5.1.1 20150618 (Red Hat 5.1.1-4))
Joint work with Alexander Duyck.
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The network stack defers SKBs free, in-case free happens in IRQ or
when IRQs are disabled. This happens in __dev_kfree_skb_irq() that
writes SKBs that were free'ed during IRQ to the softirq completion
queue (softnet_data.completion_queue).
These SKBs are naturally delayed, and cleaned up during NET_TX_SOFTIRQ
in function net_tx_action(). Take advantage of this a use the skb
defer and flush API, as we are already in softirq context.
For modern drivers this rarely happens. Although most drivers do call
dev_kfree_skb_any(), which detects the situation and calls
__dev_kfree_skb_irq() when needed. This due to netpoll can call from
IRQ context.
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Discovered that network stack were hitting the kmem_cache/SLUB
slowpath when freeing SKBs. Doing bulk free with kmem_cache_free_bulk
can speedup this slowpath.
NAPI context is a bit special, lets take advantage of that for bulk
free'ing SKBs.
In NAPI context we are running in softirq, which gives us certain
protection. A softirq can run on several CPUs at once. BUT the
important part is a softirq will never preempt another softirq running
on the same CPU. This gives us the opportunity to access per-cpu
variables in softirq context.
Extend napi_alloc_cache (before only contained page_frag_cache) to be
a struct with a small array based stack for holding SKBs. Introduce a
SKB defer and flush API for accessing this.
Introduce napi_consume_skb() as replacement for e.g. dev_consume_skb_any()
when running in NAPI context. A small trick to handle/detect if we
are called from netpoll is to see if budget is 0. In that case, we
need to invoke dev_consume_skb_irq().
Joint work with Alexander Duyck.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Nikolay Aleksandrov says:
====================
virtio_net: better ethtool setting validation
This small set is a follow-up for the recent patches that added ethtool
get/set settings. Patch 1 changes the speed validation routine to check
if the speed is between 0 and INT_MAX (or SPEED_UNKNOWN) and patch 2 adds
port validation to virtio_net and better validation comment.
This set is on top of Michael's patch which explains that speeds from 0
to INT_MAX are valid:
http://patchwork.ozlabs.org/patch/578911/
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
We should validate the port setting that we got from the user and check
if it's what we've set it to (PORT_OTHER), also add explanation that
ignoring advertising is good as long as we don't have autonegotiation.
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Devices these days can have any speed and as was recently pointed out
any speed from 0 to INT_MAX is valid so adjust speed validation to
accept such values.
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Andrew F. Davis says:
====================
net: phy: dp83848: Add support for TI TLK10x Ethernet PHYs
This series is [0] split into its logical components.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Andrew F. Davis <afd@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The TI TLK10x Ethernet PHYs are similar in the interrupt relevant
registers and so are compatible with the DP83848x devices already
supported.
Signed-off-by: Andrew F. Davis <afd@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Reorganize code by moving the desired interrupt mask definition
out of function. Also rearrange the enable/disable interrupt function
to prevent accidental over-writing of values in registers.
Signed-off-by: Andrew F. Davis <afd@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
After acquiring National Semiconductor, TI appears to have
changed the Vendor Model Number for the DP83848C PHYs,
add this new ID to supported IDs.
Signed-off-by: Andrew F. Davis <afd@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add a helper macro for defining dp83848 compatible phy devices.
Update copyright info.
Signed-off-by: Andrew F. Davis <afd@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The EVB (virtual bridge) functionality should be disabled on older BE3
and Lancer chips if SR-IOV is disabled in the NIC's BIOS. This setting
is identified by the zero value of total VFs reported by the card.
The GET_HSW_CONFIG command cannot be used as it is not supported by
these older chipset's FW.
v2: added the comment
Cc: Sathya Perla <sathya.perla@broadcom.com>
Cc: Ajit Khaparde <ajit.khaparde@broadcom.com>
Cc: Padmanabh Ratnakar <padmanabh.ratnakar@broadcom.com>
Cc: Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com>
Cc: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Ivan Vecera <ivecera@redhat.com>
Acked-by: Sathya Perla <sathya.perla@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Helmut Buchsbaum says:
====================
Add support for MICREL KSZ8795CLX 5-port switch
This patch series refactors the spi-ks8995 driver to finally add support
for the MICREL KSZ8795CLX. Additionally support for controlling a GPIO
line for resetting the switch is added.
Helmut
Changes since v2:
- use GPIO_ACTIVE_LOW according to Andrew's remark.
- use ePAPR compliant node name in example, thanks to Sergei for
pointing out
Changes since v1:
- removed initializing registers from Device Tree following Florian's
advice
- fixed GPIO handling for reset according to Andrew's remark.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Helmut Buchsbaum <helmut.buchsbaum@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add support for MICREL KSZ8795CLX Integrated 5-Port, 10-/100-Managed
Ethernet Switch with Gigabit GMII/RGMII and MII/RMII interfaces.
Signed-off-by: Helmut Buchsbaum <helmut.buchsbaum@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Prepare creating SPI reads and writes for other switch families.
The KS8995 family uses the straight forward
<8bit CMD><8bit ADDR>
sequence.
To be able to support KSZ8795 family, which uses
<3bit CMD><12bit ADDR><1 bit TR>
make the SPI command creation chip variant dependent.
Signed-off-by: Helmut Buchsbaum <helmut.buchsbaum@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When using device tree it is no more possible to reset the PHY at board
level. Furthermore, doing in the driver allows to power down the switch
when it is not used any more.
The patch introduces a new optional property "reset-gpios" denoting an
appropriate GPIO handle, e.g.:
reset-gpios = <&gpio0 46 GPIO_ACTIVE_LOW>
Signed-off-by: Helmut Buchsbaum <helmut.buchsbaum@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Since the chip variant is now determined by spi_device_id, verify
family and chip id and determine the revision id.
Signed-off-by: Helmut Buchsbaum <helmut.buchsbaum@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Refactor to use spi_device_id table to facilitate easy
extendability.
Signed-off-by: Helmut Buchsbaum <helmut.buchsbaum@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Sunil Goutham says:
====================
net: thunderx: Setting IRQ affinity hints and other optimizations
This patch series contains changes
- To add support for virtual function's irq affinity hint
- Replace napi_schedule() with napi_schedule_irqoff()
- Reduce page allocation overhead by allocating pages
of higher order when pagesize is 4KB.
- Add couple of stats which helps in debugging
- Some miscellaneous changes to BGX driver.
Changes from v1:
- As suggested changed MAC address invalid log message
to dev_err() instead of dev_warn().
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Allocate higher order pages when pagesize is small, this will
reduce number of calls to page allocator and wastage of memory.
Signed-off-by: Sunil Goutham <sgoutham@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Robert Richter <rrichter@cavium.com>
Signed-off-by: Sunil Goutham <sgoutham@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
In the case of OF device tree, the firmware information is attached to
the BGX device structure in the standard manner, so use the firmware
iterators and accessors where possible.
Signed-off-by: David Daney <david.daney@cavium.com>
Signed-off-by: Sunil Goutham <sgoutham@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This affinity hint can be used by user space irqbalance tool to set
preferred CPU mask for irqs registered by this VF. Irqbalance needs
to be in 'exact' mode to set irq affinity same as indicated by
affinity hint.
Signed-off-by: Sunil Goutham <sgoutham@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
napi_schedule is being called from hard irq context, hence
switch to napi_schedule_irqoff which avoids unneeded call
to local_irq_save and local_irq_restore.
Signed-off-by: Sunil Goutham <sgoutham@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When system is low on atomic memory, too many error messages are logged.
Since this is not a total failure but a simple switch to non-atomic allocation
better to have a stat.
Also add a stat for reset, kicked due to transmit watchdog timeout.
Signed-off-by: Thanneeru Srinivasulu <tsrinivasulu@caviumnetworks.com>
Signed-off-by: Sunil Goutham <sgoutham@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
When checking specific attribute from a bit mask, need to use bitwise
AND and not logical AND, fixed that.
Fixes: 145d9c541032 ('IB/core: Display extended counter set if
available')
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Matan Barak <matanb@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
|
|
The while loop after err_slaves should use post-decrement; otherwise
we'll fail to do the kfrees for i==0, and will run into out-of-bounds
accesses if the setup above failed already at i==0.
[I'm not sure why one even bothers populating the ->vlan_filter array:
mlx4.h isn't #included by anything outside
drivers/net/ethernet/mellanox/mlx4/, and "git grep -C2 -w vlan_filter
drivers/net/ethernet/mellanox/mlx4/" seems to suggest that the
vlan_filter elements aren't used at all.]
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Doug Ledford <dledford@redhat.com>
|
|
If the LightNVM subsystem is not compiled into the kernel, and the
null_blk device driver requests lightnvm to be initialized. The call to
nvm_register fails and the null_add_dev function cleans up the
initialization. However, at this point the null block device has
already been added to the nullb_list and thus a second cleanup will
occur when the function has returned, that leads to a double call to
blk_cleanup_queue.
Signed-off-by: Matias Bjørling <m@bjorling.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
|
|
This reverts commit 829b6962f7e3cfc06f7c5c26269fd47ad48cf503.
Revert this change as it causes a sysfs path to change and therefore
introduces and ABI regression. More precisely Android's vold is not being
able to access /sys/module/mmcblk/parameters/perdev_minors any more, since
the path becomes changed to: "/sys/module/mmc_block/..."
Fixes: 829b6962f7e3 ("mmc: block: don't use parameter prefix if built as
module")
Reported-by: John Stultz <john.stultz@linaro.org>
Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
|
|
In __cpufreq_cooling_register() we allocate the arrays for time_in_idle
and time_in_idle_timestamp to be as big as the number of cpus in this
cpufreq device. However, in get_load() we access this array using the
cpu number as index, which can result in an out of bound access.
Index time_in_idle{,_timestamp} using the index in the cpufreq_device's
allowed_cpus mask, as we do for the load_cpu array in
cpufreq_get_requested_power()
Reported-by: Nicolas Boichat <drinkcat@chromium.org>
Cc: Amit Daniel Kachhap <amit.kachhap@gmail.com>
Cc: Zhang Rui <rui.zhang@intel.com>
Cc: Eduardo Valentin <edubezval@gmail.com>
Tested-by: Nicolas Boichat <drinkcat@chromium.org>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Javi Merino <javi.merino@arm.com>
Signed-off-by: Eduardo Valentin <edubezval@gmail.com>
|
|
The value of ctx->pos in the last readdir call is supposed to be set to
INT_MAX due to 32bit compatibility, unless 'pos' is intentially set to a
larger value, then it's LLONG_MAX.
There's a report from PaX SIZE_OVERFLOW plugin that "ctx->pos++"
overflows (https://forums.grsecurity.net/viewtopic.php?f=1&t=4284), on a
64bit arch, where the value is 0x7fffffffffffffff ie. LLONG_MAX before
the increment.
We can get to that situation like that:
* emit all regular readdir entries
* still in the same call to readdir, bump the last pos to INT_MAX
* next call to readdir will not emit any entries, but will reach the
bump code again, finds pos to be INT_MAX and sets it to LLONG_MAX
Normally this is not a problem, but if we call readdir again, we'll find
'pos' set to LLONG_MAX and the unconditional increment will overflow.
The report from Victor at
(http://thread.gmane.org/gmane.comp.file-systems.btrfs/49500) with debugging
print shows that pattern:
Overflow: e
Overflow: 7fffffff
Overflow: 7fffffffffffffff
PAX: size overflow detected in function btrfs_real_readdir
fs/btrfs/inode.c:5760 cicus.935_282 max, count: 9, decl: pos; num: 0;
context: dir_context;
CPU: 0 PID: 2630 Comm: polkitd Not tainted 4.2.3-grsec #1
Hardware name: Gigabyte Technology Co., Ltd. H81ND2H/H81ND2H, BIOS F3 08/11/2015
ffffffff81901608 0000000000000000 ffffffff819015e6 ffffc90004973d48
ffffffff81742f0f 0000000000000007 ffffffff81901608 ffffc90004973d78
ffffffff811cb706 0000000000000000 ffff8800d47359e0 ffffc90004973ed8
Call Trace:
[<ffffffff81742f0f>] dump_stack+0x4c/0x7f
[<ffffffff811cb706>] report_size_overflow+0x36/0x40
[<ffffffff812ef0bc>] btrfs_real_readdir+0x69c/0x6d0
[<ffffffff811dafc8>] iterate_dir+0xa8/0x150
[<ffffffff811e6d8d>] ? __fget_light+0x2d/0x70
[<ffffffff811dba3a>] SyS_getdents+0xba/0x1c0
Overflow: 1a
[<ffffffff811db070>] ? iterate_dir+0x150/0x150
[<ffffffff81749b69>] entry_SYSCALL_64_fastpath+0x12/0x83
The jump from 7fffffff to 7fffffffffffffff happens when new dir entries
are not yet synced and are processed from the delayed list. Then the code
could go to the bump section again even though it might not emit any new
dir entries from the delayed list.
The fix avoids entering the "bump" section again once we've finished
emitting the entries, both for synced and delayed entries.
References: https://forums.grsecurity.net/viewtopic.php?f=1&t=4284
Reported-by: Victor <services@swwu.com>
CC: stable@vger.kernel.org
Signed-off-by: David Sterba <dsterba@suse.com>
Tested-by: Holger Hoffstätte <holger.hoffstaette@googlemail.com>
Signed-off-by: Chris Mason <clm@fb.com>
|
|
Nikolay Borisov says:
====================
Make igmp sysctl knobs namespace aware
This series continue making more of the net related sysctls
namespace aware. The first 2 and last patches are straight
forward and convert sysctls which weren't defined to be
namespace aware. The only thing in them is that each removes
a define which is used in only one place (to initialise
the respective sysctl) so I don't think this is a huge loss.
The third patch however, converts igmp_llm_reports which was
already defined in the ipv4_net_table but wasn't using any of
the net namespace infrastructure.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Nikolay Borisov <kernel@kyup.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
This was initially introduced in df2cf4a78e488d26 ("IGMP: Inhibit
reports for local multicast groups") by defining the sysctl in the
ipv4_net_table array, however it was never implemented to be
namespace aware. Fix this by changing the code accordingly.
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Nikolay Borisov <kernel@kyup.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Nikolay Borisov <kernel@kyup.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Replace 'goto' with 'return' to remove unnecessary check at label:
err_undo_flags.
The reason is that 'err_undo_flags' do two things for the first slave device:
1.revert bond mac address if it is set by the slave device.
2.revert bond device type if it's not ARPHRD_ETHER.
It's not necessary for the following three places, they changed neither bond
mac address nor type. It's straightforward to return directly.
Signed-off-by: Zhang Shengju <zhangshengju@cmss.chinamobile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Tested on Acqua A5 SoM (http://www.acmesystems.it/acqua).
Signed-off-by: Sergio Prado <sergio.prado@e-labworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
32 bit systems using 'struct timeval' will break in the year 2038, so
we replace the code appropriately. However, this driver is not broken
in 2038 since we are only using microseconds portion of the time.
This patch replaces 'struct timeval' with 'struct timespec64'. We only
need to find elapsed microseconds rather than absolute time, so it's
better to use monotonic time, so using ktime_get_ts64() makes the code
more efficient and more robust against concurrent settimeofday()
calls.
Signed-off-by: Amitoj Kaur Chawla <amitoj1606@gmail.com>
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Thomas Sailer <t.sailer@alumni.ethz.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Operations with the GENL_ADMIN_PERM flag fail permissions checks because
this flag means we call netlink_capable, which uses the init user ns.
Instead, let's introduce a new flag, GENL_UNS_ADMIN_PERM for operations
which should be allowed inside a user namespace.
The motivation for this is to be able to run openvswitch in unprivileged
containers. I've tested this and it seems to work, but I really have no
idea about the security consequences of this patch, so thoughts would be
much appreciated.
v2: use the GENL_UNS_ADMIN_PERM flag instead of a check in each function
v3: use separate ifs for UNS_ADMIN_PERM and ADMIN_PERM, instead of one
massive one
Reported-by: James Page <james.page@canonical.com>
Signed-off-by: Tycho Andersen <tycho.andersen@canonical.com>
CC: Eric Biederman <ebiederm@xmission.com>
CC: Pravin Shelar <pshelar@ovn.org>
CC: Justin Pettit <jpettit@nicira.com>
CC: "David S. Miller" <davem@davemloft.net>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Many virtual and not quite virtual devices allow any speed to be set
through ethtool. In particular, this applies to the virtio-net devices.
Document this fact to make sure people don't assume the enum lists all
possible values. Reserve values greater than INT_MAX for future
extension and to avoid conflict with SPEED_UNKNOWN.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Duplicate include detected.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The return value of kzalloc on failure of allocation of memory should
be -ENOMEM and not -1.
Found using Coccinelle. A simplified version of the semantic patch
used is:
//<smpl>
@@
expression *e;
@@
e = kzalloc(...);
if (e == NULL) {
...
return
- -1
+ -ENOMEM
;
}
//</smpl>
The single call site only checks that the return value is not 0,
hence no change is required at the call site.
Signed-off-by: Amitoj Kaur Chawla <amitoj1606@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Since the dawn of time the ICST code has only supported divide
by one or hang in an eternal loop. Luckily we were always dividing
by one because the reference frequency for the systems using
the ICSTs is 24MHz and the [min,max] values for the PLL input
if [10,320] MHz for ICST307 and [6,200] for ICST525, so the loop
will always terminate immediately without assigning any divisor
for the reference frequency.
But for the code to make sense, let's insert the missing i++
Reported-by: David Binderman <dcb314@hotmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
|
|
Alexander Duyck says:
====================
Add GSO support for outer checksum w/ inner checksum offloads
This patch series updates the existing segmentation offload code for
tunnels to make better use of existing and updated GSO checksum
computation. This is done primarily through two mechanisms. First we
maintain a separate checksum in the GSO context block of the sk_buff. This
allows us to maintain two checksum values, one offloaded with values stored
in csum_start and csum_offset, and one computed and tracked in
SKB_GSO_CB(skb)->csum. By maintaining these two values we are able to take
advantage of the same sort of math used in local checksum offload so that
we can provide both inner and outer checksums with minimal overhead.
Below is the performance for a netperf session between an ixgbe PF and VF
on the same host but in different namespaces. As can be seen a significant
gain in performance can be had from allowing the use of Tx checksum offload
on the inner headers while performing a software offload on the outer
header computation:
Recv Send Send Utilization Service Demand
Socket Socket Message Elapsed Send Recv Send Recv
Size Size Size Time Throughput local remote local remote
bytes bytes bytes secs. 10^6bits/s % S % U us/KB us/KB
Before:
87380 16384 16384 10.00 12844.38 9.30 -1.00 0.712 -1.00
After:
87380 16384 16384 10.00 13216.63 6.78 -1.00 0.504 -1.000
Changes from v1:
* Dropped use of CHECKSUM_UNNECESSARY for remote checksum offload
* Left encap_hdr_csum as it will likely be needed in future for SCTP GSO
* Broke the changes out over many more patches
* Updated GRE segmentation to more closely match UDP tunnel segmentation
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
