Shan Wei [Tue, 13 Nov 2012 12:36:00 +0000 (20:36 +0800)]
net: xfrm: use __this_cpu_read per-cpu helper
this_cpu_ptr/this_cpu_read is faster than per_cpu_ptr(p, smp_processor_id())
and can reduce memory accesses.
The latter helper needs to find the offset for current cpu,
and needs more assembler instructions which objdump shows in following.
this_cpu_ptr relocates and address. this_cpu_read() relocates the address
and performs the fetch. this_cpu_read() saves you more instructions
since it can do the relocation and the fetch in one instruction.
David S. Miller [Thu, 8 Nov 2012 00:08:42 +0000 (19:08 -0500)]
Merge tag 'batman-adv-for-davem' of git://git.open-mesh.org/linux-merge
Included changes:
- minimal fixes to the packet layout to avoid the __packed attribute when not
needed
- new packet type called UNICAST_4ADDR: in this packet it is possible to find
both source and destination node (in the classic UNICAST header only the
destination field exists).
- a new feature: Distributed ARP Table (D.A.T.). It aims to reduce ARP lookups
latency by means of a simil-DHT approach.
Merav Sicron [Wed, 7 Nov 2012 00:45:48 +0000 (00:45 +0000)]
bnx2x: Support loading cnic resources at run-time
This patch replaces the BCM_CNIC define with a flag which can change at run-time
and which does not use the CONFIG_CNIC kconfig option.
For the PF/hypervisor driver cnic is always supported, however allocation of
cnic resources and configuration of the HW for offload mode is done only when
the cnic module registers bnx2x.
Merav Sicron [Wed, 7 Nov 2012 00:45:47 +0000 (00:45 +0000)]
bnx2x: HSI change for 'update' ramrod
This patch updates the driver-FW HSI to support changes to the 'update' ramrod
(FW supports this change since 7.8.2). This ramrod is sent when the cnic module
registers bnx2x, to enable changing the nic_mode configuration in HW at
run-time.
Sathya Perla [Tue, 6 Nov 2012 17:48:59 +0000 (17:48 +0000)]
be2net: fix access to SEMAPHORE reg
The SEMAPHORE register was being accessed from the csr BAR space. This BAR
may not be available in some Skyhawk-R configurations. Instead, access this
register via the PCI config space (it's available there too).
Sathya Perla [Tue, 6 Nov 2012 17:48:58 +0000 (17:48 +0000)]
be2net: re-factor bar mapping code
1) separate NIC and roce bar mapping code
2) parse sli_intf::if_type inside be_map_pci_bars() as if_type must be
used only to identify bars.
3) Use pci_iomap/unmap() routines
Eric Dumazet [Mon, 5 Nov 2012 16:20:42 +0000 (16:20 +0000)]
mlx4: change TX coalescing defaults
mlx4 currently uses a too high tx coalescing setting, deferring
TX completion interrupts by up to 128 us.
With the recent skb_orphan() removal in commit 8112ec3b872,
performance of a single TCP flow is capped to ~4 Gbps, unless
we increase tcp_limit_output_bytes.
I suggest using 16 us instead of 128 us, allowing a finer control.
Performance of a single TCP flow is restored to previous levels,
while keeping TCP small queues fully enabled with default sysctl.
batman-adv: enable fast client detection using unicast_4addr packets
The "early client detection mechanism" can be extended to find new clients by
means of unicast_4addr packets.
The unicast_4addr packet contains as well as the broadcast packet (which is
currently used in this mechanism) the address of the originating node and can
therefore be used to install new entries in the Global Translation Table
This patch makes it possible to decide whether to include DAT within the
batman-adv binary or not.
It is extremely useful when the user wants to reduce the size of the resulting
module by cutting off any not needed feature.
In case of an ARP message going in or out the soft_iface, it is intercepted and
a special action is performed. In particular the DHT helper functions previously
implemented are used to store all the ARP entries belonging to the network in
order to provide a fast and unicast lookup instead of the classic broadcast
flooding mechanism.
Each node stores the entries it is responsible for (following the DHT rules) in
its soft_iface ARP table. This makes it possible to reuse the kernel data
structures and functions for ARP management.
batman-adv: Distributed ARP Table - implement local storage
Since batman-adv cannot inter-operate with the host ARP table, this patch
introduces a batman-adv private storage for ARP entries exchanged within DAT.
This storage will represent the node local cache in the DAT protocol.
Add all the relevant functions in order to manage a Distributed Hash Table over
the B.A.T.M.A.N.-adv network. It will later be used to store several ARP entries
and implement DAT (Distributed ARP Table)
batman-adv: Distributed ARP Table - add a new debug log level
A new log level has been added to concentrate messages regarding DAT: ARP
snooping, requests, response and DHT related messages.
The new log level is named BATADV_DBG_DAT
The current unicast packet type does not contain the orig source address. This
patches add a new unicast packet (called UNICAST_4ADDR) which provides two new
fields: the originator source address and the subtype (the type of the data
contained in the packet payload). The former is useful to identify the node
which injected the packet into the network and the latter is useful to avoid
creating new unicast packet types in the future: a macro defining a new subtype
will be enough.
Sven Eckelmann [Mon, 5 Nov 2012 20:25:26 +0000 (21:25 +0100)]
batman-adv: Mark correctly aligned headers not as __packed
Headers which are already perfectly aligned and create a 4 byte boundary
non-ethernet header payload can have the __packed attribute removed. The
__packed attribute doesn't change the appeareance of the packet for these
headers because no extra padding is necessary to align the data members. The
compiler will also create slightly faster code for loads of multi-byte members.
Sven Eckelmann [Sun, 4 Nov 2012 16:11:45 +0000 (17:11 +0100)]
batman-adv: Reserve extra bytes in skb for better alignment
The ethernet header is 14 bytes long. Therefore, the data after it is not 4
byte aligned and may cause problems on systems without unaligned data access.
Reserving NET_IP_ALIGN more byes can fix the misalignment of the ethernet
header.
Ming Lei [Tue, 6 Nov 2012 04:53:08 +0000 (04:53 +0000)]
usbnet: runtime wake up device before calling usbnet_{read|write}_cmd
This patch gets the runtime PM reference count before calling
usbnet_{read|write}_cmd, and puts it after completion of the
usbnet_{read|write}_cmd, so that the usb control message can always
be sent to one active device in the non-PM context.
Ming Lei [Tue, 6 Nov 2012 04:53:07 +0000 (04:53 +0000)]
usbnet: smsc95xx: apply the introduced usbnet_{read|write}_cmd_nopm
This patch applies the introduced usbnet_read_cmd_nopm() and
usbnet_write_cmd_nopm() in the callback of resume and suspend
to avoid deadlock if USB runtime PM is considered into
usbnet_read_cmd() and usbnet_write_cmd().
Ming Lei [Tue, 6 Nov 2012 04:53:05 +0000 (04:53 +0000)]
usbnet: smsc75xx: apply the introduced usbnet_{read|write}_cmd_nopm
This patch applies the introduced usbnet_read_cmd_nopm() and
usbnet_write_cmd_nopm() in the callback of resume and suspend
to avoid deadlock if USB runtime PM is considered into
usbnet_read_cmd() and usbnet_write_cmd().
Ming Lei [Tue, 6 Nov 2012 04:53:04 +0000 (04:53 +0000)]
usbnet: introduce usbnet_{read|write}_cmd_nopm
This patch introduces the below two helpers to prepare for solving
the usbnet runtime PM problem, which may cause some network utilities
(ifconfig, ethtool,...) touch a suspended device.
usbnet_read_cmd_nopm()
usbnet_write_cmd_nopm()
The above two helpers should be called by usbnet resume/suspend
callback to avoid deadlock.
Rob Herring [Mon, 5 Nov 2012 06:22:24 +0000 (06:22 +0000)]
net: calxedaxgmac: ip align receive buffers
On gcc 4.7, we will get alignment traps in the ip stack if we don't align
the ip headers on receive. The h/w can support this, so use ip aligned
allocations.
Cut down the unnecessary padding on the allocation. The buffer can start on
any byte alignment, but the size including the begining offset must be 8
byte aligned. So the h/w buffer size must include the NET_IP_ALIGN offset.
Thanks to Eric Dumazet for the initial patch highlighting the padding issues.
Rob Herring [Mon, 5 Nov 2012 06:22:23 +0000 (06:22 +0000)]
net: calxedaxgmac: rework transmit ring handling
Only generate tx interrupts on every ring size / 4 descriptors. Move the
netif_stop_queue call to the end of the xmit function rather than
checking at the beginning.
Rob Herring [Mon, 5 Nov 2012 06:22:22 +0000 (06:22 +0000)]
net: calxedaxgmac: drop some unnecessary register writes
The interrupts have already been cleared, so we don't need to clear them
again. Also, we could miss interrupts if they are cleared, but we don't
process the packet.
Rob Herring [Mon, 5 Nov 2012 06:22:21 +0000 (06:22 +0000)]
net: calxedaxgmac: use raw i/o accessors in rx and tx paths
The standard readl/writel accessors involve a spinlock and cache sync
operation on ARM platforms with an outer cache. Only DMA triggering
accesses need this, so use the raw variants instead in the critical paths.
The relaxed variants would be more appropriate, but don't exist on all
arches.
New received frames will trigger the rx DMA to poll the DMA descriptors,
so there is no need to tell the h/w to poll. We also want to enable
dropping frames from the fifo when there is no buffer.
Eric Dumazet [Mon, 5 Nov 2012 16:40:49 +0000 (16:40 +0000)]
htb: fix two bugs
Commit 56b765b79e9 (htb: improved accuracy at high rates)
introduced two bugs :
1) one bstats_update() was inadvertently removed from
htb_dequeue_tree(), breaking statistics/rate estimation.
2) Missing qdisc_put_rtab() calls in htb_change_class(),
leaking kernel memory, now struct htb_class no longer
retains pointers to qdisc_rate_table structs.
Since only rate is used, dont use qdisc_get_rtab() calls
copying data we ignore anyway.
Lee Jones [Sat, 3 Nov 2012 22:02:30 +0000 (23:02 +0100)]
bridge: Avoid 'statement with no effect' compiler warnings
Instead of issuing (0) statements when !CONFIG_SYSFS which will cause
'warning: ', we'll use inline statements instead. This will effectively
do the same thing, but suppress any unnecessary warnings.
Paul Bolle [Fri, 2 Nov 2012 23:53:15 +0000 (23:53 +0000)]
atp: remove set_rx_mode_8012()
Building atp.o triggers this GCC warning:
drivers/net/ethernet/realtek/atp.c: In function ‘set_rx_mode’:
drivers/net/ethernet/realtek/atp.c:871:26: warning: ‘mc_filter[0]’ may be used uninitialized in this function [-Wuninitialized]
GCC is correct. In promiscuous mode 'mc_filter' will be used
uninitialized in set_rx_mode_8012(), which is apparently inlined into
set_rx_mode().
But it turns out set_rx_mode_8012() will never be called, since
net_local.chip_type will always be RTL8002. So we can just remove
set_rx_mode_8012() and do some related cleanups.
Richard Cochran [Fri, 2 Nov 2012 22:25:30 +0000 (22:25 +0000)]
cpsw: fix leaking IO mappings
The CPSW driver remaps two different IO regions, but fails to unmap them
both. This patch fixes the issue by calling iounmap in the appropriate
places.
Richard Cochran [Fri, 2 Nov 2012 22:25:29 +0000 (22:25 +0000)]
cpsw: rename register banks to match the reference manual, part 2
The code mixes up the CPSW_SS and the CPSW_WR register naming. This patch
changes the names to conform to the published Technical Reference Manual
from TI, in order to make working on the code less confusing.
John Fastabend [Fri, 2 Nov 2012 16:32:36 +0000 (16:32 +0000)]
net: fix bridge notify hook to manage flags correctly
The bridge notify hook rtnl_bridge_notify() was not handling the
case where the master flags was set or with both flags set. First
flags are not being passed correctly and second the logic to parse
them is broken.
This patch passes the original flags value and fixes the
logic.
macb: Keep driver's speed/duplex in sync with actual NCFGR
When underlying phy driver restores its state very fast after being brought
down and up so that macb driver function macb_handle_link_change() was never
called with link state "down", driver's internal representation of phy speed
and duplex (bp->speed and bp->duplex) didn't change. So, macb driver sees no
reason to perform actual write to the NCFGR register, although the speed and
duplex settings in that register were reset when interface was brought down
and up. In that case actual phy speed and duplex differ from NCFGR settings.
The patch fixes that by keeping internal driver representation of speed and
duplex in sync with actual content of NCFGR.
Richard Cochran [Thu, 1 Nov 2012 23:57:38 +0000 (23:57 +0000)]
ptp: fixup Kconfig for two PHC drivers.
Ben Hutchings recently came up with a better way to handle the kconfig
dependencies for the PTP hardware clocks. This patch converts one new and
one older driver to the new scheme.
Vimalkumar [Wed, 31 Oct 2012 06:04:11 +0000 (06:04 +0000)]
htb: improved accuracy at high rates
Current HTB (and TBF) uses rate table computed by the "tc"
userspace program, which has the following issue:
The rate table has 256 entries to map packet lengths
to token (time units). With TSO sized packets, the
256 entry granularity leads to loss/gain of rate,
making the token bucket inaccurate.
Thus, instead of relying on rate table, this patch
explicitly computes the time and accounts for packet
transmission times with nanosecond granularity.
This greatly improves accuracy of HTB with a wide
range of packet sizes.
tc class add dev $dev classid 1:1 parent 1: \
rate 5Gbit mtu 64k
Here is an example of inaccuracy:
$ iperf -c host -t 10 -i 1
With old htb:
eth4: 34.76 Mb/s In 5827.98 Mb/s Out - 65836.0 p/s In 481273.0 p/s Out
[SUM] 9.0-10.0 sec 669 MBytes 5.61 Gbits/sec
[SUM] 0.0-10.0 sec 6.50 GBytes 5.58 Gbits/sec
With new htb:
eth4: 28.36 Mb/s In 5208.06 Mb/s Out - 53704.0 p/s In 430076.0 p/s Out
[SUM] 9.0-10.0 sec 594 MBytes 4.98 Gbits/sec
[SUM] 0.0-10.0 sec 5.80 GBytes 4.98 Gbits/sec
The bits per second on the wire is still 5200Mb/s with new HTB
because qdisc accounts for packet length using skb->len, which
is smaller than total bytes on the wire if GSO is used. But
that is for another patch regardless of how time is accounted.
Many thanks to Eric Dumazet for review and feedback.
Vincent Bernat [Tue, 30 Oct 2012 10:27:16 +0000 (10:27 +0000)]
vxlan: allow a user to set TTL value
"ip link add ... type vxlan ... ttl X" allows a user to set the TTL
used by a VXLAN for encapsulation. The provided value was ignored by
vxlan module and the default value of 1 was used when encapsulating
multicast packets.
Add Ethertype 0x4305 (not an officially registered id).
This Ethertype is used by every frame generated by B.A.T.M.A.N.-Advanced. Its
definition is currently batman-adv local only and since it is not officially
registered it is better to make its definition kernel-wide so that we avoid
collisions given by future unofficial uses of the same Ethertype.
Neil Horman [Mon, 29 Oct 2012 08:32:13 +0000 (08:32 +0000)]
sctp: Clean up type-punning in sctp_cmd_t union
Lots of points in the sctp_cmd_interpreter function treat the sctp_cmd_t arg as
a void pointer, even though they are written as various other types. Theres no
need for this as doing so just leads to possible type-punning issues that could
cause crashes, and if we remain type-consistent we can actually just remove the
void * member of the union entirely.
Change Notes:
v2)
* Dropped chunk that modified SCTP_NULL to create a marker pattern
should anyone try to use a SCTP_NULL() assigned sctp_arg_t, Assigning
to .zero provides the same effect and should be faster, per Vlad Y.
v3)
* Reverted part of V2, opting to use memset instead of .zero, so that
the entire union is initalized thus avoiding the i164 speculative load
problems previously encountered, per Dave M.. Also rewrote
SCTP_[NO]FORCE so as to use common infrastructure a little more
Daniel Borkmann [Sun, 28 Oct 2012 08:27:19 +0000 (08:27 +0000)]
pktgen: clean up ktime_t helpers
Some years ago, the ktime_t helper functions ktime_now() and ktime_lt()
have been introduced. Instead of defining them inside pktgen.c, they
should either use ktime_t library functions or, if not available, they
should be defined in ktime.h, so that also others can benefit from them.
ktime_compare() is introduced with a similar notion as in timespec_compare().
Eric Dumazet [Sat, 27 Oct 2012 23:16:46 +0000 (23:16 +0000)]
tcp: better retrans tracking for defer-accept
For passive TCP connections using TCP_DEFER_ACCEPT facility,
we incorrectly increment req->retrans each time timeout triggers
while no SYNACK is sent.
SYNACK are not sent for TCP_DEFER_ACCEPT that were established (for
which we received the ACK from client). Only the last SYNACK is sent
so that we can receive again an ACK from client, to move the req into
accept queue. We plan to change this later to avoid the useless
retransmit (and potential problem as this SYNACK could be lost)
TCP_INFO later gives wrong information to user, claiming imaginary
retransmits.
Decouple req->retrans field into two independent fields :
num_retrans : number of retransmit
num_timeout : number of timeouts
num_timeout is the counter that is incremented at each timeout,
regardless of actual SYNACK being sent or not, and used to
compute the exponential timeout.
Introduce inet_rtx_syn_ack() helper to increment num_retrans
only if ->rtx_syn_ack() succeeded.
Use inet_rtx_syn_ack() from tcp_check_req() to increment num_retrans
when we re-send a SYNACK in answer to a (retransmitted) SYN.
Prior to this patch, we were not counting these retransmits.
Change tcp_v[46]_rtx_synack() to increment TCP_MIB_RETRANSSEGS
only if a synack packet was successfully queued.
Ben Hutchings [Fri, 2 Nov 2012 12:56:52 +0000 (12:56 +0000)]
net: Fix continued iteration in rtnl_bridge_getlink()
Commit e5a55a898720096f43bc24938f8875c0a1b34cd7 ('net: create generic
bridge ops') broke the handling of a non-zero starting index in
rtnl_bridge_getlink() (based on the old br_dump_ifinfo()).
When the starting index is non-zero, we need to increment the current
index for each entry that we are skipping. Also, we need to check the
index before both cases, since we may previously have stopped
iteration between getting information about a device from its master
and from itself.
Ben Hutchings [Thu, 1 Nov 2012 09:12:02 +0000 (09:12 +0000)]
eth: Rename and properly align br_reserved_address array
Since this array is no longer part of the bridge driver, it should
have an 'eth' prefix not 'br'.
We also assume that either it's 16-bit-aligned or the architecture has
efficient unaligned access. Ensure the first of these is true by
explicitly aligning it.
Ben Hutchings [Thu, 1 Nov 2012 11:22:22 +0000 (11:22 +0000)]
sfc: Select PTP_1588_CLOCK
This was missed in commit a24006ed12616bde1bbdb26868495906a212d8dc
('ptp: Enable clock drivers along with associated net/PHY drivers')
which enabled sfc's clock driver unconditionally.
It seems that to avoid deadlocks it is enough to poll vq before
we are going to use the last buffer. This is faster than c70aa540c7a9f67add11ad3161096fb95233aa2e.
Even when vhost-net is in zero-copy transmit mode,
net core might still decide to copy the skb later
which is somewhat slower than a copy in user
context: data copy overhead is added to the cost of
page pin/unpin. The result is that enabling tx zero copy
option leads to higher CPU utilization for guest to guest
and guest to host traffic.
To fix this, suppress zero copy tx after a given number of
packets triggered late data copy. Re-enable periodically
to detect workload changes.
tun: report orphan frags errors to zero copy callback
When tun transmits a zero copy skb, it orphans the frags
which might need to allocate extra memory, in atomic context.
If that fails, notify ubufs callback before freeing the skb
as a hint that device should disable zerocopy mode.
Orphaning frags for zero copy skbs needs to allocate data in atomic
context so is has a chance to fail. If it does we currently discard
the skb which is safe, but we don't report anything to the caller,
so it can not recover by e.g. disabling zero copy.
Add an API to free skb reporting such errors: this is used
by tun in case orphaning frags fails.
Even if skb is marked for zero copy, net core might still decide
to copy it later which is somewhat slower than a copy in user context:
besides copying the data we need to pin/unpin the pages.
Add a parameter reporting such cases through zero copy callback:
if this happens a lot, device can take this into account
and switch to copying in user context.
This patch updates all users but ignores the passed value for now:
it will be used by follow-up patches.
Frank Li [Tue, 30 Oct 2012 18:25:31 +0000 (18:25 +0000)]
FEC: Add time stamping code and a PTP hardware clock
This patch adds a driver for the FEC(MX6) that offers time
stamping and a PTP haderware clock. Because FEC\ENET(MX6)
hardware frequency adjustment is complex, we have implemented
this in software by changing the multiplication factor of the
timecounter.
projects / linux.git / log