This enables the use of ethtool --set-channels devname combined N to
change the number of vRSS queues. Separate rx, tx, and other parameters
are not supported. The maximum is rsscap.num_recv_que. It passes the
given value to rndis_filter_device_add through the device_info->num_chn
field.
If the procedure fails, it attempts to recover to the prior state. If
the recovery fails, it logs an error and aborts.
Current num_chn is saved and restored when changing the MTU.
Signed-off-by: Andrew Schwartzmeyer <andschwa@microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Uses device_info->num_chn to pass user provided number of vRSS
queues (from ethtool --set-channels) to rndis_filter_device_add. If
nonzero and less than the maximum, set net_device->num_chn to the given
value; else default to prior algorithm.
Always initialize struct device_info to 0, otherwise not all its fields
are guaranteed to be 0, which is necessary when checking if num_chn has
been purposefully set.
Signed-off-by: Andrew Schwartzmeyer <andschwa@microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds data structures and handlers for messages related
to SRIOV Virtual Function.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current code returns from probe without waiting for the proper handling
of subchannels that may be requested. If the netvsc driver were to be rapidly
loaded/unloaded, we can trigger a panic as the unload will be tearing
down state that may not have been fully setup yet. We fix this issue by making
sure that we return from the probe call only after ensuring that the
sub-channel offers in flight are properly handled.
Reviewed-and-tested-by: Haiyang Zhang <haiyangz@microsoft.com
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current change mtu call only stops tx before removing RNDIS filter.
In case ringbufer is not empty, the rndis_filter_device_remove() may
hang on removing the buffers.
This patch adds close of RNDIS filter before removing it, also a
gradual waiting loop until the ring is empty. The change_mtu hang
issue under heavy traffic is solved by this patch.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When packet encapsulation is in use, the MTU needs to be reduced for
headroom reservation.
The existing code takes the updated MTU value only from the host side.
But vSwitch extensions, such as Open vSwitch, require the flexibility
to change the MTU to different values from within a guest during the
lifecycle of a vNIC, when the encapsulation protocol is changed. The
patch supports this kind of MTU changes.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull networking updates from David Miller:
1) Add TX fast path in mac80211, from Johannes Berg.
2) Add TSO/GRO support to ibmveth, from Thomas Falcon
3) Move away from cached routes in ipv6, just like ipv4, from Martin
KaFai Lau.
4) Lots of new rhashtable tests, from Thomas Graf.
5) Run ingress qdisc lockless, from Alexei Starovoitov.
6) Allow servers to fetch TCP packet headers for SYN packets of new
connections, for fingerprinting. From Eric Dumazet.
7) Add mode parameter to pktgen, for testing receive. From Alexei
Starovoitov.
8) Cache access optimizations via simplifications of build_skb(), from
Alexander Duyck.
9) Move page frag allocator under mm/, also from Alexander.
10) Add xmit_more support to hv_netvsc, from KY Srinivasan.
11) Add a counter guard in case we try to perform endless reclassify
loops in the packet scheduler.
12) Extern flow dissector to be programmable and use it in new "Flower"
classifier. From Jiri Pirko.
13) AF_PACKET fanout rollover fixes, performance improvements, and new
statistics. From Willem de Bruijn.
14) Add netdev driver for GENEVE tunnels, from John W Linville.
15) Add ingress netfilter hooks and filtering, from Pablo Neira Ayuso.
16) Fix handling of epoll edge triggers in TCP, from Eric Dumazet.
17) Add an ECN retry fallback for the initial TCP handshake, from Daniel
Borkmann.
18) Add tail call support to BPF, from Alexei Starovoitov.
19) Add several pktgen helper scripts, from Jesper Dangaard Brouer.
20) Add zerocopy support to AF_UNIX, from Hannes Frederic Sowa.
21) Favor even port numbers for allocation to connect() requests, and
odd port numbers for bind(0), in an effort to help avoid
ip_local_port_range exhaustion. From Eric Dumazet.
22) Add Cavium ThunderX driver, from Sunil Goutham.
23) Allow bpf programs to access skb_iif and dev->ifindex SKB metadata,
from Alexei Starovoitov.
24) Add support for T6 chips in cxgb4vf driver, from Hariprasad Shenai.
25) Double TCP Small Queues default to 256K to accomodate situations
like the XEN driver and wireless aggregation. From Wei Liu.
26) Add more entropy inputs to flow dissector, from Tom Herbert.
27) Add CDG congestion control algorithm to TCP, from Kenneth Klette
Jonassen.
28) Convert ipset over to RCU locking, from Jozsef Kadlecsik.
29) Track and act upon link status of ipv4 route nexthops, from Andy
Gospodarek.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next: (1670 commits)
bridge: vlan: flush the dynamically learned entries on port vlan delete
bridge: multicast: add a comment to br_port_state_selection about blocking state
net: inet_diag: export IPV6_V6ONLY sockopt
stmmac: troubleshoot unexpected bits in des0 & des1
net: ipv4 sysctl option to ignore routes when nexthop link is down
net: track link-status of ipv4 nexthops
net: switchdev: ignore unsupported bridge flags
net: Cavium: Fix MAC address setting in shutdown state
drivers: net: xgene: fix for ACPI support without ACPI
ip: report the original address of ICMP messages
net/mlx5e: Prefetch skb data on RX
net/mlx5e: Pop cq outside mlx5e_get_cqe
net/mlx5e: Remove mlx5e_cq.sqrq back-pointer
net/mlx5e: Remove extra spaces
net/mlx5e: Avoid TX CQE generation if more xmit packets expected
net/mlx5e: Avoid redundant dev_kfree_skb() upon NOP completion
net/mlx5e: Remove re-assignment of wq type in mlx5e_enable_rq()
net/mlx5e: Use skb_shinfo(skb)->gso_segs rather than counting them
net/mlx5e: Static mapping of netdev priv resources to/from netdev TX queues
net/mlx4_en: Use HW counters for rx/tx bytes/packets in PF device
...
Nothing in <asm/io.h> uses anything from <linux/vmalloc.h>, so
remove it from there and fix up the resulting build problems
triggered on x86 {64|32}-bit {def|allmod|allno}configs.
The breakages were triggering in places where x86 builds relied
on vmalloc() facilities but did not include <linux/vmalloc.h>
explicitly and relied on the implicit inclusion via <asm/io.h>.
Also add:
- <linux/init.h> to <linux/io.h>
- <asm/pgtable_types> to <asm/io.h>
... which were two other implicit header file dependencies.
Suggested-by: David Miller <davem@davemloft.net>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
[ Tidied up the changelog. ]
Acked-by: David Miller <davem@davemloft.net>
Acked-by: Takashi Iwai <tiwai@suse.de>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Vinod Koul <vinod.koul@intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Anton Vorontsov <anton@enomsg.org>
Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Cc: Colin Cross <ccross@android.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: James E.J. Bottomley <JBottomley@odin.com>
Cc: Jaroslav Kysela <perex@perex.cz>
Cc: K. Y. Srinivasan <kys@microsoft.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Kristen Carlson Accardi <kristen@linux.intel.com>
Cc: Len Brown <lenb@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rafael J. Wysocki <rjw@rjwysocki.net>
Cc: Suma Ramars <sramars@cisco.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Tony Luck <tony.luck@intel.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Allocate the receive bufer from the NUMA node assigned to the primary
channel.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current algorithm for deciding on the number of VRSS channels is
not optimal since we open up the min of number of CPUs online and the
number of VRSS channels the host is offering. So on a 32 VCPU guest
we could potentially open 32 VRSS subchannels. Experimentation has
shown that it is best to limit the number of VRSS channels to the number
of CPUs within a NUMA node.
Here is the new algorithm for deciding on the number of sub-channels we
would open up:
1) Pick the minimum of what the host is offering and what the driver
in the guest is specifying as the default value.
2) Pick the minimum of (1) and the numbers of CPUs in the NUMA
node the primary channel is bound to.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently the struct netvsc_stats has a member s_sync
of type u64_stats_sync.
This definition will break kernel build as the macro
netdev_alloc_pcpu_stats requires this member name to be syncp.
(see netdev_alloc_pcpu_stats definition in ./include/linux/netdevice.h)
This patch changes netvsc_stats's member name from s_sync to syncp to fix
the build break.
Signed-off-by: Simon Xiao <sixiao@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Current code does not lock anything when calculating the TX and RX stats.
As a result, the RX and TX data reported by ifconfig are not accuracy in a
system with high network throughput and multiple CPUs (in my test,
RX/TX = 83% between 2 HyperV VM nodes which have 8 vCPUs and 40G Ethernet).
This patch fixed the above issue by using per_cpu stats.
netvsc_get_stats64() summarizes TX and RX data by iterating over all CPUs
to get their respective stats.
This v2 patch addressed David's comments on the cleanup path when
netdev_alloc_pcpu_stats() failed.
Signed-off-by: Simon Xiao <sixiao@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Four minor merge conflicts:
1) qca_spi.c renamed the local variable used for the SPI device
from spi_device to spi, meanwhile the spi_set_drvdata() call
got moved further up in the probe function.
2) Two changes were both adding new members to codel params
structure, and thus we had overlapping changes to the
initializer function.
3) 'net' was making a fix to sk_release_kernel() which is
completely removed in 'net-next'.
4) In net_namespace.c, the rtnl_net_fill() call for GET operations
had the command value fixed, meanwhile 'net-next' adjusted the
argument signature a bit.
This also matches example merge resolutions posted by Stephen
Rothwell over the past two days.
Signed-off-by: David S. Miller <davem@davemloft.net>
Based on the information given to this driver (via the xmit_more skb flag),
we can defer signaling the host if more packets are on the way. This will help
make the host more efficient since it can potentially process a larger batch of
packets. Implement this optimization.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With commit b56fc3c536 ("hv_netvsc: Fix a bug in netvsc_start_xmit()"),
skb variable is no longer used in netvsc_send. Remove variable and dead
code that depended on it.
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Jerry Snitselaar <jsnitsel@redhat.com>
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit b08cc79155 eliminated memory
allocation in the packet send path:
"hv_netvsc: Eliminate memory allocation in the packet send path
The network protocol used to communicate with the host is the remote ndis (rndis)
protocol. We need to decorate each outgoing packet with a rndis header and
additional rndis state (rndis per-packet state). To manage this state, we
currently allocate memory in the transmit path. Eliminate this allocation by
requesting additional head room in the skb."
This commit introduced a bug since it did not account for the case if the skb
was cloned. Fix this bug.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Tested-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
1. Introduce netif-msg to netvsc to control debug logging output
and keep msg_enable in netvsc_device_context so that it is
kept persistently.
2. Only call dump_rndis_message() when NETIF_MSG_RX_ERR or above
is specified in netvsc module debug param.
In non-debug mode, in current code, dump_rndis_message() will not
dump anything but it still initialize some local variables and
process the switch logic which is unnecessary, especially in
high network throughput situation.
Signed-off-by: Simon Xiao <sixiao@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If remaining space in a send buffer slot is too small for the whole message,
we only copy the RNDIS header and PPI data into send buffer, so we can batch
one more packet each time. It reduces the vmbus per-message overhead.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In netvsc_start_xmit() we can handle packets which are scattered around not
more than MAX_PAGE_BUFFER_COUNT-2 pages. It is, however, easy to create a
packet which is not big in size but occupies more pages (e.g. if it uses frags
on compound pages boundaries). When we drop such packet it cases sender to try
resending it but in most cases it will try resending the same packet which will
also get dropped, this will cause the particular connection to stick. To solve
the issue we can try linearizing skb.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
... which validly uses dev_kfree_skb_any() instead of dev_kfree_skb().
Setting ret to -EFAULT and -ENOMEM have no real meaning here (we need to set
it to anything but -EAGAIN) as we drop the packet and return NETDEV_TX_OK
anyway.
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the two places changed, we now use netvsc_xmit_completion() which properly
frees hv_netvsc_packet in or not in skb headroom.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The sum of RNDIS msg and PPI struct sizes is used in multiple places, so we define
a macro for them.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The network protocol used to communicate with the host is the remote ndis (rndis)
protocol. We need to decorate each outgoing packet with a rndis header and
additional rndis state (rndis per-packet state). To manage this state, we
currently allocate memory in the transmit path. Eliminate this allocation by
requesting additional head room in the skb.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation for embedding the rndis state and other packet state into
the skb, cleanup the test for freeing the skb.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The vmbus_are_subchannels_present() also involves opening the channels, which
may be too early at this point. Checking for subchannels is not necessary here.
So this patch removes it. Subchannels will be opened when offer messages arrive.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With this patch, we can send out multiple RNDIS data packets in one send buffer
slot and one VMBus message. It reduces the overhead associated with VMBus messages.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds support for reporting the actual and maximum combined channels
count of the hv_netvsc driver via 'ethtool --show-channels'.
This required adding 'max_chn' to 'struct netvsc_device', and assigning
it 'rsscap.num_recv_que' in 'rndis_filter_device_add'. Now we can access
the combined maximum channel count via 'struct netvsc_device' in the
ethtool callback.
Signed-off-by: Andrew Schwartzmeyer <andrew@schwartzmeyer.com>
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
this patch fixes following sparse warnings:
netvsc.c:688:5: warning: symbol 'netvsc_copy_to_send_buf' was not declared. Should it be static?
rndis_filter.c:627:5: warning: symbol 'rndis_filter_set_offload_params' was not declared. Should it be static?
rndis_filter.c:702:5: warning: symbol 'rndis_filter_set_rss_param' was not declared. Should it be static?
Signed-off-by: Lad, Prabhakar <prabhakar.csengg@gmail.com>
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/vxlan.c
drivers/vhost/net.c
include/linux/if_vlan.h
net/core/dev.c
The net/core/dev.c conflict was the overlap of one commit marking an
existing function static whilst another was adding a new function.
In the include/linux/if_vlan.h case, the type used for a local
variable was changed in 'net', whereas the function got rewritten
to fix a stacked vlan bug in 'net-next'.
In drivers/vhost/net.c, Al Viro's iov_iter conversions in 'net-next'
overlapped with an endainness fix for VHOST 1.0 in 'net'.
In drivers/net/vxlan.c, vxlan_find_vni() added a 'flags' parameter
in 'net-next' whereas in 'net' there was a bug fix to pass in the
correct network namespace pointer in calls to this function.
Signed-off-by: David S. Miller <davem@davemloft.net>
The existing code frees the skb in EAGAIN case, in which the skb will be
retried from upper layer and used again.
Also, the existing code doesn't free send buffer slot in error case, because
there is no completion message for unsent packets.
This patch fixes these problems.
(Please also include this patch for stable trees. Thanks!)
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
return type of wait_for_completion_timeout is unsigned long not int, this
patch just fixes up the declarations.
Signed-off-by: Nicholas Mc Guire <der.herr@hofr.at>
Signed-off-by: David S. Miller <davem@davemloft.net>
The changed names are union fields with the same size, so the existing code
still works. But, we now update these variables to the correct names.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds proper handling of the vNIC hot removal event, which includes
a rescind-channel-offer message from the host side that triggers vNIC close and
removal. In this case, the notices to the host during close and removal is not
necessary because the channel is rescinded. This patch blocks these unnecessary
messages, and lets vNIC removal process complete normally.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The vfree() function performs also input parameter validation.
Thus the test around the call is not needed.
This issue was detected by using the Coccinelle software.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the host uses packet encapsulation feature, the MTU may be reduced by the
host due to headroom reservation for encapsulation. This patch handles this
new MTU value.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This will allow the workload spreading via vRSS for IPv6.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
total_data_buflen is used by netvsc_send() to decide if a packet can be put
into send buffer. It should also include the size of RNDIS message before the
Ethernet frame. Otherwise, a messge with total size bigger than send_section_size
may be copied into the send buffer, and cause data corruption.
[Request to include this patch to the Stable branches]
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case that the IP header has optional field at the end, this patch will
get the port numbers after that field, and compute the hash. The general
parser skb_flow_dissect() is used here.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After the packet is successfully sent, we should not touch the packet
as it may have been freed. This patch is based on the work done by
Long Li <longli@microsoft.com>.
David, please queue this up for stable.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Reported-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/usb/r8152.c
net/netfilter/nfnetlink.c
Both r8152 and nfnetlink conflicts were simple overlapping changes.
Signed-off-by: David S. Miller <davem@davemloft.net>
After the packet is successfully sent, we should not touch the skb
as it may have been freed. This patch is based on the work done by
Long Li <longli@microsoft.com>.
In this version of the patch I have fixed issues pointed out by David.
David, please queue this up for stable.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Tested-by: Long Li <longli@microsoft.com>
Tested-by: Sitsofe Wheeler <sitsofe@yahoo.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We try to call free_netvsc_device(net_device) when "net_device" is NULL.
It leads to an Oops.
Fixes: f90251c8a6 ('hyperv: Increase the buffer length for netvsc_channel_cb()')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When the buffer is too small for a packet from VMBus, a bigger buffer will be
allocated in netvsc_channel_cb() and retry reading the packet from VMBus.
Increasing this buffer size will reduce the retry overhead.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: Dexuan Cui <decui@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
WS2008R2 is a supported platform and it turns out that the maximum sendbuf
size that ws2008R2 can support is only 15MB. Make the necessary
adjustment.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Intel did some benchmarking on our network throughput when Linux on Hyper-V
is as used as a gateway. This fix gave us almost a 1 Gbps additional throughput
on about 5Gbps base throughput we hadi, prior to increasing the sendbuf size.
The sendbuf mechanism is a copy based transport that we have which is clearly
more optimal than the copy-free page flipping mechanism (for small packets).
In the forwarding scenario, we deal only with MTU sized packets,
and increasing the size of the senbuf area gave us the additional performance.
For what it is worth, Windows guests on Hyper-V, I am told use similar sendbuf
size as well.
The exact value of sendbuf I think is less important than the fact that it needs
to be larger than what Linux can allocate as physically contiguous memory.
Thus the change over to allocating via vmalloc().
We currently allocate 16MB receive buffer and we use vmalloc there for allocation.
Also the low level channel code has already been modified to deal with physically
dis-contiguous memory in the ringbuffer setup.
Based on experimentation Intel did, they say there was some improvement in throughput
as the sendbuf size was increased up to 16MB and there was no effect on throughput
beyond 16MB. Thus I have chosen 16MB here.
Increasing the sendbuf value makes a material difference in small packet handling
In this version of the patch, based on David's feedback, I have added
additional details in the commit log.
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>