This patch correct a bug in the delay of pktgen. 
It makes sure the inter-packet interval is accurate.
Signed-off-by: NDaniel Turull <daniel.turull@gmail.com>
Signed-off-by: NRobert Olsson <robert.olsson@its.uu.se>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Here is patch to manipulate packet node allocation and implicitly
how packets are DMA'd etc.

The flag NODE_ALLOC enables the function and numa_node_id();
when enabled it can also be explicitly controlled via a new
node parameter

Tested this with 10 Intel 82599 ports w. TYAN S7025 E5520 CPU's.
Was able to TX/DMA ~80 Gbit/s to Ethernet wires.
Signed-off-by: NRobert Olsson <robert.olsson@its.uu.se>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

pass mark to all SA lookups to prepare them for when we add code
to have them search.
Signed-off-by: NJamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add missing try_to_freeze() to one of the pktgen_thread_worker() code
paths so that it doesn't block suspend/hibernation.

Fixes http://bugzilla.kernel.org/show_bug.cgi?id=15006Signed-off-by: NRafael J. Wysocki <rjw@sisk.pl>
Reported-and-tested-by: NCiprian Dorin Craciun <ciprian.craciun@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This updates pktgen so that it does not decrement skb->users
when it receives valid NET_XMIT_xxx values.  These are now
valid return values from ndo_start_xmit in net-next-2.6.
They also indicate the skb has been consumed.

This fixes pktgen to work correctly with vlan devices.
Signed-off-by: NJohn Fastabend <john.r.fastabend@intel.com>
Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Not including net/atm/

Compiled tested x86 allyesconfig only
Added a > 80 column line or two, which I ignored.
Existing checkpatch plaints willfully, cheerfully ignored.
Signed-off-by: NJoe Perches <joe@perches.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

pktgen threads are bound to given CPU, we can allocate memory for
these threads in a NUMA aware way.

After a pktgen session on two threads, we can check flows memory was
allocated on right node, instead of a not related one.

# grep pktgen_thread_write /proc/vmallocinfo
0xffffc90007204000-0xffffc90007385000 1576960 pktgen_thread_write+0x3a4/0x6b0 [pktgen] pages=384 vmalloc N0=384
0xffffc90007386000-0xffffc90007507000 1576960 pktgen_thread_write+0x3a4/0x6b0 [pktgen] pages=384 vmalloc N1=384
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When multi queue compatable names are used by pktgen (eg eth0@0),
we currently cannot unload a NIC driver if one of its device
is currently in use.

Allow pktgen_find_dev() to find pktgen devices by their suffix (netdev name)
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Commit e6fce5b9 (pktgen: multiqueue etc.) tried to relax
the pktgen restriction of one device per kernel thread, adding a '@'
tag to device names.

Problem is we dont perform check on full pktgen device name.
This allows adding many time same 'device' to pktgen thread

 pgset "add_device eth0@0"

one session later :

 pgset "add_device eth0@0"

(This doesnt find previous device)

This consumes ~1.5 MBytes of vmalloc memory per round and also triggers
this warning :

[  673.186380] proc_dir_entry 'pktgen/eth0@0' already registered
[  673.186383] Modules linked in: pktgen ixgbe ehci_hcd psmouse mdio mousedev evdev [last unloaded: pktgen]
[  673.186406] Pid: 6219, comm: bash Tainted: G        W  2.6.32-rc7-03302-g41cec6f1-dirty #16
[  673.186410] Call Trace:
[  673.186417]  [<ffffffff8104a29b>] warn_slowpath_common+0x7b/0xc0
[  673.186422]  [<ffffffff8104a341>] warn_slowpath_fmt+0x41/0x50
[  673.186426]  [<ffffffff8114e789>] proc_register+0x109/0x210
[  673.186433]  [<ffffffff8100bf2e>] ? apic_timer_interrupt+0xe/0x20
[  673.186438]  [<ffffffff8114e905>] proc_create_data+0x75/0xd0
[  673.186444]  [<ffffffffa006ad38>] pktgen_thread_write+0x568/0x640 [pktgen]
[  673.186449]  [<ffffffffa006a7d0>] ? pktgen_thread_write+0x0/0x640 [pktgen]
[  673.186453]  [<ffffffff81149144>] proc_reg_write+0x84/0xc0
[  673.186458]  [<ffffffff810f5a58>] vfs_write+0xb8/0x180
[  673.186463]  [<ffffffff810f5c11>] sys_write+0x51/0x90
[  673.186468]  [<ffffffff8100b51b>] system_call_fastpath+0x16/0x1b
[  673.186470] ---[ end trace ccbb991b0a8d994d ]---

Solution to this problem is to use a odevname field (includes @ tag and suffix),
instead of using netdevice name.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NRobert Olsson <robert.olsson@its.uu.se>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

cur_pkt_size can be changed in proc fs while pktgen is running,
we better use a private field to get precise tx-bytes counter.
Signed-off-by: NBen Greear <greearb@candelatech.com>
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

While playing with pktgen, I realized IP ID was not filled and a
random value was taken, possibly leaking 2 bytes of kernel memory.
 
We can use an increasing ID, this can help diagnostics anyway.

Also clear packet payload, instead of leaking kernel memory.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We can avoid two atomic ops on skb->users if packet is not going to be
sent to the device (because hardware txqueue is full)
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Acked-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Commit fd29cf72 (pktgen: convert to use ktime_t)
inadvertantly converted "delay" parameter from nanosec to microsec.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

It is not currently possible to instruct pktgen to use one selected tx queue.

When Robert added multiqueue support in commit 45b270f8, he added
an interval (queue_map_min, queue_map_max), and his code doesnt take
into account the case of min = max, to select one tx queue exactly.

I suspect a high performance setup on a eight txqueue device wants
to use exactly eight cpus, and assign one tx queue to each sender.

This patchs makes pktgen select the right tx queue, not the first one.

Also updates Documentation to reflect Robert changes.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NRobert Olsson <robert.olsson@its.uu.se>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

After last pktgen changes, delay handling is wrong.

pktgen actually sends packets at full line speed.

Fix is to update pkt_dev->next_tx even if spin() returns early,
so that next spin() calls have a chance to see a positive delay.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Previous update did not resched in inner loop causing watchdogs.
Rewrite inner loop to:
  * account for delays better with less clock calls
  * more accurate timing of delay:
    - only delay if packet was successfully sent
    - if delay is 100ns and it takes 10ns to build packet then
      account for that
  * use wait_event_interruptible_timeout rather than open coding it.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Get rid of unused flag bit.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

It looks like after rename device proc entry is unusable,
because of no ->read_proc or ->proc_fops.

And create_proc_entry() is deprecated.
Signed-off-by: NAlexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Increase module version, and cleanup module info.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Simpler to have one place that spins and accounts for delays,
this will also make the last packet be detected faster for more
repeatable timing.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This changes how the pktgen thread spins/waits between
packets if delay is configured. It uses a high res timer to
wait for time to arrive.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The kernel ktime_t is a nice generic infrastructure for mananging
high resolution times, as is done in pktgen.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If not using delay then no need to update next_tx after
each packet sent. This allows pktgen to send faster especially
on systems with slower clock sources.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Handle standard (and non-standard) return values in a switch.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

netdev_alloc_skb is NUMA node aware.
Also, don't exhaust atomic emergency pool. Don't want pktgen
to cause OOM behaviour.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The if statement to test for "should a new packet be used"
can be simplified.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Do some reorganization of transmit logic path:
   * move transmit queue full idle to separate routine
   * add a cpu_relax()
   * eliminate some of the uneeded goto's
   * if queue is still stopped, go back to main thread loop.
   * don't give up transmitting if quantum is exhausted (be greedy)
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

All the callers were freeing skb after stopping device.
Remove unneeded forward decl.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Don't force inlining where not needed. Gcc does better job
of deciding to inline local functions.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

A couple of minor functions can be written more compactly.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

vfree() does its own 'NULL' check, so no need for check before
calling it.
Signed-off-by: NFigo.zhang <figo1802@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We would like to get rid of netdev->trans_start = jiffies; that about all net
drivers have to use in their start_xmit() function, and use txq->trans_start
instead.

This can be done generically in core network, as suggested by David.

Some devices, (particularly loopback) dont need trans_start update, because
they dont have transmit watchdog. We could add a new device flag, or rely
on fact that txq->tran_start can be updated is txq->xmit_lock_owner is
different than -1. Use a helper function to hide our choice.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

typo -- pkt_dev->nflows is for stats only, the number of concurrent
flows is stored in cflows.
Reported-By: NVladimir Ivashchenko <hazard@francoudi.com>
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Setting ->owner as done currently (pde->owner = THIS_MODULE) is racy
as correctly noted at bug #12454. Someone can lookup entry with NULL
->owner, thus not pinning enything, and release it later resulting
in module refcount underflow.

We can keep ->owner and supply it at registration time like ->proc_fops
and ->data.

But this leaves ->owner as easy-manipulative field (just one C assignment)
and somebody will forget to unpin previous/pin current module when
switching ->owner. ->proc_fops is declared as "const" which should give
some thoughts.

->read_proc/->write_proc were just fixed to not require ->owner for
protection.

rmmod'ed directories will be empty and return "." and ".." -- no harm.
And directories with tricky enough readdir and lookup shouldn't be modular.
We definitely don't want such modular code.

Removing ->owner will also make PDE smaller.

So, let's nuke it.

Kudos to Jeff Layton for reminding about this, let's say, oversight.

http://bugzilla.kernel.org/show_bug.cgi?id=12454Signed-off-by: NAlexey Dobriyan <adobriyan@gmail.com>

Remove some pointless conditionals before kfree_skb().
Signed-off-by: NWei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NAlexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch moves neigh_setup and hard_start_xmit into the network device ops
structure. For bisection, fix all the previously converted drivers as well.
Bonding driver took the biggest hit on this.

Added a prefetch of the hard_start_xmit in the fast path to try and reduce
any impact this would have.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>