It is useful to be able to see the hash configuration when running tests.
This patch adds a debugfs node for that purpose.
Signed-off-by: NPaul Durrant <paul.durrant@citrix.com>
Cc: Wei Liu <wei.liu2@citrix.com>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

My recent patch to include/xen/interface/io/netif.h defines a new shared
ring (in addition to the rx and tx rings) for passing control messages
from a VM frontend driver to a backend driver.

A previous patch added the necessary boilerplate for mapping the control
ring from the frontend, should it be created. This patch adds
implementations for each of the defined protocol messages.
Signed-off-by: NPaul Durrant <paul.durrant@citrix.com>
Cc: Wei Liu <wei.liu2@citrix.com>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

My recent patch to include/xen/interface/io/netif.h defines a new shared
ring (in addition to the rx and tx rings) for passing control messages
from a VM frontend driver to a backend driver.

This patch adds the necessary code to xen-netback to map this new shared
ring, should it be created by a frontend, but does not add implementations
for any of the defined protocol messages. These are added in a subsequent
patch for clarity.
Signed-off-by: NPaul Durrant <paul.durrant@citrix.com>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The code does not currently support a frontend passing multiple extra info
fragments to the backend in a tx request. The xenvif_get_extras() function
handles multiple extra_info fragments but make_tx_response() assumes there
is only ever a single extra info fragment.

This patch modifies xenvif_get_extras() to pass back a count of extra
info fragments, which is then passed to make_tx_response() (after
possibly being stashed in pending_tx_info for deferred responses).
Signed-off-by: NPaul Durrant <paul.durrant@citrix.com>
Cc: Wei Liu <wei.liu2@citrix.com>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

My recent patch to the Xen Project documents a protocol for 'dynamic
multicast control' in netif.h. This extends the previous multicast control
protocol to not require a shared ring reconnection to turn the feature off.
Instead the backend watches the "request-multicast-control" key in xenstore
and turns the feature off if the key value is written to zero.

This patch adds support for dynamic multicast control in xen-netback.
Signed-off-by: NPaul Durrant <paul.durrant@citrix.com>
Cc: Ian Campbell <ian.campbell@citrix.com>
Cc: Wei Liu <wei.liu2@citrix.com>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The PV network protocol is using 4KB page granularity. The goal of this
patch is to allow a Linux using 64KB page granularity working as a
network backend on a non-modified Xen.

It's only necessary to adapt the ring size and break skb data in small
chunk of 4KB. The rest of the code is relying on the grant table code.
Signed-off-by: NJulien Grall <julien.grall@citrix.com>
Reviewed-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid Vrabel <david.vrabel@citrix.com>

Commit f48da8b1 (xen-netback: fix
unlimited guest Rx internal queue and carrier flapping) introduced a
regression.

The PV frontend in IPXE only places 4 requests on the guest Rx ring.
Since netback required at least (MAX_SKB_FRAGS + 1) slots, IPXE could
not receive any packets.

a) If GSO is not enabled on the VIF, fewer guest Rx slots are required
   for the largest possible packet.  Calculate the required slots
   based on the maximum GSO size or the MTU.

   This calculation of the number of required slots relies on
   1650d545 (xen-netback: always fully coalesce guest Rx packets)
   which present in 4.0-rc1 and later.

b) Reduce the Rx stall detection to checking for at least one
   available Rx request.  This is fine since we're predominately
   concerned with detecting interfaces which are down and thus have
   zero available Rx requests.
Signed-off-by: NDavid Vrabel <david.vrabel@citrix.com>
Reviewed-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Xen's PV network protocol includes messages to add/remove ethernet
multicast addresses to/from a filter list in the backend. This allows
the frontend to request the backend only forward multicast packets
which are of interest thus preventing unnecessary noise on the shared
ring.

The canonical netif header in git://xenbits.xen.org/xen.git specifies
the message format (two more XEN_NETIF_EXTRA_TYPEs) so the minimal
necessary changes have been pulled into include/xen/interface/io/netif.h.

To prevent the frontend from extending the multicast filter list
arbitrarily a limit (XEN_NETBK_MCAST_MAX) has been set to 64 entries.
This limit is not specified by the protocol and so may change in future.
If the limit is reached then the next XEN_NETIF_EXTRA_TYPE_MCAST_ADD
sent by the frontend will be failed with NETIF_RSP_ERROR.
Signed-off-by: NPaul Durrant <paul.durrant@citrix.com>
Cc: Ian Campbell <ian.campbell@citrix.com>
Cc: Wei Liu <wei.liu2@citrix.com>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

There are two duplicated xenvif_zerocopy_callback() definitions.
Remove one of them.
Signed-off-by: NLiang Li <liang.z.li@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

With the current netback, the bandwidth limiter's parameters are only
settable during vif setup time.  This patch register a watch on them, and
thus makes them runtime changeable.

When the watch fires, the timer is reset.  The timer's mutex is used for
fencing the change.

Cc: Anthony Liguori <aliguori@amazon.com>
Signed-off-by: NImre Palik <imrep@amazon.de>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Always fully coalesce guest Rx packets into the minimum number of ring
slots.  Reducing the number of slots per packet has significant
performance benefits when receiving off-host traffic.

Results from XenServer's performance benchmarks:

                         Baseline    Full coalesce
Interhost VM receive      7.2 Gb/s   11 Gb/s
Interhost aggregate      24 Gb/s     24 Gb/s
Intrahost single stream  14 Gb/s     14 Gb/s
Intrahost aggregate      34 Gb/s     34 Gb/s

However, this can increase the number of grant ops per packet which
decreases performance of backend (dom0) to VM traffic (by ~10%)
/unless/ grant copy has been optimized for adjacent ops with the same
source or destination (see "grant-table: defer releasing pages
acquired in a grant copy"[1] expected in Xen 4.6).

[1] http://lists.xen.org/archives/html/xen-devel/2015-01/msg01118.htmlSigned-off-by: NDavid Vrabel <david.vrabel@citrix.com>
Acked-by: NIan Campbell <ian.campbell@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Commit bc96f648 (xen-netback: make
feature-rx-notify mandatory) incorrectly assumed that there were no
frontends in use that did not support this feature.  But the frontend
driver in MiniOS does not and since this is used by (qemu) stubdoms,
these stopped working.

Netback sort of works as-is in this mode except:

- If there are no Rx requests and the internal Rx queue fills, only
  the drain timeout will wake the thread.  The default drain timeout
  of 10 s would give unacceptable pauses.

- If an Rx stall was detected and the internal Rx queue is drained,
  then the Rx thread would never wake.

Handle these two cases (when feature-rx-notify is disabled) by:

- Reducing the drain timeout to 30 ms.

- Disabling Rx stall detection.
Reported-by: NJohn <jw@nuclearfallout.net>
Tested-by: NJohn <jw@nuclearfallout.net>
Signed-off-by: NDavid Vrabel <david.vrabel@citrix.com>
Reviewed-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If a frontend not receiving packets it is useful to detect this and
turn off the carrier so packets are dropped early instead of being
queued and drained when they expire.

A to-guest queue is stalled if it doesn't have enough free slots for a
an extended period of time (default 60 s).

If at least one queue is stalled, the carrier is turned off (in the
expectation that the other queues will soon stall as well).  The
carrier is only turned on once all queues are ready.

When the frontend connects, all the queues start in the stalled state
and only become ready once the frontend queues enough Rx requests.
Signed-off-by: NDavid Vrabel <david.vrabel@citrix.com>
Reviewed-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Netback needs to discard old to-guest skb's (guest Rx queue drain) and
it needs detect guest Rx stalls (to disable the carrier so packets are
discarded earlier), but the current implementation is very broken.

1. The check in hard_start_xmit of the slot availability did not
   consider the number of packets that were already in the guest Rx
   queue.  This could allow the queue to grow without bound.

   The guest stops consuming packets and the ring was allowed to fill
   leaving S slot free.  Netback queues a packet requiring more than S
   slots (ensuring that the ring stays with S slots free).  Netback
   queue indefinately packets provided that then require S or fewer
   slots.

2. The Rx stall detection is not triggered in this case since the
   (host) Tx queue is not stopped.

3. If the Tx queue is stopped and a guest Rx interrupt occurs, netback
   will consider this an Rx purge event which may result in it taking
   the carrier down unnecessarily.  It also considers a queue with
   only 1 slot free as unstalled (even though the next packet might
   not fit in this).

The internal guest Rx queue is limited by a byte length (to 512 Kib,
enough for half the ring).  The (host) Tx queue is stopped and started
based on this limit.  This sets an upper bound on the amount of memory
used by packets on the internal queue.

This allows the estimatation of the number of slots for an skb to be
removed (it wasn't a very good estimate anyway).  Instead, the guest
Rx thread just waits for enough free slots for a maximum sized packet.

skbs queued on the internal queue have an 'expires' time (set to the
current time plus the drain timeout).  The guest Rx thread will detect
when the skb at the head of the queue has expired and discard expired
skbs.  This sets a clear upper bound on the length of time an skb can
be queued for.  For a guest being destroyed the maximum time needed to
wait for all the packets it sent to be dropped is still the drain
timeout (10 s) since it will not be sending new packets.

Rx stall detection is reintroduced in a later commit.
Signed-off-by: NDavid Vrabel <david.vrabel@citrix.com>
Reviewed-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Frontends that do not provide feature-rx-notify may stall because
netback depends on the notification from frontend to wake the guest Rx
thread (even if can_queue is false).

This could be fixed but feature-rx-notify was introduced in 2006 and I
am not aware of any frontends that do not implement this.
Signed-off-by: NDavid Vrabel <david.vrabel@citrix.com>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Reference count the number of packets in host stack, so that we don't
stop the deallocation thread too early. If not, we can end up with
xenvif_free permanently waiting for deallocation thread to unmap grefs.
Reported-by: NThomas Leonard <talex5@gmail.com>
Signed-off-by: NWei Liu <wei.liu2@citrix.com>
Cc: Ian Campbell <ian.campbell@citrix.com>
Cc: Zoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Currently when the guest is not able to receive more packets, qdisc layer starts
a timer, and when it goes off, qdisc is started again to deliver a packet again.
This is a very slow way to drain the queues, consumes unnecessary resources and
slows down other guests shutdown.
This patch change the behaviour by turning the carrier off when that timer
fires, so all the packets are freed up which were stucked waiting for that vif.
Instead of the rx_queue_purge bool it uses the VIF_STATUS_RX_PURGE_EVENT bit to
signal the thread that either the timeout happened or an RX interrupt arrived,
so the thread can check what it should do. It also disables NAPI, so the guest
can't transmit, but leaves the interrupts on, so it can resurrect.
Only the queues which brought down the interface can enable it again, the bit
QUEUE_STATUS_RX_STALLED makes sure of that.
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: NDavid Vrabel <david.vrabel@citrix.com>
Cc: netdev@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: xen-devel@lists.xenproject.org
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch introduces a new state bit VIF_STATUS_CONNECTED to track whether the
vif is in a connected state. Using carrier will not work with the next patch
in this series, which aims to turn the carrier temporarily off if the guest
doesn't seem to be able to receive packets.
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: NDavid Vrabel <david.vrabel@citrix.com>
Cc: netdev@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: xen-devel@lists.xenproject.org

v2:
- rename the bitshift type to "enum state_bit_shift" here, not in the next patch
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adds debugfs capabilities to netback. There used to be a similar
patch floating around for classic kernel, but it used procfs. It is based on a
very similar blkback patch.
It creates xen-netback/[vifname]/io_ring_q[queueno] files, reading them output
various ring variables etc. Writing "kick" into it imitates an interrupt
happened, it can be useful to check whether the ring is just stalled due to a
missed interrupt.
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Cc: netdev@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: xen-devel@lists.xenproject.org
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The original code uses netdev->real_num_tx_queues to bookkeep number of
queues and invokes netif_set_real_num_tx_queues to set the number of
queues. However, netif_set_real_num_tx_queues doesn't allow
real_num_tx_queues to be smaller than 1, which means setting the number
to 0 will not work and real_num_tx_queues is untouched.

This is bogus when xenvif_free is invoked before any number of queues is
allocated. That function needs to iterate through all queues to free
resources. Using the wrong number of queues results in NULL pointer
dereference.

So we bookkeep the number of queues in xen-netback to solve this
problem. This fixes a regression introduced by multiqueue patchset in
3.16-rc1.

There's another bug in original code that the real number of RX queues
is never set. In current Xen multiqueue design, the number of TX queues
and RX queues are in fact the same. We need to set the numbers of TX and
RX queues to the same value.

Also remove xenvif_select_queue and leave queue selection to core
driver, as suggested by David Miller.
Reported-by: NBoris Ostrovsky <boris.ostrovsky@oracle.com>
Signed-off-by: NWei Liu <wei.liu2@citrix.com>
CC: Ian Campbell <ian.campbell@citrix.com>
CC: Paul Durrant <paul.durrant@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Builds on the refactoring of the previous patch to implement multiple
queues between xen-netfront and xen-netback.

Writes the maximum supported number of queues into XenStore, and reads
the values written by the frontend to determine how many queues to use.

Ring references and event channels are read from XenStore on a per-queue
basis and rings are connected accordingly.

Also adds code to handle the cleanup of any already initialised queues
if the initialisation of a subsequent queue fails.
Signed-off-by: NAndrew J. Bennieston <andrew.bennieston@citrix.com>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In preparation for multi-queue support in xen-netback, move the
queue-specific data from struct xenvif into struct xenvif_queue, and
update the rest of the code to use this.

Also adds loops over queues where appropriate, even though only one is
configured at this point, and uses alloc_netdev_mq() and the
corresponding multi-queue netif wake/start/stop functions in preparation
for multiple active queues.

Finally, implements a trivial queue selection function suitable for
ndo_select_queue, which simply returns 0 for a single queue and uses
skb_get_hash() to compute the queue index otherwise.
Signed-off-by: NAndrew J. Bennieston <andrew.bennieston@citrix.com>
Signed-off-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This array was allocated separately in commit ac3d5ac2 ("xen-netback:
fix guest-receive-side array sizes") due to it being very large, and a
struct xenvif is allocated as the netdev_priv part of a struct
net_device, i.e. via kmalloc() but falling back to vmalloc() if the
initial alloc. fails.

In preparation for the multi-queue patches, where this array becomes
part of struct xenvif_queue and is always allocated through vzalloc(),
move this back into the struct xenvif.
Signed-off-by: NAndrew J. Bennieston <andrew.bennieston@citrix.com>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When the NAPI budget was not all used, xenvif_poll() would call
napi_complete() /after/ enabling the interrupt.  This resulted in a
race between the napi_complete() and the napi_schedule() in the
interrupt handler.  The use of local_irq_save/restore() avoided by
race iff the handler is running on the same CPU but not if it was
running on a different CPU.

Fix this properly by calling napi_complete() before reenabling
interrupts (in the xenvif_napi_schedule_or_enable_irq() call).
Signed-off-by: NDavid Vrabel <david.vrabel@citrix.com>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Acked-by: NIan Campbell <ian.campbell@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

An old inefficiency of the TX path that we are grant mapping the first slot,
and then copy the header part to the linear area. Instead, doing a grant copy
for that header straight on is more reasonable. Especially because there are
ongoing efforts to make Xen avoiding TLB flush after unmap when the page were
not touched in Dom0. In the original way the memcpy ruined that.
The key changes:
- the vif has a tx_copy_ops array again
- xenvif_tx_build_gops sets up the grant copy operations
- we don't have to figure out whether the header and first frag are on the same
  grant mapped page or not
Note, we only grant copy PKT_PROT_LEN bytes from the first slot, the rest (if
any) will be on the first frag, which is grant mapped. If the first slot is
smaller than PKT_PROT_LEN, then we grant copy that, and later __pskb_pull_tail
will pull more from the frags (if any)
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Reviewed-by: NPaul Durrant <paul.durrant@citrix.com>
Acked-by: NIan Campbell <ian.campbell@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When netback discovers frontend is sending malformed packet it will
disables the interface which serves that frontend.

However disabling a network interface involving taking a mutex which
cannot be done in softirq context, so we need to defer this process to
kthread context.

This patch does the following:
1. introduce a flag to indicate the interface is disabled.
2. check that flag in TX path, don't do any work if it's true.
3. check that flag in RX path, turn off that interface if it's true.

The reason to disable it in RX path is because RX uses kthread. After
this change the behavior of netback is still consistent -- it won't do
any TX work for a rogue frontend, and the interface will be eventually
turned off.

Also change a "continue" to "break" after xenvif_fatal_tx_err, as it
doesn't make sense to continue processing packets if frontend is rogue.

This is a fix for XSA-90.
Reported-by: NTörök Edwin <edwin@etorok.net>
Signed-off-by: NWei Liu <wei.liu2@citrix.com>
Cc: Ian Campbell <ian.campbell@citrix.com>
Reviewed-by: NDavid Vrabel <david.vrabel@citrix.com>
Acked-by: NIan Campbell <ian.campbell@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Since the early days TX stops if there isn't enough free pending slots to
consume a maximum sized (slot-wise) packet. Probably the reason for that is to
avoid the case when we don't have enough free pending slot in the ring to finish
the packet. But if we make sure that the pending ring has the same size as the
shared ring, that shouldn't really happen. The frontend can only post packets
which fit the to the free space of the shared ring. If it doesn't, the frontend
has to stop, as it can only increase the req_prod when the whole packet fits
onto the ring.
This patch avoid using this checking, makes sure the 2 ring has the same size,
and remove a checking from the callback. As now we don't stop the NAPI instance
on this condition, we don't have to wake it up if we free pending slots up.
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This reverts commit e9275f5e. This commit is the
last in the netback grant mapping series, and it tries to do more aggressive
aggreagtion of unmap operations. However practical use showed almost no
positive effect, whilst with certain frontends it causes significant performance
regression.
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Acked-by: NIan Campbell <ian.campbell@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Unmapping causes TLB flushing, therefore we should make it in the largest
possible batches. However we shouldn't starve the guest for too long. So if
the guest has space for at least two big packets and we don't have at least a
quarter ring to unmap, delay it for at most 1 milisec.
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

A malicious or buggy guest can leave its queue filled indefinitely, in which
case qdisc start to queue packets for that VIF. If those packets came from an
another guest, it can block its slots and prevent shutdown. To avoid that, we
make sure the queue is drained in every 10 seconds.
The QDisc queue in worst case takes 3 round to flush usually.
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Xen network protocol had implicit dependency on MAX_SKB_FRAGS. Netback has to
handle guests sending up to XEN_NETBK_LEGACY_SLOTS_MAX slots. To achieve that:
- create a new skb
- map the leftover slots to its frags (no linear buffer here!)
- chain it to the previous through skb_shinfo(skb)->frag_list
- map them
- copy and coalesce the frags into a brand new one and send it to the stack
- unmap the 2 old skb's pages

It's also introduces new stat counters, which help determine how often the guest
sends a packet with more than MAX_SKB_FRAGS frags.

NOTE: if bisect brought you here, you should apply the series up until
"xen-netback: Timeout packets in RX path", otherwise malicious guests can block
other guests by not releasing their sent packets.
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

These counters help determine how often the buffers had to be copied. Also
they help find out if packets are leaked, as if "sent != success + fail",
there are probably packets never freed up properly.

NOTE: if bisect brought you here, you should apply the series up until
"xen-netback: Timeout packets in RX path", otherwise Windows guests can't work
properly and malicious guests can block other guests by not releasing their sent
packets.
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

These became obsolete with grant mapping. I've left intentionally the
indentations in this way, to improve readability of previous patches.

NOTE: if bisect brought you here, you should apply the series up until
"xen-netback: Timeout packets in RX path", otherwise Windows guests can't work
properly and malicious guests can block other guests by not releasing their sent
packets.
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch introduces grant mapping on netback TX path. It replaces grant copy
operations, ditching grant copy coalescing along the way. Another solution for
copy coalescing is introduced in "xen-netback: Handle guests with too many
frags", older guests and Windows can broke before that patch applies.
There is a callback (xenvif_zerocopy_callback) from core stack to release the
slots back to the guests when kfree_skb or skb_orphan_frags called. It feeds a
separate dealloc thread, as scheduling NAPI instance from there is inefficient,
therefore we can't do dealloc from the instance.
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

RX path need to know if the SKB fragments are stored on pages from another
domain.
Logically this patch should be after introducing the grant mapping itself, as
it makes sense only after that. But to keep bisectability, I moved it here. It
shouldn't change any functionality here. xenvif_zerocopy_callback and
ubuf_to_vif are just stubs here, they will be introduced properly later on.
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch contains a few bits of refactoring before introducing the grant
mapping changes:
- introducing xenvif_tx_pending_slots_available(), as this is used several
  times, and will be used more often
- rename the thread to vifX.Y-guest-rx, to signify it does RX work from the
  guest point of view
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The recent patch to fix receive side flow control
(11b57f90: xen-netback: stop vif thread
spinning if frontend is unresponsive) solved the spinning thread problem,
however caused an another one. The receive side can stall, if:
- [THREAD] xenvif_rx_action sets rx_queue_stopped to true
- [INTERRUPT] interrupt happens, and sets rx_event to true
- [THREAD] then xenvif_kthread sets rx_event to false
- [THREAD] rx_work_todo doesn't return true anymore

Also, if interrupt sent but there is still no room in the ring, it take quite a
long time until xenvif_rx_action realize it. This patch ditch that two variable,
and rework rx_work_todo. If the thread finds it can't fit more skb's into the
ring, it saves the last slot estimation into rx_last_skb_slots, otherwise it's
kept as 0. Then rx_work_todo will check if:
- there is something to send to the ring (like before)
- there is space for the topmost packet in the queue

I think that's more natural and optimal thing to test than two bool which are
set somewhere else.
Signed-off-by: NZoltan Kiss <zoltan.kiss@citrix.com>
Reviewed-by: NPaul Durrant <paul.durrant@citrix.com>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The recent patch to improve guest receive side flow control (ca2f09f2) had a
slight flaw in the wait condition for the vif thread in that any remaining
skbs in the guest receive side netback internal queue would prevent the
thread from sleeping. An unresponsive frontend can lead to a permanently
non-empty internal queue and thus the thread will spin. In this case the
thread should really sleep until the frontend becomes responsive again.

This patch adds an extra flag to the vif which is set if the shared ring
is full and cleared when skbs are drained into the shared ring. Thus,
if the thread runs, finds the shared ring full and can make no progress the
flag remains set. If the flag remains set then the thread will sleep,
regardless of a non-empty queue, until the next event from the frontend.
Signed-off-by: NPaul Durrant <paul.durrant@citrix.com>
Cc: Wei Liu <wei.liu2@citrix.com>
Cc: Ian Campbell <ian.campbell@citrix.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The sizes chosen for the metadata and grant_copy_op arrays on the guest
receive size are wrong;

- The meta array is needlessly twice the ring size, when we only ever
  consume a single array element per RX ring slot
- The grant_copy_op array is way too small. It's sized based on a bogus
  assumption: that at most two copy ops will be used per ring slot. This
  may have been true at some point in the past but it's clear from looking
  at start_new_rx_buffer() that a new ring slot is only consumed if a frag
  would overflow the current slot (plus some other conditions) so the actual
  limit is MAX_SKB_FRAGS grant_copy_ops per ring slot.

This patch fixes those two sizing issues and, because grant_copy_ops grows
so much, it pulls it out into a separate chunk of vmalloc()ed memory.
Signed-off-by: NPaul Durrant <paul.durrant@citrix.com>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Cc: Ian Campbell <ian.campbell@citrix.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The way that flow control works without this patch is that, in start_xmit()
the code uses xenvif_count_skb_slots() to predict how many slots
xenvif_gop_skb() will consume and then adds this to a 'req_cons_peek'
counter which it then uses to determine if the shared ring has that amount
of space available by checking whether 'req_prod' has passed that value.
If the ring doesn't have space the tx queue is stopped.
xenvif_gop_skb() will then consume slots and update 'req_cons' and issue
responses, updating 'rsp_prod' as it goes. The frontend will consume those
responses and post new requests, by updating req_prod. So, req_prod chases
req_cons which chases rsp_prod, and can never exceed that value. Thus if
xenvif_count_skb_slots() ever returns a number of slots greater than
xenvif_gop_skb() uses, req_cons_peek will get to a value that req_prod cannot
possibly achieve (since it's limited by the 'real' req_cons) and, if this
happens enough times, req_cons_peek gets more than a ring size ahead of
req_cons and the tx queue then remains stopped forever waiting for an
unachievable amount of space to become available in the ring.

Having two routines trying to calculate the same value is always going to be
fragile, so this patch does away with that. All we essentially need to do is
make sure that we have 'enough stuff' on our internal queue without letting
it build up uncontrollably. So start_xmit() makes a cheap optimistic check
of how much space is needed for an skb and only turns the queue off if that
is unachievable. net_rx_action() is the place where we could do with an
accurate predicition but, since that has proven tricky to calculate, a cheap
worse-case (but not too bad) estimate is all we really need since the only
thing we *must* prevent is xenvif_gop_skb() consuming more slots than are
available.

Without this patch I can trivially stall netback permanently by just doing
a large guest to guest file copy between two Windows Server 2008R2 VMs on a
single host.

Patch tested with frontends in:
- Windows Server 2008R2
- CentOS 6.0
- Debian Squeeze
- Debian Wheezy
- SLES11
Signed-off-by: NPaul Durrant <paul.durrant@citrix.com>
Cc: Wei Liu <wei.liu2@citrix.com>
Cc: Ian Campbell <ian.campbell@citrix.com>
Cc: David Vrabel <david.vrabel@citrix.com>
Cc: Annie Li <annie.li@oracle.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Acked-by: NWei Liu <wei.liu2@citrix.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>