The virtqueue_pop() implementation needs to check if the avail ring
contains some pending buffers. To perform this check, it is not
always necessary to fetch the avail_idx in the VQ memory, which is
expensive. This patch introduces a shadow variable tracking avail_idx
and modifies virtio_queue_empty() to access avail_idx in physical
memory only when necessary.
Signed-off-by: NVincenzo Maffione <v.maffione@gmail.com>
Message-Id: <b617d6459902773d9f4ab843bfaca764f5af8eda.1450218353.git.v.maffione@gmail.com>
Reviewed-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

Accessing used_idx in the VQ requires an expensive access to
guest physical memory. Before this patch, 3 accesses are normally
done for each pop/push/notify call. However, since the used_idx is
only written by us, we can track it in our internal data structure.
Signed-off-by: NVincenzo Maffione <v.maffione@gmail.com>
Message-Id: <3d062ec54e9a7bf9fb325c1fd693564951f2b319.1450218353.git.v.maffione@gmail.com>
Reviewed-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

Compared to vring, virtio has a performance penalty of 10%.  Fix it
by combining all the reads for a descriptor in a single address_space_read
call.  This also simplifies the code nicely.
Reviewed-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

Build the addresses and s/g lists on the stack, and then copy them
to a VirtQueueElement that is just as big as required to contain this
particular s/g list.  The cost of the copy is minimal compared to that
of a large malloc.

When virtqueue_map is used on the destination side of migration or on
loadvm, the iovecs have already been split at memory region boundary,
so we can just reuse the out_num/in_num we find in the file.
Reviewed-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

Allocate the arrays for in_addr/out_addr/in_sg/out_sg outside the
VirtQueueElement.  For now, virtqueue_pop and vring_pop keep
allocating a very large VirtQueueElement.
Reviewed-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

Move allocation to virtio functions also when loading/saving a
VirtQueueElement.  This will also let the load/save functions
keep backwards compatibility when the VirtQueueElement layout
is changed.
Reviewed-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

The return code of virtqueue_pop/vring_pop is unused except to check for
errors or 0.  We can thus easily move allocation inside the functions
and just return a pointer to the VirtQueueElement.

The advantage is that we will be able to allocate only the space that
is needed for the actual size of the s/g list instead of the full
VIRTQUEUE_MAX_SIZE items.  Currently VirtQueueElement takes about 48K
of memory, and this kind of allocation puts a lot of stress on malloc.
By cutting the size by two or three orders of magnitude, malloc can
use much more efficient algorithms.

The patch is pretty large, but changes to each device are testable
more or less independently.  Splitting it would mostly add churn.
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NCornelia Huck <cornelia.huck@de.ibm.com>

I misunderstood the vmstate macro definition when I reworked the
virtio .get/.put.
The VMSTATE_STRUCT_VARRAY_KNOWN, was described as being for "a
variable length array (i.e. _type *_field) but we know the
length".  However it actually specified operation for arrays embedded in
the struct (i.e. _type _field[]) since it lacked the VMS_POINTER
flag. This caused offset calculation to be completely off, examining and
potentially sending random data instead of the VirtQueue content.

Replace the otherwise unused VMSTATE_STRUCT_VARRAY_KNOWN with a
VMSTATE_STRUCT_VARRAY_POINTER_KNOWN that includes the VMS_POINTER flag
(so now actually doing what it advertises) and use it in the virtio
migration code.

Fixes and description as per Sascha's suggestions/debug.
Signed-off-by: NDr. David Alan Gilbert <dgilbert@redhat.com>
Reported-by: NSascha Silbe <silbe@linux.vnet.ibm.com>
Tested-By: NSascha Silbe <silbe@linux.vnet.ibm.com>
Reviewed-By: NSascha Silbe <silbe@linux.vnet.ibm.com>

Fixes: 50e5ae4d
Fixes: 2cf01486Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Tested-by: NCornelia Huck <cornelia.huck@de.ibm.com>

Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.

This commit was created with scripts/clean-includes.
Signed-off-by: NPeter Maydell <peter.maydell@linaro.org>
Message-id: 1453832250-766-15-git-send-email-peter.maydell@linaro.org

There's no such thing as "PCI queues" in the virtio core.
Signed-off-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

The 'virtqueue_state' and 'ringsize' can be saved using VMSTATE
macros rather than hand coded .get/.put
Signed-off-by: NDr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NAmit Shah <amit.shah@redhat.com>

We don't migrate the followings fields for virtio-pci:

uint32_t dfselect;
uint32_t gfselect;
uint32_t guest_features[2];
struct {
    uint16_t num;
    bool enabled;
    uint32_t desc[2];
    uint32_t avail[2];
    uint32_t used[2];
} vqs[VIRTIO_QUEUE_MAX];

This will confuse driver if migrating during initialization. Solves
this issue by:

- introduce transport specific callbacks to load and store extra
  virtqueue states.
- add a new subsection for virtio to migrate transport specific modern
  device state.
- implement pci specific callbacks.
- add a new property for virtio-pci for whether or not to migrate
  extra state.
- compat the migration for 2.4 and elder machine types

Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: NJason Wang <jasowang@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NCornelia Huck <cornelia.huck@de.ibm.com>

Deprecated in favor of virtqueue_map.
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NIgor Mammedov <imammedo@redhat.com>

Drop use of the deprecated virtio_map_sg in virtio core.
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NIgor Mammedov <imammedo@redhat.com>

virtio_map_sg currently fails if one of the entries it's mapping is
contigious in GPA but not HVA address space.  Introduce virtio_map which
handles this by splitting sg entries.

This new API generally turns out to be a good idea since it's harder to
misuse: at least in one case the existing one was used incorrectly.

This will still fail if there's no space left in the sg, but luckily max
queue size in use is currently 256, while max sg size is 1024, so we
should be OK even is all entries happen to cross a single DIMM boundary.

Won't work well with very small DIMM sizes, unfortunately:
e.g. this will fail with 4K DIMMs where a single
request might span a large number of DIMMs.

Let's hope these are uncommon - at least we are not breaking things.

Note: virtio-scsi calls virtio_map_sg on data loaded from network, and
validates input, asserting on failure.  Copy the validating code here -
it will be dropped from virtio-scsi in a follow-up patch.
Reported-by: NIgor Mammedov <imammedo@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: NIgor Mammedov <imammedo@redhat.com>

This patch introduces virtqueue_discard() to discard a descriptor and
unmap the sgs. This will be used by the patch that will discard
descriptor when packet is truncated.

Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: NJason Wang <jasowang@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

Factor out sg unmapping logic. This will be reused by the patch that
can discard descriptor.

Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Andrew James <andrew.james@hpe.com>
Signed-off-by: NJason Wang <jasowang@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

We allow guests to change the size of the virtqueue rings by supplying
a number of buffers that is different from the number of buffers the
device was initialized with. Current code has some problems, however,
since reset does not reset the ringsizes to the default values (as this
is not saved anywhere).

Let's extend the core code to keep track of the default ringsizes and
migrate them once the guest changed them for any of the virtqueues
for a device.
Reviewed-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

Being working on dataplane I notice something strange:

virtio_queue_get_avail_size() used a 64bit size index
for the calculation of the available ring size.

It is quite strange but it did work with the old calculation
of the avail ring, at most with performance penalty,
and I wonder where I missed something.

This patch let use a 16bit size as defined in virtio_ring.h
Signed-off-by: NPierre Morel <pmorel@linux.vnet.ibm.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

Commit ef546f12 ("virtio: add
feature checking helpers") introduced a helper __virtio_has_feature.
We don't want to use reserved identifiers, though, so let's
rename __virtio_has_feature to virtio_has_feature and virtio_has_feature
to virtio_vdev_has_feature.
Signed-off-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

1.0 does not requires physically-contiguous pages layout for a
virtqueue. So we could not infer avail and used from desc. This means
we need to migrate vring.avail and vring.used when host support virtio
1.0. This fixes malfunction of virtio 1.0 device after migration.

Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Cornelia Huck <cornelia.huck@de.ibm.com>
Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: NJason Wang <jasowang@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

We create optional sections with this patch.  But we already have
optional subsections.  Instead of having two mechanism that do the
same, we can just generalize it.

For subsections we just change:

- Add a needed function to VMStateDescription
- Remove VMStateSubsection (after removal of the needed function
  it is just a VMStateDescription)
- Adjust the whole tree, moving the needed function to the corresponding
  VMStateDescription
Signed-off-by: NJuan Quintela <quintela@redhat.com>

virtio 1.0 defines config space as LE,
as opposed to pre-1.0 which was native endian.

Add API for transports to execute word/dword accesses in
little endian format - will be useful for mmio
and pci (byte access is also wrapped, for completeness).

For simplicity, we still keep config in host native
endian format, byteswap to LE on guest access.
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>

Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>

virtio-1 allow setting of the FEATURES_OK status bit to fail if
the negotiated feature bits are inconsistent: let's fail
virtio_set_status() in that case and update virtio-ccw to post an
error to the guest.
Signed-off-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>

For virtio-1 devices, the driver must not attempt to set feature bits
after it set FEATURES_OK in the device status. Simply reject it in
that case.
Signed-off-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>

For virtio-1 devices, we allow a more complex queue layout that doesn't
require descriptor table and rings on a physically-contigous memory area:
add virtio_queue_set_rings() to allow transports to set this up.
Signed-off-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>

Add code that checks for the VERSION_1 feature bit in order to make
decisions about the device's endianness. This allows us to support
transitional devices.
Signed-off-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>

Commit "019a3edb virtio: make features 64bit wide" missed a few changes,
as I've noticed while trying to rebase the virtio-1 branch to latest
master.  This patch adds them.
Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

Make features 64bit wide everywhere.

On migration a full 64bit guest_features field is sent if one of the
high bits is set, in addition to the lower 32bit guest_features field
which must stay for compatibility reasons.  That way we send the lower
32 feature bits twice, but the code is simpler because we don't have
to split and compose the 64bit features into two 32bit fields.
Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

VIRTIO_PCI_QUEUE_MAX is not only used for pci, so rename it be generic.

Cc: Amit Shah <amit.shah@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: NJason Wang <jasowang@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

This patch introduces virtio_get_num_queues() which iterates the vqs
array and return the number of virtqueues used by device.
Signed-off-by: NJason Wang <jasowang@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

This patch passes error pointer to transport specific device_plugged()
callback. Through this way, device_plugged() can do some transport
specific check and fail. This will be uesd by following patches that
check the number of virtqueues against the transport limitation.

Cc: Cornelia Huck <cornelia.huck@de.ibm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Alexander Graf <agraf@suse.de>
Signed-off-by: NJason Wang <jasowang@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

Move host_features from the individual transport proxies into
the virtio device. Transports may continue to add feature bits
during device plugging.

This should it make easier to offer different sets of host features
for virtio-1/transitional support.
Tested-by: NShannon Zhao <shannon.zhao@linaro.org>
Signed-off-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

Currently we will try to traverse all virtqueues to find a subset that
using a specific vector. This is sub optimal when we will support
hundreds or even thousands of virtqueues. So this patch introduces a
method which could be used by transport to get all virtqueues that
using a same vector. This is done through QLISTs and the number of
QLISTs was queried through a transport specific method. When guest
setting vectors, the virtqueue will be linked and helpers for traverse
the list was also introduced.

The first user will be virtio pci which will use this to speed up
MSI-X masking and unmasking handling.

Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: NJason Wang <jasowang@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

We don't validate the existence of handle_output which may let a buggy
guest to trigger a SIGSEV easily. E.g:

1) write 10 to queue_sel to a virtio net device with only 1 queue
2) setup an arbitrary pfn
3) then notify queue 10

Fixing this by validating the existence of handle_output before.

Cc: qemu-stable@nongnu.org
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: NJason Wang <jasowang@redhat.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NDon Koch <dkoch@verizon.com>
Reviewed-by: NFam Zheng <famz@redhat.com>

Add a helper function for checking whether a bit is set in the guest
features for a vdev as well as one that works on a feature bit set.

Convert code that open-coded this: It cleans up the code and makes it
easier to extend the guest feature bits.
Signed-off-by: NCornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>

Drop a bunch of code duplicated from virtio_config.h and virtio_ring.h.
This makes us rename event index accessors which conflict,
as reusing the ones from virtio_ring.h isn't trivial.
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NThomas Huth <thuth@linux.vnet.ibm.com>

One of the annoyances of the current migration format is the fact that
it's not self-describing. In fact, it's not properly describing at all.
Some code randomly scattered throughout QEMU elaborates roughly how to
read and write a stream of bytes.

We discussed an idea during KVM Forum 2013 to add a JSON description of
the migration protocol itself to the migration stream. This patch
adds a section after the VM_END migration end marker that contains
description data on what the device sections of the stream are composed of.

This approach is backwards compatible with any QEMU version reading the
stream, because QEMU just stops reading after the VM_END marker and ignores
any data following it.

With an additional external program this allows us to decipher the
contents of any migration stream and hopefully make migration bugs easier
to track down.
Signed-off-by: NAlexander Graf <agraf@suse.de>
Signed-off-by: NAmit Shah <amit.shah@redhat.com>
Signed-off-by: NJuan Quintela <quintela@redhat.com>

The event idx in virtio is an effective way to reduce the number of
interrupts and exits of the guest. When the guest puts an request
into the virtio ring, it doesn't exit immediately to inform the
backend. Instead, the guest checks the "avail" event idx to determine
the notification.

In virtqueue_pop, when a request is poped, the current avail event
idx should be set to the number of vq->last_avail_idx.
Signed-off-by: NBin Wu <wu.wubin@huawei.com>
Reviewed-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Reviewed-by: NStefan Hajnoczi <stefanha@redhat.com>