If the packet uses multiple descriptors and its descriptor indices are
wrapped, the first descriptor flag is not updated last, which may cause
virtio read the incomplete packet. For example, given a packet uses 64
descriptors, and virtio ring size is 256, and its descriptor indices are
224~255 and 0~31, current implementation will update 224~255 descriptor
flags earlier than 0~31 descriptor flags.
This patch fixes this issue by updating descriptor flags in one loop,
so that the first descriptor flag is always updated last.
Fixes: 873e8dad6f ("vhost: support packed ring in async datapath")
Signed-off-by: Jiayu Hu <jiayu.hu@intel.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Fix spelling errors in comments including doxygen found using codespell.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Vladimir Medvedkin <vladimir.medvedkin@intel.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
Acked-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
Acked-by: Chenbo Xia <chenbo.xia@intel.com>
As previously announced, this patch renames struct
vhost_device_ops to struct rte_vhost_device_ops.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
This patch marks the vDPA driver APIs as internal and
rename the corresponding header file to vdpa_driver.h.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
This patch increases the number of IO vectors for the
asynchronous data path from 512 to 2048. It has been
reported during testing the starvation of IO vectors
during iperf benchmark with 64KB packet size.
As there are no direct relationship between
VHOST_MAX_ASYNC_VEC and BUF_VECTOR_MAX, this patch also
assign VHOST_MAX_ASYNC_VEC value directly instead of being
a multiple of BUF_VECTOR_MAX.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
This patches merges copy_mbuf_to_desc() used by the sync
path with async_mbuf_to_desc() used by the async path.
Most of these complex functions are identical, so merging
them will make the maintenance easier.
In order not to degrade performance, the patch introduces
a boolean function parameter to specify whether it is called
in async context. This boolean is statically passed to this
always-inlined function, so the compiler will optimize this
out.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
This patch extracts the descriptors buffers filling
from copy_mbuf_to_desc() into a dedicated function as a
preliminary step of merging copy_mubf_to_desc() and
async_mbuf_to_desc().
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
This patch extracts the IO vectors filling from
async_mbuf_to_desc() into a dedicated function as a
preliminary step of merging copy_mubf_to_desc() and
async_mbuf_to_desc().
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
This patch reworks the function getting the index
for the first packet in-flight.
When this index turns out to be zero, let's use the simple
path. Doing that avoid having to do a modulo with the
virtqueue size.
The patch also rename the function for better clarification,
and only pass the virtqueue metadata pointer, as all the
needed information are stored there.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
vhost_poll_enqueue_completed() assumes some inflight
packets could have been completed in a previous call but
not returned to the application. But this is not the case,
since check_completed_copies callback is never called with
more than the current count as argument.
In other words, async->last_pkts_n is always 0. Removing it
greatly simplifies the function.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
Now that IO vectors iterator have been simplified, the
rte_vhost_async_desc struct only contains a pointer on
the iterator array stored in the async metadata.
This patch removes it, and pass directly the iterators
array pointer to the transfer_data callback. Doing that,
we avoid declaring the descriptor array in the stack, and
also avoid the cost of filling it.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
IO vectors and their iterators arrays were part of the
async metadata but not their indexes.
In order to makes this more consistent, the patch adds the
indexes to the async metadata. Doing that, we can avoid
triggering DMA transfer within the loop as it IO vector
index overflow is now prevented in the async_mbuf_to_desc()
function.
Note that previous detection mechanism was broken
since the overflow already happened when detected, so OOB
memory access would already have happened.
With this changes done, virtio_dev_rx_async_submit_split()
and virtio_dev_rx_async_submit_packed() can be further
simplified.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
Offset and count fields are unused and so can be removed.
The offset field was actually in the Vhost example, but
in a way that does not make sense.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
This patch introduces rte_vhost_iovec struct that contains
both source and destination addresses since we always have
a 1:1 mapping between source and destination. While using
the standard iovec struct might have seemed better, having
to duplicate IO vectors and its iterators is memory
inefficient and make the implementation more complex.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
Reaching the async batch threshold was one of the condition
to trigger the DMA transfer. However, this condition was
never met since the threshold value is 32, same as the
MAX_PKT_BURST value.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
This patch splits the iterator arrays in two, one for
source and one for destination. The goal is make the code
easier to understand.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
IO vectors implementation is unnecessarily complex, mixing
source and destinations vectors in the same array.
This patch declares two arrays, one for the source and one
for the destination. It also gets rid of seg_awaits variable
in both packed and split implementation, which is the same
as iovec_idx.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
This patch moves async_inflight_info struct to internal
header since it should not be part of the API.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
This patch moves async-related metadata from vhost_virtqueue
to a dedicated struct. It makes it clear which fields are
async related, and also saves some memory when async feature
is not in use.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
This commit defines rte_vhost_power_monitor_cond which is used to pass
some information to vhost driver. The information is including the
address to monitor, the expected value, the mask to extract value read
from 'addr', the value size of monitor address, the match flag used to
distinguish the value used to match something or not match something.
Vhost driver can use these information to fill rte_power_monitor_cond.
Signed-off-by: Miao Li <miao.li@intel.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Acked-by: David Hunt <david.hunt@intel.com>
Async DMA map status flag was added to prevent the unnecessary unmap
when DMA devices bound to kernel driver. This brings maintenance cost
for a lot of code. This patch removes the DMA map status by using
rte_errno instead.
This patch relies on the following patch to fix a partial
unmap check in vfio unmapping API.
[1] https://www.mail-archive.com/dev@dpdk.org/msg226464.html
Signed-off-by: Xuan Ding <xuan.ding@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Fix the mbuf offload flags namespace by adding an RTE_ prefix to the
name. The old flags remain usable, but a deprecation warning is issued
at compilation.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>
Acked-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Acked-by: Somnath Kotur <somnath.kotur@broadcom.com>
The use of IOMMU has many advantages, such as isolation and address
translation. This patch extends the capability of DMA engine to use
IOMMU if the DMA engine is bound to vfio.
When set memory table, the guest memory will be mapped
into the default container of DPDK.
Signed-off-by: Xuan Ding <xuan.ding@intel.com>
Tested-by: Yvonne Yang <yvonnex.yang@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
The index in rte_vhost_set_last_inflight_io_split is from
the frontend driver, check if it's in the virtqueue range.
Fixes: bb0c2de960 ("vhost: add APIs to operate inflight ring")
Cc: stable@dpdk.org
Signed-off-by: Li Feng <fengli@smartx.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Fix the mempool flags namespace by adding an RTE_ prefix to the name.
The old flags remain usable, to be deprecated in the future.
Flag MEMPOOL_F_NON_IO added in the release is just renamed to have RTE_
prefix.
Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
In some function definitions, adjust return type and function name on
a separate line to be consistent with DPDK coding style.
Signed-off-by: Xuan Ding <xuan.ding@intel.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Caught by code review, this copy is unnecessary.
Signed-off-by: David Marchand <david.marchand@redhat.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Old IOVA cache entries are left when there is a change on virtio driver
in VM. In case that all these old entries have iova addresses lesser
than new iova entries, vhost code will need to iterate all the cache to
find the new ones. In case of just a new iova entry needed for the new
translations, this condition will last forever.
This has been observed in virtio-net to testpmd's vfio-pci driver
transition, reducing the performance from more than 10Mpps to less than
0.07Mpps if the hugepage address was higher than the networking
buffers. Since all new buffers are contained in this new gigantic page,
vhost needs to scan IOTLB_CACHE_SIZE - 1 for each translation at worst.
Fixes: 69c90e98f4 ("vhost: enable IOMMU support")
Cc: stable@dpdk.org
Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Reported-by: Pei Zhang <pezhang@redhat.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Currently there are some public headers that include 'sys/queue.h', which
is not POSIX, but usually provided by the Linux/BSD system library.
(Not in POSIX.1, POSIX.1-2001, or POSIX.1-2008. Present on the BSDs.)
The file is missing on Windows. During the Windows build, DPDK uses a
bundled copy, so building a DPDK library works fine. But when OVS or other
applications use DPDK as a library, because some DPDK public headers
include 'sys/queue.h', on Windows, it triggers an error due to no such
file.
One solution is to install the 'lib/eal/windows/include/sys/queue.h' into
Windows environment, such as [1]. However, this means DPDK exports the
functionalities of 'sys/queue.h' into the environment, which might cause
symbols, macros, headers clashing with other applications.
The patch fixes it by removing the "#include <sys/queue.h>" from
DPDK public headers, so programs including DPDK headers don't depend
on the system to provide 'sys/queue.h'. When these public headers use
macros such as TAILQ_xxx, we replace it by the ones with RTE_ prefix.
For Windows, we copy the definitions from <sys/queue.h> to rte_os.h
in Windows EAL. Note that these RTE_ macros are compatible with
<sys/queue.h>, both at the level of API (to use with <sys/queue.h>
macros in C files) and ABI (to avoid breaking it).
Additionally, the TAILQ_FOREACH_SAFE is not part of <sys/queue.h>,
the patch replaces it with RTE_TAILQ_FOREACH_SAFE.
[1] http://mails.dpdk.org/archives/dev/2021-August/216304.html
Suggested-by: Nick Connolly <nick.connolly@mayadata.io>
Suggested-by: Dmitry Kozlyuk <dmitry.kozliuk@gmail.com>
Signed-off-by: William Tu <u9012063@gmail.com>
Acked-by: Dmitry Kozlyuk <dmitry.kozliuk@gmail.com>
Acked-by: Narcisa Vasile <navasile@linux.microsoft.com>
As reported by symbol bot, APIs listed in this patch have been
experimental for more than two years. This patch promotes these
18 APIs to stable.
Signed-off-by: Chenbo Xia <chenbo.xia@intel.com>
Acked-by: Ray Kinsella <mdr@ashroe.eu>
Acked-by: Kevin Traynor <ktraynor@redhat.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Add log print of socket path in vhost_user_add_connection.
It's useful when adding a mass of socket connections,
because the information of every connection is clearer.
Fixes: 8f972312b8 ("vhost: support vhost-user")
Cc: stable@dpdk.org
Signed-off-by: Gaoxiang Liu <liugaoxiang@huawei.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
The rte_vhost_driver_unregister() and vhost_user_read_cb()
can be called at the same time by 2 threads.
when memory of vsocket is freed in rte_vhost_driver_unregister(),
the invalid memory of vsocket is accessed in vhost_user_read_cb().
It's a bug of both mode for vhost as server or client.
E.g., vhostuser port is created as server.
Thread1 calls rte_vhost_driver_unregister().
Before the listen fd is deleted from poll waiting fds,
"vhost-events" thread then calls vhost_user_server_new_connection(),
then a new conn fd is added in fdset when trying to reconnect.
"vhost-events" thread then calls vhost_user_read_cb() and
accesses invalid memory of socket while thread1 frees the memory of
vsocket.
E.g., vhostuser port is created as client.
Thread1 calls rte_vhost_driver_unregister().
Before vsocket of reconn is deleted from reconn list,
"vhost_reconn" thread then calls vhost_user_add_connection()
then a new conn fd is added in fdset when trying to reconnect.
"vhost-events" thread then calls vhost_user_read_cb() and
accesses invalid memory of socket while thread1 frees the memory of
vsocket.
The fix is to move the "fdset_try_del" in front of free memory of conn,
then avoid the race condition.
The core trace is:
Program terminated with signal 11, Segmentation fault.
Fixes: 52d874dc67 ("vhost: fix crash on closing in client mode")
Cc: stable@dpdk.org
Signed-off-by: Gaoxiang Liu <liugaoxiang@huawei.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Copy threshold has been introduced in async vhost data
path to select the appropriate copy engine to do copies
for higher efficiency.
However, it may cause packets ordering issues and also
introduces performance unpredictability.
Therefore, this patch removes copy threshold support in
async vhost data path.
Signed-off-by: Jiayu Hu <jiayu.hu@intel.com>
Signed-off-by: Cheng Jiang <cheng1.jiang@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Added macro to print six bytes of MAC address.
The MAC addresses will be printed in upper case
hexadecimal format.
In case there is a specific check for lower case
MAC address, the user may need to make a change in
such test case after this patch.
Signed-off-by: Aman Deep Singh <aman.deep.singh@intel.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Start a new release cycle with empty release notes.
The ABI version becomes 22.0.
The map files are updated to the new ABI major number (22).
The ABI exceptions are dropped and CI ABI checks are disabled because
compatibility is not preserved.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: David Marchand <david.marchand@redhat.com>
When the vhost-user frontend like Virtio-user tries to
reconnect to the restarted Vhost backend, the Vhost backend
segfaults when multiqueue is enabled.
This is caused by VHOST_USER_GET_VRING_BASE being called for
a virtqueue that has not been created before, causing a NULL
pointer dereferencing.
This patch adds the VHOST_USER_GET_VRING_BASE requests to
the list of requests that trigger queue pair allocations.
Fixes: 160cbc815b ("vhost: remove a hack on queue allocation")
Cc: stable@dpdk.org
Reported-by: Yinan Wang <yinan.wang@intel.com>
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Tested-by: Yinan Wang <yinan.wang@intel.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
When the guest memory is hotplugged, the vhost application which
enables DMA acceleration must stop DMA transfers before the vhost
re-maps the guest memory.
This patch is to notify the vhost application of stopping DMA
transfers.
Signed-off-by: Jiayu Hu <jiayu.hu@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Applications need to stop DMA transfers and finish all the inflight
packets when in VM memory hot-plug case and async vhost is used. This
patch is to provide an unsafe API to clear inflight packets which
are submitted to DMA engine in vhost async data path. Update the
program guide and release notes for virtqueue inflight packets clear
API in vhost lib.
Signed-off-by: Cheng Jiang <cheng1.jiang@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
The async vhost callback ops should return negative value when there
are something wrong in the callback, so the return type should be
changed into int32_t. The issue in vhost example is also fixed.
Fixes: cd6760da10 ("vhost: introduce async enqueue for split ring")
Fixes: 819a716858 ("vhost: fix async callback return type")
Fixes: 6b3c81db8b ("vhost: simplify async copy completion")
Fixes: abec60e711 ("examples/vhost: support vhost async data path")
Fixes: 6e9a9d2a02 ("examples/vhost: fix ioat dependency")
Fixes: 873e8dad6f ("vhost: support packed ring in async datapath")
Cc: stable@dpdk.org
Signed-off-by: Cheng Jiang <cheng1.jiang@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
This patch adds thread unsafe version for async register and
unregister functions.
Signed-off-by: Jiayu Hu <jiayu.hu@intel.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
This patch reworks the async configuration structure to improve code
readability. In addition, add preserved padding fields on the structure
for future usage.
Signed-off-by: Jiayu Hu <jiayu.hu@intel.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
The vhost notifies the application of device readiness via
vhost_user_notify_queue_state(), but calling this function
is not protected by the lock. This patch is to make this
function call lock protected.
Fixes: d0fcc38f5f ("vhost: improve device readiness notifications")
Cc: stable@dpdk.org
Signed-off-by: Jiayu Hu <jiayu.hu@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Unlike split ring, packed ring does not mandate the ring size
to be a power of 2. So we have to use a modulo operation when
wrapping ring index.
Fixes: 873e8dad6f ("vhost: support packed ring in async datapath")
Cc: stable@dpdk.org
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Acked-by: Cheng Jiang <cheng1.jiang@intel.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
This patch allows to check the amount of in-flight packets
for the vhost queue using async acceleration.
Signed-off-by: Jiayu Hu <jiayu.hu@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
We assume that in the sync path, if there is no buffer wrap in the
avail descriptors fetched in a batch, there is no buffer wrap in the
used descriptors which need to be written back in this batch, but
this assumption is wrong in the async path since there are inflight
descriptors which are processed by the DMA device.
This patch refactors the batch copy code and adds used ring buffer
wrap check as a batch copy condition to fix this issue.
Fixes: 873e8dad6f ("vhost: support packed ring in async datapath")
Cc: stable@dpdk.org
Signed-off-by: Cheng Jiang <cheng1.jiang@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
We introduced some new indexes in packed ring of async vhost. They
will eventually overflow and lead to errors if the ring size is not
a power of 2. This patch is to check and keep these indexes within a
reasonable range.
Fixes: 873e8dad6f ("vhost: support packed ring in async datapath")
Cc: stable@dpdk.org
Signed-off-by: Cheng Jiang <cheng1.jiang@intel.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
When parsing the virtio net header and packet header for dequeue offload,
we need to perform sanity check on the packet header to ensure:
- No out-of-boundary memory access.
- The packet header and virtio_net header are valid and aligned.
Fixes: d0cf91303d ("vhost: add Tx offload capabilities")
Cc: stable@dpdk.org
Signed-off-by: Xiao Wang <xiao.w.wang@intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Inflight metadata are allocated using glibc's calloc.
This patch converts them to rte_zmalloc_socket to take
care of the NUMA affinity.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
This patch saves the NUMA node the virtqueue is allocated
on at init time, in order to allocate all other data on the
same node.
While most of the data are allocated before numa_realloc()
is called and so the data will be reallocated properly, some
data like the log cache are most likely allocated after.
For the virtio device metadata, we decide to allocate them
on the same node as the VQ 0.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
This patch improves the numa_realloc() function by making use
of rte_realloc_socket(), which takes care of the memory copy
and freeing of the old data.
Suggested-by: David Marchand <david.marchand@redhat.com>
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Reviewed-by: Chenbo Xia <chenbo.xia@intel.com>
