DPDK 17.11 removed all rte_config.h includes
from rte_*.h headers, meaning we should either
use gcc param -include rte_config.h (just
like DPDK does), or include this file before
each other rte_*.h include. Since we're using
the latter approach in many places already,
I decided to follow it.
While here, also removed rte_vdev.h dependency
from rte_virtio/virtio_user.c. It's not used
anyway.
Change-Id: I865ee9f828211c03a60fd0446f7a418d5dddd140
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/387653
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
There are two separate abstraction layers:
* vsocket - which represents a unix domain socket
* virtio_net - which represents a vsocket connection
There can be many connections
on the same socket. vsocket
provides an API to enable/disable
particular virtio features on
the fly, but it's the virtio_net
that uses these features.
virtio_net used to rely on
vsocket->features during
feature negotiation, breaking
the layer encapsulation (and
yet causing a deadlock - two
locks were being locked in a
separate order). Now each
virtio_net device has it's own
copy of vsocket features, created
at the time of virtio_net creation.
vsocket->features have to be
still present, as features can be
enabled/disabled while no
virtio_net device has been
created yet.
Fixes#214
Change-Id: Ic4b2dd8cae6050813fc9a420b2ed30bc5ae60393
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/386294
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Pawel Wodkowski <pawelx.wodkowski@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
The vhost connection can be closed
concurrently from 2 places:
* the connection thread itself
* rte_vhost_driver_unregister
The connection thread will terminate
the connection if any recv error
occured. The unregister function
will terminate the connection
together with the thread.
However, there is no sychronization
between those two. The connection
thread runs in the background
without any mutex.
The rte_vhost_driver_unregister
now signals the connection thread
to terminate itself and waits
until it's killed.
Change-Id: I012e97ebb8a79edcb2c17c28b2fc7e8041bf92b3
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/383085
Reviewed-by: Pawel Wodkowski <pawelx.wodkowski@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
The rte_malloc_socket() call just above that allocates vq is only
allocating sizeof(*vq), but the memcpy() would have tried to copy
sizeof(*vq) * 2.
This code is under #ifdef RTE_LIBRTE_VHOST_NUMA, so it was not normally
enabled with DPDK 17.05, but it breaks when DPDK 17.08 turns on libnuma
support by default.
Change-Id: I75c0c8666a9147346038d313fb419350988d8187
Signed-off-by: Daniel Verkamp <daniel.verkamp@intel.com>
Reviewed-on: https://review.gerrithub.io/377596
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: John Kariuki <John.K.Kariuki@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Some rte_vhost files use #ifdef RTE_LIBRTE_VHOST_NUMA, but they don't
explicitly include rte_config.h, which defines this macro. Instruct
the compiler to pre-include rte_config.h in the same way DPDK's build
system does.
Change-Id: Iddde76b8c3d0956ccd5f481956cede650d858586
Signed-off-by: Daniel Verkamp <daniel.verkamp@intel.com>
Reviewed-on: https://review.gerrithub.io/377595
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: John Kariuki <John.K.Kariuki@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Added new callbacks to notify about socket connection status.
As destroy_device is used for virtqueue processing *pause* as well as
connection close, the user has no distinction between those.
Consider the following scenario:
rte_vhost: received SET_VRING_BASE message,
calling destroy_device() as usual
user: end-user asks to remove the device (together with socket file),
OK, device is not *in use* - that's NOT the behavior we want
calling rte_vhost_driver_unregister() etc.
Instead of changing new_device/destroy_device callbacks and breaking
the ABI, a set of new functions new_connection/destroy_connection
has been added.
Change-Id: I50a8ca4035045892d6c658da7df58c0c97025ec3
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/372074
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Pawel Wodkowski <pawelx.wodkowski@intel.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
2 locks are executed in 2 places in opposite orders.
Consider the following scenario, threads A and B:
(A)
* fdset_event_dispatch() start
* pfdentry->busy = 1; (lock #1)
* vhost_user_read_cb() start
* vhost_destroy_device() start
(B)
* rte_vhost_driver_unregister() start
* pthread_mutex_lock(&vsocket->conn_mutex); (lock #2)
* fdset_del()
* endless loop, waiting for pfdentry->busy == 0 (lock #1)
(A)
* vhost_destroy_device() end
* pthread_mutex_lock(&vsocket->conn_mutex); (lock #2)
(mutex already locked - deadlock at this point)
Thread B has locked vsocket->conn_mutex and is in while(1)
loop waiting for given fd to change it's busy flag to 0.
Thread A would have to finish vhost_user_read_cb() in order
to set busy flag back to 0, but that can't happen due to
the vsocket->conn_mutex lock.
This patch defers the fdset_del(), so that it's called outside of
vsocket->conn_mutex.
Change-Id: Ifb5d4699bdafe96a573444c11ad4eae3adc359f5
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/375910
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Pawel Wodkowski <pawelx.wodkowski@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This reverts commit 697389816422e61f9295e60d48be0bd39f0dec14.
This solution was incomplete, see the next patch which properly
fixes the deadlock issue.
Change-Id: Ib3cc609814276f1c48b05280379b8c2849ad831f
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/375909
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Pawel Wodkowski <pawelx.wodkowski@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Since vhost_user_set_features failure is not handled in any way, a
single error log has been added to at least to let the user know that
something has gone wrong.
Change-Id: Ifcf27320af75ba74347b742643b23e43b7c01149
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Reviewed-on: https://review.gerrithub.io/365807
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
For each virt queue's kickfd and callfd, there are 2 invalid
status: VIRTIO_UNINITIALIZED_EVENTFD and VIRTIO_INVALID_EVENTFD.
Don't set the virt queue to ready status until got the valid
descriptor.
This is safe for polling mode drivers in Guest OS, the backend
vhost process will not post notification to interrupt vector for
PMD mode in Guest, but the interrupt vector still valid.
Change-Id: Icdf1e67f3c4e8da221843eb1383469ca1fba485c
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Reviewed-on: https://review.gerrithub.io/365327
Reviewed-by: Pawel Wodkowski <pawelx.wodkowski@intel.com>
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
After changes done in commit f325e71c, closing old FD is delayed until
VHOST_USER_SET_VRING_ADDR. If VM is closed before this call, original FDs
remain during vhost_backend_cleanup. This resolves issue #162.
This patch closes second set of FDs during vhost backend cleanup.
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Change-Id: Ieb9d123c987009ac451b6214bb74d2720d852781
Reviewed-on: https://review.gerrithub.io/361787
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Piotr Pelpliński <piotr.pelplinski@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Pawel Wodkowski <pawelx.wodkowski@intel.com>
This fixes spontaneous vhost hangs on SIGINT shutdown.
Apperently during vhost_destroy_device(conn->vid) from
line #284 another QEMU message might arrive, causing
vsocket->conn_mutex deadlock. (line #286)
Change-Id: I4f1c31a52facffd1eb1e1192591095f00da55031
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
LOG_DEBUG is a symbol defined by POSIX, so if sys/log.h
is included the symbols conflict.
We'll need to push this patch to upstream DPDK too.
Change-Id: Ib263731864aca4791226ea6e3abb5ddfe42e97d8
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Since we keep a copy of DPDK vhost library, the header file don't
have dependency on DPDK vhost library.
Change-Id: I14d48e10227633547231e4f429e7375ffa76128d
Signed-off-by: Changpeng Liu <changpeng.liu@intel.com>
Fix up all existing spacing errors in comments and add an automated
check for patterns like /*comment*/.
Change-Id: I28f61c93612dc0f8aed66bd509da78e91ea9737e
Signed-off-by: Daniel Verkamp <daniel.verkamp@intel.com>
First step is do not destroy an existing device in
vhost_user_set_mem_table(). This is because we may
still be processing I/O via INT13 while QEMU is setting
up the mem tables for OS boot.
The primary part of this patch though is to defer
using the new mem table until after we receive the
first SET_VRING_ADDR message. SET_VRING_ADDR will be
sent by QEMU when guest OS virtio-scsi driver starts
initialization. At this point it is safe to invalidate
the old mem tables because there will be no more
INT13 I/O at this point.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I45fb5910f45e7fd2cf4a325341ad105a57d8ea40
This patch adds a library, application and test scripts for extending
SPDK to present virtio-scsi controllers to QEMU-based VMs and
process I/O submitted to devices attached to those controllers.
This functionality is dependent on QEMU patches to enable
vhost-scsi in userspace - those patches are currently working their
way through the QEMU mailing list, but temporary patches to enable
this functionality in QEMU will be made available shortly through the
SPDK github repository.
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Signed-off-by: Krzysztof Jakimiak <krzysztof.jakimiak@intel.com>
Signed-off-by: Michal Kosciowski <michal.kosciowski@intel.com>
Signed-off-by: Karol Latecki <karolx.latecki@intel.com>
Signed-off-by: Piotr Pelplinski <piotr.pelplinski@intel.com>
Signed-off-by: Daniel Verkamp <daniel.verkamp@intel.com>
Signed-off-by: Pawel Wodkowski <pawelx.wodkowski@intel.com>
Signed-off-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
Signed-off-by: Krzysztof Jakimiak <krzysztof.jakimiak@intel.com>
Change-Id: I138e4021f0ac4b1cd9a6e4041783cdf06e6f0efb
