The SCSI library already provides a callback when the task is released
(free_fn), so the user can update their own task counter.
Change-Id: I7fb13f6fff66dbba2315fd03fb06e49f793be123
Signed-off-by: Daniel Verkamp <daniel.verkamp@intel.com>
The task free callback function is required, so make its assignment part
of task construction.
Change-Id: I2f5fdf73b064653ee85b4e7961cb1653a0a4107d
Signed-off-by: Daniel Verkamp <daniel.verkamp@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>
If VRING_AVAIL_F_NO_INTERRUPT flag for queue is negotiated, guest can
set this flag in avail ring to suppress interrupts from host.
Change-Id: I2043067de33da009a974be5915ae81b1343fc424
Signed-off-by: Pawel Wodkowski <pawelx.wodkowski@intel.com>
Unit tests will be added as part of a separate patch updating all UT for
nvme.c. Global used for timeout value so it can be easily overwritten
by the upcoming unit tests for this function.
Change-Id: I7fc15aab91601ac57c94cae266b212c0998d2495
Signed-off-by: paul luse <paul.e.luse@intel.com>
This was implemented as two functions, but it
is much simpler as one. Also, the public function
was way at the bottom of the file instead of near
spdk_bdev_put_io_buf.
Change-Id: I3a90688910b0542cc77b6333bab15132cf514eeb
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
This was implemented as 3 separate functions but
it is simpler as 1.
Also, this wasn't previously freeing the buffer pools.
Change-Id: Ic1b2b3a0596e745a223099cb2a79bea6ef5c69cc
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
This was broken into three functions, but it is
a lot simpler as one.
Change-Id: If58ad50fe7d4f65c598b62f24e9e1ce7a64fdd8e
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
This is better organizationally, but also will serve as
an io_device in the future.
Change-Id: I6d65cf39df59e874d13f5fccc5a489720e86c48f
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Return types should be on a separate line for definitions.
Change-Id: Iaa38dd00042359fc6640fc67053bd69ebbb7af03
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Make the buffer allocation work for all types of
commands, not just read.
Change-Id: I72d8f67a724566630e7c4a74759fcb08449f7de4
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Eliminate rte_memcpy dependency by replacing it with
regular memcpy. This may impact performance, but the only
use of rte_memcpy was in the malloc bdev which is for
testing only.
Change-Id: I3e8592cb08262272518ec3d29ea165b4e8f48a5c
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Blockdevs already indicate support for unmap via
spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_UNMAP).
Change-Id: I634f27a281fd900bb3a6da2e4ff8a74e43579578
Signed-off-by: Daniel Verkamp <daniel.verkamp@intel.com>
We plan to use these buffers for more than just reads.
Change-Id: I8fa6cb432a6cfe4406fbf240cd3aa2ae4ab5f3d5
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
The user can get there via the bdev, so this didn't
have a purpose.
Change-Id: I7f85bb71d5ee238d37ba3624d0ac68a161c95e49
Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Older kernel headers don't have the definition of this macro, so define
it if necessary.
This is the same workaround as used in rte_vhost/vhost.h.
Change-Id: I01e0661db05de517adf8e24a47c63d32853cd385
Signed-off-by: Daniel Verkamp <daniel.verkamp@intel.com>
vhost_net.c file is not needed and fail scan build so remove it.
Change-Id: I5817201373f7253cc8bc1a9bdc5884197e166a14
Signed-off-by: Pawel Wodkowski <pawelx.wodkowski@intel.com>
guest_pages is being allocated in vhost_setup_mem_table(), reallocated
in add_one_guest_page(), but never freed. This patch fixes a memory
leak.
Change-Id: Ie381c43bafea5cdea2ac9f057c0282044a340dce
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
This prevents from destroying & recreating user device in "incomplete"
vring state. virtio_is_ready() was returning true for devices with
vrings which did not have valid callfd (their VHOST_USER_SET_VRING_CALL
hasn't arrived yet)
Change-Id: Idc4b41efd544ff5c6b093a5a48798b41c55bbe06
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
vhost-net devices might keep track of last descriptors indices by
themselves, and assuming they initially start at 0, but that is not the
case for vhost-scsi. Initial last descriptor indices are set via
VHOST_USER_SET_VRING_BASE message, and we cannot possibly predict what
will they be. Setting these to vqueue->used->idx is also not an option,
because there might be some yet unprocessed requests between these and
the actual last_idx. This patch adds API for getting/setting last
descriptor indices of vrings, so that they can be synchronized between
user-device and rte_vhost.
The last_idx flow could be as following:
* vhost start
* received SET_VRING_BASE msg, last_idx is set on rte_vhost side
* created user-device, last_idx pulled from rte_vhost
* requests are being processed by user-device, last_idx changes
* destroyed user-device, last_idx pushed to rte_vhost
* *at this point, vrings could be recreated and another SET_VRING_BASE
message could arrive, so last_idx would be set*
* recreated user-device, last_idx pulled from rte_vhost
Change-Id: I247ba4e461a2a2b524ccade364f5b7bf260f7538
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
There is an issue when QEMU sets new memory table just after guest OS
starts booting. Then, if guest OS tries to issue any I/O to device (e.g.
using BIOS INT13h - EDD) it will get stuck because previous addresses of
mmaped memory might change.
To fix this issue, 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 (e.g. virtio-scsi) driver starts initialization.
At this point it is safe to invalidate the old mem tables because there
will be no more outstanding IO at this point.
Change-Id: I24772be87a8b6c8781868b9b7773317761499748
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
For now DPDK assumes that callfd, kickfd and last_idx are being set just
once during vring initialization and device cannot be running while DPDK
receives SET_VRING_KICK, SET_VRING_CALL and SET_VRING_BASE messages.
However, that assumption is wrong. For Vhost SCSI messages might arrive
at any point of time, possibly multiple times, one after another.
QEMU issues SET_VRING_CALL once during device initialization, then again
during device start. The second message will close previous callfd,
which is still being used by the user-implementation of vhost device.
This results in writing to invalid (closed) callfd.
This patch destroys vhost device before setting callfd, kickfd and last
vring indices. It will be recreated right after (with updated vring
data).
Change-Id: I293bd91106f53f6c2f65d8b8a41f47ae7548cddc
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
This will be decoupled from the build to start. Next
patches will modify this code to prepare it for use with
SPDK vhost-scsi. The final patch will replace the existing
v17.02-based code with this version, and make the necessary
SPDK vhost changes to use it.
This enables to better track the differences between upstream
DPDK and our internal copy, while not breaking the build at
any point in the git history.
While here, expand the POSIX include file check to exclude
any directory starting with lib/vhost/rte_vhost (which would
include this new directory).
Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: Icf1202c1b7a898edff12aa226943a08b578cf962
Scan the source for POSIX includes outside of the
allowed locations in check_format.sh. This only
tests for POSIX headers - not Linux Standards Base.
Also, fix one bug that was caught by this addition.
Change-Id: Ib0ca93fe6ac552dc49d95b27b4803e40282027e8
Signed-off-by: Ben Walker <benjamin.walker@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>
FOREACH_DEVICE_ON_PCIBUS macro has been defined since rc2.
Change-Id: Iad61401520735dfde4e5715c32e74a54a2dff7da
Signed-off-by: Dariusz Stojaczyk <dariuszx.stojaczyk@intel.com>
Replace with it with check the returned req
via spdk_unlikely macro
Change-Id: I1202b3955af9a68496d8ced7cf66c20cf26f7fff
Signed-off-by: Ziye Yang <ziye.yang@intel.com>
The SCSI layer always passes task->iovs to spdk_bdev_readv(), so there
is no way for task->iovs != bdev_io->u.read.iovs to be true.
Change-Id: I4c0a2075c6e50e4304d62707a29bededa37b4e5c
Signed-off-by: Daniel Verkamp <daniel.verkamp@intel.com>
The SCSI task bdev I/O should never be pending when spdk_scsi_task_put()
is called, and just setting the status to failed is not correct (when
the bdev eventually completes the I/O, it will write into the now-freed
bdev_io, which may be reused by someone else).
Change-Id: Iaad6ce9ab41539652abc40147fed47c5012109dc
Signed-off-by: Daniel Verkamp <daniel.verkamp@intel.com>
