If the xhci uses msix, it doesn't free the corresponding
memory, thus leading a memory leak. This patch avoid this.
Signed-off-by: Li Qiang <liqiang6-s@360.cn>
Message-id: 57d7d2e0.d4301c0a.d13e9.9a55@mx.google.com
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Right after main_loop ends, we release various things but keep iothread
alive. The latter is not prepared to the sudden change of resources.
Specifically, after bdrv_close_all(), virtio-scsi dataplane get a
surprise at the empty BlockBackend:
(gdb) bt
at /usr/src/debug/qemu-2.6.0/hw/scsi/virtio-scsi.c:543
at /usr/src/debug/qemu-2.6.0/hw/scsi/virtio-scsi.c:577
It is because the d->conf.blk->root is set to NULL, then
blk_get_aio_context() returns qemu_aio_context, whereas s->ctx is still
pointing to the iothread:
hw/scsi/virtio-scsi.c:543:
if (s->dataplane_started) {
assert(blk_get_aio_context(d->conf.blk) == s->ctx);
}
To fix this, let's stop iothreads before doing bdrv_close_all().
Cc: qemu-stable@nongnu.org
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 1473326931-9699-1-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
vq->avail.idx and vq->avail->ring[] are a 16bit values,
so read and write them with readw()/writew() instead of
readl()/writel().
To read/write a 16bit value with a 32bit accessor works fine
on little-endian CPU but not on big endian CPU.
[An equivalent patch for the writew() calls was also sent by
Zhang Shuai <zhangshuai13@huawei.com>.
--Stefan]
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Message-id: 1472330054-22607-1-git-send-email-lvivier@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
As per Stefan's suggestion, add Wen and I as co-maintainers
of replication.
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Changlong Xie <xiecl.fnst@cn.fujitsu.com>
Message-id: 1469602913-20979-13-git-send-email-xiecl.fnst@cn.fujitsu.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
[Rename get_error test cases to get_error_all to avoid tripping up
scripts that grep for "error:" in test output. It also reflects the
actual replication API function name better.
-Stefan]
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Changlong Xie <xiecl.fnst@cn.fujitsu.com>
Signed-off-by: Wang WeiWei <wangww.fnst@cn.fujitsu.com>
Message-id: 1469602913-20979-11-git-send-email-xiecl.fnst@cn.fujitsu.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This commit introduces six replication interfaces(for block, network etc).
Firstly we can use replication_(new/remove) to create/destroy replication
instances, then in migration we can use replication_(start/stop/do_checkpoint
/get_error)_all to handle all replication operations. More detail please
refer to replication.h
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Changlong Xie <xiecl.fnst@cn.fujitsu.com>
Signed-off-by: Wang WeiWei <wangww.fnst@cn.fujitsu.com>
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Message-id: 1469602913-20979-9-git-send-email-xiecl.fnst@cn.fujitsu.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
configure --(enable/disable)-replication to switch replication
support on/off, and it is on by default.
We later introduce replation support.
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Changlong Xie <xiecl.fnst@cn.fujitsu.com>
Signed-off-by: Wang WeiWei <wangww.fnst@cn.fujitsu.com>
Message-id: 1469602913-20979-8-git-send-email-xiecl.fnst@cn.fujitsu.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Auto complete mirror job in background to prevent from
blocking synchronously
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Changlong Xie <xiecl.fnst@cn.fujitsu.com>
Signed-off-by: Wang WeiWei <wangww.fnst@cn.fujitsu.com>
Message-id: 1469602913-20979-7-git-send-email-xiecl.fnst@cn.fujitsu.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Some programs that add a dependency on it will use
the block layer directly.
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Changlong Xie <xiecl.fnst@cn.fujitsu.com>
Signed-off-by: Wang WeiWei <wangww.fnst@cn.fujitsu.com>
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Message-id: 1469602913-20979-5-git-send-email-xiecl.fnst@cn.fujitsu.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Normal backup(sync='none') workflow:
step 1. NBD peformance I/O write from client to server
qcow2_co_writev
bdrv_co_writev
...
bdrv_aligned_pwritev
notifier_with_return_list_notify -> backup_do_cow
bdrv_driver_pwritev // write new contents
step 2. drive-backup sync=none
backup_do_cow
{
wait_for_overlapping_requests
cow_request_begin
for(; start < end; start++) {
bdrv_co_readv_no_serialising //read old contents from Secondary disk
bdrv_co_writev // write old contents to hidden-disk
}
cow_request_end
}
step 3. Then roll back to "step 1" to write new contents to Secondary disk.
And for replication, we must make sure that we only read the old contents from
Secondary disk in order to keep contents consistent.
1) Replication workflow of Secondary
virtio-blk
^
-------> 1 NBD |
|| server 3 replication
|| ^ ^
|| | backing backing |
|| Secondary disk 6<-------- hidden-disk 5 <-------- active-disk 4
|| | ^
|| '-------------------------'
|| drive-backup sync=none 2
Hence, we need these interfaces to implement coarse-grained serialization between
COW of Secondary disk and the read operation of replication.
Example codes about how to use them:
*#include "block/block_backup.h"
static coroutine_fn int xxx_co_readv()
{
CowRequest req;
BlockJob *job = secondary_disk->bs->job;
if (job) {
backup_wait_for_overlapping_requests(job, start, end);
backup_cow_request_begin(&req, job, start, end);
ret = bdrv_co_readv();
backup_cow_request_end(&req);
goto out;
}
ret = bdrv_co_readv();
out:
return ret;
}
Signed-off-by: Changlong Xie <xiecl.fnst@cn.fujitsu.com>
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Wang WeiWei <wangww.fnst@cn.fujitsu.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1469602913-20979-4-git-send-email-xiecl.fnst@cn.fujitsu.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The old one is confusing with @virtio_device_info in virtio.c,
so make it more appropriate.
Signed-off-by: Changlong Xie <xiecl.fnst@cn.fujitsu.com>
Message-id: 1470214147-32560-1-git-send-email-xiecl.fnst@cn.fujitsu.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
In order to reduce completion latency it makes sense to harvest completed
requests ASAP. Very fast backend device can complete requests just after
submission, so it is worth trying to check ring buffer in order to peek
completed requests directly after io_submit() has been called.
Indeed, this patch reduces the completions latencies and increases the
overall throughput, e.g. the following is the percentiles of number of
completed requests at once:
1th 10th 20th 30th 40th 50th 60th 70th 80th 90th 99.99th
Before 2 4 42 112 128 128 128 128 128 128 128
After 1 1 4 14 33 45 47 48 50 51 108
That means, that before the current patch is applied the ring buffer is
observed as full (128 requests were consumed at once) in 60% of calls.
After patch is applied the distribution of number of completed requests
is "smoother" and the queue (requests in-flight) is almost never full.
The fio read results are the following (write results are almost the
same and are not showed here):
Before
------
job: (groupid=0, jobs=8): err= 0: pid=2227: Tue Jul 19 11:29:50 2016
Description : [Emulation of Storage Server Access Pattern]
read : io=54681MB, bw=1822.7MB/s, iops=179779, runt= 30001msec
slat (usec): min=172, max=16883, avg=338.35, stdev=109.66
clat (usec): min=1, max=21977, avg=1051.45, stdev=299.29
lat (usec): min=317, max=22521, avg=1389.83, stdev=300.73
clat percentiles (usec):
| 1.00th=[ 346], 5.00th=[ 596], 10.00th=[ 708], 20.00th=[ 852],
| 30.00th=[ 932], 40.00th=[ 996], 50.00th=[ 1048], 60.00th=[ 1112],
| 70.00th=[ 1176], 80.00th=[ 1256], 90.00th=[ 1384], 95.00th=[ 1496],
| 99.00th=[ 1800], 99.50th=[ 1928], 99.90th=[ 2320], 99.95th=[ 2672],
| 99.99th=[ 4704]
bw (KB /s): min=205229, max=553181, per=12.50%, avg=233278.26, stdev=18383.51
After
------
job: (groupid=0, jobs=8): err= 0: pid=2220: Tue Jul 19 11:31:51 2016
Description : [Emulation of Storage Server Access Pattern]
read : io=57637MB, bw=1921.2MB/s, iops=189529, runt= 30002msec
slat (usec): min=169, max=20636, avg=329.61, stdev=124.18
clat (usec): min=2, max=19592, avg=988.78, stdev=251.04
lat (usec): min=381, max=21067, avg=1318.42, stdev=243.58
clat percentiles (usec):
| 1.00th=[ 310], 5.00th=[ 580], 10.00th=[ 748], 20.00th=[ 876],
| 30.00th=[ 908], 40.00th=[ 948], 50.00th=[ 1012], 60.00th=[ 1064],
| 70.00th=[ 1080], 80.00th=[ 1128], 90.00th=[ 1224], 95.00th=[ 1288],
| 99.00th=[ 1496], 99.50th=[ 1608], 99.90th=[ 1960], 99.95th=[ 2256],
| 99.99th=[ 5408]
bw (KB /s): min=212149, max=390160, per=12.49%, avg=245746.04, stdev=11606.75
Throughput increased from 1822MB/s to 1921MB/s, average completion latencies
decreased from 1051us to 988us.
Signed-off-by: Roman Pen <roman.penyaev@profitbricks.com>
Message-id: 1468931263-32667-4-git-send-email-roman.penyaev@profitbricks.com
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: qemu-devel@nongnu.org
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Prepare processing events function to be called from ioq_submit(),
thus split function on two parts: the first harvests completed IO
requests, the second submits pending requests.
Signed-off-by: Roman Pen <roman.penyaev@profitbricks.com>
Message-id: 1468931263-32667-3-git-send-email-roman.penyaev@profitbricks.com
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: qemu-devel@nongnu.org
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
AIO context in userspace is represented as a simple ring buffer, which
can be consumed directly without entering the kernel, which obviously
can bring some performance gain. QEMU does not use timeout value for
waiting for events completions, so we can consume all events from
userspace.
Signed-off-by: Roman Pen <roman.penyaev@profitbricks.com>
Message-id: 1468931263-32667-2-git-send-email-roman.penyaev@profitbricks.com
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: qemu-devel@nongnu.org
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Section "7.1.4 Use of library functions" in the C99 standard says:
If an argument to a function has an invalid value (such as [...]
a null pointer [...]) [...] the behavior is undefined.
Additionally the "searching and sorting" functions are specified as
requiring valid pointer values as described in 7.1.4.
This patch fixes the following sanitizer errors:
block/qcow2.c:1807:41: runtime error: null pointer passed as argument 2, which is declared to never be null
block/qcow2-cluster.c:86:26: runtime error: null pointer passed as argument 2, which is declared to never be null
Reported-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1473758138-19260-1-git-send-email-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
We can't hotplug display adapters in qemu, tag virtio-gpu-pci
accordingly (virtio-vga already has this).
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Message-id: 1473319037-27645-1-git-send-email-kraxel@redhat.com
When processing svga command DEFINE_CURSOR in vmsvga_fifo_run,
the computed BITMAP and PIXMAP size are checked against the
'cursor.mask[]' and 'cursor.image[]' array sizes in bytes.
Correct these checks to avoid OOB memory access.
Reported-by: Qinghao Tang <luodalongde@gmail.com>
Reported-by: Li Qiang <liqiang6-s@360.cn>
Signed-off-by: Prasad J Pandit <pjp@fedoraproject.org>
Message-id: 1473338754-15430-1-git-send-email-ppandit@redhat.com
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
The companion descriptor is present on all usb3 devices, not only
those with streams support. We need to check attributes to see
whenever the device uses streams or not.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Message-id: 1473406890-30164-1-git-send-email-kraxel@redhat.com
Previously the code would incorrectly report the remainder as 8 bytes. A
remainder of 0 bytes should be reported when the SETUP packet is
successfully transferred. Found using FreeBSD's XHCI driver.
Signed-off-by: Hans Petter Selasky <hps@selasky.org>
[ kraxel: codestyle fixup ]
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
since f1d3e58, the code had changed the default value to 'off', so this patch
make document and code are consistent.
Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Message-id: 1470024419-10886-1-git-send-email-lizhijian@cn.fujitsu.com
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Coverity identifies that at the top of the while(1) loop
in curses_refresh() the variable nextchr is always ERR,
and so the else case of the first if() is dead code.
Remove this dead code, and narrow the scope of the
nextchr variable to the place where it's used.
(This confused logic has been present since the curses
code was added to QEMU in 2008.)
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1470925407-23850-3-git-send-email-peter.maydell@linaro.org
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Coverity spots that there is no bounds check before we
access the curses2qemu[] array. Add one, bringing this
code path into line with the one that looks up entries
in curses2keysym[].
In theory getch() shouldn't return out of range keycodes,
but it's better not to assume this.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1470925407-23850-2-git-send-email-peter.maydell@linaro.org
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Added documentation describing relation between GlusterServer and
SocketAddress qapi schemas.
Thanks to Markus Armbruster <armbru@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Prasanna Kumar Kalever <prasanna.kalever@redhat.com>
Message-id: 1471715924-3642-1-git-send-email-prasanna.kalever@redhat.com
Signed-off-by: Jeff Cody <jcody@redhat.com>
currently all the libgfapi logs defaults to '/dev/stderr' as it was hardcoded
in a call to glfs logging api. When the debug level is chosen to DEBUG/TRACE,
gfapi logs will be huge and fill/overflow the console view.
This patch provides a commandline option to mention log file path which helps
in logging to the specified file and also help in persisting the gfapi logs.
Usage:
-----
*URI Style:
---------
-drive file=gluster://hostname/volname/image.qcow2,file.debug=9,\
file.logfile=/var/log/qemu/qemu-gfapi.log
*JSON Style:
----------
'json:{
"driver":"qcow2",
"file":{
"driver":"gluster",
"volume":"volname",
"path":"image.qcow2",
"debug":"9",
"logfile":"/var/log/qemu/qemu-gfapi.log",
"server":[
{
"type":"tcp",
"host":"1.2.3.4",
"port":24007
},
{
"type":"unix",
"socket":"/var/run/glusterd.socket"
}
]
}
}'
Reviewed-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Prasanna Kumar Kalever <prasanna.kalever@redhat.com>
Signed-off-by: Jeff Cody <jcody@redhat.com>
Several error messages print out the raw enum value, which
is less than helpful to users, as these values are not
documented, nor stable across QEMU releases. Switch to use
the enum string instead.
The nettle impl also had two typos where it mistakenly
said "algorithm" instead of "mode", and actually reported
the algorithm value too.
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
gnutls commit 846753877d renamed LIBGNUTLS_VERSION_NUMBER to GNUTLS_VERSION_NUMBER.
If using gnutls before that verion, we'll get the below warning:
crypto/tlscredsx509.c:618:5: warning: "GNUTLS_VERSION_NUMBER" is not defined
Because gnutls 3.x still defines LIBGNUTLS_VERSION_NUMBER for back compat, Let's
use LIBGNUTLS_VERSION_NUMBER instead of GNUTLS_VERSION_NUMBER to fix building
complaint.
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The XTS cipher mode needs to be used with a cipher which has
a block size of 16 bytes. If a mis-matching block size is used,
the code will either corrupt memory beyond the IV array, or
not fully encrypt/decrypt the IV.
This fixes a memory corruption crash when attempting to use
cast5-128 with xts, since the former has an 8 byte block size.
A test case is added to ensure the cipher creation fails with
such an invalid combination.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Implement the new virtio sockets device for host<->guest communication
using the Sockets API. Most of the work is done in a vhost kernel
driver so that virtio-vsock can hook into the AF_VSOCK address family.
The QEMU vhost-vsock device handles configuration and live migration
while the rx/tx happens in the vhost_vsock.ko Linux kernel driver.
The vsock device must be given a CID (host-wide unique address):
# qemu -device vhost-vsock-pci,id=vhost-vsock-pci0,guest-cid=3 ...
For more information see:
http://qemu-project.org/Features/VirtioVsock
[Endianness fixes and virtio-ccw support by Claudio Imbrenda
<imbrenda@linux.vnet.ibm.com>]
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
[mst: rebase to master]
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Use kvm acceleration if available.
Disable kernel-irqchip and use qemu64 cpu
for both kvm and tcg cases.
Using kvm acceleration saves about a second
and disabling kernel-irqchip has no visible
performance impact.
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Marcel Apfelbaum <marcel@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
!legacy && !modern is shorter than !(legacy || modern).
I also perfer this (less ()s) as a matter of taste.
Cc: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
We used to set vring call fd unconditionally even if guest driver does
not use MSIX for this vritqueue at all. This will cause lots of
unnecessary userspace access and other checks for drivers does not use
interrupt at all (e.g virtio-net pmd). So check and clean vring call
fd if guest does not use any vector for this virtqueue at
all.
Perf diffs (on rx) shows lots of cpus wasted on vhost_signal() were saved:
#
28.12% -27.82% [vhost] [k] vhost_signal
14.44% -1.69% [kernel.vmlinux] [k] copy_user_generic_string
7.05% +1.53% [kernel.vmlinux] [k] __free_page_frag
6.51% +5.53% [vhost] [k] vhost_get_vq_desc
...
Pktgen tests shows 15.8% improvement on rx pps and 6.5% on tx pps.
Before: RX 2.08Mpps TX 1.35Mpps
After: RX 2.41Mpps TX 1.44Mpps
Signed-off-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Without presuming if we got there because of a user mistake or some
more subtle bug in the tooling, it really does not make sense to
implement a non-functional device.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Marcel Apfelbaum <marcel@redhat.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The statistics virtqueue is not migrated properly because virtio-balloon
does not include s->stats_vq_elem in the migration stream.
After migration the statistics virtqueue hangs because the host never
completes the last element (s->stats_vq_elem is NULL on the destination
QEMU). Therefore the guest never submits new elements and the virtqueue
is hung.
Instead of changing the migration stream format in an incompatible way,
detect the migration case and rewind the virtqueue so the last element
can be completed.
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Roman Kagan <rkagan@virtuozzo.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Suggested-by: Roman Kagan <rkagan@virtuozzo.com>
Signed-off-by: Ladi Prosek <lprosek@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
virtqueue_discard() requires a VirtQueueElement but virtio-balloon does
not migrate its in-use element. Introduce a new function that is
similar to virtqueue_discard() but doesn't require a VirtQueueElement.
This will allow virtio-balloon to access element again after migration
with the usual proviso that the guest may have modified the vring since
last time.
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Roman Kagan <rkagan@virtuozzo.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ladi Prosek <lprosek@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The one pending element is being freed but not discarded on device
reset, which causes svq->inuse to creep up, eventually hitting the
"Virtqueue size exceeded" error.
Properly discarding the element on device reset makes sure that its
buffers are unmapped and the inuse counter stays balanced.
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Roman Kagan <rkagan@virtuozzo.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ladi Prosek <lprosek@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
vq->inuse must be zeroed upon device reset like most other virtqueue
fields.
In theory, virtio_reset() just needs assert(vq->inuse == 0) since
devices must clean up in-flight requests during reset (requests cannot
not be leaked!).
In practice, it is difficult to achieve vq->inuse == 0 across reset
because balloon, blk, 9p, etc implement various different strategies for
cleaning up requests. Most devices call g_free(elem) directly without
telling virtio.c that the VirtQueueElement is cleaned up. Therefore
vq->inuse is not decremented during reset.
This patch zeroes vq->inuse and trusts that devices are not leaking
VirtQueueElements across reset.
I will send a follow-up series that refactors request life-cycle across
all devices and converts vq->inuse = 0 into assert(vq->inuse == 0) but
this more invasive approach is not appropriate for stable trees.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Cc: qemu-stable <qemu-stable@nongnu.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Ladi Prosek <lprosek@redhat.com>
Currently each VQ Notification Virtio Capability is allocated
on a different page. The idea is to enable split drivers within
guests, however there are no known plans to do that.
The allocation will result in a 8MB BAR, more than various
guest firmwares pre-allocates for PCI Bridges hotplug process.
Reserve 4 bytes per VQ by default and add a new parameter
"page-per-vq" to be used with split drivers.
Signed-off-by: Marcel Apfelbaum <marcel@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Some software algorithms are based on the hardware's cache info, for example,
for x86 linux kernel, when cpu1 want to wakeup a task on cpu2, cpu1 will trigger
a resched IPI and told cpu2 to do the wakeup if they don't share low level
cache. Oppositely, cpu1 will access cpu2's runqueue directly if they share llc.
The relevant linux-kernel code as bellow:
static void ttwu_queue(struct task_struct *p, int cpu)
{
struct rq *rq = cpu_rq(cpu);
......
if (... && !cpus_share_cache(smp_processor_id(), cpu)) {
......
ttwu_queue_remote(p, cpu); /* will trigger RES IPI */
return;
}
......
ttwu_do_activate(rq, p, 0); /* access target's rq directly */
......
}
In real hardware, the cpus on the same socket share L3 cache, so one won't
trigger a resched IPIs when wakeup a task on others. But QEMU doesn't present a
virtual L3 cache info for VM, then the linux guest will trigger lots of RES IPIs
under some workloads even if the virtual cpus belongs to the same virtual socket.
For KVM, there will be lots of vmexit due to guest send IPIs.
The workload is a SAP HANA's testsuite, we run it one round(about 40 minuates)
and observe the (Suse11sp3)Guest's amounts of RES IPIs which triggering during
the period:
No-L3 With-L3(applied this patch)
cpu0: 363890 44582
cpu1: 373405 43109
cpu2: 340783 43797
cpu3: 333854 43409
cpu4: 327170 40038
cpu5: 325491 39922
cpu6: 319129 42391
cpu7: 306480 41035
cpu8: 161139 32188
cpu9: 164649 31024
cpu10: 149823 30398
cpu11: 149823 32455
cpu12: 164830 35143
cpu13: 172269 35805
cpu14: 179979 33898
cpu15: 194505 32754
avg: 268963.6 40129.8
The VM's topology is "1*socket 8*cores 2*threads".
After present virtual L3 cache info for VM, the amounts of RES IPIs in guest
reduce 85%.
For KVM, vcpus send IPIs will cause vmexit which is expensive, so it can cause
severe performance degradation. We had tested the overall system performance if
vcpus actually run on sparate physical socket. With L3 cache, the performance
improves 7.2%~33.1%(avg:15.7%).
Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
