While we cannot check against the type of the full array, we can check
against the type of the fields.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Force shutdown on migration socket on cancel to cause the cancel
to complete even if the socket is blocked on a dead network.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
libvirt prefers opening the TCP connection itself, for two reasons.
First, connection failed errors can be detected easier, without having
to parse qemu's error output.
Second, libvirt might be asked to secure the transfer by tunnelling the
communication through an TLS layer.
Therefore, libvirt opens the TCP connection itself and passes an FD to qemu
using QMP and a POSIX-specific mechanism.
Hence, in order to make the reverse-path work in such cases, qemu needs to
distinguish if the transmitted FD is a socket (reverse-path available)
or not (reverse-path might not be available) and use the corresponding
abstraction.
Signed-off-by: Cristian Klein <cristian.klein@cs.umu.se>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Add QEMUFile interface to allow a socket to be 'shut down' - i.e. any
reads/writes will fail (and any blocking read/write will be woken).
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Modify some of tests/test-vmstate.c due to qemu_bufopen() change.
If you create a QEMUSizedBuffer yourself, you have to explicitly
free it.
Signed-off-by: Yang Hongyang <yanghy@cn.fujitsu.com>
Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
Cc: Juan Quintela <quintela@redhat.com>
Cc: Amit Shah <amit.shah@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Only free qsb that qemu_bufopen allocated, and also allow
qemu_bufopen accept qsb as input for write operation. It
will make the API more logical:
1.If you create the QEMUSizedBuffer yourself, you need to
free it by using qsb_free() but not depends on other API
like qemu_fclose.
2.allow qemu_bufopen() accept QEMUSizedBuffer as input for
write operation, otherwise, it will be a little strange
for this API won't accept the second parameter.
This brings API change, since there are only 3
users of this API currently, this change only impact the
first one which will be fixed in patch 2 of this patchset,
so I think it is safe to do this change.
1 70 tests/test-vmstate.c <<open_mem_file_read>>
return qemu_bufopen("r", qsb);
2 404 tests/test-vmstate.c <<test_save_noskip>>
QEMUFile *fsave = qemu_bufopen("w", NULL);
3 424 tests/test-vmstate.c <<test_save_skip>>
QEMUFile *fsave = qemu_bufopen("w", NULL);
Signed-off-by: Yang Hongyang <yanghy@cn.fujitsu.com>
Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
Cc: Juan Quintela <quintela@redhat.com>
Cc: Amit Shah <amit.shah@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Rebuild the cache_is_cached function by cache_get_by_addr. And
drops the asserts because the caller is also asserting the same
thing.
Signed-off-by: ChenLiang <chenliang88@huawei.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Avoid hot pages being replaced by others to remarkably decrease cache
misses
Sample results with the test program which quote from xbzrle.txt ran in
vm:(migrate bandwidth:1GE and xbzrle cache size 8MB)
the test program:
include <stdlib.h>
include <stdio.h>
int main()
{
char *buf = (char *) calloc(4096, 4096);
while (1) {
int i;
for (i = 0; i < 4096 * 4; i++) {
buf[i * 4096 / 4]++;
}
printf(".");
}
}
before this patch:
virsh qemu-monitor-command test_vm '{"execute": "query-migrate"}'
{"return":{"expected-downtime":1020,"xbzrle-cache":{"bytes":1108284,
"cache-size":8388608,"cache-miss-rate":0.987013,"pages":18297,"overflow":8,
"cache-miss":1228737},"status":"active","setup-time":10,"total-time":52398,
"ram":{"total":12466991104,"remaining":1695744,"mbps":935.559472,
"transferred":5780760580,"dirty-sync-counter":271,"duplicate":2878530,
"dirty-pages-rate":29130,"skipped":0,"normal-bytes":5748592640,
"normal":1403465}},"id":"libvirt-706"}
18k pages sent compressed in 52 seconds.
cache-miss-rate is 98.7%, totally miss.
after optimizing:
virsh qemu-monitor-command test_vm '{"execute": "query-migrate"}'
{"return":{"expected-downtime":2054,"xbzrle-cache":{"bytes":5066763,
"cache-size":8388608,"cache-miss-rate":0.485924,"pages":194823,"overflow":0,
"cache-miss":210653},"status":"active","setup-time":11,"total-time":18729,
"ram":{"total":12466991104,"remaining":3895296,"mbps":937.663549,
"transferred":1615042219,"dirty-sync-counter":98,"duplicate":2869840,
"dirty-pages-rate":58781,"skipped":0,"normal-bytes":1588404224,
"normal":387794}},"id":"libvirt-266"}
194k pages sent compressed in 18 seconds.
The value of cache-miss-rate decrease to 48.59%.
Signed-off-by: ChenLiang <chenliang88@huawei.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQEcBAABAgAGBQJUt3myAAoJEL7lnXSkw9fbKXkH/00rjLDjazEuCNlitZDEDbPi
gSNEHVk65xSmhhHURpvilKfA2IfXt67H9zDKp8FYFR0qfyLjgm5nMZQ7JWAcBk8e
UD851zL3anuJyjxQUB//vK5VmNyoG29IFyFXQAqXwuvZDZAflvAuXxsQYHh+tB/f
mIcQZNUmIAjLeg/JdY55s3b5ZLUcmd8Mmtzi1wZQPHeZQq5CxG3GGTKMB7wI4zJ/
76NnP163Gdvr8J1J00icjE8sl7f0o/tJzhR7lbK5QtzySMtRVj8lkDBYC77lOe6U
CaWTZgDCLtxtEVJ7xP/1hjg/TmSiPuOiAAvPFaUjdzb641Adshk4EaKF8ww2g9I=
=YF9X
-----END PGP SIGNATURE-----
Merge remote-tracking branch 'remotes/mjt/tags/pull-trivial-patches-2015-01-15' into staging
trivial patches for 2015-01-15
# gpg: Signature made Thu 15 Jan 2015 08:26:26 GMT using RSA key ID A4C3D7DB
# gpg: Good signature from "Michael Tokarev <mjt@tls.msk.ru>"
# gpg: aka "Michael Tokarev <mjt@corpit.ru>"
# gpg: aka "Michael Tokarev <mjt@debian.org>"
* remotes/mjt/tags/pull-trivial-patches-2015-01-15:
vl.c: fix some alignment issues
blizzard: do not depend on VGA internals
Makefile: Remove config.status and common.env during 'make distclean'
target-openrisc: bugfix for dec_sys to decode instructions correctly
Do not hang on full PTY
misc: Fix new typos in comments
target-arm: Fix typo in comment (seperately -> separately)
target-tricore: Fix new typos
migration/qemu-file.c: Don't shift left into sign bit
translate-all: Mark map_exec() with the 'unused' attribute
tests/hd-geo-test.c: Remove unused test_image variable
vt82c686: avoid out-of-bounds read
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The misalignment was caused by tabs which were used instead of spaces.
Signed-off-by: Marcel Apfelbaum <marcel@redhat.com>
Reviewed-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
There is nothing that is used by this ARM-specific device.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
config.status and tests/qemu-iotests/common.env are generated files
that should be deleted during 'make distclean'.
Signed-off-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Fixed the decoding of "system" instructions (starting with 0x2)
in dec_sys() in translate.c. In particular, the l.trap instruction
is now correctly decoded, which enables for singlestepping and
breakpoints to be set in GDB.
Signed-off-by: David R. Morrison <dmorrison@invlim.com>
Acked-by: Jia Liu <proljc@gmail.com>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
recieve -> receive
suprise -> surprise
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: John Snow <jsnow@redhat.com>
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Cc: Peter Maydell <peter.maydell@linaro.org>
Cc: Greg Bellows <greg.bellows@linaro.org>
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Add a cast in qemu_get_be32() to avoid shifting left into the sign
bit of a signed integer (which is undefined behaviour in C).
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Mark map_exec() with the 'unused' attribute to avoid '-Wunused-function'
warnings on clang 3.4 or later. This means we don't need to mark it
'inline', which is what we were previously using to suppress the warning
(a trick which only works with gcc, not clang).
Signed-off-by: SeokYeon Hwang <syeon.hwang@samsung.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
[PMM: tweaked comment message a little]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Remove unused variable test_image; this silences a clang warning.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
superio_ioport_readb can read the 256th element of the array.
Coverity reports an out-of-bounds write in superio_ioport_writeb,
but it does not show the corresponding out-of-bounds read
because it cannot prove that it can happen. Fix the root
cause of the problem (zhanghailang's patch instead fixes
the logic in superio_ioport_writeb).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Cc: qemu-stable@nongnu.org
the record/replay series, and a few SCSI and i386 patches as well.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)
iQEcBAABAgAGBQJUtjlCAAoJEL/70l94x66Dy5gH/0QIHoXVH/2wuA9apNK2/gBj
2U7g08QGKlc2wQGF4a48sQf523lSt5eirVxrwta0wmvFeznrdR84d4YGpolHM67A
Q9Y5J2i+v1H6cfQH6ylq61QQ7rEC3+isa65wblLeMSCAb2W1CcV7avSKu4BSPZw2
jGr3jd2Ve7pOsULpPhiNsmmltYSeZc7sQBYc9C7fQEoxOGsNnRoKOUKPnIk1mJTc
iYH480L1MnOL3enIz13K34lQofNRhJxJBLYKhYsBydQbOh0/Ls1eifOY4xEegXZ0
IUODy6c2pk+s/IUPARpBucKGKzDxdv0DLXDV60uGn5EsYT0CjCl9/sRs3bZvaQE=
=eT8u
-----END PGP SIGNATURE-----
Merge remote-tracking branch 'remotes/bonzini/tags/for-upstream' into staging
Mostly bugfixes and cleanups from qemu-devel. Yet another small patch from
the record/replay series, and a few SCSI and i386 patches as well.
# gpg: Signature made Wed 14 Jan 2015 09:39:14 GMT using RSA key ID 78C7AE83
# gpg: Good signature from "Paolo Bonzini <bonzini@gnu.org>"
# gpg: aka "Paolo Bonzini <pbonzini@redhat.com>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg: It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 46F5 9FBD 57D6 12E7 BFD4 E2F7 7E15 100C CD36 69B1
# Subkey fingerprint: F133 3857 4B66 2389 866C 7682 BFFB D25F 78C7 AE83
* remotes/bonzini/tags/for-upstream:
cpus: consistently use QEMU_CLOCK_VIRTUAL_RT for icount_warp_rt timer
qemu-timer: rename timer_init to timer_init_tl
scsi: fix cancellation when I/O was completed but DMA was not.
rules.mak: Fix module build
hw/scsi/lsi53c895a: add support for additional diag / debug registers
qemu-common.h: optimise muldiv64 if int128 is available
target-i386: do not memcpy in and out of xmm_regs
target-i386: fix movntsd on big-endian hosts
vl.c: fix regression when reading memory size from config file
vl: Don't silently change topology when all -smp options were set
vl: fix max_cpus check
vl: Avoid unnecessary 'if' nesting
9pfs: changed to use event_notifier instead of qemu_pipe
vl.c: fix regression when reading machine type from config file
char: restore stdio echo on resume from suspend.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Fix mismatch between timer_new_ms and timer_mod.
Signed-off-by: Pavel Dovgalyuk <Pavel.Dovgaluk@ispras.ru>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Commit d577646 (scsi: Introduce scsi_req_cancel_complete, 2014-09-25)
was supposed to have no semantic change, but it missed a case. When
r->aiocb has already been NULLed, but DMA was not complete and the
SCSI layer was waiting for scsi_req_continue, after the patch the
SCSI layer will not call the .cancel callback of SCSIBusInfo.
Fixes: d5776465ee
Cc: qemu-stable@nongnu.org
Reported-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Tested-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Module build is broken since commit c261d774fb ( rules.mak: Fix DSO
build by pulling in archive symbols). That commit added .mo placeholders
of DSO to -y variables, in order to pull stub symbols to executable. But
the placeholders are unintentionally expanded in -y, rather than
filtered out while linking.
Fix it by moving the -objs expanding to before inserting .mo
placeholders. Note that passing -cflags and -libs to member objects are
also moved to keep it happening before object expanding.
Reported-by: Bharata B Rao <bharata.rao@gmail.com>
Tested-by: Bharata B Rao <bharata.rao@gmail.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Some ancient Linux kernels read from registers 0x09 and 0x3c-3f during
boot. According to the spec these registers are for diag and debug
purposes only. If they are absend qemu aborts on read.
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Let compiler do the job to optimise the function.
Signed-off-by: Frediano Ziglio <frediano.ziglio@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
After the next patch, we will move the high parts of AVX and AVX512 registers
in the same array as the SSE registers. This will make it impossible to
memcpy an array of 128-bit values in and out of xmm_regs in one swoop.
Use a for loop instead.
Similarly, always use XMM_Q in translate.c. This avoids introducing bugs
such as the one fixed in the previous patch.
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This was accessing an XMM register's low half without going through XMM_Q.
Cc: qemu-stable@nongnu.org
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This is happening because an actual logic is performed on the memory
arguments inside the main's switch, disregarding the config file content.
Solved by extracting the logic on a separate function and calling it
after the switch.
Signed-off-by: Marcel Apfelbaum <marcel@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Increase maxmem before calling xc_domain_populate_physmap_exact to
avoid the risk of running out of guest memory. This way we can also
avoid complex memory calculations in libxl at domain construction
time.
This patch fixes an abort() when assigning more than 4 NICs to a VM.
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Don Slutz <dslutz@verizon.com>
According to NVMe specifications Bits 15:08 represent Minor Version number.
Signed-off-by: Anubhav Rakshit <anubhav.rakshit@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
We are moving block-migration.c to the separated migration directory,
keep this file watched by block maintainers is a good idea.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Remove first email address and let the one from which I am contributing.
Signed-off-by: Chrysostomos Nanakos <chris@include.gr>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
According to the specification, the low 16 bits should contain the number of
I/O submission queues, and the high 16 bits should contain the number of
I/O completion queues.
Signed-off-by: Alex Friedman <alex@e8storage.com>
Acked-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
SCSI devices have multiple kinds of queries they need to respond
to, as defined in the "cmd inquiry" section in MMC-6 and SPC-3.
Relevent sections:
MMC-6 revision 2g:
Non-VPD response data and pointer to SPC-3;
Section 6.8 "Inquiry Command"
SPC-3 revision 23:
Inquiry command and error handling:
Section 6.4 "INQUIRY command"
VPD data pages format:
Section 7.6 "Vital product data parameters"
We implement these Vital Product Data queries for SCSI, but not for
ATAPI through IDE. The result is that if you are looking for the WWN
identifier via tools such as sg3_utils, you will be unable to query
our CD/DVD rom device to obtain it.
This patch adds the minimum number of mandatory responses as defined
by SPC-3, which include the "supported pages" response (page 0x00)
and the "Device Identification" response (page 0x83). It also correctly
responds when it receives a request for an illegal page to improve
error output from related tools.
The Device ID page contains an arbitrary list of identification
strings of various formats; the ID strings included in this patch
were chosen to mimic those provided by the libata driver when
emulating this SCSI query (model, serial, and wwn when present.)
Example:
# libata emulated response
[root@localhost ~]# sg_inq --id /dev/sda
VPD INQUIRY: Device Identification page
Designation descriptor number 1, descriptor length: 24
designator_type: vendor specific [0x0], code_set: ASCII
associated with the addressed logical unit
vendor specific: QM00001
Designation descriptor number 2, descriptor length: 72
designator_type: T10 vendor identification, code_set: ASCII
associated with the addressed logical unit
vendor id: ATA
vendor specific: QEMU HARDDISK QM00001
# QEMU generated ATAPI response, with WWN
[root@localhost ~]# sg_inq --id /dev/sr0
VPD INQUIRY: Device Identification page
Designation descriptor number 1, descriptor length: 24
designator_type: vendor specific [0x0], code_set: ASCII
associated with the addressed logical unit
vendor specific: QM00005
Designation descriptor number 2, descriptor length: 72
designator_type: T10 vendor identification, code_set: ASCII
associated with the addressed logical unit
vendor id: ATA
vendor specific: QEMU DVD-ROM QM00005
Designation descriptor number 3, descriptor length: 12
designator_type: NAA, code_set: Binary
associated with the addressed logical unit
NAA 5, IEEE Company_id: 0xc50
Vendor Specific Identifier: 0x15ea71bb
[0x5000c50015ea71bb]
See also: hw/scsi/scsi-disk.c, scsi_disk_emulate_inquiry()
Signed-off-by: John Snow <jsnow@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Like BLOCK_OP_TYPE_BACKUP_SOURCE and BLOCK_OP_TYPE_BACKUP_TARGET,
block-commit involves two asymmetric devices.
This change is not user-visible (yet), because commit only works with
device names.
But once we enable backing reference in blockdev-add, or specifying
node-name in block-commit command, we don't want the user to start two
commit jobs on the same backing chain, which will corrupt things because
of the final bdrv_swap.
Before we have per category blockers, splitting this type is still
better.
[Resolved virtio-blk dataplane conflict by replacing
BLOCK_OP_TYPE_COMMIT with both BLOCK_OP_TYPE_COMMIT_{SOURCE, TARGET}.
They are safe since the block job runs in the same AioContext as the
dataplane IOThread.
--Stefan]
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
If bs->bl.max_write_zeroes is large and we end up in the unsupported
path we might allocate a lot of memory for the iovector and/or even
generate an oversized requests.
Fix this by limiting the request by the minimum of the reported
maximum transfer size or 16MB (32768 sectors).
Reported-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Peter Lieven <pl@kamp.de>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Message-id: 1420457389-16332-1-git-send-email-pl@kamp.de
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Placing coroutines on the global pool should be preferrable, because it
can help all threads. But if the global pool is full, we can still
try to save some allocations by stashing completed coroutines on the
local pool. This is quite cheap too, because it does not require
atomic operations, and provides a gain of 15% in the best case.
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 1417518350-6167-8-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This is not needed anymore. The new TLS-based algorithm is adaptive.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 1417518350-6167-7-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This patch removes the mutex by using fancy lock-free manipulation of
the pool. Lock-free stacks and queues are not hard, but they can suffer
from the ABA problem so they are better avoided unless you have some
deferred reclamation scheme like RCU. Otherwise you have to stick
with adding to a list, and emptying it completely. This is what this
patch does, by coupling a lock-free global list of available coroutines
with per-CPU lists that are actually used on coroutine creation.
Whenever the destruction pool is big enough, the next thread that runs
out of coroutines will steal the whole destruction pool. This is positive
in two ways:
1) the allocation does not have to do any atomic operation in the fast
path, it's entirely using thread-local storage. Once every POOL_BATCH_SIZE
allocations it will do a single atomic_xchg. Release does an atomic_cmpxchg
loop, that hopefully doesn't cause any starvation, and an atomic_inc.
A later patch will also remove atomic operations from the release path,
and try to avoid the atomic_xchg altogether---succeeding in doing so if
all devices either use ioeventfd or are not submitting requests actively.
2) in theory this should be completely adaptive. The number of coroutines
around should be a little more than POOL_BATCH_SIZE * number of allocating
threads; so this also empties qemu_coroutine_adjust_pool_size. (The previous
pool size was POOL_BATCH_SIZE * number of block backends, so it was a bit
more generous. But if you actually have many high-iodepth disks, it's better
to put them in different iothreads, which will also use separate thread
pools and aio=native file descriptors).
This speeds up perf/cost (in tests/test-coroutine) by a factor of ~1.33.
No matter if we end with some kind of coroutine bypass scheme or not,
it cannot hurt to optimize hot code.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 1417518350-6167-6-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
These operations are trivial to implement and do not have ABA problems.
They are enough to implement simple multiple-producer, single consumer
lock-free lists or, as in the next patch, the multiple consumers can
steal a whole batch of elements and process them at their leisure.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 1417518350-6167-5-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
unsigned long is not large enough to represent 1000000000 * duration there.
Just use floating point.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 1417518350-6167-4-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Destructors are the main additional feature of pthread TLS compared
to __thread. If we were using C++ (hint, hint!) we could have used
thread-local objects with a destructor. Since we are not, instead,
we add a simple Notifier-based API.
Note that the notifier must be per-thread as well. We can add a
global list as well later, perhaps.
The Win32 implementation has some complications because a) detached
threads used not to have a QemuThreadData; b) the main thread does
not go through win32_start_routine, so we have to use atexit too.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 1417518350-6167-3-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
ELF thread local storage is about 10% faster on tests/test-coroutine's
perf/cost test. The timing on my machine is 190ns per iteration with
pthread TLS, 170 with ELF TLS.
Based on a patch by Kevin Wolf and Peter Lieven, but redone to follow
the model of coroutine-win32.c (including the important "noinline"
attribute!).
Platforms without thread-local storage (OpenBSD probably?) will need
a new-enough GCC for this to compile, in order to use the same emutls
support that Windows already relies on.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 1417518350-6167-2-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
If I understand correctly, qemu-iotests never meant to be portable. We
only support Linux for all the shell cases, but didn't specify it for
python tests. Now add this and default all the python tests as Linux
only. If we cares enough later, we can override the parameter in
individual cases.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Other cases have this, and this test is not portable as well, as we want
to add "make check-block" to "make check", it shouldn't fail on Mac OS
X.
Reported-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Paolo Bonzini