This fixes a crash caused by picking the wrong memory region in
address_space_lookup_region seen with client code accessing a device
model that uses alias memory regions. The expensive part of
address_space_lookup_region anyway is phys_page_find; performance-wise
it is okay to repeat the subsequent subpage lookup.
Signed-off-by: BALATON Zoltan <balaton@eik.bme.hu>
Message-Id: <20171114225941.072707456B5@zero.eik.bme.hu>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
GCC 4.9 and newer stopped warning for missing braces around the
"universal" C zero initializer {0}. One such initializer sneaked
into scsi/qemu-pr-helper.c and is breaking the build with such
older GCC versions.
Detect the lack of support for the idiom, and disable the warning
in that case.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Legacy PCI device assignment has been removed from Linux in 4.12,
and had been deprecated 2 years ago there. We can remove it from
QEMU as well.
The ROM loading code was shared with Xen PCI passthrough, so move
it to hw/xen.
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
libmultipath has recently changed its API. The new API supports multi-threaded
clients better. Unfortunately there is no backwards-compatibility, so we just
switch to the new one. Running QEMU compiled with the new library on the old
library will likely crash, while doing the opposite will cause QEMU not to
start at all (because udev, get_multipath_config and put_multipath_config
are undefined).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Aligned 8-byte memory writes by a 64-bit target on a 64-bit host should
always turn into atomic 8-byte writes on the host, however a write
write watchpoint would end up tearing the 8-byte write into two 4-byte
writes in access_with_adjusted_size().
Reported-by: Andrew Baumann <Andrew.Baumann@microsoft.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Besides being more correct, arbitrarily long instruction allow the
generation of a translation block that spans three pages. This
confuses the generator and even allows ring 3 code to poison the
translation block cache and inject code into other processes that are
in guest ring 3.
This is an improved (and more invasive) fix for commit 30663fd ("tcg/i386:
Check the size of instruction being translated", 2017-03-24). In addition
to being more precise (and generating the right exception, which is #GP
rather than #UD), it distinguishes better between page faults and too long
instructions, as shown by this test case:
#include <sys/mman.h>
#include <string.h>
#include <stdio.h>
int main()
{
char *x = mmap(NULL, 8192, PROT_READ|PROT_WRITE|PROT_EXEC,
MAP_PRIVATE|MAP_ANON, -1, 0);
memset(x, 0x66, 4096);
x[4096] = 0x90;
x[4097] = 0xc3;
char *i = x + 4096 - 15;
mprotect(x + 4096, 4096, PROT_READ|PROT_WRITE);
((void(*)(void)) i) ();
}
... which produces a #GP without the mprotect, and a #PF with it.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
These take care of advancing s->pc, and will provide a unified point
where to check for the 15-byte instruction length limit.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This should be done by all target and, since commit 53f6672bcf
("gen-icount: use tcg_ctx.tcg_env instead of cpu_env", 2017-06-30),
is causing the NIOS2 target to hang.
This is because the test for "should I exit to the main loop"
was being done with the correct offset to the icount decrementer,
but using TCG temporary 0 (the frame pointer) rather than the
env pointer.
Cc: qemu-stable@nongnu.org
Cc: Marek Vasut <marex@denx.de>
Reported-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
It is used by all PPC targets; we can give the directory its own
Makefile.objs file, and include it directly from target/ppc.
target/s390 can do the same when it starts using it.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
stgit produces patch files that lack the ".patch" extensions. Others
might be using ".diff" too. But since we are already limiting source files
to only a handful of extensions, we can reuse that in the mode selection
code.
While at it, do not match "../foo" as a branch name.
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
"check" is full of qemu-iotests--specific details. Separating it
from "common" does not make much sense anymore.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The variable is almost unused, and one of the two uses is actually
uninitialized.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The variable is used in "common" but defined only after the file
is sourced.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Split "check" parts from tests part.
For the directory setup, the actual computation of directories goes
in "check", while the sanity checks go in the tests.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
These are never used by "check", with one exception that does not need
$QEMU_OPTIONS. Keep them in common.rc, which will be soon included only
by the tests.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Instead of ./check failing when a binary is missing, we try each test
case now and each one fails with tons of test case diffs. Also, all the
variables were initialized by "check" prior to "common" being sourced,
and then (uselessly) checked for emptiness again in "check".
Centralize the search for programs in "common" (which will soon be
one with "check"), including the "realpath" invocation which can be done
just once in "check" rather than in the tests.
For qnio_server, move the detection to "common", simplifying
set_prog_path to stop handling the unused second argument, and
embedding the "realpath" pass.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Some functions in common.rc are never used by the tests. Move
them out of that file and into common, which is already included
only by "check".
Code that actually *is* common to "check" and tests can be placed in
common.config.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This includes shell function, shell variables and command line options
(randomize.awk does not exist).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Proper support of persistent reservation for multipath devices requires
communication with the multipath daemon, so that the reservation is
registered and applied when a path comes up. The device mapper
utilities provide a library to do so; this patch makes qemu-pr-helper.c
detect multipath devices and, when one is found, delegate the operation
to libmpathpersist.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Introduce a privileged helper to run persistent reservation commands.
This lets virtual machines send persistent reservations without using
CAP_SYS_RAWIO or out-of-tree patches. The helper uses Unix permissions
and SCM_RIGHTS to restrict access to processes that can access its socket
and prove that they have an open file descriptor for a raw SCSI device.
The next patch will also correct the usage of persistent reservations
with multipath devices.
It would also be possible to support for Linux's IOC_PR_* ioctls in
the future, to support NVMe devices. For now, however, only SCSI is
supported.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
It is a common requirement for virtual machine to send persistent
reservations, but this currently requires either running QEMU with
CAP_SYS_RAWIO, or using out-of-tree patches that let an unprivileged
QEMU bypass Linux's filter on SG_IO commands.
As an alternative mechanism, the next patches will introduce a
privileged helper to run persistent reservation commands without
expanding QEMU's attack surface unnecessarily.
The helper is invoked through a "pr-manager" QOM object, to which
file-posix.c passes SG_IO requests for PERSISTENT RESERVE OUT and
PERSISTENT RESERVE IN commands. For example:
$ qemu-system-x86_64
-device virtio-scsi \
-object pr-manager-helper,id=helper0,path=/var/run/qemu-pr-helper.sock
-drive if=none,id=hd,driver=raw,file.filename=/dev/sdb,file.pr-manager=helper0
-device scsi-block,drive=hd
or:
$ qemu-system-x86_64
-device virtio-scsi \
-object pr-manager-helper,id=helper0,path=/var/run/qemu-pr-helper.sock
-blockdev node-name=hd,driver=raw,file.driver=host_device,file.filename=/dev/sdb,file.pr-manager=helper0
-device scsi-block,drive=hd
Multiple pr-manager implementations are conceivable and possible, though
only one is implemented right now. For example, a pr-manager could:
- talk directly to the multipath daemon from a privileged QEMU
(i.e. QEMU links to libmpathpersist); this makes reservation work
properly with multipath, but still requires CAP_SYS_RAWIO
- use the Linux IOC_PR_* ioctls (they require CAP_SYS_ADMIN though)
- more interestingly, implement reservations directly in QEMU
through file system locks or a shared database (e.g. sqlite)
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
A container can be used instead of an alias to allow switching between
multiple subregions. In this case we cannot directly share the
subregions (since they only belong to a single parent), but if the
subregions are aliases we can in turn walk those.
This is not enough to remove all source of quadratic FlatView creation,
but it enables sharing of the PCI bus master FlatViews (and their
AddressSpaceDispatch structures) across all PCI devices. For 112
virtio-net-pci devices, boot time is reduced from 25 to 10 seconds and
memory consumption from 1.4 to 1 G.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
It's possible for address_space_get_flatview() as it currently stands
to cause a use-after-free for the returned FlatView, if the reference
count is incremented after the FlatView has been replaced by a writer:
thread 1 thread 2 RCU thread
-------------------------------------------------------------
rcu_read_lock
read as->current_map
set as->current_map
flatview_unref
'--> call_rcu
flatview_ref
[ref=1]
rcu_read_unlock
flatview_destroy
<badness>
Since FlatViews are not updated very often, we can just detect the
situation using a new atomic op atomic_fetch_inc_nonzero, similar to
Linux's atomic_inc_not_zero, which performs the refcount increment only if
it hasn't already hit zero. This is similar to Linux commit de09a9771a53
("CRED: Fix get_task_cred() and task_state() to not resurrect dead
credentials", 2010-07-29).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The pixman submodule does not exist anymore, and its removal broke
docker-based tests. Fix it.
Cc: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This test provides its own mocks, so do not use the "standard"
stubs in libqemustub.a or the event loop implementation in
libqemuutil.a.
This is required on OS X, which otherwise brings in qemu-timer.o,
async.o and main-loop.o from libqemuutil.a.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Complete the transition by renaming this header, which was
shared by block/iscsi.c and the SCSI emulation code.
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move more knowledge of SG_IO out of hw/scsi/scsi-generic.c, for
reusability.
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move more knowledge of sense data format out of hw/scsi/scsi-bus.c
for reusability.
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
util/scsi.c includes some SCSI code that is shared by block/iscsi.c and
hw/scsi, but the introduction of the persistent reservation helper
will add many more instances of this. There is also include/block/scsi.h,
which actually is not part of the core block layer.
The persistent reservation manager will also need a home. A scsi/
directory provides one for both the aforementioned shared code and
the PR manager code.
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
After introducing the scsi/ subdirectory, there will be a scsi_build_sense
function that is the same as scsi_req_build_sense but without needing
a SCSIRequest. The existing scsi_build_sense function gets in the way,
remove it.
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This reverts commit a59629fcc6.
This is not needed anymore because the IOThread mutex is not
"magic" anymore (need not kick the CPU thread)and also because
fork callbacks are only enabled at the very beginning of
QEMU's execution.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Because of -daemonize, system mode QEMU sometimes needs to fork() and
keep RCU enabled in the child. However, there is a possible deadlock
with synchronize_rcu:
- the CPU thread is inside a RCU critical section and wants to take
the BQL in order to do MMIO
- the monitor thread, which is owning the BQL, calls rcu_init_lock
which tries to take the rcu_sync_lock
- the call_rcu thread has taken rcu_sync_lock in synchronize_rcu, but
synchronize_rcu needs the CPU thread to end the critical section
before returning.
This cannot happen for user-mode emulation, because it does not have
a BQL.
To fix it, assume that system mode QEMU only forks in preparation for
exec (except when daemonizing) and disable pthread_atfork as soon as
the double fork has happened.
Reported-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Tested-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
In some cases, the guest can observe the wrong ordering of UIP and
interrupts. This can happen if the VCPU exit is timed like this:
iothread VCPU
... wait for interrupt ...
t-100ns read register A
t wake up, take BQL
t+100ns update_in_progress
return false
return UIP=0
trigger interrupt
The interrupt is late; the VCPU expected the falling edge of UIP to
happen after the interrupt. update_in_progress is already trying to
cover this case by latching UIP if the timer is going to fire soon,
and the fix is documented in the commit message for commit 56038ef623
("RTC: Update the RTC clock only when reading it", 2012-09-10). It
cannot be tested with qtest, because its timing of interrupts vs. reads
is exact.
However, the implementation was incorrect because UIP cmos_ioport_read
cleared register A instead of leaving that to rtc_update_timer. Fixing
the implementation of cmos_ioport_read to match the commit message,
however, breaks the "uip-stuck" test case from the previous patch.
To fix it, skip update timer optimizations if UIP has been latched.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Clang static analyzer reports a memory leak. Actually, the allocated
memory escapes here:
record->attribute_list[record->attributes].pair = data;
but clang is correct that the memory might leak if len is zero. We
know it isn't; assert that it is the case.
The craziness doesn't end there. The memory is freed by
bt_l2cap_sdp_close_ch:
g_free(sdp->service_list[i].attribute_list->pair);
which actually should have been written like this:
g_free(sdp->service_list[i].attribute_list[0].pair);
The attribute_list is sorted with qsort; but indeed the first
entry of attribute_list should point to "data" even after the qsort,
because the first record has id SDP_ATTR_RECORD_HANDLE, whose
numeric value is zero.
But hang on. The qsort function is
static int sdp_attributeid_compare(
const struct sdp_service_attribute_s *a,
const struct sdp_service_attribute_s *b)
{
return (int) b->attribute_id - a->attribute_id;
}
but no one ever writes attribute_id. So it only works if qsort is
stable, and who knows what else is broken, but we can fix it by
setting attribute_id in the while loop.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Because global environment variables can be overridden when .travis.yml
is processed by the docker-travis target, the effect of this patch is
that docker-travis now obeys the "J=n" option.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
This is useful so that we can do builds at higher than -j3 when running
travis.py locally.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
The VirtualBox driver is using a mutex to order all allocating writes,
but it is not protecting accesses to the bitmap because they implicitly
happen under the AioContext mutex. Change this to use a CoRwlock
explicitly.
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20170629132749.997-4-pbonzini@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
This warning is included in -Wall by clang, but not by GCC (which only
enables it for -Wextra). Include it in the list of warnings we enable
to minimize the differences between the compilers:
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This was used to extract .txt documentation for QMP. This was
changed to use the QAPI schema instead, so zap it.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
This function calls tlb_set_page_with_attrs, which is not available
when TCG is disabled. Move it to excp_helper.c.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
translate-common.c will not be available anymore with --disable-tcg,
so we cannot leave cpu_interrupt_handler there.
Move the TCG-specific handler to accel/tcg/tcg-all.c, and adopt
KVM's handler as the default one, since it works just as well for
Xen and qtest.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This lets you build without TCG (hardware accelerationor qtest only). When
this flag is passed to configure, it will automatically filter out the target
list to only those that support KVM or Xen or HAX.
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Check for unsupported targets in target_list, and print an
error early in the configuration process.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This will be useful when the functions are called, early in the configure
process, to filter out targets that do not support hardware acceleration.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Using signal to establish a signal handler is not portable; on
SysV systems, the signal handler would be reset to SIG_DFL after
delivery, while BSD preserves the signal handler. Daniel Berrange
reported that (to complicate matters further) the signal system call
has SysV behavior, but glibc signal() actually calls the sigaction
system call to provide BSD behavior.
However, using signal() to set a signal's disposition to SIG_DFL
or SIG_IGN is portable and is a relatively common occurrence in
QEMU source code, so allow that.
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
Reviewed-by: Richard W.M. Jones <rjones@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move the handling of conforming code segments before the handling
of stack switch.
Because dpl == cpl after the new "if", it's now unnecessary to check
the C bit when testing dpl < cpl. Furthermore, dpl > cpl is checked
slightly above the modified code, so the final "else" is unreachable
and we can remove it.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When attaching the NBD QIOChannel to an AioContext, the TLS channel should
be used, not the underlying socket channel. This is because, trivially,
the TLS channel will be the one that we read/write to and thus the one
that will get the qio_channel_yield() call.
Fixes: ff82911cd3
Cc: qemu-stable@nongnu.org
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
Tested-by: Daniel P. Berrange <berrange@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
block_acct_destroy is called unconditionally in blk_delete, but there is
no BlockAcctStats function that is called unconditionally in blk_new.
Split block_acct_init in two, so that it will be possible to create a
QemuMutex in block_acct_init and destroy it in block_acct_cleanup.
Cc: Alberto Garcia <berto@igalia.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20170605123908.18777-19-pbonzini@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Another possibility is to use tg->lock, which we're holding anyway in
both schedule_next_request and throttle_group_co_io_limits_intercept.
This would require open-coding the CoQueue however, so I've chosen this
alternative.
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20170605123908.18777-10-pbonzini@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Starting all waiting coroutines from bdrv_drain_all is unnecessary;
throttle_group_co_io_limits_intercept calls schedule_next_request as
soon as the coroutine restarts, which in turn will restart the next
request if possible.
If we only start the first request and let the coroutines dance from
there the code is simpler and there is more reuse between
throttle_group_config, throttle_group_restart_blk and timer_cb. The
next patch will benefit from this.
We also stop accessing from throttle_group_restart_blk the
blkp->throttled_reqs CoQueues even when there was no
attached throttling group. This worked but is not pretty.
The only thing that can interrupt the dance is the QEMU_CLOCK_VIRTUAL
timer when switching from one block device to the next, because the
timer is set to "now + 1" but QEMU_CLOCK_VIRTUAL might not be running.
Set that timer to point in the present ("now") rather than the future
and things work.
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20170605123908.18777-8-pbonzini@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Paolo Bonzini