Peter Maydell [Fri, 5 Aug 2016 10:03:59 +0000 (11:03 +0100)]
hw/sparc/leon3: Don't call get_image_size() on a NULL pointer
get_image_size() doesn't handle being passed a NULL pointer, so
avoid doing that. Spotted by the clang ub sanitizer (which notices
the attempt to pass NULL to open()).
Peter Maydell [Mon, 8 Aug 2016 11:41:38 +0000 (12:41 +0100)]
Merge remote-tracking branch 'remotes/elmarco/tags/leaks-for-2.7-pull-request' into staging
# gpg: Signature made Sun 07 Aug 2016 21:03:14 BST
# gpg: using RSA key 0xDAE8E10975969CE5
# gpg: Good signature from "Marc-André Lureau <[email protected]>"
# gpg: aka "Marc-André Lureau <[email protected]>"
# 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: 87A9 BD93 3F87 C606 D276 F62D DAE8 E109 7596 9CE5
* remotes/elmarco/tags/leaks-for-2.7-pull-request:
ahci: fix sglist leak on retry
usb: free leaking path
usb: free USBDevice.strings
virtio-input: free config list
qjson: free str
ahci: free irqs array
char: free MuxDriver when closing
char: free the tcp connection data when closing
numa: do not leak NumaOptions
Peter Maydell [Mon, 8 Aug 2016 10:32:01 +0000 (11:32 +0100)]
Merge remote-tracking branch 'remotes/dgibson/tags/ppc-for-2.7-20160808' into staging
ppc patch queue 2016-08-08
This batch has several last minute bug fixes to be merged for
qemu-2.7.
# gpg: Signature made Mon 08 Aug 2016 03:40:58 BST
# gpg: using RSA key 0x6C38CACA20D9B392
# gpg: Good signature from "David Gibson <[email protected]>"
# gpg: aka "David Gibson (Red Hat) <[email protected]>"
# gpg: aka "David Gibson (ozlabs.org) <[email protected]>"
# 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: 75F4 6586 AE61 A66C C44E 87DC 6C38 CACA 20D9 B392
* remotes/dgibson/tags/ppc-for-2.7-20160808:
spapr: Fix undefined behaviour in spapr_tce_reset()
macio: set res_count value to 0 after non-block ATAPI DMA transfers
spapr: Correctly set query_hotpluggable_cpus hook based on machine version
* remotes/rth/tags/pull-tcg-20160805:
tcg: Lower indirect registers in a separate pass
tcg: Require liveness analysis
tcg: Include liveness info in the dumps
tcg: Compress dead_temps and mem_temps into a single array
tcg: Fold life data into TCGOp
tcg: Reorg TCGOp chaining
tcg: Compress liveness data to 16 bits
checkpatch: Fix newline detection in error_setg() & friends
Commit 5d596c2's regexp assumes the error message string is the first
argument. Correct for error_report(), wrong for all the others.
Relax the regexp to match newline in anywhere. This might cause
additional false positives.
While there, update the list of error_reporting functions.
error: Strip trailing '\n' from error string arguments (again)
Commit 9af9e0f, 6daf194d, be62a2eb and 312fd5f got rid of a bunch, but
they keep coming back. checkpatch.pl tries to flag them since commit 5d596c2, but it's not very good at it. Offenders tracked down with
Coccinelle script scripts/coccinelle/err-bad-newline.cocci, an updated
version of the script from commit 312fd5f.
David Gibson [Mon, 8 Aug 2016 00:06:25 +0000 (10:06 +1000)]
spapr: Fix undefined behaviour in spapr_tce_reset()
When a TCE table (sPAPR IOMMU context) is in disabled state (which is true
by default for the 64-bit window), it has tcet->nb_table == 0 and
tcet->table == NULL. However, on system reset, spapr_tce_reset() executes,
which unconditionally calls
memset(tcet->table, 0, table_size);
We get away with this in practice, because it's a zero length memset(),
but memset() on a NULL pointer is undefined behaviour, so we should not
call it in this case.
macio: set res_count value to 0 after non-block ATAPI DMA transfers
res_count should be set to the number of outstanding bytes after a DBDMA
request. Unfortunately this wasn't being set to zero by the non-block
transfer codepath meaning drivers that checked the descriptor result for
such requests (e.g reading the CDROM TOC) would assume from a non-zero result
that the transfer had failed.
David Gibson [Fri, 5 Aug 2016 06:25:33 +0000 (16:25 +1000)]
spapr: Correctly set query_hotpluggable_cpus hook based on machine version
Prior to c8721d3 "spapr: Error out when CPU hotplug is attempted on older
pseries machines", attempting to use query-hotpluggable-cpus on pseries-2.6
and earlier machine types would SEGV.
That change fixed that, but due to some unexpected interactions in init
order and a brown-paper-bag worthy failure to test, it accidentally
disabled query-hotpluggable-cpus for all pseries machine types, including
the current one which should allow it.
In fact, query_hotpluggable_cpus needs to be non-NULL when and only when
the dr_cpu_enabled flag in sPAPRMachineClass is set, which makes
dr_cpu_enabled itself redundant.
This patch removes dr_cpu_enabled, instead directly setting
query_hotpluggable_cpus from the machine class_init functions, and using
that to determine the availability of CPU hotplug when necessary.
ahci-test /x86_64/ahci/io/dma/lba28/retry triggers the following leak:
Direct leak of 16 byte(s) in 1 object(s) allocated from:
#0 0x7fc4b2a25e20 in malloc (/lib64/libasan.so.3+0xc6e20)
#1 0x7fc4993bce58 in g_malloc (/lib64/libglib-2.0.so.0+0x4ee58)
#2 0x556a187d4b34 in ahci_populate_sglist hw/ide/ahci.c:896
#3 0x556a187d8237 in ahci_dma_prepare_buf hw/ide/ahci.c:1367
#4 0x556a187b5a1a in ide_dma_cb hw/ide/core.c:844
#5 0x556a187d7eec in ahci_start_dma hw/ide/ahci.c:1333
#6 0x556a187b650b in ide_start_dma hw/ide/core.c:921
#7 0x556a187b61e6 in ide_sector_start_dma hw/ide/core.c:911
#8 0x556a187b9e26 in cmd_write_dma hw/ide/core.c:1486
#9 0x556a187bd519 in ide_exec_cmd hw/ide/core.c:2027
#10 0x556a187d71c5 in handle_reg_h2d_fis hw/ide/ahci.c:1204
#11 0x556a187d7681 in handle_cmd hw/ide/ahci.c:1254
#12 0x556a187d168a in check_cmd hw/ide/ahci.c:510
#13 0x556a187d0afc in ahci_port_write hw/ide/ahci.c:314
#14 0x556a187d105d in ahci_mem_write hw/ide/ahci.c:435
#15 0x556a1831d959 in memory_region_write_accessor /home/elmarco/src/qemu/memory.c:525
#16 0x556a1831dc35 in access_with_adjusted_size /home/elmarco/src/qemu/memory.c:591
#17 0x556a18323ce3 in memory_region_dispatch_write /home/elmarco/src/qemu/memory.c:1262
#18 0x556a1828cf67 in address_space_write_continue /home/elmarco/src/qemu/exec.c:2578
#19 0x556a1828d20b in address_space_write /home/elmarco/src/qemu/exec.c:2635
#20 0x556a1828d92b in address_space_rw /home/elmarco/src/qemu/exec.c:2737
#21 0x556a1828daf7 in cpu_physical_memory_rw /home/elmarco/src/qemu/exec.c:2746
#22 0x556a183068d3 in cpu_physical_memory_write /home/elmarco/src/qemu/include/exec/cpu-common.h:72
#23 0x556a18308194 in qtest_process_command /home/elmarco/src/qemu/qtest.c:382
#24 0x556a18309999 in qtest_process_inbuf /home/elmarco/src/qemu/qtest.c:573
#25 0x556a18309a4a in qtest_read /home/elmarco/src/qemu/qtest.c:585
#26 0x556a18598b85 in qemu_chr_be_write_impl /home/elmarco/src/qemu/qemu-char.c:387
#27 0x556a18598c52 in qemu_chr_be_write /home/elmarco/src/qemu/qemu-char.c:399
#28 0x556a185a2afa in tcp_chr_read /home/elmarco/src/qemu/qemu-char.c:2902
#29 0x556a18cbaf52 in qio_channel_fd_source_dispatch io/channel-watch.c:84
Follow John Snow recommendation:
Everywhere else ncq_err is used, it is accompanied by a list cleanup
except for ncq_cb, which is the case you are fixing here.
Move the sglist destruction inside of ncq_err and then delete it from
the other two locations to keep it tidy.
Call dma_buf_commit in ide_dma_cb after the early return. Though, this
is also a little wonky because this routine does more than clear the
list, but it is at the moment the centralized "we're done with the
sglist" function and none of the other side effects that occur in
dma_buf_commit will interfere with the reset that occurs from
ide_restart_bh, I think
Clear the list when finalizing. The list is created during realize with
virtio_input_idstr_config() and later by further calls to
virtio_input_init_config() and virtio_input_add_config().
This leak can be reproduced with device-introspect-test -p
/x86_64/device/introspect/concrete.
Make sure the connection data got freed when closing the chardev, to
avoid leaks. Introduce tcp_chr_free_connection() to clean all connection
related data, and move some tcp_chr_close() clean-ups there.
(while at it, set write_msgfds_num to 0 when clearing array in
tcp_set_msgfds())
Rather than rely on recursion during the middle of register allocation,
lower indirect registers to loads and stores off the indirect base into
plain temps.
For an x86_64 host, with sufficient registers, this results in identical
code, modulo the actual register assignments.
For an i686 host, with insufficient registers, this means that temps can
be (temporarily) spilled to the stack in order to satisfy an allocation.
This as opposed to the possibility of not being able to spill, to allocate
a register for the indirect base, in order to perform a spill.
Peter Maydell [Thu, 4 Aug 2016 10:39:38 +0000 (11:39 +0100)]
tests: Rename qtests which have names ending "error"
We have three qtest tests which have test names ending with "error".
This is awkward because the output of verbose test runs looks like
/crypto/task/error: OK
/crypto/task/thread_error: OK
which gives false positives if you are grepping build logs for
errors by looking for "error:". Since there are only three tests
with this problem, just rename them all to 'failure' instead.
Peter Maydell [Fri, 5 Aug 2016 10:44:56 +0000 (11:44 +0100)]
Merge remote-tracking branch 'remotes/stefanha/tags/block-pull-request' into staging
# gpg: Signature made Fri 05 Aug 2016 10:24:34 BST
# gpg: using RSA key 0x9CA4ABB381AB73C8
# gpg: Good signature from "Stefan Hajnoczi <[email protected]>"
# gpg: aka "Stefan Hajnoczi <[email protected]>"
# Primary key fingerprint: 8695 A8BF D3F9 7CDA AC35 775A 9CA4 ABB3 81AB 73C8
* remotes/stefanha/tags/block-pull-request:
virtio-blk: Remove stale comment about draining
virtio-blk: Release s->rq queue at system_reset
throttle: Test burst limits lower than the normal limits
throttle: Don't allow burst limits to be lower than the normal limits
block/parallels: check new image size
Peter Maydell [Fri, 5 Aug 2016 10:16:30 +0000 (11:16 +0100)]
Merge remote-tracking branch 'remotes/famz/tags/docker-pull-request' into staging
# gpg: Signature made Fri 05 Aug 2016 09:58:50 BST
# gpg: using RSA key 0xCA35624C6A9171C6
# gpg: Good signature from "Fam Zheng <[email protected]>"
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg: There is no indication that the signature belongs to the owner.
# Primary key fingerprint: 5003 7CB7 9706 0F76 F021 AD56 CA35 624C 6A91 71C6
* remotes/famz/tags/docker-pull-request:
docker: Add "--enable-werror" to configure command line
docker: Be compatible with older docker
Alberto Garcia [Thu, 28 Jul 2016 08:08:13 +0000 (11:08 +0300)]
throttle: Test burst limits lower than the normal limits
This checks that making FOO_max lower than FOO is not allowed.
We could also forbid having FOO_max == FOO, but that doesn't have
any odd side effects and it would require us to update several other
tests, so let's keep it simple.
Alberto Garcia [Thu, 28 Jul 2016 08:08:12 +0000 (11:08 +0300)]
throttle: Don't allow burst limits to be lower than the normal limits
Setting FOO_max to a value that is lower than FOO does not make
sense, and it produces odd results depending on the value of
FOO_max_length. Although the user should not set that configuration
in the first place it's better to reject it explicitly.
Before this patch incorrect image could be created via qemu-img
(Example: qemu-img create -f parallels -o size=4096T hack.img),
incorrect images cannot be used due to overflow in main image structure.
This patch add check of size in image creation.
After reading size it compare it with UINT32_MAX * cluster_size.
The broken Identify implementation in earlier Qemu versions means we
need to blacklist it from issueing the NVMe 1.1 Identify Namespace List
command. As we want to be able to use it in newer Qemu versions we need
a way to identify those. Bump the PCI revision as a guest visible
indicator of this bug fix.
NVMe 1.1 requires devices to implement a Namespace List subcommand of
the identify command. Qemu not only not implements this features, but
also misinterprets it as an Identify Controller request. Due to this
any OS trying to use the Namespace List will fail the probe.
Kevin Wolf [Tue, 2 Aug 2016 17:22:04 +0000 (19:22 +0200)]
block: Accept any target node for transactional blockdev-backup
Commit 0d978913 changed blockdev-backup to accept arbitrary node names
instead of device names (i.e. root nodes) for the backup target.
However, it forgot to make the same change in transactions and to update
the documentation. This patch fixes these omissions.
Bruce Rogers [Wed, 27 Jul 2016 15:30:48 +0000 (09:30 -0600)]
Xen PCI passthrough: fix passthrough failure when no interrupt pin
Commit 5a11d0f7 mistakenly converted a log message into an error
condition when no pin interrupt is found for the pci device being
passed through. Revert that part of the commit.
Peter Maydell [Thu, 4 Aug 2016 17:36:05 +0000 (18:36 +0100)]
Merge remote-tracking branch 'remotes/riku/tags/pull-linux-user-20160804' into staging
linux-user important fixes for 2.7
# gpg: Signature made Thu 04 Aug 2016 15:10:57 BST
# gpg: using RSA key 0xB44890DEDE3C9BC0
# gpg: Good signature from "Riku Voipio <[email protected]>"
# gpg: aka "Riku Voipio <[email protected]>"
# Primary key fingerprint: FF82 03C8 C391 98AE 0581 41EF B448 90DE DE3C 9BC0
* remotes/riku/tags/pull-linux-user-20160804:
linux-user: Handle brk() attempts with very large sizes
linux-user: Fix target_semid_ds structure definition
linux-user: Don't write off end of new_utsname buffer
linux-user: Fix memchr() argument in open_self_cmdline()
linux-user: Use correct alignment for long long on i386 guests
Peter Maydell [Fri, 22 Jul 2016 12:41:52 +0000 (13:41 +0100)]
linux-user: Handle brk() attempts with very large sizes
In do_brk(), we were inadvertently truncating the size
of a requested brk() from the guest by putting it into an
'int' variable. This meant that we would incorrectly report
success back to the guest rather than a failed allocation,
typically resulting in the guest then segfaulting. Use
abi_ulong instead.
This fixes a crash in the '31370.cc' test in the gcc libstdc++ test
suite (the test case starts by trying to allocate a very large
size and reduces the size until the allocation succeeds).
The target_semid_ds structure is not correct for all
architectures: the padding fields should only exist for:
* 32-bit ABIs
* x86
It is also misnamed, since it is following the kernel
semid64_ds structure (QEMU doesn't support the legacy
semid_ds structure at all). Rename the struct, provide
a correct generic definition and allow the oddball x86
architecture to provide its own version.
This fixes broken SYSV semaphores for all our 64-bit
architectures except x86 and ppc.
Peter Maydell [Tue, 12 Jul 2016 12:02:16 +0000 (13:02 +0100)]
linux-user: Don't write off end of new_utsname buffer
Use g_strlcpy() rather than strcpy() to copy the uname string
into the structure we return to the guest for the uname syscall.
This avoids overrunning the buffer if the user passed us an
overlong string via the QEMU command line.
We fix a comment typo while we're in the neighbourhood.
Peter Maydell [Tue, 12 Jul 2016 12:02:19 +0000 (13:02 +0100)]
linux-user: Fix memchr() argument in open_self_cmdline()
In open_self_cmdline() we look for a 0 in the buffer we read
from /prc/self/cmdline. We were incorrectly passing the length
of our buf[] array to memchr() as the length to search, rather
than the number of bytes we actually read into it, which could
be shorter. This was spotted by Coverity (because it could
result in our trying to pass a negative length argument to
write()).
Peter Maydell [Thu, 28 Jul 2016 11:57:59 +0000 (12:57 +0100)]
linux-user: Use correct alignment for long long on i386 guests
For i386, the ABI specifies that 'long long' (8 byte values)
need only be 4 aligned, but we were requiring them to be
8-aligned. This meant we were laying out the target_epoll_event
structure wrongly. Add a suitable ifdef to abitypes.h to
specify the i386-specific alignment requirement.
* remotes/bonzini/tags/for-upstream: (25 commits)
util: Fix assertion in iov_copy() upon zero 'bytes' and non-zero 'offset'
qdev: Fix use after free in qdev_init_nofail error path
Reorganize help output of '-display' option
x86: ioapic: add support for explicit EOI
x86: ioapic: ignore level irq during processing
apic: fix broken migration for kvm-apic
fw_cfg: Make base type "fw_cfg" abstract
block: Cater to iscsi with non-power-of-2 discard
osdep: Document differences in rounding macros
nbd: Limit nbdflags to 16 bits
nbd: Fix bad flag detection on server
i2c: fix migration regression introduced by broadcast support
mptsas: really fix migration compatibility
qdist: return "(empty)" instead of NULL when printing an empty dist
qdist: use g_renew and g_new instead of g_realloc and g_malloc.
qdist: fix memory leak during binning
target-i386: fix typo in xsetbv implementation
qht: do not segfault when gathering stats from an uninitialized qht
util: Drop inet_listen()
util: drop unix_nonblocking_connect()
...
* remotes/kraxel/tags/pull-vnc-20160803-1:
vnc: ensure connection sharing/limits is always configured
vnc: fix crash when vnc_server_info_get has an error
vnc: don't crash getting server info if lsock is NULL
Fam Zheng [Tue, 2 Aug 2016 03:41:41 +0000 (11:41 +0800)]
qdev: Fix use after free in qdev_init_nofail error path
Since 69382d8b (qdev: Fix object reference leak in case device.realize()
fails), object_property_set_bool could release the object. The error
path wants the type name, so hold an reference before realizing it.
Robert Ho [Tue, 26 Jul 2016 10:17:11 +0000 (18:17 +0800)]
Reorganize help output of '-display' option
The '-display' help information is not very correct. This patch sort
it a little.
Also, in its help information, reveals what implicit display option
will be chosen if no definition.
Peter Xu [Mon, 1 Aug 2016 13:59:19 +0000 (21:59 +0800)]
x86: ioapic: add support for explicit EOI
Some old Linux kernels (upstream before v4.0), or any released RHEL
kernels has problem in sending APIC EOI when IR is enabled. Meanwhile,
many of them only support explicit EOI for IOAPIC, which is only
introduced in IOAPIC version 0x20. This patch provide a way to boost
QEMU IOAPIC to version 0x20, in order for QEMU to correctly receive EOI
messages.
Without boosting IOAPIC version to 0x20, kernels before commit d32932d
("x86/irq: Convert IOAPIC to use hierarchical irqdomain interfaces")
will have trouble enabling both IR and level-triggered interrupt devices
(like e1000).
To upgrade IOAPIC to version 0x20, we need to specify:
-global ioapic.version=0x20
To be compatible with old systems, 0x11 will still be the default IOAPIC
version. Here 0x11 and 0x20 are the only versions to be supported.
One thing to mention: this patch only applies to emulated IOAPIC. It
does not affect kernel IOAPIC behavior.
Peter Xu [Sun, 31 Jul 2016 14:18:05 +0000 (22:18 +0800)]
x86: ioapic: ignore level irq during processing
For level triggered interrupts, we will get Remote IRR bit cleared after
guest kernel finished processing specific request. Before that, we
should ignore the same interrupt from triggering again.
Igor Mammedov [Fri, 29 Jul 2016 13:55:42 +0000 (15:55 +0200)]
apic: fix broken migration for kvm-apic
commit f6e98444 (apic: Use apic_id as apic's migration instance_id)
breaks migration when in kernel irqchip is used for 2.6 and older
machine types.
It applies compat property only for userspace 'apic' type
instead of applying it to all apic types inherited from
'apic-common' type as it was supposed to do.
Fix it by setting compat property 'legacy-instance-id' for
'apic-common' type which affects inherited types (i.e. not
only 'apic' but also 'kvm-apic' types)
Eric Blake [Thu, 21 Jul 2016 19:34:48 +0000 (13:34 -0600)]
block: Cater to iscsi with non-power-of-2 discard
Dell Equallogic iSCSI SANs have a very unusual advertised geometry:
$ iscsi-inq -e 1 -c $((0xb0)) iscsi://XXX/0
wsnz:0
maximum compare and write length:1
optimal transfer length granularity:0
maximum transfer length:0
optimal transfer length:0
maximum prefetch xdread xdwrite transfer length:0
maximum unmap lba count:30720
maximum unmap block descriptor count:2
optimal unmap granularity:30720
ugavalid:1
unmap granularity alignment:0
maximum write same length:30720
which says that both the maximum and the optimal discard size
is 15M. It is not immediately apparent if the device allows
discard requests not aligned to the optimal size, nor if it
allows discards at a finer granularity than the optimal size.
I tried to find details in the SCSI Commands Reference Manual
Rev. A on what valid values of maximum and optimal sizes are
permitted, but while that document mentions a "Block Limits
VPD Page", I couldn't actually find documentation of that page
or what values it would have, or if a SCSI device has an
advertisement of its minimal unmap granularity. So it is not
obvious to me whether the Dell Equallogic device is compliance
with the SCSI specification.
Fortunately, it is easy enough to support non-power-of-2 sizing,
even if it means we are less efficient than truly possible when
targetting that device (for example, it means that we refuse to
unmap anything that is not a multiple of 15M and aligned to a
15M boundary, even if the device truly does support a smaller
granularity where unmapping actually works).
Eric Blake [Thu, 21 Jul 2016 19:34:46 +0000 (13:34 -0600)]
nbd: Limit nbdflags to 16 bits
Rather than asserting that nbdflags is within range, just give
it the correct type to begin with :) nbdflags corresponds to
the per-export portion of NBD Protocol "transmission flags", which
is 16 bits in response to NBD_OPT_EXPORT_NAME and NBD_OPT_GO.
Furthermore, upstream NBD has never passed the global flags to
the kernel via ioctl(NBD_SET_FLAGS) (the ioctl was first
introduced in NBD 2.9.22; then a latent bug in NBD 3.1 actually
tried to OR the global flags with the transmission flags, with
the disaster that the addition of NBD_FLAG_NO_ZEROES in 3.9
caused all earlier NBD 3.x clients to treat every export as
read-only; NBD 3.10 and later intentionally clip things to 16
bits to pass only transmission flags). Qemu should follow suit,
since the current two global flags (NBD_FLAG_FIXED_NEWSTYLE
and NBD_FLAG_NO_ZEROES) have no impact on the kernel's behavior
during transmission.
Igor Mammedov [Wed, 27 Jul 2016 12:39:58 +0000 (14:39 +0200)]
i2c: fix migration regression introduced by broadcast support
QEMU fails migration with following error:
qemu-system-x86_64: Missing section footer for i2c_bus
qemu-system-x86_64: load of migration failed: Invalid argument
when migrating from:
qemu-system-x86_64-v2.6.0 -m 256M rhel72.img -M pc-i440fx-2.6
to
qemu-system-x86_64-v2.7.0-rc0 -m 256M rhel72.img -M pc-i440fx-2.6
Regression is added by commit 2293c27f (i2c: implement broadcast write)
Fix it by dropping 'broadcast' VMState introduced by 2293c27f and
reuse broadcast 0x00 address as broadcast flag in bus->saved_address.
Then if there were ongoing broadcast at migration time, set
bus->saved_address to it and at i2c_slave_post_load() time check
for it instead of transfering and using 'broadcast' VMState.
As result of reusing existing saved_address VMState, no compat
glue will be needed to keep forward/backward compatiblity. which
makes fix much less intrusive.
* remotes/kraxel/tags/pull-usb-20160803-1:
xen: use a common function for pv and hvm guest backend register calls
xen: drain submit queue in xen-usb before removing device
xen: when removing a backend don't remove many of them
ehci: faster frame index calculation for skipped frames
wxx: Fix compilation of host-libusb.c
wxx: Fix compiler warning for host-libusb.c
vnc: ensure connection sharing/limits is always configured
The connection sharing / limits are only set in the
vnc_display_open() method and so missed when VNC is running
with '-vnc none'. This in turn prevents clients being added
to the VNC server with the QMP "add_client" command.
vnc: fix crash when vnc_server_info_get has an error
The vnc_server_info_get will allocate the VncServerInfo
struct and then call vnc_init_basic_info_from_server_addr
to populate the basic fields. If this returns an error
though, the qapi_free_VncServerInfo call will then crash
because the VncServerInfo struct instance was not properly
NULL-initialized and thus contains random stack garbage.
#0 0x00007f1987c8e6f5 in raise () at /lib64/libc.so.6
#1 0x00007f1987c902fa in abort () at /lib64/libc.so.6
#2 0x00007f1987ccf600 in __libc_message () at /lib64/libc.so.6
#3 0x00007f1987cd7d4a in _int_free () at /lib64/libc.so.6
#4 0x00007f1987cdb2ac in free () at /lib64/libc.so.6
#5 0x00007f198b654f6e in g_free () at /lib64/libglib-2.0.so.0
#6 0x0000559193cdcf54 in visit_type_str (v=v@entry=
0x5591972f14b0, name=name@entry=0x559193de1e29 "host", obj=obj@entry=0x5591961dbfa0, errp=errp@entry=0x7fffd7899d80)
at qapi/qapi-visit-core.c:255
#7 0x0000559193cca8f3 in visit_type_VncBasicInfo_members (v=v@entry=
0x5591972f14b0, obj=obj@entry=0x5591961dbfa0, errp=errp@entry=0x7fffd7899dc0) at qapi-visit.c:12307
#8 0x0000559193ccb523 in visit_type_VncServerInfo_members (v=v@entry=
0x5591972f14b0, obj=0x5591961dbfa0, errp=errp@entry=0x7fffd7899e00) at qapi-visit.c:12632
#9 0x0000559193ccb60b in visit_type_VncServerInfo (v=v@entry=
0x5591972f14b0, name=name@entry=0x0, obj=obj@entry=0x7fffd7899e48, errp=errp@entry=0x0) at qapi-visit.c:12658
#10 0x0000559193cb53d8 in qapi_free_VncServerInfo (obj=<optimized out>) at qapi-types.c:3970
#11 0x0000559193c1e6ba in vnc_server_info_get (vd=0x7f1951498010) at ui/vnc.c:233
#12 0x0000559193c24275 in vnc_connect (vs=0x559197b2f200, vs=0x559197b2f200, event=QAPI_EVENT_VNC_CONNECTED) at ui/vnc.c:284
#13 0x0000559193c24275 in vnc_connect (vd=vd@entry=0x7f1951498010, sioc=sioc@entry=0x559196bf9c00, skipauth=skipauth@entry=tru e, websocket=websocket@entry=false) at ui/vnc.c:3039
#14 0x0000559193c25806 in vnc_display_add_client (id=<optimized out>, csock=<optimized out>, skipauth=<optimized out>)
at ui/vnc.c:3877
#15 0x0000559193a90c28 in qmp_marshal_add_client (args=<optimized out>, ret=<optimized out>, errp=0x7fffd7899f90)
at qmp-marshal.c:105
#16 0x000055919399c2b7 in handle_qmp_command (parser=<optimized out>, tokens=<optimized out>)
at /home/berrange/src/virt/qemu/monitor.c:3971
#17 0x0000559193ce3307 in json_message_process_token (lexer=0x559194ab0838, input=0x559194a6d940, type=JSON_RCURLY, x=111, y=1 2) at qobject/json-streamer.c:105
#18 0x0000559193cfa90d in json_lexer_feed_char (lexer=lexer@entry=0x559194ab0838, ch=125 '}', flush=flush@entry=false)
at qobject/json-lexer.c:319
#19 0x0000559193cfaa1e in json_lexer_feed (lexer=0x559194ab0838, buffer=<optimized out>, size=<optimized out>)
at qobject/json-lexer.c:369
#20 0x0000559193ce33c9 in json_message_parser_feed (parser=<optimized out>, buffer=<optimized out>, size=<optimized out>)
at qobject/json-streamer.c:124
#21 0x000055919399a85b in monitor_qmp_read (opaque=<optimized out>, buf=<optimized out>, size=<optimized out>)
at /home/berrange/src/virt/qemu/monitor.c:3987
#22 0x0000559193a87d00 in tcp_chr_read (chan=<optimized out>, cond=<optimized out>, opaque=0x559194a7d900)
at qemu-char.c:2895
#23 0x00007f198b64f703 in g_main_context_dispatch () at /lib64/libglib-2.0.so.0
#24 0x0000559193c484b3 in main_loop_wait () at main-loop.c:213
#25 0x0000559193c484b3 in main_loop_wait (timeout=<optimized out>) at main-loop.c:258
#26 0x0000559193c484b3 in main_loop_wait (nonblocking=<optimized out>) at main-loop.c:506
#27 0x0000559193964c55 in main () at vl.c:1908
#28 0x0000559193964c55 in main (argc=<optimized out>, argv=<optimized out>, envp=<optimized out>) at vl.c:4603
vnc: don't crash getting server info if lsock is NULL
When VNC is started with '-vnc none' there will be no
listener socket present. When we try to populate the
VncServerInfo we'll crash accessing a NULL 'lsock'
field.
#0 qio_channel_socket_get_local_address (ioc=0x0, errp=errp@entry=0x7ffd5b8aa0f0) at io/channel-socket.c:33
#1 0x00007f4b9a297d6f in vnc_init_basic_info_from_server_addr (errp=0x7ffd5b8aa0f0, info=0x7f4b9d425460, ioc=<optimized out>) at ui/vnc.c:146
#2 vnc_server_info_get (vd=0x7f4b9e858000) at ui/vnc.c:223
#3 0x00007f4b9a29d318 in vnc_qmp_event (vs=0x7f4b9ef82000, vs=0x7f4b9ef82000, event=QAPI_EVENT_VNC_CONNECTED) at ui/vnc.c:279
#4 vnc_connect (vd=vd@entry=0x7f4b9e858000, sioc=sioc@entry=0x7f4b9e8b3a20, skipauth=skipauth@entry=true, websocket=websocket @entry=false) at ui/vnc.c:2994
#5 0x00007f4b9a29e8c8 in vnc_display_add_client (id=<optimized out>, csock=<optimized out>, skipauth=<optimized out>) at ui/v nc.c:3825
#6 0x00007f4b9a18d8a1 in qmp_marshal_add_client (args=<optimized out>, ret=<optimized out>, errp=0x7ffd5b8aa230) at qmp-marsh al.c:123
#7 0x00007f4b9a0b53f5 in handle_qmp_command (parser=<optimized out>, tokens=<optimized out>) at /usr/src/debug/qemu-2.6.0/mon itor.c:3922
#8 0x00007f4b9a348580 in json_message_process_token (lexer=0x7f4b9c78dfe8, input=0x7f4b9c7350e0, type=JSON_RCURLY, x=111, y=5 9) at qobject/json-streamer.c:94
#9 0x00007f4b9a35cfeb in json_lexer_feed_char (lexer=lexer@entry=0x7f4b9c78dfe8, ch=125 '}', flush=flush@entry=false) at qobj ect/json-lexer.c:310
#10 0x00007f4b9a35d0ae in json_lexer_feed (lexer=0x7f4b9c78dfe8, buffer=<optimized out>, size=<optimized out>) at qobject/json -lexer.c:360
#11 0x00007f4b9a348679 in json_message_parser_feed (parser=<optimized out>, buffer=<optimized out>, size=<optimized out>) at q object/json-streamer.c:114
#12 0x00007f4b9a0b3a1b in monitor_qmp_read (opaque=<optimized out>, buf=<optimized out>, size=<optimized out>) at /usr/src/deb ug/qemu-2.6.0/monitor.c:3938
#13 0x00007f4b9a186751 in tcp_chr_read (chan=<optimized out>, cond=<optimized out>, opaque=0x7f4b9c7add40) at qemu-char.c:2895
#14 0x00007f4b92b5c79a in g_main_context_dispatch () from /lib64/libglib-2.0.so.0
#15 0x00007f4b9a2bb0c0 in glib_pollfds_poll () at main-loop.c:213
#16 os_host_main_loop_wait (timeout=<optimized out>) at main-loop.c:258
#17 main_loop_wait (nonblocking=<optimized out>) at main-loop.c:506
#18 0x00007f4b9a0835cf in main_loop () at vl.c:1934
#19 main (argc=<optimized out>, argv=<optimized out>, envp=<optimized out>) at vl.c:4667
Do an upfront check for a NULL lsock and report an error to
the caller, which matches behaviour from before
Juergen Gross [Tue, 2 Aug 2016 06:32:32 +0000 (08:32 +0200)]
xen: use a common function for pv and hvm guest backend register calls
Instead of calling xen_be_register() for each supported backend type
for hvm and pv guests in their machine init functions use a common
function in order not to have to add new backends twice.
This at once fixes the error that hvm domains couldn't use the qusb
backend.
Peter Maydell [Wed, 3 Aug 2016 10:52:53 +0000 (11:52 +0100)]
Merge remote-tracking branch 'remotes/dgibson/tags/ppc-for-2.7-20160803' into staging
qemu-2.7: ppc patch queue 2016-08-03
Here's the current set of patches (only 2) for spapr, ppc and related
things. These are important bugfixes for the stabilizing 2.7 tree.
One is for a regression where confusion between x86 only and generic
KVM irq handling resulted in breakage on KVM/Power. The other is
fixing (yet another) problem in the vcpu hotplug code: older pseries
machine types which don't support vcpu hotplug weren't correctly
advertising that, potentially leading to crashes or other problems.
# gpg: Signature made Wed 03 Aug 2016 06:23:40 BST
# gpg: using RSA key 0x6C38CACA20D9B392
# gpg: Good signature from "David Gibson <[email protected]>"
# gpg: aka "David Gibson (Red Hat) <[email protected]>"
# gpg: aka "David Gibson (ozlabs.org) <[email protected]>"
# 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: 75F4 6586 AE61 A66C C44E 87DC 6C38 CACA 20D9 B392
* remotes/dgibson/tags/ppc-for-2.7-20160803:
kvm-irqchip: only commit route when irqchip is used
spapr: Error out when CPU hotplug is attempted on older pseries machines
The QTask struct is just a standalone struct, not a QOM Object,
so calling object_ref() on it is not appropriate. This results
in mangling the 'destroy' field in the QTask struct, causing
the later call to qtask_free() to try to call the function
at address 0x1, with predictably segfault happy results.
There is in fact no need for ref counting with QTask, as the
call to qtask_abort() or qtask_complete() will automatically
free associated memory.
Juergen Gross [Tue, 2 Aug 2016 12:14:04 +0000 (14:14 +0200)]
xen: drain submit queue in xen-usb before removing device
When unplugging a device in the Xen pvusb backend drain the submit
queue before deallocation of the control structures. Otherwise there
will be bogus memory accesses when I/O contracts are finished.
Correlated to this issue is the handling of cancel requests: a packet
cancelled will still lead to the call of complete, so add a flag
to the request indicating it should be just dropped on complete.
Juergen Gross [Tue, 2 Aug 2016 12:14:03 +0000 (14:14 +0200)]
xen: when removing a backend don't remove many of them
When a Xenstore watch fires indicating a backend has to be removed
don't remove all backends for that domain with the specified device
index, but just the one which has the correct type.
The easiest way to achieve this is to use the already determined
xendev as parameter for xen_be_del_xendev() instead of only the domid
and device index.
This at once removes the open coded QTAILQ_FOREACH_SAVE() in
xen_be_del_xendev() as there is no need to search for the correct
xendev any longer.
Peter Xu [Wed, 3 Aug 2016 03:07:21 +0000 (11:07 +0800)]
kvm-irqchip: only commit route when irqchip is used
Reported from Alexey Kardashevskiy:
3f1fea0fb5bf "kvm-irqchip: do explicit commit when update irq" produces
a crash on pseries guest running with VFIO on POWER8 machine as it does
not support KVM_CAP_IRQCHIP (KVM_CAP_IRQ_XICS is there instead). At the
result, KVMState::irq_routes is NULL when VFIO calls
kvm_irqchip_commit_routes.
Bharata B Rao [Wed, 3 Aug 2016 03:08:23 +0000 (13:08 +1000)]
spapr: Error out when CPU hotplug is attempted on older pseries machines
CPU hotplug and coldplug aren't supported prior to pseries-2.7. Further,
earlier machine types don't use CPU core objects at all. These mean that
query-hotpluggable-cpus and coldplug on older pseries machines will crash
QEMU. It also means that hotpluggable_cpus flag in query-machines will
be incorrectly set to true for pseries < 2.7, since it is based on the
presence of the query_hotpluggable_cpus hook.
- Don't assign the query_hotpluggable_cpus hook for pseries < 2.7
- query_hotpluggable_cpus should therefore never be called on pseries <
2.7, so add an assert
- spapr_core_pre_plug() should fail hot/cold plug attempts for pseries <
2.7, since core objects are never used there
- spapr_core_plug() should therefore never be called for pseries < 2.7, so
add an assert.
Signed-off-by: Bharata B Rao <[email protected]>
[dwg: Change from query_hotpluggable_cpus returning NULL for pseries < 2.7
to not being called at all, reword commit message for accuracy] Signed-off-by: David Gibson <[email protected]>
ehci: faster frame index calculation for skipped frames
ehci_update_frindex takes time linearly proportional to a number
of uframes to calculate new frame index and raise FLR interrupts,
which is a problem for large amounts of uframes.
If we experience large delays between echi timer callbacks (i.e. because
other periodic handlers have taken a lot of time to complete) we
get a lot of skipped frames which then delay ehci timer callback more
and this leads to deadlocking the system when ehci schedules next
callback to be too soon.
Observable behaviour is qemu consuming 100% host CPU time while guest
is unresponsive. This misbehavior could happen for a while and QEMU does
not get out from this state automatically without the patch.
This change makes ehci_update_frindex execute in constant time.
Eduardo Habkost [Mon, 1 Aug 2016 13:52:42 +0000 (10:52 -0300)]
MAINTAINERS: Add Host Memory Backends section
The hostmem code is closely related to the NUMA code, so I am
willing to handle patches to those files and share the work with
Igor (the original author of that code).
Dave Hansen [Sat, 23 Jul 2016 07:50:25 +0000 (09:50 +0200)]
target-i386: fix typo in xsetbv implementation
QEMU 2.6 added support for the XSAVE family of instructions, which
includes the XSETBV instruction which allows setting the XCR0
register.
But, when booting Linux kernels with XSAVE support enabled, I was
getting very early crashes where the instruction pointer was set
to 0x3. I tracked it down to a jump instruction generated by this:
gen_jmp_im(s->pc - pc_start);
where s->pc is pointing to the instruction after XSETBV and pc_start
is pointing _at_ XSETBV. Subtract the two and you get 0x3. Whoops.
The fix is to replace this typo with the pattern found everywhere
else in the file when folks want to end the translation buffer.
Richard Henderson confirmed that this is a bug and that this is the
correct fix.
qht: do not segfault when gathering stats from an uninitialized qht
So far, QHT functions assume that the passed qht has previously been
initialized--otherwise they segfault.
This patch makes an exception for qht_statistics_init, with the goal
of simplifying calling code. For instance, qht_statistics_init is
called from the 'info jit' dump, and given that under KVM the TB qht
is never initialized, we get a segfault. Thus, instead of complicating
the 'info jit' code with additional checks, let's allow passing an
uninitialized qht to qht_statistics_init.
While at it, add a test for this to test-qht.
Before the patch (for $ qemu -enable-kvm [...]):
(qemu) info jit
[...]
direct jump count 0 (0%) (2 jumps=0 0%)
Program received signal SIGSEGV, Segmentation fault.
After the patch the "TB hash buckets", "TB hash occupancy"
and "TB hash avg chain" lines are omitted.
(qemu) info jit
[...]
direct jump count 0 (0%) (2 jumps=0 0%)
TB hash buckets 0/0 (-nan% head buckets used)
TB hash occupancy nan% avg chain occ. Histogram: (null)
TB hash avg chain nan buckets. Histogram: (null)
[...]
Igor Mammedov [Wed, 20 Jul 2016 09:54:03 +0000 (11:54 +0200)]
fix qemu exit on memory hotplug when allocation fails at prealloc time
When adding hostmem backend at runtime, QEMU might exit with error:
"os_mem_prealloc: Insufficient free host memory pages available to allocate guest RAM"
It happens due to os_mem_prealloc() not handling errors gracefully.
Fix it by passing errp argument so that os_mem_prealloc() could
report error to callers and undo performed allocation when
os_mem_prealloc() fails.
Greg Kurz [Tue, 19 Jul 2016 08:28:35 +0000 (10:28 +0200)]
numa: set the memory backend "is_mapped" field
Commit 2aece63 "hostmem: detect host backend memory is being used properly"
added a way to know if a memory backend is busy or available for use. It
caused a slight regression if we pass the same backend to a NUMA node and
to a pc-dimm device:
Paolo Bonzini [Fri, 15 Jul 2016 17:58:41 +0000 (20:58 +0300)]
util/qht: Document memory ordering assumptions
It is naturally expected that some memory ordering should be provided
around qht_insert() and qht_lookup(). Document these assumptions in the
header file and put some comments in the source to denote how that
memory ordering requirements are fulfilled.
This avoids a segfault like the following for at least some 4.8 versions
of gcc when configured with --static if avx2 instructions are also
enabled:
Program received signal SIGSEGV, Segmentation fault.
buffer_find_nonzero_offset_ifunc () at ./util/cutils.c:333
333 {
(gdb) bt
#0 buffer_find_nonzero_offset_ifunc () at ./util/cutils.c:333
#1 0x0000000000939c58 in __libc_start_main ()
#2 0x0000000000419337 in _start ()
projects / qemu.git / log