There is now a trivial check on entry of if_start for pending packets,
so we can drop the additional tracking via if_queued.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Another attempt to get this right: We need to carefully walk both the
fastq and the batchq in if_start while trying to send packets to
possibly not yet resolved hosts on the virtual network.
So far we just requeued a delayed packet where it was and then started
walking the queues from the top again - that couldn't work. Now we pre-
calculate the next packet in the queue so that the current one can
safely be removed if it was sent successfully. We also need to take into
account that the next packet can be from the same session if the current
one was sent and there are no other sessions.
CC: Fabien Chouteau <chouteau@adacore.com>
CC: Zhi Yong Wu <wuzhy@linux.vnet.ibm.com>
CC: Stefan Weil <sw@weilnetz.de>
Tested-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
if_start can be called recursively via if_encap. Avoid this as our
scheme of dequeuing packets is not compatible with this.
CC: Fabien Chouteau <chouteau@adacore.com>
CC: Zhi Yong Wu <wuzhy@linux.vnet.ibm.com>
CC: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Make sure that next_m always points to a packet if batchq is non-empty.
This will simplify walking the queues in if_start.
CC: Fabien Chouteau <chouteau@adacore.com>
CC: Zhi Yong Wu <wuzhy@linux.vnet.ibm.com>
CC: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
* kwolf/for-anthony:
test-coroutine: add performance test for nesting
coroutine: adding configure option for sigaltstack coroutine backend
coroutine: adding configure choose mechanism for coroutine backend
coroutine: adding sigaltstack method (.c source)
qcow2: Reduce number of I/O requests
qcow2: Add qcow2_alloc_clusters_at()
qcow2: Factor out count_cow_clusters
qmp: convert blockdev-snapshot-sync to a wrapper around transactions
add mode field to blockdev-snapshot-sync transaction item
rename blockdev-group-snapshot-sync
qapi: complete implementation of unions
use QSIMPLEQ_FOREACH_SAFE when freeing list elements
Add 'make check-block'
make check: Add qemu-iotests subset
qemu-iotests: Mark some tests as quick
qcow2: Add error messages in qcow2_truncate
block: handle -EBUSY in bdrv_commit_all()
qcow2: Add some tracing
qed: do not evict in-use L2 table cache entries
Group snapshot: Fix format name for backing file
* qemu-kvm/memory/core:
memory: get rid of cpu_register_io_memory()
memory: dispatch directly via MemoryRegion
exec: fix code tlb entry misused as iotlb in get_page_addr_code()
memory: store section indices in iotlb instead of io indices
memory: make phys_page_find() return an unadjusted section
* stefanha/tracing:
vga: add trace event for ppm_save
console: add some trace events
maintainers: Add docs/tracing.txt to Tracing
docs: correct ./configure line in tracing.txt
trace: make trace_thread_create() use its function arg
tracetool: Omit useless QEMU_*_ENABLED() check
trace: Provide a per-event status define for conditional compilation
* qmp/queue/qmp:
qapi-schema.json: fix comment for type ObjectPropretyInfo
qapi-schema: fix typos and explain 'spice' auth
qjson.h: include compiler.h for GCC_FMT_ATTR
guest-sync leaves it as an exercise to the user as to how to reliably
obtain the response to guest-sync if the client had previously read in a
partial response (due qemu-ga previously being restarted mid-"sentence"
due to reboot, forced restart, etc).
qemu-ga handles this situation on its end by having a client precede
their guest-sync request with a 0xFF byte (invalid UTF-8), which
qemu-ga/QEMU JSON parsers will treat as a flush event. Thus we can
reliably flush the qemu-ga parser state in preparation for receiving
the guest-sync request.
guest-sync-delimited provides the same functionality for a client: when
a guest-sync-delimited is issued, qemu-ga will precede it's response
with a 0xFF byte that the client can use as an indicator to flush its
buffer/parser state in preparation for reliably receiving the
guest-sync-delimited response.
It is also useful as an optimization for clients, since, after issuing a
guest-sync-delimited, clients can safely discard all stale data read
from the channel until the 0xFF is found.
More information available on the wiki:
http://wiki.qemu.org/Features/QAPI/GuestAgent#QEMU_Guest_Agent_Protocol
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
This command returns an array of:
[ifname, hwaddr, [ipaddr, ipaddr_family, prefix] ]
for each interface in the system.
Currently, only IPv4 and IPv6 are supported.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Implement guest-suspend-disk RPC for Windows. Functionally this should be
equivalent to the posix implementation.
Signed-off-by: Gal Hammer <ghammer@redhat.com>
As the command name implies, this command suspends the guest to disk.
The suspend operation is implemented by two functions: bios_supports_mode()
and guest_suspend(). Both functions are generic enough to be used by
other suspend modes (introduced by next commits).
Both functions will try to use the scripts provided by the pm-utils
package if it's available. If it's not available, a manual method,
which consists of directly writing to '/sys/power/state', will be used.
To reap terminated children, a new signal handler is installed in the
parent to catch SIGCHLD signals and a non-blocking call to waitpid()
is done to collect their exit statuses. The statuses, however, are
discarded.
The approach used to query the guest for suspend support deserves some
explanation. It's implemented by bios_supports_mode() and shown below:
qemu-ga
|
create pipe
|
fork()
-----------------
| |
| |
| fork()
| --------------------------
| | |
| | |
| | exec('pm-is-supported')
| |
| wait()
| write exit status to pipe
| exit
|
read pipe
This might look complex, but the resulting code is quite simple.
The purpose of that approach is to allow qemu-ga to reap its children
(semi-)automatically from its SIGCHLD handler.
Implementing this the obvious way, that's, doing the exec() call from
the first child process, would force us to introduce a more complex way
to reap qemu-ga's children. Like registering PIDs to be reaped and
having a way to wait for them when returning their exit status to
qemu-ga is necessary. The approach explained above avoids that complexity.
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Cast the argument of the g2h() macro to a target_ulong so that
it isn't accidentally sign-extended if it is a signed 32 bit
type and long is a 64 bit type. In particular, this fixes a
bug where it would return the wrong value for 32 bit guests
on 64 bit hosts when passed in one of the arg* values from
do_syscall() [which are all abi_long and thus signed types].
This could result in spurious failure of mlock(), among others.
Reviewed-by: Andreas F=E4rber <afaerber@suse.de>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
The comment is wrong, we have to do something in the setdata callback.
Changing the framebuffer backing storage (happens when the guest pans
the display) renders the whole screen content invalid.
Trigger #1: cirrus vga + 32bit linux guest + vesafb with ypan enabled.
Trigger #2: std vga + http://patchwork.ozlabs.org/patch/145479/
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
To properly load cpudefs using -readconfig, we have to call
cpudef_init() after finishing the command-line option handling.
Consequently, the handling of "-cpu ?" has to be done after the
command-line option handling loop, too.
Without this patch, "-readconfig configfile -cpu ?" fails to list the
CPU definitions read from 'configfile'.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
This patch addes a Bulldozer-based Opteron_G4 CPU model.
This version has the ffxsr bit actually disabled, to match what was
documented below. Thanks to Andre Przywara for spotting the bug.
I am trying to be conservative with the new model, so I am enabling only
features known to be useful to guests, and not enabling anything that
was not tested or found to be useful to a guest.
List of missing flags in comparison to real hardware:
- vme: host-specific feature.
- osxsave: it is not set here because it is set by the guest OS, not by KVM
- monitor: this is filtered out by the KVM module, so no point in
enabling it.
- mmxext: untested, so not enabled.
- Perf*, Topology*, lwp, ibs: not emulated by KVM.
- wdt, skinit, osvw, altmovcr8, extapicspace, cmplegacy: untested,
so not enabled.
List of new flags, in comparison to the Opteron_G3 model:
- xsave: xsave feature, already implemented by Qemu
- avx, aes, sse4.x, ssse3, pclmulqdq: all new state the new instructions
could use is handled by the xsave state loading/saving code on Qemu.
- pdpe1gb: 1GB pages, supported by the KVM kernel module.
- ffxsr: untested, so not enabled
- fma4, xop: all new state the new instructions could use is handled by
the xsave loading/saving code on Qemu.
- 3dnowprefetch: safe to pass through, though the flag is not used by
Linux guests, at least.
Below is the comparison between the current Opteron_G3 model
and the new model being added.
- The "full" line contains the flags found on actual hardware.
- The "missing" line shows the flags that are present on actual
hardware, but not on the added Opteron_G4 model.
- The "new" line shows the flags that were not on the Opteron_G3 model
but are on Opteron_G4.
feature_edx:
Opteron_G3: sse2 sse fxsr mmx clflush pse36 pat cmov mca pge mtrr sep apic cx8 mce pae msr tsc pse de fpu
full: sse2 sse fxsr mmx clflush pse36 pat cmov mca pge mtrr sep apic cx8 mce pae msr tsc pse de vme fpu
Opteron_G4: sse2 sse fxsr mmx clflush pse36 pat cmov mca pge mtrr sep apic cx8 mce pae msr tsc pse de fpu
missing: vme
feature_ecx:
Opteron_G3: popcnt cx16 monitor sse3
full: avx osxsave xsave aes popcnt sse4.2 sse4.1 cx16 ssse3 monitor pclmulqdq sse3
Opteron_G4: avx xsave aes popcnt sse4.2 sse4.1 cx16 ssse3 pclmulqdq sse3
missing: osxsave monitor
new: avx xsave aes sse4.2 sse4.1 ssse3 pclmulqdq
extfeature_edx:
Opteron_G3: lm rdtscp fxsr mmx nx pse36 pat cmov mca pge mtrr syscall apic cx8 mce pae msr tsc pse de fpu
full: lm rdtscp pdpe1gb ffxsr fxsr mmx mmxext nx pse36 pat cmov mca pge mtrr syscall apic cx8 mce pae msr tsc pse de vme fpu
Opteron_G4: lm rdtscp pdpe1gb fxsr mmx nx pse36 pat cmov mca pge mtrr syscall apic cx8 mce pae msr tsc pse de fpu
missing: mmxext vme
new: pdpe1gb
extfeature_ecx:
Opteron_G3: misalignsse sse4a abm svm lahf_lm
full: Perf* Topology* fma4 lwp wdt skinit xop ibs osvw 3dnowprefetch misalignsse sse4a abm altmovcr8 extapicspace svm cmplegacy lahf_lm
Opteron_G4: fma4 xop 3dnowprefetch misalignsse sse4a abm svm lahf_lm
new: fma4 xop 3dnowprefetch
missing: Perf* Topology* lwp wdt skinit ibs osvw altmovcr8 extapicspace cmplegacy
Changes v1 -> v2:
- Actually disable ffxsr bit
Cc: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
This patches add the definition of a SandyBridge CPU model.
Summary of differences:
Flags present on actual hardware, but not on the added model definition:
- pbe, tm, ht, ss, acpi, vme, xTPR, tm2, eist, smx: host-specific
features, not exposed to guest.
- ds, ds-cpl, dtes64, pdcm: emulation not supported by KVM (although it
may be added in the future if implementing PMU virtualization)
- pcid, vmx, monitor: not emulated by Qemu/KVM right now.
- osxsave: set by the guest OS, not by Qemu.
Flags added, that were not present on Westmere model:
- xsave: already supported by Qemu
- avx, pclmulqdq: all new state the new instructions could use is
handled by xsave state loading/saving code.
- tsc-deadline, x2apic, rdtscp: already supported by Qemu/KVM.
Below there's a comparison of the features on the current Westmere CPU
model, and the SandyBridge CPU model.
- The "full" line contains the flags found on actual hardware.
- The "missing" line shows the flags that are present on actual
hardware, but not on the added SandyBridge model.
- The "new" line shows the flags that were not on the Westmere model,
but are on SandyBridge.
feature_edx:
Westmere: sse2 sse fxsr mmx clflush pse36 pat cmov mca pge mtrr sep apic cx8 mce pae msr tsc pse de fpu
full: pbe tm ht ss sse2 sse fxsr mmx ds acpi clflush pse36 pat cmov mca pge mtrr sep apic cx8 mce pge msr tsc pse de vme fpu
SandyBridge: sse2 sse fxsr mmx clflush pse36 pat cmov mca pge mtrr sep apic cx8 mce pae msr tsc pse de fpu
missing: pbe tm ht ss ds acpi vme
feature_ecx:
Westmere: aes popcnt sse4.2 sse4.1 cx16 ssse3 sse3
full: avx osxsave xsave aes tsc-deadline popcnt x2apic sse4.2 sse4.1 pcid pdcm xTPR cx16 ssse3 tm2 eist smx vmx ds-cpl monitor dtes64 pclmulqdq sse3
SandyBridge: avx xsave aes tsc-deadline popcnt x2apic sse4.2 sse4.1 cx16 ssse3 pclmulqdq sse3
missing: osxsave pcid pdcm xTPR tm2 eist smx vmx ds-cpl monitor dtes64
new: avx xsave tsc-deadline x2apic pclmulqdq
extfeature_edx:
Westmere: i64 nx syscall
full: i64 rdtscp nx syscall
SandyBridge: i64 rdtscp nx syscall
new: rdtscp
extfeature_ecx:
Westmere: lahf_lm
full: lahf_lm
SandyBridge: lahf_lm
Cc: "Dugger, Donald D" <donald.d.dugger@intel.com>
Cc: "Zhang, Xiantao" <xiantao.zhang@intel.com>
Acked-by: Xiantao Zhang <xiantao.zhang@intel.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Property removal modifies the list, so it is not safe to continue
iteration. We know anyway that each object can have only one
parent (see object_property_add_child), so exit after finding
the requested object.
Reported-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
These were stored as NULL due to wrong cut-and-paste from set_pointer.
Reported-by: Gerhard Wiesinger <lists@wiesinger.com>
Tested-by: Gerhard Wiesinger <lists@wiesinger.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
The performance test will also check for nesting. It will do
a certain quantity of cycles, and each of one will do a depth
nesting process.
This is useful for benchmarking the creation of coroutines,
given that nesting is creation-intensive (and the other perf
test does not benchmark that).
Signed-off-by: Alex Barcelo <abarcelo@ac.upc.edu>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
It's possible to use sigaltstack backend with --with-coroutine=sigaltstack
v2: changed from enable/disable configure flags
Signed-off-by: Alex Barcelo <abarcelo@ac.upc.edu>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Configure tries, as a default, ucontext functions for the
coroutines. But now the user can force another backend by
--with-coroutine=BACKEND option
v2: Using --with-coroutine=BACKEND instead of enable
disable individual configure options
Signed-off-by: Alex Barcelo <abarcelo@ac.upc.edu>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This file is based in both coroutine-ucontext.c and
pth_mctx.c (from the GNU Portable Threads library).
The mechanism used to change stacks is the sigaltstack
function (variant 2 of the pth library).
v2: Some corrections. Moving global variables into
thread storage (CoroutineThreadState).
Signed-off-by: Alex Barcelo <abarcelo@ac.upc.edu>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If the first part of a write request is allocated, but the second isn't
and it can be allocated so that the resulting area is contiguous, handle
it at once. This is a common case for sequential writes.
After this patch, alloc_cluster_offset() only checks if the clusters are
already allocated or how many new clusters can be allocated contigouosly.
The actual cluster allocation is split off into a new function
do_alloc_cluster_offset().
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
This function allows to allocate clusters at a given offset in the image
file. This is useful if you want to allocate the second part of an area
that must be contiguous.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Simplify the blockdev-snapshot-sync code and gain failsafe operation
by turning it into a wrapper around the new transaction command. A new
option is also added matching "mode".
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The mode field lets a management application create the snapshot
destination outside QEMU.
Right now, the only modes are "existing" and "absolute-paths". Mirroring
introduces "no-backing-file". In the future "relative-paths" could be
implemented too.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We will add other kinds of operation. Prepare for this by adjusting
the schema.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This creates a new test group 'quick' for some test case that take at
most a couple of seconds each, so that the group can be run during a
quick 'make check'
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
qemu-img resize has some limitations with qcow2, but the user is only
told that "this image format does not support resize". Quite confusing,
so add some more detailed error_report() calls and change "this image
format" into "this image".
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Monitor operations that manipulate image files must not execute while a
background job (like image streaming) is in progress. This prevents
corruptions from happening when two pieces of code are manipulating the
image file without knowledge of each other.
The monitor "commit" command raises QERR_DEVICE_IN_USE when
bdrv_commit() returns -EBUSY but "commit all" has no error handling.
This is easy to fix, although note that we do not deliver a detailed
error about which device was busy in the "commit all" case.
Suggested-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The L2 table cache reduces QED metadata reads that would be required
when translating LBAs to offsets into the image file. Since requests
execute in parallel it is possible to share an L2 table between multiple
requests.
There is a potential data corruption issue when an in-use L2 table is
evicted from the cache because the following situation occurs:
1. An allocating write performs an update to L2 table "A".
2. Another request needs L2 table "B" and causes table "A" to be
evicted.
3. A new read request needs L2 table "A" but it is not cached.
As a result the L2 update from #1 can overlap with the L2 fetch from #3.
We must avoid doing overlapping I/O requests here since the worst case
outcome is that the L2 fetch completes before the L2 update and yields
stale data. In that case we would effectively discard the L2 update and
lose data clusters!
Thanks to Benoît Canet <benoit.canet@gmail.com> for extensive testing
and debugging which lead to discovery of this bug.
Reported-by: Benoît Canet <benoit.canet@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Tested-by: Benoît Canet <benoit.canet@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The topic of whether and by whom docs/tracing.txt is maintained was
brought up. It currently does not have an official maintainer.
Add it to the tracing section so that Stefan gets cc'ed on patches.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Acked-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Cc: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
This patch corrects the configure's trace option in docs/tracing.txt.
Signed-off-by: Jun Koi <junkoi2004@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
This patch makes trace_thread_create() to use its function arg to
initialize thread. The other choice is to make this a function to use
void arg, but i prefer this way.
Signed-off-by: Jun Koi <junkoi2004@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
SystemTap provides a "semaphore" that can optionally be tested before
executing a trace event. The purpose of this mechanism is to skip
expensive tracing code when the trace event is disabled.
For example, some applications may have trace events that format or
convert strings for trace events. This expensive processing should only
be done in the case where the trace event is enabled.
Since QEMU's generated trace events never have such special-purpose
code, there is no reason to add the semaphore check.
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
