We abort a few lines above if kernel_xics_fd == -1.
This is only code cleanup.
Signed-off-by: Greg Kurz <groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The 9P spec at http://man.cat-v.org/plan_9/5/intro says:
All directories must support walks to the directory .. (dot-dot) meaning
parent directory, although by convention directories contain no explicit
entry for .. or . (dot). The parent of the root directory of a server's
tree is itself.
This means that a client cannot walk further than the root directory
exported by the server. In other words, if the client wants to walk
"/.." or "/foo/../..", the server should answer like the request was
to walk "/".
This patch just does that:
- we cache the QID of the root directory at attach time
- during the walk we compare the QID of each path component with the root
QID to detect if we're in a "/.." situation
- if so, we skip the current component and go to the next one
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
According to the 9P spec http://man.cat-v.org/plan_9/5/open about the
create request:
The names . and .. are special; it is illegal to create files with these
names.
This patch causes the create and lcreate requests to fail with EINVAL if
the file name is either "." or "..".
Even if it isn't explicitly written in the spec, this patch extends the
checking to all requests that may cause a directory entry to be created:
- mknod
- rename
- renameat
- mkdir
- link
- symlink
The unlinkat request also gets patched for consistency (even if
rmdir("foo/..") is expected to fail according to POSIX.1-2001).
The various error values come from the linux manual pages.
Suggested-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Empty path components don't make sense for most commands and may cause
undefined behavior, depending on the backend.
Also, the walk request described in the 9P spec [1] clearly shows that
the client is supposed to send individual path components: the official
linux client never sends portions of path containing the / character for
example.
Moreover, the 9P spec [2] also states that a system can decide to restrict
the set of supported characters used in path components, with an explicit
mention "to remove slashes from name components".
This patch introduces a new name_is_illegal() helper that checks the
names sent by the client are not empty and don't contain unwanted chars.
Since 9pfs is only supported on linux hosts, only the / character is
checked at the moment. When support for other hosts (AKA. win32) is added,
other chars may need to be blacklisted as well.
If a client sends an illegal path component, the request will fail and
ENOENT is returned to the client.
[1] http://man.cat-v.org/plan_9/5/walk
[2] http://man.cat-v.org/plan_9/5/intro
Suggested-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Considering that features are converted to global properties and
global properties are automatically applied to every new instance
of created CPU (at object_new() time), there is no point in
parsing cpu_model string every time a CPU created. So move
parsing outside CPU creation loop and do it only once.
Parsing also should be done before any CPU is created so that
features would affect the first CPU a well.
This patch does that for all PowerPC machine types.
It is based on previous work from Bharata:
https://lists.nongnu.org/archive/html/qemu-devel/2016-06/msg07564.html
Signed-off-by: Greg Kurz <groug@kaod.org>
[clg: only kept the fix for the spapr platform. support for other
platform will be added in 2.8 ]
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Tested-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
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:
-m 1G,slots=2,maxmem=2G \
-object memory-backend-ram,size=1G,id=mem-mem1 \
-device pc-dimm,id=dimm-mem1,memdev=mem-mem1 \
-numa node,nodeid=0,memdev=mem-mem1
Before commit 2aece63, this would cause QEMU to print an error message and
to exit gracefully:
qemu-system-ppc64: -device pc-dimm,id=dimm-mem1,memdev=mem-mem1:
can't use already busy memdev: mem-mem1
Since commit 2aece63, QEMU hits an assertion in the memory code:
qemu-system-ppc64: memory.c:1934: memory_region_add_subregion_common:
Assertion `!subregion->container' failed.
Aborted
This happens because pc-dimm devices don't use memory_region_is_mapped()
anymore and cannot guess the backend is already used by a NUMA node.
Let's revert to the previous behavior by turning the NUMA code to also
call host_memory_backend_set_mapped() when it uses a backend.
Fixes: 2aece63c8a
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <146891691503.15642.9817215371777203794.stgit@bahia.lan>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When passing '-global driver=host-powerpc64-cpu,property=compat,value=foo'
on the command line, without this patch, we get the following warning per
device (which means many lines if the guests has many cpus):
qemu-system-ppc64: Warning: can't apply global host-powerpc64-cpu.compat=foo:
Invalid compatibility mode "foo"
... and QEMU continues execution, ignoring the property.
With this patch, we get a single line:
qemu-system-ppc64: can't apply global host-powerpc64-cpu.compat=foo:
Invalid compatibility mode "foo"
... and QEMU exits.
The previous behavior is kept for hotplugged devices since we don't want
QEMU to exit when doing device_add.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
This patch ensures QEMU won't terminate while hotplugging a device if the
global property cannot be set and errp points to error_fatal or error_abort.
While here, it also fixes indentation of the typename argument.
Suggested-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
We're not supposed to abort when the user passes a bogus value.
Since the checking is done in visit_type_OnOffSplit(), the call
to abort() is legitimate. Let's add a comment to make it
explicit.
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
The goal of this patch is to have a stable core-id which does not depend
on any DT related semantics, which involve non-obvious computations on
modern PowerPC server cpus.
With this patch, the DT core id is computed on-demand as:
(core-id / smp_threads) * smt
where smt is the number of threads per core in the host.
This formula should be consolidated in a helper since it is needed in
several places.
Other uses for core-id includes: compute a stable cpu_index (which
allows random order hotplug/unplug without breaking migration) and
NUMA.
Signed-off-by: Greg Kurz <groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
It is not possible to set the compat property to an unknown value with
powerpc_set_compat(). Something must have gone terribly wrong in QEMU,
if we detect an "Internal error" in powerpc_get_compat(). Let's abort then.
This patch also drops the "max_compat ? *max_compat : -1" construct. It is
useless since max_compat is dereferenced a few lines above.
Signed-off-by: Greg Kurz <groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
If the host has 8 threads/core and the guest is started with:
-smp cores=1,threads=4,maxcpus=12
It is possible to crash QEMU by doing:
(qemu) device_add host-spapr-cpu-core,core-id=16,id=foo
(qemu) device_del foo
Segmentation fault
This happens because spapr_core_unplug() assumes cpu_dt_id == core_id.
As long as cpu_dt_id is derived from the non-table cpu_index, this is
only true when you plug cores with contiguous ids.
It is safer to be consistent: the DR connector was created with an
index that is immediately written to cc->core_id, and spapr_core_plug()
also relies on cc->core_id.
Let's use it also in spapr_core_unplug().
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The <sys/uio.h> system header doesn't exist on all host platforms. Code
should include "qemu/osdep.h" instead to avoid build breaks on plafforms
that don't define CONFIG_IOVEC (like win32, if it is to support 9p one day).
Acked-by: Cédric Le Goater <clg@kaod.org>
Acked-by: Michael Fritscher <michael@fritscher.net>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
This patch changes spapr_cpu_core_realize_child() to have a local error
pointer and use error_propagate() as it is supposed to be done.
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
When a core is being realized, we create a child object for each thread
of the core.
The child is first initialized with object_initialize() which sets its ref
count to 1, and then added to the core with object_property_add_child()
which bumps the ref count to 2.
When the core gets released, object_unparent() decreases the ref count to 1,
and we g_free() the object: we hence loose the reference on an unfinalized
object. This is likely to cause random crashes.
Let's drop the extra reference as soon as we don't need it, after the
thread is added to the core.
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
As stated in linux/Documentation/virtual/kvm/api.txt:
The maximum possible value for max_vcpu_id can be retrieved using the
KVM_CAP_MAX_VCPU_ID of the KVM_CHECK_EXTENSION ioctl() at run-time.
If the KVM_CAP_MAX_VCPU_ID does not exist, you should assume that
max_vcpu_id is the same as the value returned from KVM_CAP_MAX_VCPUS.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Message-Id: <146424974323.5666.5471538288045048119.stgit@bahia.huguette.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Paolo's recent cpu.h cleanups broke legacy virtio for ppc64 LE guests (and
arm BE guests as well, even if I have not verified that). Especially, commit
"33c11879fd42 qemu-common: push cpu.h inclusion out of qemu-common.h" has
the side-effect of silently hiding the TARGET_IS_BIENDIAN macro from the
virtio memory accessors, and thus fully disabling support of endian changing
targets.
To be sure this cannot happen again, let's gather all the bi-endian bits
where they belong in include/hw/virtio/virtio-access.h.
The changes in hw/virtio/vhost.c are safe because vhost_needs_vring_endian()
is not called on a hot path and non bi-endian targets will return false
anyway.
While here, also rename TARGET_IS_BIENDIAN to be more precise: it is only for
legacy virtio and bi-endian guests.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
This patch changes the 9p code to use readdir() again instead of
readdir_r(), which is deprecated in glibc 2.24.
All the locking was put in place by a previous patch.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
If several threads concurrently call readdir() with the same directory
stream pointer, it is possible that they all get a pointer to the same
dirent structure, whose content is overwritten each time readdir() is
called.
We must thus serialize accesses to the dirent structure.
This may be achieved with a mutex like below:
lock_mutex();
readdir();
// work with the dirent
unlock_mutex();
This patch adds all the locking, to prepare the switch to readdir().
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
If we are to switch back to readdir(), we need a more complex type than
DIR * to be able to serialize concurrent accesses to the directory stream.
This patch introduces a placeholder type and fixes all users.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Most of the 9p code is now virtio agnostic. This patch does a final cleanup:
- drop references to Virtio from the header comments
- fix includes
Also drop a couple of leading empty lines while here.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
The "9p-attr.h" header isn't needed by 9p synth and virtio 9p.
While here, also drop last references to virtio from 9p synth since it is
now transport agnostic code.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
The KVM API restricts vcpu ids to be < KVM_CAP_MAX_VCPUS. On PowerPC
targets, depending on the number of threads per core in the host and
in the guest, some topologies do generate higher vcpu ids actually.
When this happens, QEMU bails out with the following error:
kvm_init_vcpu failed: Invalid argument
The KVM_CREATE_VCPU ioctl has several EINVAL return paths, so it is
not possible to fully disambiguate.
This patch adds a check in the code that computes vcpu ids, so that
we can detect the error earlier, and print a friendlier message instead
of calling KVM_CREATE_VCPU with an obviously bogus vcpu id.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
We currently have an error path during migration that can cause
the source QEMU to abort:
migration_thread()
migration_completion()
runstate_is_running() ----------------> true if guest is running
bdrv_inactivate_all() ----------------> inactivate images
qemu_savevm_state_complete_precopy()
... qemu_fflush()
socket_writev_buffer() --------> error because destination fails
qemu_fflush() -------------------> set error on migration stream
migration_completion() -----------------> set migrate state to FAILED
migration_thread() -----------------------> break migration loop
vm_start() -----------------------------> restart guest with inactive
images
and you get:
qemu-system-ppc64: socket_writev_buffer: Got err=104 for (32768/18446744073709551615)
qemu-system-ppc64: /home/greg/Work/qemu/qemu-master/block/io.c:1342:bdrv_co_do_pwritev: Assertion `!(bs->open_flags & 0x0800)' failed.
Aborted (core dumped)
If we try postcopy with a similar scenario, we also get the writev error
message but QEMU leaves the guest paused because entered_postcopy is true.
We could possibly do the same with precopy and leave the guest paused.
But since the historical default for migration errors is to restart the
source, this patch adds a call to bdrv_invalidate_cache_all() instead.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Message-Id: <146357896785.6003.11983081732454362715.stgit@bahia.huguette.org>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
QEMU has currently two ways to prevent migration to occur:
- migration blocker when it depends on runtime state
- VMStateDescription.unmigratable when migration is not supported at all
This patch gathers all the logic into a single function to be called from
both the savevm and the migrate paths.
This fixes a bug with 9p, at least, where savevm would succeed and the
following would happen in the guest after loadvm:
$ ls /host
ls: cannot access /host: Protocol error
With this patch:
(qemu) savevm foo
Migration is disabled when VirtFS export path '/' is mounted in the guest
using mount_tag 'host'
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <146239057139.11271.9011797645454781543.stgit@bahia.huguette.org>
[Update subject according to Paolo's suggestion - Amit]
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Commit "ebac1202c95a virtio-9p: use QEMU thread pool" dropped function
v9fs_init_worker_threads.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Since commit "60253ed1e6ec rng: add request queue support to rng-random",
the use of a spapr_rng device may hang vCPU threads.
The following path is taken without holding the lock to the main loop mutex:
h_random()
rng_backend_request_entropy()
rng_random_request_entropy()
qemu_set_fd_handler()
The consequence is that entropy_available() may be called before the vCPU
thread could even queue the request: depending on the scheduling, it may
happen that entropy_available() does not call random_recv()->qemu_sem_post().
The vCPU thread will then sleep forever in h_random()->qemu_sem_wait().
This could not happen before 60253ed1e6 because entropy_available() used
to call random_recv() unconditionally.
This patch ensures the lock is held to avoid the race.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Cédric Le Goater <clg@fr.ibm.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Also fix a typo in the virtio_balloon_handle_output() trace while here.
[The double-quoting was a limitation of the old tracetool.sh script.
The modern tracetool.py script does not require double-quotes at the end
of the line. See commit cf85cf8e97
("trace: Format strings must begin/end with double quotes").
--Stefan]
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Message-id: 20160111173036.24764.59878.stgit@bahia.huguette.org
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Using the return value to report errors is error prone:
- xics_alloc() returns -1 on error but spapr_vio_busdev_realize() errors
on 0
- xics_alloc_block() returns the unclear value of ics->offset - 1 on error
but both rtas_ibm_change_msi() and spapr_phb_realize() error on 0
This patch adds an errp argument to xics_alloc() and xics_alloc_block() to
report errors. The return value of these functions is a valid IRQ number
if errp is NULL. It is undefined otherwise.
The corresponding error traces get promotted to error messages. Note that
the "can't allocate IRQ" error message in spapr_vio_busdev_realize() also
moves to xics_alloc(). Similar error message consolidation isn't really
applicable to xics_alloc_block() because callers have extra context (device
config address, MSI or MSIX).
This fixes the issues mentioned above.
Based on previous work from Brian W. Hart.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Migration of pseries-2.3 doesn't have configuration section. Unfortunately,
QEMU 2.4/2.4.1/2.5 are buggy and always stream and expect the configuration
section, and break migration both ways.
This patch introduces a property which allows to enforce a configuration
section for machines who don't have one.
It can be set at startup:
-machine enforce-config-section=on
or later from the QEMU monitor:
qom-set /machine enforce-config-section on
It is up to the tooling to set or unset this property according to the
version of the QEMU at the other end of the pipe.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Since QEMU 2.4, we have a configuration section in the migration stream.
This must be skipped for older machines, like it is already done for x86.
This patch fixes the migration of pseries-2.3 from/to QEMU 2.3, but it
breaks migration of the same machine from/to QEMU 2.4/2.4.1/2.5. We do
that anyway because QEMU 2.3 is likely to be more widely deployed than
newer QEMU versions.
Fixes: 61964c23e5
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Since QEMU 2.3, we have a vmdesc section in the migration stream.
This section is not mandatory but when migrating a pseries-2.2
machine from QEMU 2.2, you get a warning at the destination:
qemu-system-ppc64: Expected vmdescription section, but got 0
The warning goes away if we decide to skip vmdesc as well for
older pseries, like it is already done for pc's.
This can only be observed with -cpu POWER7 because POWER8
cannot migrate from QEMU 2.2 to 2.3 (insns_flags2 mismatch).
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This RTAS call is used to request new interrupts or to free all interrupts.
If the driver has already allocated interrupts and asks again for a non-null
number of irqs, then the rtas_ibm_change_msi() function will silently leak
the previous interrupts.
It happens because xics_free() is only called when the driver releases all
interrupts (!req_num case). Note that the previously allocated spapr_pci_msi
is not leaked because the GHashTable is created with destroy functions and
g_hash_table_insert() hence frees the old value.
This patch makes sure any previously allocated MSIs are released when a
new allocation succeeds.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
It is currently possible to hotplug a spapr_rng device but QEMU crashes
when we try to hot unplug:
ERROR:hw/core/qdev.c:295:qdev_unplug: assertion failed: (hotplug_ctrl)
Aborted
This happens because spapr_rng isn't plugged to any bus and sPAPR does
not provide hotplug support for it: qdev_get_hotplug_handler() hence
return NULL and we hit the assertion.
And anyway, it doesn't make much sense to unplug this device since hcalls
cannot be unregistered. Even the idea of hotplugging a RNG device instead
of declaring it on the QEMU command line looks weird.
This patch simply disables hotpluggability for the spapr-rng class.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
When adding cross-endian support, we introduced the TARGET_IS_BIENDIAN macro
and the virtio_access_is_big_endian() helper to have a branchless fast path
in the virtio memory accessors for targets that don't switch endian.
This was considered as a strong requirement at the time.
Now we have added a runtime check for virtio 1.0, which ruins the benefit
of the virtio_access_is_big_endian() helper for always little-endian targets.
With this patch, always little-endian targets stop checking for virtio 1.0,
since the result is little-endian in all cases.
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
After the call to virtio_vdev_has_feature(), we only care for legacy
devices, so we don't need the extra check in virtio_is_big_endian().
Also the device_endian field is always set (VIRTIO_DEVICE_ENDIAN_UNKNOWN
may only happen on a virtio_load() path that cannot lead here), so we
don't need the assert() either.
This open codes the device_endian checking in vhost_needs_vring_endian().
It also adds a comment to explain the logic, as recent reviews showed the
cross-endian tweaks aren't that obvious.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Indeed vhost doesn't need to ask for vring endian fixing if the device is
virtio 1.0, since it is already handled by the in-kernel vhost driver. This
patch simply consolidates the logic into the existing helper.
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
If target is bi-endian (ppc64, arm), the virtio_legacy_is_cross_endian()
indeed returns the runtime state of the virtio device. However, it returns
false unconditionally in the general case. This sounds a bit strange
given the name of the function.
This helper is only useful for vhost actually, where indeed non bi-endian
targets don't have to deal with cross-endian issues.
This patch moves the helper to vhost.c and gives it a more appropriate name.
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Cross-endian is now handled by the core virtio-net code.
This patch reverts:
commit 5be7d9f1b1
vhost-net: tell tap backend about the vnet endianness
and
commit cf0a628f6e81bfc9b7a944fa0b80c3594836df56
net: set endianness on all backend devices
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
When running a fully emulated device in cross-endian conditions, including
a virtio 1.0 device offered to a big endian guest, we need to fix the vnet
headers. This is currently handled by the virtio_net_hdr_swap() function
in the core virtio-net code but it should actually be handled by the net
backend.
With this patch, virtio-net now tries to configure the backend to do the
endian fixing when the device starts (i.e. drivers sets the CONFIG_OK bit).
If the backend cannot support the requested endiannes, we have to fallback
onto virtio_net_hdr_swap(): this is recorded in the needs_vnet_hdr_swap flag,
to be used in the TX and RX paths.
Note that we reset the backend to the default behaviour (guest native
endianness) when the device stops (i.e. device status had CONFIG_OK bit and
driver unsets it). This is needed, with the linux tap backend at least,
otherwise the guest may lose network connectivity if rebooted into a
different endianness.
The current vhost-net code also tries to configure net backends. This will
be no more needed and will be reverted in a subsequent patch.
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Let's reuse the ppc_maybe_bswap_register() helper, like we already do
with the general registers.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Altivec registers are 128-bit wide. They are stored in memory as two
64-bit values that must be byteswapped when the guest is little-endian.
Let's reuse the ppc_maybe_bswap_register() helper for this.
We also need to fix the ordering of the 64-bit elements according to
the target endianness, for both system and user mode.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This helper will be used to support Altivec registers in little-endian guests.
This patch does not change functionnality.
Note: I had to put the helper some lines away from the gdb_*_avr_reg()
routines to get a more readable patch.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Let's reuse the ppc_maybe_bswap_register() helper, like we already do
with the general registers.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This helper will be used to support FP, Altivec and VSX registers when
the guest is little-endian.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
On VSX capable CPUs, the 32 FP registers are mapped to the high-bits
of the 32 first VSX registers. So if you have:
VSR31 = (uint128) 0x0102030405060708090a0b0c0d0e0f00
then
FPR31 = (uint64) 0x0102030405060708
The kernel stores the VSX registers in the fp_state struct following the
host endian element ordering.
On big-endian:
fp_state.fpr[31][0] = 0x0102030405060708
fp_state.fpr[31][1] = 0x090a0b0c0d0e0f00
On little-endian:
fp_state.fpr[31][0] = 0x090a0b0c0d0e0f00
fp_state.fpr[31][1] = 0x0102030405060708
The KVM_GET_ONE_REG and KVM_SET_ONE_REG ioctls preserve this ordering, but
QEMU considers it as big-endian and always copies element [0] to the
fpr[] array and element [1] to the vsr[] array. This does not work with
little-endian hosts, and you will get:
(qemu) p $f31
0x90a0b0c0d0e0f00
instead of:
(qemu) p $f31
0x102030405060708
This patch fixes the element ordering for little-endian hosts.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The aio_context_new() function does not allocate a thread pool. This is
deferred to the first call to the aio_get_thread_pool() accessor. It is
hence forbidden to access the thread_pool field directly, as it may be
NULL. The accessor *must* be used always.
Fixes: ebac1202c9
Reviewed-by: Michael Tokarev <mjt@tls.msk.ru>
Tested-by: Michael Tokarev <mjt@tls.msk.ru>
Cc: qemu-stable@nongnu.org
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Since commit 4652f1640e "virtio-9p: add savevm
handlers", if the user hot-unplugs a quiescent 9p device and live
migrates, the source QEMU crashes before migration completetion...
This happens because virtio-9p devices have a realize handler which
calls virtio_init() and register_savevm(). Both calls store pointers
to the device internals, that get dereferenced during migration even
if the device got unplugged.
This patch simply adds an unrealize handler to perform minimal
cleanup and avoid the crash. Hot unplug of non-quiescent 9p devices
is still not supported in QEMU, and not supported by linux guests
either.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Message-id: 20151208155457.27775.69441.stgit@bahia.huguette.org
[PMM: rewrapped long lines in commit message]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
We don't support migration of mounted 9p shares. This is handled by a
migration blocker.
One would expect, however, to be able to migrate if the share is unmounted.
Unfortunately virtio-9p-device does not register savevm handlers at all !
Migration succeeds and leaves the guest with a dangling device...
This patch simply registers migration handlers for virtio-9p-device. Whether
migration is possible or not still depends on the migration blocker.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The fact that these enums have matching values is pure coincidence. We
actually need to translate from the PAPR definition to the QEMU one.
This patch doesn't fix any bug, it is only code cleanup.
Suggested-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
The tce_list variable is not a TCE but the address to a TCE: we shouldn't
clear permission bits as we do now. And this is dead code anyway since we
check tce_list is 4K aligned a few lines above.
This patch doesn't fix any bug, it is only code cleanup.
Suggested-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
XICS needs to know the upper value for cpu_index as it is used to compute
the number of servers:
smp_cpus * kvmppc_smt_threads() / smp_threads
When passing -smp cpus=1,threads=9 on a POWER8 host, we end up with:
1 * 8 / 9 = 0
... which leads to an assertion in both emulated:
Number of servers needs to be greater 0
Aborted (core dumped)
... and in-kernel XICS:
xics_kvm_realize: Assertion `icp->nr_servers' failed.
Aborted (core dumped)
With this patch, we are sure that nr_servers > 0. Passing the same bogus
-smp option then leads to:
qemu-system-ppc64: Cannot support more than 8 threads on PPC with KVM
... which is a lot more explicit than the XICS errors.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
Accesses to vring_avail_event and vring_used_event must honor the queue
endianness.
This patch allows cross-endian setups to use dataplane (tested with ppc64
on ppc64le, and vice-versa).
Suggested-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
The default behaviour for TAP/MACVTAP is to consider vnet as native endian.
This patch handles the cases when this is not true:
- virtio 1.0: always little-endian
- legacy cross-endian
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The linux tap and macvtap backends can be told to parse vnet headers
according to little or big endian. This is done through the TUNSETVNETLE
and TUNSETVNETBE ioctls.
This patch brings all the plumbing for QEMU to use these APIs.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Legacy virtio is native endian: if the guest and host endianness differ,
we have to tell vhost so it can swap bytes where appropriate. This is
done through a vhost ring ioctl.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This helper will be used by vhost and tap to detect cross-endianness in
the legacy virtio case.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
It fixes the following error:
Traceback (most recent call last):
File "./scripts/analyze-migration.py", line 584, in <module>
dump.read(dump_memory = args.memory)
File "./scripts/analyze-migration.py", line 528, in read
self.sections[section_id].read()
File "./scripts/analyze-migration.py", line 250, in read
self.file.readvar(n_valid * HASH_PTE_SIZE_64)
NameError: global name 'HASH_PTE_SIZE_64' is not defined
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
On sPAPR, virtio devices are connected to the PCI bus and use MSI-X.
Commit cc943c36fa has modified MSI-X
so that writes are made using the bus master address space and follow
the IOMMU path.
Unfortunately, the IOMMU address space address space does not have an
MSI window: the notification is silently dropped in unassigned_mem_write
instead of reaching the guest... The most visible effect is that all
virtio devices are non-functional on sPAPR since then. :(
This patch does the following:
1) map the MSI window into the IOMMU address space for each PHB
- since each PHB instantiates its own IOMMU address space, we
can safely map the window at a fixed address (SPAPR_PCI_MSI_WINDOW)
- no real need to keep the MSI window setup in a separate function,
the spapr_pci_msi_init() code moves to spapr_phb_realize().
2) kill the global MSI window as it is not needed in the end
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Hotplug of a virtio scsi disk is currently broken: no disk appears in the
guest (verified with a fedora 20 host running a fedora 20 guest with KVM).
Bisect leeds to Paolo's patches to support any_layout, especially this
commit:
commit 36b15c79aa
Author: Paolo Bonzini <pbonzini@redhat.com>
Date: Tue Jun 10 16:21:18 2014 +0200
virtio-scsi: start preparing for any_layout
It modifies virtio_scsi_pop_req() so that it is up to the callers to parse
the virtio scsi request. It seems that virtio_scsi_push_event() was not
modified accordingly...
This patch adds a call to virtio_scsi_parse_req(). It also drops some
sanity checks that are already performed by virtio_scsi_parse_req().
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
As of today, vhost assumes guest and host have the same endianness.
This is definitely not compatible with modern PPC64 and ARM that
can change endianness at runtime. Let's disable vhost-net and print
an error message when we detect such a case:
qemu-system-ppc64: vhost-net does not support cross-endian
qemu-system-ppc64: unable to start vhost net: 38: falling back on userspace virtio
This way users can continue to run VMs without changing their setup and
have a chance to know that performance will be impacted.
Suggested-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The device endianness is the cpu endianness at device reset time.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Note that st*_raw and ld*_raw are effectively replaced by st*_p and ld*_p.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This is the virtio-access.h header file taken from Rusty's "endian-ambivalent
targets using legacy virtio" patch. It introduces helpers that should be used
when accessing vring data or by drivers for data that contains headers.
The virtio config space is also target endian, but the current code already
handles that with the virtio_is_big_endian() helper. There is no obvious
benefit at using the virtio accessors in this case.
Now we have two distinct paths: a fast inline one for fixed endian targets,
and a slow out-of-line one for targets that define the new TARGET_IS_BIENDIAN
macro.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
[ relicensed virtio-access.h to GPLv2+ on Rusty's request,
pass &address_space_memory to physical memory accessors,
per-device endianness,
virtio tswap16 and tswap64 helpers,
faspath for fixed endian targets,
Greg Kurz <gkurz@linux.vnet.ibm.com> ]
Cc: Cédric Le Goater <clg@fr.ibm.com>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Some CPU families can dynamically change their endianness. This means we
can have little endian ppc or big endian arm guests for example. This has
an impact on legacy virtio data structures since they are target endian.
We hence introduce a new property to track the endianness of each virtio
device. It is reasonnably assumed that endianness won't change while the
device is in use : we hence capture the device endianness when it gets
reset.
We migrate this property in a subsection, after the device descriptor. This
means the load code must not rely on it until it is restored. As a consequence,
the vring sanity checks had to be moved after the call to vmstate_load_state().
We enforce paranoia by poisoning the property at the begining of virtio_load().
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
If we want to support targets that can change endianness (modern PPC and
ARM for the moment), we need to add a per-CPU class method to be called
from the virtio code. The virtio_ prefix in the name is a hint for people
to avoid misusage (aka. anywhere but from the virtio code).
The default behaviour is to return the compile-time default target
endianness.
Suggested-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
We currently have a virtio_is_big_endian() helper that provides the target
endianness to the virtio code. As of today, the helper returns a fixed
compile-time value. Of course, this will have to change if we want to
support target endianness changes at run-time.
Let's move the TARGET_WORDS_BIGENDIAN bits out to a new helper and have
virtio_is_big_endian() implemented on top of it.
This patch doesn't change any functionality.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
There is a need to add some more fields to VirtIODevice that should be
migrated (broken status, endianness). The problem is that we do not
want to break compatibility while adding a new feature... This issue has
been addressed in the generic VMState code with the use of optional
subsections. As a *temporary* alternative to port the whole virtio
migration code to VMState, this patch mimics a similar subsectionning
ability for virtio, using the VMState code.
Since each virtio device is streamed in its own section, the idea is to
stream subsections between the end of the device section and the start
of the next sections. This allows an older QEMU to complain and exit
when fed with subsections:
Unknown savevm section type 5
load of migration failed
Suggested-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
In order to migrate virtio subsections, they should be streamed after
the device itself. We need the device specific code to be called from
the common migration code to achieve this. This patch introduces load
and save methods for this purpose.
Suggested-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
POWER7, POWER7+ and POWER8 families use the ILE bit of the LPCR
special purpose register to decide the endianness to use when
entering interrupt handlers. When running a Linux guest, this
provides a hint on the endianness used by the kernel. And when
it comes to dumping a guest, the information is needed to write
ELF headers using the kernel endianness.
Suggested-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
[agraf: change subject line]
Signed-off-by: Alexander Graf <agraf@suse.de>
In the past, IO space could not be mapped into the memory address space
so we introduced a workaround for that. Nowadays it does not look
necessary so we can remove the workaround and make sPAPR PCI
configuration simplier.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Acked-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
The LPCR special purpose register was introduced with the PowerPC 970MP family.
This patch initializes LPCR for the following families:
- 970 MP
- POWER5+
- POWER7
- POWER8
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
The latest update to v3.13-rc3 (bf63839f) breaks the
ppc build with KVM:
kvm-all.o: In function `kvm_update_guest_debug':
kvm-all.c:1910: undefined reference to `kvm_arch_update_guest_debug'
kvm-all.o: In function `kvm_insert_breakpoint':
kvm-all.c:1937: undefined reference to `kvm_arch_insert_sw_breakpoint'
kvm-all.c:1945: undefined reference to `kvm_arch_insert_hw_breakpoint'
kvm-all.o: In function `kvm_remove_breakpoint':
kvm-all.c:1977: undefined reference to `kvm_arch_remove_sw_breakpoint'
kvm-all.c:1985: undefined reference to `kvm_arch_remove_hw_breakpoint'
kvm-all.o: In function `kvm_remove_all_breakpoints':
kvm-all.c:2009: undefined reference to `kvm_arch_remove_sw_breakpoint'
kvm-all.c:2006: undefined reference to `kvm_arch_remove_sw_breakpoint'
kvm-all.c:2017: undefined reference to `kvm_arch_remove_all_hw_breakpoints'
We need stubs until something gets implemented.
Signed-off-by: Greg Kurz <gkurz@linux.vnet.ibm.com>
Reviewed-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Alexander Graf <agraf@suse.de>
Greg Kurz
Greg Kurz