The next step to centralize all NUMA/associativity handling in
the spapr machine is to create a 'one stop place' for all
things ibm,associativity.
This patch introduces numa_assoc_array, a 2 dimensional array
that will store all ibm,associativity arrays of all NUMA nodes.
This array is initialized in a new spapr_numa_associativity_init()
function, called in spapr_machine_init(). It is being initialized
with the same values used in other ibm,associativity properties
around spapr files (i.e. all zeros, last value is node_id).
The idea is to remove all hardcoded definitions and FDT writes
of ibm,associativity arrays, doing instead a call to the new
helper spapr_numa_write_associativity_dt() helper, that will
be able to write the DT with the correct values.
We'll start small, handling the trivial cases first. The
remaining instances of ibm,associativity will be handled
next.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20200903220639.563090-2-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The spapr_phb_realize() function has a local_err variable which
is used to:
1) check failures of spapr_irq_findone() and spapr_irq_claim()
2) prepend extra information to the error message
Recent work from Markus Armbruster highlighted we get better
code when testing the return value of a function, rather than
setting up all the local_err boiler plate. For similar reasons,
it is now preferred to use ERRP_GUARD() and error_prepend()
rather than error_propagate_prepend().
Since spapr_irq_findone() and spapr_irq_claim() return negative
values in case of failure, do both changes.
This is just cleanup, no functional impact.
Signed-off-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Message-Id: <159707843851.1489912.6108405733810934642.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
NUMA nodes corresponding to GPU memory currently have the same
affinity/distance as normal memory nodes. Add a third NUMA associativity
reference point enabling us to give GPU nodes more distance.
This is guest visible information, which shouldn't change under a
running guest across migration between different qemu versions, so make
the change effective only in new (pseries > 5.0) machine types.
Before, `numactl -H` output in a guest with 4 GPUs (nodes 2-5):
node distances:
node 0 1 2 3 4 5
0: 10 40 40 40 40 40
1: 40 10 40 40 40 40
2: 40 40 10 40 40 40
3: 40 40 40 10 40 40
4: 40 40 40 40 10 40
5: 40 40 40 40 40 10
After:
node distances:
node 0 1 2 3 4 5
0: 10 40 80 80 80 80
1: 40 10 80 80 80 80
2: 80 80 10 80 80 80
3: 80 80 80 10 80 80
4: 80 80 80 80 10 80
5: 80 80 80 80 80 10
These are the same distances as on the host, mirroring the change made
to host firmware in skiboot commit f845a648b8cb ("numa/associativity:
Add a new level of NUMA for GPU's").
Signed-off-by: Reza Arbab <arbab@linux.ibm.com>
Message-Id: <20200716225655.24289-1-arbab@linux.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Some recent error handling cleanups unveiled issues with our support of
PCI bridges:
1) QEMU aborts when using non-standard PCI bridge types,
unveiled by commit 7ef1553dac "spapr_pci: Drop some dead error handling"
$ qemu-system-ppc64 -M pseries -device pcie-pci-bridge
Unexpected error in object_property_find() at qom/object.c:1240:
qemu-system-ppc64: -device pcie-pci-bridge: Property '.chassis_nr' not found
Aborted (core dumped)
This happens because we assume all PCI bridge types to have a "chassis_nr"
property. This property only exists with the standard PCI bridge type
"pci-bridge" actually. We could possibly revert 7ef1553dac but it seems
much simpler to check the presence of "chassis_nr" earlier.
2) QEMU abort if same "chassis_nr" value is used several times,
unveiled by commit d2623129a7 "qom: Drop parameter @errp of
object_property_add() & friends"
$ qemu-system-ppc64 -M pseries -device pci-bridge,chassis_nr=1 \
-device pci-bridge,chassis_nr=1
Unexpected error in object_property_try_add() at qom/object.c:1167:
qemu-system-ppc64: -device pci-bridge,chassis_nr=1: attempt to add duplicate property '40000100' to object (type 'container')
Aborted (core dumped)
This happens because we assume that "chassis_nr" values are unique, but
nobody enforces that and we end up generating duplicate DRC ids. The PCI
code doesn't really care for duplicate "chassis_nr" properties since it
is only used to initialize the "Chassis Number Register" of the bridge,
with no functional impact on QEMU. So, even if passing the same value
several times might look weird, it never broke anything before, so
I guess we don't necessarily want to enforce strict checking in the PCI
code now.
Workaround both issues in the PAPR code: check that the bridge has a
unique and non null "chassis_nr" when plugging it into its parent bus.
Fixes: 05929a6c5d ("spapr: Don't use bus number for building DRC ids")
Fixes: 7ef1553dac ("spapr_pci: Drop some dead error handling")
Fixes: d2623129a7 ("qom: Drop parameter @errp of object_property_add() & friends")
Reported-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <159431476748.407044.16711294833569014964.stgit@bahia.lan>
[dwg: Move check slightly to a better place]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Replace
error_setg(&err, ...);
error_propagate(errp, err);
by
error_setg(errp, ...);
Related pattern:
if (...) {
error_setg(&err, ...);
goto out;
}
...
out:
error_propagate(errp, err);
return;
When all paths to label out are that way, replace by
if (...) {
error_setg(errp, ...);
return;
}
and delete the label along with the error_propagate().
When we have at most one other path that actually needs to propagate,
and maybe one at the end that where propagation is unnecessary, e.g.
foo(..., &err);
if (err) {
goto out;
}
...
bar(..., &err);
out:
error_propagate(errp, err);
return;
move the error_propagate() to where it's needed, like
if (...) {
foo(..., &err);
error_propagate(errp, err);
return;
}
...
bar(..., errp);
return;
and transform the error_setg() as above.
In some places, the transformation results in obviously unnecessary
error_propagate(). The next few commits will eliminate them.
Bonus: the elimination of gotos will make later patches in this series
easier to review.
Candidates for conversion tracked down with this Coccinelle script:
@@
identifier err, errp;
expression list args;
@@
- error_setg(&err, args);
+ error_setg(errp, args);
... when != err
error_propagate(errp, err);
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200707160613.848843-34-armbru@redhat.com>
The object_property_set_FOO() setters take property name and value in
an unusual order:
void object_property_set_FOO(Object *obj, FOO_TYPE value,
const char *name, Error **errp)
Having to pass value before name feels grating. Swap them.
Same for object_property_set(), object_property_get(), and
object_property_parse().
Convert callers with this Coccinelle script:
@@
identifier fun = {
object_property_get, object_property_parse, object_property_set_str,
object_property_set_link, object_property_set_bool,
object_property_set_int, object_property_set_uint, object_property_set,
object_property_set_qobject
};
expression obj, v, name, errp;
@@
- fun(obj, v, name, errp)
+ fun(obj, name, v, errp)
Chokes on hw/arm/musicpal.c's lcd_refresh() with the unhelpful error
message "no position information". Convert that one manually.
Fails to convert hw/arm/armsse.c, because Coccinelle gets confused by
ARMSSE being used both as typedef and function-like macro there.
Convert manually.
Fails to convert hw/rx/rx-gdbsim.c, because Coccinelle gets confused
by RXCPU being used both as typedef and function-like macro there.
Convert manually. The other files using RXCPU that way don't need
conversion.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20200707160613.848843-27-armbru@redhat.com>
[Straightforwad conflict with commit 2336172d9b "audio: set default
value for pcspk.iobase property" resolved]
qbus_set_hotplug_handler() is a simple wrapper around
object_property_set_link().
object_property_set_link() fails when the property doesn't exist, is
not settable, or its .check() method fails. These are all programming
errors here, so passing &error_abort to qbus_set_hotplug_handler() is
appropriate.
Most of its callers do. Exceptions:
* pcie_cap_slot_init(), shpc_init(), spapr_phb_realize() pass NULL,
i.e. they ignore errors.
* spapr_machine_init() passes &error_fatal.
* s390_pcihost_realize(), virtio_serial_device_realize(),
s390_pcihost_plug() pass the error to their callers. The latter two
keep going after the error, which looks wrong.
Drop the @errp parameter, and instead pass &error_abort to
object_property_set_link().
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Daniel P. Berrangé" <berrange@redhat.com>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200630090351.1247703-15-armbru@redhat.com>
chassis_from_bus() uses object_property_get_uint() to get property
"chassis_nr" of the bridge device. Failure would be a programming
error. Pass &error_abort, and simplify its callers.
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: qemu-ppc@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Acked-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-18-armbru@redhat.com>
Devices may have component devices and buses.
Device realization may fail. Realization is recursive: a device's
realize() method realizes its components, and device_set_realized()
realizes its buses (which should in turn realize the devices on that
bus, except bus_set_realized() doesn't implement that, yet).
When realization of a component or bus fails, we need to roll back:
unrealize everything we realized so far. If any of these unrealizes
failed, the device would be left in an inconsistent state. Must not
happen.
device_set_realized() lets it happen: it ignores errors in the roll
back code starting at label child_realize_fail.
Since realization is recursive, unrealization must be recursive, too.
But how could a partly failed unrealize be rolled back? We'd have to
re-realize, which can fail. This design is fundamentally broken.
device_set_realized() does not roll back at all. Instead, it keeps
unrealizing, ignoring further errors.
It can screw up even for a device with no buses: if the lone
dc->unrealize() fails, it still unregisters vmstate, and calls
listeners' unrealize() callback.
bus_set_realized() does not roll back either. Instead, it stops
unrealizing.
Fortunately, no unrealize method can fail, as we'll see below.
To fix the design error, drop parameter @errp from all the unrealize
methods.
Any unrealize method that uses @errp now needs an update. This leads
us to unrealize() methods that can fail. Merely passing it to another
unrealize method cannot cause failure, though. Here are the ones that
do other things with @errp:
* virtio_serial_device_unrealize()
Fails when qbus_set_hotplug_handler() fails, but still does all the
other work. On failure, the device would stay realized with its
resources completely gone. Oops. Can't happen, because
qbus_set_hotplug_handler() can't actually fail here. Pass
&error_abort to qbus_set_hotplug_handler() instead.
* hw/ppc/spapr_drc.c's unrealize()
Fails when object_property_del() fails, but all the other work is
already done. On failure, the device would stay realized with its
vmstate registration gone. Oops. Can't happen, because
object_property_del() can't actually fail here. Pass &error_abort
to object_property_del() instead.
* spapr_phb_unrealize()
Fails and bails out when remove_drcs() fails, but other work is
already done. On failure, the device would stay realized with some
of its resources gone. Oops. remove_drcs() fails only when
chassis_from_bus()'s object_property_get_uint() fails, and it can't
here. Pass &error_abort to remove_drcs() instead.
Therefore, no unrealize method can fail before this patch.
device_set_realized()'s recursive unrealization via bus uses
object_property_set_bool(). Can't drop @errp there, so pass
&error_abort.
We similarly unrealize with object_property_set_bool() elsewhere,
always ignoring errors. Pass &error_abort instead.
Several unrealize methods no longer handle errors from other unrealize
methods: virtio_9p_device_unrealize(),
virtio_input_device_unrealize(), scsi_qdev_unrealize(), ...
Much of the deleted error handling looks wrong anyway.
One unrealize methods no longer ignore such errors:
usb_ehci_pci_exit().
Several realize methods no longer ignore errors when rolling back:
v9fs_device_realize_common(), pci_qdev_unrealize(),
spapr_phb_realize(), usb_qdev_realize(), vfio_ccw_realize(),
virtio_device_realize().
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-17-armbru@redhat.com>
Currently, we can't properly handle unplug of NVLink2 devices, because we
don't have code to tear down their special memory resources. There's not
a lot of impetus to implement that: since hardware NVLink2 devices can't
be hot unplugged, the guest side drivers don't usually support unplug
anyway.
Therefore, simply prevent unplug of NVLink2 devices.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
For various technical reasons we can't currently allow unplug a PCI to PCI
bridge on the pseries machine. spapr_pci_unplug_request() correctly
generates an error message if that's attempted.
But.. if the given errp is not error_abort or error_fatal, it doesn't
actually stop trying to unplug the bridge anyway.
Fixes: 14e714900f "spapr: Allow hot plug/unplug of PCI bridges and devices under PCI bridges"
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Provide a temporary device_legacy_reset function doing what
device_reset does to prepare for the transition with Resettable
API.
All occurrence of device_reset in the code tree are also replaced
by device_legacy_reset.
The new resettable API has different prototype and semantics
(resetting child buses as well as the specified device). Subsequent
commits will make the changeover for each call site individually; once
that is complete device_legacy_reset() will be removed.
Signed-off-by: Damien Hedde <damien.hedde@greensocs.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Acked-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Cornelia Huck <cohuck@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20200123132823.1117486-2-damien.hedde@greensocs.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Local Error * variables are conventionally named @err or @local_err,
and Error ** parameters @errp. Naming local variables like parameters
is confusing. Clean that up.
Naming parameters like local variables is also confusing. Left for
another day.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20191204093625.14836-17-armbru@redhat.com>
There are three page size in qemu:
real host page size
host page size
target page size
All of them have dedicate variable to represent. For the last two, we
use the same form in the whole qemu project, while for the first one we
use two forms: qemu_real_host_page_size and getpagesize().
qemu_real_host_page_size is defined to be a replacement of
getpagesize(), so let it serve the role.
[Note] Not fully tested for some arch or device.
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Message-Id: <20191013021145.16011-3-richardw.yang@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The nr_msis value we use here has to line up with whether we're using
legacy or modern irq allocation. Therefore it's safer to derive it based
on legacy_irq_allocation rather than having SpaprIrq contain a canned
value.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
No point having a two-line helper that's used exactly once, and not likely
to be used anywhere else in future.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
QEMU does not allocate PCI resources (BARs) in any case - coldplug devices
are configured by the firmware and hotplug devices rely on the guest
system to do the assignment via the PCI rescan mechanism. Also in order
to create non empty "assigned-addresses", the device has to be enabled
(i.e. PCI_COMMAND needs the MMIO bit set) first as otherwise
io_regions[i].addr are -1, and devices are not enabled at this point.
This removes "assigned-addresses" and leaves it to those who actually
do resource allocation.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Message-Id: <20190927022651.71642-1-aik@ozlabs.ru>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Bug fixes:
* Fix die-id validation regression (Eduardo Habkost)
* vmmouse: Properly reset state (Jan Kiszka)
* hostmem-file: fix pmem file size check (Stefan Hajnoczi)
* Keep query-hotpluggable-cpus output compatible with older QEMU
if '-smp dies' is not set (Igor Mammedov)
* migration: Do not re-read the clock on pre_save in case of paused guest
(Maxiwell S. Garcia)
Cleanups:
* NUMA code cleanups (Tao Xu)
* Remove stale externs from includes (Alex Bennée)
Features:
* qapi: report the default CPU type for each machine (Daniel P. Berrangé)
-----BEGIN PGP SIGNATURE-----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=u5HZ
-----END PGP SIGNATURE-----
Merge remote-tracking branch 'remotes/ehabkost/tags/machine-next-pull-request' into staging
Machine + x86 queue, 2019-09-03
Bug fixes:
* Fix die-id validation regression (Eduardo Habkost)
* vmmouse: Properly reset state (Jan Kiszka)
* hostmem-file: fix pmem file size check (Stefan Hajnoczi)
* Keep query-hotpluggable-cpus output compatible with older QEMU
if '-smp dies' is not set (Igor Mammedov)
* migration: Do not re-read the clock on pre_save in case of paused guest
(Maxiwell S. Garcia)
Cleanups:
* NUMA code cleanups (Tao Xu)
* Remove stale externs from includes (Alex Bennée)
Features:
* qapi: report the default CPU type for each machine (Daniel P. Berrangé)
# gpg: Signature made Tue 03 Sep 2019 21:57:37 BST
# gpg: using RSA key 5A322FD5ABC4D3DBACCFD1AA2807936F984DC5A6
# gpg: issuer "ehabkost@redhat.com"
# gpg: Good signature from "Eduardo Habkost <ehabkost@redhat.com>" [full]
# Primary key fingerprint: 5A32 2FD5 ABC4 D3DB ACCF D1AA 2807 936F 984D C5A6
* remotes/ehabkost/tags/machine-next-pull-request:
migration: Do not re-read the clock on pre_save in case of paused guest
x86: do not advertise die-id in query-hotpluggbale-cpus if '-smp dies' is not set
i386/vmmouse: Properly reset state
hostmem-file: fix pmem file size check
qapi: report the default CPU type for each machine
pc: Don't make die-id mandatory unless necessary
pc: Improve error message when die-id is omitted
pc: Fix error message on die-id validation
numa: move numa global variable numa_info into MachineState
numa: move numa global variable have_numa_distance into MachineState
numa: move numa global variable nb_numa_nodes into MachineState
hw/arm: simplify arm_load_dtb
includes: remove stale [smp|max]_cpus externs
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Move existing numa global numa_info (renamed as "nodes") into NumaState.
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Suggested-by: Igor Mammedov <imammedo@redhat.com>
Suggested-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Tao Xu <tao3.xu@intel.com>
Message-Id: <20190809065731.9097-5-tao3.xu@intel.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
The QEMU coding style requires:
- to typedef structured types (HACKING)
- to use CamelCase for types and structure names (CODING_STYLE)
Do that for PCI and Nvlink2 code.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156701644465.505236.2850655823182656869.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
There is nothing wrong with how sPAPR handles multifunction PCI
hot unplugs. The problem is that x86 does it simpler. Instead of
removing each non-zero function and then removing function zero,
x86 can remove any function of the slot to trigger the hot unplug.
Libvirt will be directly impacted by this difference, in the
(hopefully soon) PCI Multifunction hot plug/unplug support. For
hot plugs, both x86 and sPAPR will operate the same way: a XML
with all desired functions to be added, then consecutive hotplugs
of all non-zero functions first, zero last. For hot unplugs, at
least in the current state, a XML with the devices to be removed
must also be provided because of how sPAPR operates - x86 does
not need it - since any function unplug will unplug the whole
PCIe slot. This difference puts extra strain in the management
layer, which needs to either handle both archs differently in
the unplug scenario or choose treat x86 like sPAPR, forcing x86
users to cope with sPAPR internals.
This patch changes spapr_pci_unplug_request to handle the
unplug of function zero differently. When removing function zero,
instead of error-ing out if there are any remaining function
DRCs which needs detaching, detach those. This has no effect in
any existing scripts that are detaching the non-zero functions
before function zero, and can be used by management as a shortcut
to remove the whole PCI multifunction device without specifying
each child function.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20190822195918.3307-1-danielhb413@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
When the machine is reset, the MSI bitmap is cleared but the allocated
MSIs are not freed. Some operating systems, such as AIX, can detect the
previous configuration and assert.
Empty the MSI cache, this performs the needed cleanup.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156415228410.1064338.4486161194061636096.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
When freeing MSIs, we need to:
- remove them from the machine's MSI bitmap
- remove them from the IC backend
- remove them from the PHB's MSI cache
This is currently open coded in two places in rtas_ibm_change_msi(),
and we're about to need this in spapr_phb_reset() as well. Instead of
duplicating this code again, make it a destroy function for the PHB's
MSI cache. Removing an MSI device from the cache will call the destroy
function internally.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156415227855.1064338.5657793835271464648.stgit@bahia.lan>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
We've had the qemu and kernel KVM infrastructure to handle larger TCE
page sizes for a while, but forgot to update the defaults to actually
allow them. This turns that change on.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
In my "build everything" tree, changing hw/qdev-properties.h triggers
a recompile of some 2700 out of 6600 objects (not counting tests and
objects that don't depend on qemu/osdep.h).
Many places including hw/qdev-properties.h (directly or via hw/qdev.h)
actually need only hw/qdev-core.h. Include hw/qdev-core.h there
instead.
hw/qdev.h is actually pointless: all it does is include hw/qdev-core.h
and hw/qdev-properties.h, which in turn includes hw/qdev-core.h.
Replace the remaining uses of hw/qdev.h by hw/qdev-properties.h.
While there, delete a few superfluous inclusions of hw/qdev-core.h.
Touching hw/qdev-properties.h now recompiles some 1200 objects.
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Daniel P. Berrangé" <berrange@redhat.com>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <20190812052359.30071-22-armbru@redhat.com>
In my "build everything" tree, changing hw/hw.h triggers a recompile
of some 2600 out of 6600 objects (not counting tests and objects that
don't depend on qemu/osdep.h).
The previous commits have left only the declaration of hw_error() in
hw/hw.h. This permits dropping most of its inclusions. Touching it
now recompiles less than 200 objects.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20190812052359.30071-19-armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
In my "build everything" tree, changing migration/vmstate.h triggers a
recompile of some 2700 out of 6600 objects (not counting tests and
objects that don't depend on qemu/osdep.h).
hw/hw.h supposedly includes it for convenience. Several other headers
include it just to get VMStateDescription. The previous commit made
that unnecessary.
Include migration/vmstate.h only where it's still needed. Touching it
now recompiles only some 1600 objects.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20190812052359.30071-16-armbru@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
In my "build everything" tree, changing hw/irq.h triggers a recompile
of some 5400 out of 6600 objects (not counting tests and objects that
don't depend on qemu/osdep.h).
hw/hw.h supposedly includes it for convenience. Several other headers
include it just to get qemu_irq and.or qemu_irq_handler.
Move the qemu_irq and qemu_irq_handler typedefs from hw/irq.h to
qemu/typedefs.h, and then include hw/irq.h only where it's still
needed. Touching it now recompiles only some 500 objects.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190812052359.30071-13-armbru@redhat.com>
Hot-unplugging a PHB with a VFIO device connected to it crashes QEMU:
-device spapr-pci-host-bridge,index=1,id=phb1 \
-device vfio-pci,host=0034:01:00.3,id=vfio0
(qemu) device_del phb1
[ 357.207183] iommu: Removing device 0001:00:00.0 from group 1
[ 360.375523] rpadlpar_io: slot PHB 1 removed
qemu-system-ppc64: memory.c:2742:
do_address_space_destroy: Assertion `QTAILQ_EMPTY(&as->listeners)' failed.
'as' is the IOMMU address space, which indeed has a listener registered
to by vfio_connect_container() when the VFIO device is realized. This
listener is supposed to be unregistered by vfio_disconnect_container()
when the VFIO device is finalized. Unfortunately, the VFIO device hasn't
reached finalize yet at the time the PHB unrealize function is called,
and address_space_destroy() gets called with the VFIO listener still
being registered.
All regions have just been unmapped from the address space. Listeners
aren't needed anymore at this point. Remove them before destroying the
address space.
The VFIO code will try to remove them _again_ at device finalize,
but it is okay since memory_listener_unregister() is idempotent.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156110925375.92514.11649846071216864570.stgit@bahia.lan>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
[dwg: Correct spelling error pointed out by aik]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
kvm_enabled() expands to (0) when CONFIG_KVM is not defined.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156051052977.224162.17306829691809502082.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
spapr_dt_drc() scans the aliases of all DRConnector objects and filters
the ones that it will use to generate OF properties according to their
owner and type.
Passing bus->parent_dev _works_ if bus belongs to a PCI bridge, but it is
NULL if it is the PHB's root bus. This causes all allocated PCI DRCs to
be associated to all PHBs (visible in their "ibm,drc-types" properties).
As a consequence, hot unplugging a PHB results in PCI devices from the
other PHBs to be unplugged as well, and likely confuses the guest.
Use the same logic as in add_drcs() to ensure the correct owner is passed
to spapr_dt_drc().
Fixes: 14e714900f "spapr: Allow hot plug/unplug of PCI bridges and devices under PCI bridges"
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <156084737348.512412.3552825999605902691.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Commit 14e714900f refactored the call to spapr_dt_drc(),
introducing a potential NULL pointer dereference while
accessing bus->parent_dev.
A trivial audit show 'bus' is not null in the two places
the static function spapr_dt_drc() is called.
Since the 'bus' parameter is not NULL in both callers, remove
remove the test on if (bus), and add an assert() to silent
static analyzers.
This fixes:
/hw/ppc/spapr_pci.c: 1367 in spapr_dt_pci_bus()
>>> CID 1401933: Null pointer dereferences (FORWARD_NULL)
>>> Dereferencing null pointer "bus".
1367 ret = spapr_dt_drc(fdt, offset, OBJECT(bus->parent_dev),
1368 SPAPR_DR_CONNECTOR_TYPE_PCI);
Fixes: 14e714900f
Reported-by: Coverity (CID 1401933)
Suggested-by: Greg Kurz <groug@kaod.org>
Suggested-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190613213406.22053-1-philmd@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Next pull request against qemu-4.1. The big thing here is adding
support for hot plug of P2P bridges, and PCI devices under P2P bridges
on the "pseries" machine (which doesn't use SHPC). Other than that
there's just a handful of fixes and small enhancements.
-----BEGIN PGP SIGNATURE-----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=ERgu
-----END PGP SIGNATURE-----
Merge remote-tracking branch 'remotes/dgibson/tags/ppc-for-4.1-20190612' into staging
ppc patch queue 2019-06-12
Next pull request against qemu-4.1. The big thing here is adding
support for hot plug of P2P bridges, and PCI devices under P2P bridges
on the "pseries" machine (which doesn't use SHPC). Other than that
there's just a handful of fixes and small enhancements.
# gpg: Signature made Wed 12 Jun 2019 06:47:56 BST
# gpg: using RSA key 75F46586AE61A66CC44E87DC6C38CACA20D9B392
# gpg: Good signature from "David Gibson <david@gibson.dropbear.id.au>" [full]
# gpg: aka "David Gibson (Red Hat) <dgibson@redhat.com>" [full]
# gpg: aka "David Gibson (ozlabs.org) <dgibson@ozlabs.org>" [full]
# gpg: aka "David Gibson (kernel.org) <dwg@kernel.org>" [unknown]
# Primary key fingerprint: 75F4 6586 AE61 A66C C44E 87DC 6C38 CACA 20D9 B392
* remotes/dgibson/tags/ppc-for-4.1-20190612:
ppc/xive: Make XIVE generate the proper interrupt types
ppc/pnv: activate the "dumpdtb" option on the powernv machine
target/ppc: Use tcg_gen_gvec_bitsel
spapr: Allow hot plug/unplug of PCI bridges and devices under PCI bridges
spapr: Direct all PCI hotplug to host bridge, rather than P2P bridge
spapr: Don't use bus number for building DRC ids
spapr: Clean up DRC index construction
spapr: Clean up spapr_drc_populate_dt()
spapr: Clean up dt creation for PCI buses
spapr: Clean up device tree construction for PCI devices
spapr: Clean up device node name generation for PCI devices
target/ppc: Fix lxvw4x, lxvh8x and lxvb16x
spapr_pci: Improve error message
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The pseries machine type already allows PCI hotplug and unplug via the
PAPR mechanism, but only on the root bus of each PHB. This patch extends
this to allow PCI to PCI bridges to be hotplugged, and devices to be
hotplugged or unplugged under P2P bridges.
For now we disallow hot unplugging P2P bridges. I tried doing that, but
haven't managed to get it working, I think due to some guest side problems
that need further investigation.
To do this we dynamically construct DRCs when bridges are hot (or cold)
added, which can in turn be used to hotplug devices under the bridge.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
DRC ids are more or less arbitrary, as long as they're consistent. For
PCI, we notionally build them from the phb's index along with PCI bus
number, slot and function number.
Using bus number is broken, however, because it can change if the guest
re-enumerates the PCI topology for whatever reason (e.g. due to hotplug
of a bridge, which we don't support yet but want to).
Fortunately, there's an alternative. Bridges are required to have a unique
non-zero "chassis number" that we can use instead. Adjust the code to
use that instead.
This looks like it would introduce a guest visible breaking change, but
in fact it does not because we don't yet ever use non-zero bus numbers.
Both chassis and bus number are always 0 for the root bus, so there's no
change for the existing cases.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
spapr_pci.c currently has several confusingly similarly named functions for
various conversions between representations of DRCs. Make things clearer
by renaming things in a more consistent XXX_from_YYY() manner and remove
some called-only-once variants in favour of open coding.
While we're at it, move this code together in the file to avoid some extra
forward references, and split out construction and removal of DRCs for the
host bridge into helper functions.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
This makes some minor cleanups to spapr_drc_populate_dt(), renaming it to
the shorter and more idiomatic spapr_dt_drc() along the way.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Device nodes for PCI bridges (both host and P2P) describe both the bridge
device itself and the bus hanging off it, handling of this is a bit of a
mess.
spapr_dt_pci_device() has a few things it only adds for non-bridges, but
always adds #address-cells and #size-cells which should only appear for
bridges. But the walking down the subordinate PCI bus is done in one of
its callers spapr_populate_pci_devices_dt(). The PHB dt creation in
spapr_populate_pci_dt() open codes some similar logic to the bridge case.
This patch consolidates things in a bunch of ways:
* Bus specific dt info is now created in spapr_dt_pci_bus() used for both
P2P bridges and the host bridge. This includes walking subordinate
devices
* spapr_dt_pci_device() now calls spapr_dt_pci_bus() when called on a
P2P bridge
* We do detection of bridges with the is_bridge field of the device class,
rather than checking PCI config space directly, for consistency with
qemu's core PCI code.
* Several things are renamed for brevity and clarity
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
spapr_create_pci_child_dt() is a trivial wrapper around
spapr_populate_pci_child_dt(), but is the latter's only caller. So fold
them together into spapr_dt_pci_device(), which closer matches our modern
naming convention.
While there, make a number of cleanups to the function itself. This is
mostly using more temporary locals to avoid awkwardly long lines, and in
some cases avoiding double reads of PCI config space variables.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
spapr_populate_pci_child_dt() adds a 'name' property to the device tree
node for PCI devices. This is never necessary for a flattened device tree,
it is implicit in the name added when the node is constructed. In fact
anything we do add to a 'name' property will be overwritten with something
derived from the structural name in the guest firmware (but in fact it is
exactly the same bytes).
So, remove that. In addition, pci_get_node_name() is very simple, so fold
it into its (also simple) sole caller spapr_create_pci_child_dt().
While we're there rename pci_find_device_name() to the shorter and more
accurate dt_name_from_class().
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Every PHB must have a unique index. This is checked at realize but when
a duplicate index is detected, an error message mentioning BUIDs is
printed. This doesn't help much, especially since BUID is an internal
concept that is no longer exposed to the user.
Fix the message to mention the index property instead of BUID. As a bonus
print a list of indexes already in use.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <155915010892.2061314.10485622810149098411.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Since c2077e2c "pci: Adjust PCI config limit based on bus topology",
pci_adjust_config_limit() has been used in the config space read and write
paths to only permit access to extended config space on buses which permit
it. Specifically it prevents access on devices below a vanilla-PCI bus via
some combination of bridges, even if both the host bridge and the device
itself are PCI-E.
It accomplishes this with a somewhat complex call up the chain of bridges
to see if any of them prohibit extended config space access. This is
overly complex, since we can always know if the bus will support such
access at the point it is constructed.
This patch simplifies the test by using a flag in the PCIBus instance
indicating whether extended configuration space is accessible. It is
false for vanilla PCI buses. For PCI-E buses, it is true for root
buses and equal to the parent bus's's capability otherwise.
For the special case of sPAPR's paravirtualized PCI root bus, which
acts mostly like vanilla PCI, but does allow extended config space
access, we override the default value of the flag from the host bridge
code.
This should cause no behavioural change.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Message-Id: <20190513061939.3464-4-david@gibson.dropbear.id.au>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
LSI mapping in spapr currently open-codes standard PCI swizzling. It thus
duplicates the code of pci_swizzle_map_irq_fn().
Expose the swizzling formula so that it can be used with a slot number
when building the device tree. Simply drop pci_spapr_map_irq() and call
pci_swizzle_map_irq_fn() instead.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <155448184841.8446.13959787238854054119.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
According to the changelog of 298a971024, SpaprPhbState::dtbusname was
introduced to "make it easier to relate the guest and qemu views of memory
to each other", hence its name.
Use it when creating the PHB node to avoid code duplication.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <155448184292.8446.8225650773162648595.stgit@bahia.lan>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
NVIDIA V100 GPUs have on-board RAM which is mapped into the host memory
space and accessible as normal RAM via an NVLink bus. The VFIO-PCI driver
implements special regions for such GPUs and emulates an NVLink bridge.
NVLink2-enabled POWER9 CPUs also provide address translation services
which includes an ATS shootdown (ATSD) register exported via the NVLink
bridge device.
This adds a quirk to VFIO to map the GPU memory and create an MR;
the new MR is stored in a PCI device as a QOM link. The sPAPR PCI uses
this to get the MR and map it to the system address space.
Another quirk does the same for ATSD.
This adds additional steps to sPAPR PHB setup:
1. Search for specific GPUs and NPUs, collect findings in
sPAPRPHBState::nvgpus, manage system address space mappings;
2. Add device-specific properties such as "ibm,npu", "ibm,gpu",
"memory-block", "link-speed" to advertise the NVLink2 function to
the guest;
3. Add "mmio-atsd" to vPHB to advertise the ATSD capability;
4. Add new memory blocks (with extra "linux,memory-usable" to prevent
the guest OS from accessing the new memory until it is onlined) and
npuphb# nodes representing an NPU unit for every vPHB as the GPU driver
uses it for link discovery.
This allocates space for GPU RAM and ATSD like we do for MMIOs by
adding 2 new parameters to the phb_placement() hook. Older machine types
set these to zero.
This puts new memory nodes in a separate NUMA node to as the GPU RAM
needs to be configured equally distant from any other node in the system.
Unlike the host setup which assigns numa ids from 255 downwards, this
adds new NUMA nodes after the user configures nodes or from 1 if none
were configured.
This adds requirement similar to EEH - one IOMMU group per vPHB.
The reason for this is that ATSD registers belong to a physical NPU
so they cannot invalidate translations on GPUs attached to another NPU.
It is guaranteed by the host platform as it does not mix NVLink bridges
or GPUs from different NPU in the same IOMMU group. If more than one
IOMMU group is detected on a vPHB, this disables ATSD support for that
vPHB and prints a warning.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
[aw: for vfio portions]
Acked-by: Alex Williamson <alex.williamson@redhat.com>
Message-Id: <20190312082103.130561-1-aik@ozlabs.ru>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Recent commit 5cf0d326a0 fixed a regression which was preventing the
guest to access the extended config space of a PCIe device. This was
done by introducing a new PCI bus subtype for PAPR. The original fix
was causing PCI busses to be named "spapr-pci-host-bridge-root-bus.N"
instead of "pci.N", which was making upper layers unhappy of course.
This got worked around by hardcoding the PCI bus name to "pci.0", but
this only works for the default PHB. And we're now hitting:
# qemu-system-ppc64 \
-device spapr-pci-host-bridge,index=1 \
-device e1000e,bus=pci.0 \
-device e1000e,bus=pci.1
qemu-system-ppc64: -device e1000e,bus=pci.1: Bus 'pci.1' not found
David already posted some patches [1] to control PCI extended config
space accesses with a new flag in the base PCI bus class instead of
subtyping. These patches are a bit more intrusive though, and
are targetted for 4.1.
When no name is passed to pci_register_bus(), the core device code
generates a lowercase name based on the QOM typename. The typename
for the base PCI bus class is "PCI", hence the "pci.0", "pci.1"
bus names. Rename the type of the PAPR PCI bus to "pci", so that
the QOM code can generate proper names. This is a hack but it is
enough to fix the regression. And all this will be reworked properly
in 4.1.
[1] https://patchwork.ozlabs.org/project/qemu-devel/list/?series=100486
Fixes: 5cf0d326a0
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <155500034416.646888.1307366522340665522.stgit@bahia.lab.toulouse-stg.fr.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The PAPR PHB acts as a legacy PCI bus but it allows PCIe extended
config space accesses anyway (for pseries-2.9 and newer machine
types).
Introduce a specific PCI bus subtype to inform the common PCI code
about that.
Fixes: c2077e2ca0
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <155414130834.574858.16502276132110219890.stgit@bahia.lan>
[dwg: Apply fix so we don't rename the default pci bus, breaking everything]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
