Originally CPU threads were by default assigned in
round-robin fashion. However it was causing issues in
guest since CPU threads from the same socket/core could
be placed on different NUMA nodes.
Commit fb43b73b (pc: fix default VCPU to NUMA node mapping)
fixed it by grouping threads within a socket on the same node
introducing cpu_index_to_socket_id() callback and commit
20bb648d (spapr: Fix default NUMA node allocation for threads)
reused callback to fix similar issues for SPAPR machine
even though socket doesn't make much sense there.
As result QEMU ended up having 3 default distribution rules
used by 3 targets /virt-arm, spapr, pc/.
In effort of moving NUMA mapping for CPUs into possible_cpus,
generalize default mapping in numa.c by making boards decide
on default mapping and let them explicitly tell generic
numa code to which node a CPU thread belongs to by replacing
cpu_index_to_socket_id() with @cpu_index_to_instance_props()
which provides default node_id assigned by board to specified
cpu_index.
Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <1494415802-227633-2-git-send-email-imammedo@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
When there are more nodes than available memory to put the minimum
allowed memory by node, all the memory is put on the last node.
This is because we put (ram_size / nb_numa_nodes) &
~((1 << mc->numa_mem_align_shift) - 1); on each node, and in this
case the value is 0. This is particularly true with pseries,
as the memory must be aligned to 256MB.
To avoid this problem, this patch uses an error diffusion algorithm [1]
to distribute equally the memory on nodes.
We introduce numa_auto_assign_ram() function in MachineClass
to keep compatibility between machine type versions.
The legacy function is used with pseries-2.9, pc-q35-2.9 and
pc-i440fx-2.9 (and previous), the new one with all others.
Example:
qemu-system-ppc64 -S -nographic -nodefaults -monitor stdio -m 1G -smp 8 \
-numa node -numa node -numa node \
-numa node -numa node -numa node
Before:
(qemu) info numa
6 nodes
node 0 cpus: 0 6
node 0 size: 0 MB
node 1 cpus: 1 7
node 1 size: 0 MB
node 2 cpus: 2
node 2 size: 0 MB
node 3 cpus: 3
node 3 size: 0 MB
node 4 cpus: 4
node 4 size: 0 MB
node 5 cpus: 5
node 5 size: 1024 MB
After:
(qemu) info numa
6 nodes
node 0 cpus: 0 6
node 0 size: 0 MB
node 1 cpus: 1 7
node 1 size: 256 MB
node 2 cpus: 2
node 2 size: 0 MB
node 3 cpus: 3
node 3 size: 256 MB
node 4 cpus: 4
node 4 size: 256 MB
node 5 cpus: 5
node 5 size: 256 MB
[1] https://en.wikipedia.org/wiki/Error_diffusion
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Message-Id: <20170502162955.1610-2-lvivier@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
[ehabkost: s/ram_size/size/ at numa_default_auto_assign_ram()]
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
xics_system_init() does not need 'nr_servers' anymore as it is only
used to define the 'interrupt-controller' node in the device tree. So
let's just compute the value when calling spapr_dt_xics().
This also gives us an opportunity to simplify the xics_system_init()
routine and introduce a specific spapr_ics_create() helper to create
the sPAPR ICS object.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Today, all the ICPs are created before the CPUs, stored in an array
under the sPAPR machine and linked to the CPU when the core threads
are realized. This modeling brings some complexity when a lookup in
the array is required and it can be simplified by allocating the ICPs
when the CPUs are.
This is the purpose of this proposal which introduces a new 'icp_type'
field under the machine and creates the ICP objects of the right type
(KVM or not) before the PowerPCCPU object are.
This change allows more cleanups : the removal of the icps array under
the sPAPR machine and the removal of the xics_get_cpu_index_by_dt_id()
helper.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This is the second step to abstract the IRQ 'server' number of the
XICS layer. Now that the prereq cleanups have been done in the
previous patch, we can move down the 'cpu_dt_id' to 'cpu_index'
mapping in the sPAPR machine handler.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This enables in-kernel handling of H_PUT_TCE_INDIRECT and
H_STUFF_TCE hypercalls. The host kernel support is there since v4.6,
in particular d3695aa4f452
("KVM: PPC: Add support for multiple-TCE hcalls").
H_PUT_TCE is already accelerated and does not need any special enablement.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
For a little while around 4.9, Linux kernels that saw the radix bit in
ibm,pa-features would attempt to set up the MMU as if they were a
hypervisor, even if they were a guest, which would cause them to
crash.
Work around this by detecting pre-ISA 3.0 guests by their lack of that
bit in option vector 1, and then removing the radix bit from
ibm,pa-features. Note: This now requires regeneration of that node
after CAS negotiation.
Signed-off-by: Sam Bobroff <sam.bobroff@au1.ibm.com>
[dwg: Fix style nits]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Add the new node, /chosen/ibm,arch-vec-5-platform-support to the
device tree. This allows the guest to determine which modes are
supported by the hypervisor.
Update the option vector processing in h_client_architecture_support()
to handle the new MMU bits. This allows guests to request hash or
radix mode and QEMU to create the guest's HPT at this time if it is
necessary but hasn't yet been done. QEMU will terminate the guest if
it requests an unavailable mode, as required by the architecture.
Extend the ibm,pa-features node with the new ISA 3.0 values
and set the radix bit if KVM supports radix mode. This probably won't
be used directly by guests to determine the availability of radix mode
(that is indicated by the new node added above) but the architecture
requires that it be set when the hardware supports it.
If QEMU is using KVM, and KVM is capable of running in radix mode,
guests can be run in real-mode without allocating a HPT (because KVM
will use a minimal RPT). So in this case, we avoid creating the HPT
at reset time and later (during CAS) create it if it is necessary.
ISA 3.0 guests will now begin to call h_register_process_table(),
which has been added previously.
Signed-off-by: Sam Bobroff <sam.bobroff@au1.ibm.com>
[dwg: Strip some unneeded prefix from error messages]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
In the next patch, spapr_fixup_cpu_dt() will need to call
spapr_populate_pa_features() so move it's definition up without making
any other changes.
Signed-off-by: Sam Bobroff <sam.bobroff@au1.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The H_REGISTER_PROCESS_TABLE H_CALL is used by a guest to indicate to the
hypervisor where in memory its process table is and how translation should
be performed using this process table.
Provide the implementation of this H_CALL for a guest.
We first check for invalid flags, then parse the flags to determine the
operation, and then check the other parameters for valid values based on
the operation (register new table/deregister table/maintain registration).
The process table is then stored in the appropriate location and registered
with the hypervisor (if running under KVM), and the LPCR_[UPRT/GTSE] bits
are updated as required.
Signed-off-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com>
Signed-off-by: Sam Bobroff <sam.bobroff@au1.ibm.com>
[dwg: Correct missing prototype and uninitialized variable]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Use the new ioctl, KVM_PPC_GET_RMMU_INFO, to fetch radix MMU
information from KVM and present the page encodings in the device tree
under ibm,processor-radix-AP-encodings. This provides page size
information to the guest which is necessary for it to use radix mode.
Signed-off-by: Sam Bobroff <sam.bobroff@au1.ibm.com>
[dwg: Compile fix for 32-bit targets, style nit fix]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Also use an 'sPAPRRTCState' attribute under the sPAPR machine to hold
the RTC object. Overall, these changes remove an unnecessary and
implicit dependency on SysBus.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
For reasons that may be useful in future, CPU core objects, as used on the
pseries machine type have their own nr-threads property, potentially
allowing cores with different numbers of threads in the same system.
If the user/management uses the values specified in query-hotpluggable-cpus
as they're expected to do, this will never matter in pratice. But that's
not actually enforced - it's possible to manually specify a core with
a different number of threads from that in -smp. That will confuse the
platform - most immediately, this can be used to create a CPU thread with
index above max_cpus which leads to an assertion failure in
spapr_cpu_core_realize().
For now, enforce that all cores must have the same, standard, number of
threads.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Running postcopy-test with ASAN produces the following error:
QTEST_QEMU_BINARY=ppc64-softmmu/qemu-system-ppc64 tests/postcopy-test
...
=================================================================
==23641==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x7f1556600000 at pc 0x55b8e9d28208 bp 0x7f1555f4d3c0 sp 0x7f1555f4d3b0
READ of size 8 at 0x7f1556600000 thread T6
#0 0x55b8e9d28207 in htab_save_first_pass /home/elmarco/src/qq/hw/ppc/spapr.c:1528
#1 0x55b8e9d2939c in htab_save_iterate /home/elmarco/src/qq/hw/ppc/spapr.c:1665
#2 0x55b8e9beae3a in qemu_savevm_state_iterate /home/elmarco/src/qq/migration/savevm.c:1044
#3 0x55b8ea677733 in migration_thread /home/elmarco/src/qq/migration/migration.c:1976
#4 0x7f15845f46c9 in start_thread (/lib64/libpthread.so.0+0x76c9)
#5 0x7f157d9d0f7e in clone (/lib64/libc.so.6+0x107f7e)
0x7f1556600000 is located 0 bytes to the right of 2097152-byte region [0x7f1556400000,0x7f1556600000)
allocated by thread T0 here:
#0 0x7f159bb76980 in posix_memalign (/lib64/libasan.so.3+0xc7980)
#1 0x55b8eab185b2 in qemu_try_memalign /home/elmarco/src/qq/util/oslib-posix.c:106
#2 0x55b8eab186c8 in qemu_memalign /home/elmarco/src/qq/util/oslib-posix.c:122
#3 0x55b8e9d268a8 in spapr_reallocate_hpt /home/elmarco/src/qq/hw/ppc/spapr.c:1214
#4 0x55b8e9d26e04 in ppc_spapr_reset /home/elmarco/src/qq/hw/ppc/spapr.c:1261
#5 0x55b8ea12e913 in qemu_system_reset /home/elmarco/src/qq/vl.c:1697
#6 0x55b8ea13fa40 in main /home/elmarco/src/qq/vl.c:4679
#7 0x7f157d8e9400 in __libc_start_main (/lib64/libc.so.6+0x20400)
Thread T6 created by T0 here:
#0 0x7f159bae0488 in __interceptor_pthread_create (/lib64/libasan.so.3+0x31488)
#1 0x55b8eab1d9cb in qemu_thread_create /home/elmarco/src/qq/util/qemu-thread-posix.c:465
#2 0x55b8ea67874c in migrate_fd_connect /home/elmarco/src/qq/migration/migration.c:2096
#3 0x55b8ea66cbb0 in migration_channel_connect /home/elmarco/src/qq/migration/migration.c:500
#4 0x55b8ea678f38 in socket_outgoing_migration /home/elmarco/src/qq/migration/socket.c:87
#5 0x55b8eaa5a03a in qio_task_complete /home/elmarco/src/qq/io/task.c:142
#6 0x55b8eaa599cc in gio_task_thread_result /home/elmarco/src/qq/io/task.c:88
#7 0x7f15823e38e6 (/lib64/libglib-2.0.so.0+0x468e6)
SUMMARY: AddressSanitizer: heap-buffer-overflow /home/elmarco/src/qq/hw/ppc/spapr.c:1528 in htab_save_first_pass
index seems to be wrongly incremented, unless I miss something that
would be worth a comment.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Since commit 224245b ("spapr: Add LMB DR connectors"), NUMA node
memory size must be aligned to 256MB (SPAPR_MEMORY_BLOCK_SIZE).
But when "-numa" option is provided without "mem" parameter,
the memory is equally divided between nodes, but 8MB aligned.
This can be not valid for pseries.
In that case we can have:
$ ./ppc64-softmmu/qemu-system-ppc64 -m 4G -numa node -numa node -numa node
qemu-system-ppc64: Node 0 memory size 0x55000000 is not aligned to 256 MiB
With this patch, we have:
(qemu) info numa
3 nodes
node 0 cpus: 0
node 0 size: 1280 MB
node 1 cpus:
node 1 size: 1280 MB
node 2 cpus:
node 2 size: 1536 MB
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
bb9986452 "spapr_pci: Advertise access to PCIe extended config space"
allowed guests to access the extended config space of PCI Express devices
via the PAPR interfaces, even though the paravirtualized bus mostly acts
like plain PCI.
However, that patch enabled access unconditionally, including for existing
machine types, which is an unwise change in behaviour. This patch limits
the change to pseries-2.9 (and later) machine types.
Suggested-by: Andrea Bolognani <abologna@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The recent changes on the XICS layer removed the XICSState object to
let the sPAPR machine handle the ICP and ICS directly. The reset of
these objects was previously handled by XICSState, which was a SysBus
device, and to keep the same behavior, the ICP and ICS were assigned
to SysbBus.
But that broke the 'info qtree' command in the monitor. 'qtree'
performs a loop on the children of a bus to print their properties and
SysBus devices are expected to be found under SysBus, which is not the
case anymore.
The fix for this problem is to register reset handlers for the ICP and
ICS objects and stop using SysBus for such devices.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Tested-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The PPC MMU types are sometimes treated as if they were a bit field
and sometime as if they were an enum which causes maintenance
problems: flipping bits in the MMU type (which is done on both the 1TB
segment and 64K segment bits) currently produces new MMU type
values that are not handled in every "switch" on it, sometimes causing
an abort().
This patch provides some macros that can be used to filter out the
"bit field-like" bits so that the remainder of the value can be
switched on, like an enum. This allows removal of all of the
"degraded" types from the list and should ease maintenance.
Signed-off-by: Sam Bobroff <sam.bobroff@au1.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Add a pa-features definition which includes all of the new fields which
have been added, note we don't claim support for any of these new features
at this stage.
Signed-off-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
ISA v3.00 adds the idea of a partition table which is used to store the
address translation details for all partitions on the system. The partition
table consists of double word entries indexed by partition id where the second
double word contains the location of the process table in guest memory. The
process table is registered by the guest via a h-call.
We need somewhere to store the address of the process table so we add an entry
to the sPAPRMachineState struct called patb_entry to represent the second
doubleword of a single partition table entry corresponding to the current
guest. We need to store this value so we know if the guest is using radix or
hash translation and the location of the corresponding process table in guest
memory. Since we only have a single guest per qemu instance, we only need one
entry.
Since the partition table is technically a hypervisor resource we require that
access to it is abstracted by the virtual hypervisor through the get_patbe()
call. Currently the value of the entry is never set (and thus
defaults to 0 indicating hash), but it will be required to both implement
POWER9 kvm support and tcg radix support.
We also add this field to be migrated as part of the sPAPRMachineState as we
will need it on the receiving side as the guest will never tell us this
information again and we need it to perform translation.
Signed-off-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
It provides a better monitor output of the ICP and ICS objects, else
the objects are printed out of order.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The ICS object uses a post_load() handler which is implicitly relying
on the fact that the internal state of the ICS and ICP objects has
been restored but this is not guaranteed. So, let's move the code
under the post_load() handler of the machine where we know the objects
have been fully restored.
The icp_resend() handler of the XICSFabric QOM interface is also
removed as it is now obsolete.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The XICSState classes are not used anymore. They have now been fully
deprecated by the XICSFabric QOM interface. Do the cleanups.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
There is nothing left related to the XICS object in the realize
functions of the KVMXICSState and XICSState class. So adapt the
interfaces to call these routines directly from the sPAPR machine init
sequence.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This is the last step to remove the XICSState abstraction and have the
machine hold all the objects related to interrupts : ICSs and ICPs.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The reset of the ICP objects is currently handled by XICS but this can
be done for each individual ICP.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
spapr_dt_xics() only needs the number of servers to build the device
tree nodes. Let's change the routine interface to reflect that.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Also introduce a xics_icp_get() helper to simplify the changes.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Let's add two new handlers for ICPs. One is to get an ICP object from
a server number and a second is to resend the irqs when needed.
The icp_resend() handler is a temporary workaround needed by the
ics-simple post_load() handler. It will be removed when the post_load
portion can be done at the machine level.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This is not used anymore.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The reset of the ICS objects is currently handled by XICS but this can
be done for each individual ICS. This also reduces the use of the XICS
list of ICS.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Also change the ICPState 'xics' backlink to be a XICSFabric, this
removes the need of using qdev_get_machine() to get the QOM interface
in some of the routines.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Add 'ics_get' and 'ics_resend' handlers to the sPAPR machine. These
are relatively simple for a single ICS.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
A list of ICS objects was introduced under the XICS object for the
PowerNV machine but, for the sPAPR machine, it brings extra complexity
as there is only a single ICS. To simplify the code, let's add the ICS
pointer under the sPAPR machine and try to reduce the use of this list
where possible.
Also, change the xics_spapr_*() routines to use an ICS object instead
of an XICSState and change their name to reflect that these are
specific to the sPAPR ICS object.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Today, the ICP (Interrupt Controller Presenter) objects are created by
the 'nr_servers' property handler of the XICS object and a class
handler. They are realized in the XICS object realize routine.
Let's simplify the process by creating the ICP objects along with the
XICS object at the machine level.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Today, the ICS (Interrupt Controller Source) object is created and
realized by the init and realize routines of the XICS object, but some
of the parameters are only known at the machine level.
These parameters are passed from the sPAPR machine to the ICS object
in a rather convoluted way using property handlers and a class handler
of the XICS object. The number of irqs required to allocate the IRQ
state objects in the ICS realize routine is one of them.
Let's simplify the process by creating the ICS object along with the
XICS object at the machine level and link the ICS into the XICS list
of ICSs at this level also. In the sPAPR machine, there is only a
single ICS but that will change with the PowerNV machine.
Also, QOMify the creation of the objects and get rid of the
superfluous code.
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Currently xics - the component of the IBM POWER interrupt controller
representing the overall interrupt fabric / architecture is
represented as a descendent of SysBusDevice. However, this is not
really correct - the xics presents nothing in MMIO space so it should
be an "unattached" device in the current QOM model.
Since this device will always be created by the machine type, not created
specifically from the command line, and because it has no migrated state
it should be safe to move it around the device composition tree.
Therefore this patch changes it to a descendent of TYPE_DEVICE, and
makes it an unattached device. So that its reset handler still gets
called correctly, we add a qdev_set_parent_bus() to attach it to
sysbus. It's not really clear that's correct (instead of using
register_reset()) but it appears to a common technique.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
[clg corrected problems with reset]
Signed-off-by: Cédric Le Goater <clg@kaod.org>
[dwg folded together and updated commit message]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The pseries machine type implements the behaviour of a PAPR compliant
hypervisor, without actually executing such a hypervisor on the virtual
CPU. To do this we need some hooks in the CPU code to make hypervisor
facilities get redirected to the machine instead of emulated internally.
For hypercalls this is managed through the cpu->vhyp field, which points
to a QOM interface with a method implementing the hypercall.
For the hashed page table (HPT) - also a hypervisor resource - we use an
older hack. CPUPPCState has an 'external_htab' field which when non-NULL
indicates that the HPT is stored in qemu memory, rather than within the
guest's address space.
For consistency - and to make some future extensions easier - this merges
the external HPT mechanism into the vhyp mechanism. Methods are added
to vhyp for the basic operations the core hash MMU code needs: map_hptes()
and unmap_hptes() for reading the HPT, store_hpte() for updating it and
hpt_mask() to retrieve its size.
To match this, the pseries machine now sets these vhyp fields in its
existing vhyp class, rather than reaching into the cpu object to set the
external_htab field.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com>
Some systems can already provide more than 255 hardware threads.
Bumping the QEMU limit to 1024 seems reasonable:
- it has no visible overhead in top;
- the limit itself has no effect on hot paths.
Cc: Greg Kurz <gkurz@linux.vnet.ibm.com>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This finally allows TCG to benefit from the iothread introduction: Drop
the global mutex while running pure TCG CPU code. Reacquire the lock
when entering MMIO or PIO emulation, or when leaving the TCG loop.
We have to revert a few optimization for the current TCG threading
model, namely kicking the TCG thread in qemu_mutex_lock_iothread and not
kicking it in qemu_cpu_kick. We also need to disable RAM block
reordering until we have a more efficient locking mechanism at hand.
Still, a Linux x86 UP guest and my Musicpal ARM model boot fine here.
These numbers demonstrate where we gain something:
20338 jan 20 0 331m 75m 6904 R 99 0.9 0:50.95 qemu-system-arm
20337 jan 20 0 331m 75m 6904 S 20 0.9 0:26.50 qemu-system-arm
The guest CPU was fully loaded, but the iothread could still run mostly
independent on a second core. Without the patch we don't get beyond
32206 jan 20 0 330m 73m 7036 R 82 0.9 1:06.00 qemu-system-arm
32204 jan 20 0 330m 73m 7036 S 21 0.9 0:17.03 qemu-system-arm
We don't benefit significantly, though, when the guest is not fully
loading a host CPU.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Message-Id: <1439220437-23957-10-git-send-email-fred.konrad@greensocs.com>
[FK: Rebase, fix qemu_devices_reset deadlock, rm address_space_* mutex]
Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
[EGC: fixed iothread lock for cpu-exec IRQ handling]
Signed-off-by: Emilio G. Cota <cota@braap.org>
[AJB: -smp single-threaded fix, clean commit msg, BQL fixes]
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Reviewed-by: Pranith Kumar <bobby.prani@gmail.com>
[PM: target-arm changes]
Acked-by: Peter Maydell <peter.maydell@linaro.org>
On POWER, the valid page sizes that the guest can use are bound
to the CPU and not to the memory region. QEMU already has some
fancy logic to find out the right maximum memory size to tell
it to the guest during boot (see getrampagesize() in the file
target/ppc/kvm.c for more information).
However, once we're booted and the guest is using huge pages
already, it is currently still possible to hot-plug memory regions
that does not support huge pages - which of course does not work
on POWER, since the guest thinks that it is possible to use huge
pages everywhere. The KVM_RUN ioctl will then abort with -EFAULT,
QEMU spills out a not very helpful error message together with
a register dump and the user is annoyed that the VM unexpectedly
died.
To avoid this situation, we should check the page size of hot-plugged
DIMMs to see whether it is possible to use it in the current VM.
If it does not fit, we can print out a better error message and
refuse to add it, so that the VM does not die unexpectely and the
user has a second chance to plug a DIMM with a matching memory
backend instead.
Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=1419466
Signed-off-by: Thomas Huth <thuth@redhat.com>
[dwg: Fix a build error on 32-bit builds with KVM]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Generic helper machine_query_hotpluggable_cpus() replaced
target specific query_hotpluggable_cpus() callbacks so
there is no need in it anymore. However inon NULL callback
value is used to detect/report hotpluggable cpus support,
therefore it can be removed completely.
Replace it with MachineClass.has_hotpluggable_cpus boolean
which is sufficient for the task.
Suggested-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
All callbacks FOO_query_hotpluggable_cpus() are practically
the same except of setting vcpus_count to different values.
Convert them to a generic machine_query_hotpluggable_cpus()
callback by moving vcpus_count initialization to per machine
specific callback possible_cpu_arch_ids().
Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Replace SPAPR specific cores[] array with generic
machine->possible_cpus and store core objects there.
It makes cores bookkeeping similar to x86 cpus and
will allow to unify similar code.
It would allow to replace cpu_index based NUMA node
mapping with iproperty based one (for -device created
cores) since possible_cpus carries board defined
topology/layout.
Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Acked-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
spapr_core_unplug() were essentially spapr_core_unplug_request()
handler that requested CPU removal and registered callback
which did actual cpu core removali but it was called from
spapr_machine_device_unplug() which is intended for actual object
removal. Commit (cf632463 spapr: Memory hot-unplug support)
sort of fixed it introducing spapr_machine_device_unplug_request()
and calling spapr_core_unplug() but it hasn't renamed callback and
by mistake calls it from spapr_machine_device_unplug().
However spapr_machine_device_unplug() isn't ever called for
cpu core since spapr_core_release() doesn't follow expected
hotunplug call flow which is:
1: device_del() ->
hotplug_handler_unplug_request() ->
set destroy_cb()
2: destroy_cb() ->
hotplug_handler_unplug() ->
object_unparent // actual device removal
Fix it by renaming spapr_core_unplug() to spapr_core_unplug_request()
which is called from spapr_machine_device_unplug_request() and
making spapr_core_release() call hotplug_handler_unplug() which
will call spapr_machine_device_unplug() -> spapr_core_unplug()
to remove cpu core.
Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Reveiwed-by: Bharata B Rao <bharata@linux.vnet.ibm.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
spapr_core_pre_plug/spapr_core_plug/spapr_core_unplug() are managing
wiring CPU core into spapr machine state and not internal CPU core state.
So move them from spapr_cpu_core.c to spapr.c where other similar
(spapr_memory_[foo]plug()) callbacks are located, which also matches
x86 target practice.
Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
We are switching BUILD_BUG_ON to verify that it's parameter is a
compile-time constant, and it turns out that some gcc versions
(specifically gcc (Ubuntu 5.4.0-6ubuntu1~16.04.4) 5.4.0 20160609) are
not smart enough to figure it out for expressions involving local
variables. This is harmless but means that the check is ineffective for
these platforms. To fix, replace the variable with macros.
Reported-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
[dwg: Correct a printf format warning]
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
This is a port to ppc of the i386 commit:
00f4d64 kvmclock: clock should count only if vm is running
We remove timebase_post_load function, and use the VM state
change handler to save and restore the guest_timebase (on stop
and continue).
We keep timebase_pre_save to reduce the clock difference on
migration like in:
6053a86 kvmclock: reduce kvmclock difference on migration
Time base offset has originally been introduced by commit
98a8b52 spapr: Add support for time base offset migration
So while VM is paused, the time is stopped. This allows to have
the same result with date (based on Time Base Register) and
hwclock (based on "get-time-of-day" RTAS call).
Moreover in TCG mode, the Time Base is always paused, so this
patch also adjust the behavior between TCG and KVM.
VM state field "time_of_the_day_ns" is now useless but we keep
it to be able to migrate to older version of the machine.
As vmstate_ppc_timebase structure (with timebase_pre_save() and
timebase_post_load() functions) was only used by vmstate_spapr,
we register the VM state change handler only in ppc_spapr_init().
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>