memory: Share FlatView's and dispatch trees between address spaces
This allows sharing flat views between address spaces (AS) when
the same root memory region is used when creating a new address space.
This is done by walking through all ASes and caching one FlatView per
a physical root MR (i.e. not aliased).
This removes search for duplicates from address_space_init_shareable() as
FlatViews are shared elsewhere and keeping as::ref_count correct seems
an unnecessary and useless complication.
This should cause no change and memory use or boot time yet.
Address spaces get to keep a root MR (alias or not) but FlatView stores
the actual MR as this is going to be used later on to decide whether to
share a particular FlatView or not.
memory: Move AddressSpaceDispatch from AddressSpace to FlatView
As we are going to share FlatView's between AddressSpace's,
and AddressSpaceDispatch is a structure to perform quick lookup
in FlatView, this moves ASD to FlatView.
After previosly open coded ASD rendering, we can also remove
as->next_dispatch as the new FlatView pointer is stored
on a stack and set to an AS atomically.
flatview_destroy() is executed under RCU instead of
address_space_dispatch_free() now.
This makes mem_begin/mem_commit to work with ASD and mem_add with FV
as later on mem_add will be taking FV as an argument anyway.
This moves a FlatView allocation and initialization to a helper.
While we are nere, replace g_new with g_new0 to not to bother if we add
new fields in the future.
We are going to share FlatView's between AddressSpace's and per-AS
memory listeners won't suit the purpose anymore so open code
the dispatch tree rendering.
Since there is a good chance that dispatch_listener was the only
listener, this avoids address_space_update_topology_pass() if there is
no registered listeners; this should improve starting time.
Paolo Bonzini [Thu, 21 Sep 2017 12:32:47 +0000 (14:32 +0200)]
memory: avoid "resurrection" of dead FlatViews
It's possible for address_space_get_flatview() as it currently stands
to cause a use-after-free for the returned FlatView, if the reference
count is incremented after the FlatView has been replaced by a writer:
Since FlatViews are not updated very often, we can just detect the
situation using a new atomic op atomic_fetch_inc_nonzero, similar to
Linux's atomic_inc_not_zero, which performs the refcount increment only if
it hasn't already hit zero. This is similar to Linux commit de09a9771a53
("CRED: Fix get_task_cred() and task_state() to not resurrect dead
credentials", 2010-07-29).
Peter Maydell [Thu, 21 Sep 2017 16:42:27 +0000 (17:42 +0100)]
Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20170921' into staging
target-arm queue:
* more preparatory work for v8M support
* convert some omap devices away from old_mmio
* remove out of date ARM ARM section references in comments
* add the Smartfusion2 board
* remotes/pmaydell/tags/pull-target-arm-20170921: (31 commits)
msf2: Add Emcraft's Smartfusion2 SOM kit
msf2: Add Smartfusion2 SoC
msf2: Add Smartfusion2 SPI controller
msf2: Microsemi Smartfusion2 System Register block
msf2: Add Smartfusion2 System timer
hw/arm/omap2.c: Don't use old_mmio
hw/i2c/omap_i2c.c: Don't use old_mmio
hw/timer/omap_gptimer: Don't use old_mmio
hw/timer/omap_synctimer.c: Don't use old_mmio
hw/gpio/omap_gpio.c: Don't use old_mmio
hw/arm/palm.c: Don't use old_mmio for static_ops
target/arm: Remove out of date ARM ARM section references in A64 decoder
nvic: Support banked exceptions in acknowledge and complete
nvic: Make SHCSR banked for v8M
nvic: Make ICSR banked for v8M
target/arm: Handle banking in negative-execution-priority check in cpu_mmu_index()
nvic: Handle v8M changes in nvic_exec_prio()
nvic: Disable the non-secure HardFault if AIRCR.BFHFNMINS is clear
nvic: Implement v8M changes to fixed priority exceptions
nvic: In escalation to HardFault, support HF not being priority -1
...
Peter Maydell [Fri, 15 Sep 2017 15:08:49 +0000 (16:08 +0100)]
target/arm: Remove out of date ARM ARM section references in A64 decoder
In the A64 decoder, we have a lot of references to section numbers
from version A.a of the v8A ARM ARM (DDI0487). This version of the
document is now long obsolete (we are currently on revision B.a),
and various intervening versions renumbered all the sections.
The most recent B.a version of the document doesn't assign
section numbers at all to the individual instruction classes
in the way that the various A.x versions did. The simplest thing
to do is just to delete all the out of date C.x.x references.
Peter Maydell [Tue, 12 Sep 2017 18:14:06 +0000 (19:14 +0100)]
nvic: Support banked exceptions in acknowledge and complete
Update armv7m_nvic_acknowledge_irq() and armv7m_nvic_complete_irq()
to handle banked exceptions:
* acknowledge needs to use the correct vector, which may be
in sec_vectors[]
* acknowledge needs to return to its caller whether the
exception should be taken to secure or non-secure state
* complete needs its caller to tell it whether the exception
being completed is a secure one or not
Peter Maydell [Tue, 12 Sep 2017 18:14:04 +0000 (19:14 +0100)]
nvic: Make ICSR banked for v8M
The ICSR NVIC register is banked for v8M. This doesn't
require any new state, but it does mean that some bits
are controlled by BFHNFNMINS and some bits must work
with the correct banked exception. There is also a new
in v8M PENDNMICLR bit.
Peter Maydell [Tue, 12 Sep 2017 18:14:03 +0000 (19:14 +0100)]
target/arm: Handle banking in negative-execution-priority check in cpu_mmu_index()
Now that we have a banked FAULTMASK register and banked exceptions,
we can implement the correct check in cpu_mmu_index() for whether
the MPU_CTRL.HFNMIENA bit's effect should apply. This bit causes
handlers which have requested a negative execution priority to run
with the MPU disabled. In v8M the test has to check this for the
current security state and so takes account of banking.
Peter Maydell [Tue, 12 Sep 2017 18:14:02 +0000 (19:14 +0100)]
nvic: Handle v8M changes in nvic_exec_prio()
Update nvic_exec_prio() to support the v8M changes:
* BASEPRI, FAULTMASK and PRIMASK are all banked
* AIRCR.PRIS can affect NS priorities
* AIRCR.BFHFNMINS affects FAULTMASK behaviour
These changes mean that it's no longer possible to
definitely say that if FAULTMASK is set it overrides
PRIMASK, and if PRIMASK is set it overrides BASEPRI
(since if PRIMASK_NS is set and AIRCR.PRIS is set then
whether that 0x80 priority should take effect or the
priority in BASEPRI_S depends on the value of BASEPRI_S,
for instance). So we switch to the same approach used
by the pseudocode of working through BASEPRI, PRIMASK
and FAULTMASK and overriding the previous values if
needed.
Peter Maydell [Tue, 12 Sep 2017 18:14:01 +0000 (19:14 +0100)]
nvic: Disable the non-secure HardFault if AIRCR.BFHFNMINS is clear
If AIRCR.BFHFNMINS is clear, then although NonSecure HardFault
can still be pended via SHCSR.HARDFAULTPENDED it mustn't actually
preempt execution. The simple way to achieve this is to clear the
enable bit for it, since the enable bit isn't guest visible.
Peter Maydell [Tue, 12 Sep 2017 18:14:00 +0000 (19:14 +0100)]
nvic: Implement v8M changes to fixed priority exceptions
In v7M, the fixed-priority exceptions are:
Reset: -3
NMI: -2
HardFault: -1
In v8M, this changes because Secure HardFault may need
to be prioritised above NMI:
Reset: -4
Secure HardFault if AIRCR.BFHFNMINS == 1: -3
NMI: -2
Secure HardFault if AIRCR.BFHFNMINS == 0: -1
NonSecure HardFault: -1
Make these changes, including support for changing the
priority of Secure HardFault as AIRCR.BFHFNMINS changes.
Peter Maydell [Tue, 12 Sep 2017 18:13:59 +0000 (19:13 +0100)]
nvic: In escalation to HardFault, support HF not being priority -1
When escalating to HardFault, we must go into Lockup if we
can't take the synchronous HardFault because the current
execution priority is already at or below the priority of
HardFault. In v7M HF is always priority -1 so a simple < 0
comparison sufficed; in v8M the priority of HardFault can
vary depending on whether it is a Secure or NonSecure
HardFault, so we must check against the priority of the
HardFault exception vector we're about to use.
Peter Maydell [Tue, 12 Sep 2017 18:13:58 +0000 (19:13 +0100)]
nvic: Compare group priority for escalation to HF
In armv7m_nvic_set_pending() we have to compare the
priority of an exception against the execution priority
to decide whether it needs to be escalated to HardFault.
In the specification this is a comparison against the
exception's group priority; for v7M we implemented it
as a comparison against the raw exception priority
because the two comparisons will always give the
same answer. For v8M the existence of AIRCR.PRIS and
the possibility of different PRIGROUP values for secure
and nonsecure exceptions means we need to explicitly
calculate the vector's group priority for this check.
Peter Maydell [Tue, 12 Sep 2017 18:13:57 +0000 (19:13 +0100)]
nvic: Make SHPR registers banked
Make the set_prio() function take a bool indicating
whether to pend the secure or non-secure version of a banked
interrupt, and use this to implement the correct banking
semantics for the SHPR registers.
Peter Maydell [Tue, 12 Sep 2017 18:13:56 +0000 (19:13 +0100)]
nvic: Make set_pending and clear_pending take a secure parameter
Make the armv7m_nvic_set_pending() and armv7m_nvic_clear_pending()
functions take a bool indicating whether to pend the secure
or non-secure version of a banked interrupt, and update the
callsites accordingly.
In most callsites we can simply pass the correct security
state in; in a couple of cases we use TODO comments to indicate
that we will return the code in a subsequent commit.
Peter Maydell [Tue, 12 Sep 2017 18:13:54 +0000 (19:13 +0100)]
nvic: Implement NVIC_ITNS<n> registers
For v8M, the NVIC has a new set of registers per interrupt,
NVIC_ITNS<n>. These determine whether the interrupt targets Secure
or Non-secure state. Implement the register read/write code for
these, and make them cause NVIC_IABR, NVIC_ICER, NVIC_ISER,
NVIC_ICPR, NVIC_IPR and NVIC_ISPR to RAZ/WI for non-secure
accesses to fields corresponding to interrupts which are
configured to target secure state.
Peter Maydell [Tue, 12 Sep 2017 18:13:52 +0000 (19:13 +0100)]
nvic: Implement AIRCR changes for v8M
The Application Interrupt and Reset Control Register has some changes
for v8M:
* new bits SYSRESETREQS, BFHFNMINS and PRIS: these all have
real state if the security extension is implemented and otherwise
are constant
* the PRIGROUP field is banked between security states
* non-secure code can be blocked from using the SYSRESET bit
to reset the system if SYSRESETREQS is set
Implement the new state and the changes to register read and write.
For the moment we ignore the effects of the secure PRIGROUP.
We will implement the effects of PRIS and BFHFNMIS later.
Peter Maydell [Tue, 12 Sep 2017 18:13:51 +0000 (19:13 +0100)]
nvic: Add cached vectpending_prio state
Instead of looking up the pending priority
in nvic_pending_prio(), cache it in a new state struct
field. The calculation of the pending priority given
the interrupt number is more complicated in v8M with
the security extension, so the caching will be worthwhile.
This changes nvic_pending_prio() from returning a full
(group + subpriority) priority value to returning a group
priority. This doesn't require changes to its callsites
because we use it only in comparisons of the form
execution_prio > nvic_pending_prio()
and execution priority is always a group priority, so
a test (exec prio > full prio) is true if and only if
(execprio > group_prio).
(Architecturally the expected comparison is with the
group priority for this sort of "would we preempt" test;
we were only doing a test with a full priority as an
optimisation to avoid the mask, which is possible
precisely because the two comparisons always give the
same answer.)
Peter Maydell [Tue, 12 Sep 2017 18:13:50 +0000 (19:13 +0100)]
nvic: Add cached vectpending_is_s_banked state
With banked exceptions, just the exception number in
s->vectpending is no longer sufficient to uniquely identify
the pending exception. Add a vectpending_is_s_banked bool
which is true if the exception is using the sec_vectors[]
array.
Peter Maydell [Thu, 21 Sep 2017 15:28:59 +0000 (16:28 +0100)]
nvic: Add banked exception states
For the v8M security extension, some exceptions must be banked
between security states. Add the new vecinfo array which holds
the state for the banked exceptions and migrate it if the
CPU the NVIC is attached to implements the security extension.
Peter Maydell [Tue, 12 Sep 2017 18:13:48 +0000 (19:13 +0100)]
target/arm: Implement MSR/MRS access to NS banked registers
In v8M the MSR and MRS instructions have extra register value
encodings to allow secure code to access the non-secure banked
version of various special registers.
(We don't implement the MSPLIM_NS or PSPLIM_NS aliases, because
we don't currently implement the stack limit registers at all.)
Peter Maydell [Thu, 21 Sep 2017 13:40:30 +0000 (14:40 +0100)]
Merge remote-tracking branch 'remotes/yongbok/tags/mips-20170921' into staging
MIPS patches 2017-09-21
Changes:
QOMify MIPS cpu
Improve macro parenthesization
# gpg: Signature made Thu 21 Sep 2017 13:50:37 BST
# gpg: using RSA key 0x2238EB86D5F797C2
# gpg: Good signature from "Yongbok Kim <[email protected]>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg: It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 8600 4CF5 3415 A5D9 4CFA 2B5C 2238 EB86 D5F7 97C2
* remotes/yongbok/tags/mips-20170921:
mips: Improve macro parenthesization
mips: replace cpu_mips_init() with cpu_generic_init()
mips: MIPSCPU model subclasses
mips: call cpu_mips_realize_env() from mips_cpu_realizefn()
mips: split cpu_mips_realize_env() out of cpu_mips_init()
mips: introduce internal.h and cleanup cpu.h
mips: move hw/mips/cputimer.c to target/mips/
Eric Blake [Tue, 19 Sep 2017 14:13:07 +0000 (09:13 -0500)]
mips: Improve macro parenthesization
Although none of the existing macro call-sites were broken,
it's always better to write macros that properly parenthesize
arguments that can be complex expressions, so that the intended
order of operations is not broken.
Igor Mammedov [Wed, 20 Sep 2017 19:49:33 +0000 (16:49 -0300)]
mips: MIPSCPU model subclasses
Register separate QOM types for each mips cpu model,
so it would be possible to reuse generic CPU creation
routines.
Signed-off-by: Igor Mammedov <[email protected]> Signed-off-by: Philippe Mathieu-Daudé <[email protected]>
[PMD: use internal.h, use void* to hold cpu_def in MIPSCPUClass,
mark MIPSCPU abstract, address Eduardo Habkost review] Tested-by: James Hogan <[email protected]> Reviewed-by: Eduardo Habkost <[email protected]> Signed-off-by: Yongbok Kim <[email protected]>
mips: call cpu_mips_realize_env() from mips_cpu_realizefn()
This changes the order between cpu_mips_realize_env() and
cpu_exec_initfn(), but cpu_exec_initfn() don't have anything that
depends on cpu_mips_realize_env() being called first.
This timer is a required part of the MIPS32/MIPS64 System Control coprocessor
(CP0). Moving it with the other architecture related files will allow an opaque
use of CPUMIPSState* in the next commit (introduce "internal.h").
also remove it from 'user' targets, remove an unnecessary include.
kvm: drop wrong assertion creating problems with pflash
pflash toggles mr->romd_mode. So this assert does not always hold.
1) a device was added with !mr->romd_mode, therefore effectively not
creating a kvm slot as we want to trap every access (add = false).
2) mr->romd_mode was toggled on before remove it. There is now
actually no slot to remove and the assert is wrong.
Pavel Butsykin [Tue, 19 Sep 2017 12:07:33 +0000 (15:07 +0300)]
virtio-serial: add enable_backend callback
We should guarantee that RAM will not be modified while VM has a stopped
state, otherwise it can lead to negative consequences during post-copy
migration. In RUN_STATE_FINISH_MIGRATE step, it's expected that RAM on
source side will not be modified as this could lead to non-consistent vm state
on the destination side. Also RAM access during postcopy-ram migration with
enabled release-ram capability can lead to sad consequences.
Let's add enable_backend() callback to avoid undesirable virtioqueue changes
in the guest memory.
Roger Pau Monne [Thu, 24 Aug 2017 15:07:03 +0000 (16:07 +0100)]
xen/pt: allow QEMU to request MSI unmasking at bind time
When a MSI interrupt is bound to a guest using
xc_domain_update_msi_irq (XEN_DOMCTL_bind_pt_irq) the interrupt is
left masked by default.
This causes problems with guests that first configure interrupts and
clean the per-entry MSIX table mask bit and afterwards enable MSIX
globally. In such scenario the Xen internal msixtbl handlers would not
detect the unmasking of MSIX entries because vectors are not yet
registered since MSIX is not enabled, and vectors would be left
masked.
Introduce a new flag in the gflags field to signal Xen whether a MSI
interrupt should be unmasked after being bound.
This also requires to track the mask register for MSI interrupts, so
QEMU can also notify to Xen whether the MSI interrupt should be bound
masked or unmasked
Olaf Hering [Fri, 28 Jul 2017 13:11:51 +0000 (15:11 +0200)]
xen-disk: use g_new0 to fix build
g_malloc0_n is available since glib-2.24. To allow build with older glib
versions use the generic g_new0, which is already used in many other
places in the code.
Fixes commit 3284fad728 ("xen-disk: add support for multi-page shared rings")
Peter Maydell [Wed, 20 Sep 2017 19:33:48 +0000 (20:33 +0100)]
Merge remote-tracking branch 'remotes/gkurz/tags/for-upstream' into staging
These patches fix regressions in 2.10
# gpg: Signature made Wed 20 Sep 2017 07:51:07 BST
# gpg: using DSA key 0x02FC3AEB0101DBC2
# gpg: Good signature from "Greg Kurz <[email protected]>"
# gpg: aka "Greg Kurz <[email protected]>"
# gpg: aka "Greg Kurz <[email protected]>"
# gpg: aka "Gregory Kurz (Groug) <[email protected]>"
# gpg: aka "[jpeg image of size 3330]"
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg: There is no indication that the signature belongs to the owner.
# Primary key fingerprint: 2BD4 3B44 535E C0A7 9894 DBA2 02FC 3AEB 0101 DBC2
* remotes/gkurz/tags/for-upstream:
9pfs: check the size of transport buffer before marshaling
9pfs: fix name_to_path assertion in v9fs_complete_rename()
9pfs: fix readdir() for 9p2000.u
Peter Maydell [Wed, 20 Sep 2017 16:35:36 +0000 (17:35 +0100)]
Merge remote-tracking branch 'remotes/ehabkost/tags/machine-next-pull-request' into staging
Machine/CPU/NUMA queue, 2017-09-19
# gpg: Signature made Tue 19 Sep 2017 21:17:01 BST
# gpg: using RSA key 0x2807936F984DC5A6
# gpg: Good signature from "Eduardo Habkost <[email protected]>"
# Primary key fingerprint: 5A32 2FD5 ABC4 D3DB ACCF D1AA 2807 936F 984D C5A6
* remotes/ehabkost/tags/machine-next-pull-request:
MAINTAINERS: Update git URLs for my trees
hw/acpi-build: Fix SRAT memory building in case of node 0 without RAM
NUMA: Replace MAX_NODES with nb_numa_nodes in for loop
numa: cpu: calculate/set default node-ids after all -numa CLI options are parsed
arm: drop intermediate cpu_model -> cpu type parsing and use cpu type directly
pc: use generic cpu_model parsing
vl.c: convert cpu_model to cpu type and set of global properties before machine_init()
cpu: make cpu_generic_init() abort QEMU on error
qom: cpus: split cpu_generic_init() on feature parsing and cpu creation parts
hostmem-file: Add "discard-data" option
osdep: Define QEMU_MADV_REMOVE
vl: Clean up user-creatable objects when exiting
Jan Dakinevich [Wed, 20 Sep 2017 06:48:52 +0000 (08:48 +0200)]
9pfs: check the size of transport buffer before marshaling
v9fs_do_readdir_with_stat() should check for a maximum buffer size
before an attempt to marshal gathered data. Otherwise, buffers assumed
as misconfigured and the transport would be broken.
The patch brings v9fs_do_readdir_with_stat() in conformity with
v9fs_do_readdir() behavior.
Jan Dakinevich [Wed, 20 Sep 2017 06:48:51 +0000 (08:48 +0200)]
9pfs: fix readdir() for 9p2000.u
If the client is using 9p2000.u, the following occurs:
$ cd ${virtfs_shared_dir}
$ mkdir -p a/b/c
$ ls a/b
ls: cannot access 'a/b/a': No such file or directory
ls: cannot access 'a/b/b': No such file or directory
a b c
instead of the expected:
$ ls a/b
c
This is a regression introduced by commit f57f5878578a;
local_name_to_path() now resolves ".." and "." in paths,
and v9fs_do_readdir_with_stat()->stat_to_v9stat() then
copies the basename of the resulting path to the response.
With the example above, this means that "." and ".." are
turned into "b" and "a" respectively...
stat_to_v9stat() currently assumes it is passed a full
canonicalized path and uses it to do two different things:
1) to pass it to v9fs_co_readlink() in case the file is a symbolic
link
2) to set the name field of the V9fsStat structure to the basename
part of the given path
It only has two users: v9fs_stat() and v9fs_do_readdir_with_stat().
v9fs_stat() really needs 1) and 2) to be performed since it starts
with the full canonicalized path stored in the fid. It is different
for v9fs_do_readdir_with_stat() though because the name we want to
put into the V9fsStat structure is the d_name field of the dirent
actually (ie, we want to keep the "." and ".." special names). So,
we only need 1) in this case.
This patch hence adds a basename argument to stat_to_v9stat(), to
be used to set the name field of the V9fsStat structure, and moves
the basename logic to v9fs_stat().
Signed-off-by: Jan Dakinevich <[email protected]>
(groug, renamed old name argument to path and updated changelog) Signed-off-by: Greg Kurz <[email protected]>
List the branches where I queue patches for Machine Core, NUMA,
Memory Backends, and X86. Update the NUMA section to list the
"machine-next" branch instead of "numa".
hw/acpi-build: Fix SRAT memory building in case of node 0 without RAM
Currently, Using the fisrt node without memory on the machine makes
QEMU unhappy. With this example command line:
... \
-m 1024M,slots=4,maxmem=32G \
-numa node,nodeid=0 \
-numa node,mem=1024M,nodeid=1 \
-numa node,nodeid=2 \
-numa node,nodeid=3 \
Guest reports "No NUMA configuration found" and the NUMA topology is
wrong.
This is because when QEMU builds ACPI SRAT, it regards node 0 as the
default node to deal with the memory hole(640K-1M). this means the
node0 must have some memory(>1M), but, actually it can have no
memory.
Fix this problem by cut out the 640K hole in the same way the PCI
4G hole does.
Dou Liyang [Tue, 22 Aug 2017 07:45:36 +0000 (15:45 +0800)]
NUMA: Replace MAX_NODES with nb_numa_nodes in for loop
In QEMU, the number of the NUMA nodes is determined by parse_numa_opts().
Then, QEMU uses it for iteration, for example:
for (i = 0; i < nb_numa_nodes; i++)
However, in memory_region_allocate_system_memory(), it uses MAX_NODES
not nb_numa_nodes.
So, replace MAX_NODES with nb_numa_nodes to keep code consistency and
reduce the loop times.
Igor Mammedov [Thu, 1 Jun 2017 10:53:28 +0000 (12:53 +0200)]
numa: cpu: calculate/set default node-ids after all -numa CLI options are parsed
Calculating default node-ids for CPUs in possible_cpu_arch_ids()
is rather fragile since defaults calculation uses nb_numa_nodes but
callback might be potentially called early before all -numa CLI
options are parsed, which would lead to cpus assigned only upto
nb_numa_nodes at the time possible_cpu_arch_ids() is called.
Issue was introduced by
(7c88e65 numa: mirror cpu to node mapping in MachineState::possible_cpus)
and for example CLI:
-smp 4 -numa node,cpus=0 -numa node
would set props.node-id in possible_cpus array for every non
explicitly mapped CPU to the first node.
Issue is not visible to guest nor to mgmt interface due to
1) implictly mapped cpus are forced to the first node in
case of partial mapping
2) in case of default mapping possible_cpu_arch_ids() is
called after all -numa options are parsed (resulting
in correct mapping).
However it's fragile to rely on late execution of
possible_cpu_arch_ids(), therefore add machine specific
callback that returns node-id for CPU and use it to calculate/
set defaults at machine_numa_finish_init() time when all -numa
options are parsed.
Peter Maydell [Tue, 19 Sep 2017 17:08:48 +0000 (18:08 +0100)]
Merge remote-tracking branch 'remotes/cohuck/tags/s390x-20170919-v2' into staging
Assorted s390x patches:
- introduce virtio-gpu-ccw, with virtio-gpu endian fixes
- lots of cleanup in the s390x code
- make device_add work for s390x cpus
- enable seccomp on s390x
- an ivshmem endian fix
- set the reserved DHCP client architecture id for netboot
- fixes in the css and pci support
* remotes/cohuck/tags/s390x-20170919-v2: (38 commits)
MAINTAINERS/s390x: add terminal3270.c
virtio-ccw: Create a virtio gpu device for the ccw bus
virtio-gpu: Handle endian conversion
s390x/ccw: create s390 phb for compat reasons as well
configure: Allow --enable-seccomp on s390x, too
virtio-ccw: remove stale comments on endianness
s390x: allow CPU hotplug in random core-id order
s390x: generate sclp cpu information from possible_cpus
s390x: get rid of cpu_s390x_create()
s390x: get rid of cpu_states and use possible_cpus instead
s390x: implement query-hotpluggable-cpus
s390x: CPU hot unplug via device_del cannot work for now
s390x: allow cpu hotplug via device_add
s390x: print CPU definitions in sorted order
target/s390x: rename next_cpu_id to next_core_id
target/s390x: use "core-id" for cpu number/address/id handling
target/s390x: set cpu->id for linux user when realizing
s390x: allow only 1 CPU with TCG
target/s390x: use program_interrupt() in per_check_exception()
target/s390x: use trigger_pgm_exception() in s390_cpu_handle_mmu_fault()
...
s390x/ccw: create s390 phb for compat reasons as well
d32bd032d8 ("s390x/ccw: create s390 phb conditionally") made
registering the s390 pci host bridge conditional on presense
of the zpci facility bit. Sadly, that breaks migration from
machines that did not use the cpu model (2.7 and previous).
Create the s390 phb for pre-cpu model machines as well: We can
tweak s390_has_feat() to always indicate the zpci facility bit
when no cpu model is available (on 2.7 and previous compat machines).
Thomas Huth [Thu, 14 Sep 2017 10:36:03 +0000 (12:36 +0200)]
configure: Allow --enable-seccomp on s390x, too
libseccomp supports s390x since version 2.3.0, and I was able to start
a VM with "-sandbox on" without any obvious problems by using this patch,
so it should be safe to allow --enable-seccomp on s390x nowadays, too.
Halil Pasic [Thu, 14 Sep 2017 10:55:35 +0000 (12:55 +0200)]
virtio-ccw: remove stale comments on endianness
We have two stale comments suggesting one should think about virtio
config space endianness a bit longer. We have just done that, and came to
the conclusion we are fine as is: it's the responsibility of the virtio
device and not of the transport (and that is how it works now). Putting
the responsibility into the transport isn't even possible, because the
transport would have to know about the config space layout of each
device.
SCLP correctly indicates the core-id aka. CPU address for each available
CPU.
As the core-id corresponds to cpu_index, also a newly created kvm vcpu
gets assigned this core-id as vcpu id. So SIGP in the kernel works
correctly (it uses the vcpu id to lookup the correct CPU).
So there should be nothing hindering us from hotplugging CPUs in random
core-id order.
This now makes sure that the output from "query-hotpluggable-cpus"
is completely true. Until now, a specific order is implicit. Performance
vice, hotplugging CPUs in non-sequential order might not be the best thing
to do, as VCPU lookup inside KVM might be a little slower. But that
doesn't hinder us from supporting it.
s390x: generate sclp cpu information from possible_cpus
This is the first step to allow hot plugging of CPUs in a non-sequential
order. If a cpu is available ("plugged") can directly be decided by
looking at the cpu state pointer.
This makes sure, that really only cpus attached to the machine are
reported.
Now that there is only one user of cpu_s390x_create() left, make cpu
creation look like on x86.
- Perform the model/properties split and checks in s390_init_cpus()
- Parse features only once without having to remember if already parsed
- Pass only the typename to s390x_new_cpu()
- Use the typename of an existing CPU for hotplug via cpu-add
s390x: CPU hot unplug via device_del cannot work for now
device_del on a CPU will currently do nothing. Let's emit an error
telling that this is will currently not work (there is no architecture
support on s390x). Error message copied from ppc.
(qemu) device_del cpu1
device_del cpu1
CPU hot unplug not supported on this machine
E.g. the following now works:
device_add host-s390-cpu,id=cpu1,core-id=1
The system will perform the same checks as when using cpu_add:
- If the core_id is already in use
- If the next sequential core_id isn't used
- If core-id >= max_cpu is specified
In addition, mixed CPU models are checked. E.g. if starting with
-cpu host and trying to hotplug "qemu-s390-cpu":
"Mixed CPU models are not supported on s390x."
target/s390x: use "core-id" for cpu number/address/id handling
Some time ago we discussed that using "id" as property name is not the
right thing to do, as it is a reserved property for other devices and
will not work with device_add.
Switch to the term "core-id" instead, and use it as an equivalent to
"CPU address" mentioned in the PoP. There is no such thing as cpu number,
so rename env.cpu_num to env.core_id. We use "core-id" as this is the
common term to use for device_add later on (x86 and ppc).
We can get rid of cpu->id now. Keep cpu_index and env->core_id in sync.
cpu_index was already implicitly used by e.g. cpu_exists(), so keeping
both in sync seems to be the right thing to do.
cpu_index will now no longer automatically get set via
cpu_exec_realizefn(). For now, we were lucky that both implicitly stayed
in sync.
Our new cpu property "core-id" can be a static property. Range checks can
be avoided by using the correct type and the "setting after realized"
check is done implicitly.
device_add will later need the reserved "id" property. Hotplugging a CPU
on s390x will then be: "device_add host-s390-cpu,id=cpu2,core-id=2".
Specifying more than 1 CPU (e.g. -smp 5) leads to SIGP errors (the
guest tries to bring these CPUs up but fails), because we don't support
multiple CPUs on s390x under TCG.
Let's bail out if more than 1 is specified, so we don't raise people's
hope.
projects / qemu.git / log