target_phys_addr_t is unwieldly, violates the C standard (_t suffixes are
reserved) and its purpose doesn't match the name (most target_phys_addr_t
addresses are not target specific). Replace it with a finger-friendly,
standards conformant hwaddr.
Outstanding patchsets can be fixed up with the command
git rebase -i --exec 'find -name "*.[ch]"
| xargs s/target_phys_addr_t/hwaddr/g' origin
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
KVM performs TPR raising asynchronously to QEMU, specifically outside
QEMU's global lock. When an interrupt is injected into the APIC and TPR
is checked to decide if this can be delivered, a stale TPR value may be
used, causing spurious interrupts in the end.
Fix this by deferring apic_update_irq to the context of the target VCPU.
We introduce a new interrupt flag for this, CPU_INTERRUPT_POLL. When it
is set, the VCPU calls apic_poll_irq before checking for further pending
interrupts. To avoid special-casing KVM, we also implement this logic
for TCG mode.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
When the guest modifies the LVT_LINT0 register, we need to check if some
pending PIC interrupt can now be delivered.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Commit d96e173769 refactored the reinjection of pending PIC interrupts.
However, it missed the potential loop of apic_update_irq ->
apic_deliver_pic_intr -> apic_local_deliver -> apic_set_irq ->
apic_update_irq that /could/ occur if LINT0 is injected as APIC_DM_FIXED
and that vector is currently blocked via TPR.
Resolve this by reinjecting only where it matters: inside
apic_get_interrupt. This function may clear a vector while a
PIC-originated reason still exists.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This patch move the msi definition from apic.c to apic-msidef.h. So it can be
used also by other .c files.
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Push msi_supported enabling to the APIC implementations where we can
encapsulate the decision more cleanly, hiding the details from the
generic code.
Acked-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
This enables acceleration for MMIO-based TPR registers accesses of
32-bit Windows guest systems. It is mostly useful with KVM enabled,
either on older Intel CPUs (without flexpriority feature, can also be
manually disabled for testing) or any current AMD processor.
The approach introduced here is derived from the original version of
qemu-kvm. It was refactored, documented, and extended by support for
user space APIC emulation, both with and without KVM acceleration. The
VMState format was kept compatible, so was the ABI to the option ROM
that implements the guest-side para-virtualized driver service. This
enables seamless migration from qemu-kvm to upstream or, one day,
between KVM and TCG mode.
The basic concept goes like this:
- VAPIC PV interface consisting of I/O port 0x7e and (for KVM in-kernel
irqchip) a vmcall hypercall is registered
- VAPIC option ROM is loaded into guest
- option ROM activates TPR MMIO access reporting via port 0x7e
- TPR accesses are trapped and patched in the guest to call into option
ROM instead, VAPIC support is enabled
- option ROM TPR helpers track state in memory and invoke hypercall to
poll for pending IRQs if required
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Replace device_init() with generalized type_init().
While at it, unify naming convention: type_init([$prefix_]register_types)
Also, type_init() is a function, so add preceding blank line where
necessary and don't put a semicolon after the closing brace.
Signed-off-by: Andreas Färber <afaerber@suse.de>
Cc: Anthony Liguori <anthony@codemonkey.ws>
Cc: malc <av1474@comtv.ru>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
This was done in a mostly automated fashion. I did it in three steps and then
rebased it into a single step which avoids repeatedly touching every file in
the tree.
The first step was a sed-based addition of the parent type to the subclass
registration functions.
The second step was another sed-based removal of subclass registration functions
while also adding virtual functions from the base class into a class_init
function as appropriate.
Finally, a python script was used to convert the DeviceInfo structures and
qdev_register_subclass functions to TypeInfo structures, class_init functions,
and type_register_static calls.
We are almost fully converted to QOM after this commit.
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
This converts three devices because apic and ioapic are subclasses of sysbus.
Converting subclasses independently of their base class is prohibitively hard.
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
To enable migration between accelerated and non-accelerated APIC models,
we will need to handle the timer saving and restoring specially and can
no longer rely on the automatics of VMSTATE_TIMER. Specifically,
accelerated model will not start any QEMUTimer.
This patch therefore factors out the generic bits into apic_next_timer
and use a post-load callback to implemented model-specific logic.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
The KVM in-kernel APIC model will reuse parts of the user space model
while providing the same frontend view to guest and most management
interfaces.
Factor out an APIC base class to encapsulate those parts that will be
shared by user space and KVM model. This class offers callback hooks for
init, base/tpr setting, and the external NMI delivery that will be
set via APICCommonInfo structure and implemented specifically in the
subclasses.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
On real hardware, NMI button events are injected via the LINT1 line of
the APICs. E.g. kdump expect this wiring and gets upset if the per-APIC
LINT1 mask is not respected, i.e. if NMIs are injected to VCPUs that
should not receive them. Change the APIC emulation code to reflect this.
Based on qemu-kvm patch by Lai Jiangshan.
CC: Lai Jiangshan <laijs@cn.fujitsu.com>
Reported-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
All LVTs are masked on reset, so the timer becomes ineffective. Letting
it tick nevertheless is harmless, but will at least create a spurious
trace event.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
The master PIC is connected to the LINTIN0 of the APICs. As the APIC
currently does not track the state of that line, we have to ask the PIC
to reinject its IRQ after the CPU picked up an event from the APIC.
This introduces pic_get_output to read the master PIC IRQ line state
without changing it. The APIC uses this function to decide if a PIC IRQ
should be reinjected on apic_update_irq. This reflects better how the
real hardware works.
The patch fixes some failures of the kvm unit tests apic and eventinj by
allowing to enable the proper CPU IRQ deassertion when the guest masks
some pending IRQs at PIC level.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Polarity of external interrupts needs to be handled in the IOAPIC.
Passing it to the APIC is pointless. So remove all these arguments.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
This was done with:
sed -i 's/qemu_get_clock\>/qemu_get_clock_ns/' \
$(git grep -l 'qemu_get_clock\>' )
sed -i 's/qemu_new_timer\>/qemu_new_timer_ns/' \
$(git grep -l 'qemu_new_timer\>' )
after checking that get_clock and new_timer never occur twice
on the same line. There were no missed occurrences; however, even
if there had been, they would have been caught by the compiler.
There was exactly one false positive in qemu_run_timers:
- current_time = qemu_get_clock (clock);
+ current_time = qemu_get_clock_ns (clock);
which is of course not in this patch.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
TPR blocks all interrupts in a priority class, so simple "less or
equal" check is not enough.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Reviewed-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Add the missing EOI broadcast from local APIC to the IOAPICs on
completion of level-triggered IRQs. This ensures that a still asserted
IRQ source properly re-triggers an APIC IRQ.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
As stated before, devices can be little, big or native endian. The
target endianness is not of their concern, so we need to push things
down a level.
This patch adds a parameter to cpu_register_io_memory that allows a
device to choose its endianness. For now, all devices simply choose
native endian, because that's the same behavior as before.
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
local_apics are allocated sequentially and never removed, so
we can stop any iterations that go to MAX_APICS as soon as we
hit the first NULL. Looking at a small guest running a virtio-net
workload with oprofile, this drops apic_get_delivery_bitmask()
from #3 in the profile to down in the noise.
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Replace debug printf statements with tracepoints.
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Make APICState completely private to apic.c by using DeviceState
in external APIs.
Move apic_init() to pc.c.
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Convert to qdev.
Use an opaque CPUState pointer because of missing VMState
implementation for CPUState.
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Move the actual CPUState contents handling to cpu.h and cpuid.c.
Handle CPU reset and set env->halted in pc.c.
Add a function to get the local APIC state of the current
CPU for the MMIO.
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
Add a DPRINTF macro. Use TARGET_FMT_plx for printing target_phys_addr_t
items. Add a separate flag for debugging coalescing interrupts.
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
x86 definitions (especially CPUState uses) prevent many files from
being compiled within libhw.
Move x86 specific declarations (APIC stuff) to a separate file.
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
This grand cleanup drops all reset and vmsave/load related
synchronization points in favor of four(!) generic hooks:
- cpu_synchronize_all_states in qemu_savevm_state_complete
(initial sync from kernel before vmsave)
- cpu_synchronize_all_post_init in qemu_loadvm_state
(writeback after vmload)
- cpu_synchronize_all_post_init in main after machine init
- cpu_synchronize_all_post_reset in qemu_system_reset
(writeback after system reset)
These writeback points + the existing one of VCPU exec after
cpu_synchronize_state map on three levels of writeback:
- KVM_PUT_RUNTIME_STATE (during runtime, other VCPUs continue to run)
- KVM_PUT_RESET_STATE (on synchronous system reset, all VCPUs stopped)
- KVM_PUT_FULL_STATE (on init or vmload, all VCPUs stopped as well)
This level is passed to the arch-specific VCPU state writing function
that will decide which concrete substates need to be written. That way,
no writer of load, save or reset functions that interact with in-kernel
KVM states will ever have to worry about synchronization again. That
also means that a lot of reasons for races, segfaults and deadlocks are
eliminated.
cpu_synchronize_state remains untouched, just as Anthony suggested. We
continue to need it before reading or writing of VCPU states that are
also tracked by in-kernel KVM subsystems.
Consequently, this patch removes many cpu_synchronize_state calls that
are now redundant, just like remaining explicit register syncs.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
There is absolutely no need to call reset functions when initializing
devices. Since we are already registering them, calling qemu_system_reset()
should suffice. Actually, it is what happens when we reboot the machine,
and using the same process instead of a special case semantics will even
allow us to find bugs easier.
Furthermore, the fact that we initialize things like the cpu quite early,
leads to the need to introduce synchronization stuff like qemu_system_cond.
This patch removes it entirely. All we need to do is call qemu_system_reset()
only when we're already sure the system is up and running
I tested it with qemu (with and without io-thread) and qemu-kvm, and it
seems to be doing okay - although qemu-kvm uses a slightly different patch.
[ v2: user mode still needs cpu_reset, so put it in ifdef. ]
[ v3: leave qemu_system_cond for now. ]
Signed-off-by: Glauber Costa <glommer@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
In the very least, a change like this requires discussion on the list.
The naming convention is goofy and it causes a massive merge problem. Something
like this _must_ be presented on the list first so people can provide input
and cope with it.
This reverts commit 99a0949b72.
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Doing this will make the vcpu ioctl be issued from the I/O thread, instead
of cpu thread. The correct behaviour is to call it from within the cpu thread,
as soon as we are ready to go.
Signed-off-by: Glauber Costa <glommer@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
cpu_synchronize_state() is a little unreadable since the 'modified'
argument isn't self-explanatory. Simplify it by making it always
synchronize the kernel state into qemu, and automatically flush the
registers back to the kernel if they've been synchronized on this
exit.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
This reverts commit 8217606e6e (and
updates later added users of qemu_register_reset), we solved the
problem it originally addressed less invasively.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
During startup and after reset we have to synchronize user space to the
in-kernel KVM state. Namely, we need to transfer the VCPU registers when
they change due to VCPU as well as APIC reset.
This patch refactors the required hooks so that kvm_init_vcpu registers
its own per-VCPU reset handler and adds a cpu_synchronize_state to the
APIC reset. That way we no longer depend on the new reset order (and can
drop this disliked interface again) and we can even drop a KVM hook in
main().
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Implement MSI support in APIC. Note that MSI and MMIO APIC registers
are at the same memory location, but actually not on the global bus: MSI
is on PCI bus, APIC is connected directly to the CPU. We map them on the
global bus at the same address which happens to work because MSI
registers are reserved in APIC MMIO and vice versa.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
This should fix compilation problem in case of CONFIG_USER_ONLY.
Currently INIT/SIPI is handled in the context of CPU that sends IPI.
This patch changes this to handle them like all other events in a main
cpu exec loop. When KVM will gain thread per vcpu capability it will
be much more clear to handle those event by cpu thread itself and not
modify one cpu's state from the context of the other.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
The parameter is always zero except when registering the three internal
io regions (ROM, unassigned, notdirty). Remove the parameter to reduce
the API's power, thus facilitating future change.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
(patch is on top of "Apic creation should not depend on pci" series)
Currently cpu_index is used as cpu apic id on x86. This is incorrect
since apic ids not have to be continuous (they can also encode cpu
hierarchy information). This patch uses cpuid_apic_id for initial apic id
value. For now cpuid_apic_id is set to be equal to cpu_index so behaviour
is fully backward compatible, but it allows us to add qemu option to
provide other values for cpu apic id.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
