Convert the fixed-address IO_MEM_RAM, IO_MEM_ROM, IO_MEM_UNASSIGNED,
and IO_MEM_NOTDIRTY io handlers to MemoryRegions. These aren't real
regions, since they are never added to the memory hierarchy, but they
allow reuse of the dispatch functionality.
Signed-off-by: Avi Kivity <avi@redhat.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Its use of IO_MEM_ROM and friends will later cause #include loops; and it
is too large to merit inlining.
Signed-off-by: Avi Kivity <avi@redhat.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
The code sometimes uses range comparisons on io indexes (e.g.
index =< IO_MEM_ROM). Avoid these as they make moving to objects harder.
Signed-off-by: Avi Kivity <avi@redhat.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
cpu_register_physical_memory_log() does not update region_offset
if a page was previously registered for the same address. This
could cause mmio accesses going to the wrong place, by using the
old region_offset.
Signed-off-by: Avi Kivity <avi@redhat.com>
Acked-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Richard Henderson <rth@twiddle.net>
backend_registered was used to lazify the process of registering an
mmio region, since the it is different for the I/O address space and
the memory address space. However, it also makes registration dependent
on the region being visible in the address space. This is not the case
for "fake" regions, like watchpoints or IO_MEM_UNASSIGNED.
Remove backend_registered and always initialize the region. If it turns
out to be part of the I/O address space, we've wasted an I/O slot, but
that's not too bad. In any case this will be optimized later on.
Signed-off-by: Avi Kivity <avi@redhat.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Currently mmio access goes directly to the io_mem_{read,write} arrays.
In preparation for eliminating them, add indirection via a function.
Signed-off-by: Avi Kivity <avi@redhat.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Instead of returning a PhysPageDesc pointer, return a temporary.
This lets us move away from actually storing PhysPageDesc's, and
instead sythesising them when needed.
Signed-off-by: Avi Kivity <avi@redhat.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Instead of doing device endianness compensation in cpu_register_io_memory(),
do it in the memory core.
Signed-off-by: Avi Kivity <avi@redhat.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
The getter is no longer used, so it is completely removed.
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Version 3 ram_load depends on ram_addrs, which are not stable. Version 4
was introduced in 0.13 (and RHEL 6), so this means live migration from 0.12
and earlier to 1.1 or later will not work.
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
ram_addr is (a) unstable (b) going away. Sort by idstr instead.
Commit b2e0a138e initially introduced the sorting for the purpose
of improving debuggability. After this patch, the order is still
stable, but perhaps less usable by a human.
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Avoid using ram_addr_t, instead use (MemoryRegion *, offset) pairs.
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
As a step in moving live migration from RAMBlocks to MemoryRegions,
store the MemoryRegion in a RAMBlock.
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Currently creating a memory region automatically registers it for
live migration. This differs from other state (which is enumerated
in a VMStateDescription structure) and ties the live migration code
into the memory core.
Decouple the two by introducing a separate API, vmstate_register_ram(),
for registering a RAM block for migration. Currently the same
implementation is reused, but later it can be moved into a separate list,
and registrations can be moved to VMStateDescription blocks.
Signed-off-by: Avi Kivity <avi@redhat.com>
* qemu-kvm/memory/page_desc: (22 commits)
Remove cpu_get_physical_page_desc()
sparc: avoid cpu_get_physical_page_desc()
virtio-balloon: avoid cpu_get_physical_page_desc()
vhost: avoid cpu_get_physical_page_desc()
kvm: avoid cpu_get_physical_page_desc()
memory: remove CPUPhysMemoryClient
xen: convert to MemoryListener API
memory: temporarily add memory_region_get_ram_addr()
xen, vga: add API for registering the framebuffer
vhost: convert to MemoryListener API
kvm: convert to MemoryListener API
kvm: switch kvm slots to use host virtual address instead of ram_addr_t
memory: add API for observing updates to the physical memory map
memory: replace cpu_physical_sync_dirty_bitmap() with a memory API
framebuffer: drop use of cpu_physical_sync_dirty_bitmap()
loader: remove calls to cpu_get_physical_page_desc()
framebuffer: drop use of cpu_get_physical_page_desc()
memory: introduce memory_region_find()
memory: add memory_region_is_logging()
memory: add memory_region_is_rom()
...
Make's multiple output syntax
x.c x.h: x.template
gen < x.template
actually invokes the command once for x.c and once for x.h (with differing $@
in each invocation). During a parallel build, the two commands may be invoked
in parallel; this opens up a race, where the second invocation trashes a file
supposedly produced during the first, and now in use by a dependent command.
The various qapi code generators are susceptible to this; fix by making them
generate just one file per invocation.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
* aneesh/for-upstream:
scripts/analyse-9p-simpletrace.py: Add symbolic names for 9p operations.
hw/9pfs: iattr_valid flags are kernel internal flags map them to 9p values.
hw/9pfs: Use the correct signed type for different variables
hw/9pfs: replace iovec manipulation with QEMUIOVector
* bonzini/nbd-for-anthony: (26 commits)
nbd: add myself as maintainer
qemu-nbd: throttle requests
qemu-nbd: asynchronous operation
qemu-nbd: add client pointer to NBDRequest
qemu-nbd: move client handling to nbd.c
qemu-nbd: use common main loop
link the main loop and its dependencies into the tools
qemu-nbd: introduce NBDRequest
qemu-nbd: introduce NBDExport
qemu-nbd: introduce nbd_do_receive_request
qemu-nbd: more robust handling of invalid requests
qemu-nbd: introduce nbd_do_send_reply
qemu-nbd: simplify nbd_trip
move corking functions to osdep.c
qemu-nbd: remove data_size argument to nbd_trip
qemu-nbd: remove offset argument to nbd_trip
Update ioctl order in nbd_init() to detect EBUSY
nbd: add support for NBD_CMD_TRIM
nbd: add support for NBD_CMD_FLUSH
nbd: add support for NBD_CMD_FLAG_FUA
...
qemu-kvm passes numa/SRAT topology information for smp_cpus to SeaBIOS. However
SeaBIOS always expects to setup max_cpus number of SRAT cpu entries
(MaxCountCPUs variable in build_srat function of Seabios). When qemu-kvm runs
with smp_cpus != max_cpus (e.g. -smp 2,maxcpus=4), Seabios will mistakenly use
memory SRAT info for setting up CPU SRAT entries for the offline CPUs. Wrong
SRAT memory entries are also created. This breaks NUMA in a guest.
Fix by setting up SRAT info for max_cpus in qemu-kvm.
Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
The latter was already commented out, the former is redundant as well.
We always get the latest changes after return from the guest via
kvm_arch_post_run.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Keep a per-VCPU xsave buffer for kvm_put/get_xsave instead of
continuously allocating and freeing it on state sync.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Field 0 (FCW+FSW) and 1 (FTW+FOP) were hard-coded so far.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Limiting the number of in-flight requests is implemented very simply
with a can_read callback. It does not require a semaphore, unlike the
client side in block/nbd.c, because we can throttle directly the creation
of coroutines. The client side can have a coroutine created at any time
when an I/O request is made.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Using coroutines enable asynchronous operation on both the network and
the block side. Network can be owned by two coroutines at the same time,
one writing and one reading. On the send side, mutual exclusion is
guaranteed by a CoMutex. On the receive side, mutual exclusion is
guaranteed because new coroutines immediately start receiving data,
and no new coroutines are created as long as the previous one is receiving.
Between receive and send, qemu-nbd can have an arbitrary number of
in-flight block transfers. Throttling is implemented by the next
patch.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
By attaching a client to an NBDRequest, we can avoid passing around the
socket descriptor and data buffer.
Also, we can now manage the reference count for the client in
nbd_request_get/put request instead of having to do it ourselved in
nbd_read. This simplifies things when coroutines are used.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch sets up the fd handler in nbd.c instead of qemu-nbd.c. It
introduces NBDClient, which wraps the arguments to nbd_trip in a single
structure, so that we can add a notifier to it. This way, qemu-nbd can
know about disconnections.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Using a single main loop for sockets will help yielding from the socket
coroutine back to the main loop, and later reentering it.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Using the main loop code from QEMU enables tools to operate fully
asynchronously. Advantages include better Windows portability (for some
definition of portability) over glib's.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move the buffer from NBDExport to a new structure, so that it will be
possible to have multiple in-flight requests for the same export
(and for the same client too---we get that for free).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Group the sending of a reply and the associated data into a new function.
Without corking, the caller would be forced to leave 12 free bytes at the
beginning of the data pointer. Not too ugly, but still ugly. :)
Using nbd_do_send_reply everywhere will help when the routine will set up
the write handler that re-enters the send coroutine.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Avi Kivity