Merge remote-tracking branch 'remotes/elmarco/tags/ivshmem-pull-request' into staging

ivshmem series

# gpg: Signature made Mon 26 Oct 2015 09:27:46 GMT using RSA key ID 75969CE5
# gpg: Good signature from "Marc-André Lureau <marcandre.lureau@redhat.com>"
# gpg:                 aka "Marc-André Lureau <marcandre.lureau@gmail.com>"
# 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: 87A9 BD93 3F87 C606 D276  F62D DAE8 E109 7596 9CE5

* remotes/elmarco/tags/ivshmem-pull-request: (51 commits)
  doc: document ivshmem & hugepages
  ivshmem: use little-endian int64_t for the protocol
  ivshmem: use kvm irqfd for msi notifications
  ivshmem: rename MSI eventfd_table
  ivshmem: remove EventfdEntry.vector
  ivshmem: add hostmem backend
  ivshmem: use qemu_strtosz()
  ivshmem: do not keep shm_fd open
  tests: add ivshmem qtest
  qtest: add qtest_add_abrt_handler()
  msix: implement pba write (but read-only)
  contrib: remove unnecessary strdup()
  ivshmem: add check on protocol version in QEMU
  docs: update ivshmem device spec
  ivshmem-server: fix hugetlbfs support
  ivshmem-server: use a uint16 for client ID
  ivshmem-client: check the number of vectors
  contrib: add ivshmem client and server
  util: const event_notifier_get_fd() argument
  ivshmem: reset mask on device reset
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

doc: document ivshmem & hugepages

Document and give some examples of hugepages support with ivshmem device
and server.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>

ivshmem: use little-endian int64_t for the protocol

The current ivshmem protocol uses 'long' for integers. But the
sizeof(long) depends on the host and the endianess is not defined, which
may cause portability troubles.

Instead, switch to using little-endian int64_t. This breaks the
protocol, except on x64 little-endian host where this change
should be compatible.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: use kvm irqfd for msi notifications

Use irqfd for improving context switch when notifying the guest.
If the host doesn't support kvm irqfd, regular msi notifications are
still supported.

Note: the ivshmem implementation doesn't allow switching between MSI and
IO interrupts, this patch doesn't either.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>

ivshmem: rename MSI eventfd_table

The array is used to have vector specific data, so use a more
descriptive name.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: remove EventfdEntry.vector

No need to store an extra int for the vector number when it can be
computed easily by looking at the position in the array.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: add hostmem backend

Instead of handling allocation, teach ivshmem to use a memory backend.
This allows to use hugetlbfs backed memory now.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: use qemu_strtosz()

Use the common qemu utility function to parse the memory size.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: do not keep shm_fd open

Remove shm_fd from device state, closing it as early as possible to avoid leaks.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

tests: add ivshmem qtest

Adds 4 ivshmemtests:
- single qemu instance and basic IO
- pair of instances, check memory sharing
- pair of instances with server, and MSIX
- hot plug/unplug

A temporary shm is created as well as a directory to place server
socket, both should be clear on exit and abort.

Cc: Cam Macdonell <cam@cs.ualberta.ca>
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>

qtest: add qtest_add_abrt_handler()

Allow a test to add abort handlers, use GHook for all handlers.

There is currently no way to remove a handler, but it could be
later added if needed.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

msix: implement pba write (but read-only)

qpci_msix_pending() writes on pba region, causing qemu to SEGV:

  Program received signal SIGSEGV, Segmentation fault.
  [Switching to Thread 0x7ffff7fba8c0 (LWP 25882)]
  0x0000000000000000 in ?? ()
  (gdb) bt
  #0  0x0000000000000000 in  ()
  #1  0x00005555556556c5 in memory_region_oldmmio_write_accessor (mr=0x5555579f3f80, addr=0, value=0x7fffffffbf68, size=4, shift=0, mask=4294967295, attrs=...) at /home/elmarco/src/qemu/memory.c:434
  #2  0x00005555556558e1 in access_with_adjusted_size (addr=0, value=0x7fffffffbf68, size=4, access_size_min=1, access_size_max=4, access=0x55555565563e <memory_region_oldmmio_write_accessor>, mr=0x5555579f3f80, attrs=...) at /home/elmarco/src/qemu/memory.c:506
  #3  0x00005555556581eb in memory_region_dispatch_write (mr=0x5555579f3f80, addr=0, data=0, size=4, attrs=...) at /home/elmarco/src/qemu/memory.c:1176
  #4  0x000055555560b6f9 in address_space_rw (as=0x555555eff4e0 <address_space_memory>, addr=3759147008, attrs=..., buf=0x7fffffffc1b0 "", len=4, is_write=true) at /home/elmarco/src/qemu/exec.c:2439
  #5  0x000055555560baa2 in cpu_physical_memory_rw (addr=3759147008, buf=0x7fffffffc1b0 "", len=4, is_write=1) at /home/elmarco/src/qemu/exec.c:2534
  #6  0x000055555564c005 in cpu_physical_memory_write (addr=3759147008, buf=0x7fffffffc1b0, len=4) at /home/elmarco/src/qemu/include/exec/cpu-common.h:80
  #7  0x000055555564cd9c in qtest_process_command (chr=0x55555642b890, words=0x5555578de4b0) at /home/elmarco/src/qemu/qtest.c:378
  #8  0x000055555564db77 in qtest_process_inbuf (chr=0x55555642b890, inbuf=0x55555641b340) at /home/elmarco/src/qemu/qtest.c:569
  #9  0x000055555564dc07 in qtest_read (opaque=0x55555642b890, buf=0x7fffffffc2e0 "writel 0xe0100800 0x0\n", size=22) at /home/elmarco/src/qemu/qtest.c:581
  #10 0x000055555574ce3e in qemu_chr_be_write (s=0x55555642b890, buf=0x7fffffffc2e0 "writel 0xe0100800 0x0\n", len=22) at qemu-char.c:306
  #11 0x0000555555751263 in tcp_chr_read (chan=0x55555642bcf0, cond=G_IO_IN, opaque=0x55555642b890) at qemu-char.c:2876
  #12 0x00007ffff64c9a8a in g_main_context_dispatch (context=0x55555641c400) at gmain.c:3122

(without this patch, this can be reproduced with the ivshmem qtest)

Implement an empty mmio write to avoid the crash.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>

contrib: remove unnecessary strdup()

getopt() optarg points to argv memory, no need to dup those values,
fixes small leaks detected by clang-analyzer.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>

ivshmem: add check on protocol version in QEMU

Send a protocol version as the first message from server, clients must
close communication if they don't support this protocol version.  Older
QEMUs should be fine with this change in the protocol since they
overrides their own vm_id on reception of an id associated to no
eventfd.

Signed-off-by: David Marchand <david.marchand@6wind.com>
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
[use fifo_update_and_get()]
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

docs: update ivshmem device spec

Add some notes on the parts needed to use ivshmem devices: more specifically,
explain the purpose of an ivshmem server and the basic concept to use the
ivshmem devices in guests.
Move some parts of the documentation and re-organise it.

Signed-off-by: David Marchand <david.marchand@6wind.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>

ivshmem-server: fix hugetlbfs support

As pointed out on the ML by Andrew Jones, glibc no longer permits
creating POSIX shm on hugetlbfs directly. When given a hugetlbfs path,
create a shareable file there.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>

ivshmem-server: use a uint16 for client ID

In practice, the number of VM is limited to MAXUINT16 in ivshmem, so use
the same limit on the server (removes a theorical infinite loop)

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem-client: check the number of vectors

Check the number of vectors received from the server, to avoid
out of bound array access.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

contrib: add ivshmem client and server

When using ivshmem devices, notifications between guests can be sent as
interrupts using a ivshmem-server (typical use described in documentation).
The client is provided as a debug tool.

Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Signed-off-by: David Marchand <david.marchand@6wind.com>
[fix a valgrind warning, option and server_close() segvs, extra server
headers includes, getopt() return type, out-of-tree build, use qemu
event_notifier instead of eventfd, fix x86/osx warnings - Marc-André]
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>

util: const event_notifier_get_fd() argument

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>

ivshmem: reset mask on device reset

The interrupt mask is a state value, it should be reset, like the
interrupt status.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: error on too many eventfd received

The number of eventfd that can be handled per peer is limited by the
number of vectors. Return an error when receiving too many of them.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: replace 'guest' for 'peer' appropriately

The terms 'guest' and 'peer' are used sometime interchangeably which may
be confusing. Instead, use 'peer' for the remote instances of ivshmem
clients, and 'guest' for the local VM.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: fix pci_ivshmem_exit()

Free all objects owned by the device, making sure the device is free,
fixing hot-unplug.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: add device description

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: check shm isn't already initialized

The server should not change the shm, and this isn't handled by qemu and
we should should verify this in qemu.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: shmfd can be 0

0 is a valid fd value, so change conditions and set -1 value early

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: migrate with VMStateDescription

load_state_old() is used to keep compatibility with version 0.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: use common is_power_of_2()

The common version correctly checks for 0 value case.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: use common return

Both if branches return, move this out to common end.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: simplify a bit the code

Use some more explicit variables to simplify the code.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: print error on invalid peer id

The server shouldn't send invalid peer id, so print an error if it's the
case.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: improve error handling

The test whether the chardev is an AF_UNIX socket rejects
"-chardev socket,id=chr0,path=/tmp/foo,server,nowait -device
ivshmem,chardev=chr0", but fails to explain why.

Use an explicit error on why a chardev may be rejected.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: improve debug messages

Some misc improvements to ivshmem debug.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: remove max_peer field

max_peer isn't really useful, it tracks the maximum received VM id, but
that quickly matches nb_peers, the size of the peers array. Since VM
come and go, there might be sparse peers so it doesn't help much in
general to have this value around.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: initialize max_peer to -1

There is no peer when device is initialized, do not let doorbell for
inexisting peer 0.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: remove useless ivshmem_update_irq() val argument

val isn't used in ivshmem_update_irq() function.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: allocate eventfds in resize_peers()

It simplifies a bit the code to allocate the array when setting the
number of peers instead of lazily when receiving the first vector.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: simplify around increase_dynamic_storage()

Set the number of peers and array allocation in a single place. Rename
to better reflect the function content.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: limit maximum number of peers to G_MAXUINT16

Limit the maximum number of peers to MAXUINT16. This is more realistic
and better matches the limit of the doorbell register.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: remove last exit(1)

Failing to create a chardev shouldn't be fatal.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: more qdev conversion

Use the latest qemu device modeling API, in particular, convert to
realize to fix the error handling; right now a botched device_add
ivhsmem command kills the VM.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: remove useless doorbell field

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: remove superflous ivshmem_attr field

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: remove unnecessary dup()

qemu_chr_fe_get_msgfd() transfers ownership, there is no need to dup the
fd.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: factor out the incoming fifo handling

Make a new function fifo_update_and_get() that can be reused by other
functions (in next commits).

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivshmem: fix number of bytes to push to fifo

If the fifo has 0 bytes, and the read is of size 1, the call to
fifo8_push_all() will copy off boundary data.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

ivhsmem: read do not accept more than sizeof(long)

ivshmem_read() only reads sizeof(long) from the input buffer.  Accepting
more could lead to fifo8 abort() on 32bit systems if fifo is not empty.

A following patch will change the protocol to 64-bit little-endian
instead.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

msix: add VMSTATE_MSIX_TEST

ivshmem is going to use MSIX state conditionally.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

char: add qemu_chr_free()

If a chardev is allowed to be created outside of QMP, then it must be
also possible to free it. This is useful for ivshmem that creates
chardev anonymously and must be able to free them.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Claudio Fontana <claudio.fontana@huawei.com>

tests: Add ivshmem qtest

Note that it launches two instances, as sharing memory is the purpose of
ivshmem.

Cc: Cam Macdonell <cam@cs.ualberta.ca>
Cc: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
[ Remove Nahanni codename, add test to pci set - Marc-André ]
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>

config: enable ivshmem on POSIX

ivshmem doesn't actually require kvm, so enable it when POSIX is
enabled. (it is required however when ioeventfd is enabled)

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>

Merge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging

Block layer patches

# gpg: Signature made Fri 23 Oct 2015 17:59:56 BST using RSA key ID C88F2FD6
# gpg: Good signature from "Kevin Wolf <kwolf@redhat.com>"

* remotes/kevin/tags/for-upstream: (37 commits)
  tests: Add test case for aio_disable_external
  block: Add "drained begin/end" for internal snapshot
  block: Add "drained begin/end" for transactional blockdev-backup
  block: Add "drained begin/end" for transactional backup
  block: Add "drained begin/end" for transactional external snapshot
  block: Introduce "drained begin/end" API
  aio: introduce aio_{disable,enable}_external
  dataplane: Mark host notifiers' client type as "external"
  nbd: Mark fd handlers client type as "external"
  aio: Add "is_external" flag for event handlers
  throttle: Remove throttle_group_lock/unlock()
  blockdev: Allow more options for BB-less BDS tree
  blockdev: Pull out blockdev option extraction
  blockdev: Do not create BDS for empty drive
  block: Prepare for NULL BDS
  block: Add blk_insert_bs()
  block: Prepare remaining BB functions for NULL BDS
  block: Fail requests to empty BlockBackend
  block: Make some BB functions fall back to BBRS
  block: Add BlockBackendRootState
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

tests: Add test case for aio_disable_external

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add "drained begin/end" for internal snapshot

This ensures the atomicity of the transaction by avoiding processing of
external requests such as those from ioeventfd.

state->bs is assigned right after bdrv_drained_begin. Because it was
used as the flag for deletion or not in abort, now we need a separate
flag - InternalSnapshotState.created.

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add "drained begin/end" for transactional blockdev-backup

Similar to the previous patch, make sure that external events are not
dispatched during transaction operations.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add "drained begin/end" for transactional backup

This ensures the atomicity of the transaction by avoiding processing of
external requests such as those from ioeventfd.

Move the assignment to state->bs up right after bdrv_drained_begin, so
that we can use it in the clean callback. The abort callback will still
check bs->job and state->job, so it's OK.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add "drained begin/end" for transactional external snapshot

This ensures the atomicity of the transaction by avoiding processing of
external requests such as those from ioeventfd.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Introduce "drained begin/end" API

The semantics is that after bdrv_drained_begin(bs), bs will not get new external
requests until the matching bdrv_drained_end(bs).

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

aio: introduce aio_{disable,enable}_external

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

dataplane: Mark host notifiers' client type as "external"

They will be excluded by type in the nested event loops in block layer,
so that unwanted events won't be processed there.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

nbd: Mark fd handlers client type as "external"

So we could distinguish it from internal used fds, thus avoid handling
unwanted events in nested aio polls.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

aio: Add "is_external" flag for event handlers

All callers pass in false, and the real external ones will switch to
true in coming patches.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

throttle: Remove throttle_group_lock/unlock()

The group throttling code was always meant to handle its locking
internally. However, bdrv_swap() was touching the ThrottleGroup
structure directly and therefore needed an API for that.

Now that bdrv_swap() no longer exists there's no need for the
throttle_group_lock() API anymore.

Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockdev: Allow more options for BB-less BDS tree

Most of the options which blockdev_init() parses for both the
BlockBackend and the root BDS are valid for just the root BDS as well
(e.g. read-only). This patch allows specifying these options even if not
creating a BlockBackend.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockdev: Pull out blockdev option extraction

Extract some of the blockdev option extraction code from blockdev_init()
into its own function. This simplifies blockdev_init() and will allow
reusing the code in a different function added in a follow-up patch.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockdev: Do not create BDS for empty drive

Do not use "rudimentary" BDSs for empty drives any longer (for
freshly created drives).

After a follow-up patch, empty drives will generally use a NULL BDS, not
only the freshly created drives.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Prepare for NULL BDS

blk_bs() will not necessarily return a non-NULL value any more (unless
blk_is_available() is true or it can be assumed to otherwise, e.g.
because it is called immediately after a successful blk_new_with_bs() or
blk_new_open()).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add blk_insert_bs()

This function associates the given BlockDriverState with the given
BlockBackend.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Prepare remaining BB functions for NULL BDS

There are several BlockBackend functions which, in theory, cannot fail.
This patch makes them cope with the BlockDriverState pointer being NULL
by making them fall back to some default action like ignoring the value
in setters and returning the default in getters.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Fail requests to empty BlockBackend

If there is no BlockDriverState in a BlockBackend or if the tray of the
guest device is open, fail all requests (where that is possible) with
-ENOMEDIUM.

The reason the status of the guest device is taken into account is
because once the guest device's tray is opened, any request on the same
BlockBackend as the guest uses should fail. If the BDS tree is supposed
to be usable even after ejecting it from the guest, a different
BlockBackend must be used.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Make some BB functions fall back to BBRS

If there is no BDS tree attached to a BlockBackend, functions that can
do so should fall back to the BlockBackendRootState structure.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add BlockBackendRootState

This structure will store some of the state of the root BDS if the BDS
tree is removed, so that state can be restored once a new BDS tree is
inserted.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/throttle-groups: Make incref/decref public

Throttle groups are not necessarily referenced by BDSs alone; a later
patch will essentially allow BBs to reference them, too. Make the
ref/unref functions public so that reference can be properly accounted
for.

Their interface is slightly adjusted in that they return and take a
ThrottleState pointer, respectively, instead of a ThrottleGroup pointer.
Functionally, they are equivalent, but since ThrottleGroup is not meant
to be used outside of block/throttle-groups.c, ThrottleState is easier
to handle.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Move I/O status and error actions into BB

These options are only relevant for the user of a whole BDS tree (like a
guest device or a block job) and should thus be moved into the
BlockBackend.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Move BlockAcctStats into BlockBackend

As the comment above bdrv_get_stats() says, BlockAcctStats is something
which belongs to the device instead of each BlockDriverState. This patch
therefore moves it into the BlockBackend.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Remove wr_highest_sector from BlockAcctStats

BlockAcctStats contains statistics about the data transferred from and
to the device; wr_highest_sector does not fit in with the rest.

Furthermore, those statistics are supposed to be specific for a certain
device and not necessarily for a BDS (see the comment above
bdrv_get_stats()); on the other hand, wr_highest_sector may be a rather
important information to know for each BDS. When BlockAcctStats is
finally removed from the BDS, we will want to keep wr_highest_sector in
the BDS.

Finally, wr_highest_sector is renamed to wr_highest_offset and given the
appropriate meaning. Externally, it is represented as an offset so there
is no point in doing something different internally. Its definition is
changed to match that in qapi/block-core.json which is "the offset after
the greatest byte written to". Doing so should not cause any harm since
if external programs tried to calculate the volume usage by
(wr_highest_offset + 512) / volume_size, after this patch they will just
assume the volume to be full slightly earlier than before.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Move guest_block_size into BlockBackend

guest_block_size is a guest device property so it should be moved into
the interface between block layer and guest devices, which is the
BlockBackend.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Fix BB AIOCB AioContext without BDS

Fix the BlockBackend's AIOCB AioContext for aborting AIO in case there
is no BDS. If there is no implementation of AIOCBInfo::get_aio_context()
the AioContext is derived from the BDS the AIOCB belongs to. If that BDS
is NULL (because it has been removed from the BB) this will not work.

This patch makes blk_get_aio_context() fall back to the main loop
context if the BDS pointer is NULL and implements
AIOCBInfo::get_aio_context() (blk_aiocb_get_aio_context()) which invokes
blk_get_aio_context().

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

hw/usb-storage: Check whether BB is inserted

Only call bdrv_add_key() on the BlockDriverState if it is not NULL.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

hw/block/fdc: Implement tray status

The tray of an FDD is open iff there is no medium inserted (there are
only two states for an FDD: "medium inserted" or "no medium inserted").

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Invoke change media CB before NULLing drv

In order to handle host device passthrough, some guest device models
may call blk_is_inserted() to check whether the medium is inserted on
the host, when checking the guest tray status.

This tray status is inquired by blk_dev_change_media_cb(); because
bdrv_is_inserted() (invoked by blk_is_inserted()) always returns false
for BDS with drv set to NULL, blk_dev_change_media_cb() should therefore
be called before drv is set to NULL.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/raw_bsd: Drop raw_is_inserted()

With the new automatically-recursive implementation of
bdrv_is_inserted() checking by default whether all the children of a BDS
are inserted, we can drop raw's own implementation.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Make bdrv_is_inserted() recursive

If bdrv_is_inserted() is called on the top level BDS, it should make
sure all nodes in the BDS tree are actually inserted.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add blk_is_available()

blk_is_available() returns true iff the BDS is inserted (which means
blk_bs() is not NULL and bdrv_is_inserted() returns true) and if the
tray of the guest device is closed.

blk_is_inserted() is changed to return true only if blk_bs() is not
NULL.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Make bdrv_is_inserted() return a bool

Make bdrv_is_inserted(), blk_is_inserted(), and the callback
BlockDriver.bdrv_is_inserted() return a bool.

Suggested-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests: Only create BB if necessary

Tests 071 and 081 test giving references in blockdev-add. It is not
necessary to create a BlockBackend here, so omit it.

While at it, fix up some blockdev-add invocations in the vicinity
(s/raw/$IMGFMT/ in 081, drop the format BDS for blkverify's raw child in
071).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

blockdev: Allow creation of BDS trees without BB

If the "id" field is missing from the options given to blockdev-add,
just omit the BlockBackend and create the BlockDriverState tree alone.

However, if "id" is missing, "node-name" must be specified; otherwise,
the BDS tree would no longer be accessible.

Many BDS options which are not parsed by bdrv_open() (like caching)
cannot be specified for these BB-less BDS trees yet. A future patch will
remove this limitation.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Set BDRV_O_INCOMING in bdrv_fill_options()

This flag should not be set for the root BDS only, but for any BDS that
is being created while incoming migration is pending, so setting it is
moved from blockdev_init() to bdrv_fill_options().

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Remove host floppy support

It has been deprecated as of 2.3, so we can now remove it.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

Merge remote-tracking branch 'remotes/ehabkost/tags/x86-pull-request' into staging

X86 queue, 2015-10-23

# gpg: Signature made Fri 23 Oct 2015 16:30:58 BST using RSA key ID 984DC5A6
# gpg: Good signature from "Eduardo Habkost <ehabkost@redhat.com>"

* remotes/ehabkost/tags/x86-pull-request:
  vl: trivial: minor tweaks to a max-cpu error msg
  target-i386: Use 1UL for bit shift
  target-i386: Add DE to TCG_FEATURES
  target-i386: Ensure always-1 bits on DR6 can't be cleared
  target-i386: Check CR4[DE] for processing DR4/DR5
  target-i386: Handle I/O breakpoints
  target-i386: Optimize setting dr[0-3]
  target-i386: Move hw_*breakpoint_* functions
  target-i386: Ensure bit 10 on DR7 is never cleared
  target-i386: Re-introduce optimal breakpoint removal
  target-i386: Introduce cpu_x86_update_dr7
  target-i386: Disable cache info passthrough by default
  target-i386: allow any alignment for SMBASE

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

vl: trivial: minor tweaks to a max-cpu error msg

Signed-off-by: Andrew Jones <drjones@redhat.com>

target-i386: Use 1UL for bit shift

Fix undefined behavior detected by clang runtime check:

  qemu/target-i386/cpu.c:1494:15: runtime error:
    left shift of 1 by 31 places cannot be represented in type 'int'

While doing that, add extra parenthesis for clarity.

Reported-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

target-i386: Add DE to TCG_FEATURES

Now DE is supported by TCG so it can be enabled in CPUID bits.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

target-i386: Ensure always-1 bits on DR6 can't be cleared

Bits 4-11 and 16-31 on DR6 are documented as always 1, so ensure they
can't be cleared by software.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

target-i386: Check CR4[DE] for processing DR4/DR5

Introduce helper_get_dr so that we don't have to put CR4[DE]
into the scarce HFLAGS resource.  At the same time, rename
helper_movl_drN_T0 to helper_set_dr and set the helper flags.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

target-i386: Handle I/O breakpoints

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

target-i386: Optimize setting dr[0-3]

If the debug register is not enabled, we need
do nothing besides update the register.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

target-i386: Move hw_*breakpoint_* functions

They're only used from bpt_helper.c now.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

target-i386: Ensure bit 10 on DR7 is never cleared

Bit 10 of DR7 is documented as always set to 1, so ensure that's
always the case.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>