Use the function provided rather than spying on the struct.
Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
It is never used, since ISA device are not hot-unpluggable.
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The function is empty after the previous patch, so remove it.
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Explicitly call object_unparent in the few places where we
will re-create the memory region. If the memory region is
simply being destroyed as part of device teardown, let QOM
handle it.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Of the two functions portio_list_del and portio_list_destroy,
the latter is just freeing a memory area. However, portio_list_del
is the logical equivalent of memory_region_del_subregion so
destruction of memory regions does not belong there.
Actually, neither of these APIs are in use; portio is mostly used by
ISA devices or VGAs, and neither of these is currently hot-unpluggable.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The memory regions should be destroyed in the unrealize function;
since these NICs are not even qdev-ified, they cannot be unplugged
and they do not have to do anything to destroy their memory regions.
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Instead, add a boolean variable to indicate the presence of the region.
This avoids a repeated malloc/free (later we can also avoid the
add_child/unparent by changing the offset/size of the alias).
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This ensures that the unparent callback is called automatically
when the parent object is finalized.
Note that there's no need to keep a reference neither in
object_unparent nor in object_finalize_child_property. The
reference held by the child property itself will do.
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This ensures that the children's unparent callback will still
have a usable parent.
Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The callback lets the bus provide the direction and transfer count
for passthrough commands, enabling passthrough of vendor-specific
commands.
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This will be used for both scsi_block_new_request and the scsi-block
implementation of parse_cdb.
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
These callbacks will let devices do their own request parsing, or
defer it to the bus. If the bus does not provide an implementation,
in turn, fall back to the default parsing routine.
Swap the first two arguments to scsi_req_parse, and rename it to
scsi_req_parse_cdb, for consistency.
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The per-SCSIDevice parse_cdb callback must not be called if the
request will go through special SCSIReqOps, so detect the special
cases early enough.
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
vhost-user-test uses the linux/vhost.h header, so it must only be
enabled if CONFIG_LINUX is defined. (Previously it was enabled
for CONFIG_POSIX, which broke 'make check' on MacOSX.)
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
* remotes/riku/linux-user-for-upstream:
linux-user: use TARGET_SA_ONSTACK in get_sigframe
alloca one extra byte sockets
linux-user: handle AF_PACKET sockaddrs in target_to_host_sockaddr
qemu-user: Impl. setsockopt(SO_BINDTODEVICE)
SIOCGIFINDEX: fix typo
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The permission of TCE entry should exclude physical base address.
Otherwise, unmapping TCE entry can be interpreted to mapping TCE
entry wrongly for VFIO devices.
Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Acked-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
0b183fc87 "memory: move mem_path handling to
memory_region_allocate_system_memory" disabled -mempath use for all
machines that do not use memory_region_allocate_system_memory() to
register RAM. Since SPAPR uses memory_region_init_ram(), the huge pages
support was disabled for it.
This replaces memory_region_init_ram()+vmstate_register_ram_global() with
memory_region_allocate_system_memory() to get huge pages back.
This changes RAM size from (ram_limit - rma_alloc_size) to ram_limit as
the previous patch moved RMA memory region allocation after RAM allocation
and therefore this change does not have immediate effect but simplifies
the code.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
PPC970 does not support VRMA (virtual RMA) so real memory required
for SLOF to execute must be allocated by the KVM_ALLOCATE_RMA ioctl.
Later this memory is used as a part of the guest RAM area.
The RMA allocating code also registers a memory region for this piece
of RAM.
We are going to simplify memory regions layout: RMA memory region
will be a subregion in the RAM memory region, both starting from zero.
This way we will not have to take care of start address alignment for
the piece of RAM next to the RMA.
This moves memory region business closer to the RAM memory region
creation/allocation code.
As this is a mechanical patch, no change in behaviour is expected.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
[agraf: fix compilation on non-kvm systems]
Signed-off-by: Alexander Graf <agraf@suse.de>
Commit 0b183fc871:"memory: move mem_path handling to
memory_region_allocate_system_memory" split memory_region_init_ram and
memory_region_init_ram_from_file. Also it moved mem-path handling a step
up from memory_region_init_ram to memory_region_allocate_system_memory.
Therefore for any board that uses memory_region_init_ram directly,
-mem-path is not supported.
Fix this by replacing memory_region_init_ram with
memory_region_allocate_system_memory.
Signed-off-by: Shreyas B. Prabhu <shreyas@linux.vnet.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
The number of threads per core is different for POWER6/7/8 CPUs.
Guest systems do not expect to see more threads per core than
a specific CPU supports so we need to limit this number.
This limit is implemented by ppc_get_compat_smt_threads().
However it has a problem as it checks for PCR (Processor Compatibility
Register) mask, 2.05 means 2 threads per core, 2.06 - 4 threads.
For POWER8 one would expect PCR_COMPAT_2_07 bit set and
ppc_get_compat_smt_threads() checking for it to return 8 threads
per core. But the latest PowerISA spec now is 2.07 and there is
no 2.07 compatibility mode defined, QEMU does not define it either
(will be in PowerISA 2.08).
Instead of relying on a PCR mask, this uses kvmppc_smt_threads()
which returns the maximum supported threads number for KVM or
1 for TCG.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
As reported by Laurent, which should use TARGET_SA_ONSTACK
on arm, microblaze and openrisc targets like we do on all
others. Practical matter is minimal as for almost all archs
SA_ONSTACK is 0x08000000:
http://lxr.free-electrons.com/ident?i=SA_ONSTACK
Reported-by: Laurent Desnogues <laurent.desnogues@gmail.com>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
Now requests are submitted as a batch, so it is natural
to notify guest as a batch too.
This may suppress interrupt notification to VM a lot:
- in my test, decreased by ~13K/sec
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The callback has to be saved and reset in virtio_blk_data_plane_start(),
otherwise dataplane's requests will be completed in qemu aio context.
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
when hotplug virtio-scsi disks using laio, the aio_nr will
increase in laio_init() by io_setup(), we can see the number by
# cat /proc/sys/fs/aio-nr
128
if the aio_nr attach the maxnum, which found from
# cat /proc/sys/fs/aio-max-nr
65536
the hotplug process will fail because of aio context leak.
Fix it by io_destroy in laio_cleanup().
Reported-by: daifulai <daifulai@huawei.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
target_to_host_sockaddr() may increase the lenth with 1 byte
for AF_UNIX sockets so allocate 1 extra byte.
Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
Wrong type was used in ioctl definition.
Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Riku Voipio <riku.voipio@linaro.org>
The libqos implementation of io_read{b,w,l} and io_write{b,w,l} hooks
was relying on qtest_mem{read,write}() respectively. With d81d410 (usb:
improve ehci/uhci test) this resulted in assertion failures on ppc hosts:
ERROR:tests/usb-hcd-ehci-test.c:78:ehci_port_test: assertion failed: ((value & mask) == (expect & mask))
ERROR:tests/usb-hcd-ehci-test.c:128:pci_uhci_port_2: assertion failed: (pcibus != NULL)
ERROR:tests/usb-hcd-ehci-test.c:150:pci_ehci_port_2: assertion failed: (pcibus != NULL)
qtest_read{b,w,l,q}() and qtest_write{b,w,l,q}() had been introduced
as endian-safe replacement for qtest_mem{read,write}() in I2C in
872536b (qtest: Add MMIO support). Use them for PCI as well.
Cc: Anthony Liguori <aliguori@amazon.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Fixes: c4efe1c qtest: add libqos including PCI support
Fixes: d81d410 usb: improve ehci/uhci test
Signed-off-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2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=8zIe
-----END PGP SIGNATURE-----
Merge remote-tracking branch 'remotes/bonzini/tags/for-upstream' into staging
Misc 2.1 fixes regarding character/serial devices and SCSI.
# gpg: Signature made Mon 14 Jul 2014 16:26:08 BST using RSA key ID 9B4D86F2
# gpg: Can't check signature: public key not found
* remotes/bonzini/tags/for-upstream:
serial-pci: remove memory regions from BAR before destroying them
virtio-scsi: fix with -M pc-i440fx-2.0
serial: change retry logic to avoid concurrency
qemu-char: fix deadlock with "-monitor pty"
scsi: Report error when lun number is in use
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Right now starting a machine with virtio-scsi and a <= 2.0 machine type
fails with:
qemu-system-x86_64: -device virtio-scsi-pci: Property .any_layout not found
This is because the any_layout bit was actually never set after
virtio-scsi was changed to support arbitrary layout for virtio buffers.
(This was just a cleanup and a preparation for virtio 1.0; no guest
actually checks the bit, but the new request parsing algorithms are
tested even with old guest).
Reported-by: David Gilbert <dgilbert@redhat.com>
Reviewed-by: David Gilbert <dgilbert@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Whenever serial_xmit fails to transmit a byte it adds a watch that would
call it again when the "line" becomes ready. This results in a retry
chain:
serial_xmit -> add_watch -> serial_xmit
Each chain is able to transmit one character, and for every character
passed to serial by the guest driver a new chain is spawned.
The problem lays with the fact that a new chain is spawned even when
there is one already waiting on the watch. So there can be several retry
chains waiting concurrently on one "line". Every chain tries to transmit
current character, so character order is not messed up. But also every
chain increases retry counter (tsr_retry). If there are enough
concurrent chains this counter will hit MAX_XMIT_RETRY value and
the character will be dropped.
To reproduce this bug you need to feed serial output to some program
consuming it slowly enough. A python script from bug #1335444
description is an example of such program.
This commit changes retry logic in the following way to avoid
concurrency: instead of spawning a new chain for each character being
transmitted spawn only one and make it transmit characters until FIFO is
empty.
The change consists of two parts:
- add a do {} while () loop in serial_xmit (diff is a bit erratic
for this part, diff -w will show actual change),
- do not call serial_xmit from serial_ioport_write if there is one
waiting on the watch already.
This should fix another issue causing bug #1335444.
Signed-off-by: Kirill Batuzov <batuzovk@ispras.ru>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
qemu_chr_be_generic_open cannot be called with the write lock taken,
because it calls client code that may call qemu_chr_fe_write. This
actually happens for the monitor:
0x00007ffff27dbf79 in __GI_raise (sig=sig@entry=6)
0x00007ffff27df388 in __GI_abort ()
0x00005555555ef489 in error_exit (err=<optimized out>, msg=msg@entry=0x5555559796d0 <__func__.5959> "qemu_mutex_lock")
0x00005555558f9080 in qemu_mutex_lock (mutex=mutex@entry=0x555556248a30)
0x0000555555713936 in qemu_chr_fe_write (s=0x555556248a30, buf=buf@entry=0x5555563d8870 "QEMU 2.0.90 monitor - type 'help' for more information\r\n", len=56)
0x00005555556217fd in monitor_flush_locked (mon=mon@entry=0x555556251fd0)
0x0000555555621a12 in monitor_flush_locked (mon=0x555556251fd0)
monitor_puts (mon=mon@entry=0x555556251fd0, str=0x55555634bfa7 "", str@entry=0x55555634bf70 "QEMU 2.0.90 monitor - type 'help' for more information\n")
0x0000555555624359 in monitor_vprintf (mon=0x555556251fd0, fmt=<optimized out>, ap=<optimized out>)
0x0000555555624414 in monitor_printf (mon=<optimized out>, fmt=fmt@entry=0x5555559105a0 "QEMU %s monitor - type 'help' for more information\n")
0x0000555555629806 in monitor_event (opaque=0x555556251fd0, event=<optimized out>)
0x000055555571343c in qemu_chr_be_generic_open (s=0x555556248a30)
To avoid this, defer the call to an idle callback, which will be
called as soon as the main loop is re-entered. In order to simplify
the cleanup and do it in one place only, change pty_chr_close to
call pty_chr_state.
To reproduce, run with "-monitor pty", then try to read from the
slave /dev/pts/FOO that it creates.
Fixes: 9005b2a758
Reported-by: Li Liang <liangx.z.li@intel.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)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=YxHy
-----END PGP SIGNATURE-----
Merge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging
Block patches for 2.1.0-rc2 (v2)
# gpg: Signature made Mon 14 Jul 2014 11:04:12 BST using RSA key ID C88F2FD6
# gpg: Good signature from "Kevin Wolf <kwolf@redhat.com>"
* remotes/kevin/tags/for-upstream: (22 commits)
ide: Treat read/write beyond end as invalid
virtio-blk: Treat read/write beyond end as invalid
virtio-blk: Bypass error action and I/O accounting on invalid r/w
virtio-blk: Factor common checks out of virtio_blk_handle_read/write()
dma-helpers: Fix too long qiov
qtest: fix vhost-user-test compilation with old GLib
tests: Fix unterminated string output visitor enum human string
AioContext: do not rely on aio_poll(ctx, true) result to end a loop
virtio-blk: embed VirtQueueElement in VirtIOBlockReq
virtio-blk: avoid g_slice_new0() for VirtIOBlockReq and VirtQueueElement
dataplane: do not free VirtQueueElement in vring_push()
virtio-blk: avoid dataplane VirtIOBlockReq early free
block: Assert qiov length matches request length
qed: Make qiov match request size until backing file EOF
qcow2: Make qiov match request size until backing file EOF
block: Make qiov match the request size until EOF
AioContext: speed up aio_notify
test-aio: fix GSource-based timer test
block: drop aio functions that operate on the main AioContext
block: prefer aio_poll to qemu_aio_wait
...
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The block layer fails such reads and writes just fine. However, they
then get treated like valid operations that fail: the error action
gets executed. Unwanted; reporting the error to the guest is the only
sensible action.
Reject them before passing them to the block layer. This bypasses the
error action and I/O accounting. Not quite correct for DMA, because
DMA can fail after some success, and when that happens, the part that
succeeded isn't counted. Tolerable, because I/O accounting is an
inconsistent mess anyway.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The block layer fails such reads and writes just fine. However, they
then get treated like valid operations that fail: the error action
gets executed. Unwanted; reporting the error to the guest is the only
sensible action.
Reject them before passing them to the block layer. This bypasses the
error action and I/O accounting.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
When a device model's I/O operation fails, we execute the error
action. This lets layers above QEMU implement thin provisioning, or
attempt to correct errors before they reach the guest. But when the
I/O operation fails because it's invalid, reporting the error to the
guest is the only sensible action.
If the guest's read or write asks for an invalid sector range, fail
the request right away, without considering the error action. No
change with error action BDRV_ACTION_REPORT.
Furthermore, bypass I/O accounting, because we want to track only I/O
that actually reaches the block layer.
The next commit will extend "invalid sector range" to cover attempts
to read/write beyond the end of the medium.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If the size of the scatter/gather list isn't a multiple of 512, the
number of sectors for the block layer request is rounded down, resulting
in a qiov that doesn't match the request length. Truncate the qiov to the
new length of the request.
This fixes the IDE qtest case /x86_64/ide/bmdma/short_prdt.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Mising G_TIME_SPAN_SECOND definition breaks the RHEL6 compilation as GLib
version before 2.26 does not have it. In such case just define it.
Reported-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Nikolay Nikolaev <n.nikolaev@virtualopensystems.com>
Tested-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The buffer was being allocated of size string length plus two.
Around the string two quotes were being added, but no terminating NUL.
It was then compared using g_assert_cmpstr(), resulting in fairly random
assertion failures:
ERROR:tests/test-string-output-visitor.c:213:test_visitor_out_enum: assertion failed (str == str_human): ("\"value1\"" == "\"value1\"\001EEEEEEEEEEEEEE\0171")
There is no g_assert_cmpnstr() counterpart, so use g_strdup_printf()
for safely assembling the string in the first place.
Cc: Hu Tao <hutao@cn.fujitsu.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Suggested-by: Eric Blake <eblake@redhat.com>
Fixes: b4900c0 tests: add human format test for string output visitor
Signed-off-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Currently, whenever aio_poll(ctx, true) has completed all pending
work it returns true *and* the next call to aio_poll(ctx, true)
will not block.
This invariant has its roots in qemu_aio_flush()'s implementation
as "while (qemu_aio_wait()) {}". However, qemu_aio_flush() does
not exist anymore and bdrv_drain_all() is implemented differently;
and this invariant is complicated to maintain and subtly different
from the return value of GMainLoop's g_main_context_iteration.
All calls to aio_poll(ctx, true) except one are guarded by a
while() loop checking for a request to be incomplete, or a
BlockDriverState to be idle. The one remaining call (in
iothread.c) uses this to delay the aio_context_release/acquire
pair until the AioContext is quiescent, however:
- we can do the same just by using non-blocking aio_poll,
similar to how vl.c invokes main_loop_wait
- it is buggy, because it does not ensure that the AioContext
is released between an aio_notify and the next time the
iothread goes to sleep. This leads to hangs when stopping
the dataplane thread.
In the end, these semantics are a bad match for the current
users of AioContext. So modify that one exception in iothread.c,
which also fixes the hangs, as well as the testcase so that
it use the same idiom as the actual QEMU code.
Reported-by: Christian Borntraeger <borntraeger@de.ibm.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The memory allocation between hw/block/virtio-blk.c,
hw/block/dataplane/virtio-blk.c, and hw/virtio/dataplane/vring.c is
messy. Structs are allocated in different files than they are freed in.
This is risky and makes memory leaks easier.
Embed VirtQueueElement in VirtIOBlockReq to reduce the amount of memory
allocation we need to juggle. This also makes vring.c and virtio.c
slightly more similar.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
