sdk/emulator/qemu.git
13 years agoMerge remote-tracking branch 'stefanha/trivial-patches' into staging
Anthony Liguori [Tue, 31 May 2011 13:23:11 +0000 (08:23 -0500)]
Merge remote-tracking branch 'stefanha/trivial-patches' into staging

13 years agoMerge remote-tracking branch 'amit/for-anthony' into staging
Anthony Liguori [Tue, 31 May 2011 13:22:03 +0000 (08:22 -0500)]
Merge remote-tracking branch 'amit/for-anthony' into staging

13 years agoMerge remote-tracking branch 'bonzini/scsi.2' into staging
Anthony Liguori [Tue, 31 May 2011 13:20:56 +0000 (08:20 -0500)]
Merge remote-tracking branch 'bonzini/scsi.2' into staging

Conflicts:
hw/usb-msd.c

13 years agoMerge remote-tracking branch 'kraxel/usb.14.pull' into staging
Anthony Liguori [Tue, 31 May 2011 13:17:15 +0000 (08:17 -0500)]
Merge remote-tracking branch 'kraxel/usb.14.pull' into staging

13 years agoFix spelling in comment (additon -> addition)
Stefan Weil [Sun, 22 May 2011 12:02:40 +0000 (14:02 +0200)]
Fix spelling in comment (additon -> addition)

Signed-off-by: Stefan Weil <weil@mail.berlios.de>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
13 years agopflash_cfi02: Fix a typo in debug code (TARGET_FMT_pld -> TARGET_FMT_plx)
Stefan Weil [Sun, 22 May 2011 12:02:39 +0000 (14:02 +0200)]
pflash_cfi02: Fix a typo in debug code (TARGET_FMT_pld -> TARGET_FMT_plx)

Thanks to Tobias Hoffmann <th55@gmx.de> for this patch.

Signed-off-by: Stefan Weil <weil@mail.berlios.de>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
13 years agobitbang_i2c: Fix spurious slave read after NACK
Marcus Comstedt [Sat, 28 May 2011 14:55:52 +0000 (16:55 +0200)]
bitbang_i2c: Fix spurious slave read after NACK

After NACKing a read operation, a raising SCL should not trigger a new
read from the slave.  Introduce a new state which just waits for a stop
or start condition after NACK.

Signed-off-by: Marcus Comstedt <marcus@mc.pp.se>
Signed-off-by: Andrzej Zaborowski <andrew.zaborowski@intel.com>
13 years agoMove user emulator stuff from cpu-exec.c to user-exec.c
Blue Swirl [Sun, 8 May 2011 11:22:38 +0000 (11:22 +0000)]
Move user emulator stuff from cpu-exec.c to user-exec.c

Simplify cpu-exec.c by refactoring.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
13 years agocpu-exec: prepare for user and softmmu split
Blue Swirl [Sat, 21 May 2011 08:42:35 +0000 (08:42 +0000)]
cpu-exec: prepare for user and softmmu split

There is little in common with user and softmmu versions of cpu_resume_signal(),
split them.

Fix coding style for the user emulator part.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
13 years agovirtio-console: Simplify init callbacks
Markus Armbruster [Wed, 25 May 2011 12:21:14 +0000 (14:21 +0200)]
virtio-console: Simplify init callbacks

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
13 years agovirtio-serial: Drop redundant VirtIOSerialPort member info
Markus Armbruster [Wed, 25 May 2011 12:21:13 +0000 (14:21 +0200)]
virtio-serial: Drop redundant VirtIOSerialPort member info

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
13 years agovirtio-serial: Drop useless property is_console
Markus Armbruster [Wed, 25 May 2011 12:21:12 +0000 (14:21 +0200)]
virtio-serial: Drop useless property is_console

All you could ever achieve with it is break stuff, so removing it
should be safe.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
13 years agovirtio-serial: Clean up virtconsole detection
Markus Armbruster [Wed, 25 May 2011 12:21:11 +0000 (14:21 +0200)]
virtio-serial: Clean up virtconsole detection

virtio-serial-bus needs to treat "virtconsole" devices specially.  It
uses VirtIOSerialPort member is_console to recognize them.  It gets
its value via property initialization.  Cute hack, except it lets
users mess with it: "-device virtconsole,is_console=0" isn't plugged
into port 0 as it should.

Move the flag to VirtIOSerialPortInfo.  Keep the property for backward
compatibility; its value has no effect.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
13 years agovirtio-serial: Plug memory leak on qdev exit()
Markus Armbruster [Wed, 25 May 2011 12:21:10 +0000 (14:21 +0200)]
virtio-serial: Plug memory leak on qdev exit()

virtio_serial_init() allocates the VirtIOSerialBus dynamically, but
virtio_serial_exit() doesn't free it.

Fix by getting rid of the allocation.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
13 years agovirtio-serial-bus: use bh for unthrottling
Alon Levy [Fri, 29 Apr 2011 11:25:06 +0000 (14:25 +0300)]
virtio-serial-bus: use bh for unthrottling

Instead of calling flush_queued_data when unthrottling, schedule
a bh. That way we can return immediately to the caller, and the
flush uses the same call path as a have_data for callbackee.

No migration change is required because bh are called from vm_stop.

Signed-off-by: Alon Levy <alevy@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
13 years agoscsi: ignore LUN field in the CDB
Paolo Bonzini [Thu, 19 May 2011 14:47:28 +0000 (16:47 +0200)]
scsi: ignore LUN field in the CDB

The LUN field in the CDB is a historical relic.  Ignore it as reserved,
which is what modern SCSI specifications actually say.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi: rename arguments to the new callbacks
Paolo Bonzini [Fri, 20 May 2011 18:18:07 +0000 (20:18 +0200)]
scsi: rename arguments to the new callbacks

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
13 years agoscsi: split command_complete callback in two
Paolo Bonzini [Fri, 22 Apr 2011 10:27:30 +0000 (12:27 +0200)]
scsi: split command_complete callback in two

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
13 years agoesp: rename sense to status
Paolo Bonzini [Fri, 20 May 2011 18:10:02 +0000 (20:10 +0200)]
esp: rename sense to status

This mirrors the LSI patch that was recently committed.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
13 years agoscsi-generic: Handle queue full
Paolo Bonzini [Mon, 18 Apr 2011 12:29:16 +0000 (14:29 +0200)]
scsi-generic: Handle queue full

The sg driver currently has a hardcoded limit of commands it
can handle simultaneously. When this limit is reached the
driver will return -EDOM. So we need to capture this to
enable proper return values here.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi: make write_data return void
Paolo Bonzini [Fri, 22 Apr 2011 07:39:16 +0000 (09:39 +0200)]
scsi: make write_data return void

The return value is unused anyway.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi-disk: add data direction checking
Hannes Reinecke [Mon, 18 Apr 2011 10:57:22 +0000 (12:57 +0200)]
scsi-disk: add data direction checking

scsi_req_parse() already provides for a data direction setting,
so we should be using it to check for correct direction.
And we should return the sense code 'INVALID FIELD IN CDB'
in these cases.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi: Implement 'get_sense' callback
Hannes Reinecke [Mon, 18 Apr 2011 11:36:02 +0000 (13:36 +0200)]
scsi: Implement 'get_sense' callback

The get_sense callback copies existing sense information into
the provided buffer. This is required if sense information
should be transferred together with the command response.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi: introduce scsi_req_get_buf
Paolo Bonzini [Thu, 21 Apr 2011 11:21:02 +0000 (13:21 +0200)]
scsi: introduce scsi_req_get_buf

... and remove some SCSIDevice variables or fields that now become unused.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi: introduce scsi_req_continue
Paolo Bonzini [Mon, 18 Apr 2011 13:28:11 +0000 (15:28 +0200)]
scsi: introduce scsi_req_continue

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
13 years agoscsi: introduce scsi_req_new
Paolo Bonzini [Mon, 18 Apr 2011 17:09:55 +0000 (19:09 +0200)]
scsi: introduce scsi_req_new

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi: do not call send_command directly
Paolo Bonzini [Mon, 18 Apr 2011 17:07:23 +0000 (19:07 +0200)]
scsi: do not call send_command directly

Move the common part of scsi-disk.c and scsi-generic.c to the SCSI layer.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi: Update sense code handling
Hannes Reinecke [Mon, 18 Apr 2011 10:53:14 +0000 (12:53 +0200)]
scsi: Update sense code handling

The SCSI spec has a quite detailed list of sense codes available.
It even mandates the use of specific ones for some failure cases.
The current implementation just has one type of generic error
which is actually a violation of the spec in certain cases.
This patch introduces various predefined sense codes to have the
sense code reporting more in line with the spec.

On top of Hannes's patch I fixed the reply to REQUEST SENSE commands
with DESC=0 and a small (<18) length.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi: use scsi_req_complete
Paolo Bonzini [Mon, 18 Apr 2011 15:06:17 +0000 (17:06 +0200)]
scsi: use scsi_req_complete

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi: introduce scsi_req_cancel
Paolo Bonzini [Mon, 18 Apr 2011 20:53:08 +0000 (22:53 +0200)]
scsi: introduce scsi_req_cancel

This is for when the request must be dropped in the void,
but still memory should be freed.  To this end, the devices
register a second callback in SCSIBusOps.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi: introduce scsi_req_abort
Paolo Bonzini [Mon, 18 Apr 2011 15:14:51 +0000 (17:14 +0200)]
scsi: introduce scsi_req_abort

This covers the case of canceling a request's I/O and still
completing it.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi: commonize purging requests
Paolo Bonzini [Mon, 18 Apr 2011 16:47:12 +0000 (18:47 +0200)]
scsi: commonize purging requests

The code for canceling requests upon reset is already the same.  Clean
it up and move it to scsi-bus.c.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi: Use 'SCSIRequest' directly
Hannes Reinecke [Mon, 18 Apr 2011 10:35:39 +0000 (12:35 +0200)]
scsi: Use 'SCSIRequest' directly

Currently the SCSIRequest structure is abstracted away and cannot accessed
directly from the driver. This requires the handler to do a lookup on
an abstract 'tag' which identifies the SCSIRequest structure.

With this patch the SCSIRequest structure is exposed to the driver. This
allows use to use it directly as an argument to the SCSIDeviceInfo
callback functions and remove the lookup.

A new callback function 'alloc_req' is introduced matching 'free
req'; unref'ing to free up resources after use is moved into the
scsi_command_complete callbacks.

This temporarily introduces a leak of requests that are cancelled,
when they are removed from the queue and not from the driver.  This
is fixed later by introducing scsi_req_cancel.  That patch in turn
depends on this one, because the argument to scsi_req_cancel is a
SCSIRequest.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agolsi: extract lsi_find_by_tag
Paolo Bonzini [Mon, 18 Apr 2011 10:01:45 +0000 (12:01 +0200)]
lsi: extract lsi_find_by_tag

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi: reference-count requests
Paolo Bonzini [Mon, 18 Apr 2011 14:01:56 +0000 (16:01 +0200)]
scsi: reference-count requests

With the next patch, a device may hold SCSIRequest for an indefinite
time.  Split a rather big patch, and protect against access errors,
by reference counting them.

There is some ugliness in scsi_send_command implementation due to
the need to unref the request when it fails.  This will go away
with the next patches, which move the unref'ing to the devices.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
13 years agoscsi-generic: do not use a stale aiocb
Paolo Bonzini [Wed, 25 May 2011 14:53:46 +0000 (16:53 +0200)]
scsi-generic: do not use a stale aiocb

If a request is canceled after it has been completed, scsi_cancel_io
would pass a stale aiocb to bdrv_aio_cancel.  Avoid this.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
13 years agoscsi: introduce SCSIBusOps
Paolo Bonzini [Mon, 18 Apr 2011 15:11:14 +0000 (17:11 +0200)]
scsi: introduce SCSIBusOps

There are more operations than a SCSI bus can handle, besides completing
commands.  One example, which this series will introduce, is cleaning up
after a request is cancelled.

More long term, a "SCSI bus" can represent the LUNs attached to a
target; in this case, while all commands will ultimately reach a logical
unit, it is the target who is in charge of answering REPORT LUNs.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
13 years agoscsi: introduce scsi_req_data
Paolo Bonzini [Mon, 18 Apr 2011 12:59:13 +0000 (14:59 +0200)]
scsi: introduce scsi_req_data

This abstracts calling the command_complete callback, reducing churn
in the following patches.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi-generic: Remove bogus double complete
Paolo Bonzini [Tue, 3 May 2011 12:15:59 +0000 (14:15 +0200)]
scsi-generic: Remove bogus double complete

scsi-generic scsi_read_complete() should not -both- call the client
complete callback with SCSI_REASON_DATA -and- call
scsi_command_complete().  The former will cause the client to queue a
new read or write request, while the later will free the request data
structure, thus causing the new read or write request to use a
freed/stale structure when it completes.

This patch fixes the bug, fixing a crash with scsi-generic & RHEL5.5
installer.

Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agoscsi: add tracing of scsi requests
Paolo Bonzini [Fri, 15 Apr 2011 09:51:13 +0000 (11:51 +0200)]
scsi: add tracing of scsi requests

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
13 years agousb: add ehci adapter
Gerd Hoffmann [Fri, 3 Dec 2010 15:17:28 +0000 (16:17 +0100)]
usb: add ehci adapter

This patch finally merges the EHCI host adapter aka USB 2.0 support.

Based on the ehci bits collected @ git://git.kiszka.org/qemu.git ehci

EHCI has a long out-of-tree history.  Project was started by Mark
Burkley, with contributions by Niels de Vos.  David S. Ahern continued
working on it.  Kevin Wolf, Jan Kiszka and Vincent Palatin contributed
bugfixes.

/me (Gerd Hoffmann) picked it up where it left off, prepared the code
for merge, fixed a few bugs and added basic user docs.

Cc: David S. Ahern <daahern@cisco.com>
Cc: Jan Kiszka <jan.kiszka@web.de>
Cc: Kevin Wolf <mail@kevin-wolf.de>
Cc: Vincent Palatin <vincent.palatin_qemu@m4x.org>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb: move cancel callback to USBDeviceInfo
Gerd Hoffmann [Mon, 16 May 2011 08:34:53 +0000 (10:34 +0200)]
usb: move cancel callback to USBDeviceInfo

Remove the cancel callback from the USBPacket struct, move it over
to USBDeviceInfo.  Zap usb_defer_packet() which is obsolete now.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb: keep track of packet owner.
Gerd Hoffmann [Thu, 12 May 2011 11:48:13 +0000 (13:48 +0200)]
usb: keep track of packet owner.

Keep track of the device which owns the usb packet for async processing.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb: add usb_handle_packet
Gerd Hoffmann [Thu, 12 May 2011 11:20:39 +0000 (13:20 +0200)]
usb: add usb_handle_packet

Add a usb_handle_packet function, put it into use everywhere.
Right now it just calls dev->info->handle_packet(), that will
change in future patches though.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb-storage: don't call usb_packet_complete twice
Gerd Hoffmann [Mon, 16 May 2011 12:25:59 +0000 (14:25 +0200)]
usb-storage: don't call usb_packet_complete twice

usb_msd_copy_data() may cause a recursive call to
usb_msd_command_complete() which in turn may complete
the packet, setting s->packet to NULL in case it does.
Recheck s->packet before calling usb_packet_complete()
to fix the double call.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb-linux: fix max_packet_size for highspeed.
Gerd Hoffmann [Mon, 16 May 2011 09:30:57 +0000 (11:30 +0200)]
usb-linux: fix max_packet_size for highspeed.

Calculate the max packet size correctly.  Only bits 0..11 specify the size,
bits 11+12 specify the number of (highspeed) microframes the endpoint wants
to use.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb-linux: split large xfers
Gerd Hoffmann [Mon, 16 May 2011 08:21:51 +0000 (10:21 +0200)]
usb-linux: split large xfers

Add support for splitting large transfers into multiple smaller ones.
This is needed for the upcoming EHCI emulation which allows guests
to submit requests up to 20k in size.  The linux kernel allows 16k
max size though.

Based on a patch from David Ahern, see
http://www.mail-archive.com/qemu-devel@nongnu.org/msg30337.html

Cc: David Ahern <daahern@cisco.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb-linux: walk async urb list in cancel
Gerd Hoffmann [Mon, 16 May 2011 07:20:06 +0000 (09:20 +0200)]
usb-linux: walk async urb list in cancel

Lookup async urbs which are to be canceled using the linked list
instead of the direct opaque pointer.  There are two reasons we
are doing that:  First, to avoid the opaque poiner to the callback,
which is needed for upcoming cleanups.  Second, because we might
need multiple urbs per request for highspeed support, so a single
opaque pointer doesn't cut it any more anyway.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb-linux: track aurbs in list
Gerd Hoffmann [Mon, 16 May 2011 07:13:05 +0000 (09:13 +0200)]
usb-linux: track aurbs in list

This patch adds code to track all async urbs in a linked list,
so we can find them without having to pass around a opaque
pointer to them.  Prerequisite for the cleanups.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb-linux: add hostport property
Gerd Hoffmann [Tue, 10 May 2011 10:07:42 +0000 (12:07 +0200)]
usb-linux: add hostport property

This patch adds a hostport property which allows to specify the host usb
devices to pass through by bus number and physical port.  This means you
can basically hand over one (or more) of the usb plugs on your host to
the guest and whatever device is plugged in there will show up in the
guest.

Usage:

  -device usb-host,hostbus=1,hostport=1

You can figure the port numbers by plugging in some usb device, then
find it in "info usbhost" and pick bus and port specified there.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb-linux: fix device path aka physical port handling
Gerd Hoffmann [Tue, 10 May 2011 09:43:57 +0000 (11:43 +0200)]
usb-linux: fix device path aka physical port handling

The device path isn't just a number.  It specifies the physical port
the device is connected to and in case the device is connected via
usb hub you'll have two numbers there, like this: "5.1".  The first
specifies the root port where the hub is plugged into, the second
specifies the port number of the hub where the device is plugged in.
With multiple hubs chained the string can become longer.

This patch renames devpath to port and makes it a string.   It also
adapts the sysfs parsing code accordingly.  The parser code is also more
strict now and skips the root hubs (which can't be assigned anyway).

The "info usbhost" monitor command now prints bus number, (os-assigned)
device address and physical port for each device.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb-linux: use usb_generic_handle_packet()
Hans de Goede [Wed, 2 Feb 2011 16:36:29 +0000 (17:36 +0100)]
usb-linux: use usb_generic_handle_packet()

Make the linux usb host passthrough code use the usb_generic_handle_packet()
function, rather then the curent DYI code. This removes 200 lines of almost
identical code.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
13 years agousb: Pass the packet to the device's handle_control callback
Hans de Goede [Wed, 2 Feb 2011 15:33:13 +0000 (16:33 +0100)]
usb: Pass the packet to the device's handle_control callback

This allows using the generic usb_generic_handle_packet function from
device code which does ASYNC control requests (such as the linux host
pass through code).

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
13 years agoBug #757654: UHCI fails to signal stall response patch
Jan Vesely [Mon, 9 May 2011 10:16:50 +0000 (12:16 +0200)]
Bug #757654: UHCI fails to signal stall response patch

UHCI host controller status register indicates error and
an interrupt is triggered on BABBLE and STALL errors.

Signed-off-by: Jan Vesely <jano.vesely@gmail.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb: add support for "grouped" interfaces and the Interface Association Descriptor
Brad Hards [Sun, 3 Apr 2011 05:33:21 +0000 (15:33 +1000)]
usb: add support for "grouped" interfaces and the Interface Association Descriptor

This is used for some devices that have multiple interfaces that form a logic
device. An example is Video Class, which has a Control interface and a
Streaming interface. There can be additional interfaces on the same (physical)
devices (e.g. a microphone), and Interface Association Descriptor handles this
case.

Signed-off-by: Brad Hards <bradh@frogmouth.net>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb: remove fallback to bNumInterfaces if no .nif
Brad Hards [Sun, 3 Apr 2011 05:33:20 +0000 (15:33 +1000)]
usb: remove fallback to bNumInterfaces if no .nif

All callers have been updated.

Signed-off-by: Brad Hards <bradh@frogmouth.net>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb: update config descriptors to identify number of interfaces
Brad Hards [Sun, 3 Apr 2011 05:33:19 +0000 (15:33 +1000)]
usb: update config descriptors to identify number of interfaces

Previously we relied on the .bNumInterfaces, but that won't always be
accurate after the introduction of grouped interfaces.

Signed-off-by: Brad Hards <bradh@frogmouth.net>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agousb: Add Interface Association Descriptor descriptor type
Brad Hards [Sun, 3 Apr 2011 05:33:18 +0000 (15:33 +1000)]
usb: Add Interface Association Descriptor descriptor type

Signed-off-by: Brad Hards <bradh@frogmouth.net>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
13 years agoMerge remote-tracking branch 'kraxel/CVE-2011-1751' into staging
Anthony Liguori [Wed, 25 May 2011 12:04:13 +0000 (07:04 -0500)]
Merge remote-tracking branch 'kraxel/CVE-2011-1751' into staging

13 years agoconfigure: Document --disable-slirp option in --help
Peter Maydell [Mon, 23 May 2011 09:13:47 +0000 (10:13 +0100)]
configure: Document --disable-slirp option in --help

The --disable-slirp option was undocumented; add it to configure's
--help output.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
13 years agotarget-arm/exec.h: Remove unused #define of M0
Peter Maydell [Mon, 23 May 2011 15:04:42 +0000 (16:04 +0100)]
target-arm/exec.h: Remove unused #define of M0

Remove a preprocessor #define which is never used.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
13 years agotarget-arm: Signal InvalidOp for Neon GE and GT compares of QNaN
Peter Maydell [Thu, 19 May 2011 13:46:16 +0000 (14:46 +0100)]
target-arm: Signal InvalidOp for Neon GE and GT compares of QNaN

If the input to a Neon float comparison is a quiet NaN, the ARM ARM
specifies that we should raise InvalidOp if the comparison is GE or GT
but not for EQ. (Signaling NaNs raise InvalidOp regardless). This means
only EQ should use the _quiet version of the comparison function.

We implement this by cleaning up the comparison helpers to call the
appopriate versions of the softfloat simple comparison functions
(float32_le and friends) rather than the generic float32_compare functions.
This makes them simple enough that they are clearer opencoded rather
than macroised.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
13 years agotarget-arm: Use correct float status for Neon int-float conversions
Peter Maydell [Thu, 19 May 2011 13:46:19 +0000 (14:46 +0100)]
target-arm: Use correct float status for Neon int-float conversions

The Neon versions of int-float conversions must use the "standard FPSCR"
rather than the default FPSCR. Implement this by having the helper
functions take a pointer to the appropriate float_status value rather
than simply taking a pointer to the entire CPUState, and making
translate.c pass a pointer to vfp.fp_status or vfp.standard_fp_status
appropriately for whether the instruction being translated is Neon
or VFP.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
13 years agotarget-arm: Signal Underflow when denormal flushed to zero on output
Peter Maydell [Thu, 19 May 2011 13:46:18 +0000 (14:46 +0100)]
target-arm: Signal Underflow when denormal flushed to zero on output

On ARM the architecture mandates that when an output denormal is flushed to
zero we must set the FPSCR UFC (underflow) bit, so map softfloat's
float_flag_output_denormal accordingly.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
13 years agosoftfloat: Add new flag for when denormal result is flushed to zero
Peter Maydell [Thu, 19 May 2011 13:46:17 +0000 (14:46 +0100)]
softfloat: Add new flag for when denormal result is flushed to zero

Add a new float_flag_output_denormal which is set when the result
of a floating point operation would be denormal but is flushed to
zero because we are in flush_to_zero mode. This is necessary because
some architectures signal this condition as an underflow and others
signal it as an inexact result.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
13 years agotarget-arm: Signal InputDenormal for VRECPE, VRSQRTE, VRECPS, VRSQRTS
Peter Maydell [Thu, 19 May 2011 13:46:15 +0000 (14:46 +0100)]
target-arm: Signal InputDenormal for VRECPE, VRSQRTE, VRECPS, VRSQRTS

The helpers for VRECPE.F32, VSQRTE.F32, VRECPS and VRSQRTS handle denormals
as special cases, so we must set the InputDenormal exception flag ourselves.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
13 years agotarget-arm: Don't set FP exceptions in recip, recip_sqrt estimate fns
Peter Maydell [Thu, 19 May 2011 13:46:14 +0000 (14:46 +0100)]
target-arm: Don't set FP exceptions in recip, recip_sqrt estimate fns

The functions which do the core estimation algorithms for the VRSQRTE
and VRECPE instructions should not set floating point exception flags,
so use a local fp status for doing these calculations.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
13 years agotcg: don't keep dead outputs in registers
Aurelien Jarno [Tue, 17 May 2011 16:25:45 +0000 (18:25 +0200)]
tcg: don't keep dead outputs in registers

If an op with dead outputs is not removed, because it has side effects
or has multiple output and only one dead, mark the registers as dead
instead of saving them. This avoid a few register spills on TCG targets
with low register count, especially with div2 and mul2 ops, or when a
qemu_ld* result is not used (prefetch emulation for example).

Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
13 years agotcg: mark dead output argument in op_dead_args
Aurelien Jarno [Tue, 17 May 2011 16:25:45 +0000 (18:25 +0200)]
tcg: mark dead output argument in op_dead_args

If an op is not removed and has dead output arguments, mark it
in op_dead_args similarly to what is done for input arguments.

Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
13 years agotcg: replace op_dead_iargs by op_dead_args
Aurelien Jarno [Tue, 17 May 2011 16:25:45 +0000 (18:25 +0200)]
tcg: replace op_dead_iargs by op_dead_args

Allow all args to be dead by replacing the input specific op_dead_iargs
variable by op_dead_args. Note this is a purely mechanical change.

Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
13 years agoMerge branch 'trivial-patches' of git://repo.or.cz/qemu/stefanha
Aurelien Jarno [Mon, 23 May 2011 20:36:17 +0000 (22:36 +0200)]
Merge branch 'trivial-patches' of git://repo.or.cz/qemu/stefanha

* 'trivial-patches' of git://repo.or.cz/qemu/stefanha:
  Fix typos in comments (chek -> check)
  hw/sd.c: Don't complain about SDIO commands CMD52/CMD53
  hw/realview.c: Remove duplicate #include line
  piix_pci: fix piix3_set_irq_pic()

13 years agoMerge branch 'ppc-next' of git://repo.or.cz/qemu/agraf
Aurelien Jarno [Mon, 23 May 2011 20:35:37 +0000 (22:35 +0200)]
Merge branch 'ppc-next' of git://repo.or.cz/qemu/agraf

* 'ppc-next' of git://repo.or.cz/qemu/agraf:
  Fix a bug in mtsr/mtsrin emulation on ppc64
  pSeries: Clean up write-only variables
  w32: Fix compilation and replace non-portable usage of ulong

13 years agoMerge branch 's390-next' of git://repo.or.cz/qemu/agraf
Aurelien Jarno [Mon, 23 May 2011 20:33:39 +0000 (22:33 +0200)]
Merge branch 's390-next' of git://repo.or.cz/qemu/agraf

* 's390-next' of git://repo.or.cz/qemu/agraf:
  s390x: complain when allocating ram fails
  s390x: fix memory detection for guests > 64GB
  s390x: change mapping base to allow guests > 2GB
  s390x: Fix debugging for unknown sigp order codes
  s390x: build s390x by default
  s390x: remove compatibility cc field
  s390x: Adjust GDB stub
  s390x: translate engine for s390x CPU
  s390x: Adjust internal kvm code
  s390x: Implement opcode helpers
  s390x: helper functions for system emulation
  s390x: Shift variables in CPUState for memset(0)
  s390x: keep hint on virtio managing size
  s390x: make kvm exported functions conditional on kvm
  s390x: s390x-linux-user support
  tcg: extend max tcg opcodes when using 64-on-32bit
  s390x: fix smp support for kvm

13 years agoFix typos in comments (chek -> check)
Stefan Weil [Sat, 30 Apr 2011 20:49:26 +0000 (22:49 +0200)]
Fix typos in comments (chek -> check)

Signed-off-by: Stefan Weil <weil@mail.berlios.de>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
13 years agohw/sd.c: Don't complain about SDIO commands CMD52/CMD53
Peter Maydell [Fri, 20 May 2011 09:11:53 +0000 (10:11 +0100)]
hw/sd.c: Don't complain about SDIO commands CMD52/CMD53

The SDIO specification introduces new commands 52 and 53.
Handle as illegal command but do not complain on stderr,
as SDIO-aware OSes (including Linux) may legitimately use
these in their probing for presence of an SDIO card.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
13 years agohw/realview.c: Remove duplicate #include line
Peter Maydell [Thu, 19 May 2011 15:21:57 +0000 (16:21 +0100)]
hw/realview.c: Remove duplicate #include line

Remove a duplicate #include of sysbus.h.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
13 years agopiix_pci: fix piix3_set_irq_pic()
TeLeMan [Mon, 16 May 2011 11:50:55 +0000 (19:50 +0800)]
piix_pci: fix piix3_set_irq_pic()

If pic_irq is greater than 7, the irq level is always 0 on 32bits.

Signed-off-by: TeLeMan <geleman@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
13 years agoDelete unused tb_invalidate_page_range
Blue Swirl [Sat, 14 May 2011 11:55:30 +0000 (11:55 +0000)]
Delete unused tb_invalidate_page_range

tb_invalidate_page_range() was intended to be used to invalidate an
area of a TB which the guest explicitly flushes from i-cache. However,
QEMU detects writes to code areas where TBs have been generated, so
his has never been useful.

Delete the function, adjust callers.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
13 years agoFix a bug in mtsr/mtsrin emulation on ppc64
David Gibson [Fri, 20 May 2011 03:34:59 +0000 (13:34 +1000)]
Fix a bug in mtsr/mtsrin emulation on ppc64

Early ppc64 CPUs include a hack to partially simulate the ppc32 segment
registers, by translating writes to them into writes to the SLB.  This is
not used by any current Linux kernel, but it is used by the openbios used
in the qemu mac99 model.

Commit 81762d6dd0d430d87024f2c83e9c4dcc4329fb7d, cleaning up the SLB
handling introduced a bug in this code, breaking the openbios currently in
qemu.  Specifically, there was an off by one error bitshuffling the
register format used by mtsr into the format needed for the SLB load,
causing the flag bits to end up in the wrong place.  This caused the
storage keys to be wrong under openbios, meaning that the translation code
incorrectly thought a legitimate access was a permission violation.

This patch fixes the bug, at the same time it fixes some build bug in the
MMU debugging code (only exposed when DEBUG_MMU is enabled).

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agopSeries: Clean up write-only variables
David Gibson [Tue, 17 May 2011 06:47:04 +0000 (16:47 +1000)]
pSeries: Clean up write-only variables

A few pieces of the pSeries emulation code have variables which are set
but never used, which causes warnings on gcc 4.6.  This patch removes
these instances.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agow32: Fix compilation and replace non-portable usage of ulong
Stefan Weil [Sat, 14 May 2011 23:38:04 +0000 (23:38 +0000)]
w32: Fix compilation and replace non-portable usage of ulong

ulong is undefined for w32 (and maybe other) compilations.
Replace it by uintptr_t (which also fixes compilation for w64
and is a better choice for pointer to integer conversions).

Cc: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Stefan Weil <weil@mail.berlios.de>
Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: complain when allocating ram fails
Alexander Graf [Fri, 20 May 2011 15:33:28 +0000 (17:33 +0200)]
s390x: complain when allocating ram fails

While trying out the > 64GB guest RAM patch, I hit some virtual address
limitations of my host system, which resulted in mmap failing. Unfortunately,
qemu didn't tell me about this failure, but just used the NULL pointer
happily, resulting in either segmentation faults or other fun errors.

To spare other users from tracing this down, let's print a nice message
instead so the user can figure out what's wrong from there.

Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: fix memory detection for guests > 64GB
Christian Borntraeger [Thu, 12 May 2011 08:50:44 +0000 (10:50 +0200)]
s390x: fix memory detection for guests > 64GB

the s390 memory detection has a 16bit field that specifies the amount of
increments. This patch adopts the memory size to always fit into that
scheme. This also fixes virtio detection for these guests, since the
descriptor page is located after the main memory.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: change mapping base to allow guests > 2GB
Christian Borntraeger [Tue, 10 May 2011 12:49:10 +0000 (14:49 +0200)]
s390x: change mapping base to allow guests > 2GB

the current s390x qemu memory layout is

0x1000000: guest start
0x80000000: qemu binary

which limits the amount of available memory to <2GB.
This patch moves the guest pages to 32GB to not collide with the binary
and to leave some space for the program break of qemu.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: Fix debugging for unknown sigp order codes
Christian Borntraeger [Wed, 4 May 2011 08:30:12 +0000 (10:30 +0200)]
s390x: Fix debugging for unknown sigp order codes

On unknown sigp order codes we print a debug message. This patch
fixes the output, since we want to see the order_code and not
the register numbers.
Patch applies on agraf tree.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: build s390x by default
Alexander Graf [Wed, 23 Mar 2011 07:23:47 +0000 (08:23 +0100)]
s390x: build s390x by default

This patch enables building of s390x-softmmu and s390x-linux-user
targets by default.

Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: remove compatibility cc field
Alexander Graf [Wed, 13 Apr 2011 11:30:00 +0000 (13:30 +0200)]
s390x: remove compatibility cc field

Remove the now unused cc field that was only required to not break
bisectability.

Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: Adjust GDB stub
Alexander Graf [Wed, 23 Mar 2011 09:58:07 +0000 (10:58 +0100)]
s390x: Adjust GDB stub

We have successfully lazilized cc computation, so we need to manually
trigger its calculation when gdb wants to fetch it. We also changed the
variable name, so writing it writes into a different field now.

Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: translate engine for s390x CPU
Alexander Graf [Wed, 23 Mar 2011 09:58:07 +0000 (10:58 +0100)]
s390x: translate engine for s390x CPU

This is the main meat part of the patch set. It implements emulation for an
s390x CPU.

The code does all the optimizations that are common for TCG code:

 - direct branches
 - cc optimization
 - unrolling of simple microcode loops

I'm still open for suggestions on speedups of course :).

Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: Adjust internal kvm code
Alexander Graf [Wed, 23 Mar 2011 09:58:07 +0000 (10:58 +0100)]
s390x: Adjust internal kvm code

We're now finally emulating an s390x CPU, so we can move quite some logic
from the kvm code out into generic CPU code.

This patch does this and adjusts the interfaces according to what the code
around now expects to be able to call.

Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: Implement opcode helpers
Alexander Graf [Wed, 23 Mar 2011 09:58:07 +0000 (10:58 +0100)]
s390x: Implement opcode helpers

There are some instructions that can't (or shouldn't) be expressed by pure
tcg code. For those, we call into externally compiled C functions.

This patch implements those C functions.

Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: helper functions for system emulation
Alexander Graf [Wed, 23 Mar 2011 09:58:07 +0000 (10:58 +0100)]
s390x: helper functions for system emulation

When running system emulation, we need to transverse through the MMU and
deliver interrupts according to the specification.

This patch implements those two pieces and in addition adjusts the CPU
initialization code to account for the new fields in CPUState.

Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: Shift variables in CPUState for memset(0)
Alexander Graf [Mon, 2 May 2011 08:11:40 +0000 (10:11 +0200)]
s390x: Shift variables in CPUState for memset(0)

The default reset handler does a memset(0) until right in between CPU_COMMON.
I incorrectly changed that behavior on the s390x port, so let's move the fields
in CPUState around to reflect the correct split up to which point memset(0)
zeros out everything.

Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: keep hint on virtio managing size
Alexander Graf [Wed, 13 Apr 2011 08:55:11 +0000 (10:55 +0200)]
s390x: keep hint on virtio managing size

The s390x virtio bus keeps management information on virtio after the top
of the guest's RAM. We need to be able to tell the guest the size of its
RAM (without virtio stuff), but also be able to trap when the guest accesses
RAM outside of its scope (including virtio stuff).

So we need a variable telling us the size of the virtio stuff, so we can
calculate the highest available RAM address from that.

While at it, also increase the maximum number of virtio pages, so we play
along well with more recent kernels that spawn a ridiculous number of virtio
console adapters.

Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: make kvm exported functions conditional on kvm
Alexander Graf [Fri, 15 Apr 2011 13:16:40 +0000 (15:16 +0200)]
s390x: make kvm exported functions conditional on kvm

We have some helper functions we use to directly invoke KVM
functionality from device emulation code.

This patch replaces those exported functions with static inline
stubs when not building with KVM enabled.

Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: s390x-linux-user support
Ulrich Hecht [Fri, 24 Jul 2009 14:57:31 +0000 (16:57 +0200)]
s390x: s390x-linux-user support

This patch adds support for running s390x binaries in the linux-user emulation
code.

Signed-off-by: Ulrich Hecht <uli@suse.de>
Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agotcg: extend max tcg opcodes when using 64-on-32bit
Alexander Graf [Wed, 13 Apr 2011 11:08:44 +0000 (13:08 +0200)]
tcg: extend max tcg opcodes when using 64-on-32bit

When running a 64 bit guest on a 32 bit host, we tend to use more TCG ops
than on a 64 bit host. Reflect that in the reserved opcode amount constant.

Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agos390x: fix smp support for kvm
Christian Borntraeger [Thu, 5 May 2011 07:29:57 +0000 (09:29 +0200)]
s390x: fix smp support for kvm

Currently smp support for kvm does not work. Qemu does a kvm run even on
secondary CPUs which dont have a sane state (initial psw == 0)
triggering some program faults. Architecturally these cpus are in the stopped
state, so we should not do the kvm run ioctl. (these CPUs will be started
by a SIGP restart later during the boot process)

We need to tell the loop that this cpu should not run. Jan Kiszka pointed
out that kvm_arch_process_async_events is the right place to do.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
13 years agoahci: Fix non-NCQ accesses for LBA > 16bits
Alexander Graf [Thu, 19 May 2011 09:57:09 +0000 (11:57 +0200)]
ahci: Fix non-NCQ accesses for LBA > 16bits

AHCI provides two ways of reading/writing data:

 1) NCQ
 2) ATA commands with the LBA in the command FIS

In the second code path, we didn't handle any LBAs that were bigger than
16 bits, so whenever a guest that used high LBA numbers wanted to access
data, the LBA got truncated down to 16 bits, giving the guest garbage.

This patch adds support for LBAs higher than 16 bits. I've tested that it
works just fine with SeaBIOS and Linux guests. This patch also unbreaks
the often reported grub errors people have seen with AHCI.

Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
13 years agoIgnore pci unplug requests for unpluggable devices (CVE-2011-1751)
Gerd Hoffmann [Wed, 27 Apr 2011 09:05:34 +0000 (11:05 +0200)]
Ignore pci unplug requests for unpluggable devices (CVE-2011-1751)

This patch makes qemu ignore unplug requests from the guest for pci
devices which are tagged as non-hotpluggable.  Trouble spot is the
piix4 chipset with the ISA bridge.  Requests to unplug that one will
make it go away together with all ISA bus devices, which are not
prepared to be unplugged and thus don't cleanup, leaving active
qemu timers behind in free'ed memory.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>