linux-aio: remove process requests callback

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: avoid SIGUSR2

Now that iothread is always compiled sending a signal seems only an
additional step. This patch also avoid writing to two pipe (one from signal
and one in qemu_service_io).

Work with kvm enabled or disabled. strace output is more readable (less syscalls).

[ kwolf: Merged build fix by Paolo Bonzini ]

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

scsi: fix sign extension problems

When assigning a 32-bit value to cmd->xfer (which is 64-bits)
it can be erroneously sign extended because the intermediate
32-bit computation is signed.  Fix this by standardizing on
the ld*_be_p functions.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

rbd: call flush, if available

librbd recently added async writeback and flush support.  If the new
rbd_flush() call is available, call it.

Signed-off-by: Sage Weil <sage@newdream.net>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

rbd: update comment heading

Properly document the configuration string syntax and semantics.  Remove
(out of date) details about the librbd implementation.

Signed-off-by: Sage Weil <sage@newdream.net>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

rbd: ignore failures when reading from default conf location

If we are reading from the default config location, ignore any failures.
It is perfectly legal for the user to specify exactly the options they need
and to not rely on any config file.

Signed-off-by: Sage Weil <sage@newdream.net>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

AHCI Port Interrupt Enable register cleaning on soft reset

I've found that FreeBSD AHCI driver doesn't work with AHCI hardware
emulation of QEMU 0.15.0. I believe the problem is on QEMU's side. As I
see, it clears port's Interrupt Enable register each time when reset of
any level happens. Is is reasonable for the global controller reset. It
is probably not good, but acceptable for FreeBSD driver for the port
hard reset. But it is IMO wrong for the device soft reset. None of real
hardware I know behaves that way.

Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

posix-aio-compat: Removed unused offset variable

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

VMDK: fix leak of extent_file

Release extent_file on error in vmdk_parse_extents. Added closing files
in freeing extents.

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

scsi-disk: lazily allocate bounce buffer

It will not be needed for reads and writes if the HBA provides a sglist.
In addition, this lets scsi-disk refuse commands with an excessive
allocation length, as well as limit memory on usual well-behaved guests.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

scsi-disk: commonize iovec creation between reads and writes

Also, consistently use qiov.size instead of iov.iov_len.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

dma-helpers: rewrite completion/cancellation

This fixes various problems with completion/cancellation:

* if the io_func fails to get an AIOCB, the callback wasn't called

* If DMA encounters a bounce buffer conflict, and the DMA operation is
canceled before the bottom half fires, bad things happen.

* memory is not unmapped after cancellation, again causing problems
when doing DMA to I/O areas

* cancellation could leak the iovec

* the callback was missed if the I/O operation failed without returning
an AIOCB

and probably more that I've missed.  The patch fixes them by sharing
the cleanup code between completion and cancellation.  The dma_bdrv_cb
now returns a boolean completed/not completed flag, and the wrapper
dma_continue takes care of tasks to do upon completion.

Most of these are basically impossible in practice, but it is better
to be tidy...

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

dma-helpers: allow including from target-independent code

Target-independent code cannot construct sglists, but it can take
them from the outside as a black box.  Allow this.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

dma-helpers: rename is_write to to_dev

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

scsi-generic: do not disable FUA

I found no rationale for this in the logs, and it is quite bad because
it will make scsi-generic unsafe WRT power failures.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

raw-posix: Fix bdrv_flush error return values

bdrv_flush is supposed to use 0/-errno return values

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

nbd: support NBD_SET_FLAGS ioctl

The nbd kernel module cannot enable DISCARD requests unless it is
informed about it.  The flags field in the header is used for this,
and this patch adds support for it.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

nbd: sync API definitions with upstream

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

nbd: support feature negotiation

nbd supports writing flags in bytes 24...27 of the header,
and uses that for the read-only flag.  Add support for it
in qemu-nbd.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

sun4u: don't set up isa_mem_base

Since we use memory API in sun4u.c, after
71579cae30b53c910cd6c47ab4e683f647d36519, setting up isa_mem_base
puts vga.chain4 outside of the physical address space.

Fix by removing obsolete isa_mem_base set up.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

tcg/ppc64: Only one call output register needed for 64 bit hosts

The second register is only needed for 32 bit hosts.

Cc: Vassili Karpov <av1474@comtv.ru>
Fine-with-me'd-by: Vassili Karpov <av1474@comtv.ru>
Signed-off-by: Stefan Weil <weil@mail.berlios.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

tcg/sparc: Only one call output register needed for 64 bit hosts

The second register is only needed for 32 bit hosts.

Cc: Blue Swirl <blauwirbel@gmail.com>
Signed-off-by: Stefan Weil <weil@mail.berlios.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

tcg/s390: Only one call output register needed for 64 bit hosts

The second register is only needed for 32 bit hosts.

Cc: Alexander Graf <agraf@suse.de>
Acked-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Stefan Weil <weil@mail.berlios.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

tcg/ia64: Only one call output register needed for 64 bit hosts

The second register is never used for ia64 hosts.

Cc: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Stefan Weil <weil@mail.berlios.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

tcg/i386: Only one call output register needed for 64 bit hosts

The second register is only needed for 32 bit hosts.

Signed-off-by: Stefan Weil <weil@mail.berlios.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-i386: Fix several SSE3 instructions.

haddp[sd], hsubp[sd] and addsubp[sd] operate on floats, thus it is
necessary to use the appropriate floating point calculation functions.
If this is not done, those functions operate merely on integers, which
is not correct.

Signed-off-by: Max Reitz <max@tyndur.org>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

trace: add virtio_set_status() trace event

The virtio device lifecycle can be observed by looking at the sequence
of set status operations.  This is especially important for catching the
reset operation (status value 0), which resets the device and all
virtqueues.

Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

trace: remove trailing double quotes after PRI*64

Now that format strings can end in a PRI*64 macro, remove the
workarounds from the trace-events file.

Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

trace: allow PRI*64 at beginning and ending of format string

The tracetool parser only picks up PRI*64 and other format string macros
when enclosed between double quoted strings.  Lift this restriction by
extracting everything after the closing ')' as the format string:

  cpu_set_apic_base(uint64_t val) "%016"PRIx64
                                  ^^        ^^

One trick here: it turns out that backslashes in the format string like
"\n" were being interpreted by echo(1).  Fix this by using the POSIX
printf(1) command instead.  Although it normally does not make sense to
include backslashes in trace event format strings, an injected newline
causes tracetool to emit a broken header file and I want to eliminate
cases where broken output is emitted, even if the input was bad.

Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

trace: remove newline from grlib_irqmp_check_irqs format string

There is no need to put a newline in trace event format strings.  The
backend may use the format string within some context and takes care of
how to display the event.  The stderr backend automatically appends "\n"
whereas the ust backend does not want a newline at all.

Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

Makefile: Fix broken build

make -C mybuilddir no longer works (regression caused by commit)
388d475815c23901010a25c845eb078d47ee0740.

PWD is the directory of the caller (not mybuilddir),
so BUILD_DIR is set to the wrong value.

GNU make sets CURDIR to the correct value.
Use this macro instead of PWD.

Cc: Lluís Vilanova <vilanova@ac.upc.edu>
Cc: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Stefan Weil <weil@mail.berlios.de>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

qdev: print bus properties too

Make qdev_device_help print both device and bus properties.
Helps libvirt to figure whenever bus properties such as
PCI.multifunction are supported present or not.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vga: Respect isa_mem_base when registering chain4 alias

This does not yet unbreak PPC (which has its own problems) but
potentially other non-x86 systems where isa_mem_base is != 0.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

hw/integratorcp: Fix bugs in writes to CM_CTRL system register

Fix a number of bugs in the implementation of writes to the CM_CTRL
system register:
 * write to cm_ctrl, not cm_init !
 * an '&' vs '^' typo meant we would write the inverse of the bits
 * handling the LED via printf() meant we spew lots of output
   to stdout when Linux uses the LED as a heartbeat indicator
 * we would hw_error() if a reset was requested rather than
   actually resetting

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

bswap.h: build fix

qemu build fails when CONFIG_MACHINE_BSWAP_H is defined
because float32, float64, etc. are not defined.
This makes qemu build.

Signed-off-by: Christoph Egger <Christoph.Egger@amd.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

console: Properly switch consoles for screen dumps

Do not mess with active_console, use console_select instead. This fixes
corrupt virtual monitor consoles after issuing the screendump command.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

vga: Fix text mode screendumps

In text mode, even a full refresh of the screen takes multiple updates.
As we reset the dump file pointer after the first call, we only wrote
the first line.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Fix termination by signal with -no-shutdown

On signals such as SIGTERM qemu should exit instead of just stopping the VM
even with -no-shutdown.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Remove blanks before \n in output strings

Those blanks violate the coding conventions, see
scripts/checkpatch.pl.

Blanks missing after colons in the changed lines were added.

This patch does not try to fix tabs, long lines and other
problems in the changed lines, therefore checkpatch.pl reports
many violations.

Signed-off-by: Stefan Weil <weil@mail.berlios.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

build: Move tracing objects into libuser on usermode emulation targets

This will apply libuser-specific compilation flags (like the ones added by
--enable-user-pie), but keep softmmu emulation targets "as-is".

Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

build: Always have 'BUILD_DIR' variable pointing to the root of the build tree

Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Signed-off-by: Lluís Vilanova <vilanova@ac.upc.edu>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

Merge remote-tracking branch 'qemu-kvm-tmp/memory/core' into staging

Merge remote-tracking branch 'aneesh/for-upstream-4' into staging

Fix subtle integer overflow bug in memory API

It is quite common to have a MemoryRegion with size of INT64_MAX.
When processing alias regions in render_memory_region() it's quite
easy to find a case where it will construct a temporary AddrRange with
a non-zero start, and size still of INT64_MAX.  When means attempting
to compute the end of such a range as start + size will result in
signed integer overflow.

This integer overflow means that addrrange_intersects() can
incorrectly report regions as not intersecting when they do.  For
example consider the case of address ranges {0x10000000000,
0x7fffffffffffffff} and {0x10010000000, 0x10000000} where the second
is in fact included completely in the first.

This patch rearranges addrrange_intersects() to avoid the integer
overflow, correcting this behaviour.

Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Avi Kivity <avi@redhat.com>

qcow2: fix range check

QCowL2Meta::offset is not cluster aligned but only sector aligned
however nb_clusters count cluster from cluster start.
This fix range check. Note that old code have no corruption issues
related to this check cause it only cause intersection to occur
when shouldn't.

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: align cluster_data to block to improve performance using O_DIRECT

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

ahci: Remove unused struct member

Member variable is_read is written, but never read
(contrary to its name). Remove it.

Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Weil <weil@mail.berlios.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: initialize metadata before inserting in cluster_allocs

QCow2Meta structure was inserted into list before many fields are
initialized. Currently is not a problem cause all occur in a lock
but if qcow2_alloc_clusters would in a future unlock this lock
some issues could arise.
Initializing fields before inserting fix the problem.

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

ide/atapi scsi-disk: Make monitor eject -f, then change work

change fails while the tray is locked by the guest.  eject -f forces
it open and removes any media.  Unfortunately, the tray closes again
instantly.  Since the lock remains as it is, there is no way to insert
another medium unless the guest voluntarily unlocks.

Fix by leaving the tray open after monitor eject.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: New change_media_cb() parameter load

To let device models distinguish between eject and load.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

nbd: Clean up use of block_int.h

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Reset buffer alignment on detach

BlockDriverState member buffer_alignment is initially 512.  The device
model may set them, with bdrv_set_buffer_alignment().  If the device
model gets detached (hot unplug), the device's alignment is left
behind.  Only okay because device hot unplug automatically destroys
the BlockDriverState.  But that's a questionable feature, best not to
rely on it.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: New bdrv_set_buffer_alignment()

Device models should be able to set it without an unclean include of
block_int.h.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

hw: Trim superfluous #include "block_int.h"

Including it in device models is unclean, including it without a
reason adds insult to injury.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Move BlockConf & friends from block_int.h to block.h

It's convenience stuff for block device models, so block.h isn't the
ideal home either, but better than block_int.h.

Permits moving some #include "block_int.h" from device model .h into
.c.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Show whether the virtual tray is open in info block

Need to ask the device, so this requires new BlockDevOps member
is_tray_open().

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Drop BlockDriverState member removable

It's a confused mess (see previous commit).  No users remain.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Clean up remaining users of "removable"

BlockDriverState member removable is a confused mess.  It is true when
an ide-cd, scsi-cd or floppy qdev is attached, or when the
BlockDriverState was created with -drive if={floppy,sd} or -drive
if={ide,scsi,xen,none},media=cdrom ("created removable"), except when
an ide-hd, scsi-hd, scsi-generic or virtio-blk qdev is attached.

Three users remain:

1. eject_device(), via bdrv_is_removable() uses it to determine
   whether a block device can eject media.

2. bdrv_info() is monitor command "info block".  QMP documentation
   says "true if the device is removable, false otherwise".  From the
   monitor user's point of view, the only sensible interpretation of
   "is removable" is "can eject media with monitor commands eject and
   change".

A block device can eject media unless a device is attached that
doesn't support it.  Switch the two users over to new
bdrv_dev_has_removable_media() that returns exactly that.

3. bdrv_getlength() uses to suppress its length cache when media can
   change (see commit 46a4e4e6).  Media change is either monitor
   command change (updates the length cache), monitor command eject
   (doesn't update the length cache, easily fixable), or physical
   media change (invalidates length cache, not so easily fixable).

I'm refraining from improving anything here, because this series is
long enough already.  Instead, I simply switch it over to
bdrv_dev_has_removable_media() as well.

This changes the behavior of the length cache and of monitor commands
eject and change in two cases:

a. drive not created removable, no device attached

   The commit makes the drive removable, and defeats the length cache.

   Example: -drive if=none

b. drive created removable, but the attached drive is non-removable,
   and doesn't call bdrv_set_removable(..., 0) (most devices don't)

   The commit makes the drive non-removable, and enables the length
   cache.

   Example: -drive if=xen,media=cdrom -M xenpv

   The other non-removable devices that don't call
   bdrv_set_removable() can't currently use a drive created removable,
   either because they aren't qdevified, or because they lack a drive
   property.  Won't stay that way.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

ide/atapi: Preserve tray state on migration

Use a subsection, so that migration to older version still works,
provided the tray is closed and unlocked.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

scsi-disk: Fix START_STOP to fail when it can't eject

Don't fail when tray is already open.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

ide/atapi: Don't fail eject when tray is already open

MMC-5 6.40.2.6 specifies that START STOP UNIT succeeds when the drive
already has the requested state.  cmd_start_stop_unit() fails when
asked to eject while the tray is open and locked.  Fix that.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

rbd: fix leak in qemu_rbd_open failure paths

Fix leak of s->snap in failure path.  Simplify error paths for the whole
function.

Reported-by: Stefan Hajnoczi <stefanha@gmail.com>
Signed-off-by: Sage Weil <sage@newdream.net>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

rbd: clean up, fix style

No assignment in condition.  Remove duplicate ret > 0 check.

Signed-off-by: Sage Weil <sage@newdream.net>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

rbd: allow client id to be specified in config string

Allow the client id to be specified in the config string via 'id=' so that
users can control who they authenticate as.  Currently they are stuck with
the default ('admin').  This is necessary for anyone using authentication
in their environment.

Signed-off-by: Sage Weil <sage@newdream.net>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Rename bdrv_set_locked() to bdrv_lock_medium()

While there, make the locked parameter bool.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Drop medium lock tracking, ask device models instead

Requires new BlockDevOps member is_medium_locked().  Implement for IDE
and SCSI CD-ROMs.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Leave enforcing tray lock to device models

The device model knows best when to accept the guest's eject command.
No need to detour through the block layer.

bdrv_eject() can't fail anymore.  Make it void.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

scsi-disk: Track tray locked state

We already track it in BlockDriverState.  Just like tray open/close
state, we should track it in the device models instead, because it's
device state.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

ide/atapi: Track tray locked state

We already track it in BlockDriverState.  Just like tray open/close
state, we should track it in the device models instead, because it's
device state.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Drop tray status tracking, no longer used

Commit 4be9762a is now completely redone.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Revert entanglement of bdrv_is_inserted() with tray status

Commit 4be9762a changed bdrv_is_inserted() to fail when the tray is
open.  Unfortunately, there are two different kinds of users, with
conflicting needs.

1. Device models using bdrv_eject(), currently ide-cd and scsi-cd.
They expect bdrv_is_inserted() to reflect the tray status.  Commit
4be9762a makes them happy.

2. Code that wants to know whether a BlockDriverState has media, such
as find_image_format(), bdrv_flush_all().  Commit 4be9762a makes them
unhappy.  In particular, it breaks flush on VM stop for media ejected
by the guest.

Revert the change to bdrv_is_inserted().  Check the tray status in the
device models instead.

Note on IDE: Since only ATAPI devices have a tray, and they don't
accept ATA commands since the recent commit "ide: Reject ATA commands
specific to drive kinds", checking in atapi.c suffices.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

scsi-disk: Track tray open/close state

We already track it in BlockDriverState since commit 4be9762a.  As
discussed in that commit's message, we should track it in the device
device models instead, because it's device state.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

scsi-disk: Factor out scsi_disk_emulate_start_stop()

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

ide/atapi: Track tray open/close state

We already track it in BlockDriverState since commit 4be9762a.  As
discussed in that commit's message, we should track it in the device
device models instead, because it's device state.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

ide/atapi: Clean up misleading name in cmd_start_stop_unit()

"eject" is misleading; it means "eject" when start is clear, but
"load" when start is set.  Rename to loej, because that's how MMC-5
calls it, in section 6.40.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

ide: Reject ATA commands specific to drive kinds

ACS-2 Table B.2 explicitly prohibits ATAPI devices from implementing
WIN_RECAL, WIN_READ_EXT, WIN_READDMA_EXT, WIN_READ_NATIVE_MAX,
WIN_MULTREAD_EXT, WIN_WRITE, WIN_WRITE_ONCE, WIN_WRITE_EXT,
WIN_WRITEDMA_EXT, WIN_MULTWRITE_EXT, WIN_WRITE_VERIFY, WIN_VERIFY,
WIN_VERIFY_ONCE, WIN_VERIFY_EXT, WIN_SPECIFY, WIN_MULTREAD,
WIN_MULTWRITE, WIN_SETMULT, WIN_READDMA, WIN_READDMA_ONCE,
WIN_WRITEDMA, WIN_WRITEDMA_ONCE, WIN_FLUSH_CACHE_EXT.  Restrict them
to IDE_HD and IDE_CFATA.

Same for CFA_WRITE_SECT_WO_ERASE, CFA_WRITE_MULTI_WO_ERASE.  Restrict
them to IDE_CFATA, like the other CFA_ commands.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

ide: Use a table to declare which drive kinds accept each command

No functional change.

It would be nice to have handler functions in the table, like commit
e1a064f9 did for ATAPI.  Left for another day.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

ide: Fix ATA command READ to set ATAPI signature for CD-ROM

Must set the ATAPI device signature, see ATA4 8.27.5.2 Outputs for
PACKET Command feature set devices, and ACS-2 7.36.6 Outputs for
PACKET feature set devices.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qcow2: removed unused depends_on field

Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

Gdbstub: Fix back-trace on SPARC32

Gdb expects all registers windows to be flushed in ram, which is not the case
in Qemu. Therefore the back-trace generation doesn't work. This patch adds a
function to handle reads (and only read) in stack frames as if windows were
flushed.

Signed-off-by: Fabien Chouteau <chouteau@adacore.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: add regression testsuite

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

MAINTAINERS: add xtensa maintainer

Add myself as target-xtensa and DC232B maintainer.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: add dc232b core and board

This is Diamond 232L Standard Core Rev.B (LE).

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement boolean option

See ISA, 4.3.9

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement memory protection options

- TLB opcode group;
- region protection option (ISA, 4.6.3);
- region translation option (ISA, 4.6.4);
- MMU option (ISA, 4.6.5).

Cache control attribute bits are not used by this implementation.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: add gdb support

Specific xtensa processor overlay for GDB contains register map in
the gdb/xtensa-config.c. This description is used by the GDB to e.g.
parse 'g' response packets and it may be reused in the qemu's gdbstub
(only XTREG definitions for non-pseudoregisters are needed).

Currently mainline GDB does not support operations with privileged SRs
(see http://sourceware.org/ml/gdb/2011-07/msg00075.html). This support
may be enabled, see NUM_CORE_REGS comment in the gdbstub.c

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement relocatable vectors

See ISA, 4.4.3 for details.

Vector addresses recorded in core configuration are absolute values that
correspond to default VECBASE value.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement CPENABLE and PRID SRs

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement accurate window check

See ISA, 4.7.1.3 for details.

Window check is inserted before commands that push "used register
watermark" beyond its current level. Used register watermark is reset on
instructions that change WINDOW_BASE/WINDOW_START SRs.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement interrupt option

See ISA, 4.4.6 (interrupt option), 4.4.7 (high priority interrupt
option) and 4.4.8 (timer interrupt option) for details.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement SIMCALL

Tensilica iss provides support for applications running in freestanding
environment through SIMCALL command. It is used by Tensilica libc to
access argc/argv, for file I/O, etc.

Note that simcalls that accept buffer addresses expect virtual addresses.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement unaligned exception option

See ISA, 4.4.4 for details.

Correct (aligned as per ISA) address for unaligned access is generated
in case this option is not enabled.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement extended L32R

See ISA, 4.3.3 for details.

TB flag XTENSA_TBFLAG_LITBASE is used to track enable bit of LITBASE SR.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement loop option

See ISA, 4.3.2 for details.

Operations that change LEND SR value invalidate TBs at the old and at
the new LEND. LEND value at TB compilation time is considered constant
and loop instruction is generated based on this value.

Invalidation may be avoided for the TB at the old LEND address, since
looping code verifies actual LEND value.

Invalidation may be avoided for the TB at the new LEND address if
there's a way to associate LEND address with TB at compilation time and
later verify that it doesn't change.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement windowed registers

See ISA, 4.7.1 for details.

Physical registers and currently visible window are separate fields in
CPUEnv. Only current window is accessible to TCG. On operations that
change window base helpers copy current window to and from physical
registers.

Window overflow check described in 4.7.1.3 is in separate patch.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement RST2 group (32 bit mul/div/rem)

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement exceptions

- mark privileged opcodes with ring check;
- make debug exception on exception handler entry.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: add PS register and access control

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement CACHE group

All operations in this group are no-ops, because cache ought to be
transparent to applications. However cache may be abused, then we'll
need to actually implement these opcodes.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>

target-xtensa: implement SYNC group

All operations in this group are no-ops, because there are no delayed
side effects.

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>