hw: Use QEMU_IS_ALIGNED() on parallel flash block size

Use the QEMU_IS_ALIGNED() macro to verify the flash block size
is properly aligned. It is quicker to process when reviewing.

Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200511205246.24621-1-philmd@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests/030: Reduce run time by unthrottling job earlier

test_overlapping_3() throttles its active commit job so it can be sure
the job is still busy when it checks that you can't start a conflicting
streaming job.

However, it only sets the commit job back to full speed when it is
ready, which takes a few seconds while it's throttled. We can already
reset the limit after having checked that block-stream returns an error
and save these seconds.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20200513100025.33543-1-kwolf@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

hw/ide/ahci: Log lost IRQs

One might find interesting to look at AHCI IRQs.

Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20200504094858.5975-1-f4bug@amsat.org>
Reviewed-by: John Snow <jsnow@redhat.com>
Acked-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests: log messages from notrun()

Shift the logging initialization up to occur prior to validation checks,
so that notrun() messages still get printed to console.

(Also, remove the "debugging messages active" message, because we don't
need to see that hundreds of times per iotest suite run.)

Signed-off-by: John Snow <jsnow@redhat.com>
Message-Id: <20200514201614.19941-2-jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/block-copy: Simplify block_copy_do_copy()

block_copy_do_copy() is static, only used in block_copy_task_entry
with the error_is_read argument set. No need to check for it,
simplify.

Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200507121129.29760-3-philmd@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/block-copy: Fix uninitialized variable in block_copy_task_entry

Fix when building with -Os:

    CC      block/block-copy.o
  block/block-copy.c: In function ‘block_copy_task_entry’:
  block/block-copy.c:428:38: error: ‘error_is_read’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
    428 |         t->call_state->error_is_read = error_is_read;
        |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~

Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200507121129.29760-2-philmd@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Drop @child_class from bdrv_child_perm()

Implementations should decide the necessary permissions based on @role.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-35-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Pass BdrvChildRole in remaining cases

These calls have no real use for the child role yet, but it will not
harm to give one.

Notably, the bdrv_root_attach_child() call in blockjob.c is left
unmodified because there is not much the generic BlockJob object wants
from its children.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-34-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Drop child_file

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-33-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Drop bdrv_format_default_perms()

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-32-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Make bdrv_filter_default_perms() static

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-31-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Use bdrv_default_perms()

bdrv_default_perms() can decide which permission profile to use based on
the BdrvChildRole, so block drivers do not need to select it explicitly.

The blkverify driver now no longer shares the WRITE permission for the
image to verify.  We thus have to adjust two places in
test-block-iothread not to take it.  (Note that in theory, blkverify
should behave like quorum in this regard and share neither WRITE nor
RESIZE for both of its children.  In practice, it does not really
matter, because blkverify is used only for debugging, so we might as
well keep its permissions rather liberal.)

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-30-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

tests: Use child_of_bds instead of child_file

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-29-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Use child_of_bds in remaining places

Replace child_file by child_of_bds in all remaining places (excluding
tests).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-28-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Make filter drivers use child_of_bds

Note that some filters have secondary children, namely blkverify (the
image to be verified) and blklogwrites (the log).  This patch does not
touch those children.

Note that for blkverify, the filtered child should not be format-probed.
While there is nothing enforcing this here, in practice, it will not be:
blkverify implements .bdrv_file_open.  The block layer ensures (and in
fact, asserts) that BDRV_O_PROTOCOL is set for every BDS whose driver
implements .bdrv_file_open.  This flag will then be bequeathed to
blkverify's children, and they will thus (by default) not be probed
either.

("By default" refers to the fact that blkverify's other child (the
non-filtered one) will have BDRV_O_PROTOCOL force-unset, because that is
what happens for all non-filtered children of non-format drivers.)

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-27-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Make format drivers use child_of_bds

Commonly, they need to pass the BDRV_CHILD_IMAGE set as the
BdrvChildRole; but there are exceptions for drivers with external data
files (qcow2 and vmdk).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-26-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Drop child_backing

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-25-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Make backing files child_of_bds children

Make all parents of backing files pass the appropriate BdrvChildRole.
By doing so, we can switch their BdrvChildClass over to the generic
child_of_bds, which will do the right thing when given a correct
BdrvChildRole.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-24-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Drop child_format

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-23-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Switch child_format users to child_of_bds

Both users (quorum and blkverify) use child_format for
not-really-filtered children, so the appropriate BdrvChildRole in both
cases is DATA.  (Note that this will cause bdrv_inherited_options() to
force-allow format probing.)

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-22-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

raw-format: Split raw_read_options()

Split raw_read_options() into one function that actually just reads the
options, and another that applies them.  This will allow us to detect
whether the user has specified any options before attaching the file
child (so we can decide on its role based on the options).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-21-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add bdrv_default_perms()

This callback can be used by BDSs that use child_of_bds with the
appropriate BdrvChildRole for their children.

Also, make bdrv_format_default_perms() use it for child_of_bds children
(just a temporary solution until we can drop bdrv_format_default_perms()
altogether).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-20-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Relax *perms_for_storage for data children

We can be less restrictive about pure data children than those with
metadata on them, so let bdrv_default_perms_for_storage() handle
metadata children differently from pure data children.

As explained in the code, the restrictions on metadata children are
strictly stricter than those for pure data children, so in theory we
just have to distinguish between pure-data and all other storage
children (pure metadata or data+metadata).  In practice, that is not
obvious, though, so we have two independent code paths for metadata and
for data children, and data+metadata children will go through both
(without the path for data children doing anything meaningful).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-19-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Pull out bdrv_default_perms_for_storage()

Right now, bdrv_format_default_perms() is used by format parents
(generally). We want to switch to a model where most parents use a
single BdrvChildClass, which then decides the permissions based on the
child role. To do so, we have to split bdrv_format_default_perms() into
separate functions for each such role.

Note that bdrv_default_perms_for_storage() currently handles all DATA |
METADATA children.  A follow-up patch is going to split it further into
one function for each case.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-18-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Pull out bdrv_default_perms_for_cow()

Right now, bdrv_format_default_perms() is used by format parents
(generally). We want to switch to a model where most parents use a
single BdrvChildClass, which then decides the permissions based on the
child role. To do so, we have to split bdrv_format_default_perms() into
separate functions for each such role.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-17-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Distinguish paths in *_format_default_perms

bdrv_format_default_perms() has one code path for backing files, and one
for storage files.  We want to pull them out into their own functions,
so make sure they are completely distinct before so the next patches
will be a bit cleaner.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-16-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add child_of_bds

Any current user of child_file, child_format, and child_backing can and
should use this generic BdrvChildClass instead, as it can handle all of
these cases.  However, to be able to do so, the users must pass the
appropriate BdrvChildRole when the child is created/attached.  (The
following commits will take care of that.)

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-15-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Unify bdrv_child_cb_detach()

Make bdrv_child_cb_detach() call bdrv_backing_detach() for children with
a COW role (and drop the reverse call from bdrv_backing_detach()), so it
can be used for any child (with a proper role set).

Because so far no child has a proper role set, we need a temporary new
callback for child_backing.detach that ensures bdrv_backing_detach() is
called for all COW children that do not have their role set yet.

(Also, move bdrv_child_cb_detach() down to group it with
bdrv_inherited_options() and bdrv_child_cb_attach().)

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-14-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Unify bdrv_child_cb_attach()

Make bdrv_child_cb_attach() call bdrv_backing_attach() for children with
a COW role (and drop the reverse call from bdrv_backing_attach()), so it
can be used for any child (with a proper role set).

Because so far no child has a proper role set, we need a temporary new
callback for child_backing.attach that ensures bdrv_backing_attach() is
called for all COW children that do not have their role set yet.

(Also, move bdrv_child_cb_attach() down to group it with
bdrv_inherited_options().)

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-13-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Use bdrv_inherited_options()

Let child_file's, child_format's, and child_backing's .inherit_options()
implementations fall back to bdrv_inherited_options() to show that it
would really work for all of these cases, if only the parents passed the
appropriate BdrvChildRole and parent_is_format values.

(Also, make bdrv_open_inherit(), the only place to explicitly call
bdrv_backing_options(), call bdrv_inherited_options() instead.)

This patch should incur only two visible changes, both for child_format
children, both of which are effectively bug fixes:

First, they no longer have discard=unmap set by default.  This reason it
was set is because bdrv_inherited_fmt_options() fell through to
bdrv_protocol_options(), and that set it because "format drivers take
care to send flushes and respect unmap policy".  None of the drivers
that use child_format for their children (quorum and blkverify) are
format drivers, though, so this reasoning does not apply here.

Second, they no longer have BDRV_O_NO_IO force-cleared.  child_format
was used solely for children that do not store any metadata and as such
will not be accessed by their parents as long as those parents do not
receive I/O themselves.  Thus, such children should inherit
BDRV_O_NO_IO.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-12-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add generic bdrv_inherited_options()

After the series this patch belongs to, we want to have a common
BdrvChildClass that encompasses all of child_file, child_format, and
child_backing.  Such a single class needs a single .inherit_options()
implementation, and this patch introduces it.

The next patch will show how the existing implementations can fall back
to it just by passing appropriate BdrvChildRole and parent_is_format
values.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-11-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Rename bdrv_inherited_options()

The other two .inherit_options implementations specify exactly for what
case they are used in their name, so do it for this one as well.

(The actual intention behind this patch is to follow it up with a
generic bdrv_inherited_options() that works for all three cases.)

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-10-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Pass parent_is_format to .inherit_options()

We plan to unify the generic .inherit_options() functions.  The
resulting common function will need to decide whether to force-enable
format probing, force-disable it, or leave it as-is.  To make this
decision, it will need to know whether the parent node is a format node
or not (because we never want format probing if the parent is a format
node already (except for the backing chain)).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-9-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Pass BdrvChildRole to .inherit_options()

For now, all callers (effectively) pass 0 and no callee evaluates thie
value.  Later patches will change both.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-8-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Pass BdrvChildRole to bdrv_child_perm()

For now, all callers pass 0 and no callee evaluates this value.  Later
patches will change both.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-7-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add BdrvChildRole to BdrvChild

For now, it is always set to 0.  Later patches in this series will
ensure that all callers pass an appropriate combination of flags.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-6-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add BdrvChildRole and BdrvChildRoleBits

This mask will supplement BdrvChildClass when it comes to what role (or
combination of roles) a child takes for its parent.  It consists of
BdrvChildRoleBits values (which is an enum).

Because empty enums are not allowed, let us just start with it filled.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-5-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Rename BdrvChildRole to BdrvChildClass

This structure nearly only contains parent callbacks for child state
changes.  It cannot really reflect a child's role, because different
roles may overlap (as we will see when real roles are introduced), and
because parents can have custom callbacks even when the child fulfills a
standard role.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-Id: <20200513110544.176672-4-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add BlockDriver.is_format

We want to unify child_format and child_file at some point.  One of the
important things that set format drivers apart from other drivers is
that they do not expect other format nodes under them (except in the
backing chain), i.e. we must not probe formats inside of formats.  That
means we need something on which to distinguish format drivers from
others, and hence this flag.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-Id: <20200513110544.176672-3-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Mark commit, mirror, blkreplay as filters

The commit, mirror, and blkreplay block nodes are filters, so they should
be marked as such.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-2-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Use bdrv_make_empty() where possible

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20200429141126.85159-3-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

replication: Avoid blk_make_empty() on read-only child

This is just a bandaid to keep tests/test-replication working after
bdrv_make_empty() starts to assert that we're not trying to call it on a
read-only child.

For the real solution in the future, replication should not steal the
BdrvChild from its backing file (this is never correct to do!), but
instead have its own child node references, with the appropriate
permissions.

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

block: Use blk_make_empty() after commits

bdrv_commit() already has a BlockBackend pointing to the BDS that we
want to empty, it just has the wrong permissions.

qemu-img commit has no BlockBackend pointing to the old backing file
yet, but introducing one is simple.

After this commit, bdrv_make_empty() is the only remaining caller of
BlockDriver.bdrv_make_empty().

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200429141126.85159-5-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[kwolf: Fixed up reference output for 098]
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add blk_make_empty()

Two callers of BlockDriver.bdrv_make_empty() remain that should not call
this method directly.  Both do not have access to a BdrvChild, but they
can use a BlockBackend, so we add this function that lets them use it.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200429141126.85159-4-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add bdrv_make_empty()

Right now, all users of bdrv_make_empty() call the BlockDriver method
directly.  That is not only bad style, it is also wrong, unless the
caller has a BdrvChild with a WRITE or WRITE_UNCHANGED permission.
(WRITE_UNCHANGED suffices, because callers generally use this function
to clear a node with a backing file after a commit operation.)

Introduce bdrv_make_empty() that verifies that it does.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200429141126.85159-2-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests: Run pylint and mypy in a testcase

We made sure that iotests.py passes pylint. It would be a shame if we
allowed new patches in that break this again, so let's just add a
meta-test case that runs pylint on it.

While we don't pass mypy --strict yet, we can already run it with a few
options that would be part of --strict to make sure that we won't
regress on these aspects at least until we can enable the full thing.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20200511163529.349329-3-kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests: Fix incomplete type declarations

We need to fix only a few places so that iotests.py can pass
mypy --disallow-incomplete-defs, which seems to be a desirable option to
have enabled in the long run.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20200511163529.349329-2-kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/replication.c: Avoid cancelling the job twice

If qemu in colo secondary mode is stopped, it crashes because
s->backup_job is canceled twice: First with job_cancel_sync_all()
in qemu_cleanup() and then in replication_stop().

Fix this by assigning NULL to s->backup_job when the job completes
so replication_stop() and replication_do_checkpoint() won't touch
the job.

Signed-off-by: Lukas Straub <lukasstraub2@web.de>
Message-Id: <20200511090801.7ed5d8f3@luklap>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests: Mirror with different source/target size

This tests that the mirror job catches situations where the target node
has a different size than the source node. It must also forbid resize
operations when the job is already running.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200511135825.219437-5-kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

mirror: Make sure that source and target size match

If the target is shorter than the source, mirror would copy data until
it reaches the end of the target and then fail with an I/O error when
trying to write past the end.

If the target is longer than the source, the mirror job would complete
successfully, but the target wouldn't actually be an accurate copy of
the source image (it would contain some additional garbage at the end).

Fix this by checking that both images have the same size when the job
starts.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200511135825.219437-4-kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests/229: Use blkdebug to inject an error

229 relies on the mirror running into an I/O error when the target is
smaller than the source. After changing mirror to catch this condition
while starting the job, this test case won't get a job that is paused
for an I/O error any more. Use blkdebug instead to inject an error.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200511135825.219437-3-kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

iotests/109: Don't mirror with mismatched size

This patch makes the raw image the same size as the file in a different
format that is mirrored as raw to it to avoid errors when mirror starts
to enforce that source and target are the same size.

We check only that the first 512 bytes are zeroed (instead of 64k)
because some image formats create image files that are smaller than 64k,
so trying to read 64k would result in I/O errors. Apart from this, 512
is more appropriate anyway because the raw format driver protects
specifically the first 512 bytes.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20200511135825.219437-2-kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

Merge remote-tracking branch 'remotes/rth/tags/pull-fpu-20200515' into staging

floatx80 x87 pseudo-denormal fixes

# gpg: Signature made Fri 15 May 2020 19:10:27 BST
# gpg:                using RSA key 7A481E78868B4DB6A85A05C064DF38E8AF7E215F
# gpg:                issuer "richard.henderson@linaro.org"
# gpg: Good signature from "Richard Henderson <richard.henderson@linaro.org>" [full]
# Primary key fingerprint: 7A48 1E78 868B 4DB6 A85A  05C0 64DF 38E8 AF7E 215F

* remotes/rth/tags/pull-fpu-20200515:
  softfloat: fix floatx80 pseudo-denormal round to integer
  softfloat: fix floatx80 pseudo-denormal comparisons
  softfloat: fix floatx80 pseudo-denormal addition / subtraction
  softfloat: silence sNaN for conversions to/from floatx80

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

softfloat: fix floatx80 pseudo-denormal round to integer

The softfloat function floatx80_round_to_int incorrectly handles the
case of a pseudo-denormal where only the high bit of the significand
is set, ignoring that bit (treating the number as an exact zero)
rather than treating the number as an alternative representation of
+/- 2^-16382 (which may round to +/- 1 depending on the rounding mode)
as hardware does.  Fix this check (simplifying the code in the
process).

Signed-off-by: Joseph Myers <joseph@codesourcery.com>
Message-Id: <alpine.DEB.2.21.2005042339420.22972@digraph.polyomino.org.uk>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

softfloat: fix floatx80 pseudo-denormal comparisons

The softfloat floatx80 comparisons fail to allow for pseudo-denormals,
which should compare equal to corresponding values with biased
exponent 1 rather than 0.  Add an adjustment for that case when
comparing numbers with the same sign.

Signed-off-by: Joseph Myers <joseph@codesourcery.com>
Message-Id: <alpine.DEB.2.21.2005042338470.22972@digraph.polyomino.org.uk>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

softfloat: fix floatx80 pseudo-denormal addition / subtraction

The softfloat function addFloatx80Sigs, used for addition of values
with the same sign and subtraction of values with opposite sign, fails
to handle the case where the two values both have biased exponent zero
and there is a carry resulting from adding the significands, which can
occur if one or both values are pseudo-denormals (biased exponent
zero, explicit integer bit 1).  Add a check for that case, so making
the results match those seen on x86 hardware for pseudo-denormals.

Signed-off-by: Joseph Myers <joseph@codesourcery.com>
Message-Id: <alpine.DEB.2.21.2005042337570.22972@digraph.polyomino.org.uk>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

softfloat: silence sNaN for conversions to/from floatx80

Conversions between IEEE floating-point formats should convert
signaling NaNs to quiet NaNs.  Most of those in QEMU's softfloat code
do so, but those for floatx80 fail to.  Fix those conversions to
silence signaling NaNs as well.

Signed-off-by: Joseph Myers <joseph@codesourcery.com>
Message-Id: <alpine.DEB.2.21.2005042336170.22972@digraph.polyomino.org.uk>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Merge remote-tracking branch 'remotes/stsquad/tags/pull-testing-tcg-plugins-150520-2' into staging

Various testing, tcg and plugin updates

  - fix bug in gdbstub tests that leave hanging QEMUs
  - tweak s390x travis test
  - re-factor guest_base handling
  - support "notes" in disassembler output
  - include guest address notes in out_asm
  - cleanup plugin headers and and constify hwaddr
  - updates MAINTAINERS for cpu-common.c

# gpg: Signature made Fri 15 May 2020 15:40:40 BST
# gpg:                using RSA key 6685AE99E75167BCAFC8DF35FBD0DB095A9E2A44
# gpg: Good signature from "Alex Bennée (Master Work Key) <alex.bennee@linaro.org>" [full]
# Primary key fingerprint: 6685 AE99 E751 67BC AFC8  DF35 FBD0 DB09 5A9E 2A44

* remotes/stsquad/tags/pull-testing-tcg-plugins-150520-2:
  MAINTAINERS: update the orphaned cpus-common.c file
  qemu/qemu-plugin: Make qemu_plugin_hwaddr_is_io() hwaddr argument const
  qemu/plugin: Move !CONFIG_PLUGIN stubs altogether
  qemu/plugin: Trivial code movement
  translate-all: include guest address in out_asm output
  disas: add optional note support to cap_disas
  disas: include an optional note for the start of disassembly
  accel/tcg: don't disable exec_tb trace events
  accel/tcg: Relax va restrictions on 64-bit guests
  exec/cpu-all: Use bool for have_guest_base
  linux-user: completely re-write init_guest_space
  travis.yml: Improve the --disable-tcg test on s390x
  tests/guest-debug: catch hanging guests

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

MAINTAINERS: update the orphaned cpus-common.c file

We forgot to update MAINTAINERS when this code was re-factored.

Fixes: 267f685b8b
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200513173200.11830-5-alex.bennee@linaro.org>

qemu/qemu-plugin: Make qemu_plugin_hwaddr_is_io() hwaddr argument const

Rename qemu_plugin_hwaddr_is_io() address argument 'haddr'
similarly to qemu_plugin_hwaddr_device_offset(), and make
it const.

Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20200510171119.20827-4-f4bug@amsat.org>
Message-Id: <20200513173200.11830-4-alex.bennee@linaro.org>

qemu/plugin: Move !CONFIG_PLUGIN stubs altogether

Simplify the ifdef'ry by moving all stubs together.

Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20200510171119.20827-3-f4bug@amsat.org>
Message-Id: <20200513173200.11830-3-alex.bennee@linaro.org>

qemu/plugin: Trivial code movement

Move the qemu_plugin_event enum declaration earlier.
This will make the next commit easier to review.

Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20200510171119.20827-2-f4bug@amsat.org>
Message-Id: <20200513173200.11830-2-alex.bennee@linaro.org>

translate-all: include guest address in out_asm output

We already have information about where each guest instructions
representation starts stored in the tcg_ctx->gen_insn_data so we can
rectify the PC for faults. We can re-use this information to annotate
the out_asm output with guest instruction address which makes it a bit
easier to work out where you are especially with longer blocks. A
minor wrinkle is that some instructions get optimised away so we have
to scan forward until we find some actual generated code.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20200513175134.19619-11-alex.bennee@linaro.org>

disas: add optional note support to cap_disas

Include support for outputting a note at the top of a chunk of
disassembly to capstone as well.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200513175134.19619-10-alex.bennee@linaro.org>

disas: include an optional note for the start of disassembly

This will become useful shortly for providing more information about
output assembly inline. While there fix up the indenting and code
formatting in disas().

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200513175134.19619-9-alex.bennee@linaro.org>

accel/tcg: don't disable exec_tb trace events

I doubt the well predicted trace event check is particularly special in
the grand context of TCG code execution.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200513175134.19619-8-alex.bennee@linaro.org>

accel/tcg: Relax va restrictions on 64-bit guests

We cannot at present limit a 64-bit guest to a virtual address
space smaller than the host.  It will mostly work to ignore this
limitation, except if the guest uses high bits of the address
space for tags.  But it will certainly work better, as presently
we can wind up failing to allocate the guest stack.

Widen our user-only page tree to the host or abi pointer width.
Remove the workaround for this problem from target/alpha.
Always validate guest addresses vs reserved_va, as there we
control allocation ourselves.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20200513175134.19619-7-alex.bennee@linaro.org>

exec/cpu-all: Use bool for have_guest_base

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Message-Id: <20200513175134.19619-6-alex.bennee@linaro.org>

linux-user: completely re-write init_guest_space

First we ensure all guest space initialisation logic comes through
probe_guest_base once we understand the nature of the binary we are
loading. The convoluted init_guest_space routine is removed and
replaced with a number of pgb_* helpers which are called depending on
what requirements we have when loading the binary.

We first try to do what is requested by the host. Failing that we try
and satisfy the guest requested base address. If all those options
fail we fall back to finding a space in the memory map using our
recently written read_self_maps() helper.

There are some additional complications we try and take into account
when looking for holes in the address space. We try not to go directly
after the system brk() space so there is space for a little growth. We
also don't want to have to use negative offsets which would result in
slightly less efficient code on x86 when it's unable to use the
segment offset register.

Less mind-binding gotos and hopefully clearer logic throughout.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Acked-by: Laurent Vivier <laurent@vivier.eu>
Message-Id: <20200513175134.19619-5-alex.bennee@linaro.org>

travis.yml: Improve the --disable-tcg test on s390x

Since the s390x containers do not allow KVM, we only compile-test
the --disable-tcg build on s390x and do not run the qtests. Thus,
it does not make sense to install genisoimage here, and it also does
not make sense to build the s390-ccw.img here again - it is simply
not used without the qtests.
On the other hand, if we do not build the s390-ccw.img anymore, we
can also compile with Clang - so let's use that compiler here to
get some additional test coverage.

Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20200512133849.10624-1-thuth@redhat.com>
Message-Id: <20200513175134.19619-3-alex.bennee@linaro.org>

tests/guest-debug: catch hanging guests

If gdb never actually connected with the guest we need to catch that
and clean-up after ourselves.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20200513175134.19619-2-alex.bennee@linaro.org>

Merge remote-tracking branch 'remotes/kraxel/tags/ui-20200515-pull-request' into staging

ui: sdl bugfix, -show-cursor deprecation message

# gpg: Signature made Fri 15 May 2020 09:21:29 BST
# gpg:                using RSA key 4CB6D8EED3E87138
# gpg: Good signature from "Gerd Hoffmann (work) <kraxel@redhat.com>" [full]
# gpg:                 aka "Gerd Hoffmann <gerd@kraxel.org>" [full]
# gpg:                 aka "Gerd Hoffmann (private) <kraxel@gmail.com>" [full]
# Primary key fingerprint: A032 8CFF B93A 17A7 9901  FE7D 4CB6 D8EE D3E8 7138

* remotes/kraxel/tags/ui-20200515-pull-request:
  ui/sdl2: fix segment fault caused by null pointer dereference
  ui: improve -show-cursor deprecation message

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

Merge remote-tracking branch 'remotes/armbru/tags/pull-qom-2020-05-15' into staging

QOM patches for 2020-05-15

# gpg: Signature made Fri 15 May 2020 06:58:29 BST
# gpg:                using RSA key 354BC8B3D7EB2A6B68674E5F3870B400EB918653
# gpg:                issuer "armbru@redhat.com"
# gpg: Good signature from "Markus Armbruster <armbru@redhat.com>" [full]
# gpg:                 aka "Markus Armbruster <armbru@pond.sub.org>" [full]
# Primary key fingerprint: 354B C8B3 D7EB 2A6B 6867  4E5F 3870 B400 EB91 8653

* remotes/armbru/tags/pull-qom-2020-05-15: (21 commits)
  hw: Remove unnecessary DEVICE() cast
  various: Remove unnecessary OBJECT() cast
  target: Remove unnecessary CPU() cast
  qom: Drop @errp parameter of object_property_del()
  spapr_pci: Drop some dead error handling
  qdev: Unrealize must not fail
  Drop more @errp parameters after previous commit
  qom: Drop parameter @errp of object_property_add() & friends
  qdev: Clean up qdev_connect_gpio_out_named()
  hw/arm/bcm2835: Drop futile attempts at QOM-adopting memory
  e1000: Don't run e1000_instance_init() twice
  hw/isa/superio: Make the components QOM children
  s390x/cpumodel: Fix UI to CPU features pcc-cmac-{aes,eaes}-256
  tests/check-qom-proplist: Improve iterator coverage
  qom: Drop object_property_set_description() parameter @errp
  qom: Make all the object_property_add_FOO() return the property
  qom: Drop convenience method object_property_get_uint16List()
  qom: Simplify object_property_get_enum()
  qom: Drop object_property_del_child()'s unused parameter @errp
  qom: Clean up inconsistent use of gchar * vs. char *
  ...

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

hw: Remove unnecessary DEVICE() cast

The DEVICE() macro is defined as:

  #define DEVICE(obj) OBJECT_CHECK(DeviceState, (obj), TYPE_DEVICE)

which expands to:

  ((DeviceState *)object_dynamic_cast_assert((Object *)(obj), (name),
                                             __FILE__, __LINE__,
                                             __func__))

This assertion can only fail when @obj points to something other
than its stated type, i.e. when we're in undefined behavior country.

Remove the unnecessary DEVICE() casts when we already know the
pointer is of DeviceState type.

Patch created mechanically using spatch with this script:

  @@
  typedef DeviceState;
  DeviceState *s;
  @@
  -   DEVICE(s)
  +   s

Acked-by: David Gibson <david@gibson.dropbear.id.au>
Acked-by: Paul Durrant <paul@xen.org>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Acked-by: John Snow <jsnow@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20200512070020.22782-4-f4bug@amsat.org>

various: Remove unnecessary OBJECT() cast

The OBJECT() macro is defined as:

  #define OBJECT(obj) ((Object *)(obj))

Remove the unnecessary OBJECT() casts when we already know the
pointer is of Object type.

Patch created mechanically using spatch with this script:

  @@
  typedef Object;
  Object *o;
  @@
  -   OBJECT(o)
  +   o

Acked-by: Cornelia Huck <cohuck@redhat.com>
Acked-by: Corey Minyard <cminyard@mvista.com>
Acked-by: John Snow <jsnow@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20200512070020.22782-3-f4bug@amsat.org>
[Trivial rebase conflict in hw/s390x/sclp.c resolved]

target: Remove unnecessary CPU() cast

The CPU() macro is defined as:

  #define CPU(obj) ((CPUState *)(obj))

which expands to:

  ((CPUState *)object_dynamic_cast_assert((Object *)(obj), (name),
                                          __FILE__, __LINE__, __func__))

This assertion can only fail when @obj points to something other
than its stated type, i.e. when we're in undefined behavior country.

Remove the unnecessary CPU() casts when we already know the pointer
is of CPUState type.

Patch created mechanically using spatch with this script:

  @@
  typedef CPUState;
  CPUState *s;
  @@
  -   CPU(s)
  +   s

Acked-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20200512070020.22782-2-f4bug@amsat.org>

qom: Drop @errp parameter of object_property_del()

Same story as for object_property_add(): the only way
object_property_del() can fail is when the property with this name
does not exist.  Since our property names are all hardcoded, failure
is a programming error, and the appropriate way to handle it is
passing &error_abort.  Most callers do that, the commit before
previous fixed one that didn't (and got the error handling wrong), and
the two remaining exceptions ignore errors.

Drop the @errp parameter.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-19-armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>

spapr_pci: Drop some dead error handling

chassis_from_bus() uses object_property_get_uint() to get property
"chassis_nr" of the bridge device.  Failure would be a programming
error.  Pass &error_abort, and simplify its callers.

Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: qemu-ppc@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Acked-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Greg Kurz <groug@kaod.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-18-armbru@redhat.com>

qdev: Unrealize must not fail

Devices may have component devices and buses.

Device realization may fail.  Realization is recursive: a device's
realize() method realizes its components, and device_set_realized()
realizes its buses (which should in turn realize the devices on that
bus, except bus_set_realized() doesn't implement that, yet).

When realization of a component or bus fails, we need to roll back:
unrealize everything we realized so far.  If any of these unrealizes
failed, the device would be left in an inconsistent state.  Must not
happen.

device_set_realized() lets it happen: it ignores errors in the roll
back code starting at label child_realize_fail.

Since realization is recursive, unrealization must be recursive, too.
But how could a partly failed unrealize be rolled back?  We'd have to
re-realize, which can fail.  This design is fundamentally broken.

device_set_realized() does not roll back at all.  Instead, it keeps
unrealizing, ignoring further errors.

It can screw up even for a device with no buses: if the lone
dc->unrealize() fails, it still unregisters vmstate, and calls
listeners' unrealize() callback.

bus_set_realized() does not roll back either.  Instead, it stops
unrealizing.

Fortunately, no unrealize method can fail, as we'll see below.

To fix the design error, drop parameter @errp from all the unrealize
methods.

Any unrealize method that uses @errp now needs an update.  This leads
us to unrealize() methods that can fail.  Merely passing it to another
unrealize method cannot cause failure, though.  Here are the ones that
do other things with @errp:

* virtio_serial_device_unrealize()

  Fails when qbus_set_hotplug_handler() fails, but still does all the
  other work.  On failure, the device would stay realized with its
  resources completely gone.  Oops.  Can't happen, because
  qbus_set_hotplug_handler() can't actually fail here.  Pass
  &error_abort to qbus_set_hotplug_handler() instead.

* hw/ppc/spapr_drc.c's unrealize()

  Fails when object_property_del() fails, but all the other work is
  already done.  On failure, the device would stay realized with its
  vmstate registration gone.  Oops.  Can't happen, because
  object_property_del() can't actually fail here.  Pass &error_abort
  to object_property_del() instead.

* spapr_phb_unrealize()

  Fails and bails out when remove_drcs() fails, but other work is
  already done.  On failure, the device would stay realized with some
  of its resources gone.  Oops.  remove_drcs() fails only when
  chassis_from_bus()'s object_property_get_uint() fails, and it can't
  here.  Pass &error_abort to remove_drcs() instead.

Therefore, no unrealize method can fail before this patch.

device_set_realized()'s recursive unrealization via bus uses
object_property_set_bool().  Can't drop @errp there, so pass
&error_abort.

We similarly unrealize with object_property_set_bool() elsewhere,
always ignoring errors.  Pass &error_abort instead.

Several unrealize methods no longer handle errors from other unrealize
methods: virtio_9p_device_unrealize(),
virtio_input_device_unrealize(), scsi_qdev_unrealize(), ...
Much of the deleted error handling looks wrong anyway.

One unrealize methods no longer ignore such errors:
usb_ehci_pci_exit().

Several realize methods no longer ignore errors when rolling back:
v9fs_device_realize_common(), pci_qdev_unrealize(),
spapr_phb_realize(), usb_qdev_realize(), vfio_ccw_realize(),
virtio_device_realize().

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-17-armbru@redhat.com>

Drop more @errp parameters after previous commit

Several functions can't fail anymore: ich9_pm_add_properties(),
device_add_bootindex_property(), ppc_compat_add_property(),
spapr_caps_add_properties(), PropertyInfo.create().  Drop their @errp
parameter.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-16-armbru@redhat.com>

qom: Drop parameter @errp of object_property_add() & friends

The only way object_property_add() can fail is when a property with
the same name already exists.  Since our property names are all
hardcoded, failure is a programming error, and the appropriate way to
handle it is passing &error_abort.

Same for its variants, except for object_property_add_child(), which
additionally fails when the child already has a parent.  Parentage is
also under program control, so this is a programming error, too.

We have a bit over 500 callers.  Almost half of them pass
&error_abort, slightly fewer ignore errors, one test case handles
errors, and the remaining few callers pass them to their own callers.

The previous few commits demonstrated once again that ignoring
programming errors is a bad idea.

Of the few ones that pass on errors, several violate the Error API.
The Error ** argument must be NULL, &error_abort, &error_fatal, or a
pointer to a variable containing NULL.  Passing an argument of the
latter kind twice without clearing it in between is wrong: if the
first call sets an error, it no longer points to NULL for the second
call.  ich9_pm_add_properties(), sparc32_ledma_realize(),
sparc32_dma_realize(), xilinx_axidma_realize(), xilinx_enet_realize()
are wrong that way.

When the one appropriate choice of argument is &error_abort, letting
users pick the argument is a bad idea.

Drop parameter @errp and assert the preconditions instead.

There's one exception to "duplicate property name is a programming
error": the way object_property_add() implements the magic (and
undocumented) "automatic arrayification".  Don't drop @errp there.
Instead, rename object_property_add() to object_property_try_add(),
and add the obvious wrapper object_property_add().

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-15-armbru@redhat.com>
[Two semantic rebase conflicts resolved]

qdev: Clean up qdev_connect_gpio_out_named()

Both qdev_connect_gpio_out_named() and device_set_realized() put
objects without a parent into the "/machine/unattached/" orphanage.

qdev_connect_gpio_out_named() needs a lengthy comment to explain how
it works.  It exploits that object_property_add_child() can fail only
when we got a parent already, and ignoring that error does what we
want.  True.  If it failed due to "duplicate property", we'd be in
trouble, but that would be a programming error.

device_set_realized() is cleaner: it checks whether we need a parent,
then calls object_property_add_child(), aborting on failure.  No need
for a comment, and programming errors get caught.

Change qdev_connect_gpio_out_named() to match.

Cc: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200505152926.18877-14-armbru@redhat.com>

hw/arm/bcm2835: Drop futile attempts at QOM-adopting memory

The "bcm2835-peripherals" device's .instance_init() method
bcm2835_peripherals_init() attempts to make two memory regions QOM
children of the device.  This is futile, because memory_region_init()
already did.  The errors are ignored (a later commit will change
that).  Drop the useless calls.

Cc: Peter Maydell <peter.maydell@linaro.org>
Cc: Andrew Baumann <Andrew.Baumann@microsoft.com>
Cc: "Philippe Mathieu-Daudé" <philmd@redhat.com>
Cc: qemu-arm@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200505152926.18877-13-armbru@redhat.com>

e1000: Don't run e1000_instance_init() twice

QOM object initialization runs .instance_init() for the type and all
its supertypes; see object_init_with_type().

Both TYPE_E1000_BASE and its concrete subtypes set .instance_init() to
e1000_instance_init().  For the concrete subtypes, it duly gets run
twice.  The second run fails, but the error gets ignored (a later
commit will change that).

Remove it from the subtypes.

Cc: Jason Wang <jasowang@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-12-armbru@redhat.com>

hw/isa/superio: Make the components QOM children

isa_superio_realize() attempts to make isa-parallel and isa-serial QOM
children, but this does not work, because it calls
object_property_add_child() after realizing with qdev_init_nofail().
Realizing a device without a parent gives it one: it gets put into the
"/machine/unattached/" orphanage.  The extra
object_property_add_child() fails, and isa_superio_realize() ignores
the error.

Move the object_property_add_child() before qdev_init_nofail(), and
pass &error_abort.

For the other components, isa_superio_realize() doesn't even try.  Add
object_property_add_child() there.

This affects machines 40p, clipper and fulong2e.

For instance, fulong2e has its vt82c686b-superio (which is an
isa-superio) at /machine/unattached/device[9].  Before the patch, its
components are at /machine/unattached/device[10] .. [14].  Afterwards,
they are at
/machine/unattached/device[9]/{parallel0,serial0,serial1,isa-fdc,i8042}.

Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-11-armbru@redhat.com>

s390x/cpumodel: Fix UI to CPU features pcc-cmac-{aes,eaes}-256

Both s390_features[S390_FEAT_PCC_CMAC_AES_256].name and
s390_features[S390_FEAT_PCC_CMAC_EAES_256].name is
"pcc-cmac-eaes-256".  The former is obviously a pasto.

Impact:

* s390_feat_bitmap_to_ascii() misidentifies S390_FEAT_PCC_CMAC_AES_256
  as "pcc-cmac-eaes-256".  Affects QMP commands query-cpu-definitions,
  query-cpu-model-expansion, query-cpu-model-baseline,
  query-cpu-model-comparison, and the error message when
  s390_realize_cpu_model() fails in check_compatibility().

* s390_cpu_list() also misidentifies it.  Affects -cpu help.

* s390_cpu_model_register_props() creates CPU property
  "pcc-cmac-eaes-256" twice.  The second one fails, but the error is
  ignored (a later commit will change that).  Results in a single
  property "pcc-cmac-eaes-256" with the description for
  S390_FEAT_PCC_CMAC_AES_256, and no property for
  S390_FEAT_PCC_CMAC_EAES_256.  CPU properties are visible in CLI -cpu
  and -device, QMP & HMP device_add, QMP device-list-properties, and
  QOM introspection.

The two features are almost always used via their group msa4.  Such
use is not affected by this bug.

Fix by deleting the wayward 'e'.

Fixes: 782417446279 ("s390x/cpumodel: introduce CPU features")
Cc: Halil Pasic <pasic@linux.ibm.com>
Cc: Cornelia Huck <cohuck@redhat.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: David Hildenbrand <david@redhat.com>
Cc: qemu-s390x@nongnu.org
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com>
Message-Id: <20200505152926.18877-10-armbru@redhat.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
[Lost paragraph in commit message restored, Fixes: tweaked]

tests/check-qom-proplist: Improve iterator coverage

The tests' "qemu-dummy" device has only class properties.  Turn one of
them into an instance property.  test_dummy_class_iterator() expects
one fewer property than test_dummy_iterator().  Rewrite
test_dummy_prop_iterator() to take expected properties as argument.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-9-armbru@redhat.com>

qom: Drop object_property_set_description() parameter @errp

object_property_set_description() and
object_class_property_set_description() fail only when property @name
is not found.

There are 85 calls of object_property_set_description() and
object_class_property_set_description().  None of them can fail:

* 84 immediately follow the creation of the property.

* The one in spapr_rng_instance_init() refers to a property created in
  spapr_rng_class_init(), from spapr_rng_properties[].

Every one of them still gets to decide what to pass for @errp.

51 calls pass &error_abort, 32 calls pass NULL, one receives the error
and propagates it to &error_abort, and one propagates it to
&error_fatal.  I'm actually surprised none of them violates the Error
API.

What are we gaining by letting callers handle the "property not found"
error?  Use when the property is not known to exist is simpler: you
don't have to guard the call with a check.  We haven't found such a
use in 5+ years.  Until we do, let's make life a bit simpler and drop
the @errp parameter.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-8-armbru@redhat.com>
[One semantic rebase conflict resolved]

qom: Make all the object_property_add_FOO() return the property

Some object_property_add_FOO() return the newly added property, some
don't.  Clean that up.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-7-armbru@redhat.com>

qom: Drop convenience method object_property_get_uint16List()

qom/object.c provides object_property_get_TYPE() and
object_property_set_TYPE() for a number of common types.  These are
all convenience wrappers around object_property_get_qobject() and
object_property_set_qobject().

Except for object_property_get_uint16List(), which is unusual in two ways:

* It bypasses object_property_get_qobject().  Fixable; the previous
  commit did it for object_property_get_enum())

* It stores the value through a parameter.  Its contract claims it
  returns the value, like the other functions do.  Also fixable.

Fixing is not worthwhile, though: object_property_get_uint16List() has
seen exactly one user in six years.

Convert the lone user to do its job with the generic
object_property_get_qobject(), and drop object_property_get_uint16List().

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200505152926.18877-6-armbru@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
[Commit message typo fixed]

qom: Simplify object_property_get_enum()

Reuse object_property_get_str().  Switches from the string to the
qobject visitor under the hood.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20200505152926.18877-5-armbru@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>

qom: Drop object_property_del_child()'s unused parameter @errp

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-4-armbru@redhat.com>

qom: Clean up inconsistent use of gchar * vs. char *

Uses of gchar * in qom/object.h:

* ObjectProperty member @name

  Functions that take a property name argument all use char *.  Change
  the member to match.

* ObjectProperty member @type

  Functions that take a property type argument or return it all use
  char *.  Change the member to match.

* ObjectProperty member @description

  Functions that take a property description argument all use char *.
  Change the member to match.

* object_resolve_path_component() parameter @part

  Path components are property names.  Most callers pass char *
  arguments.  Change the parameter to match.  Adjust the few callers
  that pass gchar * to pass char *.

* Return value of object_get_canonical_path_component(),
  object_get_canonical_path()

  Most callers convert their return values right back to char *.
  Change the return value to match.  Adjust the few callers where that
  would add a conversion to gchar * to use char * instead.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-3-armbru@redhat.com>

qom: Clearer reference counting in object_initialize_childv()

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20200505152926.18877-2-armbru@redhat.com>

Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20200514' into staging

target-arm queue:
 * target/arm: Use correct GDB XML for M-profile cores
 * target/arm: Code cleanup to use gvec APIs better
 * aspeed: Add support for the sonorapass-bmc board
 * target/arm: Support reporting KVM host memory errors
   to the guest via ACPI notifications
 * target/arm: Finish conversion of Neon 3-reg-same insns to decodetree

# gpg: Signature made Thu 14 May 2020 15:19:15 BST
# gpg:                using RSA key E1A5C593CD419DE28E8315CF3C2525ED14360CDE
# gpg:                issuer "peter.maydell@linaro.org"
# gpg: Good signature from "Peter Maydell <peter.maydell@linaro.org>" [ultimate]
# gpg:                 aka "Peter Maydell <pmaydell@gmail.com>" [ultimate]
# gpg:                 aka "Peter Maydell <pmaydell@chiark.greenend.org.uk>" [ultimate]
# Primary key fingerprint: E1A5 C593 CD41 9DE2 8E83  15CF 3C25 25ED 1436 0CDE

* remotes/pmaydell/tags/pull-target-arm-20200514: (45 commits)
  target/arm: Convert NEON VFMA, VFMS 3-reg-same insns to decodetree
  target/arm: Convert Neon fp VMAX/VMIN/VMAXNM/VMINNM/VRECPS/VRSQRTS to decodetree
  target/arm: Move 'env' argument of recps_f32 and rsqrts_f32 helpers to usual place
  target/arm: Convert Neon 3-reg-same compare insns to decodetree
  target/arm: Convert Neon fp VMUL, VMLA, VMLS 3-reg-same insns to decodetree
  target/arm: Convert Neon VPMIN/VPMAX/VPADD float 3-reg-same insns to decodetree
  target/arm: Convert Neon VADD, VSUB, VABD 3-reg-same insns to decodetree
  target/arm: Convert Neon VQDMULH/VQRDMULH 3-reg-same to decodetree
  target/arm: Convert Neon VPADD 3-reg-same insns to decodetree
  target/arm: Convert Neon VPMAX/VPMIN 3-reg-same insns to decodetree
  target/arm: Convert Neon VQSHL, VRSHL, VQRSHL 3-reg-same insns to decodetree
  target/arm: Convert Neon VRHADD, VHSUB 3-reg-same insns to decodetree
  target/arm: Convert Neon VABA/VABD 3-reg-same to decodetree
  target/arm: Convert Neon VHADD 3-reg-same insns
  target/arm: Convert Neon 64-bit element 3-reg-same insns
  target/arm: Convert Neon 3-reg-same SHA to decodetree
  target/arm: Convert Neon 3-reg-same VQRDMLAH/VQRDMLSH to decodetree
  MAINTAINERS: Add ACPI/HEST/GHES entries
  target-arm: kvm64: handle SIGBUS signal from kernel or KVM
  ACPI: Record Generic Error Status Block(GESB) table
  ...

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

Merge remote-tracking branch 'remotes/edgar/tags/edgar/xilinx-next-2020-05-14.for-upstream' into staging

Upstream

# gpg: Signature made Thu 14 May 2020 15:04:44 BST
# gpg:                using RSA key AC44FEDC14F7F1EBEDBF415129C596780F6BCA83
# gpg: Good signature from "Edgar E. Iglesias (Xilinx key) <edgar.iglesias@xilinx.com>" [unknown]
# gpg:                 aka "Edgar E. Iglesias <edgar.iglesias@gmail.com>" [full]
# Primary key fingerprint: AC44 FEDC 14F7 F1EB EDBF  4151 29C5 9678 0F6B CA83

* remotes/edgar/tags/edgar/xilinx-next-2020-05-14.for-upstream:
  target/microblaze: monitor: Increase the number of registers reported
  target/microblaze: gdb: Fix incorrect SReg reporting
  target/microblaze: gdb: Extend the number of registers presented to GDB
  target/microblaze: Fix FPU2 instruction check
  target/microblaze: Add MFS Rd,EDR translation
  MAINTAINERS: Add myself as streams maintainer
  hw/dma/xilinx_axidma: s2mm: Support stream fragments
  hw/dma/xilinx_axidma: mm2s: Stream descriptor by descriptor
  hw/net/xilinx_axienet: Handle fragmented packets from DMA
  hw/core: stream: Add an end-of-packet flag
  hw/dma/xilinx_axidma: Add DMA memory-region property
  hw/net/xilinx_axienet: Remove unncessary cast
  hw/net/xilinx_axienet: Cleanup stream->push assignment
  hw/net/xilinx_axienet: Auto-clear PHY Autoneg

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

target/arm: Convert NEON VFMA, VFMS 3-reg-same insns to decodetree

Convert the Neon floating point VFMA and VFMS insn to decodetree.
These are the last insns in the 3-reg-same group so we can
remove all the support/loop code from the old decoder.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20200512163904.10918-18-peter.maydell@linaro.org

target/arm: Convert Neon fp VMAX/VMIN/VMAXNM/VMINNM/VRECPS/VRSQRTS to decodetree

Convert the Neon fp VMAX/VMIN/VMAXNM/VMINNM/VRECPS/VRSQRTS 3-reg-same
insns to decodetree. (These are all the remaining non-accumulation
instructions in this group.)

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20200512163904.10918-17-peter.maydell@linaro.org

target/arm: Move 'env' argument of recps_f32 and rsqrts_f32 helpers to usual place

The usual location for the env argument in the argument list of a TCG helper
is immediately after the return-value argument. recps_f32 and rsqrts_f32
differ in that they put it at the end.

Move the env argument to its usual place; this will allow us to
more easily use these helper functions with the gvec APIs.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20200512163904.10918-16-peter.maydell@linaro.org

target/arm: Convert Neon 3-reg-same compare insns to decodetree

Convert the Neon integer 3-reg-same compare insns VCGE, VCGT,
VCEQ, VACGE and VACGT to decodetree.

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20200512163904.10918-15-peter.maydell@linaro.org