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

Block layer patches

- Fix permission update order problems with block graph changes
- qemu-img convert: Unshare write permission for source
- vhost-user-blk: Fail gracefully on too large queue size

# gpg: Signature made Fri 30 Apr 2021 11:27:51 BST
# gpg:                using RSA key DC3DEB159A9AF95D3D7456FE7F09B272C88F2FD6
# gpg:                issuer "kwolf@redhat.com"
# gpg: Good signature from "Kevin Wolf <kwolf@redhat.com>" [full]
# Primary key fingerprint: DC3D EB15 9A9A F95D 3D74  56FE 7F09 B272 C88F 2FD6

* remotes/kevin/tags/for-upstream: (39 commits)
  vhost-user-blk: Fail gracefully on too large queue size
  qemu-img convert: Unshare write permission for source
  block: Add BDRV_O_NO_SHARE for blk_new_open()
  block: refactor bdrv_node_check_perm()
  block: rename bdrv_replace_child_safe() to bdrv_replace_child()
  block: refactor bdrv_child_set_perm_safe() transaction action
  block: inline bdrv_replace_child()
  block: inline bdrv_check_perm_common()
  block: drop unused permission update functions
  block: bdrv_reopen_multiple: refresh permissions on updated graph
  block: bdrv_reopen_multiple(): move bdrv_flush to separate pre-prepare
  block: add bdrv_set_backing_noperm() transaction action
  block: make bdrv_refresh_limits() to be a transaction action
  block: make bdrv_unset_inherits_from to be a transaction action
  block: drop ignore_children for permission update functions
  block/backup-top: drop .active
  block: introduce bdrv_drop_filter()
  block: add bdrv_remove_filter_or_cow transaction action
  block: adapt bdrv_append() for inserting filters
  block: split out bdrv_replace_node_noperm()
  ...

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

Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20210430' into staging

target-arm queue:
 * hw/pci-host/gpex: Don't fault for unmapped parts of MMIO and PIO windows
 * hw: add compat machines for 6.1
 * Fault misaligned accesses where the architecture requires it
 * Fix some corner cases of MTE faults (notably with misaligned accesses)
 * Make Thumb store insns UNDEF for Rn==1111
 * hw/arm/smmuv3: Support 16K translation granule

# gpg: Signature made Fri 30 Apr 2021 11:33:45 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-20210430: (43 commits)
  hw/pci-host/gpex: Don't fault for unmapped parts of MMIO and PIO windows
  hw: add compat machines for 6.1
  target/arm: Enforce alignment for sve LD1R
  target/arm: Enforce alignment for aa64 vector LDn/STn (single)
  target/arm: Enforce alignment for aa64 vector LDn/STn (multiple)
  target/arm: Use MemOp for size + endian in aa64 vector ld/st
  target/arm: Enforce alignment for aa64 load-acq/store-rel
  target/arm: Use finalize_memop for aa64 fpr load/store
  target/arm: Use finalize_memop for aa64 gpr load/store
  target/arm: Enforce alignment for VLDn/VSTn (single)
  target/arm: Enforce alignment for VLDn/VSTn (multiple)
  target/arm: Enforce alignment for VLDn (all lanes)
  target/arm: Enforce alignment for VLDR/VSTR
  target/arm: Enforce alignment for VLDM/VSTM
  target/arm: Enforce alignment for SRS
  target/arm: Enforce alignment for RFE
  target/arm: Enforce alignment for LDM/STM
  target/arm: Enforce alignment for LDA/LDAH/STL/STLH
  target/arm: Enforce word alignment for LDRD/STRD
  target/arm: Adjust gen_aa32_{ld, st}_i64 for align+endianness
  ...

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

vhost-user-blk: Fail gracefully on too large queue size

virtio_add_queue() aborts when queue_size > VIRTQUEUE_MAX_SIZE, so
vhost_user_blk_device_realize() should check this before calling it.

Simple reproducer:

qemu-system-x86_64 \
    -chardev null,id=foo \
    -device vhost-user-blk-pci,queue-size=4096,chardev=foo

Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1935014
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210413165654.50810-1-kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qemu-img convert: Unshare write permission for source

For a successful conversion of an image, we must make sure that its
content doesn't change during the conversion.

A special case of this is using the same image file both as the source
and as the destination. If both input and output format are raw, the
operation would just be useless work, with other formats it is a sure
way to destroy the image. This will now fail because the image file
can't be opened a second time for the output when opening it for the
input has already acquired file locks to unshare BLK_PERM_WRITE.

Nevertheless, if there is some reason in a special case why it is
actually okay to allow writes to the image while it is being converted,
-U can still be used to force sharing all permissions.

Note that for most image formats, BLK_PERM_WRITE would already be
unshared by the format driver, so this only really makes a difference
for raw source images (but any output format).

Reported-by: Xueqiang Wei <xuwei@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20210422164344.283389-3-kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Add BDRV_O_NO_SHARE for blk_new_open()

Normally, blk_new_open() just shares all permissions. This was fine
originally when permissions only protected against uses in the same
process because no other part of the code would actually get to access
the block nodes opened with blk_new_open(). However, since we use it for
file locking now, unsharing permissions becomes desirable.

Add a new BDRV_O_NO_SHARE flag that is used in blk_new_open() to unshare
any permissions that can be unshared.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210422164344.283389-2-kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: refactor bdrv_node_check_perm()

Now, bdrv_node_check_perm() is called only with fresh cumulative
permissions, so its actually "refresh_perm".

Move permission calculation to the function. Also, drop unreachable
error message and rewrite the remaining one to be more generic (as now
we don't know which node is added and which was already here).

Add also Virtuozzo copyright, as big work is done at this point.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-37-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: rename bdrv_replace_child_safe() to bdrv_replace_child()

We don't have bdrv_replace_child(), so it's time for
bdrv_replace_child_safe() to take its place.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-36-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: refactor bdrv_child_set_perm_safe() transaction action

Old interfaces dropped, nobody directly calls
bdrv_child_set_perm_abort() and bdrv_child_set_perm_commit(), so we can
use personal state structure for the action and stop exploiting
BdrvChild structure. Also, drop "_safe" suffix which is redundant now.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-35-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: inline bdrv_replace_child()

bdrv_replace_child() has only one caller, the second argument is
unused. Inline it now. This triggers deletion of some more unused
interfaces.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-34-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: inline bdrv_check_perm_common()

bdrv_check_perm_common() has only one caller, so no more sense in
"common".

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-33-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: drop unused permission update functions

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-32-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: bdrv_reopen_multiple: refresh permissions on updated graph

Move bdrv_reopen_multiple to new paradigm of permission update:
first update graph relations, then do refresh the permissions.

We have to modify reopen process in file-posix driver: with new scheme
we don't have prepared permissions in raw_reopen_prepare(), so we
should reconfigure fd in raw_check_perm(). Still this seems more native
and simple anyway.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-31-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: bdrv_reopen_multiple(): move bdrv_flush to separate pre-prepare

During reopen we may add backing bs from other aio context, which may
lead to changing original context of top bs.

We are going to move graph modification to prepare stage. So, it will
be possible that bdrv_flush() in bdrv_reopen_prepare called on bs in
non-original aio context, which we didn't aquire which leads to crash.

To avoid this problem move bdrv_flush() to be a separate reopen stage
before bdrv_reopen_prepare().

This doesn't seem correct to acquire only one aio context and not all
contexts participating in reopen. But it's not obvious how to do it
correctly, keeping in mind:

 1. rules of bdrv_set_aio_context_ignore() that requires new_context
    lock not being held

 2. possible deadlocks because of holding all (or several?) AioContext
    locks

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-30-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: add bdrv_set_backing_noperm() transaction action

Split out no-perm part of bdrv_set_backing_hd() as a separate
transaction action. Note the in case of existing BdrvChild we reuse it,
not recreate, just to do less actions.

We don't need to create extra reference to backing_hd as we don't lose
it in bdrv_attach_child().

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-29-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: make bdrv_refresh_limits() to be a transaction action

To be used in further commit.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-28-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: make bdrv_unset_inherits_from to be a transaction action

To be used in the further commit.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-27-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: drop ignore_children for permission update functions

This argument is always NULL. Drop it.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-26-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/backup-top: drop .active

We don't need this workaround anymore: bdrv_append is already smart
enough and we can use new bdrv_drop_filter().

This commit efficiently reverts also recent 705dde27c6c53b73, which
checked .active on io path. Still it said that the problem should be
theoretical. And the logic of filter removement is changed anyway.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-25-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: introduce bdrv_drop_filter()

Using bdrv_replace_node() for removing filter is not good enough: it
keeps child reference of the filter, which may conflict with original
top node during permission update.

Instead let's create new interface, which will do all graph
modifications first and then update permissions.

Let's modify bdrv_replace_node_common(), allowing it additionally drop
backing chain child link pointing to new node. This is quite
appropriate for bdrv_drop_intermediate() and makes possible to add
new bdrv_drop_filter() as a simple wrapper.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-24-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: add bdrv_remove_filter_or_cow transaction action

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-23-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: adapt bdrv_append() for inserting filters

bdrv_append is not very good for inserting filters: it does extra
permission update as part of bdrv_set_backing_hd(). During this update
filter may conflict with other parents of top_bs.

Instead, let's first do all graph modifications and after it update
permissions.

append-greedy-filter test-case in test-bdrv-graph-mod is now works, so
move it out of debug option.

Note: bdrv_append() is still only works for backing-child based
filters. It's something to improve later.

Note2: we use the fact that bdrv_append() is used to append new nodes,
without backing child, so we don't need frozen check and inherits_from
logic from bdrv_set_backing_hd().

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-22-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: split out bdrv_replace_node_noperm()

Split part of bdrv_replace_node_common() to be used separately.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-21-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: add bdrv_attach_child_noperm() transaction action

Split no-perm part of bdrv_attach_child as separate transaction action.
It will be used in later commits.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-20-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: add bdrv_attach_child_common() transaction action

Split out no-perm part of bdrv_root_attach_child() into separate
transaction action. bdrv_root_attach_child() now moves to new
permission update paradigm: first update graph relations then update
permissions.

qsd-jobs test output updated. Seems now permission update goes in
another order. Still, the test comment say that we only want to check
that command doesn't crash, and it's still so.

Error message is a bit misleading as it looks like job was added first.
But actually in new paradigm of graph update we can't distinguish such
things. We should update the error message, but let's not do it now.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20210428151804.439460-19-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: fix bdrv_replace_node_common

inore_children thing doesn't help to track all propagated permissions
of children we want to ignore. The simplest way to correctly update
permissions is update graph first and then do permission update. In
this case we just referesh permissions for the whole subgraph (in
topological-sort defined order) and everything is correctly calculated
automatically without any ignore_children.

So, refactor bdrv_replace_node_common to first do graph update and then
refresh the permissions.

Test test_parallel_exclusive_write() now pass, so move it out of
debugging "if".

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-18-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: add bdrv_replace_child_safe() transaction action

To be used in the following commit.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-17-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: add bdrv_list_* permission update functions

Add new interface, allowing use of existing node list. It will be used
to fix bdrv_replace_node() in the further commit.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-16-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: add bdrv_drv_set_perm transaction action

Refactor calling driver callbacks to a separate transaction action to
be used later.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-15-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: use topological sort for permission update

Rewrite bdrv_check_perm(), bdrv_abort_perm_update() and bdrv_set_perm()
to update nodes in topological sort order instead of simple DFS. With
topologically sorted nodes, we update a node only when all its parents
already updated. With DFS it's not so.

Consider the following example:

    A -+
    |  |
    |  v
    |  B
    |  |
    v  |
    C<-+

A is parent for B and C, B is parent for C.

Obviously, to update permissions, we should go in order A B C, so, when
we update C, all parent permissions already updated. But with current
approach (simple recursion) we can update in sequence A C B C (C is
updated twice). On first update of C, we consider old B permissions, so
doing wrong thing. If it succeed, all is OK, on second C update we will
finish with correct graph. But if the wrong thing failed, we break the
whole process for no reason (it's possible that updated B permission
will be less strict, but we will never check it).

Also new approach gives a way to simultaneously and correctly update
several nodes, we just need to run bdrv_topological_dfs() several times
to add all nodes and their subtrees into one topologically sorted list
(next patch will update bdrv_replace_node() in this manner).

Test test_parallel_perm_update() is now passing, so move it out of
debugging "if".

We also need to support ignore_children in
bdrv_parent_perms_conflict()

For test 283 order of conflicting parents check is changed.

Note also that in bdrv_check_perm() we don't check for parents conflict
at root bs, as we may be in the middle of permission update in
bdrv_reopen_multiple(). bdrv_reopen_multiple() will be updated soon.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-14-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: inline bdrv_child_*() permission functions calls

Each of them has only one caller. Open-coding simplifies further
pemission-update system changes.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-13-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: rewrite bdrv_child_try_set_perm() using bdrv_refresh_perms()

We are going to drop recursive bdrv_child_* functions, so stop use them
in bdrv_child_try_set_perm() as a first step.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-12-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: refactor bdrv_child* permission functions

Split out non-recursive parts, and refactor as block graph transaction
action.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-11-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: bdrv_refresh_perms: check for parents permissions conflict

Add additional check that node parents do not interfere with each
other. This should not hurt existing callers and allows in further
patch use bdrv_refresh_perms() to update a subtree of changed
BdrvChild (check that change is correct).

New check will substitute bdrv_check_update_perm() in following
permissions refactoring, so keep error messages the same to avoid
unit test result changes.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-10-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

util: add transactions.c

Add simple transaction API to use in further update of block graph
operations.

Supposed usage is:

- "prepare" is main function of the action and it should make the main
  effect of the action to be visible for the following actions, keeping
  possibility of roll-back, saving necessary things in action state,
  which is prepended to the action list (to do that, prepare func
  should call tran_add()). So, driver struct doesn't include "prepare"
  field, as it is supposed to be called directly.

- commit/rollback is supposed to be called for the list of action
  states, to commit/rollback all the actions in reverse order

- When possible "commit" should not make visible effect for other
  actions, which make possible transparent logical interaction between
  actions.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-9-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: make bdrv_reopen_{prepare,commit,abort} private

These functions are called only from bdrv_reopen_multiple() in block.c.
No reason to publish them.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-8-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: drop ctx argument from bdrv_root_attach_child

Passing parent aio context is redundant, as child_class and parent
opaque pointer are enough to retrieve it. Drop the argument and use new
bdrv_child_get_parent_aio_context() interface.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-7-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: BdrvChildClass: add .get_parent_aio_context handler

Add new handler to get aio context and implement it in all child
classes. Add corresponding public interface to be used soon.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-6-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: bdrv_append(): don't consume reference

We have too much comments for this feature. It seems better just don't
do it. Most of real users (tests don't count) have to create additional
reference.

Drop also comment in external_snapshot_prepare:
 - bdrv_append doesn't "remove" old bs in common sense, it sounds
   strange
 - the fact that bdrv_append can fail is obvious from the context
 - the fact that we must rollback all changes in transaction abort is
   known (it's the direct role of abort)

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-5-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

tests/test-bdrv-graph-mod: add test_append_greedy_filter

bdrv_append() is not quite good for inserting filters: it does extra
permission update in intermediate state, where filter get it filtered
child but is not yet replace it in a backing chain.

Some filters (for example backup-top) may want permissions even when
have no parents. And described intermediate state becomes invalid.

That's (half a) reason, why we need "inactive" state for backup-top
filter.

bdrv_append() will be improved later, now let's add a unit test.

Now test fails, so it runs only with -d flag. To run do

  ./test-bdrv-graph-mod -d -p /bdrv-graph-mod/append-greedy-filter

from <build-directory>/tests.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-4-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

tests/test-bdrv-graph-mod: add test_parallel_perm_update

Add test to show that simple DFS recursion order is not correct for
permission update. Correct order is topological-sort order, which will
be introduced later.

Consider the block driver which has two filter children: one active
with exclusive write access and one inactive with no specific
permissions.

And, these two children has a common base child, like this:

┌─────┐     ┌──────┐
│ fl2 │ ◀── │ top  │
└─────┘     └──────┘
  │           │
  │           │ w
  │           ▼
  │         ┌──────┐
  │         │ fl1  │
  │         └──────┘
  │           │
  │           │ w
  │           ▼
  │         ┌──────┐
  └───────▶ │ base │
            └──────┘

So, exclusive write is propagated.

Assume, we want to make fl2 active instead of fl1.
So, we set some option for top driver and do permission update.

If permission update (remember, it's DFS) goes first through
top->fl1->base branch it will succeed: it firstly drop exclusive write
permissions and than apply them for another BdrvChildren.
But if permission update goes first through top->fl2->base branch it
will fail, as when we try to update fl2->base child, old not yet
updated fl1->base child will be in conflict.

Now test fails, so it runs only with -d flag. To run do

  ./test-bdrv-graph-mod -d -p /bdrv-graph-mod/parallel-perm-update

from <build-directory>/tests.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-3-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

tests/test-bdrv-graph-mod: add test_parallel_exclusive_write

Add the test that shows that concept of ignore_children is incomplete.
Actually, when we want to update something, ignoring permission of some
existing BdrvChild, we should ignore also the propagated effect of this
child to the other children. But that's not done. Better approach
(update permissions on already updated graph) will be implemented
later.

Now the test fails, so it's added with -d argument to not break make
check.

Test fails with

 "Conflicts with use by fl1 as 'backing', which does not allow 'write' on base"

because when updating permissions we can ignore original top->fl1
BdrvChild. But we don't ignore exclusive write permission in fl1->base
BdrvChild, which is propagated. Correct thing to do is make graph
change first and then do permission update from the top node.

To run test do

  ./test-bdrv-graph-mod -d -p /bdrv-graph-mod/parallel-exclusive-write

from <build-directory>/tests.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210428151804.439460-2-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

hw/pci-host/gpex: Don't fault for unmapped parts of MMIO and PIO windows

Currently the gpex PCI controller implements no special behaviour for
guest accesses to areas of the PIO and MMIO where it has not mapped
any PCI devices, which means that for Arm you end up with a CPU
exception due to a data abort.

Most host OSes expect "like an x86 PC" behaviour, where bad accesses
like this return -1 for reads and ignore writes.  In the interests of
not being surprising, make host CPU accesses to these windows behave
as -1/discard where there's no mapped PCI device.

The old behaviour generally didn't cause any problems, because
almost always the guest OS will map the PCI devices and then only
access where it has mapped them. One corner case where you will see
this kind of access is if Linux attempts to probe legacy ISA
devices via a PIO window access. So far the only case where we've
seen this has been via the syzkaller fuzzer.

Reported-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Message-id: 20210325163315.27724-1-peter.maydell@linaro.org
Fixes: https://bugs.launchpad.net/qemu/+bug/1918917
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

hw: add compat machines for 6.1

Add 6.1 machine types for arm/i440fx/q35/s390x/spapr.

Signed-off-by: Cornelia Huck <cohuck@redhat.com>
Acked-by: Greg Kurz <groug@kaod.org>
Message-id: 20210331111900.118274-1-cohuck@redhat.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Enforce alignment for sve LD1R

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-32-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Enforce alignment for aa64 vector LDn/STn (single)

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-31-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Enforce alignment for aa64 vector LDn/STn (multiple)

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-30-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Use MemOp for size + endian in aa64 vector ld/st

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-29-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Enforce alignment for aa64 load-acq/store-rel

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-28-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Use finalize_memop for aa64 fpr load/store

For 128-bit load/store, use 16-byte alignment.  This
requires that we perform the two operations in the
correct order so that we generate the alignment fault
before modifying memory.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-27-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Use finalize_memop for aa64 gpr load/store

In the case of gpr load, merge the size and is_signed arguments;
otherwise, simply convert size to memop.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-26-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Enforce alignment for VLDn/VSTn (single)

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-25-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Enforce alignment for VLDn/VSTn (multiple)

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-24-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Enforce alignment for VLDn (all lanes)

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-23-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Enforce alignment for VLDR/VSTR

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-22-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Enforce alignment for VLDM/VSTM

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-21-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Enforce alignment for SRS

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-20-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Enforce alignment for RFE

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-19-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Enforce alignment for LDM/STM

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-18-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Enforce alignment for LDA/LDAH/STL/STLH

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-17-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Enforce word alignment for LDRD/STRD

Buglink: https://bugs.launchpad.net/qemu/+bug/1905356
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-16-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Adjust gen_aa32_{ld, st}_i64 for align+endianness

Adjust the interface to match what has been done to the
TCGv_i32 load/store functions.

This is less obvious, because at present the only user of
these functions, trans_VLDST_multiple, also wants to manipulate
the endianness to speed up loading multiple bytes.  Thus we
retain an "internal" interface which is identical to the
current gen_aa32_{ld,st}_i64 interface.

The "new" interface will gain users as we remove the legacy
interfaces, gen_aa32_ld64 and gen_aa32_st64.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-15-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Fix SCTLR_B test for TCGv_i64 load/store

Just because operating on a TCGv_i64 temporary does not
mean that we're performing a 64-bit operation.  Restrict
the frobbing to actual 64-bit operations.

This bug is not currently visible because all current
users of these two functions always pass MO_64.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-14-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Merge gen_aa32_frob64 into gen_aa32_ld_i64

This is the only caller.  Adjust some commentary to talk
about SCTLR_B instead of the vanishing function.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-13-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Adjust gen_aa32_{ld, st}_i32 for align+endianness

Create a finalize_memop function that computes alignment and
endianness and returns the final MemOp for the operation.

Split out gen_aa32_{ld,st}_internal_i32 which bypasses any special
handling of endianness or alignment.  Adjust gen_aa32_{ld,st}_i32
so that s->be_data is not added by the callers.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-12-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Add ALIGN_MEM to TBFLAG_ANY

Use this to signal when memory access alignment is required.
This value comes from the CCR register for M-profile, and
from the SCTLR register for A-profile.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-11-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Move TBFLAG_ANY bits to the bottom

Now that other bits have been moved out of tb->flags,
there's no point in filling from the top.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-10-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Move TBFLAG_AM32 bits to the top

Now that these bits have been moved out of tb->flags,
where TBFLAG_ANY was filling from the top, move AM32
to fill from the top, and A32 and M32 to fill from the
bottom.  This means fewer changes when adding new bits.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-9-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Move mode specific TB flags to tb->cs_base

Now that we have all of the proper macros defined, expanding
the CPUARMTBFlags structure and populating the two TB fields
is relatively simple.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-7-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Introduce CPUARMTBFlags

In preparation for splitting tb->flags across multiple
fields, introduce a structure to hold the value(s).
So far this only migrates the one uint32_t and fixes
all of the places that require adjustment to match.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-6-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Add wrapper macros for accessing tbflags

We're about to split tbflags into two parts.  These macros
will ensure that the correct part is used with the correct
set of bits.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-5-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Rename TBFLAG_ANY, PSTATE_SS

We're about to rearrange the macro expansion surrounding tbflags,
and this field name will be expanded using the bit definition of
the same name, resulting in a token pasting error.

So PSTATE_SS -> PSTATE__SS in the uses, and document it.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-4-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Rename TBFLAG_A32, SCTLR_B

We're about to rearrange the macro expansion surrounding tbflags,
and this field name will be expanded using the bit definition of
the same name, resulting in a token pasting error.

So SCTLR_B -> SCTLR__B in the 3 uses, and document it.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-3-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Fix decode of align in VLDST_single

The encoding of size = 2 and size = 3 had the incorrect decode
for align, overlapping the stride field.  This error was hidden
by what should have been unnecessary masking in translate.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210419202257.161730-2-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Remove log2_esize parameter to gen_mte_checkN

The log2_esize parameter is not used except trivially.
Drop the parameter and the deferral to gen_mte_check1.

This fixes a bug in that the parameters as documented
in the header file were the reverse from those in the
implementation.  Which meant that translate-sve.c was
passing the parameters in the wrong order.

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210416183106.1516563-10-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Simplify sve mte checking

Now that mte_check1 and mte_checkN have been merged, we can
merge sve_cont_ldst_mte_check1 and sve_cont_ldst_mte_checkN.

Which means that we can eliminate the function pointer into
sve_ldN_r and sve_stN_r, calling sve_cont_ldst_mte_check directly.

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210416183106.1516563-9-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Rename mte_probe1 to mte_probe

For consistency with the mte_check1 + mte_checkN merge
to mte_check, rename the probe function as well.

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210416183106.1516563-8-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Merge mte_check1, mte_checkN

The mte_check1 and mte_checkN functions are now identical.
Drop mte_check1 and rename mte_checkN to mte_check.

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210416183106.1516563-7-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Replace MTEDESC ESIZE+TSIZE with SIZEM1

After recent changes, mte_checkN does not use ESIZE,
and mte_check1 never used TSIZE.  We can combine the
two into a single field: SIZEM1.

Choose to pass size - 1 because size == 0 is never used,
our immediate need in mte_probe_int is for the address
of the last byte (ptr + size - 1), and since almost all
operations are powers of 2, this makes the immediate
constant one bit smaller.

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210416183106.1516563-6-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

test/tcg/aarch64: Add mte-5

Buglink: https://bugs.launchpad.net/bugs/1921948
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210416183106.1516563-5-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Fix unaligned checks for mte_check1, mte_probe1

We were incorrectly assuming that only the first byte of an MTE access
is checked against the tags.  But per the ARM, unaligned accesses are
pre-decomposed into single-byte accesses.  So by the time we reach the
actual MTE check in the ARM pseudocode, all accesses are aligned.

We cannot tell a priori whether or not a given scalar access is aligned,
therefore we must at least check.  Use mte_probe_int, which is already
set up for checking multiple granules.

Buglink: https://bugs.launchpad.net/bugs/1921948
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210416183106.1516563-4-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Split out mte_probe_int

Split out a helper function from mte_checkN to perform
all of the checking and address manpulation.  So far,
just use this in mte_checkN itself.

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210416183106.1516563-3-richard.henderson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Fix mte_checkN

We were incorrectly assuming that only the first byte of an MTE access
is checked against the tags.  But per the ARM, unaligned accesses are
pre-decomposed into single-byte accesses.  So by the time we reach the
actual MTE check in the ARM pseudocode, all accesses are aligned.

Therefore, the first failure is always either the first byte of the
access, or the first byte of the granule.

In addition, some of the arithmetic is off for last-first -> count.
This does not become directly visible until a later patch that passes
single bytes into this function, so ptr == ptr_last.

Buglink: https://bugs.launchpad.net/bugs/1921948
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210416183106.1516563-2-richard.henderson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
[PMM: tweaked a comment]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Make Thumb store insns UNDEF for Rn==1111

The Arm ARM specifies that for Thumb encodings of the various plain
store insns, if the Rn field is 1111 then we must UNDEF.  This is
different from the Arm encodings, where this case is either
UNPREDICTABLE or has well-defined behaviour.  The exclusive stores,
store-release and STRD do not have this UNDEF case for any encoding.

Enforce the UNDEF for this case in the Thumb plain store insns.

Fixes: https://bugs.launchpad.net/qemu/+bug/1922887
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210408162402.5822-1-peter.maydell@linaro.org

hw/arm/smmuv3: Support 16K translation granule

The driver can query some bits in SMMUv3 IDR5 to learn which
translation granules are supported. Arm recommends that SMMUv3
implementations support at least 4K and 64K granules. But in
the vSMMUv3, there seems to be no reason not to support 16K
translation granule. In addition, if 16K is not supported,
vSVA will failed to be enabled in the future for 16K guest
kernel. So it'd better to support it.

Signed-off-by: Kunkun Jiang <jiangkunkun@huawei.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Tested-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Open 6.1 development tree

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

Update version for v6.0.0 release

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

Update version for v6.0.0-rc5 release

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

hw/block/nvme: fix invalid msix exclusive uninit

Commit 1901b4967c3f changed the nvme device from using a bar exclusive
for MSI-x to sharing it on bar0.

Unfortunately, the msix_uninit_exclusive_bar() call remains in
nvme_exit() which causes havoc when the device is removed with, say,
device_del. Fix this.

Additionally, a subregion is added but it is not removed on exit which
causes a reference to linger and the drive to never be unlocked.

Fixes: 1901b4967c3f ("hw/block/nvme: move msix table and pba to BAR 0")
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/s390x: fix s390_probe_access to check PAGE_WRITE_ORG for writeability

We can remove PAGE_WRITE when (internally) marking a page read-only
because it contains translated code. This can get confused when we are
executing signal return code on signal stacks.

Fixes: e56552cf07 ("target/s390x: Implement the MVPG condition-code-option bit")
Found-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Message-id: 20210422154427.13038-1-alex.bennee@linaro.org
Cc: Cornelia Huck <cohuck@redhat.com>
Cc: Thomas Huth <thuth@redhat.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

net: check the existence of peer before trying to pad

There could be case that peer is NULL. This can happen when during
network device hot-add where net device needs to be added first. So
the patch check the existence of peer before trying to do the pad.

Fixes: 969e50b61a285 ("net: Pad short frames to minimum size before sending from SLiRP/TAP")
Signed-off-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Stefan Weil <sw@weilnetz.de>
Message-id: 20210423031803.1479-1-jasowang@redhat.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Update version for v6.0.0-rc4 release

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

qga: fix guest-get-disks regression

Commit 54aa3de72 ("qapi: Use QAPI_LIST_PREPEND() where possible")
inadvertently removed the has_dependencies from the partition disk
info, resulting in empty list being returned.

Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1950833
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Message-id: 20210420125831.233092-1-marcandre.lureau@redhat.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/mips/rel6_translate: Change license to GNU LGPL v2.1 (or later)

When adding this file and its new content in commit 3f7a927847a
("target/mips: LSA/DLSA R6 decodetree helpers") I did 2 mistakes:

1: Listed authors who haven't been involved in its development,
2: Used an incorrect GNU GPLv2 license text (using 'and' instead
   of 'or').

Instead of correcting the GNU GPLv2 license text, replace the license
by the 'GNU LGPL v2.1 or later' one, to be coherent with the other
translation files in the target/mips/ folder.

Suggested-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20210420100633.1752440-1-f4bug@amsat.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

migration: Deprecate redundant query-migrate result @blocked

Result @blocked is true when and only when result @blocked-reasons is
present.  It's always non-empty when present.  @blocked is redundant.
It was introduced in commit 3af8554bd0 "migration: Add blocker
information", and has not been released.  This gives us a chance to
fix the interface with minimal fuss.

Unfortunately, we're already too close to the release to risk dropping
it.  Deprecate it instead.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-id: 20210420051907.891470-1-armbru@redhat.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

mptsas: Remove unused MPTSASState 'pending' field (CVE-2021-3392)

While processing SCSI i/o requests in mptsas_process_scsi_io_request(),
the Megaraid emulator appends new MPTSASRequest object 'req' to
the 's->pending' queue. In case of an error, this same object gets
dequeued in mptsas_free_request() only if SCSIRequest object
'req->sreq' is initialised. This may lead to a use-after-free issue.

Since s->pending is actually not used, simply remove it from
MPTSASState.

Cc: qemu-stable@nongnu.org
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reported-by: Cheolwoo Myung <cwmyung@snu.ac.kr>
Message-id: 20210419134247.1467982-1-f4bug@amsat.org
Message-Id: <20210416102243.1293871-1-mjt@msgid.tls.msk.ru>
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Reported-by: Cheolwoo Myung <cwmyung@snu.ac.kr>
BugLink: https://bugs.launchpad.net/qemu/+bug/1914236
Fixes: e351b826112 ("hw: Add support for LSI SAS1068 (mptsas) device")
[PMD: Reworded description, added more tags]
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20210417' into staging

Fixes for rc4:
 * Fix compile failures of C++ files with new glib headers
 * mps3-an547: Use correct Cortex-M55 CPU and don't disable its FPU
 * accel/tcg: Fix assertion failure executing from non-RAM with -icount

# gpg: Signature made Sat 17 Apr 2021 20:39:58 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-20210417:
  accel/tcg: avoid re-translating one-shot instructions
  target/arm: drop CF_LAST_IO/dc->condjump check
  hw/arm/armsse: Make SSE-300 use Cortex-M55
  hw/arm/armsse: Give SSE-300 its own Property array
  include/qemu/osdep.h: Move system includes to top
  osdep: protect qemu/osdep.h with extern "C"
  osdep: include glib-compat.h before other QEMU headers

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

accel/tcg: avoid re-translating one-shot instructions

By definition a single instruction is capable of being an IO
instruction. This avoids a problem of triggering a cpu_io_recompile on
a non-recorded translation which then fails because it expects
tcg_tb_lookup() to succeed unconditionally. The normal use case
requires a TB to be able to resolve machine state.

The other users of tcg_tb_lookup() are able to tolerate a missing TB
if the machine state has been resolved by other means - which in the
single-shot case is always true because machine state is synced at the
start of a block.

Reported-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210415162454.22056-1-alex.bennee@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: drop CF_LAST_IO/dc->condjump check

This is a left over erroneous check from the days front-ends handled
io start/end themselves. Regardless just because IO could be performed
on the last instruction doesn't obligate the front end to do so.

This fixes an abort faced by the aspeed execute-in-place support which
will necessarily trigger this state (even before the one-shot
CF_LAST_IO fix). The test still seems to hang once it attempts to boot
the Linux kernel but I suspect this is an unrelated issue with icount
and the timer handling code.

The original intention of the cpu_abort (added in commit 2e70f6efa8b9
when the icount stuff was first added) seems to have been to act as
an assert() to catch an unhandled corner case where the generated code
would be something like:
    conditional branch to condlabel if its cc failed
    implementation of the insn (a conditional branch or trap)
    code emitted by gen_io_end()
 condlabel:
    gen_goto_tb or equivalent thing to go to next insn

At runtime the cc-failed case would skip over the code emitted by
gen_io_end(), leaving the can_do_io flag incorrectly set.

In commit ba3e7926691ed33 we switched to an implementation which
always clears can_do_io at the start of the following TB instead
of trying to clear it at the end of a TB that did IO. So the corner
case that this cpu_abort() was trying to flag is no longer possible,
because the gen_io_end() call has been deleted. We can therefore
safely remove the no-longer-valid assertion.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20210416170207.12504-1-alex.bennee@linaro.org
Cc: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

hw/arm/armsse: Make SSE-300 use Cortex-M55

The SSE-300 has a Cortex-M55 (which was the whole reason for us
modelling it), but we forgot to actually update the code to let it
have a different CPU type from the IoTKit and SSE-200.  Add CPU type
as a field for ARMSSEInfo instead of hardcoding it to always use a
Cortex-M33.

Buglink: https://bugs.launchpad.net/qemu/+bug/1923861
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210416104010.13228-1-peter.maydell@linaro.org

hw/arm/armsse: Give SSE-300 its own Property array

SSE-300 currently shares the SSE-200 Property array. This is
bad principally because the default values of the CPU0_FPU
and CPU0_DSP properties disable the FPU and DSP on the CPU.
That is correct for the SSE-200 but not the SSE-300.
Give the SSE-300 its own Property array with the correct
SSE-300 specific settings:
 * SSE-300 has only one CPU, so no CPU1* properties
 * SSE-300 CPU has FPU and DSP

Buglink: https://bugs.launchpad.net/qemu/+bug/1923861
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20210415182353.8173-1-peter.maydell@linaro.org