OSDN Git Service

qmiga/qemu.git
6 years agoMerge remote-tracking branch 'remotes/mdroth/tags/qga-pull-2017-10-26-tag' into staging
Peter Maydell [Sat, 28 Oct 2017 08:59:38 +0000 (09:59 +0100)]
Merge remote-tracking branch 'remotes/mdroth/tags/qga-pull-2017-10-26-tag' into staging

qemu-ga patch queue for 2.11

* support for network interface stats
* w32: improvements for guest-set-time
* w32: fix a hang with guest-fsfreeze-freeze when timeout occurs
  during heavy I/O
* w32: fix faulty error-handling in VSS/fsfreeze COM registration

# gpg: Signature made Fri 27 Oct 2017 02:11:53 BST
# gpg:                using RSA key 0x3353C9CEF108B584
# gpg: Good signature from "Michael Roth <flukshun@gmail.com>"
# gpg:                 aka "Michael Roth <mdroth@utexas.edu>"
# gpg:                 aka "Michael Roth <mdroth@linux.vnet.ibm.com>"
# Primary key fingerprint: CEAC C9E1 5534 EBAB B82D  3FA0 3353 C9CE F108 B584

* remotes/mdroth/tags/qga-pull-2017-10-26-tag:
  qga-win: fix error-handling in getNameByStringSID()
  qga: add network stats to guest-network-get-interfaces
  qga-win: Updating guest_set_time action
  qga-win: don't hang if vss hold writes timeout

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/sstabellini/tags/xen-20171026-tag' into staging
Peter Maydell [Fri, 27 Oct 2017 13:43:02 +0000 (14:43 +0100)]
Merge remote-tracking branch 'remotes/sstabellini/tags/xen-20171026-tag' into staging

Xen 2017/10/26

# gpg: Signature made Thu 26 Oct 2017 23:57:16 BST
# gpg:                using RSA key 0x894F8F4870E1AE90
# gpg: Good signature from "Stefano Stabellini <stefano.stabellini@eu.citrix.com>"
# gpg:                 aka "Stefano Stabellini <sstabellini@kernel.org>"
# Primary key fingerprint: D04E 33AB A51F 67BA 07D3  0AEA 894F 8F48 70E1 AE90

* remotes/sstabellini/tags/xen-20171026-tag:
  xen: Log errno rather than return value
  xen: dont try setting max grants multiple times
  xen: add a global indicator for grant copy being available

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging
Peter Maydell [Fri, 27 Oct 2017 08:29:05 +0000 (09:29 +0100)]
Merge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging

Block layer patches

# gpg: Signature made Thu 26 Oct 2017 14:02:54 BST
# gpg:                using RSA key 0x7F09B272C88F2FD6
# gpg: Good signature from "Kevin Wolf <kwolf@redhat.com>"
# Primary key fingerprint: DC3D EB15 9A9A F95D 3D74  56FE 7F09 B272 C88F 2FD6

* remotes/kevin/tags/for-upstream: (35 commits)
  iotests: Add cluster_size=64k to 125
  qcow2: Always execute preallocate() in a coroutine
  qcow2: Fix unaligned preallocated truncation
  qcow2: Emit errp when truncating the image tail
  iotests: Filter actual image size in 184 and 191
  iotests: Pull _filter_actual_image_size from 67/87
  iotests: Add test for dataplane mirroring
  qcow2: Use BDRV_SECTOR_BITS instead of its literal value
  qemu-img.1: Image invalidation on qemu-img commit
  qemu-io: Relax 'alloc' now that block-status doesn't assert
  qcow2: Reduce is_zero() rounding
  block: Reduce bdrv_aligned_preadv() rounding
  block: Align block status requests
  qemu-img: Change img_compare() to be byte-based
  qemu-img: Change img_rebase() to be byte-based
  qemu-img: Change compare_sectors() to be byte-based
  qemu-img: Change check_empty_sectors() to byte-based
  qemu-img: Drop redundant error message in compare
  qemu-img: Add find_nonzero()
  qemu-img: Speed up compare on pre-allocated larger file
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/rth/tags/pull-dis-20171026' into staging
Peter Maydell [Fri, 27 Oct 2017 07:04:51 +0000 (08:04 +0100)]
Merge remote-tracking branch 'remotes/rth/tags/pull-dis-20171026' into staging

Capstone disassembler

# gpg: Signature made Thu 26 Oct 2017 10:57:27 BST
# gpg:                using RSA key 0x64DF38E8AF7E215F
# gpg: Good signature from "Richard Henderson <richard.henderson@linaro.org>"
# Primary key fingerprint: 7A48 1E78 868B 4DB6 A85A  05C0 64DF 38E8 AF7E 215F

* remotes/rth/tags/pull-dis-20171026:
  disas: Add capstone as submodule
  disas: Remove monitor_disas_is_physical
  ppc: Support Capstone in disas_set_info
  arm: Support Capstone in disas_set_info
  i386: Support Capstone in disas_set_info
  disas: Support the Capstone disassembler library
  disas: Remove unused flags arguments
  target/arm: Don't set INSN_ARM_BE32 for CONFIG_USER_ONLY
  target/arm: Move BE32 disassembler fixup
  target/ppc: Convert to disas_set_info hook
  target/i386: Convert to disas_set_info hook

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
# Conflicts:
# target/i386/cpu.c
# target/ppc/translate_init.c

6 years agoqga-win: fix error-handling in getNameByStringSID()
Michael Roth [Fri, 27 Oct 2017 00:53:45 +0000 (19:53 -0500)]
qga-win: fix error-handling in getNameByStringSID()

In one case we misconstrue a BOOL return as an HRESULT, and in the
other case we don't check the BOOL return from LookupAccountSidW()
before extracting the HRESULT from GetLastError(). Both can lead to
getNameByStringSID() misreporting an error.

Reported-by: Chen Hanxiao <chenhanxiao@gmail.com>
Suggested-by: Tomáš Golembiovský <tgolembi@redhat.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
6 years agoxen: Log errno rather than return value
Ross Lagerwall [Wed, 11 Oct 2017 15:52:03 +0000 (16:52 +0100)]
xen: Log errno rather than return value

xen_modified_memory() sets errno to communicate what went wrong so log
this rather than the return value which is not interesting.

Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Acked-by: Anthony PERARD <anthony.perard@citrix.com>
Signed-off-by: Stefano Stabellini <sstabellini@kernel.org>
6 years agoxen: dont try setting max grants multiple times
Juergen Gross [Fri, 22 Sep 2017 12:07:25 +0000 (14:07 +0200)]
xen: dont try setting max grants multiple times

Trying to call xengnttab_set_max_grants() with the same file handle
might fail on some kernels, as this operation is allowed only once.

This is a problem for the qdisk backend as blk_connect() can be
called multiple times for a domain, e.g. in case grub-xen is being
used to boot it.

So instead of letting the generic backend code open the gnttab device
do it in blk_connect() and close it again in blk_disconnect.

Signed-off-by: Juergen Gross <jgross@suse.com>
Acked-by: Anthony PERARD <anthony.perard@citrix.com>
Signed-off-by: Stefano Stabellini <sstabellini@kernel.org>
6 years agoxen: add a global indicator for grant copy being available
Juergen Gross [Fri, 22 Sep 2017 12:07:24 +0000 (14:07 +0200)]
xen: add a global indicator for grant copy being available

The Xen qdisk backend needs to test whether grant copy operations is
available in the kernel. Unfortunately this collides with using
xengnttab_set_max_grants() on some kernels as this operation has to
be the first one after opening the gnttab device.

In order to solve this problem test for the availability of grant copy
in xen_be_init() opening the gnttab device just for that purpose and
closing it again afterwards. Advertise the availability via a global
flag and use that flag in the qdisk backend.

Signed-off-by: Juergen Gross <jgross@suse.com>
Acked-by: Anthony PERARD <anthony.perard@citrix.com>
Signed-off-by: Stefano Stabellini <sstabellini@kernel.org>
6 years agotcg: Avoid setting tcg_initialize if !CONFIG_TCG
Richard Henderson [Thu, 26 Oct 2017 13:58:14 +0000 (15:58 +0200)]
tcg: Avoid setting tcg_initialize if !CONFIG_TCG

Fix the build for --disable-tcg.

Fixes: 55c3ceef61fcf06fc98ddc752b7cce788ce7680b
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Message-id: 20171026135814.20773-1-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'mreitz/tags/pull-block-2017-10-26' into queue-block
Kevin Wolf [Thu, 26 Oct 2017 13:02:40 +0000 (15:02 +0200)]
Merge remote-tracking branch 'mreitz/tags/pull-block-2017-10-26' into queue-block

Block patches

# gpg: Signature made Thu Oct 26 15:01:20 2017 CEST
# gpg:                using RSA key F407DB0061D5CF40
# gpg: Good signature from "Max Reitz <mreitz@redhat.com>"
# Primary key fingerprint: 91BE B60A 30DB 3E88 57D1  1829 F407 DB00 61D5 CF40

* mreitz/tags/pull-block-2017-10-26:
  iotests: Add cluster_size=64k to 125
  qcow2: Always execute preallocate() in a coroutine
  qcow2: Fix unaligned preallocated truncation
  qcow2: Emit errp when truncating the image tail
  iotests: Filter actual image size in 184 and 191
  iotests: Pull _filter_actual_image_size from 67/87
  iotests: Add test for dataplane mirroring
  qcow2: Use BDRV_SECTOR_BITS instead of its literal value

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoiotests: Add cluster_size=64k to 125
Max Reitz [Mon, 9 Oct 2017 21:55:33 +0000 (23:55 +0200)]
iotests: Add cluster_size=64k to 125

Apparently it would be a good idea to test that, too.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20171009215533.12530-4-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
6 years agoqcow2: Always execute preallocate() in a coroutine
Max Reitz [Mon, 9 Oct 2017 21:55:32 +0000 (23:55 +0200)]
qcow2: Always execute preallocate() in a coroutine

Some qcow2 functions (at least perform_cow()) expect s->lock to be
taken.  Therefore, if we want to make use of them, we should execute
preallocate() (as "preallocate_co") in a coroutine so that we can use
the qemu_co_mutex_* functions.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20171009215533.12530-3-mreitz@redhat.com
Cc: qemu-stable@nongnu.org
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
6 years agoqcow2: Fix unaligned preallocated truncation
Max Reitz [Mon, 9 Oct 2017 21:55:31 +0000 (23:55 +0200)]
qcow2: Fix unaligned preallocated truncation

A qcow2 image file's length is not required to have a length that is a
multiple of the cluster size.  However, qcow2_refcount_area() expects an
aligned value for its @start_offset parameter, so we need to round
@old_file_size up to the next cluster boundary.

Reported-by: Ping Li <pingl@redhat.com>
Bug: https://bugzilla.redhat.com/show_bug.cgi?id=1414049
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20171009215533.12530-2-mreitz@redhat.com
Cc: qemu-stable@nongnu.org
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
6 years agoqcow2: Emit errp when truncating the image tail
Max Reitz [Mon, 9 Oct 2017 15:54:31 +0000 (17:54 +0200)]
qcow2: Emit errp when truncating the image tail

bdrv_truncate() has an errp parameter which is always set when an error
occurs.  Let's use that instead of a plain strerror().

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20171009155431.14093-1-mreitz@redhat.com
Reviewed-by: Pavel Butsykin <pbutsykin@virtuozzo.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
6 years agoiotests: Filter actual image size in 184 and 191
Max Reitz [Mon, 9 Oct 2017 16:34:56 +0000 (18:34 +0200)]
iotests: Filter actual image size in 184 and 191

Whenever the actual image size is not part of the test, it should be
filtered as it depends on the host filesystem.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20171009163456.485-3-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
6 years agoiotests: Pull _filter_actual_image_size from 67/87
Max Reitz [Mon, 9 Oct 2017 16:34:55 +0000 (18:34 +0200)]
iotests: Pull _filter_actual_image_size from 67/87

Tests 067 and 087 filter the actual image size because it depends on the
host filesystem (and is not part of the respective test).  Since this is
generally true, we should have a common filter function for this, so
let's pull out the sed line from both tests into such a function.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20171009163456.485-2-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
6 years agoiotests: Add test for dataplane mirroring
Max Reitz [Fri, 29 Sep 2017 17:08:43 +0000 (19:08 +0200)]
iotests: Add test for dataplane mirroring

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20170929170843.3711-1-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
6 years agoqcow2: Use BDRV_SECTOR_BITS instead of its literal value
Alberto Garcia [Mon, 9 Oct 2017 15:38:56 +0000 (17:38 +0200)]
qcow2: Use BDRV_SECTOR_BITS instead of its literal value

BDRV_SECTOR_BITS is defined to be 9 in block.h (and BDRV_SECTOR_SIZE
is calculated from that), but there are still a couple of places where
we are using the literal value instead of the macro.

Signed-off-by: Alberto Garcia <berto@igalia.com>
Message-id: 20171009153856.20387-1-berto@igalia.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
6 years agoqemu-img.1: Image invalidation on qemu-img commit
Max Reitz [Thu, 26 Oct 2017 07:59:47 +0000 (09:59 +0200)]
qemu-img.1: Image invalidation on qemu-img commit

qemu-img commit invalidates all images between base and top.  This
should be mentioned in the man page.

Suggested-by: Ping Li <pingl@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoqemu-io: Relax 'alloc' now that block-status doesn't assert
Eric Blake [Thu, 12 Oct 2017 03:47:20 +0000 (22:47 -0500)]
qemu-io: Relax 'alloc' now that block-status doesn't assert

Previously, the alloc command required that input parameters be
sector-aligned and clamped to 32 bits, because the underlying
bdrv_is_allocated used a 32-bit parameter and asserted aligned
inputs.  But now that we have fixed block status to report a
64-bit bytes value, and to properly round requests on behalf of
guests, we can pass any values, and can use qemu-io to add
coverage that our rounding is correct regardless of the guest
alignment constraints.

Update iotest 177 to intentionally probe block status at
unaligned boundaries as well as with a bytes value that does not
map to 32-bit sectors, which also required tweaking the image
prep to leave an unallocated portion to the image under test.

Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoqcow2: Reduce is_zero() rounding
Eric Blake [Thu, 12 Oct 2017 03:47:19 +0000 (22:47 -0500)]
qcow2: Reduce is_zero() rounding

Now that bdrv_is_allocated accepts non-aligned inputs, we can
remove the TODO added in earlier refactoring.

Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoblock: Reduce bdrv_aligned_preadv() rounding
Eric Blake [Thu, 12 Oct 2017 03:47:18 +0000 (22:47 -0500)]
block: Reduce bdrv_aligned_preadv() rounding

Now that bdrv_is_allocated accepts non-aligned inputs, we can
remove the TODO added in commit d6a644bb.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoblock: Align block status requests
Eric Blake [Thu, 12 Oct 2017 03:47:17 +0000 (22:47 -0500)]
block: Align block status requests

Any device that has request_alignment greater than 512 should be
unable to report status at a finer granularity; it may also be
simpler for such devices to be guaranteed that the block layer
has rounded things out to the granularity boundary (the way the
block layer already rounds all other I/O out).  Besides, getting
the code correct for super-sector alignment also benefits us
for the fact that our public interface now has byte granularity,
even though none of our drivers have byte-level callbacks.

Add an assertion in blkdebug that proves that the block layer
never requests status of unaligned sections, similar to what it
does on other requests (while still keeping the generic helper
in place for when future patches add a throttle driver).  Note
that iotest 177 already covers this (it would fail if you use
just the blkdebug.c hunk without the io.c changes).  Meanwhile,
we can drop assertions in callers that no longer have to pass
in sector-aligned addresses.

There is a mid-function scope added for 'count' and 'longret',
for a couple of reasons: first, an upcoming patch will add an
'if' statement that checks whether a driver has an old- or
new-style callback, and can conveniently use the same scope for
less indentation churn at that time.  Second, since we are
trying to get rid of sector-based computations, wrapping things
in a scope makes it easier to group and see what will be
deleted in a final cleanup patch once all drivers have been
converted to the new-style callback.

Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoqemu-img: Change img_compare() to be byte-based
Eric Blake [Thu, 12 Oct 2017 03:47:16 +0000 (22:47 -0500)]
qemu-img: Change img_compare() to be byte-based

In the continuing quest to make more things byte-based, change
the internal iteration of img_compare().  We can finally drop the
TODO assertions added earlier, now that the entire algorithm is
byte-based and no longer has to shift from bytes to sectors.

Most of the change is mechanical ('total_sectors' becomes
'total_size', 'sector_num' becomes 'offset', 'nb_sectors' becomes
'chunk', 'progress_base' goes from sectors to bytes); some of it
is also a cleanup (sectors_to_bytes() is now unused, loss of
variable 'count' added earlier in commit 51b0a488).

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoqemu-img: Change img_rebase() to be byte-based
Eric Blake [Thu, 12 Oct 2017 03:47:15 +0000 (22:47 -0500)]
qemu-img: Change img_rebase() to be byte-based

In the continuing quest to make more things byte-based, change
the internal iteration of img_rebase().  We can finally drop the
TODO assertion added earlier, now that the entire algorithm is
byte-based and no longer has to shift from bytes to sectors.

Most of the change is mechanical ('num_sectors' becomes 'size',
'sector' becomes 'offset', 'n' goes from sectors to bytes); some
of it is also a cleanup (use of MIN() instead of open-coding,
loss of variable 'count' added earlier in commit d6a644bb).

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoqemu-img: Change compare_sectors() to be byte-based
Eric Blake [Thu, 12 Oct 2017 03:47:14 +0000 (22:47 -0500)]
qemu-img: Change compare_sectors() to be byte-based

In the continuing quest to make more things byte-based, change
compare_sectors(), renaming it to compare_buffers() in the
process.  Note that one caller (qemu-img compare) only cares
about the first difference, while the other (qemu-img rebase)
cares about how many consecutive sectors have the same
equal/different status; however, this patch does not bother to
micro-optimize the compare case to avoid the comparisons of
sectors beyond the first mismatch.  Both callers are always
passing valid buffers in, so the initial check for buffer size
can be turned into an assertion.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoqemu-img: Change check_empty_sectors() to byte-based
Eric Blake [Thu, 12 Oct 2017 03:47:13 +0000 (22:47 -0500)]
qemu-img: Change check_empty_sectors() to byte-based

Continue on the quest to make more things byte-based instead of
sector-based.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoqemu-img: Drop redundant error message in compare
Eric Blake [Thu, 12 Oct 2017 03:47:12 +0000 (22:47 -0500)]
qemu-img: Drop redundant error message in compare

If a read error is encountered during 'qemu-img compare', we
were printing the "Error while reading offset ..." message twice;
this was because our helper function was awkward, printing output
on some but not all paths.  Fix it to consistently report errors
on all paths, so that the callers do not risk a redundant message,
and update the testsuite for the improved output.

Further simplify the code by hoisting the conversion from an error
message to an exit code into the helper function, rather than
repeating that logic at all callers (yes, the helper function is
now less generic, but it's a net win in lines of code).

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoqemu-img: Add find_nonzero()
Eric Blake [Thu, 12 Oct 2017 03:47:11 +0000 (22:47 -0500)]
qemu-img: Add find_nonzero()

During 'qemu-img compare', when we are checking that an allocated
portion of one file is all zeros, we don't need to waste time
computing how many additional sectors after the first non-zero
byte are also non-zero.  Create a new helper find_nonzero() to do
the check for a first non-zero sector, and rebase
check_empty_sectors() to use it.

The new interface intentionally uses bytes in its interface, even
though it still crawls the buffer a sector at a time; it is robust
to a partial sector at the end of the buffer.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoqemu-img: Speed up compare on pre-allocated larger file
Eric Blake [Thu, 12 Oct 2017 03:47:10 +0000 (22:47 -0500)]
qemu-img: Speed up compare on pre-allocated larger file

Compare the following images with all-zero contents:
$ truncate --size 1M A
$ qemu-img create -f qcow2 -o preallocation=off B 1G
$ qemu-img create -f qcow2 -o preallocation=metadata C 1G

On my machine, the difference is noticeable for pre-patch speeds,
with more than an order of magnitude in difference caused by the
choice of preallocation in the qcow2 file:

$ time ./qemu-img compare -f raw -F qcow2 A B
Warning: Image size mismatch!
Images are identical.

real 0m0.014s
user 0m0.007s
sys 0m0.007s

$ time ./qemu-img compare -f raw -F qcow2 A C
Warning: Image size mismatch!
Images are identical.

real 0m0.341s
user 0m0.144s
sys 0m0.188s

Why? Because bdrv_is_allocated() returns false for image B but
true for image C, throwing away the fact that both images know
via lseek(SEEK_HOLE) that the entire image still reads as zero.
From there, qemu-img ends up calling bdrv_pread() for every byte
of the tail, instead of quickly looking for the next allocation.
The solution: use block_status instead of is_allocated, giving:

$ time ./qemu-img compare -f raw -F qcow2 A C
Warning: Image size mismatch!
Images are identical.

real 0m0.014s
user 0m0.011s
sys 0m0.003s

which is on par with the speeds for no pre-allocation.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoqemu-img: Simplify logic in img_compare()
Eric Blake [Thu, 12 Oct 2017 03:47:09 +0000 (22:47 -0500)]
qemu-img: Simplify logic in img_compare()

As long as we are querying the status for a chunk smaller than
the known image size, we are guaranteed that a successful return
will have set pnum to a non-zero size (pnum is zero only for
queries beyond the end of the file).  Use that to slightly
simplify the calculation of the current chunk size being compared.
Likewise, we don't have to shrink the amount of data operated on
until we know we have to read the file, and therefore have to fit
in the bounds of our buffer.  Also, note that 'total_sectors_over'
is equivalent to 'progress_base'.

With these changes in place, sectors_to_process() is now dead code,
and can be removed.

Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoblock: Convert bdrv_get_block_status_above() to bytes
Eric Blake [Thu, 12 Oct 2017 03:47:08 +0000 (22:47 -0500)]
block: Convert bdrv_get_block_status_above() to bytes

We are gradually moving away from sector-based interfaces, towards
byte-based.  In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.

Changing the name of the function from bdrv_get_block_status_above()
to bdrv_block_status_above() ensures that the compiler enforces that
all callers are updated.  Likewise, since it a byte interface allows
an offset mapping that might not be sector aligned, split the mapping
out of the return value and into a pass-by-reference parameter.  For
now, the io.c layer still assert()s that all uses are sector-aligned,
but that can be relaxed when a later patch implements byte-based
block status in the drivers.

For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_block_status(), plus
updates for the new split return interface.  But some code,
particularly bdrv_block_status(), gets a lot simpler because it no
longer has to mess with sectors.  Likewise, mirror code no longer
computes s->granularity >> BDRV_SECTOR_BITS, and can therefore drop
an assertion about alignment because the loop no longer depends on
alignment (never mind that we don't really have a driver that
reports sub-sector alignments, so it's not really possible to test
the effect of sub-sector mirroring).  Fix a neighboring assertion to
use is_power_of_2 while there.

For ease of review, bdrv_get_block_status() was tackled separately.

Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoblock: Switch bdrv_co_get_block_status_above() to byte-based
Eric Blake [Thu, 12 Oct 2017 03:47:07 +0000 (22:47 -0500)]
block: Switch bdrv_co_get_block_status_above() to byte-based

We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based.  Convert another internal
type (no semantic change), and rename it to match the corresponding
public function rename.

Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoblock: Switch bdrv_common_block_status_above() to byte-based
Eric Blake [Thu, 12 Oct 2017 03:47:06 +0000 (22:47 -0500)]
block: Switch bdrv_common_block_status_above() to byte-based

We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based.  Convert another internal
function (no semantic change).

Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoblock: Switch BdrvCoGetBlockStatusData to byte-based
Eric Blake [Thu, 12 Oct 2017 03:47:05 +0000 (22:47 -0500)]
block: Switch BdrvCoGetBlockStatusData to byte-based

We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based.  Convert another internal
type (no semantic change), and rename it to match the corresponding
public function rename.

Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoblock: Switch bdrv_co_get_block_status() to byte-based
Eric Blake [Thu, 12 Oct 2017 03:47:04 +0000 (22:47 -0500)]
block: Switch bdrv_co_get_block_status() to byte-based

We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based.  Convert another internal
function (no semantic change); and as with its public counterpart,
rename to bdrv_co_block_status() and split the offset return, to
make the compiler enforce that we catch all uses.  For now, we
assert that callers and the return value still use aligned data,
but ultimately, this will be the function where we hand off to a
byte-based driver callback, and will eventually need to add logic
to ensure we round calls according to the driver's
request_alignment then touch up the result handed back to the
caller, to start permitting a caller to pass unaligned offsets.

Note that we are now prepared to accepts 'bytes' larger than INT_MAX;
this is okay as long as we clamp things internally before violating
any 32-bit limits, and makes no difference to how a client will
use the information (clients looping over the entire file must
already be prepared for consecutive calls to return the same status,
as drivers are already free to return shorter-than-maximal status
due to any other convenient split points, such as when the L2 table
crosses cluster boundaries in qcow2).

Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoblock: Convert bdrv_get_block_status() to bytes
Eric Blake [Thu, 12 Oct 2017 03:47:03 +0000 (22:47 -0500)]
block: Convert bdrv_get_block_status() to bytes

We are gradually moving away from sector-based interfaces, towards
byte-based.  In the common case, allocation is unlikely to ever use
values that are not naturally sector-aligned, but it is possible
that byte-based values will let us be more precise about allocation
at the end of an unaligned file that can do byte-based access.

Changing the name of the function from bdrv_get_block_status() to
bdrv_block_status() ensures that the compiler enforces that all
callers are updated.  For now, the io.c layer still assert()s that
all callers are sector-aligned, but that can be relaxed when a later
patch implements byte-based block status in the drivers.

There was an inherent limitation in returning the offset via the
return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which
means an offset can only be mapped for sector-aligned queries (or,
if we declare that non-aligned input is at the same relative position
modulo 512 of the answer), so the new interface also changes things to
return the offset via output through a parameter by reference rather
than mashed into the return value.  We'll have some glue code that
munges between the two styles until we finish converting all uses.

For the most part this patch is just the addition of scaling at the
callers followed by inverse scaling at bdrv_block_status(), coupled
with the tweak in calling convention.  But some code, particularly
bdrv_is_allocated(), gets a lot simpler because it no longer has to
mess with sectors.

For ease of review, bdrv_get_block_status_above() will be tackled
separately.

Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoqemu-img: Switch get_block_status() to byte-based
Eric Blake [Thu, 12 Oct 2017 03:47:02 +0000 (22:47 -0500)]
qemu-img: Switch get_block_status() to byte-based

We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based.  Continue by converting
an internal function (no semantic change), and simplifying its
caller accordingly.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoblock: Switch bdrv_make_zero() to byte-based
Eric Blake [Thu, 12 Oct 2017 03:47:01 +0000 (22:47 -0500)]
block: Switch bdrv_make_zero() to byte-based

We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based.  Change the internal
loop iteration of zeroing a device to track by bytes instead of
sectors (although we are still guaranteed that we iterate by steps
that are sector-aligned).

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoqcow2: Switch is_zero_sectors() to byte-based
Eric Blake [Thu, 12 Oct 2017 03:47:00 +0000 (22:47 -0500)]
qcow2: Switch is_zero_sectors() to byte-based

We are gradually converting to byte-based interfaces, as they are
easier to reason about than sector-based.  Convert another internal
function (no semantic change), and rename it to is_zero() in the
process.

Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoblock: Make bdrv_round_to_clusters() signature more useful
Eric Blake [Thu, 12 Oct 2017 03:46:59 +0000 (22:46 -0500)]
block: Make bdrv_round_to_clusters() signature more useful

In the process of converting sector-based interfaces to bytes,
I'm finding it easier to represent a byte count as a 64-bit
integer at the block layer (even if we are internally capped
by SIZE_MAX or even INT_MAX for individual transactions, it's
still nicer to not have to worry about truncation/overflow
issues on as many variables).  Update the signature of
bdrv_round_to_clusters() to uniformly use int64_t, matching
the signature already chosen for bdrv_is_allocated and the
fact that off_t is also a signed type, then adjust clients
according to the required fallout (even where the result could
now exceed 32 bits, no client is directly assigning the result
into a 32-bit value without breaking things into a loop first).

Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoblock: Add flag to avoid wasted work in bdrv_is_allocated()
Eric Blake [Thu, 12 Oct 2017 03:46:58 +0000 (22:46 -0500)]
block: Add flag to avoid wasted work in bdrv_is_allocated()

Not all callers care about which BDS owns the mapping for a given
range of the file, or where the zeroes lie within that mapping.  In
particular, bdrv_is_allocated() cares more about finding the
largest run of allocated data from the guest perspective, whether
or not that data is consecutive from the host perspective, and
whether or not the data reads as zero.  Therefore, doing subsequent
refinements such as checking how much of the format-layer
allocation also satisfies BDRV_BLOCK_ZERO at the protocol layer is
wasted work - in the best case, it just costs extra CPU cycles
during a single bdrv_is_allocated(), but in the worst case, it
results in a smaller *pnum, and forces callers to iterate through
more status probes when visiting the entire file for even more
extra CPU cycles.

This patch only optimizes the block layer (no behavior change when
want_zero is true, but skip unnecessary effort when it is false).
Then when subsequent patches tweak the driver callback to be
byte-based, we can also pass this hint through to the driver.

Tweak BdrvCoGetBlockStatusData to declare arguments in parameter
order, rather than mixing things up (minimizing padding is not
necessary here).

Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoblock: Allow NULL file for bdrv_get_block_status()
Eric Blake [Thu, 12 Oct 2017 03:46:57 +0000 (22:46 -0500)]
block: Allow NULL file for bdrv_get_block_status()

Not all callers care about which BDS owns the mapping for a given
range of the file.  This patch merely simplifies the callers by
consolidating the logic in the common call point, while guaranteeing
a non-NULL file to all the driver callbacks, for no semantic change.
The only caller that does not care about pnum is bdrv_is_allocated,
as invoked by vvfat; we can likewise add assertions that the rest
of the stack does not have to worry about a NULL pnum.

Furthermore, this will also set the stage for a future cleanup: when
a caller does not care about which BDS owns an offset, it would be
nice to allow the driver to optimize things to not have to return
BDRV_BLOCK_OFFSET_VALID in the first place.  In the case of fragmented
allocation (for example, it's fairly easy to create a qcow2 image
where consecutive guest addresses are not at consecutive host
addresses), the current contract requires bdrv_get_block_status()
to clamp *pnum to the limit where host addresses are no longer
consecutive, but allowing a NULL file means that *pnum could be
set to the full length of known-allocated data.

Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agoqemu-iotests: Test backing_fmt with backing node reference
Kevin Wolf [Tue, 17 Oct 2017 15:08:51 +0000 (17:08 +0200)]
qemu-iotests: Test backing_fmt with backing node reference

This changes test case 191 to include a backing image that has
backing_fmt set in the image file, but is referenced by node name in the
qemu command line.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
6 years agoblock: don't add 'driver' to options when referring to backing via node name
Peter Krempa [Thu, 12 Oct 2017 14:14:10 +0000 (16:14 +0200)]
block: don't add 'driver' to options when referring to backing via node name

When referring to a backing file of an image via node name
bdrv_open_backing_file would add the 'driver' option to the option list
filling it with the backing format driver. This breaks construction of
the backing chain via -blockdev, as bdrv_open_inherit reports an error
if both 'reference' and 'options' are provided.

$ qemu-img create -f raw /tmp/backing.raw 64M
$ qemu-img create -f qcow2 -F raw -b /tmp/backing.raw /tmp/test.qcow2
$ qemu-system-x86_64 \
  -blockdev driver=file,filename=/tmp/backing.raw,node-name=backing \
  -blockdev driver=qcow2,file.driver=file,file.filename=/tmp/test.qcow2,node-name=root,backing=backing
qemu-system-x86_64: -blockdev driver=qcow2,file.driver=file,file.filename=/tmp/test.qcow2,node-name=root,backing=backing: Could not open backing file: Cannot reference an existing block device with additional options or a new filename

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
6 years agodisas: Add capstone as submodule
Richard Henderson [Thu, 28 Sep 2017 16:01:23 +0000 (09:01 -0700)]
disas: Add capstone as submodule

Do not require the submodule, but use it if present.  Allow the
command-line to override system or git submodule either way.

Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/stefanberger/tags/pull-tpm-2017-10-24-1' into...
Peter Maydell [Thu, 26 Oct 2017 08:20:11 +0000 (09:20 +0100)]
Merge remote-tracking branch 'remotes/stefanberger/tags/pull-tpm-2017-10-24-1' into staging

Merge tpm 2017/10/24 v1

# gpg: Signature made Wed 25 Oct 2017 06:06:55 BST
# gpg:                using RSA key 0x75AD65802A0B4211
# gpg: Good signature from "Stefan Berger <stefanb@linux.vnet.ibm.com>"
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg:          There is no indication that the signature belongs to the owner.
# Primary key fingerprint: B818 B9CA DF90 89C2 D5CE  C66B 75AD 6580 2A0B 4211

* remotes/stefanberger/tags/pull-tpm-2017-10-24-1:
  tpm: print buffers received from TPM when debugging
  vl: remove unnecessary #ifdef CONFIG_TPM
  tpm: remove unnecessary #ifdef CONFIG_TPM
  tpm: add stubs
  tpm: add missing include

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoqga: add network stats to guest-network-get-interfaces
ZhiPeng Lu [Tue, 12 Sep 2017 08:54:26 +0000 (16:54 +0800)]
qga: add network stats to guest-network-get-interfaces

we can get the network interface statistics inside a virtual machine by
guest-network-get-interfaces command. it is very useful for us tomonitor
and analyze network traffic.

Signed-off-by: ZhiPeng Lu <lu.zhipeng@zte.com.cn>
* don't rely on sizeof(wchar[]) for wchar[] indexing
* avoid camelCase variable names
* fix up getline() usage
* condensed commit subject line
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
6 years agoqga-win: Updating guest_set_time action
Bishara AbuHattoum [Tue, 22 Aug 2017 13:55:04 +0000 (16:55 +0300)]
qga-win: Updating guest_set_time action

  At the moment, Windows libraries don't provide a way to access
  RTC, so, a workaround is to use the Windows w32tm command to
  resync the time.
  Related bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1183874

Signed-off-by: Bishara AbuHattoum <bishara@daynix.com>
Reviewed-by: Sameeh Jubran <sameeh@daynix.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
6 years agoqga-win: don't hang if vss hold writes timeout
Chen Hanxiao [Tue, 17 Oct 2017 06:41:42 +0000 (14:41 +0800)]
qga-win: don't hang if vss hold writes timeout

When VM is in a heavy IO, if the command "guest-fsfreeze-freeze"
is executed, VSS may timeout when trying to hold writes.

Inside guest, Event ID 12298(VSS_ERROR_HOLD_WRITES_TIMEOUT)
is logged in the Event Viewer.

At that time, if we call AbortBackup, qga may hang forever.

This patch will solve this issue.

Cc: Michael Roth <mdroth@linux.vnet.ibm.com>
Cc: Tomoki Sekiyama <tomoki.sekiyama@gmail.com>
Signed-off-by: Chen Hanxiao <chenhanxiao@gmail.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
6 years agoMerge remote-tracking branch 'remotes/rth/tags/pull-tcg-20171025' into staging
Peter Maydell [Wed, 25 Oct 2017 15:38:57 +0000 (16:38 +0100)]
Merge remote-tracking branch 'remotes/rth/tags/pull-tcg-20171025' into staging

TCG patch queue

# gpg: Signature made Wed 25 Oct 2017 10:30:18 BST
# gpg:                using RSA key 0x64DF38E8AF7E215F
# gpg: Good signature from "Richard Henderson <richard.henderson@linaro.org>"
# Primary key fingerprint: 7A48 1E78 868B 4DB6 A85A  05C0 64DF 38E8 AF7E 215F

* remotes/rth/tags/pull-tcg-20171025: (51 commits)
  translate-all: exit from tb_phys_invalidate if qht_remove fails
  tcg: Initialize cpu_env generically
  tcg: enable multiple TCG contexts in softmmu
  tcg: introduce regions to split code_gen_buffer
  translate-all: use qemu_protect_rwx/none helpers
  osdep: introduce qemu_mprotect_rwx/none
  tcg: allocate optimizer temps with tcg_malloc
  tcg: distribute profiling counters across TCGContext's
  tcg: introduce **tcg_ctxs to keep track of all TCGContext's
  gen-icount: fold exitreq_label into TCGContext
  tcg: define tcg_init_ctx and make tcg_ctx a pointer
  tcg: take tb_ctx out of TCGContext
  translate-all: report correct avg host TB size
  exec-all: rename tb_free to tb_remove
  translate-all: use a binary search tree to track TBs in TBContext
  tcg: Remove CF_IGNORE_ICOUNT
  tcg: Add CF_LAST_IO + CF_USE_ICOUNT to CF_HASH_MASK
  cpu-exec: lookup/generate TB outside exclusive region during step_atomic
  tcg: check CF_PARALLEL instead of parallel_cpus
  target/sparc: check CF_PARALLEL instead of parallel_cpus
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agoMerge remote-tracking branch 'remotes/juanquintela/tags/migration/20171023' into...
Peter Maydell [Wed, 25 Oct 2017 14:24:08 +0000 (15:24 +0100)]
Merge remote-tracking branch 'remotes/juanquintela/tags/migration/20171023' into staging

migration/next for 20171023

# gpg: Signature made Mon 23 Oct 2017 17:05:14 BST
# gpg:                using RSA key 0xF487EF185872D723
# gpg: Good signature from "Juan Quintela <quintela@redhat.com>"
# gpg:                 aka "Juan Quintela <quintela@trasno.org>"
# Primary key fingerprint: 1899 FF8E DEBF 58CC EE03  4B82 F487 EF18 5872 D723

* remotes/juanquintela/tags/migration/20171023: (21 commits)
  migration: Improve migration thread error handling
  qapi: Fix grammar in x-multifd-page-count descriptions
  migration: add bitmap for received page
  migration: introduce qemu_ufd_copy_ioctl helper
  migration: postcopy_place_page factoring out
  migration: new ram_init_bitmaps()
  migration: clean up xbzrle cache init/destroy
  migration: provide ram_state_cleanup
  migration: provide ram_state_init()
  migration: pause-before-switchover for postcopy
  migration: allow cancel to unpause
  migrate: HMP migate_continue
  migration: migrate-continue
  migration: Wait for semaphore before completing migration
  migration: Add 'pre-switchover' and 'device' statuses
  migration: Add 'pause-before-switchover' capability
  migration: Make cache_init() take an error parameter
  migration: Move xbzrle cache resize error handling to xbzrle_cache_resize
  migration: Make cache size elements use the right types
  migratiom: Remove max_item_age parameter
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
6 years agodisas: Remove monitor_disas_is_physical
Richard Henderson [Tue, 19 Sep 2017 14:40:40 +0000 (09:40 -0500)]
disas: Remove monitor_disas_is_physical

Even though there is only one monitor, and thus no race on this
global data object, there is also no point in having it.  We can
just as well record the decision in the read_memory_function that
we select.

Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agoppc: Support Capstone in disas_set_info
Richard Henderson [Thu, 14 Sep 2017 17:38:40 +0000 (10:38 -0700)]
ppc: Support Capstone in disas_set_info

Cc: qemu-ppc@nongnu.org
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agoarm: Support Capstone in disas_set_info
Richard Henderson [Thu, 14 Sep 2017 16:51:06 +0000 (09:51 -0700)]
arm: Support Capstone in disas_set_info

Tested-by: Alex Bennée <alex.bennee@linaro.org>
Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agoi386: Support Capstone in disas_set_info
Richard Henderson [Thu, 14 Sep 2017 16:50:05 +0000 (09:50 -0700)]
i386: Support Capstone in disas_set_info

Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agodisas: Support the Capstone disassembler library
Richard Henderson [Thu, 14 Sep 2017 16:41:12 +0000 (09:41 -0700)]
disas: Support the Capstone disassembler library

If configured, prefer this over our rather dated copy of the
GPLv2-only binutils.  This will be especially apparent with
the proposed vector extensions to TCG, as disas/i386.c does
not handle AVX.

Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agodisas: Remove unused flags arguments
Richard Henderson [Thu, 14 Sep 2017 15:38:35 +0000 (08:38 -0700)]
disas: Remove unused flags arguments

Now that every target is using the disas_set_info hook,
the flags argument is unused.  Remove it.

Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotarget/arm: Don't set INSN_ARM_BE32 for CONFIG_USER_ONLY
Richard Henderson [Thu, 19 Oct 2017 21:13:02 +0000 (14:13 -0700)]
target/arm: Don't set INSN_ARM_BE32 for CONFIG_USER_ONLY

This matches translator behaviour in arm_lduw_code.

Fixes: https://bugs.launchpad.net/qemu/+bug/1724485
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotarget/arm: Move BE32 disassembler fixup
Richard Henderson [Thu, 19 Oct 2017 21:10:45 +0000 (14:10 -0700)]
target/arm: Move BE32 disassembler fixup

The Capstone disassembler has its own big-endian fixup.
Doing this twice does not work, of course.  Move our current
fixup from target/arm/cpu.c to disas/arm.c.

This makes read_memory_inner_func unused and can be removed.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotarget/ppc: Convert to disas_set_info hook
Richard Henderson [Thu, 14 Sep 2017 15:21:40 +0000 (08:21 -0700)]
target/ppc: Convert to disas_set_info hook

Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotarget/i386: Convert to disas_set_info hook
Richard Henderson [Thu, 14 Sep 2017 15:10:53 +0000 (08:10 -0700)]
target/i386: Convert to disas_set_info hook

Tested-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotpm: print buffers received from TPM when debugging
Stefan Berger [Wed, 11 Oct 2017 19:46:14 +0000 (15:46 -0400)]
tpm: print buffers received from TPM when debugging

Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
6 years agovl: remove unnecessary #ifdef CONFIG_TPM
Philippe Mathieu-Daudé [Tue, 24 Oct 2017 12:20:45 +0000 (09:20 -0300)]
vl: remove unnecessary #ifdef CONFIG_TPM

a stub is now provided.

Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Tested-by: Richard W.M. Jones <rjones@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
6 years agotpm: remove unnecessary #ifdef CONFIG_TPM
Philippe Mathieu-Daudé [Tue, 24 Oct 2017 12:20:44 +0000 (09:20 -0300)]
tpm: remove unnecessary #ifdef CONFIG_TPM

Makefile.objs now checks for $(CONFIG_TPM).

Suggested-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
6 years agotpm: add stubs
Philippe Mathieu-Daudé [Tue, 24 Oct 2017 12:20:43 +0000 (09:20 -0300)]
tpm: add stubs

Commit c37cacabf22 moved tpm_cleanup() in the main loop exit, however this
function is not available when compiling with --disable-tpm.

Provides necessary stubs to keep code clean of #ifdef'fery.

Reported-by: BALATON Zoltan <balaton@eik.bme.hu>
Message-Id: <20171023102903.256AF7456A0@zero.eik.bme.hu>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Tested-by: Richard W.M. Jones <rjones@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
6 years agotranslate-all: exit from tb_phys_invalidate if qht_remove fails
Emilio G. Cota [Thu, 19 Oct 2017 20:31:54 +0000 (16:31 -0400)]
translate-all: exit from tb_phys_invalidate if qht_remove fails

Two or more threads might race while invalidating the same TB. We currently
do not check for this at all despite taking tb_lock, which means we would
wrongly invalidate the same TB more than once. This bug has actually been
hit by users: I recently saw a report on IRC, although I have yet to see
the corresponding test case.

Fix this by using qht_remove as the synchronization point; if it fails,
that means the TB has already been invalidated, and therefore there
is nothing left to do in tb_phys_invalidate.

Note that this solution works now that we still have tb_lock, and will
continue working once we remove tb_lock.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Message-Id: <1508445114-4717-1-git-send-email-cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: Initialize cpu_env generically
Richard Henderson [Tue, 10 Oct 2017 21:34:37 +0000 (14:34 -0700)]
tcg: Initialize cpu_env generically

This is identical for each target.  So, move the initialization to
common code.  Move the variable itself out of tcg_ctx and name it
cpu_env to minimize changes within targets.

This also means we can remove tcg_global_reg_new_{ptr,i32,i64},
since there are no longer global-register temps created by targets.

Reviewed-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: enable multiple TCG contexts in softmmu
Emilio G. Cota [Wed, 19 Jul 2017 22:57:58 +0000 (18:57 -0400)]
tcg: enable multiple TCG contexts in softmmu

This enables parallel TCG code generation. However, we do not take
advantage of it yet since tb_lock is still held during tb_gen_code.

In user-mode we use a single TCG context; see the documentation
added to tcg_region_init for the rationale.

Note that targets do not need any conversion: targets initialize a
TCGContext (e.g. defining TCG globals), and after this initialization
has finished, the context is cloned by the vCPU threads, each of
them keeping a separate copy.

TCG threads claim one entry in tcg_ctxs[] by atomically increasing
n_tcg_ctxs. Do not be too annoyed by the subsequent atomic_read's
of that variable and tcg_ctxs; they are there just to play nice with
analysis tools such as thread sanitizer.

Note that we do not allocate an array of contexts (we allocate
an array of pointers instead) because when tcg_context_init
is called, we do not know yet how many contexts we'll use since
the bool behind qemu_tcg_mttcg_enabled() isn't set yet.

Previous patches folded some TCG globals into TCGContext. The non-const
globals remaining are only set at init time, i.e. before the TCG
threads are spawned. Here is a list of these set-at-init-time globals
under tcg/:

Only written by tcg_context_init:
- indirect_reg_alloc_order
- tcg_op_defs
Only written by tcg_target_init (called from tcg_context_init):
- tcg_target_available_regs
- tcg_target_call_clobber_regs
- arm: arm_arch, use_idiv_instructions
- i386: have_cmov, have_bmi1, have_bmi2, have_lzcnt,
        have_movbe, have_popcnt
- mips: use_movnz_instructions, use_mips32_instructions,
        use_mips32r2_instructions, got_sigill (tcg_target_detect_isa)
- ppc: have_isa_2_06, have_isa_3_00, tb_ret_addr
- s390: tb_ret_addr, s390_facilities
- sparc: qemu_ld_trampoline, qemu_st_trampoline (build_trampolines),
         use_vis3_instructions

Only written by tcg_prologue_init:
- 'struct jit_code_entry one_entry'
- aarch64: tb_ret_addr
- arm: tb_ret_addr
- i386: tb_ret_addr, guest_base_flags
- ia64: tb_ret_addr
- mips: tb_ret_addr, bswap32_addr, bswap32u_addr, bswap64_addr

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: introduce regions to split code_gen_buffer
Emilio G. Cota [Fri, 7 Jul 2017 23:24:20 +0000 (19:24 -0400)]
tcg: introduce regions to split code_gen_buffer

This is groundwork for supporting multiple TCG contexts.

The naive solution here is to split code_gen_buffer statically
among the TCG threads; this however results in poor utilization
if translation needs are different across TCG threads.

What we do here is to add an extra layer of indirection, assigning
regions that act just like pages do in virtual memory allocation.
(BTW if you are wondering about the chosen naming, I did not want
to use blocks or pages because those are already heavily used in QEMU).

We use a global lock to serialize allocations as well as statistics
reporting (we now export the size of the used code_gen_buffer with
tcg_code_size()). Note that for the allocator we could just use
a counter and atomic_inc; however, that would complicate the gathering
of tcg_code_size()-like stats. So given that the region operations are
not a fast path, a lock seems the most reasonable choice.

The effectiveness of this approach is clear after seeing some numbers.
I used the bootup+shutdown of debian-arm with '-tb-size 80' as a benchmark.
Note that I'm evaluating this after enabling per-thread TCG (which
is done by a subsequent commit).

* -smp 1, 1 region (entire buffer):
    qemu: flush code_size=83885014 nb_tbs=154739 avg_tb_size=357
    qemu: flush code_size=83884902 nb_tbs=153136 avg_tb_size=363
    qemu: flush code_size=83885014 nb_tbs=152777 avg_tb_size=364
    qemu: flush code_size=83884950 nb_tbs=150057 avg_tb_size=373
    qemu: flush code_size=83884998 nb_tbs=150234 avg_tb_size=373
    qemu: flush code_size=83885014 nb_tbs=154009 avg_tb_size=360
    qemu: flush code_size=83885014 nb_tbs=151007 avg_tb_size=370
    qemu: flush code_size=83885014 nb_tbs=151816 avg_tb_size=367

That is, 8 flushes.

* -smp 8, 32 regions (80/32 MB per region) [i.e. this patch]:

    qemu: flush code_size=76328008 nb_tbs=141040 avg_tb_size=356
    qemu: flush code_size=75366534 nb_tbs=138000 avg_tb_size=361
    qemu: flush code_size=76864546 nb_tbs=140653 avg_tb_size=361
    qemu: flush code_size=76309084 nb_tbs=135945 avg_tb_size=375
    qemu: flush code_size=74581856 nb_tbs=132909 avg_tb_size=375
    qemu: flush code_size=73927256 nb_tbs=135616 avg_tb_size=360
    qemu: flush code_size=78629426 nb_tbs=142896 avg_tb_size=365
    qemu: flush code_size=76667052 nb_tbs=138508 avg_tb_size=368

Again, 8 flushes. Note how buffer utilization is not 100%, but it
is close. Smaller region sizes would yield higher utilization,
but we want region allocation to be rare (it acquires a lock), so
we do not want to go too small.

* -smp 8, static partitioning of 8 regions (10 MB per region):
    qemu: flush code_size=21936504 nb_tbs=40570 avg_tb_size=354
    qemu: flush code_size=11472174 nb_tbs=20633 avg_tb_size=370
    qemu: flush code_size=11603976 nb_tbs=21059 avg_tb_size=365
    qemu: flush code_size=23254872 nb_tbs=41243 avg_tb_size=377
    qemu: flush code_size=28289496 nb_tbs=52057 avg_tb_size=358
    qemu: flush code_size=43605160 nb_tbs=78896 avg_tb_size=367
    qemu: flush code_size=45166552 nb_tbs=82158 avg_tb_size=364
    qemu: flush code_size=63289640 nb_tbs=116494 avg_tb_size=358
    qemu: flush code_size=51389960 nb_tbs=93937 avg_tb_size=362
    qemu: flush code_size=59665928 nb_tbs=107063 avg_tb_size=372
    qemu: flush code_size=38380824 nb_tbs=68597 avg_tb_size=374
    qemu: flush code_size=44884568 nb_tbs=79901 avg_tb_size=376
    qemu: flush code_size=50782632 nb_tbs=90681 avg_tb_size=374
    qemu: flush code_size=39848888 nb_tbs=71433 avg_tb_size=372
    qemu: flush code_size=64708840 nb_tbs=119052 avg_tb_size=359
    qemu: flush code_size=49830008 nb_tbs=90992 avg_tb_size=362
    qemu: flush code_size=68372408 nb_tbs=123442 avg_tb_size=368
    qemu: flush code_size=33555560 nb_tbs=59514 avg_tb_size=378
    qemu: flush code_size=44748344 nb_tbs=80974 avg_tb_size=367
    qemu: flush code_size=37104248 nb_tbs=67609 avg_tb_size=364

That is, 20 flushes. Note how a static partitioning approach uses
the code buffer poorly, leading to many unnecessary flushes.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotranslate-all: use qemu_protect_rwx/none helpers
Emilio G. Cota [Sat, 15 Jul 2017 06:38:57 +0000 (02:38 -0400)]
translate-all: use qemu_protect_rwx/none helpers

The helpers require the address and size to be page-aligned, so
do that before calling them.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agoosdep: introduce qemu_mprotect_rwx/none
Emilio G. Cota [Sat, 15 Jul 2017 06:28:47 +0000 (02:28 -0400)]
osdep: introduce qemu_mprotect_rwx/none

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: allocate optimizer temps with tcg_malloc
Emilio G. Cota [Wed, 19 Jul 2017 18:32:24 +0000 (14:32 -0400)]
tcg: allocate optimizer temps with tcg_malloc

Groundwork for supporting multiple TCG contexts.

While at it, also allocate temps_used directly as a bitmap of the
required size, instead of using a bitmap of TCG_MAX_TEMPS via
TCGTempSet.

Performance-wise we lose about 1.12% in a translation-heavy workload
such as booting+shutting down debian-arm:

Performance counter stats for 'taskset -c 0 arm-softmmu/qemu-system-arm \
-machine type=virt -nographic -smp 1 -m 4096 \
-netdev user,id=unet,hostfwd=tcp::2222-:22 \
-device virtio-net-device,netdev=unet \
-drive file=die-on-boot.qcow2,id=myblock,index=0,if=none \
-device virtio-blk-device,drive=myblock \
-kernel kernel.img -append console=ttyAMA0 root=/dev/vda1 \
-name arm,debug-threads=on -smp 1' (10 runs):

             exec time (s)  Relative slowdown wrt original (%)
---------------------------------------------------------------
 original     20.213321616                                  0.
 tcg_malloc   20.441130078                           1.1270214
 TCGContext   20.477846517                           1.3086662
 g_malloc     20.780527895                           2.8061013

The other two alternatives shown in the table are:
- TCGContext: embed temps[TCG_MAX_TEMPS] and TCGTempSet used_temps
  in TCGContext. This is simple enough but it isn't faster than using
  tcg_malloc; moreover, it wastes memory.
- g_malloc: allocate/deallocate both temps and used_temps every time
  tcg_optimize is executed.

Suggested-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: distribute profiling counters across TCGContext's
Emilio G. Cota [Wed, 5 Jul 2017 23:35:06 +0000 (19:35 -0400)]
tcg: distribute profiling counters across TCGContext's

This is groundwork for supporting multiple TCG contexts.

To avoid scalability issues when profiling info is enabled, this patch
makes the profiling info counters distributed via the following changes:

1) Consolidate profile info into its own struct, TCGProfile, which
   TCGContext also includes. Note that tcg_table_op_count is brought
   into TCGProfile after dropping the tcg_ prefix.
2) Iterate over the TCG contexts in the system to obtain the total counts.

This change also requires updating the accessors to TCGProfile fields to
use atomic_read/set whenever there may be conflicting accesses (as defined
in C11) to them.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: introduce **tcg_ctxs to keep track of all TCGContext's
Emilio G. Cota [Wed, 12 Jul 2017 22:26:40 +0000 (18:26 -0400)]
tcg: introduce **tcg_ctxs to keep track of all TCGContext's

Groundwork for supporting multiple TCG contexts.

Note that having n_tcg_ctxs is unnecessary. However, it is
convenient to have it, since it will simplify iterating over the
array: we'll have just a for loop instead of having to iterate
over a NULL-terminated array (which would require n+1 elems)
or having to check with ifdef's for usermode/softmmu.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agogen-icount: fold exitreq_label into TCGContext
Emilio G. Cota [Tue, 4 Jul 2017 17:54:21 +0000 (13:54 -0400)]
gen-icount: fold exitreq_label into TCGContext

Groundwork for supporting multiple TCG contexts.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: define tcg_init_ctx and make tcg_ctx a pointer
Emilio G. Cota [Wed, 12 Jul 2017 21:15:52 +0000 (17:15 -0400)]
tcg: define tcg_init_ctx and make tcg_ctx a pointer

Groundwork for supporting multiple TCG contexts.

The core of this patch is this change to tcg/tcg.h:

> -extern TCGContext tcg_ctx;
> +extern TCGContext tcg_init_ctx;
> +extern TCGContext *tcg_ctx;

Note that for now we set *tcg_ctx to whatever TCGContext is passed
to tcg_context_init -- in this case &tcg_init_ctx.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: take tb_ctx out of TCGContext
Emilio G. Cota [Sat, 24 Jun 2017 00:04:43 +0000 (20:04 -0400)]
tcg: take tb_ctx out of TCGContext

Groundwork for supporting multiple TCG contexts.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotranslate-all: report correct avg host TB size
Emilio G. Cota [Sat, 24 Jun 2017 00:57:44 +0000 (20:57 -0400)]
translate-all: report correct avg host TB size

Since commit 6e3b2bfd6 ("tcg: allocate TB structs before the
corresponding translated code") we are not fully utilizing
code_gen_buffer for translated code, and therefore are
incorrectly reporting the amount of translated code as well as
the average host TB size. Address this by:

- Making the conscious choice of misreporting the total translated code;
  doing otherwise would mislead users into thinking "-tb-size" is not
  honoured.

- Expanding tb_tree_stats to accurately count the bytes of translated code on
  the host, and using this for reporting the average tb host size,
  as well as the expansion ratio.

In the future we might want to consider reporting the accurate numbers for
the total translated code, together with a "bookkeeping/overhead" field to
account for the TB structs.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agoexec-all: rename tb_free to tb_remove
Emilio G. Cota [Wed, 12 Jul 2017 18:40:28 +0000 (14:40 -0400)]
exec-all: rename tb_free to tb_remove

We don't really free anything in this function anymore; we just remove
the TB from the binary search tree.

Suggested-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotranslate-all: use a binary search tree to track TBs in TBContext
Emilio G. Cota [Fri, 23 Jun 2017 23:00:11 +0000 (19:00 -0400)]
translate-all: use a binary search tree to track TBs in TBContext

This is a prerequisite for supporting multiple TCG contexts, since
we will have threads generating code in separate regions of
code_gen_buffer.

For this we need a new field (.size) in struct tb_tc to keep
track of the size of the translated code. This field uses a size_t
to avoid adding a hole to the struct, although really an unsigned
int would have been enough.

The comparison function we use is optimized for the common case:
insertions. Profiling shows that upon booting debian-arm, 98%
of comparisons are between existing tb's (i.e. a->size and b->size
are both !0), which happens during insertions (and removals, but
those are rare). The remaining cases are lookups. From reading the glib
sources we see that the first key is always the lookup key. However,
the code does not assume this to always be the case because this
behaviour is not guaranteed in the glib docs. However, we embed
this knowledge in the code as a branch hint for the compiler.

Note that tb_free does not free space in the code_gen_buffer anymore,
since we cannot easily know whether the tb is the last one inserted
in code_gen_buffer. The next patch in this series renames tb_free
to tb_remove to reflect this.

Performance-wise, lookups in tb_find_pc are the same as before:
O(log n). However, insertions are O(log n) instead of O(1), which
results in a small slowdown when booting debian-arm:

Performance counter stats for 'build/arm-softmmu/qemu-system-arm \
-machine type=virt -nographic -smp 1 -m 4096 \
-netdev user,id=unet,hostfwd=tcp::2222-:22 \
-device virtio-net-device,netdev=unet \
-drive file=img/arm/jessie-arm32.qcow2,id=myblock,index=0,if=none \
-device virtio-blk-device,drive=myblock \
-kernel img/arm/aarch32-current-linux-kernel-only.img \
-append console=ttyAMA0 root=/dev/vda1 \
-name arm,debug-threads=on -smp 1' (10 runs):

- Before:

       8048.598422      task-clock (msec)         #    0.931 CPUs utilized            ( +-  0.28% )
            16,974      context-switches          #    0.002 M/sec                    ( +-  0.12% )
                 0      cpu-migrations            #    0.000 K/sec
            10,125      page-faults               #    0.001 M/sec                    ( +-  1.23% )
    35,144,901,879      cycles                    #    4.367 GHz                      ( +-  0.14% )
   <not supported>      stalled-cycles-frontend
   <not supported>      stalled-cycles-backend
    65,758,252,643      instructions              #    1.87  insns per cycle          ( +-  0.33% )
    10,871,298,668      branches                  # 1350.707 M/sec                    ( +-  0.41% )
       192,322,212      branch-misses             #    1.77% of all branches          ( +-  0.32% )

       8.640869419 seconds time elapsed                                          ( +-  0.57% )

- After:
       8146.242027      task-clock (msec)         #    0.923 CPUs utilized            ( +-  1.23% )
            17,016      context-switches          #    0.002 M/sec                    ( +-  0.40% )
                 0      cpu-migrations            #    0.000 K/sec
            18,769      page-faults               #    0.002 M/sec                    ( +-  0.45% )
    35,660,956,120      cycles                    #    4.378 GHz                      ( +-  1.22% )
   <not supported>      stalled-cycles-frontend
   <not supported>      stalled-cycles-backend
    65,095,366,607      instructions              #    1.83  insns per cycle          ( +-  1.73% )
    10,803,480,261      branches                  # 1326.192 M/sec                    ( +-  1.95% )
       195,601,289      branch-misses             #    1.81% of all branches          ( +-  0.39% )

       8.828660235 seconds time elapsed                                          ( +-  0.38% )

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: Remove CF_IGNORE_ICOUNT
Richard Henderson [Fri, 13 Oct 2017 19:22:28 +0000 (12:22 -0700)]
tcg: Remove CF_IGNORE_ICOUNT

Now that we have curr_cflags, we can include CF_USE_ICOUNT
early and then remove it as necessary.

Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: Add CF_LAST_IO + CF_USE_ICOUNT to CF_HASH_MASK
Richard Henderson [Fri, 13 Oct 2017 19:15:06 +0000 (12:15 -0700)]
tcg: Add CF_LAST_IO + CF_USE_ICOUNT to CF_HASH_MASK

These flags are used by target/*/translate.c,
and affect code generation.

Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agocpu-exec: lookup/generate TB outside exclusive region during step_atomic
Emilio G. Cota [Fri, 14 Jul 2017 21:56:30 +0000 (17:56 -0400)]
cpu-exec: lookup/generate TB outside exclusive region during step_atomic

Now that all code generation has been converted to check CF_PARALLEL, we can
generate !CF_PARALLEL code without having yet set !parallel_cpus --
and therefore without having to be in the exclusive region during
cpu_exec_step_atomic.

While at it, merge cpu_exec_step into cpu_exec_step_atomic.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: check CF_PARALLEL instead of parallel_cpus
Emilio G. Cota [Sun, 16 Jul 2017 19:13:52 +0000 (15:13 -0400)]
tcg: check CF_PARALLEL instead of parallel_cpus

Thereby decoupling the resulting translated code from the current state
of the system.

The tb->cflags field is not passed to tcg generation functions. So
we add a field to TCGContext, storing there a copy of tb->cflags.

Most architectures have <= 32 registers, which results in a 4-byte hole
in TCGContext. Use this hole for the new field.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotarget/sparc: check CF_PARALLEL instead of parallel_cpus
Emilio G. Cota [Fri, 14 Jul 2017 22:47:23 +0000 (18:47 -0400)]
target/sparc: check CF_PARALLEL instead of parallel_cpus

Thereby decoupling the resulting translated code from the current state
of the system.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotarget/sh4: check CF_PARALLEL instead of parallel_cpus
Emilio G. Cota [Thu, 20 Jul 2017 00:08:38 +0000 (20:08 -0400)]
target/sh4: check CF_PARALLEL instead of parallel_cpus

Thereby decoupling the resulting translated code from the current state
of the system.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotarget/s390x: check CF_PARALLEL instead of parallel_cpus
Emilio G. Cota [Fri, 14 Jul 2017 22:43:35 +0000 (18:43 -0400)]
target/s390x: check CF_PARALLEL instead of parallel_cpus

Thereby decoupling the resulting translated code from the current state
of the system.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotarget/m68k: check CF_PARALLEL instead of parallel_cpus
Emilio G. Cota [Fri, 14 Jul 2017 22:37:21 +0000 (18:37 -0400)]
target/m68k: check CF_PARALLEL instead of parallel_cpus

Thereby decoupling the resulting translated code from the current state
of the system.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotarget/i386: check CF_PARALLEL instead of parallel_cpus
Emilio G. Cota [Fri, 14 Jul 2017 22:30:49 +0000 (18:30 -0400)]
target/i386: check CF_PARALLEL instead of parallel_cpus

Thereby decoupling the resulting translated code from the current state
of the system.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotarget/hppa: check CF_PARALLEL instead of parallel_cpus
Emilio G. Cota [Fri, 14 Jul 2017 22:29:47 +0000 (18:29 -0400)]
target/hppa: check CF_PARALLEL instead of parallel_cpus

Thereby decoupling the resulting translated code from the current state
of the system.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotarget/arm: check CF_PARALLEL instead of parallel_cpus
Emilio G. Cota [Fri, 14 Jul 2017 22:20:49 +0000 (18:20 -0400)]
target/arm: check CF_PARALLEL instead of parallel_cpus

Thereby decoupling the resulting translated code from the current state
of the system.

Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: convert tb->cflags reads to tb_cflags(tb)
Emilio G. Cota [Wed, 19 Jul 2017 00:46:52 +0000 (20:46 -0400)]
tcg: convert tb->cflags reads to tb_cflags(tb)

Convert all existing readers of tb->cflags to tb_cflags, so that we
use atomic_read and therefore avoid undefined behaviour in C11.

Note that the remaining setters/getters of the field are protected
by tb_lock, and therefore do not need conversion.

Luckily all readers access the field via 'tb->cflags' (so no foo.cflags,
bar->cflags in the code base), which makes the conversion easily
scriptable:

FILES=$(git grep 'tb->cflags' target include/exec/gen-icount.h \
 accel/tcg/translator.c | cut -f1 -d':' | sort | uniq)

perl -pi -e 's/([^.>])tb->cflags/$1tb_cflags(tb)/g' $FILES
perl -pi -e 's/([a-z->.]*)(->|\.)tb->cflags/tb_cflags($1$2tb)/g' $FILES

Then manually fixed the few errors that checkpatch reported.

Compile-tested for all targets.

Suggested-by: Richard Henderson <rth@twiddle.net>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: Include CF_COUNT_MASK in CF_HASH_MASK
Richard Henderson [Fri, 13 Oct 2017 18:22:57 +0000 (11:22 -0700)]
tcg: Include CF_COUNT_MASK in CF_HASH_MASK

Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: Add CPUState cflags_next_tb
Richard Henderson [Fri, 13 Oct 2017 17:50:02 +0000 (10:50 -0700)]
tcg: Add CPUState cflags_next_tb

We were generating code during tb_invalidate_phys_page_range,
check_watchpoint, cpu_io_recompile, and (seemingly) discarding
the TB, assuming that it would magically be picked up during
the next iteration through the cpu_exec loop.

Instead, record the desired cflags in CPUState so that we request
the proper TB so that there is no more magic.

Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: define CF_PARALLEL and use it for TB hashing along with CF_COUNT_MASK
Emilio G. Cota [Tue, 11 Jul 2017 18:29:37 +0000 (14:29 -0400)]
tcg: define CF_PARALLEL and use it for TB hashing along with CF_COUNT_MASK

This will enable us to decouple code translation from the value
of parallel_cpus at any given time. It will also help us minimize
TB flushes when generating code via EXCP_ATOMIC.

Note that the declaration of parallel_cpus is brought to exec-all.h
to be able to define there the "curr_cflags" inline.

Signed-off-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: Use offsets not indices for TCGv_*
Richard Henderson [Fri, 20 Oct 2017 19:08:19 +0000 (12:08 -0700)]
tcg: Use offsets not indices for TCGv_*

Using the offset of a temporary, relative to TCGContext, rather than
its index means that we don't use 0.  That leaves offset 0 free for
a NULL representation without having to leave index 0 unused.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agoqom: Introduce CPUClass.tcg_initialize
Richard Henderson [Mon, 16 Oct 2017 02:02:42 +0000 (19:02 -0700)]
qom: Introduce CPUClass.tcg_initialize

Move target cpu tcg initialization to common code,
called from cpu_exec_realizefn.

Acked-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: Remove TCGV_EQUAL*
Richard Henderson [Fri, 20 Oct 2017 03:27:27 +0000 (20:27 -0700)]
tcg: Remove TCGV_EQUAL*

When we used structures for TCGv_*, we needed a macro in order to
perform a comparison.  Now that we use pointers, this is just clutter.

Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6 years agotcg: Remove GET_TCGV_* and MAKE_TCGV_*
Richard Henderson [Fri, 20 Oct 2017 07:30:24 +0000 (00:30 -0700)]
tcg: Remove GET_TCGV_* and MAKE_TCGV_*

The GET and MAKE functions weren't really specific enough.
We now have a full complement of functions that convert exactly
between temporaries, arguments, tcgv pointers, and indices.

The target/sparc change is also a bug fix, which would have affected
a host that defines TCG_TARGET_HAS_extr[lh]_i64_i32, i.e. MIPS64.

Reviewed-by: Emilio G. Cota <cota@braap.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>