fpu/softfloat: Define floatN_silence_nan in terms of parts_silence_nan

Isolate the target-specific choice to 3 functions instead of 6.

The code in floatx80_default_nan tried to be over-general.  There are
only two targets that support this format: x86 and m68k.  Thus there
is no point in inventing a mechanism for snan_bit_is_one.

Move routines that no longer have ifdefs out of softfloat-specialize.h.

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>

fpu/softfloat: Clean up parts_default_nan

Reduce the number of ifdefs.  Correct the result for OpenRISC
and TriCore (although TriCore fixed in target-specific code).

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>

fpu/softfloat: Define floatN_default_nan in terms of parts_default_nan

Isolate the target-specific choice to 2 functions instead of 6.

The code in float16_default_nan was only correct for ARM, MIPS, and X86.
Though float16 support is rare among our targets.

The code in float128_default_nan was arguably wrong for Sparc.  While
QEMU supports the Sparc 128-bit insns, no real cpu enables it.

The code in floatx80_default_nan tried to be over-general.  There are
only two targets that support this format: x86 and m68k.  Thus there
is no point in inventing a value for snan_bit_is_one.

Move routines that no longer have ifdefs out of softfloat-specialize.h.

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>

fpu/softfloat: Pass FloatClass to pickNaNMulAdd

For each operand, pass a single enumeration instead of a pair of booleans.
The commit also merges multiple different ifdef-selected implementations
of pickNaNMulAdd into a single function whose body is ifdef-selected.

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

fpu/softfloat: Pass FloatClass to pickNaN

For each operand, pass a single enumeration instead of a pair of booleans.
The commit also merges multiple different ifdef-selected implementations
of pickNaN into a single function whose body is ifdef-selected.

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

fpu/softfloat: Make is_nan et al available to softfloat-specialize.h

We will need these helpers within softfloat-specialize.h, so move
the definitions above the include.  After specialization, they will
not always be used so mark them to avoid the Werror.

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

fpu/softfloat: Specialize on snan_bit_is_one

Only MIPS requires snan_bit_is_one to be variable.  While we are
specializing softfloat behaviour, allow other targets to eliminate
this runtime check.

Cc: Aurelien Jarno <aurelien@aurel32.net>
Cc: Yongbok Kim <yongbok.kim@mips.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Alexander Graf <agraf@suse.de>
Cc: Guan Xuetao <gxt@mprc.pku.edu.cn>
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

fpu/softfloat: Remove floatX_maybe_silence_nan

These functions are now unused.

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

fpu/softfloat: Use float*_silence_nan in propagateFloat*NaN

We have already checked the arguments for SNaN;
we don't need to do it again.

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

target/s390x: Remove floatX_maybe_silence_nan from conversions

This is now handled properly by the generic softfloat code.

Cc: Alexander Graf <agraf@suse.de>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/riscv: Remove floatX_maybe_silence_nan from conversions

This is now handled properly by the generic softfloat code.

Cc: Palmer Dabbelt <palmer@sifive.com>
Cc: Sagar Karandikar <sagark@eecs.berkeley.edu>
Cc: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
Reviewed-by: Michael Clark <mjc@sifive.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/mips: Remove floatX_maybe_silence_nan from conversions

This is now handled properly by the generic softfloat code.

Cc: Aurelien Jarno <aurelien@aurel32.net>
Cc: Yongbok Kim <yongbok.kim@mips.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/m68k: Use floatX_silence_nan when we have already checked for SNaN

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/hppa: Remove floatX_maybe_silence_nan from conversions

This is now handled properly by the generic softfloat code.

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

target/arm: Remove floatX_maybe_silence_nan from conversions

This is now handled properly by the generic softfloat code.

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

target/arm: Use floatX_silence_nan when we have already checked for SNaN

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

fpu/softfloat: re-factor float to float conversions

This allows us to delete a lot of additional boilerplate
code which is no longer needed.

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

fpu/softfloat: Partial support for ARM Alternative half-precision

For float16 ARM supports an alternative half-precision format which
sacrifices the ability to represent NaN/Inf in return for a higher
dynamic range.  The new FloatFmt flag, arm_althp, is then used to
modify the behaviour of canonicalize and round_canonical with respect
to representation and exception raising.

Usage of this new flag waits until we re-factor float-to-float conversions.

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

target/arm: squash FZ16 behaviour for conversions

The ARM ARM specifies FZ16 is suppressed for conversions. Rather than
pushing this logic into the softfloat code we can simply save the FZ
state and temporarily disable it for the softfloat call.

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

target/arm: convert conversion helpers to fpst/ahp_flag

Instead of passing env and leaving it up to the helper to get the
right fpstatus we pass it explicitly. There was already a get_fpstatus
helper for neon for the 32 bit code. We also add an get_ahp_flag() for
passing the state of the alternative FP16 format flag. This leaves
scope for later tracking the AHP state in translation flags.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

fpu/softfloat: Replace float_class_msnan with parts_silence_nan

With a canonical representation of NaNs, we can silence an SNaN
immediately rather than delay until the final format is known.

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

fpu/softfloat: Replace float_class_dnan with parts_default_nan

With a canonical representation of NaNs, we can return the
default nan directly rather than delay the expansion until
the final format is known.

Note one case where we uselessly assigned to a.sign, which was
overwritten/ignored later when expanding float_class_dnan.

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

fpu/softfloat: Introduce parts_is_snan_frac

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

fpu/softfloat: Canonicalize NaN fraction

Shift the NaN fraction to a canonical position, much like we
do for the fraction of normal numbers.  This will facilitate
manipulation of NaNs within the shared code paths.

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

fpu/softfloat: Move softfloat-specialize.h below FloatParts definition

We want to be able to specialize on the canonical representation.

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

fpu/softfloat: Split floatXX_silence_nan from floatXX_maybe_silence_nan

The new function assumes that the input is an SNaN and
does not double-check.

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

fpu/softfloat: Merge NO_SIGNALING_NANS definitions

Move the ifdef inside the relevant functions instead of
duplicating the function declarations.

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

fpu/softfloat: Fix conversion from uint64 to float128

The significand is passed to normalizeRoundAndPackFloat128() as high
first, low second. The current code passes the integer first, so the
result is incorrectly shifted left by 64 bits.

This bug affects the emulation of s390x instruction CXLGBR (convert
from logical 64-bit binary-integer operand to extended BFP result).

Cc: qemu-stable@nongnu.org
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Petr Tesarik <ptesarik@suse.com>
Message-Id: <20180511071052.1443-1-ptesarik@suse.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Merge remote-tracking branch 'remotes/cody/tags/block-pull-request' into staging

# gpg: Signature made Wed 16 May 2018 18:38:15 BST
# gpg:                using RSA key BDBE7B27C0DE3057
# gpg: Good signature from "Jeffrey Cody <jcody@redhat.com>"
# gpg:                 aka "Jeffrey Cody <jeff@codyprime.org>"
# gpg:                 aka "Jeffrey Cody <codyprime@gmail.com>"
# Primary key fingerprint: 9957 4B4D 3474 90E7 9D98  D624 BDBE 7B27 C0DE 3057

* remotes/cody/tags/block-pull-request:
  nfs: Remove processed options from QDict
  nfs: Fix error path in nfs_options_qdict_to_qapi()
  blockjob: do not cancel timer in resume
  qemu-iotests: reduce chance of races in 185

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

Merge remote-tracking branch 'remotes/juanquintela/tags/migration/20180515' into staging

migration/next for 20180515

# gpg: Signature made Tue 15 May 2018 22:54:38 BST
# gpg:                using RSA key F487EF185872D723
# 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/20180515: (40 commits)
  Migration+TLS: Fix crash due to double cleanup
  migration: Textual fixups for blocktime
  migration: update index field when delete or qsort RDMALocalBlock
  migration: update docs
  migration/hmp: add migrate_pause command
  migration/qmp: add command migrate-pause
  migration: introduce lock for to_dst_file
  hmp/migration: add migrate_recover command
  qmp/migration: new command migrate-recover
  migration: init dst in migration_object_init too
  migration: final handshake for the resume
  migration: setup ramstate for resume
  migration: synchronize dirty bitmap for resume
  migration: introduce SaveVMHandlers.resume_prepare
  migration: new message MIG_RP_MSG_RESUME_ACK
  migration: new cmd MIG_CMD_POSTCOPY_RESUME
  migration: new message MIG_RP_MSG_RECV_BITMAP
  migration: new cmd MIG_CMD_RECV_BITMAP
  migration: wakeup dst ram-load-thread for recover
  migration: new state "postcopy-recover"
  ...

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

Merge remote-tracking branch 'remotes/ehabkost/tags/x86-next-pull-request' into staging

x86 queue, 2018-05-15

* KnightsMill CPU model
* CLDEMOTE(Demote Cache Line) cpu feature
* pc-i440fx-2.13 and pc-q35-2.13 machine-types
* Add model-specific cache information to EPYC CPU model

# gpg: Signature made Tue 15 May 2018 22:53:12 BST
# gpg:                using RSA key 2807936F984DC5A6
# gpg: Good signature from "Eduardo Habkost <ehabkost@redhat.com>"
# Primary key fingerprint: 5A32 2FD5 ABC4 D3DB ACCF  D1AA 2807 936F 984D C5A6

* remotes/ehabkost/tags/x86-next-pull-request:
  i386: Add new property to control cache info
  pc: add 2.13 machine types
  i386: Initialize cache information for EPYC family processors
  i386: Add cache information in X86CPUDefinition
  i386: Helpers to encode cache information consistently
  x86/cpu: Enable CLDEMOTE(Demote Cache Line) cpu feature
  i386: add KnightsMill cpu model

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

nfs: Remove processed options from QDict

Commit c22a03454 QAPIfied option parsing in the NFS block driver, but
forgot to remove all the options we processed. Therefore, we get an
error in bdrv_open_inherit(), which thinks the remaining options are
invalid. Trying to open an NFS image will result in an error like this:

    Block protocol 'nfs' doesn't support the option 'server.host'

Remove all options from the QDict to make the NFS driver work again.

Cc: qemu-stable@nongnu.org
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 20180516160816.26259-1-kwolf@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Jeff Cody <jcody@redhat.com>

nfs: Fix error path in nfs_options_qdict_to_qapi()

Don't throw away local_err, but propagate it to errp.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 20180516161034.27440-1-kwolf@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Jeff Cody <jcody@redhat.com>

blockjob: do not cancel timer in resume

Currently the timer is cancelled and the block job is entered by
block_job_resume().  This behavior causes drain to run extra blockjob
iterations when the job was sleeping due to the ratelimit.

This patch leaves the job asleep when block_job_resume() is called.
Jobs can still be forcibly woken up using block_job_enter(), which is
used to cancel jobs.

After this patch drain no longer runs extra blockjob iterations.  This
is the expected behavior that qemu-iotests 185 used to rely on.  We
temporarily changed the 185 test output to make it pass for the QEMU
2.12 release but now it's time to address this issue.

Cc: QingFeng Hao <haoqf@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: QingFeng Hao <haoqf@linux.vnet.ibm.com>
Message-id: 20180508135436.30140-3-stefanha@redhat.com
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Jeff Cody <jcody@redhat.com>

qemu-iotests: reduce chance of races in 185

Commit 8565c3ab537e78f3e69977ec2c609dc9417a806e ("qemu-iotests: fix
185") identified a race condition in a sub-test.

Similar issues also affect the other sub-tests.  If disk I/O completes
quickly, it races with the QMP 'quit' command.  This causes spurious
test failures because QMP events are emitted in an unpredictable order.

This test relies on QEMU internals and there is no QMP API for getting
deterministic behavior needed to make this test 100% reliable.  At the
same time, the test is useful and it would be a shame to remove it.

Add sleep 0.5 to reduce the chance of races.  This is not a real fix but
appears to reduce spurious failures in practice.

Cc: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20180508135436.30140-2-stefanha@redhat.com
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Jeff Cody <jcody@redhat.com>

Migration+TLS: Fix crash due to double cleanup

During a TLS connect we see:
  migration_channel_connect calls
  migration_tls_channel_connect
  (calls after TLS setup)
  migration_channel_connect

My previous error handling fix made migration_channel_connect
call migrate_fd_connect in all cases; unfortunately the above
means it gets called twice and crashes doing double cleanup.

Fixes: 688a3dcba98

Reported-by: Peter Krempa <pkrempa@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20180430185943.35714-1-dgilbert@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: Textual fixups for blocktime

Blank lines and comments as suggested by Eric.

Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180427111502.9822-1-dgilbert@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: update index field when delete or qsort RDMALocalBlock

rdma_delete_block function deletes RDMALocalBlock base on index field,
but not update the index field. So when next time invoke rdma_delete_block,
it will not work correctly.

If start and cancel migration repeatedly, some RDMALocalBlock not invoke
ibv_dereg_mr to decrease kernel mm_struct vmpin. When vmpin is large than
max locked memory limitation, ibv_reg_mr will failed, and migration can not
start successfully again.

Signed-off-by: Lidong Chen <lidongchen@tencent.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <1525618499-1560-1-git-send-email-lidongchen@tencent.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Lidong Chen <jemmy858585@gmail.com>

migration: update docs

Update the migration docs:

Among other changes:
  * Added a general list of advice for device authors
  * Reordered the section on conditional state (subsections etc)
    into the order we prefer.
  * Add a note about firmware

Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Balamuruhan S <bala24@linux.vnet.ibm.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Message-Id: <20180503191059.19576-1-dgilbert@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration/hmp: add migrate_pause command

Wrapper for QMP command "migrate-pause".

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-25-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration/qmp: add command migrate-pause

It pauses an ongoing migration.  Currently it only supports postcopy.
Note that this command will work on either side of the migration.
Basically when we trigger this on one side, it'll interrupt the other
side as well since the other side will get notified on the disconnect
event.

However, it's still possible that the other side is not notified, for
example, when the network is totally broken, or due to some firewall
configuration changes.  In that case, we will also need to run the same
command on the other side so both sides will go into the paused state.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-24-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
---
s/2.12/2.13/

migration: introduce lock for to_dst_file

Let's introduce a lock for that QEMUFile since we are going to operate
on it in multiple threads.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-23-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

hmp/migration: add migrate_recover command

Sister command to migrate-recover in QMP.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-22-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

qmp/migration: new command migrate-recover

The first allow-oob=true command.  It's used on destination side when
the postcopy migration is paused and ready for a recovery.  After
execution, a new migration channel will be established for postcopy to
continue.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-21-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
---
s/2.12/2.13/

migration: init dst in migration_object_init too

Though we may not need it, now we init both the src/dst migration
objects in migration_object_init() so that even incoming migration
object would be thread safe (it was not).

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-20-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: final handshake for the resume

Finish the last step to do the final handshake for the recovery.

First source sends one MIG_CMD_RESUME to dst, telling that source is
ready to resume.

Then, dest replies with MIG_RP_MSG_RESUME_ACK to source, telling that
dest is ready to resume (after switch to postcopy-active state).

When source received the RESUME_ACK, it switches its state to
postcopy-active, and finally the recovery is completed.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-19-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: setup ramstate for resume

After we updated the dirty bitmaps of ramblocks, we also need to update
the critical fields in RAMState to make sure it is ready for a resume.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-18-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: synchronize dirty bitmap for resume

This patch implements the first part of core RAM resume logic for
postcopy. ram_resume_prepare() is provided for the work.

When the migration is interrupted by network failure, the dirty bitmap
on the source side will be meaningless, because even the dirty bit is
cleared, it is still possible that the sent page was lost along the way
to destination. Here instead of continue the migration with the old
dirty bitmap on source, we ask the destination side to send back its
received bitmap, then invert it to be our initial dirty bitmap.

The source side send thread will issue the MIG_CMD_RECV_BITMAP requests,
once per ramblock, to ask for the received bitmap. On destination side,
MIG_RP_MSG_RECV_BITMAP will be issued, along with the requested bitmap.
Data will be received on the return-path thread of source, and the main
migration thread will be notified when all the ramblock bitmaps are
synchronized.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-17-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: introduce SaveVMHandlers.resume_prepare

This is hook function to be called when a postcopy migration wants to
resume from a failure. For each module, it should provide its own
recovery logic before we switch to the postcopy-active state.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-16-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: new message MIG_RP_MSG_RESUME_ACK

Creating new message to reply for MIG_CMD_POSTCOPY_RESUME. One uint32_t
is used as payload to let the source know whether destination is ready
to continue the migration.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-15-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: new cmd MIG_CMD_POSTCOPY_RESUME

Introducing this new command to be sent when the source VM is ready to
resume the paused migration.  What the destination does here is
basically release the fault thread to continue service page faults.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-14-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: new message MIG_RP_MSG_RECV_BITMAP

Introducing new return path message MIG_RP_MSG_RECV_BITMAP to send
received bitmap of ramblock back to source.

This is the reply message of MIG_CMD_RECV_BITMAP, it contains not only
the header (including the ramblock name), and it was appended with the
whole ramblock received bitmap on the destination side.

When the source receives such a reply message (MIG_RP_MSG_RECV_BITMAP),
it parses it, convert it to the dirty bitmap by inverting the bits.

One thing to mention is that, when we send the recv bitmap, we are doing
these things in extra:

- converting the bitmap to little endian, to support when hosts are
  using different endianess on src/dst.

- do proper alignment for 8 bytes, to support when hosts are using
  different word size (32/64 bits) on src/dst.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-13-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: new cmd MIG_CMD_RECV_BITMAP

Add a new vm command MIG_CMD_RECV_BITMAP to request received bitmap for
one ramblock.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-12-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: wakeup dst ram-load-thread for recover

On the destination side, we cannot wake up all the threads when we got
reconnected. The first thing to do is to wake up the main load thread,
so that we can continue to receive valid messages from source again and
reply when needed.

At this point, we switch the destination VM state from postcopy-paused
back to postcopy-recover.

Now we are finally ready to do the resume logic.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-11-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: new state "postcopy-recover"

Introducing new migration state "postcopy-recover". If a migration
procedure is paused and the connection is rebuilt afterward
successfully, we'll switch the source VM state from "postcopy-paused" to
the new state "postcopy-recover", then we'll do the resume logic in the
migration thread (along with the return path thread).

This patch only do the state switch on source side. Another following up
patch will handle the state switching on destination side using the same
status bit.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-10-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
---
s/2.11/2.13/

migration: rebuild channel on source

This patch detects the "resume" flag of migration command, rebuild the
channels only if the flag is set.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-9-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

qmp: hmp: add migrate "resume" option

It will be used when we want to resume one paused migration.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-8-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
---
s/2.12/2.13/

migration: allow fault thread to pause

Allows the fault thread to stop handling page faults temporarily. When
network failure happened (and if we expect a recovery afterwards), we
should not allow the fault thread to continue sending things to source,
instead, it should halt for a while until the connection is rebuilt.

When the dest main thread noticed the failure, it kicks the fault thread
to switch to pause state.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-7-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: allow src return path to pause

Let the thread pause for network issues.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-6-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: allow dst vm pause on postcopy

When there is IO error on the incoming channel (e.g., network down),
instead of bailing out immediately, we allow the dst vm to switch to the
new POSTCOPY_PAUSE state. Currently it is still simple - it waits the
new semaphore, until someone poke it for another attempt.

One note is that here on ram loading thread we cannot detect the
POSTCOPY_ACTIVE state, but we need to detect the more specific
POSTCOPY_INCOMING_RUNNING state, to make sure we have already loaded all
the device states.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-5-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: implement "postcopy-pause" src logic

Now when network down for postcopy, the source side will not fail the
migration. Instead we convert the status into this new paused state, and
we will try to wait for a rescue in the future.

If a recovery is detected, migration_thread() will reset its local
variables to prepare for that.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-4-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: new postcopy-pause state

Introducing a new state "postcopy-paused", which can be used when the
postcopy migration is paused. It is targeted for postcopy network
failure recovery.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-3-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: let incoming side use thread context

The old incoming migration is running in main thread and default
gcontext.  With the new qio_channel_add_watch_full() we can now let it
run in the thread's own gcontext (if there is one).

Currently this patch does nothing alone.  But when any of the incoming
migration is run in another iothread (e.g., the upcoming migrate-recover
command), this patch will bind the incoming logic to the iothread
instead of the main thread (which may already get page faulted and
hanged).

RDMA is not considered for now since it's not even using the QIO watch
framework at all.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180502104740.12123-2-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

migration: Define MultifdRecvParams sooner

Once there, we don't need the struct names anywhere, just the
typedefs.  And now also document all fields.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: Transmit initial package through the multifd channels

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
--

Be network agnostic.
Add error checking for all values.

migration: Delay start of migration main routines

We need to make sure that we have started all the multifd threads.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

migration: Create multifd channels

In both sides.  We still don't transmit anything through them.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

migration: Export functions to create send channels

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

migration: Be sure all recv channels are created

We need them before we start migration.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

migration: terminate_* can be called for other threads

Once there, make  count field to always be accessed with atomic
operations.  To make blocking operations, we need to know that the
thread is running, so create a bool to indicate that.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
--

Once here, s/terminate_multifd_*-threads/multifd_*_terminate_threads/
This is consistente with every other function

migration: Introduce multifd_recv_new_channel()

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

migration: Set error state in case of error

Signed-off-by: Juan Quintela <quintela@redhat.com>

tests: Migration ppc now inlines its program

No need to write it to a file.  Just need a proper firmware O:-)

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>

tests: Add migration precopy test

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>

migration: fix saving normal page even if it's been compressed

Fix the bug introduced by da3f56cb2e767016 (migration: remove
ram_save_compressed_page()), It should be 'return' rather than
'res'

Sorry for this stupid mistake :(

Signed-off-by: Xiao Guangrong <xiaoguangrong@tencent.com>
Message-Id: <20180428081045.8878-1-xiaoguangrong@tencent.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>

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

Block layer patches:

- Switch AIO/callback based block drivers to a byte-based interface
- Block jobs: Expose error string via query-block-jobs
- Block job cleanups and fixes
- hmp: Allow using a qdev id in block_set_io_throttle

# gpg: Signature made Tue 15 May 2018 16:33:10 BST
# gpg:                using RSA key 7F09B272C88F2FD6
# 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: (37 commits)
  iotests: Add test for -U/force-share conflicts
  qemu-img: Use only string options in img_open_opts
  qemu-io: Use purely string blockdev options
  block: Document BDRV_REQ_WRITE_UNCHANGED support
  qemu-img: Check post-truncation size
  iotests: Add test for COR across nodes
  iotests: Copy 197 for COR filter driver
  iotests: Clean up wrap image in 197
  block: Support BDRV_REQ_WRITE_UNCHANGED in filters
  block/quorum: Support BDRV_REQ_WRITE_UNCHANGED
  block: Set BDRV_REQ_WRITE_UNCHANGED for COR writes
  block: Add BDRV_REQ_WRITE_UNCHANGED flag
  block: BLK_PERM_WRITE includes ..._UNCHANGED
  block: Add COR filter driver
  iotests: Skip 181 and 201 without userfaultfd
  iotests: Add failure matching to common.qemu
  docs: Document the new default sizes of the qcow2 caches
  qcow2: Give the refcount cache the minimum possible size by default
  specs/qcow2: Clarify that compressed clusters have the COPIED bit reset
  Fix error message about compressed clusters with OFLAG_COPIED
  ...

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

i386: Add new property to control cache info

The property legacy-cache will be used to control the cache information.
If user passes "-cpu legacy-cache" then older information will
be displayed even if the hardware supports new information. Otherwise
use the statically loaded cache definitions if available.

Renamed the previous cache structures to legacy_*. If there is any change in
the cache information, then it needs to be initialized in builtin_x86_defs.

Signed-off-by: Babu Moger <babu.moger@amd.com>
Tested-by: Geoffrey McRae <geoff@hostfission.com>
Message-Id: <20180514164156.27034-3-babu.moger@amd.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

pc: add 2.13 machine types

Add pc-q35-2.13 and pc-i440fx-2.13 machine types

Signed-off-by: Babu Moger <babu.moger@amd.com>
Message-Id: <20180514164156.27034-2-babu.moger@amd.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

i386: Initialize cache information for EPYC family processors

Initialize pre-determined cache information for EPYC processors.

Signed-off-by: Babu Moger <babu.moger@amd.com>
Tested-by: Geoffrey McRae <geoff@hostfission.com>
Message-Id: <20180510204148.11687-5-babu.moger@amd.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

i386: Add cache information in X86CPUDefinition

Add cache information in X86CPUDefinition and CPUX86State.

Signed-off-by: Babu Moger <babu.moger@amd.com>
Tested-by: Geoffrey McRae <geoff@hostfission.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <20180510204148.11687-3-babu.moger@amd.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

i386: Helpers to encode cache information consistently

Instead of having a collection of macros that need to be used in
complex expressions to build CPUID data, define a CPUCacheInfo
struct that can hold information about a given cache.  Helper
functions will take a CPUCacheInfo struct as input to encode
CPUID leaves for a cache.

This will help us ensure consistency between cache information
CPUID leaves, and make the existing inconsistencies in CPUID info
more visible.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Babu Moger <babu.moger@amd.com>
Tested-by: Geoffrey McRae <geoff@hostfission.com>
Message-Id: <20180510204148.11687-2-babu.moger@amd.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

x86/cpu: Enable CLDEMOTE(Demote Cache Line) cpu feature

The CLDEMOTE instruction hints to hardware that the cache line that
contains the linear address should be moved("demoted") from
the cache(s) closest to the processor core to a level more distant
from the processor core. This may accelerate subsequent accesses
to the line by other cores in the same coherence domain,
especially if the line was written by the core that demotes the line.

Intel Snow Ridge has added new cpu feature, CLDEMOTE.
The new cpu feature needs to be exposed to guest VM.

The bit definition:
CPUID.(EAX=7,ECX=0):ECX[bit 25] CLDEMOTE

The release document ref below link:
https://software.intel.com/sites/default/files/managed/c5/15/\
architecture-instruction-set-extensions-programming-reference.pdf

Signed-off-by: Jingqi Liu <jingqi.liu@intel.com>
Message-Id: <1525406253-54846-1-git-send-email-jingqi.liu@intel.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

i386: add KnightsMill cpu model

A new cpu model called "KnightsMill" is added to model Knights Mill
processors.  Compared to "Skylake-Server" cpu model, the following
features are added:

avx512_4vnniw avx512_4fmaps avx512pf avx512er avx512_vpopcntdq

and the following features are removed:

pcid invpcid clflushopt avx512dq avx512bw clwb smap rtm mpx
xsavec xgetbv1 hle

Signed-off-by: Boqun Feng <boqun.feng@intel.com>
Message-Id: <20180320000821.8337-1-boqun.feng@intel.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>

Merge remote-tracking branch 'mreitz/tags/pull-block-2018-05-15' into queue-block

- Copy-on-read block driver
- The qcow2 default refcount cache size has been decreased
- Various bug fixes

# gpg: Signature made Tue May 15 16:18:25 2018 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-2018-05-15: (21 commits)
  iotests: Add test for -U/force-share conflicts
  qemu-img: Use only string options in img_open_opts
  qemu-io: Use purely string blockdev options
  block: Document BDRV_REQ_WRITE_UNCHANGED support
  qemu-img: Check post-truncation size
  iotests: Add test for COR across nodes
  iotests: Copy 197 for COR filter driver
  iotests: Clean up wrap image in 197
  block: Support BDRV_REQ_WRITE_UNCHANGED in filters
  block/quorum: Support BDRV_REQ_WRITE_UNCHANGED
  block: Set BDRV_REQ_WRITE_UNCHANGED for COR writes
  block: Add BDRV_REQ_WRITE_UNCHANGED flag
  block: BLK_PERM_WRITE includes ..._UNCHANGED
  block: Add COR filter driver
  iotests: Skip 181 and 201 without userfaultfd
  iotests: Add failure matching to common.qemu
  docs: Document the new default sizes of the qcow2 caches
  qcow2: Give the refcount cache the minimum possible size by default
  specs/qcow2: Clarify that compressed clusters have the COPIED bit reset
  Fix error message about compressed clusters with OFLAG_COPIED
  ...

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

iotests: Add test for -U/force-share conflicts

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180502202051.15493-4-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

qemu-img: Use only string options in img_open_opts

img_open_opts() takes a QemuOpts and converts them to a QDict, so all
values therein are strings.  Then it may try to call qdict_get_bool(),
however, which will fail with a segmentation fault every time:

$ ./qemu-img info -U --image-opts \
    driver=file,filename=/dev/null,force-share=off
[1]    27869 segmentation fault (core dumped)  ./qemu-img info -U
--image-opts driver=file,filename=/dev/null,force-share=off

Fix this by using qdict_get_str() and comparing the value as a string.
Also, when adding a force-share value to the QDict, add it as a string
so it fits the rest of the dict.

Cc: qemu-stable@nongnu.org
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180502202051.15493-3-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

qemu-io: Use purely string blockdev options

Currently, qemu-io only uses string-valued blockdev options (as all are
converted directly from QemuOpts) -- with one exception: -U adds the
force-share option as a boolean.  This in itself is already a bit
questionable, but a real issue is that it also assumes the value already
existing in the options QDict would be a boolean, which is wrong.

That has the following effect:

$ ./qemu-io -r -U --image-opts \
    driver=file,filename=/dev/null,force-share=off
[1]    15200 segmentation fault (core dumped)  ./qemu-io -r -U
--image-opts driver=file,filename=/dev/null,force-share=off

Since @opts is converted from QemuOpts, the value must be a string, and
we have to compare it as such.  Consequently, it makes sense to also set
it as a string instead of a boolean.

Cc: qemu-stable@nongnu.org
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180502202051.15493-2-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block: Document BDRV_REQ_WRITE_UNCHANGED support

Add BDRV_REQ_WRITE_UNCHANGED to the list of flags honored during pwrite
and pwrite_zeroes, and also add a note on when you absolutely need to
support it.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180502140359.18222-1-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

qemu-img: Check post-truncation size

Some block drivers (iscsi and file-posix when dealing with device files)
do not actually support truncation, even though they provide a
.bdrv_truncate() method and will happily return success when providing a
new size that does not exceed the current size.  This is because these
drivers expect the user to resize the image outside of qemu and then
provide qemu with that information through the block_resize command
(compare cb1b83e740384b4e0d950f3d7c81c02b8ce86c2e).

Of course, anyone using qemu-img resize will find that behavior useless.
So we should check the actual size of the image after the supposedly
successful truncation took place, emit an error if nothing changed and
emit a warning if the target size was not met.

Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=1523065
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180421163957.29872-1-mreitz@redhat.com
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Add test for COR across nodes

COR across nodes (that is, you have some filter node between the
actually COR target and the node that performs the COR) cannot reliably
work together with the permission system when there is no explicit COR
node that can request the WRITE_UNCHANGED permission for its child.
This is because COR (currently) sneaks its requests by the usual
permission checks, so it can work without a WRITE* permission; but if
there is a filter node in between, that will re-issue the request, which
then passes through the usual check -- and if nobody has requested a
WRITE_UNCHANGED permission, that check will fail.

There is no real direct fix apart from hoping that there is someone who
has requested that permission; in case of just the qemu-io HMP command
(and no guest device), however, that is not the case.  The real real fix
is to implement the copy-on-read flag through an implicitly added COR
node.  Such a node can request the necessary permissions as shown in
this test.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180421132929.21610-10-mreitz@redhat.com
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Copy 197 for COR filter driver

iotest 197 tests copy-on-read using the (now old) copy-on-read flag.
Copy it to 215 and modify it to use the COR filter driver instead.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180421132929.21610-9-mreitz@redhat.com
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Clean up wrap image in 197

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20180421132929.21610-8-mreitz@redhat.com
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block: Support BDRV_REQ_WRITE_UNCHANGED in filters

Update the rest of the filter drivers to support
BDRV_REQ_WRITE_UNCHANGED.  They already forward write request flags to
their children, so we just have to announce support for it.

This patch does not cover the replication driver because that currently
does not support flags at all, and because it just grabs the WRITE
permission for its children when it can, so we should be fine just
submitting the incoming WRITE_UNCHANGED requests as normal writes.

It also does not cover format drivers for similar reasons.  They all use
bdrv_format_default_perms() as their .bdrv_child_perm() implementation
so they just always grab the WRITE permission for their file children
whenever possible.  In addition, it often would be difficult to
ascertain whether incoming unchanging writes end up as unchanging writes
in their files.  So we just leave them as normal potentially changing
writes.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20180421132929.21610-7-mreitz@redhat.com
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block/quorum: Support BDRV_REQ_WRITE_UNCHANGED

We just need to forward it to quorum's children (except in case of a
rewrite because of corruption), but for that we first have to support
flags in child requests at all.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20180421132929.21610-6-mreitz@redhat.com
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block: Set BDRV_REQ_WRITE_UNCHANGED for COR writes

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20180421132929.21610-5-mreitz@redhat.com
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block: Add BDRV_REQ_WRITE_UNCHANGED flag

This flag signifies that a write request will not change the visible
disk content.  With this flag set, it is sufficient to have the
BLK_PERM_WRITE_UNCHANGED permission instead of BLK_PERM_WRITE.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20180421132929.21610-4-mreitz@redhat.com
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block: BLK_PERM_WRITE includes ..._UNCHANGED

Currently we never actually check whether the WRITE_UNCHANGED
permission has been taken for unchanging writes.  But the one check that
is commented out checks both WRITE and WRITE_UNCHANGED; and considering
that WRITE_UNCHANGED is already documented as being weaker than WRITE,
we should probably explicitly document WRITE to include WRITE_UNCHANGED.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20180421132929.21610-3-mreitz@redhat.com
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block: Add COR filter driver

This adds a simple copy-on-read filter driver.  It relies on the already
existing COR functionality in the central block layer code, which may be
moved here once we no longer need it there.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180421132929.21610-2-mreitz@redhat.com
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Skip 181 and 201 without userfaultfd

userfaultfd support depends on the host kernel, so it may not be
available.  If so, 181 and 201 should be skipped.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180406151731.4285-3-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Add failure matching to common.qemu

Currently, common.qemu only allows to match for results indicating
success.  The only way to fail is by provoking a timeout.  However,
sometimes we do have a defined failure output and can match for that,
which saves us from having to wait for the timeout in case of failure.
Because failure can sometimes just result in a _notrun in the test, it
is actually important to care about being able to fail quickly.

Also, sometimes we simply do not get any specific output in case of
success.  The only way to handle this currently would be to define an
error message as the string to look for, which means that actual success
results in a timeout.  This is really bad because it unnecessarily slows
down a succeeding test.

Therefore, this patch adds a new parameter $success_or_failure to
_timed_wait_for and _send_qemu_cmd.  Setting this to a non-empty string
makes both commands expect two match parameters: If the first matches,
the function succeeds.  If the second matches, the function fails.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180406151731.4285-2-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>