target/riscv: rationalise softfloat includes

We should avoid including the whole of softfloat headers in cpu.h and
explicitly include it only where we will be calling softfloat
functions. We can use the -types.h and -helpers.h in cpu.h for the few
bits that are global.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Acked-by: Palmer Dabbelt <palmer@sifive.com>

target/mips: rationalise softfloat includes

We should avoid including the whole of softfloat headers in cpu.h and
explicitly include it only where we will be calling softfloat
functions. We can use the -types.h in cpu.h for the few bits that are
global. We also move the restore_snan_bit_mode into internal.h and
include -helpers.h there.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Aleksandar Markovic <amarkovic@wavecomp.com>

fpu: rename softfloat-specialize.h -> .inc.c

This is not a normal header and should only be included in the main
softfloat.c file to bring in the various target specific
specialisations. Indeed as it contains non-inlined C functions it is
not even a legal header. Rename it to match our included C convention.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>

fpu: make softfloat-macros "self-contained"

The macros use the "flags" type and to be consistent if anyone just
needs the macros we should bring in the header we need. There is an
outstanding TODO to audit the use of "flags" and replace with bool at
which point this include could be dropped.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Acked-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>

fpu: move inline helpers into a separate header

There are a bunch of users of the inline helpers who do not need
access to the entire softfloat API. Move those inline helpers into a
new header file which can be included without bringing in the rest of
the world.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>

fpu: remove the LIT64 macro

Now the rest of the code has been cleaned up we can remove this.

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

target/m68k: replace LIT64 with UINT64_C macros

In our quest to eliminate the home rolled LIT64 macro we fixup usage
inside for m68k's many constants.

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

fpu: replace LIT64 with UINT64_C macros

In our quest to eliminate the home rolled LIT64 macro we fixup usage
inside the softfloat code. While we are at it we remove some of the
extraneous spaces to closer fit the house style.

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

fpu: use min/max values from stdint.h for integral overflow

Remove some more use of LIT64 while making the meaning more clear. We
also avoid the need of casts as the results by definition fit into the
return type.

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

fpu: convert float[16/32/64]_squash_denormal to new modern style

This also allows us to remove the extractFloat16exp/frac helpers. We
avoid using the floatXX_pack_raw functions as they are slight overkill
for masking out all but the top bit of the number. The generated code
is almost exactly the same as makes no difference to the
pre-conversion code.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>

fpu: replace LIT64 usage with UINT64_C for specialize constants

We have a wrapper that does the right thing from stdint.h so lets use
it for our constants in softfloat-specialize.h

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

Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20190816' into staging

target-arm queue:
 * target/arm: generate a custom MIDR for -cpu max
 * hw/misc/zynq_slcr: refactor to use standard register definition
 * Set ENET_BD_BDU in I.MX FEC controller
 * target/arm: Fix routing of singlestep exceptions
 * refactor a32/t32 decoder handling of PC
 * minor optimisations/cleanups of some a32/t32 codegen
 * target/arm/cpu64: Ensure kvm really supports aarch64=off
 * target/arm/cpu: Ensure we can use the pmu with kvm
 * target/arm: Minor cleanups preparatory to KVM SVE support

# gpg: Signature made Fri 16 Aug 2019 14:15:55 BST
# gpg:                using RSA key E1A5C593CD419DE28E8315CF3C2525ED14360CDE
# gpg:                issuer "peter.maydell@linaro.org"
# gpg: Good signature from "Peter Maydell <peter.maydell@linaro.org>" [ultimate]
# gpg:                 aka "Peter Maydell <pmaydell@gmail.com>" [ultimate]
# gpg:                 aka "Peter Maydell <pmaydell@chiark.greenend.org.uk>" [ultimate]
# Primary key fingerprint: E1A5 C593 CD41 9DE2 8E83  15CF 3C25 25ED 1436 0CDE

* remotes/pmaydell/tags/pull-target-arm-20190816: (29 commits)
  target/arm: Use tcg_gen_extrh_i64_i32 to extract the high word
  target/arm: Simplify SMMLA, SMMLAR, SMMLS, SMMLSR
  target/arm: Use tcg_gen_rotri_i32 for gen_swap_half
  target/arm: Use ror32 instead of open-coding the operation
  target/arm: Remove redundant shift tests
  target/arm: Use tcg_gen_deposit_i32 for PKHBT, PKHTB
  target/arm: Use tcg_gen_extract_i32 for shifter_out_im
  target/arm/kvm64: Move the get/put of fpsimd registers out
  target/arm/kvm64: Fix error returns
  target/arm/cpu: Use div-round-up to determine predicate register array size
  target/arm/helper: zcr: Add build bug next to value range assumption
  target/arm/cpu: Ensure we can use the pmu with kvm
  target/arm/cpu64: Ensure kvm really supports aarch64=off
  target/arm: Remove helper_double_saturate
  target/arm: Use unallocated_encoding for aarch32
  target/arm: Remove offset argument to gen_exception_bkpt_insn
  target/arm: Replace offset with pc in gen_exception_internal_insn
  target/arm: Replace offset with pc in gen_exception_insn
  target/arm: Replace s->pc with s->base.pc_next
  target/arm: Remove redundant s->pc & ~1
  ...

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

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

Block layer patches:

- file-posix: Fix O_DIRECT alignment detection
- Fixes for concurrent block jobs
- block-backend: Queue requests while drained (fix IDE vs. job crashes)
- qemu-img convert: Deprecate using -n and -o together
- iotests: Migration tests with filter nodes
- iotests: More media change tests

# gpg: Signature made Fri 16 Aug 2019 10:29:18 BST
# gpg:                using RSA key 7F09B272C88F2FD6
# gpg: Good signature from "Kevin Wolf <kwolf@redhat.com>" [full]
# Primary key fingerprint: DC3D EB15 9A9A F95D 3D74  56FE 7F09 B272 C88F 2FD6

* remotes/kevin/tags/for-upstream:
  file-posix: Handle undetectable alignment
  qemu-img convert: Deprecate using -n and -o together
  block-backend: Queue requests while drained
  mirror: Keep mirror_top_bs drained after dropping permissions
  block: Remove blk_pread_unthrottled()
  iotests: Add test for concurrent stream/commit
  tests: Test mid-drain bdrv_replace_child_noperm()
  tests: Test polling in bdrv_drop_intermediate()
  block: Reduce (un)drains when replacing a child
  block: Keep subtree drained in drop_intermediate
  block: Simplify bdrv_filter_default_perms()
  iotests: Test migration with all kinds of filter nodes
  iotests: Move migration helpers to iotests.py
  iotests/118: Add -blockdev based tests
  iotests/118: Create test classes dynamically
  iotests/118: Test media change for scsi-cd

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

Merge remote-tracking branch 'remotes/ericb/tags/pull-nbd-2019-08-15' into staging

nbd patches for 2019-08-15

- Addition of InetSocketAddress keep-alive
- Addition of BDRV_REQ_PREFETCH for more efficient copy-on-read
- Initial refactoring in preparation of NBD reconnect

# gpg: Signature made Thu 15 Aug 2019 19:28:41 BST
# gpg:                using RSA key A7A16B4A2527436A
# gpg: Good signature from "Eric Blake <eblake@redhat.com>" [full]
# gpg:                 aka "Eric Blake (Free Software Programmer) <ebb9@byu.net>" [full]
# gpg:                 aka "[jpeg image of size 6874]" [full]
# Primary key fingerprint: 71C2 CC22 B1C4 6029 27D2  F3AA A7A1 6B4A 2527 436A

* remotes/ericb/tags/pull-nbd-2019-08-15:
  block/nbd: refactor nbd connection parameters
  block/nbd: add cmdline and qapi parameter reconnect-delay
  block/nbd: move from quit to state
  block/nbd: use non-blocking io channel for nbd negotiation
  block/nbd: split connection_co start out of nbd_client_connect
  nbd: improve CMD_CACHE: use BDRV_REQ_PREFETCH
  block/stream: use BDRV_REQ_PREFETCH
  block: implement BDRV_REQ_PREFETCH
  qapi: Add InetSocketAddress member keep-alive

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

Merge remote-tracking branch 'remotes/armbru/tags/pull-include-2019-08-13-v2' into staging

Header cleanup patches for 2019-08-13

# gpg: Signature made Fri 16 Aug 2019 12:39:12 BST
# gpg:                using RSA key 354BC8B3D7EB2A6B68674E5F3870B400EB918653
# gpg:                issuer "armbru@redhat.com"
# gpg: Good signature from "Markus Armbruster <armbru@redhat.com>" [full]
# gpg:                 aka "Markus Armbruster <armbru@pond.sub.org>" [full]
# Primary key fingerprint: 354B C8B3 D7EB 2A6B 6867  4E5F 3870 B400 EB91 8653

* remotes/armbru/tags/pull-include-2019-08-13-v2: (29 commits)
  sysemu: Split sysemu/runstate.h off sysemu/sysemu.h
  sysemu: Move the VMChangeStateEntry typedef to qemu/typedefs.h
  Include sysemu/sysemu.h a lot less
  Clean up inclusion of sysemu/sysemu.h
  numa: Move remaining NUMA declarations from sysemu.h to numa.h
  Include sysemu/hostmem.h less
  numa: Don't include hw/boards.h into sysemu/numa.h
  Include hw/boards.h a bit less
  Include hw/qdev-properties.h less
  Include qemu/main-loop.h less
  Include qemu/queue.h slightly less
  Include hw/hw.h exactly where needed
  Include qom/object.h slightly less
  Include exec/memory.h slightly less
  Include migration/vmstate.h less
  migration: Move the VMStateDescription typedef to typedefs.h
  Clean up inclusion of exec/cpu-common.h
  Include hw/irq.h a lot less
  typedefs: Separate incomplete types and function types
  ide: Include hw/ide/internal a bit less outside hw/ide/
  ...

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

target/arm: Use tcg_gen_extrh_i64_i32 to extract the high word

Separate shift + extract low will result in one extra insn
for hosts like RISC-V, MIPS, and Sparc.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20190808202616.13782-8-richard.henderson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Simplify SMMLA, SMMLAR, SMMLS, SMMLSR

All of the inputs to these instructions are 32-bits.  Rather than
extend each input to 64-bits and then extract the high 32-bits of
the output, use tcg_gen_muls2_i32 and other 32-bit generator functions.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20190808202616.13782-7-richard.henderson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Use tcg_gen_rotri_i32 for gen_swap_half

Rotate is the more compact and obvious way to swap 16-bit
elements of a 32-bit word.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20190808202616.13782-6-richard.henderson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Use ror32 instead of open-coding the operation

The helper function is more documentary, and also already
handles the case of rotate by zero.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20190808202616.13782-5-richard.henderson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Remove redundant shift tests

The immediate shift generator functions already test for,
and eliminate, the case of a shift by zero.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20190808202616.13782-4-richard.henderson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Use tcg_gen_deposit_i32 for PKHBT, PKHTB

Use deposit as the composit operation to merge the
bits from the two inputs.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20190808202616.13782-3-richard.henderson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Use tcg_gen_extract_i32 for shifter_out_im

Extract is a compact combination of shift + and.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20190808202616.13782-2-richard.henderson@linaro.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm/kvm64: Move the get/put of fpsimd registers out

Move the getting/putting of the fpsimd registers out of
kvm_arch_get/put_registers() into their own helper functions
to prepare for alternatively getting/putting SVE registers.

No functional change.

Signed-off-by: Andrew Jones <drjones@redhat.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm/kvm64: Fix error returns

A couple return -EINVAL's forgot their '-'s.

Signed-off-by: Andrew Jones <drjones@redhat.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm/cpu: Use div-round-up to determine predicate register array size

Unless we're guaranteed to always increase ARM_MAX_VQ by a multiple of
four, then we should use DIV_ROUND_UP to ensure we get an appropriate
array size.

Signed-off-by: Andrew Jones <drjones@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm/helper: zcr: Add build bug next to value range assumption

The current implementation of ZCR_ELx matches the architecture, only
implementing the lower four bits, with the rest RAZ/WI. This puts
a strict limit on ARM_MAX_VQ of 16. Make sure we don't let ARM_MAX_VQ
grow without a corresponding update here.

Suggested-by: Dave Martin <Dave.Martin@arm.com>
Signed-off-by: Andrew Jones <drjones@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm/cpu: Ensure we can use the pmu with kvm

We first convert the pmu property from a static property to one with
its own accessors. Then we use the set accessor to check if the PMU is
supported when using KVM. Indeed a 32-bit KVM host does not support
the PMU, so this check will catch an attempt to use it at property-set
time.

Signed-off-by: Andrew Jones <drjones@redhat.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm/cpu64: Ensure kvm really supports aarch64=off

If -cpu <cpu>,aarch64=off is used then KVM must also be used, and it
and the host must support running the vcpu in 32-bit mode. Also, if
-cpu <cpu>,aarch64=on is used, then it doesn't matter if kvm is
enabled or not.

Signed-off-by: Andrew Jones <drjones@redhat.com>
Reviewed-by: Eric Auger <eric.auger@redhat.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Remove helper_double_saturate

Replace x = double_saturate(y) with x = add_saturate(y, y).
There is no need for a separate more specialized helper.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20190807045335.1361-12-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Use unallocated_encoding for aarch32

Promote this function from aarch64 to fully general use.
Use it to unify the code sequences for generating illegal
opcode exceptions.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20190807045335.1361-11-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Remove offset argument to gen_exception_bkpt_insn

Unlike the other more generic gen_exception{,_internal}_insn
interfaces, breakpoints always refer to the current instruction.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20190807045335.1361-10-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Replace offset with pc in gen_exception_internal_insn

The offset is variable depending on the instruction set.
Passing in the actual value is clearer in intent.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20190807045335.1361-9-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Replace offset with pc in gen_exception_insn

The offset is variable depending on the instruction set, whereas
we have stored values for the current pc and the next pc.  Passing
in the actual value is clearer in intent.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20190807045335.1361-8-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Replace s->pc with s->base.pc_next

We must update s->base.pc_next when we return from the translate_insn
hook to the main translator loop.  By incrementing s->base.pc_next
immediately after reading the insn word, "pc_next" contains the address
of the next instruction throughout translation.

All remaining uses of s->pc are referencing the address of the next insn,
so this is now a simple global replacement.  Remove the "s->pc" field.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20190807045335.1361-7-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Remove redundant s->pc & ~1

The thumb bit has already been removed from s->pc, and is always even.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20190807045335.1361-6-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Introduce add_reg_for_lit

Provide a common routine for the places that require ALIGN(PC, 4)
as the base address as opposed to plain PC.  The two are always
the same for A32, but the difference is meaningful for thumb mode.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20190807045335.1361-5-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Introduce read_pc

We currently have 3 different ways of computing the architectural
value of "PC" as seen in the ARM ARM.

The value of s->pc has been incremented past the current insn,
but that is all.  Thus for a32, PC = s->pc + 4; for t32, PC = s->pc;
for t16, PC = s->pc + 2.  These differing computations make it
impossible at present to unify the various code paths.

With the newly introduced s->pc_curr, we can compute the correct
value for all cases, using the formula given in the ARM ARM.

This changes the behaviour for load_reg() and load_reg_var()
when called with reg==15 from a 32-bit Thumb instruction:
previously they would have returned the incorrect value
of pc_curr + 6, and now they will return the architecturally
correct value of PC, which is pc_curr + 4. This will not
affect well-behaved guest software, because all of the places
we call these functions from T32 code are instructions where
using r15 is UNPREDICTABLE. Using the architectural PC value
here is more consistent with the T16 and A32 behaviour.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20190807045335.1361-4-richard.henderson@linaro.org
[PMM: added commit message note about UNPREDICTABLE T32 cases]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Introduce pc_curr

Add a new field to retain the address of the instruction currently
being translated.  The 32-bit uses are all within subroutines used
by a32 and t32.  This will become less obvious when t16 support is
merged with a32+t32, and having a clear definition will help.

Convert aarch64 as well for consistency.  Note that there is one
instance of a pre-assert fprintf that used the wrong value for the
address of the current instruction.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20190807045335.1361-3-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Pass in pc to thumb_insn_is_16bit

This function is used in two different contexts, and it will be
clearer if the function is given the address to which it applies.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-id: 20190807045335.1361-2-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: Fix routing of singlestep exceptions

When generating an architectural single-step exception we were
routing it to the "default exception level", which is to say
the same exception level we execute at except that EL0 exceptions
go to EL1. This is incorrect because the debug exception level
can be configured by the guest for situations such as single
stepping of EL0 and EL1 code by EL2.

We have to track the target debug exception level in the TB
flags, because it is dependent on CPU state like HCR_EL2.TGE
and MDCR_EL2.TDE. (That we were previously calling the
arm_debug_target_el() function to determine dc->ss_same_el
is itself a bug, though one that would only have manifested
as incorrect syndrome information.) Since we are out of TB
flag bits unless we want to expand into the cs_base field,
we share some bits with the M-profile only HANDLER and
STACKCHECK bits, since only A-profile has this singlestep.

Fixes: https://bugs.launchpad.net/qemu/+bug/1838913
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Message-id: 20190805130952.4415-3-peter.maydell@linaro.org

target/arm: Factor out 'generate singlestep exception' function

Factor out code to 'generate a singlestep exception', which is
currently repeated in four places.

To do this we need to also pull the identical copies of the
gen-exception() function out of translate-a64.c and translate.c
into translate.h.

(There is a bug in the code: we're taking the exception to the wrong
target EL.  This will be simpler to fix if there's only one place to
do it.)

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Message-id: 20190805130952.4415-2-peter.maydell@linaro.org

Set ENET_BD_BDU in I.MX FEC controller

This commit properly sets the ENET_BD_BDU flag once the emulated FEC controller
has finished processing the last descriptor. This is done for both transmit
and receive descriptors.

This allows the QNX 7.0.0 BSP for the Sabrelite board (which can be
found at http://blackberry.qnx.com/en/developers/bsp) to properly
control the FEC. Without this patch, the BSP ethernet driver will never
re-use FEC descriptors, as the unset ENET_BD_BDU flag will cause
it to believe that the descriptors are still in use by the NIC.

Note that Linux does not appear to use this field at all, and is
unaffected by this patch.

Without this patch, QNX will think that the NIC is still processing its
transaction descriptors, and won't send any more data over the network.

For reference:

On page 1192 of the I.MX 6DQ reference manual revision (Rev. 5, 06/2018),
which can be found at https://www.nxp.com/products/processors-and-microcontrollers/arm-based-processors-and-mcus/i.mx-applications-processors/i.mx-6-processors/i.mx-6quad-processors-high-performance-3d-graphics-hd-video-arm-cortex-a9-core:i.MX6Q?&tab=Documentation_Tab&linkline=Application-Note

the 'BDU' field is described as follows for the 'Enhanced transmit
buffer descriptor':

'Last buffer descriptor update done. Indicates that the last BD data has been updated by
uDMA. This field is written by the user (=0) and uDMA (=1).'

The same description is used for the receive buffer descriptor.

Signed-off-by: Aaron Hill <aa1ronham@gmail.com>
Message-id: 20190805142417.10433-1-aaron.hill@alertinnovation.com
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

hw/misc/zynq_slcr: use standard register definition

Replace the zynq_slcr registers enum and macros using the
hw/registerfields.h macros.

Signed-off-by: Damien Hedde <damien.hedde@greensocs.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-id: 20190729145654.14644-30-damien.hedde@greensocs.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

target/arm: generate a custom MIDR for -cpu max

While most features are now detected by probing the ID_* registers
kernels can (and do) use MIDR_EL1 for working out of they have to
apply errata. This can trip up warnings in the kernel as it tries to
work out if it should apply workarounds to features that don't
actually exist in the reported CPU type.

Avoid this problem by synthesising our own MIDR value.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20190726113950.7499-1-alex.bennee@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Merge remote-tracking branch 'remotes/huth-gitlab/tags/pull-request-2019-08-15' into staging

- Fix for ctrl queue in the virtio-net QOS driver
- Improve Valgrind reports in the tests that use the null-co driver
- Get rid of global_qtest related code in libqtest and libqos

# gpg: Signature made Thu 15 Aug 2019 18:28:16 BST
# gpg:                using RSA key 27B88847EEE0250118F3EAB92ED9D774FE702DB5
# gpg:                issuer "thuth@redhat.com"
# gpg: Good signature from "Thomas Huth <th.huth@gmx.de>" [full]
# gpg:                 aka "Thomas Huth <thuth@redhat.com>" [full]
# gpg:                 aka "Thomas Huth <huth@tuxfamily.org>" [full]
# gpg:                 aka "Thomas Huth <th.huth@posteo.de>" [unknown]
# Primary key fingerprint: 27B8 8847 EEE0 2501 18F3  EAB9 2ED9 D774 FE70 2DB5

* remotes/huth-gitlab/tags/pull-request-2019-08-15:
  tests/libqtest: Make qmp_assert_success() independent from global_qtest
  tests/libqtest: Make qtest_qmp_device_add/del independent from global_qtest
  tests/libqtest: Clean up qtest_cb_for_every_machine() wrt global_qtest
  tests/libqtest: Remove unused function hmp()
  tests/libqos: Make virtio-pci code independent from global_qtest
  tests/libqos: Make generic virtio code independent from global_qtest
  tests: Set read-zeroes on for null-co driver
  libqos: Account for the ctrl queue in virtio-net
  qtest: Rename qtest.c:qtest_init()

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

sysemu: Split sysemu/runstate.h off sysemu/sysemu.h

sysemu/sysemu.h is a rather unfocused dumping ground for stuff related
to the system-emulator.  Evidence:

* It's included widely: in my "build everything" tree, changing
  sysemu/sysemu.h still triggers a recompile of some 1100 out of 6600
  objects (not counting tests and objects that don't depend on
  qemu/osdep.h, down from 5400 due to the previous two commits).

* It pulls in more than a dozen additional headers.

Split stuff related to run state management into its own header
sysemu/runstate.h.

Touching sysemu/sysemu.h now recompiles some 850 objects.  qemu/uuid.h
also drops from 1100 to 850, and qapi/qapi-types-run-state.h from 4400
to 4200.  Touching new sysemu/runstate.h recompiles some 500 objects.

Since I'm touching MAINTAINERS to add sysemu/runstate.h anyway, also
add qemu/main-loop.h.

Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20190812052359.30071-30-armbru@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
[Unbreak OS-X build]

sysemu: Move the VMChangeStateEntry typedef to qemu/typedefs.h

In my "build everything" tree, changing sysemu/sysemu.h triggers a
recompile of some 1800 out of 6600 objects (not counting tests and
objects that don't depend on qemu/osdep.h, down from 5400 due to the
previous commit).

Several headers include sysemu/sysemu.h just to get typedef
VMChangeStateEntry.  Move it from sysemu/sysemu.h to qemu/typedefs.h.
Spell its structure tag the same while there.  Drop the now
superfluous includes of sysemu/sysemu.h from headers.

Touching sysemu/sysemu.h now recompiles some 1100 objects.
qemu/uuid.h also drops from 1800 to 1100, and
qapi/qapi-types-run-state.h from 5000 to 4400.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20190812052359.30071-29-armbru@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>

Include sysemu/sysemu.h a lot less

In my "build everything" tree, changing sysemu/sysemu.h triggers a
recompile of some 5400 out of 6600 objects (not counting tests and
objects that don't depend on qemu/osdep.h).

hw/qdev-core.h includes sysemu/sysemu.h since recent commit e965ffa70a
"qdev: add qdev_add_vm_change_state_handler()".  This is a bad idea:
hw/qdev-core.h is widely included.

Move the declaration of qdev_add_vm_change_state_handler() to
sysemu/sysemu.h, and drop the problematic include from hw/qdev-core.h.

Touching sysemu/sysemu.h now recompiles some 1800 objects.
qemu/uuid.h also drops from 5400 to 1800.  A few more headers show
smaller improvement: qemu/notify.h drops from 5600 to 5200,
qemu/timer.h from 5600 to 4500, and qapi/qapi-types-run-state.h from
5500 to 5000.

Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20190812052359.30071-28-armbru@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

Clean up inclusion of sysemu/sysemu.h

In my "build everything" tree, changing sysemu/sysemu.h triggers a
recompile of some 5400 out of 6600 objects (not counting tests and
objects that don't depend on qemu/osdep.h).

Almost a third of its inclusions are actually superfluous.  Delete
them.  Downgrade two more to qapi/qapi-types-run-state.h, and move one
from char/serial.h to char/serial.c.

hw/semihosting/config.c, monitor/monitor.c, qdev-monitor.c, and
stubs/semihost.c define variables declared in sysemu/sysemu.h without
including it.  The compiler is cool with that, but include it anyway.

This doesn't reduce actual use much, as it's still included into
widely included headers.  The next commit will tackle that.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20190812052359.30071-27-armbru@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

numa: Move remaining NUMA declarations from sysemu.h to numa.h

Commit e35704ba9c "numa: Move NUMA declarations from sysemu.h to
numa.h" left a few NUMA-related macros behind.  Move them now.

Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <20190812052359.30071-26-armbru@redhat.com>

Include sysemu/hostmem.h less

Move the HostMemoryBackend typedef from sysemu/hostmem.h to
qemu/typedefs.h.  This renders a few inclusions of sysemu/hostmem.h
superfluous; drop them.

Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190812052359.30071-25-armbru@redhat.com>

numa: Don't include hw/boards.h into sysemu/numa.h

sysemu/numa.h includes hw/boards.h just for the CPUArchId typedef, at
the cost of pulling in more than two dozen extra headers indirectly.

I could move the typedef from hw/boards.h to qemu/typedefs.h.  But
it's used in just two headers: boards.h and numa.h.

I could move it to another header both its users include.
exec/cpu-common.h seems to be the least bad fit.

But I'm keeping this simple & stupid: declare the struct tag in
numa.h.

Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <20190812052359.30071-24-armbru@redhat.com>

Include hw/boards.h a bit less

hw/boards.h pulls in almost 60 headers.  The less we include it into
headers, the better.  As a first step, drop superfluous inclusions,
and downgrade some more to what's actually needed.  Gets rid of just
one inclusion into a header.

Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20190812052359.30071-23-armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>

Include hw/qdev-properties.h less

In my "build everything" tree, changing hw/qdev-properties.h triggers
a recompile of some 2700 out of 6600 objects (not counting tests and
objects that don't depend on qemu/osdep.h).

Many places including hw/qdev-properties.h (directly or via hw/qdev.h)
actually need only hw/qdev-core.h.  Include hw/qdev-core.h there
instead.

hw/qdev.h is actually pointless: all it does is include hw/qdev-core.h
and hw/qdev-properties.h, which in turn includes hw/qdev-core.h.
Replace the remaining uses of hw/qdev.h by hw/qdev-properties.h.

While there, delete a few superfluous inclusions of hw/qdev-core.h.

Touching hw/qdev-properties.h now recompiles some 1200 objects.

Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Daniel P. Berrangé" <berrange@redhat.com>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Message-Id: <20190812052359.30071-22-armbru@redhat.com>

Include qemu/main-loop.h less

In my "build everything" tree, changing qemu/main-loop.h triggers a
recompile of some 5600 out of 6600 objects (not counting tests and
objects that don't depend on qemu/osdep.h).  It includes block/aio.h,
which in turn includes qemu/event_notifier.h, qemu/notify.h,
qemu/processor.h, qemu/qsp.h, qemu/queue.h, qemu/thread-posix.h,
qemu/thread.h, qemu/timer.h, and a few more.

Include qemu/main-loop.h only where it's needed.  Touching it now
recompiles only some 1700 objects.  For block/aio.h and
qemu/event_notifier.h, these numbers drop from 5600 to 2800.  For the
others, they shrink only slightly.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20190812052359.30071-21-armbru@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>

Include qemu/queue.h slightly less

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190812052359.30071-20-armbru@redhat.com>

Include hw/hw.h exactly where needed

In my "build everything" tree, changing hw/hw.h triggers a recompile
of some 2600 out of 6600 objects (not counting tests and objects that
don't depend on qemu/osdep.h).

The previous commits have left only the declaration of hw_error() in
hw/hw.h.  This permits dropping most of its inclusions.  Touching it
now recompiles less than 200 objects.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20190812052359.30071-19-armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>

Include qom/object.h slightly less

hw/hw.h used to include headers hardware emulation "usually" needs.
The previous commits removed all but one of them, to good effect.
Only qom/object.h is left.  Remove that one, too.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190812052359.30071-18-armbru@redhat.com>

Include exec/memory.h slightly less

Drop unnecessary inclusions from headers.  Downgrade a few more to
exec/hwaddr.h.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190812052359.30071-17-armbru@redhat.com>

Include migration/vmstate.h less

In my "build everything" tree, changing migration/vmstate.h triggers a
recompile of some 2700 out of 6600 objects (not counting tests and
objects that don't depend on qemu/osdep.h).

hw/hw.h supposedly includes it for convenience.  Several other headers
include it just to get VMStateDescription.  The previous commit made
that unnecessary.

Include migration/vmstate.h only where it's still needed.  Touching it
now recompiles only some 1600 objects.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20190812052359.30071-16-armbru@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>

migration: Move the VMStateDescription typedef to typedefs.h

We declare incomplete struct VMStateDescription in a couple of places
so we don't have to include migration/vmstate.h for the typedef.
That's fine with me.  However, the next commit will drop
migration/vmstate.h from a massive number of compiles.  Move the
typedef to qemu/typedefs.h now, so I don't have to insert struct in
front of VMStateDescription all over the place then.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190812052359.30071-15-armbru@redhat.com>

Clean up inclusion of exec/cpu-common.h

migration/qemu-file.h neglects to include it even though it needs
ram_addr_t.  Fix that.  Drop a few superfluous inclusions elsewhere.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190812052359.30071-14-armbru@redhat.com>

Include hw/irq.h a lot less

In my "build everything" tree, changing hw/irq.h triggers a recompile
of some 5400 out of 6600 objects (not counting tests and objects that
don't depend on qemu/osdep.h).

hw/hw.h supposedly includes it for convenience.  Several other headers
include it just to get qemu_irq and.or qemu_irq_handler.

Move the qemu_irq and qemu_irq_handler typedefs from hw/irq.h to
qemu/typedefs.h, and then include hw/irq.h only where it's still
needed.  Touching it now recompiles only some 500 objects.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190812052359.30071-13-armbru@redhat.com>

typedefs: Separate incomplete types and function types

While there, rewrite the obsolete file comment.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190812052359.30071-12-armbru@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
[File comment rewritten, commit message adjusted]

ide: Include hw/ide/internal a bit less outside hw/ide/

According to hw/ide/internal's file comment, only files in hw/ide/ are
supposed to include it.  Drag reality slightly closer to supposition.

Three includes outside hw/ide remain: hw/arm/sbsa-ref.c,
include/hw/ide/pci.h, and include/hw/misc/macio/macio.h.  Turns out
board code needs ide-internal.h to wire up IDE stuff.  More cleanup is
needed.  Left for another day.

Cc: John Snow <jsnow@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190812052359.30071-11-armbru@redhat.com>

Include migration/qemu-file-types.h a lot less

In my "build everything" tree, changing migration/qemu-file-types.h
triggers a recompile of some 2600 out of 6600 objects (not counting
tests and objects that don't depend on qemu/osdep.h).

The culprit is again hw/hw.h, which supposedly includes it for
convenience.

Include migration/qemu-file-types.h only where it's needed.  Touching
it now recompiles less than 200 objects.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20190812052359.30071-10-armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>

Include sysemu/reset.h a lot less

In my "build everything" tree, changing sysemu/reset.h triggers a
recompile of some 2600 out of 6600 objects (not counting tests and
objects that don't depend on qemu/osdep.h).

The main culprit is hw/hw.h, which supposedly includes it for
convenience.

Include sysemu/reset.h only where it's needed.  Touching it now
recompiles less than 200 objects.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190812052359.30071-9-armbru@redhat.com>

trace: Do not include qom/cpu.h into generated trace.h

docs/devel/tracing.txt explains "since many source files include
trace.h, [the generated trace.h use] a minimum of types and other
header files included to keep the namespace clean and compile times
and dependencies down."

Commit 4815185902 "trace: Add per-vCPU tracing states for events with
the 'vcpu' property" made them all include qom/cpu.h via
control-internal.h.  qom/cpu.h in turn includes about thirty headers.
Ouch.

Per-vCPU tracing is currently not supported in sub-directories'
trace-events.  In other words, qom/cpu.h can only be used in
trace-root.h, not in any trace.h.

Split trace/control-vcpu.h off trace/control.h and
trace/control-internal.h.  Have the generated trace.h include
trace/control.h (which no longer includes qom/cpu.h), and trace-root.h
include trace/control-vcpu.h (which includes it).

The resulting improvement is a bit disappointing: in my "build
everything" tree, some 1100 out of 6600 objects (not counting tests
and objects that don't depend on qemu/osdep.h) depend on a trace.h,
and about 600 of them no longer depend on qom/cpu.h.  But more than
1300 others depend on trace-root.h.  More work is clearly needed.
Left for another day.

Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190812052359.30071-8-armbru@redhat.com>

trace: Eliminate use of TARGET_FMT_plx

hw/tpm/trace-events uses TARGET_FMT_plx formats with uint64_t
arguments.  That's wrong, TARGET_FMT_plx takes hwaddr.  Since hwaddr
happens to be uint64_t, it works anyway.  Messed up in commit
ec427498da5, v2.12.0.  Clean up by replacing TARGET_FMT_plx with its
macro expansion.

scripts/tracetool/format/log_stap.py (commit 62dd1048c0b, v4.0.0) has
a special case for TARGET_FMT_plx.  Delete it.

Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20190812052359.30071-7-armbru@redhat.com>

queue: Drop superfluous #include qemu/atomic.h

When commit 5f7d05ecfda added QLIST_INSERT_HEAD_RCU() to qemu/queue.h,
it had to include qemu/atomic.h.  Commit 341774fe6cc removed
QLIST_INSERT_HEAD_RCU() again, but neglected to remove the #include.
Do that now.

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20190812052359.30071-6-armbru@redhat.com>

memory: Fix type of IOMMUMemoryRegionClass member @parent_class

TYPE_IOMMU_MEMORY_REGION is a direct subtype of TYPE_MEMORY_REGION.
Its instance struct is IOMMUMemoryRegion, and its first member is a
MemoryRegion.  Correct.  Its class struct is IOMMUMemoryRegionClass,
and its first member is a DeviceClass.  Wrong.  Messed up when commit
1221a474676 introduced the QOM type.  It even included hw/qdev-core.h
just for that.

TYPE_MEMORY_REGION doesn't bother to define a class struct.  This is
fine, it simply defaults to its super-type TYPE_OBJECT's class struct
ObjectClass.  Changing IOMMUMemoryRegionClass's first member's type to
ObjectClass would be a minimal fix, if a bit brittle: if
TYPE_MEMORY_REGION ever acquired own class struct, we'd have to update
IOMMUMemoryRegionClass to use it.

Fix it the clean and robust way instead: give TYPE_MEMORY_REGION its
own class struct MemoryRegionClass now, and use it for
IOMMUMemoryRegionClass's first member.

Revert the include of hw/qdev-core.h, and fix the few files that have
come to rely on it.

Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20190812052359.30071-5-armbru@redhat.com>

qapi: Split error.json off common.json

In my "build everything" tree, changing a type in qapi/common.json
triggers a recompile of some 3600 out of 6600 objects (not counting
tests and objects that don't depend on qemu/osdep.h).

One common dependency is QapiErrorClass: it's used only in in
qapi/error.h, which uses nothing else, and is widely included.

Move QapiErrorClass from common.json to new error.json.  Touching
common.json now recompiles only some 2900 objects.

Cc: Eric Blake <eblake@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190812052359.30071-4-armbru@redhat.com>

Include generated QAPI headers less

Some of the generated qapi-types-MODULE.h are included all over the
place.  Changing a QAPI type can trigger massive recompiling.  Top
scorers recompile more than 1000 out of some 6600 objects (not
counting tests and objects that don't depend on qemu/osdep.h):

    6300 qapi/qapi-builtin-types.h
    5700 qapi/qapi-types-run-state.h
    3900 qapi/qapi-types-common.h
    3300 qapi/qapi-types-sockets.h
    3000 qapi/qapi-types-misc.h
    3000 qapi/qapi-types-crypto.h
    3000 qapi/qapi-types-job.h
    3000 qapi/qapi-types-block-core.h
    2800 qapi/qapi-types-block.h
    1300 qapi/qapi-types-net.h

Clean up headers to include generated QAPI headers only where needed.
Impact is negligible except for hw/qdev-properties.h.

This header includes qapi/qapi-types-block.h and
qapi/qapi-types-misc.h.  They are used only in expansions of property
definition macros such as DEFINE_PROP_BLOCKDEV_ON_ERROR() and
DEFINE_PROP_OFF_AUTO().  Moving their inclusion from
hw/qdev-properties.h to the users of these macros avoids pointless
recompiles.  This is how other property definition macros, such as
DEFINE_PROP_NETDEV(), already work.

Improves things for some of the top scorers:

    3600 qapi/qapi-types-common.h
    2800 qapi/qapi-types-sockets.h
     900 qapi/qapi-types-misc.h
    2200 qapi/qapi-types-crypto.h
    2100 qapi/qapi-types-job.h
    2100 qapi/qapi-types-block-core.h
     270 qapi/qapi-types-block.h

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20190812052359.30071-3-armbru@redhat.com>

include: Make headers more self-contained

Back in 2016, we discussed[1] rules for headers, and these were
generally liked:

1. Have a carefully curated header that's included everywhere first.  We
   got that already thanks to Peter: osdep.h.

2. Headers should normally include everything they need beyond osdep.h.
   If exceptions are needed for some reason, they must be documented in
   the header.  If all that's needed from a header is typedefs, put
   those into qemu/typedefs.h instead of including the header.

3. Cyclic inclusion is forbidden.

This patch gets include/ closer to obeying 2.

It's actually extracted from my "[RFC] Baby steps towards saner
headers" series[2], which demonstrates a possible path towards
checking 2 automatically.  It passes the RFC test there.

[1] Message-ID: <87h9g8j57d.fsf@blackfin.pond.sub.org>
    https://lists.nongnu.org/archive/html/qemu-devel/2016-03/msg03345.html
[2] Message-Id: <20190711122827.18970-1-armbru@redhat.com>
    https://lists.nongnu.org/archive/html/qemu-devel/2019-07/msg02715.html

Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Message-Id: <20190812052359.30071-2-armbru@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>

Merge remote-tracking branch 'remotes/dgilbert/tags/pull-migration-20190814a' into staging

Migration pull 2019-08-15

Marcel's vmxnet3 live migraiton fix (that breaks vmxnet3 compatibility
but makes it work)

Error description improvements from Yury.

Multifd fixes from Ivan and Juan.

A load of small cleanups from Wei.

A small cleanup from Marc-André for a future patch.

# gpg: Signature made Wed 14 Aug 2019 19:00:39 BST
# gpg:                using RSA key 45F5C71B4A0CB7FB977A9FA90516331EBC5BFDE7
# gpg: Good signature from "Dr. David Alan Gilbert (RH2) <dgilbert@redhat.com>" [full]
# Primary key fingerprint: 45F5 C71B 4A0C B7FB 977A  9FA9 0516 331E BC5B FDE7

* remotes/dgilbert/tags/pull-migration-20190814a: (33 commits)
  migration: add some multifd traces
  migration: Make global sem_sync semaphore by channel
  migration: Add traces for multifd terminate threads
  qemu-file: move qemu_{get,put}_counted_string() declarations
  migration/postcopy: use mis->bh instead of allocating a QEMUBH
  migration: rename migration_bitmap_sync_range to ramblock_sync_dirty_bitmap
  migration: update ram_counters for multifd sync packet
  migration: add speed limit for multifd migration
  migration: add qemu_file_update_transfer interface
  migration: always initialise ram_counters for a new migration
  migration: remove unused field bytes_xfer
  hmp: Remove migration capabilities from "info migrate"
  migration/postcopy: use QEMU_IS_ALIGNED to replace host_offset
  migration/postcopy: simplify calculation of run_start and fixup_start_addr
  migration/postcopy: make PostcopyDiscardState a static variable
  migration: extract ram_load_precopy
  migration: return -EINVAL directly when version_id mismatch
  migration: equation is more proper than and to check LOADVM_QUIT
  migration: just pass RAMBlock is enough
  migration: use migration_in_postcopy() to check POSTCOPY_ACTIVE
  ...

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

file-posix: Handle undetectable alignment

In some cases buf_align or request_alignment cannot be detected:

1. With Gluster, buf_align cannot be detected since the actual I/O is
   done on Gluster server, and qemu buffer alignment does not matter.
   Since we don't have alignment requirement, buf_align=1 is the best
   value.

2. With local XFS filesystem, buf_align cannot be detected if reading
   from unallocated area. In this we must align the buffer, but we don't
   know what is the correct size. Using the wrong alignment results in
   I/O error.

3. With Gluster backed by XFS, request_alignment cannot be detected if
   reading from unallocated area. In this case we need to use the
   correct alignment, and failing to do so results in I/O errors.

4. With NFS, the server does not use direct I/O, so both buf_align cannot
   be detected. In this case we don't need any alignment so we can use
   buf_align=1 and request_alignment=1.

These cases seems to work when storage sector size is 512 bytes, because
the current code starts checking align=512. If the check succeeds
because alignment cannot be detected we use 512. But this does not work
for storage with 4k sector size.

To determine if we can detect the alignment, we probe first with
align=1. If probing succeeds, maybe there are no alignment requirement
(cases 1, 4) or we are probing unallocated area (cases 2, 3). Since we
don't have any way to tell, we treat this as undetectable alignment. If
probing with align=1 fails with EINVAL, but probing with one of the
expected alignments succeeds, we know that we found a working alignment.

Practically the alignment requirements are the same for buffer
alignment, buffer length, and offset in file. So in case we cannot
detect buf_align, we can use request alignment. If we cannot detect
request alignment, we can fallback to a safe value. To use this logic,
we probe first request alignment instead of buf_align.

Here is a table showing the behaviour with current code (the value in
parenthesis is the optimal value).

Case    Sector    buf_align (opt)   request_alignment (opt)     result
======================================================================
1       512       512   (1)          512   (512)                 OK
1       4096      512   (1)          4096  (4096)                FAIL
----------------------------------------------------------------------
2       512       512   (512)        512   (512)                 OK
2       4096      512   (4096)       4096  (4096)                FAIL
----------------------------------------------------------------------
3       512       512   (1)          512   (512)                 OK
3       4096      512   (1)          512   (4096)                FAIL
----------------------------------------------------------------------
4       512       512   (1)          512   (1)                   OK
4       4096      512   (1)          512   (1)                   OK

Same cases with this change:

Case    Sector    buf_align (opt)   request_alignment (opt)     result
======================================================================
1       512       512   (1)          512   (512)                 OK
1       4096      4096  (1)          4096  (4096)                OK
----------------------------------------------------------------------
2       512       512   (512)        512   (512)                 OK
2       4096      4096  (4096)       4096  (4096)                OK
----------------------------------------------------------------------
3       512       4096  (1)          4096  (512)                 OK
3       4096      4096  (1)          4096  (4096)                OK
----------------------------------------------------------------------
4       512       4096  (1)          4096  (1)                   OK
4       4096      4096  (1)          4096  (1)                   OK

I tested that provisioning VMs and copying disks on local XFS and
Gluster with 4k bytes sector size work now, resolving bugs [1],[2].
I tested also on XFS, NFS, Gluster with 512 bytes sector size.

[1] https://bugzilla.redhat.com/1737256
[2] https://bugzilla.redhat.com/1738657

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

qemu-img convert: Deprecate using -n and -o together

bdrv_create options specified with -o have no effect when skipping image
creation with -n, so this doesn't make sense. Warn against the misuse
and deprecate the combination so we can make it a hard error later.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>

block-backend: Queue requests while drained

This fixes devices like IDE that can still start new requests from I/O
handlers in the CPU thread while the block backend is drained.

The basic assumption is that in a drain section, no new requests should
be allowed through a BlockBackend (blk_drained_begin/end don't exist,
we get drain sections only on the node level). However, there are two
special cases where requests should not be queued:

1. Block jobs: We already make sure that block jobs are paused in a
   drain section, so they won't start new requests. However, if the
   drain_begin is called on the job's BlockBackend first, it can happen
   that we deadlock because the job stays busy until it reaches a pause
   point - which it can't if its requests aren't processed any more.

   The proper solution here would be to make all requests through the
   job's filter node instead of using a BlockBackend. For now, just
   disabling request queuing on the job BlockBackend is simpler.

2. In test cases where making requests through bdrv_* would be
   cumbersome because we'd need a BdrvChild. As we already got the
   functionality to disable request queuing from 1., use it in tests,
   too, for convenience.

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

mirror: Keep mirror_top_bs drained after dropping permissions

mirror_top_bs is currently implicitly drained through its connection to
the source or the target node. However, the drain section for target_bs
ends early after moving mirror_top_bs from src to target_bs, so that
requests can already be restarted while mirror_top_bs is still present
in the chain, but has dropped all permissions and therefore runs into an
assertion failure like this:

    qemu-system-x86_64: block/io.c:1634: bdrv_co_write_req_prepare:
    Assertion `child->perm & BLK_PERM_WRITE' failed.

Keep mirror_top_bs drained until all graph changes have completed.

Cc: qemu-stable@nongnu.org
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>

block: Remove blk_pread_unthrottled()

The functionality offered by blk_pread_unthrottled() goes back to commit
498e386c584. Then, we couldn't perform I/O throttling with synchronous
requests because timers wouldn't be executed in polling loops. So the
commit automatically disabled I/O throttling as soon as a synchronous
request was issued.

However, for geometry detection during disk initialisation, we always
used (and still use) synchronous requests even if guest requests use AIO
later. Geometry detection was not wanted to disable I/O throttling, so
bdrv_pread_unthrottled() was introduced which disabled throttling only
temporarily.

All of this isn't necessary any more because we do run timers in polling
loop and even synchronous requests are now using coroutine
infrastructure internally. For this reason, commit 90c78624f already
removed the automatic disabling of I/O throttling.

It's time to get rid of the workaround for the removed code, and its
abuse of blk_root_drained_begin()/end(), as well.

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

iotests: Add test for concurrent stream/commit

We already have 030 for that in general, but this tests very specific
cases of both jobs finishing concurrently.

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

tests: Test mid-drain bdrv_replace_child_noperm()

Add a test for what happens when you call bdrv_replace_child_noperm()
for various drain situations ({old,new} child {drained,not drained}).

Most importantly, if both the old and the new child are drained, the
parent must not be undrained at any point.

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

tests: Test polling in bdrv_drop_intermediate()

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

block: Reduce (un)drains when replacing a child

Currently, bdrv_replace_child_noperm() undrains the parent until it is
completely undrained, then re-drains it after attaching the new child
node.

This is a problem with bdrv_drop_intermediate(): We want to keep the
whole subtree drained, including parents, while the operation is
under way.  bdrv_replace_child_noperm() breaks this by allowing every
parent to become unquiesced briefly, and then redraining it.

In fact, there is no reason why the parent should become unquiesced and
be allowed to submit requests to the new child node if that new node is
supposed to be kept drained.  So if anything, we have to drain the
parent before detaching the old child node.  Conversely, we have to
undrain it only after attaching the new child node.

Thus, change the whole drain algorithm here: Calculate the number of
times we have to drain/undrain the parent before replacing the child
node then drain it (if necessary), replace the child node, and then
undrain it.

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

block: Keep subtree drained in drop_intermediate

bdrv_drop_intermediate() calls BdrvChildRole.update_filename().  That
may poll, thus changing the graph, which potentially breaks the
QLIST_FOREACH_SAFE() loop.

Just keep the whole subtree drained.  This is probably the right thing
to do anyway (dropping nodes while the subtree is not drained seems
wrong).

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

block: Simplify bdrv_filter_default_perms()

The same change as commit 2b23f28639 ('block/copy-on-read: Fix
permissions for inactive node') made for the copy-on-read driver can be
made for bdrv_filter_default_perms(): Retaining the old permissions from
the BdrvChild if it is given complicates things unnecessarily when in
the end this only means that the options set in the c == NULL case (i.e.
during child creation) are retained.

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

iotests: Test migration with all kinds of filter nodes

This test case is motivated by commit 2b23f28639 ('block/copy-on-read:
Fix permissions for inactive node'). Instead of just testing
copy-on-read on migration, let's stack all sorts of filter nodes on top
of each other and try if the resulting VM can still migrate
successfully. For good measure, put everything into an iothread, because
why not?

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

iotests: Move migration helpers to iotests.py

234 implements functions that are useful for doing migration between two
VMs. Move them to iotests.py so that other test cases can use them, too.

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

iotests/118: Add -blockdev based tests

The code path for -device drive=<node-name> or without a drive=...
option for empty drives, which is supposed to be used with -blockdev
differs enough from the -drive based path with a user-owned
BlockBackend, so we want to test both paths at least for the basic tests
implemented by TestInitiallyFilled and TestInitiallyEmpty.

This would have caught the bug recently fixed for inserting read-only
nodes into a scsi-cd created without a drive=... option.

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

iotests/118: Create test classes dynamically

We're getting a ridiculous number of child classes of
TestInitiallyFilled and TestInitiallyEmpty that differ only in a few
attributes that we want to test in all combinations.

Instead of explicitly writing down every combination, let's use a loop
and create those classes dynamically.

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

iotests/118: Test media change for scsi-cd

The test covered only floppy and ide-cd. Add scsi-cd as well.

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

block/nbd: refactor nbd connection parameters

We'll need some connection parameters to be available all the time to
implement nbd reconnect. So, let's refactor them: define additional
parameters in BDRVNBDState, drop them from function parameters, drop
nbd_client_init and separate options parsing instead from nbd_open.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20190618114328.55249-6-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: Drop useless 'if' before object_unref]
Signed-off-by: Eric Blake <eblake@redhat.com>

block/nbd: add cmdline and qapi parameter reconnect-delay

Reconnect will be implemented in the following commit, so for now,
in semantics below, disconnect itself is a "serious error".

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20190618114328.55249-5-vsementsov@virtuozzo.com>
[eblake: slipped from 4.1 to 4.2]
Signed-off-by: Eric Blake <eblake@redhat.com>

block/nbd: move from quit to state

To implement reconnect we need several states for the client:
CONNECTED, QUIT and two different CONNECTING states. CONNECTING states
will be added in the following patches. This patch implements CONNECTED
and QUIT.

QUIT means, that we should close the connection and fail all current
and further requests (like old quit = true).

CONNECTED means that connection is ok, we can send requests (like old
quit = false).

For receiving loop we use a comparison of the current state with QUIT,
because reconnect will be in the same loop, so it should be looping
until the end.

Opposite, for requests we use a comparison of the current state with
CONNECTED, as we don't want to send requests in future CONNECTING
states.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20190618114328.55249-4-vsementsov@virtuozzo.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

block/nbd: use non-blocking io channel for nbd negotiation

No reason to use blocking channel for negotiation and we'll benefit in
further reconnect feature, as qio_channel reads and writes will do
qemu_coroutine_yield while waiting for io completion.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20190618114328.55249-3-vsementsov@virtuozzo.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

block/nbd: split connection_co start out of nbd_client_connect

nbd_client_connect is going to be used from connection_co, so, let's
refactor nbd_client_connect in advance, leaving io channel
configuration all in nbd_client_connect.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20190618114328.55249-2-vsementsov@virtuozzo.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

nbd: improve CMD_CACHE: use BDRV_REQ_PREFETCH

This helps to avoid extra io, allocations and memory copying.
We assume here that CMD_CACHE is always used with copy-on-read, as
otherwise it's a noop.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20190725100550.33801-4-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>

block/stream: use BDRV_REQ_PREFETCH

This helps to avoid extra io, allocations and memory copying.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20190725100550.33801-3-vsementsov@virtuozzo.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
[eblake: fix comment grammar]
Signed-off-by: Eric Blake <eblake@redhat.com>

block: implement BDRV_REQ_PREFETCH

Do effective copy-on-read request when we don't need data actually. It
will be used for block-stream and NBD_CMD_CACHE.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20190725100550.33801-2-vsementsov@virtuozzo.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
[eblake: comment grammar fix]
Signed-off-by: Eric Blake <eblake@redhat.com>

qapi: Add InetSocketAddress member keep-alive

It's needed to provide keepalive for nbd client to track server
availability.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20190725094937.32454-1-vsementsov@virtuozzo.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Acked-by: Daniel P. Berrangé <berrange@redhat.com>
[eblake: Fix error message typo]
Signed-off-by: Eric Blake <eblake@redhat.com>