Merge remote-tracking branch 'remotes/rth/tags/pull-tcg-20191028' into staging

Improvements for TARGET_PAGE_BITS_VARY
Fix for TCI ld16u_i64.
Fix for segv on icount execute from i/o memory.
Two misc cleanups.

# gpg: Signature made Mon 28 Oct 2019 14:55:08 GMT
# gpg:                using RSA key 7A481E78868B4DB6A85A05C064DF38E8AF7E215F
# gpg:                issuer "richard.henderson@linaro.org"
# gpg: Good signature from "Richard Henderson <richard.henderson@linaro.org>" [full]
# Primary key fingerprint: 7A48 1E78 868B 4DB6 A85A  05C0 64DF 38E8 AF7E 215F

* remotes/rth/tags/pull-tcg-20191028:
  translate-all: Remove tb_alloc
  translate-all: fix uninitialized tb->orig_tb
  cputlb: Fix tlb_vaddr_to_host
  exec: Cache TARGET_PAGE_MASK for TARGET_PAGE_BITS_VARY
  exec: Promote TARGET_PAGE_MASK to target_long
  exec: Restrict TARGET_PAGE_BITS_VARY assert to CONFIG_DEBUG_TCG
  exec: Use const alias for TARGET_PAGE_BITS_VARY
  configure: Detect compiler support for __attribute__((alias))
  exec: Split out variable page size support to exec-vary.c
  cpu: use ROUND_UP() to define xxx_PAGE_ALIGN
  cputlb: ensure _cmmu helper functions follow the naming standard
  tci: Add implementation for INDEX_op_ld16u_i64

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

Merge remote-tracking branch 'remotes/palmer/tags/riscv-for-master-4.2-sf2' into staging

RISC-V Patches for the 4.2 Soft Freeze, Part 2

This patch set contains a handful of small fixes for RISC-V targets that
I'd like to target for the 4.2 soft freeze.  They include:

* A fix to allow the debugger to access the state of all privilege
  modes, as opposed to just the currently executing one.
* A pair of cleanups to implement cpu_do_transaction_failed.
* Fixes to the device tree.
* The addition of various memory regions to make the sifive_u machine
  more closely match the HiFive Unleashed board.
* Fixes to our GDB interface to allow CSRs to be accessed.
* A fix to a memory leak pointed out by coverity.
* A fix that prevents PMP checks from firing incorrectly.

This passes "make chcek" and boots Open Embedded for me.

# gpg: Signature made Mon 28 Oct 2019 15:47:52 GMT
# gpg:                using RSA key 00CE76D1834960DFCE886DF8EF4CA1502CCBAB41
# gpg:                issuer "palmer@dabbelt.com"
# gpg: Good signature from "Palmer Dabbelt <palmer@dabbelt.com>" [unknown]
# gpg:                 aka "Palmer Dabbelt <palmer@sifive.com>" [unknown]
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg:          There is no indication that the signature belongs to the owner.
# Primary key fingerprint: 00CE 76D1 8349 60DF CE88  6DF8 EF4C A150 2CCB AB41

* remotes/palmer/tags/riscv-for-master-4.2-sf2:
  target/riscv: PMP violation due to wrong size parameter
  riscv/boot: Fix possible memory leak
  target/riscv: Make the priv register writable by GDB
  target/riscv: Expose "priv" register for GDB for reads
  target/riscv: Tell gdbstub the correct number of CSRs
  riscv/virt: Jump to pflash if specified
  riscv/virt: Add the PFlash CFI01 device
  riscv/virt: Manually define the machine
  riscv/sifive_u: Add the start-in-flash property
  riscv/sifive_u: Manually define the machine
  riscv/sifive_u: Add QSPI memory region
  riscv/sifive_u: Add L2-LIM cache memory
  linux-user/riscv: Propagate fault address
  riscv: sifive_u: Add ethernet0 to the aliases node
  riscv: hw: Drop "clock-frequency" property of cpu nodes
  RISC-V: Implement cpu_do_transaction_failed
  RISC-V: Handle bus errors in the page table walker
  riscv: Skip checking CSR privilege level in debugger mode

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

target/riscv: PMP violation due to wrong size parameter

riscv_cpu_tlb_fill() uses the `size` parameter to check PMP violation
using pmp_hart_has_privs().
However, if the size is unknown (=0), the ending address will be
`addr - 1` as it is `addr + size - 1` in `pmp_hart_has_privs()`.
This always causes a false PMP violation on the starting address of the
range, as `addr - 1` is not in the range.

In order to fix, we just assume that all bytes from addr to the end of
the page will be accessed if the size is unknown.

Signed-off-by: Dayeol Lee <dayeol@berkeley.edu>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

riscv/boot: Fix possible memory leak

Coverity (CID 1405786) thinks that there is a possible memory leak as
we don't guarantee that the memory allocated from riscv_find_firmware()
is freed. This is a false positive, but let's tidy up the code to fix
the warning.

Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

target/riscv: Make the priv register writable by GDB

Currently only PRV_U, PRV_S and PRV_M are supported, so this patch ensures that
the privilege mode is set to one of them. Once support for the H-extension is
added, this code will also need to properly update the virtualization status
when switching between VU/VS-modes and M-mode.

Signed-off-by: Jonathan Behrens <jonathan@fintelia.io>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

target/riscv: Expose "priv" register for GDB for reads

This patch enables a debugger to read the current privilege level via a virtual
"priv" register. When compiled with CONFIG_USER_ONLY the register is still
visible but always reports the value zero.

Signed-off-by: Jonathan Behrens <jonathan@fintelia.io>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

target/riscv: Tell gdbstub the correct number of CSRs

If the number of registers reported to the gdbstub code does not match the
number in the associated XML file, then the register numbers used by the stub
may get out of sync with a remote GDB instance.

Signed-off-by: Jonathan Behrens <jonathan@fintelia.io>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

riscv/virt: Jump to pflash if specified

If the user supplied pflash to QEMU then change the reset code to jump
to the pflash base address instead of the DRAM base address.

Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

riscv/virt: Add the PFlash CFI01 device

Add the CFI01 PFlash to the RISC-V virt board. This is the same PFlash
from the ARM Virt board and the implementation is based on the ARM Virt
board. This allows users to specify flash files from the command line.

Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

riscv/virt: Manually define the machine

Instead of using the DEFINE_MACHINE() macro to define the machine let's
do it manually. This allows us to use the machine object to create
RISCVVirtState. This is required to add children and aliases to the
machine.

This patch is no functional change.

Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

riscv/sifive_u: Add the start-in-flash property

Add a property that when set to true QEMU will jump from the ROM code to
the start of flash memory instead of DRAM which is the default
behaviour.

Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

riscv/sifive_u: Manually define the machine

Instead of using the DEFINE_MACHINE() macro to define the machine let's
do it manually. This allows us to specify machine properties.

This patch is no functional change.

Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng <bmeng.cn@gmail.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

riscv/sifive_u: Add QSPI memory region

The HiFive Unleashed uses is25wp256 SPI NOR flash. There is currently no
model of this in QEMU, so to allow boot firmware developers to use QEMU
to target the Unleashed let's add a chunk of memory to represent the QSPI0
memory mapped flash. This can be targeted using QEMU's -device loader
command line option.

In the future we can look at adding a model for the is25wp256 flash.

Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

riscv/sifive_u: Add L2-LIM cache memory

On reset only a single L2 cache way is enabled, the others are exposed
as memory that can be used by early boot firmware. This L2 region is
generally disabled using the WayEnable register at a later stage in the
boot process. To allow firmware to target QEMU and the HiFive Unleashed
let's add the L2 LIM (LooselyIntegrated Memory).

Ideally we would want to adjust the size of this chunk of memory as the
L2 Cache Controller WayEnable register is incremented. Unfortunately I
don't see a nice way to handle reducing or blocking out the L2 LIM while
still allowing it be re returned to all enabled from a reset.

Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Bin Meng <bmeng.cn@gmail.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

linux-user/riscv: Propagate fault address

The CPU loop tagged all the queued signals as QEMU_SI_KILL while it was
filling the `_sigfault` part of `siginfo`: this caused QEMU to copy the
wrong fields over to the userspace program.

Make sure the fault address recorded by the MMU is is stored in the CPU
environment structure.

In case of memory faults store the exception address into `siginfo`.

Signed-off-by: Giuseppe Musacchio <thatlemon@gmail.com>
Reviewed-by: Palmer Dabbelt <palmer@sifive.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

riscv: sifive_u: Add ethernet0 to the aliases node

U-Boot expects this alias to be in place in order to fix up the mac
address of the ethernet node.

This is to keep in sync with Linux kernel commit below:
https://patchwork.kernel.org/patch/11133033/

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

riscv: hw: Drop "clock-frequency" property of cpu nodes

The "clock-frequency" property of cpu nodes isn't required. Drop it.

This is to keep in sync with Linux kernel commit below:
https://patchwork.kernel.org/patch/11133031/

Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

RISC-V: Implement cpu_do_transaction_failed

This converts our port over from cpu_do_unassigned_access to
cpu_do_transaction_failed, as cpu_do_unassigned_access has been
deprecated.

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

RISC-V: Handle bus errors in the page table walker

We directly access physical memory while walking the page tables on
RISC-V, but while doing so we were using cpu_ld*() which does not report
bus errors.  This patch converts the page table walker over to use
address_space_ld*(), which allows bus errors to be detected.

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

riscv: Skip checking CSR privilege level in debugger mode

If we are in debugger mode, skip the CSR privilege level checking
so that we can read/write all CSRs. Otherwise we get:

(gdb) p/x $mtvec
Could not fetch register "mtvec"; remote failure reply 'E14'

when the hart is currently in S-mode.

Reported-by: Zong Li <zong.li@sifive.com>
Signed-off-by: Bin Meng <bmeng.cn@gmail.com>
Reviewed-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>

Merge remote-tracking branch 'remotes/maxreitz/tags/pull-block-2019-10-28' into staging

Block patches for softfreeze:
- iotest patches
- Improve performance of the mirror block job in write-blocking mode
- Limit memory usage for the backup block job
- Add discard and write-zeroes support to the NVMe host block driver
- Fix a bug in the mirror job
- Prevent the qcow2 driver from creating technically non-compliant qcow2
  v3 images (where there is not enough extra data for snapshot table
  entries)
- Allow callers of bdrv_truncate() (etc.) to determine whether the file
  must be resized to the exact given size or whether it is OK for block
  devices not to shrink

# gpg: Signature made Mon 28 Oct 2019 12:13:53 GMT
# gpg:                using RSA key 91BEB60A30DB3E8857D11829F407DB0061D5CF40
# gpg:                issuer "mreitz@redhat.com"
# gpg: Good signature from "Max Reitz <mreitz@redhat.com>" [full]
# Primary key fingerprint: 91BE B60A 30DB 3E88 57D1  1829 F407 DB00 61D5 CF40

* remotes/maxreitz/tags/pull-block-2019-10-28: (69 commits)
  qemu-iotests: restrict 264 to qcow2 only
  Revert "qemu-img: Check post-truncation size"
  block: Pass truncate exact=true where reasonable
  block: Let format drivers pass @exact
  block: Evaluate @exact in protocol drivers
  block: Add @exact parameter to bdrv_co_truncate()
  block: Do not truncate file node when formatting
  block/cor: Drop cor_co_truncate()
  block: Handle filter truncation like native impl.
  iotests: Test qcow2's snapshot table handling
  iotests: Add peek_file* functions
  qcow2: Fix v3 snapshot table entry compliancy
  qcow2: Repair snapshot table with too many entries
  qcow2: Fix overly long snapshot tables
  qcow2: Keep track of the snapshot table length
  qcow2: Fix broken snapshot table entries
  qcow2: Add qcow2_check_fix_snapshot_table()
  qcow2: Separate qcow2_check_read_snapshot_table()
  qcow2: Write v3-compliant snapshot list on upgrade
  qcow2: Put qcow2_upgrade() into its own function
  ...

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

Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging

virtio: features, tests

libqos update with support for virtio 1.
Packed ring support for virtio.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
# gpg: Signature made Fri 25 Oct 2019 12:47:59 BST
# gpg:                using RSA key 281F0DB8D28D5469
# gpg: Good signature from "Michael S. Tsirkin <mst@kernel.org>" [full]
# gpg:                 aka "Michael S. Tsirkin <mst@redhat.com>" [full]
# Primary key fingerprint: 0270 606B 6F3C DF3D 0B17  0970 C350 3912 AFBE 8E67
#      Subkey fingerprint: 5D09 FD08 71C8 F85B 94CA  8A0D 281F 0DB8 D28D 5469

* remotes/mst/tags/for_upstream: (25 commits)
  virtio: drop unused virtio_device_stop_ioeventfd() function
  libqos: add VIRTIO PCI 1.0 support
  libqos: extract Legacy virtio-pci.c code
  libqos: make the virtio-pci BAR index configurable
  libqos: expose common virtqueue setup/cleanup functions
  libqos: add MSI-X callbacks to QVirtioPCIDevice
  libqos: pass full QVirtQueue to set_queue_address()
  libqos: add iteration support to qpci_find_capability()
  libqos: access VIRTIO 1.0 vring in little-endian
  libqos: implement VIRTIO 1.0 FEATURES_OK step
  libqos: enforce Device Initialization order
  libqos: add missing virtio-9p feature negotiation
  tests/virtio-blk-test: set up virtqueue after feature negotiation
  virtio-scsi-test: add missing feature negotiation
  libqos: extend feature bits to 64-bit
  libqos: read QVIRTIO_MMIO_VERSION register
  tests/virtio-blk-test: read config space after feature negotiation
  virtio: add property to enable packed virtqueue
  vhost_net: enable packed ring support
  virtio: event suppression support for packed ring
  ...

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

qemu-iotests: restrict 264 to qcow2 only

264 is unprepared to run with different formats, for example luks needs
handling keys, cloop doesn't support image creation, vpc creates image
larger than requested (which breaks "Backup completed: 5242880" in test
output).

The test is here to check nbd-reconnect feature and we actually don't
need it for all formats. Let's restrict it to qcow2 only.

Reported-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-id: 20191025145023.6182-1-vsementsov@virtuozzo.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

Revert "qemu-img: Check post-truncation size"

This reverts commit 5279b30392da7a3248b320c75f20c61e3a95863c.

We no longer need this check because exact=true forces the block driver
to give the image the exact size requested by the user.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190918095144.955-9-mreitz@redhat.com
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block: Pass truncate exact=true where reasonable

This is a change in behavior, so all instances need a good
justification.  The comments added here should explain my reasoning.

qed already had a comment that suggests it always expected
bdrv_truncate()/blk_truncate() to behave as if exact=true were passed
(c743849bee7 came eight months before 55b949c8476), so it was simply
broken until now.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190918095144.955-8-mreitz@redhat.com
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
[mreitz: Changed comment in qed.c to explain why a new QED file must be
         empty, as requested and suggested by Maxim]
Signed-off-by: Max Reitz <mreitz@redhat.com>

block: Let format drivers pass @exact

When truncating a format node, the @exact parameter is generally handled
simply by virtue of the format storing the new size in the image
metadata.  Such formats do not need to pass on the parameter to their
file nodes.

There are exceptions, though:
- raw and crypto cannot store the image size, and thus must pass on
  @exact.

- When using qcow2 with an external data file, it just makes sense to
  keep its size in sync with the qcow2 virtual disk (because the
  external data file is the virtual disk).  Therefore, we should pass
  @exact when truncating it.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190918095144.955-7-mreitz@redhat.com
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block: Evaluate @exact in protocol drivers

We have two protocol drivers that return success when trying to shrink a
block device even though they cannot shrink it.  This behavior is now
only allowed with exact=false, so they should return an error with
exact=true.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190918095144.955-6-mreitz@redhat.com
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block: Add @exact parameter to bdrv_co_truncate()

We have two drivers (iscsi and file-posix) that (in some cases) return
success from their .bdrv_co_truncate() implementation if the block
device is larger than the requested offset, but cannot be shrunk.  Some
callers do not want that behavior, so this patch adds a new parameter
that they can use to turn off that behavior.

This patch just adds the parameter and lets the block/io.c and
block/block-backend.c functions pass it around.  All other callers
always pass false and none of the implementations evaluate it, so that
this patch does not change existing behavior.  Future patches take care
of that.

Suggested-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190918095144.955-5-mreitz@redhat.com
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block: Do not truncate file node when formatting

There is no reason why the format drivers need to truncate the protocol
node when formatting it.  When using the old .bdrv_co_create_ops()
interface, the file will be created with no size option anyway, which
generally gives it a size of 0.  (Exceptions are block devices, which
cannot be truncated anyway.)

When using blockdev-create, the user must have given the file node some
size anyway, so there is no reason why we should override that.

qed is an exception, it needs the file to start completely empty (as
explained by c743849bee7333c7ef256b7e12e34ed6f907064f).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190918095144.955-4-mreitz@redhat.com
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block/cor: Drop cor_co_truncate()

No other filter driver has a .bdrv_co_truncate() implementation, and
there is no need to because the general block layer code can handle it
just as well.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190918095144.955-3-mreitz@redhat.com
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block: Handle filter truncation like native impl.

Make the filter truncation (passing it through to bs->file) a
first-class citizen and handle it exactly as if it was the filter
driver's native implementation of .bdrv_co_truncate().

I do not see a reason not to, it makes the code a bit shorter, and may
be even more correct because this gets us to finish the write_req that
we prepared before (may be important to e.g. bring dirty bitmaps to the
correct size).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190918095144.955-2-mreitz@redhat.com
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Test qcow2's snapshot table handling

Add a test how our qcow2 driver handles extra data in snapshot table
entries, and how it repairs overly long snapshot tables.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-17-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Add peek_file* functions

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-16-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

qcow2: Fix v3 snapshot table entry compliancy

qcow2 v3 images require every snapshot table entry to have at least 16
bytes of extra data.  If they do not, let qemu-img check -r all fix it.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-15-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

qcow2: Repair snapshot table with too many entries

The user cannot choose which snapshots are removed.  This is fine
because we have chosen the maximum snapshot table size to be so large
(65536 entries) that it cannot be reasonably reached.  If the snapshot
table exceeds this size, the image has probably been corrupted in some
way; in this case, it is most important to just make the image usable
such that the user can copy off at least the active layer.
(Also note that the snapshots will be removed only with "-r all", so a
plain "check" or "check -r leaks" will not delete any data.)

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-14-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

qcow2: Fix overly long snapshot tables

We currently refuse to open qcow2 images with overly long snapshot
tables.  This patch makes qemu-img check -r all drop all offending
entries past what we deem acceptable.

The user cannot choose which snapshots are removed.  This is fine
because we have chosen the maximum snapshot table size to be so large
(64 MB) that it cannot be reasonably reached.  If the snapshot table
exceeds this size, the image has probably been corrupted in some way; in
this case, it is most important to just make the image usable such that
the user can copy off at least the active layer.
(Also note that the snapshots will be removed only with "-r all", so a
plain "check" or "check -r leaks" will not delete any data.)

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-13-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

qcow2: Keep track of the snapshot table length

When repairing the snapshot table, we truncate entries that have too
much extra data.  This frees up space that we do not have to count
towards the snapshot table size.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-12-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

qcow2: Fix broken snapshot table entries

The only case where we currently reject snapshot table entries is when
they have too much extra data.  Fix them with qemu-img check -r all by
counting it as a corruption, reducing their extra_data_size, and then
letting qcow2_check_fix_snapshot_table() do the rest.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-11-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

qcow2: Add qcow2_check_fix_snapshot_table()

qcow2_check_read_snapshot_table() can perform consistency checks, but it
cannot fix everything.  Specifically, it cannot allocate new clusters,
because that should wait until the refcount structures are known to be
consistent (i.e., after qcow2_check_refcounts()).  Thus, it cannot call
qcow2_write_snapshots().

Do that in qcow2_check_fix_snapshot_table(), which is called after
qcow2_check_refcounts().

Currently, there is nothing that would set result->corruptions, so this
is a no-op.  A follow-up patch will change that.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-10-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

qcow2: Separate qcow2_check_read_snapshot_table()

Reading the snapshot table can fail.  That is a problem when we want to
repair the image.

Therefore, stop reading the snapshot table in qcow2_do_open() in check
mode.  Instead, add a new function qcow2_check_read_snapshot_table()
that reads the snapshot table at a later point.  In the future, we want
to handle errors here and fix them.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-9-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

qcow2: Write v3-compliant snapshot list on upgrade

qcow2 v3 requires every snapshot table entry to have two extra data
fields: The 64-bit VM state size, and the virtual disk size.  Both are
optional for v2 images, so they may not be present.

qcow2_upgrade() therefore should update the snapshot table to ensure all
entries have these extra data fields.

Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=1727347
Reported-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-8-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

qcow2: Put qcow2_upgrade() into its own function

This does not make sense right now, but it will make sense once we need
to do more than to just update s->qcow_version.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-7-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

qcow2: Make qcow2_write_snapshots() public

Updating the snapshot list will be useful when upgrading a v2 image to
v3, so we will need to call this function in qcow2.c.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-6-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

qcow2: Keep unknown extra snapshot data

The qcow2 specification says to ignore unknown extra data fields in
snapshot table entries.  Currently, we discard it whenever we update the
image, which is a bit different from "ignore".

This patch makes the qcow2 driver keep all unknown extra data fields
when updating an image's snapshot table.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-5-mreitz@redhat.com
[mreitz: Adjusted comments as proposed by Eric]
Signed-off-by: Max Reitz <mreitz@redhat.com>

qcow2: Add Error ** to qcow2_read_snapshots()

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-4-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

qcow2: Use endof()

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-3-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

include: Move endof() up from hw/virtio/virtio.h

endof() is a useful macro, we can make use of it outside of virtio.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191011152814.14791-2-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

mirror: Do not dereference invalid pointers

mirror_exit_common() may be called twice (if it is called from
mirror_prepare() and fails, it will be called from mirror_abort()
again).

In such a case, many of the pointers in the MirrorBlockJob object will
already be freed.  This can be seen most reliably for s->target, which
is set to NULL (and then dereferenced by blk_bs()).

Cc: qemu-stable@nongnu.org
Fixes: 737efc1eda23b904fbe0e66b37715fb0e5c3e58b
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-id: 20191014153931.20699-2-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

block/nvme: add support for discard

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-id: 20190913133627.28450-3-mlevitsk@redhat.com
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block/nvme: add support for write zeros

Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-id: 20190913133627.28450-2-mlevitsk@redhat.com
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block/block-copy: increase buffered copy request

No reason to limit buffered copy to one cluster. Let's allow up to 1
MiB.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20191022111805.3432-7-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

block/block-copy: add memory limit

Currently total allocation for parallel requests to block-copy instance
is unlimited. Let's limit it to 128 MiB.

For now block-copy is used only in backup, so actually we limit total
allocation for backup job.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20191022111805.3432-6-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

util: introduce SharedResource

Introduce an API for some shared splittable resource, like memory.
It's going to be used by backup. Backup uses both read/write io and
copy_range. copy_range may consume memory implictly, so the new API is
abstract: it doesn't allocate any real memory but only hands out
tickets.

The idea is that we have some total amount of something and callers
should wait in coroutine queue if there is not enough of the resource
at the moment.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20191022111805.3432-5-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

block/block-copy: refactor copying

Merge copying code into one function block_copy_do_copy, which only
calls bdrv_ io functions and don't do any synchronization (like dirty
bitmap set/reset).

Refactor block_copy() function so that it takes full decision about
size of chunk to be copied and does all the synchronization (checking
intersecting requests, set/reset dirty bitmaps).

It will help:
 - introduce parallel processing of block_copy iterations: we need to
   calculate chunk size, start async chunk copying and go to the next
   iteration
 - simplify synchronization improvement (like memory limiting in
   further commit and reducing critical section (now we lock the whole
   requested range, when actually we need to lock only dirty region
   which we handle at the moment))

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20191022111805.3432-4-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

block/block-copy: limit copy_range_size to 16 MiB

Large copy range may imply memory allocation and large io effort, so
using 2G copy range request may be bad idea. Let's limit it to 16 MiB.
It also helps the following patch to refactor copy-with-offload
fallback to copy-with-bounce-buffer.

Note, that total memory usage of backup is still not limited, it will
be fixed in further commit.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20191022111805.3432-3-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

block/block-copy: allocate buffer in block_copy_with_bounce_buffer

Move bounce_buffer allocation block_copy_with_bounce_buffer. This
commit simplifies further work on implementing copying by larger chunks
(of different size) and further asynchronous handling of block_copy
iterations (with help of block/aio_task API).

Allocation works fast, a lot faster than disk io, so it's not a problem
that we now allocate/free bounce_buffer more times. And we anyway will
have to allocate several bounce_buffers for parallel execution of loop
iterations in future.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20191022111805.3432-2-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Drop TEST_DIR filter from _filter_nbd

Sockets should be placed into $SOCK_DIR instead of $TEST_DIR, so remove
the $TEST_DIR filter from _filter_nbd.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191017133155.5327-24-mreitz@redhat.com
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/267: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191017133155.5327-23-mreitz@redhat.com
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/240: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191017133155.5327-22-mreitz@redhat.com
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/223: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191017133155.5327-21-mreitz@redhat.com
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/222: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191017133155.5327-20-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/209: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191017133155.5327-19-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/208: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191017133155.5327-18-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/205: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191017133155.5327-17-mreitz@redhat.com
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/201: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191017133155.5327-16-mreitz@redhat.com
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/194: Create sockets in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191017133155.5327-15-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/192: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191017133155.5327-14-mreitz@redhat.com
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/183: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-id: 20191017133155.5327-13-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/182: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-id: 20191017133155.5327-12-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/181: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-id: 20191017133155.5327-11-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/147: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-id: 20191017133155.5327-10-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/143: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-id: 20191017133155.5327-9-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/140: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-id: 20191017133155.5327-8-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests/083: Create socket in $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-id: 20191017133155.5327-7-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Let common.nbd create socket in $SOCK_DIR

In addition, drop the nbd_unix_socket assignment in 241 because it does
not really do anything.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191017133155.5327-6-mreitz@redhat.com
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Filter $SOCK_DIR

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191017133155.5327-5-mreitz@redhat.com
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests.py: Add @base_dir to FilePaths etc.

Specifying this optional parameter allows creating temporary files in
other directories than the test_dir; for example in sock_dir.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20191017133155.5327-4-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests.py: Store socket files in $SOCK_DIR

iotests.py itself does not store socket files, but machine.py and
qtest.py do.  iotests.py needs to pass the respective path to them, and
they need to adhere to it.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-id: 20191017133155.5327-3-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Introduce $SOCK_DIR

Unix sockets generally have a maximum path length.  Depending on your
$TEST_DIR, it may be exceeded and then all tests that create and use
Unix sockets there may fail.

Circumvent this by adding a new scratch directory specifically for
Unix socket files.  It defaults to a temporary directory (mktemp -d)
that is completely removed after the iotests are done.

(By default, mktemp -d creates a /tmp/tmp.XXXXXXXXXX directory, which
should be short enough for our use cases.)

Use mkdir -p to create the directory (because it seems right), and do
the same for $TEST_DIR (because there is no reason for that to be
created in any different way).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20191017133155.5327-2-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

Revert "mirror: Only mirror granularity-aligned chunks"

This reverts commit 9adc1cb49af8d4e54f57980b1eed5c0a4b2dafa6.
    "mirror: Only mirror granularity-aligned chunks"

Since previous commit unaligned chunks are supported by
do_sync_target_write.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20191011090711.19940-6-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

block/mirror: support unaligned write in active mirror

Prior 9adc1cb49af8d do_sync_target_write had a bug: it reset aligned-up
region in the dirty bitmap, which means that we may not copy some bytes
and assume them copied, which actually leads to producing corrupted
target.

So 9adc1cb49af8d forced dirty bitmap granularity to be
request_alignment for mirror-top filter, so we are not working with
unaligned requests. However forcing large alignment obviously decreases
performance of unaligned requests.

This commit provides another solution for the problem: if unaligned
padding is already dirty, we can safely ignore it, as
1. It's dirty, it will be copied by mirror_iteration anyway
2. It's dirty, so skipping it now we don't increase dirtiness of the
   bitmap and therefore don't damage "synchronicity" of the
   write-blocking mirror.

If unaligned padding is not dirty, we just write it, no reason to touch
dirty bitmap if we succeed (on failure we'll set the whole region
ofcourse, but we loss "synchronicity" on failure anyway).

Note: we need to disable dirty_bitmap, otherwise we will not be able to
see in do_sync_target_write bitmap state before current operation. We
may of course check dirty bitmap before the operation in
bdrv_mirror_top_do_write and remember it, but we don't need active
dirty bitmap for write-blocking mirror anyway.

New code-path is unused until the following commit reverts
9adc1cb49af8d.

Suggested-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-id: 20191011090711.19940-5-vsementsov@virtuozzo.com
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

block/block-backend: add blk_co_pwritev_part

Add blk write function with qiov_offset parameter. It's needed for the
following commit.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20191011090711.19940-4-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

block/mirror: simplify do_sync_target_write

do_sync_target_write is called from bdrv_mirror_top_do_write after
write/discard operation, all inside active_write/active_write_settle
protecting us from mirror iteration. So the whole area is dirty for
sure, no reason to examine dirty bitmap.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20191011090711.19940-3-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

hbitmap: handle set/reset with zero length

Passing zero length to these functions leads to unpredicted results.
Zero-length set/reset may occur in active-mirror, on zero-length write
(which is unlikely, but not guaranteed to never happen).

Let's just do nothing on zero-length request.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-id: 20191011090711.19940-2-vsementsov@virtuozzo.com
Reviewed-by: Max Reitz <mreitz@redhat.com>
Cc: qemu-stable@nongnu.org
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Cache supported_formats()

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Message-id: 20190917092004.999-8-mreitz@redhat.com
Reviewed-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Test driver whitelisting in 136

null-aio may not be whitelisted.  Skip all test cases that require it.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190917092004.999-7-mreitz@redhat.com
Reviewed-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Test driver whitelisting in 093

null-aio may not be whitelisted.  Skip all test cases that require it.

(And skip the whole test if null-co is not whitelisted.)

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190917092004.999-6-mreitz@redhat.com
Reviewed-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Let skip_if_unsupported accept a function

This lets tests use skip_if_unsupported() with a potentially variable
list of required formats.

Suggested-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190917092004.999-5-mreitz@redhat.com
Reviewed-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Use case_skip() in skip_if_unsupported()

skip_if_unsupported() should use the stronger variant case_skip(),
because this allows it to be used even with setUp() (in a meaningful
way).

In the process, make it explicit what we expect the first argument of
the func_wrapper to be (namely something derived of QMPTestCase).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190917092004.999-4-mreitz@redhat.com
Reviewed-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Allow skipping test cases

case_notrun() does not actually skip the current test case.  It just
adds a "notrun" note and then returns to the caller, who manually has to
skip the test.  Generally, skipping a test case is as simple as
returning from the current function, but not always: For example, this
model does not allow skipping tests already in the setUp() function.

Thus, add a QMPTestCase.case_skip() function that invokes case_notrun()
and then self.skipTest().  To make this work, we need to filter the
information on how many test cases were skipped from the unittest
output.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Reviewed-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
Message-id: 20190917092004.999-3-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>

iotests: Prefer null-co over null-aio

We use null-co basically everywhere in the iotests.  Unless we want to
test null-aio specifically, we should use it instead (for consistency).

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-id: 20190917092004.999-2-mreitz@redhat.com
Reviewed-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>

translate-all: Remove tb_alloc

Since 2ac01d6dafab, this function does only two things: assert a
lock is held, and call tcg_tb_alloc.  It is used exactly once,
and its user has already done the assert.

Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Clement Deschamps <clement.deschamps@greensocs.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

translate-all: fix uninitialized tb->orig_tb

This fixes a segmentation fault in icount mode when executing
from an IO region.

TB is marked as CF_NOCACHE but tb->orig_tb is not initialized
(equals previous value in code_gen_buffer).

The issue happens in cpu_io_recompile() when it tries to invalidate orig_tb.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Clement Deschamps <clement.deschamps@greensocs.com>
Message-Id: <20191022140016.918371-1-clement.deschamps@greensocs.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

cputlb: Fix tlb_vaddr_to_host

Using uintptr_t instead of target_ulong meant that, for 64-bit guest
and 32-bit host, we truncated the guest address comparator and so may
not hit the tlb when we should.

Fixes: 4811e9095c0
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

exec: Cache TARGET_PAGE_MASK for TARGET_PAGE_BITS_VARY

This eliminates a set of runtime shifts.  It turns out that we
require TARGET_PAGE_MASK more often than TARGET_PAGE_SIZE, so
redefine TARGET_PAGE_SIZE based on TARGET_PAGE_MASK instead of
the other way around.

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

exec: Promote TARGET_PAGE_MASK to target_long

There are some uint64_t uses that expect TARGET_PAGE_MASK to
extend for a 32-bit, so this must continue to be a signed type.
Define based on TARGET_PAGE_BITS not TARGET_PAGE_SIZE; this
will make a following patch more clear.

This should not have a functional effect so far.

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

exec: Restrict TARGET_PAGE_BITS_VARY assert to CONFIG_DEBUG_TCG

This reduces the size of a release build by about 10k.
Noticably, within the tlb miss helpers.

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

exec: Use const alias for TARGET_PAGE_BITS_VARY

Using a variable that is declared "const" for this tells the
compiler that it may read the value once and assume that it
does not change across function calls.

For target_page_size, this means we have only one assert per
function, and one read of the variable.

This reduces the size of qemu-system-aarch64 by 8k.

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

configure: Detect compiler support for __attribute__((alias))

Such support is present almost everywhere, except for Xcode 9.
It is added in Xcode 10, but travis uses xcode9 by default,
so we should support it for a while yet.

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

exec: Split out variable page size support to exec-vary.c

The next patch will play a trick with "const" that will
confuse the compiler about the uses of target_page_bits
within exec.c.  Moving everything to a new file prevents
this confusion.

No functional change so far.

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