Adaptive polling has a big problem: It doesn't consider that an event
loop can wait for many different events that may have very different
typical latencies.
For example, think of a guest that tends to send a new I/O request soon
after the previous I/O request completes, but the storage on the host is
rather slow. In this case, getting the new request from guest quickly
means that polling is enabled, but the next thing is performing the I/O
request on the backend, which is slow and disables polling again for the
next guest request. This means that in such a scenario, polling could
help for every other event, but is only ever enabled when it can't
succeed.
In order to fix this, keep a separate AioPolledEvent for each
AioHandler. We will then know that the backend file descriptor always
has a high latency and isn't worth polling for, but we also know that
the guest is always fast and we should poll for it. This solves at least
half of the problem, we can now keep polling for those cases where it
makes sense and get the improved performance from it.
Since the event loop doesn't know which event will be next, we still do
some unnecessary polling while we're waiting for the slow disk. I made
some attempts to be more clever than just randomly growing and shrinking
the polling time, and even to let callers be explicit about when they
expect a new event, but so far this hasn't resulted in improved
performance or even caused performance regressions. For now, let's just
fix the part that is easy enough to fix, we can revisit the rest later.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250307221634.71951-6-kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Until now, FUA was always emulated with a separate flush after the write
for file-posix. The overhead of processing a second request can reduce
performance significantly for a guest disk that has disabled the write
cache, especially if the host disk is already write through, too, and
the flush isn't actually doing anything.
Advertise support for REQ_FUA in write requests and implement it for
Linux AIO and io_uring using the RWF_DSYNC flag for write requests. The
thread pool still performs a separate fdatasync() call. This can be
improved later by using the pwritev2() syscall if available.
As an example, this is how fio numbers can be improved in some scenarios
with this patch (all using virtio-blk with cache=directsync on an nvme
block device for the VM, fio with ioengine=libaio,direct=1,sync=1):
| old | with FUA support
------------------------------+---------------+-------------------
bs=4k, iodepth=1, numjobs=1 | 45.6k iops | 56.1k iops
bs=4k, iodepth=1, numjobs=16 | 183.3k iops | 236.0k iops
bs=4k, iodepth=16, numjobs=1 | 258.4k iops | 311.1k iops
However, not all scenarios are clear wins. On another slower disk I saw
little to no improvment. In fact, in two corner case scenarios, I even
observed a regression, which I however consider acceptable:
1. On slow host disks in a write through cache mode, when the guest is
using virtio-blk in a separate iothread so that polling can be
enabled, and each completion is quickly followed up with a new
request (so that polling gets it), it can happen that enabling FUA
makes things slower - the additional very fast no-op flush we used to
have gave the adaptive polling algorithm a success so that it kept
polling. Without it, we only have the slow write request, which
disables polling. This is a problem in the polling algorithm that
will be fixed later in this series.
2. With a high queue depth, it can be beneficial to have flush requests
for another reason: The optimisation in bdrv_co_flush() that flushes
only once per write generation acts as a synchronisation mechanism
that lets all requests complete at the same time. This can result in
better batching and if the disk is very fast (I only saw this with a
null_blk backend), this can make up for the overhead of the flush and
improve throughput. In theory, we could optionally introduce a
similar artificial latency in the normal completion path to achieve
the same kind of completion batching. This is not implemented in this
series.
Compatibility is not a concern for the kernel side of io_uring, it has
supported RWF_DSYNC from the start. However, io_uring_prep_writev2() is
not available before liburing 2.2.
Linux AIO started supporting it in Linux 4.13 and libaio 0.3.111. The
kernel is not a problem for any supported build platform, so it's not
necessary to add runtime checks. However, openSUSE is still stuck with
an older libaio version that would break the build.
We must detect the presence of the writev2 functions in the user space
libraries at build time to avoid build failures.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20250307221634.71951-2-kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Generic CPUs / accelerators patch queue
- Merge "qemu/clang-tsa.h" within "qemu/compiler.h"
- Various cleanups around accelerators initialization code
(better user/system split)
- Various trivial cleanups in accel/tcg/,
Guard few TCG calls with tcg_enabled()
- Explicit disassemble_info endianness
- Improve dual-endianness support for MicroBlaze
# -----BEGIN PGP SIGNATURE-----
#
# iQIzBAABCAAdFiEE+qvnXhKRciHc/Wuy4+MsLN6twN4FAmfJw08ACgkQ4+MsLN6t
# wN70whAAtfcdWtqseFfb6fvDtjflgxN51Ui0iaOECXUA18USKriGy34eBcMYMiM2
# +eKgU7+jI6JGE4+burcgWUsPpFFF951/A8+lyIbFgO5yToTDmC+qNe4XfmMAIyXq
# uf9Obr2c0Xk9luh4odb+jPAQodw/7G1fKgcCVIJNDCl/xEcPhS9eNpTaHwcVnkWI
# K6KrxWXOsqG6+evJBPWYoXtOOyt0+JcwAsJoGhprwtGm3P9+jSVXsgeGsJVyZcna
# f32JtjWL754O8XeMkOn4x6rt58VrCIMKI9xT7keDyuhTCq0Zki9RO2nMU2dSw5mN
# AfL9hxqUy0Nijnyslg3ugujDfTePsNyLdwwH7n0mnoD72ELi6WnhDsmOThuEB3Rd
# 4/kdwTJfA/rlWk/GF1tbKW7AvQZokRARtzmL3V0HmGJu57lX+2JuszEdYBkqDEP7
# GH1I10B2yANUm+C9y3X8qWOU7Ws433ebJeJoZuyfnbZ9Me+UfRmql/oS+V8ata2i
# fArEItpldUFrWRyYLkTbXrh2dgyV9yJTEir/lzOzeAZZzyabTbjf2z9qnh976GGO
# 1QnDy5QA4f54kDBUZe7JK26TZsHPch7cgqXW6f8tRlJF7A9hxGK8d2TUV/lC3/vx
# LUOlWNu03PhiruYmZEcWOsY3Jt9jRCF6lIryrnaJsqnVOVmMUMM=
# =3TRh
# -----END PGP SIGNATURE-----
# gpg: Signature made Thu 06 Mar 2025 23:46:23 HKT
# gpg: using RSA key FAABE75E12917221DCFD6BB2E3E32C2CDEADC0DE
# gpg: Good signature from "Philippe Mathieu-Daudé (F4BUG) <f4bug@amsat.org>" [full]
# Primary key fingerprint: FAAB E75E 1291 7221 DCFD 6BB2 E3E3 2C2C DEAD C0DE
* tag 'accel-cpus-20250306' of https://github.com/philmd/qemu: (54 commits)
include: Poison TARGET_PHYS_ADDR_SPACE_BITS definition
system: Open-code qemu_init_arch_modules() using target_name()
target/i386: Mark WHPX APIC region as little-endian
target/alpha: Do not mix exception flags and FPCR bits
target/riscv: Convert misa_mxl_max using GLib macros
target/riscv: Declare RISCVCPUClass::misa_mxl_max as RISCVMXL
target/xtensa: Finalize config in xtensa_register_core()
target/sparc: Constify SPARCCPUClass::cpu_def
target/i386: Constify X86CPUModel uses
disas: Remove target_words_bigendian() call in initialize_debug_target()
target/xtensa: Set disassemble_info::endian value in disas_set_info()
target/sh4: Set disassemble_info::endian value in disas_set_info()
target/riscv: Set disassemble_info::endian value in disas_set_info()
target/ppc: Set disassemble_info::endian value in disas_set_info()
target/mips: Set disassemble_info::endian value in disas_set_info()
target/microblaze: Set disassemble_info::endian value in disas_set_info
target/arm: Set disassemble_info::endian value in disas_set_info()
target: Set disassemble_info::endian value for big-endian targets
target: Set disassemble_info::endian value for little-endian targets
target/mips: Fix possible MSA int overflow
...
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Migration code wants to manage device data sending threads in one place.
QEMU has an existing thread pool implementation, however it is limited
to queuing AIO operations only and essentially has a 1:1 mapping between
the current AioContext and the AIO ThreadPool in use.
Implement generic (non-AIO) ThreadPool by essentially wrapping Glib's
GThreadPool.
This brings a few new operations on a pool:
* thread_pool_wait() operation waits until all the submitted work requests
have finished.
* thread_pool_set_max_threads() explicitly sets the maximum thread count
in the pool.
* thread_pool_adjust_max_threads_to_work() adjusts the maximum thread count
in the pool to equal the number of still waiting in queue or unfinished work.
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Link: https://lore.kernel.org/qemu-devel/b1efaebdbea7cb7068b8fb74148777012383e12b.1741124640.git.maciej.szmigiero@oracle.com
Signed-off-by: Cédric Le Goater <clg@redhat.com>
This patch introduces temperature event notification support to the UFS
emulation. It enables the emulated UFS device to generate
temperature-related events, including high and low temperature
notifications, in compliance with the UFS specification.
With this feature, UFS drivers can now handle temperature exception events
during testing and development within the emulated environment.
This enhances validation and debugging capabilities for thermal event
handling in UFS implementations.
Signed-off-by: Keoseong Park <keosung.park@samsung.com>
Reviewed-by: Jeuk Kim <jeuk20.kim@samsung.com>
Message-ID: <20250225064146epcms2p50889cb0066e2d4734f2386de325bcdf6@epcms2p5>
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
The nvme_aio_err() does not handle Verify, Compare, Copy and other misc
commands and defaults to setting the error status code to Internal
Device Error. For some of these commands, we know better, so set it
explicitly.
For the commands using the nvme_misc_cb() callback (Copy, Flush, ...),
if no status code has explicitly been set by the lower handlers, default
to Internal Device Error as previously.
Reviewed-by: Jesper Wendel Devantier <foss@defmacro.it>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
The controller incorrectly allows a zoned namespace to be attached even
if CS.CSS is configured to only support the NVM command set for I/O
queues.
Rework handling of namespace command sets in general by attaching
supported namespaces when the controller is started instead of, like
now, statically when realized.
Reviewed-by: Jesper Wendel Devantier <foss@defmacro.it>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
The specification states that,
> The controller shall set all three processing limit fields (i.e., the
> DMRL, DMRSL and DMSL fields) to non-zero values or shall clear all
> three processing limit fields to 0h.
So, set the DMRL and DMSL fields in addition to DMRSL.
Reviewed-by: Jesper Wendel Devantier <foss@defmacro.it>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
The status codes related to I/O Command Sets are in the wrong group.
Reviewed-by: Jesper Wendel Devantier <foss@defmacro.it>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Add a 'dbcs' knob to allow Doorbell Buffer Config command to be
disabled.
Reviewed-by: Jesper Wendel Devantier <foss@defmacro.it>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Virtualization Management needs sriov-related parameters. Only report
support for the command when that conditions are true.
Reviewed-by: Jesper Wendel Devantier <foss@defmacro.it>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
The Open Compute Project [1] includes a Datacenter NVMe
SSD Specification [2]. The most recent version of this specification
(as of November 2024) is 2.6.1. This specification layers on top of
the NVM Express specifications [3] to provide additional
functionality. A key part of of this is the 512 Byte OCP SMART / Health
Information Extended log page that is defined in Section 4.8.6 of the
specification.
We add a controller argument (ocp) that toggles on/off the SMART log
extended structure. To accommodate different vendor specific specifications
like OCP, we add a multiplexing function (nvme_vendor_specific_log) which
will route to the different log functions based on arguments and log ids.
We only return the OCP extended SMART log when the command is 0xC0 and ocp
has been turned on in the nvme argumants.
Though we add the whole nvme SMART log extended structure, we only populate
the physical_media_units_{read,written}, log_page_version and
log_page_uuid.
This patch is based on work done by Joel but has been modified enough
that he requested a co-developed-by tag rather than a signed-off-by.
[1]: https://www.opencompute.org/
[2]: https://www.opencompute.org/documents/datacenter-nvme-ssd-specification-v2-6-1-pdf
[3]: https://nvmexpress.org/specifications/
Signed-off-by: Stephen Bates <sbates@raithlin.com>
Co-developed-by: Joel Granados <j.granados@samsung.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Although defaulting the handshake limit to 10 seconds was a nice QoI
change to weed out intentionally slow clients, it can interfere with
integration testing done with manual NBD_OPT commands over 'nbdsh
--opt-mode'. Expose a QMP knob 'handshake-max-secs' to allow the user
to alter the timeout away from the default.
The parameter name here intentionally matches the spelling of the
constant added in commit fb1c2aaa98, and not the command-line spelling
added in the previous patch for qemu-nbd; that's because in QMP,
longer names serve as good self-documentation, and unlike the command
line, machines don't have problems generating longer spellings.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-ID: <20250203222722.650694-6-eblake@redhat.com>
[eblake: s/max-secs/max-seconds/ in QMP]
Acked-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
The general expectation is that header files should follow the same
file/path naming scheme as the corresponding source file. There are
various historical exceptions to this practice in QEMU, with one of
the most notable being the include/qapi/qmp/ directory. Most of the
headers there correspond to source files in qobject/.
This patch corrects most of that inconsistency by creating
include/qobject/ and moving the headers for qobject/ there.
This also fixes MAINTAINERS for include/qapi/qmp/dispatch.h:
scripts/get_maintainer.pl now reports "QAPI" instead of "No
maintainers found".
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Zhao Liu <zhao1.liu@intel.com>
Acked-by: Halil Pasic <pasic@linux.ibm.com> #s390x
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-ID: <20241118151235.2665921-2-armbru@redhat.com>
[Rebased]
BLOCK_OP_TYPE_DATAPLANE prevents BlockDriverState from being used by
virtio-blk/virtio-scsi with IOThread. Commit b112a65c52 ("block:
declare blockjobs and dataplane friends!") eliminated the main reason
for this blocker in 2014.
Nowadays the block layer supports I/O from multiple AioContexts, so
there is even less reason to block IOThread users. Any legitimate
reasons related to interference would probably also apply to
non-IOThread users.
The only remaining users are bdrv_op_unblock(BLOCK_OP_TYPE_DATAPLANE)
calls after bdrv_op_block_all(). If we remove BLOCK_OP_TYPE_DATAPLANE
their behavior doesn't change.
Existing bdrv_op_block_all() callers that don't explicitly unblock
BLOCK_OP_TYPE_DATAPLANE seem to do so simply because no one bothered to
rather than because it is necessary to keep BLOCK_OP_TYPE_DATAPLANE
blocked.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-ID: <20250203182529.269066-1-stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Add an option in BlockExportOptions to allow creating an export on an
inactive node without activating the node. This mode needs to be
explicitly supported by the export type (so that it doesn't perform any
operations that are forbidden for inactive nodes), so this patch alone
doesn't allow this option to be successfully used yet.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Acked-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-ID: <20250204211407.381505-13-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The system emulator tries to automatically activate and inactivate block
nodes at the right point during migration. However, there are still
cases where it's necessary that the user can do this manually.
Images are only activated on the destination VM of a migration when the
VM is actually resumed. If the VM was paused, this doesn't happen
automatically. The user may want to perform some operation on a block
device (e.g. taking a snapshot or starting a block job) without also
resuming the VM yet. This is an example where a manual command is
necessary.
Another example is VM migration when the image files are opened by an
external qemu-storage-daemon instance on each side. In this case, the
process that needs to hand over the images isn't even part of the
migration and can't know when the migration completes. Management tools
need a way to explicitly inactivate images on the source and activate
them on the destination.
This adds a new blockdev-set-active QMP command that lets the user
change the status of individual nodes (this is necessary in
qemu-storage-daemon because it could be serving multiple VMs and only
one of them migrates at a time). For convenience, operating on all
devices (like QEMU does automatically during migration) is offered as an
option, too, and can be used in the context of single VM.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Acked-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-ID: <20250204211407.381505-9-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
In QEMU, nodes are automatically created inactive while expecting an
incoming migration (i.e. RUN_STATE_INMIGRATE). In qemu-storage-daemon,
the notion of runstates doesn't exist. It also wouldn't necessarily make
sense to introduce it because a single daemon can serve multiple VMs
that can be in different states.
Therefore, allow the user to explicitly open images as inactive with a
new option. The default is as before: Nodes are usually active, except
when created during RUN_STATE_INMIGRATE.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Acked-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-ID: <20250204211407.381505-8-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Kevin Wolf
Stefan Hajnoczi
Philippe Mathieu-Daudé
Maciej S. SzmigieroandCédric Le Goater
Keoseong ParkandPhilippe Mathieu-Daudé
Klaus Jensen
Stephen BatesandKlaus Jensen
Eric Blake
Daniel P. BerrangéandMarkus Armbruster
Matt Borgerson