This QEMU_VM_COMMAND sub-command and its switchover_start SaveVMHandler is
used to mark the switchover point in main migration stream.
It can be used to inform the destination that all pre-switchover main
migration stream data has been sent/received so it can start to process
post-switchover data that it might have received via other migration
channels like the multifd ones.
Add also the relevant MigrationState bit stream compatibility property and
its hw_compat entry.
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Zhang Chen <zhangckid@gmail.com> # for the COLO part
Signed-off-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Link: https://lore.kernel.org/qemu-devel/311be6da85fc7e49a7598684d80aa631778dcbce.1741124640.git.maciej.szmigiero@oracle.com
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Remove all instances of _fd_ from the migration generic code. These
functions have grown over time and the _fd_ part is now just
confusing.
migration_fd_error() -> migration_error() makes it a little
vague. Since it's only used for migration_connect() failures, change
it to migration_connect_set_error().
Reviewed-by: Peter Xu <peterx@redhat.com>
Message-ID: <20250213175927.19642-4-farosas@suse.de>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
There's no need for two separate functions and this _fd_ is a historic
artifact that makes little sense nowadays.
Reviewed-by: Peter Xu <peterx@redhat.com>
Message-ID: <20250213175927.19642-3-farosas@suse.de>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Postcopy invokes qemu_savevm_state_complete_precopy() twice, that means
it'll invoke COMPLETE notify twice.. also twice the tracepoints that
marking precopy complete.
Move that notification (along with the tracepoint) out to the caller, so
that postcopy will only notify once right at the start of switchover phase
from precopy. When at it, rename it to suite the file now it locates.
For precopy, there should have no functional change except the tracepoint
has a name change.
For the other two users of qemu_savevm_state_complete_precopy(), namely:
qemu_savevm_state() and qemu_savevm_live_state(): the notifier shouldn't
matter because they're not precopy at all. Now in these two contexts (aka,
"savevm", and "colo") sometimes the precopy notifiers will still be
invoked, but that's outside the scope of this patch.
Signed-off-by: Peter Xu <peterx@redhat.com>
Tested-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Juraj Marcin <jmarcin@redhat.com>
Link: https://lore.kernel.org/r/20250114230746.3268797-12-peterx@redhat.com
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Add functions to create a QEMUFile based on a unix URI, for saving or
loading, for use by cpr-transfer mode to preserve CPR state.
Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Link: https://lore.kernel.org/r/1736967650-129648-16-git-send-email-steven.sistare@oracle.com
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Define functions to put/get file descriptors to/from a QEMUFile, for qio
channels that support SCM_RIGHTS. Maintain ordering such that
put(A), put(fd), put(B)
followed by
get(A), get(fd), get(B)
always succeeds. Other get orderings may succeed but are not guaranteed.
Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Link: https://lore.kernel.org/r/1736967650-129648-14-git-send-email-steven.sistare@oracle.com
Signed-off-by: Fabiano Rosas <farosas@suse.de>
CPR must save state that is needed after QEMU is restarted, when devices
are realized. Thus the extra state cannot be saved in the migration
channel, as objects must already exist before that channel can be loaded.
Instead, define auxilliary state structures and vmstate descriptions, not
associated with any registered object, and serialize the aux state to a
cpr-specific channel in cpr_state_save. Deserialize in cpr_state_load
after QEMU restarts, before devices are realized.
Provide accessors for clients to register file descriptors for saving.
The mechanism for passing the fd's to the new process will be specific
to each migration mode, and added in subsequent patches.
Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Peter Xu <peterx@redhat.com>
Link: https://lore.kernel.org/r/1736967650-129648-8-git-send-email-steven.sistare@oracle.com
Signed-off-by: Fabiano Rosas <farosas@suse.de>
This patch proposes a flag to maintain disk activation status globally. It
mostly rewrites disk activation mgmt for QEMU, including COLO and QMP
command xen_save_devices_state.
Backgrounds
===========
We have two problems on disk activations, one resolved, one not.
Problem 1: disk activation recover (for switchover interruptions)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
When migration is either cancelled or failed during switchover, especially
when after the disks are inactivated, QEMU needs to remember re-activate
the disks again before vm starts.
It used to be done separately in two paths: one in qmp_migrate_cancel(),
the other one in the failure path of migration_completion().
It used to be fixed in different commits, all over the places in QEMU. So
these are the relevant changes I saw, I'm not sure if it's complete list:
- In 2016, commit fe904ea824 ("migration: regain control of images when
migration fails to complete")
- In 2017, commit 1d2acc3162 ("migration: re-active images while migration
been canceled after inactive them")
- In 2023, commit 6dab4c93ec ("migration: Attempt disk reactivation in
more failure scenarios")
Now since we have a slightly better picture maybe we can unify the
reactivation in a single path.
One side benefit of doing so is, we can move the disk operation outside QMP
command "migrate_cancel". It's possible that in the future we may want to
make "migrate_cancel" be OOB-compatible, while that requires the command
doesn't need BQL in the first place. This will already do that and make
migrate_cancel command lightweight.
Problem 2: disk invalidation on top of invalidated disks
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
This is an unresolved bug for current QEMU. Link in "Resolves:" at the
end. It turns out besides the src switchover phase (problem 1 above), QEMU
also needs to remember block activation on destination.
Consider two continuous migration in a row, where the VM was always paused.
In that scenario, the disks are not activated even until migration
completed in the 1st round. When the 2nd round starts, if QEMU doesn't
know the status of the disks, it needs to try inactivate the disk again.
Here the issue is the block layer API bdrv_inactivate_all() will crash a
QEMU if invoked on already inactive disks for the 2nd migration. For
detail, see the bug link at the end.
Implementation
==============
This patch proposes to maintain disk activation with a global flag, so we
know:
- If we used to inactivate disks for migration, but migration got
cancelled, or failed, QEMU will know it should reactivate the disks.
- On incoming side, if the disks are never activated but then another
migration is triggered, QEMU should be able to tell that inactivate is
not needed for the 2nd migration.
We used to have disk_inactive, but it only solves the 1st issue, not the
2nd. Also, it's done in completely separate paths so it's extremely hard
to follow either how the flag changes, or the duration that the flag is
valid, and when we will reactivate the disks.
Convert the existing disk_inactive flag into that global flag (also invert
its naming), and maintain the disk activation status for the whole
lifecycle of qemu. That includes the incoming QEMU.
Put both of the error cases of source migration (failure, cancelled)
together into migration_iteration_finish(), which will be invoked for
either of the scenario. So from that part QEMU should behave the same as
before. However with such global maintenance on disk activation status, we
not only cleanup quite a few temporary paths that we try to maintain the
disk activation status (e.g. in postcopy code), meanwhile it fixes the
crash for problem 2 in one shot.
For freshly started QEMU, the flag is initialized to TRUE showing that the
QEMU owns the disks by default.
For incoming migrated QEMU, the flag will be initialized to FALSE once and
for all showing that the dest QEMU doesn't own the disks until switchover.
That is guaranteed by the "once" variable.
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2395
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Message-Id: <20241206230838.1111496-7-peterx@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
When VM is configured with huge memory, the current throttle logic
doesn't look like to scale, because migration_trigger_throttle()
is only called for each iteration, so it won't be invoked for a long
time if one iteration can take a long time.
The periodic dirty sync aims to fix the above issue by synchronizing
the ramblock from remote dirty bitmap and, when necessary, triggering
the CPU throttle multiple times during a long iteration.
This is a trade-off between synchronization overhead and CPU throttle
impact.
Signed-off-by: Hyman Huang <yong.huang@smartx.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/f61f1b3653f2acf026901103e1c73d157d38b08f.1729146786.git.yong.huang@smartx.com
[peterx: make prev_cnt global, and reset for each migration]
Signed-off-by: Peter Xu <peterx@redhat.com>
Move cpu-throttle.c from system to migration since it's
only used for migration; this makes us avoid exporting the
util functions and variables in misc.h but export them in
migration.h when implementing the periodic ramblock dirty
sync feature in the upcoming commits.
Since CPU throttle timers are only used in migration, move
their registry to migration_object_init.
Signed-off-by: Hyman Huang <yong.huang@smartx.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/c1b3efaa0cb49e03d422e9da97bdb65cc3d234d1.1729146786.git.yong.huang@smartx.com
[peterx: Fix build on MacOS on cocoa.m, not move cpu-throttle.h yet]
[peterx: Fix subject spelling, per pm215]
Signed-off-by: Peter Xu <peterx@redhat.com>
While we cannot yet disentangle the multifd packet from page data, we
can make the code a bit cleaner by setting the page-related fields in
a separate function.
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
The total_normal_pages and total_zero_pages elements are used only for
the end tracepoints of the multifd threads. These are not super useful
since they record per-channel numbers and are just the sum of all the
pages that are transmitted per-packet, for which we already have
tracepoints. Remove the totals from the tracing.
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
1. Add zero_pages field in MultiFDPacket_t.
2. Implements the zero page detection and handling on the multifd
threads for non-compression, zlib and zstd compression backends.
3. Added a new value 'multifd' in ZeroPageDetection enumeration.
4. Adds zero page counters and updates multifd send/receive tracing
format to track the newly added counters.
Signed-off-by: Hao Xiang <hao.xiang@bytedance.com>
Acked-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240311180015.3359271-5-hao.xiang@linux.dev
Signed-off-by: Peter Xu <peterx@redhat.com>
Some minor cleanups and documentation for multifd_recv_sync_main.
Use thread_count as done in other parts of the code. Remove p->id from
the multifd_recv_state sync, since that is global and not tied to a
channel. Add documentation for the sync steps.
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240229153017.2221-2-farosas@suse.de
Signed-off-by: Peter Xu <peterx@redhat.com>
Split multifd_send_terminate_threads() into two functions:
- multifd_send_set_error(): used when an error happened on the sender
side, set error and quit state only
- multifd_send_terminate_threads(): used only by the main thread to kick
all multifd send threads out of sleep, for the last recycling.
Use multifd_send_set_error() in the three old call sites where only the
error will be set.
Use multifd_send_terminate_threads() in the last one where the main thread
will kick the multifd threads at last in multifd_save_cleanup().
Both helpers will need to set quitting=1.
Suggested-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240202102857.110210-16-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>
rp_state.error was a boolean used to show error happened in return path
thread. That's not only duplicating error reporting (migrate_set_error),
but also not good enough in that we only do error_report() and set it to
true, we never can keep a history of the exact error and show it in
query-migrate.
To make this better, a few things done:
- Use error_setg() rather than error_report() across the whole lifecycle
of return path thread, keeping the error in an Error*.
- With above, no need to have mark_source_rp_bad(), remove it, alongside
with rp_state.error itself.
- Use migrate_set_error() to apply that captured error to the global
migration object when error occured in this thread.
- Do the same when detected qemufile error in source return path
We need to re-export qemu_file_get_error_obj() to do the last one.
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Message-ID: <20231017202633.296756-2-peterx@redhat.com>
This patch is inspired by Joao Martin's patch here:
https://lore.kernel.org/r/20230926161841.98464-1-joao.m.martins@oracle.com
Add tracepoints for major downtime checkpoints on both src and dst. They
share the same tracepoint with a string showing its stage.
Besides the checkpoints in the previous patch, this patch also added
destination checkpoints.
On src, we have these checkpoints added:
- src-downtime-start: right before vm stops on src
- src-vm-stopped: after vm is fully stopped
- src-iterable-saved: after all iterables saved (END sections)
- src-non-iterable-saved: after all non-iterable saved (FULL sections)
- src-downtime-stop: migration fully completed
On dst, we have these checkpoints added:
- dst-precopy-loadvm-completes: after loadvm all done for precopy
- dst-precopy-bh-*: record BH steps to resume VM for precopy
- dst-postcopy-bh-*: record BH steps to resume VM for postcopy
On dst side, we don't have a good way to trace total time consumed by
iterable or non-iterable for now. We can mark it by 1st time receiving a
FULL / END section, but rather than that let's just rely on the other
tracepoints added for vmstates to back up the information.
With this patch, one can enable "vmstate_downtime*" tracepoints and it'll
enable all tracepoints for downtime measurements necessary.
Drop loadvm_postcopy_handle_run_bh() tracepoint alongside, because they
service the same purpose, which was only for postcopy. We then have
unified prefix for all downtime relevant tracepoints.
Co-developed-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Message-ID: <20231030163346.765724-6-peterx@redhat.com>
We have a bunch of savevm_section* tracepoints, they're good to analyze
migration stream, but not always suitable if someone would like to analyze
the migration downtime. Two major problems:
- savevm_section* tracepoints are dumping all sections, we only care
about the sections that contribute to the downtime
- They don't have an identifier to show the type of sections, so no way
to filter downtime information either easily.
We can add type into the tracepoints, but instead of doing so, this patch
kept them untouched, instead of adding a bunch of downtime specific
tracepoints, so one can enable "vmstate_downtime*" tracepoints and get a
full picture of how the downtime is distributed across iterative and
non-iterative vmstate save/load.
Note that here both save() and load() need to be traced, because both of
them may contribute to the downtime. The contribution is not a simple "add
them together", though: consider when the src is doing a save() of device1
while the dest can be load()ing for device2, so they can happen
concurrently.
Tracking both sides make sense because device load() and save() can be
imbalanced, one device can save() super fast, but load() super slow, vice
versa. We can't figure that out without tracing both.
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Message-ID: <20231030163346.765724-4-peterx@redhat.com>
The function is currently called from two sites, one always gives it a
NULL Error and the other always gives it a non-NULL Error.
In the non-NULL case, all it does it trace the error and return. One
of the callers already have tracing, add a tracepoint to the other and
stop passing the error into the function.
Cc: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Message-ID: <20231012134343.23757-4-farosas@suse.de>
Functions are long enough even without this.
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Li Zhijian <lizhijian@fujitsu.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Message-ID: <20231011203527.9061-10-quintela@redhat.com>
Migration bandwidth is a very important value to live migration. It's
because it's one of the major factors that we'll make decision on when to
switchover to destination in a precopy process.
This value is currently estimated by QEMU during the whole live migration
process by monitoring how fast we were sending the data. This can be the
most accurate bandwidth if in the ideal world, where we're always feeding
unlimited data to the migration channel, and then it'll be limited to the
bandwidth that is available.
However in reality it may be very different, e.g., over a 10Gbps network we
can see query-migrate showing migration bandwidth of only a few tens of
MB/s just because there are plenty of other things the migration thread
might be doing. For example, the migration thread can be busy scanning
zero pages, or it can be fetching dirty bitmap from other external dirty
sources (like vhost or KVM). It means we may not be pushing data as much
as possible to migration channel, so the bandwidth estimated from "how many
data we sent in the channel" can be dramatically inaccurate sometimes.
With that, the decision to switchover will be affected, by assuming that we
may not be able to switchover at all with such a low bandwidth, but in
reality we can.
The migration may not even converge at all with the downtime specified,
with that wrong estimation of bandwidth, keeping iterations forever with a
low estimation of bandwidth.
The issue is QEMU itself may not be able to avoid those uncertainties on
measuing the real "available migration bandwidth". At least not something
I can think of so far.
One way to fix this is when the user is fully aware of the available
bandwidth, then we can allow the user to help providing an accurate value.
For example, if the user has a dedicated channel of 10Gbps for migration
for this specific VM, the user can specify this bandwidth so QEMU can
always do the calculation based on this fact, trusting the user as long as
specified. It may not be the exact bandwidth when switching over (in which
case qemu will push migration data as fast as possible), but much better
than QEMU trying to wildly guess, especially when very wrong.
A new parameter "avail-switchover-bandwidth" is introduced just for this.
So when the user specified this parameter, instead of trusting the
estimated value from QEMU itself (based on the QEMUFile send speed), it
trusts the user more by using this value to decide when to switchover,
assuming that we'll have such bandwidth available then.
Note that specifying this value will not throttle the bandwidth for
switchover yet, so QEMU will always use the full bandwidth possible for
sending switchover data, assuming that should always be the most important
way to use the network at that time.
This can resolve issues like "unconvergence migration" which is caused by
hilarious low "migration bandwidth" detected for whatever reason.
Reported-by: Zhiyi Guo <zhguo@redhat.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Message-ID: <20231010221922.40638-1-peterx@redhat.com>
qemu_rdma_dump_id() dumps RDMA device details to stdout.
rdma_start_outgoing_migration() calls it via qemu_rdma_source_init()
and qemu_rdma_resolve_host() to show source device details.
rdma_start_incoming_migration() arranges its call via
rdma_accept_incoming_migration() and qemu_rdma_accept() to show
destination device details.
Two issues:
1. rdma_start_outgoing_migration() can run in HMP context. The
information should arguably go the monitor, not stdout.
2. ibv_query_port() failure is reported as error. Its callers remain
unaware of this failure (qemu_rdma_dump_id() can't fail), so
reporting this to the user as an error is problematic.
Fixable, but the device detail dump is noise, except when
troubleshooting. Tracing is a better fit. Similar function
qemu_rdma_dump_id() was converted to tracing in commit
733252deb8 (Tracify migration/rdma.c).
Convert qemu_rdma_dump_id(), too.
While there, touch up qemu_rdma_dump_gid()'s outdated comment.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Message-ID: <20230928132019.2544702-54-armbru@redhat.com>
We use int instead of uint64_t in a few places. Change them to
uint64_t.
This cleans up a comparison of signed qemu_rdma_block_for_wrid()
parameter @wrid_requested with unsigned @wr_id. Harmless, because the
actual arguments are non-negative enumeration constants.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Li Zhijian <lizhijian@fujitsu.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Message-ID: <20230928132019.2544702-6-armbru@redhat.com>
wrid_desc[] uses 4001 pointers to map four integer values to strings.
print_wrid() accesses wrid_desc[] out of bounds when passed a negative
argument. It returns null for values 2..1999 and 2001..3999.
qemu_rdma_poll() and qemu_rdma_block_for_wrid() print wrid_desc[wr_id]
and passes print_wrid(wr_id) to tracepoints. Could conceivably crash
trying to format a null string. I believe access out of bounds is not
possible.
Not worth cleaning up. Dumb down to show just numeric wr_id.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Li Zhijian <lizhijian@fujitsu.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Message-ID: <20230928132019.2544702-5-armbru@redhat.com>
For both save/load we actually share the logic on deciding whether a field
should exist. Merge the checks into a helper and use it for both save and
load. When doing so, add documentations and reformat the code to make it
much easier to read.
The real benefit here (besides code cleanups) is we add a trace-point for
this; this is a known spot where we can easily break migration
compatibilities between binaries, and this trace point will be critical for
us to identify such issues.
For example, this will be handy when debugging things like:
https://gitlab.com/qemu-project/qemu/-/issues/932
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Message-ID: <20230906204722.514474-1-peterx@redhat.com>
Extend the migration URI to support file:<filename>. This can be used for
any migration scenario that does not require a reverse path. It can be
used as an alternative to 'exec:cat > file' in minimized containers that
do not contain /bin/sh, and it is easier to use than the fd:<fdname> URI.
It can be used in HMP commands, and as a qemu command-line parameter.
For best performance, guest ram should be shared and x-ignore-shared
should be true, so guest pages are not written to the file, in which case
the guest may remain running. If ram is not so configured, then the user
is advised to stop the guest first. Otherwise, a busy guest may re-dirty
the same page, causing it to be appended to the file multiple times,
and the file may grow unboundedly. That issue is being addressed in the
"fixed-ram" patch series.
Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Tested-by: Michael Galaxy <mgalaxy@akamai.com>
Reviewed-by: Michael Galaxy <mgalaxy@akamai.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Message-ID: <1694182931-61390-2-git-send-email-steven.sistare@oracle.com>
Just create a variable for it, the same way that multifd does. This
way it is safe to use for other thread, etc, etc.
Reviewed-by: Leonardo Bras <leobras@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Message-Id: <20230515195709.63843-11-quintela@redhat.com>
Implement dirty-limit convergence algo for live migration,
which is kind of like auto-converge algo but using dirty-limit
instead of cpu throttle to make migration convergent.
Enable dirty page limit if dirty_rate_high_cnt greater than 2
when dirty-limit capability enabled, Disable dirty-limit if
migration be canceled.
Note that "set_vcpu_dirty_limit", "cancel_vcpu_dirty_limit"
commands are not allowed during dirty-limit live migration.
Signed-off-by: Hyman Huang(黄勇) <yong.huang@smartx.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Message-ID: <168733225273.5845.15871826788879741674-7@git.sr.ht>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Implement switchover ack logic. This prevents the source from stopping
the VM and completing the migration until an ACK is received from the
destination that it's OK to do so.
To achieve this, a new SaveVMHandlers handler switchover_ack_needed()
and a new return path message MIG_RP_MSG_SWITCHOVER_ACK are added.
The switchover_ack_needed() handler is called during migration setup in
the destination to check if switchover ack is used by the migrated
device.
When switchover is approved by all migrated devices in the destination
that support this capability, the MIG_RP_MSG_SWITCHOVER_ACK return path
message is sent to the source to notify it that it's OK to do
switchover.
Signed-off-by: Avihai Horon <avihaih@nvidia.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Tested-by: YangHang Liu <yanghliu@redhat.com>
Acked-by: Alex Williamson <alex.williamson@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
This significantly reduces overhead of dirty page
rate calculation in sampling mode.
Tested using 32GiB VM on E5-2690 CPU.
With CRC32:
total_pages=8388608 sampled_pages=16384 millis=71
With xxHash:
total_pages=8388608 sampled_pages=16384 millis=14
Signed-off-by: Andrei Gudkov <gudkov.andrei@huawei.com>
Message-Id: <cd115a89fc81d5f2eeb4ea7d57a98b84f794f340.1682598010.git.gudkov.andrei@huawei.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Nothing assigns to it after previous commit.
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Finish the conversion from commit fe80c0241d
("migration: using trace_ to replace DPRINTF").
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Until previous commit, save_live_pending() was used for ram. Now with
the split into state_pending_estimate() and state_pending_exact() it
is not needed anymore, so remove them.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
We split the function into to:
- state_pending_estimate: We estimate the remaining state size without
stopping the machine.
- state pending_exact: We calculate the exact amount of remaining
state.
The only "device" that implements different functions for _estimate()
and _exact() is ram.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
This reverts commit cfd66f30fb.
The simplification of unqueue_page() introduced a bug that sometimes
breaks migration on s390x hosts.
The problem is not fully understood yet, but since we are already in
the freeze for QEMU 7.1 and we need something working there, let's
revert this patch for the upcoming release. The optimization can be
redone later again in a proper way if necessary.
Buglink: https://bugzilla.redhat.com/show_bug.cgi?id=2099934
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20220802061949.331576-1-thuth@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
This patch is based on the async preempt channel creation. It continues
wiring up the new channel with TLS handshake to destionation when enabled.
Note that only the src QEMU needs such operation; the dest QEMU does not
need any change for TLS support due to the fact that all channels are
established synchronously there, so all the TLS magic is already properly
handled by migration_tls_channel_process_incoming().
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20220707185518.27529-1-peterx@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
To allow postcopy recovery, the ram fast load (preempt-only) dest QEMU thread
needs similar handling on fault tolerance. When ram_load_postcopy() fails,
instead of stopping the thread it halts with a semaphore, preparing to be
kicked again when recovery is detected.
A mutex is introduced to make sure there's no concurrent operation upon the
socket. To make it simple, the fast ram load thread will take the mutex during
its whole procedure, and only release it if it's paused. The fast-path socket
will be properly released by the main loading thread safely when there's
network failures during postcopy with that mutex held.
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20220707185506.27257-1-peterx@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
This patch enables postcopy-preempt feature.
It contains two major changes to the migration logic:
(1) Postcopy requests are now sent via a different socket from precopy
background migration stream, so as to be isolated from very high page
request delays.
(2) For huge page enabled hosts: when there's postcopy requests, they can now
intercept a partial sending of huge host pages on src QEMU.
After this patch, we'll live migrate a VM with two channels for postcopy: (1)
PRECOPY channel, which is the default channel that transfers background pages;
and (2) POSTCOPY channel, which only transfers requested pages.
There's no strict rule of which channel to use, e.g., if a requested page is
already being transferred on precopy channel, then we will keep using the same
precopy channel to transfer the page even if it's explicitly requested. In 99%
of the cases we'll prioritize the channels so we send requested page via the
postcopy channel as long as possible.
On the source QEMU, when we found a postcopy request, we'll interrupt the
PRECOPY channel sending process and quickly switch to the POSTCOPY channel.
After we serviced all the high priority postcopy pages, we'll switch back to
PRECOPY channel so that we'll continue to send the interrupted huge page again.
There's no new thread introduced on src QEMU.
On the destination QEMU, one new thread is introduced to receive page data from
the postcopy specific socket (done in the preparation patch).
This patch has a side effect: after sending postcopy pages, previously we'll
assume the guest will access follow up pages so we'll keep sending from there.
Now it's changed. Instead of going on with a postcopy requested page, we'll go
back and continue sending the precopy huge page (which can be intercepted by a
postcopy request so the huge page can be sent partially before).
Whether that's a problem is debatable, because "assuming the guest will
continue to access the next page" may not really suite when huge pages are
used, especially if the huge page is large (e.g. 1GB pages). So that locality
hint is much meaningless if huge pages are used.
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20220707185504.27203-1-peterx@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Create a new socket for postcopy to be prepared to send postcopy requested
pages via this specific channel, so as to not get blocked by precopy pages.
A new thread is also created on dest qemu to receive data from this new channel
based on the ram_load_postcopy() routine.
The ram_load_postcopy(POSTCOPY) branch and the thread has not started to
function, and that'll be done in follow up patches.
Cleanup the new sockets on both src/dst QEMUs, meanwhile look after the new
thread too to make sure it'll be recycled properly.
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20220707185502.27149-1-peterx@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
dgilbert: With Peter's fix to quieten compiler warning on
start_migration
Remove the old two tracepoints and they're even near each other:
trace_loadvm_postcopy_handle_run_cpu_sync()
trace_loadvm_postcopy_handle_run_vmstart()
Add trace_loadvm_postcopy_handle_run_bh() with a finer granule trace.
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20220301083925.33483-4-peterx@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
The enablement of postcopy listening has a few steps, add a few tracepoints to
be there ready for some basic measurements for them.
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20220301083925.33483-3-peterx@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
It'll be easier to read the name rather than index of sub-cmd when debugging.
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20220301083925.33483-2-peterx@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
This patch simplifies unqueue_page() on both sides of it (itself, and caller).
Firstly, due to the fact that right after unqueue_page() returned true, we'll
definitely send a huge page (see ram_save_huge_page() call - it will _never_
exit before finish sending that huge page), so unqueue_page() does not need to
jump in small page size if huge page is enabled on the ramblock. IOW, it's
destined that only the 1st 4K page will be valid, when unqueue the 2nd+ time
we'll notice the whole huge page has already been sent anyway. Switching to
operating on huge page reduces a lot of the loops of redundant unqueue_page().
Meanwhile, drop the dirty check. It's not helpful to call test_bit() every
time to jump over clean pages, as ram_save_host_page() has already done so,
while in a faster way (see commit ba1b7c812c ("migration/ram: Optimize
ram_save_host_page()", 2021-05-13)). So that's not necessary too.
Drop the two tracepoints along the way - based on above analysis it's very
possible that no one is really using it..
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
We are only sending normal pages through multifd channels.
Later on this series, we are going to also send zero pages.
We are going to detect if a page is zero or non zero in the multifd
channel thread, not on the main thread.
So we receive an array of pages page->offset[N]
And we will end with:
p->normal[N - zero_pages]
p->zero[zero_pages].
In this patch, we just copy all the pages in offset to normal.
for (i = 0; i < pages->num; i++) {
p->narmal[p->normal_num] = pages->offset[i];
p->normal_num++:
}
Later in the series this becomes:
for (i = 0; i < pages->num; i++) {
if (buffer_is_zero(page->offset[i])) {
p->zerol[p->zero_num] = pages->offset[i];
p->zero_num++:
} else {
p->narmal[p->normal_num] = pages->offset[i];
p->normal_num++:
}
}
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
---
Improving comment (dave)
Renaming num_normal_pages to total_normal_pages (peter)
So printing it as %d is wrong. Notice that for the channel id, that
is an uint8_t, but I changed it anyways for consistency.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
