My fix (84e7b80a) replaced the last_sent_block update that I'd
removed earlier; however it was too aggressive in the xbzrle case.
save_xbzrle_page might return '0' to mean that the page didn't
need sending since it was the same as the last sent version;
in this case we can't update 'last_sent_block' since we didn't
actually send it.
Symptom: 'Illegal RAM offset 1018000' as we try and send a page
to the wrong RAMBlock; potentially that could be a data
corruption if you were really unlucky.
Fixes: 84e7b80a05
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Message-id: 1449765106-6528-1-git-send-email-dgilbert@redhat.com
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
In a82d593b61 I accidentally removed the setting of
last_sent_block, put it back.
Symptoms:
Multithreaded compression only uses one thread.
Migration is a bit less efficient since it won't use 'cont' flags.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Fixes: a82d593b61
Signed-off-by: Juan Quintela <quintela@redhat.com>
Where the target page size is different from the host page
we special case it, but I messed up on the zero case check.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Prior to the start of postcopy, ensure that everything that will
be transferred later is a whole host-page in size.
This is accomplished by discarding partially transferred host pages
and marking any that are partially dirty as fully dirty.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Once we're in postcopy the source processors are stopped and memory
shouldn't change any more, so there's no need to look at the dirty
map.
There are two notes to this:
1) If we do resync and a page had changed then the page would get
sent again, which the destination wouldn't allow (since it might
have also modified the page)
2) Before disabling this I'd seen very rare cases where a page had been
marked dirtied although the memory contents are apparently identical
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Ensure that target pages received within a host page are in order.
This shouldn't trigger, but in the cases where the sender goes
wrong and sends stuff out of order it produces a corruption that's
really nasty to debug.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
In postcopy, the destination guest is running at the same time
as it's receiving pages; as we receive new pages we must put
them into the guests address space atomically to avoid a running
CPU accessing a partially written page.
Use the helpers in postcopy-ram.c to map these pages.
qemu_get_buffer_in_place is used to avoid a copy out of qemu_file
in the case that postcopy is going to do a copy anyway.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
When transmitting RAM pages, consume pages that have been queued by
MIG_RPCOMM_REQPAGE commands and send them ahead of normal page scanning.
Note:
a) After a queued page the linear walk carries on from after the
unqueued page; there is a reasonable chance that the destination
was about to ask for other closeby pages anyway.
b) We have to be careful of any assumptions that the page walking
code makes, in particular it does some short cuts on its first linear
walk that break as soon as we do a queued page.
c) We have to be careful to not break up host-page size chunks, since
this makes it harder to place the pages on the destination.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
On receiving MIG_RPCOMM_REQ_PAGES look up the address and
queue the page.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Soon we'll be in either ACTIVE or POSTCOPY_ACTIVE when we
complete migration, and we need to know which we expect to be
in to change state safely.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Maintain an 'unsentmap' of pages that have yet to be sent.
This is used in the following patches to discard some set of
the pages already sent as we enter postcopy mode.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Add qemu_savevm_state_complete_postcopy to complement
qemu_savevm_state_complete_precopy together with a new
save_live_complete_postcopy method on devices.
The save_live_complete_precopy method is called on
all devices during a precopy migration, and all non-postcopy
devices during a postcopy migration at the transition.
The save_live_complete_postcopy method is called at
the end of postcopy for all postcopiable devices.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Modify save_live_pending to return separate postcopiable and
non-postcopiable counts.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
In postcopy we're going to need to perform the complete phase
for postcopiable devices at a different point, start out by
renaming all of the 'complete's to make the difference obvious.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
The main RAM load loop has a call to host_from_stream_offset for
each page type that actually loads data with the same test;
factor it out before the switch.
The host = NULL is to silence a bogus gcc warning of
an unitialised in the RAM_SAVE_COMPRESS_PAGE case, it
doesn't seem to realise that host is always initialised by the if at
the top in the cases the switch takes.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Useful for debugging the migration bitmap and other bitmaps
of the same format (including the sentmap in postcopy).
The bitmap is printed to stderr.
Lines that are all the expected value are excluded so the output
can be quite compact for many bitmaps.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Add a function to find a RAMBlock by name; use it in two
of the places that already open code that loop; we've
got another use later in postcopy.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Just clean up code, no behavior change.
Signed-off-by: Liang Li <liang.z.li@intel.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>al3
Reviewed-by: Amit Shah <amit.shah@redhat.com>al3
Signed-off-by: Juan Quintela <quintela@redhat.com>al3
'cleanup' seems more appropriate than 'cancel'.
Signed-off-by: Liang Li <liang.z.li@intel.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>al3
Reviewed-by: Amit Shah <amit.shah@redhat.com>al3
Signed-off-by: Juan Quintela <quintela@redhat.com>al3
Because of the patch 3ea3b7fa9af067982f34b of kvm, which introduces a
lazy collapsing of small sptes into large sptes mechanism, now
migration_end() is a time consuming operation because it calls
memroy_global_dirty_log_stop(), which will trigger the dropping of small
sptes operation and takes about dozens of milliseconds, so call
migration_end() before all the vmsate data has already been transferred
to the destination will prolong VM downtime. This operation should be
deferred after all the data has been transferred to the destination.
blk_mig_cleanup() can be deferred too.
For a VM with 8G RAM, this patch can reduce the VM downtime about 30 ms.
Signed-off-by: Liang Li <liang.z.li@intel.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>al3
Reviewed-by: Amit Shah <amit.shah@redhat.com>al3
Signed-off-by: Juan Quintela <quintela@redhat.com>al3
Release qemu global mutex before call synchronize_rcu().
synchronize_rcu() waiting for all readers to finish their critical
sections. There is at least one critical section in which we try
to get QGM (critical section is in address_space_rw() and
prepare_mmio_access() is trying to aquire QGM).
Both functions (migration_end() and migration_bitmap_extend())
are called from main thread which is holding QGM.
Thus there is a race condition that ends up with deadlock:
main thread working thread
Lock QGA |
| Call KVM_EXIT_IO handler
| |
| Open rcu reader's critical section
Migration cleanup bh |
| |
synchronize_rcu() is |
waiting for readers |
| prepare_mmio_access() is waiting for QGM
\ /
deadlock
The patch changes bitmap freeing from direct g_free after synchronize_rcu
to free inside call_rcu.
Signed-off-by: Denis V. Lunev <den@openvz.org>
Reported-by: Igor Redko <redkoi@virtuozzo.com>
Tested-by: Igor Redko <redkoi@virtuozzo.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
CC: Anna Melekhova <annam@virtuozzo.com>
CC: Juan Quintela <quintela@redhat.com>
CC: Amit Shah <amit.shah@redhat.com>
CC: Paolo Bonzini <pbonzini@redhat.com>
CC: Wen Congyang <wency@cn.fujitsu.com>
Remove traditional auto-converge static 30ms throttling code and replace it
with a dynamic throttling algorithm.
Additionally, be more aggressive when deciding when to start throttling.
Previously we waited until four unproductive memory passes. Now we begin
throttling after only two unproductive memory passes. Four seemed quite
arbitrary and only waiting for two passes allows us to complete the migration
faster.
Signed-off-by: Jason J. Herne <jjherne@linux.vnet.ibm.com>
Reviewed-by: Matthew Rosato <mjrosato@linux.vnet.ibm.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Split out the finding of the dirty page and all the wrap detection
into a separate function since it was getting a bit hairy.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <1443018431-11170-3-git-send-email-dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
[Fix comment -- Amit]
Signed-off-by: Amit Shah <amit.shah@redhat.com>
Pull the search state for one iteration of the dirty page
search into a structure.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Message-Id: <1443018431-11170-2-git-send-email-dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
RAM migration mainly works on RAMBlocks but in a few places
uses data from MemoryRegions to access the same information that's
already held in RAMBlocks; clean it up just to avoid the
MemoryRegion use.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <1439463094-5394-2-git-send-email-dgilbert@redhat.com>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit.shah@redhat.com>
'strlen' is called three times in 'save_page_header', it's
inefficient.
Signed-off-by: Liang Li <liang.z.li@intel.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
migration_end calls synchronize_rcu() within a critical section.
That causes a deadlock; move the call after rcu_read_unlock().
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Prevously, if we hotplug a device(e.g. device_add e1000) during
migration is processing in source side, qemu will add a new ram
block but migration_bitmap is not extended.
In this case, migration_bitmap will overflow and lead qemu abort
unexpectedly.
Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
We need the names of RAMBlocks as they're loaded for RDMA,
reuse a slightly modified ram_control_load_hook:
a) Pass a 'data' parameter to use for the name in the block-reg
case
b) Only some hook types now require the presence of a hook function.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
There are two places that define 'len' variable, It's OK for compiling,
but makes it difficult for reading.
Remove the local one which defined in the inside 'while' loop.
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
To make changes easier, with the copy, I maintained almost all include
files. Now I remove the unnecessary ones on this patch. This compiles
on linux x64 with all architectures configured, and cross-compiles for
windows 32 and 64 bits.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
If anyone feels like adding himself to the list, just sent me a patch.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
For historic reasons, ram migration have been on arch_init.c. Just
split it into migration/ram.c, the same that happened with block.c.
There is only code movement, no changes altogether.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>