Re: [PATCH RFC] memory: Don't allow to resize RAM while migrating

[David Hildenbrand <david@redhat.com>]

>> diff --git a/migration/ram.c b/migration/ram.c
>> index ed23ed1c7c..f86f32b453 100644
>> --- a/migration/ram.c
>> +++ b/migration/ram.c
>> @@ -52,6 +52,7 @@
>>  #include "migration/colo.h"
>>  #include "block.h"
>>  #include "sysemu/sysemu.h"
>> +#include "sysemu/runstate.h"
>>  #include "savevm.h"
>>  #include "qemu/iov.h"
>>  #include "multifd.h"
>> @@ -3710,8 +3711,49 @@ static SaveVMHandlers savevm_ram_handlers = {
>>      .resume_prepare = ram_resume_prepare,
>>  };
>>  
>> +static void ram_mig_ram_block_resized(RAMBlockNotifier *n, void *host,
>> +                                      size_t old_size, size_t new_size)
>> +{
>> +    /*
>> +     * We don't care about resizes triggered on incoming migration (when
>> +     * syncing ram blocks), or of course, when no migration is going on.
>> +     */
>> +    if (migration_is_idle() || !runstate_is_running()) {
>> +        return;
>> +    }
>> +
>> +    if (!postcopy_is_running()) {
>> +        Error *err = NULL;
>> +
>> +        /*
>> +         * Precopy code cannot deal with the size of ram blocks changing at
>> +         * random points in time. We're still running on the source, abort
>> +         * the migration and continue running here. Make sure to wait until
>> +         * migration was canceled.
>> +         */
>> +        error_setg(&err, "RAM resized during precopy.");
>> +        migrate_set_error(migrate_get_current(), err);
>> +        error_free(err);
>> +        migration_cancel();
> 
> If we can't do anything else, this is reasonable.
> 
> But as discussed before, it is still not fully clear to me _why_ are
> ramblocks changing if we have disabled add/remove memory during migration.


Ramblock add/remove is ties to device add/remove, which we block.

Resize, however, it not. Here, the resize happens while the guest is
booting up. The content/size of the ram block depends also on previous
guest action AFAIK. There is no way from stopping the guest from doing
that. It's required for the guest to continue booting (with ACPI).

I'm currently working on a project which reuses resizable ram blocks in
different context. There, I can simply defer/avoid resizing when
migration is active. In the ACPI case, however, we cannot avoid it.

Hope that answers your question

> 
>> +    } else {
>> +        /*
>> +         * Postcopy code cannot deal with the size of ram blocks changing at
>> +         * random points in time. We're running on the target. Fail hard.
>> +         *
>> +         * TODO: How to handle this in a better way?
>> +         */
>> +        error_report("RAM resized during postcopy.");
>> +        exit(-1);
> 
> Idea is good, but we also need to exit destination, not only source, no?

@Dave, any idea what could be the right thing to do here?

> 
>> +    }
>> +}
> 
> 
> 
>> +static RAMBlockNotifier ram_mig_ram_notifier = {
>> +    .ram_block_resized = ram_mig_ram_block_resized,
>> +};
>> +
>>  void ram_mig_init(void)
>>  {
>>      qemu_mutex_init(&XBZRLE.lock);
>>      register_savevm_live("ram", 0, 4, &savevm_ram_handlers, &ram_state);
>> +    ram_block_notifier_add(&ram_mig_ram_notifier);
>>  }
> 
> Shouldn't we remove the notifier when we finish the migration.

It's called from main() unconditionally (so not when migration starts),
so the notifier remains active the whole QEMU lifetime (which should be
fine AFAIKT).

-- 
Thanks,

David / dhildenb