Re: [Qemu-devel] [PATCH COLO-Frame v12 31/38] COLO: Separate the process of saving/loading ram and device state

["Dr. David Alan Gilbert" <dgilbert@redhat.com>]

* zhanghailiang (zhang.zhanghailiang@huawei.com) wrote:
> We separate the process of saving/loading ram and device state when do checkpoint,
> we add new helpers for save/load ram/device. With this change, we can directly
> transfer ram from master to slave without using QEMUSizeBuffer as assistant,
> which also reduce the size of extra memory been used during checkpoint.
> 
> Besides, we move the colo_flush_ram_cache to the proper position after the
> above change.
> 
> Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
> Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
> ---
> v11:
> - Remove load configuration section in qemu_loadvm_state_begin()
> 
> Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
> ---
>  include/sysemu/sysemu.h |   6 +++
>  migration/colo.c        |  43 ++++++++++++----
>  migration/ram.c         |   5 --
>  migration/savevm.c      | 132 ++++++++++++++++++++++++++++++++++++++++++++++--
>  4 files changed, 168 insertions(+), 18 deletions(-)
> 
> diff --git a/include/sysemu/sysemu.h b/include/sysemu/sysemu.h
> index 91eeda3..5deae53 100644
> --- a/include/sysemu/sysemu.h
> +++ b/include/sysemu/sysemu.h
> @@ -133,7 +133,13 @@ void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name,
>                                             uint64_t *start_list,
>                                             uint64_t *length_list);
>  
> +int qemu_save_ram_precopy(QEMUFile *f);
> +int qemu_save_device_state(QEMUFile *f);
> +
>  int qemu_loadvm_state(QEMUFile *f);
> +int qemu_loadvm_state_begin(QEMUFile *f);
> +int qemu_load_ram_state(QEMUFile *f);
> +int qemu_load_device_state(QEMUFile *f);
>  
>  typedef enum DisplayType
>  {
> diff --git a/migration/colo.c b/migration/colo.c
> index 62a0444..d253d64 100644
> --- a/migration/colo.c
> +++ b/migration/colo.c
> @@ -272,21 +272,32 @@ static int colo_do_checkpoint_transaction(MigrationState *s,
>          goto out;
>      }
>  
> +    ret = colo_put_cmd(s->to_dst_file, COLO_COMMAND_VMSTATE_SEND);
> +    if (ret < 0) {
> +        goto out;
> +    }
>      /* Disable block migration */
>      s->params.blk = 0;
>      s->params.shared = 0;
> -    qemu_savevm_state_header(trans);
> -    qemu_savevm_state_begin(trans, &s->params);
> -    qemu_mutex_lock_iothread();
> -    qemu_savevm_state_complete_precopy(trans, false);
> -    qemu_mutex_unlock_iothread();
> -
> -    qemu_fflush(trans);
> +    qemu_savevm_state_begin(s->to_dst_file, &s->params);
> +    ret = qemu_file_get_error(s->to_dst_file);
> +    if (ret < 0) {
> +        error_report("save vm state begin error\n");

You don't need \n in error_report (Markus is trying to get rid
of all the existing cases where people do that!)

> +        goto out;
> +    }
>  
> -    ret = colo_put_cmd(s->to_dst_file, COLO_COMMAND_VMSTATE_SEND);
> +    qemu_mutex_lock_iothread();
> +    /* Note: device state is saved into buffer */
> +    ret = qemu_save_device_state(trans);
>      if (ret < 0) {
> +        error_report("save device state error\n");
> +        qemu_mutex_unlock_iothread();
>          goto out;
>      }
> +    qemu_fflush(trans);
> +    qemu_save_ram_precopy(s->to_dst_file);
> +    qemu_mutex_unlock_iothread();
> +

It's interesting you're saving the devices and then saving the RAM,
where as in a normal migration we always save the RAM first and then
the devices;  I don't _think_ it makes any difference but I thought
I'd point it out.

>      /* we send the total size of the vmstate first */
>      size = qsb_get_length(buffer);
>      ret = colo_put_cmd_value(s->to_dst_file, COLO_COMMAND_VMSTATE_SIZE, size);
> @@ -545,6 +556,16 @@ void *colo_process_incoming_thread(void *opaque)
>              goto out;
>          }
>  
> +        ret = qemu_loadvm_state_begin(mis->from_src_file);
> +        if (ret < 0) {
> +            error_report("load vm state begin error, ret=%d", ret);
> +            goto out;
> +        }
> +        ret = qemu_load_ram_state(mis->from_src_file);
> +        if (ret < 0) {
> +            error_report("load ram state error");
> +            goto out;
> +        }
>          /* read the VM state total size first */
>          ret = colo_get_cmd_value(mis->from_src_file,
>                                   COLO_COMMAND_VMSTATE_SIZE, &value);
> @@ -577,8 +598,10 @@ void *colo_process_incoming_thread(void *opaque)
>          qemu_mutex_lock_iothread();
>          qemu_system_reset(VMRESET_SILENT);
>          vmstate_loading = true;
> -        if (qemu_loadvm_state(fb) < 0) {
> -            error_report("COLO: loadvm failed");
> +        colo_flush_ram_cache();
> +        ret = qemu_load_device_state(fb);
> +        if (ret < 0) {
> +            error_report("COLO: load device state failed\n");
>              vmstate_loading = false;
>              qemu_mutex_unlock_iothread();
>              goto out;
> diff --git a/migration/ram.c b/migration/ram.c
> index 8ff7f7c..45d9332 100644
> --- a/migration/ram.c
> +++ b/migration/ram.c
> @@ -2458,7 +2458,6 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
>       * be atomic
>       */
>      bool postcopy_running = postcopy_state_get() >= POSTCOPY_INCOMING_LISTENING;
> -    bool need_flush = false;
>  
>      seq_iter++;
>  
> @@ -2493,7 +2492,6 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
>              /* After going into COLO, we should load the Page into colo_cache */
>              if (ram_cache_enable) {
>                  host = colo_cache_from_block_offset(block, addr);
> -                need_flush = true;
>              } else {
>                  host = host_from_ram_block_offset(block, addr);
>              }
> @@ -2588,9 +2586,6 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
>  
>      rcu_read_unlock();
>  
> -    if (!ret  && ram_cache_enable && need_flush) {
> -        colo_flush_ram_cache();
> -    }
>      DPRINTF("Completed load of VM with exit code %d seq iteration "
>              "%" PRIu64 "\n", ret, seq_iter);
>      return ret;
> diff --git a/migration/savevm.c b/migration/savevm.c
> index c7c26d8..94c0d10 100644
> --- a/migration/savevm.c
> +++ b/migration/savevm.c
> @@ -50,6 +50,7 @@
>  #include "qemu/iov.h"
>  #include "block/snapshot.h"
>  #include "block/qapi.h"
> +#include "migration/colo.h"
>  
>  
>  #ifndef ETH_P_RARP
> @@ -923,6 +924,10 @@ void qemu_savevm_state_begin(QEMUFile *f,
>              break;
>          }
>      }
> +    if (migration_in_colo_state()) {
> +        qemu_put_byte(f, QEMU_VM_EOF);
> +        qemu_fflush(f);
> +    }
>  }
>  
>  /*
> @@ -1192,13 +1197,44 @@ static int qemu_savevm_state(QEMUFile *f, Error **errp)
>      return ret;
>  }
>  
> -static int qemu_save_device_state(QEMUFile *f)
> +int qemu_save_ram_precopy(QEMUFile *f)
>  {
>      SaveStateEntry *se;
> +    int ret = 0;
>  
> -    qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
> -    qemu_put_be32(f, QEMU_VM_FILE_VERSION);
> +    QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
> +        if (!se->ops || !se->ops->save_live_complete_precopy) {
> +            continue;
> +        }
> +        if (se->ops && se->ops->is_active) {
> +            if (!se->ops->is_active(se->opaque)) {
> +                continue;
> +            }
> +        }
> +        trace_savevm_section_start(se->idstr, se->section_id);

Please update the trace_ names to match the function.

> +
> +        save_section_header(f, se, QEMU_VM_SECTION_END);
>  
> +        ret = se->ops->save_live_complete_precopy(f, se->opaque);
> +        trace_savevm_section_end(se->idstr, se->section_id, ret);
> +        save_section_footer(f, se);
> +        if (ret < 0) {
> +            qemu_file_set_error(f, ret);
> +            return ret;
> +        }
> +    }
> +    qemu_put_byte(f, QEMU_VM_EOF);
> +
> +    return 0;
> +}

OK, that function is a bit odd - you're relying on a device having
save_live_complete_precopy to let you know that it's a RAM block; it's
currently true but it'll get more interesting if anyone tries to
add any other postcopy devices.  Please add a comment at least
to point that out.

> +
> +int qemu_save_device_state(QEMUFile *f)
> +{
> +    SaveStateEntry *se;
> +
> +    if (!migration_in_colo_state()) {
> +        qemu_savevm_state_header(f);
> +    }
>      cpu_synchronize_all_states();
>  
>      QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
> @@ -1938,6 +1974,96 @@ int qemu_loadvm_state(QEMUFile *f)
>      return ret;
>  }
>  
> +int qemu_loadvm_state_begin(QEMUFile *f)
> +{
> +    uint8_t section_type;
> +    int ret = -1;
> +    MigrationIncomingState *mis = migration_incoming_get_current();
> +
> +    if (!mis) {
> +        error_report("qemu_loadvm_state_begin");
> +        return -EINVAL;
> +    }

an odd error; how can that happen?

> +    /* CleanUp */
> +    loadvm_free_handlers(mis);

I don't understand why you do that here?

> +
> +    if (qemu_savevm_state_blocked(NULL)) {
> +        return -EINVAL;
> +    }

The other calls to that function print the error it returns!

> +    while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) {
> +        if (section_type != QEMU_VM_SECTION_START) {
> +            error_report("QEMU_VM_SECTION_START");
> +            ret = -EINVAL;
> +            goto out;
> +        }
> +        ret = qemu_loadvm_section_start_full(f, mis);
> +        if (ret < 0) {
> +            goto out;
> +        }
> +    }
> +    ret = qemu_file_get_error(f);
> +    if (ret == 0) {
> +        return 0;
> +     }

That 'if' isn't needed - just remove the 3 lines and it will do the same
thing!

> +out:
> +    return ret;
> +}
> +
> +int qemu_load_ram_state(QEMUFile *f)
> +{
> +    uint8_t section_type;
> +    MigrationIncomingState *mis = migration_incoming_get_current();
> +    int ret = -1;
> +
> +    while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) {
> +        if (section_type != QEMU_VM_SECTION_PART &&
> +            section_type != QEMU_VM_SECTION_END) {
> +            error_report("load ram state, not get "
> +                         "QEMU_VM_SECTION_FULL or QEMU_VM_SECTION_END");
> +            return -EINVAL;
> +        }
> +        ret = qemu_loadvm_section_part_end(f, mis);
> +        if (ret < 0) {
> +            goto out;
> +        }
> +    }
> +    ret = qemu_file_get_error(f);
> +    if (ret == 0) {
> +        return 0;
> +     }
> +out:
> +    return ret;
> +}
> +
> +int qemu_load_device_state(QEMUFile *f)
> +{
> +    uint8_t section_type;
> +    MigrationIncomingState *mis = migration_incoming_get_current();
> +    int ret = -1;
> +
> +    while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) {
> +        if (section_type != QEMU_VM_SECTION_FULL) {
> +            error_report("load device state error: "
> +                         "Not get QEMU_VM_SECTION_FULL");
> +            return -EINVAL;
> +        }
> +        ret = qemu_loadvm_section_start_full(f, mis);
> +        if (ret < 0) {
> +            goto out;
> +        }
> +    }
> +
> +    ret = qemu_file_get_error(f);
> +
> +    cpu_synchronize_all_post_init();
> +    if (ret == 0) {
> +        return 0;
> +    }
> +out:
> +    return ret;
> +}
> +

These three functions are all very similar;  would it be easier
just to call qemu_loadvm_state_main ?  Perhaps add a flag/enum
parameter to it for it to check which section_types are allowed
in the 3 different cases?

Dave

>  void hmp_savevm(Monitor *mon, const QDict *qdict)
>  {
>      BlockDriverState *bs, *bs1;
> -- 
> 1.8.3.1
> 
> 
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK