Re: [Qemu-devel] [PATCH v10 05/14] migration: Create ram_multifd_page

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

* Juan Quintela (quintela@redhat.com) wrote:
> The function still don't use multifd, but we have simplified
> ram_save_page, xbzrle and RDMA stuff is gone.  We have added a new
> counter and a new flag for this type of pages.
> 
> Signed-off-by: Juan Quintela <quintela@redhat.com>
> 
> --
> Add last_page parameter
> Add commets for done and address
> Remove multifd field, it is the same than normal pages
> Merge next patch, now we send multiple pages at a time
> Remove counter for multifd pages, it is identical to normal pages
> Use iovec's instead of creating the equivalent.
> Clear memory used by pages (dave)
> Use g_new0(danp)
> define MULTIFD_CONTINUE
> now pages member is a pointer
> Fix off-by-one in number of pages in one packet
> ---
>  migration/ram.c        | 159 ++++++++++++++++++++++++++++++++++++++++++++++++-
>  migration/trace-events |   2 +
>  2 files changed, 160 insertions(+), 1 deletion(-)
> 
> diff --git a/migration/ram.c b/migration/ram.c
> index aef5a323f3..5d6b46ac23 100644
> --- a/migration/ram.c
> +++ b/migration/ram.c
> @@ -52,6 +52,7 @@
>  #include "migration/block.h"
>  #include "sysemu/sysemu.h"
>  #include "qemu/uuid.h"
> +#include "qemu/iov.h"
>  
>  /***********************************************************/
>  /* ram save/restore */
> @@ -71,6 +72,7 @@
>  #define RAM_SAVE_FLAG_XBZRLE   0x40
>  /* 0x80 is reserved in migration.h start with 0x100 next */
>  #define RAM_SAVE_FLAG_COMPRESS_PAGE    0x100
> +#define RAM_SAVE_FLAG_MULTIFD_PAGE     0x200
>  
>  static inline bool is_zero_range(uint8_t *p, uint64_t size)
>  {
> @@ -395,14 +397,36 @@ static void compress_threads_save_setup(void)
>  
>  /* Multiple fd's */
>  
> +/* used to continue on the same multifd group */
> +#define MULTIFD_CONTINUE UINT16_MAX
> +
> +typedef struct {
> +    /* number of used pages */

      Telling us it's used doesn't help much; if I understand
right, this is for accumulating them into one large block
before actually sending them?

> +    uint32_t used;
> +    /* number of allocated pages */
> +    uint32_t allocated;
> +    /* global number of generated multifd packets */
> +    uint32_t seq;
> +    struct iovec *iov;
> +    RAMBlock *block;
> +} multifd_pages_t;
> +
>  struct MultiFDSendParams {
> +    /* not changed */

???

>      uint8_t id;
>      char *name;
>      QemuThread thread;
>      QIOChannel *c;
>      QemuSemaphore sem;
>      QemuMutex mutex;
> +    /* protected by param mutex */
>      bool quit;
> +    multifd_pages_t *pages;
> +    /* how many patches has sent this channel */

s/patches/packets/

> +    uint32_t packets_sent;
> +    /* protected by multifd mutex */
> +    /* has the thread finish the last submitted job */
> +    bool done;
>  };
>  typedef struct MultiFDSendParams MultiFDSendParams;
>  
> @@ -410,8 +434,31 @@ struct {
>      MultiFDSendParams *params;
>      /* number of created threads */
>      int count;
> +    QemuMutex mutex;
> +    QemuSemaphore sem;
> +    multifd_pages_t *pages;
>  } *multifd_send_state;
>  
> +static void multifd_pages_init(multifd_pages_t **ppages, size_t size)

What is the 'size' here - it's allocated pages for something?

> +{
> +    multifd_pages_t *pages = g_new0(multifd_pages_t, 1);
> +
> +    pages->allocated = size;
> +    pages->iov = g_new0(struct iovec, size);
> +    *ppages = pages;
> +}
> +
> +static void multifd_pages_clear(multifd_pages_t *pages)
> +{
> +    pages->used = 0;
> +    pages->allocated = 0;
> +    pages->seq = 0;
> +    pages->block = NULL;
> +    g_free(pages->iov);
> +    pages->iov = NULL;
> +    g_free(pages);
> +}
> +
>  static void terminate_multifd_send_threads(Error *errp)
>  {
>      int i;
> @@ -453,9 +500,13 @@ int multifd_save_cleanup(Error **errp)
>          socket_send_channel_destroy(p->c);
>          g_free(p->name);
>          p->name = NULL;
> +        multifd_pages_clear(p->pages);
> +        p->pages = NULL;
>      }
>      g_free(multifd_send_state->params);
>      multifd_send_state->params = NULL;
> +    multifd_pages_clear(multifd_send_state->pages);
> +    multifd_send_state->pages = NULL;
>      g_free(multifd_send_state);
>      multifd_send_state = NULL;
>      return ret;
> @@ -482,6 +533,7 @@ static void *multifd_send_thread(void *opaque)
>          terminate_multifd_send_threads(local_err);
>          return NULL;
>      }
> +    qemu_sem_post(&multifd_send_state->sem);
>  
>      while (true) {
>          qemu_mutex_lock(&p->mutex);
> @@ -489,9 +541,24 @@ static void *multifd_send_thread(void *opaque)
>              qemu_mutex_unlock(&p->mutex);
>              break;
>          }
> +        if (p->pages->used) {
> +            p->pages->used = 0;
> +            qemu_mutex_unlock(&p->mutex);
> +
> +            trace_multifd_send(p->id, p->pages->seq, p->pages->used);

but p->pages->used is just been set to 0?

> +            /* ToDo: send page here */
> +
> +            qemu_mutex_lock(&multifd_send_state->mutex);
> +            p->done = true;
> +            p->packets_sent++;
> +            qemu_mutex_unlock(&multifd_send_state->mutex);
> +            qemu_sem_post(&multifd_send_state->sem);
> +            continue;
> +        }
>          qemu_mutex_unlock(&p->mutex);
>          qemu_sem_wait(&p->sem);
>      }
> +    trace_multifd_send_thread(p->id, p->packets_sent);
>  
>      return NULL;
>  }
> @@ -529,6 +596,10 @@ int multifd_save_setup(void)
>      multifd_send_state = g_malloc0(sizeof(*multifd_send_state));
>      multifd_send_state->params = g_new0(MultiFDSendParams, thread_count);
>      multifd_send_state->count = 0;
> +    qemu_mutex_init(&multifd_send_state->mutex);
> +    qemu_sem_init(&multifd_send_state->sem, 0);
> +    multifd_pages_init(&multifd_send_state->pages,
> +                       migrate_multifd_page_count());
>      for (i = 0; i < thread_count; i++) {
>          MultiFDSendParams *p = &multifd_send_state->params[i];
>  
> @@ -536,12 +607,58 @@ int multifd_save_setup(void)
>          qemu_sem_init(&p->sem, 0);
>          p->quit = false;
>          p->id = i;
> +        p->done = true;
> +        multifd_pages_init(&p->pages, migrate_multifd_page_count());
>          p->name = g_strdup_printf("multifdsend_%d", i);
>          socket_send_channel_create(multifd_new_send_channel_async, p);
>      }
>      return 0;
>  }
>  
> +static uint16_t multifd_send_page(RAMBlock *block, ram_addr_t offset,
> +                                  bool last_page)
> +{
> +    int i;
> +    MultiFDSendParams *p = NULL; /* make happy gcc */

(English: Make gcc happy)

> +    multifd_pages_t *pages = multifd_send_state->pages;
> +
> +    if (!pages->block) {
> +        pages->block = block;
> +    }
> +
> +    pages->iov[pages->used].iov_base = block->host + offset;
> +    pages->iov[pages->used].iov_len = TARGET_PAGE_SIZE;
> +    pages->used++;
> +
> +    if (!last_page) {
> +        if (pages->used < pages->allocated) {
> +            return MULTIFD_CONTINUE;
> +        }
> +    }

I'm confused by this a bit.
Isn't the next bit waiting for a free thread?

> +    qemu_sem_wait(&multifd_send_state->sem);
> +    qemu_mutex_lock(&multifd_send_state->mutex);
> +    for (i = 0; i < multifd_send_state->count; i++) {
> +        p = &multifd_send_state->params[i];
> +
> +        if (p->done) {
> +            p->done = false;
> +            break;
> +        }
> +    }
> +    qemu_mutex_unlock(&multifd_send_state->mutex);
> +    qemu_mutex_lock(&p->mutex);
> +    p->pages->used = 0;

If we're handing the block of pages to the thread, I don't understand
why we zero used here.

> +    p->pages->seq = pages->seq + 1;
> +    p->pages->block = NULL;
> +    multifd_send_state->pages = p->pages;
> +    p->pages = pages;
> +    qemu_mutex_unlock(&p->mutex);
> +    qemu_sem_post(&p->sem);
> +
> +    return i;
> +}
> +
>  struct MultiFDRecvParams {
>      uint8_t id;
>      char *name;
> @@ -1070,6 +1187,31 @@ static int ram_save_page(RAMState *rs, PageSearchStatus *pss, bool last_stage)
>      return pages;
>  }
>  
> +static int ram_multifd_page(RAMState *rs, PageSearchStatus *pss,
> +                            bool last_stage)
> +{
> +    int pages;
> +    uint8_t *p;
> +    RAMBlock *block = pss->block;
> +    ram_addr_t offset = pss->page << TARGET_PAGE_BITS;
> +
> +    p = block->host + offset;
> +
> +    pages = save_zero_page(rs, block, offset);
> +    if (pages == -1) {
> +        ram_counters.transferred +=
> +            save_page_header(rs, rs->f, block,
> +                             offset | RAM_SAVE_FLAG_MULTIFD_PAGE);
> +        multifd_send_page(block, offset, rs->migration_dirty_pages == 1);
> +        qemu_put_buffer(rs->f, p, TARGET_PAGE_SIZE);

So that's temporary - we're hoping multifd_send_page will do that?

> +        ram_counters.transferred += TARGET_PAGE_SIZE;
> +        pages = 1;
> +        ram_counters.normal++;
> +    }
> +
> +    return pages;
> +}
> +
>  static int do_compress_ram_page(QEMUFile *f, RAMBlock *block,
>                                  ram_addr_t offset)
>  {
> @@ -1498,6 +1640,8 @@ static int ram_save_target_page(RAMState *rs, PageSearchStatus *pss,
>          if (migrate_use_compression() &&
>              (rs->ram_bulk_stage || !migrate_use_xbzrle())) {
>              res = ram_save_compressed_page(rs, pss, last_stage);
> +        } else if (migrate_use_multifd()) {
> +            res = ram_multifd_page(rs, pss, last_stage);
>          } else {
>              res = ram_save_page(rs, pss, last_stage);
>          }
> @@ -2878,6 +3022,10 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
>      if (!migrate_use_compression()) {
>          invalid_flags |= RAM_SAVE_FLAG_COMPRESS_PAGE;
>      }
> +
> +    if (!migrate_use_multifd()) {
> +        invalid_flags |= RAM_SAVE_FLAG_MULTIFD_PAGE;
> +    }
>      /* This RCU critical section can be very long running.
>       * When RCU reclaims in the code start to become numerous,
>       * it will be necessary to reduce the granularity of this
> @@ -2902,13 +3050,17 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
>              if (flags & invalid_flags & RAM_SAVE_FLAG_COMPRESS_PAGE) {
>                  error_report("Received an unexpected compressed page");
>              }
> +            if (flags & invalid_flags  & RAM_SAVE_FLAG_MULTIFD_PAGE) {
> +                error_report("Received an unexpected multifd page");
> +            }
>  
>              ret = -EINVAL;
>              break;
>          }
>  
>          if (flags & (RAM_SAVE_FLAG_ZERO | RAM_SAVE_FLAG_PAGE |
> -                     RAM_SAVE_FLAG_COMPRESS_PAGE | RAM_SAVE_FLAG_XBZRLE)) {
> +                     RAM_SAVE_FLAG_COMPRESS_PAGE | RAM_SAVE_FLAG_XBZRLE |
> +                     RAM_SAVE_FLAG_MULTIFD_PAGE)) {
>              RAMBlock *block = ram_block_from_stream(f, flags);
>  
>              host = host_from_ram_block_offset(block, addr);
> @@ -2997,6 +3149,11 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
>                  break;
>              }
>              break;
> +
> +        case RAM_SAVE_FLAG_MULTIFD_PAGE:
> +            qemu_get_buffer(f, host, TARGET_PAGE_SIZE);
> +            break;
> +
>          case RAM_SAVE_FLAG_EOS:
>              /* normal exit */
>              break;
> diff --git a/migration/trace-events b/migration/trace-events
> index 141e773305..61ee21a13e 100644
> --- a/migration/trace-events
> +++ b/migration/trace-events
> @@ -77,6 +77,8 @@ ram_load_postcopy_loop(uint64_t addr, int flags) "@%" PRIx64 " %x"
>  ram_postcopy_send_discard_bitmap(void) ""
>  ram_save_page(const char *rbname, uint64_t offset, void *host) "%s: offset: 0x%" PRIx64 " host: %p"
>  ram_save_queue_pages(const char *rbname, size_t start, size_t len) "%s: start: 0x%zx len: 0x%zx"
> +multifd_send(char id, int seq, int num) "channel %d sequence %d num pages %d"
> +multifd_send_thread(char id, uint32_t packets) "channel %d packets %d"
>  
>  # migration/migration.c
>  await_return_path_close_on_source_close(void) ""
> -- 
> 2.14.3
> 
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK