Re: [Qemu-devel] [PATCH v2 3/3] migration: add bitmap for received page

[Dr. David Alan Gilbert]

* Alexey Perevalov (address@hidden) wrote:
> On 06/16/2017 12:06 PM, Peter Xu wrote:
> > On Thu, Jun 15, 2017 at 07:36:15PM +0300, Alexey Perevalov wrote:
> > 
> > [...]
> > 
> > > +void init_receivedmap(void)
> > > +{
> > > +    RAMBlock *rb;
> > > +
> > > +    RAMBLOCK_FOREACH(rb) {
> > > +        unsigned long pages;
> > > +        pages = rb->max_length >> TARGET_PAGE_BITS;
> > Nit: I would assert(!rb->receivedmap) before new it in case for leak.
> ok,
> > 
> > > +        rb->receivedmap = bitmap_new(pages);
> > > +    }
> > > +}
> > > +
> > > +static unsigned long int get_received_bit_offset(void *host_addr, 
> > > RAMBlock *rb)
> > > +{
> > > +    uint64_t host_addr_offset = (uint64_t)(uintptr_t)(host_addr
> > > +                                                      - (void 
> > > *)rb->host);
> > > +    return host_addr_offset >> TARGET_PAGE_BITS;
> > > +}
> > Can we rename this function to ramblock_get_pfn() and export it in
> > ramblock API (exec/ram_addr.h)?
> if you need it, I'll place export into ram_addr.h
> but name, PFN could confuse, because in case of hugepage
> address this function will not return real PFN, due to
> 
> TARGET_PAGE_BITS usage.
> 
> > 
> > static inline uint64_t ramblock_get_pfn(RAMBlock *rb, void *host)
> > {
> >      return (host - rb->host) >> TARGET_PAGE_BITS;
> > }
> > 
> > > +
> > > +int test_receivedmap_by_addr(void *host_addr, RAMBlock *rb)
> > > +{
> > > +    return test_bit(get_received_bit_offset(host_addr, rb), 
> > > rb->receivedmap);
> > > +}
> > > +
> > > +void set_receivedmap_by_addr(void *host_addr, RAMBlock *rb)
> > > +{
> > > +    set_bit_atomic(get_received_bit_offset(host_addr, rb), 
> > > rb->receivedmap);
> > > +}
> > > +
> > >   /*
> > >    * An outstanding page request, on the source, having been received
> > >    * and queued
> > > @@ -2324,8 +2352,13 @@ static int ram_load_setup(QEMUFile *f, void 
> > > *opaque)
> > >   static int ram_load_cleanup(void *opaque)
> > >   {
> > > +    RAMBlock *rb;
> > >       xbzrle_load_cleanup();
> > >       compress_threads_load_cleanup();
> > > +
> > > +    RAMBLOCK_FOREACH(rb) {
> > Nit: I'd prefer: if (rb->receivedmap) {...}
> g_free already contains it
> > 
> > > +        g_free(rb->receivedmap);
> > and: rb->receivedmap = NULL later.
> why not )
> > 
> > > +    }
> > >       return 0;
> > >   }
> > > diff --git a/migration/ram.h b/migration/ram.h
> > > index c081fde..7048ff9 100644
> > > --- a/migration/ram.h
> > > +++ b/migration/ram.h
> > > @@ -52,4 +52,9 @@ int ram_discard_range(const char *block_name, uint64_t 
> > > start, size_t length);
> > >   int ram_postcopy_incoming_init(MigrationIncomingState *mis);
> > >   void ram_handle_compressed(void *host, uint8_t ch, uint64_t size);
> > > +
> > > +void init_receivedmap(void);
> > > +int test_receivedmap_by_addr(void *host_addr, RAMBlock *rb);
> > > +void set_receivedmap_by_addr(void *host_addr, RAMBlock *rb);
> > Nit: I would name these names as: ramblock_recv_map_*().
> > 
> > > +
> > >   #endif
> > > diff --git a/migration/savevm.c b/migration/savevm.c
> > > index 31158da..668d3bb 100644
> > > --- a/migration/savevm.c
> > > +++ b/migration/savevm.c
> > > @@ -1372,6 +1372,7 @@ static int 
> > > loadvm_postcopy_handle_advise(MigrationIncomingState *mis)
> > >           return -1;
> > >       }
> > > +    init_receivedmap();
> > >       remote_pagesize_summary = qemu_get_be64(mis->from_src_file);
> > >       local_pagesize_summary = ram_pagesize_summary();
> > > -- 
> > > 1.9.1
> > > 
> > Again, did you trap precopy?
> sorry, I really forgot, now it's not copied bitmap, it's
> received bitmap. So, I know about
> case RAM_SAVE_FLAG_PAGE, but it seems another cases exist
> (compressed/xbzrle)
> and there is no single point where coping is doing.

Yes we probably should trap precopy as well; I'm not sure yet
whether it'll trigger during the network cases; but certainly
in the recovery-from-postcopy-failure case you want to know
which pages have been received.
You'd also want to clear flags in this bitmap during the discard
phase.

Dave

> 
> 
> > 
> > Thanks,
> > 
> 
> -- 
> Best regards,
> Alexey Perevalov
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK