migration: ram_handle_compressed

ram_handle_compressed() should be aware size > TARGET_PAGE_SIZE
migration-rdma can call it with larger size.

Signed-off-by: Isaku Yamahata <yamahata@private.email.ne.jp>
---
 arch_init.c |   21 ++++++++++++++-------
 1 file changed, 14 insertions(+), 7 deletions(-)

Isaku Yamahata <yamahata@private.email.ne.jp> wrote:
> ram_handle_compressed() should be aware size > TARGET_PAGE_SIZE
> migration-rdma can call it with larger size.
>
> Signed-off-by: Isaku Yamahata <yamahata@private.email.ne.jp>
> ---
>  arch_init.c |   21 ++++++++++++++-------
>  1 file changed, 14 insertions(+), 7 deletions(-)
>
> diff --git a/arch_init.c b/arch_init.c
> index e47e139..64c81b0 100644
> --- a/arch_init.c
> +++ b/arch_init.c
> @@ -844,15 +844,22 @@ static inline void *host_from_stream_offset(QEMUFile *f,
>   */
>  void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
>  {
> -    if (ch != 0 || !is_zero_page(host)) {
> -        memset(host, ch, size);

I assume that size is always a multiple of pagesize, right?

> +    uint64_t pagesize = getpagesize();
> +    while (size > 0) {
> +        uint64_t length = MIN(pagesize, size);
> +
> +        if (ch !=0 || buffer_find_nonzero_offset(host, length) != length) {

This is a different optimization, we were't doing this before.

> +            memset(host, ch, length);
>  #ifndef _WIN32
> -        if (ch == 0 &&
> -            (!kvm_enabled() || kvm_has_sync_mmu()) &&
> -            getpagesize() <= TARGET_PAGE_SIZE) {
> -            qemu_madvise(host, TARGET_PAGE_SIZE, QEMU_MADV_DONTNEED);
> -        }
> +            if (ch == 0 &&
> +                (!kvm_enabled() || kvm_has_sync_mmu()) && pagesize <= length) {
> +                qemu_madvise(host, size, QEMU_MADV_DONTNEED);

Here don't we need s/size/length/?

> +            }
>  #endif
> +        }
> +
> +        size -= length;
> +        host += length;
>      }
>  }

On 09/18/2013 06:07 AM, Isaku Yamahata wrote:
> ram_handle_compressed() should be aware size > TARGET_PAGE_SIZE
> migration-rdma can call it with larger size.
>
> Signed-off-by: Isaku Yamahata <yamahata@private.email.ne.jp>
> ---
>   arch_init.c |   21 ++++++++++++++-------
>   1 file changed, 14 insertions(+), 7 deletions(-)
>
> diff --git a/arch_init.c b/arch_init.c
> index e47e139..64c81b0 100644
> --- a/arch_init.c
> +++ b/arch_init.c
> @@ -844,15 +844,22 @@ static inline void *host_from_stream_offset(QEMUFile *f,
>    */
>   void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
>   {
> -    if (ch != 0 || !is_zero_page(host)) {
> -        memset(host, ch, size);
> +    uint64_t pagesize = getpagesize();
> +    while (size > 0) {
> +        uint64_t length = MIN(pagesize, size);
> +
> +        if (ch !=0 || buffer_find_nonzero_offset(host, length) != length) {
> +            memset(host, ch, length);
>   #ifndef _WIN32
> -        if (ch == 0 &&
> -            (!kvm_enabled() || kvm_has_sync_mmu()) &&
> -            getpagesize() <= TARGET_PAGE_SIZE) {
> -            qemu_madvise(host, TARGET_PAGE_SIZE, QEMU_MADV_DONTNEED);
> -        }
> +            if (ch == 0 &&
> +                (!kvm_enabled() || kvm_has_sync_mmu()) && pagesize <= length) {
> +                qemu_madvise(host, size, QEMU_MADV_DONTNEED);
> +            }
>   #endif
> +        }
> +
> +        size -= length;
> +        host += length;
>       }
>   }
>

What is the reason for breaking the madvise() into smaller pieces?

- Michael

On Wed, Sep 18, 2013 at 02:08:40PM +0200, Juan Quintela wrote:
> Isaku Yamahata <yamahata@private.email.ne.jp> wrote:
> > ram_handle_compressed() should be aware size > TARGET_PAGE_SIZE
> > migration-rdma can call it with larger size.
> >
> > Signed-off-by: Isaku Yamahata <yamahata@private.email.ne.jp>
> > ---
> >  arch_init.c |   21 ++++++++++++++-------
> >  1 file changed, 14 insertions(+), 7 deletions(-)
> >
> > diff --git a/arch_init.c b/arch_init.c
> > index e47e139..64c81b0 100644
> > --- a/arch_init.c
> > +++ b/arch_init.c
> > @@ -844,15 +844,22 @@ static inline void *host_from_stream_offset(QEMUFile *f,
> >   */
> >  void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
> >  {
> > -    if (ch != 0 || !is_zero_page(host)) {
> > -        memset(host, ch, size);
> 
> I assume that size is always a multiple of pagesize, right?

Right. The issues is that is_zero_page() checks only first page.
The following page needs to be checked.

> 
> > +    uint64_t pagesize = getpagesize();
> > +    while (size > 0) {
> > +        uint64_t length = MIN(pagesize, size);
> > +
> > +        if (ch !=0 || buffer_find_nonzero_offset(host, length) != length) {
> 
> This is a different optimization, we were't doing this before.

Okay, I'll drop loop and use size.

thanks,

> > +            memset(host, ch, length);
> >  #ifndef _WIN32
> > -        if (ch == 0 &&
> > -            (!kvm_enabled() || kvm_has_sync_mmu()) &&
> > -            getpagesize() <= TARGET_PAGE_SIZE) {
> > -            qemu_madvise(host, TARGET_PAGE_SIZE, QEMU_MADV_DONTNEED);
> > -        }
> > +            if (ch == 0 &&
> > +                (!kvm_enabled() || kvm_has_sync_mmu()) && pagesize <= length) {
> > +                qemu_madvise(host, size, QEMU_MADV_DONTNEED);
> 
> Here don't we need s/size/length/?
> 
> > +            }
> >  #endif
> > +        }
> > +
> > +        size -= length;
> > +        host += length;
> >      }
> >  }
>