Re: [Qemu-devel] [PATCH v3 03/29] postcopy: use UFFDIO_ZEROPAGE only when available

[Dr. David Alan Gilbert]

* Peter Xu (address@hidden) wrote:
> On Fri, Feb 16, 2018 at 01:15:59PM +0000, Dr. David Alan Gilbert (git) wrote:
> > From: "Dr. David Alan Gilbert" <address@hidden>
> > 
> > Use a flag on the RAMBlock to state whether it has the
> > UFFDIO_ZEROPAGE capability, use it when it's available.
> > 
> > This allows the use of postcopy on tmpfs as well as hugepage
> > backed files.
> > 
> > Signed-off-by: Dr. David Alan Gilbert <address@hidden>
> > ---
> >  exec.c                    | 15 +++++++++++++++
> >  include/exec/cpu-common.h |  3 +++
> >  migration/postcopy-ram.c  | 13 ++++++++++---
> >  3 files changed, 28 insertions(+), 3 deletions(-)
> > 
> > diff --git a/exec.c b/exec.c
> > index 0ec73bc917..1dc15298c2 100644
> > --- a/exec.c
> > +++ b/exec.c
> > @@ -99,6 +99,11 @@ static MemoryRegion io_mem_unassigned;
> >   */
> >  #define RAM_RESIZEABLE (1 << 2)
> >  
> > +/* UFFDIO_ZEROPAGE is available on this RAMBlock to atomically
> > + * zero the page and wake waiting processes.
> > + * (Set during postcopy)
> > + */
> > +#define RAM_UF_ZEROPAGE (1 << 3)
> >  #endif
> >  
> >  #ifdef TARGET_PAGE_BITS_VARY
> > @@ -1767,6 +1772,16 @@ bool qemu_ram_is_shared(RAMBlock *rb)
> >      return rb->flags & RAM_SHARED;
> >  }
> >  
> > +bool qemu_ram_is_uf_zeroable(RAMBlock *rb)
> > +{
> > +    return rb->flags & RAM_UF_ZEROPAGE;
> > +}
> > +
> > +void qemu_ram_set_uf_zeroable(RAMBlock *rb)
> > +{
> > +    rb->flags |= RAM_UF_ZEROPAGE;
> > +}
> > +
> >  /* Called with iothread lock held.  */
> >  void qemu_ram_set_idstr(RAMBlock *new_block, const char *name, DeviceState 
> > *dev)
> >  {
> > diff --git a/include/exec/cpu-common.h b/include/exec/cpu-common.h
> > index 0d861a6289..24d335f95d 100644
> > --- a/include/exec/cpu-common.h
> > +++ b/include/exec/cpu-common.h
> > @@ -73,6 +73,9 @@ void qemu_ram_set_idstr(RAMBlock *block, const char 
> > *name, DeviceState *dev);
> >  void qemu_ram_unset_idstr(RAMBlock *block);
> >  const char *qemu_ram_get_idstr(RAMBlock *rb);
> >  bool qemu_ram_is_shared(RAMBlock *rb);
> > +bool qemu_ram_is_uf_zeroable(RAMBlock *rb);
> > +void qemu_ram_set_uf_zeroable(RAMBlock *rb);
> > +
> >  size_t qemu_ram_pagesize(RAMBlock *block);
> >  size_t qemu_ram_pagesize_largest(void);
> >  
> > diff --git a/migration/postcopy-ram.c b/migration/postcopy-ram.c
> > index bec6c2c66b..6297979700 100644
> > --- a/migration/postcopy-ram.c
> > +++ b/migration/postcopy-ram.c
> > @@ -490,6 +490,10 @@ static int ram_block_enable_notify(const char 
> > *block_name, void *host_addr,
> >          error_report("%s userfault: Region doesn't support COPY", 
> > __func__);
> >          return -1;
> >      }
> > +    if (reg_struct.ioctls & ((__u64)1 << _UFFDIO_ZEROPAGE)) {
> > +        RAMBlock *rb = qemu_ram_block_by_name(block_name);
> > +        qemu_ram_set_uf_zeroable(rb);
> > +    }
> 
> So the zeroable flag is only set after a listening operation of
> postcopy migration.  One thing I am a bit worried is that if someone
> else wants to use the flag for a RAMBlock he/she may not notice this.
> Say, qemu_ram_is_uf_zeroable() is not valid if there is no such an
> incoming postcopy migration.

Yes, the problem is you don't get to know until you register it with
userfault.

> Maybe worth add a comment in the flag definition about this?

Yes, I've added:
/* Note: Only set at the start of postcopy */

above the qemu_ram_is_uf_zeroable() defintiion; also see the existing
'(Set during postcopy)' comment above the #define.


> Not a big deal (considering that I see no potential QEMU user for
> userfaultfd in short peroid), so no matter what:
> 
> Reviewed-by: Peter Xu <address@hidden>

Thanks.

Dave

> 
> >  
> >      return 0;
> >  }
> > @@ -699,11 +703,14 @@ int postcopy_place_page(MigrationIncomingState *mis, 
> > void *host, void *from,
> >  int postcopy_place_page_zero(MigrationIncomingState *mis, void *host,
> >                               RAMBlock *rb)
> >  {
> > +    size_t pagesize = qemu_ram_pagesize(rb);
> >      trace_postcopy_place_page_zero(host);
> >  
> > -    if (qemu_ram_pagesize(rb) == getpagesize()) {
> > -        if (qemu_ufd_copy_ioctl(mis->userfault_fd, host, NULL, 
> > getpagesize(),
> > -                                rb)) {
> > +    /* Normal RAMBlocks can zero a page using UFFDIO_ZEROPAGE
> > +     * but it's not available for everything (e.g. hugetlbpages)
> > +     */
> > +    if (qemu_ram_is_uf_zeroable(rb)) {
> > +        if (qemu_ufd_copy_ioctl(mis->userfault_fd, host, NULL, pagesize, 
> > rb)) {
> >              int e = errno;
> >              error_report("%s: %s zero host: %p",
> >                           __func__, strerror(e), host);
> > -- 
> > 2.14.3
> > 
> 
> -- 
> Peter Xu
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK