Re: [Xen-devel] [RFC XEN PATCH 01/16] x86_64/mm: explicitly specify the location to place the frame table

[Haozhong Zhang <haozhong.zhang@xxxxxxxxx>]

On 12/09/16 16:35 -0500, Konrad Rzeszutek Wilk wrote:
> On Mon, Oct 10, 2016 at 08:32:20AM +0800, Haozhong Zhang wrote:
> > A reserved area on each pmem region is used to place the frame table.
> > However, it's not at the beginning of the pmem region, so we need to
> > specify the location explicitly when extending the frame table.
> >
> > Signed-off-by: Haozhong Zhang <haozhong.zhang@xxxxxxxxx>
> > ---
> > Cc: Jan Beulich <jbeulich@xxxxxxxx>
> > Cc: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
> > ---
> >  xen/arch/x86/x86_64/mm.c | 9 +++++----
> >  1 file changed, 5 insertions(+), 4 deletions(-)
> >
> > diff --git a/xen/arch/x86/x86_64/mm.c b/xen/arch/x86/x86_64/mm.c
> > index b8b6b70..33f226a 100644
> > --- a/xen/arch/x86/x86_64/mm.c
> > +++ b/xen/arch/x86/x86_64/mm.c
> > @@ -792,7 +792,8 @@ static int setup_frametable_chunk(void *start, void *end,
> >      return 0;
> >  }
> >
> > -static int extend_frame_table(struct mem_hotadd_info *info)
> > +static int extend_frame_table(struct mem_hotadd_info *info,
> This looks like it could be 'const struct mem_hotadd_info *info' ?
yes

> > +                              struct mem_hotadd_info *alloc_info)
> >  {
> >      unsigned long cidx, nidx, eidx, spfn, epfn;
> >
> > @@ -818,9 +819,9 @@ static int extend_frame_table(struct mem_hotadd_info *info)
> >          nidx = find_next_bit(pdx_group_valid, eidx, cidx);
> >          if ( nidx >= eidx )
> >              nidx = eidx;
> > -        err = setup_frametable_chunk(pdx_to_page(cidx * PDX_GROUP_COUNT ),
> > +        err = setup_frametable_chunk(pdx_to_page(cidx * PDX_GROUP_COUNT),
> >                                       pdx_to_page(nidx * PDX_GROUP_COUNT),
> > -                                     info);
> > +                                     alloc_info);
> Granted this one modifies the 'alloc_info' in
> alloc_hotadd_mfn, and 'alloc_info'
> >          if ( err )
> >              return err;
> >
> > @@ -1413,7 +1414,7 @@ int memory_add(unsigned long spfn, unsigned long epfn, 
> > unsigned int pxm)
> >      info.epfn = epfn;
> >      info.cur = spfn;
> >
> > -    ret = extend_frame_table(&info);
> > +    ret = extend_frame_table(&info, &info);
> is equivalant to 'info' so I am not sure I understand the purpose
> behind this patch?
Yes, they are identical for the ordinary RAM here, and the frame table
is allocated at the begin of the hot-added RAM. For NVDIMM, the
hypervisor does not know which part is used for data, so the second
parameter 'alloc_info' is used to indicate which part can be used for
the frame table, and might be different from 'info'.

Thanks,
Haozhong

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel