Re: [PATCHv5 01/19] mm/sparsemem: Allocate mem_section at runtime for SPARSEMEM_EXTREME

[Kirill A. Shutemov]

On Tue, Aug 22, 2017 at 06:28:26PM +0200, Borislav Petkov wrote:
> On Mon, Aug 21, 2017 at 06:28:58PM +0300, Kirill A. Shutemov wrote:
> > Size of mem_section array depends on size of physical address space.
> > 
> > In preparation for boot-time switching between paging modes on x86-64
> > we need to make allocation of mem_section dynamic.
> > 
> > The patch allocates the array on the first call to
> > sparse_memory_present_with_active_regions().
> > 
> > Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
> > ---
> >  include/linux/mmzone.h |  6 +++++-
> >  mm/page_alloc.c        | 10 ++++++++++
> >  mm/sparse.c            | 17 +++++++++++------
> >  3 files changed, 26 insertions(+), 7 deletions(-)
> 
> This patch needs running through checkpatch:
> 
> ERROR: code indent should use tabs where possible
> #53: FILE: include/linux/mmzone.h:1148:
> +        if (!mem_section)$
> 
> WARNING: please, no spaces at the start of a line
> #53: FILE: include/linux/mmzone.h:1148:
> +        if (!mem_section)$
> 
> ERROR: code indent should use tabs where possible
> #54: FILE: include/linux/mmzone.h:1149:
> +                return NULL;$
> 
> WARNING: please, no spaces at the start of a line
> #54: FILE: include/linux/mmzone.h:1149:
> +                return NULL;$
> 
> ERROR: "foo* bar" should be "foo *bar"
> #99: FILE: mm/sparse.c:106:
> +       struct mem_section* root = NULL;
> 
> ERROR: do not initialise statics to 0
> #118: FILE: mm/sparse.c:335:
> +       static unsigned long old_usemap_snr = 0;
> 
> ERROR: do not initialise statics to 0
> #119: FILE: mm/sparse.c:336:
> +       static unsigned long old_pgdat_snr = 0;
> 
> You should integrate it into your patch creation workflow.

Sorry for this.

> > diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
> > index fc14b8b3f6ce..9799c2c58ce6 100644
> > --- a/include/linux/mmzone.h
> > +++ b/include/linux/mmzone.h
> > @@ -1137,13 +1137,17 @@ struct mem_section {
> >  #define SECTION_ROOT_MASK	(SECTIONS_PER_ROOT - 1)
> >  
> >  #ifdef CONFIG_SPARSEMEM_EXTREME
> > -extern struct mem_section *mem_section[NR_SECTION_ROOTS];
> > +extern struct mem_section **mem_section;
> >  #else
> >  extern struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT];
> >  #endif
> >  
> >  static inline struct mem_section *__nr_to_section(unsigned long nr)
> >  {
> > +#ifdef CONFIG_SPARSEMEM_EXTREME
> > +        if (!mem_section)
> > +                return NULL;
> > +#endif
> >  	if (!mem_section[SECTION_NR_TO_ROOT(nr)])
> >  		return NULL;
> >  	return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK];
> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > index 6d30e914afb6..639fd2dce0c4 100644
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -5681,6 +5681,16 @@ void __init sparse_memory_present_with_active_regions(int nid)
> >  	unsigned long start_pfn, end_pfn;
> >  	int i, this_nid;
> >  
> > +#ifdef CONFIG_SPARSEMEM_EXTREME
> > +	if (!mem_section) {
> 
> Any chance this ifdeffery and above can use IS_ENABLED() instead?

Unfortunately, no.

This case cannot be changed to IS_ENABLED() as we don't define mem_section
and NR_SECTION_ROOTS for !SPARSEMEM.

The case above cannot be changed as GCC would complain in case of
!SPARSEMEM_EXTREME:

warning: the address of âmem_sectionâ will always evaluate as âtrueâ 

-- 
 Kirill A. Shutemov