Re: [PATCH v3 03/12] mm: Make HPAGE_PXD_* macros even if !THP

[Peter Xu]

On Fri, Mar 22, 2024 at 10:14:56AM -0700, SeongJae Park wrote:
> Hi Peter,

Hi, SeongJae,

> 
> On Thu, 21 Mar 2024 18:07:53 -0400 peterx@xxxxxxxxxx wrote:
> 
> > From: Peter Xu <peterx@xxxxxxxxxx>
> > 
> > These macros can be helpful when we plan to merge hugetlb code into generic
> > code.  Move them out and define them even if !THP.
> > 
> > We actually already defined HPAGE_PMD_NR for other reasons even if !THP.
> > Reorganize these macros.
> > 
> > Reviewed-by: Christoph Hellwig <hch@xxxxxxxxxxxxx>
> > Reviewed-by: Jason Gunthorpe <jgg@xxxxxxxxxx>
> > Signed-off-by: Peter Xu <peterx@xxxxxxxxxx>
> > ---
> >  include/linux/huge_mm.h | 17 ++++++-----------
> >  1 file changed, 6 insertions(+), 11 deletions(-)
> > 
> > diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
> > index de0c89105076..3bcdfc7e5d57 100644
> > --- a/include/linux/huge_mm.h
> > +++ b/include/linux/huge_mm.h
> > @@ -64,9 +64,6 @@ ssize_t single_hugepage_flag_show(struct kobject *kobj,
> >  				  enum transparent_hugepage_flag flag);
> >  extern struct kobj_attribute shmem_enabled_attr;
> >  
> > -#define HPAGE_PMD_ORDER (HPAGE_PMD_SHIFT-PAGE_SHIFT)
> > -#define HPAGE_PMD_NR (1<<HPAGE_PMD_ORDER)
> > -
> >  /*
> >   * Mask of all large folio orders supported for anonymous THP; all orders up to
> >   * and including PMD_ORDER, except order-0 (which is not "huge") and order-1
> > @@ -87,14 +84,19 @@ extern struct kobj_attribute shmem_enabled_attr;
> >  #define thp_vma_allowable_order(vma, vm_flags, smaps, in_pf, enforce_sysfs, order) \
> >  	(!!thp_vma_allowable_orders(vma, vm_flags, smaps, in_pf, enforce_sysfs, BIT(order)))
> >  
> > -#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> >  #define HPAGE_PMD_SHIFT PMD_SHIFT
> >  #define HPAGE_PMD_SIZE	((1UL) << HPAGE_PMD_SHIFT)
> >  #define HPAGE_PMD_MASK	(~(HPAGE_PMD_SIZE - 1))
> > +#define HPAGE_PMD_ORDER (HPAGE_PMD_SHIFT-PAGE_SHIFT)
> > +#define HPAGE_PMD_NR (1<<HPAGE_PMD_ORDER)
> >  
> >  #define HPAGE_PUD_SHIFT PUD_SHIFT
> >  #define HPAGE_PUD_SIZE	((1UL) << HPAGE_PUD_SHIFT)
> >  #define HPAGE_PUD_MASK	(~(HPAGE_PUD_SIZE - 1))
> > +#define HPAGE_PUD_ORDER (HPAGE_PUD_SHIFT-PAGE_SHIFT)
> > +#define HPAGE_PUD_NR (1<<HPAGE_PUD_ORDER)
> 
> I just found latest mm-unstable fails one of my build configurations[1] with
> below error.  'git bisect' says this is the first patch set started the
> failure.  I haven't looked in deep, but just reporting first.
> 
>     In file included from .../include/linux/mm.h:1115,
>                      from .../mm/vmstat.c:14:
>     .../mm/vmstat.c: In function 'zoneinfo_show_print':
>     .../include/linux/huge_mm.h:87:25: error: 'PMD_SHIFT' undeclared (first use in this function); did you mean 'PUD_SHIFT'?
>        87 | #define HPAGE_PMD_SHIFT PMD_SHIFT
>           |                         ^~~~~~~~~
> 
> [1] https://github.com/awslabs/damon-tests/blob/next/corr/tests/build_m68k.sh

Apologies for the issue.  This is caused by !CONFIG_MMU, I think.

I thought this would be fairly easy to fix by putting these macros under
CONFIG_PGTABLE_HAS_HUGE_LEAVES, however when doing this I could have found
some other issue that violates this rule.. mm/vmstat.c has referenced
HPAGE_PMD_NR even if vmstat_item_print_in_thp() wanted to guard it only
with CONFIG_THP.

/home/peterx/git/linux/mm/vmstat.c: In function 'zoneinfo_show_print':
/home/peterx/git/linux/mm/vmstat.c:1689:42: error: 'HPAGE_PMD_NR' undeclared (first use in this function)
 1689 |                                 pages /= HPAGE_PMD_NR;
      |                                          ^~~~~~~~~~~~
/home/peterx/git/linux/mm/vmstat.c:1689:42: note: each undeclared identifier is reported only once for each function it appears in
  CC      drivers/tty/tty_port.o
/home/peterx/git/linux/mm/vmstat.c: In function 'vmstat_start':
/home/peterx/git/linux/mm/vmstat.c:1822:33: error: 'HPAGE_PMD_NR' undeclared (first use in this function)
 1822 |                         v[i] /= HPAGE_PMD_NR;
      |                                 ^~~~~~~~~~~~
make[4]: *** [/home/peterx/git/linux/scripts/Makefile.build:243: mm/vmstat.o] Error 1

static __always_inline bool vmstat_item_print_in_thp(enum node_stat_item item)
{
        if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
                return false;
        ...
}

I think the problem is vmstat_item_print_in_thp() uses IS_ENABLED() however
that won't stop compiler from looking into the "if".. so it'll still try to
find the HPAGE_PMD_NR macro.

It means, I may need to further change vmstat_item_print_in_thp() to make
this work in the clean way.. by properly switching to a #ifdef.

For that I'll need to post a formal patch and add people to review.  I'll
keep you posted.

Side note: thank you for your script, just to mention make.cross has been
moved to kbuild/, and it'll also need kbuild.sh now to work.  So you may
want to fix your test script (and it worked for you because you kept the
old make.cross around), like:

  wget https://raw.githubusercontent.com/intel/lkp-tests/master/kbuild/make.cross -O ./bin/make.cross
  wget https://raw.githubusercontent.com/intel/lkp-tests/master/kbuild/kbuild.sh -O ./bin/kbuild.sh

Thanks,

-- 
Peter Xu