On Thu, Feb 27, 2020 at 9:57 AM Geert Uytterhoeven [off-list ref] wrote:

Hi Arjun,

On Thu, Feb 27, 2020 at 6:45 PM Arjun Roy [off-list ref] wrote:

quoted

On Thu, Feb 27, 2020 at 9:13 AM Arjun Roy [off-list ref] wrote:

quoted

On Thu, Feb 27, 2020 at 1:03 AM Geert Uytterhoeven [off-list ref] wrote:

quoted

On Thu, Feb 27, 2020 at 5:12 AM Stephen Rothwell [off-list ref] wrote:

quoted

After merging the akpm tree, today's linux-next build (sparc defconfig)
failed like this:

In file included from include/linux/list.h:9:0,
                 from include/linux/smp.h:12,
                 from include/linux/kernel_stat.h:5,
                 from mm/memory.c:42:
mm/memory.c: In function 'insert_pages':
mm/memory.c:1523:41: error: implicit declaration of function 'pte_index'; did you mean 'page_index'? [-Werror=implicit-function-declaration]
   remaining_pages_total, PTRS_PER_PTE - pte_index(addr));
                                         ^
include/linux/kernel.h:842:40: note: in definition of macro '__typecheck'
   (!!(sizeof((typeof(x) *)1 == (typeof(y) *)1)))
                                        ^
include/linux/kernel.h:866:24: note: in expansion of macro '__safe_cmp'
  __builtin_choose_expr(__safe_cmp(x, y), \
                        ^~~~~~~~~~
include/linux/kernel.h:934:27: note: in expansion of macro '__careful_cmp'
 #define min_t(type, x, y) __careful_cmp((type)(x), (type)(y), <)
                           ^~~~~~~~~~~~~
mm/memory.c:1522:26: note: in expansion of macro 'min_t'
  pages_to_write_in_pmd = min_t(unsigned long,
                          ^~~~~

Same issue on m68k, as per a report from kisskb.

quoted

Caused by patch

  "mm/memory.c: add vm_insert_pages()"

sparc32 does not implement pte_index at all :-(

Seems like about only half of the architectures do.

:/ I begin to suspect the only sane way to make this work is to have a
per-arch header defined method, returning a bool saying whether
pte_index() is meaningful or not on that arch, and early on in
vm_insert_pages() if that bool returns true, to just call
vm_insert_page() in a loop.

So, here is what I propose: something like the following macro in a
per-arch header:

#define PTE_INDEX_DEFINED 1 // or 0 if it is not

pte_index is already a #define on architectures where it exists, so
you can just use that.

quoted

In mm/memory.c, another macro:

#ifndef PTE_INDEX_DEFINED
#define PTE_INDEX_DEFINED 0
#endifndef

No need for the above...

quoted

And inside vm_insert_pages:

int vm_insert_pages() {

#if PTE_INDEX_DEFINED

... if you use "#ifdef" here.

Sounds good, thanks. I'll cook up a patch and send it along.

-Arjun

quoted

// The existing method

#else

for (i=0; i<n; ++i)
        vm_insert_page(i)

#endif
}

That way:
1. No playing whack-a-mole with different architectures
2. Architecture that knows pte_index is meaningful works can define
this explicitly
3. Can remove the sparc patches modifying pte_index that Stephen and I
contributed.

If that sounds acceptable I can cook a patch.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds