Re: [PATCH v1] mm: page_alloc: Add debug log in free_reserved_area for static memory

[David Hildenbrand]

On 06.10.21 14:13, Faiyaz Mohammed wrote:
> Hi,
>
> Sorry for delayed response.
>
> On 9/29/2021 10:33 PM, David Hildenbrand wrote:
> > On 29.09.21 10:58, Faiyaz Mohammed wrote:
> > > On 9/28/2021 4:46 PM, David Hildenbrand wrote:
> > > > On 28.09.21 12:53, Faiyaz Mohammed wrote:
> > > > > On 9/28/2021 4:09 PM, David Hildenbrand wrote:
> > > > > > On 28.09.21 11:04, Faiyaz Mohammed wrote:
> > > > > > > For INITRD and initmem memory is reserved through "memblock_reserve"
> > > > > > > during boot up but it is free via "free_reserved_area" instead
> > > > > > > of "memblock_free".
> > > > > > > For example:
> > > > > > > [    0.294848] Freeing initrd memory: 12K.
> > > > > > > [    0.696688] Freeing unused kernel memory: 4096K.
> > > > > > >
> > > > > > > To get the start and end address of the above freed memory and to
> > > > > > > account
> > > > > > > proper memblock added memblock_dbg log in "free_reserved_area".
> > > > > > > After adding log:
> > > > > > > [    0.294837] memblock_free: [0x00000083600000-0x00000083603000]
> > > > > > > free_initrd_mem+0x20/0x28
> > > > > > > [    0.294848] Freeing initrd memory: 12K.
> > > > > > > [    0.695246] memblock_free: [0x00000081600000-0x00000081a00000]
> > > > > > > free_initmem+0x70/0xc8
> > > > > > > [    0.696688] Freeing unused kernel memory: 4096K.
> > > > > > >
> > > > > > > Signed-off-by: Faiyaz Mohammed <faiyazm@xxxxxxxxxxxxxx>
> > > > > > > ---
> > > > > > >      mm/page_alloc.c | 5 +++++
> > > > > > >      1 file changed, 5 insertions(+)
> > > > > > >
> > > > > > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > > > > > > index b37435c..f85c3b2 100644
> > > > > > > --- a/mm/page_alloc.c
> > > > > > > +++ b/mm/page_alloc.c
> > > > > > > @@ -8129,6 +8129,11 @@ unsigned long free_reserved_area(void *start,
> > > > > > > void *end, int poison, const char
> > > > > > >              pr_info("Freeing %s memory: %ldK\n",
> > > > > > >                  s, pages << (PAGE_SHIFT - 10));
> > > > > > >      +#ifdef CONFIG_HAVE_MEMBLOCK
> > > > > > > +        memblock_dbg("memblock_free: [%#016llx-%#016llx] %pS\n",
> > > > > > > +            __pa(start), __pa(end), (void *)_RET_IP_);
> > > > > > > +#endif
> > > > > >
> > > > > > IMHO, the "memblock_free" part is misleading. Something was allocated
> > > > > > early via memblock, then we transitioned to the buddy, now we're
> > > > > > freeing
> > > > > > that early allocation via the buddy.
> > > > > > Yes, we're freeing the early allocation via buddy, but for proper
> > > > > memblock accounting we need this debug print.
> > > >
> > > > What do you mean with "accounting" ? These are debug statements.
> > > Yes, these are debug statements, which help to know the a-b address
> > > belongs to x callsite. This info is required when memblock=debug is
> > > passed through command line and CONFIG_HAVE_MEMBLOCK is enabled.
> >
> > The issue I'm having is talking in the name of memblock "memblock_dbg,
> > memblock_free", when memblock might no longer be around. We have other
> > places where we free early memblock allocations back to the buddy.
> I didn't find place where we free early memblock allocation back to the
> buddy.

One example I know is

section_deactivate()->free_map_bootmem()->vmemmap_free()-> ... 
free_pagetable()->free_reserved_page().

when we free the vmemmap allocated via memblock back to the buddy.

> Why "memblock_dbg" print with "memblock_free" string?.
> - After buddy took over, buddy will free memblock reserved memory
> through free_reserved_area and it will print the freed memory size, but
> the freed memory through buddy still be part of memblock.reserved.regions.
> - To know the address ranges, added the "memblock_dbg" print along with
> "membloc_free" string.
> - If it is misleading or confusing, we can remove the "memblock_free"
> string from the "memblock_dgb" print and we can just print the address
> range when "memlock=debug" pass through command line.

That would be better, but do we really have to depend on 
"memlock=debug"? Can't we do pr_debug() ?

--
Thanks,

David / dhildenb