[PATCH] arm64: mm: fix max_mapnr is assigned the wrong value

[PATCH] arm64: mm: fix max_mapnr is assigned the wrong value

[Muchun Song]

When we not use flat memory, the mem_map will be NULL and
pfn_to_page(max_pfn) is a pointer which is located in kernel space. So
max_mapnr is assigned a very large number(e.g., 0xffffxxxx_xxxxxxxx) - fix
it.

Signed-off-by: Muchun Song <smuchun@gmail.com>
---
 arch/arm64/mm/init.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/arm64/mm/init.c b/arch/arm64/mm/init.c
index bc02818fa48b..e86c21a44c88 100644
--- a/arch/arm64/mm/init.c
+++ b/arch/arm64/mm/init.c
@@ -535,7 +535,7 @@ void __init mem_init(void)
 	else
 		swiotlb_force = SWIOTLB_NO_FORCE;
 
-	set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
+	set_max_mapnr(max_pfn - PHYS_PFN_OFFSET);
 
 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 	free_unused_memmap();
-- 
2.17.1

Re: [PATCH] arm64: mm: fix max_mapnr is assigned the wrong value

[Catalin Marinas]

On Sat, Mar 30, 2019 at 09:13:46PM +0800, Muchun Song wrote:
> When we not use flat memory, the mem_map will be NULL and
> pfn_to_page(max_pfn) is a pointer which is located in kernel space. So
> max_mapnr is assigned a very large number(e.g., 0xffffxxxx_xxxxxxxx) - fix
> it.
> 
> Signed-off-by: Muchun Song <smuchun@gmail.com>
> ---
>  arch/arm64/mm/init.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/arm64/mm/init.c b/arch/arm64/mm/init.c
> index bc02818fa48b..e86c21a44c88 100644
> --- a/arch/arm64/mm/init.c
> +++ b/arch/arm64/mm/init.c
> @@ -535,7 +535,7 @@ void __init mem_init(void)
>  	else
>  		swiotlb_force = SWIOTLB_NO_FORCE;
>  
> -	set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
> +	set_max_mapnr(max_pfn - PHYS_PFN_OFFSET);
>  
>  #ifndef CONFIG_SPARSEMEM_VMEMMAP
>  	free_unused_memmap();

The patch looks fine but did you actually hit any problem? max_mapnr
seems to only be used in the generic pfn_valid() which we do not use on
arm64 (just wondering if it needs a cc stable; it doesn't look like as
it's not a regression).

-- 
Catalin

Re: [PATCH] arm64: mm: fix max_mapnr is assigned the wrong value

[Muchun Song]

Hi Catalin,

Thanks for your reply.  In fact, I didn't hit any problem.  I just read the code
and found the problem. I see max_mapnr is to only be used in the
generic pfn_valid().
As you said, we do not use it on arm64.  So in a sense, the patch is
meaningless.

But I think since it is a problem, why not fix it (Even if max_mapnr
is not be used)?
What is your opinion?

Catalin Marinas <catalin.marinas@arm.com> 于2019年4月1日周一 下午6:44写道：
>
> On Sat, Mar 30, 2019 at 09:13:46PM +0800, Muchun Song wrote:
> > When we not use flat memory, the mem_map will be NULL and
> > pfn_to_page(max_pfn) is a pointer which is located in kernel space. So
> > max_mapnr is assigned a very large number(e.g., 0xffffxxxx_xxxxxxxx) - fix
> > it.
> >
> > Signed-off-by: Muchun Song <smuchun@gmail.com>
> > ---
> >  arch/arm64/mm/init.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/arch/arm64/mm/init.c b/arch/arm64/mm/init.c
> > index bc02818fa48b..e86c21a44c88 100644
> > --- a/arch/arm64/mm/init.c
> > +++ b/arch/arm64/mm/init.c
> > @@ -535,7 +535,7 @@ void __init mem_init(void)
> >       else
> >               swiotlb_force = SWIOTLB_NO_FORCE;
> >
> > -     set_max_mapnr(pfn_to_page(max_pfn) - mem_map);
> > +     set_max_mapnr(max_pfn - PHYS_PFN_OFFSET);
> >
> >  #ifndef CONFIG_SPARSEMEM_VMEMMAP
> >       free_unused_memmap();
>
> The patch looks fine but did you actually hit any problem? max_mapnr
> seems to only be used in the generic pfn_valid() which we do not use on
> arm64 (just wondering if it needs a cc stable; it doesn't look like as
> it's not a regression).
>
> --
> Catalin

Re: [PATCH] arm64: mm: fix max_mapnr is assigned the wrong value

[Catalin Marinas]

On Mon, Apr 01, 2019 at 10:21:58PM +0800, Muchun Song wrote:
> Thanks for your reply.  In fact, I didn't hit any problem.  I just read the code
> and found the problem. I see max_mapnr is to only be used in the
> generic pfn_valid().
> As you said, we do not use it on arm64.  So in a sense, the patch is
> meaningless.
> 
> But I think since it is a problem, why not fix it (Even if max_mapnr
> is not be used)?
> What is your opinion?

I'm ok with fixing, only that it's not urgent. Will can pick it up for
the 5.2 merging window.

Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>