Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1730845 > unrolled thread

Re: [PATCH] mm/memory_hotplug: fix wrong casting for __remove_section()

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH] mm/memory_hotplug: fix wrong casting for  __remove_section() Michal Hocko <mhocko@kernel.org> - 2017-09-12 15:00 +0200
    Re: [PATCH] mm/memory_hotplug: fix wrong casting for  __remove_section() YASUAKI ISHIMATSU <yasu.isimatu@gmail.com> - 2017-09-12 19:10 +0200
      Re: [PATCH] mm/memory_hotplug: fix wrong casting for  __remove_section() Michal Hocko <mhocko@kernel.org> - 2017-09-13 08:00 +0200
        Re: [PATCH] mm/memory_hotplug: fix wrong casting for  __remove_section() YASUAKI ISHIMATSU <yasu.isimatu@gmail.com> - 2017-09-14 17:50 +0200
          Re: [PATCH] mm/memory_hotplug: fix wrong casting for  __remove_section() Michal Hocko <mhocko@kernel.org> - 2017-09-15 11:40 +0200

#1730845 — Re: [PATCH] mm/memory_hotplug: fix wrong casting for __remove_section()

On Fri 08-09-17 16:43:04, YASUAKI ISHIMATSU wrote:
> __remove_section() calls __remove_zone() to shrink zone and pgdat.
> But due to wrong castings, __remvoe_zone() cannot shrink zone
> and pgdat correctly if pfn is over 0xffffffff.
> 
> So the patch fixes the following 3 wrong castings.
> 
>   1. find_smallest_section_pfn() returns 0 or start_pfn which defined
>      as unsigned long. But the function always returns 32bit value
>      since the function is defined as int.
> 
>   2. find_biggest_section_pfn() returns 0 or pfn which defined as
>      unsigned long. the function always returns 32bit value
>      since the function is defined as int.

this is indeed wrong. Pfns over would be really broken 15TB. Not that
unrealistic these days

> 
>   3. __remove_section() calculates start_pfn using section_nr_to_pfn()
>      and scn_nr. section_nr_to_pfn() just shifts scn_nr by
>      PFN_SECTION_SHIFT bit. But since scn_nr is defined as int,
>      section_nr_to_pfn() always return 32 bit value.

Dohh, those nasty macros. This is hidden quite well. It seems other
callers are using unsigned long properly. But I would rather make sure
we won't repeat that error again. Can we instead make section_nr_to_pfn
resp. pfn_to_section_nr static inline and enfore proper types?

I would also split this into two patches. 

Thanks!

> The patch fixes the wrong castings.
> 
> Signed-off-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
> ---
>  mm/memory_hotplug.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
> index 73bf17d..3514ef2 100644
> --- a/mm/memory_hotplug.c
> +++ b/mm/memory_hotplug.c
> @@ -331,7 +331,7 @@ int __ref __add_pages(int nid, unsigned long phys_start_pfn,
> 
>  #ifdef CONFIG_MEMORY_HOTREMOVE
>  /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
> -static int find_smallest_section_pfn(int nid, struct zone *zone,
> +static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
>  				     unsigned long start_pfn,
>  				     unsigned long end_pfn)
>  {
> @@ -356,7 +356,7 @@ static int find_smallest_section_pfn(int nid, struct zone *zone,
>  }
> 
>  /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
> -static int find_biggest_section_pfn(int nid, struct zone *zone,
> +static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
>  				    unsigned long start_pfn,
>  				    unsigned long end_pfn)
>  {
> @@ -544,7 +544,7 @@ static int __remove_section(struct zone *zone, struct mem_section *ms,
>  		return ret;
> 
>  	scn_nr = __section_nr(ms);
> -	start_pfn = section_nr_to_pfn(scn_nr);
> +	start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
>  	__remove_zone(zone, start_pfn);
> 
>  	sparse_remove_one_section(zone, ms, map_offset);
> -- 
> 1.8.3.1
> 

-- 
Michal Hocko
SUSE Labs

[toc] | [next] | [standalone]

#1731048

Hi Michal,

Thanks you for reviewing my patch.

On 09/12/2017 08:49 AM, Michal Hocko wrote:
> On Fri 08-09-17 16:43:04, YASUAKI ISHIMATSU wrote:
>> __remove_section() calls __remove_zone() to shrink zone and pgdat.
>> But due to wrong castings, __remvoe_zone() cannot shrink zone
>> and pgdat correctly if pfn is over 0xffffffff.
>>
>> So the patch fixes the following 3 wrong castings.
>>
>>   1. find_smallest_section_pfn() returns 0 or start_pfn which defined
>>      as unsigned long. But the function always returns 32bit value
>>      since the function is defined as int.
>>
>>   2. find_biggest_section_pfn() returns 0 or pfn which defined as
>>      unsigned long. the function always returns 32bit value
>>      since the function is defined as int.
> 
> this is indeed wrong. Pfns over would be really broken 15TB. Not that
> unrealistic these days

Why 15TB?

Actually, all callers use pfn which defined as unsigned long to receive
the return value of find_{smallest|biggest}_section_nr(). So it will break
over 16TB.

> 
>>
>>   3. __remove_section() calculates start_pfn using section_nr_to_pfn()
>>      and scn_nr. section_nr_to_pfn() just shifts scn_nr by
>>      PFN_SECTION_SHIFT bit. But since scn_nr is defined as int,
>>      section_nr_to_pfn() always return 32 bit value.
> 
> Dohh, those nasty macros. This is hidden quite well. It seems other
> callers are using unsigned long properly. But I would rather make sure
> we won't repeat that error again. Can we instead make section_nr_to_pfn
> resp. pfn_to_section_nr static inline and enfore proper types?

I'll update it.

> 
> I would also split this into two patches. 

I'll update it.

Thanks,
Yasuaki Ishimatsu

> 
> Thanks!
> 
>> The patch fixes the wrong castings.
>>
>> Signed-off-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
>> ---
>>  mm/memory_hotplug.c | 6 +++---
>>  1 file changed, 3 insertions(+), 3 deletions(-)
>>
>> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
>> index 73bf17d..3514ef2 100644
>> --- a/mm/memory_hotplug.c
>> +++ b/mm/memory_hotplug.c
>> @@ -331,7 +331,7 @@ int __ref __add_pages(int nid, unsigned long phys_start_pfn,
>>
>>  #ifdef CONFIG_MEMORY_HOTREMOVE
>>  /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
>> -static int find_smallest_section_pfn(int nid, struct zone *zone,
>> +static unsigned long find_smallest_section_pfn(int nid, struct zone *zone,
>>  				     unsigned long start_pfn,
>>  				     unsigned long end_pfn)
>>  {
>> @@ -356,7 +356,7 @@ static int find_smallest_section_pfn(int nid, struct zone *zone,
>>  }
>>
>>  /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
>> -static int find_biggest_section_pfn(int nid, struct zone *zone,
>> +static unsigned long find_biggest_section_pfn(int nid, struct zone *zone,
>>  				    unsigned long start_pfn,
>>  				    unsigned long end_pfn)
>>  {
>> @@ -544,7 +544,7 @@ static int __remove_section(struct zone *zone, struct mem_section *ms,
>>  		return ret;
>>
>>  	scn_nr = __section_nr(ms);
>> -	start_pfn = section_nr_to_pfn(scn_nr);
>> +	start_pfn = section_nr_to_pfn((unsigned long)scn_nr);
>>  	__remove_zone(zone, start_pfn);
>>
>>  	sparse_remove_one_section(zone, ms, map_offset);
>> -- 
>> 1.8.3.1
>>
> 

[toc] | [prev] | [next] | [standalone]

#1731387

On Tue 12-09-17 13:05:39, YASUAKI ISHIMATSU wrote:
> Hi Michal,
> 
> Thanks you for reviewing my patch.
> 
> On 09/12/2017 08:49 AM, Michal Hocko wrote:
> > On Fri 08-09-17 16:43:04, YASUAKI ISHIMATSU wrote:
> >> __remove_section() calls __remove_zone() to shrink zone and pgdat.
> >> But due to wrong castings, __remvoe_zone() cannot shrink zone
> >> and pgdat correctly if pfn is over 0xffffffff.
> >>
> >> So the patch fixes the following 3 wrong castings.
> >>
> >>   1. find_smallest_section_pfn() returns 0 or start_pfn which defined
> >>      as unsigned long. But the function always returns 32bit value
> >>      since the function is defined as int.
> >>
> >>   2. find_biggest_section_pfn() returns 0 or pfn which defined as
> >>      unsigned long. the function always returns 32bit value
> >>      since the function is defined as int.
> > 
> > this is indeed wrong. Pfns over would be really broken 15TB. Not that
> > unrealistic these days
> 
> Why 15TB?

0xffffffff>>28

-- 
Michal Hocko
SUSE Labs

[toc] | [prev] | [next] | [standalone]

#1732362

Hi Michal,

On 09/13/2017 01:59 AM, Michal Hocko wrote:
> On Tue 12-09-17 13:05:39, YASUAKI ISHIMATSU wrote:
>> Hi Michal,
>>
>> Thanks you for reviewing my patch.
>>
>> On 09/12/2017 08:49 AM, Michal Hocko wrote:
>>> On Fri 08-09-17 16:43:04, YASUAKI ISHIMATSU wrote:
>>>> __remove_section() calls __remove_zone() to shrink zone and pgdat.
>>>> But due to wrong castings, __remvoe_zone() cannot shrink zone
>>>> and pgdat correctly if pfn is over 0xffffffff.
>>>>
>>>> So the patch fixes the following 3 wrong castings.
>>>>
>>>>   1. find_smallest_section_pfn() returns 0 or start_pfn which defined
>>>>      as unsigned long. But the function always returns 32bit value
>>>>      since the function is defined as int.
>>>>
>>>>   2. find_biggest_section_pfn() returns 0 or pfn which defined as
>>>>      unsigned long. the function always returns 32bit value
>>>>      since the function is defined as int.
>>>
>>> this is indeed wrong. Pfns over would be really broken 15TB. Not that
>>> unrealistic these days
>>
>> Why 15TB?
> 
> 0xffffffff>>28
> 

Even thought I see your explanation, I cannot understand.

In my understanding, find_{smallest|biggest}_section_pfn() return integer.
So the functions always return 0x00000000 - 0xffffffff. Therefore if pfn is over
0xffffffff (under 16TB), then the function cannot work correctly.

What am I wrong?

Thanks,
Yasuaki Ishimatsu

[toc] | [prev] | [next] | [standalone]

#1732800

On Thu 14-09-17 11:43:10, YASUAKI ISHIMATSU wrote:
> Hi Michal,
> 
> On 09/13/2017 01:59 AM, Michal Hocko wrote:
> > On Tue 12-09-17 13:05:39, YASUAKI ISHIMATSU wrote:
> >> Hi Michal,
> >>
> >> Thanks you for reviewing my patch.
> >>
> >> On 09/12/2017 08:49 AM, Michal Hocko wrote:
> >>> On Fri 08-09-17 16:43:04, YASUAKI ISHIMATSU wrote:
> >>>> __remove_section() calls __remove_zone() to shrink zone and pgdat.
> >>>> But due to wrong castings, __remvoe_zone() cannot shrink zone
> >>>> and pgdat correctly if pfn is over 0xffffffff.
> >>>>
> >>>> So the patch fixes the following 3 wrong castings.
> >>>>
> >>>>   1. find_smallest_section_pfn() returns 0 or start_pfn which defined
> >>>>      as unsigned long. But the function always returns 32bit value
> >>>>      since the function is defined as int.
> >>>>
> >>>>   2. find_biggest_section_pfn() returns 0 or pfn which defined as
> >>>>      unsigned long. the function always returns 32bit value
> >>>>      since the function is defined as int.
> >>>
> >>> this is indeed wrong. Pfns over would be really broken 15TB. Not that
> >>> unrealistic these days
> >>
> >> Why 15TB?
> > 
> > 0xffffffff>>28
> > 
> 
> Even thought I see your explanation, I cannot understand.
> 
> In my understanding, find_{smallest|biggest}_section_pfn() return integer.
> So the functions always return 0x00000000 - 0xffffffff. Therefore if pfn is over
> 0xffffffff (under 16TB), then the function cannot work correctly.
> 
> What am I wrong?

You are not wrong. We are talking about the same thing AFAICS. I was
just less precise...

-- 
Michal Hocko
SUSE Labs

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web