diff mbox series

[2/2] mm: don't hide potentially null memmap pointer in sparse_remove_section

Message ID 20190827053656.32191-3-alastair@au1.ibm.com (mailing list archive)
State New, archived
Headers show
Series mm: Minor cleanup | expand

Commit Message

Alastair D'Silva Aug. 27, 2019, 5:36 a.m. UTC 
From: Alastair D'Silva <alastair@d-silva.org>

By adding offset to memmap before passing it in to clear_hwpoisoned_pages,
we hide a theoretically null memmap from the null check inside
clear_hwpoisoned_pages.

This patch passes the offset to clear_hwpoisoned_pages instead, allowing
memmap to successfully perform it's null check.

Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
---
 mm/sparse.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

Comments

Michal Hocko Aug. 27, 2019, 6:24 a.m. UTC | #1 
On Tue 27-08-19 15:36:55, Alastair D'Silva wrote:
> From: Alastair D'Silva <alastair@d-silva.org>
> 
> By adding offset to memmap before passing it in to clear_hwpoisoned_pages,
> we hide a theoretically null memmap from the null check inside
> clear_hwpoisoned_pages.

Isn't that other way around? Calculating the offset struct page pointer
will actually make the null check effective. Besides that I cannot
really see how pfn_to_page would return NULL. I have to confess that I
cannot really see how offset could lead to a NULL struct page either and
I strongly suspect that the NULL check is not really needed. Maybe it
used to be in the past.

> This patch passes the offset to clear_hwpoisoned_pages instead, allowing
> memmap to successfully perform it's null check.

I do not see any improvement in this patch. It just adds a new argument
unnecessarily.

> Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
> ---
>  mm/sparse.c | 9 +++++----
>  1 file changed, 5 insertions(+), 4 deletions(-)
> 
> diff --git a/mm/sparse.c b/mm/sparse.c
> index e41917a7e844..3ff84e627e58 100644
> --- a/mm/sparse.c
> +++ b/mm/sparse.c
> @@ -882,7 +882,7 @@ int __meminit sparse_add_section(int nid, unsigned long start_pfn,
>  }
>  
>  #ifdef CONFIG_MEMORY_FAILURE
> -static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
> +static void clear_hwpoisoned_pages(struct page *memmap, int start, int count)
>  {
>  	int i;
>  
> @@ -898,7 +898,7 @@ static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
>  	if (atomic_long_read(&num_poisoned_pages) == 0)
>  		return;
>  
> -	for (i = 0; i < nr_pages; i++) {
> +	for (i = start; i < start + count; i++) {
>  		if (PageHWPoison(&memmap[i])) {
>  			num_poisoned_pages_dec();
>  			ClearPageHWPoison(&memmap[i]);
> @@ -906,7 +906,8 @@ static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
>  	}
>  }
>  #else
> -static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
> +static inline void clear_hwpoisoned_pages(struct page *memmap, int start,
> +		int count)
>  {
>  }
>  #endif
> @@ -915,7 +916,7 @@ void sparse_remove_section(struct mem_section *ms, unsigned long pfn,
>  		unsigned long nr_pages, unsigned long map_offset,
>  		struct vmem_altmap *altmap)
>  {
> -	clear_hwpoisoned_pages(pfn_to_page(pfn) + map_offset,
> +	clear_hwpoisoned_pages(pfn_to_page(pfn), map_offset,
>  			nr_pages - map_offset);
>  	section_deactivate(pfn, nr_pages, altmap);
>  }
> -- 
> 2.21.0
>
Alastair D'Silva Aug. 27, 2019, 7 a.m. UTC | #2 
On Tue, 2019-08-27 at 08:24 +0200, Michal Hocko wrote:
> On Tue 27-08-19 15:36:55, Alastair D'Silva wrote:
> > From: Alastair D'Silva <alastair@d-silva.org>
> > 
> > By adding offset to memmap before passing it in to
> > clear_hwpoisoned_pages,
> > we hide a theoretically null memmap from the null check inside
> > clear_hwpoisoned_pages.
> 
> Isn't that other way around? Calculating the offset struct page
> pointer
> will actually make the null check effective. Besides that I cannot
> really see how pfn_to_page would return NULL. I have to confess that
> I
> cannot really see how offset could lead to a NULL struct page either
> and
> I strongly suspect that the NULL check is not really needed. Maybe it
> used to be in the past.
> 

You're probably right, but I didn't feel confident in removing the NULL
check. 

While the NULL check remains though, I can't see how adding the offset
would turn a non-NULL pointer into a NULL unless the pointer is invalid
in the first place, and if this is the case, we should have a comment
explaining this.

The NULL check was added in commit:
95a4774d055c ("memory-hotplug: update mce_bad_pages when removing the
memory")
where memmap was originally inited to NULL, and only conditionally
given a value.

With this in mind, since that situation is no longer true, I think we
could instead drop the NULL check.
David Hildenbrand Aug. 27, 2019, 7:08 a.m. UTC | #3 
On 27.08.19 09:00, Alastair D'Silva wrote:
> On Tue, 2019-08-27 at 08:24 +0200, Michal Hocko wrote:
>> On Tue 27-08-19 15:36:55, Alastair D'Silva wrote:
>>> From: Alastair D'Silva <alastair@d-silva.org>
>>>
>>> By adding offset to memmap before passing it in to
>>> clear_hwpoisoned_pages,
>>> we hide a theoretically null memmap from the null check inside
>>> clear_hwpoisoned_pages.
>>
>> Isn't that other way around? Calculating the offset struct page
>> pointer
>> will actually make the null check effective. Besides that I cannot
>> really see how pfn_to_page would return NULL. I have to confess that
>> I
>> cannot really see how offset could lead to a NULL struct page either
>> and
>> I strongly suspect that the NULL check is not really needed. Maybe it
>> used to be in the past.
>>
> 
> You're probably right, but I didn't feel confident in removing the NULL
> check. 
> 
> While the NULL check remains though, I can't see how adding the offset
> would turn a non-NULL pointer into a NULL unless the pointer is invalid
> in the first place, and if this is the case, we should have a comment
> explaining this.
> 
> The NULL check was added in commit:
> 95a4774d055c ("memory-hotplug: update mce_bad_pages when removing the
> memory")
> where memmap was originally inited to NULL, and only conditionally
> given a value.
> 
> With this in mind, since that situation is no longer true, I think we
> could instead drop the NULL check.
> 

Makes sense to me.
Michal Hocko Aug. 27, 2019, 7:22 a.m. UTC | #4 
On Tue 27-08-19 17:00:16, Alastair D'Silva wrote:
[...]
> The NULL check was added in commit:
> 95a4774d055c ("memory-hotplug: update mce_bad_pages when removing the
> memory")
> where memmap was originally inited to NULL, and only conditionally
> given a value.
> 
> With this in mind, since that situation is no longer true, I think we
> could instead drop the NULL check.

This would be much more preferable to the original patch.
diff mbox series

Patch

diff --git a/mm/sparse.c b/mm/sparse.c
index e41917a7e844..3ff84e627e58 100644
--- a/mm/sparse.c
+++ b/mm/sparse.c
@@ -882,7 +882,7 @@  int __meminit sparse_add_section(int nid, unsigned long start_pfn,
 }
 
 #ifdef CONFIG_MEMORY_FAILURE
-static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
+static void clear_hwpoisoned_pages(struct page *memmap, int start, int count)
 {
 	int i;
 
@@ -898,7 +898,7 @@  static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
 	if (atomic_long_read(&num_poisoned_pages) == 0)
 		return;
 
-	for (i = 0; i < nr_pages; i++) {
+	for (i = start; i < start + count; i++) {
 		if (PageHWPoison(&memmap[i])) {
 			num_poisoned_pages_dec();
 			ClearPageHWPoison(&memmap[i]);
@@ -906,7 +906,8 @@  static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
 	}
 }
 #else
-static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
+static inline void clear_hwpoisoned_pages(struct page *memmap, int start,
+		int count)
 {
 }
 #endif
@@ -915,7 +916,7 @@  void sparse_remove_section(struct mem_section *ms, unsigned long pfn,
 		unsigned long nr_pages, unsigned long map_offset,
 		struct vmem_altmap *altmap)
 {
-	clear_hwpoisoned_pages(pfn_to_page(pfn) + map_offset,
+	clear_hwpoisoned_pages(pfn_to_page(pfn), map_offset,
 			nr_pages - map_offset);
 	section_deactivate(pfn, nr_pages, altmap);
 }