[PATCH] mm/page_owner: Record the timestamp of all pages during free

[PATCH] mm/page_owner: Record the timestamp of all pages during free

[Georgi Djakov]

Collect the time when each allocation is freed, to help with memory
analysis with kdump/ramdump.

Having another timestamp when we free the page helps for debugging
page migration issues. For example both alloc and free timestamps
being the same can gave hints that there is an issue with migrating
memory, as opposed to a page just being dropped during migration.

Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>
---
 Documentation/vm/page_owner.rst | 2 +-
 mm/page_owner.c                 | 5 +++++
 2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/Documentation/vm/page_owner.rst b/Documentation/vm/page_owner.rst
index 4e67c2e9bbed..5d7a62c2be28 100644
--- a/Documentation/vm/page_owner.rst
+++ b/Documentation/vm/page_owner.rst
@@ -47,7 +47,7 @@ size change due to this facility.
 
    text    data     bss     dec     hex filename
    48800   2445     644   51889    cab1 mm/page_alloc.o
-   6574     108      29    6711    1a37 mm/page_owner.o
+   6644     108      29    6777    1a79 mm/page_owner.o
    1025       8       8    1041     411 mm/page_ext.o
 
 Although, roughly, 8 KB code is added in total, page_alloc.o increase by
diff --git a/mm/page_owner.c b/mm/page_owner.c
index d15c7c4994f5..fbdf064e7494 100644
--- a/mm/page_owner.c
+++ b/mm/page_owner.c
@@ -27,6 +27,7 @@ struct page_owner {
 	depot_stack_handle_t handle;
 	depot_stack_handle_t free_handle;
 	u64 ts_nsec;
+	u64 free_ts_nsec;
 	pid_t pid;
 };
 
@@ -148,6 +149,7 @@ void __reset_page_owner(struct page *page, unsigned int order)
 	struct page_ext *page_ext;
 	depot_stack_handle_t handle = 0;
 	struct page_owner *page_owner;
+	u64 free_ts_nsec = local_clock();
 
 	handle = save_stack(GFP_NOWAIT | __GFP_NOWARN);
 
@@ -158,6 +160,7 @@ void __reset_page_owner(struct page *page, unsigned int order)
 		__clear_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags);
 		page_owner = get_page_owner(page_ext);
 		page_owner->free_handle = handle;
+		page_owner->free_ts_nsec = free_ts_nsec;
 		page_ext = page_ext_next(page_ext);
 	}
 }
@@ -177,6 +180,7 @@ static inline void __set_page_owner_handle(struct page *page,
 		page_owner->last_migrate_reason = -1;
 		page_owner->pid = current->pid;
 		page_owner->ts_nsec = local_clock();
+		page_owner->free_ts_nsec = 0;
 		__set_bit(PAGE_EXT_OWNER, &page_ext->flags);
 		__set_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags);
 
@@ -243,6 +247,7 @@ void __copy_page_owner(struct page *oldpage, struct page *newpage)
 	new_page_owner->handle = old_page_owner->handle;
 	new_page_owner->pid = old_page_owner->pid;
 	new_page_owner->ts_nsec = old_page_owner->ts_nsec;
+	new_page_owner->free_ts_nsec = old_page_owner->ts_nsec;
 
 	/*
 	 * We don't clear the bit on the oldpage as it's going to be freed

Re: [PATCH] mm/page_owner: Record the timestamp of all pages during free

[Vlastimil Babka]

On 1/30/21 12:45 AM, Georgi Djakov wrote:
> Collect the time when each allocation is freed, to help with memory
> analysis with kdump/ramdump.
> 
> Having another timestamp when we free the page helps for debugging
> page migration issues. For example both alloc and free timestamps
> being the same can gave hints that there is an issue with migrating
> memory, as opposed to a page just being dropped during migration.

Besides crash dump analysis, we should also provide this timestamp in the
page_owner file and dump_page()?

> Signed-off-by: Georgi Djakov <georgi.djakov@linaro.org>
> ---
>  Documentation/vm/page_owner.rst | 2 +-
>  mm/page_owner.c                 | 5 +++++
>  2 files changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/Documentation/vm/page_owner.rst b/Documentation/vm/page_owner.rst
> index 4e67c2e9bbed..5d7a62c2be28 100644
> --- a/Documentation/vm/page_owner.rst
> +++ b/Documentation/vm/page_owner.rst
> @@ -47,7 +47,7 @@ size change due to this facility.
>  
>     text    data     bss     dec     hex filename
>     48800   2445     644   51889    cab1 mm/page_alloc.o
> -   6574     108      29    6711    1a37 mm/page_owner.o
> +   6644     108      29    6777    1a79 mm/page_owner.o
>     1025       8       8    1041     411 mm/page_ext.o
>  
>  Although, roughly, 8 KB code is added in total, page_alloc.o increase by
> diff --git a/mm/page_owner.c b/mm/page_owner.c
> index d15c7c4994f5..fbdf064e7494 100644
> --- a/mm/page_owner.c
> +++ b/mm/page_owner.c
> @@ -27,6 +27,7 @@ struct page_owner {
>  	depot_stack_handle_t handle;
>  	depot_stack_handle_t free_handle;
>  	u64 ts_nsec;
> +	u64 free_ts_nsec;
>  	pid_t pid;
>  };
>  
> @@ -148,6 +149,7 @@ void __reset_page_owner(struct page *page, unsigned int order)
>  	struct page_ext *page_ext;
>  	depot_stack_handle_t handle = 0;
>  	struct page_owner *page_owner;
> +	u64 free_ts_nsec = local_clock();
>  
>  	handle = save_stack(GFP_NOWAIT | __GFP_NOWARN);
>  
> @@ -158,6 +160,7 @@ void __reset_page_owner(struct page *page, unsigned int order)
>  		__clear_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags);
>  		page_owner = get_page_owner(page_ext);
>  		page_owner->free_handle = handle;
> +		page_owner->free_ts_nsec = free_ts_nsec;
>  		page_ext = page_ext_next(page_ext);
>  	}
>  }
> @@ -177,6 +180,7 @@ static inline void __set_page_owner_handle(struct page *page,
>  		page_owner->last_migrate_reason = -1;
>  		page_owner->pid = current->pid;
>  		page_owner->ts_nsec = local_clock();
> +		page_owner->free_ts_nsec = 0;

Might be more useful not to reset the time of previous free when the page is
reallocated?

>  		__set_bit(PAGE_EXT_OWNER, &page_ext->flags);
>  		__set_bit(PAGE_EXT_OWNER_ALLOCATED, &page_ext->flags);
>  
> @@ -243,6 +247,7 @@ void __copy_page_owner(struct page *oldpage, struct page *newpage)
>  	new_page_owner->handle = old_page_owner->handle;
>  	new_page_owner->pid = old_page_owner->pid;
>  	new_page_owner->ts_nsec = old_page_owner->ts_nsec;
> +	new_page_owner->free_ts_nsec = old_page_owner->ts_nsec;
>  
>  	/*
>  	 * We don't clear the bit on the oldpage as it's going to be freed
>