All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH v2] mm/page_owner: Record the timestamp of all pages during free
@ 2021-02-03 17:59 Georgi Djakov
  2021-02-05 12:02 ` Vlastimil Babka
  0 siblings, 1 reply; 2+ messages in thread
From: Georgi Djakov @ 2021-02-03 17:59 UTC (permalink / raw)
  To: akpm, linux-mm; +Cc: vbabka, corbet, linux-doc, linux-kernel, georgi.djakov

Collect the time when each allocation is freed, to help with memory
analysis with kdump/ramdump. Add the timestamp also in the page_owner
debugfs file and print it in dump_page().

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>
---
v2:
 * Don't reset the time of previous free when page is reallocated. (Vlastimil)
 * Add the timestamp in the page_owner file and dump_page(). (Vlastimil)

v1: https://lore.kernel.org/r/20210129234548.10054-1-georgi.djakov@linaro.org

 Documentation/vm/page_owner.rst |  2 +-
 mm/page_owner.c                 | 12 ++++++++----
 2 files changed, 9 insertions(+), 5 deletions(-)

diff --git a/Documentation/vm/page_owner.rst b/Documentation/vm/page_owner.rst
index 4e67c2e9bbed..2175465c9bf2 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
+   6662     108      29    6799    1a8f 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..9ceae81ba7b8 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);
 	}
 }
@@ -243,6 +246,7 @@ 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
@@ -356,10 +360,10 @@ print_page_owner(char __user *buf, size_t count, unsigned long pfn,
 		return -ENOMEM;
 
 	ret = snprintf(kbuf, count,
-			"Page allocated via order %u, mask %#x(%pGg), pid %d, ts %llu ns\n",
+			"Page allocated via order %u, mask %#x(%pGg), pid %d, ts %llu ns, free_ts %llu ns\n",
 			page_owner->order, page_owner->gfp_mask,
 			&page_owner->gfp_mask, page_owner->pid,
-			page_owner->ts_nsec);
+			page_owner->ts_nsec, page_owner->free_ts_nsec);
 
 	if (ret >= count)
 		goto err;
@@ -435,9 +439,9 @@ void __dump_page_owner(struct page *page)
 	else
 		pr_alert("page_owner tracks the page as freed\n");
 
-	pr_alert("page last allocated via order %u, migratetype %s, gfp_mask %#x(%pGg), pid %d, ts %llu\n",
+	pr_alert("page last allocated via order %u, migratetype %s, gfp_mask %#x(%pGg), pid %d, ts %llu, free_ts %llu\n",
 		 page_owner->order, migratetype_names[mt], gfp_mask, &gfp_mask,
-		 page_owner->pid, page_owner->ts_nsec);
+		 page_owner->pid, page_owner->ts_nsec, page_owner->free_ts_nsec);
 
 	handle = READ_ONCE(page_owner->handle);
 	if (!handle) {

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [PATCH v2] mm/page_owner: Record the timestamp of all pages during free
  2021-02-03 17:59 [PATCH v2] mm/page_owner: Record the timestamp of all pages during free Georgi Djakov
@ 2021-02-05 12:02 ` Vlastimil Babka
  0 siblings, 0 replies; 2+ messages in thread
From: Vlastimil Babka @ 2021-02-05 12:02 UTC (permalink / raw)
  To: Georgi Djakov, akpm, linux-mm; +Cc: corbet, linux-doc, linux-kernel

On 2/3/21 6:59 PM, Georgi Djakov wrote:
> Collect the time when each allocation is freed, to help with memory
> analysis with kdump/ramdump. Add the timestamp also in the page_owner
> debugfs file and print it in dump_page().
> 
> 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>

Acked-by: Vlastimil Babka <vbabka@suse.cz>

Thanks.

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2021-02-05 12:07 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-02-03 17:59 [PATCH v2] mm/page_owner: Record the timestamp of all pages during free Georgi Djakov
2021-02-05 12:02 ` Vlastimil Babka

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.