Re: [PATCH 2/2] mm, page_owner: Skip unnecessary stack_trace entries
From: Prakash Gupta
Date: Fri Sep 01 2017 - 01:00:38 EST
On 8/31/2017 1:04 PM, Vlastimil Babka wrote:
On 08/30/2017 09:32 AM, Prakash Gupta wrote:
The page_owner stacktrace always begin as follows:
[<ffffff987bfd48f4>] save_stack+0x40/0xc8
[<ffffff987bfd4da8>] __set_page_owner+0x3c/0x6c
Hmm, on x86_64 it looks like this:
save_stack_trace+0x16/0x20
save_stack+0x43/0xe0
__set_page_owner+0x24/0x50
So after your patch there's still __set_page_owner. Seems x86 needs
something similar to your arm64 patch 1/2?
Yes, that's correct.
These two entries do not provide any useful information and limits the
available stacktrace depth. The page_owner stacktrace was skipping caller
function from stack entries but this was missed with commit f2ca0b557107
("mm/page_owner: use stackdepot to store stacktrace")
Example page_owner entry after the patch:
Page allocated via order 0, mask 0x8(ffffff80085fb714)
PFN 654411 type Movable Block 639 type CMA Flags 0x0(ffffffbe5c7f12c0)
[<ffffff9b64989c14>] post_alloc_hook+0x70/0x80
...
[<ffffff9b651216e8>] msm_comm_try_state+0x5f8/0x14f4
[<ffffff9b6512486c>] msm_vidc_open+0x5e4/0x7d0
[<ffffff9b65113674>] msm_v4l2_open+0xa8/0x224
Fixes: f2ca0b557107 ("mm/page_owner: use stackdepot to store stacktrace")
Signed-off-by: Prakash Gupta <guptap@xxxxxxxxxxxxxx>
The patch itself improves the output regardless of whether we fix the
x86 internals, so:
Acked-by: Vlastimil Babka <vbabka@xxxxxxx>
Thanks.
---
mm/page_owner.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/mm/page_owner.c b/mm/page_owner.c
index 10d16fc45bd9..75b7c39bf1df 100644
--- a/mm/page_owner.c
+++ b/mm/page_owner.c
@@ -139,7 +139,7 @@ static noinline depot_stack_handle_t save_stack(gfp_t flags)
.nr_entries = 0,
.entries = entries,
.max_entries = PAGE_OWNER_STACK_DEPTH,
- .skip = 0
+ .skip = 2
};
depot_stack_handle_t handle;