[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <311acc28-50fd-4acd-b5e5-fa03ae4ab2cd@lucifer.local>
Date: Fri, 4 Oct 2024 15:58:00 +0100
From: Lorenzo Stoakes <lorenzo.stoakes@...cle.com>
To: Bert Karwatzki <spasswolf@....de>
Cc: "Liam R . Howlett" <Liam.Howlett@...cle.com>,
Andrew Morton <akpm@...ux-foundation.org>, linux-mm@...ck.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH v8 14/21] mm/mmap: Avoid zeroing vma tree in mmap_region()
On Fri, Oct 04, 2024 at 03:32:28PM +0100, Lorenzo Stoakes wrote:
> On Fri, Oct 04, 2024 at 03:26:26PM +0100, Lorenzo Stoakes wrote:
> > On Fri, Oct 04, 2024 at 03:23:37PM +0100, Lorenzo Stoakes wrote:
> > > On Fri, Oct 04, 2024 at 11:35:44AM +0200, Bert Karwatzki wrote:
> > > > Here's the log procduced by this kernel:
> > > >
> > > > c9e7f76815d3 (HEAD -> maple_tree_debug_4) hack: set of info stuff v5
> > > > 7e3bb072761a mm: correct error handling in mmap_region()
> > > > 77df9e4bb222 (tag: next-20241001, origin/master, origin/HEAD, master) Add linux-next specific files for 20241001
> > > >
> > > > Again it took two attempts to trigger the bug.
> > > >
> > > > Bert Karwatzki
> > > >
> > >
[snip]
OK sorry to keep on updating so quick, but been busy testing out the fix.
This version passes all tests so shooould work generally.
I'll do a proper write-up, etc. but generally it's because we can have a
condition where one node is empty other than the end node that is to be
overwritten, but we weren't being careful about that on a spanning store,
so were merging this AND writing in the new entry resulting in the observed
duplicate.
You need very specific circumstances to trigger this which the MAP_FIXED
series + steam seemed to provide :)
Please give this a try and let me know how it works.
Huge thanks for all your help throughout this! :)
----8<----
>From 3d01e0f97a2a62992b5f2bd466908628ae72da95 Mon Sep 17 00:00:00 2001
From: Lorenzo Stoakes <lorenzo.stoakes@...cle.com>
Date: Fri, 4 Oct 2024 15:18:58 +0100
Subject: [PATCH] fix v3
This now passes all maple tree tests, so should be good to go...
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@...cle.com>
---
lib/maple_tree.c | 16 ++++++++++++----
1 file changed, 12 insertions(+), 4 deletions(-)
diff --git a/lib/maple_tree.c b/lib/maple_tree.c
index 37abf0fe380b..4c0bb56a3142 100644
--- a/lib/maple_tree.c
+++ b/lib/maple_tree.c
@@ -3527,6 +3527,7 @@ static bool mas_wr_walk(struct ma_wr_state *wr_mas)
return true;
}
+// Return whether the node actually contains entries at or greater than wr_mas->mas->index.
static bool mas_wr_walk_index(struct ma_wr_state *wr_mas)
{
struct ma_state *mas = wr_mas->mas;
@@ -3535,8 +3536,11 @@ static bool mas_wr_walk_index(struct ma_wr_state *wr_mas)
mas_wr_walk_descend(wr_mas);
wr_mas->content = mas_slot_locked(mas, wr_mas->slots,
mas->offset);
- if (ma_is_leaf(wr_mas->type))
- return true;
+ if (ma_is_leaf(wr_mas->type)) {
+ unsigned long pivot = wr_mas->pivots[mas->offset];
+
+ return pivot == 0 || mas->index <= pivot;
+ }
mas_wr_walk_traverse(wr_mas);
}
@@ -3696,6 +3700,7 @@ static noinline void mas_wr_spanning_store(struct ma_wr_state *wr_mas)
struct maple_big_node b_node;
struct ma_state *mas;
unsigned char height;
+ bool r_populated;
/* Left and Right side of spanning store */
MA_STATE(l_mas, NULL, 0, 0);
@@ -3737,7 +3742,7 @@ static noinline void mas_wr_spanning_store(struct ma_wr_state *wr_mas)
r_mas.last++;
r_mas.index = r_mas.last;
- mas_wr_walk_index(&r_wr_mas);
+ r_populated = mas_wr_walk_index(&r_wr_mas);
r_mas.last = r_mas.index = mas->last;
/* Set up left side. */
@@ -3759,9 +3764,12 @@ static noinline void mas_wr_spanning_store(struct ma_wr_state *wr_mas)
memset(&b_node, 0, sizeof(struct maple_big_node));
/* Copy l_mas and store the value in b_node. */
+
mas_store_b_node(&l_wr_mas, &b_node, l_mas.end);
+
/* Copy r_mas into b_node. */
- if (r_mas.offset <= r_mas.end)
+
+ if (r_populated && r_mas.offset <= r_mas.end)
mas_mab_cp(&r_mas, r_mas.offset, r_mas.end,
&b_node, b_node.b_end + 1);
else
--
2.46.2
Powered by blists - more mailing lists