maple_tree: Fix handle of invalidated state in mas_wr_store_setup()

Commit Message

Liam R. Howlett Jan. 9, 2023, 4:55 p.m. UTC

If an invalidated maple state is encountered during write, reset the
maple state to MAS_START.  This will result in a re-walk of the tree to
the correct location for the write.

Link: https://lore.kernel.org/all/[e-mail]/
Reported-by: SeongJae Park <[e-mail]>
Signed-off-by: Liam R. Howlett <[e-mail]>
---
 lib/maple_tree.c | 3 +++
 1 file changed, 3 insertions(+)

Comments

SeongJae Park Jan. 9, 2023, 7:22 p.m. UTC | #1

On Mon, 9 Jan 2023 16:55:42 +0000 Liam Howlett <[e-mail]> wrote:

> If an invalidated maple state is encountered during write, reset the
> maple state to MAS_START.  This will result in a re-walk of the tree to
> the correct location for the write.
> 
> Link: https://lore.kernel.org/all/[e-mail]/
> Reported-by: SeongJae Park <[e-mail]>
> Signed-off-by: Liam R. Howlett <[e-mail]>

Tested-by: SeongJae Park <[e-mail]>

I confirmed this patch fixes the boot failure issue.  Nevertheless, the 'make
install' issue on my system is still not fixed.


Thanks,
SJ

> ---
>  lib/maple_tree.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/lib/maple_tree.c b/lib/maple_tree.c
> index 14cff87cf058..59c4879ac714 100644
> --- a/lib/maple_tree.c
> +++ b/lib/maple_tree.c
> @@ -5620,6 +5620,9 @@ static inline void mte_destroy_walk(struct maple_enode *enode,
>  
>  static void mas_wr_store_setup(struct ma_wr_state *wr_mas)
>  {
> +	if (unlikely(mas_is_paused(wr_mas->mas)))
> +		mas_reset(wr_mas->mas);
> +
>  	if (!mas_is_start(wr_mas->mas)) {
>  		if (mas_is_none(wr_mas->mas)) {
>  			mas_reset(wr_mas->mas);
> -- 
> 2.35.1
>

Andrew Morton Jan. 9, 2023, 11:40 p.m. UTC | #2

On Mon,  9 Jan 2023 19:22:48 +0000 SeongJae Park <[e-mail]> wrote:

> On Mon, 9 Jan 2023 16:55:42 +0000 Liam Howlett <[e-mail]> wrote:
> 
> > If an invalidated maple state is encountered during write, reset the
> > maple state to MAS_START.  This will result in a re-walk of the tree to
> > the correct location for the write.

Can we think of a Fixes: for this?

> > Link: https://lore.kernel.org/all/[e-mail]/
> > Reported-by: SeongJae Park <[e-mail]>
> > Signed-off-by: Liam R. Howlett <[e-mail]>
> 
> Tested-by: SeongJae Park <[e-mail]>
> 
> I confirmed this patch fixes the boot failure issue.  Nevertheless, the 'make
> install' issue on my system is still not fixed.

Is this a mainline issue or an mm-unstable issue?

SeongJae Park Jan. 9, 2023, 11:43 p.m. UTC | #3

Hi Andrew,

On Mon, 9 Jan 2023 15:40:12 -0800 Andrew Morton <[e-mail]> wrote:

> On Mon,  9 Jan 2023 19:22:48 +0000 SeongJae Park <[e-mail]> wrote:
> 
> > On Mon, 9 Jan 2023 16:55:42 +0000 Liam Howlett <[e-mail]> wrote:
> > 
> > > If an invalidated maple state is encountered during write, reset the
> > > maple state to MAS_START.  This will result in a re-walk of the tree to
> > > the correct location for the write.
> 
> Can we think of a Fixes: for this?
> 
> > > Link: https://lore.kernel.org/all/[e-mail]/
> > > Reported-by: SeongJae Park <[e-mail]>
> > > Signed-off-by: Liam R. Howlett <[e-mail]>
> > 
> > Tested-by: SeongJae Park <[e-mail]>
> > 
> > I confirmed this patch fixes the boot failure issue.  Nevertheless, the 'make
> > install' issue on my system is still not fixed.
> 
> Is this a mainline issue or an mm-unstable issue?

It's mm-unstable issue:
https://lore.kernel.org/linux-mm/[e-mail]/


Thanks,
SJ

Liam R. Howlett Jan. 10, 2023, 2:21 p.m. UTC | #4

* SeongJae Park <[e-mail]> [230109 18:43]:
> Hi Andrew,
> 
> On Mon, 9 Jan 2023 15:40:12 -0800 Andrew Morton <[e-mail]> wrote:
> 
> > On Mon,  9 Jan 2023 19:22:48 +0000 SeongJae Park <[e-mail]> wrote:
> > 
> > > On Mon, 9 Jan 2023 16:55:42 +0000 Liam Howlett <[e-mail]> wrote:
> > > 
> > > > If an invalidated maple state is encountered during write, reset the
> > > > maple state to MAS_START.  This will result in a re-walk of the tree to
> > > > the correct location for the write.
> > 
> > Can we think of a Fixes: for this?

It wasn't really an issue until I wrote the vma iterator store
function.

You can put this before (mm-unstable git id..)
9df3d25a97ec ("mm: expand vma iterator interface")

> > 
> > > > Link: https://lore.kernel.org/all/[e-mail]/
> > > > Reported-by: SeongJae Park <[e-mail]>
> > > > Signed-off-by: Liam R. Howlett <[e-mail]>
> > > 
> > > Tested-by: SeongJae Park <[e-mail]>
> > > 
> > > I confirmed this patch fixes the boot failure issue.  Nevertheless, the 'make
> > > install' issue on my system is still not fixed.
> > 
> > Is this a mainline issue or an mm-unstable issue?
> 
> It's mm-unstable issue:
> https://lore.kernel.org/linux-mm/[e-mail]/

Yes, just mm-unstable.

Andrew Morton Jan. 10, 2023, 11:18 p.m. UTC | #5

On Tue, 10 Jan 2023 14:21:15 +0000 Liam Howlett <[e-mail]> wrote:

> * SeongJae Park <[e-mail]> [230109 18:43]:
> > Hi Andrew,
> > 
> > On Mon, 9 Jan 2023 15:40:12 -0800 Andrew Morton <[e-mail]> wrote:
> > 
> > > On Mon,  9 Jan 2023 19:22:48 +0000 SeongJae Park <[e-mail]> wrote:
> > > 
> > > > On Mon, 9 Jan 2023 16:55:42 +0000 Liam Howlett <[e-mail]> wrote:
> > > > 
> > > > > If an invalidated maple state is encountered during write, reset the
> > > > > maple state to MAS_START.  This will result in a re-walk of the tree to
> > > > > the correct location for the write.
> > > 
> > > Can we think of a Fixes: for this?
> 
> It wasn't really an issue until I wrote the vma iterator store
> function.
> 
> You can put this before (mm-unstable git id..)
> 9df3d25a97ec ("mm: expand vma iterator interface")

I have dropped "mm: expand vma iterator interface"'s series.  Should we
retain this patch as a standalone thing and is the changelog still
appropriate?

Liam R. Howlett Jan. 11, 2023, 1:07 a.m. UTC | #6

* Andrew Morton <[e-mail]> [230110 18:18]:
> On Tue, 10 Jan 2023 14:21:15 +0000 Liam Howlett <[e-mail]> wrote:
> 
> > * SeongJae Park <[e-mail]> [230109 18:43]:
> > > Hi Andrew,
> > > 
> > > On Mon, 9 Jan 2023 15:40:12 -0800 Andrew Morton <[e-mail]> wrote:
> > > 
> > > > On Mon,  9 Jan 2023 19:22:48 +0000 SeongJae Park <[e-mail]> wrote:
> > > > 
> > > > > On Mon, 9 Jan 2023 16:55:42 +0000 Liam Howlett <[e-mail]> wrote:
> > > > > 
> > > > > > If an invalidated maple state is encountered during write, reset the
> > > > > > maple state to MAS_START.  This will result in a re-walk of the tree to
> > > > > > the correct location for the write.
> > > > 
> > > > Can we think of a Fixes: for this?
> > 
> > It wasn't really an issue until I wrote the vma iterator store
> > function.
> > 
> > You can put this before (mm-unstable git id..)
> > 9df3d25a97ec ("mm: expand vma iterator interface")
> 
> I have dropped "mm: expand vma iterator interface"'s series.  Should we
> retain this patch as a standalone thing and is the changelog still
> appropriate?

No, just drop it and I'll re-spin v3 to include maple tree fixes on the
top.

Patch

diff --git a/lib/maple_tree.c b/lib/maple_tree.c
index 14cff87cf058..59c4879ac714 100644
--- a/lib/maple_tree.c
+++ b/lib/maple_tree.c
@@ -5620,6 +5620,9 @@  static inline void mte_destroy_walk(struct maple_enode *enode,
 
 static void mas_wr_store_setup(struct ma_wr_state *wr_mas)
 {
+	if (unlikely(mas_is_paused(wr_mas->mas)))
+		mas_reset(wr_mas->mas);
+
 	if (!mas_is_start(wr_mas->mas)) {
 		if (mas_is_none(wr_mas->mas)) {
 			mas_reset(wr_mas->mas);