* Re: [BUG] Swap xarray workingset eviction warning.
2018-07-02 2:50 ` [BUG] Swap xarray workingset eviction warning Matthew Wilcox
@ 2018-07-02 3:11 ` Gao Xiang
2018-07-05 17:00 ` Johannes Weiner
1 sibling, 0 replies; 5+ messages in thread
From: Gao Xiang @ 2018-07-02 3:11 UTC (permalink / raw)
To: Matthew Wilcox, Peter Geis; +Cc: linux-fsdevel, Johannes Weiner, linux-mm
Hi Matthew,
On 2018/7/2 10:50, Matthew Wilcox wrote:
> On Sun, Jul 01, 2018 at 07:09:41PM -0400, Peter Geis wrote:
>> The warning is as follows:
>> [10409.408904] ------------[ cut here ]------------
>> [10409.408912] WARNING: CPU: 0 PID: 38 at ./include/linux/xarray.h:53
>> workingset_eviction+0x14c/0x154
> This is interesting. Here's the code that leads to the warning:
>
> static void *pack_shadow(int memcgid, pg_data_t *pgdat, unsigned long eviction)
> {
> eviction >>= bucket_order;
> eviction = (eviction << MEM_CGROUP_ID_SHIFT) | memcgid;
> eviction = (eviction << NODES_SHIFT) | pgdat->node_id;
>
> return xa_mk_value(eviction);
> }
>
> The warning itself comes from:
>
> static inline void *xa_mk_value(unsigned long v)
> {
> WARN_ON((long)v < 0);
> return (void *)((v << 1) | 1);
> }
Sorry for breaking in, how do you think about considering
[RFC PATCH v4] <linux/tagptr.h>: Introduce tagged pointer
https://marc.info/?l=linux-kernel&m=153035209012070&w=2
to replace these masks? It seems boths for the XArray or old radix trees has many hacked code...
or if you think this implmentation is not ok, could you please give some suggestions or alternatives on tagptr...
>
> The fact that we haven't seen this on other architectures makes me wonder
> if NODES_SHIFT or MEM_CGROUP_ID_SHIFT are messed up on Tegra?
>
> Johannes, I wonder if you could help out here? I'm not terribly familiar
> with this part of the workingset code.
Thanks,
Gao Xiang
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: [BUG] Swap xarray workingset eviction warning.
2018-07-02 2:50 ` [BUG] Swap xarray workingset eviction warning Matthew Wilcox
2018-07-02 3:11 ` Gao Xiang
@ 2018-07-05 17:00 ` Johannes Weiner
2018-07-05 17:53 ` Matthew Wilcox
1 sibling, 1 reply; 5+ messages in thread
From: Johannes Weiner @ 2018-07-05 17:00 UTC (permalink / raw)
To: Matthew Wilcox; +Cc: Peter Geis, linux-fsdevel, linux-mm
Hello,
On Sun, Jul 01, 2018 at 07:50:59PM -0700, Matthew Wilcox wrote:
> On Sun, Jul 01, 2018 at 07:09:41PM -0400, Peter Geis wrote:
> > The warning is as follows:
> > [10409.408904] ------------[ cut here ]------------
> > [10409.408912] WARNING: CPU: 0 PID: 38 at ./include/linux/xarray.h:53
> > workingset_eviction+0x14c/0x154
>
> This is interesting. Here's the code that leads to the warning:
>
> static void *pack_shadow(int memcgid, pg_data_t *pgdat, unsigned long eviction)
> {
> eviction >>= bucket_order;
> eviction = (eviction << MEM_CGROUP_ID_SHIFT) | memcgid;
> eviction = (eviction << NODES_SHIFT) | pgdat->node_id;
>
> return xa_mk_value(eviction);
> }
>
> The warning itself comes from:
>
> static inline void *xa_mk_value(unsigned long v)
> {
> WARN_ON((long)v < 0);
> return (void *)((v << 1) | 1);
> }
>
> The fact that we haven't seen this on other architectures makes me wonder
> if NODES_SHIFT or MEM_CGROUP_ID_SHIFT are messed up on Tegra?
>
> Johannes, I wonder if you could help out here? I'm not terribly familiar
> with this part of the workingset code.
This could be a matter of uptime, but the warning triggers on a thing
that is supposed to happen everywhere eventually. Let's fix it.
The eviction timestamp is from a monotonically increasing counter that
will eventually reach values high enough that the left-shifts for
memcg id and node id will truncate the upper bits.
The bucketing logic isn't supposed to prevent this truncation, it's
just making sure that the namespace of the truncated timestamp field
is big enough to cover the full range of actionable refault pages.
xa_mk_value() doesn't understand that we're okay with it chopping off
our upper-most bit. We shouldn't make this an API behavior, either, so
let's fix the workingset code to always clear those bits before hand.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
---
diff --git a/mm/workingset.c b/mm/workingset.c
index a466e731231d..1da19c04b6f7 100644
--- a/mm/workingset.c
+++ b/mm/workingset.c
@@ -173,6 +173,7 @@ static unsigned int bucket_order __read_mostly;
static void *pack_shadow(int memcgid, pg_data_t *pgdat, unsigned long eviction)
{
eviction >>= bucket_order;
+ eviction &= EVICTION_MASK;
eviction = (eviction << MEM_CGROUP_ID_SHIFT) | memcgid;
eviction = (eviction << NODES_SHIFT) | pgdat->node_id;
^ permalink raw reply related [flat|nested] 5+ messages in thread