Re: [PATCH v2] mm: page_alloc: dump migrate-failed pages

From: Minchan Kim
Date: Wed Mar 10 2021 - 11:07:01 EST


On Wed, Mar 10, 2021 at 02:07:05PM +0100, Michal Hocko wrote:
> On Tue 09-03-21 23:42:46, Minchan Kim wrote:
> > On Tue, Mar 09, 2021 at 08:15:41AM -0800, Minchan Kim wrote:
> >
> > < snip >
> >
> > > > [...]
> > > > > +void dump_migrate_failure_pages(struct list_head *page_list)
> > > > > +{
> > > > > + DEFINE_DYNAMIC_DEBUG_METADATA(descriptor,
> > > > > + "migrate failure");
> > > > > + if (DYNAMIC_DEBUG_BRANCH(descriptor) &&
> > > > > + alloc_contig_ratelimit()) {
> > > > > + struct page *page;
> > > > > +
> > > > > + WARN(1, "failed callstack");
> > > > > + list_for_each_entry(page, page_list, lru)
> > > > > + dump_page(page, "migration failure");
> > > > > + }
> > > >
> > > > Apart from the above, do we have to warn for something that is a
> > > > debugging aid? A similar concern wrt dump_page which uses pr_warn and
> > >
> > > Make sense.
> > >
> > > > page owner is using even pr_alert.
> > > > Would it make sense to add a loglevel parameter both into __dump_page
> > > > and dump_page_owner?
> > >
> > > Let me try it.
> >
> > I looked though them and made first draft to clean them up.
> >
> > It's bigger than my initial expectaion because there are many callsites
> > to use dump_page and stack_trace_print inconsistent loglevel.
> > Since it's not a specific problem for this work, I'd like to deal with
> > it as separate patchset since I don't want to be stuck on here for my
> > initial goal.
> >
> > FYI,
> >
> > Subject: [RFC 0/5] make dump_page aware of loglevel
> >
> > - Forked from [1]
> >
> > dump_page uses __dump_page and dump_page_owner internally to
> > print various information. However, their printk loglevel are
> > inconsistent in that
> >
> > __dump_page: KERN_WARNING
> > __dump_page_owner: KERN_ALERT
> > stack_trace_print: KERN_DEFAULT
> >
> > To make them consistent from dump_page, this patch introduces
> > pr_loglevel in printk and make the utility functions aware of
> > loglevel. Finally, last patch changes dump_page to support
> > loglevel to make the printing level consistent.
> >
> > [1] https://lore.kernel.org/linux-mm/YEdAw6gnp9XxoWUQ@xxxxxxxxxxxxxx/
> >
> > Minchan Kim (5):
> > mm: introduce pr_loglevel for __dump_[page]_owner
> > stacktrace: stack_trace_print aware of loglevel
> > mm: page_owner: dump_page_owner aware of loglevel
> > mm: debug: __dump_page aware of loglevel
> > mm: debug: dump_page aware of loglevel
> > drivers/md/dm-bufio.c | 2 +-
> > drivers/virtio/virtio_mem.c | 2 +-
> > fs/btrfs/ref-verify.c | 2 +-
> > fs/fuse/dev.c | 2 +-
> > include/linux/mmdebug.h | 10 ++++++----
> > include/linux/page_owner.h | 8 ++++----
> > include/linux/printk.h | 12 +++++++++++
> > include/linux/stacktrace.h | 4 ++--
> > kernel/backtracetest.c | 2 +-
> > kernel/dma/debug.c | 3 ++-
> > kernel/kcsan/report.c | 7 ++++---
> > kernel/locking/lockdep.c | 3 ++-
> > kernel/stacktrace.c | 5 +++--
> > mm/debug.c | 40 ++++++++++++++++++-------------------
> > mm/filemap.c | 2 +-
> > mm/gup_test.c | 4 ++--
> > mm/huge_memory.c | 4 ++--
> > mm/kasan/report.c | 4 ++--
> > mm/kfence/report.c | 3 ++-
> > mm/kmemleak.c | 2 +-
> > mm/memory.c | 2 +-
> > mm/memory_hotplug.c | 4 ++--
> > mm/page_alloc.c | 4 ++--
> > mm/page_isolation.c | 2 +-
> > mm/page_owner.c | 24 +++++++++++-----------
> > 25 files changed, 88 insertions(+), 69 deletions(-)
>
> The is a lot of churn indeed. Have you considered adding $FOO_lglvl
> variants for those so that you can use them for your particular case
> without affecting most of existing users? Something similar we have
> discussed in other email thread regarding lru_add_drain_all?

I thought that way but didn't try since it couldn't make them
atomic(For example, other printk place in other context will
affect by the $FOO_lglvl).