On Fri, 15 Mar 2019, Oscar Salvador wrote:
> On Fri, Mar 15, 2019 at 01:47:33PM +0100, Michal Hocko wrote:
> > diff --git a/mm/debug.c b/mm/debug.c
> > index 1611cf00a137..499c26d5ebe5 100644
> > --- a/mm/debug.c
> > +++ b/mm/debug.c
> > @@ -78,6 +78,9 @@ void __dump_page(struct page *page, const char *reason)
> > else if (PageKsm(page))
> > pr_warn("ksm ");
> > else if (mapping) {
> > + if (PageSwapCache(page))
> > + mapping = page_swap_info(page)->swap_file->f_mapping;
> > +
> > pr_warn("%ps ", mapping->a_ops);
> > if (mapping->host->i_dentry.first) {
> > struct dentry *dentry;
>
> This looks like a much nicer fix, indeed.
> I gave it a spin and it works.
>
> Since the mapping is set during the swapon, I would assume that this should
> always work for swap.
> Although I am not sure if once you start playing with e.g zswap the picture can
> change.
>
> Let us wait for Hugh and Jan.
>
> Thanks Michal
Sorry, I don't agree that Michal's more sophisticated patch is nicer:
the appropriate patch was your original, just checking for NULL.
Though, would I be too snarky to suggest that your patch description
would be better at 2 lines than 90? Swap mapping->host is NULL,
so of course __dump_page() needs to be careful about that.
I was a little disturbed to see __dump_page() now getting into dentries,
but admit that it can sometimes be very helpful to see the name of the
file involved; so if that is not in danger of breaking anything, okay.
It is very often useful to see if a page is PageSwapCache (typically
because that should account for 1 of its refcount); I cannot think of
a time when it's been useful to know the name of the underlying swap
device (if that's indeed what f_mapping leads to: it's new to me).
And if you need swp_type and swp_offset, they're in the raw output.
The cleverer __dump_page() tries to get, the more likely that it will
itself crash just when you need it most. Please just keep it simple.
Thanks,
Hugh
