Message-ID: <CAJuCfpE670s5=QAbuqCLB3XuOkfL=L44r93cwJdrhHn=bYNd-Q@mail.gmail.com>
Date: Fri, 9 May 2025 21:03:06 -0700
From: Suren Baghdasaryan <surenb@...gle.com>
To: David Wang <00107082@....com>
Cc: Tim Chen <tim.c.chen@...ux.intel.com>, kent.overstreet@...ux.dev, 
	akpm@...ux-foundation.org, linux-kernel@...r.kernel.org, linux-mm@...ck.org
Subject: Re: [PATCH v2 2/2] alloc_tag: keep codetag iterator active between
 read() calls

On Fri, May 9, 2025 at 8:58 PM David Wang <00107082@....com> wrote:
>
>
>
> At 2025-05-10 11:30:50, "Suren Baghdasaryan" <surenb@...gle.com> wrote:
> >On Fri, May 9, 2025 at 8:10 PM David Wang <00107082@....com> wrote:
> >>
> >>
> >> At 2025-05-10 05:15:43, "Suren Baghdasaryan" <surenb@...gle.com> wrote:
> >> >On Fri, May 9, 2025 at 1:46 PM Suren Baghdasaryan <surenb@...gle.com> wrote:
> >> >>
> >> >> On Fri, May 9, 2025 at 12:46 PM Tim Chen <tim.c.chen@...ux.intel.com> wrote:
> >> >> >
> >> >> > On Fri, 2025-05-09 at 12:36 -0700, Suren Baghdasaryan wrote:
> >> >> > > On Fri, May 9, 2025 at 11:33 AM Tim Chen <tim.c.chen@...ux.intel.com> wrote:
> >> >> > > >
> >> >> > > > On Sat, 2025-05-10 at 01:39 +0800, David Wang wrote:
> >> >> > > > >
> >> >> > > > >
> >> >> > > > > Signed-off-by: David Wang <00107082@....com>
> >> >> > >
> >> >> > > Acked-by: Suren Baghdasaryan <surenb@...gle.com>
> >> >> > >
> >> >> > > > > ---
> >> >> > > > >  lib/alloc_tag.c | 29 ++++++++++-------------------
> >> >> > > > >  1 file changed, 10 insertions(+), 19 deletions(-)
> >> >> > > > >
> >> >> > > > > diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
> >> >> > > > > index 25ecc1334b67..fdd5887769a6 100644
> >> >> > > > > --- a/lib/alloc_tag.c
> >> >> > > > > +++ b/lib/alloc_tag.c
> >> >> > > > > @@ -45,21 +45,16 @@ struct allocinfo_private {
> >> >> > > > >  static void *allocinfo_start(struct seq_file *m, loff_t *pos)
> >> >> > > > >  {
> >> >> > > > >       struct allocinfo_private *priv;
> >> >> > > > > -     struct codetag *ct;
> >> >> > > > >       loff_t node = *pos;
> >> >> > > > >
> >> >> > > > > -     priv = kzalloc(sizeof(*priv), GFP_KERNEL);
> >> >> > > > > -     m->private = priv;
> >> >> > > > > -     if (!priv)
> >> >> > > > > -             return NULL;
> >> >> > > > > -
> >> >> > > > > -     priv->print_header = (node == 0);
> >> >> > > > > +     priv = (struct allocinfo_private *)m->private;
> >> >> > > > >       codetag_lock_module_list(alloc_tag_cttype, true);
> >> >> > > > > -     priv->iter = codetag_get_ct_iter(alloc_tag_cttype);
> >> >> > > > > -     while ((ct = codetag_next_ct(&priv->iter)) != NULL && node)
> >> >> > > > > -             node--;
> >> >> > > > > -
> >> >> > > > > -     return ct ? priv : NULL;
> >> >> > > > > +     if (node == 0) {
> >> >> > > > > +             priv->print_header = true;
> >> >> > > > > +             priv->iter = codetag_get_ct_iter(alloc_tag_cttype);
> >> >> > > > > +             codetag_next_ct(&priv->iter);
> >> >> > > > > +     }
> >> >> > > >
> >> >> > > > Do you need to skip print header when *pos != 0? i.e add
> >> >> > >
> >> >> > > Technically not needed since proc_create_seq_private() allocates
> >> >> > > seq->private using kzalloc(), so the initial value of
> >> >> > > priv->print_header is always false.
> >> >> >
> >> >> > But we'll start with first call to allocinfo_start() with *pos == 0,
> >> >>
> >> >> Usually but not always if we do lseek() to a non-zero position beforehand.
> >> >
> >> >Actually, this change will break the lseek() case. We can't always
> >> >assume that we start reading from *pos == 0. Current patch will fail
> >> >to initialize priv if we start reading with *pos != 0.
> >> >priv->iter should be tracking current position and allocinfo_start()
> >> >should detect a mismatch between *pos and iter->pos and re-walk the
> >> >tags if there was a position change.
> >>
> >> seq_file works line by line,  I think even if it support lseek, seq_file would still start with line #0,
> >> since seq_file have on clue the byte size for each line.
> >>
> >> I will check the code,  make some tests and update later.
> >
> >Ah, yes. You are correct.
> >seq_lseek() will traverse restarting from 0:
> >https://elixir.bootlin.com/linux/v6.14.6/source/fs/seq_file.c#L323.
> >Position jumps are similarly handled with traversal from 0:
> >https://elixir.bootlin.com/linux/v6.14.6/source/fs/seq_file.c#L194.
> >
>
> Actually I was expecting EOPNOTSUPP when lseek on seq files, surprised to see it works...... :)
>
> If seq_file somehow skips start(0),  then nothing would be displayed since
> priv->iter.ct would be 0 and `return priv->iter.ct ? priv : NULL;` would return NULL;
> But I think that case  would be  seq_file's bug,   starting with 0 is kind of protocol promised by seq_file.

Yes, I think that is the correct assumption. Even pread() with a jump
in position should work the same way, restarting at 0.

>
>
> >>
>
> >>
> >> >
> >> >>
> >> >> > then print_header will be initialized to true.
> >> >>
> >> >> After the first call to allocinfo_show() print_header will be reset
> >> >> back to false.
> >> >>
> >> >> > Will there be subsequent calls of allocinfo_start() with *pos !=0,
> >> >> > but priv->print_header stays at 0?
> >> >>
> >> >> Yes, there will be subsequent calls to allocinfo_start() with *pos !=0
> >> >> and priv->print_header=false, which is what we want, right? We want to
> >> >> print the header only at the beginning of the file (node == 0).
> >> >>
> >> >> >
> >> >> > Tim
> >> >> > >
> >> >> > > >
> >> >> > > >         } else {
> >> >> > > >                 priv->print_header = false;
> >> >> > > >         }
> >> >> > > >
> >> >> > > > Tim
> >> >> > > >
> >> >> > > > > +     return priv->iter.ct ? priv : NULL;
> >> >> > > > >  }
> >> >> > > > >
> >> >> > > >
> >> >> >

Powered by blists - more mailing lists