Re: [PATCH v3] alloc_tag: check mem_profiling_support in alloc_tag_init

[Casey Chen <cachen@purestorage.com>]

On Mon, May 12, 2025 at 1:53 PM Suren Baghdasaryan <surenb@google.com> wrote:
>
> On Mon, May 12, 2025 at 12:42 PM Casey Chen <cachen@purestorage.com> wrote:
> >
> > If mem_profiling_support is false, for example by
> > sysctl.vm.mem_profiling=never, alloc_tag_init should skip
> > module tags allocation, codetag type registration and
> > procfs init.
> >
> > Signed-off-by: Casey Chen <cachen@purestorage.com>
> > Reviewed-by: Yuanyuan Zhong <yzhong@purestorage.com>
> > ---
> >  lib/alloc_tag.c | 34 +++++++++++++++++++---------------
> >  1 file changed, 19 insertions(+), 15 deletions(-)
> >
> > diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
> > index 25ecc1334b67..7c798778669c 100644
> > --- a/lib/alloc_tag.c
> > +++ b/lib/alloc_tag.c
> > @@ -244,17 +244,6 @@ static void shutdown_mem_profiling(bool remove_file)
> >         mem_profiling_support = false;
> >  }
> >
> > -static void __init procfs_init(void)
> > -{
> > -       if (!mem_profiling_support)
> > -               return;
> > -
> > -       if (!proc_create_seq(ALLOCINFO_FILE_NAME, 0400, NULL, &allocinfo_seq_op)) {
> > -               pr_err("Failed to create %s file\n", ALLOCINFO_FILE_NAME);
> > -               shutdown_mem_profiling(false);
> > -       }
> > -}
> > -
> >  void __init alloc_tag_sec_init(void)
> >  {
> >         struct alloc_tag *last_codetag;
> > @@ -762,19 +751,34 @@ static int __init alloc_tag_init(void)
> >         };
> >         int res;
> >
> > +       sysctl_init();
> > +
> > +       if (!mem_profiling_support) {
> > +               pr_info("Memory allocation profiling is not supported!\n");
> > +               return 0;
> > +       }
> > +
> > +       if (!proc_create_seq(ALLOCINFO_FILE_NAME, 0400, NULL, &allocinfo_seq_op)) {
>
> I suggest keeping file creation as the last step so that we don't have
> to remove it if later steps fail.
>

If the file creation is the last step, how could it clean up codetag
type registration ? As we miss codetag unregistration function, I
think codetag type registration should be the last step.

> > +               pr_err("Failed to create %s file\n", ALLOCINFO_FILE_NAME);
> > +               shutdown_mem_profiling(false);
> > +               return -EAGAIN;
>
> Why EAGAIN? I would just let it fall through returning 0 (after moving
> this to the end as I suggested in my previous comment).
>

return 0 is fine to me if this fails. EAGAIN sounds fine too.

> > +       }
> > +
> >         res = alloc_mod_tags_mem();
> > -       if (res)
> > +       if (res) {
> > +               pr_err("Failed to allocate module tags\n");
>
> The error message is a bit misleading here since we are not allocating
> module tags but reserving virtual space for them. Maybe:
> "Failed to reserve address space for module tags"
>

Sounds good!

> > +               shutdown_mem_profiling(true);
> >                 return res;
> > +       }
> >
> >         alloc_tag_cttype = codetag_register_type(&desc);
> >         if (IS_ERR(alloc_tag_cttype)) {
> > +               pr_err("Failed to register codetag type with desc %s\n", desc.section);
>
> pr_err("Allocation tags registration failed, errno = %d\n", ret);
>

Sounds good too.

> >                 free_mod_tags_mem();
> > +               shutdown_mem_profiling(true);
> >                 return PTR_ERR(alloc_tag_cttype);
> >         }
> >
> > -       sysctl_init();
> > -       procfs_init();
> > -
> >         return 0;
> >  }
> >  module_init(alloc_tag_init);
> > --
> > 2.49.0
> >