Re: [PATCH v2] mm: cma: print cma name as well in cma_alloc debug

[Pintu Agarwal]

On Fri, 7 Jul 2023 at 19:40, Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
>
> On Fri, Jul 07, 2023 at 07:36:20PM +0530, Pintu Agarwal wrote:
> > On Fri, 7 Jul 2023 at 18:16, Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
> > >
> > > On Fri, Jul 07, 2023 at 03:57:42PM +0530, Anshuman Khandual wrote:
> > > > LGTM, cma->name is an identifying attribute for the region for which the allocation
> > > > request was made. But how about using cma_get_name() helper instead ? Very few call
> > > > sites have been using the helper.
> > >
> > > It's not really a "helper", is it?  The function name is longer than
> > > its implementation.
> > >
> > > cma_get_name(cma)
> > > vs
> > > cma->name
> > >
> > > Plus there's the usual question about whether a "got" name needs to be
> > > "put" (does it grab a refcount?)
> > >
> > > I think it's useful that this function exists since it lets us not expose
> > > struct cma outside of mm/, but it really should be called cma_name()
> > > and I don't think we should be encouraging its use within cma.c.
> >
> > Also, cma_get_name() is a trivial assignment.
> > And in one of the previous patches we avoided function calls with
> > trivial assignments.
> > https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/kernel/dma/contiguous.c?h=next-20230705&id=5af638931eb374aa0894d8343cee72f50307ef20
> > dma-contiguous: remove dev_set_cma_area
> >
> > One more question from here:
> > pr_debug("%s(cma %p, name: %s, count %lu, align %d)\n", __func__,
> >                 (void *)cma, cma->name, count, align);
> >
> > Do we really need this "cma %p" printing ?
> > I hardly check it and simply rely on name and count.
>
> Printing pointers is almost always a bad idea.  Printing the base_pfn
> might be a good idea to distinguish CMAs which happen to have the
> same name?
>
No there is no name there, it's just a ptrval
cma: cma_alloc(cma (ptrval), name: reserved, count 64, align 6)