Re: [PATCH v5 2/2] mailbox: mediatek: Add mtk-vcp-mailbox driver

[Nicolas Frattaroli <nicolas.frattaroli@collabora.com>]

On Tuesday, 23 September 2025 04:35:59 Central European Summer Time Jjian Zhou (周建) wrote:
> On Mon, 2025-09-22 at 15:10 +0200, Nicolas Frattaroli wrote:
> > External email : Please do not click links or open attachments until
> > you have verified the sender or the content.
> > 
> > 
> > On Monday, 22 September 2025 09:17:27 Central European Summer Time
> > Jjian Zhou (周建) wrote:
> > > On Sat, 2025-09-20 at 23:02 -0500, Jassi Brar wrote:
> > > > External email : Please do not click links or open attachments
> > > > until
> > > > you have verified the sender or the content.
> > > > 
> > > > 
> > > > On Fri, Sep 19, 2025 at 2:02 PM Nicolas Frattaroli
> > > > <nicolas.frattaroli@collabora.com> wrote:
> > > > > 
> > > > > On Friday, 19 September 2025 18:32:12 Central European Summer
> > > > > Time
> > > > > Jassi Brar wrote:
> > > > > > On Fri, Sep 19, 2025 at 3:31 AM Chen-Yu Tsai <
> > > > > > wenst@chromium.org>
> > > > > > wrote:
> > > > > > > 
> > > > > > > On Fri, Sep 19, 2025 at 7:50 AM Jassi Brar <
> > > > > > > jassisinghbrar@gmail.com> wrote:
> > > > > > > > 
> > > > > > > > On Thu, Aug 21, 2025 at 9:12 PM Jjian Zhou <
> > > > > > > > jjian.zhou@mediatek.com> wrote:
> > > > > > > > 
> > > > > > > > .....
> > > > > > > > 
> > > > > > > > > +#include <linux/module.h>
> > > > > > > > > +#include <linux/of.h>
> > > > > > > > > +#include <linux/platform_device.h>
> > > > > > > > > +#include <linux/slab.h>
> > > > > > > > > +
> > > > > > > > > +struct mtk_vcp_mbox_priv {
> > > > > > > > 
> > > > > > > > Maybe 'mtk_vcp_mbox' is a more appropriate name ?
> > > > > > > > 
> > > > > > > > > +       void __iomem *base;
> > > > > > > > > +       struct device *dev;
> > > > > > > > > +       struct mbox_controller mbox;
> > > > > > > > > +       const struct mtk_vcp_mbox_cfg *cfg;
> > > > > > > > > +       struct mtk_ipi_info ipi_recv;
> > > > > > > > 
> > > > > > > > Maybe also have "struct mbox_chan chan[1]; " so that you
> > > > > > > > don't have to
> > > > > > > > allocate one during the probe.
> > > > > > > > Also if you have  "struct mbox_controller mbox;" as the
> > > > > > > > first
> > > > > > > > member,
> > > > > > > > you could simply typecast that to get this structure.
> > > > > > > > Something like "struct mpfs_mbox" in mailbox-mpfs.c
> > > > > > > 
> > > > > > > I read somewhere that this way of subclassing is not
> > > > > > > recommended.
> > > > > > > Instead the base class should explicitly not be the first
> > > > > > > member.
> > > > > > > And then container_of() should be used.
> > > > > > > 
> > > > > > > I don't remember where I read this though. But I think the
> > > > > > > explicit
> > > > > > > container_of() is easier for understanding the intent.
> > > > > > > 
> > > > > > 
> > > > > > And how does container_of() work ? :)
> > > > > > typcasting the first member to its parent is the simplest
> > > > > > form of
> > > > > > container_of.
> > > > > > 
> > > > > > -j
> > > > > > 
> > > > > > 
> > > > > 
> > > > > Which is why it's completely equivalent and since code is
> > > > > supposed
> > > > > to communicate meaning to humans, container_of should be used.
> > > > > 
> > > > 
> > > > Nobody is suggesting typecasting cfg, dev or anything else.
> > > > Typecasting between mailbox controllers is fine and arguably
> > > > easier
> > > > on
> > > > the eyes than using a container_of.
> > > > 
> > > > -j
> > > 
> > > OK. How about:
> > > struct mtk_vcp_mbox *priv = (struct mtk_vcp_mbox *)chan-
> > > > con_priv;
> > > 
> > > Thanks.
> > > 
> > 
> > An explicit cast would be worse, as at that point you're telling
> > C to completely ignore any semblance of a type system it has.
> > 
> > 
> > 
>      struct mtk_vcp_mbox *priv; 
>      priv->dev = dev;
>      priv->chans[0].con_priv = priv;
> The type of con_priv is "void *". 
> Would the conversion mentioned above also have the issue you mentioned?
> 
> Thanks.
> 

No, in that case the cast is implicit. While void pointers do
subvert the type system, they are needed in this case because
the con_priv member needs to point at structs of any type.

The problem is that when you do something like

  struct apple *a = something;
  struct orange *o = (struct orange *)a;

then if the two structs (apple and orange) are incompatible,
the compiler won't even yell at you, because you're explicitly
casting.

With an implicit cast:

  struct apple *a = something;
  struct orange *o = a;

the compiler will tell you if you're doing something wrong.
Here's a userspace code example to illustrate the point:

    #include <stdio.h>

    struct apple {
            const char *name;
            unsigned int weight;
    };

    struct orange {
            int x;
            int y;
            int z;
    };

    int main(int argc, char** argv)
    {
            struct apple a = {"Granny Smith", 200};
            // won't compile, good!
            /* struct orange *o = &a; */
            // will compile, bad!
            struct orange *o = (struct orange *)&a;

            printf("%d\n", o->x);

            return 0;
    }

If you comment out the second struct orange line and uncomment the
first, then you'll get a compilation error, which is what we want
because the two structs are incompatible and we don't want the
assignment to work in this case, as that would be a bug.

The second struct orange line always compiles, even though the two
structs are incompatible, and will cause nonsense to be printed.

I hope this illustrates the point I was trying to make, which is
that explicit casts make it harder to find issues because they
force the language to simply accept the cast rather than give us
a compilation error when something nonsensical is being done.

Kind regards,
Nicolas Frattaroli