Re: [PATCH net v4 01/12] net: core: limit nested device depth

[Taehee Yoo <ap420073@gmail.com>]

On Tue, 1 Oct 2019 at 22:57, Johannes Berg <johannes@sipsolutions.net> wrote:
>
> Hi,
>

Hi!

> (jumping out now, forgive me for being so brief)
>
> > If I understand correctly, you said about the alignment of
> > "lower_level" and "upper_level".
> > I thought this place is a fine position for variables as regards the
> > alignment and I didn't try to put each variable in different places.
> >
> > If I misunderstood your mention, please let me know.
>
> Not sure what you mean, alignment doesn't matter for them (they're u8).
>
> I was thinking of the packing for the overall struct, we have:
>
>         unsigned int            max_mtu;
>         unsigned short          type;
>         unsigned short          hard_header_len;
>         unsigned char           min_header_len;
>
> +       unsigned char           upper_level, lower_level;
>
>         unsigned short          needed_headroom;
>         unsigned short          needed_tailroom;
>
>
> Previously, there was a one byte hole at that spot due to a single
> "unsigned char" (after something aligned at least 4 bytes) followed by
> "unsigned short" - now you push that out a bit.
>
> If you place the variables a bit lower, below "name_assign_type", you
> probably fill a hole instead.
>
> Check out the 'pahole' tool.
>

Thank you for the detailed explanation.
I tested the pahole and found holes.

$ pahole ./vmlinux.o -C net_device
        unsigned char              addr_assign_type;     /*   598     1 */
        unsigned char              addr_len;             /*   599     1 */
        short unsigned int         neigh_priv_len;       /*   600     2 */
        short unsigned int         dev_id;               /*   602     2 */
        short unsigned int         dev_port;             /*   604     2 */

        /* XXX 2 bytes hole, try to pack */

I will place the variables here.

> johannes
>

Thank you so much!
Taehee