On 06.06.2024 13:38:24, Geert Uytterhoeven wrote:
> > > > > @@ -545,8 +539,8 @@ struct rcar_canfd_global {
> > > > > struct platform_device *pdev; /* Respective platform device */
> > > > > struct clk *clkp; /* Peripheral clock */
> > > > > struct clk *can_clk; /* fCAN clock */
> > > > > - enum rcar_canfd_fcanclk fcan; /* CANFD or Ext clock */
> > > > > unsigned long channels_mask; /* Enabled channels mask */
> > > > > + bool extclk; /* CANFD or Ext clock */
> > > > > bool fdmode; /* CAN FD or Classical CAN only mode */
> > > >
> > > > Notwithstanding comment: you may consider to replace those two booleans by a:
> > > >
> > > > unsigned int flags;
> > > >
> > > > This way, no more fields would be needed in the future if more quirks are added.
> > >
> > > Using "unsigned int flags" and BIT(x) flags would increase code size
> > > by 8 bytes (arm/arm64).
> >
> > I am not sure where you derive your figure from, but looking at the pahole:
>
> pahole shows the size of data structures.
>
> > > Using "unsigned int foo:1" bitfields would increase code size by 16
> > > (arm) or 12 (arm64) bytes.
> > > So as long as we can fit more bools inside the hole, it is more
> > > efficient to do so...
> >
> > I do not get this either. Where did you get your 16 bytes from? If I do:
>
> I also looked at code size[*]: while storing bits takes less space than
> storing bytes, processing bits may require more instructions than
> processing bytes (depending on the architecture).
>
> [*] size drivers/net/can/rcar/rcar_canfd.o
You have probably used "scripts/bloat-o-meter" from the kernel source
for this, right?
Marc
--
Pengutronix e.K. | Marc Kleine-Budde |
Embedded Linux | https://www.pengutronix.de |
Vertretung Nürnberg | Phone: +49-5121-206917-129 |
Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-9 |
Attachment:
signature.asc
Description: PGP signature
