Hi! > On Tue, Jan 5, 2021 at 12:40 AM Pavel Machek <[email protected]> wrote: > > > > > > .name = "rpc-if-spi", > > > > > > - .pm = DEV_PM_OPS, > > > > > > + .pm = &rpcif_spi_pm_ops, > > > > > > > > You're aware rpcif_spi_pm_ops is now always referenced and thus > > > > > emitted, > > > > > increasing kernel size by 92 bytes if CONFIG_PM_SLEEP=n? > > > > > This may matter for RZ/A SoCs running from internal SRAM. > > > > > > > Hmm didn't realise this would be an issue on RZ/A. > > > > > > > Mark, could you please drop this patch from your branch. > > > > > > Please send an incremental patch with an appropriate changelog. > > > > Let's fix this properly. I'm pretty sure we have some macros that can > > solve this without re-introducing the ifdefs... > > There's pm_ptr(), but it uses CONFIG_PM as a selector, not CONFIG_PM_SLEEP.
Okay; so we could introduce pm_sleep_ptr().
Or we could just put single #ifdef CONFIG_PM_SLEEP around the .pm
assignment? That would be improvement on the original, and still
result in the same binary, right?
> > (Besides... 92 bytes. How big is kernel these days? 4MB? More? How
> > much SRAM do you have?)
>
> 92 bytes is indeed not much (on 64-bit it would be doubled).
> Still, it's good to make people think about innocent looking changes,
> once in a while.
>
> RZ/A1H and RZ/A1M have 10 resp. 5 MiB of SRAM.
> RZ/A2 has 4 MiB SRAM, which is sufficient to run Linux when used with
> XIP (requires a one-line Kconfig change rmk has been vetoing for
> years).
Aha, that is a bit smaller than I expected.
Best regards,
Pavel
--
DENX Software Engineering GmbH, Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
signature.asc
Description: Digital signature
