Re: [PATCH v2 4/4] serial: Add kserial_rs485 to avoid wasted space due to .padding

[Ilpo Järvinen]

On Tue, 30 Aug 2022, Greg Kroah-Hartman wrote:

> On Tue, Aug 30, 2022 at 10:29:56AM +0300, Ilpo Järvinen wrote:
> > The struct serial_rs485 has a .padding field to make uapi updates
> > easier. It wastes space, however. Create struct kserial_rs485 which is
> > a kerner counterpart w/o padding.
> 
> "kernel"?
> 
> And what is the size difference now?

Roughly 16B of padding is eliminated. That saving is then multiplied by 
the times it appears in in-kernel structs (2x per uart_port, 1x per rs485 
supporting driver). As said in my other reply, if you feel it's too 
little gained by eliminating the padding, I can drop this patch, just let 
me know.

> > +/**
> > + * struct kserial_rs485 - kernel-side struct for controlling RS485 settings.
> > + * @flags:			RS485 feature flags
> > + * @delay_rts_before_send:	Delay before send (milliseconds)
> > + * @delay_rts_after_send:	Delay after send (milliseconds)
> > + * @addr_recv:			Receive filter for RS485 addressing mode
> > + *				(used only when %SER_RS485_ADDR_RECV is set).
> > + * @addr_dest:			Destination address for RS485 addressing mode
> > + *				(used only when %SER_RS485_ADDR_DEST is set).
> > + *
> > + * Must match with struct serial_rs485 in include/uapi/linux/serial.h excluding
> > + * the padding.
> 
> Why must this match?

Because serial_rs485_from_user() and serial_rs485_to_user() just copy 
things over from one struct type to another w/o considering the fields 
individually. If that's not acceptable, I could make it copy field by 
field but it didn't feel necessary to allow "real" fields to differ to 
achieve padding elimination...

> And how is that going to be enforced?

With static_assert()s in serial_core.c. I'll add a note about that into 
the comment.

> > + */
> > +struct kserial_rs485 {
> > +	__u32	flags;
> > +	__u32	delay_rts_before_send;
> > +	__u32	delay_rts_after_send;
> > +	struct {
> > +		__u8    addr_recv;
> > +		__u8    addr_dest;
> > +	};
> 
> As this is an in-kernel structure, this should be "u32" and "u8" now.

Right, I'll change those.

-- 
 i.