Re: CVS commit: src/sys/dev/pci

[Taylor R Campbell <campbell+netbsd-source-changes-d%mumble.net@localhost>]

> Date: Sat, 23 Jan 2021 22:59:22 -0000 (UTC)
> From: christos%astron.com@localhost (Christos Zoulas)
> 
> In article <23974.1611441140%splode.eterna.com.au@localhost>,
> matthew green  <mrg%eterna.com.au@localhost> wrote:
> >this seems dangerous to me.  we don't define it on
> >some platforms because we can't, so having it faked
> >out here seems like someone later will be confused
> >and the wrong thing will happen.
> >
> >i would rather have something like
> >
> >virtio_write8(...)
> >{
> >#ifdef __HAVE_BUS_SPACE_8
> >	just use the real thing
> >#else
> >	use the dual-_4 version that is ok _for this device_
> >#endif
> >}
> >
> >and then use this wrapper in the rest of the code.
> 
> This implementation is internal to virtio_pci and is guaranteed to work
> by the spec, how will someone else us it?

Conversely, how do you know whether this hacked-up implementation
which tears the write into two will actually work?  Maybe it works for
virtio but there are likely other devices out there for which it will
fail or have weird side effects if the architecture doesn't have
native 8-byte bus I/O.