Re: [PATCH] busybox: add devmem

[Tom Rini <trini@konsulko.com>]

[-- Attachment #1: Type: text/plain, Size: 2270 bytes --]

On Thu, Jan 31, 2019 at 09:04:53PM +0200, Adrian Bunk wrote:
> On Wed, Jan 30, 2019 at 05:47:03PM -0500, Tom Rini wrote:
> > 
> > I would also ask if we should be enabling more stuff in busybox, period.
> 
> I don't think the current status quo would be a good starting point for that.
> 
> Before submitting I saw CONFIG_PATCH=y, which is a lot more of a
> "Why would I ever need this on an embedded device?".
> 
> Not saying that you are wrong, but the starting point should be a review
> of the current config.

Yes, somewhere between an audit of busybox and making it easier /
clearer on how to remove busybox on a "regular" image (core-image-*,
etc) are on my TODO list.  You might also get surprised btw where things
like patch get, erm, differently-used in systems in automated ways.

> > Customizing busybox for what you're trying to _do_ with a custom setup
> > is one of those top 5 TODO items with making a cut down image and
> > customizing the kernel config.  Outside of -tiny and initramfs/similar
> > cases, there's not a great reason to use
> > almost-but-not-quite-complete-busybox-applet compared with the regular
> > app.
> 
> I frequently end up in projects where I do have great reason for that:
> 
> For root filesystems of the 32-64 MB size it is not necessary to use a 
> different libc or use -Os, but using the busybox applets when they are 
> sufficient saves > 10% filesystem size compared to the full versions.
> 
> E.g. 0.45 MB for a standalone tar is much when the little functionality 
> you actually need is also provided by the busybox applet.
> 
> Plus there is also the convenience point that most of the tools you need 
> are already installed just by adding busybox with the default config.

Right.  This is the -tiny and similar cases.  Busybox is great for a lot
of things.  But there's also common cases where it's not.

But the reasons I object here are, aside from what others have brought
up about devmem vs devmem2 vs putting stuff in DTS files, etc, is that I
really don't want to see more implicit deps to busybox added in.
busybox is in virtually every single OE fs image, so I am going to argue
that every new utility we add there needs a real good justification.

-- 
Tom

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 836 bytes --]