linux-kernel - Re: [patch 2.6.24-rc1] resource

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <200710241939.17576.david-b@pacbell.net>
Date:	Wed, 24 Oct 2007 19:39:17 -0700
From:	David Brownell <david-b@...bell.net>
To:	Alan Cox <alan@...rguk.ukuu.org.uk>
Cc:	Linux Kernel list <linux-kernel@...r.kernel.org>,
	Andrew Morton <akpm@...ux-foundation.org>
Subject: Re: [patch 2.6.24-rc1] resource_len() utility function

On Wednesday 24 October 2007, Alan Cox wrote:
> On Wed, 24 Oct 2007 18:20:52 -0700
> David Brownell <david-b@...bell.net> wrote:
> 
> > Add a new resource_len() function, so drivers can start using this
> > instead of driver-private code for a common idiom.  The call can be
> > useful with at least:
> > 
> >  - request_region(), release_region()
> >  - request_mem_region(), release_mem_region()
> >  - ioremap()
> > 
> > Candidate drivers include those using platform or PNP busses, and
> > maybe some others.  PCI already has a similar function.
> > 
> > This patch also updates a representative set of drivers in two
> > subsystems to use this call (SPI, and USB peripheral/gadget).
> 
> PCI also increasingly is using functions that allow the user to choose to
> map a resource as a resource (eg pci_iomap). So is it better to have
> functions request_mem_resource(res)  free_mem_resource(res) and similar
> instead or as well ?

This was intended to be a minor band-aid.  ;)

We already have request_resource(), which does something
different than the request_*region() calls.  I think calls
with those names would complicate an already-too-strange
interface, adding oddball siblings to request_resource().

I'd hope that when those resource calls were defined they
made sense ... but to me, they don't do so today.  Consider
that the *typical* caller is given a "struct resource", and
then to claim the specified address space it must convert
it into a "start + length" representation before getting
back a *NEW* "struct resource" ... with identical contents,
other than the value of one all-but-undocumented flag bit.
Then, if it's I/O space the address is usable already; but
for memory space, it still needs an ioremap()...

Oh, and PCI has its own resource structs ("BAR") that don't
look or act the same as other resources.

So while I like the notion of starting to abolish that
conversion step, this wasn't an attempt to fix all the
bizarre behaviors of the resource API.

I could imagine a call taking a resource and returning
a "void __iomem *" to use for IO, which implicitly claims
the region (in either memory or i/o space) and does any
ioremap needed for memory space.  With a sibling call to
undo all that.  If that's the answer, someone else should
develop the patch and update drivers...

- Dave
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/