linux-kernel - Re: [PATCH RFC] uio: allow use on nommu systems

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]

Date:	Thu, 14 Jul 2016 21:58:29 -0400
From:	Rich Felker <dalias@...c.org>
To:	Greg Kroah-Hartman <gregkh@...uxfoundation.org>
Cc:	"Hans J. Koch" <hjk@...sjkoch.de>, linux-kernel@...r.kernel.org
Subject: Re: [PATCH RFC] uio: allow use on nommu systems

On Fri, Jul 15, 2016 at 10:42:37AM +0900, Greg Kroah-Hartman wrote:
> On Thu, Jul 14, 2016 at 08:53:06PM -0400, Rich Felker wrote:
> > mmap of uio devices does not seem to work without further
> > nommu-specific support, but interrupt handling already works, and
> > userspace drivers for nommu systems can simply use physical mmio
> > addresses from userspace directly anyway without mmap.
> > 
> > Signed-off-by: Rich Felker <dalias@...c.org>
> > ---
> > 
> > I don't particularly expect this to be accepted upstream as-is, but
> > since we're actually trying to use the UIO subsystem on nommu (J2),
> > I'd like to start a discussion of what an acceptable patch would
> > entail.
> > 
> > The uio_pdrv_genirq driver was tested on J2 with this patch and
> > interrupt handling works as expected.
> > 
> >  drivers/uio/Kconfig | 1 -
> >  1 file changed, 1 deletion(-)
> > 
> > diff --git a/drivers/uio/Kconfig b/drivers/uio/Kconfig
> > index 52c98ce..387b2bb 100644
> > --- a/drivers/uio/Kconfig
> > +++ b/drivers/uio/Kconfig
> > @@ -1,6 +1,5 @@
> >  menuconfig UIO
> >  	tristate "Userspace I/O drivers"
> > -	depends on MMU
> 
> This doesn't seem to do much, don't you need to modify uio-specific
> drivers?  Or given that the core uio code doesn't rely on mmu, is this
> safe?

As noted above, mmap of the device does not work (returns error); I
don't understand why. But select/read/write for interrupt handling
works fine, and the uio system is fully usable if you just do mmio via
the raw addresses from userspace (this is nommu after all). I tested
it with uio_pdrv_genirq bound to a fake DT node for the J2 IPI
interrupt, disabling SMP and generating IPI interrupts manually to
check that it worked.

Rich