linux-kernel - Re: [PATCH v5 8/8] rpmsg: Turn name service into a stand alone driver

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: <20201106175332.GB3203364@xps15>
Date:   Fri, 6 Nov 2020 10:53:32 -0700
From:   Mathieu Poirier <mathieu.poirier@...aro.org>
To:     Guennadi Liakhovetski <guennadi.liakhovetski@...ux.intel.com>
Cc:     ohad@...ery.com, bjorn.andersson@...aro.org,
        linux-remoteproc@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v5 8/8] rpmsg: Turn name service into a stand alone driver

On Fri, Nov 06, 2020 at 03:00:28PM +0100, Guennadi Liakhovetski wrote:
> On Fri, Nov 06, 2020 at 02:15:45PM +0100, Guennadi Liakhovetski wrote:
> > Hi Mathieu, Arnaud,
> > 
> > On Thu, Nov 05, 2020 at 03:50:28PM -0700, Mathieu Poirier wrote:
> > > From: Arnaud Pouliquen <arnaud.pouliquen@...com>
> > > 
> > > Make the RPMSG name service announcement a stand alone driver so that it
> > > can be reused by other subsystems.  It is also the first step in making the
> > > functionatlity transport independent, i.e that is not tied to virtIO.
> > 
> > Sorry, I just realised that my testing was incomplete. I haven't tested 
> > automatic module loading and indeed it doesn't work. If rpmsg_ns is loaded 
> > it probes and it's working, but if it isn't loaded and instead the rpmsg 
> > bus driver is probed (e.g. virtio_rpmsg_bus), calling 
> > rpmsg_ns_register_device() to create a new rpmsg_ns device doesn't cause 
> > rpmsg_ns to be loaded.
> 
> A simple fix for that is using MODULE_ALIAS("rpmsg:rpmsg_ns"); in rpmsg_ns.c 
> but that alone doesn't fix the problem completely - the module does load then 
> but not quickly enough, the NS announcement from the host / remote arrives 
> before rpmsg_ns has properly registered. I think the best solution would be 
> to link rpmsg_ns.c together with rpmsg_core.c. You'll probably want to keep 
> the module name, so you could rename them to just core.c and ns.c.

I'm pretty sure it is because virtio_device_ready() in rpmsg_probe() is called
before the kernel has finished loading the name space driver.  There has to be
a way to prevent that from happening - I will investigate further.

Thanks for reporting this,
Mathieu

> 
> Thanks
> Guennadi