linux-kernel - Re: [PATCH v4 11/12] [RFC] firmware: arm_scmi: Add sync_cmds_atomic

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: <20210923150319.GC6510@e120937-lin>
Date:   Thu, 23 Sep 2021 16:03:19 +0100
From:   Cristian Marussi <cristian.marussi@....com>
To:     Jim Quinlan <james.quinlan@...adcom.com>
Cc:     Florian Fainelli <f.fainelli@...il.com>,
        open list <linux-kernel@...r.kernel.org>,
        "moderated list:BROADCOM BCM2711/BCM2835 ARM ARCHITECTURE" 
        <linux-arm-kernel@...ts.infradead.org>,
        Sudeep Holla <sudeep.holla@....com>,
        Jonathan Cameron <Jonathan.Cameron@...wei.com>,
        etienne.carriere@...aro.org,
        Vincent Guittot <vincent.guittot@...aro.org>,
        Souvik Chakravarty <souvik.chakravarty@....com>
Subject: Re: [PATCH v4 11/12] [RFC] firmware: arm_scmi: Add
 sync_cmds_atomic_replies transport flag

On Thu, Aug 26, 2021 at 02:29:21PM -0400, Jim Quinlan wrote:
> On Wed, Aug 25, 2021 at 2:49 PM Cristian Marussi
> <cristian.marussi@....com> wrote:
> >
> > On Wed, Aug 25, 2021 at 01:17:47PM -0400, Jim Quinlan wrote:
> > > On Wed, Aug 25, 2021 at 12:38 PM Florian Fainelli <f.fainelli@...il.com> wrote:
> > > >
> > > >
> > > >
> >
> > Hi Florian and Jim,
> >
> > > > On 8/24/2021 3:59 PM, Cristian Marussi wrote:
> > > > > A flag is added to let the transport signal the core that its handling of
> > > > > synchronous command messages implies that, after .send_message has returned
> > > > > successfully, the requested command can be assumed to be fully and
> > > > > completely executed on SCMI platform side so that any possible response
> > > > > value is already immediately available to be retrieved by a .fetch_reponse:
> > > > > in other words the polling phase can be skipped in such a case and the
> > > > > response values accessed straight away.
> > > > >
> > > > > Note that all of the above applies only when polling mode of operation was
> > > > > selected by the core: if instead a completion IRQ was found to be available
> > > > > the normal response processing path based on completions will still be
> > > > > followed.
> > > >
> > > > This might actually have to be settable on a per-message basis ideally
> > > > since we may be transporting short lived SCMI messages for which the
> > > > completion can be done at SMC time, and long lived SCMI messages (e.g.:
> > > > involving a voltage change) for which we would prefer a completion
> > > > interrupt. Jim, what do you think?
> > > Even if the SCMI main driver could be configured this way in an
> > > elegant manner, I'm not sure that there is a clean way of specifying
> > > this  attribute on a per-message basis.  Certainly we could do this
> > > with our own protocols, but  many of our "long lived" messages are the
> > > Perf protocol's set_level command.  At any rate, let me give it some
> > > thought.
> > >
> >
> > The new flag .sync_cmds_atomic_replies applies only when polling mode
> > has been selected for a specific cmd transaction, which means when no
> > completion IRQ was found available OR if xfer.poll_completion was
> > excplicitly set for a specific command.
> >
> > At the moment in this series (unknown bugs apart :D), if you have a
> > channel configured with a completion IRQ and the .sync_cmds_atomic_replies
> > set for the transport, this latter flag would be generally ignored and a
> > wait_for_completion() will be normally used upon reception of the
> > completionIRQ, UNLESS you specify that one specific command has to be
> > polled using the per message xfer.poll_completion flag: so you should be
> > already able to selectively use a polling which immediately returns after
> > the smc by setting xfer.poll_completion for that specific short lived
> > message (since sync_cmds_atomic_replies is set and applies to pollmode).
> > On the other side any other LONG lived message will be naturally handled
> > via completionIRQ + wait_for_completion. (at least that was the aim..)
> >
> > !!! NOTE that you'll have also to drop
> >
> >  [PATCH v4 10/12] [RFC] firmware: arm_scmi: Make smc transport atomic
> >
> > from this series for the wait_completion to happen as you wish.
> 
> Hi Cristian,
> 

Hi Jim,

> I've tested all commits on our SMC-based system.  I also tested all commits
> minus  "10/12 [RFC] firmware: arm_scmi: Make smc transport atomic".
> This was a basic stress test, not a comprehensive one.  So
> 
> Tested-by: Jim Quinlan <james.quinlan@...adcom.com>
> 
> Of course I have a strong preference for omitting  "10/12 [RFC]" :-).
> FWIW, if you are not planning on dropping this commit, perhaps there
> could be a transport
> node in the DT, and that could contain the  a bool  property
> "smc-atomic-capable"?
> 

I just posted V5 on this SCMI atomic transport series, where the atomic
mode behaviour of a transport can be selected by a Kconfig which is defined
as default N: so this new series should behave out-of-the-box like with the
previous one when you had dropped as a whole the SMC atomic patch.

Any feedback welcome.

Thanks,
Cristian