linux-kernel - Re: [PATCH] soc: qcom: rpmh: Dirt can only make you dirtier, not cleaner

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] [day] [month] [year] [list]

Message-ID: <20200421065722.GV576963@builder.lan>
Date:   Mon, 20 Apr 2020 23:57:22 -0700
From:   Bjorn Andersson <bjorn.andersson@...aro.org>
To:     Doug Anderson <dianders@...omium.org>
Cc:     Andy Gross <agross@...nel.org>, Stephen Boyd <swboyd@...omium.org>,
        Maulik Shah <mkshah@...eaurora.org>,
        Evan Green <evgreen@...omium.org>,
        Srinivas Rao L <lsrao@...eaurora.org>,
        linux-arm-msm <linux-arm-msm@...r.kernel.org>,
        LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] soc: qcom: rpmh: Dirt can only make you dirtier, not
 cleaner

On Mon 20 Apr 13:47 PDT 2020, Doug Anderson wrote:

> Hi,
> 
> On Mon, Apr 20, 2020 at 1:30 PM Bjorn Andersson
> <bjorn.andersson@...aro.org> wrote:
> >
> > On Fri 17 Apr 14:15 PDT 2020, Douglas Anderson wrote:
> >
> > > Adding an item into the cache should never be able to make the cache
> > > cleaner.  Use "|=" rather than "=" to update the dirty flag.
> > >
> >
> > This is correct...
> >
> > Reviewed-by: Bjorn Andersson <bjorn.andersson@...aro.org>
> >
> > > Fixes: bb7000677a1b ("soc: qcom: rpmh: Update dirty flag only when data changes")
> > > Reported-by: Stephen Boyd <swboyd@...omium.org>
> > > Signed-off-by: Douglas Anderson <dianders@...omium.org>
> > > ---
> > >
> > >  drivers/soc/qcom/rpmh.c | 8 ++++----
> > >  1 file changed, 4 insertions(+), 4 deletions(-)
> > >
> > > diff --git a/drivers/soc/qcom/rpmh.c b/drivers/soc/qcom/rpmh.c
> > > index 3abbb08cd6e1..d1626a1328d7 100644
> > > --- a/drivers/soc/qcom/rpmh.c
> > > +++ b/drivers/soc/qcom/rpmh.c
> > > @@ -151,10 +151,10 @@ static struct cache_req *cache_rpm_request(struct rpmh_ctrlr *ctrlr,
> > >               break;
> > >       }
> > >
> > > -     ctrlr->dirty = (req->sleep_val != old_sleep_val ||
> > > -                     req->wake_val != old_wake_val) &&
> > > -                     req->sleep_val != UINT_MAX &&
> > > -                     req->wake_val != UINT_MAX;
> > > +     ctrlr->dirty |= (req->sleep_val != old_sleep_val ||
> > > +                      req->wake_val != old_wake_val) &&
> > > +                      req->sleep_val != UINT_MAX &&
> > > +                      req->wake_val != UINT_MAX;
> >
> > ...but this logic says dirty "if either sleep or wake has changed and
> > both sleep and wake are requested".
> >
> > So what if we have an entry with only sleep wake changed, then the
> > controller won't be dirty and hence the hardware won't know about this
> > request - until another "fully specified" request comes in, which would
> > cause the controller to be dirty and flush out the "partially specified"
> > request as well.
> >
> > Is this really the expected behavior?
> 
> IIRC, this has to do with how is_req_valid() works and how it's called
> from rpmh_flush().  ...but since I clearly screwed up the "|=" in the
> past let's see if I screwed this up too.  ...errr, what I mean to say
> is that I never make any mistakes.  Ever.  How dare you accuse me of
> such a thing?  ;-)
> 
> So is_req_valid() says that a request is valid if all three of these are true
> * Sleep is not UINT_MAX
> * Wake is not UINT_MAX
> * Sleep is not equal to wake
> 
> If a request is not valid then rpmh_flush will ignore it, it won't be
> sent, and it's as if it wasn't even in the cache.  Also: It's not
> expected that anyone will ever change a value _to_ UINT_MAX, so once
> something is initialized it's never uninitialized (NOTE: I don't think
> anything actually enforces this).
> 
> The above means that until both sleep and wake have changed away from
> their default of UINT_MAX that they can't really dirty the cache
> because we act as if they're not even in the cache.  Once they get
> something then the cache gets dirty if either sleep changes or wake
> changes.
> 
> 
> So I think the logic is right...  Does that make sense?
> 

Okay, that makes sense. Iirc how the RPMh deals with entering and
exiting the sleep state having one without the other shouldn't be valid,
so it sounds like we're good here.

Thanks for the explanation, applied the patch.

Regards,
Bjorn