linux-kernel - Re: [PATCH] powerpc/powernv: Clear LPCR[PECE1] via stop-api only for deep state offline

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: <20170901203837.765f1875@roar.ozlabs.ibm.com>
Date:   Fri, 1 Sep 2017 20:38:37 +1000
From:   Nicholas Piggin <npiggin@...il.com>
To:     Akshay Adiga <akshay.adiga@...ux.vnet.ibm.com>
Cc:     linux-kernel@...r.kernel.org,
        Shilpasri G Bhat <shilpa.bhat@...ux.vnet.ibm.com>,
        Pavithra Prakash <pavirampu@...ux.vnet.ibm.com>,
        linuxppc-dev@...ts.ozlabs.org,
        "Gautham R. Shenoy" <ego@...ux.vnet.ibm.com>
Subject: Re: [PATCH] powerpc/powernv: Clear LPCR[PECE1] via stop-api only
 for deep state offline

On Fri, 1 Sep 2017 15:38:59 +0530
Akshay Adiga <akshay.adiga@...ux.vnet.ibm.com> wrote:

> On 08/31/2017 05:37 PM, Nicholas Piggin wrote:
> > On Thu, 31 Aug 2017 17:17:41 +0530
> > "Gautham R. Shenoy" <ego@...ux.vnet.ibm.com> wrote:
> >  
> > > From: "Gautham R. Shenoy" <ego@...ux.vnet.ibm.com>
> > >
> > > commit 24be85a23d1f ("powerpc/powernv: Clear PECE1 in LPCR via
> > > stop-api only on Hotplug") clears the PECE1 bit of the LPCR via
> > > stop-api during CPU-Hotplug to prevent wakeup due to a decrementer on
> > > an offlined CPU which is in a deep stop state.
> > >
> > > In the case where the stop-api support is found to be lacking, the
> > > commit 785a12afdb4a ("powerpc/powernv/idle: Disable LOSE_FULL_CONTEXT
> > > states when stop-api fails") disables deep states that lose hypervisor
> > > context. Thus in this case, the offlined CPU will be put to some
> > > shallow idle state.
> > >
> > > However, we currently unconditionally clear the PECE1 in LPCR via
> > > stop-api during CPU-Hotplug even when deep states are disabled due to
> > > stop-api failure.
> > >
> > > Fix this by clearing PECE1 of LPCR via stop-api during CPU-Hotplug
> > > *only* when the offlined CPU will be put to a deep state that loses
> > > hypervisor context.  
> >
> > This looks okay to me. The bug is due to calling opal_slw_set_reg when
> > firmware has not enabled that feature, right?  
> Yes,
> 
> In the case where the stop-api support is found to be lacking, the 
> commit 785a12afdb4a ("powerpc/powernv/idle: Disable LOSE_FULL_CONTEXT 
> states when stop-api fails") disables deep states that lose hypervisor 
> context. Thus in this case, the offlined CPU will be put to some shallow 
> idle state.
> 
> If a shallow state ( < stop4 ) is being chosen for cpu hotplug, then :
> 1) this opal call is not required.
> 2) may not be supported.
> 
> Hence  should call opal_slw_set_reg() only if a deep state chosen for 
> cpu hotplug.

That makes sense, thanks for confirming.

> >  
> > >
> > > Fixes: commit 24be85a23d1f ("powerpc/powernv: Clear PECE1 in LPCR via
> > > stop-api only on Hotplug")
> > >
> > > Reported-by: Pavithra Prakash <pavirampu@...ux.vnet.ibm.com>
> > > Signed-off-by: Gautham R. Shenoy <ego@...ux.vnet.ibm.com>


Reviewed-by: Nicholas Piggin <npiggin@...il.com>


> > > ---
> > >  arch/powerpc/platforms/powernv/idle.c | 8 +++++++-
> > >  1 file changed, 7 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/arch/powerpc/platforms/powernv/idle.c  
> > b/arch/powerpc/platforms/powernv/idle.c  
> > > index 9f59041..23f8fba 100644
> > > --- a/arch/powerpc/platforms/powernv/idle.c
> > > +++ b/arch/powerpc/platforms/powernv/idle.c
> > > @@ -393,7 +393,13 @@ static void  
> > pnv_program_cpu_hotplug_lpcr(unsigned int cpu, u64 lpcr_val)  
> > >      u64 pir = get_hard_smp_processor_id(cpu);
> > >
> > >      mtspr(SPRN_LPCR, lpcr_val);
> > > -    opal_slw_set_reg(pir, SPRN_LPCR, lpcr_val);
> > > +
> > > +    /*
> > > +     * Program the LPCR via stop-api only for deepest stop state
> > > +     * can lose hypervisor context.
> > > +     */
> > > +    if (supported_cpuidle_states & OPAL_PM_LOSE_FULL_CONTEXT)
> > > +        opal_slw_set_reg(pir, SPRN_LPCR, lpcr_val);
> > >  }
> > >
> > >  /*  
> >  
>