Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Michael Ellerman <mpe@ellerman.id.au>
To:     Fabiano Rosas <farosas@linux.ibm.com>,
        Laurent Dufour <ldufour@linux.ibm.com>
Cc:     linuxppc-dev@lists.ozlabs.org, linux-kernel@vger.kernel.org,
        stable@vger.kernel.org, Nicholas Piggin <npiggin@gmail.com>
Subject: Re: [PATCH v2] powerpc/rtas: Keep MSR[RI] set when calling RTAS
In-Reply-To: <87levia8wy.fsf@linux.ibm.com>
References: <20220401140634.65726-1-ldufour@linux.ibm.com>
 <87r15aveny.fsf@mpe.ellerman.id.au> <87levia8wy.fsf@linux.ibm.com>
Date:   Wed, 04 May 2022 14:26:17 +1000
Message-ID: <87k0b1vs7q.fsf@mpe.ellerman.id.au>
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk

Fabiano Rosas <farosas@linux.ibm.com> writes:
> Michael Ellerman <mpe@ellerman.id.au> writes:
>>> diff --git a/arch/powerpc/kernel/entry_64.S b/arch/powerpc/kernel/entry_64.S
>>> index 9581906b5ee9..65cb14b56f8d 100644
>>> --- a/arch/powerpc/kernel/entry_64.S
>>> +++ b/arch/powerpc/kernel/entry_64.S
>>> @@ -330,22 +330,18 @@ _GLOBAL(enter_rtas)
>>>  	clrldi	r4,r4,2			/* convert to realmode address */
>>>         	mtlr	r4
>>>  
>>> -	li	r0,0
>>> -	ori	r0,r0,MSR_EE|MSR_SE|MSR_BE|MSR_RI
>>> -	andc	r0,r6,r0
>>> -	
>>> -        li      r9,1
>>> -        rldicr  r9,r9,MSR_SF_LG,(63-MSR_SF_LG)
>>> -	ori	r9,r9,MSR_IR|MSR_DR|MSR_FE0|MSR_FE1|MSR_FP|MSR_RI|MSR_LE
>>> -	andc	r6,r0,r9
>>  
>> One advantage of the old method is it can adapt to new MSR bits being
>> set by the kernel.
>>
>> For example we used to use RTAS on powernv, and this code didn't need
>> updating to cater to MSR_HV being set. We will probably never use RTAS
>> on bare-metal again, so that's OK.
>>
>> But your change might break secure virtual machines, because it clears
>> MSR_S whereas the old code didn't. I think SVMs did use RTAS, but I
>> don't know whether it matters if it's called with MSR_S set or not?
>>
>> Not sure if anyone will remember, or has a working setup they can test.
>> Maybe for now we just copy MSR_S from the kernel MSR the way the
>> current code does.
>
> Would the kernel even be able to change the bit? I think only urfid can
> clear MSR_S.

Good point :)

cheers