Re: [PATCH] ARM: virt: Relax arch timer version check during early boot

Re: [PATCH] ARM: virt: Relax arch timer version check during early boot

[Vladimir Murzin]

+ Marc
+ kvmarm@lists.cs.columbia.edu

On 1/15/20 2:16 PM, Vladimir Murzin wrote:
> Updates to the Generic Timer architecture allow ID_PFR1.GenTimer to
> have values other than 0 or 1. At the moment, Linux is quite strict in
> the way it handles this field at early boot and will not configure
> arch timer if it doesn't find the value 1.
> 
> Since here use ubfx for arch timer version extraction (hyb-stub build
> with -march=armv7-a, so it is safe)
> 
> To help backports (even though the code was correct at the time of writing)
> Fixes: 8ec58be9f3ff ("ARM: virt: arch_timers: enable access to physical timers")
> Signed-off-by: Vladimir Murzin <vladimir.murzin@arm.com>
> ---
>  arch/arm/kernel/hyp-stub.S | 7 +++----
>  1 file changed, 3 insertions(+), 4 deletions(-)
> 
> diff --git a/arch/arm/kernel/hyp-stub.S b/arch/arm/kernel/hyp-stub.S
> index ae50203..6607fa8 100644
> --- a/arch/arm/kernel/hyp-stub.S
> +++ b/arch/arm/kernel/hyp-stub.S
> @@ -146,10 +146,9 @@ ARM_BE8(orr	r7, r7, #(1 << 25))     @ HSCTLR.EE
>  #if !defined(ZIMAGE) && defined(CONFIG_ARM_ARCH_TIMER)
>  	@ make CNTP_* and CNTPCT accessible from PL1
>  	mrc	p15, 0, r7, c0, c1, 1	@ ID_PFR1
> -	lsr	r7, #16
> -	and	r7, #0xf
> -	cmp	r7, #1
> -	bne	1f
> +	ubfx	r7, r7, #16, #4
> +	teq	r7, #0
> +	beq	1f
>  	mrc	p15, 4, r7, c14, c1, 0	@ CNTHCTL
>  	orr	r7, r7, #3		@ PL1PCEN | PL1PCTEN
>  	mcr	p15, 4, r7, c14, c1, 0	@ CNTHCTL
> 

_______________________________________________
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm

Re: [PATCH] ARM: virt: Relax arch timer version check during early boot

[Marc Zyngier]

Hi Vladimir,

On 2020-01-20 11:46, Vladimir Murzin wrote:
> + Marc
> + kvmarm@lists.cs.columbia.edu
> 
> On 1/15/20 2:16 PM, Vladimir Murzin wrote:
>> Updates to the Generic Timer architecture allow ID_PFR1.GenTimer to
>> have values other than 0 or 1. At the moment, Linux is quite strict in
>> the way it handles this field at early boot and will not configure
>> arch timer if it doesn't find the value 1.
>> 
>> Since here use ubfx for arch timer version extraction (hyb-stub build
>> with -march=armv7-a, so it is safe)
>> 
>> To help backports (even though the code was correct at the time of 
>> writing)
>> Fixes: 8ec58be9f3ff ("ARM: virt: arch_timers: enable access to 
>> physical timers")
>> Signed-off-by: Vladimir Murzin <vladimir.murzin@arm.com>

I'm not opposed to such a change, but it'd be good to document what 
other values
are expected here, as the current (Rev E_a) ARM ARM only mentions values 
0 and 1.

Thanks,

         M.

>> ---
>>  arch/arm/kernel/hyp-stub.S | 7 +++----
>>  1 file changed, 3 insertions(+), 4 deletions(-)
>> 
>> diff --git a/arch/arm/kernel/hyp-stub.S b/arch/arm/kernel/hyp-stub.S
>> index ae50203..6607fa8 100644
>> --- a/arch/arm/kernel/hyp-stub.S
>> +++ b/arch/arm/kernel/hyp-stub.S
>> @@ -146,10 +146,9 @@ ARM_BE8(orr	r7, r7, #(1 << 25))     @ HSCTLR.EE
>>  #if !defined(ZIMAGE) && defined(CONFIG_ARM_ARCH_TIMER)
>>  	@ make CNTP_* and CNTPCT accessible from PL1
>>  	mrc	p15, 0, r7, c0, c1, 1	@ ID_PFR1
>> -	lsr	r7, #16
>> -	and	r7, #0xf
>> -	cmp	r7, #1
>> -	bne	1f
>> +	ubfx	r7, r7, #16, #4
>> +	teq	r7, #0
>> +	beq	1f
>>  	mrc	p15, 4, r7, c14, c1, 0	@ CNTHCTL
>>  	orr	r7, r7, #3		@ PL1PCEN | PL1PCTEN
>>  	mcr	p15, 4, r7, c14, c1, 0	@ CNTHCTL
>> 

-- 
Jazz is not dead. It just smells funny...
_______________________________________________
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm

Re: [PATCH] ARM: virt: Relax arch timer version check during early boot

[Vladimir Murzin]

Hi Marc,

On 1/20/20 11:14 AM, Marc Zyngier wrote:
> Hi Vladimir,
> 
> On 2020-01-20 11:46, Vladimir Murzin wrote:
>> + Marc
>> + kvmarm@lists.cs.columbia.edu
>>
>> On 1/15/20 2:16 PM, Vladimir Murzin wrote:
>>> Updates to the Generic Timer architecture allow ID_PFR1.GenTimer to
>>> have values other than 0 or 1. At the moment, Linux is quite strict in
>>> the way it handles this field at early boot and will not configure
>>> arch timer if it doesn't find the value 1.
>>>
>>> Since here use ubfx for arch timer version extraction (hyb-stub build
>>> with -march=armv7-a, so it is safe)
>>>
>>> To help backports (even though the code was correct at the time of writing)
>>> Fixes: 8ec58be9f3ff ("ARM: virt: arch_timers: enable access to physical timers")
>>> Signed-off-by: Vladimir Murzin <vladimir.murzin@arm.com>
> 
> I'm not opposed to such a change, but it'd be good to document what other values
> are expected here, as the current (Rev E_a) ARM ARM only mentions values 0 and 1.

That true, ARM ARM doesn't mention it yet. OTOH, should we really care about exact
values as soon it stays compatible?

Cheers
Vladimir

> 
> Thanks,
> 
>         M.
> 
>>> ---
>>>  arch/arm/kernel/hyp-stub.S | 7 +++----
>>>  1 file changed, 3 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/arch/arm/kernel/hyp-stub.S b/arch/arm/kernel/hyp-stub.S
>>> index ae50203..6607fa8 100644
>>> --- a/arch/arm/kernel/hyp-stub.S
>>> +++ b/arch/arm/kernel/hyp-stub.S
>>> @@ -146,10 +146,9 @@ ARM_BE8(orr    r7, r7, #(1 << 25))     @ HSCTLR.EE
>>>  #if !defined(ZIMAGE) && defined(CONFIG_ARM_ARCH_TIMER)
>>>      @ make CNTP_* and CNTPCT accessible from PL1
>>>      mrc    p15, 0, r7, c0, c1, 1    @ ID_PFR1
>>> -    lsr    r7, #16
>>> -    and    r7, #0xf
>>> -    cmp    r7, #1
>>> -    bne    1f
>>> +    ubfx    r7, r7, #16, #4
>>> +    teq    r7, #0
>>> +    beq    1f
>>>      mrc    p15, 4, r7, c14, c1, 0    @ CNTHCTL
>>>      orr    r7, r7, #3        @ PL1PCEN | PL1PCTEN
>>>      mcr    p15, 4, r7, c14, c1, 0    @ CNTHCTL
>>>
> 

_______________________________________________
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm

Re: [PATCH] ARM: virt: Relax arch timer version check during early boot

[Marc Zyngier]

On 2020-01-20 12:34, Vladimir Murzin wrote:
> Hi Marc,
> 
> On 1/20/20 11:14 AM, Marc Zyngier wrote:
>> Hi Vladimir,
>> 
>> On 2020-01-20 11:46, Vladimir Murzin wrote:
>>> + Marc
>>> + kvmarm@lists.cs.columbia.edu
>>> 
>>> On 1/15/20 2:16 PM, Vladimir Murzin wrote:
>>>> Updates to the Generic Timer architecture allow ID_PFR1.GenTimer to
>>>> have values other than 0 or 1. At the moment, Linux is quite strict 
>>>> in
>>>> the way it handles this field at early boot and will not configure
>>>> arch timer if it doesn't find the value 1.
>>>> 
>>>> Since here use ubfx for arch timer version extraction (hyb-stub 
>>>> build
>>>> with -march=armv7-a, so it is safe)
>>>> 
>>>> To help backports (even though the code was correct at the time of 
>>>> writing)
>>>> Fixes: 8ec58be9f3ff ("ARM: virt: arch_timers: enable access to 
>>>> physical timers")
>>>> Signed-off-by: Vladimir Murzin <vladimir.murzin@arm.com>
>> 
>> I'm not opposed to such a change, but it'd be good to document what 
>> other values
>> are expected here, as the current (Rev E_a) ARM ARM only mentions 
>> values 0 and 1.
> 
> That true, ARM ARM doesn't mention it yet. OTOH, should we really care
> about exact values as soon it stays compatible?

That's for you to say, really. But given that you hint at some changes,
it'd be good to have at least a short sentence explaining that, for 
example,
"upcoming revisions of the architecture will allow different 
ID_PFR1.GenTimer
values while preserving backward compatibility".

Other than that, feel free to add my

Acked-by: Marc Zyngier <maz@kernel.org>

Thanks,

         M.
-- 
Jazz is not dead. It just smells funny...
_______________________________________________
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm