[v1] arm/vgic-v3: Fix write_ignore_64's check in __vgic_v3_rdistr_rd_mmio_write()

[PATCH] arm/vgic-v3: Fix write_ignore_64's check in __vgic_v3_rdistr_rd_mmio_write()

Posted by Oleksandr Tyshchenko 5 months, 1 week ago

An attempt to write access the register (i.e. GICR_PROPBASER, GICR_PENDBASER)
which should be ignored (i.e. no virtual ITS present) causes the data about
due to incorrect check at the write_ignore_64 label. The check should be
inverted.

Fixes: c4d6bbdc12e5 ("xen/arm: vgic-v3: Support 32-bit access for 64-bit registers")
Signed-off-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
---
 xen/arch/arm/vgic-v3.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/xen/arch/arm/vgic-v3.c b/xen/arch/arm/vgic-v3.c
index 2eaa48fadb..b366b046a2 100644
--- a/xen/arch/arm/vgic-v3.c
+++ b/xen/arch/arm/vgic-v3.c
@@ -649,7 +649,7 @@ bad_width:
     return 0;
 
 write_ignore_64:
-    if ( vgic_reg64_check_access(dabt) ) goto bad_width;
+    if ( !vgic_reg64_check_access(dabt) ) goto bad_width;
     return 1;
 
 write_ignore_32:
-- 
2.34.1

Re: [PATCH] arm/vgic-v3: Fix write_ignore_64's check in __vgic_v3_rdistr_rd_mmio_write()

Posted by Julien Grall 5 months, 1 week ago

Hi Oleksandr,

On 20/05/2025 14:47, Oleksandr Tyshchenko wrote:
> An attempt to write access the register (i.e. GICR_PROPBASER, GICR_PENDBASER)
> which should be ignored (i.e. no virtual ITS present) causes the data about

I assume, this is a guest data abort, rather than Xen crash?

> due to incorrect check at the write_ignore_64 label. 
> The check should be
> inverted.

OOI, why would a guest try to write to GICR_PROPBASER if the ITS is not 
present? Was it a bug in the OS?

> 
> Fixes: c4d6bbdc12e5 ("xen/arm: vgic-v3: Support 32-bit access for 64-bit registers")
> Signed-off-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>

With the commit message clarified and Andrew's comments addressed:

Acked-by: Julien Grall <jgrall@amazon.com>

Cheers,

-- 
Julien Grall

Re: [PATCH] arm/vgic-v3: Fix write_ignore_64's check in __vgic_v3_rdistr_rd_mmio_write()

Posted by Oleksandr Tyshchenko 5 months, 1 week ago


On 20.05.25 18:02, Julien Grall wrote:
> Hi Oleksandr,

Hello Julien

> 
> On 20/05/2025 14:47, Oleksandr Tyshchenko wrote:
>> An attempt to write access the register (i.e. GICR_PROPBASER, 
>> GICR_PENDBASER)
>> which should be ignored (i.e. no virtual ITS present) causes the data 
>> about
> 
> I assume, this is a guest data abort, rather than Xen crash?

yes

> 
>> due to incorrect check at the write_ignore_64 label. The check should be
>> inverted.
> 
> OOI, why would a guest try to write to GICR_PROPBASER if the ITS is not 
> present? Was it a bug in the OS?

no, it was just me experimenting with redistributor registers.

> 
>>
>> Fixes: c4d6bbdc12e5 ("xen/arm: vgic-v3: Support 32-bit access for 
>> 64-bit registers")
>> Signed-off-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
> 
> With the commit message clarified and Andrew's comments addressed:
> 
> Acked-by: Julien Grall <jgrall@amazon.com>

thanks

> 
> Cheers,
>

Re: [PATCH] arm/vgic-v3: Fix write_ignore_64's check in __vgic_v3_rdistr_rd_mmio_write()

Posted by Andrew Cooper 5 months, 1 week ago

On 20/05/2025 2:47 pm, Oleksandr Tyshchenko wrote:
> An attempt to write access the register (i.e. GICR_PROPBASER, GICR_PENDBASER)
> which should be ignored (i.e. no virtual ITS present) causes the data about

Do you mean "data abort" here?  If not, I can't parse the sentence.

> due to incorrect check at the write_ignore_64 label. The check should be
> inverted.
>
> Fixes: c4d6bbdc12e5 ("xen/arm: vgic-v3: Support 32-bit access for 64-bit registers")
> Signed-off-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
> ---
>  xen/arch/arm/vgic-v3.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/xen/arch/arm/vgic-v3.c b/xen/arch/arm/vgic-v3.c
> index 2eaa48fadb..b366b046a2 100644
> --- a/xen/arch/arm/vgic-v3.c
> +++ b/xen/arch/arm/vgic-v3.c
> @@ -649,7 +649,7 @@ bad_width:
>      return 0;
>  
>  write_ignore_64:
> -    if ( vgic_reg64_check_access(dabt) ) goto bad_width;
> +    if ( !vgic_reg64_check_access(dabt) ) goto bad_width;

As you're modifying anyway, the goto should be on the next line.

~Andrew

Re: [PATCH] arm/vgic-v3: Fix write_ignore_64's check in __vgic_v3_rdistr_rd_mmio_write()

Posted by Oleksandr Tyshchenko 5 months, 1 week ago


On 20.05.25 17:24, Andrew Cooper wrote:

Hello Andrew

> On 20/05/2025 2:47 pm, Oleksandr Tyshchenko wrote:
>> An attempt to write access the register (i.e. GICR_PROPBASER, GICR_PENDBASER)
>> which should be ignored (i.e. no virtual ITS present) causes the data about
> 
> Do you mean "data abort" here?

yes

   If not, I can't parse the sentence.
> 
>> due to incorrect check at the write_ignore_64 label. The check should be
>> inverted.
>>
>> Fixes: c4d6bbdc12e5 ("xen/arm: vgic-v3: Support 32-bit access for 64-bit registers")
>> Signed-off-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
>> ---
>>   xen/arch/arm/vgic-v3.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/xen/arch/arm/vgic-v3.c b/xen/arch/arm/vgic-v3.c
>> index 2eaa48fadb..b366b046a2 100644
>> --- a/xen/arch/arm/vgic-v3.c
>> +++ b/xen/arch/arm/vgic-v3.c
>> @@ -649,7 +649,7 @@ bad_width:
>>       return 0;
>>   
>>   write_ignore_64:
>> -    if ( vgic_reg64_check_access(dabt) ) goto bad_width;
>> +    if ( !vgic_reg64_check_access(dabt) ) goto bad_width;
> 
> As you're modifying anyway, the goto should be on the next line.

ok, will move

> 
> ~Andrew