diff mbox

[v4,11/14] ARM64: KVM: fix vgic_bitmap_get_reg function for BE 64bit case

Message ID 1402590613-3341-12-git-send-email-victor.kamensky@linaro.org (mailing list archive)
State New, archived
Headers show

Commit Message

Victor Kamensky June 12, 2014, 4:30 p.m. UTC
Fix vgic_bitmap_get_reg function to return 'right' word address of
'unsigned long' bitmap value in case of BE 64bit image.

Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
---
 virt/kvm/arm/vgic.c | 24 ++++++++++++++++++++++--
 1 file changed, 22 insertions(+), 2 deletions(-)

Comments

Christoffer Dall June 14, 2014, 3:04 p.m. UTC | #1
On Thu, Jun 12, 2014 at 09:30:10AM -0700, Victor Kamensky wrote:
> Fix vgic_bitmap_get_reg function to return 'right' word address of
> 'unsigned long' bitmap value in case of BE 64bit image.
> 
> Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
> ---
>  virt/kvm/arm/vgic.c | 24 ++++++++++++++++++++++--
>  1 file changed, 22 insertions(+), 2 deletions(-)
> 
> diff --git a/virt/kvm/arm/vgic.c b/virt/kvm/arm/vgic.c
> index 529c336..b4ffd82 100644
> --- a/virt/kvm/arm/vgic.c
> +++ b/virt/kvm/arm/vgic.c
> @@ -101,14 +101,34 @@ static u32 vgic_nr_lr;
>  
>  static unsigned int vgic_maint_irq;
>  
> +/*
> + * struct vgic_bitmap contains unions that provide two views of
> + * the same data. In one case it is an array of registers of
> + * u32's, and in the other case it is a bitmap of unsigned
> + * longs.
> + *
> + * This does not work on 64-bit BE systems, because the bitmap access
> + * will store two consecutive 32-bit words with the higher-addressed
> + * register's bits at the lower index and the lower-addressed register's
> + * bits at the higher index.
> + *
> + * Therefore, swizzle the register index when accessing the 32-bit word
> + * registers to access the right register's value.
> + */
> +#if defined(CONFIG_CPU_BIG_ENDIAN) && BITS_PER_LONG == 64
> +#define REG_OFFSET_SWIZZLE	1
> +#else
> +#define REG_OFFSET_SWIZZLE	0
> +#endif
> +
>  static u32 *vgic_bitmap_get_reg(struct vgic_bitmap *x,
>  				int cpuid, u32 offset)
>  {
>  	offset >>= 2;
>  	if (!offset)
> -		return x->percpu[cpuid].reg;
> +		return x->percpu[cpuid].reg + (offset ^ REG_OFFSET_SWIZZLE);
>  	else
> -		return x->shared.reg + offset - 1;
> +		return x->shared.reg + ((offset - 1) ^ REG_OFFSET_SWIZZLE);
>  }
>  
>  static int vgic_bitmap_get_irq_val(struct vgic_bitmap *x,
> -- 
> 1.8.1.4
> 

I had wished there was a cleaner way to abstract this, but I can't think
of a cleaner solution, and restructuring the whole thing would be very
invasive.

I'm curious to hear from Marc on this one, but you can add:

Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>

because it looks correct.

Thanks for picking up my suggestion on the commenting.

-Christoffer
Marc Zyngier June 21, 2014, 9:58 a.m. UTC | #2
On Sat, Jun 14 2014 at 04:04:51 PM, Christoffer Dall <christoffer.dall@linaro.org> wrote:
> On Thu, Jun 12, 2014 at 09:30:10AM -0700, Victor Kamensky wrote:
>> Fix vgic_bitmap_get_reg function to return 'right' word address of
>> 'unsigned long' bitmap value in case of BE 64bit image.
>> 
>> Signed-off-by: Victor Kamensky <victor.kamensky@linaro.org>
>> ---
>>  virt/kvm/arm/vgic.c | 24 ++++++++++++++++++++++--
>>  1 file changed, 22 insertions(+), 2 deletions(-)
>> 
>> diff --git a/virt/kvm/arm/vgic.c b/virt/kvm/arm/vgic.c
>> index 529c336..b4ffd82 100644
>> --- a/virt/kvm/arm/vgic.c
>> +++ b/virt/kvm/arm/vgic.c
>> @@ -101,14 +101,34 @@ static u32 vgic_nr_lr;
>>  
>>  static unsigned int vgic_maint_irq;
>>  
>> +/*
>> + * struct vgic_bitmap contains unions that provide two views of
>> + * the same data. In one case it is an array of registers of
>> + * u32's, and in the other case it is a bitmap of unsigned
>> + * longs.
>> + *
>> + * This does not work on 64-bit BE systems, because the bitmap access
>> + * will store two consecutive 32-bit words with the higher-addressed
>> + * register's bits at the lower index and the lower-addressed register's
>> + * bits at the higher index.
>> + *
>> + * Therefore, swizzle the register index when accessing the 32-bit word
>> + * registers to access the right register's value.
>> + */
>> +#if defined(CONFIG_CPU_BIG_ENDIAN) && BITS_PER_LONG == 64
>> +#define REG_OFFSET_SWIZZLE	1
>> +#else
>> +#define REG_OFFSET_SWIZZLE	0
>> +#endif
>> +
>>  static u32 *vgic_bitmap_get_reg(struct vgic_bitmap *x,
>>  				int cpuid, u32 offset)
>>  {
>>  	offset >>= 2;
>>  	if (!offset)
>> -		return x->percpu[cpuid].reg;
>> +		return x->percpu[cpuid].reg + (offset ^ REG_OFFSET_SWIZZLE);
>>  	else
>> -		return x->shared.reg + offset - 1;
>> +		return x->shared.reg + ((offset - 1) ^ REG_OFFSET_SWIZZLE);
>>  }
>>  
>>  static int vgic_bitmap_get_irq_val(struct vgic_bitmap *x,
>> -- 
>> 1.8.1.4
>> 
>
> I had wished there was a cleaner way to abstract this, but I can't think
> of a cleaner solution, and restructuring the whole thing would be very
> invasive.
>
> I'm curious to hear from Marc on this one, but you can add:
>
> Reviewed-by: Christoffer Dall <christoffer.dall@linaro.org>
>
> because it looks correct.
>
> Thanks for picking up my suggestion on the commenting.

I tried hard to find another solution, but I think this is unfortunately
the best we can have... ;-)

So:
        Acked-by: Marc Zyngier <narc.zyngier@arm.com>

I will have to fold in a similar solution for the vgic-dyn series.

Thanks,

	M.
diff mbox

Patch

diff --git a/virt/kvm/arm/vgic.c b/virt/kvm/arm/vgic.c
index 529c336..b4ffd82 100644
--- a/virt/kvm/arm/vgic.c
+++ b/virt/kvm/arm/vgic.c
@@ -101,14 +101,34 @@  static u32 vgic_nr_lr;
 
 static unsigned int vgic_maint_irq;
 
+/*
+ * struct vgic_bitmap contains unions that provide two views of
+ * the same data. In one case it is an array of registers of
+ * u32's, and in the other case it is a bitmap of unsigned
+ * longs.
+ *
+ * This does not work on 64-bit BE systems, because the bitmap access
+ * will store two consecutive 32-bit words with the higher-addressed
+ * register's bits at the lower index and the lower-addressed register's
+ * bits at the higher index.
+ *
+ * Therefore, swizzle the register index when accessing the 32-bit word
+ * registers to access the right register's value.
+ */
+#if defined(CONFIG_CPU_BIG_ENDIAN) && BITS_PER_LONG == 64
+#define REG_OFFSET_SWIZZLE	1
+#else
+#define REG_OFFSET_SWIZZLE	0
+#endif
+
 static u32 *vgic_bitmap_get_reg(struct vgic_bitmap *x,
 				int cpuid, u32 offset)
 {
 	offset >>= 2;
 	if (!offset)
-		return x->percpu[cpuid].reg;
+		return x->percpu[cpuid].reg + (offset ^ REG_OFFSET_SWIZZLE);
 	else
-		return x->shared.reg + offset - 1;
+		return x->shared.reg + ((offset - 1) ^ REG_OFFSET_SWIZZLE);
 }
 
 static int vgic_bitmap_get_irq_val(struct vgic_bitmap *x,