The logic was checking the sizeof the structure being allocated to
determine whether an alignment fixup was required. This isn't right;
what we actually care about is the alignment of the actual pointer that's
about to be returned. This became an issue recently because struct
edac_mc_layer has a size that is not zero modulo eight, so we were
taking the correctly-aligned pointer and forcing it to be misaligned.
On Tile this caused an alignment exception.
Signed-off-by: Chris Metcalf <[email protected]>
---
Mauro, I assume you will push this through your tree? I am also happy
to push it through the tile tree. Let me know!
drivers/edac/edac_mc.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/drivers/edac/edac_mc.c b/drivers/edac/edac_mc.c
index 10f3750..de5ba86 100644
--- a/drivers/edac/edac_mc.c
+++ b/drivers/edac/edac_mc.c
@@ -164,7 +164,7 @@ void *edac_align_ptr(void **p, unsigned size, int n_elems)
else
return (char *)ptr;
- r = size % align;
+ r = (unsigned long)p % align;
if (r == 0)
return (char *)ptr;
--
1.6.5.2
On 6/6/2012 1:24 PM, Chris Metcalf wrote:
> The logic was checking the sizeof the structure being allocated to
> determine whether an alignment fixup was required. This isn't right;
> what we actually care about is the alignment of the actual pointer that's
> about to be returned. This became an issue recently because struct
> edac_mc_layer has a size that is not zero modulo eight, so we were
> taking the correctly-aligned pointer and forcing it to be misaligned.
> On Tile this caused an alignment exception.
>
> Signed-off-by: Chris Metcalf <[email protected]>
> ---
> Mauro, I assume you will push this through your tree? I am also happy
> to push it through the tile tree. Let me know!
Ping - I see this didn't make it into 3.5-rc2. I'm happy to push the change
through the tile tree if that's helpful (and someone wants to give their
Acked-by).
>
> drivers/edac/edac_mc.c | 2 +-
> 1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/drivers/edac/edac_mc.c b/drivers/edac/edac_mc.c
> index 10f3750..de5ba86 100644
> --- a/drivers/edac/edac_mc.c
> +++ b/drivers/edac/edac_mc.c
> @@ -164,7 +164,7 @@ void *edac_align_ptr(void **p, unsigned size, int n_elems)
> else
> return (char *)ptr;
>
> - r = size % align;
> + r = (unsigned long)p % align;
>
> if (r == 0)
> return (char *)ptr;
--
Chris Metcalf, Tilera Corp.
http://www.tilera.com
Em 11-06-2012 12:35, Chris Metcalf escreveu:
> On 6/6/2012 1:24 PM, Chris Metcalf wrote:
>> The logic was checking the sizeof the structure being allocated to
>> determine whether an alignment fixup was required. This isn't right;
>> what we actually care about is the alignment of the actual pointer that's
>> about to be returned. This became an issue recently because struct
>> edac_mc_layer has a size that is not zero modulo eight, so we were
>> taking the correctly-aligned pointer and forcing it to be misaligned.
>> On Tile this caused an alignment exception.
>>
>> Signed-off-by: Chris Metcalf <[email protected]>
>> ---
>> Mauro, I assume you will push this through your tree? I am also happy
>> to push it through the tile tree. Let me know!
>
> Ping - I see this didn't make it into 3.5-rc2. I'm happy to push the change
> through the tile tree if that's helpful (and someone wants to give their
> Acked-by).
Sorry, I took some days off to rest. I'll be applying it today on my -next tree
and likely tomorrow upstream, together with a few other EDAC fixes.
Regards,
Mauro
>
>>
>> drivers/edac/edac_mc.c | 2 +-
>> 1 files changed, 1 insertions(+), 1 deletions(-)
>>
>> diff --git a/drivers/edac/edac_mc.c b/drivers/edac/edac_mc.c
>> index 10f3750..de5ba86 100644
>> --- a/drivers/edac/edac_mc.c
>> +++ b/drivers/edac/edac_mc.c
>> @@ -164,7 +164,7 @@ void *edac_align_ptr(void **p, unsigned size, int n_elems)
>> else
>> return (char *)ptr;
>>
>> - r = size % align;
>> + r = (unsigned long)p % align;
>>
>> if (r == 0)
>> return (char *)ptr;
>