Re: [PATCH] powerpc/lpar: Don't crash on H_PROTECT errors

[Anshuman Khandual]

On 03/01/2016 10:57 AM, Aneesh Kumar K.V wrote:
> Tyrel Datwyler  writes:
> 
>> > On 02/29/2016 03:52 AM, Anshuman Khandual wrote:
>>> >> There are certain condition in which H_PROTECT can return error code
>>> >> other than H_NOT_FOUND and H_SUCCESS. One such being an attempt to
>>> >> update an hpte owned by adjunct partition. Return 0 in that case so
>>> >> that user space will retry the access. In adjunct case this mean we
>>> >> will not make much progress in the user space. But atleast we get a
>>> >> chance to kill the task and avoid taking down the entire box.
>>> >> 
>>> >> Signed-off-by: Aneesh Kumar K.V 
>>> >> Signed-off-by: Anshuman Khandual 
>>> >> ---
>>> >>  arch/powerpc/platforms/pseries/lpar.c | 2 +-
>>> >>  1 file changed, 1 insertion(+), 1 deletion(-)
>>> >> 
>>> >> diff --git a/arch/powerpc/platforms/pseries/lpar.c 
>>> >> b/arch/powerpc/platforms/pseries/lpar.c
>>> >> index 477290a..31bcdaf 100644
>>> >> --- a/arch/powerpc/platforms/pseries/lpar.c
>>> >> +++ b/arch/powerpc/platforms/pseries/lpar.c
>>> >> @@ -310,7 +310,7 @@ static long pSeries_lpar_hpte_updatepp(unsigned long 
>>> >> slot,
>>> >>  
>>> >>  pr_devel("ok\n");
>>> >>  
>>> >> -BUG_ON(lpar_rc != H_SUCCESS);
>>> >> +WARN_RATELIMIT(lpar_rc != H_SUCCESS, "H_PROTECT returned 
>>> >> %lu\n", lpar_rc);
>> >
>> > In the event that we don't get H_NOT_FOUND (which is handled earlier in
>> > the function) or H_SUCCESS this patch assumes H_RESOURCE. It fails to
>> > handle H_PARAMETER which is also a valid return from the H_PROTECT
>> > hypercall.
> One of the possible thing we could do is sent SIGBUS to the application
> rather than taking down the system ?

But in case of H_PARAMETER, its the kernel which generated wrong arguments
which was in it's control, hence BUG_ON will not more appropriate for it ?

___
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev

Re: [PATCH] powerpc/lpar: Don't crash on H_PROTECT errors

[Aneesh Kumar K.V]

Tyrel Datwyler  writes:

> On 02/29/2016 03:52 AM, Anshuman Khandual wrote:
>> There are certain condition in which H_PROTECT can return error code
>> other than H_NOT_FOUND and H_SUCCESS. One such being an attempt to
>> update an hpte owned by adjunct partition. Return 0 in that case so
>> that user space will retry the access. In adjunct case this mean we
>> will not make much progress in the user space. But atleast we get a
>> chance to kill the task and avoid taking down the entire box.
>> 
>> Signed-off-by: Aneesh Kumar K.V 
>> Signed-off-by: Anshuman Khandual 
>> ---
>>  arch/powerpc/platforms/pseries/lpar.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>> 
>> diff --git a/arch/powerpc/platforms/pseries/lpar.c 
>> b/arch/powerpc/platforms/pseries/lpar.c
>> index 477290a..31bcdaf 100644
>> --- a/arch/powerpc/platforms/pseries/lpar.c
>> +++ b/arch/powerpc/platforms/pseries/lpar.c
>> @@ -310,7 +310,7 @@ static long pSeries_lpar_hpte_updatepp(unsigned long 
>> slot,
>>  
>>  pr_devel("ok\n");
>>  
>> -BUG_ON(lpar_rc != H_SUCCESS);
>> +WARN_RATELIMIT(lpar_rc != H_SUCCESS, "H_PROTECT returned %lu\n", 
>> lpar_rc);
>
> In the event that we don't get H_NOT_FOUND (which is handled earlier in
> the function) or H_SUCCESS this patch assumes H_RESOURCE. It fails to
> handle H_PARAMETER which is also a valid return from the H_PROTECT
> hypercall.

One of the possible thing we could do is sent SIGBUS to the application
rather than taking down the system ?

-aneesh

___
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev

Re: [PATCH] powerpc/lpar: Don't crash on H_PROTECT errors

[Anshuman Khandual]

On 02/29/2016 10:20 PM, Tyrel Datwyler wrote:
> On 02/29/2016 03:52 AM, Anshuman Khandual wrote:
>> There are certain condition in which H_PROTECT can return error code
>> other than H_NOT_FOUND and H_SUCCESS. One such being an attempt to
>> update an hpte owned by adjunct partition. Return 0 in that case so
>> that user space will retry the access. In adjunct case this mean we
>> will not make much progress in the user space. But atleast we get a
>> chance to kill the task and avoid taking down the entire box.
>>
>> Signed-off-by: Aneesh Kumar K.V 
>> Signed-off-by: Anshuman Khandual 
>> ---
>>  arch/powerpc/platforms/pseries/lpar.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/arch/powerpc/platforms/pseries/lpar.c 
>> b/arch/powerpc/platforms/pseries/lpar.c
>> index 477290a..31bcdaf 100644
>> --- a/arch/powerpc/platforms/pseries/lpar.c
>> +++ b/arch/powerpc/platforms/pseries/lpar.c
>> @@ -310,7 +310,7 @@ static long pSeries_lpar_hpte_updatepp(unsigned long 
>> slot,
>>  
>>  pr_devel("ok\n");
>>  
>> -BUG_ON(lpar_rc != H_SUCCESS);
>> +WARN_RATELIMIT(lpar_rc != H_SUCCESS, "H_PROTECT returned %lu\n", 
>> lpar_rc);
> 
> In the event that we don't get H_NOT_FOUND (which is handled earlier in
> the function) or H_SUCCESS this patch assumes H_RESOURCE. It fails to
> handle H_PARAMETER which is also a valid return from the H_PROTECT
> hypercall.

Right, I guess its okay to BUG_ON() if we detect H_PARAMETER return code
as the kernel should not have created wrong arguments to the hcall in
the first place. Although we will still want to return 0 from here for
H_RESOURCE return code with a WARN_RATELIMIT message. Will change the
code accordingly next time around.

___
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev

Re: [PATCH] powerpc/lpar: Don't crash on H_PROTECT errors

[Tyrel Datwyler]

On 02/29/2016 03:52 AM, Anshuman Khandual wrote:
> There are certain condition in which H_PROTECT can return error code
> other than H_NOT_FOUND and H_SUCCESS. One such being an attempt to
> update an hpte owned by adjunct partition. Return 0 in that case so
> that user space will retry the access. In adjunct case this mean we
> will not make much progress in the user space. But atleast we get a
> chance to kill the task and avoid taking down the entire box.
> 
> Signed-off-by: Aneesh Kumar K.V 
> Signed-off-by: Anshuman Khandual 
> ---
>  arch/powerpc/platforms/pseries/lpar.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/powerpc/platforms/pseries/lpar.c 
> b/arch/powerpc/platforms/pseries/lpar.c
> index 477290a..31bcdaf 100644
> --- a/arch/powerpc/platforms/pseries/lpar.c
> +++ b/arch/powerpc/platforms/pseries/lpar.c
> @@ -310,7 +310,7 @@ static long pSeries_lpar_hpte_updatepp(unsigned long slot,
>  
>   pr_devel("ok\n");
>  
> - BUG_ON(lpar_rc != H_SUCCESS);
> + WARN_RATELIMIT(lpar_rc != H_SUCCESS, "H_PROTECT returned %lu\n", 
> lpar_rc);

In the event that we don't get H_NOT_FOUND (which is handled earlier in
the function) or H_SUCCESS this patch assumes H_RESOURCE. It fails to
handle H_PARAMETER which is also a valid return from the H_PROTECT
hypercall.

-Tyrel

>  
>   return 0;
>  }
> 

___
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev

[PATCH] powerpc/lpar: Don't crash on H_PROTECT errors

[Anshuman Khandual]

There are certain condition in which H_PROTECT can return error code
other than H_NOT_FOUND and H_SUCCESS. One such being an attempt to
update an hpte owned by adjunct partition. Return 0 in that case so
that user space will retry the access. In adjunct case this mean we
will not make much progress in the user space. But atleast we get a
chance to kill the task and avoid taking down the entire box.

Signed-off-by: Aneesh Kumar K.V 
Signed-off-by: Anshuman Khandual 
---
 arch/powerpc/platforms/pseries/lpar.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/powerpc/platforms/pseries/lpar.c 
b/arch/powerpc/platforms/pseries/lpar.c
index 477290a..31bcdaf 100644
--- a/arch/powerpc/platforms/pseries/lpar.c
+++ b/arch/powerpc/platforms/pseries/lpar.c
@@ -310,7 +310,7 @@ static long pSeries_lpar_hpte_updatepp(unsigned long slot,
 
pr_devel("ok\n");
 
-   BUG_ON(lpar_rc != H_SUCCESS);
+   WARN_RATELIMIT(lpar_rc != H_SUCCESS, "H_PROTECT returned %lu\n", 
lpar_rc);
 
return 0;
 }
-- 
2.1.0

___
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev