[v2] nptl: Unconditionally use a 32-byte rseq area

If the kernel headers provide a larger struct rseq, we used that
size as the argument to the rseq system call.  As a result,
rseq registration would fail on older kernels which only accept
size 32.

Tested on x86_64-linux-gnu.  Built with build-many-glibcs.py.
This needs to be backported all the way to glibc 2.35.

---
v2: Drop further struct rseq reference on ia64.  Remove #include <sys/rseq.h>.
 nptl/descr.h | 18 ++++++++++++++----
 1 file changed, 14 insertions(+), 4 deletions(-)


base-commit: 3edc4ff2ceff4a59587ebecb94148d3bcfa1df62

On 7/20/23 07:46, Florian Weimer wrote:
> If the kernel headers provide a larger struct rseq, we used that
> size as the argument to the rseq system call.  As a result,
> rseq registration would fail on older kernels which only accept
> size 32.

Yes, this change appears to be needed considering the way the upstream 
kernel uapi has evolved.

Thanks,

Mathieu

> 
> Tested on x86_64-linux-gnu.  Built with build-many-glibcs.py.
> This needs to be backported all the way to glibc 2.35.
> 
> ---
> v2: Drop further struct rseq reference on ia64.  Remove #include <sys/rseq.h>.
>   nptl/descr.h | 18 ++++++++++++++----
>   1 file changed, 14 insertions(+), 4 deletions(-)
> 
> diff --git a/nptl/descr.h b/nptl/descr.h
> index d06abd6ad9..0171576c23 100644
> --- a/nptl/descr.h
> +++ b/nptl/descr.h
> @@ -34,7 +34,6 @@
>   #include <bits/types/res_state.h>
>   #include <kernel-features.h>
>   #include <tls-internal-struct.h>
> -#include <sys/rseq.h>
>   #include <internal-sigset.h>
>   
>   #ifndef TCB_ALIGNMENT
> @@ -405,14 +404,25 @@ struct pthread
>     /* Used on strsignal.  */
>     struct tls_internal_t tls_state;
>   
> -  /* rseq area registered with the kernel.  */
> -  struct rseq rseq_area;
> +  /* rseq area registered with the kernel.  Use a custom definition
> +     here to isolate from kernel struct rseq changes.  The
> +     implementation of sched_getcpu needs acccess to the cpu_id field;
> +     the other fields are unused and not included here.  */
> +  union
> +  {
> +    struct
> +    {
> +      uint32_t cpu_id_start;
> +      uint32_t cpu_id;
> +    };
> +    char pad[32];		/* Original rseq area size.  */
> +  } rseq_area __attribute__ ((aligned (32)));
>   
>     /* Amount of end padding, if any, in this structure.
>        This definition relies on rseq_area being last.  */
>   #define PTHREAD_STRUCT_END_PADDING \
>     (sizeof (struct pthread) - offsetof (struct pthread, rseq_area) \
> -   + sizeof (struct rseq))
> +   + sizeof ((struct pthread) {}.rseq_area))
>   } __attribute ((aligned (TCB_ALIGNMENT)));
>   
>   static inline bool
> 
> base-commit: 3edc4ff2ceff4a59587ebecb94148d3bcfa1df62
>

* Mathieu Desnoyers:

> On 7/20/23 07:46, Florian Weimer wrote:
>> If the kernel headers provide a larger struct rseq, we used that
>> size as the argument to the rseq system call.  As a result,
>> rseq registration would fail on older kernels which only accept
>> size 32.
>
> Yes, this change appears to be needed considering the way the upstream
> kernel uapi has evolved.

Andreas, is it okay to push this from an RM perspective?

Thanks,
Florian

>> Tested on x86_64-linux-gnu.  Built with build-many-glibcs.py.
>> This needs to be backported all the way to glibc 2.35.
>> ---
>> v2: Drop further struct rseq reference on ia64.  Remove #include <sys/rseq.h>.
>>   nptl/descr.h | 18 ++++++++++++++----
>>   1 file changed, 14 insertions(+), 4 deletions(-)
>> diff --git a/nptl/descr.h b/nptl/descr.h
>> index d06abd6ad9..0171576c23 100644
>> --- a/nptl/descr.h
>> +++ b/nptl/descr.h
>> @@ -34,7 +34,6 @@
>>   #include <bits/types/res_state.h>
>>   #include <kernel-features.h>
>>   #include <tls-internal-struct.h>
>> -#include <sys/rseq.h>
>>   #include <internal-sigset.h>
>>     #ifndef TCB_ALIGNMENT
>> @@ -405,14 +404,25 @@ struct pthread
>>     /* Used on strsignal.  */
>>     struct tls_internal_t tls_state;
>>   -  /* rseq area registered with the kernel.  */
>> -  struct rseq rseq_area;
>> +  /* rseq area registered with the kernel.  Use a custom definition
>> +     here to isolate from kernel struct rseq changes.  The
>> +     implementation of sched_getcpu needs acccess to the cpu_id field;
>> +     the other fields are unused and not included here.  */
>> +  union
>> +  {
>> +    struct
>> +    {
>> +      uint32_t cpu_id_start;
>> +      uint32_t cpu_id;
>> +    };
>> +    char pad[32];		/* Original rseq area size.  */
>> +  } rseq_area __attribute__ ((aligned (32)));
>>       /* Amount of end padding, if any, in this structure.
>>        This definition relies on rseq_area being last.  */
>>   #define PTHREAD_STRUCT_END_PADDING \
>>     (sizeof (struct pthread) - offsetof (struct pthread, rseq_area) \
>> -   + sizeof (struct rseq))
>> +   + sizeof ((struct pthread) {}.rseq_area))
>>   } __attribute ((aligned (TCB_ALIGNMENT)));
>>     static inline bool
>> base-commit: 3edc4ff2ceff4a59587ebecb94148d3bcfa1df62
>>

Am Freitag, 21. Juli 2023, 16:01:51 CEST schrieb Florian Weimer:
> * Mathieu Desnoyers:
> 
> > On 7/20/23 07:46, Florian Weimer wrote:
> >> If the kernel headers provide a larger struct rseq, we used that
> >> size as the argument to the rseq system call.  As a result,
> >> rseq registration would fail on older kernels which only accept
> >> size 32.
> >
> > Yes, this change appears to be needed considering the way the upstream
> > kernel uapi has evolved.
> 
> Andreas, is it okay to push this from an RM perspective?

Yes, please do it!

Best,
Andreas