[musl] [PATCH v2] Define user_regs, elf_greg_t and elf_fpregset_t for or1k

[musl] [PATCH v2] Define user_regs, elf_greg_t and elf_fpregset_t for or1k

[Florian Fainelli]

The OpenRISC architecture is currently missing a definition of its
user_regs structure, elf_greg_t and elf_fpregset_t as well as ELF_NGREG,
add those.
---
Changes in v2:

- Follow Stafford's recommendations based upon glibc

 arch/or1k/bits/user.h | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/arch/or1k/bits/user.h b/arch/or1k/bits/user.h
index e69de29bb2d1..227f022e2fec 100644
--- a/arch/or1k/bits/user.h
+++ b/arch/or1k/bits/user.h
@@ -0,0 +1,10 @@
+struct user_regs {
+	unsigned long gpr[32];
+	unsigned long pc;
+	unsigned long sr;
+};
+
+#define ELF_NGREG 32
+typedef unsigned long elf_greg_t, elf_gregset_t[ELF_NGREG];
+
+typedef elf_greg_t elf_fpregset_t[ELF_NGREG];
-- 
2.25.1

[musl] Re: [PATCH v2] Define user_regs, elf_greg_t and elf_fpregset_t for or1k

[Stafford Horne]

On Tue, May 17, 2022 at 02:58:26PM -0700, Florian Fainelli wrote:
> The OpenRISC architecture is currently missing a definition of its
> user_regs structure, elf_greg_t and elf_fpregset_t as well as ELF_NGREG,
> add those.
> ---
> Changes in v2:
> 
> - Follow Stafford's recommendations based upon glibc
> 
>  arch/or1k/bits/user.h | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/arch/or1k/bits/user.h b/arch/or1k/bits/user.h
> index e69de29bb2d1..227f022e2fec 100644
> --- a/arch/or1k/bits/user.h
> +++ b/arch/or1k/bits/user.h
> @@ -0,0 +1,10 @@
> +struct user_regs {
> +	unsigned long gpr[32];
> +	unsigned long pc;
> +	unsigned long sr;
> +};
> +
> +#define ELF_NGREG 32
> +typedef unsigned long elf_greg_t, elf_gregset_t[ELF_NGREG];
> +

I think ELF_NGREG should be 34 here.

On glibc we defined this like this, but I am beginning to doubt it is correct:

    (sys/ucontext.h)
    #define __NGREG	32

    (bits/procfs.h)
    #define ELF_NGREG __NGREG
    typedef elf_greg_t elf_gregset_t[34];

In linux and some other architectures they define ELF_NGREG as below (so that
would mean 34 is right):

    #define ELF_NGREG (sizeof (struct user_regs_struct) / sizeof (elf_greg_t))

In linux we define this, so yours matches.

    (arch/openrisc/include/uapi/asm/ptrace.h)
    truct user_regs_struct {
	    /* GPR R0-R31... */
	    unsigned long gpr[32];
	    unsigned long pc;
	    unsigned long sr;
    };

> +typedef elf_greg_t elf_fpregset_t[ELF_NGREG];

This should be:

    typedef elf_greg_t elf_fpregset_t[32];

However, I have tested all of glibc with the definition as above with ELF_NGREG
32 and elf_gregset_t[34].  But, I think I will like to change that and test
again.

So in the end I think we should use this.

   #define ELF_NGREG (sizeof (struct user_regs) / sizeof (elf_greg_t))
   typedef elf_greg_t elf_gregset_t[ELF_NGREG];
   typedef elf_greg_t elf_fpregset_t[32];


Side Point

I am not sure why cpulimit needs elf_gregset_t anyway.  I see it includes
procfs.h, I removed that include and it seems to compile ok.

-Stafford

[musl] Re: [PATCH v2] Define user_regs, elf_greg_t and elf_fpregset_t for or1k

[Florian Fainelli]

On 5/18/2022 9:18 PM, Stafford Horne wrote:
> On Tue, May 17, 2022 at 02:58:26PM -0700, Florian Fainelli wrote:
>> The OpenRISC architecture is currently missing a definition of its
>> user_regs structure, elf_greg_t and elf_fpregset_t as well as ELF_NGREG,
>> add those.
>> ---
>> Changes in v2:
>>
>> - Follow Stafford's recommendations based upon glibc
>>
>>   arch/or1k/bits/user.h | 10 ++++++++++
>>   1 file changed, 10 insertions(+)
>>
>> diff --git a/arch/or1k/bits/user.h b/arch/or1k/bits/user.h
>> index e69de29bb2d1..227f022e2fec 100644
>> --- a/arch/or1k/bits/user.h
>> +++ b/arch/or1k/bits/user.h
>> @@ -0,0 +1,10 @@
>> +struct user_regs {
>> +	unsigned long gpr[32];
>> +	unsigned long pc;
>> +	unsigned long sr;
>> +};
>> +
>> +#define ELF_NGREG 32
>> +typedef unsigned long elf_greg_t, elf_gregset_t[ELF_NGREG];
>> +
> 
> I think ELF_NGREG should be 34 here.
> 
> On glibc we defined this like this, but I am beginning to doubt it is correct:
> 
>      (sys/ucontext.h)
>      #define __NGREG	32
> 
>      (bits/procfs.h)
>      #define ELF_NGREG __NGREG
>      typedef elf_greg_t elf_gregset_t[34];
> 
> In linux and some other architectures they define ELF_NGREG as below (so that
> would mean 34 is right):
> 
>      #define ELF_NGREG (sizeof (struct user_regs_struct) / sizeof (elf_greg_t))
> 
> In linux we define this, so yours matches.
> 
>      (arch/openrisc/include/uapi/asm/ptrace.h)
>      truct user_regs_struct {
> 	    /* GPR R0-R31... */
> 	    unsigned long gpr[32];
> 	    unsigned long pc;
> 	    unsigned long sr;
>      };
> 
>> +typedef elf_greg_t elf_fpregset_t[ELF_NGREG];
> 
> This should be:
> 
>      typedef elf_greg_t elf_fpregset_t[32];
> 
> However, I have tested all of glibc with the definition as above with ELF_NGREG
> 32 and elf_gregset_t[34].  But, I think I will like to change that and test
> again.
> 
> So in the end I think we should use this.
> 
>     #define ELF_NGREG (sizeof (struct user_regs) / sizeof (elf_greg_t))
>     typedef elf_greg_t elf_gregset_t[ELF_NGREG];
>     typedef elf_greg_t elf_fpregset_t[32];

OK, thanks!

> 
> 
> Side Point
> 
> I am not sure why cpulimit needs elf_gregset_t anyway.  I see it includes
> procfs.h, I removed that include and it seems to compile ok.

Yes, I think there must have been some confusion that procfs.h might be 
related to manipulating the /proc filesystem on Linux when it is not. I 
will submit a patch to buildroot backporting the pull request:

https://github.com/opsengine/cpulimit/pull/110

Thanks!
-- 
Florian