[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <307db9de-e1e6-714e-8ccb-7a9b98b4a348@gmail.com>
Date: Sat, 28 Sep 2019 16:39:11 -0700
From: Khem Raj <raj.khem@...il.com>
To: musl@...ts.openwall.com, Rich Felker <dalias@...c.org>
Subject: Re: [PATCH] Change container for riscv floating-point state to
__riscv_mc_fp_state
On 9/28/19 4:24 PM, Rich Felker wrote:
> On Sat, Sep 28, 2019 at 04:03:48PM -0700, Khem Raj wrote:
>> Match the struct elements with glibc
>> glibc also uses __riscv_mc prefix, some packages like gdb uses that
>>
>> Signed-off-by: Khem Raj <raj.khem@...il.com>
>> ---
>> arch/riscv64/bits/signal.h | 18 +++++++++---------
>> arch/riscv64/bits/user.h | 2 +-
>> 2 files changed, 10 insertions(+), 10 deletions(-)
>>
>> diff --git a/arch/riscv64/bits/signal.h b/arch/riscv64/bits/signal.h
>> index 76d7ad80..aba7ab00 100644
>> --- a/arch/riscv64/bits/signal.h
>> +++ b/arch/riscv64/bits/signal.h
>> @@ -13,25 +13,25 @@ typedef unsigned long greg_t;
>> typedef unsigned long gregset_t[32];
>>
>> struct __riscv_mc_f_ext_state {
>> - unsigned int f[32];
>> - unsigned int fcsr;
>> + unsigned int __f[32];
>> + unsigned int __fcsr;
>> };
>>
>> struct __riscv_mc_d_ext_state {
>> - unsigned long long f[32];
>> - unsigned int fcsr;
>> + unsigned long long __f[32];
>> + unsigned int __fcsr;
>> };
>>
>> struct __riscv_mc_q_ext_state {
>> - unsigned long long f[64] __attribute__((aligned(16)));
>> - unsigned int fcsr;
>> + unsigned long long __f[64] __attribute__((aligned(16)));
>> + unsigned int __fcsr;
>> unsigned int reserved[3];
>> };
>>
>> union __riscv_mc_fp_state {
>> - struct __riscv_mc_f_ext_state f;
>> - struct __riscv_mc_d_ext_state d;
>> - struct __riscv_mc_q_ext_state q;
>> + struct __riscv_mc_f_ext_state __f;
>> + struct __riscv_mc_d_ext_state __d;
>> + struct __riscv_mc_q_ext_state __q;
>> };
>
> This looks like an improvement for another major reason too -- the
> names f, d, q, are really bad namespace pollution to have in the
> default profile.
>
> There are actually more namespace problems in this file too, some of
> them actually conformance issues.
>
>> diff --git a/arch/riscv64/bits/user.h b/arch/riscv64/bits/user.h
>> index d2e383b2..9918a2fe 100644
>> --- a/arch/riscv64/bits/user.h
>> +++ b/arch/riscv64/bits/user.h
>> @@ -5,4 +5,4 @@ struct user_fpregs_struct {
>>
>> #define ELF_NGREG 32
>> typedef unsigned long elf_greg_t, elf_gregset_t[ELF_NGREG];
>> -typedef struct user_fpregs_struct elf_fpregset_t;
>> +typedef union __riscv_mc_fp_state elf_fpregset_t;
>> --
>> 2.23.0
>
> This part unfortunately depends on signal.h, which sys/user.h does not
> include. Since sys/user.h is a junk header it might be acceptable to
> just make the rv64 bits include signal.h, or to make it always include
> signal.h.
I think including bitss/signal.h in sys/user.h on rv64 is a good
compromise. Do you want a v2 with this change ? or I can send the
include part as a separate patch on top
>
> Rich
>
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
