On Wed, Sep 14, 2016 at 3:29 PM, Dmitry Safonov <0x7f454...@gmail.com> wrote:
> 2016-09-15 1:08 GMT+03:00 Kyle Huey <m...@kylehuey.com>:
>> On Wed, Sep 14, 2016 at 2:59 PM, Dmitry Safonov <0x7f454...@gmail.com> wrote:
>>> 2016-09-15 0:08 GMT+03:00 Kyle Huey <m...@kylehuey.com>:
>>>> Signed-off-by: Kyle Huey <kh...@kylehuey.com>
>>>> ---
>>>> arch/x86/entry/syscalls/syscall_32.tbl | 1 +
>>>> arch/x86/kernel/process.c | 80
>>>> ++++++++++++++++++++++++++++++++++
>>>> arch/x86/kernel/process_64.c | 66 ----------------------------
>>>> 3 files changed, 81 insertions(+), 66 deletions(-)
>>>>
>>>> diff --git a/arch/x86/entry/syscalls/syscall_32.tbl
>>>> b/arch/x86/entry/syscalls/syscall_32.tbl
>>>> index f848572..3b6965b 100644
>>>> --- a/arch/x86/entry/syscalls/syscall_32.tbl
>>>> +++ b/arch/x86/entry/syscalls/syscall_32.tbl
>>>> @@ -386,3 +386,4 @@
>>>> 377 i386 copy_file_range sys_copy_file_range
>>>> 378 i386 preadv2 sys_preadv2
>>>> compat_sys_preadv2
>>>> 379 i386 pwritev2 sys_pwritev2
>>>> compat_sys_pwritev2
>>>> +380 i386 arch_prctl sys_arch_prctl
>>>
>>> Why not define it as other 32-bit syscalls with compat_sys_ prefix
>>> with the help of COMPAT_SYSCALL_DEFINE() macro?
>>> Then you could omit code moving, drop is_32 helper.
>>> I miss something obvious?
>>
>> The code will have to move regardless, because right now do_arch_prctl
>> is in process-64.c which is only compiled on a 64 bit kernel.
>
> Why? This code will not work anyway for 32-bit in your patches
> by obscuring it with is_32.
>
>> As I told Dave Hansen in the non-RESEND thread (not sure why
>> git-send-email didn't put him in this one ...) I considered doing a
>> compat_sys_arch_prctl that would reject the relevant arch_prctls that
>> don't apply on 32 bit but I didn't see any prior art for it (in my
>> admittedly non-exhaustive search).
>
> Well, you could just add to 64-bit do_arch_prctl() new cases for your
> prctls - that would be just a two-lines for each new prctl.
> Also add compat_sys_ and define *only* what's needed there for you,
> do not add there ARCH_{SET,GET}_{FS,GS}.
> Does this make sense?
Yeah, I should have spoken more clearly. We'll need some
implementation of the syscall outside of process_64.c. But we could
leave the 64 bit specific stuff behind in it. Dave Hansen suggested
something similar (though without the compat_sys_bit)
>FWIW, I don't think it would be horrible to leave the existing
> do_arch_prctl() code in process_64.h and call it
> do_64_bit_only_something_arch_prctl(), and only call in to it from the
> generic do_arch_prctl(). You really have one reason for all the "if
> (is_32)"'s and it would be nice to document why in one single place.
- Kyle
