Re: [PATCH 24/25] arm64:ilp32: add vdso-ilp32 and use for signal return
From: Andrew Pinski
Date: Thu May 05 2016 - 02:40:53 EST
On Wed, May 4, 2016 at 7:24 PM, Zhangjian (Bamvor)
<bamvor.zhangjian@xxxxxxxxxx> wrote:
> Hi,
>
>
> On 2016/5/5 7:23, Andrew Pinski wrote:
>>
>> On Wed, May 4, 2016 at 2:49 PM, Yury Norov <ynorov@xxxxxxxxxxxxxxxxxx>
>> wrote:
>>>
>>> On Tue, May 03, 2016 at 08:41:25PM +0800, Zhangjian (Bamvor) wrote:
>>>>
>>>> Hi, all
>>>>
>>>> After apply this patch with my small testcase, the vsyscall of
>>>> gettimeofday in
>>>> ilp32 works in both big endian and small endian. In this patch, I use
>>>> the
>>>> different register and offset for ilp32 and lp64. Actually, the
>>>> COMPAT_TVAL_TV_SEC is same as TVAL_TV_SEC(so as to COMPAT_TSPEC_TV_SEC
>>>> and
>>>> TSPEC_TV_SEC). I add it to keep the logic clear. I also change the
>>>> version
>>>> of vdso to 4.6. It should change to 2.6.39 if glibc is not update.
>>>>
>>>
>>> [...]
>>>
>>> Hi Bamvor,
>>>
>>> It works for me as well. Thank you.
>>> I'll incorporate it in next submission.
>>
>>
>> We should add the following so we are complaint to the ILP32 ABI
>> dealing with pointers don't have to be zero extended for arguments,
>> Note ZERO should most likely be ZERO_PTR or something to do that
>> effect, I am not so good with names.
>> #ifdef __LP64__
>> #define ZERO(n)
>> #else
>> #define ZERO(n) mov w##n, w##n
>> #endif
>>
>> ...
>> ENTRY(__kernel_gettimeofday)
>> .cfi_startproc
>> ZERO(0)
>> ZERO(1)
>> ...
>> ENTRY(__kernel_clock_gettime)
>> .cfi_startproc
>> ZERO(1)
>> ...
>> ENTRY(__kernel_clock_getres)
>> .cfi_startproc
>> ZERO(1)
>
> Thanks. I will test and send a new version.
Note there is one more patch needed to __kernel_clock_getres to store
using 32bit rather than 64bit too.
@@ -213,7 +213,7 @@ ENTRY(__kernel_clock_getres)
ldr x2, 6f
2:
cbz w1, 3f
- stp xzr, x2, [x1]
+ stp PTR_REG(zr), PTR_REG(2), [x1]
3: /* res == NULL. */
mov w0, wzr
Thanks,
Andrew
>
> Thanks.
>
> Bamvor
>>
>> Thanks,
>> Andrew Pinski
>>
>>>
>>> Yury.
>
>