Re: [PATCH v19 02/13] linux-user: Add LoongArch signal support

[gaosong <gaosong@loongson.cn>]

[-- Attachment #1: Type: text/plain, Size: 1547 bytes --]


On 2022/6/24 上午9:20, Richard Henderson wrote:
> On 6/23/22 17:45, maobibo wrote:
>>
>>
>> 在 2022/6/24 07:34, Richard Henderson 写道:
>>> On 6/23/22 01:55, Song Gao wrote:
>>>> +static void setup_sigcontext(CPULoongArchState *env,
>>>> +                             struct target_sigcontext *sc,
>>>> +                             struct extctx_layout *extctx)
>>>> +{
>>>> +    int i;
>>>> +
>>>> +    if (extctx->flags & SC_USED_FP) {
>>>> +        __put_user(extctx->fpu.addr, &sc->sc_extcontext[0]);
>>>> +    } else {
>>>> +        __put_user(extctx->end.addr, &sc->sc_extcontext[0]);
>>>> +    }
>>>
>>> This is incorrect.  Where did this come from?  It certainly doesn't 
>>> appear in the kernel's version of setup_sigcontext. The only reason 
>>> the result works for you is that this is overwritten by 
>>> copy_fpu_to_sigframe within setup_sigframe.
>>
>> The val of flags is SC_USED_FP alway in function setup_extcontext in 
>> this version. We want to optimization in future if FP is not used for 
>> application where sigcontext for FP is not necessary. Also it can 
>> will be extended for 128bit/256bit vector FPU.
>
> No, that's not what I mean.  The store of the address is incorrect.
> This is no such address stored in the frame at this location.
>
> If you believe that I am incorrect, please point to the line within 
> the kernel source to which this corresponds.

You're right,   I'll corrrect on next version.

Thanks.
Song Gao

[-- Attachment #2: Type: text/html, Size: 2668 bytes --]