Re: sigset_t for new arches

["H.J. Lu" <hjl dot tools at gmail dot com>]

On Mon, Nov 13, 2017 at 4:28 PM, Vineet Gupta
<Vineet.Gupta1@synopsys.com> wrote:
> On 11/13/2017 02:54 PM, H.J. Lu wrote:
>>
>> On Mon, Nov 13, 2017 at 2:29 PM, Vineet Gupta
>> <Vineet.Gupta1@synopsys.com> wrote:
>>>
>>> Hi,
>>>
>>> This came to light when initial testing ARC port to glibc.
>>>
>>> ARC glibc sigaction wrapper was failing as Linux kernel expects sa_mask
>>> (sigset_t) to be 2 words (for the asm-generic syscall ABI mandated for
>>> newer
>>> arches) while glibc defines it to be 32 words.
>>>
>>> The quick fix is to lie to kernel about it - and following works. This is
>>> also approach taken by few other arches.
>>>
>>>          return INLINE_SYSCALL(rt_sigaction, 4,
>>> -                       sig, act, oact, sizeof(act->sa_mask));
>>> +                      sig, act, oact, _NSIG / 8);
>>>
>>>
>>> However, IMHO, it would be nicer to just define sigset_t to be 2 words
>>> wide
>>> for the relevant arch. If nothing else it would conserver some memory.
>>>
>>> My proposal is to define _SIGSET_NWORDS per arch in some header, with a
>>> fallback header defaulting to what the current value is. Does this make
>>> sense ? If yes, should this be in a new header or put in existing hdr
>>> etc.
>>
>>
>> No.  See:
>>
>> https://sourceware.org/ml/libc-alpha/2017-11/msg00412.html
>
>
> I don't see how this is solving the case of sigaction() where kernel is
> expecting 2 * 4 bytes, but glibc has it defined as 32 * 4. Perhaps your work
> needs to be extended to more syscalls.
>

In glibc, user-space sigset_t is much bigger kernel sigset_t, my
patch depends on it.   For system calls, you should pass what kernel
expects.


-- 
H.J.