On 12/21/18 4:05 AM, Adhemerval Zanella wrote:
>> Thing is ARC signal return depends on a restorer. Meaning !SA_RESTORER doesn't
>> effectively exist (between libc and kernel). We currently honor user provided
>> value, instead we could simply overwrite it with default_rt_restorer and on the
>> way out, zero it out to avoid leaking the glibc guts to curious users.
>>
> Yes, this is what I suggest ARC should do.
So I was able to
(1) switch ARC to __sigset_t [2] vs. [128]
(2) create a generic patch to breakout struct sigaction into bits/sigaction-arch.h
(3) update generic __libc_sigaction to avoid itemized copy if sa_mask, sa_flags et
al offsets matched for struct kernel_signal and struct sigaction
(4) create arc/bits/sigaction-arch.h
However it turned out to be futile as the passthru won't work anyways :-(
@act in sigaction(sig, act, oact) is a const and can't be modified in place for
sticking in sa_restorer for kernel, thus needs to be copied over anyways.
This means we have [1] but have to drop [2-4]. Oh well !
_______________________________________________
linux-snps-arc mailing list
linux-snps-arc@xxxxxxxxxxxxxxxxxxx
http://lists.infradead.org/mailman/listinfo/linux-snps-arc
