Re: [PATCH 2/2] riscv/include/uapi: Define a custom __SIGINFO struct for RV32

[Alistair Francis]

On Wed, Jul 3, 2019 at 12:47 PM Arnd Bergmann <arnd@xxxxxxxx> wrote:
>
> On Wed, Jul 3, 2019 at 8:45 PM Alistair Francis <alistair23@xxxxxxxxx> wrote:
> >
> > On Wed, Jul 3, 2019 at 1:41 AM Arnd Bergmann <arnd@xxxxxxxx> wrote:
> > >
> > > On Wed, Jul 3, 2019 at 2:21 AM Alistair Francis
> > > <alistair.francis@xxxxxxx> wrote:
> > > >
> > > > The glibc implementation of siginfo_t results in an allignment of 8 bytes
> > > > for the union _sifields on RV32. The kernel has an allignment of 4 bytes
> > > > for the _sifields union. This results in information being lost when
> > > > glibc parses the siginfo_t struct.
> > >
> > > I think the problem is that you incorrectly defined clock_t to 64-bit,
> > > while it is 32 bit in the kernel. You should fix the clock_t definition
> > > instead, it would otherwise cause additional problems.
> >
> > That is the problem. I assume we want to change the kernel to use a
> > 64-bit clock_t.
>
> Certainly not!
>
> Doing so would likely involve deprecating all system calls that
> take a clock_t (anything passing a struct siginfo or struct tms) and
> replacements based on clock64_t.

Ah, that's really easy to fix then.

>
> > What I don't understand though is how that impacted this struct, it
> > doesn't use clock_t at all, everything in the struct is an int or
> > void*.
>
> si_utime/si_stime in siginfo are clock_t.

But they are further down the struct. I just assumed that GCC would
align those as required, I guess it aligns the start of the struct to
match some 64-bit members which seems strange.

Alistair

>
>       Arnd