From mboxrd@z Thu Jan  1 00:00:00 1970
MIME-Version: 1.0
In-Reply-To: <51d4d8ec419f841a99baf30ede300c25@brasstown.quanstro.net>
References: <51d4d8ec419f841a99baf30ede300c25@brasstown.quanstro.net>
Date: Fri,  9 May 2014 09:28:53 -0500
Message-ID: <CAGGHmKF-upNXMZjzhrmp4zGsFvS25hx2BJeZapndYB0QbusgJw@mail.gmail.com>
From: Steven Stallion <sstallion@gmail.com>
To: Fans of the OS Plan 9 from Bell Labs <9fans@9fans.net>
Content-Type: text/plain; charset=UTF-8
Subject: Re: [9fans] ureg alignment
Topicbox-Message-UUID: e4c6439e-ead8-11e9-9d60-3106f5b1d025

On Fri, May 9, 2014 at 9:21 AM, erik quanstrom <quanstro@quanstro.net> wrote:
> on 64-bit machines, the unions in the ureg.h can lead to
> internal padding.  (power64 avoids this issue because everything
> is 64-bit aligned anyway.)  to sidestep the issue, i think
> it might make sense to use #defines.  for example, for arm
> the conversion would look something like this:
>
> typedef struct Ureg {
>         ulong   r0;
>         ulong   r1;
>         ulong   r2;
>         ulong   r3;
>         ulong   r4;
>         ulong   r5;
>         ulong   r6;
>         ulong   r7;
>         ulong   r8;
>         ulong   r9;
>         ulong   r10;
>         ulong   r11;
>         ulong   r12;    /* sb */
>         ulong   r13;
> #define sp      r13
>         ulong   r14;
> #define link    r14
>         ulong   type;   /* of exception */
>         ulong   psr;
>         ulong   pc;     /* interrupted addr */
> } Ureg;
>
> is there any reason not to do this?

Ugh, no! Is there a case where the padding is a problem? Normally
registers belonging to the same union are uniform in size. Those
defines will expand anywhere and not behave as expected:

ulong sp = ureg->sp;

This could lead to all sorts of entertaining problems.

Steve