Re: [PATCH] AARCH64: Change cpsr type to be 64bit.

[Yufeng Zhang <Yufeng dot Zhang at arm dot com>]

On 12/23/13 18:04, Mark Kettenis wrote:
> > Date: Mon, 23 Dec 2013 17:39:00 +0000
> > From: Yufeng Zhang<Yufeng.Zhang@arm.com>
> >
> > On 12/20/13 17:47, Pedro Alves wrote:
> > > On 12/18/2013 07:08 PM, Andrew Pinski wrote:
> > > > As mentioned in http://www.spinics.net/lists/arm-kernel/msg290896.html, we should change gdb's notion of cpsr to be 64bit.
> > > >
> > > > OK?  Build and tested for aarch64-linux-gnu with no regressions.
> > > > I also double checked to make sure that what is passed down from the kernel (via sigcontext), is the full 64bit.
> > >
> > > With this, the register will be exposed as 64-bit to the
> > > user.  Is that desirable?  The fact that the original
> > > description used 32-bit makes it sounds like it's not.
> > > What's the real register width at the (asm visible)
> > > machine level?
> >
> > There is no access to CPSR as a single register in AArch64 (*).
> > Instead, process states can be accessed/modified via system
> > instructions.  I guess the kernel synthesizes one CPSR register, so if
> > kernel defines it as a 64-bit register, it is reasonable for gdb to
> > treat it of the same size as well.
>
> Basing GDB's fundamentals on a particular OS's ptrace(2)
> implementation is a bad idea.

Do you have a suggestion on how to do it more properly?

> > Reference:
> >
> > * ARMv8 Instruction Set Overview - Chapter 3 A64 Overview
>
> Wow, is the ARMv8 instruction set documentation really only available
> to "ARM Custumers"?  How are we supposed to review patches for an
> architecture for which we don't have access to the fundamental
> documentation?

It is 'customer' in a broader sense.  I think the requirement on having 
an account is for the purpose of protecting ARM IP.  While it is fairly 
easy to open an account, I understand that it still causes 
inconvenience.  I hope the situation will be improved in the future.

Thanks,
Yufeng