Re: [PATCH 11/24] x86/entry: Replace the most of asm code of error_entry to C code

[Lai Jiangshan]

On 2021/9/2 18:16, Peter Zijlstra wrote:
> On Wed, Sep 01, 2021 at 01:50:12AM +0800, Lai Jiangshan wrote:
> > diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S
> > index 42d2918f5646..bc9e2f5ad370 100644
> > --- a/arch/x86/entry/entry_64.S
> > +++ b/arch/x86/entry/entry_64.S
> > @@ -972,83 +972,14 @@ SYM_CODE_START_LOCAL(error_entry)
> >   	cld
> >   	PUSH_AND_CLEAR_REGS save_ret=1
> >   	ENCODE_FRAME_POINTER 8
> >   
> >   	popq	%r12				/* save return addr in %12 */
> >   	movq	%rsp, %rdi			/* arg0 = pt_regs pointer */
> > +	call	do_error_entry
> >   	movq	%rax, %rsp			/* switch stack */
> >   	ENCODE_FRAME_POINTER
> >   	pushq	%r12
> >   	ret
>
> There's only a single error_entry callsite, which is idtentry_body. One
> of the things I wanted to do is change this lot so we change to the
> task_stack in 'C', using an adaptation of call_on_irqstack() and
> basically don't return frrom C until we're done with \cfunc.
>
> That is, once we call C, stay there, and don't do this back and forth
> between C and asm.

I haven't figured out how can an adaptation of call_on_irqstack() can do it.
The original stack need to be "free" for next task.  And we can't switch
the stack before error_entry() since the CR3 is not switched.

I believe the ASM code here can be simplified and clearer further. But I don't
think going back and forth between C and ASM is real issue if the ASM code is
short and simple enough.


> As is, the resulting asm in error_entry is somewhat confusing given that
> we sometimes don't actually switch stacks.