Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 2 Sep 2021 14:02:38 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Lai Jiangshan <laijs@linux.alibaba.com>
Cc:     Lai Jiangshan <jiangshanlai@gmail.com>,
        linux-kernel@vger.kernel.org, Andy Lutomirski <luto@kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
        x86@kernel.org, "H. Peter Anvin" <hpa@zytor.com>,
        Joerg Roedel <jroedel@suse.de>,
        Youquan Song <youquan.song@intel.com>,
        Tony Luck <tony.luck@intel.com>
Subject: Re: [PATCH 22/24] x86/entry: Implement and use do_paranoid_entry()
 and paranoid_exit()
Message-ID: <YTC9Xv5K6npINezG@hirez.programming.kicks-ass.net>
References: <20210831175025.27570-1-jiangshanlai@gmail.com>
 <20210831175025.27570-23-jiangshanlai@gmail.com>
 <YTCoenvIaHjLQmAC@hirez.programming.kicks-ass.net>
 <465e2f89-402d-6f05-ab36-46a6f005f2b3@linux.alibaba.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <465e2f89-402d-6f05-ab36-46a6f005f2b3@linux.alibaba.com>
Precedence: bulk

On Thu, Sep 02, 2021 at 06:42:16PM +0800, Lai Jiangshan wrote:
> On 2021/9/2 18:33, Peter Zijlstra wrote:
> > On Wed, Sep 01, 2021 at 01:50:23AM +0800, Lai Jiangshan wrote:
> > 
> > > +	call	do_paranoid_entry
> > >   	ret
> > 
> > That's normally spelled like:
> > 
> > 	jmp do_paranoid_entry
> > 
> > But the same comment as for error_entry but more; pretty much all that's
> > left in asm is things like:
> > 
> > 
> > 	call paranoid_entry;
> > 
> > 	# setup args
> > 	call \cfunc
> > 
> > 	call paranoid_exit
> > 
> > which seems like prime material to also pull into C to avoid the
> > back-and-forth thing. In fact, why can't you call paranoid_entry/exit
> > from \cfunc itself? The IDT macros should be able to help.
> > 
> 
> It sounds better.
> 
> I should have moved the code of pushing pt_regs out of paranoid_entry(),
> so that I could also have seen this.
> (and we don't need do_paranoid_entry(), paranoid_entry() itself can be in C)
> 
> The \cfunc would need to be marked as entry_text, right?

Yes I think so. The distinction between .entry.text and .noinstr.text is
that the former it mapped into the userspace mapping, while the latter
is not. Seeing how the PTI swizzling still has to happen when calling
cfunc, that had bettern be .entry.text.

If we care about a strict minimum of .entry.text the IDT macros can
generate a noinstr function to be called the moment we've done the PTI
munging I suppose.