Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 22 Mar 2022 13:07:25 +0100
From:   Peter Zijlstra <peterz@infradead.org>
To:     Masami Hiramatsu <mhiramat@kernel.org>
Cc:     Stephen Rothwell <sfr@canb.auug.org.au>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>,
        "H. Peter Anvin" <hpa@zytor.com>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Linux Next Mailing List <linux-next@vger.kernel.org>,
        rostedt@goodmis.org, ast@kernel.org, hjl.tools@gmail.com,
        rick.p.edgecombe@intel.com, rppt@kernel.org,
        linux-toolchains@vger.kernel.org, Andrew.Cooper3@citrix.com,
        ndesaulniers@google.com
Subject: Re: linux-next: build warnings after merge of the tip tree
Message-ID: <Yjm7/YbuUwjEP43r@hirez.programming.kicks-ass.net>
References: <20220321140327.777f9554@canb.auug.org.au>
 <Yjh11UjDZogc3foM@hirez.programming.kicks-ass.net>
 <Yjh3xZuuY3QcZ1Bn@hirez.programming.kicks-ass.net>
 <YjisdqdofbDIYj2U@hirez.programming.kicks-ass.net>
 <20220322143136.0e78366c3521b54b7b9385b8@kernel.org>
 <YjmD9ks5OAiZkjoL@hirez.programming.kicks-ass.net>
 <20220322181454.659b15269d8c2e2348f19ba1@kernel.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220322181454.659b15269d8c2e2348f19ba1@kernel.org>
Precedence: bulk

On Tue, Mar 22, 2022 at 06:14:54PM +0900, Masami Hiramatsu wrote:
> On Tue, 22 Mar 2022 09:08:22 +0100
> Peter Zijlstra <peterz@infradead.org> wrote:
> 
> > On Tue, Mar 22, 2022 at 02:31:36PM +0900, Masami Hiramatsu wrote:
> > 
> > > > Also, I think both should fix regs->ss.
> > > 
> > > I'm not sure this part. Since the return trampoline should run in the same
> > > context of the called function, isn't ss same there too?
> > 
> > It creates pt_regs on the stack, so the trampolines do:
> > 
> > 	push $arch_rethook_trampoline
> > 	push %rsp
> > 	pushf
> > 	sub $24, %rsp /* cs, ip, orig_ax */
> > 	push %rdi
> > 	...
> > 	push %r15
> > 
> > That means that if anybody looks at regs->ss, it'll find
> > $arch_rethook_trampoline, which isn't a valid segment descriptor, or am
> > I just really bad at counting today?
> 
> Ah, got it. It seems that the ss was skipped from the beginning, and
> no one argued that.

Yeah, this is a long-standing issue, but I noticed it when looking at
the code yesterday.

> > I'm thinking you want a copy of __KERNEL_DS in that stack slot, not a
> > function pointer.
> 
> The function pointer is for unwinding stack which involves the kretprobe.
> Anyway, I can add a slot for ss if it is neeeded. But if it always be
> __KERNEL_DS, is it worth to save it?

Probably, to save someone future head-aches. The insn-eval.c stuff will
actually look at SS when it tries to decode BP/SP fields, and I've got
vague memories of actually using that a while ago. I think I was playing
around with double-fault and the whole espfix64 mess and hit the ESPFIX
segment.