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;
MIME-Version: 1.0
References: <20220925175356.681-1-jszhang@kernel.org> <20220925175356.681-4-jszhang@kernel.org>
 <CAJF2gTRFXwu5QtA8GywGaSVRPA7zeexcDHRzcngcY1zD3MFjog@mail.gmail.com> <YzJBXsiWrbmxM6Q2@xhacker>
In-Reply-To: <YzJBXsiWrbmxM6Q2@xhacker>
From:   Guo Ren <guoren@kernel.org>
Date:   Tue, 27 Sep 2022 10:00:19 +0800
Message-ID: <CAJF2gTRkPc7AQsUNZopevey61PD8o6BQBZKjgeLHmOgX+pj6dA@mail.gmail.com>
Subject: Re: [PATCH 3/4] riscv: fix race when vmap stack overflow and remove shadow_stack
To:     Jisheng Zhang <jszhang@kernel.org>
Cc:     Paul Walmsley <paul.walmsley@sifive.com>,
        Palmer Dabbelt <palmer@dabbelt.com>,
        Albert Ou <aou@eecs.berkeley.edu>,
        Nathan Chancellor <nathan@kernel.org>,
        Nick Desaulniers <ndesaulniers@google.com>,
        linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org,
        llvm@lists.linux.dev
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Sep 27, 2022 at 8:28 AM Jisheng Zhang <jszhang@kernel.org> wrote:
>
>
> > >  #ifdef CONFIG_VMAP_STACK
> > > -static DEFINE_PER_CPU(unsigned long [OVERFLOW_STACK_SIZE/sizeof(long)],
> > > -               overflow_stack)__aligned(16);
> > > -/*
> > > - * shadow stack, handled_ kernel_ stack_ overflow(in kernel/entry.S) is used
> > > - * to get per-cpu overflow stack(get_overflow_stack).
> > > - */
> > > -long shadow_stack[SHADOW_OVERFLOW_STACK_SIZE/sizeof(long)];
> > > -asmlinkage unsigned long get_overflow_stack(void)
> > > -{
> > > -       return (unsigned long)this_cpu_ptr(overflow_stack) +
> > > -               OVERFLOW_STACK_SIZE;
> > > -}
> > > +unsigned long overflow_stack[NR_CPUS][OVERFLOW_STACK_SIZE/sizeof(long)] __aligned(16);
>
> If NR_CPUS is large, there's a non-trival memory waste, I have a
> solution for this case, will send a new version today.
Er...

Yes, we can't bypass the percpu mechanism. I also forgot the percpu
basic concept. In the end, I prefer the previous solution, maybe just
simply giving an atomic flag would be okay. (But we only have one
register (sp) which could be used, it seems not simple.)

>
> Thanks


-- 
Best Regards
 Guo Ren