Re: [PATCH -next 0/3] replace open coded VA->PA calculation

From: Linus Walleij <linus.walleij@linaro.org>
To: Ard Biesheuvel <ardb@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>,
	cuigaosheng <cuigaosheng1@huawei.com>,
	Russell King - ARM Linux <linux@armlinux.org.uk>,
	Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com>,
	Andrew Lunn <andrew@lunn.ch>,
	Gregory CLEMENT <gregory.clement@bootlin.com>,
	Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>,
	Viresh Kumar <vireshk@kernel.org>,
	Shiraz Hashim <shiraz.linux.kernel@gmail.com>,
	SoC Team <soc@kernel.org>,
	Linux ARM <linux-arm-kernel@lists.infradead.org>,
	"moderated list:ARM/SAMSUNG EXYNOS ARM ARCHITECTURES" 
	<linux-samsung-soc@vger.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	gongruiqi1@huawei.com, wangweiyang2@huawei.com
Subject: Re: [PATCH -next 0/3] replace open coded VA->PA calculation
Date: Fri, 24 Dec 2021 05:05:46 +0100	[thread overview]
Message-ID: <CACRpkdafoc_erHyfkby9X1LhkYgkBQsad=qNyXh+Tq_32zr_yg@mail.gmail.com> (raw)
In-Reply-To: <CAMj1kXHgptq47g3mSPTKW9KMnqTH4K2Q7riBVwX4iA1V9YWYiA@mail.gmail.com>

On Wed, Dec 22, 2021 at 10:30 AM Ard Biesheuvel <ardb@kernel.org> wrote:
> On Wed, 22 Dec 2021 at 03:31, Linus Walleij <linus.walleij@linaro.org> wrote:
> > On Tue, Dec 21, 2021 at 10:16 AM Arnd Bergmann <arnd@arndb.de> wrote:
> >
> > > I think for CPUs that can use LPAE, we want to eventually move to the 4G:4G
> > > memory model, which in turn depends on having the kernel in vmalloc space, as
> > > implemented by Linus Walleij in
> > > https://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-integrator.git/log/?h=kernel-in-vmalloc-v5.14-rc1
> >
> > Yeah I'm still working on this series. The 4/4 split works pretty well,

Actually it's just kernel-in-vmalloc, 4/4 comes after that.

> > but it breaks KASan and I was in the process of fixing that when I left
> > for parental leave.
> >
> > The reason it breaks would be related to KASan not expecting
> > executable code in the vmalloc area, and since the shadowing
> > algorithm is a simple pointer offset, and since we go to lengths
> > to chisel shadow memory out of lowmem at a fixed offset from
> > TEXT_OFFSET, we have a problem.
> >
>
> Vmap'ed stacks actually has a similar problem, which is why it is
> disabled when KAsan is enabled. But this can be fixed by enabling arch
> support for KASAN_VMALLOC, and I suspect it may address the vmap'ed
> kernel as well.

Yep after seeing the other convo on the topic I realized that this
is indeed the same as I'm seeing. I can't disable KASAN just
because the kernel is in VMALLOC though, so I suppose when
I finally get back to this I have to fix KASAN_VMALLOC too
if noone beats me to it.

(It'd be great if someone could beat me to it...)

Yours,
Linus Walleij