Re: [PATCH v4 4/8] arm: use fixmap for text patching when text is RO

From: Will Deacon <will.deacon@arm.com>
To: Kees Cook <keescook@chromium.org>
Cc: "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	Rabin Vincent <rabin@rab.in>, Rob Herring <robh@kernel.org>,
	Laura Abbott <lauraa@codeaurora.org>,
	Leif Lindholm <leif.lindholm@linaro.org>,
	Stephen Boyd <sboyd@codeaurora.org>,
	"msalter@redhat.com" <msalter@redhat.com>,
	Liu hua <sdu.liu@huawei.com>,
	Nikolay Borisov <Nikolay.Borisov@arm.com>,
	Nicolas Pitre <nicolas.pitre@linaro.org>,
	Tomasz Figa <t.figa@samsung.com>,
	Doug Anderson <dianders@google.com>,
	Jason Wessel <jason.wessel@windriver.com>,
	Catalin Marinas <Catalin.Marinas@arm.com>,
	Russell King - ARM Linux <linux@arm.linux.org.uk>,
	"linux-arm-kernel@lists.infradead.org" 
	<linux-arm-kernel@lists.infradead.org>,
	"linux-doc@vger.kernel.org" <linux-doc@vger.kernel.org>
Subject: Re: [PATCH v4 4/8] arm: use fixmap for text patching when text is RO
Date: Thu, 4 Sep 2014 10:27:21 +0100	[thread overview]
Message-ID: <20140904092720.GA7156@arm.com> (raw)
In-Reply-To: <CAGXu5jJ890ruLO8Nit2exMSWigSM71kfZOkLoF0DcY1viFiX_A@mail.gmail.com>

On Wed, Sep 03, 2014 at 10:43:58PM +0100, Kees Cook wrote:
> On Wed, Aug 20, 2014 at 5:28 AM, Kees Cook <keescook@chromium.org> wrote:
> > On Tue, Aug 19, 2014 at 7:29 AM, Will Deacon <will.deacon@arm.com> wrote:
> >> On Wed, Aug 13, 2014 at 06:06:29PM +0100, Kees Cook wrote:
> >>> +static void __kprobes *patch_map(void *addr, int fixmap, unsigned long *flags)
> >>> +     __acquires(&patch_lock)
> >>> +{
> >>> +     unsigned int uintaddr = (uintptr_t) addr;
> >>> +     bool module = !core_kernel_text(uintaddr);
> >>> +     struct page *page;
> >>> +
> >>> +     if (module && IS_ENABLED(CONFIG_DEBUG_SET_MODULE_RONX))
> >>> +             page = vmalloc_to_page(addr);
> >>> +     else if (!module && IS_ENABLED(CONFIG_DEBUG_RODATA))
> >>> +             page = virt_to_page(addr);
> >>> +     else
> >>> +             return addr;
> >>> +
> >>> +     if (flags)
> >>> +             spin_lock_irqsave(&patch_lock, *flags);
> >>> +     else
> >>> +             __acquire(&patch_lock);
> >>
> >> I don't understand the locking here. Why is it conditional, why do we need
> >> to disable interrupts, and are you just racing against yourself?
> >
> > AIUI, the locking is here to avoid multiple users of the text poking
> > fixmaps. It's conditional because there are two fixmaps
> > (FIX_TEXT_POKE0 and FIX_TEXT_POKE1). Locking happens around 0 so
> > locking around 1 is not needed since it is only ever used when 0 is in
> > use. (__patch_text_real locks patch_lock before setting 0 when it uses
> > remapping, and if it also needs 1, it doesn't have to lock since the
> > lock is already held.)
> >
> >>> +     set_fixmap(fixmap, page_to_phys(page));
> >>
> >> set_fixmap does TLB invalidation, right? I think that means it can block on
> >> 11MPCore and A15 w/ the TLBI erratum, so it's not safe to call this with
> >> interrupts disabled anyway.
> >
> > Oh right. Hrm.
> >
> > In an earlier version of this series set_fixmap did not perform TLB
> > invalidation. I wonder if this is not needed at all? (Wouldn't that be
> > nice...)
> 
> As suspected, my tests fail spectacularly without the TLB flush.
> Adding WARN_ON(!irqs_disabled()) doesn't warn, so I think we're safe
> here. Should I leave the WARN_ON in place for clarity, or some other
> comments?

I thought there was a potential call to spin_lock_irqsave right before
this TLB flush?

Will