Re: [PATCH v5 2/3] x86: Remap GDT tables in the Fixmap section

From: Andy Lutomirski <luto@amacapital.net>
To: Andrew Cooper <andrew.cooper3@citrix.com>
Cc: "Michal Hocko" <mhocko@suse.com>,
	"Stanislaw Gruszka" <sgruszka@redhat.com>,
	"kvm list" <kvm@vger.kernel.org>,
	"linux-doc@vger.kernel.org" <linux-doc@vger.kernel.org>,
	"Matt Fleming" <matt@codeblueprint.co.uk>,
	"Frederic Weisbecker" <fweisbec@gmail.com>,
	"Josh Poimboeuf" <jpoimboe@redhat.com>,
	"Chris Wilson" <chris@chris-wilson.co.uk>,
	"linux-mm@kvack.org" <linux-mm@kvack.org>,
	"Dave Hansen" <dave.hansen@intel.com>,
	"Radim Krčmář" <rkrcmar@redhat.com>,
	"linux-efi@vger.kernel.org" <linux-efi@vger.kernel.org>,
	"Alexander Potapenko" <glider@google.com>,
	"Pavel Machek" <pavel@ucw.cz>, "H . Peter Anvin" <hpa@zytor.com>,
	"kernel-hardening@lists.openwall.com"
	<kernel-hardening@lists.openwall.com>,
	"Jiri Olsa" <jolsa@redhat.com>,
	"Thomas Gleixner" <tglx@linutronix.de>,
	zijun_hu <zijun_hu@htc.com>,
	"Prarit Bhargava" <prarit@redhat.com>,
	"Andi Kleen" <ak@linux.intel.com>
Subject: Re: [PATCH v5 2/3] x86: Remap GDT tables in the Fixmap section
Date: Thu, 9 Mar 2017 13:46:20 -0800	[thread overview]
Message-ID: <CALCETrWv-u7OdjWDY+5eF7p-ngPun-yYf0QegMzYc6MGVQd-4w__46710.2304487073$1489096071$gmane$org@mail.gmail.com> (raw)
In-Reply-To: <17ffcc5b-1c9a-51b6-272a-5eaecf1bc0c4@citrix.com>

On Thu, Mar 9, 2017 at 1:43 PM, Andrew Cooper <andrew.cooper3@citrix.com> wrote:
> On 09/03/2017 21:32, Andy Lutomirski wrote:
>> On Mon, Mar 6, 2017 at 2:03 PM, Thomas Garnier <thgarnie@google.com> wrote:
>>
>>> --- a/arch/x86/xen/enlighten.c
>>> +++ b/arch/x86/xen/enlighten.c
>>> @@ -710,7 +710,7 @@ static void load_TLS_descriptor(struct thread_struct *t,
>>>
>>>         *shadow = t->tls_array[i];
>>>
>>> -       gdt = get_cpu_gdt_table(cpu);
>>> +       gdt = get_cpu_gdt_rw(cpu);
>>>         maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
>>>         mc = __xen_mc_entry(0);
>> Boris, is this right?  I don't see why it wouldn't be, but Xen is special.
>
> Under Xen PV, the GDT is already read-only at this point.  (It is not
> safe to let the guest have writeable access to system tables, so the
> guest must relinquish write access to the frames wishing to be used as
> LDTs or GDTs.)
>
> The hypercall acts on the frame, not a virtual address, so either alias
> should be fine here.
>
> Under this new scheme, there will be two read-only aliases.  I guess
> this is easier to maintain the split consistently across Linux, than to
> special case Xen PV because it doesn't need the second alias.
>

I think we would gain nothing at all by special-casing Xen PV -- Linux
allocates the fixmap vaddrs at compile time, so we'd still allocate
them even if we rejigger all the helpers to avoid using them.

--Andy

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel