Re: [PATCH 3/4] Add 32 bit VDSO time support for 32 bit kernel

["H. Peter Anvin" <hpa@zytor.com>]

Yes, that we can move, of course.  

On February 1, 2014 4:30:17 PM PST, Andy Lutomirski <luto@amacapital.net> wrote:
>On Sat, Feb 1, 2014 at 4:26 PM, H. Peter Anvin <hpa@zytor.com> wrote:
>> On 02/01/2014 03:59 PM, Andy Lutomirski wrote:
>>>
>>>   If it is, indeed, okay to use non-fixed maps on 32-bit, it might
>>> also be okay on 64-bit.  If so, it could be useful to implement
>that,
>>> which would remove a bit of a wart and allow PR_SET_TSC to work
>>> usefully for 64-bit userspace.  (This would remove the need for the
>>> VVAR macro and would allow shorter rip-relative address modes.)
>>>
>>
>> We can't really move the 64-bit legacy vsyscall area, though, as it
>is
>> an ABI.  It can be disabled with vsyscall=none, but Linus has
>vehemently
>> vetoed removing them.
>
>VVAR != vsyscall.  They've been different pages since I wrote the
>vsyscall emulation stuff.  Any userspace code that relies on any of
>the contents of the VVAR page is totally screwed already, since the
>layout changes semi-regularly and depends on whether lockdep is
>enabled.
>
>>
>>>  (Note that those fixmaps are a security problem on native 32-bit if
>>> NX is not available.  We may not care.)
>>
>> Not only on native 32 bit... although the amount of 64-bit hardware
>> without NX is quite small, the same is true for anywhere near modern
>> 32-bit hardware.
>
><irrelevant>It can't be a problem for 32-bit compat mode, though,
>since userspace can't address the fixmap anyway.</irrelevant>
>
>>
>>>>
>>>> -#define VDSO_HIGH_BASE         0xffffe000U /* CONFIG_COMPAT_VDSO
>address */
>>>> +#define VDSO_HIGH_BASE         0xffffc000U /* CONFIG_COMPAT_VDSO
>address */
>>>
>>> This is odd.  Can you explain it?
>>>
>>
>> He needs 3 pages instead of 1 after his changes.
>
>Right.  But there's some obscure ABI reason for CONFIG_COMPAT_VDSO,
>and if this breaks it, then it's no good.  From extremely vague
>memory, there's some version of SuSE that breaks if the 32-bit vdso
>moves.  I have no idea what the bug is, but moving a "compat" address
>seems suspect.
>
>--Andy

-- 
Sent from my mobile phone.  Please pardon brevity and lack of formatting.