lkml.org 
[lkml]   [2014]   [Mar]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH 3/3] x86, vdso32: handle 32 bit vDSO larger one page
On Thu, Mar 13, 2014 at 11:21 AM, H. Peter Anvin <hpa@zytor.com> wrote:
> On 03/13/2014 11:08 AM, H. Peter Anvin wrote:
>> On 03/13/2014 10:28 AM, Andy Lutomirski wrote:
>>>
>>> Does this mean you prefer the relocation approach to the compat vdso
>>> removal approach? It seems like Linus is okay with either one.
>>>
>>
>> Actually, thinking about it, removing it is probably better:
>>
>> a) gets rid of legacy code, making room for unification;
>> b) either way enabling compat support (either relocation or disabling
>> the vdso) has a performance penalty for *all* processes.
>>
>> The only way to avoid that is to have a vdso at a fixed addresses across
>> all processes, either in the fixmap or in the user area (presumably at
>> the very top.)
>>
>
> So going back and re-reading all the threads, the consensus was to
> remove the compat vdso, but recycling the CONFIG_COMPAT_VDSO
> configuration option name for the default-disable option.
>

I don't recall anyone suggesting keeping the name. CONFIG_COMPAT_VDSO
will henceforth be a silly name, since the "compat" mode won't have a
vdso.

The consensus was that the default should be no vdso, right? My most
recent patchset calls it CONFIG_ENABLE_VDSO32_BY_DEFAULT and defaults
to n.

--Andy

> It is important that anyone who actually cares about performance unsets
> the option.

Should there be a warning at startup if the option is set?

--Andy


\
 
 \ /
  Last update: 2014-03-13 20:21    [W:0.155 / U:0.092 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site