Messages in this thread | | | Date | Fri, 24 Feb 2012 00:04:30 +0100 | From | Willy Tarreau <> | Subject | Re: [PATCH 0/5] i387: stable kernel backport |
| |
On Thu, Feb 23, 2012 at 02:55:11PM -0800, H. Peter Anvin wrote: > On 02/23/2012 02:52 PM, Willy Tarreau wrote: > > On Thu, Feb 23, 2012 at 02:48:51PM -0800, H. Peter Anvin wrote: > >> On 02/23/2012 02:38 PM, Linus Torvalds wrote: > >>> > >>> You'd still need an x86-32 machine to test on, because x86-64 was > >>> immune to this issue. > >>> > >>> But yeah, the impact of this seems to be small enough that for older > >>> kernels (which are likely used on older systems for maintenance > >>> anyway) disabling AES-NI on x86-32 really might be the way to go. > >>> > >> > >> That would really suck for users of encrypted hard disks. > > > > Peter, do you really think there are that many ? I think I only saw > > AES-NI on recent 64-bit capable chips, and it's been a while that > > users have been installing 64-bit distros on such machines. Note that > > I'm not advocating for breaking existing setups, just that I'm surprized > > by this combination (aes-ni + 32-bit). > > > > There are still people running 32-bit systems because they have some odd > compatibility constraints but now have to deal with corporate or other > security constraints; they may also have been using disk encryption > since before AES-NI was in but doing it on the integer side is way slower.
Indeed the combination looks plausible :-)
> This is not AES-NI in the interrupt path, but I don't think there is a > knob for that.
OK. So let's hope this works then !
Willy
| |