Re: [linux41] regression with 'x86/cacheinfo: Move cacheinfo sysfs code to generic infrastructure' on AMD i686

[Greg KH <gregkh@linuxfoundation.org>]

On Thu, Sep 17, 2015 at 09:15:04AM +0200, Borislav Petkov wrote:
> On Wed, Sep 16, 2015 at 07:52:47PM -0400, Josh Boyer wrote:
> > On Sun, Jul 26, 2015 at 10:42 AM, Borislav Petkov <bp@alien8.de> wrote:
> > > On Sun, Jul 26, 2015 at 12:54:55PM +0200, Philip Müller wrote:
> > >> I can confirm your patch working. However, it might be good to use yours
> > >> and Thomas' in combination to solve this properly.
> > >
> > > Please do not top-post.
> > >
> > > We could use Thomas' too although from looking at it,
> > > detect_cache_attributes() allocates a per-CPU per_cpu_cacheinfo thing
> > > for each CPU. By the time we hit cache_shared_cpu_map_remove() in
> > > free_cache_attributes(), those per_cpu_cacheinfo(cpu) things are still
> > > allocated. We kfree them in the next step only.
> > >
> > > But I like the moving of the check from the CPU hotplug callback to
> > > free_cache_attributes().
> > >
> > > So I'll merge the two patches and write up a proper commit message,
> > > unless someone objects.
> > >
> > > I'll add your Tested-by too.
> > 
> > Did this actually happen?  I don't see either fix in Linus' tree yet,
> > the merge window is closed, and the bug happens on 4.1 and 4.2 stable
> > kernels..
> 
> Greg wanted to pick it up...
> 
> Greg, what's up?

It's in my "to-apply" queue, let me go dig it up now...

thanks for the reminder.

greg k-h