Re: [PATCH v6 6/6] drivers/node: Show in sysfs node's crypto capabilities

[Tom Lendacky]

On 2/4/22 07:21, Martin Fernandez wrote:
> On 2/4/22, Limonciello, Mario <mario.limonciello@xxxxxxx> wrote:
> > On 2/3/2022 10:43, Martin Fernandez wrote:
> > > +static ssize_t crypto_capable_show(struct device *dev,
> > > +				   struct device_attribute *attr, char *buf)
> > > +{
> > > +	struct pglist_data *pgdat = NODE_DATA(dev->id);
> > > +
> > > +	return sysfs_emit(buf, "%d\n", pgdat->crypto_capable);
> >
> > As there is interest in seeing these capabilities from userspace, it
> > seems like a logical time to also expose a `crypto_active` attribute.
>
> I planned to do something similar to this, but to show (or actually
> hide if inactive) tme in cpuinfo, just as Borislav Petkov suggested a
> few versions back.
>
> https://lore.kernel.org/linux-efi/YXrnkxgdjWbcPlJA@xxxxxxx/
>
> > Then userspace can make a judgement call if the system supports crypto
> > memory (`crypto_capable`) and then also whether or not it's been turned
> > on (`crypto_active`).
> >
> > `crypto_active` could be detected with some existing support in the
> > kernel of `mem_encrypt_active()`.  This will then work for a variety of
> > architectures too that offer `mem_encrypt_active()`.
>
> I need a hand with this, I grepped for mem_encrypt_active and nothing
> showed up...

The mem_encrypt_active() function has been replaced by 
cc_platform_has(CC_ATTR_MEM_ENCRYPT).

> > As it stands today the only reliable way to tell from userspace (at
> > least for AMD's x86 implementation) is by grepping the system log for
> > the line "AMD Memory Encryption Features active".
>
> Isn't enough to grep for sme/sev in cpuinfo?

No, it's not enough. Cpuinfo shows a processors capabilities and not 
necessarily whether that capability is being used.

Thanks,
Tom