Re: [RFC][PATCH 00/10] Use global pages with PTI

[Linus Torvalds]

On Fri, Feb 23, 2018 at 5:49 PM, Dave Hansen
<dave.hansen@xxxxxxxxxxxxxxx> wrote:
> On 02/22/2018 01:52 PM, Linus Torvalds wrote:
>> Side note - and this may be crazy talk - I wonder if it might make
>> sense to have a mode where we allow executable read-only kernel pages
>> to be marked global too (but only in the kernel mapping).
>
> We did that accidentally, somewhere.  It causes machine checks on K8's
> iirc, which is fun (52994c256df fixed it).  So, we'd need to make sure
> we avoid it there, or just make it global in the user mapping too.

They'd be missing _entirely_ in the user maps, which should be fine.
The problem that commit 52994c256df3 fixed was that they actually
existed in the user maps, just with different data, and then you can
have a ITLB and a DTLB entry for the same address that don't match
(because one has been loaded from the kernel mapping and the other
from the user one).

But when the address isn't mapped at all in the user map, that should
be fine - because there's no associated TLB entry to get mixed up
about.

It's no different from clearing a page from the page table before then
flushing the TLB entry later - which is the normal (and required)
behavior for unmapping a page. For a while it exists in the TLB
without existing in the page tables.

> Just for fun, I tried a 4-core Skylake system with KPTI and nopcid and
> compiled a random kernel 10 times.  I did three configs: no global, all
> kernel text global + cpu_entry_area, and only cpu_entry_area + entry
> text.  The delta percentages are from the Baseline.  The deltas are
> measurable, but the largest bang for our buck is obviously the entry text.
>
>                         User Time       Kernel Time     Clock Elapsed
> Baseline (33 GLB PTEs)  907.6           81.6            264.7
> Entry    (28 GLB PTEs)  910.9 (+0.4%)   84.0 (+2.9%)    265.2 (+0.2%)
> No global( 0 GLB PTEs)  914.2 (+0.7%)   89.2 (+9.3%)    267.8 (+1.2%)

That's actually noticeable. Maybe not so much in the final elapsed
time itself, but almost 3% for just the kernel side sounds meaningful.

Of course, that's with nopcid, so it's a fairly small special case, but still.

> It's a single line of code to go from the "33" to "28" configuration, so
> it's totally doable.  But, it means having and parsing another boot
> option that confuses people and then I have to go write actual
> documentation, which I detest. :)

Heh.

Ok, maybe the complexity isn't in the code, but in the concept.

               Linus