| 1 |
On Fri, Jan 12, 2018 at 2:58 PM, Corbin Bird <corbinbird@×××××××.net> wrote: |
| 2 |
> |
| 3 |
> The Fam16h and Fam17h microcode updates were new to Gentoo? |
| 4 |
> I don't recall ever seeing them before. |
| 5 |
> |
| 6 |
|
| 7 |
They're new in general - they first appeared last week and they're |
| 8 |
being treated as if they're related to Spectre. I've yet to see any |
| 9 |
kind of official release of them, but that seems to be par for the |
| 10 |
course for AMD the more I hunt around for documentation. It seems |
| 11 |
like Suse first released it, and claimed that it disabled branch |
| 12 |
prediction, whatever that means (certainly it can't completely disable |
| 13 |
it without tanking performance). |
| 14 |
|
| 15 |
I'm not sure if it has been merged upstream yet. The 17h loader |
| 16 |
updates were merged into linux 4.4.13 (Gentoo backported it as well), |
| 17 |
but I'm not sure about the microcode itself. |
| 18 |
|
| 19 |
Separately there are some lkml patches proposed by AMD to use an MSR |
| 20 |
setting to make LFENCE block speculative execution. It looks like the |
| 21 |
strategy is to enable that on amd64, and verify that the setting took |
| 22 |
(a guest VM wouldn't be able to set the MSR). If the setting isn't |
| 23 |
effective then it would fall back to Retpolines (which is the Intel |
| 24 |
fix for Spectre), which have more overhead (this is in addition to the |
| 25 |
much larger overhead for the meltdown fix on Intel). |
| 26 |
|
| 27 |
If somebody actually sees anything official from AMD clearly giving a |
| 28 |
checklist for Spectre remediation I'm all ears. To its credit, Intel |
| 29 |
at least published one of those (even if it amounts to "pound sand" |
| 30 |
for older CPUs). |
| 31 |
|
| 32 |
-- |
| 33 |
Rich |
Archives