Re: L1TFv8

Re: L1TFv8

[Thomas Gleixner]

On Wed, 13 Jun 2018, speck for Andi Kleen wrote:

Thanks for adding Message-IDs this time. Though now the remailer messed it
up due to a recent change dealing with windooze mails.. Fixed now. I'm
replaying the series. Sorry for the extra spam.

Thanks,

	tglx

Re: L1TFv8

[Thomas Gleixner]

[-- Attachment #1: Type: text/plain, Size: 718 bytes --]

On Wed, 13 Jun 2018, speck for Andi Kleen wrote:

> This is v8 of the native OS patchkit to mitigate the L1TF side channel.
> It does not cover KVM.
> 
> This version addresses the latest review feedbacks. The mitigation
> setup has been moved into check_bugs, and the memory size checking
> patch is now integrated into the standard setup. The swap mitigation
> has been split into two patches. Various other changes.
> For more details see the individual changelogs.

I've picked up the whole set with a few edits and applied it to the master
branch of the speck git repository and backported it to the linux-4.17.y,
linux-4.16.y and linux-4.14.y branches.

Git bundle against Linus tree is attached.

Thanks,

	tglx

[-- Attachment #2: Type: application/octet-stream, Size: 17631 bytes --]

[MODERATED] Re: L1TFv8

[Josh Poimboeuf]

On Fri, Jun 15, 2018 at 04:50:38PM +0200, speck for Thomas Gleixner wrote:
> On Wed, 13 Jun 2018, speck for Andi Kleen wrote:
> 
> > This is v8 of the native OS patchkit to mitigate the L1TF side channel.
> > It does not cover KVM.
> > 
> > This version addresses the latest review feedbacks. The mitigation
> > setup has been moved into check_bugs, and the memory size checking
> > patch is now integrated into the standard setup. The swap mitigation
> > has been split into two patches. Various other changes.
> > For more details see the individual changelogs.
> 
> I've picked up the whole set with a few edits and applied it to the master
> branch of the speck git repository and backported it to the linux-4.17.y,
> linux-4.16.y and linux-4.14.y branches.
> 
> Git bundle against Linus tree is attached.

Has anyone tested these patches on a multi-node system with NUMA
balancing enabled?  We've see some panics on RHEL because _PAGE_NUMA ==
_PAGE_PROTNONE.  Not sure if this problem also exists upstream.

Andrea is looking into it (and I think he'll be joining this list
shortly).

-- 
Josh

Re: L1TFv8

[Thomas Gleixner]

On Fri, 15 Jun 2018, speck for Josh Poimboeuf wrote:
> On Fri, Jun 15, 2018 at 04:50:38PM +0200, speck for Thomas Gleixner wrote:
> > On Wed, 13 Jun 2018, speck for Andi Kleen wrote:
> > 
> > > This is v8 of the native OS patchkit to mitigate the L1TF side channel.
> > > It does not cover KVM.
> > > 
> > > This version addresses the latest review feedbacks. The mitigation
> > > setup has been moved into check_bugs, and the memory size checking
> > > patch is now integrated into the standard setup. The swap mitigation
> > > has been split into two patches. Various other changes.
> > > For more details see the individual changelogs.
> > 
> > I've picked up the whole set with a few edits and applied it to the master
> > branch of the speck git repository and backported it to the linux-4.17.y,
> > linux-4.16.y and linux-4.14.y branches.
> > 
> > Git bundle against Linus tree is attached.
> 
> Has anyone tested these patches on a multi-node system with NUMA
> balancing enabled?  We've see some panics on RHEL because _PAGE_NUMA ==
> _PAGE_PROTNONE.  Not sure if this problem also exists upstream.
> 
> Andrea is looking into it (and I think he'll be joining this list
> shortly).

He's subscribed, so he should have received your mail already.

Thanks,

	tglx

[MODERATED] Re: L1TFv8

[Andi Kleen]

On Fri, Jun 15, 2018 at 02:48:23PM -0500, speck for Josh Poimboeuf wrote:
> On Fri, Jun 15, 2018 at 04:50:38PM +0200, speck for Thomas Gleixner wrote:
> > On Wed, 13 Jun 2018, speck for Andi Kleen wrote:
> > 
> > > This is v8 of the native OS patchkit to mitigate the L1TF side channel.
> > > It does not cover KVM.
> > > 
> > > This version addresses the latest review feedbacks. The mitigation
> > > setup has been moved into check_bugs, and the memory size checking
> > > patch is now integrated into the standard setup. The swap mitigation
> > > has been split into two patches. Various other changes.
> > > For more details see the individual changelogs.
> > 
> > I've picked up the whole set with a few edits and applied it to the master
> > branch of the speck git repository and backported it to the linux-4.17.y,
> > linux-4.16.y and linux-4.14.y branches.
> > 
> > Git bundle against Linus tree is attached.
> 
> Has anyone tested these patches on a multi-node system with NUMA
> balancing enabled?  We've see some panics on RHEL because _PAGE_NUMA ==
> _PAGE_PROTNONE.  Not sure if this problem also exists upstream.

Not sure if there was a test with numa balancing on. I'll check.
-Andi

[MODERATED] Re: L1TFv8

[Andi Kleen]

On Fri, Jun 15, 2018 at 02:48:23PM -0500, speck for Josh Poimboeuf wrote:
> On Fri, Jun 15, 2018 at 04:50:38PM +0200, speck for Thomas Gleixner wrote:
> > On Wed, 13 Jun 2018, speck for Andi Kleen wrote:
> > 
> > > This is v8 of the native OS patchkit to mitigate the L1TF side channel.
> > > It does not cover KVM.
> > > 
> > > This version addresses the latest review feedbacks. The mitigation
> > > setup has been moved into check_bugs, and the memory size checking
> > > patch is now integrated into the standard setup. The swap mitigation
> > > has been split into two patches. Various other changes.
> > > For more details see the individual changelogs.
> > 
> > I've picked up the whole set with a few edits and applied it to the master
> > branch of the speck git repository and backported it to the linux-4.17.y,
> > linux-4.16.y and linux-4.14.y branches.
> > 
> > Git bundle against Linus tree is attached.
> 
> Has anyone tested these patches on a multi-node system with NUMA
> balancing enabled?  We've see some panics on RHEL because _PAGE_NUMA ==
> _PAGE_PROTNONE.  Not sure if this problem also exists upstream.
> 
> Andrea is looking into it (and I think he'll be joining this list
> shortly).

I did some testing and autonuma seems to work.

Shouldn't really anything special about numa hinting ptes. They get
inverted of course, but the access macros all handle it.

-Andi

[MODERATED] Re: L1TFv8

[Michal Hocko]

On Fri 15-06-18 14:48:23, speck for Josh Poimboeuf wrote:
> On Fri, Jun 15, 2018 at 04:50:38PM +0200, speck for Thomas Gleixner wrote:
> > On Wed, 13 Jun 2018, speck for Andi Kleen wrote:
> > 
> > > This is v8 of the native OS patchkit to mitigate the L1TF side channel.
> > > It does not cover KVM.
> > > 
> > > This version addresses the latest review feedbacks. The mitigation
> > > setup has been moved into check_bugs, and the memory size checking
> > > patch is now integrated into the standard setup. The swap mitigation
> > > has been split into two patches. Various other changes.
> > > For more details see the individual changelogs.
> > 
> > I've picked up the whole set with a few edits and applied it to the master
> > branch of the speck git repository and backported it to the linux-4.17.y,
> > linux-4.16.y and linux-4.14.y branches.
> > 
> > Git bundle against Linus tree is attached.
> 
> Has anyone tested these patches on a multi-node system with NUMA
> balancing enabled?  We've see some panics on RHEL because _PAGE_NUMA ==
> _PAGE_PROTNONE.  Not sure if this problem also exists upstream.
> 
> Andrea is looking into it (and I think he'll be joining this list
> shortly).

The numa balancing pte encoding has changed not that long ago. Have a
look at 21d9ee3eda77 and the pile from the same patch series (sorry I do
not have the reference handy).
-- 
Michal Hocko
SUSE Labs

[MODERATED] Re: L1TFv8

[Konrad Rzeszutek Wilk]

On Tue, Jun 19, 2018 at 04:54:56PM +0200, speck for Michal Hocko wrote:
> On Fri 15-06-18 14:48:23, speck for Josh Poimboeuf wrote:
> > On Fri, Jun 15, 2018 at 04:50:38PM +0200, speck for Thomas Gleixner wrote:
> > > On Wed, 13 Jun 2018, speck for Andi Kleen wrote:
> > > 
> > > > This is v8 of the native OS patchkit to mitigate the L1TF side channel.
> > > > It does not cover KVM.
> > > > 
> > > > This version addresses the latest review feedbacks. The mitigation
> > > > setup has been moved into check_bugs, and the memory size checking
> > > > patch is now integrated into the standard setup. The swap mitigation
> > > > has been split into two patches. Various other changes.
> > > > For more details see the individual changelogs.
> > > 
> > > I've picked up the whole set with a few edits and applied it to the master
> > > branch of the speck git repository and backported it to the linux-4.17.y,
> > > linux-4.16.y and linux-4.14.y branches.
> > > 
> > > Git bundle against Linus tree is attached.


Hi Thomas,

I see you pushed that in the tree but I think you need to rebase, mainly:

commit 43d5c223374710cc41617ca6e33284eb1f646277 (speck/master, l1tf)
Author: speck for Andi Kleen <speck@linutronix.de>		<== 'speck for'
Date:   Wed Jun 13 15:48:28 2018 -0700

    x86/speculation/l1tf: Limit swap file size to MAX_PA/2

is on them.

Re: L1TFv8

[Thomas Gleixner]

On Wed, 20 Jun 2018, speck for Konrad Rzeszutek Wilk wrote:

> On Tue, Jun 19, 2018 at 04:54:56PM +0200, speck for Michal Hocko wrote:
> > On Fri 15-06-18 14:48:23, speck for Josh Poimboeuf wrote:
> > > On Fri, Jun 15, 2018 at 04:50:38PM +0200, speck for Thomas Gleixner wrote:
> > > > On Wed, 13 Jun 2018, speck for Andi Kleen wrote:
> > > > 
> > > > > This is v8 of the native OS patchkit to mitigate the L1TF side channel.
> > > > > It does not cover KVM.
> > > > > 
> > > > > This version addresses the latest review feedbacks. The mitigation
> > > > > setup has been moved into check_bugs, and the memory size checking
> > > > > patch is now integrated into the standard setup. The swap mitigation
> > > > > has been split into two patches. Various other changes.
> > > > > For more details see the individual changelogs.
> > > > 
> > > > I've picked up the whole set with a few edits and applied it to the master
> > > > branch of the speck git repository and backported it to the linux-4.17.y,
> > > > linux-4.16.y and linux-4.14.y branches.
> > > > 
> > > > Git bundle against Linus tree is attached.
> 
> 
> Hi Thomas,
> 
> I see you pushed that in the tree but I think you need to rebase, mainly:
> 
> commit 43d5c223374710cc41617ca6e33284eb1f646277 (speck/master, l1tf)
> Author: speck for Andi Kleen <speck@linutronix.de>		<== 'speck for'
> Date:   Wed Jun 13 15:48:28 2018 -0700
> 
>     x86/speculation/l1tf: Limit swap file size to MAX_PA/2

Blush. /me goes to fix

Re: L1TFv8

[Thomas Gleixner]

[-- Attachment #1: Type: text/plain, Size: 352 bytes --]

On Wed, 20 Jun 2018, speck for Thomas Gleixner wrote:
> Blush. /me goes to fix

The branches are rebased to 4.18-rc1 and the latest stable versions. The
fixed up series have been applied on top. Stable branches also have the new
upstream commit references updated. All pushed out.

Bundle against 4.18-rc1 attached. Sorry for the churn.

Thanks,

	tglx

[-- Attachment #2: Type: application/octet-stream, Size: 17521 bytes --]

[MODERATED] Re: L1TFv8

[Michal Hocko]

On Fri 15-06-18 14:48:23, speck for Josh Poimboeuf wrote:
> On Fri, Jun 15, 2018 at 04:50:38PM +0200, speck for Thomas Gleixner wrote:
> > On Wed, 13 Jun 2018, speck for Andi Kleen wrote:
> > 
> > > This is v8 of the native OS patchkit to mitigate the L1TF side channel.
> > > It does not cover KVM.
> > > 
> > > This version addresses the latest review feedbacks. The mitigation
> > > setup has been moved into check_bugs, and the memory size checking
> > > patch is now integrated into the standard setup. The swap mitigation
> > > has been split into two patches. Various other changes.
> > > For more details see the individual changelogs.
> > 
> > I've picked up the whole set with a few edits and applied it to the master
> > branch of the speck git repository and backported it to the linux-4.17.y,
> > linux-4.16.y and linux-4.14.y branches.
> > 
> > Git bundle against Linus tree is attached.
> 
> Has anyone tested these patches on a multi-node system with NUMA
> balancing enabled?  We've see some panics on RHEL because _PAGE_NUMA ==
> _PAGE_PROTNONE.  Not sure if this problem also exists upstream.
> 
> Andrea is looking into it (and I think he'll be joining this list
> shortly).

This didn't have any follow up but I guess you have figured out in the
meantime. So JFTR. We have encountered oopses on older 4.4 kernels
as well. And it turned out that fd7e315988b7 ("x86/mm:  Simplify
p[g4um]d_page() macros") is really needed.

HTH
-- 
Michal Hocko
SUSE Labs