[5.15.y,v2,0/2] fix softlockups in stage2_apply_range()

Message

Krister Johansen March 6, 2024, 12:49 a.m. UTC

Hi Stable Team,
In 5.15, unmapping large kvm vms on arm64 can generate softlockups.  My team has
been hitting this when tearing down VMs > 100Gb in size.

Oliver fixed this with the attached patches.  They've been in mainline since
6.1.

I tested on 5.15.150 with these patches applied. When they're present,
both the dirty_log_perf_test detailed in the second patch, and
kvm_page_table_test no longer generate softlockups when unmapping VMs
with large memory configurations.

Would you please consider these patches for inclusion in an upcoming 5.15
release?

Change in v2:  I ran format-patch without the --from option which incorrectly
generated the first series without leaving Oliver in place as the author.  The
v2 should retain the correct authorship.  Apologies for the mistake.

-K

Oliver Upton (2):
  KVM: arm64: Work out supported block level at compile time
  KVM: arm64: Limit stage2_apply_range() batch size to largest block

 arch/arm64/include/asm/kvm_pgtable.h    | 18 +++++++++++++-----
 arch/arm64/include/asm/stage2_pgtable.h | 20 --------------------
 arch/arm64/kvm/mmu.c                    |  9 ++++++++-
 3 files changed, 21 insertions(+), 26 deletions(-)

Comments

Marc Zyngier March 6, 2024, 10:08 a.m. UTC | #1

On Wed, 06 Mar 2024 00:49:34 +0000,
Krister Johansen <[e-mail]> wrote:
> 
> Hi Stable Team,
> In 5.15, unmapping large kvm vms on arm64 can generate softlockups.  My team has
> been hitting this when tearing down VMs > 100Gb in size.
> 
> Oliver fixed this with the attached patches.  They've been in mainline since
> 6.1.
> 
> I tested on 5.15.150 with these patches applied. When they're present,
> both the dirty_log_perf_test detailed in the second patch, and
> kvm_page_table_test no longer generate softlockups when unmapping VMs
> with large memory configurations.
> 
> Would you please consider these patches for inclusion in an upcoming 5.15
> release?
> 
> Change in v2:  I ran format-patch without the --from option which incorrectly
> generated the first series without leaving Oliver in place as the author.  The
> v2 should retain the correct authorship.  Apologies for the mistake.

Thanks for this.

FWIW,

Acked-by: Marc Zyngier <[e-mail]>

	M.

Greg Kroah-Hartman March 29, 2024, 12:21 p.m. UTC | #2

On Wed, Mar 06, 2024 at 10:08:40AM +0000, Marc Zyngier wrote:
> On Wed, 06 Mar 2024 00:49:34 +0000,
> Krister Johansen <[e-mail]> wrote:
> > 
> > Hi Stable Team,
> > In 5.15, unmapping large kvm vms on arm64 can generate softlockups.  My team has
> > been hitting this when tearing down VMs > 100Gb in size.
> > 
> > Oliver fixed this with the attached patches.  They've been in mainline since
> > 6.1.
> > 
> > I tested on 5.15.150 with these patches applied. When they're present,
> > both the dirty_log_perf_test detailed in the second patch, and
> > kvm_page_table_test no longer generate softlockups when unmapping VMs
> > with large memory configurations.
> > 
> > Would you please consider these patches for inclusion in an upcoming 5.15
> > release?
> > 
> > Change in v2:  I ran format-patch without the --from option which incorrectly
> > generated the first series without leaving Oliver in place as the author.  The
> > v2 should retain the correct authorship.  Apologies for the mistake.
> 
> Thanks for this.
> 
> FWIW,
> 
> Acked-by: Marc Zyngier <[e-mail]>

Now queued up,t hanks.

greg k-h