Skip to contentSkip to conversation

KVM patches applied in weird order in -stable

DISCUSSION Matt Fleming 3 participants 6 messages Sep 13, 2016 – Sep 14, 2016 Stable branch

lore
Share & tools
lore
#0 lore raw

Folks,

While hunting down a performance issue involving KVM I was surprised to see "native_set_debugreg()" as the first entry in perf top.

Digging deeper, it looks as though the following patches were applied in the wrong order in -stable. This is the order as they appear in Linus' tree,

[0] commit 4e422bdd2f84 ("KVM: x86: fix missed hardware breakpoints")
[1] commit 172b2386ed16 ("KVM: x86: fix missed hardware breakpoints")
[2] commit 70e4da7a8ff6 ("KVM: x86: fix root cause for missed hardware breakpoints")

but this is the order for linux-4.4.y

[1] commit fc90441e728a ("KVM: x86: fix missed hardware breakpoints")
[2] commit 25e8618619a5 ("KVM: x86: fix root cause for missed hardware breakpoints")
[0] commit 0f6e5e26e68f ("KVM: x86: fix missed hardware breakpoints")

The upshot is that KVM_DEBUGREG_RELOAD is always set when returning from kvm_arch_vcpu_load() in stable, but not in Linus' tree.

maintainer
#1 lore raw

[adding stable@ as this is a stable issue, not a 'normal' issue]

Quoted 13 lines · Matt Fleming

Folks,

While hunting down a performance issue involving KVM I was surprised to see "native_set_debugreg()" as the first entry in perf top.

Digging deeper, it looks as though the following patches were applied in the wrong order in -stable. This is the order as they appear in Linus' tree,

[0] commit 4e422bdd2f84 ("KVM: x86: fix missed hardware breakpoints")
[1] commit 172b2386ed16 ("KVM: x86: fix missed hardware breakpoints")
[2] commit 70e4da7a8ff6 ("KVM: x86: fix root cause for missed hardware breakpoints")

Quoted 8 lines

but this is the order for linux-4.4.y

[1] commit fc90441e728a ("KVM: x86: fix missed hardware breakpoints")
[2] commit 25e8618619a5 ("KVM: x86: fix root cause for missed hardware breakpoints")
[0] commit 0f6e5e26e68f ("KVM: x86: fix missed hardware breakpoints")

The upshot is that KVM_DEBUGREG_RELOAD is always set when returning from kvm_arch_vcpu_load() in stable, but not in Linus' tree.

How would applying these in a different order cause breakage?

And if this is a problem, can you please send me a patch to fix it up?

thanks,

greg k-h

#2 lore raw
Quoted 18 lines · Greg KH

[adding stable@ as this is a stable issue, not a 'normal' issue]

Quoted 15 lines · Matt Fleming

Folks,

While hunting down a performance issue involving KVM I was surprised to see "native_set_debugreg()" as the first entry in perf top.

Digging deeper, it looks as though the following patches were applied in the wrong order in -stable. This is the order as they appear in Linus' tree,

[0] commit 4e422bdd2f84 ("KVM: x86: fix missed hardware breakpoints")
[1] commit 172b2386ed16 ("KVM: x86: fix missed hardware breakpoints")
[2] commit 70e4da7a8ff6 ("KVM: x86: fix root cause for missed hardware breakpoints")

but this is the order for linux-4.4.y

Quoted 8 lines
Quoted 6 lines

[1] commit fc90441e728a ("KVM: x86: fix missed hardware breakpoints")
[2] commit 25e8618619a5 ("KVM: x86: fix root cause for missed hardware breakpoints")
[0] commit 0f6e5e26e68f ("KVM: x86: fix missed hardware breakpoints")

The upshot is that KVM_DEBUGREG_RELOAD is always set when returning from kvm_arch_vcpu_load() in stable, but not in Linus' tree.

How would applying these in a different order cause breakage?

[2] is reverting [0]+[1]. Stable is not due to the different order.

Quoted 1 line

And if this is a problem, can you please send me a patch to fix it up?

Yup, on the way.

Paolo

maintainer
#3 lore raw
Quoted 25 lines · Paolo Bonzini
Quoted 22 lines · Greg KH

[adding stable@ as this is a stable issue, not a 'normal' issue]

Quoted 19 lines · Matt Fleming

Folks,

While hunting down a performance issue involving KVM I was surprised to see "native_set_debugreg()" as the first entry in perf top.

Digging deeper, it looks as though the following patches were applied in the wrong order in -stable. This is the order as they appear in Linus' tree,

[0] commit 4e422bdd2f84 ("KVM: x86: fix missed hardware breakpoints")
[1] commit 172b2386ed16 ("KVM: x86: fix missed hardware breakpoints")
[2] commit 70e4da7a8ff6 ("KVM: x86: fix root cause for missed hardware breakpoints")

but this is the order for linux-4.4.y

[1] commit fc90441e728a ("KVM: x86: fix missed hardware breakpoints")
[2] commit 25e8618619a5 ("KVM: x86: fix root cause for missed hardware breakpoints")
[0] commit 0f6e5e26e68f ("KVM: x86: fix missed hardware breakpoints")

Quoted 6 lines
Quoted 4 lines
Quoted 2 lines

The upshot is that KVM_DEBUGREG_RELOAD is always set when returning from kvm_arch_vcpu_load() in stable, but not in Linus' tree.

How would applying these in a different order cause breakage?

[2] is reverting [0]+[1]. Stable is not due to the different order.

Really? Are you sure that [0] and [1] isn't just the same commit? It looks like that to me.

Quoted 3 lines
Quoted 1 line

And if this is a problem, can you please send me a patch to fix it up?

Yup, on the way.

thanks,

greg k-h

#4 lore raw
Quoted 10 lines · Greg KH
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines

[0] commit 4e422bdd2f84 ("KVM: x86: fix missed hardware breakpoints")
[1] commit 172b2386ed16 ("KVM: x86: fix missed hardware breakpoints")
[2] commit 70e4da7a8ff6 ("KVM: x86: fix root cause for missed hardware breakpoints")

but this is the order for linux-4.4.y

[1] commit fc90441e728a ("KVM: x86: fix missed hardware breakpoints")
[2] commit 25e8618619a5 ("KVM: x86: fix root cause for missed hardware breakpoints")
[0] commit 0f6e5e26e68f ("KVM: x86: fix missed hardware breakpoints")

Quoted 8 lines
Quoted 6 lines
Quoted 6 lines
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines

The upshot is that KVM_DEBUGREG_RELOAD is always set when returning from kvm_arch_vcpu_load() in stable, but not in Linus' tree.

How would applying these in a different order cause breakage?

[2] is reverting [0]+[1]. Stable is not due to the different order.

Really? Are you sure that [0] and [1] isn't just the same commit? It looks like that to me.

It is; "git" automatically resolved the conflicts when merging [1], and then [2] reverted the change. In stable, changing the order created a different conflict resolution.

Paolo

maintainer
#5 lore raw
Quoted 18 lines · Paolo Bonzini
Quoted 15 lines · Greg KH
Quoted 15 lines
Quoted 15 lines
Quoted 14 lines
Quoted 14 lines
Quoted 4 lines
Quoted 4 lines

[0] commit 4e422bdd2f84 ("KVM: x86: fix missed hardware breakpoints")
[1] commit 172b2386ed16 ("KVM: x86: fix missed hardware breakpoints")
[2] commit 70e4da7a8ff6 ("KVM: x86: fix root cause for missed hardware breakpoints")

Quoted 8 lines
Quoted 8 lines

but this is the order for linux-4.4.y

[1] commit fc90441e728a ("KVM: x86: fix missed hardware breakpoints")
[2] commit 25e8618619a5 ("KVM: x86: fix root cause for missed hardware breakpoints")
[0] commit 0f6e5e26e68f ("KVM: x86: fix missed hardware breakpoints")

The upshot is that KVM_DEBUGREG_RELOAD is always set when returning from kvm_arch_vcpu_load() in stable, but not in Linus' tree.

How would applying these in a different order cause breakage?

Quoted 7 lines
Quoted 3 lines
Quoted 1 line
Quoted 1 line

[2] is reverting [0]+[1]. Stable is not due to the different order.

Really? Are you sure that [0] and [1] isn't just the same commit? It looks like that to me.

It is; "git" automatically resolved the conflicts when merging [1], and then [2] reverted the change. In stable, changing the order created a different conflict resolution.

Yes, given that I turn them into individual patches, the order I used was really the only one that would work, and is how this happened :)

thanks,

greg k-h