Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1542899 > unrolled thread

[PATCH v3 3/4] KVM: x86: make interrupt delivery fast and slow path behave the same

Started byRadim Krčmář <rkrcmar@redhat.com>
First post2016-12-15 18:10 +0100
Last post2016-12-19 16:00 +0100
Articles 5 — 3 participants

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  [PATCH v3 3/4] KVM: x86: make interrupt delivery fast and slow path behave the same Radim Krčmář <rkrcmar@redhat.com> - 2016-12-15 18:10 +0100
    Re: [PATCH v3 3/4] KVM: x86: make interrupt delivery fast and slow  path behave the same David Hildenbrand <david@redhat.com> - 2016-12-16 17:50 +0100
      Re: [PATCH v3 3/4] KVM: x86: make interrupt delivery fast and slow  path behave the same Paolo Bonzini <pbonzini@redhat.com> - 2016-12-16 18:00 +0100
        [PATCH v4 3/4] KVM: x86: make interrupt delivery fast and slow path  behave the same Radim Krčmář <rkrcmar@redhat.com> - 2016-12-16 18:10 +0100
          Re: [PATCH v4 3/4] KVM: x86: make interrupt delivery fast and slow  path behave the same David Hildenbrand <david@redhat.com> - 2016-12-19 16:00 +0100

#1542899 — [PATCH v3 3/4] KVM: x86: make interrupt delivery fast and slow path behave the same

FromRadim Krčmář <rkrcmar@redhat.com>
Date2016-12-15 18:10 +0100
Subject[PATCH v3 3/4] KVM: x86: make interrupt delivery fast and slow path behave the same
Message-ID<sOHHA-8tl-29@gated-at.bofh.it>
Slow path tried to prevent IPIs from x2APIC VCPUs from being delivered
to xAPIC VCPUs and vice-versa.  Make slow path behave like fast path,
which never distinguished that.

Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
---
 arch/x86/kvm/lapic.c | 14 ++++++--------
 1 file changed, 6 insertions(+), 8 deletions(-)

diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index 81cc93580c40..519b72bbf947 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -606,7 +606,7 @@ static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
 	if (apic_x2apic_mode(apic))
 		return mda == kvm_x2apic_id(apic);
 
-	return mda == SET_APIC_DEST_FIELD(kvm_xapic_id(apic));
+	return mda == kvm_xapic_id(apic);
 }
 
 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
@@ -623,7 +623,6 @@ static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
 		       && (logical_id & mda & 0xffff) != 0;
 
 	logical_id = GET_APIC_LOGICAL_ID(logical_id);
-	mda = GET_APIC_DEST_FIELD(mda);
 
 	switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
 	case APIC_DFR_FLAT:
@@ -640,9 +639,9 @@ static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
 
 /* The KVM local APIC implementation has two quirks:
  *
- *  - the xAPIC MDA stores the destination at bits 24-31, while this
- *    is not true of struct kvm_lapic_irq's dest_id field.  This is
- *    just a quirk in the API and is not problematic.
+ *  - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
+ *    in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
+ *    KVM doesn't do that aliasing.
  *
  *  - in-kernel IOAPIC messages have to be delivered directly to
  *    x2APIC, because the kernel does not support interrupt remapping.
@@ -658,13 +657,12 @@ static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
 		struct kvm_lapic *source, struct kvm_lapic *target)
 {
 	bool ipi = source != NULL;
-	bool x2apic_mda = apic_x2apic_mode(ipi ? source : target);
 
 	if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
-	    !ipi && dest_id == APIC_BROADCAST && x2apic_mda)
+	    !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
 		return X2APIC_BROADCAST;
 
-	return x2apic_mda ? dest_id : SET_APIC_DEST_FIELD(dest_id);
+	return dest_id;
 }
 
 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
-- 
2.11.0

[toc] | [next] | [standalone]

#1543587 — Re: [PATCH v3 3/4] KVM: x86: make interrupt delivery fast and slow path behave the same

FromDavid Hildenbrand <david@redhat.com>
Date2016-12-16 17:50 +0100
SubjectRe: [PATCH v3 3/4] KVM: x86: make interrupt delivery fast and slow path behave the same
Message-ID<sP3RM-69y-19@gated-at.bofh.it>
In reply to#1542899
Am 15.12.2016 um 18:06 schrieb Radim Krčmář:
> Slow path tried to prevent IPIs from x2APIC VCPUs from being delivered
> to xAPIC VCPUs and vice-versa.  Make slow path behave like fast path,
> which never distinguished that.
>
> Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>

Just to verify:

The GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST is still correct in 
kvm_apic_broadcast() ?


Apart from that thing confusing me, this looks good to me :)

-- 

David

[toc] | [prev] | [next] | [standalone]

#1543591 — Re: [PATCH v3 3/4] KVM: x86: make interrupt delivery fast and slow path behave the same

FromPaolo Bonzini <pbonzini@redhat.com>
Date2016-12-16 18:00 +0100
SubjectRe: [PATCH v3 3/4] KVM: x86: make interrupt delivery fast and slow path behave the same
Message-ID<sP41r-6cL-15@gated-at.bofh.it>
In reply to#1543587

On 16/12/2016 17:41, David Hildenbrand wrote:
> 
> Just to verify:
> 
> The GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST is still correct in
> kvm_apic_broadcast() ?

No, it seems wrong.

Paolo

[toc] | [prev] | [next] | [standalone]

#1543601 — [PATCH v4 3/4] KVM: x86: make interrupt delivery fast and slow path behave the same

FromRadim Krčmář <rkrcmar@redhat.com>
Date2016-12-16 18:10 +0100
Subject[PATCH v4 3/4] KVM: x86: make interrupt delivery fast and slow path behave the same
Message-ID<sP4b7-6vt-19@gated-at.bofh.it>
In reply to#1543591
2016-12-16 17:53+0100, Paolo Bonzini:
> On 16/12/2016 17:41, David Hildenbrand wrote:
>> 
>> Just to verify:
>> 
>> The GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST is still correct in
>> kvm_apic_broadcast() ?
> 
> No, it seems wrong.

Definitely, sending v4 of this patch.
I'll also add broadcast address check into kvm-unit-tests ...
---8<---
Slow path tried to prevent IPIs from x2APIC VCPUs from being delivered
to xAPIC VCPUs and vice-versa.  Make slow path behave like fast path,
which never distinguished that.

Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
---
 v4: fix xAPIC broadcast address detection [David]
---
 arch/x86/kvm/lapic.c | 20 ++++++++------------
 1 file changed, 8 insertions(+), 12 deletions(-)

diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index 81cc93580c40..dba19cb82c3e 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -592,10 +592,8 @@ static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
 
 static bool kvm_apic_broadcast(struct kvm_lapic *apic, u32 mda)
 {
-	if (apic_x2apic_mode(apic))
-		return mda == X2APIC_BROADCAST;
-
-	return GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST;
+	return mda == (apic_x2apic_mode(apic) ?
+			X2APIC_BROADCAST : APIC_BROADCAST);
 }
 
 static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
@@ -606,7 +604,7 @@ static bool kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 mda)
 	if (apic_x2apic_mode(apic))
 		return mda == kvm_x2apic_id(apic);
 
-	return mda == SET_APIC_DEST_FIELD(kvm_xapic_id(apic));
+	return mda == kvm_xapic_id(apic);
 }
 
 static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
@@ -623,7 +621,6 @@ static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
 		       && (logical_id & mda & 0xffff) != 0;
 
 	logical_id = GET_APIC_LOGICAL_ID(logical_id);
-	mda = GET_APIC_DEST_FIELD(mda);
 
 	switch (kvm_lapic_get_reg(apic, APIC_DFR)) {
 	case APIC_DFR_FLAT:
@@ -640,9 +637,9 @@ static bool kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
 
 /* The KVM local APIC implementation has two quirks:
  *
- *  - the xAPIC MDA stores the destination at bits 24-31, while this
- *    is not true of struct kvm_lapic_irq's dest_id field.  This is
- *    just a quirk in the API and is not problematic.
+ *  - Real hardware delivers interrupts destined to x2APIC ID > 0xff to LAPICs
+ *    in xAPIC mode if the "destination & 0xff" matches its xAPIC ID.
+ *    KVM doesn't do that aliasing.
  *
  *  - in-kernel IOAPIC messages have to be delivered directly to
  *    x2APIC, because the kernel does not support interrupt remapping.
@@ -658,13 +655,12 @@ static u32 kvm_apic_mda(struct kvm_vcpu *vcpu, unsigned int dest_id,
 		struct kvm_lapic *source, struct kvm_lapic *target)
 {
 	bool ipi = source != NULL;
-	bool x2apic_mda = apic_x2apic_mode(ipi ? source : target);
 
 	if (!vcpu->kvm->arch.x2apic_broadcast_quirk_disabled &&
-	    !ipi && dest_id == APIC_BROADCAST && x2apic_mda)
+	    !ipi && dest_id == APIC_BROADCAST && apic_x2apic_mode(target))
 		return X2APIC_BROADCAST;
 
-	return x2apic_mda ? dest_id : SET_APIC_DEST_FIELD(dest_id);
+	return dest_id;
 }
 
 bool kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
-- 
2.11.0

[toc] | [prev] | [next] | [standalone]

#1544595 — Re: [PATCH v4 3/4] KVM: x86: make interrupt delivery fast and slow path behave the same

FromDavid Hildenbrand <david@redhat.com>
Date2016-12-19 16:00 +0100
SubjectRe: [PATCH v4 3/4] KVM: x86: make interrupt delivery fast and slow path behave the same
Message-ID<sQ7zY-ML-25@gated-at.bofh.it>
In reply to#1543601
Am 16.12.2016 um 18:07 schrieb Radim Krčmář:
> 2016-12-16 17:53+0100, Paolo Bonzini:
>> On 16/12/2016 17:41, David Hildenbrand wrote:
>>>
>>> Just to verify:
>>>
>>> The GET_APIC_DEST_FIELD(mda) == APIC_BROADCAST is still correct in
>>> kvm_apic_broadcast() ?
>>
>> No, it seems wrong.
>
> Definitely, sending v4 of this patch.
> I'll also add broadcast address check into kvm-unit-tests ...
> ---8<---
> Slow path tried to prevent IPIs from x2APIC VCPUs from being delivered
> to xAPIC VCPUs and vice-versa.  Make slow path behave like fast path,
> which never distinguished that.
>
> Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
> ---
>  v4: fix xAPIC broadcast address detection [David]
> ---
>  arch/x86/kvm/lapic.c | 20 ++++++++------------
>  1 file changed, 8 insertions(+), 12 deletions(-)
>

Looks good to me.

Reviewed-by: David Hildenbrand <david@redhat.com>

-- 

David

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web