KVM: arm64: Replace KVM_ARM_PMU with HW_PERF_EVENTS
Marc Zyngier
maintainerPATCH 1/1 KVM: arm64: Replace KVM_ARM_PMU with HW_PERF_EVENTS
KVM_ARM_PMU only existed for the benefit of 32bit ARM hosts, and makes no sense now that we are 64bit only. Get rid of it.
--- arch/arm64/kvm/Kconfig | 8 -------- arch/arm64/kvm/Makefile | 2 +- include/kvm/arm_pmu.h | 2 +- 3 files changed, 2 insertions(+), 10 deletions(-)
arch/arm64/kvm/Kconfig +0 -8
@@ -49,14 +49,6 @@ if KVM source "virt/kvm/Kconfig" -config KVM_ARM_PMU - bool "Virtual Performance Monitoring Unit (PMU) support" - depends on HW_PERF_EVENTS - default y - help - Adds support for a virtual Performance Monitoring Unit (PMU) in - virtual machines. - endif # KVM endif # VIRTUALIZATION
arch/arm64/kvm/Makefile +1 -1
@@ -24,4 +24,4 @@ kvm-y := $(KVM)/kvm_main.o $(KVM)/coalesced_mmio.o $(KVM)/eventfd.o \ vgic/vgic-mmio-v3.o vgic/vgic-kvm-device.o \ vgic/vgic-its.o vgic/vgic-debug.o -kvm-$(CONFIG_KVM_ARM_PMU) += pmu-emul.o +kvm-$(CONFIG_HW_PERF_EVENTS) += pmu-emul.o
include/kvm/arm_pmu.h +1 -1
@@ -13,7 +13,7 @@#define ARMV8_PMU_CYCLE_IDX (ARMV8_PMU_MAX_COUNTERS - 1)#define ARMV8_PMU_MAX_COUNTER_PAIRS ((ARMV8_PMU_MAX_COUNTERS + 1) >> 1) -#ifdef CONFIG_KVM_ARM_PMU +#ifdef CONFIG_HW_PERF_EVENTS struct kvm_pmc { u8 idx; /* index into the pmu->pmc array */
Signature · 6 lines
2.29.2 _______________________________________________ kvmarm mailing list [email protected] https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
Hi Marc,
Quoted 44 lines · Marc Zyngier
KVM_ARM_PMUonly existed for the benefit of 32bit ARM hosts, and makes no sense now that we are 64bit only. Get rid of it.Signed-off-by Marc Zyngier <[email protected]>--- arch/arm64/kvm/Kconfig | 8 -------- arch/arm64/kvm/Makefile | 2 +- include/kvm/arm_pmu.h | 2 +- 3 files changed, 2 insertions(+), 10 deletions(-)arch/arm64/kvm/Kconfig +0
-8@@ -49,14 +49,6 @@ if KVM source "virt/kvm/Kconfig" -config KVM_ARM_PMU - bool "Virtual Performance Monitoring Unit (PMU) support" - depends on HW_PERF_EVENTS - default y - help - Adds support for a virtual Performance Monitoring Unit (PMU) in - virtual machines. - endif # KVM endif # VIRTUALIZATIONarch/arm64/kvm/Makefile +1
-1@@ -24,4 +24,4 @@ kvm-y := $(KVM)/kvm_main.o $(KVM)/coalesced_mmio.o $(KVM)/eventfd.o \ vgic/vgic-mmio-v3.o vgic/vgic-kvm-device.o \ vgic/vgic-its.o vgic/vgic-debug.o -kvm-$(CONFIG_KVM_ARM_PMU) += pmu-emul.o +kvm-$(CONFIG_HW_PERF_EVENTS) += pmu-emul.oinclude/kvm/arm_pmu.h +0
-0
include/kvm/arm_pmu.h line 19
Quoted 8 lines
include/kvm/arm_pmu.h +1
-1@@ -13,7 +13,7 @@#define ARMV8_PMU_CYCLE_IDX (ARMV8_PMU_MAX_COUNTERS - 1)#define ARMV8_PMU_MAX_COUNTER_PAIRS ((ARMV8_PMU_MAX_COUNTERS + 1) >> 1) -#ifdef CONFIG_KVM_ARM_PMU +#ifdef CONFIG_HW_PERF_EVENTS struct kvm_pmc { u8 idx; /* index into the pmu->pmc array */
I grep'ed for KVM_ARM_PMU in the Linux sources, this patch takes care of all its occurrences.
A few things popped into my mind when I saw the patch.
1. Replacing KVM_ARM_PMU with CONFIG_HW_PERF_EVENTS means it's not possible anymore for the host to have perf support while KVM does not support emulating a PMU. In this scenario, functions which would have been empty functions if KVM_ARM_PMU was still around (I only found kvm_pmu_flush_hwstate() and kvm_pmu_sync_hwstate() on the KVM_RUN path) will now be called and return early after kvm_vcpu_has_pmu() returns 0. The overhead looks negligible to me, and I don't think this configuration was common (especially since the default was y).
2. I did a grep for the files that include arm_pmu.h, and all the files were in arch/arm64. I suppose arm_pmu.h exists in include/kvm instead of arch/arm64/include/asm because it was shared with KVM/arm when it was still around, right? Or is there another reason for that?
[1] https://www.spinics.net/lists/kvm-arm/msg44184.html
Thanks,
Alex
_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
Marc Zyngier
maintainerHi Alex,
Quoted 63 lines · Alexandru Elisei
Hi Marc,
Quoted 53 lines · Marc Zyngier
KVM_ARM_PMUonly existed for the benefit of 32bit ARM hosts, and makes no sense now that we are 64bit only. Get rid of it.Signed-off-by Marc Zyngier <[email protected]>--- arch/arm64/kvm/Kconfig | 8 -------- arch/arm64/kvm/Makefile | 2 +- include/kvm/arm_pmu.h | 2 +- 3 files changed, 2 insertions(+), 10 deletions(-)arch/arm64/kvm/Kconfig +0
-8@@ -49,14 +49,6 @@ if KVM source "virt/kvm/Kconfig" -config KVM_ARM_PMU - bool "Virtual Performance Monitoring Unit (PMU) support" - depends on HW_PERF_EVENTS - default y - help - Adds support for a virtual Performance Monitoring Unit (PMU) in - virtual machines. - endif # KVM endif # VIRTUALIZATIONarch/arm64/kvm/Makefile +1
-1@@ -24,4 +24,4 @@ kvm-y := $(KVM)/kvm_main.o $(KVM)/coalesced_mmio.o $(KVM)/eventfd.o \ vgic/vgic-mmio-v3.o vgic/vgic-kvm-device.o \ vgic/vgic-its.o vgic/vgic-debug.o -kvm-$(CONFIG_KVM_ARM_PMU) += pmu-emul.o +kvm-$(CONFIG_HW_PERF_EVENTS) += pmu-emul.oinclude/kvm/arm_pmu.h +1
-1@@ -13,7 +13,7 @@#define ARMV8_PMU_CYCLE_IDX (ARMV8_PMU_MAX_COUNTERS - 1)#define ARMV8_PMU_MAX_COUNTER_PAIRS ((ARMV8_PMU_MAX_COUNTERS + 1) >> 1) -#ifdef CONFIG_KVM_ARM_PMU +#ifdef CONFIG_HW_PERF_EVENTS struct kvm_pmc {Quoted 1 line
u8 idx; /* index into the
pmu->pmcarray */I grep'ed for
KVM_ARM_PMUin the Linux sources, this patch takes care of all its occurrences.A few things popped into my mind when I saw the patch.
Quoted 7 lines
1. Replacing
KVM_ARM_PMUwithCONFIG_HW_PERF_EVENTSmeans it's not possible anymore for the host to have perf support while KVM does not support emulating a PMU. In this scenario, functions which would have been empty functions ifKVM_ARM_PMUwas still around (I only foundkvm_pmu_flush_hwstate()andkvm_pmu_sync_hwstate()on theKVM_RUNpath) will now be called and return early afterkvm_vcpu_has_pmu()returns 0. The overhead looks negligible to me, and I don't think this configuration was common (especially since the default was y).
I don't think this is either common nor useful. If the kernel supports the PMU, then finding a PMU enables all the uses of the PMU, including for guests. And userspace is still in control of what it exposes to the guest. Yes, it's a tiny more overhead (one extra load) on the exit/entry path. Should we care? I don't think so.
Quoted 4 lines
2. I did a grep for the files that include
arm_pmu.h, and all the files were inarch/arm64. I supposearm_pmu.hexists ininclude/kvminstead ofarch/arm64/include/asmbecause it was shared with KVM/arm when it was still around, right? Or is there another reason for that?
No, that's basically the only reason. That was the easy landing spot for anything shared (including things like GIC, timers and co). I'm not sure it is worth the move, TBH...
Quoted 1 line
Yuo, and that's the reason I posted this patch. I have a couple more that I'll post by the end of the day.
Thanks,
M.
Signature · 5 lines
Jazz is not dead. It just smells funny... _______________________________________________ kvmarm mailing list [email protected] https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
Alexandru Elisei
Reviewed-byHi Marc,
Quoted 73 lines · Marc Zyngier
Hi Alex,
Quoted 62 lines · Alexandru Elisei
Hi Marc,
Quoted 52 lines · Marc Zyngier
KVM_ARM_PMUonly existed for the benefit of 32bit ARM hosts, and makes no sense now that we are 64bit only. Get rid of it.Signed-off-by Marc Zyngier <[email protected]>---
arch/arm64/kvm/Kconfig | 8 --------
arch/arm64/kvm/Makefile | 2 +-
include/kvm/arm_pmu.h | 2 +-3 files changed, 2 insertions(+), 10 deletions(-)arch/arm64/kvm/Kconfig +0
-0@@ -49,14 +49,6 @@ if KVMsource "
virt/kvm/Kconfig"arch/arm64/kvm/Kconfig +0
-8-config KVM_ARM_PMU - bool "Virtual Performance Monitoring Unit (PMU) support" - depends on HW_PERF_EVENTS - default y - help - Adds support for a virtual Performance Monitoring Unit (PMU) in - virtual machines. -endif # KVM
endif # VIRTUALIZATION
arch/arm64/kvm/Makefile +0
-0@@ -24,4 +24,4 @@ kvm-y := $(KVM)/kvm_main.o $(KVM)/coalesced_mmio.o $(KVM)/eventfd.o \ vgic/vgic-mmio-v3.o vgic/vgic-kvm-device.o \
vgic/vgic-its.ovgic/vgic-debug.oarch/arm64/kvm/Makefile +1
-1-kvm-$(CONFIG_KVM_ARM_PMU) += pmu-emul.o +kvm-$(CONFIG_HW_PERF_EVENTS) += pmu-emul.oinclude/kvm/arm_pmu.h +1
-1@@ -13,7 +13,7 @@ #define ARMV8_PMU_CYCLE_IDX (ARMV8_PMU_MAX_COUNTERS - 1) #define ARMV8_PMU_MAX_COUNTER_PAIRS ((ARMV8_PMU_MAX_COUNTERS + 1) >> 1) -#ifdef CONFIG_KVM_ARM_PMU +#ifdef CONFIG_HW_PERF_EVENTS struct kvm_pmc {Quoted 1 line
u8 idx; /* index into the
pmu->pmcarray */I grep'ed for
KVM_ARM_PMUin the Linux sources, this patch takes care of all its occurrences.A few things popped into my mind when I saw the patch.
Quoted 7 lines
1. Replacing
KVM_ARM_PMUwithCONFIG_HW_PERF_EVENTSmeans it's not possible anymore for the host to have perf support while KVM does not support emulating a PMU. In this scenario, functions which would have been empty functions ifKVM_ARM_PMUwas still around (I only foundkvm_pmu_flush_hwstate()andkvm_pmu_sync_hwstate()on theKVM_RUNpath) will now be called and return early afterkvm_vcpu_has_pmu()returns 0. The overhead looks negligible to me, and I don't think this configuration was common (especially since the default was y).
Quoted 5 lines
I don't think this is either common nor useful. If the kernel supports the PMU, then finding a PMU enables all the uses of the PMU, including for guests. And userspace is still in control of what it exposes to the guest. Yes, it's a tiny more overhead (one extra load) on the exit/entry path. Should we care? I don't think so.
Yes, I agree:
Quoted 9 lines
Quoted 5 lines
2. I did a grep for the files that include
arm_pmu.h, and all the files were inarch/arm64. I supposearm_pmu.hexists ininclude/kvminstead ofarch/arm64/include/asmbecause it was shared with KVM/arm when it was still around, right? Or is there another reason for that?No, that's basically the only reason. That was the easy landing spot for anything shared (including things like GIC, timers and co). I'm not sure it is worth the move, TBH...
I feel the same way, I was asking more in the context of new code. I admit to also having an ulterior motive, as one of the patches I picked up for the SPE series added a header file in include/kvm and it looked a bit out of place.
Thanks,
Alex
Quoted 9 lines
Quoted 2 lines
Yuo, and that's the reason I posted this patch. I have a couple more that I'll post by the end of the day.
Thanks,
M.
_______________________________________________
kvmarm mailing list
[email protected]
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm