The arm64 kernel doesn't yet support the full generic entry
infrastructure. That being said, KVM/arm64 doesn't properly handle
TIF_NOTIFY_RESUME and could pick this up by switching to the generic
guest entry infrasturture.
Patch 1 adds a missing vCPU stat to ARM64 to record the number of signal
exits to userspace.
Patch 2 unhitches entry-kvm from entry-generic, as ARM64 doesn't
currently support the generic infrastructure.
Patch 3 replaces the open-coded entry handling with the generic xfer
function.
This series was tested on an Ampere Mt. Jade reference system. The
series cleanly applies to kvm/queue (note that this is deliberate as the
generic kvm stats patches have not yet propagated to kvm-arm/queue) at
the following commit:
8ad5e63649ff ("KVM: Don't take mmu_lock for range invalidation unless necessary")
v1 -> v2:
- Address Jing's comment
- Carry Jing's r-b tag
v2 -> v3:
- Roll all exit conditions into kvm_vcpu_exit_request() (Marc)
- Avoid needlessly checking for work twice (Marc)
v1: http://lore.kernel.org/r/[email protected]
v2: http://lore.kernel.org/r/[email protected]
Oliver Upton (3):
KVM: arm64: Record number of signal exits as a vCPU stat
entry: KVM: Allow use of generic KVM entry w/o full generic support
KVM: arm64: Use generic KVM xfer to guest work function
arch/arm64/include/asm/kvm_host.h | 1 +
arch/arm64/kvm/Kconfig | 1 +
arch/arm64/kvm/arm.c | 71 +++++++++++++++++++------------
arch/arm64/kvm/guest.c | 1 +
include/linux/entry-kvm.h | 6 ++-
5 files changed, 52 insertions(+), 28 deletions(-)
--
2.32.0.554.ge1b32706d8-goog
Most other architectures that implement KVM record a statistic
indicating the number of times a vCPU has exited due to a pending
signal. Add support for that stat to arm64.
Reviewed-by: Jing Zhang <[email protected]>
Signed-off-by: Oliver Upton <[email protected]>
---
arch/arm64/include/asm/kvm_host.h | 1 +
arch/arm64/kvm/arm.c | 1 +
arch/arm64/kvm/guest.c | 1 +
3 files changed, 3 insertions(+)
diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
index 41911585ae0c..70e129f2b574 100644
--- a/arch/arm64/include/asm/kvm_host.h
+++ b/arch/arm64/include/asm/kvm_host.h
@@ -576,6 +576,7 @@ struct kvm_vcpu_stat {
u64 wfi_exit_stat;
u64 mmio_exit_user;
u64 mmio_exit_kernel;
+ u64 signal_exits;
u64 exits;
};
diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
index e9a2b8f27792..60d0a546d7fd 100644
--- a/arch/arm64/kvm/arm.c
+++ b/arch/arm64/kvm/arm.c
@@ -783,6 +783,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
if (signal_pending(current)) {
ret = -EINTR;
run->exit_reason = KVM_EXIT_INTR;
+ ++vcpu->stat.signal_exits;
}
/*
diff --git a/arch/arm64/kvm/guest.c b/arch/arm64/kvm/guest.c
index 1dfb83578277..853d1e8d2e73 100644
--- a/arch/arm64/kvm/guest.c
+++ b/arch/arm64/kvm/guest.c
@@ -50,6 +50,7 @@ const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
STATS_DESC_COUNTER(VCPU, wfi_exit_stat),
STATS_DESC_COUNTER(VCPU, mmio_exit_user),
STATS_DESC_COUNTER(VCPU, mmio_exit_kernel),
+ STATS_DESC_COUNTER(VCPU, signal_exits),
STATS_DESC_COUNTER(VCPU, exits)
};
static_assert(ARRAY_SIZE(kvm_vcpu_stats_desc) ==
--
2.32.0.554.ge1b32706d8-goog
Some architectures (e.g. arm64) have yet to adopt the generic entry
infrastructure. Despite that, it would be nice to use some common
plumbing for guest entry/exit handling. For example, KVM/arm64 currently
does not handle TIF_NOTIFY_PENDING correctly.
Allow use of only the generic KVM entry code by tightening up the
include list. No functional change intended.
Signed-off-by: Oliver Upton <[email protected]>
---
include/linux/entry-kvm.h | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/include/linux/entry-kvm.h b/include/linux/entry-kvm.h
index 136b8d97d8c0..0d7865a0731c 100644
--- a/include/linux/entry-kvm.h
+++ b/include/linux/entry-kvm.h
@@ -2,7 +2,11 @@
#ifndef __LINUX_ENTRYKVM_H
#define __LINUX_ENTRYKVM_H
-#include <linux/entry-common.h>
+#include <linux/static_call_types.h>
+#include <linux/tracehook.h>
+#include <linux/syscalls.h>
+#include <linux/seccomp.h>
+#include <linux/sched.h>
#include <linux/tick.h>
/* Transfer to guest mode work */
--
2.32.0.554.ge1b32706d8-goog
Clean up handling of checks for pending work by switching to the generic
infrastructure to do so.
We pick up handling for TIF_NOTIFY_RESUME from this switch, meaning that
task work will be correctly handled.
Signed-off-by: Oliver Upton <[email protected]>
---
arch/arm64/kvm/Kconfig | 1 +
arch/arm64/kvm/arm.c | 72 ++++++++++++++++++++++++++----------------
2 files changed, 45 insertions(+), 28 deletions(-)
diff --git a/arch/arm64/kvm/Kconfig b/arch/arm64/kvm/Kconfig
index a4eba0908bfa..8bc1fac5fa26 100644
--- a/arch/arm64/kvm/Kconfig
+++ b/arch/arm64/kvm/Kconfig
@@ -26,6 +26,7 @@ menuconfig KVM
select HAVE_KVM_ARCH_TLB_FLUSH_ALL
select KVM_MMIO
select KVM_GENERIC_DIRTYLOG_READ_PROTECT
+ select KVM_XFER_TO_GUEST_WORK
select SRCU
select KVM_VFIO
select HAVE_KVM_EVENTFD
diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
index 60d0a546d7fd..8245efc6e88f 100644
--- a/arch/arm64/kvm/arm.c
+++ b/arch/arm64/kvm/arm.c
@@ -6,6 +6,7 @@
#include <linux/bug.h>
#include <linux/cpu_pm.h>
+#include <linux/entry-kvm.h>
#include <linux/errno.h>
#include <linux/err.h>
#include <linux/kvm_host.h>
@@ -714,6 +715,45 @@ static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
static_branch_unlikely(&arm64_mismatched_32bit_el0);
}
+/**
+ * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
+ * @vcpu: The VCPU pointer
+ * @ret: Pointer to write optional return code
+ *
+ * Returns: true if the VCPU needs to return to a preemptible + interruptible
+ * and skip guest entry.
+ *
+ * This function disambiguates between two different types of exits: exits to a
+ * preemptible + interruptible kernel context and exits to userspace. For an
+ * exit to userspace, this function will write the return code to ret and return
+ * true. For an exit to preemptible + interruptible kernel context (i.e. check
+ * for pending work and re-enter), return true without writing to ret.
+ */
+static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
+{
+ struct kvm_run *run = vcpu->run;
+
+ /*
+ * If we're using a userspace irqchip, then check if we need
+ * to tell a userspace irqchip about timer or PMU level
+ * changes and if so, exit to userspace (the actual level
+ * state gets updated in kvm_timer_update_run and
+ * kvm_pmu_update_run below).
+ */
+ if (static_branch_unlikely(&userspace_irqchip_in_use)) {
+ if (kvm_timer_should_notify_user(vcpu) ||
+ kvm_pmu_should_notify_user(vcpu)) {
+ *ret = -EINTR;
+ run->exit_reason = KVM_EXIT_INTR;
+ return true;
+ }
+ }
+
+ return kvm_request_pending(vcpu) ||
+ need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
+ xfer_to_guest_mode_work_pending();
+}
+
/**
* kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
* @vcpu: The VCPU pointer
@@ -757,7 +797,9 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
/*
* Check conditions before entering the guest
*/
- cond_resched();
+ ret = xfer_to_guest_mode_handle_work(vcpu);
+ if (!ret)
+ ret = 1;
update_vmid(&vcpu->arch.hw_mmu->vmid);
@@ -776,31 +818,6 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
kvm_vgic_flush_hwstate(vcpu);
- /*
- * Exit if we have a signal pending so that we can deliver the
- * signal to user space.
- */
- if (signal_pending(current)) {
- ret = -EINTR;
- run->exit_reason = KVM_EXIT_INTR;
- ++vcpu->stat.signal_exits;
- }
-
- /*
- * If we're using a userspace irqchip, then check if we need
- * to tell a userspace irqchip about timer or PMU level
- * changes and if so, exit to userspace (the actual level
- * state gets updated in kvm_timer_update_run and
- * kvm_pmu_update_run below).
- */
- if (static_branch_unlikely(&userspace_irqchip_in_use)) {
- if (kvm_timer_should_notify_user(vcpu) ||
- kvm_pmu_should_notify_user(vcpu)) {
- ret = -EINTR;
- run->exit_reason = KVM_EXIT_INTR;
- }
- }
-
/*
* Ensure we set mode to IN_GUEST_MODE after we disable
* interrupts and before the final VCPU requests check.
@@ -809,8 +826,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
*/
smp_store_mb(vcpu->mode, IN_GUEST_MODE);
- if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) ||
- kvm_request_pending(vcpu)) {
+ if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
vcpu->mode = OUTSIDE_GUEST_MODE;
isb(); /* Ensure work in x_flush_hwstate is committed */
kvm_pmu_sync_hwstate(vcpu);
--
2.32.0.554.ge1b32706d8-goog
Friendly ping :)
--
Thanks,
Oliver
On Mon, Aug 2, 2021 at 12:28 PM Oliver Upton <[email protected]> wrote:
>
> The arm64 kernel doesn't yet support the full generic entry
> infrastructure. That being said, KVM/arm64 doesn't properly handle
> TIF_NOTIFY_RESUME and could pick this up by switching to the generic
> guest entry infrasturture.
>
> Patch 1 adds a missing vCPU stat to ARM64 to record the number of signal
> exits to userspace.
>
> Patch 2 unhitches entry-kvm from entry-generic, as ARM64 doesn't
> currently support the generic infrastructure.
>
> Patch 3 replaces the open-coded entry handling with the generic xfer
> function.
>
> This series was tested on an Ampere Mt. Jade reference system. The
> series cleanly applies to kvm/queue (note that this is deliberate as the
> generic kvm stats patches have not yet propagated to kvm-arm/queue) at
> the following commit:
>
> 8ad5e63649ff ("KVM: Don't take mmu_lock for range invalidation unless necessary")
>
> v1 -> v2:
> - Address Jing's comment
> - Carry Jing's r-b tag
>
> v2 -> v3:
> - Roll all exit conditions into kvm_vcpu_exit_request() (Marc)
> - Avoid needlessly checking for work twice (Marc)
>
> v1: http://lore.kernel.org/r/[email protected]
> v2: http://lore.kernel.org/r/[email protected]
>
> Oliver Upton (3):
> KVM: arm64: Record number of signal exits as a vCPU stat
> entry: KVM: Allow use of generic KVM entry w/o full generic support
> KVM: arm64: Use generic KVM xfer to guest work function
>
> arch/arm64/include/asm/kvm_host.h | 1 +
> arch/arm64/kvm/Kconfig | 1 +
> arch/arm64/kvm/arm.c | 71 +++++++++++++++++++------------
> arch/arm64/kvm/guest.c | 1 +
> include/linux/entry-kvm.h | 6 ++-
> 5 files changed, 52 insertions(+), 28 deletions(-)
>
> --
> 2.32.0.554.ge1b32706d8-goog
>
On Sun, 15 Aug 2021 08:50:09 +0100,
Oliver Upton <[email protected]> wrote:
>
> Friendly ping :)
Ah, thanks for reminding me of this. I'll have a look again early next
week.
M.
--
Without deviation from the norm, progress is not possible.
On Mon, 2 Aug 2021 19:28:06 +0000, Oliver Upton wrote:
> The arm64 kernel doesn't yet support the full generic entry
> infrastructure. That being said, KVM/arm64 doesn't properly handle
> TIF_NOTIFY_RESUME and could pick this up by switching to the generic
> guest entry infrasturture.
>
> Patch 1 adds a missing vCPU stat to ARM64 to record the number of signal
> exits to userspace.
>
> [...]
Applied to next, thanks!
[1/3] KVM: arm64: Record number of signal exits as a vCPU stat
commit: fe5161d2c39b8c2801f0e786631460c6e8a1cae4
[2/3] entry: KVM: Allow use of generic KVM entry w/o full generic support
commit: e1c6b9e1669e44fb7f9688e34e460b759e3b9187
[3/3] KVM: arm64: Use generic KVM xfer to guest work function
commit: 6caa5812e2d126a0aa8a17816c1ba6f0a0c2b309
Cheers,
M.
--
Without deviation from the norm, progress is not possible.