On 2020-01-09 05:45, Jianyong Wu wrote:
Hi Marc,
-----Original Message-----
From: Marc Zyngier <maz@xxxxxxxxxx>
Sent: Tuesday, January 7, 2020 5:16 PM
To: Jianyong Wu <Jianyong.Wu@xxxxxxx>
Cc: netdev@xxxxxxxxxxxxxxx; yangbo.lu@xxxxxxx; john.stultz@xxxxxxxxxx;
tglx@xxxxxxxxxxxxx; pbonzini@xxxxxxxxxx;
sean.j.christopherson@xxxxxxxxx;
richardcochran@xxxxxxxxx; Mark Rutland <Mark.Rutland@xxxxxxx>;
will@xxxxxxxxxx; Suzuki Poulose <Suzuki.Poulose@xxxxxxx>; Steven Price
<Steven.Price@xxxxxxx>; linux-kernel@xxxxxxxxxxxxxxx; linux-arm-
kernel@xxxxxxxxxxxxxxxxxxx; kvmarm@xxxxxxxxxxxxxxxxxxxxx;
kvm@xxxxxxxxxxxxxxx; Steve Capper <Steve.Capper@xxxxxxx>; Kaly Xin
<Kaly.Xin@xxxxxxx>; Justin He <Justin.He@xxxxxxx>; nd <nd@xxxxxxx>
Subject: Re: [RFC PATCH v9 6/8] psci: Add hvc call service for
ptp_kvm.
On 2019-12-10 03:40, Jianyong Wu wrote:
> ptp_kvm modules will call hvc to get this service.
> The service offers real time and counter cycle of host for guest.
>
> Signed-off-by: Jianyong Wu <jianyong.wu@xxxxxxx>
> ---
> include/linux/arm-smccc.h | 12 ++++++++++++
> virt/kvm/arm/psci.c | 22 ++++++++++++++++++++++
> 2 files changed, 34 insertions(+)
>
> diff --git a/include/linux/arm-smccc.h b/include/linux/arm-smccc.h
> index 6f82c87308ed..aafb6bac167d 100644
> --- a/include/linux/arm-smccc.h
> +++ b/include/linux/arm-smccc.h
> @@ -94,6 +94,7 @@
>
> /* KVM "vendor specific" services */
> #define ARM_SMCCC_KVM_FUNC_FEATURES 0
> +#define ARM_SMCCC_KVM_PTP 1
> #define ARM_SMCCC_KVM_FUNC_FEATURES_2 127
> #define ARM_SMCCC_KVM_NUM_FUNCS 128
>
> @@ -103,6 +104,17 @@
> ARM_SMCCC_OWNER_VENDOR_HYP,
\
> ARM_SMCCC_KVM_FUNC_FEATURES)
>
> +/*
> + * This ID used for virtual ptp kvm clock and it will pass second
> value
> + * and nanosecond value of host real time and system counter by vcpu
> + * register to guest.
> + */
> +#define ARM_SMCCC_VENDOR_HYP_KVM_PTP_FUNC_ID
\
> + ARM_SMCCC_CALL_VAL(ARM_SMCCC_FAST_CALL,
\
> + ARM_SMCCC_SMC_32,
\
> + ARM_SMCCC_OWNER_VENDOR_HYP,
\
> + ARM_SMCCC_KVM_PTP)
> +
All of this depends on patches that have never need posted to any ML,
and
just linger in Will's tree. You need to pick them up and post them as
part of
this series so that they can at least be reviewed.
Ok, I will add them next version.
> #ifndef __ASSEMBLY__
>
> #include <linux/linkage.h>
> diff --git a/virt/kvm/arm/psci.c b/virt/kvm/arm/psci.c index
> 0debf49bf259..682d892d6717 100644
> --- a/virt/kvm/arm/psci.c
> +++ b/virt/kvm/arm/psci.c
> @@ -9,6 +9,7 @@
> #include <linux/kvm_host.h>
> #include <linux/uaccess.h>
> #include <linux/wait.h>
> +#include <linux/clocksource_ids.h>
>
> #include <asm/cputype.h>
> #include <asm/kvm_emulate.h>
> @@ -389,6 +390,8 @@ static int kvm_psci_call(struct kvm_vcpu *vcpu)
>
> int kvm_hvc_call_handler(struct kvm_vcpu *vcpu) {
> + struct system_time_snapshot systime_snapshot;
> + u64 cycles;
> u32 func_id = smccc_get_function(vcpu);
> u32 val[4] = {};
> u32 option;
> @@ -431,6 +434,25 @@ int kvm_hvc_call_handler(struct kvm_vcpu *vcpu)
> case ARM_SMCCC_VENDOR_HYP_KVM_FEATURES_FUNC_ID:
> val[0] = BIT(ARM_SMCCC_KVM_FUNC_FEATURES);
> break;
> + /*
> + * This will used for virtual ptp kvm clock. three
> + * values will be passed back.
> + * reg0 stores high 32-bit host ktime;
> + * reg1 stores low 32-bit host ktime;
> + * reg2 stores high 32-bit difference of host cycles and cntvoff;
> + * reg3 stores low 32-bit difference of host cycles and cntvoff.
That's either two or four values, and not three as you claim above.
Sorry, I'm not sure what do you mean "three", the registers here is 4
from reg0 to reg3.
Please read the comment you have written above...
Also, I fail to understand the meaning of the host cycle vs cntvoff
comparison.
This is something that guest can perform on its own (it has access to
both
physical and virtual timers, and can compute cntvoff without
intervention of
the hypervisor).
To keep consistency and precision, clock time and counter cycle must
captured at the same time. It will perform at ktime_get_snapshot.
Fair enough. It would vertainly help if you documented it. It would also
help if you explained why it is so much worse to read the counter in the
guest before *and* after the call, and assume that the clock time read
happened right in the middle?
That aside, what you are returning is something that *looks* like
the virtual counter. What if the guest is using the physical counter,
which is likely to be the case with nested virt? Do you expect the guest
to always use the virtual counter? This isn't going to fly.
Finally, how does it work with nested virt, where cntvoff is for the
the vEL2
guest?
For now, I have not considered ptp_kvm in nested virtualization. Also
I'm not sure about if nested virtualization is
ready on arm64 , as I need test ptp_kvm on it. If so, I can consider
it.
It is not about testing. It is about taking the architecture into
account.
And ready or not doesn't come into play here. What you're defining here
is
an ABI, and it better be totally future proof.
But if you want to test, help yourself to [1] and have fun!
> + */
> + case ARM_SMCCC_VENDOR_HYP_KVM_PTP_FUNC_ID:
> + ktime_get_snapshot(&systime_snapshot);
> + if (systime_snapshot.cs_id != CSID_ARM_ARCH_COUNTER)
> + return kvm_psci_call(vcpu);
What does this mean? Calling PSCI because you've failed to identify
the clock
source? What result do you expect from this? Hint: this isn't a PSCI
call.
Sorry, what I want to do here is that return to guest with the error
info.
Maybe I should set val[0] to -1 and break to let the guest know that
error comes, as
the guest will check if val[0] is positive to determine the next step.
What you should do is handle it like a normal SMCCC failure.
M.
[1]
https://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms.git/log/?h=kvm-arm64/nv-5.5-rc4-WIP
--
Jazz is not dead. It just smells funny...
