From: Paolo Bonzini <pbonzini@redhat.com>
To: "Radim Krčmář" <rkrcmar@redhat.com>, kvm@vger.kernel.org
Cc: linux-kernel@vger.kernel.org, Gleb Natapov <gleb@kernel.org>,
Raghavendra KT <raghavendra.kt@linux.vnet.ibm.com>,
Vinod Chegu <chegu_vinod@hp.com>, Hui-Zhi <hui-zhi.zhao@hp.com>,
Christian Borntraeger <borntraeger@de.ibm.com>
Subject: Re: [PATCH v2 4/6] KVM: VMX: dynamise PLE window
Date: Thu, 21 Aug 2014 10:26:38 +0200 [thread overview]
Message-ID: <53F5AD3E.3050800@redhat.com> (raw)
In-Reply-To: <1408567997-21222-5-git-send-email-rkrcmar@redhat.com>
Il 20/08/2014 22:53, Radim Krčmář ha scritto:
> Window is increased on every PLE exit and decreased on every sched_in.
> The idea is that we don't want to PLE exit if there is no preemption
> going on.
> We do this with sched_in() because it does not hold rq lock.
>
> There are two new kernel parameters for changing the window:
> ple_window_grow and ple_window_shrink
> ple_window_grow affects the window on PLE exit and ple_window_shrink
> does it on sched_in; depending on their value, the window is modifier
> like this: (ple_window is kvm_intel's global)
>
> ple_window_shrink/ |
> ple_window_grow | PLE exit | sched_in
> -------------------+--------------------+---------------------
> < 1 | = ple_window | = ple_window
> < ple_window | *= ple_window_grow | /= ple_window_shrink
> otherwise | += ple_window_grow | -= ple_window_shrink
>
> A third new parameter, ple_window_max, controls a maximal ple_window.
> A minimum equals to ple_window.
>
> Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
> ---
> arch/x86/kvm/vmx.c | 80 ++++++++++++++++++++++++++++++++++++++++++++++++++++--
> 1 file changed, 78 insertions(+), 2 deletions(-)
>
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index 18e0e52..e63d7ac 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -125,14 +125,32 @@ module_param(nested, bool, S_IRUGO);
> * Time is measured based on a counter that runs at the same rate as the TSC,
> * refer SDM volume 3b section 21.6.13 & 22.1.3.
> */
> -#define KVM_VMX_DEFAULT_PLE_GAP 128
> -#define KVM_VMX_DEFAULT_PLE_WINDOW 4096
> +#define KVM_VMX_DEFAULT_PLE_GAP 128
> +#define KVM_VMX_DEFAULT_PLE_WINDOW 4096
> +#define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
> +#define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
> +#define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
> + INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
> +
> static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
> module_param(ple_gap, int, S_IRUGO);
>
> static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
> module_param(ple_window, int, S_IRUGO);
>
> +/* Default doubles per-vcpu window every exit. */
> +static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
> +module_param(ple_window_grow, int, S_IRUGO);
> +
> +/* Default resets per-vcpu window every exit to ple_window. */
> +static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
> +module_param(ple_window_shrink, int, S_IRUGO);
> +
> +/* Default is to compute the maximum so we can never overflow. */
> +static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
> +static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
> +module_param(ple_window_max, int, S_IRUGO);
> +
> extern const ulong vmx_return;
>
> #define NR_AUTOLOAD_MSRS 8
> @@ -5679,12 +5697,66 @@ out:
> return ret;
> }
>
> +static int __grow_ple_window(int val)
> +{
> + if (ple_window_grow < 1)
> + return ple_window;
> +
> + val = min(val, ple_window_actual_max);
> +
> + if (ple_window_grow < ple_window)
> + val *= ple_window_grow;
> + else
> + val += ple_window_grow;
> +
> + return val;
> +}
> +
> +static int __shrink_ple_window(int val, int shrinker, int minimum)
s/shrinker/factor/ or s/shrinker/param/ (shrinker has another meaning in
the kernel).
> +{
> + if (shrinker < 1)
> + return ple_window;
> +
> + if (shrinker < ple_window)
> + val /= shrinker;
> + else
> + val -= shrinker;
> +
> + return max(val, minimum);
Any reason to use anything but ple_window as the minimum, even in
update_ple_window_actual_max?
> +}
> +
> +static void modify_ple_window(struct kvm_vcpu *vcpu, int grow)
> +{
> + struct vcpu_vmx *vmx = to_vmx(vcpu);
> + int new;
> +
> + if (grow)
> + new = __grow_ple_window(vmx->ple_window);
> + else
> + new = __shrink_ple_window(vmx->ple_window, ple_window_shrink,
> + ple_window);
> +
> + vmx->ple_window = max(new, ple_window);
> +}
> +#define grow_ple_window(vcpu) modify_ple_window(vcpu, 1)
> +#define shrink_ple_window(vcpu) modify_ple_window(vcpu, 0)
No macros please. :)
Paolo
> +
> +static void update_ple_window_actual_max(void)
> +{
> + ple_window_actual_max =
> + __shrink_ple_window(max(ple_window_max, ple_window),
> + ple_window_grow, INT_MIN);
> +}
> +
> /*
> * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
> * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
> */
> static int handle_pause(struct kvm_vcpu *vcpu)
> {
> + if (ple_gap)
> + grow_ple_window(vcpu);
> +
> skip_emulated_instruction(vcpu);
> kvm_vcpu_on_spin(vcpu);
>
> @@ -8854,6 +8926,8 @@ static int vmx_check_intercept(struct kvm_vcpu *vcpu,
>
> void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
> {
> + if (ple_gap)
> + shrink_ple_window(vcpu);
> }
>
> static struct kvm_x86_ops vmx_x86_ops = {
> @@ -9077,6 +9151,8 @@ static int __init vmx_init(void)
> } else
> kvm_disable_tdp();
>
> + update_ple_window_actual_max();
> +
> return 0;
>
> out7:
>
next prev parent reply other threads:[~2014-08-21 8:26 UTC|newest]
Thread overview: 28+ messages / expand[flat|nested] mbox.gz Atom feed top
2014-08-20 20:53 [PATCH v2 0/6] Dynamic Pause Loop Exiting window Radim Krčmář
2014-08-20 20:53 ` [PATCH v2 1/6] KVM: add kvm_arch_sched_in Radim Krčmář
2014-08-21 8:29 ` Paolo Bonzini
2014-08-21 11:38 ` Radim Krčmář
2014-08-21 12:27 ` Paolo Bonzini
2014-08-21 12:50 ` Radim Krčmář
2014-08-21 13:25 ` Paolo Bonzini
2014-08-20 20:53 ` [PATCH v2 2/6] KVM: x86: introduce sched_in to kvm_x86_ops Radim Krčmář
2014-08-20 20:53 ` [PATCH v2 3/6] KVM: VMX: make PLE window per-VCPU Radim Krčmář
2014-08-21 8:25 ` Paolo Bonzini
2014-08-21 11:38 ` Radim Krčmář
2014-08-20 20:53 ` [PATCH v2 4/6] KVM: VMX: dynamise PLE window Radim Krčmář
2014-08-21 8:24 ` Paolo Bonzini
2014-08-21 11:47 ` Radim Krčmář
2014-08-21 8:26 ` Paolo Bonzini [this message]
2014-08-21 11:54 ` Radim Krčmář
2014-08-21 12:29 ` Paolo Bonzini
2014-08-21 12:42 ` Radim Krčmář
2014-08-21 13:18 ` Paolo Bonzini
2014-08-21 13:46 ` Radim Krčmář
2014-08-20 20:53 ` [PATCH v2 5/6] KVM: trace kvm_ple_window Radim Krčmář
2014-08-21 8:29 ` Paolo Bonzini
2014-08-21 11:56 ` Radim Krčmář
2014-08-21 13:22 ` Paolo Bonzini
2014-08-21 13:49 ` Radim Krčmář
2014-08-20 20:53 ` [PATCH v2 6/6] KVM: VMX: runtime knobs for dynamic PLE window Radim Krčmář
2014-08-21 6:49 ` [PATCH v2 0/6] Dynamic Pause Loop Exiting window Zhao, Hui-Zhi (Steven, HPservers-Core-OE-PSC)
2014-08-21 6:49 ` Zhao, Hui-Zhi (Steven, HPservers-Core-OE-PSC)
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=53F5AD3E.3050800@redhat.com \
--to=pbonzini@redhat.com \
--cc=borntraeger@de.ibm.com \
--cc=chegu_vinod@hp.com \
--cc=gleb@kernel.org \
--cc=hui-zhi.zhao@hp.com \
--cc=kvm@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=raghavendra.kt@linux.vnet.ibm.com \
--cc=rkrcmar@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.