From: Jan Kiszka <jan.kiszka@siemens.com>
To: Alexander Graf <agraf@suse.de>
Cc: "kvm@vger.kernel.org" <kvm@vger.kernel.org>
Subject: Re: [PATCH 1/5] Implement #NMI exiting for nested SVM
Date: Fri, 18 Sep 2009 18:01:52 +0200 [thread overview]
Message-ID: <4AB3AEF0.8020500@siemens.com> (raw)
In-Reply-To: <C29673BB-58BE-4918-90B1-4D7C838A60F1@suse.de>
Alexander Graf wrote:
> Am 18.09.2009 um 15:33 schrieb Jan Kiszka <jan.kiszka@siemens.com>:
>
>> Alexander Graf wrote:
>>> When injecting an NMI to the l1 guest while it was running the l2
>>> guest, we
>>> didn't #VMEXIT but just injected the NMI to the l2 guest.
>>>
>>> Let's be closer to real hardware and #VMEXIT if we're supposed to
>>> do so.
>>>
>>> Signed-off-by: Alexander Graf <agraf@suse.de>
>>> ---
>>> arch/x86/kvm/svm.c | 38 ++++++++++++++++++++++++++++++++------
>>> 1 files changed, 32 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
>>> index 9a4daca..f12a669 100644
>>> --- a/arch/x86/kvm/svm.c
>>> +++ b/arch/x86/kvm/svm.c
>>> @@ -1375,6 +1375,21 @@ static int nested_svm_check_exception(struct
>>> vcpu_svm *svm, unsigned nr,
>>> return nested_svm_exit_handled(svm);
>>> }
>>>
>>> +static inline int nested_svm_nmi(struct vcpu_svm *svm)
>>> +{
>>> + if (!is_nested(svm))
>>> + return 0;
>>> +
>>> + svm->vmcb->control.exit_code = SVM_EXIT_NMI;
>>> +
>>> + if (nested_svm_exit_handled(svm)) {
>>> + nsvm_printk("VMexit -> NMI\n");
>>> + return 1;
>>> + }
>>> +
>>> + return 0;
>>> +}
>>> +
>>> static inline int nested_svm_intr(struct vcpu_svm *svm)
>>> {
>>> if (!is_nested(svm))
>>> @@ -2462,7 +2477,9 @@ static int svm_nmi_allowed(struct kvm_vcpu
>>> *vcpu)
>>> struct vcpu_svm *svm = to_svm(vcpu);
>>> struct vmcb *vmcb = svm->vmcb;
>>> return !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
>>> - !(svm->vcpu.arch.hflags & HF_NMI_MASK);
>>> + !(svm->vcpu.arch.hflags & HF_NMI_MASK) &&
>>> + gif_set(svm) &&
>> ^^^^^^^^^^^^
>> I'm not claiming to be up-to-date with the SVM code around NMI
>> injection, but this addition irritates me. Can you explain why I don't
>> have to worry that a cleared IF could now defer NMI injections for L1
>> guests?
>
> It's not about IF, but GIF, which is a special SVM addition that also
> masks NMIs.
Ah, now I got it: That's normally a host thing but, due to nesting, we
need to consider it for L1, too. OK, something learned today. :)
>
>>> + !is_nested(svm);
>>> }
>>>
>>> static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
>>> @@ -2488,22 +2505,31 @@ static void enable_irq_window(struct
>>> kvm_vcpu *vcpu)
>>> struct vcpu_svm *svm = to_svm(vcpu);
>>> nsvm_printk("Trying to open IRQ window\n");
>>>
>>> - nested_svm_intr(svm);
>>> + if (nested_svm_intr(svm))
>>> + return;
>>>
>>> /* In case GIF=0 we can't rely on the CPU to tell us when
>>> * GIF becomes 1, because that's a separate STGI/VMRUN intercept.
>>> * The next time we get that intercept, this function will be
>>> * called again though and we'll get the vintr intercept. */
>>> - if (gif_set(svm)) {
>>> - svm_set_vintr(svm);
>>> - svm_inject_irq(svm, 0x0);
>>> - }
>>> + if (!gif_set(svm))
>>> + return;
>>> +
>>> + svm_set_vintr(svm);
>>> + svm_inject_irq(svm, 0x0);
>> The last change is pure refactoring that should not belong into this
>> patch, should it?
>
> It went along the same function and makes the code more alike. But if
> you feel like this really should be separate, I can split it.
I've been slapped a few times for mixing both, but I'm not the person
who will merge it.
>
>>> }
>>>
>>> static void enable_nmi_window(struct kvm_vcpu *vcpu)
>>> {
>>> struct vcpu_svm *svm = to_svm(vcpu);
>>>
>>> + if (nested_svm_nmi(svm))
>>> + return;
>>> +
>>> + /* NMI is deferred until GIF == 1. Setting GIF will cause a
>>> #VMEXIT */
>>> + if (!gif_set(svm))
>>> + return;
>>> +
>> The second half is an unrelated optimization? Then please file a
>> separate patch.
>
> Nope, it's about not injecting NMIs while GIF is not set again.
Yes, got it.
Jan
--
Siemens AG, Corporate Technology, CT SE 2
Corporate Competence Center Embedded Linux
next prev parent reply other threads:[~2009-09-18 16:01 UTC|newest]
Thread overview: 19+ messages / expand[flat|nested] mbox.gz Atom feed top
2009-09-18 13:00 [PATCH 0/5] Nested SVM Interrupt Fixes Alexander Graf
2009-09-18 13:00 ` [PATCH 1/5] Implement #NMI exiting for nested SVM Alexander Graf
2009-09-18 13:00 ` [PATCH 2/5] Don't call svm_complete_interrupts for nested guests Alexander Graf
2009-09-18 13:00 ` [PATCH 3/5] Don't #VMEXIT(INTR) if we still have event_inj waiting Alexander Graf
2009-09-18 13:00 ` [PATCH 4/5] Don't bail when injecting an event in nested SVM Alexander Graf
2009-09-18 13:00 ` [PATCH 5/5] Notify nested hypervisor of lost event injections Alexander Graf
2009-09-23 1:22 ` Joerg Roedel
2009-09-27 14:18 ` Joerg Roedel
2009-09-23 1:39 ` [PATCH 3/5] Don't #VMEXIT(INTR) if we still have event_inj waiting Joerg Roedel
2009-09-23 8:09 ` Alexander Graf
2009-09-18 13:39 ` [PATCH 2/5] Don't call svm_complete_interrupts for nested guests Jan Kiszka
2009-09-23 1:26 ` Joerg Roedel
2009-09-23 8:04 ` Alexander Graf
2009-09-23 8:05 ` Alexander Graf
2009-09-23 8:28 ` Gleb Natapov
2009-09-18 13:33 ` [PATCH 1/5] Implement #NMI exiting for nested SVM Jan Kiszka
2009-09-18 15:44 ` Alexander Graf
2009-09-18 16:01 ` Jan Kiszka [this message]
2009-09-23 1:06 ` Joerg Roedel
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=4AB3AEF0.8020500@siemens.com \
--to=jan.kiszka@siemens.com \
--cc=agraf@suse.de \
--cc=kvm@vger.kernel.org \
/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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).