linux-kernel - Re: [PATCH v4 1/4] x86/kvm: Add "nopvspin" parameter to disable PV spinlocks

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <b0d6df7b-00ff-cdd8-f9f2-26af73256f5b@oracle.com>
Date:   Sun, 6 Oct 2019 15:49:01 +0800
From:   Zhenzhong Duan <zhenzhong.duan@...cle.com>
To:     Boris Ostrovsky <boris.ostrovsky@...cle.com>,
        linux-kernel@...r.kernel.org
Cc:     vkuznets@...hat.com, linux-hyperv@...r.kernel.org,
        kvm@...r.kernel.org, kys@...rosoft.com, haiyangz@...rosoft.com,
        sthemmin@...rosoft.com, sashal@...nel.org, tglx@...utronix.de,
        mingo@...hat.com, bp@...en8.de, pbonzini@...hat.com,
        rkrcmar@...hat.com, sean.j.christopherson@...el.com,
        wanpengli@...cent.com, jmattson@...gle.com, joro@...tes.org,
        jgross@...e.com, sstabellini@...nel.org, peterz@...radead.org,
        Jonathan Corbet <corbet@....net>,
        "H. Peter Anvin" <hpa@...or.com>, Will Deacon <will@...nel.org>
Subject: Re: [PATCH v4 1/4] x86/kvm: Add "nopvspin" parameter to disable PV
 spinlocks

On 2019/10/4 22:52, Boris Ostrovsky wrote:

> On 10/3/19 10:02 AM, Zhenzhong Duan wrote:
>>   void __init kvm_spinlock_init(void)
>>   {
>> -	/* Does host kernel support KVM_FEATURE_PV_UNHALT? */
>> -	if (!kvm_para_has_feature(KVM_FEATURE_PV_UNHALT))
>> -		return;
>> -
>> -	if (kvm_para_has_hint(KVM_HINTS_REALTIME))
>> +	/*
>> +	 * Don't use the pvqspinlock code if no KVM_FEATURE_PV_UNHALT feature
>> +	 * support, or there is REALTIME hints or only 1 vCPU.
>> +	 */
>> +	if (!kvm_para_has_feature(KVM_FEATURE_PV_UNHALT) ||
>> +	    kvm_para_has_hint(KVM_HINTS_REALTIME) ||
>> +	    num_possible_cpus() == 1) {
>> +		pr_info("PV spinlocks disabled\n");
>>   		return;
>> +	}
>>   
>> -	/* Don't use the pvqspinlock code if there is only 1 vCPU. */
>> -	if (num_possible_cpus() == 1)
>> +	if (nopvspin) {
>> +		pr_info("PV spinlocks disabled forced by \"nopvspin\" parameter.\n");
>> +		static_branch_disable(&virt_spin_lock_key);
> Would it make sense to bring here the other site where the key is
> disabled (in kvm_smp_prepare_cpus())?

Thanks for point out, I'll do it. Just not clear if I should do that in a separate patch,
there is a history about that code:

Its original place was here and then moved to kvm_smp_prepare_cpus() by below commit:
34226b6b ("KVM: X86: Fix setup the virt_spin_lock_key before static key get initialized")
which fixed jump_label_init() calling late issue.

Then 8990cac6 ("x86/jump_label: Initialize static branching early") move jump_label_init()
early, so commit 34226b6b could be reverted.

>
> (and, in fact, shouldn't all of the checks that result in early return
> above disable the key?)

I think we should enable he key for !kvm_para_has_feature(KVM_FEATURE_PV_UNHALT) case,
there is lock holder preemption issue as qspinlock is fair lock, virt_spin_lock()
is an optimization to that, imaging one pcpu running 10 vcpus of same guest
contending a same lock.

For kvm_para_has_hint(KVM_HINTS_REALTIME) case, hypervisor hints there is
no preemption and we should disable virt_spin_lock_key to use native qspinlock.

For the UP case, we don't care virt_spin_lock_key value.

For nopvspin case, we intentionally check native qspinlock code performance,
compare it with PV qspinlock, etc. So virt_spin_lock() optimization should be disabled.

Let me know if anything wrong with above understanding. Thanks

Zhenzhong