linux-kernel - Re: [PATCH v3] KVM: remove buggy vcpu id check on vcpu creation

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] [day] [month] [year] [list]

Message-ID: <CANRm+CxBUps+pKARfXse-s09Qrsc6pERNjBgVOjUBVCV+B2XPw@mail.gmail.com>
Date:	Sun, 24 Apr 2016 06:54:15 +0800
From:	Wanpeng Li <kernellwp@...il.com>
To:	Radim Krčmář <rkrcmar@...hat.com>
Cc:	Greg Kurz <gkurz@...ux.vnet.ibm.com>,
	Paolo Bonzini <pbonzini@...hat.com>, james.hogan@...tec.com,
	Ingo Molnar <mingo@...hat.com>, linux-mips@...ux-mips.org,
	kvm <kvm@...r.kernel.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	qemu-ppc@...gnu.org, Cornelia Huck <cornelia.huck@...ibm.com>,
	Paul Mackerras <paulus@...ba.org>,
	David Gibson <david@...son.dropbear.id.au>
Subject: Re: [PATCH v3] KVM: remove buggy vcpu id check on vcpu creation

2016-04-22 21:07 GMT+08:00 Radim Krčmář <rkrcmar@...hat.com>:
> 2016-04-22 09:40+0800, Wanpeng Li:
>> 2016-04-21 23:29 GMT+08:00 Radim Krčmář <rkrcmar@...hat.com>:
>>> x86 vcpu_id encodes APIC ID and APIC ID encodes CPU topology by
>>> reserving blocks of bits for socket/core/thread, so if core or thread
>>> count isn't a power of two, then the set of valid APIC IDs is sparse,
>>
>>              ^^^^^^^^^^^^^^^^^^^
>>              ^^^^^^^
>> Is this the root reason why recommand max vCPUs per vm is 160 and the
>> KVM_MAX_VCPUS is 255 instead of due to perforamnce concern?
>
> No, the recommended amout of VCPUs is 160 because I didn't bump it after
> PLE stopped killing big guests. :/
>
> You can get full 255 VCPU guest with a proper configuration, e.g.
> "-smp 255" or "-smp 255,cores=8" and the only problem is scalability,
> but I don't know of anything that doesn't scale to that point.
>
> (Scaling up to 2^32 is harder, because you don't want O(N) search, nor
>  full allocation on smaller guests.  Neither is a big problem now.)

I see, thanks Radim.

Regards,
Wanpeng Li