linux-kernel - Re: [PATCH v2 00/59] KVM: iommu: Overhaul device posted IRQs support

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: <55b5827b-6809-47a9-b5d9-57fa68736e9f@amd.com>
Date: Thu, 12 Jun 2025 16:33:52 +0530
From: Sairaj Kodilkar <sarunkod@....com>
To: Sean Christopherson <seanjc@...gle.com>
CC: <baolu.lu@...ux.intel.com>, <dmatlack@...gle.com>, <dwmw2@...radead.org>,
	<francescolavra.fl@...il.com>, <iommu@...ts.linux.dev>,
	<joao.m.martins@...cle.com>, <joro@...tes.org>, <kvm@...r.kernel.org>,
	<linux-kernel@...r.kernel.org>, <mlevitsk@...hat.com>, <pbonzini@...hat.com>,
	<vasant.hegde@....com>
Subject: Re: [PATCH v2 00/59] KVM: iommu: Overhaul device posted IRQs support



On 6/9/2025 8:04 PM, Sean Christopherson wrote:
> On Mon, Jun 09, 2025, Sairaj Kodilkar wrote:
>> Hi Sean,
>>
>> Sorry for the delay in testing. All sanity tests are OK. I reran the performance
>> test on the V2 and noticed that V2 has significantly more GALOG entries than V1
>> for all three cases. I also noticed that the Guest Nvme interrupt rate has
>> dropped for the 192 VCPUS.
> 
> Hmm, I don't see any obvious bugs or differences (based on a code diffed between
> v1 and v2).  I'll poke at the GALogIntr behavior just to double check, but my
> guess is that the differences are due to exernal factors, e.g. guest behavior,
> timing, scheduling, etc.
> 
> IOPS are all nearly identical, so I'm not terribly concerned.
>   
>

Yep you are right. I was indeed using different guest kernel to test V2.
Keeping it same, I can produce almost identical results for both V1 and
V2.

Only one case that still stands out is with 32 vCPUS, where posted
interrupt count has increased from 200 to 7000. But IOPS and NVME
interrupt rate is identical hence I am not concerned about it as well.


Thanks
Sairaj