linux-kernel - Re: [PATCH V4 5/6] perf/amd/iommu: Enable support for multiple IOMMUs

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: <20160222222457.2824f7e0@t450s.home>
Date:	Mon, 22 Feb 2016 22:24:57 -0700
From:	Alex Williamson <alex.williamson@...hat.com>
To:	Suravee Suthikulpanit <Suravee.Suthikulpanit@....com>
Cc:	Peter Zijlstra <peterz@...radead.org>,
	iommu@...ts.linux-foundation.org, linux-kernel@...r.kernel.org,
	acme@...nel.org, andihartmann@...enet.de, mingo@...hat.com,
	bp@...en8.de
Subject: Re: [PATCH V4 5/6] perf/amd/iommu: Enable support for multiple
 IOMMUs

On Tue, 23 Feb 2016 12:12:42 +0700
Suravee Suthikulpanit <Suravee.Suthikulpanit@....com> wrote:

> Hi
> 
> On 02/22/2016 09:07 PM, Peter Zijlstra wrote:
> > On Mon, Feb 22, 2016 at 03:00:31PM +0700, Suravee Suthikulpanit wrote:  
> >>> So I really don't have time to review new muck while I'm hunting perf
> >>> core fail, but Boris made me look at this.
> >>>
> >>> This is crazy, if you have multiple IOMMUs then create an event per
> >>> IOMMU, do _NOT_ fold them all into a single event.  
> >>
> >> These are system-wide events, which are programmed on every IOMMU the same
> >> way. I am not sure what you meant by creating an event per IOMMU. Do you
> >> mean I should create internal per-IOMMU struct perf_event for each event?  
> >
> > No, I meant to expose each IOMMU individually to userspace, as a
> > separate device.
> >
> > Is there never a case to profile just one of the IOMMUs ?
> >  
> 
> I see. That's definitely doable and simpler to implement.
> 
> I was not sure if making users specify the IOMMU instance (e.g. 
> amd_iommu_0/<ev name> , amd_iommu_1/<ev_name>, ....) would be too 
> tedious. However, this would actually give users better control of the 
> performance events, which is a good trade-off. I think it is acceptable.
> 
> I'll make the change and send this out in V5.

We already expose individual IOMMU hardware units in /sys/class/iommu/,
you might consider trying to match the names there for the convenience
of the user.  Looks like we use ivhd%d for AMD.  Thanks,

Alex