linux-kernel - Re: [virtio-dev] Re: [PATCH v1 2/2] virtio-mmio: add features for virtio-mmio specification version 3

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]

Date:   Mon, 6 Jan 2020 10:51:11 +0800
From:   Jason Wang <jasowang@...hat.com>
To:     "Michael S. Tsirkin" <mst@...hat.com>
Cc:     "Liu, Jiang" <gerry@...ux.alibaba.com>,
        "Liu, Jing2" <jing2.liu@...ux.intel.com>,
        Zha Bin <zhabin@...ux.alibaba.com>,
        linux-kernel@...r.kernel.org, slp@...hat.com,
        virtio-dev@...ts.oasis-open.org, jing2.liu@...el.com,
        chao.p.peng@...el.com
Subject: Re: [virtio-dev] Re: [PATCH v1 2/2] virtio-mmio: add features for
 virtio-mmio specification version 3


On 2020/1/5 下午7:25, Michael S. Tsirkin wrote:
> On Fri, Jan 03, 2020 at 05:12:38PM +0800, Jason Wang wrote:
>> On 2020/1/3 下午2:14, Liu, Jiang wrote:
>>>> Ok, I get you now.
>>>>
>>>> But still, having fixed number of MSIs is less flexible. E.g:
>>>>
>>>> - for x86, processor can only deal with about 250 interrupt vectors.
>>>> - driver may choose to share MSI vectors [1] (which is not merged but we will for sure need it)
>>> Thanks for the info:)
>>> X86 systems roughly have NCPU * 200 vectors available for device interrupts.
>>> The proposed patch tries to map multiple event sources to an interrupt vector, to avoid running out of x86 CPU vectors.
>>> Many virtio mmio devices may have several or tens of event sources, and it’s rare to have hundreds of event sources.
>>> So could we treat the dynamic mapping between event sources and interrupt vectors as an advanced optional feature?
>>>
>> Maybe, but I still prefer to implement it if it is not too complex. Let's
>> see Michael's opinion on this.
>>
>> Thanks
> I think a way for the device to limit # of vectors in use by driver is
> useful. But sharing of vectors doesn't really need any special
> registers, just program the same vector for multiple Qs/interrupts.


Right, but sine the #vectors is limited, we still need dynamic mapping 
like what is done in PCI.

Thanks


>