linux-kernel - Re: [Xen-devel] [PATCH 07/11] swiotlb-xen: provide a single page-coherent.h header

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: <21746bbf-618a-d12b-c767-f9e865f4dd20@arm.com>
Date:   Sat, 17 Aug 2019 19:20:27 +0100
From:   Julien Grall <julien.grall@....com>
To:     Christoph Hellwig <hch@....de>
Cc:     Stefano Stabellini <sstabellini@...nel.org>,
        Konrad Rzeszutek Wilk <konrad.wilk@...cle.com>,
        xen-devel@...ts.xenproject.org, iommu@...ts.linux-foundation.org,
        x86@...nel.org, linux-kernel@...r.kernel.org,
        linux-arm-kernel@...ts.infradead.org
Subject: Re: [Xen-devel] [PATCH 07/11] swiotlb-xen: provide a single
 page-coherent.h header

Hi Christoph,

On 8/17/19 7:50 AM, Christoph Hellwig wrote:
> On Fri, Aug 16, 2019 at 11:40:43PM +0100, Julien Grall wrote:
>> I am not sure I agree with this rename. The implementation of the helpers
>> are very Arm specific as this is assuming Dom0 is 1:1 mapped.
>>
>> This was necessary due to the lack of IOMMU on Arm platforms back then.
>> But this is now a pain to get rid of it on newer platform...
> 
> So if you look at the final version of the header after the whole
> series, what assumes a 1:1 mapping?  It all just is
> 
> 	if (pfn_valid())
> 		local cache sync;
> 	else
> 		call into the arch code;

In the context of Xen Arm, the dev_addr is a host physical address. From 
my understanding pfn_valid() is dealing with a guest physical frame.

Therefore by passing PFN_DOWN(dev_addr) in argument you assume that the 
host and guest address spaces are the same.

> 
> are you concerned that the local cache sync might have to be split
> up more for a non-1:1 map in that case?  We could just movea
> the xen_dma_* routines into the arch instead of __xen_dma, but it
> really helps to have a common interface header.
Moving xen_dma_* routines into the arch would be a good option. 
Although, I would still consider a stub version for arch not requiring 
specific DMA.

Cheers,

-- 
Julien Grall