[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <59c365d8-4b76-4e57-794b-e1d4b3f0e2a5@arm.com>
Date: Thu, 23 Jun 2022 10:01:34 +0100
From: Robin Murphy <robin.murphy@....com>
To: Dexuan Cui <decui@...rosoft.com>, Christoph Hellwig <hch@....de>
Cc: "m.szyprowski@...sung.com" <m.szyprowski@...sung.com>,
"iommu@...ts.linux-foundation.org" <iommu@...ts.linux-foundation.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"linux-hyperv@...r.kernel.org" <linux-hyperv@...r.kernel.org>,
"Michael Kelley (LINUX)" <mikelley@...rosoft.com>,
Andrea Parri <Andrea.Parri@...rosoft.com>,
Tianyu Lan <Tianyu.Lan@...rosoft.com>
Subject: Re: [PATCH] dma-direct: use the correct size for dma_set_encrypted()
On 2022-06-23 08:00, Dexuan Cui wrote:
>> From: Christoph Hellwig <hch@....de>
>> Sent: Wednesday, June 22, 2022 10:44 PM
>> To: Dexuan Cui <decui@...rosoft.com>
>> ...
>> On Wed, Jun 22, 2022 at 12:14:24PM -0700, Dexuan Cui wrote:
>>> The third parameter of dma_set_encrypted() is a size in bytes rather than
>>> the number of pages.
>>>
>>> Fixes: 4d0564785bb0 ("dma-direct: factor out dma_set_{de,en}crypted
>> helpers")
>>> Signed-off-by: Dexuan Cui <decui@...rosoft.com>
>>
>> see:
>>
>> commit 4a37f3dd9a83186cb88d44808ab35b78375082c9 (tag:
>> dma-mapping-5.19-2022-05-25)
>> Author: Robin Murphy <robin.murphy@....com>
>> Date: Fri May 20 18:10:13 2022 +0100
>>
>> dma-direct: don't over-decrypt memory
>
> It looks like commit 4a37f3dd9a831 fixed a different issue?
>
> Here my patch is for the latest mainline:
>
> In dma_direct_alloc()'s error handling path, we pass 'size' to dma_set_encrypted():
> out_encrypt_pages:
> if (dma_set_encrypted(dev, page_address(page), size))
>
> However, in dma_direct_free(), we pass ' 1 << page_order ' to dma_set_encrypted().
> I think the ' 1 << page_order' is incorrect and it should be 'size' as well?
I think technically you're both right - these instances clearly have a
history tracing back to the original bug that my patch addressed, but
the refactoring then made them into their own distinct bug in terms of
the internal dma_set_encrypted() interface, per the commit message here.
Apparently I failed to spot this when forward-porting 4a37f3dd9a831 from
5.10 (as the commit message says, don't ask... ;) ) - I guess I was only
looking at where the set_memory_*() callsites had moved to. For this patch,
Reviewed-by: Robin Murphy <robin.murphy@....com>
Thanks
Robin.
Powered by blists - more mailing lists