Message-ID: <b6b55688-1390-4e92-8184-770154a8955d@linux.intel.com>
Date:   Fri, 27 Jan 2023 21:50:29 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:102.0) Gecko/20100101
 Thunderbird/102.6.1
Cc:     baolu.lu@linux.intel.com, joro@8bytes.org, will@kernel.org,
        hch@lst.de, iommu@lists.linux.dev, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 1/8] iommu: Decouple iommu_present() from bus ops
Content-Language: en-US
To:     Jason Gunthorpe <jgg@nvidia.com>,
        Robin Murphy <robin.murphy@arm.com>
References: <cover.1673978700.git.robin.murphy@arm.com>
 <1fb168b22cbbb5c24162d29d2a9aca339cda2c72.1673978700.git.robin.murphy@arm.com>
 <c96aaa6c-0f46-39dc-0c72-f38394e37cc3@linux.intel.com>
 <ce25dcdc-99a9-61ff-0cad-6c6cd9552680@arm.com> <Y9KRBRKdwSIRrvQw@nvidia.com>
From:   Baolu Lu <baolu.lu@linux.intel.com>
In-Reply-To: <Y9KRBRKdwSIRrvQw@nvidia.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 2023/1/26 22:41, Jason Gunthorpe wrote:
> On Thu, Jan 26, 2023 at 02:21:29PM +0000, Robin Murphy wrote:
> 
>> The "check" is inherent in the fact that it's been called at all. VFIO
>> noiommu*is*  an IOMMU driver in the sense that it provides a bare minimum of
>> IOMMU API functionality (i.e. creating groups), sufficient to support
>> (careful) usage by VFIO drivers. There would not seem to be a legitimate
>> reason for some*other*  driver to be specifically querying a device while it
>> is already bound to a VFIO driver (and thus may have a noiommu group).
> Yes, the devices that VFIO assigns to its internal groups never leak
> outside VFIO's control during their assignment - ie they are
> continuously bound to VFIO never another driver.
> 
> So no other driver can ever see the internal groups unless it is
> messing around with devices it is not bound to ????

Fair enough. I was thinking that probably we could make it like below:

/**
  * device_iommu_mapped - Returns true when the device DMA is translated
  *                       by an IOMMU
  * @dev: Device to perform the check on
  */
static inline bool device_iommu_mapped(struct device *dev)
{
         return (dev->iommu && dev->iommu->iommu_dev);
}

The iommu probe device code guarantees that dev->iommu->iommu_dev is
valid only after the IOMMU driver's .probe_device returned successfully.

Any thoughts?

Best regards,
baolu