KVM Archive on lore.kernel.org
 help / color / Atom feed
From: Lu Baolu <baolu.lu@linux.intel.com>
To: Alex Williamson <alex.williamson@redhat.com>
Cc: baolu.lu@linux.intel.com, "Tian, Kevin" <kevin.tian@intel.com>,
	Jacob Pan <jacob.jun.pan@linux.intel.com>,
	Joerg Roedel <joro@8bytes.org>,
	Jean-Philippe Brucker <jean-philippe@linaro.org>,
	"Jiang, Dave" <dave.jiang@intel.com>,
	"Raj, Ashok" <ashok.raj@intel.com>,
	"kvm@vger.kernel.org" <kvm@vger.kernel.org>,
	Cornelia Huck <cohuck@redhat.com>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	"iommu@lists.linux-foundation.org"
	<iommu@lists.linux-foundation.org>,
	Robin Murphy <robin.murphy@arm.com>
Subject: Re: [PATCH v3 2/4] iommu: Add iommu_aux_at(de)tach_group()
Date: Mon, 3 Aug 2020 09:57:24 +0800
Message-ID: <9ee76708-f50c-adb1-0b3f-67871c799cab@linux.intel.com> (raw)
In-Reply-To: <20200731120537.2e1d8916@x1.home>

Hi Alex,

On 8/1/20 2:05 AM, Alex Williamson wrote:
> On Fri, 31 Jul 2020 13:47:57 +0800
> Lu Baolu <baolu.lu@linux.intel.com> wrote:
> 
>> Hi Alex,
>>
>> On 2020/7/31 3:46, Alex Williamson wrote:
>>> On Wed, 29 Jul 2020 23:34:40 +0000
>>> "Tian, Kevin" <kevin.tian@intel.com> wrote:
>>>    
>>>>> From: Alex Williamson <alex.williamson@redhat.com>
>>>>> Sent: Thursday, July 30, 2020 4:04 AM
>>>>>
>>>>> On Thu, 16 Jul 2020 09:07:46 +0800
>>>>> Lu Baolu <baolu.lu@linux.intel.com> wrote:
>>>>>       
>>>>>> Hi Jacob,
>>>>>>
>>>>>> On 7/16/20 12:01 AM, Jacob Pan wrote:
>>>>>>> On Wed, 15 Jul 2020 08:47:36 +0800
>>>>>>> Lu Baolu <baolu.lu@linux.intel.com> wrote:
>>>>>>>      
>>>>>>>> Hi Jacob,
>>>>>>>>
>>>>>>>> On 7/15/20 12:39 AM, Jacob Pan wrote:
>>>>>>>>> On Tue, 14 Jul 2020 13:57:01 +0800
>>>>>>>>> Lu Baolu<baolu.lu@linux.intel.com>  wrote:
>>>>>>>>>      
>>>>>>>>>> This adds two new aux-domain APIs for a use case like vfio/mdev
>>>>>>>>>> where sub-devices derived from an aux-domain capable device are
>>>>>>>>>> created and put in an iommu_group.
>>>>>>>>>>
>>>>>>>>>> /**
>>>>>>>>>>      * iommu_aux_attach_group - attach an aux-domain to an
>>>>> iommu_group
>>>>>>>>>> which
>>>>>>>>>>      *                          contains sub-devices (for example
>>>>>>>>>> mdevs) derived
>>>>>>>>>>      *                          from @dev.
>>>>>>>>>>      * @domain: an aux-domain;
>>>>>>>>>>      * @group:  an iommu_group which contains sub-devices derived
>>>>> from
>>>>>>>>>> @dev;
>>>>>>>>>>      * @dev:    the physical device which supports
>>>>> IOMMU_DEV_FEAT_AUX.
>>>>>>>>>>      *
>>>>>>>>>>      * Returns 0 on success, or an error value.
>>>>>>>>>>      */
>>>>>>>>>> int iommu_aux_attach_group(struct iommu_domain *domain,
>>>>>>>>>>                                struct iommu_group *group,
>>>>>>>>>>                                struct device *dev)
>>>>>>>>>>
>>>>>>>>>> /**
>>>>>>>>>>      * iommu_aux_detach_group - detach an aux-domain from an
>>>>>>>>>> iommu_group *
>>>>>>>>>>      * @domain: an aux-domain;
>>>>>>>>>>      * @group:  an iommu_group which contains sub-devices derived
>>>>> from
>>>>>>>>>> @dev;
>>>>>>>>>>      * @dev:    the physical device which supports
>>>>> IOMMU_DEV_FEAT_AUX.
>>>>>>>>>>      *
>>>>>>>>>>      * @domain must have been attached to @group via
>>>>>>>>>> iommu_aux_attach_group(). */
>>>>>>>>>> void iommu_aux_detach_group(struct iommu_domain *domain,
>>>>>>>>>>                                 struct iommu_group *group,
>>>>>>>>>>                                 struct device *dev)
>>>>>>>>>>
>>>>>>>>>> It also adds a flag in the iommu_group data structure to identify
>>>>>>>>>> an iommu_group with aux-domain attached from those normal ones.
>>>>>>>>>>
>>>>>>>>>> Signed-off-by: Lu Baolu<baolu.lu@linux.intel.com>
>>>>>>>>>> ---
>>>>>>>>>>      drivers/iommu/iommu.c | 58
>>>>>>>>>> +++++++++++++++++++++++++++++++++++++++++++
>>>>> include/linux/iommu.h |
>>>>>>>>>> 17 +++++++++++++ 2 files changed, 75 insertions(+)
>>>>>>>>>>
>>>>>>>>>> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
>>>>>>>>>> index e1fdd3531d65..cad5a19ebf22 100644
>>>>>>>>>> --- a/drivers/iommu/iommu.c
>>>>>>>>>> +++ b/drivers/iommu/iommu.c
>>>>>>>>>> @@ -45,6 +45,7 @@ struct iommu_group {
>>>>>>>>>>      	struct iommu_domain *default_domain;
>>>>>>>>>>      	struct iommu_domain *domain;
>>>>>>>>>>      	struct list_head entry;
>>>>>>>>>> +	unsigned int aux_domain_attached:1;
>>>>>>>>>>      };
>>>>>>>>>>
>>>>>>>>>>      struct group_device {
>>>>>>>>>> @@ -2759,6 +2760,63 @@ int iommu_aux_get_pasid(struct
>>>>> iommu_domain
>>>>>>>>>> *domain, struct device *dev) }
>>>>>>>>>>      EXPORT_SYMBOL_GPL(iommu_aux_get_pasid);
>>>>>>>>>>
>>>>>>>>>> +/**
>>>>>>>>>> + * iommu_aux_attach_group - attach an aux-domain to an
>>>>> iommu_group
>>>>>>>>>> which
>>>>>>>>>> + *                          contains sub-devices (for example
>>>>>>>>>> mdevs) derived
>>>>>>>>>> + *                          from @dev.
>>>>>>>>>> + * @domain: an aux-domain;
>>>>>>>>>> + * @group:  an iommu_group which contains sub-devices derived
>>>>> from
>>>>>>>>>> @dev;
>>>>>>>>>> + * @dev:    the physical device which supports
>>>>> IOMMU_DEV_FEAT_AUX.
>>>>>>>>>> + *
>>>>>>>>>> + * Returns 0 on success, or an error value.
>>>>>>>>>> + */
>>>>>>>>>> +int iommu_aux_attach_group(struct iommu_domain *domain,
>>>>>>>>>> +			   struct iommu_group *group, struct
>>>>>>>>>> device *dev) +{
>>>>>>>>>> +	int ret = -EBUSY;
>>>>>>>>>> +
>>>>>>>>>> +	mutex_lock(&group->mutex);
>>>>>>>>>> +	if (group->domain)
>>>>>>>>>> +		goto out_unlock;
>>>>>>>>>> +
>>>>>>>>> Perhaps I missed something but are we assuming only one mdev per
>>>>>>>>> mdev group? That seems to change the logic where vfio does:
>>>>>>>>> iommu_group_for_each_dev()
>>>>>>>>> 	iommu_aux_attach_device()
>>>>>>>>>      
>>>>>>>>
>>>>>>>> It has been changed in PATCH 4/4:
>>>>>>>>
>>>>>>>> static int vfio_iommu_attach_group(struct vfio_domain *domain,
>>>>>>>>                                        struct vfio_group *group)
>>>>>>>> {
>>>>>>>>             if (group->mdev_group)
>>>>>>>>                     return iommu_aux_attach_group(domain->domain,
>>>>>>>>                                                   group->iommu_group,
>>>>>>>>                                                   group->iommu_device);
>>>>>>>>             else
>>>>>>>>                     return iommu_attach_group(domain->domain,
>>>>>>>> group->iommu_group);
>>>>>>>> }
>>>>>>>>
>>>>>>>> So, for both normal domain and aux-domain, we use the same concept:
>>>>>>>> attach a domain to a group.
>>>>>>>>      
>>>>>>> I get that, but don't you have to attach all the devices within the
>>>>>>
>>>>>> This iommu_group includes only mediated devices derived from an
>>>>>> IOMMU_DEV_FEAT_AUX-capable device. Different from
>>>>> iommu_attach_group(),
>>>>>> iommu_aux_attach_group() doesn't need to attach the domain to each
>>>>>> device in group, instead it only needs to attach the domain to the
>>>>>> physical device where the mdev's were created from.
>>>>>>      
>>>>>>> group? Here you see the group already has a domain and exit.
>>>>>>
>>>>>> If the (group->domain) has been set, that means a domain has already
>>>>>> attached to the group, so it returns -EBUSY.
>>>>>
>>>>> I agree with Jacob, singleton groups should not be built into the IOMMU
>>>>> API, we're not building an interface just for mdevs or current
>>>>> limitations of mdevs.  This also means that setting a flag on the group
>>>>> and passing a device that's assumed to be common for all devices within
>>>>> the group, don't really make sense here.  Thanks,
>>>>>
>>>>> Alex
>>>>
>>>> Baolu and I discussed about this assumption before. The assumption is
>>>> not based on singleton groups. We do consider multiple mdevs in one
>>>> group. But our feeling at the moment is that all mdevs (or other AUX
>>>> derivatives) in the same group should come from the same parent
>>>> device, thus comes with above design. Does it sound a reasonable
>>>> assumption to you?
>>>
>>> No, the approach in this series doesn't really make sense to me.  We
>>> currently have the following workflow as Baolu notes in the cover
>>> letter:
>>>
>>> 	domain = iommu_domain_alloc(bus);
>>>
>>> 	iommu_group_for_each_dev(group...
>>>
>>> 		iommu_device = mdev-magic()
>>>
>>> 		if (iommu_dev_feature_enabled(iommu_device,
>>> 						IOMMU_DEV_FEAT_AUX))
>>> 			iommu_aux_attach_device(domain, iommu_device);
>>>
>>> And we want to convert this to a group function, like we have for
>>> non-aux domains:
>>>
>>> 	domain = iommu_domain_alloc(bus);
>>>
>>> 	iommu_device = mdev-magic()
>>>
>>> 	iommu_aux_attach_group(domain, group, iommu_device);
>>>
>>> And I think we want to do that largely because iommu_group.domain is
>>> private to iommu.c (therefore vfio code cannot set it), but we need it
>>> set in order for iommu_get_domain_for_dev() to work with a group
>>> attached to an aux domain.  Passing an iommu_device avoids the problem
>>> that IOMMU API code doesn't know how to derive an iommu_device for each
>>> device in the group, but while doing so it ignores the fundamental
>>> nature of a group as being a set of one or more devices.  Even if we
>>> can make the leap that all devices within the group would use the same
>>> iommu_device, an API that sets and aux domain for a group while
>>> entirely ignoring the devices within the group seems very broken.
>>
>> Agreed. We couldn't assume that all devices in an iommu group shares a
>> same iommu_device, especially when it comes to PF/VF wrapped mediated
>> device case.
>>
>>>
>>> So, barring adding an abstraction at struct device where an IOMMU API
>>> could retrieve the iommu_device backing anther device (which seems a
>>> very abstract concept for the base class), why not have the caller
>>> provide a lookup function?  Ex:
>>>
>>> int iommu_aux_attach_group(struct iommu_domain *domain,
>>> 			   struct iommu_group *group,
>>> 			   struct device *(*iommu_device_lookup)(
>>> 				struct device *dev));
>>>
>>> Thus vfio could could simply provide &vfio_mdev_get_iommu_device and
>>> we'd have equivalent functionality to what we have currently, but with
>>> the domain pointer set in the iommu_group.
>>
>> This looks good to me.
>>
>>>
>>> This also however highlights that our VF backed mdevs will have the
>>> same issue, so maybe this new IOMMU API interface should mimic
>>> vfio_mdev_attach_domain() more directly, testing whether the resulting
>>> device supports IOMMU_DEV_FEAT_AUX and using an aux vs non-aux attach.
>>> I'm not sure what the name of this combined function should be,
>>> iommu_attach_group_with_lookup()?  This could be the core
>>> implementation of iommu_attach_group() where the existing function
>>> simply wraps the call with a NULL function pointer.
>>>
>>> Anyway, I think there are ways to implement this that are more in line
>>> with the spirit of groups.
>>
>> Another possible implementation, just for discussion purpose:
>>
>> 1. Add a member in group_device to save the iommu_device if it exists:
>>
>> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
>> index b6858adc4f17..6474e82cf4b4 100644
>> --- a/drivers/iommu/iommu.c
>> +++ b/drivers/iommu/iommu.c
>> @@ -47,9 +47,16 @@ struct iommu_group {
>>           struct list_head entry;
>>    };
>>
>> +/*
>> + * dma_alias: The device put in this group might depends on another
>> + *            physical device to do the DMA remapping. At(de)taching
>> + *            the domain to/from @dma_alias instead of @dev if
>> + *            @dma_alias is set.
>> + */
>>    struct group_device {
>>           struct list_head list;
>>           struct device *dev;
>> +       struct device *dma_alias;
>>           char *name;
>>    };
>>
>> 2. Pass in the iommu_device when calling iommu_group_add_device().
>>
>> int iommu_group_add_device(struct iommu_group *group,
>>                              struct device *dev,
>>                              struct device *dma_alias)
>>
>> Hence, the iommu core could get a chance to set the iommu_device in the
>> group device.
>>
>> 3. Mimic vfio_mdev_attach_domain() logic in iommu_group_do_attach_device():
>>
>> if (group->dma_alias) {
>> 	if (iommu_dev_feature_enabled(group->dma_alias, IOMMU_DEV_FEAT_AUX))
>> 		iommu_aux_attach_device(domain, group->dma_alias);
>> 	else
>> 		__iommu_attach_device(domain, group->dma_alias);
>> } else {
>> 	__iommu_attach_device(domain, dev);
>> }
>>
>> One limitation is that the driver should call mdev_set_iommu_device()
>> before the mdev_probe() get called.
> 
> That's an option, but DMA aliases are an existing thing within our
> IOMMU/PCI constructs, so I'd steer away from "dma_alias" terminology.  I
> thought the callback was a little less invasive to the IOMMU layer for
> now as aux domains are still a rather unique use case, and I'm not sure
> we can justify the hack of otherwise IOMMU backed mdevs formally within
> the IOMMU API.

Fair enough. I will start with lookup callback solution.

Best regards,
baolu

  reply index

Thread overview: 36+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-07-14  5:56 [PATCH v3 0/4] iommu aux-domain APIs extensions Lu Baolu
2020-07-14  5:57 ` [PATCH v3 1/4] iommu: Check IOMMU_DEV_FEAT_AUX feature in aux api's Lu Baolu
2020-07-29 20:03   ` Alex Williamson
2020-07-30  1:46     ` Lu Baolu
2020-07-14  5:57 ` [PATCH v3 2/4] iommu: Add iommu_aux_at(de)tach_group() Lu Baolu
2020-07-14 16:39   ` Jacob Pan
2020-07-15  0:47     ` Lu Baolu
2020-07-15 16:01       ` Jacob Pan
2020-07-16  1:07         ` Lu Baolu
2020-07-29 20:03           ` Alex Williamson
2020-07-29 23:34             ` Tian, Kevin
2020-07-30 19:46               ` Alex Williamson
2020-07-31  5:47                 ` Lu Baolu
2020-07-31 18:05                   ` Alex Williamson
2020-08-03  1:57                     ` Lu Baolu [this message]
2020-07-14  5:57 ` [PATCH v3 3/4] iommu: Add iommu_aux_get_domain_for_dev() Lu Baolu
2020-07-29 20:25   ` Alex Williamson
2020-07-29 23:49     ` Tian, Kevin
2020-07-30 20:17       ` Alex Williamson
2020-07-31  0:26         ` Tian, Kevin
2020-07-31  2:17         ` Tian, Kevin
2020-07-31  6:30     ` Lu Baolu
2020-07-31 18:14       ` Alex Williamson
2020-08-03  2:15         ` Lu Baolu
2020-07-14  5:57 ` [PATCH v3 4/4] vfio/type1: Use iommu_aux_at(de)tach_group() APIs Lu Baolu
2020-07-14  8:25   ` Christoph Hellwig
2020-07-14 16:29     ` Jacob Pan
2020-07-15  1:00       ` Lu Baolu
2020-07-15  1:23         ` Tian, Kevin
2020-07-29 20:32   ` Alex Williamson
2020-07-30  2:41     ` Lu Baolu
2020-07-30 21:17       ` Alex Williamson
2020-07-31  1:37         ` Lu Baolu
2020-07-30  9:36   ` Liu, Yi L
2020-07-31  1:39     ` Lu Baolu
2020-07-23 13:55 ` [PATCH v3 0/4] iommu aux-domain APIs extensions Lu Baolu

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=9ee76708-f50c-adb1-0b3f-67871c799cab@linux.intel.com \
    --to=baolu.lu@linux.intel.com \
    --cc=alex.williamson@redhat.com \
    --cc=ashok.raj@intel.com \
    --cc=cohuck@redhat.com \
    --cc=dave.jiang@intel.com \
    --cc=iommu@lists.linux-foundation.org \
    --cc=jacob.jun.pan@linux.intel.com \
    --cc=jean-philippe@linaro.org \
    --cc=joro@8bytes.org \
    --cc=kevin.tian@intel.com \
    --cc=kvm@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=robin.murphy@arm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

KVM Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/kvm/0 kvm/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 kvm kvm/ https://lore.kernel.org/kvm \
		kvm@vger.kernel.org
	public-inbox-index kvm

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.kvm


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git