[RFC PATCH 0/5] iommu: APIs for paravirtual PASID allocation

[RFC PATCH 0/5] iommu: APIs for paravirtual PASID allocation

[Lu Baolu]

This adds an uniformed API set for global PASIDs used by IOMMU
and device drivers which depend on IOMMU. It works for drivers
running on bare metal, full virtualized environments and para-
virtualized environment.

When PASID requests come from components running on the bare
metal hardware, the requests will be routed to the system wide
I/O ASID allocator.


            ^   ^  iommu_pasid_init() 
            |   |  iommu_pasid_exit() 
            |   |  iommu_pasid_alloc()
            |   |  iommu_pasid_free() 
      .-------------------------------.
      |       IOMMU PASID APIs        |
      '-------------------------------'
           |   ^
           v   |
      .------------.  .---------------.
      | I/O ASID   |  |               |
      | allocator  |  | IOMMU drivers |
      | (drivers/  |  |      or       |
      | base/      |  | virtio IOMMU  |
      | ioasid.c)  |  |               |
      '------------'  '---------------'

When PASID requests come from components running in full-virtualized
or para-virtualized environments, they will be routed to the vendor-
specific or virtio IOMMU driver, where the requests will be intercepted
and routed to the Host Linux via VFIO interfaces.

    ......................................
    .            Guest Linux             .
    ......................................
    .                                    .
    . .----------------------------.     .
    . |       IOMMU PASID APIs     |     .
    . '----------------------------'     .
    .                           |        .
    .                           v        .
    .  .----------. .---------------.    ..---------.
    .  | I/O ASID | | IOMMU drivers |    .|  QEMU   |
    .  '----------' | /virtio IOMMU |---->'---------'
    .               '---------------'    .| virtio  |
    ......................................'---------'
                                               |
                                               |
    ...........................................|....
    .                  Host Linux              |   .
    ...........................................|....
    .                      |-------------------'   .
    .                      v                       .
    .         .----------------------------.       .
    .         |       IOMMU PASID APIs     |       .
    .         '----------------------------'       .
    .               |                              .
    .               v       .---------------.      .
    .         .----------.  | IOMMU drivers |      .
    .         | I/O ASID |  | /virtio IOMMU |      .
    .         '----------'  '---------------'      .
    ................................................

Below APIs are introduced:
    
* iommu_pasid_init(pasid)
  - Initialize a PASID consumer. The vendor specific IOMMU
    drivers are able to set the PASID range imposed by IOMMU
    hardware through a callback in iommu_ops.

* iommu_pasid_exit(pasid)
  - The PASID consumer stops consuming any PASID.

* iommu_pasid_alloc(pasid, min, max, private, *ioasid)
  - Allocate a PASID and associate a @private data with this
    PASID. The PASID value is stored in @ioaisd if returning
    success.

* iommu_pasid_free(pasid, ioasid)
  - Free a PASID to the pool so that it could be consumed by
    others.

It also adds below helpers to lookup or iterate PASID items.

* iommu_pasid_for_each(pasid, func, data)
  - Iterate PASID items of the consumer identified by @pasid,
    and call @func() against each item. An error returned from
    @func() will break the iteration.

* iommu_pasid_find(pasid, ioasid)
  - Retrieve the private data associated with @ioasid.

This patch set depends on the I/O APSID allocator posted here
[1] for discussion.

[1] https://www.spinics.net/lists/iommu/msg30639.html

Best regards,
Lu Baolu

Lu Baolu (5):
  iommu: Add APIs for IOMMU PASID management
  iommu/vt-d: Initialize a PASID consumer
  iommu/vt-d: Enlightened PASID allocation
  iommu/vt-d: Allocate and free a pasid
  iommu/vt-d: Use global pasid allocator

 drivers/iommu/Kconfig       |  1 +
 drivers/iommu/intel-iommu.c | 83 +++++++++++++++++++++++++++++++---
 drivers/iommu/intel-pasid.c | 88 +++++++++++++++++++++++++-----------
 drivers/iommu/intel-pasid.h | 16 +++++--
 drivers/iommu/intel-svm.c   | 24 +++++-----
 drivers/iommu/iommu.c       | 89 +++++++++++++++++++++++++++++++++++++
 include/linux/intel-iommu.h |  5 +++
 include/linux/iommu.h       | 73 ++++++++++++++++++++++++++++++
 8 files changed, 332 insertions(+), 47 deletions(-)

-- 
2.17.1

[RFC PATCH 1/5] iommu: Add APIs for IOMMU PASID management

[Lu Baolu]

This adds APIs for IOMMU drivers and device drivers to manage
the PASIDs used for DMA transfer and translation. It bases on
I/O ASID allocator for PASID namespace management and relies
on vendor specific IOMMU drivers for paravirtual PASIDs.

Below APIs are added:

* iommu_pasid_init(pasid)
  - Initialize a PASID consumer. The vendor specific IOMMU
    drivers are able to set the PASID range imposed by IOMMU
    hardware through a callback in iommu_ops.

* iommu_pasid_exit(pasid)
  - The PASID consumer stops consuming any PASID.

* iommu_pasid_alloc(pasid, min, max, private, *ioasid)
  - Allocate a PASID and associate a @private data with this
    PASID. The PASID value is stored in @ioaisd if returning
    success.

* iommu_pasid_free(pasid, ioasid)
  - Free a PASID to the pool so that it could be consumed by
    others.

This also adds below helpers to lookup or iterate PASID items
associated with a consumer.

* iommu_pasid_for_each(pasid, func, data)
  - Iterate PASID items of the consumer identified by @pasid,
    and call @func() against each item. An error returned from
    @func() will break the iteration.

* iommu_pasid_find(pasid, ioasid)
  - Retrieve the private data associated with @ioasid.

Cc: Ashok Raj <ashok.raj@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Kevin Tian <kevin.tian@intel.com>
Cc: Jean-Philippe Brucker <jean-philippe.brucker@arm.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
---
 drivers/iommu/Kconfig |  1 +
 drivers/iommu/iommu.c | 89 +++++++++++++++++++++++++++++++++++++++++++
 include/linux/iommu.h | 73 +++++++++++++++++++++++++++++++++++
 3 files changed, 163 insertions(+)

diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
index d9a25715650e..39f2bb76c7b8 100644
--- a/drivers/iommu/Kconfig
+++ b/drivers/iommu/Kconfig
@@ -1,6 +1,7 @@
 # IOMMU_API always gets selected by whoever wants it.
 config IOMMU_API
 	bool
+	select IOASID
 
 menuconfig IOMMU_SUPPORT
 	bool "IOMMU Hardware Support"
diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
index 0b7c96d1425e..570b244897bb 100644
--- a/drivers/iommu/iommu.c
+++ b/drivers/iommu/iommu.c
@@ -2082,3 +2082,92 @@ void iommu_detach_device_aux(struct iommu_domain *domain, struct device *dev)
 	}
 }
 EXPORT_SYMBOL_GPL(iommu_detach_device_aux);
+
+/*
+ * APIs for PASID used by IOMMU and the device drivers which depend
+ * on IOMMU.
+ */
+struct iommu_pasid *iommu_pasid_init(struct bus_type *bus)
+{
+	struct iommu_pasid *pasid;
+	int ret;
+
+	if (!bus || !bus->iommu_ops)
+		return NULL;
+
+	pasid = kzalloc(sizeof(*pasid), GFP_KERNEL);
+	if (!pasid)
+		return NULL;
+
+	pasid->ops = bus->iommu_ops;
+	/*
+	 * The default range of an IOMMU PASID is from 0 to the full
+	 * 20bit integer.
+	 */
+	pasid->min = 0;
+	pasid->max = 0x100000;
+	/*
+	 * Give vendor specific iommu drivers a chance to set the pasid
+	 * limits imposed by the iommu hardware.
+	 */
+	if (bus->iommu_ops->pasid_init) {
+		ret = bus->iommu_ops->pasid_init(pasid);
+		if (ret) {
+			kfree(pasid);
+			return NULL;
+		}
+	}
+
+	return pasid;
+}
+EXPORT_SYMBOL_GPL(iommu_pasid_init);
+
+void iommu_pasid_exit(struct iommu_pasid *pasid)
+{
+	kfree(pasid);
+}
+EXPORT_SYMBOL_GPL(iommu_pasid_exit);
+
+int iommu_pasid_alloc(struct iommu_pasid *pasid, ioasid_t min,
+		      ioasid_t max, void *private, ioasid_t *ioasid)
+{
+	ioasid_t start, end, hw, val;
+	int ret = -EAGAIN;
+
+	start = max_t(int, min, pasid->min);
+	end = min_t(int, max, pasid->max);
+
+	if (pasid->ops->pasid_alloc)
+		ret = pasid->ops->pasid_alloc(pasid, start, end, &hw);
+
+	if (ret == -EAGAIN)
+                val = ioasid_alloc(&pasid->set, start, end, private);
+	else if (ret == 0)
+		val = ioasid_alloc(&pasid->set, hw, hw + 1, private);
+	else
+		goto hw_ret;
+
+	if (val == INVALID_IOASID)
+		goto ioasid_ret;
+
+	*ioasid = val;
+
+        return 0;
+
+ioasid_ret:
+	if (pasid->ops->pasid_free)
+		pasid->ops->pasid_free(pasid, hw);
+
+hw_ret:
+	return -ENODEV;
+}
+EXPORT_SYMBOL_GPL(iommu_pasid_alloc);
+
+void iommu_pasid_free(struct iommu_pasid *pasid, ioasid_t ioasid)
+{
+	if (pasid->ops->pasid_free)
+		pasid->ops->pasid_free(pasid, ioasid);
+
+	ioasid_free(ioasid);
+}
+EXPORT_SYMBOL_GPL(iommu_pasid_free);
diff --git a/include/linux/iommu.h b/include/linux/iommu.h
index 9bf1b3f2457a..4f5202c8170b 100644
--- a/include/linux/iommu.h
+++ b/include/linux/iommu.h
@@ -20,6 +20,7 @@
 #define __LINUX_IOMMU_H
 
 #include <linux/scatterlist.h>
+#include <linux/ioasid.h>
 #include <linux/device.h>
 #include <linux/types.h>
 #include <linux/errno.h>
@@ -48,6 +49,7 @@ struct bus_type;
 struct device;
 struct iommu_domain;
 struct notifier_block;
+struct iommu_pasid;
 
 /* iommu fault flags */
 #define IOMMU_FAULT_READ	0x0
@@ -194,6 +196,9 @@ enum iommu_dev_attr {
  * @of_xlate: add OF master IDs to iommu grouping
  * @get_dev_attr: get per device IOMMU attributions
  * @set_dev_attr: set per device IOMMU attributions
+ * @pasid_init: initialize a pasid consumer
+ * @pasid_alloc: allocate a pasid from low level driver
+ * @pasid_free: free a pasid to low level driver
  * @pgsize_bitmap: bitmap of all possible supported page sizes
  */
 struct iommu_ops {
@@ -246,6 +251,12 @@ struct iommu_ops {
 	int (*attach_dev_aux)(struct iommu_domain *domain, struct device *dev);
 	void (*detach_dev_aux)(struct iommu_domain *domain, struct device *dev);
 
+	/* IOMMU pasid callbacks */
+	int (*pasid_init)(struct iommu_pasid *pasid);
+	int (*pasid_alloc)(struct iommu_pasid *pasid, ioasid_t start,
+			   ioasid_t end, ioasid_t *ioasid);
+	void (*pasid_free)(struct iommu_pasid *pasid, ioasid_t ioasid);
+
 	unsigned long pgsize_bitmap;
 };
 
@@ -428,12 +439,41 @@ extern int iommu_attach_device_aux(struct iommu_domain *domain,
 extern void iommu_detach_device_aux(struct iommu_domain *domain,
 				    struct device *dev);
 
+/*
+ * Per IOMMU PASID consumer data.
+ */
+struct iommu_pasid {
+	ioasid_t		max;
+	ioasid_t		min;
+	struct ioasid_set	set;
+	const struct iommu_ops	*ops;
+
+	/* vendor specific iommu private data */
+	void			*priv;
+};
+
+struct iommu_pasid *iommu_pasid_init(struct bus_type *bus);
+void iommu_pasid_exit(struct iommu_pasid *pasid);
+int iommu_pasid_alloc(struct iommu_pasid *pasid, ioasid_t min,
+		      ioasid_t max, void *private, ioasid_t *ioasid);
+void iommu_pasid_free(struct iommu_pasid *pasid, ioasid_t iosid);
+static inline int
+iommu_pasid_for_each(struct iommu_pasid *pasid, ioasid_iter_t func, void *data)
+{
+	return ioasid_for_each(&pasid->set, func, data);
+}
+static inline void*
+iommu_pasid_find(struct iommu_pasid *pasid, ioasid_t ioasid)
+{
+	return ioasid_find(&pasid->set, ioasid);
+}
 #else /* CONFIG_IOMMU_API */
 
 struct iommu_ops {};
 struct iommu_group {};
 struct iommu_fwspec {};
 struct iommu_device {};
+struct iommu_pasid {};
 
 static inline bool iommu_present(struct bus_type *bus)
 {
@@ -734,6 +774,39 @@ static inline void
 iommu_detach_device_aux(struct iommu_domain *domain, struct device *dev)
 {
 }
+
+static inline struct iommu_pasid *
+iommu_pasid_init(struct bus_type *bus)
+{
+	return NULL;
+}
+
+static inline void iommu_pasid_exit(struct iommu_pasid *pasid)
+{
+}
+
+static inline int
+iommu_pasid_alloc(struct iommu_pasid *pasid, ioasid_t min,
+		  ioasid_t max, void *private, ioasid_t *ioasid)
+{
+	return -ENODEV;
+}
+
+static inline void iommu_pasid_free(struct iommu_pasid *pasid, ioasid_t iosid)
+{
+}
+
+static inline int
+iommu_pasid_for_each(struct iommu_pasid *pasid, ioasid_iter_t func, void *data)
+{
+	return -ENODEV;
+}
+
+static inline void*
+iommu_pasid_find(struct iommu_pasid *pasid, ioasid_t ioasid)
+{
+	return NULL;
+}
 #endif /* CONFIG_IOMMU_API */
 
 #ifdef CONFIG_IOMMU_DEBUGFS
-- 
2.17.1

RE: [RFC PATCH 1/5] iommu: Add APIs for IOMMU PASID management

[Liu, Yi L]

> From: Lu Baolu [mailto:baolu.lu@linux.intel.com]
> Sent: Sunday, November 11, 2018 10:45 PM
> Subject: [RFC PATCH 1/5] iommu: Add APIs for IOMMU PASID management
> 
> This adds APIs for IOMMU drivers and device drivers to manage the PASIDs used for
> DMA transfer and translation. It bases on I/O ASID allocator for PASID namespace
> management and relies on vendor specific IOMMU drivers for paravirtual PASIDs.
> 
> Below APIs are added:
> 
> * iommu_pasid_init(pasid)
>   - Initialize a PASID consumer. The vendor specific IOMMU
>     drivers are able to set the PASID range imposed by IOMMU
>     hardware through a callback in iommu_ops.
> 
> * iommu_pasid_exit(pasid)
>   - The PASID consumer stops consuming any PASID.
> 
> * iommu_pasid_alloc(pasid, min, max, private, *ioasid)
>   - Allocate a PASID and associate a @private data with this
>     PASID. The PASID value is stored in @ioaisd if returning
>     success.
> 
> * iommu_pasid_free(pasid, ioasid)
>   - Free a PASID to the pool so that it could be consumed by
>     others.
> 
> This also adds below helpers to lookup or iterate PASID items associated with a
> consumer.
> 
> * iommu_pasid_for_each(pasid, func, data)
>   - Iterate PASID items of the consumer identified by @pasid,
>     and call @func() against each item. An error returned from
>     @func() will break the iteration.
> 
> * iommu_pasid_find(pasid, ioasid)
>   - Retrieve the private data associated with @ioasid.
> 
> Cc: Ashok Raj <ashok.raj@intel.com>
> Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
> Cc: Kevin Tian <kevin.tian@intel.com>
> Cc: Jean-Philippe Brucker <jean-philippe.brucker@arm.com>
> Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
> ---
>  drivers/iommu/Kconfig |  1 +
>  drivers/iommu/iommu.c | 89 +++++++++++++++++++++++++++++++++++++++++++
>  include/linux/iommu.h | 73 +++++++++++++++++++++++++++++++++++
>  3 files changed, 163 insertions(+)
> 
> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig index
> d9a25715650e..39f2bb76c7b8 100644
> --- a/drivers/iommu/Kconfig
> +++ b/drivers/iommu/Kconfig
> @@ -1,6 +1,7 @@
>  # IOMMU_API always gets selected by whoever wants it.
>  config IOMMU_API
>  	bool
> +	select IOASID
> 
>  menuconfig IOMMU_SUPPORT
>  	bool "IOMMU Hardware Support"
> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c index
> 0b7c96d1425e..570b244897bb 100644
> --- a/drivers/iommu/iommu.c
> +++ b/drivers/iommu/iommu.c
> @@ -2082,3 +2082,92 @@ void iommu_detach_device_aux(struct iommu_domain
> *domain, struct device *dev)
>  	}
>  }
>  EXPORT_SYMBOL_GPL(iommu_detach_device_aux);
> +
> +/*
> + * APIs for PASID used by IOMMU and the device drivers which depend
> + * on IOMMU.
> + */
> +struct iommu_pasid *iommu_pasid_init(struct bus_type *bus) {

I'm thinking about if using struct iommu_domain here is better
than struct bus_type. The major purpose is to pass iommu_ops
in it and route into iommu-sublayer. iommu_domain may be
better since some modules like vfio_iommu_type1 would use
iommu_domain more than bus type.

Thanks,
Yi Liu

Re: [RFC PATCH 1/5] iommu: Add APIs for IOMMU PASID management

[Lu Baolu]

Hi,

On 12/16/18 6:38 AM, Liu, Yi L wrote:
>> From: Lu Baolu [mailto:baolu.lu@linux.intel.com]
>> Sent: Sunday, November 11, 2018 10:45 PM
>> Subject: [RFC PATCH 1/5] iommu: Add APIs for IOMMU PASID management
>>
>> This adds APIs for IOMMU drivers and device drivers to manage the PASIDs used for
>> DMA transfer and translation. It bases on I/O ASID allocator for PASID namespace
>> management and relies on vendor specific IOMMU drivers for paravirtual PASIDs.
>>
>> Below APIs are added:
>>
>> * iommu_pasid_init(pasid)
>>    - Initialize a PASID consumer. The vendor specific IOMMU
>>      drivers are able to set the PASID range imposed by IOMMU
>>      hardware through a callback in iommu_ops.
>>
>> * iommu_pasid_exit(pasid)
>>    - The PASID consumer stops consuming any PASID.
>>
>> * iommu_pasid_alloc(pasid, min, max, private, *ioasid)
>>    - Allocate a PASID and associate a @private data with this
>>      PASID. The PASID value is stored in @ioaisd if returning
>>      success.
>>
>> * iommu_pasid_free(pasid, ioasid)
>>    - Free a PASID to the pool so that it could be consumed by
>>      others.
>>
>> This also adds below helpers to lookup or iterate PASID items associated with a
>> consumer.
>>
>> * iommu_pasid_for_each(pasid, func, data)
>>    - Iterate PASID items of the consumer identified by @pasid,
>>      and call @func() against each item. An error returned from
>>      @func() will break the iteration.
>>
>> * iommu_pasid_find(pasid, ioasid)
>>    - Retrieve the private data associated with @ioasid.
>>
>> Cc: Ashok Raj <ashok.raj@intel.com>
>> Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
>> Cc: Kevin Tian <kevin.tian@intel.com>
>> Cc: Jean-Philippe Brucker <jean-philippe.brucker@arm.com>
>> Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
>> ---
>>   drivers/iommu/Kconfig |  1 +
>>   drivers/iommu/iommu.c | 89 +++++++++++++++++++++++++++++++++++++++++++
>>   include/linux/iommu.h | 73 +++++++++++++++++++++++++++++++++++
>>   3 files changed, 163 insertions(+)
>>
>> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig index
>> d9a25715650e..39f2bb76c7b8 100644
>> --- a/drivers/iommu/Kconfig
>> +++ b/drivers/iommu/Kconfig
>> @@ -1,6 +1,7 @@
>>   # IOMMU_API always gets selected by whoever wants it.
>>   config IOMMU_API
>>   	bool
>> +	select IOASID
>>
>>   menuconfig IOMMU_SUPPORT
>>   	bool "IOMMU Hardware Support"
>> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c index
>> 0b7c96d1425e..570b244897bb 100644
>> --- a/drivers/iommu/iommu.c
>> +++ b/drivers/iommu/iommu.c
>> @@ -2082,3 +2082,92 @@ void iommu_detach_device_aux(struct iommu_domain
>> *domain, struct device *dev)
>>   	}
>>   }
>>   EXPORT_SYMBOL_GPL(iommu_detach_device_aux);
>> +
>> +/*
>> + * APIs for PASID used by IOMMU and the device drivers which depend
>> + * on IOMMU.
>> + */
>> +struct iommu_pasid *iommu_pasid_init(struct bus_type *bus) {
> 
> I'm thinking about if using struct iommu_domain here is better
> than struct bus_type. The major purpose is to pass iommu_ops
> in it and route into iommu-sublayer. iommu_domain may be
> better since some modules like vfio_iommu_type1 would use
> iommu_domain more than bus type.

But drivers might call this during initialization when it doesn't has
any domain yet.

Best regards,
Lu Baolu

Re: [RFC PATCH 1/5] iommu: Add APIs for IOMMU PASID management

[Jacob Pan]

On Mon, 12 Nov 2018 14:44:57 +0800
Lu Baolu <baolu.lu@linux.intel.com> wrote:

> This adds APIs for IOMMU drivers and device drivers to manage
> the PASIDs used for DMA transfer and translation. It bases on
> I/O ASID allocator for PASID namespace management and relies
> on vendor specific IOMMU drivers for paravirtual PASIDs.
> 
Trying to integrate this with VT-d SVM code had some thoughts.
First of all, it seems to me the problem we are trying to solve here is
to allow custom PASID/IOASID allocator.
IOASID, as in driver base, is a common infrastructure of its own right.
So it is OK to let device drivers such as VT-d driver directly
communicate with IOASID w/o going through common IOMMU layer. Therefore
I see little value of adding this wrapper to ioasid code, instead I
feel it might be less work to enhance ioasid with the following:

1. allow ioasid code to register a system wide custom asid allocator,
which takes precedence over the idr allocator
e.g.
typedef ioasid_t (*ioasid_alloc_fn_t)(ioasid_t min, ioasid_t max, void *data);
void ioasid_set_allocator(ioasid_alloc_fn_t fn)
{}
We can still use idr for tracking ioasid and private data lookup, since
the ioasid idr is exclusively owned by ioasid_alloc, we can rest and be
sure there is no conflict with other callers. See code comments below.

2. support setting asid private data _after_ asid is allocated. The use
case is that PASID allocation and mm bind can be split into separate
stages. Private data (mm related) are not available at the time of
allocation, only bind time.
Since IDR needs the data pointer at allocation time, we can still
create an empty ioasid_data for ioasid tracking at alloc time. i.e.
struct ioasid_data {
	void *ptr; /* private data */
};


3. allow NULL ioasid_set
I also had a hard time understanding the use of ioasid_set, since there
is already a private data associated with each ioasid, is it not enough
to group ioasid based on private data?

So the usage scenarios can be.
1. during boot, vt-d driver register a custom ioasid allocator (vcmd) if
it detects its running as a guest.

2. Running as a guest, all pasid allocation via ioasid_alloc() will go
to this custom allocators and tracked

3. for native case, there is no custom allocator, ioasid just use IDR
alloc

4. for native bind mm, pasid allocation already has private data filled
in when calling ioasid_alloc

5. for guest bind, pasid is allocated with empty private data (called
by VFIO layer), but private data is filled in later by bind guest
pasid inside the vt-d driver.

So my thinking is that we can avoid having another layer of APIs as
below and keep everything within ioasid. Also allows private data to be
managed at lower level as compared to VFIO.


Thanks,

Jacob

> Below APIs are added:
> 
> * iommu_pasid_init(pasid)
>   - Initialize a PASID consumer. The vendor specific IOMMU
>     drivers are able to set the PASID range imposed by IOMMU
>     hardware through a callback in iommu_ops.
> 
> * iommu_pasid_exit(pasid)
>   - The PASID consumer stops consuming any PASID.
> 
> * iommu_pasid_alloc(pasid, min, max, private, *ioasid)
>   - Allocate a PASID and associate a @private data with this
>     PASID. The PASID value is stored in @ioaisd if returning
>     success.
> 
> * iommu_pasid_free(pasid, ioasid)
>   - Free a PASID to the pool so that it could be consumed by
>     others.
> 
> This also adds below helpers to lookup or iterate PASID items
> associated with a consumer.
> 
> * iommu_pasid_for_each(pasid, func, data)
>   - Iterate PASID items of the consumer identified by @pasid,
>     and call @func() against each item. An error returned from
>     @func() will break the iteration.
> 
> * iommu_pasid_find(pasid, ioasid)
>   - Retrieve the private data associated with @ioasid.
> 
> Cc: Ashok Raj <ashok.raj@intel.com>
> Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
> Cc: Kevin Tian <kevin.tian@intel.com>
> Cc: Jean-Philippe Brucker <jean-philippe.brucker@arm.com>
> Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
> ---
>  drivers/iommu/Kconfig |  1 +
>  drivers/iommu/iommu.c | 89
> +++++++++++++++++++++++++++++++++++++++++++ include/linux/iommu.h |
> 73 +++++++++++++++++++++++++++++++++++ 3 files changed, 163
> insertions(+)
> 
> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
> index d9a25715650e..39f2bb76c7b8 100644
> --- a/drivers/iommu/Kconfig
> +++ b/drivers/iommu/Kconfig
> @@ -1,6 +1,7 @@
>  # IOMMU_API always gets selected by whoever wants it.
>  config IOMMU_API
>  	bool
> +	select IOASID
>  
>  menuconfig IOMMU_SUPPORT
>  	bool "IOMMU Hardware Support"
> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
> index 0b7c96d1425e..570b244897bb 100644
> --- a/drivers/iommu/iommu.c
> +++ b/drivers/iommu/iommu.c
> @@ -2082,3 +2082,92 @@ void iommu_detach_device_aux(struct
> iommu_domain *domain, struct device *dev) }
>  }
>  EXPORT_SYMBOL_GPL(iommu_detach_device_aux);
> +
> +/*
> + * APIs for PASID used by IOMMU and the device drivers which depend
> + * on IOMMU.
> + */
> +struct iommu_pasid *iommu_pasid_init(struct bus_type *bus)
> +{
> +	struct iommu_pasid *pasid;
> +	int ret;
> +
> +	if (!bus || !bus->iommu_ops)
> +		return NULL;
> +
> +	pasid = kzalloc(sizeof(*pasid), GFP_KERNEL);
> +	if (!pasid)
> +		return NULL;
> +
> +	pasid->ops = bus->iommu_ops;
> +	/*
> +	 * The default range of an IOMMU PASID is from 0 to the full
> +	 * 20bit integer.
> +	 */
> +	pasid->min = 0;
> +	pasid->max = 0x100000;
> +	/*
> +	 * Give vendor specific iommu drivers a chance to set the
> pasid
> +	 * limits imposed by the iommu hardware.
> +	 */
> +	if (bus->iommu_ops->pasid_init) {
> +		ret = bus->iommu_ops->pasid_init(pasid);
> +		if (ret) {
> +			kfree(pasid);
> +			return NULL;
> +		}
> +	}
> +
> +	return pasid;
> +}
> +EXPORT_SYMBOL_GPL(iommu_pasid_init);
> +
> +void iommu_pasid_exit(struct iommu_pasid *pasid)
> +{
> +	kfree(pasid);
> +}
> +EXPORT_SYMBOL_GPL(iommu_pasid_exit);
> +
> +int iommu_pasid_alloc(struct iommu_pasid *pasid, ioasid_t min,
> +		      ioasid_t max, void *private, ioasid_t *ioasid)
> +{
> +	ioasid_t start, end, hw, val;
> +	int ret = -EAGAIN;
> +
> +	start = max_t(int, min, pasid->min);
> +	end = min_t(int, max, pasid->max);
> +
> +	if (pasid->ops->pasid_alloc)
> +		ret = pasid->ops->pasid_alloc(pasid, start, end,
> &hw); +
> +	if (ret == -EAGAIN)
> +                val = ioasid_alloc(&pasid->set, start, end, private);
> +	else if (ret == 0)
> +		val = ioasid_alloc(&pasid->set, hw, hw + 1, private);
this may fail due to conflict with other callers of ioasic_alloc(), but
we should really respect the iommu ops allocator. If we move the
pasid_alloc op into ioasid code, then we don't need to worry about the
conflict.
> +	else
> +		goto hw_ret;
> +
> +	if (val == INVALID_IOASID)
> +		goto ioasid_ret;
> +
> +	*ioasid = val;
> +
> +        return 0;
> +
> +ioasid_ret:
> +	if (pasid->ops->pasid_free)
> +		pasid->ops->pasid_free(pasid, hw);
> +
> +hw_ret:
> +	return -ENODEV;
> +}
> +EXPORT_SYMBOL_GPL(iommu_pasid_alloc);
> +
> +void iommu_pasid_free(struct iommu_pasid *pasid, ioasid_t ioasid)
> +{
> +	if (pasid->ops->pasid_free)
> +		pasid->ops->pasid_free(pasid, ioasid);
> +
> +	ioasid_free(ioasid);
> +}
> +EXPORT_SYMBOL_GPL(iommu_pasid_free);
> diff --git a/include/linux/iommu.h b/include/linux/iommu.h
> index 9bf1b3f2457a..4f5202c8170b 100644
> --- a/include/linux/iommu.h
> +++ b/include/linux/iommu.h
> @@ -20,6 +20,7 @@
>  #define __LINUX_IOMMU_H
>  
>  #include <linux/scatterlist.h>
> +#include <linux/ioasid.h>
>  #include <linux/device.h>
>  #include <linux/types.h>
>  #include <linux/errno.h>
> @@ -48,6 +49,7 @@ struct bus_type;
>  struct device;
>  struct iommu_domain;
>  struct notifier_block;
> +struct iommu_pasid;
>  
>  /* iommu fault flags */
>  #define IOMMU_FAULT_READ	0x0
> @@ -194,6 +196,9 @@ enum iommu_dev_attr {
>   * @of_xlate: add OF master IDs to iommu grouping
>   * @get_dev_attr: get per device IOMMU attributions
>   * @set_dev_attr: set per device IOMMU attributions
> + * @pasid_init: initialize a pasid consumer
> + * @pasid_alloc: allocate a pasid from low level driver
> + * @pasid_free: free a pasid to low level driver
>   * @pgsize_bitmap: bitmap of all possible supported page sizes
>   */
>  struct iommu_ops {
> @@ -246,6 +251,12 @@ struct iommu_ops {
>  	int (*attach_dev_aux)(struct iommu_domain *domain, struct
> device *dev); void (*detach_dev_aux)(struct iommu_domain *domain,
> struct device *dev); 
> +	/* IOMMU pasid callbacks */
> +	int (*pasid_init)(struct iommu_pasid *pasid);
> +	int (*pasid_alloc)(struct iommu_pasid *pasid, ioasid_t start,
> +			   ioasid_t end, ioasid_t *ioasid);
> +	void (*pasid_free)(struct iommu_pasid *pasid, ioasid_t
> ioasid); +
>  	unsigned long pgsize_bitmap;
>  };
>  
> @@ -428,12 +439,41 @@ extern int iommu_attach_device_aux(struct
> iommu_domain *domain, extern void iommu_detach_device_aux(struct
> iommu_domain *domain, struct device *dev);
>  
> +/*
> + * Per IOMMU PASID consumer data.
> + */
> +struct iommu_pasid {
> +	ioasid_t		max;
> +	ioasid_t		min;
> +	struct ioasid_set	set;
> +	const struct iommu_ops	*ops;
> +
> +	/* vendor specific iommu private data */
> +	void			*priv;
> +};
> +
> +struct iommu_pasid *iommu_pasid_init(struct bus_type *bus);
> +void iommu_pasid_exit(struct iommu_pasid *pasid);
> +int iommu_pasid_alloc(struct iommu_pasid *pasid, ioasid_t min,
> +		      ioasid_t max, void *private, ioasid_t *ioasid);
> +void iommu_pasid_free(struct iommu_pasid *pasid, ioasid_t iosid);
> +static inline int
> +iommu_pasid_for_each(struct iommu_pasid *pasid, ioasid_iter_t func,
> void *data) +{
> +	return ioasid_for_each(&pasid->set, func, data);
> +}
> +static inline void*
> +iommu_pasid_find(struct iommu_pasid *pasid, ioasid_t ioasid)
> +{
> +	return ioasid_find(&pasid->set, ioasid);
> +}
>  #else /* CONFIG_IOMMU_API */
>  
>  struct iommu_ops {};
>  struct iommu_group {};
>  struct iommu_fwspec {};
>  struct iommu_device {};
> +struct iommu_pasid {};
>  
>  static inline bool iommu_present(struct bus_type *bus)
>  {
> @@ -734,6 +774,39 @@ static inline void
>  iommu_detach_device_aux(struct iommu_domain *domain, struct device
> *dev) {
>  }
> +
> +static inline struct iommu_pasid *
> +iommu_pasid_init(struct bus_type *bus)
> +{
> +	return NULL;
> +}
> +
> +static inline void iommu_pasid_exit(struct iommu_pasid *pasid)
> +{
> +}
> +
> +static inline int
> +iommu_pasid_alloc(struct iommu_pasid *pasid, ioasid_t min,
> +		  ioasid_t max, void *private, ioasid_t *ioasid)
> +{
> +	return -ENODEV;
> +}
> +
> +static inline void iommu_pasid_free(struct iommu_pasid *pasid,
> ioasid_t iosid) +{
> +}
> +
> +static inline int
> +iommu_pasid_for_each(struct iommu_pasid *pasid, ioasid_iter_t func,
> void *data) +{
> +	return -ENODEV;
> +}
> +
> +static inline void*
> +iommu_pasid_find(struct iommu_pasid *pasid, ioasid_t ioasid)
> +{
> +	return NULL;
> +}
>  #endif /* CONFIG_IOMMU_API */
>  
>  #ifdef CONFIG_IOMMU_DEBUGFS

[Jacob Pan]

Re: [RFC PATCH 1/5] iommu: Add APIs for IOMMU PASID management

[Jean-Philippe Brucker]

Hi Jacob, Lu,

On 30/01/2019 19:05, Jacob Pan wrote:
> On Mon, 12 Nov 2018 14:44:57 +0800
> Lu Baolu <baolu.lu@linux.intel.com> wrote:
> 
>> This adds APIs for IOMMU drivers and device drivers to manage
>> the PASIDs used for DMA transfer and translation. It bases on
>> I/O ASID allocator for PASID namespace management and relies
>> on vendor specific IOMMU drivers for paravirtual PASIDs.
>>
> Trying to integrate this with VT-d SVM code had some thoughts.
> First of all, it seems to me the problem we are trying to solve here is
> to allow custom PASID/IOASID allocator.
> IOASID, as in driver base, is a common infrastructure of its own right.
> So it is OK to let device drivers such as VT-d driver directly
> communicate with IOASID w/o going through common IOMMU layer. Therefore
> I see little value of adding this wrapper to ioasid code, instead I
> feel it might be less work to enhance ioasid with the following:
> 
> 1. allow ioasid code to register a system wide custom asid allocator,
> which takes precedence over the idr allocator
> e.g.
> typedef ioasid_t (*ioasid_alloc_fn_t)(ioasid_t min, ioasid_t max, void *data);
> void ioasid_set_allocator(ioasid_alloc_fn_t fn)
> {}
> We can still use idr for tracking ioasid and private data lookup, since
> the ioasid idr is exclusively owned by ioasid_alloc, we can rest and be
> sure there is no conflict with other callers. See code comments below.

Yes, I think we need the alloc function to be registered at boot time.
Because even when using a PV allocation instead of idr allocation, we do
need a way to quickly retrieve PASID->mm when handling an I/O page
fault. In the ioasid_alloc() function, we could do, roughly:

static ioasid_t (*ioasid_alloc_fn)(...);

ioasid_t ioasid_alloc(... min, max...)
{
	if (ioasid_alloc_fn) {
	   ioasid = ioasid_alloc_fn(min, max);

	   /* With PV allocation, only use the IDR for storage */
	   min = ioasid;
	   max = ioasid + 1;
	}

	idr_lock
	ioasid = idr_alloc(... min, max... );
	idr_unlock
}

> 2. support setting asid private data _after_ asid is allocated. The use
> case is that PASID allocation and mm bind can be split into separate
> stages. Private data (mm related) are not available at the time of
> allocation, only bind time.
> Since IDR needs the data pointer at allocation time, we can still
> create an empty ioasid_data for ioasid tracking at alloc time. i.e.
> struct ioasid_data {
> 	void *ptr; /* private data */
> };

I think I agree as well (though finding the right ordering and locking
in SVA allocation is giving me the biggest headache these days). If the
PASID becomes reachable early we need to make sure that concurrent
threads working on the PASID space can deal with incomplete data, but I
think that might be one of the easiest alternatives.

> 3. allow NULL ioasid_set
> I also had a hard time understanding the use of ioasid_set, since there
> is already a private data associated with each ioasid, is it not enough
> to group ioasid based on private data?

The idea behind ioasid_set is that different modules compete for the
single IOASID space, and associate different data types to their IOASIDs.

(a) iommu-sva associates an mm_struct
(b) IOMMU driver allocates a PASIDs for each auxiliary domain, and
probably associates an iommu_domain to it.
(c) device drivers (e.g. AMD KFD) want some PASIDs for their own use in
the same space, and will have their own private data in there.

For (a) we also want the io_pgfault code to find the mm associated with
a PASID:
  iopf_handle_mm_fault()
    iommu_sva_find(pasid)
      ioasid_find(&shared_pasid_set, pasid) -> mm struct

And in this case we really need the search to be only on the
shared_pasid_set, or else the IOPF code gets something that's not an mm
struct.

The easy alternative is to let the IOMMU driver handle both (a) and (b),
and store the same data type for both. I've been considering this a few
times this week, but it doesn't solve case (c).

Thanks,
Jean

> So the usage scenarios can be.
> 1. during boot, vt-d driver register a custom ioasid allocator (vcmd) if
> it detects its running as a guest.
> 
> 2. Running as a guest, all pasid allocation via ioasid_alloc() will go
> to this custom allocators and tracked
> 
> 3. for native case, there is no custom allocator, ioasid just use IDR
> alloc
> 
> 4. for native bind mm, pasid allocation already has private data filled
> in when calling ioasid_alloc
> 
> 5. for guest bind, pasid is allocated with empty private data (called
> by VFIO layer), but private data is filled in later by bind guest
> pasid inside the vt-d driver.
> 
> So my thinking is that we can avoid having another layer of APIs as
> below and keep everything within ioasid. Also allows private data to be
> managed at lower level as compared to VFIO.
> 
> 
> Thanks,
> 
> Jacob
> 
>> Below APIs are added:
>>
>> * iommu_pasid_init(pasid)
>>   - Initialize a PASID consumer. The vendor specific IOMMU
>>     drivers are able to set the PASID range imposed by IOMMU
>>     hardware through a callback in iommu_ops.
>>
>> * iommu_pasid_exit(pasid)
>>   - The PASID consumer stops consuming any PASID.
>>
>> * iommu_pasid_alloc(pasid, min, max, private, *ioasid)
>>   - Allocate a PASID and associate a @private data with this
>>     PASID. The PASID value is stored in @ioaisd if returning
>>     success.
>>
>> * iommu_pasid_free(pasid, ioasid)
>>   - Free a PASID to the pool so that it could be consumed by
>>     others.
>>
>> This also adds below helpers to lookup or iterate PASID items
>> associated with a consumer.
>>
>> * iommu_pasid_for_each(pasid, func, data)
>>   - Iterate PASID items of the consumer identified by @pasid,
>>     and call @func() against each item. An error returned from
>>     @func() will break the iteration.
>>
>> * iommu_pasid_find(pasid, ioasid)
>>   - Retrieve the private data associated with @ioasid.
>>
>> Cc: Ashok Raj <ashok.raj@intel.com>
>> Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
>> Cc: Kevin Tian <kevin.tian@intel.com>
>> Cc: Jean-Philippe Brucker <jean-philippe.brucker@arm.com>
>> Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
>> ---
>>  drivers/iommu/Kconfig |  1 +
>>  drivers/iommu/iommu.c | 89
>> +++++++++++++++++++++++++++++++++++++++++++ include/linux/iommu.h |
>> 73 +++++++++++++++++++++++++++++++++++ 3 files changed, 163
>> insertions(+)
>>
>> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
>> index d9a25715650e..39f2bb76c7b8 100644
>> --- a/drivers/iommu/Kconfig
>> +++ b/drivers/iommu/Kconfig
>> @@ -1,6 +1,7 @@
>>  # IOMMU_API always gets selected by whoever wants it.
>>  config IOMMU_API
>>  	bool
>> +	select IOASID
>>  
>>  menuconfig IOMMU_SUPPORT
>>  	bool "IOMMU Hardware Support"
>> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
>> index 0b7c96d1425e..570b244897bb 100644
>> --- a/drivers/iommu/iommu.c
>> +++ b/drivers/iommu/iommu.c
>> @@ -2082,3 +2082,92 @@ void iommu_detach_device_aux(struct
>> iommu_domain *domain, struct device *dev) }
>>  }
>>  EXPORT_SYMBOL_GPL(iommu_detach_device_aux);
>> +
>> +/*
>> + * APIs for PASID used by IOMMU and the device drivers which depend
>> + * on IOMMU.
>> + */
>> +struct iommu_pasid *iommu_pasid_init(struct bus_type *bus)
>> +{
>> +	struct iommu_pasid *pasid;
>> +	int ret;
>> +
>> +	if (!bus || !bus->iommu_ops)
>> +		return NULL;
>> +
>> +	pasid = kzalloc(sizeof(*pasid), GFP_KERNEL);
>> +	if (!pasid)
>> +		return NULL;
>> +
>> +	pasid->ops = bus->iommu_ops;
>> +	/*
>> +	 * The default range of an IOMMU PASID is from 0 to the full
>> +	 * 20bit integer.
>> +	 */
>> +	pasid->min = 0;
>> +	pasid->max = 0x100000;
>> +	/*
>> +	 * Give vendor specific iommu drivers a chance to set the
>> pasid
>> +	 * limits imposed by the iommu hardware.
>> +	 */
>> +	if (bus->iommu_ops->pasid_init) {
>> +		ret = bus->iommu_ops->pasid_init(pasid);
>> +		if (ret) {
>> +			kfree(pasid);
>> +			return NULL;
>> +		}
>> +	}
>> +
>> +	return pasid;
>> +}
>> +EXPORT_SYMBOL_GPL(iommu_pasid_init);
>> +
>> +void iommu_pasid_exit(struct iommu_pasid *pasid)
>> +{
>> +	kfree(pasid);
>> +}
>> +EXPORT_SYMBOL_GPL(iommu_pasid_exit);
>> +
>> +int iommu_pasid_alloc(struct iommu_pasid *pasid, ioasid_t min,
>> +		      ioasid_t max, void *private, ioasid_t *ioasid)
>> +{
>> +	ioasid_t start, end, hw, val;
>> +	int ret = -EAGAIN;
>> +
>> +	start = max_t(int, min, pasid->min);
>> +	end = min_t(int, max, pasid->max);
>> +
>> +	if (pasid->ops->pasid_alloc)
>> +		ret = pasid->ops->pasid_alloc(pasid, start, end,
>> &hw); +
>> +	if (ret == -EAGAIN)
>> +                val = ioasid_alloc(&pasid->set, start, end, private);
>> +	else if (ret == 0)
>> +		val = ioasid_alloc(&pasid->set, hw, hw + 1, private);
> this may fail due to conflict with other callers of ioasic_alloc(), but
> we should really respect the iommu ops allocator. If we move the
> pasid_alloc op into ioasid code, then we don't need to worry about the
> conflict.
>> +	else
>> +		goto hw_ret;
>> +
>> +	if (val == INVALID_IOASID)
>> +		goto ioasid_ret;
>> +
>> +	*ioasid = val;
>> +
>> +        return 0;
>> +
>> +ioasid_ret:
>> +	if (pasid->ops->pasid_free)
>> +		pasid->ops->pasid_free(pasid, hw);
>> +
>> +hw_ret:
>> +	return -ENODEV;
>> +}
>> +EXPORT_SYMBOL_GPL(iommu_pasid_alloc);
>> +
>> +void iommu_pasid_free(struct iommu_pasid *pasid, ioasid_t ioasid)
>> +{
>> +	if (pasid->ops->pasid_free)
>> +		pasid->ops->pasid_free(pasid, ioasid);
>> +
>> +	ioasid_free(ioasid);
>> +}
>> +EXPORT_SYMBOL_GPL(iommu_pasid_free);
>> diff --git a/include/linux/iommu.h b/include/linux/iommu.h
>> index 9bf1b3f2457a..4f5202c8170b 100644
>> --- a/include/linux/iommu.h
>> +++ b/include/linux/iommu.h
>> @@ -20,6 +20,7 @@
>>  #define __LINUX_IOMMU_H
>>  
>>  #include <linux/scatterlist.h>
>> +#include <linux/ioasid.h>
>>  #include <linux/device.h>
>>  #include <linux/types.h>
>>  #include <linux/errno.h>
>> @@ -48,6 +49,7 @@ struct bus_type;
>>  struct device;
>>  struct iommu_domain;
>>  struct notifier_block;
>> +struct iommu_pasid;
>>  
>>  /* iommu fault flags */
>>  #define IOMMU_FAULT_READ	0x0
>> @@ -194,6 +196,9 @@ enum iommu_dev_attr {
>>   * @of_xlate: add OF master IDs to iommu grouping
>>   * @get_dev_attr: get per device IOMMU attributions
>>   * @set_dev_attr: set per device IOMMU attributions
>> + * @pasid_init: initialize a pasid consumer
>> + * @pasid_alloc: allocate a pasid from low level driver
>> + * @pasid_free: free a pasid to low level driver
>>   * @pgsize_bitmap: bitmap of all possible supported page sizes
>>   */
>>  struct iommu_ops {
>> @@ -246,6 +251,12 @@ struct iommu_ops {
>>  	int (*attach_dev_aux)(struct iommu_domain *domain, struct
>> device *dev); void (*detach_dev_aux)(struct iommu_domain *domain,
>> struct device *dev); 
>> +	/* IOMMU pasid callbacks */
>> +	int (*pasid_init)(struct iommu_pasid *pasid);
>> +	int (*pasid_alloc)(struct iommu_pasid *pasid, ioasid_t start,
>> +			   ioasid_t end, ioasid_t *ioasid);
>> +	void (*pasid_free)(struct iommu_pasid *pasid, ioasid_t
>> ioasid); +
>>  	unsigned long pgsize_bitmap;
>>  };
>>  
>> @@ -428,12 +439,41 @@ extern int iommu_attach_device_aux(struct
>> iommu_domain *domain, extern void iommu_detach_device_aux(struct
>> iommu_domain *domain, struct device *dev);
>>  
>> +/*
>> + * Per IOMMU PASID consumer data.
>> + */
>> +struct iommu_pasid {
>> +	ioasid_t		max;
>> +	ioasid_t		min;
>> +	struct ioasid_set	set;
>> +	const struct iommu_ops	*ops;
>> +
>> +	/* vendor specific iommu private data */
>> +	void			*priv;
>> +};
>> +
>> +struct iommu_pasid *iommu_pasid_init(struct bus_type *bus);
>> +void iommu_pasid_exit(struct iommu_pasid *pasid);
>> +int iommu_pasid_alloc(struct iommu_pasid *pasid, ioasid_t min,
>> +		      ioasid_t max, void *private, ioasid_t *ioasid);
>> +void iommu_pasid_free(struct iommu_pasid *pasid, ioasid_t iosid);
>> +static inline int
>> +iommu_pasid_for_each(struct iommu_pasid *pasid, ioasid_iter_t func,
>> void *data) +{
>> +	return ioasid_for_each(&pasid->set, func, data);
>> +}
>> +static inline void*
>> +iommu_pasid_find(struct iommu_pasid *pasid, ioasid_t ioasid)
>> +{
>> +	return ioasid_find(&pasid->set, ioasid);
>> +}
>>  #else /* CONFIG_IOMMU_API */
>>  
>>  struct iommu_ops {};
>>  struct iommu_group {};
>>  struct iommu_fwspec {};
>>  struct iommu_device {};
>> +struct iommu_pasid {};
>>  
>>  static inline bool iommu_present(struct bus_type *bus)
>>  {
>> @@ -734,6 +774,39 @@ static inline void
>>  iommu_detach_device_aux(struct iommu_domain *domain, struct device
>> *dev) {
>>  }
>> +
>> +static inline struct iommu_pasid *
>> +iommu_pasid_init(struct bus_type *bus)
>> +{
>> +	return NULL;
>> +}
>> +
>> +static inline void iommu_pasid_exit(struct iommu_pasid *pasid)
>> +{
>> +}
>> +
>> +static inline int
>> +iommu_pasid_alloc(struct iommu_pasid *pasid, ioasid_t min,
>> +		  ioasid_t max, void *private, ioasid_t *ioasid)
>> +{
>> +	return -ENODEV;
>> +}
>> +
>> +static inline void iommu_pasid_free(struct iommu_pasid *pasid,
>> ioasid_t iosid) +{
>> +}
>> +
>> +static inline int
>> +iommu_pasid_for_each(struct iommu_pasid *pasid, ioasid_iter_t func,
>> void *data) +{
>> +	return -ENODEV;
>> +}
>> +
>> +static inline void*
>> +iommu_pasid_find(struct iommu_pasid *pasid, ioasid_t ioasid)
>> +{
>> +	return NULL;
>> +}
>>  #endif /* CONFIG_IOMMU_API */
>>  
>>  #ifdef CONFIG_IOMMU_DEBUGFS
> 
> [Jacob Pan]
> 

[RFC PATCH 2/5] iommu/vt-d: Initialize a PASID consumer

[Lu Baolu]

This adds the Intel vt-d specific ops to initialize a PASID
consumer.

Cc: Ashok Raj <ashok.raj@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Kevin Tian <kevin.tian@intel.com>
Signed-off-by: Liu Yi L <yi.l.liu@intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
---
 drivers/iommu/intel-iommu.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c
index 49a278a699b0..769b7059d52f 100644
--- a/drivers/iommu/intel-iommu.c
+++ b/drivers/iommu/intel-iommu.c
@@ -5597,6 +5597,35 @@ static int intel_iommu_domain_get_attr(struct iommu_domain *domain,
 	return ret;
 }
 
+static int intel_iommu_pasid_init(struct iommu_pasid *pasid)
+{
+	struct dmar_drhd_unit *drhd;
+	struct intel_iommu *iommu = NULL, *temp;
+
+	/*
+	 *  Iterate the iommu units and get a scalable mode capable one
+	 *  for virtual command usages. Return failure if scalable mode
+	 *  doesn't support by any unit.
+	 */
+	rcu_read_lock();
+	for_each_active_iommu(temp, drhd) {
+		if (sm_supported(temp)) {
+			iommu = temp;
+			break;
+		}
+	}
+	rcu_read_unlock();
+
+	if (!iommu)
+		return -ENODEV;
+
+	pasid->max = intel_pasid_max_id;
+	pasid->min = PASID_MIN;
+	pasid->priv = iommu;
+
+	return 0;
+}
+
 const struct iommu_ops intel_iommu_ops = {
 	.capable		= intel_iommu_capable,
 	.domain_alloc		= intel_iommu_domain_alloc,
@@ -5616,6 +5645,7 @@ const struct iommu_ops intel_iommu_ops = {
 	.domain_get_attr	= intel_iommu_domain_get_attr,
 	.get_dev_attr		= intel_iommu_get_dev_attr,
 	.set_dev_attr		= intel_iommu_set_dev_attr,
+	.pasid_init		= intel_iommu_pasid_init,
 	.pgsize_bitmap		= INTEL_IOMMU_PGSIZES,
 };
 
-- 
2.17.1

[RFC PATCH 3/5] iommu/vt-d: Enlightened PASID allocation

[Lu Baolu]

If Intel IOMMU runs in caching mode, a.k.a. virtual IOMMU, the
IOMMU driver should rely on the emulation software to allocate
and free PASID IDs. The Intel vt-d spec revision 3.0 defines a
register set to support this. This includes a capability register,
a virtual command register and a virtual response register. Refer
to section 10.4.42, 10.4.43, 10.4.44 for more information.

This patch adds the enlightened PASID allocation/free interfaces
via the virtual command register.

Cc: Ashok Raj <ashok.raj@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Kevin Tian <kevin.tian@intel.com>
Signed-off-by: Liu Yi L <yi.l.liu@intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
---
 drivers/iommu/intel-pasid.c | 70 +++++++++++++++++++++++++++++++++++++
 drivers/iommu/intel-pasid.h | 13 ++++++-
 include/linux/intel-iommu.h |  2 ++
 3 files changed, 84 insertions(+), 1 deletion(-)

diff --git a/drivers/iommu/intel-pasid.c b/drivers/iommu/intel-pasid.c
index 839f83974aca..fb42f0c2493e 100644
--- a/drivers/iommu/intel-pasid.c
+++ b/drivers/iommu/intel-pasid.c
@@ -28,6 +28,76 @@ static DEFINE_SPINLOCK(pasid_lock);
 u32 intel_pasid_max_id = PASID_MAX;
 static DEFINE_IDR(pasid_idr);
 
+int vcmd_alloc_pasid(struct intel_iommu *iommu, unsigned int *pasid)
+{
+	u64 res;
+	u64 cap;
+	u8 err_code;
+	unsigned long flags;
+	int ret = 0;
+
+	if (!ecap_vcs(iommu->ecap)) {
+		pr_warn("IOMMU: %s: Hardware doesn't support virtual command\n",
+			iommu->name);
+		return -ENODEV;
+	}
+
+	cap = dmar_readq(iommu->reg + DMAR_VCCAP_REG);
+	if (!(cap & DMA_VCS_PAS)) {
+		pr_warn("IOMMU: %s: Emulation software doesn't support PASID allocation\n",
+			iommu->name);
+		return -ENODEV;
+	}
+
+	raw_spin_lock_irqsave(&iommu->register_lock, flags);
+	dmar_writeq(iommu->reg + DMAR_VCMD_REG, VCMD_CMD_ALLOC);
+	IOMMU_WAIT_OP(iommu, DMAR_VCRSP_REG, dmar_readq,
+		      !(res & VCMD_VRSP_IP), res);
+	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
+
+	err_code = VCMD_VRSP_EC(res);
+	switch (err_code) {
+	case VCMD_VRSP_EC_SUCCESS:
+		*pasid = VCMD_VRSP_RESULE(res);
+		break;
+	case VCMD_VRSP_EC_UNAVAIL:
+		pr_info("IOMMU: %s: No PASID available\n", iommu->name);
+		ret = -ENOMEM;
+		break;
+	default:
+		ret = -ENODEV;
+		pr_warn("IOMMU: %s: Unkonwn error code %d\n",
+			iommu->name, err_code);
+	}
+
+	return ret;
+}
+
+void vcmd_free_pasid(struct intel_iommu *iommu, unsigned int pasid)
+{
+	u64 res;
+	u8 err_code;
+	unsigned long flags;
+
+	raw_spin_lock_irqsave(&iommu->register_lock, flags);
+	dmar_writeq(iommu->reg + DMAR_VCMD_REG, (pasid << 8) | VCMD_CMD_FREE);
+	IOMMU_WAIT_OP(iommu, DMAR_VCRSP_REG, dmar_readq,
+		      !(res & VCMD_VRSP_IP), res);
+	raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
+
+	err_code = VCMD_VRSP_EC(res);
+	switch (err_code) {
+	case VCMD_VRSP_EC_SUCCESS:
+		break;
+	case VCMD_VRSP_EC_INVAL:
+		pr_info("IOMMU: %s: Invalid PASID\n", iommu->name);
+		break;
+	default:
+		pr_warn("IOMMU: %s: Unkonwn error code %d\n",
+			iommu->name, err_code);
+	}
+}
+
 int intel_pasid_alloc_id(void *ptr, int start, int end, gfp_t gfp)
 {
 	int ret, min, max;
diff --git a/drivers/iommu/intel-pasid.h b/drivers/iommu/intel-pasid.h
index 381545ff9fb7..c80787d02e2d 100644
--- a/drivers/iommu/intel-pasid.h
+++ b/drivers/iommu/intel-pasid.h
@@ -19,6 +19,16 @@
 #define PASID_PDE_SHIFT			6
 #define MAX_NR_PASID_BITS		20
 
+/* Virtual command interface for enlightened pasid management. */
+#define VCMD_CMD_ALLOC			0x1
+#define VCMD_CMD_FREE			0x2
+#define VCMD_VRSP_IP			0x1
+#define VCMD_VRSP_EC(e)			(((e) >> 1) & 0x3)
+#define VCMD_VRSP_EC_SUCCESS		0
+#define VCMD_VRSP_EC_UNAVAIL		1
+#define VCMD_VRSP_EC_INVAL		1
+#define VCMD_VRSP_RESULE(e)		(((e) >> 8) & 0xfffff)
+
 /*
  * Domain ID reserved for pasid entries programmed for first-level
  * only and pass-through transfer modes.
@@ -69,5 +79,6 @@ int intel_pasid_setup_pass_through(struct intel_iommu *iommu,
 				   struct device *dev, int pasid);
 void intel_pasid_tear_down_entry(struct intel_iommu *iommu,
 				 struct device *dev, int pasid);
-
+int vcmd_alloc_pasid(struct intel_iommu *iommu, unsigned int *pasid);
+void vcmd_free_pasid(struct intel_iommu *iommu, unsigned int pasid);
 #endif /* __INTEL_PASID_H */
diff --git a/include/linux/intel-iommu.h b/include/linux/intel-iommu.h
index b563a61a6c39..4605eef3686c 100644
--- a/include/linux/intel-iommu.h
+++ b/include/linux/intel-iommu.h
@@ -173,6 +173,7 @@
 #define ecap_smpwc(e)		(((e) >> 48) & 0x1)
 #define ecap_flts(e)		(((e) >> 47) & 0x1)
 #define ecap_slts(e)		(((e) >> 46) & 0x1)
+#define ecap_vcs(e)		(((e) >> 45) & 0x1)
 #define ecap_smts(e)		(((e) >> 43) & 0x1)
 #define ecap_dit(e)		((e >> 41) & 0x1)
 #define ecap_pasid(e)		((e >> 40) & 0x1)
@@ -289,6 +290,7 @@
 
 /* PRS_REG */
 #define DMA_PRS_PPR	((u32)1)
+#define DMA_VCS_PAS	((u64)1)
 
 #define IOMMU_WAIT_OP(iommu, offset, op, cond, sts)			\
 do {									\
-- 
2.17.1

[RFC PATCH 4/5] iommu/vt-d: Allocate and free a pasid

[Lu Baolu]

This adds the Intel vt-d specific ops to allocate and free a
pasid value.

Cc: Ashok Raj <ashok.raj@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Kevin Tian <kevin.tian@intel.com>
Signed-off-by: Liu Yi L <yi.l.liu@intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
---
 drivers/iommu/intel-iommu.c | 33 +++++++++++++++++++++++++++++++++
 1 file changed, 33 insertions(+)

diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c
index 769b7059d52f..8dbd0c601dab 100644
--- a/drivers/iommu/intel-iommu.c
+++ b/drivers/iommu/intel-iommu.c
@@ -5626,6 +5626,37 @@ static int intel_iommu_pasid_init(struct iommu_pasid *pasid)
 	return 0;
 }
 
+static int intel_iommu_pasid_alloc(struct iommu_pasid *pasid, ioasid_t start,
+				   ioasid_t end, ioasid_t *ioasid)
+{
+	struct intel_iommu *iommu;
+
+	iommu = pasid->priv;
+	if (!iommu)
+		return -EINVAL;
+
+	/*
+	 * In caching mode, PASID ID should be allocated and freed
+	 * through the virtual command registers. Otherwise, rely
+	 * on the iommu global idr.
+	 */
+	if (!cap_caching_mode(iommu->cap))
+		return -EAGAIN;
+
+	return vcmd_alloc_pasid(iommu, ioasid);
+}
+
+static void intel_iommu_pasid_free(struct iommu_pasid *pasid, ioasid_t ioasid)
+{
+	struct intel_iommu *iommu;
+
+	iommu = pasid->priv;
+	if (!iommu || !cap_caching_mode(iommu->cap))
+		return;
+
+	vcmd_free_pasid(iommu, ioasid);
+}
+
 const struct iommu_ops intel_iommu_ops = {
 	.capable		= intel_iommu_capable,
 	.domain_alloc		= intel_iommu_domain_alloc,
@@ -5646,6 +5677,8 @@ const struct iommu_ops intel_iommu_ops = {
 	.get_dev_attr		= intel_iommu_get_dev_attr,
 	.set_dev_attr		= intel_iommu_set_dev_attr,
 	.pasid_init		= intel_iommu_pasid_init,
+	.pasid_alloc		= intel_iommu_pasid_alloc,
+	.pasid_free		= intel_iommu_pasid_free,
 	.pgsize_bitmap		= INTEL_IOMMU_PGSIZES,
 };
 
-- 
2.17.1

[RFC PATCH 5/5] iommu/vt-d: Use global pasid allocator

[Lu Baolu]

This uses global pasid allocator in the Intel iommu driver.

Cc: Ashok Raj <ashok.raj@intel.com>
Cc: Jacob Pan <jacob.jun.pan@linux.intel.com>
Cc: Kevin Tian <kevin.tian@intel.com>
Signed-off-by: Liu Yi L <yi.l.liu@intel.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
---
 drivers/iommu/intel-iommu.c | 20 +++++++++++++++-----
 drivers/iommu/intel-pasid.c | 36 ------------------------------------
 drivers/iommu/intel-pasid.h |  3 ---
 drivers/iommu/intel-svm.c   | 24 +++++++++++++-----------
 include/linux/intel-iommu.h |  3 +++
 5 files changed, 31 insertions(+), 55 deletions(-)

diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c
index 8dbd0c601dab..e15b81afcbbc 100644
--- a/drivers/iommu/intel-iommu.c
+++ b/drivers/iommu/intel-iommu.c
@@ -1698,7 +1698,11 @@ static void free_dmar_iommu(struct intel_iommu *iommu)
 		if (ecap_prs(iommu->ecap))
 			intel_svm_finish_prq(iommu);
 	}
+
+	iommu_pasid_exit(iommu->svm_pasid);
 #endif
+
+	iommu_pasid_exit(iommu->aux_pasid);
 }
 
 static struct dmar_domain *alloc_domain(int flags)
@@ -5050,7 +5054,7 @@ static void auxiliary_unlink_device(struct dmar_domain *domain,
 	domain->auxd_refcnt--;
 
 	if (!domain->auxd_refcnt && domain->default_pasid > 0)
-		intel_pasid_free_id(domain->default_pasid);
+		iommu_pasid_free(info->iommu->aux_pasid, domain->default_pasid);
 }
 
 static int domain_add_dev_auxd(struct dmar_domain *domain,
@@ -5067,9 +5071,10 @@ static int domain_add_dev_auxd(struct dmar_domain *domain,
 
 	spin_lock_irqsave(&device_domain_lock, flags);
 	if (domain->default_pasid <= 0) {
-		domain->default_pasid = intel_pasid_alloc_id(domain, PASID_MIN,
-				pci_max_pasids(to_pci_dev(dev)), GFP_ATOMIC);
-		if (domain->default_pasid < 0) {
+		ret = iommu_pasid_alloc(iommu->aux_pasid, PASID_MIN,
+					pci_max_pasids(to_pci_dev(dev)),
+					domain, &domain->default_pasid);
+		if (ret) {
 			pr_err("Can't allocate default pasid\n");
 			ret = -ENODEV;
 			goto pasid_failed;
@@ -5099,7 +5104,7 @@ static int domain_add_dev_auxd(struct dmar_domain *domain,
 attach_failed:
 	spin_unlock(&iommu->lock);
 	if (!domain->auxd_refcnt && domain->default_pasid > 0)
-		intel_pasid_free_id(domain->default_pasid);
+		iommu_pasid_free(iommu->aux_pasid, domain->default_pasid);
 pasid_failed:
 	spin_unlock_irqrestore(&device_domain_lock, flags);
 
@@ -5525,6 +5530,7 @@ static int intel_iommu_enable_auxd(struct device *dev)
 {
 	struct device_domain_info *info;
 	struct dmar_domain *domain;
+	struct intel_iommu *iommu;
 	unsigned long flags;
 
 	if (!scalable_mode_support())
@@ -5539,6 +5545,10 @@ static int intel_iommu_enable_auxd(struct device *dev)
 	info->auxd_enabled = 1;
 	spin_unlock_irqrestore(&device_domain_lock, flags);
 
+	iommu = info->iommu;
+	if (!iommu->aux_pasid)
+		iommu->aux_pasid = iommu_pasid_init(&pci_bus_type);
+
 	return 0;
 }
 
diff --git a/drivers/iommu/intel-pasid.c b/drivers/iommu/intel-pasid.c
index fb42f0c2493e..68eaef3bd4fd 100644
--- a/drivers/iommu/intel-pasid.c
+++ b/drivers/iommu/intel-pasid.c
@@ -26,7 +26,6 @@
  */
 static DEFINE_SPINLOCK(pasid_lock);
 u32 intel_pasid_max_id = PASID_MAX;
-static DEFINE_IDR(pasid_idr);
 
 int vcmd_alloc_pasid(struct intel_iommu *iommu, unsigned int *pasid)
 {
@@ -98,41 +97,6 @@ void vcmd_free_pasid(struct intel_iommu *iommu, unsigned int pasid)
 	}
 }
 
-int intel_pasid_alloc_id(void *ptr, int start, int end, gfp_t gfp)
-{
-	int ret, min, max;
-
-	min = max_t(int, start, PASID_MIN);
-	max = min_t(int, end, intel_pasid_max_id);
-
-	WARN_ON(in_interrupt());
-	idr_preload(gfp);
-	spin_lock(&pasid_lock);
-	ret = idr_alloc(&pasid_idr, ptr, min, max, GFP_ATOMIC);
-	spin_unlock(&pasid_lock);
-	idr_preload_end();
-
-	return ret;
-}
-
-void intel_pasid_free_id(int pasid)
-{
-	spin_lock(&pasid_lock);
-	idr_remove(&pasid_idr, pasid);
-	spin_unlock(&pasid_lock);
-}
-
-void *intel_pasid_lookup_id(int pasid)
-{
-	void *p;
-
-	spin_lock(&pasid_lock);
-	p = idr_find(&pasid_idr, pasid);
-	spin_unlock(&pasid_lock);
-
-	return p;
-}
-
 /*
  * Per device pasid table management:
  */
diff --git a/drivers/iommu/intel-pasid.h b/drivers/iommu/intel-pasid.h
index c80787d02e2d..029f82ce9f4a 100644
--- a/drivers/iommu/intel-pasid.h
+++ b/drivers/iommu/intel-pasid.h
@@ -60,9 +60,6 @@ struct pasid_table {
 };
 
 extern u32 intel_pasid_max_id;
-int intel_pasid_alloc_id(void *ptr, int start, int end, gfp_t gfp);
-void intel_pasid_free_id(int pasid);
-void *intel_pasid_lookup_id(int pasid);
 int intel_pasid_alloc_table(struct device *dev);
 void intel_pasid_free_table(struct device *dev);
 struct pasid_table *intel_pasid_get_table(struct device *dev);
diff --git a/drivers/iommu/intel-svm.c b/drivers/iommu/intel-svm.c
index 27cf1963207d..bb57a024eae7 100644
--- a/drivers/iommu/intel-svm.c
+++ b/drivers/iommu/intel-svm.c
@@ -261,6 +261,9 @@ int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_
 		BUG_ON(!mm);
 	}
 
+	if (!iommu->svm_pasid)
+		iommu->svm_pasid = iommu_pasid_init(&pci_bus_type);
+
 	mutex_lock(&pasid_mutex);
 	if (pasid && !(flags & SVM_FLAG_PRIVATE_PASID)) {
 		struct intel_svm *t;
@@ -325,15 +328,14 @@ int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_
 			pasid_max = intel_pasid_max_id;
 
 		/* Do not use PASID 0 in caching mode (virtualised IOMMU) */
-		ret = intel_pasid_alloc_id(svm,
-					   !!cap_caching_mode(iommu->cap),
-					   pasid_max - 1, GFP_KERNEL);
-		if (ret < 0) {
+		ret = iommu_pasid_alloc(iommu->svm_pasid,
+					!!cap_caching_mode(iommu->cap),
+					pasid_max - 1, svm, &svm->pasid);
+		if (ret) {
 			kfree(svm);
 			kfree(sdev);
 			goto out;
 		}
-		svm->pasid = ret;
 		svm->notifier.ops = &intel_mmuops;
 		svm->mm = mm;
 		svm->flags = flags;
@@ -343,7 +345,7 @@ int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_
 		if (mm) {
 			ret = mmu_notifier_register(&svm->notifier, mm);
 			if (ret) {
-				intel_pasid_free_id(svm->pasid);
+				iommu_pasid_free(iommu->svm_pasid, svm->pasid);
 				kfree(svm);
 				kfree(sdev);
 				goto out;
@@ -356,7 +358,7 @@ int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_
 		if (ret) {
 			if (mm)
 				mmu_notifier_unregister(&svm->notifier, mm);
-			intel_pasid_free_id(svm->pasid);
+			iommu_pasid_free(iommu->svm_pasid, svm->pasid);
 			kfree(svm);
 			kfree(sdev);
 			goto out;
@@ -389,7 +391,7 @@ int intel_svm_unbind_mm(struct device *dev, int pasid)
 	if (!iommu)
 		goto out;
 
-	svm = intel_pasid_lookup_id(pasid);
+	svm = iommu_pasid_find(iommu->svm_pasid, pasid);
 	if (!svm)
 		goto out;
 
@@ -411,7 +413,7 @@ int intel_svm_unbind_mm(struct device *dev, int pasid)
 				kfree_rcu(sdev, rcu);
 
 				if (list_empty(&svm->devs)) {
-					intel_pasid_free_id(svm->pasid);
+					iommu_pasid_free(iommu->svm_pasid, svm->pasid);
 					if (svm->mm)
 						mmu_notifier_unregister(&svm->notifier, svm->mm);
 
@@ -446,7 +448,7 @@ int intel_svm_is_pasid_valid(struct device *dev, int pasid)
 	if (!iommu)
 		goto out;
 
-	svm = intel_pasid_lookup_id(pasid);
+	svm = iommu_pasid_find(iommu->svm_pasid, pasid);
 	if (!svm)
 		goto out;
 
@@ -545,7 +547,7 @@ static irqreturn_t prq_event_thread(int irq, void *d)
 
 		if (!svm || svm->pasid != req->pasid) {
 			rcu_read_lock();
-			svm = intel_pasid_lookup_id(req->pasid);
+			svm = iommu_pasid_find(iommu->svm_pasid, req->pasid);
 			/* It *can't* go away, because the driver is not permitted
 			 * to unbind the mm while any page faults are outstanding.
 			 * So we only need RCU to protect the internal idr code. */
diff --git a/include/linux/intel-iommu.h b/include/linux/intel-iommu.h
index 4605eef3686c..d79f99388293 100644
--- a/include/linux/intel-iommu.h
+++ b/include/linux/intel-iommu.h
@@ -552,6 +552,7 @@ struct intel_iommu {
 #ifdef CONFIG_INTEL_IOMMU_SVM
 	struct page_req_dsc *prq;
 	unsigned char prq_name[16];    /* Name for PRQ interrupt */
+	struct iommu_pasid *svm_pasid;	/* SVM PASID consumer */
 #endif
 	struct q_inval  *qi;            /* Queued invalidation info */
 	u32 *iommu_state; /* Store iommu states between suspend and resume.*/
@@ -564,6 +565,8 @@ struct intel_iommu {
 	struct iommu_device iommu;  /* IOMMU core code handle */
 	int		node;
 	u32		flags;      /* Software defined flags */
+
+	struct iommu_pasid *aux_pasid; /* Aux-domain pasid consumer */
 };
 
 /* PCI domain-device relationship */
-- 
2.17.1

Re: [RFC PATCH 0/5] iommu: APIs for paravirtual PASID allocation

[Konrad Rzeszutek Wilk]

On Mon, Nov 12, 2018 at 02:44:56PM +0800, Lu Baolu wrote:
> This adds an uniformed API set for global PASIDs used by IOMMU
> and device drivers which depend on IOMMU. It works for drivers
> running on bare metal, full virtualized environments and para-
> virtualized environment.
> 

Are there also some QEMU patches for this in RFC state?

Re: [RFC PATCH 0/5] iommu: APIs for paravirtual PASID allocation

[Yi Sun]

On 18-11-19 11:36:48, Konrad Rzeszutek Wilk wrote:
> On Mon, Nov 12, 2018 at 02:44:56PM +0800, Lu Baolu wrote:
> > This adds an uniformed API set for global PASIDs used by IOMMU
> > and device drivers which depend on IOMMU. It works for drivers
> > running on bare metal, full virtualized environments and para-
> > virtualized environment.
> > 
> 
> Are there also some QEMU patches for this in RFC state?

Yes, we already have draft codes and are working on refining
them. Will send patches out for your review. Thanks!