[v3,3/4] iommufd: Add IOMMU_DEVICE_GET_HW_INFO

Commit Message

Yi Liu May 11, 2023, 2:30 p.m. UTC

Under nested IOMMU translation, userspace owns the stage-1 translation
table (e.g. the stage-1 page table of Intel VT-d or the context table
of ARM SMMUv3, and etc.). Stage-1 translation tables are vendor specific,
and needs to be compatiable with the underlying IOMMU hardware. Hence,
userspace should know the IOMMU hardware capability before creating and
configuring the stage-1 translation table to kernel.

This adds IOMMU_DEVICE_GET_HW_INFO to query the IOMMU hardware information
for a given device. The returned data is vendor specific, userspace needs
to decode it with the structure mapped by the @out_data_type field.

As only physical devices have IOMMU hardware, so this will return error
if the given device is not a physical device.

Co-developed-by: Nicolin Chen <[e-mail]>
Signed-off-by: Nicolin Chen <[e-mail]>
Signed-off-by: Yi Liu <[e-mail]>
---
 drivers/iommu/iommufd/device.c          | 72 +++++++++++++++++++++++++
 drivers/iommu/iommufd/iommufd_private.h |  1 +
 drivers/iommu/iommufd/main.c            |  3 ++
 include/uapi/linux/iommufd.h            | 37 +++++++++++++
 4 files changed, 113 insertions(+)

Comments

Baolu Lu May 12, 2023, 5:38 a.m. UTC | #1

On 5/11/23 10:30 PM, Yi Liu wrote:
> Under nested IOMMU translation, userspace owns the stage-1 translation
> table (e.g. the stage-1 page table of Intel VT-d or the context table
> of ARM SMMUv3, and etc.). Stage-1 translation tables are vendor specific,
> and needs to be compatiable with the underlying IOMMU hardware. Hence,
> userspace should know the IOMMU hardware capability before creating and
> configuring the stage-1 translation table to kernel.
> 
> This adds IOMMU_DEVICE_GET_HW_INFO to query the IOMMU hardware information
> for a given device. The returned data is vendor specific, userspace needs
> to decode it with the structure mapped by the @out_data_type field.
> 
> As only physical devices have IOMMU hardware, so this will return error
> if the given device is not a physical device.
> 
> Co-developed-by: Nicolin Chen <[e-mail]>
> Signed-off-by: Nicolin Chen <[e-mail]>
> Signed-off-by: Yi Liu <[e-mail]>
> ---
>   drivers/iommu/iommufd/device.c          | 72 +++++++++++++++++++++++++
>   drivers/iommu/iommufd/iommufd_private.h |  1 +
>   drivers/iommu/iommufd/main.c            |  3 ++
>   include/uapi/linux/iommufd.h            | 37 +++++++++++++
>   4 files changed, 113 insertions(+)
> 
> diff --git a/drivers/iommu/iommufd/device.c b/drivers/iommu/iommufd/device.c
> index 051bd8e99858..bc99d092de8f 100644
> --- a/drivers/iommu/iommufd/device.c
> +++ b/drivers/iommu/iommufd/device.c
> @@ -263,6 +263,78 @@ u32 iommufd_device_to_id(struct iommufd_device *idev)
>   }
>   EXPORT_SYMBOL_NS_GPL(iommufd_device_to_id, IOMMUFD);
>   
> +static int iommufd_zero_fill_user(u64 ptr, int bytes)
> +{
> +	int index = 0;
> +
> +	for (; index < bytes; index++) {
> +		if (put_user(0, (uint8_t __user *)u64_to_user_ptr(ptr + index)))
> +			return -EFAULT;
> +	}
> +	return 0;
> +}
> +
> +int iommufd_device_get_hw_info(struct iommufd_ucmd *ucmd)
> +{
> +	struct iommu_hw_info *cmd = ucmd->cmd;
> +	unsigned int length = 0, data_len;
> +	struct iommufd_device *idev;
> +	const struct iommu_ops *ops;
> +	void *data = NULL;
> +	int rc = 0;
> +
> +	if (cmd->flags || cmd->__reserved || !cmd->data_len)
> +		return -EOPNOTSUPP;
> +
> +	idev = iommufd_get_device(ucmd, cmd->dev_id);
> +	if (IS_ERR(idev))
> +		return PTR_ERR(idev);
> +
> +	ops = dev_iommu_ops(idev->dev);
> +	if (!ops->hw_info)
> +		goto done;

If the iommu driver doesn't provide a hw_info callback, it still
returns success?

> +
> +	/* driver has hw_info callback should have a unique hw_info_type */
> +	if (ops->hw_info_type == IOMMU_HW_INFO_TYPE_NONE) {
> +		pr_warn_ratelimited("iommu driver set an invalid type\n");
> +		rc = -ENODEV;
> +		goto out_err;
> +	}
> +
> +	data = ops->hw_info(idev->dev, &data_len);
> +	if (IS_ERR(data)) {
> +		rc = PTR_ERR(data);
> +		goto out_err;
> +	}
> +
> +	length = min(cmd->data_len, data_len);
> +	if (copy_to_user(u64_to_user_ptr(cmd->data_ptr), data, length)) {
> +		rc = -EFAULT;
> +		goto out_err;
> +	}
> +
> +	/*
> +	 * Zero the trailing bytes if the user buffer is bigger than the
> +	 * data size kernel actually has.
> +	 */
> +	if (length < cmd->data_len) {
> +		rc = iommufd_zero_fill_user(cmd->data_ptr + length,
> +					    cmd->data_len - length);
> +		if (rc)
> +			goto out_err;
> +	}
> +
> +done:
> +	cmd->data_len = length;
> +	cmd->out_data_type = ops->hw_info_type;
> +	rc = iommufd_ucmd_respond(ucmd, sizeof(*cmd));
> +
> +out_err:
> +	kfree(data);
> +	iommufd_put_object(&idev->obj);
> +	return rc;
> +}
> +
>   static int iommufd_group_setup_msi(struct iommufd_group *igroup,
>   				   struct iommufd_hw_pagetable *hwpt)
>   {

Best regards,
baolu

Yi Liu May 15, 2023, 6:14 a.m. UTC | #2

> -----Original Message-----
> From: Baolu Lu <[e-mail]>
> Sent: Friday, May 12, 2023 1:39 PM
> To: Liu, Yi L <[e-mail]>; [e-mail]; [e-mail];
> [e-mail]; Tian, Kevin <[e-mail]>; [e-mail]
> Cc: [e-mail]; [e-mail]; [e-mail];
> [e-mail]; [e-mail]; [e-mail];
> [e-mail]; [e-mail]; [e-mail];
> [e-mail]; [e-mail]; [e-mail];
> [e-mail]; [e-mail]; [e-mail];
> [e-mail]; Duan, Zhenzhong <[e-mail]>
> Subject: Re: [PATCH v3 3/4] iommufd: Add IOMMU_DEVICE_GET_HW_INFO
> 
> On 5/11/23 10:30 PM, Yi Liu wrote:
> > Under nested IOMMU translation, userspace owns the stage-1 translation
> > table (e.g. the stage-1 page table of Intel VT-d or the context table
> > of ARM SMMUv3, and etc.). Stage-1 translation tables are vendor specific,
> > and needs to be compatiable with the underlying IOMMU hardware. Hence,
> > userspace should know the IOMMU hardware capability before creating and
> > configuring the stage-1 translation table to kernel.
> >
> > This adds IOMMU_DEVICE_GET_HW_INFO to query the IOMMU hardware
> information
> > for a given device. The returned data is vendor specific, userspace needs
> > to decode it with the structure mapped by the @out_data_type field.
> >
> > As only physical devices have IOMMU hardware, so this will return error
> > if the given device is not a physical device.
> >
> > Co-developed-by: Nicolin Chen <[e-mail]>
> > Signed-off-by: Nicolin Chen <[e-mail]>
> > Signed-off-by: Yi Liu <[e-mail]>
> > ---
> >   drivers/iommu/iommufd/device.c          | 72 +++++++++++++++++++++++++
> >   drivers/iommu/iommufd/iommufd_private.h |  1 +
> >   drivers/iommu/iommufd/main.c            |  3 ++
> >   include/uapi/linux/iommufd.h            | 37 +++++++++++++
> >   4 files changed, 113 insertions(+)
> >
> > diff --git a/drivers/iommu/iommufd/device.c b/drivers/iommu/iommufd/device.c
> > index 051bd8e99858..bc99d092de8f 100644
> > --- a/drivers/iommu/iommufd/device.c
> > +++ b/drivers/iommu/iommufd/device.c
> > @@ -263,6 +263,78 @@ u32 iommufd_device_to_id(struct iommufd_device *idev)
> >   }
> >   EXPORT_SYMBOL_NS_GPL(iommufd_device_to_id, IOMMUFD);
> >
> > +static int iommufd_zero_fill_user(u64 ptr, int bytes)
> > +{
> > +	int index = 0;
> > +
> > +	for (; index < bytes; index++) {
> > +		if (put_user(0, (uint8_t __user *)u64_to_user_ptr(ptr + index)))
> > +			return -EFAULT;
> > +	}
> > +	return 0;
> > +}
> > +
> > +int iommufd_device_get_hw_info(struct iommufd_ucmd *ucmd)
> > +{
> > +	struct iommu_hw_info *cmd = ucmd->cmd;
> > +	unsigned int length = 0, data_len;
> > +	struct iommufd_device *idev;
> > +	const struct iommu_ops *ops;
> > +	void *data = NULL;
> > +	int rc = 0;
> > +
> > +	if (cmd->flags || cmd->__reserved || !cmd->data_len)
> > +		return -EOPNOTSUPP;
> > +
> > +	idev = iommufd_get_device(ucmd, cmd->dev_id);
> > +	if (IS_ERR(idev))
> > +		return PTR_ERR(idev);
> > +
> > +	ops = dev_iommu_ops(idev->dev);
> > +	if (!ops->hw_info)
> > +		goto done;
> 
> If the iommu driver doesn't provide a hw_info callback, it still
> returns success?

Yes, as noted in the cover letter. It's for a remark from Jason. In such
case, the out_data_type is NULL, it means no specific data is filled
in the buffer pointed by cmd->data_ptr.

- Let IOMMU_DEVICE_GET_HW_INFO succeed even the underlying iommu driver
   does not have driver-specific data to report per below remark.
   https://lore.kernel.org/kvm/[e-mail]/

Regards,
Yi Liu

> > +
> > +	/* driver has hw_info callback should have a unique hw_info_type */
> > +	if (ops->hw_info_type == IOMMU_HW_INFO_TYPE_NONE) {
> > +		pr_warn_ratelimited("iommu driver set an invalid type\n");
> > +		rc = -ENODEV;
> > +		goto out_err;
> > +	}
> > +
> > +	data = ops->hw_info(idev->dev, &data_len);
> > +	if (IS_ERR(data)) {
> > +		rc = PTR_ERR(data);
> > +		goto out_err;
> > +	}
> > +
> > +	length = min(cmd->data_len, data_len);
> > +	if (copy_to_user(u64_to_user_ptr(cmd->data_ptr), data, length)) {
> > +		rc = -EFAULT;
> > +		goto out_err;
> > +	}
> > +
> > +	/*
> > +	 * Zero the trailing bytes if the user buffer is bigger than the
> > +	 * data size kernel actually has.
> > +	 */
> > +	if (length < cmd->data_len) {
> > +		rc = iommufd_zero_fill_user(cmd->data_ptr + length,
> > +					    cmd->data_len - length);
> > +		if (rc)
> > +			goto out_err;
> > +	}
> > +
> > +done:
> > +	cmd->data_len = length;
> > +	cmd->out_data_type = ops->hw_info_type;
> > +	rc = iommufd_ucmd_respond(ucmd, sizeof(*cmd));
> > +
> > +out_err:
> > +	kfree(data);
> > +	iommufd_put_object(&idev->obj);
> > +	return rc;
> > +}
> > +
> >   static int iommufd_group_setup_msi(struct iommufd_group *igroup,
> >   				   struct iommufd_hw_pagetable *hwpt)
> >   {
> 
> Best regards,
> baolu

Baolu Lu May 16, 2023, 1:49 a.m. UTC | #3

On 5/15/23 2:14 PM, Liu, Yi L wrote:
>> -----Original Message-----
>> From: Baolu Lu<[e-mail]>
>> Sent: Friday, May 12, 2023 1:39 PM
>> To: Liu, Yi L<[e-mail]>;[e-mail];[e-mail];
>> [e-mail]; Tian, Kevin<[e-mail]>;[e-mail]
>> Cc:[e-mail];[e-mail];[e-mail];
>> [e-mail];[e-mail];[e-mail];
>> [e-mail];[e-mail];[e-mail];
>> [e-mail];[e-mail];[e-mail];
>> [e-mail];[e-mail];[e-mail];
>> [e-mail]; Duan, Zhenzhong<[e-mail]>
>> Subject: Re: [PATCH v3 3/4] iommufd: Add IOMMU_DEVICE_GET_HW_INFO
>>
>> On 5/11/23 10:30 PM, Yi Liu wrote:
>>> Under nested IOMMU translation, userspace owns the stage-1 translation
>>> table (e.g. the stage-1 page table of Intel VT-d or the context table
>>> of ARM SMMUv3, and etc.). Stage-1 translation tables are vendor specific,
>>> and needs to be compatiable with the underlying IOMMU hardware. Hence,
>>> userspace should know the IOMMU hardware capability before creating and
>>> configuring the stage-1 translation table to kernel.
>>>
>>> This adds IOMMU_DEVICE_GET_HW_INFO to query the IOMMU hardware
>> information
>>> for a given device. The returned data is vendor specific, userspace needs
>>> to decode it with the structure mapped by the @out_data_type field.
>>>
>>> As only physical devices have IOMMU hardware, so this will return error
>>> if the given device is not a physical device.
>>>
>>> Co-developed-by: Nicolin Chen<[e-mail]>
>>> Signed-off-by: Nicolin Chen<[e-mail]>
>>> Signed-off-by: Yi Liu<[e-mail]>
>>> ---
>>>    drivers/iommu/iommufd/device.c          | 72 +++++++++++++++++++++++++
>>>    drivers/iommu/iommufd/iommufd_private.h |  1 +
>>>    drivers/iommu/iommufd/main.c            |  3 ++
>>>    include/uapi/linux/iommufd.h            | 37 +++++++++++++
>>>    4 files changed, 113 insertions(+)
>>>
>>> diff --git a/drivers/iommu/iommufd/device.c b/drivers/iommu/iommufd/device.c
>>> index 051bd8e99858..bc99d092de8f 100644
>>> --- a/drivers/iommu/iommufd/device.c
>>> +++ b/drivers/iommu/iommufd/device.c
>>> @@ -263,6 +263,78 @@ u32 iommufd_device_to_id(struct iommufd_device *idev)
>>>    }
>>>    EXPORT_SYMBOL_NS_GPL(iommufd_device_to_id, IOMMUFD);
>>>
>>> +static int iommufd_zero_fill_user(u64 ptr, int bytes)
>>> +{
>>> +	int index = 0;
>>> +
>>> +	for (; index < bytes; index++) {
>>> +		if (put_user(0, (uint8_t __user *)u64_to_user_ptr(ptr + index)))
>>> +			return -EFAULT;
>>> +	}
>>> +	return 0;
>>> +}
>>> +
>>> +int iommufd_device_get_hw_info(struct iommufd_ucmd *ucmd)
>>> +{
>>> +	struct iommu_hw_info *cmd = ucmd->cmd;
>>> +	unsigned int length = 0, data_len;
>>> +	struct iommufd_device *idev;
>>> +	const struct iommu_ops *ops;
>>> +	void *data = NULL;
>>> +	int rc = 0;
>>> +
>>> +	if (cmd->flags || cmd->__reserved || !cmd->data_len)
>>> +		return -EOPNOTSUPP;
>>> +
>>> +	idev = iommufd_get_device(ucmd, cmd->dev_id);
>>> +	if (IS_ERR(idev))
>>> +		return PTR_ERR(idev);
>>> +
>>> +	ops = dev_iommu_ops(idev->dev);
>>> +	if (!ops->hw_info)
>>> +		goto done;
>> If the iommu driver doesn't provide a hw_info callback, it still
>> returns success?
> Yes, as noted in the cover letter. It's for a remark from Jason. In such
> case, the out_data_type is NULL, it means no specific data is filled
> in the buffer pointed by cmd->data_ptr.
> 
> - Let IOMMU_DEVICE_GET_HW_INFO succeed even the underlying iommu driver
>     does not have driver-specific data to report per below remark.
>     https://lore.kernel.org/kvm/[e-mail]/

Oh, I overlooked that. Thanks for the explanation. It's fair enough.

Best regards,
baolu

Tian, Kevin May 19, 2023, 8:42 a.m. UTC | #4

> From: Liu, Yi L <[e-mail]>
> Sent: Thursday, May 11, 2023 10:30 PM
> +
> +int iommufd_device_get_hw_info(struct iommufd_ucmd *ucmd)
> +{
> +	struct iommu_hw_info *cmd = ucmd->cmd;
> +	unsigned int length = 0, data_len;
> +	struct iommufd_device *idev;
> +	const struct iommu_ops *ops;
> +	void *data = NULL;
> +	int rc = 0;
> +
> +	if (cmd->flags || cmd->__reserved || !cmd->data_len)
> +		return -EOPNOTSUPP;
> +
> +	idev = iommufd_get_device(ucmd, cmd->dev_id);
> +	if (IS_ERR(idev))
> +		return PTR_ERR(idev);
> +
> +	ops = dev_iommu_ops(idev->dev);
> +	if (!ops->hw_info)
> +		goto done;
> +
> +	/* driver has hw_info callback should have a unique hw_info_type */
> +	if (ops->hw_info_type == IOMMU_HW_INFO_TYPE_NONE) {
> +		pr_warn_ratelimited("iommu driver set an invalid type\n");
> +		rc = -ENODEV;
> +		goto out_err;
> +	}

this should be a WARN_ON_ONCE() since it's a driver bug.

> +
> +	data = ops->hw_info(idev->dev, &data_len);
> +	if (IS_ERR(data)) {
> +		rc = PTR_ERR(data);
> +		goto out_err;
> +	}
> +
> +	length = min(cmd->data_len, data_len);
> +	if (copy_to_user(u64_to_user_ptr(cmd->data_ptr), data, length)) {
> +		rc = -EFAULT;
> +		goto out_err;
> +	}
> +
> +	/*
> +	 * Zero the trailing bytes if the user buffer is bigger than the
> +	 * data size kernel actually has.
> +	 */
> +	if (length < cmd->data_len) {
> +		rc = iommufd_zero_fill_user(cmd->data_ptr + length,
> +					    cmd->data_len - length);
> +		if (rc)
> +			goto out_err;
> +	}
> +
> +done:
> +	cmd->data_len = length;
> +	cmd->out_data_type = ops->hw_info_type;
> +	rc = iommufd_ucmd_respond(ucmd, sizeof(*cmd));

the 'done' label should be moved before above zero_fill. Otherwise
in !ops->hw_info case the user buffer is not cleared.

>  union ucmd_buffer {
>  	struct iommu_destroy destroy;
>  	struct iommu_hwpt_alloc hwpt;
> +	struct iommu_hw_info info;

follow alphabetic order this should be ahead of hwpt.

> @@ -302,6 +303,8 @@ static const struct iommufd_ioctl_op
> iommufd_ioctl_ops[] = {
>  	IOCTL_OP(IOMMU_DESTROY, iommufd_destroy, struct
> iommu_destroy, id),
>  	IOCTL_OP(IOMMU_HWPT_ALLOC, iommufd_hwpt_alloc, struct
> iommu_hwpt_alloc,
>  		 __reserved),
> +	IOCTL_OP(IOMMU_DEVICE_GET_HW_INFO,
> iommufd_device_get_hw_info,
> +		 struct iommu_hw_info, __reserved),

before IOMMU_HWPT_ALLOC

> +
> +/**
> + * struct iommu_hw_info - ioctl(IOMMU_DEVICE_GET_HW_INFO)
> + * @size: sizeof(struct iommu_hw_info)
> + * @flags: Must be 0
> + * @dev_id: The device bound to the iommufd
> + * @data_len: Input the length of the user buffer in bytes. Output the
> + *            length of data filled in the user buffer.
> + * @data_ptr: Pointer to the user buffer
> + * @out_data_type: Output the iommu hardware info type as defined by
> + *                 enum iommu_hw_info_type.
> + * @__reserved: Must be 0
> + *
> + * Query the hardware iommu information for given device which has been
> + * bound to iommufd. @data_len is the size of the buffer which captures
> + * iommu type specific data and the data will be filled. Trailing bytes
> + * are zeroed if the user buffer is larger than the data kernel has.
> + *
> + * The type specific data would be used to sync capability between the
> + * virtual IOMMU and the hardware IOMMU. e.g. nested translation
> requires
> + * to check the hardware IOMMU capability so guest stage-1 page table
> + * uses a format compatible to the hardware IOMMU.
> + *
> + * The @out_data_type will be filled if the ioctl succeeds. It would
> + * be used to decode the data filled in the buffer pointed by @data_ptr.
> + */
> +struct iommu_hw_info {
> +	__u32 size;
> +	__u32 flags;
> +	__u32 dev_id;
> +	__u32 data_len;
> +	__aligned_u64 data_ptr;
> +	__u32 out_data_type;
> +	__u32 __reserved;

it's unusual to have reserved field in the end. It makes more sense
to move data_ptr to the end to make it meaningful.

> +};
> +#define IOMMU_DEVICE_GET_HW_INFO _IO(IOMMUFD_TYPE,
> IOMMUFD_CMD_DEVICE_GET_HW_INFO)
>  #endif

Here we have a naming confusion.

'IOMMU' is the prefix of iommufd ioctls.

'DEVICE' is the subjective.

Then "GET_HW_INFO" implies getting hardware info related to
this device. then it should not be restricted to the iommu info.

with that it's clearer to call it IOMMU_DEVICE_GET_IOMMU_INFO.

similarly for struct iommu_hw_info.

'iommu' is the prefix for all iommufd ioctl structures.

then 'hw_info' is too broard.

iommu_device_iommu_info reads better? though having two
iommu's in the name is a little bit annoying...

Tian, Kevin May 19, 2023, 9:15 a.m. UTC | #5

> From: Tian, Kevin
> Sent: Friday, May 19, 2023 4:42 PM
> > +struct iommu_hw_info {
> > +	__u32 size;
> > +	__u32 flags;
> > +	__u32 dev_id;
> > +	__u32 data_len;
> > +	__aligned_u64 data_ptr;
> > +	__u32 out_data_type;
> > +	__u32 __reserved;
> 
> it's unusual to have reserved field in the end. It makes more sense
> to move data_ptr to the end to make it meaningful.
> 

Please ignore this comment. typed too fast...

Nicolin Chen May 19, 2023, 6:30 p.m. UTC | #6

Hi Kevin,

On Fri, May 19, 2023 at 08:42:07AM +0000, Tian, Kevin wrote:

> > +};
> > +#define IOMMU_DEVICE_GET_HW_INFO _IO(IOMMUFD_TYPE,
> > IOMMUFD_CMD_DEVICE_GET_HW_INFO)
> >  #endif
> 
> Here we have a naming confusion.
> 
> 'IOMMU' is the prefix of iommufd ioctls.
> 
> 'DEVICE' is the subjective.
> 
> Then "GET_HW_INFO" implies getting hardware info related to
> this device. then it should not be restricted to the iommu info.
> 
> with that it's clearer to call it IOMMU_DEVICE_GET_IOMMU_INFO.

Though the entire ioctl is tied to the input "dev_id", I think
it isn't really about the device corresponding to the dev_id,
similar to the IOMMU_HWPT_ALLOC having a dev_id input too. So,
I think the "IOMMU_DEVICE" here should be interpreted simply
as "an iommu device". We could also highlight this somewhere
in the header.

With that being said, IOMMU_DEVICE_SET/UNSET_DATA should be
renamed to IOMMU_DEVICE_SET/UNSET_DEV_DATA -- "DEVICE" is the
iommu device while the "DEV_DATA" is a given device that's
behind the iommu.

> similarly for struct iommu_hw_info.
> 
> 'iommu' is the prefix for all iommufd ioctl structures.
> 
> then 'hw_info' is too broard.
> 
> iommu_device_iommu_info reads better? though having two
> iommu's in the name is a little bit annoying...

How about:
IOMMU_DEVICE_GET_FEATURES
struct iommu_device_features
?

Thanks
Nic

Tian, Kevin May 24, 2023, 5 a.m. UTC | #7

> From: Nicolin Chen <[e-mail]>
> Sent: Saturday, May 20, 2023 2:30 AM
> 
> Hi Kevin,
> 
> On Fri, May 19, 2023 at 08:42:07AM +0000, Tian, Kevin wrote:
> 
> > > +};
> > > +#define IOMMU_DEVICE_GET_HW_INFO _IO(IOMMUFD_TYPE,
> > > IOMMUFD_CMD_DEVICE_GET_HW_INFO)
> > >  #endif
> >
> > Here we have a naming confusion.
> >
> > 'IOMMU' is the prefix of iommufd ioctls.
> >
> > 'DEVICE' is the subjective.
> >
> > Then "GET_HW_INFO" implies getting hardware info related to
> > this device. then it should not be restricted to the iommu info.
> >
> > with that it's clearer to call it IOMMU_DEVICE_GET_IOMMU_INFO.
> 
> Though the entire ioctl is tied to the input "dev_id", I think
> it isn't really about the device corresponding to the dev_id,
> similar to the IOMMU_HWPT_ALLOC having a dev_id input too. So,
> I think the "IOMMU_DEVICE" here should be interpreted simply
> as "an iommu device". We could also highlight this somewhere
> in the header.

yes this is a good view of it. with that it's not necessary to have
a 'DEVICE' notation in the name which looks confusing with dev_id.

Just IOMMU_GET_HW_INFO for the iommu behind the specified dev_id.

then keep the structure name as iommu_hw_info.

> 
> With that being said, IOMMU_DEVICE_SET/UNSET_DATA should be
> renamed to IOMMU_DEVICE_SET/UNSET_DEV_DATA -- "DEVICE" is the
> iommu device while the "DEV_DATA" is a given device that's
> behind the iommu.

this then becomes IOMMU_SET/UNSET_DEV_DATA.

> 
> > similarly for struct iommu_hw_info.
> >
> > 'iommu' is the prefix for all iommufd ioctl structures.
> >
> > then 'hw_info' is too broard.
> >
> > iommu_device_iommu_info reads better? though having two
> > iommu's in the name is a little bit annoying...
> 
> How about:
> IOMMU_DEVICE_GET_FEATURES
> struct iommu_device_features
> ?
> 
> Thanks
> Nic

Nicolin Chen May 24, 2023, 5:19 a.m. UTC | #8

On Wed, May 24, 2023 at 05:00:40AM +0000, Tian, Kevin wrote:
> > > > +};
> > > > +#define IOMMU_DEVICE_GET_HW_INFO _IO(IOMMUFD_TYPE,
> > > > IOMMUFD_CMD_DEVICE_GET_HW_INFO)
> > > >  #endif
> > >
> > > Here we have a naming confusion.
> > >
> > > 'IOMMU' is the prefix of iommufd ioctls.
> > >
> > > 'DEVICE' is the subjective.
> > >
> > > Then "GET_HW_INFO" implies getting hardware info related to
> > > this device. then it should not be restricted to the iommu info.
> > >
> > > with that it's clearer to call it IOMMU_DEVICE_GET_IOMMU_INFO.
> >
> > Though the entire ioctl is tied to the input "dev_id", I think
> > it isn't really about the device corresponding to the dev_id,
> > similar to the IOMMU_HWPT_ALLOC having a dev_id input too. So,
> > I think the "IOMMU_DEVICE" here should be interpreted simply
> > as "an iommu device". We could also highlight this somewhere
> > in the header.
> 
> yes this is a good view of it. with that it's not necessary to have
> a 'DEVICE' notation in the name which looks confusing with dev_id.
> 
> Just IOMMU_GET_HW_INFO for the iommu behind the specified dev_id.
> 
> then keep the structure name as iommu_hw_info.

That'd be neat.

> > With that being said, IOMMU_DEVICE_SET/UNSET_DATA should be
> > renamed to IOMMU_DEVICE_SET/UNSET_DEV_DATA -- "DEVICE" is the
> > iommu device while the "DEV_DATA" is a given device that's
> > behind the iommu.
> 
> this then becomes IOMMU_SET/UNSET_DEV_DATA.

Ack.

Thanks
Nic

Patch

diff --git a/drivers/iommu/iommufd/device.c b/drivers/iommu/iommufd/device.c
index 051bd8e99858..bc99d092de8f 100644
--- a/drivers/iommu/iommufd/device.c
+++ b/drivers/iommu/iommufd/device.c
@@ -263,6 +263,78 @@  u32 iommufd_device_to_id(struct iommufd_device *idev)
 }
 EXPORT_SYMBOL_NS_GPL(iommufd_device_to_id, IOMMUFD);
 
+static int iommufd_zero_fill_user(u64 ptr, int bytes)
+{
+	int index = 0;
+
+	for (; index < bytes; index++) {
+		if (put_user(0, (uint8_t __user *)u64_to_user_ptr(ptr + index)))
+			return -EFAULT;
+	}
+	return 0;
+}
+
+int iommufd_device_get_hw_info(struct iommufd_ucmd *ucmd)
+{
+	struct iommu_hw_info *cmd = ucmd->cmd;
+	unsigned int length = 0, data_len;
+	struct iommufd_device *idev;
+	const struct iommu_ops *ops;
+	void *data = NULL;
+	int rc = 0;
+
+	if (cmd->flags || cmd->__reserved || !cmd->data_len)
+		return -EOPNOTSUPP;
+
+	idev = iommufd_get_device(ucmd, cmd->dev_id);
+	if (IS_ERR(idev))
+		return PTR_ERR(idev);
+
+	ops = dev_iommu_ops(idev->dev);
+	if (!ops->hw_info)
+		goto done;
+
+	/* driver has hw_info callback should have a unique hw_info_type */
+	if (ops->hw_info_type == IOMMU_HW_INFO_TYPE_NONE) {
+		pr_warn_ratelimited("iommu driver set an invalid type\n");
+		rc = -ENODEV;
+		goto out_err;
+	}
+
+	data = ops->hw_info(idev->dev, &data_len);
+	if (IS_ERR(data)) {
+		rc = PTR_ERR(data);
+		goto out_err;
+	}
+
+	length = min(cmd->data_len, data_len);
+	if (copy_to_user(u64_to_user_ptr(cmd->data_ptr), data, length)) {
+		rc = -EFAULT;
+		goto out_err;
+	}
+
+	/*
+	 * Zero the trailing bytes if the user buffer is bigger than the
+	 * data size kernel actually has.
+	 */
+	if (length < cmd->data_len) {
+		rc = iommufd_zero_fill_user(cmd->data_ptr + length,
+					    cmd->data_len - length);
+		if (rc)
+			goto out_err;
+	}
+
+done:
+	cmd->data_len = length;
+	cmd->out_data_type = ops->hw_info_type;
+	rc = iommufd_ucmd_respond(ucmd, sizeof(*cmd));
+
+out_err:
+	kfree(data);
+	iommufd_put_object(&idev->obj);
+	return rc;
+}
+
 static int iommufd_group_setup_msi(struct iommufd_group *igroup,
 				   struct iommufd_hw_pagetable *hwpt)
 {
diff --git a/drivers/iommu/iommufd/iommufd_private.h b/drivers/iommu/iommufd/iommufd_private.h
index dba730129b8c..69d6bb61d387 100644
--- a/drivers/iommu/iommufd/iommufd_private.h
+++ b/drivers/iommu/iommufd/iommufd_private.h
@@ -308,6 +308,7 @@  iommufd_get_device(struct iommufd_ucmd *ucmd, u32 id)
 }
 
 void iommufd_device_destroy(struct iommufd_object *obj);
+int iommufd_device_get_hw_info(struct iommufd_ucmd *ucmd);
 
 struct iommufd_access {
 	struct iommufd_object obj;
diff --git a/drivers/iommu/iommufd/main.c b/drivers/iommu/iommufd/main.c
index 3932fe26522b..5c24e8971f09 100644
--- a/drivers/iommu/iommufd/main.c
+++ b/drivers/iommu/iommufd/main.c
@@ -269,6 +269,7 @@  static int iommufd_option(struct iommufd_ucmd *ucmd)
 union ucmd_buffer {
 	struct iommu_destroy destroy;
 	struct iommu_hwpt_alloc hwpt;
+	struct iommu_hw_info info;
 	struct iommu_ioas_alloc alloc;
 	struct iommu_ioas_allow_iovas allow_iovas;
 	struct iommu_ioas_copy ioas_copy;
@@ -302,6 +303,8 @@  static const struct iommufd_ioctl_op iommufd_ioctl_ops[] = {
 	IOCTL_OP(IOMMU_DESTROY, iommufd_destroy, struct iommu_destroy, id),
 	IOCTL_OP(IOMMU_HWPT_ALLOC, iommufd_hwpt_alloc, struct iommu_hwpt_alloc,
 		 __reserved),
+	IOCTL_OP(IOMMU_DEVICE_GET_HW_INFO, iommufd_device_get_hw_info,
+		 struct iommu_hw_info, __reserved),
 	IOCTL_OP(IOMMU_IOAS_ALLOC, iommufd_ioas_alloc_ioctl,
 		 struct iommu_ioas_alloc, out_ioas_id),
 	IOCTL_OP(IOMMU_IOAS_ALLOW_IOVAS, iommufd_ioas_allow_iovas,
diff --git a/include/uapi/linux/iommufd.h b/include/uapi/linux/iommufd.h
index 99bf0715f545..e9d42838dcbd 100644
--- a/include/uapi/linux/iommufd.h
+++ b/include/uapi/linux/iommufd.h
@@ -46,6 +46,7 @@  enum {
 	IOMMUFD_CMD_OPTION,
 	IOMMUFD_CMD_VFIO_IOAS,
 	IOMMUFD_CMD_HWPT_ALLOC,
+	IOMMUFD_CMD_DEVICE_GET_HW_INFO,
 };
 
 /**
@@ -377,4 +378,40 @@  struct iommu_hwpt_alloc {
 enum iommu_hw_info_type {
 	IOMMU_HW_INFO_TYPE_NONE,
 };
+
+/**
+ * struct iommu_hw_info - ioctl(IOMMU_DEVICE_GET_HW_INFO)
+ * @size: sizeof(struct iommu_hw_info)
+ * @flags: Must be 0
+ * @dev_id: The device bound to the iommufd
+ * @data_len: Input the length of the user buffer in bytes. Output the
+ *            length of data filled in the user buffer.
+ * @data_ptr: Pointer to the user buffer
+ * @out_data_type: Output the iommu hardware info type as defined by
+ *                 enum iommu_hw_info_type.
+ * @__reserved: Must be 0
+ *
+ * Query the hardware iommu information for given device which has been
+ * bound to iommufd. @data_len is the size of the buffer which captures
+ * iommu type specific data and the data will be filled. Trailing bytes
+ * are zeroed if the user buffer is larger than the data kernel has.
+ *
+ * The type specific data would be used to sync capability between the
+ * virtual IOMMU and the hardware IOMMU. e.g. nested translation requires
+ * to check the hardware IOMMU capability so guest stage-1 page table
+ * uses a format compatible to the hardware IOMMU.
+ *
+ * The @out_data_type will be filled if the ioctl succeeds. It would
+ * be used to decode the data filled in the buffer pointed by @data_ptr.
+ */
+struct iommu_hw_info {
+	__u32 size;
+	__u32 flags;
+	__u32 dev_id;
+	__u32 data_len;
+	__aligned_u64 data_ptr;
+	__u32 out_data_type;
+	__u32 __reserved;
+};
+#define IOMMU_DEVICE_GET_HW_INFO _IO(IOMMUFD_TYPE, IOMMUFD_CMD_DEVICE_GET_HW_INFO)
 #endif