Skip to contentSkip to conversation

Update vfio_pin/unpin_pages API

PATCH Superseded Nicolin Chen 7 participants 51 messages Jul 6, 2022 – Jul 8, 2022

Subsystems (12)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v2 cover Update vfio_pin/unpin_pages API

This is a preparatory series for IOMMUFD v2 patches. It prepares for replacing vfio_iommu_type1 implementations of vfio_pin/unpin_pages() with IOMMUFD version.

There's a gap between these two versions: the vfio_iommu_type1 version inputs a non-contiguous PFN list and outputs another PFN list for the pinned physical page list, while the IOMMUFD version only supports a contiguous address input by accepting the starting IO virtual address of a set of pages to pin and by outputting to a physical page list.

The nature of existing callers mostly aligns with the IOMMUFD version, except s390's vfio_ccw_cp code where some additional change is needed along with this series. Overall, updating to "iova" and "phys_page" does improve the caller side to some extent.

Also fix a misuse of physical address and virtual address in the s390's crypto code. And update the input naming at the adjacent vfio_dma_rw().

This is on github:
https://github.com/nicolinc/iommufd/commits/vfio_pin_pages

Request for testing: I only did build for s390 and i915 code, so it'd be nice to have people who have environment to run sanity accordingly.

Thanks!

Changelog
v2:
* Added a patch to make vfio_unpin_pages return void
* Added two patches to remove PFN list from two s390 callers
* Renamed "phys_page" parameter to "pages" for vfio_pin_pages
* Updated commit log of kmap_local_page() patch
* Added Harald's "Reviewed-by" to pa_ind patch
* Rebased on top of Alex's extern removal path
v1: https://lore.kernel.org/kvm/[email protected]/

Nicolin Chen (9):
vfio: Make vfio_unpin_pages() return void vfio/ap: Pass in physical address of ind to ap_aqic() vfio/ccw: Only pass in contiguous pages
vfio: Pass in starting IOVA to vfio_pin/unpin_pages API vfio/ap: Remove redundant pfn
vfio/ccw: Change pa_pfn list to pa_iova list
vfio: Rename user_iova of vfio_dma_rw() vfio/ccw: Add kmap_local_page() for memcpy
vfio: Replace phys_pfn with pages for vfio_pin_pages()

 .../driver-api/vfio-mediated-device.rst       |   6 +-
 arch/s390/include/asm/ap.h                    |   6 +-
 drivers/gpu/drm/i915/gvt/kvmgt.c              |  46 ++---
 drivers/s390/cio/vfio_ccw_cp.c                | 195 +++++++++++-------
 drivers/s390/crypto/ap_queue.c                |   2 +-
 drivers/s390/crypto/vfio_ap_ops.c             |  54 +++--
 drivers/s390/crypto/vfio_ap_private.h         |   4 +-
 drivers/vfio/vfio.c                           |  55 +++--
 drivers/vfio/vfio.h                           |   8 +-
 drivers/vfio/vfio_iommu_type1.c               |  46 +++--
 include/linux/vfio.h                          |   9 +-
 11 files changed, 218 insertions(+), 213 deletions(-)
Signature · 1 line
2.17.1
#1 lore raw

PATCH v2 1/9 vfio: Make vfio_unpin_pages() return void

There's only one caller that checks its return value with a WARN_ON_ONCE, while all other callers do not check return value at all. So simplify the API to return void by embedding similar WARN_ON_ONCEs.

---
 .../driver-api/vfio-mediated-device.rst       |  2 +-
 drivers/gpu/drm/i915/gvt/kvmgt.c              |  5 +---
 drivers/vfio/vfio.c                           | 24 ++++++++-----------
 drivers/vfio/vfio.h                           |  2 +-
 drivers/vfio/vfio_iommu_type1.c               | 16 ++++++-------
 include/linux/vfio.h                          |  4 ++--
 6 files changed, 23 insertions(+), 30 deletions(-)
Documentation/driver-api/vfio-mediated-device.rst +1 -1
index 1c57815619fd..b0fdf76b339a 100644
@@ -265,7 +265,7 @@ driver::
 	int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
 				  int npage, int prot, unsigned long *phys_pfn);
 
-	int vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
+	void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
 				    int npage);
 
 These functions call back into the back-end IOMMU module by using the pin_pages
drivers/gpu/drm/i915/gvt/kvmgt.c +1 -4
index e2f6c56ab342..8c67c9aba82d 100644
@@ -231,18 +231,15 @@ static void intel_gvt_cleanup_vgpu_type_groups(struct intel_gvt *gvt)
 static void gvt_unpin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
 		unsigned long size)
 {
-	struct drm_i915_private *i915 = vgpu->gvt->gt->i915;
 	int total_pages;
 	int npage;
-	int ret;
 
 	total_pages = roundup(size, PAGE_SIZE) / PAGE_SIZE;
 
 	for (npage = 0; npage < total_pages; npage++) {
 		unsigned long cur_gfn = gfn + npage;
 
-		ret = vfio_unpin_pages(&vgpu->vfio_device, &cur_gfn, 1);
-		drm_WARN_ON(&i915->drm, ret != 1);
+		vfio_unpin_pages(&vgpu->vfio_device, &cur_gfn, 1);
 	}
 }
 
drivers/vfio/vfio.c +10 -14
index 61e71c1154be..01f45ec70a3d 100644
@@ -1959,31 +1959,27 @@ EXPORT_SYMBOL(vfio_pin_pages);
  *		   PFNs should not be greater than VFIO_PIN_PAGES_MAX_ENTRIES.
  * @npage [in]   : count of elements in user_pfn array.  This count should not
  *                 be greater than VFIO_PIN_PAGES_MAX_ENTRIES.
- * Return error or number of pages unpinned.
  */
-int vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
-		     int npage)
+void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
+		      int npage)
 {
 	struct vfio_container *container;
 	struct vfio_iommu_driver *driver;
-	int ret;
 
-	if (!user_pfn || !npage || !vfio_assert_device_open(device))
-		return -EINVAL;
+	if (WARN_ON_ONCE(!user_pfn || !npage || !vfio_assert_device_open(device)))
+		return;
 
-	if (npage > VFIO_PIN_PAGES_MAX_ENTRIES)
-		return -E2BIG;
+	if (WARN_ON_ONCE(npage > VFIO_PIN_PAGES_MAX_ENTRIES))
+		return;
 
 	/* group->container cannot change while a vfio device is open */
 	container = device->group->container;
 	driver = container->iommu_driver;
-	if (likely(driver && driver->ops->unpin_pages))
-		ret = driver->ops->unpin_pages(container->iommu_data, user_pfn,
-					       npage);
-	else
-		ret = -ENOTTY;
 
-	return ret;
+	if (WARN_ON_ONCE(unlikely(!driver || !driver->ops->unpin_pages)))
+		return;
+
+	driver->ops->unpin_pages(container->iommu_data, user_pfn, npage);
 }
 EXPORT_SYMBOL(vfio_unpin_pages);
 
drivers/vfio/vfio.h +1 -1
index a67130221151..bef4edf58138 100644
@@ -53,7 +53,7 @@ struct vfio_iommu_driver_ops {
 				     unsigned long *user_pfn,
 				     int npage, int prot,
 				     unsigned long *phys_pfn);
-	int		(*unpin_pages)(void *iommu_data,
+	void		(*unpin_pages)(void *iommu_data,
 				       unsigned long *user_pfn, int npage);
 	int		(*register_notifier)(void *iommu_data,
 					     unsigned long *events,
drivers/vfio/vfio_iommu_type1.c +8 -8
index c13b9290e357..08613edaf722 100644
@@ -948,20 +948,19 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
 	return ret;
 }
 
-static int vfio_iommu_type1_unpin_pages(void *iommu_data,
-					unsigned long *user_pfn,
-					int npage)
+static void vfio_iommu_type1_unpin_pages(void *iommu_data,
+					 unsigned long *user_pfn, int npage)
 {
 	struct vfio_iommu *iommu = iommu_data;
 	bool do_accounting;
 	int i;
 
-	if (!iommu || !user_pfn || npage <= 0)
-		return -EINVAL;
+	if (WARN_ON_ONCE(!iommu || !user_pfn || npage <= 0))
+		return;
 
 	/* Supported for v2 version only */
-	if (!iommu->v2)
-		return -EACCES;
+	if (WARN_ON_ONCE(!iommu->v2))
+		return;
 
 	mutex_lock(&iommu->lock);
 
@@ -979,7 +978,8 @@ static int vfio_iommu_type1_unpin_pages(void *iommu_data,
 	}
 
 	mutex_unlock(&iommu->lock);
-	return i > 0 ? i : -EINVAL;
+
+	WARN_ON_ONCE(i != npage);
 }
 
 static long vfio_sync_unpin(struct vfio_dma *dma, struct vfio_domain *domain,
include/linux/vfio.h +2 -2
index 49580fa2073a..d0844ecdc961 100644
@@ -149,8 +149,8 @@ bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
 
 int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
 		   int npage, int prot, unsigned long *phys_pfn);
-int vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
-		     int npage);
+void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
+		      int npage);
 int vfio_dma_rw(struct vfio_device *device, dma_addr_t user_iova,
 		void *data, size_t len, bool write);
 
Signature · 1 line
2.17.1
Reviewed-by on v2 1/9
#2 lore raw

Re: [RFT][PATCH v2 1/9] vfio: Make vfio_unpin_pages() return void

drivers/vfio/vfio.c EXPORT_SYMBOL(vfio_pin_pages); · line 1969

Quoted 10 lines
drivers/vfio/vfio.c +3 -3
+void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
+		      int npage)
 {
 	struct vfio_container *container;
 	struct vfio_iommu_driver *driver;
-	int ret;
 
-	if (!user_pfn || !npage || !vfio_assert_device_open(device))
-		return -EINVAL;
+	if (WARN_ON_ONCE(!user_pfn || !npage || !vfio_assert_device_open(device)))

This adds an overly long line. Note that I think in general it is better to have an individual WARN_ON per condition anyway, as that allows to directly pinpoint what went wrong when it triggers.

drivers/vfio/vfio.c EXPORT_SYMBOL(vfio_pin_pages); · line 1980

Quoted 2 lines
drivers/vfio/vfio.c +2 -0
+	if (WARN_ON_ONCE(unlikely(!driver || !driver->ops->unpin_pages)))
+		return;

I'd just skip this check an let it crash. If someone calls unpin on something totally random that wasn't even pinned we don't need to handle that gracefully.

Otherwise looks good:

#3 lore raw

Re: [RFT][PATCH v2 1/9] vfio: Make vfio_unpin_pages() return void

Quoted 7 lines · Christoph Hellwig
Quoted 7 lines
drivers/vfio/vfio.c +2 -1
+void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
+		      int npage)
 {
 	struct vfio_container *container;
 	struct vfio_iommu_driver *driver;
-	int ret;
Quoted 7 lines

drivers/vfio/vfio.c EXPORT_SYMBOL(vfio_pin_pages); · line 1969

Quoted 3 lines
drivers/vfio/vfio.c +1 -2
-	if (!user_pfn || !npage || !vfio_assert_device_open(device))
-		return -EINVAL;
+	if (WARN_ON_ONCE(!user_pfn || !npage || !vfio_assert_device_open(device)))

This adds an overly long line. Note that I think in general it is better to have an individual WARN_ON per condition anyway, as that allows to directly pinpoint what went wrong when it triggers.

Following patches are touching this line too. And it'll be shrunk to a shorter line eventually by the end of PATCH-9.

Yet, I can separate them as you pointed out.

Quoted 6 lines

drivers/vfio/vfio.c EXPORT_SYMBOL(vfio_pin_pages); · line 1980

Quoted 2 lines
drivers/vfio/vfio.c +2 -0
+	if (WARN_ON_ONCE(unlikely(!driver || !driver->ops->unpin_pages)))
+		return;

I'd just skip this check an let it crash. If someone calls unpin on something totally random that wasn't even pinned we don't need to handle that gracefully.

Makes sense. I can drop that in next version.

Quoted 1 line

Will add to v3. Thanks for the review!

Reviewed-by on v2 1/9
#4 lore raw

Re: [RFT][PATCH v2 1/9] vfio: Make vfio_unpin_pages() return void

In reply to Nicolin Chen
Quoted 6 lines · Nicolin Chen

There's only one caller that checks its return value with a WARN_ON_ONCE, while all other callers do not check return value at all. So simplify the API to return void by embedding similar WARN_ON_ONCEs.

Quoted 8 lines
---
 .../driver-api/vfio-mediated-device.rst       |  2 +-
 drivers/gpu/drm/i915/gvt/kvmgt.c              |  5 +---
 drivers/vfio/vfio.c                           | 24 ++++++++-----------
 drivers/vfio/vfio.h                           |  2 +-
 drivers/vfio/vfio_iommu_type1.c               | 16 ++++++-------
 include/linux/vfio.h                          |  4 ++--
 6 files changed, 23 insertions(+), 30 deletions(-)

Jason

maintainer Reviewed-by on v2 1/9
#5 lore raw

Re: [RFT][PATCH v2 1/9] vfio: Make vfio_unpin_pages() return void

In reply to Nicolin Chen
Quoted 167 lines · Nicolin Chen

There's only one caller that checks its return value with a WARN_ON_ONCE, while all other callers do not check return value at all. So simplify the API to return void by embedding similar WARN_ON_ONCEs.

---
  .../driver-api/vfio-mediated-device.rst       |  2 +-
  drivers/gpu/drm/i915/gvt/kvmgt.c              |  5 +---
  drivers/vfio/vfio.c                           | 24 ++++++++-----------
  drivers/vfio/vfio.h                           |  2 +-
  drivers/vfio/vfio_iommu_type1.c               | 16 ++++++-------
  include/linux/vfio.h                          |  4 ++--
  6 files changed, 23 insertions(+), 30 deletions(-)
Documentation/driver-api/vfio-mediated-device.rst +1 -1
index 1c57815619fd..b0fdf76b339a 100644
@@ -265,7 +265,7 @@ driver::
  	int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
  				  int npage, int prot, unsigned long *phys_pfn);
 
-	int vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
+	void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
  				    int npage);
 
  These functions call back into the back-end IOMMU module by using the pin_pages
drivers/gpu/drm/i915/gvt/kvmgt.c +1 -4
index e2f6c56ab342..8c67c9aba82d 100644
@@ -231,18 +231,15 @@ static void intel_gvt_cleanup_vgpu_type_groups(struct intel_gvt *gvt)
  static void gvt_unpin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
  		unsigned long size)
  {
-	struct drm_i915_private *i915 = vgpu->gvt->gt->i915;
  	int total_pages;
  	int npage;
-	int ret;
 
  	total_pages = roundup(size, PAGE_SIZE) / PAGE_SIZE;
 
  	for (npage = 0; npage < total_pages; npage++) {
  		unsigned long cur_gfn = gfn + npage;
 
-		ret = vfio_unpin_pages(&vgpu->vfio_device, &cur_gfn, 1);
-		drm_WARN_ON(&i915->drm, ret != 1);
+		vfio_unpin_pages(&vgpu->vfio_device, &cur_gfn, 1);
  	}
  }
 
drivers/vfio/vfio.c +10 -14
index 61e71c1154be..01f45ec70a3d 100644
@@ -1959,31 +1959,27 @@ EXPORT_SYMBOL(vfio_pin_pages);
   *		   PFNs should not be greater than VFIO_PIN_PAGES_MAX_ENTRIES.
   * @npage [in]   : count of elements in user_pfn array.  This count should not
   *                 be greater than VFIO_PIN_PAGES_MAX_ENTRIES.
- * Return error or number of pages unpinned.
   */
-int vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
-		     int npage)
+void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
+		      int npage)
  {
  	struct vfio_container *container;
  	struct vfio_iommu_driver *driver;
-	int ret;
 
-	if (!user_pfn || !npage || !vfio_assert_device_open(device))
-		return -EINVAL;
+	if (WARN_ON_ONCE(!user_pfn || !npage || !vfio_assert_device_open(device)))
+		return;
 
-	if (npage > VFIO_PIN_PAGES_MAX_ENTRIES)
-		return -E2BIG;
+	if (WARN_ON_ONCE(npage > VFIO_PIN_PAGES_MAX_ENTRIES))
+		return;
 
  	/* group->container cannot change while a vfio device is open */
  	container = device->group->container;
  	driver = container->iommu_driver;
-	if (likely(driver && driver->ops->unpin_pages))
-		ret = driver->ops->unpin_pages(container->iommu_data, user_pfn,
-					       npage);
-	else
-		ret = -ENOTTY;
 
-	return ret;
+	if (WARN_ON_ONCE(unlikely(!driver || !driver->ops->unpin_pages)))
+		return;
+
+	driver->ops->unpin_pages(container->iommu_data, user_pfn, npage);
  }
  EXPORT_SYMBOL(vfio_unpin_pages);
 
drivers/vfio/vfio.h +1 -1
index a67130221151..bef4edf58138 100644
@@ -53,7 +53,7 @@ struct vfio_iommu_driver_ops {
  				     unsigned long *user_pfn,
  				     int npage, int prot,
  				     unsigned long *phys_pfn);
-	int		(*unpin_pages)(void *iommu_data,
+	void		(*unpin_pages)(void *iommu_data,
  				       unsigned long *user_pfn, int npage);
  	int		(*register_notifier)(void *iommu_data,
  					     unsigned long *events,
drivers/vfio/vfio_iommu_type1.c +8 -8
index c13b9290e357..08613edaf722 100644
@@ -948,20 +948,19 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
  	return ret;
  }
 
-static int vfio_iommu_type1_unpin_pages(void *iommu_data,
-					unsigned long *user_pfn,
-					int npage)
+static void vfio_iommu_type1_unpin_pages(void *iommu_data,
+					 unsigned long *user_pfn, int npage)
  {
  	struct vfio_iommu *iommu = iommu_data;
  	bool do_accounting;
  	int i;
 
-	if (!iommu || !user_pfn || npage <= 0)
-		return -EINVAL;
+	if (WARN_ON_ONCE(!iommu || !user_pfn || npage <= 0))
+		return;
 
  	/* Supported for v2 version only */
-	if (!iommu->v2)
-		return -EACCES;
+	if (WARN_ON_ONCE(!iommu->v2))
+		return;
 
  	mutex_lock(&iommu->lock);
 
@@ -979,7 +978,8 @@ static int vfio_iommu_type1_unpin_pages(void *iommu_data,
  	}
 
  	mutex_unlock(&iommu->lock);
-	return i > 0 ? i : -EINVAL;
+
+	WARN_ON_ONCE(i != npage);
  }
 
  static long vfio_sync_unpin(struct vfio_dma *dma, struct vfio_domain *domain,
include/linux/vfio.h +2 -2
index 49580fa2073a..d0844ecdc961 100644
@@ -149,8 +149,8 @@ bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
 
  int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
  		   int npage, int prot, unsigned long *phys_pfn);
-int vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
-		     int npage);
+void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
+		      int npage);
  int vfio_dma_rw(struct vfio_device *device, dma_addr_t user_iova,
  		void *data, size_t len, bool write);
 
#6 lore raw

RE: [RFT][PATCH v2 1/9] vfio: Make vfio_unpin_pages() return void

In reply to Nicolin Chen
Quoted 6 lines

From: Nicolin Chen <[email protected]>
Sent: Wednesday, July 6, 2022 2:28 PM

There's only one caller that checks its return value with a WARN_ON_ONCE, while all other callers do not check return value at all. So simplify the API to return void by embedding similar WARN_ON_ONCEs.

While this change keeps the similar effect as before it leads to different policy for same type of errors between pin and unpin paths:

e.g.

vfio_unpin_pages():

	if (WARN_ON_ONCE(!user_pfn || !npage || !vfio_assert_device_open(device)))
		return;

vfio_pin_pages():

	if (!user_pfn || !phys_pfn || !npage ||
	    !vfio_assert_device_open(device))
		return -EINVAL;

It sounds a bit weird when reading related code...

#7 lore raw

Re: [RFT][PATCH v2 1/9] vfio: Make vfio_unpin_pages() return void

Quoted 6 lines · Kevin

External email: Use caution opening links or attachments

Quoted 3 lines

From: Nicolin Chen <[email protected]>
Sent: Wednesday, July 6, 2022 2:28 PM

Quoted 6 lines
Quoted 3 lines

There's only one caller that checks its return value with a WARN_ON_ONCE, while all other callers do not check return value at all. So simplify the API to return void by embedding similar WARN_ON_ONCEs.

While this change keeps the similar effect as before it leads to different policy for same type of errors between pin and unpin paths:

I think it's because of the policy that an undo function should not fail. Meanwhile, indulging faulty inputs isn't good either.

Quoted 6 lines

e.g.

vfio_unpin_pages():

        if (WARN_ON_ONCE(!user_pfn || !npage || !vfio_assert_device_open(device)))
                return;
Quoted 6 lines

vfio_pin_pages():

        if (!user_pfn || !phys_pfn || !npage ||
            !vfio_assert_device_open(device))
                return -EINVAL;

It sounds a bit weird when reading related code...

Any better way to handle this?

#8 lore raw

Re: [RFT][PATCH v2 1/9] vfio: Make vfio_unpin_pages() return void

Quoted 23 lines · Nicolin Chen
Quoted 6 lines · Kevin

External email: Use caution opening links or attachments

Quoted 3 lines

From: Nicolin Chen <[email protected]>
Sent: Wednesday, July 6, 2022 2:28 PM

Quoted 6 lines
Quoted 3 lines

There's only one caller that checks its return value with a WARN_ON_ONCE, while all other callers do not check return value at all. So simplify the API to return void by embedding similar WARN_ON_ONCEs.

While this change keeps the similar effect as before it leads to different policy for same type of errors between pin and unpin paths:

I think it's because of the policy that an undo function should not fail. Meanwhile, indulging faulty inputs isn't good either.

Quoted 6 lines

e.g.

vfio_unpin_pages():

        if (WARN_ON_ONCE(!user_pfn || !npage || !vfio_assert_device_open(device)))
                return;
Quoted 8 lines
Quoted 6 lines

vfio_pin_pages():

        if (!user_pfn || !phys_pfn || !npage ||
            !vfio_assert_device_open(device))
                return -EINVAL;

It sounds a bit weird when reading related code...

Any better way to handle this?

They should all be WARN_ON's, that is the standard pattern to assert that function arguments must be correctly formed.

I would also drop the tests that obviously will oops on their on anyone, like NULL pointer checks. This is a semi-performance path.

Jason

#9 lore raw

Re: [RFT][PATCH v2 1/9] vfio: Make vfio_unpin_pages() return void

Quoted 29 lines · Jason Gunthorpe
Quoted 28 lines · Nicolin Chen
Quoted 6 lines · Kevin

External email: Use caution opening links or attachments

Quoted 3 lines

From: Nicolin Chen <[email protected]>
Sent: Wednesday, July 6, 2022 2:28 PM

Quoted 6 lines
Quoted 3 lines

There's only one caller that checks its return value with a WARN_ON_ONCE, while all other callers do not check return value at all. So simplify the API to return void by embedding similar WARN_ON_ONCEs.

While this change keeps the similar effect as before it leads to different policy for same type of errors between pin and unpin paths:

I think it's because of the policy that an undo function should not fail. Meanwhile, indulging faulty inputs isn't good either.

Quoted 11 lines

e.g.

vfio_unpin_pages():

        if (WARN_ON_ONCE(!user_pfn || !npage || !vfio_assert_device_open(device)))
                return;

vfio_pin_pages():

        if (!user_pfn || !phys_pfn || !npage ||
            !vfio_assert_device_open(device))
                return -EINVAL;
Quoted 6 lines
Quoted 3 lines
Quoted 1 line

It sounds a bit weird when reading related code...

Any better way to handle this?

They should all be WARN_ON's, that is the standard pattern to assert that function arguments must be correctly formed.

OK. I can change that. I assume that, not confined to arguments, we might want to have a WARN_ON for the return value check also.

Quoted 2 lines

I would also drop the tests that obviously will oops on their on anyone, like NULL pointer checks. This is a semi-performance path.

OK. I will simply remove those NULL pointer checks. Actually, that !user_pfn check is gone anyway in the following patch, as user_pfn is replaced with iova.

#10 lore raw

PATCH v2 2/9 vfio/ap: Pass in physical address of ind to ap_aqic()

In reply to Nicolin Chen

The ap_aqic() is called by vfio_ap_irq_enable() where it passes in a virt value that's casted from a physical address "h_nib". Inside the ap_aqic(), it does virt_to_phys() again.

Since ap_aqic() needs a physical address, let's just pass in a pa of ind directly. So change the "ind" to "pa_ind".

---
 arch/s390/include/asm/ap.h        | 6 +++---
 drivers/s390/crypto/ap_queue.c    | 2 +-
 drivers/s390/crypto/vfio_ap_ops.c | 7 ++++---
 3 files changed, 8 insertions(+), 7 deletions(-)
arch/s390/include/asm/ap.h +3 -3
index b515cfa62bd9..f508f5025e38 100644
@@ -227,13 +227,13 @@ struct ap_qirq_ctrl {
  * ap_aqic(): Control interruption for a specific AP.
  * @qid: The AP queue number
  * @qirqctrl: struct ap_qirq_ctrl (64 bit value)
- * @ind: The notification indicator byte
+ * @pa_ind: Physical address of the notification indicator byte
  *
  * Returns AP queue status.
  */
 static inline struct ap_queue_status ap_aqic(ap_qid_t qid,
 					     struct ap_qirq_ctrl qirqctrl,
-					     void *ind)
+					     phys_addr_t pa_ind)
 {
 	unsigned long reg0 = qid | (3UL << 24);  /* fc 3UL is AQIC */
 	union {
@@ -241,7 +241,7 @@ static inline struct ap_queue_status ap_aqic(ap_qid_t qid,
 		struct ap_qirq_ctrl qirqctrl;
 		struct ap_queue_status status;
 	} reg1;
-	unsigned long reg2 = virt_to_phys(ind);
+	unsigned long reg2 = pa_ind;
 
 	reg1.qirqctrl = qirqctrl;
 
drivers/s390/crypto/ap_queue.c +1 -1
index c48b0db824e3..a32457b4cbb8 100644
@@ -34,7 +34,7 @@ static int ap_queue_enable_irq(struct ap_queue *aq, void *ind)
 
 	qirqctrl.ir = 1;
 	qirqctrl.isc = AP_ISC;
-	status = ap_aqic(aq->qid, qirqctrl, ind);
+	status = ap_aqic(aq->qid, qirqctrl, virt_to_phys(ind));
 	switch (status.response_code) {
 	case AP_RESPONSE_NORMAL:
 	case AP_RESPONSE_OTHERWISE_CHANGED:
drivers/s390/crypto/vfio_ap_ops.c +4 -3
index a7d2a95796d3..bb869b28cebd 100644
@@ -154,7 +154,7 @@ static struct ap_queue_status vfio_ap_irq_disable(struct vfio_ap_queue *q)
 	int retries = 5;
 
 	do {
-		status = ap_aqic(q->apqn, aqic_gisa, NULL);
+		status = ap_aqic(q->apqn, aqic_gisa, 0);
 		switch (status.response_code) {
 		case AP_RESPONSE_OTHERWISE_CHANGED:
 		case AP_RESPONSE_NORMAL:
@@ -245,7 +245,8 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
 	struct kvm_s390_gisa *gisa;
 	int nisc;
 	struct kvm *kvm;
-	unsigned long h_nib, g_pfn, h_pfn;
+	unsigned long g_pfn, h_pfn;
+	phys_addr_t h_nib;
 	int ret;
 
 	/* Verify that the notification indicator byte address is valid */
@@ -290,7 +291,7 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
 	aqic_gisa.ir = 1;
 	aqic_gisa.gisa = (uint64_t)gisa >> 4;
 
-	status = ap_aqic(q->apqn, aqic_gisa, (void *)h_nib);
+	status = ap_aqic(q->apqn, aqic_gisa, h_nib);
 	switch (status.response_code) {
 	case AP_RESPONSE_NORMAL:
 		/* See if we did clear older IRQ configuration */
Signature · 1 line
2.17.1
Reviewed-by on v2 2/9
#11 lore raw

Re: [RFT][PATCH v2 2/9] vfio/ap: Pass in physical address of ind to ap_aqic()

Quoted 7 lines · Nicolin Chen

The ap_aqic() is called by vfio_ap_irq_enable() where it passes in a virt value that's casted from a physical address "h_nib". Inside the ap_aqic(), it does virt_to_phys() again.

Since ap_aqic() needs a physical address, let's just pass in a pa of ind directly. So change the "ind" to "pa_ind".

Quoted 7 lines
---
 arch/s390/include/asm/ap.h        | 6 +++---
 drivers/s390/crypto/ap_queue.c    | 2 +-
 drivers/s390/crypto/vfio_ap_ops.c | 7 ++++---
 3 files changed, 8 insertions(+), 7 deletions(-)

Jason

#12 lore raw

PATCH v2 3/9 vfio/ccw: Only pass in contiguous pages

In reply to Nicolin Chen

This driver is the only caller of vfio_pin/unpin_pages that might pass in a non-contiguous PFN list, but in many cases it has a contiguous PFN list to process. So letting VFIO API handle a non-contiguous PFN list is actually counterproductive.

Add a pair of simple loops to pass in contiguous PFNs only, to have an efficient implementation in VFIO.

---
 drivers/s390/cio/vfio_ccw_cp.c | 70 +++++++++++++++++++++++++++-------
 1 file changed, 56 insertions(+), 14 deletions(-)
drivers/s390/cio/vfio_ccw_cp.c +56 -14
index 0c2be9421ab7..3b94863ad24e 100644
@@ -90,6 +90,38 @@ static int pfn_array_alloc(struct pfn_array *pa, u64 iova, unsigned int len)
 	return 0;
 }
 
+/*
+ * pfn_array_unpin() - Unpin user pages in memory
+ * @pa: pfn_array on which to perform the operation
+ * @vdev: the vfio device to perform the operation
+ * @pa_nr: number of user pages to unpin
+ *
+ * Only unpin if any pages were pinned to begin with, i.e. pa_nr > 0,
+ * otherwise only clear pa->pa_nr
+ */
+static void pfn_array_unpin(struct pfn_array *pa,
+			    struct vfio_device *vdev, int pa_nr)
+{
+	int unpinned = 0, npage = 1;
+
+	while (unpinned < pa_nr) {
+		unsigned long *first = &pa->pa_iova_pfn[unpinned];
+		unsigned long *last = &first[npage];
+
+		if (unpinned + npage < pa_nr &&
+		    *first + npage == *last) {
+			npage++;
+			continue;
+		}
+
+		vfio_unpin_pages(vdev, first, npage);
+		unpinned += npage;
+		npage = 1;
+	}
+
+	pa->pa_nr = 0;
+}
+
 /*
  * pfn_array_pin() - Pin user pages in memory
  * @pa: pfn_array on which to perform the operation
@@ -101,34 +133,44 @@ static int pfn_array_alloc(struct pfn_array *pa, u64 iova, unsigned int len)
  */
 static int pfn_array_pin(struct pfn_array *pa, struct vfio_device *vdev)
 {
+	int pinned = 0, npage = 1;
 	int ret = 0;
 
-	ret = vfio_pin_pages(vdev, pa->pa_iova_pfn, pa->pa_nr,
-			     IOMMU_READ | IOMMU_WRITE, pa->pa_pfn);
+	while (pinned < pa->pa_nr) {
+		unsigned long *first = &pa->pa_iova_pfn[pinned];
+		unsigned long *last = &first[npage];
 
-	if (ret < 0) {
-		goto err_out;
-	} else if (ret > 0 && ret != pa->pa_nr) {
-		vfio_unpin_pages(vdev, pa->pa_iova_pfn, ret);
-		ret = -EINVAL;
-		goto err_out;
+		if (pinned + npage < pa->pa_nr &&
+		    *first + npage == *last) {
+			npage++;
+			continue;
+		}
+
+		ret = vfio_pin_pages(vdev, first, npage,
+				     IOMMU_READ | IOMMU_WRITE,
+				     &pa->pa_pfn[pinned]);
+		if (ret < 0) {
+			goto err_out;
+		} else if (ret > 0 && ret != npage) {
+			pinned += ret;
+			ret = -EINVAL;
+			goto err_out;
+		}
+		pinned += npage;
+		npage = 1;
 	}
 
 	return ret;
 
 err_out:
-	pa->pa_nr = 0;
-
+	pfn_array_unpin(pa, vdev, pinned);
 	return ret;
 }
 
 /* Unpin the pages before releasing the memory. */
 static void pfn_array_unpin_free(struct pfn_array *pa, struct vfio_device *vdev)
 {
-	/* Only unpin if any pages were pinned to begin with */
-	if (pa->pa_nr)
-		vfio_unpin_pages(vdev, pa->pa_iova_pfn, pa->pa_nr);
-	pa->pa_nr = 0;
+	pfn_array_unpin(pa, vdev, pa->pa_nr);
 	kfree(pa->pa_iova_pfn);
 }
 
Signature · 1 line
2.17.1
Reviewed-by on v2 3/9
#13 lore raw

Re: [RFT][PATCH v2 3/9] vfio/ccw: Only pass in contiguous pages

Quoted 5 lines · Nicolin Chen

This driver is the only caller of vfio_pin/unpin_pages that might pass in a non-contiguous PFN list, but in many cases it has a contiguous PFN list to process. So letting VFIO API handle a non-contiguous PFN list is actually counterproductive.

Quoted 7 lines

Add a pair of simple loops to pass in contiguous PFNs only, to have an efficient implementation in VFIO.

---
 drivers/s390/cio/vfio_ccw_cp.c | 70 +++++++++++++++++++++++++++-------
 1 file changed, 56 insertions(+), 14 deletions(-)

I think this is fine as-is for this series, but someone who knows and can test ccw should go in and fix things so that pfn_array_alloc() doesn't exist. Allocating memory and filling it with consecutive integers is kind of silly given we can just call vfio_pin_pages() with pa_nr directly.

	pa->pa_iova_pfn[0] = pa->pa_iova >> PAGE_SHIFT;
	pa->pa_pfn[0] = -1ULL;
	for (i = 1; i < pa->pa_nr; i++) {
		pa->pa_iova_pfn[i] = pa->pa_iova_pfn[i - 1] + 1;

It looks like only the 'ccw_is_idal' flow can actually create non-continuities. Also the loop in copy_from_iova() should ideally be using the much faster 'rw' interface, and not a pin/unpin cycle just to memcpy.

If I guess right these changes would significantly speed this driver up.

Anyhow,

Jason

maintainer
#14 lore raw

Re: [RFT][PATCH v2 3/9] vfio/ccw: Only pass in contiguous pages

Quoted 26 lines · Jason Gunthorpe
Quoted 12 lines · Nicolin Chen

This driver is the only caller of vfio_pin/unpin_pages that might pass
in a non-contiguous PFN list, but in many cases it has a contiguous PFN
list to process. So letting VFIO API handle a non-contiguous PFN list
is actually counterproductive.

Add a pair of simple loops to pass in contiguous PFNs only, to have an
efficient implementation in VFIO.

Quoted 5 lines
---
 drivers/s390/cio/vfio_ccw_cp.c | 70 +++++++++++++++++++++++++++---
----
 1 file changed, 56 insertions(+), 14 deletions(-)

I think this is fine as-is for this series, but someone who knows and can test ccw should go in and fix things so that pfn_array_alloc() doesn't exist. Allocating memory and filling it with consecutive integers is kind of silly given we can just call vfio_pin_pages() with
pa_nr directly.

Quoted 7 lines
	pa->pa_iova_pfn[0] = pa->pa_iova >> PAGE_SHIFT;
	pa->pa_pfn[0] = -1ULL;
	for (i = 1; i < pa->pa_nr; i++) {
		pa->pa_iova_pfn[i] = pa->pa_iova_pfn[i - 1] + 1;

It looks like only the 'ccw_is_idal' flow can actually create non-continuities.

This code is certainly not my favorite, but you're right that it's the IDAL flow that generates the non-contiguous requests and the code you reference is simply an initialization for the !IDAL case. As I have a todo in this code anyway, I'll register your suggestion to see if they can be untangled.

Quoted 3 lines

Also the loop in copy_from_iova() should ideally be
using the much faster 'rw' interface, and not a pin/unpin cycle just to memcpy.

I guess I missed when that was added. This looks like low hanging fruit for some ollllld code regardless of the above. Will get to this once I'm back. Thank you!

Eric

Quoted 9 lines

If I guess right these changes would significantly speed this driver up.

Anyhow,

Jason

maintainer Reviewed-by on v2 3/9
#15 lore raw

Re: [RFT][PATCH v2 3/9] vfio/ccw: Only pass in contiguous pages

In reply to Nicolin Chen
Quoted 7 lines · Nicolin Chen

This driver is the only caller of vfio_pin/unpin_pages that might pass
in a non-contiguous PFN list, but in many cases it has a contiguous PFN
list to process. So letting VFIO API handle a non-contiguous PFN list is actually counterproductive.

Quoted 5 lines

Add a pair of simple loops to pass in contiguous PFNs only, to have an
efficient implementation in VFIO.

Quoted 113 lines
---
 drivers/s390/cio/vfio_ccw_cp.c | 70 +++++++++++++++++++++++++++-----
--
 1 file changed, 56 insertions(+), 14 deletions(-)
drivers/s390/cio/vfio_ccw_cp.c +32 -0
index 0c2be9421ab7..3b94863ad24e 100644
@@ -90,6 +90,38 @@ static int pfn_array_alloc(struct pfn_array *pa,
 u64 iova, unsigned int len)
 	return 0;
 }
 
+/*
+ * pfn_array_unpin() - Unpin user pages in memory
+ * @pa: pfn_array on which to perform the operation
+ * @vdev: the vfio device to perform the operation
+ * @pa_nr: number of user pages to unpin
+ *
+ * Only unpin if any pages were pinned to begin with, i.e. pa_nr >
 0,
+ * otherwise only clear pa->pa_nr
+ */
+static void pfn_array_unpin(struct pfn_array *pa,
+			    struct vfio_device *vdev, int pa_nr)
+{
+	int unpinned = 0, npage = 1;
+
+	while (unpinned < pa_nr) {
+		unsigned long *first = &pa->pa_iova_pfn[unpinned];
+		unsigned long *last = &first[npage];
+
+		if (unpinned + npage < pa_nr &&
+		    *first + npage == *last) {
+			npage++;
+			continue;
+		}
+
+		vfio_unpin_pages(vdev, first, npage);
+		unpinned += npage;
+		npage = 1;
+	}
+
+	pa->pa_nr = 0;
+}
+
 /*

* pfn_array_pin() - Pin user pages in memory
* @pa: pfn_array on which to perform the operation

drivers/s390/cio/vfio_ccw_cp.c +24 -14
@@ -101,34 +133,44 @@ static int pfn_array_alloc(struct pfn_array
 *pa, u64 iova, unsigned int len)
  */
 static int pfn_array_pin(struct pfn_array *pa, struct vfio_device
 *vdev)
 {
+	int pinned = 0, npage = 1;
 	int ret = 0;
 
-	ret = vfio_pin_pages(vdev, pa->pa_iova_pfn, pa->pa_nr,
-			     IOMMU_READ | IOMMU_WRITE, pa->pa_pfn);
+	while (pinned < pa->pa_nr) {
+		unsigned long *first = &pa->pa_iova_pfn[pinned];
+		unsigned long *last = &first[npage];
 
-	if (ret < 0) {
-		goto err_out;
-	} else if (ret > 0 && ret != pa->pa_nr) {
-		vfio_unpin_pages(vdev, pa->pa_iova_pfn, ret);
-		ret = -EINVAL;
-		goto err_out;
+		if (pinned + npage < pa->pa_nr &&
+		    *first + npage == *last) {
+			npage++;
+			continue;
+		}
+
+		ret = vfio_pin_pages(vdev, first, npage,
+				     IOMMU_READ | IOMMU_WRITE,
+				     &pa->pa_pfn[pinned]);
+		if (ret < 0) {
+			goto err_out;
+		} else if (ret > 0 && ret != npage) {
+			pinned += ret;
+			ret = -EINVAL;
+			goto err_out;
+		}
+		pinned += npage;
+		npage = 1;
 	}
 
 	return ret;
 
 err_out:
-	pa->pa_nr = 0;
-
+	pfn_array_unpin(pa, vdev, pinned);
 	return ret;
 }
 
 /* Unpin the pages before releasing the memory. */
 static void pfn_array_unpin_free(struct pfn_array *pa, struct
 vfio_device *vdev)
 {
-	/* Only unpin if any pages were pinned to begin with */
-	if (pa->pa_nr)
-		vfio_unpin_pages(vdev, pa->pa_iova_pfn, pa->pa_nr);
-	pa->pa_nr = 0;
+	pfn_array_unpin(pa, vdev, pa->pa_nr);

kfree(pa->pa_iova_pfn);
}

#16 lore raw

PATCH v2 4/9 vfio: Pass in starting IOVA to vfio_pin/unpin_pages API

In reply to Nicolin Chen

The vfio_pin/unpin_pages() so far accepted arrays of PFNs of user IOVA. Among all three callers, there was only one caller possibly passing in a non-contiguous PFN list, which is now ensured to have contiguous PFN inputs too.

Pass in the starting address with "iova" alone to simplify things, so callers no longer need to maintain a PFN list or to pin/unpin one page at a time. This also allows VFIO to use more efficient implementations of pin/unpin_pages.

For now, also update vfio_iommu_type1 to fit this new parameter too, while keeping its input intact (being user_iova) since we don't want to spend too much effort swapping its parameters and local variables at that level.

---
 .../driver-api/vfio-mediated-device.rst       |  4 +--
 drivers/gpu/drm/i915/gvt/kvmgt.c              | 24 ++++++-----------
 drivers/s390/cio/vfio_ccw_cp.c                |  4 +--
 drivers/s390/crypto/vfio_ap_ops.c             |  9 +++----
 drivers/vfio/vfio.c                           | 27 +++++++++----------
 drivers/vfio/vfio.h                           |  4 +--
 drivers/vfio/vfio_iommu_type1.c               | 17 ++++++------
 include/linux/vfio.h                          |  5 ++--
 8 files changed, 40 insertions(+), 54 deletions(-)
Documentation/driver-api/vfio-mediated-device.rst +2 -2
index b0fdf76b339a..ea32a0f13ddb 100644
@@ -262,10 +262,10 @@ Translation APIs for Mediated Devices
 The following APIs are provided for translating user pfn to host pfn in a VFIO
 driver::
 
-	int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
+	int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
 				  int npage, int prot, unsigned long *phys_pfn);
 
-	void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
+	void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova,
 				    int npage);
 
 These functions call back into the back-end IOMMU module by using the pin_pages
drivers/gpu/drm/i915/gvt/kvmgt.c +8 -16
index 8c67c9aba82d..ea6041fa48ac 100644
@@ -231,16 +231,8 @@ static void intel_gvt_cleanup_vgpu_type_groups(struct intel_gvt *gvt)
 static void gvt_unpin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
 		unsigned long size)
 {
-	int total_pages;
-	int npage;
-
-	total_pages = roundup(size, PAGE_SIZE) / PAGE_SIZE;
-
-	for (npage = 0; npage < total_pages; npage++) {
-		unsigned long cur_gfn = gfn + npage;
-
-		vfio_unpin_pages(&vgpu->vfio_device, &cur_gfn, 1);
-	}
+	vfio_unpin_pages(&vgpu->vfio_device, gfn << PAGE_SHIFT,
+			 roundup(size, PAGE_SIZE) / PAGE_SIZE);
 }
 
 /* Pin a normal or compound guest page for dma. */
@@ -258,14 +250,14 @@ static int gvt_pin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
 	 * on stack to hold pfns.
 	 */
 	for (npage = 0; npage < total_pages; npage++) {
-		unsigned long cur_gfn = gfn + npage;
+		dma_addr_t cur_iova = (gfn + npage) << PAGE_SHIFT;
 		unsigned long pfn;
 
-		ret = vfio_pin_pages(&vgpu->vfio_device, &cur_gfn, 1,
+		ret = vfio_pin_pages(&vgpu->vfio_device, cur_iova, 1,
 				     IOMMU_READ | IOMMU_WRITE, &pfn);
 		if (ret != 1) {
-			gvt_vgpu_err("vfio_pin_pages failed for gfn 0x%lx, ret %d\n",
-				     cur_gfn, ret);
+			gvt_vgpu_err("vfio_pin_pages failed for iova %pad, ret %d\n",
+				     &cur_iova, ret);
 			goto err;
 		}
 
@@ -309,7 +301,7 @@ static int gvt_dma_map_page(struct intel_vgpu *vgpu, unsigned long gfn,
 	if (dma_mapping_error(dev, *dma_addr)) {
 		gvt_vgpu_err("DMA mapping failed for pfn 0x%lx, ret %d\n",
 			     page_to_pfn(page), ret);
-		gvt_unpin_guest_page(vgpu, gfn, size);
+		gvt_unpin_guest_page(vgpu, gfn << PAGE_SHIFT, size);
 		return -ENOMEM;
 	}
 
@@ -322,7 +314,7 @@ static void gvt_dma_unmap_page(struct intel_vgpu *vgpu, unsigned long gfn,
 	struct device *dev = vgpu->gvt->gt->i915->drm.dev;
 
 	dma_unmap_page(dev, dma_addr, size, DMA_BIDIRECTIONAL);
-	gvt_unpin_guest_page(vgpu, gfn, size);
+	gvt_unpin_guest_page(vgpu, gfn << PAGE_SHIFT, size);
 }
 
 static struct gvt_dma *__gvt_cache_find_dma_addr(struct intel_vgpu *vgpu,
drivers/s390/cio/vfio_ccw_cp.c +2 -2
index 3b94863ad24e..a739262f988d 100644
@@ -114,7 +114,7 @@ static void pfn_array_unpin(struct pfn_array *pa,
 			continue;
 		}
 
-		vfio_unpin_pages(vdev, first, npage);
+		vfio_unpin_pages(vdev, *first << PAGE_SHIFT, npage);
 		unpinned += npage;
 		npage = 1;
 	}
@@ -146,7 +146,7 @@ static int pfn_array_pin(struct pfn_array *pa, struct vfio_device *vdev)
 			continue;
 		}
 
-		ret = vfio_pin_pages(vdev, first, npage,
+		ret = vfio_pin_pages(vdev, *first << PAGE_SHIFT, npage,
 				     IOMMU_READ | IOMMU_WRITE,
 				     &pa->pa_pfn[pinned]);
 		if (ret < 0) {
drivers/s390/crypto/vfio_ap_ops.c +4 -5
index bb869b28cebd..8a2018ab3cf0 100644
@@ -124,7 +124,7 @@ static void vfio_ap_free_aqic_resources(struct vfio_ap_queue *q)
 		q->saved_isc = VFIO_AP_ISC_INVALID;
 	}
 	if (q->saved_pfn && !WARN_ON(!q->matrix_mdev)) {
-		vfio_unpin_pages(&q->matrix_mdev->vdev, &q->saved_pfn, 1);
+		vfio_unpin_pages(&q->matrix_mdev->vdev, q->saved_pfn << PAGE_SHIFT, 1);
 		q->saved_pfn = 0;
 	}
 }
@@ -258,7 +258,7 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
 		return status;
 	}
 
-	ret = vfio_pin_pages(&q->matrix_mdev->vdev, &g_pfn, 1,
+	ret = vfio_pin_pages(&q->matrix_mdev->vdev, g_pfn << PAGE_SHIFT, 1,
 			     IOMMU_READ | IOMMU_WRITE, &h_pfn);
 	switch (ret) {
 	case 1:
@@ -301,7 +301,7 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
 		break;
 	case AP_RESPONSE_OTHERWISE_CHANGED:
 		/* We could not modify IRQ setings: clear new configuration */
-		vfio_unpin_pages(&q->matrix_mdev->vdev, &g_pfn, 1);
+		vfio_unpin_pages(&q->matrix_mdev->vdev, g_pfn << PAGE_SHIFT, 1);
 		kvm_s390_gisc_unregister(kvm, isc);
 		break;
 	default:
@@ -1248,9 +1248,8 @@ static int vfio_ap_mdev_iommu_notifier(struct notifier_block *nb,
 
 	if (action == VFIO_IOMMU_NOTIFY_DMA_UNMAP) {
 		struct vfio_iommu_type1_dma_unmap *unmap = data;
-		unsigned long g_pfn = unmap->iova >> PAGE_SHIFT;
 
-		vfio_unpin_pages(&matrix_mdev->vdev, &g_pfn, 1);
+		vfio_unpin_pages(&matrix_mdev->vdev, unmap->iova, 1);
 		return NOTIFY_OK;
 	}
 
drivers/vfio/vfio.c +12 -15
index 01f45ec70a3d..813b517c973e 100644
@@ -1910,17 +1910,17 @@ int vfio_set_irqs_validate_and_prepare(struct vfio_irq_set *hdr, int num_irqs,
 EXPORT_SYMBOL(vfio_set_irqs_validate_and_prepare);
 
 /*
- * Pin a set of guest PFNs and return their associated host PFNs for local
+ * Pin contiguous guest pages and return their associated host pages for local
  * domain only.
  * @device [in]  : device
- * @user_pfn [in]: array of user/guest PFNs to be pinned.
- * @npage [in]   : count of elements in user_pfn array.  This count should not
+ * @iova [in]    : starting IOVA of user/guest pages to be pinned.
+ * @npage [in]   : count of pages to be pinned.  This count should not
  *		   be greater VFIO_PIN_PAGES_MAX_ENTRIES.
  * @prot [in]    : protection flags
  * @phys_pfn[out]: array of host PFNs
  * Return error or number of pages pinned.
  */
-int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
+int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
 		   int npage, int prot, unsigned long *phys_pfn)
 {
 	struct vfio_container *container;
@@ -1928,8 +1928,7 @@ int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
 	struct vfio_iommu_driver *driver;
 	int ret;
 
-	if (!user_pfn || !phys_pfn || !npage ||
-	    !vfio_assert_device_open(device))
+	if (!phys_pfn || !npage || !vfio_assert_device_open(device))
 		return -EINVAL;
 
 	if (npage > VFIO_PIN_PAGES_MAX_ENTRIES)
@@ -1943,7 +1942,7 @@ int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
 	driver = container->iommu_driver;
 	if (likely(driver && driver->ops->pin_pages))
 		ret = driver->ops->pin_pages(container->iommu_data,
-					     group->iommu_group, user_pfn,
+					     group->iommu_group, iova,
 					     npage, prot, phys_pfn);
 	else
 		ret = -ENOTTY;
@@ -1953,20 +1952,18 @@ int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
 EXPORT_SYMBOL(vfio_pin_pages);
 
 /*
- * Unpin set of host PFNs for local domain only.
+ * Unpin contiguous host pages for local domain only.
  * @device [in]  : device
- * @user_pfn [in]: array of user/guest PFNs to be unpinned. Number of user/guest
- *		   PFNs should not be greater than VFIO_PIN_PAGES_MAX_ENTRIES.
- * @npage [in]   : count of elements in user_pfn array.  This count should not
+ * @iova [in]    : starting address of user/guest pages to be unpinned.
+ * @npage [in]   : count of pages to be unpinned.  This count should not
  *                 be greater than VFIO_PIN_PAGES_MAX_ENTRIES.
  */
-void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
-		      int npage)
+void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage)
 {
 	struct vfio_container *container;
 	struct vfio_iommu_driver *driver;
 
-	if (WARN_ON_ONCE(!user_pfn || !npage || !vfio_assert_device_open(device)))
+	if (WARN_ON_ONCE(!npage || !vfio_assert_device_open(device)))
 		return;
 
 	if (WARN_ON_ONCE(npage > VFIO_PIN_PAGES_MAX_ENTRIES))
@@ -1979,7 +1976,7 @@ void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
 	if (WARN_ON_ONCE(unlikely(!driver || !driver->ops->unpin_pages)))
 		return;
 
-	driver->ops->unpin_pages(container->iommu_data, user_pfn, npage);
+	driver->ops->unpin_pages(container->iommu_data, iova, npage);
 }
 EXPORT_SYMBOL(vfio_unpin_pages);
 
drivers/vfio/vfio.h +2 -2
index bef4edf58138..dbcd0e8c031b 100644
@@ -50,11 +50,11 @@ struct vfio_iommu_driver_ops {
 					struct iommu_group *group);
 	int		(*pin_pages)(void *iommu_data,
 				     struct iommu_group *group,
-				     unsigned long *user_pfn,
+				     dma_addr_t user_iova,
 				     int npage, int prot,
 				     unsigned long *phys_pfn);
 	void		(*unpin_pages)(void *iommu_data,
-				       unsigned long *user_pfn, int npage);
+				       dma_addr_t user_iova, int npage);
 	int		(*register_notifier)(void *iommu_data,
 					     unsigned long *events,
 					     struct notifier_block *nb);
drivers/vfio/vfio_iommu_type1.c +8 -9
index 08613edaf722..f10d0c4b1f26 100644
@@ -828,7 +828,7 @@ static int vfio_unpin_page_external(struct vfio_dma *dma, dma_addr_t iova,
 
 static int vfio_iommu_type1_pin_pages(void *iommu_data,
 				      struct iommu_group *iommu_group,
-				      unsigned long *user_pfn,
+				      dma_addr_t user_iova,
 				      int npage, int prot,
 				      unsigned long *phys_pfn)
 {
@@ -840,7 +840,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
 	bool do_accounting;
 	dma_addr_t iova;
 
-	if (!iommu || !user_pfn || !phys_pfn)
+	if (!iommu || !phys_pfn)
 		return -EINVAL;
 
 	/* Supported for v2 version only */
@@ -856,7 +856,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
 again:
 	if (iommu->vaddr_invalid_count) {
 		for (i = 0; i < npage; i++) {
-			iova = user_pfn[i] << PAGE_SHIFT;
+			iova = user_iova + PAGE_SIZE * i;
 			ret = vfio_find_dma_valid(iommu, iova, PAGE_SIZE, &dma);
 			if (ret < 0)
 				goto pin_done;
@@ -881,7 +881,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
 	for (i = 0; i < npage; i++) {
 		struct vfio_pfn *vpfn;
 
-		iova = user_pfn[i] << PAGE_SHIFT;
+		iova = user_iova + PAGE_SIZE * i;
 		dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
 		if (!dma) {
 			ret = -EINVAL;
@@ -938,7 +938,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
 	for (j = 0; j < i; j++) {
 		dma_addr_t iova;
 
-		iova = user_pfn[j] << PAGE_SHIFT;
+		iova = user_iova + PAGE_SIZE * j;
 		dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
 		vfio_unpin_page_external(dma, iova, do_accounting);
 		phys_pfn[j] = 0;
@@ -949,13 +949,13 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
 }
 
 static void vfio_iommu_type1_unpin_pages(void *iommu_data,
-					 unsigned long *user_pfn, int npage)
+					dma_addr_t user_iova, int npage)
 {
 	struct vfio_iommu *iommu = iommu_data;
 	bool do_accounting;
 	int i;
 
-	if (WARN_ON_ONCE(!iommu || !user_pfn || npage <= 0))
+	if (WARN_ON_ONCE(!iommu || npage <= 0))
 		return;
 
 	/* Supported for v2 version only */
@@ -966,10 +966,9 @@ static void vfio_iommu_type1_unpin_pages(void *iommu_data,
 
 	do_accounting = list_empty(&iommu->domain_list);
 	for (i = 0; i < npage; i++) {
+		dma_addr_t iova = user_iova + PAGE_SIZE * i;
 		struct vfio_dma *dma;
-		dma_addr_t iova;
 
-		iova = user_pfn[i] << PAGE_SHIFT;
 		dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
 		if (!dma)
 			break;
include/linux/vfio.h +2 -3
index d0844ecdc961..c3e441c0bf4b 100644
@@ -147,10 +147,9 @@ bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
 
 #define VFIO_PIN_PAGES_MAX_ENTRIES	(PAGE_SIZE/sizeof(unsigned long))
 
-int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
+int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
 		   int npage, int prot, unsigned long *phys_pfn);
-void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
-		      int npage);
+void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage);
 int vfio_dma_rw(struct vfio_device *device, dma_addr_t user_iova,
 		void *data, size_t len, bool write);
 
Signature · 1 line
2.17.1
Reviewed-by on v2 4/9
#17 lore raw

Re: [RFT][PATCH v2 4/9] vfio: Pass in starting IOVA to vfio_pin/unpin_pages API

drivers/s390/crypto/vfio_ap_ops.c static void vfio_ap_free_aqic_resources(struct vfio_ap_queue *q) · line 127

Quoted 2 lines
drivers/s390/crypto/vfio_ap_ops.c +1 -1
-		vfio_unpin_pages(&q->matrix_mdev->vdev, &q->saved_pfn, 1);
+		vfio_unpin_pages(&q->matrix_mdev->vdev, q->saved_pfn << PAGE_SHIFT, 1);

Overly long line here.

Otherwise looks good:

#18 lore raw

Re: [RFT][PATCH v2 4/9] vfio: Pass in starting IOVA to vfio_pin/unpin_pages API

Quoted 4 lines · Christoph Hellwig

drivers/s390/crypto/vfio_ap_ops.c static void vfio_ap_free_aqic_resources(struct vfio_ap_queue *q) · line 127

Quoted 2 lines
drivers/s390/crypto/vfio_ap_ops.c +1 -1
-		vfio_unpin_pages(&q->matrix_mdev->vdev, &q->saved_pfn, 1);
+		vfio_unpin_pages(&q->matrix_mdev->vdev, q->saved_pfn << PAGE_SHIFT, 1);

Overly long line here.

The following PATCH-5 drops the "<< PAGE_SHIFT", addressing this.

To not wrapping and unwrapping a line back and forth, I think we are fine here?

Quoted 1 line

Will add to v3. Thanks for the review!

maintainer
#19 lore raw

Re: [RFT][PATCH v2 4/9] vfio: Pass in starting IOVA to vfio_pin/unpin_pages API

In reply to Nicolin Chen

Reviewed by: Kirti Wankhede <[email protected]>

Quoted 360 lines · Nicolin Chen

The vfio_pin/unpin_pages() so far accepted arrays of PFNs of user IOVA. Among all three callers, there was only one caller possibly passing in a non-contiguous PFN list, which is now ensured to have contiguous PFN inputs too.

Pass in the starting address with "iova" alone to simplify things, so callers no longer need to maintain a PFN list or to pin/unpin one page at a time. This also allows VFIO to use more efficient implementations of pin/unpin_pages.

For now, also update vfio_iommu_type1 to fit this new parameter too, while keeping its input intact (being user_iova) since we don't want to spend too much effort swapping its parameters and local variables at that level.

---
  .../driver-api/vfio-mediated-device.rst       |  4 +--
  drivers/gpu/drm/i915/gvt/kvmgt.c              | 24 ++++++-----------
  drivers/s390/cio/vfio_ccw_cp.c                |  4 +--
  drivers/s390/crypto/vfio_ap_ops.c             |  9 +++----
  drivers/vfio/vfio.c                           | 27 +++++++++----------
  drivers/vfio/vfio.h                           |  4 +--
  drivers/vfio/vfio_iommu_type1.c               | 17 ++++++------
  include/linux/vfio.h                          |  5 ++--
  8 files changed, 40 insertions(+), 54 deletions(-)
Documentation/driver-api/vfio-mediated-device.rst +2 -2
index b0fdf76b339a..ea32a0f13ddb 100644
@@ -262,10 +262,10 @@ Translation APIs for Mediated Devices
  The following APIs are provided for translating user pfn to host pfn in a VFIO
  driver::
 
-	int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
+	int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
  				  int npage, int prot, unsigned long *phys_pfn);
 
-	void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
+	void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova,
  				    int npage);
 
  These functions call back into the back-end IOMMU module by using the pin_pages
drivers/gpu/drm/i915/gvt/kvmgt.c +8 -16
index 8c67c9aba82d..ea6041fa48ac 100644
@@ -231,16 +231,8 @@ static void intel_gvt_cleanup_vgpu_type_groups(struct intel_gvt *gvt)
  static void gvt_unpin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
  		unsigned long size)
  {
-	int total_pages;
-	int npage;
-
-	total_pages = roundup(size, PAGE_SIZE) / PAGE_SIZE;
-
-	for (npage = 0; npage < total_pages; npage++) {
-		unsigned long cur_gfn = gfn + npage;
-
-		vfio_unpin_pages(&vgpu->vfio_device, &cur_gfn, 1);
-	}
+	vfio_unpin_pages(&vgpu->vfio_device, gfn << PAGE_SHIFT,
+			 roundup(size, PAGE_SIZE) / PAGE_SIZE);
  }
 
  /* Pin a normal or compound guest page for dma. */
@@ -258,14 +250,14 @@ static int gvt_pin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
  	 * on stack to hold pfns.
  	 */
  	for (npage = 0; npage < total_pages; npage++) {
-		unsigned long cur_gfn = gfn + npage;
+		dma_addr_t cur_iova = (gfn + npage) << PAGE_SHIFT;
  		unsigned long pfn;
 
-		ret = vfio_pin_pages(&vgpu->vfio_device, &cur_gfn, 1,
+		ret = vfio_pin_pages(&vgpu->vfio_device, cur_iova, 1,
  				     IOMMU_READ | IOMMU_WRITE, &pfn);
  		if (ret != 1) {
-			gvt_vgpu_err("vfio_pin_pages failed for gfn 0x%lx, ret %d\n",
-				     cur_gfn, ret);
+			gvt_vgpu_err("vfio_pin_pages failed for iova %pad, ret %d\n",
+				     &cur_iova, ret);
  			goto err;
  		}
 
@@ -309,7 +301,7 @@ static int gvt_dma_map_page(struct intel_vgpu *vgpu, unsigned long gfn,
  	if (dma_mapping_error(dev, *dma_addr)) {
  		gvt_vgpu_err("DMA mapping failed for pfn 0x%lx, ret %d\n",
  			     page_to_pfn(page), ret);
-		gvt_unpin_guest_page(vgpu, gfn, size);
+		gvt_unpin_guest_page(vgpu, gfn << PAGE_SHIFT, size);
  		return -ENOMEM;
  	}
 
@@ -322,7 +314,7 @@ static void gvt_dma_unmap_page(struct intel_vgpu *vgpu, unsigned long gfn,
  	struct device *dev = vgpu->gvt->gt->i915->drm.dev;
 
  	dma_unmap_page(dev, dma_addr, size, DMA_BIDIRECTIONAL);
-	gvt_unpin_guest_page(vgpu, gfn, size);
+	gvt_unpin_guest_page(vgpu, gfn << PAGE_SHIFT, size);
  }
 
  static struct gvt_dma *__gvt_cache_find_dma_addr(struct intel_vgpu *vgpu,
drivers/s390/cio/vfio_ccw_cp.c +2 -2
index 3b94863ad24e..a739262f988d 100644
@@ -114,7 +114,7 @@ static void pfn_array_unpin(struct pfn_array *pa,
  			continue;
  		}
 
-		vfio_unpin_pages(vdev, first, npage);
+		vfio_unpin_pages(vdev, *first << PAGE_SHIFT, npage);
  		unpinned += npage;
  		npage = 1;
  	}
@@ -146,7 +146,7 @@ static int pfn_array_pin(struct pfn_array *pa, struct vfio_device *vdev)
  			continue;
  		}
 
-		ret = vfio_pin_pages(vdev, first, npage,
+		ret = vfio_pin_pages(vdev, *first << PAGE_SHIFT, npage,
  				     IOMMU_READ | IOMMU_WRITE,
  				     &pa->pa_pfn[pinned]);
  		if (ret < 0) {
drivers/s390/crypto/vfio_ap_ops.c +4 -5
index bb869b28cebd..8a2018ab3cf0 100644
@@ -124,7 +124,7 @@ static void vfio_ap_free_aqic_resources(struct vfio_ap_queue *q)
  		q->saved_isc = VFIO_AP_ISC_INVALID;
  	}
  	if (q->saved_pfn && !WARN_ON(!q->matrix_mdev)) {
-		vfio_unpin_pages(&q->matrix_mdev->vdev, &q->saved_pfn, 1);
+		vfio_unpin_pages(&q->matrix_mdev->vdev, q->saved_pfn << PAGE_SHIFT, 1);
  		q->saved_pfn = 0;
  	}
  }
@@ -258,7 +258,7 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
  		return status;
  	}
 
-	ret = vfio_pin_pages(&q->matrix_mdev->vdev, &g_pfn, 1,
+	ret = vfio_pin_pages(&q->matrix_mdev->vdev, g_pfn << PAGE_SHIFT, 1,
  			     IOMMU_READ | IOMMU_WRITE, &h_pfn);
  	switch (ret) {
  	case 1:
@@ -301,7 +301,7 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
  		break;
  	case AP_RESPONSE_OTHERWISE_CHANGED:
  		/* We could not modify IRQ setings: clear new configuration */
-		vfio_unpin_pages(&q->matrix_mdev->vdev, &g_pfn, 1);
+		vfio_unpin_pages(&q->matrix_mdev->vdev, g_pfn << PAGE_SHIFT, 1);
  		kvm_s390_gisc_unregister(kvm, isc);
  		break;
  	default:
@@ -1248,9 +1248,8 @@ static int vfio_ap_mdev_iommu_notifier(struct notifier_block *nb,
 
  	if (action == VFIO_IOMMU_NOTIFY_DMA_UNMAP) {
  		struct vfio_iommu_type1_dma_unmap *unmap = data;
-		unsigned long g_pfn = unmap->iova >> PAGE_SHIFT;
 
-		vfio_unpin_pages(&matrix_mdev->vdev, &g_pfn, 1);
+		vfio_unpin_pages(&matrix_mdev->vdev, unmap->iova, 1);
  		return NOTIFY_OK;
  	}
 
drivers/vfio/vfio.c +12 -15
index 01f45ec70a3d..813b517c973e 100644
@@ -1910,17 +1910,17 @@ int vfio_set_irqs_validate_and_prepare(struct vfio_irq_set *hdr, int num_irqs,
  EXPORT_SYMBOL(vfio_set_irqs_validate_and_prepare);
 
  /*
- * Pin a set of guest PFNs and return their associated host PFNs for local
+ * Pin contiguous guest pages and return their associated host pages for local
   * domain only.
   * @device [in]  : device
- * @user_pfn [in]: array of user/guest PFNs to be pinned.
- * @npage [in]   : count of elements in user_pfn array.  This count should not
+ * @iova [in]    : starting IOVA of user/guest pages to be pinned.
+ * @npage [in]   : count of pages to be pinned.  This count should not
   *		   be greater VFIO_PIN_PAGES_MAX_ENTRIES.
   * @prot [in]    : protection flags
   * @phys_pfn[out]: array of host PFNs
   * Return error or number of pages pinned.
   */
-int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
+int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
  		   int npage, int prot, unsigned long *phys_pfn)
  {
  	struct vfio_container *container;
@@ -1928,8 +1928,7 @@ int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
  	struct vfio_iommu_driver *driver;
  	int ret;
 
-	if (!user_pfn || !phys_pfn || !npage ||
-	    !vfio_assert_device_open(device))
+	if (!phys_pfn || !npage || !vfio_assert_device_open(device))
  		return -EINVAL;
 
  	if (npage > VFIO_PIN_PAGES_MAX_ENTRIES)
@@ -1943,7 +1942,7 @@ int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
  	driver = container->iommu_driver;
  	if (likely(driver && driver->ops->pin_pages))
  		ret = driver->ops->pin_pages(container->iommu_data,
-					     group->iommu_group, user_pfn,
+					     group->iommu_group, iova,
  					     npage, prot, phys_pfn);
  	else
  		ret = -ENOTTY;
@@ -1953,20 +1952,18 @@ int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
  EXPORT_SYMBOL(vfio_pin_pages);
 
  /*
- * Unpin set of host PFNs for local domain only.
+ * Unpin contiguous host pages for local domain only.
   * @device [in]  : device
- * @user_pfn [in]: array of user/guest PFNs to be unpinned. Number of user/guest
- *		   PFNs should not be greater than VFIO_PIN_PAGES_MAX_ENTRIES.
- * @npage [in]   : count of elements in user_pfn array.  This count should not
+ * @iova [in]    : starting address of user/guest pages to be unpinned.
+ * @npage [in]   : count of pages to be unpinned.  This count should not
   *                 be greater than VFIO_PIN_PAGES_MAX_ENTRIES.
   */
-void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
-		      int npage)
+void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage)
  {
  	struct vfio_container *container;
  	struct vfio_iommu_driver *driver;
 
-	if (WARN_ON_ONCE(!user_pfn || !npage || !vfio_assert_device_open(device)))
+	if (WARN_ON_ONCE(!npage || !vfio_assert_device_open(device)))
  		return;
 
  	if (WARN_ON_ONCE(npage > VFIO_PIN_PAGES_MAX_ENTRIES))
@@ -1979,7 +1976,7 @@ void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
  	if (WARN_ON_ONCE(unlikely(!driver || !driver->ops->unpin_pages)))
  		return;
 
-	driver->ops->unpin_pages(container->iommu_data, user_pfn, npage);
+	driver->ops->unpin_pages(container->iommu_data, iova, npage);
  }
  EXPORT_SYMBOL(vfio_unpin_pages);
 
drivers/vfio/vfio.h +2 -2
index bef4edf58138..dbcd0e8c031b 100644
@@ -50,11 +50,11 @@ struct vfio_iommu_driver_ops {
  					struct iommu_group *group);
  	int		(*pin_pages)(void *iommu_data,
  				     struct iommu_group *group,
-				     unsigned long *user_pfn,
+				     dma_addr_t user_iova,
  				     int npage, int prot,
  				     unsigned long *phys_pfn);
  	void		(*unpin_pages)(void *iommu_data,
-				       unsigned long *user_pfn, int npage);
+				       dma_addr_t user_iova, int npage);
  	int		(*register_notifier)(void *iommu_data,
  					     unsigned long *events,
  					     struct notifier_block *nb);
drivers/vfio/vfio_iommu_type1.c +8 -9
index 08613edaf722..f10d0c4b1f26 100644
@@ -828,7 +828,7 @@ static int vfio_unpin_page_external(struct vfio_dma *dma, dma_addr_t iova,
 
  static int vfio_iommu_type1_pin_pages(void *iommu_data,
  				      struct iommu_group *iommu_group,
-				      unsigned long *user_pfn,
+				      dma_addr_t user_iova,
  				      int npage, int prot,
  				      unsigned long *phys_pfn)
  {
@@ -840,7 +840,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
  	bool do_accounting;
  	dma_addr_t iova;
 
-	if (!iommu || !user_pfn || !phys_pfn)
+	if (!iommu || !phys_pfn)
  		return -EINVAL;
 
  	/* Supported for v2 version only */
@@ -856,7 +856,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
  again:
  	if (iommu->vaddr_invalid_count) {
  		for (i = 0; i < npage; i++) {
-			iova = user_pfn[i] << PAGE_SHIFT;
+			iova = user_iova + PAGE_SIZE * i;
  			ret = vfio_find_dma_valid(iommu, iova, PAGE_SIZE, &dma);
  			if (ret < 0)
  				goto pin_done;
@@ -881,7 +881,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
  	for (i = 0; i < npage; i++) {
  		struct vfio_pfn *vpfn;
 
-		iova = user_pfn[i] << PAGE_SHIFT;
+		iova = user_iova + PAGE_SIZE * i;
  		dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
  		if (!dma) {
  			ret = -EINVAL;
@@ -938,7 +938,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
  	for (j = 0; j < i; j++) {
  		dma_addr_t iova;
 
-		iova = user_pfn[j] << PAGE_SHIFT;
+		iova = user_iova + PAGE_SIZE * j;
  		dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
  		vfio_unpin_page_external(dma, iova, do_accounting);
  		phys_pfn[j] = 0;
@@ -949,13 +949,13 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
  }
 
  static void vfio_iommu_type1_unpin_pages(void *iommu_data,
-					 unsigned long *user_pfn, int npage)
+					dma_addr_t user_iova, int npage)
  {
  	struct vfio_iommu *iommu = iommu_data;
  	bool do_accounting;
  	int i;
 
-	if (WARN_ON_ONCE(!iommu || !user_pfn || npage <= 0))
+	if (WARN_ON_ONCE(!iommu || npage <= 0))
  		return;
 
  	/* Supported for v2 version only */
@@ -966,10 +966,9 @@ static void vfio_iommu_type1_unpin_pages(void *iommu_data,
 
  	do_accounting = list_empty(&iommu->domain_list);
  	for (i = 0; i < npage; i++) {
+		dma_addr_t iova = user_iova + PAGE_SIZE * i;
  		struct vfio_dma *dma;
-		dma_addr_t iova;
 
-		iova = user_pfn[i] << PAGE_SHIFT;
  		dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
  		if (!dma)
  			break;
include/linux/vfio.h +2 -3
index d0844ecdc961..c3e441c0bf4b 100644
@@ -147,10 +147,9 @@ bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
 
  #define VFIO_PIN_PAGES_MAX_ENTRIES	(PAGE_SIZE/sizeof(unsigned long))
 
-int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
+int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
  		   int npage, int prot, unsigned long *phys_pfn);
-void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
-		      int npage);
+void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage);
  int vfio_dma_rw(struct vfio_device *device, dma_addr_t user_iova,
  		void *data, size_t len, bool write);
 
Reviewed-by on v2 4/9
#20 lore raw

Re: [RFT][PATCH v2 4/9] vfio: Pass in starting IOVA to vfio_pin/unpin_pages API

In reply to Nicolin Chen
Quoted 11 lines · Nicolin Chen

These functions call back into the back-end IOMMU module by using the pin_pages

drivers/gpu/drm/i915/gvt/kvmgt.c +0 -2
index 8c67c9aba82d..ea6041fa48ac 100644
@@ -231,16 +231,8 @@ static void intel_gvt_cleanup_vgpu_type_groups(struct intel_gvt *gvt)
 static void gvt_unpin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
 		unsigned long size)
 {
-	int total_pages;
-	int npage;

drivers/gpu/drm/i915/gvt/kvmgt.c static void intel_gvt_cleanup_vgpu_type_groups(struct intel_gvt *gvt) · line 235

Quoted 10 lines
drivers/gpu/drm/i915/gvt/kvmgt.c +2 -8
-
-	total_pages = roundup(size, PAGE_SIZE) / PAGE_SIZE;
-
-	for (npage = 0; npage < total_pages; npage++) {
-		unsigned long cur_gfn = gfn + npage;
-
-		vfio_unpin_pages(&vgpu->vfio_device, &cur_gfn, 1);
-	}
+	vfio_unpin_pages(&vgpu->vfio_device, gfn << PAGE_SHIFT,
+			 roundup(size, PAGE_SIZE) / PAGE_SIZE);

These maths are DIV_ROUND_UP()

Otherwise,

Jason

#21 lore raw

Re: [RFT][PATCH v2 4/9] vfio: Pass in starting IOVA to vfio_pin/unpin_pages API

Quoted 17 lines · Jason Gunthorpe
Quoted 15 lines · Nicolin Chen

These functions call back into the back-end IOMMU module by using the pin_pages

drivers/gpu/drm/i915/gvt/kvmgt.c +0 -6
index 8c67c9aba82d..ea6041fa48ac 100644
@@ -231,16 +231,8 @@ static void intel_gvt_cleanup_vgpu_type_groups(struct intel_gvt *gvt)
 static void gvt_unpin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
 		unsigned long size)
 {
-	int total_pages;
-	int npage;
-
-	total_pages = roundup(size, PAGE_SIZE) / PAGE_SIZE;
-
-	for (npage = 0; npage < total_pages; npage++) {
Quoted 8 lines

drivers/gpu/drm/i915/gvt/kvmgt.c static void intel_gvt_cleanup_vgpu_type_groups(struct intel_gvt *gvt) · line 235

Quoted 6 lines
drivers/gpu/drm/i915/gvt/kvmgt.c +2 -4
-		unsigned long cur_gfn = gfn + npage;
-
-		vfio_unpin_pages(&vgpu->vfio_device, &cur_gfn, 1);
-	}
+	vfio_unpin_pages(&vgpu->vfio_device, gfn << PAGE_SHIFT,
+			 roundup(size, PAGE_SIZE) / PAGE_SIZE);

These maths are DIV_ROUND_UP()

Will change in v3.

Quoted 1 line

Thanks!

#22 lore raw

Re: [RFT][PATCH v2 4/9] vfio: Pass in starting IOVA to vfio_pin/unpin_pages API

In reply to Jason Gunthorpe
Quoted 8 lines · Jason Gunthorpe
Quoted 5 lines · Nicolin Chen

These functions call back into the back-end IOMMU module by using the pin_pages

drivers/gpu/drm/i915/gvt/kvmgt.c +0 -0
index 8c67c9aba82d..ea6041fa48ac 100644
Quoted 4 lines

drivers/gpu/drm/i915/gvt/kvmgt.c static void intel_gvt_cleanup_vgpu_type_groups(struct intel_gvt *gvt) · line 235

Quoted 2 lines
drivers/gpu/drm/i915/gvt/kvmgt.c +2 -0
+	vfio_unpin_pages(&vgpu->vfio_device, gfn << PAGE_SHIFT,
+			 roundup(size, PAGE_SIZE) / PAGE_SIZE);

These maths are DIV_ROUND_UP()

Actually I see two places in this file doing the same roundup. So I am going to add a prep patch in v3 to fix them separately.

Reviewed-by on v2 4/9
#23 lore raw

RE: [RFT][PATCH v2 4/9] vfio: Pass in starting IOVA to vfio_pin/unpin_pages API

In reply to Nicolin Chen
Quoted 6 lines

From: Nicolin Chen
Sent: Wednesday, July 6, 2022 2:28 PM
/*
- * Pin a set of guest PFNs and return their associated host PFNs for local
+ * Pin contiguous guest pages and return their associated host pages for local

can we replace 'guest' with 'user'?

Quoted 6 lines

* domain only.
* @device [in] : device
- * @user_pfn [in]: array of user/guest PFNs to be pinned.
- * @npage [in] : count of elements in user_pfn array. This count should
not
+ * @iova [in] : starting IOVA of user/guest pages to be pinned.

remove 'guest'.

Quoted 2 lines

+ * @npage [in] : count of pages to be pinned. This count should not
* be greater VFIO_PIN_PAGES_MAX_ENTRIES.

greater 'than' ...

otherwise looks good:

#24 lore raw

Re: [RFT][PATCH v2 4/9] vfio: Pass in starting IOVA to vfio_pin/unpin_pages API

Quoted 19 lines · Kevin

External email: Use caution opening links or attachments

Quoted 6 lines

From: Nicolin Chen
Sent: Wednesday, July 6, 2022 2:28 PM
/*
- * Pin a set of guest PFNs and return their associated host PFNs for local
+ * Pin contiguous guest pages and return their associated host pages for local

can we replace 'guest' with 'user'?

Quoted 6 lines

* domain only.
* @device [in] : device
- * @user_pfn [in]: array of user/guest PFNs to be pinned.
- * @npage [in] : count of elements in user_pfn array. This count should
not
+ * @iova [in] : starting IOVA of user/guest pages to be pinned.

Quoted 6 lines

remove 'guest'.

Quoted 2 lines

+ * @npage [in] : count of pages to be pinned. This count should not
* be greater VFIO_PIN_PAGES_MAX_ENTRIES.

greater 'than' ...

Will update them in v3. Thanks

maintainer Acked-by on v2 4/9
#25 lore raw

Re: [RFT][PATCH v2 4/9] vfio: Pass in starting IOVA to vfio_pin/unpin_pages API

In reply to Nicolin Chen
Quoted 20 lines · Nicolin Chen

The vfio_pin/unpin_pages() so far accepted arrays of PFNs of user IOVA.
Among all three callers, there was only one caller possibly passing in
a non-contiguous PFN list, which is now ensured to have contiguous PFN
inputs too.

Pass in the starting address with "iova" alone to simplify things, so callers no longer need to maintain a PFN list or to pin/unpin one page
at a time. This also allows VFIO to use more efficient implementations
of pin/unpin_pages.

For now, also update vfio_iommu_type1 to fit this new parameter too, while keeping its input intact (being user_iova) since we don't want to spend too much effort swapping its parameters and local variables at that level.

Quoted 5 lines
---
 .../driver-api/vfio-mediated-device.rst       |  4 +--
 drivers/gpu/drm/i915/gvt/kvmgt.c              | 24 ++++++-----------
 drivers/s390/cio/vfio_ccw_cp.c                |  4 +--
Quoted 410 lines
 drivers/s390/crypto/vfio_ap_ops.c             |  9 +++----
 drivers/vfio/vfio.c                           | 27 +++++++++------
----
 drivers/vfio/vfio.h                           |  4 +--
 drivers/vfio/vfio_iommu_type1.c               | 17 ++++++------
 include/linux/vfio.h                          |  5 ++--
 8 files changed, 40 insertions(+), 54 deletions(-)
Documentation/driver-api/vfio-mediated-device.rst +0 -0
index b0fdf76b339a..ea32a0f13ddb 100644
@@ -262,10 +262,10 @@ Translation APIs for Mediated Devices
 The following APIs are provided for translating user pfn to host pfn

in a VFIO
driver::

Documentation/driver-api/vfio-mediated-device.rst +0 -1
-	int vfio_pin_pages(struct vfio_device *device, unsigned long

*user_pfn,

Documentation/driver-api/vfio-mediated-device.rst +1 -0
+	int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
 				  int npage, int prot, unsigned long

*phys_pfn);

Documentation/driver-api/vfio-mediated-device.rst +0 -1
-	void vfio_unpin_pages(struct vfio_device *device, unsigned long

*user_pfn,

Documentation/driver-api/vfio-mediated-device.rst +1 -0
+	void vfio_unpin_pages(struct vfio_device *device, dma_addr_t

iova,
int npage);

These functions call back into the back-end IOMMU module by using
the pin_pages

drivers/gpu/drm/i915/gvt/kvmgt.c +2 -10
index 8c67c9aba82d..ea6041fa48ac 100644
@@ -231,16 +231,8 @@ static void
 intel_gvt_cleanup_vgpu_type_groups(struct intel_gvt *gvt)
 static void gvt_unpin_guest_page(struct intel_vgpu *vgpu, unsigned
 long gfn,
 		unsigned long size)
 {
-	int total_pages;
-	int npage;
-
-	total_pages = roundup(size, PAGE_SIZE) / PAGE_SIZE;
-
-	for (npage = 0; npage < total_pages; npage++) {
-		unsigned long cur_gfn = gfn + npage;
-
-		vfio_unpin_pages(&vgpu->vfio_device, &cur_gfn, 1);
-	}
+	vfio_unpin_pages(&vgpu->vfio_device, gfn << PAGE_SHIFT,
+			 roundup(size, PAGE_SIZE) / PAGE_SIZE);
 }

/* Pin a normal or compound guest page for dma. */

drivers/gpu/drm/i915/gvt/kvmgt.c +4 -4
@@ -258,14 +250,14 @@ static int gvt_pin_guest_page(struct intel_vgpu
 *vgpu, unsigned long gfn,
 	 * on stack to hold pfns.
 	 */
 	for (npage = 0; npage < total_pages; npage++) {
-		unsigned long cur_gfn = gfn + npage;
+		dma_addr_t cur_iova = (gfn + npage) << PAGE_SHIFT;
 		unsigned long pfn;
 
-		ret = vfio_pin_pages(&vgpu->vfio_device, &cur_gfn, 1,
+		ret = vfio_pin_pages(&vgpu->vfio_device, cur_iova, 1,
 				     IOMMU_READ | IOMMU_WRITE, &pfn);
 		if (ret != 1) {
-			gvt_vgpu_err("vfio_pin_pages failed for gfn
 0x%lx, ret %d\n",
-				     cur_gfn, ret);
+			gvt_vgpu_err("vfio_pin_pages failed for iova
 %pad, ret %d\n",
+				     &cur_iova, ret);
drivers/gpu/drm/i915/gvt/kvmgt.c +1 -1
 			goto err;
 		}
@@ -309,7 +301,7 @@ static int gvt_dma_map_page(struct intel_vgpu
 *vgpu, unsigned long gfn,
 	if (dma_mapping_error(dev, *dma_addr)) {
 		gvt_vgpu_err("DMA mapping failed for pfn 0x%lx, ret
 %d\n",
 			     page_to_pfn(page), ret);
-		gvt_unpin_guest_page(vgpu, gfn, size);
+		gvt_unpin_guest_page(vgpu, gfn << PAGE_SHIFT, size);
 		return -ENOMEM;
drivers/gpu/drm/i915/gvt/kvmgt.c +1 -1
 	}
@@ -322,7 +314,7 @@ static void gvt_dma_unmap_page(struct intel_vgpu
 *vgpu, unsigned long gfn,
 	struct device *dev = vgpu->gvt->gt->i915->drm.dev;
 
 	dma_unmap_page(dev, dma_addr, size, DMA_BIDIRECTIONAL);
-	gvt_unpin_guest_page(vgpu, gfn, size);
+	gvt_unpin_guest_page(vgpu, gfn << PAGE_SHIFT, size);
 }
 

static struct gvt_dma *__gvt_cache_find_dma_addr(struct intel_vgpu
*vgpu,

drivers/s390/cio/vfio_ccw_cp.c +2 -2
index 3b94863ad24e..a739262f988d 100644
@@ -114,7 +114,7 @@ static void pfn_array_unpin(struct pfn_array *pa,
 			continue;
 		}
 
-		vfio_unpin_pages(vdev, first, npage);
+		vfio_unpin_pages(vdev, *first << PAGE_SHIFT, npage);
 		unpinned += npage;
 		npage = 1;
 	}
@@ -146,7 +146,7 @@ static int pfn_array_pin(struct pfn_array *pa,
 struct vfio_device *vdev)
 			continue;
 		}
 
-		ret = vfio_pin_pages(vdev, first, npage,
+		ret = vfio_pin_pages(vdev, *first << PAGE_SHIFT, npage,
 				     IOMMU_READ | IOMMU_WRITE,
 				     &pa->pa_pfn[pinned]);
drivers/s390/cio/vfio_ccw_cp.c +0 -0
 		if (ret < 0) {
drivers/s390/crypto/vfio_ap_ops.c +1 -1
index bb869b28cebd..8a2018ab3cf0 100644
@@ -124,7 +124,7 @@ static void vfio_ap_free_aqic_resources(struct
 vfio_ap_queue *q)
 		q->saved_isc = VFIO_AP_ISC_INVALID;
 	}
 	if (q->saved_pfn && !WARN_ON(!q->matrix_mdev)) {
-		vfio_unpin_pages(&q->matrix_mdev->vdev, &q->saved_pfn,
 1);
+		vfio_unpin_pages(&q->matrix_mdev->vdev, q->saved_pfn <<
 PAGE_SHIFT, 1);
drivers/s390/crypto/vfio_ap_ops.c +1 -1
 		q->saved_pfn = 0;
 	}
 }
@@ -258,7 +258,7 @@ static struct ap_queue_status
 vfio_ap_irq_enable(struct vfio_ap_queue *q,
 		return status;
 	}
 
-	ret = vfio_pin_pages(&q->matrix_mdev->vdev, &g_pfn, 1,
+	ret = vfio_pin_pages(&q->matrix_mdev->vdev, g_pfn <<
 PAGE_SHIFT, 1,
 			     IOMMU_READ | IOMMU_WRITE, &h_pfn);
drivers/s390/crypto/vfio_ap_ops.c +0 -0
 	switch (ret) {
 	case 1:
@@ -301,7 +301,7 @@ static struct ap_queue_status
 vfio_ap_irq_enable(struct vfio_ap_queue *q,
 		break;
 	case AP_RESPONSE_OTHERWISE_CHANGED:
 		/* We could not modify IRQ setings: clear new

configuration */

drivers/s390/crypto/vfio_ap_ops.c +1 -1
-		vfio_unpin_pages(&q->matrix_mdev->vdev, &g_pfn, 1);
+		vfio_unpin_pages(&q->matrix_mdev->vdev, g_pfn <<

PAGE_SHIFT, 1);

drivers/s390/crypto/vfio_ap_ops.c +1 -2
 		kvm_s390_gisc_unregister(kvm, isc);
 		break;
 	default:
@@ -1248,9 +1248,8 @@ static int vfio_ap_mdev_iommu_notifier(struct
 notifier_block *nb,
 
 	if (action == VFIO_IOMMU_NOTIFY_DMA_UNMAP) {
 		struct vfio_iommu_type1_dma_unmap *unmap = data;
-		unsigned long g_pfn = unmap->iova >> PAGE_SHIFT;
 
-		vfio_unpin_pages(&matrix_mdev->vdev, &g_pfn, 1);
+		vfio_unpin_pages(&matrix_mdev->vdev, unmap->iova, 1);
 		return NOTIFY_OK;
 	}
drivers/vfio/vfio.c +0 -1
index 01f45ec70a3d..813b517c973e 100644
@@ -1910,17 +1910,17 @@ int vfio_set_irqs_validate_and_prepare(struct
 vfio_irq_set *hdr, int num_irqs,
 EXPORT_SYMBOL(vfio_set_irqs_validate_and_prepare);
 
 /*
- * Pin a set of guest PFNs and return their associated host PFNs for

local
+ * Pin contiguous guest pages and return their associated host pages for local
* domain only.
* @device [in] : device
- * @user_pfn [in]: array of user/guest PFNs to be pinned.
- * @npage [in] : count of elements in user_pfn array. This count
should not
+ * @iova [in] : starting IOVA of user/guest pages to be pinned.
+ * @npage [in] : count of pages to be pinned. This count should
not
* be greater VFIO_PIN_PAGES_MAX_ENTRIES.
* @prot [in] : protection flags
* @phys_pfn[out]: array of host PFNs
* Return error or number of pages pinned.
*/
-int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,

drivers/vfio/vfio.c +2 -2
+int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
 		   int npage, int prot, unsigned long *phys_pfn)
 {
 	struct vfio_container *container;
@@ -1928,8 +1928,7 @@ int vfio_pin_pages(struct vfio_device *device,
 unsigned long *user_pfn,
 	struct vfio_iommu_driver *driver;
 	int ret;
 
-	if (!user_pfn || !phys_pfn || !npage ||
-	    !vfio_assert_device_open(device))
+	if (!phys_pfn || !npage || !vfio_assert_device_open(device))
 		return -EINVAL;
 
drivers/vfio/vfio.c +1 -1
 	if (npage > VFIO_PIN_PAGES_MAX_ENTRIES)
@@ -1943,7 +1942,7 @@ int vfio_pin_pages(struct vfio_device *device,
 unsigned long *user_pfn,
 	driver = container->iommu_driver;
 	if (likely(driver && driver->ops->pin_pages))
 		ret = driver->ops->pin_pages(container->iommu_data,
-					     group->iommu_group,
 user_pfn,
+					     group->iommu_group, iova,
 					     npage, prot, phys_pfn);
drivers/vfio/vfio.c +1 -2
 	else
 		ret = -ENOTTY;
@@ -1953,20 +1952,18 @@ int vfio_pin_pages(struct vfio_device
 *device, unsigned long *user_pfn,
 EXPORT_SYMBOL(vfio_pin_pages);
 
 /*
- * Unpin set of host PFNs for local domain only.
+ * Unpin contiguous host pages for local domain only.
  * @device [in]  : device
- * @user_pfn [in]: array of user/guest PFNs to be unpinned. Number

of user/guest
- * PFNs should not be greater than
VFIO_PIN_PAGES_MAX_ENTRIES.
- * @npage [in] : count of elements in user_pfn array. This count
should not
+ * @iova [in] : starting address of user/guest pages to be
unpinned.
+ * @npage [in] : count of pages to be unpinned. This count should
not
* be greater than VFIO_PIN_PAGES_MAX_ENTRIES.
*/
-void vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,

drivers/vfio/vfio.c +1 -1
-		      int npage)
+void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova,

int npage)
{

drivers/vfio/vfio.c +0 -1
 	struct vfio_container *container;
 	struct vfio_iommu_driver *driver;
 
-	if (WARN_ON_ONCE(!user_pfn || !npage ||

!vfio_assert_device_open(device)))

drivers/vfio/vfio.c +2 -1
+	if (WARN_ON_ONCE(!npage || !vfio_assert_device_open(device)))
 		return;
 
 	if (WARN_ON_ONCE(npage > VFIO_PIN_PAGES_MAX_ENTRIES))
@@ -1979,7 +1976,7 @@ void vfio_unpin_pages(struct vfio_device
 *device, unsigned long *user_pfn,
 	if (WARN_ON_ONCE(unlikely(!driver || !driver->ops-
 >unpin_pages)))
 		return;
 
-	driver->ops->unpin_pages(container->iommu_data, user_pfn,
 npage);
+	driver->ops->unpin_pages(container->iommu_data, iova, npage);

}
EXPORT_SYMBOL(vfio_unpin_pages);

drivers/vfio/vfio.h +2 -2
index bef4edf58138..dbcd0e8c031b 100644
@@ -50,11 +50,11 @@ struct vfio_iommu_driver_ops {
 					struct iommu_group *group);
 	int		(*pin_pages)(void *iommu_data,
 				     struct iommu_group *group,
-				     unsigned long *user_pfn,
+				     dma_addr_t user_iova,
 				     int npage, int prot,
 				     unsigned long *phys_pfn);
 	void		(*unpin_pages)(void *iommu_data,
-				       unsigned long *user_pfn, int
 npage);
+				       dma_addr_t user_iova, int
 npage);
 	int		(*register_notifier)(void *iommu_data,
drivers/vfio/vfio.h +0 -0
 					     unsigned long *events,
 					     struct notifier_block

*nb);

drivers/vfio/vfio_iommu_type1.c +1 -1
index 08613edaf722..f10d0c4b1f26 100644
@@ -828,7 +828,7 @@ static int vfio_unpin_page_external(struct
 vfio_dma *dma, dma_addr_t iova,
 
 static int vfio_iommu_type1_pin_pages(void *iommu_data,
 				      struct iommu_group *iommu_group,
-				      unsigned long *user_pfn,
+				      dma_addr_t user_iova,
 				      int npage, int prot,
 				      unsigned long *phys_pfn)

{

drivers/vfio/vfio_iommu_type1.c +1 -1
@@ -840,7 +840,7 @@ static int vfio_iommu_type1_pin_pages(void
 *iommu_data,
 	bool do_accounting;
 	dma_addr_t iova;
 
-	if (!iommu || !user_pfn || !phys_pfn)
+	if (!iommu || !phys_pfn)
 		return -EINVAL;
 
drivers/vfio/vfio_iommu_type1.c +1 -1
 	/* Supported for v2 version only */
@@ -856,7 +856,7 @@ static int vfio_iommu_type1_pin_pages(void
 *iommu_data,
 again:
 	if (iommu->vaddr_invalid_count) {
 		for (i = 0; i < npage; i++) {
-			iova = user_pfn[i] << PAGE_SHIFT;
+			iova = user_iova + PAGE_SIZE * i;
 			ret = vfio_find_dma_valid(iommu, iova,
 PAGE_SIZE, &dma);
drivers/vfio/vfio_iommu_type1.c +1 -1
 			if (ret < 0)
 				goto pin_done;
@@ -881,7 +881,7 @@ static int vfio_iommu_type1_pin_pages(void
 *iommu_data,
 	for (i = 0; i < npage; i++) {
 		struct vfio_pfn *vpfn;
 
-		iova = user_pfn[i] << PAGE_SHIFT;
+		iova = user_iova + PAGE_SIZE * i;
 		dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
 		if (!dma) {
drivers/vfio/vfio_iommu_type1.c +1 -1
 			ret = -EINVAL;
@@ -938,7 +938,7 @@ static int vfio_iommu_type1_pin_pages(void
 *iommu_data,
 	for (j = 0; j < i; j++) {
 		dma_addr_t iova;
 
-		iova = user_pfn[j] << PAGE_SHIFT;
+		iova = user_iova + PAGE_SIZE * j;
 		dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
 		vfio_unpin_page_external(dma, iova, do_accounting);
drivers/vfio/vfio_iommu_type1.c +2 -2
 		phys_pfn[j] = 0;
@@ -949,13 +949,13 @@ static int vfio_iommu_type1_pin_pages(void
 *iommu_data,
 }
 
 static void vfio_iommu_type1_unpin_pages(void *iommu_data,
-					 unsigned long *user_pfn, int
 npage)
+					dma_addr_t user_iova, int
 npage)
 {
 	struct vfio_iommu *iommu = iommu_data;
 	bool do_accounting;
 	int i;
 
-	if (WARN_ON_ONCE(!iommu || !user_pfn || npage <= 0))
+	if (WARN_ON_ONCE(!iommu || npage <= 0))
drivers/vfio/vfio_iommu_type1.c +1 -2
 		return;
 
 	/* Supported for v2 version only */
@@ -966,10 +966,9 @@ static void vfio_iommu_type1_unpin_pages(void
 *iommu_data,
 
 	do_accounting = list_empty(&iommu->domain_list);
 	for (i = 0; i < npage; i++) {
+		dma_addr_t iova = user_iova + PAGE_SIZE * i;
 		struct vfio_dma *dma;
-		dma_addr_t iova;
 
-		iova = user_pfn[i] << PAGE_SHIFT;
 		dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
 		if (!dma)
drivers/vfio/vfio_iommu_type1.c +0 -0
 			break;
include/linux/vfio.h +2 -3
index d0844ecdc961..c3e441c0bf4b 100644
@@ -147,10 +147,9 @@ bool vfio_file_has_dev(struct file *file, struct
 vfio_device *device);
 
 #define VFIO_PIN_PAGES_MAX_ENTRIES	(PAGE_SIZE/sizeof(unsigned
 long))
 
-int vfio_pin_pages(struct vfio_device *device, unsigned long
 *user_pfn,
+int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
 		   int npage, int prot, unsigned long *phys_pfn);
-void vfio_unpin_pages(struct vfio_device *device, unsigned long
 *user_pfn,
-		      int npage);
+void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova,

int npage);
int vfio_dma_rw(struct vfio_device *device, dma_addr_t user_iova,
void *data, size_t len, bool write);

#26 lore raw

PATCH v2 5/9 vfio/ap: Remove redundant pfn

In reply to Nicolin Chen

The vfio_ap_ops code maintains both nib address and its PFN, which is redundant, merely because vfio_pin/unpin_pages API wanted pfn. Since vfio_pin/unpin_pages() now accept "iova", remove duplicated pfn code in their callers too.

---
 drivers/s390/crypto/vfio_ap_ops.c     | 42 +++++++++++----------------
 drivers/s390/crypto/vfio_ap_private.h |  4 +--
 2 files changed, 19 insertions(+), 27 deletions(-)
drivers/s390/crypto/vfio_ap_ops.c +17 -25
index 8a2018ab3cf0..6945e0ddc08c 100644
@@ -112,7 +112,7 @@ static void vfio_ap_wait_for_irqclear(int apqn)
  *
  * Unregisters the ISC in the GIB when the saved ISC not invalid.
  * Unpins the guest's page holding the NIB when it exists.
- * Resets the saved_pfn and saved_isc to invalid values.
+ * Resets the saved_iova and saved_isc to invalid values.
  */
 static void vfio_ap_free_aqic_resources(struct vfio_ap_queue *q)
 {
@@ -123,9 +123,9 @@ static void vfio_ap_free_aqic_resources(struct vfio_ap_queue *q)
 		kvm_s390_gisc_unregister(q->matrix_mdev->kvm, q->saved_isc);
 		q->saved_isc = VFIO_AP_ISC_INVALID;
 	}
-	if (q->saved_pfn && !WARN_ON(!q->matrix_mdev)) {
-		vfio_unpin_pages(&q->matrix_mdev->vdev, q->saved_pfn << PAGE_SHIFT, 1);
-		q->saved_pfn = 0;
+	if (q->saved_iova && !WARN_ON(!q->matrix_mdev)) {
+		vfio_unpin_pages(&q->matrix_mdev->vdev, q->saved_iova, 1);
+		q->saved_iova = 0;
 	}
 }
 
@@ -189,27 +189,19 @@ static struct ap_queue_status vfio_ap_irq_disable(struct vfio_ap_queue *q)
  *
  * @vcpu: the object representing the vcpu executing the PQAP(AQIC) instruction.
  * @nib: the location for storing the nib address.
- * @g_pfn: the location for storing the page frame number of the page containing
- *	   the nib.
  *
  * When the PQAP(AQIC) instruction is executed, general register 2 contains the
  * address of the notification indicator byte (nib) used for IRQ notification.
- * This function parses the nib from gr2 and calculates the page frame
- * number for the guest of the page containing the nib. The values are
- * stored in @nib and @g_pfn respectively.
- *
- * The g_pfn of the nib is then validated to ensure the nib address is valid.
+ * This function parses and validate the nib from gr2.
  *
  * Return: returns zero if the nib address is a valid; otherwise, returns
  *	   -EINVAL.
  */
-static int vfio_ap_validate_nib(struct kvm_vcpu *vcpu, unsigned long *nib,
-				unsigned long *g_pfn)
+static int vfio_ap_validate_nib(struct kvm_vcpu *vcpu, dma_addr_t *nib)
 {
 	*nib = vcpu->run->s.regs.gprs[2];
-	*g_pfn = *nib >> PAGE_SHIFT;
 
-	if (kvm_is_error_hva(gfn_to_hva(vcpu->kvm, *g_pfn)))
+	if (kvm_is_error_hva(gfn_to_hva(vcpu->kvm, *nib >> PAGE_SHIFT)))
 		return -EINVAL;
 
 	return 0;
@@ -239,34 +231,34 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
 						 int isc,
 						 struct kvm_vcpu *vcpu)
 {
-	unsigned long nib;
 	struct ap_qirq_ctrl aqic_gisa = {};
 	struct ap_queue_status status = {};
 	struct kvm_s390_gisa *gisa;
 	int nisc;
 	struct kvm *kvm;
-	unsigned long g_pfn, h_pfn;
+	unsigned long h_pfn;
 	phys_addr_t h_nib;
+	dma_addr_t nib;
 	int ret;
 
 	/* Verify that the notification indicator byte address is valid */
-	if (vfio_ap_validate_nib(vcpu, &nib, &g_pfn)) {
-		VFIO_AP_DBF_WARN("%s: invalid NIB address: nib=%#lx, g_pfn=%#lx, apqn=%#04x\n",
-				 __func__, nib, g_pfn, q->apqn);
+	if (vfio_ap_validate_nib(vcpu, &nib)) {
+		VFIO_AP_DBF_WARN("%s: invalid NIB address: nib=%pad, apqn=%#04x\n",
+				 __func__, &nib, q->apqn);
 
 		status.response_code = AP_RESPONSE_INVALID_ADDRESS;
 		return status;
 	}
 
-	ret = vfio_pin_pages(&q->matrix_mdev->vdev, g_pfn << PAGE_SHIFT, 1,
+	ret = vfio_pin_pages(&q->matrix_mdev->vdev, nib, 1,
 			     IOMMU_READ | IOMMU_WRITE, &h_pfn);
 	switch (ret) {
 	case 1:
 		break;
 	default:
 		VFIO_AP_DBF_WARN("%s: vfio_pin_pages failed: rc=%d,"
-				 "nib=%#lx, g_pfn=%#lx, apqn=%#04x\n",
-				 __func__, ret, nib, g_pfn, q->apqn);
+				 "nib=%pad, apqn=%#04x\n",
+				 __func__, ret, &nib, q->apqn);
 
 		status.response_code = AP_RESPONSE_INVALID_ADDRESS;
 		return status;
@@ -296,12 +288,12 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
 	case AP_RESPONSE_NORMAL:
 		/* See if we did clear older IRQ configuration */
 		vfio_ap_free_aqic_resources(q);
-		q->saved_pfn = g_pfn;
+		q->saved_iova = nib;
 		q->saved_isc = isc;
 		break;
 	case AP_RESPONSE_OTHERWISE_CHANGED:
 		/* We could not modify IRQ setings: clear new configuration */
-		vfio_unpin_pages(&q->matrix_mdev->vdev, g_pfn << PAGE_SHIFT, 1);
+		vfio_unpin_pages(&q->matrix_mdev->vdev, nib, 1);
 		kvm_s390_gisc_unregister(kvm, isc);
 		break;
 	default:
drivers/s390/crypto/vfio_ap_private.h +2 -2
index a26efd804d0d..479b205179bd 100644
@@ -102,13 +102,13 @@ struct ap_matrix_mdev {
  * struct vfio_ap_queue - contains the data associated with a queue bound to the
  *			  vfio_ap device driver
  * @matrix_mdev: the matrix mediated device
- * @saved_pfn: the guest PFN pinned for the guest
+ * @saved_iova: the notification indicator byte (nib) address
  * @apqn: the APQN of the AP queue device
  * @saved_isc: the guest ISC registered with the GIB interface
  */
 struct vfio_ap_queue {
 	struct ap_matrix_mdev *matrix_mdev;
-	unsigned long saved_pfn;
+	dma_addr_t saved_iova;
 	int	apqn;
 #define VFIO_AP_ISC_INVALID 0xff
 	unsigned char saved_isc;
Signature · 1 line
2.17.1
Reviewed-by on v2 5/9
#27 lore raw

Re: [RFT][PATCH v2 5/9] vfio/ap: Remove redundant pfn

Quoted 4 lines · Nicolin Chen

The vfio_ap_ops code maintains both nib address and its PFN, which is redundant, merely because vfio_pin/unpin_pages API wanted pfn. Since vfio_pin/unpin_pages() now accept "iova", remove duplicated pfn code in their callers too.

I would describe this as renaming saved_pfn to saved_iova

Quoted 18 lines

*
* @vcpu: the object representing the vcpu executing the PQAP(AQIC) instruction.
* @nib: the location for storing the nib address.
- * @g_pfn: the location for storing the page frame number of the page containing
- * the nib.
*
* When the PQAP(AQIC) instruction is executed, general register 2 contains the
* address of the notification indicator byte (nib) used for IRQ notification.
- * This function parses the nib from gr2 and calculates the page frame
- * number for the guest of the page containing the nib. The values are
- * stored in @nib and @g_pfn respectively.
- *
- * The g_pfn of the nib is then validated to ensure the nib address is valid.
+ * This function parses and validate the nib from gr2.
*
* Return: returns zero if the nib address is a valid; otherwise, returns
* -EINVAL.
*/

drivers/s390/crypto/vfio_ap_ops.c static struct ap_queue_status vfio_ap_irq_disable(struct vfio_ap_queue *q) · line 205

Quoted 10 lines
drivers/s390/crypto/vfio_ap_ops.c +2 -4
-static int vfio_ap_validate_nib(struct kvm_vcpu *vcpu, unsigned long *nib,
-				unsigned long *g_pfn)
+static int vfio_ap_validate_nib(struct kvm_vcpu *vcpu, dma_addr_t *nib)
 {
 	*nib = vcpu->run->s.regs.gprs[2];
-	*g_pfn = *nib >> PAGE_SHIFT;
 
-	if (kvm_is_error_hva(gfn_to_hva(vcpu->kvm, *g_pfn)))
+	if (kvm_is_error_hva(gfn_to_hva(vcpu->kvm, *nib >> PAGE_SHIFT)))
 		return -EINVAL;

This existing code is fishy. nib is either an IOVA passed to vfio_pin_pages() or a GFN passed to gfn_to_hva(). These are not the same thing and are not interchangable - writing code like this assumes that the guest is running with iommu=pt or no iommu.

Someone should look at it..

Otherwise it looks OK

Jason

#28 lore raw

PATCH v2 6/9 vfio/ccw: Change pa_pfn list to pa_iova list

In reply to Nicolin Chen

The vfio_ccw_cp code maintains both iova and its PFN list because the vfio_pin/unpin_pages API wanted pfn list. Since vfio_pin/unpin_pages() now accept "iova", change to maintain only pa_iova list and rename all "pfn_array" strings to "page_array", so as to simplify the code.

---
 drivers/s390/cio/vfio_ccw_cp.c | 135 ++++++++++++++++-----------------
 1 file changed, 64 insertions(+), 71 deletions(-)
drivers/s390/cio/vfio_ccw_cp.c +64 -71
index a739262f988d..3854c3d573f5 100644
@@ -18,11 +18,9 @@
 #include "vfio_ccw_cp.h"
 #include "vfio_ccw_private.h"
 
-struct pfn_array {
-	/* Starting guest physical I/O address. */
-	unsigned long		pa_iova;
-	/* Array that stores PFNs of the pages need to pin. */
-	unsigned long		*pa_iova_pfn;
+struct page_array {
+	/* Array that stores pages need to pin. */
+	dma_addr_t		*pa_iova;
 	/* Array that receives PFNs of the pages pinned. */
 	unsigned long		*pa_pfn;
 	/* Number of pages pinned from @pa_iova. */
@@ -37,53 +35,50 @@ struct ccwchain {
 	/* Count of the valid ccws in chain. */
 	int			ch_len;
 	/* Pinned PAGEs for the original data. */
-	struct pfn_array	*ch_pa;
+	struct page_array	*ch_pa;
 };
 
 /*
- * pfn_array_alloc() - alloc memory for PFNs
- * @pa: pfn_array on which to perform the operation
+ * page_array_alloc() - alloc memory for page array
+ * @pa: page_array on which to perform the operation
  * @iova: target guest physical address
  * @len: number of bytes that should be pinned from @iova
  *
- * Attempt to allocate memory for PFNs.
+ * Attempt to allocate memory for page array.
  *
- * Usage of pfn_array:
- * We expect (pa_nr == 0) and (pa_iova_pfn == NULL), any field in
+ * Usage of page_array:
+ * We expect (pa_nr == 0) and (pa_iova == NULL), any field in
  * this structure will be filled in by this function.
  *
  * Returns:
- *         0 if PFNs are allocated
- *   -EINVAL if pa->pa_nr is not initially zero, or pa->pa_iova_pfn is not NULL
+ *         0 if page array is allocated
+ *   -EINVAL if pa->pa_nr is not initially zero, or pa->pa_iova is not NULL
  *   -ENOMEM if alloc failed
  */
-static int pfn_array_alloc(struct pfn_array *pa, u64 iova, unsigned int len)
+static int page_array_alloc(struct page_array *pa, u64 iova, unsigned int len)
 {
 	int i;
 
-	if (pa->pa_nr || pa->pa_iova_pfn)
+	if (pa->pa_nr || pa->pa_iova)
 		return -EINVAL;
 
-	pa->pa_iova = iova;
-
 	pa->pa_nr = ((iova & ~PAGE_MASK) + len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 	if (!pa->pa_nr)
 		return -EINVAL;
 
-	pa->pa_iova_pfn = kcalloc(pa->pa_nr,
-				  sizeof(*pa->pa_iova_pfn) +
-				  sizeof(*pa->pa_pfn),
-				  GFP_KERNEL);
-	if (unlikely(!pa->pa_iova_pfn)) {
+	pa->pa_iova = kcalloc(pa->pa_nr,
+			      sizeof(*pa->pa_iova) + sizeof(*pa->pa_pfn),
+			      GFP_KERNEL);
+	if (unlikely(!pa->pa_iova)) {
 		pa->pa_nr = 0;
 		return -ENOMEM;
 	}
-	pa->pa_pfn = pa->pa_iova_pfn + pa->pa_nr;
+	pa->pa_pfn = (unsigned long *)&pa->pa_iova[pa->pa_nr];
 
-	pa->pa_iova_pfn[0] = pa->pa_iova >> PAGE_SHIFT;
+	pa->pa_iova[0] = iova;
 	pa->pa_pfn[0] = -1ULL;
 	for (i = 1; i < pa->pa_nr; i++) {
-		pa->pa_iova_pfn[i] = pa->pa_iova_pfn[i - 1] + 1;
+		pa->pa_iova[i] = pa->pa_iova[i - 1] + PAGE_SIZE;
 		pa->pa_pfn[i] = -1ULL;
 	}
 
@@ -91,30 +86,30 @@ static int pfn_array_alloc(struct pfn_array *pa, u64 iova, unsigned int len)
 }
 
 /*
- * pfn_array_unpin() - Unpin user pages in memory
- * @pa: pfn_array on which to perform the operation
+ * page_array_unpin() - Unpin user pages in memory
+ * @pa: page_array on which to perform the operation
  * @vdev: the vfio device to perform the operation
  * @pa_nr: number of user pages to unpin
  *
  * Only unpin if any pages were pinned to begin with, i.e. pa_nr > 0,
  * otherwise only clear pa->pa_nr
  */
-static void pfn_array_unpin(struct pfn_array *pa,
-			    struct vfio_device *vdev, int pa_nr)
+static void page_array_unpin(struct page_array *pa,
+			     struct vfio_device *vdev, int pa_nr)
 {
 	int unpinned = 0, npage = 1;
 
 	while (unpinned < pa_nr) {
-		unsigned long *first = &pa->pa_iova_pfn[unpinned];
-		unsigned long *last = &first[npage];
+		dma_addr_t *first = &pa->pa_iova[unpinned];
+		dma_addr_t *last = &first[npage];
 
 		if (unpinned + npage < pa_nr &&
-		    *first + npage == *last) {
+		    *first + npage * PAGE_SIZE == *last) {
 			npage++;
 			continue;
 		}
 
-		vfio_unpin_pages(vdev, *first << PAGE_SHIFT, npage);
+		vfio_unpin_pages(vdev, *first, npage);
 		unpinned += npage;
 		npage = 1;
 	}
@@ -123,30 +118,30 @@ static void pfn_array_unpin(struct pfn_array *pa,
 }
 
 /*
- * pfn_array_pin() - Pin user pages in memory
- * @pa: pfn_array on which to perform the operation
+ * page_array_pin() - Pin user pages in memory
+ * @pa: page_array on which to perform the operation
  * @mdev: the mediated device to perform pin operations
  *
  * Returns number of pages pinned upon success.
  * If the pin request partially succeeds, or fails completely,
  * all pages are left unpinned and a negative error value is returned.
  */
-static int pfn_array_pin(struct pfn_array *pa, struct vfio_device *vdev)
+static int page_array_pin(struct page_array *pa, struct vfio_device *vdev)
 {
 	int pinned = 0, npage = 1;
 	int ret = 0;
 
 	while (pinned < pa->pa_nr) {
-		unsigned long *first = &pa->pa_iova_pfn[pinned];
-		unsigned long *last = &first[npage];
+		dma_addr_t *first = &pa->pa_iova[pinned];
+		dma_addr_t *last = &first[npage];
 
 		if (pinned + npage < pa->pa_nr &&
-		    *first + npage == *last) {
+		    *first + npage * PAGE_SIZE == *last) {
 			npage++;
 			continue;
 		}
 
-		ret = vfio_pin_pages(vdev, *first << PAGE_SHIFT, npage,
+		ret = vfio_pin_pages(vdev, *first, npage,
 				     IOMMU_READ | IOMMU_WRITE,
 				     &pa->pa_pfn[pinned]);
 		if (ret < 0) {
@@ -163,32 +158,30 @@ static int pfn_array_pin(struct pfn_array *pa, struct vfio_device *vdev)
 	return ret;
 
 err_out:
-	pfn_array_unpin(pa, vdev, pinned);
+	page_array_unpin(pa, vdev, pinned);
 	return ret;
 }
 
 /* Unpin the pages before releasing the memory. */
-static void pfn_array_unpin_free(struct pfn_array *pa, struct vfio_device *vdev)
+static void page_array_unpin_free(struct page_array *pa, struct vfio_device *vdev)
 {
-	pfn_array_unpin(pa, vdev, pa->pa_nr);
-	kfree(pa->pa_iova_pfn);
+	page_array_unpin(pa, vdev, pa->pa_nr);
+	kfree(pa->pa_iova);
 }
 
-static bool pfn_array_iova_pinned(struct pfn_array *pa, unsigned long iova)
+static bool page_array_iova_pinned(struct page_array *pa, unsigned long iova)
 {
-	unsigned long iova_pfn = iova >> PAGE_SHIFT;
 	int i;
 
 	for (i = 0; i < pa->pa_nr; i++)
-		if (pa->pa_iova_pfn[i] == iova_pfn)
+		if (pa->pa_iova[i] == iova)
 			return true;
 
 	return false;
 }
-/* Create the list of IDAL words for a pfn_array. */
-static inline void pfn_array_idal_create_words(
-	struct pfn_array *pa,
-	unsigned long *idaws)
+/* Create the list of IDAL words for a page_array. */
+static inline void page_array_idal_create_words(struct page_array *pa,
+						unsigned long *idaws)
 {
 	int i;
 
@@ -204,7 +197,7 @@ static inline void pfn_array_idal_create_words(
 		idaws[i] = pa->pa_pfn[i] << PAGE_SHIFT;
 
 	/* Adjust the first IDAW, since it may not start on a page boundary */
-	idaws[0] += pa->pa_iova & (PAGE_SIZE - 1);
+	idaws[0] += pa->pa_iova[0] & (PAGE_SIZE - 1);
 }
 
 static void convert_ccw0_to_ccw1(struct ccw1 *source, unsigned long len)
@@ -236,18 +229,18 @@ static void convert_ccw0_to_ccw1(struct ccw1 *source, unsigned long len)
 static long copy_from_iova(struct vfio_device *vdev, void *to, u64 iova,
 			   unsigned long n)
 {
-	struct pfn_array pa = {0};
+	struct page_array pa = {0};
 	u64 from;
 	int i, ret;
 	unsigned long l, m;
 
-	ret = pfn_array_alloc(&pa, iova, n);
+	ret = page_array_alloc(&pa, iova, n);
 	if (ret < 0)
 		return ret;
 
-	ret = pfn_array_pin(&pa, vdev);
+	ret = page_array_pin(&pa, vdev);
 	if (ret < 0) {
-		pfn_array_unpin_free(&pa, vdev);
+		page_array_unpin_free(&pa, vdev);
 		return ret;
 	}
 
@@ -268,7 +261,7 @@ static long copy_from_iova(struct vfio_device *vdev, void *to, u64 iova,
 			break;
 	}
 
-	pfn_array_unpin_free(&pa, vdev);
+	page_array_unpin_free(&pa, vdev);
 
 	return l;
 }
@@ -371,7 +364,7 @@ static struct ccwchain *ccwchain_alloc(struct channel_program *cp, int len)
 	chain->ch_ccw = (struct ccw1 *)data;
 
 	data = (u8 *)(chain->ch_ccw) + sizeof(*chain->ch_ccw) * len;
-	chain->ch_pa = (struct pfn_array *)data;
+	chain->ch_pa = (struct page_array *)data;
 
 	chain->ch_len = len;
 
@@ -555,7 +548,7 @@ static int ccwchain_fetch_direct(struct ccwchain *chain,
 	struct vfio_device *vdev =
 		&container_of(cp, struct vfio_ccw_private, cp)->vdev;
 	struct ccw1 *ccw;
-	struct pfn_array *pa;
+	struct page_array *pa;
 	u64 iova;
 	unsigned long *idaws;
 	int ret;
@@ -589,13 +582,13 @@ static int ccwchain_fetch_direct(struct ccwchain *chain,
 	}
 
 	/*
-	 * Allocate an array of pfn's for pages to pin/translate.
+	 * Allocate an array of pages to pin/translate.
 	 * The number of pages is actually the count of the idaws
 	 * required for the data transfer, since we only only support
 	 * 4K IDAWs today.
 	 */
 	pa = chain->ch_pa + idx;
-	ret = pfn_array_alloc(pa, iova, bytes);
+	ret = page_array_alloc(pa, iova, bytes);
 	if (ret < 0)
 		goto out_free_idaws;
 
@@ -606,21 +599,21 @@ static int ccwchain_fetch_direct(struct ccwchain *chain,
 			goto out_unpin;
 
 		/*
-		 * Copy guest IDAWs into pfn_array, in case the memory they
+		 * Copy guest IDAWs into page_array, in case the memory they
 		 * occupy is not contiguous.
 		 */
 		for (i = 0; i < idaw_nr; i++)
-			pa->pa_iova_pfn[i] = idaws[i] >> PAGE_SHIFT;
+			pa->pa_iova[i] = idaws[i];
 	} else {
 		/*
-		 * No action is required here; the iova addresses in pfn_array
-		 * were initialized sequentially in pfn_array_alloc() beginning
+		 * No action is required here; the iova addresses in page_array
+		 * were initialized sequentially in page_array_alloc() beginning
 		 * with the contents of ccw->cda.
 		 */
 	}
 
 	if (ccw_does_data_transfer(ccw)) {
-		ret = pfn_array_pin(pa, vdev);
+		ret = page_array_pin(pa, vdev);
 		if (ret < 0)
 			goto out_unpin;
 	} else {
@@ -630,13 +623,13 @@ static int ccwchain_fetch_direct(struct ccwchain *chain,
 	ccw->cda = (__u32) virt_to_phys(idaws);
 	ccw->flags |= CCW_FLAG_IDA;
 
-	/* Populate the IDAL with pinned/translated addresses from pfn */
-	pfn_array_idal_create_words(pa, idaws);
+	/* Populate the IDAL with pinned/translated addresses from page */
+	page_array_idal_create_words(pa, idaws);
 
 	return 0;
 
 out_unpin:
-	pfn_array_unpin_free(pa, vdev);
+	page_array_unpin_free(pa, vdev);
 out_free_idaws:
 	kfree(idaws);
 out_init:
@@ -742,7 +735,7 @@ void cp_free(struct channel_program *cp)
 	cp->initialized = false;
 	list_for_each_entry_safe(chain, temp, &cp->ccwchain_list, next) {
 		for (i = 0; i < chain->ch_len; i++) {
-			pfn_array_unpin_free(chain->ch_pa + i, vdev);
+			page_array_unpin_free(chain->ch_pa + i, vdev);
 			ccwchain_cda_free(chain, i);
 		}
 		ccwchain_free(chain);
@@ -918,7 +911,7 @@ bool cp_iova_pinned(struct channel_program *cp, u64 iova)
 
 	list_for_each_entry(chain, &cp->ccwchain_list, next) {
 		for (i = 0; i < chain->ch_len; i++)
-			if (pfn_array_iova_pinned(chain->ch_pa + i, iova))
+			if (page_array_iova_pinned(chain->ch_pa + i, iova))
 				return true;
 	}
 
Signature · 1 line
2.17.1
Reviewed-by on v2 6/9
#29 lore raw

Re: [RFT][PATCH v2 6/9] vfio/ccw: Change pa_pfn list to pa_iova list

Quoted 9 lines · Nicolin Chen

The vfio_ccw_cp code maintains both iova and its PFN list because the vfio_pin/unpin_pages API wanted pfn list. Since vfio_pin/unpin_pages() now accept "iova", change to maintain only pa_iova list and rename all "pfn_array" strings to "page_array", so as to simplify the code.

---
 drivers/s390/cio/vfio_ccw_cp.c | 135 ++++++++++++++++-----------------
 1 file changed, 64 insertions(+), 71 deletions(-)

Jason

maintainer Reviewed-by on v2 6/9
#30 lore raw

Re: [RFT][PATCH v2 6/9] vfio/ccw: Change pa_pfn list to pa_iova list

In reply to Nicolin Chen
Quoted 8 lines · Nicolin Chen

The vfio_ccw_cp code maintains both iova and its PFN list because the vfio_pin/unpin_pages API wanted pfn list. Since vfio_pin/unpin_pages()
now accept "iova", change to maintain only pa_iova list and rename all
"pfn_array" strings to "page_array", so as to simplify the code.

Quoted 392 lines
---
 drivers/s390/cio/vfio_ccw_cp.c | 135 ++++++++++++++++---------------
--
 1 file changed, 64 insertions(+), 71 deletions(-)
drivers/s390/cio/vfio_ccw_cp.c +9 -13
index a739262f988d..3854c3d573f5 100644
@@ -18,11 +18,9 @@
 #include "vfio_ccw_cp.h"
 #include "vfio_ccw_private.h"
 
-struct pfn_array {
-	/* Starting guest physical I/O address. */
-	unsigned long		pa_iova;
-	/* Array that stores PFNs of the pages need to pin. */
-	unsigned long		*pa_iova_pfn;
+struct page_array {
+	/* Array that stores pages need to pin. */
+	dma_addr_t		*pa_iova;
 	/* Array that receives PFNs of the pages pinned. */
 	unsigned long		*pa_pfn;
 	/* Number of pages pinned from @pa_iova. */
@@ -37,53 +35,50 @@ struct ccwchain {
 	/* Count of the valid ccws in chain. */
 	int			ch_len;
 	/* Pinned PAGEs for the original data. */
-	struct pfn_array	*ch_pa;
+	struct page_array	*ch_pa;
 };
 
 /*
- * pfn_array_alloc() - alloc memory for PFNs
- * @pa: pfn_array on which to perform the operation
+ * page_array_alloc() - alloc memory for page array
+ * @pa: page_array on which to perform the operation
  * @iova: target guest physical address
  * @len: number of bytes that should be pinned from @iova
  *
- * Attempt to allocate memory for PFNs.
+ * Attempt to allocate memory for page array.
  *
- * Usage of pfn_array:
- * We expect (pa_nr == 0) and (pa_iova_pfn == NULL), any field in
+ * Usage of page_array:
+ * We expect (pa_nr == 0) and (pa_iova == NULL), any field in
  * this structure will be filled in by this function.
  *
  * Returns:
- *         0 if PFNs are allocated
- *   -EINVAL if pa->pa_nr is not initially zero, or pa->pa_iova_pfn

is not NULL
+ * 0 if page array is allocated
+ * -EINVAL if pa->pa_nr is not initially zero, or pa->pa_iova is
not NULL
* -ENOMEM if alloc failed
*/
-static int pfn_array_alloc(struct pfn_array *pa, u64 iova, unsigned int len)
+static int page_array_alloc(struct page_array *pa, u64 iova, unsigned int len)
{

drivers/s390/cio/vfio_ccw_cp.c +1 -3
 	int i;
 
-	if (pa->pa_nr || pa->pa_iova_pfn)
+	if (pa->pa_nr || pa->pa_iova)
 		return -EINVAL;
 
-	pa->pa_iova = iova;
-
 	pa->pa_nr = ((iova & ~PAGE_MASK) + len + (PAGE_SIZE - 1)) >>

PAGE_SHIFT;

drivers/s390/cio/vfio_ccw_cp.c +2 -5
 	if (!pa->pa_nr)
 		return -EINVAL;
 
-	pa->pa_iova_pfn = kcalloc(pa->pa_nr,
-				  sizeof(*pa->pa_iova_pfn) +
-				  sizeof(*pa->pa_pfn),
-				  GFP_KERNEL);
-	if (unlikely(!pa->pa_iova_pfn)) {
+	pa->pa_iova = kcalloc(pa->pa_nr,
+			      sizeof(*pa->pa_iova) + sizeof(*pa-
Quoted 1 line

pa_pfn),

drivers/s390/cio/vfio_ccw_cp.c +13 -11
+			      GFP_KERNEL);
+	if (unlikely(!pa->pa_iova)) {
 		pa->pa_nr = 0;
 		return -ENOMEM;
 	}
-	pa->pa_pfn = pa->pa_iova_pfn + pa->pa_nr;
+	pa->pa_pfn = (unsigned long *)&pa->pa_iova[pa->pa_nr];
 
-	pa->pa_iova_pfn[0] = pa->pa_iova >> PAGE_SHIFT;
+	pa->pa_iova[0] = iova;
 	pa->pa_pfn[0] = -1ULL;
 	for (i = 1; i < pa->pa_nr; i++) {
-		pa->pa_iova_pfn[i] = pa->pa_iova_pfn[i - 1] + 1;
+		pa->pa_iova[i] = pa->pa_iova[i - 1] + PAGE_SIZE;
 		pa->pa_pfn[i] = -1ULL;
 	}
@@ -91,30 +86,30 @@ static int pfn_array_alloc(struct pfn_array *pa,
 u64 iova, unsigned int len)
 }
 
 /*
- * pfn_array_unpin() - Unpin user pages in memory
- * @pa: pfn_array on which to perform the operation
+ * page_array_unpin() - Unpin user pages in memory
+ * @pa: page_array on which to perform the operation
  * @vdev: the vfio device to perform the operation
  * @pa_nr: number of user pages to unpin
  *
  * Only unpin if any pages were pinned to begin with, i.e. pa_nr >
 0,
  * otherwise only clear pa->pa_nr
  */
-static void pfn_array_unpin(struct pfn_array *pa,
-			    struct vfio_device *vdev, int pa_nr)
+static void page_array_unpin(struct page_array *pa,
+			     struct vfio_device *vdev, int pa_nr)
 {
 	int unpinned = 0, npage = 1;
 
 	while (unpinned < pa_nr) {
-		unsigned long *first = &pa->pa_iova_pfn[unpinned];
-		unsigned long *last = &first[npage];
+		dma_addr_t *first = &pa->pa_iova[unpinned];
+		dma_addr_t *last = &first[npage];
 
 		if (unpinned + npage < pa_nr &&
-		    *first + npage == *last) {
+		    *first + npage * PAGE_SIZE == *last) {
 			npage++;
 			continue;
 		}
 
-		vfio_unpin_pages(vdev, *first << PAGE_SHIFT, npage);
+		vfio_unpin_pages(vdev, *first, npage);
 		unpinned += npage;
drivers/s390/cio/vfio_ccw_cp.c +2 -2
 		npage = 1;
 	}
@@ -123,30 +118,30 @@ static void pfn_array_unpin(struct pfn_array
 *pa,
 }
 
 /*
- * pfn_array_pin() - Pin user pages in memory
- * @pa: pfn_array on which to perform the operation
+ * page_array_pin() - Pin user pages in memory
+ * @pa: page_array on which to perform the operation
  * @mdev: the mediated device to perform pin operations
  *
  * Returns number of pages pinned upon success.
  * If the pin request partially succeeds, or fails completely,
  * all pages are left unpinned and a negative error value is

returned.
*/
-static int pfn_array_pin(struct pfn_array *pa, struct vfio_device *vdev)
+static int page_array_pin(struct page_array *pa, struct vfio_device *vdev)
{

drivers/s390/cio/vfio_ccw_cp.c +12 -15
 	int pinned = 0, npage = 1;
 	int ret = 0;
 
 	while (pinned < pa->pa_nr) {
-		unsigned long *first = &pa->pa_iova_pfn[pinned];
-		unsigned long *last = &first[npage];
+		dma_addr_t *first = &pa->pa_iova[pinned];
+		dma_addr_t *last = &first[npage];
 
 		if (pinned + npage < pa->pa_nr &&
-		    *first + npage == *last) {
+		    *first + npage * PAGE_SIZE == *last) {
 			npage++;
 			continue;
 		}
 
-		ret = vfio_pin_pages(vdev, *first << PAGE_SHIFT, npage,
+		ret = vfio_pin_pages(vdev, *first, npage,
 				     IOMMU_READ | IOMMU_WRITE,
 				     &pa->pa_pfn[pinned]);
 		if (ret < 0) {
@@ -163,32 +158,30 @@ static int pfn_array_pin(struct pfn_array *pa,
 struct vfio_device *vdev)
 	return ret;
 
 err_out:
-	pfn_array_unpin(pa, vdev, pinned);
+	page_array_unpin(pa, vdev, pinned);
 	return ret;
 }
 
 /* Unpin the pages before releasing the memory. */
-static void pfn_array_unpin_free(struct pfn_array *pa, struct
 vfio_device *vdev)
+static void page_array_unpin_free(struct page_array *pa, struct
 vfio_device *vdev)
 {
-	pfn_array_unpin(pa, vdev, pa->pa_nr);
-	kfree(pa->pa_iova_pfn);
+	page_array_unpin(pa, vdev, pa->pa_nr);
+	kfree(pa->pa_iova);
 }
 
-static bool pfn_array_iova_pinned(struct pfn_array *pa, unsigned
 long iova)
+static bool page_array_iova_pinned(struct page_array *pa, unsigned
 long iova)
 {
-	unsigned long iova_pfn = iova >> PAGE_SHIFT;
 	int i;
 
 	for (i = 0; i < pa->pa_nr; i++)
-		if (pa->pa_iova_pfn[i] == iova_pfn)
+		if (pa->pa_iova[i] == iova)
 			return true;
 
 	return false;
 }
-/* Create the list of IDAL words for a pfn_array. */
-static inline void pfn_array_idal_create_words(
-	struct pfn_array *pa,
-	unsigned long *idaws)
+/* Create the list of IDAL words for a page_array. */
+static inline void page_array_idal_create_words(struct page_array

*pa,

drivers/s390/cio/vfio_ccw_cp.c +1 -0
+						unsigned long *idaws)
 {
 	int i;
@@ -204,7 +197,7 @@ static inline void pfn_array_idal_create_words(
 		idaws[i] = pa->pa_pfn[i] << PAGE_SHIFT;
 
 	/* Adjust the first IDAW, since it may not start on a page

boundary */

drivers/s390/cio/vfio_ccw_cp.c +1 -1
-	idaws[0] += pa->pa_iova & (PAGE_SIZE - 1);
+	idaws[0] += pa->pa_iova[0] & (PAGE_SIZE - 1);
 }
 
 static void convert_ccw0_to_ccw1(struct ccw1 *source, unsigned long

len)

drivers/s390/cio/vfio_ccw_cp.c +4 -4
@@ -236,18 +229,18 @@ static void convert_ccw0_to_ccw1(struct ccw1
 *source, unsigned long len)
 static long copy_from_iova(struct vfio_device *vdev, void *to, u64
 iova,
 			   unsigned long n)
 {
-	struct pfn_array pa = {0};
+	struct page_array pa = {0};
 	u64 from;
 	int i, ret;
 	unsigned long l, m;
 
-	ret = pfn_array_alloc(&pa, iova, n);
+	ret = page_array_alloc(&pa, iova, n);
 	if (ret < 0)
 		return ret;
 
-	ret = pfn_array_pin(&pa, vdev);
+	ret = page_array_pin(&pa, vdev);
 	if (ret < 0) {
-		pfn_array_unpin_free(&pa, vdev);
+		page_array_unpin_free(&pa, vdev);
 		return ret;
drivers/s390/cio/vfio_ccw_cp.c +1 -1
 	}
@@ -268,7 +261,7 @@ static long copy_from_iova(struct vfio_device
 *vdev, void *to, u64 iova,
 			break;
 	}
 
-	pfn_array_unpin_free(&pa, vdev);
+	page_array_unpin_free(&pa, vdev);
 
 	return l;

}

drivers/s390/cio/vfio_ccw_cp.c +2 -2
@@ -371,7 +364,7 @@ static struct ccwchain *ccwchain_alloc(struct
 channel_program *cp, int len)
 	chain->ch_ccw = (struct ccw1 *)data;
 
 	data = (u8 *)(chain->ch_ccw) + sizeof(*chain->ch_ccw) * len;
-	chain->ch_pa = (struct pfn_array *)data;
+	chain->ch_pa = (struct page_array *)data;
 
 	chain->ch_len = len;
@@ -555,7 +548,7 @@ static int ccwchain_fetch_direct(struct ccwchain
 *chain,
 	struct vfio_device *vdev =
 		&container_of(cp, struct vfio_ccw_private, cp)->vdev;
 	struct ccw1 *ccw;
-	struct pfn_array *pa;
+	struct page_array *pa;
 	u64 iova;
 	unsigned long *idaws;
drivers/s390/cio/vfio_ccw_cp.c +2 -3
 	int ret;
@@ -589,13 +582,13 @@ static int ccwchain_fetch_direct(struct
 ccwchain *chain,
 	}
 
 	/*
-	 * Allocate an array of pfn's for pages to pin/translate.
+	 * Allocate an array of pages to pin/translate.
 	 * The number of pages is actually the count of the idaws
 	 * required for the data transfer, since we only only support
 	 * 4K IDAWs today.
 	 */
 	pa = chain->ch_pa + idx;
-	ret = pfn_array_alloc(pa, iova, bytes);
+	ret = page_array_alloc(pa, iova, bytes);
 	if (ret < 0)
 		goto out_free_idaws;
@@ -606,21 +599,21 @@ static int ccwchain_fetch_direct(struct
 ccwchain *chain,
 			goto out_unpin;
 
 		/*
-		 * Copy guest IDAWs into pfn_array, in case the memory

they

drivers/s390/cio/vfio_ccw_cp.c +1 -0
+		 * Copy guest IDAWs into page_array, in case the memory

they

drivers/s390/cio/vfio_ccw_cp.c +1 -2
 		 * occupy is not contiguous.
 		 */
 		for (i = 0; i < idaw_nr; i++)
-			pa->pa_iova_pfn[i] = idaws[i] >> PAGE_SHIFT;
+			pa->pa_iova[i] = idaws[i];
 	} else {
 		/*
-		 * No action is required here; the iova addresses in

pfn_array

drivers/s390/cio/vfio_ccw_cp.c +0 -1
-		 * were initialized sequentially in pfn_array_alloc()

beginning

drivers/s390/cio/vfio_ccw_cp.c +1 -0
+		 * No action is required here; the iova addresses in

page_array

drivers/s390/cio/vfio_ccw_cp.c +1 -0
+		 * were initialized sequentially in page_array_alloc()

beginning

drivers/s390/cio/vfio_ccw_cp.c +4 -4
 		 * with the contents of ccw->cda.
 		 */
 	}
 
 	if (ccw_does_data_transfer(ccw)) {
-		ret = pfn_array_pin(pa, vdev);
+		ret = page_array_pin(pa, vdev);
 		if (ret < 0)
 			goto out_unpin;
 	} else {
@@ -630,13 +623,13 @@ static int ccwchain_fetch_direct(struct
 ccwchain *chain,
 	ccw->cda = (__u32) virt_to_phys(idaws);
 	ccw->flags |= CCW_FLAG_IDA;
 
-	/* Populate the IDAL with pinned/translated addresses from pfn
 */
-	pfn_array_idal_create_words(pa, idaws);
+	/* Populate the IDAL with pinned/translated addresses from page
 */
+	page_array_idal_create_words(pa, idaws);
 
 	return 0;
 
 out_unpin:
-	pfn_array_unpin_free(pa, vdev);
+	page_array_unpin_free(pa, vdev);

out_free_idaws:

drivers/s390/cio/vfio_ccw_cp.c +1 -1
 	kfree(idaws);
 out_init:
@@ -742,7 +735,7 @@ void cp_free(struct channel_program *cp)
 	cp->initialized = false;
 	list_for_each_entry_safe(chain, temp, &cp->ccwchain_list, next)
 {
 		for (i = 0; i < chain->ch_len; i++) {
-			pfn_array_unpin_free(chain->ch_pa + i, vdev);
+			page_array_unpin_free(chain->ch_pa + i, vdev);
 			ccwchain_cda_free(chain, i);
 		}
drivers/s390/cio/vfio_ccw_cp.c +1 -1
 		ccwchain_free(chain);
@@ -918,7 +911,7 @@ bool cp_iova_pinned(struct channel_program *cp,
 u64 iova)
 
 	list_for_each_entry(chain, &cp->ccwchain_list, next) {
 		for (i = 0; i < chain->ch_len; i++)
-			if (pfn_array_iova_pinned(chain->ch_pa + i,
 iova))
+			if (page_array_iova_pinned(chain->ch_pa + i,
 iova))
drivers/s390/cio/vfio_ccw_cp.c +0 -0
 				return true;
 	}
#31 lore raw

PATCH v2 7/9 vfio: Rename user_iova of vfio_dma_rw()

In reply to Nicolin Chen

Following the updated vfio_pin/unpin_pages(), use the simpler "iova".

---
 drivers/vfio/vfio.c  | 6 +++---
 include/linux/vfio.h | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)
drivers/vfio/vfio.c +3 -3
index 813b517c973e..b95ec2d78966 100644
@@ -1991,13 +1991,13 @@ EXPORT_SYMBOL(vfio_unpin_pages);
  * not a real device DMA, it is not necessary to pin the user space memory.
  *
  * @device [in]		: VFIO device
- * @user_iova [in]	: base IOVA of a user space buffer
+ * @iova [in]		: base IOVA of a user space buffer
  * @data [in]		: pointer to kernel buffer
  * @len [in]		: kernel buffer length
  * @write		: indicate read or write
  * Return error code on failure or 0 on success.
  */
-int vfio_dma_rw(struct vfio_device *device, dma_addr_t user_iova, void *data,
+int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova, void *data,
 		size_t len, bool write)
 {
 	struct vfio_container *container;
@@ -2013,7 +2013,7 @@ int vfio_dma_rw(struct vfio_device *device, dma_addr_t user_iova, void *data,
 
 	if (likely(driver && driver->ops->dma_rw))
 		ret = driver->ops->dma_rw(container->iommu_data,
-					  user_iova, data, len, write);
+					  iova, data, len, write);
 	else
 		ret = -ENOTTY;
 	return ret;
include/linux/vfio.h +1 -1
index c3e441c0bf4b..9108de34a79b 100644
@@ -150,7 +150,7 @@ bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
 int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
 		   int npage, int prot, unsigned long *phys_pfn);
 void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage);
-int vfio_dma_rw(struct vfio_device *device, dma_addr_t user_iova,
+int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova,
 		void *data, size_t len, bool write);
 
 /* each type has independent events */
Signature · 1 line
2.17.1
#35 lore raw

PATCH v2 8/9 vfio/ccw: Add kmap_local_page() for memcpy

In reply to Nicolin Chen

A PFN is not secure enough to promise that the memory is not IO. And direct access via memcpy() that only handles CPU memory will crash on S390 if the PFN is an IO PFN, as we have to use the memcpy_to/fromio() that uses the special S390 IO access instructions. On the other hand, a "struct page *" is always a CPU coherent thing that fits memcpy().

Also, casting a PFN to "void *" for memcpy() is not a proper practice, kmap_local_page() is the correct API to call here, though S390 doesn't use highmem, which means kmap_local_page() is a NOP.

There's a following patch changing the vfio_pin_pages() API to return a list of "struct page *" instead of PFNs. It will block any IO memory from ever getting into this call path, for such a security purpose. In this patch, add kmap_local_page() to prepare for that.

---
 drivers/s390/cio/vfio_ccw_cp.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
drivers/s390/cio/vfio_ccw_cp.c +6 -3
index 3854c3d573f5..cd4ec4f6d6ff 100644
@@ -11,6 +11,7 @@
 #include <linux/ratelimit.h>
 #include <linux/mm.h>
 #include <linux/slab.h>
+#include <linux/highmem.h>
 #include <linux/iommu.h>
 #include <linux/vfio.h>
 #include <asm/idals.h>
@@ -230,7 +231,6 @@ static long copy_from_iova(struct vfio_device *vdev, void *to, u64 iova,
 			   unsigned long n)
 {
 	struct page_array pa = {0};
-	u64 from;
 	int i, ret;
 	unsigned long l, m;
 
@@ -246,7 +246,9 @@ static long copy_from_iova(struct vfio_device *vdev, void *to, u64 iova,
 
 	l = n;
 	for (i = 0; i < pa.pa_nr; i++) {
-		from = pa.pa_pfn[i] << PAGE_SHIFT;
+		struct page *page = pfn_to_page(pa.pa_pfn[i]);
+		void *from = kmap_local_page(page);
+
 		m = PAGE_SIZE;
 		if (i == 0) {
 			from += iova & (PAGE_SIZE - 1);
@@ -254,7 +256,8 @@ static long copy_from_iova(struct vfio_device *vdev, void *to, u64 iova,
 		}
 
 		m = min(l, m);
-		memcpy(to + (n - l), (void *)from, m);
+		memcpy(to + (n - l), from, m);
+		kunmap_local(from);
 
 		l -= m;
 		if (l == 0)
Signature · 1 line
2.17.1
Reviewed-by on v2 8/9
#36 lore raw

Re: [RFT][PATCH v2 8/9] vfio/ccw: Add kmap_local_page() for memcpy

Quoted 15 lines · Nicolin Chen

A PFN is not secure enough to promise that the memory is not IO. And direct access via memcpy() that only handles CPU memory will crash on S390 if the PFN is an IO PFN, as we have to use the memcpy_to/fromio() that uses the special S390 IO access instructions. On the other hand, a "struct page *" is always a CPU coherent thing that fits memcpy().

Also, casting a PFN to "void *" for memcpy() is not a proper practice, kmap_local_page() is the correct API to call here, though S390 doesn't use highmem, which means kmap_local_page() is a NOP.

There's a following patch changing the vfio_pin_pages() API to return a list of "struct page *" instead of PFNs. It will block any IO memory from ever getting into this call path, for such a security purpose. In this patch, add kmap_local_page() to prepare for that.

Quoted 5 lines
---
 drivers/s390/cio/vfio_ccw_cp.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

Jason

maintainer Acked-by on v2 8/9
#37 lore raw

Re: [RFT][PATCH v2 8/9] vfio/ccw: Add kmap_local_page() for memcpy

In reply to Nicolin Chen
Quoted 13 lines · Nicolin Chen

A PFN is not secure enough to promise that the memory is not IO. And direct access via memcpy() that only handles CPU memory will crash on S390 if the PFN is an IO PFN, as we have to use the memcpy_to/fromio()
that uses the special S390 IO access instructions. On the other hand, a "struct page *" is always a CPU coherent thing that fits memcpy().

Also, casting a PFN to "void *" for memcpy() is not a proper practice,
kmap_local_page() is the correct API to call here, though S390 doesn't
use highmem, which means kmap_local_page() is a NOP.

Quoted 6 lines

There's a following patch changing the vfio_pin_pages() API to return a list of "struct page *" instead of PFNs. It will block any IO memory
from ever getting into this call path, for such a security purpose. In
this patch, add kmap_local_page() to prepare for that.

This all sounds like it's conflating vfio-ccw with vfio-pci, and configuration-wise I have a hard time picturing the situation described above. But in the interest of the change in the next patch, I suppose it's fine.

Quoted 52 lines
---
 drivers/s390/cio/vfio_ccw_cp.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
drivers/s390/cio/vfio_ccw_cp.c +4 -2
index 3854c3d573f5..cd4ec4f6d6ff 100644
@@ -11,6 +11,7 @@
 #include <linux/ratelimit.h>
 #include <linux/mm.h>
 #include <linux/slab.h>
+#include <linux/highmem.h>
 #include <linux/iommu.h>
 #include <linux/vfio.h>
 #include <asm/idals.h>
@@ -230,7 +231,6 @@ static long copy_from_iova(struct vfio_device
 *vdev, void *to, u64 iova,
 			   unsigned long n)
 {
 	struct page_array pa = {0};
-	u64 from;
 	int i, ret;
 	unsigned long l, m;
@@ -246,7 +246,9 @@ static long copy_from_iova(struct vfio_device
 *vdev, void *to, u64 iova,
 
 	l = n;
 	for (i = 0; i < pa.pa_nr; i++) {
-		from = pa.pa_pfn[i] << PAGE_SHIFT;
+		struct page *page = pfn_to_page(pa.pa_pfn[i]);
+		void *from = kmap_local_page(page);
+
 		m = PAGE_SIZE;
 		if (i == 0) {
drivers/s390/cio/vfio_ccw_cp.c +2 -1
 			from += iova & (PAGE_SIZE - 1);
@@ -254,7 +256,8 @@ static long copy_from_iova(struct vfio_device
 *vdev, void *to, u64 iova,
 		}
 
 		m = min(l, m);
-		memcpy(to + (n - l), (void *)from, m);
+		memcpy(to + (n - l), from, m);
+		kunmap_local(from);
 
 		l -= m;

if (l == 0)

#38 lore raw

PATCH v2 9/9 vfio: Replace phys_pfn with pages for vfio_pin_pages()

In reply to Nicolin Chen

Most of the callers of vfio_pin_pages() want "struct page *" and the low-level mm code to pin pages returns a list of "struct page *" too. So there's no gain in converting "struct page *" to PFN in between.

Replace the output parameter "phys_pfn" list with a "pages" list, to simplify callers. This also allows us to replace the vfio_iommu_type1 implementation with a more efficient one.

For now, also update vfio_iommu_type1 to fit this new parameter too.

---
 .../driver-api/vfio-mediated-device.rst       |  2 +-
 drivers/gpu/drm/i915/gvt/kvmgt.c              | 19 ++++++-------------
 drivers/s390/cio/vfio_ccw_cp.c                | 19 +++++++++----------
 drivers/s390/crypto/vfio_ap_ops.c             |  6 +++---
 drivers/vfio/vfio.c                           |  8 ++++----
 drivers/vfio/vfio.h                           |  2 +-
 drivers/vfio/vfio_iommu_type1.c               | 19 +++++++++++--------
 include/linux/vfio.h                          |  2 +-
 8 files changed, 36 insertions(+), 41 deletions(-)
Documentation/driver-api/vfio-mediated-device.rst +1 -1
index ea32a0f13ddb..ba5fefcdae1a 100644
@@ -263,7 +263,7 @@ The following APIs are provided for translating user pfn to host pfn in a VFIO
 driver::
 
 	int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
-				  int npage, int prot, unsigned long *phys_pfn);
+				  int npage, int prot, struct page **pages);
 
 	void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova,
 				    int npage);
drivers/gpu/drm/i915/gvt/kvmgt.c +6 -13
index ea6041fa48ac..3a49471dcc16 100644
@@ -239,7 +239,7 @@ static void gvt_unpin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
 static int gvt_pin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
 		unsigned long size, struct page **page)
 {
-	unsigned long base_pfn = 0;
+	struct page *base_page = NULL;
 	int total_pages;
 	int npage;
 	int ret;
@@ -251,26 +251,19 @@ static int gvt_pin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
 	 */
 	for (npage = 0; npage < total_pages; npage++) {
 		dma_addr_t cur_iova = (gfn + npage) << PAGE_SHIFT;
-		unsigned long pfn;
+		struct page *cur_page;
 
 		ret = vfio_pin_pages(&vgpu->vfio_device, cur_iova, 1,
-				     IOMMU_READ | IOMMU_WRITE, &pfn);
+				     IOMMU_READ | IOMMU_WRITE, &cur_page);
 		if (ret != 1) {
 			gvt_vgpu_err("vfio_pin_pages failed for iova %pad, ret %d\n",
 				     &cur_iova, ret);
 			goto err;
 		}
 
-		if (!pfn_valid(pfn)) {
-			gvt_vgpu_err("pfn 0x%lx is not mem backed\n", pfn);
-			npage++;
-			ret = -EFAULT;
-			goto err;
-		}
-
 		if (npage == 0)
-			base_pfn = pfn;
-		else if (base_pfn + npage != pfn) {
+			base_page = cur_page;
+		else if (base_page + npage != cur_page) {
 			gvt_vgpu_err("The pages are not continuous\n");
 			ret = -EINVAL;
 			npage++;
@@ -278,7 +271,7 @@ static int gvt_pin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
 		}
 	}
 
-	*page = pfn_to_page(base_pfn);
+	*page = base_page;
 	return 0;
 err:
 	gvt_unpin_guest_page(vgpu, gfn, npage * PAGE_SIZE);
drivers/s390/cio/vfio_ccw_cp.c +9 -10
index cd4ec4f6d6ff..8963f452f963 100644
@@ -22,8 +22,8 @@
 struct page_array {
 	/* Array that stores pages need to pin. */
 	dma_addr_t		*pa_iova;
-	/* Array that receives PFNs of the pages pinned. */
-	unsigned long		*pa_pfn;
+	/* Array that receives the pinned pages. */
+	struct page		**pa_page;
 	/* Number of pages pinned from @pa_iova. */
 	int			pa_nr;
 };
@@ -68,19 +68,19 @@ static int page_array_alloc(struct page_array *pa, u64 iova, unsigned int len)
 		return -EINVAL;
 
 	pa->pa_iova = kcalloc(pa->pa_nr,
-			      sizeof(*pa->pa_iova) + sizeof(*pa->pa_pfn),
+			      sizeof(*pa->pa_iova) + sizeof(*pa->pa_page),
 			      GFP_KERNEL);
 	if (unlikely(!pa->pa_iova)) {
 		pa->pa_nr = 0;
 		return -ENOMEM;
 	}
-	pa->pa_pfn = (unsigned long *)&pa->pa_iova[pa->pa_nr];
+	pa->pa_page = (struct page **)&pa->pa_iova[pa->pa_nr];
 
 	pa->pa_iova[0] = iova;
-	pa->pa_pfn[0] = -1ULL;
+	pa->pa_page[0] = NULL;
 	for (i = 1; i < pa->pa_nr; i++) {
 		pa->pa_iova[i] = pa->pa_iova[i - 1] + PAGE_SIZE;
-		pa->pa_pfn[i] = -1ULL;
+		pa->pa_page[i] = NULL;
 	}
 
 	return 0;
@@ -144,7 +144,7 @@ static int page_array_pin(struct page_array *pa, struct vfio_device *vdev)
 
 		ret = vfio_pin_pages(vdev, *first, npage,
 				     IOMMU_READ | IOMMU_WRITE,
-				     &pa->pa_pfn[pinned]);
+				     &pa->pa_page[pinned]);
 		if (ret < 0) {
 			goto err_out;
 		} else if (ret > 0 && ret != npage) {
@@ -195,7 +195,7 @@ static inline void page_array_idal_create_words(struct page_array *pa,
 	 */
 
 	for (i = 0; i < pa->pa_nr; i++)
-		idaws[i] = pa->pa_pfn[i] << PAGE_SHIFT;
+		idaws[i] = page_to_phys(pa->pa_page[i]);
 
 	/* Adjust the first IDAW, since it may not start on a page boundary */
 	idaws[0] += pa->pa_iova[0] & (PAGE_SIZE - 1);
@@ -246,8 +246,7 @@ static long copy_from_iova(struct vfio_device *vdev, void *to, u64 iova,
 
 	l = n;
 	for (i = 0; i < pa.pa_nr; i++) {
-		struct page *page = pfn_to_page(pa.pa_pfn[i]);
-		void *from = kmap_local_page(page);
+		void *from = kmap_local_page(pa.pa_page[i]);
 
 		m = PAGE_SIZE;
 		if (i == 0) {
drivers/s390/crypto/vfio_ap_ops.c +3 -3
index 6945e0ddc08c..b0972ca0dfa3 100644
@@ -234,9 +234,9 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
 	struct ap_qirq_ctrl aqic_gisa = {};
 	struct ap_queue_status status = {};
 	struct kvm_s390_gisa *gisa;
+	struct page *h_page;
 	int nisc;
 	struct kvm *kvm;
-	unsigned long h_pfn;
 	phys_addr_t h_nib;
 	dma_addr_t nib;
 	int ret;
@@ -251,7 +251,7 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
 	}
 
 	ret = vfio_pin_pages(&q->matrix_mdev->vdev, nib, 1,
-			     IOMMU_READ | IOMMU_WRITE, &h_pfn);
+			     IOMMU_READ | IOMMU_WRITE, &h_page);
 	switch (ret) {
 	case 1:
 		break;
@@ -267,7 +267,7 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
 	kvm = q->matrix_mdev->kvm;
 	gisa = kvm->arch.gisa_int.origin;
 
-	h_nib = (h_pfn << PAGE_SHIFT) | (nib & ~PAGE_MASK);
+	h_nib = page_to_phys(h_page) | (nib & ~PAGE_MASK);
 	aqic_gisa.gisc = isc;
 
 	nisc = kvm_s390_gisc_register(kvm, isc);
drivers/vfio/vfio.c +4 -4
index b95ec2d78966..96b758e06c4a 100644
@@ -1917,18 +1917,18 @@ EXPORT_SYMBOL(vfio_set_irqs_validate_and_prepare);
  * @npage [in]   : count of pages to be pinned.  This count should not
  *		   be greater VFIO_PIN_PAGES_MAX_ENTRIES.
  * @prot [in]    : protection flags
- * @phys_pfn[out]: array of host PFNs
+ * @pages[out]   : array of host pages
  * Return error or number of pages pinned.
  */
 int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
-		   int npage, int prot, unsigned long *phys_pfn)
+		   int npage, int prot, struct page **pages)
 {
 	struct vfio_container *container;
 	struct vfio_group *group = device->group;
 	struct vfio_iommu_driver *driver;
 	int ret;
 
-	if (!phys_pfn || !npage || !vfio_assert_device_open(device))
+	if (!pages || !npage || !vfio_assert_device_open(device))
 		return -EINVAL;
 
 	if (npage > VFIO_PIN_PAGES_MAX_ENTRIES)
@@ -1943,7 +1943,7 @@ int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
 	if (likely(driver && driver->ops->pin_pages))
 		ret = driver->ops->pin_pages(container->iommu_data,
 					     group->iommu_group, iova,
-					     npage, prot, phys_pfn);
+					     npage, prot, pages);
 	else
 		ret = -ENOTTY;
 
drivers/vfio/vfio.h +1 -1
index dbcd0e8c031b..dbfad8e20581 100644
@@ -52,7 +52,7 @@ struct vfio_iommu_driver_ops {
 				     struct iommu_group *group,
 				     dma_addr_t user_iova,
 				     int npage, int prot,
-				     unsigned long *phys_pfn);
+				     struct page **pages);
 	void		(*unpin_pages)(void *iommu_data,
 				       dma_addr_t user_iova, int npage);
 	int		(*register_notifier)(void *iommu_data,
drivers/vfio/vfio_iommu_type1.c +11 -8
index f10d0c4b1f26..de342d82b154 100644
@@ -830,7 +830,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
 				      struct iommu_group *iommu_group,
 				      dma_addr_t user_iova,
 				      int npage, int prot,
-				      unsigned long *phys_pfn)
+				      struct page **pages)
 {
 	struct vfio_iommu *iommu = iommu_data;
 	struct vfio_iommu_group *group;
@@ -840,7 +840,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
 	bool do_accounting;
 	dma_addr_t iova;
 
-	if (!iommu || !phys_pfn)
+	if (!iommu || !pages)
 		return -EINVAL;
 
 	/* Supported for v2 version only */
@@ -879,6 +879,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
 	do_accounting = list_empty(&iommu->domain_list);
 
 	for (i = 0; i < npage; i++) {
+		unsigned long phys_pfn;
 		struct vfio_pfn *vpfn;
 
 		iova = user_iova + PAGE_SIZE * i;
@@ -895,23 +896,25 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
 
 		vpfn = vfio_iova_get_vfio_pfn(dma, iova);
 		if (vpfn) {
-			phys_pfn[i] = vpfn->pfn;
+			pages[i] = pfn_to_page(vpfn->pfn);
 			continue;
 		}
 
 		remote_vaddr = dma->vaddr + (iova - dma->iova);
-		ret = vfio_pin_page_external(dma, remote_vaddr, &phys_pfn[i],
+		ret = vfio_pin_page_external(dma, remote_vaddr, &phys_pfn,
 					     do_accounting);
 		if (ret)
 			goto pin_unwind;
 
-		ret = vfio_add_to_pfn_list(dma, iova, phys_pfn[i]);
+		ret = vfio_add_to_pfn_list(dma, iova, phys_pfn);
 		if (ret) {
-			if (put_pfn(phys_pfn[i], dma->prot) && do_accounting)
+			if (put_pfn(phys_pfn, dma->prot) && do_accounting)
 				vfio_lock_acct(dma, -1, true);
 			goto pin_unwind;
 		}
 
+		pages[i] = pfn_to_page(phys_pfn);
+
 		if (iommu->dirty_page_tracking) {
 			unsigned long pgshift = __ffs(iommu->pgsize_bitmap);
 
@@ -934,14 +937,14 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
 	goto pin_done;
 
 pin_unwind:
-	phys_pfn[i] = 0;
+	pages[i] = NULL;
 	for (j = 0; j < i; j++) {
 		dma_addr_t iova;
 
 		iova = user_iova + PAGE_SIZE * j;
 		dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
 		vfio_unpin_page_external(dma, iova, do_accounting);
-		phys_pfn[j] = 0;
+		pages[j] = NULL;
 	}
 pin_done:
 	mutex_unlock(&iommu->lock);
include/linux/vfio.h +1 -1
index 9108de34a79b..c76a2c492bbd 100644
@@ -148,7 +148,7 @@ bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
 #define VFIO_PIN_PAGES_MAX_ENTRIES	(PAGE_SIZE/sizeof(unsigned long))
 
 int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
-		   int npage, int prot, unsigned long *phys_pfn);
+		   int npage, int prot, struct page **pages);
 void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage);
 int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova,
 		void *data, size_t len, bool write);
Signature · 1 line
2.17.1
maintainer Reviewed-by on v2 9/9
#40 lore raw

Re: [RFT][PATCH v2 9/9] vfio: Replace phys_pfn with pages for vfio_pin_pages()

In reply to Nicolin Chen
Quoted 326 lines · Nicolin Chen

Most of the callers of vfio_pin_pages() want "struct page *" and the low-level mm code to pin pages returns a list of "struct page *" too. So there's no gain in converting "struct page *" to PFN in between.

Replace the output parameter "phys_pfn" list with a "pages" list, to simplify callers. This also allows us to replace the vfio_iommu_type1 implementation with a more efficient one.

For now, also update vfio_iommu_type1 to fit this new parameter too.

---
  .../driver-api/vfio-mediated-device.rst       |  2 +-
  drivers/gpu/drm/i915/gvt/kvmgt.c              | 19 ++++++-------------
  drivers/s390/cio/vfio_ccw_cp.c                | 19 +++++++++----------
  drivers/s390/crypto/vfio_ap_ops.c             |  6 +++---
  drivers/vfio/vfio.c                           |  8 ++++----
  drivers/vfio/vfio.h                           |  2 +-
  drivers/vfio/vfio_iommu_type1.c               | 19 +++++++++++--------
  include/linux/vfio.h                          |  2 +-
  8 files changed, 36 insertions(+), 41 deletions(-)
Documentation/driver-api/vfio-mediated-device.rst +1 -1
index ea32a0f13ddb..ba5fefcdae1a 100644
@@ -263,7 +263,7 @@ The following APIs are provided for translating user pfn to host pfn in a VFIO
  driver::
 
  	int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
-				  int npage, int prot, unsigned long *phys_pfn);
+				  int npage, int prot, struct page **pages);
 
  	void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova,
  				    int npage);
drivers/gpu/drm/i915/gvt/kvmgt.c +6 -13
index ea6041fa48ac..3a49471dcc16 100644
@@ -239,7 +239,7 @@ static void gvt_unpin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
  static int gvt_pin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
  		unsigned long size, struct page **page)
  {
-	unsigned long base_pfn = 0;
+	struct page *base_page = NULL;
  	int total_pages;
  	int npage;
  	int ret;
@@ -251,26 +251,19 @@ static int gvt_pin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
  	 */
  	for (npage = 0; npage < total_pages; npage++) {
  		dma_addr_t cur_iova = (gfn + npage) << PAGE_SHIFT;
-		unsigned long pfn;
+		struct page *cur_page;
 
  		ret = vfio_pin_pages(&vgpu->vfio_device, cur_iova, 1,
-				     IOMMU_READ | IOMMU_WRITE, &pfn);
+				     IOMMU_READ | IOMMU_WRITE, &cur_page);
  		if (ret != 1) {
  			gvt_vgpu_err("vfio_pin_pages failed for iova %pad, ret %d\n",
  				     &cur_iova, ret);
  			goto err;
  		}
 
-		if (!pfn_valid(pfn)) {
-			gvt_vgpu_err("pfn 0x%lx is not mem backed\n", pfn);
-			npage++;
-			ret = -EFAULT;
-			goto err;
-		}
-
  		if (npage == 0)
-			base_pfn = pfn;
-		else if (base_pfn + npage != pfn) {
+			base_page = cur_page;
+		else if (base_page + npage != cur_page) {
  			gvt_vgpu_err("The pages are not continuous\n");
  			ret = -EINVAL;
  			npage++;
@@ -278,7 +271,7 @@ static int gvt_pin_guest_page(struct intel_vgpu *vgpu, unsigned long gfn,
  		}
  	}
 
-	*page = pfn_to_page(base_pfn);
+	*page = base_page;
  	return 0;
  err:
  	gvt_unpin_guest_page(vgpu, gfn, npage * PAGE_SIZE);
drivers/s390/cio/vfio_ccw_cp.c +9 -10
index cd4ec4f6d6ff..8963f452f963 100644
@@ -22,8 +22,8 @@
  struct page_array {
  	/* Array that stores pages need to pin. */
  	dma_addr_t		*pa_iova;
-	/* Array that receives PFNs of the pages pinned. */
-	unsigned long		*pa_pfn;
+	/* Array that receives the pinned pages. */
+	struct page		**pa_page;
  	/* Number of pages pinned from @pa_iova. */
  	int			pa_nr;
  };
@@ -68,19 +68,19 @@ static int page_array_alloc(struct page_array *pa, u64 iova, unsigned int len)
  		return -EINVAL;
 
  	pa->pa_iova = kcalloc(pa->pa_nr,
-			      sizeof(*pa->pa_iova) + sizeof(*pa->pa_pfn),
+			      sizeof(*pa->pa_iova) + sizeof(*pa->pa_page),
  			      GFP_KERNEL);
  	if (unlikely(!pa->pa_iova)) {
  		pa->pa_nr = 0;
  		return -ENOMEM;
  	}
-	pa->pa_pfn = (unsigned long *)&pa->pa_iova[pa->pa_nr];
+	pa->pa_page = (struct page **)&pa->pa_iova[pa->pa_nr];
 
  	pa->pa_iova[0] = iova;
-	pa->pa_pfn[0] = -1ULL;
+	pa->pa_page[0] = NULL;
  	for (i = 1; i < pa->pa_nr; i++) {
  		pa->pa_iova[i] = pa->pa_iova[i - 1] + PAGE_SIZE;
-		pa->pa_pfn[i] = -1ULL;
+		pa->pa_page[i] = NULL;
  	}
 
  	return 0;
@@ -144,7 +144,7 @@ static int page_array_pin(struct page_array *pa, struct vfio_device *vdev)
 
  		ret = vfio_pin_pages(vdev, *first, npage,
  				     IOMMU_READ | IOMMU_WRITE,
-				     &pa->pa_pfn[pinned]);
+				     &pa->pa_page[pinned]);
  		if (ret < 0) {
  			goto err_out;
  		} else if (ret > 0 && ret != npage) {
@@ -195,7 +195,7 @@ static inline void page_array_idal_create_words(struct page_array *pa,
  	 */
 
  	for (i = 0; i < pa->pa_nr; i++)
-		idaws[i] = pa->pa_pfn[i] << PAGE_SHIFT;
+		idaws[i] = page_to_phys(pa->pa_page[i]);
 
  	/* Adjust the first IDAW, since it may not start on a page boundary */
  	idaws[0] += pa->pa_iova[0] & (PAGE_SIZE - 1);
@@ -246,8 +246,7 @@ static long copy_from_iova(struct vfio_device *vdev, void *to, u64 iova,
 
  	l = n;
  	for (i = 0; i < pa.pa_nr; i++) {
-		struct page *page = pfn_to_page(pa.pa_pfn[i]);
-		void *from = kmap_local_page(page);
+		void *from = kmap_local_page(pa.pa_page[i]);
 
  		m = PAGE_SIZE;
  		if (i == 0) {
drivers/s390/crypto/vfio_ap_ops.c +3 -3
index 6945e0ddc08c..b0972ca0dfa3 100644
@@ -234,9 +234,9 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
  	struct ap_qirq_ctrl aqic_gisa = {};
  	struct ap_queue_status status = {};
  	struct kvm_s390_gisa *gisa;
+	struct page *h_page;
  	int nisc;
  	struct kvm *kvm;
-	unsigned long h_pfn;
  	phys_addr_t h_nib;
  	dma_addr_t nib;
  	int ret;
@@ -251,7 +251,7 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
  	}
 
  	ret = vfio_pin_pages(&q->matrix_mdev->vdev, nib, 1,
-			     IOMMU_READ | IOMMU_WRITE, &h_pfn);
+			     IOMMU_READ | IOMMU_WRITE, &h_page);
  	switch (ret) {
  	case 1:
  		break;
@@ -267,7 +267,7 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
  	kvm = q->matrix_mdev->kvm;
  	gisa = kvm->arch.gisa_int.origin;
 
-	h_nib = (h_pfn << PAGE_SHIFT) | (nib & ~PAGE_MASK);
+	h_nib = page_to_phys(h_page) | (nib & ~PAGE_MASK);
  	aqic_gisa.gisc = isc;
 
  	nisc = kvm_s390_gisc_register(kvm, isc);
drivers/vfio/vfio.c +4 -4
index b95ec2d78966..96b758e06c4a 100644
@@ -1917,18 +1917,18 @@ EXPORT_SYMBOL(vfio_set_irqs_validate_and_prepare);
   * @npage [in]   : count of pages to be pinned.  This count should not
   *		   be greater VFIO_PIN_PAGES_MAX_ENTRIES.
   * @prot [in]    : protection flags
- * @phys_pfn[out]: array of host PFNs
+ * @pages[out]   : array of host pages
   * Return error or number of pages pinned.
   */
  int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
-		   int npage, int prot, unsigned long *phys_pfn)
+		   int npage, int prot, struct page **pages)
  {
  	struct vfio_container *container;
  	struct vfio_group *group = device->group;
  	struct vfio_iommu_driver *driver;
  	int ret;
 
-	if (!phys_pfn || !npage || !vfio_assert_device_open(device))
+	if (!pages || !npage || !vfio_assert_device_open(device))
  		return -EINVAL;
 
  	if (npage > VFIO_PIN_PAGES_MAX_ENTRIES)
@@ -1943,7 +1943,7 @@ int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
  	if (likely(driver && driver->ops->pin_pages))
  		ret = driver->ops->pin_pages(container->iommu_data,
  					     group->iommu_group, iova,
-					     npage, prot, phys_pfn);
+					     npage, prot, pages);
  	else
  		ret = -ENOTTY;
 
drivers/vfio/vfio.h +1 -1
index dbcd0e8c031b..dbfad8e20581 100644
@@ -52,7 +52,7 @@ struct vfio_iommu_driver_ops {
  				     struct iommu_group *group,
  				     dma_addr_t user_iova,
  				     int npage, int prot,
-				     unsigned long *phys_pfn);
+				     struct page **pages);
  	void		(*unpin_pages)(void *iommu_data,
  				       dma_addr_t user_iova, int npage);
  	int		(*register_notifier)(void *iommu_data,
drivers/vfio/vfio_iommu_type1.c +11 -8
index f10d0c4b1f26..de342d82b154 100644
@@ -830,7 +830,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
  				      struct iommu_group *iommu_group,
  				      dma_addr_t user_iova,
  				      int npage, int prot,
-				      unsigned long *phys_pfn)
+				      struct page **pages)
  {
  	struct vfio_iommu *iommu = iommu_data;
  	struct vfio_iommu_group *group;
@@ -840,7 +840,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
  	bool do_accounting;
  	dma_addr_t iova;
 
-	if (!iommu || !phys_pfn)
+	if (!iommu || !pages)
  		return -EINVAL;
 
  	/* Supported for v2 version only */
@@ -879,6 +879,7 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
  	do_accounting = list_empty(&iommu->domain_list);
 
  	for (i = 0; i < npage; i++) {
+		unsigned long phys_pfn;
  		struct vfio_pfn *vpfn;
 
  		iova = user_iova + PAGE_SIZE * i;
@@ -895,23 +896,25 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
 
  		vpfn = vfio_iova_get_vfio_pfn(dma, iova);
  		if (vpfn) {
-			phys_pfn[i] = vpfn->pfn;
+			pages[i] = pfn_to_page(vpfn->pfn);
  			continue;
  		}
 
  		remote_vaddr = dma->vaddr + (iova - dma->iova);
-		ret = vfio_pin_page_external(dma, remote_vaddr, &phys_pfn[i],
+		ret = vfio_pin_page_external(dma, remote_vaddr, &phys_pfn,
  					     do_accounting);
  		if (ret)
  			goto pin_unwind;
 
-		ret = vfio_add_to_pfn_list(dma, iova, phys_pfn[i]);
+		ret = vfio_add_to_pfn_list(dma, iova, phys_pfn);
  		if (ret) {
-			if (put_pfn(phys_pfn[i], dma->prot) && do_accounting)
+			if (put_pfn(phys_pfn, dma->prot) && do_accounting)
  				vfio_lock_acct(dma, -1, true);
  			goto pin_unwind;
  		}
 
+		pages[i] = pfn_to_page(phys_pfn);
+
  		if (iommu->dirty_page_tracking) {
  			unsigned long pgshift = __ffs(iommu->pgsize_bitmap);
 
@@ -934,14 +937,14 @@ static int vfio_iommu_type1_pin_pages(void *iommu_data,
  	goto pin_done;
 
  pin_unwind:
-	phys_pfn[i] = 0;
+	pages[i] = NULL;
  	for (j = 0; j < i; j++) {
  		dma_addr_t iova;
 
  		iova = user_iova + PAGE_SIZE * j;
  		dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
  		vfio_unpin_page_external(dma, iova, do_accounting);
-		phys_pfn[j] = 0;
+		pages[j] = NULL;
  	}
  pin_done:
  	mutex_unlock(&iommu->lock);
include/linux/vfio.h +1 -1
index 9108de34a79b..c76a2c492bbd 100644
@@ -148,7 +148,7 @@ bool vfio_file_has_dev(struct file *file, struct vfio_device *device);
  #define VFIO_PIN_PAGES_MAX_ENTRIES	(PAGE_SIZE/sizeof(unsigned long))
 
  int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
-		   int npage, int prot, unsigned long *phys_pfn);
+		   int npage, int prot, struct page **pages);
  void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova, int npage);
  int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova,
  		void *data, size_t len, bool write);
Reviewed-by on v2 9/9
#41 lore raw

Re: [RFT][PATCH v2 9/9] vfio: Replace phys_pfn with pages for vfio_pin_pages()

In reply to Nicolin Chen
Quoted 11 lines · Nicolin Chen

Most of the callers of vfio_pin_pages() want "struct page *" and the low-level mm code to pin pages returns a list of "struct page *" too. So there's no gain in converting "struct page *" to PFN in between.

Replace the output parameter "phys_pfn" list with a "pages" list, to simplify callers. This also allows us to replace the vfio_iommu_type1 implementation with a more efficient one.

For now, also update vfio_iommu_type1 to fit this new parameter too.

Quoted 10 lines
---
 .../driver-api/vfio-mediated-device.rst       |  2 +-
 drivers/gpu/drm/i915/gvt/kvmgt.c              | 19 ++++++-------------
 drivers/s390/cio/vfio_ccw_cp.c                | 19 +++++++++----------
 drivers/s390/crypto/vfio_ap_ops.c             |  6 +++---
 drivers/vfio/vfio.c                           |  8 ++++----
 drivers/vfio/vfio.h                           |  2 +-
 drivers/vfio/vfio_iommu_type1.c               | 19 +++++++++++--------
 include/linux/vfio.h                          |  2 +-
 8 files changed, 36 insertions(+), 41 deletions(-)

Jason

Reviewed-by on v2 9/9
#42 lore raw

RE: [RFT][PATCH v2 9/9] vfio: Replace phys_pfn with pages for vfio_pin_pages()

In reply to Nicolin Chen
Quoted 10 lines

From: Nicolin Chen <[email protected]>
Sent: Wednesday, July 6, 2022 2:28 PM

Most of the callers of vfio_pin_pages() want "struct page *" and the low-level mm code to pin pages returns a list of "struct page *" too. So there's no gain in converting "struct page *" to PFN in between.

Replace the output parameter "phys_pfn" list with a "pages" list, to simplify callers. This also allows us to replace the vfio_iommu_type1 implementation with a more efficient one.

worth mentioning that vfio pin is only for struct page * hence the pfn_valid() check in gvt can be removed.

Quoted 4 lines

For now, also update vfio_iommu_type1 to fit this new parameter too.

#43 lore raw

Re: [RFT][PATCH v2 9/9] vfio: Replace phys_pfn with pages for vfio_pin_pages()

Quoted 10 lines · Kevin

External email: Use caution opening links or attachments

Quoted 7 lines

From: Nicolin Chen <[email protected]>
Sent: Wednesday, July 6, 2022 2:28 PM

Most of the callers of vfio_pin_pages() want "struct page *" and the low-level mm code to pin pages returns a list of "struct page *" too. So there's no gain in converting "struct page *" to PFN in between.

Quoted 6 lines
Quoted 3 lines

Replace the output parameter "phys_pfn" list with a "pages" list, to simplify callers. This also allows us to replace the vfio_iommu_type1 implementation with a more efficient one.

worth mentioning that vfio pin is only for struct page * hence the pfn_valid() check in gvt can be removed.

Will add that.

Quoted 6 lines
Quoted 4 lines

For now, also update vfio_iommu_type1 to fit this new parameter too.

Thanks for the review!

maintainer Acked-by on v2 9/9
#44 lore raw

Re: [RFT][PATCH v2 9/9] vfio: Replace phys_pfn with pages for vfio_pin_pages()

In reply to Nicolin Chen
Quoted 10 lines · Nicolin Chen

Most of the callers of vfio_pin_pages() want "struct page *" and the low-level mm code to pin pages returns a list of "struct page *" too. So there's no gain in converting "struct page *" to PFN in between.

Replace the output parameter "phys_pfn" list with a "pages" list, to simplify callers. This also allows us to replace the vfio_iommu_type1 implementation with a more efficient one.

For now, also update vfio_iommu_type1 to fit this new parameter too.

Quoted 7 lines
---
 .../driver-api/vfio-mediated-device.rst       |  2 +-
 drivers/gpu/drm/i915/gvt/kvmgt.c              | 19 ++++++-----------
--
 drivers/s390/cio/vfio_ccw_cp.c                | 19 +++++++++------

----

Quoted 355 lines
 drivers/s390/crypto/vfio_ap_ops.c             |  6 +++---
 drivers/vfio/vfio.c                           |  8 ++++----
 drivers/vfio/vfio.h                           |  2 +-
 drivers/vfio/vfio_iommu_type1.c               | 19 +++++++++++----
----
 include/linux/vfio.h                          |  2 +-
 8 files changed, 36 insertions(+), 41 deletions(-)
Documentation/driver-api/vfio-mediated-device.rst +0 -0
index ea32a0f13ddb..ba5fefcdae1a 100644
@@ -263,7 +263,7 @@ The following APIs are provided for translating

user pfn to host pfn in a VFIO
driver::

Documentation/driver-api/vfio-mediated-device.rst +0 -1
 	int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
-				  int npage, int prot, unsigned long

*phys_pfn);

Documentation/driver-api/vfio-mediated-device.rst +1 -0
+				  int npage, int prot, struct page

**pages);

void vfio_unpin_pages(struct vfio_device *device, dma_addr_t
iova,

Documentation/driver-api/vfio-mediated-device.rst +0 -0
 				    int npage);
drivers/gpu/drm/i915/gvt/kvmgt.c +1 -1
index ea6041fa48ac..3a49471dcc16 100644
@@ -239,7 +239,7 @@ static void gvt_unpin_guest_page(struct
 intel_vgpu *vgpu, unsigned long gfn,
 static int gvt_pin_guest_page(struct intel_vgpu *vgpu, unsigned long
 gfn,
 		unsigned long size, struct page **page)
 {
-	unsigned long base_pfn = 0;
+	struct page *base_page = NULL;
 	int total_pages;
drivers/gpu/drm/i915/gvt/kvmgt.c +5 -12
 	int npage;
 	int ret;
@@ -251,26 +251,19 @@ static int gvt_pin_guest_page(struct intel_vgpu
 *vgpu, unsigned long gfn,
 	 */
 	for (npage = 0; npage < total_pages; npage++) {
 		dma_addr_t cur_iova = (gfn + npage) << PAGE_SHIFT;
-		unsigned long pfn;
+		struct page *cur_page;
 
 		ret = vfio_pin_pages(&vgpu->vfio_device, cur_iova, 1,
-				     IOMMU_READ | IOMMU_WRITE, &pfn);
+				     IOMMU_READ | IOMMU_WRITE,
 &cur_page);
 		if (ret != 1) {
 			gvt_vgpu_err("vfio_pin_pages failed for iova
 %pad, ret %d\n",
 				     &cur_iova, ret);
 			goto err;
 		}
 
-		if (!pfn_valid(pfn)) {
-			gvt_vgpu_err("pfn 0x%lx is not mem backed\n",
 pfn);
-			npage++;
-			ret = -EFAULT;
-			goto err;
-		}
-
 		if (npage == 0)
-			base_pfn = pfn;
-		else if (base_pfn + npage != pfn) {
+			base_page = cur_page;
+		else if (base_page + npage != cur_page) {
 			gvt_vgpu_err("The pages are not continuous\n");
 			ret = -EINVAL;
 			npage++;
@@ -278,7 +271,7 @@ static int gvt_pin_guest_page(struct intel_vgpu
 *vgpu, unsigned long gfn,
 		}
 	}
 
-	*page = pfn_to_page(base_pfn);
+	*page = base_page;
 	return 0;
 err:
drivers/gpu/drm/i915/gvt/kvmgt.c +0 -0
 	gvt_unpin_guest_page(vgpu, gfn, npage * PAGE_SIZE);
drivers/s390/cio/vfio_ccw_cp.c +6 -6
index cd4ec4f6d6ff..8963f452f963 100644
@@ -22,8 +22,8 @@
 struct page_array {
 	/* Array that stores pages need to pin. */
 	dma_addr_t		*pa_iova;
-	/* Array that receives PFNs of the pages pinned. */
-	unsigned long		*pa_pfn;
+	/* Array that receives the pinned pages. */
+	struct page		**pa_page;
 	/* Number of pages pinned from @pa_iova. */
 	int			pa_nr;
 };
@@ -68,19 +68,19 @@ static int page_array_alloc(struct page_array
 *pa, u64 iova, unsigned int len)
 		return -EINVAL;
 
 	pa->pa_iova = kcalloc(pa->pa_nr,
-			      sizeof(*pa->pa_iova) + sizeof(*pa-
 >pa_pfn),
+			      sizeof(*pa->pa_iova) + sizeof(*pa-
 >pa_page),
 			      GFP_KERNEL);
 	if (unlikely(!pa->pa_iova)) {
 		pa->pa_nr = 0;
 		return -ENOMEM;
 	}
-	pa->pa_pfn = (unsigned long *)&pa->pa_iova[pa->pa_nr];
+	pa->pa_page = (struct page **)&pa->pa_iova[pa->pa_nr];
 
 	pa->pa_iova[0] = iova;
-	pa->pa_pfn[0] = -1ULL;
+	pa->pa_page[0] = NULL;
 	for (i = 1; i < pa->pa_nr; i++) {
 		pa->pa_iova[i] = pa->pa_iova[i - 1] + PAGE_SIZE;
-		pa->pa_pfn[i] = -1ULL;
+		pa->pa_page[i] = NULL;
drivers/s390/cio/vfio_ccw_cp.c +1 -1
 	}
 
 	return 0;
@@ -144,7 +144,7 @@ static int page_array_pin(struct page_array *pa,
 struct vfio_device *vdev)
 
 		ret = vfio_pin_pages(vdev, *first, npage,
 				     IOMMU_READ | IOMMU_WRITE,
-				     &pa->pa_pfn[pinned]);
+				     &pa->pa_page[pinned]);
 		if (ret < 0) {
 			goto err_out;
drivers/s390/cio/vfio_ccw_cp.c +1 -1
 		} else if (ret > 0 && ret != npage) {
@@ -195,7 +195,7 @@ static inline void
 page_array_idal_create_words(struct page_array *pa,
 	 */
 
 	for (i = 0; i < pa->pa_nr; i++)
-		idaws[i] = pa->pa_pfn[i] << PAGE_SHIFT;
+		idaws[i] = page_to_phys(pa->pa_page[i]);
 
 	/* Adjust the first IDAW, since it may not start on a page

boundary */

drivers/s390/cio/vfio_ccw_cp.c +1 -2
 	idaws[0] += pa->pa_iova[0] & (PAGE_SIZE - 1);
@@ -246,8 +246,7 @@ static long copy_from_iova(struct vfio_device
 *vdev, void *to, u64 iova,
 
 	l = n;
 	for (i = 0; i < pa.pa_nr; i++) {
-		struct page *page = pfn_to_page(pa.pa_pfn[i]);
-		void *from = kmap_local_page(page);
+		void *from = kmap_local_page(pa.pa_page[i]);
 
 		m = PAGE_SIZE;
drivers/s390/cio/vfio_ccw_cp.c +0 -0
 		if (i == 0) {
drivers/s390/crypto/vfio_ap_ops.c +1 -1
index 6945e0ddc08c..b0972ca0dfa3 100644
@@ -234,9 +234,9 @@ static struct ap_queue_status
 vfio_ap_irq_enable(struct vfio_ap_queue *q,
 	struct ap_qirq_ctrl aqic_gisa = {};
 	struct ap_queue_status status = {};
 	struct kvm_s390_gisa *gisa;
+	struct page *h_page;
 	int nisc;
 	struct kvm *kvm;
-	unsigned long h_pfn;
 	phys_addr_t h_nib;
 	dma_addr_t nib;
drivers/s390/crypto/vfio_ap_ops.c +1 -1
 	int ret;
@@ -251,7 +251,7 @@ static struct ap_queue_status
 vfio_ap_irq_enable(struct vfio_ap_queue *q,
 	}
 
 	ret = vfio_pin_pages(&q->matrix_mdev->vdev, nib, 1,
-			     IOMMU_READ | IOMMU_WRITE, &h_pfn);
+			     IOMMU_READ | IOMMU_WRITE, &h_page);
 	switch (ret) {
 	case 1:
drivers/s390/crypto/vfio_ap_ops.c +1 -1
 		break;
@@ -267,7 +267,7 @@ static struct ap_queue_status
 vfio_ap_irq_enable(struct vfio_ap_queue *q,
 	kvm = q->matrix_mdev->kvm;
 	gisa = kvm->arch.gisa_int.origin;
 
-	h_nib = (h_pfn << PAGE_SHIFT) | (nib & ~PAGE_MASK);
+	h_nib = page_to_phys(h_page) | (nib & ~PAGE_MASK);
 	aqic_gisa.gisc = isc;
 
drivers/s390/crypto/vfio_ap_ops.c +0 -0
 	nisc = kvm_s390_gisc_register(kvm, isc);
drivers/vfio/vfio.c +0 -0
index b95ec2d78966..96b758e06c4a 100644
@@ -1917,18 +1917,18 @@
 EXPORT_SYMBOL(vfio_set_irqs_validate_and_prepare);
  * @npage [in]   : count of pages to be pinned.  This count should

not
* be greater VFIO_PIN_PAGES_MAX_ENTRIES.
* @prot [in] : protection flags
- * @phys_pfn[out]: array of host PFNs
+ * @pages[out] : array of host pages
* Return error or number of pages pinned.
*/
int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,

drivers/vfio/vfio.c +3 -3
-		   int npage, int prot, unsigned long *phys_pfn)
+		   int npage, int prot, struct page **pages)
 {
 	struct vfio_container *container;
 	struct vfio_group *group = device->group;
 	struct vfio_iommu_driver *driver;
 	int ret;
 
-	if (!phys_pfn || !npage || !vfio_assert_device_open(device))
+	if (!pages || !npage || !vfio_assert_device_open(device))
 		return -EINVAL;
 
 	if (npage > VFIO_PIN_PAGES_MAX_ENTRIES)
@@ -1943,7 +1943,7 @@ int vfio_pin_pages(struct vfio_device *device,
 dma_addr_t iova,
 	if (likely(driver && driver->ops->pin_pages))
 		ret = driver->ops->pin_pages(container->iommu_data,
 					     group->iommu_group, iova,
-					     npage, prot, phys_pfn);
+					     npage, prot, pages);
 	else
 		ret = -ENOTTY;
drivers/vfio/vfio.h +1 -1
index dbcd0e8c031b..dbfad8e20581 100644
@@ -52,7 +52,7 @@ struct vfio_iommu_driver_ops {
 				     struct iommu_group *group,
 				     dma_addr_t user_iova,
 				     int npage, int prot,
-				     unsigned long *phys_pfn);
+				     struct page **pages);
 	void		(*unpin_pages)(void *iommu_data,
 				       dma_addr_t user_iova, int
 npage);
drivers/vfio/vfio.h +0 -0
 	int		(*register_notifier)(void *iommu_data,
drivers/vfio/vfio_iommu_type1.c +1 -1
index f10d0c4b1f26..de342d82b154 100644
@@ -830,7 +830,7 @@ static int vfio_iommu_type1_pin_pages(void
 *iommu_data,
 				      struct iommu_group *iommu_group,
 				      dma_addr_t user_iova,
 				      int npage, int prot,
-				      unsigned long *phys_pfn)
+				      struct page **pages)
 {
 	struct vfio_iommu *iommu = iommu_data;
drivers/vfio/vfio_iommu_type1.c +1 -1
 	struct vfio_iommu_group *group;
@@ -840,7 +840,7 @@ static int vfio_iommu_type1_pin_pages(void
 *iommu_data,
 	bool do_accounting;
 	dma_addr_t iova;
 
-	if (!iommu || !phys_pfn)
+	if (!iommu || !pages)
 		return -EINVAL;
 
drivers/vfio/vfio_iommu_type1.c +1 -0
 	/* Supported for v2 version only */
@@ -879,6 +879,7 @@ static int vfio_iommu_type1_pin_pages(void
 *iommu_data,
 	do_accounting = list_empty(&iommu->domain_list);
 
 	for (i = 0; i < npage; i++) {
+		unsigned long phys_pfn;
 		struct vfio_pfn *vpfn;
 
drivers/vfio/vfio_iommu_type1.c +6 -4
 		iova = user_iova + PAGE_SIZE * i;
@@ -895,23 +896,25 @@ static int vfio_iommu_type1_pin_pages(void
 *iommu_data,
 
 		vpfn = vfio_iova_get_vfio_pfn(dma, iova);
 		if (vpfn) {
-			phys_pfn[i] = vpfn->pfn;
+			pages[i] = pfn_to_page(vpfn->pfn);
 			continue;
 		}
 
 		remote_vaddr = dma->vaddr + (iova - dma->iova);
-		ret = vfio_pin_page_external(dma, remote_vaddr,
 &phys_pfn[i],
+		ret = vfio_pin_page_external(dma, remote_vaddr,
 &phys_pfn,
 					     do_accounting);
 		if (ret)
 			goto pin_unwind;
 
-		ret = vfio_add_to_pfn_list(dma, iova, phys_pfn[i]);
+		ret = vfio_add_to_pfn_list(dma, iova, phys_pfn);
 		if (ret) {
-			if (put_pfn(phys_pfn[i], dma->prot) &&
 do_accounting)
+			if (put_pfn(phys_pfn, dma->prot) &&
 do_accounting)
 				vfio_lock_acct(dma, -1, true);
 			goto pin_unwind;
 		}
 
+		pages[i] = pfn_to_page(phys_pfn);
+
 		if (iommu->dirty_page_tracking) {
 			unsigned long pgshift = __ffs(iommu-
Quoted 1 line

pgsize_bitmap);

drivers/vfio/vfio_iommu_type1.c +2 -2
@@ -934,14 +937,14 @@ static int vfio_iommu_type1_pin_pages(void
 *iommu_data,
 	goto pin_done;
 
 pin_unwind:
-	phys_pfn[i] = 0;
+	pages[i] = NULL;
 	for (j = 0; j < i; j++) {
 		dma_addr_t iova;
 
 		iova = user_iova + PAGE_SIZE * j;
 		dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
 		vfio_unpin_page_external(dma, iova, do_accounting);
-		phys_pfn[j] = 0;
+		pages[j] = NULL;
 	}
 pin_done:
drivers/vfio/vfio_iommu_type1.c +0 -0
 	mutex_unlock(&iommu->lock);
include/linux/vfio.h +1 -1
index 9108de34a79b..c76a2c492bbd 100644
@@ -148,7 +148,7 @@ bool vfio_file_has_dev(struct file *file, struct
 vfio_device *device);
 #define VFIO_PIN_PAGES_MAX_ENTRIES	(PAGE_SIZE/sizeof(unsigned
 long))
 
 int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
-		   int npage, int prot, unsigned long *phys_pfn);
+		   int npage, int prot, struct page **pages);
 void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova,

int npage);
int vfio_dma_rw(struct vfio_device *device, dma_addr_t iova,
void *data, size_t len, bool write);

#45 lore raw
In reply to Nicolin Chen
Quoted 18 lines

From: Nicolin Chen
Sent: Wednesday, July 6, 2022 2:28 PM

This is a preparatory series for IOMMUFD v2 patches. It prepares for replacing vfio_iommu_type1 implementations of vfio_pin/unpin_pages() with IOMMUFD version.

There's a gap between these two versions: the vfio_iommu_type1 version inputs a non-contiguous PFN list and outputs another PFN list for the pinned physical page list, while the IOMMUFD version only supports a contiguous address input by accepting the starting IO virtual address of a set of pages to pin and by outputting to a physical page list.

The nature of existing callers mostly aligns with the IOMMUFD version, except s390's vfio_ccw_cp code where some additional change is needed along with this series. Overall, updating to "iova" and "phys_page" does improve the caller side to some extent.

Quoted 9 lines

Also fix a misuse of physical address and virtual address in the s390's crypto code. And update the input naming at the adjacent vfio_dma_rw().

This is on github:
https://github.com/nicolinc/iommufd/commits/vfio_pin_pages

Request for testing: I only did build for s390 and i915 code, so it'd be nice to have people who have environment to run sanity accordingly.

+Terrence who is testing it for i915 now...

#46 lore raw
Quoted 5 lines · Kevin
Quoted 3 lines

Request for testing: I only did build for s390 and i915 code, so it'd be nice to have people who have environment to run sanity accordingly.

+Terrence who is testing it for i915 now...

Hi Terrence, would it be possible for you to pull v3 to test on?
https://github.com/nicolinc/iommufd/commits/dev/vfio_pin_pages-v3

They are basically same but there's a new DIV_ROUND_UP change, which shouldn't result in any functional difference, IMHO. If v3 passes, I can simply add your Tested-by when I respin it.

Thanks
Nic

#47 lore raw
Quoted 8 lines

-----Original Message-----
From: intel-gvt-dev <[email protected]> On Behalf Of

Quoted 4 lines · Kevin
Quoted 3 lines

Request for testing: I only did build for s390 and i915 code, so it'd be nice to have people who have environment to run sanity accordingly.

Quoted 8 lines
Quoted 1 line

+Terrence who is testing it for i915 now...

Hi Terrence, would it be possible for you to pull v3 to test on?
https://github.com/nicolinc/iommufd/commits/dev/vfio_pin_pages-v3

They are basically same but there's a new DIV_ROUND_UP change, which shouldn't result in any functional difference, IMHO. If v3 passes, I can simply add your Tested-by when I respin it.

Hi Nicolin, I already completed KVMGT key feature testing based on your v3 repo, VM booted up successfully and run smoothly, but there is a call trace during each time VM booting up, as the attachment.

[ 4379.881770] ------------[ cut here ]------------
[ 4379.881776] WARNING: CPU: 1 PID: 4042 at drivers/vfio/vfio_iommu_type1.c:1167 vfio_remove_dma+0xb3/0xc0 [vfio_iommu_type1]
[ 4379.881818] Modules linked in: binfmt_misc nfnetlink bridge stp llc rfcomm cmac algif_hash algif_skcipher af_alg bnep snd_hda_codec_hdmi snd_hda_codec_realtek snd_hda_codec_generic ledtrig_audio snd_hda_intel snd_intel_dspcfg snd_hda_codec snd_hda_core snd_hwdep snd_pcm intel_rapl_msr intel_rapl_common x86_pkg_temp_thermal intel_powerclamp coretemp snd_seq_midi snd_seq_midi_event snd_rawmidi kvm_intel btusb snd_seq btrtl rapl btbcm btintel snd_seq_device intel_cstate input_leds snd_timer bluetooth serio_raw mei_me ecdh_generic mxm_wmi ecc snd soundcore mei mac_hid acpi_pad sch_fq_codel kvmgt mdev vfio_iommu_type1 vfio kvm irqbypass parport_pc ppdev lp parport nfsd auth_rpcgss nfs_acl lockd grace sunrpc ip_tables x_tables autofs4 btrfs blake2b_generic zstd_compress raid10 raid456 async_raid6_recov async_memcpy async_pq async_xor async_tx xor raid6_pq libcrc32c raid1 raid0 multipath linear dm_mirror dm_region_hash dm_log i915 i2c_algo_bit cec rc_core drm_buddy ttm drm_display_helper
[ 4379.882092]  drm_kms_helper crct10dif_pclmul syscopyarea sysfillrect crc32_pclmul sysimgblt ghash_clmulni_intel fb_sys_fops aesni_intel crypto_simd e1000e cryptd i2c_i801 psmouse ptp drm i2c_smbus ahci pps_core libahci hid_generic wmi video usbhid hid
[ 4379.882173] CPU: 1 PID: 4042 Comm: qemu-system-x86 Tainted: G        W         5.19.0-rc1-next-20220606+ #1
[ 4379.882187] Hardware name: Supermicro C7Z370-CG-IW/C7Z370-CG-IW, BIOS 1.1 02/08/2018
[ 4379.882194] RIP: 0010:vfio_remove_dma+0xb3/0xc0 [vfio_iommu_type1]
[ 4379.882218] Code: d5 48 89 df e8 ee 28 67 d5 41 83 44 24 78 01 5b 41 5c 5d c3 e8 8e 57 3f d5 eb b3 be 03 00 00 00 48 89 d7 e8 df 11 94 d5 eb a4 <0f> 0b e9 65 ff ff ff 66 0f 1f 44 00 00 0f 1f 44 00 00 55 48 89 e5
[ 4379.882230] RSP: 0018:ffffad03835a7bc8 EFLAGS: 00010282
[ 4379.882242] RAX: ffff8df213fa0380 RBX: ffff8df1d2d28060 RCX: 00000000802a0009
[ 4379.882251] RDX: 0000000000000000 RSI: ffff8df1d2d28060 RDI: ffff8df2c2ce1180
[ 4379.882259] RBP: ffffad03835a7bd8 R08: ffff8df1d2d28300 R09: 00000000802a0009
[ 4379.882267] R10: ffff8df213d1dc00 R11: fffffffffffffffe R12: ffff8df2c2ce1180
[ 4379.882275] R13: ffff8df1c87cf900 R14: ffff8df2c2ce11a0 R15: ffff8df1c71a9900
[ 4379.882284] FS:  0000000000000000(0000) GS:ffff8df91aa40000(0000) knlGS:0000000000000000
[ 4379.882294] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 4379.882302] CR2: 00007fbf5c950290 CR3: 00000002b5886003 CR4: 00000000003726e0
[ 4379.882311] Call Trace:
[ 4379.882316]  <TASK>
[ 4379.882325]  vfio_iommu_unmap_unpin_all+0x20/0x40 [vfio_iommu_type1]
[ 4379.882348]  vfio_iommu_type1_detach_group+0x123/0x560 [vfio_iommu_type1]
[ 4379.882369]  ? security_file_free+0x54/0x60
[ 4379.882388]  __vfio_group_unset_container+0x4c/0x1b0 [vfio]
[ 4379.882415]  vfio_group_fops_release+0x52/0x80 [vfio]
[ 4379.882438]  __fput+0x99/0x260
[ 4379.882451]  ____fput+0xe/0x10
[ 4379.882460]  task_work_run+0x6c/0xa0
[ 4379.882475]  do_exit+0x355/0xb70
[ 4379.882492]  do_group_exit+0x35/0xa0
[ 4379.882506]  get_signal+0x946/0x950
[ 4379.882519]  arch_do_signal_or_restart+0x37/0x770
[ 4379.882536]  ? do_futex+0x118/0x1a0
[ 4379.882554]  exit_to_user_mode_prepare+0x9b/0x180
[ 4379.882569]  syscall_exit_to_user_mode+0x29/0x40
[ 4379.882582]  do_syscall_64+0x46/0x80
[ 4379.882597]  entry_SYSCALL_64_after_hwframe+0x46/0xb0
[ 4379.882610] RIP: 0033:0x7f9ffcb97376
[ 4379.882621] Code: Unable to access opcode bytes at RIP 0x7f9ffcb9734c.
[ 4379.882627] RSP: 002b:00007f9ff9d275b0 EFLAGS: 00000282 ORIG_RAX: 00000000000000ca
[ 4379.882639] RAX: fffffffffffffe00 RBX: 0000000000000000 RCX: 00007f9ffcb97376
[ 4379.882647] RDX: 0000000000000000 RSI: 0000000000000080 RDI: 00005567e17a8f2c
[ 4379.882655] RBP: 00005567e17a8f00 R08: 0000000000000000 R09: 0000000000000004
[ 4379.882662] R10: 0000000000000000 R11: 0000000000000282 R12: 00005567e17a8f24
[ 4379.882669] R13: 00005567df8f5200 R14: 00007f9ff9d275f0 R15: 00005567e17a8f2c
[ 4379.882686]  </TASK>
[ 4379.882692] ---[ end trace 0000000000000000 ]---
[ 4423.824057] device tap1 left promiscuous mode
[ 4423.824085] sw0: port 3(tap1) entered disabled state
#48 lore raw
Quoted 16 lines · Terrence

External email: Use caution opening links or attachments

Quoted 13 lines

-----Original Message-----
From: intel-gvt-dev <[email protected]> On Behalf Of

Quoted 5 lines · Kevin
Quoted 3 lines

Request for testing: I only did build for s390 and i915 code, so it'd be nice to have people who have environment to run sanity accordingly.

+Terrence who is testing it for i915 now...

Hi Terrence, would it be possible for you to pull v3 to test on?
https://github.com/nicolinc/iommufd/commits/dev/vfio_pin_pages-v3

Quoted 7 lines
Quoted 3 lines

They are basically same but there's a new DIV_ROUND_UP change, which shouldn't result in any functional difference, IMHO. If v3 passes, I can simply add your Tested-by when I respin it.

Hi Nicolin, I already completed KVMGT key feature testing based on your v3 repo, VM booted up successfully and run smoothly, but there is a call trace during each time VM booting up, as the attachment.

Nice! Thank you for the testing. I will add your Tested-by in v3.

maintainer Tested-by on the v2 series
#49 lore raw
In reply to Nicolin Chen
Quoted 22 lines · Nicolin Chen

This is a preparatory series for IOMMUFD v2 patches. It prepares for replacing vfio_iommu_type1 implementations of vfio_pin/unpin_pages() with IOMMUFD version.

There's a gap between these two versions: the vfio_iommu_type1 version
inputs a non-contiguous PFN list and outputs another PFN list for the pinned physical page list, while the IOMMUFD version only supports a contiguous address input by accepting the starting IO virtual address of a set of pages to pin and by outputting to a physical page list.

The nature of existing callers mostly aligns with the IOMMUFD version,
except s390's vfio_ccw_cp code where some additional change is needed along with this series. Overall, updating to "iova" and "phys_page" does improve the caller side to some extent.

Also fix a misuse of physical address and virtual address in the s390's
crypto code. And update the input naming at the adjacent vfio_dma_rw().

Quoted 6 lines

This is on github:
https://github.com/nicolinc/iommufd/commits/vfio_pin_pages

Request for testing: I only did build for s390 and i915 code, so it'd be nice to have people who have environment to run sanity accordingly.

Quoted 39 lines

Thanks!

Changelog
v2:
* Added a patch to make vfio_unpin_pages return void
* Added two patches to remove PFN list from two s390 callers
* Renamed "phys_page" parameter to "pages" for vfio_pin_pages
* Updated commit log of kmap_local_page() patch
* Added Harald's "Reviewed-by" to pa_ind patch
* Rebased on top of Alex's extern removal path
v1:
https://lore.kernel.org/kvm/[email protected]/

Nicolin Chen (9):
vfio: Make vfio_unpin_pages() return void vfio/ap: Pass in physical address of ind to ap_aqic() vfio/ccw: Only pass in contiguous pages
vfio: Pass in starting IOVA to vfio_pin/unpin_pages API vfio/ap: Remove redundant pfn
vfio/ccw: Change pa_pfn list to pa_iova list
vfio: Rename user_iova of vfio_dma_rw() vfio/ccw: Add kmap_local_page() for memcpy
vfio: Replace phys_pfn with pages for vfio_pin_pages()

 .../driver-api/vfio-mediated-device.rst       |   6 +-
 arch/s390/include/asm/ap.h                    |   6 +-
 drivers/gpu/drm/i915/gvt/kvmgt.c              |  46 ++---
 drivers/s390/cio/vfio_ccw_cp.c                | 195 +++++++++++-----
--
 drivers/s390/crypto/ap_queue.c                |   2 +-
 drivers/s390/crypto/vfio_ap_ops.c             |  54 +++--
 drivers/s390/crypto/vfio_ap_private.h         |   4 +-
 drivers/vfio/vfio.c                           |  55 +++--
 drivers/vfio/vfio.h                           |   8 +-
 drivers/vfio/vfio_iommu_type1.c               |  46 +++--
 include/linux/vfio.h                          |   9 +-
 11 files changed, 218 insertions(+), 213 deletions(-)
#50 lore raw
Quoted 26 lines · Eric Farman

External email: Use caution opening links or attachments

Quoted 22 lines · Nicolin Chen

This is a preparatory series for IOMMUFD v2 patches. It prepares for replacing vfio_iommu_type1 implementations of vfio_pin/unpin_pages() with IOMMUFD version.

There's a gap between these two versions: the vfio_iommu_type1 version
inputs a non-contiguous PFN list and outputs another PFN list for the pinned physical page list, while the IOMMUFD version only supports a contiguous address input by accepting the starting IO virtual address of a set of pages to pin and by outputting to a physical page list.

The nature of existing callers mostly aligns with the IOMMUFD version,
except s390's vfio_ccw_cp code where some additional change is needed along with this series. Overall, updating to "iova" and "phys_page" does improve the caller side to some extent.

Also fix a misuse of physical address and virtual address in the s390's
crypto code. And update the input naming at the adjacent vfio_dma_rw().

Quoted 8 lines
Quoted 6 lines

This is on github:
https://github.com/nicolinc/iommufd/commits/vfio_pin_pages

Request for testing: I only did build for s390 and i915 code, so it'd be nice to have people who have environment to run sanity accordingly.

Thank you for the review and test!