Merge tag 'vfio-v6.9-rc1' of https://github.com/awilliam/linux-vfio

Pull VFIO updates from Alex Williamson:

 - Add warning in unlikely case that device is not captured with
   driver_override (Kunwu Chan)

 - Error handling improvements in mlx5-vfio-pci to detect firmware
   tracking object error states, logging of firmware error syndrom, and
   releasing of firmware resources in aborted migration sequence (Yishai
   Hadas)

 - Correct an un-alphabetized VFIO MAINTAINERS entry (Alex Williamson)

 - Make the mdev_bus_type const and also make the class struct const for
   a couple of the vfio-mdev sample drivers (Ricardo B. Marliere)

 - Addition of a new vfio-pci variant driver for the GPU of NVIDIA's
   Grace-Hopper superchip. During initialization of the chip-to-chip
   interconnect in this hardware module, the PCI BARs of the device
   become unused in favor of a faster, coherent mechanism for exposing
   device memory. This driver primarily changes the VFIO representation
   of the device to masquerade this coherent aperture to replace the
   physical PCI BARs for userspace drivers. This also incorporates use
   of a new vma flag allowing KVM to use write combining attributes for
   uncached device memory (Ankit Agrawal)

 - Reset fixes and cleanups for the pds-vfio-pci driver. Save and
   restore files were previously leaked if the device didn't pass
   through an error state, this is resolved and later re-fixed to
   prevent access to the now freed files. Reset handling is also
   refactored to remove the complicated deferred reset mechanism (Brett
   Creeley)

 - Remove some references to pl330 in the vfio-platform amba driver
   (Geert Uytterhoeven)

 - Remove twice redundant and ugly code to unpin incidental pins of the
   zero-page (Alex Williamson)

 - Deferred reset logic is also removed from the hisi-acc-vfio-pci
   driver as a simplification (Shameer Kolothum)

 - Enforce that mlx5-vfio-pci devices must support PRE_COPY and remove
   resulting unnecessary code. There is no device firmware that has been
   available publicly without this support (Yishai Hadas)

 - Switch over to using the .remove_new callback for vfio-platform in
   support of the broader transition for a void remove function (Uwe
   Kleine-König)

 - Resolve multiple issues in interrupt code for VFIO bus drivers that
   allow calling eventfd_signal() on a NULL context. This also remove a
   potential race in INTx setup on certain hardware for vfio-pci, races
   with various mechanisms to mask INTx, and leaked virqfds in
   vfio-platform (Alex Williamson)

* tag 'vfio-v6.9-rc1' of https://github.com/awilliam/linux-vfio: (29 commits)
  vfio/fsl-mc: Block calling interrupt handler without trigger
  vfio/platform: Create persistent IRQ handlers
  vfio/platform: Disable virqfds on cleanup
  vfio/pci: Create persistent INTx handler
  vfio: Introduce interface to flush virqfd inject workqueue
  vfio/pci: Lock external INTx masking ops
  vfio/pci: Disable auto-enable of exclusive INTx IRQ
  vfio/pds: Refactor/simplify reset logic
  vfio/pds: Make sure migration file isn't accessed after reset
  vfio/platform: Convert to platform remove callback returning void
  vfio/mlx5: Enforce PRE_COPY support
  vfio/mbochs: make mbochs_class constant
  vfio/mdpy: make mdpy_class constant
  hisi_acc_vfio_pci: Remove the deferred_reset logic
  Revert "vfio/type1: Unpin zero pages"
  vfio/nvgrace-gpu: Convey kvm to map device memory region as noncached
  vfio: amba: Rename pl330_ids[] to vfio_amba_ids[]
  vfio/pds: Always clear the save/restore FDs on reset
  vfio/nvgrace-gpu: Add vfio pci variant module for grace hopper
  vfio/pci: rename and export range_intersect_range
  ...

29 files changed, 1416 insertions, 426 deletions

diff --git a/drivers/vfio/fsl-mc/vfio_fsl_mc_intr.c b/drivers/vfio/fsl-mc/vfio_fsl_mc_intr.c
index d62fbfff20b8..82b2afa9b7e3 100644
--- a/drivers/vfio/fsl-mc/vfio_fsl_mc_intr.c
+++ b/drivers/vfio/fsl-mc/vfio_fsl_mc_intr.c
@@ -141,13 +141,14 @@ static int vfio_fsl_mc_set_irq_trigger(struct vfio_fsl_mc_device *vdev,
 	irq = &vdev->mc_irqs[index];
 
 	if (flags & VFIO_IRQ_SET_DATA_NONE) {
-		vfio_fsl_mc_irq_handler(hwirq, irq);
+		if (irq->trigger)
+			eventfd_signal(irq->trigger);
 
 	} else if (flags & VFIO_IRQ_SET_DATA_BOOL) {
 		u8 trigger = *(u8 *)data;
 
-		if (trigger)
-			vfio_fsl_mc_irq_handler(hwirq, irq);
+		if (trigger && irq->trigger)
+			eventfd_signal(irq->trigger);
 	}
 
 	return 0;
diff --git a/drivers/vfio/mdev/mdev_driver.c b/drivers/vfio/mdev/mdev_driver.c
index 7825d83a55f8..b98322966b3e 100644
--- a/drivers/vfio/mdev/mdev_driver.c
+++ b/drivers/vfio/mdev/mdev_driver.c
@@ -40,7 +40,7 @@ static int mdev_match(struct device *dev, struct device_driver *drv)
 	return 0;
 }
 
-struct bus_type mdev_bus_type = {
+const struct bus_type mdev_bus_type = {
 	.name		= "mdev",
 	.probe		= mdev_probe,
 	.remove		= mdev_remove,
diff --git a/drivers/vfio/mdev/mdev_private.h b/drivers/vfio/mdev/mdev_private.h
index af457b27f607..63a1316b08b7 100644
--- a/drivers/vfio/mdev/mdev_private.h
+++ b/drivers/vfio/mdev/mdev_private.h
@@ -13,7 +13,7 @@
 int  mdev_bus_register(void);
 void mdev_bus_unregister(void);
 
-extern struct bus_type mdev_bus_type;
+extern const struct bus_type mdev_bus_type;
 extern const struct attribute_group *mdev_device_groups[];
 
 #define to_mdev_type_attr(_attr)	\
diff --git a/drivers/vfio/pci/Kconfig b/drivers/vfio/pci/Kconfig
index 18c397df566d..15821a2d77d2 100644
--- a/drivers/vfio/pci/Kconfig
+++ b/drivers/vfio/pci/Kconfig
@@ -67,4 +67,6 @@ source "drivers/vfio/pci/pds/Kconfig"
 
 source "drivers/vfio/pci/virtio/Kconfig"
 
+source "drivers/vfio/pci/nvgrace-gpu/Kconfig"
+
 endmenu
diff --git a/drivers/vfio/pci/Makefile b/drivers/vfio/pci/Makefile
index 046139a4eca5..ce7a61f1d912 100644
--- a/drivers/vfio/pci/Makefile
+++ b/drivers/vfio/pci/Makefile
@@ -15,3 +15,5 @@ obj-$(CONFIG_HISI_ACC_VFIO_PCI) += hisilicon/
 obj-$(CONFIG_PDS_VFIO_PCI) += pds/
 
 obj-$(CONFIG_VIRTIO_VFIO_PCI) += virtio/
+
+obj-$(CONFIG_NVGRACE_GPU_VFIO_PCI) += nvgrace-gpu/
diff --git a/drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c b/drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c
index 4d27465c8f1a..9a3e97108ace 100644
--- a/drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c
+++ b/drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c
@@ -630,25 +630,11 @@ static void hisi_acc_vf_disable_fds(struct hisi_acc_vf_core_device *hisi_acc_vde
 	}
 }
 
-/*
- * This function is called in all state_mutex unlock cases to
- * handle a 'deferred_reset' if exists.
- */
-static void
-hisi_acc_vf_state_mutex_unlock(struct hisi_acc_vf_core_device *hisi_acc_vdev)
+static void hisi_acc_vf_reset(struct hisi_acc_vf_core_device *hisi_acc_vdev)
 {
-again:
-	spin_lock(&hisi_acc_vdev->reset_lock);
-	if (hisi_acc_vdev->deferred_reset) {
-		hisi_acc_vdev->deferred_reset = false;
-		spin_unlock(&hisi_acc_vdev->reset_lock);
-		hisi_acc_vdev->vf_qm_state = QM_NOT_READY;
-		hisi_acc_vdev->mig_state = VFIO_DEVICE_STATE_RUNNING;
-		hisi_acc_vf_disable_fds(hisi_acc_vdev);
-		goto again;
-	}
-	mutex_unlock(&hisi_acc_vdev->state_mutex);
-	spin_unlock(&hisi_acc_vdev->reset_lock);
+	hisi_acc_vdev->vf_qm_state = QM_NOT_READY;
+	hisi_acc_vdev->mig_state = VFIO_DEVICE_STATE_RUNNING;
+	hisi_acc_vf_disable_fds(hisi_acc_vdev);
 }
 
 static void hisi_acc_vf_start_device(struct hisi_acc_vf_core_device *hisi_acc_vdev)
@@ -804,8 +790,10 @@ static long hisi_acc_vf_precopy_ioctl(struct file *filp,
 
 	info.dirty_bytes = 0;
 	info.initial_bytes = migf->total_length - *pos;
+	mutex_unlock(&migf->lock);
+	mutex_unlock(&hisi_acc_vdev->state_mutex);
 
-	ret = copy_to_user((void __user *)arg, &info, minsz) ? -EFAULT : 0;
+	return copy_to_user((void __user *)arg, &info, minsz) ? -EFAULT : 0;
 out:
 	mutex_unlock(&migf->lock);
 	mutex_unlock(&hisi_acc_vdev->state_mutex);
@@ -1071,7 +1059,7 @@ hisi_acc_vfio_pci_set_device_state(struct vfio_device *vdev,
 			break;
 		}
 	}
-	hisi_acc_vf_state_mutex_unlock(hisi_acc_vdev);
+	mutex_unlock(&hisi_acc_vdev->state_mutex);
 	return res;
 }
 
@@ -1092,7 +1080,7 @@ hisi_acc_vfio_pci_get_device_state(struct vfio_device *vdev,
 
 	mutex_lock(&hisi_acc_vdev->state_mutex);
 	*curr_state = hisi_acc_vdev->mig_state;
-	hisi_acc_vf_state_mutex_unlock(hisi_acc_vdev);
+	mutex_unlock(&hisi_acc_vdev->state_mutex);
 	return 0;
 }
 
@@ -1104,21 +1092,9 @@ static void hisi_acc_vf_pci_aer_reset_done(struct pci_dev *pdev)
 				VFIO_MIGRATION_STOP_COPY)
 		return;
 
-	/*
-	 * As the higher VFIO layers are holding locks across reset and using
-	 * those same locks with the mm_lock we need to prevent ABBA deadlock
-	 * with the state_mutex and mm_lock.
-	 * In case the state_mutex was taken already we defer the cleanup work
-	 * to the unlock flow of the other running context.
-	 */
-	spin_lock(&hisi_acc_vdev->reset_lock);
-	hisi_acc_vdev->deferred_reset = true;
-	if (!mutex_trylock(&hisi_acc_vdev->state_mutex)) {
-		spin_unlock(&hisi_acc_vdev->reset_lock);
-		return;
-	}
-	spin_unlock(&hisi_acc_vdev->reset_lock);
-	hisi_acc_vf_state_mutex_unlock(hisi_acc_vdev);
+	mutex_lock(&hisi_acc_vdev->state_mutex);
+	hisi_acc_vf_reset(hisi_acc_vdev);
+	mutex_unlock(&hisi_acc_vdev->state_mutex);
 }
 
 static int hisi_acc_vf_qm_init(struct hisi_acc_vf_core_device *hisi_acc_vdev)
diff --git a/drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h b/drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h
index dcabfeec6ca1..5bab46602fad 100644
--- a/drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h
+++ b/drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h
@@ -98,8 +98,8 @@ struct hisi_acc_vf_migration_file {
 
 struct hisi_acc_vf_core_device {
 	struct vfio_pci_core_device core_device;
-	u8 match_done:1;
-	u8 deferred_reset:1;
+	u8 match_done;
+
 	/* For migration state */
 	struct mutex state_mutex;
 	enum vfio_device_mig_state mig_state;
@@ -109,8 +109,6 @@ struct hisi_acc_vf_core_device {
 	struct hisi_qm vf_qm;
 	u32 vf_qm_state;
 	int vf_id;
-	/* For reset handler */
-	spinlock_t reset_lock;
 	struct hisi_acc_vf_migration_file *resuming_migf;
 	struct hisi_acc_vf_migration_file *saving_migf;
 };
diff --git a/drivers/vfio/pci/mlx5/cmd.c b/drivers/vfio/pci/mlx5/cmd.c
index efd1d252cdc9..41a4b0cf4297 100644
--- a/drivers/vfio/pci/mlx5/cmd.c
+++ b/drivers/vfio/pci/mlx5/cmd.c
@@ -108,8 +108,9 @@ int mlx5vf_cmd_query_vhca_migration_state(struct mlx5vf_pci_core_device *mvdev,
 		ret = wait_for_completion_interruptible(&mvdev->saving_migf->save_comp);
 		if (ret)
 			return ret;
-		if (mvdev->saving_migf->state ==
-		    MLX5_MIGF_STATE_PRE_COPY_ERROR) {
+		/* Upon cleanup, ignore previous pre_copy error state */
+		if (mvdev->saving_migf->state == MLX5_MIGF_STATE_PRE_COPY_ERROR &&
+		    !(query_flags & MLX5VF_QUERY_CLEANUP)) {
 			/*
 			 * In case we had a PRE_COPY error, only query full
 			 * image for final image
@@ -121,6 +122,11 @@ int mlx5vf_cmd_query_vhca_migration_state(struct mlx5vf_pci_core_device *mvdev,
 			}
 			query_flags &= ~MLX5VF_QUERY_INC;
 		}
+		/* Block incremental query which is state-dependent */
+		if (mvdev->saving_migf->state == MLX5_MIGF_STATE_ERROR) {
+			complete(&mvdev->saving_migf->save_comp);
+			return -ENODEV;
+		}
 	}
 
 	MLX5_SET(query_vhca_migration_state_in, in, opcode,
@@ -149,6 +155,12 @@ int mlx5vf_cmd_query_vhca_migration_state(struct mlx5vf_pci_core_device *mvdev,
 	return 0;
 }
 
+static void set_tracker_change_event(struct mlx5vf_pci_core_device *mvdev)
+{
+	mvdev->tracker.object_changed = true;
+	complete(&mvdev->tracker_comp);
+}
+
 static void set_tracker_error(struct mlx5vf_pci_core_device *mvdev)
 {
 	/* Mark the tracker under an error and wake it up if it's running */
@@ -189,7 +201,7 @@ void mlx5vf_cmd_close_migratable(struct mlx5vf_pci_core_device *mvdev)
 	/* Must be done outside the lock to let it progress */
 	set_tracker_error(mvdev);
 	mutex_lock(&mvdev->state_mutex);
-	mlx5vf_disable_fds(mvdev);
+	mlx5vf_disable_fds(mvdev, NULL);
 	_mlx5vf_free_page_tracker_resources(mvdev);
 	mlx5vf_state_mutex_unlock(mvdev);
 }
@@ -221,6 +233,10 @@ void mlx5vf_cmd_set_migratable(struct mlx5vf_pci_core_device *mvdev,
 	if (!MLX5_CAP_GEN(mvdev->mdev, migration))
 		goto end;
 
+	if (!(MLX5_CAP_GEN_2(mvdev->mdev, migration_multi_load) &&
+	      MLX5_CAP_GEN_2(mvdev->mdev, migration_tracking_state)))
+		goto end;
+
 	mvdev->vf_id = pci_iov_vf_id(pdev);
 	if (mvdev->vf_id < 0)
 		goto end;
@@ -250,17 +266,14 @@ void mlx5vf_cmd_set_migratable(struct mlx5vf_pci_core_device *mvdev,
 	mvdev->migrate_cap = 1;
 	mvdev->core_device.vdev.migration_flags =
 		VFIO_MIGRATION_STOP_COPY |
-		VFIO_MIGRATION_P2P;
+		VFIO_MIGRATION_P2P |
+		VFIO_MIGRATION_PRE_COPY;
+
 	mvdev->core_device.vdev.mig_ops = mig_ops;
 	init_completion(&mvdev->tracker_comp);
 	if (MLX5_CAP_GEN(mvdev->mdev, adv_virtualization))
 		mvdev->core_device.vdev.log_ops = log_ops;
 
-	if (MLX5_CAP_GEN_2(mvdev->mdev, migration_multi_load) &&
-	    MLX5_CAP_GEN_2(mvdev->mdev, migration_tracking_state))
-		mvdev->core_device.vdev.migration_flags |=
-			VFIO_MIGRATION_PRE_COPY;
-
 	if (MLX5_CAP_GEN_2(mvdev->mdev, migration_in_chunks))
 		mvdev->chunk_mode = 1;
 
@@ -402,6 +415,50 @@ void mlx5vf_free_data_buffer(struct mlx5_vhca_data_buffer *buf)
 	kfree(buf);
 }
 
+static int mlx5vf_add_migration_pages(struct mlx5_vhca_data_buffer *buf,
+				      unsigned int npages)
+{
+	unsigned int to_alloc = npages;
+	struct page **page_list;
+	unsigned long filled;
+	unsigned int to_fill;
+	int ret;
+
+	to_fill = min_t(unsigned int, npages, PAGE_SIZE / sizeof(*page_list));
+	page_list = kvzalloc(to_fill * sizeof(*page_list), GFP_KERNEL_ACCOUNT);
+	if (!page_list)
+		return -ENOMEM;
+
+	do {
+		filled = alloc_pages_bulk_array(GFP_KERNEL_ACCOUNT, to_fill,
+						page_list);
+		if (!filled) {
+			ret = -ENOMEM;
+			goto err;
+		}
+		to_alloc -= filled;
+		ret = sg_alloc_append_table_from_pages(
+			&buf->table, page_list, filled, 0,
+			filled << PAGE_SHIFT, UINT_MAX, SG_MAX_SINGLE_ALLOC,
+			GFP_KERNEL_ACCOUNT);
+
+		if (ret)
+			goto err;
+		buf->allocated_length += filled * PAGE_SIZE;
+		/* clean input for another bulk allocation */
+		memset(page_list, 0, filled * sizeof(*page_list));
+		to_fill = min_t(unsigned int, to_alloc,
+				PAGE_SIZE / sizeof(*page_list));
+	} while (to_alloc > 0);
+
+	kvfree(page_list);
+	return 0;
+
+err:
+	kvfree(page_list);
+	return ret;
+}
+
 struct mlx5_vhca_data_buffer *
 mlx5vf_alloc_data_buffer(struct mlx5_vf_migration_file *migf,
 			 size_t length,
@@ -608,8 +665,13 @@ static void mlx5vf_save_callback(int status, struct mlx5_async_work *context)
 
 err:
 	/* The error flow can't run from an interrupt context */
-	if (status == -EREMOTEIO)
+	if (status == -EREMOTEIO) {
 		status = MLX5_GET(save_vhca_state_out, async_data->out, status);
+		/* Failed in FW, print cmd out failure details */
+		mlx5_cmd_out_err(migf->mvdev->mdev, MLX5_CMD_OP_SAVE_VHCA_STATE, 0,
+				 async_data->out);
+	}
+
 	async_data->status = status;
 	queue_work(migf->mvdev->cb_wq, &async_data->work);
 }
@@ -623,6 +685,7 @@ int mlx5vf_cmd_save_vhca_state(struct mlx5vf_pci_core_device *mvdev,
 	u32 in[MLX5_ST_SZ_DW(save_vhca_state_in)] = {};
 	struct mlx5_vhca_data_buffer *header_buf = NULL;
 	struct mlx5vf_async_data *async_data;
+	bool pre_copy_cleanup = false;
 	int err;
 
 	lockdep_assert_held(&mvdev->state_mutex);
@@ -633,6 +696,10 @@ int mlx5vf_cmd_save_vhca_state(struct mlx5vf_pci_core_device *mvdev,
 	if (err)
 		return err;
 
+	if ((migf->state == MLX5_MIGF_STATE_PRE_COPY ||
+	     migf->state == MLX5_MIGF_STATE_PRE_COPY_ERROR) && !track && !inc)
+		pre_copy_cleanup = true;
+
 	if (migf->state == MLX5_MIGF_STATE_PRE_COPY_ERROR)
 		/*
 		 * In case we had a PRE_COPY error, SAVE is triggered only for
@@ -651,29 +718,27 @@ int mlx5vf_cmd_save_vhca_state(struct mlx5vf_pci_core_device *mvdev,
 
 	async_data = &migf->async_data;
 	async_data->buf = buf;
-	async_data->stop_copy_chunk = !track;
+	async_data->stop_copy_chunk = (!track && !pre_copy_cleanup);
 	async_data->out = kvzalloc(out_size, GFP_KERNEL);
 	if (!async_data->out) {
 		err = -ENOMEM;
 		goto err_out;
 	}
 
-	if (MLX5VF_PRE_COPY_SUPP(mvdev)) {
-		if (async_data->stop_copy_chunk) {
-			u8 header_idx = buf->stop_copy_chunk_num ?
-				buf->stop_copy_chunk_num - 1 : 0;
+	if (async_data->stop_copy_chunk) {
+		u8 header_idx = buf->stop_copy_chunk_num ?
+			buf->stop_copy_chunk_num - 1 : 0;
 
-			header_buf = migf->buf_header[header_idx];
-			migf->buf_header[header_idx] = NULL;
-		}
+		header_buf = migf->buf_header[header_idx];
+		migf->buf_header[header_idx] = NULL;
+	}
 
-		if (!header_buf) {
-			header_buf = mlx5vf_get_data_buffer(migf,
-				sizeof(struct mlx5_vf_migration_header), DMA_NONE);
-			if (IS_ERR(header_buf)) {
-				err = PTR_ERR(header_buf);
-				goto err_free;
-			}
+	if (!header_buf) {
+		header_buf = mlx5vf_get_data_buffer(migf,
+			sizeof(struct mlx5_vf_migration_header), DMA_NONE);
+		if (IS_ERR(header_buf)) {
+			err = PTR_ERR(header_buf);
+			goto err_free;
 		}
 	}
 
@@ -900,6 +965,29 @@ static int mlx5vf_cmd_modify_tracker(struct mlx5_core_dev *mdev,
 	return mlx5_cmd_exec(mdev, in, sizeof(in), out, sizeof(out));
 }
 
+static int mlx5vf_cmd_query_tracker(struct mlx5_core_dev *mdev,
+				    struct mlx5_vhca_page_tracker *tracker)
+{
+	u32 out[MLX5_ST_SZ_DW(query_page_track_obj_out)] = {};
+	u32 in[MLX5_ST_SZ_DW(general_obj_in_cmd_hdr)] = {};
+	void *obj_context;
+	void *cmd_hdr;
+	int err;
+
+	cmd_hdr = MLX5_ADDR_OF(modify_page_track_obj_in, in, general_obj_in_cmd_hdr);
+	MLX5_SET(general_obj_in_cmd_hdr, cmd_hdr, opcode, MLX5_CMD_OP_QUERY_GENERAL_OBJECT);
+	MLX5_SET(general_obj_in_cmd_hdr, cmd_hdr, obj_type, MLX5_OBJ_TYPE_PAGE_TRACK);
+	MLX5_SET(general_obj_in_cmd_hdr, cmd_hdr, obj_id, tracker->id);
+
+	err = mlx5_cmd_exec(mdev, in, sizeof(in), out, sizeof(out));
+	if (err)
+		return err;
+
+	obj_context = MLX5_ADDR_OF(query_page_track_obj_out, out, obj_context);
+	tracker->status = MLX5_GET(page_track, obj_context, state);
+	return 0;
+}
+
 static int alloc_cq_frag_buf(struct mlx5_core_dev *mdev,
 			     struct mlx5_vhca_cq_buf *buf, int nent,
 			     int cqe_size)
@@ -957,9 +1045,11 @@ static int mlx5vf_event_notifier(struct notifier_block *nb, unsigned long type,
 		mlx5_nb_cof(nb, struct mlx5_vhca_page_tracker, nb);
 	struct mlx5vf_pci_core_device *mvdev = container_of(
 		tracker, struct mlx5vf_pci_core_device, tracker);
+	struct mlx5_eqe_obj_change *object;
 	struct mlx5_eqe *eqe = data;
 	u8 event_type = (u8)type;
 	u8 queue_type;
+	u32 obj_id;
 	int qp_num;
 
 	switch (event_type) {
@@ -975,6 +1065,12 @@ static int mlx5vf_event_notifier(struct notifier_block *nb, unsigned long type,
 			break;
 		set_tracker_error(mvdev);
 		break;
+	case MLX5_EVENT_TYPE_OBJECT_CHANGE:
+		object = &eqe->data.obj_change;
+		obj_id = be32_to_cpu(object->obj_id);
+		if (obj_id == tracker->id)
+			set_tracker_change_event(mvdev);
+		break;
 	default:
 		break;
 	}
@@ -1634,6 +1730,11 @@ int mlx5vf_tracker_read_and_clear(struct vfio_device *vdev, unsigned long iova,
 		goto end;
 	}
 
+	if (tracker->is_err) {
+		err = -EIO;
+		goto end;
+	}
+
 	mdev = mvdev->mdev;
 	err = mlx5vf_cmd_modify_tracker(mdev, tracker->id, iova, length,
 					MLX5_PAGE_TRACK_STATE_REPORTING);
@@ -1652,6 +1753,12 @@ int mlx5vf_tracker_read_and_clear(struct vfio_device *vdev, unsigned long iova,
 						      dirty, &tracker->status);
 			if (poll_err == CQ_EMPTY) {
 				wait_for_completion(&mvdev->tracker_comp);
+				if (tracker->object_changed) {
+					tracker->object_changed = false;
+					err = mlx5vf_cmd_query_tracker(mdev, tracker);
+					if (err)
+						goto end;
+				}
 				continue;
 			}
 		}
diff --git a/drivers/vfio/pci/mlx5/cmd.h b/drivers/vfio/pci/mlx5/cmd.h
index f2c7227fa683..df421dc6de04 100644
--- a/drivers/vfio/pci/mlx5/cmd.h
+++ b/drivers/vfio/pci/mlx5/cmd.h
@@ -13,9 +13,6 @@
 #include <linux/mlx5/cq.h>
 #include <linux/mlx5/qp.h>
 
-#define MLX5VF_PRE_COPY_SUPP(mvdev) \
-	((mvdev)->core_device.vdev.migration_flags & VFIO_MIGRATION_PRE_COPY)
-
 enum mlx5_vf_migf_state {
 	MLX5_MIGF_STATE_ERROR = 1,
 	MLX5_MIGF_STATE_PRE_COPY_ERROR,
@@ -25,7 +22,6 @@ enum mlx5_vf_migf_state {
 };
 
 enum mlx5_vf_load_state {
-	MLX5_VF_LOAD_STATE_READ_IMAGE_NO_HEADER,
 	MLX5_VF_LOAD_STATE_READ_HEADER,
 	MLX5_VF_LOAD_STATE_PREP_HEADER_DATA,
 	MLX5_VF_LOAD_STATE_READ_HEADER_DATA,
@@ -162,6 +158,7 @@ struct mlx5_vhca_page_tracker {
 	u32 id;
 	u32 pdn;
 	u8 is_err:1;
+	u8 object_changed:1;
 	struct mlx5_uars_page *uar;
 	struct mlx5_vhca_cq cq;
 	struct mlx5_vhca_qp *host_qp;
@@ -196,6 +193,7 @@ struct mlx5vf_pci_core_device {
 enum {
 	MLX5VF_QUERY_INC = (1UL << 0),
 	MLX5VF_QUERY_FINAL = (1UL << 1),
+	MLX5VF_QUERY_CLEANUP = (1UL << 2),
 };
 
 int mlx5vf_cmd_suspend_vhca(struct mlx5vf_pci_core_device *mvdev, u16 op_mod);
@@ -226,12 +224,11 @@ struct mlx5_vhca_data_buffer *
 mlx5vf_get_data_buffer(struct mlx5_vf_migration_file *migf,
 		       size_t length, enum dma_data_direction dma_dir);
 void mlx5vf_put_data_buffer(struct mlx5_vhca_data_buffer *buf);
-int mlx5vf_add_migration_pages(struct mlx5_vhca_data_buffer *buf,
-			       unsigned int npages);
 struct page *mlx5vf_get_migration_page(struct mlx5_vhca_data_buffer *buf,
 				       unsigned long offset);
 void mlx5vf_state_mutex_unlock(struct mlx5vf_pci_core_device *mvdev);
-void mlx5vf_disable_fds(struct mlx5vf_pci_core_device *mvdev);
+void mlx5vf_disable_fds(struct mlx5vf_pci_core_device *mvdev,
+			enum mlx5_vf_migf_state *last_save_state);
 void mlx5vf_mig_file_cleanup_cb(struct work_struct *_work);
 void mlx5vf_mig_file_set_save_work(struct mlx5_vf_migration_file *migf,
 				   u8 chunk_num, size_t next_required_umem_size);
diff --git a/drivers/vfio/pci/mlx5/main.c b/drivers/vfio/pci/mlx5/main.c
index fe09a8c8af95..61d9b0f9146d 100644
--- a/drivers/vfio/pci/mlx5/main.c
+++ b/drivers/vfio/pci/mlx5/main.c
@@ -65,50 +65,6 @@ mlx5vf_get_migration_page(struct mlx5_vhca_data_buffer *buf,
 	return NULL;
 }
 
-int mlx5vf_add_migration_pages(struct mlx5_vhca_data_buffer *buf,
-			       unsigned int npages)
-{
-	unsigned int to_alloc = npages;
-	struct page **page_list;
-	unsigned long filled;
-	unsigned int to_fill;
-	int ret;
-
-	to_fill = min_t(unsigned int, npages, PAGE_SIZE / sizeof(*page_list));
-	page_list = kvzalloc(to_fill * sizeof(*page_list), GFP_KERNEL_ACCOUNT);
-	if (!page_list)
-		return -ENOMEM;
-
-	do {
-		filled = alloc_pages_bulk_array(GFP_KERNEL_ACCOUNT, to_fill,
-						page_list);
-		if (!filled) {
-			ret = -ENOMEM;
-			goto err;
-		}
-		to_alloc -= filled;
-		ret = sg_alloc_append_table_from_pages(
-			&buf->table, page_list, filled, 0,
-			filled << PAGE_SHIFT, UINT_MAX, SG_MAX_SINGLE_ALLOC,
-			GFP_KERNEL_ACCOUNT);
-
-		if (ret)
-			goto err;
-		buf->allocated_length += filled * PAGE_SIZE;
-		/* clean input for another bulk allocation */
-		memset(page_list, 0, filled * sizeof(*page_list));
-		to_fill = min_t(unsigned int, to_alloc,
-				PAGE_SIZE / sizeof(*page_list));
-	} while (to_alloc > 0);
-
-	kvfree(page_list);
-	return 0;
-
-err:
-	kvfree(page_list);
-	return ret;
-}
-
 static void mlx5vf_disable_fd(struct mlx5_vf_migration_file *migf)
 {
 	mutex_lock(&migf->lock);
@@ -777,36 +733,6 @@ mlx5vf_append_page_to_mig_buf(struct mlx5_vhca_data_buffer *vhca_buf,
 	return 0;
 }
 
-static int
-mlx5vf_resume_read_image_no_header(struct mlx5_vhca_data_buffer *vhca_buf,
-				   loff_t requested_length,
-				   const char __user **buf, size_t *len,
-				   loff_t *pos, ssize_t *done)
-{
-	int ret;
-
-	if (requested_length > MAX_LOAD_SIZE)
-		return -ENOMEM;
-
-	if (vhca_buf->allocated_length < requested_length) {
-		ret = mlx5vf_add_migration_pages(
-			vhca_buf,
-			DIV_ROUND_UP(requested_length - vhca_buf->allocated_length,
-				     PAGE_SIZE));
-		if (ret)
-			return ret;
-	}
-
-	while (*len) {
-		ret = mlx5vf_append_page_to_mig_buf(vhca_buf, buf, len, pos,
-						    done);
-		if (ret)
-			return ret;
-	}
-
-	return 0;
-}
-
 static ssize_t
 mlx5vf_resume_read_image(struct mlx5_vf_migration_file *migf,
 			 struct mlx5_vhca_data_buffer *vhca_buf,
@@ -1038,13 +964,6 @@ static ssize_t mlx5vf_resume_write(struct file *filp, const char __user *buf,
 			migf->load_state = MLX5_VF_LOAD_STATE_READ_IMAGE;
 			break;
 		}
-		case MLX5_VF_LOAD_STATE_READ_IMAGE_NO_HEADER:
-			ret = mlx5vf_resume_read_image_no_header(vhca_buf,
-						requested_length,
-						&buf, &len, pos, &done);
-			if (ret)
-				goto out_unlock;
-			break;
 		case MLX5_VF_LOAD_STATE_READ_IMAGE:
 			ret = mlx5vf_resume_read_image(migf, vhca_buf,
 						migf->record_size,
@@ -1114,21 +1033,16 @@ mlx5vf_pci_resume_device_data(struct mlx5vf_pci_core_device *mvdev)
 	}
 
 	migf->buf[0] = buf;
-	if (MLX5VF_PRE_COPY_SUPP(mvdev)) {
-		buf = mlx5vf_alloc_data_buffer(migf,
-			sizeof(struct mlx5_vf_migration_header), DMA_NONE);
-		if (IS_ERR(buf)) {
-			ret = PTR_ERR(buf);
-			goto out_buf;
-		}
-
-		migf->buf_header[0] = buf;
-		migf->load_state = MLX5_VF_LOAD_STATE_READ_HEADER;
-	} else {
-		/* Initial state will be to read the image */
-		migf->load_state = MLX5_VF_LOAD_STATE_READ_IMAGE_NO_HEADER;
+	buf = mlx5vf_alloc_data_buffer(migf,
+		sizeof(struct mlx5_vf_migration_header), DMA_NONE);
+	if (IS_ERR(buf)) {
+		ret = PTR_ERR(buf);
+		goto out_buf;
 	}
 
+	migf->buf_header[0] = buf;
+	migf->load_state = MLX5_VF_LOAD_STATE_READ_HEADER;
+
 	stream_open(migf->filp->f_inode, migf->filp);
 	mutex_init(&migf->lock);
 	INIT_LIST_HEAD(&migf->buf_list);
@@ -1146,7 +1060,8 @@ end:
 	return ERR_PTR(ret);
 }
 
-void mlx5vf_disable_fds(struct mlx5vf_pci_core_device *mvdev)
+void mlx5vf_disable_fds(struct mlx5vf_pci_core_device *mvdev,
+			enum mlx5_vf_migf_state *last_save_state)
 {
 	if (mvdev->resuming_migf) {
 		mlx5vf_disable_fd(mvdev->resuming_migf);
@@ -1157,6 +1072,8 @@ void mlx5vf_disable_fds(struct mlx5vf_pci_core_device *mvdev)
 	if (mvdev->saving_migf) {
 		mlx5_cmd_cleanup_async_ctx(&mvdev->saving_migf->async_ctx);
 		cancel_work_sync(&mvdev->saving_migf->async_data.work);
+		if (last_save_state)
+			*last_save_state = mvdev->saving_migf->state;
 		mlx5vf_disable_fd(mvdev->saving_migf);
 		wake_up_interruptible(&mvdev->saving_migf->poll_wait);
 		mlx5fv_cmd_clean_migf_resources(mvdev->saving_migf);
@@ -1217,12 +1134,34 @@ mlx5vf_pci_step_device_state_locked(struct mlx5vf_pci_core_device *mvdev,
 		return migf->filp;
 	}
 
-	if ((cur == VFIO_DEVICE_STATE_STOP_COPY && new == VFIO_DEVICE_STATE_STOP) ||
-	    (cur == VFIO_DEVICE_STATE_PRE_COPY && new == VFIO_DEVICE_STATE_RUNNING) ||
+	if (cur == VFIO_DEVICE_STATE_STOP_COPY && new == VFIO_DEVICE_STATE_STOP) {
+		mlx5vf_disable_fds(mvdev, NULL);
+		return NULL;
+	}
+
+	if ((cur == VFIO_DEVICE_STATE_PRE_COPY && new == VFIO_DEVICE_STATE_RUNNING) ||
 	    (cur == VFIO_DEVICE_STATE_PRE_COPY_P2P &&
 	     new == VFIO_DEVICE_STATE_RUNNING_P2P)) {
-		mlx5vf_disable_fds(mvdev);
-		return NULL;
+		struct mlx5_vf_migration_file *migf = mvdev->saving_migf;
+		struct mlx5_vhca_data_buffer *buf;
+		enum mlx5_vf_migf_state state;
+		size_t size;
+
+		ret = mlx5vf_cmd_query_vhca_migration_state(mvdev, &size, NULL,
+					MLX5VF_QUERY_INC | MLX5VF_QUERY_CLEANUP);
+		if (ret)
+			return ERR_PTR(ret);
+		buf = mlx5vf_get_data_buffer(migf, size, DMA_FROM_DEVICE);
+		if (IS_ERR(buf))
+			return ERR_CAST(buf);
+		/* pre_copy cleanup */
+		ret = mlx5vf_cmd_save_vhca_state(mvdev, migf, buf, false, false);
+		if (ret) {
+			mlx5vf_put_data_buffer(buf);
+			return ERR_PTR(ret);
+		}
+		mlx5vf_disable_fds(mvdev, &state);
+		return (state != MLX5_MIGF_STATE_ERROR) ? NULL : ERR_PTR(-EIO);
 	}
 
 	if (cur == VFIO_DEVICE_STATE_STOP && new == VFIO_DEVICE_STATE_RESUMING) {
@@ -1237,14 +1176,7 @@ mlx5vf_pci_step_device_state_locked(struct mlx5vf_pci_core_device *mvdev,
 	}
 
 	if (cur == VFIO_DEVICE_STATE_RESUMING && new == VFIO_DEVICE_STATE_STOP) {
-		if (!MLX5VF_PRE_COPY_SUPP(mvdev)) {
-			ret = mlx5vf_cmd_load_vhca_state(mvdev,
-							 mvdev->resuming_migf,
-							 mvdev->resuming_migf->buf[0]);
-			if (ret)
-				return ERR_PTR(ret);
-		}
-		mlx5vf_disable_fds(mvdev);
+		mlx5vf_disable_fds(mvdev, NULL);
 		return NULL;
 	}
 
@@ -1289,7 +1221,7 @@ again:
 		mvdev->deferred_reset = false;
 		spin_unlock(&mvdev->reset_lock);
 		mvdev->mig_state = VFIO_DEVICE_STATE_RUNNING;
-		mlx5vf_disable_fds(mvdev);
+		mlx5vf_disable_fds(mvdev, NULL);
 		goto again;
 	}
 	mutex_unlock(&mvdev->state_mutex);
diff --git a/drivers/vfio/pci/nvgrace-gpu/Kconfig b/drivers/vfio/pci/nvgrace-gpu/Kconfig
new file mode 100644
index 000000000000..a7f624b37e41
--- /dev/null
+++ b/drivers/vfio/pci/nvgrace-gpu/Kconfig
@@ -0,0 +1,10 @@
+# SPDX-License-Identifier: GPL-2.0-only
+config NVGRACE_GPU_VFIO_PCI
+	tristate "VFIO support for the GPU in the NVIDIA Grace Hopper Superchip"
+	depends on ARM64 || (COMPILE_TEST && 64BIT)
+	select VFIO_PCI_CORE
+	help
+	  VFIO support for the GPU in the NVIDIA Grace Hopper Superchip is
+	  required to assign the GPU device to userspace using KVM/qemu/etc.
+
+	  If you don't know what to do here, say N.
diff --git a/drivers/vfio/pci/nvgrace-gpu/Makefile b/drivers/vfio/pci/nvgrace-gpu/Makefile
new file mode 100644
index 000000000000..3ca8c187897a
--- /dev/null
+++ b/drivers/vfio/pci/nvgrace-gpu/Makefile
@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: GPL-2.0-only
+obj-$(CONFIG_NVGRACE_GPU_VFIO_PCI) += nvgrace-gpu-vfio-pci.o
+nvgrace-gpu-vfio-pci-y := main.o
diff --git a/drivers/vfio/pci/nvgrace-gpu/main.c b/drivers/vfio/pci/nvgrace-gpu/main.c
new file mode 100644
index 000000000000..a7fd018aa548
--- /dev/null
+++ b/drivers/vfio/pci/nvgrace-gpu/main.c
@@ -0,0 +1,888 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved
+ */
+
+#include <linux/sizes.h>
+#include <linux/vfio_pci_core.h>
+
+/*
+ * The device memory usable to the workloads running in the VM is cached
+ * and showcased as a 64b device BAR (comprising of BAR4 and BAR5 region)
+ * to the VM and is represented as usemem.
+ * Moreover, the VM GPU device driver needs a non-cacheable region to
+ * support the MIG feature. This region is also exposed as a 64b BAR
+ * (comprising of BAR2 and BAR3 region) and represented as resmem.
+ */
+#define RESMEM_REGION_INDEX VFIO_PCI_BAR2_REGION_INDEX
+#define USEMEM_REGION_INDEX VFIO_PCI_BAR4_REGION_INDEX
+
+/* Memory size expected as non cached and reserved by the VM driver */
+#define RESMEM_SIZE SZ_1G
+
+/* A hardwired and constant ABI value between the GPU FW and VFIO driver. */
+#define MEMBLK_SIZE SZ_512M
+
+/*
+ * The state of the two device memory region - resmem and usemem - is
+ * saved as struct mem_region.
+ */
+struct mem_region {
+	phys_addr_t memphys;    /* Base physical address of the region */
+	size_t memlength;       /* Region size */
+	size_t bar_size;        /* Reported region BAR size */
+	__le64 bar_val;         /* Emulated BAR offset registers */
+	union {
+		void *memaddr;
+		void __iomem *ioaddr;
+	};                      /* Base virtual address of the region */
+};
+
+struct nvgrace_gpu_pci_core_device {
+	struct vfio_pci_core_device core_device;
+	/* Cached and usable memory for the VM. */
+	struct mem_region usemem;
+	/* Non cached memory carved out from the end of device memory */
+	struct mem_region resmem;
+	/* Lock to control device memory kernel mapping */
+	struct mutex remap_lock;
+};
+
+static void nvgrace_gpu_init_fake_bar_emu_regs(struct vfio_device *core_vdev)
+{
+	struct nvgrace_gpu_pci_core_device *nvdev =
+		container_of(core_vdev, struct nvgrace_gpu_pci_core_device,
+			     core_device.vdev);
+
+	nvdev->resmem.bar_val = 0;
+	nvdev->usemem.bar_val = 0;
+}
+
+/* Choose the structure corresponding to the fake BAR with a given index. */
+static struct mem_region *
+nvgrace_gpu_memregion(int index,
+		      struct nvgrace_gpu_pci_core_device *nvdev)
+{
+	if (index == USEMEM_REGION_INDEX)
+		return &nvdev->usemem;
+
+	if (index == RESMEM_REGION_INDEX)
+		return &nvdev->resmem;
+
+	return NULL;
+}
+
+static int nvgrace_gpu_open_device(struct vfio_device *core_vdev)
+{
+	struct vfio_pci_core_device *vdev =
+		container_of(core_vdev, struct vfio_pci_core_device, vdev);
+	struct nvgrace_gpu_pci_core_device *nvdev =
+		container_of(core_vdev, struct nvgrace_gpu_pci_core_device,
+			     core_device.vdev);
+	int ret;
+
+	ret = vfio_pci_core_enable(vdev);
+	if (ret)
+		return ret;
+
+	if (nvdev->usemem.memlength) {
+		nvgrace_gpu_init_fake_bar_emu_regs(core_vdev);
+		mutex_init(&nvdev->remap_lock);
+	}
+
+	vfio_pci_core_finish_enable(vdev);
+
+	return 0;
+}
+
+static void nvgrace_gpu_close_device(struct vfio_device *core_vdev)
+{
+	struct nvgrace_gpu_pci_core_device *nvdev =
+		container_of(core_vdev, struct nvgrace_gpu_pci_core_device,
+			     core_device.vdev);
+
+	/* Unmap the mapping to the device memory cached region */
+	if (nvdev->usemem.memaddr) {
+		memunmap(nvdev->usemem.memaddr);
+		nvdev->usemem.memaddr = NULL;
+	}
+
+	/* Unmap the mapping to the device memory non-cached region */
+	if (nvdev->resmem.ioaddr) {
+		iounmap(nvdev->resmem.ioaddr);
+		nvdev->resmem.ioaddr = NULL;
+	}
+
+	mutex_destroy(&nvdev->remap_lock);
+
+	vfio_pci_core_close_device(core_vdev);
+}
+
+static int nvgrace_gpu_mmap(struct vfio_device *core_vdev,
+			    struct vm_area_struct *vma)
+{
+	struct nvgrace_gpu_pci_core_device *nvdev =
+		container_of(core_vdev, struct nvgrace_gpu_pci_core_device,
+			     core_device.vdev);
+	struct mem_region *memregion;
+	unsigned long start_pfn;
+	u64 req_len, pgoff, end;
+	unsigned int index;
+	int ret = 0;
+
+	index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT);
+
+	memregion = nvgrace_gpu_memregion(index, nvdev);
+	if (!memregion)
+		return vfio_pci_core_mmap(core_vdev, vma);
+
+	/*
+	 * Request to mmap the BAR. Map to the CPU accessible memory on the
+	 * GPU using the memory information gathered from the system ACPI
+	 * tables.
+	 */
+	pgoff = vma->vm_pgoff &
+		((1U << (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT)) - 1);
+
+	if (check_sub_overflow(vma->vm_end, vma->vm_start, &req_len) ||
+	    check_add_overflow(PHYS_PFN(memregion->memphys), pgoff, &start_pfn) ||
+	    check_add_overflow(PFN_PHYS(pgoff), req_len, &end))
+		return -EOVERFLOW;
+
+	/*
+	 * Check that the mapping request does not go beyond available device
+	 * memory size
+	 */
+	if (end > memregion->memlength)
+		return -EINVAL;
+
+	/*
+	 * The carved out region of the device memory needs the NORMAL_NC
+	 * property. Communicate as such to the hypervisor.
+	 */
+	if (index == RESMEM_REGION_INDEX) {
+		/*
+		 * The nvgrace-gpu module has no issues with uncontained
+		 * failures on NORMAL_NC accesses. VM_ALLOW_ANY_UNCACHED is
+		 * set to communicate to the KVM to S2 map as NORMAL_NC.
+		 * This opens up guest usage of NORMAL_NC for this mapping.
+		 */
+		vm_flags_set(vma, VM_ALLOW_ANY_UNCACHED);
+
+		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
+	}
+
+	/*
+	 * Perform a PFN map to the memory and back the device BAR by the
+	 * GPU memory.
+	 *
+	 * The available GPU memory size may not be power-of-2 aligned. The
+	 * remainder is only backed by vfio_device_ops read/write handlers.
+	 *
+	 * During device reset, the GPU is safely disconnected to the CPU
+	 * and access to the BAR will be immediately returned preventing
+	 * machine check.
+	 */
+	ret = remap_pfn_range(vma, vma->vm_start, start_pfn,
+			      req_len, vma->vm_page_prot);
+	if (ret)
+		return ret;
+
+	vma->vm_pgoff = start_pfn;
+
+	return 0;
+}
+
+static long
+nvgrace_gpu_ioctl_get_region_info(struct vfio_device *core_vdev,
+				  unsigned long arg)
+{
+	struct nvgrace_gpu_pci_core_device *nvdev =
+		container_of(core_vdev, struct nvgrace_gpu_pci_core_device,
+			     core_device.vdev);
+	unsigned long minsz = offsetofend(struct vfio_region_info, offset);
+	struct vfio_info_cap caps = { .buf = NULL, .size = 0 };
+	struct vfio_region_info_cap_sparse_mmap *sparse;
+	struct vfio_region_info info;
+	struct mem_region *memregion;
+	u32 size;
+	int ret;
+
+	if (copy_from_user(&info, (void __user *)arg, minsz))
+		return -EFAULT;
+
+	if (info.argsz < minsz)
+		return -EINVAL;
+
+	/*
+	 * Request to determine the BAR region information. Send the
+	 * GPU memory information.
+	 */
+	memregion = nvgrace_gpu_memregion(info.index, nvdev);
+	if (!memregion)
+		return vfio_pci_core_ioctl(core_vdev,
+					   VFIO_DEVICE_GET_REGION_INFO, arg);
+
+	size = struct_size(sparse, areas, 1);
+
+	/*
+	 * Setup for sparse mapping for the device memory. Only the
+	 * available device memory on the hardware is shown as a
+	 * mappable region.
+	 */
+	sparse = kzalloc(size, GFP_KERNEL);
+	if (!sparse)
+		return -ENOMEM;
+
+	sparse->nr_areas = 1;
+	sparse->areas[0].offset = 0;
+	sparse->areas[0].size = memregion->memlength;
+	sparse->header.id = VFIO_REGION_INFO_CAP_SPARSE_MMAP;
+	sparse->header.version = 1;
+
+	ret = vfio_info_add_capability(&caps, &sparse->header, size);
+	kfree(sparse);
+	if (ret)
+		return ret;
+
+	info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
+	/*
+	 * The region memory size may not be power-of-2 aligned.
+	 * Given that the memory  as a BAR and may not be
+	 * aligned, roundup to the next power-of-2.
+	 */
+	info.size = memregion->bar_size;
+	info.flags = VFIO_REGION_INFO_FLAG_READ |
+		     VFIO_REGION_INFO_FLAG_WRITE |
+		     VFIO_REGION_INFO_FLAG_MMAP;
+
+	if (caps.size) {
+		info.flags |= VFIO_REGION_INFO_FLAG_CAPS;
+		if (info.argsz < sizeof(info) + caps.size) {
+			info.argsz = sizeof(info) + caps.size;
+			info.cap_offset = 0;
+		} else {
+			vfio_info_cap_shift(&caps, sizeof(info));
+			if (copy_to_user((void __user *)arg +
+					 sizeof(info), caps.buf,
+					 caps.size)) {
+				kfree(caps.buf);
+				return -EFAULT;
+			}
+			info.cap_offset = sizeof(info);
+		}
+		kfree(caps.buf);
+	}
+	return copy_to_user((void __user *)arg, &info, minsz) ?
+			    -EFAULT : 0;
+}
+
+static long nvgrace_gpu_ioctl(struct vfio_device *core_vdev,
+			      unsigned int cmd, unsigned long arg)
+{
+	switch (cmd) {
+	case VFIO_DEVICE_GET_REGION_INFO:
+		return nvgrace_gpu_ioctl_get_region_info(core_vdev, arg);
+	case VFIO_DEVICE_IOEVENTFD:
+		return -ENOTTY;
+	case VFIO_DEVICE_RESET:
+		nvgrace_gpu_init_fake_bar_emu_regs(core_vdev);
+		fallthrough;
+	default:
+		return vfio_pci_core_ioctl(core_vdev, cmd, arg);
+	}
+}
+
+static __le64
+nvgrace_gpu_get_read_value(size_t bar_size, u64 flags, __le64 val64)
+{
+	u64 tmp_val;
+
+	tmp_val = le64_to_cpu(val64);
+	tmp_val &= ~(bar_size - 1);
+	tmp_val |= flags;
+
+	return cpu_to_le64(tmp_val);
+}
+
+/*
+ * Both the usable (usemem) and the reserved (resmem) device memory region
+ * are exposed as a 64b fake device BARs in the VM. These fake BARs must
+ * respond to the accesses on their respective PCI config space offsets.
+ *
+ * resmem BAR owns PCI_BASE_ADDRESS_2 & PCI_BASE_ADDRESS_3.
+ * usemem BAR owns PCI_BASE_ADDRESS_4 & PCI_BASE_ADDRESS_5.
+ */
+static ssize_t
+nvgrace_gpu_read_config_emu(struct vfio_device *core_vdev,
+			    char __user *buf, size_t count, loff_t *ppos)
+{
+	struct nvgrace_gpu_pci_core_device *nvdev =
+		container_of(core_vdev, struct nvgrace_gpu_pci_core_device,
+			     core_device.vdev);
+	u64 pos = *ppos & VFIO_PCI_OFFSET_MASK;
+	struct mem_region *memregion = NULL;
+	__le64 val64;
+	size_t register_offset;
+	loff_t copy_offset;
+	size_t copy_count;
+	int ret;
+
+	ret = vfio_pci_core_read(core_vdev, buf, count, ppos);
+	if (ret < 0)
+		return ret;
+
+	if (vfio_pci_core_range_intersect_range(pos, count, PCI_BASE_ADDRESS_2,
+						sizeof(val64),
+						&copy_offset, &copy_count,
+						&register_offset))
+		memregion = nvgrace_gpu_memregion(RESMEM_REGION_INDEX, nvdev);
+	else if (vfio_pci_core_range_intersect_range(pos, count,
+						     PCI_BASE_ADDRESS_4,
+						     sizeof(val64),
+						     &copy_offset, &copy_count,
+						     &register_offset))
+		memregion = nvgrace_gpu_memregion(USEMEM_REGION_INDEX, nvdev);
+
+	if (memregion) {
+		val64 = nvgrace_gpu_get_read_value(memregion->bar_size,
+						   PCI_BASE_ADDRESS_MEM_TYPE_64 |
+						   PCI_BASE_ADDRESS_MEM_PREFETCH,
+						   memregion->bar_val);
+		if (copy_to_user(buf + copy_offset,
+				 (void *)&val64 + register_offset, copy_count)) {
+			/*
+			 * The position has been incremented in
+			 * vfio_pci_core_read. Reset the offset back to the
+			 * starting position.
+			 */
+			*ppos -= count;
+			return -EFAULT;
+		}
+	}
+
+	return count;
+}
+
+static ssize_t
+nvgrace_gpu_write_config_emu(struct vfio_device *core_vdev,
+			     const char __user *buf, size_t count, loff_t *ppos)
+{
+	struct nvgrace_gpu_pci_core_device *nvdev =
+		container_of(core_vdev, struct nvgrace_gpu_pci_core_device,
+			     core_device.vdev);
+	u64 pos = *ppos & VFIO_PCI_OFFSET_MASK;
+	struct mem_region *memregion = NULL;
+	size_t register_offset;
+	loff_t copy_offset;
+	size_t copy_count;
+
+	if (vfio_pci_core_range_intersect_range(pos, count, PCI_BASE_ADDRESS_2,
+						sizeof(u64), &copy_offset,
+						&copy_count, &register_offset))
+		memregion = nvgrace_gpu_memregion(RESMEM_REGION_INDEX, nvdev);
+	else if (vfio_pci_core_range_intersect_range(pos, count, PCI_BASE_ADDRESS_4,
+						     sizeof(u64), &copy_offset,
+						     &copy_count, &register_offset))
+		memregion = nvgrace_gpu_memregion(USEMEM_REGION_INDEX, nvdev);
+
+	if (memregion) {
+		if (copy_from_user((void *)&memregion->bar_val + register_offset,
+				   buf + copy_offset, copy_count))
+			return -EFAULT;
+		*ppos += copy_count;
+		return copy_count;
+	}
+
+	return vfio_pci_core_write(core_vdev, buf, count, ppos);
+}
+
+/*
+ * Ad hoc map the device memory in the module kernel VA space. Primarily needed
+ * as vfio does not require the userspace driver to only perform accesses through
+ * mmaps of the vfio-pci BAR regions and such accesses should be supported using
+ * vfio_device_ops read/write implementations.
+ *
+ * The usemem region is cacheable memory and hence is memremaped.
+ * The resmem region is non-cached and is mapped using ioremap_wc (NORMAL_NC).
+ */
+static int
+nvgrace_gpu_map_device_mem(int index,
+			   struct nvgrace_gpu_pci_core_device *nvdev)
+{
+	struct mem_region *memregion;
+	int ret = 0;
+
+	memregion = nvgrace_gpu_memregion(index, nvdev);
+	if (!memregion)
+		return -EINVAL;
+
+	mutex_lock(&nvdev->remap_lock);
+
+	if (memregion->memaddr)
+		goto unlock;
+
+	if (index == USEMEM_REGION_INDEX)
+		memregion->memaddr = memremap(memregion->memphys,
+					      memregion->memlength,
+					      MEMREMAP_WB);
+	else
+		memregion->ioaddr = ioremap_wc(memregion->memphys,
+					       memregion->memlength);
+
+	if (!memregion->memaddr)
+		ret = -ENOMEM;
+
+unlock:
+	mutex_unlock(&nvdev->remap_lock);
+
+	return ret;
+}
+
+/*
+ * Read the data from the device memory (mapped either through ioremap
+ * or memremap) into the user buffer.
+ */
+static int
+nvgrace_gpu_map_and_read(struct nvgrace_gpu_pci_core_device *nvdev,
+			 char __user *buf, size_t mem_count, loff_t *ppos)
+{
+	unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
+	u64 offset = *ppos & VFIO_PCI_OFFSET_MASK;
+	int ret;
+
+	if (!mem_count)
+		return 0;
+
+	/*
+	 * Handle read on the BAR regions. Map to the target device memory
+	 * physical address and copy to the request read buffer.
+	 */
+	ret = nvgrace_gpu_map_device_mem(index, nvdev);
+	if (ret)
+		return ret;
+
+	if (index == USEMEM_REGION_INDEX) {
+		if (copy_to_user(buf,
+				 (u8 *)nvdev->usemem.memaddr + offset,
+				 mem_count))
+			ret = -EFAULT;
+	} else {
+		/*
+		 * The hardware ensures that the system does not crash when
+		 * the device memory is accessed with the memory enable
+		 * turned off. It synthesizes ~0 on such read. So there is
+		 * no need to check or support the disablement/enablement of
+		 * BAR through PCI_COMMAND config space register. Pass
+		 * test_mem flag as false.
+		 */
+		ret = vfio_pci_core_do_io_rw(&nvdev->core_device, false,
+					     nvdev->resmem.ioaddr,
+					     buf, offset, mem_count,
+					     0, 0, false);
+	}
+
+	return ret;
+}
+
+/*
+ * Read count bytes from the device memory at an offset. The actual device
+ * memory size (available) may not be a power-of-2. So the driver fakes
+ * the size to a power-of-2 (reported) when exposing to a user space driver.
+ *
+ * Reads starting beyond the reported size generate -EINVAL; reads extending
+ * beyond the actual device size is filled with ~0; reads extending beyond
+ * the reported size are truncated.
+ */
+static ssize_t
+nvgrace_gpu_read_mem(struct nvgrace_gpu_pci_core_device *nvdev,
+		     char __user *buf, size_t count, loff_t *ppos)
+{
+	u64 offset = *ppos & VFIO_PCI_OFFSET_MASK;
+	unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
+	struct mem_region *memregion;
+	size_t mem_count, i;
+	u8 val = 0xFF;
+	int ret;
+
+	/* No need to do NULL check as caller does. */
+	memregion = nvgrace_gpu_memregion(index, nvdev);
+
+	if (offset >= memregion->bar_size)
+		return -EINVAL;
+
+	/* Clip short the read request beyond reported BAR size */
+	count = min(count, memregion->bar_size - (size_t)offset);
+
+	/*
+	 * Determine how many bytes to be actually read from the device memory.
+	 * Read request beyond the actual device memory size is filled with ~0,
+	 * while those beyond the actual reported size is skipped.
+	 */
+	if (offset >= memregion->memlength)
+		mem_count = 0;
+	else
+		mem_count = min(count, memregion->memlength - (size_t)offset);
+
+	ret = nvgrace_gpu_map_and_read(nvdev, buf, mem_count, ppos);
+	if (ret)
+		return ret;
+
+	/*
+	 * Only the device memory present on the hardware is mapped, which may
+	 * not be power-of-2 aligned. A read to an offset beyond the device memory
+	 * size is filled with ~0.
+	 */
+	for (i = mem_count; i < count; i++) {
+		ret = put_user(val, (unsigned char __user *)(buf + i));
+		if (ret)
+			return ret;
+	}
+
+	*ppos += count;
+	return count;
+}
+
+static ssize_t
+nvgrace_gpu_read(struct vfio_device *core_vdev,
+		 char __user *buf, size_t count, loff_t *ppos)
+{
+	unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
+	struct nvgrace_gpu_pci_core_device *nvdev =
+		container_of(core_vdev, struct nvgrace_gpu_pci_core_device,
+			     core_device.vdev);
+
+	if (nvgrace_gpu_memregion(index, nvdev))
+		return nvgrace_gpu_read_mem(nvdev, buf, count, ppos);
+
+	if (index == VFIO_PCI_CONFIG_REGION_INDEX)
+		return nvgrace_gpu_read_config_emu(core_vdev, buf, count, ppos);
+
+	return vfio_pci_core_read(core_vdev, buf, count, ppos);
+}
+
+/*
+ * Write the data to the device memory (mapped either through ioremap
+ * or memremap) from the user buffer.
+ */
+static int
+nvgrace_gpu_map_and_write(struct nvgrace_gpu_pci_core_device *nvdev,
+			  const char __user *buf, size_t mem_count,
+			  loff_t *ppos)
+{
+	unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
+	loff_t pos = *ppos & VFIO_PCI_OFFSET_MASK;
+	int ret;
+
+	if (!mem_count)
+		return 0;
+
+	ret = nvgrace_gpu_map_device_mem(index, nvdev);
+	if (ret)
+		return ret;
+
+	if (index == USEMEM_REGION_INDEX) {
+		if (copy_from_user((u8 *)nvdev->usemem.memaddr + pos,
+				   buf, mem_count))
+			return -EFAULT;
+	} else {
+		/*
+		 * The hardware ensures that the system does not crash when
+		 * the device memory is accessed with the memory enable
+		 * turned off. It drops such writes. So there is no need to
+		 * check or support the disablement/enablement of BAR
+		 * through PCI_COMMAND config space register. Pass test_mem
+		 * flag as false.
+		 */
+		ret = vfio_pci_core_do_io_rw(&nvdev->core_device, false,
+					     nvdev->resmem.ioaddr,
+					     (char __user *)buf, pos, mem_count,
+					     0, 0, true);
+	}
+
+	return ret;
+}
+
+/*
+ * Write count bytes to the device memory at a given offset. The actual device
+ * memory size (available) may not be a power-of-2. So the driver fakes the
+ * size to a power-of-2 (reported) when exposing to a user space driver.
+ *
+ * Writes extending beyond the reported size are truncated; writes starting
+ * beyond the reported size generate -EINVAL.
+ */
+static ssize_t
+nvgrace_gpu_write_mem(struct nvgrace_gpu_pci_core_device *nvdev,
+		      size_t count, loff_t *ppos, const char __user *buf)
+{
+	unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
+	u64 offset = *ppos & VFIO_PCI_OFFSET_MASK;
+	struct mem_region *memregion;
+	size_t mem_count;
+	int ret = 0;
+
+	/* No need to do NULL check as caller does. */
+	memregion = nvgrace_gpu_memregion(index, nvdev);
+
+	if (offset >= memregion->bar_size)
+		return -EINVAL;
+
+	/* Clip short the write request beyond reported BAR size */
+	count = min(count, memregion->bar_size - (size_t)offset);
+
+	/*
+	 * Determine how many bytes to be actually written to the device memory.
+	 * Do not write to the offset beyond available size.
+	 */
+	if (offset >= memregion->memlength)
+		goto exitfn;
+
+	/*
+	 * Only the device memory present on the hardware is mapped, which may
+	 * not be power-of-2 aligned. Drop access outside the available device
+	 * memory on the hardware.
+	 */
+	mem_count = min(count, memregion->memlength - (size_t)offset);
+
+	ret = nvgrace_gpu_map_and_write(nvdev, buf, mem_count, ppos);
+	if (ret)
+		return ret;
+
+exitfn:
+	*ppos += count;
+	return count;
+}
+
+static ssize_t
+nvgrace_gpu_write(struct vfio_device *core_vdev,
+		  const char __user *buf, size_t count, loff_t *ppos)
+{
+	struct nvgrace_gpu_pci_core_device *nvdev =
+		container_of(core_vdev, struct nvgrace_gpu_pci_core_device,
+			     core_device.vdev);
+	unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
+
+	if (nvgrace_gpu_memregion(index, nvdev))
+		return nvgrace_gpu_write_mem(nvdev, count, ppos, buf);
+
+	if (index == VFIO_PCI_CONFIG_REGION_INDEX)
+		return nvgrace_gpu_write_config_emu(core_vdev, buf, count, ppos);
+
+	return vfio_pci_core_write(core_vdev, buf, count, ppos);
+}
+
+static const struct vfio_device_ops nvgrace_gpu_pci_ops = {
+	.name		= "nvgrace-gpu-vfio-pci",
+	.init		= vfio_pci_core_init_dev,
+	.release	= vfio_pci_core_release_dev,
+	.open_device	= nvgrace_gpu_open_device,
+	.close_device	= nvgrace_gpu_close_device,
+	.ioctl		= nvgrace_gpu_ioctl,
+	.device_feature	= vfio_pci_core_ioctl_feature,
+	.read		= nvgrace_gpu_read,
+	.write		= nvgrace_gpu_write,
+	.mmap		= nvgrace_gpu_mmap,
+	.request	= vfio_pci_core_request,
+	.match		= vfio_pci_core_match,
+	.bind_iommufd	= vfio_iommufd_physical_bind,
+	.unbind_iommufd	= vfio_iommufd_physical_unbind,
+	.attach_ioas	= vfio_iommufd_physical_attach_ioas,
+	.detach_ioas	= vfio_iommufd_physical_detach_ioas,
+};
+
+static const struct vfio_device_ops nvgrace_gpu_pci_core_ops = {
+	.name		= "nvgrace-gpu-vfio-pci-core",
+	.init		= vfio_pci_core_init_dev,
+	.release	= vfio_pci_core_release_dev,
+	.open_device	= nvgrace_gpu_open_device,
+	.close_device	= vfio_pci_core_close_device,
+	.ioctl		= vfio_pci_core_ioctl,
+	.device_feature	= vfio_pci_core_ioctl_feature,
+	.read		= vfio_pci_core_read,
+	.write		= vfio_pci_core_write,
+	.mmap		= vfio_pci_core_mmap,
+	.request	= vfio_pci_core_request,
+	.match		= vfio_pci_core_match,
+	.bind_iommufd	= vfio_iommufd_physical_bind,
+	.unbind_iommufd	= vfio_iommufd_physical_unbind,
+	.attach_ioas	= vfio_iommufd_physical_attach_ioas,
+	.detach_ioas	= vfio_iommufd_physical_detach_ioas,
+};
+
+static int
+nvgrace_gpu_fetch_memory_property(struct pci_dev *pdev,
+				  u64 *pmemphys, u64 *pmemlength)
+{
+	int ret;
+
+	/*
+	 * The memory information is present in the system ACPI tables as DSD
+	 * properties nvidia,gpu-mem-base-pa and nvidia,gpu-mem-size.
+	 */
+	ret = device_property_read_u64(&pdev->dev, "nvidia,gpu-mem-base-pa",
+				       pmemphys);
+	if (ret)
+		return ret;
+
+	if (*pmemphys > type_max(phys_addr_t))
+		return -EOVERFLOW;
+
+	ret = device_property_read_u64(&pdev->dev, "nvidia,gpu-mem-size",
+				       pmemlength);
+	if (ret)
+		return ret;
+
+	if (*pmemlength > type_max(size_t))
+		return -EOVERFLOW;
+
+	/*
+	 * If the C2C link is not up due to an error, the coherent device
+	 * memory size is returned as 0. Fail in such case.
+	 */
+	if (*pmemlength == 0)
+		return -ENOMEM;
+
+	return ret;
+}
+
+static int
+nvgrace_gpu_init_nvdev_struct(struct pci_dev *pdev,
+			      struct nvgrace_gpu_pci_core_device *nvdev,
+			      u64 memphys, u64 memlength)
+{
+	int ret = 0;
+
+	/*
+	 * The VM GPU device driver needs a non-cacheable region to support
+	 * the MIG feature. Since the device memory is mapped as NORMAL cached,
+	 * carve out a region from the end with a different NORMAL_NC
+	 * property (called as reserved memory and represented as resmem). This
+	 * region then is exposed as a 64b BAR (region 2 and 3) to the VM, while
+	 * exposing the rest (termed as usable memory and represented using usemem)
+	 * as cacheable 64b BAR (region 4 and 5).
+	 *
+	 *               devmem (memlength)
+	 * |-------------------------------------------------|
+	 * |                                           |
+	 * usemem.memphys                              resmem.memphys
+	 */
+	nvdev->usemem.memphys = memphys;
+
+	/*
+	 * The device memory exposed to the VM is added to the kernel by the
+	 * VM driver module in chunks of memory block size. Only the usable
+	 * memory (usemem) is added to the kernel for usage by the VM
+	 * workloads. Make the usable memory size memblock aligned.
+	 */
+	if (check_sub_overflow(memlength, RESMEM_SIZE,
+			       &nvdev->usemem.memlength)) {
+		ret = -EOVERFLOW;
+		goto done;
+	}
+
+	/*
+	 * The USEMEM part of the device memory has to be MEMBLK_SIZE
+	 * aligned. This is a hardwired ABI value between the GPU FW and
+	 * VFIO driver. The VM device driver is also aware of it and make
+	 * use of the value for its calculation to determine USEMEM size.
+	 */
+	nvdev->usemem.memlength = round_down(nvdev->usemem.memlength,
+					     MEMBLK_SIZE);
+	if (nvdev->usemem.memlength == 0) {
+		ret = -EINVAL;
+		goto done;
+	}
+
+	if ((check_add_overflow(nvdev->usemem.memphys,
+				nvdev->usemem.memlength,
+				&nvdev->resmem.memphys)) ||
+	    (check_sub_overflow(memlength, nvdev->usemem.memlength,
+				&nvdev->resmem.memlength))) {
+		ret = -EOVERFLOW;
+		goto done;
+	}
+
+	/*
+	 * The memory regions are exposed as BARs. Calculate and save
+	 * the BAR size for them.
+	 */
+	nvdev->usemem.bar_size = roundup_pow_of_two(nvdev->usemem.memlength);
+	nvdev->resmem.bar_size = roundup_pow_of_two(nvdev->resmem.memlength);
+done:
+	return ret;
+}
+
+static int nvgrace_gpu_probe(struct pci_dev *pdev,
+			     const struct pci_device_id *id)
+{
+	const struct vfio_device_ops *ops = &nvgrace_gpu_pci_core_ops;
+	struct nvgrace_gpu_pci_core_device *nvdev;
+	u64 memphys, memlength;
+	int ret;
+
+	ret = nvgrace_gpu_fetch_memory_property(pdev, &memphys, &memlength);
+	if (!ret)
+		ops = &nvgrace_gpu_pci_ops;
+
+	nvdev = vfio_alloc_device(nvgrace_gpu_pci_core_device, core_device.vdev,
+				  &pdev->dev, ops);
+	if (IS_ERR(nvdev))
+		return PTR_ERR(nvdev);
+
+	dev_set_drvdata(&pdev->dev, &nvdev->core_device);
+
+	if (ops == &nvgrace_gpu_pci_ops) {
+		/*
+		 * Device memory properties are identified in the host ACPI
+		 * table. Set the nvgrace_gpu_pci_core_device structure.
+		 */
+		ret = nvgrace_gpu_init_nvdev_struct(pdev, nvdev,
+						    memphys, memlength);
+		if (ret)
+			goto out_put_vdev;
+	}
+
+	ret = vfio_pci_core_register_device(&nvdev->core_device);
+	if (ret)
+		goto out_put_vdev;
+
+	return ret;
+
+out_put_vdev:
+	vfio_put_device(&nvdev->core_device.vdev);
+	return ret;
+}
+
+static void nvgrace_gpu_remove(struct pci_dev *pdev)
+{
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(&pdev->dev);
+
+	vfio_pci_core_unregister_device(core_device);
+	vfio_put_device(&core_device->vdev);
+}
+
+static const struct pci_device_id nvgrace_gpu_vfio_pci_table[] = {
+	/* GH200 120GB */
+	{ PCI_DRIVER_OVERRIDE_DEVICE_VFIO(PCI_VENDOR_ID_NVIDIA, 0x2342) },
+	/* GH200 480GB */
+	{ PCI_DRIVER_OVERRIDE_DEVICE_VFIO(PCI_VENDOR_ID_NVIDIA, 0x2345) },
+	{}
+};
+
+MODULE_DEVICE_TABLE(pci, nvgrace_gpu_vfio_pci_table);
+
+static struct pci_driver nvgrace_gpu_vfio_pci_driver = {
+	.name = KBUILD_MODNAME,
+	.id_table = nvgrace_gpu_vfio_pci_table,
+	.probe = nvgrace_gpu_probe,
+	.remove = nvgrace_gpu_remove,
+	.err_handler = &vfio_pci_core_err_handlers,
+	.driver_managed_dma = true,
+};
+
+module_pci_driver(nvgrace_gpu_vfio_pci_driver);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Ankit Agrawal <ankita@nvidia.com>");
+MODULE_AUTHOR("Aniket Agashe <aniketa@nvidia.com>");
+MODULE_DESCRIPTION("VFIO NVGRACE GPU PF - User Level driver for NVIDIA devices with CPU coherently accessible device memory");
diff --git a/drivers/vfio/pci/pds/dirty.c b/drivers/vfio/pci/pds/dirty.c
index 8ddf4346fcd5..68e8f006dfdb 100644
--- a/drivers/vfio/pci/pds/dirty.c
+++ b/drivers/vfio/pci/pds/dirty.c
@@ -607,7 +607,7 @@ int pds_vfio_dma_logging_report(struct vfio_device *vdev, unsigned long iova,
 
 	mutex_lock(&pds_vfio->state_mutex);
 	err = pds_vfio_dirty_sync(pds_vfio, dirty, iova, length);
-	pds_vfio_state_mutex_unlock(pds_vfio);
+	mutex_unlock(&pds_vfio->state_mutex);
 
 	return err;
 }
@@ -624,7 +624,7 @@ int pds_vfio_dma_logging_start(struct vfio_device *vdev,
 	mutex_lock(&pds_vfio->state_mutex);
 	pds_vfio_send_host_vf_lm_status_cmd(pds_vfio, PDS_LM_STA_IN_PROGRESS);
 	err = pds_vfio_dirty_enable(pds_vfio, ranges, nnodes, page_size);
-	pds_vfio_state_mutex_unlock(pds_vfio);
+	mutex_unlock(&pds_vfio->state_mutex);
 
 	return err;
 }
@@ -637,7 +637,7 @@ int pds_vfio_dma_logging_stop(struct vfio_device *vdev)
 
 	mutex_lock(&pds_vfio->state_mutex);
 	pds_vfio_dirty_disable(pds_vfio, true);
-	pds_vfio_state_mutex_unlock(pds_vfio);
+	mutex_unlock(&pds_vfio->state_mutex);
 
 	return 0;
 }
diff --git a/drivers/vfio/pci/pds/lm.c b/drivers/vfio/pci/pds/lm.c
index 79fe2e66bb49..6b94cc0bf45b 100644
--- a/drivers/vfio/pci/pds/lm.c
+++ b/drivers/vfio/pci/pds/lm.c
@@ -92,8 +92,10 @@ static void pds_vfio_put_lm_file(struct pds_vfio_lm_file *lm_file)
 {
 	mutex_lock(&lm_file->lock);
 
+	lm_file->disabled = true;
 	lm_file->size = 0;
 	lm_file->alloc_size = 0;
+	lm_file->filep->f_pos = 0;
 
 	/* Free scatter list of file pages */
 	sg_free_table(&lm_file->sg_table);
@@ -183,6 +185,12 @@ static ssize_t pds_vfio_save_read(struct file *filp, char __user *buf,
 	pos = &filp->f_pos;
 
 	mutex_lock(&lm_file->lock);
+
+	if (lm_file->disabled) {
+		done = -ENODEV;
+		goto out_unlock;
+	}
+
 	if (*pos > lm_file->size) {
 		done = -EINVAL;
 		goto out_unlock;
@@ -283,6 +291,11 @@ static ssize_t pds_vfio_restore_write(struct file *filp, const char __user *buf,
 
 	mutex_lock(&lm_file->lock);
 
+	if (lm_file->disabled) {
+		done = -ENODEV;
+		goto out_unlock;
+	}
+
 	while (len) {
 		size_t page_offset;
 		struct page *page;
diff --git a/drivers/vfio/pci/pds/lm.h b/drivers/vfio/pci/pds/lm.h
index 13be893198b7..9511b1afc6a1 100644
--- a/drivers/vfio/pci/pds/lm.h
+++ b/drivers/vfio/pci/pds/lm.h
@@ -27,6 +27,7 @@ struct pds_vfio_lm_file {
 	struct scatterlist *last_offset_sg;	/* Iterator */
 	unsigned int sg_last_entry;
 	unsigned long last_offset;
+	bool disabled;
 };
 
 struct pds_vfio_pci_device;
diff --git a/drivers/vfio/pci/pds/pci_drv.c b/drivers/vfio/pci/pds/pci_drv.c
index a34dda516629..16e93b11ab1b 100644
--- a/drivers/vfio/pci/pds/pci_drv.c
+++ b/drivers/vfio/pci/pds/pci_drv.c
@@ -21,16 +21,13 @@
 
 static void pds_vfio_recovery(struct pds_vfio_pci_device *pds_vfio)
 {
-	bool deferred_reset_needed = false;
-
 	/*
 	 * Documentation states that the kernel migration driver must not
 	 * generate asynchronous device state transitions outside of
 	 * manipulation by the user or the VFIO_DEVICE_RESET ioctl.
 	 *
 	 * Since recovery is an asynchronous event received from the device,
-	 * initiate a deferred reset. Issue a deferred reset in the following
-	 * situations:
+	 * initiate a reset in the following situations:
 	 *   1. Migration is in progress, which will cause the next step of
 	 *	the migration to fail.
 	 *   2. If the device is in a state that will be set to
@@ -42,24 +39,8 @@ static void pds_vfio_recovery(struct pds_vfio_pci_device *pds_vfio)
 	     pds_vfio->state != VFIO_DEVICE_STATE_ERROR) ||
 	    (pds_vfio->state == VFIO_DEVICE_STATE_RUNNING &&
 	     pds_vfio_dirty_is_enabled(pds_vfio)))
-		deferred_reset_needed = true;
+		pds_vfio_reset(pds_vfio, VFIO_DEVICE_STATE_ERROR);
 	mutex_unlock(&pds_vfio->state_mutex);
-
-	/*
-	 * On the next user initiated state transition, the device will
-	 * transition to the VFIO_DEVICE_STATE_ERROR. At this point it's the user's
-	 * responsibility to reset the device.
-	 *
-	 * If a VFIO_DEVICE_RESET is requested post recovery and before the next
-	 * state transition, then the deferred reset state will be set to
-	 * VFIO_DEVICE_STATE_RUNNING.
-	 */
-	if (deferred_reset_needed) {
-		mutex_lock(&pds_vfio->reset_mutex);
-		pds_vfio->deferred_reset = true;
-		pds_vfio->deferred_reset_state = VFIO_DEVICE_STATE_ERROR;
-		mutex_unlock(&pds_vfio->reset_mutex);
-	}
 }
 
 static int pds_vfio_pci_notify_handler(struct notifier_block *nb,
@@ -185,7 +166,9 @@ static void pds_vfio_pci_aer_reset_done(struct pci_dev *pdev)
 {
 	struct pds_vfio_pci_device *pds_vfio = pds_vfio_pci_drvdata(pdev);
 
-	pds_vfio_reset(pds_vfio);
+	mutex_lock(&pds_vfio->state_mutex);
+	pds_vfio_reset(pds_vfio, VFIO_DEVICE_STATE_RUNNING);
+	mutex_unlock(&pds_vfio->state_mutex);
 }
 
 static const struct pci_error_handlers pds_vfio_pci_err_handlers = {
diff --git a/drivers/vfio/pci/pds/vfio_dev.c b/drivers/vfio/pci/pds/vfio_dev.c
index 4c351c59d05a..76a80ae7087b 100644
--- a/drivers/vfio/pci/pds/vfio_dev.c
+++ b/drivers/vfio/pci/pds/vfio_dev.c
@@ -26,37 +26,14 @@ struct pds_vfio_pci_device *pds_vfio_pci_drvdata(struct pci_dev *pdev)
 			    vfio_coredev);
 }
 
-void pds_vfio_state_mutex_unlock(struct pds_vfio_pci_device *pds_vfio)
+void pds_vfio_reset(struct pds_vfio_pci_device *pds_vfio,
+		    enum vfio_device_mig_state state)
 {
-again:
-	mutex_lock(&pds_vfio->reset_mutex);
-	if (pds_vfio->deferred_reset) {
-		pds_vfio->deferred_reset = false;
-		if (pds_vfio->state == VFIO_DEVICE_STATE_ERROR) {
-			pds_vfio_put_restore_file(pds_vfio);
-			pds_vfio_put_save_file(pds_vfio);
-			pds_vfio_dirty_disable(pds_vfio, false);
-		}
-		pds_vfio->state = pds_vfio->deferred_reset_state;
-		pds_vfio->deferred_reset_state = VFIO_DEVICE_STATE_RUNNING;
-		mutex_unlock(&pds_vfio->reset_mutex);
-		goto again;
-	}
-	mutex_unlock(&pds_vfio->state_mutex);
-	mutex_unlock(&pds_vfio->reset_mutex);
-}
-
-void pds_vfio_reset(struct pds_vfio_pci_device *pds_vfio)
-{
-	mutex_lock(&pds_vfio->reset_mutex);
-	pds_vfio->deferred_reset = true;
-	pds_vfio->deferred_reset_state = VFIO_DEVICE_STATE_RUNNING;
-	if (!mutex_trylock(&pds_vfio->state_mutex)) {
-		mutex_unlock(&pds_vfio->reset_mutex);
-		return;
-	}
-	mutex_unlock(&pds_vfio->reset_mutex);
-	pds_vfio_state_mutex_unlock(pds_vfio);
+	pds_vfio_put_restore_file(pds_vfio);
+	pds_vfio_put_save_file(pds_vfio);
+	if (state == VFIO_DEVICE_STATE_ERROR)
+		pds_vfio_dirty_disable(pds_vfio, false);
+	pds_vfio->state = state;
 }
 
 static struct file *
@@ -97,8 +74,7 @@ pds_vfio_set_device_state(struct vfio_device *vdev,
 			break;
 		}
 	}
-	pds_vfio_state_mutex_unlock(pds_vfio);
-	/* still waiting on a deferred_reset */
+	mutex_unlock(&pds_vfio->state_mutex);
 	if (pds_vfio->state == VFIO_DEVICE_STATE_ERROR)
 		res = ERR_PTR(-EIO);
 
@@ -114,7 +90,7 @@ static int pds_vfio_get_device_state(struct vfio_device *vdev,
 
 	mutex_lock(&pds_vfio->state_mutex);
 	*current_state = pds_vfio->state;
-	pds_vfio_state_mutex_unlock(pds_vfio);
+	mutex_unlock(&pds_vfio->state_mutex);
 	return 0;
 }
 
@@ -156,7 +132,6 @@ static int pds_vfio_init_device(struct vfio_device *vdev)
 	pds_vfio->vf_id = vf_id;
 
 	mutex_init(&pds_vfio->state_mutex);
-	mutex_init(&pds_vfio->reset_mutex);
 
 	vdev->migration_flags = VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P;
 	vdev->mig_ops = &pds_vfio_lm_ops;
@@ -178,7 +153,6 @@ static void pds_vfio_release_device(struct vfio_device *vdev)
 			     vfio_coredev.vdev);
 
 	mutex_destroy(&pds_vfio->state_mutex);
-	mutex_destroy(&pds_vfio->reset_mutex);
 	vfio_pci_core_release_dev(vdev);
 }
 
@@ -194,7 +168,6 @@ static int pds_vfio_open_device(struct vfio_device *vdev)
 		return err;
 
 	pds_vfio->state = VFIO_DEVICE_STATE_RUNNING;
-	pds_vfio->deferred_reset_state = VFIO_DEVICE_STATE_RUNNING;
 
 	vfio_pci_core_finish_enable(&pds_vfio->vfio_coredev);
 
diff --git a/drivers/vfio/pci/pds/vfio_dev.h b/drivers/vfio/pci/pds/vfio_dev.h
index e7b01080a1ec..803d99d69c73 100644
--- a/drivers/vfio/pci/pds/vfio_dev.h
+++ b/drivers/vfio/pci/pds/vfio_dev.h
@@ -18,20 +18,16 @@ struct pds_vfio_pci_device {
 	struct pds_vfio_dirty dirty;
 	struct mutex state_mutex; /* protect migration state */
 	enum vfio_device_mig_state state;
-	struct mutex reset_mutex; /* protect reset_done flow */
-	u8 deferred_reset;
-	enum vfio_device_mig_state deferred_reset_state;
 	struct notifier_block nb;
 
 	int vf_id;
 	u16 client_id;
 };
 
-void pds_vfio_state_mutex_unlock(struct pds_vfio_pci_device *pds_vfio);
-
 const struct vfio_device_ops *pds_vfio_ops_info(void);
 struct pds_vfio_pci_device *pds_vfio_pci_drvdata(struct pci_dev *pdev);
-void pds_vfio_reset(struct pds_vfio_pci_device *pds_vfio);
+void pds_vfio_reset(struct pds_vfio_pci_device *pds_vfio,
+		    enum vfio_device_mig_state state);
 
 struct pci_dev *pds_vfio_to_pci_dev(struct pds_vfio_pci_device *pds_vfio);
 struct device *pds_vfio_to_dev(struct pds_vfio_pci_device *pds_vfio);
diff --git a/drivers/vfio/pci/vfio_pci_config.c b/drivers/vfio/pci/vfio_pci_config.c
index 7e2e62ab0869..97422aafaa7b 100644
--- a/drivers/vfio/pci/vfio_pci_config.c
+++ b/drivers/vfio/pci/vfio_pci_config.c
@@ -1966,3 +1966,45 @@ ssize_t vfio_pci_config_rw(struct vfio_pci_core_device *vdev, char __user *buf,
 
 	return done;
 }
+
+/**
+ * vfio_pci_core_range_intersect_range() - Determine overlap between a buffer
+ *					   and register offset ranges.
+ * @buf_start:		start offset of the buffer
+ * @buf_cnt:		number of buffer bytes
+ * @reg_start:		start register offset
+ * @reg_cnt:		number of register bytes
+ * @buf_offset:	start offset of overlap in the buffer
+ * @intersect_count:	number of overlapping bytes
+ * @register_offset:	start offset of overlap in register
+ *
+ * Returns: true if there is overlap, false if not.
+ * The overlap start and size is returned through function args.
+ */
+bool vfio_pci_core_range_intersect_range(loff_t buf_start, size_t buf_cnt,
+					 loff_t reg_start, size_t reg_cnt,
+					 loff_t *buf_offset,
+					 size_t *intersect_count,
+					 size_t *register_offset)
+{
+	if (buf_start <= reg_start &&
+	    buf_start + buf_cnt > reg_start) {
+		*buf_offset = reg_start - buf_start;
+		*intersect_count = min_t(size_t, reg_cnt,
+					 buf_start + buf_cnt - reg_start);
+		*register_offset = 0;
+		return true;
+	}
+
+	if (buf_start > reg_start &&
+	    buf_start < reg_start + reg_cnt) {
+		*buf_offset = 0;
+		*intersect_count = min_t(size_t, buf_cnt,
+					 reg_start + reg_cnt - buf_start);
+		*register_offset = buf_start - reg_start;
+		return true;
+	}
+
+	return false;
+}
+EXPORT_SYMBOL_GPL(vfio_pci_core_range_intersect_range);
diff --git a/drivers/vfio/pci/vfio_pci_core.c b/drivers/vfio/pci/vfio_pci_core.c
index df6f99bdf70d..d94d61b92c1a 100644
--- a/drivers/vfio/pci/vfio_pci_core.c
+++ b/drivers/vfio/pci/vfio_pci_core.c
@@ -2064,6 +2064,7 @@ static int vfio_pci_bus_notifier(struct notifier_block *nb,
 			 pci_name(pdev));
 		pdev->driver_override = kasprintf(GFP_KERNEL, "%s",
 						  vdev->vdev.ops->name);
+		WARN_ON(!pdev->driver_override);
 	} else if (action == BUS_NOTIFY_BOUND_DRIVER &&
 		   pdev->is_virtfn && physfn == vdev->pdev) {
 		struct pci_driver *drv = pci_dev_driver(pdev);
diff --git a/drivers/vfio/pci/vfio_pci_intrs.c b/drivers/vfio/pci/vfio_pci_intrs.c
index 237beac83809..fb5392b749ff 100644
--- a/drivers/vfio/pci/vfio_pci_intrs.c
+++ b/drivers/vfio/pci/vfio_pci_intrs.c
@@ -90,22 +90,28 @@ static void vfio_send_intx_eventfd(void *opaque, void *unused)
 
 	if (likely(is_intx(vdev) && !vdev->virq_disabled)) {
 		struct vfio_pci_irq_ctx *ctx;
+		struct eventfd_ctx *trigger;
 
 		ctx = vfio_irq_ctx_get(vdev, 0);
 		if (WARN_ON_ONCE(!ctx))
 			return;
-		eventfd_signal(ctx->trigger);
+
+		trigger = READ_ONCE(ctx->trigger);
+		if (likely(trigger))
+			eventfd_signal(trigger);
 	}
 }
 
 /* Returns true if the INTx vfio_pci_irq_ctx.masked value is changed. */
-bool vfio_pci_intx_mask(struct vfio_pci_core_device *vdev)
+static bool __vfio_pci_intx_mask(struct vfio_pci_core_device *vdev)
 {
 	struct pci_dev *pdev = vdev->pdev;
 	struct vfio_pci_irq_ctx *ctx;
 	unsigned long flags;
 	bool masked_changed = false;
 
+	lockdep_assert_held(&vdev->igate);
+
 	spin_lock_irqsave(&vdev->irqlock, flags);
 
 	/*
@@ -143,6 +149,17 @@ out_unlock:
 	return masked_changed;
 }
 
+bool vfio_pci_intx_mask(struct vfio_pci_core_device *vdev)
+{
+	bool mask_changed;
+
+	mutex_lock(&vdev->igate);
+	mask_changed = __vfio_pci_intx_mask(vdev);
+	mutex_unlock(&vdev->igate);
+
+	return mask_changed;
+}
+
 /*
  * If this is triggered by an eventfd, we can't call eventfd_signal
  * or else we'll deadlock on the eventfd wait queue.  Return >0 when
@@ -194,12 +211,21 @@ out_unlock:
 	return ret;
 }
 
-void vfio_pci_intx_unmask(struct vfio_pci_core_device *vdev)
+static void __vfio_pci_intx_unmask(struct vfio_pci_core_device *vdev)
 {
+	lockdep_assert_held(&vdev->igate);
+
 	if (vfio_pci_intx_unmask_handler(vdev, NULL) > 0)
 		vfio_send_intx_eventfd(vdev, NULL);
 }
 
+void vfio_pci_intx_unmask(struct vfio_pci_core_device *vdev)
+{
+	mutex_lock(&vdev->igate);
+	__vfio_pci_intx_unmask(vdev);
+	mutex_unlock(&vdev->igate);
+}
+
 static irqreturn_t vfio_intx_handler(int irq, void *dev_id)
 {
 	struct vfio_pci_core_device *vdev = dev_id;
@@ -231,97 +257,100 @@ static irqreturn_t vfio_intx_handler(int irq, void *dev_id)
 	return ret;
 }
 
-static int vfio_intx_enable(struct vfio_pci_core_device *vdev)
+static int vfio_intx_enable(struct vfio_pci_core_device *vdev,
+			    struct eventfd_ctx *trigger)
 {
+	struct pci_dev *pdev = vdev->pdev;
 	struct vfio_pci_irq_ctx *ctx;
+	unsigned long irqflags;
+	char *name;
+	int ret;
 
 	if (!is_irq_none(vdev))
 		return -EINVAL;
 
-	if (!vdev->pdev->irq)
+	if (!pdev->irq)
 		return -ENODEV;
 
+	name = kasprintf(GFP_KERNEL_ACCOUNT, "vfio-intx(%s)", pci_name(pdev));
+	if (!name)
+		return -ENOMEM;
+
 	ctx = vfio_irq_ctx_alloc(vdev, 0);
 	if (!ctx)
 		return -ENOMEM;
 
+	ctx->name = name;
+	ctx->trigger = trigger;
+
 	/*
-	 * If the virtual interrupt is masked, restore it.  Devices
-	 * supporting DisINTx can be masked at the hardware level
-	 * here, non-PCI-2.3 devices will have to wait until the
-	 * interrupt is enabled.
+	 * Fill the initial masked state based on virq_disabled.  After
+	 * enable, changing the DisINTx bit in vconfig directly changes INTx
+	 * masking.  igate prevents races during setup, once running masked
+	 * is protected via irqlock.
+	 *
+	 * Devices supporting DisINTx also reflect the current mask state in
+	 * the physical DisINTx bit, which is not affected during IRQ setup.
+	 *
+	 * Devices without DisINTx support require an exclusive interrupt.
+	 * IRQ masking is performed at the IRQ chip.  Again, igate protects
+	 * against races during setup and IRQ handlers and irqfds are not
+	 * yet active, therefore masked is stable and can be used to
+	 * conditionally auto-enable the IRQ.
+	 *
+	 * irq_type must be stable while the IRQ handler is registered,
+	 * therefore it must be set before request_irq().
 	 */
 	ctx->masked = vdev->virq_disabled;
-	if (vdev->pci_2_3)
-		pci_intx(vdev->pdev, !ctx->masked);
+	if (vdev->pci_2_3) {
+		pci_intx(pdev, !ctx->masked);
+		irqflags = IRQF_SHARED;
+	} else {
+		irqflags = ctx->masked ? IRQF_NO_AUTOEN : 0;
+	}
 
 	vdev->irq_type = VFIO_PCI_INTX_IRQ_INDEX;
 
+	ret = request_irq(pdev->irq, vfio_intx_handler,
+			  irqflags, ctx->name, vdev);
+	if (ret) {
+		vdev->irq_type = VFIO_PCI_NUM_IRQS;
+		kfree(name);
+		vfio_irq_ctx_free(vdev, ctx, 0);
+		return ret;
+	}
+
 	return 0;
 }
 
-static int vfio_intx_set_signal(struct vfio_pci_core_device *vdev, int fd)
+static int vfio_intx_set_signal(struct vfio_pci_core_device *vdev,
+				struct eventfd_ctx *trigger)
 {
 	struct pci_dev *pdev = vdev->pdev;
-	unsigned long irqflags = IRQF_SHARED;
 	struct vfio_pci_irq_ctx *ctx;
-	struct eventfd_ctx *trigger;
-	unsigned long flags;
-	int ret;
+	struct eventfd_ctx *old;
 
 	ctx = vfio_irq_ctx_get(vdev, 0);
 	if (WARN_ON_ONCE(!ctx))
 		return -EINVAL;
 
-	if (ctx->trigger) {
-		free_irq(pdev->irq, vdev);
-		kfree(ctx->name);
-		eventfd_ctx_put(ctx->trigger);
-		ctx->trigger = NULL;
-	}
-
-	if (fd < 0) /* Disable only */
-		return 0;
+	old = ctx->trigger;
 
-	ctx->name = kasprintf(GFP_KERNEL_ACCOUNT, "vfio-intx(%s)",
-			      pci_name(pdev));
-	if (!ctx->name)
-		return -ENOMEM;
+	WRITE_ONCE(ctx->trigger, trigger);
 
-	trigger = eventfd_ctx_fdget(fd);
-	if (IS_ERR(trigger)) {
-		kfree(ctx->name);
-		return PTR_ERR(trigger);
+	/* Releasing an old ctx requires synchronizing in-flight users */
+	if (old) {
+		synchronize_irq(pdev->irq);
+		vfio_virqfd_flush_thread(&ctx->unmask);
+		eventfd_ctx_put(old);
 	}
 
-	ctx->trigger = trigger;
-
-	if (!vdev->pci_2_3)
-		irqflags = 0;
-
-	ret = request_irq(pdev->irq, vfio_intx_handler,
-			  irqflags, ctx->name, vdev);
-	if (ret) {
-		ctx->trigger = NULL;
-		kfree(ctx->name);
-		eventfd_ctx_put(trigger);
-		return ret;
-	}
-
-	/*
-	 * INTx disable will stick across the new irq setup,
-	 * disable_irq won't.
-	 */
-	spin_lock_irqsave(&vdev->irqlock, flags);
-	if (!vdev->pci_2_3 && ctx->masked)
-		disable_irq_nosync(pdev->irq);
-	spin_unlock_irqrestore(&vdev->irqlock, flags);
-
 	return 0;
 }
 
 static void vfio_intx_disable(struct vfio_pci_core_device *vdev)
 {
+	struct pci_dev *pdev = vdev->pdev;
 	struct vfio_pci_irq_ctx *ctx;
 
 	ctx = vfio_irq_ctx_get(vdev, 0);
@@ -329,10 +358,13 @@ static void vfio_intx_disable(struct vfio_pci_core_device *vdev)
 	if (ctx) {
 		vfio_virqfd_disable(&ctx->unmask);
 		vfio_virqfd_disable(&ctx->mask);
+		free_irq(pdev->irq, vdev);
+		if (ctx->trigger)
+			eventfd_ctx_put(ctx->trigger);
+		kfree(ctx->name);
+		vfio_irq_ctx_free(vdev, ctx, 0);
 	}
-	vfio_intx_set_signal(vdev, -1);
 	vdev->irq_type = VFIO_PCI_NUM_IRQS;
-	vfio_irq_ctx_free(vdev, ctx, 0);
 }
 
 /*
@@ -560,11 +592,11 @@ static int vfio_pci_set_intx_unmask(struct vfio_pci_core_device *vdev,
 		return -EINVAL;
 
 	if (flags & VFIO_IRQ_SET_DATA_NONE) {
-		vfio_pci_intx_unmask(vdev);
+		__vfio_pci_intx_unmask(vdev);
 	} else if (flags & VFIO_IRQ_SET_DATA_BOOL) {
 		uint8_t unmask = *(uint8_t *)data;
 		if (unmask)
-			vfio_pci_intx_unmask(vdev);
+			__vfio_pci_intx_unmask(vdev);
 	} else if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
 		struct vfio_pci_irq_ctx *ctx = vfio_irq_ctx_get(vdev, 0);
 		int32_t fd = *(int32_t *)data;
@@ -591,11 +623,11 @@ static int vfio_pci_set_intx_mask(struct vfio_pci_core_device *vdev,
 		return -EINVAL;
 
 	if (flags & VFIO_IRQ_SET_DATA_NONE) {
-		vfio_pci_intx_mask(vdev);
+		__vfio_pci_intx_mask(vdev);
 	} else if (flags & VFIO_IRQ_SET_DATA_BOOL) {
 		uint8_t mask = *(uint8_t *)data;
 		if (mask)
-			vfio_pci_intx_mask(vdev);
+			__vfio_pci_intx_mask(vdev);
 	} else if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
 		return -ENOTTY; /* XXX implement me */
 	}
@@ -616,19 +648,23 @@ static int vfio_pci_set_intx_trigger(struct vfio_pci_core_device *vdev,
 		return -EINVAL;
 
 	if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
+		struct eventfd_ctx *trigger = NULL;
 		int32_t fd = *(int32_t *)data;
 		int ret;
 
-		if (is_intx(vdev))
-			return vfio_intx_set_signal(vdev, fd);
+		if (fd >= 0) {
+			trigger = eventfd_ctx_fdget(fd);
+			if (IS_ERR(trigger))
+				return PTR_ERR(trigger);
+		}
 
-		ret = vfio_intx_enable(vdev);
-		if (ret)
-			return ret;
+		if (is_intx(vdev))
+			ret = vfio_intx_set_signal(vdev, trigger);
+		else
+			ret = vfio_intx_enable(vdev, trigger);
 
-		ret = vfio_intx_set_signal(vdev, fd);
-		if (ret)
-			vfio_intx_disable(vdev);
+		if (ret && trigger)
+			eventfd_ctx_put(trigger);
 
 		return ret;
 	}
diff --git a/drivers/vfio/pci/vfio_pci_rdwr.c b/drivers/vfio/pci/vfio_pci_rdwr.c
index 07fea08ea8a2..03b8f7ada1ac 100644
--- a/drivers/vfio/pci/vfio_pci_rdwr.c
+++ b/drivers/vfio/pci/vfio_pci_rdwr.c
@@ -96,10 +96,10 @@ VFIO_IOREAD(32)
  * reads with -1.  This is intended for handling MSI-X vector tables and
  * leftover space for ROM BARs.
  */
-static ssize_t do_io_rw(struct vfio_pci_core_device *vdev, bool test_mem,
-			void __iomem *io, char __user *buf,
-			loff_t off, size_t count, size_t x_start,
-			size_t x_end, bool iswrite)
+ssize_t vfio_pci_core_do_io_rw(struct vfio_pci_core_device *vdev, bool test_mem,
+			       void __iomem *io, char __user *buf,
+			       loff_t off, size_t count, size_t x_start,
+			       size_t x_end, bool iswrite)
 {
 	ssize_t done = 0;
 	int ret;
@@ -201,6 +201,7 @@ static ssize_t do_io_rw(struct vfio_pci_core_device *vdev, bool test_mem,
 
 	return done;
 }
+EXPORT_SYMBOL_GPL(vfio_pci_core_do_io_rw);
 
 int vfio_pci_core_setup_barmap(struct vfio_pci_core_device *vdev, int bar)
 {
@@ -279,8 +280,8 @@ ssize_t vfio_pci_bar_rw(struct vfio_pci_core_device *vdev, char __user *buf,
 		x_end = vdev->msix_offset + vdev->msix_size;
 	}
 
-	done = do_io_rw(vdev, res->flags & IORESOURCE_MEM, io, buf, pos,
-			count, x_start, x_end, iswrite);
+	done = vfio_pci_core_do_io_rw(vdev, res->flags & IORESOURCE_MEM, io, buf, pos,
+				      count, x_start, x_end, iswrite);
 
 	if (done >= 0)
 		*ppos += done;
@@ -348,7 +349,8 @@ ssize_t vfio_pci_vga_rw(struct vfio_pci_core_device *vdev, char __user *buf,
 	 * probing, so we don't currently worry about access in relation
 	 * to the memory enable bit in the command register.
 	 */
-	done = do_io_rw(vdev, false, iomem, buf, off, count, 0, 0, iswrite);
+	done = vfio_pci_core_do_io_rw(vdev, false, iomem, buf, off, count,
+				      0, 0, iswrite);
 
 	vga_put(vdev->pdev, rsrc);
 
diff --git a/drivers/vfio/pci/virtio/main.c b/drivers/vfio/pci/virtio/main.c
index d5af683837d3..b5d3a8c5bbc9 100644
--- a/drivers/vfio/pci/virtio/main.c
+++ b/drivers/vfio/pci/virtio/main.c
@@ -132,33 +132,6 @@ end:
 	return ret ? ret : count;
 }
 
-static bool range_intersect_range(loff_t range1_start, size_t count1,
-				  loff_t range2_start, size_t count2,
-				  loff_t *start_offset,
-				  size_t *intersect_count,
-				  size_t *register_offset)
-{
-	if (range1_start <= range2_start &&
-	    range1_start + count1 > range2_start) {
-		*start_offset = range2_start - range1_start;
-		*intersect_count = min_t(size_t, count2,
-					 range1_start + count1 - range2_start);
-		*register_offset = 0;
-		return true;
-	}
-
-	if (range1_start > range2_start &&
-	    range1_start < range2_start + count2) {
-		*start_offset = 0;
-		*intersect_count = min_t(size_t, count1,
-					 range2_start + count2 - range1_start);
-		*register_offset = range1_start - range2_start;
-		return true;
-	}
-
-	return false;
-}
-
 static ssize_t virtiovf_pci_read_config(struct vfio_device *core_vdev,
 					char __user *buf, size_t count,
 					loff_t *ppos)
@@ -178,16 +151,18 @@ static ssize_t virtiovf_pci_read_config(struct vfio_device *core_vdev,
 	if (ret < 0)
 		return ret;
 
-	if (range_intersect_range(pos, count, PCI_DEVICE_ID, sizeof(val16),
-				  &copy_offset, &copy_count, &register_offset)) {
+	if (vfio_pci_core_range_intersect_range(pos, count, PCI_DEVICE_ID,
+						sizeof(val16), &copy_offset,
+						&copy_count, &register_offset)) {
 		val16 = cpu_to_le16(VIRTIO_TRANS_ID_NET);
 		if (copy_to_user(buf + copy_offset, (void *)&val16 + register_offset, copy_count))
 			return -EFAULT;
 	}
 
 	if ((le16_to_cpu(virtvdev->pci_cmd) & PCI_COMMAND_IO) &&
-	    range_intersect_range(pos, count, PCI_COMMAND, sizeof(val16),
-				  &copy_offset, &copy_count, &register_offset)) {
+	    vfio_pci_core_range_intersect_range(pos, count, PCI_COMMAND,
+						sizeof(val16), &copy_offset,
+						&copy_count, &register_offset)) {
 		if (copy_from_user((void *)&val16 + register_offset, buf + copy_offset,
 				   copy_count))
 			return -EFAULT;
@@ -197,16 +172,18 @@ static ssize_t virtiovf_pci_read_config(struct vfio_device *core_vdev,
 			return -EFAULT;
 	}
 
-	if (range_intersect_range(pos, count, PCI_REVISION_ID, sizeof(val8),
-				  &copy_offset, &copy_count, &register_offset)) {
+	if (vfio_pci_core_range_intersect_range(pos, count, PCI_REVISION_ID,
+						sizeof(val8), &copy_offset,
+						&copy_count, &register_offset)) {
 		/* Transional needs to have revision 0 */
 		val8 = 0;
 		if (copy_to_user(buf + copy_offset, &val8, copy_count))
 			return -EFAULT;
 	}
 
-	if (range_intersect_range(pos, count, PCI_BASE_ADDRESS_0, sizeof(val32),
-				  &copy_offset, &copy_count, &register_offset)) {
+	if (vfio_pci_core_range_intersect_range(pos, count, PCI_BASE_ADDRESS_0,
+						sizeof(val32), &copy_offset,
+						&copy_count, &register_offset)) {
 		u32 bar_mask = ~(virtvdev->bar0_virtual_buf_size - 1);
 		u32 pci_base_addr_0 = le32_to_cpu(virtvdev->pci_base_addr_0);
 
@@ -215,8 +192,9 @@ static ssize_t virtiovf_pci_read_config(struct vfio_device *core_vdev,
 			return -EFAULT;
 	}
 
-	if (range_intersect_range(pos, count, PCI_SUBSYSTEM_ID, sizeof(val16),
-				  &copy_offset, &copy_count, &register_offset)) {
+	if (vfio_pci_core_range_intersect_range(pos, count, PCI_SUBSYSTEM_ID,
+						sizeof(val16), &copy_offset,
+						&copy_count, &register_offset)) {
 		/*
 		 * Transitional devices use the PCI subsystem device id as
 		 * virtio device id, same as legacy driver always did.
@@ -227,8 +205,9 @@ static ssize_t virtiovf_pci_read_config(struct vfio_device *core_vdev,
 			return -EFAULT;
 	}
 
-	if (range_intersect_range(pos, count, PCI_SUBSYSTEM_VENDOR_ID, sizeof(val16),
-				  &copy_offset, &copy_count, &register_offset)) {
+	if (vfio_pci_core_range_intersect_range(pos, count, PCI_SUBSYSTEM_VENDOR_ID,
+						sizeof(val16), &copy_offset,
+						&copy_count, &register_offset)) {
 		val16 = cpu_to_le16(PCI_VENDOR_ID_REDHAT_QUMRANET);
 		if (copy_to_user(buf + copy_offset, (void *)&val16 + register_offset,
 				 copy_count))
@@ -270,19 +249,20 @@ static ssize_t virtiovf_pci_write_config(struct vfio_device *core_vdev,
 	loff_t copy_offset;
 	size_t copy_count;
 
-	if (range_intersect_range(pos, count, PCI_COMMAND, sizeof(virtvdev->pci_cmd),
-				  &copy_offset, &copy_count,
-				  &register_offset)) {
+	if (vfio_pci_core_range_intersect_range(pos, count, PCI_COMMAND,
+						sizeof(virtvdev->pci_cmd),
+						&copy_offset, &copy_count,
+						&register_offset)) {
 		if (copy_from_user((void *)&virtvdev->pci_cmd + register_offset,
 				   buf + copy_offset,
 				   copy_count))
 			return -EFAULT;
 	}
 
-	if (range_intersect_range(pos, count, PCI_BASE_ADDRESS_0,
-				  sizeof(virtvdev->pci_base_addr_0),
-				  &copy_offset, &copy_count,
-				  &register_offset)) {
+	if (vfio_pci_core_range_intersect_range(pos, count, PCI_BASE_ADDRESS_0,
+						sizeof(virtvdev->pci_base_addr_0),
+						&copy_offset, &copy_count,
+						&register_offset)) {
 		if (copy_from_user((void *)&virtvdev->pci_base_addr_0 + register_offset,
 				   buf + copy_offset,
 				   copy_count))
diff --git a/drivers/vfio/platform/vfio_amba.c b/drivers/vfio/platform/vfio_amba.c
index 6464b3939ebc..485c6f9161a9 100644
--- a/drivers/vfio/platform/vfio_amba.c
+++ b/drivers/vfio/platform/vfio_amba.c
@@ -122,16 +122,16 @@ static const struct vfio_device_ops vfio_amba_ops = {
 	.detach_ioas	= vfio_iommufd_physical_detach_ioas,
 };
 
-static const struct amba_id pl330_ids[] = {
+static const struct amba_id vfio_amba_ids[] = {
 	{ 0, 0 },
 };
 
-MODULE_DEVICE_TABLE(amba, pl330_ids);
+MODULE_DEVICE_TABLE(amba, vfio_amba_ids);
 
 static struct amba_driver vfio_amba_driver = {
 	.probe = vfio_amba_probe,
 	.remove = vfio_amba_remove,
-	.id_table = pl330_ids,
+	.id_table = vfio_amba_ids,
 	.drv = {
 		.name = "vfio-amba",
 		.owner = THIS_MODULE,
diff --git a/drivers/vfio/platform/vfio_platform.c b/drivers/vfio/platform/vfio_platform.c
index 8cf22fa65baa..42d1462c5e19 100644
--- a/drivers/vfio/platform/vfio_platform.c
+++ b/drivers/vfio/platform/vfio_platform.c
@@ -85,14 +85,13 @@ static void vfio_platform_release_dev(struct vfio_device *core_vdev)
 	vfio_platform_release_common(vdev);
 }
 
-static int vfio_platform_remove(struct platform_device *pdev)
+static void vfio_platform_remove(struct platform_device *pdev)
 {
 	struct vfio_platform_device *vdev = dev_get_drvdata(&pdev->dev);
 
 	vfio_unregister_group_dev(&vdev->vdev);
 	pm_runtime_disable(vdev->device);
 	vfio_put_device(&vdev->vdev);
-	return 0;
 }
 
 static const struct vfio_device_ops vfio_platform_ops = {
@@ -113,7 +112,7 @@ static const struct vfio_device_ops vfio_platform_ops = {
 
 static struct platform_driver vfio_platform_driver = {
 	.probe		= vfio_platform_probe,
-	.remove		= vfio_platform_remove,
+	.remove_new	= vfio_platform_remove,
 	.driver	= {
 		.name	= "vfio-platform",
 	},
diff --git a/drivers/vfio/platform/vfio_platform_irq.c b/drivers/vfio/platform/vfio_platform_irq.c
index 61a1bfb68ac7..ef41ecef83af 100644
--- a/drivers/vfio/platform/vfio_platform_irq.c
+++ b/drivers/vfio/platform/vfio_platform_irq.c
@@ -136,6 +136,16 @@ static int vfio_platform_set_irq_unmask(struct vfio_platform_device *vdev,
 	return 0;
 }
 
+/*
+ * The trigger eventfd is guaranteed valid in the interrupt path
+ * and protected by the igate mutex when triggered via ioctl.
+ */
+static void vfio_send_eventfd(struct vfio_platform_irq *irq_ctx)
+{
+	if (likely(irq_ctx->trigger))
+		eventfd_signal(irq_ctx->trigger);
+}
+
 static irqreturn_t vfio_automasked_irq_handler(int irq, void *dev_id)
 {
 	struct vfio_platform_irq *irq_ctx = dev_id;
@@ -155,7 +165,7 @@ static irqreturn_t vfio_automasked_irq_handler(int irq, void *dev_id)
 	spin_unlock_irqrestore(&irq_ctx->lock, flags);
 
 	if (ret == IRQ_HANDLED)
-		eventfd_signal(irq_ctx->trigger);
+		vfio_send_eventfd(irq_ctx);
 
 	return ret;
 }
@@ -164,52 +174,40 @@ static irqreturn_t vfio_irq_handler(int irq, void *dev_id)
 {
 	struct vfio_platform_irq *irq_ctx = dev_id;
 
-	eventfd_signal(irq_ctx->trigger);
+	vfio_send_eventfd(irq_ctx);
 
 	return IRQ_HANDLED;
 }
 
 static int vfio_set_trigger(struct vfio_platform_device *vdev, int index,
-			    int fd, irq_handler_t handler)
+			    int fd)
 {
 	struct vfio_platform_irq *irq = &vdev->irqs[index];
 	struct eventfd_ctx *trigger;
-	int ret;
 
 	if (irq->trigger) {
-		irq_clear_status_flags(irq->hwirq, IRQ_NOAUTOEN);
-		free_irq(irq->hwirq, irq);
-		kfree(irq->name);
+		disable_irq(irq->hwirq);
 		eventfd_ctx_put(irq->trigger);
 		irq->trigger = NULL;
 	}
 
 	if (fd < 0) /* Disable only */
 		return 0;
-	irq->name = kasprintf(GFP_KERNEL_ACCOUNT, "vfio-irq[%d](%s)",
-			      irq->hwirq, vdev->name);
-	if (!irq->name)
-		return -ENOMEM;
 
 	trigger = eventfd_ctx_fdget(fd);
-	if (IS_ERR(trigger)) {
-		kfree(irq->name);
+	if (IS_ERR(trigger))
 		return PTR_ERR(trigger);
-	}
 
 	irq->trigger = trigger;
 
-	irq_set_status_flags(irq->hwirq, IRQ_NOAUTOEN);
-	ret = request_irq(irq->hwirq, handler, 0, irq->name, irq);
-	if (ret) {
-		kfree(irq->name);
-		eventfd_ctx_put(trigger);
-		irq->trigger = NULL;
-		return ret;
-	}
-
-	if (!irq->masked)
-		enable_irq(irq->hwirq);
+	/*
+	 * irq->masked effectively provides nested disables within the overall
+	 * enable relative to trigger.  Specifically request_irq() is called
+	 * with NO_AUTOEN, therefore the IRQ is initially disabled.  The user
+	 * may only further disable the IRQ with a MASK operations because
+	 * irq->masked is initially false.
+	 */
+	enable_irq(irq->hwirq);
 
 	return 0;
 }
@@ -228,7 +226,7 @@ static int vfio_platform_set_irq_trigger(struct vfio_platform_device *vdev,
 		handler = vfio_irq_handler;
 
 	if (!count && (flags & VFIO_IRQ_SET_DATA_NONE))
-		return vfio_set_trigger(vdev, index, -1, handler);
+		return vfio_set_trigger(vdev, index, -1);
 
 	if (start != 0 || count != 1)
 		return -EINVAL;
@@ -236,7 +234,7 @@ static int vfio_platform_set_irq_trigger(struct vfio_platform_device *vdev,
 	if (flags & VFIO_IRQ_SET_DATA_EVENTFD) {
 		int32_t fd = *(int32_t *)data;
 
-		return vfio_set_trigger(vdev, index, fd, handler);
+		return vfio_set_trigger(vdev, index, fd);
 	}
 
 	if (flags & VFIO_IRQ_SET_DATA_NONE) {
@@ -260,6 +258,14 @@ int vfio_platform_set_irqs_ioctl(struct vfio_platform_device *vdev,
 		    unsigned start, unsigned count, uint32_t flags,
 		    void *data) = NULL;
 
+	/*
+	 * For compatibility, errors from request_irq() are local to the
+	 * SET_IRQS path and reflected in the name pointer.  This allows,
+	 * for example, polling mode fallback for an exclusive IRQ failure.
+	 */
+	if (IS_ERR(vdev->irqs[index].name))
+		return PTR_ERR(vdev->irqs[index].name);
+
 	switch (flags & VFIO_IRQ_SET_ACTION_TYPE_MASK) {
 	case VFIO_IRQ_SET_ACTION_MASK:
 		func = vfio_platform_set_irq_mask;
@@ -280,7 +286,7 @@ int vfio_platform_set_irqs_ioctl(struct vfio_platform_device *vdev,
 
 int vfio_platform_irq_init(struct vfio_platform_device *vdev)
 {
-	int cnt = 0, i;
+	int cnt = 0, i, ret = 0;
 
 	while (vdev->get_irq(vdev, cnt) >= 0)
 		cnt++;
@@ -292,37 +298,70 @@ int vfio_platform_irq_init(struct vfio_platform_device *vdev)
 
 	for (i = 0; i < cnt; i++) {
 		int hwirq = vdev->get_irq(vdev, i);
+		irq_handler_t handler = vfio_irq_handler;
 
-		if (hwirq < 0)
+		if (hwirq < 0) {
+			ret = -EINVAL;
 			goto err;
+		}
 
 		spin_lock_init(&vdev->irqs[i].lock);
 
 		vdev->irqs[i].flags = VFIO_IRQ_INFO_EVENTFD;
 
-		if (irq_get_trigger_type(hwirq) & IRQ_TYPE_LEVEL_MASK)
+		if (irq_get_trigger_type(hwirq) & IRQ_TYPE_LEVEL_MASK) {
 			vdev->irqs[i].flags |= VFIO_IRQ_INFO_MASKABLE
 						| VFIO_IRQ_INFO_AUTOMASKED;
+			handler = vfio_automasked_irq_handler;
+		}
 
 		vdev->irqs[i].count = 1;
 		vdev->irqs[i].hwirq = hwirq;
 		vdev->irqs[i].masked = false;
+		vdev->irqs[i].name = kasprintf(GFP_KERNEL_ACCOUNT,
+					       "vfio-irq[%d](%s)", hwirq,
+					       vdev->name);
+		if (!vdev->irqs[i].name) {
+			ret = -ENOMEM;
+			goto err;
+		}
+
+		ret = request_irq(hwirq, handler, IRQF_NO_AUTOEN,
+				  vdev->irqs[i].name, &vdev->irqs[i]);
+		if (ret) {
+			kfree(vdev->irqs[i].name);
+			vdev->irqs[i].name = ERR_PTR(ret);
+		}
 	}
 
 	vdev->num_irqs = cnt;
 
 	return 0;
 err:
+	for (--i; i >= 0; i--) {
+		if (!IS_ERR(vdev->irqs[i].name)) {
+			free_irq(vdev->irqs[i].hwirq, &vdev->irqs[i]);
+			kfree(vdev->irqs[i].name);
+		}
+	}
 	kfree(vdev->irqs);
-	return -EINVAL;
+	return ret;
 }
 
 void vfio_platform_irq_cleanup(struct vfio_platform_device *vdev)
 {
 	int i;
 
-	for (i = 0; i < vdev->num_irqs; i++)
-		vfio_set_trigger(vdev, i, -1, NULL);
+	for (i = 0; i < vdev->num_irqs; i++) {
+		vfio_virqfd_disable(&vdev->irqs[i].mask);
+		vfio_virqfd_disable(&vdev->irqs[i].unmask);
+		if (!IS_ERR(vdev->irqs[i].name)) {
+			free_irq(vdev->irqs[i].hwirq, &vdev->irqs[i]);
+			if (vdev->irqs[i].trigger)
+				eventfd_ctx_put(vdev->irqs[i].trigger);
+			kfree(vdev->irqs[i].name);
+		}
+	}
 
 	vdev->num_irqs = 0;
 	kfree(vdev->irqs);
diff --git a/drivers/vfio/vfio_iommu_type1.c b/drivers/vfio/vfio_iommu_type1.c
index b2854d7939ce..b5c15fe8f9fc 100644
--- a/drivers/vfio/vfio_iommu_type1.c
+++ b/drivers/vfio/vfio_iommu_type1.c
@@ -567,18 +567,6 @@ static int vaddr_get_pfns(struct mm_struct *mm, unsigned long vaddr,
 	ret = pin_user_pages_remote(mm, vaddr, npages, flags | FOLL_LONGTERM,
 				    pages, NULL);
 	if (ret > 0) {
-		int i;
-
-		/*
-		 * The zero page is always resident, we don't need to pin it
-		 * and it falls into our invalid/reserved test so we don't
-		 * unpin in put_pfn().  Unpin all zero pages in the batch here.
-		 */
-		for (i = 0 ; i < ret; i++) {
-			if (unlikely(is_zero_pfn(page_to_pfn(pages[i]))))
-				unpin_user_page(pages[i]);
-		}
-
 		*pfn = page_to_pfn(pages[0]);
 		goto done;
 	}
diff --git a/drivers/vfio/virqfd.c b/drivers/vfio/virqfd.c
index 29c564b7a6e1..532269133801 100644
--- a/drivers/vfio/virqfd.c
+++ b/drivers/vfio/virqfd.c
@@ -101,6 +101,13 @@ static void virqfd_inject(struct work_struct *work)
 		virqfd->thread(virqfd->opaque, virqfd->data);
 }
 
+static void virqfd_flush_inject(struct work_struct *work)
+{
+	struct virqfd *virqfd = container_of(work, struct virqfd, flush_inject);
+
+	flush_work(&virqfd->inject);
+}
+
 int vfio_virqfd_enable(void *opaque,
 		       int (*handler)(void *, void *),
 		       void (*thread)(void *, void *),
@@ -124,6 +131,7 @@ int vfio_virqfd_enable(void *opaque,
 
 	INIT_WORK(&virqfd->shutdown, virqfd_shutdown);
 	INIT_WORK(&virqfd->inject, virqfd_inject);
+	INIT_WORK(&virqfd->flush_inject, virqfd_flush_inject);
 
 	irqfd = fdget(fd);
 	if (!irqfd.file) {
@@ -213,3 +221,16 @@ void vfio_virqfd_disable(struct virqfd **pvirqfd)
 	flush_workqueue(vfio_irqfd_cleanup_wq);
 }
 EXPORT_SYMBOL_GPL(vfio_virqfd_disable);
+
+void vfio_virqfd_flush_thread(struct virqfd **pvirqfd)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&virqfd_lock, flags);
+	if (*pvirqfd && (*pvirqfd)->thread)
+		queue_work(vfio_irqfd_cleanup_wq, &(*pvirqfd)->flush_inject);
+	spin_unlock_irqrestore(&virqfd_lock, flags);
+
+	flush_workqueue(vfio_irqfd_cleanup_wq);
+}
+EXPORT_SYMBOL_GPL(vfio_virqfd_flush_thread);