[PATCH V1 mlx5-next 0/4] Improve mlx5 live migration driver

[PATCH V1 mlx5-next 0/4] Improve mlx5 live migration driver

[Yishai Hadas]

This series improves mlx5 live migration driver in few aspects as of
below.

Refactor to enable running migration commands in parallel over the PF
command interface.

To achieve that we exposed from mlx5_core an API to let the VF be
notified before that the PF command interface goes down/up. (e.g. PF
reload upon health recovery).

Once having the above functionality in place mlx5 vfio doesn't need any
more to obtain the global PF lock upon using the command interface but
can rely on the above mechanism to be in sync with the PF.

This can enable parallel VFs migration over the PF command interface
from kernel driver point of view.

In addition,
Moved to use the PF async command mode for the SAVE state command.
This enables returning earlier to user space upon issuing successfully
the command and improve latency by let things run in parallel.

Alex, as this series touches mlx5_core we may need to send this in a
pull request format to VFIO to avoid conflicts before acceptance.

V1:
- Put the net/mlx5 patch as the first patch based on Jason's note.
- Refactor and combine the previous first patch with the third patch to
  have a cleaner readable code, this follows Alex's notes on V0.

V0: https://lore.kernel.org/netdev/20220504213309.GM49344@nvidia.com/T/

Yishai
Yishai Hadas (4):
  net/mlx5: Expose mlx5_sriov_blocking_notifier_register /  unregister
    APIs
  vfio/mlx5: Manage the VF attach/detach callback from the PF
  vfio/mlx5: Refactor to enable VFs migration in parallel
  vfio/mlx5: Run the SAVE state command in an async mode

 .../net/ethernet/mellanox/mlx5/core/sriov.c   |  65 ++++-
 drivers/vfio/pci/mlx5/cmd.c                   | 226 +++++++++++++-----
 drivers/vfio/pci/mlx5/cmd.h                   |  50 +++-
 drivers/vfio/pci/mlx5/main.c                  | 132 +++++-----
 include/linux/mlx5/driver.h                   |  12 +
 5 files changed, 352 insertions(+), 133 deletions(-)

-- 
2.18.1

[PATCH V1 mlx5-next 1/4] net/mlx5: Expose mlx5_sriov_blocking_notifier_register / unregister APIs

[Yishai Hadas]

Expose mlx5_sriov_blocking_notifier_register / unregister APIs to let a
VF register to be notified for its enablement / disablement by the PF.

Upon VF probe it will call mlx5_sriov_blocking_notifier_register() with
its notifier block and upon VF remove it will call
mlx5_sriov_blocking_notifier_unregister() to drop its registration.

This can give a VF the ability to clean some resources upon disable
before that the command interface goes down and on the other hand sets
some stuff before that it's enabled.

This may be used by a VF which is migration capable in few cases.(e.g.
PF load/unload upon an health recovery).

Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 .../net/ethernet/mellanox/mlx5/core/sriov.c   | 65 ++++++++++++++++++-
 include/linux/mlx5/driver.h                   | 12 ++++
 2 files changed, 76 insertions(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/sriov.c b/drivers/net/ethernet/mellanox/mlx5/core/sriov.c
index 887ee0f729d1..2935614f6fa9 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/sriov.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/sriov.c
@@ -87,6 +87,11 @@ static int mlx5_device_enable_sriov(struct mlx5_core_dev *dev, int num_vfs)
 enable_vfs_hca:
 	num_msix_count = mlx5_get_default_msix_vec_count(dev, num_vfs);
 	for (vf = 0; vf < num_vfs; vf++) {
+		/* Notify the VF before its enablement to let it set
+		 * some stuff.
+		 */
+		blocking_notifier_call_chain(&sriov->vfs_ctx[vf].notifier,
+					     MLX5_PF_NOTIFY_ENABLE_VF, dev);
 		err = mlx5_core_enable_hca(dev, vf + 1);
 		if (err) {
 			mlx5_core_warn(dev, "failed to enable VF %d (%d)\n", vf, err);
@@ -127,6 +132,11 @@ mlx5_device_disable_sriov(struct mlx5_core_dev *dev, int num_vfs, bool clear_vf)
 	for (vf = num_vfs - 1; vf >= 0; vf--) {
 		if (!sriov->vfs_ctx[vf].enabled)
 			continue;
+		/* Notify the VF before its disablement to let it clean
+		 * some resources.
+		 */
+		blocking_notifier_call_chain(&sriov->vfs_ctx[vf].notifier,
+					     MLX5_PF_NOTIFY_DISABLE_VF, dev);
 		err = mlx5_core_disable_hca(dev, vf + 1);
 		if (err) {
 			mlx5_core_warn(dev, "failed to disable VF %d\n", vf);
@@ -257,7 +267,7 @@ int mlx5_sriov_init(struct mlx5_core_dev *dev)
 {
 	struct mlx5_core_sriov *sriov = &dev->priv.sriov;
 	struct pci_dev *pdev = dev->pdev;
-	int total_vfs;
+	int total_vfs, i;
 
 	if (!mlx5_core_is_pf(dev))
 		return 0;
@@ -269,6 +279,9 @@ int mlx5_sriov_init(struct mlx5_core_dev *dev)
 	if (!sriov->vfs_ctx)
 		return -ENOMEM;
 
+	for (i = 0; i < total_vfs; i++)
+		BLOCKING_INIT_NOTIFIER_HEAD(&sriov->vfs_ctx[i].notifier);
+
 	return 0;
 }
 
@@ -281,3 +294,53 @@ void mlx5_sriov_cleanup(struct mlx5_core_dev *dev)
 
 	kfree(sriov->vfs_ctx);
 }
+
+/**
+ * mlx5_sriov_blocking_notifier_unregister - Unregister a VF from
+ * a notification block chain.
+ *
+ * @mdev: The mlx5 core device.
+ * @vf_id: The VF id.
+ * @nb: The notifier block to be unregistered.
+ */
+void mlx5_sriov_blocking_notifier_unregister(struct mlx5_core_dev *mdev,
+					     int vf_id,
+					     struct notifier_block *nb)
+{
+	struct mlx5_vf_context *vfs_ctx;
+	struct mlx5_core_sriov *sriov;
+
+	sriov = &mdev->priv.sriov;
+	if (WARN_ON(vf_id < 0 || vf_id >= sriov->num_vfs))
+		return;
+
+	vfs_ctx = &sriov->vfs_ctx[vf_id];
+	blocking_notifier_chain_unregister(&vfs_ctx->notifier, nb);
+}
+EXPORT_SYMBOL(mlx5_sriov_blocking_notifier_unregister);
+
+/**
+ * mlx5_sriov_blocking_notifier_register - Register a VF notification
+ * block chain.
+ *
+ * @mdev: The mlx5 core device.
+ * @vf_id: The VF id.
+ * @nb: The notifier block to be called upon the VF events.
+ *
+ * Returns 0 on success or an error code.
+ */
+int mlx5_sriov_blocking_notifier_register(struct mlx5_core_dev *mdev,
+					  int vf_id,
+					  struct notifier_block *nb)
+{
+	struct mlx5_vf_context *vfs_ctx;
+	struct mlx5_core_sriov *sriov;
+
+	sriov = &mdev->priv.sriov;
+	if (vf_id < 0 || vf_id >= sriov->num_vfs)
+		return -EINVAL;
+
+	vfs_ctx = &sriov->vfs_ctx[vf_id];
+	return blocking_notifier_chain_register(&vfs_ctx->notifier, nb);
+}
+EXPORT_SYMBOL(mlx5_sriov_blocking_notifier_register);
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index 9424503eb8d3..3d1594bad4ec 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -445,6 +445,11 @@ struct mlx5_qp_table {
 	struct radix_tree_root	tree;
 };
 
+enum {
+	MLX5_PF_NOTIFY_DISABLE_VF,
+	MLX5_PF_NOTIFY_ENABLE_VF,
+};
+
 struct mlx5_vf_context {
 	int	enabled;
 	u64	port_guid;
@@ -455,6 +460,7 @@ struct mlx5_vf_context {
 	u8	port_guid_valid:1;
 	u8	node_guid_valid:1;
 	enum port_state_policy	policy;
+	struct blocking_notifier_head notifier;
 };
 
 struct mlx5_core_sriov {
@@ -1155,6 +1161,12 @@ int mlx5_dm_sw_icm_dealloc(struct mlx5_core_dev *dev, enum mlx5_sw_icm_type type
 struct mlx5_core_dev *mlx5_vf_get_core_dev(struct pci_dev *pdev);
 void mlx5_vf_put_core_dev(struct mlx5_core_dev *mdev);
 
+int mlx5_sriov_blocking_notifier_register(struct mlx5_core_dev *mdev,
+					  int vf_id,
+					  struct notifier_block *nb);
+void mlx5_sriov_blocking_notifier_unregister(struct mlx5_core_dev *mdev,
+					     int vf_id,
+					     struct notifier_block *nb);
 #ifdef CONFIG_MLX5_CORE_IPOIB
 struct net_device *mlx5_rdma_netdev_alloc(struct mlx5_core_dev *mdev,
 					  struct ib_device *ibdev,
-- 
2.18.1

[PATCH V1 mlx5-next 2/4] vfio/mlx5: Manage the VF attach/detach callback from the PF

[Yishai Hadas]

Manage the VF attach/detach callback from the PF.

This lets the driver to enable parallel VFs migration as will be
introduced in the next patch.

As part of this, reorganize the VF is migratable code to be in a
separate function and rename it to be set_migratable() to match its
functionality.

Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/vfio/pci/mlx5/cmd.c  | 63 ++++++++++++++++++++++++++++++++++++
 drivers/vfio/pci/mlx5/cmd.h  | 22 +++++++++++++
 drivers/vfio/pci/mlx5/main.c | 40 ++++-------------------
 3 files changed, 91 insertions(+), 34 deletions(-)

diff --git a/drivers/vfio/pci/mlx5/cmd.c b/drivers/vfio/pci/mlx5/cmd.c
index 5c9f9218cc1d..5031978ae63a 100644
--- a/drivers/vfio/pci/mlx5/cmd.c
+++ b/drivers/vfio/pci/mlx5/cmd.c
@@ -71,6 +71,69 @@ int mlx5vf_cmd_query_vhca_migration_state(struct pci_dev *pdev, u16 vhca_id,
 	return ret;
 }
 
+static int mlx5fv_vf_event(struct notifier_block *nb,
+			   unsigned long event, void *data)
+{
+	struct mlx5vf_pci_core_device *mvdev =
+		container_of(nb, struct mlx5vf_pci_core_device, nb);
+
+	mutex_lock(&mvdev->state_mutex);
+	switch (event) {
+	case MLX5_PF_NOTIFY_ENABLE_VF:
+		mvdev->mdev_detach = false;
+		break;
+	case MLX5_PF_NOTIFY_DISABLE_VF:
+		mvdev->mdev_detach = true;
+		break;
+	default:
+		break;
+	}
+	mlx5vf_state_mutex_unlock(mvdev);
+	return 0;
+}
+
+void mlx5vf_cmd_remove_migratable(struct mlx5vf_pci_core_device *mvdev)
+{
+	mlx5_sriov_blocking_notifier_unregister(mvdev->mdev, mvdev->vf_id,
+						&mvdev->nb);
+}
+
+void mlx5vf_cmd_set_migratable(struct mlx5vf_pci_core_device *mvdev)
+{
+	struct pci_dev *pdev = mvdev->core_device.pdev;
+	int ret;
+
+	if (!pdev->is_virtfn)
+		return;
+
+	mvdev->mdev = mlx5_vf_get_core_dev(pdev);
+	if (!mvdev->mdev)
+		return;
+
+	if (!MLX5_CAP_GEN(mvdev->mdev, migration))
+		goto end;
+
+	mvdev->vf_id = pci_iov_vf_id(pdev);
+	if (mvdev->vf_id < 0)
+		goto end;
+
+	mutex_init(&mvdev->state_mutex);
+	spin_lock_init(&mvdev->reset_lock);
+	mvdev->nb.notifier_call = mlx5fv_vf_event;
+	ret = mlx5_sriov_blocking_notifier_register(mvdev->mdev, mvdev->vf_id,
+						    &mvdev->nb);
+	if (ret)
+		goto end;
+
+	mvdev->migrate_cap = 1;
+	mvdev->core_device.vdev.migration_flags =
+		VFIO_MIGRATION_STOP_COPY |
+		VFIO_MIGRATION_P2P;
+
+end:
+	mlx5_vf_put_core_dev(mvdev->mdev);
+}
+
 int mlx5vf_cmd_get_vhca_id(struct pci_dev *pdev, u16 function_id, u16 *vhca_id)
 {
 	struct mlx5_core_dev *mdev = mlx5_vf_get_core_dev(pdev);
diff --git a/drivers/vfio/pci/mlx5/cmd.h b/drivers/vfio/pci/mlx5/cmd.h
index 1392a11a9cc0..340a06b98007 100644
--- a/drivers/vfio/pci/mlx5/cmd.h
+++ b/drivers/vfio/pci/mlx5/cmd.h
@@ -7,6 +7,7 @@
 #define MLX5_VFIO_CMD_H
 
 #include <linux/kernel.h>
+#include <linux/vfio_pci_core.h>
 #include <linux/mlx5/driver.h>
 
 struct mlx5_vf_migration_file {
@@ -24,13 +25,34 @@ struct mlx5_vf_migration_file {
 	unsigned long last_offset;
 };
 
+struct mlx5vf_pci_core_device {
+	struct vfio_pci_core_device core_device;
+	int vf_id;
+	u16 vhca_id;
+	u8 migrate_cap:1;
+	u8 deferred_reset:1;
+	/* protect migration state */
+	struct mutex state_mutex;
+	enum vfio_device_mig_state mig_state;
+	/* protect the reset_done flow */
+	spinlock_t reset_lock;
+	struct mlx5_vf_migration_file *resuming_migf;
+	struct mlx5_vf_migration_file *saving_migf;
+	struct notifier_block nb;
+	struct mlx5_core_dev *mdev;
+	u8 mdev_detach:1;
+};
+
 int mlx5vf_cmd_suspend_vhca(struct pci_dev *pdev, u16 vhca_id, u16 op_mod);
 int mlx5vf_cmd_resume_vhca(struct pci_dev *pdev, u16 vhca_id, u16 op_mod);
 int mlx5vf_cmd_query_vhca_migration_state(struct pci_dev *pdev, u16 vhca_id,
 					  size_t *state_size);
 int mlx5vf_cmd_get_vhca_id(struct pci_dev *pdev, u16 function_id, u16 *vhca_id);
+void mlx5vf_cmd_set_migratable(struct mlx5vf_pci_core_device *mvdev);
+void mlx5vf_cmd_remove_migratable(struct mlx5vf_pci_core_device *mvdev);
 int mlx5vf_cmd_save_vhca_state(struct pci_dev *pdev, u16 vhca_id,
 			       struct mlx5_vf_migration_file *migf);
 int mlx5vf_cmd_load_vhca_state(struct pci_dev *pdev, u16 vhca_id,
 			       struct mlx5_vf_migration_file *migf);
+void mlx5vf_state_mutex_unlock(struct mlx5vf_pci_core_device *mvdev);
 #endif /* MLX5_VFIO_CMD_H */
diff --git a/drivers/vfio/pci/mlx5/main.c b/drivers/vfio/pci/mlx5/main.c
index bbec5d288fee..9716c87e31f9 100644
--- a/drivers/vfio/pci/mlx5/main.c
+++ b/drivers/vfio/pci/mlx5/main.c
@@ -17,7 +17,6 @@
 #include <linux/uaccess.h>
 #include <linux/vfio.h>
 #include <linux/sched/mm.h>
-#include <linux/vfio_pci_core.h>
 #include <linux/anon_inodes.h>
 
 #include "cmd.h"
@@ -25,20 +24,6 @@
 /* Arbitrary to prevent userspace from consuming endless memory */
 #define MAX_MIGRATION_SIZE (512*1024*1024)
 
-struct mlx5vf_pci_core_device {
-	struct vfio_pci_core_device core_device;
-	u16 vhca_id;
-	u8 migrate_cap:1;
-	u8 deferred_reset:1;
-	/* protect migration state */
-	struct mutex state_mutex;
-	enum vfio_device_mig_state mig_state;
-	/* protect the reset_done flow */
-	spinlock_t reset_lock;
-	struct mlx5_vf_migration_file *resuming_migf;
-	struct mlx5_vf_migration_file *saving_migf;
-};
-
 static struct page *
 mlx5vf_get_migration_page(struct mlx5_vf_migration_file *migf,
 			  unsigned long offset)
@@ -444,7 +429,7 @@ mlx5vf_pci_step_device_state_locked(struct mlx5vf_pci_core_device *mvdev,
  * This function is called in all state_mutex unlock cases to
  * handle a 'deferred_reset' if exists.
  */
-static void mlx5vf_state_mutex_unlock(struct mlx5vf_pci_core_device *mvdev)
+void mlx5vf_state_mutex_unlock(struct mlx5vf_pci_core_device *mvdev)
 {
 again:
 	spin_lock(&mvdev->reset_lock);
@@ -596,24 +581,7 @@ static int mlx5vf_pci_probe(struct pci_dev *pdev,
 	if (!mvdev)
 		return -ENOMEM;
 	vfio_pci_core_init_device(&mvdev->core_device, pdev, &mlx5vf_pci_ops);
-
-	if (pdev->is_virtfn) {
-		struct mlx5_core_dev *mdev =
-			mlx5_vf_get_core_dev(pdev);
-
-		if (mdev) {
-			if (MLX5_CAP_GEN(mdev, migration)) {
-				mvdev->migrate_cap = 1;
-				mvdev->core_device.vdev.migration_flags =
-					VFIO_MIGRATION_STOP_COPY |
-					VFIO_MIGRATION_P2P;
-				mutex_init(&mvdev->state_mutex);
-				spin_lock_init(&mvdev->reset_lock);
-			}
-			mlx5_vf_put_core_dev(mdev);
-		}
-	}
-
+	mlx5vf_cmd_set_migratable(mvdev);
 	ret = vfio_pci_core_register_device(&mvdev->core_device);
 	if (ret)
 		goto out_free;
@@ -622,6 +590,8 @@ static int mlx5vf_pci_probe(struct pci_dev *pdev,
 	return 0;
 
 out_free:
+	if (mvdev->migrate_cap)
+		mlx5vf_cmd_remove_migratable(mvdev);
 	vfio_pci_core_uninit_device(&mvdev->core_device);
 	kfree(mvdev);
 	return ret;
@@ -632,6 +602,8 @@ static void mlx5vf_pci_remove(struct pci_dev *pdev)
 	struct mlx5vf_pci_core_device *mvdev = dev_get_drvdata(&pdev->dev);
 
 	vfio_pci_core_unregister_device(&mvdev->core_device);
+	if (mvdev->migrate_cap)
+		mlx5vf_cmd_remove_migratable(mvdev);
 	vfio_pci_core_uninit_device(&mvdev->core_device);
 	kfree(mvdev);
 }
-- 
2.18.1

[PATCH V1 mlx5-next 3/4] vfio/mlx5: Refactor to enable VFs migration in parallel

[Yishai Hadas]

Refactor to enable different VFs to run their commands over the PF
command interface in parallel and to not block one each other.

This is done by not using the global PF lock that was used before but
relying on the VF attach/detach mechanism to sync.

Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/vfio/pci/mlx5/cmd.c  | 103 +++++++++++++++--------------------
 drivers/vfio/pci/mlx5/cmd.h  |  11 ++--
 drivers/vfio/pci/mlx5/main.c |  44 ++++-----------
 3 files changed, 59 insertions(+), 99 deletions(-)

diff --git a/drivers/vfio/pci/mlx5/cmd.c b/drivers/vfio/pci/mlx5/cmd.c
index 5031978ae63a..9a6e3d3e0d44 100644
--- a/drivers/vfio/pci/mlx5/cmd.c
+++ b/drivers/vfio/pci/mlx5/cmd.c
@@ -5,70 +5,65 @@
 
 #include "cmd.h"
 
-int mlx5vf_cmd_suspend_vhca(struct pci_dev *pdev, u16 vhca_id, u16 op_mod)
+static int mlx5vf_cmd_get_vhca_id(struct mlx5_core_dev *mdev, u16 function_id,
+				  u16 *vhca_id);
+
+int mlx5vf_cmd_suspend_vhca(struct mlx5vf_pci_core_device *mvdev, u16 op_mod)
 {
-	struct mlx5_core_dev *mdev = mlx5_vf_get_core_dev(pdev);
 	u32 out[MLX5_ST_SZ_DW(suspend_vhca_out)] = {};
 	u32 in[MLX5_ST_SZ_DW(suspend_vhca_in)] = {};
-	int ret;
 
-	if (!mdev)
+	lockdep_assert_held(&mvdev->state_mutex);
+	if (mvdev->mdev_detach)
 		return -ENOTCONN;
 
 	MLX5_SET(suspend_vhca_in, in, opcode, MLX5_CMD_OP_SUSPEND_VHCA);
-	MLX5_SET(suspend_vhca_in, in, vhca_id, vhca_id);
+	MLX5_SET(suspend_vhca_in, in, vhca_id, mvdev->vhca_id);
 	MLX5_SET(suspend_vhca_in, in, op_mod, op_mod);
 
-	ret = mlx5_cmd_exec_inout(mdev, suspend_vhca, in, out);
-	mlx5_vf_put_core_dev(mdev);
-	return ret;
+	return mlx5_cmd_exec_inout(mvdev->mdev, suspend_vhca, in, out);
 }
 
-int mlx5vf_cmd_resume_vhca(struct pci_dev *pdev, u16 vhca_id, u16 op_mod)
+int mlx5vf_cmd_resume_vhca(struct mlx5vf_pci_core_device *mvdev, u16 op_mod)
 {
-	struct mlx5_core_dev *mdev = mlx5_vf_get_core_dev(pdev);
 	u32 out[MLX5_ST_SZ_DW(resume_vhca_out)] = {};
 	u32 in[MLX5_ST_SZ_DW(resume_vhca_in)] = {};
-	int ret;
 
-	if (!mdev)
+	lockdep_assert_held(&mvdev->state_mutex);
+	if (mvdev->mdev_detach)
 		return -ENOTCONN;
 
 	MLX5_SET(resume_vhca_in, in, opcode, MLX5_CMD_OP_RESUME_VHCA);
-	MLX5_SET(resume_vhca_in, in, vhca_id, vhca_id);
+	MLX5_SET(resume_vhca_in, in, vhca_id, mvdev->vhca_id);
 	MLX5_SET(resume_vhca_in, in, op_mod, op_mod);
 
-	ret = mlx5_cmd_exec_inout(mdev, resume_vhca, in, out);
-	mlx5_vf_put_core_dev(mdev);
-	return ret;
+	return mlx5_cmd_exec_inout(mvdev->mdev, resume_vhca, in, out);
 }
 
-int mlx5vf_cmd_query_vhca_migration_state(struct pci_dev *pdev, u16 vhca_id,
+int mlx5vf_cmd_query_vhca_migration_state(struct mlx5vf_pci_core_device *mvdev,
 					  size_t *state_size)
 {
-	struct mlx5_core_dev *mdev = mlx5_vf_get_core_dev(pdev);
 	u32 out[MLX5_ST_SZ_DW(query_vhca_migration_state_out)] = {};
 	u32 in[MLX5_ST_SZ_DW(query_vhca_migration_state_in)] = {};
 	int ret;
 
-	if (!mdev)
+	lockdep_assert_held(&mvdev->state_mutex);
+	if (mvdev->mdev_detach)
 		return -ENOTCONN;
 
 	MLX5_SET(query_vhca_migration_state_in, in, opcode,
 		 MLX5_CMD_OP_QUERY_VHCA_MIGRATION_STATE);
-	MLX5_SET(query_vhca_migration_state_in, in, vhca_id, vhca_id);
+	MLX5_SET(query_vhca_migration_state_in, in, vhca_id, mvdev->vhca_id);
 	MLX5_SET(query_vhca_migration_state_in, in, op_mod, 0);
 
-	ret = mlx5_cmd_exec_inout(mdev, query_vhca_migration_state, in, out);
+	ret = mlx5_cmd_exec_inout(mvdev->mdev, query_vhca_migration_state, in,
+				  out);
 	if (ret)
-		goto end;
+		return ret;
 
 	*state_size = MLX5_GET(query_vhca_migration_state_out, out,
 			       required_umem_size);
-
-end:
-	mlx5_vf_put_core_dev(mdev);
-	return ret;
+	return 0;
 }
 
 static int mlx5fv_vf_event(struct notifier_block *nb,
@@ -117,6 +112,10 @@ void mlx5vf_cmd_set_migratable(struct mlx5vf_pci_core_device *mvdev)
 	if (mvdev->vf_id < 0)
 		goto end;
 
+	if (mlx5vf_cmd_get_vhca_id(mvdev->mdev, mvdev->vf_id + 1,
+				   &mvdev->vhca_id))
+		goto end;
+
 	mutex_init(&mvdev->state_mutex);
 	spin_lock_init(&mvdev->reset_lock);
 	mvdev->nb.notifier_call = mlx5fv_vf_event;
@@ -134,23 +133,18 @@ void mlx5vf_cmd_set_migratable(struct mlx5vf_pci_core_device *mvdev)
 	mlx5_vf_put_core_dev(mvdev->mdev);
 }
 
-int mlx5vf_cmd_get_vhca_id(struct pci_dev *pdev, u16 function_id, u16 *vhca_id)
+static int mlx5vf_cmd_get_vhca_id(struct mlx5_core_dev *mdev, u16 function_id,
+				  u16 *vhca_id)
 {
-	struct mlx5_core_dev *mdev = mlx5_vf_get_core_dev(pdev);
 	u32 in[MLX5_ST_SZ_DW(query_hca_cap_in)] = {};
 	int out_size;
 	void *out;
 	int ret;
 
-	if (!mdev)
-		return -ENOTCONN;
-
 	out_size = MLX5_ST_SZ_BYTES(query_hca_cap_out);
 	out = kzalloc(out_size, GFP_KERNEL);
-	if (!out) {
-		ret = -ENOMEM;
-		goto end;
-	}
+	if (!out)
+		return -ENOMEM;
 
 	MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
 	MLX5_SET(query_hca_cap_in, in, other_function, 1);
@@ -168,8 +162,6 @@ int mlx5vf_cmd_get_vhca_id(struct pci_dev *pdev, u16 function_id, u16 *vhca_id)
 
 err_exec:
 	kfree(out);
-end:
-	mlx5_vf_put_core_dev(mdev);
 	return ret;
 }
 
@@ -214,21 +206,23 @@ static int _create_state_mkey(struct mlx5_core_dev *mdev, u32 pdn,
 	return err;
 }
 
-int mlx5vf_cmd_save_vhca_state(struct pci_dev *pdev, u16 vhca_id,
+int mlx5vf_cmd_save_vhca_state(struct mlx5vf_pci_core_device *mvdev,
 			       struct mlx5_vf_migration_file *migf)
 {
-	struct mlx5_core_dev *mdev = mlx5_vf_get_core_dev(pdev);
 	u32 out[MLX5_ST_SZ_DW(save_vhca_state_out)] = {};
 	u32 in[MLX5_ST_SZ_DW(save_vhca_state_in)] = {};
+	struct mlx5_core_dev *mdev;
 	u32 pdn, mkey;
 	int err;
 
-	if (!mdev)
+	lockdep_assert_held(&mvdev->state_mutex);
+	if (mvdev->mdev_detach)
 		return -ENOTCONN;
 
+	mdev = mvdev->mdev;
 	err = mlx5_core_alloc_pd(mdev, &pdn);
 	if (err)
-		goto end;
+		return err;
 
 	err = dma_map_sgtable(mdev->device, &migf->table.sgt, DMA_FROM_DEVICE,
 			      0);
@@ -242,7 +236,7 @@ int mlx5vf_cmd_save_vhca_state(struct pci_dev *pdev, u16 vhca_id,
 	MLX5_SET(save_vhca_state_in, in, opcode,
 		 MLX5_CMD_OP_SAVE_VHCA_STATE);
 	MLX5_SET(save_vhca_state_in, in, op_mod, 0);
-	MLX5_SET(save_vhca_state_in, in, vhca_id, vhca_id);
+	MLX5_SET(save_vhca_state_in, in, vhca_id, mvdev->vhca_id);
 	MLX5_SET(save_vhca_state_in, in, mkey, mkey);
 	MLX5_SET(save_vhca_state_in, in, size, migf->total_length);
 
@@ -250,37 +244,28 @@ int mlx5vf_cmd_save_vhca_state(struct pci_dev *pdev, u16 vhca_id,
 	if (err)
 		goto err_exec;
 
-	migf->total_length =
-		MLX5_GET(save_vhca_state_out, out, actual_image_size);
-
-	mlx5_core_destroy_mkey(mdev, mkey);
-	mlx5_core_dealloc_pd(mdev, pdn);
-	dma_unmap_sgtable(mdev->device, &migf->table.sgt, DMA_FROM_DEVICE, 0);
-	mlx5_vf_put_core_dev(mdev);
-
-	return 0;
-
+	migf->total_length = MLX5_GET(save_vhca_state_out, out,
+				      actual_image_size);
 err_exec:
 	mlx5_core_destroy_mkey(mdev, mkey);
 err_create_mkey:
 	dma_unmap_sgtable(mdev->device, &migf->table.sgt, DMA_FROM_DEVICE, 0);
 err_dma_map:
 	mlx5_core_dealloc_pd(mdev, pdn);
-end:
-	mlx5_vf_put_core_dev(mdev);
 	return err;
 }
 
-int mlx5vf_cmd_load_vhca_state(struct pci_dev *pdev, u16 vhca_id,
+int mlx5vf_cmd_load_vhca_state(struct mlx5vf_pci_core_device *mvdev,
 			       struct mlx5_vf_migration_file *migf)
 {
-	struct mlx5_core_dev *mdev = mlx5_vf_get_core_dev(pdev);
+	struct mlx5_core_dev *mdev;
 	u32 out[MLX5_ST_SZ_DW(save_vhca_state_out)] = {};
 	u32 in[MLX5_ST_SZ_DW(save_vhca_state_in)] = {};
 	u32 pdn, mkey;
 	int err;
 
-	if (!mdev)
+	lockdep_assert_held(&mvdev->state_mutex);
+	if (mvdev->mdev_detach)
 		return -ENOTCONN;
 
 	mutex_lock(&migf->lock);
@@ -289,6 +274,7 @@ int mlx5vf_cmd_load_vhca_state(struct pci_dev *pdev, u16 vhca_id,
 		goto end;
 	}
 
+	mdev = mvdev->mdev;
 	err = mlx5_core_alloc_pd(mdev, &pdn);
 	if (err)
 		goto end;
@@ -304,7 +290,7 @@ int mlx5vf_cmd_load_vhca_state(struct pci_dev *pdev, u16 vhca_id,
 	MLX5_SET(load_vhca_state_in, in, opcode,
 		 MLX5_CMD_OP_LOAD_VHCA_STATE);
 	MLX5_SET(load_vhca_state_in, in, op_mod, 0);
-	MLX5_SET(load_vhca_state_in, in, vhca_id, vhca_id);
+	MLX5_SET(load_vhca_state_in, in, vhca_id, mvdev->vhca_id);
 	MLX5_SET(load_vhca_state_in, in, mkey, mkey);
 	MLX5_SET(load_vhca_state_in, in, size, migf->total_length);
 
@@ -316,7 +302,6 @@ int mlx5vf_cmd_load_vhca_state(struct pci_dev *pdev, u16 vhca_id,
 err_reg:
 	mlx5_core_dealloc_pd(mdev, pdn);
 end:
-	mlx5_vf_put_core_dev(mdev);
 	mutex_unlock(&migf->lock);
 	return err;
 }
diff --git a/drivers/vfio/pci/mlx5/cmd.h b/drivers/vfio/pci/mlx5/cmd.h
index 340a06b98007..2a20b7435393 100644
--- a/drivers/vfio/pci/mlx5/cmd.h
+++ b/drivers/vfio/pci/mlx5/cmd.h
@@ -43,16 +43,15 @@ struct mlx5vf_pci_core_device {
 	u8 mdev_detach:1;
 };
 
-int mlx5vf_cmd_suspend_vhca(struct pci_dev *pdev, u16 vhca_id, u16 op_mod);
-int mlx5vf_cmd_resume_vhca(struct pci_dev *pdev, u16 vhca_id, u16 op_mod);
-int mlx5vf_cmd_query_vhca_migration_state(struct pci_dev *pdev, u16 vhca_id,
+int mlx5vf_cmd_suspend_vhca(struct mlx5vf_pci_core_device *mvdev, u16 op_mod);
+int mlx5vf_cmd_resume_vhca(struct mlx5vf_pci_core_device *mvdev, u16 op_mod);
+int mlx5vf_cmd_query_vhca_migration_state(struct mlx5vf_pci_core_device *mvdev,
 					  size_t *state_size);
-int mlx5vf_cmd_get_vhca_id(struct pci_dev *pdev, u16 function_id, u16 *vhca_id);
 void mlx5vf_cmd_set_migratable(struct mlx5vf_pci_core_device *mvdev);
 void mlx5vf_cmd_remove_migratable(struct mlx5vf_pci_core_device *mvdev);
-int mlx5vf_cmd_save_vhca_state(struct pci_dev *pdev, u16 vhca_id,
+int mlx5vf_cmd_save_vhca_state(struct mlx5vf_pci_core_device *mvdev,
 			       struct mlx5_vf_migration_file *migf);
-int mlx5vf_cmd_load_vhca_state(struct pci_dev *pdev, u16 vhca_id,
+int mlx5vf_cmd_load_vhca_state(struct mlx5vf_pci_core_device *mvdev,
 			       struct mlx5_vf_migration_file *migf);
 void mlx5vf_state_mutex_unlock(struct mlx5vf_pci_core_device *mvdev);
 #endif /* MLX5_VFIO_CMD_H */
diff --git a/drivers/vfio/pci/mlx5/main.c b/drivers/vfio/pci/mlx5/main.c
index 9716c87e31f9..5bda6c0e194c 100644
--- a/drivers/vfio/pci/mlx5/main.c
+++ b/drivers/vfio/pci/mlx5/main.c
@@ -208,8 +208,8 @@ mlx5vf_pci_save_device_data(struct mlx5vf_pci_core_device *mvdev)
 	stream_open(migf->filp->f_inode, migf->filp);
 	mutex_init(&migf->lock);
 
-	ret = mlx5vf_cmd_query_vhca_migration_state(
-		mvdev->core_device.pdev, mvdev->vhca_id, &migf->total_length);
+	ret = mlx5vf_cmd_query_vhca_migration_state(mvdev,
+						    &migf->total_length);
 	if (ret)
 		goto out_free;
 
@@ -218,8 +218,7 @@ mlx5vf_pci_save_device_data(struct mlx5vf_pci_core_device *mvdev)
 	if (ret)
 		goto out_free;
 
-	ret = mlx5vf_cmd_save_vhca_state(mvdev->core_device.pdev,
-					 mvdev->vhca_id, migf);
+	ret = mlx5vf_cmd_save_vhca_state(mvdev, migf);
 	if (ret)
 		goto out_free;
 	return migf;
@@ -346,8 +345,7 @@ mlx5vf_pci_step_device_state_locked(struct mlx5vf_pci_core_device *mvdev,
 	int ret;
 
 	if (cur == VFIO_DEVICE_STATE_RUNNING_P2P && new == VFIO_DEVICE_STATE_STOP) {
-		ret = mlx5vf_cmd_suspend_vhca(
-			mvdev->core_device.pdev, mvdev->vhca_id,
+		ret = mlx5vf_cmd_suspend_vhca(mvdev,
 			MLX5_SUSPEND_VHCA_IN_OP_MOD_SUSPEND_RESPONDER);
 		if (ret)
 			return ERR_PTR(ret);
@@ -355,8 +353,7 @@ mlx5vf_pci_step_device_state_locked(struct mlx5vf_pci_core_device *mvdev,
 	}
 
 	if (cur == VFIO_DEVICE_STATE_STOP && new == VFIO_DEVICE_STATE_RUNNING_P2P) {
-		ret = mlx5vf_cmd_resume_vhca(
-			mvdev->core_device.pdev, mvdev->vhca_id,
+		ret = mlx5vf_cmd_resume_vhca(mvdev,
 			MLX5_RESUME_VHCA_IN_OP_MOD_RESUME_RESPONDER);
 		if (ret)
 			return ERR_PTR(ret);
@@ -364,8 +361,7 @@ mlx5vf_pci_step_device_state_locked(struct mlx5vf_pci_core_device *mvdev,
 	}
 
 	if (cur == VFIO_DEVICE_STATE_RUNNING && new == VFIO_DEVICE_STATE_RUNNING_P2P) {
-		ret = mlx5vf_cmd_suspend_vhca(
-			mvdev->core_device.pdev, mvdev->vhca_id,
+		ret = mlx5vf_cmd_suspend_vhca(mvdev,
 			MLX5_SUSPEND_VHCA_IN_OP_MOD_SUSPEND_INITIATOR);
 		if (ret)
 			return ERR_PTR(ret);
@@ -373,8 +369,7 @@ mlx5vf_pci_step_device_state_locked(struct mlx5vf_pci_core_device *mvdev,
 	}
 
 	if (cur == VFIO_DEVICE_STATE_RUNNING_P2P && new == VFIO_DEVICE_STATE_RUNNING) {
-		ret = mlx5vf_cmd_resume_vhca(
-			mvdev->core_device.pdev, mvdev->vhca_id,
+		ret = mlx5vf_cmd_resume_vhca(mvdev,
 			MLX5_RESUME_VHCA_IN_OP_MOD_RESUME_INITIATOR);
 		if (ret)
 			return ERR_PTR(ret);
@@ -409,8 +404,7 @@ mlx5vf_pci_step_device_state_locked(struct mlx5vf_pci_core_device *mvdev,
 	}
 
 	if (cur == VFIO_DEVICE_STATE_RESUMING && new == VFIO_DEVICE_STATE_STOP) {
-		ret = mlx5vf_cmd_load_vhca_state(mvdev->core_device.pdev,
-						 mvdev->vhca_id,
+		ret = mlx5vf_cmd_load_vhca_state(mvdev,
 						 mvdev->resuming_migf);
 		if (ret)
 			return ERR_PTR(ret);
@@ -517,34 +511,16 @@ static int mlx5vf_pci_open_device(struct vfio_device *core_vdev)
 	struct mlx5vf_pci_core_device *mvdev = container_of(
 		core_vdev, struct mlx5vf_pci_core_device, core_device.vdev);
 	struct vfio_pci_core_device *vdev = &mvdev->core_device;
-	int vf_id;
 	int ret;
 
 	ret = vfio_pci_core_enable(vdev);
 	if (ret)
 		return ret;
 
-	if (!mvdev->migrate_cap) {
-		vfio_pci_core_finish_enable(vdev);
-		return 0;
-	}
-
-	vf_id = pci_iov_vf_id(vdev->pdev);
-	if (vf_id < 0) {
-		ret = vf_id;
-		goto out_disable;
-	}
-
-	ret = mlx5vf_cmd_get_vhca_id(vdev->pdev, vf_id + 1, &mvdev->vhca_id);
-	if (ret)
-		goto out_disable;
-
-	mvdev->mig_state = VFIO_DEVICE_STATE_RUNNING;
+	if (mvdev->migrate_cap)
+		mvdev->mig_state = VFIO_DEVICE_STATE_RUNNING;
 	vfio_pci_core_finish_enable(vdev);
 	return 0;
-out_disable:
-	vfio_pci_core_disable(vdev);
-	return ret;
 }
 
 static void mlx5vf_pci_close_device(struct vfio_device *core_vdev)
-- 
2.18.1

[PATCH V1 mlx5-next 4/4] vfio/mlx5: Run the SAVE state command in an async mode

[Yishai Hadas]

Use the PF asynchronous command mode for the SAVE state command.

This enables returning earlier to user space upon issuing successfully
the command and improve latency by let things run in parallel.

Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
---
 drivers/vfio/pci/mlx5/cmd.c  | 72 ++++++++++++++++++++++++++++++++++--
 drivers/vfio/pci/mlx5/cmd.h  | 17 +++++++++
 drivers/vfio/pci/mlx5/main.c | 56 +++++++++++++++++++++++++---
 3 files changed, 136 insertions(+), 9 deletions(-)

diff --git a/drivers/vfio/pci/mlx5/cmd.c b/drivers/vfio/pci/mlx5/cmd.c
index 9a6e3d3e0d44..0538e44e7eac 100644
--- a/drivers/vfio/pci/mlx5/cmd.c
+++ b/drivers/vfio/pci/mlx5/cmd.c
@@ -78,6 +78,7 @@ static int mlx5fv_vf_event(struct notifier_block *nb,
 		mvdev->mdev_detach = false;
 		break;
 	case MLX5_PF_NOTIFY_DISABLE_VF:
+		mlx5vf_disable_fds(mvdev);
 		mvdev->mdev_detach = true;
 		break;
 	default:
@@ -206,11 +207,56 @@ static int _create_state_mkey(struct mlx5_core_dev *mdev, u32 pdn,
 	return err;
 }
 
+void mlx5vf_mig_file_cleanup_cb(struct work_struct *_work)
+{
+	struct mlx5vf_async_data *async_data = container_of(_work,
+		struct mlx5vf_async_data, work);
+	struct mlx5_vf_migration_file *migf = container_of(async_data,
+		struct mlx5_vf_migration_file, async_data);
+	struct mlx5_core_dev *mdev = migf->mvdev->mdev;
+
+	mutex_lock(&migf->lock);
+	if (async_data->status) {
+		migf->is_err = true;
+		wake_up_interruptible(&migf->poll_wait);
+	}
+	mutex_unlock(&migf->lock);
+
+	mlx5_core_destroy_mkey(mdev, async_data->mkey);
+	dma_unmap_sgtable(mdev->device, &migf->table.sgt, DMA_FROM_DEVICE, 0);
+	mlx5_core_dealloc_pd(mdev, async_data->pdn);
+	kvfree(async_data->out);
+	fput(migf->filp);
+}
+
+static void mlx5vf_save_callback(int status, struct mlx5_async_work *context)
+{
+	struct mlx5vf_async_data *async_data = container_of(context,
+			struct mlx5vf_async_data, cb_work);
+	struct mlx5_vf_migration_file *migf = container_of(async_data,
+			struct mlx5_vf_migration_file, async_data);
+
+	if (!status) {
+		WRITE_ONCE(migf->total_length,
+			   MLX5_GET(save_vhca_state_out, async_data->out,
+				    actual_image_size));
+		wake_up_interruptible(&migf->poll_wait);
+	}
+
+	/*
+	 * The error and the cleanup flows can't run from an
+	 * interrupt context
+	 */
+	async_data->status = status;
+	queue_work(migf->mvdev->cb_wq, &async_data->work);
+}
+
 int mlx5vf_cmd_save_vhca_state(struct mlx5vf_pci_core_device *mvdev,
 			       struct mlx5_vf_migration_file *migf)
 {
-	u32 out[MLX5_ST_SZ_DW(save_vhca_state_out)] = {};
+	u32 out_size = MLX5_ST_SZ_BYTES(save_vhca_state_out);
 	u32 in[MLX5_ST_SZ_DW(save_vhca_state_in)] = {};
+	struct mlx5vf_async_data *async_data;
 	struct mlx5_core_dev *mdev;
 	u32 pdn, mkey;
 	int err;
@@ -240,13 +286,31 @@ int mlx5vf_cmd_save_vhca_state(struct mlx5vf_pci_core_device *mvdev,
 	MLX5_SET(save_vhca_state_in, in, mkey, mkey);
 	MLX5_SET(save_vhca_state_in, in, size, migf->total_length);
 
-	err = mlx5_cmd_exec_inout(mdev, save_vhca_state, in, out);
+	async_data = &migf->async_data;
+	async_data->out = kvzalloc(out_size, GFP_KERNEL);
+	if (!async_data->out) {
+		err = -ENOMEM;
+		goto err_out;
+	}
+
+	/* no data exists till the callback comes back */
+	migf->total_length = 0;
+	get_file(migf->filp);
+	async_data->mkey = mkey;
+	async_data->pdn = pdn;
+	err = mlx5_cmd_exec_cb(&migf->async_ctx, in, sizeof(in),
+			       async_data->out,
+			       out_size, mlx5vf_save_callback,
+			       &async_data->cb_work);
 	if (err)
 		goto err_exec;
 
-	migf->total_length = MLX5_GET(save_vhca_state_out, out,
-				      actual_image_size);
+	return 0;
+
 err_exec:
+	fput(migf->filp);
+	kvfree(async_data->out);
+err_out:
 	mlx5_core_destroy_mkey(mdev, mkey);
 err_create_mkey:
 	dma_unmap_sgtable(mdev->device, &migf->table.sgt, DMA_FROM_DEVICE, 0);
diff --git a/drivers/vfio/pci/mlx5/cmd.h b/drivers/vfio/pci/mlx5/cmd.h
index 2a20b7435393..d053d314b745 100644
--- a/drivers/vfio/pci/mlx5/cmd.h
+++ b/drivers/vfio/pci/mlx5/cmd.h
@@ -10,10 +10,20 @@
 #include <linux/vfio_pci_core.h>
 #include <linux/mlx5/driver.h>
 
+struct mlx5vf_async_data {
+	struct mlx5_async_work cb_work;
+	struct work_struct work;
+	int status;
+	u32 pdn;
+	u32 mkey;
+	void *out;
+};
+
 struct mlx5_vf_migration_file {
 	struct file *filp;
 	struct mutex lock;
 	bool disabled;
+	u8 is_err:1;
 
 	struct sg_append_table table;
 	size_t total_length;
@@ -23,6 +33,10 @@ struct mlx5_vf_migration_file {
 	struct scatterlist *last_offset_sg;
 	unsigned int sg_last_entry;
 	unsigned long last_offset;
+	struct mlx5vf_pci_core_device *mvdev;
+	wait_queue_head_t poll_wait;
+	struct mlx5_async_ctx async_ctx;
+	struct mlx5vf_async_data async_data;
 };
 
 struct mlx5vf_pci_core_device {
@@ -38,6 +52,7 @@ struct mlx5vf_pci_core_device {
 	spinlock_t reset_lock;
 	struct mlx5_vf_migration_file *resuming_migf;
 	struct mlx5_vf_migration_file *saving_migf;
+	struct workqueue_struct *cb_wq;
 	struct notifier_block nb;
 	struct mlx5_core_dev *mdev;
 	u8 mdev_detach:1;
@@ -54,4 +69,6 @@ int mlx5vf_cmd_save_vhca_state(struct mlx5vf_pci_core_device *mvdev,
 int mlx5vf_cmd_load_vhca_state(struct mlx5vf_pci_core_device *mvdev,
 			       struct mlx5_vf_migration_file *migf);
 void mlx5vf_state_mutex_unlock(struct mlx5vf_pci_core_device *mvdev);
+void mlx5vf_disable_fds(struct mlx5vf_pci_core_device *mvdev);
+void mlx5vf_mig_file_cleanup_cb(struct work_struct *_work);
 #endif /* MLX5_VFIO_CMD_H */
diff --git a/drivers/vfio/pci/mlx5/main.c b/drivers/vfio/pci/mlx5/main.c
index 5bda6c0e194c..625ed5743f26 100644
--- a/drivers/vfio/pci/mlx5/main.c
+++ b/drivers/vfio/pci/mlx5/main.c
@@ -134,12 +134,22 @@ static ssize_t mlx5vf_save_read(struct file *filp, char __user *buf, size_t len,
 		return -ESPIPE;
 	pos = &filp->f_pos;
 
+	if (!(filp->f_flags & O_NONBLOCK)) {
+		if (wait_event_interruptible(migf->poll_wait,
+			     READ_ONCE(migf->total_length) || migf->is_err))
+			return -ERESTARTSYS;
+	}
+
 	mutex_lock(&migf->lock);
+	if ((filp->f_flags & O_NONBLOCK) && !READ_ONCE(migf->total_length)) {
+		done = -EAGAIN;
+		goto out_unlock;
+	}
 	if (*pos > migf->total_length) {
 		done = -EINVAL;
 		goto out_unlock;
 	}
-	if (migf->disabled) {
+	if (migf->disabled || migf->is_err) {
 		done = -ENODEV;
 		goto out_unlock;
 	}
@@ -179,9 +189,28 @@ static ssize_t mlx5vf_save_read(struct file *filp, char __user *buf, size_t len,
 	return done;
 }
 
+static __poll_t mlx5vf_save_poll(struct file *filp,
+				 struct poll_table_struct *wait)
+{
+	struct mlx5_vf_migration_file *migf = filp->private_data;
+	__poll_t pollflags = 0;
+
+	poll_wait(filp, &migf->poll_wait, wait);
+
+	mutex_lock(&migf->lock);
+	if (migf->disabled || migf->is_err)
+		pollflags = EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
+	else if (READ_ONCE(migf->total_length))
+		pollflags = EPOLLIN | EPOLLRDNORM;
+	mutex_unlock(&migf->lock);
+
+	return pollflags;
+}
+
 static const struct file_operations mlx5vf_save_fops = {
 	.owner = THIS_MODULE,
 	.read = mlx5vf_save_read,
+	.poll = mlx5vf_save_poll,
 	.release = mlx5vf_release_file,
 	.llseek = no_llseek,
 };
@@ -207,7 +236,9 @@ mlx5vf_pci_save_device_data(struct mlx5vf_pci_core_device *mvdev)
 
 	stream_open(migf->filp->f_inode, migf->filp);
 	mutex_init(&migf->lock);
-
+	init_waitqueue_head(&migf->poll_wait);
+	mlx5_cmd_init_async_ctx(mvdev->mdev, &migf->async_ctx);
+	INIT_WORK(&migf->async_data.work, mlx5vf_mig_file_cleanup_cb);
 	ret = mlx5vf_cmd_query_vhca_migration_state(mvdev,
 						    &migf->total_length);
 	if (ret)
@@ -218,6 +249,7 @@ mlx5vf_pci_save_device_data(struct mlx5vf_pci_core_device *mvdev)
 	if (ret)
 		goto out_free;
 
+	migf->mvdev = mvdev;
 	ret = mlx5vf_cmd_save_vhca_state(mvdev, migf);
 	if (ret)
 		goto out_free;
@@ -323,7 +355,7 @@ mlx5vf_pci_resume_device_data(struct mlx5vf_pci_core_device *mvdev)
 	return migf;
 }
 
-static void mlx5vf_disable_fds(struct mlx5vf_pci_core_device *mvdev)
+void mlx5vf_disable_fds(struct mlx5vf_pci_core_device *mvdev)
 {
 	if (mvdev->resuming_migf) {
 		mlx5vf_disable_fd(mvdev->resuming_migf);
@@ -331,6 +363,8 @@ static void mlx5vf_disable_fds(struct mlx5vf_pci_core_device *mvdev)
 		mvdev->resuming_migf = NULL;
 	}
 	if (mvdev->saving_migf) {
+		mlx5_cmd_cleanup_async_ctx(&mvdev->saving_migf->async_ctx);
+		cancel_work_sync(&mvdev->saving_migf->async_data.work);
 		mlx5vf_disable_fd(mvdev->saving_migf);
 		fput(mvdev->saving_migf->filp);
 		mvdev->saving_migf = NULL;
@@ -558,6 +592,13 @@ static int mlx5vf_pci_probe(struct pci_dev *pdev,
 		return -ENOMEM;
 	vfio_pci_core_init_device(&mvdev->core_device, pdev, &mlx5vf_pci_ops);
 	mlx5vf_cmd_set_migratable(mvdev);
+	if (mvdev->migrate_cap) {
+		mvdev->cb_wq = alloc_ordered_workqueue("mlx5vf_wq", 0);
+		if (!mvdev->cb_wq) {
+			ret = -ENOMEM;
+			goto out_free;
+		}
+	}
 	ret = vfio_pci_core_register_device(&mvdev->core_device);
 	if (ret)
 		goto out_free;
@@ -566,8 +607,11 @@ static int mlx5vf_pci_probe(struct pci_dev *pdev,
 	return 0;
 
 out_free:
-	if (mvdev->migrate_cap)
+	if (mvdev->migrate_cap) {
 		mlx5vf_cmd_remove_migratable(mvdev);
+		if (mvdev->cb_wq)
+			destroy_workqueue(mvdev->cb_wq);
+	}
 	vfio_pci_core_uninit_device(&mvdev->core_device);
 	kfree(mvdev);
 	return ret;
@@ -578,8 +622,10 @@ static void mlx5vf_pci_remove(struct pci_dev *pdev)
 	struct mlx5vf_pci_core_device *mvdev = dev_get_drvdata(&pdev->dev);
 
 	vfio_pci_core_unregister_device(&mvdev->core_device);
-	if (mvdev->migrate_cap)
+	if (mvdev->migrate_cap) {
 		mlx5vf_cmd_remove_migratable(mvdev);
+		destroy_workqueue(mvdev->cb_wq);
+	}
 	vfio_pci_core_uninit_device(&mvdev->core_device);
 	kfree(mvdev);
 }
-- 
2.18.1

Re: [PATCH V1 mlx5-next 2/4] vfio/mlx5: Manage the VF attach/detach callback from the PF

[Alex Williamson]

On Sun, 8 May 2022 16:10:51 +0300
Yishai Hadas <yishaih@nvidia.com> wrote:

> Manage the VF attach/detach callback from the PF.
> 
> This lets the driver to enable parallel VFs migration as will be
> introduced in the next patch.
> 
> As part of this, reorganize the VF is migratable code to be in a
> separate function and rename it to be set_migratable() to match its
> functionality.
> 
> Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
> Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> ---
>  drivers/vfio/pci/mlx5/cmd.c  | 63 ++++++++++++++++++++++++++++++++++++
>  drivers/vfio/pci/mlx5/cmd.h  | 22 +++++++++++++
>  drivers/vfio/pci/mlx5/main.c | 40 ++++-------------------
>  3 files changed, 91 insertions(+), 34 deletions(-)
> 
> diff --git a/drivers/vfio/pci/mlx5/cmd.c b/drivers/vfio/pci/mlx5/cmd.c
> index 5c9f9218cc1d..5031978ae63a 100644
> --- a/drivers/vfio/pci/mlx5/cmd.c
> +++ b/drivers/vfio/pci/mlx5/cmd.c
> @@ -71,6 +71,69 @@ int mlx5vf_cmd_query_vhca_migration_state(struct pci_dev *pdev, u16 vhca_id,
>  	return ret;
>  }
>  
> +static int mlx5fv_vf_event(struct notifier_block *nb,
> +			   unsigned long event, void *data)
> +{
> +	struct mlx5vf_pci_core_device *mvdev =
> +		container_of(nb, struct mlx5vf_pci_core_device, nb);
> +
> +	mutex_lock(&mvdev->state_mutex);
> +	switch (event) {
> +	case MLX5_PF_NOTIFY_ENABLE_VF:
> +		mvdev->mdev_detach = false;
> +		break;
> +	case MLX5_PF_NOTIFY_DISABLE_VF:
> +		mvdev->mdev_detach = true;
> +		break;
> +	default:
> +		break;
> +	}
> +	mlx5vf_state_mutex_unlock(mvdev);
> +	return 0;
> +}
> +
> +void mlx5vf_cmd_remove_migratable(struct mlx5vf_pci_core_device *mvdev)
> +{
> +	mlx5_sriov_blocking_notifier_unregister(mvdev->mdev, mvdev->vf_id,
> +						&mvdev->nb);
> +}
> +
> +void mlx5vf_cmd_set_migratable(struct mlx5vf_pci_core_device *mvdev)
> +{
> +	struct pci_dev *pdev = mvdev->core_device.pdev;
> +	int ret;
> +
> +	if (!pdev->is_virtfn)
> +		return;
> +
> +	mvdev->mdev = mlx5_vf_get_core_dev(pdev);
> +	if (!mvdev->mdev)
> +		return;
> +
> +	if (!MLX5_CAP_GEN(mvdev->mdev, migration))
> +		goto end;
> +
> +	mvdev->vf_id = pci_iov_vf_id(pdev);
> +	if (mvdev->vf_id < 0)
> +		goto end;
> +
> +	mutex_init(&mvdev->state_mutex);
> +	spin_lock_init(&mvdev->reset_lock);
> +	mvdev->nb.notifier_call = mlx5fv_vf_event;
> +	ret = mlx5_sriov_blocking_notifier_register(mvdev->mdev, mvdev->vf_id,
> +						    &mvdev->nb);
> +	if (ret)
> +		goto end;
> +
> +	mvdev->migrate_cap = 1;
> +	mvdev->core_device.vdev.migration_flags =
> +		VFIO_MIGRATION_STOP_COPY |
> +		VFIO_MIGRATION_P2P;
> +
> +end:
> +	mlx5_vf_put_core_dev(mvdev->mdev);
> +}
> +
>  int mlx5vf_cmd_get_vhca_id(struct pci_dev *pdev, u16 function_id, u16 *vhca_id)
>  {
>  	struct mlx5_core_dev *mdev = mlx5_vf_get_core_dev(pdev);
> diff --git a/drivers/vfio/pci/mlx5/cmd.h b/drivers/vfio/pci/mlx5/cmd.h
> index 1392a11a9cc0..340a06b98007 100644
> --- a/drivers/vfio/pci/mlx5/cmd.h
> +++ b/drivers/vfio/pci/mlx5/cmd.h
> @@ -7,6 +7,7 @@
>  #define MLX5_VFIO_CMD_H
>  
>  #include <linux/kernel.h>
> +#include <linux/vfio_pci_core.h>
>  #include <linux/mlx5/driver.h>
>  
>  struct mlx5_vf_migration_file {
> @@ -24,13 +25,34 @@ struct mlx5_vf_migration_file {
>  	unsigned long last_offset;
>  };
>  
> +struct mlx5vf_pci_core_device {
> +	struct vfio_pci_core_device core_device;
> +	int vf_id;
> +	u16 vhca_id;
> +	u8 migrate_cap:1;
> +	u8 deferred_reset:1;
> +	/* protect migration state */
> +	struct mutex state_mutex;
> +	enum vfio_device_mig_state mig_state;
> +	/* protect the reset_done flow */
> +	spinlock_t reset_lock;
> +	struct mlx5_vf_migration_file *resuming_migf;
> +	struct mlx5_vf_migration_file *saving_migf;
> +	struct notifier_block nb;
> +	struct mlx5_core_dev *mdev;
> +	u8 mdev_detach:1;


This should be packed with the other bit fields, there's plenty of
space there.


> +};
> +
>  int mlx5vf_cmd_suspend_vhca(struct pci_dev *pdev, u16 vhca_id, u16 op_mod);
>  int mlx5vf_cmd_resume_vhca(struct pci_dev *pdev, u16 vhca_id, u16 op_mod);
>  int mlx5vf_cmd_query_vhca_migration_state(struct pci_dev *pdev, u16 vhca_id,
>  					  size_t *state_size);
>  int mlx5vf_cmd_get_vhca_id(struct pci_dev *pdev, u16 function_id, u16 *vhca_id);
> +void mlx5vf_cmd_set_migratable(struct mlx5vf_pci_core_device *mvdev);
> +void mlx5vf_cmd_remove_migratable(struct mlx5vf_pci_core_device *mvdev);
>  int mlx5vf_cmd_save_vhca_state(struct pci_dev *pdev, u16 vhca_id,
>  			       struct mlx5_vf_migration_file *migf);
>  int mlx5vf_cmd_load_vhca_state(struct pci_dev *pdev, u16 vhca_id,
>  			       struct mlx5_vf_migration_file *migf);
> +void mlx5vf_state_mutex_unlock(struct mlx5vf_pci_core_device *mvdev);
>  #endif /* MLX5_VFIO_CMD_H */
> diff --git a/drivers/vfio/pci/mlx5/main.c b/drivers/vfio/pci/mlx5/main.c
> index bbec5d288fee..9716c87e31f9 100644
> --- a/drivers/vfio/pci/mlx5/main.c
> +++ b/drivers/vfio/pci/mlx5/main.c
> @@ -17,7 +17,6 @@
>  #include <linux/uaccess.h>
>  #include <linux/vfio.h>
>  #include <linux/sched/mm.h>
> -#include <linux/vfio_pci_core.h>
>  #include <linux/anon_inodes.h>
>  
>  #include "cmd.h"
> @@ -25,20 +24,6 @@
>  /* Arbitrary to prevent userspace from consuming endless memory */
>  #define MAX_MIGRATION_SIZE (512*1024*1024)
>  
> -struct mlx5vf_pci_core_device {
> -	struct vfio_pci_core_device core_device;
> -	u16 vhca_id;
> -	u8 migrate_cap:1;
> -	u8 deferred_reset:1;
> -	/* protect migration state */
> -	struct mutex state_mutex;
> -	enum vfio_device_mig_state mig_state;
> -	/* protect the reset_done flow */
> -	spinlock_t reset_lock;
> -	struct mlx5_vf_migration_file *resuming_migf;
> -	struct mlx5_vf_migration_file *saving_migf;
> -};
> -
>  static struct page *
>  mlx5vf_get_migration_page(struct mlx5_vf_migration_file *migf,
>  			  unsigned long offset)
> @@ -444,7 +429,7 @@ mlx5vf_pci_step_device_state_locked(struct mlx5vf_pci_core_device *mvdev,
>   * This function is called in all state_mutex unlock cases to
>   * handle a 'deferred_reset' if exists.
>   */
> -static void mlx5vf_state_mutex_unlock(struct mlx5vf_pci_core_device *mvdev)
> +void mlx5vf_state_mutex_unlock(struct mlx5vf_pci_core_device *mvdev)
>  {
>  again:
>  	spin_lock(&mvdev->reset_lock);
> @@ -596,24 +581,7 @@ static int mlx5vf_pci_probe(struct pci_dev *pdev,
>  	if (!mvdev)
>  		return -ENOMEM;
>  	vfio_pci_core_init_device(&mvdev->core_device, pdev, &mlx5vf_pci_ops);
> -
> -	if (pdev->is_virtfn) {
> -		struct mlx5_core_dev *mdev =
> -			mlx5_vf_get_core_dev(pdev);
> -
> -		if (mdev) {
> -			if (MLX5_CAP_GEN(mdev, migration)) {
> -				mvdev->migrate_cap = 1;
> -				mvdev->core_device.vdev.migration_flags =
> -					VFIO_MIGRATION_STOP_COPY |
> -					VFIO_MIGRATION_P2P;
> -				mutex_init(&mvdev->state_mutex);
> -				spin_lock_init(&mvdev->reset_lock);
> -			}
> -			mlx5_vf_put_core_dev(mdev);
> -		}
> -	}
> -
> +	mlx5vf_cmd_set_migratable(mvdev);
>  	ret = vfio_pci_core_register_device(&mvdev->core_device);
>  	if (ret)
>  		goto out_free;
> @@ -622,6 +590,8 @@ static int mlx5vf_pci_probe(struct pci_dev *pdev,
>  	return 0;
>  
>  out_free:
> +	if (mvdev->migrate_cap)
> +		mlx5vf_cmd_remove_migratable(mvdev);
>  	vfio_pci_core_uninit_device(&mvdev->core_device);
>  	kfree(mvdev);
>  	return ret;
> @@ -632,6 +602,8 @@ static void mlx5vf_pci_remove(struct pci_dev *pdev)
>  	struct mlx5vf_pci_core_device *mvdev = dev_get_drvdata(&pdev->dev);
>  
>  	vfio_pci_core_unregister_device(&mvdev->core_device);
> +	if (mvdev->migrate_cap)
> +		mlx5vf_cmd_remove_migratable(mvdev);
>  	vfio_pci_core_uninit_device(&mvdev->core_device);
>  	kfree(mvdev);
>  }


Personally, I'd push the test into the function, ie.

void mlx5vf_cmd_remove_migratable(struct mlx5vf_pci_core_device *mvdev)
{
	if (!mvdev->migrate_cap)
		return;

	...
}

But it's clearly functional this way as well.  Please do fix the bit
field packing though.  Thanks,

Alex

Re: [PATCH V1 mlx5-next 4/4] vfio/mlx5: Run the SAVE state command in an async mode

[Alex Williamson]

On Sun, 8 May 2022 16:10:53 +0300
Yishai Hadas <yishaih@nvidia.com> wrote:
> diff --git a/drivers/vfio/pci/mlx5/cmd.h b/drivers/vfio/pci/mlx5/cmd.h
> index 2a20b7435393..d053d314b745 100644
> --- a/drivers/vfio/pci/mlx5/cmd.h
> +++ b/drivers/vfio/pci/mlx5/cmd.h
> @@ -10,10 +10,20 @@
>  #include <linux/vfio_pci_core.h>
>  #include <linux/mlx5/driver.h>
>  
> +struct mlx5vf_async_data {
> +	struct mlx5_async_work cb_work;
> +	struct work_struct work;
> +	int status;
> +	u32 pdn;
> +	u32 mkey;
> +	void *out;
> +};
> +
>  struct mlx5_vf_migration_file {
>  	struct file *filp;
>  	struct mutex lock;
>  	bool disabled;
> +	u8 is_err:1;

Convert @disabled to bit field as well to pack these?

...
> @@ -558,6 +592,13 @@ static int mlx5vf_pci_probe(struct pci_dev *pdev,
>  		return -ENOMEM;
>  	vfio_pci_core_init_device(&mvdev->core_device, pdev, &mlx5vf_pci_ops);
>  	mlx5vf_cmd_set_migratable(mvdev);
> +	if (mvdev->migrate_cap) {
> +		mvdev->cb_wq = alloc_ordered_workqueue("mlx5vf_wq", 0);
> +		if (!mvdev->cb_wq) {
> +			ret = -ENOMEM;
> +			goto out_free;
> +		}
> +	}

Should this be rolled into mlx5vf_cmd_set_migratable(), updating the
function to return -errno?

>  	ret = vfio_pci_core_register_device(&mvdev->core_device);
>  	if (ret)
>  		goto out_free;
> @@ -566,8 +607,11 @@ static int mlx5vf_pci_probe(struct pci_dev *pdev,
>  	return 0;
>  
>  out_free:
> -	if (mvdev->migrate_cap)
> +	if (mvdev->migrate_cap) {
>  		mlx5vf_cmd_remove_migratable(mvdev);
> +		if (mvdev->cb_wq)
> +			destroy_workqueue(mvdev->cb_wq);
> +	}
>  	vfio_pci_core_uninit_device(&mvdev->core_device);
>  	kfree(mvdev);
>  	return ret;
> @@ -578,8 +622,10 @@ static void mlx5vf_pci_remove(struct pci_dev *pdev)
>  	struct mlx5vf_pci_core_device *mvdev = dev_get_drvdata(&pdev->dev);
>  
>  	vfio_pci_core_unregister_device(&mvdev->core_device);
> -	if (mvdev->migrate_cap)
> +	if (mvdev->migrate_cap) {
>  		mlx5vf_cmd_remove_migratable(mvdev);
> +		destroy_workqueue(mvdev->cb_wq);
> +	}
>  	vfio_pci_core_uninit_device(&mvdev->core_device);
>  	kfree(mvdev);
>  }

This looks like more evidence for expanding remove_migratable(),
rolling this in as well.  If this workqueue were setup in
set_migratable() we'd not need the special condition to test if cb_wq
is NULL while migrate_cap is set.  Thanks,

Alex

Re: [PATCH V1 mlx5-next 2/4] vfio/mlx5: Manage the VF attach/detach callback from the PF

[Yishai Hadas]

On 09/05/2022 20:29, Alex Williamson wrote:
> On Sun, 8 May 2022 16:10:51 +0300
> Yishai Hadas <yishaih@nvidia.com> wrote:
>
>> Manage the VF attach/detach callback from the PF.
>>
>> This lets the driver to enable parallel VFs migration as will be
>> introduced in the next patch.
>>
>> As part of this, reorganize the VF is migratable code to be in a
>> separate function and rename it to be set_migratable() to match its
>> functionality.
>>
>> Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
>> Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
>> ---
>>   drivers/vfio/pci/mlx5/cmd.c  | 63 ++++++++++++++++++++++++++++++++++++
>>   drivers/vfio/pci/mlx5/cmd.h  | 22 +++++++++++++
>>   drivers/vfio/pci/mlx5/main.c | 40 ++++-------------------
>>   3 files changed, 91 insertions(+), 34 deletions(-)
>>
>> diff --git a/drivers/vfio/pci/mlx5/cmd.c b/drivers/vfio/pci/mlx5/cmd.c
>> index 5c9f9218cc1d..5031978ae63a 100644
>> --- a/drivers/vfio/pci/mlx5/cmd.c
>> +++ b/drivers/vfio/pci/mlx5/cmd.c
>> @@ -71,6 +71,69 @@ int mlx5vf_cmd_query_vhca_migration_state(struct pci_dev *pdev, u16 vhca_id,
>>   	return ret;
>>   }
>>   
>> +static int mlx5fv_vf_event(struct notifier_block *nb,
>> +			   unsigned long event, void *data)
>> +{
>> +	struct mlx5vf_pci_core_device *mvdev =
>> +		container_of(nb, struct mlx5vf_pci_core_device, nb);
>> +
>> +	mutex_lock(&mvdev->state_mutex);
>> +	switch (event) {
>> +	case MLX5_PF_NOTIFY_ENABLE_VF:
>> +		mvdev->mdev_detach = false;
>> +		break;
>> +	case MLX5_PF_NOTIFY_DISABLE_VF:
>> +		mvdev->mdev_detach = true;
>> +		break;
>> +	default:
>> +		break;
>> +	}
>> +	mlx5vf_state_mutex_unlock(mvdev);
>> +	return 0;
>> +}
>> +
>> +void mlx5vf_cmd_remove_migratable(struct mlx5vf_pci_core_device *mvdev)
>> +{
>> +	mlx5_sriov_blocking_notifier_unregister(mvdev->mdev, mvdev->vf_id,
>> +						&mvdev->nb);
>> +}
>> +
>> +void mlx5vf_cmd_set_migratable(struct mlx5vf_pci_core_device *mvdev)
>> +{
>> +	struct pci_dev *pdev = mvdev->core_device.pdev;
>> +	int ret;
>> +
>> +	if (!pdev->is_virtfn)
>> +		return;
>> +
>> +	mvdev->mdev = mlx5_vf_get_core_dev(pdev);
>> +	if (!mvdev->mdev)
>> +		return;
>> +
>> +	if (!MLX5_CAP_GEN(mvdev->mdev, migration))
>> +		goto end;
>> +
>> +	mvdev->vf_id = pci_iov_vf_id(pdev);
>> +	if (mvdev->vf_id < 0)
>> +		goto end;
>> +
>> +	mutex_init(&mvdev->state_mutex);
>> +	spin_lock_init(&mvdev->reset_lock);
>> +	mvdev->nb.notifier_call = mlx5fv_vf_event;
>> +	ret = mlx5_sriov_blocking_notifier_register(mvdev->mdev, mvdev->vf_id,
>> +						    &mvdev->nb);
>> +	if (ret)
>> +		goto end;
>> +
>> +	mvdev->migrate_cap = 1;
>> +	mvdev->core_device.vdev.migration_flags =
>> +		VFIO_MIGRATION_STOP_COPY |
>> +		VFIO_MIGRATION_P2P;
>> +
>> +end:
>> +	mlx5_vf_put_core_dev(mvdev->mdev);
>> +}
>> +
>>   int mlx5vf_cmd_get_vhca_id(struct pci_dev *pdev, u16 function_id, u16 *vhca_id)
>>   {
>>   	struct mlx5_core_dev *mdev = mlx5_vf_get_core_dev(pdev);
>> diff --git a/drivers/vfio/pci/mlx5/cmd.h b/drivers/vfio/pci/mlx5/cmd.h
>> index 1392a11a9cc0..340a06b98007 100644
>> --- a/drivers/vfio/pci/mlx5/cmd.h
>> +++ b/drivers/vfio/pci/mlx5/cmd.h
>> @@ -7,6 +7,7 @@
>>   #define MLX5_VFIO_CMD_H
>>   
>>   #include <linux/kernel.h>
>> +#include <linux/vfio_pci_core.h>
>>   #include <linux/mlx5/driver.h>
>>   
>>   struct mlx5_vf_migration_file {
>> @@ -24,13 +25,34 @@ struct mlx5_vf_migration_file {
>>   	unsigned long last_offset;
>>   };
>>   
>> +struct mlx5vf_pci_core_device {
>> +	struct vfio_pci_core_device core_device;
>> +	int vf_id;
>> +	u16 vhca_id;
>> +	u8 migrate_cap:1;
>> +	u8 deferred_reset:1;
>> +	/* protect migration state */
>> +	struct mutex state_mutex;
>> +	enum vfio_device_mig_state mig_state;
>> +	/* protect the reset_done flow */
>> +	spinlock_t reset_lock;
>> +	struct mlx5_vf_migration_file *resuming_migf;
>> +	struct mlx5_vf_migration_file *saving_migf;
>> +	struct notifier_block nb;
>> +	struct mlx5_core_dev *mdev;
>> +	u8 mdev_detach:1;
>
> This should be packed with the other bit fields, there's plenty of
> space there.
>
Sure, will be part of V2.

>> +};
>> +
>>   int mlx5vf_cmd_suspend_vhca(struct pci_dev *pdev, u16 vhca_id, u16 op_mod);
>>   int mlx5vf_cmd_resume_vhca(struct pci_dev *pdev, u16 vhca_id, u16 op_mod);
>>   int mlx5vf_cmd_query_vhca_migration_state(struct pci_dev *pdev, u16 vhca_id,
>>   					  size_t *state_size);
>>   int mlx5vf_cmd_get_vhca_id(struct pci_dev *pdev, u16 function_id, u16 *vhca_id);
>> +void mlx5vf_cmd_set_migratable(struct mlx5vf_pci_core_device *mvdev);
>> +void mlx5vf_cmd_remove_migratable(struct mlx5vf_pci_core_device *mvdev);
>>   int mlx5vf_cmd_save_vhca_state(struct pci_dev *pdev, u16 vhca_id,
>>   			       struct mlx5_vf_migration_file *migf);
>>   int mlx5vf_cmd_load_vhca_state(struct pci_dev *pdev, u16 vhca_id,
>>   			       struct mlx5_vf_migration_file *migf);
>> +void mlx5vf_state_mutex_unlock(struct mlx5vf_pci_core_device *mvdev);
>>   #endif /* MLX5_VFIO_CMD_H */
>> diff --git a/drivers/vfio/pci/mlx5/main.c b/drivers/vfio/pci/mlx5/main.c
>> index bbec5d288fee..9716c87e31f9 100644
>> --- a/drivers/vfio/pci/mlx5/main.c
>> +++ b/drivers/vfio/pci/mlx5/main.c
>> @@ -17,7 +17,6 @@
>>   #include <linux/uaccess.h>
>>   #include <linux/vfio.h>
>>   #include <linux/sched/mm.h>
>> -#include <linux/vfio_pci_core.h>
>>   #include <linux/anon_inodes.h>
>>   
>>   #include "cmd.h"
>> @@ -25,20 +24,6 @@
>>   /* Arbitrary to prevent userspace from consuming endless memory */
>>   #define MAX_MIGRATION_SIZE (512*1024*1024)
>>   
>> -struct mlx5vf_pci_core_device {
>> -	struct vfio_pci_core_device core_device;
>> -	u16 vhca_id;
>> -	u8 migrate_cap:1;
>> -	u8 deferred_reset:1;
>> -	/* protect migration state */
>> -	struct mutex state_mutex;
>> -	enum vfio_device_mig_state mig_state;
>> -	/* protect the reset_done flow */
>> -	spinlock_t reset_lock;
>> -	struct mlx5_vf_migration_file *resuming_migf;
>> -	struct mlx5_vf_migration_file *saving_migf;
>> -};
>> -
>>   static struct page *
>>   mlx5vf_get_migration_page(struct mlx5_vf_migration_file *migf,
>>   			  unsigned long offset)
>> @@ -444,7 +429,7 @@ mlx5vf_pci_step_device_state_locked(struct mlx5vf_pci_core_device *mvdev,
>>    * This function is called in all state_mutex unlock cases to
>>    * handle a 'deferred_reset' if exists.
>>    */
>> -static void mlx5vf_state_mutex_unlock(struct mlx5vf_pci_core_device *mvdev)
>> +void mlx5vf_state_mutex_unlock(struct mlx5vf_pci_core_device *mvdev)
>>   {
>>   again:
>>   	spin_lock(&mvdev->reset_lock);
>> @@ -596,24 +581,7 @@ static int mlx5vf_pci_probe(struct pci_dev *pdev,
>>   	if (!mvdev)
>>   		return -ENOMEM;
>>   	vfio_pci_core_init_device(&mvdev->core_device, pdev, &mlx5vf_pci_ops);
>> -
>> -	if (pdev->is_virtfn) {
>> -		struct mlx5_core_dev *mdev =
>> -			mlx5_vf_get_core_dev(pdev);
>> -
>> -		if (mdev) {
>> -			if (MLX5_CAP_GEN(mdev, migration)) {
>> -				mvdev->migrate_cap = 1;
>> -				mvdev->core_device.vdev.migration_flags =
>> -					VFIO_MIGRATION_STOP_COPY |
>> -					VFIO_MIGRATION_P2P;
>> -				mutex_init(&mvdev->state_mutex);
>> -				spin_lock_init(&mvdev->reset_lock);
>> -			}
>> -			mlx5_vf_put_core_dev(mdev);
>> -		}
>> -	}
>> -
>> +	mlx5vf_cmd_set_migratable(mvdev);
>>   	ret = vfio_pci_core_register_device(&mvdev->core_device);
>>   	if (ret)
>>   		goto out_free;
>> @@ -622,6 +590,8 @@ static int mlx5vf_pci_probe(struct pci_dev *pdev,
>>   	return 0;
>>   
>>   out_free:
>> +	if (mvdev->migrate_cap)
>> +		mlx5vf_cmd_remove_migratable(mvdev);
>>   	vfio_pci_core_uninit_device(&mvdev->core_device);
>>   	kfree(mvdev);
>>   	return ret;
>> @@ -632,6 +602,8 @@ static void mlx5vf_pci_remove(struct pci_dev *pdev)
>>   	struct mlx5vf_pci_core_device *mvdev = dev_get_drvdata(&pdev->dev);
>>   
>>   	vfio_pci_core_unregister_device(&mvdev->core_device);
>> +	if (mvdev->migrate_cap)
>> +		mlx5vf_cmd_remove_migratable(mvdev);
>>   	vfio_pci_core_uninit_device(&mvdev->core_device);
>>   	kfree(mvdev);
>>   }
>
> Personally, I'd push the test into the function, ie.
>
> void mlx5vf_cmd_remove_migratable(struct mlx5vf_pci_core_device *mvdev)
> {
> 	if (!mvdev->migrate_cap)
> 		return;
>
> 	...
> }


Makes sense, this will keep the caller code symmetric/clean for both the 
set/remove calls.

Will be part of V2.

> But it's clearly functional this way as well.  Please do fix the bit
> field packing though.  Thanks,
>
> Alex
>

Re: [PATCH V1 mlx5-next 4/4] vfio/mlx5: Run the SAVE state command in an async mode

[Yishai Hadas]

On 09/05/2022 20:29, Alex Williamson wrote:
> On Sun, 8 May 2022 16:10:53 +0300
> Yishai Hadas <yishaih@nvidia.com> wrote:
>> diff --git a/drivers/vfio/pci/mlx5/cmd.h b/drivers/vfio/pci/mlx5/cmd.h
>> index 2a20b7435393..d053d314b745 100644
>> --- a/drivers/vfio/pci/mlx5/cmd.h
>> +++ b/drivers/vfio/pci/mlx5/cmd.h
>> @@ -10,10 +10,20 @@
>>   #include <linux/vfio_pci_core.h>
>>   #include <linux/mlx5/driver.h>
>>   
>> +struct mlx5vf_async_data {
>> +	struct mlx5_async_work cb_work;
>> +	struct work_struct work;
>> +	int status;
>> +	u32 pdn;
>> +	u32 mkey;
>> +	void *out;
>> +};
>> +
>>   struct mlx5_vf_migration_file {
>>   	struct file *filp;
>>   	struct mutex lock;
>>   	bool disabled;
>> +	u8 is_err:1;
> Convert @disabled to bit field as well to pack these?

OK

>
> ...
>> @@ -558,6 +592,13 @@ static int mlx5vf_pci_probe(struct pci_dev *pdev,
>>   		return -ENOMEM;
>>   	vfio_pci_core_init_device(&mvdev->core_device, pdev, &mlx5vf_pci_ops);
>>   	mlx5vf_cmd_set_migratable(mvdev);
>> +	if (mvdev->migrate_cap) {
>> +		mvdev->cb_wq = alloc_ordered_workqueue("mlx5vf_wq", 0);
>> +		if (!mvdev->cb_wq) {
>> +			ret = -ENOMEM;
>> +			goto out_free;
>> +		}
>> +	}
> Should this be rolled into mlx5vf_cmd_set_migratable(), updating the
> function to return -errno?

This can be done, however, I would still keep the function as void as 
you previously suggested.

In case the WQ somehow couldn't be created it just means that migratable 
functionality couldn't be activated and its cap won't be set.

>>   	ret = vfio_pci_core_register_device(&mvdev->core_device);
>>   	if (ret)
>>   		goto out_free;
>> @@ -566,8 +607,11 @@ static int mlx5vf_pci_probe(struct pci_dev *pdev,
>>   	return 0;
>>   
>>   out_free:
>> -	if (mvdev->migrate_cap)
>> +	if (mvdev->migrate_cap) {
>>   		mlx5vf_cmd_remove_migratable(mvdev);
>> +		if (mvdev->cb_wq)
>> +			destroy_workqueue(mvdev->cb_wq);
>> +	}
>>   	vfio_pci_core_uninit_device(&mvdev->core_device);
>>   	kfree(mvdev);
>>   	return ret;
>> @@ -578,8 +622,10 @@ static void mlx5vf_pci_remove(struct pci_dev *pdev)
>>   	struct mlx5vf_pci_core_device *mvdev = dev_get_drvdata(&pdev->dev);
>>   
>>   	vfio_pci_core_unregister_device(&mvdev->core_device);
>> -	if (mvdev->migrate_cap)
>> +	if (mvdev->migrate_cap) {
>>   		mlx5vf_cmd_remove_migratable(mvdev);
>> +		destroy_workqueue(mvdev->cb_wq);
>> +	}
>>   	vfio_pci_core_uninit_device(&mvdev->core_device);
>>   	kfree(mvdev);
>>   }
> This looks like more evidence for expanding remove_migratable(),
> rolling this in as well.  If this workqueue were setup in
> set_migratable() we'd not need the special condition to test if cb_wq
> is NULL while migrate_cap is set.  Thanks,
>
> Alex
>
Makes sense, will be part of V2.

Yishai