Skip to contentSkip to conversation

add multi queue support to vDPA ifc driver

PATCH Superseded Andy Pei 4 participants 181 messages Aug 23, 2022 – Oct 26, 2022 virtio core Page 2 of 2

lore
Share & tools
lore
author
#131 lore raw

RE: [PATCH v8 07/12] vdpa/ifc: change internal function name

In reply to Maxime Coquelin

Hi Maxime,

Thanks for your efforts.

Quoted 46 lines

-----Original Message-----
From: Maxime Coquelin <[email protected]>
Sent: Tuesday, October 18, 2022 9:45 PM
To: Pei, Andy <[email protected]>; [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>
Subject: Re: [PATCH v8 07/12] vdpa/ifc: change internal function name

Quoted 22 lines · Andy Pei

Change internal function name "find_internal_resource_by_dev" to "find_internal_resource_by_pci_dev".

---
  drivers/vdpa/ifc/ifcvf_vdpa.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)
(quoted diff) +0 -0

b/drivers/vdpa/ifc/ifcvf_vdpa.c index 9c49f9c..73d04ed 100644

drivers/vdpa/ifc/ifcvf_vdpa.c +1 -1
@@ -121,7 +121,7 @@ struct rte_vdpa_dev_info {
  }
 
  static struct internal_list *
-find_internal_resource_by_dev(struct rte_pci_device *pdev)
+find_internal_resource_by_pci_dev(struct rte_pci_device *pdev)
  {
  	int found = 0;
  	struct internal_list *list;

drivers/vdpa/ifc/ifcvf_vdpa.c struct rte_vdpa_dev_info dev_info[] = { · line 1752

Quoted 8 lines
drivers/vdpa/ifc/ifcvf_vdpa.c +1 -1
@@ -1746,7 +1746,7 @@ struct rte_vdpa_dev_info dev_info[] = {
  	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
  		return 0;
 
-	list = find_internal_resource_by_dev(pci_dev);
+	list = find_internal_resource_by_pci_dev(pci_dev);
  	if (list == NULL) {
  		DRV_LOG(ERR, "Invalid device: %s", pci_dev->name);
  		return -1;

Thanks,
Maxime

Reviewed-by on v8 7/12
#132 lore raw

RE: [PATCH v8 07/12] vdpa/ifc: change internal function name

In reply to Andy Pei
Quoted 40 lines

-----Original Message-----
From: Pei, Andy <[email protected]>
Sent: Tuesday, October 18, 2022 8:08 PM
To: [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: [PATCH v8 07/12] vdpa/ifc: change internal function name

Change internal function name "find_internal_resource_by_dev" to "find_internal_resource_by_pci_dev".

---
 drivers/vdpa/ifc/ifcvf_vdpa.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/vdpa/ifc/ifcvf_vdpa.c +2 -3
index 9c49f9c..73d04ed 100644
@@ -121,7 +121,7 @@ struct rte_vdpa_dev_info {
 }
 
 static struct internal_list *
-find_internal_resource_by_dev(struct rte_pci_device *pdev)
+find_internal_resource_by_pci_dev(struct rte_pci_device *pdev)
 {
 	int found = 0;
 	struct internal_list *list;
@@ -1746,7 +1746,7 @@ struct rte_vdpa_dev_info dev_info[] = {
 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 		return 0;
 
-	list = find_internal_resource_by_dev(pci_dev);
+	list = find_internal_resource_by_pci_dev(pci_dev);
 	if (list == NULL) {
 		DRV_LOG(ERR, "Invalid device: %s", pci_dev->name);
 		return -1;
--
Quoted 1 line

1.8.3.1

author
#133 lore raw

RE: [PATCH v8 07/12] vdpa/ifc: change internal function name

HI Chenbo,

Thanks for your review.

Quoted 50 lines

-----Original Message-----
From: Xia, Chenbo <[email protected]>
Sent: Wednesday, October 19, 2022 2:59 PM
To: Pei, Andy <[email protected]>; [email protected]
Cc: Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: RE: [PATCH v8 07/12] vdpa/ifc: change internal function name

Quoted 39 lines

-----Original Message-----
From: Pei, Andy <[email protected]>
Sent: Tuesday, October 18, 2022 8:08 PM
To: [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: [PATCH v8 07/12] vdpa/ifc: change internal function name

Change internal function name "find_internal_resource_by_dev" to "find_internal_resource_by_pci_dev".

---
 drivers/vdpa/ifc/ifcvf_vdpa.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
(quoted diff) +0 -0

b/drivers/vdpa/ifc/ifcvf_vdpa.c index 9c49f9c..73d04ed 100644

drivers/vdpa/ifc/ifcvf_vdpa.c +2 -3
@@ -121,7 +121,7 @@ struct rte_vdpa_dev_info {  }
 
 static struct internal_list *
-find_internal_resource_by_dev(struct rte_pci_device *pdev)
+find_internal_resource_by_pci_dev(struct rte_pci_device *pdev)
 {
 	int found = 0;
 	struct internal_list *list;
@@ -1746,7 +1746,7 @@ struct rte_vdpa_dev_info dev_info[] = {
 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 		return 0;
 
-	list = find_internal_resource_by_dev(pci_dev);
+	list = find_internal_resource_by_pci_dev(pci_dev);
 	if (list == NULL) {
 		DRV_LOG(ERR, "Invalid device: %s", pci_dev->name);
 		return -1;
--
Quoted 1 line

1.8.3.1

author
#134 lore raw

PATCH v8 8/12 vdpa/ifc: add internal API to get device

In reply to Andy Pei

Add new internal API "find_internal_resource_by_rte_dev" to get device.

---
 drivers/vdpa/ifc/ifcvf_vdpa.c | 28 ++++++++++++++++++++++++++--
 1 file changed, 26 insertions(+), 2 deletions(-)
drivers/vdpa/ifc/ifcvf_vdpa.c +26 -2
index 73d04ed..c16e263 100644
@@ -144,6 +144,29 @@ struct rte_vdpa_dev_info {
 	return list;
 }
 
+static struct internal_list *
+find_internal_resource_by_rte_dev(struct rte_device *rte_dev)
+{
+	int found = 0;
+	struct internal_list *list;
+
+	pthread_mutex_lock(&internal_list_lock);
+
+	TAILQ_FOREACH(list, &internal_list, next) {
+		if (rte_dev == &list->internal->pdev->device) {
+			found = 1;
+			break;
+		}
+	}
+
+	pthread_mutex_unlock(&internal_list_lock);
+
+	if (!found)
+		return NULL;
+
+	return list;
+}
+
 static int
 ifcvf_vfio_setup(struct ifcvf_internal *internal)
 {
@@ -1398,10 +1421,11 @@ struct rte_vdpa_dev_info {
 {
 	struct ifcvf_internal *internal;
 	struct internal_list *list;
+	struct rte_device *rte_dev = vdev->device;
 
-	list = find_internal_resource_by_vdev(vdev);
+	list = find_internal_resource_by_rte_dev(rte_dev);
 	if (list == NULL) {
-		DRV_LOG(ERR, "Invalid vDPA device: %p", vdev);
+		DRV_LOG(ERR, "Invalid rte device: %p", rte_dev);
 		return -1;
 	}
 
Signature · 1 line
1.8.3.1
Reviewed-by on v8 8/12
#135 lore raw

Re: [PATCH v8 08/12] vdpa/ifc: add internal API to get device

Quoted 8 lines · Andy Pei

Add new internal API "find_internal_resource_by_rte_dev" to get device.

---
  drivers/vdpa/ifc/ifcvf_vdpa.c | 28 ++++++++++++++++++++++++++--
  1 file changed, 26 insertions(+), 2 deletions(-)

Thanks,
Maxime

author
#136 lore raw

RE: [PATCH v8 08/12] vdpa/ifc: add internal API to get device

Hi Maxime,

Thanks for your efforts.

Quoted 24 lines

-----Original Message-----
From: Maxime Coquelin <[email protected]>
Sent: Tuesday, October 18, 2022 9:48 PM
To: Pei, Andy <[email protected]>; [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>
Subject: Re: [PATCH v8 08/12] vdpa/ifc: add internal API to get device

Quoted 8 lines · Andy Pei

Add new internal API "find_internal_resource_by_rte_dev" to get device.

---
  drivers/vdpa/ifc/ifcvf_vdpa.c | 28 ++++++++++++++++++++++++++--
  1 file changed, 26 insertions(+), 2 deletions(-)

Thanks,
Maxime

Reviewed-by on v8 8/12
#137 lore raw

RE: [PATCH v8 08/12] vdpa/ifc: add internal API to get device

In reply to Andy Pei
Quoted 66 lines

-----Original Message-----
From: Pei, Andy <[email protected]>
Sent: Tuesday, October 18, 2022 8:08 PM
To: [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: [PATCH v8 08/12] vdpa/ifc: add internal API to get device

Add new internal API "find_internal_resource_by_rte_dev" to get device.

---
 drivers/vdpa/ifc/ifcvf_vdpa.c | 28 ++++++++++++++++++++++++++--
 1 file changed, 26 insertions(+), 2 deletions(-)
drivers/vdpa/ifc/ifcvf_vdpa.c +26 -3
index 73d04ed..c16e263 100644
@@ -144,6 +144,29 @@ struct rte_vdpa_dev_info {
 	return list;
 }
 
+static struct internal_list *
+find_internal_resource_by_rte_dev(struct rte_device *rte_dev)
+{
+	int found = 0;
+	struct internal_list *list;
+
+	pthread_mutex_lock(&internal_list_lock);
+
+	TAILQ_FOREACH(list, &internal_list, next) {
+		if (rte_dev == &list->internal->pdev->device) {
+			found = 1;
+			break;
+		}
+	}
+
+	pthread_mutex_unlock(&internal_list_lock);
+
+	if (!found)
+		return NULL;
+
+	return list;
+}
+
 static int
 ifcvf_vfio_setup(struct ifcvf_internal *internal)
 {
@@ -1398,10 +1421,11 @@ struct rte_vdpa_dev_info {
 {
 	struct ifcvf_internal *internal;
 	struct internal_list *list;
+	struct rte_device *rte_dev = vdev->device;
 
-	list = find_internal_resource_by_vdev(vdev);
+	list = find_internal_resource_by_rte_dev(rte_dev);
 	if (list == NULL) {
-		DRV_LOG(ERR, "Invalid vDPA device: %p", vdev);
+		DRV_LOG(ERR, "Invalid rte device: %p", rte_dev);
 		return -1;
 	}
 
--
Quoted 1 line

1.8.3.1

author
#138 lore raw

RE: [PATCH v8 08/12] vdpa/ifc: add internal API to get device

Hi Chenbo,
Thanks for your review.

Quoted 74 lines

-----Original Message-----
From: Xia, Chenbo <[email protected]>
Sent: Wednesday, October 19, 2022 3:04 PM
To: Pei, Andy <[email protected]>; [email protected]
Cc: Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: RE: [PATCH v8 08/12] vdpa/ifc: add internal API to get device

Quoted 63 lines

-----Original Message-----
From: Pei, Andy <[email protected]>
Sent: Tuesday, October 18, 2022 8:08 PM
To: [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: [PATCH v8 08/12] vdpa/ifc: add internal API to get device

Add new internal API "find_internal_resource_by_rte_dev" to get device.

---
 drivers/vdpa/ifc/ifcvf_vdpa.c | 28 ++++++++++++++++++++++++++--
 1 file changed, 26 insertions(+), 2 deletions(-)
(quoted diff) +0 -0

b/drivers/vdpa/ifc/ifcvf_vdpa.c index 73d04ed..c16e263 100644

drivers/vdpa/ifc/ifcvf_vdpa.c +23 -0
@@ -144,6 +144,29 @@ struct rte_vdpa_dev_info {
 	return list;
 }
 
+static struct internal_list *
+find_internal_resource_by_rte_dev(struct rte_device *rte_dev) {
+	int found = 0;
+	struct internal_list *list;
+
+	pthread_mutex_lock(&internal_list_lock);
+
+	TAILQ_FOREACH(list, &internal_list, next) {
+		if (rte_dev == &list->internal->pdev->device) {
+			found = 1;
+			break;
+		}
+	}
+
+	pthread_mutex_unlock(&internal_list_lock);
+
+	if (!found)
+		return NULL;
+
+	return list;
+}
+
 static int
 ifcvf_vfio_setup(struct ifcvf_internal *internal)  { @@ -1398,10
+1421,11 @@ struct rte_vdpa_dev_info {  {
 	struct ifcvf_internal *internal;
drivers/vdpa/ifc/ifcvf_vdpa.c +3 -3
 	struct internal_list *list;
+	struct rte_device *rte_dev = vdev->device;
 
-	list = find_internal_resource_by_vdev(vdev);
+	list = find_internal_resource_by_rte_dev(rte_dev);
 	if (list == NULL) {
-		DRV_LOG(ERR, "Invalid vDPA device: %p", vdev);
+		DRV_LOG(ERR, "Invalid rte device: %p", rte_dev);
 		return -1;
 	}
 
--
Quoted 1 line

1.8.3.1

author
#139 lore raw

PATCH v8 9/12 vdpa/ifc: change some driver logic

In reply to Andy Pei

Insert internal list element to internal list before register vdpa device, in order to call vdpa ops during vdpa device registration.

---
 drivers/vdpa/ifc/ifcvf_vdpa.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
drivers/vdpa/ifc/ifcvf_vdpa.c +7 -4
index c16e263..8dfd493 100644
@@ -1737,17 +1737,20 @@ struct rte_vdpa_dev_info dev_info[] = {
 	}
 	internal->sw_lm = sw_fallback_lm;
 
+	pthread_mutex_lock(&internal_list_lock);
+	TAILQ_INSERT_TAIL(&internal_list, list, next);
+	pthread_mutex_unlock(&internal_list_lock);
+
 	internal->vdev = rte_vdpa_register_device(&pci_dev->device,
 				dev_info[internal->hw.device_type].ops);
 	if (internal->vdev == NULL) {
 		DRV_LOG(ERR, "failed to register device %s", pci_dev->name);
+		pthread_mutex_lock(&internal_list_lock);
+		TAILQ_REMOVE(&internal_list, list, next);
+		pthread_mutex_unlock(&internal_list_lock);
 		goto error;
 	}
 
-	pthread_mutex_lock(&internal_list_lock);
-	TAILQ_INSERT_TAIL(&internal_list, list, next);
-	pthread_mutex_unlock(&internal_list_lock);
-
 	rte_atomic32_set(&internal->started, 1);
 	update_datapath(internal);
 
Signature · 1 line
1.8.3.1
Reviewed-by on v8 9/12
#140 lore raw

Re: [PATCH v8 09/12] vdpa/ifc: change some driver logic

Quoted 9 lines · Andy Pei

Insert internal list element to internal list before register vdpa device, in order to call vdpa ops during vdpa device registration.

---
  drivers/vdpa/ifc/ifcvf_vdpa.c | 11 +++++++----
  1 file changed, 7 insertions(+), 4 deletions(-)

Thanks,
Maxime

author
#141 lore raw

RE: [PATCH v8 09/12] vdpa/ifc: change some driver logic

Hi Maxime,

Thanks.

Quoted 24 lines

-----Original Message-----
From: Maxime Coquelin <[email protected]>
Sent: Tuesday, October 18, 2022 9:57 PM
To: Pei, Andy <[email protected]>; [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>
Subject: Re: [PATCH v8 09/12] vdpa/ifc: change some driver logic

Quoted 8 lines · Andy Pei

Insert internal list element to internal list before register vdpa device, in order to call vdpa ops during vdpa device registration.

---
  drivers/vdpa/ifc/ifcvf_vdpa.c | 11 +++++++----
  1 file changed, 7 insertions(+), 4 deletions(-)

Thanks,
Maxime

#142 lore raw

RE: [PATCH v8 09/12] vdpa/ifc: change some driver logic

In reply to Andy Pei
Quoted 9 lines

-----Original Message-----
From: Pei, Andy <[email protected]>
Sent: Tuesday, October 18, 2022 8:08 PM
To: [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: [PATCH v8 09/12] vdpa/ifc: change some driver logic

Quoted 3 lines

Insert internal list element to internal list before register vdpa device, in order to call vdpa ops during vdpa device registration.

Patch seems good but title may be too general..

Maybe: improve internal list logic ?

Thanks,
Chenbo

Quoted 37 lines
---
 drivers/vdpa/ifc/ifcvf_vdpa.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
drivers/vdpa/ifc/ifcvf_vdpa.c +7 -5
index c16e263..8dfd493 100644
@@ -1737,17 +1737,20 @@ struct rte_vdpa_dev_info dev_info[] = {
 	}
 	internal->sw_lm = sw_fallback_lm;
 
+	pthread_mutex_lock(&internal_list_lock);
+	TAILQ_INSERT_TAIL(&internal_list, list, next);
+	pthread_mutex_unlock(&internal_list_lock);
+
 	internal->vdev = rte_vdpa_register_device(&pci_dev->device,
 				dev_info[internal->hw.device_type].ops);
 	if (internal->vdev == NULL) {
 		DRV_LOG(ERR, "failed to register device %s", pci_dev->name);
+		pthread_mutex_lock(&internal_list_lock);
+		TAILQ_REMOVE(&internal_list, list, next);
+		pthread_mutex_unlock(&internal_list_lock);
 		goto error;
 	}
 
-	pthread_mutex_lock(&internal_list_lock);
-	TAILQ_INSERT_TAIL(&internal_list, list, next);
-	pthread_mutex_unlock(&internal_list_lock);
-
 	rte_atomic32_set(&internal->started, 1);
 	update_datapath(internal);
 
--

1.8.3.1

author
#143 lore raw

RE: [PATCH v8 09/12] vdpa/ifc: change some driver logic

Hi Chenbo,

Thanks for your review, I will send v9 to do this.

Quoted 64 lines

-----Original Message-----
From: Xia, Chenbo <[email protected]>
Sent: Wednesday, October 19, 2022 5:13 PM
To: Pei, Andy <[email protected]>; [email protected]
Cc: Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: RE: [PATCH v8 09/12] vdpa/ifc: change some driver logic

Quoted 11 lines

-----Original Message-----
From: Pei, Andy <[email protected]>
Sent: Tuesday, October 18, 2022 8:08 PM
To: [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: [PATCH v8 09/12] vdpa/ifc: change some driver logic

Insert internal list element to internal list before register vdpa device, in order to call vdpa ops during vdpa device registration.

Patch seems good but title may be too general..

Maybe: improve internal list logic ?

Thanks,
Chenbo

Quoted 37 lines
---
 drivers/vdpa/ifc/ifcvf_vdpa.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
(quoted diff) +0 -0

b/drivers/vdpa/ifc/ifcvf_vdpa.c index c16e263..8dfd493 100644

drivers/vdpa/ifc/ifcvf_vdpa.c +7 -5
@@ -1737,17 +1737,20 @@ struct rte_vdpa_dev_info dev_info[] = {
 	}
 	internal->sw_lm = sw_fallback_lm;
 
+	pthread_mutex_lock(&internal_list_lock);
+	TAILQ_INSERT_TAIL(&internal_list, list, next);
+	pthread_mutex_unlock(&internal_list_lock);
+
 	internal->vdev = rte_vdpa_register_device(&pci_dev->device,
 				dev_info[internal->hw.device_type].ops);
 	if (internal->vdev == NULL) {
 		DRV_LOG(ERR, "failed to register device %s", pci_dev->name);
+		pthread_mutex_lock(&internal_list_lock);
+		TAILQ_REMOVE(&internal_list, list, next);
+		pthread_mutex_unlock(&internal_list_lock);
 		goto error;
 	}
 
-	pthread_mutex_lock(&internal_list_lock);
-	TAILQ_INSERT_TAIL(&internal_list, list, next);
-	pthread_mutex_unlock(&internal_list_lock);
-
 	rte_atomic32_set(&internal->started, 1);
 	update_datapath(internal);
 
--

1.8.3.1

author
#144 lore raw

PATCH v8 10/12 vhost: add type to rte vdpa device

In reply to Andy Pei

Add type to rte_vdpa_device to store device type. Call vdpa ops get_dev_type to fill type when register vdpa device.

---
 lib/vhost/socket.c      | 15 +--------------
 lib/vhost/vdpa.c        | 15 +++++++++++++++
 lib/vhost/vdpa_driver.h |  2 ++
 3 files changed, 18 insertions(+), 14 deletions(-)
lib/vhost/socket.c +1 -14
index 608ae57..863a6f6 100644
@@ -627,7 +627,6 @@ struct rte_vdpa_device *
 {
 	struct vhost_user_socket *vsocket;
 	struct rte_vdpa_device *vdpa_dev;
-	uint32_t vdpa_type = 0;
 	int ret = 0;
 
 	pthread_mutex_lock(&vhost_user.mutex);
@@ -644,19 +643,7 @@ struct rte_vdpa_device *
 		goto unlock_exit;
 	}
 
-	if (vdpa_dev->ops->get_dev_type) {
-		ret = vdpa_dev->ops->get_dev_type(vdpa_dev, &vdpa_type);
-		if (ret) {
-			VHOST_LOG_CONFIG(path, ERR,
-				"failed to get vdpa dev type for socket file.\n");
-			ret = -1;
-			goto unlock_exit;
-		}
-	} else {
-		vdpa_type = RTE_VHOST_VDPA_DEVICE_TYPE_NET;
-	}
-
-	*type = vdpa_type;
+	*type = vdpa_dev->type;
 
 unlock_exit:
 	pthread_mutex_unlock(&vhost_user.mutex);
lib/vhost/vdpa.c +15 -0
index bb82857..577cb00 100644
@@ -73,6 +73,7 @@ struct rte_vdpa_device *
 		struct rte_vdpa_dev_ops *ops)
 {
 	struct rte_vdpa_device *dev;
+	int ret = 0;
 
 	if (ops == NULL)
 		return NULL;
@@ -101,6 +102,20 @@ struct rte_vdpa_device *
 
 	dev->device = rte_dev;
 	dev->ops = ops;
+
+	if (ops->get_dev_type) {
+		ret = ops->get_dev_type(dev, &dev->type);
+		if (ret) {
+			VHOST_LOG_CONFIG(rte_dev->name, ERR,
+					 "Failed to get vdpa dev type.\n");
+			ret = -1;
+			goto out_unlock;
+		}
+	} else {
+		/** by default, we assume vdpa device is a net device */
+		dev->type = RTE_VHOST_VDPA_DEVICE_TYPE_NET;
+	}
+
 	TAILQ_INSERT_TAIL(&vdpa_device_list, dev, next);
 out_unlock:
 	rte_spinlock_unlock(&vdpa_device_list_lock);
lib/vhost/vdpa_driver.h +2 -0
index 8b88a53..8db4ab9 100644
@@ -92,6 +92,8 @@ struct rte_vdpa_device {
 	struct rte_device *device;
 	/** vdpa device operations */
 	struct rte_vdpa_dev_ops *ops;
+	/** vdpa device type: net, blk... */
+	uint32_t type;
 };
 
 /**
Signature · 1 line
1.8.3.1
Reviewed-by on v8 10/12
#145 lore raw

Re: [PATCH v8 10/12] vhost: add type to rte vdpa device

Quoted 11 lines · Andy Pei

Add type to rte_vdpa_device to store device type. Call vdpa ops get_dev_type to fill type when register vdpa device.

---
  lib/vhost/socket.c      | 15 +--------------
  lib/vhost/vdpa.c        | 15 +++++++++++++++
  lib/vhost/vdpa_driver.h |  2 ++
  3 files changed, 18 insertions(+), 14 deletions(-)

Thanks for doing the change!
Maxime

author
#146 lore raw

RE: [PATCH v8 10/12] vhost: add type to rte vdpa device

Hi Maxime,

Thanks for your efforts.

Quoted 26 lines

-----Original Message-----
From: Maxime Coquelin <[email protected]>
Sent: Tuesday, October 18, 2022 9:15 PM
To: Pei, Andy <[email protected]>; [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>
Subject: Re: [PATCH v8 10/12] vhost: add type to rte vdpa device

Quoted 10 lines · Andy Pei

Add type to rte_vdpa_device to store device type.
Call vdpa ops get_dev_type to fill type when register vdpa device.

---
  lib/vhost/socket.c      | 15 +--------------
  lib/vhost/vdpa.c        | 15 +++++++++++++++
  lib/vhost/vdpa_driver.h |  2 ++
  3 files changed, 18 insertions(+), 14 deletions(-)

Thanks for doing the change!
Maxime

author
#147 lore raw

PATCH v8 11/12 vhost: vDPA blk device gets ready when the first queue is ready

In reply to Andy Pei

When boot from virtio blk device, seabios in QEMU only enables one queue. To work in this scenario, vDPA BLK device back-end configure device when the first queue is ready.

---
 lib/vhost/vhost_user.c | 33 +++++++++++++++++----------------
 1 file changed, 17 insertions(+), 16 deletions(-)
lib/vhost/vhost_user.c +17 -16
index cd65257..e0ff79d 100644
@@ -1441,11 +1441,14 @@
 }
 
 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
+#define VIRTIO_BLK_NUM_VQS_TO_BE_READY 1u
 
 static int
 virtio_is_ready(struct virtio_net *dev)
 {
+	struct rte_vdpa_device *vdpa_dev;
 	struct vhost_virtqueue *vq;
+	uint32_t vdpa_type;
 	uint32_t i, nr_vring = dev->nr_vring;
 
 	if (dev->flags & VIRTIO_DEV_READY)
@@ -1454,13 +1457,22 @@
 	if (!dev->nr_vring)
 		return 0;
 
-	if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
-		nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
+	vdpa_dev = dev->vdpa_dev;
+	if (vdpa_dev)
+		vdpa_type = vdpa_dev->type;
+	else
+		vdpa_type = -1;
 
-		if (dev->nr_vring < nr_vring)
-			return 0;
+	if (vdpa_type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK) {
+		nr_vring = VIRTIO_BLK_NUM_VQS_TO_BE_READY;
+	} else {
+		if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET)
+			nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
 	}
 
+	if (dev->nr_vring < nr_vring)
+		return 0;
+
 	for (i = 0; i < nr_vring; i++) {
 		vq = dev->virtqueue[i];
 
@@ -2958,7 +2970,6 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	int ret;
 	int unlock_required = 0;
 	bool handled;
-	uint32_t vdpa_type = 0;
 	uint32_t request;
 	uint32_t i;
 
@@ -3170,17 +3181,7 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	if (!vdpa_dev)
 		goto out;
 
-	if (vdpa_dev->ops->get_dev_type) {
-		ret = vdpa_dev->ops->get_dev_type(vdpa_dev, &vdpa_type);
-		if (ret) {
-			VHOST_LOG_CONFIG(dev->ifname, ERR, "failed to get vdpa dev type.\n");
-			ret = -1;
-			goto out;
-		}
-	} else {
-		vdpa_type = RTE_VHOST_VDPA_DEVICE_TYPE_NET;
-	}
-	if (vdpa_type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
+	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
 		&& request != VHOST_USER_SET_VRING_CALL)
 		goto out;
 
Signature · 1 line
1.8.3.1
Reviewed-by on v8 11/12
#148 lore raw

Re: [PATCH v8 11/12] vhost: vDPA blk device gets ready when the first queue is ready

Quoted 10 lines · Andy Pei

When boot from virtio blk device, seabios in QEMU only enables one queue. To work in this scenario, vDPA BLK device back-end configure device when the first queue is ready.

---
  lib/vhost/vhost_user.c | 33 +++++++++++++++++----------------
  1 file changed, 17 insertions(+), 16 deletions(-)

Thanks,
Maxime

author
#149 lore raw

RE: [PATCH v8 11/12] vhost: vDPA blk device gets ready when the first queue is ready

HI Maxime,

Thanks for your review.

Quoted 27 lines

-----Original Message-----
From: Maxime Coquelin <[email protected]>
Sent: Tuesday, October 18, 2022 10:09 PM
To: Pei, Andy <[email protected]>; [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>
Subject: Re: [PATCH v8 11/12] vhost: vDPA blk device gets ready when the first queue is ready

Quoted 10 lines · Andy Pei

When boot from virtio blk device, seabios in QEMU only enables one queue. To work in this scenario, vDPA BLK device back-end configure device when the first queue is ready.

---
  lib/vhost/vhost_user.c | 33 +++++++++++++++++----------------
  1 file changed, 17 insertions(+), 16 deletions(-)

Thanks,
Maxime

Reviewed-by on v8 11/12
#150 lore raw

RE: [PATCH v8 11/12] vhost: vDPA blk device gets ready when the first queue is ready

In reply to Andy Pei
Quoted 95 lines

-----Original Message-----
From: Pei, Andy <[email protected]>
Sent: Tuesday, October 18, 2022 8:08 PM
To: [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: [PATCH v8 11/12] vhost: vDPA blk device gets ready when the first queue is ready

When boot from virtio blk device, seabios in QEMU only enables one queue. To work in this scenario, vDPA BLK device back-end configure device when the first queue is ready.

---
 lib/vhost/vhost_user.c | 33 +++++++++++++++++----------------
 1 file changed, 17 insertions(+), 16 deletions(-)
lib/vhost/vhost_user.c +17 -17
index cd65257..e0ff79d 100644
@@ -1441,11 +1441,14 @@
 }
 
 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
+#define VIRTIO_BLK_NUM_VQS_TO_BE_READY 1u
 
 static int
 virtio_is_ready(struct virtio_net *dev)
 {
+	struct rte_vdpa_device *vdpa_dev;
 	struct vhost_virtqueue *vq;
+	uint32_t vdpa_type;
 	uint32_t i, nr_vring = dev->nr_vring;
 
 	if (dev->flags & VIRTIO_DEV_READY)
@@ -1454,13 +1457,22 @@
 	if (!dev->nr_vring)
 		return 0;
 
-	if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
-		nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
+	vdpa_dev = dev->vdpa_dev;
+	if (vdpa_dev)
+		vdpa_type = vdpa_dev->type;
+	else
+		vdpa_type = -1;
 
-		if (dev->nr_vring < nr_vring)
-			return 0;
+	if (vdpa_type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK) {
+		nr_vring = VIRTIO_BLK_NUM_VQS_TO_BE_READY;
+	} else {
+		if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET)
+			nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
 	}
 
+	if (dev->nr_vring < nr_vring)
+		return 0;
+
 	for (i = 0; i < nr_vring; i++) {
 		vq = dev->virtqueue[i];
 
@@ -2958,7 +2970,6 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	int ret;
 	int unlock_required = 0;
 	bool handled;
-	uint32_t vdpa_type = 0;
 	uint32_t request;
 	uint32_t i;
 
@@ -3170,17 +3181,7 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	if (!vdpa_dev)
 		goto out;
 
-	if (vdpa_dev->ops->get_dev_type) {
-		ret = vdpa_dev->ops->get_dev_type(vdpa_dev, &vdpa_type);
-		if (ret) {
-			VHOST_LOG_CONFIG(dev->ifname, ERR, "failed to get vdpa
 dev type.\n");
-			ret = -1;
-			goto out;
-		}
-	} else {
-		vdpa_type = RTE_VHOST_VDPA_DEVICE_TYPE_NET;
-	}
-	if (vdpa_type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
+	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
 		&& request != VHOST_USER_SET_VRING_CALL)
 		goto out;
 
--
Quoted 1 line

1.8.3.1

author
#151 lore raw

RE: [PATCH v8 11/12] vhost: vDPA blk device gets ready when the first queue is ready

Hi Chenbo,

Thanks for your review.

Quoted 108 lines

-----Original Message-----
From: Xia, Chenbo <[email protected]>
Sent: Wednesday, October 19, 2022 5:15 PM
To: Pei, Andy <[email protected]>; [email protected]
Cc: Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: RE: [PATCH v8 11/12] vhost: vDPA blk device gets ready when the first queue is ready

Quoted 47 lines

-----Original Message-----
From: Pei, Andy <[email protected]>
Sent: Tuesday, October 18, 2022 8:08 PM
To: [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: [PATCH v8 11/12] vhost: vDPA blk device gets ready when the first queue is ready

When boot from virtio blk device, seabios in QEMU only enables one queue. To work in this scenario, vDPA BLK device back-end configure device when the first queue is ready.

---
 lib/vhost/vhost_user.c | 33 +++++++++++++++++----------------
 1 file changed, 17 insertions(+), 16 deletions(-)
lib/vhost/vhost_user.c index renamed from lib/vhost/vhost_user.c +0 -0

cd65257..e0ff79d 100644

lib/vhost/vhost_user.c +6 -2
@@ -1441,11 +1441,14 @@
 }
 
 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
+#define VIRTIO_BLK_NUM_VQS_TO_BE_READY 1u
 
 static int
 virtio_is_ready(struct virtio_net *dev)  {
+	struct rte_vdpa_device *vdpa_dev;
 	struct vhost_virtqueue *vq;
+	uint32_t vdpa_type;
 	uint32_t i, nr_vring = dev->nr_vring;
 
 	if (dev->flags & VIRTIO_DEV_READY)
@@ -1454,13 +1457,22 @@
 	if (!dev->nr_vring)
 		return 0;
 
-	if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
-		nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
+	vdpa_dev = dev->vdpa_dev;
+	if (vdpa_dev)
+		vdpa_type = vdpa_dev->type;

lib/vhost/vhost_user.c line 1470

Quoted 10 lines
lib/vhost/vhost_user.c +7 -2
+	else
+		vdpa_type = -1;
 
-		if (dev->nr_vring < nr_vring)
-			return 0;
+	if (vdpa_type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK) {
+		nr_vring = VIRTIO_BLK_NUM_VQS_TO_BE_READY;
+	} else {
+		if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET)
+			nr_vring =

VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;

Quoted 17 lines
lib/vhost/vhost_user.c +3 -1
 	}
 
+	if (dev->nr_vring < nr_vring)
+		return 0;
+
 	for (i = 0; i < nr_vring; i++) {
 		vq = dev->virtqueue[i];
 
@@ -2958,7 +2970,6 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	int ret;
 	int unlock_required = 0;
 	bool handled;
-	uint32_t vdpa_type = 0;
 	uint32_t request;
 	uint32_t i;
 

lib/vhost/vhost_user.c static int is_vring_iotlb(struct virtio_net *dev, · line 3176

Quoted 7 lines
lib/vhost/vhost_user.c +0 -4
@@ -3170,17 +3181,7 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	if (!vdpa_dev)
 		goto out;
 
-	if (vdpa_dev->ops->get_dev_type) {
-		ret = vdpa_dev->ops->get_dev_type(vdpa_dev, &vdpa_type);
-		if (ret) {
-			VHOST_LOG_CONFIG(dev->ifname, ERR, "failed to

get vdpa

Quoted 13 lines
lib/vhost/vhost_user.c +1 -8
 dev type.\n");
-			ret = -1;
-			goto out;
-		}
-	} else {
-		vdpa_type = RTE_VHOST_VDPA_DEVICE_TYPE_NET;
-	}
-	if (vdpa_type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
+	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
 		&& request != VHOST_USER_SET_VRING_CALL)
 		goto out;
 
--
Quoted 1 line

1.8.3.1

author
#152 lore raw

PATCH v8 12/12 vhost: improve vDPA blk device configure condition

In reply to Andy Pei

To support multi-queue, configure device
after call fd of all queues are set.

---
 lib/vhost/vhost_user.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)
lib/vhost/vhost_user.c +10 -3
index e0ff79d..9902ae9 100644
@@ -2972,6 +2972,7 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	bool handled;
 	uint32_t request;
 	uint32_t i;
+	uint16_t blk_call_fd;
 
 	dev = get_device(vid);
 	if (dev == NULL)
@@ -3181,9 +3182,15 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	if (!vdpa_dev)
 		goto out;
 
-	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
-		&& request != VHOST_USER_SET_VRING_CALL)
-		goto out;
+	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK) {
+		if (request == VHOST_USER_SET_VRING_CALL) {
+			blk_call_fd = ctx.msg.payload.u64 & VHOST_USER_VRING_IDX_MASK;
+			if (blk_call_fd != dev->nr_vring - 1)
+				goto out;
+		} else {
+			goto out;
+		}
+	}
 
 	if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
 		if (vdpa_dev->ops->dev_conf(dev->vid))
Signature · 1 line
1.8.3.1
Reviewed-by on v8 12/12
#153 lore raw

Re: [PATCH v8 12/12] vhost: improve vDPA blk device configure condition

Quoted 30 lines · Andy Pei

To support multi-queue, configure device
after call fd of all queues are set.

---
  lib/vhost/vhost_user.c | 13 ++++++++++---
  1 file changed, 10 insertions(+), 3 deletions(-)
lib/vhost/vhost_user.c +3 -3
index e0ff79d..9902ae9 100644
@@ -2972,6 +2972,7 @@ static int is_vring_iotlb(struct virtio_net *dev,
  	bool handled;
  	uint32_t request;
  	uint32_t i;
+	uint16_t blk_call_fd;
 
  	dev = get_device(vid);
  	if (dev == NULL)
@@ -3181,9 +3182,15 @@ static int is_vring_iotlb(struct virtio_net *dev,
  	if (!vdpa_dev)
  		goto out;
 
-	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
-		&& request != VHOST_USER_SET_VRING_CALL)
-		goto out;
+	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK) {
+		if (request == VHOST_USER_SET_VRING_CALL) {

lib/vhost/vhost_user.c static int is_vring_iotlb(struct virtio_net *dev, · line 3196

Quoted 10 lines
lib/vhost/vhost_user.c +7 -0
+			blk_call_fd = ctx.msg.payload.u64 & VHOST_USER_VRING_IDX_MASK;
+			if (blk_call_fd != dev->nr_vring - 1)
+				goto out;
+		} else {
+			goto out;
+		}
+	}
 
  	if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
  		if (vdpa_dev->ops->dev_conf(dev->vid))

Thanks,
Maxime

author
#154 lore raw

RE: [PATCH v8 12/12] vhost: improve vDPA blk device configure condition

Hi Maxime,

Thanks for your review.

Quoted 58 lines

-----Original Message-----
From: Maxime Coquelin <[email protected]>
Sent: Tuesday, October 18, 2022 10:15 PM
To: Pei, Andy <[email protected]>; [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>
Subject: Re: [PATCH v8 12/12] vhost: improve vDPA blk device configure condition

Quoted 22 lines · Andy Pei

To support multi-queue, configure device after call fd of all queues are set.

---
  lib/vhost/vhost_user.c | 13 ++++++++++---
  1 file changed, 10 insertions(+), 3 deletions(-)
lib/vhost/vhost_user.c index renamed from lib/vhost/vhost_user.c +0 -0

e0ff79d..9902ae9 100644

lib/vhost/vhost_user.c +1 -0
@@ -2972,6 +2972,7 @@ static int is_vring_iotlb(struct virtio_net *dev,
  	bool handled;
  	uint32_t request;
  	uint32_t i;
+	uint16_t blk_call_fd;
 
  	dev = get_device(vid);
  	if (dev == NULL)

lib/vhost/vhost_user.c static int is_vring_iotlb(struct virtio_net *dev, · line 3187

Quoted 9 lines
lib/vhost/vhost_user.c +3 -3
@@ -3181,9 +3182,15 @@ static int is_vring_iotlb(struct virtio_net *dev,
  	if (!vdpa_dev)
  		goto out;
 
-	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
-		&& request != VHOST_USER_SET_VRING_CALL)
-		goto out;
+	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK) {
+		if (request == VHOST_USER_SET_VRING_CALL) {
+			blk_call_fd = ctx.msg.payload.u64 &

VHOST_USER_VRING_IDX_MASK;

lib/vhost/vhost_user.c static int is_vring_iotlb(struct virtio_net *dev, · line 3196

Quoted 9 lines
lib/vhost/vhost_user.c +6 -0
+			if (blk_call_fd != dev->nr_vring - 1)
+				goto out;
+		} else {
+			goto out;
+		}
+	}
 
  	if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
  		if (vdpa_dev->ops->dev_conf(dev->vid))

Thanks,
Maxime

Reviewed-by on v8 12/12
#155 lore raw

RE: [PATCH v8 12/12] vhost: improve vDPA blk device configure condition

In reply to Andy Pei
Quoted 50 lines

-----Original Message-----
From: Pei, Andy <[email protected]>
Sent: Tuesday, October 18, 2022 8:08 PM
To: [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: [PATCH v8 12/12] vhost: improve vDPA blk device configure condition

To support multi-queue, configure device
after call fd of all queues are set.

---
 lib/vhost/vhost_user.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)
lib/vhost/vhost_user.c +10 -3
index e0ff79d..9902ae9 100644
@@ -2972,6 +2972,7 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	bool handled;
 	uint32_t request;
 	uint32_t i;
+	uint16_t blk_call_fd;
 
 	dev = get_device(vid);
 	if (dev == NULL)
@@ -3181,9 +3182,15 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	if (!vdpa_dev)
 		goto out;
 
-	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
-		&& request != VHOST_USER_SET_VRING_CALL)
-		goto out;
+	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK) {
+		if (request == VHOST_USER_SET_VRING_CALL) {
+			blk_call_fd = ctx.msg.payload.u64 &
 VHOST_USER_VRING_IDX_MASK;
+			if (blk_call_fd != dev->nr_vring - 1)
+				goto out;
+		} else {
+			goto out;
+		}
+	}
 
 	if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
Quoted 3 lines

if (vdpa_dev->ops->dev_conf(dev->vid))
--
1.8.3.1

author
#156 lore raw

RE: [PATCH v8 12/12] vhost: improve vDPA blk device configure condition

Hi Chenbo,

Thanks for your review.

Quoted 64 lines

-----Original Message-----
From: Xia, Chenbo <[email protected]>
Sent: Wednesday, October 19, 2022 5:15 PM
To: Pei, Andy <[email protected]>; [email protected]
Cc: Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: RE: [PATCH v8 12/12] vhost: improve vDPA blk device configure condition

Quoted 50 lines

-----Original Message-----
From: Pei, Andy <[email protected]>
Sent: Tuesday, October 18, 2022 8:08 PM
To: [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: [PATCH v8 12/12] vhost: improve vDPA blk device configure condition

To support multi-queue, configure device after call fd of all queues are set.

---
 lib/vhost/vhost_user.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)
lib/vhost/vhost_user.c index renamed from lib/vhost/vhost_user.c +0 -0

e0ff79d..9902ae9 100644

lib/vhost/vhost_user.c +10 -3
@@ -2972,6 +2972,7 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	bool handled;
 	uint32_t request;
 	uint32_t i;
+	uint16_t blk_call_fd;
 
 	dev = get_device(vid);
 	if (dev == NULL)
@@ -3181,9 +3182,15 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	if (!vdpa_dev)
 		goto out;
 
-	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
-		&& request != VHOST_USER_SET_VRING_CALL)
-		goto out;
+	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK) {
+		if (request == VHOST_USER_SET_VRING_CALL) {
+			blk_call_fd = ctx.msg.payload.u64 &
 VHOST_USER_VRING_IDX_MASK;
+			if (blk_call_fd != dev->nr_vring - 1)
+				goto out;
+		} else {
+			goto out;
+		}
+	}
 
 	if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
Quoted 3 lines

if (vdpa_dev->ops->dev_conf(dev->vid))
--
1.8.3.1

author
#157 lore raw

v9 cover vdpa/ifc: add multi queue support

In reply to Andy Pei

v9:
fix some commit message.

v8:
change "vdpa_device_type" in "rte_vdpa_device" to "type".

v7:
Fill vdpa_device_type in vdpa device registration.

v6:
Add vdpa_device_type to rte_vdpa_device to store vDPA device type.

v5:
fix some commit message.
rework some code logic.

v4:
fix some commit message.
add some commets to code.
fix some code to reduce confusion.

v3:

 rename device ID macro name.

fix some patch title and commit message.
delete some used marco.
rework some code logic.

v2:
fix some coding style issue.
support dynamic enable/disable queue at run time.

Andy Pei (10):
vdpa/ifc: add multi-queue support
vdpa/ifc: set max queues based on virtio spec vdpa/ifc: write queue count to MQ register
vdpa/ifc: only configure enabled queue
vdpa/ifc: change internal function name
vdpa/ifc: add internal API to get device
vdpa/ifc: improve internal list logic
vhost: add type to rte vdpa device
vhost: vDPA blk device gets ready when the first queue is ready
vhost: improve vDPA blk device configure condition

Huang Wei (2):
vdpa/ifc: add new device ID for legacy network device vdpa/ifc: support dynamic enable/disable queue

 drivers/vdpa/ifc/base/ifcvf.c | 144 ++++++++++++++++++++++++++++++++
 drivers/vdpa/ifc/base/ifcvf.h |  16 +++-
 drivers/vdpa/ifc/ifcvf_vdpa.c | 185 +++++++++++++++++++++++++++++++++++-------
 lib/vhost/socket.c            |  15 +---
 lib/vhost/vdpa.c              |  15 ++++
 lib/vhost/vdpa_driver.h       |   2 +
 lib/vhost/vhost_user.c        |  38 +++++----
 7 files changed, 354 insertions(+), 61 deletions(-)
Signature · 1 line
1.8.3.1
author
#158 lore raw

PATCH v9 1/12 vdpa/ifc: add new device ID for legacy network device

From: Huang Wei <[email protected]>

Add new device id to support IFCVF_NET_TRANSITIONAL_DEVICE_ID (0x1000). Rename macro from "IFCVF_BLK_DEVICE_ID" to "IFCVF_SUBSYS_BLK_DEVICE_ID".

---
 drivers/vdpa/ifc/base/ifcvf.h |  6 ++++--
 drivers/vdpa/ifc/ifcvf_vdpa.c | 13 ++++++++++---
 2 files changed, 14 insertions(+), 5 deletions(-)
drivers/vdpa/ifc/base/ifcvf.h +4 -2
index 9d95aac..ef7697a 100644
@@ -12,12 +12,14 @@
 #define IFCVF_BLK	1
 
 #define IFCVF_VENDOR_ID                     0x1AF4
-#define IFCVF_NET_DEVICE_ID                 0x1041
+#define IFCVF_NET_MODERN_DEVICE_ID          0x1041
 #define IFCVF_BLK_MODERN_DEVICE_ID          0x1042
+#define IFCVF_NET_TRANSITIONAL_DEVICE_ID    0x1000
 #define IFCVF_BLK_TRANSITIONAL_DEVICE_ID    0x1001
 #define IFCVF_SUBSYS_VENDOR_ID              0x8086
 #define IFCVF_SUBSYS_DEVICE_ID              0x001A
-#define IFCVF_BLK_DEVICE_ID                 0x0002
+#define IFCVF_SUBSYS_NET_DEVICE_ID          0x0001
+#define IFCVF_SUBSYS_BLK_DEVICE_ID          0x0002
 
 #define IFCVF_MAX_QUEUES		1
 
drivers/vdpa/ifc/ifcvf_vdpa.c +10 -3
index d5ac583..b4389a0 100644
@@ -1684,23 +1684,30 @@ struct rte_vdpa_dev_info dev_info[] = {
 static const struct rte_pci_id pci_id_ifcvf_map[] = {
 	{ .class_id = RTE_CLASS_ANY_ID,
 	  .vendor_id = IFCVF_VENDOR_ID,
-	  .device_id = IFCVF_NET_DEVICE_ID,
+	  .device_id = IFCVF_NET_MODERN_DEVICE_ID,
 	  .subsystem_vendor_id = IFCVF_SUBSYS_VENDOR_ID,
 	  .subsystem_device_id = IFCVF_SUBSYS_DEVICE_ID,
 	},
 
 	{ .class_id = RTE_CLASS_ANY_ID,
 	  .vendor_id = IFCVF_VENDOR_ID,
+	  .device_id = IFCVF_NET_TRANSITIONAL_DEVICE_ID,
+	  .subsystem_vendor_id = IFCVF_SUBSYS_VENDOR_ID,
+	  .subsystem_device_id = IFCVF_SUBSYS_NET_DEVICE_ID,
+	},
+
+	{ .class_id = RTE_CLASS_ANY_ID,
+	  .vendor_id = IFCVF_VENDOR_ID,
 	  .device_id = IFCVF_BLK_TRANSITIONAL_DEVICE_ID,
 	  .subsystem_vendor_id = IFCVF_SUBSYS_VENDOR_ID,
-	  .subsystem_device_id = IFCVF_BLK_DEVICE_ID,
+	  .subsystem_device_id = IFCVF_SUBSYS_BLK_DEVICE_ID,
 	},
 
 	{ .class_id = RTE_CLASS_ANY_ID,
 	  .vendor_id = IFCVF_VENDOR_ID,
 	  .device_id = IFCVF_BLK_MODERN_DEVICE_ID,
 	  .subsystem_vendor_id = IFCVF_SUBSYS_VENDOR_ID,
-	  .subsystem_device_id = IFCVF_BLK_DEVICE_ID,
+	  .subsystem_device_id = IFCVF_SUBSYS_BLK_DEVICE_ID,
 	},
 
 	{ .vendor_id = 0, /* sentinel */
Signature · 1 line
1.8.3.1
author
#159 lore raw

PATCH v9 2/12 vdpa/ifc: add multi-queue support

In reply to Andy Pei

Enable VHOST_USER_PROTOCOL_F_MQ feature. Expose IFCVF_MQ_OFFSET register to enable multi-queue.

---
 drivers/vdpa/ifc/base/ifcvf.c | 9 +++++++++
 drivers/vdpa/ifc/base/ifcvf.h | 2 ++
 drivers/vdpa/ifc/ifcvf_vdpa.c | 1 +
 3 files changed, 12 insertions(+)
drivers/vdpa/ifc/base/ifcvf.c +9 -0
index f1e1474..81c68c0 100644
@@ -90,6 +90,15 @@
 	if (!hw->lm_cfg)
 		WARNINGOUT("HW support live migration not support!\n");
 
+	/* For some hardware implementation, for example:
+	 * the BAR 4 of PF is NULL, while BAR 4 of VF is not.
+	 * This code makes sure hw->mq_cfg is a valid address.
+	 */
+	if (hw->mem_resource[4].addr)
+		hw->mq_cfg = hw->mem_resource[4].addr + IFCVF_MQ_OFFSET;
+	else
+		hw->mq_cfg = NULL;
+
 	if (hw->common_cfg == NULL || hw->notify_base == NULL ||
 			hw->isr == NULL || hw->dev_cfg == NULL) {
 		DEBUGOUT("capability incomplete\n");
drivers/vdpa/ifc/base/ifcvf.h +2 -0
index ef7697a..d16d9ab 100644
@@ -50,6 +50,7 @@
 
 #define IFCVF_LM_CFG_SIZE		0x40
 #define IFCVF_LM_RING_STATE_OFFSET	0x20
+#define IFCVF_MQ_OFFSET			0x28
 
 #define IFCVF_LM_LOGGING_CTRL		0x0
 
@@ -149,6 +150,7 @@ struct ifcvf_hw {
 	u16    *notify_base;
 	u16    *notify_addr[IFCVF_MAX_QUEUES * 2];
 	u8     *lm_cfg;
+	u8     *mq_cfg;
 	struct vring_info vring[IFCVF_MAX_QUEUES * 2];
 	u8 nr_vring;
 	int device_type;
drivers/vdpa/ifc/ifcvf_vdpa.c +1 -0
index b4389a0..008cf89 100644
@@ -1248,6 +1248,7 @@ struct rte_vdpa_dev_info {
 		 1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD | \
 		 1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER | \
 		 1ULL << VHOST_USER_PROTOCOL_F_LOG_SHMFD | \
+		 1ULL << VHOST_USER_PROTOCOL_F_MQ | \
 		 1ULL << VHOST_USER_PROTOCOL_F_STATUS)
 
 #define VDPA_BLK_PROTOCOL_FEATURES \
Signature · 1 line
1.8.3.1
author
#160 lore raw

PATCH v9 3/12 vdpa/ifc: set max queues based on virtio spec

In reply to Andy Pei

Set max_queues according to virtio spec.
For virtio BLK device, set max_queues to the value of num_queues in struct virtio_blk_config.
For virtio NET device, read num_queues from struct ifcvf_pci_common_cfg, get the queue pair number using num_queues and set max_queues to it.

---
 drivers/vdpa/ifc/base/ifcvf.h |  2 +-
 drivers/vdpa/ifc/ifcvf_vdpa.c | 19 ++++++++++++++++++-
 2 files changed, 19 insertions(+), 2 deletions(-)
drivers/vdpa/ifc/base/ifcvf.h +1 -1
index d16d9ab..1e133c0 100644
@@ -21,7 +21,7 @@
 #define IFCVF_SUBSYS_NET_DEVICE_ID          0x0001
 #define IFCVF_SUBSYS_BLK_DEVICE_ID          0x0002
 
-#define IFCVF_MAX_QUEUES		1
+#define IFCVF_MAX_QUEUES		32
 
 #ifndef VIRTIO_F_IOMMU_PLATFORM
 #define VIRTIO_F_IOMMU_PLATFORM		33
drivers/vdpa/ifc/ifcvf_vdpa.c +18 -1
index 008cf89..5a24204 100644
@@ -26,6 +26,12 @@
 
 #include "base/ifcvf.h"
 
+/*
+ * RTE_MIN() cannot be used since braced-group within expression allowed
+ * only inside a function.
+ */
+#define MIN(v1, v2)	((v1) < (v2) ? (v1) : (v2))
+
 RTE_LOG_REGISTER(ifcvf_vdpa_logtype, pmd.vdpa.ifcvf, NOTICE);
 #define DRV_LOG(level, fmt, args...) \
 	rte_log(RTE_LOG_ ## level, ifcvf_vdpa_logtype, \
@@ -1512,6 +1518,7 @@ struct rte_vdpa_dev_info dev_info[] = {
 	uint64_t capacity = 0;
 	uint8_t *byte;
 	uint32_t i;
+	uint16_t queue_pairs;
 
 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 		return 0;
@@ -1559,7 +1566,6 @@ struct rte_vdpa_dev_info dev_info[] = {
 	}
 
 	internal->configured = 0;
-	internal->max_queues = IFCVF_MAX_QUEUES;
 	features = ifcvf_get_features(&internal->hw);
 
 	device_id = ifcvf_pci_get_device_type(pci_dev);
@@ -1570,6 +1576,14 @@ struct rte_vdpa_dev_info dev_info[] = {
 
 	if (device_id == VIRTIO_ID_NET) {
 		internal->hw.device_type = IFCVF_NET;
+		/*
+		 * ifc device always has CTRL_VQ,
+		 * and supports VIRTIO_NET_F_CTRL_VQ feature.
+		 */
+		queue_pairs = (internal->hw.common_cfg->num_queues - 1) / 2;
+		DRV_LOG(INFO, "%s support %u queue pairs", pci_dev->name,
+			queue_pairs);
+		internal->max_queues = MIN(IFCVF_MAX_QUEUES, queue_pairs);
 		internal->features = features &
 					~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
 		internal->features |= dev_info[IFCVF_NET].features;
@@ -1609,6 +1623,9 @@ struct rte_vdpa_dev_info dev_info[] = {
 			internal->hw.blk_cfg->geometry.sectors);
 		DRV_LOG(DEBUG, "num_queues: 0x%08x",
 			internal->hw.blk_cfg->num_queues);
+
+		internal->max_queues = MIN(IFCVF_MAX_QUEUES,
+			internal->hw.blk_cfg->num_queues);
 	}
 
 	list->internal = internal;
Signature · 1 line
1.8.3.1
author
#161 lore raw

PATCH v9 4/12 vdpa/ifc: write queue count to MQ register

In reply to Andy Pei

Write queue count to IFCVF_MQ_OFFSET register to enable multi-queue feature.

---
 drivers/vdpa/ifc/base/ifcvf.c | 32 ++++++++++++++++++++++++++++++++
 1 file changed, 32 insertions(+)
drivers/vdpa/ifc/base/ifcvf.c +32 -0
index 81c68c0..b377126 100644
@@ -202,6 +202,37 @@
 	IFCVF_WRITE_REG32(val >> 32, hi);
 }
 
+STATIC void
+ifcvf_enable_mq(struct ifcvf_hw *hw)
+{
+	u8 *mq_cfg;
+	u8 qid;
+	int nr_queue = 0;
+
+	for (qid = 0; qid < hw->nr_vring; qid++) {
+		if (!hw->vring[qid].enable)
+			continue;
+		nr_queue++;
+	}
+
+	if (nr_queue == 0) {
+		WARNINGOUT("no enabled vring\n");
+		return;
+	}
+
+	mq_cfg = hw->mq_cfg;
+	if (mq_cfg) {
+		if (hw->device_type == IFCVF_BLK) {
+			*(u32 *)mq_cfg = nr_queue;
+			RTE_LOG(INFO, PMD, "%d queues are enabled\n", nr_queue);
+		} else {
+			*(u32 *)mq_cfg = nr_queue / 2;
+			RTE_LOG(INFO, PMD, "%d queue pairs are enabled\n",
+				nr_queue / 2);
+		}
+	}
+}
+
 STATIC int
 ifcvf_hw_enable(struct ifcvf_hw *hw)
 {
@@ -219,6 +250,7 @@
 		return -1;
 	}
 
+	ifcvf_enable_mq(hw);
 	for (i = 0; i < hw->nr_vring; i++) {
 		IFCVF_WRITE_REG16(i, &cfg->queue_select);
 		io_write64_twopart(hw->vring[i].desc, &cfg->queue_desc_lo,
Signature · 1 line
1.8.3.1
author
#162 lore raw

PATCH v9 5/12 vdpa/ifc: only configure enabled queue

In reply to Andy Pei

When configuring the hardware queue, we only configure queues which have been enabled by vhost.

---
 drivers/vdpa/ifc/base/ifcvf.c |  3 +++
 drivers/vdpa/ifc/ifcvf_vdpa.c | 16 ++++++++++++++--
 2 files changed, 17 insertions(+), 2 deletions(-)
drivers/vdpa/ifc/base/ifcvf.c +3 -0
index b377126..30bb8cb 100644
@@ -252,6 +252,9 @@
 
 	ifcvf_enable_mq(hw);
 	for (i = 0; i < hw->nr_vring; i++) {
+		if (!hw->vring[i].enable)
+			continue;
+
 		IFCVF_WRITE_REG16(i, &cfg->queue_select);
 		io_write64_twopart(hw->vring[i].desc, &cfg->queue_desc_lo,
 				&cfg->queue_desc_hi);
drivers/vdpa/ifc/ifcvf_vdpa.c +14 -2
index 5a24204..0c3407a 100644
@@ -284,6 +284,8 @@ struct rte_vdpa_dev_info {
 	rte_vhost_get_negotiated_features(vid, &hw->req_features);
 
 	for (i = 0; i < nr_vring; i++) {
+		if (!hw->vring[i].enable)
+			continue;
 		rte_vhost_get_vhost_vring(vid, i, &vq);
 		gpa = hva_to_gpa(vid, (uint64_t)(uintptr_t)vq.desc);
 		if (gpa == 0) {
@@ -499,6 +501,8 @@ struct rte_vdpa_dev_info {
 
 	vring.kickfd = -1;
 	for (qid = 0; qid < q_num; qid++) {
+		if (!hw->vring[qid].enable)
+			continue;
 		ev.events = EPOLLIN | EPOLLPRI;
 		rte_vhost_get_vhost_vring(internal->vid, qid, &vring);
 		ev.data.u64 = qid | (uint64_t)vring.kickfd << 32;
@@ -1058,6 +1062,8 @@ struct rte_vdpa_dev_info {
 	struct rte_vdpa_device *vdev;
 	struct internal_list *list;
 	struct ifcvf_internal *internal;
+	struct ifcvf_hw *hw;
+	uint16_t i;
 
 	vdev = rte_vhost_get_vdpa_device(vid);
 	list = find_internal_resource_by_vdev(vdev);
@@ -1071,11 +1077,17 @@ struct rte_vdpa_dev_info {
 	rte_atomic32_set(&internal->dev_attached, 1);
 	update_datapath(internal);
 
-	if (rte_vhost_host_notifier_ctrl(vid, RTE_VHOST_QUEUE_ALL, true) != 0)
-		DRV_LOG(NOTICE, "vDPA (%s): software relay is used.",
+	hw = &internal->hw;
+	for (i = 0; i < hw->nr_vring; i++) {
+		if (!hw->vring[i].enable)
+			continue;
+		if (rte_vhost_host_notifier_ctrl(vid, i, true) != 0)
+			DRV_LOG(NOTICE, "vDPA (%s): software relay is used.",
 				vdev->device->name);
+	}
 
 	internal->configured = 1;
+	DRV_LOG(INFO, "vDPA device %s is configured", vdev->device->name);
 	return 0;
 }
 
Signature · 1 line
1.8.3.1
author
#163 lore raw

PATCH v9 6/12 vdpa/ifc: support dynamic enable/disable queue

In reply to Andy Pei

From: Huang Wei <[email protected]>

Support dynamic enable or disable queue.
For front end, like QEMU, user can use ethtool to configure queue. For example, "ethtool -L eth0 combined 3" to enable 3 queues pairs.

---
 drivers/vdpa/ifc/base/ifcvf.c | 100 ++++++++++++++++++++++++++++++++++++++++++
 drivers/vdpa/ifc/base/ifcvf.h |   6 +++
 drivers/vdpa/ifc/ifcvf_vdpa.c |  93 ++++++++++++++++++++++++++++++++-------
 3 files changed, 184 insertions(+), 15 deletions(-)
drivers/vdpa/ifc/base/ifcvf.c +100 -0
index 30bb8cb..869ddd6 100644
@@ -233,6 +233,106 @@
 	}
 }
 
+int
+ifcvf_enable_vring_hw(struct ifcvf_hw *hw, int i)
+{
+	struct ifcvf_pci_common_cfg *cfg;
+	u8 *lm_cfg;
+	u16 notify_off;
+	int msix_vector;
+
+	if (i >= (int)hw->nr_vring)
+		return -1;
+
+	cfg = hw->common_cfg;
+	if (!cfg) {
+		RTE_LOG(ERR, PMD, "common_cfg in HW is NULL.\n");
+		return -1;
+	}
+
+	ifcvf_enable_mq(hw);
+
+	IFCVF_WRITE_REG16(i, &cfg->queue_select);
+	msix_vector = IFCVF_READ_REG16(&cfg->queue_msix_vector);
+	if (msix_vector != (i + 1)) {
+		IFCVF_WRITE_REG16(i + 1, &cfg->queue_msix_vector);
+		msix_vector = IFCVF_READ_REG16(&cfg->queue_msix_vector);
+		if (msix_vector == IFCVF_MSI_NO_VECTOR) {
+			RTE_LOG(ERR, PMD, "queue %d, msix vec alloc failed\n",
+				i);
+			return -1;
+		}
+	}
+
+	io_write64_twopart(hw->vring[i].desc, &cfg->queue_desc_lo,
+			&cfg->queue_desc_hi);
+	io_write64_twopart(hw->vring[i].avail, &cfg->queue_avail_lo,
+			&cfg->queue_avail_hi);
+	io_write64_twopart(hw->vring[i].used, &cfg->queue_used_lo,
+			&cfg->queue_used_hi);
+	IFCVF_WRITE_REG16(hw->vring[i].size, &cfg->queue_size);
+
+	lm_cfg = hw->lm_cfg;
+	if (lm_cfg) {
+		if (hw->device_type == IFCVF_BLK)
+			*(u32 *)(lm_cfg + IFCVF_LM_RING_STATE_OFFSET +
+				i * IFCVF_LM_CFG_SIZE) =
+				(u32)hw->vring[i].last_avail_idx |
+				((u32)hw->vring[i].last_used_idx << 16);
+		else
+			*(u32 *)(lm_cfg + IFCVF_LM_RING_STATE_OFFSET +
+				(i / 2) * IFCVF_LM_CFG_SIZE +
+				(i % 2) * 4) =
+				(u32)hw->vring[i].last_avail_idx |
+				((u32)hw->vring[i].last_used_idx << 16);
+	}
+
+	notify_off = IFCVF_READ_REG16(&cfg->queue_notify_off);
+	hw->notify_addr[i] = (void *)((u8 *)hw->notify_base +
+			notify_off * hw->notify_off_multiplier);
+	IFCVF_WRITE_REG16(1, &cfg->queue_enable);
+
+	return 0;
+}
+
+void
+ifcvf_disable_vring_hw(struct ifcvf_hw *hw, int i)
+{
+	struct ifcvf_pci_common_cfg *cfg;
+	u32 ring_state;
+	u8 *lm_cfg;
+
+	if (i >= (int)hw->nr_vring)
+		return;
+
+	cfg = hw->common_cfg;
+	if (!cfg) {
+		RTE_LOG(ERR, PMD, "common_cfg in HW is NULL.\n");
+		return;
+	}
+
+	IFCVF_WRITE_REG16(i, &cfg->queue_select);
+	IFCVF_WRITE_REG16(0, &cfg->queue_enable);
+
+	lm_cfg = hw->lm_cfg;
+	if (lm_cfg) {
+		if (hw->device_type == IFCVF_BLK) {
+			ring_state = *(u32 *)(lm_cfg +
+					IFCVF_LM_RING_STATE_OFFSET +
+					i * IFCVF_LM_CFG_SIZE);
+			hw->vring[i].last_avail_idx =
+				(u16)(ring_state & IFCVF_16_BIT_MASK);
+		} else {
+			ring_state = *(u32 *)(lm_cfg +
+					IFCVF_LM_RING_STATE_OFFSET +
+					(i / 2) * IFCVF_LM_CFG_SIZE +
+					(i % 2) * 4);
+			hw->vring[i].last_avail_idx = (u16)(ring_state >> 16);
+		}
+		hw->vring[i].last_used_idx = (u16)(ring_state >> 16);
+	}
+}
+
 STATIC int
 ifcvf_hw_enable(struct ifcvf_hw *hw)
 {
drivers/vdpa/ifc/base/ifcvf.h +6 -0
index 1e133c0..3726da7 100644
@@ -164,6 +164,12 @@ struct ifcvf_hw {
 ifcvf_get_features(struct ifcvf_hw *hw);
 
 int
+ifcvf_enable_vring_hw(struct ifcvf_hw *hw, int i);
+
+void
+ifcvf_disable_vring_hw(struct ifcvf_hw *hw, int i);
+
+int
 ifcvf_start_hw(struct ifcvf_hw *hw);
 
 void
drivers/vdpa/ifc/ifcvf_vdpa.c +78 -15

124 lines not on this page · raw message

Signature · 1 line
1.8.3.1
author
#164 lore raw

PATCH v9 7/12 vdpa/ifc: change internal function name

In reply to Andy Pei

Change internal function name "find_internal_resource_by_dev" to "find_internal_resource_by_pci_dev".

---
 drivers/vdpa/ifc/ifcvf_vdpa.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/vdpa/ifc/ifcvf_vdpa.c +2 -2
index 9c49f9c..73d04ed 100644
@@ -121,7 +121,7 @@ struct rte_vdpa_dev_info {
 }
 
 static struct internal_list *
-find_internal_resource_by_dev(struct rte_pci_device *pdev)
+find_internal_resource_by_pci_dev(struct rte_pci_device *pdev)
 {
 	int found = 0;
 	struct internal_list *list;
@@ -1746,7 +1746,7 @@ struct rte_vdpa_dev_info dev_info[] = {
 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 		return 0;
 
-	list = find_internal_resource_by_dev(pci_dev);
+	list = find_internal_resource_by_pci_dev(pci_dev);
 	if (list == NULL) {
 		DRV_LOG(ERR, "Invalid device: %s", pci_dev->name);
 		return -1;
Signature · 1 line
1.8.3.1
author
#165 lore raw

PATCH v9 8/12 vdpa/ifc: add internal API to get device

In reply to Andy Pei

Add new internal API "find_internal_resource_by_rte_dev" to get device.

---
 drivers/vdpa/ifc/ifcvf_vdpa.c | 28 ++++++++++++++++++++++++++--
 1 file changed, 26 insertions(+), 2 deletions(-)
drivers/vdpa/ifc/ifcvf_vdpa.c +26 -2
index 73d04ed..c16e263 100644
@@ -144,6 +144,29 @@ struct rte_vdpa_dev_info {
 	return list;
 }
 
+static struct internal_list *
+find_internal_resource_by_rte_dev(struct rte_device *rte_dev)
+{
+	int found = 0;
+	struct internal_list *list;
+
+	pthread_mutex_lock(&internal_list_lock);
+
+	TAILQ_FOREACH(list, &internal_list, next) {
+		if (rte_dev == &list->internal->pdev->device) {
+			found = 1;
+			break;
+		}
+	}
+
+	pthread_mutex_unlock(&internal_list_lock);
+
+	if (!found)
+		return NULL;
+
+	return list;
+}
+
 static int
 ifcvf_vfio_setup(struct ifcvf_internal *internal)
 {
@@ -1398,10 +1421,11 @@ struct rte_vdpa_dev_info {
 {
 	struct ifcvf_internal *internal;
 	struct internal_list *list;
+	struct rte_device *rte_dev = vdev->device;
 
-	list = find_internal_resource_by_vdev(vdev);
+	list = find_internal_resource_by_rte_dev(rte_dev);
 	if (list == NULL) {
-		DRV_LOG(ERR, "Invalid vDPA device: %p", vdev);
+		DRV_LOG(ERR, "Invalid rte device: %p", rte_dev);
 		return -1;
 	}
 
Signature · 1 line
1.8.3.1
author
#166 lore raw

PATCH v9 9/12 vdpa/ifc: improve internal list logic

In reply to Andy Pei

Insert internal list element to internal list before register vdpa device, in order to call vdpa ops during vdpa device registration.

---
 drivers/vdpa/ifc/ifcvf_vdpa.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
drivers/vdpa/ifc/ifcvf_vdpa.c +7 -4
index c16e263..8dfd493 100644
@@ -1737,17 +1737,20 @@ struct rte_vdpa_dev_info dev_info[] = {
 	}
 	internal->sw_lm = sw_fallback_lm;
 
+	pthread_mutex_lock(&internal_list_lock);
+	TAILQ_INSERT_TAIL(&internal_list, list, next);
+	pthread_mutex_unlock(&internal_list_lock);
+
 	internal->vdev = rte_vdpa_register_device(&pci_dev->device,
 				dev_info[internal->hw.device_type].ops);
 	if (internal->vdev == NULL) {
 		DRV_LOG(ERR, "failed to register device %s", pci_dev->name);
+		pthread_mutex_lock(&internal_list_lock);
+		TAILQ_REMOVE(&internal_list, list, next);
+		pthread_mutex_unlock(&internal_list_lock);
 		goto error;
 	}
 
-	pthread_mutex_lock(&internal_list_lock);
-	TAILQ_INSERT_TAIL(&internal_list, list, next);
-	pthread_mutex_unlock(&internal_list_lock);
-
 	rte_atomic32_set(&internal->started, 1);
 	update_datapath(internal);
 
Signature · 1 line
1.8.3.1
Reviewed-by on v9 9/12
#167 lore raw

RE: [PATCH v9 09/12] vdpa/ifc: improve internal list logic

Quoted 49 lines

-----Original Message-----
From: Pei, Andy <[email protected]>
Sent: Wednesday, October 19, 2022 4:41 PM
To: [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: [PATCH v9 09/12] vdpa/ifc: improve internal list logic

Insert internal list element to internal list before register vdpa device, in order to call vdpa ops during vdpa device registration.

---
 drivers/vdpa/ifc/ifcvf_vdpa.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
drivers/vdpa/ifc/ifcvf_vdpa.c +7 -5
index c16e263..8dfd493 100644
@@ -1737,17 +1737,20 @@ struct rte_vdpa_dev_info dev_info[] = {
 	}
 	internal->sw_lm = sw_fallback_lm;
 
+	pthread_mutex_lock(&internal_list_lock);
+	TAILQ_INSERT_TAIL(&internal_list, list, next);
+	pthread_mutex_unlock(&internal_list_lock);
+
 	internal->vdev = rte_vdpa_register_device(&pci_dev->device,
 				dev_info[internal->hw.device_type].ops);
 	if (internal->vdev == NULL) {
 		DRV_LOG(ERR, "failed to register device %s", pci_dev->name);
+		pthread_mutex_lock(&internal_list_lock);
+		TAILQ_REMOVE(&internal_list, list, next);
+		pthread_mutex_unlock(&internal_list_lock);
 		goto error;
 	}
 
-	pthread_mutex_lock(&internal_list_lock);
-	TAILQ_INSERT_TAIL(&internal_list, list, next);
-	pthread_mutex_unlock(&internal_list_lock);
-
 	rte_atomic32_set(&internal->started, 1);
 	update_datapath(internal);
 
--
Quoted 1 line

1.8.3.1

author
#168 lore raw

RE: [PATCH v9 09/12] vdpa/ifc: improve internal list logic

Hi Chenbo,

Thanks.

Quoted 59 lines

-----Original Message-----
From: Xia, Chenbo <[email protected]>
Sent: Thursday, October 20, 2022 11:22 AM
To: Pei, Andy <[email protected]>; [email protected]
Cc: Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: RE: [PATCH v9 09/12] vdpa/ifc: improve internal list logic

Quoted 48 lines

-----Original Message-----
From: Pei, Andy <[email protected]>
Sent: Wednesday, October 19, 2022 4:41 PM
To: [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: [PATCH v9 09/12] vdpa/ifc: improve internal list logic

Insert internal list element to internal list before register vdpa device, in order to call vdpa ops during vdpa device registration.

---
 drivers/vdpa/ifc/ifcvf_vdpa.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
(quoted diff) +0 -0

b/drivers/vdpa/ifc/ifcvf_vdpa.c index c16e263..8dfd493 100644

drivers/vdpa/ifc/ifcvf_vdpa.c +7 -5
@@ -1737,17 +1737,20 @@ struct rte_vdpa_dev_info dev_info[] = {
 	}
 	internal->sw_lm = sw_fallback_lm;
 
+	pthread_mutex_lock(&internal_list_lock);
+	TAILQ_INSERT_TAIL(&internal_list, list, next);
+	pthread_mutex_unlock(&internal_list_lock);
+
 	internal->vdev = rte_vdpa_register_device(&pci_dev->device,
 				dev_info[internal->hw.device_type].ops);
 	if (internal->vdev == NULL) {
 		DRV_LOG(ERR, "failed to register device %s", pci_dev->name);
+		pthread_mutex_lock(&internal_list_lock);
+		TAILQ_REMOVE(&internal_list, list, next);
+		pthread_mutex_unlock(&internal_list_lock);
 		goto error;
 	}
 
-	pthread_mutex_lock(&internal_list_lock);
-	TAILQ_INSERT_TAIL(&internal_list, list, next);
-	pthread_mutex_unlock(&internal_list_lock);
-
 	rte_atomic32_set(&internal->started, 1);
 	update_datapath(internal);
 
--
Quoted 1 line

1.8.3.1

author
#169 lore raw

PATCH v9 10/12 vhost: add type to rte vdpa device

In reply to Andy Pei

Add type to rte_vdpa_device to store device type. Call vdpa ops get_dev_type to fill type when register vdpa device.

---
 lib/vhost/socket.c      | 15 +--------------
 lib/vhost/vdpa.c        | 15 +++++++++++++++
 lib/vhost/vdpa_driver.h |  2 ++
 3 files changed, 18 insertions(+), 14 deletions(-)
lib/vhost/socket.c +1 -14
index 608ae57..863a6f6 100644
@@ -627,7 +627,6 @@ struct rte_vdpa_device *
 {
 	struct vhost_user_socket *vsocket;
 	struct rte_vdpa_device *vdpa_dev;
-	uint32_t vdpa_type = 0;
 	int ret = 0;
 
 	pthread_mutex_lock(&vhost_user.mutex);
@@ -644,19 +643,7 @@ struct rte_vdpa_device *
 		goto unlock_exit;
 	}
 
-	if (vdpa_dev->ops->get_dev_type) {
-		ret = vdpa_dev->ops->get_dev_type(vdpa_dev, &vdpa_type);
-		if (ret) {
-			VHOST_LOG_CONFIG(path, ERR,
-				"failed to get vdpa dev type for socket file.\n");
-			ret = -1;
-			goto unlock_exit;
-		}
-	} else {
-		vdpa_type = RTE_VHOST_VDPA_DEVICE_TYPE_NET;
-	}
-
-	*type = vdpa_type;
+	*type = vdpa_dev->type;
 
 unlock_exit:
 	pthread_mutex_unlock(&vhost_user.mutex);
lib/vhost/vdpa.c +15 -0
index bb82857..577cb00 100644
@@ -73,6 +73,7 @@ struct rte_vdpa_device *
 		struct rte_vdpa_dev_ops *ops)
 {
 	struct rte_vdpa_device *dev;
+	int ret = 0;
 
 	if (ops == NULL)
 		return NULL;
@@ -101,6 +102,20 @@ struct rte_vdpa_device *
 
 	dev->device = rte_dev;
 	dev->ops = ops;
+
+	if (ops->get_dev_type) {
+		ret = ops->get_dev_type(dev, &dev->type);
+		if (ret) {
+			VHOST_LOG_CONFIG(rte_dev->name, ERR,
+					 "Failed to get vdpa dev type.\n");
+			ret = -1;
+			goto out_unlock;
+		}
+	} else {
+		/** by default, we assume vdpa device is a net device */
+		dev->type = RTE_VHOST_VDPA_DEVICE_TYPE_NET;
+	}
+
 	TAILQ_INSERT_TAIL(&vdpa_device_list, dev, next);
 out_unlock:
 	rte_spinlock_unlock(&vdpa_device_list_lock);
lib/vhost/vdpa_driver.h +2 -0
index 8b88a53..8db4ab9 100644
@@ -92,6 +92,8 @@ struct rte_vdpa_device {
 	struct rte_device *device;
 	/** vdpa device operations */
 	struct rte_vdpa_dev_ops *ops;
+	/** vdpa device type: net, blk... */
+	uint32_t type;
 };
 
 /**
Signature · 1 line
1.8.3.1
author
#170 lore raw

PATCH v9 11/12 vhost: vDPA blk device gets ready when the first queue is ready

In reply to Andy Pei

When boot from virtio blk device, seabios in QEMU only enables one queue. To work in this scenario, vDPA BLK device back-end configure device when the first queue is ready.

---
 lib/vhost/vhost_user.c | 33 +++++++++++++++++----------------
 1 file changed, 17 insertions(+), 16 deletions(-)
lib/vhost/vhost_user.c +17 -16
index cd65257..e0ff79d 100644
@@ -1441,11 +1441,14 @@
 }
 
 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
+#define VIRTIO_BLK_NUM_VQS_TO_BE_READY 1u
 
 static int
 virtio_is_ready(struct virtio_net *dev)
 {
+	struct rte_vdpa_device *vdpa_dev;
 	struct vhost_virtqueue *vq;
+	uint32_t vdpa_type;
 	uint32_t i, nr_vring = dev->nr_vring;
 
 	if (dev->flags & VIRTIO_DEV_READY)
@@ -1454,13 +1457,22 @@
 	if (!dev->nr_vring)
 		return 0;
 
-	if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
-		nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
+	vdpa_dev = dev->vdpa_dev;
+	if (vdpa_dev)
+		vdpa_type = vdpa_dev->type;
+	else
+		vdpa_type = -1;
 
-		if (dev->nr_vring < nr_vring)
-			return 0;
+	if (vdpa_type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK) {
+		nr_vring = VIRTIO_BLK_NUM_VQS_TO_BE_READY;
+	} else {
+		if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET)
+			nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
 	}
 
+	if (dev->nr_vring < nr_vring)
+		return 0;
+
 	for (i = 0; i < nr_vring; i++) {
 		vq = dev->virtqueue[i];
 
@@ -2958,7 +2970,6 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	int ret;
 	int unlock_required = 0;
 	bool handled;
-	uint32_t vdpa_type = 0;
 	uint32_t request;
 	uint32_t i;
 
@@ -3170,17 +3181,7 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	if (!vdpa_dev)
 		goto out;
 
-	if (vdpa_dev->ops->get_dev_type) {
-		ret = vdpa_dev->ops->get_dev_type(vdpa_dev, &vdpa_type);
-		if (ret) {
-			VHOST_LOG_CONFIG(dev->ifname, ERR, "failed to get vdpa dev type.\n");
-			ret = -1;
-			goto out;
-		}
-	} else {
-		vdpa_type = RTE_VHOST_VDPA_DEVICE_TYPE_NET;
-	}
-	if (vdpa_type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
+	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
 		&& request != VHOST_USER_SET_VRING_CALL)
 		goto out;
 
Signature · 1 line
1.8.3.1
author
#171 lore raw

PATCH v9 12/12 vhost: improve vDPA blk device configure condition

In reply to Andy Pei

To support multi-queue, configure device after call fd of all queues are set.

---
 lib/vhost/vhost_user.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)
lib/vhost/vhost_user.c +10 -3
index e0ff79d..9902ae9 100644
@@ -2972,6 +2972,7 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	bool handled;
 	uint32_t request;
 	uint32_t i;
+	uint16_t blk_call_fd;
 
 	dev = get_device(vid);
 	if (dev == NULL)
@@ -3181,9 +3182,15 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	if (!vdpa_dev)
 		goto out;
 
-	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
-		&& request != VHOST_USER_SET_VRING_CALL)
-		goto out;
+	if (vdpa_dev->type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK) {
+		if (request == VHOST_USER_SET_VRING_CALL) {
+			blk_call_fd = ctx.msg.payload.u64 & VHOST_USER_VRING_IDX_MASK;
+			if (blk_call_fd != dev->nr_vring - 1)
+				goto out;
+		} else {
+			goto out;
+		}
+	}
 
 	if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
 		if (vdpa_dev->ops->dev_conf(dev->vid))
Signature · 1 line
1.8.3.1
#172 lore raw

RE: [PATCH v9 00/12] vdpa/ifc: add multi queue support

In reply to Andy Pei
Quoted 66 lines

-----Original Message-----
From: Pei, Andy <[email protected]>
Sent: Wednesday, October 19, 2022 4:41 PM
To: [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: [PATCH v9 00/12] vdpa/ifc: add multi queue support

v9:
fix some commit message.

v8:
change "vdpa_device_type" in "rte_vdpa_device" to "type".

v7:
Fill vdpa_device_type in vdpa device registration.

v6:
Add vdpa_device_type to rte_vdpa_device to store vDPA device type.

v5:
fix some commit message.
rework some code logic.

v4:
fix some commit message.
add some commets to code.
fix some code to reduce confusion.

v3:

 rename device ID macro name.

fix some patch title and commit message.
delete some used marco.
rework some code logic.

v2:
fix some coding style issue.
support dynamic enable/disable queue at run time.

Andy Pei (10):
vdpa/ifc: add multi-queue support
vdpa/ifc: set max queues based on virtio spec vdpa/ifc: write queue count to MQ register
vdpa/ifc: only configure enabled queue
vdpa/ifc: change internal function name
vdpa/ifc: add internal API to get device
vdpa/ifc: improve internal list logic
vhost: add type to rte vdpa device
vhost: vDPA blk device gets ready when the first queue is ready
vhost: improve vDPA blk device configure condition

Huang Wei (2):
vdpa/ifc: add new device ID for legacy network device vdpa/ifc: support dynamic enable/disable queue

 drivers/vdpa/ifc/base/ifcvf.c | 144 ++++++++++++++++++++++++++++++++
 drivers/vdpa/ifc/base/ifcvf.h |  16 +++-
 drivers/vdpa/ifc/ifcvf_vdpa.c | 185 +++++++++++++++++++++++++++++++++++--
-----
 lib/vhost/socket.c            |  15 +---
 lib/vhost/vdpa.c              |  15 ++++
 lib/vhost/vdpa_driver.h       |   2 +
 lib/vhost/vhost_user.c        |  38 +++++----
 7 files changed, 354 insertions(+), 61 deletions(-)
Quoted 2 lines

--
1.8.3.1

Change title ' vhost: add type to rte vdpa device ' to ' vhost: add type to vDPA device '

Series applied to next-virtio/main, Thanks

author
#173 lore raw

RE: [PATCH v9 00/12] vdpa/ifc: add multi queue support

Quoted 78 lines

-----Original Message-----
From: Xia, Chenbo <[email protected]>
Sent: Wednesday, October 26, 2022 5:00 PM
To: Pei, Andy <[email protected]>; [email protected]
Cc: Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: RE: [PATCH v9 00/12] vdpa/ifc: add multi queue support

Quoted 52 lines

-----Original Message-----
From: Pei, Andy <[email protected]>
Sent: Wednesday, October 19, 2022 4:41 PM
To: [email protected]
Cc: Xia, Chenbo <[email protected]>; Xu, Rosen <[email protected]>; Huang, Wei <[email protected]>; Cao, Gang <[email protected]>; [email protected]
Subject: [PATCH v9 00/12] vdpa/ifc: add multi queue support

v9:
fix some commit message.

v8:
change "vdpa_device_type" in "rte_vdpa_device" to "type".

v7:
Fill vdpa_device_type in vdpa device registration.

v6:
Add vdpa_device_type to rte_vdpa_device to store vDPA device type.

v5:
fix some commit message.
rework some code logic.

v4:
fix some commit message.
add some commets to code.
fix some code to reduce confusion.

v3:

 rename device ID macro name.

fix some patch title and commit message.
delete some used marco.
rework some code logic.

v2:
fix some coding style issue.
support dynamic enable/disable queue at run time.

Andy Pei (10):
vdpa/ifc: add multi-queue support
vdpa/ifc: set max queues based on virtio spec vdpa/ifc: write queue count to MQ register
vdpa/ifc: only configure enabled queue
vdpa/ifc: change internal function name
vdpa/ifc: add internal API to get device
vdpa/ifc: improve internal list logic
vhost: add type to rte vdpa device
vhost: vDPA blk device gets ready when the first queue is ready
vhost: improve vDPA blk device configure condition

Quoted 5 lines

Huang Wei (2):
vdpa/ifc: add new device ID for legacy network device vdpa/ifc: support dynamic enable/disable queue

 drivers/vdpa/ifc/base/ifcvf.c | 144

++++++++++++++++++++++++++++++++

Quoted 10 lines

drivers/vdpa/ifc/base/ifcvf.h | 16 +++-
drivers/vdpa/ifc/ifcvf_vdpa.c | 185
+++++++++++++++++++++++++++++++++++--
-----

 lib/vhost/socket.c            |  15 +---
 lib/vhost/vdpa.c              |  15 ++++
 lib/vhost/vdpa_driver.h       |   2 +
 lib/vhost/vhost_user.c        |  38 +++++----
 7 files changed, 354 insertions(+), 61 deletions(-)
Quoted 6 lines
Quoted 2 lines

--
1.8.3.1

Change title ' vhost: add type to rte vdpa device ' to ' vhost: add type to vDPA device '

OK. Thanks Chenbo.

Quoted 1 line

Series applied to next-virtio/main, Thanks

author
#174 lore raw

PATCH 2/8 vdpa/ifc: add multi queue suppoort

In reply to Andy Pei

Enable VHOST_USER_PROTOCOL_F_MQ feature. ExposeIFCVF_MQ_OFFSET register to enable multi queue.

---
 drivers/vdpa/ifc/base/ifcvf.c | 5 +++++
 drivers/vdpa/ifc/base/ifcvf.h | 2 ++
 drivers/vdpa/ifc/ifcvf_vdpa.c | 1 +
 3 files changed, 8 insertions(+)
drivers/vdpa/ifc/base/ifcvf.c +5 -0
index f1e1474..34c8226 100644
@@ -90,6 +90,11 @@
 	if (!hw->lm_cfg)
 		WARNINGOUT("HW support live migration not support!\n");
 
+	if (hw->mem_resource[4].addr)
+		hw->mq_cfg = hw->mem_resource[4].addr + IFCVF_MQ_OFFSET;
+	else
+		hw->mq_cfg = NULL;
+
 	if (hw->common_cfg == NULL || hw->notify_base == NULL ||
 			hw->isr == NULL || hw->dev_cfg == NULL) {
 		DEBUGOUT("capability incomplete\n");
drivers/vdpa/ifc/base/ifcvf.h +2 -0
index 7ede738..ad505f1 100644
@@ -50,6 +50,7 @@
 
 #define IFCVF_LM_CFG_SIZE		0x40
 #define IFCVF_LM_RING_STATE_OFFSET	0x20
+#define IFCVF_MQ_OFFSET			0x28
 
 #define IFCVF_LM_LOGGING_CTRL		0x0
 
@@ -149,6 +150,7 @@ struct ifcvf_hw {
 	u16    *notify_base;
 	u16    *notify_addr[IFCVF_MAX_QUEUES * 2];
 	u8     *lm_cfg;
+	u8     *mq_cfg;
 	struct vring_info vring[IFCVF_MAX_QUEUES * 2];
 	u8 nr_vring;
 	int device_type;
drivers/vdpa/ifc/ifcvf_vdpa.c +1 -0
index 61d0250..2d165c0 100644
@@ -1248,6 +1248,7 @@ struct rte_vdpa_dev_info {
 		 1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD | \
 		 1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER | \
 		 1ULL << VHOST_USER_PROTOCOL_F_LOG_SHMFD | \
+		 1ULL << VHOST_USER_PROTOCOL_F_MQ | \
 		 1ULL << VHOST_USER_PROTOCOL_F_STATUS)
 
 #define VDPA_BLK_PROTOCOL_FEATURES \
Signature · 1 line
1.8.3.1
author
#175 lore raw

PATCH 3/8 vdpa/ifc: set max queues according to HW spec

In reply to Andy Pei

Set max_queues according to virtio HW spec.
For virtio BLK device, set max_queues to the value of "num_queues". "num_queues" is element of struct virtio_blk_config.

---
 drivers/vdpa/ifc/base/ifcvf.h |  2 +-
 drivers/vdpa/ifc/ifcvf_vdpa.c | 18 +++++++++++++++++-
 2 files changed, 18 insertions(+), 2 deletions(-)
drivers/vdpa/ifc/base/ifcvf.h +1 -1
index ad505f1..c17bf2a 100644
@@ -21,7 +21,7 @@
 #define IFCVF_NET_DEVICE_ID                 0x0001
 #define IFCVF_BLK_DEVICE_ID                 0x0002
 
-#define IFCVF_MAX_QUEUES		1
+#define IFCVF_MAX_QUEUES		32
 
 #ifndef VIRTIO_F_IOMMU_PLATFORM
 #define VIRTIO_F_IOMMU_PLATFORM		33
drivers/vdpa/ifc/ifcvf_vdpa.c +17 -1
index 2d165c0..34aea6c 100644
@@ -26,6 +26,18 @@
 
 #include "base/ifcvf.h"
 
+/**
+** RTE_MAX() and RTE_MIN() cannot be used since braced-group within
+** expression allowed only inside a function, but MAX() is used as
+** a number of elements in array.
+**/
+#ifndef MAX
+#define MAX(v1, v2)	((v1) > (v2) ? (v1) : (v2))
+#endif
+#ifndef MIN
+#define MIN(v1, v2)	((v1) < (v2) ? (v1) : (v2))
+#endif
+
 RTE_LOG_REGISTER(ifcvf_vdpa_logtype, pmd.vdpa.ifcvf, NOTICE);
 #define DRV_LOG(level, fmt, args...) \
 	rte_log(RTE_LOG_ ## level, ifcvf_vdpa_logtype, \
@@ -1559,7 +1571,6 @@ struct rte_vdpa_dev_info dev_info[] = {
 	}
 
 	internal->configured = 0;
-	internal->max_queues = IFCVF_MAX_QUEUES;
 	features = ifcvf_get_features(&internal->hw);
 
 	device_id = ifcvf_pci_get_device_type(pci_dev);
@@ -1570,6 +1581,8 @@ struct rte_vdpa_dev_info dev_info[] = {
 
 	if (device_id == VIRTIO_ID_NET) {
 		internal->hw.device_type = IFCVF_NET;
+		internal->max_queues = MIN(IFCVF_MAX_QUEUES,
+			(internal->hw.common_cfg->num_queues - 1)/2);
 		internal->features = features &
 					~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
 		internal->features |= dev_info[IFCVF_NET].features;
@@ -1609,6 +1622,9 @@ struct rte_vdpa_dev_info dev_info[] = {
 			internal->hw.blk_cfg->geometry.sectors);
 		DRV_LOG(DEBUG, "num_queues: 0x%08x",
 			internal->hw.blk_cfg->num_queues);
+
+		internal->max_queues = MIN(IFCVF_MAX_QUEUES,
+			internal->hw.blk_cfg->num_queues);
 	}
 
 	list->internal = internal;
Signature · 1 line
1.8.3.1
author
#176 lore raw

PATCH 4/8 vdpa/ifc: write queue count to MQ register

In reply to Andy Pei

Write queue count to IFCVF_MQ_OFFSET register
to enable multi queue feature.

---
 drivers/vdpa/ifc/base/ifcvf.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)
drivers/vdpa/ifc/base/ifcvf.c +14 -0
index 34c8226..1b50df6 100644
@@ -198,6 +198,19 @@
 	IFCVF_WRITE_REG32(val >> 32, hi);
 }
 
+STATIC void
+ifcvf_enable_multiqueue(struct ifcvf_hw *hw, u16 nr_queue_pair)
+{
+	u8 *mq_cfg;
+
+	if (hw->device_type == IFCVF_NET)
+		nr_queue_pair = (nr_queue_pair + 1) / 2;
+
+	mq_cfg = hw->mq_cfg;
+	if (mq_cfg)
+		*(u32 *)mq_cfg = nr_queue_pair;
+}
+
 STATIC int
 ifcvf_hw_enable(struct ifcvf_hw *hw)
 {
@@ -215,6 +228,7 @@
 		return -1;
 	}
 
+	ifcvf_enable_multiqueue(hw, hw->nr_vring);
 	for (i = 0; i < hw->nr_vring; i++) {
 		IFCVF_WRITE_REG16(i, &cfg->queue_select);
 		io_write64_twopart(hw->vring[i].desc, &cfg->queue_desc_lo,
Signature · 1 line
1.8.3.1
author
#177 lore raw

PATCH 5/8 vdpa/ifc: only configure enabled queue

In reply to Andy Pei

when configure the hardware queue, we only configure queues which have been enabled by vhost.

---
 drivers/vdpa/ifc/base/ifcvf.c | 5 ++++-
 drivers/vdpa/ifc/ifcvf_vdpa.c | 4 ++++
 2 files changed, 8 insertions(+), 1 deletion(-)
drivers/vdpa/ifc/base/ifcvf.c +4 -1
index 1b50df6..ca5f677 100644
@@ -230,6 +230,8 @@
 
 	ifcvf_enable_multiqueue(hw, hw->nr_vring);
 	for (i = 0; i < hw->nr_vring; i++) {
+		if (!hw->vring[i].enable)
+			continue;
 		IFCVF_WRITE_REG16(i, &cfg->queue_select);
 		io_write64_twopart(hw->vring[i].desc, &cfg->queue_desc_lo,
 				&cfg->queue_desc_hi);
@@ -264,7 +266,8 @@
 		notify_off = IFCVF_READ_REG16(&cfg->queue_notify_off);
 		hw->notify_addr[i] = (void *)((u8 *)hw->notify_base +
 				notify_off * hw->notify_off_multiplier);
-		IFCVF_WRITE_REG16(1, &cfg->queue_enable);
+		if (hw->vring[i].enable)
+			IFCVF_WRITE_REG16(1, &cfg->queue_enable);
 	}
 
 	return 0;
drivers/vdpa/ifc/ifcvf_vdpa.c +4 -0
index 34aea6c..a62bcec 100644
@@ -290,6 +290,8 @@ struct rte_vdpa_dev_info {
 	rte_vhost_get_negotiated_features(vid, &hw->req_features);
 
 	for (i = 0; i < nr_vring; i++) {
+		if (!hw->vring[i].enable)
+			continue;
 		rte_vhost_get_vhost_vring(vid, i, &vq);
 		gpa = hva_to_gpa(vid, (uint64_t)(uintptr_t)vq.desc);
 		if (gpa == 0) {
@@ -505,6 +507,8 @@ struct rte_vdpa_dev_info {
 
 	vring.kickfd = -1;
 	for (qid = 0; qid < q_num; qid++) {
+		if (!hw->vring[qid].enable)
+			continue;
 		ev.events = EPOLLIN | EPOLLPRI;
 		rte_vhost_get_vhost_vring(internal->vid, qid, &vring);
 		ev.data.u64 = qid | (uint64_t)vring.kickfd << 32;
Signature · 1 line
1.8.3.1
author
#178 lore raw

PATCH 6/8 vdpa/ifc: set vring state callback update data path

In reply to Andy Pei

To support multi queue, in the case that first queue is ready and device is configured, when more queues need to be configured, we just close and restart data path.
This also fix the situation that using set_vring_state callback to disable one queue will cause all vfio interrupts being disabled.

---
 drivers/vdpa/ifc/ifcvf_vdpa.c | 27 +++++++++------------------
 1 file changed, 9 insertions(+), 18 deletions(-)
drivers/vdpa/ifc/ifcvf_vdpa.c +9 -18
index a62bcec..94c8ef1 100644
@@ -1286,8 +1286,6 @@ struct rte_vdpa_dev_info {
 	struct internal_list *list;
 	struct ifcvf_internal *internal;
 	struct ifcvf_hw *hw;
-	struct ifcvf_pci_common_cfg *cfg;
-	int ret = 0;
 
 	vdev = rte_vhost_get_vdpa_device(vid);
 	list = find_internal_resource_by_vdev(vdev);
@@ -1303,27 +1301,20 @@ struct rte_vdpa_dev_info {
 	}
 
 	hw = &internal->hw;
-	if (!internal->configured)
-		goto exit;
 
-	cfg = hw->common_cfg;
-	IFCVF_WRITE_REG16(vring, &cfg->queue_select);
-	IFCVF_WRITE_REG16(!!state, &cfg->queue_enable);
+	hw->vring[vring].enable = !!state;
 
-	if (!state && hw->vring[vring].enable) {
-		ret = vdpa_disable_vfio_intr(internal);
-		if (ret)
-			return ret;
-	}
+	if (!internal->configured)
+		goto exit;
 
-	if (state && !hw->vring[vring].enable) {
-		ret = vdpa_enable_vfio_intr(internal, false);
-		if (ret)
-			return ret;
-	}
+	/* close data path */
+	rte_atomic32_set(&internal->dev_attached, 0);
+	update_datapath(internal);
 
+	/* restart data path */
+	rte_atomic32_set(&internal->dev_attached, 1);
+	update_datapath(internal);
 exit:
-	hw->vring[vring].enable = !!state;
 	return 0;
 }
 
Signature · 1 line
1.8.3.1
author
#179 lore raw

PATCH 7/8 vhost: configure device when any queue is ready for BLK device

In reply to Andy Pei

When boot from virtio blk device, seabois in QEMU only enables one queue. To work in this scenario, vDPA BLK device back-end conf_dev when any queue is ready.

---
 lib/vhost/vhost_user.c | 56 +++++++++++++++++++++++++++++++++++++++-----------
 1 file changed, 44 insertions(+), 12 deletions(-)
lib/vhost/vhost_user.c +44 -12
index 4ad28ba..b65fba3 100644
@@ -1451,6 +1451,25 @@
 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
 
 static int
+virtio_has_queue_ready(struct virtio_net *dev)
+{
+	struct vhost_virtqueue *vq;
+	uint32_t i, nr_vring = dev->nr_vring;
+
+	if (!dev->nr_vring)
+		return 0;
+
+	for (i = 0; i < nr_vring; i++) {
+		vq = dev->virtqueue[i];
+
+		if (vq_is_ready(dev, vq))
+			return 1;
+	}
+
+	return 0;
+}
+
+static int
 virtio_is_ready(struct virtio_net *dev)
 {
 	struct vhost_virtqueue *vq;
@@ -3167,9 +3186,33 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	if (unlock_required)
 		vhost_user_unlock_all_queue_pairs(dev);
 
-	if (ret != 0 || !virtio_is_ready(dev))
+	if (ret != 0)
 		goto out;
 
+	vdpa_dev = dev->vdpa_dev;
+	if (vdpa_dev) {
+		if (vdpa_dev->ops->get_dev_type) {
+			ret = vdpa_dev->ops->get_dev_type(vdpa_dev, &vdpa_type);
+			if (ret) {
+				VHOST_LOG_CONFIG(dev->ifname, ERR,
+					"failed to get vdpa dev type.\n");
+				ret = -1;
+				goto out;
+			}
+		} else {
+			vdpa_type = RTE_VHOST_VDPA_DEVICE_TYPE_NET;
+		}
+	}
+
+	if (!virtio_is_ready(dev)) {
+		if (vdpa_type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK) {
+			if (!virtio_has_queue_ready(dev))
+				goto out;
+		} else {
+			goto out;
+		}
+	}
+
 	/*
 	 * Virtio is now ready. If not done already, it is time
 	 * to notify the application it can process the rings and
@@ -3181,20 +3224,9 @@ static int is_vring_iotlb(struct virtio_net *dev,
 			dev->flags |= VIRTIO_DEV_RUNNING;
 	}
 
-	vdpa_dev = dev->vdpa_dev;
 	if (!vdpa_dev)
 		goto out;
 
-	if (vdpa_dev->ops->get_dev_type) {
-		ret = vdpa_dev->ops->get_dev_type(vdpa_dev, &vdpa_type);
-		if (ret) {
-			VHOST_LOG_CONFIG(dev->ifname, ERR, "failed to get vdpa dev type.\n");
-			ret = -1;
-			goto out;
-		}
-	} else {
-		vdpa_type = RTE_VHOST_VDPA_DEVICE_TYPE_NET;
-	}
 	if (vdpa_type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
 		&& request != VHOST_USER_SET_VRING_CALL)
 		goto out;
Signature · 1 line
1.8.3.1
author
#180 lore raw

PATCH 8/8 vhost: vDPA BLK devices configure device when all queue callfds are set

In reply to Andy Pei

In the virtio blk vDPA live migration use case, before the live migration process, QEMU will set call fd to vDPA back-end. QEMU and vDPA back-end stand by until live migration starts. During live migration process, QEMU sets kick fd and a new call fd. However, after the kick fd is set to the vDPA back-end, the vDPA back-end configures device and data path starts. The new call fd will cause some kind of "re-configuration", this kind of "re-configuration" cause IO drop.
After this patch, vDPA back-end configures device after kick fd and call fd are well set and make sure no IO drops. This patch only impact virtio blk vDPA device and does not impact net device.

---
 lib/vhost/vhost_user.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)
lib/vhost/vhost_user.c +10 -3
index b65fba3..568030a 100644
@@ -2994,6 +2994,7 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	uint32_t vdpa_type = 0;
 	uint32_t request;
 	uint32_t i;
+	uint16_t blk_call_fd;
 
 	dev = get_device(vid);
 	if (dev == NULL)
@@ -3227,9 +3228,15 @@ static int is_vring_iotlb(struct virtio_net *dev,
 	if (!vdpa_dev)
 		goto out;
 
-	if (vdpa_type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK
-		&& request != VHOST_USER_SET_VRING_CALL)
-		goto out;
+	if (vdpa_type == RTE_VHOST_VDPA_DEVICE_TYPE_BLK) {
+		if (request == VHOST_USER_SET_VRING_CALL) {
+			blk_call_fd = ctx.msg.payload.u64 & VHOST_USER_VRING_IDX_MASK;
+			if (blk_call_fd != dev->nr_vring - 1)
+				goto out;
+		} else {
+			goto out;
+		}
+	}
 
 	if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
 		if (vdpa_dev->ops->dev_conf(dev->vid))
Signature · 1 line
1.8.3.1