Skip to contentSkip to conversation

bugfix some driver issues

PATCH Superseded Longfang Liu 2 participants 10 messages Mar 13, 2025 – Mar 18, 2025

Subsystems (4)

v5 of 8 Superseded Latest is v8 →

lore
Share & tools
lore

Longfang Liu

author maintainer
#0 lore raw

v5 cover bugfix some driver issues

As the test scenarios for the live migration function become more and more extensive. Some previously undiscovered driver issues were found.
Update and fix through this patchset.

Change v4 -> v5
Update version matching strategy

Change v3 -> v4
Modify version matching scheme

Change v2 -> v3
Modify the magic digital field segment

Change v1 -> v2
Add fixes line for patch comment

Longfang Liu (5):
hisi_acc_vfio_pci: fix XQE dma address error
hisi_acc_vfio_pci: add eq and aeq interruption restore
hisi_acc_vfio_pci: bugfix cache write-back issue
hisi_acc_vfio_pci: bugfix the problem of uninstalling driver
hisi_acc_vfio_pci: bugfix live migration function without VF device
driver

 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 97 +++++++++++++++----
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.h    | 14 ++-
 2 files changed, 89 insertions(+), 22 deletions(-)
Signature · 1 line
2.24.0
author maintainer
#1 lore raw

PATCH v5 1/5 hisi_acc_vfio_pci: fix XQE dma address error

The dma addresses of EQE and AEQE are wrong after migration and results in guest kernel-mode encryption services failure. Comparing the definition of hardware registers, we found that there was an error when the data read from the register was combined into an address. Therefore, the address combination sequence needs to be corrected.

Even after fixing the above problem, we still have an issue where the Guest from an old kernel can get migrated to new kernel and may result in wrong data.

In order to ensure that the address is correct after migration, if an old magic number is detected, the dma address needs to be updated.

Fixes b0eed085903e ("hisi_acc_vfio_pci: Add support for VFIO live migration")
---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 41 ++++++++++++++++---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.h    | 14 ++++++-
 2 files changed, 47 insertions(+), 8 deletions(-)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +35 -6
index 451c639299eb..304dbdfa0e95 100644
@@ -350,6 +350,32 @@ static int vf_qm_func_stop(struct hisi_qm *qm)
 	return hisi_qm_mb(qm, QM_MB_CMD_PAUSE_QM, 0, 0, 0);
 }
 
+static int vf_qm_version_check(struct acc_vf_data *vf_data, struct device *dev)
+{
+	switch (vf_data->acc_magic) {
+	case ACC_DEV_MAGIC_V2:
+		if (vf_data->major_ver != ACC_DRV_MAJOR_VER) {
+			dev_info(dev, "migration driver version<%u.%u> not match!\n",
+				 vf_data->major_ver, vf_data->minor_ver);
+			return -EINVAL;
+		}
+		break;
+	case ACC_DEV_MAGIC_V1:
+		/* Correct dma address */
+		vf_data->eqe_dma = vf_data->qm_eqc_dw[QM_XQC_ADDR_HIGH];
+		vf_data->eqe_dma <<= QM_XQC_ADDR_OFFSET;
+		vf_data->eqe_dma |= vf_data->qm_eqc_dw[QM_XQC_ADDR_LOW];
+		vf_data->aeqe_dma = vf_data->qm_aeqc_dw[QM_XQC_ADDR_HIGH];
+		vf_data->aeqe_dma <<= QM_XQC_ADDR_OFFSET;
+		vf_data->aeqe_dma |= vf_data->qm_aeqc_dw[QM_XQC_ADDR_LOW];
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
 static int vf_qm_check_match(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 			     struct hisi_acc_vf_migration_file *migf)
 {
@@ -363,7 +389,8 @@ static int vf_qm_check_match(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 	if (migf->total_length < QM_MATCH_SIZE || hisi_acc_vdev->match_done)
 		return 0;
 
-	if (vf_data->acc_magic != ACC_DEV_MAGIC) {
+	ret = vf_qm_version_check(vf_data, dev);
+	if (ret) {
 		dev_err(dev, "failed to match ACC_DEV_MAGIC\n");
 		return -EINVAL;
 	}
@@ -418,7 +445,9 @@ static int vf_qm_get_match_data(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 	int vf_id = hisi_acc_vdev->vf_id;
 	int ret;
 
-	vf_data->acc_magic = ACC_DEV_MAGIC;
+	vf_data->acc_magic = ACC_DEV_MAGIC_V2;
+	vf_data->major_ver = ACC_DRV_MAJOR_VER;
+	vf_data->minor_ver = ACC_DRV_MINOR_VER;
 	/* Save device id */
 	vf_data->dev_id = hisi_acc_vdev->vf_dev->device;
 
@@ -496,12 +525,12 @@ static int vf_qm_read_data(struct hisi_qm *vf_qm, struct acc_vf_data *vf_data)
 		return -EINVAL;
 
 	/* Every reg is 32 bit, the dma address is 64 bit. */
-	vf_data->eqe_dma = vf_data->qm_eqc_dw[1];
+	vf_data->eqe_dma = vf_data->qm_eqc_dw[QM_XQC_ADDR_HIGH];
 	vf_data->eqe_dma <<= QM_XQC_ADDR_OFFSET;
-	vf_data->eqe_dma |= vf_data->qm_eqc_dw[0];
-	vf_data->aeqe_dma = vf_data->qm_aeqc_dw[1];
+	vf_data->eqe_dma |= vf_data->qm_eqc_dw[QM_XQC_ADDR_LOW];
+	vf_data->aeqe_dma = vf_data->qm_aeqc_dw[QM_XQC_ADDR_HIGH];
 	vf_data->aeqe_dma <<= QM_XQC_ADDR_OFFSET;
-	vf_data->aeqe_dma |= vf_data->qm_aeqc_dw[0];
+	vf_data->aeqe_dma |= vf_data->qm_aeqc_dw[QM_XQC_ADDR_LOW];
 
 	/* Through SQC_BT/CQC_BT to get sqc and cqc address */
 	ret = qm_get_sqc(vf_qm, &vf_data->sqc_dma);
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h +12 -2
index 245d7537b2bc..91002ceeebc1 100644
@@ -39,6 +39,9 @@
 #define QM_REG_ADDR_OFFSET	0x0004
 
 #define QM_XQC_ADDR_OFFSET	32U
+#define QM_XQC_ADDR_LOW	0x1
+#define QM_XQC_ADDR_HIGH	0x2
+
 #define QM_VF_AEQ_INT_MASK	0x0004
 #define QM_VF_EQ_INT_MASK	0x000c
 #define QM_IFC_INT_SOURCE_V	0x0020
@@ -50,10 +53,15 @@
 #define QM_EQC_DW0		0X8000
 #define QM_AEQC_DW0		0X8020
 
+#define ACC_DRV_MAJOR_VER 1
+#define ACC_DRV_MINOR_VER 0
+
+#define ACC_DEV_MAGIC_V1	0XCDCDCDCDFEEDAACC
+#define ACC_DEV_MAGIC_V2	0xAACCFEEDDECADEDE
+
 struct acc_vf_data {
 #define QM_MATCH_SIZE offsetofend(struct acc_vf_data, qm_rsv_state)
 	/* QM match information */
-#define ACC_DEV_MAGIC	0XCDCDCDCDFEEDAACC
 	u64 acc_magic;
 	u32 qp_num;
 	u32 dev_id;
@@ -61,7 +69,9 @@ struct acc_vf_data {
 	u32 qp_base;
 	u32 vf_qm_state;
 	/* QM reserved match information */
-	u32 qm_rsv_state[3];
+	u16 major_ver;
+	u16 minor_ver;
+	u32 qm_rsv_state[2];
 
 	/* QM RW regs */
 	u32 aeq_int_mask;
Signature · 1 line
2.24.0
Reviewed-by on v5 1/5
#2 lore raw

RE: [PATCH v5 1/5] hisi_acc_vfio_pci: fix XQE dma address error

Quoted 23 lines

-----Original Message-----
From: liulongfang <[email protected]>
Sent: Thursday, March 13, 2025 7:20 AM
To: [email protected]; [email protected]; Shameerali Kolothum Thodi <[email protected]>; Jonathan Cameron <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; liulongfang <[email protected]>
Subject: [PATCH v5 1/5] hisi_acc_vfio_pci: fix XQE dma address error

The dma addresses of EQE and AEQE are wrong after migration and results in guest kernel-mode encryption services failure. Comparing the definition of hardware registers, we found that there was an error when the data read from the register was combined into an address. Therefore, the address combination sequence needs to be corrected.

Even after fixing the above problem, we still have an issue where the Guest from an old kernel can get migrated to new kernel and may result in wrong data.

Quoted 7 lines

In order to ensure that the address is correct after migration, if an old magic number is detected, the dma address needs to be updated.

Fixes b0eed085903e ("hisi_acc_vfio_pci: Add support for VFIO live

migration")

LGTM,

Thanks,
Shameer

Quoted 139 lines
---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 41 ++++++++++++++++---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.h    | 14 ++++++-
 2 files changed, 47 insertions(+), 8 deletions(-)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +15 -0
index 451c639299eb..304dbdfa0e95 100644
@@ -350,6 +350,32 @@ static int vf_qm_func_stop(struct hisi_qm *qm)
 	return hisi_qm_mb(qm, QM_MB_CMD_PAUSE_QM, 0, 0, 0);
 }
 
+static int vf_qm_version_check(struct acc_vf_data *vf_data, struct device
 *dev)
+{
+	switch (vf_data->acc_magic) {
+	case ACC_DEV_MAGIC_V2:
+		if (vf_data->major_ver != ACC_DRV_MAJOR_VER) {
+			dev_info(dev, "migration driver version<%u.%u> not
 match!\n",
+				 vf_data->major_ver, vf_data->minor_ver);
+			return -EINVAL;
+		}
+		break;
+	case ACC_DEV_MAGIC_V1:
+		/* Correct dma address */
+		vf_data->eqe_dma = vf_data-
 >qm_eqc_dw[QM_XQC_ADDR_HIGH];
+		vf_data->eqe_dma <<= QM_XQC_ADDR_OFFSET;
+		vf_data->eqe_dma |= vf_data-
Quoted 1 line

qm_eqc_dw[QM_XQC_ADDR_LOW];

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
+		vf_data->aeqe_dma = vf_data-
Quoted 1 line

qm_aeqc_dw[QM_XQC_ADDR_HIGH];

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +2 -0
+		vf_data->aeqe_dma <<= QM_XQC_ADDR_OFFSET;
+		vf_data->aeqe_dma |= vf_data-
Quoted 1 line

qm_aeqc_dw[QM_XQC_ADDR_LOW];

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +8 -0
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
 static int vf_qm_check_match(struct hisi_acc_vf_core_device

*hisi_acc_vdev,

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +2 -1
 			     struct hisi_acc_vf_migration_file *migf)
 {
@@ -363,7 +389,8 @@ static int vf_qm_check_match(struct
 hisi_acc_vf_core_device *hisi_acc_vdev,
 	if (migf->total_length < QM_MATCH_SIZE || hisi_acc_vdev-
 >match_done)
 		return 0;
 
-	if (vf_data->acc_magic != ACC_DEV_MAGIC) {
+	ret = vf_qm_version_check(vf_data, dev);
+	if (ret) {
 		dev_err(dev, "failed to match ACC_DEV_MAGIC\n");
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +7 -5
 		return -EINVAL;
 	}
@@ -418,7 +445,9 @@ static int vf_qm_get_match_data(struct
 hisi_acc_vf_core_device *hisi_acc_vdev,
 	int vf_id = hisi_acc_vdev->vf_id;
 	int ret;
 
-	vf_data->acc_magic = ACC_DEV_MAGIC;
+	vf_data->acc_magic = ACC_DEV_MAGIC_V2;
+	vf_data->major_ver = ACC_DRV_MAJOR_VER;
+	vf_data->minor_ver = ACC_DRV_MINOR_VER;
 	/* Save device id */
 	vf_data->dev_id = hisi_acc_vdev->vf_dev->device;
@@ -496,12 +525,12 @@ static int vf_qm_read_data(struct hisi_qm
 *vf_qm, struct acc_vf_data *vf_data)
 		return -EINVAL;
 
 	/* Every reg is 32 bit, the dma address is 64 bit. */
-	vf_data->eqe_dma = vf_data->qm_eqc_dw[1];
+	vf_data->eqe_dma = vf_data->qm_eqc_dw[QM_XQC_ADDR_HIGH];
 	vf_data->eqe_dma <<= QM_XQC_ADDR_OFFSET;
-	vf_data->eqe_dma |= vf_data->qm_eqc_dw[0];
-	vf_data->aeqe_dma = vf_data->qm_aeqc_dw[1];
+	vf_data->eqe_dma |= vf_data->qm_eqc_dw[QM_XQC_ADDR_LOW];
+	vf_data->aeqe_dma = vf_data-
 >qm_aeqc_dw[QM_XQC_ADDR_HIGH];
 	vf_data->aeqe_dma <<= QM_XQC_ADDR_OFFSET;
-	vf_data->aeqe_dma |= vf_data->qm_aeqc_dw[0];
+	vf_data->aeqe_dma |= vf_data-
 >qm_aeqc_dw[QM_XQC_ADDR_LOW];
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +0 -0
 	/* Through SQC_BT/CQC_BT to get sqc and cqc address */
 	ret = qm_get_sqc(vf_qm, &vf_data->sqc_dma);
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h +12 -3
index 245d7537b2bc..91002ceeebc1 100644
@@ -39,6 +39,9 @@
 #define QM_REG_ADDR_OFFSET	0x0004
 
 #define QM_XQC_ADDR_OFFSET	32U
+#define QM_XQC_ADDR_LOW	0x1
+#define QM_XQC_ADDR_HIGH	0x2
+
 #define QM_VF_AEQ_INT_MASK	0x0004
 #define QM_VF_EQ_INT_MASK	0x000c
 #define QM_IFC_INT_SOURCE_V	0x0020
@@ -50,10 +53,15 @@
 #define QM_EQC_DW0		0X8000
 #define QM_AEQC_DW0		0X8020
 
+#define ACC_DRV_MAJOR_VER 1
+#define ACC_DRV_MINOR_VER 0
+
+#define ACC_DEV_MAGIC_V1	0XCDCDCDCDFEEDAACC
+#define ACC_DEV_MAGIC_V2	0xAACCFEEDDECADEDE
+
 struct acc_vf_data {
 #define QM_MATCH_SIZE offsetofend(struct acc_vf_data, qm_rsv_state)
 	/* QM match information */
-#define ACC_DEV_MAGIC	0XCDCDCDCDFEEDAACC
 	u64 acc_magic;
 	u32 qp_num;
 	u32 dev_id;
@@ -61,7 +69,9 @@ struct acc_vf_data {
 	u32 qp_base;
 	u32 vf_qm_state;
 	/* QM reserved match information */
-	u32 qm_rsv_state[3];
+	u16 major_ver;
+	u16 minor_ver;
+	u32 qm_rsv_state[2];
 
 	/* QM RW regs */
 	u32 aeq_int_mask;
--

2.24.0

author maintainer
#3 lore raw

Re: [PATCH v5 1/5] hisi_acc_vfio_pci: fix XQE dma address error

Quoted 31 lines · Shameerali Kolothum Thodi
Quoted 23 lines

-----Original Message-----
From: liulongfang <[email protected]>
Sent: Thursday, March 13, 2025 7:20 AM
To: [email protected]; [email protected]; Shameerali Kolothum Thodi <[email protected]>; Jonathan Cameron <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; liulongfang <[email protected]>
Subject: [PATCH v5 1/5] hisi_acc_vfio_pci: fix XQE dma address error

The dma addresses of EQE and AEQE are wrong after migration and results in guest kernel-mode encryption services failure. Comparing the definition of hardware registers, we found that there was an error when the data read from the register was combined into an address. Therefore, the address combination sequence needs to be corrected.

Even after fixing the above problem, we still have an issue where the Guest from an old kernel can get migrated to new kernel and may result in wrong data.

Quoted 7 lines

In order to ensure that the address is correct after migration, if an old magic number is detected, the dma address needs to be updated.

Fixes b0eed085903e ("hisi_acc_vfio_pci: Add support for VFIO live

migration")

Quoted 7 lines

LGTM,

Thanks,
Shameer

Thanks.
Longfang.

Quoted 142 lines
Quoted 138 lines
---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 41 ++++++++++++++++---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.h    | 14 ++++++-
 2 files changed, 47 insertions(+), 8 deletions(-)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +15 -0
index 451c639299eb..304dbdfa0e95 100644
@@ -350,6 +350,32 @@ static int vf_qm_func_stop(struct hisi_qm *qm)
 	return hisi_qm_mb(qm, QM_MB_CMD_PAUSE_QM, 0, 0, 0);
 }
 
+static int vf_qm_version_check(struct acc_vf_data *vf_data, struct device
 *dev)
+{
+	switch (vf_data->acc_magic) {
+	case ACC_DEV_MAGIC_V2:
+		if (vf_data->major_ver != ACC_DRV_MAJOR_VER) {
+			dev_info(dev, "migration driver version<%u.%u> not
 match!\n",
+				 vf_data->major_ver, vf_data->minor_ver);
+			return -EINVAL;
+		}
+		break;
+	case ACC_DEV_MAGIC_V1:
+		/* Correct dma address */
+		vf_data->eqe_dma = vf_data-
 > qm_eqc_dw[QM_XQC_ADDR_HIGH];
+		vf_data->eqe_dma <<= QM_XQC_ADDR_OFFSET;
+		vf_data->eqe_dma |= vf_data-
Quoted 1 line

qm_eqc_dw[QM_XQC_ADDR_LOW];

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
+		vf_data->aeqe_dma = vf_data-
Quoted 1 line

qm_aeqc_dw[QM_XQC_ADDR_HIGH];

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +2 -0
+		vf_data->aeqe_dma <<= QM_XQC_ADDR_OFFSET;
+		vf_data->aeqe_dma |= vf_data-
Quoted 1 line

qm_aeqc_dw[QM_XQC_ADDR_LOW];

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +8 -0
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
 static int vf_qm_check_match(struct hisi_acc_vf_core_device

*hisi_acc_vdev,

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +2 -1
 			     struct hisi_acc_vf_migration_file *migf)
 {
@@ -363,7 +389,8 @@ static int vf_qm_check_match(struct
 hisi_acc_vf_core_device *hisi_acc_vdev,
 	if (migf->total_length < QM_MATCH_SIZE || hisi_acc_vdev-
 > match_done)
 		return 0;
 
-	if (vf_data->acc_magic != ACC_DEV_MAGIC) {
+	ret = vf_qm_version_check(vf_data, dev);
+	if (ret) {
 		dev_err(dev, "failed to match ACC_DEV_MAGIC\n");
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +7 -5
 		return -EINVAL;
 	}
@@ -418,7 +445,9 @@ static int vf_qm_get_match_data(struct
 hisi_acc_vf_core_device *hisi_acc_vdev,
 	int vf_id = hisi_acc_vdev->vf_id;
 	int ret;
 
-	vf_data->acc_magic = ACC_DEV_MAGIC;
+	vf_data->acc_magic = ACC_DEV_MAGIC_V2;
+	vf_data->major_ver = ACC_DRV_MAJOR_VER;
+	vf_data->minor_ver = ACC_DRV_MINOR_VER;
 	/* Save device id */
 	vf_data->dev_id = hisi_acc_vdev->vf_dev->device;
@@ -496,12 +525,12 @@ static int vf_qm_read_data(struct hisi_qm
 *vf_qm, struct acc_vf_data *vf_data)
 		return -EINVAL;
 
 	/* Every reg is 32 bit, the dma address is 64 bit. */
-	vf_data->eqe_dma = vf_data->qm_eqc_dw[1];
+	vf_data->eqe_dma = vf_data->qm_eqc_dw[QM_XQC_ADDR_HIGH];
 	vf_data->eqe_dma <<= QM_XQC_ADDR_OFFSET;
-	vf_data->eqe_dma |= vf_data->qm_eqc_dw[0];
-	vf_data->aeqe_dma = vf_data->qm_aeqc_dw[1];
+	vf_data->eqe_dma |= vf_data->qm_eqc_dw[QM_XQC_ADDR_LOW];
+	vf_data->aeqe_dma = vf_data-
 > qm_aeqc_dw[QM_XQC_ADDR_HIGH];
 	vf_data->aeqe_dma <<= QM_XQC_ADDR_OFFSET;
-	vf_data->aeqe_dma |= vf_data->qm_aeqc_dw[0];
+	vf_data->aeqe_dma |= vf_data-
 > qm_aeqc_dw[QM_XQC_ADDR_LOW];
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +0 -0
 	/* Through SQC_BT/CQC_BT to get sqc and cqc address */
 	ret = qm_get_sqc(vf_qm, &vf_data->sqc_dma);
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h +12 -3
index 245d7537b2bc..91002ceeebc1 100644
@@ -39,6 +39,9 @@
 #define QM_REG_ADDR_OFFSET	0x0004
 
 #define QM_XQC_ADDR_OFFSET	32U
+#define QM_XQC_ADDR_LOW	0x1
+#define QM_XQC_ADDR_HIGH	0x2
+
 #define QM_VF_AEQ_INT_MASK	0x0004
 #define QM_VF_EQ_INT_MASK	0x000c
 #define QM_IFC_INT_SOURCE_V	0x0020
@@ -50,10 +53,15 @@
 #define QM_EQC_DW0		0X8000
 #define QM_AEQC_DW0		0X8020
 
+#define ACC_DRV_MAJOR_VER 1
+#define ACC_DRV_MINOR_VER 0
+
+#define ACC_DEV_MAGIC_V1	0XCDCDCDCDFEEDAACC
+#define ACC_DEV_MAGIC_V2	0xAACCFEEDDECADEDE
+
 struct acc_vf_data {
 #define QM_MATCH_SIZE offsetofend(struct acc_vf_data, qm_rsv_state)
 	/* QM match information */
-#define ACC_DEV_MAGIC	0XCDCDCDCDFEEDAACC
 	u64 acc_magic;
 	u32 qp_num;
 	u32 dev_id;
@@ -61,7 +69,9 @@ struct acc_vf_data {
 	u32 qp_base;
 	u32 vf_qm_state;
 	/* QM reserved match information */
-	u32 qm_rsv_state[3];
+	u16 major_ver;
+	u16 minor_ver;
+	u32 qm_rsv_state[2];
 
 	/* QM RW regs */
 	u32 aeq_int_mask;
--
Quoted 1 line

2.24.0

.

author maintainer
#4 lore raw

PATCH v5 2/5 hisi_acc_vfio_pci: add eq and aeq interruption restore

In reply to Longfang Liu

In order to ensure that the task packets of the accelerator device are not lost during the migration process, it is necessary to send an EQ and AEQ command to the device after the live migration is completed and to update the completion position of the task queue.

Let the device recheck the completed tasks data and if there are uncollected packets, device resend a task completion interrupt to the software.

Fixes b0eed085903e ("hisi_acc_vfio_pci: Add support for VFIO live migration")
---
 drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +16 -0
index 304dbdfa0e95..80217aea5475 100644
@@ -470,6 +470,19 @@ static int vf_qm_get_match_data(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 	return 0;
 }
 
+static void vf_qm_xeqc_save(struct hisi_qm *qm,
+			    struct hisi_acc_vf_migration_file *migf)
+{
+	struct acc_vf_data *vf_data = &migf->vf_data;
+	u16 eq_head, aeq_head;
+
+	eq_head = vf_data->qm_eqc_dw[0] & 0xFFFF;
+	qm_db(qm, 0, QM_DOORBELL_CMD_EQ, eq_head, 0);
+
+	aeq_head = vf_data->qm_aeqc_dw[0] & 0xFFFF;
+	qm_db(qm, 0, QM_DOORBELL_CMD_AEQ, aeq_head, 0);
+}
+
 static int vf_qm_load_data(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 			   struct hisi_acc_vf_migration_file *migf)
 {
@@ -578,6 +591,9 @@ static int vf_qm_state_save(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 		return -EINVAL;
 
 	migf->total_length = sizeof(struct acc_vf_data);
+	/* Save eqc and aeqc interrupt information */
+	vf_qm_xeqc_save(vf_qm, migf);
+
 	return 0;
 }
 
Signature · 1 line
2.24.0
author maintainer
#5 lore raw

PATCH v5 3/5 hisi_acc_vfio_pci: bugfix cache write-back issue

In reply to Longfang Liu

At present, cache write-back is placed in the device data copy stage after stopping the device operation. Writing back to the cache at this stage will cause the data obtained by the cache to be written back to be empty.

In order to ensure that the cache data is written back successfully, the data needs to be written back into the stop device stage.

Fixes b0eed085903e ("hisi_acc_vfio_pci: Add support for VFIO live migration")
---
 drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +7 -7
index 80217aea5475..d96446f499ed 100644
@@ -566,7 +566,6 @@ static int vf_qm_state_save(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 {
 	struct acc_vf_data *vf_data = &migf->vf_data;
 	struct hisi_qm *vf_qm = &hisi_acc_vdev->vf_qm;
-	struct device *dev = &vf_qm->pdev->dev;
 	int ret;
 
 	if (unlikely(qm_wait_dev_not_ready(vf_qm))) {
@@ -580,12 +579,6 @@ static int vf_qm_state_save(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 	vf_data->vf_qm_state = QM_READY;
 	hisi_acc_vdev->vf_qm_state = vf_data->vf_qm_state;
 
-	ret = vf_qm_cache_wb(vf_qm);
-	if (ret) {
-		dev_err(dev, "failed to writeback QM Cache!\n");
-		return ret;
-	}
-
 	ret = vf_qm_read_data(vf_qm, vf_data);
 	if (ret)
 		return -EINVAL;
@@ -1012,6 +1005,13 @@ static int hisi_acc_vf_stop_device(struct hisi_acc_vf_core_device *hisi_acc_vdev
 		dev_err(dev, "failed to check QM INT state!\n");
 		return ret;
 	}
+
+	ret = vf_qm_cache_wb(vf_qm);
+	if (ret) {
+		dev_err(dev, "failed to writeback QM cache!\n");
+		return ret;
+	}
+
 	return 0;
 }
 
Signature · 1 line
2.24.0
author maintainer
#6 lore raw

PATCH v5 4/5 hisi_acc_vfio_pci: bugfix the problem of uninstalling driver

In reply to Longfang Liu

In a live migration scenario. If the number of VFs at the destination is greater than the source, the recovery operation will fail and qemu will not be able to complete the process and exit after shutting down the device FD.

This will cause the driver to be unable to be unloaded normally due to abnormal reference counting of the live migration driver caused by the abnormal closing operation of fd.

Fixes b0eed085903e ("hisi_acc_vfio_pci: Add support for VFIO live migration")
---
 drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c | 1 +
 1 file changed, 1 insertion(+)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
index d96446f499ed..cadc82419dca 100644
@@ -1508,6 +1508,7 @@ static void hisi_acc_vfio_pci_close_device(struct vfio_device *core_vdev)
 	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(core_vdev);
 	struct hisi_qm *vf_qm = &hisi_acc_vdev->vf_qm;
 
+	hisi_acc_vf_disable_fds(hisi_acc_vdev);
 	mutex_lock(&hisi_acc_vdev->open_mutex);
 	hisi_acc_vdev->dev_opened = false;
 	iounmap(vf_qm->io_base);
Signature · 1 line
2.24.0
author maintainer
#7 lore raw

PATCH v5 5/5 hisi_acc_vfio_pci: bugfix live migration function without VF device driver

In reply to Longfang Liu

If the VF device driver is not loaded in the Guest OS and we attempt to perform device data migration, the address of the migrated data will be NULL.
The live migration recovery operation on the destination side will access a null address value, which will cause access errors.

Therefore, live migration of VMs without added VF device drivers does not require device data migration.
In addition, when the queue address data obtained by the destination is empty, device queue recovery processing will not be performed.

Fixes b0eed085903e ("hisi_acc_vfio_pci: Add support for VFIO live migration")
---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 25 +++++++++++++------
 1 file changed, 18 insertions(+), 7 deletions(-)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +18 -7
index cadc82419dca..44fa2d16bbcc 100644
@@ -426,13 +426,6 @@ static int vf_qm_check_match(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 		return -EINVAL;
 	}
 
-	ret = qm_write_regs(vf_qm, QM_VF_STATE, &vf_data->vf_qm_state, 1);
-	if (ret) {
-		dev_err(dev, "failed to write QM_VF_STATE\n");
-		return ret;
-	}
-
-	hisi_acc_vdev->vf_qm_state = vf_data->vf_qm_state;
 	hisi_acc_vdev->match_done = true;
 	return 0;
 }
@@ -498,6 +491,13 @@ static int vf_qm_load_data(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 	if (migf->total_length < sizeof(struct acc_vf_data))
 		return -EINVAL;
 
+	ret = qm_write_regs(qm, QM_VF_STATE, &vf_data->vf_qm_state, 1);
+	if (ret) {
+		dev_err(dev, "failed to write QM_VF_STATE\n");
+		return -EINVAL;
+	}
+	hisi_acc_vdev->vf_qm_state = vf_data->vf_qm_state;
+
 	qm->eqe_dma = vf_data->eqe_dma;
 	qm->aeqe_dma = vf_data->aeqe_dma;
 	qm->sqc_dma = vf_data->sqc_dma;
@@ -506,6 +506,12 @@ static int vf_qm_load_data(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 	qm->qp_base = vf_data->qp_base;
 	qm->qp_num = vf_data->qp_num;
 
+	if (!vf_data->eqe_dma || !vf_data->aeqe_dma ||
+	    !vf_data->sqc_dma || !vf_data->cqc_dma) {
+		dev_err(dev, "resume dma addr is NULL!\n");
+		return -EINVAL;
+	}
+
 	ret = qm_set_regs(qm, vf_data);
 	if (ret) {
 		dev_err(dev, "set VF regs failed\n");
@@ -726,8 +732,12 @@ static int hisi_acc_vf_load_state(struct hisi_acc_vf_core_device *hisi_acc_vdev)
 {
 	struct device *dev = &hisi_acc_vdev->vf_dev->dev;
 	struct hisi_acc_vf_migration_file *migf = hisi_acc_vdev->resuming_migf;
+	struct acc_vf_data *vf_data = &migf->vf_data;
 	int ret;
 
+	if (vf_data->vf_qm_state != QM_READY)
+		return 0;
+
 	/* Recover data to VF */
 	ret = vf_qm_load_data(hisi_acc_vdev, migf);
 	if (ret) {
@@ -1531,6 +1541,7 @@ static int hisi_acc_vfio_pci_migrn_init_dev(struct vfio_device *core_vdev)
 	hisi_acc_vdev->vf_id = pci_iov_vf_id(pdev) + 1;
 	hisi_acc_vdev->pf_qm = pf_qm;
 	hisi_acc_vdev->vf_dev = pdev;
+	hisi_acc_vdev->vf_qm_state = QM_NOT_READY;
 	mutex_init(&hisi_acc_vdev->state_mutex);
 	mutex_init(&hisi_acc_vdev->open_mutex);
 
Signature · 1 line
2.24.0
Reviewed-by on v5 5/5
#8 lore raw

RE: [PATCH v5 5/5] hisi_acc_vfio_pci: bugfix live migration function without VF device driver

Quoted 82 lines

-----Original Message-----
From: liulongfang <[email protected]>
Sent: Thursday, March 13, 2025 7:20 AM
To: [email protected]; [email protected]; Shameerali Kolothum Thodi <[email protected]>; Jonathan Cameron <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; liulongfang <[email protected]>
Subject: [PATCH v5 5/5] hisi_acc_vfio_pci: bugfix live migration function without VF device driver

If the VF device driver is not loaded in the Guest OS and we attempt to perform device data migration, the address of the migrated data will be NULL.
The live migration recovery operation on the destination side will access a null address value, which will cause access errors.

Therefore, live migration of VMs without added VF device drivers does not require device data migration.
In addition, when the queue address data obtained by the destination is empty, device queue recovery processing will not be performed.

Fixes b0eed085903e ("hisi_acc_vfio_pci: Add support for VFIO live

migration")

---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 25 +++++++++++++------
 1 file changed, 18 insertions(+), 7 deletions(-)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +0 -7
index cadc82419dca..44fa2d16bbcc 100644
@@ -426,13 +426,6 @@ static int vf_qm_check_match(struct
 hisi_acc_vf_core_device *hisi_acc_vdev,
 		return -EINVAL;
 	}
 
-	ret = qm_write_regs(vf_qm, QM_VF_STATE, &vf_data->vf_qm_state,
 1);
-	if (ret) {
-		dev_err(dev, "failed to write QM_VF_STATE\n");
-		return ret;
-	}
-
-	hisi_acc_vdev->vf_qm_state = vf_data->vf_qm_state;
 	hisi_acc_vdev->match_done = true;
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +7 -0
 	return 0;
 }
@@ -498,6 +491,13 @@ static int vf_qm_load_data(struct
 hisi_acc_vf_core_device *hisi_acc_vdev,
 	if (migf->total_length < sizeof(struct acc_vf_data))
 		return -EINVAL;
 
+	ret = qm_write_regs(qm, QM_VF_STATE, &vf_data->vf_qm_state, 1);
+	if (ret) {
+		dev_err(dev, "failed to write QM_VF_STATE\n");
+		return -EINVAL;
+	}
+	hisi_acc_vdev->vf_qm_state = vf_data->vf_qm_state;
+
 	qm->eqe_dma = vf_data->eqe_dma;
 	qm->aeqe_dma = vf_data->aeqe_dma;
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +6 -0
 	qm->sqc_dma = vf_data->sqc_dma;
@@ -506,6 +506,12 @@ static int vf_qm_load_data(struct
 hisi_acc_vf_core_device *hisi_acc_vdev,
 	qm->qp_base = vf_data->qp_base;
 	qm->qp_num = vf_data->qp_num;
 
+	if (!vf_data->eqe_dma || !vf_data->aeqe_dma ||
+	    !vf_data->sqc_dma || !vf_data->cqc_dma) {
+		dev_err(dev, "resume dma addr is NULL!\n");
+		return -EINVAL;
+	}
+
 	ret = qm_set_regs(qm, vf_data);
 	if (ret) {
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +0 -0
 		dev_err(dev, "set VF regs failed\n");

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static int hisi_acc_vf_load_state(struct · line 741

Quoted 10 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +3 -0
@@ -726,8 +732,12 @@ static int hisi_acc_vf_load_state(struct
 hisi_acc_vf_core_device *hisi_acc_vdev)
 {
 	struct device *dev = &hisi_acc_vdev->vf_dev->dev;
 	struct hisi_acc_vf_migration_file *migf = hisi_acc_vdev-
 >resuming_migf;
+	struct acc_vf_data *vf_data = &migf->vf_data;
 	int ret;
 
+	if (vf_data->vf_qm_state != QM_READY)
+		return 0;

I don't think we need to check the above. In vf_qm_satte_save(), If vf_qm_state != QM_READY, we set the
migf->total_length = QM_MATCH_SIZE.

Hence it will return 0 in the below vf_qm_load_data() anyway.

With that corrected,

Thanks,
Shameer

Quoted 15 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
+
 	/* Recover data to VF */
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -1
 	ret = vf_qm_load_data(hisi_acc_vdev, migf);
 	if (ret) {
@@ -1531,6 +1541,7 @@ static int hisi_acc_vfio_pci_migrn_init_dev(struct
 vfio_device *core_vdev)
 	hisi_acc_vdev->vf_id = pci_iov_vf_id(pdev) + 1;
 	hisi_acc_vdev->pf_qm = pf_qm;
 	hisi_acc_vdev->vf_dev = pdev;
+	hisi_acc_vdev->vf_qm_state = QM_NOT_READY;
 	mutex_init(&hisi_acc_vdev->state_mutex);
 	mutex_init(&hisi_acc_vdev->open_mutex);
 
--

2.24.0

author maintainer
#9 lore raw

Re: [PATCH v5 5/5] hisi_acc_vfio_pci: bugfix live migration function without VF device driver

Quoted 90 lines · Shameerali Kolothum Thodi
Quoted 90 lines

-----Original Message-----
From: liulongfang <[email protected]>
Sent: Thursday, March 13, 2025 7:20 AM
To: [email protected]; [email protected]; Shameerali Kolothum Thodi <[email protected]>; Jonathan Cameron <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; liulongfang <[email protected]>
Subject: [PATCH v5 5/5] hisi_acc_vfio_pci: bugfix live migration function without VF device driver

If the VF device driver is not loaded in the Guest OS and we attempt to perform device data migration, the address of the migrated data will be NULL.
The live migration recovery operation on the destination side will access a null address value, which will cause access errors.

Therefore, live migration of VMs without added VF device drivers does not require device data migration.
In addition, when the queue address data obtained by the destination is empty, device queue recovery processing will not be performed.

Fixes b0eed085903e ("hisi_acc_vfio_pci: Add support for VFIO live

migration")

---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 25 +++++++++++++------
 1 file changed, 18 insertions(+), 7 deletions(-)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +0 -7
index cadc82419dca..44fa2d16bbcc 100644
@@ -426,13 +426,6 @@ static int vf_qm_check_match(struct
 hisi_acc_vf_core_device *hisi_acc_vdev,
 		return -EINVAL;
 	}
 
-	ret = qm_write_regs(vf_qm, QM_VF_STATE, &vf_data->vf_qm_state,
 1);
-	if (ret) {
-		dev_err(dev, "failed to write QM_VF_STATE\n");
-		return ret;
-	}
-
-	hisi_acc_vdev->vf_qm_state = vf_data->vf_qm_state;
 	hisi_acc_vdev->match_done = true;
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +7 -0
 	return 0;
 }
@@ -498,6 +491,13 @@ static int vf_qm_load_data(struct
 hisi_acc_vf_core_device *hisi_acc_vdev,
 	if (migf->total_length < sizeof(struct acc_vf_data))
 		return -EINVAL;
 
+	ret = qm_write_regs(qm, QM_VF_STATE, &vf_data->vf_qm_state, 1);
+	if (ret) {
+		dev_err(dev, "failed to write QM_VF_STATE\n");
+		return -EINVAL;
+	}
+	hisi_acc_vdev->vf_qm_state = vf_data->vf_qm_state;
+
 	qm->eqe_dma = vf_data->eqe_dma;
 	qm->aeqe_dma = vf_data->aeqe_dma;
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +6 -0
 	qm->sqc_dma = vf_data->sqc_dma;
@@ -506,6 +506,12 @@ static int vf_qm_load_data(struct
 hisi_acc_vf_core_device *hisi_acc_vdev,
 	qm->qp_base = vf_data->qp_base;
 	qm->qp_num = vf_data->qp_num;
 
+	if (!vf_data->eqe_dma || !vf_data->aeqe_dma ||
+	    !vf_data->sqc_dma || !vf_data->cqc_dma) {
+		dev_err(dev, "resume dma addr is NULL!\n");
+		return -EINVAL;
+	}
+
 	ret = qm_set_regs(qm, vf_data);
 	if (ret) {
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
 		dev_err(dev, "set VF regs failed\n");
@@ -726,8 +732,12 @@ static int hisi_acc_vf_load_state(struct
 hisi_acc_vf_core_device *hisi_acc_vdev)
 {
 	struct device *dev = &hisi_acc_vdev->vf_dev->dev;
 	struct hisi_acc_vf_migration_file *migf = hisi_acc_vdev-
 > resuming_migf;
+	struct acc_vf_data *vf_data = &migf->vf_data;
 	int ret;
 
Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static int hisi_acc_vf_load_state(struct hisi_acc_vf_core_device *hisi_acc_vdev) · line 739

Quoted 2 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +2 -0
+	if (vf_data->vf_qm_state != QM_READY)
+		return 0;

I don't think we need to check the above. In vf_qm_satte_save(), If vf_qm_state != QM_READY, we set the
migf->total_length = QM_MATCH_SIZE.

Hence it will return 0 in the below vf_qm_load_data() anyway.

After removing this judgment code, the live migration function works normally without loading the VM driver.
I will remove this judgment in the next version.

Thanks.
Longfang.

Quoted 25 lines

With that corrected,

Thanks,
Shameer

Quoted 14 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
+
 	/* Recover data to VF */
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -1
 	ret = vf_qm_load_data(hisi_acc_vdev, migf);
 	if (ret) {
@@ -1531,6 +1541,7 @@ static int hisi_acc_vfio_pci_migrn_init_dev(struct
 vfio_device *core_vdev)
 	hisi_acc_vdev->vf_id = pci_iov_vf_id(pdev) + 1;
 	hisi_acc_vdev->pf_qm = pf_qm;
 	hisi_acc_vdev->vf_dev = pdev;
+	hisi_acc_vdev->vf_qm_state = QM_NOT_READY;
 	mutex_init(&hisi_acc_vdev->state_mutex);
 	mutex_init(&hisi_acc_vdev->open_mutex);
 
--
Quoted 1 line

2.24.0

.