Skip to contentSkip to conversation

add debugfs to hisilicon migration driver

PATCH Superseded Longfang Liu 2 participants 12 messages Apr 2, 2024 – Apr 24, 2024

Subsystems (4)

v4 of 15 Superseded Latest is v15 →

lore
Share & tools
lore

Longfang Liu

author maintainer
#0 lore raw

v4 cover add debugfs to hisilicon migration driver

Add a debugfs function to the hisilicon migration driver in VFIO to provide intermediate state values and data during device migration.

When the execution of live migration fails, the user can view the status and data during the migration process separately from the source and the destination, which is convenient for users to analyze and locate problems.

Changes v3 -> v4
Rebased on kernel6.9

Changes v2 -> v3
Solve debugfs serialization problem.

Changes v1 -> v2
Solve the racy problem of io_base.

Longfang Liu (4):
hisi_acc_vfio_pci: extract public functions for container_of
hisi_acc_vfio_pci: Create subfunction for data reading
hisi_acc_vfio_pci: register debugfs for hisilicon migration driver
Documentation: add debugfs description for hisi migration

 .../ABI/testing/debugfs-hisi-migration        |  34 +++
 MAINTAINERS                                   |   1 +
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 281 ++++++++++++++++--
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.h    |  14 +
 4 files changed, 298 insertions(+), 32 deletions(-)
 create mode 100644 Documentation/ABI/testing/debugfs-hisi-migration
Signature · 1 line
2.24.0
author maintainer
#1 lore raw

PATCH v4 1/4 hisi_acc_vfio_pci: extract public functions for container_of

In the current driver, vdev is obtained from struct hisi_acc_vf_core_device through the container_of function. This method is used in many places in the driver. In order to reduce this repetitive operation, It was extracted into a public function.

---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 21 ++++++++++---------
 1 file changed, 11 insertions(+), 10 deletions(-)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +11 -10
index 9a3e97108ace..45351be8e270 100644
@@ -630,6 +630,12 @@ static void hisi_acc_vf_disable_fds(struct hisi_acc_vf_core_device *hisi_acc_vde
 	}
 }
 
+static struct hisi_acc_vf_core_device *hisi_acc_get_vf_dev(struct vfio_device *vdev)
+{
+	return container_of(vdev, struct hisi_acc_vf_core_device,
+			    core_device.vdev);
+}
+
 static void hisi_acc_vf_reset(struct hisi_acc_vf_core_device *hisi_acc_vdev)
 {
 	hisi_acc_vdev->vf_qm_state = QM_NOT_READY;
@@ -1033,8 +1039,7 @@ static struct file *
 hisi_acc_vfio_pci_set_device_state(struct vfio_device *vdev,
 				   enum vfio_device_mig_state new_state)
 {
-	struct hisi_acc_vf_core_device *hisi_acc_vdev = container_of(vdev,
-			struct hisi_acc_vf_core_device, core_device.vdev);
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
 	enum vfio_device_mig_state next_state;
 	struct file *res = NULL;
 	int ret;
@@ -1075,8 +1080,7 @@ static int
 hisi_acc_vfio_pci_get_device_state(struct vfio_device *vdev,
 				   enum vfio_device_mig_state *curr_state)
 {
-	struct hisi_acc_vf_core_device *hisi_acc_vdev = container_of(vdev,
-			struct hisi_acc_vf_core_device, core_device.vdev);
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
 
 	mutex_lock(&hisi_acc_vdev->state_mutex);
 	*curr_state = hisi_acc_vdev->mig_state;
@@ -1280,8 +1284,7 @@ static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int
 
 static int hisi_acc_vfio_pci_open_device(struct vfio_device *core_vdev)
 {
-	struct hisi_acc_vf_core_device *hisi_acc_vdev = container_of(core_vdev,
-			struct hisi_acc_vf_core_device, core_device.vdev);
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(core_vdev);
 	struct vfio_pci_core_device *vdev = &hisi_acc_vdev->core_device;
 	int ret;
 
@@ -1304,8 +1307,7 @@ static int hisi_acc_vfio_pci_open_device(struct vfio_device *core_vdev)
 
 static void hisi_acc_vfio_pci_close_device(struct vfio_device *core_vdev)
 {
-	struct hisi_acc_vf_core_device *hisi_acc_vdev = container_of(core_vdev,
-			struct hisi_acc_vf_core_device, core_device.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;
 
 	iounmap(vf_qm->io_base);
@@ -1320,8 +1322,7 @@ static const struct vfio_migration_ops hisi_acc_vfio_pci_migrn_state_ops = {
 
 static int hisi_acc_vfio_pci_migrn_init_dev(struct vfio_device *core_vdev)
 {
-	struct hisi_acc_vf_core_device *hisi_acc_vdev = container_of(core_vdev,
-			struct hisi_acc_vf_core_device, core_device.vdev);
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(core_vdev);
 	struct pci_dev *pdev = to_pci_dev(core_vdev->dev);
 	struct hisi_qm *pf_qm = hisi_acc_get_pf_qm(pdev);
 
Signature · 1 line
2.24.0
author maintainer
#2 lore raw

PATCH v4 2/4 hisi_acc_vfio_pci: Create subfunction for data reading

In reply to Longfang Liu

During the live migration process. It needs to obtain various status data of drivers and devices. In order to facilitate calling it in the debugfs function. For all operations that read data from device registers, the driver creates a subfunction.
Also fixed the location of address data.

---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 56 +++++++++++--------
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.h    |  3 +
 2 files changed, 37 insertions(+), 22 deletions(-)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +34 -22
index 45351be8e270..bf358ba94b5d 100644
@@ -486,6 +486,39 @@ static int vf_qm_load_data(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 	return 0;
 }
 
+static int vf_qm_read_data(struct hisi_qm *vf_qm, struct acc_vf_data *vf_data)
+{
+	struct device *dev = &vf_qm->pdev->dev;
+	int ret;
+
+	ret = qm_get_regs(vf_qm, vf_data);
+	if (ret)
+		return -EINVAL;
+
+	/* Every reg is 32 bit, the dma address is 64 bit. */
+	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];
+
+	/* Through SQC_BT/CQC_BT to get sqc and cqc address */
+	ret = qm_get_sqc(vf_qm, &vf_data->sqc_dma);
+	if (ret) {
+		dev_err(dev, "failed to read SQC addr!\n");
+		return -EINVAL;
+	}
+
+	ret = qm_get_cqc(vf_qm, &vf_data->cqc_dma);
+	if (ret) {
+		dev_err(dev, "failed to read CQC addr!\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
 static int vf_qm_state_save(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 			    struct hisi_acc_vf_migration_file *migf)
 {
@@ -511,31 +544,10 @@ static int vf_qm_state_save(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 		return ret;
 	}
 
-	ret = qm_get_regs(vf_qm, vf_data);
+	ret = vf_qm_read_data(vf_qm, vf_data);
 	if (ret)
 		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 <<= 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->aeqe_dma <<= QM_XQC_ADDR_OFFSET;
-	vf_data->aeqe_dma |= vf_data->qm_aeqc_dw[0];
-
-	/* Through SQC_BT/CQC_BT to get sqc and cqc address */
-	ret = qm_get_sqc(vf_qm, &vf_data->sqc_dma);
-	if (ret) {
-		dev_err(dev, "failed to read SQC addr!\n");
-		return -EINVAL;
-	}
-
-	ret = qm_get_cqc(vf_qm, &vf_data->cqc_dma);
-	if (ret) {
-		dev_err(dev, "failed to read CQC addr!\n");
-		return -EINVAL;
-	}
-
 	migf->total_length = sizeof(struct acc_vf_data);
 	return 0;
 }
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h +3 -0
index 5bab46602fad..7a9dc87627cd 100644
@@ -38,6 +38,9 @@
 #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
Signature · 1 line
2.24.0
#3 lore raw

Re: [PATCH v4 2/4] hisi_acc_vfio_pci: Create subfunction for data reading

Quoted 5 lines · On Tue, 2 Apr 2024 11:24:30 +0800 Longfang Liu <[email protected]> wrote:

During the live migration process. It needs to obtain various status data of drivers and devices. In order to facilitate calling it in the debugfs function. For all operations that read data from device registers, the driver creates a subfunction.
Also fixed the location of address data.

Cédric noted privately and I agree, 1) fixes should be provided in separate patches with a Fixes: tag rather than subtly included in a minor refactoring, and 2) what does this imply about the existing functionality of migration? This would seem to suggest existing migration data is bogus if we're offset by a register reading the DMA address. The commit log for the Fixes patch should describe this.

Quoted 98 lines
---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 56 +++++++++++--------
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.h    |  3 +
 2 files changed, 37 insertions(+), 22 deletions(-)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +34 -22
index 45351be8e270..bf358ba94b5d 100644
@@ -486,6 +486,39 @@ static int vf_qm_load_data(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 	return 0;
 }
 
+static int vf_qm_read_data(struct hisi_qm *vf_qm, struct acc_vf_data *vf_data)
+{
+	struct device *dev = &vf_qm->pdev->dev;
+	int ret;
+
+	ret = qm_get_regs(vf_qm, vf_data);
+	if (ret)
+		return -EINVAL;
+
+	/* Every reg is 32 bit, the dma address is 64 bit. */
+	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];
+
+	/* Through SQC_BT/CQC_BT to get sqc and cqc address */
+	ret = qm_get_sqc(vf_qm, &vf_data->sqc_dma);
+	if (ret) {
+		dev_err(dev, "failed to read SQC addr!\n");
+		return -EINVAL;
+	}
+
+	ret = qm_get_cqc(vf_qm, &vf_data->cqc_dma);
+	if (ret) {
+		dev_err(dev, "failed to read CQC addr!\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
 static int vf_qm_state_save(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 			    struct hisi_acc_vf_migration_file *migf)
 {
@@ -511,31 +544,10 @@ static int vf_qm_state_save(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 		return ret;
 	}
 
-	ret = qm_get_regs(vf_qm, vf_data);
+	ret = vf_qm_read_data(vf_qm, vf_data);
 	if (ret)
 		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 <<= 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->aeqe_dma <<= QM_XQC_ADDR_OFFSET;
-	vf_data->aeqe_dma |= vf_data->qm_aeqc_dw[0];
-
-	/* Through SQC_BT/CQC_BT to get sqc and cqc address */
-	ret = qm_get_sqc(vf_qm, &vf_data->sqc_dma);
-	if (ret) {
-		dev_err(dev, "failed to read SQC addr!\n");
-		return -EINVAL;
-	}
-
-	ret = qm_get_cqc(vf_qm, &vf_data->cqc_dma);
-	if (ret) {
-		dev_err(dev, "failed to read CQC addr!\n");
-		return -EINVAL;
-	}
-
 	migf->total_length = sizeof(struct acc_vf_data);
 	return 0;
 }
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h +3 -0
index 5bab46602fad..7a9dc87627cd 100644
@@ -38,6 +38,9 @@
 #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
author maintainer
#4 lore raw

Re: [PATCH v4 2/4] hisi_acc_vfio_pci: Create subfunction for data reading

Quoted 9 lines · Alex Williamson
Quoted 5 lines · On Tue, 2 Apr 2024 11:24:30 +0800 Longfang Liu <[email protected]> wrote:

During the live migration process. It needs to obtain various status data of drivers and devices. In order to facilitate calling it in the debugfs function. For all operations that read data from device registers, the driver creates a subfunction.
Also fixed the location of address data.

Quoted 7 lines

Cédric noted privately and I agree, 1) fixes should be provided in separate patches with a Fixes: tag rather than subtly included in a minor refactoring, and 2) what does this imply about the existing functionality of migration? This would seem to suggest existing migration data is bogus if we're offset by a register reading the DMA address. The commit log for the Fixes patch should describe this.

Okay, the modification of the DMA address offset part is split into a new patch, and the modification of this part is explained clearly.

Thanks,
Longfang.

Quoted 101 lines
Quoted 89 lines
---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 56 +++++++++++--------
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.h    |  3 +
 2 files changed, 37 insertions(+), 22 deletions(-)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +34 -22
index 45351be8e270..bf358ba94b5d 100644
@@ -486,6 +486,39 @@ static int vf_qm_load_data(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 	return 0;
 }
 
+static int vf_qm_read_data(struct hisi_qm *vf_qm, struct acc_vf_data *vf_data)
+{
+	struct device *dev = &vf_qm->pdev->dev;
+	int ret;
+
+	ret = qm_get_regs(vf_qm, vf_data);
+	if (ret)
+		return -EINVAL;
+
+	/* Every reg is 32 bit, the dma address is 64 bit. */
+	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];
+
+	/* Through SQC_BT/CQC_BT to get sqc and cqc address */
+	ret = qm_get_sqc(vf_qm, &vf_data->sqc_dma);
+	if (ret) {
+		dev_err(dev, "failed to read SQC addr!\n");
+		return -EINVAL;
+	}
+
+	ret = qm_get_cqc(vf_qm, &vf_data->cqc_dma);
+	if (ret) {
+		dev_err(dev, "failed to read CQC addr!\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
 static int vf_qm_state_save(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 			    struct hisi_acc_vf_migration_file *migf)
 {
@@ -511,31 +544,10 @@ static int vf_qm_state_save(struct hisi_acc_vf_core_device *hisi_acc_vdev,
 		return ret;
 	}
 
-	ret = qm_get_regs(vf_qm, vf_data);
+	ret = vf_qm_read_data(vf_qm, vf_data);
 	if (ret)
 		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 <<= 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->aeqe_dma <<= QM_XQC_ADDR_OFFSET;
-	vf_data->aeqe_dma |= vf_data->qm_aeqc_dw[0];
-
-	/* Through SQC_BT/CQC_BT to get sqc and cqc address */
-	ret = qm_get_sqc(vf_qm, &vf_data->sqc_dma);
-	if (ret) {
-		dev_err(dev, "failed to read SQC addr!\n");
-		return -EINVAL;
-	}
-
-	ret = qm_get_cqc(vf_qm, &vf_data->cqc_dma);
-	if (ret) {
-		dev_err(dev, "failed to read CQC addr!\n");
-		return -EINVAL;
-	}
-
 	migf->total_length = sizeof(struct acc_vf_data);
 	return 0;
 }
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h +0 -0
index 5bab46602fad..7a9dc87627cd 100644

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h line 46

Quoted 9 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h +3 -0
@@ -38,6 +38,9 @@
 #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

.

author maintainer
#5 lore raw

PATCH v4 3/4 hisi_acc_vfio_pci: register debugfs for hisilicon migration driver

In reply to Longfang Liu

On the debugfs framework of VFIO, if the CONFIG_VFIO_DEBUGFS macro is enabled, the debug function is registered for the live migration driver of the HiSilicon accelerator device.

After registering the HiSilicon accelerator device on the debugfs framework of live migration of vfio, a directory file "hisi_acc" of debugfs is created, and then three debug function files are created in this directory:

vfio

    |
    +---<dev_name1>
    |    +---migration
    |        +--state
    |        +--hisi_acc
    |            +--attr
    |            +--data
    |            +--save
    |            +--cmd_state
    |
    +---<dev_name2>
         +---migration
             +--state
             +--hisi_acc
                 +--attr
                 +--data
                 +--save
                 +--cmd_state

data file: used to get the migration data from the driver attr file: used to get device attributes parameters from the driver save file: used to read the data of the live migration device and save it to the driver.
cmd_state: used to get the cmd channel state for the device.

+----------------+        +--------------+       +---------------+
| migration dev  |        |   src  dev   |       |   dst  dev    |
+-------+--------+        +------+-------+       +-------+-------+
        |                        |                       |
        |                        |                       |
        |                        |                       |
        |                        |                       |
  save  |                 +------v-------+       +-------v-------+
        |                 |  saving_mif  |       | resuming_migf |
        |                 |     file     |       |     file      |
        |                 +------+-------+       +-------+-------+
        |                        |                       |
        |        mutex           |                       |
+-------v--------+               |                       |
|                |               |                       |
| debug_migf file<---------------+-----------------------+
|                |             copy
+-------+--------+
        |
   cat  |
        |
+-------v--------+
|     user       |
+----------------+

In debugfs scheme. The driver creates a separate debug_migf file. It is completely separated from the two files of live migration, thus preventing debugfs data from interfering with migration data. Moreover, it only performs read operations on the device.

For serialization of debugfs:
First, it only writes data when performing a debugfs save operation. Second, it is only copied from the file on the migration device when the live migration is complete.
These two operations are mutually exclusive through mutex.

---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 204 ++++++++++++++++++
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.h    |  11 +
 2 files changed, 215 insertions(+)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +204 -0
index bf358ba94b5d..9f563a31a2a1 100644
@@ -15,6 +15,7 @@
 #include <linux/anon_inodes.h>
 
 #include "hisi_acc_vfio_pci.h"
+#include "../../vfio.h"
 
 /* Return 0 on VM acc device ready, -ETIMEDOUT hardware timeout */
 static int qm_wait_dev_not_ready(struct hisi_qm *qm)
@@ -618,6 +619,22 @@ hisi_acc_check_int_state(struct hisi_acc_vf_core_device *hisi_acc_vdev)
 	}
 }
 
+static void hisi_acc_vf_migf_save(struct hisi_acc_vf_core_device *hisi_acc_vdev,
+	struct hisi_acc_vf_migration_file *src_migf)
+{
+	struct hisi_acc_vf_migration_file *dst_migf = hisi_acc_vdev->debug_migf;
+
+	if (!dst_migf)
+		return;
+
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
+	dst_migf->disabled = src_migf->disabled;
+	dst_migf->total_length = src_migf->total_length;
+	memcpy(&dst_migf->vf_data, &src_migf->vf_data,
+		    sizeof(struct acc_vf_data));
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
+}
+
 static void hisi_acc_vf_disable_fd(struct hisi_acc_vf_migration_file *migf)
 {
 	mutex_lock(&migf->lock);
@@ -630,12 +647,14 @@ static void hisi_acc_vf_disable_fd(struct hisi_acc_vf_migration_file *migf)
 static void hisi_acc_vf_disable_fds(struct hisi_acc_vf_core_device *hisi_acc_vdev)
 {
 	if (hisi_acc_vdev->resuming_migf) {
+		hisi_acc_vf_migf_save(hisi_acc_vdev, hisi_acc_vdev->resuming_migf);
 		hisi_acc_vf_disable_fd(hisi_acc_vdev->resuming_migf);
 		fput(hisi_acc_vdev->resuming_migf->filp);
 		hisi_acc_vdev->resuming_migf = NULL;
 	}
 
 	if (hisi_acc_vdev->saving_migf) {
+		hisi_acc_vf_migf_save(hisi_acc_vdev, hisi_acc_vdev->saving_migf);
 		hisi_acc_vf_disable_fd(hisi_acc_vdev->saving_migf);
 		fput(hisi_acc_vdev->saving_migf->filp);
 		hisi_acc_vdev->saving_migf = NULL;
@@ -1144,6 +1163,7 @@ static int hisi_acc_vf_qm_init(struct hisi_acc_vf_core_device *hisi_acc_vdev)
 	if (!vf_qm->io_base)
 		return -EIO;
 
+	mutex_init(&hisi_acc_vdev->enable_mutex);
 	vf_qm->fun_type = QM_HW_VF;
 	vf_qm->pdev = vf_dev;
 	mutex_init(&vf_qm->mailbox_lock);
@@ -1294,6 +1314,181 @@ static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int
 	return vfio_pci_core_ioctl(core_vdev, cmd, arg);
 }
 
+static int hisi_acc_vf_debug_check(struct seq_file *seq, struct vfio_device *vdev)
+{
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *migf = hisi_acc_vdev->debug_migf;
+	struct hisi_qm *vf_qm = &hisi_acc_vdev->vf_qm;
+	int ret;
+
+	if (!vdev->mig_ops || !migf) {
+		seq_printf(seq, "%s\n", "device does not support live migration!");
+		return -EINVAL;
+	}
+
+	/**
+	 * When the device is not opened, the io_base is not mapped.
+	 * The driver cannot perform device read and write operations.
+	 */
+	if (hisi_acc_vdev->dev_opened != DEV_OPEN) {
+		seq_printf(seq, "%s\n", "device not opened!");
+		return -EINVAL;
+	}
+
+	ret = qm_wait_dev_not_ready(vf_qm);
+	if (ret) {
+		seq_printf(seq, "%s\n", "VF device not ready!");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int hisi_acc_vf_debug_cmd(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_qm *vf_qm = &hisi_acc_vdev->vf_qm;
+	u64 value;
+	int ret;
+
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
+	ret = hisi_acc_vf_debug_check(seq, vdev);
+	if (ret) {
+		mutex_unlock(&hisi_acc_vdev->enable_mutex);
+		return 0;
+	}
+
+	value = readl(vf_qm->io_base + QM_MB_CMD_SEND_BASE);
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
+	seq_printf(seq, "%s:0x%llx\n", "mailbox cmd channel state is OK", value);
+
+	return 0;
+}
+
+static int hisi_acc_vf_debug_save(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *migf = hisi_acc_vdev->debug_migf;
+	struct acc_vf_data *vf_data = &migf->vf_data;
+	int ret;
+
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
+	ret = hisi_acc_vf_debug_check(seq, vdev);
+	if (ret) {
+		mutex_unlock(&hisi_acc_vdev->enable_mutex);
+		return 0;
+	}
+
+	vf_data->vf_qm_state = QM_READY;
+	ret = vf_qm_read_data(&hisi_acc_vdev->vf_qm, vf_data);
+	if (ret) {
+		mutex_unlock(&hisi_acc_vdev->enable_mutex);
+		seq_printf(seq, "%s\n", "failed to save device data!");
+		return 0;
+	}
+
+	migf->total_length = sizeof(struct acc_vf_data);
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
+	seq_printf(seq, "%s\n", "successful to save device data!");
+
+	return 0;
+}
+
+static int hisi_acc_vf_data_read(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *debug_migf = hisi_acc_vdev->debug_migf;
+	size_t vf_data_sz = offsetofend(struct acc_vf_data, padding);
+
+	if (debug_migf && debug_migf->total_length)
+		seq_hex_dump(seq, "Mig Data:", DUMP_PREFIX_OFFSET, 16, 1,
+				(unsigned char *)&debug_migf->vf_data,
+				vf_data_sz, false);
+	else
+		seq_printf(seq, "%s\n", "device not migrated!");
+
+	return 0;
+}
+
+static int hisi_acc_vf_attr_read(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *debug_migf = hisi_acc_vdev->debug_migf;
+
+	if (debug_migf && debug_migf->total_length) {
+		seq_printf(seq,
+			 "acc device:\n"
+			 "device  state: %d\n"
+			 "device  ready: %u\n"
+			 "data    valid: %d\n"
+			 "data     size: %lu\n",
+			 hisi_acc_vdev->mig_state,
+			 hisi_acc_vdev->vf_qm_state,
+			 debug_migf->disabled,
+			 debug_migf->total_length);
+	} else {
+		seq_printf(seq, "%s\n", "device not migrated!");
+	}
+
+	return 0;
+}
+
+static int hisi_acc_vfio_debug_init(struct hisi_acc_vf_core_device *hisi_acc_vdev)
+{
+	struct vfio_device *vdev = &hisi_acc_vdev->core_device.vdev;
+	struct dentry *vfio_dev_migration = NULL;
+	struct dentry *vfio_hisi_acc = NULL;
+	struct device *dev = vdev->dev;
+	void *migf = NULL;
+
+	if (!debugfs_initialized())
+		return 0;
+
+	migf = kzalloc(sizeof(struct hisi_acc_vf_migration_file), GFP_KERNEL);
+	if (!migf)
+		return -ENOMEM;
+	hisi_acc_vdev->debug_migf = migf;
+
+	vfio_dev_migration = debugfs_lookup("migration", vdev->debug_root);
+	if (!vfio_dev_migration) {
+		kfree(migf);
+		dev_err(dev, "failed to lookup migration debugfs file!\n");
+		return -ENODEV;
+	}
+
+	vfio_hisi_acc = debugfs_create_dir("hisi_acc", vfio_dev_migration);
+	debugfs_create_devm_seqfile(dev, "data", vfio_hisi_acc,
+				  hisi_acc_vf_data_read);
+	debugfs_create_devm_seqfile(dev, "attr", vfio_hisi_acc,
+				  hisi_acc_vf_attr_read);
+	debugfs_create_devm_seqfile(dev, "cmd_state", vfio_hisi_acc,
+				  hisi_acc_vf_debug_cmd);
+	debugfs_create_devm_seqfile(dev, "save", vfio_hisi_acc,
+				  hisi_acc_vf_debug_save);
+
+	return 0;
+}
+
+static void hisi_acc_vf_debugfs_exit(struct hisi_acc_vf_core_device *hisi_acc_vdev)
+{
+	if (!debugfs_initialized())
+		return;
+
+	kfree(hisi_acc_vdev->debug_migf);
+}
+
 static int hisi_acc_vfio_pci_open_device(struct vfio_device *core_vdev)
 {
 	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(core_vdev);
@@ -1311,9 +1506,11 @@ static int hisi_acc_vfio_pci_open_device(struct vfio_device *core_vdev)
 			return ret;
 		}
 		hisi_acc_vdev->mig_state = VFIO_DEVICE_STATE_RUNNING;
+		hisi_acc_vdev->dev_opened = DEV_OPEN;
 	}
 
 	vfio_pci_core_finish_enable(vdev);
+
 	return 0;
 }
 
@@ -1322,7 +1519,10 @@ 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_vdev->dev_opened = DEV_CLOSE;
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
 	iounmap(vf_qm->io_base);
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
 	vfio_pci_core_close_device(core_vdev);
 }
 
@@ -1413,6 +1613,9 @@ static int hisi_acc_vfio_pci_probe(struct pci_dev *pdev, const struct pci_device
 	ret = vfio_pci_core_register_device(&hisi_acc_vdev->core_device);
 	if (ret)
 		goto out_put_vdev;
+
+	if (ops == &hisi_acc_vfio_pci_migrn_ops)
+		hisi_acc_vfio_debug_init(hisi_acc_vdev);
 	return 0;
 
 out_put_vdev:
@@ -1425,6 +1628,7 @@ static void hisi_acc_vfio_pci_remove(struct pci_dev *pdev)
 	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_drvdata(pdev);
 
 	vfio_pci_core_unregister_device(&hisi_acc_vdev->core_device);
+	hisi_acc_vf_debugfs_exit(hisi_acc_vdev);
 	vfio_put_device(&hisi_acc_vdev->core_device.vdev);
 }
 
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h +11 -0
index 7a9dc87627cd..3a20d81d105c 100644
@@ -52,6 +52,11 @@
 #define QM_EQC_DW0		0X8000
 #define QM_AEQC_DW0		0X8020
 
+enum acc_dev_state {
+	DEV_CLOSE = 0x0,
+	DEV_OPEN,
+};
+
 struct acc_vf_data {
 #define QM_MATCH_SIZE offsetofend(struct acc_vf_data, qm_rsv_state)
 	/* QM match information */
@@ -114,5 +119,11 @@ struct hisi_acc_vf_core_device {
 	int vf_id;
 	struct hisi_acc_vf_migration_file *resuming_migf;
 	struct hisi_acc_vf_migration_file *saving_migf;
+
+	/* To make sure the device is enabled */
+	struct mutex enable_mutex;
+	bool dev_opened;
+	/* For debugfs */
+	struct hisi_acc_vf_migration_file *debug_migf;
 };
 #endif /* HISI_ACC_VFIO_PCI_H */
Signature · 1 line
2.24.0
#6 lore raw

Re: [PATCH v4 3/4] hisi_acc_vfio_pci: register debugfs for hisilicon migration driver

Quoted 60 lines · On Tue, 2 Apr 2024 11:24:31 +0800 Longfang Liu <[email protected]> wrote:

On the debugfs framework of VFIO, if the CONFIG_VFIO_DEBUGFS macro is enabled, the debug function is registered for the live migration driver of the HiSilicon accelerator device.

After registering the HiSilicon accelerator device on the debugfs framework of live migration of vfio, a directory file "hisi_acc" of debugfs is created, and then three debug function files are created in this directory:

   vfio
    |
    +---<dev_name1>
    |    +---migration
    |        +--state
    |        +--hisi_acc
    |            +--attr
    |            +--data
    |            +--save
    |            +--cmd_state
    |
    +---<dev_name2>
         +---migration
             +--state
             +--hisi_acc
                 +--attr
                 +--data
                 +--save
                 +--cmd_state

data file: used to get the migration data from the driver attr file: used to get device attributes parameters from the driver save file: used to read the data of the live migration device and save it to the driver.
cmd_state: used to get the cmd channel state for the device.

+----------------+        +--------------+       +---------------+
| migration dev  |        |   src  dev   |       |   dst  dev    |
+-------+--------+        +------+-------+       +-------+-------+
        |                        |                       |
        |                        |                       |
        |                        |                       |
        |                        |                       |
  save  |                 +------v-------+       +-------v-------+
        |                 |  saving_mif  |       | resuming_migf |
        |                 |     file     |       |     file      |
        |                 +------+-------+       +-------+-------+
        |                        |                       |
        |        mutex           |                       |
+-------v--------+               |                       |
|                |               |                       |
| debug_migf file<---------------+-----------------------+
|                |             copy
+-------+--------+
        |
   cat  |
        |
+-------v--------+
|     user       |
+----------------+
Quoted 7 lines

In debugfs scheme. The driver creates a separate debug_migf file. It is completely separated from the two files of live migration, thus preventing debugfs data from interfering with migration data. Moreover, it only performs read operations on the device.

For serialization of debugfs:
First, it only writes data when performing a debugfs save operation.

This distinction between "writing" and "copying" is very confusing.

Quoted 2 lines

Second, it is only copied from the file on the migration device when the live migration is complete.

Why does it do this at all? If you're looking for a postmortem of the user generated buffer, that should be explicitly stated.

Quoted 1 line

These two operations are mutually exclusive through mutex.

The mutual exclusion between debugfs operations is not the concern, the question is whether there's serialization that prevents the debugfs operations from interfering with the user migration flow. Nothing here seems to prevent concurrent use of the debugfs interface proposed here with a user migration.

Quoted 12 lines
---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 204 ++++++++++++++++++
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.h    |  11 +
 2 files changed, 215 insertions(+)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +0 -0
index bf358ba94b5d..9f563a31a2a1 100644

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c line 18

Quoted 4 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
@@ -15,6 +15,7 @@
 #include <linux/anon_inodes.h>
 
 #include "hisi_acc_vfio_pci.h"
+#include "../../vfio.h"

This include seems not to be required.

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c hisi_acc_check_int_state(struct hisi_acc_vf_core_device *hisi_acc_vdev) · line 623

Quoted 9 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +2 -0
 /* Return 0 on VM acc device ready, -ETIMEDOUT hardware timeout */
 static int qm_wait_dev_not_ready(struct hisi_qm *qm)
@@ -618,6 +619,22 @@ hisi_acc_check_int_state(struct hisi_acc_vf_core_device *hisi_acc_vdev)
 	}
 }
 
+static void hisi_acc_vf_migf_save(struct hisi_acc_vf_core_device *hisi_acc_vdev,
+	struct hisi_acc_vf_migration_file *src_migf)

Seems this should be named something relative to debug since it's only purpose is to copy the migration file to the debug migration file.

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c hisi_acc_check_int_state(struct hisi_acc_vf_core_device *hisi_acc_vdev) · line 631

Quoted 8 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +8 -0
+{
+	struct hisi_acc_vf_migration_file *dst_migf = hisi_acc_vdev->debug_migf;
+
+	if (!dst_migf)
+		return;
+
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
+	dst_migf->disabled = src_migf->disabled;

In the cases where this is called, the caller is about to call hisi_acc_vf_disable_fd() which sets disabled = true and then hisi_acc_vf_debug_save() doesn't touch the value! What does it even mean to copy this value, let alone print it as part of the debugfs output later?

Quoted 11 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +6 -0
+	dst_migf->total_length = src_migf->total_length;
+	memcpy(&dst_migf->vf_data, &src_migf->vf_data,
+		    sizeof(struct acc_vf_data));
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
+}
+
 static void hisi_acc_vf_disable_fd(struct hisi_acc_vf_migration_file *migf)
 {
 	mutex_lock(&migf->lock);
@@ -630,12 +647,14 @@ static void hisi_acc_vf_disable_fd(struct hisi_acc_vf_migration_file *migf)
 static void hisi_acc_vf_disable_fds(struct hisi_acc_vf_core_device *hisi_acc_vdev)

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static void hisi_acc_vf_disable_fd(struct hisi_acc_vf_migration_file *migf) · line 657

Quoted 10 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +2 -0
 {
 	if (hisi_acc_vdev->resuming_migf) {
+		hisi_acc_vf_migf_save(hisi_acc_vdev, hisi_acc_vdev->resuming_migf);
 		hisi_acc_vf_disable_fd(hisi_acc_vdev->resuming_migf);
 		fput(hisi_acc_vdev->resuming_migf->filp);
 		hisi_acc_vdev->resuming_migf = NULL;
 	}
 
 	if (hisi_acc_vdev->saving_migf) {
+		hisi_acc_vf_migf_save(hisi_acc_vdev, hisi_acc_vdev->saving_migf);

Why are these buffers copied to the debug_migf in this case? This can happen asynchronous to accessing the debugfs migration file and there's no serialization.

Quoted 12 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
 		hisi_acc_vf_disable_fd(hisi_acc_vdev->saving_migf);
 		fput(hisi_acc_vdev->saving_migf->filp);
 		hisi_acc_vdev->saving_migf = NULL;
@@ -1144,6 +1163,7 @@ static int hisi_acc_vf_qm_init(struct hisi_acc_vf_core_device *hisi_acc_vdev)
 	if (!vf_qm->io_base)
 		return -EIO;
 
+	mutex_init(&hisi_acc_vdev->enable_mutex);
 	vf_qm->fun_type = QM_HW_VF;
 	vf_qm->pdev = vf_dev;
 	mutex_init(&vf_qm->mailbox_lock);

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1323

Quoted 10 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +7 -0
@@ -1294,6 +1314,181 @@ static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int
 	return vfio_pci_core_ioctl(core_vdev, cmd, arg);
 }
 
+static int hisi_acc_vf_debug_check(struct seq_file *seq, struct vfio_device *vdev)
+{
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *migf = hisi_acc_vdev->debug_migf;
+	struct hisi_qm *vf_qm = &hisi_acc_vdev->vf_qm;
+	int ret;
+

lockdep_assert_held(...)

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1326

Quoted 3 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +3 -0
+	if (!vdev->mig_ops || !migf) {
+		seq_printf(seq, "%s\n", "device does not support live migration!");
+		return -EINVAL;

Isn't the -EINVAL sufficient?

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1333

Quoted 7 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +7 -0
+	}
+
+	/**
+	 * When the device is not opened, the io_base is not mapped.
+	 * The driver cannot perform device read and write operations.
+	 */
+	if (hisi_acc_vdev->dev_opened != DEV_OPEN) {

Why are we assigning and testing a bool against and enum?!

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1341

Quoted 8 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +8 -0
+		seq_printf(seq, "%s\n", "device not opened!");
+		return -EINVAL;
+	}
+
+	ret = qm_wait_dev_not_ready(vf_qm);
+	if (ret) {
+		seq_printf(seq, "%s\n", "VF device not ready!");
+		return -EINVAL;

-EBUSY? Again, not sure why we need the seq_printf() in addition to the error value.

Quoted 10 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +10 -0
+	}
+
+	return 0;
+}
+
+static int hisi_acc_vf_debug_cmd(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1361

Quoted 10 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +10 -0
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_qm *vf_qm = &hisi_acc_vdev->vf_qm;
+	u64 value;
+	int ret;
+
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
+	ret = hisi_acc_vf_debug_check(seq, vdev);
+	if (ret) {
+		mutex_unlock(&hisi_acc_vdev->enable_mutex);
+		return 0;

Why do we squash the error return here and throughout?

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1366

Quoted 5 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +5 -0
+	}
+
+	value = readl(vf_qm->io_base + QM_MB_CMD_SEND_BASE);
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
+	seq_printf(seq, "%s:0x%llx\n", "mailbox cmd channel state is OK", value);

We didn't test the value, what makes the state OK? Can this readl() or those in qm_wait_dev_not_ready() interfere with the main device flow?

Quoted 41 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +41 -0
+
+	return 0;
+}
+
+static int hisi_acc_vf_debug_save(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *migf = hisi_acc_vdev->debug_migf;
+	struct acc_vf_data *vf_data = &migf->vf_data;
+	int ret;
+
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
+	ret = hisi_acc_vf_debug_check(seq, vdev);
+	if (ret) {
+		mutex_unlock(&hisi_acc_vdev->enable_mutex);
+		return 0;
+	}
+
+	vf_data->vf_qm_state = QM_READY;
+	ret = vf_qm_read_data(&hisi_acc_vdev->vf_qm, vf_data);
+	if (ret) {
+		mutex_unlock(&hisi_acc_vdev->enable_mutex);
+		seq_printf(seq, "%s\n", "failed to save device data!");
+		return 0;
+	}
+
+	migf->total_length = sizeof(struct acc_vf_data);
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
+	seq_printf(seq, "%s\n", "successful to save device data!");
+
+	return 0;
+}
+
+static int hisi_acc_vf_data_read(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1417

Quoted 10 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +10 -0
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *debug_migf = hisi_acc_vdev->debug_migf;
+	size_t vf_data_sz = offsetofend(struct acc_vf_data, padding);
+
+	if (debug_migf && debug_migf->total_length)
+		seq_hex_dump(seq, "Mig Data:", DUMP_PREFIX_OFFSET, 16, 1,
+				(unsigned char *)&debug_migf->vf_data,
+				vf_data_sz, false);
+	else
+		seq_printf(seq, "%s\n", "device not migrated!");

"device state not saved"? Although I don't recall why this doesn't just return an errno.

Quoted 10 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +10 -0
+
+	return 0;
+}
+
+static int hisi_acc_vf_attr_read(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1437

Quoted 10 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +10 -0
+	struct hisi_acc_vf_migration_file *debug_migf = hisi_acc_vdev->debug_migf;
+
+	if (debug_migf && debug_migf->total_length) {
+		seq_printf(seq,
+			 "acc device:\n"
+			 "device  state: %d\n"
+			 "device  ready: %u\n"
+			 "data    valid: %d\n"
+			 "data     size: %lu\n",
+			 hisi_acc_vdev->mig_state,

This is redundant to migration/state, however note hisi_acc_vfio_pci_get_device_state() protects the value with state mutex while reading it.

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1438

Quoted 1 line
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
+			 hisi_acc_vdev->vf_qm_state,

What's the purpose of this, it's ready or not, what does that mean?

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1439

Quoted 1 line
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
+			 debug_migf->disabled,

What's the purpose of this?

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1440

Quoted 1 line
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
+			 debug_migf->total_length);

Why not just have this printed or inferred via the above data_read function, this all seems unnecessary.

Quoted 14 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +14 -0
+	} else {
+		seq_printf(seq, "%s\n", "device not migrated!");
+	}
+
+	return 0;
+}
+
+static int hisi_acc_vfio_debug_init(struct hisi_acc_vf_core_device *hisi_acc_vdev)
+{
+	struct vfio_device *vdev = &hisi_acc_vdev->core_device.vdev;
+	struct dentry *vfio_dev_migration = NULL;
+	struct dentry *vfio_hisi_acc = NULL;
+	struct device *dev = vdev->dev;
+	void *migf = NULL;

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1464

Quoted 10 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +10 -0
+
+	if (!debugfs_initialized())
+		return 0;
+
+	migf = kzalloc(sizeof(struct hisi_acc_vf_migration_file), GFP_KERNEL);
+	if (!migf)
+		return -ENOMEM;
+	hisi_acc_vdev->debug_migf = migf;
+
+	vfio_dev_migration = debugfs_lookup("migration", vdev->debug_root);

Fails to build without CONFIG_DEBUG_FS=y There should be a dependency on CONFIG_VFIO_DEBUGFS here somewhere.

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1466

Quoted 2 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +2 -0
+	if (!vfio_dev_migration) {
+		kfree(migf);

hisi_acc_vdev->debug_migf still points at this freed buffer, the return value of this function is not tested, allows a use-after-free in all of the below debugfs interfaces.

Quoted 13 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +13 -0
+		dev_err(dev, "failed to lookup migration debugfs file!\n");
+		return -ENODEV;
+	}
+
+	vfio_hisi_acc = debugfs_create_dir("hisi_acc", vfio_dev_migration);
+	debugfs_create_devm_seqfile(dev, "data", vfio_hisi_acc,
+				  hisi_acc_vf_data_read);
+	debugfs_create_devm_seqfile(dev, "attr", vfio_hisi_acc,
+				  hisi_acc_vf_attr_read);
+	debugfs_create_devm_seqfile(dev, "cmd_state", vfio_hisi_acc,
+				  hisi_acc_vf_debug_cmd);
+	debugfs_create_devm_seqfile(dev, "save", vfio_hisi_acc,
+				  hisi_acc_vf_debug_save);

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1489

Quoted 10 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +10 -0
+
+	return 0;
+}
+
+static void hisi_acc_vf_debugfs_exit(struct hisi_acc_vf_core_device *hisi_acc_vdev)
+{
+	if (!debugfs_initialized())
+		return;
+
+	kfree(hisi_acc_vdev->debug_migf);

Double free if the lookup in init fails.

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static int hisi_acc_vfio_pci_open_device(struct vfio_device *core_vdev) · line 1509

Quoted 10 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +3 -0
+}
+
 static int hisi_acc_vfio_pci_open_device(struct vfio_device *core_vdev)
 {
 	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(core_vdev);
@@ -1311,9 +1506,11 @@ static int hisi_acc_vfio_pci_open_device(struct vfio_device *core_vdev)
 			return ret;
 		}
 		hisi_acc_vdev->mig_state = VFIO_DEVICE_STATE_RUNNING;
+		hisi_acc_vdev->dev_opened = DEV_OPEN;

= true!

Quoted 64 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +8 -0
 	}
 
 	vfio_pci_core_finish_enable(vdev);
+
 	return 0;
 }
 
@@ -1322,7 +1519,10 @@ 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_vdev->dev_opened = DEV_CLOSE;
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
 	iounmap(vf_qm->io_base);
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
 	vfio_pci_core_close_device(core_vdev);
 }
 
@@ -1413,6 +1613,9 @@ static int hisi_acc_vfio_pci_probe(struct pci_dev *pdev, const struct pci_device
 	ret = vfio_pci_core_register_device(&hisi_acc_vdev->core_device);
 	if (ret)
 		goto out_put_vdev;
+
+	if (ops == &hisi_acc_vfio_pci_migrn_ops)
+		hisi_acc_vfio_debug_init(hisi_acc_vdev);
 	return 0;
 
 out_put_vdev:
@@ -1425,6 +1628,7 @@ static void hisi_acc_vfio_pci_remove(struct pci_dev *pdev)
 	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_drvdata(pdev);
 
 	vfio_pci_core_unregister_device(&hisi_acc_vdev->core_device);
+	hisi_acc_vf_debugfs_exit(hisi_acc_vdev);
 	vfio_put_device(&hisi_acc_vdev->core_device.vdev);
 }
 
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h +11 -0
index 7a9dc87627cd..3a20d81d105c 100644
@@ -52,6 +52,11 @@
 #define QM_EQC_DW0		0X8000
 #define QM_AEQC_DW0		0X8020
 
+enum acc_dev_state {
+	DEV_CLOSE = 0x0,
+	DEV_OPEN,
+};
+
 struct acc_vf_data {
 #define QM_MATCH_SIZE offsetofend(struct acc_vf_data, qm_rsv_state)
 	/* QM match information */
@@ -114,5 +119,11 @@ struct hisi_acc_vf_core_device {
 	int vf_id;
 	struct hisi_acc_vf_migration_file *resuming_migf;
 	struct hisi_acc_vf_migration_file *saving_migf;
+
+	/* To make sure the device is enabled */
+	struct mutex enable_mutex;
+	bool dev_opened;
+	/* For debugfs */
+	struct hisi_acc_vf_migration_file *debug_migf;
 };
 #endif /* HISI_ACC_VFIO_PCI_H */
author maintainer
#7 lore raw

Re: [PATCH v4 3/4] hisi_acc_vfio_pci: register debugfs for hisilicon migration driver

Quoted 68 lines · Alex Williamson
Quoted 65 lines · On Tue, 2 Apr 2024 11:24:31 +0800 Longfang Liu <[email protected]> wrote:

On the debugfs framework of VFIO, if the CONFIG_VFIO_DEBUGFS macro is enabled, the debug function is registered for the live migration driver of the HiSilicon accelerator device.

After registering the HiSilicon accelerator device on the debugfs framework of live migration of vfio, a directory file "hisi_acc" of debugfs is created, and then three debug function files are created in this directory:

   vfio
    |
    +---<dev_name1>
    |    +---migration
    |        +--state
    |        +--hisi_acc
    |            +--attr
    |            +--data
    |            +--save
    |            +--cmd_state
    |
    +---<dev_name2>
         +---migration
             +--state
             +--hisi_acc
                 +--attr
                 +--data
                 +--save
                 +--cmd_state

data file: used to get the migration data from the driver attr file: used to get device attributes parameters from the driver save file: used to read the data of the live migration device and save it to the driver.
cmd_state: used to get the cmd channel state for the device.

+----------------+        +--------------+       +---------------+
| migration dev  |        |   src  dev   |       |   dst  dev    |
+-------+--------+        +------+-------+       +-------+-------+
        |                        |                       |
        |                        |                       |
        |                        |                       |
        |                        |                       |
  save  |                 +------v-------+       +-------v-------+
        |                 |  saving_mif  |       | resuming_migf |
        |                 |     file     |       |     file      |
        |                 +------+-------+       +-------+-------+
        |                        |                       |
        |        mutex           |                       |
+-------v--------+               |                       |
|                |               |                       |
| debug_migf file<---------------+-----------------------+
|                |             copy
+-------+--------+
        |
   cat  |
        |
+-------v--------+
|     user       |
+----------------+

In debugfs scheme. The driver creates a separate debug_migf file. It is completely separated from the two files of live migration, thus preventing debugfs data from interfering with migration data. Moreover, it only performs read operations on the device.

Quoted 5 lines
Quoted 2 lines

For serialization of debugfs:
First, it only writes data when performing a debugfs save operation.

This distinction between "writing" and "copying" is very confusing.

"Writing" means reading data from the device and writing it to debug_migf. "Copying" is to copy the data that has been saved in other migf files to debug_migf.
The destination of both operations is the same.
The data sources are different.

Quoted 6 lines
Quoted 2 lines

Second, it is only copied from the file on the migration device when the live migration is complete.

Why does it do this at all? If you're looking for a postmortem of the user generated buffer, that should be explicitly stated.

debug_migf is a data buffer. Used to cache debugfs data for users

Quoted 8 lines
Quoted 1 line

These two operations are mutually exclusive through mutex.

The mutual exclusion between debugfs operations is not the concern, the question is whether there's serialization that prevents the debugfs operations from interfering with the user migration flow. Nothing here seems to prevent concurrent use of the debugfs interface proposed here with a user migration.

Reading data directly from the device does not cause any problems. The device supports concurrent requests for read operations. Therefore, there is no mutual exclusion between the debugfs save operation and the user migration operation.

Quoted 14 lines
Quoted 14 lines
---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 204 ++++++++++++++++++
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.h    |  11 +
 2 files changed, 215 insertions(+)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +0 -0
index bf358ba94b5d..9f563a31a2a1 100644
@@ -15,6 +15,7 @@
 #include <linux/anon_inodes.h>
 
Quoted 5 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c line 18

Quoted 2 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
 #include "hisi_acc_vfio_pci.h"
+#include "../../vfio.h"

This include seems not to be required.

OK, let me modify it and verify it.

Quoted 7 lines
Quoted 7 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +0 -0
 /* Return 0 on VM acc device ready, -ETIMEDOUT hardware timeout */
 static int qm_wait_dev_not_ready(struct hisi_qm *qm)
@@ -618,6 +619,22 @@ hisi_acc_check_int_state(struct hisi_acc_vf_core_device *hisi_acc_vdev)
 	}
 }
 
Quoted 5 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c hisi_acc_check_int_state(struct hisi_acc_vf_core_device *hisi_acc_vdev) · line 623

Quoted 2 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +2 -0
+static void hisi_acc_vf_migf_save(struct hisi_acc_vf_core_device *hisi_acc_vdev,
+	struct hisi_acc_vf_migration_file *src_migf)

Seems this should be named something relative to debug since it's only purpose is to copy the migration file to the debug migration file.

How about hisi_acc_debug_migf_copy?

Quoted 16 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c hisi_acc_check_int_state(struct hisi_acc_vf_core_device *hisi_acc_vdev) · line 631

Quoted 9 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +8 -0
+{
+	struct hisi_acc_vf_migration_file *dst_migf = hisi_acc_vdev->debug_migf;
+
+	if (!dst_migf)
+		return;
+
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
+	dst_migf->disabled = src_migf->disabled;

In the cases where this is called, the caller is about to call hisi_acc_vf_disable_fd() which sets disabled = true and then hisi_acc_vf_debug_save() doesn't touch the value! What does it even mean to copy this value, let alone print it as part of the debugfs output later?

Yes, the disable assignment of debug_migf needs to be processed in hisi_acc_vf_debug_save.

Quoted 20 lines
Quoted 20 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +7 -0
+	dst_migf->total_length = src_migf->total_length;
+	memcpy(&dst_migf->vf_data, &src_migf->vf_data,
+		    sizeof(struct acc_vf_data));
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
+}
+
 static void hisi_acc_vf_disable_fd(struct hisi_acc_vf_migration_file *migf)
 {
 	mutex_lock(&migf->lock);
@@ -630,12 +647,14 @@ static void hisi_acc_vf_disable_fd(struct hisi_acc_vf_migration_file *migf)
 static void hisi_acc_vf_disable_fds(struct hisi_acc_vf_core_device *hisi_acc_vdev)
 {
 	if (hisi_acc_vdev->resuming_migf) {
+		hisi_acc_vf_migf_save(hisi_acc_vdev, hisi_acc_vdev->resuming_migf);
 		hisi_acc_vf_disable_fd(hisi_acc_vdev->resuming_migf);
 		fput(hisi_acc_vdev->resuming_migf->filp);
 		hisi_acc_vdev->resuming_migf = NULL;
 	}
 
Quoted 7 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static void hisi_acc_vf_disable_fd(struct hisi_acc_vf_migration_file *migf) · line 657

Quoted 2 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
 	if (hisi_acc_vdev->saving_migf) {
+		hisi_acc_vf_migf_save(hisi_acc_vdev, hisi_acc_vdev->saving_migf);

Why are these buffers copied to the debug_migf in this case? This can happen asynchronous to accessing the debugfs migration file and there's no serialization.

We can try to copy when accessing debugfs.

Quoted 25 lines
Quoted 12 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
 		hisi_acc_vf_disable_fd(hisi_acc_vdev->saving_migf);
 		fput(hisi_acc_vdev->saving_migf->filp);
 		hisi_acc_vdev->saving_migf = NULL;
@@ -1144,6 +1163,7 @@ static int hisi_acc_vf_qm_init(struct hisi_acc_vf_core_device *hisi_acc_vdev)
 	if (!vf_qm->io_base)
 		return -EIO;
 
+	mutex_init(&hisi_acc_vdev->enable_mutex);
 	vf_qm->fun_type = QM_HW_VF;
 	vf_qm->pdev = vf_dev;
 	mutex_init(&vf_qm->mailbox_lock);

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1323

Quoted 10 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +7 -0
@@ -1294,6 +1314,181 @@ static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int
 	return vfio_pci_core_ioctl(core_vdev, cmd, arg);
 }
 
+static int hisi_acc_vf_debug_check(struct seq_file *seq, struct vfio_device *vdev)
+{
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *migf = hisi_acc_vdev->debug_migf;
+	struct hisi_qm *vf_qm = &hisi_acc_vdev->vf_qm;
+	int ret;
+

lockdep_assert_held(...)

Quoted 6 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1326

Quoted 3 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +3 -0
+	if (!vdev->mig_ops || !migf) {
+		seq_printf(seq, "%s\n", "device does not support live migration!");
+		return -EINVAL;

Isn't the -EINVAL sufficient?

Which one do you think is better?

Quoted 10 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1333

Quoted 7 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +7 -0
+	}
+
+	/**
+	 * When the device is not opened, the io_base is not mapped.
+	 * The driver cannot perform device read and write operations.
+	 */
+	if (hisi_acc_vdev->dev_opened != DEV_OPEN) {

Why are we assigning and testing a bool against and enum?!

OK, change to true and false assignment.

Quoted 12 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1341

Quoted 8 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +8 -0
+		seq_printf(seq, "%s\n", "device not opened!");
+		return -EINVAL;
+	}
+
+	ret = qm_wait_dev_not_ready(vf_qm);
+	if (ret) {
+		seq_printf(seq, "%s\n", "VF device not ready!");
+		return -EINVAL;

-EBUSY? Again, not sure why we need the seq_printf() in addition to the error value.

OK, -EBUSY is better.
seq_printf() allows users to directly obtain the cause of the error without checking dmesg.

Quoted 14 lines
Quoted 14 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +14 -0
+	}
+
+	return 0;
+}
+
+static int hisi_acc_vf_debug_cmd(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_qm *vf_qm = &hisi_acc_vdev->vf_qm;
+	u64 value;
+	int ret;
Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1361

Quoted 6 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +6 -0
+
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
+	ret = hisi_acc_vf_debug_check(seq, vdev);
+	if (ret) {
+		mutex_unlock(&hisi_acc_vdev->enable_mutex);
+		return 0;

Why do we squash the error return here and throughout?

seq_printf() gives the user a failure message.
This can be changed to "return ret;"

Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1366

Quoted 5 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +5 -0
+	}
+
+	value = readl(vf_qm->io_base + QM_MB_CMD_SEND_BASE);
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
+	seq_printf(seq, "%s:0x%llx\n", "mailbox cmd channel state is OK", value);

We didn't test the value, what makes the state OK? Can this readl() or those in qm_wait_dev_not_ready() interfere with the main device flow?

If the cmd channel is normal, it will return a non-all-F value. Add exception checking in the next version.

cmd channel read operation will not affect the main migration process.

Quoted 46 lines
Quoted 46 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +46 -0
+
+	return 0;
+}
+
+static int hisi_acc_vf_debug_save(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *migf = hisi_acc_vdev->debug_migf;
+	struct acc_vf_data *vf_data = &migf->vf_data;
+	int ret;
+
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
+	ret = hisi_acc_vf_debug_check(seq, vdev);
+	if (ret) {
+		mutex_unlock(&hisi_acc_vdev->enable_mutex);
+		return 0;
+	}
+
+	vf_data->vf_qm_state = QM_READY;
+	ret = vf_qm_read_data(&hisi_acc_vdev->vf_qm, vf_data);
+	if (ret) {
+		mutex_unlock(&hisi_acc_vdev->enable_mutex);
+		seq_printf(seq, "%s\n", "failed to save device data!");
+		return 0;
+	}
+
+	migf->total_length = sizeof(struct acc_vf_data);
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
+	seq_printf(seq, "%s\n", "successful to save device data!");
+
+	return 0;
+}
+
+static int hisi_acc_vf_data_read(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *debug_migf = hisi_acc_vdev->debug_migf;
+	size_t vf_data_sz = offsetofend(struct acc_vf_data, padding);
+
+	if (debug_migf && debug_migf->total_length)
Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1417

Quoted 5 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +5 -0
+		seq_hex_dump(seq, "Mig Data:", DUMP_PREFIX_OFFSET, 16, 1,
+				(unsigned char *)&debug_migf->vf_data,
+				vf_data_sz, false);
+	else
+		seq_printf(seq, "%s\n", "device not migrated!");

"device state not saved"? Although I don't recall why this doesn't just return an errno.

OK, those who exit directly without migration will be processed according to the error mode and an error code will be returned.

Quoted 14 lines
Quoted 14 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +14 -0
+
+	return 0;
+}
+
+static int hisi_acc_vf_attr_read(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *debug_migf = hisi_acc_vdev->debug_migf;
+
+	if (debug_migf && debug_migf->total_length) {
+		seq_printf(seq,
Quoted 8 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1437

Quoted 6 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +6 -0
+			 "acc device:\n"
+			 "device  state: %d\n"
+			 "device  ready: %u\n"
+			 "data    valid: %d\n"
+			 "data     size: %lu\n",
+			 hisi_acc_vdev->mig_state,

This is redundant to migration/state, however note

OK!

Quoted 7 lines

hisi_acc_vfio_pci_get_device_state() protects the value with state mutex while reading it.

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1438

Quoted 1 line
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
+			 hisi_acc_vdev->vf_qm_state,

What's the purpose of this, it's ready or not, what does that mean?

If this status is not ready, the live migration process will exit directly. It indicates that there are two possible exceptions:
1. The acc device driver in the VM does not have insmod.
2. The acc device driver in the VM is insmoded, but the cmd channel is abnormal.

Quoted 3 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1439

Quoted 1 line
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
+			 debug_migf->disabled,

What's the purpose of this?

Get the enable status of migf file in the driver.

Quoted 6 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1440

Quoted 2 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
+			 debug_migf->total_length);

Why not just have this printed or inferred via the above data_read function, this all seems unnecessary.

This is used to obtain some key status of the live migration driver. It is more important than data in problem location.
So it is output in key-value pairs.

The above data is directly output in the form of hexadecimal data. It is used for more detailed analysis when there are no abnormalities in the key status.

Quoted 19 lines
Quoted 19 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +19 -0
+	} else {
+		seq_printf(seq, "%s\n", "device not migrated!");
+	}
+
+	return 0;
+}
+
+static int hisi_acc_vfio_debug_init(struct hisi_acc_vf_core_device *hisi_acc_vdev)
+{
+	struct vfio_device *vdev = &hisi_acc_vdev->core_device.vdev;
+	struct dentry *vfio_dev_migration = NULL;
+	struct dentry *vfio_hisi_acc = NULL;
+	struct device *dev = vdev->dev;
+	void *migf = NULL;
+
+	if (!debugfs_initialized())
+		return 0;
+
+	migf = kzalloc(sizeof(struct hisi_acc_vf_migration_file), GFP_KERNEL);
Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1464

Quoted 5 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +5 -0
+	if (!migf)
+		return -ENOMEM;
+	hisi_acc_vdev->debug_migf = migf;
+
+	vfio_dev_migration = debugfs_lookup("migration", vdev->debug_root);

Fails to build without CONFIG_DEBUG_FS=y There should be a dependency on CONFIG_VFIO_DEBUGFS here somewhere.

Yes, the driver needs to add "!IS_ENABLED(CONFIG_VFIO_DEBUGFS)" behind "debugfs_initialized()" above.

Quoted 7 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1466

Quoted 2 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +2 -0
+	if (!vfio_dev_migration) {
+		kfree(migf);

hisi_acc_vdev->debug_migf still points at this freed buffer, the return value of this function is not tested, allows a use-after-free in all of the below debugfs interfaces.

Yes, there needs to add "hisi_acc_vdev->debug_migf = NULL"

Quoted 17 lines
Quoted 17 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +17 -0
+		dev_err(dev, "failed to lookup migration debugfs file!\n");
+		return -ENODEV;
+	}
+
+	vfio_hisi_acc = debugfs_create_dir("hisi_acc", vfio_dev_migration);
+	debugfs_create_devm_seqfile(dev, "data", vfio_hisi_acc,
+				  hisi_acc_vf_data_read);
+	debugfs_create_devm_seqfile(dev, "attr", vfio_hisi_acc,
+				  hisi_acc_vf_attr_read);
+	debugfs_create_devm_seqfile(dev, "cmd_state", vfio_hisi_acc,
+				  hisi_acc_vf_debug_cmd);
+	debugfs_create_devm_seqfile(dev, "save", vfio_hisi_acc,
+				  hisi_acc_vf_debug_save);
+
+	return 0;
+}
+
Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1489

Quoted 6 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +6 -0
+static void hisi_acc_vf_debugfs_exit(struct hisi_acc_vf_core_device *hisi_acc_vdev)
+{
+	if (!debugfs_initialized())
+		return;
+
+	kfree(hisi_acc_vdev->debug_migf);

Double free if the lookup in init fails.

After adding "hisi_acc_vdev->debug_migf = NULL" above. These processing codes need to be added here:

if (hisi_acc_vdev->debug_migf)

	kfree(hisi_acc_vdev->debug_migf);
Quoted 4 lines
Quoted 4 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +2 -0
+}
+
 static int hisi_acc_vfio_pci_open_device(struct vfio_device *core_vdev)
 {
Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static int hisi_acc_vfio_pci_open_device(struct vfio_device *core_vdev) · line 1509

Quoted 6 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
 	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(core_vdev);
@@ -1311,9 +1506,11 @@ static int hisi_acc_vfio_pci_open_device(struct vfio_device *core_vdev)
 			return ret;
 		}
 		hisi_acc_vdev->mig_state = VFIO_DEVICE_STATE_RUNNING;
+		hisi_acc_vdev->dev_opened = DEV_OPEN;

= true!

OK, the next version will not use enumeration values. will use true/false assignment.

Quoted 58 lines
Quoted 58 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +8 -0
 	}
 
 	vfio_pci_core_finish_enable(vdev);
+
 	return 0;
 }
 
@@ -1322,7 +1519,10 @@ 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_vdev->dev_opened = DEV_CLOSE;
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
 	iounmap(vf_qm->io_base);
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
 	vfio_pci_core_close_device(core_vdev);
 }
 
@@ -1413,6 +1613,9 @@ static int hisi_acc_vfio_pci_probe(struct pci_dev *pdev, const struct pci_device
 	ret = vfio_pci_core_register_device(&hisi_acc_vdev->core_device);
 	if (ret)
 		goto out_put_vdev;
+
+	if (ops == &hisi_acc_vfio_pci_migrn_ops)
+		hisi_acc_vfio_debug_init(hisi_acc_vdev);
 	return 0;
 
 out_put_vdev:
@@ -1425,6 +1628,7 @@ static void hisi_acc_vfio_pci_remove(struct pci_dev *pdev)
 	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_drvdata(pdev);
 
 	vfio_pci_core_unregister_device(&hisi_acc_vdev->core_device);
+	hisi_acc_vf_debugfs_exit(hisi_acc_vdev);
 	vfio_put_device(&hisi_acc_vdev->core_device.vdev);
 }
 
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h +7 -0
index 7a9dc87627cd..3a20d81d105c 100644
@@ -52,6 +52,11 @@
 #define QM_EQC_DW0		0X8000
 #define QM_AEQC_DW0		0X8020
 
+enum acc_dev_state {
+	DEV_CLOSE = 0x0,
+	DEV_OPEN,
+};
+
 struct acc_vf_data {
 #define QM_MATCH_SIZE offsetofend(struct acc_vf_data, qm_rsv_state)
 	/* QM match information */
@@ -114,5 +119,11 @@ struct hisi_acc_vf_core_device {
 	int vf_id;
 	struct hisi_acc_vf_migration_file *resuming_migf;
 	struct hisi_acc_vf_migration_file *saving_migf;
+
+	/* To make sure the device is enabled */
Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h struct hisi_acc_vf_core_device { · line 129

Quoted 6 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h +4 -0
+	struct mutex enable_mutex;
+	bool dev_opened;
+	/* For debugfs */
+	struct hisi_acc_vf_migration_file *debug_migf;
 };
 #endif /* HISI_ACC_VFIO_PCI_H */

.

Thank you very much for your careful inspection. I will revise the inspection issues you mentioned above one by one and publish them in the next version.

Thanks again!
Longfang.

author maintainer
#8 lore raw

Re: [PATCH v4 3/4] hisi_acc_vfio_pci: register debugfs for hisilicon migration driver

Quoted 189 lines · liulongfang
Quoted 68 lines · Alex Williamson
Quoted 65 lines · On Tue, 2 Apr 2024 11:24:31 +0800 Longfang Liu <[email protected]> wrote:

On the debugfs framework of VFIO, if the CONFIG_VFIO_DEBUGFS macro is enabled, the debug function is registered for the live migration driver of the HiSilicon accelerator device.

After registering the HiSilicon accelerator device on the debugfs framework of live migration of vfio, a directory file "hisi_acc" of debugfs is created, and then three debug function files are created in this directory:

   vfio
    |
    +---<dev_name1>
    |    +---migration
    |        +--state
    |        +--hisi_acc
    |            +--attr
    |            +--data
    |            +--save
    |            +--cmd_state
    |
    +---<dev_name2>
         +---migration
             +--state
             +--hisi_acc
                 +--attr
                 +--data
                 +--save
                 +--cmd_state

data file: used to get the migration data from the driver attr file: used to get device attributes parameters from the driver save file: used to read the data of the live migration device and save it to the driver.
cmd_state: used to get the cmd channel state for the device.

+----------------+        +--------------+       +---------------+
| migration dev  |        |   src  dev   |       |   dst  dev    |
+-------+--------+        +------+-------+       +-------+-------+
        |                        |                       |
        |                        |                       |
        |                        |                       |
        |                        |                       |
  save  |                 +------v-------+       +-------v-------+
        |                 |  saving_mif  |       | resuming_migf |
        |                 |     file     |       |     file      |
        |                 +------+-------+       +-------+-------+
        |                        |                       |
        |        mutex           |                       |
+-------v--------+               |                       |
|                |               |                       |
| debug_migf file<---------------+-----------------------+
|                |             copy
+-------+--------+
        |
   cat  |
        |
+-------v--------+
|     user       |
+----------------+

In debugfs scheme. The driver creates a separate debug_migf file. It is completely separated from the two files of live migration, thus preventing debugfs data from interfering with migration data. Moreover, it only performs read operations on the device.

Quoted 5 lines
Quoted 2 lines

For serialization of debugfs:
First, it only writes data when performing a debugfs save operation.

This distinction between "writing" and "copying" is very confusing.

"Writing" means reading data from the device and writing it to debug_migf. "Copying" is to copy the data that has been saved in other migf files to debug_migf.
The destination of both operations is the same.
The data sources are different.

Quoted 6 lines
Quoted 2 lines

Second, it is only copied from the file on the migration device when the live migration is complete.

Why does it do this at all? If you're looking for a postmortem of the user generated buffer, that should be explicitly stated.

debug_migf is a data buffer. Used to cache debugfs data for users

Quoted 8 lines
Quoted 1 line

These two operations are mutually exclusive through mutex.

The mutual exclusion between debugfs operations is not the concern, the question is whether there's serialization that prevents the debugfs operations from interfering with the user migration flow. Nothing here seems to prevent concurrent use of the debugfs interface proposed here with a user migration.

Reading data directly from the device does not cause any problems. The device supports concurrent requests for read operations. Therefore, there is no mutual exclusion between the debugfs save operation and the user migration operation.

Quoted 14 lines
Quoted 14 lines
---
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c    | 204 ++++++++++++++++++
 .../vfio/pci/hisilicon/hisi_acc_vfio_pci.h    |  11 +
 2 files changed, 215 insertions(+)
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +0 -0
index bf358ba94b5d..9f563a31a2a1 100644
@@ -15,6 +15,7 @@
 #include <linux/anon_inodes.h>
 
Quoted 5 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c line 18

Quoted 2 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
 #include "hisi_acc_vfio_pci.h"
+#include "../../vfio.h"

This include seems not to be required.

OK, let me modify it and verify it.

Quoted 7 lines
Quoted 7 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +0 -0
 /* Return 0 on VM acc device ready, -ETIMEDOUT hardware timeout */
 static int qm_wait_dev_not_ready(struct hisi_qm *qm)
@@ -618,6 +619,22 @@ hisi_acc_check_int_state(struct hisi_acc_vf_core_device *hisi_acc_vdev)
 	}
 }
 
Quoted 5 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c hisi_acc_check_int_state(struct hisi_acc_vf_core_device *hisi_acc_vdev) · line 623

Quoted 2 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +2 -0
+static void hisi_acc_vf_migf_save(struct hisi_acc_vf_core_device *hisi_acc_vdev,
+	struct hisi_acc_vf_migration_file *src_migf)

Seems this should be named something relative to debug since it's only purpose is to copy the migration file to the debug migration file.

How about hisi_acc_debug_migf_copy?

Quoted 16 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c hisi_acc_check_int_state(struct hisi_acc_vf_core_device *hisi_acc_vdev) · line 631

Quoted 9 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +8 -0
+{
+	struct hisi_acc_vf_migration_file *dst_migf = hisi_acc_vdev->debug_migf;
+
+	if (!dst_migf)
+		return;
+
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
+	dst_migf->disabled = src_migf->disabled;

In the cases where this is called, the caller is about to call hisi_acc_vf_disable_fd() which sets disabled = true and then hisi_acc_vf_debug_save() doesn't touch the value! What does it even mean to copy this value, let alone print it as part of the debugfs output later?

Yes, the disable assignment of debug_migf needs to be processed in hisi_acc_vf_debug_save.

Quoted 20 lines
Quoted 20 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +7 -0
+	dst_migf->total_length = src_migf->total_length;
+	memcpy(&dst_migf->vf_data, &src_migf->vf_data,
+		    sizeof(struct acc_vf_data));
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
+}
+
 static void hisi_acc_vf_disable_fd(struct hisi_acc_vf_migration_file *migf)
 {
 	mutex_lock(&migf->lock);
@@ -630,12 +647,14 @@ static void hisi_acc_vf_disable_fd(struct hisi_acc_vf_migration_file *migf)
 static void hisi_acc_vf_disable_fds(struct hisi_acc_vf_core_device *hisi_acc_vdev)
 {
 	if (hisi_acc_vdev->resuming_migf) {
+		hisi_acc_vf_migf_save(hisi_acc_vdev, hisi_acc_vdev->resuming_migf);
 		hisi_acc_vf_disable_fd(hisi_acc_vdev->resuming_migf);
 		fput(hisi_acc_vdev->resuming_migf->filp);
 		hisi_acc_vdev->resuming_migf = NULL;
 	}
 
Quoted 7 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static void hisi_acc_vf_disable_fd(struct hisi_acc_vf_migration_file *migf) · line 657

Quoted 2 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
 	if (hisi_acc_vdev->saving_migf) {
+		hisi_acc_vf_migf_save(hisi_acc_vdev, hisi_acc_vdev->saving_migf);

Why are these buffers copied to the debug_migf in this case? This can happen asynchronous to accessing the debugfs migration file and there's no serialization.

Quoted 2 lines

We can try to copy when accessing debugfs.

I tried migrating the file to save data when accessing debugfs.
This solution will not work. resuming_migf and saving_migf will be
kfree after completing the migration, and their data will no longer
be saved.
Therefore, the data needs to be saved to debug_migf when calling "hisi_acc_vf_disable_fd".

In addition, reading the migration status data of device can be read directly during debugfs access, without using debug_migf.

Thanks,
Longfang.

Quoted 391 lines
Quoted 25 lines
Quoted 12 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
 		hisi_acc_vf_disable_fd(hisi_acc_vdev->saving_migf);
 		fput(hisi_acc_vdev->saving_migf->filp);
 		hisi_acc_vdev->saving_migf = NULL;
@@ -1144,6 +1163,7 @@ static int hisi_acc_vf_qm_init(struct hisi_acc_vf_core_device *hisi_acc_vdev)
 	if (!vf_qm->io_base)
 		return -EIO;
 
+	mutex_init(&hisi_acc_vdev->enable_mutex);
 	vf_qm->fun_type = QM_HW_VF;
 	vf_qm->pdev = vf_dev;
 	mutex_init(&vf_qm->mailbox_lock);

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1323

Quoted 10 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +7 -0
@@ -1294,6 +1314,181 @@ static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int
 	return vfio_pci_core_ioctl(core_vdev, cmd, arg);
 }
 
+static int hisi_acc_vf_debug_check(struct seq_file *seq, struct vfio_device *vdev)
+{
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *migf = hisi_acc_vdev->debug_migf;
+	struct hisi_qm *vf_qm = &hisi_acc_vdev->vf_qm;
+	int ret;
+

lockdep_assert_held(...)

Quoted 6 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1326

Quoted 3 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +3 -0
+	if (!vdev->mig_ops || !migf) {
+		seq_printf(seq, "%s\n", "device does not support live migration!");
+		return -EINVAL;

Isn't the -EINVAL sufficient?

Which one do you think is better?

Quoted 10 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1333

Quoted 7 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +7 -0
+	}
+
+	/**
+	 * When the device is not opened, the io_base is not mapped.
+	 * The driver cannot perform device read and write operations.
+	 */
+	if (hisi_acc_vdev->dev_opened != DEV_OPEN) {

Why are we assigning and testing a bool against and enum?!

OK, change to true and false assignment.

Quoted 12 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1341

Quoted 8 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +8 -0
+		seq_printf(seq, "%s\n", "device not opened!");
+		return -EINVAL;
+	}
+
+	ret = qm_wait_dev_not_ready(vf_qm);
+	if (ret) {
+		seq_printf(seq, "%s\n", "VF device not ready!");
+		return -EINVAL;

-EBUSY? Again, not sure why we need the seq_printf() in addition to the error value.

OK, -EBUSY is better.
seq_printf() allows users to directly obtain the cause of the error without checking dmesg.

Quoted 14 lines
Quoted 14 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +14 -0
+	}
+
+	return 0;
+}
+
+static int hisi_acc_vf_debug_cmd(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_qm *vf_qm = &hisi_acc_vdev->vf_qm;
+	u64 value;
+	int ret;
Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1361

Quoted 6 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +6 -0
+
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
+	ret = hisi_acc_vf_debug_check(seq, vdev);
+	if (ret) {
+		mutex_unlock(&hisi_acc_vdev->enable_mutex);
+		return 0;

Why do we squash the error return here and throughout?

seq_printf() gives the user a failure message.
This can be changed to "return ret;"

Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1366

Quoted 5 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +5 -0
+	}
+
+	value = readl(vf_qm->io_base + QM_MB_CMD_SEND_BASE);
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
+	seq_printf(seq, "%s:0x%llx\n", "mailbox cmd channel state is OK", value);

We didn't test the value, what makes the state OK? Can this readl() or those in qm_wait_dev_not_ready() interfere with the main device flow?

If the cmd channel is normal, it will return a non-all-F value. Add exception checking in the next version.

cmd channel read operation will not affect the main migration process.

Quoted 46 lines
Quoted 46 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +46 -0
+
+	return 0;
+}
+
+static int hisi_acc_vf_debug_save(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *migf = hisi_acc_vdev->debug_migf;
+	struct acc_vf_data *vf_data = &migf->vf_data;
+	int ret;
+
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
+	ret = hisi_acc_vf_debug_check(seq, vdev);
+	if (ret) {
+		mutex_unlock(&hisi_acc_vdev->enable_mutex);
+		return 0;
+	}
+
+	vf_data->vf_qm_state = QM_READY;
+	ret = vf_qm_read_data(&hisi_acc_vdev->vf_qm, vf_data);
+	if (ret) {
+		mutex_unlock(&hisi_acc_vdev->enable_mutex);
+		seq_printf(seq, "%s\n", "failed to save device data!");
+		return 0;
+	}
+
+	migf->total_length = sizeof(struct acc_vf_data);
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
+	seq_printf(seq, "%s\n", "successful to save device data!");
+
+	return 0;
+}
+
+static int hisi_acc_vf_data_read(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *debug_migf = hisi_acc_vdev->debug_migf;
+	size_t vf_data_sz = offsetofend(struct acc_vf_data, padding);
+
+	if (debug_migf && debug_migf->total_length)
Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1417

Quoted 5 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +5 -0
+		seq_hex_dump(seq, "Mig Data:", DUMP_PREFIX_OFFSET, 16, 1,
+				(unsigned char *)&debug_migf->vf_data,
+				vf_data_sz, false);
+	else
+		seq_printf(seq, "%s\n", "device not migrated!");

"device state not saved"? Although I don't recall why this doesn't just return an errno.

OK, those who exit directly without migration will be processed according to the error mode and an error code will be returned.

Quoted 14 lines
Quoted 14 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +14 -0
+
+	return 0;
+}
+
+static int hisi_acc_vf_attr_read(struct seq_file *seq, void *data)
+{
+	struct device *vf_dev = seq->private;
+	struct vfio_pci_core_device *core_device = dev_get_drvdata(vf_dev);
+	struct vfio_device *vdev = &core_device->vdev;
+	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(vdev);
+	struct hisi_acc_vf_migration_file *debug_migf = hisi_acc_vdev->debug_migf;
+
+	if (debug_migf && debug_migf->total_length) {
+		seq_printf(seq,
Quoted 8 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1437

Quoted 6 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +6 -0
+			 "acc device:\n"
+			 "device  state: %d\n"
+			 "device  ready: %u\n"
+			 "data    valid: %d\n"
+			 "data     size: %lu\n",
+			 hisi_acc_vdev->mig_state,

This is redundant to migration/state, however note

OK!

Quoted 7 lines

hisi_acc_vfio_pci_get_device_state() protects the value with state mutex while reading it.

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1438

Quoted 1 line
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
+			 hisi_acc_vdev->vf_qm_state,

What's the purpose of this, it's ready or not, what does that mean?

If this status is not ready, the live migration process will exit directly. It indicates that there are two possible exceptions:
1. The acc device driver in the VM does not have insmod.
2. The acc device driver in the VM is insmoded, but the cmd channel is abnormal.

Quoted 3 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1439

Quoted 1 line
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
+			 debug_migf->disabled,

What's the purpose of this?

Get the enable status of migf file in the driver.

Quoted 6 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1440

Quoted 2 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
+			 debug_migf->total_length);

Why not just have this printed or inferred via the above data_read function, this all seems unnecessary.

This is used to obtain some key status of the live migration driver. It is more important than data in problem location.
So it is output in key-value pairs.

The above data is directly output in the form of hexadecimal data. It is used for more detailed analysis when there are no abnormalities in the key status.

Quoted 19 lines
Quoted 19 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +19 -0
+	} else {
+		seq_printf(seq, "%s\n", "device not migrated!");
+	}
+
+	return 0;
+}
+
+static int hisi_acc_vfio_debug_init(struct hisi_acc_vf_core_device *hisi_acc_vdev)
+{
+	struct vfio_device *vdev = &hisi_acc_vdev->core_device.vdev;
+	struct dentry *vfio_dev_migration = NULL;
+	struct dentry *vfio_hisi_acc = NULL;
+	struct device *dev = vdev->dev;
+	void *migf = NULL;
+
+	if (!debugfs_initialized())
+		return 0;
+
+	migf = kzalloc(sizeof(struct hisi_acc_vf_migration_file), GFP_KERNEL);
Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1464

Quoted 5 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +5 -0
+	if (!migf)
+		return -ENOMEM;
+	hisi_acc_vdev->debug_migf = migf;
+
+	vfio_dev_migration = debugfs_lookup("migration", vdev->debug_root);

Fails to build without CONFIG_DEBUG_FS=y There should be a dependency on CONFIG_VFIO_DEBUGFS here somewhere.

Yes, the driver needs to add "!IS_ENABLED(CONFIG_VFIO_DEBUGFS)" behind "debugfs_initialized()" above.

Quoted 7 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1466

Quoted 2 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +2 -0
+	if (!vfio_dev_migration) {
+		kfree(migf);

hisi_acc_vdev->debug_migf still points at this freed buffer, the return value of this function is not tested, allows a use-after-free in all of the below debugfs interfaces.

Yes, there needs to add "hisi_acc_vdev->debug_migf = NULL"

Quoted 17 lines
Quoted 17 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +17 -0
+		dev_err(dev, "failed to lookup migration debugfs file!\n");
+		return -ENODEV;
+	}
+
+	vfio_hisi_acc = debugfs_create_dir("hisi_acc", vfio_dev_migration);
+	debugfs_create_devm_seqfile(dev, "data", vfio_hisi_acc,
+				  hisi_acc_vf_data_read);
+	debugfs_create_devm_seqfile(dev, "attr", vfio_hisi_acc,
+				  hisi_acc_vf_attr_read);
+	debugfs_create_devm_seqfile(dev, "cmd_state", vfio_hisi_acc,
+				  hisi_acc_vf_debug_cmd);
+	debugfs_create_devm_seqfile(dev, "save", vfio_hisi_acc,
+				  hisi_acc_vf_debug_save);
+
+	return 0;
+}
+
Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static long hisi_acc_vfio_pci_ioctl(struct vfio_device *core_vdev, unsigned int · line 1489

Quoted 6 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +6 -0
+static void hisi_acc_vf_debugfs_exit(struct hisi_acc_vf_core_device *hisi_acc_vdev)
+{
+	if (!debugfs_initialized())
+		return;
+
+	kfree(hisi_acc_vdev->debug_migf);

Double free if the lookup in init fails.

After adding "hisi_acc_vdev->debug_migf = NULL" above. These processing codes need to be added here:

if (hisi_acc_vdev->debug_migf)

	kfree(hisi_acc_vdev->debug_migf);
Quoted 4 lines
Quoted 4 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +2 -0
+}
+
 static int hisi_acc_vfio_pci_open_device(struct vfio_device *core_vdev)
 {
Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c static int hisi_acc_vfio_pci_open_device(struct vfio_device *core_vdev) · line 1509

Quoted 6 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +1 -0
 	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_get_vf_dev(core_vdev);
@@ -1311,9 +1506,11 @@ static int hisi_acc_vfio_pci_open_device(struct vfio_device *core_vdev)
 			return ret;
 		}
 		hisi_acc_vdev->mig_state = VFIO_DEVICE_STATE_RUNNING;
+		hisi_acc_vdev->dev_opened = DEV_OPEN;

= true!

OK, the next version will not use enumeration values. will use true/false assignment.

Quoted 58 lines
Quoted 58 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +8 -0
 	}
 
 	vfio_pci_core_finish_enable(vdev);
+
 	return 0;
 }
 
@@ -1322,7 +1519,10 @@ 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_vdev->dev_opened = DEV_CLOSE;
+	mutex_lock(&hisi_acc_vdev->enable_mutex);
 	iounmap(vf_qm->io_base);
+	mutex_unlock(&hisi_acc_vdev->enable_mutex);
 	vfio_pci_core_close_device(core_vdev);
 }
 
@@ -1413,6 +1613,9 @@ static int hisi_acc_vfio_pci_probe(struct pci_dev *pdev, const struct pci_device
 	ret = vfio_pci_core_register_device(&hisi_acc_vdev->core_device);
 	if (ret)
 		goto out_put_vdev;
+
+	if (ops == &hisi_acc_vfio_pci_migrn_ops)
+		hisi_acc_vfio_debug_init(hisi_acc_vdev);
 	return 0;
 
 out_put_vdev:
@@ -1425,6 +1628,7 @@ static void hisi_acc_vfio_pci_remove(struct pci_dev *pdev)
 	struct hisi_acc_vf_core_device *hisi_acc_vdev = hisi_acc_drvdata(pdev);
 
 	vfio_pci_core_unregister_device(&hisi_acc_vdev->core_device);
+	hisi_acc_vf_debugfs_exit(hisi_acc_vdev);
 	vfio_put_device(&hisi_acc_vdev->core_device.vdev);
 }
 
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h +7 -0
index 7a9dc87627cd..3a20d81d105c 100644
@@ -52,6 +52,11 @@
 #define QM_EQC_DW0		0X8000
 #define QM_AEQC_DW0		0X8020
 
+enum acc_dev_state {
+	DEV_CLOSE = 0x0,
+	DEV_OPEN,
+};
+
 struct acc_vf_data {
 #define QM_MATCH_SIZE offsetofend(struct acc_vf_data, qm_rsv_state)
 	/* QM match information */
@@ -114,5 +119,11 @@ struct hisi_acc_vf_core_device {
 	int vf_id;
 	struct hisi_acc_vf_migration_file *resuming_migf;
 	struct hisi_acc_vf_migration_file *saving_migf;
+
+	/* To make sure the device is enabled */
Quoted 9 lines

drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h struct hisi_acc_vf_core_device { · line 129

Quoted 6 lines
drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.h +4 -0
+	struct mutex enable_mutex;
+	bool dev_opened;
+	/* For debugfs */
+	struct hisi_acc_vf_migration_file *debug_migf;
 };
 #endif /* HISI_ACC_VFIO_PCI_H */

.

Thank you very much for your careful inspection. I will revise the inspection issues you mentioned above one by one and publish them in the next version.

Thanks again!
Longfang.

.

author maintainer
#9 lore raw

PATCH v4 4/4 Documentation: add debugfs description for hisi migration

In reply to Longfang Liu

Add a debugfs document description file to help users understand how to use the hisilicon accelerator live migration driver's debugfs.

Update the file paths that need to be maintained in MAINTAINERS

---
 .../ABI/testing/debugfs-hisi-migration        | 34 +++++++++++++++++++
 MAINTAINERS                                   |  1 +
 2 files changed, 35 insertions(+)
 create mode 100644 Documentation/ABI/testing/debugfs-hisi-migration
Documentation/ABI/testing/debugfs-hisi-migration new file +34 -0
new file mode 100644
index 000000000000..3d7339276e6f
@@ -0,0 +1,34 @@
+What:		/sys/kernel/debug/vfio/<device>/migration/hisi_acc/data
+Date:		Apr 2024
+KernelVersion:  6.9
+Contact:	Longfang Liu <[email protected]>
+Description:	Read the live migration data of the vfio device.
+		These data include device status data, queue configuration
+		data and some task configuration data.
+		The output format of the data is defined by the live
+		migration driver.
+
+What:		/sys/kernel/debug/vfio/<device>/migration/hisi_acc/attr
+Date:		Apr 2024
+KernelVersion:  6.9
+Contact:	Longfang Liu <[email protected]>
+Description:	Read the live migration attributes of the vfio device.
+		it include device status attributes and data length attributes
+		The output format of the attributes is defined by the live
+		migration driver.
+
+What:		/sys/kernel/debug/vfio/<device>/migration/hisi_acc/cmd_state
+Date:		Apr 2024
+KernelVersion:  6.9
+Contact:	Longfang Liu <[email protected]>
+Description:	Used to obtain the device command sending and receiving
+		channel status. If successful, returns the command value.
+		If failed, return error log.
+
+What:		/sys/kernel/debug/vfio/<device>/migration/hisi_acc/save
+Date:		Apr 2024
+KernelVersion:  6.9
+Contact:	Longfang Liu <[email protected]>
+Description:	Trigger the Hisilicon accelerator device to perform
+		the state saving operation of live migration through the read
+		operation, and output the operation log results.
MAINTAINERS +1 -0
index 7625911ec2f1..8c2d13b13273 100644
@@ -23072,6 +23072,7 @@ M:	Longfang Liu <[email protected]>
 M:	Shameer Kolothum <[email protected]>
 L:	[email protected]
 S:	Maintained
+F:	Documentation/ABI/testing/debugfs-hisi-migration
 F:	drivers/vfio/pci/hisilicon/
 
 VFIO MEDIATED DEVICE DRIVERS
Signature · 1 line
2.24.0
#10 lore raw

Re: [PATCH v4 4/4] Documentation: add debugfs description for hisi migration

Quoted 19 lines · On Tue, 2 Apr 2024 11:24:32 +0800 Longfang Liu <[email protected]> wrote:

Add a debugfs document description file to help users understand how to use the hisilicon accelerator live migration driver's debugfs.

Update the file paths that need to be maintained in MAINTAINERS

---
 .../ABI/testing/debugfs-hisi-migration        | 34 +++++++++++++++++++
 MAINTAINERS                                   |  1 +
 2 files changed, 35 insertions(+)
 create mode 100644 Documentation/ABI/testing/debugfs-hisi-migration
Documentation/ABI/testing/debugfs-hisi-migration new file +0 -0
new file mode 100644
index 000000000000..3d7339276e6f

Documentation/ABI/testing/debugfs-hisi-migration line 3

Quoted 3 lines
Documentation/ABI/testing/debugfs-hisi-migration new file +3 -0
@@ -0,0 +1,34 @@
+What:		/sys/kernel/debug/vfio/<device>/migration/hisi_acc/data
+Date:		Apr 2024
+KernelVersion:  6.9

At best 6.10 with a merge window in May.

Documentation/ABI/testing/debugfs-hisi-migration line 9

Quoted 6 lines
Documentation/ABI/testing/debugfs-hisi-migration new file +6 -0
+Contact:	Longfang Liu <[email protected]>
+Description:	Read the live migration data of the vfio device.
+		These data include device status data, queue configuration
+		data and some task configuration data.
+		The output format of the data is defined by the live
+		migration driver.

"Dumps the device debug migration buffer, state must first be saved using the 'save' attribute."

Documentation/ABI/testing/debugfs-hisi-migration line 18

Quoted 9 lines
Documentation/ABI/testing/debugfs-hisi-migration new file +9 -0
+
+What:		/sys/kernel/debug/vfio/<device>/migration/hisi_acc/attr
+Date:		Apr 2024
+KernelVersion:  6.9
+Contact:	Longfang Liu <[email protected]>
+Description:	Read the live migration attributes of the vfio device.
+		it include device status attributes and data length attributes
+		The output format of the attributes is defined by the live
+		migration driver.

AFAICT from the previous patch, this attribute is useless.

Documentation/ABI/testing/debugfs-hisi-migration line 27

Quoted 9 lines
Documentation/ABI/testing/debugfs-hisi-migration new file +9 -0
+
+What:		/sys/kernel/debug/vfio/<device>/migration/hisi_acc/cmd_state
+Date:		Apr 2024
+KernelVersion:  6.9
+Contact:	Longfang Liu <[email protected]>
+Description:	Used to obtain the device command sending and receiving
+		channel status. If successful, returns the command value.
+		If failed, return error log.
+

Seems like it statically returns "OK" plus the actual value.

Documentation/ABI/testing/debugfs-hisi-migration line 34

Quoted 7 lines
Documentation/ABI/testing/debugfs-hisi-migration new file +7 -0
+What:		/sys/kernel/debug/vfio/<device>/migration/hisi_acc/save
+Date:		Apr 2024
+KernelVersion:  6.9
+Contact:	Longfang Liu <[email protected]>
+Description:	Trigger the Hisilicon accelerator device to perform
+		the state saving operation of live migration through the read
+		operation, and output the operation log results.

These interfaces are confusing, attr and data only work if there has either been a previous save OR the user migration process closed saving or resuming fds in the interim, and the user doesn't know which one they get. Note that debug_migf isn't even discarded between open/close, only cmd and save require the device to be opened by a user, data and attr might continue to return data from some previous user save, resume, or debugfs save.

Quoted 12 lines
MAINTAINERS +1 -0
index 7625911ec2f1..8c2d13b13273 100644
@@ -23072,6 +23072,7 @@ M:	Longfang Liu <[email protected]>
 M:	Shameer Kolothum <[email protected]>
 L:	[email protected]
 S:	Maintained
+F:	Documentation/ABI/testing/debugfs-hisi-migration
 F:	drivers/vfio/pci/hisilicon/
 
 VFIO MEDIATED DEVICE DRIVERS
author maintainer
#11 lore raw

Re: [PATCH v4 4/4] Documentation: add debugfs description for hisi migration

Quoted 63 lines · Alex Williamson
Quoted 19 lines · On Tue, 2 Apr 2024 11:24:32 +0800 Longfang Liu <[email protected]> wrote:

Add a debugfs document description file to help users understand how to use the hisilicon accelerator live migration driver's debugfs.

Update the file paths that need to be maintained in MAINTAINERS

---
 .../ABI/testing/debugfs-hisi-migration        | 34 +++++++++++++++++++
 MAINTAINERS                                   |  1 +
 2 files changed, 35 insertions(+)
 create mode 100644 Documentation/ABI/testing/debugfs-hisi-migration
Documentation/ABI/testing/debugfs-hisi-migration new file +0 -0
new file mode 100644
index 000000000000..3d7339276e6f

Documentation/ABI/testing/debugfs-hisi-migration line 3

Quoted 3 lines
Documentation/ABI/testing/debugfs-hisi-migration new file +3 -0
@@ -0,0 +1,34 @@
+What:		/sys/kernel/debug/vfio/<device>/migration/hisi_acc/data
+Date:		Apr 2024
+KernelVersion:  6.9

At best 6.10 with a merge window in May.

Documentation/ABI/testing/debugfs-hisi-migration line 9

Quoted 6 lines
Documentation/ABI/testing/debugfs-hisi-migration new file +6 -0
+Contact:	Longfang Liu <[email protected]>
+Description:	Read the live migration data of the vfio device.
+		These data include device status data, queue configuration
+		data and some task configuration data.
+		The output format of the data is defined by the live
+		migration driver.

"Dumps the device debug migration buffer, state must first be saved using the 'save' attribute."

Documentation/ABI/testing/debugfs-hisi-migration line 18

Quoted 9 lines
Documentation/ABI/testing/debugfs-hisi-migration new file +9 -0
+
+What:		/sys/kernel/debug/vfio/<device>/migration/hisi_acc/attr
+Date:		Apr 2024
+KernelVersion:  6.9
+Contact:	Longfang Liu <[email protected]>
+Description:	Read the live migration attributes of the vfio device.
+		it include device status attributes and data length attributes
+		The output format of the attributes is defined by the live
+		migration driver.

AFAICT from the previous patch, this attribute is useless.

Documentation/ABI/testing/debugfs-hisi-migration line 27

Quoted 9 lines
Documentation/ABI/testing/debugfs-hisi-migration new file +9 -0
+
+What:		/sys/kernel/debug/vfio/<device>/migration/hisi_acc/cmd_state
+Date:		Apr 2024
+KernelVersion:  6.9
+Contact:	Longfang Liu <[email protected]>
+Description:	Used to obtain the device command sending and receiving
+		channel status. If successful, returns the command value.
+		If failed, return error log.
+

Seems like it statically returns "OK" plus the actual value.

Quoted 16 lines

Documentation/ABI/testing/debugfs-hisi-migration line 34

Quoted 7 lines
Documentation/ABI/testing/debugfs-hisi-migration new file +7 -0
+What:		/sys/kernel/debug/vfio/<device>/migration/hisi_acc/save
+Date:		Apr 2024
+KernelVersion:  6.9
+Contact:	Longfang Liu <[email protected]>
+Description:	Trigger the Hisilicon accelerator device to perform
+		the state saving operation of live migration through the read
+		operation, and output the operation log results.

These interfaces are confusing, attr and data only work if there has either been a previous save OR the user migration process closed saving or resuming fds in the interim, and the user doesn't know which one they get. Note that debug_migf isn't even discarded between open/close, only cmd and save require the device to be opened by a user, data and attr might continue to return data from some previous user save, resume, or debugfs save.

data: Indicates the device migration data obtained after the migration is completed. This data is saved in debug_migf. The user reads it through "cat" and presents it to the user in the form of hexadecimal pure data.

attr: Indicates the configuration parameters of the migration process after the migration is completed. These parameters are saved in vfio device and debug_migf. The user reads it through "cat" and presents it to the user in the form of key-value pairs such as <attribute name, attribute value>.

Save is an action process. After "cat" it, a migration save operation will be performed and the result data will be updated to debug_migf.

There is still a big difference between data and attr, and the data formats are also different. Not merging makes it easier for users to obtain information. If you feel confused about save, it is recommended to use migrate_save.

Thanks,
Longfang.

Quoted 17 lines
Quoted 7 lines
MAINTAINERS +0 -0
index 7625911ec2f1..8c2d13b13273 100644

MAINTAINERS M: Longfang Liu <[email protected]> · line 23078

Quoted 7 lines
MAINTAINERS +1 -0
@@ -23072,6 +23072,7 @@ M:	Longfang Liu <[email protected]>
 M:	Shameer Kolothum <[email protected]>
 L:	[email protected]
 S:	Maintained
+F:	Documentation/ABI/testing/debugfs-hisi-migration
 F:	drivers/vfio/pci/hisilicon/
 
 VFIO MEDIATED DEVICE DRIVERS

.