[RFC PATCH v1 1/2] block: bsg: resume scsi device before accessing

[RFC PATCH v1 1/2] block: bsg: resume scsi device before accessing

[Asutosh Das]

Resumes the scsi device before accessing it.

Change-Id: I2929af60f2a92c89704a582fcdb285d35b429fde
Signed-off-by: Asutosh Das <asutoshd@codeaurora.org>
Signed-off-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Bao D. Nguyen <nguyenb@codeaurora.org>
---
 block/bsg.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/block/bsg.c b/block/bsg.c
index d7bae94..f4c197f 100644
--- a/block/bsg.c
+++ b/block/bsg.c
@@ -306,12 +306,16 @@ static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
 static int bsg_open(struct inode *inode, struct file *file)
 {
 	struct bsg_device *bd;
+	struct scsi_device *sd;
 
 	bd = bsg_get_device(inode, file);
 
 	if (IS_ERR(bd))
 		return PTR_ERR(bd);
 
+	sd = (struct scsi_device *) bd->queue->queuedata;
+	if (scsi_autopm_get_device(sd))
+		return -EIO;
 	file->private_data = bd;
 	return 0;
 }
@@ -319,8 +323,12 @@ static int bsg_open(struct inode *inode, struct file *file)
 static int bsg_release(struct inode *inode, struct file *file)
 {
 	struct bsg_device *bd = file->private_data;
+	struct scsi_device *sd;
 
 	file->private_data = NULL;
+	sd = (struct scsi_device *) bd->queue->queuedata;
+	scsi_autopm_put_device(sd);
+
 	return bsg_put_device(bd);
 }
 
-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project.

[RFC PATCH v1 2/2] scsi: ufs: Fix deadlock while suspending ufs host

[Asutosh Das]

During runtime-suspend of ufs host, the scsi devices are
already suspended and so are the queues associated with them.
But the ufs host sends SSU to wlun during its runtime-suspend.
During the process blk_queue_enter checks if the queue is not in
suspended state. If so, it waits for the queue to resume, and never
comes out of it.
The commit
(d55d15a33: scsi: block: Do not accept any requests while suspended)
adds the check if the queue is in suspended state in blk_queue_enter().

Fix this, by decoupling wlun scsi devices from block layer pm.
The runtime-pm for these devices would be managed by bsg and sg drivers.

Call trace:
 __switch_to+0x174/0x2c4
 __schedule+0x478/0x764
 schedule+0x9c/0xe0
 blk_queue_enter+0x158/0x228
 blk_mq_alloc_request+0x40/0xa4
 blk_get_request+0x2c/0x70
 __scsi_execute+0x60/0x1c4
 ufshcd_set_dev_pwr_mode+0x124/0x1e4
 ufshcd_suspend+0x208/0x83c
 ufshcd_runtime_suspend+0x40/0x154
 ufshcd_pltfrm_runtime_suspend+0x14/0x20
 pm_generic_runtime_suspend+0x28/0x3c
 __rpm_callback+0x80/0x2a4
 rpm_suspend+0x308/0x614
 rpm_idle+0x158/0x228
 pm_runtime_work+0x84/0xac
 process_one_work+0x1f0/0x470
 worker_thread+0x26c/0x4c8
 kthread+0x13c/0x320
 ret_from_fork+0x10/0x18

Change-Id: Id777fd52493c8b5522d1ebcad73cd30dac33e8a4
Signed-off-by: Asutosh Das <asutoshd@codeaurora.org>
Signed-off-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Bao D. Nguyen <nguyenb@codeaurora.org>
---
 drivers/scsi/ufs/ufshcd.c | 18 ++----------------
 1 file changed, 2 insertions(+), 16 deletions(-)

diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index 9c691e4..b7e7f81 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -7217,16 +7217,6 @@ static void ufshcd_set_active_icc_lvl(struct ufs_hba *hba)
 	kfree(desc_buf);
 }
 
-static inline void ufshcd_blk_pm_runtime_init(struct scsi_device *sdev)
-{
-	scsi_autopm_get_device(sdev);
-	blk_pm_runtime_init(sdev->request_queue, &sdev->sdev_gendev);
-	if (sdev->rpm_autosuspend)
-		pm_runtime_set_autosuspend_delay(&sdev->sdev_gendev,
-						 RPM_AUTOSUSPEND_DELAY_MS);
-	scsi_autopm_put_device(sdev);
-}
-
 /**
  * ufshcd_scsi_add_wlus - Adds required W-LUs
  * @hba: per-adapter instance
@@ -7265,7 +7255,6 @@ static int ufshcd_scsi_add_wlus(struct ufs_hba *hba)
 		hba->sdev_ufs_device = NULL;
 		goto out;
 	}
-	ufshcd_blk_pm_runtime_init(hba->sdev_ufs_device);
 	scsi_device_put(hba->sdev_ufs_device);
 
 	hba->sdev_rpmb = __scsi_add_device(hba->host, 0, 0,
@@ -7274,17 +7263,14 @@ static int ufshcd_scsi_add_wlus(struct ufs_hba *hba)
 		ret = PTR_ERR(hba->sdev_rpmb);
 		goto remove_sdev_ufs_device;
 	}
-	ufshcd_blk_pm_runtime_init(hba->sdev_rpmb);
 	scsi_device_put(hba->sdev_rpmb);
 
 	sdev_boot = __scsi_add_device(hba->host, 0, 0,
 		ufshcd_upiu_wlun_to_scsi_wlun(UFS_UPIU_BOOT_WLUN), NULL);
-	if (IS_ERR(sdev_boot)) {
+	if (IS_ERR(sdev_boot))
 		dev_err(hba->dev, "%s: BOOT WLUN not found\n", __func__);
-	} else {
-		ufshcd_blk_pm_runtime_init(sdev_boot);
+	else
 		scsi_device_put(sdev_boot);
-	}
 	goto out;
 
 remove_sdev_ufs_device:
-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project.

RE: [RFC PATCH v1 1/2] block: bsg: resume scsi device before accessing

[Avri Altman]

> 
> Resumes the scsi device before accessing it.
> 
> Change-Id: I2929af60f2a92c89704a582fcdb285d35b429fde
> Signed-off-by: Asutosh Das <asutoshd@codeaurora.org>
> Signed-off-by: Can Guo <cang@codeaurora.org>
> Signed-off-by: Bao D. Nguyen <nguyenb@codeaurora.org>
Following this patch, is it possible to revert commit 74e5e468b664d?

Thanks,
Avri

Re: [RFC PATCH v1 1/2] block: bsg: resume scsi device before accessing

[Can Guo]

On 2021-01-27 15:09, Avri Altman wrote:
>> 
>> Resumes the scsi device before accessing it.
>> 
>> Change-Id: I2929af60f2a92c89704a582fcdb285d35b429fde
>> Signed-off-by: Asutosh Das <asutoshd@codeaurora.org>
>> Signed-off-by: Can Guo <cang@codeaurora.org>
>> Signed-off-by: Bao D. Nguyen <nguyenb@codeaurora.org>
> Following this patch, is it possible to revert commit 74e5e468b664d?
> 

No, but this is a good finding... This change assumes
that the queue->queue_data is a scsi_device, which is
why we call scsi_auto_pm_get(). But for ufs_bsg's case,
queue->queue_data is a device...

Thanks,
Can Guo.

> Thanks,
> Avri

Re: [RFC PATCH v1 1/2] block: bsg: resume scsi device before accessing

[Can Guo]

On 2021-01-27 15:59, Can Guo wrote:
> On 2021-01-27 15:09, Avri Altman wrote:
>>> 
>>> Resumes the scsi device before accessing it.
>>> 
>>> Change-Id: I2929af60f2a92c89704a582fcdb285d35b429fde
>>> Signed-off-by: Asutosh Das <asutoshd@codeaurora.org>
>>> Signed-off-by: Can Guo <cang@codeaurora.org>
>>> Signed-off-by: Bao D. Nguyen <nguyenb@codeaurora.org>
>> Following this patch, is it possible to revert commit 74e5e468b664d?
>> 
> 
> No, but this is a good finding... This change assumes
> that the queue->queue_data is a scsi_device, which is
> why we call scsi_auto_pm_get(). But for ufs_bsg's case,
> queue->queue_data is a device...
> 

If we call pm_runtime_get/put_sync(bcd->class_dev->parent) in
bsg_get/put_device(), commit 74e5e468b664d can be reverted.
This is just a rough idea.

> Thanks,
> Can Guo.
> 
>> Thanks,
>> Avri

[RFC PATCH v2 1/2] block: bsg: resume platform device before accessing

[Asutosh Das]

It may happen that the underlying device's runtime-pm is
not controlled by block-pm. So it's possible that when
commands are sent to the device, it's suspended and may not
be resumed by blk-pm. Hence explicitly resume the parent
which is the platform device.

Signed-off-by: Asutosh Das <asutoshd@codeaurora.org>
Signed-off-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Bao D. Nguyen <nguyenb@codeaurora.org>
---
 block/bsg.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/block/bsg.c b/block/bsg.c
index d7bae94..e9fc896 100644
--- a/block/bsg.c
+++ b/block/bsg.c
@@ -12,6 +12,7 @@
 #include <linux/idr.h>
 #include <linux/bsg.h>
 #include <linux/slab.h>
+#include <linux/pm_runtime.h>
 
 #include <scsi/scsi.h>
 #include <scsi/scsi_ioctl.h>
@@ -306,12 +307,15 @@ static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
 static int bsg_open(struct inode *inode, struct file *file)
 {
 	struct bsg_device *bd;
+	struct bsg_class_device *bcd;
 
 	bd = bsg_get_device(inode, file);
 
 	if (IS_ERR(bd))
 		return PTR_ERR(bd);
 
+	bcd = &bd->queue->bsg_dev;
+	pm_runtime_get_sync(bcd->class_dev->parent);
 	file->private_data = bd;
 	return 0;
 }
@@ -319,8 +323,12 @@ static int bsg_open(struct inode *inode, struct file *file)
 static int bsg_release(struct inode *inode, struct file *file)
 {
 	struct bsg_device *bd = file->private_data;
+	struct bsg_class_device *bcd;
 
 	file->private_data = NULL;
+
+	bcd = &bd->queue->bsg_dev;
+	pm_runtime_put_sync(bcd->class_dev->parent);
 	return bsg_put_device(bd);
 }
 
-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project.

[RFC PATCH v2 2/2] scsi: ufs: Fix deadlock while suspending ufs host

[Asutosh Das]

During runtime-suspend of ufs host, the scsi devices are
already suspended and so are the queues associated with them.
But the ufs host sends SSU to wlun during its runtime-suspend.
During the process blk_queue_enter checks if the queue is not in
suspended state. If so, it waits for the queue to resume, and never
comes out of it.
The commit
(d55d15a33: scsi: block: Do not accept any requests while suspended)
adds the check if the queue is in suspended state in blk_queue_enter().

Fix this, by decoupling wlun scsi devices from block layer pm.
The runtime-pm for these devices would be managed by bsg and sg drivers.

Call trace:
 __switch_to+0x174/0x2c4
 __schedule+0x478/0x764
 schedule+0x9c/0xe0
 blk_queue_enter+0x158/0x228
 blk_mq_alloc_request+0x40/0xa4
 blk_get_request+0x2c/0x70
 __scsi_execute+0x60/0x1c4
 ufshcd_set_dev_pwr_mode+0x124/0x1e4
 ufshcd_suspend+0x208/0x83c
 ufshcd_runtime_suspend+0x40/0x154
 ufshcd_pltfrm_runtime_suspend+0x14/0x20
 pm_generic_runtime_suspend+0x28/0x3c
 __rpm_callback+0x80/0x2a4
 rpm_suspend+0x308/0x614
 rpm_idle+0x158/0x228
 pm_runtime_work+0x84/0xac
 process_one_work+0x1f0/0x470
 worker_thread+0x26c/0x4c8
 kthread+0x13c/0x320
 ret_from_fork+0x10/0x18

Signed-off-by: Asutosh Das <asutoshd@codeaurora.org>
Signed-off-by: Can Guo <cang@codeaurora.org>
Signed-off-by: Bao D. Nguyen <nguyenb@codeaurora.org>
---
 drivers/scsi/ufs/ufshcd.c | 18 ++----------------
 1 file changed, 2 insertions(+), 16 deletions(-)

diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c
index 9c691e4..b7e7f81 100644
--- a/drivers/scsi/ufs/ufshcd.c
+++ b/drivers/scsi/ufs/ufshcd.c
@@ -7217,16 +7217,6 @@ static void ufshcd_set_active_icc_lvl(struct ufs_hba *hba)
 	kfree(desc_buf);
 }
 
-static inline void ufshcd_blk_pm_runtime_init(struct scsi_device *sdev)
-{
-	scsi_autopm_get_device(sdev);
-	blk_pm_runtime_init(sdev->request_queue, &sdev->sdev_gendev);
-	if (sdev->rpm_autosuspend)
-		pm_runtime_set_autosuspend_delay(&sdev->sdev_gendev,
-						 RPM_AUTOSUSPEND_DELAY_MS);
-	scsi_autopm_put_device(sdev);
-}
-
 /**
  * ufshcd_scsi_add_wlus - Adds required W-LUs
  * @hba: per-adapter instance
@@ -7265,7 +7255,6 @@ static int ufshcd_scsi_add_wlus(struct ufs_hba *hba)
 		hba->sdev_ufs_device = NULL;
 		goto out;
 	}
-	ufshcd_blk_pm_runtime_init(hba->sdev_ufs_device);
 	scsi_device_put(hba->sdev_ufs_device);
 
 	hba->sdev_rpmb = __scsi_add_device(hba->host, 0, 0,
@@ -7274,17 +7263,14 @@ static int ufshcd_scsi_add_wlus(struct ufs_hba *hba)
 		ret = PTR_ERR(hba->sdev_rpmb);
 		goto remove_sdev_ufs_device;
 	}
-	ufshcd_blk_pm_runtime_init(hba->sdev_rpmb);
 	scsi_device_put(hba->sdev_rpmb);
 
 	sdev_boot = __scsi_add_device(hba->host, 0, 0,
 		ufshcd_upiu_wlun_to_scsi_wlun(UFS_UPIU_BOOT_WLUN), NULL);
-	if (IS_ERR(sdev_boot)) {
+	if (IS_ERR(sdev_boot))
 		dev_err(hba->dev, "%s: BOOT WLUN not found\n", __func__);
-	} else {
-		ufshcd_blk_pm_runtime_init(sdev_boot);
+	else
 		scsi_device_put(sdev_boot);
-	}
 	goto out;
 
 remove_sdev_ufs_device:
-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project.

RE: [RFC PATCH v2 2/2] scsi: ufs: Fix deadlock while suspending ufs host

[Avri Altman]

> 
> During runtime-suspend of ufs host, the scsi devices are
> already suspended and so are the queues associated with them.
> But the ufs host sends SSU to wlun during its runtime-suspend.
Do you possible meant: "sends request-sense while clearing UAC to...."

Thanks,
Avri

Re: [RFC PATCH v2 2/2] scsi: ufs: Fix deadlock while suspending ufs host

[Asutosh Das]

On Thu, Jan 28 2021 at 04:21 -0800, Avri Altman wrote:
>>
>> During runtime-suspend of ufs host, the scsi devices are
>> already suspended and so are the queues associated with them.
>> But the ufs host sends SSU to wlun during its runtime-suspend.
>Do you possible meant: "sends request-sense while clearing UAC to...."
>

The idea was to show that there's a scsi command that's sent during
suspend which may deadlock.

Yes, I agree to your comment and would change the message to reflect it in
the next version.

>
>Thanks,
>Avri

Re: [RFC PATCH v1 1/2] block: bsg: resume scsi device before accessing

[Bart Van Assche]

On 1/26/21 8:00 PM, Asutosh Das wrote:
> Resumes the scsi device before accessing it.
> 
> Change-Id: I2929af60f2a92c89704a582fcdb285d35b429fde
> Signed-off-by: Asutosh Das <asutoshd@codeaurora.org>
> Signed-off-by: Can Guo <cang@codeaurora.org>
> Signed-off-by: Bao D. Nguyen <nguyenb@codeaurora.org>

No Change-Id tags in upstream patches please.

Thanks,

Bart.