scsi: pm: Leave runtime resume along if block layer PM is enabled
Subsystems (2)
cover scsi: pm: Leave runtime resume along if block layer PM is enabled
In current UFS driver, if the hba device is in RPM_SUSPENDED status when system suspend happens, UFS driver chooses not to resume hba device during system resume [1], which is to achieve power efficiency by leaving the runtime PM to block layer PM.
Similarly, if block layer runtime PM is enabled for one SCSI device, then there is no need to forcibly change the SCSI device and its request queue's runtime PM status to RPM_ACTIVE in scsi_dev_type_resume(), since block layer PM shall resume the SCSI device on the demand of bios.
Without this change, if situation [1] happens, when scsi_dev_type_resume() invokes blk_pm_set_active(), we will get an error since it is trying to set the SCSI device's runtime PM status to RPM_ACTIVE even when its parent is not RPM_ACTIVE, which actually is not fatal. This change can avoid the annoying error logs when situation [1] happens.
However, there are some rare cases where upper layers start to submit bios into the request queue of one SCSI device even when scsi_dev_type_resume() is still running on that SCSI device. In these rare cases, if situation [1] happens, when a bio is sent to block layer, block layer runtime PM shall try to runtime resume the SCSI device (invoked from blk_queue_enter()) and if meanwhile scsi_dev_type_resume() has ran over line #83 but before #85 (see below code snippet), block layer runtime PM won't be able to resume it due to runtime PM is disabled on this device at this moment. And later if no more bios kick blk_queue_enter(), the bio will be stuck at blk_queue_enter() forever. This change can fix the problem too.
<--code snippet-->
72 static int scsi_dev_type_resume(struct device *dev,
73 int (*cb)(struct device *, const struct dev_pm_ops *))
74 {
...
83 pm_runtime_disable(dev);
84 err = pm_runtime_set_active(dev);
85 pm_runtime_enable(dev);
...
<--code snippet-->
Can Guo (1):
scsi: pm: Leave runtime resume along if block layer PM is enabled
drivers/scsi/scsi_pm.c | 25 ++++++++++++------------- 1 file changed, 12 insertions(+), 13 deletions(-)
Signature · 1 line
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project.
RFC PATCH 1/1 scsi: pm: Leave runtime resume along if block layer PM is enabled
If block layer runtime PM is enabled for one SCSI device, then there is no need to forcibly change the SCSI device and its request queue's runtime PM status to active in scsi_dev_type_resume(), since block layer PM shall resume the SCSI device on the demand of bios.
--- drivers/scsi/scsi_pm.c | 25 ++++++++++++------------- 1 file changed, 12 insertions(+), 13 deletions(-)
drivers/scsi/scsi_pm.c +12 -13
@@ -79,23 +79,22 @@ static int scsi_dev_type_resume(struct device *dev,scsi_device_resume (to_scsi_device (dev));dev_dbg (dev,"scsi resume: %d\n" , err); -if (err ==0 ) { -pm_runtime_disable (dev); - err =pm_runtime_set_active (dev); -pm_runtime_enable (dev); +if (scsi_is_sdev_device (dev)) { +struct scsi_device *sdev; + sdev =to_scsi_device (dev);/* -* Forcibly set runtime PM status of request queue to "active" -* to make sure we can again get requests from the queue -* (see also blk_pm_peek_request()). -* -* The resume hook will correct runtime PM status of the disk. +* If block layer runtime PM is enabled for the SCSI device, +* let block layer PM handle its runtime PM routines. */ -if (!err &&scsi_is_sdev_device (dev)) { -struct scsi_device *sdev =to_scsi_device (dev); +if (sdev->request_queue->dev) +return err; + } -blk_set_runtime_active (sdev->request_queue); - } +if (err ==0 ) { +pm_runtime_disable (dev); + err =pm_runtime_set_active (dev); +pm_runtime_enable (dev); }return err;
Signature · 1 line
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project.
Quoted 4 lines · Can Guo
If block layer runtime PM is enabled for one SCSI device, then there is no need to forcibly change the SCSI device and its request queue's runtime PM status to active in
scsi_dev_type_resume(), since block layer PM shall resume the SCSI device on the demand of bios.
Please change "along" into "alone" in the subject of this patch (if that is what you meant).
drivers/scsi/scsi_pm.c static int scsi_dev_type_resume(struct device *dev, · line 85
Quoted 4 lines
drivers/scsi/scsi_pm.c +3
-0+if (scsi_is_sdev_device (dev)) { +struct scsi_device *sdev; + sdev =to_scsi_device (dev);
A minor comment: I think that "struct scsi_device *sdev = to_scsi_device(dev);" fits on a single line.
drivers/scsi/scsi_pm.c static int scsi_dev_type_resume(struct device *dev, · line 88
Quoted 2 lines
drivers/scsi/scsi_pm.c +2
-0+* If block layer runtime PM is enabled for the SCSI device, +* let block layer PM handle its runtime PM routines.
Please change "its runtime PM routines" into "runtime resume" or similar. I think that will make the comment more clear.
drivers/scsi/scsi_pm.c static int scsi_dev_type_resume(struct device *dev, · line 92
Quoted 3 lines
drivers/scsi/scsi_pm.c +3
-0+if (sdev->request_queue->dev) +return err; + }
The 'dev' member only exists in struct request_queue if CONFIG_PM=y so the above won't compile if CONFIG_PM=n. How about adding a function in include/linux/blk-pm.h to check whether or not runtime PM has been enabled?
Otherwise this patch looks good to me.
Thanks,
Bart.
Hi Bart,
Quoted 9 lines · Bart Van Assche
Quoted 4 lines · Can Guo
If block layer runtime PM is enabled for one SCSI device, then there is no need to forcibly change the SCSI device and its request queue's runtime PM status to active in
scsi_dev_type_resume(), since block layer PM shall resume the SCSI device on the demand of bios.Please change "along" into "alone" in the subject of this patch (if that is what you meant).
Aha, sorry, a typo here.
Quoted 8 lines
drivers/scsi/scsi_pm.c static int scsi_dev_type_resume(struct device *dev, · line 85
Quoted 4 lines
drivers/scsi/scsi_pm.c +3
-0+if (scsi_is_sdev_device (dev)) { +struct scsi_device *sdev; + sdev =to_scsi_device (dev);A minor comment: I think that "struct
scsi_device*sdev =to_scsi_device(dev);" fits on a single line.
Sure.
Quoted 6 lines
drivers/scsi/scsi_pm.c static int scsi_dev_type_resume(struct device *dev, · line 88
Quoted 2 lines
drivers/scsi/scsi_pm.c +2
-0+* If block layer runtime PM is enabled for the SCSI device, +* let block layer PM handle its runtime PM routines. Please change "its runtime PM routines" into "runtime resume" or similar. I think that will make the comment more clear.
Yes, thanks.
Quoted 8 lines
drivers/scsi/scsi_pm.c static int scsi_dev_type_resume(struct device *dev, · line 92
Quoted 3 lines
drivers/scsi/scsi_pm.c +3
-0+if (sdev->request_queue->dev) +return err; + }The 'dev' member only exists in struct
request_queueifCONFIG_PM=yso the above won't compile ifCONFIG_PM=n. How about adding a function ininclude/linux/blk-pm.hto check whether or not runtime PM has been enabled?
You are right.
Quoted 1 line
Otherwise this patch looks good to me.
Actually, I am thinking about removing all the pm_runtime_set_active() codes in both scsi_bus_resume_common() and scsi_dev_type_resume() - we don't need to forcibly set the runtime PM status to RPM_ACTIVE for either SCSI host/target or SCSI devices.
Whenever we access one SCSI device, either block layer or somewhere in the path (e.g. throgh sg IOCTL, sg_open() calls scsi_autopm_get_device()) should runtime resume the device first, and the runtime PM framework makes sure device's gets resumed as well. Thus, the pm_runtime_set_active() seems redundant. What do you think?
Thanks,
Can Guo.
Quoted 4 lines
Thanks,
Bart.
_______________________________________________
Linux-mediatek mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-mediatek
Hi Bart,
Resent, typo fixed.
Quoted 9 lines · Bart Van Assche
Quoted 4 lines · Can Guo
If block layer runtime PM is enabled for one SCSI device, then there is no need to forcibly change the SCSI device and its request queue's runtime PM status to active in
scsi_dev_type_resume(), since block layer PM shall resume the SCSI device on the demand of bios.Please change "along" into "alone" in the subject of this patch (if that is what you meant).
Aha, sorry, a typo here.
Quoted 8 lines
drivers/scsi/scsi_pm.c static int scsi_dev_type_resume(struct device *dev, · line 85
Quoted 4 lines
drivers/scsi/scsi_pm.c +3
-0+if (scsi_is_sdev_device (dev)) { +struct scsi_device *sdev; + sdev =to_scsi_device (dev);A minor comment: I think that "struct
scsi_device*sdev =to_scsi_device(dev);" fits on a single line.
Sure.
Quoted 6 lines
drivers/scsi/scsi_pm.c static int scsi_dev_type_resume(struct device *dev, · line 88
Quoted 2 lines
drivers/scsi/scsi_pm.c +2
-0+* If block layer runtime PM is enabled for the SCSI device, +* let block layer PM handle its runtime PM routines. Please change "its runtime PM routines" into "runtime resume" or similar. I think that will make the comment more clear.
Yes, thanks.
Quoted 8 lines
drivers/scsi/scsi_pm.c static int scsi_dev_type_resume(struct device *dev, · line 92
Quoted 3 lines
drivers/scsi/scsi_pm.c +3
-0+if (sdev->request_queue->dev) +return err; + }The 'dev' member only exists in struct
request_queueifCONFIG_PM=yso the above won't compile ifCONFIG_PM=n. How about adding a function ininclude/linux/blk-pm.hto check whether or not runtime PM has been enabled?
You are right.
Quoted 2 lines
Otherwise this patch looks good to me.
Actually, I am thinking about removing all the pm_runtime_set_active() codes in both scsi_bus_resume_common() and scsi_dev_type_resume() - we don't need to forcibly set the runtime PM status to RPM_ACTIVE for either SCSI host/target or SCSI devices.
Whenever we access one SCSI device, either block layer or somewhere in the path (e.g. throgh sg IOCTL, sg_open() calls scsi_autopm_get_device()) should runtime resume the device first, and the runtime PM framework makes sure device's parent (and its parent's parent and so on)gets resumed as well. Thus, the pm_runtime_set_active() seems redundant. What do you think?
Thanks,
Can Guo.
Quoted 3 lines
Thanks,
Bart.
_______________________________________________
Linux-mediatek mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-mediatek
Quoted 11 lines · Can Guo
Actually, I am thinking about removing all the
pm_runtime_set_active()codes in bothscsi_bus_resume_common()andscsi_dev_type_resume()- we don't need to forcibly set the runtime PM status toRPM_ACTIVEfor either SCSI host/target or SCSI devices.Whenever we access one SCSI device, either block layer or somewhere in the path (e.g. throgh sg IOCTL,
sg_open()callsscsi_autopm_get_device()) should runtime resume the device first, and the runtime PM framework makes sure device's parent (and its parent's parent and so on)gets resumed as well.
Thus, thepm_runtime_set_active()seems redundant. What do you think?
Hi Can,
It is not clear to me why the pm_runtime_set_active() calls occur in the scsi_pm.c source file since the block layer automatically activates block devices if necessary. Maybe these calls are a leftover from a time when runtime suspended devices were not resumed automatically by the block layer? Anyway, I'm fine with removing these calls.
Thanks,
Bart.
Hi Bart,
Quoted 21 lines · Bart Van Assche
Quoted 11 lines · Can Guo
Actually, I am thinking about removing all the
pm_runtime_set_active()codes in bothscsi_bus_resume_common()andscsi_dev_type_resume()- we don't need to forcibly set the runtime PM status toRPM_ACTIVEfor either SCSI host/target or SCSI devices.Whenever we access one SCSI device, either block layer or somewhere in the path (e.g. throgh sg IOCTL,
sg_open()callsscsi_autopm_get_device()) should runtime resume the device first, and the runtime PM framework makes sure device's parent (and its parent's parent and so on)gets resumed as well.
Thus, thepm_runtime_set_active()seems redundant. What do you think?Hi Can,
It is not clear to me why the
pm_runtime_set_active()calls occur in thescsi_pm.csource file since the block layer automatically activates block devices if necessary. Maybe these calls are a leftover from a time when runtime suspended devices were not resumed automatically by the block layer? Anyway, I'm fine with removing these calls.
Quoted 3 lines
Thanks,
Bart.
Yes, I agree with you. Let me test the new patch (which removes all the pm_runtime_set_active() calls) first, if no issue found, I will upload it for review.
Thanks,
Can Guo.