| Message ID | 1605861443-11459-1-git-send-email-cang@codeaurora.org |
|---|---|
| State | New |
| Headers | show |
| Series | [RFC,v2,1/1] scsi: pm: Leave runtime PM status alone during system resume/thaw/restore | expand |
On Sat, Nov 21, 2020 at 09:00:02AM -0800, Bart Van Assche wrote: > On 11/20/20 8:35 AM, Alan Stern wrote: > > On Fri, Nov 20, 2020 at 12:37:22AM -0800, Can Guo wrote: > >> Runtime resume is handled by runtime PM framework, no need to forcibly > >> set runtime PM status to RPM_ACTIVE during system resume/thaw/restore. > > > > Sorry, I don't understand this explanation at all. > > > > Sure, runtime resume is handled by the runtime PM framework. But this > > patch changes the code for system resume, which is completely different. > > > > Following a system resume, the hardware will be at full power. We don't > > want the kernel to think that the device is still in runtime suspend; > > otherwise is would never put the device back into low-power mode. > > Hi Alan, > > Does this mean that every driver needs similar code for handling runtime > suspended devices upon system resume? If so, would it be possible to > move that code into the power management core (drivers/base/power)? That's a complicated story. In short, many drivers need to do this, but not all. There is a complex collection of settings available for subsystems or drivers that would like their devices to remain in runtime system across a system sleep. For the subsystems/drivers that don't care to deal with this complexity or don't have any special requirements -- yes, they all need to include code like this in their system-resume paths. I had a very long discussion with Rafael Wysocki about all this starting last March; you can find the relevant emails beginning roughly here: https://marc.info/?l=linux-pm&m=158516934924947&w=2 and continuing through a few different threads. Rafael ended up making a large number of changes to the PM core and API to simplify things, straighten them out, and improve the documentation. But we never did try to add this automatic set-runtime-active thing into the core. Probably we wanted all the other changes to settle down before trying to do it, and then just forgot about it. In fact, I'm not certain that it is possible now, but we should look into it. Alan Stern
Hi Alan, On 2020-11-21 00:35, Alan Stern wrote: > On Fri, Nov 20, 2020 at 12:37:22AM -0800, Can Guo wrote: >> Runtime resume is handled by runtime PM framework, no need to forcibly >> set runtime PM status to RPM_ACTIVE during system resume/thaw/restore. > > Sorry, I don't understand this explanation at all. > > Sure, runtime resume is handled by the runtime PM framework. But this > patch changes the code for system resume, which is completely > different. > > Following a system resume, the hardware will be at full power. We > don't > want the kernel to think that the device is still in runtime suspend; > otherwise is would never put the device back into low-power mode. How about adding below lines to the patch? diff --git a/drivers/scsi/scsi_pm.c b/drivers/scsi/scsi_pm.c index 908f27f..7ebe582 100644 --- a/drivers/scsi/scsi_pm.c +++ b/drivers/scsi/scsi_pm.c @@ -75,9 +75,11 @@ static int scsi_dev_type_resume(struct device *dev, const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL; int err = 0; - err = cb(dev, pm); - scsi_device_resume(to_scsi_device(dev)); - dev_dbg(dev, "scsi resume: %d\n", err); + if (pm_runtime_active(dev)) { + err = cb(dev, pm); + scsi_device_resume(to_scsi_device(dev)); + dev_dbg(dev, "scsi resume: %d\n", err); + } return err; } Whenever a device is accessed, the issuer or somewhere in the path should do something like pm_runtime_get_sync (e.g. in sg_open()) or pm_runtime_resume() (e.g. in blk_queue_enter()), in either sync or async way. After the job (read/write/ioctl or whatever) is done, either a pm_runtime_put_sync() or auto runtime suspend puts the device back into runtime suspended/low-power mode. Since the func scsi_bus_suspend_common() does nothing if device is already in runtime suspended mode, scsi_dev_type_resume() should only resume the device if it is runtime active. Thanks, Can Guo. > Alan Stern > >> Cc: Stanley Chu <stanley.chu@mediatek.com> >> Cc: Bart Van Assche <bvanassche@acm.org> >> Cc: Alan Stern <stern@rowland.harvard.edu> >> Signed-off-by: Can Guo <cang@codeaurora.org> >> --- >> >> Changes since v1: >> - Incorporated Bart's comments >> >> --- >> drivers/scsi/scsi_pm.c | 24 +----------------------- >> 1 file changed, 1 insertion(+), 23 deletions(-) >> >> diff --git a/drivers/scsi/scsi_pm.c b/drivers/scsi/scsi_pm.c >> index 3717eea..908f27f 100644 >> --- a/drivers/scsi/scsi_pm.c >> +++ b/drivers/scsi/scsi_pm.c >> @@ -79,25 +79,6 @@ 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); >> - >> - /* >> - * 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 (!err && scsi_is_sdev_device(dev)) { >> - struct scsi_device *sdev = to_scsi_device(dev); >> - >> - blk_set_runtime_active(sdev->request_queue); >> - } >> - } >> - >> return err; >> } >> >> @@ -165,11 +146,8 @@ static int scsi_bus_resume_common(struct device >> *dev, >> */ >> if (strncmp(scsi_scan_type, "async", 5) != 0) >> async_synchronize_full_domain(&scsi_sd_pm_domain); >> - } else { >> - pm_runtime_disable(dev); >> - pm_runtime_set_active(dev); >> - pm_runtime_enable(dev); >> } >> + >> return 0; >> } >> >> -- >> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a >> Linux Foundation Collaborative Project. >>
On Mon, Nov 23, 2020 at 09:23:53AM +0800, Can Guo wrote: > Hi Alan, > > On 2020-11-21 00:35, Alan Stern wrote: > > On Fri, Nov 20, 2020 at 12:37:22AM -0800, Can Guo wrote: > > > Runtime resume is handled by runtime PM framework, no need to forcibly > > > set runtime PM status to RPM_ACTIVE during system resume/thaw/restore. > > > > Sorry, I don't understand this explanation at all. > > > > Sure, runtime resume is handled by the runtime PM framework. But this > > patch changes the code for system resume, which is completely different. > > > > Following a system resume, the hardware will be at full power. We don't > > want the kernel to think that the device is still in runtime suspend; > > otherwise is would never put the device back into low-power mode. > > How about adding below lines to the patch? > > diff --git a/drivers/scsi/scsi_pm.c b/drivers/scsi/scsi_pm.c > index 908f27f..7ebe582 100644 > --- a/drivers/scsi/scsi_pm.c > +++ b/drivers/scsi/scsi_pm.c > @@ -75,9 +75,11 @@ static int scsi_dev_type_resume(struct device *dev, > const struct dev_pm_ops *pm = dev->driver ? dev->driver->pm : NULL; > int err = 0; > > - err = cb(dev, pm); > - scsi_device_resume(to_scsi_device(dev)); > - dev_dbg(dev, "scsi resume: %d\n", err); > + if (pm_runtime_active(dev)) { > + err = cb(dev, pm); > + scsi_device_resume(to_scsi_device(dev)); > + dev_dbg(dev, "scsi resume: %d\n", err); > + } > > return err; > } > > Whenever a device is accessed, the issuer or somewhere in the path > should do something like pm_runtime_get_sync (e.g. in sg_open()) or > pm_runtime_resume() (e.g. in blk_queue_enter()), in either sync or > async way. After the job (read/write/ioctl or whatever) is done, > either a pm_runtime_put_sync() or auto runtime suspend puts the device > back into runtime suspended/low-power mode. Since the func > scsi_bus_suspend_common() does nothing if device is already in runtime > suspended mode, scsi_dev_type_resume() should only resume the device > if it is runtime active. You're starting to think along the right lines, but you are ignoring all the other work that people have already done for handling these cases. Please read Documentation/driver-api/pm/devices.rst very carefully, especially the parts about returning a positive value from the ->prepare callback (also known as "direct-complete" and related to the DPM_FLAG_NO_DIRECT_COMPLETE and DPM_FLAG_SMART_PREPARE flags) and the parts about the DPM_FLAG_SMART_SUSPEND and DPM_FLAG_MAY_SKIP_RESUME flags. Then think about what you want to accomplish and write a patch that takes all this information into account. Key point: At no time should any part of the kernel think that the device is in a low-power state when it is actually in a high-power state, or vice versa. Alan Stern
diff --git a/drivers/scsi/scsi_pm.c b/drivers/scsi/scsi_pm.c index 3717eea..908f27f 100644 --- a/drivers/scsi/scsi_pm.c +++ b/drivers/scsi/scsi_pm.c @@ -79,25 +79,6 @@ 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); - - /* - * 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 (!err && scsi_is_sdev_device(dev)) { - struct scsi_device *sdev = to_scsi_device(dev); - - blk_set_runtime_active(sdev->request_queue); - } - } - return err; } @@ -165,11 +146,8 @@ static int scsi_bus_resume_common(struct device *dev, */ if (strncmp(scsi_scan_type, "async", 5) != 0) async_synchronize_full_domain(&scsi_sd_pm_domain); - } else { - pm_runtime_disable(dev); - pm_runtime_set_active(dev); - pm_runtime_enable(dev); } + return 0; }
Runtime resume is handled by runtime PM framework, no need to forcibly set runtime PM status to RPM_ACTIVE during system resume/thaw/restore. Cc: Stanley Chu <stanley.chu@mediatek.com> Cc: Bart Van Assche <bvanassche@acm.org> Cc: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Can Guo <cang@codeaurora.org> --- Changes since v1: - Incorporated Bart's comments --- drivers/scsi/scsi_pm.c | 24 +----------------------- 1 file changed, 1 insertion(+), 23 deletions(-)