| Message ID | cover.1574179738.git.leonard.crestez@nxp.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | PM / devfreq: Add dev_pm_qos support with minimal changes | expand |
On Tuesday, November 19, 2019 5:12:12 PM CET Leonard Crestez wrote: > Add dev_pm_qos notifiers to devfreq core in order to support frequency > limits via dev_pm_qos_add_request. > > Unlike the rest of devfreq the dev_pm_qos frequency is measured in kHz, > this is consistent with current dev_pm_qos usage for cpufreq and > allows frequencies above 2Ghz (pm_qos expresses limits as s32). > > Like with cpufreq the handling of min_freq/max_freq is moved to the > dev_pm_qos mechanism. Constraints from userspace are no longer clamped on > store, instead all values can be written and we only check against OPPs in a > new devfreq_get_freq_range function. This is consistent with the design of > dev_pm_qos. > > Notifiers from pm_qos are executed under a single global dev_pm_qos_mtx and > need to take devfreq->lock, this means that calls into dev_pm_qos while holding > devfreq->lock are not allowed (lockdep warns about possible deadlocks). > > Fix this by only adding the qos request and notifiers after devfreq->lock is > released inside devfreq_add_device. In theory this means sysfs writes > are possible before the min/max requests are initialized so we guard > against that explictly. The dev_pm_qos_update_request function would > otherwise print a big WARN splat. > > Alternatively devfreq initialization could be refactored to avoid taking > devfreq->lock but that requires several intricate changes: > > https://patchwork.kernel.org/cover/11242865/ > > I considered making dev_pm_qos call notifiers outside the lock but > that's another complex refactoring and it's difficult to ensure > correctness. If two identical qos requests are made in parallel then the > second shouldn't return until all notifiers are completely executed for > the first and QOS is enforced; otherwise it mostly defeats the purpose > of making proactive requests. > > This series implements the minimal changes in order to implement dev_pm_qos > support for devfreq. It only costs a little defensive programming. > > This series is also marked as [RFC] because it depends on restoring > DEV_PM_QOS_MIN/MAX_FREQUENCY inside the pm core: > > https://patchwork.kernel.org/cover/11250413/ > > --- > Changes since "big version" v10: > * Drop accepted cleanups > * Work with current locking approach (split cleanups into other series) > * Drop acks and deliberately relabel as a new series. It still incorporates > most previous discussion but takes a different approach to locking. > * Don't print errors if devfreq_dev_release is called on error cleanup from > devfreq_add_device, just accept that requests and notifiers might not be > registered yet. I wish dev_pm_qos cleanups behaved like standard "kfree" and > silently did nothing when there's nothing to be done. > Link to v10: https://patchwork.kernel.org/project/linux-arm-kernel/list/?series=196443 > > Leonard Crestez (2): > PM / devfreq: Add PM QoS support > PM / devfreq: Use PM QoS for sysfs min/max_freq > > drivers/devfreq/devfreq.c | 151 ++++++++++++++++++++++++++++++++++---- > include/linux/devfreq.h | 14 +++- > 2 files changed, 145 insertions(+), 20 deletions(-) Please resend this series as non-RFC with the ACKs from Chanwoo included. It may still be viable to push it for 5.5 during the -rc period. Thanks!
On 12/4/19 7:46 PM, Rafael J. Wysocki wrote: > On Tuesday, November 19, 2019 5:12:12 PM CET Leonard Crestez wrote: >> Add dev_pm_qos notifiers to devfreq core in order to support frequency >> limits via dev_pm_qos_add_request. >> >> Unlike the rest of devfreq the dev_pm_qos frequency is measured in kHz, >> this is consistent with current dev_pm_qos usage for cpufreq and >> allows frequencies above 2Ghz (pm_qos expresses limits as s32). >> >> Like with cpufreq the handling of min_freq/max_freq is moved to the >> dev_pm_qos mechanism. Constraints from userspace are no longer clamped on >> store, instead all values can be written and we only check against OPPs in a >> new devfreq_get_freq_range function. This is consistent with the design of >> dev_pm_qos. >> >> Notifiers from pm_qos are executed under a single global dev_pm_qos_mtx and >> need to take devfreq->lock, this means that calls into dev_pm_qos while holding >> devfreq->lock are not allowed (lockdep warns about possible deadlocks). >> >> Fix this by only adding the qos request and notifiers after devfreq->lock is >> released inside devfreq_add_device. In theory this means sysfs writes >> are possible before the min/max requests are initialized so we guard >> against that explictly. The dev_pm_qos_update_request function would >> otherwise print a big WARN splat. >> >> Alternatively devfreq initialization could be refactored to avoid taking >> devfreq->lock but that requires several intricate changes: >> >> https://protect2.fireeye.com/url?k=0cd18bf1-51473286-0cd000be-0cc47a312ab0-56d7069e720c53c0&u=https://patchwork.kernel.org/cover/11242865/ >> >> I considered making dev_pm_qos call notifiers outside the lock but >> that's another complex refactoring and it's difficult to ensure >> correctness. If two identical qos requests are made in parallel then the >> second shouldn't return until all notifiers are completely executed for >> the first and QOS is enforced; otherwise it mostly defeats the purpose >> of making proactive requests. >> >> This series implements the minimal changes in order to implement dev_pm_qos >> support for devfreq. It only costs a little defensive programming. >> >> This series is also marked as [RFC] because it depends on restoring >> DEV_PM_QOS_MIN/MAX_FREQUENCY inside the pm core: >> >> https://protect2.fireeye.com/url?k=304e4bce-6dd8f2b9-304fc081-0cc47a312ab0-ae5dd24df5aed564&u=https://patchwork.kernel.org/cover/11250413/ >> >> --- >> Changes since "big version" v10: >> * Drop accepted cleanups >> * Work with current locking approach (split cleanups into other series) >> * Drop acks and deliberately relabel as a new series. It still incorporates >> most previous discussion but takes a different approach to locking. >> * Don't print errors if devfreq_dev_release is called on error cleanup from >> devfreq_add_device, just accept that requests and notifiers might not be >> registered yet. I wish dev_pm_qos cleanups behaved like standard "kfree" and >> silently did nothing when there's nothing to be done. >> Link to v10: https://protect2.fireeye.com/url?k=221a9dd6-7f8c24a1-221b1699-0cc47a312ab0-2fdb90bdac06bfe8&u=https://patchwork.kernel.org/project/linux-arm-kernel/list/?series=196443 >> >> Leonard Crestez (2): >> PM / devfreq: Add PM QoS support >> PM / devfreq: Use PM QoS for sysfs min/max_freq >> >> drivers/devfreq/devfreq.c | 151 ++++++++++++++++++++++++++++++++++---- >> include/linux/devfreq.h | 14 +++- >> 2 files changed, 145 insertions(+), 20 deletions(-) > > Please resend this series as non-RFC with the ACKs from Chanwoo included. > > It may still be viable to push it for 5.5 during the -rc period. > If Leonard resend this patch with Ack, I'll send the pull-request for -rc period with the related minimal patches. Thanks.